Handbook of Teichmuller Theory

titelei11_papadopoulos 30.4.
2007 10:00 Uhr Seite 1
IRMA Lectures in Mathematics and Theoretical Physics 11

Series Editor: Vladimir G. Turaev
Institut de Recherche Mathématique Avancée

Université Louis Pasteur et CNRS
7 rue René Descartes
67084 Strasbourg Cedex
France
titelei11_papadopoulos 30.4.2007 10:00 Uhr Seite 2
IRMA Lectures in Mathematics and Theoretical Physics

Edited by Vladimir G. Turaev
This series is devoted to the publication of research monographs, lecture notes, and other
materials arising from programs of the Institut de Recherche Mathématique Avancée
(Strasbourg, France). The goal is to promote recent advances in mathematics and theoretical
physics and to make them accessible to wide circles of mathematicians, physicists, and
students of these disciplines.
Previously published in this series:
1 Deformation Quantization, Gilles Halbout (Ed.)

2 Locally Compact Quantum Groups and Groupoids, Leonid Vainerman (Ed.)
3 From Combinatorics to Dynamical Systems, Frédéric Fauvet and Claude Mitschi (Eds.)
4 Three courses on Partial Differential Equations, Eric Sonnendrücker (Ed.)
5 Infinite Dimensional Groups and Manifolds, Tilman Wurzbacher (Ed.)
6 Athanase Papadopoulos, Metric Spaces, Convexity and Nonpositive Curvature
7 Numerical Methods for Hyperbolic and Kinetic Problems, Stéphane Cordier,
Thierry Goudon, Michaël Gutnic and Eric Sonnendrücker (Eds.)
8 AdS/CFT Correspondence: Einstein Metrics and Their Conformal Boundaries,
Oliver Biquard (Ed.)
9 Differential Equations and Quantum Groups, D. Bertrand, B. Enriquez,
C. Mitschi, C. Sabbah and R. Schäfke (Eds.)
10 Physics and Number Theory, Louise Nyssen (Ed.)
Volumes 1–5 are available from Walter de Gruyter (www.degruyter.de)

Handbook of
Teichmüller Theory
Volume I
Athanase Papadopoulos
Editor
Editor:
Institut de Recherche Mathématique Avancée
Université Louis Pasteur et CNRS
7 Rue René Descartes
67084 Strasbourg Cedex
France
2000 Mathematics Subject Classification: Primary 30-00, 32G15, 30F60; secondary 30C62, 57N16, 53A35,
20F65, 30F20, 30F25, 30F10, 30F30, 30F40, 30F45, 14H15, 20H10, 30F15, 53B35, 57M07, 57M60, 14H60,
14D20, 57M20, 20F38.
ISBN 978-3-03719-029-6
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. For any kind of use permission
of the copyright owner must be obtained.
© 2007 European Mathematical Society

Contact address:
European Mathematical Society Publishing House

Seminar for Applied Mathematics
ETH-Zentrum FLI C4
CH-8092 Zürich
Switzerland
Phone: +41 (0)44 632 34 36

Email: info@ems-ph.org
Homepage: www.ems-ph.org
Typeset using the author’s TE X files: I. Zimmermann, Freiburg

Printed in Germany
987654321
Foreword
In a broad sense, the subject of Teichmüller theory is the study of moduli spaces
for geometric structures on surfaces. This subject makes important connections be-
tween several areas in mathematics, including low-dimensional topology, hyperbolic
geometry, dynamical systems theory, differential geometry, algebraic topology, repre-
sentations of discrete groups in Lie groups, symplectic geometry, topological quantum
field theory, string theory, and there are others.
The central object in this theory is the Teichmüller space of a surface. As is
well-known, this space can be seen from different points of view. It is a space of
equivalence classes of conformal structures on the surface, a space of equivalence
classes of hyperbolic metrics on this surface, and a space of equivalence classes of
representations of the fundamental group of the surface into a Lie group (primarily,
PSL(2, R), but there are several others). These three points of view are equally impor-
tant in the study of Teichmüller space and, in several situations, they cannot be clearly
separated. Furthermore, Teichmüller space inherits from these different points of view
various structures, including several interesting metrics (Teichmüller, Weil–Petersson,
Thurston, Bergman, Carathéodory, Kähler–Einstein, McMullen, etc.), a natural com-
plex structure, a symplectic structure, a real analytic structure, the structure of an
algebraic set, cellular structures, various boundary structures, a natural discrete action
by the mapping class group, a quantization theory of its Poisson and symplectic struc-
tures, a measure-preserving geodesic flow, a horocyclic flow, and the list of structures
goes on and on. Teichmüller theory is growing at a fantastic rate, and the richness of
this theory is to a large extent a consequence of the diversity and the richness of the
structures that Teichmüller space itself carries.
The purpose of this Handbook is to give a comprehensive picture of the classical
and of the recent developments in Teichmüller theory. The range of this picture will
hopefully include all the aspects mentioned above although, because new ideas and
new connections come out regularly in this theory, the picture will necessarily be
incomplete. Nevertheless we hope that the Handbook will reflect the beauty, the
liveliness and the richness of Teichmüller theory.
I have tried to divide this first volume of the Handbook into parts. There were
several possibilities, none of which imposed itself as being more natural or more
efficient than the others. I finally ended up with the following division into four
parts.1
• The metric and the analytic theory, 1
• The group theory, 1
• Surfaces with singularities and discrete Riemann surfaces
• The quantum theory, 1
1 The titles of some parts are followed by the numeral 1 because they will be continued in Volume II of the
Handbook.
vi Foreword
Volume II of the Handbook will contain more material on the analytic and the
metric theory, on the group theory, on the quantum theory, as well as sections on
cluster algebras, on representation theory and higher Teichmüller theory, on complex
projective structures, on the Teichmüller geodesic flow and other dynamical aspects,
and on the Grothendieck–Teichmüller theory.
I would like to take this opportunity to thank Vladimir Turaev for his encour-
agement in this project, Manfred Karbe and Irene Zimmermann for their invaluable
collaboration in the final editing process and, of course, all the authors who contributed
to the Handbook. It was a pleasure to work with all these people.
Strasbourg, April 2007 Athanase Papadopoulos

Contents
Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Introduction to Teichmüller theory, old and new

by Athanase Papadopoulos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Part A. The metric and the analytic theory, 1

Chapter 1. Harmonic maps and Teichmüller theory
by Georgios D. Daskalopoulos and Richard A. Wentworth . . . . . . . . . . . . . . . . . . . . . . 33
Chapter 2. On Teichmüller’s metric and Thurston’s asymmetric metric

on Teichmüller space
by Athanase Papadopoulos and Guillaume Théret . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Chapter 3. Surfaces, circles, and solenoids

by Robert C. Penner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Chapter 4. About the embedding of Teichmüller space in the space

of geodesic Hölder distributions
by Jean-Pierre Otal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Chapter 5. Teichmüller spaces, triangle groups and Grothendieck dessins

by William J. Harvey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Chapter 6. On the boundary of Teichmüller disks in Teichmüller

and in Schottky space
by Frank Herrlich and Gabriela Schmithüsen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Part B. The group theory, 1

Chapter 7. Introduction to mapping class groups of surfaces and related groups
by Shigeyuki Morita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Chapter 8. Geometric survey of subgroups of mapping class groups

by Lee Mosher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
viii Contents
Chapter 9. Deformations of Kleinian groups

by Albert Marden . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Chapter 10. Geometry of the complex of curves and of Teichmüller space

by Ursula Hamenstädt. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .447
Part C. Surfaces with singularities and discrete Riemann surfaces

Chapter 11. Parameters for generalized Teichmüller spaces
by Charalampos Charitos and Ioannis Papadoperakis . . . . . . . . . . . . . . . . . . . . . . . . 471
Chapter 12. On the moduli space of singular euclidean surfaces

by Marc Troyanov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
Chapter 13. Discrete Riemann surfaces

by Christian Mercat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
Part D. The quantum theory, 1

Chapter 14. On quantizing Teichmüller and Thurston theories
by Leonid O. Chekhov and Robert C. Penner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
Chapter 15. Dual Teichmüller and lamination spaces

by Vladimir V. Fock and Alexander B. Goncharov . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
Chapter 16. An analog of a modular functor from quantized Teichmüller theory

by Jörg Teschner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685
Chapter 17. On quantum moduli space of flat PSL2 (R)-connections

by Rinat M. Kashaev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761
List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 785
Introduction to Teichmüller theory, old and new
Teichmüller theory is one of those few wonderful subjects which bring together, at an
equally important level, fundamental ideas coming from different fields. Among the
fields related to Teichmüller theory, one can surely mention complex analysis, hyper-
bolic geometry, the theory of discrete groups, algebraic geometry, low-dimensional
topology, differential geometry, Lie group theory, symplectic geometry, dynamical
systems, number theory, topological quantum field theory, string theory, and there are
many others.
Let us start by recalling a few definitions.
Let Sg,p be a connected orientable topological surface of genus g ≥ 0 with p ≥ 0
punctures. Any such surface admits a complex structure, that is, an atlas of charts with
values in the complex plane C and whose coordinate changes are holomorphic. In
the classical theory, one considers complex structures Sg,p for which each puncture of
Sg,p has a neighborhood which is holomorphically equivalent to a punctured disk in C.
To simplify the exposition, we shall suppose that the orientation induced on Sg,p by the
complex structure coincides with the orientation of this surface. Homeomorphisms
of the surface act in a natural manner on atlases, and two complex structures on Sg,p
are said to be equivalent if there exists a homeomorphism of the surface which is
homotopic to the identity and which sends one structure to the other. The surface Sg,p
admits infinitely many non-equivalent complex structures, except if this surface is a
sphere with at most three punctures.
To say things precisely, we introduce some notation. Let Cg,p be the space of all
complex structures on Sg,p and let Diff+ (Sg,p ) be the group of orientation-preserving
diffeomorphisms of Sg,p . We consider the action of Diff+ (Sg,p ) by pullback on
Cg,p . The quotient space Mg,p = Cg,p /Diff+ (Sg,p ) is called Riemann’s moduli
space of deformations of complex structures on Sg,p . This space was considered by
G. F. B. Riemann in his famous paper on Abelian functions, Theorie der Abel’schen
Functionen, Crelle’s Journal, Band 54 (1857), in which he studied moduli for algebraic
curves. In that paper, Riemann stated, without giving a formal proof, that the space
of deformations of equivalence classes of conformal structures on a closed orientable
surface of genus g ≥ 2 is of complex dimension 3g − 3.
The Teichmüller space Tg,p of Sg,p was introduced in the 1930s by Oswald Teich-
müller. It is defined as the quotient of the space Cg,p of complex structures by the
group Diff+ 0 (Sg,p ) of orientation-preserving diffeomorphisms of Sg,p that are isotopic
to the identity. The group Diff+ +
0 (Sg,p ) is a normal subgroup of Diff (Sg,p ), and the
+ +
quotient group g,p = Diff (Sg,p )/Diff0 (Sg,p ) is called the mapping class group of
Sg,p (sometimes also called the modular group, or the Teichmüller modular group)
2 Athanase Papadopoulos
of Sg,p . The mapping class group g,p acts on the Teichmüller space Tg,p , and the
quotient of Tg,p by this action is Riemann’s moduli space Mg,p .
During a remarkably brief period of time (1935–1941), Teichmüller wrote about
thirty papers which laid the foundations of the theory which now bears his name. After
Teichmüller’s death in 1943 (at the age of 30), L. V. Ahlfors, L. Bers, H. E. Rauch
and several of their students and collaborators started a project that provided a solid
grounding for Teichmüller’s ideas. The realization of this project took more than
two decades, during which the whole complex-analytic theory of Teichmüller space
was built. In the 1970s, W. P. Thurston opened a new and wide area of research by
introducing beautiful techniques of hyperbolic geometry in the study of Teichmüller
space and of its asymptotic geometry. Thurston’s work highlighted this space as a
central object in the field of low-dimensional topology. Thurston’s ideas are still
being developed and extended today by his students and followers. In the 1980s,
there also evolved an essentially combinatorial treatment of Teichmüller and moduli
spaces, involving techniques and ideas from high-energy physics, namely from string
theory. The current research interests in Teichmüller theory from the point of view of
mathematical physics include the quantization of Teichmüller space using the Weil–
Petersson symplectic and Poisson geometry of this space, as well as gauge-theoretic
extensions of these structures. The quantization theories are also developed with the
purpose of finding new invariants of hyperbolic 3-manifolds.
My aim in the rest of this introduction is twofold: first, to give an exposition of the
structure of Teichmüller theory that is studied in this volume of the Handbook, and
second, to give an overview of the material contained in this volume. The goal is to
give a flavour of the subject. Of course, the exposition will be very sketchy at some
places.1
1 An overview of the structure of Teichmüller space

Teichmüller space Tg,p has a natural topology which makes it homeomorphic to an
open ball of dimension 6g − 6 + 2p. It has a complex structure which is induced by
an embedding of this space as a bounded pseudo-convex domain of holomorphy in a
Banach space. It carries several interesting metrics, in addition to the usual metrics
that are associated to a bounded domain of holomorphy viz. the Carathéodory, the
Kobayashi and the Bergman metrics. It also has a symplectic and a Poisson structure,
various boundary structures and several other structures which I now briefly describe.
1 In order not to overload this introduction, I decided not to give the complete references for the results that
I cite. Most of the references are given in the chapters that follow this introduction and, furthermore, with the
availability of electronic supports, it is nowadays easy to find the exact reference for a result, given its statement,
its author’s name and the year it was written.
Introduction to Teichmüller theory, old and new 3
1.1 Complex structure

Let Sg,p be as before a surface of genus g ≥ 0 with p ≥ 0 punctures. If the Euler
characteristic of Sg,p is negative, then the Teichmüller space Tg,p of Sg,p is a com-
plex manifold of complex dimension 3g − 3 + p. The bases of the complex analytic
theory of Teichmüller space were developed by Ahlfors and Bers. There are several
ways of defining the complex structure of Teichmüller space. All of them use deep
results from analysis, and giving the details of this theory would be far beyond the
scope of this introduction. We only mention that there is a famous embedding (called
Bers’ embedding) which realizes Tg,p as an open bounded pseudo-convex (but not
strictly pseudo-convex) domain in C3g−3+p . This embedding depends on the choice
of a basepoint in Teichmüller space. Classically, the holomorphic cotangent space at
a point of Teichmüller space is described by the Kodaira–Spencer deformation theory,
which parametrizes the deformations of a conformal structure in terms of the sheaf
of holomorphic vector fields on the given Riemann surface. In Bers’ embedding of
Tg,p , the holomorphic cotangent space at a point is identified with the vector space of
holomorphic quadratic differentials with simple poles at the punctures on a Riemann
surface representing the point. By the theory of quasiconformal mappings, any com-
plex structure on a surface is specified by a Beltrami differential of norm less than
one, and this leads to a description of the holomorphic tangent space at a point of
Teichmüller space as a vector space of Beltrami differentials of norm less than one
divided by a subspace of differentials which induce trivial deformations. We note
that besides providing a description of the complex structure of Teichmüller space,
these descriptions of the tangent and cotangent spaces are used to give an infinitesimal
description of the Teichmüller metric which we recall below, and that as such, the
tangent and cotangent spaces are Banach spaces and not Hilbert spaces, which reflects
the fact that the Teichmüller metric is a Finsler metric and not a Riemannian metric.
We also note that the descriptions of the tangent and cotangent spaces by holomor-
phic quadratic differentials and Beltrami differentials are key tools for the definition
of Hermitian metrics on Teichmüller space, for instance the Weil–Petersson metric
which we discuss below.
It follows from the definitions that the mapping class group g,p preserves this
complex structure, that is, it acts on Teichmüller space by biholomorphic maps. Con-
versely, by a result of H. L. Royden, any biholomorphic automorphism of Teichmüller
space is induced by an element of the mapping class group. The quotient moduli space
Mg,p is a complex orbifold of dimension 3g − 3 + p.
1.2 Metric structures

Teichmüller space carries various metrics, each arising naturally from a particular
viewpoint on the space. There are three metrics on Teichmüller space which will be
considered in some detail in this volume: Teichmüller’s metric, the Weil–Petersson
metric and Thurston’s asymmetric metric (which, because of its asymmetry, is not a
metric in the usual sense of the word). I will say a few words about each of these
metrics.
Teichmüller’s metric. This metric is obtained by first defining the distance between
two conformal structures g and h on the surface Sg,p to be 21 inff log K(f ), where the
infimum is taken over all quasiconformal homeomorphisms f : (Sg,p , g) → (Sg,p , h)
that are isotopic to the identity and where K(f ) is the quasiconformal dilatation of f .
Teichmüller showed that the infimum is realized by a quasiconformal homeomor-
phism, and he gave a description of this homeomorphism in terms of a quadratic
differential on the domain conformal surface (Sg,p , g). This distance function on the
set of conformal structures is invariant by the action of the group of diffeomorphisms
isotopic to the identity on each factor, and it induces a distance function on Teichmüller
space Tg,p , which is Teichmüller’s metric. Teichmüller’s metric is a complete Finsler
metric which is not Riemannian unless the surface is a torus, in which case Teichmüller
space, equipped with Teichmüller’s metric, is isometric to the 2-dimensional hyper-
bolic plane. Teichmüller’s metric is geodesically convex, that is, any two points are
joined by a unique geodesic segment. The metric is also uniquely geodesic, that is, the
geodesic segment joining two arbitrary points is unique. Furthermore, any geodesic
segment can be extended in a unique way (up to parametrization) to a geodesic line.
For some time, it was believed that Teichmüller’s metric was nonpositively curved in
the sense of Busemann, that is, that the distance function between two parametrized
geodesics is convex. Indeed, in 1959, S. Kravetz published a paper containing this
statement as one of the main results. In 1971, M. Linch found a mistake in the argu-
ments of Kravetz and in 1975, H. Masur showed that the statement is false, by providing
examples of two geodesic rays whose associated distance function is bounded, and
therefore is not convex. By a result of Masur and Wolf (1995), Teichmüller’s metric
is not Gromov hyperbolic. Despite these facts, much of the work that has been done
on Teichmüller’s metric is motivated by analogies with metrics on simply connected
manifolds of negative curvature. We can mention in this connection that there is a
Teichmüller geodesic flow which preserves a natural measure, which induces a quo-
tient geodesic flow on moduli space, and that this quotient flow, like the geodesic
flow of a finite volume hyperbolic manifold, is ergodic. (This result was obtained
independently in 1982 by H. Masur and W. Veech.)
The mapping class group g,p of Sg,p acts on Tg,p by isometries of the Teich-
müller metric and, conversely, by a result of Royden (1970), every isometry group
of the Teichmüller metric is induced by an element of the mapping class group. At
the same time, Royden showed that the Teichmüller metric can be recovered directly
from the complex structure of Teichmüller space, by proving that this metric coincides
with the Kobayashi metric of this space. (In fact, Royden proved the first result as a
corollary of the second.)
The Weil–Petersson metric. This metric is a Riemannian metric which is also closely
related to the complex structure of Teichmüller space. Its definition starts with an L2 -
norm on the space of quadratic differentials at each point X of Teichmüller space,
considered as the holomorphic cotangent space at that point. The definition of this
norm makes use of the hyperbolic metric of a surface representing
−2 this point. More
precisely, this norm is given by the formula φ = X ρ (z)|φ(z)|2 |dz|2 , with
2
ρ(z)|dz| being the length element on the point X represented by a hyperbolic sur-
face. The Weil–Petersson inner product on the tangent space is then defined by tak-
ing a dual of this L2 inner product, the duality between the tangent and cotangent
spaces being defined through a natural pairing between quadratic differentials and
Beltrami differentials. Several chapters of this Handbook deal with this metric, and
we shall list here some of its important geometric features. L. Ahlfors showed that
the Weil–Petersson metric is Kähler. It has variable negative sectional curvature (a
result proved independently by A. Tromba and S. Wolpert). It is geodesically convex
but it is not complete (a result obtained by Masur and by Wolpert). Masur stud-
ied the completion of this metric, and he described the points on the frontier of this
completion as Riemann surfaces with nodes 2 . More precisely, Masur identified the
Weil–Petersson completion with the augmented Teichmüller space, whose frontier is
a stratified union of lower-dimensional spaces, each of which is the Teichmüller space
of a nodal surface. He showed that in some precise sense the tangential component of
the Weil–Petersson metric on Teichmüller space extends to the Weil–Petersson metric
of these boundary Teichmüller spaces. The sectional curvature of the Weil–Petersson
metric is unbounded (its supremum is zero and its infimum is −∞), except in the
cases where the complex dimension of Teichmüller space is 1. Ahlfors showed that
the holomorphic sectional curvature of the Weil–Petersson metric is negative. Wolpert
showed that the holomorphic sectional curvature and the Ricci curvature of this metric
have negative upper bounds (proving a conjecture made by Royden), with these up-
per bounds expressible in terms of the topological type of the surface. The mapping
class group acts by biholomorphic isometries on Teichmüller space equipped with the
Weil–Petersson metric, and this metric descends to a metric on moduli space. Masur
and Wolf proved (2002) that conversely, every Weil–Petersson isometry is induced by
an element of the mapping class group, except for a few surfaces of low genus. The
untreated cases were completely analyzed later on by J. Brock and D. Margalit (2004).
J. Brock and B. Farb (2004) showed that the Weil–Petersson metric is not Gromov
hyperbolic, except if the surface Sg,p is a torus with at most two punctures or a sphere
with at most five punctures. In all the other cases, the space has higher rank in the
sense of Gromov. S.Yamada (2001) showed that the Weil–Petersson completion of the
Weil–Petersson metric is a CAT(0)-space (that is, a non-positively curved space in the
sense of Cartan–Alexandrov–Toponogov). Brock initiated a new point of view on the
2 We note that surfaces with nodes appear naturally in complex analysis. A node is the simplest type of a
singularity of a complex curve. More precisely, while a smooth point of a complex curve is a point which admits
a neighborhood which is biholomorphically equivalent to an open disk in C, a node is a point which admits
a neighborhood which is biholomorphically equivalent to a space obtained by taking a disjoint union of two
disks in C and gluing them along a point. Riemann surfaces with nodes are complex curves whose points are
either smooth points or nodes, and they appear naturally as degenerations of smooth complex curves (that is, of
Riemann surfaces). Thus, it is not surprising that surfaces with nodes appear as boundary structures of spaces of
Riemann spaces.
Weil–Petersson metric, by establishing a quasi-isometry between this metric and the

metric on the vertices of the pants graph of the surface induced by the simplicial metric
on the 1-skeleton. This work has connections with volumes of quasi-Fuchsian hyper-
bolic three-manifolds, and it also leads to beautiful results on the global behaviour
of Weil–Petersson geodesics, quasi-geodesics of the Weil–Petersson metric being in
correspondence with quasi-geodesics in the pants graph. There is a result which has a
similar flavour, on the comparison between quasi-geodesics of the Teichmüller metric,
and the quasi-geodesics in the complex of curves, which is described in Chapter 10
by Ursula Hamenstädt.
There have been other recent breakthroughs on the Weil–Petersson metric, by
S. Wolpert, R. Wentworth, S. Yamada, M. Mirzakhani, B-Y. Chen, Z. Huang, joint
work by S. Y. Cheng and S. T. Yau and by K. Liu, X. Sun and S. T. Yau, and by others.
We also mention that the Weil–Petersson metric can be recovered through Bona-
hon’s embedding of Teichmüller space into the space of geodesic currents.
Recent work on the Weil–Petersson metric will be reported on in Volume II of this
Handbook.
Thurston’s asymmetric metric. This metric was introduced by Thurston in 1986,
in his paper Minimal Stretch maps between hyperbolic surfaces (which is still in an
unpublished form). Here, one considers Teichmüller space as the set of isotopy classes
of complete finite-area hyperbolic structures on Sg,p . If g and h are such structures,
one defines their “distance” to be the logarithm of the infimum of Lipschitz constants
of all diffeomorphisms ϕ : (Sg,p , g) → (Sg,p , h) which
are isotopic
to the identity,
dh ϕ(x),ϕ(y)
the Lipschitz constant of ϕ being, by definition, sup , the supremum being
dg x,y
taken over all distinct pairs of points x and y in Sg,p . This distance function on the set of
hyperbolic structures is invariant by the action on each factor of the group Diff+ 0 (Sg,p )
of diffeomorphisms of Sg,p that are isotopic to the identity, and it induces a function
on the Teichmüller space Tg,p of Sg,p , which we call Thurston’s asymmetric metric.
This is not a metric in the usual sense of the word since, as its name indicates, it does
not satisfy the symmetry axiom. Despite this asymmetry (and in some cases because
of this asymmetry), Thurston’s metric has beautiful geometric properties. Several of
these properties are expounded in Chapter 2 of this volume.
Let us note here that the study of asymmetric metrics has a long history, and we can
mention in this respect the name of Herbert Busemann, who did a systematic study of
such metrics, starting in the 1940s. The methods of synthetic geometry in the sense
of Busemann can be used to investigate Thurston’s asymmetric metric, for instance in
the study of the behaviour of its geodesics, its isometries, its visual boundaries and so
on. The techniques that are used by Thurston in the study of his asymmetric metric
are those of basic hyperbolic geometry, and it is most satisfying to see that with these
basic elementary techniques, a beautiful picture of a whole theory arises. There are
several good questions which one can ask about this metric, and some of them are
listed in Chapter 2.
The Carathéodory, the Kobayashi and the Bergman metrics. The Carathéodory
and the Kobayashi “metrics” are semi-metrics3 that are classically defined on any
complex space, and that are invariant under biholomorphic self-mappings of this space.
In the case of Teichmüller space, these semi-metrics are genuine metrics. Let us first
recall the definitions, since they are easy to state and are most appealing from the point
of view of synthetic geometry.
Let X be a complex space and let D be the unit disk in C equipped with its Poincaré
metric. The Carathéodory
semi-metric on X is defined by the formula dC (x, y) =
supf dD f (x), f (y) for all x and y in X, where the supremum is taken over all
holomorphic maps f : X → D.
The Kobayashi semi-metric on X is defined by a dual construction. One first
defines a map dK : X × X → R by d (x, y) = inf
K f,a,b dD (a, b) for all x and y in X,
the infimum being taken over all holomorphic maps f : D → X and over all a and
b in D satisfying f (a) = x and f (b) = y. The map dK does not necessarily satisfy
the triangle inequality, and the Kobayashi semi-metric dK is defined as the largest
semi-metric on X satisfying dK ≤ dK .
We already mentioned that H. L. Royden proved in 1970 that the Kobayashi met-
ric of Teichmüller space coincides with the Teichmüller metric. In 1974, C. J. Earle
proved that the Carathéodory metric on Teichmüller space is complete, and he raised
the question of whether this metric coincides with the Teichmüller metric. He also
asked the weaker question of whether this metric on Teichmüller space is proper (that
is, whether every closed bounded set is compact). Both questions were answered by
S. L. Krushkal’. In 1976, Krushkal’ proved that the Carathéodory metric on Teich-
müller space is proper, and in 1981 he proved that the Carathéodory metric does not
coincide with the Teichmüller metric, unless the complex dimension of this space is
one.
We also note that the Carathéodory and the Kobayashi metrics on a complex space
have infinitesimal descriptions, and that in general they are Finsler and not Riemannian
metrics.
The Bergman metric is a Kähler semi-metric that is associated to any bounded
domain of holomorphy. Its definition uses the Bergman Kernel. The Bergman metric
of Teichmüller space (seen as a bounded domain through Bers’embedding) is a genuine
metric.
By a result of K. T. Hahn (1976), the Carathéodory metric on any bounded domain
in CN is bounded from above by the Bergman metric. Therefore, by using Earle’s
result on the completeness of the Carathéodory metric, the Bergman metric is complete.
B.-Y. Chen showed (2004) that the Bergman metric of Teichmüller space is equivalent
to the Teichmüller metric in the sense that there exists a positive constant C such that
C dT ≤ dB ≤ CdT , where dT and dB denote respectively the Teichmüller and the
1
Bergman metrics. This result was also obtained, at about the same time, by K. Liu,
X. Sun and S.-T. Yau.
3 Recall that a semi-metric d : X × X → [0, ∞[ satisfies the axioms of a metric except possibly for the
separation axiom d(x, y) = 0 ⇒ x = y.
A survey of the Carathéodory, the Kobayashi and the Bergman metrics on Teich-
müller space is contained in a chapter by A. Fletcher and V. Markovich in Volume II
of this Handbook.
Other metrics. There are many other interesting metrics on Teichmüller space. We
mention a few of them, although they are not considered in this volume. S. Y. Cheng,
and S. T. Yau showed that there exists a complete Kähler–Einstein metric of negative
Ricci curvature on any bounded pseudo-convex domain in CN (1975). Using Bers’
embedding, this result shows that there is a unique complete Kähler–Einstein metric
on Teichmüller space. This metric is invariant under the action of the mapping class
group. It has constant negative scalar curvature, it descends to a complete Kähler–
Einstein metric on moduli space and it has been thoroughly studied by S. Y. Cheng,
N. Mok and S. T. Yau, and by algebraic geometers. In 2004, K. Liu, X. Sun and
S.-T. Yau showed that this Kähler–Einstein metric is equivalent to the Teichmüller
metric (a result which had been conjectured by Yau in 1986), which implies that the
Bergman and the Kähler–Einstein metrics are equivalent.
We also mention a metric defined in the late 1990s by C. McMullen’s (a metric
which is now called McMullen’s metric), which is a complete Kähler metric of bounded
sectional curvature. It is obtained as a kind of interpolation between the Teichmüller
and the Weil–Petersson metric. McMullen showed that this metric is equivalent to
Teichmüller’s metric and that it is Kähler-hyperbolic in the sense of Gromov.
Other complete Kähler metrics on Teichmüller space have been considered quite
recently by K. Liu, X. Sun and S.-T. Yau. These authors asked a number of interesting
questions about these metrics: computation of the L2 -cohomology groups, Ricci flow
between these metrics, index theory, representation of the mapping class group in the
middle-dimensional L2 -cohomology, and so on.
Let us note that a recent paper by S.-K.Yeung (Quasi-isometry of metrics on Teich-
müller space, 2005) reports on the state of the art concerning quasi-isometries and
equivalences between several of the metrics on Teichmüller space that we mentioned
above. The same paper adds new results on that subject.
1.3 Symplectic and Poisson structures

Teichmüller space has a natural symplectic structure ω which is induced by its Weil–
Petersson Kähler metric. This structure is invariant by the action of the mapping
class group and it descends to a symplectic structure on moduli space. The most
geometrically appealing description of this structure is certainly the one given by Scott
Wolpert in the early 1980s, in terms of the Fenchel–Nielsen coordinates. Specifically,
if α1 , . . . , αn is a set of homotopy classes of simple closed curves defining a pants
decomposition of the surface, and if τ1 , . . . , τn and l1 , . . . , ln are respectively the twist
and length parameters associated to these curves, then it is known that the parameters
(τi , li ), i = 1, . . . , n are global coordinates for Teichmüller space; they are called the
Fenchel–Nielsen (twist-length) coordinates. Wolpert proved that the Weil–Petersson

symplectic form is given by w = ni=1 dτi ∧ dli (which implies in particular that
this twist-length description of the symplectic structure is independent of the choice
of the pants decomposition). We note that in the definition of the twist parameter, one
should be careful in counting right twists positively, and in twisting at time t a total
length of t.
Wolpert obtained several other beautiful formulae concerning this symplectic struc-
ture, and we state a few of them. Let t (α) denote the Fenchel–Nielsen vector field on
Teichmüller space associated to a homotopy class of simple closed curves α on the
surface and l(α) denote the hyperbolic length function on T associated to α. Then,
Wolpert proved the following:
– A “duality formula”: ω(t (α), ·) = −dt (α).
– A “cosine formula”: w(t (α), t (β)) = t (α)l(β) = p cos θp . Here, t (α)l(β) is
the natural action of a vector field on a function, p varies over the set of intersection
points of two fixed representatives of α and β which are in minimal position and θp is
the angle of the intersection at the point p of these representatives, with appropriate
orientations.
– A formula for the Poisson bracket. First, Wolpert defines a normalized vector
field Tα = 4(sinh l(α)/2)t (α). The Tα ’s form a Lie algebra over the integers, and the
formula for the Lie bracket is [Tα , Tβ ] = p T(αp β)+ − T(αp β)− , where p varies as
above over the set of intersection points of two fixed closed curves in minimal position
representing α and β and where (αp β)+ and (αp β)− are homotopy classes of closed
curves obtained by modifying by local surgeries the union of the curves representing
α and β at the intersection point p.
We finally note that there are nice formulae for the pull-back of the symplectic
form ω on decorated Teichmüller space of a punctured surface, obtained by Penner,
which also have a geometric flavour. As we shall see below, Penner’s formulae have
been used in the quantization theory of Teichmüller spaces of punctured surfaces.
1.4 Boundary structures

Although the topology of Teichmüller space is very simple (the space is homeomorphic
to a ball), the boundary structure of this space is very rich and highly nontrivial. In fact,
there are several interesting boundaries. Every time Teichmüller space is embedded
in some function space (for instance, by Bers’ embedding as a bounded domain in a
Banach space, or by Thurston’s embedding in the space of geodesic length functions,
or by Bonahon’s embedding in the space of geodesic currents, or by its embeddings in
various spaces of representations), one can define a boundary structure for this space,
by taking the closure of the image of the embedding. In all the cases mentioned,
the boundary points have a geometric significance, as degenerate Riemann surfaces
(for instance surfaces with nodes), or as projective classes of measured foliations,
or as degenerate representations defined by group actions on R-trees, and so on. In
some cases, the closure of the image is compact, and the boundary structure defines a
compactification of Teichmüller space. In some cases, the boundary is homeomorphic
to a sphere, and in other cases it is not. It can happen that the boundary structure of
Teichmüller space defines a boundary structure to Riemann’s moduli space and vice
versa. We shall review here some of the boundary structures of Teichmüller space and
moduli space.
Thurston’s boundary. Thurston’s boundary of Teichmüller space is induced from

the embedding of this space in the space P RS+ of projective classes of nonnegative
functions on the set S of homotopy classes of essential simple closed curves on the
surface, by means of the geodesic length functions. The boundary points are projective
classes of compactly supported measured geodesic laminations, embedded in P RS+
by means of the geometric intersection functions. The mapping class group action
on Teichmüller space extends continuously to the union of Teichmüller space with is
Thurston boundary.
We note that Thurston’s boundary can also be recovered using Bonahon’s embed-
ding of Teichmüller space into the space of geodesic currents.
Bers’ boundary. Bers’ boundary of Teichmüller space is induced from the Bers
embedding of this space in the Banach space of quadratic differentials on a given
Riemann surface. Bers showed that the closure of this embedding is compact, and
this defines Bers’ compactification of Teichmüller space. Bers’ embedding depends
upon the choice of a basepoint in Teichmüller space, but any two such embeddings
are biholomorphically equivalent. However, Kerckhoff and Thurston showed that
the boundary structure that one obtains in this way depends upon the choice of the
basepoint. They proved that the mapping class group action on Teichmüller space
does not extend continuously to an action on the union of this space with its Bers
boundary. In fact, Kerckhoff and Thurston’s work (1990) gives a new description of
Bers’ boundary in the setting of geometric convergence in the space of representations
of the fundamental group of the surface in PSL(2, C). Their results on Bers’ boundary
are obtained after translating the questions into the context of quasi-Fuchsian groups.
By Kerckhoff and Thurston’s results, Bers’ boundary and Thurston’s boundary are
distinct.
The visual boundaries. Different visual boundaries of Teichmüller space can be

associated to the various metrics on this space. We recall that the visual boundary
of a complete metric space (X, d) is the space of equivalence classes of geodesic
rays, where two geodesic rays γ1 : [0, ∞[→ X and γ2 : [0, ∞[→ X are considered
equivalent if the set {d(γ1 (t), γ2 (t)), t ∈ [0, ∞[} is bounded. As a matter of fact, one
usually considers the visual boundary at a given point x in X, by taking the set of
geodesic rays emanating from x. There is a natural topology (the “cone topology”)
on the visual boundary at x. In the case where the space X is uniquely geodesic, there
is a topology on the union of X with its visual boundary at x (which makes this visual
“boundary” indeed a boundary). This topology is obtained by embedding X into the
space of all (that is, finite or infinite) geodesics emanating from x, each point in X
being identified with the geodesic that joins it to x. Of course, the first natural question
arising from such a definition is: To what extent does the visual boundary depend on
the choice of x?
The visual boundary of Teichmüller’s space equipped with Teichmüller’s metric
was studied in the late 1970s by Kerckhoff, who showed that the action of the mapping
class group on Teichmüller space does not extend continuously to the Teichmüller
visual boundary.
The Weil–Petersson visual boundary has recently been studied by J. Brock, who
worked on the analogies of this visual boundary with the boundary defined by taking
the completion of the Weil–Petersson metric, and with the Bers boundary of Teich-
müller space. Masur had already worked on the comparison between these last two
boundaries in 1982. Brock’s work starts with the known observation that some Weil–
Petersson geodesic rays have finite length, a fact related to the non-completeness
of the Weil–Petersson metric. In this work, the definition of the visual boundary is
adapted to the setting of a non-complete space. The finite-length geodesics can be
made to converge to surfaces with cusps, and this convergence is comparable to the
convergence to surfaces with nodes that appears in the works of Bers and Abikoff.
Brock also showed that the action of the mapping class group on Teichmüller space
does not extend continuously to the Weil–Petersson visual boundary. Again, this is
analogous to the result by Kerckhoff and Thurston that we mentioned above, showing
that the action of the mapping class group does not extend continuously to Bers’
boundary. Brock showed that despite these analogies, the Weil–Petersson boundary
is distinct from Bers’ boundary.
Finally, we note that the visual boundary of Thurston’s asymmetric metric is another
interesting object which so far has not been given the attention it deserves.
Other boundary structures. There are boundary structures for Riemann’s moduli
space, some of them induced from boundary structures for Teichmüller space, and
some of them belonging only to the realm of moduli space. The compactification of
moduli space by stable curves is a projective variety (studied as such by Wolpert and
others). It is a quotient space of Teichmüller space equipped with its bordification by
surfaces with nodes (sometimes called augmented Teichmüller space, with boundary
called Abikoff’s augmented boundary) by the action of the mapping class group. Note
that this boundary structure of Teichmüller space is not a compactification of this space,
but the quotient of this structure by the mapping class group is a compactification of
moduli space.
Sometimes, the same boundary structure for Teichmüller space is studied from
various points of view. When Riemann surfaces are viewed as algebraic surfaces,
the boundary structure can be studied in terms of degeneration of algebraic surfaces,
and so on. For instance, the augmented moduli space (and finite coverings of this
space), whose boundary points are surfaces with nodes, has been studied by Deligne,
Mumford, Knudsen and Mayer from the point of view of algebraic geometry, by
Abikoff from the point of view of (geometric) complex analysis, and by Marden in
relation with hyperbolic 3-manifold theory. Due to the fact that people working in
different domains use different languages, it is sometime difficult to see the relations
between the various results.
Let us finally mention that there are boundary structures for Schottky space, for gen-
eral representation spaces, and for several other spaces related to Teichmüller space.
1.5 The harmonic maps approach

The study of harmonic maps is a classical subject in analysis and geometry, and these
maps have been used in the last three decades as an important tool in the study of
Teichmüller space. The definition of a harmonic map is based on the notion of the
energy of a map. The energy of a map between two (say compact smooth) manifolds
is the integral over the domain of the square of the derivative of the map. This quantity
is regarded as a measure of the average squared stretching of the map. A map between
two Riemannian manifolds is harmonic if it is a critical point of the energy functional.
The collection of maps between the two manifolds over which the critical point is
searched for is usually taken to be the set of maps in a given homotopy class of maps.
Harmonic maps are solutions of an Euler–Lagrange equation, defined by a second-
order elliptic partial differential operator. In some sense, harmonic maps generalize
closed geodesics (which are harmonic maps with the domain manifold being the
circle). They also generalize totally geodesic maps and harmonic functions (which are
harmonic maps whose range is Euclidean space). The study of Teichmüller space using
harmonic maps started with the work of M. Gerstenhaber and H. E. Rauch in 1954.
In particular, there are harmonic map approaches to the Teichmüller metric and to the
Weil–Petersson metric. The list of people that have worked in this domain is very long;
just to give a few names, we mention E. Reich, C. Earle, J. Eells, J. Jost, A. Tromba,
M. Wolf, Y. Minsky, C. Mese, G. Daskalopoulos, R. Wentworth and L. Katzarkov.
The classes of harmonic maps that have been used in Teichmüller theory include
harmonic maps between surfaces, harmonic maps from surfaces to R-trees (as limits
of harmonic maps between surfaces; these maps can be used to describe degenerations
of elements of Teichmüller space), and, more recently, harmonic maps from surfaces
to Teichmüller space itself (equipped with various metrics).
A basic feature of the study of harmonic maps between surfaces is that the energy of
such a map only depends on the conformal class of the domain surface, and not on its
metric. In the harmonic maps approach to Teichmüller theory, the domain surface is a
Riemann surface and the target surface is generally equipped with a hyperbolic metric.
Sometimes, the target surface is equipped with a singular flat metric. For instance,
E. Kuwert (1996) proved that a Teichmüller map is the unique harmonic map in its
isotopy class when the target surface is equipped with the singular flat metric defined
by the terminal quadratic differential of the map.
Thus, one can distinguish the following three approaches to Teichmüller theory:
– the approach where the elements of Teichmüller space are homotopy classes of
conformal structures on a surface, and the optimal maps between two such structures
are those that have minimum quasiconformal constant;
– the approach where the elements of Teichmüller space are homotopy classes of
hyperbolic structures, the optimal maps between two such structures being those that
have minimum Lipschitz constant;
– the harmonic maps approach where the domain surface is equipped with a con-
formal structure and the target surface is equipped with a hyperbolic structure, and
where the harmonic maps are the optimal maps in the sense that they minimize energy.
Of course, there are very good questions concerning relations between the three
approaches. The harmonic maps approach has the advantage of using, at the same
time, the conformal and the hyperbolic point of view on Teichmüller space, and making
links between them.
M. Wolf and A. E. Fischer together with A. Tromba developed theories that de-
scribe Teichmüller space in terms of harmonic maps. The approach of Fischer and
Tromba uses techniques of analysis, whereas Wolf’s approach is more geometric. Wolf
gave a harmonic map description of the Weil–Petersson metric. He also studied har-
monic maps between a fixed Riemann surface and families of degenerating hyperbolic
surfaces. His approach includes a harmonic map description of Thurston’s boundary.
This approach involves maps from surfaces to R-trees, and it uses the work of Gromov
and Schoen on harmonic maps from surfaces to singular spaces. We can also mention
here the work of Y. Minsky, who studied the behaviour of families of harmonic maps
between surfaces when the domain surface varies along a Teichmüller geodesic, and
other related problems.
We finally mention joint work by Daskalopoulos, Katzarkov and Wentworth on
harmonic maps from surfaces to Teichmüller space equipped with the Weil–Petersson
metric and with McMullen’s metric, and joint work by Daskalopoulos, Dostoglou and
Wentworth in which they give a harmonic map interpretation of the Morgan–Shalen
compactification of the character variety of representations of the fundamental group
of a surface into SL(2, C).
By looking at the work done by the various people working in the area, one gets
the impression that all the important features of Teichmüller theory can be recovered
using the theory of harmonic maps.
1.6 The group theory

There are several classes of groups that are intimately connected to Teichmüller the-
ory. Of course, one thinks primarily of mapping class groups of surfaces and their
subgroups, but one can consider more generally the mapping class groups of general
C ∞ manifolds, and their subgroups. Another important class of groups related to
Teichmüller theory is the class of Kleinian groups, with the related Fuchsian, quasi-
Fuchsian, Schottky groups, and other groups defined by representations of the funda-
mental group of the surface. Of course, all these groups are also studied for their own
sake, that is, independently of their relation to Teichmüller space.
Mapping class groups. The groups that play the most prominent role in Teich-
müller theory are certainly the mapping class groups.4 Important special cases of
mapping class groups include the mapping class groups of punctured disks, which can
be identified with braid groups.
The study of the mapping class group has a long history that starts with that of low-
dimensional topology. There is a large number of interesting open problems related
to the algebraic structure of mapping class groups, to their action on various spaces,
and to the structure of their subgroups. An impressive list of such problems appears
in a book edited by B. Farb (American Mathematical Society, 2006). An important
subgroup of the mapping class group is the Torelli subgroup, consisting of the mapping
classes that induce the identity on the homology of the surface, but there are several
others. In Chapter 8 of this volume, L. Mosher gives an overview of the geometry
and dynamics of actions of several classes of subgroups of the mapping class group
on Teichmüller space equipped with its Thurston boundary.
As already mentioned, the mapping class group of a surface is also the isometry
group of the Teichmüller metric and of the Weil–Petersson metric on the Teichmüller
space of that surface, and it is conceivable that similar results hold for other metrics
on that space. The most enlightening study of the mapping class group is certainly
the one that Thurston made through the analysis of the action of that group on the
compactification of Teichmüller space by the space of projective classes of measured
foliations on the surface. Thurston’s topological classification of the elements of the
mapping class group, which is surveyed in Chapter 8 by Mosher, has counterparts
in terms of the metric structures of Teichmüller space. In this connection, we recall
that after Thurston’s work was completed, Bers worked out a similar classification of
mapping classes which is based on the action of the mapping class group equipped with
the Teichmüller metric. Bers obtained this classification by analyzing the minimal set
and the displacement function associated to a mapping class acting by isometries. A
similar metric classification was recently obtained by Daskalopoulos and Wentworth,
this time with respect to the Weil–Petersson metric. This classification is surveyed in
Chapter 1 of this volume.
There are several other interesting actions of the mapping class group that were
studied in the last few decades. For instance, Hatcher and Thurston proved in 1980 that
the mapping class group of a closed surface is finitely presented, by studying the action
of that group on a two-dimensional simplicial complex whose vertices are cut systems
on the surface, that is, systems of isotopy classes of disjoint closed curves whose
complement is a sphere with holes. The edges of this complex correspond to certain
moves between cut systems, called elementary moves, and the two-dimensional cells
correspond to certain cycles of moves, which are of three types: triangles, rectangles
and pentagons. E. Irmak and M. Korkmaz proved recently that the automorphism
4 There are some variations in the terminology of mapping class groups. Usually, the term extended mapping
class group of a surface designates the group of isotopy classes of diffeomorphisms of that surface. The mapping
class group is then the group of isotopy classes of orientation-preserving diffeomorphisms of an oriented surface.
There are also various subclasses, depending on whether the surface is closed or not, with or without boundary
components, whether the group fixes a distinguished point or not, and so on.
group of the Hatcher–Thurston complex of a compact oriented surface of positive

genus is isomorphic to the extended mapping class group of the surface modulo its
center. Other actions of mapping class groups which have been thoroughly studied
include the action on the complex of curves, a flag complex whose n-simplices are
collections of (n + 1) distinct isotopy classes of essential disjoint simple closed curves
on the surface. This complex was introduced in 1979 by W. Harvey, with the idea of
including it as a boundary structure of Teichmüller space. We also mention the actions
of the mapping class group on the complex of pants, on the complex of nonseparating
curves, on the complex of separating curves and the complex of domains, and there
are several others. We already mentioned that Brock found a precise relation between
the Weil–Petersson metric on Teichmüller space and the metric on the vertices of the
complex of pants, induced by the simplicial metric on the 1-skeleton. The coarse
geometries of these complexes equipped with their natural simplicial metric have also
been studied for their own sake, that is, independently of their relation to the mapping
class group. In Chapter 10 of this volume, Ursula Hamenstädt studies the action of
the mapping class group on the complex of curves. She gives a new proof of a result
of Masur and Minsky stating that this complex is hyperbolic in the sense of Gromov,
and she discusses the relation between the geometry of this complex and the geometry
of Teichmüller space.
Outer automorphism groups. An inner automorphism of a group G is an auto-
morphism of the form g → h−1 gh, where h is a fixed element of G. The outer
automorphism group of G is the quotient group of the automorphism group of G by
the action of the inner automorphism group of G.
Automorphism and outer automorphism groups of free groups are basic objects of
study in combinatorial group theory. They were already extensively investigated by
J. Nielsen and W. Magnus, in the first quarter of the twentieth century. There is a close
relation between mapping class groups of surfaces and outer automorphism groups.
For instance, in the particular case of a closed surface, the extended mapping class
group of the surface is isomorphic to the outer automorphism group of its fundamental
group. In the case of a surface with one puncture, the fundamental group is a free
group, and there is a natural map from the extended mapping class group of the surface
to the outer automorphism group of its fundamental group. This map is explicited in
Chapter 7, written by S. Morita. The study of the outer automorphism of a free group
acquired a very geometric character after the introduction in the 1980s, by M. Culler
and K. Vogtmann, of a space called Outer space, on which this group acts. There are
parallels between the action of the outer automorphism of a free group on Outer space,
and the action of the mapping class group on Teichmüller space.
Kleinian groups A Kleinian group G is a discrete group of orientation-preserving
Möbius transformation of hyperbolic 3-space H3 .5 A Fuchsian group is a Kleinian
5 The terminology here is not universally established. For instance, some authors ask that a Kleinian group
acts properly discontinuously on some non-empty open subset on the Riemann sphere seen as the boundary at
infinity of H3 , and some other authors require that the group is finitely generated.
group whose action on S 2 = ∂ H3 preserves a round disk. A quasi-Fuchsian group

is obtained from a Fuchsian group by conjugation of the action by a quasiconformal
homeomorphism of H2 . The quotient space of the action of a Kleinian group on H3
is a hyperbolic 3-orbifold (and it is a manifold if is torsion-free). Under certain
conditions, the picture of the action can be made more complete: the Kleinian group
on H3 ∪ S 2 gives rise to a hyperbolic 3-manifold with nonempty boundary consisting
of a finite number of surfaces of finite type carrying a canonical conformal structure,
and the deformation theory of the hyperbolic 3-manifold can be studied through the
deformation theory of its boundary surfaces. More precisely, given a Kleinian group G,
it has an induced action on the boundary at infinity ∂ H3 = S 2 . This action decomposes
S 2 into the union of two subsets, the ordinary set and the limit set. The ordinary set
(or domain of discontinuity) ⊂ S 2 is the largest open subset of the sphere on
which the group G acts properly discontinuously. The limit set of the action is the
set of accumulation points in S 2 of the orbit of any point in H3 . From the point of
view of the theory of hyperbolic 3-manifolds with boundary, one considers the action
of G on the union H3 ∪ . Ahlfors’ finiteness theorem ensures that under some mild
hypotheses, the quotient of H3 ∪ by this action is a 3-manifold with boundary, with
the boundary consisting of a finite union of connected Riemann surfaces of finite type.
The 3-manifold is equipped with a quotient hyperbolic structure induced from that
of H3 , and each boundary component is equipped with a complex structure (in fact,
with a complex projective structure) induced from that of S 2 . There is an interplay
between the deformation theory of the hyperbolic 3-manifold and the union of the
Teichmüller spaces of the boundary components. The classical approach to this study
consists in applying Teichmüller’s quasiconformal deformation theory in succession
to the boundary components. For instance, in the case where the Kleinian group
is a quasi-Fuchsian group, a theorem of Bers says that the hyperbolic structure of
the 3-manifold is completely determined by the conformal structure of the boundary,
which as a matter of fact consists of two conformal structures on the same topological
surface. Thus, quasi-Fuchsian space is parametrized by the product of two copies of
Teichmüller space.
We also note that the relation of Teichmüller theory with Kleinian groups involves
the deformation theory of the projective structure of the boundary, and that the gen-
eral deformation theory of Kleinian groups can be studied from the point of view of
representation theory.
1.7 Representation theory

By the uniformization theorem, every Riemann surface of negative Euler characteristic
can be defined as a complex structure induced by some hyperbolic metric. In the case
of a surface without boundary, the hyperbolic metric can be regarded, by lifting the
hyperbolic structure to the universal cover, as a discrete group of orientation-preserving
isometries of hyperbolic 2-space H2 , and such a group is well-defined up to conjugacy.
From this observation, Teichmüller space can be studied as a set of conjugacy classes
of representations of the fundamental group of the surface into the group PSL(2, R)
of orientation-preserving isometries of H2 .
More generally, one can study the moduli spaces of representations of the funda-
mental group of a surface S in a Lie group G, that is, the space of homomorphisms of
π1 (S) into G up to conjugation. The classical case is G = PSL(2, R), as described
above. The Teichmüller space of a closed surface S of genus ≥ 2 is a connected
component of this moduli space. This component consists of the conjugacy classes
of discrete and faithful representations. W. Goldman showed that in this case, the
number of connected components of the moduli space of representations is equal to
4g − 3.
There are other important cases for the Lie group G. These include the case where
G = PSL(2, C) (which is the setting of Kleinian representations, which includes
several subsettings, like quasi-Fuchsian representations, Schottky representations, and
so on). There is also the case where G is a split semisimple real group, whose study is
the subject of “higher Teichmüller theory”. It is worth saying a few words about this
theory, since it has recently attracted much interest.
It is generally considered that higher Teichmüller theory was initiated by the work
of Nigel Hitchin, who proved in the early 1990s that the space of conjugacy classes of
discrete faithful representations in the space of conjugacy classes of all representations
of the fundamental group of a compact oriented surface of genus ≥ 2 in a semisimple
split real Lie group (namely SL(n, R) or Sp(2n, R)) is a contractible component,
a result which is regarded as a generalization of the fact that Teichmüller space is
contractible. As mentioned above, in the case n = 2, this component is the Teich-
müller space of the surface. In the case n = 3, the corresponding component is, by
a result of Goldman and Choi, the moduli space of convex real projective structures
on the surface. There are several good questions in higher Teichmüller theory, some
of them independent of classical Teichmüller theory; some concern the action of the
mapping class group on components other than the one discovered by Hitchin. V. Fock
and A. Gontcharov recently studied “positive” representations of the fundamental
group of a closed surface into a split semisimple algebraic group G with trivial center.
They proved that these representations are faithful and discrete and that the moduli
space of such representations is an open cell in the space of all representations. For
G = SL(2, R), this space is the classical Teichmüller space. Higher Teichmüller
theory has also been developed by F. Labourie. M. Burger, A. Iozzi and A. Weinhard
recently undertook a study of representations of the fundamental group of the surface
into semisimple Lie groups of Hermitian type. Let us note that one can also study
representations of fundamental groups of surfaces in compact Lie groups like SU(2);
in this case, the space of representations is compact.
1.8 Dessins d’enfants

A natural question in the algebro-geometric point of view on Teichmüller space is to
understand the Riemann surfaces that can be defined by polynomials with coefficients
in number fields. For instance, one would like to understand how the collection of such
surfaces is embedded in Riemann’s moduli space, or how to recognize the algebraic
and, more specially, the arithmetic surfaces. There are also interesting questions about
the automorphisms of such surfaces, e.g. what are the stabilizers of these surfaces in
the mapping class group, and so on.
Grothendieck’s introduction of his theory of dessins d’enfants was a major step in
this direction. This theory makes connections between Teichmüller theory, complex
algebraic geometry, arithmetic geometry and the study of a class of combinatorial
objects on the surface.
We now briefly discuss this theory, and the related space of Belyi affine curves.
This space is an analytic variety equipped with a natural action of the absolute Galois
group (also called the universal Galois group Gal(Q)) which, by definition, is the auto-
morphism group of the field Q of algebraic numbers. We note that the absolute Galois
group Gal(Q) is considered as one of the most mysterious groups in mathematics, and
for this reason, producing new actions of this group is always interesting.
Grothendieck’s approach to Teichmüller theory is based on the idea that the com-
binatorial techniques of hyperbolic geometry (starting with surface decompositions
into hyperbolic pairs of pants) have a parallel in algebraic geometry over Q. In this ap-
proach, dessins d’enfants play a prominent role. A dessin d’enfant (“child’s drawing”)
is a connected graph embedded in a compact topological surface, whose complemen-
tary components are simply connected and whose edges are colored black and white
such that no edge has its two vertices of the same color.
This theory is outlined in Grothendieck’s Esquisse d’un programme, a manuscript
which has been widely circulated in the form of mimeographed notes since it was writ-
ten in 1984 and which finally appeared in print (together with an English translation)
in 1997.
The correspondence between dessins d’enfants and algebraic curves over Q is a
consequence of a theorem of G. Belyi. This theorem states that any algebraic curve
over Q can be realized as a ramified finite covering of the Riemann sphere P1 (C),
with ramification locus contained in the set {0, 1, ∞}. In this way, an algebraic curve
over Q defines a dessin d’enfant, obtained as the preimage of [0, 1] ⊂ P1 (C) by the
covering map. The color of a vertex is specified according to whether the image of
that vertex is 0 or 1. The relation with algebraic curves over C stems then from the
fact that a complex algebraic curve X can be represented by an algebraic curve over
Q if and only if there exists a non-constant holomorphic map f : X → P1 (C) whose
critical values are in Q. The surprising contribution of Belyi was to reduce the number
of critical values to three. This result was first announced at the Helsinki ICM in 1978.
It had been conjectured by Grothendieck, who later on declared that he was amazed
by the simplicity of Belyi’s proof of this result. The converse result, that is, the fact
that to a dessin d’enfant one can associate a ramified covering over P1 (C) is due to
Grothendieck himself (who proved it using Belyi’s work).
Grothendieck’s ideas on Teichmüller space have been essentially exploited by
algebraic geometers, because the original language of Grothendieck is that of algebraic
geometry, although Grothendieck claimed on several occasions (in particular in his

Esquisse d’un programme) that he was amazed by Thurston’s approach to Teichmüller
space, and that he would have liked to see his theory expressed in a similar geometric
language.
Volume II of this Handbook contains a chapter on the theory of dessins d’enfants,
written by F. Herrlich and G. Schmithüsen.
Links between Grothendieck’s theory and the geometric Teichmüller theory will
certainly become more apparent in the near future.
1.9 Physics and quantum theory
E. Verlinde and H. Verlinde conjectured in 1990 that the space of conformal blocks
of Liouville conformal field theory (which is a version of a two-dimensional Einstein
quantum gravity) can be obtained by the quantization of the Teichmüller space of
the corresponding Riemann surface. This was motivated by the developments of
the quantization of Chern–Simons theory. The quantization theory of Teichmüller
space makes the Verlindes idea precise and it provides an explicit Hilbert space and
observable algebra for 2-dimensional quantum gravity.
More explicitly, the quantization of Teichmüller space Tg,p produces a one-para-
meter family of noncommutative ∗-algebras, with an action of the mapping class group
of the surface by ∗-algebra automorphisms. These algebras are parametrized by a
quantization parameter h̄, having the property that when h̄ = 0 the algebra coincides
with the original commutative algebra of functions on Tg,p while the first derivative of
the commutator with respect to h̄ at zero gives the Weil–Petersson Poisson structure
on Tg,p .
Quantization theories of Teichmüller space were developed independently during
the 1990s in joint work by L. Chekhov and V. Fock, and by R. Kashaev. The two
theories are essentially equivalent up to technical details.
The quantization theory developed by Chekhov and Fock considers the Teich-
müller space of a surface with holes as a (degenerate) Poisson manifold and uses
Thurston’s shear coordinates on that surface, whereas the quantization theory devel-
oped by Kashaev considers the decorated Teichmüller space and uses Penner’s λ-length
coordinates.
The quantization procedures produce (for h̄ = 0) semi-simple infinite-dimensional
∗-algebras, and unitary projective representations of a subgroup of the mapping class
group in the ∗-representation spaces of these algebras. Technically, this representa-
tion acts on a Hilbert space of functions on Rn by certain compositions of Fourier
transformations, multiplications by quadratic forms and multiplication by quantum
dilogarithms.
Let us also mention that Kashaev recently developed a quantization theory of the
moduli space of flat PSL(2, R)-connections on a punctured surface with parabolic
holonomy around the punctures. This moduli space is closely related to the classical
Teichmüller space of the surface, and in fact, two of its connected components are
isomorphic to Teichmüller space.
J. Teschner’s work proves the Verlindes’ conjecture relating the quantization the-
ories developed by Chekhov–Fock and by Kashaev to the Liouville conformal field
theory.
Furthermore, the quantization of Teichmüller space may lead to new invariants of
hyperbolic three-manifolds. A famous example in this respect is the volume conjecture
by R. Kashaev, H. Murakami and J. Murakami which expresses the volume of any
hyperbolic knot as a certain limit of a specific value of the colored Jones invariant
associated with the quantum group SU(2)q , which in turn can be computed using
Kashaev’s dilogarithmic quantization theory.
Thus, the quantization theories establish new links between Teichmüller theory,
algebra, algebraic geometry, representation theory and mathematical physics.
The present volume contains chapters written by the various people involved in
the quantization theory, namely, Chekhov, Fock, Goncharov, Kashaev, Penner and
Teschner.
We mention that S. Baseilhac and R. Benedetti worked out a finite-dimensional
combinatorial version of the quantization of Teichmüller space, with the aim of con-
structing new invariants for 3-manifolds, and with a view towards relating these results
to the volume conjecture (with the hope of proving this conjecture). Their techniques
are partly inspired by those used by V. Turaev in his work on Chern–Simons theory.
Volume II of this Handbook will contain a chapter written by Baseilhac and Benedetti
on that subject.
Let us also note that F. Bonahon and X. Liu worked out a finite-dimensional version
of the quantization theory of Chekhov and Fock.
2 An overview of the content of this volume

The overview of the various sections of this Handbook that I will give now is also
intended to give an idea of the large variety of ideas that Teichmüller theory involves.
2.1 The metric and the analytic theory, 1

Chapter 1, written by Georgios Daskalopoulos and Richard Wentworth, gives an
overview of several important aspects of Teichmüller theory from the point of view
of the theory of harmonic maps. The subjects that are treated in this chapter in-
clude the geometry of the Weil–Petersson metric (geodesics, curvature, isometries,
completion, length functionals, convexity and superrigidity) and the study of group
actions on R-trees. In particular, the authors describe a harmonic map construction
of the Teichmüller map, a harmonic map interpretation of the theorem of Hubbard
and Masur on the existence and uniqueness of a quadratic differential with a given
equivalence class of horizontal foliation, and a harmonic map interpretation of the

Thurston–Morgan–Shalen compactification of Teichmüller space. The chapter also
contains an introduction to several basic topics in Teichmüller theory, including the
uniformization theorem, quasiconformal maps, Beltrami differentials, the solution to
the Beltrami equation, several proofs (including one that is based on harmonic maps)
of Teichmüller’s theorem stating that Teichmüller space is a cell, Nielsen realization,
a Higgs bundle interpretation of Teichmüller space in relation to flat SL(2, C) connec-
tions, and holomorphic convexity. The authors also discuss similarities and differences
between the mapping class group and arithmetic lattices.
To make the exposition more accessible, the authors also give an introduction to the
theory of harmonic maps, with an emphasis on harmonic maps between surfaces, be-
tween surfaces and trees and between surfaces and Teichmüller space. The exposition
includes some classical results by Gerstenhaber and Rauch, the results of Eells and
Sampson on the existence of harmonic maps in a given homotopy class of maps in the
case where the range is compact and has nonpositive sectional curvature, Hartmann’s
uniqueness result in the case where the curvature is negative, and an overview of the
relatively recent theory of Gromov and Schoen on harmonic maps between singular
spaces and of the recent work by C. Mese on the proof of a conjecture made in 1954
by Gerstenhaber and Rauch on realizing extremal energy maps by Teichmüller maps.
Chapter 2 by Guillaume Théret and myself has two parts. The first part concerns
Teichmüller’s metric on Teichmüller space, and the second part concerns Thurston’s
asymmetric metric on this space. Teichmüller’s metric regards Teichmüller space from
the point of view of conformal geometry, whereas Thurston’s asymmetric metric is
based on considerations on Teichmüller space from the point of view of hyperbolic
geometry. The two metrics are Finsler metrics, and they are studied in the same chapter
with the aim of drawing parallels and differences between them. This chapter contains
basic facts about the two metrics and about the asymptotic geometry they induce on
Teichmüller space. The main results that are presented concern the behaviour of
geodesic rays for the two metrics, in particular the convergence or non-convergence
of certain classes of geodesic rays to points in Thurston’s boundary. This includes
results on the limiting behaviour of stretch lines, which are geodesics for Thurston’s
asymmetric metric, and on the limiting behaviour of anti-stretch lines, which are stretch
lines traversed in the negative direction, and which in general are not geodesic lines for
Thurston’s asymmetric metric. The chapter also contains a review of a parametrization
of Teichmüller space by a space of measured geodesic laminations that was introduced
by Thurston, and which Thurston calls “cataclysm coordinates”.
The results concerning Teichmüller’s metric which are surveyed in Chapter 2 are
classical, whereas some of the results concerning Thurston’s asymmetric metric are
new. We also discuss some open problems related to Thurston’s asymmetric metric.
Chapter 3, written by Robert Penner, contains an exposition of a generalized notion

of a decorated hyperbolic structure.
A decoration of a hyperbolic surface with cusps is the choice of a closed horocycle

around each cusp. The space of homotopy classes of decorated hyperbolic structures
is a fibre bundle over Teichmüller space, called decorated Teichmüller space. This
space was introduced by R. Penner in 1987, and the idea turned out to be extremely
fruitful. A particularly useful set of parameters that Penner defined are the λ-length
coordinates of this space. These are defined as signed distances between distinguished
horocycles, measured on the edges of an ideal triangulation of the surface. There is a
very simple and useful description of the pull-back of the Weil–Petersson symplectic
form on decorated Teichmüller space in terms of λ-length coordinates. These coor-
dinates, together with their transformation laws (the so-called “Ptolemy equations”),
were used successfully in the study of the combinatorial structure of Teichmüller
space. For instance, Penner used them to construct a mapping class group invariant
cell-decomposition of decorated Teichmüller space and to compute Weil–Petersson
volumes of moduli spaces. We already mentioned that the decorated theory, together
with the λ-length coordinates, have been used as essential tools in the quantization
theory of Teichmüller space, in particular in the work of R. Kashaev.
In Chapter 3, λ-length coordinates are reviewed and are generalized in two direc-
tions, namely, in the setting of homeomorphisms of the circle, and in what the author
calls the decorated Teichmüller theory of the punctured solenoid. Let us say a few
words about these two theories.
Universal Teichmüller theory first appeared in the work of Ahlfors and Bers. This
theory can be formulated in terms of quasisymmetric homeomorphisms of the unit
circle. These mappings arise as homeomorphisms which are boundary values of
quasiconformal maps of the hyperbolic disk. In the context of universal Teichmüller
theory, quasisymmetric homeomorphisms of the unit circle are used to parametrize
hyperbolic structures on the unit disk relative to the boundary, which in particular
parametrize Teichmüller spaces of all finite type surfaces. λ-length coordinates are
used to parametrize the space of cosets of the subgroup of Möbius transformations in
the group of orientation-preserving homeomorphisms of the circle, which is identified
with a suitable space of tessellations of the hyperbolic disk. Some elements of universal
Teichmüller theory are surveyed in Chapter 3 of this volume.
The (hyperbolic) solenoid H was introduced in the 1990s by D. Sullivan as an
inverse limit of the space of all branched covering of a closed Riemann surface. More
precisely, let (S, x) be a pointed surface of genus > 1. If πi : (Si , xi ) → (S, x) and
πj : (Sj , xj ) → (S, x) are two pointed coverings of (S, x), then we write πi ≤ πj if the
covering πj factors through the covering πj,i : (Sj , xj ) → (Si , xi ) with πj = πi ◦ πj,i .
Equipped with the relation ≤, the set of all coverings of (S, x) is inverse directed, and
the solenoid H is the inverse limit of this directed set. This space H is equipped with a
topology, viz. the subspace topology induced from the product topology on the infinite
product of all closed surfaces in the finite coverings of (S, x). With this topology, the
local structure of H is that of a plane times a Cantor set. The solenoid H can also
be equipped with a complex structures, and in fact, like surfaces of negative Euler
characteristic, it can be equipped with a family of inequivalent complex structures.
The space of complex structures on the solenoid was studied by Sullivan and others.
There is a Teichmüller space T (H) of H, which can be defined as a certain closure
of the stack of the Teichmüller spaces of all closed surfaces of genus > 1. This
space admits a complete metric, in analogy to Teichmüller’s metric. The mapping
class group of the solenoid H is the set of isotopy classes of quasiconformal maps
h : H → H, and there is a natural action of this group on T (H ). Volume II of this
Handbook contains a chapter on the solenoid, written by D. Šarić.
In Chapter 3, Penner discusses joint work with Šarić on the punctured solenoid.
In analogy to Sullivan’s definition, the punctured hyperbolic solenoid is defined in
this setting as the inverse limit of all finite unbranched pointed covers, the branching
being permitted only over the punctures. The Teichmüller space of the punctured
solenoid is a separable Banach space admitting a complete Teichmüller metric. This
is a universal object in the sense that one can canonically embed in this space every
classical Teichmüller space over a hyperbolic punctured surface. There are again
λ-length coordinates on an appropriate decorated Teichmüller space of the punctured
solenoid, and the structure of decorated Teichmüller space for finite type surfaces
survives mutatis mutandis for this punctured solenoid.
Chapter 4 by Jean-Pierre Otal contains an exposition of the quasiconformal defor-
mation theory of Riemann surfaces, of Bers’ construction of the complex structure of
Teichmüller space and of the theory of geodesic currents and of Hölder distributions.
The space of geodesic currents is a subspace of the space of Hölder distributions. Both
spaces were introduced by F. Bonahon in the setting of Fuchsian groups of finite co-
area. According to Bonahon’s definition, a geodesic current is a positive locally finite
measure on the space of geodesic lines in the universal covering of the surface, which
is invariant by the action of the group of covering transformations on that space. We
recall that if S 1 is the boundary at infinity of the unit disk, considered as the universal
covering of the surface, then the space of geodesic lines in the universal covering of
the surface is naturally identified with the set (S 1 × S 1 \ )/(Z/2), where is the
diagonal set of S 1 × S 1 and where Z/2 acts by exchanging coordinates. The space of
geodesic currents is a complete uniform space. The terminology of currents was used
by Bonahon after Sullivan who introduced it in the study of dynamical systems, in
analogy to de Rham’s currents. Bonahon defined a topological embedding of Teich-
müller space into the space of geodesic currents, which he called the Liouville map.
This embedding gives a new and unifying point of view on Teichmüller space, from
which one can recover, for instance, Thurston’s compactification by adjoining to this
space the space of asymptotic rays, which are identified with measured geodesic lami-
nations on the surface. The space of Hölder distributions is the dual space of the space
of Hölder continuous functions with compact support. The definition of the space
of Hölder distributions was extended to the case of an arbitrary Fuchsian group by
Šarić, who showed that the Liouville map, generalized to that setting, is also a topo-
logical embedding. As in Bonahon’s setting, Šarić defined a Thurston-type boundary
by adding to the image of the Liouville map the set of asymptotic rays, which can be
identified with bounded measured laminations on the surface, but here, the closure of
the image is in general not compact. Chapter 4 contains an exposition of this work of
Šarić. The last part of this chapter contains new results by Otal on the analyticity of
the Liouville map.
In Chapter 5, William Harvey considers the complex theory of moduli space of

complex algebraic curves, and its relation with algebraic geometry and arithmetic
geometry. This study mainly involves the following two interrelated topics:
— A discussion of Grothendieck’s theory of dessins d’enfants and of Belyi’s com-
plex affine algebraic curves, with the action of the absolute Galois group of algebraic
numbers Gal(Q) on these objects. One of the advantages of Harvey’s review of
Grothendieck’s theory is that it is done by means of the classical tools of the theory
of Teichmüller space, that is, the techniques introduced by Ahlfors and Bers.
— A study of Teichmüller disks in Teichmüller space, of their stabilizer groups
and of their images in moduli space. In particular, the author addresses the question
of the existence and the description of Teichmüller disks with large stabilizer groups,
and he presents a construction of such disks that arises from hyperbolic tessellations
of surfaces, defined by Hecke triangle groups and their subgroups.
We recall that a Teichmüller disk is a holomorphic and isometric embedding of the
unit disk in the complex plane equipped with its hyperbolic metric into Teichmüller
space equipped with Teichmüller’s metric. Teichmüller disks were already studied
by Teichmüller himself, and their study was revived by works of Thurston, Veech,
McMullen and others.
As a matter of fact, Harvey describes several relations between Grothendieck’s
theory and the study of Teichmüller disks. The first relation is through a classical con-
struction, due to Thurston, of pseudo-Anosov maps as products of Dehn twists along
simple closed curves whose union fills the surface. This construction provides both a
dessin d’enfant and a Teichmüller disk. Harvey then describes a general procedure for
constructing Teichmüller curves in moduli space and he gives a characterization of all
the curves of a given genus that are definable over Q. He presents a theorem stating
that for any point in moduli space that corresponds to a curve defined over a number
field (that is, a finite extension of Q), there exists a Teichmüller disk of an arithmetic
nature passing through it.
Chpater 5 also contains background material on the Ahlfors–Bers deformation
theory of complex structures on surfaces as solutions of Beltrami partial differential
equations, and an overview of some elements of the theory of universal Teichmüller
space. The author also mentions works by C. McMullen, by P. Hubert, S. Lelièvre
and T. A. Schmidt on Veech disk, and by P. B. Cohen and J. Wolfart on deformations
of algebraic curves.
In Chapter 6, Frank Herrlich and Gabriela Schmithüsen study, among others, the
following three questions:
— When is the image of a Teichmüller disk in moduli space an algebraic curve?
— What is the limiting behaviour of such a disk at the boundary of Teichmüller
space and in Schottky space and its compactification?
— What are the stabilizer groups of Teichmüller disks in the mapping class group,
and what is their relation with discrete subgroups of PSL(2, R) such as the Veech
groups?
The authors present three different approaches to Teichmüller disks. First, such a
disk can be considered as a complex geodesic in Teichmüller space, and in particular
as a collection of (real) geodesic rays. The second approach starts with a (singular) flat
structure that is induced on a Riemann surface by a holomorphic quadratic differential;
a Teichmüller disk is then obtained by an affine deformation of this structure. Finally,
a Teichmüller disk can also be obtained by varying the Beltrami differential of the
quasiconformal mapping corresponding to a point in Teichmüller space. The authors
show that all three approaches lead to the same object.
If the image of a Teichmüller disk in moduli space is an algebraic curve, it is called
a Teichmüller curve. Whether this happens or not can be seen from the Veech group,
which is introduced in this chapter geometrically as a Fuchsian group and equivalently
as a subgroup of the mapping class group.
Chapter 6 also contains a report on W. Abikoff’s construction of the augmented
Teichmüller space and on an alternative point of view on that space, due to V. Braun-
gardt, using a general complex-analytic device. As a matter of fact, Braungardt’s
construction is done in the category of locally complex ringed spaces.
A central aspect of Chapter 6 is the investigation of limit points of Teichmüller
curves in the Abikoff–Braungardt boundary. The main result is that these points can
be described as endpoints of Strebel rays corresponding to parabolic elements in the
Veech group. The techniques build on Masur’s analysis of Strebel rays.
Finally, a partial compactification of Schottky space is introduced in two ways. The
relation between Teichmüller space, Schottky space, moduli space, their boundaries
and the groups acting on them are worked out. Using the boundary of Schottky space,
properties of the image of a Teichmüller disk in Schottky space are obtained.
2.2 The group theory, 1

In Chapter 7, Shigeyuki Morita studies several classes of groups which either general-
ize mapping class groups of surfaces, or are subgroups of mapping class groups. The
classes of groups that are involved in this study include the following:
— Diffeotopy groups (that is, groups of isotopy classes of diffeomorphisms) of
arbitrary smooth manifolds. When the manifold is a surface, then the diffeotopy group
is the extended mapping class of this surface (that is, the group of isotopy classes of
diffeomorphisms which do not necessarily preserve an orientation).
— Groups of isotopy classes of diffeomorphisms of smooth manifolds that preserve
a volume form, respectively a symplectic form.
— The Torelli group, that is the subgroup of the mapping class group consisting
of isotopy classes which act trivially on homology. Morita discusses the action of the
mapping class group on the homology of the surface, and he presents properties of the
Torelli group, including the fundamental work of D. Johnson on that group.
— The outer automorphism group of a free group. The exposition starts with a
description of work done by Jacob Nielsen on this outer automorphism group. The
outer automorphism group of a free group acts on a space introduced in the 1980s
by M. Culler and K. Vogtmann, called Outer space. This space, with respect to its
action by the outer automorphism group, is considered to play a role analogous to that
of Teichmüller space, with respect to its action by the mapping class group. Morita
explains important parallels between the two theories.
In this chapter, Morita presents important aspects of the algebraic structure of all
these groups, in particular on the interplay between their cohomology groups, and he
discusses some related open problems.
Chapter 8 by Lee Mosher expounds on the dynamics of subgroups of the mapping

class group from the point of view of their action on Teichmüller space compactified
by its Thurston boundary. The author presents important dynamical, geometric and
algebraic properties of this action, with the aim of highlighting analogies between these
actions and the actions of discrete groups of isometries of hyperbolic 3-space H3 on
the union of that space with its boundary at infinity. From the purely dynamical point
of view, the analogy starts with the notions of limit set and of domain of discontinuity.
From the geometric point of view, the analogy is based on the notions of convex–
compact subgroups and of Schottky subgroups. Teichmüller disks intervene in this
study as isometric images of hyperbolic planes in Teichmüller space, their stabilizers
being analogous to Fuchsian subgroups which in the classical case are isometry groups
of H3 that stabilize isometrically embedded 2-dimensional hyperbolic spaces. From
the algebraic point of view, the author presents the Tits alternative for subgroups of the
mapping class group (proved by J. McCarthy) and the Leininger–Ried combination
theorem which provides a method for building closed surface subgroups of mapping
class groups. This last theorem is reminiscent of Maskit’s combination theorem that
is used in the construction of discrete subgroups of isometries of hyperbolic spaces
by combining simplest subgroups. The chapter also contains a survey of Thurston’s
classification theorem of mapping classes.
Chapter 9 by Albert Marden concerns the deformation theory of Kleinian groups,

specifically the interplay between the deformation space of a hyperbolic 3-manifold
and the Teichmüller space of its boundary surfaces. In his report, Marden recalls
several deep results related to this theory, including Thurston’s hyperbolization the-
orem, Mostow rigidity theorem and a theorem of Sullivan on the relation between
the conformal structure of a simply connected region in the Riemann sphere and the
geometry of a certain surface embedded in H3 . Sullivan’s result says that under some
mild conditions, the boundary of the convex hull in H3 of the complement of an open
subset U of the Riemann sphere (seen as the boundary at infinity of H3 ) is quasi-
conformally equivalent to U , with a quasiconformality constant being a universal
constant. Chapter 9 includes a report on Bers’ embedding of Teichmüller space into
the Banach space of quadratic differentials, which in Kleinian groups theory leads to
the simultaneous uniformization theorem. It also includes a report on the augmented
space of representation which, when quotiented by the mapping class group action,
gives a compactification of moduli space. The discussion on the representation space
includes a review of the various notions of convergence in the study of the represen-
tation variety of the 3-manifold group in PSL(2, C).
Chapter 9 also contains a review of earthquakes and of their generalization to
complex earthquakes, and the recent activity on that subject that stems from Sullivan’s
theorem and that has been carried out by D. Epstein, A. Marden and V. Markovic.
Chapter 10, by Ursula Hamenstädt, concerns the geometry of the complex of
curves of a surface and its relation to the geometry of Teichmüller space. The natural
action on this complex by the mapping class group of the surface has been thoroughly
investigated in the last two decades by several people. The bulk of Chapter 10 is about
the coarse geometry of this complex. Hamenstädt gives a new proof of a theorem due
to H. Masur and Y. Minsky stating that the complex of curves is Gromov hyperbolic,
and she reports on a result due to E. Klarreich describing the Gromov boundary of
this complex. She describes a map from Teichmüller space to the complex of curves
which is coarsely equivariant with respect to the action of the mapping class group on
both spaces, with the property that the image under this map of a Teichmüller geodesic
is a quasi-geodesic in the complex of curves.
Chapter 10 also contains an exposition of some basic facts on surface topology,
namely, a description of the Hausdorff topology of the space of (not necessarily mea-
sured) laminations and of train track coordinates for this space.
2.3 Surfaces with singularities and discrete Riemann surfaces

Singular metrics on surfaces of constant curvature with conical singularities have been
studied by several authors, including A. D. Alexandrov, W. P. Thurston, W. A. Veech,
I. Rivin and B. Bowditch. Such structures naturally appear in the study of cellullar
decompositions of Teichmüller space and of moduli space, but also in other geometric
contexts such as the study of patterns of circles. The existence and uniqueness results
of M. Troyanov and his work on the analytic theory of spaces of singular flat metrics
are often quoted as basic references on that subject.
In Chapter 11 of this volume, Marc Troyanov describes classical and new results
on the deformation theory of flat metrics with cone singularities on surfaces. The
chapter starts with a general introduction to the deformation theory of geometric
structures on compact manifolds with boundary, based on the notions of developing
map and holonomy homomorphism. The chapter also contains an introduction to the
representation theory of a finitely presented group into a real algebraic group. This
theory is applied to the special case of flat metrics with cone singularities on surfaces,
in which the representation group is the group SE(2) of rigid motions of the Euclidean
plane. The developing map and holonomy homomorphism of such structures are
used to define an explicit structure on Teichmüller space that makes it a real algebraic
variety, and a corresponding orbifold structure on the quotient moduli space. In the
special case where the surface has genus zero, this structure has already been studied
by Deligne and Mostow, using techniques of algebraic geometry, and by Thurston.
The techniques used by Troyanov are close to those of Thurston.
In Chapter 12, Charalampos Charitos and Ioannis Papadoperakis consider gen-

eralizations of hyperbolic structures (that is, metrics of constant curvature −1) on
surfaces. This chapter is divided into two distinct parts that concern respectively
hyperbolic structures with cone singularities on surfaces, and hyperbolic structures on
simplicial 2-complexes.
The hyperbolic structures with cone singularities that are considered here have their
cone angles ≥ 2π . With this condition, the surfaces satisfy the CAT(−1) condition
(the so-called Cartan–Alexandrov–Toponogov negative curvature condition), and the
techniques of the theory of CAT(−1)-spaces can therefore be used in this study. The
authors work out in detail coordinates for the moduli space of these structures in the
case where the surface is a pair of pants with a unique cone point. In this case, the
parameter space is homeomorphic to R6 .
Let us note that a pair of pants equipped with a hyperbolic metric with a unique
cone point is the simplest surface that one can study in this context. In some sense, this
case is a building block for the general theory of hyperbolic structures with conical
singular points and with cone angles ≥ 2π. Indeed, it is conceivable that the space of
hyperbolic metrics with cone singularities can be parametrized by decomposing the
surface into pairs of pants, in analogy to the case of non-singular hyperbolic structures,
except that in this singular setting one has to deal with several kinds of degeneration
of pairs of pants.
The other class of singular hyperbolic structures studied in Chapter 12 is the class
of complete length metrics on 2-dimensional simplicial complexes in which each
2-simplex with its vertices deleted is isometric to a hyperbolic ideal triangle. Such a
metric space is called a 2-dimensional ideal simplicial complex, and it also satisfies the
CAT(−1)-condition. The authors describe parameters for the moduli space of ideal
simplicial complexes, using lengths of closed geodesics, in analogy to Thurston’s
parametrization of the Teichmüller space of a surface by embedding it in the space
RS+ of functions on the set S of homotopy classes of essential simple closed curves on
the surface.
Chapter 13, by Christian Mercat, is an introduction to the theory of discrete Rie-

mann surfaces. Here, a discrete Riemann surface is a topological surface equipped
with a cell-decomposition whose faces are quadrilaterals, with positive weights as-
signed to the diagonals in such a way that for each quadrilateral the product of the
weights of the two diagonals is equal to one. A discrete holomorphic map on a discrete
Riemann surface is a function on the set of vertices satisfying a discrete version of
the Cauchy–Riemann equations. These definitions lead to the notions of differential
forms, holomorphic forms, wedge products, Dirichlet energy, a Hodge-star operator,
harmonicity and period matrices for discrete Riemann surfaces. Using these no-
tions, there are discrete analogs of a number of classical theorems in Riemann surface
theory, e.g. Hodge decomposition theorems, existence theorems for holomorphic

forms with prescribed holonomies, and several approximation theorems. The results
that are presented in Chapter 13 include a convergence theorem for the period matrices
as a smooth Riemann surface is approximated by continuous ones, and results con-
necting a discrete version of the exponential function to the Bäcklund (or Darboux)
transform for discrete holomorphic maps. The author also mentions a relation between
discrete analytic functions and circle patterns.
2.4 The quantum theory of Teichmüller space, 1

In Chapter 14, Leonid Chekhov and Robert Penner present details of the Chekhov–
Fock version of the quantization theory of the Teichmüller space of a punctured
surface. They furthermore present results towards the quantization of its Thurston
boundary, whose elements are projective measured foliations of compact support, and
they show that the required operatorial limits exist weakly in the special case of the
once-punctured torus. They introduce two classes of quantum operators, defined on
the set of homotopy classes of essential simple closed curves on the surface, namely,
quantum versions of the geodesic length operators associated to hyperbolic structures
and quantum versions of the intersection function operators associated to measured
foliations of compact support. Relating the quantization of Teichmüller space to the
quantization of Thurston’s boundary is realized by showing that if a sequence of hy-
perbolic structures (gn ) converges to a projective class of a measured foliation λ,
then the sequence of quantum operators associated to (gn ) converges weakly to the
quantum operator associated to λ. This chapter also contains some basic material on
Teichmüller theory, including an introduction to measured foliations, to train tracks
and to shear coordinates associated to ideal triangulations.
Chapter 15 by Vladimir Fock and Alexander Goncharov is a report on mostly
introductory material related to the geometry of surfaces with boundary and with dis-
tinguished points on their boundaries (which the authors call ciliated surfaces), and of
the corresponding spaces of measured laminations. The authors give detailed descrip-
tions of coordinates for these spaces, with formulae for the Poisson and the symplectic
structures on Teichmüller and lamination spaces, of pairing between lamination and
Teichmüller spaces, and of the action of the mapping class group in these coordinates.
Their exposition also contains an interpretation of lamination spaces as tropical limits
of Teichmüller spaces.
Chapter 16 by Jörg Teschner contains a construction of a modular functor out of
quantum Teichmüller space. A modular functor is a functor from the category of
Riemann surfaces with isotopy classes of embeddings as morphisms, to the category
of vector spaces with linear maps as morphisms. In this work, the vector spaces
are infinite-dimensional. Modular functors are natural generalizations of represen-
tations of mapping class groups. In his work, Teschner extends the set of operators
corresponding to morphisms of a surface into itself (and arising from the quantum
Teichmüller space) to the morphisms between different surfaces in a compatible way.

This chapter also contains a review of the coordinates for Teichmüller space that were
used by Penner, by Fock and by Kashaev, as well as a self-contained presentation of
the quantization theory of Teichmüller space.
Chapter 17 by Rinat Kashaev is a review of his quantization theory of the moduli
space of irreducible flat PSL(2, R)-connections on a punctured surface, or, equiva-
lently, of the space of conjugacy classes of irreducible representations of the funda-
mental group of the surface in PSL(2, R) with parabolicity conditions at the punctures.
Teichmüller space embeds as a connected component of this space. The moduli space
of irreducible flat PSL(2, R)-connections is equipped with a symplectic form defined
by Goldman, which restricts to the Weil–Petersson form on Teichmüller space. This
quantization theory leads to an infinite dimensional projective unitary representation
of the mapping class group.
Part A
The metric and the analytic theory, 1

Chapter 1
Harmonic maps and Teichmüller theory
Georgios D. Daskalopoulos and Richard A. Wentworth
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2 Teichmüller space and extremal maps . . . . . . . . . . . . . . . . . . . . 37
2.1 The Teichmüller theorems . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.1 Uniformization and the Fricke space . . . . . . . . . . . . . . 38
2.1.2 Quasiconformal maps . . . . . . . . . . . . . . . . . . . . . . 39
2.1.3 Quadratic differentials and Teichmüller maps . . . . . . . . . 40
2.1.4 The Teichmüller space . . . . . . . . . . . . . . . . . . . . . 42
2.1.5 Metric definition of Teichmüller space . . . . . . . . . . . . . 44
2.2 Harmonic maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2.2 Existence and uniqueness . . . . . . . . . . . . . . . . . . . . 49
2.2.3 Two dimensional domains . . . . . . . . . . . . . . . . . . . 51
2.2.4 A second proof of Teichmüller’s theorem . . . . . . . . . . . . 53
2.3 Singular space targets . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.1 The Gerstenhaber–Rauch approach . . . . . . . . . . . . . . . 54
2.3.2 R-trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3.3 Harmonic maps to NPC spaces . . . . . . . . . . . . . . . . . 59
3 Harmonic maps and representations . . . . . . . . . . . . . . . . . . . . . 63
3.1 Equivariant harmonic maps . . . . . . . . . . . . . . . . . . . . . . . 63
3.1.1 Reductive representations . . . . . . . . . . . . . . . . . . . . 63
3.1.2 Measured foliations and Hopf differentials . . . . . . . . . . . 66
3.2 Higgs bundles and character varieties . . . . . . . . . . . . . . . . . 69
3.2.1 Stability and the Hitchin–Simpson theorem . . . . . . . . . . 69
3.2.2 Higgs bundle proof of Teichmüller’s theorem . . . . . . . . . 72
3.2.3 The Thurston–Morgan–Shalen compactification . . . . . . . . 74
4 Weil–Petersson geometry and mapping class groups . . . . . . . . . . . . 77
4.1 Weil–Petersson geodesics and isometries . . . . . . . . . . . . . . . . 77
4.1.1 The Weil–Petersson metric and its completion . . . . . . . . . 78
4.1.2 The mapping class group . . . . . . . . . . . . . . . . . . . . 81
4.1.3 Classification of Weil–Petersson isometries . . . . . . . . . . 83
4.2 Energy of harmonic maps . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2.1 Nielsen realization . . . . . . . . . . . . . . . . . . . . . . . 85
34 Georgios D. Daskalopoulos and Richard A. Wentworth
4.2.2 Properness of the energy . . . . . . . . . . . . . . . . . . . . 86

4.2.3 Convexity of energy and length functionals . . . . . . . . . . 87
4.2.4 Further applications . . . . . . . . . . . . . . . . . . . . . . . 88
5 Harmonic maps to Teichmüller space . . . . . . . . . . . . . . . . . . . . 90
5.1 Existence of equivariant harmonic maps . . . . . . . . . . . . . . . . 90
5.1.1 Maps to the completion . . . . . . . . . . . . . . . . . . . . . 90
5.1.2 Surface domains . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.1.3 Holomorphic maps from Riemann surfaces . . . . . . . . . . 93
5.2 Superrigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.2.1 The Ivanov–Farb–Kaimanovich–Masur theorem . . . . . . . . 95
5.2.2 Harmonic maps from singular domains . . . . . . . . . . . . . 96
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
1 Introduction
Teichmüller theory is rich in applications to topology and physics. By way of the
mapping class group the subject is closely related to knot theory and three-manifolds.
From the uniformization theorem, Teichmüller theory is part of the more general
study of Kleinian groups and character varieties. Conformal field theory and quantum
cohomology make use of the algebraic and geometric properties of the Riemann moduli
space.
At the same time, analytic techniques have been important in Teichmüller theory
almost from the very beginning of the subject. Extremal maps and special metrics
give alternative perspectives to moduli problems and clarify our understanding of a
wide range of results. In some cases they can be used to obtain new properties.
The goal of this exposition is to present some of the more recent activity using
analysis, and in particular harmonic maps, in the context of Teichmüller theory, rep-
resentations of surface groups, mapping class groups, and Weil–Petersson geometry.
Topics have been selected in order to illustrate the theme that the analytic and topo-
logical points of view complement each other in a useful way. For example, we shall
present four different proofs of the fact that Teichmüller space is a cell, and we shall
discuss the recent completion of a harmonic maps approach to Teichmüller’s existence
and uniqueness theorems on extremal quasiconformal maps. Instead of a systematic
survey of the subject, we have chosen to present the ideas behind the results through
examples and in a rather informal way. There are very few proofs, but hopefully the
references given at the end will provide the interested reader sufficient recourse for
more details.
We make no attempt to exhaust all aspects of this subject. In particular, no mention
is made of the work on quasiconformal harmonic maps of the disk and Schoen’s
conjecture (see [117], [118], [159], [181], [190]), or of the universal Teichmüller
space in general. Other topics that have been covered in great detail in the literature
Chapter 1. Harmonic maps and Teichmüller theory 35
have also been omitted or only briefly touched upon. For example, there is little
discussion of the complex analytic theory of Teichmüller space, the Bers embedding,
Royden’s theorem on automorphisms, etc. We refer instead to the other chapters in
this Handbook, in particular to [62] by A. Fletcher and V. Markovic. For the same
reason, our summary of Weil–Petersson geometry is rather brief, and instead we refer
to Wolpert’s recent survey [206].
Finally, while we have tried to give complete and accurate references to the re-
sults stated in this chapter, given the expanse of the subject there will inevitably be
omissions. For these we offer our apologies in advance. Two useful surveys of ear-
lier results on harmonic maps are [50] and [166]. Relatively recent general texts on
Teichmüller theory are [1], [86], [147]. The point of view taken in Tromba’s book
[186] is especially relevant to the material presented here. For an interesting account
of Teichmüller’s life and work, see Abikoff [2].
Georgios D. Daskalopoulos was partially supported by NSF Grant DMS-0204191.
Richard A. Wentworth was partially supported by NSF Grants DMS-0204496 and
DMS-0505512.
Notation
For simplicity, we shall deal with connected compact oriented surfaces without bound-
ary and of genus p ≥ 2. The notation we shall use is the following: S will denote
the underlying smooth surface, and j will denote a complex structure on S. Hence, a
Riemann surface is a pair (S, j ). The hyperbolic metric on S will be denoted by σ .
Since it is uniquely determined by and uniquely determines the complex structure,
the notation (S, j ) and (S, σ ) will both be understood to represent a Riemann surface
structure. When the complex structure is understood we shall use letters S and R
alone to denote Riemann surfaces, and hopefully this will not cause confusion. The
following are some of the commonly used symbols in this chapter:
• id = identity map;
• I = identity endomorphism;
• f ∼ f homotopic maps;
• deg(f ) = the degree of a map between surfaces;
• K(f ) = the dilatation of a quasiconformal map (Section 2.1.2);
• = π1 (S), or = π1 (M) for a manifold M;
• M̃ = the universal cover of M;
• p = the space of smooth p-forms;
• = a Fuchsian group (Section 2.1.1);
• D = the unit disk in C;
• H = the upper half plane in C;
• H3 = hyperbolic 3-space SL(2, C)/ SU(2);

• i(a, b) = the geometric intersection number of simple closed curves a, b on S;
• c [σ ] = the length of a simple closed curve on S with respect to the hyperbolic
metric σ ;
• F = a measured foliation on S (Section 2.3.2);
• i([c], F ) = the intersection number of an isotopy class of simple closed curves
with a measured foliation F (see Section 2.3.2);
• i(F1 , F2 ) = the intersection number of a pair of measured foliations (see Sec-
tion 2.3.2);
• TF = the R-tree dual to a measured foliation F (Section 2.3.2);
• MF(S) (resp. PMF(S)) = the spaces of measured (resp. projective measured)
foliations on S (Section 2.3.2);
• KS = the canonical line bundle on a Riemann surface S;
• χS = the Euler characteristic of S;
• ∇ = the covariant derivative, or a connection on a vector bundle V ;
• d∇ = the de Rham operator, twisted by a connection ∇;
• ∇H = the Chern connection on a holomorphic bundle with hermitan metric H
(Section 3.2.1);
• F∇ = the curvature of a connection ∇;
• = the Laplace–Beltrami operator;
• μ = a Beltrami differential (Section 2.1.2);
• μ∞ = the L∞ norm of a Beltrami differential μ;
• ϕ = a holomorphic quadratic differential (Section 2.1.3);
• ϕ1 (resp. ϕ2 ) = the L1 (resp. L2 ) norms of a quadratic differential ϕ (see
eqs. (2.6) and (2.17));
• Tϕ = the R-tree dual to the horizontal foliation of ϕ (Section 2.3.2);
• QD(S) = the space of holomorphic quadratic differentials;
• F(S) = the Fricke space (Section 2.1.1);
• χ (, r) (resp. χ()) = the SL(r, C) (resp. SL(2, C)) character varieties of
(Section 3.2.1);
• T (S) = Teichmüller space (Section 2.1.4);
• T (S) = the Weil–Petersson completion of T (S) (Section 4.1.1);
• Diff(S), Diff + (S), Diff 0 (S) = the diffeomorphisms, orientation preserving dif-
feomorphisms, and diffeomorphisms connected to the identity of a surface S;
• Mod(S) = the mapping class group (Section 4.1.2);
• M(S) = the Riemann moduli space (Section 4.1.2);

• M(S) = the Deligne–Mumford compactification of M(S) (Section 4.1.2);
• dT = the Teichmüller metric on T (S) (see eq. (2.9));
• dwp = the Weil–Petersson metric on T (S) (Section 4.1.1);
• Iso(X) = the isometry group of a metric space (X, d);
• ∂X = the ideal boundary of an NPC space X (Section 3.1);
• Lρ = the translation length function of a representation (3.1);
• Lwp [φ] = the Weil–Petersson translation length of [φ] ∈ Mod(S) (see eq. (4.4));
• H 1 (resp. Hloc
1 ) = the Sobolev space of square integrable (resp. locally square
integrable) functions with square integrable (resp. locally square integrable) dis-
tributional derivatives;
• e(f ) = the energy density of a map f (see eq. (2.19));
• παβ = the directional energy tensor (see eq. (2.48));
• E(f ) = the energy of a map f (see eq. (2.20));
• End(V ) (resp. End0 (V )) = the endomorphism (resp. traceless endomorphism)
bundle of a complex vector bundle V (Section 3.2.1);
• ad(V ) (resp. ad0 (V )) = the skew-hermitian (resp. traceless skew-hermitian) en-
domorphism bundle of a hermitian vector bundle V (Section 3.2.1);
• = a Higgs field (Section 3.2.1);
• M(S, r) = the moduli space of polystable Higgs bundles of rank r on S (Sec-
tion 3.2.1).
2 Teichmüller space and extremal maps

2.1 The Teichmüller theorems
In this section we give a summary of the basics of Teichmüller theory from the point
of view of quasiconformal maps. In Section 2.1.1 we review the uniformization
theorem and the Fricke space. In Section 2.1.2 we introduce quasiconformal maps,
Beltrami differentials, and we state the basic existence theorem for solutions to the Bel-
trami equation. We also formulate the extremal problem. In Section 2.1.3 we review
quadratic differentials, Teichmüller maps and Teichmüller’s existence and uniqueness
theorems. In Section 2.1.4 we define the Teichmüller space based on a Riemann sur-
face and discuss the first approach to Teichmüller’s theorem on the contractibility of
Teichmüller space. The proof that we give here is based on the notion of extremal
maps, i.e. quasiconformal maps that minimize dilatation in their homotopy class. The
connection between extremal and harmonic maps will be explained in Section 2.3.1.
Finally, in Section 2.1.5, we provide an alternative definition of Teichmüller space via
hyperbolic metrics.
2.1.1 Uniformization and the Fricke space. The famous uniformization theorem
of Poincaré, Klein, and Koebe states that every closed Riemann surface S of genus
at least 2 is biholomorphic to a quotient H/, where H denotes the upper half plane
and is a group of holomorphic automorphisms of H acting freely and properly
discontinuously. Such a group can be identified with a discrete subgroup of PSL(2, R),
i.e. a Fuchsian group (cf. [63], [86]).
On H we have the Poincaré metric
|dz|2
dsH
2
= .
(Im z)2
2 = h∗ ds 2 ,
Under the biholomorphism h : H → D given by h(z) = (z−i)/(z+i), dsH D
where
4|dz|2
dsD
2
= .
(1 − |z|2 )2
By a straightforward calculation the curvature of the Poincaré metric is constant equal
to −1, and by Pick’s theorem its isometry group is PSL(2, R) (cf. [86]). In particular,
this metric descends to the quotient H/. Hence, every Riemann surface of genus
≥ 2 has a hyperbolic metric, and this metric is unique. On the other hand, any
Riemannian metric induces a unique complex structure. This is a consequence of
Gauss’ theorem on the existence of isothermal coordinates: if (S, g) is an oriented
surface with Riemannian metric g, then S admits a unique complex structure j such
that in local complex coordinates g = g(z)|dz|2 , where g(z) is a smooth, positive
(local) function. Hence, specifying a complex structure on the topological surface S
is equivalent to specifying a hyperbolic metric. We shall use Greek letters, e.g. σ =
σ (z)|dz|2 , to distinguish the hyperbolic from arbitrary Riemannian metrics g.
Let F(S) denote the Fricke space of conjugacy classes of discrete embeddings
= π1 (S) → PSL(2, R). Then F(S) inherits a topology as a character variety
(cf. [33], [72], [71] and Section 3.2.1 below, and the chapter by W. M. Goldman in
Volume II of this Handbook [72]). The idea is to choose a marking of the genus p
surface S, namely, a presentation
p
= a1 , . . . , ap , b1 , . . . , bp : i=1 [ai , bi ] = id .
where the ai and bi are represented by simple closed curves on S with geometric
intersection numbers satisfying i(ai , bj ) = δij , i(ai , aj ) = i(bi , bj ) = 0. A homo-
morphism ρ : → PSL(2, R) isdetermined by specifying 2p elements Ai , Bi ∈
p
PSL(2, R) satisfying the relation i=1 [Ai , Bi ] = I. A naive dimension count (which
can easily be made precise at irreducible representations ρ) suggests that the dimension
of the space of such homomorphisms is 6p − 3. Since PSL(2, R) acts by conjugation,
producing a 3-dimensional orbit, we have dim F(S) = 6p − 6. Indeed, since the
Fricke space consists of discrete embeddings, a more precise analysis can be given
which realizes F(S) as a subset of R6p−6 (cf. [1]).
Proposition 2.1. The Fricke space F(S) is embedded in R6p−6 .

It is this embedding (the details of which will not be important) that we shall use
to define the topology on F(S). We shall see below that F(S) is homeomorphic to
Teichmüller space (Theorem 2.9).
2.1.2 Quasiconformal maps. An orientation preserving homeomorphism f of a do-

main ⊂ C into C is called K-quasiconformal (or K-qc) if
(1) f is of Sobolev class Hloc 1 , i.e. the distributional derivatives f , f are locally
z z̄
square integrable on ;
(2) there exists a constant 0 ≤ k < 1 such that |fz̄ | ≤ k|fz |, almost everywhere on
, where K = (1 + k)/(1 − k).
The infimum of K ≥ 1 such that f is K-qc is called the dilatation of f , and it is
denoted by K(f ). Clearly, 1-qc is equivalent to conformal.
An orientation preserving homeomorphism f : S → R between two Riemann
surfaces is called K-qc if its lift to the universal cover f˜ : H → H is K-qc. We define
the dilatation K(f ) of f to be K(f˜). Given such a map f , let
QC[f ] = {f : S → R : f is a qc homeomorphism homotopic to f }. (2.1)
The main extremal problem in Teichmüller theory is a generalization to closed
surfaces of Grötzsche’s problem for rectangles (see [1]): given a qc map f : S → R,
let
K ∗ [f ] = inf K(f ). (2.2)
f ∈QC[f ]
Teichmüller’s extremal problem. Is K ∗ [f ] realized as the dilatation of a qc map,

and if so, what are the properties of the map?
A qc homeomorphism f such that K(f ) = K ∗ [f ] is called an extremal map. The
existence of extremal maps is a relatively easy consequence of compactness properties
of quasiconformal maps. The emphasis of this problem is therefore on the uniqueness
and characterization of extremal maps. We shall give Teichmüller’s answer to this
question in the next section.
Choose coordinates (U, z) on S and (V , w) on R and set F = w f z−1 . Define
the Beltrami coefficient of f with respect to the choice of coordinates by
μf = μf (z)d z̄ ⊗ (dz)−1 = Fz̄ /Fz d z̄ ⊗ (dz)−1 .
By (2), |μf (z)| < 1 almost everywhere. The above expression is independent of
the choice of coordinates w and transforms tensorially with respect to coordinate
changes in z. More precisely, μf may be regarded as an L∞ -section of the bundle
KS ⊗ KS−1 , where KS is the canonical line bundle of S. Notice, however, that |μf (z)|
is independent of a choice of conformal coordinates. Set μ∞ to be the essential
supremum of |μf | over S.
Let B(S) denote the Banach space of L∞ -sections of KS ⊗KS−1 with the L∞ -norm.
Set
B1 (S) = {μ ∈ B(S) : μ∞ < 1}.
For any qc map f : S → R we associate μf ∈ B1 (S). If S = H/ a Beltrami

differential on S can be identified with an L∞ function μ̃f on H satisfying the equations
of automorphy
γ (z)
μ̃f (γ z) = μ̃f (z), z ∈ H, γ ∈ . (2.3)
γ (z)
Furthermore, qc homeomorphisms f˜ of H whose Beltrami coefficients satisfy (2.3)
give deformations of Fuchsian groups via
−1
; μ : γ ∈ → f˜ γ f˜ ∈ PSL(2, R). (2.4)
Specifying the Beltrami coefficient and solving for a qc map is called Beltrami’s equa-
tion. The following is the fundamental existence theorem for solutions to Beltrami’s
equation. The seminal reference is Ahlfors [5]. See also [86, Chapter 4].
Theorem 2.2. For any Beltrami differential μ ∈ B1 (C) there exists a unique qc
homeomorphism f μ of H, extending continuously to Ĥ = H ∪ {∞}, whose Beltrami
coefficient is μf μ = μ, and which fixes the points 0, 1, and ∞. Furthermore, f μ
depends complex analytically on μ.
Corollary 2.3. For any Beltrami differential μ ∈ B1 (S) there exists a unique qc
homeomorphism f μ : S → R, for some Riemann surface R. More precisely, if
μ
S = H/, then R = H/f˜ (f˜μ )−1 , where f˜μ is the solution in Theorem 2.2
for the pullback Beltrami differential. Furthermore, f μ depends complex analytically
on μ.
Hence, Beltrami differentials can be used to parametrize the Fricke space F(S).
Of course, there is an infinite dimensional family of Beltrami differentials giving
conjugate Fuchsian groups.
2.1.3 Quadratic differentials and Teichmüller maps. By a holomorphic quadratic

differential on a Riemann surface S we mean a holomorphic section of the line bundle
KS2 . Set QD(S) = H 0 (S, KS2 ). By the Riemann–Roch theorem, QD(S) is a complex
vector space of dimension 3p − 3, where p is the genus of S. If ϕ ∈ QD(S), then
in local conformal coordinates (centered at z0 , say) ϕ = ϕ(z)dz2 , where ϕ(z) is a
local holomorphic function. By a coordinate change we can write ϕ = zk dz2 , where
k = 0, 1, 2, . . . is the order of vanishing of ϕ at z0 . The coordinate system
z z
√ 2 k+2
w(z) = ϕ= ϕ(z)dz = z 2
z0 z0 k+2
will be called the ϕ-coordinates around z0 (if m is odd, this is multi-valued). Writing
w = u + iv, the foliations v = constant and u = constant are called the horizontal
and vertical foliations of ϕ, respectively.
ϕ(z0 ) = 0 ϕ(z0 ) = 0, k = 1 ϕ(z0 ) = 0, k = 2
Figure 1
For more details, we refer to Strebel’s treatise on quadratic differentials [179].

A holomorphic quadratic differential on S = H/ is given by ϕ = ϕ̃(z)dz2 , where
ϕ̃ is a holomorphic function on H satisfying the equations of automorphy
ϕ̃(γ z)γ (z)2 = ϕ̃(z), z ∈ H, γ ∈ . (2.5)
Set
QD1 (S) = {ϕ ∈ QD(S) : ϕ1 < 1},
where · 1 denotes the L1 -norm

ϕ1 = |ϕ(z)| dxdy. (2.6)
S
Given ϕ ∈ QD1 (S)\{0}, we say that a qc homeomorphism f : S → R is a Teichmüller
map for ϕ if the Beltrami coefficient of f satisfies
ϕ̄
μf = k , k = ϕ1 . (2.7)
|ϕ|
We are now in a position to give Teichmüller’s solution to the extremal problem
stated in the previous section. First, a Teichmüller map is uniquely extremal.
Theorem 2.4 (Teichmüller’s Uniqueness Theorem). Let f : S → R be a Teichmüller

map. Then every f ∈ QC[f ] satisfies
μf ∞ ≥ μf ∞ (equivalently, K(f ) ≥ K(f )).
Moreover, the equality holds if and only if f = f .
The second result asserts that Teichmüller maps always exist.
Theorem 2.5 (Teichmüller’s Existence Theorem). In the homotopy class of every qc

homeomorphism f : S → R there is either a conformal map or a Teichmüller map.
Theorems 2.4 and 2.5 were first stated by Teichmüller (see [182]). His papers
are “generally considered unreadable” (Abikoff, [1, p. 36]). Subsequent proofs were
given in [3] and [13] (see also [66], [77]). Below we outline a proof of these two
fundamental results based on harmonic maps to singular spaces (see Section 2.3.1).
Teichmüller maps are essentially affine with respect to a natural choice of coordi-
nates (see [86]):
Theorem 2.6. Fix ϕ ∈ QD1 (S) \ {0}, k = ϕ1 < 1, and let f : S → R be a
Teichmüller map for ϕ. Then there exists a unique holomorphic quadratic differential
ψ on R satisfying the following conditions.
(1) If z is a zero of ϕ, then f (z) is a zero of ψ of the same order.
(2) If z is not a zero of ϕ and ζ is a ϕ-coordinate about z, then there exists a ψ-
coordinate w at f (z) such that
ζ + k ζ̄
wf = . (2.8)
1−k
The quadratic differentials ϕ and ψ are called the initial and terminal differentials
of the Teichmüller map f , respectively.
2.1.4 The Teichmüller space. We now come to the definition of Teichmüller space.
Let S be a closed Riemann surface of genus p ≥ 2. Consider triples (S, f, R), where
R is a Riemann surface and f : S → R is an orientation preserving diffeomorphism.
Triples (S, f1 , R1 ) and (S, f2 , R2 ) are said to be equivalent if f2 f1−1 : R1 → R2
is homotopic to a biholomorphism. The set of all equivalence classes [S, f, R] of
triples (S, f, R) is denoted T (S) and is called the Teichmüller space based on S. The
definition of T (S) turns out to be independent of the complex structure on S (see
Theorem 2.7 below). Since any homeomorphism (in particular quasiconformal ones)
is homotopic to a diffeomorphism, one obtains the same space if one considers pairs
(S, f, R) where f is quasiconformal. This is a point of subtlety when dealing with
Riemann surfaces with punctures.
Restricting as we are to the case of closed surfaces, Teichmüller space may be
regarded as parametrizing complex structures up to biholomorphisms connected to the
identity. Indeed, if S0 = (S, j0 ) denotes the basepoint and (S, j ) is another complex
structure on the underlying surface S, then by choosing f = id and R = (S, j ) there is
an associated point [j ] = [S0 , id, R] ∈ T (S). Two points [S0 , id, R1 ] and [S0 , id, R2 ]
obtained in this way are equivalent if and only if (S, j1 ) and (S, j2 ) are biholomorphic
via a map connected to the identity. Conversely, given any triple (S0 , f, R), let j
denote the pullback by f of the complex structure on R to the underlying surface S.
Then by definition f : (S, j ) → R is a biholomorphism; hence, (S0 , id, (S, j )) is
equivalent to (S0 , f, R). With this understood, we sometimes represent points in
T (S) by equivalence classes [j ].
Given [j1 ], [j2 ] ∈ T (S), recall that QC[id] is the set of all qc homeomorphisms
(S, j1 ) → (S, j2 ) homotopic to the identity. The Teichmüller metric dT is defined by
dT ([j1 ], [j2 ]) = inf log K(f ). (2.9)
f ∈QC[id]
For the next result we refer to [86, §5.1].
Theorem 2.7. T (S) is a complete metric space with respect to the Teichmüller met-
ric dT . Furthermore, given [S, f, R] ∈ T (S), the map [f ]∗ : T (S) → T (R) given
by [S, f , R ] → [R, f f −1 , R ] is an isometry.
Henceforth, the topology on T (S) is that given by the metric dT . Also, in light
of the theorem we identify all Teichmüller spaces independent of the choice of base
point. Now we are ready for Teichmüller’s third result.
Theorem 2.8 (Teichmüller’s theorem). T (S) is homeomorphic to a cell of dimension

6p − 6.
By Corollary 2.3 on solutions to Beltrami’s equation, Teichmüller maps with initial

differential ϕ exist for any ϕ ∈ QD1 (S). Hence, we may define a map
τ : QD1 (S) −→ T (S), τ (ϕ) = [S, f, R], (2.10)
where f is a Teichmüller map for ϕ = 0, and f = id, R = S, for ϕ = 0. Theorem 2.8
follows from
Theorem 2.9. The map τ in (2.10) is a homeomorphism. Moreover, T (S) is homeo-

morphic to F(S).
Proof. First, note that there is a natural bijection F : T (S) → F(S) defined as follows:
given [S, f, R] ∈ T (S), by the uniformization theorem applied to the Riemann surface
R there is a discrete embedding ρR : π1 (R) → PSL(2, R), determined up to conjuga-
tion. Since the diffeomorphism f induces an isomorphism f∗ : = π1 (S) −− ∼→ π1 (R),
we obtain a discrete embedding ρ = ρR f∗ : → PSL(2, R). Notice that if
[S, f1 , R1 ] = [S, f2 , R2 ], then the corresponding homomorphisms are conjugate.
Hence, there is a well-defined point F [S, f, R] ∈ F(S). Conversely, given a dis-
crete embedding ρ : → PSL(2, R), consider the Riemann surface R = H/ρ().
The Poincaré polygon theorem realizes the boundary of a fundamental domain for
the action of ρ() as the lift of simple closed curves αi , βi on R satisfying the re-
lations i(αi , βj ) = δij , i(αi , αj ) = i(βi , βj ) = 0 (cf. [11]). The identification of
ai , bi with αi , βi fixes a homotopy class of diffeomorphisms f : S → R, and it is
clear that F [S, f, R] = [ρ]. Hence, F is a bijection. Moreover, F is continuous by
Corollary 2.3, since a qc map of small dilatation is close to the identity, hence the
corresponding deformation of the Fuchsian groups is small. Consider the following
diagram
τ
/ T (S)
QD1 (S)G
KK
KK
KK
KK F
K%
F(S)
where G = F τ . By Theorems 2.4 and 2.5, τ is a bijection. It is also continuous.
Indeed, dT (τ (0), τ (ϕ)) = log((1+k)/(1−k)), where k = ϕ1 (recall that τ (0) = S),
so τ is clearly continuous at the origin. Continuity at general points follows from the
change of basepoints in Theorem 2.7. It follows that G is a continuous bijection. By
the embedding F(S) → R6p−6 (Proposition 2.1) and Invariance of Domain, G is a
homeomorphism; hence, so are F and τ .
We have proven Teichmüller’s Theorem via his existence and uniqueness results
(Theorems 2.4 and 2.5). The proof uses the Fricke space F(S) and the finite dimen-
sionality of the space of holomorphic quadratic differentials. In Sections 2.2.4, 3.2.2,
and 4.2.2 we shall give three alternative proofs of Theorem 2.8 using harmonic maps
and the metric description of Teichmüller space.
2.1.5 Metric definition of Teichmüller space. Let S be an oriented surface of genus

p ≥ 2. Let Methyp (S) be the space of metrics with constant curvature −1. This has a
smooth structure inherited as a smooth submanifold of the space Met(S) of all smooth
metrics on S. As discussed in Section 2.1.1, a hyperbolic metric defines a complex
structure on S via Gauss’ theorem, and conversely, in every conformal class of metrics
compatible with a given complex structure there is a unique hyperbolic metric. The
group Diff 0 (S) of diffeomorphisms isotopic to the identity acts on Met hyp (S) by
pullback. Define
Thyp (S) = Methyp (S)/Diff 0 (S), (2.11)
with the quotient topology. By constructing a slice for the action of Diff 0 (S) on
Met hyp (S) it is not hard to prove (see [48], [60], [154], [186])
Proposition 2.10. Thyp (S) is a smooth manifold of dimension 6p − 6.
To elaborate on this statement, we review the description of the tangent and cotan-
gent spaces to T (S) (for this approach, cf. [41], [60]). Let ∇ denote the covariant
derivative for a metric g on S. On the tangent space Tg Met(S) there is a natural
L2 -pairing:
δg, δg = (g αβ g μν δgαμ δgβν

) d vol(S,g) (2.12)
S
where the metrics and variations are expressed with respect to local coordinates {x α },
z = x 1 + ix 2 = x + iy, and repeated indices are summed. For σ a hyperbolic metric,
the condition that δσ be tangent to Methyp (S) is
0 = (− + 1) Tr(δσ ) + ∇ α ∇ β (δσαβ ) , (2.13)
where is the Laplace–Beltrami operator associated to σ . Finally, the tangent space

to the orbit Diff 0 (S) · σ at σ consists of Lie derivatives of the metric:
δσαβ = (Lv σ )αβ = ∇α vβ + ∇β vα (2.14)
for smooth vector fields {v α }. From (2.12) and (2.14), the L2 -orthogonal in Tσ Met(S)
to Tσ (Diff 0 (S) · σ ) consists of variations satisfying
∇ α δσαβ = 0. (2.15)
Restricting to hyperbolic metrics, it then follows from (2.13) that these variations must
also be traceless. Hence, we have an identification of T[σ∗ ] Thyp (S) with the space of
traceless symmetric 2-tensors satisfying (2.15). Now the bundle of traceless symmetric
2-tensors is real isomorphic to KS2 via 2ϕ(z) = δσ11 − iδσ12 . Moreover, (2.15) is
precisely the statement that the corresponding quadratic differential ϕ = ϕ(z)dz2 is
holomorphic. Hence, T[σ∗ ] T (S) QD(S).
This description makes contact with the Kodaira–Spencer theory of deformations
of a complex structure (cf. [108]). Indeed, infinitesimal deformations of a complex
structure are parametrized by smooth sections μ of KS ⊗KS−1 . These are just (smooth)
Beltrami differentials. Note that there is a natural pairing between Beltrami differ-
entials and holomorphic quadratic differentials on a Riemann surface S obtained by
raising indices in (2.12):

μ, ϕ = μ(z)ϕ(z)|dz|2 , (2.16)
S
where μ = μ(z)d z̄ ⊗ (dz)−1 and ϕ = ϕ(z)dz2 . Let H B(S) denote the space of
harmonic Beltrami differentials, i.e.
HB(S) = {μ ∈ B(S) : ∂¯ ∗ μ = 0},
where the adjoint ∂¯ ∗ is defined with respect to the hyperbolic metric. For any holo-
morphic quadratic differential ϕ, the Beltrami differential μ = σ −1 ϕ̄ is harmonic.
Moreover, μ, ϕ = ϕ22 , where · 2 denotes the L2 -norm with respect to the
metric σ :

ϕ22 = |ϕ(z)|2 σ (z)−1 dxdy. (2.17)
S
It follows that the pairing

·, · : HB(S) × QD(S) −→ C (2.18)
is nondegenerate and that the tangent space is given by T[σ ] T (S) = H B(S).
To complete this circle of ideas, one can directly compute the Beltrami differential
associated to δσ . Let σt be a differentiable family of hyperbolic metrics with σ0 = σ ,
(dσt /dt)|t=0 = δσ , and let νt be the Beltrami differentials associated to the identity
map (S, σ ) → (S, σt ), ν0 = 0, (dνt /dt)|t=0 = μ. If w = wt , w0 = z is a family of
conformal coordinates such that

dsσ2t = σt (w)|dw|2 = σt (w)|wz |2 |dz + νt d z̄|2 ,
then since δσαβ is traceless,

d
σt (w)|wz |2 = 0.
dt t=0
It then follows that 2σ μ = δσ11 +iδσ12 , in agreement with the correspondence above.
There is a canonical map c : Thyp (S) → T (S) obtained by associating to an equiv-
alence class of hyperbolic metrics the corresponding equivalence classes of complex
structures obtained via Gauss’ theorem (see Section 2.1.1). This map is continuous,
for if two hyperbolic metrics are close in the smooth topology, then the identity has
small dilatation. Furthermore, c is a bijection by the uniformization theorem. With this
understood, we now see that the two definitions of Teichmüller space are equivalent.
Theorem 2.11. The canonical map c : Thyp (S) → T (S) obtained by associating to
the hyperbolic metric its conformal class is a homeomorphism.
Proof. Recall from the proof of Theorem 2.9 that the map F : T (S) → F(S) is also
a continuous bijection. Since F(S) ⊂ R6p−6 , it follows by Proposition 2.10 and
Invariance of Domain that the composition
F c : Thyp (S) → F(S) → R6p−6
is a homeomorphism; hence, both F and c are as well.
Remark 2.12. (1) We emphasize that the proof of the homeomorphism Thyp (S)
T (S) given above is independent of the Teichmüller Theorems 2.5, 2.4, and 2.8.
(2) By Theorem 2.11, we may regard the topological space T (S) either as equiva-
lence classes of marked Riemann surfaces or as the moduli space of hyperbolic metrics.
In particular, for the alternative proofs of Theorem 2.8 given below, it suffices to prove
that Thyp (S) is homeomorphic to a cell.
(3) The L2 -metric (2.17) is the Weil–Petersson cometric on T (S) (see Section 4.1.1
below). In this description, it is easy to see that the Teichmüller metric (2.9) is a Finsler
metric defined by the L1 -norm (2.6).
2.2 Harmonic maps

This section is a brief summary of the theory of harmonic maps with an emphasis
on those aspects that relate to Teichmüller theory. In Section 2.2.1 we give the basic
definitions and present the variational formulation along with some examples. In Sec-
tion 2.2.2 we state the existence and uniqueness theorem of Eells–Sampson–Hartman
for nonpositively curved targets, and we indicate the importance of the Bochner for-
mula. In Section 2.2.3 we specialize to the case of surface domains. We discuss
conformal invariance, the Hopf differential, and some applications. In Section 2.2.4
we present another proof that Teichmüller space is a cell using harmonic maps.
2.2.1 Definitions. Let (M, g) and (N, h) be Riemannian manifolds. With respect to
coordinates {x α } on M and {y i } on N , write g = (gαβ ), h = (hij ). Given a smooth
map f : M → N, its differential
(df )kα = (∂f k /∂x α )dx α ⊗ (∂/∂y k ),
is a section of the bundle T ∗ M ⊗ f ∗ T N with the induced metric and connection.
Define the energy density and energy of f by
1 ∂f i ∂f j αβ
e(f ) = 21 df, df T ∗ M⊗f ∗ T N = g hij f, (2.19)
2 ∂x α ∂x β
E(f ) = e(f ) d volM , (2.20)
M
respectively (repeated indices are summed). The energy can be viewed as a functional
on the space of smooth maps between M and N .
The second extremal problem, analogous to the Teichmüller problem in Sec-
tion 2.1.2, may now be formulated as follows: given a smooth map f : (M, g) →
(N, h), let
E ∗ [f ] = inf{E(f ) : f smooth, f ∼ f }. (2.21)
Energy extremal problem. Is E ∗ [f ] is realized as the energy of a smooth map, and

if so, what are the properties of the map?
A smooth map f such that E(f ) = E ∗ [f ] is called an energy minimizer. Unlike
the problem for quasiconformal maps, existence of energy minimizers is not obvious.
We shall discuss this at greater length in the next section.
The covariant derivative ∇df is a section of Sym2 (T ∗ M) ⊗ f ∗ T N, where Sym2
denotes symmetric 2-tensors. The trace τ (f ) = Tr g ∇df is called the tension field
of f . Let denote the Laplace–Beltrami operator on (M, g). Then
∂f i ∂f j αβ
τ (f )k = f k + (ijk f ) g .
∂x α ∂x β
Here, ijk denotes the Christoffel symbols of N. A smooth map f : M → N is called
harmonic if τ (f ) ≡ 0. Let
d∇ : p (f ∗ T N) −→ p+1 (f ∗ T N )
denote the exterior derivative coupled with pulled-back Levi-Cività connection on N .
It is easily seen that d∇ (df ) = 0 for all differentiable maps. The equations for
harmonic maps are then equivalent to
d∇ (∗df ) = 0, (2.22)
i.e. df is a harmonic form (cf. [51], [50]). Here are some examples.
• Harmonic maps S 1 → N are closed geodesics in N.
• When N = Rn the harmonic map equations are equivalent to the harmonicity of
the coordinate functions.
• Totally geodesic maps satisfy ∇df = 0, and so are harmonic.
• Holomorphic or anti-holomorphic maps between Kähler manifolds are harmonic.
• Minimal isometric immersions are harmonic.
Now let us consider variational formulas for the energy E(f ). A smooth vector field
v along f , i.e. v ∈ C ∞ (f ∗ T N), defines a variation of f by ft (x) = expf (x) (tv(x)).
Since N is assumed to be complete, this defines a smooth map M × R → N with
f0 = f . The first variational formula is

dE(ft )
δv E(f ) = = − τ (f ), vh d volM . (2.23)
dt t=0 M
It follows that the Euler–Lagrange equations for E are precisely the harmonic map
equations (2.22).
In general there is a distinction between energy minimizers, smooth minimizers
of E which then necessarily satisfy (2.22), and smooth solutions to (2.22) which may
represent higher critical points of the energy functional. We shall see below that this
distinction vanishes when the target manifold N has nonpositive curvature. Another
case where minimizers can be detected is the following: let S be a compact Riemann
surface and N a compact Kähler manifold.
Proposition 2.13. If f : S → N is holomorphic or anti-holomorphic, then for any

conformal metric on S, f is harmonic and is energy minimizing in its homotopy class.
Indeed, a computation in local coordinates as above shows that for any smooth
map f : S → N,

∗
E(f ) = f ω + 2 |∂f ¯ |2 d volS (2.24)
S S

= − f ∗ ω + 2 |∂f |2 d volS (2.25)
S S
where ω is the Kähler form on N. Since the first terms on the right hand sides depend
only on the homotopy class of f , the result follows.
Now let v, w ∈ C ∞ (f ∗ T N) and fs,t be a two-parameter family of maps such
that f0,0 = f , v = (∂fs,t /∂s)|s=t=0 , w = (∂fs,t /∂t)|(s,t)=(0,0) , where f is harmonic.
Then
∂ 2 E(fs,t )
Hf (v, w) = s=t=0
= − Jf v, wh d volM , (2.26)
∂s∂t M
where
Jf (v) = Tr g (∇ 2 v + RiemN (df, v)df ) (2.27)
is the Jacobi operator, and RiemN is the Riemannian curvature of (N, h). In particular,
if N has nonpositive Riemannian sectional curvature, then

Hf (v, v) ≥ |∇v|2 d volM ≥ 0,
M
and hence every harmonic map is a local minimum of the energy.
Given smooth maps f : M → N and ψ : N → P , one has the composition formula
∇d(ψ f ) = dψ ∇df + ∇dψ(df, df ).
Taking traces we obtain the formula for the tension (cf. [51])
τ (ψ f ) = dψ τ (f ) + Tr g ∇dψ(df, df ). (2.28)
In particular, if f is harmonic and ψ is totally geodesic then ψ f is also harmonic.
If P = R and f is harmonic, then (2.28) becomes
(ψ f ) = Tr g ∇dψ(df, df ),
and therefore a harmonic map pulls back germs of convex functions to germs of
subharmonic functions. The converse is also true:
Theorem 2.14 (Ishihara [87]). A map is harmonic if and only if it pulls back germs
of convex functions to germs of subharmonic functions.
2.2.2 Existence and uniqueness. In the case of nonpositively curved targets the
energy extremal problem has a solution. The basic existence result is the following
Theorem 2.15 (Eells–Sampson [51]). Let f : M → N be a continuous map be-

tween compact Riemannian manifolds, and suppose that N has nonpositive sectional
curvature. Then there exists an energy minimizing harmonic map homotopic to f .
The proof is based on the heat equation method to deform a map to a harmonic
one (cf. [78]). Namely, one solves the initial value problem for a nonlinear parabolic
equation
∂f
(x, t) = τ (f )(x, t), f (x, 0) = f (x). (2.29)
∂t
Stationary solutions to (2.29) satisfy the harmonic map equations. Furthermore, by
taking the inner product on both sides in (2.29) with τ (f )(x, t) and integrating over M,
one observes, using (2.23), that the energy of the map x → f (x, t) is decreasing in t.
Hence, one hopes that as t → ∞, f ( ·, t) converges to a harmonic map. Unfortunately,
it turns out that this procedure does not always work. In general, even existence of a
solution to (2.29) for all t ≥ 0 is not guaranteed (cf. [23], [31]). However, we have
Theorem 2.16. Assume M, N are compact Riemannian manifolds and N has non-
positive sectional curvature. Given a smooth map f : M → N, then the solution to
(2.29) exists for all t ∈ [0, ∞) and converges as t → ∞ uniformly to a harmonic map
homotopic to f .
The key to this theorem is following parabolic Bochner formula. Suppose f (x, t)
is a solution to (2.29) for 0 ≤ t < T , and let e(f )(x, t) denote the energy density of
the map x → f (x, t). Then for any orthonormal frame {uα } at a point x ∈ M we
have the following pointwise identity:
∂e(f )
− + e(f ) = |∇df |2 + df RicM (uα ), df (uα )
∂t (2.30)
− RiemN (df (uα ), df (uβ ))df (uβ ), df (uα ).
where RicM is the Ricci curvature of (M, g). In particular, if M is compact and N is
nonpositively curved, then
∂e(f )
≤ e(f ) + Ce(f ) (2.31)
∂t
for some constant C ≥ 0. If the solution f (x, t) exists for 0 ≤ t < T , then it follows
easily from (2.31) that e(f ) is uniformly bounded in x and t < T . The bound on
the energy density means that the maps f ( ·, t) form an equicontinuous family from
which convergence as t → T can be deduced. Resolving the initial problem at t = T
then allows one to extend the solution for some small time t > T . This is the rough
idea behind the existence for all 0 ≤ t < ∞. In fact, more sophisticated methods
show that e(f ) is bounded for all time (cf. [145]).
The Bochner formula for harmonic maps, i.e. stationary solutions of (2.29), is
e(f ) = |∇df |2 + df RicM (uα ), df (uα )
(2.32)
− RiemN (df (uα ), df (uβ ))df (uβ ), df (uα ).
As before, this implies
e(f ) ≥ −Ce(f ). (2.33)
Inequality (2.33) is the key to regularity of weakly harmonic maps to nonpositively
curved spaces (cf. [166]). To state the result precisely, we note that if ⊂ M is
a domain with smooth boundary, one can solve the Dirichlet problem for an energy
minimizing map f : → N with prescribed boundary conditions. If f : M → N
is energy minimizing then it is automatically energy minimizing with respect to its
boundary values for any ⊂ M. This is what is meant by locally energy mini-
mizing. The following Lipschitz bound follows from (2.33) by iterating the Sobolev
embedding.
Proposition 2.17. If f : → N is harmonic with energy E(f ) and N has nonpositive

curvature, then for any U ⊂⊂ ,
sup e(f )(x) ≤ C(U )E(f )
x∈U
for some constant C(U ) independent of f .
Next, we have the following result on uniqueness.
Theorem 2.18 (Hartman [79]). Assume M, N are compact Riemannian manifolds and
N has nonpositive sectional curvature. Let f0 , f1 : M → N be homotopic harmonic
maps, and let fs : M → N be a geodesic homotopy where s ∈ [0, 1] is proportional
to arc length. Then:
(1) for every s, fs is a harmonic map with E(fs ) = E(f0 ) = E(f1 ); and
(2) the length of the geodesic s → fs (x) is independent of x.
In case N has negative sectional curvature, any nonconstant harmonic map f : M →
N is unique in its homotopy class unless f maps onto a geodesic, in which case all
homotopic harmonic maps are translations of f along the geodesic.
A good reference for these results is Jost’s book [93].

The theorems above fail to hold if the curvature assumption on N is relaxed. In
this case the analytic complexity increases substantially, and there is no satisfactory
existence result in general. There is something of an exception in the case of surface
domains (see [95], [162], [163], [169]), where the conformal invariance with respect
to the domain metric leads to bubbling phenomena. We shall not attempt to present any
results for the case of higher dimensional domains, since the relation with Teichmüller
theory is less important.
2.2.3 Two dimensional domains. We now specialize to the case where the domain
is a Riemann surface. Here the salient feature, as we have just mentioned above, is
that the energy functional is invariant under conformal changes of metric on S, i.e.
g → eφ g. Hence, the harmonic map equations for surface domains depend only on
the complex structure on S.
Let f : (S, σ ) → (N, h) be a smooth map, where N is an arbitrary Riemannian
manifold. Then ϕ = (f ∗ h)2,0 = Hopf(f ) is a quadratic differential, called the Hopf
differential of f . A key fact is that ϕ is holomorphic if f is harmonic. Indeed, in local
coordinates, ϕ = ϕ(z)dz2 , where

ϕ(z) = fz , fz = 41 |fx |2 − |fy |2 − 2i fx , fy . (2.34)

Notice that ϕ ≡ 0 if and only if f is conformal. In normal coordinates at f (z), the
harmonic map equations are f k = 0, for all k. Together with the vanishing of the
derivatives of the metric, this implies
j
j
ϕz̄ (z) = fz , fz z̄ = hij (f (z))fzi fz z̄ = 2hij (f (z))fzi fzz̄ = 0.
Several results in this chapter depend on the holomorphicity of the Hopf differential.
In Section 2.3.3 we shall present a different argument due to Schoen [167] which
works for a more general class of metric space targets.
To see how holomorphicity can have topological consequences, take for example
the case where the target is also a Riemann surface R. Writing the metric h on R in
local conformal coordinates w, the energy of a map f is then

Eh (f ) = h(f (z))(|fz |2 + |fz̄ |2 ) dxdy, z = x + iy (2.35)
S
where we have confused the notation f and w f . When the metric h is understood,
we shall simply write E(f ). The harmonic map equations are (cf. [170, Ch. 1])
hw
fzz̄ + fz fz̄ = 0. (2.36)
h
As an immediate application of (2.36) it follows that if f : S → R is harmonic, then
¯ | are either identically zero or have a well-defined order. Indeed, if H = fz
|∂f | and |∂f
and G = −(hw / h)fz̄ , and ζ satisfies the equation ζz̄ = −G, then it is easily checked
that H eζ is holomorphic. By setting np = ordp H eζ we obtain |∂f | = |z|np k(z),
where k(z) is a smooth strictly positive function. We call np the order of |∂f | at p.
This leads to
Theorem 2.19 (Eells–Wood [52]). Let f : S → R be a harmonic map between

surfaces. If |∂f | is not identically zero, then

np = deg(f )χR − χS .
|∂f |(p)=0
¯ is not identically zero, then

If |∂u|

mp = − deg(f )χR − χS .
|∂f |(p)=0
¯ | at p, respectively. An immediate
Here np and mp are the orders of |∂f |, |∂f
consequence of this is Kneser’s theorem:
Corollary 2.20 (Kneser [106]). Let f : S → R be a continuous map between surfaces,

χR < 0. Then | deg(f )|χR ≥ χS .
Pushing these ideas further, Schoen–Yau and Sampson proved
Theorem 2.21 (Schoen–Yau [168], Sampson [164], see also Jost–Schoen [98]). Sup-
pose f : S → R is a harmonic map between surfaces of the same genus. If deg f = 1
and R has negative curvature, then f is a diffeomorphism.
Theorems 2.19 and 2.21 depend on the following formulas for a harmonic map
between surfaces.
¯ | = KR J (f ) + KS ,
log |∂f | = −KR J (f ) + KS , log |∂f (2.37)
where KS , KR are the Gaussian curvatures of S and R, and J (f ) = |∂f |2 − |∂f ¯ |2
is the Jacobian of f . The equations (2.37) are related to the Bochner formula (2.32).
The proof is a simple calculation which can be found, for example, in [170, Ch. 1].
As an application, the next theorem regarding the quotient (2.11) is due to Earle
and Eells (cf. [47] and also [48]).
Theorem 2.22. The bundle given by the quotient map p : Met hyp (S) → Thyp (S) is
trivial, i.e. there exists a homeomorphism H with the property that the diagram
Met hyp (S)p H / Thyp (S) × Diff 0 (S)

QQQ
QQQ
QQQ π
QQQ
Q(
Thyp (S)
commutes, where π is the projection onto the first factor.
The map H can be constructed as follows: fix a metric σ0 ∈ Methyp (S). For any
other σ ∈ Methyp (S), let fσ : (S, σ0 ) → (S, σ ) be the harmonic diffeomorphism ∼ id
from Theorem 2.21. Then H is defined by F (σ ) = (p(σ ), fσ−1 ).
2.2.4 A second proof of Teichmüller’s theorem. We now give a second proof that
Teichmüller space is a cell (Theorem 2.8) using harmonic maps and Hopf differentials
as opposed to Teichmüller maps. Let ϕσ = Hopf(fσ ) be the Hopf differential of
the map fσ defined above. By uniqueness of the harmonic diffeomorphism in its
homotopy class (Theorem 2.18) we obtain a well-defined map
H : Thyp (S) −→ QD(S), [σ ] → H [σ ] = ϕσ . (2.38)
Then we have
Theorem 2.23 (Wolf [196]). The map H is a diffeomorphism.
The fact that H is 1-1 is due to Sampson [164]. The smooth dependence of H
follows easily as in [51]. This seems to have been first observed also by Sampson.
That H is proper is due to Wolf. The idea is based on the following energy bound (see
also [137]):
E(fσ ) ≤ 2 |ϕσ | − 2π χS . (2.39)
S
To see this, let f : (S, σ0 ) → (S, σ ) be any quasiconformal map with Beltrami coef-
ficient μ and Hopf differential ϕ. Then
¯ |2 d vol = σfz̄ f¯z |dz|2 = σfz f¯z fz̄

|∂f |dz|2 = ϕμ|dz|2 ≤ |ϕ|,
fz
since |μ| < 1. Then since fσ has degree 1, (2.39) is a consequence of the above
inequality and (2.24). Similarly, using (2.25), one has by the same argument

2 |ϕσ | + 2π χS ≤ E(fσ ). (2.40)
S
Properness of H now follows from (2.39) and properness of the energy. The latter is
due to Schoen and Yau [169]. In Section 4.2.2 we shall sketch the proof. Finally, that
H is onto follows from the properness and the fact that both the domain and target are
smooth manifolds of dimension 6p − 6. Hence, Theorem 2.8 is a consequence of the
theorem above, along with Theorem 2.11.
2.3 Singular space targets

Harmonic maps to singular spaces were first introduced in a systematic way in the
paper of Gromov and Schoen [75] in connection with arithmetic superrigidity. Since
then the subject has played an important role in Teichmüller theory and is one of the
main themes of this review. In Section 2.3.1 we shall indicate how singular space
targets make a connection between the extremal maps discussed in Section 2.1.4 and
the harmonic maps of Section 2.2.1. The highlight is the proof of Teichmüller’s
existence and uniqueness theorems. The idea, going back to Gerstenhaber and Rauch,
provides a clear motivation for the use of singular targets from the point of view
of Teichmüller theory. We shall defer the technical aspects of the general theory to
Section 2.3.3. In Section 2.3.2 we discuss the notion of R-trees and their connection
to measured foliations and quadratic differentials. We also state the famous Hubbard–
Masur theorem. Section 2.3.3 contains all of the technical results on harmonic maps
to metric spaces that we shall need. There we give an outline of the main results of
[75], [109], [110]. In addition, we describe several results that are special to harmonic
maps to trees.
2.3.1 The Gerstenhaber–Rauch approach to Teichmüller’s extremal problem.

Teichmüller’s extremal problem (Section 2.1.2) and the energy extremal problem (Sec-
tion 2.2.1) bear obvious similarities; hence, the natural
Question. Are Teichmüller maps harmonic for some metric?
This leads to the notion of energy minimizing maps to singular space targets,
which is the subject of this section. We begin with a simple example. Given a
holomorphic quadratic differential ψ on a Riemann surface R, |ψ| defines a singular
flat metric with conical singularities at the zeros of ψ (cf. [179]). Indeed, away from
the zeros we may write |ψ| = |dw|2 for some conformal coordinate w, whereas at
a zero of order m ≥ 1, |ψ| = |w|m |dw|2 . Notice that for h(w) = |w|m , the Gauss
curvature
1
K = − log h ≤ 0, (2.41)
2h
in the sense of distributions. We say that S with the metric h = |ψ| is a nonpositively
curved space.
Let S be another Riemann surface. Given a map f : S → R one can define the
Sobolev class H 1 and the energy of f with respect to the singular conformal metric
|ψ| on R by (2.35). Following the definitions of Section 2.2.1 we call such a map
harmonic if it is an energy minimizer (see also Theorem 2.18). This is a special
case of the general theory of Gromov, Korevaar, and Schoen that we shall describe
below; in particular, such minimizers always exist and are Lipschitz by Theorem 2.31.
The following result builds on earlier, weaker versions due to Miyahara [138] and
Leite [115].
Theorem 2.24 (Kuwert [113]). A Teichmüller map f0 : S → R is the unique harmonic

map in its homotopy class when R is endowed with the singular flat metric h = |ψ|
defined by the terminal quadratic differential of f0 .
Let us show how this gives a
Proof of Teichmüller’s Uniqueness Theorem 2.4. Let f : S → R be any quasiconfor-

mal map with Beltrami differential μf . Then by (2.35) we have

E(f ) = (|fz |2 + |fz̄ |2 )|ψ(f (z))| dxdy
S
= (1 + |μf |2 )|ψ(f (z))||fz |2 dxdy
S

≤ (1 + μf 2∞ ) |∂f |2 d volS .
S
Now by (2.25), which continues to hold for the singular metric,
1
E(f ) ≤ (1 + μf 2∞ )(E(f ) + C[f ]),
2
1 + μf 2∞
E(f ) ≤ C[f ], (2.42)
1 − μf 2∞
where C[f ] is a constant depending only on the homotopy class of [f ] and the area
of the metric |ψ|. On the other hand, for the Teichmüller map f0 we have by the same
computation
1 + μf0 2∞
E(f0 ) = C[f0 ]. (2.43)
1 − μf0 2∞
If f ∼ f0 , then C[f ] = C[f0 ]. By Theorem 2.24, E(f0 ) ≤ E(f ), which by (2.42)
and (2.43) implies μf0 ∞ ≤ μf ∞ , with equality if and only if f = f0 .
This result does not answer the question of existence of extremal maps by harmonic
map methods. In their 1954 paper, Gerstenhaber and Rauch proposed a minimax
method of finding a Teichmüller map [69]. Let CM(R) denote the space of conformal
metrics on R with unit area and with at most conical singularities (see below for more
details). For each h ∈ CM(R), let Eh (f ) be defined as in (2.35), where f : S → R
is in H 1 . Gerstenhaber–Rauch conjectured that

1 ∗ 1
sup inf Eh (f ) = K [f1 ] + ∗ , (2.44)
h∈CM(R) f ∼f1 2 K [f1 ]
and that the sup-inf in (2.44) is realized by the Teichmüller map homotopic to f1 . The
problem was investigated further by Reich and Reich–Strebel in the case where S, R
are both the disk [157], [158]. Kuwert, assuming the existence of the Teichmüller
map, proved
Theorem 2.25 (Kuwert [113]). The Teichmüller map f0 and the singular metric
h0 = |ψ| defined by its terminal differential realize the sup-inf in (2.44).
The full Gerstenhaber–Rauch conjecture was recently proved by Mese (cf. [133],
[135]) using the harmonic map theory of Gromov, Korevaar, and Schoen. Before we
state Mese’s theorem we need to set up some notation and terminology. Let (X, d)
be a metric space which is also a length space, i.e. for all pairs p, q ∈ X there exists
a rectifiable curve γpq whose length equals d(p, q) (which we sometimes write dpq ).
We call γpq a geodesic from p to q. Then X is NPC (= nonpositively curved) if every
point of X is contained in a neighborhood U so that for all p, q, r ∈ U ,
dpq
2
τ
≤ (1 − τ )dpq
2
+ τ dpr
2
− τ (1 − τ )dqr
2
,
where qτ is the point on γqr so that dqqτ = τ dqr . Note that equality is achieved for
every triple p, q, r ∈ R2 . More generally, one defines a length space with curvature
bounded above by κ by making comparisons with geodesic triangles in surfaces of
constant curvature κ (cf. [132]). It follows from (2.41) that if h(w)|dw|2 is a conformal
metric on R with
log h ≥ −2κh (2.45)
then the induced metric space has curvature bounded above by κ (cf. [131]). We shall
use this fact when we give a harmonic map construction of the Teichmüller map.
Let CMα,κ (R) denote the set of metrics h = h(w)|dw|2 on R where h ≥ 0 is
bounded of Sobolev class H 1 , satisfies (2.45) weakly, and has area = α. Let dh denote
the distance function associated to the above metric. As we have discussed before it
is not hard to see that (R, dh ) has curvature bounded above by κ. The key result is the
following
Theorem 2.26 (Mese [135]). Let hi ∈ CMα,κ (R), κ > 0, and fi : S → (R, hi ) be
such that
(1) fi is harmonic;
(2) lim Ehi (fi ) = sup inf Eh (f ).
i→∞ h∈CMα,κ (R) f ∼f1
Then the fi converge in the pullback sense to the Teichmüller map f0 .

Convergence in the pullback sense is essentially Gromov–Hausdorff convergence.

This will be explained in greater detail below (see Section 3.2.3). Theorem 2.26, along
with earlier work, gives a proof of Teichmüller’s Existence Theorem 2.5.
2.3.2 R-trees. The use of singular metrics to prove the Teichmüller theorems is mo-
tivation to study energy minimizing maps for other metric space targets. Here we
discuss another ubiquitous example. An R-tree is a length space such that any two
points can be joined by a unique path parametrized by arc length. This path is called
the geodesic between the points, say p, q, and it is denoted pq. An equivalent def-
inition is that an R-tree is a simply connected length space with curvature bounded
above by κ for any κ ∈ R (cf. [180]).
Example 2.27. Let T be a simplicial tree, i.e. a simply connected 1-dimensional

simplicial complex. Then T can be thought of as an R-tree by assigning to each edge
a unit length. An R-tree is called simplicial if it is obtained from a simplicial tree in
this way. Note that we do not assume the simplicial tree is locally finite, although the
set of vertices clearly is.
Example 2.28. Take T = R2 and define d(p, q) = |p − q| if p, q lie on some ray

from the origin, and d(p, q) = |p| + |q|, otherwise. Clearly, T with this metric is not
locally compact, though it is simplicial.
Example 2.29. A slight modification of the above yields a non-simplicial tree. Again
take T = R2 and define d(p, q) = |p − q| if p and q lie on the same vertical line.
In all other cases, let d(p, q) = d(p, p ) + d(p , q ) + d(q, q ), where p , q are the
projections of p, q to the x-axis. Then every point on the x-axis becomes a vertex.
R-trees appear in Teichmüller theory in several ways. The primary example is

the leaf space of the horizontal and vertical foliations of a holomorphic quadratic
differential. First recall that a measured foliation F on a surface S with singularities
at the points z1 , . . . , z and multiplicities k1 , . . . , k is described by the following (cf.
[57]): an open cover {Ui } of S \{z1 , . . . , z } and open sets V1 , . . . , V about z1 , . . . , z
along with smooth real valued functions ui defined on Ui such that
(1) |dui | = |duj | on Ui ∩ Uj ;
(2) |dui | = | Im(z − zj )kj /2 dz| on Ui ∩ Vj .
Clearly, ker dui defines a vector field on S which integrates to give a foliation away
from {z1 , . . . , z }, with (kj + 2)-pronged singularities at zj (see Figure 1). A leaf
containing a singularity is called a critical trajectory, whereas the other leaves are
called noncritical. An important attribute of measured foliations is that they carry a
transverse measure. More precisely, if c is a rectifiable path then we denote by ν(c),
the number ν(c) = c |du|, where |du| is defined by |du|Ui = |dui |. An important
feature of this measure is its translation invariance along the leaves. Namely, if c0
is a path transverse to the foliation F , and if we deform c0 to c1 via an isotopy that
maintains the transversality to the foliation at every time, then ν(c1 ) = ν(c0 ). For the
free homotopy class [c] of a simple closed curve we define
i([c], F ) = inf{ν(c) : c ∈ [c]}. (2.46)
Two measured foliations (F , ν) and (F , ν ) are called equivalent if i([c], F ) =
i([c], F ) for all free homotopy classes of simple closed curves. We denote the space
of equivalence classes of measured foliations on S by MF(S). Then the collection of
intersection numbers (2.46), as c ranges over isotopy classes of simple closed curves,
endows MF(S) with a topology. We call F and F projectively equivalent if there
is b > 0 such that i([c], F ) = b i([c], F ) for all free homotopy classes of simple
closed curves. In this case, we write F = bF . The space of projective equivalence
classes will be denoted PMF(S).
Given a measured foliation (F , ν) we can associate a dual tree TF to the foliation
with an isometric action of = π1 (S). Explicitly, let (F˜ , ν̃) denote the pullback of
(F , ν) to the universal cover H of S. On H we define a pseudodistance d̃ via
d̃(p, q) = inf{ν̃(c) : c a rectifiable path between p, q}.
It follows by [18, Corollary 2.6] that the Hausdorffication of (H, d̃) is an R-tree
with an isometric action of . Strictly speaking, the setup in [18] works for measured
foliations on arbitrary 2-complexes. The approach is useful in that it avoids introducing
the notion of a geodesic lamination. For a proof using laminations, see [142], [153].
For a holomorphic quadratic differential ϕ = 0 on S we have seen in Section 2.1.3
how to define horizontal and vertical foliations. If the ϕ-coordinate is locally given by
w = u + iv, then transverse measures may be defined by |du| and |dv|, respectively.
In other words, a nonzero quadratic differential defines a measured foliation via its
horizontal foliation. We denote the corresponding dual tree by Tϕ .
The following fundamental theorem, due to Hubbard–Masur and also announced
by Thurston, asserts that every measured foliation on S arises in this way:
Theorem 2.30 (Hubbard–Masur [84]). Given a measured foliation (F , ν) on a closed

Riemann surface S of genus p ≥ 2 there is a unique holomorphic quadratic differ-
ential whose horizontal foliation is equivalent to (F , ν). In particular, MF(S) is
homeomorphic to R6p−6 \ {0}, and PMF(S) S 6p−6 .
In Section 3.1.2 we shall sketch how we can interpret the Hubbard–Masur theorem
via harmonic maps to trees (see [198], [199]).
There is a particular class of quadratic differentials on S called Jenkins–Strebel
differentials (cf. [179]). They are characterized by the property that the noncritical
trajectories are all closed and they partition the complement of the critical trajectories
in S into cylinders with the standard foliations (see Figure 2). Notice that in this case
the dual tree Tϕ is a simplicial tree with a action. The quotient Tϕ / is a graph
Gϕ , and the quotient map p̃ : H → Tϕ descends to a map p : S → Gϕ , as indicated
in Figure 2.
S =
Gϕ =
Figure 2
Hence, the intersection number (2.46) number is a generalization of the geometric

intersection number of simple closed curves. Let us point out two facts (cf. [57],
[179]).
• There are examples of measured foliations where all the noncritical leaves are
noncompact. For example, the fixed points in PMF(S) of pseudo-Anosov map-
ping classes. These give rise to R-trees which are not simplicial.
• However, the measured foliations whose associated trees are simplicial are dense
in MF(S). Furthermore, the intersection number (2.46) extends continuously to
MF(S) × MF(S).
Prior to a rigorous definition of energy minimizers to NPC spaces, we first introduce
the notion of a harmonic map to a tree. This definition is due to Wolf [197] and is
motivated by Ishihara’s Theorem 2.14. As we shall see in the next subsection, it turns
out that for the case of trees it is equivalent to the definition of energy minimizers due
to Korevaar–Schoen.
Let S be a Riemann surface and let (T , d) be a minimal R-tree with an isometric
action of = π1 (S). Let f : H → T be a -equivariant, continuous map. We say that
f is harmonic if it pulls back germs of convex functions on T to germs of subharmonic
functions on H. Notice that a function f : U → R, where U is a convex open subset
of an R-tree, is called convex if for any segment pq ⊂ U and r ∈ pq we have
(f (r) − f (p))d(q, r) ≤ (f (q) − f (r))d(p, r).
A basic example of a harmonic map to a tree is the projection
p : H −→ Tϕ (2.47)
where Tϕ is the dual tree to the horizontal foliation of a holomorphic quadratic differ-
ential. It is not hard to see by direct observation that p is harmonic (cf. [197]).
2.3.3 Harmonic maps to NPC spaces. For the purpose of this subsection (, g)
will be a bounded Riemannian domain of dimension m with Lipschitz boundary and
(X, d) any complete NPC space. References for the following are [75], [96], [109],
[110]. The generalization to the case where X is assumed only to have curvature
bounded from above can be found in [132].
A Borel measurable map f : → X is said to be in L2 (, X) if for p ∈ X,

d 2 (p, f (x)) d vol (x) < ∞.

By the triangle inequality, the condition is independent of the choice of point p. For
f ∈ L2 (, X) we construct an ε-approximate energy function eε (f ) : ε → R,
where ε = {x ∈ : d(x, ∂) > ε} by

1 d 2 (f (x), f (y)) dσ (y)
eε (f )(x) = ,
2ωm ∂Bε (x) ε2 εm−1
where ωm is the volume of the unit sphere in Rm and dσ is the induced volume on the
sphere ∂Bε (x) ⊂ of radius ε about x. Setting eε (f )(x) = 0 for x ∈ \ ε , we can
consider eε (f ) to be an L1 function on . In particular, it defines a linear functional
Eε : Cc () → R. We say that f has finite energy (or that f ∈ H 1 (, X)) if
E(f ) ≡ sup lim sup Eε (ϕ) < ∞.
0≤ϕ≤1 ε→0
It can be shown that if f has finite energy, the measures eε (f )(x)d vol (x) converge
weakly to a measure that is absolutely continuous with respect to Lebesgue measure
on . Therefore, there is a well-defined integrable function e(f )(x), which we call
the energy density, so that for each ϕ ∈ Cc (),

lim eε (f )(x)ϕ(x) d vol (x) = e(f )(x)ϕ(x) d vol (x).
ε→0
By analogy with the case of smooth maps we write e(f )(x) = 21 |∇f |2 (x) with total
energy

1
E(f ) = |∇f |2 d vol .
2
Similarly, the directional energy measures |f∗ (Z)|2 d vol for Z ∈ (T ) is a
Lipschitz tangent vector field can also be defined as a weak-∗ limit of measures
Z e (f ) d vol . Here,
ε
Z d 2 (f (x), f (x(x, ε))

eε (f )(x) = ,
ε2
where x(x, ε) denotes the flow along Z at time ε, starting at x. For almost all x ∈ ,

1
|∇f | (x) =
2
|f∗ (v)|2 dσ (v),
ωm S m−1
where S m−1 is the unit sphere in Tx . This definition of the Sobolev space H 1 (, X)
is consistent with the usual definition when X is a Riemannian manifold.
For any map f ∈ H 1 (, X) we can also make sense of the notion of the pullback
metric
π : (T ) × (T ) −→ L1 () (2.48)
defined by
π(V , W ) = 41 |f∗ (V + W )|2 − 41 |f∗ (V − W )|2 , V , W ∈ (T ).
If the tangent space to (, g) has a local frame (u1 , . . . , um ), we write παβ =
π(uα , uβ ), and
e(f ) = 21 |∇f |2 = 21 g αβ παβ . (2.49)
The L1 -tensor will be used in the next section to define the Hopf differential.
A finite energy map f : → X is said to be harmonic if it is locally energy
minimizing. In other words, for each point x ∈ and each neighborhood of x, all
comparison maps agreeing with f outside this neighborhood have total energy no less
than f . The following are the basic existence and regularity results. For an alternative
approach, see [96].
Theorem 2.31 (Korevaar–Schoen [109], see also [172]). Let (X, d) be an NPC space.
If f : → X is harmonic, then f is locally
√ Lipschitz continuous. The Lipschitz
constant on U ⊂⊂ is of the form C(U ) E(f ), where C(U ) is independent of the
map f (cf. Proposition 2.17).
Theorem 2.32 (Korevaar–Schoen [109]). Let (X, d) be compact and NPC. Let M be
a compact Riemannian manifold without boundary, and f : M → X a continuous
map. Then there exists a Lipschitz harmonic map homotopic to f .
Note that Theorem 2.32 is a generalization of the Eells–Sampson Theorem 2.15.

The uniqueness result in the singular case is due to Mese.
Theorem 2.33 (Mese [134]). Let M be a compact Riemannian manifold and X a

compact metric space with curvature bounded above by a constant κ < 0. If f : M →
X is a nonconstant harmonic map, then f is unique in its homotopy class unless it
maps onto a geodesic.
An important tool in understanding the structure of harmonic maps is the mono-

tonicity formula for energy minimizers. The idea goes back to Almgren [7]. The
statement is that for nonconstant energy minimizers, the quantity

Cε
ε Bε (x) |∇f |2 d vol
e , (2.50)
∂Bε (x) d (f (x), f (y))ds(y)
2
is monotone increasing in ε, for some constant C. The extension of this to singular

space targets was obtained in [75], [167], and further developed in [132]. The basic
idea is that since the derivation of the formula depends only on domain variations, and
not on any differentiability of the target space, it continues to hold for maps to metric
space targets.
The monotonicity (2.50) can be used to construct linear approximations to harmonic
maps, and in some cases further regularity can be derived. A key quantity is the
order function. Roughly speaking, the order of a harmonic map f : → X at a
point x measures the degree of the dominant homogeneous harmonic polynomial
which approximates f − f (x). This is precisely true when X is a smooth manifold.
In the general case, it is defined as follows. Define

ε Bε (x) |∇f |2 d vol
ordx (f ) = lim . (2.51)
ε↓0 ∂B (x) d 2 (f (x), f (y))ds(y)
ε
It follows from the monotonicity formula (2.50) that the above limit exists and is
≥ 1 for nonconstant maps. We call this limit the order of f at x. It is not an integer in
general. For example, let p : H → Tϕ be the projection map (2.47). If x is not a zero
of ϕ, then p is locally a harmonic function and ordx (p) is the order of vanishing. If x
is a zero of order k, then ordx (p) = (k + 2)/2. The order is related to the eigenvalues
of the Laplacian on subdomains of ∂Br (x), as explained in [75, Theorem 5.5]. In
Figure 3, ordx (p) is equal to the first Dirichlet eigenvalue of the domain Di in the
circle around x. It is clear in this case that it is equal to 3/2.
D1
•y
x p
• • p(x)
D3 D2
Figure 3
On the other hand, if y ∈ p−1 (p(x)) is not a zero, then ordy (p) = 1, and indeed
locally near y, p maps to an interval.
This can be generalized. Let f : → T be a harmonic map to an R-tree. A point
x ∈ is called regular if there exists r > 0 such that f (Br (x)) is an embedded
arc. In particular, f restricted to Br (x) is then a harmonic function. Nonregular
points are called singular. In the case of two-dimensional domains, the harmonic map
p : H → Tϕ has singularities precisely at the zeros of ϕ. In particular, they are of
codimension 2. The next result was proven in [75] for simplicial trees and in [180]
for R-trees.
Theorem 2.34. Let f : → T be a harmonic map to an R-tree. Then x ∈ is

regular if ordx (f ) = 1. Moreover, the Hausdorff codimension of the singular set is at
least 2.
X. Sun also proved the following useful fact.
Theorem 2.35 (Sun [180]). Let f : → T be a harmonic map to an R-tree. Then

for any point x ∈ there is r > 0 such that f (Br (x)) lies in a locally finite subtree.
3 Harmonic maps and representations

3.1 Equivariant harmonic maps
In this section we describe the equivariant harmonic map problem and its applications.
In Section 3.1.1 we introduce the notion of reductivity (or semisimplicity) in different
contexts and indicate how it is related to the existence problem for equivariant harmonic
maps. In Section 3.1.2 we discuss the holomorphicity of the Hopf differential for
harmonic maps and show how it can be used to simplify the proofs of the Hubbard–
Masur and Skora theorems. We also give the first variation harmonic maps with respect
to the domain metric and apply this to derive Gardiner’s formula.
3.1.1 Reductive representations. Throughout this section, unless otherwise noted,

(M, g) is a closed Riemannian manifold with = π1 (M), and (X, d) is a simply
connected NPC space. Let M̃ denote the universal cover of M. We assume that acts
on X via isometries, i.e. that there is a homomorphism ρ : → Iso(X). Associated
to ρ is a translation length function
Lρ : −→ R+ , γ → inf d(x, ρ(γ )x). (3.1)
x∈X
Let f : M̃ → X be a ρ-equivariant map. Provided that f is a locally in H 1 , the energy

density |∇f |2 is -invariant, and therefore we can define the energy by

1
E(f ) = |∇f |2 d volM . (3.2)
2 M=M̃/
Finite energy maps always exist, and indeed energy minimizing sequences can be
taken to be uniformly Lipschitz [110]. Under conditions that will be made precise
below and which we shall always assume, there exist maps with finite energy. A ρ-
equivariant map f : M̃ → X which is locally in H 1 is called harmonic if it minimizes
the energy (3.2) among all other equivariant maps in Hloc 1 .
It follows from the trace theory in [109] that equivariant harmonic maps are locally
energy minimizers. Therefore, in the case where X is a smooth manifold the first
variational formula (2.23) implies that a ρ-equivariant harmonic map is equivalent to
a smooth ρ-equivariant map that satisfies the harmonic map equations (2.22). For
general NPC targets it follows from Theorem 2.31 that ρ-equivariant harmonic maps
are Lipschitz.
The existence of equivariant harmonic maps is more complicated than in the case
of compact targets. The reason for this is that in the process of choosing an energy
minimizing sequence, e.g. using the heat flow as in the Eells–Sampson theory, the map
can “escape to infinity”, and fail to converge. An example of this phenomenon can
be found in [51]. One naturally looks for a condition on the homomorphism ρ which
rules out this kind of behavior. For example, it is reasonable to rule out the existence
of a sequence of points escaping to infinity whose translates by fixed elements in the
image of ρ remain bounded. This is the notion of a proper action (see below).
Before making this more precise we introduce the notion of the ideal boundary of
an NPC space. By a ray in X we mean a geodesic α parametrized by arc length on the
interval [0, ∞). Two rays α1 , α2 , are said to be equivalent if the Hausdorff distance
between them is finite. Denote by ∂X the set of equivalence classes of rays. Notice
that since acts by isometries, also acts on ∂X. We have the following facts:
(1) (cf. [19]) If X is locally compact then X = X ∪ ∂X can be topologized so that
it becomes a compact metric space.
(2) (cf. [32]) If (X, d) is an R-tree (not necessarily locally compact) then two rays
α1 and α2 are equivalent if and only if α1 ∩ α2 is another ray.
(3) (cf. [32]) If (X, d) is an R-tree with action, then fixes a point on ∂X if and
only if Lρ (γ ) = |r(γ )| where r : → R is a homomorphism.
We now state
Theorem 3.1 (Korevaar–Schoen [110]). Suppose ρ : → Iso(X) is a homomorphism

that does not fix a point of ∂X. If either (i) X is locally compact, or (ii) X has curvature
bounded above by κ < 0, then there exists a ρ-equivariant harmonic map f : M̃ → X.
The equivariant version of Theorem 2.33 also holds:
Theorem 3.2 (Mese [134]). If X has curvature bounded above by a constant κ < 0,
and if f : M̃ → X is a nonconstant equivariant harmonic map, then f is unique in
its equivariant homotopy class unless it maps onto a geodesic.
Special cases of Theorem 3.1 had been proven earlier:

• The Corlette–Donaldson theorem.
Theorem 3.3 (cf. [29], [30], [43]). Let X be a Riemannian symmetric space of non-
compact type X = G/K, where G is a semisimple Lie group and K a maximal
compact subgroup. Let ρ : → G be a homomorphism with Zariski dense image.
Then there is a ρ-equivariant (smooth) harmonic map f : M̃ → X.
This theorem is implied by Theorem 3.1, since if ρ() fixes a point [α] ∈ ∂X, then
ρ() would be closed subgroup contained in the stabilizer of [α], which is a proper
subgroup of G. See also [99].
• Labourie’s theorem. In the Riemannian case, the criterion for existence in terms
of fixing a point in the ideal boundary was conjectured in [29] and proved in [114]
(see also [97]). A homomorphism ρ : → Iso(X) is called semisimple (or reductive)
if either ρ() does not fix a point in ∂X or it fixes a geodesic. Then we have the
following
Theorem 3.4 (Labourie [114]). Let X be a Riemannian manifold with negative sec-
tional curvature. Then there exists a ρ-equivariant harmonic map f : M̃ → X if and
only if ρ is semisimple.
• R-trees. Let (T , d) be an R-tree and ρ : → Iso(T ) a homomorphism. We assume

(without loss of generality) that the action of on T is minimal.
Theorem 3.5 (Culler–Morgan [32]). Let ρ1 , ρ2 be nontrivial semisimple actions on R-

trees T1 , T2 with the same translation length functions. Then there exists an equivariant
isometry T1 T2 . If either action is not isometric to an action on R, then the
equivariant isometry is unique.
Then we have the following generalization of Theorem 3.4 to trees.
Theorem 3.6. Let (T , d) be a minimal R-tree and ρ : → Iso(T ). Then there exists
a ρ-equivariant harmonic map u : M̃ → T if an only if ρ is semisimple.
Proof. The sufficiency follows from Theorem 3.1 (see Section 2.3.2). For the con-
verse, suppose fixes a point in ∂T . If there is a ρ-equivariant harmonic map there
would necessarily be a family of distinct such maps (see [34]). By the uniqueness
Theorem 3.2 and the minimality of T , it follows that T in this case is equivariantly
isometric to R.
In the case where X is not locally compact, the condition of not fixing a point at
infinity does not seem to be sufficient to guarantee existence. Korevaar and Schoen
developed a slightly stronger condition to cover this case. Let ρ : → Iso(X) be a
homomorphism. To each set of generators G of we associate a function on X:

Dρ (x) = max d(x, ρ(γ )x) : γ ∈ G .
A homomorphism ρ : → Iso(X) is called proper if for every B ≥ 0, the set
{x ∈ X : Dρ (x) ≤ B} is bounded. Clearly, this condition is independent of the choice
of generating set G. For complete manifolds of nonpositive curvature, the existence
of two hyperbolic isometries in the image of ρ with nonasymptotic axes is sufficient
to prove properness. More generally, ρ being proper implies that ρ has no fixed end,
for if R is a fixed ray then Dρ is bounded along R.
Theorem 3.7 (Korevaar–Schoen [110]). Suppose ρ : → Iso(X) is proper. Then

there exists a ρ-equivariant harmonic map f : M̃ → X.
In case X is locally compact this is implied by Theorem 3.1, but for nonlocally
compact spaces it is not. Yet another sufficient condition is introduced in [111].
To end this section, we connect the definition of harmonicity given in this section
with that at the end of Section 2.3.2.
Theorem 3.8. Let S be a Riemann surface, (T , d) an R-tree, and ρ : = π1 (S) →

Iso(T ) a reductive action. A ρ-equivariant map f : M̃ → T is harmonic if and only
if f pulls back germs of convex functions to germs of subharmonic functions.
Proof. The fact that harmonic maps pull back functions to subharmonic ones is the
content of [56, Prop. 3.2] (see also [109]). For the converse, we argue as follows:
suppose f : H → T is a ρ-equivariant map that pulls back germs of convex functions
to subharmonic ones. Let f : H → T be a ρ-equivariant harmonic map. Since
both f , f pull back germs of convex functions to subharmonic functions, it follows
that the same is true for f × f : H → T × T . Hence, d(f, f ) is -equivariant
and subharmonic, hence constant. But because of the 1-dimensionality of trees it is
easy to see that the energy densities of f and f must be equal, so that f is energy
minimizing.
So far as we know, this result for general NPC targets is open.
3.1.2 Measured foliations and Hopf differentials. Recall from Section 2.3.3 that
if X is a metric space target and f : (M, g) → (X, d) is a finite energy map, then
one can associate an integrable symmetric 2-tensor παβ on S with the property that
the energy density |∇f |2 = g αβ παβ . Hence, while the energy density may not be
the square of the norm of a derivative, it is a trace of directional energies. Let us
specialize to the case where the domain is a Riemann surface, and let f be an energy
minimizer. By varying among finite energy maps obtained from pulling f back by a
local diffeomorphism defined by a vector field v, we arrive at

0= π, Lv g − (1/2) Tr g (Lv g)g d volM .
M
Note that the integrand is well defined since π is integrable. By a particular choice
of v, and using Weyl’s lemma on integrable weakly holomorphic functions, we obtain
[167] that
ϕ(z)dz2 = 41 (π11 − π22 − 2iπ12 )dz2
is a holomorphic quadratic differential on S (cf. Section 2.2.2). We call ϕ the Hopf
differential of f . Since these computations are local, they apply as well to the case of
equivariant harmonic maps.
Thus far we have seen that a measured foliation F on a surface S gives rise to an
R-tree TF with an isometric action of . This action has the following properties:
(1) the action is minimal in the sense that no proper subtree is invariant under (cf.
[153] – strictly speaking, the proof there uses geodesic laminations but it can be
easily adapted to the case of measured foliations);
(2) the action is small in the sense that the edge stabilizer subgroups do not contain
free groups on 2-generators (cf. [140] – more precisely, the stabilizers are cyclic,
since leaves on the quotient surface are either lines or circles).
Shalen conjectured [173] that every minimal, small action of a surface group on
an R-tree is dual to a measured foliation. This conjecture, which plays an important
role in Thurston’s hyperbolization theorem for fibered 3-manifolds (see [153]), was
proved by R. Skora, building upon previous work of Morgan–Otal [140].
Theorem 3.9 (Skora [178]). Let S be a surface of genus at least 2. Then if (T , d) is

an R-tree with a minimal, small isometric action of = π1 (S), there is a measured
foliation F on S such that (T , d) is equivariantly isometric to TF .
For example, we have seen that if (T , d) is dual to a measured foliation on S then

the action is small. It is also a simple matter to see that a small action is semisimple.
Indeed, choose γi ∈ , i = 1, . . . , 4 such that the commutators [γ1 , γ2 ] and [γ3 , γ4 ]
generate a group G containing a free group on 2-generators. Then if ρ had a fixed
end, then ρ(γi ) would act by translations along a common ray. In particular, ρ(G)
would stabilize this ray, contradicting the assumption of smallness. Hence, small
actions are semisimple, and by Theorem 3.6 there exists a ρ-equivariant harmonic
map f : H → T .
In general, let ϕ̃ be the Hopf differential of an equivariant harmonic map. Then
ϕ̃ is the lift of a holomorphic quadratic differential ϕ on S. Let Tϕ denote the dual
tree to the vertical foliation of ϕ̃. It is not hard to see (cf. [35], [56]) that there is a
-equivariant map F : Tϕ → T such that the following diagram commutes
p
Hf / Tϕ
AA
AA
AA (3.3)
AA F
AA
A
T
where p : H → Tϕ is the natural projection. Moreover, this is a morphism of trees,

meaning that any segment xy ∈ Tϕ decomposes into a finite union of subsegments
along which p is an isometry. By [140], it follows that F is either an equivariant
isometry, or F folds at some point. This means there is an identification of two or
more segments z y and z y in Tϕ to a single segment zy in T . An example of a
folding is shown in Figure 4.
y
• x • y
x x F
y
Figure 4
Interestingly, the harmonicity of the map f precludes a whole class of undesirable

foldings. For example, the following is a consequence of the maximum principle.
Proposition 3.10 ([35], [56]). Suppose Tϕ → T arises from an equivariant harmonic

map to T , as in (3.3). Then folding occurs only at points in Tϕ corresponding to zeros
of ϕ̃ of multiplicity at least two. Moreover, adjacent edges may not be identified under
such a folding.
This type of resolution of the tree T by the dual tree Tϕ to a measured foliation,
with the folding properties of the proposition, had been obtained by Morgan–Otal in
[140], and it is the first step in proving Skora’s Theorem 3.9.
By an ingenious counting argument using interval exchanges, Skora went on to
show that provided the action of is small, folding at vertices cannot occur either,
and in fact F is an isometry. This completes the proof of Theorems 2.30 and 3.9.
An alternative source for the counting argument is [153, §8.4]. The reader may also
consult [56], [198], [199].
As a second application, consider a measured foliation F on a Riemann surface
(S, σ ). We have seen above that there is a unique holomorphic quadratic differential
ϕF = ϕ(σ, F ) whose horizontal foliation is measure equivalent to F . The extremal
length of F is defined by
ExtF [σ ] = |ϕF | (3.4)
S
and is a well-defined function on T (S). It is a generalization of the extremal length of
a simple closed curve to the case of arbitrary measured foliations. In [64], Gardiner
gave a formula for the first variation of ExtF [σ ]. Here we show how this formula
arises naturally as the variation of the energy of harmonic maps.
Let X be an NPC space, and suppose ρ : → Iso(X) is proper in the sense of
Section 3.1.1. For simplicity, assume X has curvature bounded above by some κ < 0.
Then for each complex structure j on S, Theorems 3.7 and 3.2 guarantee the existence
of a unique ρ-equivariant harmonic map f : S̃ → X. The energy of f = f[j ],ρ gives
a well-defined function depending upon [j ] and ρ:
Eρ− : T (S) → R+ , [j ] → E(f[j ],ρ ). (3.5)

Theorem 3.11. The function Eρ− is differentiable on T (S). If σt , −1 ≤ t ≤ 1, is a

differentiable family of metrics on S with Beltrami differential μ at t = 0, and ϕ is
the Hopf differential of a ρ-equivariant energy minimizer (S̃, σ0 ) → X, then

d
Eρ− [σt ] = −4 Reμ, ϕ, (3.6)
dt
t=0
where the pairing is as in (2.16).
In the case where X is a smooth Riemannian manifold, this formula has some
history. Wolf [200] provides a derivation and refers to earlier notes of Schoen, as well
as [185], [186], [95]. The earliest computation of this sort may be due to Douglas (cf.
[45, eq. (12.29)]). Formally, the proof of (3.6) goes as follows. The total energy is
the contraction of the energy density tensor πij with the metric on S. Hence, the first
variation involves varying first πij , i.e. the harmonic map, and then the metric. But
the term associated to the variation of the map is necessarily zero, since the map is
energy minimizing. It follows that the only contribution comes from variations with
respect to the metric. Formula (3.6) then follows easily. Some care must be taken to
justify this in the case of metric space targets (see [194]).
Now consider a measured foliation F on S with associated dual tree TF . The
energy of the unique equivariant harmonic map f : S̃ → TF is precisely the extremal
length ExtF . From (3.6) we have
Theorem 3.12 (Gardiner [64], [65]). For any measured foliation F , ExtF is differ-
entiable on T (S) with derivative

d
ExtF [σt ] = 2 Reμ, ϕF .
dt t=0
Here, ϕF is the Hubbard–Masur differential for F at σ0 .
3.2 Higgs bundles and character varieties

This section discusses the relationship between character varieties and certain special
metrics on holomorphic vector bundles. The link between these two comes via the
equivariant harmonic map problem of the previous section. In Section 3.2.1 we intro-
duce the notion of a Higgs bundle and discuss the correspondence between stable Higgs
bundles, the self-duality equations, and flat SL(2, C) connections. In Section 3.2.2
we give a Higgs bundle interpretation of the Teichmüller space and another proof of
Theorem 2.8 using the self-duality equations. Finally, in Section 3.2.3, we discuss the
notion of convergence in the pullback sense and give a harmonic maps interpretation
of the Morgan–Shalen-Thurston compactification of character varieties.
3.2.1 Stability and the Hitchin–Simpson theorem. By a Higgs bundle on a Rie-

mann surface S we mean a pair (V , ), where V → S is a holomorphic vector bundle
and is a holomorphic section of the associated bundle End(V ) ⊗ KS . Two Higgs

bundles (V , ), (V , ) are isomorphic if there exists an isomorphism ı : V → V
of holomorphic structures such ı = ı .
Recall that a complex bundle has a well-defined degree,

deg(V ) = c1 (V ),
S
where c1 (V ) denotes the first Chern class. The slope is defined by
slope(V ) = deg(V )/ rk(V ),
where rk(V ) is the rank of V . A Higgs bundle (V , ) is called stable if slope(V ) <
slope(V ) for all nontrivial -invariant proper subbundles V ⊂ V , i.e. V = 0, V and
(V ) ⊂ V ⊗ KS . A Higgs bundle is called polystable if can be written as a direct
sum of stable Higgs bundles.
Given a hermitian metric H on a holomorphic bundle V we shall denote the Chern
connection by ∇H , i.e. ∇H is the unique connection compatible with H and the
holomorphic structure (cf. [26]). The curvature F∇H takes values in ad(V ) ⊗ 2 (S),
where ad(V ) ⊂ End(V ) is the bundle of skew-hermitian endomorphisms. Let ω be a
Kähler form on S normalized so that S ω = 1. The following result is due to Hitchin,
who first introduced Higgs bundles in this form [81], [83]. The result for higher
dimensional Kähler manifolds is due to Simpson [174]. The case ≡ 0 corresponds
to stable bundles on Riemann surfaces and was proved first by Narasimhan–Seshadri
[148] and later, using very different methods, by Donaldson [42]. Higher dimensional
versions of the Narasimhan–Seshadri theorem were obtained by Donaldson [43] and
Uhlenbeck–Yau [189].
Theorem 3.13 (Hitchin, Simpson). Let (V , ) be a Higgs bundle on a closed Riemann

surface S. Then (V , ) is polystable if and only if there exists a hermitian metric H
on V solving the self-duality equations
i
F∇ + [, ∗H ] = s I ⊗ ω, (3.7)
2π H
where ∗H is the adjoint of with respect to H , and s = slope(V ). Furthermore, H
is unique up to scalars.
From both the algebro-geometric and topological points of view, it is preferable to

fix determinants. In other words, fix a holomorphic line bundle L → S with hermitian
metric h such that det(V ) = L, and let ∈ H 0 (End0 V ⊗ KS ), where End0 (V ) is
the bundle of traceless endomorphisms. We shall call (V , ) a Higgs bundle of fixed
determinant L.
Corollary 3.14. A Higgs bundle (V , ) of fixed determinant L is polystable if and

only if there exists a hermitian metric H on V with det H = h and such that (3.7)
holds. In the case such an H exists, it is unique.
Following Corlette [29] we call a flat SL(r, C) connection ∇ on the trivial rank r
bundle on S reductive if any ∇-invariant subbundle has a ∇-invariant complement.
Clearly, a reductive flat SL(r, C) connection is a direct sum of irreducible flat SL(r , C)
connections for values r < r.
Define M(S, r) to be the moduli space of isomorphism classes of polystable Higgs
bundles on S of rank r and fixed trivial determinant. We denote the space of equivalence
classes of reductive flat SL(r, C) connections on the trivial rank r bundle V → S by
χ (, r). We have the following
Theorem 3.15 (Corlette, Donaldson). The map

: M(S, r) → χ (, r)
(V , ) → ∇H + + ∗H ,
is a bijection, where H satisfies (3.7).
That is well defined follows from Corollary 3.14, and the injectivity is a conse-
quence of the uniqueness of the solution H . The surjectivity part was first conjectured
by Hitchin in [81] and was subsequently proven for rank 2 by Donaldson [44] and in
general by Corlette [29]. It is equivalent to the Corlette–Donaldson Theorem 3.3 on
equivariant harmonic maps discussed above.
Indeed, given a reductive flat connection ∇, let ρ : → SL(r, C) denote its holon-
omy representation. Since reductive representations split into irreducible factors, we
may assume without loss of generality that ρ is irreducible. By Theorem 3.3 there
exists an equivariant harmonic map H : H → SL(r, C)/ SU(r). Equivalently, we can
view H as a section of the “twisted bundle”
H ×ρ SL(r, C)/ SU(r) −→ S,
i.e. H is nothing but the choice of a hermitian metric on V . Therefore, we can split
∇ = ∇H + + ∗H , where ∇H is a hermitian connection with respect to H , and
is a smooth section of End0 (V ) ⊗ KS . Clearly, the flatness of ∇ is equivalent to the
equations (3.7) together with the Bianchi identity
d∇H ( + ∗H ) = 0. (3.8)
The harmonicity of H is equivalent to the condition [29], [44]
(d∇H )∗H ( + ∗H ) = 0. (3.9)
Conditions (3.8) and (3.9) are together equivalent to the holomorphicity of . Hence,
V with the induced holomorphic structure from ∇H and define a Higgs bundle with
(V , ) = ∇.
Notice that in the above argument we indicated that the existence of a ρ-equivariant
harmonic map H → SL(r, C)/ SU(r) was equivalent to the reductivity of the flat
connection ∇. Therefore, by Labourie’s Theorem 3.4, it is also equivalent to the
reductivity of the holonomy representation ρ : → SL(2, C), in the case of H3 =
SL(2, C)/ SU(2).
The question of the complex structure on the spaces M(S, r) χ (, r), originally
addressed by Hitchin [81], is an extremely interesting one. As a character variety,
χ (, r) is an affine algebraic variety. For example, given any γ ∈ we define a regular
function τγ : χ(, 2) → C by τγ [ρ] = Tr ρ(γ ). Here, [ρ] denotes the conjugacy class
of representations containing ρ. By [33] the ring generated by all elements τγ , γ ∈ ,
is finitely generated. Fix a generating set associated to {γ1 , . . . , γm }, and define
t : χ(, 2) −→ Cm , [ρ] → (τγ1 (ρ), . . . , τγm (ρ)).
Then t is a bijection onto its image and gives χ (, 2) the structure of an affine variety.
For higher rank, one needs to consider other invariant polynomials in addition to
traces. On the other hand, Nitsure and Simpson have shown that M(S, r) with its
complex structure induced as a moduli space over the Riemann surface S has the
structure of a quasiprojective algebraic scheme [151], [175], [176]. The bijection
M(S, r) χ(, r) is not complex analytic. On the contrary, Hitchin shows that the
two complex structures are part of a hyperkähler family. For more details, we refer to
[81], [175].
A consequence of the realization of χ(, r) as a moduli space of Higgs bundles is
that there is a natural C∗ -action. Indeed, if (V , ) is a polystable Higgs bundle then so
is (V , t), t ∈ C∗ . This defines a holomorphic action on M(S, r), and therefore also
an action (not holomorphic) on χ(, r). This action depends on the complex structure
on S and is not apparent from the point of view of representations. Nevertheless, we
shall see in the next section that it has some connection with Teichmüller theory.
3.2.2 Higgs bundle proof of Teichmüller’s theorem. For the purposes of this sec-
tion we specialize to the case r = 2 and set M(S) = M(S, 2) and χ () = χ (, 2).
Define the Hitchin map
1
det : M(S) −→ QD(S), [V , ] → det = − Tr 2 . (3.10)
2
Hitchin proved that det is a proper, surjective map with generic fibers being half-
dimensional tori. This last property in fact realizes M(S) as a completely integrable
system (see [82]). More importantly for us, notice that under the Corlette–Donaldson
correspondence : M(S) → χ(), det −1 is just the Hopf differential of the
associated harmonic map (cf. [35]). Indeed, for [ρ] ∈ χ () with an associated
equivariant harmonic map fρ , ϕρ = Hopf(fρ ) is given by
ϕρ = ∇fρ1,0 , ∇fρ1,0 = − Tr(∇fρ1,0 )2 = 2 det −1 [ρ].
In order to realize the Teichmüller space inside M(S), let ı : M(S) → M(S) denote
the involution ı(V , ) = (V , −). Notice that ı is a restriction of the full C∗ -action on
M(S) described at the end of the previous section. Also notice that under the Corlette–
Donaldson correspondence , ı corresponds to complex conjugation. Hence, the fixed
points of ı are either SU(2) or SL(2, R) representations. The former correspond under
the Narasimhan–Seshadri theorem to the Higgs pair (V , 0), i.e. ≡ 0. If (V , ) is a
fixed point of ı with ≡ 0, Hitchin shows that V must be a split holomorphic bundle
L ⊕ L∗ , and with respect to this splitting is of the form

0 a
= ,
b 0
where a ∈ H 0 (S, L2 ⊗ KS ), and b ∈ H 0 (S, L−2 ⊗ KS ). Stability implies b = 0, and
hence by vanishing of cohomology, deg L ≤ p − 1, where p is the genus of S. This
fact, as pointed out in [81], turns out to be equivalent to the Milnor–Wood inequality
which states that the Euler class of any PSL(2, R) bundle on S is ≤ 2p − 2 (cf. [136],
[208]).
We next restrict ourselves to the components of the fixed point set of ı corresponding
to line bundles L of maximal degree p − 1. In this case, L must be a spin structure,
i.e. L2 = KS , for otherwise b = 0, contradicting stability. We denote this moduli
space by NL (S). After normalizing by automorphisms of L ⊕ L−1 , we can write

0 a
= ,
1 0
for some quadratic differential a ∈ QD(S). It follows that the restriction of the Hitchin
map to NL (S) defines a homeomorphism det : NL (S) −− ∼
→ QD(S).
The following gives another proof of Theorem 2.8.
Theorem 3.16 (Hitchin). Given a Higgs bundle

−1 0 a
L⊕L ,
1 0
in NL (S), let H denote the metric on L ⊕ L−1 solving the self-duality equations, and
let h be the induced metric on KS−1 = L−2 = T 1,0 S. Then
(1) the tensor
ĥ = a + (h + h−1 aa) + a ∈ 0 (S, Sym2 (T ∗ S) ⊗ C)
is a Riemannian metric on S of constant curvature −4;
(2) any metric of constant curvature −4 on S is isometric to one of this form for
some a ∈ QD(S).
The new ingredient in this theorem is the use of the existence of solutions to the self-
duality equations (3.7). Notice that in the reducible case described in Theorem 3.13 the
self-duality equations reduce to the abelian vortex equations Fh = −2(1 − a2L2 )ω
(cf. [92]). The relation between the vortex equations and curvature of metrics on
surfaces had been noted previously in the work of Kazdan and Warner [103].
Notice that the definition of NL (S) depends on a choice of spin structure L, and
there are #H 1 (S, Z2 ) = 22p such choices. This reflects the fact that on χ () there is
2p
an action of Z2 , and the quotient is
2p
χ()/Z2 = Hom(, PSL(2, C))// PSL(2, C),
the character variety of PSL(2, C), of which the Fricke space F(S) is a natural subset.
The preimage of F(S) in χ() is the disjoint union of the NL (S), and each of these is
homeomorphic to Teichmüller space.
3.2.3 The Thurston–Morgan–Shalen compactification. Let us first explain the no-

tion of convergence in the pullback sense, due to Korevaar–Schoen, that appears
in the statement of Theorem 2.26. Let be a set and f : → X a map into
a simply connected NPC space (X, d). Use f to define a pseudometric on ,
df (x, y) = d(f (x), f (y)), x, y ∈ . To obtain convergence in an NPC setting,
some convexity is needed. This is achieved by enlarging to a space ∞ , defined
recursively by:
0 = ,
k+1 = k × k × [0, 1],
∞

∞ = k ∼,
k=0
where the identification ∼ is generated by an inclusion k → k+1 , x → (x, x, 0).
The map f extends to ∞ recursively by setting f (x, y, t), where x, y ∈ k+1 ,
equal to the point on the geodesic t of the way from f (x) to f (y). Let d∞ denote
the pullback pseudometric on ∞ . After identifying points of zero pseudodistance in
(∞ , d∞ ) and completing, one obtains a metric space (Z, dZ ) isometric to the closed
convex hull C(f ()) ⊂ X (see [110]).
Given a sequence fi : → Xi of maps into simply connected NPC spaces Xi , we
say that fi → f in the pullback sense if the pullback pseudodistances di,∞ on ∞
converge locally uniformly to a pseudometric d∞ , and if the map f is the quotient
→ ∞ → (Z, dZ ).
This notion is equivalent to Gromov–Hausdorff convergence (cf. [197]). Indeed,
(uniform) Gromov–Hausdorff convergence (Zi , di ) → (Z, dZ ) means that for any
ε > 0 there are relations Ri ⊂ Zi × Z whose projections surject onto Zi and Z, and
such that if (zi , z), (zi , z ) ∈ Ri , then
|di (zi , zi ) − dZ (z, z )| < ε.
Convergence of the maps fi : → Zi to f : → Z imposes the additional re-
quirement that (fi (x), f (x)) ∈ Ri for all x ∈ . It is easy to see that fi : → Xi
converges in the pullback sense if and only if the convex hulls Zi = C(fi ()) and the
maps fi converge in the Gromov–Hausdorff sense. Indeed, pulling everything back
to ∞ , the relations Ri can be taken to be the diagonal. We also point out that it is
easy to extend these notions equivariantly in the presence of isometric group actions.
We have the following compactness property:
Proposition 3.17. Let be a metric space, and let fi : → Xi be a sequence

of maps into NPC spaces such there is a uniform modulus of continuity: i.e. for
each x ∈ there is a monotone function ωx so that limR→0 ωx (R) = 0, and

maxy∈B(x,R) dfi (x, y) ≤ ωx (R). Then fi converges (after passing to a subsequence)
in the pullback sense to a map f : → Z, where Z is an NPC space.
We call the NPC space (Z, dZ ) a Korevaar–Schoen limit. Strictly speaking, the
target surfaces (R, hi ) in Theorem 2.26 are not simply connected and are not NPC.
To deal with the former, consider equivariant convergence of the lifts to the universal
covers as mentioned above. For the latter, one shows that under the assumption that
curvature is bounded from above, geodesics are locally unique, so the construction of
∞ above works at a local level.
As usual, we denote by the fundamental group of a hyperbolic surface. Let
χ () be the SL(2, C)-character variety of . As we have seen, χ () is a noncompact
algebraic variety. In this section we describe a construction, introduced by Thurston
in the case of SL(2, R) representations, to compactify χ (). It is important to note
that this is not a compactification in an algebro-geometric sense, and indeed χ () will
not be an complex analytic space.
Let C denote the set of conjugacy classes of , and let

P(C) = [0, ∞)C \ {0} R+ ,
where R+ acts by homotheties. Topologize P(C) with the product topology. We define
a map

ϑ : χ() −→ P(C), [ρ] → log(| Tr ρ(γ )| + 2) γ ∈C
The purpose of the “+2” in the formula is to truncate the logarithm so that it goes
to infinity only when the trace goes to infinity. It is easy to see (cf. [28]) that
log(| Tr ρ(γ )| + 2) is asymptotic to Lρ (γ ), where Lρ is the translation length function
of ρ acting on hyperbolic space H3 (see (3.1) and recall that Iso(H3 ) = PSL(2, C)).
In case ρ is a discrete faithful SL(2, R) representation, hence defining an element of
Teichmüller space, Lρ (γ ) is just the length of the closed geodesic in the hyperbolic
surface S = H/ρ() in the free homotopy class of γ .
Next, recall from Section 3.2.1 that by definition of the affine variety structure on
χ (), coordinate functions are of the form τγ , where τγ (ρ) = Tr ρ(γ ). Hence, τγ ,
γ ∈ C generate the coordinate ring of χ() as a C-algebra, and it follows that ϑ is a
continuous injection. Define χ() to be the closure of the image of ϑ as a subset of
P(C). It follows, essentially from the finite generation of the coordinate ring of χ (),
that χ () is compact (cf. [141]). We call χ () the Morgan–Shalen compactification
of χ (), and set ∂χ() = χ() \ χ() to be the set of ideal points. The really useful
ingredient in this construction is that the ideal points are not arbitrary but are translation
length functions for isometric actions of on R-trees. Another important property
is the following: the group Aut() of automorphisms of clearly acts continuously
on χ () and this action admits a continuous extension to χ () (this is essentially the
action of the mapping class group to be discussed in Section 4.1.2 below).
Let T (S) ⊂ χ() denote the Teichmüller space, viewed as a component of the
th
discrete faithful representations. The closure T (S) of T (S) in χ () is called
th
the Thurston compactification of T (S) and ∂T th (S) = T (S) \ T (S) is called
the Thurston boundary of T (S). The action of Aut() extends continuously to
th
T (S) ⊂ T (S), and indeed this was part of the motivation for Thurston’s com-
pactification.
In terms of a finite set γ1 , . . . , γm ∈ , where {τγi } generate the coordinate ring
of χ (), we can rephrase the compactness of χ () as follows. Given a sequence of
representations ρi : → SL(2, C), only one of the following can occur:
(1) For some subsequence {i }, all traces ρi (γj ), j = 1, . . . , m, are bounded (in this
case, we call the sequence ρi bounded). Then [ρi ] converges (after possibly
passing to a further subsequence) in χ ().
(2) For any subsequence {i } there is some s = 1, . . . , m such that Tr ρi (γs ) → ∞
as i → ∞. Then there is a function : C → R+ , = 0, such that (after possibly
passing to a further subsequence) ρi → , projectively.
In terms of the relationship between representations and equivariant harmonic maps
we have the following simple but important observation (cf. [34]):
Proposition 3.18. A sequence of representations ρi : → SL(2, C) with associated

ρi -equivariant harmonic maps fi : H → H3 is bounded (up to conjugation) if and
only if the energy of the harmonic maps fi is uniformly bounded.
We now assume that ρi is an unbounded sequence of representations with fi as

above. Consider the sequence of ρi -equivariant harmonic maps
fî : H −→ (H3 , di ), (3.11)
where the hyperbolic metric d on H3 is scaled by the square-root of the energy:
di (x, y) = d(x, y)/E 1/2 (fi ), and fî = fi . Then because of the scaling the fî have
uniform modulus of continuity. Furthermore, by properties of thin triangles in H3 and
the fact that E(fi ) → ∞, one can see that geodesic triangles in the convex hull of the
image of fî become infinitely thin (cf. [16], [156]).
Figure 5
Using these ideas we have

Theorem 3.19 (Daskalopoulos–Dostoglou–Wentworth [34]). For an unbounded se-

quence of irreducible SL(2, C) representations ρi the corresponding harmonic maps
fî in (3.11) converge (after possibly passing to a subsequence) in the pullback sense
to a -equivariant harmonic map fˆ : H → X, where X is an R-tree with isometric
action such that
(1) acts on X without fixed points;
(2) the length function of the action of on X is in the projective class of the
Morgan–Shalen limit of the sequence ρi ;
(3) the image of fˆ is a minimal tree.
Let χdf () ⊂ χ() denote the subspace of discrete faithful representations. It
is a consequence of Jorgenson type inequalities (cf. [141]) that the Morgan–Shalen
limit of a sequence of discrete faithful representations is the length function of a small
action on an R-tree. By Skora’s Theorem 3.9, the tree is dual to a measured foliation,
and therefore ∂χdf () ⊂ PMF(S). We actually have
∂χdf () = PMF(S) QD(S), (3.12)
∂T th (S) = PMF(S) QD(S). (3.13)
The second equality (3.13), first proven by Thurston using the density of Jenkins–
Strebel differentials, was also proven by Wolf using harmonic maps [196]. We show
how this result follows from the discussion above. Recall from Theorem 2.23 that the
map H : Thyp (S) → QD(S) defined in (2.38) is a homeomorphism. Choose tj → ∞
and a sequence {ϕj } ∈ QD(S), ϕj 1 = 1. Without loss of generality, we may assume
{ϕj } converges to some nonzero ϕ ∈ QD(S). Let [σj ] = H −1 (tj ϕj ), and let fj be the
associated harmonic maps. By definition, the Hopf differentials of the rescaled maps
fˆj converge to (1/2)ϕ. Indeed, Hopf(fˆj ) = tj ϕj /E(fj ), and by (2.39) and (2.40),
E(fj ) ∼ 2tj . On the other hand, the Hopf differentials of fˆj converge to the Hopf
differential of the limiting equivariant map H → X. By the smallness of the action of
on X, Skora’s theorem implies that X is dual to a measured foliation. This measured
foliation must coincide with the horizontal foliation of (1/2)ϕ. Hence, we have shown
the equality (3.13) and that the map H defined in (2.38) extends continuously as a map
from ∂T (S) to the sphere at infinity in QD(S). Equality (3.12) follows from (3.13)
and the fact that ∂T th (S) ⊂ ∂χdf () ⊂ PMF(S).
4 Weil–Petersson geometry and mapping class groups

4.1 Weil–Petersson geodesics and isometries
Teichmüller space has a length space structure given by the Teichmüller distance (2.9).
An alternative Riemannian structure arises from the description of Teichmüller space
via hyperbolic metrics presented in Section 2.1.5. This is the Weil–Petersson metric,
and its properties continue to be the subject of much research. In this section, we
present a short review of some of the aspects of Weil–Petersson geometry that will
be relevant later on. The basic definitions as well as properties of the Weil–Petersson
completion are discussed in Section 4.1.1. In Section 4.1.2 we introduce the mapping
class group, and in Section 4.1.3 we indicate how the classification of individual
mapping classes follows from the structure of Weil–Petersson geodesics.
4.1.1 The Weil–Petersson metric and its completion. Recall from Section 2.1.5
that the cotangent space T[σ∗ ] T (S) is identified with the space of holomorphic quadratic
differentials on (S, σ ). The complete hyperbolic metric on (S, σ ) can be expressed in
local conformal coordinates as ds 2 = σ (z)|dz|2 . Similarly, a quadratic differential has
a local expression ϕ = ϕ(z)dz2 . Then for ϕ ∈ T[σ∗ ] T (S), the Weil–Petersson cometric
is given by ϕwp = ϕ2 (see (2.17)). While there exist a wide variety of invariant
metrics, the Weil–Petersson metric is in a real sense the most useful for applications.
We refer the reader to Wolpert’s recent survey [206]. The two most important facts
for us here are that (1) the Weil–Petersson metric has negative sectional curvature and
(2) it is incomplete.
The curvature properties of Teichmüller space with the Weil–Petersson or Teich-
müller metrics have an interesting history. It was long thought that the Teichmüller
metric had negative curvature in the sense of triangle comparisons (see [112]). This
was disproven by Masur in [123] (see also [119], and more recently [91], [129], [130],
[125]). For the Weil–Petersson metric, the first step was taken by Ahlfors [4], who
showed that the first variation of the area element induced by the hyperbolic metric
vanishes. This implies the kählerity. He also established the negativity of the Ricci
and holomorphic sectional curvatures. The following result was established later:
Theorem 4.1 (Tromba [183], Wolpert [203], see also [95], [177].). The curvature of
the Weil–Petersson metric has
(1) holomorphic sectional curvatures and Ricci curvatures bounded above by
−1/2π(p − 1), and
(2) negative sectional curvature.
Incompleteness is a consequence of the nature of degenerating Riemann surfaces.

This was first recognized in the work of Bers, Chu, Wolpert and Masur (cf. [27], [124],
[202]). A model for degeneration is given by the “plumbing construction”. Here is
a simple version: let S1 and S2 be compact surfaces of genera p1 , p2 . Choose local
coordinates z1 , z2 centered at points x1 ∈ S1 , x2 ∈ S2 . Fix 0 < t < 1 and construct a
new surface from the following three pieces: S1 \ {|z1 | ≤ 1}, S2 \ {|z2 | ≤ 1}, and the
annulus {(z1 , z2 ) : z1 z2 = t}. The boundary of the annulus is identified in the obvious
way with the boundaries of the surfaces with disks deleted. In this way, one obtains a
compact Riemann surface St of genus p1 + p2 .
S1 S2
St =
Figure 6
As t → 0, the points in Teichmüller space corresponding to St diverge, because

the annulus is begin “pinched”. This can also be seen from the hyperbolic geometry.
Using the maximal principle, one can approximate the behavior of the hyperbolic
metric on St (cf. [205]). In the pinching region, it is roughly approximated by the
hyperbolic metric on the annulus given by
|dz|2 2t
dst2 = ,
|z|2 (log |z|)2 (sin t )2
where t = π log |z|/ log |t|, and z is either z1 or z2 . As a result the length of the
“waist” of the annulus is shrinking to zero as t → 0. In fact, the length is of order
∼ 1/ log(1/t) (see [205]). Notice that every curve passing through the annulus
must then become rather long. This is a general fact in the hyperbolic geometry of
surfaces. The following rough statement of the Collar Lemma indicates that around
short geodesics on a hyperbolic surface one always can find long cylinders. For a
more precise statement, see [104].
Lemma 4.2. Let (S, σ ) be a hyperbolic surface and c a simple closed geodesic of
length = 0. Then any simple closed essential curve having nonzero geometric
intersection with c has length on the order ∼ log(1/).
The behavior of the Weil–Petersson metric at points in T (S) described by these

degenerations has the following model due to Masur, Yamada, and Wolpert (for a
review, see [206]). Define an incomplete metric space

M = (ξ, θ) ∈ R2 : ξ > 0 , dsM 2
= 4dξ 2 + ξ 6 dθ 2 . (4.1)
The metric completion M of M is obtained by adding a single point ∂ M corresponding
to the entire real axis ξ = 0. The completion is then an NPC space which is, however,
not locally compact. Indeed, an ε-neighborhood of ∂ M contains all points of the form
(ξ, θ ), ξ < ε, and θ arbitrary.
The importance of M is that it is a model for the normal space to the boundary
strata. Let T (S) denote the metric completion of T (S). We have the following local
description (cf. [124]): ∂T (S) = T (S)\T (S) is a disjoint union of smooth connected
strata formed by collapsing a collection of disjoint simple closed essential curves on
S to points. Associated to the nodal surface is another Teichmüller space which is
by definition the set of equivalence classes of complex structures on the normalized

(possibly disconnected) surface, with the preimages of the nodes as additional marked
points. It is therefore naturally isomorphic to a product of lower dimensional Teich-
müller spaces. A neighborhood of a point in the boundary is then homeomorphic to
an open set in the lower dimensional product crossed with as many multiples of M as
there are collapsed curves. Metrically, the statement is that the Weil–Petersson metric
in this neighborhood is equal to the product metric up to third order in the ξ variables
(see [39], [124], [152], [205], [206], [201], [210]). Moreover, by Wolpert’s theorem
(see Theorem 4.16 below) T (S) with the Weil–Petersson metric is geodesically convex
and the boundary strata are totally geodesically embedded.
The following observation is also due to Yamada:
Theorem 4.3. The completion T (S) of T (S) is a complete NPC space.
While this follows on general principles (cf. [19]), the identification of the bound-
ary strata of the completion with lower dimensional Teichmüller spaces (and Weil–
Petersson metrics) is especially useful.
Let us point out two properties of the geometry of the Weil–Petersson completion
that are consequences of this expansion. These were first stated by Yamada [210].
The first result, dubbed nonrefraction by Wolpert, is the statement that geodesics from
points in Teichmüller space to the boundary touch the boundary only at their endpoints
(see Figure 7 (a)). It is easy to see that this is true for the model space above. Indeed,
the equations for a unit speed geodesic α(t) = (ξ(t), θ (t)) in M are
ξ ξ̈ = (3/4)ξ 6 θ̇ 2 , ξ 6 θ̇ = constant, 2|ξ̇ |, ξ 3 |θ̇ | ≤ 1. (4.2)
If ξ(t) → 0 as t → 1, say, then the second and third equations imply that the
constant above must vanish. In other words, θ (t) is constant and ξ(t) is linear.
The proof of the statement for geodesics in T (S) involves a scaling argument to
approximate geodesics in T (S) by corresponding geodesics in the model space.
The third order approximation of the Weil–Petersson metric by the model metric
is sufficient to show that the approximation of geodesics is also to high order, and
the qualitative behavior of geodesics in T (S) is the same as for the model space.
D[c1 ]
∂T (S)
D[c2 ]
(a) (b)
Figure 7
Another application of this approximation gives the second important result: the
different strata of the boundary of T (S) intersect transversely. For example, consider
disjoint nonisotopic simple closed essential curves {c1 , c2 } on a closed compact surface
S with isotopy classes [c1 ], [c2 ]. Let [σ1 ] denote a point in the boundary component
D[c1 ] of T (S) corresponding to pinching c1 . Similarly, let [σ2 ] denote a point in the
boundary component D[c2 ] of T (S) corresponding to pinching c2 . Since c1 and c2 are
disjoint, the intersection of the closures D[c1 ]∩D[c2 ] is nonempty, and in fact contains
D([c1 ], [c2 ]), the stratum where both c1 and c2 are pinched. In particular, there is a
path in T (S) from σ1 to σ2 , lying completely in the boundary, which corresponds to
first pinching c2 , and then “opening up” c1 . The theorem states that this path has a
“corner” at its intersection with D([c1 ], [c2 ]), and is therefore not length minimizing.
In fact, the geodesic from [σ1 ] to [σ2 ] intersects the boundary of T (S) only in its
endpoints (see Figure 7 (b)).
4.1.2 The mapping class group. Denote by Diff(S) (resp. Diff + (S)) denote the
group of smooth diffeomorphisms (resp. orientation preserving diffeomorphisms) of
S with the smooth topology. Recall that Diff 0 (S) denotes the identity component of
Diff(S), that is, the group of all diffeomorphisms isotopic to the identity. The mapping
class group of S is the quotient
Mod(S) = Diff + (S)/Diff 0 (S).
See [54] for a recent survey on mapping class groups. From any of the several def-
initions of Teichmüller space given previously, it is clear that Mod(S) acts on T (S).
The first important result about this action is the following
Theorem 4.4. The mapping class group acts properly discontinuously on T (S).
This result is commonly attributed to Fricke. One method of proof follows from
the general fact that the action of Diff(S) on the space Met(S) of smooth Rieman-
nian metrics is properly discontinuous (cf. [48], [47]). In particular the restriction
to the action of Diff 0 (S) on Methyp (S) is properly discontinuous, and Teichmüller
space, which is the quotient Thyp (S) = Methyp (S)/Diff 0 (S), inherits such an action
of Mod(S).
Diffeomorphisms of S determine automorphisms of = π1 (S) as follows. Let
x0 ∈ S be a fixed basepoint. A diffeomorphism φ : S → S determines an automor-
phism of the fundamental group π1 (S, x0 ) if φ(x0 ) = x0 . Now any diffeomorphism is
isotopic to one which fixes x0 . Different choices of isotopy define automorphisms
of which differ by an inner automorphism. Hence, there is a homomorphism
Diff(S)/Diff 0 (S) → Out(S) where Out(S) is the outer automorphism group of π1 (S).
Theorem 4.5 (Dehn–Nielsen [149]). The homomorphism described above gives an

isomorphism Diff(S)/Diff 0 (S) Out(S).
The quotient M(S) = T (S)/Mod(S) is the classical Riemann moduli space. Since
by Teichmüller’s theorem T (S) is contractible (and in particular, simply connected),
Mod(S) may be regarded as the fundamental group of M(S). However, Mod(S) does
not quite act freely, so this interpretation holds only in the orbifold sense. Indeed,
M(S) is actually simply connected [121]. The compactification
M(S) = T (S)/Mod(S)
is homeomorphic to the Deligne–Mumford compactification of M(S), and it is a
projective algebraic variety (with orbifold singularities) [40]. An important measure
of the interior regions of M(S) is given by the Mumford–Mahler compactness theorem:
Theorem 4.6 (Mumford [146]). The set M(S)ε ⊂ M(S) consisting of equivalence
classes of Riemann surfaces where the hyperbolic lengths of all closed geodesics are
bounded below by ε > 0 is compact.
Note the condition in the theorem is Diff(S)-invariant and so is valid on the moduli
space M(S). The corresponding result is, of course, not true for T (S) because of
the proper action of the infinite discrete group Mod(S). For example, the orbit of
a point in T (S) by Mod(S) is unbounded, but projects to a single point in M(S).
This, however, is the only distinction between T (S) and M(S), and so the Mumford–
Mahler compactness theorem can be used effectively to address convergence questions
in T (S) as well.
We illustrate this by proving a fact that will be useful later on. Given a simple
closed curve c ⊂ S, let c [σ ] denote the length of the geodesic in the homotopy class
of c with respect to the hyperbolic metric σ . Note that this is independent of the choice
of σ up to the action of Diff 0 (S). Hence, c gives a well-defined function
c : T (S) −→ R+ . (4.3)
Then we have the following
Corollary 4.7. If [σj ] is a sequence in T (S) contained in no compact subset then

there is a simple closed curved c ⊂ S such that c [σj ] is unbounded.
Proof. For a point [σ ] ∈ T (S), let [[σ ]] ∈ M(S) denote the corresponding point in
M(S). Without loss of generality, we may assume [σj ] has no convergent subsequence
in T (S). The same may or may not be true for the sequence [[σj ]] ⊂ M(S). Indeed,
by Theorem 4.6, there are two cases: (1) there are elements [φj ] ∈ Mod(S) and a point
[σ∞ ] ∈ T (S) such that [φj ][σj ] → [σ∞ ] (after passing to a subsequence); (2) there
are simple closed curves cj such that cj [σj ] → 0 (after passing to a subsequence).
In the first case, our assumptions imply that infinitely many [φj ] are distinct. It
follows that there is a simple closed curve c such that fj (c) [σ∞ ] → ∞. But then
c [σj ] → ∞, as desired. In the second case, we may assume cj converges projectively
to a nontrivial measured foliation F (see Section 2.3.2). If c is any simple closed curve
with i([c], F ) = 0, then i(c, cj ) = 0 for j large. But since cj [σj ] → 0, c [σj ] → ∞
by the Collar Lemma 4.2.
Thurston’s classification of surface diffeomorphisms may be described in terms of

the natural action of Mod(S) on MF(S) and PMF(S): an element [φ] ∈ Mod(S) is
called reducible if [φ] fixes (up to isotopy) some collection of disjoint simple closed
essential curves on S. It is called pseudo-Anosov if there is r > 1 and transverse
measured foliations F+ , F− on S such that [φ]F+ is measure equivalent to rF+ , and
[φ]F− is measure equivalent to r −1 F− . F+ and F− are called the stable and unstable
foliations of [φ], respectively. The classification states that any [φ] ∈ Mod(S) is either
periodic (i.e. finite order), infinite order and reducible, or pseudo-Anosov. Moreover,
these are mutually exclusive possibilities.
4.1.3 Classification of Weil–Petersson isometries. We now indicate how the Thur-

ston classification of mapping classes is mirrored by the Weil–Petersson geometry.
The action of Mod(S) on T (S) is isometric with respect to the Weil–Petersson metric.
Conversely, every Weil–Petersson isometry is essentially given by a mapping class (see
[21], [129], [126], [206]). Since the Weil–Petersson metric has negative curvature it
is a natural to classify individual mapping classes in a manner similar to isometries of
Cartan–Hadamard manifolds.
Theorem 4.8 (Daskalopoulos–Wentworth [39], Wolpert [206]). If [φ] ∈ Mod(S) is

infinite order and irreducible, then there is a unique [φ]-invariant complete Weil–
Petersson geodesic in T (S).
Here is a very rough idea of proof of this result. Let α̃j : [0, 1] → T (S), α̃j (1) =
[φ]α̃j (0) be a sequence of curves minimizing the translation length of [φ], i.e.
1
lim α̃˙j wp dt = Lwp [φ].
j →∞ 0
Let αj : S 1 → M(S) be the projection of α̃j . Since M(S) is compact one can show
using Ascoli’s theorem that, after passing to a subsequence, αj converge uniformly to
some curve α : S 1 → M(S). The trick now is to show that this curve admits a lift
α̃ : [0, 1] → T (S), α̃(1) = [φ]α̃(0). Then α̃ must be an invariant geodesic. Since [φ]
is irreducible, by the nonrefraction results we know that α̃ must have image in T (S).
The existence of a lift is not obvious, since T (S) → M(S) is “branched” to infinite
order along the boundary ∂T (S). One needs to exploit the fact that α is the limit of
curves that are liftable. We refer to [39] for more details.
The existence of invariant geodesics for infinite order irreducible mapping classes
allows for the precise classification of Weil–Petersson isometries in terms of translation
length that we have given in Table 1. For [φ] ∈ Mod(S), define the Weil–Petersson
translation length by
Lwp [φ] = inf dwp ([σ ], [φ][σ ]) . (4.4)
[σ ]∈T (S)
Table 1. Classification of Weil–Petersson Isometries.
semisimple not semisimple

Lwp = 0 periodic strictly pseudoperiodic
Lwp = 0 infinite order irreducible reducible but not pseudoperiodic
First, let us clarify the terminology used there: [φ] ∈ Mod(S) is pseudoperiodic
if it is either periodic, or it is reducible and periodic on the reduced components; it is
strictly pseudoperiodic if it is pseudoperiodic but not periodic. Furthermore, we say
that [φ] is semisimple if there is [σ ] ∈ T (S) such that Lwp [φ] = dwp ([σ ], [φ][σ ]).
Here is a sketch of the proof: first, note that it is a consequence of Theorem 4.4 that
Lwp [φ] = 0 if and only if [φ] is pseudoperiodic. The first row of Table 1 then follows
from this and the fact that [φ] has a fixed point in T (S) if and only if [φ] is periodic.
If [φ] is infinite order irreducible, then as a consequence of Theorem 4.8, Lwp [φ]
is attained along an invariant geodesic, so these mapping classes are semisimple.
Conversely, suppose the translation length is attained at [σ ] ∈ T (S), but [φ][σ ] = [σ ].
Then we argue as in Bers [14] (see also, [9, p. 81]) to show that the geodesic from
[σ ] to [φ][σ ], which exists by the geodesic convexity of the Weil–Petersson metric,
may be extended to a complete [φ]-invariant geodesic. On the other hand, if there is a
complete, nonconstant Weil–Petersson geodesic in T (S) that is invariant with respect
to a mapping class [φ] ∈ Mod(S), the negative curvature implies that [φ] must be
infinite order and irreducible.
It is worth mentioning that no properties of pseudo-Anosov’s other than the fact
that they have infinite order and are irreducible were used in the proof above. In
particular, the description given in Table 1 is independent of Thurston’s classification.
We point out a further property of the axes of pseudo-Anosov’s.
Theorem 4.9 (Daskalopoulos–Wentworth [39], Wolpert [206]). Let A[φ] and A[φ ] be
the axes for independent pseudo-Anosov mapping classes [φ] and [φ ]. Then A[φ] and
A[φ ] diverge.
This result is also not completely obvious because of the noncompleteness of

T (S). More to the point, there exist flats, i.e. a totally geodesically embedded copy of
Rm → T (S). which potentially hinder the divergence. A much more detailed discus-
sion of asymptotics of complete Weil–Petersson geodesics is forthcoming (see [22]).
4.2 Energy of harmonic maps

In this section we return to harmonic maps and show how they can be used to probe
the action of the mapping class group on Teichmüller space. In Section 4.2.1 we
discuss Nielsen’s realization problem for finite subgroups of the mapping class group.
In Section 4.2.2 we introduce two classes of functions on Teichmüller space that are
constructed using the energy of harmonic maps, and we indicate when these functions
are proper. In Section 4.2.3 we discuss the convexity of one of the two classes and show
how this resolves the Nielsen conjecture. We also state Wolpert’s result on convexity
of length functions. Finally, in Section 4.2.4, we indicate some other applications of
the energy functionals.
4.2.1 Nielsen realization. Here we discuss the classical question of Nielsen [150].
The exact sequence
π
1 −→ Diff 0 (S) −→ Diff + (S) −→ Mod(S) −→ 1 (4.5)
which defines the mapping class group does not split in general (see [122], [143],
[144]). The realization problem asks for which subgroups G ⊂ Mod(S) does there
exist a homomorphism j : G → Diff + (S) such that π j = id.
Let S be a closed Riemann surface of negative Euler characteristic. Then we have
the following two important facts. First, if φ is a holomorphic automorphism of S
homotopic to the identity, then φ is in fact equal to the identity. Indeed, if this were
not the case then since complex curves in a complex surface intersect positively the
number of fixed points of φ, counted with multiplicity, would necessarily be positive.
On the other hand, if φ ∼ id, then by the Lefschetz fixed point theorem the total
intersection number is just the Euler characteristic of S, which we have assumed is
negative.
From this fact we arrive at Fenchel’s observation that if a subgroup G ⊂ Mod(S)
fixes a point [j ] ∈ T (S), then G can be realized as the automorphism group of a
Riemann surface (S, j ) with j in the class [j ]. For if φ1 , . . . , φm are holomorphic
lifts to Diff + (S) of generators [φ1 ], . . . , [φm ] of G, then any relation on the [φj ]’s,
applied to the φj ’s, is a holomorphic map ∼ id, and so by the previous paragraph the
relations in the group also lift. In particular, (4.5) splits over G.
The second fact is that the automorphism group of a Riemann surface of genus
p ≥ 2 is finite. This is because on the one hand it is the isometry group of the
hyperbolic metric, which is compact, and on the other hand it is discrete, since
there are no holomorphic vector fields. Hence, any subgroup of the mapping class
group which fixes a point in Teichmüller space is finite and (4.5) splits over it. These
two facts motivate the following result, which is known as the Nielsen Realization
Theorem.
Theorem 4.10 (Kerckhoff [105]). The sequence (4.5) splits over all finite subgroups
of Mod(S).
From the discussion above, the idea of the proof is to show the following
Theorem 4.11. Let G ⊂ Mod(S) be a finite subgroup of the mapping class group.
Then G has a fixed point in T (S).
The complete proof of Theorem 4.10 was first obtained by Kerckhoff in [105] and
later by Wolpert [204]. Both proofs proceed via Theorem 4.11. Partial results had
been found earlier by Fenchel [58], [59] and Zieschang [214]. See also Tromba [187].
4.2.2 Properness of the energy. Let M be an arbitrary compact Riemannian man-

ifold and S a closed hyperbolic surface with negative Euler characteristic. Now if
ρ : π1 (M) → π1 (S) is a given homomorphism it follows by Theorem 2.15 that there
is a harmonic map f : M → S such that the induced action f∗ : π1 (M) → π1 (S)
coincides with ρ. The energy E(f ) then depends only on the equivalence class of
hyperbolic metrics [σ ] ∈ T (S) (see Theorem 2.18). In other words, there is a well-
defined function
Eρ+ : T (S) −→ R+ .
The existence of a minimum is in turn a reflection of the homomorphism ρ. One way
to guarantee a minimum is to show that Eρ+ diverges at infinity. In this context, we
have the following
Proposition 4.12. If ρ is surjective then the associated function Eρ+ is proper.
Proof. This is easy to see, given the Lipschitz bound Proposition 2.17 and the Mum-
ford–Mahler Compactness Theorem 4.6 (or more precisely, Corollary 4.7). Indeed, if
Eρ+ is not proper, there is a sequence [σj ] and harmonic maps fj : M → (S, σj ) in the
homotopy class defined by ρ, such that E[σj ] ≤ B for some constant B. Furthermore,
we may assume there is a simple closed curve c with c [σj ] → ∞. Let s be a closed
curve in M with fj (s) homotopic to c. Then since the fj are uniformly Lipschitz,
c [σj ] ≤ length(fj (s)) ≤ B̃ length(s).
Since the right hand side is fixed independent of j and the left hand side diverges
with j , we derive a contradiction.
The superscript + on Eρ+ is to remind us that this is a function of the hyperbolic

metric on the target. It is also interesting to consider the energy as a function of the
domain metric (cf. (3.5)). Let M be a compact Riemannian manifold with nonpositive
sectional curvature. Let S be a closed surface and let ρ : π1 (S) → π1 (M) is a
homomorphism. Then for each complex structure σ on S there is a harmonic map
f : (S, σ ) → M whose induced action on π1 coincides with ρ. The energy of this
map gives a well-defined function
Eρ− −→ T (S) −→ R+ .
Again, the existence of minima can be deduced from the properness of this func-
tional. The following can be proved using the same ideas as in the proof of Proposi-
tion 4.12 and Corollary 4.7.
Proposition 4.13 (see [163], [169]). If ρ is injective then Eρ− is proper.
As we have seen in Theorem 3.11, the function Eρ− is differentiable. From the
discussion in Section 2.1.5 (see esp. (2.18)), critical points correspond to conformal
harmonic maps, i.e. those for which the Hopf differential vanishes. According to
Sacks–Uhlenbeck [163], these are branched minimal surfaces in M.
There is a remarkable connection between these functionals and the Weil–Petersson
metric. If we take M = (S, σ0 ) for some hyperbolic metric σ0 and ρ = id, we have
defined two functions Eid± on T (S), both of which clearly have critical points at [σ0 ].
We have
Theorem 4.14 (Tromba [184], Wolf [196], Jost [94]). The second variation of either
Eid± at [σ0 ] is a positive definite hermitian form on T[σ0 ] T (S) which coincides with the
Weil–Petersson metric.
A critical point of Eid− is a holomorphic map (S, σ ) → (S, σ0 ) homotopic to the

identity. As argued in Section 4.2.1, this must be the identity and σ = σ0 . We conclude
that Eid− is a proper function on T (S) with a unique critical point. By Theorem 4.14,
it is also nondegenerate. It follows that T (S) is diffeomorphic to Rn ; hence, we have
a fourth a proof of Theorem 2.8 (see [61]).
4.2.3 Convexity of energy and length functionals
Theorem 4.15 (Tromba [187],Yamada [209]). The energy Eρ+ defined above is strictly
convex along Weil–Petersson geodesics.
This result was first obtained by Tromba in the case where M is homeomorphic
to S. It was later generalized to the statement above by Yamada. It follows that the
minimum of Eρ+ is unique if it exists.
The conclusion is that there exists an abundance of convex exhaustion functions on
Teichmüller space and an explicit method to construct them. Any one of these gives
a solution to the Nielsen problem! For the average of such a function over a finite
subgroup G ⊂ Mod(S) is again strictly convex and G-invariant. Hence, its unique
minimum is also G-invariant, i.e. a fixed point of G, and Theorem 4.11 is proven. The
easiest example is to take M = (S, σ0 ), for any complex structure σ0 , and ρ = id, as
in the previous section.
It turns out that the analogous statement Theorem 4.15 for Eρ− is false. For example,
we could choose M to be a fibered hyperbolic 3-manifold with S a fiber and ρ the
homomorphism coming from the inclusion. Then ρ is invariant by conjugation of
the monodromy of the fibration, which by a theorem of Thurston is represented by a
pseudo-Anosov diffeomorphism (cf. [153]). In particular, it has infinite order. This

fact leads to infinitely many minima of Eρ− , whereas if Eρ− were strictly convex, it would
have a unique minimum. It is certainly an interesting question to find conditions where
convexity holds.
A very special case of the previous discussion is when M is a circle. Harmonic maps
from a circle correspond to geodesics. Historically, geodesic length functions were
considered before the energy of harmonic maps from higher dimensional domains. In
particular, we have the following important result of Wolpert.
Theorem 4.16 (Wolpert [204], [207]). For any simple closed curve c, the function
c : T (S) → R+ defined in (4.3) is strictly convex along Weil–Petersson geodesics.
The extension of the length function to geodesic currents is also strictly convex.
One consequence of this is the geodesic convexity of Teichmüller space, i.e. be-
tween any two points in T (S) there exists a unique Weil–Petersson geodesic. One
can also construct convex exhaustion functions, although in a manner slightly differ-
ent from that of the previous section. If we choose a collection c1 , . . . , cm of simple
closed curves which are filling in S in the sense that any other simple closed essential
curve has nontrivial intersection with at least one cj , then the function
β = c1 + · · · + cm , (4.6)
is an exhaustion function. This again follows by Mumford–Mahler compactness and
the Collar Lemma. Since β is also strictly convex, this gives a solution to Nielsen’s
problem as above, and indeed this is Wolpert’s method.
Finally, we point out that Kerckhoff’s proof of Theorem 4.10 was the first to lay
out this type of argument. The difference is that he proved convexity not with respect
to the Weil–Petersson geometry but along Thurston’s earthquake deformations.
4.2.4 Further applications. We now enumerate some other applications of the ideas
developed in previous section.
• Convex cocompact representations. Note that Proposition 4.13 can also be adapted
to the equivariant case and metric space targets. Here, ρ : = π1 (S) → Iso(X),
where X is a simply connected NPC space. Injectivity is replaced by the condition
that the translation length of any isometry in the image is bounded below by a uniform
constant.
A discrete embedding ρ : → Iso(X) is convex cocompact if there exists a ρ-
invariant closed geodesically convex subset N ⊂ X such that N/ρ is compact.
Theorem 4.17 (Goldman–Wentworth [73]). Mod(S) acts properly discontinuously

on the space of convex cocompact embeddings ρ : → Iso(X).
When Iso(X) = PSL(2, C), a convex cocompact representation is quasi-Fuchsian,

that is, a discrete embedding whose action on S 2 = ∂ H3 is topologically conjugate
to the action of a discrete subgroup of PSL(2, C). In this case, Theorem 4.17 is just
the known fact that Mod(S) acts properly on the space QF (S) of quasi-Fuchsian
embeddings. Indeed, Bers’ simultaneous uniformization theorem [12] provides a
Mod(S)-equivariant homeomorphism
QF (S) −→ T (S) × T (S).
Properness of the action of Mod(S) on T (S), Theorem 4.4, implies properness on
QF (S).
The idea of the proof of Theorem 4.17 is to show that if ρ is convex cocompact
(1) then there exists a ρ-equivariant harmonic map f : S̃ → X, and
(2) the corresponding energy functional Eρ− : T (S) → R+ is proper.
Then one associates to each ρ the compact subset of minima of Eρ− in T (S), and
properness of the action of Mod(S) on T (S) implies the result. See [73] for more
details.
• Filling foliations. Recall from Section 3.1.2 that by the Hubbard–Masur theorem
any measured foliation can be realized as the horizontal foliation of a holomorphic
quadratic differential. As a second application, consider the problem of realizing a pair
of measured foliations as the horizontal and vertical foliations of a single quadratic
differential on some Riemann surface. A pair F+ , F− of measured foliations on S is
called filling if for any third measured foliation G
i(F+ , G) + i(F− , G) = 0 ,
where i( ·, ·) denotes the intersection number (see Section 2.3.2).
Theorem 4.18 (Gardiner–Masur [67]). F+ , F− are filling if and only if there is a

complex structure j and a holomorphic quadratic differential ϕ on (S, j ) such that
F+ and F− are measure equivalent to the vertical and horizontal foliations of ϕ,
respectively. Moreover, [j ] ∈ T (S), and ϕ for each j ∈ [j ], are uniquely determined
by F± .
It is relatively easy to see that the horizontal and vertical trajectories of a holomor-
phic quadratic differential are filling (cf. [67, Lemma 5.3]). The proof of the converse
follows by showing, using arguments similar to those in the proof of Proposition 4.13,
that Ext F+ + ExtF− is a proper function on T (S). The first variational formula Theo-
rem 3.11 shows that a local minimum is a point at which the quadratic differentials
for F+ and F− are related by a minus sign. On the other hand, by the argument in
Section 3.1.2, F± are therefore vertical and horizontal foliations of one and the same
differential. Uniqueness can also be proven by analytic methods (see [194] for more
details).
• Holomorphic convexity of T (S). The convex exhaustion functions constructed in the
previous sections are, in particular, strictly plurisubharmonic (Tromba [188] showed
that this is true for Eid− as well). This gives a new proof of the following
Theorem 4.19 (Bers–Ehrenpreis [15]). Teichmüller space is a Stein manifold.
By a slight modification of length functions, we also have
Theorem 4.20 (Yeung [212]). T (S) admits a bounded strictly plurisubharmonic func-
tion.
Explicitly, one may take −β −ε , where β is the function in (4.6) and 0 < ε < 1.
The existence of a bounded plurisubharmonic function has important implications for
the equivalence of invariant metrics on Teichmüller space (see [24], [120], [212]).
5 Harmonic maps to Teichmüller space

5.1 Existence of equivariant harmonic maps
In many ways this last chapter combines ideas from all of the previous ones. Because
of the nonpositive curvature of the Weil–Petersson metric, harmonic maps with Teich-
müller space as a target have good regularity properties. The isometry group is the
mapping class group, so the equivariant problem gives a way to study representations
of fundamental groups to Mod(S). Since the Weil–Petersson metric is not complete,
we need to pass to the completion T (S) and use the theory of singular space targets of
Gromov–Korevaar–Schoen. In Section 5.1.1 we show how the results of Section 4.1.3
can be used to prove existence of equivariant harmonic maps to T (S), and in Sec-
tion 5.1.2 we state a result on the regularity of energy minimizing maps for surface
domains. Finally, in Section 5.1.3, we discuss the special case of holomorphic maps
from surfaces to Teichmüller space. An a priori bound on the energy of such maps
gives rise to the Arakelov–Paršin finiteness result (see Theorem 5.9).
5.1.1 Maps to the completion. As an application of the previous results, we consider

the problem of finding energy minimizing equivariant maps to Teichmüller space with
the Weil–Petersson metric. Recall the set-up: let M be a compact Riemannian mani-
fold with universal cover M̃, and let ρ : = π1 (M) → Mod(S) be a homomorphism.
Since Mod(S) acts on T (S) by isometries, we may ask under what conditions does
there exist a ρ-equivariant energy minimizing map f : M̃ → T (S).
Note that these may be regarded as harmonic maps M → M(S), although there
are two points of caution. The first is that strictly speaking M(S) is not a manifold,
but has orbifold singularities at those points corresponding to Riemann surfaces with
automorphisms. Hence, the smoothness of the map, and the harmonic map equations,
should be understood on a smooth finite (local) cover of M(S). The second (more
important) point is that the homotopy class of a map M → M(S) should be taken in
the orbifold sense (i.e. equivariantly with respect to a homomorphism → Mod(S)).
Indeed, by the simple connectivity of M(S) remarked on in Section 4.1.2, homotopy

classes of maps to M(S) are very different from equivariant homotopy classes of maps
to T (S).
As in Section 3.1.1, the answer to the existence question depends on the asymp-
totic dynamics of the image subgroup ρ() ⊂ Mod(S). In general, the asymptotic
behavior of Weil–Petersson geodesics is quite complicated (see [20], [22]). As an
approximation, one can consider the action on the Thurston boundary PMF(S) of
projective measured foliations. From this point of view there derives a complete clas-
sification, analogous to the Thurston classification, of subgroups of the mapping class
group.
Theorem 5.1 (McCarthy–Papadopoulos [128]). A subgroup of Mod(S) is exactly one

of the following types:
(1) finite;
(2) infinite irreducible and virtually cyclic;
(3) infinite reducible;
(4) sufficiently large.
By sufficiently large we mean that the subgroup contains two pseudo-Anosov’s

with distinct fixed point sets in PMF(S). These groups contain free groups on two
generators.
We apply this theorem to the image G = ρ() of the homomorphism ρ. By the
Nielsen Realization Theorem 4.10, if G is finite then it fixes a point [σ ] in Teichmüller
space. Hence, the constant map f (x) = [σ ] is equivariant and clearly harmonic.
Case (2) arises when G has a finite index subgroup [φ] Z generated by
a pseudo-Anosov [φ]. By Theorem 4.8 this stabilizes a complete Weil–Petersson
geodesic A[φ] ⊂ T (S). The corresponding finite index subgroup ˆ ⊂ defines a
finite cover Mˆ → M, and the group of deck transformations then acts on S 1 . Hence,
it suffices to find an equivariant harmonic map Mˆ → S 1 → T (S)/[φ]. This can
be done using the heat equation approach, since equivariance is preserved under the
flow (2.29).
In Case (3), G fixes a stratum in the boundary ∂T (S) isomorphic to a product of
lower dimensional Teichmüller spaces. Since the boundary strata are totally geodesi-
cally embedded, the problem of finding an energy minimizer to T (S) is reduced to
Cases (1), (2), and (4) for lower dimensions.
Finally, we come to Case (4).
Theorem 5.2 (Daskalopoulos–Wentworth [39]). If ρ : → Mod(S) is sufficiently

large then it is proper in the sense of Korevaar–Schoen (see Section 3.1.1).
This is a consequence of Theorem 4.9. Using Theorem 3.7, it follows that there
exist equivariant harmonic maps in this case as well. Putting all of these considerations
together, we have
Corollary 5.3. Let ρ : π1 (M) → Mod(S) be a homomorphism, where M is a compact

Riemannian manifold. Then there exists a finite energy ρ-equivariant harmonic map
f : M̃ → T (S). Moreover, f is uniformly Lipschitz.
We note that this statement, apparently stronger than the one appearing in [39],
follows from considering the possibilities in Theorem 5.1.
It is certainly expected that uniqueness holds in the corollary under certain as-
sumptions. Some generalization of Theorem 2.33 is needed. Roughly speaking, one
expects uniqueness to fail only if the image of f lies in a flat. Alternatively, one should
be able to prove a priori that if ρ is sufficiently large then some point of the image of
f lies in the interior T (S). Then the strictly negative curvature implies that the image
is a geodesic, which again contradicts the assumption of sufficiently large.
5.1.2 Surface domains. A natural question arises from the statement of Corollary 5.3.
Under what conditions does the image of a harmonic map to T (S) actually lie in T (S)?
This is an important issue, since if f (x) ∈ T (S), then since T (S) is a manifold f is
smooth near the point x. More generally, one would at least like to have control over
the size of the singular set (cf. Theorem 2.35).
The first result in this direction is the following
Theorem 5.4 (Wentworth [193]). Let ⊂ R2 be a bounded domain, and suppose

f : → T (S) is energy minimizing with respect to its boundary conditions. If
f (x) ∈ T (S) for some x ∈ , then f () ⊂ T (S).
To give a rough idea of why this should be the case, we again consider the model for
the Weil–Petersson geometry near the boundary ∂T (S) discussed in Section 4.1.1. Let
f : B1 (0) → M be a finite energy harmonic map. By Theorem 2.31, f is uniformly
Lipschitz. The generalization of (4.2) are the equations
ξ ξ = 43 ξ 6 |∇θ |2 , div(ξ 6 ∇θ) = 0, |∇ξ |, ξ 3 |∇θ | are locally bounded.
Because of the singularities, ξ(x, y) and θ(x, y) are only weak solutions of these
equations. We may assume that f is nonconstant with f (0) ∈ ∂ M. Furthermore,
suppose the origin is not a zero of the Hopf differential ϕ. It is not hard to show that
the singular set, i.e. f −1 (∂ M) is a leaf of the horizontal foliation of ϕ. In ϕ-coordinates
(x, y), one shows with some more analysis of the situation that ξ(x, y) ∼ y. Then the
second equation above becomes essentially div(y 6 ∇θ ) = 0. This kind of degenerate
equation appears in the study of the porous medium equation [107], and one can show
that θ (x, y) itself is locally bounded. Then a scaling argument using the monotonicity
formula (2.50) can be used to derive a contradiction.
To go from a regularity result for harmonic maps to M to a result for maps to T (S)
requires an approximation of harmonic maps to targets with asymptotically product
metrics. This is similar to the discussion of geodesics above. For more details we
refer to [193].
The local regularity implies
Corollary 5.5. Let ρ : π1 (B) → Mod(S) be irreducible, where B is a compact

Riemann surface. Then there exists a smooth ρ-equivariant harmonic map f : B̃ →
T (S). Moreover, if ρ is sufficiently large, then f is unique.
An interesting potential application of this result pertains to the following

Question. Let B be a closed surface. Does there exist an injective homomorphism
ρ : π1 (B) → Mod(S) such that the image of ρ consists entirely of pseudo-Anosov’s?
Examples of all pseudo-Anosov subgroups of Mod(S) have been constructed in
[195], but these are not surface groups. Such groups, should they exist, would admit
minimal surface representations in Mod(S):
Corollary 5.6. Let B be a closed surface and ρ : π1 (B) → Mod(S). In addition,

we assume that for every simple closed essential curve in B, the image by ρ of the
associated conjugacy class in π1 (B) is pseudo-Anosov. Then there is a conformal
harmonic ρ-equivariant map f : (B̃, j ) → T (S) for some complex structure j on B.
The argument proceeds as in the proof of Proposition 4.13. Note that there is a
lower bound, depending only on the genus, of the Weil–Petersson translation length
of any pseudo-Anosov (see [39]).
5.1.3 Holomorphic maps from Riemann surfaces. By Proposition 2.13 (see esp.
(2.25)), since the Weil–Petersson metric is Kähler, equivariant holomorphic maps from
surfaces to T (S) are examples of energy minimizers; in particular, harmonic maps.
These are given by holomorphic curves in M(S) that are locally liftable to T (S).
Alternatively, consider a family X → B, where B is a compact Riemann surface, and
X is a locally liftable holomorphic fibration of genus p Riemann surfaces. Associated
to this is a monodromy homomorphism ρ : π1 (B) → Mod(S). By Corollary 5.5, if
ρ is irreducible there is a ρ-equivariant harmonic map f : B̃ → T (S). In general,
this will not be holomorphic for any choice of complex structure on B. By the
essential uniqueness of the harmonic map, we see that the issue of holomorphicity is
a property of the (conjugacy class) of the monodromy representation ρ. Let us call
a homomorphism ρ : π1 (B) → Mod(S) holomorphic if there exists a ρ-equivariant
holomorphic map B̃ → T (S).
A simple example occurs when the monodromy has finite image. Then by the
Nielsen realization theorem, ρ fixes a point in T (S). In particular, there is a (constant)
holomorphic map. In terms of the family X → B, this is precisely the case where
the lift of the fibration p ∗ X → B̂ to some finite cover p : B̂ → B is trivial. Such a
fibration is called isotrivial.
The harmonic map point of view provides a tool to study holomorphic families.
Here is one property:
Theorem 5.7. If ρ : π1 (B) → Mod(S) is holomorphic and nonisotrivial, then ρ is

sufficiently large.
Proof. Suppose not. By the classification of subgroups of the mapping class group
Theorem 5.1, ρ is either reducible or virtually cyclic. In the former case, there is a
proper totally geodesic stratum S ⊂ ∂T (S) that is invariant under ρ. Since projection
to S from the interior T (S) is strictly distance decreasing, the geodesic homotopy of f
to S is both ρ-equivariant and strictly energy decreasing. This contradicts the fact that
u is the energy minimizer. If ρ is virtually cyclic, then the energy minimizer maps onto
a geodesic. Since the image is one dimensional, this contradicts holomorphicity.
The following is also a consequence of the uniqueness of harmonic maps to T (S)

discussed in the proof above. This is sometimes called the rigidity theorem.
Theorem 5.8. Holomorphic families with the same monodromy (up to conjugation)
are equivalent.
The main finiteness result is the following
Theorem 5.9 (Arakelov [8], Paršin [155]). Fix a closed Riemann surface B, and
let Mod(S) denote the mapping class group of a compact surface of genus p ≥ 2.
Then there are at most finitely many conjugacy classes of non-isotrivial holomorphic
homomorphisms ρ : π1 (B) → Mod(S).
We note that this can be extended to the case where B is a Riemann surface with
punctures. The punctures correspond to singularities in the surface fibration, and in
the holomorphic case the local monodromy around the punctures is pseudoperiodic.
Finite energy maps always exist in this case (see [36]).
The key to Theorem 5.9 is a uniform bound on the energy. Since T (S) has holo-
morphic sectional curvature bounded above by a negative constant (see Theorem 4.1),
Royden’s version of the Yau-Schwartz lemma implies that if f : B̃ → T (S) is holo-
morphic, then
f ∗ dswp
2
≤ CdsB̃2 ,
for a uniform constant C (see [160]). In particular, by (2.25), the energy of a holo-
morphic map is uniformly bounded. Since by Proposition 2.17 the Lipschitz constant
of harmonic maps is bounded by the total energy, a sequence of holomorphic maps
to M(S) is necessarily equicontinuous (see also [74]). This allows one to construct
convergent subsequences for the maps M(S). As in the argument in Section 4.1.3
there is the issue of lifting the limiting map. In this way, one derives a contradiction
to the existence of infinitely many distinct conjugacy classes of holomorphic ρ. For
a fuller account of this approach to the Arakelov–Paršin theorem, we refer to [100]
and [85].
5.2 Superrigidity
In this final section we briefly describe how equivariant harmonic map theory can be
used to study homomorphisms of fundamental groups of compact manifolds to the
mapping class group. The link between superrigidity and harmonic maps uses a tech-
nique which can be traced back to Bochner and Calabi–Weil and was first fully utilized
in connection with the Margulis superrigidity theorem. In fact, as mentioned earlier,
many of the ideas presented here were inspired by the attempt to give a harmonic maps
proof of superrigidity. In Section 5.2.1 we state the Ivanov–Farb–Kaimanovich–Masur
theorem for homomorphisms of superrigid lattices into mapping class groups. In Sec-
tion 5.2.2 we describe two approaches in generalizing harmonic maps by allowing the
domain to be singular as well. The first is the analytic approach along the lines for
smooth domains, the second is the combinatorial approach. As an application one can
prove a statement on the non-Archimedean superigidity of lattices in mapping class
groups.
5.2.1 The Ivanov–Farb–Kaimanovich–Masur theorem. Harvey originally asked

whether the mapping class group could be isomorphic to a lattice in a symmetric
space [80]. This was shown not to be the case by Ivanov [88], [89]. For some of the
similarities and differences between Mod(S) and arithmetic lattices, see [53], [88],
[127] and Ivanov’s survey article [90]. Indeed, a stronger statement is true:
Theorem 5.10. Let be a cocompact lattice in any symmetric space with nonpositive
curvature other than the real or complex hyperbolic spaces. Then any homomorphism
→ Mod(S) has finite image.
For symmetric spaces of rank ≥ 2 this result is due to Farb–Masur [55], following
earlier work of Kaimanovich–Masur [102]. Ivanov has announced an independent
proof. Bestvina–Fujiwara [17] gave a proof using bounded cohomology, and for
hermitian symmetric spaces an independent proof can be found in Hain [76]. Us-
ing the method of [30], [101], [139] the remaining rank 1 cases were proven by
S.-K. Yeung [213].
Geometric superrigidity uses harmonic maps to prove results of this type. The
basic philosophy is to show that equivariant harmonic maps f : G/K → N, where
G/K is a symmetric space of higher rank and N has nonpositive curvature, would
necessarily be totally geodesic. Recall from Section 2.2.1 that the harmonic map
equations are of the form Tr ∇df = 0, whereas the equations for a totally geodesic
map are ∇df = 0. Curvature conditions must be used to show that the stronger
(overdetermined) set of equations are automatically satisfied. One then attempts to
use geometric considerations to rule out the existence of nonconstant totally geodesic
maps.
To give a simple example of how this might come about, consider the following
result.
Theorem 5.11 (Eells–Sampson [51]). If f : M̃ → N is an equivariant harmonic

map, N is a Riemannian manifold of nonpositive curvature, and M is closed compact
with non-negative Ricci curvature, then f is totally geodesic. If the Ricci curvature
of M is positive at one point, f is constant. If the sectional curvature of N is negative
then f is either constant or maps to a geodesic.
Indeed, the statement easily follows by integrating both sides of the Bochner for-
mula (2.32) and using the divergence theorem. When the domain does not satisfy this
curvature restriction, the proof fails. Nevertheless, more sophisticated forms of the
Bochner formulas have been derived in the case of domains with Einstein metrics,
or more generally, certain parallel tensors. For more details, we refer to [30], [101],
[139].
In light of Corollary 5.3, one is tempted to prove Theorem 5.10 using harmonic
maps to T (S). The difficulty is in the singular nature of the NPC space T (S). However,
the idea that these techniques could be generalized to singular space targets is one of
the major contributions of [75]. The argument based on the Bochner formula given
above continues to be valid, so long as the singular set of f is relatively small, e.g. has
codimension at least 2, so that the integration by parts needed to apply the divergence
theorem holds. All of this is motivation to extend the regularity result of Theorem 5.4
to higher dimensional domains.
5.2.2 Harmonic maps from singular domains. Thus far we have discussed the
theory of harmonic maps from smooth domains into (possibly singular) metric space
targets. These included singular surfaces, R-trees, and the Weil–Petersson completion
of Teichmüller space. In this section we sketch two generalizations of this study to
the case where the domain is also allowed to be singular.
We start with an analytic approach closely related to the techniques discussed
above. Let be a finite 2-dimensional simplicial complex. The restriction to two
dimensions is not essential and most of the following results hold in general. It is
important, however, to assume that is admissible (cf. [25], [49]), meaning that it
satisfies the following conditions:
(1) Every simplex is contained in a face (i.e. a 2-simplex).
(2) Every pair of faces can be joined by a sequence of pairwise adjacent faces.
(3) has no boundary, i.e. every edge is contained in at least two faces.
(4) is flat in that every open face is isometric to an equilateral triangle in R2 .
We also allow ourselves a choice w of weights w(F ) > 0 for each face. This is an
important technical point. Given an NPC space (X, d) and a map f : → X, define
the w-energy

1
Ew (f ) = w(F ) |∇f |2 (x) dx,
2 F
F
where the sum is over all faces F of . A map f is called w-harmonic if it is

locally energy minimizing among all maps of finite w-energy. As before, we also
consider the equivariant theory, where f is a map from the universal cover ˜ of
that is equivariant with respect to a homomorphism ρ : = π1 ( ) → Iso(X). The
existence Theorem 3.1 then holds for domains as well (cf. [49], [37]).
In the following, we shall assume a fixed choice of weights and omit w from
the notation. Perhaps the most interesting feature of harmonic maps from simplicial
domains is the Hölder continuity. This was first proven by J. Chen [25] for flat metrics
and in a more general context by Eells–Fuglende [49]. The following stronger version
describes the singular behavior near the vertices.
Theorem 5.12 (Daskalopoulos–Mese [37]). Let f : → X be harmonic. Then for

domains U ⊂⊂ ⊂ the following holds:
(1) f is Lipschitz continuous on U away from the vertices of , where the Lipschitz
constant depends only on U , the total energy on , and the distance to the vertex
set.
(2) Let v be a vertex with α = ordv (f ), where the order is defined as in (2.51). Then
there exists r0 > 0 and C depending only on the energy of f such that
sup |∇f |2 (x) ≤ Cr 2α−2
x∈Br (v)
for all 0 < r ≤ r0 .
The important point here is that, unlike the case of smooth domains, α need not
be ≥ 1. One application of Theorem 5.12 is the compactification of character varieties
for arbitrarily finitely presented groups along the lines of Theorem 3.19. Indeed, one
can always realize such a group as the fundamental group of an admissible 2-complex.
Other potential consequences use the notion of a Hopf differential. Clearly, for energy
minimizers, ϕ = Hopf(f ) is a holomorphic quadratic differential on the interior of
each face (cf. Section 3.1.2). For points x on an edge e, we have the following
balancing condition:

Im ϕF (x) = 0,
F
where the sum is over all faces F adjacent to e at x. An important open question is
whether zeros of ϕ can accumulate along the edges. If not, then the Hopf differentials
of w-harmonic maps define geometric or track foliations on (cf. [18], [46], [116]).
Another important issue is the asymptotic behavior of the induced foliation on ˜ . More
generally, one might ask under what conditions one can generalize to this setting the
results for surface groups discussed previously in this chapter.
We now return to the relationship between regularity and rigidity. We have the
following
Theorem 5.13 (Daskalopoulos–Mese [37]). Let f : ⊂ → X be energy minimiz-

ing, where X is a smooth manifold of nonpositive curvature. For any x ∈ which
is not a vertex, then there is a neighborhood U of x such that for any face F the
restriction of f to F ∩ U is smooth.
Using this, one has a nontrivial generalization of Theorem 5.11 to the case of
singular domains:
Theorem 5.14 (Daskalopoulos–Mese [38]). Suppose is an admissible 2-complex

and X is a complete Riemannian manifold of nonpositive curvature. If f : → X is
harmonic and |∇f |2 bounded, then f is totally geodesic on each simplex of X. If the
sectional curvature of X is strictly negative, then either f is constant or it maps each
simplex to a geodesic.
This is a kind of rigidity result for the group = π1 ( ), and a combinatorial

version was first proven by M.-T. Wang (see below). The result follows by the Bochner
formula (2.32), the vanishing of the Ricci curvature on the domain, and the fact that
|∇f |2 allows us to integrate by parts. Global boundedness of the energy density is
guaranteed by a combinatorial condition on . Namely, the first eigenvalue of the
discrete Laplacian on the link of every vertex with the induced weights should be
≥ 1/2 (see [38]). This condition is a generalization of the notion of p-adic curvature
that first appeared in the work of Garland (cf. [68]).
The second approach discretizes the notion of an energy minimizer. Let be
an admissible 2-complex and X an NPC space as above and ρ : → Iso(X) a
homomorphism. Given a system of weights on the faces of there is a standard way
to induce weights on the lower dimensional simplices. For example, the weight of an
edge is the sum of the weights of adjacent faces. Let i , ˜ i denote the i-skeletons.
Given a ρ-equivariant map f : ˜ 0 → X define its energy by
1
Ecomb (f ) = w(exy )d 2 (f (x̃), f (ỹ)),
2
exy ∈ 1
where exy denotes an edge with adjacent vertices x and y, and x̃, ỹ are adjacent
vertices of a lift of exy to ˜ . We say that f is a ρ-equivariant combinatorial harmonic
map if it minimizes Ecomb (f ). Under the assumption that X is locally compact and
that ρ() does not fix a point in ∂X one can prove the existence of combinatorial
harmonic maps (see Wang [191], [192]). Furthermore, assuming the first eigenvalue
of the combinatorial Laplacian of the link of every vertex with the induced weights is
> 1/2, one can deduce rigidity results as in the first approach. This can be used to
deduce non-Archimedean generalizations of Theorem 5.10 [171].
References
[1] W. Abikoff, The real analytic theory of Teichmüller space. Lecture Notes in Math. 820,
Springer-Verlag, Berlin 1980. 35, 38, 39, 42
[2] W. Abikoff, Oswald Teichmüller. Math. Intelligencer 8 (1986), 8–16, 33. 35
[3] L. V. Ahlfors, On quasiconformal mappings. J. Analyse Math. 3 (1954), 1–58; correction,
207–208. 42
[4] L. V. Ahlfors, Curvature properties of Teichmüller’s space. J. Analyse Math. 9
(1961/1962), 161–176. 78
[5] L. V. Ahlfors, Lectures on quasiconformal mappings. Van Nostrand, Inc., New York
1966. 40
[6] L. V. Ahlfors and L. Sario, Riemann surfaces. Princeton Math. Ser. 26, Princeton Uni-
versity Press, Princeton, N.J., 1960.
[7] F. J. Almgren, Jr., Almgren’s big regularity paper. World Sci. Monogr. Ser. Math. 1,
World Scientific, Inc., River Edge, N.J., 2000. 61
[8] S. J.Arakelov, Families of algebraic curves with fixed degeneracies. Izv. Akad. Nauk SSSR
Ser. Mat. 35 (1971), 1269–1293; English transl. Math. USSR, Izv. 5 (1971), 1277–1302.
94
[9] W. Ballmann, M. Gromov, and V. Schroeder, Manifolds of nonpositive curvature. Progr.
Math. 61, Birkhäuser, Boston 1985. 84
[10] W. Ballmann and J. Światkowski, On L2 -cohomology and property (T) for automorphism
‘ complexes. Geom. Funct. Anal. 7 (1997), 615–645.
groups of polyhedral cell
[11] A. F. Beardon, The geometry of discrete groups. Grad. Texts in Math. 91, Springer-Verlag,
New York 1983. 43
[12] L. Bers, Simultaneous uniformization. Bull. Amer. Math. Soc. 66 (1960), 94–97. 89
[13] L. Bers, Quasiconformal mappings and Teichmüller’s theorem. In Analytic functions,
Princeton University Press, Princeton, N.J., 1960, 89–119. 42
[14] L. Bers, An extremal problem for quasiconformal mappings and a theorem by Thurston.
Acta Math. 141 (1978), 73–98. 84
[15] L. Bers and L. Ehrenpreis, Holomorphic convexity of Teichmüller spaces. Bull. Amer.
Math. Soc. 70 (1964), 761–764. 90
[16] M. Bestvina, Degenerations of the hyperbolic space. Duke Math. J. 56 (1988), 143–161.
76
[17] M. Bestvina and K. Fujiwara, Bounded cohomology of subgroups of mapping class
groups. Geom. Topol. 6 (2002), 69–89. 95
[18] B. H. Bowditch, Group actions on trees and dendrons. Topology 37 (1998), 1275–1298.
58, 97
[19] M. R. Bridson and A. Haefliger, Metric spaces of non-positive curvature. Grundlehren
Math. Wiss. 319, Springer-Verlag, Berlin 1999. 64, 80
[20] J. Brock, The Weil-Petersson visual sphere. Geom. Dedicata 115 (2005), 1–18. 91
[21] J. Brock and D. Margalit, Weil–Petersson isometries via the pants complex. Proc. Amer.
Math. Soc. 135 (2007), 795–803. 83
[22] J. Brock, H. Masur, and Y. Minsky. Asymptotics of Weil–Petersson geodesics. In prepa-

ration. 84, 91
[23] K.-C. Chang, W. Y. Ding, and R. Ye, Finite-time blow-up of the heat flow of harmonic
maps from surfaces. J. Differential Geom. 36 (1992), 507–515. 49
[24] B.-Y. Chen, Equivalence of the Bergman and Teichmüller metrics on Teichmüller spaces.
Preprint, 2004. 90
[25] J. Chen, On energy minimizing mappings between and into singular spaces. Duke Math.
J. 79 (1995), 77–99. 96, 97
[26] S. S. Chern, Complex manifolds without potential theory. 2nd edn., Universitext,
Springer-Verlag, New York 1979. 70
[27] T. Chu, The Weil-Petersson metric in the moduli space. Chinese J. Math. 4 (1976), 29–51.
78
[28] D. Cooper. Degenerations of representations into SL2 (C). Preprint 1994. 75
[29] K. Corlette, Flat G-bundles with canonical metrics. J. Differential Geom. 28 (1988),
361–382. 64, 65, 71
[30] K. Corlette, Archimedean superrigidity and hyperbolic geometry. Ann. of Math. (2) 135
(1992), 165–182. 64, 95, 96
[31] J.-M. Coron and J.-M. Ghidaglia, Explosion en temps fini pour le flot des applications
harmoniques. C. R. Acad. Sci. Paris Sér. I Math. 308 (1989), 339–344. 49
[32] M. Culler and J. W. Morgan, Group actions on R-trees. Proc. London Math. Soc. (3) 55
(1987), 571–604. 64, 65
[33] M. Culler and P. B. Shalen, Varieties of group representations and splittings of 3-
manifolds. Ann. of Math. (2) 117 (1983), 109–146. 38, 72
[34] G. Daskalopoulos, S. Dostoglou, and R. Wentworth, Character varieties and harmonic
maps to R-trees. Math. Res. Lett. 5 (1998), 523–533. 65, 76, 77
[35] G. Daskalopoulos, S. Dostoglou, and R. Wentworth, On the Morgan-Shalen compactifi-
cation of the SL(2, C) character varieties of surface groups. Duke Math. J. 101 (2000),
189–207. 67, 68, 72
[36] G. Daskalopoulos, L. Katzarkov, and R. Wentworth, Harmonic maps to Teichmüller
space. Math. Res. Lett. 7 (2000), 133–146. 94
[37] G. Daskalopoulos and C. Mese. Harmonic maps from two-complexes. Comm. Anal.
Geom. 14 (3) (2006), 497–549. 97, 98
[38] G. Daskalopoulos and C. Mese. Harmonic maps from a simplicial complexes and geo-
metric rigidity. Preprint. 98
[39] G. Daskalopoulos and R. Wentworth, Classification of Weil-Petersson isometries. Amer.
J. Math. 125 (2003), 941–975. 80, 83, 84, 91, 92, 93
[40] P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus.
Inst. Hautes Études Sci. Publ. Math. (1969), 75–109. 82
[41] E. D’Hoker and D. H. Phong, Multiloop amplitudes for the bosonic Polyakov string.
Nuclear Phys. B 269 (1986), 205–234. 44
[42] S. K. Donaldson, A new proof of a theorem of Narasimhan and Seshadri. J. Differential
Geom. 18 (1983), 269–277. 70
[43] S. K. Donaldson, Anti self-dualYang-Mills connections over complex algebraic surfaces

and stable vector bundles. Proc. London Math. Soc. (3) 50 (1985), 1–26. 64, 70
[44] S. K. Donaldson, Twisted harmonic maps and the self-duality equations. Proc. London
Math. Soc. (3) 55 (1987), 127–131. 71
[45] J. Douglas, Minimal surfaces of higher topological structure. Ann. of Math. (2) 40 (1939),
205–298. 69
[46] M. J. Dunwoody, Groups acting on R-trees. Comm. Algebra 19 (1991), 2125–2136. 97
[47] C. J. Earle and J. Eells, A fibre bundle description of Teichmüller theory. J. Differential
Geometry 3 (1969), 19–43. 53, 81
[48] D. G. Ebin, The manifold of Riemannian metrics. In Global Analysis (Berkeley, Calif.,
1968), Proc. Sympos. Pure Math. 15, Amer. Math. Soc., Providence, R.I., 1970, 11–40.
44, 53, 81
[49] J. Eells and B. Fuglede, Harmonic maps between Riemannian polyhedra. Cambridge
Tracts in Math. 142, Cambridge University Press, Cambridge 2001. 96, 97
[50] J. Eells and L. Lemaire, Two reports on harmonic maps. World Scientific, River Edge,
N.J., 1995. 35, 48
[51] J. Eells and J. H. Sampson, Harmonic mappings of Riemannian manifolds. Amer. J.
Math. 86 (1964), 109–160. 48, 49, 53, 64, 96
[52] J. Eells and J. C. Wood, Restrictions on harmonic maps of surfaces. Topology 15 (1976),
263–266. 52
[53] B. Farb, A. Lubotzky, andY. Minsky, Rank-1 phenomena for mapping class groups. Duke
Math. J. 106 (2001), 581–597. 95
[54] B. Farb and D. Margalit, A primer on mapping class groups. In preparation. 81
[55] B. Farb and H. Masur, Superrigidity and mapping class groups. Topology 37 (1998),
1169–1176. 95
[56] B. Farb and M. Wolf, Harmonic splittings of surfaces. Topology 40 (2001), 1395–1414.
66, 67, 68
[57] A. Fathi, F. Laudenbach and V. Poénaru (eds.), Travaux de Thurston sur les surfaces.
Astérisque 66–67 (1979). 57, 59
[58] W. Fenchel, Estensioni di gruppi discontinui e trasformazioni periodiche delle superficie.
Atti Accad. Naz Lincei. Rend. Cl. Sci. Fis. Mat. Nat. (8) 5 (1948), 326–329. 86
[59] W. Fenchel, Remarks on finite groups of mapping classes. Mat. Tidsskr. B. 1950 (1950),
90–95 (in Danish). 86
[60] A. E. Fischer and A. J. Tromba, On a purely “Riemannian” proof of the structure and
dimension of the unramified moduli space of a compact Riemann surface. Math. Ann.
267 (1984), 311–345. 44
[61] A. E. Fischer and A. J. Tromba, A new proof that Teichmüller space is a cell. Trans.
Amer. Math. Soc. 303 (1987), 257–262. 87
[62] A. Fletcher and V. Markovic, Infinite dimensional Teichmüller spaces. In Handbook of
Teichmüller theory (A. Papadopoulos, ed.), Volume II, EMS Publishing House, Zurich
2008. 35
[63] O. Forster, Lectures on Riemann surfaces. Grad. Texts in Math. 81, Springer-Verlag,
New York 1991. 38
[64] F. P. Gardiner, Measured foliations and the minimal norm property for quadratic differ-
entials. Acta Math. 152 (1984), 57–76. 68, 69
[65] F. P. Gardiner, Teichmüller theory and quadratic differentials. John Wiley & Sons, New
York 1987. 69
[66] F. P. Gardiner and N. Lakic, Quasiconformal Teichmüller theory. Math. Surveys Monogr.
76, Amer. Math. Soc., Providence, R.I., 2000. 42
[67] F. P. Gardiner and H. Masur, Extremal length geometry of Teichmüller space. Complex
Variables Theory Appl. 16 (1991), 209–237. 89
[68] H. Garland, p-adic curvature and the cohomology of discrete subgroups of p-adic groups.
Ann. of Math. (2) 97 (1973), 375–423. 98
[69] M. Gerstenhaber and H. E. Rauch, On extremal quasi-conformal mappings. I; II. Proc.
Nat. Acad. Sci. U.S.A. 40 (1954), 808–812; 991–994. 55
[70] W. M. Goldman, Representations of fundamental groups of surfaces. In Geometry and
topology (College Park, Md., 1983/84), Lecture Notes in Math. 1167, Springer-Verlag,
Berlin 1985, 95–117. 38
[71] W. M. Goldman and J. J. Millson, The deformation theory of representations of fun-
damental groups of compact Kähler manifolds. Inst. Hautes Études Sci. Publ. Math.
(1988), 43–96. 38
[72] W. M. Goldman, Trace coordinates on Fricke space of some simple hyperbolic surfaces.
In Handbook of Teichmüller theory (A. Papadopoulos, ed.), Volume II, EMS Publishing
House, Zurich 2008. 38
[73] W. M. Goldman and R. Wentworth, Energy of twisted harmonic maps of Riemann sur-
faces. In Proceedings of the Ahlfors-Bers Colloquium (Ann Arbor, Michigan, 2005), to
appear. 88, 89
[74] H. Grauert and H. Reckziegel, Hermitesche Metriken und normale Familien holomorpher
Abbildungen. Math. Z. 89 (1965), 108–125. 94
[75] M. Gromov and R. Schoen, Harmonic maps into singular spaces and p-adic superrigidity
for lattices in groups of rank 1. Inst. Hautes Études Sci. Publ. Math. 76 (1992), 165–246.
54, 60, 61, 62, 96
[76] R. Hain, Locally symmetric families of curves and Jacobians. In Moduli of curves and
abelian varieties, Aspects Math. E33, Vieweg, Braunschweig 1999, 91–108. 95
[77] R. S. Hamilton, Extremal quasiconformal mappings with prescribed boundary values.
Trans. Amer. Math. Soc. 138 (1969), 399–406. 42
[78] R. S. Hamilton, Harmonic maps of manifolds with boundary. Lecture Notes in Math.
471, Springer-Verlag, Berlin 1975. 49
[79] P. Hartman, On homotopic harmonic maps. Canad. J. Math. 19 (1967), 673–687. 51
[80] W. J. Harvey, Geometric structure of surface mapping class groups. In Homological
group theory (Proc. Sympos., Durham, 1977), London Math. Soc. Lecture Note Ser. 36,
Cambridge University Press, Cambridge 1979, 255–269. 95
[81] N. J. Hitchin, The self-duality equations on a Riemann surface. Proc. London Math. Soc.
(3) 55 (1987), 59–126. 70, 71, 72, 73
[82] N. Hitchin, Stable bundles and integrable systems. Duke Math. J. 54 (1987), 91–114. 72
[83] N. J. Hitchin, Lie groups and Teichmüller space. Topology 31 (1992), 449–473. 70
[84] J. Hubbard and H. Masur, Quadratic differentials and foliations. Acta Math. 142 (1979),
221–274. 58
[85] Y. Imayoshi and H. Shiga, A finiteness theorem for holomorphic families of Riemann
surfaces. In Holomorphic functions and moduli, vol. II (Berkeley, CA, 1986), Math. Sci.
Res. Inst. Publ. 11, Springer-Verlag, New York 1988, 207–219. 94
[86] Y. Imayoshi and M. Taniguchi, An introduction to Teichmüller spaces. Springer-Verlag,
Tokyo 1992. 35, 38, 40, 42, 43
[87] T. Ishihara, A mapping of Riemannian manifolds which preserves harmonic functions.
J. Math. Kyoto Univ. 19 (1979), 215–229. 49
[88] N. V. Ivanov, Algebraic properties of the Teichmüller modular group. Dokl. Akad. Nauk
SSSR 275 (1984), 786–789; English transl. Soviet Math. Dokl. 29 (1984), 288–291. 95
[89] N. V. Ivanov, Teichmüller modular groups and arithmetic groups. Zap. Nauchn. Sem.
Leningrad. Otdel. Mat. Inst. Steklov. (LOMI) 167 (1988), 95–110, 190–191; English
transl. J. Soviet Math. 52 (1990), 2809–2818. 95
[90] N. V. Ivanov, Mapping class groups. In Handbook of geometric topology, North-Holland,
Amsterdam 2002, 523–633. 95
[91] N. V. Ivanov, A short proof of non-Gromov hyperbolicity of Teichmüller spaces. Ann.
Acad. Sci. Fenn. Math. 27 (2002), 3–5. 78
[92] A. Jaffe and C. Taubes, Vortices and monopoles. Progr. Phys. 2, Birkhäuser, Boston 1980.
73
[93] J. Jost, Nonlinear methods in Riemannian and Kählerian geometry. DMV Seminar 10,
Birkhäuser, Basel 1988. 51
[94] J. Jost, Harmonic maps and curvature computations in Teichmüller theory. Ann. Acad.
Sci. Fenn. Ser. A I Math. 16 (1991), 13–46. 87
[95] J. Jost, Two-dimensional geometric variational problems. Pure Appl. Math., John Wiley
& Sons, Chichester 1991. 51, 69, 78
[96] J. Jost, Equilibrium maps between metric spaces. Calc. Var. Partial Differential Equations
2 (1994), 173–204. 60, 61
[97] J. Jost, Nonpositive curvature: geometric and analytic aspects. Lectures Math. ETH
Zürich, Birkhäuser, Basel 1997. 65
[98] J. Jost and R. Schoen, On the existence of harmonic diffeomorphisms. Invent. Math. 66
(1982), 353–359. 52
[99] J. Jost and S.-T. Yau, Harmonic maps and group representations. In Differential geome-
try, Pitman Monogr. Surveys Pure Appl. Math. 52, Longman Sci. Tech., Harlow 1991,
241–259. 65
[100] J. Jost and S.-T. Yau, Harmonic mappings and algebraic varieties over function fields.
Amer. J. Math. 115 (1993), 1197–1227. 94
[101] J. Jost and S.-T.Yau, Harmonic maps and superrigidity. In Tsing Hua lectures on geometry
& analysis (Hsinchu, 1990–1991), International Press, Cambridge, MA, 1997, 213–246.
95, 96
[102] V. A. Kaimanovich and H. Masur, The Poisson boundary of the mapping class group.
Invent. Math. 125 (1996), 221–264. 95
[103] J. L. Kazdan and F. W. Warner, Curvature functions for open 2-manifolds. Ann. of Math.
(2) 99 (1974), 203–219. 73
[104] L. Keen, Collars on Riemann surfaces. In Discontinuous groups and Riemann surfaces
(Proc. Conf., Univ. Maryland, College Park, Md., 1973), Ann. of Math. Stud. 79, Prince-
ton University Press, Princeton, N.J. 1974, 263–268. 79
[105] S. P. Kerckhoff, The Nielsen realization problem. Ann. of Math. (2) 117 (1983), 235–265.
85, 86
[106] H. Kneser, Die kleinste Bedeckungszahl innerhalb einer Klasse von Flächenabbildungen.
Math. Ann. 103 (1930), 347–358. 52
[107] H. Koch, Non-euclidean singular integrals and the porous medium equation. Habilitation
thesis, 1999. 92
[108] K. Kodaira, Complex manifolds and deformation of complex structures. Grundlehren
Math. Wiss. 283, Springer-Verlag, Berlin 1986. 45
[109] N. J. Korevaar and R. M. Schoen, Sobolev spaces and harmonic maps for metric space
targets. Comm. Anal. Geom. 1 (1993), 561–659. 54, 60, 61, 63, 66
[110] N. J. Korevaar and R. M. Schoen, Global existence theorems for harmonic maps to non-
locally compact spaces. Comm. Anal. Geom. 5 (1997), 333–387. 54, 60, 63, 64, 66,
74
[111] N. J. Korevaar and R. M. Schoen, Global existence theorems for harmonic maps: finite
rank spaces and an approach to rigidity for smooth actions. Preprint. 66
[112] S. Kravetz, On the geometry of Teichmüller spaces and the structure of their modular
groups. Ann. Acad. Sci. Fenn. Ser. A I No. 278 (1959). 78
[113] E. Kuwert, Harmonic maps between flat surfaces with conical singularities. Math. Z. 221
(1996), 421–436. 55, 56
[114] F. Labourie, Existence d’applications harmoniques tordues à valeurs dans les variétés à
courbure négative. Proc. Amer. Math. Soc. 111 (1991), 877–882. 65
[115] M. L. Leite, Harmonic mappings of surfaces with respect to degenerate metrics. Amer.
J. Math. 110 (1988), 399–412. 55
[116] G. Levitt and F. Paulin, Geometric group actions on trees. Amer. J. Math. 119 (1997),
83–102. 97
[117] P. Li and L.-F. Tam, Uniqueness and regularity of proper harmonic maps. Ann. of Math.
(2) 137 (1993), 167–201. 34
[118] P. Li and L.-F. Tam, Uniqueness and regularity of proper harmonic maps. II. Indiana
Univ. Math. J. 42 (1993), 591–635. 34
[119] M. Linch, On metrics in Teichmüller spaces. Ph.D. Thesis, Columbia University, 1971.
78
[120] K. Liu, X. Sun, and S.-T. Yau, Canonical metrics on the moduli space of Riemann
surfaces. I; II. J. Differential Geom. 68 (2004), 571–637; 69 (2005), 163–216. 90
[121] C. Maclachlan, Modulus space is simply-connected. Proc. Amer. Math. Soc. 29 (1971),
85–86. 82
[122] V. Markovic, Realization of the mapping class group by homeomorphisms. Preprint
2005. 85
[123] H. Masur, On a class of geodesics in Teichmüller space. Ann. of Math. (2) 102 (1975),
205–221. 78
[124] H. Masur, Extension of the Weil-Petersson metric to the boundary of Teichmüller space.
Duke Math. J. 43 (1976), 623–635. 78, 79, 80
[125] H. A. Masur and M. Wolf, Teichmüller space is not Gromov hyperbolic. Ann. Acad. Sci.
Fenn. Ser. A I Math. 20 (1995), 259–267. 78
[126] H. Masur and M. Wolf, The Weil-Petersson isometry group. Geom. Dedicata 93 (2002),
177–190. 83
[127] J. McCarthy, A “Tits-alternative” for subgroups of surface mapping class groups. Trans.
Amer. Math. Soc. 291 (1985), 583–612. 95
[128] J. McCarthy and A. Papadopoulos, Dynamics on Thurston’s sphere of projective mea-
sured foliations. Comment. Math. Helv. 64 (1989), 133–166. 91
[129] J. D. McCarthy and A. Papadopoulos, The visual sphere of Teichmüller space and a
theorem of Masur-Wolf. Ann. Acad. Sci. Fenn. Math. 24 (1999), 147–154. 78, 83
[130] J. D. McCarthy and A. Papadopoulos, The mapping class group and a theorem of Masur-
Wolf. Topology Appl. 96 (1999), 75–84. 78
[131] C. Mese, The curvature of minimal surfaces in singular spaces. Comm. Anal. Geom. 9
(2001), 3–34. 56
[132] C. Mese, Harmonic maps between surfaces and Teichmüller spaces. Amer. J. Math. 124
(2002), 451–481. 56, 60, 61
[133] C. Mese, Harmonic maps into spaces with an upper curvature bound in the sense of
Alexandrov. Math. Z. 242 (2002), 633–661. 56
[134] C. Mese, Uniqueness theorems for harmonic maps into metric spaces. Commun. Con-
temp. Math. 4 (2002), 725–750. 61, 64
[135] C. Mese,A variational construction of the Teichmüller map. Calc. Var. Partial Differential
Equations 21 (2004), 15–46. 56
[136] J. Milnor, On the existence of a connection with curvature zero. Comment. Math. Helv.
32 (1958), 215–223. 73
[137] Y. N. Minsky, Harmonic maps, length, and energy in Teichmüller space. J. Differential
Geom. 35 (1992), 151–217. 53
[138] Y. Miyahara, On some properties of a Teichmüller mapping. TRU Math. 4 (1968), 36–43.
55
[139] N. Mok, Y. T. Siu, and S.-K. Yeung, Geometric superrigidity. Invent. Math. 113 (1993),
57–83. 95, 96
[140] J. W. Morgan and J.-P. Otal, Relative growth rates of closed geodesics on a surface under
varying hyperbolic structures. Comment. Math. Helv. 68 (1993), 171–208. 67, 68
[141] J. W. Morgan and P. B. Shalen, Valuations, trees, and degenerations of hyperbolic struc-
tures. I. Ann. of Math. (2) 120 (1984), 401–476. 75, 77
[142] J. W. Morgan and P. B. Shalen, Free actions of surface groups on R-trees. Topology 30
(1991), 143–154. 58
[143] S. Morita, Characteristic classes of surface bundles. Invent. Math. 90 (1987), 551–577.
85
[144] S. Morita, Geometry of characteristic classes. Transl. Math. Monogr. 199, Amer. Math.
Soc., Providence, R.I., 2001. 85
[145] J. Moser, A Harnack inequality for parabolic differential equations. Comm. Pure Appl.
Math. 17 (1964), 101–134. 50
[146] D. Mumford, A remark on Mahler’s compactness theorem. Proc. Amer. Math. Soc. 28
(1971), 289–294. 82
[147] S. Nag, The complex analytic theory of Teichmüller spaces. Canad. Math. Soc. Ser.
Monogr. Adv. Texts, John Wiley & Sons, New York 1988. 35
[148] M. S. Narasimhan and C. S. Seshadri, Stable and unitary vector bundles on a compact
Riemann surface. Ann. of Math. (2) 82 (1965), 540–567. 70
[149] J. Nielsen, Untersuchungen zur Topologie der geschlossenen zweiseitigen Flächen. I.
Acta Math. 50 (1927), 189–358. 81
[150] J. Nielsen, Abbildungsklassen endlicher Ordnung. Acta Math. 75 (1943), 23–115. 85
[151] N. Nitsure, Moduli space of semistable pairs on a curve. Proc. London Math. Soc. (3)
62 (1991), 275–300. 72
[152] K. Obitsu and S. Wolpert, in preparation. 80
[153] J.-P. Otal, Le théorème d’hyperbolisation pour les variétés fibrées de dimension 3.
Astérisque 235 (1996). 58, 67, 68, 88
[154] R. Palais, unpublished. 44
[155] A. N. Paršin, Algebraic curves over function fields. Dokl. Akad. Nauk SSSR 183 (1968),
524–526; English transl. Soviet Math. Dokl. 9 (1968), 1419–1422. 94
[156] F. Paulin, Topologie de Gromov équivariante, structures hyperboliques et arbres réels.
Invent. Math. 94 (1988), 53–80. 76
[157] E. Reich, On the variational principle of Gerstenhaber and Rauch. Ann. Acad. Sci. Fenn.
Ser. A I Math. 10 (1985), 469–475. 56
[158] E. Reich and K. Strebel, On the Gerstenhaber-Rauch principle. Israel J. Math. 57 (1987),
89–100. 56
[159] B. Rivet, Difféomorphismes harmoniques du plan hyperbolique. Preprint 2001. 34
[160] H. L. Royden, The Ahlfors-Schwarz lemma in several complex variables. Comment.
Math. Helv. 55 (1980), 547–558. 94
[161] H. Royden, unpublished.
[162] J. Sacks and K. Uhlenbeck, The existence of minimal immersions of 2-spheres. Ann. of
Math. (2) 113 (1981), 1–24. 51
[163] J. Sacks and K. Uhlenbeck, Minimal immersions of closed Riemann surfaces. Trans.
Amer. Math. Soc. 271 (1982), 639–652. 51, 87
[164] J. H. Sampson, Some properties and applications of harmonic mappings. Ann. Sci. École
Norm. Sup. (4) 11 (1978), 211–228. 52, 53
[165] J. H. Sampson, Applications of harmonic maps to Kähler geometry. In Complex differen-
tial geometry and nonlinear differential equations (Brunswick, Maine, 1984), Contemp.
Math. 49, Amer. Math. Soc., Providence, R.I. 1986, 125–134.
[166] R. Schoen, Analytic aspects of the harmonic map problem. In Seminar on nonlinear
partial differential equations (Berkeley, Calif., 1983), Math. Sci. Res. Inst. Publ. 2,
Springer, New York 1984, 321–358. 35, 50
[167] R. Schoen, The effect of curvature on the behavior of harmonic functions and mappings.
In Nonlinear partial differential equations in differential geometry (Park City, UT, 1992),
IAS/Park City Math. Ser. 2, Amer. Math. Soc., Providence, R.I. 1996, 127–184. 51, 61,
66
[168] R. Schoen and S. T. Yau, On univalent harmonic maps between surfaces. Invent. Math.
44 (1978), 265–278. 52
[169] R. Schoen and S. T. Yau, Existence of incompressible minimal surfaces and the topology
of three-dimensional manifolds with nonnegative scalar curvature. Ann. of Math. (2) 110
(1979), 127–142. 51, 54, 87
[170] R. Schoen and S. T. Yau, Lectures on harmonic maps. Conference Proceedings and
Lecture Notes in Geometry and Topology, II, International Press, Cambridge, MA, 1997.
52
[171] R. Schoen and M.-T. Wang, in preparation. 98
[172] T. Serbinowski, Boundary regularity of harmonic maps to nonpositively curved metric
spaces. Comm. Anal. Geom. 2 (1994), 139–153. 61
[173] P. B. Shalen, Dendrology of groups: an introduction. In Essays in group theory, Math.
Sci. Res. Inst. Publ. 8, Springer-Verlag, New York 1987, 265–319. 67
[174] C. T. Simpson, Constructing variations of Hodge structure using Yang-Mills theory and
applications to uniformization. J. Amer. Math. Soc. 1 (1988), 867–918. 70
[175] C. T. Simpson, Higgs bundles and local systems. Inst. Hautes Études Sci. Publ. Math.
(1992), 5–95. 72
[176] C. T. Simpson, Moduli of representations of the fundamental group of a smooth projective
variety. I. Inst. Hautes Études Sci. Publ. Math. (1994), 47–129. 72
[177] Y. T. Siu, Curvature of the Weil-Petersson metric in the moduli space of compact Kähler-
Einstein manifolds of negative first Chern class. In Contributions to several complex
variables, Aspects Math. E9, Vieweg, Braunschweig 1986, 261–298. 78
[178] R. K. Skora, Splittings of surfaces. J. Amer. Math. Soc. 9 (1996), 605–616. 67
[179] K. Strebel, Quadratic differentials. Ergeb. Math. Grenzgeb. (3) 5, Springer-Verlag, Berlin
1984. 41, 54, 58, 59
[180] X. Sun, Regularity of harmonic maps to trees. Amer. J. Math. 125 (2003), 737–771. 57,
62, 63
[181] L.-F. Tam and T.Y. H. Wan, Quasi-conformal harmonic diffeomorphism and the universal
Teichmüller space. J. Differential Geom. 42 (1995), 368–410. 34
[182] O. Teichmüller, Gesammelte Abhandlungen. Springer-Verlag, Berlin 1982. 42
[183] A. J. Tromba, On a natural algebraic affine connection on the space of almost complex
structures and the curvature of Teichmüller space with respect to its Weil-Petersson
metric. Manuscripta Math. 56 (1986), 475–497. 78
[184] A. J. Tromba, On an energy function for the Weil-Petersson metric on Teichmüller space.
Manuscripta Math. 59 (1987), 249–260. 87
[185] A. J. Tromba, A new proof that Teichmüller space is a cell. Trans. Amer. Math. Soc. 303
(1990), 257–262. 69
[186] A. J. Tromba, Teichmüller theory in Riemannian geometry. Lectures Math. ETH Zürich,
Birkhäuser, Basel 1992. 35, 44, 69
[187] A. J. Tromba, Dirichlet’s energy on Teichmüller’s moduli space and the Nielsen realiza-
tion problem. Math. Z. 222 (1996), 451–464. 86, 87
[188] A. J. Tromba, Dirichlet’s energy of Teichmüller moduli space is strictly pluri-
subharmonic. In Geometric analysis and the calculus of variations, International Press,
Cambridge, MA, 1996, 315–341. 89
[189] K. Uhlenbeck and S.-T. Yau, On the existence of Hermitian-Yang-Mills connections in
stable vector bundles. Comm. Pure Appl. Math. 39 (1986), S257–S293. 70
[190] T. Y.-H. Wan, Constant mean curvature surface, harmonic maps, and universal Teich-
müller space. J. Differential Geom. 35 (1992), 643–657. 34
[191] M.-T. Wang, A fixed point theorem of discrete group actions on Riemannian manifolds.
J. Differential Geom. 50 (1998), 249–267. 98
[192] M.-T. Wang, Generalized harmonic maps and representations of discrete groups. Comm.
Anal. Geom. 8 (2000), 545–563. 98
[193] R. Wentworth, Regularity of harmonic maps from Riemann surfaces to the Weil–
Petersson completion of Teichmüller space. Preprint 2004. 92
[194] R. Wentworth, Energy of harmonic maps and Gardiner’s formula. In Proceedings of the
Ahlfors-Bers Colloquium (Ann Arbor, Michigan, 2005), to appear. 69, 89
[195] K. Whittlesey, Normal all pseudo-Anosov subgroups of mapping class groups. Geom.
Topol. 4 (2000), 293–307. 93
[196] M. Wolf, The Teichmüller theory of harmonic maps. J. Differential Geom. 29 (1989),
449–479. 53, 77, 87
[197] M. Wolf, Harmonic maps from surfaces to R-trees. Math. Z. 218 (1995), 577–593. 59,
74
[198] M. Wolf, On the existence of Jenkins-Strebel differentials using harmonic maps from
surfaces to graphs. Ann. Acad. Sci. Fenn. Ser. A I Math. 20 (1995), 269–278. 58, 68
[199] M. Wolf, On realizing measured foliations via quadratic differentials of harmonic maps
to R-trees. J. Anal. Math. 68 (1996), 107–120. 58, 68
[200] M. Wolf, Measured foliations and harmonic maps of surfaces. J. Differential Geom. 49
(1998), 437–467. 69
[201] M. Wolf and S. Wolpert, Real analytic structures on the moduli space of curves. Amer.
J. Math. 114 (1992), 1079–1102. 80
[202] S. Wolpert, Noncompleteness of the Weil-Petersson metric for Teichmüller space. Pacific
J. Math. 61 (1975), 573–577. 78
[203] S. A. Wolpert, Chern forms and the Riemann tensor for the moduli space of curves.
Invent. Math. 85 (1986), 119–145. 78
[204] S. A. Wolpert, Geodesic length functions and the Nielsen problem. J. Differential Geom.
25 (1987), 275–296. 86, 88
[205] S. A. Wolpert, The hyperbolic metric and the geometry of the universal curve. J. Differ-
ential Geom. 31 (1990), 417–472. 79, 80
[206] S. A. Wolpert, Geometry of the Weil-Petersson completion of Teichmüller space. In
Surveys in differential geometry (Boston, MA, 2002), Surv. Differ. Geom. 8, International
Press, Somerville, MA, 2003, 357–393. 35, 78, 79, 80, 83, 84
[207] S. A. Wolpert, Convexity of geodesic-length functions: a reprise. In Spaces of Kleinian
groups (Cambridge, UK, 2003), London Math. Soc. Lecture Note Ser. 329, Cambridge
University Press, Cambridge 2006, 233–245. 88
[208] J. W. Wood, Bundles with totally disconnected structure group. Comment. Math. Helv.
46 (1971), 257–273. 73
[209] S. Yamada, Weil-Peterson convexity of the energy functional on classical and universal
Teichmüller spaces. J. Differential Geom. 51 (1999), 35–96. 87
[210] S.Yamada, On the geometry of Weil-Petersson completion of Teichmüller spaces. Math.
Res. Lett. 11 (2004), 327–344. 80
[211] S. T. Yau, A general Schwarz lemma for Kähler manifolds. Amer. J. Math. 100 (1978),
197–203.
[212] S.-K. Yeung, Bounded smooth strictly plurisubharmonic exhaustion functions on Teich-
müller spaces. Math. Res. Lett. 10 (2003), 391–400. 90
[213] S.-K. Yeung, Representations of semisimple lattices in mapping class groups. Internat.
Math. Res. Notices (2003), 1677–1686. 95
[214] H. Zieschang, Finite groups of mapping classes of surfaces. Lecture Notes in Math. 875,
Springer-Verlag, Berlin 1981. 86
Chapter 2
On Teichmüller’s metric and Thurston’s asymmetric

metric on Teichmüller space
Athanase Papadopoulos and Guillaume Théret
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
2 Teichmüller’s metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
2.1 Measured foliations . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
2.2 Conformal structures . . . . . . . . . . . . . . . . . . . . . . . . . . 120
2.3 Moduli and extremal lengths . . . . . . . . . . . . . . . . . . . . . . 124
2.4 The Teichmüller metric . . . . . . . . . . . . . . . . . . . . . . . . . 133
2.5 Teichmüller maps and Teichmüller geodesics . . . . . . . . . . . . . 134
2.6 On the asymptotic behaviour of Teichmüller rays . . . . . . . . . . . 137
2.7 The Teichmüller boundary and convergent rays . . . . . . . . . . . . 138
2.8 Convergence to Thurston’s boundary . . . . . . . . . . . . . . . . . . 139
2.9 On the visual boundary of the Teichmüller metric . . . . . . . . . . . 141
3 Thurston’s asymmetric metric . . . . . . . . . . . . . . . . . . . . . . . . 142
3.1 Hyperbolic structures . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.2 Decompositions by generalized pairs of pants and by ideal triangles . 143
3.3 Geodesic laminations . . . . . . . . . . . . . . . . . . . . . . . . . . 147
3.4 Horocyclic foliations . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.5 Thurston’s cataclysm coordinates . . . . . . . . . . . . . . . . . . . . 154
3.6 Stretch lines, stretch rays, anti-stretch rays and stretch maps . . . . . 164
3.7 Thurston’s asymmetric metric . . . . . . . . . . . . . . . . . . . . . 165
3.8 Comparison with other metrics . . . . . . . . . . . . . . . . . . . . . 172
3.9 Example I of a stretch line: the four-punctured sphere . . . . . . . . . 173
3.10 Example II of a stretch line: the punctured torus . . . . . . . . . . . . 175
3.11 Example III of a stretch line: the closed surface of genus 2 . . . . . . 178
3.12 The behaviour of the lengths of measured geodesic laminations along
stretch and anti-stretch lines . . . . . . . . . . . . . . . . . . . . . . 191
3.13 Thurston’s boundary . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.14 Converging to Thurston’s boundary . . . . . . . . . . . . . . . . . . 193
3.15 Stretch lines and earthquakes . . . . . . . . . . . . . . . . . . . . . . 195
4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
112 Athanase Papadopoulos and Guillaume Théret
1 Introduction
In this chapter, g and n are two nonnegative integers and S = Sg,n is a connected
oriented surface obtained from a closed surface of genus g (which we shall denote by Ŝ)
by removing n points called the punctures.1 We assume that the Euler characteristic
of S is negative. T = Tg,n denotes the Teichmüller space of S. This space carries
several interesting metrics. In this chapter, we shall consider two of them, namely,
Teichmüller’s metric and Thurston’s asymmetric metric. We shall study these metrics
respectively in Sections 2 and 3 below. These two metrics are Finsler metrics, that is,
the distance between two points in Teichmüller space can be defined by minimizing
the lengths of paths joining these points, and lengths of paths are computed by using
a norm defined on the tangent space at each point of T . In the case of Thurston’s
asymmetric metric, the norms on the tangent spaces are not symmetric.
Teichmüller space can be defined either as a space of equivalence classes of con-
formal structures on S or as a space of equivalence classes of hyperbolic structures
on S. Each of the two metrics considered here is natural from one of these points
of view: Teichmüller’s metric from the point of view of conformal geometry and
Thurston’s asymmetric metric from the point of view of hyperbolic geometry. Indeed,
Teichmüller’s metric is natural as a measure of distances between (equivalence classes
of) conformal structures, since it is defined as the logarithm of the least quasiconfor-
mal dilatation of a homeomorphism isotopic to the identity between these conformal
structures, whereas Thurston’s asymmetric metric is natural from the point of view of
measuring distances between (equivalence classes of) hyperbolic structures, since it is
defined as the logarithm of the smallest Lipschitz constant of homeomorphisms iso-
topic to the identity from one hyperbolic structure to the other one. (The order in which
we take the hyperbolic surfaces is important in the last definition because the smallest
Lipschitz constant in one direction is generally different from the smallest Lipschitz
constant in the other direction.) The difference between the conformal and the hy-
perbolic points of view makes the techniques used in the study of these two metrics
on Teichmüller space of different natures: on the one hand, we use complex analysis
(quasiconformal mappings, quadratic differentials, extremal length and so on), and,
on the other hand, we use two-dimensional hyperbolic geometry. But the problems
that we try to solve are formally the same: the global behaviour of geodesic lines (that
is, isometric images of R) in Teichmüller space, e.g. the question of whether they are
properly embedded or not, the convergence of geodesic rays to a point on Thurston’s
boundary, the study of visual boundaries, and other related problems. There are also
several analogies between the results obtained so far for the two metrics. Finally,
there are natural questions about the comparison between the two metrics. These are
reasons for which we present the two metrics in parallel.
Let us start with a few words about some analogies and some differences between
the general features of the two metrics.
1 Such a surface S is said to be of finite type, in reference to the fact that its fundamental group is of finite
type.
Chapter 2. On Teichmüller’s metric and Thurston’s asymmetric metric 113
We recall two descriptions of the Teichmüller distance between two conformal

structures. Consider two conformal structures on a surface. On the one hand, the
Teichmüller distance is the logarithm of the infimum of quasiconformal dilatations
of homeomorphisms isotopic to the identity between the two conformal structures.
On the other hand, this distance is the logarithm of the supremum of quotients of
extremal lengths of closed curves with respect to these structures. Likewise, there are
two descriptions of Thurston’s asymmetric distance from one hyperbolic surface to
another. On the one hand, this distance is the logarithm of the infimum of Lipschitz
constants of homeomorphisms isotopic to the identity between the two hyperbolic
surfaces, and on the other hand it is the logarithm of the supremum of quotients of
hyperbolic lengths of closed geodesics, with respect to the two hyperbolic surfaces.
Teichmüller proved that given two conformal structures on S, there exists, in each
isotopy class of homeomorphisms, a “best quasiconformal stretch homeomorphism”,
that is, a homeomorphism for which the infimum of the quasiconformal dilatation in the
definition of the Teichmüller distance between the two conformal structures is attained.
Likewise, Thurston proved that given two hyperbolic structures on S, there exists, in
each isotopy class of homeomorphisms, a “best Lipschitz stretch homeomorphism”,
that is, a homeomorphism for which the infimum of the Lipschitz constant in the
definition of Thurston’s asymmetric distance is attained.
Teichmüller space, equipped with the Teichmüller metric, is a straight G-space in
the sense of Busemann (cf. Kravetz [33]). This means that any two distinct points in
that space lie on a unique geodesic line. Thurston’s asymmetric metric has a different
character: it is not symmetric (as its name indicates). Furthermore, any two distinct
points in the space lie on a geodesic line, but this geodesic line is not necessarily
unique.
Let us look more closely at the geodesic lines for the two metrics.
Geodesics for Teichmüller’s metric are usually described (since the work of Teich-
müller himself) in terms of quadratic differentials. In this chapter, we have avoided
talking about quadratic differentials, but we have used instead (the equivalent point
of view of) pairs of transverse measured foliations. This is not because we do not
like quadratic differentials, but it is for the sake of stressing a further analogy between
Teichmüller’s metric and Thurston’s asymmetric metric. To describe this analogy, let
us be more precise.
To characterize a geodesic for Teichmüller’s metric, we represent each point in
Teichmüller space by a conformal structure defined by a pair (F1 , F2 ) of transverse
measured foliations on the surface S. The pair of measured foliations defines a “grid”
on the surface, and there is a natural notion of distance measured along the leaves of
each of these foliations, coming from the transverse measure of the other foliation.
Thus, to a pair of transverse measured foliations, it is easy to associate holomorphic
local parameters z = x + iy, the x-direction defined by the leaves of the first foliation
and the y-direction by those of the other one. A geodesic line for Teichmüller’s metric
is then described as a family of surfaces S t = (e−t F1 , et F2 ), t ∈ R. The measured
foliations F1 and F2 are called respectively the horizontal and vertical foliations as-
sociated to the geodesic line (S t )t∈R . This description makes it easy to visualize the
holomorphic coordinates of the surface S t when this surface varies along a geodesic
line. In this way, each conformal structure S t is represented by a Euclidean structure
with isolated conical singular points on the surface, where the transverse measure of
the foliation e−t F1 (respectively et F2 ) determines the Euclidean length on the leaves
of the transverse foliation et F2 (respectively e−t F1 ).
In the case of Thurston’s asymmetric metric, a distinguished class of geodesics
can be described using an object which is less symmetrical than a pair of measured
foliations. This object is a pair consisting of a complete (not necessarily measured)
geodesic lamination, and a measured foliation which is transverse to it. More precisely,
Thurston showed that any two points in Teichmüller space can be joined by a geodesic
(for his asymmetric metric) made up of a concatenation of pieces of “stretch lines”,
a stretch line being a parametrized family of hyperbolic structures S t = (μ, et F ),
t ∈ R, where μ is a complete geodesic lamination on S t and F a measured foliation
on S = S 0 which is transverse to μ. Here also, the pair (μ, et F ) defines a privileged
set of directions on the hyperbolic surface S t , viz the geodesic directions of the leaves
of μ, and the perpendicular directions which are the leaves of et F and which are
made out of pieces of horocycles, with the transverse measure of et F coinciding on
the leaves of μ with hyperbolic length. Thus, the measured foliation et F plays here
the role of a “vertical foliation”, and the complete lamination μ plays the role of a
“horizontal foliation”, associated to the stretch line. Varying the parameter t describes
the stretch line.
These descriptions of geodesic lines for the two metrics on Teichmüller space
lead naturally to similar questions concerning both metrics, as well as to questions
concerning the comparison between them. We now enumerate some of these questions.
• The descriptions of the geodesic rays for Teichmüller’s metric and for Thurston’s
asymmetric metric lead to two distinct natural parametrizations of Teichmüller space
equipped with actions of the group R∗+ of positive reals on the parameter spaces. A
first question related to these parametrizations concerns the study of the extension of
the parameters by adjoining limit points to the orbits of the R∗+ -actions. This leads
to defining boundaries to Teichmüller space, and one natural question is about the
dependence of such a boundary on the chosen parametrization in each case. Other
questions concern the comparison of these boundaries among themselves and with
other geometrically defined boundaries. More precisely, for each of the two metrics,
there is a collection of parameter spaces, each of which being a set of equivalence
classes of measured foliations. In the case of Teichmüller’s metric, a parameter space
is a space of equivalence classes of measured foliations that are transverse to a fixed
measured foliation, and in the case of Thurston’s asymmetric metric, a parameter
space is a space of equivalence classes of measured foliations that are transverse
to a fixed complete geodesic lamination. As such, the two parameter spaces admit
natural R∗+ -action (induced by the action of R∗+ on measures), and the orbits of these
actions correspond to geodesic lines in Teichmüller space, for each of these metrics
respectively. The questions about the extension of the parameters to the boundary
involve the study of the asymptotic behaviour of geodesic rays for each of these
metrics. In each case, the R∗+ -orbits are properly embedded in the parameter space. In
these notes, we present these facts in some detail, as well as results on the following
questions.
• Convergence of geodesic rays. Some of the convergence results are formulated
in terms of a boundary of T which Steve Kerckhoff called Teichmüller boundary,
others in terms of Thurston’s boundary, and others in terms of the visual boundaries of
Teichmüller space. In the early 1980s, Howard Masur obtained results on the conver-
gence of some particular classes of geodesic rays for Teichmüller’s metric to points
on Thurston’s boundary, and Kerckhoff obtained results on the relative behaviour of
pairs of Teichmüller geodesic rays. We present these results below. It seems that there
were no other significant results of that type until a recent work by Anna Lenzhen,
in which she produces a geodesic ray that does not converge to a point on Thurston’s
boundary. The question of the description of the behaviour of an arbitrary geodesic
ray for Teichmüller’s metric with respect to Thurston’s boundary is still open. Con-
cerning the convergence of geodesic rays for Thurston’s asymmetric metric to a point
on Thurston’s boundary, we shall present some recent results.
• The asymptotic behaviour of “anti-stretch” rays. An anti-stretch ray is the neg-
ative part of a stretch line, oriented in the direction opposite to the one given by the
parametrization of the stretch line. Due to the fact that Thurston’s metric is not sym-
metric, an anti-stretch ray is in general not a stretch ray (even after reparametrization).
We note that an anti-stretch ray is (up to reparametrization) a geodesic ray for the
asymmetric metric on Teichmüller space which is “dual” to Thurston’s asymmetric
metric. Here, the dual K ∗ of an asymmetric metric K is defined by the formula
K ∗ (x, y) = K(y, x).
• The asymptotic behaviour of the length of an arbitrary measured geodesic lam-
ination of compact support under a stretch or an anti-stretch ray. More precisely, for
a given family of hyperbolic surfaces (S t )t∈R parametrized by a stretch line and for
any compactly supported measured geodesic lamination α, we are interested in the
existence of the limits limt→∞ lS t (α) and limt→−∞ lS t (α), and whether these limits
are finite or infinite.
We note that in the case of the Teichmüller metric, we have stated some of the
results for the restricted case of closed surfaces, because the written sources for these
results exist only in that special case (although most of these results are certainly valid
in the larger context of surfaces of finite type). For Thurston’s asymmetric metric, we
present the results in the case of surfaces with or without punctures.
The results on the Teichmüller metric that we give are classical, but some of those
on Thurston’s asymmetric metric are new.
At the end of this chapter, we formulate some open problems which concern espe-
cially Thurston’s asymmetric metric.
2 Teichmüller’s metric
References for Teichmüller’s metric can be found in Teichmüller’s collected papers
[62], as well as in those of Ahlfors and of Bers [3], [12], who rewrote part of the theory
and developed it. There are several introductory books on the subject of Teichmüller’s
theory, for instance the books by Abikoff [1] and by Imayoshi and Taniguchi [27]. We
also recommend the recent books by Hubbard [25] and by Fletcher and Markovic [21].
2.1 Measured foliations

We start by recalling a few facts about measured foliations that are used in the theory
of deformation of conformal structures which we present below.
Definition 2.1 (Measured foliation). A measured foliation on S is a foliation with

isolated singularities, equipped with a positive measure on each transverse arc that is
equivalent to the Lebesgue measure of a closed interval of R. (An arc in a surface
is, by definition, a homeomorphic image of the interval [0, 1].) These measures are
invariant by isotopies of the transverse arcs during which each point stays on the same
leaf. The isolated singularities are of the type suggested in Figure 1, and we call them
s-prong singularities, where s can be any integer ≥ 3. We require that at the punctures
of S, the foliation extends as a measured foliation of the unpunctured surface Ŝ in such
a way that each puncture becomes either a nonsingular point, or an s-prong singular
point with s being here any integer ≥ 1 (see Figure 2). Note that in the case s = 2,
the foliation extends as a nonsingular point at the puncture.
Figure 1. The four pictures represent s-prong singular points with s = 3, 4, 5, 6 respectively.
There is an equivalence relation between measured foliations, called Whitehead-

equivalence. It is generated by the following transformations:
• Homeomorphisms of the surface which are isotopic to the identity sending one
foliation to the other and preserving the transverse measures. By abuse of language, we
shall sometimes call a homeomorphism isotopic to the identity an isotopy. We recall
that two homeomorphisms of a surface are isotopic if and only if they are homotopic.
This is a result of Baer [6]; see also Mangler [39] and Epstein [19].
• Whitehead moves: These are deformations of the surface that take place in a
neighborhood of arcs that join two singular points and whose effect is to collapse such
(a) (b) (c)
Figure 2. The three pictures represent the local model for a measured foliation at a puncture
of S. The foliation extends in three possible ways: either as a 1-prong singular point (case (a)),
or as a regular point (case (b)), or as a singular point with at least 3 prongs (case (c)), that is,
like the singular points at interior points of the surface.
an arc to a point. (Remember that by our definition, an arc is embedded, which implies
that after such a collapse, the surface remains a surface.) An example of a Whitehead
move is given in Figure 3. Again, these moves are required to respect the transverse
measures. The inverse move of a Whitehead move is also called a Whitehead move.
Figure 3. Whitehead move: collapsing or creating an arc joining two singular points.
Note that a singular point involved in a Whitehead move can be at a puncture of S,

and that it is sometimes possible to eliminate a 1-prong singularity at a puncture by
using a Whitehead move; see for instance Figure 4.
We let MF (S) = MF denote the set of equivalence classes of measured foliations
on S. An element of MF is called a measured foliation class.
If x is a positive real number and if F is a measured foliation, then xF denotes the
foliation F (as a topological object), equipped with the transverse measure obtained
by multiplying the original transverse measure of F by the factor x. This action is
compatible with the Whitehead equivalence relation, and it induces an action of R∗+ on
the set MF . The quotient of MF by this action is denoted by PMF (S) or PMF .
An element of PMF is called a projective measured foliation class.
Figure 4. Eliminating a 1-prong singularity at a puncture of S by applying a Whitehead move.
If F is a measured foliation, then [F ] will usually denote its equivalence class

in MF . We shall also sometimes use the same notation, [F ], for the corresponding
element in PMF .
Measured foliations already appear in Teichmüller’s work as horizontal and vertical
line fields associated to holomorphic quadratic differentials on Riemann surfaces,
but it was Thurston who initiated their systematic study, and defined the space of
equivalence classes of measured foliations, in his paper On the geometry and dynamics
of diffeomorphisms of surfaces ([69], published several years after it has been written).
For a complete presentation of these results in the case of closed surfaces, we refer
the reader to [20].
A closed curve in S is called essential if it is not homotopic to a point or to a puncture,
and it is called simple if it has no self-intersection, that is, if it is an embedded image
of a circle.
Let S be the set of homotopy classes of unoriented essential simple closed curves
in S.
There is a natural embedding
S → MF (2.1)
defined as follows. For any element γ in S, we take a foliated cylinder C embedded
in S, whose leaves are simple closed curves that are in the homotopy class γ . This
foliated cylinder C defines a partial foliation on S. Here, the adjective partial means
that the support of the foliation is a subset of the surface S. We choose an arc c that
joins the two boundary components of C and which is transverse to the foliations,
and a homeomorphism between this arc and the interval [0, 1] ⊂ R, and we equip
the arc c with the pull-back of the Lebesgue measure of [0, 1]. We then take the only
invariant transverse measure for the foliation on C that induces the given Lebesgue
measure on the arc c. We shall sometimes say that C, equipped with this measured
foliation, is a foliated cylinder of height one with core curve in γ . To get a measured
foliation on S, we collapse the closure of each connected component of S \ C onto a
spine. We recall that a spine of a compact surface with boundary is a one-dimensional
complex on which the surface collapses by a homotopy equivalence (see [20], p. 90).
Here, the spine is chosen so that the singular points of the resulting foliation on S
are of the allowed type. Note that the exposition in [20] works for the case where S
is compact. To deal with the case where S is noncompact (that is, the case where S
has punctures), we can use the spine of a compact surface with boundary obtained by
replacing each puncture by a boundary closed curve and then collapsing that boundary
curve to a point. We do this in such a way that the local model of the resulting measured
foliation on the original surface S is of the allowed type, at the interior singularities and
at the punctures. The various choices of spines for the complementary components of
the cylinder C differ precisely by Whitehead moves performed on spines. Using this
fact, to any element of S, we associate a measured foliation on S which is well defined
up to isotopy and Whitehead moves. It is a fact of the classical theory of measured
foliations that the map defined in (2.1) is injective (see [20], p. 89).
Note that in the case where the surface S is the three-punctured sphere, the set S is
empty. As is well known, the Teichmüller space in that case consists of a single point,
and there is not a lot more to say. Therefore, we discard this case in what follows.
Definition 2.2 (Measured foliation with one cylinder). A measured foliation (respec-
tively a measured foliation class) which up to a constant factor is obtained from
an element of S by the map described in (2.1) will be called a measured foliation
(respectively a measured foliation class) with one cylinder.
Equivalently, a measured foliation with one cylinder is a measured foliation such

that when extended to the unpunctured surface Ŝ, the union of the leaves starting
at the singular points, with the singular points included, is a compact graph whose
complement is connected. This graph is called the critical graph of the foliation.
The height of a measured foliation with one cylinder is defined as the total mass
of a transverse arc joining the two boundary components of that cylinder.
We need to recall another description of the equivalence relation between measured
foliations.
To each measured foliation F , we associate a map i(F, · ) : S → R+ defined for
each γ in S by the formula
i(F, γ ) = inf I (F, c) (2.2)
c∈γ
where the infimum is taken over all closed curves c that are in the homotopy class γ
and that are made up of a finite concatenation of segments which are either contained
in leaves of F or are transverse to F , and where I (F, c) denotes the total mass (with
respect to the transverse measure of F ) of all the sub-segments of c that are transverse
to F .
In this way, a measured foliation defines an element of RS+ , the set of functions
from S to R+ . Two measured foliations are said to be measure-equivalent if they have
the same image in RS+ .
The next two theorems summarize some basic results of Thurston that we shall
refer to in the rest of the chapter.
Theorem 2.3 (Thurston, see [20], p. 110). Two measured foliations are Whitehead-
equivalent if and only if they are measure-equivalent.
Theorem 2.3 gives an embedding of the space MF in the function space RS+ , and
with this embedding we have the following:
Theorem 2.4 (Thurston, see [20]), p. 117 and 150). With the topology induced on
the space MF by its embedding in the function space RS+ equipped with the weak
topology, MF is homeomorphic to R6g−6+2n \ {0}. The set of elements in MF that
are of the form xγ with x ∈ R∗+ and γ ∈ S (where γ is regarded as an element
of MF by the embedding defined in (2.1) above) is a dense subset of MF . The
projectivized space PMF , equipped with the quotient topology, is homeomorphic to
a (6g − 7 + 2n)-dimensional sphere S 6g−7+2n , in which the natural image of S is
dense.
2.2 Conformal structures

We start with a word about atlases.
We shall define a conformal structure and, later on, a hyperbolic structure on S as
an atlas. An atlas is a certain collection of local charts satisfying a certain property,
and it is possible to form unions of collections of local charts. We shall say that two
atlases are compatible if their union is an atlas satisfying the required properties. A
maximal atlas is then a maximal union of compatible atlases.
Definition 2.5 (Conformal structure). A conformal structure (which we shall also

call a conformal atlas) on S is a maximal atlas {(Ui , φi )}i∈ of local charts where
for each i ∈ , Ui is an open subset of S and φi is ahomeomorphism from Ui
onto an open subset of the complex plane C satisfying i∈ Ui = S and such that
for all i and j in , the map φi φj−1 , which is called a coordinate change map or
transition map, and which is defined on φj (Ui ∩ Uj ), is conformal. Furthermore, we
suppose that each puncture of S has a neighborhood which is conformally equivalent
to a punctured disk in C. (Without this condition, the neighborhood of a puncture
could also be conformally a cylinder.) We shall sometimes use the word holomorphic
instead of conformal. Each pair (Ui , φi ) is called a holomorphic chart, and (by abuse
of language) the variable z ∈ φi (Ui ) is called a holomorphic local coordinate for the
structure. On each domain Ui of a holomorphic chart, there is a natural orientation
induced by the map φi from the usual orientation of C. The transition functions of
a conformal atlas, being holomorphic, are orientation preserving. Thus, a conformal
structure on S equips S with a canonical orientation. We shall always assume that this
orientation coincides with the orientation on S that we started with.
A surface equipped with a conformal structure is also called a Riemann surface.

Example 2.6 (Conformal structure induced by a Riemannian metric). Any Rieman-

nian metric on S has an underlying conformal structure. This is usually deduced from
the existence of the so-called isothermal coordinates, that is, local coordinates z in
which the Riemannian metric can be written as ds 2 = λ(z)|dz|2 , with λ(z) > 0,
see [5], p. 124–126, and the references therein. Isothermal coordinates have been
investigated by Gauss, who proved their existence under some restrictive conditions
which have been weakened later on. Conversely, every conformal structure on S is
induced by some Riemannian metric. Furthermore, under our requirement that the
Euler characteristic of S is negative, such a metric can be chosen to be hyperbolic,
that is, of constant Gaussian curvature −1. The existence of such a metric is the “uni-
formization theorem” for surfaces of negative Euler characteristic, attributed to Klein,
Poincaré and Koebe, which is also based on the existence of isothermal parameters,
cf. [57] and the references therein. More precisely, the uniformization theorem states
that each conformal surface of finite type and of negative Euler characteristic can be
realized as a quotient of the hyperbolic plane H2 by a discrete group of isometries
of H2 . The conformal surface is then induced by a hyperbolic metric. The hyperbolic
metric is unique up to isotopy, and this makes hyperbolic metrics play a particularly
important role in the theory of deformations of conformal structures. We shall use the
hyperbolic point of view in Section 3 below.
The connection between a hyperbolic metric and the underlying conformal struc-
ture is not easy to handle, and classical Teichmüller theory (that is, the theory based
on the techniques developed by Teichmüller) makes little use of hyperbolic geometry.
However, there is a class of metrics which is more useful in the conformal theory, be-
cause the conformal structures that underly them are convenient to manipulate; these
are the singular flat metrics that are defined by pairs of transverse measured foliations
on the surface, of which we now recall the definition.
Example 2.7 (Conformal structure defined by a pair of transverse measured foliations).

Let F1 and F2 be two transverse measured foliations on S. Recall from Definition 2.1
that each measured foliation extends to a measured foliation on the closed surface Ŝ
obtained from S by filling in the punctures. We require here that at each point of Ŝ \ S,
both F1 and F2 extend in the same manner, that is, both of them extend as s-prong
singularities with the same s ≥ 1. The local model for two transverse foliations F1
and F2 at a puncture of S is represented in Figure 6.
Such a pair of transverse measured foliations defines a conformal structure on S,
and in fact, a distinguished class of holomorphic parameters, in the following way.
We identify the neighborhood of each point on S which is a nonsingular point of the
foliations to a subset of C by using a parameter z = x + iy ∈ C, where x is a variable
along the leaves of F1 and y is a variable along the leaves of F2 , with the distance
along a leaf of each foliation being measured using the transverse measure of the other
foliation. Furthermore, we suppose that the positive orientation of x followed by the
positive orientation of y coincides with the orientation of the surface S. The local
parameter z (called a distinguished parameter) is defined up to the transformation
Figure 5. Each picture represents two transverse foliations near a singular point in S.
(a) (b) (c)
Figure 6. Each picture represents an extension of two transverse foliations at a puncture of S: in

case (a) each foliation extends as a 1-prong singular point, in case (b) it extends as a nonsingular
foliation and in case (c) it extends as a 3-prong singular point.
z
→ −z and up to a translation in C. In that way, the leaves of F1 (respectively F2 )
are locally defined by the equation y = constant (respectively x = constant). The fact
that the local parameters z = x + iy are well defined at each point up to sign and up
to the addition of a complex number implies that the local parameters associated to
the various nonsingular points are compatible with each other from the holomorphic
point of view, and they define a conformal structure in the complement of the singular
points. Now we have to see that this conformal structure extends to the whole surface.
Let p be a singular point of F1 (or, equivalently, F2 ) and let s be the number of prongs
at that point. On each small enough disk neighborhood V (p) of p in S, we can define
a map φp : V (p) → C which sends p to 0 and which at each point of V (p) \ {p}
coincides with a branch of the map z
→ z2/s , z being the distinguished parameter.
This map is well defined up to composition in the range by a rotation of C that
fixes the origin. The maximal atlas generated by the collection of distinguished local
charts at the nonsingular points of S, together with the maps (V (p), φp ) associated
to the various singular points, defines a conformal structure on S. (We note that the
condition we imposed on the way the foliations extend to the punctures of S ensures
that the requirement in Definition 2.5 that each puncture has a neighborhood which is
conformally equivalent to a punctured disk in C is satisfied.)
Since in each small enough neighborhood of each point in S, the distinguished
parameter in C is unique up to addition of a complex constant and to multiplication
by −1, such a parameter defines a metric on that neighborhood, which makes the
neighborhood isometric to an open subset of C equipped with its Euclidean metric.
Thus, if Z denotes the set of singular points of F1 and F2 , the surface S \ Z is equipped
with a canonical Euclidean metric. The leaves of the foliations F1 and F2 are geodesic
with respect to that metric, and the foliations are mutually orthogonal. Such a metric is
called a singular flat metric or a Euclidean metric with cone singularities, the singular
points being the cone points.
Let Hom(S) be the group of orientation-preserving homeomorphisms of S. We

consider the following action of Hom(S) on the set of conformal structures on S: for
any maximal atlas G = {(Ui , φi )}i∈ and for any orientation-preserving homeomor-
phism f : S → S, f ∗ G is the maximal atlas {(Ui , φi )}i∈ where for each i in ,
Ui = f −1 (Ui ) and φi = φi f|Ui . The maximal atlas f ∗ G is then a conformal atlas
on S.
If G and H are any two conformal structures on S, we shall sometimes denote
(by abuse of language) f : G → H as a map f : S → S in which the domain space
(respectively the target space) is the surface S equipped with the conformal structure
G (respectively H ).
We shall say that two conformal structures G and H are equivalent if there exists
a homeomorphism f : G → H isotopic to the identity satisfying f ∗ H = G.
Definition 2.8 (Teichmüller space). The Teichmüller space of S, which we shall denote
by Tg,n , Tg,n (S), T or T (S), is the space of equivalence classes of conformal structures
on S.
The subgroup Hom0 (S) of Hom(S) consisting of the homeomorphisms that are
isotopic to the identity is a normal subgroup, and the quotient group MCG(S) =
Hom(S)/Hom0 (S) is called the mapping class group of S. It has a natural action on
the Teichmüller space Tg,n (S), which is the quotient action of that of Hom(S) on the
set of maximal conformal atlases defined above.
The set of conformal structures on S is equipped with a natural topology in
which two conformal structures G and H are close if we can find two coverings
{(Ui , φi )}i∈ and {(Vi , ψj )}j ∈J , where for each i ∈ (respectively j ∈ J), (Ui , φi )
(respectively (Vj , ψj )) is a holomorphic chart for G (respectively H ), such that
−1
i∈ Ui = i∈J Vj = S and such that any map of the form φi ψj (with the
∞
usual convention for its domain of definition) is C -close to the identity. (We are
using the fact that a conformal structure on S defines a canonical C ∞ -structure on that
surface.) Teichmüller space Tg,n (S) is equipped with the quotient topology. There are
many nice ways of describing that topology, which show that this space is homeomor-
phic to R6g−6+2n . In particular, this topology is induced by the Teichmüller metric
which is one of the main subject matters of this chapter (Definition 2.23 below), but
we shall also see several other descriptions of this topology.
2.3 Moduli and extremal lengths

We shall talk about quadrilaterals and their moduli in Riemann surfaces, but it is natural
to start with a few words about triangles, which are simpler objects.
A topological triangle T in a Riemann surface S is an embedded closed disk
with three distinguished pairwise distinct points on its boundary. Such an object is
equipped with a conformal structure, inherited from that of S. Strictly speaking, T
is equipped with a conformal structure with boundary, which is defined as in Defini-
tion 2.5 above, except that instead of requiring the range of each homeomorphism φi
to be an open subset of C, we require it to be a relatively open subset of the closed
half-plane {z ∈ C | Im(z) ≥ 0}. To simplify the exposition, we shall suppose that
each time we consider a closed disk equipped with a conformal structure and a set of
distinguished points on its boundary, these points are taken in a cyclic order which is
compatible with the usual orientation on the boundary of the disk that is induced from
the orientation of the disk induced from the orientation of the surface. Now any two
topological triangles equipped with conformal structures are conformally equivalent,
that is, there exists a conformal homeomorphism between the two disks that respects
the distinguished points. This follows from the Riemann mapping theorem, which
says that any simply connected open subset of the plane bounded by a Jordan curve
can be mapped conformally onto the unit disk, that this mapping can be extended to a
homeomorphism between the closures of the domains, that the extended homeomor-
phism is also conformal and that if we choose arbitrarily three distinguished pairwise
distinct points in the boundary of the two disks, then the conformal map between the
two closed disks can be taken so as to respect the distinguished points, and, finally,
that this condition completely determines the map (see [34], Chapter 1).
Thus, topological triangles cannot be used to distinguish different Riemann sur-
faces, and the next objects of study along that line are topological disks with four
distinguished points on their boundary instead of three. These objects are not all
conformally equivalent, and they are classified by a conformal invariant called the
modulus. In fact, it is sometimes more convenient to deal with closed disks with two
disjoint distinguished closed arcs (instead of four distinguished points) on their bound-
ary. The four boundary points of the two arcs will be the distinguished points. Thus,
we call a quadrilateral in a Riemann surface S an embedded closed disk with two
distinguished disjoint closed arcs in its boundary. We shall call the distinguished arcs
the vertical sides of the quadrilateral. In the same way as for topological triangles, a
quadrilateral in S is equipped with a conformal structure with boundary induced from
that of S.
Definition 2.9 (Modulus of a quadrilateral). By the Riemann mapping Theorem, for

any quadrilateral Q in a Riemann surface, there exists a unique positive real number
Mod(Q) together with a conformal homeomorphism φ from Q to the rectangle R in
the Euclidean plane R2 with vertices at (0, 0), (Mod(Q), 0), (Mod(Q), 1) and (0, 1),
such that φ sends the vertical sides of Q to the vertical sides of R (that is, the sides of
length 1, see Figure 7). The value Mod(Q) is called the modulus of Q (and of R).
Figure 7. There is a conformal map from the quadrilateral Q on the left to the Euclidean rectangle
on the right, sending the vertical sides of Q (which are drawn in bold lines) to the vertical sides
of the Euclidean rectangle. The Euclidean rectangle is unique up to isometry, if we take the
lengths of its vertical sides to be equal to 1.
Remarks. 1) Explicit formulae for conformal mappings between quadrilaterals having

the same modulus are usually given by means of elliptic integrals (see [34]).
2) Another invariant of quadrilaterals in Riemann surfaces is the cross ratio of
the four distinguished points. More precisely, one starts by mapping conformally
the quadrilateral on the closed unit disk in the complex plane (by using the Riemann
mapping associated to the interior of the quadrilateral which, as we recalled, extends
to the boundary) and then taking the cross ratio of the images of the four distinguished
points that we obtain on the boundary of that disk (see [4], p. 343).
3) One can define the modulus of a quadrilateral without using the Riemann map-
ping theorem (see [34]).
4) After moduli of topological triangles and quadrilaterals, one can study moduli
of n-gons in Riemann surfaces. Here, an n-gon is defined as a topological disk with n
distinguished points on its boundary. The dimension of the space of moduli (real
parameters) of an n-gon is n − 3. More generally, Ahlfors and Beurling initiated the
study of conformal invariants (or moduli) of arbitrary domains in the plane which are
bounded by finitely many closed curves, with a finite number of distinguished points
on their boundary and in their interior. The dimension of the moduli space of such
a domain is 3n + 2p + q − 6, where n is the number of boundary components, p
the number of distinguished points in the interior and q the number of distinguished
points on the boundary (see [4], p. 342).
Now back to quadrilaterals.
If G and H are Riemann surfaces and if f : G → H is a homeomorphism, then

f transforms any quadrilateral in G into a quadrilateral in H . If f is conformal, it
preserves the moduli of quadrilaterals. If f : G → H is a general homeomorphism,
then the defect in conformality of f is measured by a quantity which is called the
quasiconformal dilatation of f , defined as follows.
Definition 2.10 (Quasiconformal homeomorphism and quasiconformal dilatation).

Let G and H be two Riemann surfaces and let f : G → H be an orientation-preserving
homeomorphism. Then f is said to be quasiconformal if we have
Mod(f (Q))
K(f ) = sup < ∞,
Q Mod(Q)
where the supremum is taken over all quadrilaterals Q in G. The value K(f ) is
called the quasiconformal dilatation of f . For every K ≥ K(f ), f is said to be
K-quasiconformal homeomorphism.
Notice that if Q is a given quadrilateral and if Q is the quadrilateral obtained

from Q by keeping the same topological disk and the same set of distinguished points,
but applying to these points an order-one translation of the indices, then we have
Mod(Q) = 1/Mod(Q ). With this remark, we can see that the value of K(f ) in
Definition 2.10 is always ≥ 1.
Definition 2.10 says in a geometric manner that a map is quasiconformal if it has
uniformly bounded distorsion. To see that this definition is equivalent to other (more
commonly used) analytic definitions of quasiconformal homeomorphisms, we refer
the reader to the paper [10] by Bers.
Remark 2.11. In the case where f is a C 1 -diffeomorphism and where the conformal
structures G and H are induced by Riemannian metrics, then, for each z in G, the
differential of f at z, being an R-linear map, takes a circle centered at the origin in the
tangent space Tz G of G at z to an ellipse centered at the origin in the tangent space
Tf (z) H of H at f (z). The (local) quasiconformal dilatation of f at z, denoted by
Kz (f ), is defined as the ratio of the major axis to the minor axis of that image ellipse.
This ratio does not depend on the choice of the circle in Tz G centered at the origin
that we started with. The quasiconformal dilatation of f is then equal to
K(f ) = sup Kz (f ).
z∈G
Finally, we note that the quasiconformal dilatation of f at z is also given by the formula
sup{dfz (u) such that u ∈ Tz S, u = 1}
Kz (f ) = .
inf{dfz (u) such that u ∈ Tz , u = 1}
In this formula, the norm of the tangent vector u (respectively dfz (u)) is measured
with respect to the Riemannian metric defining G (respectively H ).
The study of quasiconformal dilatations of homeomorphisms between Euclidean

rectangles was initiated by H. Grötzsch around 1928, who wrote several papers on that
subject. Grötzsch showed that the (real-) affine homeomorphism between two arbitrary
Euclidean rectangles realizes the minimum of the quasiconformal dilatation among all
homeomorphisms that respect the sides of these rectangles (see for instance [23]). This
result was one of the starting points of the theory of extremal quasiconformal mappings
between general Riemann surfaces and in fact the affine map between rectangles is a
building block for that general theory. Indeed, it follows from Teichmüller’s results
(which we shall recall in §2.5 below) that for any two conformal structures G and H
on S, one can find two decompositions {R1 , . . . , Rk } and {R1∗ , . . . , Rk∗ } of that surface
and a homeomorphism f : G → H (which is called the Teichmüller map) such that
the following conditions hold:
(1) f realizes the minimum of the quasiconformal dilatation in the isotopy class of
the identity, and it is the unique map that has this property;

(2) for each i = 1, . . . , k, Ri and Ri∗ are quadrilaterals satisfying ki=1 Ri =
k ∗
i=1 Ri = S;
(3) for each i = j , the interiors of Ri and Rj and of Ri∗ and Rj∗ are disjoint;
(4) for all i = 1, . . . , k, the map f sends the quadrilateral Ri to the quadrilateral

Ri∗ , and if fi : Ri → Ri (respectively fi∗ : Ri∗ → Ri∗ ) is the G- (respectively the
H -) conformal homeomorphism sending Ri (respectively Ri∗ ) onto a Euclidean
rectangle as in Definition 2.9, then, the map fi∗ f|Ri fi−1 is an affine map;
(5) the quasiconformal dilatations of the affine maps fi∗ f|Ri fi−1 are equal for
all i = 1, . . . , k.
It is in this sense that the affine map between two rectangles is a basic model for
the Teichmüller map between arbitrary Riemann surfaces.
Notice that the local quasiconformal dilatation Kz (f ) of an affine homeomorphism
f between Euclidean rectangles is constant (independent of z), and that the local qua-
siconformal dilatation on each of the quadrilaterals of the decomposition {R1 , . . . , Rk }
is the same (independently of the choice of the quadrilateral). This is a basic property
of the Teichmüller maps between Riemann surfaces.
Examples. 1) Consider the surface obtained by doubling a Euclidean rectangle along
its open edges (that is, without the vertices). This surface is the four-punctured sphere
S = S0,4 , and it is equipped with two transverse measured foliations obtained by
gluing the linear vertical (respectively horizontal) foliations of the two rectangles we
started with, equipped with the transverse measures that induce Lebesgue measure on
the edges of the rectangles. Each of these foliations extends as a one-prong singular
point at each of the four punctures. We choose a common x and y-coordinate on
the sides of the two rectangles. For each λ > 0 the real-affine map defined by
(x, y)
→ ((1/λ)x, λy) is an extremal quasiconformal map on each rectangle, and the
two maps fit together well and define a map on S = S0,4 which is a Teichmüller map.
2) More generally, we can take an arbitrary (simply connected or not) closed subset
of the Euclidean plane forming a subsurface with boundary with each boundary curve
being made of a finite number of vertical and horizontal segments (see Figure 8 for
an example). Doubling this surface, excluding the vertices, as in Example 1 above,
produces a Riemann surface equipped with a canonical product foliation structure (see
Definition 2.16), the one induced by the vertical and horizontal measured foliations
of the plane. As in the particular case considered above, for each λ > 0, one can
easily visualize a Teichmüller map on that surface, as a map induced by an affine
transformation of the Euclidean plane. Notice that in all these examples, a singular
point of each of the horizontal and vertical measured foliation is either a 1-prong or
3-prong singularity. At a 3-prong singularity, we can close the puncture, that is, we
can include in the gluing of the two planar surfaces that we started with the endpoints
of the edges that abut on that singular point. However, there are necessarily 1-prong
singularities left, and we cannot obtain closed surfaces with this kind of construction.
But it is easy to construct closed surfaces equipped with measured foliation pairs (see
for instance the examples in the chapter by Herrlich and Schmithüsen in this Handbook
[24]).
Figure 8. Doubling two such shaded regions along the open edges gives examples of surfaces
with punctures equipped with measured foliation pairs. (The foliations we are talking about are
those induced by the horizontal and vertical foliations of the plane.)
Definition 2.12 (Modulus of a topological cylinder). Let S be a Riemann surface

and let C be a topological cylinder in S, that is, a surface homeomorphic to S 1 × I
immersed in S, with its interior embedded. Such a cylinder is equipped with an induced
conformal structure with boundary, and it is conformally equivalent to a Euclidean
cylinder C ∗ which is unique up to scaling (see [2]). The modulus of C is the height
of the Euclidean cylinder C ∗ divided by its circumference. It is denoted by ModS (C)
or Mod(C).
Definition 2.13 (Modulus and extremal length of a homotopy class of curves). Let S
be a Riemann surface and let γ be a homotopy class of essential simple closed curves
in S. The modulus of γ , denoted by ModS (γ ) or Mod(γ ), is the supremum of the
moduli of topological cylinders in S with core curve in the class γ . The extremal
length of γ , denoted by ExtS (γ ) or Ext(γ ), is defined as 1/Mod(γ ).
The notion of extremal length of a family of curves in a Riemann surface was first
defined by Beurling and it was developed later on by Beurling and Ahlfors, see [4].
The following result provides examples of Riemann surfaces in which one can see
explicitly the maximal modulus cylinder in a given homotopy class.
Theorem 2.14. Let S be a closed surface equipped with a conformal structure defined
by two transverse measured foliations F1 and F2 as in Example 2.7. Suppose that F1
is a foliation with one cylinder, and let γ be the homotopy class of the core curve of
that cylinder. Then, the cylinder C in S whose interior is the complement of the critical
graph of F1 has the largest modulus among all topological cylinders in S with core
curve in the homotopy class γ . Thus, in this situation, we have Mod(γ ) = Mod(C)
or, equivalently, Ext(γ ) = 1/Mod(C).
Theorem 2.14 is stated (in an equivalent form) by Kerckhoff in [29] (Theorem 3.1)
and it is attributed there to J. A. Jenkins and K. Strebel.
There is an extension of this result to the extremal length of an arbitrary measured
foliation class instead of a homotopy class of a simple closed curve. This is Theo-
rem 2.21 below and, before stating it, we need to recall Kerckhoff’s extension of the
notion of extremal length from the set S of homotopy classes of simple closed curves
to the space MF of equivalence classes of measured foliations. (We are using here
the inclusion S ⊂ MF defined in (2.1).)
Theorem 2.15 (Kerckhoff [29]). Let S be a closed surface. The extremal length
function defined on the set S of homotopy classes of essential simple closed curves in
S extends in a unique way to a continuous function extS = ext : MF → R∗+ satisfying
ext(xF ) = x 2 ext(F ) for every x in R∗+ and for every F in MF .
Definition 2.16 (Product foliation structure). Let Q = Q(S) be the subset of

MF × MF consisting of pairs of measured foliation classes ([F1 ], [F2 ]) such that
[F1 ] and [F2 ] are representable by two foliations F1 and F2 that are transverse. We
shall call an element of Q a product foliation structure. The space Q is equipped with
the topology induced from the weak topology on MF × MF .
The following lemma will be useful in considerations about product foliation struc-
tures.
Lemma 2.17 (Masur [41]). Let S be a closed surface and let [F1 ] and [F2 ] be two
measured foliation classes on S that can be represented by two transverse foliations
F1 and F2 . Then the pair (F1 , F2 ) is unique up to isotopy.
In particular, we cannot perform Whitehead moves on F1 or on F2 while keeping

the transversality.
Remark 2.18 (Product foliation structures and quadratic differentials). We must men-
tion that a product foliation structure can be regarded as a Riemann surface equipped
with a holomorphic quadratic differential, although we do not make use of this fact
in this chapter. We recall that a holomorphic quadratic differential q on a Riemann
surface S is an invariant object that has an expression qi (z)dz2 in each holomorphic
chart (Ui , φi ), where qi (z) is a holomorphic function of z, the holomorphic local
coordinate in φi (Ui ). Invariance of q means that if (Uj , φj ) is another holomorphic
chart, if w is the holomorphic local coordinate in φj (Uj ), and if a local expression
of q in that chart is qj (w)dw 2 , then at the overlap between the two charts we have
qi (z)dz2 = qj (w)dw 2 or, equivalently, qi (z)(dz/dw)2 = qj (w). This invariance
property implies in particular that the zeroes of q and their orders are well defined.
(In other words, the zeroes of the functions qi (z) and their orders are independent
of the choice of the local chart, see [59], p. 18). Now to each pair (F1 , F2 ) of trans-
verse measured foliations on S is associated a conformal structure on that surface,
as in Example 2.7, with local coordinates z = x + iy which are well defined up to
sign and up to composition by a translation in C. From that we deduce that the local
holomorphic quadratic differential form dz2 is the local expression of a well-defined
holomorphic quadratic differential form on S (that is, the invariance property is sat-
isfied). The norm of a quadratic differential is defined as q = |q(z)|dxdy (this
uses the fact that the area element dxdy is independent of the choice of the coordinate
chart), and the condition that we imposed on the foliations F1 and F2 at the punctures
of S (see Definition 2.1) insures that the associated quadratic differential has finite
norm. Conversely, to each conformal structure on S equipped with a holomorphic
quadratic differential of finite norm, one can naturally associate a Euclidean metric
with cone singularities, and a horizontal and a vertical foliation defining an element
of the space Q(S), these two foliations being orthogonal and their leaves being local
geodesics with respect to that Euclidean metric. Let us briefly recall the definitions
of these foliations and of the Euclidean structure. Let q be a nonzero holomorphic
quadratic differential on S, let (Ui , φi ) be a holomorphic chart in S, and let qi (z)dz2
be an expression of q in that chart. If z0 = φi (x) is the image by φi of a point x
which is a nonzero point of q, then, since qi (z0 ) = 0, taking if necessary a smaller
1
neighborhood Ui of z0 , we may define a branch qi2 of the square root of qi on that
z 1
neighborhood. The integral Qi (z) = z0 qi (v) 2 dv is then a holomorphic function
in z and it determines a new holomorphic chart for S at the point x. The parameter
w = Qi (z) is called a distinguished parameter for q at the nonzero point x. In terms
of that parameter, the expression of q is dw 2 . (Remember that the differential dw is
only defined up to sign, but that dw 2 is well defined.) The distinguished parameter w
is unique up to addition of a complex constant and multiplication by −1. Therefore,
it establishes an isometry between its domain in S and an open subset of C equipped
with its Euclidean metric. Thus, if Z(q) ⊂ S denotes the set of zeroes of the quadratic
differential q, the surface S \ Z(q) is equipped with a canonical Euclidean structure
induced by q. At a zero of q of order p, the Euclidean structure has a cone singularity
of cone angle (p + 2)π . On S \ Z(q), the vertical (respectively horizontal) foliation
is defined locally by taking a distinguished parameter w = u + iv and transporting on
the domain of this local chart the foliation by vertical (respectively horizontal) lines
u = constant (respectively v = constant) in C. Since the distinguished parameters
are unique up to sign and up to translation in C, these foliations on the various chart
domains match up and give a well-defined vertical (respectively horizontal) foliation
on S \Z(q). It also follows from this construction that the leaves of these foliations are
local geodesics and are orthogonal with respect to the canonical Euclidean structure
on S \ Z(q) induced by q. Of course, this Euclidean structure is the same as the
Euclidean structure associated to the pair of transverse measured foliations defined
in Example 2.7. For more details on quadratic differentials, we refer the reader to
Strebel’s book [59] and to other chapters of this Handbook (see [18] and [24]).
Now we can define a map

p : Q(S) → T (S) (2.3)
by associating to each product foliation structure ([F1 ], [F2 ]) ∈ Q(S) the equivalence
class of conformal structures associated as in Example 2.7 to a pair (F1 , F2 ) represent-
ing ([F1 ], [F2 ]). By Lemma 2.17, the pair (F1 , F2 ) is uniquely defined up to isotopy.
Therefore, the corresponding element of Teichmüller space is well defined. The fo-
liation F1 (respectively F2 ) is called the horizontal (respectively vertical) foliation
of q. We shall say indifferently that the conformal structure is represented by the pair
(F1 , F2 ) or by the pair ([F1 ], [F2 ]).
Jenkins and Strebel studied conformal structures defined by pairs of measured
foliations, where the vertical foliation has a special property that we state in the
following definition:
Definition 2.19 (Jenkins–Strebel structure). A Jenkins–Strebel structure is a product

foliation structure ([F1 ], [F2 ]) such that [F2 ] is the equivalence class of a measured
foliation whose leaves are all compact.2
Notice that a measured foliation has all of its leaves compact if and only if the
subset of the surface consisting of the union of the leaves that start at singular points is
a compact subset of the surface. This subset is then a graph, called the critical graph
of the foliation. The complement of this graph is a finite union of cylinders foliated
by parallel leaves. The core curves of these cylinders are essential and pairwise non-
homotopic closed curves. The property for a measured foliation to have all its leaves
compact, if it holds for a given measured foliation, holds for any equivalent foliation. A
particular class of Jenkins–Strebel structures is the class of product foliation structures
whose vertical measured foliations are foliations with one cylinder, that we already
considered above (Definition 2.2).
Jenkins studied in [28] (what we now call) Jenkins–Strebel structures as solutions
of certain extremal problems concerning maps between Riemann surfaces. It follows
2 Note that the fact that in this definition we talk about the vertical (rather than the horizontal) foliation is just
a matter of convention. In comparing the statements here with statements in some of the papers that we refer to,
one has to be careful about the conventions used in those papers.
from his work and from the work of Strebel in [58] that for any Riemann surface S
and for any measured foliation F on S whose leaves are all compact, there exists a
unique Jenkins–Strebel structure whose vertical foliation is equivalent to F and whose
underlying conformal structure is the given one on S. The following more general
result was obtained later on by Hubbard and Masur:
Theorem 2.20 (Hubbard & Masur [26]). For any Riemann surface S and for any
measured foliation F on S, there exists a unique product foliation structure whose
underlying conformal structure is the one of S and whose vertical foliation is equivalent
to F . Moreover, the map ψ : MF (S)×T (S) → Q(S) that associates to any Riemann
surface S and to any measured foliation class [F ] this uniquely defined element of
Q(S) is a homeomorphism.
This implies in particular that the map p : Q(S) → T (S) defined in (2.3) is
surjective.
Theorem 2.20 was proved by Hubbard and Masur in the case where S is a closed
surface, and another proof was given by Kerckhoff [29]. An adaptation for the case
of non-closed surfaces is contained in [22].
Theorem 2.20 gives the following parametrization of Teichmüller space:
Given a measured foliation F on S, let MF (F ) be the space of measured foliation
classes that are representable by measured foliations transverse to a measured foliation
equivalent to F . For each [F ] in MF , we have a map
ψ[F ] : T → MF (F ) (2.4)
obtained by restricting the map ψ of Theorem 2.20 to the subset {[F ]} × T of
MF × T . By Theorem 2.20, the map ψ[F ] is a homeomorphism. We regard this
map as a parametrization of Teichmüller space by the subset MF (F ) of measured
foliation space. The flowlines of the natural action of R∗+ on MF correspond by this
parametrization to images of geodesic lines in Teichmüller space, for the Teichmüller
metric that we recall below.
Using Theorem 2.20, it is possible to compute explicitly the extremal length of
some particular measured foliation classes, for Kerckhoff’s extension of the extremal
length function ext : MF → R+ (Theorem 2.15).
Theorem 2.21 (Kerckhoff [29], p. 34). Let S be a closed Riemann surface defined by
two transverse measured foliations F1 and F2 as in Example 2.7. Then, ext([F1 ]) and
ext([F2 ]) are equal to the total area of the singular flat metric associated to the pair
(F1 , F2 ).
Remember that a homotopy class of essential simple closed curves considered as

an element of MF is represented by a measured foliation with one cylinder whose
height is equal to 1 (see (2.1) above). Therefore, if the foliation F1 in the statement
of Theorem 2.21 is a foliation with one cylinder whose height is h, then we have
ext([F1 ]) = h2 ExtS (γ ), which gives by Theorem 2.14 ext([F1 ]) = h2 h = h, where
is the length of a closed leaf of the foliation of the cylinder, the length being computed
in the singular flat metric defined by the two transverse foliations F1 and F2 . This
value h is the total area of that singular flat metric. Thus, the proof of Theorem 2.21
for ext([F1 ]) follows from Theorem 2.14 which contains that result in the particular
case where F1 is a foliation with one cylinder, and from the continuity of the map ψ
of Theorem 2.20. By symmetry, the same result holds for ext([F2 ]).
Note that in view of Theorem 2.21 and of the quadratic behaviour in Theorem 2.15,
extremal length should be considered as an area rather than as a length.
Remark 2.22 (Area and intersection number). The geometric intersection function
i : S × S → R+ is defined by
i(γ1 , γ2 ) = min Card{C1 ∩ C2 },
where the minimum is taken over all simple closed curves C1 and C2 in the homotopy
classes γ1 and γ2 respectively. Thurston showed that this function i extends as a
continuous function i : MF ×MF → R+ , called the geometric intersection function
for measured foliations, which extends the map i(F, γ ) defined in (2.2) above. If F1
and F2 are two transverse measured foliations, then the intersection i([F1 ], [F2 ]) is
equal to the area of the singular flat metric that this pair defines.
2.4 The Teichmüller metric

Definition 2.23 (The Teichmüller metric). Let G and H be two conformal structures
on S. The Teichmüller distance between G and H is given by
1
dT (G, H ) = inf log K(f )
2 f
where the infimum is taken over all quasiconformal homeomorphisms f : G → H
that are isotopic to the identity. Since the value dT (G, H ) remains unchanged if we
replace G or H by an isotopic conformal structure, the map dT induces a map on
Tg,n × Tg,n . This map is a metric, and it is called the Teichmüller metric. We shall
denote it by dT .
The symmetry of the map dT follows from the fact that the inverse of a quasiconfor-
mal homeomorphism is a quasiconformal homeomorphism with the same quasicon-
formal dilatation. The triangle inequality follows from the fact that the composition of
a K1 -quasiconformal homeomorphism with a K2 -quasiconformal homeomorphism is
a K1 K2 -quasiconformal homeomorphism.
From the way we introduced the quasiconformal dilatation of a homeomorphism,
Definition 2.23 gives the Teichmüller distance between two Riemann surfaces as a
comparison between moduli of quadrilaterals in these surfaces. The following theorem
gives a characterization of the Teichmüller distance as a comparison between extremal
lengths.
Theorem 2.24 (Kerckhoff [29], p. 36). Let S be a closed surface and let G and H be
two conformal structures on S. The Teichmüller distance between G and H is equal
to
1 ExtH (γ )
dT (G, H ) = log sup .
2 γ ∈S ExtG (γ )
Remarks 2.25. 1) By Kerckhoff’s extension of the extremal length function to MF

(Theorem 2.15) and the fact that the set {xγ | x ∈ R∗+ , γ ∈ S} is dense in MF
(Theorem 2.4), the Teichmüller distance is also given by
1 extH ([F ])
dT (G, H ) = log sup , (2.5)
2 [F ]∈MF extG ([F ])
and by
1 extH ([F ])
dT (G, H ) = log sup . (2.6)
2 [F ]∈PMF extG ([F ])
Note that to simplify notation, we are using in 2.6 the same notation, [F ], to denote
the equivalence class of F in MF and that in PMF .
In order to see that the last expression is meaningful (even though the value
H ([F ])
extH ([F ]) is not defined for [F ] in PMF ), notice that the ratio extextG ([F ]) in (2.6)
is defined by choosing a representative in MF of the element [F ] in PMF and that
the value of this ratio is independent of the choice that we make.
Formula (2.6) is more useful than Formula (2.5) because the space PMF is com-
pact, and therefore the supremum is attained.
H ([F ]) extG ([F ])
2) In particular, we have sup[F ]∈MF extextG ([F ]) = sup[F ]∈MF extH ([F ]) , and we
note right away that the corresponding quotients that define Thurston’s asymmetric
metric (see Section 3 below) are not equal.
3) Teichmüller’s metric is a Finsler metric.
2.5 Teichmüller maps and Teichmüller geodesics

Definition 2.26 (Teichmüller map). A Teichmüller map is the identity map
f : (S, G) → (S, H ), where G and H are two conformal structures on S that are
defined as follows: there exist two transverse measured foliations F1 and F2 and a
positive real λ, such that G is the conformal structure associated to the pair (F1 , F2 )
as in Example 2.7 and H is the structure associated to the pair ((1/λ)F1 , λF2 ).
Remark 2.27. In terms of the distinguished local coordinate z = x + iy associated

to the conformal structure determined by the pair (F1 , F2 ), the homeomorphism f of
Theorem 2.26 is defined by
(x, y)
→ ((1/λ)x, λy). (2.7)
The local quasiconformal dilatation Kz (f ) of f is constant on S, and it is equal to

max(λ2 , 1/λ2 ).
If we start with a topological surface S equipped with a pair (F1 , F2 ) of transverse

measured foliations, and a positive real number λ, then we can define an associated
Teichmüller map. This map is the identity map on S equipped with the conformal
structure (F1 , F2 ) on the domain and ((1/λ)F1 , λF2 ) on the target. In other words,
we can consider a Teichmüller map as defining a new conformal atlas on the sur-
face S, obtained by composing each local chart of the conformal structure defined by
(F1 , F2 ) with the real-affine map defined, using the distinguished local coordinates,
by Formula (2.7) (see Figure 9).
Figure 9. The effect of a Teichmüller map in local coordinates, with λ > 1 (λ is the parameter
in Definition 2.26).
Theorem 2.28 (Teichmüller [60], [61]). For any two conformal structures G and
H on S, we can find two transverse measured foliations F1 and F2 and a positive
real λ such that (F1 , F2 ) represents the structure G and ((1/λ)F1 , λF2 ) represents
the structure H . Furthermore, the Teichmüller map associated to these conformal
structures G and H is the unique homeomorphism that has the least quasiconformal
dilatation among the homeomorphisms between G and H that are isotopic to the
identity.
Remark 2.29. By Theorem 2.28, the least quasiconformal dilatation depends only on
the equivalence classes g and h of G and H respectively.
Definition 2.30 (Teichmüller line and Teichmüller ray). A Teichmüller line (respec-
tively Teichmüller ray) is a map : R → T (S) (respectively : [0, ∞[→ T (S))
defined by t
→ (t) = (e−t F1 , et F2 ) where F1 and F2 are transverse measured fo-
liations on S. (Here, as usual, we are identifying each pair of transverse measured
foliations with the equivalence class of conformal structures that it determines, as in
Example 2.7.) We say that the ray starts at the (equivalence class of the) conformal
structure determined by the pair (F1 , F2 ). By abuse of language, we shall say that F1
(respectively the measured foliation class, the projective class of F1 ) is the horizon-
tal measured foliation (respectively the measured foliation class, the projective class)
of the geodesic line (or ray), and that F2 is its vertical measured foliation (respec-
tively measured foliation class, projective class). (In fact, the pair, and therefore, the
foliations are defined up to a constant factor.)
Recall that if X is a metric space, then a geodesic (respectively a geodesic line,

geodesic ray) in X is a distance-preserving map from a compact interval of R
(respectively from R, from an interval of the form [a, ∞) with a ∈ R) into X. If
γ : [a, b] → X (respectively γ : [a, ∞) → X) is a geodesic (respectively a geodesic
ray) in X, then we say that γ starts at the point γ (a) ∈ X, and that γ connects the
points γ (a) and γ (b) in X. A local geodesic in X is a map from an interval I of
R into X such that each point in the interior of I has a neighborhood such that the
restriction of the map to that neighborhood is a geodesic.
We also recall that a metric space X is said to be geodesic if any two points in X
can be connected by a geodesic, and that a metric space is said to be proper if its closed
balls are compact.
The following is a basic result on the metric structure of Teichmüller space (cf.
Kravetz [33]).
Theorem 2.31 (Teichmüller geodesics). Teichmüller space Tg,n , equipped with the
Teichmüller metric, is complete, proper and geodesic. Furthermore, each pair of
distinct points in Tg,n is contained in the image of a geodesic line which is unique up
to reparametrization. Teichmüller lines are geodesic lines for the Teichmüller metric,
that is, for each Teichmüller line : R → Tg,n , we have, for each t1 and t2 in R,
dT ( (t1 ), (t2 )) = |t2 − t1 |. Furthermore, all the bi-infinite geodesic lines for that
metric are of this form.
The following theorem is also well known. We provide a proof because it follows
directly from previously mentioned results.
Theorem 2.32. Let S be a closed surface, let t ∈ R, let F1 and F2 be two transverse
measured foliations on S and let G and H be the conformal structures associated
respectively to the pair (F1 , F2 ) and (e−t F1 , et F2 ). Then the Teichmüller distance
between G and H is equal to |t|. Furthermore, for t > 0 (respectively t < 0), the
measured foliation F1 (respectively F2 ) realizes the supremum in Formula (2.5) for
the Teichmüller distance between the two conformal structures defined by G and by H .
Proof. By Remark 2.27, the quasiconformal dilatation of the Teichmüller map be-
tween the conformal structures defined by (F1 , F2 ) and (e−t F1 , et F2 ) in the domain
and target respectively is equal to max(e−2t , e2t ) = e2|t| . This gives
1
dT (G, H ) = log e2|t| = |t|.
2
Suppose now that t > 0. By Theorem 2.21 and Remark 2.22, we have
extG (F1 ) = i(F1 , F2 )
and
extH (e−t F1 ) = i(e−t F1 , et F2 ) = i(F1 , F2 ).
Therefore, we have, by Theorem 2.15
extH (F1 ) e2t extH (e−t F1 ) e2t i(F1 , F2 )
= = = e2t .
extG (F1 ) extG (F1 ) i(F1 , F2 )
Thus,
1 extH (F1 )
dT (G, H ) = log = t.
2 extG (F1 )
For t < 0, we can obtain the required result by the same reasoning applied to the
Teichmüller map from the structure to the structure (e−t F1 , et F2 ) to the structure
(F1 , F2 ). 2
2.6 On the asymptotic behaviour of Teichmüller rays

Definition 2.33 (Jenkins–Strebel geodesic). A Jenkins–Strebel geodesic line (respec-
tively a Jenkins–Strebel geodesic ray) is a Teichmüller geodesic line (respectively
geodesic ray) defined by t
→ (e−t F1 , et F2 ) where the vertical measured foliation F2
has all of its leaves compact.
In other words, a Jenkins–Strebel geodesic ray is a geodesic ray that starts at a

Jenkins–Strebel structure (see Definition 2.19).
Recall that if a measured foliation has all its leaves compact, then it consists
of a finite union of maximal foliated cylinders with disjoint interiors, each cylinder
foliated by homotopic leaves. The homotopy classes of leaves associated to different
maximal cylinders are distinct. Thus, to each Jenkins–Strebel geodesic line or ray we
can associate a system of homotopy classes of disjoint and pairwise nonhomotopic
essential simple closed curves on the surface, which we shall henceforth call the
vertical system of homotopy classes. In other words, these are the homotopy classes
of the nonsingular closed leaves of the associated vertical measured foliation. We
shall use the following definition:
Definition 2.34 (Similar Jenkins–Strebel rays, cf. Masur [40], p. 211). Two Jenkins–
Strebel geodesic rays are said to be similar if the associated vertical system of homo-
topy classes of closed curves are equal.
Theorem 2.35 (Masur [40]). Let S be a closed surface and let r1 : [0, ∞) → T and
r2 : [0, ∞) → T be two similar Jenkins–Strebel rays starting at the same point. Then
there exists N > 0 such that dT (r1 (t), r2 (t)) ≤ N for all t ∈ [0, ∞).
In his paper [29], Kerckhoff studied the relative asymptotic behaviour of pairs of
Jenkins–Strebel rays that do not necessarily start at the same point. To state a result,
we need the following
Definition 2.36 (Modularly equivalent Jenkins–Strebel rays, cf. Kerckhoff [29], p. 29).
Suppse that r1 : [0, ∞) → T and r2 : [0, ∞) → T are two similar Jenkins–Strebel
geodesic rays and let γ1 , . . . , γn be their associated vertical system of homotopy classes
of simple closed curves. For every t ∈ [0, ∞), we consider the sequences of moduli
m1 , . . . , mn (respectively m1 , . . . , mn ) of the homotopy classes γ1 , . . . , γn , defined
with respect to the Riemann surface r1 (t) (respectively r2 (t)). We say that the two
rays r1 and r2 are modularly equivalent if these two sequences of moduli are the same
up to a multiplicative constant. (Note that if this condition holds for some t > 0, then
it holds for any t > 0.)
Unlike the property of similarity between pairs of Jenkins–Strebel rays (Defi-

nition 2.34), modular equivalence is a property that involves both the vertical and
horizontal foliations of the rays.
Theorem 2.37 (Kerckhoff, [29], p. 29). Let S be a closed surface and let r1 : [0, ∞) →
T and r2 : [0, ∞) → T be two similar Jenkins–Strebel rays (that do not necessarily
start at the same point) whose vertical system of homotopy classes of simple closed
curves have 3g − 3 elements (which is the largest possible number of elements such
a system can have). Then the following are equivalent:
(1) r1 and r2 are modularly equivalent;

(2) limt→∞ inf M=Im(r2 (t)) d M, Im(r1 ) = 0. (Here, Im denotes the image set of a
map.)
2.7 The Teichmüller boundary and convergent rays

The following definition is contained in Kerckhoff’s paper [29].
Definition 2.38 (The Teichmüller boundary). Let M be a point of T . The Teichmüller

boundary of T relative to the basepoint M, denoted by ∂M T , is the set of geodesic
rays (for the Teichmüller metric) starting at M, equipped with the topology of uniform
convergence on compact sets.
Kerckhoff studied a compactification of Teichmüller space obtained by taking a

basepoint M and putting an endpoint at each Teichmüller ray starting at that point.
The result is a topological space TM = T ∪ ∂M T which is homeomorphic to a closed
(6g − 6 + 2n)-dimensional ball, equipped with a natural center (the point M) and a ray
structure (i.e. an action of R∗+ ), in which the closure of the image of each ray starting
at M is the union of that image with a single point in ∂M T (see [29], p. 31). Kerckhoff
calls such a compactification of T a Teichmüller compactification (the name referring

to Teichmüller’s metric and not only to Teichmüller space), and one of the main
objects of the paper [29] is the study of the dependence of such a compactification on
the basepoint. For instance, a natural question which Kerckhoff addresses is whether
a Teichmüller ray, starting at some point other than the basepoint, has also a unique
limit point in ∂M T . He makes the following definition:
Definition 2.39 (Convergent rays, cf. Kerckhoff [29], p. 28). Let M be a basepoint
in Teichmüller space T and consider the corresponding Teichmüller boundary ∂M T .
Let r : [0, ∞) → T be a geodesic ray starting at some other point M in T , and let
Im(r ) be its closure in TM . Then r is said to be convergent in TM if Im(r ) \ Im(r )
consists of a single point. In that case, Im(r )\Im(r ) is the endpoint of some geodesic
ray r starting at M, and the ray r is said to be convergent to the ray r.
For any element γ in S and for any M in T , we know, by Theorem 2.20, that there
exists a unique Jenkins–Strebel ray rM,γ that starts at M and whose vertical foliation
has one cylinder, with core curve in the homotopy class γ .
Theorem 2.40 (Kerckhoff [29], p. 29). Let S be a closed surface. Then, for any
homotopy class γ of essential simple closed curves and for any M and M in T , the
Jenkins–Strebel ray rM,γ is convergent to the Jenkins–Strebel ray rM ,γ .
Theorem 2.41 (Kerckhoff, [29], p. 29). Let S be a closed surface and let r1 : [0, ∞) →
T and r2 : [0, ∞) → T be two similar Jenkins–Strebel rays with vertical system of
homotopy classes of curves having the largest possible number of elements (that is,
3g − 3 elements). Then r1 and r2 are convergent if and only if they are modularly
equivalent.
2.8 Convergence to Thurston’s boundary

A natural question to consider is whether a Teichmüller ray converges to a point on
Thurston’s boundary PMF of Teichmüller space. We have postponed the definition
of Thurston’s boundary to Section 3 below (§3.13) since its natural setting is hyper-
bolic geometry, but we nonetheless state here a result of Masur about convergence of
Teichmüller rays to points on that boundary:
Theorem 2.42 (Limits of Jenkins–Strebel rays, Masur [43]). Let S be a closed surface,
let r : [0, ∞) → T be a Jenkins–Strebel ray and let α1 , . . . , αn be the associated
vertical system of homotopy classes. Let F = F (α1 , . . . , αn ) be a measured foliation
whose nonsingular leaves are all compact, such that the complement of the critical
graph of F consists of n cylinders whose core curves are in the classes α1 , . . . , αn
and such that the height of each of these cylinders is equal to one. Then the ray r
converges to the projective class [F ], considered as a point in Thurston’s boundary of
Teichmüller space.
It follows from Theorem 2.42 that there are pairs of points on Thurston’s boundary
that are limits of a large family of Teichmüller geodesics. Indeed, let F1 and F2 be
two transverse measured foliations whose nonsingular leaves are all compact and such
that for each of these foliations, the height of each foliated cylinder in the complement
of the critical graph is equal to 1. One can see that such pairs exist from the fact that
for each element [F ] in MF , the subset MF (F ) of equivalence classes representable
by measured foliations transverse to a representative of [F ] is open, therefore, by the
minimality of the action of the mapping class group on PMF ([20], p. 117), this
open set contains a conjugate of the equivalence class of any given measured folia-
tion. By the same argument, for any 1 ≤ m ≤ 3g − 3 and 1 ≤ n ≤ 3g − 3, we can
also choose F1 (respectively F2 ) to have its nonsingular closed leaves in m (respec-
tively n) distinct homotopy classes. Let c1 , . . . , cm (respectively c1 , . . . , cn ) be these
homotopy classes. Let F1 (respectively F2 ) be now an arbitrary measured foliation
having all of its nonsingular leaves compact and with associated homotopy classes
c1 , . . . , cm (respectively c1 , . . . , cn ). By Theorem 2.42, the Teichmüller geodesic line
t
→ (e−t F1 , et F2 ) has [F1 ] and [F2 ] as its limit points. Note that if the projective
equivalence classes of two such pairs (F1 , F2 ) and (F1 , F2 ) are distinct, then, by Teich-
müller’s uniqueness theorem, the images of the two geodesics t
→ (e−t F1 , et F2 ) and
t
→ (e−t F1 , et F2 ) have at most one common point. Thus, we have a large family of
Teichmüller geodesics whose limit points are [F1 ] and [F2 ].
We recall that a measured foliation F is said to be uniquely ergodic if F , as a
topological foliation, carries a unique transverse measure up to a multiplicative factor.
The property of being uniquely ergodic is invariant by Whitehead-equivalence and
by homothety. Therefore one can talk about uniquely ergodic projective classes of
measured foliations.
We shall also use the following terminology from [20]: a measured foliation F
is said to be arational if it does not contain any closed curve made up of segments
connecting singular points. Equivalently, up to performing Whitehead moves on F ,
every leaf of F is dense in the surface.
Theorem 2.43 (Limits of Teichmüller rays with uniquely ergodic vertical foliations,
Masur [43]). Let S be a closed surface. Then any Teichmüller ray whose associated
vertical foliation is arational and uniquely ergodic converges in the positive direction
to the projective equivalence class of that vertical foliation, considered as an element
of Thurston’s boundary of Teichmüller space.
There is a natural Lebesgue measure on the space MF , which is provided, for

instance, by train track coordinates (see [65] and [54]). By a result which was proved
independently by Masur and by Veech (see [42] and [71]), the subspace of MF
consisting of equivalence classes of uniquely ergodic measured foliations is of full
measure with respect to that measure. This measure on MF defines a measure class
on the set of geodesic rays starting at any given point M of Teichmüller space. This can
be seen by referring to the result of Hubbard and Masur stated in Theorem 2.20 which
says that for any element M ∈ T and for any measured foliation class [F1 ] ∈ MF ,
there exists a unique measured foliation class [F2 ] ∈ MF such that the conformal
structure defined by the pair (F1 , F2 ) is M. Thus, the pair (F1 , F2 ) can be regarded
as a direction in Teichmüller space, at the point M, that is, the direction of the ray
defined by t
→ (e−t F1 , et F2 ) starting M. In this sense, Theorem 2.43 says that for
any point M in Teichmüller space, almost any geodesic starting at M converges to a
point in Thurston’s boundary, which is the equivalence class of the vertical foliation
of that ray.
In a recent preprint, A. Lenzhen gave an explicit example of a Teichmüller ray that
does not have a limit in PMF . We state this as
Theorem 2.44 (Teichmüller rays with no limit in PMF , Lenzhen [35]). On the closed
surface of genus 2, there exists a Teichmüller ray that does not converge to any point
in Thurston’s boundary PMF .
The construction of that ray is based on an arithmetic property of the measured

foliations associated to the geodesic ray considered.
The asymptotic behaviour of an arbitrary Teichmüller geodesic ray in terms of the
dynamical/topological properties of its vertical foliation is still an open question.
2.9 On the visual boundary of the Teichmüller metric

We start by recalling a general definition.
Definition 2.45 (Asymptotic geodesic rays). Let (X, d) be a metric space. Two
geodesic rays r1 : [a1 , ∞) → X and r2 :[a2 , ∞) → X are said to be asymptotic if
the function defined on [0, ∞) by t
→ d r1 (t + a1 ), r2 (t + a2 ) is bounded.
It is clear that the relation of being asymptotic is an equivalence relation on the set
of geodesic rays in a metric space. We shall write r1 ∼ r2 to say that the two geodesic
rays r1 and r2 are asymptotic.
Definition 2.46 (The visual boundary). Let X be a proper geodesic metric space and
let x be a point in X. The visual boundary of X at x, denoted by ∂vis,x X, is the set of
equivalence classes (for the relation ∼) of geodesic rays starting at x, equipped with
the quotient of the topology of uniform convergence on compact sets.
Note that the visual boundary at a point in Teichmüller space equipped with the
Teichmüller metric is a quotient of the Teichmüller boundary relative to that point.
In the paper [46], based on Masur’s result stated as Theorem 2.35 above, the
following is proved
Theorem 2.47 (McCarthy–Papadopoulos [46]). Let S be a closed surface. For any

point x in the Teichmüller space T (S) equipped with the Teichmüller metric, the visual
boundary ∂vis,x T (S) is not Hausdorff.
This result was used to give a new proof of the fact due to Masur and Wolf that
Teichmüller space equipped with the Teichmüller metric is not Gromov hyperbolic
cf. [44].
3 Thurston’s asymmetric metric

3.1 Hyperbolic structures
Definition 3.1 (Hyperbolic structure). A hyperbolic structure on S is a maximal atlas
{(Ui , φi )}i∈ where for each i ∈ , Ui is an open subset of S and φi is
a homeomor-
phism from Ui onto an open subset of the hyperbolic plane H2 , satisfying i∈ Ui = S
and such that any map of the form φi φj−1 is, on each connected component of
φj (Ui ∩ Uj ), the restriction of an orientation-preserving isometry of H2 .
A surface equipped with a hyperbolic structure is called a hyperbolic surface. It

carries a length metric defined as follows. On each chart domain Ui , we pull-back by
the map φi the metric on φi (Ui ) induced from its inclusion in H2 . These metrics on the
various sets Ui give a consistent way of measuring lengths of paths in S, and the metric
we consider on S is the associated length metric. It is a Riemannian metric of constant
curvature −1. The metric universal covering of S is a subset of hyperbolic plane H2 .
If we regard the various φi ’s as diffeomorphisms (instead of homeomorphisms),
then we can pull-back the Riemannian metrics on φi (Ui ) to the various sets Ui . This
is another way of defining the Riemannian metric on S. In any case, a hyperbolic
structure is also regarded as a metric on S, called a hyperbolic metric. We shall only
consider hyperbolic metrics that are complete and of finite area. Equivalently, these
are the metrics that have the property that each puncture of S has a neighborhood
which is isometric to a cusp, that is, the quotient of a subset {x + iy | y > a} (a > 0)
of the upper half-plane model of H2 by the map z
→ z + 1. Equivalently, these are
the metrics for which the metric universal cover is H2 (see [70]).
The group Hom(S) of orientation preserving homeomorphisms of S acts on the set
of hyperbolic structures in the same way as it acts on the set of conformal structures
(cf. Definition 2.8), and, again, we say that two hyperbolic structures are equivalent
if they are related by a homeomorphism of S which is isotopic to the identity.
There is a natural one-to-one correspondence between the set of equivalence classes
of hyperbolic structures and the set of equivalence classes of conformal structures on S.
This correspondence is given by the quotient of the map which assigns to each hyper-
bolic structure its underlying conformal structure (Example 2.6 above). The fact that
the hyperbolic structure is complete and of finite area implies that the neighborhood
of each puncture is conformally a punctured disk (with respect to the conformal struc-
ture associated to that hyperbolic metric). Using this natural correspondence, one can
make the following definition, which is an equivalent form of Definition 2.8 above:
Definition 3.2 (Teichmüller space). The Teichmüller space of S, denoted by Tg,n (S) or
T (S), is the space of equivalence classes of complete finite area hyperbolic structures
on S.
By abuse of language, we shall often call an element of Teichmüller space a “hy-

perbolic structure” instead of an equivalence class of hyperbolic structures. We shall
also use interchangeably the terms “hyperbolic structure” and “hyperbolic metric”.
The topology of T (S) can be defined using several metrics. For instance, it is
induced by the “quasi-isometry” metric dqi given by
1
dqi (g, h) = log inf{K} (3.1)
2
where the infimum is taken over the set of real numbers K ≥ 1 such that there exists
a homeomorphism f : S → S satisfying
1
dg (x, y) ≤ dh f (x), f (y) ≤ Kdg (x, y)
K
(see Thurston [70], p. 266).
We can also describe the topology of T (S) by means of the length functional
l : T (S) → RS+ defined by
g
→ lg ( · ) : α
→ lg (α) for all α ∈ S, (3.2)
with lg (α) being the length of the unique g-geodesic in the homotopy class α. This
map l is an embedding, and the topology of T (S) induced by the metric defined in
(3.1) coincides with the one induced on the image of l by the weak topology on the
space RS+ . Thus, a sequence (gn )n≥0 in T (S) converges as n → ∞ to an element
g ∈ T (S) if and only if for every α in S, we have lgn (α) → lg (α) as n → ∞. In
fact, more is true: there exists a finite set {α1 , . . . , αk } of homotopy classes of simple
closed curves in S such that an arbitrary sequence (gn )n≥0 in T (S) converges to an
element g in that space if and only if for every i = 1, . . . , k we have lgn (αi ) → lg (αi )
as n → ∞. (For a proof in the case of closed surfaces, see [20].) Finally, let us recall
that this topology on T (S) is the one induced by Teichmüller’s metric, and also by
Thurston’s asymmetric metric, which will be the main object of study in this section
(see §3.7 below).
3.2 Decompositions by generalized pairs of pants and by ideal triangles

By studying the representation of conformal structures on a surface Sg,n by hyperbolic
structures one can easily get an intuition of the fact that its Teichmüller space is
homeomorphic to an open ball of dimension 6g−6+2n, a fact that is harder to conceive
when one starts with the purely conformal point of view (that is, Definition 2.8). The
statement that the space of deformations of equivalence classes of conformal structures
on a closed orientable surface of genus g ≥ 2 is of complex dimension 3g−3 was made
by G. F. B. Riemann (without giving a formal proof) in his paper Theorie der Abel’schen
Functionen, Crelle’s J. f. M., B. 54 (1857). A homeomorphism Tg,n R6g−6+2n , if
one does not use hyperbolic geometry, can be obtained using Teichmüller’s results. But
when the surface Sg,n is equipped with a hyperbolic structure, it is easy to produce
such a homeomorphism using a decomposition of the surface into objects we call
“generalized hyperbolic pairs of pants” with geodesic boundary. Another way to
realize that homeomorphism, in the case where the set of punctures of S is not empty,
can be achieved by using ideal triangulations. Our aim in this section is to discuss
these two sorts of decompositions.
We recall that a hyperbolic pair of pants is a sphere with three open disks removed,
equipped with a hyperbolic metric for which the three boundary components are
geodesic.3 A degenerate hyperbolic pair of pants is a complete finite area hyperbolic
surface which is either a sphere with three cusps, or a sphere with two cusps and one
open disk removed, or a sphere with one cusp and two open disks removed, and where
the boundary components, whenever they exist, are closed geodesics (see Figure 10).
There is a precise sense in which any degenerate hyperbolic pair of pants is a limit
of a family of hyperbolic pairs of pants. This can be seen from the construction of
hyperbolic pairs of pants using right-angled hexagons, which we now briefly recall.
Figure 10. Degenerate hyperbolic pairs of pants.
The classification of hyperbolic structures on a pair of pants with geodesic boundary

is usually done by decomposing the pair of pants into two isometric right-angled
hexagons. The isometry class of a right-angled hexagon is completely determined by
the lengths of any three alternating edges. For an exposition of this fact, we refer the
reader to Thurston’s book [70] (Exercise 2.4.11, Example 4.6.7 and Figure 4.15).
Consider a right-angled hexagon in hyperbolic space H2 . If we make the length of
one of its edges tend to zero while keeping fixed the other two which form with the
degenerating edge an alternating triple of edges, and keeping all the angles to be right
3 In talking here about closed curves in the surface S, it should be noted that the term geodesic is used in the
sense of Riemannian geometry, that is, in the sense of (the image of) a locally isometric map, as opposed to the
other possible use of that word, denoting an isometric map.
angles, the vertices of the degenerating edge converge to a single point in the boundary
of hyperbolic space, and the right-angled hexagon becomes, at the limit, an object we
call a degenerate right-angled hexagon (see Figure 11 for a picture of such a hexagon).
Figure 11. These four generalized polygons represent respectively a right-angled hexagon and
three degenerate right-angled hexagons in the upper half-plane model of the hyperbolic plane.
Likewise, we can make (successively) the lengths of two or of three alternating edges
tend to zero. We obtain in this manner degenerate right-angled hexagons with one, two
or three edges at infinity, called spikes. Note that a degenerate right-angled hexagon
with three edges at infinity is a hyperbolic ideal triangle (that is, the convex hull, in
the hyperbolic plane H2 , of three distinct points in the boundary of that plane). A
degenerate right-angled hexagon with one, two or three edges at infinity is completely
determined up to isometry by the lengths of the edges that make with the one (or
the ones) at infinity a triple of alternating edges. Since a hyperbolic pair of pants is
obtained by gluing two isometric right-angled hexagons along three alternating edges,
it is easy to see that the isometry type of the pair of pants is completely determined
by the lengths of its three boundary geodesics. Likewise, a hyperbolic structure on
a degenerate pair of pants is completely determined by the lengths of its boundary
geodesic curves, a cusp being considered as a boundary curve of length zero. In this
sense, the hyperbolic structure is rigid at the cusps. For instance, a degenerate right-
angled hexagon with three alternating edges at infinity is a hyperbolic ideal triangle,
and, as is well known, there is a unique hyperbolic ideal triangle up to isometry. In
the same way, any two degenerate hyperbolic pairs of pants that are spheres with
three cusps are isometric. We shall call a generalized hyperbolic pair of pants either
a hyperbolic pair of pants in the usual sense, or a degenerate hyperbolic pair of pants
with one, two or three cusps.
A closed surface of genus g ≥ 2 is decomposable into 2g − 2 pairs of pants, by
using 3g − 3 simple closed curves (see [20]). Using this fact, we can count the number
of curves in a generalized pair of pants decomposition of a surface of genus g with n
cusps by first replacing each cusp (if such a cusp exists) by a boundary closed curve,
obtaining a surface with boundary that is homeomorphic to a closed surface of genus g
with n open disks deleted. Doubling that surface along its boundary curves, we obtain
a closed surface of genus 2g + n − 1, which we can decompose into pairs of pants by
using 6g − 6 + 3n closed geodesics. By symmetry, we conclude that the number of
closed curves needed to decompose the surface Sg,n into generalized pairs of pants is
3g − 3 + n.
Now we can count the parameters for the Teichmüller space of a surface of genus
g with n punctures. Consider a set of closed geodesics decomposing that surface into
generalized hyperbolic pairs of pants. There are 3g − 3 + n such geodesics. There are
two parameters associated to each geodesic, one parameter (in R∗+ ) being the length
of the geodesic itself, and the other parameter (in R) describing the twist along that
curve, when we glue together its two sides. (The twist parameter is defined up to the
choice of an origin corresponding to the zero twist, and up to an orientation of the
boundary component.) This gives two sets of 3g − 3 + n real parameters each. Thus,
we have 6g − 6 + 2n parameters, which is an indication of the fact that Teichmüller
space is homeomorphic to R6g−6+2n . For a geometric and concise proof in the case
where n = 0, we refer the reader to Thurston [70], p. 271. There are also proofs of that
fact in [1], [8], [20], [27]. These length-twist parameters associated to a (generalized)
pair of pants decomposition are called (generalized) Fenchel–Nielsen parameters.
Another way of obtaining a homeomorphism Tg,n R6g−6+2n uses a decomposi-
tion of Sg,n into ideal triangles, and we now describe it in the case where n ≥ 1. Recall
that any hyperbolic ideal triangle has a well-defined center (its center of gravity), and
one distinguished point on each of its edges, which is the orthogonal projection on that
edge of the center of gravity, or, equivalently, the orthogonal projection on that edge of
the ideal vertex that is opposite to that edge. An ideal triangulation λ of S = Sg,n is a
decomposition of the associated closed surface Ŝ into triangular cells whose vertices
are all at the punctures. If S is equipped with a hyperbolic structure, then one can
make the edges of λ to be embedded bi-infinite local geodesics with limit points at the
punctures. (The bi-infinite local geodesics are, by definition, embedded images of the
bi-infinite geodesic lines in H2 realized as the metric universal covering of the surface.)
In this way, each face of λ is isometric to the interior of a hyperbolic ideal triangle, and
the hyperbolic structure on S is determined by the gluing maps between the edges of
these ideal triangles. In this gluing, there is one parameter associated to each edge of λ,
which we can take to be the algebraic distance, called the shift parameter, between
the two distinguished points on that edge, each distinguished point being associated
to one of the triangles adjacent to that edge. (Recall that using the orientation on the
surface S, one can define a notion of a left and of a right shift along the edges of μ.
The sign of the distance between the distinguished points is then defined accordingly.)
Thus, the Teichmüller space Tg,n of Sg,n is seen from that point of view as the set of
shift parameters on the edges of an ideal triangulation λ of that surface, and a rough
computation of the dimension of Tg,n can be done in the following way. Let E denote
the number of edges of λ and let T be the number of its faces (triangles). Since each
triangle has 3 boundary edges and since each edge is adjacent to 2 triangles, we have
E = 3T /2. The Euler characteristic of Sg,n is χ = 2 − 2g − n. We construct an
auxiliary foliation F on the surface Ŝ obtained from Sg,n by filling in the punctures,
such that F is transverse to the edges of the triangulation λ, with a 3-prong singularity
in the interior of each triangle, and with a center-type singularity at each puncture.
There is an index formula relating the sum of the indices at the singular points of such
a foliation to the Euler characteristic χ of the surface (see for instance [20], p. 75).
The formula says that 2χ is equal to the sum of the indices of the singular points
of the foliation. The index of a 3-prong singularity is −1 and the index of a center-
type singularity is 0. Therefore, −2χ is equal to the number of 3-prong singularities,
which is the number of triangles. In other words, we have T = −2χ . Thus, we obtain
E = 23 (−2)(2 − 2g − n) = 6g − 6 + 3n. Now for the dimension of Teichmüller space,
there is one shift parameter associated to each edge and one equation associated to
each puncture; this is the equation which ensures that the hyperbolic structure near
that puncture is complete, which is equivalent to the fact that around each puncture
of Sg,n , there is a foliation by closed leaves made up of pieces of horocycles that
are perpendicular to the edges that abut on that puncture, see Definition 3.6 below.
Therefore, the number of parameters is E − n = 6g − 6 + 3n − n = 6g − 6 + 2n, as
expected.
3.3 Geodesic laminations

Definition 3.3 (Geodesic lamination). Let S be a hyperbolic surface and let p : H2
S̃ → S be its metric universal covering. A geodesic lamination on S is a closed subset
of S which is the union of disjoint images of bi-infinite geodesics of H2 by the covering
map p such that if ⊂ S̃ is such a bi-infinite geodesic, then, either
(1) the restriction of p to is injective, or
(2) p( ) is a simple closed geodesic on S.
The decomposition of a geodesic lamination as a union of such images of bi-

infinite geodesics of H2 is unique, and each such image is called a leaf of λ. There are
several good references on geodesic laminations. We refer the reader to Thurston’s
original notes [65], to the book by Penner and Harer [54] or to the book by Casson
and Bleiler [15].
By abuse of language, we shall sometimes call the leaves of λ geodesics (rather
than local geodesics). We shall talk about bi-infinite leaves of λ to denote leaves of
kind (1) above, and about closed leaves to denote leaves of kind (2).
It is well known that there is a natural one-to-one correspondence between geodesic

laminations associated to any two hyperbolic structures on a given surface (see [36]).
This makes it possible to talk about a geodesic lamination on a surface without refer-
ence to any particular hyperbolic structure. In fact, we shall sometimes call a geodesic
lamination on a surface S an object which is a geodesic lamination for some hyperbolic
structure on S. Equivalently, a geodesic lamination on S will be an object which is
isotopic to a geodesic lamination for any hyperbolic structure on S.
A geodesic lamination is said to be complete if there is no geodesic lamination
that strictly contains it. Equivalently, a geodesic lamination λ is complete if each
connected component of S \ λ, equipped with its intrinsic metric, is isometric to the
interior of a hyperbolic ideal triangle.
An ideal triangulation of a surface S = Sg,n with n ≥ 1 is an example of a complete
geodesic lamination on S.
If λ is a geodesic lamination and if μ is any complete geodesic lamination con-
taining it, then we shall call μ a completion of λ.
Any non-complete geodesic lamination can be completed in several ways, and
it is interesting to understand these different ways because they will correspond to
various ways of constructing geodesics for Thurston’s asymmetric metric which we
shall study below. It is possible to complete any geodesic lamination λ by inserting a
finite number of bi-infinite geodesics in its complement. The surface S \ λ has a finite
number of connected components, and the metric completion of each such connected
component is a hyperbolic surface with geodesic boundary (see [15]). In particular,
if λ contains closed leaves, then, in order to obtain a completion of λ, we can add
geodesics that spiral around the closed leaves. For instance, in any hyperbolic pair
of pants (with geodesic boundary), we can draw three bi-infinite geodesics that spiral
around the boundary components so that the complement of these bi-infinite geodesics
in the pair of pants is a union of the interiors of two hyperbolic ideal triangles (see
Figure 12). In fact, there are 32 distinct ways of completing a hyperbolic pair of pants.
Figure 12. One way of obtaining complete geodesic laminations on a closed surface is to divide
that surface into hyperbolic pairs of pants, and then decompose each pair of pants by bi-infinite
leaves that spiral around the boundary components, as in the example drawn here. The stump
of the complete lamination that we obtain is the union of the closed geodesics defining the pair
of pants decomposition.
If the lamination λ that we started with does not have closed leaves, then we can
still complete it by adding a finite number of bi-infinite geodesics that “spiral along
bi-infinite leaves of λ”. Spiraling (along closed geodesics or along bi-infinite leaves)
can be defined as an operation in the universal covering of S, that consists in adding in
an equivariant manner to the lift of the lamination λ a certain number of geodesic lines
that converge in each direction to the limit point of the lift of some (finite or infinite)
leaf of λ. To be more precise, let us first suppose that all the leaves of λ are closed. Let
α and β be two closed leaves in λ and let us consider a segment c having one endpoint
on α and one endpoint on β, and whose interior is disjoint from λ. Starting from
the segment c, we define a continuous family of segments (ct )t≥0 in S, with c0 = c,
and such that for any t ≥ 0, the endpoints of ct are on α and on β, and spin around
these curves as t increases to ∞ with constant speed. For each t ≥ 0, we can take
the segment ct to be a local geodesic, its interior being disjoint from the lamination λ.
Let p : S̃ → S be again the metric universal covering of S, with S̃ identified with the
upper half-plane H2 . Consider a continuous family of segments (c̃t )t≥0 in H2 , which
lifts the family (ct )t≥0 in S. For each t ≥ 0, the endpoints of the segment c̃t are on
lifts α̃ an! d β̃ of α and β. The family of geodesic segments (c̃t )t≥0 converges in an
obvious sense to the image of a bi-infinite geodesic line c̃ : R → H2 whose endpoints
are endpoints of α̃ and β̃. The distance from the point c̃(t) to the geodesic lines α̃ and
β̃ tends to 0 as t → ∞ or t → −∞ (and the convergence is of the order of e−t ). The
covering map p : S̃ → S restricted to c̃ is injective, and the image of c̃ by this map is
a bi-infinite geodesic line in S which spirals at one end around α and at the other end
around β. We take this geodesic as a new leaf in the completion of λ.
To continue completing the geodesic lamination λ by filling-in the connected com-
ponents of S \ λ with bi-infinite geodesics, we must consider the general case (that is,
the case where leaves of λ are not all closed). Each completed connected component
of the surface S cut along λ is a surface with boundary, with each boundary component
being either a closed geodesic or a union of bi-infinite geodesics, as in Figure 13. We
can decompose each connected component of S \ λ into ideal triangles by succes-
Figure 13. A possible component of the surface S cut along the lamination μ.
sively inserting a finite number of bi-infinite geodesics which at each end either spiral
along a closed leaf or converge to one of the ends of a bi-infinite leaf or converge to
a cusp. The number of leaves that we can add is bounded in terms of the topology
of the surface since we eventually decompose the surface into ideal triangles, each of
them having a fixed area (equal to π ), and the area of the hyperbolic surface S is a
topological invariant.
Definition 3.4 (Measured geodesic lamination). A measured geodesic lamination λ

on a hyperbolic surface is a geodesic lamination which is equipped with a nonnegative
Radon measure on each transverse arc. Furthermore, the measure is invariant under
homotopies of the arc respecting the lamination λ and the support of the measure is
equal to the intersection of the arc with the support of λ.
To each measured geodesic lamination λ is associated an element of the function

space RS+ , by a rule analogous to the one that associates to each measured foliation
an element of RS+ (see (2.2) above). On a given hyperbolic surface, two distinct mea-
sured geodesic laminations define distinct elements of RS+ , unlike pairs of measured
foliations which are more flexible objects. (There is no need to talk about isotopy
or Whitehead-equivalence relation between geodesic laminations on the same hyper-
bolic surface.) The two subsets of RS+ consisting respectively of images of equivalence
classes of measured foliations and images of measured geodesic laminations coincide,
and this fact can be used to define a one-to-one correspondence between spaces of
equivalence classes of measured foliations on a surface S and spaces of measured
geodesic laminations on the surface S equipped with a hyperbolic structure. Note that
there is an alternative and more geometric procedure to associate to each measured
foliation class a measured geodesic lamination, which gives the same correspondence
between the two sets (see for instance [36]).
We already mentioned that there is a natural one-to-one correspondence between
sets of geodesic laminations on the surface S equipped with two distinct hyperbolic
structures. This correspondence leads to a metric-independent definition of a geodesic
lamination. In fact, Thurston shows in his Notes [65] that two geodesic laminations
which correspond by that natural correspondence are isotopic as topological objects,
by a global isotopy of S that fixes the cusps. Therefore, on a topological surface S, a
geodesic lamination can also be thought of as an object defined up to isotopy.
The support of a lamination is the union of its leaves. By abuse of notation, we
shall sometimes denote the support of a measured geodesic lamination λ by the same
letter λ.
We denote by ML(S) the set of measured geodesic laminations on S, and by
ML0 (S) ⊂ ML(S) the subset consisting of measured geodesic laminations whose
support is compact. Note that a compactly supported measured geodesic lamination
cannot be complete if the surface has punctures, since by the definition of the transverse
measure for a geodesic lamination, the support of the measure is equal to the support
of the lamination, and therefore the lamination cannot have leaves which converge
to punctures (the closure of such a leaf cannot be compact, and a complete geodesic
lamination has necessarily such a leaf).
There is a natural action of the group R∗+ of positive reals on the space ML(S),
and the quotient of ML(S) by this action is the set PML(S) of projective classes
of measured geodesic laminations, whose elements are called projective measured
geodesic laminations. The subset of PML(S) consisting of the projective classes of
compactly supported measured geodesic laminations is denoted by PML0 (S). One
can define topologies on the spaces ML(S) and ML0 (S) by using the inclusion maps
of these spaces in the space RS+ and taking the restriction of the weak topology. In
this way, the correspondences between spaces of measured geodesic laminations and
spaces of equivalence classes of measured foliations become homeomorphisms.
Definition 3.5 (The stump of a geodesic lamination, [63]). The stump of a geodesic
lamination μ is the support of any maximal (with respect to inclusion) compactly
supported measured geodesic sublamination λ of μ.
The stump of a geodesic lamination is itself a geodesic lamination, and it is empty

if and only if each leaf of μ converge at each end towards a cusp of the surface. For
instance, if μ is complete, then the stump of μ is empty if and only if μ is an ideal
triangulation.
Completing a compactly supported measured geodesic lamination λ by adding
spiraling leaves gives a complete geodesic lamination whose stump is λ.
We shall sometimes consider the stump of a lamination μ as being equipped with
some transverse measure, and we shall also call this measured lamination the stump
of μ, although the transverse measure it carries is not unique.
We shall use the notion of length of a (compactly supported) measured geodesic
lamination and we now recall the definition.
There is a function defined on T (S) × ML0 (S) which to every hyperbolic struc-
ture g and every compactly supported measured geodesic lamination λ on S associates
a quantity denoted by lg (λ) and called the length of λ with respect to g. This function
is a continuous extension of the notion of length of a simple closed geodesic, the set
of simple closed geodesics being considered as a subset of the space of measured
geodesic laminations, each simple closed geodesic being equipped with the Dirac
measure of mass one. The definition of lg (λ) can be made by covering the geodesic
lamination λ by a finite collection of quadrilaterals R1 , . . . , Rn with disjoint interi-
ors, such that the intersection of the leaves of λ with each quadrilateral is a union of
segments that crosses it from one side to the opposite side. We shall call these two
sides the “vertical” sides of Ri . We choose, for each quadrilateral Ri , one vertical
side, and we call it ∂v Ri . The length lg (λ) is then the sum of the integrals, over all
the quadrilaterals Ri , of the lengths of the geodesic segments of λ that cross them,
with respect to the transverse measure dλ of λ induced on the vertical sides ∂v Ri . In
formulae, if we denote by αx , x ∈ λ ∩ ∂v Ri , the collection of geodesic sub-segments
of λ that traverse the quadrilateral Ri , we have
n

lg (λ) = lg (αx )dλ(x).
i=1 ∂v Ri ∩λ
3.4 Horocyclic foliations

We need to consider measured foliations on S which are slightly different from those
we defined in Section 2.1. These foliations naturally appear in the study of hyperbolic
surfaces with cusps.
Definition 3.6 (Measured foliation trivial around the punctures). A measured foliation
F on S is said to be trivial around the punctures if F is a measured foliation in the
sense of Definition 2.1 except for the condition at the punctures, which is replaced by
the following: each puncture has a neighborhood on which the induced foliation is a
cylinder foliated by homotopic closed leaves, and any segment transverse to F and
converging to a puncture has infinite total mass with respect to the transverse measure
of F .
In what follows, we shall sometimes omit the adjective “trivial around the punc-
tures” when the context is clear. We can pass from a measured foliation trivial around
the punctures to a measured foliation in the sense of Definition 2.1 by deleting, for
each puncture of S, the maximal annulus foliated by closed leaves parallel to that
puncture, obtaining thus a partial measured foliation on the surface, and then collaps-
ing the complementary components of the support of that partial foliation in order to
get a foliation whose support is the entire surface S. The resulting measured foliation
is well defined up to equivalence, except that it may be empty. Thus, to each measured
foliation which is trivial around the punctures we can naturally associate an object
which is either a well-defined element in MF or the empty foliation.
For any fixed hyperbolic structure on S, we next define an object which is “dual” to a
complete geodesic lamination and which we call the associated “horocyclic foliation”.
We start with a construction in an ideal triangle. Any hyperbolic ideal triangle is
equipped with a canonical partial measured foliation which is called the horocyclic
foliation of the triangle. This partial foliation is characterized by the following three
properties:
(1) the leaves are pieces of horocycles that are perpendicular to the edges;
(2) the non-foliated region is a triangle bordered by three of these pieces of horocycles
(see Figure 14);
(3) the transverse measure assigned to any arc which is contained in an edge of the
ideal triangle coincides with the Lebesgue measure induced from the hyperbolic
metric.
Note that the non-foliated triangular region intersects each edge of the ideal triangle
at its distinguished point.
Definition 3.7 (Horocyclic measured foliation). Let S be a hyperbolic surface and let μ
be a complete geodesic lamination on S. The completion of each connected component
of S \ μ being an ideal triangle, we can equip it with its horocyclic measured foliation.
horocycles
perpendicular
to the boundary
horocyclic arc non-foliated region

of length one
Figure 14. The horocyclic foliation of an ideal triangle.
The union of the horocyclic measured foliations associated to the various ideal triangles
fit together smoothly, since they are perpendicular to the edges of the ideal triangles
and therefore they form a Lipschitz-line field on the surface. They define a partial
measured foliation of S, which we call the horocyclic measured foliation associated
to μ and S, and which we denote by Fμ (S). Collapsing each nonfoliated triangular
piece onto a tripod, we obtain a genuine (i.e. not partial) measured foliation on the
surface, which is well defined up to isotopy, and which we also call the horocyclic
measured foliation associated to μ.
The fact that the hyperbolic structures that we consider are complete and have
finite area is equivalent to the fact that for any complete geodesic lamination μ, the
associated horocyclic foliation Fμ (S) is trivial around the punctures.
The isotopy class of Fμ (S) depends only on the isotopy class of the hyperbolic
structure and on the complete geodesic lamination μ that we started with. Therefore,
the equivalence class of Fμ (S) is associated to the hyperbolic structure S viewed as an
element of Teichmüller space. As usual, [Fμ (S)] ∈ MF denotes the equivalence class
of the horocyclic foliation Fμ (S) after removal of the closed leaves that are parallel
to punctures (provided Fμ (S) is not the empty foliation).
Definition 3.8 (Horocyclic measured geodesic lamination). Let S be a hyperbolic

surface and let μ be a complete geodesic lamination on S. The measured geodesic
lamination that represents the (partial) measured foliation Fμ (S) (after removal of the
leaves that are parallel to the punctures, and provided the remaining measured foliation
is not the empty foliation) will be called the horocyclic measured geodesic lamination
associated to μ, and it will be denoted by λμ (S).
The horocyclic measured geodesic lamination λμ (S) is the empty lamination if and
only if the horocyclic foliation Fμ (S) is a union of cylinders foliated by leaves that
are all parallel to punctures. It is easy to construct examples of hyperbolic structures
whose associated horocyclic foliation has this property. For instance, we can start, as
in Figure 15, with an ideal quadrilateral in H2 which is the union of two ideal triangles
glued along two edges in such a way that the distinguished points on the common
edges agree. We then glue the remaining four edges by pairs, respecting the labels a
and b in Figure 15, again in such a way that the distinguished points on these edges
coincide after the gluing. The resulting hyperbolic surface is a once-punctured torus,
and the resulting lamination μ on that surface is the union of the images of the edges
of the two ideal triangles that we started with.
a b
c c
b
b a
a
Figure 15. An example of a hyperbolic structure on the punctured torus equipped with a complete
geodesic lamination μ which is an ideal triangulation. The gluing of the various ideal triangles
in the universal covering is performed in such a way that for each edge of the ideal triangulation
μ, the two distinguished points corresponding to the two triangles adjacent to it coincide. In
this case, the associated horocyclic measured geodesic lamination is empty.
3.5 Thurston’s cataclysm coordinates

In this section, we review coordinates for Teichmüller space that were introduced by
Thurston, who called them “cataclysm coordinates” and which are essential in the
study of Thurston’s asymmetric metric. The parameter space is a space of measured
foliations that are transverse to a complete geodesic lamination. We start with the
following:
Definition 3.9 (Measured foliation totally transverse to a lamination). Let μ be a

geodesic lamination on S. A measured foliation F on S is said to be totally transverse
to μ if it satisfies the following:
(1) F is transverse to μ;
(2) F is trivial around the punctures.
In this definition, we have tacitly chosen an auxiliary hyperbolic structure on S in

order to talk about geodesic laminations, but recall that we can make a definition which
is independent of that choice. In other words, we say that μ is a geodesic lamination

if μ is a geodesic lamination for some hyperbolic structure on S.
A measured foliation class is said to be totally transverse to μ if it can be represented
by a measured foliation that is totally transverse to μ.
For any geodesic lamination μ on S, we denote by MF (μ) ⊂ MF the set of
measured foliation classes that are totally transverse to μ. This subset MF (μ) is
open in MF .
If μ is complete, then, for any choice of a hyperbolic metric g on S, the associated
horocyclic measured foliation Fμ (g) is totally transverse to μ. If h is a hyperbolic
metric which is isotopic to g, then the horocyclic foliation Fμ (h) is isotopic to Fμ (g).
This enables us to associate to each element g in T (S) a well-defined element in
MF (μ). We denote by φμ : T (S) → MF (μ) the resulting map.
Theorem 3.10 (Thurston, [67], §9). For any complete geodesic lamination μ on S,
the map φμ : T (S) → MF (μ) is a homeomorphism.
The global coordinates for Teichmüller space that are provided by this map φμ are
called by Thurston cataclysm coordinates.
The proof of Theorem 3.10 is given by Thurston in §4 and §9 of the paper [67]. This
proof involves a lot of interesting details which we shall review below. In particular,
we shall discuss in length the proof of the surjectivity of the map ϕμ , that is, the fact
that for any element [F ] of MF (μ), there exists a hyperbolic structure g ∈ T (S)
such that [F ] = [Fμ (g)]. This surjectivity is contained in Proposition 9.2 of [67].
Let us first make a few remarks about the proof of the surjectivity of φμ . Given
an element [F ] ∈ MF (μ), the transverse measure of a representative F of [F ]
determines a measure on each leaf of μ, which we think of as the one-dimensional
Lebesgue measure induced on the leaves of μ by the hyperbolic metric that we seek.
Of course this is not enough to determine the hyperbolic metric; to do this, one needs
to define an isometry between each connected component of S \ μ and a hyperbolic
ideal triangle, and then define in a consistent way a gluing of the various ideal triangles
among themselves. In the case where μ is an ideal triangulation, that is, if every leaf
of μ is isolated, then the metric g can easily be defined by gluing edge-to-edge the
ideal triangles (which are finite in number). This gluing is determined by the measured
foliation by saying that the algebraic distances between the distinguished points on
each edge of μ that are associated to the hyperbolic metric and to the measured foliation
F coincide. Here, the distinguished points on an edge of μ that are associated to a
measured foliation F are the hitting points of the singular leaves starting at the singular
points of F that are contained in the ideal triangles that are on each side of that edge.
But in the case where μ has non-isolated leaves, the gluing has to be defined by an
infinite process, and the main point of the proof is the convergence of that process.
Let us fix a geodesic lamination μ on S. We start by proving some useful facts
about measured foliations totally transverse to μ.
Lemma 3.11. Let β be an infinite half-leaf of μ. Then exactly one of the following
occurs:
(1) β converges to a cusp;
(2) β has a recurrence point contained in the stump of μ (that is, there exists a point
x in the stump of μ such that β returns infinitely often in any neighborhood of x).
Proof. Suppose that β does not converge to a cusp. Then there is a nonempty com-
pactly supported geodesic sub-lamination in its closure. Since each compactly sup-
ported geodesic lamination admits a nonzero transverse measure (which may be of
smaller support), we obtain a compactly supported measured geodesic sublamination
μ0 of μ in the closure of β. The sub-lamination μ0 is contained in the stump of μ. The
lemma follows then from the fact that each point in a measured geodesic lamination
of compact support is a recurrence point. 2
Lemma 3.12 (Infinite measure for infinite half-leaves of μ). Let F be a measured
foliation totally transverse to μ. Then the F -transverse measure of any infinite half-
leaf of μ is infinite.
Proof. We use Lemma 3.11. If β converges to a cusp, then it has infinite transverse
measure, by assumption on F (Definitions 3.9 and 3.6). Suppose now that β does
not converge to a cusp and let x be a recurrence point of β in the stump γ of μ. We
choose a homeomorphism ϕ : U → [0, 1]2 such that U ⊂ S contains x in its interior,
such that the image of F|U by ϕ is made of vertical segments (which we call plaques)
defined by the equations u = constant, where u is the coordinate of the first factor of
[0, 1]2 , and such that the connected components of the intersection of μ with U are
segments joining the “vertical” sides {u = 0} and {u = 1} of [0, 1]2 . In particular,
there is a segment of γ ∩ U containing x and crossing the plaques transversely from
one vertical side to the other vertical side. Since x is a recurrence point of β, there
are infinitely many segments in β ∩ U whose images by ϕ cross the square [0, 1]2
from one vertical side to the other one. All these segments have the same transverse
measure. Therefore, the transverse measure of β is infinite. 2
In the rest of this section, the geodesic lamination μ is complete.
Let [F ] be an element of MF (μ) and let us choose a representative F of [F ] that
is totally transverse to μ.
From the transversality of F to μ and by an Euler characteristic argument, we can
see that F has exactly one singular point in each connected component of S \ μ or on
the boundary of that component, and that this singular point is a 3-prong singularity.
Up to a small homotopy of F , we can suppose that all the singular points of F lie in
the interior of the components of S \ μ.
We first show that F is homotopic (in a sense that will be made precise) in each
ideal triangle to the horocyclic foliation of some hyperbolic structure on S.
We fix an ideal triangle T ∗ in H2 , equipped with its horocyclic foliation, modified
by collapsing the non-foliated region onto a spine (that is, a tripod). We call FT ∗ this
measured foliation. We perform the collapse in such a way that it induces the identity
map on the boundary of T ∗ .
Let T be a connected component (including its boundary) of S \ μ equipped with
the measured foliation induced by F . Notice that T has the structure of a topological
ideal triangle. (In fact, it becomes a genuine ideal triangle as soon as we equip the
surface S with a hyperbolic metric, but we do not need to introduce arbitrarily such a
metric right now.) We have the following:
Lemma 3.13 (Equivalence to horocyclic foliation). There exists a homeomorphism

ϕT : T → T ∗ respecting the foliations of these triangles (that is, sending leaves to
leaves and preserving the transverse measures) which is unique up to the choice of a
one-to-one correspondence between the ideal vertices of T and those of T ∗ , and up
to post-composition of φT with a homeomorphism of T ∗ that preserves its horocyclic
foliation and that induces the identity map on the boundary of that triangle.
Proof. Let FT denote the measured foliation induced by F on T . Choose a one-to-

one correspondence between the ideal vertices of the triangles T and T ∗ . There is a
homeomorphism ϕT : T → T ∗ preserving the singular points and mapping each leaf
of FT to a leaf of FT ∗ , preserving the transverse measures. To define ϕT , it suffices
to start by mapping the three singular leaves of FT to the three singular leaves of FT ∗
(respecting the correspondence between these leaves induced by the correspondence
between the ideal vertices) and then continue mapping homeomorphically each leaf of
FT onto a leaf of FT ∗ preserving the transverse measure. This is possible because the
transverse measure induced by FT on each half-edge of T is infinite (Lemma 3.12).
We thereby get a map ϕT : T → T ∗ which is well defined up to the deformations
mentioned in the statement of the lemma. 2
A spike of a hyperbolic ideal triangle of H2 is a connected component of the com-
plement of the closure of the unfoliated region of its horocyclic measured foliation
(or a connected component of the complement of the singular graph, if the unfoli-
ated region has been collapsed onto a spine). Likewise, given a complete geodesic
lamination μ on S together with a totally transverse foliation F , a spike of S \ μ is a
connected component of the complement of the singular leaves of the restriction of F
to a connected component of S \ μ.
Using Lemma 3.13, we now define a function on each spike of S \ μ. Let q be such
a spike and let T be the triangle of S \ μ that contains it. We choose a homeomorphism
ϕT between the triangle T and a fixed hyperbolic triangle T ∗ in H2 that preserves the
measured foliations of these triangles, as in Lemma 3.13. The sharpness function
fq : q → [0, ∞) is defined by

fq (x) = − log length α(ϕT (x))
where, for each x ∈ q, α(ϕT (x)) denotes the leaf of the horocyclic foliation corre-
sponding to the leaf of FT ∗ passing through the point ϕT (x) ∈ T ∗ . In other words,
α(ϕT (x)) is the horocyclic arc that passes through the point of T ∗ that is mapped to
the point ϕT (x) by the map T ∗ → T ∗ that we used to collapse the unfoliated region
of that ideal triangle onto a tripod. It is easy to check that the function fq does not
depend on the choices involved in the definition of ϕT . Since the length of any arc
of the horocyclic foliation of an ideal triangle is bounded from above by 1 (which
is the length of the horocyclic arc that is on the boundary of the unfoliated region),
fq (x) is nonnegative for all x in q. Furthermore, this function converges to ∞ as the
point x converges to the cusp of q. (In fact, fq is a linear function with respect to
the “distance” from x to the unfoliated region, where this distance is defined by the
transverse measure of any transverse arc joining x to a point in the unfoliated region.)
Thus, associated to the measured foliation F , we have a family {fq } of functions,
one for each spike q of S \ μ. We shall use these functions to construct the hyperbolic
structure g satisfying [F ] = [Fμ (g)]. Following Thurston, to construct g, we construct
a map from the universal covering of S to H2 . This map will turn out to be the
developing map of that structure.
Let us choose a basepoint s on S and let π : S̃ → S be the topological universal
covering of S. We regard S̃ as the set of homotopy classes of paths with fixed endpoints
α : [0, 1] → S with α(0) = s; the projection map π : S̃ → S is given by [α]
→ α(1).
Given such a path α, we can replace it by a path α ∗ which is homotopic to α
by a homotopy with fixed endpoints, which is made of a finite concatenation of seg-
ments, each such segment being contained either in a leaf of F or in a leaf of μ. We
call such a path α ∗ a horogeodesic path. (Of course, we are using this terminology
because we imagine S as being equipped with a hyperbolic structure g for which μ
is geodesic, with F being the corresponding horocyclic measured foliation; in fact, g
will be the structure that we are seeking.) The construction of α ∗ may be done as
follows. We start by taking a train track approximation τ of μ, and for this we can
use an auxiliary hyperbolic metric for which μ is geodesic, and take the leaves of τ
so that they are nearly parallel to the geodesic leaves of μ, in the metric sense, as in
[65], §9.5.
We can cover μ by a finite collection of rectangles obtained by thickening the
edges of the train track τ . The horizontal sides of these rectangles are parallel to
the edges of τ , and the leaves of μ cross each such rectangle from a vertical side
to a vertical side. We also take τ to be close enough to μ in such a way that F is
transverse to τ . Moreover, up to taking smaller rectangles, we can assume that the foli-
ation induced by F on each rectangle consists in segments joining its horizontal sides.
Performing a homotopy with fixed endpoints on the path α, we can replace it by a
path whose intersection with each rectangle is a finite number of segments joining
its horizontal sides. Finally, replacing each such segment (by using again a homo-
topy with fixed endpoints) by a concatenation of (at most three) segments in leaves of
μ and in leaves of F , we obtain the desired horogeodesic path α ∗ . From this con-
struction, we can assume without loss of generality that α ∗ satisfies the following two
properties:
(1) each subsegment of α ∗ contained in a leaf of F or in a leaf of μ does not backtrack
on that leaf;
(2) there does not exist any closed immersed disk in S whose interior is embedded
in that surface and whose boundary is the union of a subsegment of α ∗ with a
segment which is either in a leaf of F or in a leaf of μ.
A horogeodesic path satisfying these two properties will be called a minimal horo-
geodesic path.
From now on, we regard S̃ as the set of homotopy classes with fixed endpoints of
minimal horogeodesic paths α ∗ : [0, 1] → S satisfying α ∗ (0) = s.
We note that in the construction that follows, a change in the auxiliary hyperbolic
metric will only affect the developing map of the hyperbolic structure that we are
seeking by conjugating it by an orientation-preserving isometry of H2 . Precisely, this
hyperbolic structure can be seen as the plane H2 mod out by a subgroup of the group
of isometries of H2 acting properly discontinuously on that space and defined up to
conjugation, and therefore the choice of the auxiliary hyperbolic metric will not affect
the final result.
We now fix a minimal horogeodesic path α ∗ . We shall say that a subpath of α ∗
is maximal if its image is either contained in a leaf of F or in a leaf of μ, and if
this subpath is maximal (with respect to inclusion) for that property. We shall call a
maximal subpath of α ∗ , if it is contained in a leaf of F (respectively of μ), a maximal
vertical (respectively horizontal) subpath. We shall consider the image of α ∗ as being
equipped with the natural orientation induced from that of the interval [0, 1].
Let c be a maximal vertical subpath of α ∗ . The connected components of the
complement of c with respect to μ form a linearly ordered set of open intervals (aj , bj ),
j ∈ J , each interval (aj , bj ) joining opposite edges of a topological ideal triangle Tj
of S \ μ. To each such interval (aj , bj ), we now associate a parabolic isometry of H2 .
Recall that to the triangle Tj containing (aj , bj ), there is an associated homeomorphism
ϕTj which is defined up to an isotopy that induces the identity map on the boundary of
that triangle and up to a permutation of the ideal vertices, such that ϕTj maps Tj onto
some fixed ideal triangle T ∗ of H2 equipped with its horocyclic foliation, respecting
the foliations and the transverse measures. We also have an associated family of
sharpness functions {fq }, one function for each spike q of S \ μ. Let us fix an oriented
geodesic line γ in H2 which we shall henceforth refer to as the base geodesic, and
let us choose a basepoint x0 on (the image of) that geodesic. For each j in J and for
each choice of a hyperbolic ideal triangle T ∗ having the base geodesic γ as one of
its edges and situated to the right of γ (we are using an orientation of the hyperbolic
plane), we consider the associated orientation-preserving map ϕTj between Tj and T ∗
that sends the edge of Tj containing aj to the base geodesic γ . (The fact that ϕTj
preserves orientations implies that the images of the vertices of Tj are well defined.)
As the ideal vertex of T ∗ that is not an endpoint of γ varies on the circle at infinity,
the image ϕTj (aj ) varies on the base geodesic γ , and we can manage so that ϕTj (aj ) is
the point x0 . Now let Pj be the parabolic isometry of H2 that fixes the endpoint of the
spike of T ∗ that contains the segment ϕTj ((aj , bj )) and that sends ϕTj (aj ) to ϕTj (bj ).
Note that the length of the horocyclic arc joining the points ϕTj (aj ) and ϕTj (bj ) equals
e−fq (aj ) . After associating a parabolic isometry Pj to each open interval (aj , bj ), we

associate to the subpath c of α ∗ the (possibly infinite) product of isometries j ∈J Pj
(see Figure 16). Note that the association of the product of isometries is in the reverse
order of the one used for composition of maps. Thus, for instance, if α crosses the
ideal triangles T1 , T2 , T3 in that order with associated parabolic isometries P1 , P2 , P3
respectively, then the product will be P1 P2 P3 .
P1 x0
x0
P1 P2 x0
P1 P2 P3 x0
Figure 16
Let Isom+ (H2 ) be the group of orientation-preserving isometries of H2 . Given a

basepoint x0 in H2 , we endow this group with the following complete left-invariant
metric
dx0 (A, B) = sup |A(x) − B(x)|e−|x0 −x| for all A, B ∈ Isom+ (H2 ),
x∈H2
where for every x and y in H2 , |x − y| denotes their hyperbolic distance. It is easy to

see that if x1 ∈ H2 is another basepoint, then we have
dx1 (A, B) ≤ dx0 (A, B)e|x0 −x1 | . (3.3)
The distance between A ∈ Isom+ (H2 ) and the identity I is denoted by Ax0 = A.
Using the left-invariance of the metric dx0 , it is easy to see that we have dx0 (A, B) =
B −1 A for all A and B in Isom+ (H2 ). We need to estimate the norm of a parabolic
element.
Lemma 3.14. Let P be a parabolic element fixing a point p in S∞ 1 , the boundary at
infinity of H2 , and let l(0) denote the length of the horocyclic arc joining x0 and P (x0 )
and contained in the horocycle centered at p. Then P ≤ l(0).
Proof. Let γ : R → H2 be the geodesic line in H2 satisfying γ (0) = x0 , having p as

one of its endpoints and oriented so that lim t→−∞ γ (t) = p. Consider the foliation
of H2 by the family {Ht : t ∈ R} of horocycles centered at p, where for each t in R,
Ht denotes the horocycle through γ (t), and let l(t) denote the length of the arc in Ht
joining γ (t) to P (γ (t)).
p
Ht
x P (γ (t))
P (x0 )
γ (t)
x0
l(0)
Figure 17. Notation used in Lemma 3.14.
Let x ∈ H2 . This point is on a unique horocycle Ht , t ∈ R. We have |P (x) − x| =

|P (γ (t)) − γ (t)|. Moreover, a geometric argument shows that the projection along
horocycles is distance-decreasing, that is, |x0 − x| ≥ |x0 − γ (t)| = |t|. Therefore,
P = supt∈R |P (γ (t)) − γ (t)|e−|t| .
A computation (for instance, using Figure 36 below) gives |P (γ (t)) − γ (t)| =
2 sinh−1 (l(t)/2), where l(t) denotes the length of the horocyclic arc joining γ (t) and
P (γ (t)). Using the fact that l(t) = e−t l(0) and using the inequality sinh−1 (x) ≤ x
for every x ≥ 0, we obtain
P = sup |P (γ (t)) − γ (t)| e−|t|

t∈R

−1 l(t) −|t|
= sup 2 sinh e
t∈R 2
≤ l(0)e−t−|t| ≤ l(0).
This proves Lemma 3.14. 2

Proposition 3.15. The product of isometries j ∈J Pj that is associated to the maximal
vertical subpath c of α ∗ converges in Isom+ (H2 ).

Proof. In order to show that the product j ∈J Pj converges, we shall use the fact
that Isom+ (H2 ) is complete. Let J0 ⊂ J1 ⊂ · · · ⊂ Jn ⊂ · · · be a sequence of finite
subsets of J converging to J , that is, such that ∪n∈N Jn = J . Without loss
of generality,
we can assume that Card(Jn ) = n. For each n ∈ N, the finite product j ∈Jn+1 Pj is

obtained from the previous finite product j ∈Jn Pj by inserting one element en . More

precisely, we can find elements An , Bn , en ∈ Isom+ (H2 ) such that j ∈Jn Pj = An Bn

and j ∈Jn+1 Pj = An en Bn . We have
dx0 (An en Bn , An Bn ) = Bn−1 en Bn

= sup |Bn−1 en Bn (x) − x|e−|x0 −x|
x∈H2
−1 (y)|
= sup |Bn−1 en (y) − Bn−1 (y)|e−|x0 −Bn
y∈H2
= sup |en (y) − y|e−|Bn (x0 )−y|

y∈H2
= dBn (x0 ) (en , I )
≤ dx0 (en , I )e|Bn (x0 )−x0 | (using 3.3)
≤ en eBn .

Let n denote j ∈Jn Pj . We have
p

dx0 (n+p , n ) ≤ dx0 (n+k , n+k−1 )
k=1
p
≤ en+k−1 eBn+k−1 .
k=1
Let q be a spike of S \ μ. The intersection of q with α is a countable family of

horocyclic arcs which, once reordered, have decreasing lengths ln , n ∈ N. There is
a positive lower bound M(F, c), depending on the compact segment c, to the set of
numbers {ln − ln+1 }. We have ln ≤ l0 e−nM(F,c) . Moreover, l(0) ≤ 1. Therefore,
for every spike q, we have ln ≤ e−nM(F,c) . Since there are finitely many spikes in
S \ μ, then, using Lemma using 3.14, we can see that there exists an integer m = m(n)
such that en ≤ lm where lm is the length of the m-th horocyclic arc of q ∩ c, for
every spike q. Therefore, for all ε > 0, there exists N = N(F,
c) ≥ 0 such that for
all n ≥ N , we have en ≤ ε. Thus, the infinite product j ∈J Pj is bounded from
above by a convergent geometric series, whence the norm of any finite subproduct
is uniformly bounded from above. This shows that the sequence {n : n ∈ N} is
Cauchy, hence converges. 2
Let c be now a maximal horizontal subsegment of α ∗ . We wish to associate to c
a hyperbolic isometry of H2 . We do this by induction, and for this, we first consider
the isometries that were already associated to the sequence of maximal subsegments
of α ∗ that precede c. (In the case where c is the initial maximal subsegment of α ∗ ,
the preceding isometry is taken to be the identity.) We let C be the product of all
these isometries, this time the product being taken in the usual order of composition of
maps. In other words, if c1 , . . . , cn is the ordered set of maximal subsegments of α ∗
that precede c (that is, c1 starts at the point s and the endpoint of cn is the starting point
of c), and if C1 , . . . , Cn is the ordered set of isometries that we already associated to
c1 , . . . , cn , then we let C = Cn . . . C1 (and if c is the first maximal subsegment of α ∗ ,

we take C to be the identity). Now if we denote by γ the image of the base geodesic
γ by the isometry C, then, we associate to c the hyperbolic isometry H ∈ Isom+ (H2 )
that fixes γ , whose translation length is equal to the transverse measure of c with
respect to F and such that the orientation of the action of H on γ coincides with the
natural orientation associated with the parametrization of α ∗ .
Finally, we associate to α ∗ the product of the finite sequence of isometries associ-
ated to the finite sequence of maximal subsegments of α ∗ . (Remember that the product
is taken here in the order of composition of maps, as already mentioned.)
Lemma 3.16. The isometry associated to the minimal horogeodesic path α ∗ does not
depend on the choice of α ∗ in its homotopy class with fixed endpoints.
Proof. Any two horogeodesic paths belonging to the same class [α] are related to each
other by a finite number of moves of the type described in Figure 18.
Figure 18. The dotted lines represent leaves of the foliation F , the horizontal lines are leaves
of μ and the bold lines represent the maximal subsegments of the path α ∗ . (There may be other
leaves of μ that are inside the rectangles shown and that are not drawn.)
Therefore, it suffices to examine the invariance of the isometry associated to α ∗

by such a move. The product of isometries associated to the left-hand rectangle of
α ∗ in Figure 18 is of the form P H , where H is the hyperbolic isometry associated
to the maximal horizontal subsegment of α ∗ that is drawn in that figure and P is the
product of parabolic isometries that are associated to the maximal vertical segment.
Likewise, in the right-hand rectangle, the product is the form H P , where H and
P are defined in an analogous way. From the definition of the products associated
to the subsegments of α ∗ that are in F , it easily follows that P = P , since these
are products of a possibly infinite family of parabolic isometries indexed by the same
set and which are all equal. From the invariance of the transverse measure of F , the
translation distances of the two hyperbolic isometries H and H are equal. From this
fact and from the way these isometries are defined, it follows that H = P −1 H P ,
that is, P H = H P , which implies the invariance by the move considered. This
completes the proof of the lemma. 2
Thus, we have a map Is,F from S̃ to Isom+ (H2 ), which associates to each homotopy
class of curves with fixed endpoints the isometry associated to a minimal horogeodesic
path representative, and which gives a map
Ds,F : S̃ −→ H2
[α]
→ Is,F ([α]) x0 .
This map is a local isometry. Changes in the choice involved in any of the parameters s
and F have the effect of conjugating the map Ds,F by an isometry (see [70]).
This map induces a representation of the fundamental group in Isom+ (H2 ), which
is free and discrete, since the quotient of H2 by the image group [F ] is the surface S.
We obtain in this way a hyperbolic structure g on S whose developing map is precisely
the map Ds,F . The hyperbolic structure g is complete because each cusp of S equipped
with that structure has a neighborhood equipped with a foliation by closed leaves that
are quotients by [F ] of horocycles in H2 , which follows from the fact that the measured
foliation F that we started with is trivial around the punctures. This measured foliation
is isotopic to the horocyclic measured foliation Fμ (g). This completes the proof of
the surjectivity of the map φμ . This map is also injective, since it admits an inverse,
which is precisely the map which we produced in the proof of the surjectivity. It is
clear from the construction of the horocyclic foliation that the map φμ is continuous.
By invariance of domain, this map is a homeomorphism. This completes the proof of
Theorem 3.10. 2
3.6 Stretch lines, stretch rays, anti-stretch rays and stretch maps
In the next definition, we use Thurston’s cataclysm coordinates φμ introduced in §3.5.
Definition 3.17 (Stretch lines, stretch rays, etc.). Let μ be a complete geodesic lami-
nation on S. For any element F in MF (μ), the map from R to T (respectively from
[0, ∞) to T ) defined by t
→ φμ−1 (et F ) is called a stretch line directed by μ and
starting at g = φμ−1 (F ). A stretch line travelled up backwards is called an anti-stretch
line. More precisely, given a stretch line t
→ φμ−1 (et F ), the map from R to T de-
fined by t
→ φμ−1 (e−t F ) is called an anti-stretch line directed by μ and starting at
g = φμ−1 (F ). A stretch ray (respectively anti-stretch ray) directed by μ and starting
at g = φμ−1 (F ) is the restriction of a stretch line (respectively an anti-stretch line)
to the half-line R+ . Given a stretch (respectively anti-stretch) ray t
→ φμ−1 (et F )
(respectively t
→ φμ−1 (e−t F )) that starts at a point g in T , we shall denote by g t the
hyperbolic surface φμ−1 (et F ) (respectively φμ−1 (e−t F )). The hyperbolic structure g 0
is the structure g that we started with. Given a stretch ray starting at g, then, for each
t ≥ 0, the identity map, considered as a map from the surface S equipped with the
hyperbolic structure g 0 to the same surface equipped with the hyperbolic structure g t ,
is called the stretch map directed by μ.
We already mentioned that there exist punctured hyperbolic surfaces g equipped

with complete geodesic laminations for which the associated horocyclic foliation is a
union of cylinders that are all parallel to punctures. A stretch ray (or an anti-stretch ray)
starting at a hyperbolic structure g with a horocyclic foliation satisfying this property
is a constant ray, that is, we have g t = g for all t ≥ 0. For instance, in the example
of Figure 15, stretching along μ does not change the hyperbolic structure. One way
of seeing this is the fact that the relative distances between the distinguished points
remain constant (equal to zero). On the other hand, any complete lamination μ which
is not an ideal triangulation (that is, which is not a finite union of bi-infinite leaves
converging at both ends to punctures) has a nonempty stump. Therefore, in this case,
for all t = 0, the hyperbolic structure g t is distinct from g (and for the same reason,
for all 0 ≤ t1 < t2 , g t1 is distinct from g t2 ). To see this, it suffices to consider the
effect of the stretch map on any nonempty sublamination of μ that can be equipped
with an invariant measure (for instance, the stump of μ). In the surface g t , the length
of that measured sub-lamination is multiplied by the factor et , which implies that the
structure g t is distinct from g.
3.7 Thurston’s asymmetric metric
An asymmetric metric on a set X is a map L : X × X → R+ that satisfies all the

axioms of a metric except the symmetry axiom, and such that the symmetry axiom is
not satisfied, i.e. there exist x and y in X such that L(x, y) = L(y, x). Teichmüller
space is equipped with an asymmetric metric that was defined by Thurston. The
stretch lines that we considered in Section 3.6 are geodesics for that metric, provided
the stump of the complete lamination μ is not empty. We now recall the definition of
this metric.
Let g and h be two hyperbolic structures on S and let ϕ : S → S be a diffeomor-
phism of S which is isotopic to the identity. The Lipschitz constant Lip(ϕ) of ϕ is
defined as

dh ϕ(x), ϕ(y)
Lip(ϕ) = sup .
x =y∈S dg x, y
The logarithm of the infimum of these Lipschitz constants over all diffeomorphisms
ϕ in the isotopy class of the identity is denoted by
L(g, h) = log inf Lip(ϕ). (3.4)
ϕ∼IdS
It is obvious that L satisfies the triangle inequality. It is less obvious that L(g, h) ≥ 0
for all g and h (this uses the fact that any two hyperbolic structures on S have the same
area, see [67], Proposition 2.1).
Making g and h vary in their respective homotopy classes does not change L(g, h)
and thus we obtain a function which is well defined on T (S) × T (S), which is
Thurston’s asymmetric metric. We shall denote it by the same letter:
L : T (S) × T (S) → R+ .
Thurston showed that the quantity L(g, h) can also be computed by comparing
lengths of homotopic closed geodesics for the metrics g and h. More precisely, for
any homotopy class α of essential simple closed curves on S, we consider the quantity
lh (α)
rg,h (α) =
lg (α)
and we set
K(g, h) = log sup rg,h (α). (3.5)
α∈S
Again, it is easy to see that the function K satisfies the triangle inequality. The fact
that we have K(g, h) > 0 for all g = h is not trivial, and it is proved by Thurston in
[67], Theorem 3.1.
Since each homeomorphism ϕ considered in Formula (3.4) is isotopic to the iden-
tity, it preserves each homotopy class of simple closed curves in S. From that obser-
vation, it is easy to see that K ≤ L. In his paper [67], Thurston proves the following
result:
Theorem 3.18 (Thurston [67], Theorem 8.5). K = L.
Note that in the same way as in the definition of the Teichmüller metric (see
Remark 2.25 above), by using the extension of the hyperbolic length function to the
space of compactly supported measured geodesic laminations, we can replace the
supremum in 3.5 by the supremum over the set of compactly supported measured
geodesic laminations, that is, we have
lh (α)
K(g, h) = log sup = log sup rg,h (α). (3.6)
α∈ML0 lg (α) α∈PL0
By compactness of the space PL0 (S), the supremum in (3.6) is attained at some
measured geodesic lamination.
Along a stretch line directed by a complete geodesic lamination μ whose stump γ
is not empty, the stretch map between g and g t is et -Lipschitz and rg,g t (γ ) = et . This
gives L(g, g t ) = K(g, g t ) = t for all t ≥ 0 (and justifies the dilatation factor et ).
A measured lamination α attaining the supremum in (3.6) is generally not unique,
even up to a scalar factor. For instance, any nonempty sub-lamination of such a lam-
ination has the same property. There exists a geodesic lamination that is canonically
associated to an ordered pair of hyperbolic structures (g, h), but this lamination is not
necessarily equipped with a transverse measure. Thurston introduced it in the course
of proving Theorem 3.18. He first defined the notion of a maximally stretched geodesic
lamination from a hyperbolic structure g to a hyperbolic structure h. This is a (chain-
recurrent) geodesic lamination λ for which there exists an eK(g,h) -Lipschitz map from
a neighborhood of λ in the surface S equipped with the metric g to a neighborhood
of λ in the surface S equipped with the metric h. Thurston proved that the union of all
maximally stretched geodesic laminations from g to h is a geodesic lamination. This
geodesic lamination is denoted by μ(g, h).
Remarks. 1) For any η in ML0 (S), we can find two hyperbolic structures g and h
such that the supremum in rg,h (α) in (3.6) is attained for α = η. This follows from
Thurston’s construction in [67] §4, taking in that construction a complete geodesic

lamination μ that contains η and taking g and h (in the right order) on a stretch line
directed by μ.
2) If an element η in ML0 (S) realizes the maximum of rg,h (α) as above, then any
other element of ML0 (S) which is topologically equal to η (that is, with a possibly
different transverse measure) also realizes the maximum. This follows from Thurston’s
definition of μ(g, h) in [67], §8, which is independent of the transverse measure of
the lamination (see Theorem 8.1 of [67]).
One can see that the function L is generally not symmetric by looking at the
following example.
Example 3.19 (The metric L is not symmetric). In Figure 19, the surface S2,0 is
equipped with two hyperbolic structures g and h for which the left and right handles
are almost isometric. One can do explicit computations of lengths of closed geodesics,
by decomposing the two hyperbolic surfaces into hyperbolic pairs of pants, with the
central curve being an element of the decomposition, and with the two other closed
curves being the core curves of the two handles represented in Figure 19. Consider
the central cylinder that joins the handles. The height of this cylinder, measured in
the structures g and h respectively, is approximately equal to c and d, with d > c.
Applying the formulae that give the distance between two boundary components of a
hyperbolic pair of pants in terms of the length of the boundary components (see e.g.
[20], p. 151), we obtain lg (α) Ae−c and lh (α) Ae−d , where A is a constant that
depends on the size of the handles. Thus, the smallest Lipschitz constant of a map
from g to h is approximately equal to d/c, whereas the smallest Lipschitz constant
of a map from h to g is bounded below by e−c /e−d = ed /ec . The latter is very
large compared to d/c when d is large compared to c. Thus, we obtain examples of
hyperbolic structures g and h satisfying L(g, h) = L(h, g).
Ae−c
g
L(f ) ed /ec L(f ) d/c
h
Ae−d
d
Figure 19. An example (due to Thurston) of hyperbolic structures g and h satisfying L(g, h) =
L(h, g).
On the other hand, there are instances of distinct hyperbolic structures g and h
satisfying K(g, h) = K(h, g) (hence L(g, h) = L(h, g)). An example of such a pair
is given in Figure 20.
Figure 20. An example of distinct hyperbolic structures g and h satisfying K(g, h) = K(h, g):
h is the image of g by an order-two isometry which is not isotopic to the identity.
The notion of a geodesic (respectively of a geodesic line, or a geodesic ray) for

an asymmetric metric space can be defined as in the case of a genuine metric, except
that one has to be careful about the order of the variables in the case of an asymmetric
metric. More precisely, we shall say that a map γ : I → T (S), (where I is a closed
interval of R and where T (S) is equipped with the asymmetric metric L) is geodesic
if for all t ≤ t in I we have t − t = L γ (t), γ (t ) . Thurston proved the following
Theorem 3.20 (Stretch lines are geodesics, Thurston [67], Theorem 8.5 and Corol-
lary 4.2). Let μ be a complete geodesic lamination on S. If the stump of μ is not the
empty lamination, then any stretch line in T (S) directed by μ is a geodesic line for
Thurston’s asymmetric metric. In other words, we have, for every 0 ≤ t ≤ t ,

L(φμ−1 (et F ), φμ−1 (et F )) = t − t.
Theorem 3.21 (Concatenations of stretches, Thurston [67], Theorem 8.5). Let g and
h be two hyperbolic structures. Then we can pass from g to h by a geodesic which is
a finite concatenation of pieces of stretch lines along complete geodesic laminations
μ1 , · · · , μk , all of them containing μ(g, h). Furthermore, the number of such pieces
needed to go from g to h is bounded by a constant which depends only on the topological
type of the surface (that is, on the genus and the number of punctures). Such a geodesic
path is obtained in the following way. We first choose an arbitrary completion μ1 of
μ(g, h) and we stretch the structure g along μ1 until we reach a first point g satisfying
μ(g , h) = μ(g, h). We necessarily have μ(g , h) ⊃ μ(g, h). Starting now with g
instead of g and continuing in the same manner, we reach h after a finite number of
steps.
Remarks. 1) That the number of pieces in this statement is bounded from above
follows from the fact that there is a bound (that depends only on the topological type
of the surface) for the length of a strictly increasing (with respect to inclusion) sequence
of geodesic laminations.
2) The geodesic path leading from g to h is in general not unique, since it depends
on the choice of the completion at each step.
3) The bound in 1) does not imply that the number of distinct geodesics from g
to h is finite, since at each step there are in general infinitely many ways of completing
the given geodesic lamination.
4) An anti-stretch ray is (up to reparametrization) a geodesic ray for the asymmetric
metric on Teichmüller space which is dual to Thurston’s asymmetric metric, that is,
the asymmetric metric defined by
lg (α)
K ∗ (g, h) = log sup (3.7)
α∈S lh (α)
for any hyperbolic metrics g and h.
Here are some questions related to the example described in Figure 20.
• What does a typical geodesic from g to h look like? Is it a segment of a stretch line?
is it a concatenation of at least two such segments? Is it something else?
• Describe a geodesic from g to h that (up to reparametrization) is also a geodesic
from h to g when traversed in the opposite direction.
We now discuss a few properties of Thurston’s asymmetric metric L with respect
to sequences in T (S) that tend to infinity. Then, we shall discuss the topologies that
this asymmetric metric induces on T (S). Here, as usual, we say that a sequence (gn )
in T (S) tends to infinity (and write gn → ∞) if for any compact subset K of T (S),
we have gn ∈ T (S) \ K for all n large enough.
It is known that Teichmüller’s metric dT is proper (that is, closed balls are compact),
which implies that a sequence of points (gn ) in T tends to infinity if and only if for
every g in T (or, equivalently, for some g in T ), we have dT (g, gn ) → ∞ as n → ∞.
In [53], we prove the following analogous result for the asymmetric metric L.
Theorem 3.22. For any sequence (gn ) in T and for any h ∈ T , we have the following
equivalences:
gn → ∞ ⇐⇒ L(h, gn ) → ∞ ⇐⇒ L(gn , h) → ∞.
We deduce the following
Corollary 3.23. Let t

→ g t be a stretch ray starting at a hyperbolic structure g and
directed by a complete geodesic lamination μ, such that either the stump of μ is not
empty (or, equivalently, μ is not an ideal triangulation) or the horocyclic measured
lamination λμ (g) is not empty. Then the map t
→ g t is proper. (In other words, as t
tends to infinity, the point g t tends to infinity in Teichmüller space.)
Proof. If the stump of μ is not empty, then we have, by Theorem 3.20, L(g, g t ) = t,
which, by Proposition 3.22, implies that g t tends to infinity. If the horocyclic measured
lamination λμ (g) is not empty, then, as t → ∞, we have lg t (λμ (g)) → 0. Indeed, by
Proposition 3.3 and Lemma 3.9 of [52], there exists a constant C > 0 that depends
only on the topological type of the surface S such that lg t (λμ (g t )) ≤ C. (In fact, by
Lemma 3.9 of [52], we can take C = −χ where χ is the Euler characteristic of S.)
Note that the setting in [52] is that of closed surfaces, but the arguments work for
general surfaces of finite type. We have λμ (g t ) = et (λμ (g)), therefore lg t (λμ (g)) ≤
Ce−t , which tends to zero as t tends to infinity. This proves the claim. Now if
l (λ (g))
lg t (λμ (g)) → 0, then rg t ,g (λμ (g)) = l gt (λμμ (g)) → ∞, which gives L(g t , g) → ∞,
g
which, again by Proposition 3.22, implies that g t tends to infinity. 2
The following was obtained by L. X. Liu in [38]. It is also proved in [53].
Proposition 3.24. For any sequence (gn ) in T (S) and for any element g in T (S), we
have the following equivalences:
gn → g ⇐⇒ L(g, gn ) → 0 ⇐⇒ L(gn , g) → 0.
Associated to the asymmetric metric L, for each g in T and for each R > 0, there
are two different notions of closed balls, which we call the left closed ball centered at
g ∈ T (S) of radius R and the right closed ball centered at g ∈ T (S) of radius R, and
which we denote respectively by g B(R) and Bg (R). These are defined respectively
by
g B(R) = {h ∈ T (S) | L(g, h) ≤ R} (3.8)
and
Bg (R) = {h ∈ T (S) | L(h, g) ≤ R}. (3.9)

In [53], we also prove the following
Proposition 3.25. For any g in T (S) and for any R > 0, the closed balls g B(R) and
Bg (R), centered at g ∈ T (S) and of radius R, are compact for the usual topology.
One can similarly define left (respectively right) open balls in T , by taking the
same definitions in (3.8) (respectively (3.9)) above, except that the large inequality is
replaced by a strict inequality.
Since the metric on Teichmüller space that we are studying here is asymmetric, it
is natural to consider two topologies on that space, the one generated by the collection
of right open balls, and the one generated by left open balls. We shall call the first of
these topologies the topology associated to Thurston’s asymmetric metric K and the
second one the topology associated to its dual asymmetric K ∗ . It is easy to see that
a sequence (gn ) of points in Teichmüller space converges to a point g in that space
for the topology generated by the left (respectively right) open balls if and only if we
have K(g, gn ) → 0 as n → ∞ (respectively K ∗ (g, gn ) → 0) (that is, we have the
same convergence criteria than for genuine metrics). Proposition 3.24 is equivalent to
the following
Corollary 3.26. The topologies on Teichmüller space generated by Thurston’s asym-

metric metric and by its dual asymmetric metric coincide with the usual topology of
that space.
From Theorem 3.24, Proposition 3.25 and from Corollary 3.26, we obtain the
following
Proposition 3.27. Teichmüller space equipped with Thurston’s asymmetric metric

is proper. More precisely, left and right closed balls are compact for the topology
generated by the asymmetric metric.
Equivalently, we have the following
Proposition 3.28. For every point g in Teichmüller space, the functions h

→ L(g, h)
and h
→ L(h, g) are proper.
Herbert Busemann developed a theory of spaces (X, δ) satisfying all the axioms
of a metric space except the symmetry axiom, and satisfying the following additional
axiom:
δ(x, xn ) → 0 ⇐⇒ δ(xn , x) → 0 for any x and for any sequence (xn ) in X.
For such spaces, there is a well-defined topology on X, the one associated to the
bona fide metric max{δ(x, y), δ(y, x)}. Theorem 3.24 insures that Thurston’s asym-
metric metric fits into that theory, and that this topology is the same as the one generated
by the collection of left (or right) open balls. Given such a space (X, δ), Busemann
introduced the following notion for such spaces (see [14], Chapter 1): (X, δ) is com-
plete if and only if for every sequence (xn )n≥0 in X satisfying δ(xn , xn+m ) → 0 as n
and m → ∞, the sequence converges to a point in X.
In that setting, the following generalization of a classical theorem of Hopf and
Rinow holds (see [14], Theorem 8, p. 4). If a locally compact space (X, δ) satisfies
Busemann’s axioms stated above, and if this “generalized metric” δ is intrinsic in the
sense that for any x and y in X, the value δ(x, y) is equal to the infimum of the lengths
of all curves joining x and y, then the following two properties are equivalent:
(1) left closed balls in X are compact;
(2) X is complete.
Here, the length of a curve is defined in a similar way as in the case of a genuine
metric space. We do not need to enter here into the details of this definition because,
by a result of Thurston (see [67], §6) that we already mentioned, Thurston’s asym-
metric metric is geodesic (that is, any two points can be joined by a geodesic), and a
generalized geodesic metric in this sense is intrinsic (see [14], p. 3).
Thus, we obtain the following
Corollary 3.29. Thurston’s asymmetric metric is complete.

3.8 Comparison with other metrics

Other results on the comparison between the Teichmüller metric dT and Thurston’s
asymmetric metric L on Teichmüller space have been obtained by L. X. Liu [37] and
in joint work by Y. Choi and K. Rafi [16]. Let us start by stating the following result
by Liu:
Theorem 3.30 (Liu [37]). Let T be the Teichmüller space of a closed surface of
negative Euler characteristic. Then,
(1) there does not exist any constant k > 0 such that dT (g, h) ≤ kL(g, h) for every
g and h in T ;
(2) there does not exist any constant k > 0 such that such that dT (g, h) ≤ kL(h, g)
for every g and h in T .
Choi and Rafi consider the following symmetrization dL of Thurston’s asymmetric

metric, which they call the Lipschitz distance on Teichmüller space:
dL (g, h) = max{L(g, h) + L(h, g)}
for g and h in T .
They obtain results which compare this metric with Teichmüller’s metric. These
results are based on a geometric study which they make, which is inspired by work of
Yair Minsky, involving a Margulis constant for Teichmüller space, and the existence of
regions in that space that have a product structure. (Whereas Minsky’s work involves
estimates of ratios of extremal lengths of curves, the work of Choi and Rafi involves
estimates of ratios of hyperbolic lengths of curves.)
More precisely, Choi and Rafi take a number ε0 > 0 which is less than the Margulis
constant of Teichmüller space, and they define the thick part of Teichmüller space to
be the set of points in that space which are represented by hyperbolic metrics whose
injectivity radius is larger than ε0 . Then, they say that a simple closed curve α which is
not homotopic to a point is short with respect to a hyperbolic surface g if the hyperbolic
length of α in g, łg (α), is less than ε0 . They prove the following:
Theorem 3.31 (Choi and Rafi, [16]). Let T be the Teichmüller space of a surface of
finite type and of negative Euler characteristic. Then,
(1) If g and h are in the thick part of T , the distances dT (g, h) and dL (g, h) are
equal up to an additive constant which is independent of g and h. More precisely,
there exists c > 0 such that for all g and h in the thick part of T , we have
dL (g, h) − c ≤ dT (g, h) ≤ dL (g, h) + C.
(2) If g and h are not in the thick part of T and do not have a common short curve,
the distances dT (g, h) and dL (g, h) are equal up to a multiplicative factor which
is independent of g and h. More precisely, there exists k > 0 such that for all g
and h in the thick part of T , we have

1
dL (g, h) ≤ dT (g, h) ≤ kdL (g, h).
k
(3) In the general situation (that is, if g and h are allowed to have common short
curves), the two distances dT (g, h) and dL (g, h) are not comparable. Precisely,
there are sequences gn and hn in T satisfying
dL (gn , hn ) → 0 and dT (gn , hn ) → ∞.
In the same paper, Choi and Rafi also show that the thin part of Teichmüller space
with the metric dL has a product structure.
More precisely, let be a collection of k disjoint and pairwise homotopically
distinct simple closed curves on S, and consider the set
Thinε (S, ) = {g ∈ T (S) | lg (γ ) ≤ ε ∀γ ∈ }.
Let
T = T (S \ ) × U1 × · · · × Uk ,
where S \ is the surface obtained from S by pinching all the curves in and where
for all i = 1, . . . , k, Ui is the subset {(x, y) | y ≥ 1/ε} of the upper-half plane H2 .
Finally, Choi and Rafi define the sup metric
dL = sup{dL(S\) , dL(A1 ) , . . . dL(Ak ) }
on T , where dL(S\) is the Lipschitz metric on T (S \ ) and where for i = 1, . . . , k,
dL(Ai ) is a modification of the hyperbolic metric on Ui . Then, the Fenchel–Nielsen
coordinates on T give rise to a natural homeomorphism
: Thinε (S, ) → T ,
and Choi and Rafi prove the following
Theorem 3.32 (Choi and Rafi, [16]). For g and h in Thinε (S, ), we have
|dL (g, h) − dL ((g), (h))| = O(1).
In the next three sections, we study explicit examples of stretch lines that are simple
enough to make computations.
3.9 Example I of a stretch line: the four-punctured sphere

In this section, S = S0,4 is the four-punctured sphere. We consider the simple case of
a hyperbolic structure on S that is obtained by gluing two ideal quadrilaterals that can
be exchanged by an order-two isometry of S that fixes pointwise each of the four edges
of the quadrilaterals. Let us call such a hyperbolic structure a symmetric hyperbolic
structure on the four-punctured sphere. Figure 21 represents the surface S as a union of
two isometric quadrilaterals, with vertices A, B, C, D. Each quadrilateral is equipped

with a diagonal in such a way that the two diagonals join the same cusps of the surface
S, say D and B, as represented in Figure 21.
A B
D C
Figure 21. The four-punctured sphere with its ideal triangulation μ.
The union of the edges of the quadrilaterals (that are glued by pairs in the surface),
together with the two diagonals that join D and B, constitute an ideal triangulation
of S which we denote by μ. In the case considered of a symmetric structure, the fact
that the hyperbolic structure is complete is equivalent to the fact that for each of the
edges AB, BC, CD and DA, the two distinguished points (in the sense defined in
§3.2) corresponding to the two ideal triangles that are adjacent to that edge coincide.
Furthermore, since the two quadrilaterals we started with are isometric, the algebraic
distance d between the distinguished points on each of the two diagonals joining the
vertices D and B coincide. As a matter of fact, the set of symmetric hyperbolic
structures on the four-punctured sphere is parametrized by R, since such a hyperbolic
structure is completely determined by the algebraic distance d.
The horocyclic foliation associated to a symmetric hyperbolic structure g together
with the ideal triangulation μ is a foliation which has one cylinder whose height is equal
to |d| (which could be equal to 0) and four other cylinders forming neighborhoods of
cusps. In other words, the associated horocyclic measured geodesic lamination λμ (g)
is a simple closed geodesic α equipped with a Dirac transverse measure of mass |d|
(and it is the empty foliation if d = 0). Stretching the hyperbolic structure g in the
direction μ always produces a symmetric hyperbolic structure. If t
→ g t , t ∈ R, is
the stretch line starting at g and directed by μ, then, provided d = 0, we have the
following:
(1) lg t (α) → 0 as t → ∞;
(2) the distance between the distinguished points on the diagonal BD for the hy-
perbolic structure g t equals et times that distance for the hyperbolic structure
g = g0 ;
(3) if ν is any compactly supported measured geodesic lamination on S with ν = α,
then, we have i(ν, α) = 0 and lg t (ν) → ∞ as t → ∞;
(4) limt→∞ g t = [α] as a point on Thurston’s boundary PMF 0 ;
(5) as t → −∞, g t does not go to infinity but converges to the hyperbolic structure
that is obtained by the completely symmetric gluing of the four ideal triangles
that we started with, that is, the gluing where the distances between the pairs of
distinguished points on the six edges of μ are all equal to zero.
3.10 Example II of a stretch line: the punctured torus

In this section, we consider the surface S = S1,1 , that is, the punctured torus, equipped
with a hyperbolic metric g satisfying some properties which we now state. Let μ be a
complete geodesic lamination on S whose leaves consist in a simple closed geodesic γ ,
together with a bi-infinite geodesic δ converging at both ends to the cusp, and in two
other bi-infinite geodesics, each having one end converging to the cusp and the other
end spiraling around γ , in such a way that the spiraling of the two geodesics takes
place in the same direction, as in Figure 22. The surface S \ μ has two connected
components, each of them, with its intrinsic metric, being the interior of a hyperbolic
δ
Figure 22. The complete geodesic lamination μ is the union of the closed geodesic γ , the
bi-infinite geodesic δ and two other bi-infinite geodesics that spiral around γ at one end and that
converge to the cusp of S at their other end.
ideal triangle. We choose the metric g so that its associated horocyclic lamination
λμ (g) is a closed geodesic α satisfying i(α, γ ) = 1, as represented in Figure 23.
In Figure 23 are represented the stump γ of μ, the closed geodesic α and the simple
closed geodesic β obtained from α by a left Dehn twist along γ .
We shall study the behaviour of the lengths of these particular closed geodesics
under the stretch and anti-stretch rays directed by μ and starting at g = g 0 . The
behaviour of the lengths of these geodesics will permit us to understand the change in
geometry of the surface g under the stretch map.
The closed geodesics α and β satisfy Card(γ ∩ β) = Card(α ∩ γ ) = 1. The
reason why we consider these geodesics is that eventually we want to get an idea of
the behaviour of the lengths of an arbitrary simple closed geodesic (and more generally
of an arbitrary compactly supported measured geodesic lamination) on S under the
stretch (respectively the anti-stretch) ray g t as t → ∞, and we expect this behaviour
to depend on the intersection pattern of that lamination with the stump of μ and with
the horocyclic measured geodesic lamination λμ (g). The general results are given
α β
Figure 23. α is the horocyclic measured geodesic lamination associated to g and μ. We are
interested in the behaviour of the lengths of the three curves α, β and γ along the stretch and
the anti-stretch ray directed by μ and starting at g.
in Theorems 3.34 and 3.35 below, and the results that we present in the example
considered here are illustrations of those general results. What makes things work
easily in this example is that we can determine the exact positions of these geodesics
for the structure g t as t varies.
The horocyclic foliation Fμ (g) is the union of a cylinder C with core curve α
and of another cylinder foliated by leaves that are parallel to the puncture of S, the
two cylinders being glued along the critical graph of Fμ (g). This critical graph is
represented in Figure 24. Figure 25 represents the cylinder C to which are attached
the nonfoliated parts of Fμ (g). In that figure, the pairs of regions with the same name
A
B
Figure 24. The critical graph of the foliation Fμ (g) with its two singular points A and B.
are identified in the surface S. In the same figure, we have drawn in bold lines the
segments induced by γ and δ on the cylinder C. This cylinder has a symmetry which
is probably more apparent if we cut the cylinder along the arc induced on it by δ. After
this cutting, we obtain Figure 26, that is, a large quadrilateral R which is the union of
two smaller isometric quadrilaterals R1 and R2 . (The internal face of the cylinder C
of Figure 25 is the quadrilateral R of Figure 26.) The quadrilateral R has an order-two
symmetry of center w, where w is the midpoint of the segment induced by γ on the
cylinder C. Let s denote the midpoint of the arc induced by δ on that cylinder.
γ
b
b
a
a
δ
Figure 25. The cylinder C and on its boundary the nonfoliated regions of Fμ (g). These
nonfoliated regions are labeled a and b, these letters corresponding to the points A and B
respectively on the critical graph of Figure 24. The two arrows indicate that the corresponding
spikes converge to the cusp.
By the uniqueness of the closed geodesic in each homotopy class of closed curves
on S, the closed geodesic α is preserved by this order-two symmetry of R. Therefore
this geodesic passes by the points s and w. The same holds for the closed geodesic β,
which passes by w (see Figure 26).
All these results on the relative positions of α and β remain true when the metric
g is replaced by the stretched (respectively anti-stretched) metric g t .
To do computations, we can use Figure 27 in which the quadrilateral R1 is repre-
sented in the upper half-plane model of hyperbolic space. In that figure, lt denotes the
length of γ for the structure g t . However, it is possible to determine the asymptotic
behaviour of the lengths of the closed curves α, β and γ along the stretch (respectively
anti-stretch) ray without computations. It suffices for that to determine the behaviour
of the lengths of the sides of R1 along these rays. We refer to Figure 26, in which
the common length of the horizontal sides of R is equal to the length of γ , and the
height of R is defined to be the length of the leaf of Fg t (μ) that is equidistant from
the boundaries of C. This last quantity is an upper bound for the length of α. Now
we observe the following facts:
(1) The distances between the non-foliated regions grow linearly with et along the
stretch ray, and they decrease to zero along the anti-stretch ray.
(2) In each spike, the lengths of the horocyclic arcs that are not on the boundary of
a non-foliated region and that are leaves of the horocyclic foliation decrease to
zero along the stretch ray and grow to 1 along the anti-stretch ray.
Using these facts, it is easy to see that along a stretch ray, the height of R tends to
zero, whereas the common length of the horizontal sides of R tends to infinity. Along
an anti-stretch line, the height of R tends to infinity whereas the length of a horizontal
side tends to zero. Now since we know the exact positions of the various curves, we
obtain the following, along the stretch ray g t :
δ
β
s
b
b R1
γ
a
R2
a
s
α β
Figure 26. The quadrilateral obtained by cutting the cylinder C along δ.
(1) limt→∞ lg t (γ ) = ∞;
(2) limt→∞ lg t (α) = 0;
(3) limt→∞ lg t (β) = ∞.
Likewise, along the anti-stretch ray g t , we have
(1) limt→∞ lg t (γ ) = 0;
(2) limt→∞ lg t (α) = ∞;
(3) limt→∞ lg t (β) = ∞.
This example is rather limited in scope because the surface considered is not large
enough so as to contain closed curves that are disjoint from the stump, or from the
horocyclic measured lamination, or from both. This will be possible in the example
which we consider next.
3.11 Example III of a stretch line: the closed surface of genus 2

In this example, S = S2,0 is a closed surface of genus 2 equipped with a complete
geodesic lamination μ whose stump consists of two nonseparating closed geodesics
γ1 and γ2 , represented in Figure 28, with the other leaves of μ spiraling around γ1 ∪ γ2
in the same direction, as represented in Figure 29. We choose a hyperbolic structure
g on S whose associated horocyclic measured geodesic lamination λμ (g) consists
of a simple closed geodesic α, satisfying Card{γ1 ∩ α} = Card{γ2 ∩ α} = 1. The
topological pattern made by the curves γ1 , γ2 and α is represented in Figure 28.
As in the preceding examples, we study the behaviour of the lengths of some
particular simple closed geodesics under a stretch or an anti-stretch ray directed by μ
e lt
b
R1
w
β
1
b
s
0 x
α
Figure 27. The quadrilateral R1 in the upper half-plane model of H2 is the non-shaded region.
and starting at g. The closed geodesics considered here are γ1 and γ2 (chosen because
they are contained in the stump of μ), the geodesic α, and the two geodesics β1 and
β2 that are represented in Figure 30, satisfying (γ1 ∪ γ2 ) ∩ β1 = ∅, (γ1 ∪ γ2 ) ∩ β2 = ∅
and α ∩ β1 = α ∩ β2 = ∅. Again, the explicit computations that we do in this example
are illustrations of Theorems 3.34 and 3.35 below.
The horocyclic foliation Fμ (g) is a foliation with one cylinder with core curve in
the class of α. In Figure 31 we have represented the critical graph of Fμ (g) (after
collapsing the non-foliated regions). Let C denote the cylinder obtained by cutting
the surface along the critical graph of Fμ (g). This cylinder is represented in Figure 33
(together with the non-foliated regions on its boundary, as in the preceding example).
Each connected component of the intersection of the leaves of μ with C is a segment
γ1
γ2
Figure 28. The union γ1 ∪ γ2 represents the stump of the complete geodesic lamination μ, and
α represents the horocyclic measured geodesic lamination λμ (g) of the surface S = S2,0 .
Figure 29. The geodesic lamination μ, with stump γ1 ∪ γ2 . There are six leaves of μ which are
not contained in the stump. Each of these leaves is a bi-infinite geodesic that spirals around γ1
or γ2 . The spiraling around each curve γ1 and γ2 is in the same direction.
γ1
β2 γ2
β1
Figure 30. We are interested in the behaviour of the lengths of the five closed geodesics repre-
sented here under the stretch line directed by μ (of stump γ1 ∪ γ2 ) and passing by the hyperbolic
structure whose horocyclic measured geodesic lamination is α.
that joins the two boundary components of that cylinder, as shown in Figure 33. From
the definition of the transverse measure of Fμ (g), it follows that all these connected
components have the same length. In particular, we have lg (γ1 ) = lg (γ2 ).
We consider the cylinder C as the union of four quadrilaterals R1 , . . . , R4 . This
decomposition into quadrilaterals is induced by γ1 and γ2 and by two other geodesic
segments that join distinguished points on the leaves of μ, whose trace on the cylinder
C cut up along γ1 is shown in Figure 34. It is easy to see that the quadrilaterals
R1 , . . . , R4 are congruent. For each of these quadrilaterals, two of the opposite sides
are geodesics contained in leaves of μ, and the remaining two sides are made out
of horocycles contained in leaves of Fμ (g). (Note that each such side is a segment
of a horocycle and not just a union of segments of horocycles. This can be seen by
examining Figure 35 which represents one of the quadrilaterals drawn in the upper
half-plane.) The cylinder C has an order-two symmetry with respect to the midpoint
of each geodesic side of any of the quadrilaterals Ri . To see this, let us consider
the quadrilateral R obtained by cutting the cylinder C along the segment γ1 . This
quadrilateral is represented in Figure 34, with its induced decomposition into the four
B D
A C
Figure 31. The critical graph of the foliation obtained from Fμ (g) by collapsing the non-foliated
regions on the points A, B, C, D.
quadrilaterals R1 , . . . , R4 . The four points s, w, u, v are the intersection points of the

leaf of Fμ (g) that is equidistant from the boundary of C (distances being measured
with respect to the transverse measure of that foliation) with the geodesic sides of the
quadrilaterals R1 , . . . , R4 . The point u is the intersection point of that leaf with γ2 .
We claim that u is a center of symmetry for the segments induced on R by each of
the curves α, β1 and β2 . To see this, first of all, note that these segments are situated
as shown in Figure 34 because of the combinatorics of the gluing of the boundary
components of R. Now, if we call α , β1 , β2 the images of the curves α, β1 , β2 by
the symmetry of center u, then α , β1 , β2 are also simple and closed since the interior
of R is symmetric with respect to u. Since these curves are closed geodesics that are
homotopic to the closed geodesics α, β1 and β2 respectively, they coincide with them.
In particular, α passes through the four points s, w, u, v.
For later use, we compute the value x, shown in Figure 35, which represents the
quadrilateral R1 in the upper half-plane. This value x is a Euclidean distance, measured
on the boundary of the upper half-plane. It is also equal to the hyperbolic length of
the horocyclic edge of the rectangle R1 lying at height 1, in that same figure. The
Euclidean distance y represented in the same figure is equal to 3, because it is equal to
the length of three horocyclic segments that are boundaries of non-foliated triangles
in the standard horocyclic foliation of an ideal triangle, and that are at height 1 in the
upper half-plane. Therefore, we have the following (see also the caption of Figure 35):
3 3el/2
x = y(1 + e−l + e−2l + · · · ) = = ,
1 − e−l 2 sinh(l/2)
where l = lg (γ1 ) = lg (γ2 ).
Now we compute the lengths of the segments labeled p, q and r in Figure 37. These
segments are induced by α, β1 and β2 respectively on the quadrilateral represented.
We use the formula of Figure 36 giving a relation between the length and the slope.
l/2
For the segment p, the slope is 2ex , therefore the length is
x
3
2 sinh−1 = 2 sinh −1
.
2el/2 4 ( l/2)
c γ2
a c a γ1 b1 d1 c
d d1 a b2 d2
b b2
b d b1 d2
a c
b d
a c a1 c2
a c
a2 c1
b a2 c2
γ1 γ2
a1 c d
b d b d 1
a c a c b1 d1 c
b1 d a b2 d2
b d b2 d1 2
c
a
b d
a c c1
a b a2 c2
a1 2 c2
b d b d a1 c 1 d
a c a c b1 d1 c
b1 d2 d2
d d1 a b2
b b2
a c
b d
a c
a c1 a2
a1 2 c2 b c2
b d b d a1 c1 d
a c a c
b1 d1 c
b1 d2
d1 d2
b d b2 a b2
a c
b d
a c c1
a2
a1 c2 b a2 c2
d b d a1 c1 d
Figure 32. Line 1 represents the passage from the singular graph to the cylinder C obtained
by cutting the surface along the singular graph (see Figure 33). On lines 2, 3 and 4 we have
repeated the cutting, showing at each line the two closed curves γ1 and γ2 in bold lines, together
with two other leaves of μ that spiral along these two closed curves.
γ1
b a
a
b a
b c
c
d d
d
c
γ2
Figure 33. The cylinder C, together with the non-foliated regions of ideal triangles on its
boundary. Non-foliated regions with the same names are identified in the surface S.
γ1
β2
s
R1 a
a
w
R2 c
b γ2
u d
d R3
v c
c R4
s a
d b
b β1 γ1
Figure 34. The quadrilateral R representing the cylinder C cut along the geodesic γ1 . This
quadrilateral is the union of the four quadrilaterals R1 , . . . , R4 that are drawn. The two horizontal
sides of these quadrilaterals that are not labeled are geodesic segments in μ that join distinguished
points on the edges of that lamination. We have also represented the leaf of Fμ (g) that is
equidistant from the two boundary components of C, together with its intersection points s, w,
u, v with the geodesic edges of the quadrilaterals R1 , . . . , R4 . Finally, we have represented the
geodesic segments induced by β1 and β2 on the quadrilateral R. Each of these segments has
two boundary points on the same vertical side of R. The trace of β2 is drawn in bold lines.
el
l
R1
0 y x
Figure 35. The quadrilateral R1 , represented in the upper half-plane model of the hyperbolic
plane, is the region between the two shaded regions. It is bounded by two vertical geodesic
segments, both of length l = lg (γ1 ) = lg (γ2 ), and two horizontal horocylic segments, the lower
one having length x and the upper one having length x/ h = xe−l . The geodesic segment
corresponding to γ1 is the segment contained in the vertical ray starting at the origin, joining the
point at height 1 to the point at height el . Both the continuous and the dotted vertical lines in the
picture are induced by the leaves of μ that cross R1 and that spiral along γ1 . This spiraling is
obtained by iterating, under the map z
→ e−l z, the right hand block delimited by the geodesics
that are based at the abscissas x and x − y. This block is made out of the three ideal triangles
suggested by the dotted lines. The (Euclidean) distance y is equal to 3.
slope = 1/sinh t
Figure 36. This is a useful ingredient for computing lg t (α), lg t (β1 ) and lg t (β2 ). In this figure, the
vertical line is a geodesic in the upper half-space model of H2 , and the oblique line is a hypercycle
(a line consisting of points at hyperbolic distance t from the geodesic line). The value t is equal
to the (hyperbolic) length of any arc of circle joining these two lines perpendicularly. The slope
of the oblique line is then equal to 1/sinh t.
e lt
q
lt
r
0 (x − 3)/2 x/2 x − 21 x
Figure 37. One of the four quadrilaterals of Figure 34, drawn in the upper half-space. The
segments labeled p, q, r are the segments induced on that quadrilateral by the geodesics α, β1
and β2 respectively. The segment p is only contained in α, the segment q is only contained in
β2 , whereas the segments r is contained in β1 and β2 . The computation of the lengths of these
segments uses the coordinates that are indicated here and the ingredient in Figure 36. The labels
(x − 3)/2, x/2 and x − 21 are the abscissas of the corners in dotted lines that are above these
labels.
For the segment q, the slope is 2

x−3 , therefore the length is

−1 x−3 3e−l/2
2 sinh = 2 sinh−1 .
2 4 ( l/2)
For the segment r, the slope is 2, therefore

√
−1 1 1+ 5
2 sinh = 2 log .
2 2
Now we stretch the structure g along μ. Recall that g t is the hyperbolic structure
defined by the equality Fμ (g t ) = et Fμ (g). The distances between the ideal triangles
of S \μ are uniformly stretched. All the remarks that we made about the quadrilaterals
R1 , . . . , R4 with respect to the metric g are valid for g t . The lengths of the bound-
ary sides of these quadrilaterals depend on t, and again, using the invariance of the
transverse measure of Fμ (g t ), we have
lg t (γ1 ) = lg t (γ2 ).
We set
lt = lg t (γ1 ) = lg t (γ2 ).
Note that lt = et lg (γ1 ) = et lg (γ2 ), since γ1 and γ2 are in the stump of μ.
γ2
γ1
d
β1
b
c a
β2 a
γ2
b d
α
c
c a
d
b
γ2
γ1
Figure 38. The universal covering of the structure g. In this picture, we can see the annuli of the
horocyclic foliation by examining the non-foliated triangular regions of the ideal triangulation.
Lifts of α, β1 and β2 are also represented.
The curve α consists of four segments of the type labeled by p in Figure 37. The
curve β1 is composed of two segments of the type r and two segments of the type q,
and the curve β2 is composed of four segments of the type r. From this we deduce
the following:

−1 3
lg t (α) = 8 sinh ,
4 sinh(lt /2)

√
−1 3e−lt /2 5
lgt (β1 ) = 4 sinh + 4 log 1 +
4 sinh(lt /2) 2
and √

1+ 5
lgt (β2 ) = 8 log .
2
Summing up, if t
→ g t is the stretch ray directed by μ and starting at g, we have
t
(1) limt→∞ lg t (α) = 0. More precisely, we have lg t ∼t→∞ 12e−lt /2 = 12e−e l/2 ;
(2) {lg t (β1 ) | t ≥ 0} and {lg t (β2 ) | t ≥ 0} are bounded subsets of R∗+ . In fact,
√
lg t (β2 ) is constant and lg t (β1 ) decreases towards 4 log 1+2 5 .
Now if t
→ g t is the anti-stretch ray directed by μ and starting at g, we have
(1) limt→∞ lg t (α) = ∞;
(2) limt→∞ lg t (γi ) = 0 for i = 1, 2;

(3) {lg t (β2 ) | t ≤ 0} is a bounded subset of R∗+ (and in fact it is constant);
(4) limt→∞ lg t (β1 ) = ∞.
From these computations, one gets the feeling that the following facts hold:
– Changing the orientation of a stretch line seems to interchange the roles of the
stump and of the horocyclic measured geodesic lamination.
– The length of any simple closed geodesic with empty intersection with the stump
and with the horocyclic measured geodesic lamination is almost constant along a
stretch line.
– The length of a simple closed geodesic intersecting the horocyclic lamination
(respectively the stump) tends to infinity under the stretch ray (respectively anti-stretch
ray).
The results stated in Section 3.12 below (Theorems 3.34 and 3.35) show that this
is indeed the case, and not only for lengths of simple closed geodesics, but more
generally for lengths of arbitrary measured geodesic laminations.
Figure 39 gives another view on the dynamical behaviour of the stretch line con-
sidered in the example that we are studying in this section. In that figure, we have
Figure 39. Each of these three pictures represents the universal covering H2 of the surface S2,0
studied in Section 3.11 successively equipped with the structures g −t , g = g 0 and g t for some
t > 0. We have represented a part of the preimage of γ1 ∪γ2 by geodesics in bold lines, as well as
the preimage of μ that decomposes in a neat way H2 into ideal triangles. In each ideal triangle,
we have represented the small non-foliated triangular region, which permits the visualization
of the horocyclic foliation. In this way, we can see that the width of the cylinder C increases
whereas its circumference decreases.
drawn the Poincaré disk realized as the universal covering of the surface S equipped
respectively with hyperbolic structures f , g and h that belong to a stretch line directed
by μ, and appearing in that order. The central disk represents the structure g that we
started with, the left-hand disk corresponds to a structure f that lies before g on the
stretch line (that is, we can obtain g by stretching f along the same complete lamina-
tion μ) and the right-hand figure represents the structure h obtained by stretching g
along μ.
From this example, we deduce the following result which says that Thurston’s
asymmetric metric and its dual are not Lipschitz-equivalent.
Proposition 3.33. There does not exist any constant C satisfying K(g, h) ≤ CK(h, g)
for all g and h in T .
Proof. We consider the stretch line studied above, directed by μ and starting at g, and
the closed geodesic α. Then we have, for every t ≥ 0,
lg (α) et l
rg t ,g (α) ∼t→∞ t = A(g)e 2
− e2 l
12e
where A(g) is a constant that depends on the metric g. Thus we obtain K(g t , g) ≥
log rg t ,g (α) ∼ et , whereas K(g, g t ) = t. This proves the proposition. 2
We can use the explicit formulae that we produced in this example to plot the graphs
of the functions t
→ rg,g t (λ) for λ ∈ {α, β1 , β2 , γ1 , γ2 }. These graphs give us an idea
of the behaviour of the hyperbolic surfaces g t as t varies. It is also interesting to draw
the graph of the function t
→ rg,g t (δ) where δ is the closed geodesic represented
in Figure 40, because the intersection pattern of δ with α satisfies δ ∩ α = ∅ and
δ ∩ γ = ∅. Figures 41 and 42 represent respectively the intersection of δ with the
δ
Figure 40
cylinder R of Figures 34 and 37.

The computations give
l 2
elt + 16
1
e t − 5 − (elt − 5)2 + 16elt
lg t (δ) = log 2
elt /2 1 + 161
elt + 3 − (elt − 5)2 + 16elt
Therefore,
lim lg t (δ) = ∞
t→∞
and √
lim lg t (δ) = 4 log(1 + 2).
t→−∞
γ1
δ
s a
R1
a
t
b R2 c
γ2 u d
d
R3
v c
c R4
s a
d
b
γ1
b
Figure 41
e lt
1
0
0
1
0
1 elt /2
0
1
lt
0
1
0
1
0
1
0
1
1 0
1
0 x
Figure 42
We have drawn below the graphs of the functions t

→ rg,g t (λ) for λ ∈ {α, β1 , δ}.
It is interesting to note that the length function t
→ lg t (α) of the horocyclic mea-
sured geodesic lamination is strictly decreasing and convex. We can see on the other
figures that in the general case, the length function along a stretch line is neither mono-
tonic nor convex. But the figures indicate that the function might be peakless in the
sense of Busemann. We recall that a function f defined on a closed interval I ⊂ R
is said to be peakless if I can be decomposed into three subintervals Il , I0 and Ir
whose interiors are pairwise disjoint and following each other in the order indicated,
such that f is strictly decreasing on Il , constant on I0 and strictly increasing on Ir .
∞ ∞
−∞ ∞
−∞ 0 ∞ −∞
Figure 43. t
→ lg t (α). Figure 44. t
→ log(rg,g t (α)).
∞ ∞
−∞ 0 ∞
−∞ 0 −∞
Figure 45. t
→ lg t (β1 ). Figure 46. t
→ log(rg,g t (β1 )).
0.5
−∞ 0 ∞ −2 −1 0 1 2 3
Figure 47. t
→ lg t (δ). Figure 48. t
→ log(rg,g t (δ)).
(Some of the intervals Il , I0 and Ir might be empty). In the case where I0 is empty, the
function f is said to be strictly peakless. This notion was introduced by Busemann
as a generalization of convexity (see [13], p. 109). Regarding the closed geodesics γ1
and γ2 , which are in the stump, we recall that their g t -lengths (which are equal) are
strictly convex and increasing.
In the sections that follow, we shall study the asymptotic behaviour of general
stretch and anti-stretch lines in Teichmüller space.
3.12 The behaviour of the lengths of measured geodesic laminations

along stretch and anti-stretch lines
In this section, we present some results that show that the estimates made in the
examples studied in Sections 3.9, 3.10 and 3.11 about the lengths of measured geodesic
laminations along stretch and anti-stretch lines, are valid in a general setting. We shall
give estimates on the behaviour of lengths of geodesic laminations under stretch and
anti-stretch rays.
The proofs are contained in [64] and [63].
In what follows, we shall use the term topological measured geodesic lamination
to denote a geodesic lamination that admits a transverse measure in the usual sense,
but which is not equipped with any particular such transverse measure. In other words,
a topological measured geodesic lamination is the support of some measured geodesic
lamination.
We first deal with the case of a stretch ray {g t | t ≥ 0}, and then we shall consider
the case of an anti-stretch ray {g t | t ≥ 0}. Our aim is to determine, in each case, the
limit, if it exists, of the length lg t (α) of a measured geodesic lamination α, as t → ∞.
Theorem 3.34 (Théret [64]). Let g be a hyperbolic structure on S, let μ be a complete

geodesic lamination, let {g t | t ≥ 0} be a stretch ray directed by μ and starting at
g = g 0 and let λ = λμ (g) be the associated horocyclic measured geodesic lamination.
Let α be an arbitrary compactly supported measured geodesic lamination on S. We
have the following:
(1) If the support of α is contained in the support of λ, then limt→∞ lg t (α) = 0.
(2) If i(α, λ) = 0, then limt→∞ lg t (α) = ∞.
(3) If the support of α is disjoint from the support of λ, then {lg t (α) | t ≥ 0} is a
bounded subset of R∗+ .
The next result concerns the limits of the same quantities, but this time along
anti-stretch rays.
Theorem 3.35 (Théret [63]). Let μ a complete geodesic lamination on S and let
{g t | t ≥ 0} be an anti-stretch ray starting at a hyperbolic structure g = g 0 and
directed by μ. Assume the stump of μ is nonempty, and call it γ . Let α be an arbitrary
compactly supported measured geodesic lamination on S. We have the following:
(1) If the support of α is contained in the support of γ , then limt→∞ lg t (α) = 0.

(2) If i(α, γ ) = 0, then limt→∞ lg t (α) = ∞. Note that in order to talk about i(α, γ ),
one has to put a transverse measure on γ . The condition i(α, γ ) = 0 is then
independent of the choice of the transverse measure on γ .
(3) If the support of α is disjoint from the support of γ , then {lg t (α) | t ≥ 0} is a
bounded subset of R∗+ .
These two theorems suggest that the two measured geodesic laminations γ and λ
play in some cases symmetric roles.
The two theorems show in particular that the length of a measured geodesic lami-
nation that is disjoint from γ and from λ remains bounded along a stretch line.
A consequence of this fact is that subsurfaces of S that have empty intersection
with the stump and the horocyclic measured geodesic lamination are distorted by a
uniformly bounded amount as one follows a stretch line. Let us be more precise.
Consider a stretch line directed by a complete geodesic lamination μ whose stump γ
is non-empty. Let λ be the support of any horocyclic measured geodesic lamination
associated to this stretch line. Now assume that there exists a subsurface S of S with
nonempty boundary satisfying the following conditions:
• the Euler characteristic of S is negative;
• S contains at least one simple closed curve which is essential in S;
• S ∩ λ = S ∩ γ = ∅.
Note that these requirements imply that any hyperbolic structure on S induces a
hyperbolic structure with geodesic boundary on S and that S , equipped with such a
structure, is isometrically embedded in S.
Let us say that two hyperbolic structures g and g on S are K-quasi-isometric if
there exists a homeomorphism f of S sending g to g , which is isotopic to the identity
of S , and a number K > 0 such that L(f ) ≤ K and L(f −1 ) ≤ K.
If g and g on S are K-quasi-isometric, then the ratios rg,g (α) are bounded from
above by K and from below by 1/K. Conversely, one can show that if the ratios
rg,g (α), as α varies over the set of essential simple closed curves of S , are bounded,
then there exists some K for which the two structures g and g are K-quasi-isometric.
Therefore, we have the following:
Corollary 3.36 (Théret [63]). Let t

→ g t be a stretch line directed by μ, with stump γ .
Assume that there exists a subsurface S of S as above. Then there exists a positive
number K such that for every t and t in R, the hyperbolic structures induced on S

by g t and g t are K-quasi-isometric.
3.13 Thurston’s boundary

Teichmüller space T (S) is embedded in the function space RS+ by the length functional
l : T → RS+ defined in (3.2) above. This embedding, composed with the natural map
from RS+ onto its projectivization P RS+ , gives an embedding (see [20])
T (S) → P RS+ . (3.10)
We also recall that the intersection number functional associates to each λ ∈
ML(S) the element i(λ, ·) ∈ RS+ , and that this defines a map from ML(S) into
RS+ . At the level of projectivizations, we obtain a map
PL(S) → P RS+
which restricts to an embedding
PL0 (S) → P RS+ , (3.11)
which induces a topology on the space PL0 (S), by restriction of the projectivization
of the weak topology on P RS+ .
We have the following result of Thurston (cf. [20], where the result is described
using measured foliations instead of measured laminations).
Theorem 3.37 (Thurston). The images of the spaces T (S) and PL0 (S) in P RS+ by
the embeddings (3.10) and (3.11) are disjoint. With the space P RS+ being equipped
with the quotient of the weak topology, the closure of the image of T (S) in P RS+ is
compact, and the complement of this image in the closure coincides with the image
of PL0 (S). Equipped with the induced topology, the union T (S) = T (S) ∪ PL0 (S)
is homeomorphic to a closed ball of dimension 6g − 6 + 2n, whose boundary is the
image of PL0 (S).
For this reason, the space PL0 (S) is also called Thurston’s boundary of Teichmüller
space.
The following convergence criterion is useful:
Let λ be an element of PL0 (S). A sequence (gn ) in T (S) converges to the projective
class [λ] ∈ PL0 (S) if and only if there exists a sequence xn ∈ R∗+ such that for all
α ∈ S, we have
lim xn lgn (α) = i(λ, α).
n→∞
3.14 Converging to Thurston’s boundary

Questions about the convergence of stretch and anti-stretch rays were already consid-
ered in [52] where the following is proved:
Theorem 3.38 (Papadopoulos [52], Theorem 5.1 and the remark following it). Let g
be an element of Teichmüller space T (S) and let μ be a complete geodesic lamination
in S. Then the stretch ray directed by μ and starting at g converges to the projective
class of the associated horocyclic measured geodesic lamination λμ (g), considered

as an element of Thurston’s boundary PL0 (S) of T (S).
We shall say that a measured geodesic lamination μ is uniquely ergodic if as a

topological lamination, μ admits a unique transverse measure up to a scalar multiple.
In particular, if μ consists of a weighted simple closed geodesic, then it is uniquely
ergodic. A uniquely ergodic measured geodesic lamination is minimal in the sense
that every leaf is dense in its support.
Note that our definitions of unique ergodicity for measured geodesic laminations
and for measured foliations (Definition 2.8 above) do not coincide under the natural
correspondence between the spaces ML0 and MF 0 . (This is so because of the case
of foliations which contain cylindrical components.)
As in the case of measured foliations, one can talk about uniquely ergodic projective
classes of measured laminations.
In the paper [52], it is shown (Proposition 5.2) that if the complete geodesic lam-
ination μ is measured and uniquely ergodic, then the anti-stretch ray directed by μ
converges to the projective class of μ. A more general result has been obtained by
Théret in [63], where the following is proved:
Theorem 3.39 (Théret [63]). Let μ be a complete geodesic lamination on S whose

stump γ is nonempty and is uniquely ergodic. Then any anti-stretch ray directed by μ
converges to the projective class of γ .
Note that there are instances where the anti-stretch line converges to a point in
Teichmüller space (and not on its boundary). This occurs for any anti-stretch line
directed by a complete geodesic lamination μ which has finitely many leaves, and
where all of these leaves converge to punctures at both ends, and it occurs only for
such complete geodesic laminations. (In other words, μ is an ideal triangulation.) Note
that μ is an ideal triangulation if and only if the stump of μ is empty. The limiting
hyperbolic structure g is the one for which all the distances between the distinguished
points on that lamination are zero, that is, λμ (g) = ∅.
To state the next result, we say that a measured geodesic lamination μ is totally
transverse to the complete geodesic lamination μ if the measure-equivalence class of
a measured foliation representing μ is totally transverse to μ in the sense of Defini-
tion 3.9.
One application of Theorems 3.38 and 3.39 is the following
Corollary 3.40. Let α and β be two points in the boundary PL0 (S) of T (S) that can
be represented by totally transverse measured geodesic laminations, and suppose that
α is uniquely ergodic. Then there exists a line in Teichmüller space which is geodesic
for Thurston’s asymmetric metric and which converges to α in the negative direction
and to β in the positive direction.
In general, this geodesic is not unique: it suffices to consider a uniquely ergodic

stump which possesses several completions (see the discussion in Section 3.3). The
stretch lines that are directed by these various complete geodesic laminations, passing
through hyperbolic structures which have the same associated horocyclic measured
geodesic laminations, converge in the negative direction and in the positive direction to
the same points, α and β respectively, in the boundary PL0 (S) of T (S). (To see that it
is possible to find such hyperbolic structures, one can use Thurston’s parametrization
φμ of Teichmüller space described in Section 3.5 above.)
It is possible to permute the points α and β, obtaining geodesics that converge to
α in the positive direction and to β in the negative direction (at least if β is uniquely
ergodic). Note that a priori these geodesics will be distinct from the preceding ones.
3.15 Stretch lines and earthquakes

There is no attempt here to present any of the basic results on earthquakes. We consider
earthquakes only in their relation to stretch maps. We start by recalling the definition.
Definition 3.41 (Fenchel–Nielsen flow). This is a flow on Teichmüller space which

is defined as follows. Let S be a hyperbolic surface and let α be a homotopy class of
an essential simple closed curve on S. The normalized left Fenchel–Nielsen flowline
through S is the one-parameter family (Eαt (S))t∈R of hyperbolic surfaces such that for
each t ≥ 0 (respectively t ≤ 0), Eαt (S) is the hyperbolic surface obtained by cutting
the hyperbolic surface S along the closed geodesic in the class α and gluing back the
two boundary geodesics of the resulting surface after a left (respectively right) twist
of magnitude |t|lg (α). We say that the hyperbolic surface Eαt (S) is obtained from S
by a time-t normalized Fenchel–Nielsen twist along α.
We note that this “normalized” parametrization of the Fenchel–Nielsen flow is

different from the parametrization used by Kerckhoff in [30]; it is the one used in
Papadopoulos [50], because it admits a non-trivial extension to Thurston’s boundary.
Definition 3.42 (Earthquake flow, see [30] and [66]). Let S be a hyperbolic surface,
let γ be a compactly supported measured geodesic lamination on S and let αn be a
sequence of elements in R∗+ × S converging to γ in the topology of ML0 (S). Then,
for each t ∈ R, the sequence of hyperbolic structures Eαt n (S) converges to a hyperbolic
structure Eγt (S) that does not depend on the choice of the sequence αn converging to γ .
We say that the hyperbolic structure Eγt (S) is obtained from S by a time-t normalized
earthquake along γ . The earthquake is said to be a left (respectively right) earthquake
if t ≥ 0 (respectively t ≤ 0).
Theorem 3.43 (Earthquake and stretch commute, Théret [63]). Let μ be a complete
geodesic lamination on S and let γ be a sublamination of μ equipped with a transverse
measure. Let Eγt denote as above the normalized earthquake flow along γ and let
Sμt denote the stretch flow directed by μ. Then, for every t and s in R, we have
Eγt Sμs = Sμs Eγt .
In the paper [52], there is a definition and a study of the extension of the normalized
earthquake flow to Thurston’s boundary PL0 (S) which is a quotient flow of a flow
defined on the unprojectivized space ML0 (S). In fact, in the paper [52], in the case
where S is closed, these flows are defined on the space MF 0 (S) of equivalence
classes of measured foliation, and on the space PMF 0 (S) of projective classes of
measured foliations on S. We can define the flows on ML0 (S) and PL0 (S) by using
the natural correspondence between foliations and laminations. As an application of
Theorem 3.43, we have the following:
Corollary 3.44 (Théret [63]). Assume that the surface S is closed. Let γ be a measured
geodesic lamination on S and let μ be a completion of γ with stump γ . Then the time-t
normalized earthquake (defined on ML0 (S)) along γ of the horocyclic geodesic
lamination λμ (S) is the horocyclic geodesic lamination λμ (Etα (S)). In other words,
we have, for all t ∈ R, λμ (Etα (S)) = Etα (λμ (S)).
Theorem 3.43 on the commutativity of the earthquake flow along components of

the stump with the stretch flow (along completions of the stump) can provide a class
of examples of converging anti-stretch lines. Indeed, as soon as we know that some
anti-stretch line directed by a complete geodesic lamination μ converges to a point
on Thurston’s boundary, all the other anti-stretch lines obtained by the action of an
earthquake along a component of the stump also converge to the same point.
4 Problems
In this last section, we have collected a few problems, which concern mainly Thurston’s
asymmetric metric. Some of them may be easy.
Problem I. On the non-symmetry of Thurston’s metric. There are several natural
questions that arise directly from the fact that Thurston’s metric K is non-symmetric.
For instance: characterize the pairs of hyperbolic structures g and h that satisfy
K(g, h) = K(h, g). In other words, study the locus in T (S) × T (S) defined by
the equation K(g, h) = K(h, g). Since K is a geodesic metric, another natural
question is: give necessary and/or sufficient conditions on the hyperbolic surfaces g
and h under which there exists a geodesic segment from g to h which is also (up to
reparametrization) a geodesic segment from h to g.
Problem II. On the symmetrization of Thurston’s asymmetric metric. There are
several definitions for the symmetrization of an asymmetric metric K, none of them
being more natural than the others. Two such options are
σ K(g, h) = max{K(g, h), K(h, g)}
and
1
SK(g, h) = (K(g, h) + K(h, g)).
2
Each of these symmetrizations is a genuine metric. In the case where K is Thurston’s

asymmetric metric, it is natural to try to relate these two associated symmetrizations
to other known metrics on that space.
In this respect, we mention that one can adapt Thurston’s definition of his asym-
metric metric to the context of the Teichmüller space T1,0 of S1,0 , that is, of the torus.
T1,0 is the space of flat metrics on the torus up to isotopy and homothety. A priori, we
have two distinct asymmetric metrics, L and K, defined as in (3.4) and (3.5) above,
on T1,0 , hyperbolic length being replaced here by Euclidean length, with a suitable
normalization that takes care of homothety. Such a study has been carried out in the
paper [7], where it is shown that K = L, as in the case of surfaces of negative Euler
characteristic, but with different techniques of proof. Recall that there is a natural
identification between T1,0 and H2 . With this identification, we obtain a nonseparat-
ing and nonsymmetric metric δ = K = L on H2 . An explicit formula for δ is given
in [7], and it is shown there that its symmetrization Sδ is the Poincaré metric of H2 ,
which, as is well known, is also the Teichmüller metric on T1,0 .
It is unlikely that for surfaces of negative Euler characteristic, some particular
symmetrization of Thurston’s asymmetric metric is Teichmüller’s metric, but one can
ask the reverse question, that is, to find an interesting asymmetric metric on Teichmüller
space whose symmetrization is Teichmüller’s metric. In fact, one can ask the same
question for the other known metrics on Teichmüller space.
Problem III. Comparing Thurston’s asymmetric metric with other metrics on
Teichmüller space. In a naive approach, one can think that Thurston’s asymmetric
metric is very different from Teichmüller’s metric, because the former is defined
using hyperbolic geometry, whereas the definition of the latter is based on complex
analysis. But, as is well known, there are many tools that make the relation between
conformal and hyperbolic geometry. One of the basic tools is a result of Wolpert ([72],
p. 326) stating that given any two hyperbolic structures g and h on S, then, for any
quasiconformal homeomorphism f : g → h with quasiconformal dilatation K(f )
and for any homotopy class γ of essential simple closed curves on S, we have
lh (f (γ )) ≤ K(f )lg (γ ).
This gives the following inequality between Thurston’s asymmetric metric L and the
Teichmüller metric:
L(g, h) ≤ 2dT (g, h).
Indeed, let f : (S, g) → (S, h) be the Teichmüller map, where g and h are considered
as conformal structures. Wolpert’s result implies
lh (γ )
≤ K(f ) = e2dT (g,h)
lg (γ )
which gives
lh (γ )
log ≤ log K(f ) = 2dT (g, h),
lg (γ )
hence L(g, h) ≤ 2dT (g, h).

We already mentioned a result by L. X. Liu (see [37]) showing that there is no
Lipschitz comparison in the reverse sense.
In this respect, we note that in [7], it is shown that in the case of the Teichmüller
space of the torus, Thurston’s asymmetric metric and Teichmüller’s metric are not
quasi-isometric. (In that paper, there is a discussion of the notion of quasi-isometry
in the context of non-symmetric metrics.)
It may also be possible to relate Thurston’s asymmetric metric to the Weil–Petersson
metric on Teichmüller space T (S). This question is motivated by a characterization
due to Thurston and Wolpert of the Weil–Petersson metric that is based, like Thurston’s
asymmetric metric, on the length-spectrum of closed curves. Indeed, Thurston intro-
duced a Riemannian metric on T (S) where the scalar product of two tangent vectors at
some hyperbolic surface (considered as an element of T (S)) is defined as the second
derivative with respect to the earthquake flows along these vectors of the length of a
uniformly distributed sequence of closed geodesics on the given hyperbolic surface,
and Wolpert showed that this metric coincides with the Weil–Petersson metric; see
[73]. Note that the Weil–Petersson geodesics that consist in pinching a closed curve to
a point have a certain resemblance to anti-stretch lines, without the parametrization.
We note that if dqi is the metric introduced in [70] by Thurston on Teichmüller space,
whose definition we recalled in (3.1) above, and if dT denotes as before Teichmüller’s
metric, then there exists a constant C > 0 such that
dT ≤ dqi ≤ CdT
(see [70], p. 268, where Thurston attributes this result to Douady and Earle).
We note finally that it is easy to make definitions of (symmetric or asymmetric)
metrics on Teichmüller space that are based on the comparison of lengths of closed
geodesics between hyperbolic surfaces, but the interesting goal is to define metrics
that have nice geometrical properties. For instance, one can take any finite collection
{γ1 , . . . , γn } of simple closed curves on the surface whose lengths for an arbitrary
hyperbolic metric completely determine that metric, and using these curves, one can
define a distance between two hyperbolic metrics g and h by taking

lg (γi ) lh (γi )
log sup , .
i=1,...,n lh (γi ) lg (γi )
Clearly, this defines a metric on Teichmüller space, but does it have interesting geo-
metric properties?
Problem IV. Isometries. Isometries between spaces equipped with asymmetric met-
rics can be defined in the same way as between usual metric spaces, except that one
has to be careful about the order of the variables in the case of an asymmetric met-
ric. A well-known result of H. L. Royden [55] states that the group of isometries
of the Teichmüller metric is the mapping class group. Masur and Wolf proved an
analogous result for the Weil–Petersson metric, see [45]. Is the same statement true
for Thurston’s asymmetric metric? It is easy to see that the elements of the mapping
class group are isometries for Thurston’s metric. Therefore, the question is about the
reverse inclusion.
We also recall that Bers obtained in [11] a classification of the isometries of the
Teichmüller metric in terms of the properties of the displacement function and of the
intersection pattern of the minimal displacement sets of the isometries. (Bers’ classi-
fication is based on Thurston’s classification of mapping classes into pseudo-Anosov,
reducible and parabolic, but it also constitutes an independent approach to the classi-
fication.) Likewise, Daskalopoulos and Wentworth obtained in [17] a classification of
isometries of the Weil–Petersson metric, again in terms of the displacement function
and the displacement sets. This result is also described in Chapter 1 of this Hand-
book, see [18]. It is natural to ask for an analogous classification for the isometries
of Thurston’s asymmetric metric. Note that this is not equivalent to the problem of
showing that the isometry group of Thurston’s asymmetric metric is the mapping class
group.
Problem V. Geodesics. Thurston showed that stretch lines are geodesics for Thurs-
ton’s asymmetric metric and that any two points in Teichmüller space can be joined by
a geodesic path that is a concatenation of stretch segments (see Theorem 8.5 of [67]
or Theorem 3.21 above for a precise statement). However, there are other types of
geodesics. Thus, an interesting problem is to describe an arbitrary geodesic. It seems
unlikely that any geodesic is a limit of a concatenations of stretch segments. Thurston
also proved that a geodesic path joining two points in Teichmüller space is in general
not unique. Therefore, another natural question is to characterize the set of ordered
pairs of points such that the geodesic joining them is unique.
Problem VI. The dual Thurston asymmetric metric. Work out an asymptotic for-
mula linking K(g, h) and K(h, g). To find a precise formula is probably not a reason-
able problem. Along a stretch line, we suspect a formula reminding the collar formula
sinh(aK(g, h)) sinh(bK(h, g)) c with some constants a, b, c depending on the
genus and on the number and punctures of the surface S, and on the “complexity” of
the complete lamination directing the stretch line. We already know that Thurston’s
asymmetric metric and its dual are not Lipschitz equivalent. In fact, there are no con-
stants C1 and C2 such that for all g and h in T , we have K(g, h) ≤ C1 K(h, g) + C2
(Proposition 3.33 above).
Problem VII. Anti-stretch lines. Find conditions on a hyperbolic metric equipped
with a complete geodesic lamination so that the anti-stretch line starting at that hyper-
bolic metric and directed by that lamination is also a stretch line (up to reparametriza-
tion).
Problem VIII. Curvature. Study the various existing notions of curvature of Teich-
müller space equipped with Thurston’s asymmetric metric (Finsler curvature, an asym-
metric version of Gromov hyperbolicity, boundedness of curvature in the sense of
Alexandroff and so on). Note that this metric is not nonpositively curved in the sense
of Busemann since there may be several geodesic segments joining two points.
Problem IX. Convergence of anti-stretch rays. For a given complete geodesic lami-
nation μ with nonempty stump γ , find weaker hypotheses than those of Theorem 3.39
that imply the convergence of an anti-stretch ray directed by μ to the projective class
of γ . The convergence result in Theorem 3.39 may suggest that if the measured
geodesic lamination γ is not uniquely ergodic, then either an anti-stretch ray directed
by μ does not converge to a definite point on Thurston’s boundary, or it converges
to the projective class of the geodesic lamination γ equipped with some transverse
measure which would be a “barycenter” of all transverse measures carried by γ . Thus,
a natural question would be to find conditions under which one of these two cases oc-
curs. One can reason by analogy with Masur’s result regarding Teichmüller’s metric
stated as Theorem 2.42 above, where a whole family of rays corresponding to different
transverse measures on a given foliation converge to a single point, corresponding to
a special transverse measure on that foliation.
Problem X. Convexity. Kerckhoff showed in [30] that the geodesic-length function on

Teichmüller space associated to a simple closed geodesic is convex along earthquake
paths and that the sum of geodesic-length functions associated to a finite family of
geodesics that fills up the surface is strictly convex. Wolpert showed in [74] that
geodesic-length functions are strictly convex along Weil–Petersson geodesics. Such
convexity results have been used by several authors to solve various problems. For
instance, Kerckhoff used the convexity of geodesic-length functions along earthquake
paths to obtain a solution of the Nielsen realization problem [30]. Wolpert, in his
paper [74], used the strict convexity of geodesic-length functions along Weil–Petersson
geodesics to obtain a new proof of the fact that Teichmüller space is Stein, and a new
solution of the Nielsen realization problem. Kerckhoff, in his paper [32] used the
convexity properties that he obtained in [30] to develop his theory of lines of minima
in Teichmüller space, which opened a new geometrical point of view on that space.
Thurston, in his preprint [68], constructed a mapping class group-equivariant spine for
the Teichmüller space of a closed surface based on the convexity of the geodesic-length
function. P. Schmutz further developed such a theory in his paper [56], where he also
used the convexity of the geodesic-length function to study what he called a systole
function on Teichmüller space. In view of all that, it is natural to study convexity
properties of length functions along stretch lines. The graphs that we plotted in 3.11
show that this function is not convex in general, but one can conjecture that it is
peakless in the sense of Busemann. At the infinitesimal level, Thurston’s asymmetric
metric is convex in the sense that it is a Finsler metric, that is, it is defined by a length
structure induced by a norm on each tangent space whose closed ball is a convex body
(which is not strictly convex).
Problem XI. The visual boundary. Describe the visual boundary at any point (or
at some class of points) of Teichmüller space equipped with Thurston’s asymmetric
metric. For that, one needs first to understand all the geodesic rays starting at a
point (cf. Problem V above). Does the visual boundary depend on the choice of the
basepoint? Does the action of the mapping class group extend to the space union its
visual boundary? One can ask similar questions about the visual boundary of the dual
asymmetric metric K ∗ (x, y) = K(y, x).
Problem XII. Stretch maps between general metric spaces. Work out a theory of
stretch maps between general (i.e. not necessarily hyperbolic) metrics on a surface.
This problem is mentioned by Thurston in his paper [67]. A particularly interesting
class of metrics on surfaces is the class of Euclidean metrics with cone singularities.
It is also an interesting problem to study stretch maps between higher-dimensional
manifolds equipped with metrics of constant curvature, or between singular spaces
(graphs, two-complexes and so on).
Problem XIII. Moduli space. Study the behaviour of stretch and anti-stretch lines in
moduli space.
References
Springer-Verlag, Berlin 1980. 116, 146
[2] L. V. Ahlfors, Lectures on quasiconformal mappings. Van Nostrand Math. Stud. 10, D. Van
Nostrand Company, Princeton, N.J., 1966. 128
[3] L. V. Ahlfors, Collected Papers. Vol. 1, 2. Birkhäuser, Basel 1982. 116
[4] L.Ahlfors andA. Beurling, Invariants conformes et problèmes extrémaux. In Den 10. Skan-
dinaviske Matematiker Kongres (København, 26.–30. August 1946), Jul. Gjellerups For-
lag, København 1947, 341–351; reprinted in Collected Works of Arne Beurling, Birkhäuser,
Basel 1989, 135–145. 125, 129
[5] L.V.Ahlfors and L. Sario, Riemann surfaces. Princeton Math. Ser. 26, Princeton University
Press, Princeton, N.J., 1960. 121
[6] R. Baer, Isotopien von Kurven auf orientierbaren, geschlossenen Flächen. J. Reine Angew.
Math. 159 (1928), 101–116. 116
[7] A. Belkhirat, A. Papadopoulos, and M. Troyanov, Thurston’s weak metric on the Teich-
müller space of the torus. Trans. Amer. Math. Soc. 357 (2005), 3311–3324. 197, 198
[8] R. Benedetti and C. Petronio, Lectures on hyperbolic geometry. Universitext, Springer-
Verlag, Berlin 1992. 146
Princeton University Press, Princeton, N.J., 1960, 89–119.
[10] L. Bers, The equivalence of two definitions of quasiconformal mappings. Comment. Math.
Helv. 37 (1962/1963), 148–154. 126
Acta Math. 141 (1978), 73–98. 199
[12] L. Bers, Selected works of Lipman Bers: papers on complex analysis. Edited by Irwin Kra
and Bernard Maskit, Amer. Math. Soc., Providence, R.I., 1998. 116
[13] H. Busemann, The geometry of geodesics. Pure Appl. Math. 6, Academic Press, NewYork,
N.Y. 1955. 191
[14] H. Busemann, Recent synthetic differential geometry. Ergeb. Math. Grenzgeb. 54,
Springer-Verlag, New York 1970. 171
[15] A. J. Casson and S. A. Bleiler, Automorphisms of surfaces after Nielsen and Thurston.
London Math. Soc. Stud. Texts 9, Cambridge University Press, Cambridge 1988. 147,
148
[16] Y-E. Choi, K. Rafi, Comparison Between Teichmüller and Lipschitz Metrics. Preprint,
2005 ;arXiv:math.GT/0510136. 172, 173
[17] G. Daskalopoulos and R. Wentworth, Classification of Weil-Petersson isometries. Amer.
J. Math. 125 (2003), 941–975. 199
[18] G. Daskalopoulos and R. Wentworth, Harmonic maps and Teichmüller theory. In Hand-
book of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing House,
Zurich 2007, 33–109. 131, 199
[19] D. B. A. Epstein, Curves on 2-manifolds and isotopies. Acta Math. 115 (1966), 83–107.
116
[20] A. Fathi, F. Laudenbach, and V. Poénaru (eds.), Travaux de Thurston sur les surfaces.
Astérisque 66–67 (1979). 118, 119, 120, 140, 143, 146, 147, 167, 193
[21] A. Fletcher and V. Markovic, Quasiconformal maps and Teichmüller theory. Oxford Grad.
Texts Math. 11, Oxford University Press, Oxford 2007.
York 1987. 116
[23] H. Grötzsch, Über die Verzerrung bei schlichten nichtkonformen Abbildungen und über
eine damit zusammenhängende Erweiterung des Picardschen Satzes. Ber. Verh. Sächs.
Akad. Wiss. Leipzig 80 (1928), 367–376. 132
[24] F. Herrlich and G. Schmithüsen, On the boundary of Teichmüller disks in Teichmüller and
in Schottky space. In Handbook of Teichmüller theory (A. Papadopoulos, ed.), Volume I,
EMS Publishing House, Zurich 2007, 293–349. 127
[25] J. H. Hubbard, Teichmüller theory and applications to geometry, topology, and dynamics.
Vol. 1, Matrix Editions, Ithaca, N.Y., 2006. 128, 131
221–274. 116
Tokyo 1992. 132
[28] J. A. Jenkins, On the existence of certain general extremal metrics. Ann. of Math. (2) 66
(1957), 440–453. 116, 146
[29] S. P. Kerckhoff, The asymptotic geometry of Teichmüller space. Topology 19 (1980),
23–41. 131
129, 132, 134, 138, 139
[31] S. P. Kerckhoff, Earthquakes are analytic. Comment. Math. Helv. 60 (1985), 17–30. 195,
200
[32] S. P. Kerckhoff, Lines of minima in Teichmüller space. Duke Math. J. 65 (1992), 187–213.
groups. Ann. Acad. Sci. Fenn. Ser. A I 278 (1959), 35. 200
[34] O. Lehto and K. I. Virtanen, Quasiconformal mappings in the plane. Grundlehren Math.
Wiss. 126, 2nd ed., Springer-Verlag, Berlin 1973. 113, 136
[35] A. Lenzhen, Teichmüller geodesics that don’t have a limit in PMF . Preprint, University
of Illinois at Chicago, Chicago 2005. 124, 125
[36] G. Levitt, Foliations and laminations on hyperbolic surfaces. Topology 22 (1983), 119–135.
141
[37] L. X. Liu, On non-quasiisometry of the Teichmüller metric and Thurston’s pseudometric.
Chinese J. Contemp. Math. 20 (1) (1999), 31–36. 148, 150
172, 198
[38] L. X. Liu, On the metrics of length spectrum in Teichmüller space. Chinese J. Contemp.
Math. 22 (1) (2001), 23–34. 170
[39] W. Mangler, Die Klassen von topologischen Abbildungen einer geschlossenen Fläche auf
sich. Math. Z. 44 (1939), 541–554. 116
205–221. 137
[41] H. Masur, Dense geodesics in moduli space. In Riemann surfaces and related topics (State
Univ. New York, Stony Brook, N.Y., 1978), Ann. of Math. Stud. 97, Princeton Univ. Press,
Princeton, N.J., 1981, 417–438. 129
[42] H. Masur, Interval exchange transformations and measured foliations. Ann. of Math. (2)
115 (1982), 169–200. 140
[43] H. Masur, Two boundaries of Teichmüller space. Duke Math. J. 49 (1982), 183–190. 139,
140
[44] H. A. Masur and M. Wolf, Teichmüller space is not Gromov hyperbolic. Ann. Acad. Sci.
Fenn. Ser. A I Math. 20 (1995), 259–267. 142
[45] H. Masur and M. Wolf, The Weil-Petersson isometry group. Geom. Dedicata 93 (2002),
177–190. 198
[46] J. D. McCarthy and A. Papadopoulos, The visual sphere of Teichmüller space and a
theorem of Masur-Wolf. Ann. Acad. Sci. Fenn. Math. A I Math. 24 (1999), 147–154. 141,
142
[47] Y. Minsky, Extremal length estimates and product regions in Teichmüller space. Duke
Math. J. 83 (1996), 249–286.
[48] J. Nielsen, Untersuchungen zur Topologie der geschlossenen zweiseitigen Flächen. Acta
Math. 50 (1927), 189–358.
[49] A. Papadopoulos, Geometric intersection functions and Hamiltonian flows on the space
of measured foliations on a surface. Pacific J. Math. 124 (1986), 375–402.
[50] A. Papadopoulos, L’extension du flot de Fenchel-Nielsen au bord de Thurston de l’espace
de Teichmüller. C. R. Acad. Sci. Paris Sér. I Math. 302 (1986), 325–327. 195
[51] A. Papadopoulos, Sur le bord de Thurston de l’espace de Teichmüller d’une surface non
compacte. Math. Ann. 282 (1988), 353–359.
[52] A. Papadopoulos, On Thurston’s boundary of Teichmüller space and the extension of
earthquakes. Topology Appl. 41 (1991), 147–177. 169, 170, 193, 194, 196
[53] A. Papadopoulos and G. Théret, On the topology defined by Thurston’s asymmetric metric.
Math. Proc. Cambridge Philos. Soc. (2007), to appear. 169, 170
[54] R. C. Penner and J. L. Harer, Combinatorics of train tracks. Ann. of Math. Stud. 125,
Princeton University Press, Princeton, N.J., 1992. 140, 147
[55] H. L. Royden, Automorphisms and isometries of Teichmüller space. In Advances in the

Theory of Riemann Surfaces (Proc. Conf., Stony Brook, N.Y., 1969), Ann. of Math. Stud.
66, Princeton University Press, Princeton, N.J., 1971, 369–383. 198
[56] P. Schmutz Schaller, A cell decomposition of Teichmüller space based on geodesic length
functions. Geom. Funct. Anal. 11 (2001), 142–174. 200
[57] G. Springer, Introduction to Riemann surfaces. Addison-Wesley, Reading, Mass., 1957.
121
[58] K. Strebel, Über quadratische Differentiale mit geschlossenen Trajektorien und extremale
quasikonforme Abbildungen. In Festband zum 70. Geburtstag von Rolf Nevanlinna
(Zürich, 4.–6. November 1965), edited by H. P. Künzi and A. Pfluger, Springer-Verlag,
Berlin 1966, 105–127. 132
1984. 130, 131
[60] O. Teichmüller, Extremale quasikonforme Abbildungen und quadratische Differentiale.
Abh. Preuss. Akad. Wiss. Math.-Nat. Kl. 1939 (1940). 135
[61] O. Teichmüller, Bestimmung der extremalen quasikonformen Abbildungen bei geschlos-
senen orientierten Riemannschen Flächen. Abh. Preuss. Akad. Wiss. Math.-Nat. Kl. 1943
(1943). 135
[62] O. Teichmüller, Gesammelte Abhandlungen. Springer-Verlag, Berlin 1982. 116
[63] G. Théret, On the negative convergence of Thurston’s negative stretch lines towards the
boundary of teichmüller space. Ann. Acad. Sci. Fenn. Math., to appear. 151, 191, 192,
194, 195, 196
[64] G. Théret, On Thurston’s stretch lines in Teichmüller space. Preprint, Université Louis
Pasteur, Strasbourg 2005. 191
[65] W. P. Thurston, The geometry and topology of three-manifolds. Mimeographed notes,
Princeton University, Princeton 1976. 140, 147, 150, 158
[66] W. P. Thurston, Earthquakes in two-dimensional hyperbolic geometry. In Low-dimensional
topology and Kleinian groups (Coventry/Durham, 1984), London Math. Soc. Lecture Note
Ser. 112, Cambridge University Press, Cambridge 1986, 91–112. 195
[67] W. P. Thurston, Minimal stretch maps between hyperbolic surfaces. Preprint 1986. 155,
165, 166, 167, 168, 171, 199, 201
[68] W. P. Thurston, A spine for Teichmüller space. Preprint 1986. 200
[69] W. P. Thurston, On the geometry and dynamics of diffeomorphisms of surfaces. Bull.
Amer. Math. Soc. (N.S.) 19 (1988), 417–431. 118
[70] W. P. Thurston, Three-dimensional geometry and topology. Vol. 1, edited by Silvio Levy,
Princeton Math. Ser. 35, Princeton University Press, Princeton, N.J., 1997. 142, 143, 144,
146, 164, 198
[71] W. A. Veech, Gauss measures for transformations on the space of interval exchange maps.
Ann. of Math. (2) 115 (1982), 201–242. 140
[72] S. Wolpert, The length spectra as moduli for compact Riemann surfaces. Ann. of Math.
(2) 109 (1979), 323–351. 197
[73] S. A. Wolpert, Thurston’s Riemannian metric for Teichmüller space. J. Differential Geom.
23 (1986), 143–174. 198
25 (1987), 275–296. 200
Chapter 3
Surfaces, circles, and solenoids
Robert C. Penner
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3 Punctured surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
4 Coordinates for circle homeomorphisms . . . . . . . . . . . . . . . . . . . 211
5 Coordinates for the solenoid . . . . . . . . . . . . . . . . . . . . . . . . . 214
6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
1 Introduction
The “lambda length” of a pair of disjoint horocycles in upper halfspace centered at
u, v ∈ R of respective diameters c, d is defined to be cd 2
|u − v| and is roughly
the exponential of the hyperbolic distance between them. (See §1 for more preci-
sion.) These invariants can be used to devise coordinates in several different guises:
for the Teichmüller space of punctured surfaces [7]; for the space of cosets of the
Möbius group Möb of real fractional linear transformations in the topological group
of all orientation-preserving homeomorphisms of the circle, which forms a general-
ized universal Teichmüller space [8]; and for the Teichmüller space of the “punctured
solenoid”, which is the punctured analogue introduced in [10] (and defined in §4) of
the space studied by Sullivan [12] to analyze dynamical properties of the mapping
class group actions on the Teichmüller spaces for closed surfaces. In fact in each case,
lambda lengths give coordinates for the “decorated” Teichmüller space rather than the
Teichmüller space. (The respective notions of decoration are defined in §§ 2, 3, 4.)
Furthermore, the manifestation of lambda lengths as coordinates on the decorated
Teichmüller space of the punctured solenoid is the first step of a larger ongoing pro-
gram with Šarić [10] to extend the decorated Teichmüller theory [7]–[9] to the solenoid.
To define the punctured solenoid H as a topological space, for definiteness fix the
“modular” group G = PSL2 (Z) of integral fractional linear transformations, let Ĝ
denote its pro-finite completion (whose definition is recalled in §4), let D denote the
open unit disk in the complex plane, and define H = (D × Ĝ)/G, where γ ∈ G
acts on (z, t) ∈ D × Ĝ by γ (z, t) = (γ z, tγ −1 ). In analogy to the case of punctured
206 Robert C. Penner
surfaces, we may produce appropriate geometric structures on H by taking suitable

quotients (D × Ĝ)/G by other actions of G on D × Ĝ. As a pro-finite completion,
the punctured solenoid itself is defined essentially number theoretically in terms of
finite-index subgroups of the modular group, and aspects of its Teichmüller theory
bear close relation to classical questions in number theory (as mentioned at the end
of §5, which also contains other concluding and speculative remarks).
We take this opportunity to correct Theorem 6.4 from [8]. See the remarks fol-
lowing Theorem 8 for the correction to the universal Teichmüller theory and Propo-
sition 12 for the corresponding affirmative statement for the solenoid. Volume II of
this Handbook contains a chapter on the Teichmüller theory of the solenoid written
by D. Šarić [12].
Acknowledgement. It was discussions with Mahmoud Zeinalian that led to the orig-
inal idea of employing lambda lengths for solenoids, and is a pleasure to thank him,
as well as Bob Guralnick for useful comments on classical number theory. The new
material in §4 on the punctured solenoid is joint work [10] with Dragomir Šarić, who
patiently explained his earlier work [11], and it is also a pleasure to thank him for
many stimulating conversations.
2 Background
Define the Minkowski inner product ·, · on R3 whose quadratic form is given by
x 2 + y 2 − z2 in the usual coordinates. The upper sheet
H = {u = (x, y, z) ∈ R3 : u, u = −1 and z > 0}
of the two-sheeted hyperboloid is isometric to the hyperbolic plane. Indeed, identify-
ing the Poincaré disk D with the open unit disk at height zero about the origin in R3 ,
central projection H → D from (0, 0, −1) ∈ R3 establishes an isometry. Moreover,
the open positive light cone
L+ = {u = (x, y, z) ∈ R3 : u, u = 0 and z > 0}
is identified with the collection of all horocycles in H via the affine duality u →
h(u) = {w ∈ H : w, u = −1}. Identifying the unit circle S 1 with the boundary of
D, the central projection extends continuously to the projection : L+ → S 1 which
maps a horocycle in L+ to its center in S 1 .
Define a “decorated geodesic” to be an unordered pair {h0 , h1 } of horocycles with
distinct centers in the hyperbolic plane, so there is a well-defined geodesic connecting
the centers of h0 and h1 ; the two horocycles may or may not be disjoint, and there is
a well-defined signed hyperbolic distance δ between them (taken to be positive if and
only if h0 ∩ h1 = ∅) as illustrated in the two cases of Figure 1. The lambda√length of
the decorated geodesic {h0 , h1 } is defined to be the transform λ(h0 , h1 ) = 2 exp δ.
Taking this particular transform renders
√ the identification h geometrically natural in
the sense that λ(h(u0 ), h(u1 )) = −u0 , u1 , for u0 , u1 ∈ L+ as one can check.
Chapter 3. Surfaces, circles, and solenoids 207
h(u1 ) h(u1 )
−δ
δ
h(u1 ) h(u1 )
Figure 1. Decorated geodesics.
Three useful lemmas (with computational proofs, which we do not reproduce here)
are as follows:
Lemma 1 ([7], Lemma 2.4). Given three rays r

0 , r
1 , r
2 ⊆ L+ from the origin which
contain linearly independent vectors and given three numbers λ0 , λ1 , λ2 ∈ R>0 ,
there are unique points ui ∈ r
i , for i = 0, 1, 2 so that λ(h(ui ), h(uj )) = λk , where
{i, j, k} = {0, 1, 2}. The points u0 , u1 , u2 depend continuously on λ0 , λ1 , λ2 and on
r
0 , r
1 , r
2 .
Lemma 2 ([7], Lemma 2.3). Given two points u0 , u1 ∈ L+ , which do not lie on
a common ray through the origin, and given two numbers λ0 , λ1 ∈ R>0 , there is a
unique point v ∈ L+ on either side of the plane through the origin containing u0 , u1
satisfying λ(h(v), h(ui )) = λi , for i = 0, 1. The point v depends continuously on
u0 , u1 and on λ0 , λ1 .
Lemma 3 ([7], Proposition 2.8). Suppose that u0 , u1 , u2 ∈ L+ are linearly indepen-

dent, let γ (ui , uj ) denote the geodesic in H with ideal vertices given by the centers of
h(ui ) and h(uj ), for i = j , and define
λi
−λ2i = uj , uk , αi = for {i, j, k} = {0, 1, 2}.
λj λk
√
Then 2αi is the hyperbolic length along the horocycle h(ui ) between γ (ui , uj ) and
γ (ui , uk ), for {i, j, k} = {0, 1, 2}.
Remark 1. Consider an ideal quadrilateral Q in D decorated so as to give four points

in L+ . The edges of Q have well defined lambda lengths, say a, b, c, d in correct
cyclic (clockwise) order about the boundary of Q. Choose a diagonal of Q, where the
diagonal has lambda length e and separates edges with lambda lengths a, b from edges
with lambda lengths c, d. The other diagonal of Q has its lambda length f given by
ef = ac + bd,
and we say that f arises from e by a Ptolemy transformation on the lambda lengths.
To see this, note that the formula for a Ptolemy transformation is independent under
scaling any of the four points in L+ , so we may alter the decoration and assume that
the four points lie in a common horizontal plane. In this plane, the Minkoswki inner
product induces a multiple of the usual Euclidean metric, and the intersection of L+
with this plane is a round circle. The formula for the Ptolemy transformation thus
follows from Ptolemy’s classical formula on Euclidean lengths of quadrilaterals that
inscribe in a circle.
Remark 2. Consider a decorated triangle, say with lambda lengths x, y, z in the cyclic
order about the boundary of the triangle determined by an orientation, and define a
2-form
ω(x, y, z) = dln x ∧ dln y + dln y ∧ dln z + dln z ∧ dln x.
A calculation shows that ω(a, b, e) + ω(c, d, e) = ω(b, c, f ) + ω(d, a, f ), thus
assigning a well-defined Ptolemy-invariant 2-form to an oriented decorated quadrilat-
eral. Regard the Poincaré disk as the open unit disk D in the complex plane in the
√ at infinity, and let
usual way so that the unit circle S 1 is identified with the circle
denote the ideal hyperbolic triangle with vertices +1, −1, − −1 ∈ S 1 as in Figure 2.
− 11
−2 − 23
− 21 3 − 21
− 13 − 31
eT U eU T
− 41 − 41
eU 2 eT 2
eU eT
0
1 = − 01 eI eS
− 01 = − 01
eU −1 eT −1
eU −2 eT −2
1
4
eT −1U −1 eU −1T −1
4 1
1 3
3 1
1 2
2 2 3 1
3 1 2
1
Figure 2
Let denote the group generated by reflections in the sides of , and define the Farey
tesselation τ∗ to be the full -orbit of the frontier of . We refer to geodesics in τ∗
as edges of τ∗ and think of τ∗ itself as a set of edges. The ideal vertices of the edges
of τ∗ are naturally identified with

√ the set Q of all rational numbers including infinity,
where for instance +1, −1, − −1 ∈ S 1 correspond respectively to ∞ = 01 , 0 = 01 ,
1 = 11 as illustrated in Figure 2. Let Q ⊆ S 1 denote the corresponding countable
dense subset of S 1 which we refer to simply as the set of rational points of S 1 . Define
the distinguished oriented edge or doe of the Farey tesselation to be the oriented edge
from 01 to 01 .
The modular group PSL2 = PSL2 (Z) of integral fractional linear transformations
is the subgroup of consisting of compositions of an even number of reflections,
and PSL2 acts simply transitively on the set of orientations on the edges of τ∗ . The
assignment
eA = (doe)A for A ∈ PSL2
establishes a bijection between PSL2 and the set of oriented edges of τ∗ as illustrated
in Figure 2. In particular, the doe of τ∗ is eI , where I denotes the identity of PSL2 .
We adopt the standard notation

0 −1 1 1 1 0
S= , T = , U=
1 0 0 1 1 1
for certain elements of PSL2 , where S is involutive and fixes the unoriented edge of
τ∗ underlying the doe while changing its orientation, and U (respectively T ) is the
parabolic transformation with fixed point 01 (respectively 01 ) which cyclically permutes
the incident edges of τ∗ in the counter-clockwise sense about 01 (respectively the
clockwise sense about 01 ). In fact, U −1 = ST S, T −1 = SU S, and any two of S, T ,
U generate PSL2 .
We shall also require the full Möbius group Möb = PSL2 (R) ⊇ PSL2 (Z) = PSL2
consisting of all real fractional linear transformations.
3 Punctured surfaces
Let F = Fgs denote a fixed smooth surface of genus g with s ≥ 1 punctures, where
2 − 2g − s < 0.
Choose any base-point to determine the fundamental group G of F , and consider
the space Hom (G, Möb) of all discrete and faithful representations ρ : G → Möb so
that no holonomy ρ(γ ) is elliptic for γ ∈ G, and the holonomies around the punctures
of F are parabolic. Define the Teichmüller space
T (F ) = Hom (G, Möb)/ Möb,
where Möb acts by conjugacy.
If ρ ∈ Hom (G, Möb), then D/ρ(G) induces a complete finite-area hyperbolic
structure on F , whose punctures are in one-to-one correspondence with the ρ(G)-
orbits of the set of fixed points of parabolic elements of ρ(G).
Define the decorated Teichmüller space T˜ (F ) → T (F ) of F to be the trivial

Rs>0 -bundle, where the fiber over a point is the set of all s-tuples of horocycles, one
horocycle about each puncture of F , say parametrized by hyperbolic length.
By an arc family in F , we mean the isotopy class of a family of essential arcs
disjointly embedded in F connecting punctures, where no two arcs in a family may
be homotopic
rel punctures. If α is a maximal arc family, so that each component of
F − α is a triangle, then we say that α is an ideal triangulation of F .
Theorem 4 ([7], Theorem 3.1). Fix an ideal triangulation τ of F . Then the assignment
of λ-lengths T˜ (F ) → Rτ>0 is a homeomorphism onto.
Proof. We must describe an inverse to the mapping and thus give the construction of
a decorated hyperbolic structure from an assignment of putative λ-lengths. To this
end, consider the topological universal cover F̃ of F and the lift τ̃ of τ to F̃ ; to each
component arc of τ̃ is associated the lambda length of its projection.
The proof proceeds by induction, and for the basis step, choose any triangle 0
of τ̃ and any ideal triangle 0 in H. The ideal vertices of 0 determine three rays in
L+ , so by Lemma 1, there are three well-defined points in L+ realizing the putative
λ-lengths on the edges of 0 . (In effect, this basis step of normalizing a triangle “kills”
the conjugacy by the Möbius group in the definition of Teichmüller space.) Of course,
the triple of points in L+ corresponds by affine duality to a triple of horocycles, one
centered at each ideal vertex of 0 , i.e., a “decoration” on 0 .
To begin the induction step, consider a triangle 1 adjacent to 0 across an arc
in τ̃ . The two ideal points which 0 and 1 share have been lifted to u, v ∈ L+ in
the basis step, and we let w ∈ L+ denote the lift of the third ideal point of 0 and
consider the plane through the origin determined by u, v. According to Lemma 2,
there is a unique lift z ∈ L+ of the third ideal point of 1 on the side of this plane not
containing the lift of w, where z realizes the putative λ-lengths. Again, u, v, z gives
rise via affine duality to another decorated triangle 1 in H sharing one edge and two
horocycles with 0 .
One continues in this manner serially applying Lemma 2 to produce a collection of
decorated triangles pairwise sharing edges in H, where any two triangles have disjoint
interiors (because of our choice of the side of the plane in Lemma 2). Thus, the
construction gives an injection F̃ → H, and we next show that in fact this mapping is
also a surjection. To this end, note first that the inductive construction has an image
which is open in H by construction. According to Lemma 3, there is some ε > 0 so
that each horocyclic arc inside of each triangle has length at least ε; indeed, there are
only finitely many values for such lengths because the surface is comprised of finitely
many triangles. Thus, each application of the inductive step moves a definite amount
along each horocycle, and it follows that the construction has an image which is closed
as well. It follows from connectivity of H that F̃ → H is surjective, and furthermore,
and can see that τ̃ is a tesselation of H, i.e., a locally finite collection of geodesics
decomposing H into ideal triangles.
Following Poincaré, the hyperbolic symmetry group of this tesselation is the re-
quired (normalized) Fuchsian group G giving a point of Teichmüller space, and the
construction likewise provides a decoration on the quotient H/G as required.
Remark 3. One thinks of the choice of ideal triangulation as a choice of “basis” for
the lambda length coordinates. Formulas for the change of basis are given by Ptolemy
transformations, and this leads [7], [8] to a faithful representation of the mapping class
group of F as well as its universal generalization to Tess, which is defined in the next
section. Furthermore as in Remark 2, the 2-form ω = −2 ω(a, b, c) is invariant
under this action and descends to the Weil–Petersson form on Riemann’s moduli space,
where the sum is over all triangles complementary to any fixed ideal triangulation and
the edges of the triangle have lambda lengths a, b, c in correct cyclic order determined
by an orientation of F . These two ingredients lead to natural quantizations [3], [4] of
Teichmüller space.
4 Coordinates for circle homeomorphisms

Define a tesselation τ of the Poincaré disk D to be a countable locally finite collection
of hyperbolic geodesics in D each of whose complementary regions is an ideal triangle.
A distinguished oriented edge or doe of τ is the specification of an orientation on one
of the geodesics in τ . Each geodesic in τ has a pair of asymptotes in S 1 , and we let
τ 0 ⊆ S 1 denote the collection of all such asymptotes of geodesics in τ and τ denote
2
the collection of all triangles complementary to τ .

Tesselations with doe are “combinatorially rigid” in the following sense. Suppose
that τ1 , τ2 are each a tesselation with doe, say the initial and terminal points of the doe
in τi are xi ∈ τi0 and yi ∈ τi0 , respectively, for i = 1, 2. There is a unique bijection
f : τ10 → τ20 so that f (x1 ) = x2 , f (y1 ) = y2 , and whenever x, y, z in correct cyclic
order span an oriented triangle in τ12 , then f (x), f (y), f (z) in correct cyclic order also
span an oriented triangle in τ22 . This mapping f : τ10 → τ20 is called the characteristic
mapping of the pair of tesselations with doe. In particular, we may fix τ1 = τ∗ to
be the Farey tesselation with doe defined in §1, so τ∗0 = Q. We may thus define the
characteristic mapping fτ : Q → τ 0 of the tesselation τ = τ2 with doe.
Define the set
Tess = {tesselations with doe of D}.
To define a topology on Tess , if τ is a tesselation with doe, then we may extend
the range of the characteristic mapping fτ : Q → τ 0 ⊆ S 1 to S 1 . The assignment
τ → fτ determines an embedding of Tess into the function space (S 1 )Q with the
compact-open topology (where Q is given its discrete topology), and we endow Tess
with the subspace topology.
Define the topological group Homeo+ = Homeo+ (S 1 ) to be the group of all
orientation-preserving homeomorphisms of the circle taken with the compact-open
topology. If f ∈ Homeo+ and e is any geodesic in D, say with ideal points x, y ∈ S 1 ,

then define f (e) to be the geodesic in D with ideal points f (x), f (y) ∈ S 1 . It is not
difficult to see that if τ is a tesselation and f ∈ Homeo+ , then f (τ ) = {f (e) : e ∈ τ }
is also a tesselation. Since a doe on τ determines a doe on f (τ ) in the natural way,
there is thus an action of Homeo+ on Tess.
Theorem 5 ([8], Theorem 2.3). The mapping Homeo+ → Tess given by f → f (τ∗ )
is a homeomorphism onto.
Proof. Injectivity follows from the fact that a homeomorphism is uniquely determined
by its values on a dense set. For surjectivity, consider any tesselation with doe τ .
Using the fact that Q and τ 0 are dense in S 1 and the characteristic mapping fτ is
order-preserving by construction, a standard point-set topology argument show that
there is a unique orientation-preserving homeomorphism fτ : S 1 → S 1 which restricts
to the characteristic mapping. Both mappings f → f (τ ) and τ → fτ are continuous
by construction.
There is the natural diagonal left action of the group Möb on (S 1 )Q , which in-
duces a left action of Möb on the subspace Tess , and we finally define the universal
Teichmüller space
Tess = Möb \Tess ≈ Möb \ Homeo+
to be the orbit space with the quotient topology.
A decoration on a tesselation τ is the specification of horocycles in D, one hororcyle
centered at each point of τ 0 . Via the affine duality discussed in §1, the characteristic
mapping fτ : Q → S 1 on a decorated tesselation τ with doe extends to a mapping
gτ : Q → L+ . The image gτ (Q) is automatically “radially dense” in L+ in the sense
that (gτ (Q)) is a dense subset of S 1 , where : L+ → S 1 is the natural projection.
Define
= {decorated tesselations τ with doe : gτ (Q) is discrete in L+ }.
Tess
The Hausdorff topology on the set of all closed subsets of L+ induces a subspace
topology on the set of all discrete subsets of L+ , and this in turn induces a compact-
. There is again a diagonal left action of Möb by Minkowski
open topology on Tess

isometries on Tess , and the decorated universal Teichmüller space is finally defined
to be the topological quotient
= Möb \Tess
Tess .
→ Tess, which is evidently continuous.
There is the natural forgetful map Tess
Given a decorated tesselation τ̃ with doe and e ∈ τ∗ , there is the corresponding
lambda length of the decorated geodesic in τ̃ with underlying geodesic fτ (e), where
fτ is the characteristic mapping of τ . Thus, lambda lengths naturally determine an
element of Rτ>0
∗
.
Theorem 6 ([8], Theorem 3.1). The assignment of lambda lengths determines an

embedding
→ R τ∗
Tess >0
onto an open set, where Rτ>0∗
is given the weak topology (compact-open on Rτ>0
∗
with

τ∗ discrete). Thus, Tess inherits the structure of a Fréchet manifold.
Proof. We say that an element τ ∈ Tess is normalized provided that {±1} ⊆ τ 0 , the
doe of τ runs from −1 to +1, and the triangle
√ in τ lying
2 to the right of the doe coincides
with the triangle spanned by −1, +1, − −1 ∈ S 1 . Since Möb acts three-effectively
on S 1 and the value of a Möbius transformation at three points of S 1 determines it
uniquely, each Möb-orbit on Tess admits a unique normalized representative. Tess is
thus canonically identified with the collection of all normalized tesselations. (Again,
we have “killed” the Möbius group by normalization.)
To define a left inverse to the assignment λ ∈ Rτ>0
∗
of lambda lengths, use Lemma 1
√
to uniquely lift the vertices ±1, − −1 of the triangle of τ∗ to the right of the doe to
points in the rays in L+ lying over these vertices realizing the lambda lengths. As in
the proof of Theorem 4, we may then uniquely extend using Lemma 2 to a function
g : Q → L+ realizing the lambda lengths.
If g(Q) ⊆ L+ is radially dense, then the order-preserving mapping Q → L+ → S 1
interpolates a unique homeomorphism f : S 1 → S 1 as before. One can always
produce a discrete decoration, say by taking the point f (p) ∈ L+ to have height i in
R3 if p ∈ Q is of Farey generation i.
It follows that the mapping Tess → Rτ∗ is indeed injective. Continuity follows
>0
from the definition of the topologies, and openness of the image follows from the
construction.
Recall [1] that a quasisymmetric homeomorphism of S 1 is the restriction to S 1

of a quasiconformal homeomorphism of D, and let Homeoqs ⊆ Homeo+ denote
the subspace of all quasisymmetric homeomorphisms of the circle. Bers’ universal
Teichmüller space [2] is the quotient
Möb \ Homeoqs ⊆ Möb \ Homeo+ ≈ Tess
and is highly studied.
As is usual in these circumstances, it is difficult to explicitly characterize the image
⊆ Rτ∗ . On the other hand, there are the following useful characterizations of
Tess >0
⊆ R τ∗ .
subsets of Tess >0
We say that λ ∈ Rτ>0 ∗
is pinched provided there is some real number K > 1 so that
K −1 ≤ λ(e) ≤ K for each e ∈ τ∗ .
Theorem 7 ([8], Theorem 6.3). If λ ∈ Rτ>0 ∗

is pinched, then there is a decorated
tesselation whose lambda lengths are given by λ, and the corresponding subset of L+
is discrete and radially dense.
Theorem 8 ([8], Theorem 6.4 (joint with Sullivan)). If λ ∈ Rτ>0

∗
is pinched, then the
corresponding homeomorphism of the circle is quasisymmetric.
In particular, consider any decoration on a marked punctured Riemann surface

F uniformized by G < Möb. Choose an ideal triangulation of D/G, and lift it to
a tesselation τ of D which inherits a G-invariant decoration. Choose a doe of τ to
The lambda lengths in F lift to G-invariant lambda lengths
determine a point of Tess.
on τ , and they are pinched since they take only finitely many values. Furthermore,
if G < PSL2 is finite-index and free of elliptics and φ : D → D is any G-invariant
quasiconformal map conjugating G to an isomorphic group, then we claim that the
boundary values of φ satisfy the smoothness conditions above. To see this, conjugate
in domain and range so that corresponding parabolic covering transformations are each

given in upper halfspace by z → z + 1 (thus, destroying the normalization in Tess),
so the monotone function φ(t) nearly agrees with the integral part of t. It follows
directly that φ(t) is differentiable at each point of Q, and the derivatives at points of
Q are uniformly near unity. (Compare with [13].)
In contrast, a quasisymmetric map φ : S 1 → S 1 arising from pinched lambda
lengths need not have these differentiability properties at Q. To see this, use the
formula in the Introduction for lambda lengths in the upper halfspace model to produce
pinched lambda lengths so that the two one-sided derivatives at infinity disagree.
This corrects the second part of Theorem 6.4 from [8]. For the corresponding result
in the setting of the solenoid, see Proposition 12.
5 Coordinates for the solenoid

Let G < PSL2 be any finite-index subgroup, and choose a base-point in the quotient
surface or orbifold M = D/G; in particular, for G = PSL2 , M is the orbifold modular
curve. Consider the category CM of all finite-sheeted orbifold covers π : F → M,
where F is a punctured Riemann surface. CM is a directed set, where π1 ≤ π2 if there
is a finite-sheeted unbranched cover π2,1 : F2 → F1 of Riemann surfaces so that the
following diagram commutes:
π2,1
F2 A / F1
AA }}
AA }}
π 2 AAA }} π
~}} 1
M
In other words by covering space theory, if i < G < PSL2 uniformizes Fi for
i = 1, 2, then π1 ≤ π2 if and only if 1 is a finite-index subgroup of 2 .
A topological space, the punctured solenoid, is defined in analogy to [12] to be the
inverse limit
HM = lim CM ;
←
a point of HM is thus determined by choices of points yi ∈ Fi for each cover πi : Fi →

M, where the choices are “compatible” in the sense that if π1 ≤ π2 , then we have in
the notation above π2,1 (y2 ) = y1 .
Since punctured surface groups are cofinal in the set of punctured orbifold groups,
we could have equivalently considered the category of orbifold covers of M in the
definition of HM . Furthermore, if < G is of finite-index, then H is naturally
homeomorphic to HG , and we may thus think of the punctured solenoid H = HPSL2
One can from first principals develop the Teichmüller theory of H along classical
lines [10] as has been done for the solenoid of closed surfaces [12], [5], [11]. Instead,
we next introduce an explicit space homeomorphic to H following [5], and we shall
then define the Teichmüller space representation theoretically in analogy to punctured
surfaces in §1.
G has characteristic subgroups

GN = { < G : [ : G] ≤ N },
for each N ≥ 1, and these are nested GN+1 < GN . Define a metric G × G → R by
γ × δ → min{N −1 : γ δ −1 ∈ GN },
and define the pro-finite completion Ĝ of G as a space to be the metric completion of G,
i.e., suitable equivalence classes of Cauchy sequences in G. Termwise multiplication
of Cauchy sequences gives Ĝ the structure of a topological group, and termwise
multiplication by G gives a continuous action of G on Ĝ.
For any sub-group G < PSL2 of finite-index, we may define the quotient
HG = D ×G Ĝ = (D × Ĝ)/G,
where γ ∈ G acts by
γ : D × Ĝ → D × Ĝ
(z, t) → (γ z, tγ −1 ).
Lemma 9 ([5]). H is homeomorphic to HG for any G < PSL2 of finite-index.
In particular, each path component, or “leaf”, of H is homeomorphic to a disk (by

residual finiteness of G), and each leaf is dense in H (since G is dense in Ĝ).
Let us for definiteness simply fix G = PSL2 and consider the collection
Hom (G × Ĝ, Möb) of all functions ρ : G × Ĝ → Möb satisfying the following
three properties.
Property 1. ρ is continuous.
Property 2. For each γ1 , γ2 ∈ G and t ∈ Ĝ, we have
ρ(γ1 γ2 , t) = ρ(γ1 , tγ2−1 ) ρ(γ2 , t).
Property 3. For every t ∈ Ĝ, there is a quasiconformal mapping φt : D → D so that

for every γ ∈ G, the following diagram commutes:
(z,t) →(γ z,tγ −1 )
D × Ĝ / D × Ĝ
φt ×id φtγ −1 ×id

D × Ĝ / D × Ĝ
(φt (z),t) →(φtγ −1 γ (z),tγ −1 )
Furthermore, φt varies continuously in t ∈ Ĝ for the common refinement of the C ∞

topology of uniform convergence on compacta in D and the usual topology on Bers’
universal Teichmüller space Möb \ Homeoqs of the extension of φt to the circle at
infinity.
As to Property 1, notice that since G is discrete, ρ is continuous if and only if it is
so in its second variable only. Property 2 is a kind of homomorphism property of ρ
mixing the leaves; notice in particular that taking γ2 = I shows that ρ(I, t) = I for
all t ∈ Ĝ. Property 3 mandates that for each t ∈ Ĝ, φt conjugates the standard action
of G on D × Ĝ at the top of the diagram to the action
γρ : (z, t) → (ρ(γ , t)z, tγ −1 )
at the bottom, and we let Gρ = {γρ : γ ∈ G} ≈ G. Notice that the action of Gρ on
D × Ĝ extends continuously to an action on S 1 × Ĝ. We finally define the solenoid
(with marked hyperbolic structure)
Hρ = (D ×ρ Ĝ) = (D × Ĝ)/Gρ .
The collection φt , for t ∈ Ĝ, thus induces a homeomorphism H → Hρ .
Define the group Cont(Ĝ, Möb) to be the collection of all continuous maps α : Ĝ →
Möb, where the product of two α, β ∈ Cont(Ĝ, Möb) is taken pointwise (αβ)(t) =
α(t) β(t) in Möb. α ∈ Cont(Ĝ, Möb) acts continuously on ρ ∈ Hom (G × Ĝ, Möb)
according to
(αρ)(γ , t) = α −1 (tγ −1 ) ρ(γ , t) α(t).
Theorem 10 ([10]). There is a natural homeomorphism of the Teichmüller space of

the solenoid H with
T (H ) = Hom (G × Ĝ, Möb)/ Cont(Ĝ, Möb).
Rather than describe the proof here, we shall for simplicity simply take this quotient
as the definition of the Teichmüller space T (H ). Again with an eye towards simplicity
here, rather than define punctures of solenoids intrinsically (as suitable equivalence
classes of ends of escaping rays), we can more simply proceed as follows. Each
φt : D → D extends continuously to a quasisymmetric mapping φt : S 1 → S 1 . We
say that a point (p, t) ∈ S 1 × Ĝ is a ρ-puncture if φt−1 (p) ∈ Q, and a puncture of Hρ

itself is a Gρ -orbit of ρ-punctures. A ρ-horocycle at a ρ-puncture (p, t) is the image
under φt of a horocycle in D centered at φt−1 (p).
A decoration on Hρ , or a “decorated hyperbolic structure” on H , is a function
ρ̃ : G × Ĝ × Q → Möb ×L+ , where
ρ̃(γ , t, q) = ρ(γ , t) × h(t, q)
with ρ(γ , t) ∈ Hom (G × Ĝ, Möb), which satisfies the following conditions.
Property 4. For each t ∈ Ĝ, the image h(t, Q) ⊆ L+ is discrete and radially dense.
Property 5. For each q ∈ Q, the restriction h(·, q) : Ĝ → L+ is continuous.
Property 6. ρ̃ is G-invariant in the sense that if δ ∈ G, then
δ ρ̃(γ , t, q) = ρ̃(δγ , tδ −1 , δq),
where δ acts diagonally δ : (γ , q) → (δγ , δq) on Möb ×L+ with δq the natural action
of G = PSL2 on L+ .
Finally, let Hom (G × Ĝ × Q, Möb ×L+ ) denote the space of all decorated hyper-
bolic structures satisfying the properties above, and define the decorated Teichmüller
space as the quotient
T˜ (H ) = Hom (G × Ĝ × Q, Möb ×L+ )/ Cont(Ĝ, Möb),
where α : Ĝ → Möb acts on ρ̃ by

(α ρ̃)(γ , t, q) = α −1 (tγ −1 ) ρ(γ , t) α(t) × h(t, α(t)q) .

It is clear that forgetting decoration induces a continuous surjection T˜ (H ) → T (H).
There is a natural mapping λ : T˜ (H ) → (Rτ>0 ∗ Ĝ
) which assigns to a function
+
ρ̃ : G × Ĝ × Q → Möb ×L the lambda length for the Gρ metric of the ρ-horocycles
determined by h at the endpoints of the geodesic in Hρ labeled by γ .
Theorem 11 ([10]). The assignment of lambda lengths determines an embedding

T˜ (H ) → Cont(Ĝ, Rτ>0
∗
)
onto an open set, where we take the strong topology on Rτ>0
∗
and on Cont(Ĝ, Rτ>0
∗
).
Sketch of Proof. To prove the mapping is injective, we must again define the construc-
tion of decorated hyperbolic structure from a continuous λt ∈ (Rτ>0 ∗ Ĝ
) . To this end,
begin the definition of ρ̃ = ρ × h on the triangle to the right of the doe in τ∗ with
lambda lengths given by λt . As usual according
√ to Lemma 1, we can uniquely lift to
a triple of points in L+ lying over ±1, − −1.
It is easily seen from the identification of G = PSL2 with the oriented edges of τ∗
that any γ ∈ PSL2 can be written uniquely in the one of the following forms:
i) γ = I ;
ii) γ lies in the free semi-group generated by either U, T or U −1 , T −1 ;
iii) γ arises from either i) or ii) by addition of the prefix S.

To define ρ(γ , t) ∈ Möb, we shall specify an ideal triangle-doe pair ( , e ) in D,
where e is oriented with to its right.
√ There is then a unique ρ ∈ Möb mapping to
the vertices of the vertices ±1, − −1 of the triangle to the right of the doe eI
in τ∗ and mapping eI to e .
Let us write γ ∈ G in one of the forms i-iii) above. Of course, if γ = I , then
ρ(γ , t) = I as follows from the functional equation in Property 2, and we take
( , e ) = (, eI ).
If γ = S, then let us begin with the lambda lengths λt ∈ Rτ>0 ∗
on the edges of
+
and employ Lemma 1 to uniquely realize a lift to L of the vertices of this decoration
on the triangle . On the triangle to the left of the doe, consider the lambda lengths
λtS −1 = λtS . It need not be that λt (eI ) = λtS (eI ), and we re-scale, taking lambda
lengths
λt (eI )
λtS (·)
λtS (eI )
on the edges eU , eT . This defines lambda lengths on the edges of the quadrilateral
√ in
τ∗ triangulated by the doe. Again using Lemma 2, there is a unique lift of −1 to L+
realizing the lambda lengths, and the projection of this point to S 1 is one of the vertices
of . The other two vertices of are ±1, and the doe is e = eS , completing the
definition in this case. Notice that this element of Möb that maps (, eI ) → ( , eS )
is necessarily involutive.
The case of any word in one of the semi-groups in ii) is handled by induction on
the length, where for instance for any such γ that has a prefix U , one begins from
the lambda lengths λt on and uses the re-scaled lambda lengths λtU −1 on the edges
eU , eT ; the edge corresponding to eU is the doe of the first step.
The remaining case iii) of a word from one of the semi-groups with prefix S is
handled in exactly the same manner completing the definition of the construction of
ρ : G × Ĝ → Möb. The functional equation in Property 2 on ρ follows by definition.
Furthermore, since λt ∈ Rτ>0 ∗
depends continuously on t (because of the definition of
the topology), ρ(γ , t) is also continuous in t as required in Property 1; indeed, the
entries of ρ(γ , t) ∈ Möb are algebraic function of finitely many lambda lengths.
As for Property 3 in the definition of T (H ), we claim that for each t ∈ Ĝ, λt ∈ Rτ>0
∗
is pinched if λ ∈ Cont(Ĝ, Rτ>0 ∗

).
To see this, note that the very definition of a continuous function λ : Ĝ → Rτ>0 ∗
means that for all K there exists an N such that we have

λt (e)
1 + K −1 ≤ ≤1+K for all e ∈ τ∗ , γ ∈ GN .
λt (γ e)
Take say K = 21 and its corresponding N . A fundamental domain for GN has only a
finite collection of values of lambda lengths, and any other lambda length is at most
a factor of 3/2 times a lambda length this finite set, and at least a factor of 1/2 times a
lambda length in this finite set. λt is therefore pinched, proving the claim.
It follows that for each t ∈ Ĝ, λt : τ∗ → R>0 is necessarily pinched. By Theorem 8,
there is a corresponding quasiconformal homeomorphism φt : D → D. Commuta-
tivity of the diagram and continuity in Property 3 follow by construction, and this
completes the proof that the function ρ constructed above lies in Hom (G × Ĝ, Möb)
and hence determines a point of T (H ).
To define a decoration on the ρ-punctures, each λt determines a decoration on
τ∗ × {t} ⊆ D × Ĝ, as required. Property 4 is guaranteed by the claim and Theorem 7.
Property 5 holds as before in the strong sense that the Euclidean coordinates of each
h(t, q) are algebraic functions of finitely many lambda lengths, and Property 6 holds
by invariance of lambda lengths under Möb.
Proposition 12. Suppose that λ ∈ Cont(Ĝ, Rτ>0 ∗

). Then for each t ∈ Ĝ, λt : τ∗ →
R>0 corresponds to a quasiconformal homeomorphism φt : D → D whose quasisym-
metric extension φt : S 1 → S 1 is differentiable at each point of Q with derivative
uniformly near unity.
Proof. As above, continuity of λt in t implies that each λt is pinched, which gives

a quasisymmetric map φt , for t ∈ Ĝ, by Theorem 8. Using the upper halfspace
model, normalize φt such that it fixes 0 and ∞, whence the geodesics of τ∗ that
limit to ∞ are mapped by φt onto geodesics which likewise limit to ∞. Again by
continuity, we conclude that for each ε > 0, e ∈ τ∗ and γ ∈ PSL2 fixing ∞, we have
|λt (e) − λt (γ n e)| < ε for n sufficiently large. The differences an = φt (n) − φt (n − 1)
are then ε1 close using continuity of the assignment of decorated ideal triangles given
lambda lengths.
We show that limn→∞ n1 φt (n) exists and is bounded, which proves the proposition.
To this end since |ai − ank+i | < ε1 for all i, k, we find

1
(a + · · · + a ) − 1 (a + · · · + a )
1 n 1 nk
n nk
n
1
≤ 1/n |ai − (ai + ai+n + · · · + ai+n(k−1) )| ≤ ε1 ,
k
i=1
and it follows that 1

n φt (n) is a Cauchy sequence, as desired.
Differentibility at the rational points, which holds in the case of punctured surfaces
thus also holds for the solenoid according to Proposition 12 but does not hold for
general decorated tesselations with pinched lambda lengths however (cf. the discussion
following Theorem 8).
6 Concluding remarks
In addition to the function h : Ĝ × Q → L+ constructed in the proof of Theorem 11,
there is another natural function h1 : Ĝ × Q → L+ defined √ as follows. Begin with
the “standard” decoration on τ∗ where all lambda lengths are 2. By Proposition 12,
each φt is smooth with bounded derivative φt (q) at each point of Q ⊆ S 1 . In an upper
halfspace model of H with the endpoint of the doe at infinity, scale the Euclidean
diameter of the horocycle centered at q ∈ Q by the absolute value of the derivative
|φt (q)| to determine the diameter of the horocycle at φt (q). This defines the function
h1 (t, q). Notice that h1 (t, Q) is again discrete and radially dense in L+ , but there
seems to be no guarantee that the function h1 discussed above satisfies Property 5.
The representation-theoretic treatment of the Teichmüller theory of the punctured
solenoid seems to us quite appealing. For example, the strong topology for the solenoid
(in Theorem 11) in contrast to the weak topology for circle homeomorphisms (in
Theorem 6) is interesting. Furthermore, one can naturally induce stronger or weaker
transverse structures in the Ĝ direction in H by imposing conditions other than con-
tinuity on the lambda length functions in Theorem 11, and we wonder in particular
what is the transverse regularity of h1 .
As was mentioned in the Introduction, Theorem 11 is the first step of an ongoing
program [10] to develop the decorated Teichmüller theory of the punctured solenoid.
Though there is an alternative to the construction of ρ : G × Ĝ → Möb in the proof of
Theorem 11, the argument given here bears a close relation to the treatment of broken
hyperbolic structures in [6].
Lambda lengths enjoy the simple transformation property of Ptolemy transforma-
tions (cf. Remark 1), and furthermore, there is a simple invariant two-form (cf. Re-
mark 2). These ingredients have been used [3], [4] to give a quantization of classical
Teichmüller theory (cf. Remark 3). These same two ingredients persist for the uni-
versal Teichmüller theory as well as for the punctured solenoid and might be used to
quantize these Teichmüller theories as well.
It is an open (but not centrally important) problem in number theory to calculate the
index of GN in G = PSL2 , and an algorithm for its calculation devolves to the “cubic
fatgraph” enumeration problem” for surfaces of fixed Euler characteristic arising as
total spaces of degree-N covers of D/G. More speculatively, there is a natural group
[10] generated by the GN -equivariant Ptolemy moves for some N , which seems closely
related to the completions [9] of the universal Ptolemy group studied in the context of
Grothendieck absolute Galois theory.
References
[1] L. V. Ahlfors, Lectures on quasiconformal mappings. Wadsworth & Brooks/Cole Ad-

vanced Books & Software, Monterey, CA, 1987; reprint of the 1966 original edition.
[2] L. Bers, Universal Teichmüller space. In Analytic methods in mathematical physics (Sym-
pos., Indiana Univ., Bloomington, Ind., 1968), Gordon and Breach, NewYork 1970, 65–83.
[3] V. V. Fock and L. O. Chekhov, A quantum Teichmüller space. Teoret. Mat. Fiz. 120
(1999), 511–528; English transl. transl. Theoret. and Math. Phys. 120 (1999), 1245–1259.
Quantum mapping class group, pentagon relation, and geodesics. Tr. Mat. Inst. Steklova
226 (1999), 163–179; English transl. Proc. Steklov Math. Inst. 226 (1999), 149–163.
[4] R. M. Kashaev, Quantization of Teichmüller spaces and the quantum dilogarithm. Lett.
Math. Phys. 43 (2) (1998), 105–115.
[5] C. Odden, Virtual automorphism group of the fundamental group of a closed surface.
Ph.D. Thesis, Duke University, 1997; The baseleaf preserving mapping class group of the
universal hyperbolic solenoid. Trans. Amer. Math. Soc. 357 (2004), 1829–1858.
[6] A. Papdopoulos and R. C. Penner, The Weil–Petersson Kähler form and affine foliations
on surfaces. Ann. Global Anal. Geom. 27 (2005), 53–77.
[7] R. C. Penner, The decorated Teichmüller space of punctured surfaces. Comm. Math. Phys.
113 (1987), 299–339.
[8] R. C. Penner, Universal constructions in Teichmüller theory. Adv. Math. 98 (1993),
143–215.
[9] R. C. Penner, The universal Ptolemy group and its completions. In Geometric galois
actions II (edited by L. Schneps and P. Lochak), London Math. Soc. Lecture Note Ser.
243, Cambridge University Press, Cambridge 1997, 293–312.
[10] R. C. Penner and D. Šarić, Teichmüller theory of the punctured solenoid. Preprint, 2005;
arXiv.math.DS/0508476.
[11] D. Šarić, On quasiconformal deformations of the universal hyperbolic solenoid. Preprint,
2003; www.math.sunysb.edu/~saric/research/solenoid2.pdf.
[12] D. Šarić, The Teichmüller theory of the solenoid. In Handbook of Teichmüller theory
(A. Papadopoulos, ed.), Volume II, EMS Publishing House, Zurich 2008. 206
[13] D. Sullivan, Linking the universalities of Milnor-Thurston, Feigenbaum and Ahlfors-
Bers. In Topological methods in modern mathematics (Stony Brook, NY, 1991), edited by
L. R. Goldberg and A. V. Phillips, Publish or Perish, Inc., Houston, TX, 1993, 543–564.
[14] P. Tukia, Differentiability and rigidity of Möbius groups. Invent. Math. 82 (1985),
557–578.
Chapter 4
About the embedding of Teichmüller space in the

space of geodesic Hölder distributions
Jean-Pierre Otal
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
2 The Teichmüller space of a Fuchsian group . . . . . . . . . . . . . . . . . 224
2.1 The quasiconformal homeomorphisms . . . . . . . . . . . . . . . . . 224
2.2 The complex structure on Teichmüller space . . . . . . . . . . . . . . 226
3 Geodesic currents and Hölder distributions . . . . . . . . . . . . . . . . . 229
3.1 The space of geodesics of H2 . . . . . . . . . . . . . . . . . . . . . . 230
3.2 Geodesic currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
3.3 The Liouville measure . . . . . . . . . . . . . . . . . . . . . . . . . 231
3.4 Geodesic Hölder distributions for a cocompact Fuchsian group . . . . 232
3.5 Geodesic Hölder distributions for a general Fuchsian group . . . . . . 233
4 Analytic properties of the Liouville map . . . . . . . . . . . . . . . . . . . 237
4.1 Hölder regularity of holomorphic motions . . . . . . . . . . . . . . . 239
4.2 Continuity properties of derivatives of Hölder functions . . . . . . . . 241
4.3 Proof of Theorems 4.3, 4.1 and 4.2 . . . . . . . . . . . . . . . . . . . 245
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
1 Introduction
Thanks to a construction due to F. Bonahon, the Teichmüller space T () of a cofinite
volume Fuchsian group can be viewed as a subspace of a Banach space: indeed
the Liouville map L is an embedding of T () into the space of geodesic currents
of . This construction allowed Bonahon to recover one important result of Thurston,
namely, that the Teichmüller space T () can be compactified by the space of pro-
jective measured laminations. This embedding however is only topological and the
space of geodesic currents needs to be enlarged in order for the Liouville map to
become differentiable. This larger space is the (Fréchet) space of geodesic Hölder
distributions H(); it was introduced by Bonahon, who showed with Sözen, that the
Liouville map L : T () → H (), is differentiable along paths. This theorem was
recently generalized by D. Šarić to the setting of an arbitrary Fuchsian group (with
224 Jean-Pierre Otal
maybe infinite covolume). Our main result in this chapter is that the Liouville map
L : T () → H() is analytic; the analytic structure on T () is the one induced by
the complex structure, the analytic structure on H () is the one induced from the
linear (Fréchet) structure.
In the first section, we review the construction due to L. Bers of the complex
structure on T (). In Section 3, we recall the definition of the space of Hölder
distributions on the space of geodesics of H2 , introduced by F. Bonahon for cocompact
Fuchsian groups and its generalization to arbitrary Fuchsian groups by D. Šarić. We
show that L is analytic in the last section.
Acknowledgement. Part of this work was written at the Bernoulli Center in Lausanne
during the Workshop “Spaces of Negative Curvature”.
2 The Teichmüller space of a Fuchsian group

The group PSL(2, C) acts by homographies on the Riemann sphere C. Any discrete
subgroup G ⊂ PSL(2, C) leaves invariant a smallest non-empty closed subset L(G) ⊂
C called the limit set of G.
We denote by a Fuchsian group, i.e. a discrete and torsion-free subgroup of
PSL(2, R).
The most efficient way to define the Teichmüller space T () is within the theory
of quasiconformal homeomorphisms.
2.1 The quasiconformal homeomorphisms

Definition 2.1. A quasiconformal homeomorphism f of the Riemann sphere C is an
orientation-preserving homeomorphism such that
∂f ∂f
(1) the distributional derivatives ∂z and ∂ z̄ are measurable functions which are lo-
cally square integrable, and
(2) for some measurable function μ ∈ L∞ (C) with μ∞ < 1, one has, for almost
all z ∈ C,
∂f ∂f
=μ . (2.1)
∂ z̄ ∂z
The function μ ∈ L∞ (U ) with this property is called the Beltrami coefficient of f .

The dilatation of f is the quantity
1 + μ∞
K(f ) = .
1 − μ∞
Equation 2.1 is called the Beltrami equation.
Chapter 4. The embedding in the space of geodesic Hölder distributions 225
The first main property of quasiconformal homeomorphisms is the following com-

pactness property [1].
Proposition 2.2. Let K > 0. The set of K-quasiconformal homeomorphisms of C

which fix 3 distinct points is compact for the topology of uniform convergence on C.
Another main property of quasiconformal homeomorphisms is that they are trac-

table by analytic tools. The importance of the following theorem for the study of
Teichmüller space was realized by L. Ahlfors and L. Bers (see the commentary of
Ahlfors in [2, p. 226]).
Theorem 2.3 (Existence of solutions to the Beltrami equation [3], [14], [13], [15]).
Let μ ∈ L∞ (C), with μ∞ < 1. There is a unique quasiconformal homeomorphism
f μ : C → C such that
(i) f μ (0) = 0 and f μ (1) = 1;
(ii) the Beltrami coefficient of f μ is equal to μ.
The uniqueness part of Theorem 2.3 and Proposition 2.2 have the following con-
sequence.
Corollary 2.4. Let k < 1. Let (μi ) be a sequence in L∞ (C) such that μi ∞ ≤ k < 1
for all i and such that (μi ) tends to μ in L∞ (C). Then the sequence (f μi ) converges
to f μ uniformly over C
Definition 2.5 (The Teichmüller space and the quasi-Fuchsian space). A quasi-Fuchs-
ian deformation of (resp. a Fuchsian deformation of ) is a pair (ρ, f ) where ρ
is a representation of into PSL(2, C) (resp. into PSL(2, R)) and f is a quasiconfor-
mal homeomorphism of C (resp. a quasiconformal homeomorphism of C leaving R
invariant) normalized by f (0) = 0 and f (1) = 1 and which conjugates and ρ():
ρ(γ ) = f γ f −1 for all γ ∈ .
One defines an equivalence relation on Fuchsian or quasi-Fuchsian deformations
by setting (ρ1 , f1 ) (ρ2 , f2 ) if and only if f1 |R = f2 |R . When the limit set of is
all R, one has (ρ1 , f1 ) (ρ2 , f2 ) if and only if ρ1 = ρ2 .
The Teichmüller space T () (resp. the quasi-Fuchsian space QF ()) is the quo-
tient of the space of Fuchsian deformations of (resp. the space of quasi-Fuchsian
deformations of ) by the equivalence relation (cf. [11], [12], [13]). The spaces
T () and QF () are complete metric spaces for the distance which is the quotient
distance of a distance on the space of Fuchsian or quasi-Fuchsian deformations, namely
1
d((ρ1 , f1 ), (ρ2 , f2 )) = log K(f2 f1−1 ).
2
The theorem of existence of solutions to the Beltrami equation allows to parame-
terize T () and QF ().
Definition 2.6 (The Beltrami differentials). A Beltrami differential for is a measur-

able function μ ∈ L∞ (C, C) such that for all γ ∈ and for almost all z ∈ C,
γ
(z)
μ(γ z) = μ(z).
γ
(z)
The space of Beltrami differentials for is denoted by B(). With the L∞ -norm, it
is a Banach space; we denote by B 1 () its open unit ball.
Let μ ∈ B 1 (), let f μ : C → C be the quasiconformal homeomorphism provided

by Theorem 2.3. The transformation rule of μ under the action of implies that the
Beltrami coefficient of the quasiconformal homeomorphism f μ γ (f μ )−1 is 0;
therefore ρ μ (γ ) = f μ γ (f μ )−1 ∈ PSL(2, C) and the map γ → ρ μ (γ ) is a
representation of into PSL(2, C). Thus μ → (ρ μ , f μ ) defines a map from B 1 ()
to the space of quasi-Fuchsian deformations of . This map is a homeomorphism.
After post-composing this map with the equivalence relation which defines QF ()
as a quotient of the space of quasi-Fuchsian deformations, we obtain a map denoted
by μ → [μ] from B 1 () onto QF (); so the notation [μ] represents the equivalence
class of (ρ μ , f μ ).
For μ ∈ B 1 (), the quasiconformal homeomorphism f μ leaves R invariant if and
only if the Beltrami differential μ ∈ B 1 () commutes with the complex involution,
i.e. if and only if μ(z) = μ(z̄) for almost all z. We shall denote by BT () ⊂ B()
the subspace of Beltrami differentials μ with this symmetry property. Then μ → [μ]
maps BT1 () onto T ().
Definition 2.7 (The change of basepoint map). Let μ0 ∈ BT1 (); denote by μ0 the
Fuchsian group ρ μ0 (). The Teichmüller spaces T () and T ( μ0 ) are homeomor-
phic: the map (ρ, f ) → ρ (ρ μ0 )−1 , f (f μ0 )−1 at the level of (quasi-)Fuchsian
deformations induces a homeomorphism from T () to T ( μ0 ) which sends [μ] to [0].
This homeomorphism is the change of basepoint map. It is induced by a homeomor-
phism ν : B 1 () → B 1 ( μ0 ),
μ
μ − μ0 fz 0
μ → ν(μ) (f μ0 )−1 .
1 − μ̄0 μ fzμ0
The homeomorphism ν restricts to a homeomorphism between BT1 () and BT1 ( μ0 ).
2.2 The complex structure on Teichmüller space

Definition 2.8 (Holomorphic family of Beltrami differentials). A holomorphic family
of Beltrami differentials parametrized by the unit disc D ⊂ C is a map w → μw
from D to L∞ (C) which satisfies
(i) there exists a k < 1 such that μw < k for all w ∈ D, and
(ii) for almost all z ∈ C, the map w → μw (z) is holomorphic.
The following theorem of Ahlfors and Bers [3] is the basic tool for the construction
of a complex structure on T ().
Theorem 2.9 (Holomorphic dependence of solutions to the Beltrami equation). Let

w → μw ∈ L∞ (C) be a holomorphic family of Beltrami differentials parametrized
by the unit disc D ⊂ C. Then, for any z ∈ C the map w → f μw (z) is holomorphic.
The existence of a complex structure on T () was established by L. Bers [5]. We

explain now this construction. Denote by L the lower half-plane {z ∈ C | z < 0}.
Definition 2.10 (Holomorphic quadratic differentials). A holomorphic quadratic dif-

ferential for is (here) a holomorphic function φ : L → C such that for all γ ∈ ,
φ(γ (z))(γ
(z))2 = φ(z). For such a φ, z → y 2 |φ(z)| defines a -invariant function
on L. We say that the holomorphic quadratic differential φ is bounded when the norm
φb = sup{y 2 |φ(z)|}
z∈H
is finite. The space of bounded holomorphic quadratic differentials for with the
norm · b is a Banach space denoted by Qb (). When the Riemann surface H/
is isomorphic to a compact surface of genus g with n points deleted, then Qb () is a
finite dimensional space of complex dimension 3g − 3 + n.
The existence of the complex structure on T () is based on the following con-
struction. For μ ∈ BT1 (), let μ̃ be the differential in B 1 () such that μ̃(z) = μ(z)
for z ∈ H and μ(z) = 0 for z ∈ L. Then, f μ̃ is holomorphic when restricted to L: it
is a univalent map from L to C. A basic invariant of a (locally) univalent map f is its
Schwarzian derivative
f
3 f
2
S(f ) =
− .
f 2 f
Since f μ̃ conjugates the action of to the action of a subgroup of PSL(2, C), the
Schwarzian S(f μ̃ ) transforms like a holomorphic quadratic differential for . Since
f μ is globally univalent, the Nehari–Kraus Lemma says that S(f μ̃ ) is a bounded
holomorphic quadratic differential and that it belongs to the ball of radius 3/2 of
Qb (). Thus μ → S(f μ̃ ) is a map from BT1 () to Qb () with image contained in
the ball of radius 3/2.
One uses now the following observation (cf. [11, p. 98], [12, p. 133]).
Claim 2.11. Let μ1 and μ2 ∈ B 1 (). Then μ1 μ2 if and only if μ̃1 μ̃2 .
Clearly μ̃1 μ̃2 is equivalent to say that the two holomorphic maps f μ̃1 |L and
f μ̃2 |
L are equal. Therefore, by the direct part of Claim 2.11, if the two differentials μ1
and μ2 are equivalent, then the Schwarzian S(f μ̃1 ) and S(f μ̃2 ) are equal. Thus the
map μ → S(f μ̃ ) factorizes through a map from T () to Qb (): S(f μ̃ ) = ([μ]).
This map is injective. To see that, recall that if two univalent maps have the same
Schwarzian, they differ by post-composition with an element of PSL(2, C). Therefore,

since f μ̃1 and f μ̃2 fix 0, 1 and ∞, S(f μ̃1 ) = S(f μ̃2 ) implies that the restrictions f μ̃1 |L
and f μ̃1 |L are equal. By the reciprocal part of Claim 2.11, this holds only when the
two differentials μ1 and μ2 are equivalent. Therefore induces a bijection between
T () and (T ()) which is indeed a homeomorphism (cf. [12], [11]; note that the
continuity follows from Corollary 2.4. We shall identify T () (as a topological space)
and (T ()).
The next tool for proving the existence of the complex structure is a construction
due to L. Ahlfors and G. Weill which provides local sections to .
Definition 2.12 (Harmonic Beltrami differentials). Let φ ∈ Qb (). Setting μφ (z) =

−2y 2 φ(z̄) for z ∈ H and μφ (z) = −2y 2 φ(z) for z ∈ L defines a Beltrami differential,
element of B(). A Beltrami differential of the form μφ is said a harmonic Beltrami
differential. The space of harmonic Beltrami differentials is denoted by BT (): it is
1
a closed subspace of B(). If φ ∈ Qb2 (), the open ball of radius 21 in Qb (), then
μφ ∈ B 1 (). It is now a theorem of L. Ahlfors and G. Weill that the Schwarzian
derivative of f μ˜φ |L is equal to φ [4]. This means that ([μφ ]) = φ; in other words,
1
φ → [μφ ] is a section to . In particular (T ()) contains the ball of radius Qb2 ().
1
The map φ → [μφ ] from the neighborhood Qb2 () of [0] ∈ T () to B1T () is a chart
for T () around [0] called a harmonic chart.
The same property holds for all other points [12, p.136]. Let μ0 ∈ BT1 (), and
let (ρ μ0 , f μ0 ) be the Fuchsian deformation of representing the point [μ0 ] ∈ T ().
The change of base point map [ν] (cf. Definition 2.7) maps neighborhoods of [μ0 ]
in T () to neighborhoods of [0] in T ( μ0 ). The preceding construction provided
a chart for a neighborhood of [0] in T ( μ0 ); precomposed with [ν], it gives a chart
which models a neighborhood of [μ0 ] in T () with a neighborhood of 0 in BT ( μ0 ).
The transition charts are holomorphic: this follows from the holomorphic dependence
of the solutions to the Beltrami equation. The complex structure on Teichmüller space
is obtained now since the Banach spaces BT ( μ ) are isomorphic. This is clear when
H2 / is a surface of finite type since BT ( μ ) is then isomorphic to C3g+n . This is
true also in general (cf. [12, p. 136]).
Definition 2.13 (The Ahlfors–Bers map). Let [μ] ∈ QF (). By the Riemann uni-
formisation theorem, there exist two Fuchsian groups ρ 1 (), and ρ 2 (), two univalent
maps f 1 : f μ (H) → C and f 2 : f μ (L) → C such that f 1 (resp. f 2 ) conjugates the
action of ρ μ () on f μ (H) (resp. on f μ (L)) to the action of ρ1 () on H (resp. to the ac-
tion of ρ2 () on L). Since f μ (R) is locally connected, Carathéodory’s theorem implies
that the homeomorphism f 1 extends to a homeomorphism from f μ (H) to H. Using
this, it is not difficult to see that the homeomorphism defined by f˜1 (z) = f 1 f μ (z)
for z ∈ H and f˜1 (z) = f 1 f μ (z̄) for z ∈ L is a quasiconformal homeomorphism
of C which conjugates the actions of and ρ 1 () on H. The same procedure gives
a quasiconformal homeomorphism f˜2 of C which conjugates the actions of and

ρ 2 () on L.
In this way, one obtains a map from the space of quasi-Fuchsian deformations of
to the product of two copies of the space of Fuchsian deformations of . This
map can be expressed as follows at the level of the Beltrami differentials. Let μ =
(μ1 , μ2 ) ∈ B(), where μ1 = μ|H and μ2 = μ|L ; then the map assigns to μ the pair
of Beltrami differentials (μ̃1 , μ̃2 ) in BT () × BT () where μ̃1 (z) = μ1 (z) if z ∈ H,
μ̃1 (z) = μ1 (z̄) if z ∈ L, and μ̃2 (z) = μ2 (z) if z ∈ L, μ̃2 (z) = μ2 (z̄) if z ∈ H. From
the construction, this map commutes with [ · ], giving a homeomorphism, called the
Ahlfors–Bers map, from QF () to T () × T ().
Definition 2.14 (Complex harmonic charts for QF ()). Using the Ahlfors–Bers map,
QF () becomes a complex manifold whose charts are the products of the harmonic
charts for T (): these charts are called the complex harmonic charts. Such a chart
1 1
identifies the neighborhood Qb2 () × Qb2 () of [0] ∈ QF () with B1T () × B1T ()
1 1
assigning to ϕ = (φ1 , φ2 ) ∈ Qb2 () × Qb2 () the point [νϕ ] ∈ QF () where
νφ ∈ B() is defined by νϕ (z) = μφ1 (z) for z ∈ H and νϕ (z) = μφ2 (z) for z ∈ L.
The Beltrami differentials of the form νφ will be called complex harmonic, in order to
differentiate them from the harmonic Beltrami differentials, elements of BT (). The
complex harmonic differentials form a complex subspace B() of B(). The space
BT () is a totally real subspace of B(), namely BT () = {(μ1 , μ2 ) | μ1 = μ2 }.
Under the complex harmonic chart, T () maps to BT (). The same charts can be
constructed near any point [μ] ∈ T () showing that T () is a totally real submanifold
of the complex manifold QF ().
3 Geodesic currents and Hölder distributions

In this section we associate to a Fuchsian group the space of geodesic currents M()
and the space of Hölder distributions H(). For a group which is cocompact, the
space M() was introduced in [6]. In [7], Francis Bonahon defined also a map
L : T () → M(), the Liouville map. One main feature of this map is that it is a
topological embedding and that it allows to recover the Thurston compactification of
the Teichmüller space by the space of measured geodesic laminations. In [8], Bonahon
and Sözen defined, for a cocompact Fuchsian group , a new space, larger than M():
the space of Hölder distributions H (). They showed that L was differentiable in a
certain sense (Theorem 3.6).
The definition of the space of Hölder distributions was generalized for a general
Fuchsian groups by Dragomir Šarić: he showed that L is a topological embedding
and that the same theorem as Theorem 3.6 did hold in this context too [17], [18].
3.1 The space of geodesics of H2

An unparametrized geodesic g of H2 is entirely determined by its two endpoints (ξ, ζ )
which are two distinct points in ∂ H2 = R. The space of geodesics of H2 , G, is the
complement of the diagonal in R×R. The action of PSL(2, R) on R induces an action
on G which is the product action on R × R −. The space G is a C ∞ -manifold covered
by the open sets γ U for γ ∈ PSL(2, R) where U = {(ξ, ζ ) ∈ R2 | ξ = ζ }. We denote
by d a distance on G corresponding to this smooth structure. The boundary R being
identified with the circle R/Z, we shall take the distance induced on R from the angular

by |ξ −
ξ
| and then the distance
distance on the circle that we shall denote induced

from the product distance on R × R: d (ξ, ζ ), (ξ , ζ ) = sup |ξ − ξ
|, |ζ − ζ
| .
Let C(G) be the space of continuous functions h : G → C which have compact
support, with the norm h∞ = supg∈G |h(g)|. If K is a compact set contained
in G, we shall denote by CK (G) the subspace of functions in C(G) whose support is
contained in K. It is a Banach space for the norm h∞ = supg∈K |h(g)|. The dual
of CK (G) is a Banach space for the norm T K = sup |T (h)|, the supremum being
taken over all the functions h ∈ CK (G) with h∞ ≤ 1.
If f is a homeomorphism of R, we denote by F the homeomorphism of G induced
by the diagonal action of f
(ξ, ζ ) → (f (ξ ), f (ζ )) = F (ξ, ζ ).
This homeomorphism F induces an isomorphism of C(G) by sending h to f ∗ (h) =
h F −1 .
3.2 Geodesic currents

A linear form on C(G) is continuous if, for each compact set K ⊂ G, its restriction to
CK (G) is continuous. We denote by M the space of continuous linear forms on C(G).
Definition 3.1 (The geodesic currents). The action of on C(G) induces by duality
an action of on M. A geodesic current is a functional T ∈ M which is -invariant,
i.e. which satisfies, for any γ ∈ , and for any h ∈ C(G), T (γ ∗ (h)) = T (h). The
space of the geodesic currents is denoted by M().
By the Riesz representation theorem [16], if T ∈ M, there exists a complex Borel

measure λ on G such that, for all h ∈ C(G), one has T (h) = hdλ. Furthermore, the
measure λ is positive if and only if the functional T is. For any compact set K ⊂ G
such that λ(∂K) = 0, one has
λ(K) = T K = sup{T (h) | h ∈ CK (G), h∞ ≤ 1}.
One of the features of the space M() is that it contains the Teichmüller space
T (). The natural map which realizes this embedding is the Liouville map.
3.3 The Liouville measure

We recall first some equivalent ways to define the Liouville measure on G.
• The group PSL(2, R) acts on G preserving a Radon measure, the Liouville mea-
sure. In the charts U = R × R −, this measure has the expression dλ = |ξdξ−ζdζ|2 .
In the charts γ U, it is given by exactly the same formula. The fact that these lo-
cal expressions patch together is a reformulation of the basic functional equation
satisfied by all elements of the Möbius group PSL(2, C): for all ξ , ζ in C
(γ (ξ ) − γ (ζ ))2 = γ
(ξ )γ
(ζ )(ξ − ζ )2 .
• The space G is also a pseudo-Riemannian manifold: the metric

1
ds 2 = dξ ⊗ dζ
|ξ − ζ |2
is invariant under the action of PSL(2, R) and the Liouville measure λ is the
(pseudo-Riemannian)-measure associated to this metric.
• Consider the differential 2-form ω on C × C − , defined in the chart C × C by
∂2
ω= log |ξ − ζ |dξ ∧ dζ.
∂ξ ∂ζ
In restriction to R × R − , ω is a volume form whose absolute value is λ.
• The measure λ is related to the cross-ratio. Let a, b, c and d be four points on R
be such that the intervals [a, b] and [c, d] are disjoint. Then [a, b] × [c, d] is a
rectangle in G and

ω = log |a − c| + log |b − d| − log |a − d| − log |b − c|.
[a,b]×[c,d]
Then λ([ab] × [cd]) = |log |(a, b, c, d)|| , where (a, b, c, d) = (a−c)(b−d)

(a−d)(b−c) is the
cross-ratio of the four points a, b, c, d. If these points occur in this order on R,
then log |(a, b, c, d)| > 0.
Definition 3.2 (The Liouville map). Let [μ] ∈ T () and let F μ denotes the home-
omorphism of G which is induced by the diagonal action of f μ , i.e. F μ (ξ, ζ ) =
(f μ (ξ ), f μ (ζ )). The group ρ μ () preserves the Liouville measure λ. Thus, since the
actions on G of the groups and ρ μ () are conjugated by F μ , the group preserves the
pullback measure (F μ )∗ (λ). Recall that the pullback measure (F μ )∗ (λ) is character-
ized by the property that for any Borel set A, one has (F μ )∗ (λ)(A) = λ(F μ (A)). Since
F μ only depends on the point [μ] in T (), one defines a map L : T () → M() by
setting L([μ]) = (F μ )∗ (λ). This is the Liouville map.
As a map from T () to M(), L has no smoothness properties. However, if one

replaces M() by the larger space H () of geodesic Hölder distributions, Bonahon
and Sözen showed that the new map L : T () → H() becomes C 1 along paths.
We introduce now the space H(), first when is cocompact like in [8], then when
is a general Fuchsian group [17], [18].
3.4 Geodesic Hölder distributions for a cocompact Fuchsian group

In this section we shall suppose that is cocompact.
Continuous linear forms on C(G) are particular examples of distributions in the
sense of Laurent Schwartz. However, if one intends to generalize the notion of geodesic
currents invariant under a Fuchsian group to the context of distributions, one is faced
with the difficulty that G, although it is a smooth manifold, is not smooth in a natural
way. Indeed let [μ] ∈ T (); then the homeomorphism f μ : R → R which conjugates
the actions of and of μ on R is not smooth in general and the same holds a fortiori
for F μ . It follows that the map that F μ induces on C(G) does not preserve the spaces
of differentiable functions of any differentiability class. In particular, there are no
natural isomorphisms between the spaces of distributions which are invariant under
and of those invariant under μ . Francis Bonahon solved this difficulty by introducing
the appropriate functional space, the space of Hölder geodesic distributions.
Definition 3.3 (Distributions of order α). Let (K, d) a metric space. Let α ∈]0, 1[. A
function h : K → C is α-Hölder if there is a constant C such that |h(x) − h(x
)| ≤
Cd(x, x
)α for all x, x
in K. For such a function h, one defines the α-Hölder norm
by
|h(x) − h(x
)|
hα = sup |h(x)| + sup .
x∈K x =x
d(x, x
)α
The space of α-Hölder functions h : K → C which have compact support is denoted
Lipα (K). For K a compact subset of G, we denote by LipαK (G) the subspace of Lipα (G)
which consists of the functions whose support is contained in K. With the norm · α ,
LipαK (G) is a Banach space. The dual of LipαK (G) is called the space of distributions
of order α on K. It is a Banach space for the norm · α which is the dual norm to
the norm · α ; it will be denoted by H α (K).
When K is a compact submanifold of G, it is a classical fact that the subspace

C ∞ (G) ∩ LipαK (G) is dense in LipαK (G) for the α-Hölder norm.
Definition 3.4 (Geodesic Hölder distribution). A geodesic Hölder distribution of order

α is a linear functional T : C ∞ (G) → C which, for each compact submanifold K
of G, can be extended continuously to LipαK (G) as an element of H α (K). The space
of geodesic Hölder distributions of order α is denoted by H α . If α
≤ α, one has

Hα ⊂ Hα.
A geodesic Hölder distribution is a linear functional T : C ∞ (G) → C which
belongs to H α for allα < 1. The space of geodesic Hölder distributions is denoted
by H. One has H = α H α .
A geodesic Hölder distribution for is a functional T ∈ H which is -invariant;

these distributions form a closed subspace H () of H . One has H () = α H α ().
Since is cocompact, there is a compact submanifold K ⊂ G such that G is covered
by the translates of K by the elements of : choose for instance a submanifold K which
contains in its interior the set of the geodesics of G which pass through a fundamental
domain for the action of on H. Define now a norm on H α () by T αK . This norm
turns H α into a Banach space and the family of norms { · αK }, as α ∈]0, 1[ turns H
into a Fréchet space. It is not difficult to see that this topology can also be defined by
1/n
the family of norms { · K }. Clearly the choice of another compact K defines the
same topology as long as K keeps the property that the union of its translates equals G.
Remark 3.5. For [μ] ∈ T (), it is known that f μ is Hölder. Let δ > 0 be such that
(f μ )−1 is Hölder with exponent δ. Then F μ (ξ, ζ ) = (f μ (ξ ), f μ (ζ )) is δ-Hölder
also. In particular, for each compact set K ⊂ G and for each α < 1, F μ acting on
C c (G) by h → h (F μ )−1 maps continuously LipαK (G) to Lipδα F μ (K) (G). By duality,
μ αδ α
this map induces a linear map H (f ) : H () → H () which is continuous.
Since for [μ] ∈ T (), (f μ )−1 is also Hölder, H (f μ ) is a linear isomorphism.
The space M() is contained in H(): for ν an element in M(), we shall use
the notation ν, h = G hdν. The Liouville map can therefore be viewed as a map
from T () to H (). The following theorem is due to Bonahon and Sözen [8].
Theorem 3.6. Let t → c(t) ∈ T () be a smooth path defined on the interval |t| ≤ ε.
Then the map t → L(c(t)) is differentiable: for any α < 1, and for any function
h ∈ LipαK (G), the function t → L(c(t)), h is differentiable.
3.5 Geodesic Hölder distributions for a general Fuchsian group

For a Fuchsian group which is not cocompact, the above definition of the space H ()
is not sufficient in order for the map L to be even a topological embedding. An extra
boundedness property for the geodesic Hölder distributions is required. The definition
which follows was introduced by D. Šarić [17], [18].
To begin with, let R be the set of rectangles [a, b] × [c, d] ⊂ G for which the
cross-ratio (a, b, c, d) equals 2: this means that the geodesics ac and bd make an angle
equal to π/2 at their intersection point. Fix an element of R, R0 = [a0 , b0 ] × [c0 , d0 ]
such that the intervals [a0 , b0 ] and [c0 , d0 ] are contained in R. For any rectangle
R = [a, b]×[c, d] in R, let q be the element of PSL(2, R) which maps (a0 , b0 , c0 , d0 )
to (a, b, c, d). If h ∈ LipαR (G), then h Q ∈ LipαR0 (G) (recall the notation Q(ξ, ζ ) =
(q(ξ ), q(ζ ))).
Definition 3.7 (Geodesic Hölder distribution). Let α < 1. A bounded geodesic Hölder
distribution of order α is a linear functional T : C ∞ (G) → C such that
(1) for each rectangle R ∈ R, T extends continuously to LipαR (G) as an element of

H α (R), and
(2) the quantity T α = sup{|T (φ)|} is finite, where the supremum is taken over all
rectangles R ∈ R and all φ ∈ LipαR (G) such that φ Qα ≤ 1.
The space of bounded geodesic Hölder distributions of order α is denoted by H α . It
is a Banach space for the norm . α .
Let be a Fuchsian group. The space of bounded geodesic Hölder distributions of
order α which are invariant under the action of is a closed subspace of H α , denoted

by H α (). For α
≤ α, one has H α () ⊂ H α ().
A geodesic Hölder distribution for is a linear functional T : C ∞ (G) → C which
belongs to H α () for each α. They form a space H, equal to the intersection α H α ;
H is a Fréchet space whose topology is defined by the family of semi-norms · α .
Remark 3.8. When is cocompact, the above definition agrees with the one given
in the preceding subsection.
Remark 3.9. If one replaces, in the definition of bounded geodesic Hölder distribution,
the set of rectangles R by the set Rβ of those rectangles [a, b]×[c, d] with (a, b, c, d)
equal to β, one obtains a set of bounded Hölder geodesic distributions which is equal
to H (). For each α < 1, the norm on H α () defined as before, after the choice of
a rectangle R0
∈ Rβ playing the same role as R0 , is equivalent to the norm · α .
Remark 3.10. Remark 3.5 can be made also in the present situation. Let [μ] ∈ T (),
and suppose that (f μ )−1 is δ-Hölder. Then the map h → h (F μ )−1 induces by
duality a map H(f μ ) : H αδ ( μ ) → H α () which is continuous for each α < 1.
Also H (f μ ) : H ( μ ) → H () is an isomorphism.
Claim 3.11. For [μ] ∈ T (), the Liouville measure L([μ]) belongs to H().
Proof.
A positive Radon measure ν on G defines a distribution by integration h →
α
G hdν. For each rectangle R ∈ R, this distribution is clearly continuous on LipR (G)
with norm less than ν(K). To prove Claim 3.11, we need to show that for each
[μ] ∈ T (), and for each R ∈ R, the measures L([μ])(R) are bounded independently
of R. Let R be a rectangle in R: R = [a, b] × [c, d] with (a, b, c, d) = 2. Let
q ∈ PSL(2, R) be the element sending R0 to R and let r ∈ PSL(2, R) be the element
such that r f μ q fixes three of the vertices of the rectangle R0 . Then the Beltrami

differential of the quasiconformal homeomorphism r f μ q equals μ (q(z)) qq

(z) (z) ;
∞
in particular, it has the same L -norm as μ. It follows from Proposition 2.2 that
λ([f μ (a), f μ (b)] × [f μ (c), f μ (d)]) is bounded from above and from below by two
positive constants depending only on μ∞ and not on R. Saying that these numbers
are bounded from above implies that L([μ]) belongs to H ().
This remark allows us to consider the Liouville map L as a map from T () to
H (). The following theorem is due to D. Šarić [18].
Theorem 3.12. The Liouville map L : T () → H () is a proper topological em-
bedding.
Proof. The map L is continuous. We begin with the continuity at [0] ∈ T (). Let
α > 1. Let R be a rectangle in R and h ∈ LipαR (G). Let q ∈ PSL(2, R) be the Möbius
map sending R0 to R. We need to show that |L([μ])(h)−L([0])(h)| ≤ C(μ)hQα
for a constant C(μ) which is independent of R and h and which tends to 0 with μ∞ .
By definition of the Liouville map, one has

μ −1
L([μ]), h = h F dλ = (h Q) Q−1 (F μ )−1 dλ.
G G
Let qμ ∈ PSL(2, R) be the Möbius map such that the quasiconformal homeomor-
phism
φμ = q μ f μ q
fixes 0, 1 and ∞; let μ (ξ, ζ ) = (φ μ (ξ ), φ μ (ζ )). Then, since the homeomorphism
Qμ , where Qμ (ξ, ζ ) = (q μ (ξ ), q μ (ζ ))) preserves λ, one has

L([μ]), h = (h Q) (μ )−1 dλ = (μ )∗ λ, h Q. (3.1)
G
Since h Q ∈ LipαR0 (G), one has

α
|L([μ]), h − L([0]), h| ≤ λ(R0 )h Qα sup d (μ )−1 (g), g .
g∈R0

Observe that the Beltrami coefficient of φ μ is q ∗ (μ) = (μ q) qq

. In particular,
q ∗ (μ)∞ = μ∞ . Corollary 2.4 implies that supξ ∈R d((φ μ )−1 (ξ ), ξ ) ≤ C(μ∞ )
where C(μ∞ ) tends to 0 with
μ∞ ; the same property is therefore also satisfied
by supg∈R0 d (μ )−1 (g), g . Thus, one has, for all h ∈ LipαR (G)
|L([μ]), h − L([0]), h| ≤ C(μ∞ )hα
where C(μ∞ ) tends to 0 with μ∞ and does not depend on R. This means that L
is continuous at [0].
The proof of the continuity at an arbitrary point [μ0 ] ∈ T () reduces to the
above argument using the change of basepoint map [ν] (cf. Definition 2.7). For any
μ ∈ B 1 (), one has f μ = f ν(μ) f μ0 . So, if h ∈ C(G), one has

∗
L([μ]), h = h (F μ0 )−1 F ν(μ) (dλ) = L([ν(μ)]), h (F μ0 )−1 . (3.2)
Equation 3.2 says that L = H(f μ0 )Lμ0 [ν], where H (f μ0 ) : H ( μ0 ) → H() is

the map induced by f μ0 (cf. Remark 3.5), Lμ0 : T ( μ0 ) → H( μ0 ) is the Liouville
map for the group μ0 , and [ν] is the change of base point map. In the remark 3.10,
we noticed that H (f μ0 ) is a linear isomorphism between the Fréchet spaces H( μ0 )
and H(). By the first part of the proof, L is continuous at [0]. Since [ν] is a
homeomorphism, it follows that L is continuous at [μ0 ].
L is injective. If (F μ1 )∗ (λ) = (F μ2 )∗ (λ), one obtains, by evaluating these measures

on rectangles [a, b] × [c, d] ⊂ R × R that f μ2 (f μ1 )−1 preserves the cross-ratio of
any 4-tuples in R. Since the homeomorphisms f μi are normalized, this implies that
f μ1 |R = f μ2 |R , and so [μ1 ] = [μ2 ].
L is an embedding. This means that L is a homeomorphism from T () to its image
with the topology induced from that of H(). To see that, it remains to prove that the
inverse map L−1 is continuous.
Suppose that ([μi ]) is a sequence in T () such that L([μi ]) tends to L([ν]) in
H(). We want to prove that [μi ] tends to [ν] in T (). Extend the characteristic
function χR0 of R0 to a positive and α-Hölder function h ≤ 1 supported on a rectangle
R0
= [a, b] × [c, d] with (a, b, c, d) = β. Let R be a rectangle in R, and let
q ∈ PSL(2, R) the element sending R0 to R. Then h Q−1 is a positive and α-Hölder
function which extends the characteristic function of R; it is bounded from above by
1 and supported in the rectangle Q(R0
), which is an element of Rβ . By Remark 3.9
and by the definition of the topology of H (), for all ε, there is an i0 such that for
i ≥ i0 , the inequality
|L([μi ]), h Q−1 − L([ν]), h Q−1 | ≤ ε
holds for any rectangle R ∈ R. By the proof of claim 3.11,
L([ν]), h Q−1 ≤ L([ν]), χqR
≤ C,
where the constant C only depends on ν∞ and on β. Also,
L([μi ]), h Q−1 ≥ L([μi ]), χR .
Therefore, for all i ≥ i0 , and for all R ∈ R, one has
λ(F μi (R)) = L([μi ]), χR ≤ C + ε. (3.3)
We now have
Claim 3.13. The sequence ([μi ]) is bounded in T ().
Proof. The “complementary” rectangle to a rectangle [a, b] × [c, d] ∈ R, i.e. the

rectangle [b, c]×[d, a] is also an element of R. Therefore Equation 3.3 implies that the
measures λ(F μi (R)) are also bounded from below by a non-zero constant independent
of i and of R ∈ R. From the definition of the Liouville measure, we deduce that for all
rectangles R = [a, b]×[c, d] in R, the cross-ratios (f μi (a), f μi (b), f μi (c), f μi (d))
are bounded between two positive constants independent of i. This implies that f μi is
S-quasisymmetric, with a constant S independent of i. Indeed, for x ∈ R and t ∈ R,
the rectangle [x − t, x] × [x + t, ∞] belongs to R since (x − t, x, x + t, ∞) = 2;
therefore
1 μi
|f (x) − f μi (x − t)| ≤ |f μi (x + t) − f μi (x)| ≤ S|f μi (x) − f μi (x − t)|,
S
for a constant S independent of x, t and i. It is known that any S-quasisymmetric

homeomorphism f of R extends to a K(S)-quasiconformal homeomorphism of H.
The extension operator constructed by Douady–Earle [10] has the property of com-
muting with the action of PSL(2, R): it provides therefore an extension of f μi |R
which is a K(S)-quasiconformal homeomorphism and which is of the form f νi ,
for νi ∈ B(). By definition, [μi ] = [νi ]: the sequence ([μi ]) is therefore bounded
in T ().
By Claim 3.13, the sequence ([μi ]) converges to [μ] ∈ T () up to possibly
extracting a subsequence. By the continuity of L, L([μ]) = L([ν]). By the injectivity
of L, [μ] = [ν]. Therefore L−1 is continuous.
L is proper. Let ([μi ])i∈N be an unbounded sequence in T (). Then, by the proof
of Claim 3.13, for some rectangle R ∈ R the sequence (L([μi ])(R)) is unbounded
as i → ∞. It follows that (L([μi ])) is unbounded in H ().
4 Analytic properties of the Liouville map

This section is devoted to the study of regularity properties of the map L with respect
to the analytic structure on T (). By definition, the space H() of the geodesic
Hölder distributions for is the intersection of the spaces H α () as α decreases to 0.
In particular, for each α < 1, L is a map from T () to H α ().
Theorem 4.1. For each 0 < α < 1, there exists an open neighborhood N α of T ()
in QF () such that L extends to a holomorphic map L̃ : N α → H α ().
Recall that a map between complex Banach manifolds is holomorphic if and only
if, when it is expressed in local charts, its restriction to each affine complex disc is
holomorphic [9, p. 28]. Saying that a map F : D → E from the unit disc to a Banach
manifold E is holomorphic can be formulated in any of the following equivalent ways:

(1) the map F (w) can be expanded as a series n∈N wn Fn which is uniformly
convergent over compact sets in D;
(2) the map F is continuous and the Cauchy formula holds: for all r < 1 and for all
|w| ≤ r, one has
1 F (z)
F (w) = dz.
2iπ |z|=r z − w
To prove Theorem 4.1 we shall construct a neighborhood N α of T () in QF (),

and a map L̃ : N α → H α () which extends L and whose restriction to any affine
holomorphic disc can be expanded as a series n∈N wn L̃n , absolutely convergent
in H α () over compact sets. This neighborhood N α depends on α and it seems
that this is necessary, i.e. it seems impossible that L : T () → H () extends to
a holomorphic map defined on some neighborhood of T () in QF (). However,

Theorem 4.1 has the following corollary, which we state as a theorem, although it
follows directly from 4.1.
Theorem 4.2. The Liouville map L : T () → H() is analytic.
We shall first prove the following “local” version of Theorem 4.1. Denote by
Gf ⊂ G, the “finite part of G ”, i.e. the set {(ξ, ζ ) ∈ R2 | ξ < ζ }. For technical
reasons – namely the expression for the Liouville measure is simpler in Gf – we
first consider rectangles contained in Gf . By restriction to functions supported in a
rectangle R, L can be viewed as a map taking values in H α (R); we continue to denote
this map by L.
Theorem 4.3. Let R be a rectangle contained in Gf . Then, there exists a neighborhood

N α of T () in QF () such that L : T () → H α (R) extends to a holomorphic map
L̃ : N α → H α (R).
The neighborhood N α provided by the proof will depend on R and on α.

To prove Theorem 4.3, we give another definition of the Liouville map.
Claim 4.4. Let [μ] ∈ T (). Then in restriction to Gf , the measure L([μ]) is equal to
the distribution ∂ξ∂∂ζ log(f μ (ζ ) − f μ (ξ )), i.e. for each function h : Gf → C of class
2
C 2 and with compact support, one has

∂ 2h
L([μ]), h = · log(f μ (ζ ) − f μ (ξ )) dξ dζ.
G ∂ξ ∂ζ
Proof. One can approximate f μ by a sequence (fi )i∈N of diffeomorphisms fi : R →

R such that, when i → ∞, (fi ) converges to f μ and (fi−1 ) converges (f μ )−1
uniformly over R. Denote by Fi the diffeomorphism (ξ, ζ ) → (fi (ξ ), fi (ζ )) of G.
Then the measure Fi∗ (λ) is absolutely continuous with respect to the Lebesgue measure
dξ dζ on Gf with density equal to ∂ξ∂∂ζ log(fi (ζ ) − fi (ξ )). Integrating by parts we
2
deduce that for each function h : Gf → C of class C 2 and with compact support

∗ ∂ 2h
Fi (λ), h = log(fi (ζ ) − fi (ξ )) dξ dζ. (4.1)
G ∂ξ ∂ζ
As i tends to ∞, the uniform convergence of Fi−1 to (F μ )−1 implies that Fi∗ (λ), h =
λ, hFi−1 tends to λ, h(F μ )−1 that is L([μ]), h. The right term of 4.1 converges
2h
to G ∂ξ∂ ∂ζ log(fi (ζ ) − fi (ξ )) dξ dζ . This proves Claim 4.4.
Definition 4.5. Denote by C 2 (R) the space of the functions h ∈ G which are of
class C 2 and whose support is contained in R. Observe for later use, that for each
0 < α < 1, C 2 (R) is dense in LipαK (G). Denote by D(R) the dual of C 2 (R); with
the norm dual to the C 2 -norm, D(R) is a Banach space.
Since M() is contained in D(R), one can consider L as a map taking values
in D(R).
The following corollary is an easy consequence of Claim 4.4.
Corollary 4.6. The map L : T () → D(R) extends to a holomorphic map

L̃ : QF () → D(R).
Proof. It suffices to see that the map BT1 () → D(R) which assigns to μ ∈ BT1 () the
Liouville measure L([μ]) extends to a map L̃ : B 1 () → D(R) which is holomorphic
and only depends on [μ] (and so defines a map on QF ()). Recall that the rectangle
R is contained in Gf ; the expression for L̃ is suggested by Claim 4.4. Let us define,
for μ ∈ B 1 (), L̃(μ) ∈ D(R) as the distribution
∂2
log(f μ (ζ ) − f μ (ξ )).
∂ξ ∂ζ
This distribution certainly only depends on the equivalence class of [μ]. Also, by
Claim 4.4, for μ ∈ BT1 (), one has L̃([μ]) = L([μ]). By Theorem 2.9, log(f μ (ζ ) −
f μ (ξ )) is a holomorphic function of μ, for each (ξ, ζ ). Since this function is also con-
tinuous in the three variables, the measure log(f μ (ζ ) − f μ (ξ ))dξ dζ is holomorphic
in μ. The corollary follows.
For proving Theorem 4.3, we shall construct for each α < 1, a neighborhood N α
of T () in QF () on which the map L̃ restricts to a holomorphic map with values
in H α (R). The next two subsections develop the tools necessary for doing this.
4.1 Hölder regularity of holomorphic motions

Let w → μw ∈ L∞ (C) be a holomorphic family of Beltrami differentials parametrized
by D. Denote by f w the normalized solution to the Beltrami equation with Beltrami
coefficient μw . The map (w, z) → f w (z) from D × C to C has the following prop-
erties:
(i) for all w ∈ D, f w fixes 0, 1 and ∞,
(ii) for all z ∈ C, w → f w (z) is holomorphic (by Theorem 2.9), and
(iii) for all w ∈ D, z → f w (z) is injective.
Notice that when μ0 = 0 ∈ L∞ (C), then f 0 (z) = z for all z ∈ C: in that case, the map
(z, w) → f w (z) is called a holomorphic motion of C. In our general case, we shall
call f w (z) the holomorphic motion associated to the family w → μw . It is classical
that the restriction to bounded sets of C of a holomorphic motion parametrized by the
unit disc is Hölder with exponent 1−|w|

1+|w| (cf. for instance [12, p. 13]). We give a proof
of this statement in our situation.
For r > 0, denote by Dr the euclidean disc of radius r: Dr = {z, |z| ≤ r}.
Proposition 4.7. Let w → μw be a holomorphic family of Beltrami differentials such

that μw ∞ ≤ k < 1 for all w ∈ D and let f w be the holomorphic motion associated
to this family. Let r > 0. Then there exists a constant C = C(r, k) such that if z1 and
z2 ∈ Dr , then for all w ∈ D, one has
1−|w|
|f w (z1 ) − f w (z2 )| ≤ C|f 0 (z1 ) − f 0 (z2 )| 1+|w| .
Proof. It follows from the holomorphic dependence in w of the map w → f w (z),

that for any distinct points z1 and z2 , the function w → log |f w (z1 ) − f w (z2 )| is
harmonic. Suppose |zi | ≤ r, for i = 1, 2. Since μw ∞ ≤ k, the continuity of the
solutions to the Beltrami equation (Proposition 2.2) implies that there is a constant
D = D(r, k) such that |f w (z1 ) − f w (z2 )| ≤ D. Therefore the function

w

f (z1 ) − f w (z2 )
w → − log

is a positive harmonic function on D. By the Harnack inequality [16], the values of a

positive harmonic function on D are controlled by its value at 0. In our situation, this
gives
w

0

f (z1 ) − f w (z2 )
1 − |w|
f (z1 ) − f 0 (z2 )
− log

≥−

log
D 1 + |w| D
Applying the exponential map to both sides of this inequality gives the required result
2|w|
where the constant C(r, k) is the supremum of D 1+|w| , that is sup(D, 1).
Remark 4.8. If μ0 = 0, then this is a classical result about holomorphic motions.

Notice however that if one does not assume that the differentials μw satisfy μw ≤ k
for some k < 1, then it could happen that the diameter of the image f w (Dr ) is not
bounded independently of r: this would prevent the constant C to be independent of w.
For instance, the holomorphic motion (w, z) → z+w z̄
1+w does not satisfy the conclusions
of the Proposition 4.7 when w tends to −1 (compare with [12, p. 13]).
Here is now a useful consequence of the Proposition 4.7. Recall the notation: for
ε > 0, B ε = {μ ∈ L∞ (C) | μ < ε}.
Proposition 4.9. Let r > 0. For any 0 < β < 1, there exists ε(β) > 0 such that for
μ ∈ B ε(β) one has
(i) the restriction f μ |Dr belongs to Lipβ (Dr ), and its β-Hölder norm is bounded in
terms of r and β;
(ii) the map μ → f μ |Dr from B ε to the Banach space Lipβ (Dr ) is holomorphic.
1−β
Proof. Set ε(β) = 2(1+β) . Applying Proposition 4.7 to the (linear) holomorphic
μ
family w → w 2μ , we deduce that for μ ≤ ε(β), the restriction f μ |Dr belongs
to Lipβ (Dr ) and its β-Hölder norm is smaller than a constant C which depends only
on r and β. This proves (i).
In order to prove (ii), we consider an affine complex disc w → μw = μ0 + wμ1
defined for w ∈ D and with image contained in B ε(β) . We denote f w the holomorphic
motion which corresponds to μw . For z ∈ Dr , w → f w (z) is holomorphic. Therefore
one can write f w (z) = Fn (z)w n , the convergence being uniform in w over the
disc |w| ≤ ρ < 1 for all ρ < 1. The Cauchy formula for the n-th derivative of
f w (z1 ) − f w (z2 ) at w = 0 gives

1 (f w (z1 ) − f w (z2 ))
Fn (z1 ) − Fn (z2 ) = dw.
2iπ |w|=ρ wn+1
Recall that for all w ∈ D, the restriction f w |Dr belongs to Lipβ (Dr ), with Hölder
norm at most C. The Cauchy formula implies therefore that the functions Fn are β-
Hölder over Dr with norm at most ρCn . Letting ρ tend to 1, we get that Fn ∈ Lipβ (Dr )

with norm at most C. Therefore, the series N Fn ( · )wn converges absolutely in
Lipβ (Dr ) uniformly on compact sets in D. This proves that the map μ → f μ |Dr
is holomorphic in restriction to any affine complex disc contained in B ε . Therefore
μ → f μ |Dr is holomorphic on B ε(β) .
Remark 4.10. For the linear family of Beltrami differentials w → μw = w.μ, the
derivative at w = 0 of the map w → f w (z) is given by

d w 1 z(z − 1)
f (z)|0 = − μ(ξ ) dveu (ξ ).
dw π C ξ(ξ − 1)(ξ − z)
It is also a classical fact (cf. [11, p. 73]) that on any compact set of C, the integral on the
right is a continuous function of z with modulus of continuity t log 1t . This is consistent
d
with Proposition 4.9 (ii), which implies that the restriction of z → dw f w (z)|0 to any
disc Dr – as well as the restrictions of all the derivatives of higher order – belongs to
Lipβ (Dr ), for all β < 1.
4.2 Continuity properties of derivatives of Hölder functions

The second ingredient in the proof of Theorems 4.3 and 4.1 is the next result of Real
Analysis. It is the key ingredient.
Recall that the rectangle R is the product I × J of two disjoint intervals of R. Let
r > 0 be such that I and J are contained in the interior of the euclidean disc Dr .
Proposition 4.11. Let 0 < α < 1 and let 1 − α/2 < β < 1. Let ε(β) be the constant
provided by Proposition 4.9. Then, if μ ∈ B ε(β) , one has
(1) the distribution L̃(μ) belongs to H α (R), and
(2) L̃(μ) − L̃(0) ≤ Cμ∞ for a constant C independent of μ.
Proof. To prove (i), we need to show that if μ satisfies the hypothesis and if h ∈ C 2 (R),
then

∂ 2h μ

∂ξ ∂ζ log f (ζ ) − f (ξ ) dξ dζ
≤ Chα
R
for a constant C which does not depend on h.
By Proposition 4.9, the restriction f μ |Dr belongs to Lipβ (Dr ). One can approx-
imate f μ by a sequence of C 2 -maps fi : R → C such that the restrictions fi |Dr
converge to f μ |Dr in Lipβ (Dr ). One has

∂ 2h μ μ ∂ 2h
log(f (ζ ) − f (ξ )) dξ dζ = lim log(fi (ζ ) − fi (ξ )) dξ dζ
R ∂ξ ∂ζ i→∞ R ∂ξ ∂ζ

h(ξ, ζ )
= lim fi
(ξ )fi
(ζ ) dξ dζ.
i→∞ R (fi (ζ ) − fi (ξ ))2
We can suppose α < 23 , so that in particular, β < α. Since the β-Hölder norm of
f μ |Dr is bounded independently of μ, due to Proposition 4.9, the same is also true for
the α-Hölder norm of f μ |Dr . Therefore, the α-Hölder norm of k(ξ, ζ ) = (f (ξh(ξ,ζ )
i )−fi (ζ ))
2
is bounded from above in terms of the α-Hölder norm of h: kα ≤ C · hα for a
constant C which is independent of h. Thus (i) will be a consequence of the following
result. Denote by Ir the interval ] − r, r[, which certainly contains I ∪ J .
Proposition 4.12. Let 0 < α < 1 and let 1 − α/2 < β < 1. Let f ∈ Lipβ (Ir ). Then
there is a constant C which depends only on α and β such that for k ∈ C 2 (R), one
has

∂ 2k

f (ξ )f (ζ ) dξ dζ
≤ Ckα f β .

∂ξ ∂ζ
This proposition means that if f ∈ Lipβ (Ir ), the distribution ∂ (f∂ξ(ξ )f (ζ ))

2
∂ζ then
belongs to H α (R) and that its norm is bounded by Cf β . In order to prove this
proposition, we begin with a 1-dimensional version of it. Denote by C 1 (I ) the set of
functions g : R → C which are C 2 and with support contained in I .
Proposition 4.13. Let α

< 1 and 1 − α
< β < 1. Let f ∈ Lipβ (Ir ). Then there is
a constant C which depends only on α
and β such that for k ∈ C 1 (I ), one has

k
(ξ )f (ξ ) dξ
≤ Ckα
f β .

Proof. After adding to k a constant function, we obtain

a C -function g on R whose
1
α-Hölder norm is smaller than 2kα and such that g(ξ ) dξ = 0. We are led to
prove the inequality with g instead of k. Since g(ξ ) dξ = 0, there is a function
G : R → C such that G
(ξ ) = g(ξ ) (the circle R is identified with R/Z). We can

normalize it so that G(ξ ) dξ . Choose also a C ∞ -function χ : Ir → Ir which has

support and is equal to 1 on I . Replacing f by χf does not change the value
compact
of R g (ξ )f (ξ ) dξ . Also χf β ≤ Cf β for a constant C depending only on r.
Recall now the definition of the Zygmund spaces t for t ∈]0, 2[ (these spaces
can be defined for any real value of t but we shall use them only for t ∈]0, 2[).
Definition 4.14 (Zygmund spaces; [19], [20]). Let t ∈]0, 2[. A function H : R → C
is t-Zygmund if it is continuous and satisfies for all ξ and θ ∈ R:
|H (ξ + θ) + H (ξ − θ ) − 2H (ξ )| ≤ A|θ|t .

The space of the t-Zygmund functions which satisfy H (ξ ) dξ = 0 is denoted t (R);
it is a Banach space for the Zygmund norm

H (ξ + θ ) + H (ξ − θ ) − 2H (ξ )
sup |H (ξ )| + sup

.
|θ |t
ξ ξ,θ
When t ∈]0, 1[, the space t (R) equals Lipt (R), and the two norms, Hölder and
Zygmund, are equivalent: we shall use the same notation · t for the Zygmund norm.
When t > 1, it is classical that G ∈ t (R) if and only if G is differentiable and its
derivative G
belongs to t−1 (R) = Lipt−1 (R).

Since g ∈ Lipα (R), G ∈ 1+α (R) and G1+α

≤ Cgα for a constant C which
depends only on α
.
One has

g (ξ )f (ξ ) dξ = G (ξ )f (ξ ) dξ = − G(ξ )f (ξ ) dξ,
R R R
where denotes the Laplace–Beltrami operator on R, identified with R/Z, i.e. simply
G = −G
. Like on any compact manifold, the operator is a self-adjoint operator

densely defined on L (R) and positive ( G(ξ )G(ξ ) dξ > 0 for any C ∞ function G).
2
This allows us to define, for any real number γ , the fractional power γ , as a positive
self-adjoint operator. It is a classical theorem of Real Analysis that the fractional power
γ acts on the Zygmund spaces of exponent t by shifting the exponent t: precisely if
t − 2γ > 0, then γ maps isomorphically t (R) to t−2γ (R) (cf. [19, p. 145], [20]).
The hypothesis on α
and β implies the existence of a γ such that 1 + α
− 2γ > 0
and β − 2(1 − γ ) > 0. Since 1−γ is self-adjoint

G(ξ )f (ξ ) dξ = γ G(ξ )1−γ f (ξ ) dξ.
R R
As
γ G∞ ≤ γ G1+α
−2γ ≤ Cγ gα
and
1−γ f ∞ ≤ 1−γ f β−2(1−γ ) ≤ C1−γ f β ,


one has
R g
(ξ )f (ξ ) dξ
≤ C · gα
f β for a constant C which is independent of
f and g.
To prove Proposition 4.12, we write

2
∂ k(ξ, ζ ) ∂ ∂k
f (ξ )f (ζ ) dξ dζ = f (ζ )dζ f (ξ ) dξ. (4.2)
R ∂ξ ∂ζ I ∂ξ J ∂ζ
For any ξ ∈ J , the function k(ξ, · ) is in C 2 (J ) and its α-Hölder norm is bounded by
kα . The β-Hölder norm of f (ξ, · ) is also bounded by f β . Therefore, applying
Proposition 4.13 with α
= α and using that β > 1 − α2 > 1 − α, one obtains, for all
ξ ∈I

∂k(ξ, ζ )

f (ζ )dζ
≤ Ckα f β .

∂ζ
J
To be able to apply Proposition 4.13, but to the variable ξ , we shall use the following
Claim 4.15. Let k ∈ Lipα (I × J ). Then for any ξ ∈ I , the restriction k(ξ, · ) belongs
α α
to Lip 2 (J ) and the map ξ → k(ξ, · ), from I to Lip 2 (J ), is α2 -Hölder with norm
smaller than 2kα .
Proof. Recall that the distance on I × J we work with is d((ξ1 , ζ1 ), (ξ2 , ζ2 )) =

sup(|ξ2 −ξ1 |, |ζ2 −ζ1 |). We need to show that for any two points (ξ1 , ζ1 ) and (ξ2 , ζ2 ) ∈
I × J with ζ1 = ζ2 , one has

(k(ξ2 , ζ2 ) − k(ξ1 , ζ2 )) − (k(ξ2 , ζ1 ) − k(ξ1 , ζ1 ))

≤ 2kα |ξ2 − ξ1 | α2 . (4.3)

α
|ζ2 − ζ1 | 2
If |ξ2 − ξ1 | ≤ |ζ2 − ζ1 |, then
|(k(ξ2 , ζ2 ) − k(ξ1 , ζ2 )) − (k(ξ2 , ζ1 ) − k(ξ1 , ζ1 ))| ≤ 2kα |ξ2 − ξ1 |α
α α
≤ 2kα |ζ2 − ζ1 | 2 |ξ2 − ξ1 | 2 ,
from which Inequality 4.3 follows.
If |ζ2 − ζ1 | ≤ |ξ2 − ξ1 |, then
|(k(ξ2 , ζ2 ) − k(ξ1 , ζ2 )) − (k(ξ2 , ζ1 ) − k(ξ1 , ζ1 ))|
= |(k(ξ2 , ζ2 ) − k(ξ2 , ζ1 )) − (k(ξ1 , ζ2 ) − k(ξ1 , ζ1 ))|
≤ 2kα |ζ2 − ζ1 |α
α α
≤ 2kα |ξ2 − ξ1 | 2 |ζ2 − ζ1 | 2 ,
proving the claim in that case also.
α
Now, Claim 4.15 and Proposition 4.13 applied with α
= 2 imply Proposition 4.12.
As we said before, Proposition 4.12 implies Proposition 4.11 (i). It implies also (ii),
using the same argument and since for μ ∈ B ε(β) , one has (f μ −Id)|Dr β ≤ Cμ∞
(Proposition 4.9).
4.3 Proof of Theorems 4.3, 4.1 and 4.2

Proof of Theorem 4.3. To prove Theorem 4.3 we shall first construct, for any point
[μ0 ] ∈ T () a neighborhood N[μ0 ] ⊂ QF () such that, if L̃ denotes the map
constructed in Corollary 4.6, the restriction L̃|N[μ0 ] has image contained in H α (R)
and is a holomorphic map from N[μ0 ] to H α (R).
We first consider the case when [μ0 ] = [0]. The proof in that case uses the
following result.
Proposition 4.16. Let R be a rectangle contained in Gf . Let α < 1 and β > 1 − α2 ;

let ε(β) be the constant provided by Proposition 4.9. Then the restriction of L to
B ε(β) ∩ T () extends to a holomorphic map L̃ : B ε(β) → H α (R).
Proof. Corollary 4.6 provides an extension of L. It is the map L̃ : B 1 () → D(R)

defined by
∂ 2 log(f μ (ζ ) − f μ (ξ ))
L̃([μ]) = ,
∂ξ ∂ζ
where the derivative is taken in the sense of distributions. By Proposition 4.11, if μ ∈
B ε(β) , then L([μ]) ∈ H α (R). We are going to show that this map L̃ is holomorphic.
Let w → μw be an affine disc w → μ0 + wμ1 parameterized by the disc D with
image contained in B ε(β) . Let h ∈ C 2 (R). For all w ∈ D and r such that |w| < r < 1,
one can write:

∂ 2h
L̃([μw ]), h = log f w (ζ ) − f w (ξ ) dξ dζ
R ∂ξ ∂ζ

∂ 2h 1 1 z z

= log f (ζ ) − f (ξ ) dz dξ dζ
R ∂ξ ∂ζ 2iπ |z|=r z − w

1 1 ∂ 2h
= log f z (ζ ) − f z (ξ ) dξ dζ dz
2iπ |z|=r z − w R ∂ξ ∂ζ

1 1
= L̃([μz ]), hdz,
2iπ |z|=r z − w
where the second equality comes from the fact that log(f w (ζ )−f w (ξ )) is holomorphic
in w and the third from the fact that it is a continuous function of the three variables w, ξ
and ζ . By Proposition 4.11, the norm in H α (R) of L̃([μz ]) is bounded independently
of z ∈ D. Therefore one has
|L̃([μw ]) − L̃([μw
]), h| ≤ C|w − w
|hα
when |w| and |w

| are smaller than r0 < r < 1; here C is a constant which only
depends on r0 , r and β. It follows that w → L̃([μw ]) is continuous as a map
from D to H α (R). In particular, the integral |z|=r z−w
1
L̃([μz ])dz is well defined as
α

an integral in H (R) and one has L̃([μw ]) = |z|=r z−w L̃([μz ])dz. It follows that
1
w → L̃([μw ]) is holomorphic on D. Therefore the map L̃ : B ε(β) → H α (R) is

holomorphic.
Fix β with 1 − α2 < β < 1.
The harmonic chart identifies a neighborhood of [0] in QF () with B1 (), the
unit ball in the space of the harmonic Beltrami differentials for . We denote by B r
and Br the balls of radius r in B() and B() respectively. Using Proposition 4.16,
one can take as neighborhood of [0], N[0] = Bε(β) (), where ε(β) is the constant
provided by Proposition 4.9. To construct the neighborhood N[μ0 ] of an arbitrary
point [μ0 ] ∈ T (), we use the change of basepoint map (cf. Definition 2.7) as in the
proof of Theorem 3.12. This map is a homeomorphism [ν] : T () → T ( μ0 ) which
maps [μ0 ] ∈ T () to [0] ∈ T ( μ0 ). We noticed during the proof of Theorem 3.12
the relation L = H(f μ0 )Lμ0 [ν]. Let δ > 0 be such that (f μ0 )−1 is δ-Hölder: this
implies that H(f μ0 ) is a continuous linear map from H αδ (F μ0 (R)) to H α (R). Let
β
> 1 − αδ ˜μ : B ε(β
) → H αδ (F μ0 (R)) is holomorphic.
2 . By Proposition 4.16, L
The formula for ν given in Definition 2.7 shows that [ν] is holomorphic. It follows

that if one sets N[μ0 ] = [ν]−1 (Bε(β ) ), then L̃ : N[μ0 ] → H α (R) is holomorphic.
To finish the proof of Theorem 4.3, it suffices to observe that if two holomorphic
extensions L̃1 and L̃2 of L are defined on a connected open neighborhood U of
[μ0 ] ∈ T (), then they are equal: L̃1 = L̃2 . This fact, which is obvious when is
cocompact – since then B() is a complex space of dimension n and T () totally real
of real dimension n – can be seen in general using the local charts for the complex
structure of QF (). Recall the isomorphism B1 ( μ0 ) B1T ( μ0 ) × B1T ( μ0 ) (cf.
Definition 2.14). Therefore we can suppose that U is a connected open neighborhood
of (0, 0) ∈ B1 ( μ0 ) × B1 ( μ0 ); the intersection U ∩ T () is then the set of pairs
(μ1 , μ2 ) ∈ U with μ2 = μ1 . Any point (μ1 , μ2 ) ∈ B1T ( μ0 ) × B1T ( μ0 ) belongs to a
C-affine plane which is the complexification of an R-affine line contained in T ( μ0 ):
precisely (μ1 , μ2 ) sits in the affine plane which is the complexification of the real
line passing through (μ1 +μ22,μ2 +μ1 ) , directed by the vector i(μ1 − μ2 , μ2 − μ1 ).
In particular, for ε > 0 sufficiently small, any pair (μ1 , μ2 ) ∈ U with μi ∞ ≤ ε
belongs to such an affine disc which is contained in U. The definition of a holomorphic
map implies that L˜ = L˜ on the open set {(μ1 , μ2 ) | μi ∞ < ε}. Since U is
connected, L˜ = L˜ on U.
The open neighborhood N α can now be constructed easily.
Remark 4.17. It follows from the proof that the image L̃(B ε(β) ) is bounded in H α (R):
its diameter is bounded in terms of α, β, and of the size of the rectangle R = I × J ,
i.e. by an upper bound on the lengths of the intervals I , J and upper and lower bounds
on the distance between them.
Proof of Theorem 4.1. Recall the notations introduced during the proof of Theo-
rem 3.12. Given a rectangle R ∈ R, not necessarily contained in Gf , q ∈ PSL(2, R)
is the map sending the reference rectangle R0 to R. Recall that R0 is a rectangle
contained in Gf , so that Theorem 4.3 applies to R0 . For μ ∈ B 1 (), q μ ∈ PSL(2, R)
is the element such that the quasiconformal homeomorphism φ μ = q μ f μ q is
normalized. Equation 3.1, for any h ∈ LipαR (G), one has

L([μ]), h = (h Q) (μ )−1 dλ = L([q ∗ (μ)]), h Q.
G
By the definition of L̃, one also has L̃(μ), h = L̃(q ∗ (μ)), h Q. Since the
neighborhood N[0] constructed in the proof of Theorem 4.3 is invariant under the map
μ → q ∗ (μ) for all q ∈ PSL(2, R), one has |L̃([μ]), h| ≤ Ch Qα the constant
C being independent of h and R ∈ R.
By definition of the space of geodesic Hölder
distributions, this means that L̃ N[0] is contained in H α (). By the remark following
the proof of Theorem 4.3, this image is also bounded. Therefore, the argument used
for proving Proposition 4.16 shows that L̃ : N[0] → H α () is holomorphic.
For a general point [μ0 ] ∈ T (), one writes: φ μ = q μ f μ q = f νq (μ) f μ0 (q) ,
with f νq (μ) = q μ f ν(μ) (q μ0 )−1 and f μ0 (q) = q μ0 f μ0 q. Therefore
∗ −1
L̃ q (μ) , h Q = L̃ νq (μ) , h Q F μ0 (q) . (4.4)
−1
Now, since q ∗ (μ)∞ = μ∞ the homeomorphisms f μ0 (q) are δ-Hölder with
the same exponent δ and with a δ-Hölder norm which
is bounded
independently
q.
of
This implies that the norms of the linear maps H (f μ0 (q) )−1 : H α F μ0 (q) (R0 ) →
H αδ (R0 ) are bounded independently of q. Since the homeomorphisms f μ0 (q) form
an equicontinuous family, the shapes of the rectangles F μ0 (q) (R0 ) are also bounded
independently of q. Let β
> 1− αδ
2 . The Beltrami coefficient νq (μ) can be computed
using the formula from Definition 2.7. It comes out that there is η(β
) > 0 which
depends only on β and on μ0 ∞ such that if μ∞ ≤ η(β
), then νq (μ)∞ ≤ ε(β
)
for all q. Therefore, using Theorem 4.3 and the remark following it, we obtain that
if μ − μ0 ∞ ≤ η(β
), then the term on the right in the equation 4.4 is bounded in
absolute value by Ch Qα , where the constant C does not depend on q nor on h.
Let N[μ0 ] be the neighborhood of [μ0 ] ∈ QF () which is {[μ] | μ∞ ≤ η(β
)}.

Then L̃ N[μ0 ] is bounded in H α (). Like above, one shows that L̃ is holomorphic
on N[μ0 ] . The neighborhood N α can now be constructed as in Theorem 4.3.
Proof of Theorem 4.2. By Theorem 4.1, L is the restriction of a holomorphic map

from a neighborhood of T (). Its restriction to T () is therefore analytic [9].
References
[1] L. V. Ahlfors, Lectures on quasiconformal mappings. Van Nostrand Math. Stud. 10, D. Van
Nostrand Company, Inc., Princeton 1966. 225
[2] L. V. Ahlfors, Collected papers. Vol. 2. Birkhäuser, Boston 1982. 225
[3] L. Ahlfors and L. Bers, Riemann’s mapping theorem for variable metrics. Ann. of Math.
(2) 72 (1960), 385–404. 225, 227
[4] L. Ahlfors and G. Weill, A uniqueness theorem for Beltrami equations. Proc. Amer. Math.
Soc. 13 (1962), 975–978. 228
[5] L. Bers, Correction to “Spaces of Riemann surfaces as bounded domains”. Bull. Amer.
Math. Soc. 67 (1961), 465–466. 227
[6] F. Bonahon, Bouts des variétés hyperboliques de dimension 3. Ann. of Math. (2) 124
(1986), 71–158. 229
[7] F. Bonahon, The geometry of Teichmüller space via geodesic currents. Invent. Math. 92
(1988), 139–162. 229
[8] F. Bonahon and Y. Sözen, Variation of the Liouville measure of a hyperbolic surface.
Ergodic Theory Dynam. Systems 23 (2003), 729–758. 229, 232, 233
[9] N. Bourbaki, Topologie differentielle. Fascicule de résultats, Hermann, Paris 1967. 237,
247
[10] A. Douady and C. J. Earle, Conformally natural extension of homeomorphisms of the
circle. Acta Math. 157 (1986), 23–48. 237
York 1987. 225, 227, 228, 241
[12] F. P. Gardiner and N. Lakic, Quasiconformal Teichmüller theory. Math. Surveys. Monogr.
76, American Mathematical Society, Providence, R.I., 2000. 225, 227, 228, 240
Tokyo 1992. 225
[14] T. Iwaniec and G. Martin, Geometric function theory and non-linear analysis. Oxford
Math. Monogr., Oxford University Press, Oxford 2001. 225
[15] J.-P. Otal, Riemann surfaces and Theta operators, Quaderns 25; http://www.crm.es/
Publications/Quaderns/Quadern25-2.pdf. 225
[16] W. Rudin, Real and complex analysis. 3rd ed., McGraw-Hill Book Co., New York 1987.
230, 240
[17] D. Šarić, Infinitesimal Liouville distributions for Teichmüller space. Proc. London Math.
Soc. (3) 88 (2004), 436–454. 229, 232, 233
[18] D. Šarić, Geodesic currents and Teichmüller space. Topology 44 (2005), 99–130. 229,
232, 233, 234
[19] E. M. Stein, Singular integrals and differentiability properties of functions. Princeton
Math. Ser. 30, Princeton University Press, Princeton, N.J., 1970. 243
[20] H. Triebel, Theory of function spaces. II. Monographs Math. 84, Birkhäuser, Basel 1992.
243
Chapter 5
Teichmüller spaces, triangle groups and

Grothendieck dessins
William J. Harvey
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
2 Historical overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
3 Grothendieck dessins and Thurston’s examples . . . . . . . . . . . . . . . 255
3.1 Dessins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
3.2 Clean dessins and Galois covers of P1 . . . . . . . . . . . . . . . . . 255
3.3 Thurston’s examples . . . . . . . . . . . . . . . . . . . . . . . . . . 256
4 Teichmüller theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
4.1 Deformations of complex structure on a surface . . . . . . . . . . . . 258
4.2 Teichmüller spaces and modular groups . . . . . . . . . . . . . . . . 260
4.3 Families of Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . 263
4.4 Hyperbolic surfaces inside moduli space . . . . . . . . . . . . . . . . 265
4.5 Example: a T-disc parametrising tori . . . . . . . . . . . . . . . . . . 267
4.6 Deformation T-discs embed . . . . . . . . . . . . . . . . . . . . . . . 268
4.7 Veech’s examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
4.8 Hecke triangle groups and polygonal surface tilings . . . . . . . . . . 270
4.9 Fermat curves represented by T-discs . . . . . . . . . . . . . . . . . . 272
4.10 Regular tesselations and smooth coverings . . . . . . . . . . . . . . . 273
4.11 Return to Thurston’s examples . . . . . . . . . . . . . . . . . . . . . 274
5 Triangle groups and modular families . . . . . . . . . . . . . . . . . . . . 275
5.1 Tesselations and triangle groups . . . . . . . . . . . . . . . . . . . . 275
5.2 Associated Teichmüller families . . . . . . . . . . . . . . . . . . . . 276
5.3 Recent work on Veech curves in Mg . . . . . . . . . . . . . . . . . . 281
5.4 Completion of a T-disc with large stabiliser . . . . . . . . . . . . . . 282
5.5 Consequences; complete curves in modular varieties . . . . . . . . . 283
6 Towards a space of all Q-curves: a universal modular family . . . . . . . . 284
6.1 A moduli space of Fuchsian groups . . . . . . . . . . . . . . . . . . . 284
6.2 A representation of the absolute Galois group . . . . . . . . . . . . . 285
6.3 Modular forms with closed trajectories . . . . . . . . . . . . . . . . . 286
6.4 Arithmetically defined curves in moduli space . . . . . . . . . . . . . 288
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
250 William J. Harvey
1 Introduction
The theory of moduli for algebraic curves began with Riemann’s famous papers on
abelian functions, but it took many years before a further ingredient, Teichmüller’s
ideas on how to deform the underlying topological surface so as to effect a change in
the complex analytic structure, produced a rigorous formulation and construction of
the variety of moduli Mg , whose points represent bijectively each isomorphism class
of non-singular curve of the given genus.
We begin by describing in outline the theory of moduli for complex algebraic
curves. The special case of genus 1 is discussed briefly, both as an avatar and in order
to describe some special types of object to be used later, classical modular forms and
Riemann surfaces, but we concentrate on the curves of genus at least 2, whose Riemann
surface is uniformised by the hyperbolic plane, using the elaborate machinery of
Teichmüller spaces introduced by Ahlfors and Bers. A method of construction is then
given for explicit complex analytic models within the metric structure of these moduli
spaces, which exhibits through the medium of hyperbolic plane geometry a large class
of curves including those with defining equation over some algebraic number field.
These Teichmüller disc models of surfaces have achieved much prominence recently
through work of W. Veech and others on dynamical properties exhibited by the set
of closed geodesics on the surface, relating to interval exchange transformations and
billiard trajectories on Euclidean polygons. Here we focus on the symmetry properties
of the complex curves so defined and the existence of Belyi structures on them; in a
later section we show that there is a collection B which comprises
a union of complex
affine algebraic curves contained in the moduli space M = Mg of all (conformal
classes of) compact Riemann surfaces (viewed as hyperbolic 2-dimensional orbifolds)
which has an intrinsic arithmetical aspect. The space B is a (reducible) analytic variety
on which the absolute Galois group of the algebraic numbers Gal(Q) acts in a natural
way.
To be more precise about the framework of the family Bg = B ∩ Mg , each mem-
ber (irreducible component) is an immersed complex affine curve in some modular
variety Mg of nonsingular curves of genus g > 1, indexed by a central base point
and unit tangent deformation vector, and the whole collection forms part of the tau-
tological geometric structure inherited from the covering Teichmüller space and the
holomorphic vector bundle 2 (Vg ) of holomorphic quadratic forms on the universal
Teichmüller-family Vg of Riemann surfaces. Veech refers to this structure, or its unit
sphere subbundle, as the Teichmüller geodesic flow. The precise relationship between
the space Bg and the Q-rational points of the moduli spaces Mg , Cg is unknown, but
we show in Section 6 that for each curve in Bg , after completion within the Deligne–
Mumford compactification M̂g , there is a ramified covering (with degree bounded
above by a linear function of g) which belongs to the conformal isomorphism class of
compact Riemann surface defined by the central point.
This survey is based on a revised version of the preprint [26]. There is little that
is original here, beyond the selection of material and point of view taken, which
Chapter 5. Teichmüller spaces, triangle groups and Grothendieck dessins 251
emphasizes wherever possible the hyperbolic geometry and group theory. However,
later results presented without attribution are the author’s responsibility.
2 Historical overview
The study of algebraic curves and their deformations has a venerable history, begin-
ning with specific investigations of simple types of curve, and progressing to very
sophisticated methods of projective geometry. With Riemann’s introduction of the
underlying topological surface and his spectacular results on theta functions and
periods of abelian integrals (Abel–Jacobi theory) ([57]), it became clear that to
answer the most fundamental questions about a given projective curve, it is neces-
sary to consider variations of the curve, and in order to maintain the link with abelian
integrals one must address the problem of how to vary the curve in all possible ways
which preserve the particular topological structure. Riemann showed that the topo-
logical genus, the number of handles of the Riemann surface of the curve, dictates the
dimension (1 if g = 1, 3g − 3 when g ≥ 2) of the moduli space of local deformations
of the curve within the (complex) projective space in which it lies. However, the
precise structure of the space of moduli remained an enigma until a more compre-
hensive account of the topological nature of surfaces became available, including the
introduction of homotopy as a refinement of the relation of homology for integration
theory along paths. Despite much progress on the analytic side with the proof of
uniformisation and the classification of Fuchsian groups, the efforts of Klein, Fricke
and (independently) Poincaré left the global moduli problem unresolved, and it was
not until the introduction of quasiconformal mappings and their relation to quadratic
differential forms by Teichmüller during the 1930s that a genuine theory of deforma-
tions of Riemann surfaces became possible. This was developed during the period after
1950 by Ahlfors and by Gerstenhaber and Rauch, but it was only after the landmark
Princeton Conference on Analytic Functions of 1957 [56] that the complex deforma-
tion theory of Riemann surfaces took flight, together almost simultaneously with the
Kodaira–Spencer theory for compact complex manifolds in general.
Grothendieck’s introduction [22] of the notion of dessin d’enfant (defined below)
brought into sharper focus the interaction between combinatorial surface theory, the
complex geometry of algebraic curves and their algebraic number coefficient fields: it
implies that the arithmetic field of definition for a curve is in some manner an intrinsic
feature of the hyperbolic geometry underlying the transcendental uniformisation of the
curve. A primary goal in this chapter is to illuminate some aspects of that perception by
systematic use of the theory of Teichmüller spaces, as developed by Ahlfors and Bers;
we shall highlight the relationship between a certain class of holomorphic quadratic
differential forms on the Riemann surface of an algebraic curve (the Jenkins–Strebel
forms) and the existence of a model of the surface (as a quotient of the hyperbolic
plane) within the corresponding modular variety. The totality of these (immersed)
model surfaces is a set on which the absolute Galois group of the algebraic numbers
acts, a pattern which is in accord with the conjectural picture of [22]. It may equally
be viewed as an organised combinatorial framework for the natural actions of the
various mapping class groups and subgroups which permute markings on the various
levels of ramified surface coverings that constitute the branching locus, or the orbifold
singularities, in each Mg .
A Grothendieck dessin or line drawing in a surface is a connected graph drawn
in a compact surface with the property that each complementary piece of surface is
a polygonal disc. During the late 1970s, Grothendieck proposed that there was an
intimate link between this purely combinatorial object and arithmetic geometry of
curves, and pointed out the relationship between dessins and two distinct, previously
unrelated notions: representations of a certain discrete (extended triangle) group on
the one hand and the absolute Galois group G of the field of algebraic numbers on the
other.
In 1979, this insight received clear confirmation when G. V. Belyi proved a fascinat-
ing theorem [5] bringing out the full significance of this idea: If a projective algebraic
curve X is defined by a set of polynomial equations with coefficients in a number field,
an algebraic extension of the rational field Q, then there is a meromorphic function
on X whose only singular values are the three points 0, 1 and ∞. The converse is also
true, following indirectly from older work of A. Weil on the field of definition for an
algebraic variety; for this aspect of the Belyi theorem, the reader may consult recent
articles by J. Wolfart [74] and G. Gonzalez-Díez [18].
Grothendieck later wrote [22] an extended account of his ideas as part of a CNRS
research proposal, placing it in an elaborate conjectural theory of ‘anabelian’ fun-
damental groups in geometry. This celebrated manuscript has received widespread
unofficial distribution; subsequently, thanks partly to an influential article by G. Shabat
and V. A. Voevodsky [64] in the Grothendieck Festschrift, the notion of dessin gained
a wider audience and established itself as an independent entity on the fringe of three
broad estates – algebraic geometry, number theory and complex analytic geometry.
Each of these areas demands heavy investment in basic language and background
theory as prerequisites for understanding and progress, and so this insertion of a new
elementary structural connection between them has been widely welcomed.
To display the link, we focus on the relationship between a subclass of dessins and
a special kind of deformation of the Riemann surface structure. The latter arises from
the initial data of a complete complex algebraic curve, carrying the dessin as part of
its uniformisation in the hyperbolic plane, and an additional choice of prescribed flat
geometric pattern coming via the Belyi ramified covering from the Euclidean plane.
These two pieces of data determine a complex affine deformation curve (Riemann
surface with punctures), immersed in the corresponding modular variety and passing
through a central point which represents the base point of the deformation, the compact
Riemann surface associated with the complete curve. It will follow from the basic
link between automorphic forms and holomorphic deformations that any point of a
modular variety which represents a curve definable over Q belongs in an appropriate
sense to at least one such family. As seen from the perspective of Belyi’s theorem,
a primary source of this type of structure comes from curves with sufficiently large
automorphism group. Simple examples of familiar curves falling into this category
include Wiman’s class of hyperelliptic curves with affine equation y 2 = 1 − x n , for
n ≥ 5, the famous Klein quartic given (in projective form) by x 3 y + y 3 z + z3 x = 0
and the Fermat curves x n + y n = zn , for n ≥ 4.
The theory of holomorphic families of compact Riemann surfaces is founded on
the analytic properties of the Teichmüller spaces Tg,n established by Lars Ahlfors and
Lipman Bers during the decade following 1960; these spaces are a countable set of
connected complex manifolds, one for each pair of non-negative integers g, n with
2g − 2 + n > 0, whose points represent marked complex structures on an n-pointed
genus g surface up to holomorphic equivalence. There is a small list of special low
values of g, n which represent surfaces and deformation spaces familiar from classical
work: T0,n is a single point for n = 1, 2, 3 and T0,4 = U = T1,0 is the upper half
plane. In this way, the totality of compact Riemann surfaces (non-singular curves)
with n-point subsets is organised into a union of disjoint spaces representing distinct
topological types. Each Teichmüller space carries intrinsically a structure of complex
manifold of dimension 3g−3+n and a complete global metric, defined by Teichmüller
as the logarithm of the least overall conformal distortion (measured in the sup norm)
involved in deforming one complex structure on a surface to another. If the surface
under consideration has genus 1, the deformation space is equivalent to the classical
identification of (conformal classes of) marked tori, X = C/L, with points of the upper
half plane, U = T1 , on which the modular group (1) = SL(2, Z) acts by fractional-
linear transformations, in effect changing the basis elements of the lattice L ∼
= π1 (X),
and preserving the Poincaré metric. Thus, in this special case, the deformation space
carries an intrinsically richer geometric structure of hyperbolic plane geometry and
we sometimes denote it by H 2 when focussing on this viewpoint. In general, the
mapping class group Modg,n = π0 Diff(X) of the (n-punctured) surface X acts as the
Teichmüller modular group by changing the marking on a reference surface. When
n = 0, the quotient of Teichmüller space Tg by this action of Modg is Riemann’s
moduli space Mg , which parametrises biholomorphic equivalence classes of (closed,
genus g) Riemann surfaces, while if n = 1 the resulting quotient is the modular curve
Cg , the total space of the fibred modular family of genus g surfaces.
Teichmüller also defined a special type of deformation within the spaces Tg , which
is now called a Teichmüller (geodesic) disc; we sometimes refer to it as a T-disc. They
are described in more detail in Section 4: for the purposes of this outline, we mention
only that inside any Tg,n they form a natural class of complete complex submanifold
of dimension 1, isometric to the Poincaré metric model of the hyperbolic plane H 2 ,
and that they are in plentiful supply, through each point and in any given direction,
forming an integrated part of the Teichmüller geodesic flow.
It is natural to consider the action of the mapping class group g,n on the set of all
T-discs in Tg,n . Because the modular action is biholomorphic and discontinuous, it
preserves the set of T-discs and the stability subgroup preserving a given disc is then
isomorphic to a Fuchsian discrete group of hyperbolic isometries. It is not hard to

show (see [13] for instance) that a sufficiently general T-disc will have trivial stabiliser;
but how large can the stability group of a T-disc be? One might naturally look first
for examples with a cocompact Fuchsian group as stabiliser, since that type of family
would be of interest for both the topology of the modular variety and the algebra
of the mapping class group, and it would provide a valuable tool for constructing
complete families of Riemann surfaces. The search is fruitless, however, since any
T-disc contains geodesic rays whose projection is divergent in the moduli space [47],
which implies at once that the quotient of a T-disc by its stabiliser cannot be compact.
Thus, there is no totally geodesic, complete, complex suborbifold in any of the higher
dimensional moduli spaces Mg,n , and one must use quite different methods to produce
complete curves in the modular varieties of closed surfaces Mg (cf. [38], [19]).
The situation with stabilisers of T-discs becomes more interesting if the compact
quotient condition is relaxed. In fact, as we explain in Section 5, in any given Tg there
is a large set of marked surfaces [Sn ] which define centre points of T-discs stabilised by
(noncompact) finite volume Fuchsian groups, so that in this case the quotient in moduli
space Mg is a finite-area immersed Riemann surface, isomorphic to a quotient of the
base surface Sn by some subgroup of its automorphism group. They are examples of
a more general class of affine algebraic curve in moduli space known as a Veech curve
to be discussed briefly later; all of the examples we describe are given by algebraic
equations defined over some number field, which suggests that the geometric action of
the Teichmüller modular group on the unit tangent bundle over the Teichmüller space
will form a distinctive part of the relationship conjectured in [22] between the absolute
Galois group G(Q) and the Grothendieck–Teichmüller tower, once the meaning of that
object has been clarified.
For higher dimensional families of algebraic curves, there are interesting but spo-
radic results deriving from work of many authors, among them E. Picard (and his
students), G. Shimura (and his students), R. Holzapfel, P. Deligne and G. D. Mostow,
Paula B. Cohen, J. Wolfart and H. Shiga; often these employ some version of the
period mapping for holomorphic 1-forms and its monodromy and have some relation
to the present work, which we shall not be able to explore here.
In the next two sections, we shall describe essential background and the examples
which motivate our viewpoint. The fourth describes a procedure for constructing one
type of Veech–Teichmüller curve in moduli space. In the last chapter, we consider the
arithmetic problem of characterising the curves of given genus g that are definable
over Q. We work within the convenient topological framework of a single space of
moduli D0 , the space of all cocompact Fuchsian groups, viewed modulo conjugacy
within the Lie group G = PSL(2, R). There is a natural structure of metric space on
D0 , using the Hausdorff metric on the space of closed subgroups of the Lie group G,
and by the uniformisation theorem it contains copies of every modular variety Mg,n
with 3g −3+n > 0. A related question asks for a precise description of all Q-curves in
a suitable compactification of the space of stable genus g curves M̂g . In emphasising
the totality of hyperbolic discs with crystallographic (i.e. discrete co-finite) Fuchsian
stabilisers, we hope to bring out a general perspective which places dessins within the
tautological geometric framework underpinning the patchwork collection of Teich-
müller spaces which cover the various modular families and which in the process
generate the space D0 . At the same time one needs the entire collection of lattices
(finite co-volume Fuchsian groups) and Teichmüller space inclusions to reflect the
wide range of possible Belyi representations of complex curves defined over Q: every
meromorphic function which has 3 ramification values determines a covering of the
complex projective line with the requisite properties and each admissible cusp form
of weight 4 produces a Teichmüller disc and, thereby, candidate members of the set
of Q-points of the space of moduli. This space may provide an appropriate starting
point for the (still incomplete) process of constructing the ultimate arithmetic modular
family described in the Grothendieck Esquisse [22].
3 Grothendieck dessins and Thurston’s examples
3.1 Dessins
A dessin d’enfant or simply dessin is defined to be a connected (finite) graph in a
compact surface, whose complement is a union of cells and which has a bipartite
structure on the vertices. It is convenient to denote the labels for vertices of the graph
with • and .
To keep the exposition simple, we restrict attention to orientable surfaces and
to a special type of dessin, concentrating on the subclass of dessins which arise from
pulling back the standard triangulation of CP 1 through a Belyi function, a holomorphic
branched covering mapping β : X → CP 1 with three critical values 0 (= •), 1 (= )
and ∞ (= ). This consists of a topological decomposition of the Riemann sphere
into two triangles U and L, with interiors the upper and lower halves of the complex
plane respectively, with vertices given by the above three symbols and disjoint edges
joining them pairwise along the real axis. In addition, we shall often assume some
regularity property for the function β, to be spelled out in the next subsection; in an
appropriate sense, the restricted class of dessins to be used forms a natural cofinal
subclass of the general pattern. For a broader study, the reader may consult a range
of recent survey articles ([33], [61]) and the chapter by Herrlich and Schmithüsen in
Volume II of this Handbook [29].
3.2 Clean dessins and Galois covers of P1

A Belyi function f : X → CP 1 is called clean if every point in the inverse image of
has ramification of order 2. The corresponding dessin in X, defined as
Kf = f −1 {•—},
is then called clean if each vertex mapping to (and labelled) has valency 4 in the
graph Kf .
It is not difficult to see that every dessin has a standard subdivision for which this
property holds: for instance, one can take the triangular subdivision (see [64]) which
arises by replacing the function f by f˜ = 4(f − f 2 ).
The transition from combinatorial surface topology to complex analysis and the
modular group is made by the classical representation of the punctured plane C −{0, 1}
as the congruence modular surface of level 2, X(2) = U/ (2). One finds with the
help of the monodromy theorem that the unbranched covering mapping f ∗ , obtained
by puncturing X at all labelled vertices, is determined by a finite index subgroup
of (2), for which f is just the projection of each -orbit to the corresponding (2)-
orbit. Consequently there is a uniformised picture of the dessin in the upper half plane
which arises as a quotient under the group of the cell subdivision of U into ideal
fundamental triangles for (2) – for more details here, consult [33] for instance.
To gain insight into the picture in the hyperbolic plane, it is helpful to begin from the
standard triangular fundamental domain for the full modular group, which determines
the barycentric subdivision of the (2)-invariant triangulation.
In the context of the analysis of general (orientable) dessins in [22], one considers
instead permutation representations of the cartographic group C2+ , which is isomor-
phic to the congruence subgroup 0 (2) containing (2) with index 2. This group
consists of all projective transformations
az + b
γ : z → ∈ (1) = PSL(2, Z) with c ≡ 0 (2).
cz + d
It is not hard to see, using a standard choice of fundamental domains for the groups
concerned, that passage from f to f˜ is defined by extending the classifying inclusion
< (2) to < 0 (2), so that there is again a tesselation of U by hyperbolic
triangles which determines a (clean) dessin on the surface X.
A dessin is called balanced when both (a) every vertex has the same valency
and (b) each cell has the same number of edges. A Galois dessin is a graph-and-
cell decomposition obtained by lifting the standard division of CP 1 through a Belyi
mapping which is also a Galois covering.
It is a simple exercise in finite group actions to see that Galois dessins are balanced,
but the converse is not necessarily true. These matters are discussed at greater length
in the collection of articles [61].
3.3 Thurston’s examples
We use the term Thurston decomposition of a compact surface S to refer to a cell

decomposition of S determined by two loops which fill it up, partitioning the surface
into polygonal cells with an even number of edges, which are segments coming from
each loop alternately.
Thurston’s original 1975 construction [69] was motivated by his work on the clas-
sification of surface mappings up to free homotopy, and in particular by the search
for a counterexample to an old conjecture of Jakob Nielsen that an algebraically
finite mapping class, i.e. one induced by a surface mapping for which all homological
eigenvalues are roots of unity, must be periodic. To construct examples of the kind of
map which contradicts this, he took a certain product of twists along a pair of loops
α, β that fill up the surface: this means that the loops, taken to be in general position
with no triple intersection points, are such that the complementary regions of S are
all simply connected cells with at least four edges. This type of decomposition by
two loops determines a dessin on S with the property that all vertices of type have
valency 4: this is achieved by placing a vertex in each cell and an edge to it from
each vertex of the graph. In fact this procedure also yields a dual partition of the
surface into quadrangles, each with two vertices labelled , which is a subdivision of
the cell partition of S dual to the original one; omitting the • vertices gives the dual
quadrilateral structure.
The method used by Thurston to produce pseudo-Anosov homeomorphisms from
this combinatorial structure is simple but very ingenious: because the surface S has
been decomposed into quadrilaterals, it has a branched piecewise Euclidean structure
given by the requirement that each cell is to be a unit square. This determines a structure
of compact Riemann surface too, by the standard removal of singularities at corners,
but the important point is that in the Euclidean structure the Dehn twist mappings along
each loop act as translation by an integer. Indeed, one can approach the construction
of the surface in reverse from a set of square tile pieces by first joining them in line
following the path α and then making the necessary identifications of edge segments
on the resulting cylinder; in the local coordinate of the resulting geometric structure,
a twist along the core of the cylinder is clearly an integral translation. For the β loop,
one follows a similar path of identifications and again we obtain as monodromy of the
geometric structure around the loop an integer translation in the orthogonal direction.
Now the group generated by the two twists is a subgroup, often of finite index, in
the modular group SL(2, Z). Furthermore, it is clear that any word in the two twist
mappings that gives a modular group element with two real eigenvalues (a hyperbolic
element) will define a pseudo-Anosov mapping class for the surface; with a more
careful choice of word, the mapping can be arranged to act trivially on homology (see
[69] for more details).
In this construction we see a first glimpse of how dessins and the mapping class
group interact, although the complex geometry of moduli space is not yet in evidence.
4 Teichmüller theory
This section presents the background theory of complex deformations of curves, mostly
without proofs but with some effort to give at least an intuitive grasp of the results
which are important. To this end, we also include some specific examples of the kind
of extremal deformation which is to be highlighted.
4.1 Deformations of complex structure on a surface

A convenient framework for studying analytic moduli of surfaces with all shapes and
sizes in an integrated manner is provided by the universal Teichmüller space developed
by Ahlfors and Bers; we provide a brief sketch only, sufficient to make it possible to
give a unified approach to the kind of deformation which we need in the last section.
Detailed accounts may be found in several textbooks [16], [54], [23].
The first fundamental problem encountered in deformation theory is the question
of specifying precisely what kind of structure can reasonably be used. We require our
object (which is to be deformed) to have no local symmetry, in the sense that there is
no object nearby that is isomorphic to it: this is local rigidity. In the case of Riemann
surfaces, it turns out that one needs, in addition to the categorical object concerned,
the additional data of a topological marking, which is defined to be a homeomorphism
f : X0 → X, from a fixed reference surface X0 , viewed up to free homotopy. This is
the appropriate condition to deploy, because rigidity of a Riemann surface X marked
in this sense follows from the fact that holomorphic self-mappings are represented
faithfully in the action induced on the homotopy (or homology) of the surface: a
conformal self-mapping (distinct from the identity) cannot be trivial on homotopy.
One prefers to employ a homotopy marking rather than a homologically defined one
because of the more direct link with covering spaces: also, with hindsight, it turns
out that this provides a clearer view of the often intricate interactions of the groups
involved.
It is a consequence of the theory of Beltrami equations, to be outlined next, coupled
to the uniformisation theorem of Riemann and Koebe, that for every (finite volume)
Riemann surface X with universal covering the upper half plane U, provided with a
marking f : X0 → X, there is a representation of (X, f ) as a normalised homeomor-
phism h : R → R with h(0) = 0, h(1) = 1. Here R ∪ {∞} ∼ S 1 is the boundary
of the upper half plane U and these maps h, normalised to fix ∞, are in fact qua-
sisymmetric homeomorphisms of the boundary real line, arising as boundary values
of quasiconformal self mappings f : U → U which cover the marking of X. All
the essential ingredients of real and complex analysis on the Riemann surface are
transferred to analogues defined on the universal cover p : U → X which transform
suitably under the discrete (Fuchsian) group of deck transformations of the covering,
which is a subgroup of the direct isometry group of the hyperbolic plane. This process
delivers within a single universal space all deformations of all Riemann surfaces.
To explain this in more detail, we make the following definitions.
Definition. A Beltrami coefficient μ on a hyperbolic Riemann surface X, is a measur-

able complex function on the universal covering U transforming under the fundamental
(covering) group as a tensor of type (−1, 1), with sup norm less than 1.
Thus μ is an element of B1 (X), the open unit ball in

L∞ ∼ ∞
−1,1 (X) = {μ ∈ L (U, C) | μ(γ (z))γ (z)/γ (z) = μ(z) for all γ ∈ }.
The corresponding Beltrami equation is a partial differential equation for a locally
homeomorphic complex-valued coordinate function w : X → C:
wz (z) = μ(z)wz . (1)
The Ahlfors–Bers theory (see [3]) provides a unique normalised solution homeo-
morphism wμ to this equation in U, called quasiconformal or simply ‘qc’, which
extends to a homeomorphism h of the boundary ∂U, compatible with the given Fuchs-
ian group . Here, the process of normalisation amounts to factoring out the natural
action on the set of solutions by post-composition with the group G = PSL2 (R) of all
real Möbius transformations, the direct hyperbolic isometries of U: if w is a solution
to the equation , then so is the composition γ w with γ ∈ G. We shall usually
employ the standard normalisation, which requires w to fix 0, 1 and ∞. Using the
above invariance property of the Beltrami form, a computation with partial derivatives
shows that each composition of the form
γμ = wμ γ wμ−1 , with γ ∈ ,
has zero conformal distortion, and it follows that there is a deformed covering group
μ = wμ wμ−1 of real Möbius transformations, quasiconformally conjugate to .
This determines a geometric isomorphism between the two Fuchsian groups, induced
by this qc-homeomorphism between the surfaces, which gives a genuine distortion of
the original complex structure if the map is not conformal.
The crucial notion of Teichmüller equivalence for two Beltrami coefficients is
introduced next; this reduces the infinite dimensional spaces of Beltrami coefficients
to the (finite dimensional) deformation spaces appropriate for these marked structures.
Definition. Two Beltrami coefficients μ and μ are called Teichmüller-equivalent if

the corresponding marked groups μ and μ are conformally equivalent, that is,
conjugate by some real Möbius transformation in G.
This condition, sometimes abbreviated to T-equivalence, implies that the corre-

sponding normalised homeomorphisms h = hμ of R, with h(0) = 0 and h(1) = 1,
depend only on the Teichmüller class of μ. The set of all the normalised maps hμ for
arbitrary measurable μ ∈ B1 (L∞ (U)), without any group compatibility condition,
is the universal Teichmüller space, denoted T (1). It transpires that there is a purely
intrinsic analytic criterion which characterises the homeomorphisms of the boundary
circle R ∼= S 1 that arise in this way: they are called quasisymmetric. See for instance
[16]. The detailed analysis of boundary behaviour need not concern us here, but
becomes highly important for infinite dimensional Teichmüller spaces, arising from
surfaces with more complicated boundary, including curves or infinitely many punc-
tures, where attention switches to asymptotic behaviour. See [17] for details of this
very different aspect of the theory.
Definition. The Teichmüller space of the Fuchsian group (and by abuse of notation,
of the Riemann surface X = U/ ) is the set of T-equivalence classes of Beltrami
coefficients for the group .
An inclusion of Fuchsian groups 1 ≤ 2 induces a natural relationship between

spaces of Beltrami forms B1 (X1 ) ⊇ B1 (X2 ), for the surfaces X1 = U/ 1 and
X2 = U/ 2 , where B1 (X1 ) ∼ = B1 (U, 1 ) ⊇ B1 (U, 2 ) ∼ = B1 (X2 ). This projects to
an inclusion T (X1 ) ⊇ T (X2 ). Thus, in particular, T (1) has the significant universal
property that it contains copies of every Teichmüller space T (). For many purposes,
however, this space is too vast and has too great a diversity of subspaces, and one has
to focus on the individual spaces to gain real insight into the geometry of the modular
families.
4.2 Teichmüller spaces and modular groups
One strongpoint of the Ahlfors–Bers formalism lies in the coordinated description it

gives for the action of the modular groups of the various Fuchsian groups and surfaces.
Recall first the standard (topological) definition of the mapping class group of a
surface (or possibly orbifold) X = U/ as the group of isotopy classes of homeo-
morphisms of X to itself preserving any punctures or cone points; equivalently one
may use homotopy classes instead, since these notions coincide in two dimensions.
For compact surfaces, it is easily shown that each homotopy class of homeomorphism
contains quasiconformal representatives and this does extend to orbifolds and surfaces
with finitely many punctures. We may apply the method of lifting in the theory of
branched coverings to represent each homeomorphism of X as the projection from a
-compatible homeomorphism f of U which induces a geometric automorphism α
of the covering group by conjugation:
γ → f γ f −1 = α(γ ).
Of course, one must verify that distinct choices in lifting, or a homotopic variation,
lead to the replacement of f by γ0 f where γ0 ∈ , and that the effect on the
automorphism α is to compose it with an inner automorphism of .
If mod() denotes the group of (geometric) automorphisms of and QC(X) de-
notes the group of quasiconformal (qc) self-homeomorphisms of the compact (possibly
orbifold) Riemann surface X, with QC0 (X) the normal subgroup of null-homotopic
ones, then from the above procedure we obtain a homomorphism from QC(X) to
Aut(π1 (X)), which projects to a homomorphism
from the geometric mapping class
group to the algebraic one: the subgroup Inn of inner automorphisms of = π1 (X)
arise from distinct liftings of a homeomorphism homotopic to the identity and writing
Mod(X) for the quotient Aut(π1 (X))/ Inn, we have

: QC(X)/QC0 (X) ∼
= π0 Diff + (X) −→ Mod(X) ≤ Out(π1 (X)).
The presence of either torsion or punctures in the Fuchsian group renders the
proofs more difficult but does not change their validity. Furthermore, the morphism
is surjective: that is, we can realise each individual automorphism α : → by a

quasiconformal homeomorphism f = fα of U/ – this is a classical theorem due to
Nielsen and Fenchel. For more details of that result, see for instance [23], Chapters 1
and 9. A more detailed discussion of the algebraic and geometric versions of modular
group for orbifold surfaces which includes the general isomorphism result is given
in [27].
Employing the geometric definition of mapping class group, we write the funda-
mental action on Teichmüller space as
ρα ([f ]) = [f fα−1 ],
with [f ] denoting Teichmüller-class, which defines a group action on T () with
kernel the subgroup Inn() of inner automorphisms. In this way, we obtain the
important representation of the quotient Mod(X) = mod()/ Inn(), as the Teich-
müller modular group Mod() of (biholomorphic) automorphisms of T (); if
represents a compact surface with genus g and n punctures, this group is often denoted
Modg,n or Modg when n = 0.
A fundamental paper of Royden [58] analyses the metric structure on T () which
is induced by using the supremum of the distortion function μ ∈ B1 to measure
the distance between marked compact surfaces, [Xj , fj ] with f1 = wμ f0 : write
k = μ∞ = ess. sup{|μ(z)| | z ∈ X0 } and set

1 1+k
d([X0 , f0 ], [X1 , f1 ]) = ln . (2)
2 1−k
This is Teichmüller’s metric on T () – see for instance [16] for a detailed account –
and it is not hard to see that it is invariant under the action of the modular group. Indeed,
one needs only to redraft the definition of the metric in terms of the qc mappings which
interconnect the two marked surfaces:
1
d([X0 , f0 ], [X1 , f1 ]) = ln K((f1 f0−1 ). (3)
2
Putting in the action of α ∈ Mod(X) leaves this expression unchanged.
Royden proved the following important result.
Theorem 1. The Teichmüller modular group Modg is the full group of Teichmüller-
metric isometries of Tg .
The result was extended to all finite dimensional Teichmüller spaces by Earle and
Kra: the notation g, n refers to the topological surface data of genus g with n ≥ 0
marked special points and is explained later on.
Isom(Tg,n ) ∼
= Mg,n . (4)
This group action is of pivotal importance in the Teichmüller theory, because the
action is properly discontinuous on T () = Tg,n with discrete orbits (see for instance
[16], [54]), producing as quotient the desired moduli space Mg,n of conformal classes
of marked genus g surfaces.
It can be proved (with some work) that the natural complex structures on the
Banach spaces of Beltrami forms project to make the Teichmüller spaces into com-
plex manifolds; one shows that the tangent space at a marked point [X0 , f ] ∈ T (X)
is dual to the complex vector space of holomorphic quadratic differential forms on X0
in the case of a closed surface, with extra vanishing conditions at any cone points or
punctures. From the Riemann–Roch theorem, it is known that this space has finite
dimension 3g −3+n, which tallies with the dimensions of the spaces T (). One then
shows that this almost complex tangential structure is integrable. Furthermore, there is
a Hermitian metric on the tangent bundle which derives from the Petersson inner prod-
uct on the quadratic differential forms and it was shown by Weil, and independently
by Ahlfors, that this metric is Kähler. In fact, Wolpert was able to prove that every
moduli space has a projective embedding, using Kodaira’s embedding theorem [76].
Note. A full account of the complex analytic theory of these spaces is best given by
considering the Bers embedding, which requires more time and space than we can
afford: the reader should consult the standard texts cited earlier for this important
work.
A compact genus g Riemann surface with a non-trivial group G of biholomorphic

automorphisms will give rise to a marked point in T (X) = Tg which has G as the
subgroup of Mod(X) stabilising the marked point; similarly, a finite group of homeo-
morphisms of a compact surface always produces a marked Riemannian metric on the
surface, hence a Teichmüller point with this as its automorphism group. The famous
Realisation Problem of Hurwitz–Nielsen asked whether, given a finite subgroup H of
Mod(X), there is a choice of marked point [μ] ∈ T (X) for which the stability group
is H ; this may be rephrased as follows.
Is there a finite group of homeomorphisms H1 which realises a given finite group of
mapping classes H , in the sense that H1 ∼= H with H1 ≤ QC(X) and
(H1 ) = H ?
This problem was studied, and partially solved, by Nielsen and Fenchel and var-
ious special cases were proved by later authors over more than 30 years until it was
finally resolved in general by S. P. Kerckhoff [35] using Thurston’s powerful theory
of earthquakes. Different proofs have been given by Wolpert (using convexity proper-
ties of the hyperbolic length functions on Teichmüller space) and more recently in an
interesting paper by D. Gabai [15], which uses methods reminiscent of Nielsen’s and
also explains how to relate it to fundamental problems in the geometry and topology
of 3-manifolds.
Notice, however that there is no natural overall choice of a realisation, in the
following sense.
Theorem 2. No homomorphism exists from mod() into the group QC(U) which
would split the epimorphism
.
For inverse homomorphisms into the subset of C ∞ diffeomorphisms of X, the

impossibility of such a splitting was proved by S. Morita [53], using his theory of
characteristic classes for surface bundles and moduli spaces. This restriction has been
removed recently by work of D. B. A. Epstein and V. Markovic.
4.3 Families of Riemann surfaces

At this point, it is appropriate to indicate how to construct the corresponding universal
holomorphic family of Riemann surfaces modelled topologically on X = U/ , with
the space T () as base. This is a tautological holomorphic fibre space V () over the
Teichmüller space, obtained as follows.
(1) Place over each point [μ] of T () the corresponding hμ -deformed hyperbolic
disc Uμ , on which the deformed group μ operates.
(2) Take the quotient of this space F () ∼
= T () × U by the discrete action of
on the fibres as μ on Uμ .
Note that although we use the notation hμ , both this normalised map and the
deformed disc depend only on the T-class [μ]. The resulting space V () is a complex
analytic manifold which is locally a topological product of the fibre orbifold and
an open disc in the base T (). However the product structure is holomorphically
nontrivial because of the nature of these families of quasiconformal mappings hmu
inter-relating the nearby fibre surfaces. This type of deformation is trivial only when
there is a conformal map of U inducing it.
The same fibre space structure is also realisable real-analytically by a standard
bundle-theoretic procedure known as homogeneous reduction, from the representation
space R0 (, G) of all discrete injective homomorphisms ρ of the Fuchsian group
into the direct-isometry group G ∼ = PSL2 (R) of the hyperbolic plane U. We assume
for simplicity that the quotient surface X = U/ is compact or possibly finite area,
with finitely many cone points or punctures; the topological type (g, n) records the
genus g and the total of n ≥ 0 special points. Since the spaces T () depend as
complex manifolds only on the topological data and not on the cone/cusp nature of
the marked points, we usually refer to the moduli spaces as Tg,n and Mg,n ; modular
groups are a little less flexible, and Modg,n will refer to the case of n punctures.
The target Lie group G operates on the space R0 by conjugation: each element ρ(γ )
in the image of the given representation is sent by t ∈ G to the conjugate tρ(γ )t −1 by
the fixed element t in G. In this way, each representation ρ ∈ R0 (, G) determines a
conjugacy class of G-similar ones.
Now this action on R0 is proper, and each G-orbit, a collection of marked groups
conformally equivalent to the group ρ(), may be viewed as the fibre of a map to the
quotient space of conjugacy classes, which is none other than our Teichmüller space
T () of normalised marked Fuchsian groups – a comparison of these two perspectives
may be found in papers of Kerckhoff [34], [35]; see also recent work of many authors
(Fock, Goncharov, Kashaev, Penner and others) on a quantisation of the theory of

moduli which exploits these two approaches to geometric structure both on surfaces
and on the Teichmüller spaces.
To convert the fibration R0 (, G) → T () into a family of Riemann surfaces,
we note that the group G = PSL2 (R) may be viewed as the unit tangent bundle of
the hyperbolic plane, a principal PSO2 -bundle; this may be converted into a holomor-
phic family of hyperbolic planes by taking the quotient modulo the rotation subgroup
PSO(2) of G. This space is denoted F (). Finally one obtains a fibration with
Riemann surfaces as fibres by taking the quotient by the action of via the represen-
tations ρ ∈ R0 () on each fibre disc, as a subgroup of G): to see this in more detail,
the reader may consult [45] or [54].
There is another description of the holomorphic fibre space F () as the space of
all pairs
(t, z) with t = [wμ ] ∈ T () and z ∈ wμ (U);
for details of this aspect of the theory, which involves the analytic properties of the
Schwarzian derivative and the Bers embedding of each Teichmüller space in the Banach
space of holomorphic functions on a half plane (with the hyperbolic L∞ norm), the
reader may consult [7] or [16].
Following on from the fibre space property, the action of the Teichmüller-modular
group on the base space – as the group Mod() (of mapping classes for the surface or
orbifold X = U/ ) on T () – lifts to an action on F () which is effective (except
when g ≤ 2), and the quotient space is a family of hyperbolic discs Uμ over Mg,n .
The discrete group acts on this family via quasiconformal deformations μ on Uμ
as fibre-preserving maps to produce the modular family Cg,n of Riemann surfaces of
(generic) genus g with n ≥ 0 special points. The quotient fibration
π : Cg,n → Mg,n
has orbifold singularities which stem from symmetry properties of certain (non-
generic) Riemann surfaces. These arise in the following way.
Proposition 1. At a point x of Mg,n which represents a Riemann surface Sx with

direct symmetry group H , the fibre π −1 (x) ⊂ Cg,n is the quotient surface Sx /H .
This fact is easily seen from the covering space properties of the hyperbolic uni-
formisation: Riemann surface symmetries lift to hyperbolic disc isometries, and these
contribute to the quotient action producing the fibre surface in the construction using
R0 (, G) outlined above; see [45] for more details.
Note. It is an important ingredient of the Fuchsian group approach to moduli that

the marking procedure for the group representing a surface with distinguished points
provides a canonical way to incorporate the forgetful map which forgets one chosen
distinguished point (or a designated set of points).
To explain this process, we recall the detailed geometric information present in

the datum of a marked Fuchsian group with quotient surface X = U/ of finite
type: there is a finite presentation for involving the standard commutator relator
arising from the canonical surface symbol and further algebraic relations, one for each
conjugacy class of torsion or boundary parabolic element. We have by implication
an ordered finite set of special points on X, with interior cone points of angle 2π
n for
order n torsion generators and a puncture for each parabolic generator, manifested as a
cusp of , a pointlike boundary component of X which may be filled in by a standard
procedure to yield a hyperbolic Riemann surface X∗ with one less boundary point.
This process makes it possible to formulate precisely (and prove) the following Bers
fiber-space theorem, which identifies the fiber space F () as a Teichmüller space in
its own right:
Theorem 3. Let X = U/ be a compact Riemann surface uniformised by a (finite

co-area) Fuchsian group . Then
F () ∼
= T ( ),
where is a Fuchsian group uniformising the punctured surface X = X − {x0 },
where X = U/ .
It leads also to an identification between Teichmüller modular groups. Under the

same hypotheses, we have
Theorem 4. The modular group mod() is a finite index subgroup of Mod( ).

In particular, mod() ∼
= Mod( ) when is a closed surface group, i.e. torsion
free and co-compact.
The latter result is the counterpart of the final terms in the topologist’s long exact
sequence for the homotopy groups of a fibration for the holomorphic fiber spaces
Fg,n → Mg,n . In general, these two modular groups are only commensurable, that
is, they possess isomorphic ‘pure’ subgroups which correspond to mapping classes
preserving all of the marked points.
4.4 Hyperbolic surfaces inside moduli space

One of our aims is to exhibit and study finite area Riemann surfaces which are totally
geodesic suborbifolds of moduli space. We saw earlier the examples discovered by
Thurston; a generalisation of this construction was obtained later by Veech and others.
In order to describe some of these results, we need a preliminary account of complex
Teichmüller geometry.
Definition. A Teichmüller geodesic disc is a set of marked complex structures on a

Riemann surface X0 arising from the following construction: fix a quadratic differen-
tial form φ on X0 . This determines a complex 1-parameter family of deformations of

X0 with several alternative descriptions:
(i) A characteristic type of parametrisation of X0 by complex affine charts is deter-
mined by φ : locally away from the zeros of φ, write φ = dw2 to get local parameters
w up to transition functions of the form w → ±w + c. Equivalently, set
z
w= φ(t).
z0
Following Veech [70], one may call such an atlas an F -structure on X0 .
Now for each ε with |ε| < 1, define a new F -structure on the underlying topological
surface S0 by rotating each chart through arg ε and expanding the real foliation of
R2 = C, while contracting the imaginary (vertical) one, by the mapping
1
−1 1 + |ε|
z = x + iy −→ w = Kε2 x + iKε 2 y where Kε = .
1 − |ε|
If arg ε = 0, so that 0 < ε < 1, the family of F -structures so defined is called the
Teichmüller ray at X0 in the direction φ.
(ii) Write νε (z) = εφ(z)/|φ(z)| and solve the Beltrami equation (1)
∂w ∂w
= νε
∂z ∂z
for this family of complex dilatations; conjugating the marking of X0 by the family ωε
of solution homeomorphisms (suitably normalised to remove the G ambiguity) yields
a one-dimensional holomorphic family of Riemann surfaces over the unit disc D.
It is an exercise in the basic workings of quasiconformal mappings to see that these

two descriptions determine the same complex 1-parameter deformation of the central
surface X0 .
There are at least two further valuable ways to view this deformation. One stems
from the affine Euclidean geometry of the system of local charts determined by a
quadratic differential; this is the point of view employed by Veech and also by Kerck-
hoff, Masur and Smillie [36]; it is described very neatly in the survey article [13] and
we describe one of Veech’s examples in these terms later in this section. The second,
perhaps more topological, view is older and was developed originally by Jenkins and
Strebel, and later by Gardiner (for details, see for instance [32], [67], [16]); it makes
use of the concept of critical trajectories for a quadratic differential. These are the
φ-horizontal curves on X0 through the zeros of φ in the initial system F0 of Euclidean
charts: the set of all (horizontal) critical trajectories determines a decomposition of
the surface into a union of annuli and spiral regions, each foliated by simple horizontal
trajectories. One may then pose an extremal stretching problem on the Riemann sur-
face S0 , determined by the topological structure of this singular foliation arising from
the quadratic differential φ: a weight is assigned to each component of the noncritical
set and one seeks a metric on the surface which minimises the distortion overall in a
sense determined by the set of weights.
This last point of view is valuable because it links the classical complex analytic
methods with the ideas of Thurston on deformations of hyperbolic structures, which
bring into play the geometry of foliations of a surface with permitted ‘finitely pronged’
singularities to set up the space of measured foliations which determine a natural real
spherical boundary of Teichmüller space. This Thurston boundary does of course
include points representing the endpoints of the T-rays emanating from the central
point [X0 ] as described above, but differs in an essential way from the Teichmüller
boundary, also a sphere, which is obtained by considering the set of all the boundary
points of T-rays from a fixed central base point, for reasons arising from the subtle
dependence of the latter sphere on the choice of base point.
The interested reader should consult Kerckhoff’s paper [34] for details of this
fundamental distinction: he establishes the crucial fact that whereas the Teichmüller
modular group action extends naturally to the topologically-based Thurston boundary
with highly significant consequences (see [69]), it cannot extend continuously to the
more rigidly contructed Teichmüller boundary.
4.5 Example: a T-disc parametrising tori
The motivating example of this basic construction comes from the deformations of a
torus: recall that marked tori Xτ correspond bijectively to τ ∈ U. We need to relate
this to the T-disc construction, in which we employed the conformally equivalent unit
disc as parameter space. To define the deformation homeomorphisms from a reference
point, say i ∈ U, to τ ∈ U, we use the real affine mapping f˜τ : x + yi → x + yτ
f˜τ : C
⏐ −→ C
⏐
⏐ ⏐

which induces fτ : C/i −→ C/τ
and this homeomorphism is extremal within its homotopy class in Teichmüller’s sense,
that is, it has the least overall distortion measured by making the supremum of the
local stretching function on X as small as possible. To obtain the local distortion, one
calculates
∂fτ (1 + iτ ) dz ∂ ∂
= , where ∂ = , ∂ =
∂fτ (1 − iτ ) dz ∂z ∂z
and z is the local parameter (z = x +iy) induced on the complex torus by the quadratic
differential form dz2 on C. This can be rewritten as ε φ̄/|φ| where ε = 1+iτ
1−iτ ∈ D, the
unit disc, and φ is the projected holomorphic form dz2 on C/i .
In the present case, the Teichmüller distance between i and τ as defined earlier
turns out to be (see for instance [2])

1 1 + |ε|
d(i, τ ) = log .
2 1 − |ε|
This is of course Poincaré’s metric on U, giving it the structure of real hyperbolic

plane.
4.6 Deformation T-discs embed

In general, let X be a Riemann surface, and let φ ∈ 2 (X), φ = 0. The Teichmüller
deformation specified by φ defines a holomorphic mapping eφ of U (or D) into T (X)
via the following map into B1 (X) ⊆ L∞ −1,1 (X)
φ
eφ (ε) = ε for |ε| < 1.
|φ|
Proposition 2. The mapping eφ is a proper holomorphic injection of the disc D into

T (X).
Proof. The map eφ is proper because as |ε| → 1, the norm eφ (ε) → 1, which
implies that the distortion is becoming arbitrarily large and so eφ (ε) tends to the
boundary ∂T (X). The map is injective, either by Teichmüller’s uniqueness theorem
[6] or by the uniqueness of solutions to Beltrami’s equation, and it is holomorphic
because the complex structure on T (X) is inherited from that of the Banach space of
distortion measures B1 (X).
The global metric between points in Teichmüller space, as defined earlier, is in
fact realised by the Poincaré metric for points in a T-disc. This is part of the original
approach introduced by Teichmüller who proved that any two points are joined by
a Teichmüller ray, and used the formula 2 to define it. As a consequence, every
embedded Teichmüller disc is totally geodesic and this leads to another result of
Royden (op. cit.) which identifies Teichmüller’s metric with the canonical Kobayashi
metric on this complex manifold. In the present context, we can re-state this classic
result, generally known as ‘Teichmüller’s Theorem’, as follows.
Theorem 5. For any two distinct marked Riemann surfaces in Tg , there is an embedded
T-disc which contains them, unique up to an isometry of the disc, and the distance
between them is given by Poincaré’s hyperbolic metric.
For a proof of this theorem, which inaugurated the global theory of moduli, see
[1], [6].
4.7 Veech’s examples

A recipe using Euclidean geometry for explicit construction of a geometric structure on
a surface using holomorphic differential forms is given next; this is part of a general
procedure for endowing a surface with a covering by Euclidean polygonal charts,
which Veech has called F -structures. They are also called translation surfaces by
some authors, since the surfaces concerned are determined by identifications of edges
of polygons using Euclidean translations.
The examples we describe involve surfaces which may be represented explicitly as
ramified coverings over CP 1 : they are taken from a large class discovered by Veech
([70], [72]).
Let Xn , with n ≥ 5 and odd, be the compact hyperelliptic surface of genus g =
(n − 1)/2 with affine equation y 2 = 1 − x n and consider the holomorphic 1-form
ω = dx/y. It has a zero of order 2g − 2 at the single point where x = ∞. The
quadratic form q = ω2 defines a Teichmüller disc in Tg , which turns out to have
particular significance.
Veech considered these structures in connection with the dynamical
study
of bil-
liard trajectories in a Euclidean isosceles triangle T with angles πn , πn , (n−2)π
n . The
relationship with Xn is based on the following construction, involving an associated
triangle Tζ with vertices at 0, 1 and ζ = e2π i/n .

Let P = n−1
=0 ζ Tζ ; multiplication by ζ determines a cyclic Euclidean rotation
symmetry of P such that the quotient mapping is an n-fold covering of the plane near 0.
Let Q be the equivalent polygon formed with T̃ζ = −Tζ , and then identify pointwise
in P ∪ Q the pairs of outside edges in corresponding triangles ζ T̃ζ , ζ Tζ using a
translation (the corresponding edges are parallel). The result is a closed surface with a
local complex analytic structure except at the projection of the corners, where the union
of all the corner angles adds up to more than 2π . After extending the complex structure
by removing these isolated singularities at the corners – using Riemann’s classical
approach to resolving isolated singularities – this process determines a closed Riemann
surface Xζ (with nontrivial conformal symmetry group) having a local structure over
P1 given by the following composite map p2 (z) := f p1 :
Xζ SSS
SSSS
SSSpS2 (z):=f p1
p1 SSSS
SSS
)
CP = Xζ /z → ±z
1 / Xζ /z → ±ζ z in P ∪ Q.
f
The arrow labelled f is the standard mapping z → zn , after moving the origin of z-
coordinates to the centre of the polygon P . It is not difficult to infer from this geometric
data that Xζ is the compact Riemann surface Xn defined above: for instance, observe
that there is an order n symmetry of X induced by lifting z → ζ z which fixes three
points, the two centre points and the single orbit of corner points. Also it follows
that the 1-form dz in P ∪ Q lifts to ω on Xζ . Veech proves the following result, a
striking link between the hyperbolic geometry of the disc and the complex geometry
of Teichmüller space and the Teichmüller modular group.
Theorem 6. The stabiliser in the modular group Modg of the Teichmüller disc deter-
mined by the differential q = ω2 is a Fuchsian triangle group

1 2 cot πn cos 2π
n − sin 2π
n
Hn = , = σ, β
0 1 sin 2π
n cos 2π
n
g+1
which is isomorphic to x1n = x22 = 1 via x1 x2 = σ , x1 = β.
4.8 Hecke triangle groups and polygonal surface tilings

In this section, we explore some of the hyperbolic geometry underlying Theorem 6.
The triangle group Hn obtained in the theorem is known as a Hecke group because of
Hecke’s classic work [28] on this kind of Fuchsian group in connection with Dirichlet
series satisfying a functional equation.
A fundamental region for the action of Hn on the hyperbolic plane U may be
obtained by a standard geometric construction: take a quadrilateral F with a reflection
symmetry which fixes the diagonal joining one ideal vertex at ∞ and a second vertex
having interior angle 2π/n. It is simplest to view F as the union of two abutting
hyperbolic triangles T , T , each with angles π/n, 0 and π/2. Place the vertices of T
with one of angle π/n at i, the zero angle at ∞ and the third at the foot ±ξ of the
hyperbolic perpendicular from i to the line Re z = cot(π/n) and let the triangle T
be the reflection of T in the imaginary axis. Just as the action of an order n rotation
divides a Euclidean regular n-gon into n congruent isosceles triangles or, equivalently,
into n quadrangles with angles {2π/n, π/2, (n − 2)π/n, π/2}, there is a hyperbolic
regular ideal n-gon P with all vertices at infinity, which is the result of applying to F
the order n hyperbolic rotation x1 fixing i.
Consider now the union of P and the (hyperbolic) translate σ (P ), which is a
congruent polygon sharing a vertical geodesic side. It is not hard to show that the two
operations x1 and σ generate a discrete group of isometries of H which turns out to
be Gn : the union of all translates of the quadrilateral F tesselate H and these two
elements are side-pairing elements of F which fit into the framework of the classical
theorem of Poincaré on fundamental polygons for discrete hyperbolic isometry groups;
for more details of this result, see [43], [45], or [52].
We can apply Proposition 2 and the homogeneity of the hyperbolic isometry group
G acting on U to arrange the present pattern neatly into the (local and global) complex
geometric structure of the Teichmüller space T (Xn ), by placing the reference point i
at the marked surface Xn and, furthermore, fixing the T-ray from i in the direction
of q to be the upper part of the imaginary axis.
It is important to maintain contact here with the symmetric properties of the Eu-
clidean tesselation picture imposed on Xn by the form q. Note that the element x2
of order 2 (which may be chosen to fix the π/2-corner point of T ) corresponds to
interchanging the Euclidean polygons P and Q via the halfturn z → −z, whereas it
is represented on the quadrangle domain F by two conjugate involutions, each fixing
one of the π/2-corners ±ξ . Also, the isometry x1 of order n fixing i corresponds to

the lifting of z → ζ z on each polygon. It is not hard to show using the geometry of the
hyperbolic plane that within the tesselation of U by ideal quadrangles γ · F , γ ∈ Hn ,
we recover the original decomposition of Xn into a pair of (hyperbolic) regular ideal
polygons P , Q = x2 (P ) with n vertices.
To produce an algebraic description of the hyperbolic surface Xn , it is convenient to
consider the quotient space X of U by the commutator subgroup Cn = [Hn , Hn ]. To
see that this is indeed the same Riemann surface, one needs to note that the commutator
quotient group Z2 ⊕ Zn acts as a group of biholomorphic automorphisms of X, with
precisely the same symmetry pattern (fixed points etc) as that of Xn = {P ∪ Q}/ ∼
taken modulo the symmetries of the dihedral n-gon. Of course, the ideal (zero-angle)
vertices, which determine either one or two points at ∞ on the quotient surface de-
pending on the parity of n, must be added to create a compact surface. By adjoining
cusps to X, we obtain a compact surface. A fundamental domain for Cn may be formed
by first taking the union of n abutting (horizontal) translations of the domain F for
Gn to produce a lift of the punctured polygon P − 0 and then adjoining a translate
of this by the involution x2 . The result is now recognisable as Xn , at least when n is
odd: observe that it has genus g = (n − 1)/2 (via the Riemann–Hurwitz formula),
an order n symmetry with three fixed points (at the Cn orbits [i], [0] and ∞) and
an involution symmetry induced by conjugation with x2 on U/Cn , which must be
hyperelliptic because it fixes the n + 1 = 2g + 2 points on the quotient surface arising
from the n right-angle corner points of P and the (single) orbit of cusp vertices.
In the piecewise-Euclidean model Xζ of the surface, the critical graph of the
holomorphic 1-form q consists of a union of vertical line segments through the vertices
of the polygons P and Q, all of which are identified to a single point of Xζ ; the
complement is a set of k = (n−1)/2 disjoint cylinders. When viewed in the hyperbolic
model as the surface Xn , the cylinders lift to strips which comprise the connected
components of the complement of the critical graph K. In this way, it follows that
a set of disjoint geodesic curves tending to the single cusp at ∞ may be taken as
representative liftings of the critical trajectories of the holomorphic form ϕ of weight 2
on U whose square defines q = ω2 on Xn . It also follows that the k-th power of the
parabolic element σ of Hn determines a covering map which preserves each cylinder,
and so defines a product of Dehn twists about the core curves of the cylinders.
These facts can also be verified directly by matrix calculations within the F -
structure, which are similar to the reasoning given earlier in Thurston’ s original
examples; see for instance [13] for a careful analysis along these lines.
Because this T-disc D = Dq has a lattice group stabiliser in the modular group
Modg , any sequence of points in D is modular-equivalent (by elements stabilising the
T-disc) to either a convergent sequence or a sequence tending to the cusp of Hn and so
divergent in Mg : it follows that the image in the quotient moduli space is a closed (but
non-compact) sub-orbifold of the complex V-manifold Mg . This proves (in outline)
the following theorem, also due essentially to Veech [70].
Theorem 7. The image in Mg of the Teichmüller disc through Xζ corresponding

to the form q is a complex analytic subspace isomorphic to the affine complex line,
U/Hn = Xζ / Aut(Xζ ).
This result motivates the following definition.
Definition. A Teichmüller–Veech disc is a Teichmüller disc in a Teichmüller space Tg

for which the stabiliser in the Teichmüller modular group is a lattice subgroup of the
hyperbolic isometry group G.
The image in Mg is then a finite area hyperbolic surface, called a Veech curve or
sometimes a Teichmüller curve.
Notes. 1. Concealed here within the orbifold structure of the moduli space is a para-
doxical appearance by the hyperelliptic involution J of the central surface Xζ : as a
modular automorphism of Teichmüller space, J fixes the point i of U corresponding
to Xζ and sends the 1-form ϕ to its negative. However this implies that it must fix
every point of the Teichmüller–Veech disc we have constructed. Thus the stabiliser
of the point i in the hyperbolic plane (representing the T-disc) U is the cyclic group
x1 and not the full automorphism group of Xζ , the surface which underlies this point
of Tg . In fact the element x2 is a representation of the hyperelliptic involution of Xζ (on
the universal covering space U) and lies in the kernel of the holonomy homomorphism
f → Df from the group of affine selfmaps of the F -structure to GL(2, R). There-
fore x2 is killed by the restriction homomorphism from the modular group stabiliser of
the T-disc U to the isometry group of the T-disc. On the other hand, the hyperelliptic
involution j is induced by this order 2 generating element x2 , when viewed as acting
on the universal covering of the central fibre of the T-disc, but this is illusory. The
point of our T-disc actually fixed by x2 represents a different Riemann surface from
Xζ , having an additional involution automorphism; Veech [70] calls this involution a
hidden symmetry.
2. The case of regular n-gons with n even follows a related but slightly different
pattern which is detailed in [71]. In fact, using the discussion of Veech’s examples
given by [13] it follows that for every n ≥ 5, the hyperelliptic curve Xn , of genus
g = [n − 1/2], enjoy the property that there are two independent Teichmüller discs
centred on the corresponding point in Mg , determined by the 1-forms ω1 = dx/y
and ω2 = xdx/y, the second corresponding to a construction of the same surface as a
translation surface quotient from a single regular polygon with 2n sides; in the latter
case the T-disc has as stabiliser a {n, ∞, ∞} triangle group.
4.9 Fermat curves represented by T-discs

The Fermat curves Yn are determined by an affine equation
xn + yn = 1
and have genus g = (n − 1)(n − 2)/2, and so when n ≥ 4 they have hyperbolic
uniformisation and fit into a similar picture with respect to the same Hecke triangle
groups Hn as above. Note that the corresponding compact Riemann surface is obtained
by adding just one or two points at infinity depending on parity.
The tesselation of Yn by regular n-gons may be seen in several ways, for instance by
using a similar description of a tailoring pattern for Yn using 2n copies of the polygon P ,
as in [71]. A second method, less reliant on diagrams and explicit pastings, comes
from consideration of the collection of finite index subgroups of a given Hecke group
Hn and the corresponding quotient surfaces covering the affine plane U/Hn [25]. One
sees by either method that the automorphic form of weight 4 for Cn = [Hn , Hn ] on U
which determines the quadratic differential form q = ω2 on Xn also induces the form
q1 = dx 2 /y 2n−2 on the covering surface Yn ; this may be viewed as arising by pullback
on forms via the holomorphic covering. In the case of the Fermat curve, however,
the involution x2 in Hn does not preserve the polygonal tesselation of the surface,
because the hyperelliptic involution does not lift to Yn . The (lattice group) stabiliser
in the modular group of the surface Yn for this Teichmüller disc is in fact precisely
the index-2 subgroup Hn that omits this element. This subgroup has fundamental set
a union of two right triangles interchanged by x2 , and hence is easily seen to be a
triangle group of type (n, n, ∞).
Theorem 8. The image in Mg of the Teichmüller disc through the complex Fermat
curve Yn corresponding to q1 is a complex curve isomorphic to the affine line, U/Hn .
See [72] for detailed proofs of this and other precise stability results.
4.10 Regular tesselations and smooth coverings

In this section, we show how to produce many other examples of hyperbolic discs in
Tg with large stabiliser, by an immediate inference from the argument used for the
Fermat surfaces.
Let X̃ be any smooth genus h covering surface of the (compactified) commutator
surface Xn for the Hecke group Hn with n ≥ 3. Then application of the same reasoning
as above with the pullback of the form ω shows that there is a corresponding T-disc
in the Teichmüller space Th centred at [X̃] with stabiliser of finite index in Hn . In
some sense, it is the same data that is involved here: when lifted to a system of
covering surfaces, the T-disc given by ω lives in some inverse limit of Teichmüller
modular objects which identifies commensurable structures occurring among all finite
coverings of Xn . Dennis Sullivan has called this type of limit object a solenoidal
surface or Riemann surface lamination; our examples are individual levels in the
profinite completion of the (complex) algebraic curve Xn .
Of course, one can proceed in analogy with this process for any example of Veech
curve, by which we mean a Teichmüller disc with stabiliser a lattice Fuchsian group,
and achieve further examples with the same property.
Note. A point which merits further exploration is the need to correlate the various
modular groups arising in this situation. The result, in the restricted case where the
coverings of moduli spaces are required to be finite cyclic and smooth, is a new type
of generalised modular group which is called the commensurability modular group by
Nag and Sullivan [55].
4.11 Return to Thurston’s examples

In this section we explain how to generalise the examples discussed earlier.
Take two distinct finite systems of loops which together fill up a compact surface,
that is, a pair α = {α1 , . . . , αm }, β = {β1 , . . . , βn }, such that loops in the same
set are pairwise disjoint and homotopically distinct, and with α-loops intersecting β-
loops transversely in such a way that the complement is a union of topological discs,
each with boundary containing an even number of segments of alternately α and β
curves and at least four in number, so as to ensure that the loops have no inessential
intersections. Choose centres for these disc components of S − {α ∪ β}; there is a
dual cell decomposition of S into quadrilaterals which arises from joining centres of
adjacent components by edges labelled as v = vertical for α, h = horizontal for β.
Now an F -structure is defined on the surface in the same way as before, by viewing
the surface as covered by unit square in the Euclidean plane, with the induced orien-
tation and α = vertical, β = horizontal labelling. This again defines an F -structure
on S. Furthermore, a twist about any α or β curve acts as an integer translation when
viewed (in the holonomy representation) in this affine structure on the real plane below;
the group generated by them is often a finite index subgroup of SL2 (Z).
We note, however, that the subgroup of the stability group of a T-disc generated by
twists may have infinite index, as in the example of a surface with genus 2 described
by Earle and Gardiner [13]. It is constructed from a 6-by-1 rectangle – six squares
in a line – with edge identifications given by a translation by 6 from one end to the
other and four involutions, operating in pairs on top and bottom sets of edges, with
two edges labelled aj of length 1 and two labelled bj of length 2, to produce a handle
from each, to give a surface symbol a1 b1 a1−1 b1−1 ca2 b2 a2−1 b2−1 c−1 . It is not difficult
to see that this structure has twist subgroup generated by the matrices

1 6 1 0
and ,
0 1 6 1
which has infinite index in the modular group (1), while the full stabiliser is com-
mensurable with (1).
It follows from Veech’s work on the Teichmüller geodesic flow on the bundle of
quadratic differentials Qg over Tg (see [71]) that the points of Tg which possess such
a representation are dense; we shall make use of this fact in the next chapter. A rather
different density property for these structures was proved in a paper of H. Masur [46]:
the set of Teichmüller-geodesics given by the original Thurston prescription (using
single α and β curves) projects to a dense subset of the unit tangent bundle to moduli
space. As a consequence, there exist Teichmüller geodesic rays whose projection is

dense in Mg .
5 Triangle groups and modular families

The Hecke groups Hn and the more general triangle (p, q, ∞)-groups appear as a
crucial ingredient in several areas of mathematics. We shall see in this section that they
are naturally related to the construction of many Teichmüller disc families of Riemann
surfaces and in particular to the examples discovered by Thurston and Veech.
5.1 Tesselations and triangle groups

Let X be a compact Riemann surface which admits a regular tesselation by hyperbolic
n-gons. We describe how such a surface admits reconstruction from a torsion-free
subgroup of finite index in a Hecke group Hn ; this follows from a fundamental
theorem about maps on surfaces (see for instance [66]). In particular, the case n = 3
relates surfaces X with a decomposition into equilateral triangles to the classical
Farey tesselation of the hyperbolic plane associated with the classical modular group,
H3 = (1) = PSL2 (Z). The compact surface X = X obtained from the (finite co-
volume) Fuchsian group has finitely many cusps added to the (finite area) surface
U/ in the usual way, one for each -orbit of boundary points at which the stabiliser
is non-trivial. The holomorphic projection π : U/ → U/Hn ∼ = C, induced from
the identity map on the universal covering U, extends to a Belyi map f : X → P1
from the compactification, with each cusp of X mapped by f to the single cusp {∞}
of Hn which compactifies the plane U/Hn to C ∪ ∞ = P1 . The tesselation of X
is then determined by projecting part of the standard Hn -invariant subdivision of U
into {2, n, ∞} triangles. Such a triangle arose in the preceding section as half of a
fundamental domain F : in that picture, the Gn -orbit of the hyperbolic line L = {z =
cot (π/n)} determines a tiling of U by ideal n-gons, projecting to a finite tesselation
of X whose vertices correspond to the cusps of . To say that the tesselation of X
by n-gons is regular means that a fixed number m of polygons surround each vertex,
which is equivalent to the statement that the total width of each cusp is the same fixed
number.
Notes. 1. The data of a regular m-valent tiling by n-gons of a compact surface can
be specified combinatorially by prescribing a torsion-free subgroup of a {2, n, m}-
triangle group, but it is the cusped version of the picture that provides the link with
Teichmüller surface families.
2. In [66] a convenient method is given to generate all hyperbolic surface tessela-
tions in similar fashion, by a universal construction in U governed by representative
subgroups of the modular (cartographic) subgroup 0 (2). This will be employed
to give a succinct description of a wide class of modular curve families in the next
subsection.
3. The equivalent combinatorial concept for more general triangle {p, q, ∞}-
groups is a hypermap (see [33]): here the corresponding universal construction in-
volves the principal congruence subgroup (2) (which has index 2 in 0 (2)) and
subgroups of finite index in that group.
5.2 Associated Teichmüller families
We wish to produce, from the covering data {X, f } of the preceding section, a corre-
sponding curve in the moduli space of X. This will apply not only to the Riemann
surfaces tiled by regular n-gons, which correspond to ‘balanced dessins’, but also to
the broader class of complex orbifold structures given as clean Belyi coverings of
P1 ; in this case the surface is the quotient of the disc by a finite index subgroup of
a (p, q, ∞)-triangle group p,q . The collection of pairs (algebraic curve & map) so
generated are part of the totality of dessins and of course, by the theorem of Belyi,
each is defined over a number field – in fact, they must be part of the Q-structure of
the modular varieties themselves by Weil’s results [73] on the field of definition for a
variety. Further discussion of this may be found in [74] and in [18].
By the classification of surface maps and hypermaps, we may assume that our
covering is given by a Fuchsian group of finite index in one or another of the groups
0 (2), (2) or p,q . As before, X will denote the compact surface obtained from
filling in the cusps on U/ .
Take a tesselation of the hyperbolic plane U by a standard fundamental set Q
for 0 (2), which is a hyperbolic quadrangle with ideal vertices at −1, 0, 1 and ∞
bisected into triangles by the imaginary axis . Now we indicate two further ways to
approach the construction of quadratic differential forms for , the first one similar to
an observation of [13].
Let ω(z) = dz2 /(z3 − z) denote the form on Y = CP 1 with simple poles at the
point set B = {0, ±1, ∞}. We consider the pullback of ω to a compact surface X
through some Belyi function F : X → P1 , which we assume factors through another
function f : X → Y , i.e. F = f 2 . The pullback to X, f∗ (ω) will have a framed
horizontal/vertical structure inherited from that of ω on Y , and if f is totally branched
over B the form will be holomorphic on X.
For instance, this structure lifts to a geometrically similar F -structure on the elliptic
curve E given by the Weierstrass equation w 2 = z3 − z, which is a double covering of
Y branched over the set B, and which has a precise description in terms of the standard
℘-function for the lattice Z + Zi corresponding to E.
It is not difficult to verify [13] (cf. [70]) that the affine stabiliser of the induced
F -structure on Y is the congruence subgroup (2). Also, one sees from this specific
model that for all such pairs X, F the stabiliser of the T-disc for the form F ∗ (ω) has
finite index in (1), by a simple topological argument. Of course, when lifted to
general coverings of Y , the form ωf will have simple poles at any unramified points
of f −1 (B).
The second method works only for surfaces X covering the modular surface
X0 (2) = U/ 0 (2). It produces forms of an arithmetic nature on X from familiar
modular forms and depends on a simple fact: the group 0 (2) possesses a holomor-
phic cusp form ω1 of weight 8 that is real on the imaginary axis with a zero of
order 3 at each of the two cusps. This form is defined to be
η(τ )η(2τ )4
where η(τ ) is Dedekind’s well-known (character-automorphic) form for (1),
∞

η(τ ) = e 2πiτ/24
(1 − q n ), for τ ∈ H and q = e2πiτ .
n=1
For any subgroup < (1) of finite index with all cusp widths even, one verifies
easily that the lifted form to X has only even order zeros at the cusps, hence has a
square root ω1 which belongs to the space S(, 4) of weight 4 cusp forms. Furthermore,
the form ω1 has critical trajectories which interconnect the cusps of the surface X in
some finite graph, which includes among its edges the projection of the axis and
whose pattern depends only on the structure of the covering. In addition, for any
conformal automorphism α of X which fixes the cusps, one can average the form over
the elements α ν to obtain a cusp form ω2 with (at least) two independent directions
in which all critical trajectories interconnect cusps and all noncritical trajectories are
closed. This property, which is shared by the form ω defined earlier on Y , turns out
to be crucial in ensuring a lattice group as Teichmüller-modular stabiliser.
We consider T-discs Dω for such forms. In order to proceed as canonically as
possible, and to avoid the confusion arising from the plethora of inclusion relations
between the various Teichmüller spaces associated to surface coverings, in what fol-
lows we place modular group activities within the universal group action of mod(1),
the group QS(S 1 ) of all quasisymmetric homeomorphisms of the boundary circle, on
Teichmüller space T (1), universal Teichmüller space, the normalised quotient by the
subgroup (isomorphic to G = PSL2 (R)) of Möbius mappings of the circle.
Let ≤ 0 (2) be a finite index normal subgroup contained in the kernel of the
finite character χω : 0 (2) → S 1 , and representing a surface X = U/ , so that
ω(γ (z)) = χω (γ )γ (z)−2 ω(z) for all γ ∈ 0 (2). If the form ω is -invariant,
we obtain an induced T-disc Dω in T () ∼ = Tg,n , where n is the total number of cusp
and elliptic torsion points. This will then be mapped injectively, via the (holomorphic)
mapping π : T () → Tg which forgets the data of all torsion and cusp points, landing
in the Teichmüller space of the compactified surface X0 .
Theorem 9. The centre point of this composite embedding is a representative marked

surface [X0 ] with a quadratic differential ω0 such that the T-disc defined by ω0 on X0
realises the embedding π eω .
Proof. The compact marked surface [X0 ] will certainly have a quadratic form ω0 given
by the obvious extension to the cusps from ω on U/ . In fact, because we have a
holomorphic image of the original T-disc, this disc is also a T-disc, since the distortion
at each point is constant, with finitely many exceptional critical points.
We would like to characterise if possible the Veech T-discs presented in this manner,
which have lattice modular stability group. From the earlier study of examples within
Hecke groups Hn , it is clear that one way to do this is to restrict attention to particular
covering surfaces X obtained from the commutator subgroup of the triangle group H ,
where there is always a good supply of (meromorphic) automorphic forms which have
the following crucial property:
Definition. An automorphic form ϕ of weight 4 (or quadratic differential) for a fi-

nite co-volume Fuchsian group is of Jenkins–Strebel type if the set of critical real
trajectories of ϕ is compact in X .
For such forms, the complementary noncritical set – which has full measure in X –
is a union of cylinders, each comprising a free homotopy class of closed trajectories.
Incidentally, the forms occurring in Veech’s examples clearly have the J–S property in
the vertical direction: one changes the direction of the trajectories from horizontal to
vertical by a quarter-turn rotation of the Euclidean structure which means switching
to the form −ϕ. It is natural to widen the class of J–S differentials to include all forms
of finite L1 norm which have some direction in which the trajectory pattern has this
property.
We note that forms of J–S type are dense in the space A14 (X) for every compact
Riemann surface [12]; their construction is usually effected by either analytic or com-
binatorial means and no clear arithmetic connection has yet been made. For more
details, including the essential part played by J–S forms in the solution of a host of
extremal problems in conformal mapping and analysis, see for instance [13], [16],
[67].
In all the examples described in Section 4, where is a finite index subgruoup
of a Hecke triangle group, it was possible to use the structure of the covering (or
equivalently a fundamental domain for the group) triangulated in canonical fashion,
to produce J–S forms from the hyperbolic geometry; our next result follows this line of
approach. We shall later (in Section 6) restrict attention to subgroups of the modular
group PSL2 (Z)(= H3 ), where a more direct approach is possible.
Theorem 10. Let H be a Fuchsian triangle group, with (at least) one torsion point and
one cusp; assume that there is a reflection symmetry fixing a cusp and a torsion point.
Let ω be an invariant holomorphic quadratic differential on some smooth Belyi cover
X1 = U/ 1 of the compact orbifold Riemann sphere XH , which is -symmetric and
corresponds to some character of H . Then the corresponding Teichmüller disc Dω in
T (X1 ) ∼
= Tg,k has stabiliser in the corresponding modular group Modg,k isomorphic
to a subgroup of H of finite index containing 1 .
Proof. We move to a conjugate of H in the unit disc so as to locate the fixed point of one
torsion generator x at the centre 0 of the disc with the real ray from 0 to 1 fixed by the
symmetry and horizontal for the form ω. In the representation Dω = π eω (D), this
implies that there is a finite order rotational symmetry for the Euclidean structure on the
central reference surface of this disc, which is a compact abelian covering surface X1
on which the form ω is defined, and with all cusps filled in. By the assumptions made
on the group H and the form, the critical trajectories of ω emanate from cusps of 1
and are contained in the H -orbit of the edge subset of some -symmetric fundamental
polygonal set P ∪ P , with P = (P ) and P containing the ray . It follows that the
critical graph of ω is compact in X1 , which implies that ω is of Jenkins–Strebel type,
with all non-critical real trajectories closed.
Now consider the variation of complex structure on the surface as we follow the
Teichmüller ray from the origin to 1 ∈ ∂ D: 1 is a cusp of 1 at which ω vanishes and
this ray determines a deformation of X1 by shrinking the lengths of all closed vertical
trajectories to zero. We claim that the disc D is invariant under a reducible (unipotent)
mapping class which effects some composition of twists about a representative set of
core loops, one from each cylinder in the non-critical trajectory pattern of the form ω
on X1 . To verify this, let t be the primitive parabolic element of H fixing the cusp 1, and
consider the effect of applying conjugation adt to the (marked) subgroup 1 : the rule
adt (k) = tkt −1 defines an element of mod 1 ∼ = Mod(1 ), the mapping class group
∗
of the pointed surface X1 . This element is reducible in the Thurston classification
of mapping classes [69], since it preserves the set of vertical core curves, but it acts
as some cyclic permutation both on the set of cusps and also on the set of trajectory
cylinders of ω. A mapping class of the required kind is induced on the compact
surface by applying such an action at the level of 1 using the smallest power t ν of the
element t that preserves each trajectory cylinder. If we denote by d the degree at the
cusp 1 of the covering X0 → P1 ∼ = XH , then 1 ≤ ν ≤ d. It follows that the modular
stabiliser of Dω contains at least the nonelementary subgroup x, t ν of H .
To see that the modular stabiliser contains the subgroup 1 , we recall the discussion
of Section 4 where the structure of the Bers–Teichmüller fibre spaces was sketched
out: in the current situation, we have a T-disc embedded in the universal Teichmüller
space T (1) on which the group H acts as a subgroup of Mod(1). To descend to T (X1 )
and its fibre space F (X1 ) involves passing to the classes of qc mapping invariant under
1 , isomorphic to the iterated fibre space F (k) (X1 ) of moduli for k-pointed surfaces.
The forgetful map π which fills in all cusps then defines a projection to the genus g
Teichmüller space Tg .
It follows from the definition of the successive Bers fibrations, beginning with the
surface X1 as fibre in F (1 ) over the point 0 = [X1 ] ∈ T (X1 ), and the Bers fiber
space theorem (discussed in Section 4.3) that the stabiliser in modg,k ∼ = Modg,k+1
of the fibre space over the disc Dω contains the group 1 . In fact this stabiliser is a
subgroup of the k-strand braid group of the surface X1 , and one has an extension of
groups within the modular group Modg of the smooth compact surface X1 :
1 −→ π1 (X1 ) → E → π1 (X1∗ ) = 1 → 1.
The action of 1 on the kernel (isomorphic to the fundamental group of the compact
fibre surface) is by conjugation on E as a subgroup of mod(X1 ). After some careful
working out, following the discussion in [13] of the relationship with the real group
of automorphisms of the Euclidean planar structure on X1 , GL(2, R), this also turns
out, to identify 1 with the subgroup which Veech in [70] calls Aff(u), the stabiliser
of the F -structure of the form ω.
Now there is a complex analytic subspace of the tangent bundle to F (k) (X1 ) that is
preserved under the action of E; it constitutes one sheet of the Teichmüller geodesic
flow on T (X1 ) lifted to the fibre space. Alternatively, one may view it as an example
of the type of construction of analytic complex surface given in [21]. It follows that
the modular stabiliser of Dω has finite index in H .
We describe one way to prove existence of the forms required in this theorem.
Work with an inclusion of co-compact Fuchsian groups 0 ≤ H0 which provides a
model of the ramified surface covering which completes the Hecke group picture: one
follows the method detailed in [52] to construct generators of fractional weight for
the ring of all holomorphic character-automorphic forms on the co-compact genus 0
Fuchsian group H0 ; each of the generating forms has a simple zero at just one vertex.
It is then possible to produce (by transferring to the punctured surface and lifting to
the universal cover) weight 4 multiplicative holomorphic forms ω of the same type
on the group H with zeros only at designated cusp vertex points of the hyperbolic
tesselation of X determined by the group inclusion 0 < H0 .
Corollary 1. Let H be a triangle group of type {2, q, ∞}. Then there are holomorphic
forms for the commutator surface of H whose modular stability group has finite index
in H .
The cusp forms of Jenkins–Strebel type for the groups arising in theVeech examples
emerge directly from the Euclidean geometry on a given polygonal model surface, as
Veech observes in [70]. The commutator surface has ε = gcd(2, q) cusps, and the
subgroup of mapping classes produced by the theorem at a torsion point of the disc
coincides in this case with either the whole group H or a triangular subgroup of index 2,
generated in each case by the two elements x ε and t.
For further details and developments from the above result, the reader may consult
[41], which discusses Teichmüller discs stabilised by a pseudo-Anosov mapping class,
and [13], where a clear description is given of some further illuminating examples of
polygonal type with detailed formulae for the relation between the T-disc deformation
and the affine polygonal F -structure from which it is defined.
For Riemann surfaces X given by finite index subgroups of general triangle
groups, one may construct cusp forms of J–S type using the methods indicated above,
but a more natural way follows a (less direct) path through the tesselations by ideal
quadrangles induced from the standard fundamental region for (2); again one sees
from the topology that the stabiliser in the Teichmüller modular group of the cusped
surface X contains the subgroup of (2) which classifies X and so is of finite
index, by examination of the structural properties of the geometric fibre space over
the Teichmüller disc.
5.3 Recent work on Veech curves in Mg
It is unclear at present how to characterise the types of Fuchsian group that occur as
the Veech (modular) stability group of a Teichmüller disc in given genus, despite the
great activity in this area in recent years. For the case of genus 2, however, the picture
is now completely understood, thanks largely to striking recent work by C. McMullen
and others, which has produced a classification partly based on an explicit type of
Euclidean decomposition of a compact (genus 2) surface X as a translation surface
with rectangular L-shape. With specific side-lengths, these structures will have lattice
Fuchsian stabiliser, and by bringing to the fore the dynamical structure of the Teich-
müller flow on such an X, he provides a striking trichotomy for the case of abelian
T-discs, where the form q = ω2 with ω an abelian 1-form. The papers [48] and [50]
bring out the presence of hidden complex rank 2 symmetry in M2 by establishing
the existence of T-discs Dq with q = ω2 , ω√∈ 1 (X) with rectangular L-shape and
side lengths in a quadratic number field Q( d), such that the projected image of the
complex curve in the Siegel moduli space of dimension 2 abelian varieties has closure
a Hilbert modular surface. This is a homogeneous algebraic surface, the quotient of a
product of upper half planes U × U by an (indiscrete) image in PSL2 (Od ), where Od
denotes the ring of integers in the real quadratic field tied to the Euclidean structure
on X, and it is a clear sign of the importance of Teichmüller–Veech curves that this
study has uncovered such rich structures within these familiar varieties of dimension
3 – it is well known that the complete varieties M̂3 and Â3 are birationally equal to
CP3 . Hubert and Lelièvre [30] have produced a highly refined analysis of the special
structures occurring for square-tiled surfaces in M2 , including asymptotic estimates
of the growth of high genus curves of this type.
Other recent work by Hubert and Schmidt [31] and by McMullen [49] provides
interesting geometrically defined examples of T-discs in low genus (g ≤ 4) with
infinitely generated Fuchsian stabilisers.
In a different but related direction concerning families of algebraic curves, work
of Cohen and Wolfart [9] showed that uniformising hyperbolic discs for a large class
of algebraic curves of genus g ≥ 2 admit modular embeddings into Shimura vari-
eties. This means that there is an embedding into a Siegel variety Ag parametrising
polarised abelian varieties of genus g, using constructions involving the hyperbolic
plane representation of the curve. One of their construction methods is related to
the special examples of Veech’s T-disc structures we examined where the quadratic
differential is the square of a 1-form, and by this (completely independent) process it
follows that, for instance, many finite index subgroups of (1) embed in the Siegel
modular group Sp(2g, Z) of the relevant genus as the symplectic monodromy group
of the closed 1-form defining the branched Euclidean structure on the surface. An
essentially equivalent construction of embedded T-discs in the Torelli–Siegel space
of a Riemann surface was given much earlier by [40]; also compare with the related
results mentioned above in the discussion of [48] on genus 2.
At the same time there have been important applications to the dynamical study
of the bundle 1 (Vg ) of holomorphic 1-forms on the Teichmüller curve Vg and the
various subspaces into which it divides under the action of SL2 (R) on the F -structure
coordinates which each point of the fibre space 1 (Vg ) represents. The (restricted)
Teichmüller geodesic flow is determined by this action and specifying the partition
of 2g − 2 defined by the set of orders occurring in the divisor of zeros of the 1-form
produces a connected component. Eskin and Okounkov [14] have computed the
volumes of some of these spaces by linking them to the numbers of distinct branched
coverings of tori of given degree. See also [39] for more precise geometric analysis of
the dynamics of this restricted geodesic flow generated on the moduli space of abelian
differentials.
5.4 Completion of a T-disc with large stabiliser

We continue with the assumptions of Section 5.2: is a group of finite index in
a group H of signature type {p, q, ∞}. Because the stability subgroup obtained in
Theorem 10 is itself a lattice group, it makes sense in the light of the discussion in
Section 4.8 to ask whether the modular image of the T-disc D in Mg has a natural
completion. This follows naturally in fact: the process of adjoining a point at each
cusp of the immersed surface occurs in the projective embedding of the moduli space
M̂g of stable genus g algebraic curves constructed by Deligne and Mumford, [11],
which is a projective (ergo compact Kähler) complex V-manifold containing Mg as a
Zariski open subset. For background on this completion of the space of non-singular
curves the reader may consult [42], [23], [24] and [75], [76].
We refer again to the discussion in Section 5.2 before Theorem 10 as prelude to the
next result, first stated in this form in [25], although it involves mainly a reinterpretation
of parts of Veech’s work.
Theorem 11. For each point {X} of the moduli space Mg represented by a compact
surface X which is a clean Galois Belyi covering of CP 1 , there is a complete algebraic
curve C in the stable compactification M̂g passing through {X} isomorphic to some
quotient surface of X by an abelian subgroup of Aut(X).
Proof. The existence of the affine T-curves was proved in Theorem 10. Examples such
as the original Veech–Wiman hyperelliptic curves and Fermat curves were met earlier.
We consider the natural completion of such a curve in the projectively embedded
variety M̂g of stable curves. The points of a completed curve C which lie in the
boundary divisor of M̂g then correspond, not necessarily bijectively, to surfaces with
nodes obtained by passing to the cusps of the stability subgroup, which produces a
noded Riemann surface by collapsing the vertical trajectories of the form. The curve C
is an abelian quotient because of the construction we used.
There is a co-compact triangle group with signature {2, m, n}, where n is the l.c.m.
of the ramification degrees at the points lying over ∞, and a corresponding finite
index subgroup representing the same curve with cyclic symmetry which has identical
ramification structure with this Belyi covering and determines a smooth surface with
the same underlying conformal structure as X. Since this is a smooth surface, the
periods will in fact satisfy the condition that m divides the l.c.m. of 2 and n while n
divides l.c.m.{2, m}.
5.5 Consequences; complete curves in modular varieties

We add some brief comments on the last theorem.
Firstly, it is worth noting that the immersed curves C in moduli space are likely in
general to have singularities such as self intersections, although these can always be
removed by passage to a finite sheeted covering of M̂g . In fact it follows immediately
from a topological property of the orbifold strucure that there is a finite Galois cover
of the modular variety which is a compact complex (projective) manifold. (This is
proved in [42].) However, the genus of the resulting lifted Teichmüller–Veech curve
is then much larger and may not be easy to estimate.
A second point concerns the general occurrence of surface groups as subgroups in
Teichmüller modular groups.
The hyperbolic plane U = X̃ ∼ = D forms the fibre over the point x = [X] of
Mg , with X = U/K, say, for some Fuchsian group K ∼ = π1 (X). This determines a
different holomorphic embedding of a disc into the space Fg,k ∼ = F (K) = T (K ) ∼=
Tg,k+1 , and in the situation of Section 5.2 the Fuchsian group K which is acting on
D, and which is naturally contained in the corresponding extended modular group
mod(K), survives the projection homomorphism to the group Mod(K) ∼ = Modg,k for
some k > 0. However, it was proved by Kra [41] (and independently by Nag) that
such fibre discs in Teichmüller spaces cannot be Teichmüller-geodesic. Consequently
these two constructions determine completely distinct ways to realise Fuchsian groups
acting on the upper half plane as geometric subobjects of the Teichmüller modular
transformation group actions Modg,k .
As a final point, we observe that the degree of any covering which may occur in
Theorem 11 for a T-curve in a given moduli space is bounded above by a linear function
of g, at most 4(g + 1) since this is the maximum order of any abelian automorphism
group for a complete curve of genus g ≥ 2. This bound may not be sharp. It is also
unclear whether other complete (not necessarily Teichmüller geodesic) curves exist in
Mg with larger automorphism group.
6 Towards a space of all Q-curves: a universal modular family

Among the candidate theories which point towards the desired goal of a universal
space of Q = Q-rational curves, so far achieved only in g = 2, there are links with
Teichmüller theory stemming from both the holomorphic and combinatorial sides of
the picture. The relationship we pursue here connects a class of finite index subgroups
of the modular group PSL(2, Z) and a subset of the space of conjugacy classes of
co-compact Fuchsian groups and there is evidence which suggests a deeper resonance
with algebraic number theory. In particular there is a clear relationship with a modular
action of Gal(Q) given in work of J. G. Thompson [68], which we summarise below.
6.1 A moduli space of Fuchsian groups

The space D of all lattice subgroups (defined to be discrete subgroups with finite
volume quotient) of the Lie group G = PSL(2, R) has a natural structure of metric
space as defined by Chabauty [8]. Within this there is a subset D0 consisting of all
lattice subgroups with co-compact quotient. According to a fundamental theorem of
A. M. Macbeath [43] this set is open in D. Furthermore the geometric isomorphism
type of the discrete groups in a connected component is fixed, and each component
D() is, by standard results on Teichmüller-theory described earlier, a fibre space over
the space of moduli for X the corresponding type of hyperbolic closed surface or
2-orbifold, with the fibre over the modular point (conformal class of surface) x = []
isomorphic to the unit tangent bundle over the quotient surface/orbifold U/ ; the
base is the moduli space M(). This follows from results by Weil, by Macbeath [43],
[45] and by Harvey (see [23], Chapter 9).
A structure of stratified space in the sense of Whitney can also be put on D0 ,
which takes note of the various topological types of orbifold, and there is a natural
way to form the relative closure of D0 in the Chabauty space of G; see again for
instance [23], Chapter 9, where it is shown that the topological boundary structure of
the Deligne–Mumford boundary can be replicated in this setting.
Later in this chapter we shall formulate a theorem that makes the subset of con-
jugacy classes of finite index subgroups of (p, q, ∞)-triangle groups into both a set
of Q-rational base points for the moduli spaces of compact surfaces representable as
some finite sheeted covering of the sphere – this is just a paraphrase of Belyi’s theo-
rem – and also a data-bank for T-disc models of all the corresponding algebraic curves
defined by these coverings within the space of G-lattices.
Thus for the case of closed surface moduli spaces, each Mg will contain (after
passing to a suitable finite sheeted covering M̃g ) complex affine models of all curves
of genus g which carry a critically framed J–S form; these occur as subvarieties defined
within a specific model of the moduli space, presumably with the standard Q-structure
that it has.
6.2 A representation of the absolute Galois group
Let (1) = PSL(2, Z), the classical modular group, act as usual on the upper half
plane by Möbius transformations and denote by U∗ the completion by the cusps of
(1), U∗ = U ∪ Q ∪ ∞ and X∗ (1) = U∗ / (1) the corresponding compact quotient
surface. This is the unique compact Riemann surface which contains the open surface
X(1) ∼ = C, so it is of course the Riemann sphere, but provided with a wealth of further
information of an arithmetic nature, stemming from the isomorphism with the moduli
space of marked complex tori, elliptic curves with their special representations coming
from the classic theory of elliptic modular functions. There are many sources which
the reader may consult on this fascinating topic, for instance [37] and [65].
Following Thompson ([68]), we construct an exact sequence of Galois groups
associated with a certain space of meromorphic modular functions on this rational
bordification, which represents from one aspect an action of the absolute Galois group
Gal(Q) on Belyi coverings of the sphere; this is related to work of Ihara and others
on braid and modular group representations of the absolute Galois group. The field
of functions to be deployed forms a part of the regular representation of the modular
group (1) on the space H (U) of functions holomorphic on the upper half plane
with convergent Laurent–Puiseux expansions at each cusp which have the additional
property that they are algebraic in the standard local parameter. For instance, near
∞ we require an expansion in the variable q 1/N = exp(2π iτ/N) for some integer
N > 0.
Let F0 denote the field of all functions meromorphic in Û which determine func-
tions on some finite sheeted cover of the modular surface X(1)∗ . Clearly F0 con-
tains the subfield C(j ), where j (τ ) denotes the classical modular invariant, and it
is well known (see for instance [63], [65]) that when the group (1) acts on F0 by
σ (γ , f ) = f γ , the fixed field is precisely C(j ).
In fact, it is known by general results on modular functions that for any subgroup
of finite index in (1) there is a (meromorphic) function invariant under precisely ;
the profinite
standard methods of Galois theory then imply that Gal(F0 /C(j )) is (1),
completion of the modular group.
Now consider the following extension field of C(j ) introduced in [68].
Definition. The field of algebraic modular functions K is the relative algebraic closure
of Q(j ) in F0 .
We write G for the Galois group of this field over Q(j ). Then there is a short exact
sequence connecting Galois groups for the field extension.

Theorem 12. The subgroup fixing the intermediate field Q(j ) is isomorphic to (1)
and there is a split exact sequence
−→ G −→ Gal(Q/Q) → 1.
1 → (1) (1)
Here a constant field Galois automorphism σ ∈ Aut Q acts on the field of algebraic
modular functions via the action on coefficients of the polynomial ring Q(j )[U ]: any
f ∈ K is a root of some polynomial P (j, U ) and, by an observation of Thompson
(see [68]), the image P σ also has a root in K.
the section s : Gal(Q) → G defined in the following way:
The extension is split by
Let f ∈ K with f = n≥M an q n/N near ∞ and let σ ∈ Gal(Q). Then

f → s σ (f ) = anσ q n/N . (2)
n
This establishes a base for a common action of the absolute Galois group and
mapping class groups of all topological types on certain subspaces of the graded ring
of modular forms. For the case of congruence subgroups, it is then possible to establish
an arithmetic model for the function fields of all levels (this is done for instance in [65])
and generators for the ring of modular forms with (algebraic) integral q-coefficients
at ∞. In this sense, at least, the action of Q on (Galois) dessins can be subsumed into
this setting of a field of functions on U.
For noncongruence subgroups, none of this arithmetic approach applies directly
due to the lack of general methods such as a satisfactory theory of Hecke operators, but
over a complex coefficient field one can define as usual the correct (integer weight k)
action of γ ∈ SL2 (R) on H (U) as

−k a b
f → f [γ ]k (z) := f (γ (z)) · (cz + d) for γ = ∈ SL2 (R). (3)
c d

The space A4 = ≤(1) A4 () of all modular forms of weight 4 with poles of
order at most 1 at cusps of some finite index subgroup and having algebraic Fourier
coefficients may then represent an appropriate arithmetic core within the cotangent
bundles to moduli spaces of curves: the linear action of (1) on spaces A2k () of
forms of weight 2k, k ≥ 1 for a subgroup intertwines with that of the absolute
Galois group on q-expansion coefficients. This territory has not been much explored
for general (i.e. non-congruence) subgroups of (1), but the reader may consult several
articles [4], [62] where this approach is applied to determining fields of definition and
other arithmetic problems for non-congruence modular subgroups.
6.3 Modular forms with closed trajectories

We return to the bridge between the class of Jenkins–Strebel modular forms and the
curves in moduli space discussed in Section 5.
Beginning with a finite index subgroup of the modular group (1), let f ∈ A4 ()
be a holomorphic automorphic form of weight 4, non-vanishing in U, with zeros at
some (non empty) finite subset of the cusps of which includes ∞.
In all the cases discussed in Section 4, where was a finite index subgroup of
some Hecke triangle group, it was possible to use the structure of the covering, or
equivalently a fundamental domain for the group, triangulated in a canonical way, to
produce J–S forms. We restrict attention now to the classical modular group where a
more direct approach is possible.
We make two observations which will be preparation for our final result. First,
given any compact surface X furnished with a Belyi covering f of the sphere, there
is a subgroup of finite index in the modular group (1) such that the covering f
is represented away from f −1 (∞) by the induced projection map of X onto X(1),
and with dessin the inverse image of the standard modular triangle decomposition of
the sphere (with barycentric triangulation). From the many alternative combinatorial
descriptions of the same ramified covering pattern which come from the various re-
finements of this polygonal decomposition of X, we choose a representation which
involves subgroups of the cartographic subgroup 0 (2) so as to coordinate with the
joint action ρ : G → Aut(F) on modular forms outlined in the exact sequence (1).
If the pattern of cells is a geometrical subdivision of a regular polygonal tesselation
with an even number 2n of sides for each polygon, then there is a dessin with • vertices
at the corners, vertices at midpoints of edges and with a vertex at the centre of each
cell and a unique form on the surface with zeros only at the points marked and with
compact critical graph. This dessin is obtained from a multivalued holomorphic form
on the complex line X0∗ = U/ 0 (2) ∪ ∞ ∼ = P1 which vanishes at ∞; it arises by use
of Singerman’s theorem on universal tesselations ([66]) which gives an isomorphism
between this covering and the canonical covering of P1 that realises the tesselation by
regular 2n-gons by a specific finite index subgroup of the {2, 2n, ∞}-Hecke group.
If the cell pattern chosen is irregular at the vertices, then we need to make a preliminary
extra choice of covering surface which does have such a regular structure; for instance
we could take the intersection ∩ 0 (2n), a normal subgroup of that congruence
group.
The kind of F -structure that we saw in sections 2 and 3 arose from the prescription
of a quadratic differential with two interacting sets of closed trajectories and in addi-
tion a finite order mapping class, represented as a rotation of the coordinates for the
F -structure. In the Thurston–Veech square-tiled examples, there is a π/2 rotational
switch from vertical to horizontal which comes in when viewed in the modular group
framework, for the reason that we need to use the paving by Euclidean squares, whereas
in the case of the patterns of regular 2n-gons one obtains a quadratic differential
for a Hecke group with closed trajectories in directions making angle π/n. In that
case, the rotation acts as a translation at the cusp ∞ by a rational fraction, which
induces a multiplication by some 2n-th root of unity in the algebraic action ρ on
q-expansions.
In this way we see that the appropriate class of modular form for our purposes
should have closed trajectories in at least two independent directions related by a root
of unity.
Definition. An automorphic form ϕ of weight 4 for is called critically framed if

every non-critical trajectory is closed for the F -structures obtained in two independent
rationally related directions for ϕ on X = U/ .
It seems quite likely that the study of these purely geometric objects should have
intrinsic arithmetic interest, as they form natural ways to parametrise curves defined
over number fields analogous to the intensively studied modular elliptic curves; the
model curves in this setting are, however, usually noncongruence quotients.
6.4 Arithmetically defined curves in moduli space

We can now formulate a final result, valid for any complete Belyi curve X of genus
g ≥ 2.
Theorem 13. Given any pair X, β with X a genus g curve and β : X → X(1) a Belyi
map, there is a corresponding Veech curve in Mg , an affine algebraic curve over Q
arising from a critically framed modular form ϕ in A14 (X) which determines a T-disc
modular parametrisation eϕ centred at X.
The proof follows directly from the definitions and preliminary discussion of the
preceding section.
all such triples X, β, ω forms a subset of a fibre space over the modular
The set of
tower M̂ = g≥2 Mg on which the absolute Galois group acts in the manner of
equation (3) on Fourier q-coefficients of weight 4 modular forms. One expects that
some curves will have many such immersed models in their moduli space, just as the
hyperelliptic curves y 2 + x n = 1 have at least two, possibly more, distinct T-disc
parametrisations.
Having seen how the complex geometry of moduli spaces is linked to dessins
d’enfant by way of the collection of modular Veech curves, we close by indicating
another approach to the relationship between moduli, combinatorial patterns on closed
surfaces and hyperbolic geometry. In a paper of Macbeath ([44]) originating from
the same period as the ideas of Grothendieck and Thurston, it was shown that the
set of points of Mg which are represented by hyperbolic structures with Dirichlet
fundamental region having an inscribed circle, that is simultaneously tangent to all
sides, corresponds precisely to the multitude of genus g permutation representations
of the modular group (1); further evidence, perhaps, of how important mathematical
ideas have a life of their own. Macbeath’s points in Mg do represent curves defined
over a number field, by Belyi’s theorem.
References
Springer, Berlin 1980. 268
[2] L. V. Ahlfors, Lectures on quasiconformal mappings. Wadsworth & Brooks/Cole Ad-
vanced Books & Software, Monterey, CA, 1987; second edition, with supplemental chap-
ters by C. J. Earle, I. Kra, M. Shishikura and J. H. Hubbard, Univ. Lecture Ser. 38, Amer.
Math. Soc., Providence, RI, 2006. 267
[3] L. Ahlfors and L. Bers, Riemann’s mapping theorem for variable metrics. Ann. of Math.
(2) 72 (1960), 385–404. 259
[4] A. O. L. Atkin and H. P. F.Swynnerton-Dyer, Modular forms on non-congruence sub-
groups. In Combinatorics (Los Angeles, Calif., 1968), Proc. Sympos. Pure Math. 19,
Amer. Math. Soc., Providence, R.I, 1971, 1–25. 286
[5] G. V. Belyı̆, On Galois extensions of a maximal cyclotomic field. Izv. Akad. Nauk SSSR
Ser. Mat. 43 (1979), 267–276; English transl. Math. USSR-Izv. 14 (1980), 247–256. 252
Princeton Math. Ser. 24, Princeton University Press, Princeton, N.J., 1960, 89–119. 268
[7] L. Bers, Fiber spaces over Teichmüller spaces. Acta. Math. 130 (1973), 89–126. 264
[8] C. Chabauty, Limite d’ensembles et géométrie des nombres. Bull. Soc. Math. France 78
(1950), 143–151. 284
[9] P. B. Cohen and J. Wolfart, Dessins de Grothendieck et variétés de Shimura. C. R. Acad.
Sci. Paris Sér. I Math. 315 (1992), 1025–1028. 281
[10] P. Cohen and J. Wolfart, Modular embeddings for some non-arithmetic Fuchsian groups.
Acta Arith. 56 (1990), 93–110.
[11] P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus. Inst.
Hautes Études Sci. Publ. Math. 36 (1969), 75–109. 282
[12] A. Douady and J. Hubbard, On the density of Strebel differentials. Invent. Math. 30 (1975),
175–179. 278
[13] C. J. Earle and F. P. Gardiner, Teichmüller disks and Veech’s F -structures. In Extremal
Riemann surfaces (San Francisco, CA, 1995), Contemp. Math. 201, Amer. Math. Soc.,
Providence, R.I., 1997, 165–189. 254, 266, 271, 272, 274, 276, 278, 280
[14] A. Eskin and A. Okounkov, Asymptotics of numbers of branched coverings of a torus and
volumes of moduli spaces of holomorphic differentials. Invent. Math. 145 (2001), 59–103.
282
[15] D. Gabai, Convergence groups are Fuchsian groups. Ann. of Math. (2) 136 (1992),
447–510. 262
York 1987. 258, 259, 261, 262, 264, 266, 278
76, American Mathematical Society, Providence, R.I., 2000. 259
[18] G. González-Diez, Variations on Belyi’s theorem. Quart. J. Math. Oxford (2) 57 (2006),
339–354. 252, 276
[19] G. González Díez and W. J. Harvey, On complete curves in moduli space I. Math. Proc.
Cambridge Philos. Soc. 110 (1991), 461–466. 254
[20] G. González-Díez and W. J. Harvey, On families of algebraic curves with automorphisms.
In Complex manifolds and hyperbolic geometry (Guanajuato, 2001), Contemp. Math. 311,
Amer. Math. Soc., Providence, R.I., 2002, 261–276.
[21] P. A. Griffiths, Complex-analytic properties of certain Zariski open sets on algebraic
varieties. Ann. of Math. (2) 94 (1971), 21–51. 280
[22] A. Grothendieck, Esquisse d’un programme. In Geometric Galois actions. 1. Around

Grothendieck’s “Esquisse d’un programme” (Luminy, France, August 1995), Lond. Math.
Soc. Lect. Note Ser. 242, Cambridge University Press, Cambridge 1997, 5–48; English
transl. “Sketch of a programme”, 241–283. 251, 252, 254, 255, 256
[23] W. J. Harvey (ed.), Discrete groups and automorphic functions (Cambridge, July 28–
August 15, 1975). Academic Press, London 1977. 258, 261, 282, 284
[24] W. J. Harvey, Modular groups and representation spaces. In Geometry of group repre-
sentations (Boulder, CO, 1987), Contemp. Math. 74, Amer. Math. Soc., Providence, R.I.,
1988, 205–214. 282
[25] W. J. Harvey, On certain families of compact Riemann surfaces. In Mapping class groups
and moduli spaces of Riemann surfaces (Göttingen, June 24–28, 1991 and Seattle, WA,
August 6-10, 1991), Contemp. Math. 150, Amer. Math. Soc., Providence, R.I., 1993,
137–148. 273, 282
[26] W. J. Harvey, Drawings, triangle groups and algebraic curves. KCL Preprint MTH-97-12,
Feb. 1997. 250
[27] W. J. Harvey and C. Maclachlan, On mapping-class groups and Teichmüller spaces. Proc.
London Math. Soc. (3) 30 (1975), 496–512. 261
[28] E. Hecke, Über die Bestimmung Dirichletscher Reihen durch ihre Funktionalgleichung.
Math. Ann. 112 (1936), 664–699; reprinted in Mathematische Werke, Vandenhoeck &
Ruprecht, Göttingen 1959, 591–626. 270
[29] F. Herrlich and G. Schmithüsen, Dessins d’enfants and origamis. In Handbook of Teich-
müller theory (A. Papadopoulos, ed.), Volume II, EMS Publishing House, Zurich 2008.
255
[30] P. Hubert and S. Lelièvre, Prime arithmetic Teichmüller discs in H(2). Israel J. Math.
151 (2006), 281–321. 281
[31] P. Hubert and T. A. Schmidt, Veech groups and polygonal coverings. J. Geom. Phys. 35
(2000), 75–91. 281
[32] J. A. Jenkins, Univalent functions and conformal mapping. Ergeb. Math. Grenzgeb. Neue
Folge 18, Springer-Verlag, Berlin 1958. 266
[33] G. Jones and D. Singerman, Belyı̆ functions, hypermaps and Galois groups. Bull. London
Math. Soc. 28 (1996), 561–590. 255, 256, 276
[34] S. P. Kerckhoff, The asymptotic geometry of Teichmüller space. Topology 19 (1980),
23–41. 263, 267
262, 263
[36] S. Kerckhoff, H. Masur, and J. Smillie, Ergodicity of billiard flows and quadratic differ-
entials. Ann. of Math. (2) 124 (1986), 293–311. 266
[37] A. W. Knapp, Elliptic curves. Math. Notes 40, Princeton University Press, Princeton, N.J.,
1992. 285
[38] K. Kodaira, A certain type of irregular algebraic surfaces. J. Analyse Math. 19 (1967),
207–215. 254
[39] M. Kontsevich and A. Zorich, Connected components of the moduli spaces of Abelian
differentials with prescribed singularities. Invent. Math. 153 (2003), 631–678. 282
[40] I. Kra, The Carathéodory metric on abelian Teichmüller disks. J. Analyse Math. 40 (1981),
129–143. 282
[41] I. Kra, On the Nielsen-Thurston-Bers type of some self-maps of Riemann surfaces. Acta
Math. 146 (1981), 231–270. 280, 283
[42] E. Looijenga, Smooth Deligne-Mumford compactifications by means of Prym level struc-
tures. J. Algebraic Geom. 3 (1994), 283–293. 282, 283
[43] A. M. Macbeath, Groups of homeomorphisms of a simply connected space. Ann. of Math.
(2) 79 (1964), 473–488. 270, 284
[44] A. M. Macbeath, Generic Dirichlet polygons and the modular group. Glasgow Math. J.
27 (1985), 129–141. 288
[45] A. M. Macbeath and D. Singerman, Spaces of subgroups and Teichmüller space. Proc.
London Math. Soc. (3) 31 (1975), 211–256. 264, 270, 284
[46] H. Masur, Dense geodesics in moduli space. In Riemann surfaces and related topics
(State University New York, Stony Brook, N.Y., 1978), Ann. of Math. Stud. 97, Princeton
University Press, Princeton, N.J., 1981, 417–438. 274
[47] H. Masur, Closed trajectories for quadratic differentials with an application to billiards.
Duke Math. J. 53 (1986), 307–314. 254
[48] C. T. McMullen, Billiards and Teichmüller curves on Hilbert modular surfaces. J. Amer.
Math. Soc. 16 (2003), 857–885. 281, 282
[49] C. T. McMullen, Teichmüller geodesics of infinite complexity. Acta Math. 191 (2003),
191–223. 281
[50] C. T. McMullen, Dynamics of SL2 (R) over moduli space in genus 2. Ann. of Math. 165
(2007), to appear. 281
[51] C. T. McMullen, Prym varieties and Teichmüller curves. Duke Math. J. 133 (2006),
569–590.
[52] J. Milnor, On the 3-dimensional Brieskorn manifolds M(p, q, r). In Knots, groups, and
3-manifolds, Ann. of Math. Stud. 84, Princeton University Press, Princeton, N.J., 1975,
175–225. 270, 280
[53] S. Morita, Geometry of characteristic classes. Transl. Math. Monographs 199, American
Mathematical Society, Providence, R.I., 2001. 263
[54] S. Nag, The complex analytic theory of Teichmüller spaces. Canad. Math. Soc. Ser. Mono-
graphs Adv. Texts, John Wiley & Sons, New York 1988. 258, 262, 264
[55] S. Nag and D. Sullivan, Teichmüller theory and the universal period mapping via quantum
calculus and the H 1/2 space on the circle. Osaka J. Math. 32 (1995), 1–34. 274
[56] R. Nevanlinna et al. (eds.), Analytic Functions. Princeton Math. Ser. 24, Princeton Uni-
versity Press, Princeton N.J., 1960. 251
[57] B. Riemann, Gesammelte mathematische Werke und wissenschaftlicher Nachlaß. Dover
Publications, Inc., New York, N.Y., 1953. 251
[58] H. L. Royden, Automorphisms and isometries of Teichmüller space. In Advances in the
Theory of Riemann Surfaces (Proc. Conf., Stony Brook, N.Y., 1969), Ann. of Math. Stud.
66, Princeton University Press, Princeton, N.J., 1971, 369–383. 261
[59] T. A. Schmidt, Klein’s cubic surface and a ‘non-arithmetic’ curve. Math. Ann. 309 (1997),
533–539.
[60] L. Schneps and P. Lochak (eds.), Geometric Galois actions. Vol. 1, 2 (Luminy, France,
August 1995). London Math. Soc. Lecture Note Ser. 242, 243, Cambridge University
Press, Cambridge 1997.
[61] L. Schneps (ed.), The Grothendieck theory of dessins d’enfants (Luminy, France, April
1993). London Math. Soc. Lecture Note Ser. 200, Cambridge University Press, Cambridge
1994. 255, 256
[62] A. J. Scholl, Modular forms on noncongruence subgroups. In Séminaire de Théorie des
Nombres, Paris 1985–86, Progr. Math. 71, Birkhäuser, Boston 1987, 199–206. 286
[63] J.-P. Serre, Cours d’arithmétique. Deuxième édition revue et corrigée. Presses Universi-
taires de France, Paris 1977. 285
[64] G. B. Shabat and V.A. Voevodsky, Drawing curves over number fields. In The Grothendieck
Festschrift, Vol. III, Progr. Math. 88, Birkhäuser, Boston 1990, 199–227. 252, 256
[65] G. Shimura, Introduction to the arithmetic theory of automorphic functions. Publ. Math.
Soc. Japan 11, Princeton University Press, Princeton, N.J., 1994. 285, 286
[66] D. Singerman, Universal tessellations. Rev. Mat. Univ. Complut. Madrid 1 (1988),
111–123. 275, 287
1984. 266, 278
[68] J. G. Thompson, Galois groups. In Groups – St Andrews 1989, Vol. 2, London Math. Soc.
Lecture Note Ser. 160, Cambridge University Press, Cambridge 1991, 455–462. 284, 285,
286
Amer. Math. Soc. (N.S.) 19 (1988), 417–431. 257, 267, 279
[70] W. A. Veech, Teichmüller curves in moduli space, Eisenstein series and an application
to triangular billiards. Invent. Math. 97 (1989), 553–583; corrigendum, ibid. 103 (1991),
447. 266, 269, 271, 272, 276, 280
[71] W. A. Veech, Moduli spaces of quadratic differentials. J. Analyse Math. 55 (1990),
117–171. 272, 273, 274
[72] W. A. Veech, The billiard in a regular polygon. Geom. Funct. Anal. 2 (1992), 341–379.
269, 273
[73] A. Weil, The field of definition of a variety. Amer. J. Math. 78 (1956), 509–524. 276
[74] J. Wolfart, The ‘obvious’ part of Belyi’s theorem and Riemann surfaces with many auto-
morphisms. In Geometric Galois actions, Vol. 1, London Math. Soc. Lecture Note Ser.
242, Cambridge University Press, Cambridge 1997, 97–112. 252, 276
[75] S. Wolpert, On the homology of the moduli space of stable curves. Ann. of Math. (2) 118
(1983), 491–523. 282
[76] S. Wolpert, On obtaining a positive line bundle from the Weil-Petersson class. Amer. J.
Math. 107 (1985), 1485–1507. 262, 282
Chapter 6
On the boundary of Teichmüller disks in Teichmüller

and in Schottky space
Frank Herrlich and Gabriela Schmithüsen
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
2 Geodesic rays, Teichmüller disks and Teichmüller curves . . . . . . . . . . 300
2.1 Teichmüller deformations . . . . . . . . . . . . . . . . . . . . . . . . 300
2.2 Geodesic rays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
2.3 Teichmüller disks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
2.3.1 Teichmüller disks as a collection of geodesic rays . . . . . . . 304
2.3.2 Teichmüller disks by affine deformations . . . . . . . . . . . . 304
2.3.3 Beltrami differentials . . . . . . . . . . . . . . . . . . . . . . 310
2.4 Teichmüller curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
2.4.1 Veech groups . . . . . . . . . . . . . . . . . . . . . . . . . . 311
2.4.2 The action of the Veech group on the Teichmüller disk . . . . 312
2.4.3 Veech groups and Teichmüller curves . . . . . . . . . . . . . 314
3 Braungardt’s construction of Tg,n . . . . . . . . . . . . . . . . . . . . . . 314
3.1 Coverings with cusps . . . . . . . . . . . . . . . . . . . . . . . . . . 315
3.2 The cusped universal covering of Mg,n . . . . . . . . . . . . . . . . . 317
3.3 Teichmüller structures . . . . . . . . . . . . . . . . . . . . . . . . . 319
4 Boundary points of Teichmüller curves . . . . . . . . . . . . . . . . . . . 321
4.1 Hitting the boundary via a Jenkins–Strebel ray . . . . . . . . . . . . . 322
4.1.1 X as patchwork of rectangles . . . . . . . . . . . . . . . . . . 323
4.1.2 Stretching the cylinders . . . . . . . . . . . . . . . . . . . . . 325
4.1.3 S as patchwork of double annuli . . . . . . . . . . . . . . . . 326
4.1.4 Contracting the central lines . . . . . . . . . . . . . . . . . . 327
4.1.5 The end point of the Jenkins–Strebel ray . . . . . . . . . . . . 328
4.1.6 Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 329
4.2 Boundary points of Teichmüller disks . . . . . . . . . . . . . . . . . 330
4.2.1 Extending Teichmüller embeddings to the cusps of the mirror
Veech group . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
4.2.2 Boundary of Teichmüller disks that lead to Teichmüller curves 334
5 Schottky spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
5.1 Schottky coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
5.2 Relation to Teichmüller space . . . . . . . . . . . . . . . . . . . . . 339
294 Frank Herrlich and Gabriela Schmithüsen
5.3 Schottky coverings of stable Riemann surfaces . . . . . . . . . . . . 341

5.4 Sg as quotient of Tg . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
5.5 Teichmüller disks in Schottky space . . . . . . . . . . . . . . . . . . 346
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
1 Introduction
One of the original motivations that led to the discovery of Teichmüller space was to
better understand the classification of Riemann surfaces. Riemann himself already
saw that the compact Riemann surfaces of genus g with n marked points on it depend
on 3g − 3 + n complex parameters (if this number is positive). More precisely, there
is a complex analytic space Mg,n whose points correspond in a natural way to the
isomorphism classes of such Riemann surfaces. Mg,n is even an algebraic variety, but
its geometry is not easy to understand. Most of the basic properties are known today,
but many questions on the finer structure of Mg,n are still open.1
Many classification problems become more accessible if the objects are endowed
with an additional structure or marking. The general strategy is to first classify the
marked objects and then, in a second step, to try to understand the equivalence relation
that forgets the marking. The markings that Teichmüller introduced for a compact
Riemann surface X consist of orientation preserving diffeomorphisms f : Xref →
X from a reference Riemann surface Xref to X. Markings (X, f ) and (X , f ) are
considered the same if f f −1 is homotopic to a biholomorphic map. Thus different
markings of a fixed Riemann surface differ by a homotopy class of diffeomorphisms
of Xref . In other words the mapping class group (or Teichmüller modular group)
g = Diffeo+ (Xref )/ Diffeo0 (Xref ) (1.1)
acts on the set Tg of all marked Riemann surfaces of genus g, and the orbit space
Tg / g is equal to Mg (here Diffeo+ (Xref ) denotes the group of orientation preserving
diffeomorphisms of Xref and Diffeo0 (Xref ) the subgroup of those that are homotopic
to the identity).
Teichmüller discovered that in each homotopy class of diffeomorphisms between
compact Riemann surfaces X and X there is a unique “extremal mapping”, i.e. a
quasiconformal map with minimal dilatation. The logarithm of this dilatation puts
a metric on Tg , the Teichmüller metric. With it Tg is a complete metric space, dif-
feomorphic to R6g−6 , and g acts on Tg by isometries. There is also a structure as
complex manifold on Tg , for which the elements of g act holomorphically and thus
make the quotient map Tg → Mg into an analytic map between complex spaces.
That the complex structure on Tg is the “right one” for the classification prob-
lem can be seen from the fact that there is a family Cg of Riemann surfaces over
Tg which in a very precise sense is universal. This family can be obtained as fol-
lows: By the uniformization theorem, the universal covering of a compact Riemann
1Although we consider this general setting in a large part of this chapter, we shall restrict ourselves in this
introduction to the case n = 0 and write, as usual, Mg instead of Mg,0 (and later Tg instead of Tg,0 ).
Chapter 6. The boundary of Teichmüller disks in Teichmüller and in Schottky space 295
surface X of genus g ≥ 2 is (isomorphic to) the upper half plane H. Any marking
f : Xref → X induces an isomorphism f∗ from πg = π1 (Xref ), the fundamental group
of the reference surface, to π1 (X). We may obtain a holomorphic action of πg on
Tg × H as follows: for γ ∈ g , x = (X, f ) ∈ Tg and z ∈ H put
γ (x, z) = (x, f∗ (γ )(z)),
where we identify π1 (X) with the group of deck transformations of the universal
covering H → X. The quotient Cg = (Tg × H)/πg is a complex manifold with a
natural projection p : Cg → Tg ; the fibre p−1 (X, f ) is isomorphic to X. Moreover p is
proper and therefore p : Cg → Tg is a family of Riemann surfaces. The representation
of Cg as a quotient of a manifold by an action of πg is called a Teichmüller structure on
this family. It follows from results of Bers on the uniformization of families (see e.g.
[4, Theorem XVII]) that this family is universal, i.e. every other family of Riemann
surfaces of genus g with a Teichmüller structure can be obtained as a pullback from
p : Cg → Tg . In a more fancy language: Tg is a fine moduli space for Riemann
surfaces of genus g with Teichmüller structure.
It follows by the same arguments that for any family π : C → S of Riemann
surfaces (over some complex space S) there is an analytic map μ = μπ : S → Mg ,
which maps s ∈ S to the point in Mg that corresponds to the isomorphism class of
the fibre π −1 (s). Unfortunately, g does not act freely on Tg ; therefore the quotient
of Cg by the action of g does not give a universal family over Mg : the fixed points
of elements in g correspond to automorphisms of the Riemann surface, and the fibre
over [X] ∈ Mg in the family Cg / g → Mg is the Riemann surface X/ Aut(X) (whose
genus is strictly less than g if Aut(X) is nontrivial). As a consequence, Mg is not a
fine moduli space for Riemann surfaces, but only a “coarse” one (see e.g. [16, 1A] for
a precise definition of fine and coarse moduli spaces).
There are several equivalent ways to define markings of Riemann surfaces and
to describe Teichmüller space. Instead of classes of diffeomorphisms f : Xref → X
often conjugacy classes of group isomorphisms πg → π1 (X) are used as markings.
For the purpose of this chapter the approach to Teichmüller space via Teichmüller
deformations is very well suited; it is developed in Section 2.1. The starting point is
the observation that a holomorphic quadratic differential q on a Riemann surface X
defines a flat structure μ on X ∗ = X − {zeroes of q}. Composing the chart maps of μ
with a certain (real) affine map yields a new point in Tg . Any point in Tg is in a unique
way such a Teichmüller deformation of a given base point (Xref , id), cf. Section 2.2.
The main objects of interest in this chapter are Teichmüller embeddings, i.e. holo-
morphic isometric embeddings ι : H → Tg (or ι : D → Tg ), where H (resp. D) is
given the hyperbolic metric and Tg the Teichmüller metric, see Definition 2.4. The
restriction of ι to a hyperbolic geodesic line in H (or D) is then a (real) geodesic line
in Tg in the usual sense. The image ι of such an embedding ι is called a Teichmüller
geodesic or Teichmüller disk in Tg . There are plenty of Teichmüller disks in Tg . To
see this note first that the tangent space to Tg at a point x = (X, f ) ∈ Tg is natu-
rally isomorphic to the vector space QX = H 0 (X, ⊗2 X ) of holomorphic quadratic
differentials on X (this results from the Bers embedding of Tg as a bounded open sub-
domain of QX ). We shall explain in Section 2.3 in three different ways how one can,
for a given holomorphic quadratic differential q on a Riemann surface X, construct a
Teichmüller embedding ι : D → Tg with ι(0) = x and ι (0) = q. This shows that for
any x ∈ Tg and any (complex) tangent vector at x there is a Teichmüller disk passing
through x in direction of the given tangent vector.
There are several natural and closely related objects attached to a Teichmüller
disk ι (or a Teichmüller embedding ι : H → Tg ): The first is a discrete subgroup
of PSL2 (R) called the (projective) Veech group ι , cf. Section 2.4.1. If q is the
quadratic differential on the Riemann surface X by which ι is induced, ι consists
of the derivatives of those diffeomorphisms of X that are affine with respect to the
flat structure defined by q. Veech showed that this subgroup of PSL2 (R) is always
discrete ([34, Proposition 2.7]).
A second group naturally attached to ι is the stabilizer
Stab(ι ) = {ϕ ∈ g : ϕ(ι ) = ι }
of ι in the Teichmüller modular group. The pointwise stabilizer
Stab0 (ι ) = {ϕ ∈ g : ϕ|ι = idι }
is a finite subgroup of Stab(ι ), and Stab(ι )/ Stab0 (ι ) then is (via ι) a group of
isometries of H and thus a subgroup of PSL2 (R). This subgroup coincides with the
projective Veech group ι , see Section 2.4.3.
Given a Teichmüller embedding ι we are also interested in the image Cι of ι in
the moduli space Mg . The map ι → Cι obviously factors through the Riemann
surface H/ι or rather through its mirror image Vι , see Section 2.4 and in particular
2.4.3. The typical case seems to be Stab(ι ) = {id} (although it is not trivial to
give explicit examples). Much attention has been given in recent years to the other
extreme case that ι is a lattice in PSL2 (R). Then Vι is of finite hyperbolic volume and
hence a Riemann surface of finite type, or equivalently an algebraic curve. In this case
the induced map Vι → Cι is birational (see [11]), i.e. Vι is the desingularization (or
normalization) of Cι . It follows from a result of Veech ([34]) that Vι (and hence also Cι )
cannot be projective. If ι is a lattice, the affine curve Cι is called a Teichmüller curve,
cf. Section 2.4. First examples were given by Veech [34]; in them, ι is a hyperbolic
triangle group. Later more examples with triangle groups as Veech groups were found,
see [20] for a comprehensive overview and [6] for recent results. Explicit examples
for Teichmüller curves also with non triangle groups as Veech groups can be found
e.g. in [27], [9] and [23]. Möller has shown ([28]) that every Teichmüller curve is, as
a subvariety of Mg , defined over a number field. This implies that there are at most
countably many Teichmüller curves.
A special class of Teichmüller curves is obtained by origamis (or square-tiled
surfaces). They arise from finite coverings of an elliptic curve that are ramified
over only one point. Given such a covering p : X → E, the quadratic differential
q = (p ∗ ωE )2 (where ωE is the invariant holomorphic differential on E) induces a

Teichmüller embedding whose Veech group is commensurable to SL2 (Z), see [15].
Lochak proposed in [23] a combinatorial construction for such coverings (which led to
the name “origami”), and Schmithüsen [30] gave a group theoretic characterization of
the Veech group. In [31], origamis and their Veech groups are systematically studied
and numerous examples are presented. Origamis in genus 2 where q has one zero are
classified in [19]. Using the description of origamis by gluing squares it is not difficult
to see that there are, for any g ≥ 2, infinitely many Teichmüller curves in Mg that
come from origamis. In genus 3 there is even an explicit example of an origami curve
that is intersected by infinitely many others, see [18].
We want to study boundary points of Teichmüller disks and Teichmüller curves;
by this we mean, for a Teichmüller embedding ι : H → Tg , the closures of ι and
Cι in suitable (partial) compactifications of Tg and Mg , respectively. For the moduli
space we shall use the compactification Mg by stable Riemann surfaces. Here a
one-dimensional connected compact complex space X is called a stable Riemann
surface if all singular points of X are ordinary double points, i.e. have a neighbourhood
isomorphic to {(z, w) ∈ C2 : z · w = 0, |z| < 1, |w| < 1}; moreover we require
that every irreducible component L of X that is isomorphic to the projective line
Ĉ = P1 (C) intersects X − L in at least three points. It was shown by Deligne and
Mumford ([10]) that stable Riemann surfaces are classified by an irreducible compact
variety Mg that, like Mg , has the quality of a coarse moduli space. In fact, with the
approach of Deligne–Mumford it is possible to classify stable algebraic curves over
an arbitrary ground field: they construct a proper scheme over Z of which Mg is the
set of complex-valued points. Some years later, Knudsen [22] showed that Mg is a
projective variety.
If ι : H → Tg is a Teichmüller embedding such that Cι is a Teichmüller curve, the
closure Cι of Cι in Mg is Zariski closed and therefore a projective curve. In particular,
Cι − Cι consists of finitely many points, called the cusps of Cι . It is very interesting to
know, for a given Teichmüller curve Cι , the number of cusps and the stable Riemann
surfaces that correspond to the cusps. In the case that ι is induced by an origami there
is an algorithm that determines (among other information) the precise number of cusps
of Cι , see [30].
The boundary ∂Mg = Mg −Mg is a divisor, i.e. a projective subvariety of (complex)
codimension 1. It has irreducible components D0 , D1 , . . . , D[ g ] ; the points in D0
2
correspond to irreducible stable Riemann surfaces with a double point, while for
i = 1, . . . , g2 , Di classifies stable Riemann surfaces consisting of two nonsingular
irreducible components that intersect transversally, one of genus i and the other of
genus g −i. The combinatorial structure of the intersections of the Di is best described
in terms of the intersection graph. For a stable Riemann surface X, we define a graph
(X) as follows: the vertices of (X) are the irreducible components of X, the edges
are the double points (connecting two irreducible components of X which need not
be distinct). For every graph let Mg () be the set of points in Mg corresponding to
stable Riemann surfaces with intersection graph isomorphic to . It is not hard to see
that for a given genus g, there are only finitely many graphs with nonempty Mg (),
and that the Mg () are the strata of a stratification of Mg . This means that each Mg ()
is a locally closed subset of Mg (for the Zariski topology), that Mg is the disjoint union
of the Mg (), and that the closure of each Mg () is a finite union of other Mg ( ). A
natural question in our context is: which Mg () contain cusps of Teichmüller curves?
In [24] Maier showed that if has no “bridge”, i.e. no edge e such that − e is
disconnected, then the stratum Mg () contains points on a compactified Teichmüller
curve Cι with a Teichmüller embedding ι that corresponds to an origami. Möller and
Schmithüsen observed that this condition on the graph is necessary if the Teichmüller
curve comes from a quadratic differential which is the square of a holomorphic 1-form
(or equivalently from a translation structure on X∗ ).
Most of our knowledge about cusps of Teichmüller curves comes from studying
boundary points of Teichmüller disks in a suitable extension of Teichmüller space.
Several different boundaries for Teichmüller space with very different properties have
been studied, like the Thurston boundary or the one coming from the Bers embedding.
In the framework of this chapter we look for a space Tg in which Tg is open and
dense such that the action of the group g extends to an action on Tg , and the quotient
space Tg / g is equal to Mg . Such a space is the “augmented” Teichmüller space Tˆg
introduced by Abikoff [1]. The points in Tˆg are equivalence classes of pairs (X, f ),
where X is a stable Riemann surface of genus g and f : Xref → X is a deformation.
This is a continuous surjective map such that there are finitely many loops c1 , . . . , ck
on Xref with the property that f is a homeomorphism outside the ci and maps each ci
to a single point Pi on X. Abikoff defined a topology on this space and showed that
the quotient for the natural action of g on the pairs (X, f ) is the moduli space Mg as
a topological space.
In his thesis [8], Braungardt introduced the concept of a covering of a complex
manifold S with cusps over a divisor D. He showed that under mild assumptions on S
there exists a universal covering X̃ of this type which extends the usual holomorphic
universal covering of S − D by attaching “cusps” over D. X̃ is no longer a complex
manifold or a complex space, but Braungardt introduced a natural notion of holomor-
phic functions in a neighbourhood of a cusp and thus defined a sheaf OX̃ of rings (of
holomorphic functions) on X̃. In this way X̃ is a locally complex ringed space, and
the quotient map X̃ → X̃/π1 (S − D) = S is analytic for this structure. When applied
to S = Mg and D = ∂Mg , Braungardt showed that the universal covering Tg of Mg
with cusps over ∂Mg is, as a topological space with an action of g , homeomorphic
to Abikoff’s augmented Teichmüller space. We shall reserve the symbol Tg exclu-
sively for this space (considered as a locally ringed space). In Section 3, we review
Braungardt’s construction and results.
Our key technique to investigate boundary points of Teichmüller disks is the use
of Jenkins–Strebel rays, see Definition 4.4. By this we mean a geodesic ray in Tg
that corresponds by the construction in Section 2.2 to a Jenkins–Strebel quadratic

differential on the Riemann surface X at the starting point of the ray. A Jenkins–Strebel
differential decomposes X into cylinders swept out by horizontal trajectories. Mainly
following [26] we give in Section 4.1 two explicit descriptions of the marked Riemann
surfaces (XK , fK ) (for K > 1) on a Jenkins–Strebel ray. This allows us to identify
the boundary point (X∞ , f∞ ) at the “end” of the ray as the stable Riemann surface
that is obtained by contracting on X the core lines of the cylinders in a prescribed way,
see Sections 4.1.5 and 4.1.6.
In the case that the Teichmüller embedding ι leads to a Teichmüller curve Cι we
show in Section 4.2 that all boundary points of ι are obtained in this way. This shows
in particular that all cusps of Teichmüller curves are obtained by contracting, on a cor-
responding Riemann surface, the center lines of the cylinders of a Jenkins–Strebel
differential. For the proof of this result we show that the Teichmüller embedding ι
can be extended to a continuous embedding ῑ : H ∪ {cusps of ι∗ } → Tg , see Propo-
sition 4.13. Moreover, if the Veech group ι is a lattice in PSL2 (R), the image of ῑ is
the closure of ι in Tg , see Proposition 4.14.
Since Tg has these cusp singularities at the boundary that prevent it from being an
ordinary complex space, whereas the boundary of Mg is a nice divisor in a projective
variety, it is interesting to look at spaces that lie properly between Teichmüller and
moduli space and to ask for a boundary that fits somehow in between Tg and Mg .
An example of such a space is provided by the Schottky space which goes back to
the paper [32] of F. Schottky from 1887. He studied discontinuous groups that are
freely generated by Möbius transformations γ1 , . . . , γg (for some g ≥ 1) chosen in
such a way that there are disjoint closed Jordan domains D1 , D1 , . . . , Dg , Dg such
that γi maps Di onto the complement of the interior of Di . The Riemann surface
of such a Schottky group is compact of genus g. It can be shown that every compact
Riemann surface X admits such a Schottky uniformization X = / (with ⊂ P1 (C)
open and a Schottky group), see Section 5.1. The covering → X is called a
Schottky covering. It is minimal for the property that is planar, i.e. biholomorphic
to a subdomain of P1 (C); here minimality means that each unramified holomorphic
covering Y → X with a planar manifold Y factors through .
Schottky coverings are classified by a complex manifold Sg of dimension 3g − 3,
called the Schottky space. The natural map from Tg to Mg factors through Sg , therefore
there is a subgroup g (α) of the mapping class group g such that Tg / g (α) = Sg .
Unfortunately the subgroup g (α) is not normal and depends on the choice of a certain
group homomorphism α. As a consequence the induced map Sg → Mg is not the
quotient for a group action. We review this classical but not so widely known material
in Sections 5.1 and 5.2.
The concept of Schottky coverings can be extended to stable Riemann surfaces. If
the analogous construction as for ordinary Riemann surfaces is applied to a surface X
with nodes, we obtain a covering space which is not planar, but on which nevertheless
a free group acts by holomorphic automorphisms with quotient space / = X.
Although the groups are no longer subgroups of PSL2 (C), it is possible to find
parameters for them in almost the same way as for Schottky groups, namely by cross
ratios of fixed points. It then turns out that these generalized Schottky coverings are
classified by a complex manifold Sg (which contains Sg as an open dense subset), see
Section 5.3. This result was originally proved in [13]; here we show that it can easily
be derived from Braungardt’s characterization of Tg as the universal covering of Mg
with cusps over ∂Mg , see Section 5.4.
Finally we wonder what the image in Sg of a Teichmüller disk ι in Tg might look
like. In the general case we have no idea. Of course, the image may depend on the
choice of the subgroup g (α) that gives the map Tg → Sg . In the special situation
that Cι is a Teichmüller curve we prove that for suitable choice of α, the image of ι
in Sg is not a disk, see Proposition 5.21.
Acknowledgments. We would like to thank Volker Braungardt for allowing us to in-
clude his results on Tg , and for his helpful comments on an earlier version of Section 3.
We are also grateful to Pierre Lochak and Martin Möller for many valuable conver-
sations on Teichmüller disks, Teichmüller curves, and their boundaries. Furthermore,
we would like to thank Bill Abikoff for his useful suggestions that helped to improve
the exposition considerably.
2 Geodesic rays, Teichmüller disks and Teichmüller curves

The aim of this section is to introduce Teichmüller disks and Teichmüller curves.
We start by recalling in 2.1 the concept of Teichmüller deformations and using them
we give a definition for the Teichmüller space Tg alternative to the one we gave in
the introduction. This will help us to define geodesic rays in the Teichmüller space
in 2.2. In 2.3 we introduce Teichmüller disks as complex version of geodesic rays
giving different alternative definitions. Finally in 2.4 we introduce the Veech group
and Teichmüller curves and summarize some facts about the interrelation between
these objects.
2.1 Teichmüller deformations

As one of numerous possibilities, one can define the Teichmüller space as the space
of Teichmüller deformations. We briefly recall this concept here. It is described e.g.
in [2, Chapter I, §3].
At the end of this subsection we extend it to the corresponding concept for punctured
Riemann surfaces and their Teichmüller space Tg,n , cf. [2, Chapter II, §1].
Let X = Xref be a fixed Riemann surface of genus g ≥ 2 and q be a holomorphic
quadratic differential on X. We refer to the zeros of q as critical points, all other points
are regular points. Then on the surface
X∗ = X − {P ∈ X : P is a critical point of q}
the differential q naturally defines a flat structure μ, i.e. an atlas such that all transition
maps are of the form z → ±z + c, with some constant c ∈ C. The charts of μ in
regular points z0 are given as
z
z → q(ξ )dξ. (2.1)
z0
One may deform this flat structure by composing each chart with the map
1 1
x + iy → Kx + iy = (K + 1)z + (K − 1)z (x, y ∈ R) (2.2)
2 2
with K an arbitrary real number > 1. This defines a new flat structure on X∗ which
can uniquely be extended to a holomorphic structure on X.
We call XK the Riemann surface that we obtain this way, X1 = X the surface with
the original complex structure and fK : X1 → XK the map that is topologically the
identity. The map fK is a Teichmüller map and has constant complex dilatation
(fK )z K −1
k(z) = = .
(fK )z K +1
1+|k(z)|
Its maximal dilatation supz∈X 1−|k(z)| (as a quasiconformal map) is equal to K.
Definition 2.1. Let q be a holomorphic quadratic differential on X and K ∈ R>1 .

The pair (XK , fK ) as defined above is called the Teichmüller deformation of X of
constant dilatation K with respect to q.
The pair (XK , fK ) defines a point in the Teichmüller space Tg which for simplicity
we also denote as (XK , fK ). Since the constant dilatation of fK is equal to K, the
Teichmüller distance between the points (X1 , id) and (XK , fK ) of Tg is log(K).
If two holomorphic quadratic differentials on X are positive scalar multiples of
each other, they define, for each K, the same point in Tg . Thus one restricts to
differentials with norm 1. By Teichmüller’s existence and uniqueness theorems, see
e.g. [2, Chapter I, (3.5), (4.2)], one can show that each point in Tg is uniquely obtained
as a Teichmüller deformation. If QX is the vector space of all holomorphic quadratic
differentials on X and if X is the unit sphere in QX , one may thus write
{(X, q, k) : q ∈ X , k ∈ (0, 1)} ∪ {0} = Tg , (2.3)
and the identification of the two sets is done by the map
1+k K −1
(X, q, k) → (XK , fK ) with K = ⇔k= and
1−k K +1 (2.4)
0 → the base point (X, id).
(XK , fK ) depends of course by its definition on the differential q. In the following
we shall denote the base point also by (X, q, 0).
The map (2.4) is a homeomorphism. Here on the left-hand side of (2.3) one takes
the topology obtained by identifying it with the open unit ball in QX . It follows in
particular, that Tg is contractible.
Teichmüller deformations can be understood as affine deformations in the following
sense: Let us here and in what follows identify C with R2 by the R-linear map sending
(1, i) to the standard basis of R2 . Then the map in (2.2) is equal to the affine map

x K 0 x
→ · .
y 0 1 y
Since composing charts with a biholomorphic map does not change the point in Teich-
müller space, one obtains the same point (XK , fK ) in Tg if one composes each chart
of the flat structure μ on X with the affine map
√
x x K 0
→ DK · with DK = √1
∈ SL2 (R). (2.5)
y y 0
K
We shall use the following notations which are compatible with those in Section 2.3.2
where we introduce the general concept of affine deformations.
Definition 2.2. Let X be a compact Riemann surface of genus g, q a holomorphic

quadratic differential, μ the flat structure defined by q. We call the flat structure
defined by (2.5) μDK and denote (X, μ) DK = (X, μDK ).
Note that (X, μDK ) is as Riemann surface isomorphic to XK . Thus the point
[(X, μDK ), id] in Tg defined by the marking id : X → (X, μDK ) is equal to (XK , fK ).
Finally, let us turn to Teichmüller deformations of punctured Riemann surfaces:
The definition is done almost in the same way as in the case without punctures, see
[2, Chapter II, §1]. Suppose that g and n are natural numbers with 3g − 3 + n > 0.
Let X be a Riemann surface of genus g with n marked points P1 , . . . , Pn , and Xref =
X0 = X − {P1 , . . . , Pn }.
In this case, one uses admissible holomorphic quadratic differentials on X0 . They
are by definition those meromorphic quadratic differentials on X that restrict to a
holomorphic quadratic differential on X0 and have at each puncture either a simple
pole or extend holomorphically across the puncture, see [2, Chapter II, (1.4)]. The
vector space of these differentials is called QX0 . For q ∈ QX0 we define the critical
points to be the marked points and all zeroes of q; the remaining points are called
regular. Now, the definition of Teichmüller deformation is done exactly as before,
just always replacing QX by QX0 . One obtains in the same way:
{(X, q, k) : q ∈ X0 , k ∈ (0, 1)} ∪ {0} = Tg,n . (2.6)
Here X0 is the unit ball in QX0 .
2.2 Geodesic rays

Let X = Xref be a Riemann surface of genus g. A holomorphic quadratic differential
q on X naturally defines a geodesic embedding of R≥0 into Tg with respect to the
Teichmüller metric on Tg as is described in the following.
Definition 2.3. Let q be a holomorphic quadratic differential on X and let γ be the

map
⎧
⎨[0, ∞) → Tg ,
⎪
γ = γq : t → (XK , fK ) = (X, μDK ) = (X, q, k) (2.7)
⎪
⎩ t K−1
with K = e , k = K+1 .
The image of γ is called the geodesic ray in Tg in direction of q (or with respect
to) q starting at (X, id).
Here we use the notation of the last section:

Def.2.2 (2.4)
(XK , fK ) = (X, μDK ) = (X, q, k)
is the point in Tg defined by the Teichmüller deformation of X of dilatation K with
respect to q. Recall from the last section that the distance between the two points
(XK , fK ) and (X, id) in Tg is log(K). Thus γ is an isometric embedding.
In fact, from the description of Tg given in (2.3) one observes that all points in Tg
which have distance log(K) to the base point (X, id) are Teichmüller deformations
of X of constant dilatation K with respect to a holomorphic quadratic differential. It
follows that each isometric embedding of [0, ∞) into Tg is of the form (2.7).
2.3 Teichmüller disks

In this section we define Teichmüller disks. They can be found defined under this name
e.g. in [29, p. 149/150] and [12, 8.1–8.2]. One may find comprehensive overviews
e.g. in [34] and [11], or more recently [27] and [23], to pick only a few of numerous
references where they occur. We introduce them here in detail comparing three dif-
ferent ways how to construct them. For completeness we have included most of the
proofs.
Definition 2.4. Let 3g − 3 + n > 0. A Teichmüller disk ι is the image of a

holomorphic isometric embedding
ι : D → Tg,n
of the complex unit disk D = {z ∈ C : |z| < 1} into the Teichmüller space. Here we
take the Poincaré metric of constant curvature −1 on D and the Teichmüller metric on
Tg,n . The embedding ι is also called Teichmüller embedding.
Instead of the unit disk D one may take as well the upper half plane H with the
hyperbolic metric. We shall switch between these two models using the holomorphic
isometry
i−t
f : H → D, t → . (2.8)
i+t
Thus Teichmüller disks are obtained equivalently as images of holomorphic isometric
embeddings H → Tg,n of the upper half plane H into the Teichmüller space Tg,n .
How does one find such embeddings? Similarly as for geodesic rays, each holo-
morphic quadratic differential q on a Riemann surface X defines a Teichmüller disk.
In the following we describe three alternative constructions starting from such a differ-
ential q that all lead to the same Teichmüller disk q . For simplicity we only consider
the case n = 0 and g ≥ 2. However the same constructions can be done in the general
case of punctured surfaces.
2.3.1 Teichmüller disks as a collection of geodesic rays
Definition 2.5. Let q be a holomorphic quadratic differential on a Riemann surface

X of genus g. Let ι1 be the map

D → Tg ,
ι1 :
z = r · eiϕ → (X, e−iϕ · q, r).
Here we use the definition of Tg given by (2.3). Hence, (X, e−iϕ · q, r) is the point
defined by the Teichmüller deformation of X of dilatation K = 1+r 1−r with respect to
q−ϕ = e −iϕ · q.
We shall show in Proposition 2.8 that ι1 is an isometric holomorphic embedding,
thus the image ι1 of ι1 is a Teichmüller disk.
The map ι1 may be considered as a collection of geodesic rays in the following
sense: Let τϕ be the geodesic ray in D starting from 0 in direction ϕ, i.e.:

[0, ∞) → D,
τϕ : t
t → r(t) · eiϕ with r(t) = eet −1
+1 .
Then ι1 τϕ : [0, ∞) → Tg is equal to the map given in (2.7) that defines the geodesic
ray to the holomorphic quadratic differential q−ϕ = e−iϕ · q on X.
Thus the Teichmüller disk ι1 is the union of all geodesic rays defined by the
differentials eiϕ · q with ϕ ∈ [0, 2π ). Furthermore, ι1 τϕ is the parameterization by
length of the restriction ι1 |Rϕ of ι1 to the ray Rϕ = {r · eiϕ : r ∈ [0, 1)}.
2.3.2 Teichmüller disks by affine deformations. We now describe a second ap-

proach that starting from a holomorphic quadratic differential q leads to the same
Teichmüller disk as in 2.3.1.
Recall from Section 2.1 that a holomorphic quadratic differential q defines on
X∗ = X −{zeroes of q} a flat structure μ. The group SL2 (R) acts on the flat structures
of X ∗ (as topological surface) in the following way: Let B ∈ SL2 (R) and μ be a flat
structure on X ∗ . Composing each chart of μ with the affine map z → B · z gives a
new flat structure on X∗ which we denote B (X, μ) or (X, μB ). In the special case
B = DK we obtain the Teichmüller deformation of dilatation K, cf. Definition 2.2.
Definition 2.6. We call (X, μB ) = B (X, μ) affine deformation of (X, μ) by the

matrix B.
Note that for B1 , B2 in SL2 (R) one may write

B1 (B2 (X, μ)) = B1 (X, μB2 ) = (X, μB1 B2 ) = B1 · B2 (X, μ).
The flat structure μB defines in particular a complex structure on X. We identify
here the complex plane C with R2 as we already did in Section 2.1. In general the new
complex structure will be different from the one defined by μ. Taking the identity
id : (X, μ) → (X, μB ) on X as marking, we obtain a point PB = [(X, μB ), id] in the
Teichmüller space Tg . By abuse of notation we shall sometimes denote this point also
just as (X, μB ).
Thus one obtains the map
ι̂2 : SL2 (R) → Tg , B → PB = [(X, μB ), id] = (X, μB ).
If however the matrix A = U is in SO2 (R) the map id : (X, μB ) → (X, μU ·B ) is
holomorphic, thus the point in Teichmüller space is not changed, i.e.
U ∈ SO2 (R) ⇒ PU A = PA for all A ∈ SL2 (R). (2.9)
Hence ι̂2 induces a map
ι2 : SO2 (R)\ SL2 (R) → Tg , SO2 (R) · B → PB = [(X, μB ), id] = (X, μB ).
Please note: The action of SL2 (R) on the flat structures {(X, μA ) : A ∈ SL2 (R)}
does not descend to the image set {PA : A ∈ SL2 (R)} in Tg ; in particular PU =
PI ⇒ PAU = PA !
The Teichmüller disk. One may identify SO2 (R)\ SL2 (R) with the upper half plane
H in the following way: Let SL2 (R) act by Möbius transformations on the upper half
plane H. This action is transitive and SO2 (R) is the stabilizing group of i. Thus the
map
p : SL2 (R) → H, A → −A−1 (i) (2.10)
induces a bijection SO2 (R)\ SL2 (R) → H. Its inverse map is induced by

1 1 Re(t)
H → SL2 (R), t → √ .
Im(t) 0 Im(t)
Composing ι2 from above with this bijection one obtains a map from H to Tg which
we also denote by ι2 .
Definition 2.7. Let q be a holomorphic quadratic differential on the Riemann surface

X and μ the flat structure defined by q. Let ι2 be the map
ι2 : H → Tg , t → PAt = [(X, μAt ), id]
with At chosen such that −A−1

t (i) = t.
Note that the identification of SO2 (R)\ SL2 (R) with H given by p may seem a bit
ponderous, but one has to compose A → A−1 (i) with the reflection at the imaginary
axis in order that ι2 becomes holomorphic. We shall see this later in 2.3.3. In fact one
has much more, as is stated in the following proposition.
Proposition 2.8. The maps ι1 and ι2 are Teichmüller embeddings. They define the
same Teichmüller disk
q = ι1 = ι1 (D) = ι2 = ι2 (H). (2.11)
Proof. The proof is given in the rest of Subsection 2.3.2 and in 2.3.3:
In Proposition 2.12 we show that ι2 = ι1 f with f from (2.8) (see also Figure 1);
thus it is sufficient to show only for one of them that it is isometric, and in the same
manner for being holomorphic.
In Proposition 2.11 it is shown that ι2 is isometric. In Subsection 2.3.3, it is
shown that ι1 is holomorphic (see Corollary 2.15). For this purpose we introduce an
embedding ι3 : D → Tg , using Beltrami differentials, for which it is not difficult to
see that it is holomorphic, and show that it is equal to ι1 .
That ι1 and ι2 define the same Teichmüller disks then also follows from Proposi-
tion 2.12.
In fact the described constructions do not only give some special examples but all
Teichmüller disks are obtained like this: Each Teichmüller disk is equal to q as in
(2.11) for some holomorphic quadratic differential q. And all Teichmüller embeddings
are of the form ι1 : D → Tg or equivalently ι2 : H → Tg , see [12, 7.4].
In order to see that ι2 from Definition 2.7 is isometric we first calculate the Teich-
müller distance between two affine deformations.
Teichmüller distance between two affine deformations. In what follows we con-

stantly use the following fact about matrices in SL2 (R):
Remark 2.9. Each matrix A ∈ SL2 (R) with A ∈ SO2 (R) can be decomposed
uniquely up to the minus signs as follows:
√
K 0
A = U1 · DK · U2 with U1 , U2 ∈ SO2 (R), DK = √1
, K > 1.
0
K (2.12)

cos(θ ) − sin(θ )
We may denote: U2 = Uθ = .
sin(θ ) cos(θ )
This fact can e.g. be seen geometrically as follows: SL2 (R) acts transitively on
the upper half plane H by Möbius transformations. The point i ∈ H can be mapped
to A(i) = i by first doing a stretching along the imaginary axis in direction ∞
and afterwards a rotation around i, i.e. A(i) = U1 (DK (i)) with suitably chosen
U1 ∈ SO2 (R) and DK with K > 1 as in the remark. Since the stabilizer of i in
SL2 (R) is SO2 (R), one has A = U1 · DK · U2 with U2 also in SO2 (R). A short
calculation gives the uniqueness claim.
In the following proposition let q again be a holomorphic quadratic differential on
X = Xref and μ the flat structure that q defines.
Proposition 2.10. Let A and B be in SL2 (R) with A · B −1 ∈ SO2 (R) and
A · B −1 = U1 · DK · U2
with U1 , U2 and DK as in (2.12). Then the Teichmüller distance between the two
points PA = [(X, μA ), id] and PB = [(X, μB ), id] in Tg is log(K).
Proof. We proceed in three steps.

a) Suppose B is the identity matrix I and
√
K 0
A = DK = √1
for some K ∈ R>1 .
0
K
Thus we have in fact that PA = [(X, μDK ), id] is the point in Tg defined by the
Teichmüller deformation of dilatation K with respect to q, see Definition 2.2. Hence
the distance between PA and the base point (Xref , id) = PI is log(K).
b) Suppose again that B = I , but A is an arbitrary matrix in SL2 (R).
Thus A = U1 · DK · U2 and the map id : (X, μ) → (X, μA ) is the composition
of three maps:
id id id
(X, μ) −
→ (X, μU2 ) −
→ (X, μDK U2 ) −
→ (X, μU1 DK U2 ).
Since the first and the third map are biholomorphic the Teichmüller distance is again
log(K).
More precisely, write U2 = Uθ as in (2.12). Then μU2 is the flat structure obtained
by composing each chart with z → eiθ · z. This is equal to the flat structure defined
by the quadratic differential q2θ = (eiθ )2 · q which is holomorphic on the Riemann
surface X. √
Now, id : (X, μU2 ) → (X, μDK U2 ) is (up to the stretching z → K · z) the
Teichmüller deformation of dilatation K with respect to the holomorphic quadratic
(2.9)
differential q2θ . Thus the distance between PA = PU1 DK U2 = PDK U2 and the base
point PB = PI is log(K).
c) Let now A, B be arbitrary in SL2 (R). The Teichmüller metric does not depend
on the chosen base point. Thus we may consider PB as base point and PA as coming
from the affine deformation defined by the matrix A · B −1 . Then with the given
decomposition A · B −1 = U1 · DK · U2 the distance is as in b) equal to log(K).
Proposition 2.11. ι2 is an isometric embedding
Proof. We denote by ρ the Poincaré distance in H and by dT the Teichmüller distance

in Tg . Let t1 and t2 be arbitrary distinct points in H. We may write t1 = p(A)
and t2 = p(B) with A, B in SL2 (R), p as in (2.10). Let AB −1 = U1 DK U2 the
decomposition of AB −1 from (2.12). (AB −1 ∈ / SO2 (R) because t1 = t2 )
ρ(t1 , t2 ) = ρ(−B −1 (i), −A−1 (i)) = ρ(B −1 (i), A−1 (i)) = ρ(AB −1 (i), i)

= ρ(U1 DK U2 (i), i) = ρ(U1 DK (i), i) = ρ(DK (i), i)
Prop. 2.10
= ρ(Ki, i) = log(K) = dT (PB , PA ) = dT (ι2 (t1 ), ι2 (t2 )).
The equality is given since U1 is a hyperbolic rotation with center i and thus does
not change the distance to i.
Now we show that ι1 and ι2 are “almost” the same map.
Proposition 2.12. ι1 and ι2 fit together. More precisely: ι1 f = ι2 , with the

isomorphism f : H → D from (2.8).
The following diagram may be helpful while reading the proof. Some parts will be
explained only after the proof; in particular the space B(X) of Beltrami differentials
will be introduced in 2.3.3.
ι1
z=r·eαi = i−t K−1 −2iθ

i+t = K+1 ·e ∈ DO NN
NNN ⎧
NNNb ⎪ ι1 (z)
q
NNN z · |q| = i−t q ⎪
⎪
i+t |q| ⎪
⎪ = (X, e−αi q, r)
A = U1 DK Uθ N& ⎨
/ = (X, e2θi q, K−1 )
SL2 (R)O f B(X) Tg K+1
OOO
OOpO pppp8 D ⎪
⎪
⎪
= ι2 (t)
p ⎪
OOO ppp ⎩= PA = [(X, μA ), id]
⎪
O' ppp = [DK U2 (X, μ), id]
H t = −A−1 (i)
ι̂2
Figure 1. Diagram for alternative definitions of the Teichmüller disk q .
Proof. We proceed in two steps:

(1) Let A ∈ SL2 (R) be decomposed as in (2.12): A = U1 · DK · U2 , U2 = Uθ .
K−1
We show that (f p)(A) = r · e−2iθ with r = K+1 .
(2) We show that ι1 (r · e−2iθ ) = ι̂2 (A).
Step 1. One may express t := p(A) in terms of K and θ as follows:

i cos(θ ) · −i
K + sin(θ )
t = −A−1 (i) = −U2−1 DK
−1
(i) = −U2−1 − =−
K − sin(θ ) · −i
K + cos(θ )
i cos(θ) − K sin(θ)
= .
i sin(θ) + K cos(θ)
Now one has:
−t + i −i cos(θ ) + K sin(θ ) + i(i sin(θ ) + K cos(θ ))
f (p(A)) = f (t) = =
t +i i cos(θ) − K sin(θ ) + i(i sin(θ ) + K cos(θ ))
(K − 1)[sin(θ) + i cos(θ )]
=
(K + 1)[− sin(θ) + i cos(θ )]
K −1 −(sin(θ ) + i cos(θ ))2
= ·
K + 1 (sin(θ) − i cos(θ ))(sin(θ ) + i cos(θ ))
K −1 K − 1 −2iθ
= (cos(θ) − i sin(θ ))2 = ·e .
K +1 K +1
Step 2. ι1 (r · e−2iθ ) = (X, e2iθ · q, r) ∈ Tg is the point in the Teichmüller space
1−r = K with respect to
that is obtained as Teichmüller deformation of dilatation 1+r
the quadratic differential e · q. Recall from the proof of Proposition 2.10 that this
2iθ
is precisely the point in Tg defined by the affine deformation DK Uθ (X, μ) =

(X, μDK Uθ ) = (X, μDK U2 ). Thus
(2.9)
(X, e2iθ · q, r) = PDK Uθ = PDK U2 = PU1 DK U2 = PA = ι̂2 (A). (2.13)
Using (2.13) one may also describe the geodesic rays ι1 τϕ from 2.3.1 in the
Teichmüller disk q = ι1 = ι2 as follows.
Corollary 2.13. Define DK , Uθ as in (2.12). The map

[0, ∞) → Tg , t → PDK Uθ = [DK (X, μUθ ), id] with K = et
is equal to ι1 τ−2θ .
It is thus by 2.3.1 the geodesic ray in direction of the quadratic differential
q2θ = e2θi q.
τ−2θ ι1 (2.13)
Proof. One has t −−→ r(t)e−2θ i −−→ (X, e2θi · q, r(t)) = PDK Uθ .
Hence, geometrically one obtains the geodesic ray to qϕ by rotating the flat structure
by U ϕ and then stretching in vertical direction with dilatation K.
2
2.3.3 Beltrami differentials. In order to see that ι1 and ι2 are holomorphic we intro-
duce an alternative way to define ι1 using Beltrami differentials. We keep this aspect
short and refer to e.g. [29] for more details.
Let
M(X) = {(X1 , f ) : X1 Riemann surface ,
f : X → X1 is a quasiconformal homeomorphism}/ ≈
with (X1 , f1 ) ≈ (X2 , f2 ) ⇔ f2 f1−1 is biholomorphic.
One has a natural projection M(X) → Tg . Furthermore M(X) can be canonically
identified with the open unit ball B(X) in the Banach space L∞
(−1,1) (X) of (−1, 1)-
forms by the bijection:
M(X) → B(X), (X1 , f ) → μf ,
where μf is the Beltrami differential (or complex dilatation) of f , cf. [29, 2.1.4].
Thus one obtains a projection : B(X) → Tg . The map is holomorphic ([29,
3.1]). Furthermore, for each quadratic differential q and for all k ∈ (0, 1) the form
q
k |q| is in B(X) ([29, 2.6.3]) Thus one may define the map
b
D → B(X) → Tg ,
ι3 : q q
z → z · |q| → z · |q| .
It is composition of two holomorphic maps and thus itself holomorphic.
We shall show in the following remark that ι3 = b = ι1 , cf. Figure 1.
Remark 2.14. For all z0 ∈ D : ι3 (z0 ) = ι1 (z0 ).
Proof. Let z0 = r · eiα ∈ D and A ∈ SL2 (R) with f (p(A)) = z0 .

Decompose A = U1 DK U2 as in (2.12) with U2 = Uθ . Then by Step 1 of the proof
of Proposition 2.12, r = K−1
K+1 and α = −2θ . Furthermore, by Proposition 2.12
ι1 (z0 ) = ι̂2 (A) = [(X, μA ), id] = [(X, μDK U2 ), id].
Let us calculate the Beltrami differential of the Teichmüller deformation f = id : X →
q
(X, μDK U2 ). We shall see that it is equal to z0 · |q| . From this it follows that ι1 (z0 ) =
ι3 (z0 ).
One has f = g h with h = id : X → (X, μU2 ) and g = id : (X, μU2 ) →
(X, μDK U2 ). Locally in the charts of the flat structure defined by q, the maps g and h
are given by
g : z → K · Re(z) + i · Im(z) and h : z → eiθ · z.
Thus in terms of these charts one has:
fz = gz · hz + gz · hz = eiθ · gz , fz = gz · hz + gz · hz = e−iθ · gz
fz g K −1
⇒ = e−2iθ · z = e−2iθ · = eiα · r = z0 .
fz gz K +1
q
Hence the Beltrami differential of f is z0 · |q| .
One obtains immediately the following conclusion.
Corollary 2.15. ι1 = ι3 is holomorphic. By Proposition 2.12 ι2 is also holomorphic.
2.4 Teichmüller curves

In this section we introduce Teichmüller curves and recall some properties of them, in
particular their relation to Veech groups. This was explored by Veech in his article [34]
and has been studied by many authors since then. Overviews and further properties
can be found e.g. in [27], [11] or [20].
Let ι : D → Tg be a Teichmüller embedding and = ι = ι(D) its image. We
may consider the image of ι in the moduli space Mg under the natural projection
Tg → Mg , cf. Section 1. In general it will be something with a large closure. But
occasionally it is an algebraic curve. Such a curve is called Teichmüller curve.
Definition 2.16. If the image of the Teichmüller disk in the moduli space Mg is an
algebraic curve C, then C is called Teichmüller curve.
A surface (X, q), with a Riemann surface X and a holomorphic quadratic differ-
ential q such that the Teichmüller disk = q defined by q projects to a Teichmüller
curve is called Veech surface.
How can one decide whether a surface (X, q) induces a Teichmüller curve or
not? An answer to this question is given by the Veech group, a subgroup of SL2 (R)
associated to (X, q). This is explained in the following two subsections.
2.4.1 Veech groups. Let X be a Riemann surface and q a holomorphic quadratic

differential on X. Let μ be the flat structure on X defined by q. One obtains a
discrete subgroup of SL2 (R) as follows: Let Aff + (X, μ) be the group of orientation
preserving diffeomorphisms which are affine with respect to the flat structure μ, i.e.
diffeomorphisms which are in terms of a local chart z of μ given by

a b
z → A · z + t, for some A = ∈ SL2 (R), t ∈ C.
c d
As above we identify the complex plane C with R2 . Furthermore, we denote for
z = x + iy: A · z = ax + by + i(cx + dy).
Since μ is a flat structure, up to change of sign the matrix A does not depend on
the charts. Thus one has a group homomorphism:
D : Aff + (X, μ) → PSL2 (R), f → [A].
For simplicity we shall denote the image [A] of the matrix A in PSL2 (R) often also
just by A.
Definition 2.17. The image (X, μ) = D(Aff + (X, μ)) of D is called the projective
Veech group of (X, μ).
We shall denote the projective Veech group also by (X, q) and ι , where ι : D →
Tg or ι : H → Tg is the Teichmüller embedding defined by q as described in 2.3.
(X, μ) is a discrete subgroup of PSL2 (R), see [34, Proposition 2.7].
2.4.2 The action of the Veech group on the Teichmüller disk. Recall that the pro-
jection Tg → Mg from the Teichmüller space to the moduli space is given by the
quotient for the action of the mapping class group
g = Diffeo+ (X)/ Diffeo0 (X),
cf. (1.1) in the introduction. The action of Diffeo+ (X) on Tg is given by
ρ : Diffeo+ (X) → Aut(Tg ) ∼
= q , ϕ → ρϕ ,
with ρϕ : Tg → Tg , (X1 , h) → (X1 , h ϕ −1 ).
The affine group Aff + (X, μ) acts as subgroup of Diffeo+ (X) on Tg . The following
remark (cf. [11, Theorem 1]) determines this action restricted to the Teichmüller disk
= q = {PB = [(X, μB ), id] ∈ Tg : B ∈ SL2 (R)}.
Remark 2.18. Aff + (X, μ) stabilizes . Its action on is given as follows:

ϕ ∈ Aff + (X, μ), B ∈ SL2 (R) ⇒ ρϕ (PB ) = PBA−1
with A ∈ SL2 (R) a preimage of D(ϕ) = [A].
Proof. Let ϕ ∈ Aff + (X), B ∈ SL2 (R) and A ∈ SL2 (R) be a preimage of D(ϕ) =
[A] ∈ PSL2 (R). In the following commutative diagram
ϕ −1
(X, μ) UU / (X, μ) id / (X, μB )
UUUU O
UUUU
UUUU
UUUU
id UUU*
(X, μBA−1 )
the map (X, μBA−1 ) → (X, μB ) is, as a composition of affine maps, itself affine. Its
derivative is D(id ϕ −1 id−1 ) = BA−1 (BA−1 )−1 = I . Thus it is biholomorphic
and ρϕ ([(X, μB ), id]) = [(X, μBA−1 ), id].
It follows from Remark 2.18 that Aff + (X, μ) is mapped by ρ to Stab(), the global
stabilizer of in g . Furthermore ρ : Aff + (X, μ) → Stab() ⊆ g is in fact an
isomorphism: It is injective, see [11, Lemma 5.2] and surjective, see [11, Theorem 1].
Thus we have Aff + (X, μ) ∼ = Stab().
From Remark 2.18 it also becomes clear that the action of ϕ ∈ Aff + (X, μ) depends
only on D(ϕ). Thus one obtains in fact an action of the projective Veech group (X, μ)
on .
Corollary 2.19. (X, μ) ⊆ PSL2 (R) acts on = {PB ∈ Tg : B ∈ SL2 (R)} by

ρ[A] (PB ) = PBA−1 (2.14)
where A is a preimage in SL2 (R) of [A].
Finally one may use the Teichmüller embedding ι2 : H → Tg defined by q (cf.

2.7) in order to compare the action of (X, μ) on = ι = ι(H) with its action on
H via Möbius transformations. One obtains the diagram in the following remark (cf.
[27, Proposition 3.2.]).
Remark 2.20. Let A ∈ PSL2 (R). Denote by A : H → H its action as Möbius

transformation on H. The following diagram is commutative:
t →−t /H ι /
H
A RAR −1 ρA
t →−t ι
H /H /
Figure 2
−1 0
Here R = 0 1 , thus R acts on P1 (C) by z → −z.
Proof. Let t ∈ H. Choose some B ∈ SL2 (R) with −B −1 (i) = −t, thus ι(−t) =
PB = [(X, μB ), id] and using (2.14) we obtain the diagram:
t →−t / −t ι / PB = [(X, μB ), id]

_t _
A ρA

A(t)
t →−t / −A(t) PBA−1 = [(X, μBA−1 ), id]
The commutativity of the diagram in Figure 2 then follows from

RAR −1 (−t) = −A(t) = −A(t)
and
−(BA−1 )−1 (i) = −A(B −1 (i)) = −A(t) = −A(t), thus ι(−A(t)) = PBA−1 . 2
2.4.3 Veech groups and Teichmüller curves. In Remark 2.18 we saw that the affine
group Aff + (X, μ) maps isomorphically to the global stabilizer of the Teichmüller disk
in g . Denote by proj : Tg → Mg the canonical projection. It then follows from
Remark 2.20 that the map
proj ι : H → proj() ⊆ Mg
factors through H/R(X, μ)R −1 . We call
∗ (X, μ) = R(X, μ)R −1
the mirror projective Veech group, since H/ ∗ (X, μ) is a mirror image of H/(X, μ),
and refer to it also as ∗ (X, q) or ι∗ .
H/ ∗ (X, μ) is a surface of finite type and hence an algebraic curve if and only if
∗ (X, μ) is a lattice in PSL2 (R). Altogether one obtains the following situation (cf.
[27, Corollary 3.3]).
Corollary 2.21. (X, q) induces a Teichmüller curve C if and only if (X, μ) is a

lattice in PSL2 (R). In this case the following diagram holds:
t →−t / H ι / = ι ⊆ Tg
H
proj proj

H/(X, μ)
antihol. / H/ ∗ (X, μ) birat. / C ⊆ Mg
In particular if (X, μ) is a lattice, then

• H/ ∗ (X, μ) is the normalization of the Teichmüller curve C,
• H/(X, μ) is antiholomorphic to H/ ∗ (X, μ).
3 Braungardt’s construction of Tg,n

Before we continue our study of Teichmüller disks and pass to the boundary, we want
to explain the partial compactification Tg,n of the Teichmüller space Tg,n that we
shall use in the subsequent chapters. As mentioned in the introduction, Tg,n will be a
locally ringed space which, as a topological space, coincides withAbikoff’s augmented
Teichmüller space Tˆg,n (see the discussion following Proposition 3.9). The points
of this space can be considered as marked stable Riemann surfaces (X, f ), where
f : Xref → X is a deformation map. The forgetful map (X, f ) → X defines a natural
map from Tg,n to the moduli space Mg,n of stable n-pointed Riemann surfaces of
genus g. This map extends the projection Tg,n → Mg,n and is in fact also the quotient
map for the natural action of the mapping class group g,n . But the stabilizers of the
boundary points are infinite, and at the boundary the topology of Tg,n is quite far from
that of a manifold.
In his thesis [8], V. Braungardt gave a construction of Tg,n which uses only the
complex structure of Mg,n and the boundary divisor ∂Mg,n . Moreover his construction
endows Tg,n with the structure of a locally ringed space and he shows that it is a fine
moduli space for “marked” stable Riemann surfaces. In this chapter we give a brief
account of his approach.
3.1 Coverings with cusps
The basic idea of Braungardt’s construction is to study, for a complex manifold S,

quotient maps W → W/G = S that have “cusps” over a divisor D in S. This
concept, which will be explained in this section, generalizes the familiar ramified
coverings. The key result is that, in the appropriate category of such quotient maps,
there exists a universal object p : W̃ → S with cusps over D.
In general W̃ cannot be a complex manifold or even a complex space. Therefore we
have to work in the larger category of locally complex ringed spaces, i.e. topological
spaces W endowed with a sheaf OW of C-algebras (called the structure sheaf) such
that at each point x ∈ W the stalk OW,x is a local C-algebra. The basic properties of
such spaces can be found e.g. in [14, Chapter 1, § 1] (where they are called C-ringed
spaces).
In our situation Braungardt constructs a normal locally complex ringed space W̃
such that the subspace W̃0 = W̃ − p −1 (D) is a complex manifold and the restriction
p|W̃0 : W̃0 → S0 = S − D is the usual universal covering.
Example 3.1. The simplest example is well known and quite typical: Take S to be
the unit disk D = {z ∈ C : |z| < 1} and D = {0}. The universal covering of S − D
is, of course, exp : H → D − {0}, z → e2πiz . It turns out that the universal covering
in Braungardt’s sense is Ĥ = H ∪ i∞} with the horocycle topology, i.e. the sets
HR = {z ∈ C : Im z > R} ∪ i∞} for R > 0 form a basis of neighbourhoods of the
point i∞. Note that this topology is not the one induced from the Euclidean topology
if H ∪ {∞} is considered as a subset of the Riemann sphere Ĉ.
Ĥ is given the structure of a normal complex ringed space by taking O(U ) to be
the holomorphic functions on U for open subsets U of H, and by defining O(HR ) to
be the set of holomorphic functions on {z ∈ C : Im z > R} that have a continuous
extension to i∞. Clearly O(HR ) contains all functions of the form z → e2πiz/n for
all n ≥ 1.
We now give the precise definitions. We begin with the class of spaces that we
need (cf. [8, 3.1.3]):
Definition 3.2. Let (W, OW ) be a locally complex ringed space whose structure sheaf
OW is a subsheaf of the sheaf C ∞ (W, C) of continuous complex valued functions
on W .
a) A subset B ⊂ W is called analytic if there is an open covering (Ui )i∈I of W and
for each i ∈ I there are finitely many elements fi,1 , . . . , fi,ni ∈ OW (Ui ) such that
B ∩ Ui is the zero set of {fi,1 , . . . , fi,ni }.
b) We call (W, OW ) an R-space if, for every open U ⊆ W and every proper closed
analytic subset B ⊂ U , a continuous function f : U → C is in OW (U ) if and only if
its restriction to U − B is in OW (U − B).
Note that all complex spaces are R-spaces: The required property is just Riemann’s
extension theorem, see [14, Chapter 7].
Definition 3.3. Let S be a complex manifold and D ⊂ S a proper closed analytic

subset. Then a surjective morphism p : W → S from an R-space (W, OW ) to S is
called a covering with cusps over D if there is a group G of automorphisms of W (as
locally complex ringed space) such that
(i) p is the quotient map W → W/G = S,
(ii) W0 = p−1 (S − D) is a complex manifold and p|W0 : W0 → S0 = S − D is an
unramified covering,
(iii) for any x ∈ W there is a basis of neighbourhoods Ux that are precisely invariant
under the stabilizer Gx of x in G (i.e. Gx (Ux ) = Ux and g(Ux ) ∩ Ux = ∅ for
each g ∈ G − Gx ).
Note that, in particular, any ramified normal covering of complex manifolds is

a covering in the sense of this definition (with cusps over the branch locus). As
mentioned before, the basic result is (see [8, Satz 3.1.9])
Theorem 3.4. (i) For any complex manifold S and any proper closed analytic subset
D ⊂ S there exists an initial object p : (W̃, OW̃ ) → S in the category of coverings
of S with cusps over D; it is called the universal covering with cusps over D. The
restriction of p to W̃0 = p−1 (S0 ) is the universal covering of S0 , and the group
G = Aut(W̃/S) is the fundamental group π1 (S0 ).
(ii) If S is an open submanifold of S and W̃ the universal covering of S with
cusps over D = D ∩ S , then W̃ /H embeds as an open subspace into W̃ , where H
is the kernel of the homomorphism π1 (S − D ) → π1 (S − D) = G.
Proof. We only sketch the construction of the space (W̃, OW̃ ). The details that it
satisfies all the required properties are worked out in [7]. For the proof of (ii) we refer
to [8].
Let S0 = S − D, G = π1 (S0 ) and p0 : W0 → S0 the universal covering. W̃ is ob-
tained from W0 by “filling in the holes above D” in such a way that the G-action
extends from W0 to W̃ . More formally, the fibre W̃s of W̃ over any point s ∈ S is
constructed as follows: let U(s) be the set of open connected neighbourhoods of s in S;

for any U ∈ U(s) denote by X(U ) the set of connected components of p0−1 (U ). Then
W̃s = {(xU )U ∈U(s) : xU ∈ X(U ), xU ∩ xU = ∅ for all U, U ∈ U(s)}.
Clearly W̃s = p0−1 (s) for s ∈ S0 . Note that by definition, G acts transitively on each
W̃s , thus W̃/G = S. For any x = (xU ) ∈ W̃ define the sets xU ∪ {x}, U ∈ U(s), to
be open neighbourhoods of x. Finally define the structure sheaf by
OW̃ (U ) = {f : U → C continuous : f holomorphic on U ∩ W̃0 } (3.1)
for any open subset U of W̃ .
A key point in Braungardt’s proof of Theorem 3.4 is the existence of neighbour-
hoods U for any point a ∈ D such that the natural homomorphism
π1 (U − D) → lim π1 (U − D)
−

−→
U ∈U(a)
is an isomorphism. He calls such neighbourhoods decent. The importance of this

notion is that if U is a decent neighbourhood of a point a ∈ D and xU a connected
component of p −1 (U ), then xU is precisely invariant under the stabilizer Gx in G of
the unique point x ∈ xU ∩ p −1 (a).
Decent neighbourhoods in the above sense do not exist in general for singular
complex spaces. For example, if S is a stable Riemann surface and s ∈ S a node,
U − {s} is not even connected for small neighbourhoods U of s. Nevertheless the
construction can be generalized to this case, and the proof of the theorem carries over
to this case as Braungardt explains in [8, Anmerkung 3.1.4]; we therefore have:
Corollary 3.5. Any stable Riemann surface has a universal covering with cusps over
the nodes.
Near the inverse image of a node, the universal covering of a stable Riemann surface
looks like two copies of Ĥ glued together in the cusps. If such a neighbourhood is
embedded into the complex plane or P1 (C) it is called a doubly cusped region, cf. [25,
VI.A.8].
3.2 The cusped universal covering of Mg,n

Let us now fix nonnegative integers g, n such that 3g −3+n > 0. We want to construct
the space Tg,n as the universal covering of Mg,n with cusps over the compactification
(or boundary) divisor ∂Mg,n . But we cannot apply Theorem 3.4 directly to Mg,n since
it is not a manifold, but only an orbifold (or smooth stack). Braungardt circumvents
this difficulty by
Definition 3.6. A morphism p : Y → Mg,n of locally complex ringed spaces is called

a covering with cusps over D = ∂Mg,n if there is an open covering (Ui )i∈I of Mg,n
and for each i ∈ I a covering qi : Ui → Ui by a complex manifold Ui such that
pi qi
p|p−1 (Ui ) factors as p−1 (Ui ) −→ Ui −→ Ui , where pi is a covering with cusps over
qi−1 (D) (in the sense of Definition 3.3).
Then one can use Theorem 3.4 to prove
Proposition 3.7. There is a universal covering Tg,n → Mg,n with cusps over ∂Mg,n .
Proof. We first construct local universal coverings and then glue them together. For
any s ∈ Mg,n choose an open neighbourhood U and a covering q : U → U with a
manifold U . Let W̃ be the universal covering of U with cusps over D = q −1 (D).
Let H be the kernel of the homomorphism from π1 (U −D ) to g,n . Theorem 3.4 (ii)
suggests that the quotient W̃ /H should be an open part of the universal covering of
Mg,n . All that remains to show is that the W̃ /H glue together to a covering with
cusps over D. This is done in [8, 3.2.1]
Locally Tg,n looks like a product of a ball with some copies of the universal covering
Ĥ of D with cusps over {0} which was explained in Section 3.1:
Corollary 3.8. Let x ∈ Tg,n correspond to a stable Riemann surface X with k nodes.
Then x has a neighbourhood that is isomorphic to
Ĥk × D3g−3+n−k .
Proof. Let s ∈ Mg,n be the image point of x. The deformation theory of stable
Riemann surfaces gives us a map from D3g−3+n onto a neighbourhood of s such that
the inverse image of D = ∂Mg,n is the union of axes D = {(z1 , . . . , z3g−3+n ) :
z1 . . . zk = 0}, see [16, Section 3B]. The fundamental group of D3g−3+n − D is a free
abelian group on k generators; they correspond to Dehn twists about the loops that are
contracted in X. Thus the homomorphism π1 (D3g−3+n − D ) → g,n is injective. By
Proposition 3.7 and its proof the universal covering W̃ of D3g−3+n with cusps over D
is therefore a neighbourhood of x. It is not hard to see that W̃ is of the given form.
Our next goal is to compare Tg,n to the augmented Teichmüller space Tˆg,n intro-
duced by Abikoff [1].
Proposition 3.9 (cf. [8], Satz 3.4.2). Tg,n is homeomorphic to the augmented Teich-
müller space Tˆg,n .
Before proving the proposition we summarize the definition and some properties
of Tˆg,n : As a point set,
Tˆg,n = {(X, f ) : X a stable Riemann surface of type (g, n),
(3.2)
f : Xref → X a deformation}/ ∼ .
As mentioned in the introduction, a deformation is a map that contracts some disjoint

loops on Xref to points (the nodes of X) and is a homeomorphism otherwise. The
equivalence relation is the same as for Tg,n : (X, f ) ∼ (X , f ) if and only if there is
a biholomorphic map h : X → X such that f is homotopic to h f .
Abikoff puts a topology on Tˆg,n by defining neighbourhoods UV ,ε of a point (X, f )
for a compact neighbourhood V of the set of nodes in X and ε > 0:
UV ,ε = {(X , f ) : there is a deformation h : X → X, (1 + ε)-quasiconformal
(3.3)
on h−1 (X − V ), such that f is homotopic to h f }/ ∼ .
The action of the mapping class group g,n extends continuously to Tˆg,n ([1, Theo-
rem 4]), and the orbit space Tˆg,n / g,n is Mg,n (as a topological space).
Proof of Proposition 3.9. Braungardt shows (see [8, Hilfssatz 3.4.4]) that the stabi-
lizer of a point (X, f ) ∈ Tˆg,n in g,n is an extension of the free abelian group generated
by the Dehn twists about the contracted
loops by the holomorphic automorphism group
Aut(X) of X. For any V and ε, σ ∈Aut(X) σ (UV ,ε ) is invariant under the stabilizer
of (X, f ), and for sufficiently small V and ε, it is precisely invariant. Therefore the
quotient map Tˆg,n → Mg,n is a covering with cusps over ∂Mg,n in the sense of Def-
inition 3.6, except that so far no structure sheaf has been defined on Tˆg,n . But this
can be done in the same way as in (3.1). The universal property of Tg,n then yields
a map p : Tg,n → Tˆg,n compatible with the action of g,n on both sides. To show
that this map is an isomorphism we compare the stabilizers in g,n for the points in
both spaces. For a point in Tˆg,n we just described this stabilizer, and the proof of
Corollary 3.8 shows that for a corresponding point in Tg,n it is also an extension of Zk
by Aut(X).
3.3 Teichmüller structures

In this section we explain how Braungardt extends the universal family of marked
Riemann surfaces that is well known to exist over Tg,n to a family over Tg,n which
still is universal for the appropriate notion of marking or Teichmüller structure.
As above we fix a reference Riemann surface Xref of type (g, n); let Q1 , . . . , Qn
be the marked points and Xref 0 = X − {Q , . . . , Q }. Let us also fix a universal
ref 1 n
covering Uref → Xref and identify πg,n = π1 (Xref
0 0 ) with the group Aut(U /X 0 ) of
ref ref
deck transformations.
A classical construction of the family Cg,n over Tg,n goes as follows (cf. [5]):
For every point x = (X, P1 , . . . , Pn , f ) ∈ Tg,n take a universal covering of X0 =
X −{P1 , . . . , Pn } and arrange them so that they form an H-bundle + over Tg,n . Then
Cg,n is obtained as the quotient of + by the natural action of πg,n . More precisely,
+ is defined as follows: to x ∈ Tg,n there corresponds the quasi-Fuchsian group
Gx = w μ G(w μ )−1 , where G = Aut(Uref /Xref 0 ) ∼ π
= g,n and wμ is the quasicon-
formal automorphism of P1 (C) associated to x, see e.g. [21, 6.1.1]. The domain of
discontinuity of Gx consists of two connected components − (x) = w μ (L) (where
L is the lower half plane) and + (x) = wμ (H). Then + (x)/Gx = X0 , whereas
− (x)/Gx = Xref 0,∗
, the mirror image of Xref0 .
To extend this family we identify Tg,n with Tˆg,n by Proposition 3.9. As explained
in [1], any point x = (X, P1 , . . . , Pn , f ) ∈ Tˆg,n − Tg,n corresponds to a regular B-
group Gx . This means that Gx is a Kleinian group isomorphic to πg,n whose domain
of discontinuity (Gx ) has a unique simply connected invariant component − (Gx )
such that − (Gx )/Gx is isomorphic to Xref 0,∗
. For the union + (Gx ) = + (x) of
the other components of (Gx ) it holds that + (Gx )/Gx ∼ = X0 − {nodes}. To every
node in X there corresponds a conjugacy class of parabolic elements in Gx , each of
which is accidental (i.e. it becomes hyperbolic in the Fuchsian group hGx h−1 , where
h : − (Gx ) → H is a conformal map). Near a fixed point of such a parabolic element,
+ (Gx ) is a doubly cusped region, cf. the remark at the end of Section 3.1. If we
denote by ˆ + (x) the union of + (Gx ) with the fixed points of the parabolic elements
in Gx (accidental or not), then ˆ + (x) → X is the universal covering of X with cusps
over the nodes (cf. Corollary 3.5).
Definition 3.10. Let

ˆ+
ˆ+
g,n = {(x, z) ∈ Tg,n × P (C) : z ∈ (x)}.
1
On ˆ+ +
g,n , πg,n acts in such a way that for fixed x ∈ Tg,n the action on (x) is that
of Gx . Cg,n = ˆ+
g,n /πg,n is called the universal family over Tg,n .
Braungardt shows ([8, Hilfssatz 4.2.1]) that + ˆ+

g,n = {(x, z) ∈ g,n : x ∈ Tg,n , z ∈
+ (x)} is an open subset of Tg,n × P1 (C) and hence has a well defined structure of a
complex ringed space. One can extend this structure sheaf to all of ˆ+
g,n in the same
way as in (3.1). Then clearly Cg,n is also a complex ringed space, and the fibre over
x is isomorphic to the stable Riemann surface X represented by x.
To justify the name “universal” family for Cg,n we introduce the notion of a
Teichmüller structure: For a single smooth Riemann surface (X, P1 , . . . , Pn ) of type
(g, n), a Teichmüller structure is just a marking: so far we used markings as classes
of mappings Xref → X; equivalently a marking can be given as an isomorphism
πg,n → π1 (X − {P1 , . . . , Pn }) inducing an isomorphism πg = π1 (Xref ) → π1 (X)
and respecting the orientation and the conjugacy classes of the loops around the Qi
resp. Pi . Yet another equivalent way to give a marking is as a universal covering
U → X 0 together with an isomorphism πg,n → Aut(U/X 0 ). This last characteri-
zation also works for a stable Riemann surface if we take for U a universal covering
with cusps over the nodes. Before we extend this definition to the relative situation
we recall the notion of a family of stable Riemann surfaces.
Definition 3.11. Let S be a complex ringed space. A family of stable Riemann surfaces
of type (g, n) over S is a complex ringed space C together with a proper flat map
π : C → S such that the fibres Xs = π −1 (s), s ∈ S, are stable Riemann surfaces
of genus g. In addition we are given n disjoint sections Pi : S → C,i = 1, . . . , n,
of π such that Pi (s) is not a node on Xs . We denote by C 0 = C − ni=1 Pi (S) the
complement of the marked sections.
Definition 3.12. Let C/S be a family of stable Riemann surfaces of type (g, n) over
a complex ringed space S. A Teichmüller structure on C is a complex ringed space U
together with a morphism U → C such that for every s ∈ S the (restriction of the)
fibre Us0 → Xs0 is a universal covering with cusps over the nodes, together with an
isomorphism πg,n → Aut(U/C 0 ).
Putting everything together we obtain
Theorem 3.13. Tg,n is a fine moduli space for stable Riemann surfaces with Teich-
ˆ+
müller structure. Cg,n → Tg,n is the universal family and ˆ+
g,n → Cg,n = g,n /πg,n
is the universal Teichmüller structure.
Finally Braungardt gives a very elegant and conceptual description of Cg,n which
extends a classical result of Bers ([5, Theorem 9]) to the boundary:
Proposition 3.14. Tg,n+1 /πg,n is in a natural way isomorphic to Cg,n .
Proof. The kernel of the obvious homomorphism g,n+1 → g,n can be identified
with πg,n , which gives the action on Tg,n+1 . The holomorphic map Tg,n+1 → Tg,n
which forgets the last marked point extends to a map Tg,n+1 → Tg,n by a general
property of universal coverings with cusps. The difficult step in Braungardt’s proof
is to show that the induced map Tg,n+1 /πg,n → Tg,n has the right fibres. For this
purpose he constructs a map ˆ+g,n → Tg,n+1 and shows that it is bijective and induces
isomorphisms on the fibres over Tg,n .
4 Boundary points of Teichmüller curves

The aim of this chapter is to study the boundary points of the Teichmüller disks and
Teichmüller curves introduced in Section 2 in Tg and Mg , respectively. Here and
later, whenever we speak about Tg and its boundary, we mean the bordification of the
Teichmüller space described in Section 3.
In particular we shall derive, in Section 4.2, the following description of the bound-
ary points of Teichmüller curves (see Proposition 4.14 and Corollary 4.15 for a more
precise formulation):
Theorem 4.1. One obtains the boundary points of a Teichmüller curve by contracting
the centers of all cylinders in Jenkins–Strebel directions. They are determined by the
parabolic elements in the associated mirror Veech group.
This statement seems to be well known to the experts although we are not aware
of a published proof.
In Section 4.1 we prepare for the proof of Theorem 4.1 by introducing Jenkins–
Strebel rays. They are special geodesic rays in Teichmüller space which always
converge to a point on the boundary. Following Masur [26], we describe this boundary
point quite explicitly using the affine structure of the quadratic differential q that
defines the Jenkins–Strebel ray.
In Section 4.2 we turn to the boundary points of Jenkins–Strebel rays that are
contained in a Teichmüller disk. In particular if the Teichmüller disk descends to
a Teichmüller curve in the moduli space, all its boundary points can be determined
explicitly with the aid of the projective Veech group. One obtains Theorem 4.1 as a
conclusion.
4.1 Hitting the boundary via a Jenkins–Strebel ray

In this section, we introduce Jenkins–Strebel rays and describe their end point on the
boundary of Tg . As before, everything might be done as well for punctured surfaces
and the moduli space Tg,n with 3g − 3 + n > 0, but for ease of notation, we restrict
to the case n = 0.
Let X be a Riemann surface of genus g ≥ 2, q a holomorphic quadratic differential
on X. Recall from Section 2.1 that with q we have chosen a natural flat structure μ
on the surface X∗ = X − {critical points of q} whose charts were given in (2.1). The
maximal real curves in X∗ which are locally mapped by these charts to horizontal
(resp. vertical) line segments are called horizontal (resp. vertical ) trajectories. A
trajectory is critical if it ends in a critical point. Otherwise it is regular.
Definition 4.2. We say that a holomorphic quadratic differential q is Jenkins–Strebel,

if all regular horizontal trajectories are closed.
Jenkins–Strebel differentials play an exceptional role in the following sense. Recall

from Section 2.2 that each holomorphic quadratic differential defines a geodesic ray.
If q is Jenkins–Strebel, then the geodesic ray defined by its negative −q converges in
Tg to an end point on the boundary. This is described more precisely in the following
proposition which was proven by Masur in [26]. We give a version of his proof with
parts of the notation and arguments adapted to the context of our exposition.
Recall also from 2.2 that we obtain the geodesic ray to −q as the image of the
isometric embedding

[0, ∞) → Tg ,
γ = γ−q : (4.1)
t → (XK , fK ) = (X, μ−q ) K0 01 , id with K = et .
Note that here (XK , fK ) is the Teichmüller deformation of X of dilatation K with

respect to −q. Furthermore, μ−q is the translation structure on X ∗ defined by −q.
Proposition 4.3. Suppose q = 0 is a Jenkins–Strebel differential. For the geodesic

ray defined by γ−q in Tg , one has:
a) The ray converges towards a unique point (X∞ , f∞ ) on the boundary of the
Teichmüller space Tg .
b) One obtains this point by contracting the central lines of the horizontal cylinders
defined by q as is described in 4.1.4.
Definition 4.4. In the previous proposition, the geodesic ray defined by −q, i.e. the
image of γ−q in Tg , is called a Jenkins–Strebel ray.
For the proof of Proposition 4.3 one may use two slightly different perspectives of
the Jenkins–Strebel ray. They are described in 4.1.1, 4.1.2 and 4.1.3, 4.1.4. In 4.1.5
we describe the boundary point (X∞ , f∞ ). In 4.1.6 we show that the Jenkins–Strebel
ray in fact converges towards this point.
Throughout Section 4.1, we assume that the differential q is Jenkins–Strebel.
4.1.1 X as patchwork of rectangles. One may regard X as a patchwork of rectangles

in the complex plane, as is described in the following.
Since q is Jenkins–Strebel, the surface X, with the critical points and critical
horizontal trajectories removed, is swept out by closed horizontal trajectories. More
precisely, it follows from the work of Strebel (cf. [33], also see [26, Theorem B] which
contains a list of the results we use here) that the surface X, except for the critical
points and critical horizontal trajectories, is covered by a finite number of maximal
horizontal cylinders Z1 , . . . , Zp , i.e. annuli that are swept out by closed horizontal
trajectories. For each Zi one may choose a vertical trajectory βi joining opposite
boundary components of Zi . If we remove βi from Zi , the remainder is mapped, by
the natural chart wi defined by μ (see (2.1)), to an open rectangle Ri in the complex
plane. The horizontal and vertical edges have lengths

1 1
ai = |q(z)| dz and bi =
2 |q(z)| 2 dz,
αi βi
where αi is any closed horizontal trajectory in the cylinder Zi .

One may extend wi−1 uniquely to a map from the closure Ri of Ri to the closure of
the annulus Zi . Then the two horizontal edges of Ri are mapped to the two horizontal
boundary components of Zi and the two vertical edges are both mapped to βi . The
critical points of q that lie on the boundary of Zi define by their preimage marked
points on the horizontal edges of Ri and decompose them into segments.
For each such segment s on a horizontal edge of Ri its image on X joins the annulus
Zi to an annulus Zj possibly with i = j .
Thus the map wj wi−1 (wi−1 is the extended map, wj is locally the inverse map
of the extended map wj −1 ) is an identification map between s and a segment on a
horizontal edge of Rj . (Images of critical points have to be excluded.)
These identification maps are of the form z → ±z + c with a constant c ∈ C.
Conversely, given the closed rectangles R1 , . . . , Rp , the marked points on their
horizontal edges and these identification maps, we may recover the surface X as
follows: for each i glue the two vertical edges of Ri by a translation and the horizontal
edges (with the marked points removed) by the identification maps. In this way, one
obtains a surface X∗ with the flat structure on it inherited from the euclidean plane
C. By filling in the punctures at vertices, we obtain the original compact Riemann
surface X.
In this sense one may consider X as a patchwork of the rectangles R1 , …, Rp . This
description depends of course on the chosen holomorphic quadratic Jenkins–Strebel
differential q.
Example 4.5. Two Riemann surfaces X given as a patchwork of rectangles. In the

two examples in Figure 3 and Figure 4, the two vertical edges of each rectangle are
glued by a translation, respectively. Horizontal segments with the same name are
glued. The direction of the arrow indicates whether the identification is a translation
or a rotation by 180 . In the example in Figure 3 one only has translations, in the
example in Figure 4 only rotations.
In the first example the surface X is of genus 2 and all marked points are identified
and thus give only one point on X. In the second example one obtains a surface of
genus 0 with four marked points indicated by the four symbols •, , and .
Surface of genus 2 with 1 marked point

s1
b2
• > •
R2
• s3
>
a2 •
Surface of genus 0 with 4 marked points
s2 s3 s2
> • > • >
s2 s2
b1
R1 b1
> <
R1
• s1
> • s2
> • s1
> <
s1 •
a1 a1
Figure 3 Figure 4
4.1.2 Stretching the cylinders. We shall now redescribe the Jenkins–Strebel ray
defined by −q by stretching the rectangles in the ’patchwork’ from 4.1.1 in the vertical
direction.
The flat structure defined by −q = eπi · q is obtained from the flat structure μ
defined by q by composing each chart with a rotation by π2 . Thus the deformation
(XK , fK ) of dilatation K with respect to −q is equal to the affine deformation

K 0 0 −1 0 −K
(X, μ) = (X, μ).
0 1 1 0 1 0
This defines by (2.9) the same point in Tg as the affine deformation

0 1 0 −K 1 0
(X, μ) = (X, μ).
−1 0 1 0 0 K
Thus the isometric embedding γ = γ−q in (4.1) is equivalently given by

[0, ∞) → Tg ,
γ−q : (4.2)
t → (XK , fK ) = 01 K0 (X, μ), id , K = et .
−q −q
Recall again that here (XK , fK ) = (XK , fK ) is the Teichmüller deformation
with respect to the differential −q.
Hence we obtain the point γ−q (t) as follows: Each chart of μ is composed with the
map x + iy → x + iKy (x, y ∈ R) with K = et , and the marking is topologically the
identity. Now, let X be given as a patchwork of the rectangles R1 , . . . , Rp as in 4.1.1.
−q
Then we obtain the surface XK = XK in the following way: We stretch each rectangle
Ri , which has horizontal and vertical edges of lengths ai and bi , into a rectangle Ri (K)
with horizontal and vertical edges of lengths ai and K · bi . The identification maps
of the horizontal segments are again translations or rotations identifying the same
−q
segments as before. The surface XK = XK then is the patchwork obtained from
R1 (K), . . . , Rp (K) as described in 4.1.1.
−q
On Ri , the diffeomorphism fK = fK has image Ri (K) and is given by
x + iy → x + iKy.
This glues to a well-defined diffeomorphism on X ∗ , which can be uniquely extended

to X.
−q
Example 4.6. K-stretched surfaces. One obtains the surface XK = XK from the
surface X in Example 4.5 as the patchwork of the stretched rectangles in Figure 5 and
Figure 6, respectively.
s1
• > •
Kb2
R2
• s3 >
a2
•
s2 s3 s2
•
> > • >
s2 s2
Kb1 > <
R1 Kb1
R1
• s1
> • s2
> • s1
> <
s1 •
a1 a1
Figure 5 Figure 6
4.1.3 S as patchwork of double annuli. Recall that, in 4.1.1, we used μ to identify

the horizontal cylinder Zi on X with the euclidean cylinder defined by the rectangle
Ri in C; we did so by adding the vertical boundary edges and identifying them by a
translation. It turns out to be easier to describe the end point of the Jenkins–Strebel
ray, if we identify the Zi with so called double annuli Ai .
Definition 4.7. A cylinder Z of length a and height b defines a double annulus A as

follows:
• Take two disjoint open annuli A1 and A2 given as
b
A1 = A2 = {z ∈ C : r ≤ |z| < 1} with r = e−π a .
• Glue their inner boundary lines {|z| = r} by the map z → 1

z · r 2.
• We call the resulting surface A the double annulus of Z.
Remark 4.8. A is biholomorphic to Z.
The identification is given explicitly as follows:

• Z is biholomorphic to the Euclidean cylinder defined by the rectangle
{z ∈ C : 0 ≤ Re(z) ≤ a, 0 < Im(z) < b}.

• Decompose the rectangle into two halves of height b2 , a lower half R 1 = z ∈

C : 0 ≤ Re(z) ≤ a, 0 < Im(z) ≤ b2 and an upper half R 2 = z ∈ C : 0 ≤

Re(z) ≤ a, b2 ≤ Im(z) < b .
z
• The cylinder defined by R 1 is mapped to A1 by z → e2πi a .
a+bi−z
The cylinder defined by R 2 is mapped to A2 by z → e2πi a .
These maps respect the identifications and define a biholomorphic map from Z
to A, as shown in Figure 7.
a+bi−z
z → e2πi a '$
A2
z
bi z
R2 &% 6
b '$
2i ? b
R1 z → e2πi a
z
?
e−2π a · 1
z
-
r 1
0 a A
&% 1
Figure 7
Consider the double annuli A1 , . . . , Ap defined by the cylinders Z1 , . . . , Zp .

The biholomorphic map Zi → Ai extends to a continuous map from the closure
of Zi to the closure Ai of Ai . The zeroes of q on the boundary of Zi define marked
points on the boundary of Ai and decompose it into segments. The surface X can
now be described as a patchwork of the closed double cylinders A1 , . . . , Ap . The
identification maps between the segments on the boundary of the Ai are essentially
the same as in 4.1.1.
4.1.4 Contracting the central lines. Suppose that X is given as a patchwork of

double annuli A1 , . . . , Ap as in 4.1.3. We may describe the points (XK , fK ) =
−q −q
(XK , fK ) on the Jenkins–Strebel ray also as a patchwork of double annuli:
Let Ai (K) = A1i (K) ∪ A2i (K) (i ∈ {1, . . . , p}) be the double annulus from Defi-
−q
nition 4.7 with r = ri (K) = riK and define XK = XK to be the surface obtained by
gluing the closures A1 (K), . . . , Ap (K) with the same maps as A1 , . . . , Ap . Further-
−q
more, define the diffeomorphism fK = fK on Ai by
−q
fK : A1i → A1i (K) and A2i → A2i (K),
where
z = r · eiϕ → r K · eiϕ on both parts.
Then the following diagram is commutative:

ai +bi i−z
2πi
z → e ai
'$ '$ bi
9 r2
z → zi Ri2
z 2πi z
y z → e ai Ri1
A2i A1i
&% &% ai
ai +Kbi i−z
fK z → e
2πi ai fK
?
Kbi
? ?
'$ '$ Ri2 (K)
9 r 2K
z → i z
j z j 2πi z
y z → e ai
A2i (K) A1i (K)
&% &%
Ri1 (K)
Figure 8 ai
where fK = (reiϕ → r K eiϕ ) on the left side and fK = (x+iy → x+Kiy) on the
right side of the diagram. Thus, in particular, we have defined here with (XK , fK ) =
−q −q
(XK , fK ) the same surface (up to isomorphism) and the same diffeomorphism as
in 4.1.2.
4.1.5 The end point of the Jenkins–Strebel ray. We use the description of the
Jenkins–Strebel ray in 4.1.4 to obtain its end point (X∞ , f∞ ) ∈ Tg . Recall from
3.2 that a point in Tg consists of a stable Riemann surface X∞ and a deformation
f∞ : X → X∞ .
If K → ∞ in 4.1.4, the interior radius ri (K) = riK of the two annuli A1i (K) and
Ai (K) that form the double annulus Ai (K) tends to 0 (i ∈ {1, . . . , p}). Ai (K) tends
2
to a double cone Ai (∞) and the whole surface XK to a stable Riemann surface X∞ .
More precisely, we define Ai (∞) and X∞ as complex spaces in the following way.
Definition 4.9. Let A1i (∞) and A2i (∞) both be the punctured disk
{z ∈ C : 0 < |z| < 1},
and let pt be an arbitrary point. The disjoint union
Ai (∞) = A1i (∞) ∪ A2i (∞) ∪ {pt}
becomes a complex cone by the following chart:
ϕ : Ai (∞) → {(z1 , z2 ) ∈ C2 : z1 · z2 = 0, |z1 |, |z2 | < 1},
ϕ|A1 (∞) : z → (0, z), ϕ|A2 (∞) : z → (z, 0), ϕ(pt) = (0, 0).
i i
The closures of the double cones A1 (∞), . . . , Ap (∞) are glued to each other by
the same identification maps as in the ’finite’ case in 4.1.4. We call the resulting
stable Riemann surface X∞ . Topologically, X∞ is obtained from the surface X by a
contraction f∞ of the middle curves of the cylinders.
'$r 2 '$
z → i
z
n
9 z n
A2i Ai 1
&% &%
f∞ :
r · eiϕ → hi,∞ (r) · eiϕ
?
'$ '$ ?
ptq9 zptq
A2i (∞) A1
i (∞)
&% &%
Figure 9
We now define the contraction f∞ as the following map: Let A1i and A2i be the two
annuli in Definition 4.7 that form the double annulus Ai (i ∈ {1, . . . , p}). Then f∞
is given by
j j
f∞ : Ai → Ai (∞) for j ∈ {1, 2},
z = r · eiϕ → f∞ (z) = hi,∞ (r) · eiϕ ,
with an arbitrary monotonously increasing diffeomorphism hi,∞ : [ri , 1) → [0, 1).
The isotopy class of f∞ is independent of the choices of hi,∞ .
4.1.6 Convergence. We now show that, in the above notation, the Jenkins–Strebel
ray γ−q converges to the point (X∞ , f∞ ) on the boundary of Tg . Recall from (3.3)
that a base of open neighbourhoods of (X∞ , f∞ ) is given by the open sets
UV ,ε (X∞ , f∞ ) = {(X , f ) : there exists a deformation ϕ : X → X∞
such that ϕ f is isotopic to f∞
and ϕ|X \ϕ −1 (V ) has dilatation < 1 + ε},
for all compact neighbourhoods V of the singular points of X∞ and for all ε > 0. We
may restrict to open neighbourhoods V of the form
V = V (κ) = V1 ∪ · · · ∪ Vp , κ = (κ1 , . . . , κp ), 0 < κi < 1
where Vi is a double cone defined by
j j
Vi = Vi1 ∪ Vi2 ∪ {pt} with Vi = {0 < |z| ≤ κi } ⊆ Ai (∞), j ∈ {1, 2}.
Lemma 4.10. For each such V = V (κ) and each ε > 0, there is some K0 ∈ R>0
−q −q
such that all points (XK , fK ) = (XK , fK ) with K > K0 are in UV ,ε (X∞ , f∞ ).
Proof. Choose K0 such that riK0 < κi for all i ∈ {1, . . . , p} and suppose that K > K0 .
j
Define the diffeomorphism ϕ : XK → X∞ on Ai (K) by
⎧ j
⎪
⎨ z ∈ Ai (∞), if 1 > |z| ≥ κi
iθ i j
ϕ : z = r · e → hK (r) · e ∈ Ai (∞), if κi ≥ |z| > riK
iθ
⎪
⎩ j
pt ∈ Ai (∞), if |z| = riK
with an arbitrary monotonously increasing diffeomorphism hiK : (riK , κi ) → (0, κi ).
Then ϕ fK is isotopic to f∞ and ϕ|XK \ϕ −1 (V ) is holomorphic, hence its dilatation
is 1. Thus (XK , fK ) is in UV ,ε (X∞ , f∞ ).
j j
Ai (K) id Ai (∞)
'$ '$
j
reiθ → hiK (r)eiθ
j - p
&% &%
1 κi riK 1 κi
Figure 10
With Lemma 4.10 we have obtained the desired result and completed the proof of
Proposition 4.3.
Corollary 4.11. The Jenkins–Strebel ray defined by −q converges to the point

(X∞ , f∞ ) on the boundary of Tg .
4.2 Boundary points of Teichmüller disks

In this section we study the boundary points of a Teichmüller disk = ι in the
bordification Tg of the Teichmüller space; in particular, we consider the case that ι
projects to an affine curve in the moduli space Mg . For convenience, we use the upper
half plane model and consider Teichmüller embeddings as maps from H to Tg . We
shall obtain Theorem 4.1 as our final result. We proceed in two steps:
• In 4.2.1, we show that a Teichmüller embedding ι : H → Tg has a natural
extension
ῑ : H ∪ {cusps of ι∗ } → Tg ,
• In 4.2.2, we show that the image of ῑ is the whole closure of ι in Tg , if the

Teichmüller disk ι projects onto a Teichmüller curve in Mg .
It will follow from this that one obtains the boundary points of ι precisely
by contracting the central lines of the cylinders in “parabolic directions”. The
parabolic directions correspond to the cusps of the projective mirror Veech
group ι∗ .
Throughout this section, we assume that ι : H → Tg is a Teichmüller embedding to
a fixed holomorphic quadratic differential q on X = Xref and that μ is the translation
structure defined by q as in Section 2.1. Recall from Section 2.4 that the associated
projective Veech group ι = (X, μ) and its mirror image ι∗ = Rι R −1 (with R as
in Remark 2.20) are both Fuchsian groups in PSL2 (R).
4.2.1 Extending Teichmüller embeddings to the cusps of ∗ . Let s̃ ∈ R∞ =

R ∪ {∞} be a cusp of the Fuchsian group ι∗ , i.e. s̃ is a fixed point of some parabolic
element Ã of ι∗ . We associate to s̃ a point ῑ(s̃) = (X∞ (s̃), f∞ (s̃)) on the boundary
of Tg in the following way:
• In a natural way we associate to s̃ a Jenkins–Strebel ray.
• We show that this Jenkins–Strebel ray is the image in Tg of the hyperbolic ray in
H from i to s̃ under ι.
• ῑ(s̃) = (S∞ (s), f∞ (s)) is defined to be the end point of the Jenkins–Strebel ray.
The Jenkins–Strebel ray associated to s̃. A = R −1 ÃR is a parabolic element in the
projective Veech group ι . Let v be its unit eigenvector.
By Proposition 2.4 in [34], the direction v is fixed by some affine diffeomorphism
h of (X, μ). The derivative of h is A and v is a Jenkins–Strebel direction. More
precisely: The trajectories in the direction of v are preserved by h and each leaf is
either closed or a saddle connection, i.e. connects two critical points.
As in 4.1.1, X decomposes into maximal cylinders of closed leaves parallel to v
and the cylinders are bounded by saddle connections. The affine diffeomorphism h
can be described nicely as follows: Passing to a power of h if necessary, one may
assume that h fixes all critical points of q. Then h is the composition of Dehn twists
along the core curves of the cylinders. Each trajectory is mapped by h to itself and
the saddle connections are fixed pointwise.
Now, let us take the matrix

1
U = Uθ ∈ SO2 (R) such that U · v = e1 =
0
with Uθ defined as in (2.12).
Consider the affine deformation id : (X, μ) → (X, μU ) = (X, μ) U as in Def-
inition 2.6. The vector v is mapped to e1 . Thus the same trajectories are now the
horizontal ones.
Recall from 2.3.2 that the flat structure (X, μU ) is defined by the quadratic differ-
ential e2θi · q. Thus e2θ i · q is Jenkins–Strebel. The ray is by (4.2) given as:
γs̃ = γ−e2θi ·q : [0, ∞) → Tg ,

t → 01 K0 (X, μUθ ), id = [(X, μAK ), id]

with K = et and AK = 01 K0 · Uθ .
The Jenkins–Strebel ray γs̃ is the image of the geodesic ray in H from i to the cusp s̃.
From Remark 2.12 (see also Figure 1) one obtains that
γs̃ (t) = [(X, μAK ), id] = ι̂(AK ) = ι(−A−1

K (i)).
Furthermore, we have
−
−A−1 −1 −1 1 −1
K (i) = −Uθ (K · i) = −Uθ (−Ki) = RUθ R (Ki).
Thus the image of γs̃ is equal to the image of the ray RUθ−1 R −1 (Ki) (K ∈ [1, ∞))
under ι. But the latter one is the geodesic ray in H from i to RU −1 R −1 (∞) =
−U −1 (∞).
Observe finally that −U −1 (∞) = s̃: Since U · v = e1 for the eigenvector v of A,
one has for the fixed point s of A that U (s) = ∞. Hence, one has for the fixed point
s̃ of Ã = RAR −1 that s̃ = −s = −U −1 (∞). Thus the Jenkins–Strebel ray defined
by γs̃ is the image of the geodesic ray from i to s̃ in H under ι.
Finally we define ῑ(s̃) = (X∞ (s̃), f∞ (s̃)) ∈ Tg to be the end point of the Jenkins–
Strebel ray γs̃ . We then define the map ῑ as follows.
Definition 4.12. ῑ is the extension of ι defined by

ῑ : H ∪ {cusps of ι∗ } → Tg ,

ι(t), if t ∈ H
t →
ῑ(t) = (X∞ (s̃), f∞ (s̃)), if t = s̃ is a cusp of ∗ .
We consider H ∪ {cusps of ι∗ } as topological space endowed with the horocycle

topology as in Example 3.1.
Proposition 4.13. ῑ is a continuous embedding.
Proof. ῑ is continuous. Let s be a cusp of ι∗ , i.e. s is a fixed point of some parabolic
element Ã ∈ ι∗ , and c : [0, ∞) → H an arbitrary path in H converging to s in the
horocycle topology.
By Remark 2.20, the action of Ã on H fits together with the action of ρ(A) ∈ g
on Tg . Both actions may be extended continuously to Hs = H ∪ {s} (endowed
with the horocycle topology) and to Tg , respectively, and one obtains the following
commutative diagram:
ῑ /T
Hs = H ∪ {s} g
pÃ p
iÃ
/M
Hs /Ã g
Here the map iÃ : H/Ã → Mg is the map induced by ῑ and H/Ã is a disk with
center pÃ (s).
Let W be a neighbourhood of
P∞ = iÃ (pÃ (s)) = p(ῑ(s)).

i be the preimages of P in T under p. One of them is
For i in an index set I , let P∞ ∞ g
ῑ(s), again by the commutativity of the diagram.
Since {P∞i : i ∈ I } is discrete we may choose the neighbourhood W in such a
manner that its preimage under p is of the form:

V = p −1 (W ) = Vi ⊆ Tg
i∈I
where the Vi are the connected components of V with P∞ i ∈ V and V is invariant

i i
i
under the stabilizer of P∞ in the mapping class group g .
Furthermore, we may choose W such that the preimage of W under iÃ is a simply
connected neighbourhood of pÃ (s). Then, again, the preimage
U = p −1 (i −1 (W ))
Ã Ã
is a neighbourhood of s in the horocycle topology.

Thus an end piece of the path c is completely contained in U , i.e. there is some
l ∈ R>0 such that c([l, ∞)) is contained in U .
Since the above diagram is commutative and the Vi are disjoint, the image of U is
one of the Vi . This Vi then contains ῑ(c[l, ∞)). In addition, Vi has to contain the end
piece of the Jenkins–Strebel ray that leads to s used to define ῑ(s). Hence, Vi is the
component that contains ῑ(s).
Making W arbitrarily small, the neighbourhood U of s becomes arbitrarily small.
Thus ι c converges to ῑ(s).
ῑ is injective. Suppose there are two cusps s1 and s2 with P∞ = ῑ(s1 ) = ῑ(s2 ).
Thus we have two Jenkins–Strebel rays defined by the negative of the Jenkins–Strebel
differentials q1 = eiθ1 · q and q2 = eiθ2 · q with initial point P0 = ι(i) and the same
end point P∞ in Tg . Let (X∞ , f∞ ) and (Y∞ , g∞ ) be the two marked stable Riemann
surfaces defined by the two Jenkins–Strebel rays, respectively. Since they define the
same point in Tg the following diagram is commutative up to homotopy with some
biholomorphic h:
< X∞
O
f∞ yyy
y
yy
yy
Xref E h
EE g
EE ∞
EE
E"
Y∞
The core curves of the cylinders relative to the flat structure on X defined by q1 are
mapped by f∞ to the singular points of X∞ . Similarly the core curves coming from
q2 are mapped to the singular points of Y∞ . Since the diagram is commutative up
to isotopy, the two systems of core curves are homotopic. Thus the two Jenkins–
Strebel rays are similar by definition, using the terminology in [26, Section 5]. From
Theorem 2 in [26] it follows that there is some constant M < ∞ such that for two
points Q = R lying on the two Jenkins–Strebel rays which are equidistant from the
initial point P0 , one has d(Q, R) ≤ M. But then, since ι is an isometric embedding,
M would have to be an upper bound for the distance of equidistant points on two
different geodesic rays in H starting from i. This cannot be true.
4.2.2 Boundary of Teichmüller disks that lead to Teichmüller curves. Let now
ι : H → Tg be a Teichmüller embedding such that its image ι projects to a Teich-
müller curve C in the moduli space Mg .
Proposition 4.14. In this situation, the extended embedding from 4.2.1
ῑ : H ∪ {cusps of ι∗ } → ι ⊆ Tg
is surjective onto the closure ι of ι in Tg .
Proof. Recall from Corollary 2.21 that if ι leads to a Teichmüller curve then the
projective Veech group = ι is a lattice in PSL2 (R), H/ ∗ is a complex algebraic
curve and H/ ∗ → C ⊂ Mg is the normalization of C. Thus it extends to a surjective
morphism
ϕ : H/ ∗ → C ⊆ Mg ,
where H/ ∗ and C are the projective closure of H/ ∗ and the closure of C in Mg ,
respectively.
Furthermore, the map H → H/ ∗ extends continuously to a surjective map
p : H ∪ {cusps of ∗ } → H/ ∗ , since ∗ is a lattice in PSL2 (R). Here we use
the horocycle topology on H ∪ {cusps of ∗ }.
Thus one has the following commutative diagram of continuous maps:

ῑ / ⊆T
H ∪ {cusps of ∗ } ι g
p p| p
ι

ϕ
/ C⊆M
H/ ∗ g
Let now P∞ be a point on the boundary of ι . Similarly as in the proof of the

continuity of ῑ we may choose a neighbourhood W of p(P∞ ) in C such that all
connected components Vi of the preimage p−1 (W ) contain only one preimage of
p(P∞ ). One of them, let’s say V0 , contains of course P∞ itself.
We choose an arbitrary path cι : [0, ∞) → W \{p(P∞ )} ⊆ C that converges to
p(P∞ ). Let ĉι : [0, ∞) → V0 be an arbitrary lift of cι via p in V0 . Since we may
choose W arbitrarily small, V0 may become arbitrarily small and ĉι converges to P∞ .
Now let c : [0, ∞) → H be the preimage of ĉι under ι, i.e. the path such that
ι c = ĉι . We project it by p to H/ ∗ , i.e. we take the path p c. Its image under
ϕ is ϕ p c = p ĉι = cι and converges to p(P∞ ) in C. Thus p c converges in
H/ ∗ , since ϕ is an open map.
Since also p is open, c converges to some t∞ ∈ H ∪ {cusps of ∗ }. By continuity
of ῑ one has ῑ(t∞ ) = P∞ . Thus ῑ is surjective onto ι .
One obtains immediately the following conclusions.
Corollary 4.15. If ι : H → Tg leads to a Teichmüller curve C, then

a) the boundary points of the Teichmüller disk ι are precisely the end points of the
Jenkins–Strebel rays in ι with initial point ι(i).
b) These boundary points correspond to the fixed points of parabolic elements in
the projective Veech group.
c) Each boundary point of the Teichmüller curve C is obtained by contracting the
core curves of the cylinders in the direction of v, where v is the eigenvector of a
parabolic element in the Veech group.
This finishes the proof of Theorem 4.1.
5 Schottky spaces
In this chapter we first recall the construction of Schottky coverings for smooth and
stable Riemann surfaces. We use them to define markings called Schottky structures.
In the smooth case they are classified by the well known Schottky space Sg , a complex
manifold of dimension 3g − 3 (if g ≥ 2). In [13] it was shown that also the Schottky
structures on stable Riemann surfaces are parameterized by a complex manifold Sg .
Here we show how to obtain Sg from Braungardt’s extension Tg of the Teichmüller
space introduced in Section 3. In the last section of this chapter we study the image
of a Teichmüller disk in the Schottky space.
5.1 Schottky coverings

We recall the basic definitions and properties of Schottky uniformization of Riemann
surfaces. We introduce the Schottky space Sg and sketch, following [13], the con-
struction of a universal family over it.
Definition 5.1. A group ⊂ PSL2 (C) of Möbius transformations on P1 (C) is called a

Schottky group if there are, for some g ≥ 1, disjoint closed simply connected domains
D1 , D1 , . . . , Dg , Dg bounded by Jordan curves Ci = ∂Di , Ci = ∂Di , and generators
γ1 , . . . , γg of such that γi (Ci ) = Ci and γi (Di ) = P1 (C) − Di for i = 1, . . . , g.
The generators γ1 , . . . , γg are called a Schottky basis of .
In Schottky’s original paper [32], the Di in the definition were disks. With the
same notation let
g
F = F () = P1 (C) − i=1 (Di ∪ Di ) and = () = γ ∈ γ (F ).
It is well known, see e.g. [25, X.H.] that is a Kleinian group, free of rank g with free
generators γ1 , . . . , γg , that is the region of discontinuity of , and that X = / is
a closed Riemann surface of genus g. The quotient map → X is called a Schottky
covering.
An important fact is the following uniformization theorem:
Proposition 5.2. Every compact Riemann surface X of genus g ≥ 1 admits a Schottky

covering by a Schottky group of rank g.
Proof. The proof is based on the following construction that we shall extend to stable
Riemann surfaces in Section 5.4: choose disjoint g simple loops c1 , . . . , cg on X which
are independent in homology, i.e. F = X − i=1 ci is connected. Then F is confor-
mally equivalent to a plane domain that is bounded by 2g closed Jordan curves. For
i = 1, . . . , g denote by Ci and Ci the two boundary components of F that result from
cutting along ci . Now let g be a free group on generators ϕ1 , . . . , ϕg , and take a
copy Fw of F for every element w ∈ g . The Fw are glued according to the following
rule: if w and w are reduced words in ϕ1 , . . . , ϕg and if w = w ϕi then the boundary
component Ci on Fw is glued to Ci on Fw ; if w ends with ϕi−1 the roles of Ci and
Ci are interchanged. By this construction we obtain a plane domain together with
a holomorphic action of g on it: an element ϕ ∈ g maps the copy Fw to Fwϕ . The
crucial step in the proof now is to show that this action extends to all of P1 (C), i.e. g
acts by Möbius transformations. For this we refer to [3, Chapter IV, Theorem 19 F].
Definition 5.3. Let S̃g be the set of all (γ1 , . . . , γg ) ∈ PSL2 (C)g that generate a
Schottky group and form a Schottky basis for . The set Sg of equivalence classes
of g-tuples (γ1 , . . . , γg ) ∈ S̃g under simultaneous conjugation is called the Schottky
space of genus g.
For a point s = (γ1 , . . . , γg ) ∈ S̃g let (s) be the Schottky group generated by
γ1 , . . . , γg , (s) the region of discontinuity of (s), and X(s) = (s)/ (s) the
associated Riemann surface. This leads to an alternative description of the Schottky
space:
Remark 5.4. Sg is the set of equivalence classes of pairs (X, σ ), where X is a Riemann
surface of genus g and σ : g → PSL2 (C) is an injective homomorphism such that
:= σ (g ) is a Schottky group and ()/ ∼ = X.
(X, σ ) and (X , σ ) are equivalent if there is some A ∈ PSL2 (C) such that σ (γ ) =
Aσ (γ )A−1 for all γ ∈ g . Note that then X is isomorphic to X.
To endow Sg with a complex structure we proceed as follows: Taking the fixed

points and the multipliers of the γi we obtain an embedding of S̃g as an open subdomain
of P1 (C)3g . For g = 1 each equivalence class contains a unique Möbius transformation
of the form z → λz for some λ ∈ C, 0 < |λ| < 1. If g ≥ 2 we find in each equivalence
class in S̃g a unique representative (γ1 , . . . , γg ) such that γ1 and γ2 have attracting
fixed points 0 and 1, respectively, and γ1 has repelling fixed point ∞. This defines a
section to the projection S̃g → Sg and embeds Sg as a closed subspace of S̃g which,
moreover, lies in {0} × {∞} × {1} × C3g−3 ⊆ P1 (C)3g . Thus we have shown
Proposition 5.5. a) S1 is a punctured disk.

b) For g ≥ 2, Sg carries a complex structure as an open subdomain of C3g−3 .
Our next goal is to show that this complex structure on Sg is natural. The main
step in this direction is
Proposition 5.6. The forgetful map μ : Sg → Mg , that sends s = (X, σ ) to the

isomorphism class of X, is analytic and surjective.
Proof. The surjectivity of μ follows from Proposition 5.2. To show that μ is analytic
we use the fact that Mg is a coarse moduli space for Riemann surfaces. Therefore
it suffices to find a holomorphic family π : Cg → Sg of Riemann surfaces over Sg
which induces μ in the sense that for s ∈ Sg , μ(s) is the isomorphism class of the
fibre Cs = π −1 (s) ⊂ Cg .
The family Cg is obtained as in Section 3.3: Let

g = {(s, z) ∈ Sg × P1 (C) : z ∈ (s)}.
g is a complex manifold on which the free group g acts holomorphically by
ϕ(s, z) = (s, σ (ϕ)(z)) for s = (X, σ ) ∈ Sg , ϕ ∈ g and z ∈ (s).
The projection pr1 : g → Sg onto the first component factors through the orbit
space Cg = g /g , and the induced map π : Cg → Sg is the family of Riemann
surfaces we were looking for.
The family Cg is in fact universal for Riemann surfaces with Schottky structure, a
kind of marking that we now recall from [13, Section 1.3]:
Definition 5.7. a) Let U → S be an analytic map of complex manifolds and ⊂

Aut(U/S) a properly discontinuous subgroup. Then the analytic quotient map U →
U/ = C is called a Schottky covering if the induced map C → S is a family of
Riemann surfaces and if for every x ∈ S the restriction Ux → Cx of the quotient map
to the fibres is a Schottky covering.
b) A Schottky structure is a Schottky covering U → U/ = C together with
an equivalence class of isomorphisms σ : g → , where σ and σ are considered
equivalent if they differ only by an inner automorphism of g .
Note that the construction in the proof of Proposition 5.6 endows the family Cg /Sg
with a Schottky structure.
A Schottky structure on a single Riemann surface X is given by a Schottky covering
→ / = X and an isomorphism σ : g → . Comparing the respective
equivalence relations we find that the points (X, σ ) in Sg correspond bijectively to
the isomorphism classes of Riemann surfaces with Schottky structure. In fact a much
stronger result holds:
Theorem 5.8. Sg is a fine moduli space for Riemann surfaces with Schottky structure.
Proof. Assume that C/S is a family of Riemann surfaces and let (U → U/ = C,

σ : g −− ∼→ ) be a Schottky structure on C. Then we have a map f : S → S which
g
maps a point x to the isomorphism class of the Schottky covering Ux → Cx . We have
to show that f is analytic. Then the other properties of a fine moduli space follow
easily from the definitions, namely that C is the fibre product Cg ×Sg S and that U is
isomorphic to g ×Cg C = g ×Sg S such that the projection U → g onto the first
factor is equivariant for the actions of and g via the isomorphism σ .
The universal property of Mg as a coarse moduli space gives us, as above for μ,
that the composition μ f is analytic. Since μ has discrete fibres, it therefore suffices
to show that f is continuous. This is quite subtle, see [13, § 3].
5.2 Relation to Teichmüller space
In this section we explain that Schottky space can be obtained as a quotient space of
the Teichmüller space which was introduced in Section 3.3. For this purpose we first
endow the universal family Cg,0 over the Teichmüller space Tg = Tg,0 with a Schottky
structure as follows:
Let a1 , b1 , . . . , ag , bg be a set of standard generators of πg , the fundamental group
of the reference surface Xref ; this means that they satisfy the relation
g

ai bi ai−1 bi−1 = 1.
i=1
Then b1 , . . . , bg are homologically independent, hence the construction in the proof

of Proposition 5.2 provides us with a corresponding Schottky covering ref → Xref .
The group Aut(ref /Xref ) of deck transformations is isomorphic to the free group
on b1 , . . . , bg . Denoting Uref → Xref the universal covering, there is a covering map
Uref → ref over Xref . The groupAut(Uref / ref ) is the kernel Nα of the homomorphism
α : πg → g which maps bi to ϕi and ai to 1; in other words, Nα is the normal closure
in πg of the subgroup generated by a1 , . . . , ag .
In Section 3.3 we described the family + g,0 → Cg,0 of universal coverings of the
surfaces in the family Cg,0 ; the fundamental group πg and hence also Nα acts on the
fibres of this covering, and we obtain:
Remark 5.9. The induced map + g,0 /Nα → Cg,0 is a Schottky covering, and the uni-
versal Teichmüller structure τ : πg −−∼→ Aut(+ /C ) (cf. Theorem 3.13) descends
g,0 g,0
via α to a Schottky structure σα : g = πg /Nα −−∼→ Aut((+ /N )/C ) on C .
g,0 α g,0 g,0
By Theorem 5.8 this Schottky structure induces an analytic map sα : Tg → Sg . To

describe sα as the quotient map for a subgroup of the mapping class group g , we first
identify g with the group Out + (πg ) of orientation preserving outer automorphisms of
πg ; then, to a diffeomorphism f : Xref → Xref , we associate the induced automorphism
ϕ = f∗ : πg → πg . It follows from the Dehn–Nielsen theorem that this gives an
isomorphism g −− ∼→ Out + (π ). In this chapter, by ϕ ∈ we always mean an
g g
+
element of Out (πg ).
Proposition 5.10. a) sα is the quotient map for the subgroup

g (α) = {ϕ ∈ g : α ϕ ≡ α mod Inn(πg )}
of the mapping class group g (where Inn(πg ) denotes the group of inner automor-
phisms).
b) sα : Tg → Sg is the universal covering of the Schottky space.
c) sα lifts to maps s̃α and ωα that make the following diagram commutative:
+
g,0
GG
GG /N
GG α
GG
GG
#
ωα
/πg +g,0 /Nα
/ g
vv
vv
vvv /g
v
{v
v s̃α
Cg,0 / Cg
sα
Tg I / Sg
II ww
II ww
II w
I ww
/ g III ww μ
$ w{ w
Mg
Proof. a) Let x = (X, f ) ∈ Tg . Recall from Section 3.3 that the fibre over x in
+ +
g,0 is the component (x) of the region of discontinuity of the quasi-Fuchsian
group Gx associated to x. The universal Teichmüller structure on Cg,0 induces an
isomorphism τx : πg → Gx = Aut(+ (x)/X). From Remark 5.9 we see that the
point sα (x) = (X, σ ) ∈ Sg is given by the restriction σα,x of σα to the fibre over x;
explicitly,
∼
σ = σα,x : g = πg /Nα −−→ Aut((+ (x)/τx (Nα ))/X) = Gx /τx (Nα ).
For ϕ ∈ g we have sα (x) = sα (ϕ(x)) if and only if σα,x = σα,ϕ(x) up to an inner
automorphism. Since τϕ(x) = τx ϕ −1 this happens if and only if ϕ induces an inner
automorphism on πg /Nα , i.e. if and only if ϕ ∈ g (α).
b) This is clear from the fact that Tg is simply connected and g (α) is torsion
free, hence sα is unramified. Using the construction in a) one can give a direct proof
which in turn provides an independent proof that Tg is simply connected, see [13,
Proposition 6].
c) It follows from Remark 5.9 that + g,0 /Nα → Cg,0 is a Schottky covering.
Therefore, by the universal property of Sg (Theorem 5.8), Cg,0 is the fibre product
Tg ×Sg Cg , and s̃α is the projection to Cg . Moreover the Schottky covering +
g,0 /Nα →
Cg,0 is a pullback of the universal Schottky covering g → Cg , i.e. + g,0 /Nα =
Cg,0 ×Cg g , and again ωα is the projection to the second factor.
In fact, the action of g (α) on Tg extends to + +

g,0 , g,0 /Nα and Cg,0 ; then s̃α and
ωα are the quotient maps for these actions.
5.3 Schottky coverings of stable Riemann surfaces

In this and the following section we introduce a partial compactification Sg of Sg that
fits in between Tg and Mg . We have presented two different ways to define Sg , and we
shall see that both are suited for extension to stable Riemann surfaces: The first way
is to construct Schottky coverings for surfaces with nodes, define Schottky structures
and find parameters for them. This approach was pursued in [13] and will be sketched
in this section. The other possibility is to extend the action of g (α) to (part of) the
boundary of Tg and show that the quotient exists and has the desired properties; this
will be done in Section 5.4.
Definition 5.11. Let X be a stable Riemann surface of genus g. A cut system on X is

a collection of disjointsimple loops c1 , . . . , cg on X, not passing through any of the
g
nodes, such that X − i=1 ci is connected.
Proposition 5.12. On any stable Riemann surface there exist cut systems.
Proof. Let f : Xref → X be a deformation; we must find disjoint and homologically

independent loops c̃1 , . . . , c̃g on Xref that are disjoint from the loops a1 , . . . , ak that are
contracted by f . For this we complete a1 , . . . , ak to a maximal system a1 , . . . , a3g−3
of homotopically independent disjoint loops (such a system decomposes Xref into
pairs of pants). Among the ai we find ai1 , . . . , aig that are homologically independent.
If iν > k we take c̃ν = aiν , and for iν ≤ k we replace aiν by a loop c̃ν that is homotopic
to aiν and disjoint from it.
Once we have found c1 , . . . , cg as above, we proceed as
in the proof of Proposi-
g
tion 5.2 to construct a Schottky covering of X: Let F = X − i=1 ci , take a copy Fw
of F for each w ∈ g , and glue these copies exactly as before to obtain a space .
Of course, neither F nor is planar whenever X has nodes. In all cases, the complex
structure on X lifts to a structure of a one-dimensional complex space on . The group
g acts on this space by holomorphic automorphisms, i.e. there is an isomorphism
g → = Aut(/X), and X is isomorphic to / as complex space.
Definition 5.13. The covering → X constructed above for a cut system c =

(c1 , . . . , cg ) on a stable Riemann surface X is called the Schottky covering of X relative
to c. A covering of X is called a Schottky covering if it is the Schottky covering relative
to some cut system.
The next goal is to define a space Sg that classifies Schottky coverings in a way
analogous to Definition 5.3. Since the covering space is in general not a subspace
of P1 (C) and thus the group of deck transformations not a subgroup of PSL2 (C), we
cannot directly extend 5.3.
A closer look at the construction of a Schottky covering → / = X of a
stable Riemann surface X shows the following:
Each irreducible component L of is an open dense subset of a projective line;

more precisely, the stabilizer of L in is a Schottky group as in Definition 5.1, and
L is its region of discontinuity. Moreover the intersection graph of the irreducible
components of is a tree (hence is called a tree of projective lines).
Therefore, for each irreducible component L, there is a well defined projection
πL : → L which is the identity on L: For an arbitrary point x ∈ there is a unique
chain L0 , L1 , . . . , Ln = L of mutually distinct components such that x ∈ L0 and Li
intersects Li+1 for i = 0, . . . , n − 1; then define πL (x) to be the intersection point of
Ln−1 and L.
An end of is an equivalence class of infinite chains L0 , L1 , L2 , . . . of irreducible
components as above (i.e. Li = Lj for i = j and Li ∩Li+1 = ∅), where two chains are
equivalent if they differ only by finitely many components. Let ∗ = ∪{ends of }.
Clearly the projection πL to a component L can be extended to ∗ .
For any three different points or ends y1 , y2 , y3 in ∗ there is a unique compo-
nent L = L(y1 , y2 , y3 ) (called the median of the three points) such that the points
πL (y1 ), πL (y2 ), πL (y3 ) are distinct. Now given any four distinct points or ends
y1 , . . . , y4 in ∗ we can define a cross ratio λ(y1 , . . . , y4 ) by taking the usual cross
ratio of πL (y1 ), . . . , πL (y4 ) on the median component L = L(y1 , y2 , y3 ) of the first
three of them; note that λ(y1 , . . . , y4 ) will be 0, 1 or ∞ if πL (y4 ) coincides with
πL (y1 ), πL (y2 ) or πL (y3 ).
To obtain parameters for the group observe that any γ ∈ , γ = 1, has exactly
two fixed points on the boundary of , where boundary points of are either points
in the closure of a component, or ends of ; one of the fixed points is attracting, the
other repelling. For any four different (primitive) elements γ1 , . . . , γ4 in we define
λ(γ1 , . . . , γ4 ) to be the cross ratio of their attracting fixed points. It is a remarkable
fact that from these cross ratios both the space and the group ⊂ Aut() can be
recovered. For any particular Schottky covering finitely many of them suffice, but
for different Schottky coverings we must take, in general, the cross ratios of different
elements of g . To parameterize all Schottky coverings we therefore have to use
infinitely many of these cross ratios. We consider them as (projective) coordinates
on an infinite product of projective lines P1 (C). The cross ratios satisfy a lot of
algebraic relations, which define a closed subset B of this huge space. Every point of
B represents a tree of projective lines as above together with an action of on it.
Sg is the open subset of B, where this action defines a Schottky covering. For details
and in particular the technical complication caused by the presence of infinitely many
variables and equations, see [13, §2] and [17]. In principle, one can proceed as in
Section 5.1 to construct a family of stable Riemann surfaces over Sg .
Given a family C/S of stable Riemann surfaces over a complex manifold S, we
can define the notion of a Schottky covering U/S → C/S and of a Schottky structure
on U exactly as in Definition 5.7, except that now U is not assumed to be a manifold,
but only a complex space. It is shown in [13, §3] that the family over Sg carries a
universal Schottky structure:
Theorem 5.14. Sg is a fine moduli space for stable Riemann surfaces with Schottky
structure.
5.4 Sg as quotient of Tg
It is not possible to extend the quotient map sα : Tg → Sg constructed in Section 5.2
to the whole boundary of Tg in Tg . Instead we shall, for each α, identify a part Tg (α)
of Tg to which the action of g (α) and hence the morphism sα can be extended. It
will turn out that the quotient space is the extended Schottky space Sg described in
the previous section.
We begin with the definition of the admissible group homomorphisms α and the
associated parts Tg (α) of Tg .
Definition 5.15. a) A surjective homomorphism α : πg → g is called symplectic

if there are standard generators a1 , b1 , . . . , ag , bg of πg (in the sense of Section 5.2)
such that α(ai ) = 1 for i = 1, . . . , g.
b) Recall from Section 3 that a point in Tg can be described as an equivalence
class of pairs (X, f ), where X is a stable Riemann surface and f : Xref → X is a
deformation (see Corollary 3.9).
For a symplectic homomorphism α : πg → g let
Tg (α) = {(X, f ) ∈ Tg : ker(π1 (f )) ⊆ ker(α)}.
Proposition 5.16. a) Tg (α) is an open subset of Tg ; it contains Tg and is invariant

under the group g (α) introduced in Proposition 5.10.
b) Tg is the union of the Tg (α), where α runs through the symplectic homomor-
phisms.
c) The restriction to Tg (α) of the universal covering p : Tg → Mg is surjective for
every symplectic α.
Proof. a) Let (X, f ) be a point in Tg and c1 , . . . , ck the loops on Xref that are contracted
under f . Then the kernel of π1 (f ) : πg → π1 (X) is the normal subgroup generated
by c1 , . . . , ck . The local description of Tg in Corollary 3.8 shows that there is a
neighbourhood U of (X, f ) in Tg such that for every (X , f ) ∈ U the map f : Xref →
X contracts a subset of {c1 , . . . , ck }. Hence the kernel of π1 (f ) is contained in
ker(π1 (f )). Thus if (X, f ) ∈ Tg (α), also U ⊆ Tg (α). The remaining assertions are
clear.
b) Again let (X, f ) be a point in Tg and c1 , . . . , ck the loops on Xref contracted
by f . By Proposition 5.12 we can find a cut system a1 , . . . , ag on X and a cor-
responding Schottky covering. This covering induces a surjective homomorphism
π1 (X) → g . Composing this homomorphism with π1 (f ) yields a homomorphism
α : πg → g which corresponds to a Schottky covering of Xref (relative to the cut
system f −1 (a1 ), . . . , f −1 (ag )) and hence is symplectic. By construction, c1 , . . . , ck

are in the kernel of α.
c) Let α : πg → g be symplectic and a1 , b1 , . . . , ag , bg standard generators of
πg such that α(ai ) = 1 for all i. For an arbitrary stable Riemann surface X choose a
deformation f : Xref → X and let c1 , . . . , ck be the loops that are contracted by f . As
in the proof of b) we find standard generators a1 , b1 , . . . , ag , bg such that the cj are
contained in the normal subgroup generated by the ai . The map ai → ai , bi → bi
defines an automorphism ϕ of πg and thus an element of g . Then by construction
(X, f ϕ) lies in Tg (α) and p(X, f ϕ) = X.
As a side remark we note that Tg (α) is not only invariant under g (α), but also
under the larger “handlebody” group
Hg (α) = {ϕ ∈ g : ϕ(Nα ) = Nα }
(where Nα = ker(α) as in Section 5.2). Note that Hg (α) is the normalizer of g (α)
in g , and that we have an exact sequence
1 → g (α) → Hg (α) → Out(g ) → 1.
The quotient space Ŝg = Tg /Hg (α) = Sg / Out(g ) is a parameter space for Schottky
groups of rank g (without any marking).
Proposition 5.17. For any symplectic homomorphism α : πg → g , the quotient

space Tg (α)/ g (α) is a complex manifold Sg (α).
Proof. This is a local statement which is clear for points (X, f ) ∈ Tg since g (α)
is torsion free. For an arbitrary x = (X, f ) ∈ Tg we saw in Section 3.2 that the
Dehn twists τ1 , . . . , τk around the loops c1 , . . . , ck that are contracted by f generate
a finite index subgroup x0 of the stabilizer x of x in g (the quotient being the
finite group Aut(X)). Let α be a symplectic homomorphism with respect to standard
generators a1 , b1 , . . . , ag , bg , and assume (X, f ) ∈ Tg (α). Since the ci are in the
normal subgroup generated by a1 , . . . , ag , they do not intersect any of the aj and thus
τi (aj ) = aj for all i and j . This shows x ⊆ g (α).
Now choose a neighbourhood U of x = (X, f ) in Tg (α) which is precisely invariant
under x . Then it follows, from Proposition 3.7 (and Definition 3.6), that U/ x is a
complex manifold.
For any two sets a1 , b1 , . . . , ag , bg and a1 , b1 , . . . , ag , bg of standard generators,
ai → ai , bi → bi defines an automorphism of πg . Therefore for any two symplectic
homomorphisms α and α there is an automorphism ψ ∈ g such that α = α
ψ. Then clearly Nα = ψ(Nα ) and g (α ) = ψg (α)ψ −1 . This shows that, as
an automorphism of Tg , ψ maps Tg (α) to Tg (α ) and descends to an isomorphism
ψ : Sg (α) → Sg (α ). We have shown:
Remark 5.18. The complex manifolds Sg (α) are isomorphic for all symplectic ho-
momorphisms α.
It remains to show that the Sg (α) coincide with the fine moduli space Sg of Sec-
tion 5.3. This is achieved by showing that Sg (α) satisfies the same universal property
as Sg :
Proposition 5.19. For any symplectic α, Sg (α) is a fine moduli space for stable
Riemann surfaces with Schottky structure and hence isomorphic to Sg .
Proof. The idea of the proof is to endow the universal family over Tg (α) with a
Schottky structure and to transfer this to a Schottky structure on the image family over
Sg (α).
Before explaining this for the whole family we consider a single stable Riemann
surface X. Let d1 , . . . , dk be the nodes on X, f : Xref → X a deformation and
α : πg → g a symplectic homomorphism such that x = (X, f ) ∈ Tg (α). In
Section 3.3 we described the universal covering ˆ + (x) → ˆ + (x)/Gx = X of X
ˆ
with cusps over the nodes. Recall that (x) is the union of the plane region + (x)
+
with the common boundary points of the doubly cusped regions lying over the nodes di ,
and that Gx is isomorphic to πg .
Remark 5.20. Using the above notation, let ρ : πg → Gx be an isomorphism and

ˆ + (x)/NαGx is a complex space, Gx /NαGx ∼
NαGx = ker(α ρ −1 ) ⊆ Gx . Then = =
g acts holomorphically on , and → /g = X is a Schottky covering.
ˆ + (x) lying
Proof. The key observation is that the stabilizer in Gx of a point d̃i ∈
over di is generated by an element γi corresponding under ρ to a conjugate of the
loop f −1 (di ). Since we assumed (X, f ) ∈ Tg (α), we have γi ∈ NαGx . This shows
that is a complex space, more precisely: a Riemann surface with nodes. The other
assertions then follow directly from the definitions.
The above construction can be carried over to families in the following way: First
ˆ+
consider the universal family Cg over Tg and the universal Teichmüller structure g →
Cg on it. Denote by Cg (α) resp. ˆ g (α) the restriction to Tg (α). Then the quotient
+
ˆ g (α)/Nα is a complex space on which g = πg /Nα acts. The quotient map

space +
ˆ+
g (α)/Nα → Cg (α) is a Schottky covering and the identification of g with the
group of deck transformations defines a Schottky structure.
The group g (α) acts not only on Tg (α), but also on ˆ+
g (α) as follows: for
ˆ + ˆ +
ϕ ∈ g (α) and (x, z) ∈ g (α) with x ∈ Tg (α) and z ∈ (x) we set
ϕ(x, z) = (ϕ(x), z).
Note that the groups Gx and Gϕ(x) are the same (only the isomorphism with πg has
changed); therefore ˆ + (x) = ˆ + (ϕ(x)). This action, which is trivial on the fibres,
descends to actions of g (α) on ˆ+
g (α)/Nα and on Cg (α). The respective orbit spaces
give a family Cg = Cg (α)/ g (α) over Sg (α) and a Schottky structure on it. Using
the universal property of the family over Tg (α) (see Theorem 3.13) and the fact that
Schottky structures are locally induced by Teichmüller structures, we find that the
Schottky structure on Cg is in fact universal.
The following diagram collects the relations between the spaces introduced and
used in this section. The horizontal maps are open embeddings, the last two vertical
maps are analytic with discrete fibres; all other maps in the diagram are quotient maps
for the groups indicated (to be precise, the map from Tg (α) to Mg is the restriction of
the orbit map for the action of g on Tg ).

Tg N / T (α)
g
3=3=NNN 33>>NNN
33== NNN 33>> NNNN
33== NNNN / (α) 33 >>> NNN/Ng (α)
33 === NNNg 33 >> NNN
33 == NNN 33 >> NNN
33 = = N N NNN > NNN
3 > NN'
33 === & 3 >
/S
33 == /H (α) Sg 33 << g
33 == g 33 <<
33 === 33 < /H
< g (α)
33 == 33/ g <<<
== / ( ) 33 << /Out(g )
/ g 33
Out g
33 == 33 <<
33 = 33 <
33 / ˆ
33 Ŝg 33 Sg
33 33
33 33
33 33
33 33
33 33

Mg /M
g
5.5 Teichmüller disks in Schottky space

Let ι : H → Tg be a Teichmüller embedding as in Definition 2.4 and = ι(H) its
image in Tg . Let Stab() be the stabilizer of in g . We have seen in Section 2.4.3
that Stab() maps surjectively to the projective Veech group ι of ι (see Defini-
tion 2.17); the kernel of this map is the pointwise stabilizer of .
In this section we assume that ι is a lattice in PSL2 (R), or equivalently that the
image Cι of in Mg is a Teichmüller curve (cf. Corollary 2.21). As mentioned in the
introduction, Veech showed that Cι is not a projective curve and thus cannot be closed
in Mg .
Proposition 5.21. Let ι : H → Tg be a Teichmüller embedding such that ι is a lattice

in PSL2 (R). Then there exists a symplectic homomorphism α : πg → g such that
Stab() ∩ g (α) = {1}.
Since g (α) is torsion free, this implies that the intersection is infinite. As a
consequence, the image of the Teichmüller disk in the Schottky space Sg is the
quotient by an infinite group and in particular not isomorphic to a disk.
Proof. Denote by and Cι the closures of and Cι in Tg and Mg , respectively. Since

Cι is not closed, we can find a point z ∈ Cι − Cι ; let x ∈ be a point above z. By
Proposition 5.16 b) there is a symplectic homomorphism α such that x ∈ Tg (α).
Let sα : Tg (α) → Sg be the quotient map for g (α) (see Proposition 5.17 and
Proposition 5.19) and let D(ι) = sα () be the image of in Sg . Then the closure
D(ι) of D(ι) in Sg contains sα (x), and we have Cι = μ(D(ι)), cf. the diagram at the
end of Section 5.4.
By our assumption, Cι is Zariski closed in Mg . Therefore μ−1 (Cι ) is an analytic
subset of Sg . D(ι) is an irreducible component of μ−1 (Cι ) and hence also an analytic
subset.
Recall, from Corollary 2.21, that / Stab() is the normalization of Cι . Fur-
thermore, by Proposition 4.14, is isomorphic to H ∪ {cusps of ι∗ }. Therefore
/ Stab() is the normalization of Cι . The restriction of the quotient map →
/ Stab() to the intersection α = ∩ Tg (α) factors through sα . If the intersection
Stab() ∩ g (α) was trivial, this restriction would be an isomorphism. But then α
would be isomorphic to an analytic subset of a complex manifold. This is impossible
since α contains x ∈ Tg − Tg and hence is not a complex space.
References
[1] W. Abikoff, Degenerating families of Riemann surfaces. Ann. of Math. (2) 105 (1977),
29–44. 298, 318, 319, 320
Springer-Verlag, Berlin 1980. 300, 301, 302
[3] L.V.Ahlfors and L. Sario, Riemann surfaces. Princeton Math. Ser. 26, Princeton University
Press, Princeton, N.J., 1960. 337
[4] L. Bers, Uniformization, moduli, and Kleinian groups. Bull. London Math. Soc. 4 (1972),
257–300. 295
[5] L. Bers, Fiber spaces over Teichmüller spaces. Acta. Math. 130 (1973), 89–126. 319, 321
[6] I. Bouw and M. Möller, Teichmüller curves, triangle groups, and Lyapunov exponents.
Preprint, 2005. arXiv:math.AG/0511738. 296
[7] V. Braungardt, Abschluß von Überlagerungen. Diploma Thesis, Karlsruhe 1996. 316
[8] V. Braungardt, Überlagerungen von Modulräumen für Kurven. PhD Thesis, Karlsruhe
2001. 298, 315, 316, 317, 318, 319, 320
[9] K. Calta, Veech surfaces and complete periodicity in genus two. J. Amer. Math. Soc. 17
(2004), 871–908. 296
[10] P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus. Inst.
Hautes Études Sci. Publ. Math. 36 (1969), 75–109. 297
[11] C. J. Earle and F. P. Gardiner, Teichmüller disks and Veech’s F -structures. In Extremal
Riemann surfaces (San Francisco, CA, 1995), Contemp. Math. 201, Amer. Math. Soc.,
Providence, R.I., 1997, 165–189. 296, 303, 311, 312
76, Amer. Math. Soc., Providence, R.I., 2000. 303, 306
[13] L. Gerritzen and F. Herrlich, The extended Schottky space. J. Reine Angew. Math. 389
(1988), 190–208. 300, 336, 338, 340, 341, 342
[14] H. Grauert and R. Remmert, Coherent analytic sheaves. Grundlehren Math. Wiss. 265,
Springer-Verlag, Berlin 1984. 315, 316
[15] E. Gutkin and C. Judge, Affine mappings of translation surfaces: geometry and arithmetic.
Duke Math. J. 103 (2000), 191–213. 297
[16] J. Harris and I. Morrison, Moduli of curves. Grad. Texts in Math. 187, Springer-Verlag,
New York 1998. 295, 318
[17] F. Herrlich, Moduli for stable marked trees of projective lines. Math. Ann. 291 (1991),
643–661. 342
[18] F. Herrlich and G. Schmithüsen, An extraordinary origami curve. Preprint, 2005.
arXiv:math.AG/0509195. 297
[19] P. Hubert and S. Lelièvre, Prime arithmetic Teichmüller discs in H(2). Israel J. Math.
151 (2006), 281–321. 297
[20] P. Hubert and T. A. Schmidt, Invariants of translation surfaces. Ann. Inst. Fourier
(Grenoble) 51 (2001), 461–495. 296, 311
Tokyo 1992. 320
[22] F. Knudsen, The projectivity of the moduli space of stable curves. I, II, III. Math. Scand.
39 (1976), 19–55; 52 (1983), 161–199, 200–212. 297
[23] P. Lochak, On arithmetic curves in the moduli spaces of curves. J. Inst. Math. Jussieu 4
(2005), 443–508. 296, 297, 303
[24] M. Maier, Schnittgraphen von stabilen Kurven, die als Randpunkte von Origamikurven
vorkommen. Diploma Thesis, Karlsruhe 2005. 298
[25] B. Maskit, Kleinian groups. Grundlehren Math. Wiss. 287, Springer-Verlag, Berlin 1988.
317, 336
205–221. 299, 322, 323, 334
[27] C. T. McMullen, Billiards and Teichmüller curves on Hilbert modular surfaces. J. Amer.
Math. Soc. 16 (2003), 857–885. 296, 303, 311, 313, 314
[28] M. Möller, Variations of Hodge structure of Teichmüller curves. J. Amer. Math. Soc. 19
(2006), 327–344. 296
[29] S. Nag, The complex analytic theory of Teichmüller spaces. Canad. Math. Soc. Ser. Mono-
graphs Adv. Texts, John Wiley & Sons, New York 1988. 303, 310
[30] G. Schmithüsen, An algorithm for finding the Veech group of an origami. Experiment.
Math. 13 (2004), 459–472. 297
[31] G. Schmithüsen, Veech groups of origamis. PhD Thesis, Karlsruhe 2005. 297
[32] F. Schottky, Ueber eine specielle Function, welche bei einer bestimmten linearen Transfor-
mation ihres Arguments unverändert bleibt. J. Reine Angew. Math. 101 (1887), 227–272.
299, 336
1984. 323
[34] W. A. Veech, Teichmüller curves in moduli space, Eisenstein series and an application to
triangular billiards. Invent. Math. 97 (1989), 553–583. 296, 303, 311, 312, 331
Part B
The group theory, 1

Chapter 7
Introduction to mapping class groups of surfaces

and related groups
Shigeyuki Morita
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
2 Diffeomorphism groups and diffeotopy groups of differentiable manifolds . 354
3 Mapping class groups of surfaces . . . . . . . . . . . . . . . . . . . . . . 357
4 Automorphism groups of free groups and IA automorphism groups . . . . 361
5 Dehn twists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
6 Mapping class groups acting on the homology of surfaces and the Torelli
groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
7 Johnson homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . 371
8 Teichmüller space and Outer Space . . . . . . . . . . . . . . . . . . . . . 376
9 Symplectomorphism groups of surfaces . . . . . . . . . . . . . . . . . . . 378
10 Extensions of the Johnson homomorphism and the flux homomorphism . . 380
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
1 Introduction
Let g denote a closed oriented surface of genus g. The mapping class group of g ,
denoted by Mg , is the group of isotopy classes of orientation preserving diffeomor-
phisms of g . This group has been investigated from various points of view for many
years.
First of all, this group has been one of the main objects in the combinatorial group
theory, the other one being the automorphism group of a finitely generated free group.
Secondly, Mg acts on the Teichmüller space and the quotient space is the moduli
space of genus g Riemann surfaces which is a very important space in both algebraic
geometry and complex analysis. Thirdly, this group has been playing crucial roles
in the theory of 3-manifolds in relation to Heegaard decompositions as well as the
geometry of surface bundles over the circle.
Reflecting this situation, there exist already many survey papers concerning various
aspects of the mapping class group. We have the famous book by Birman [7] and
several survey papers such as [8], [9], [27], [10]. Ivanov’s paper [34] gives a very nice
354 Shigeyuki Morita
introduction to the present state of the study of the mapping class group (see also [32],
[33]). Also we have the survey paper [37] by Johnson on the structure of the Torelli
group which is a very important subgroup of the mapping class group.
There are also survey papers on the cohomological structure of the mapping class
group or the moduli space of Riemann surfaces such as Harer [26], Hain–Looijenga
[24] and the author [69], [71].
The purpose of this chapter is to describe somewhat different points of view in the
study of the mapping class group and to suggest possible new directions for future
research. More precisely, we would like to consider this as a special case of the study
of the structure of the diffeomorphism group as well as the diffeotopy group of general
C ∞ manifolds. We also would like to seek for similarity and difference between the
structures of the mapping class group and some of its closely related groups such as the
automorphism groups of free groups and the symplectomorphism groups of surfaces.
We refer to Vogtmann [84] for a survey of the study of the automorphism groups of
free groups.
This work was partially supported by JSPS Grant No. 16204005.
2 Diffeomorphism groups and diffeotopy groups of

differentiable manifolds
Let M be a closed C ∞ manifold. Then it is a very important problem to determine
the set of all the isomorphism classes of differentiable M-bundles
π : E −→ X
over a given C∞ manifold X. By a standard technique in topology, this problem can
be translated into the following one in homotopy theory. We denote by Diff M the
diffeomorphism group of M equipped with the C ∞ topology and let BDiff M be its
classifying space. Then we have a natural identification
{isomorphism classes of smooth M-bundles over X} ∼
= [X, BDiff M]
where the right hand side denotes the set of all the homotopy classes of continuous
mappings from X to BDiff M. Hence we meet with the problem of determining
the homotopy type of Diff M, in particular the computation of the homotopy groups
πi (Diff M). However this is an extremely difficult problem for a general differentiable
manifold M.
Let us consider the simplest case where X = S 1 . Any smooth M-bundle
π : E −→ S 1
over S 1 can be described as follows. Choose a point x ∈ S 1 and cut the total space E
along the fiber π −1 (x) over x. Then we obtain an M-bundle over the interval I so
that it is diffeomorphic to the product I × M. Observe here that we have chosen a
Chapter 7. Introduction to mapping class groups of surfaces and related groups 355
diffeomorphism π −1 (x) ∼
= M. Now the given bundle E can be recovered as
E = I × M/(1, p) ∼ (0, ϕ(p)) (p ∈ M)
for a certain element ϕ ∈ Diff M. Namely E is obtained from I × M by pasting
{1} × M to {0} × M by the diffeomorphism ϕ. Note that the element ϕ is well-defined
up to isotopy and also note that, if we change the identification π −1 (x) ∼
= M, then the
element ϕ changes into its conjugate element ψϕψ −1 . Thus we obtain
{isomorphism classes of smooth M-bundles over S 1 }
∼
= [S 1 , BDiff M]
∼
= {isotopy classes of elements of Diff M}/conjugacy.
From the above consideration, it is natural to introduce the group consisting of all
the isotopy classes of elements of Diff M which we denote by D(M) and call it the
diffeotopy group of M. It can also be described as the group of path components of
the topological group Diff M, namely
D(M) = π0 (Diff M).
Alternatively, we can also write
D(M) = Diff M/Diff 0 M
where Diff 0 M denotes the identity component of Diff M. Thus we have an extension
1 −→ Diff 0 M −→ Diff M −→ D(M) −→ 1.
This simplest case, namely the determination of D(M) is already a very difficult
problem in general. In fact, the case where M is an n-dimensional sphere S n was one
of the most important subjects during the early years of differential topology. By virtue
of the foundational work of Cerf [13], [14] as well as the solution of the generalized
Poincaré conjecture due to Smale [81], there is a natural isomorphism
D+ (S n ) ∼
= θn+1 (n ≥ 5)
where D+ (S n ) = π0 (Diff + S n ) denotes the orientation preserving diffeotopy group
of S n and θn denotes the group of homotopy n-spheres introduced and studied by
Kervaire and Milnor [44].
Since D(M) is the quotient group of the diffeomorphism group D(M) divided by
the equivalence relation of isotopy which is stronger than (or sometimes equal to) that
of homotopy, D(M) acts on any homotopy invariants of M such as the fundamental
group π1 M and the homology group H∗ (M; Z). We first consider the case of π1 M.
For any abstract group Γ , let Aut Γ denote the automorphism group of Γ . Any
element γ ∈ Γ defines that of Aut Γ which represents the inner automorphism of Γ
by the element γ . This induces a homomorphism Γ → Aut Γ . Let Inn Γ denote the
image of this homomorphism. Clearly the group Γ is abelian if and only if Inn Γ
is the trivial subgroup of Aut Γ . It is easy to see that Inn Γ is a normal subgroup
of Aut Γ . The quotient group Aut Γ /Inn Γ is denoted by Out Γ and it is called the
outer automorphism group of Γ . In these terminologies, we can say that the action of
D(M) on π1 M induces a homomorphism
ρπ : D(M) −→ Out π1 M.
Also the action of D(M) on the homology group H ∗ (M; Z) induces another homo-
morphism
ρH : D(M) −→ Aut H∗ (M; Z).
Then there arise natural questions about these homomorphisms. For example we
could ask whether the homomorphisms ρπ , ρH are surjective or not. We could also
ask whether they are injective or not. In the cases where these questions are answered
negatively, the problem of describing the images as well as the kernels of these ho-
momorphisms arise. It turns out that these questions depend on the global topology
of the manifold M and the above problems are often very difficult to be settled.
If there is given a geometric structure on M, then the automorphism group of this
structure, which is considered to be a subgroup of Diff M, is one of the basic objects
to be studied. Here we would like to mention a few examples.
The first obvious example is the case where there is given a Riemannian metric
on M. Then the corresponding automorphism group is nothing but the isometry group
Isom M. If M is compact, then this group is known to be a compact Lie group sitting
inside Diff M. For example, for the n-sphere S n with the standard metric, we have
Isom S n = O(n + 1) ⊂ Diff S n . It has been one of the main problems in the theory
of differentiable transformation groups to study possible subgroups of Diff M which
are Lie transformation groups for a given manifold M.
The second example is the case where there is given a volume form υ on M.
Then we can consider the subgroup of Diff M, denoted by Diff υ M, which consists
of those diffeomorphisms which preserve the form υ. It is usually called the volume
preserving diffeomorphism group of M. Moser’s theorem in [72] implies that the
inclusion Diff υ M ⊂ Diff + M is a homotopy equivalence. Hence the classifying
spaces BDiff υ M and BDiff + M have the same homotopy type. In particular, we have
a bijection
π0 (Diff υ M) ∼
= π0 (Diff + M) = D+ M.
However the two groups Diff υ M and Diff + M seem to have considerably different
properties as abstract groups and there should be many interesting problems here.
The third example is the case where there is given a symplectic form ω on a 2n
dimensional manifold M. A symplectic form is, by definition, a closed 2-form ω such
that ωn is a volume form on M. The pair (M, ω) is called a symplectic manifold and the
subgroup Symp(M, ω) ⊂ Diff M consisting of those diffeomorphisms which preserve
the form ω is called the symplectomorphism group of (M, ω). Recently there have
been obtained many interesting deep results in geometry and topology of symplectic
manifolds as well as those of symplectomorphism groups (see the book [61] by McDuff
and Salamon which gives an excellent introduction to this field). Let Symp0 (M, ω)
be the identity component of Symp(M, ω) which is a normal subgroup. The quotient
group SD(M, ω) = Symp(M, ω)/Symp0 (M, ω) is called the symplectic mapping

class group of (M, ω). Thus we have the following extension.
1 −→ Symp0 (M, ω) −→ Symp(M, ω) −→ SD(M, ω) −→ 1.
We have an obvious natural homomorphism SD(M, ω) → D+ (M) and there are
many interesting problems concerning it. For example, we can ask about the structure
of the image as well as the kernel of this homomorphism. See Banyaga [3] and McDuff
[60] for basic results concerning the structures of the groups Diff υ M and Symp(M, ω).
Recall that any C ∞ manifold M admits a real analytic structure (see Whitney [86]).
Our final example here is the subgroup Diff ω M ⊂ Diff M consisting of real analytic
diffeomorphisms of M (with respect to a fixed real analytic structure). It seems to
be an interesting problem to investigate whether there exist differences in algebraic
structures between the two groups Diff ω M and Diff M.
3 Mapping class groups of surfaces

The mapping class group of a closed oriented surface g of genus g, which we denote
by Mg , is by definition the oriented diffeotopy group of g . Namely it is the group
consisting of all the isotopy classes of orientation preserving diffeomorphisms of g .
If we denote by Diff + g the group of orientation preserving diffeomorphisms of g
equipped with the C ∞ topology, then we have
Mg = π0 (Diff + g ).
It is easy to see that we have an extension
1 −→ Mg −→ D(g ) −→ Z/2 −→ 1
where the homomorphism D(g ) → Z/2 is induced by the action of D(g ) on the
set of orientations on g or equivalently on the group H2 (g ; Z) ∼= Z.
As is well known, the topology and also the geometry of surfaces g can be roughly
divided into three classes, namely the cases where g = 0, g = 1 and g ≥ 2. From
the topological point of view, the fundamental group π1 g is trivial for g = 0, rank 2
abelian for g = 1 and non-abelian for g ≥ 2. On the other hand, from the geometrical
point of view, each surface g admits a Riemannian metric of constant Gaussian
curvature K where K ≡ 1, 0, −1 for g = 0, g = 1, g ≥ 2 respectively. Furthermore
for the latter two cases, there exist plenty of such metrics up to isotopy and they fit
together to make a nice topological space called the Teichmüller space. It turns out
that the structure of the mapping class group Mg reflects this rough classification of
surfaces rather closely as follows.
First of all, we consider the case where g = 0, namely the case of the sphere S 2 .
Then a theorem of Smale [80] implies that the inclusion
O(3) ⊂ Diff S 2
is a homotopy equivalence. It follows, in particular, that the subgroup Diff + S 2 is

connected. Therefore any orientation preserving diffeomorphism of S 2 is isotopic to
the identity. Hence the genus 0 mapping class group M0 is the trivial group.
Next we assume that g ≥ 1 and consider the action of Mg on π1 g as in §2 where
we considered the case of a general manifold M. It is known that the surface g for
g ≥ 1 is an Eilenberg–MacLane space K(π1 g , 1) meaning that the higher homotopy
groups πi g vanish for all i ≥ 2. Here for a given (abstract) group π and a positive
integer n, any topological space X with the property that

π (i = n)
πi X ∼
=
0 (i
= n)
is called an Eilenberg–MacLane space K(π, n) (we assume that π is an abelian group

in the cases where n ≥ 2).
In fact, in the genus 1 case, the torus T 2 is expressed as R2 /Z2 so that its universal
covering manifold is the plane R2 . Hence all the higher homotopy groups of T 2 vanish
and T 2 is a K(Z2 , 1). In the cases where g ≥ 2, there exist Riemannian metrics on
g which have constant negative curvature −1. It follows that its universal covering
manifold is isometric to the upper half plane H = {(x, y) ∈ R2 ; y > 0} equipped with
the Poincaré metric. Hence g is a K(π1 g , 1). Here recall the standard presentation
of the fundamental group π1 g which is expressed as
π1 g = α1 , . . . αg , β1 , . . . , βg ; ζ (2g generators),
ζ = [α1 , β1 ] . . . [αg , βg ] (defining relation).
Now we state a classical theorem which is usually called the Dehn–Nielsen theorem.
Theorem 3.1 (Dehn–Nielsen, Baer). The natural action of Mg on π1 g induces an

isomorphism
Mg ∼
= Out+ π1 g .
Since π1 T 2 ∼
= Z2 is an abelian group, Aut Z2 = Out Z2 ∼ = GL(2, Z). The
subscript + in Out + Z2 means, in this case, that we consider only matrices with deter-
minant 1. Thus we can write
M1 ∼
= SL(2, Z).
In order to interpret the subscript + for the general case, we briefly recall the definition
of the homology group as well as the cohomology group of an abstract group π (see
Brown’s book [12] for details). It is known that, there exists a K(π, 1) which is a CW
complex. Furthermore it is uniquely defined up to homotopy equivalences. Hence
for any π -module M (namely M is a module and there is given a homomorphism
π → Aut M), we can define the (co)homology group of π with coefficients in M by
setting
H∗ (π ; M) = H∗ (K(π, 1); M),

H ∗ (π ; M) = H ∗ (K(π, 1); M)
where M on the right hand sides denotes the local system over K(π, 1) induced by
the given π action on the module M. Any group homomorphism ρ : π → π induces
a homomorphism ρ∗ : H∗ (π) → H∗ (π ) and similarly for the homology with twisted
coefficients as well as the cohomology group. In particular, we have a homomorphism
Aut π −→ Aut H∗ (π ; M).
It is well known (and not so difficult to see) that the inner automorphisms induce the
trivial action on the homology group so that we obtain a homomorphism
Out π −→ Aut H∗ (π ; M).
Now for any g ≥ 1, g is a K(π1 g , 1) as mentioned above. Hence we have

H∗ (π1 g ; Z) = H∗ (g ; Z). In particular H2 (π1 g ; Z) ∼
= Z and we have a homo-
morphism
Out π1 g −→ Aut H2 (π1 g ; Z) ∼

= Aut Z ∼
= Z/2.
The group Out+ π1 g in Theorem 3.1 is defined to be the kernel of the above homo-
morphism. Sometimes this group is called the orientation preserving outer automor-
phism group of π1 g because it is the subgroup of the whole group consisting of those
outer automorphisms which are induced from orientation preserving diffeomorphisms
of g .
Now we would like to mention the methods of proving Theorem 3.1 somewhat
historically.
First of all, for any topological space X, let E (X) denote the set of all the homotopy
classes of self homotopy equivalences of X. The composition of mappings induces a
natural group structure on E (X). Next for any topological manifold M, let Homeo M
denote the group of all the homeomorphisms of M equipped with the compact open
topology and let H(M) = Homeo M/Homeo0 M denote the quotient group divided
by the identity component Homeo0 M of Homeo M. It is called the homeotopy group
of M. Now if M is a C ∞ manifold, then there is a natural sequence of forgetful
homomorphisms
D(M) −→ H (M) −→ E (M).
One can also introduce the equivalence relation on Diff M, Homeo M induced by the
homotopy of mappings to obtain variants of diffeotopy or homeotopy groups.
In general, these groups are all different from each other and they have their own
meanings and properties. However, a very important and characteristic phenomenon
occurs in dimension 2 and that is the fact that they are all equal for surfaces. Namely
we have isomorphisms
D(g ) ∼= H (g ) ∼
= E (g ),
Mg = D+ (g ) = H+ (g ) ∼
∼ = E+ (g )
where the subscripts + means appropriate subgroups of index 2 consisting of orien-
tation preserving elements. Now elementary homotopy theory implies that there are
canonical isomorphisms
E (g ) ∼
= Out π1 g , E+ (g ) ∼
= Out+ π1 g .
In fact the isomorphism E (X) ∼
= Out π1 X holds for any K(π, 1) space X.
Dehn and then Nielsen [77] proved that the natural map
H+ (g ) −→ Out + π1 g
is surjective. The injectivity of the same map was proved by Baer [2] and, much later,
reproved by Epstein [19]. It may be said that Dehn and Nielsen essentially proved
that the natural map
D+ (g ) −→ Out + π1 g
is surjective, although it is unclear how they recognized the concept of diffeomor-
phisms as well as homeomorphisms which are now strictly distinguished. The injec-
tivity of the above map can be obtained by adapting the proofs of Baer and Epstein
from the context of homeomorphisms to that of diffeomorphisms which are known to
be possible in this low dimensional case.
There are variants of the mapping class group and analogues of Theorem 3.1 for
them as follows. First, if we choose a base point ∗ ∈ g , then we can consider
the subgroup Diff + (g , ∗) ⊂ Diff + g consisting of all the orientation preserving
diffeomorphisms of g which fix the base point ∗. Then we set
Mg,∗ = π0 (Diff + (g , ∗))
and call it the mapping class group of g relative to the base point. There is the
forgetful homomorphism Mg,∗ → Mg which is an isomorphism for g = 0, 1 and
in the cases g ≥ 2, the kernel of this homomorphism is known to be canonically
isomorphic to π1 g . Thus we have an extension
1 −→ π1 g −→ Mg,∗ −→ Mg −→ 1 (g ≥ 2).
Next if we choose an embedded disk D 2 ⊂ g , then we can consider the subgroup
Diff(g , D 2 ) ⊂ Diff + g consisting of all the diffeomorphisms of g which restrict
to the identity of D 2 . Then we set
Mg,1 = π0 (Diff(g , D 2 ))
and call it the mapping class group of g relative to D 2 . Alternatively, we can consider
the compact surface g0 = g \ Int D 2 and the diffeomorphism group Diff(g0 , ∂g0 )
consisting of all diffeomorphisms of g0 which restrict to the identity on the boundary.
Then we can also define

Mg,1 = π0 (Diff(g0 , ∂g0 )).
If we choose a base point ∗ in D 2 ⊂ g , then we have the forgetful homomorphism
Mg,1 → Mg,∗ , and in the case g ≥ 1 the kernel of this homomorphism is known to
be isomorphic to Z so that we have an extension
1 −→ Z −→ Mg,1 −→ Mg,∗ −→ 1 (g ≥ 1). (3.1)
Now we can state the analogue of Theorem 3.1 for the mapping class groups Mg,∗ and
Mg,1 as follows. This should also be considered as a classical theorem going back to
Magnus [54] and Zieschang [87].
Theorem 3.2. There are natural isomorphisms

Mg,∗ ∼
= Aut+ π1 g ,
Mg,1 ∼
= {ϕ ∈ Aut π1 0 ; ϕ(ζ ) = ζ }
g
where ζ = [α1 , β1 ] . . . [αg , βg ] denotes the single defining relation of π1 g with

respect to a standard generating system α1 , . . . , αg , β1 , . . . βg .
We can also consider the mapping class groups of g relative to finitely many
distinguished points as well as finitely many embedded disks on g . However here
we omit them.
4 Automorphism groups of free groups and IA automorphism

groups
Let Fn be a free group of rank n. We denote by Aut Fn the automorphism group of
the free group Fn . If n = 1, then clearly Aut Z ∼
= Z/2. Henceforth we assume that
n ≥ 2. Then the homomorphism ι : Fn → Aut Fn defined by
ι(α)γ = αγ α −1 (α, γ ∈ Fn )
is easily seen to be injective. As was already mentioned in §2 in a general setting, the
image Im ι is denoted by Inn Fn and called the inner automorphism group of Fn . It
can be checked that Inn Fn is a normal subgroup of Aut Fn so that we can consider
the quotient group
Out Fn = Aut Fn /Inn Fn
which is called the outer automorphism group of the free group Fn .
These groups Aut Fn and Out Fn have been one of the main objects of combinatorial
group theory going back to the works of Nielsen and then Magnus from the late 1910s
to the 1930s.
Aut Fn acts naturally on the abelianization of Fn , which is a free abelian group

of rank n. If we choose a system γ1 , . . . , γn of generators for Fn , then we obtain a
homomorphism
ρ0 : Aut Fn −→ GL(n, Z). (4.1)
It is easy to see that ρ0 is trivial on the subgroup Inn Fn . Hence we obtain a homo-
morphism
ρ0 : Out Fn −→ GL(n, Z). (4.2)
Nielsen [74] proved that the above homomorphism is an isomorphism for n = 2 so
that
Out F2 ∼
= GL(2, Z).
However for n ≥ 3, Nielsen [75] also observed that ρ0 is not injective and in [76] he
proved that the following four elements
(i) γ1 → γ2 , γ2 → γ1 , γi → γi (i = 3, . . . , n),
(ii) γ1 → γ1−1 ,
γi → γi (i = 2, . . . , n),
(iii) γ1 → γ1 γ2 , γi → γi (i = 2, . . . , n),
(iv) γ1 → γ2 , γ2 → γ3 , . . . , γn → γ1
generate Aut Fn and hence Out Fn . By this he was able to prove that the homomor-
phims ρ0 (4.1), (4.2) above are surjective. He also gave a finite complete set of defining
relations in terms of the above generators, for both of Aut Fn and Out Fn . Later Mc-
Cool [58] gave a simpler finite presentation for Aut Fn . Also Gersten [21] gave a finite
presentation for the subgroup Aut+ Fn which is the full inverse image under ρ0 of the
subgroup GL+ (n, Z) ⊂ GL(n, Z) consisting of matrices with determinant 1.
The kernels of the homomorphisms ρ0 are called IA (outer) automorphism groups
of Fn which we denote by IAutn and IOutn respectively. Thus we have group exten-
sions
ρ0
1 −→ IAut n −→ Aut Fn −→ GL(n, Z) −→ 1,
ρ0
1 −→ IOut n −→ Out Fn −→ GL(n, Z) −→ 1.
Magnus [53] proved that the group IAutn is finitely generated. On the other hand,
Baumslag–Taylor [4] proved that IAut n is torsion free.
There is a close connection between the mapping class group and the automorphism
groups of free groups. More precisely, we have the following two explicit relations.
One is the realization of the mapping class group Mg,1 of g relative to an embedded
disk D ⊂ g as a subgroup of Aut F2g described as
Mg,1 = {ϕ ∈ Aut F2g ; ϕ(ζ ) = ζ } ⊂ Aut F2g (4.3)
where ζ = [γ1 , γ2 ] . . . [γ2g−1 , γ2g ] ∈ F2g . This follows from Theorem 3.2 because
π1 g0 is isomorphic to F2g . The other is given as follows. The subgroup Z =
Ker(Mg,1 → Mg,∗ ) described in (3.1) is generated by the Dehn twist (see the next
section §5) along a simple closed curve parallel to the boundary of the embedded disk
D 2 ⊂ g and its action on π1 g0 is the conjugation by the element ζ above. Since
Mg,1 /Z is canonically isomorphic to Mg,∗ , we obtain a representation
Mg,∗ −→ Aut F2g /Inn F2g = Out F2g .
It is known that this representation is injective so that we can consider Mg,∗ as a
subgroup of Out F2g
Mg,∗ ⊂ Out F2g .
The comparison of various group theoretical properties between the mapping class
groups Mg,∗ , Mg and automorphism groups of free groups Aut Fn , Out Fn have been
an important subject since the very beginning of combinatorial group theory. Recently,
this tendency is strengthened in a wider framework including geometric viewpoints.
Finally we mention the following result of Laudenbach [50] which shows that, up
to a certain finite group, the (outer) automorphism groups of free groups are naturally
isomorphic to the diffeotopy groups of certain 3-manifolds. Let n S 1 × S 2 denote the
connected sum of n-copies of S 1 × S 2 . Then there are the following exact sequences
1 −→ (Z/2)n −→ Out Fn −→ D(n S 1 × S 2 ) −→ 1,
1 −→ (Z/2)n −→ Aut Fn −→ D(n S 1 × S 2 , rel D 3 ) −→ 1
where D(n S 1 × S 2 , rel D 3 ) denotes the group of path components of those diffeo-
morphisms of n S 1 × S 2 which are the identity on an embedded disk D 3 ⊂ n S 1 × S 2 .
5 Dehn twists
So far we have not mentioned explicit examples of elements of the mapping class
group Mg . Here we describe the most important construction of such elements which
is called the Dehn twist because it was introduced by Dehn.
Suppose that there is given a simple closed curve C on g and also recall that there
is specified an orientation on g . Then we can define an element τC ∈ Mg , which is
called the (right handed) Dehn twist along C, as follows. Let us choose an embedding
i : S 1 × [−1, 1] −→ g
of an annulus into g such that
(i) i(S 1 × {0}) = C and
(ii) i preserves the orientations
where we give S 1 and [−1, 1] the standard orientations and the annulus S 1 × [−1, 1]
the product orientation of them. Then we define a diffeomorphism ϕ0 of the annulus
by
ϕ0 (θ, t) = (θ + f (t), t)
where 0 ≤ θ ≤ 2π and −1 ≤ t ≤ 1 are the coordinates of S 1 and [−1, 1] respectively,

and f : [−1, 1] → R is a C ∞ function such that

0 (−1 ≤ t ≤ − 21
f (t) = 1
2π 2 ≤ t ≤ 1
1 1
and f is strictly increasing on the interval − 2 , 2 . Now recall that for any diffeo-
morphism ψ of a C ∞ manifold M, the closed set
supp(ψ) = {p ∈ M; ψ(p)
= p} ⊂ M
is called the support of ϕ. By the definition of ϕ0 , it is clear that supp(ϕ0 ) is contained
in Int (S 1 × [−1, 1]). Hence we can define a diffeomorphism ϕ ∈ Diff + g by setting
ϕ = ϕ0 on i(S 1 × [−1, 1]) and extend it by the identity over the whole of g . It can
be checked that the isotopy class of ϕ does not depend on the choice of the function f
nor the orientation preserving embedding i (observe here that the opposite embedding
ī : S 1 × [−1, 1] → g defined by ī(θ, t) = i(−θ, −t) does not change the isotopy
class of ϕ). Furthermore it depends only on the isotopy class of the simple closed
curve C. We denote the resulting mapping class by τC ∈ Mg and call it the right
handed Dehn twist along C. This is because a path, which crosses the simple closed
curve C transversely at a point, will be transferred by ϕ to a path which, after getting
near to C, goes around C once to the right direction (with respect to the orientation of
g ) and then goes on as before. The inverse τC−1 is called the left handed Dehn twist
along C. Note here that if we reverse the orientation of g , then the right handed Dehn
twist is changed into the left handed one and vice versa. Also note that the orientation
of the simple closed curve C itself has nothing to do with the definition of the Dehn
twist.
If a simple closed curve C on g bounds a disk, then it is easy to see that the
corresponding Dehn twist is the identity in Mg . A simple closed curve on g is called
essential if it does not bound a disk. We define
S(g ) = {isotopy classes of essential simple closed curves on g }.
In summary, we obtain a mapping
S(g ) [C] −→ τC ∈ Mg .
One simple but important property of the Dehn twists is that the equality
ε(ϕ)
τϕ(C) = ϕ τC ϕ −1 (5.1)
holds for any simple closed curve C and any element ϕ ∈ Diff g , where ε(ϕ) = 1
or −1 if ϕ preserves (or reverses) the orientation of g .
It may appear first that the definition of the Dehn twists is so simple that they will
cover a relatively small part of the mapping class group. However, if one observes
that there are enormously many simple closed curves on g and two (or more) simple
closed curves can meet each other in a very complicated way, one can easily understand
that products of Dehn twists along various simple closed curves can express very
complicated elements in Mg . In fact, Dehn [17] proved that finitely many Dehn twists
generate Mg . Later Lickorish [51] proved that a certain system of 3g − 1 Dehn twists,
which are now called the Lickorish generators, generates Mg . Then Humphries [30]
proved that 2g + 1 members among the Lickorish generators already generate Mg .
He also proved that this number 2g + 1 is the minimum of the number of Dehn twists
which can generate Mg .
As for the presentation of the mapping class group, McCool [59] proved that Mg
is finitely presentable without giving an explicit presentation. Hatcher and Thurston
[28] gave a method of obtaining a finite presentation and it was finally completed by
the work of Wajnryb [85].
6 Mapping class groups acting on the homology of surfaces

and the Torelli groups
The mapping class group Mg acts on the first homology group of g naturally. As-
sume here that g ≥ 1 and we denote simply by H the first integral homology group
H1 (g ; Z) of g . As an abstract group, H is a free abelian group of rank 2g. How-
ever, there exists an important additional structure on H coming from the geometry of
g . More precisely, the intersection numbers of elements of H give rise to a bilinear
mapping
μ : H × H −→ Z.
We denote by u · v (u, v ∈ H ) the intersection number μ(u, v). Then v · u =

−u · v so that μ is skew symmetric. The natural action of Mg on H comes from
orientation preserving diffeomorphisms of the surface g . Hence it clearly preserves
the intersection pairing μ so that we obtain a homomorphism
ρ0 : Mg −→ Aut(H, μ) (6.1)
where Aut(H, μ) denotes the automorphism group of H preserving μ. Namely
Aut(H, μ) = {f ∈ Aut H ; f (u) · f (v) = u · v for any u, v ∈ H }.
Let us study how this condition can be expressed in terms of that of matrices repre-
senting elements of Aut(H, μ). For this, choose a basis x1 , . . . , xg , y1 , . . . , yg of H
such that
xi · yj = δij ,
xi · xj = yi · yj = 0 (i, j = 1, . . . , g).
A basis with this property is called a symplectic basis. It is easy to see that there exist
infinitely many such bases because
x1 , . . . , xg , y1 , . . . , yg is a symplectic basis ⇒
x1 + ny1 , x2 , . . . , xg , y1 , . . . , yg is also a symplectic basis
for any n ∈ Z, for example.
Now if we fix a symplectic basis of H , then the automorphism group Aut H can
be identified as
Aut H = GL(n, Z)
by associating to each element in Aut H the corresponding matrix with respect to the
given symplectic basis. More precisely, first we express any two elements u, v ∈ H
as linear combinations
u = u1 x1 + · · · + ug xg + ug+1 y1 + · · · + u2g yg ,
v = v1 x1 + · · · + vg xg + vg+1 y1 + · · · + v2g yg
with respect to the above symplectic basis and then we identify the two elements u, v
with the following 2g-dimensional column vectors
u = t(u1 , . . . , ug , ug+1 , . . . , u2g ), v = t(v1 , . . . , vg , vg+1 , . . . , v2g )
in R2g . Now set
O E
J = ∈ GL(2g, Z).
−E O
Then we can write
u · v = u1 vg+1 + · · · + ug v2g − ug+1 v2g − · · · − u2g vg = (u, J v)
where (u, J v) denotes the standard Euclidean inner product of two vectors u, J v ∈
R2g . Now a matrix A ∈ GL(2g, Z) preserves the intersection pairing μ if and only if
the condition
Au · Av = u · v for any u, v ∈ H (6.2)
holds. On the other hand we have
Au · Av = (Au, J Av) = (u,t AJ Av),
u · v = (u, J v).
It follows that A satisfies the condition (6.2) if and only if
t
AJ A = J.
Based on the above consideration, we define a subgroup
Sp(2g, Z) = {A ∈ GL(2g, Z); tAJ A = J }
of GL(2g, Z). This group is a discrete subgroup of the symplectic group Sp(2g, R)
consisting of unimodular symplectic matrices. Sometimes the group Sp(2g, Z) is
called the Siegel modular group because it plays a fundamental role in the theory of
Siegel modular forms.
In conclusion, if we fix a symplectic basis of H , then we have an isomorphism
Aut (H, μ) ∼
= Sp(2g, Z)
and (6.1) induces a homomorphism
ρ0 : Mg −→ Sp(2g, Z).
It is easy to see that Sp(2, Z) = SL(2, Z) so that ρ0 is an isomorphism for the case
g = 1. In the cases where g ≥ 2, it was classically known, going back to a work
of Burkhardt at the end of the 19 th century and later works of Dehn and Nielsen,
that ρ0 is surjective (cf. [55]). It was recognized that ρ0 has a non-trivial kernel
which is a normal subgroup of Mg . This group was named the Torelli group after
an Italian mathematician and was known for some time among complex analysts and
algebraic geometers. However it was relatively recently that the Torelli group called
the attention of topologists. Probably Birman’s paper [6] published in 1971 is the
earliest work on this group by topologists. Then Johnson began a systematic study of
this group in the late 1970s and obtained foundational results concerning the structure
of the Torelli group within several years. We refer the readers to his survey paper [37]
as well as [36], [38] [39], [40]. Following his notation, the Torelli group is usually
denoted by g . Thus we have a group extension
ρ0
1 −→ g −→ Mg −→ Sp(2g, Z) −→ 1.
It is a classical result, going back to Grothendieck and Serre, that
the Torelli group g is torsion free.
This can be shown as follows. Suppose that there exists a non-trivial element ϕ ∈ g
which has a finite order, say d > 0. Then by Nielsen [78], there exists a diffeo-
morphism ϕ̃ : g → g such that ϕ̃ d = id and the mapping class of ϕ̃ is the given
one ϕ. Then consider the quotient g /Gϕ̃ of g divided by the action of the cyclic
group Gϕ̃ ∼= Z/d generated by the element ϕ̃. It is easy to see that this quotient
space is homeomorphic to a closed surface of some genus h, because the projection
g → g /Gϕ̃ must be a branched covering along a finite set consisting of fixed points
of ϕ̃ on g . Now there is a general fact on the rational cohomology of the quotient
space X/G divided by a properly discontinuous action of a discrete group G on X,
due originally to Grothendieck, that H ∗ (X/G; Q) ∼ = H ∗ (X; Q)G . If we apply this to
the above simplest case of a finite group action, we obtain isomorphisms
H ∗ (h ; Q) ∼
= H ∗ (g /Gϕ̃ ; Q) ∼
= H ∗ (g ; Q)Gϕ̃ ∼
= H ∗ (g ; Q)
where the last isomorphism comes from the assumption that ϕ̃ acts trivially on the
homology (and hence cohomology) of g . We can now conclude that g = h which
is a contradiction because the condition g ≥ 2 should imply that h < g. Observe
here that the genus 1 surface, namely the torus T 2 admits a free Z/d action and the
quotient space is again diffeomorphic to T 2 .
Now one of the foundational results of Johnson mentioned above is that g is
finitely generated for any g ≥ 3. To prove this, he introduced the following two types
of elements of Mg . One is the BP-map (BP for bounding pair) defined as follows.
Suppose that there are given two disjoint simple closed curves C and D on g which
satisfy the condition that if we cut g along C and D, then the resulting surface is
disconnected. In other words, the disjoint union C ∪ D bounds a subsurface of g .
We say that C and D are a bounding pair. In this case, we call the element
τC τD−1 ∈ Mg
the BP-map corresponding to the above bounding pair. The other type is the BSCC-
map (BSCC for bounding simple closed curve) defined as follows. Suppose that there
is given a simple closed curve C on g such that if we cut g along C, then the
resulting surface is disconnected. In other words, the simple closed curve C bounds a
subsurface of g . We say that C is a bounding simple closed curve. In this case, we
call the element
τC ∈ Mg
the BSCC-map corresponding to the bounding simple closed curve C.
In fact, the following important fact holds:
BP-map, BSCC-map ∈ g . (6.3)
To see this, let us study how a Dehn twist τC along a simple closed curve C acts on
H1 (g ; Z). Let u ∈ H1 (g ; Z) be a homology class and choose an oriented curve
E on g which represents u. We can assume that E intersects C transversely at
finitely many points. Let us choose an orientation on C and let v ∈ H1 (g ; Z) be the
homology class represented by C with this orientation. Locally C divides the regular
neighborhood of C into two parts. If we identify a closed regular neighborhood of
the oriented C with S 1 × [−1, 1] according to the given orientation on the surface
in such a way that the oriented C is identified with the oriented S 1 × {0}, then we
can distinguish the above two pieces by calling them the negative and positive sides
respectively. Now we count the number of the intersection points C ∩ E algebraically
by giving +1 if the oriented curve E intersects C from negative to positive direction
and −1 if E intersects C from positive to negative direction. Let m be the totality of
these ±1 numbers. Then we have
τC (u) = u + mv.
Observe here that, if we reverse the orientation of C, then both m and v change signs
so that the above formula remains unchanged. Now we can check the above fact (6.3)
as follows. First let C, D be a BP-pair. Then we can give orientations on them so
that the resulting homology classes are the same. On the other hand, in the above
computation, we have τC (u) = τD (u) for any u so that τC τD−1 acts on the homology
trivially. Next, if C is a BSCC, then clearly the corresponding homology class is

trivial, whence the claim follows.
The genus of a BP-map (or BSCC-map) is defined as follows. If we cut g along a
BP-pair C, D (or a BSCC C), then we obtain two surfaces. The genus of the relevant
map is defined to be the smaller genus of these two surfaces.
Theorem 6.1 (Johnson [38]). The Torelli group g is finitely generated for any g ≥ 3.
The method of proving this theorem was roughly as follows. Johnson constructed
a certain finite set of BP-maps of all genera between 1 and g − 2 and showed that the
subgroup of g generated by them is a normal subgroup. Since he had already proved
in [35] that g is normally generated by just one BP-map of genus 1, the proof was
completed.
Johnson considered also the subgroup
Kg = the subgroup of Mg generated by all the BSCC-maps
of the mapping class group. Since any BSCC-map is contained in g , Kg is a subgroup
of g . Also it is easy to see that property (5.1) implies any conjugate element of a
BSCC-map is again a BSCC-map. It follows that Kg is a normal subgroup of Mg
(and g ). More strongly, it is known that Kg is a characteristic subgroup of Mg (and
g ). This follows from a result of Ivanov [31] (see also [57]) that any automorphism
of Mg is induced by an inner automorphism of D(g ) for any g ≥ 3 and a similar
result for the case of the Torelli group g due to Farb and Ivanov [20].
It can be shown that Kg coincides with g for g = 2. However Johnson [39]
proved that the quotient g /Kg is an infinite group for any g ≥ 3. More precisely,
choose a symplectic basis x1 , . . . , xg , y1 , . . . , yg of H = H1 (g ; Z) as before. The
element
ω0 = x1 ∧ y1 + · · · + xg ∧ yg ∈ 2 H
is called the symplectic class. It is known that this element is well defined independent
of the choice of symplectic bases. It is easy to see that the mapping
H u −→ u ∧ ω0 ∈ 3 H
is injective. Hence H can be considered as a submodule of 3 H so that we can consider
the quotient module 3 H /H . Now Johnson [36] constructed a homomorphism
τ : g −→ 3 H /H
and showed that it is surjective and vanishes on the subgroup Kg . This homomorphism
is called now the Johnson homomorphism (see §7 for more details). Later he proved
in [39] that Ker τ is precisely the subgroup Kg . Thus we have an extension
τ
1 −→ Kg −→ g −→ 3 H /H −→ 1.
Because of these basic works, sometimes the group Kg is called the Johnson subgroup
or Johnson kernel.
To conclude this section, we would like to make a list which indicates the known
results concerning finite generation as well as finite presentability of groups such as
the automorphism group of Fn , the IA automorphism group of Fn , the mapping class
group, the Torelli group and the Johnson subgroup Kg (the groups Out Fn and IOut n
are not included in the list because the results for these groups are the same as Aut Fn
and IAutn respectively). We refer the readers to the cited original papers and also to
the survey papers [84], [10], [34], [37] for details.
Groups Generators Relations
Aut Fn finitely generated finitely presented

Nielsen [76] Nielsen [76]
McCool [58] McCool [58]
Aut+ Fn Gersten [21] Gersten [21]
IAut n finitely generated unknown (n ≥ 4)
Magnus [53] not finitely presentable (n = 3)
Krstić–McCool [48]
Mg finitely generated finitely presentable
Dehn [17] McCool [59]
Hatcher–Thurston [28]
Lickorish [51] finitely presented
Humphries [30] Wajnryb [85]
Gervais [22]
g finitely generated (g ≥ 3) unknown (g ≥ 3)
Johnson [38] infinitesimal finite presentation
Hain [23]
infinitely generated (g = 2) free group (g = 2)
Mess [62] Mess [62]
Kg infinitely generated unknown (g ≥ 3)
Biss–Farb [11] (g ≥ 3)
Problem 1. Complete the above list by filling in the “unknown” blanks.

7 Johnson homomorphisms
In this section, we define so called Johnson homomorphisms which give homomor-
phisms defined on a certain series of subgroups of the mapping class group into certain
abelian groups.
In order to do so, we first describe a method of investigating the structure of a given
abstract group Γ by approximating it by a series of nilpotent groups. This method is
due originally to Malcev [56]. The first approximation is the abelianization Γ ab of
Γ . This can be algebraically expressed as follows. Let Γ1 denote the commutator
subgroup of Γ . Namely, it is the subgroup of Γ generated by the commutators
[γ1 , γ2 ] = γ1 γ2 γ1−1 γ2−1 (γ1 , γ2 ∈ Γ ). It is easy to see that Γ1 is a normal subgroup
of Γ and, as is well known, we have a canonical isomorphism
Γ ab ∼
= Γ /Γ1 .
Next we consider the second commutator subgroup Γ2 which is defined to be the
subgroup of Γ generated by the two-fold commutators
[γ1 , [γ2 , γ3 ]] = γ1 [γ2 , γ3 ]γ1−1 [γ2 , γ3 ]−1 (γ1 , γ2 , γ3 ∈ Γ ).
Then it can be checked that Γ2 is a normal subgroup of Γ and Γ /Γ2 is a two-step
nilpotent group. More precisely, the quotient group Γ1 /Γ2 is an abelian group and
Γ /Γ2 can be described by the following group extension
0 −→ Γ1 /Γ2 −→ Γ /Γ2 −→ Γ /Γ1 = Γ ab −→ 1 (7.1)
which is a central extension of Γ /Γ1 = Γ ab by Γ1 /Γ2 . We can continue this procedure
to obtain a series of nilpotent groups Nk (k = 1, 2, . . . ) which approximate Γ as
follows, where N1 = Γ ab and N2 = Γ /Γ2 . We set Γ0 = Γ and for each k = 1, 2, . . . ,
we inductively define
Γk = [Γ, Γk−1 ]
= the subgroup of Γ generated by k-fold commutators.
It can be checked that Γk is a normal subgroup of Γ and the series {Γk }k of normal
subgroups of Γ is called the lower central series of Γ . It can be easily checked that
these subgroups of Γ are all characteristic subgroups meaning that any automorphism
f ∈ Aut Γ of Γ preserves them. Now we set Nk = Γ /Γk and call this group the k-th
nilpotent quotient of Γ . In fact, the quotient Ck = Γk−1 /Γk is easily seen to be an
abelian group and furthermore we have a central extension
0 −→ Ck −→ Nk −→ Nk−1 −→ 1 (k = 2, 3, . . . ). (7.2)
Hence Nk is a k-step nilpotent group and we obtain an inverse system
· · · −→ Nk −→ Nk−1 −→ · · · −→ N3 −→ N2 −→ N1 = Γ ab
of nilpotent groups to which there is a homomorphism from the given group Γ .
Next we ignore any torsion in the above argument and consider everything over
Q. In some sense, this procedure can be understood as taking tensor products with Q.
The first step is straightforward. Namely we just take the usual tensor product of Γ ab
with Q
N1 ⊗ Q = Γ ab ⊗ Q
which is a vector space over Q. For the second step, recall that any central extension
of a group Γ by an abelian group C is classified by its extension class which is a
second cohomology class defined in H 2 (Γ ; C) (see [12]). If we apply this to the
central extension (7.1), we obtain a certain element
χ2 (Γ ) ∈ H 2 (Γ ab ; Γ1 /Γ2 ).
Application of the natural homomorphisms Γ ab → Γ ab ⊗ Q and Γ1 /Γ2 → Γ1 /Γ2 ⊗ Q
to the above element gives
χ2Q (Γ ) ∈ H 2 (Γ ab ⊗ Q; Γ1 /Γ2 ⊗ Q).
This yields a central extension
0 −→ Γ1 /Γ2 ⊗ Q −→ N2 ⊗ Q −→ Γ ab ⊗ Q −→ 1 (7.3)
by which the group N2 ⊗ Q is defined. We can inductively continue this procedure
and we eventually obtain a series of central extensions
0 −→ Ck ⊗ Q −→ Nk ⊗ Q −→ Nk−1 ⊗ Q −→ 1 (k = 2, 3, . . . ). (7.4)
In this way, a series of nilpotent groups {Nk ⊗ Q}k is defined. The inverse system
· · · −→ Nk ⊗ Q −→ Nk−1 ⊗ Q −→ · · · −→ N3 ⊗ Q −→ N2 ⊗ Q −→ N1 ⊗ Q
of nilpotent groups is called the Malcev completion of the given group Γ .
Example 7.1. One of the most important examples of the Malcev completions which
appears in the theory of free groups as well as the mapping class group is that of free
groups. Here we briefly describe it. Let Fn denote a free group of rank n and we
denote the abelianization H1 (Fn ; Z) of Fn simply by H which is a free abelian group
of rank n. We consider the free graded Lie algebra generated by the elements of H
which we denote by
∞
L= Lk
k=1
as follows. The degree 1 part L1 is defined to be H itself. Then we consider the bracket
[u, v] ∈ L2 of two elements u, v ∈ H . The skew commutativity [v, u] = −[u, v] of
the Lie algebra implies that L2 = 2 H . Next we consider the bracket
[ , ] : L1 ⊗ L2 = H ⊗ 2 H −→ L3 .
The Jacobi identity of the Lie algebra forces that the submodule 3 H ⊂ H ⊗ 2 H
(defined by the correspondence u ∧ v ∧ w → u ⊗ [v, w] + v ⊗ [w, u] + w ⊗ [u, v])
must vanish under the above map. There are no other constraints so that L3 =
H ⊗ 2 H /3 H . Going further, the complexity of enumerating all the relations
imposed by the structure of the Lie algebra increases. However we can avoid this
difficulty by embedding L into the tensor algebra
∞

T ∗ (H ) = H ⊗k .
k=1
The degree 1 parts of L and T ∗ (H ) are the same, namely both are H . The second
term can be embedded as
L2 = 2 H [u, v] = u ∧ v −→ u ⊗ v − v ⊗ u ∈ H ⊗2 .
As for the third term, consider the linear mapping
H ⊗ 2 H u ⊗ [v, w]
−→ u ⊗ v ⊗ w − u ⊗ w ⊗ v − v ⊗ w ⊗ u + w ⊗ v ⊗ u ∈ H ⊗3 .
It is easy to check that the kernel of this map is precisely 3 H so that we obtain an
embedding
L3 = H ⊗ 2 H /3 H u ⊗ [v, w]
−→ u ⊗ v ⊗ w − u ⊗ w ⊗ v − v ⊗ w ⊗ u + w ⊗ v ⊗ u ∈ H ⊗3 .
Then we can inductively define Lk as the image of the linear mapping
H ⊗k ⊃ H ⊗ Lk−1 u ⊗ ξ −→ u ⊗ ξ − ξ ⊗ u ∈ H ⊗k .
Thus L = ⊕k Lk is realized as a submodule of T ∗ (H ). The elements in L are called
Lie elements of T ∗ (H ). Now it is a classical result that L is isomorphic to the graded
module associated to the lower central series of the free group Fn . Namely there exists
a canonical isomorphism
(Fn )k−1 /(Fn )k ∼ = Lk
where (Fn )k denotes the k-th term in the lower central series of Fn (the first one
Fn /(Fn )1 ∼
= H gives the abelianization). Thus we have a series of central extensions
0 −→ Lk −→ Nk (Fn ) −→ Nk−1 (Fn ) −→ 1 (k = 2, 3, . . . )
where Nk (Fn ) denotes the k-th nilpotent quotient of Fn . See [55] for details.
Now we define the Johnson homomorphisms. First we begin with the case of
automorphism groups of free groups. This case was considered first by Andreadakis
[1] before the works of Johnson.
We can define a series Aut Fn (k) (k = 1, 2, . . . ) of subgroups of Aut Fn as follows.
Any member (Fn )k in the lower central series of Fn is a characteristic subgroup in the
sense that it is preserved by any automorphism ϕ ∈ Aut Fn . Hence we obtain a series
of representations
pk : Aut Fn −→ Aut Nk (Fn ) (k = 1, 2, . . . ).
The first one p1 is nothing but the natural homomorphism
Aut Fn −→ Aut N1 (Fn ) = GL(n, Z).
Now we set
Aut Fn (k) = Ker pk = {ϕ ∈ Aut Fn ; ϕ acts on Nk (Fn ) trivially}.
The first one Aut Fn (1) is nothing but the subgroup IAut n .
Now let ϕ ∈ Aut Fn (k) be any element. Then for each element γ ∈ Fn ,
ϕ(γ )γ −1 ∈ (Fn )k
because, by the assumption, ϕ acts on Nk (Fn ) = Fn /(Fn )k trivially. Consider the
image of ϕ(γ )γ −1 in Lk+1 = (Fn )k /(Fn )k+1 which we denote by [ϕ(γ )γ −1 ]. This
procedure defines a mapping
Fn γ −→ [ϕ(γ )γ −1 ] ∈ Lk+1 .
It can be shown that the above mapping factors through the abelianization of Fn so
that we obtain a mapping
H = (Fn )ab −→ Lk+1 .
Hence we can define a mapping
τk : Aut Fn (k) −→ Hom(H, Lk+1 ) (7.5)
by setting
τk (ϕ)([γ ]) = [ϕ(γ )γ −1 ] (γ ∈ Fn ).
Finally it can be checked that the above mapping (7.5) is in fact a homomorphism and
this is called the k-th Johnson homomorphism for the automorphism groups of free
groups.
By the definition of the homomorphism τk , it is easy to see that
Ker τk = Aut Fn (k + 1).
Therefore we have an injection
Aut Fn (k)/Aut Fn (k + 1) ⊂ Hom(H, Lk+1 ).
If we make the direct sum over k, we obtain an injection
∞
∞

Aut Fn (k)/Aut Fn (k + 1) ⊂ Hom(H, Lk+1 ).
k=1 k=1
It is known that both of the above graded modules have the structure of graded Lie
algebras over Z and it is a very important problem to identify the left hand side as an
explicit Lie subalgebra of the right hand side.
Problem 2. Determine the graded module

∞

Aut Fn (k)/Aut Fn (k + 1)
k=1
associated to the filtration {Aut Fn (k)}k of the group Aut Fn as a Lie subalgebra of the
graded Lie algebra
∞

Hom(H, Lk+1 ).
k=1
Next we consider the case of the mapping class group. Here, for simplicity, we
consider only the case of a compact surface g0 = g \ IntD 2 with one boundary
component. Then Γ = π1 g0 is a free group of rank 2g. As before, we denote
H1 (g0 ; Z) simply by H which is a free abelian group of rank 2g. By Theorem (3.2),
the mapping class group Mg,1 of g0 is a subgroup of Aut F2g . Hence we can define
a filtration {Mg,1 (k)}k of Mg,1 by simply restricting that of Aut F2g to the subgroup
Mg,1 . The first term Mg,1 (1) in this filtration is nothing but the Torelli group g,1 . It
turns out that, in the case of the mapping class group, there are important additional
structures which do not exist in the case of Aut F2g . First notice that the Poincaré
duality theorem for the (co)homology of the surface implies that there is a natural
isomorphism
H ∗ = Hom(H, Z) = H 1 ( 0 ; Z) = ∼ H.
g
It follows that we can replace the target Hom(H, Lk+1 ) of the k-th Johnson homo-
morphism (7.5) by H ⊗ Lk+1 . Johnson [36] proved that the image of the first Johnson
homomorphism
τ1 : g,1 −→ H ⊗ 2 H
is precisely the submodule
3 H ⊂ H ⊗ 2 H.
Generalizing this fact, it was proved in [66] that the target of τk can be narrowed, for
any k, as follows. We define a submodule Hk of H ⊗ Lk+1 by setting
[ ,]
Hk = Ker H ⊗ Lk+1 −−−→ Lk+2
where [ , ] denotes the bracket operation in the graded Lie algebra L. It can be checked
that the graded submodule
∞
∞

Hk ⊂ H ⊗ Lk+1
k=1 k=1
is a graded Lie subalgebra.

Problem 3. Determine the graded module

∞

Mg,1 (k)/Mg,1 (k + 1)
k=1
associated to the filtration {Mg,1 (k)}k of the mapping class group Mg,1 as a Lie
subalgebra of the graded Lie algebra
∞

Hk .
k=1
A similar problem for the usual mapping class group Mg can be formulated by
making use of the result of Labute [49].
8 Teichmüller space and Outer Space

There are two important spaces on which the mapping class group Mg and the outer
automorphism group Out Fn of a free group act canonically. One is the classical
Teichmüller space, introduced by Teichmüller in the 1930s, and the other is the Outer
Space defined in the 1980s by Culler and Vogtmann [16]. Here we briefly describe
the definitions of them which can be given in parallel with each other.
The Teichmüller space of g , denoted by Tg , is defined to be the space of all the
orientation preserving diffeomorphisms
f : g −→ M
from g to compact Riemann surfaces M of genus g divided by a certain equivalence
relation. More precisely
Tg = {f : g → M; M is a Riemann surface of genus g}/ ∼
where two orientation preserving diffeomorphisms
f : g −→ M, f : g −→ M
are equivalent if there exists an isomorphism h : M → M of Riemann surfaces (namely
a biholomorphism) such that the following diagram
f
g −−−−→ M

⏐

⏐

h
f
g −−−−→ M
is homotopy commutative. Let [f : g → M] ∈ Tg denote the equivalence class
represented by f : g → M. If we pull back the complex structure on M by the
diffeomorphism f , we obtain a complex structure on g whose induced orientation

coincides with the given orientation on g . Since homotopic diffeomorphisms of
g are mutually isotopic, the isotopy class of the pull back complex structure on g
depends only on the element [f : g → M]. Thus we can also write
Tg = {isotopy classes of complex structure on g }.
In the cases where g ≥ 2, a complex structure on g is the same as a hyperbolic
structure (that is a Riemannian structure with constant negative curvature −1) by the
classical uniformization theorem. Therefore we have yet one more description:
Tg = {isotopy classes of hyperbolic structure on g } (g ≥ 2).
The space of all the complex (or hyperbolic) structures on g has a natural C ∞
topology and it induces a topology on Tg . It is a classical result of Teichmüller that
Tg is homeomorphic to R6g−6 for any g ≥ 2. The mapping class group Mg acts on
Tg from the right by
Tg × Mg ([f : g → M], ϕ) −→ [f ϕ̃ : g → M] ∈ Tg
where ϕ̃ ∈ Diff + g is a lift of ϕ ∈ Mg . This action is known to be properly
discontinuous. The quotient space
M g = Tg /Mg
is called the moduli space of Riemann surfaces of genus g which consists of all the
isomorphism classes of genus g Riemann surfaces.
The Outer Space Xn , which is an analogue of the Teichmüller space where we
replace Mg with Out Fn , was defined by Culler and Vogtmann [16] as follows. A
metric graph is a graph (one dimensional finite complex) such that (i) the valencies
at vertices are all ≥ 3 and (ii) there is given a length on every edge such that the sum
is equal to 1. Let Rn denote the wedge of labeled n circles Si1 (i = 1, . . . , n) so
that there is given an isomorphism π1 Rn ∼ = Fn . Then the Outer Space (of rank n) is
defined as the set of homotopy equivalences f : Rn → from Rn to metric graphs
divided by a certain equivalence relation. More precisely
Xn = {f : Rn → ; is a metric graph with π1 ∼
= Fn }/ ∼
where two homotopy equivalences
f : Rn −→ , f : Rn −→
are equivalent if there exists an isometry h : → of metric graphs such that the
following diagram
f
Rn −−−−→

⏐

⏐

h
f
Rn −−−−→
is homotopy commutative. Let [f : Rn → ] ∈ Xn denote the equivalence class

represented by f : Rn → . The group Out Fn acts on Xn from the right by
Xn × Out Fn ([f : Rn → ], ϕ) −→ [f ϕ̃ : Rn → ] ∈ Xn
where ϕ̃ : Rn → Rn is a homotopy equivalence which represents ϕ ∈ Out Fn . There
is a natural topology on Xn and Culler and Vogtmann proved that Xn is contractible
and the above action is properly discontinuous. The quotient space
Gn = Xn /Out Fn
is called the moduli space of metric graphs of rank n which consists of all the isometry
classes of metric graphs of rank n. We refer to the survey paper [84] by Vogtmann and
also Bestvina [5] for recent results concerning the Outer Space as well as Out Fn .
In general, it is a very important problem to determine the (co)homology groups
of the moduli spaces associated to various geometrical objects. In the above, we have
the moduli spaces of Riemann surfaces and the moduli space of metric graphs. There
have been obtained many results concerning the cohomology of these moduli spaces
(we refer to the survey papers [26], [45], [83], [84], [71] as well as original papers
[25], [73], [64], [65], [63], [52], [29], [15]). However the cohomological structures
of them are far from being very well understood.
Problem 4. Study the (co)homology groups of the moduli space M g of Riemann

surfaces and the moduli space Gn of metric graphs.
9 Symplectomorphism groups of surfaces

As in §1, let Diff + g denote the orientation preserving diffeomorphism group of
g . Let us choose an area form ω on g . Then, by the dimension reason, it can be
considered also as a symplectic form on g . We denote by
Symp g = {ϕ ∈ Diff + g ; ϕ ∗ ω = ω}
the subgroup of Diff + g consisting of those diffeomorphisms which preserve the
form ω. We call it the symplectomorphism group of the symplectic manifold (g , ω)
or the orientation and area preserving diffeomorphism group of g with respect to the
area form ω.
As was already mentioned in §1, in general, the volume preserving diffeomorphism
group Diff υ M of a C ∞ manifold M with respect to a given volume form υ and also
the symplectomorphism group Symp(M, ω) of a symplectic manifold (M, ω) are
both very important objects of geometry and topology. Recently there has been rapid
progress in a topological study of symplectic manifolds, under the name of symplectic
topology (see [61] for foundations and generalities of this theory).
The case of surfaces is the simplest one. However it is at the same time very
important because the symplectic and the volume preserving contexts are the same in
this case so that we can expect very rich structures here. In the following, we would
like to describe one feature of these structures, namely the one which is induced by a
basic concept in symplectic topology called the flux homomorphism.
If we apply Moser’s theorem in [72], mentioned in §2, to Symp g , we can conclude
that the inclusion
Symp g ⊂ Diff + g
is a homotopy equivalence. It follows that the symplectic mapping class group
SD(g , ω) of the symplectic manifold (g , ω) can be canonically identified with
the usual mapping class group Mg and we obtain the following exact sequence
1 −→ Symp0 g −→ Symp g −→ Mg −→ 1.
In particular, the natural homomorphism Symp g → Mg is surjective. Let us see
this fact somewhat more explicitly. One form of Moser’s theorem cited above can be
stated as follows. Let M be a closed oriented C ∞ manifold and let υ, υ be any two
volume forms on M. Then there exists a diffeomorphism ϕ of M, which can be chosen
to be isotopic to the identity, such that υ = cϕ ∗ υ where c is a constant defined by

υ =c υ.
M M
Now let ϕ ∈ Mg be any element and let ϕ̃ ∈ Diff + g be its lift. Consider the
form ϕ̃ ∗ ω which is another area form on g . Hence by the above theorem of Moser,
there exists an element ψ ∈ Diff + g , which is isotopic to the identity, such that
ϕ̃ ∗ ω = ψ ∗ ω. If we set ϕ̃ = ϕ̃ψ −1 , then (ϕ̃ )∗ ω = ω so that ϕ̃ belongs to Symp g .
On the other hand, since ψ is isotopic to the identity, the projection of ϕ̃ to Mg is
the same as that of ϕ̃ which is the given element ϕ. We can now conclude that the
mapping Symp g → Mg is surjective as required. It might be amusing to observe
here that the Dehn twist along a simple closed curve C on g , defined in §5, preserves
any area form on g whose restriction to a cylindrical neighborhood of C is equal to
the 2-form dθ ∧ dt.
Now we describe the flux homomorphism briefly (see [61] for details). It is defined
for a general symplectic manifold (M, ω). Let Symp0 (M, ω) denote the identity com-
ponent of Symp (M, ω) as before. Then the flux homomorphism is a homomorphism
Flux :
Symp0 (M, ω) −→ H 1 (M; R) (9.1)
from the universal covering group of Symp0 (M, ω) to the first real cohomology group
of M defined as follows. For each element ϕ ∈ Symp0 (M, ω), let ϕt ∈ Symp0 (M, ω)
be an isotopy such that ϕ0 = id and ϕ1 = ϕ. Then
1
Flux({ϕt }) = iϕ˙t ω dt
0
where ϕ̇t denotes the vector field associated to ϕt , which is considered as a one-
parameter family of transformations of M, and i denotes the interior product. It can
be checked that the above value depends only on the homotopy class of the curve {ϕt }
in Symp0 (M, ω) with fixed endpoints. Hence we have the induced map described in
(9.1). Furthermore it can be checked that Flux is a homomorphism and also that it is
surjective. We have an exact sequence
1 −→ π1 Symp0 (M, ω) −→
Symp0 (M, ω) −→ Symp0 (M, ω) −→ 1
and the subgroup
ω = Flux(π1 Symp0 (M, ω)) ⊂ H 1 (M; R)
is called the flux group. Then (9.1) induces the following homomorphism which is
also called the flux homomorphism
Flux : Symp0 (M, ω) −→ H 1 (M; R)/ ω . (9.2)
Very recently, Ono [79] proved a long standing conjecture that the flux group is a
discrete subgroup of H 1 (M; R) for any compact symplectic manifold M.
Now in our case of surfaces, by Moser’s theorem Symp0 g is homotopy equiv-
alent to Diff + g which in turn is known by Earle and Eells [18] to be homotopy
equivalent to T 2 for the case g = 1 and contractible for any g ≥ 2. Hence we obtain
homomorphisms
Flux : Symp0 T 2 −→ H 1 (T 2 ; R)/H 1 (T 2 ; c Z),
Flux : Symp0 g −→ H 1 (g ; R) (g ≥ 2)
where c denotes the total area of T 2 with respect to ω.
10 Extensions of the Johnson homomorphism and the flux

homomorphism
Assume that a group G acts on a module M by automorphisms. In other words,
suppose that a homomorphism
G −→ Aut M
is given. Then we can give the direct product M × G a natural structure of a group by
setting
(m, g)(n, h) = (m + g(n), gh) (m, n ∈ M, g, h ∈ G).
The resulting group is denoted by M G and called the semi-direct product of M and
G or split extension of G by M.
We mention the following three results which have a similar formal nature to
each other. The first one is given in [67], where it was proved that the first Johnson
homomorphism τ1 : g,1 −→ 3 H can be extended to a homomorphism
ρ1 : Mg,1 −→ 3 HQ Sp(2g, Z)
where HQ = H ⊗ Q. Similar results hold for the other types of the mapping class
groups Mg and Mg,∗ . The second one is due to Kawazumi [41] who proved, among
other things, that the first Johnson homomorphism τ1 : IAut n −→ H ∗ ⊗ 2 H can be
extended to a homomorphism
ρ1 : Aut Fn −→ VQ GL(n, Z)
where V = H ∗ ⊗ 2 H and VQ = V ⊗ Q. A similar result holds for Out Fn .
The third one is given in [46] where it was proved that the flux homomorphism
Flux : Symp0 g → H 1 (g ; R) can be extended to a homomorphism
: Symp g −→ H 1 (g ; R) Sp(2g, Z).
Flux
Thus we have the following three commutative diagrams.
1 −−−−→ g,1 −−−−→ Mg,1 −−−−→ Sp(2g, Z) −−−−→ 1
⏐ ⏐
⏐τ ⏐ρ
1 1
1 −−−−→ 3 H −−−−→ 3 HQ Sp(2g, Z) −−−−→ Sp(2g, Z) −−−−→ 1
1 −−−−→ IAut n −−−−→ Aut Fn −−−−→ GL(n, Z) −−−−→ 1

⏐ ⏐
⏐τ ⏐ρ
1 1
1 −−−−→ V −−−−→ VQ GL(n, Z) −−−−→ GL(n, Z) −−−−→ 1
1 −−−−→ Symp0 g −−−−→ Symp g −−−−→ Mg −−−−→ 1

⏐ ⏐
⏐ ⏐
Flux Flux
1 −−−−→ HR −−−−→ HR Mg −−−−→ Mg −−−−→ 1

where HR = H 1 (g ; R).
The first two diagrams can be extended further by considering higher nilpotent
quotients of the Torelli group g,1 and the group IAutn . However, as for the last one,
there is no such extension because the kernel of the flux homomorphism, which is
denoted by Ham g and called the Hamiltonian symplectomorphism group, is known
to be perfect by Thurston [82] (see also Banyaga [3] for the generalization of this
fact to general symplectic manifolds). There are several results which make use of
the above three commutative diagrams, see [68], [42], [43], [70], [41], [46], [47] and
references in them. However it seems likely that there should exist further interesting
facts to be uncovered along these lines.
Problem 5. Give further applications as well as generalizations of the above three

commutative diagrams.
References
[1] S. Andreadakis, On the automorphisms of free groups and free nilpotent groups. Proc.
London Math. Soc. 15 (1965), 239–268. 373
[2] R. Baer, Isotopien auf Kurven von orientierbaren, geschlossenen Flächen. J. Reine Angew.
Math. 159 (1928), 101–116. 360
[3] A. Banyaga, Sur la structure du groupe des difféomorphismes qui prèservent une forme
symplectique. Comment. Math. Helv. 53 (1978), 174–227. 357, 381
[4] G. Baumslag and T. Taylor, The centre of groups with one defining relator. Math. Ann.
175 (1968), 315–319. 362
[5] M. Bestvina, The topology of Out(Fn ). In Proceedings of the International Congress of
Mathematicians (Beijing, 2002), Vol. II, Higher Ed. Press, Beijing 2002, 373–384. 378
[6] J. Birman, On Siegel’s modular group. Math. Ann. 191 (1971), 59–68. 367
[7] J. Birman, Braids, Links and Mapping Class Groups. Ann. of Math. Stud. 82, Princeton
University Press, Princeton, N.J.; University of Tokyo Press, Tokyo 1974. 353
[8] J. Birman, Mapping class groups of surfaces: a survey. In Discontinuous Groups and
Riemann Surfaces (L. Greenberg, ed.), Ann. of Math. Stud. 79, Princeton University
Press, Princeton, N.J., 1974, 57–71. 353
[9] J. Birman, The algebraic structure of surface mapping class groups. In Discrete Groups
and Automorphic Functions (W. J. Harvey, ed.), Academic Press, London 1977, 163–198.
353
[10] J. Birman, Mapping class groups of surfaces. In Braids (J. Birman and A. Libgober, eds.),
Contemp. Math. 78, Amer. Math. Soc., Providence, RI, 1988, 13–43. 353, 370
[11] D. Biss and B. Farb, Kg is not finitely generated. Invent. Math. 163 (2006), 213–226. 370
[12] K. S. Brown, Cohomology of Groups. Grad. Texts in Math. 87, Springer-Verlag, Berlin
1982. 358, 372
[13] J. Cerf, Sur les difféomorphisms de la sphère de dimension trois (4 = 0). Lecture Notes
in Math. 53, Springer-Verlag, Berlin 1968. 355
[14] J. Cerf, La stratification naturelle des espaces de fonctions différentiables réelles et le
théorème de la pseudo-isotopie. Inst. Hautes Études Sci. Publ. Math. 39 (1970), 5–173.
355
[15] J. Conant and K. Vogtmann, Morita classes in the homology of automorphism groups of
free groups. Geom. Topol. 8 (2004), 1471–1499. 378
[16] M. Culler and K. Vogtmann, Moduli of graphs and automorphisms of free groups. Invent.
Math. 84 (1986), 91–119. 376, 377
[17] M. Dehn, Die Gruppe der Abbildungsklassen. Acta Math. 69 (1938), 135–206. 365, 370
[18] C. J. Earle and J. Eells, The diffeomorphism group of a compact Riemann surface. Bull.
Amer. Math. Soc. 73 (1967), 557–559. 380
[19] D. Epstein, Curves on 2-manifolds and isotopies. Acta Math. 115 (1966), 83–107. 360
[20] B. Farb and N. Ivanov, The Torelli geometry and its applications. Math. Res. Lett. 12
(2005), 293–301. 369
[21] S. M. Gersten, A presentation for the special automorphism group of a free group. J. Pure
Appl. Algebra 33 (1984), 269–279. 362, 370
[22] S. Gervais, A finite presentation for the mapping class group of a punctured surface.
Topology 40 (2001), 703–725. 370
[23] R. Hain, Infinitesimal presentations of the Torelli groups. J. Amer. Math. Soc. 10 (1997),
597–651. 370
[24] R. Hain and E. Looijenga, Mapping class groups and moduli spaces of curves. Proc.
Sympos. Pure Math. 62 (2) (1997), 97–142. 354
[25] J. Harer, Stability of the homology of the mapping class group of an orientable surface.
Ann. of Math. 121 (1985), 215–249. 378
[26] J. Harer, The cohomology of the moduli space of curves. Lecture Notes in Math. 1337,
Springer-Verlag, Berlin 1988, 138–221. 354, 378
[27] W. Harvey, Geometric structure of surface mapping class groups. In Homological Group
Theory (C. T. C. Wall, ed.), London Math. Soc. Lecture Note Ser. 36, Cambridge University
Press, Cambridge 1979, 255–269. 353
[28] A. Hatcher and W. Thurston, A presentation for the mapping class group of a closed
orientable surface. Topology 19 (1980), 221–237. 365, 370
[29] A. Hatcher and K. Vogtmann, Rational homology of Aut(Fn ). Math. Res. Lett. 5 (1998),
759–780. 378
[30] S. P. Humphries, Generators for the mapping class group. In Topology of low-dimensional
manifolds, Lecture Notes in Math. 722, Springer-Verlag, Berlin 1979, 44–47. 365, 370
[31] N. Ivanov, Algebraic properties of the Teichmüller modular group. Soviet Math. Dokl. 29
(1984), 288–291. 369
[32] N. Ivanov, Algebraic properties of mapping class groups of surfaces. In Geometric and
algebraic topology, Banach Center Publ. 18, PWN-Polish Scientific Publishers, Warsaw
1986, 15–35. 354
[33] N. Ivanov, Subgroups of Teichmüller Modular Groups. Transl. Math. Monogr. 115, Amer.
Math. Soc., Providence, RI, 1992. 354
[34] N. Ivanov, Mapping class groups. In Handbook of Geometric Topology, North-Holland,
Amsterdam 2002, 523–634. 353, 370
[35] D. Johnson, Homeomorphisms of a surface which act trivially on homology. Proc. Amer.
Math. Soc. 75 (1979), 119–125. 369
[36] D. Johnson, An abelian quotient of the mapping class group g . Math. Ann. 249 (1980),
225–242. 367, 369, 375
[37] D. Johnson, A survey of the Torelli group. Contemp. Math. 20 (1983), 165–179. 354, 367,
370
[38] D. Johnson, The structure of the Torelli group I: A finite set of generators for g . Ann. of
Math. 118 (1983), 423–442. 367, 369, 370
[39] D. Johnson, The structure of the Torelli group II: A characterization of the group generated
by twists on bounding simple closed curves. Topology 24 (1985), 113–126. 367, 369
[40] D. Johnson, The structure of the Torelli group III: The abelianization of g . Topology 24
(1985), 127–144. 367
[41] N. Kawazumi, Cohomological aspects of the Magnus expansions. Preprint, 2006;

arXiv:math.GT/0505497. 381
[42] N. Kawazumi and S. Morita, The primary approximation to the cohomology of the moduli
space of curves and cocycles for the stable characteristic classes. Math. Res. Lett. 3 (1996),
629–641. 381
[43] N. Kawazumi and S. Morita, The primary approximation to the cohomology of the moduli
space of curves and cocycles for the Mumford-Morita-Miller classes. Preprint. 381
[44] M. Kervaire and J. Milnor, Groups of homotopy spheres, I. Ann. of Math. 77 (1963),
504–537. 355
[45] F. Kirwan, Cohomology of moduli spaces. In Proceedings of the International Congress
of Mathematicians (Beijing, 2002), Vol. I, Higher Ed. Press, Beijing 2002, 363–382. 378
[46] D. Kotschick and S. Morita, Signature of foliated surface bundles and the group of sym-
plectomorphisms of surfaces. Topology 44 (2005), 131–149. 381
[47] D. Kotschick and S. Morita, Characteristic classes of foliated surface bundles with area-
preserving total holonomy. J. Differential Geom. 75 (2007), 273–302. 381
[48] S. Krstić and J. McCool, The non-finite presentability of IA(F3 ) and GL2 (Z[t, t −1 ]).
Invent. Math. 129 (1997), 595–606. 370
[49] J.P. Labute, On the descending central series of groups with a single defining relation. J.
Algebra 14 (1970), 16–23. 376
[50] F. Laudenbach, Sur les 2-sphère d’une variété de dimension 3. Ann. of Math. 97 (1973),
57–81. 363
[51] W. Lickorish, A finite set of generators for the homeotopy group of a 2-manifold. Proc.
Cambridge Philos. Soc. 61 (1964), 769–778. 365, 370
[52] I. Madsen and M. Weiss, The stable moduli space of Riemann surfaces: Mumford’s con-
jecture. Ann. of Math. 165 (3) (2007), to appear; arXiv:math.AT/0212321. 378
[53] W. Magnus, Über n-dimensionale Gittertransformationen. Acta Math. 64 (1934), 353–367.
362, 370
[54] W. Magnus, Über Automorphismen von Fundamentalgruppen berandeter Flächen. Math.
Ann. 109 (1934), 617–648. 361
[55] W. Magnus, A. Karrass and D. Solitar, Combinatorial Group Theory. Wiley, New York
1966. 367, 373
[56] A. Malcev, On a class of homogeneous spaces. Izv. Akad. Nauk. SSSR Ser. Mat. 13 (1949),
9–32; English transl. Amer. Math. Soc. Transl. 39 (1962), 775–790. 371
[57] J. McCarthy, Automorphisms of surface mapping class groups. A recent theorem of
N. Ivanov. Invent. Math. 84 (1986), 49–71. 369
[58] J. McCool, A presentation for the automorphism group of a free group of finite rank. J.
London Math. Soc. (2) 8 (1974), 259–266. 362, 370
[59] J. McCool, Some finitely presented subgroups of the automorphism group of a free group.
J. Algebra 35 (1975), 205–213. 365, 370
[60] D. McDuff, A survey of topological properties of groups of symplectomorphisms. In
Topology, Geometry and Quantum Field Theory, Cambridge University Press, Cambridge
2004, 173–193. 357
[61] D. McDuff and D. A. Salamon, Introduction to Symplectic Topology. Second edition,

Oxford Math. Monogr., Oxford University Press, New York 1998. 356, 378, 379
[62] G. Mess, The Torelli groups for genus 2 and 3 surfaces. Topology 31 (1992), 775–790.
370
[63] E. Y. Miller, The homology of the mapping class group. J. Differential Geom. 24 (1986),
1–14. 378
[64] S. Morita, Characteristic classes of surface bundles. Bull. Amer. Math. Soc. 11 (1984),
386–388. 378
[65] S. Morita, Characteristic classes of surface bundles. Invent. Math. 90 (1987), 551–577.
378
[66] S. Morita, Abelian quotients of subgroups of the mapping class group of surfaces. Duke
Math. J. 70 (1993), 699–726. 375
[67] S. Morita, The extension of Johnson’s homomorphism from the Torelli group to the map-
ping class group. Invent. Math. 111 (1993), 197–224. 380
[68] S. Morita, A linear representation of the mapping class group of orientable surfaces and
characteristic classes of surface bundles. In Topology and Teichmüller spaces (Katinkulta,
1995), World Scientific, River Edge, N.J., 1996, 159–186. 381
[69] S. Morita, Structure of the mapping class groups of surfaces: a survey and a prospect.
Geometry and Topology Monographs 2 (1999), Proceedings of the Kirbyfest, 349–406.
354
[70] S. Morita, Structure of the mapping class group and symplectic representation the-
ory. In Essays on geometry and related topics, Vol. 2, Monogr. Enseign. Math. 38,
L’Enseignement Math., Geneva 2001, 577–596. 381
[71] S. Morita, Cohomological structure of the mapping class group and beyond. In Problems
on Mapping Class Groups and Related Topics (Benson Farb, ed.), Proc. Sympos. Pure
Math. 74, Amer. Math. Soc., Providence, RI, 2006, 329–354. 354, 378
[72] J. Moser, On the volume forms. Trans. Amer. Math. Soc. 120 (1965), 286–294. 356, 379
[73] D. Mumford, Towards an enumerative geometry of the moduli space of curves. In Arith-
metic and Geometry, Progr. Math. 36, (1983), 271–328. 378
[74] J. Nielsen, Die Isomorphismen der allgemeinen unendlichen Gruppe mit zwei Erzeugen-
den. Math. Ann. 78 (1917), 385–397. 362
[75] J. Nielsen, Die Gruppe der dreidimensionalen Gittertransformationen. Danske Vid. Selsk.
Math. Fys. Medd. 12 (1924), 1–29. 362
[76] J. Nielsen, Die Isomorphismengruppe der freien Gruppen. Math. Ann. 91 (1924), 169–209.
362, 370
[77] J. Nielsen, Untersuchungen zur Topologie der geschlossenen zweiseitigen Flächen I. Acta.
Math. 50 (1927), 189–358. 360
[78] J. Nielsen, Abbildungklassen endlicher Ordnung. Acta. Math. 75 (1942), 23–115. 367
[79] K. Ono, Floer-Novikov cohomology and the flux conjecture. Preprint 2004. 380
[80] S. Smale, Diffeomorphisms of the 2-sphere. Proc. Amer. Math. Soc. 10 (1959), 621–626.
357
[81] S. Smale, Generalized Poincaré’s conjecture in dimensions greater than four. Ann. of Math.
74 (1961), 391–406. 355
[82] W. Thurston, On the structure of volume-preserving diffeomorphisms. Unpublished
manuscript. 381
[83] R. Vakil, The moduli space of curves and its tautological ring. Notices Amer. Math. Soc.
50 (2003), 647–658. 378
[84] K. Vogtmann, Automorphisms of free groups and Outer Space. Geom. Dedicata 94 (2002),
1–31. 354, 370, 378
[85] B. Wajnryb, A simple presentation for the mapping class group of an orientable surface.
Israel J. Math. 45 (1983), 157–174. 365, 370
[86] H. Whitney, Differentiable manifolds. Ann. of Math. 37 (1936), 645–680. 357
[87] H. Zieschang, Discrete groups of motions of the plane and plane group diagrams. Uspehi
Mat. Nauk 29 (1966), 195–212 (in Russian). 361
Chapter 8
Geometric survey of subgroups of

mapping class groups
Lee Mosher
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
2 The action of MC on compactified Teichmüller space . . . . . . . . . . 389
3 The classification of mapping classes . . . . . . . . . . . . . . . . . . . . 392
4 The Tits alternative and abelian subgroups . . . . . . . . . . . . . . . . . . 397
5 Limit sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
6 Convex cocompact subgroups . . . . . . . . . . . . . . . . . . . . . . . . 400
7 Teichmüller discs and their stabilizers . . . . . . . . . . . . . . . . . . . . 404
8 Surface subgroups and the Leininger–Reid combination theorem . . . . . . 407
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
1 Introduction
Let S be a finite type surface, obtained from a closed oriented surface of genus g ≥ 0
by removing p ≥ 0 punctures. The (orientation preserving) mapping class group
of S is denoted MC = MC(S) = Homeo+ (S)/ Homeo0 (S) where Homeo+ (S)
is the group of orientation preserving homeomorphisms and Homeo0 (S) is the nor-
mal subgroup of homeomorphisms isotopic to the identity. The Teichmüller space
of S, denoted T = T (S), is the space of conformal structures on S modulo isotopy,
or equivalently the space of complete, finite area hyperbolic structures on S mod-
ulo isotopy; the phrase “modulo isotopy” means “modulo the action of Homeo0 (S)”.
Thurston’s boundary of Teichmüller space, denoted PMF = PMF (S), is the space
of projective Whitehead classes of measured foliations on S. There is a natural topol-
ogy on the set T = T ∪ PMF which makes it homeomorphic to a ball of dimension
6g − g + 2p with interior T and boundary PMF . The group MC acts naturally
on T and on PMF , extending to an action by homeomorphisms on T .
The theme of this survey is that subgroups of MC can be studied via the ge-
ometric/dynamical properties of their action on T = T ∪ PMF , just as discrete
subgroups of the isometries of hyperbolic space Hn can be studied via their action on
388 Lee Mosher
n
H = Hn ∪ S∞ n−1 . Recent problem lists with this theme in mind include [38] and [40].
We will survey several topics exemplifying this theme, in the following sections:
§3 The Thurston–Bers classification of mapping classes of MC.
§4 The Tits alternative.
§5 Limit sets and domains of discontinuity.
§6 Convex cocompact subgroups.
§7 Stabilizers of Teichmüller discs.
§8 Combination theorems and surface subgroups.
Here is a brief description of each of these topics, to be filled out in the rest of this
chapter.
Thurston [43] and Bers [1] classified individual elements of MC, or equivalently
cyclic subgroups of MC, via the geometry and dynamics of their actions on T . Finite
order mapping classes have fixed points in T . Infinite order irreducible mapping
classes have positive translation distance in T , which is minimized along a unique
Teichmüller geodesic, and they act on T with source-sink dynamics. Finite order
mapping classes have fixed points in T , at which the translation distance of zero is
minimized. And infinite order reducible mapping classes have translation distance
in T whose infimum is not realized.
The Tits alternative, originally formulated and proved for lattices in semisimple
Lie groups, was also formulated and proved for subgroups of MC independently
by McCarthy [33] and Ivanov [25]. The Tits alternative says that each subgroup of
MC has either an abelian subgroup of finite index, or a free subgroup of rank ≥ 2.
Limit sets and domains of discontinuity of subgroups of MC were studied by
McCarthy–Papadopoulos [34], in analogy to limit sets and domains of discontinuity
for discrete subgroups of Isom(Hn ).
Convex cocompact subgroups of MC, including Schottky subgroups of MC,
were originally studied by Farb and Mosher [9] and further developed by Kent–
Leininger [27] and Hamenstädt [17], in analogy to convex cocompact subgroups and
Schottky subgroups of Isom(Hn ).
Teichmüller discs in T are isometrically embedded copies of the hyperbolic plane,
and their stabilizers are analogues in MC of Fuchsian subgroups of Isom(Hn ), which
are stabilizers of isometrically embedded copies of H2 in Hn .
The Leininger–Reid combination theorem [29] gives a method for building closed
surface subgroups of MC, by combining certain Teichmüller disc stabilizer sub-
groups, in analogy to the Maskit combination theorem and related results for building
discrete subgroups of Isom(Hn ) by combining simpler discrete subgroups.
Because of our focus on the geometric/dynamic properties of subgroups of MC,
we have nothing to say about the very rich and interesting algebraic tools that are used
to study subgroups, particularly the Torelli subgroup and the lower central series of
MC.
This work was partially supported by NSF Grant No. 0405979.
Chapter 8. Geometric survey of subgroups of mapping class groups 389
2 The action of MC on compactified Teichmüller space
We start by establishing what will be the basic notions and notations for this survey.
References for this material include [24] and [10].
Teichmüller space and the Thurston boundary. An essential curve on S is a simple

closed curve which bounds neither a disc nor a once-punctured disc. The isotopy class
of an essential curve c is denoted [c]. Let C be the set of isotopy classes of essential
curves on S. An essential curve family is a finite set of pairwise disjoint essential
curves, no two of which are isotopic. The Teichmüller space of S, denoted T (S), is
the set of complete, finite area hyperbolic structures on S, or equivalently the set of
conformal structures with removable singularities at the punctures, modulo isotopy.
Isotopy classes, in C or in T or in other contexts, will be denoted formally with square
brackets [ · ], although informally these brackets will often be dropped, as in the next
sentence. There is an embedding T (S) → RC defined by assigning to each σ ∈ T
and c ∈ C the minimal length of curves isotopic to c in the hyperbolic structure σ .
The topology on T (S) is defined so that the map T (S) → RC is a homeomorphism
onto its image. The group MC(S) acts naturally on C and on T (S), and so the
embedding T (S) → RC is MC-equivariant.
A quadratic differential on σ ∈ T is, formally, a holomorphic section of the
symmetric square of the cotangent bundle of σ , whose integral is finite. Such sections
can be added and multiplied by complex scalars, and so the quadratic differentials on σ
form a vector space QDσ over the complex numbers. Informally, to each conformal
coordinate z for σ , a quadratic differential θ assigns an expression f (z)dz2 , where
f (z) is holomorphic and such that the expression behaves correctly under coordinate
change. For p ∈ S outside of a finite singular set, θ has a local coordinate z in which
its expression is dz2 and z(p) = 0, and z is unique up to multiplication by ±1. On
each point of the singular set, including each puncture, θ has a local coordinate z in
which its expression is zk−2 dz2 for some integer k, where k ≥ 3 if p is not a puncture
and k ≥ 1 if p is a puncture, and this coordinate z is well-defined up to multiplication
by a k th root of unity. The horizontal and vertical measured foliations of θ are defined
in each regular canonical coordinate z = x + iy as follows: the horizontal foliation
has leaves on which y is constant, with transverse measure |dx|; the vertical foliation
has leaves on which x is constant, with transverse measure |dy|. At a singularity p
near which θ has the canonical expression zk−2 dz2 we say that the horizontal and
vertical measured foliations have k-pronged singularities at p.
Teichmüller’s theorem says that for any σ, σ ∈ T there exist unique choice of
quadratic differentials θ on σ called the initial quadratic differential, θ on σ called
the terminal quadratic differential, and a number d ≥ 0, such that if z = x + iy is a
regular local coordinate for θ then (up to isotopy of σ ) the expression z = ed x+ie−d y
defines a regular local coordinate for θ . This number d is called the Teichmüller
distance between σ and σ , and it defines a metric on T (S) which gives the same
topology as that defined above.
390 Lee Mosher
Composing the embedding T (S) → RC with the projectivization RC → P RC ,

the composition T (S) → P RC is still an embedding, and its image is precompact.
The closure of the image minus the image defines Thurston’s boundary for T (S), and
this boundary is characterized as follows. A measured foliation on S is a foliation with
a finite singular set, equipped with a positive transverse Borel measure, so that at each
singularity p including punctures there exists k ≥ 1 such that the singularity is locally
modelled on a k-pronged singularity of zk−2 dz2 , where k ≥ 3 if p is not a puncture.
There is an equivalence relation on measured foliations generated by isotopy and by
the Whitehead move, a move which collapses to a point each leaf segment whose
endpoints are a pair of singularities at least one of which is not a puncture. The
equivalence classes are called Whitehead classes, and the set of Whitehead classes
of measured foliations is denoted MF (again we use [ · ] to denote Whitehead class,
except that the notation is often dropped). There is an embedding MF → RC defined
by assigning to each F ∈ MF and c ∈ C the number F , c ≥ 0 which is the minimal
integral of curves homotopic to c with respect to the transverse measure on F . The
topology on MF is chosen so that the map MF → RC is a homeomorphism onto
its image. When the transverse measure on F is multiplied by a number r > 0 we
denote the result rF , and this generates the relation of projective equivalence on MF .
The space of projective equivalence classes is denoted PMF , and the resulting map
PMF → P RC is an embedding.
Thurston’s compactification theorem says that the disjoint union of T and PMF
embeds into P RC , the image is a closed ball of dimension 6g − 6 + 2p, the inte-
rior of this ball is the image of T , and its boundary sphere is the image of PMF .
There is a natural action of MC on PMF with respect to which the embedding
PMF → P RC is equivariant, and hence MC acts homeomorphically on the ball
T = T ∪ PMF .
Consider an essential closed curve c ∈ C. The enlargement of c is a measured
foliation Fc well-defined up to Whitehead equivalence as follows. First one chooses
a spine for S − c containing all punctures. The complement of this spine is an open
annulus with core c. Then one foliates this annulus by curves isotopic to c, which
together with the spine defines Fc as a singular foliation. The transverse measure on
Fc is defined so that a curve in the annulus from boundary to boundary transverse to Fc
has total transverse measure 1. This construction generalizes to the enlargement of a
weighted multicurve, a formal sum a1 c1 + · · · + ak ck where {c1 , . . . , ck } is an essential
curve system and a1 , . . . , ak > 0; in this case one chooses a spine for S −(c1 ∪· · ·∪ck )
containing all punctures, whose complement is a union of open annuli with cores
c1 , . . . , ck respectively; then one foliates these annuli by closed curves parallel to the
cores; then one assigns total transverse measures a1 , . . . , ak , respectively, to these
foliated annuli. The set of points in PMF represented by enlargements of weighted
multicurves is dense. In fact, the set of points in MF represented by enlargements of
essential closed curves is dense.
The intersection pairing MF × C → [0, +∞) taking (F , c) to F , c extends
to a continuous intersection pairing MF × MF → [0, +∞), called the inter-
section number of F , F ; uniqueness of the extension follows from denseness of C

in MF .
A pair of measured foliations F , F is said to jointly fill the surface S if for every
c ∈ C either F , c or F , c is nonzero, and F , F is also nonzero. This happens
if and only if there exists a conformal structure σ ∈ T and a quadratic differential θ
on σ such that F , F are Whitehead equivalent to the horizontal and vertical measured
foliations of θ; moreover, σ and θ are uniquely determined by F , F .
The set of geodesic lines in T for the Teichmüller metric can be parameterized as
follows. Choose measured foliations F , F which jointly fill. For each r ∈ R, the
measured foliations er F , e−r F jointly fill and determined a conformal structure σr .
The map r
→ σr is a geodesic embedding R → T . The image of this geodesic
depends only on the projective classes P [F ], P [F ], and we say that this is the
Teichmüller geodesic with directions P [F ], P [F ].
The set of geodesic rays in the Teichmüller metric can be parameterized as follows.
First we use a theorem of Hubbard and Masur [21] which says that for each σ ∈ T and
P [F ] ∈ PMF there is up to real scalar multiple a unique θ ∈ QDσ whose horizontal
measured foliation is in the given projective class P [F ]. For each d > 0 we can
then transform each regular canonical coordinate z = x + iy of θ to get a coordinate
z = ed x + ie−d y defining an isometric embedding [0, ∞) → T . The image of this
map is called the Teichmüller ray with basepoint σ and direction P [F ].
A measured foliation F is arational if F , c > 0 for every c ∈ C. Equivalently,
there is no closed loop in S which is everywhere parallel to F , nor is there any closed
path from puncture to puncture which is everywhere parallel to F . If F , F are
arational, and if F , F > 0, then F , F jointly fill S.
Virtual torsion free. Serre [42] originally proved that the group MC has a finite
index subgroup which is torsion free. For another proof see [25] Corollary 1.5.
Theorem 2.1. The kernel of the homomorphism MC(S) → GL(H1 (S; Z/3)) is
torsion free. This kernel has index bounded above by |GL(H1 (S; Z/3)))| which is
finite because H1 (S; Z/3) has finite rank ≤ 2g + p over Z/3.
Note that the bound in the index of the kernel is exponential in g and p.
Gromov hyperbolic metric spaces. Much of this survey is motivated by analogues

in the theory of discrete groups of isometries of Gromov hyperbolic metric spaces,
particularly hyperbolic spaces Hn . We recall some of the notions of this theory, whose
details can be found in, for example, [15], [13], [8], [5].
Let X be a metric space. X is proper if closed balls are compact. X is a
geodesic metric space if for any x, y ∈ X there exists an isometrically embedded
path γ : [0, d(x, y)] → X, a geodesic, from x to y. We shall use the notation [x, y]
to refer to the image of some geodesic. The Hausdorff distance between two sets
A, B ⊂ X is the infimum of δ > 0 such that A ⊂ Nδ (B) and B ⊂ Nδ (A).
392 Lee Mosher
A geodesic metric space X is hyperbolic in the sense of Gromov if there exists

δ > 0 such that for any x, y, z and any geodesics [x, y], [y, z], [z, x] in X, [x, y]
is contained in the δ-neighborhood of [y, z] ∪ [z, x]. The boundary ∂X is defined
to be the set of geodesic rays modulo the relation of finite Hausdorff distance; given
a geodesic ray ρ, let [ρ] denote the corresponding point of ∂X. There is a natural
Gromov topology on X ∪ ∂X, whose definition we shall not recount, but we shall
mention several important properties:
• If X is a proper geodesic metric space then ∂X and X ∪ ∂X are compact.
• Every geodesic ray ρ converges in X ∪ ∂X to [ρ].
• If X is proper then for any x ∈ X and any ξ ∈ ∂X there exists a geodesic ray ρ
based at x and converging to ξ , and for any ξ, η ∈ ∂X there exists a bi-infinite
geodesic whose two ends converge to ξ, η.
Of course, the primary example of the Gromov topology is the union of Hn with its
n−1 . When X is not proper then the compactness properties above
sphere at infinity S∞
can fail.
Another important example for us of a Gromov hyperbolic metric space is the curve
complex of S, discussed below.
3 The classification of mapping classes

The discrete cyclic groups of isometries of Hn or of any proper, geodesic, Gromov
hyperbolic metric space X fall into three classes: elliptic groups which are finite and
which fix a point of X; parabolic groups which are infinite cyclic and which fix a
point of ∂X but no point of X; and loxodromic groups which are infinite cyclic and
which translate along a bi-infinite geodesic of X.
Thurston discovered an analogous classification of elements of MC [43];
see also [10] and [6]. Work of Bers strengthens the analogy in terms of the action
on T [1].
There are two major classifications of a mapping class φ ∈ MC: whether φ is
finite order; and whether φ is reducible meaning that there exists an essential curve
family C, called a reduction family for φ, such that φ[C] = [C]. If φ is not reducible
then it is irreducible.
A homeomorphism : S → S is said to be pseudo-Anosov if there exists a trans-
verse pair of measured foliations F s , F u which jointly fill the surface, called the
stable and unstable measured foliations of , and there exists a number λ > 1 called
the expansion factor of , such that takes each leaf of F s to a leaf of F s , each
leaf of F u to a leaf of F u , contracts the leaves of F s by a factor of λ with respect
to the F u measure, and expands the leaves of F u by a factor of λ with respect to
the F s measure. To say this more briefly, (F s ) = λF s and (F u ) = λ−1 F u .
A mapping class φ ∈ MC(S) is said to be pseudo-Anosov if it is represented by
a pseudo-Anosov homeomorphism.
Thurston proved that every mapping class φ ∈ MC(S) falls into one of three
types: finite order, reducible, or pseudo-Anosov; in other words, every infinite order,
irreducible mapping class is pseudo-Anosov. The first two types are not mutually
exclusive, but a mapping class is pseudo-Anosov if and only if it is infinite order and
irreducible.
There are several necessary and/or sufficient conditions for the three types of map-
ping classes, expressed in terms of the geometry and dynamics of the action on com-
pactified Teichmüller space.
Finite order mapping classes. A mapping class φ ∈ MC(S) is of finite order if

and only if it has a fixed point in T , in which case there exists a conformal structure σ
on S and a homeomorphism representing φ such that (σ ) = σ . It follows that
is a finite order homeomorphism.
A sufficient condition for φ ∈ MC(S) to be of finite order, proved in [10], is the
existence of an arational measured foliation F such that φ[F ] = [F ].
It is possible for a finite order mapping class φ to be irreducible. This happens if
and only if, for a finite order homeomorphism representing φ, the quotient orbifold
S/ is a sphere with three cone points.
There is a bound on the order of a finite order element of MC(S), in fact a bound
on the order of a finite subgroup G < MC(S), depending only on the topology
of S. One way to see this is by applying Theorem 2.1, which gives a bound which is
exponential in g and p. A linear bound is obtained using the much deeper theorem of
Gabai [11] and Casson–Jungreis [7] which says that G is realized as a finite group of
homeomorphisms of S, also denoted G. We therefore have |G| = χ (S)/χ (S/G) =
(2 − 2g − p)/χ(S/G), where the denominator is the orbifold Euler characteristic
of the quotient orbifold S/G. The numerator and denominator of this fraction being
negative, |G| is maximized when χ(S/G) is maximized.
The maximum value of the Euler characteristic of a compact, oriented hyperbolic
2-orbifold is realized by the spherical orbifold with three cone points of angles 2π 2 ,

3 , 7 , whose Euler characteristic equals 2 − 1 − 2 − 1 − 3 − 1 − 7 = 42 . It
2π 2π 1 1 1 1
follows that |G| ≤ 42(2g − 2) when S is compact.

When S has one or more punctures there is a slightly better bound, because the
maximum value of the Euler characteristic of a noncompact, finite area, oriented
hyperbolic 2-orbifold is achieved by the modular space of SL(2, Z), a once-punc-
tured sphere with two cone points of angles 2π 2π
2 , 3 and with Euler characteristic

2 − 1 − 21 − 1 − 13 = 16 , and so |G| ≤ 6(2g − 2 + p).
To summarize:
Theorem 3.1. If G is a finite subgroup of MC then

84g − 84 if p = 0,
|G| ≤
12g − 12 + 6p if p > 0. 2
394 Lee Mosher
Pseudo-Anosov mapping classes. Bers proved [1] that a mapping class φ ∈ MC(S)
is pseudo-Anosov if and only if the infimum of d([σ ], φ[σ ]) is positive and is achieved
by some [σ ] ∈ T . In this case, by combining with the fact that the stable and unstable
foliations F s , F u are uniquely ergodic, it follows that the set of points [σ ] where
the infimum is achieved is a bi-infinite geodesic in T along which φ translates, with
translation distance equal to log(λ). This geodesic is called the axis for φ in T . This
axis has two ideal endpoints in PMF , namely P [F s ] and P [F u ]. The action of φ
on T is a source sink action, with source P [F s ] and sink P [F s ], meaning that for
every point x ∈ T − {P [F s ], P [F u ]},
lim φ i (x) = P [F s ] and lim φ i (x) = P [F u ].

i→−∞ i→+∞
It follows that the fixed point set of φ in T is
Fix(φ) = {P [F s ], P [F u ]}.
A sufficient (and necessary) condition for φ ∈ MC(S) to be pseudo-Anosov

is the existence of an arational measured foliation F and λ > 0, λ = 1 such that
φ[F ] = λ[F ]; if this happens then P [F ] = P [F s ] or P [F u ]. See [10] for the
proof.
Reducible mapping classes. If a mapping class φ is reducible, and if C is a reducing

family for φ, then there exists a representative of φ such that (C) = C. Given
a component S of S − C, let n ≥ 1 be the first return time of to S , meaning that
n (S ) = S and i (S ) = S for 0 < i < n. Then the first return time n and the
mapping class of the first return map n are determined by φ, independent of the
representative . These mapping classes are called the component mapping classes
of φ relative to C.
A reducing family C for a mapping class φ is complete if all of the component
mapping classes of φ relative to C are finite order or pseudo-Anosov. A minimal,
complete reducing family C for φ is unique up to isotopy; see for example [4], [18].
If C, C are any two minimal, complete reducing families for φ, if , are repre-
sentatives of φ such that (C) = C, (C ) = C , and if ∈ Homeo0 (S) satisfies
(C) = C , then conjugates the component mapping classes of φ relative to C
to the component mapping classes of φ relative to C . We may therefore speak, in a
well-defined manner, of the component mapping classes of φ, meaning the component
mapping classes of φ relative to any minimal, complete reducing family.
Bers proved [1] that a mapping class φ is infinite order and reducible if and only if
the infimum of d([σ ], φ[σ ]) is not achieved by any [σ ] ∈ T . Moreover, two subcases
are distinguished: if the infimum equals zero then each component mapping class of φ
has finite order; whereas if the infimum is positive then some component mapping class
of φ is pseudo-Anosov.
A sufficient (and necessary) condition for φ ∈ MC(S) to be reducible is the
existence of a non-arational measured foliation F such that φ[F ] = F .
A complete description of the dynamics of a reducible mapping class acting on

PMF = ∂T is obtained in [25].
Independence of pseudo-Anosov homeomorphisms. In a discrete group acting

on Hn , given two loxodromic elements φ, φ , either Fix(φ) = Fix(φ ) or Fix(φ) ∩
Fix(φ ) = ∅. The same is true for pseudo-Anosov homeomorphisms. Proofs are given
in [25], [34]. We shall give a proof here.
Lemma 3.2. For any pseudo-Anosov φ, φ ∈ MC(S), either Fix(φ) = Fix(φ ) or

Fix(φ) ∩ Fix(φ ) = ∅. In the latter case we say that φ, φ are independent.
Proof. Supposing that Fix(φ) ∩ Fix(φ ) = ∅, up to inverses of φ, φ and up to isotopy

we may assume that F := F u (φ) = F u (φ ), and that φ[F ] = λ[F ] and φ [F ] =
λ [F ] for some λ, λ > 1. Let be the intersection of Stab(P [F ]) and the kernel of
the homomorphism MC(S) → GL(H1 (S; Z/3)), and so by Theorem 2.1 it follows
that has finite index in Stab(P [F ]). It suffices to prove that is infinite cyclic,
for in that case φ, φ have positive powers in , implying that φ m = φ n for some
m, n ≥ 0, implying that [F s (φ)] = [F s (φ m )] = [F s (φ n )] = [F s (φ )].
Consider the homomorphism : → R+ which is characterized by the equation
ψ[F ] = (ψ)[F ]. Suppose that ψ ∈ kernel(). As seen above, ψ has finite order.
But ψ ∈ kernel(MC(S) → GL(H1 (S; Z/3))) and so by Theorem 2.1, ψ is the
identity. The homomorphism is thus injective. But the image of is discrete – indeed,
the set of expansion factors of pseudo-Anosov elements of MC(S) is bounded away
from 1, because each is an eigenvalue of a square integer matrix of bounded size [10].
It follows that is infinite cyclic.
The ping pong argument and free subgroups. Here is the central construction of
the Tits alternative:
Lemma 3.3. For any two independent pseudo-Anosov mapping classes φ, ψ, for suf-
ficiently large integers m, n > 0 the elements φ m , ψ n freely generate a free subgroup
of MC(S).
The proof of this lemma is a standard ping pong argument, using that φ, ψ each
act with source-sink dynamics on T , and that their fixed points sets are disjoint. A
general ping pong lemma stated in a topological context is as follows:
Lemma 3.4 (Topological ping pong). Let G act on a topological space X and let
φ1 , . . . , φk ∈ G. Suppose that there exists a pairwise disjoint collection of open sets
Ui− , Ui+ , i = 1, . . . , k such that for each i ∈ {1, . . . , k} and each ε ∈ {−, +} we have
φiε (X − Ui−ε ) ⊂ Uiε .
Then G is freely generated by φ1 , . . . , φk .
396 Lee Mosher
The way that Lemma 3.3 is proved is to note that when φ, ψ are independent
pseudo-Anosov mapping classes, and if Uφs , Uφu , Uψs , Uψu are pairwise disjoint neigh-
borhoods of the fixed points P [Fφs ], P [Fφu ], P [Fψs ], P [Fψu ], then by applying the
source–sink dynamics of φ and of ψ, the hypotheses of the Ping Pong Lemma are
satisfied for φ m , ψ n as long as m, n are sufficiently large.
One of the main themes of the rest of this survey is that by taking still more care
with the proof of Lemma 3.3, one can show that the free subgroup φ m , ψ n satisfies
stronger geometric properties reminiscent of convex cocompact discrete subgroups of
hyperbolic isometries.
The virtual centralizer and virtual normalizer of a pseudo-Anosov cyclic sub-

group. Consider a group G and a subgroup H < G. The virtual centralizer VC(H )
of H in G is the subgroup of all g ∈ G which commute with a finite index sub-
group of H . The virtual normalizer VN(H ) is the subgroup of all g ∈ G such that
gHg −1 ∩ H has finite index in gH g −1 and in H .
The virtual centralizer and virtual normalizer of an infinite cyclic pseudo-Anosov
subgroup each have a particularly nice geometric/dynamic description, and the proof
of this description uses the ping pong argument in a nice way. Given an action of a
group G on a space X, for a subspace S ∈ X denote Stab(S) = {g ∈ G | g(S) = S}.
If S is homeomorphic to a line denote Stab+ (S) to be the subgroup of Stab(S) that
preserves the orientation on S.
Theorem 3.5. Given a pseudo-Anosov φ ∈ MC(S) with stable and unstable folia-
tions [F s ], [F u ] and Teichmüller axis A, we have
VCφ = Stab(P [F s ]) = Stab(P [F u ]) = Stab+ (A),
VNφ = Stab{P [F s ], P [F u ]} = Stab(A).
The group Stab+ (A) has index at most 2 in Stab(A), it has an infinite cyclic subgroup
of finite index, and the index in Stab+ (A) of the maximal such subgroup is bounded
by the maximum order of a finite subgroup of MC(S).
Proof. The group Stab(A) acts properly discontinuously on the Teichmüller geodesic
A and so is virtually cyclic. The inclusions Stab+ (A) ⊂ VCφ and Stab(A) ⊂ VNφ
follow immediately. The inclusions Stab(P [F s ]), Stab(P [F u ]) ⊂ Stab+ (A) and the
equation Stab{P [F s ], P [F u ]} = Stab(A) are obvious.
Suppose that ψ ∈ MC(S) − Stab{P [F s ], P [F u ]}. Then φ = ψφψ −1 is
pseudo-Anosov, and it is independent of φ by Lemma 3.2. Some powers of φ and φ
therefore generate a free group of rank 2, and hence ψ ∈ VNφ .
Suppose next that ψ ∈ MC(S) − Stab(P [F u ]), and again consider the pseudo-
Anosov mapping class φ = ψφψ −1 . If ψ(P [F u ]) = P [F s ] then φ is independent
of φ and ψ ∈ VNφ , so ψ ∈ VCφ . If ψ(P [F u ]) = P [F s ] but ψ(P [F s ]) =
P [F u ] then Fix(φ), Fix(φ ) are not disjoint and not equal, violating Lemma 3.2.
And if ψ interchanges P [F u ] and P [F s ] then φ translates along A in the direction

opposite to φ and so ψ ∈ VCφ .
4 The Tits alternative and abelian subgroups

The following theorem was proved independently by Ivanov [25] and McCarthy [33]:
Theorem 4.1 (The Tits alternative). For each finite type surface S there exist constants
I , R such that for any subgroup G of MC(S), either G contains a free subgroup of
rank 2, or G contains an abelian subgroup of index ≤ I and rank ≤ R.
Ivanov gave an underlying geometric description of each subgroup of MC(S),

from which the above Tits alternative follows quickly. A subgroup G < MC(S) is
reducible if there exists an essential curve family which is invariant under the action
of each element of G. If no such curve family exists then G is irreducible.
Theorem 4.2. If G is an infinite, irreducible subgroup of MC(S) then G contains

a pseudo-Anosov element.
For the proof see Ivanov’s book [25].

Proof of Tits alternative, from Theorem 4.2. Let G be an arbitrary subgroup of MC.
Suppose first that G is irreducible. Applying Theorem 4.2, G contains a pseudo-
Anosov mapping class φ. If G ⊂ Stab(Fix(φ)) = VN(φ) then by Theorem 3.5 the
group G has an infinite cyclic subgroup of index at most twice the maximum size of
a finite subgroup of MC. If G ⊂ Stab(Fix(φ)), say ψ ∈ G − Stab(Fix(φ)), then
by Lemma 3.2 ψφψ −1 is a pseudo-Anosov element independent of φ and so by the
ping pong argument G contains a free group of rank 2.
Suppose now that G is reducible. Let C be a maximal essential curve family
invariant under G. The group G acts on the set of oriented elements of C; let GC denote
the kernel of this action, a finite index subgroup of G whose index is bounded above by
2n |C|!. Letting {Si } be the components of S − C, it follows that GC preserves each Si
up to isotopy, and so there is a well-defined restriction homomorphism ri : GC →
MC(Si ) whose image we denote G | Si . By maximality of C it follows that G | Si
is irreducible. If G | Si has two independent pseudo-Anosov elements then G | Si has
a free subgroup of rank ≥ 2, and so therefore does G.
We have reduced to the case that G | Si is either finite or is contained in VN(φi )
for some pseudo-Anosov element φi ∈ MC(Si ). In this case we find a free abelian
subgroup of finite index in G. Let Ai ⊂ G | Si denote either the trivial subgroup
or an infinite cyclic group, with index bounded as in Theorem 3.1 or 3.5. Consider
the subgroup H = i r −1 (Ai ) < G. The number of subsurfaces Si is bounded in
terms of g and p, and so we obtain a finite bound on the index of H in G, in terms
398 Lee Mosher
of g and p. Restricted to each Si , the subgroup H is either trivial or infinite cyclic,

in the latter case generated by a pseudo-Anosov element of Si . H also contains Dehn
twists about the curves in C. These pseudo-Anosov subsurface mapping classes and
the Dehn twists all commute with each other. Thus H is a free abelian group, whose
rank is bounded by |C| + |{Si }|, which is bounded in terms of g and p.
Note that this proof gives a complete description of all free abelian subgroups
of MC(S): each irreducible one is infinite cyclic generated by a pseudo-Anosov
mapping class; and each reducible one is freely generated by Dehn twists along curves
C and pseudo-Anosov mapping classes on components {Si } of S − C. To compute the
maximal rank exactly, note that on any subsurface Si that supports a pseudo-Anosov
homeomorphism, one can replace that homeomorphism by a Dehn twist, enlarging C
without decreasing the rank. The maximum rank is thus achieved when each Si is a
pair of pants, in which case the rank equals the maximum value of |C|, which equals
3g − 3 + p.
5 Limit sets
Given a Gromov hyperbolic, geodesic metric space X and a finitely generated, discrete
subgroup G < Isom(X), its limit set G ⊂ ∂X is, by definition, the accumulation set
in ∂X of any orbit of G in X. As long as G is nonelementary (that is, not virtually
abelian), the limit set may be characterized as the closure of the set of fixed points
of loxodromic elements of G, or as the unique minimal closed nonempty G-invariant
subset of ∂X. Assuming in addition that X is proper, it follows that G is compact, and
the domain of discontinuity G = ∂X − G is characterized as the unique maximal
open G-invariant set on which the action of G is properly discontinuous.
In [31], Masur investigated notions of the limit set and domain of discontinuity for
the handlebody subgroup of MC(S) when S is the boundary of a compact genus
g handlebody H ; this subgroup consists of the mapping classes of homeomorphisms
of S that extend to H .
In [34], McCarthy and Papadopoulos generalized Masur’s results to arbitrary sub-
groups of MC(S). The results are not quite as pretty as they are for discrete groups
of hyperbolic isometries. But as we shall see, the results of Kent and Leininger [27]
show that limit sets of convex cocompact subgroups of MC behavior very much
like their counterparts in hyperbolic geometry.
We start with some elementary cases. When G is a finite subgroup of MC, the
natural choice of the limit set is G = ∅, and the action of G on G = PMF
is properly discontinuous. When G is virtually cyclic, containing a pseudo-Anosov
cyclic subgroup φ with finite index, let G = {P [Fφs ], P [Fφu ]}. In the case where
the action of G does not fix each element of G , that is when VCφ is an index 2
subgroup of VNφ , then G is the unique nonempty minimal closed G-invariant
subset of PMF . In the other case, when VCφ = VNφ , then the two points of G
are the only nonempty minimal closed G-invariant subsets of PMF . In either case,
the action of G on G = PMF − G is properly discontinuous.
Consider now the case where G is an infinite, irreducible subgroup of MC and G
has no finite index cyclic subgroup. Define 0 ⊂ PMF to be the set of fixed points of
pseudo-Anosov elements of G. Define G ⊂ PMF to be the closure of 0 . Define
ZG to be the “zero set” of G , the set of all P [F ] ∈ PMF such that for some
P [F ] ∈ G we have [F ], [F ] = 0; this is well-defined independent of the choice
of [F ], [F ] in their projective Whitehead classes. Define G = PMF − ZG .
Theorem 5.1 ([34]). For G an infinite, irreducible subgroup of MC, the set G is
the unique minimal nonempty closed G-invariant subset of PMF . If G = PMF
then G = ∅, and the action of G on G is properly discontinuous.
We turn now to the case that G is an infinite, reducible subgroup of MC(S).

In general there will be no unique closed minimal G-invariant subset of PMF . For
example, if G is a free abelian group generated by Dehn twists about the curves in
an essential curve family C, then G has many, many fixed points, namely any point
in PMF having zero intersection number with each curve in C. Nevertheless, as
McCarthy and Papadopoulos show, there is still a reasonable candidate for a limit set
and a domain of discontinuity.
Let C be an essential curve family invariant under G such that if {Si } is the set
of components of S − C, then the restriction G | Si < MC(Si ) is an irreducible
subgroup of MC(Si ). As is shown in, say, [34], there exists a unique such family C
which is minimal with respect to inclusion. Reindex the Si so that S1 , . . . , Sn are the
surfaces on which the restriction groups G | Si are infinite. Each of the groups G | Si
for i = 1, . . . , n therefore contains a pseudo-Anosov element of MC(Si ). The stable
and unstable foliations of pseudo-Anosov elements of G | Si , being measured foliations
on Si , may by enlargement be regarded as measured foliations on S; let i0 ⊂ PMF
be the set of all points in PMF obtained in this manner. Let i ⊂ PMF be the
closure of i0 . Define

n
G = i ∪ P [c]
i=1 c∈C
Define the zero set ZG and the domain G = PMF − ZG as above.
Theorem 5.2 ([34]). For each infinite, reducible subgroup of MC, the set G is a
nonempty closed G-invariant subset of PMF . If G = PMF then G = ∅, and
the action of G on G is properly discontinuous.
The set G shall be called the limit set and G the domain of discontinuity of G.
400 Lee Mosher
6 Convex cocompact subgroups

Given a finitely generated, discrete subgroup G < Isom(Hn ), the convex hull of its
limit set G is the smallest closed, convex subset HG ⊂ Hn whose set of accumulation
n−1 is . The action of G on Hn preserves H , and G is said to be convex
points in S∞ G G
cocompact if G acts cocompactly on HG . Several conditions are equivalent to convex
cocompactness: some (every) orbit of G in HG is quasiconvex; the action of G on G
is cocompact; the action of G on Hn ∪ G is cocompact. Moreover, if G is convex
cocompact then G is word hyperbolic, there is a G-equivariant homeomorphism from
the Gromov boundary ∂G to G , and this homeomorphism extends to a G-equivariant
continuous map from the Gromov compactification G ∪ ∂G to HG ∪ G .
These notions can be generalized with care to a finitely generated, discrete group G
of isometries of any Gromov hyperbolic, geodesic metric space X.
The theory of convex cocompact subgroups of MC, originated in [9] and further
developed in [27], and [17], gives analogues to these results about convex cocompact
subgroups of MC. In addition, the theory develops further equivalent characteriza-
tions of convex compactness that are special to the setting of MC and are important
for applications. Thus far, the only examples of convex cocompact subgroups of
MC are free [9] or virtually free examples as constructed by Honglin Min in her
dissertation [36].
We define a subset X ⊂ T to be quasiconvex if there exist a constant C such that
for every x, y ∈ X, each point of the Teichmüller geodesic [x, y] is within distance C
of some point of X.
Given a closed subset ⊂ PMF , if every pair ξ = η ∈ jointly fills S then we
define the weak hull of to be the subset of T defined by

←−→
H() = (ξ, η) | ξ = η ∈
and we say that the weak hull of is defined. Otherwise, if there exist ξ = η ∈
which do not jointly fill S then we say that the weak hull of is undefined.
Theorem 6.1 ([9]). For any finitely generated subgroup G < MC, the following
are equivalent.
• Every (some) orbit of G on MC is quasiconvex in T .
• G is word hyperbolic, and there exists a G-equivariant embedding ∂G → PMF
with image denoted , such that the weak hull of is defined, the action of G
on Hull() is cocompact, and the extension of the map ∂G → PMF by any
G-equivariant map G → Hull() is a continuous map G ∪ ∂G → Hull() ∪ .
A group satisfying these conditions is called convex cocompact. When G is convex

cocompact then every infinite order element of G is pseudo-Anosov.
The connection of convex cocompactness with the limit set G and domain of
discontinuity G as defined by McCarthy–Papadapoulos is given in the following
results of Kent and Leininger:
Theorem 6.2 ([27]). For any finitely generated subgroup G < MC, the following
are equivalent:
• G is convex cocompact.
• The weak hull HullG of the McCarthy–Papadapoulos limit set G is defined, and
the action of G on HullG is cocompact.
• The action of G on T ∪ G is cocompact.
Furthermore, if G is convex cocompact, then the image of the embedding ∂G →
PMF is the McCarthy–Papadapoulos limit set G , and ZG = G , and so G =
PMF − G .
Convex cocompactness on the curve complex. The curve complex CC = CC(S)

is the simplicial complex whose vertex set is C, and whose n-simplices are the isotopy
classes of essential curve families of cardinality n + 1. To be precise, a set of distinct
vertices [c0 ], . . . , [cn ] ∈ C spans an n-simplex of CC if and only if there are repre-
sentatives c0 , . . . , cn which are pairwise disjoint. The complex CC is locally infinite,
in fact the link of every simplex of positive codimension is infinite.
The natural action of MC on CC has finitely many orbits of simplices. The
number of orbits of vertices is
⎧
⎪
⎪ 1 + g−1 if g ≥ 1, p = 0,
⎪
⎨ 2
p−3

m= if g = 0, p ≥ 4,
⎪
⎪
2

⎪
⎩1 + (g+1)(p+1)−4
2 if g, p ≥ 1.
When g ≥ 1 there is one orbit of nonseparating curves, but in genus zero every curve
is separating. The orbit of a separating curve is determined by how it partitions the
genus and the number of punctures. When p = 0, the relevant partition is g = g1 + g2
with g1 , g2 ∈ {1, . . . , g}, and when g = 0 the relevant partition is p = p1 + p2
with p1 , p2 ∈ {2, . . . , p}. When g ≥ 1, p ≥ 1 then the relevant partition is an
unordered partition of the ordered pair (g, p) = (g1 , p1 ) + (g2 , p2 ), where the four
values (gi , pi ) = (0, 0), (g, p), (0, 1), (g, p − 1) are forbidden but all other values
with gi ∈ {0, . . . , g} and pi ∈ {0, . . . , p} are allowed.
The large scale geometric properties of CC and of MC are related in the following
manner.
Given a finitely generated group G and a finite collection of finitely generated
subgroups H1 , . . . , Hm , assume that the generators of G include generators of each of
the subgroups H1 , . . . , Hm , so that the Cayley graph of Hi embeds, Hi -equivariantly,
in the Cayley graph of G. Define the coned off Cayley graph of G relative to
H1 , . . . , Hm by adding a new vertex to for each left coset gHi of each Hi , and
adding new edges connecting the new vertex to all the vertices of gHi . We say that
is weakly hyperbolic relative to H1 , . . . , Hm if the coned off Cayley graph is a Gromov
hyperbolic metric space.
402 Lee Mosher
Choose unique representatives [c1 ], . . . , [cm ] for the orbits of the action of MC
on C, and let Hi = Stab[ci ] < MC. It is easy to see [32] that CC is equivariantly
quasi-isometric to the coned off Cayley graph of MC relative to H1 , . . . , Hm . Much
deeper is the question of what, exactly, is the large scale geometric behavior of these
two spaces:
Theorem 6.3 ([32]). The curve complex CC is a Gromov hyperbolic metric space,
and MC is weakly hyperbolic relative to stabilizer subgroups of representatives of
orbits in C.
The boundary of the curve complex was identified in the following theorem of
Klarreich [28]; see also [16]. Two points in PMF are topologically equivalent if they
are represented by singular foliations which, after forgetting measure, are identical.
Theorem 6.4. There is an MC equivariant quotient map from the subspace of
PMF consisting of classes of arational measured foliations to the Gromov boundary
of CC, such that the decomposition elements of this quotient map are the topological
equivalence classes of arational measured foliations.
Using the fact that CC is convex cocompact, it makes sense to compare convex
cocompactness of a subgroup of MC with respect to its actions on T and on CC.
This is accomplished in the following theorem proved independently, and with different
techniques, by Kent–Leininger and Hamenstädt:
Theorem 6.5 ([27], [17]). A finitely generated subgroup G < MC is convex co-
compact if and only if its action on CC is convex cocompact.
Schottky and virtual Schottky subgroups of mapping class groups. A convex

cocompact, free subgroup of MC is called a Schottky subgroup. The ubiquity of
Schottky subgroups was first demonstrated by Farb and myself in [9], using a result
of [37] which is to be explained below.
Theorem 6.6 ([9]). For any independent set of pseudo-Anosov mapping classes
φ1 , . . . , φK ∈ MC(S), there exists an integer B ≥ 1 such that for all integers
β β
β1 , . . . , βK ≥ B, the mapping classes φ1 1 , . . . , φKK freely generates a Schottky sub-
group of MC(S).
Kent and Leininger [27] and independently Hamenstädt [17] discovered a more
direct and natural setting for this theorem, using a ping pong argument in the curve
complex. Here is their proof.
The action of φk on PMF has source sink dynamics, with source and sink each
being arational, and for k = 1, . . . , K the 2K fixed points are pairwise topologically
inequivalent, because they are pairwise distinct and each is uniquely ergodic. By
Theorem 6.4, it follows that the action of each φk on ∂CC has source sink dynamics.
and so φk is a hyperbolic isometry on the Gromov hyperbolic metric space CC, and
β β
that their fixed point sets in ∂CC are pairwise disjoint. To conclude that φ1 1 , . . . , φKK
freely generate a convex cocompact subgroup of MC, we apply the following folk
theorem:
Theorem 6.7 (Hyperbolic ping pong). If X is a Gromov hyperbolic geodesic met-

ric space, and if φ1 , . . . , φK are hyperbolic isometries of X with pairwise disjoint
fixed points in ∂X, then there exists an integer B ≥ 1 such that for all integers
β β
β1 , . . . , βK ≥ B, the isometries φ1 1 , . . . , φKK freely generate a convex cocompact
group of isometries.
It seems hard to track down an early proof of hyperbolic ping pong, although it
was certainly known to the very earliest practitioners of Gromov hyperbolicity. Here
is a sketch of a proof, pretty much the same as the proof given in [27].
One may choose λ, ε quasigeodesic lines γ1 , . . . , γK in X that are axes for
φ1 , . . . , φK , respectively (if X were proper then one could choose λ = 1, ε = 0,
in other words, geodesics). Fix a base point p ∈ X, and let D be an upper bound
for the distance from p to γ1 , . . . , γK . Also, let L be a lower bound for the length
of a fundamental domain of each φ1 , . . . , φK along its axis γ1 , . . . , γK . Fix B and
β β
β1 , . . . , βK ≥ B, let G be the group generated by φ1 1 , . . . , φKK .
We know by topological ping pong that if B is sufficiently large then G is freely
β β
generated by φ1 1 , . . . , φKK . It suffices to show that for each word w = w1 . . . wm in
the generators, a geodesic from p to w(p) stays uniformly close to the orbit G · p. We
βk
describe a path γw from p to w(p) as follows. Denote wi = φki i . Starting from p,
jump a distance at most D onto the axis γk1 , travel along γ1 for β1 fundamental
domains, thereby travelling a distance of at least BL, then jump a distance at most D
to the point w1 (p). Next, jump a distance at most D onto the axis w1 (γk2 ), travel
along this axis for β2 fundamental domains, thereby travelling a distance of at least
BL, then jump a distance at most D to the point w1 w2 (p). Continuing in this manner
we get a path γw from p to w(p), which stays in a D + BL neighborhood of G · p.
The path γw is a (BL + 2D) local λ, (ε + 2D) quasigeodesic. If B is sufficiently
large, it follows that γw is a λ, ε quasigeodesic for ε depending only on the previous
constants and the hyperbolicity constant of X [5]. Now we use the fact that γw has
Hausdorff distance at most E from any geodesic with the same endpoints p, w(p),
where E depends only on λ, ε , and the hyperbolicity constant for X. It follows that
stays within distance E + D + BL of G · p.
Virtual Schottky subgroups. A virtual Schottky subgroup of MC is a subgroup

that contains a Schottky subgroup with finite index. Clearly convex cocompactness
is preserved under passage to finite index supergroups, so every virtual Schottky sub-
group is convex cocompact. The following recent result, to appear in the dissertation
of Honglin Min, is another example of a ping pong argument in the mapping class
group:
404 Lee Mosher
Theorem 6.8 ([36]). If A, B are finite subgroups of MC, if φ ∈ MC(S) is pseudo-

Anosov, and if the virtual normalizer of φ has trivial intersection with A and B, then
for all sufficiently large n the subgroups A and φ n Bφ −n freely generate their free
product in MC, and this subgroup is virtually Schottky.
Hyperbolic extensions of surface groups When the surface S is closed and oriented,
there is a canonical short exact sequence
1 → π1 (S, p) → MC(S − p) → MC(S) → 1.
See [3].
Given a subgroup H ⊂ MC(S), let H be its preimage in MC(S − p), and
so we have an extension
1 → π1 (S, p) → H → H → 1.
Thurston’s geometrization theorem (see [39]) shows that if H is an infinite cyclic,
pseudo-Anosov subgroup of MC(S) then H is a word hyperbolic group.
The following theorem was the prequel to the theory of convex cocompact groups.
It is proved by an application of the Bestvina–Feighn combination theorem.
Theorem 6.9 ([37]). If φ1 , . . . , φK are independent pseudo-Anosov elements of MC

then sufficiently high powers of φ1 , . . . , φK freely generate a subgroup H such that
H is word hyperbolic.
With this result as motivation, in [9] Farb and I set out to develop a geometric
understanding of when H is word hyperbolic, attempting to prove that for this to
be true it was necessary and sufficient that H be convex cocompact. We proved
necessity, and when H is free we proved sufficiency by using the Bestvina–Feighn
combination theorem [2]. A completely general proof of sufficiency was recently
given by Hamenstädt [17], so that now one can state:
Theorem 6.10 ([9], [17]). For any subgroup H ⊂ MC, the group H is word
hyperbolic if and only if H is convex cocompact.
7 Teichmüller discs and their stabilizers

One very well studied class of subgroups of MC is stabilizers of Teichmüller discs.
A Teichmüller disc in T can be defined in one of the following equivalent ways.
First, consider σ ∈ T and θ ∈ QDσ . As ω varies over the unit circle in the complex
plane, consider the family of quadratic differentials ωθ . This defines a family of rays
based at σ parameterized by the circle, and the union of these rays is defined to be
the Teichmüller disc with basepoint σ and direction θ, denoted D(σ, θ ). A somewhat
more invariant way to say almost the same thing is that D(σ, θ ) is the set of points
in T obtained by choosing an element of PSL(2, R) acting affinely on the Euclidean

plane, and applying that element to the regular canonical coordinates for θ to get a
new conformal structure.
Second, a Teichmüller disc D is a maximal subset which is the image of a holo-
morphic embedding of the Poincaré disc into T . A theorem of Royden [41] in the
compact case and Gardiner [12] in the noncompact case says that any two points
σ = σ ∈ T lie on a unique such disc D, and moreover that the Teichmüller distance
between σ and σ equals their distance in the Poincaré metric on D. The identification
of D with D(σ, θ) for some θ is obtained by choosing θ to be the initial quadratic
differential from σ to σ .
To each Teichmüller disc D there is associated an embedded circle S 1 (D) ⊂
PMF , defined as follows. Choose σ ∈ T and θ ∈ QDσ so that D = D(σ, θ ). As ω
varies over the unique circle in the complex plane, the horizontal measured foliations
of ωθ trace out the desired circle S 1 (D). Alternatively, a point P [F ] ∈ PMF is
in S 1 (D) if and only if it is represented as a linear measured foliation in any regular
canonical coordinate z for θ; in other words, there exists an R-linear differential form
a dx + b dy such that P [F ] is represented in any z by |a dx + b dy|. This makes it
clear that S 1 (D) is well-defined independent of σ ∈ D, because any other σ ∈ D is
obtained by a Teichmüller deformation with initial quadratic differential ωθ for some
ω in the unit circle, resulting in a terminal quadratic differential θ on σ , and linearity
with respect to θ and to θ are clearly equivalent.
Let D ⊂ T be a Teichmüller disc, and let Stab(D) = {φ ∈ MC | φ(D) = D}
be its stabilizer. The group Stab(D), or its image in PSL(2, R) when regarding D
as the upper half plane, is sometimes called a Veech group, and the quotient orbifold
D/ Stab(D), a proper holomorphic curve in the moduli space of S, is sometimes called
a Veech surface. These terms are sometimes restricted to the case where Stab(D) is a
lattice, meaning that D/ Stab(D) has finite hyperbolic area.
Regarding D as the hyperbolic plane, the elliptic–parabolic–loxodromic trichotomy
for Isom(H2 ) becomes not just an analogy but a strict correspondence with the finite
order–reducible–pseudo-Anosov dichotomy for elements of Stab(D); for details see
for example [44]. To be precise, fix φ ∈ Stab(D). Then φ is loxodromic on D if and
only if it is pseudo-Anosov in MC, φ is parabolic on D if and only if it is reducible
in MC, and φ is elliptic on D if and only if it is finite order in MC. The case
where φ is parabolic can be analyzed more closely. In this case, letting D = D(σ, θ ),
the action of φ is a shear transformation on any regular canonical coordinate z, parallel
to the leaves of some constant slope measured foliation F called the shear foliation
for φ; this foliation corresponds to the point of S 1 (D) fixed by φ. Each leaf of F is
compact, and so F is obtained by enlarging some weighted family of simple closed
curves c1 , . . . , cn . In particular, the essential curve family {c1 , . . . , cn } is a canonical
reducing system for φ, and the first return of φ to each complementary component
of S − (c1 ∪ · · · ∪ cn ) is of finite order, so φ is the composition of a finite order
mapping class that is reduced by {c1 , . . . , cn } and a product of Dehn twists along the
components of c1 , . . . , cn .
406 Lee Mosher
Using this description, we can get a fairly precise understanding of the case when
Stab(D) is a reducible subgroup of MC. Assuming this is so, let φ ∈ Stab(D) be
of infinite order, with shear foliation F . For any other infinite order φ ∈ Stab(D),
the shear foliation of φ must have the same slope as F and so represent the same
point in S 1 (D). Otherwise, if the shear foliation F of φ is distinct from F in S 1 (D),
then high powers of φ and φ would generate a loxodromic element of Stab(D), by
the proof of the Tits alternative for H2 (this is not hyperbolic ping pong, because φ
and φ do not have source–sink dynamics, but a variation of hyperbolic pong-pong
works). Thus, every infinite order element of Stab(D) stabilizes the same point in
S 1 (D), in fact every such element acts as a shear transformation on θ with respect
to the same linear measured foliation Fω . But this implies that Stab(D) is virtually
cyclic, generated by a multiple Dehn twist about components of the curves c1 , . . . , cn
whose enlargement is Fω . In this case we can identify the McCarthy–Papadopoulos
limit set of Stab(D): it is the simplex in PMF spanned by the projective classes of
c1 , . . . , cn .
Henceforth we shall only be interested in the case when Stab(D) is infinite and
irreducible. In this case, since S 1 (D) is invariant under Stab(D) by construction, and
since S 1 (D) is obviously closed in PMF , by applying Theorem 5.1 we obtain:
Theorem 7.1. If Stab(D) is infinite and irreducible then the McCarthy–Papadopoulos

limit set of Stab(D) is contained in S 1 (D).
Thurston [43] gave the first examples of Teichmüller discs D such that Stab(D)
is a lattice, meaning that it acts on D with cofinite area. Start with the linear action
of SL(2, Z) on T 2 = R2 /R2 , then choose a finite subset A ⊂ T 2 invariant under
this action, and let S → T 2 be a branched cover of T 2 branched over A. Lifting
the conformal structure from T 2 to S defines σ ∈ T (S), and lifting the quadratic
differential (dx + i dy)2 from T 2 to S defines θ ∈ QDσ . There exists a finite index
subgroup G < SL(2, Z) that lifts to S, and G stabilizes D(σ, θ ) and acts with cofinite
area. Combining with Theorem 7.1 it follows that the McCarthy–Papadopoulos limit
set of G is the circle S 1 (D).
Veech [44] gave the first examples where Stab(D) is a lattice that does not arise
from Thurston’s construction of branching over the torus. Veech also developed some
of the general theory of Stab(D), for instance establishing that Stab(D) never acts
cocompactly on D. When Stab(D) is a lattice, it follows that there are finitely many
maximal parabolic subgroups of Stab(D) up to conjugacy, and the Veech surface
D/ Stab(D) is a finite area holomorphic curve in the moduli space of S.
McMullen [35] proved that if S is the closed surface of genus 2, if D ⊂ T (S) is
a Teichmüller disc, and if Stab(D) contains a hyperbolic element, then the limit set
of Stab(D) is all of S 1 (D). In the same paper McMullen also gave examples where
Stab(D) is infinitely generated, a phenomenon independently discovered by Hubert
and Schmidt [23].
For a further survey of stabilizers of Teichmüller discs, and references to the lit-
erature, see the section “Veech surfaces” in the problem list [22], and the chapters by
Harvey [19] and by Herrlich and Schmithüsen [20] in this Handbook.
8 Surface subgroups and the Leininger–Reid

combination theorem
The only known examples of convex cocompact subgroups are free groups and virtu-
ally free groups. Even in Isom(Hn ), constructing nonfree discrete subgroups can be
subtle. One of the pioneering tools in this regard is the Maskit combination theorem
[30].
One of the most interesting questions about subgroups of mapping class groups
is whether there exists closed surfaces S, F of genus ≥ 2 and a convex cocompact
subgroup of MC(S) isomorphic to π1 (F ). If such a subgroup existed then, by
Hamenstädt’s theorem, one would obtain the first example of a compact 4-manifold
M with word hyperbolic fundamental group that fibers over a surface. The preprint
[26] of M. Kapovich contains a theorem that says if such an M existed then M could
not have a CH2 structure.
It is even hard to construct any surface subgroup π1 (F ) of MC(S) with the genus
of F at least two. The first examples were given by Harvey and González–Díez [14],
but they have many non-pseudo-Anosov elements and so cannot be convex cocompact.
The closest approach yet to a convex cocompact surface subgroup of MC is
the following theorem of Leininger and Reid, whose statement is reminiscent of the
Maskit combination theorem.
The starting point of the theorem is the construction of a Teichmüller disc D and
finite index subgroup G < Stab(D) such that Stab(D) (and hence also G) is a lattice, G
is torsion free, and G has exactly one parabolic subgroup, generated by a multitwist τ
about a multicurve c such that no component of S − c is a three holed sphere. It
follows that there exists a homeomorphism ψ that preserves c and is pseudo-Anosov
on each component of S − c.
Theorem 8.1 ([29]). With the objects as denoted above, for each sufficiently large k
the group generated by G and φ k Gφ −k is a free product with amalgamation along
τ , and is therefore isomorphic to π1 (F ) for an oriented surface F of genus ≥ 2.
Moreover, every element of π1 (F ) < MC(S) is pseudo-Anosov except for elements
conjugate to powers of τ .
A Leininger–Reid subgroup cannot be convex cocompact, because it contains a

multitwist. In [38], a notion of geometric finiteness for subgroups of MC is pro-
posed, and the question is posed whether Leininger–Reid subgroups are geometrically
finite.
408 Lee Mosher
References
Acta Math. 141 (1978), 73–98. 388, 392, 394
[2] M. Bestvina and M. Feighn, A combination theorem for negatively curved groups. J. Dif-
ferential Geom. 35 (1) (1992), 85–101. 404
[3] J. Birman, Braids, links, and mapping class groups. Ann, of Math. Stud, 82, Princeton
University Press, Princeton, N.J., 1974. 404
[4] J. Birman, A. Lubotzky, and J. McCarthy, Abelian and solvable subgroups of the mapping
class groups. Duke Math. J. 50 (4) (1983), 1107–1120. 394
[5] J. Cannon, The theory of negatively curved spaces and groups. In Ergodic theory, symbolic
dynamics, and hyperbolic spaces (C. Series, T. Bedford, M. Keane, eds.), Oxford Sci.
Publ., Oxford University Press, New York 1991, 315–369. 391, 403
[6] A. Casson and S. Bleiler, Automorphisms of surfaces after Nielsen and Thurston. London
Math. Soc. Student Texts 9, Cambridge University Press, Cambridge 1988. 392
[7] A. Casson and D. Jungreis, Convergence groups and Seifert fibered 3-manifolds. Invent.
Math. 118 (1994), 441–456. 393
[8] M. Coornaert, T. Delzant, and A. Papadopoulos, Géométrie et théorie des groupes. Lecture
Notes in Math. 1441, Springer-Verlag, Berlin 1990. 391
[9] B. Farb and L. Mosher, Convex cocompact subgroups of mapping class groups. Geom.
Topol. 6 (2002), 91–152. 388, 400, 402, 404
Astérisque 66–67 (1979). 389, 392, 393, 394, 395
[11] D. Gabai, Convergence groups are Fuchsian groups. Ann. of Math. 136 (1992), 447–510.
393
[12] F. P. Gardiner, Approximation of infinite-dimensional Teichmüller spaces. Trans. Amer.
Math. Soc. 282 (1) (1984), 367–383. 405
[13] E. Ghys and P. de la Harpe (eds.), Sur les groupes hyperboliques d’après Mikhael Gromov.
Progr. Math. 83, Birkhäuser, Boston, MA, 1990. 391
[14] G. González-Díez and W. J. Harvey, Surface groups inside mapping class groups. Topology
38 (1) (1999), 57–69. 407
[15] M. Gromov, Hyperbolic groups. In Essays in group theory (S. Gersten, ed.), Math. Sci.
[16] U. Hamenstädt, Train tracks and the Gromov boundary of the complex of curves. In Spaces
of Kleinian groups, London Math. Soc. Lecture Note Ser. 329, 2005, 187–207. 402
[17] U. Hamenstädt, Word hyperbolic extensions of surface groups. Preprint, 2005;
arXiv:math.GT/0505244. 388, 400, 402, 404
[18] M. Handel and W. Thurston, New proofs of some results of Nielsen. Adv. in Math. 56 (2)
(1985), 173–191. 394
[19] W. Harvey, Teichmüller disks, triangle groups and Grothendieck dessins. In Handbook
of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing House, Zurich
2007, 249–292. 407
[20] F. Herrlich and G. Schmithüsen, On the boundary of Teichmüller disks in Teichmüller
space and in Schottky space. In Handbook of Teichmüller theory (A. Papadopoulos, ed.),
Volume I, EMS Publishing House, Zurich 2007, 293–349. 407
221–274. 391
[22] P. Hubert, H. Masur, T. Schmidt, and A. Zorich, Problems on billiards, flat
surfaces, and translation surfaces. Preprint, 2005; http://www.math.uic.edu/~masur/
papers.html. 407
[23] P. Hubert and T. Schmidt, Infinitely generated Veech groups. Duke Math. J. 123 (1) (2004),
49–69. 406
Tokyo 1992. 389
[25] N. V. Ivanov, Subgroups of Teichmüller modular groups. Transl. Math. Monogr. 115,
Amer. Math. Soc., Providence, RI, 1992. 388, 391, 395, 397
[26] M. Kapovich, On normal subgroups in the fundamental groups of complex surfaces.
Preprint, 1998; arXiv:math.GT/9808085. 407
[27] R. P. Kent, IV and C. J. Leininger, Shadows of mapping class groups: capturing convex
co-cocompactness. Geom. Funct. Anal. (2007), to appear; arXiv:math.GT/0505114. 388,
398, 400, 401, 402, 403
[28] E. Klarreich, The boundary at infinity of the curve complex and the relative Teichmüller
space. Preprint, 1999. 402
[29] C. Leininger and A. Reid, A combination theorem for Veech subgroups of the mapping
class group. Geom. Funct. Anal. 16 (2) (2006), 403–436. 388, 407
[30] B. Maskit, Kleinian groups. Grundlehren Math. Wiss. 287, Springer-Verlag, Berlin 1988.
407
[31] H. Masur, Measured foliations and handlebodies. Ergodic Theory Dynam. Systems 6 (1)
(1986), 99–116. 398
[32] H. Masur and Y. Minsky, Geometry of the complex of curves, I. Hyperbolicity. Invent.
Math. 138 (1) (1999), 103–149. 402
[33] J. McCarthy, A “Tits-alternative” for subgroups of surface mapping class groups. Trans.
Amer. Math. Soc. 291 (2) (1985), 582–612. 388, 397
[34] J. McCarthy and A. Papadopoulos, Dynamics on Thurston’s sphere of projective measured
foliations. Comment. Math. Helv. 64 (1) (1989), 133–166. 388, 395, 398, 399
[35] C. McMullen, Teichmüller geodesics of infinite complexity. Acta Math. 191 (2) (2003),
191–223. 406
[36] H. Min, Hyperbolic graphs of surface groups. Ph.D. thesis, Rutgers University, Newark,
2007, in preparation. 400, 404
[37] L. Mosher, A hyperbolic-by-hyperbolic hyperbolic group. Proc. Amer. Math. Soc. 125
(12) (1997), 3447–3455. 402, 404
[38] L. Mosher, Problems in the geometry of surface group extensions. In Problems on mapping
class groups and related topics, Proc. Sympos. Pure Math., Amer. Math. Soc. 74, 2006,
245–256. 388, 407
[39] J.-P. Otal, Le théorème d’hyperbolisation pour les variétés fibrées de dimension 3.
Astérisque 235 (1996). 404
410 Lee Mosher
[40] A. Reid, Surface subgroups of mapping class groups. Problems on mapping class groups
and related topics, In Problems on mapping class groups and related topics, Proc. Sympos.
Pure Math., Amer. Math. Soc. 74, 2006, 257–268. 388
[41] H. Royden, Automorphisms and isometries of Teichmüller space. In Advances in the The-
ory of Riemann Surfaces, Ann. of Math. Stud. 66, Princeton University Press, Princeton,
N.J., 1971, 369–383. 405
[42] J.-P. Serre, Rigidité du foncteur de Jacobi d’échelon n ≥ 3. Appendix to A. Grothendieck,
Techniques de construction en géométrie analytique, X. Construction de l’espace de
Teichmüller, Séminaire Henri Cartan, no. 17, Secrétariat mathématique de l’ENS, Paris,
1960/61. 391
Amer. Math. Soc. 19 (1988), 417–431. 388, 392, 406
[44] W. A. Veech, Teichmüller curves in moduli space, Eisenstein series and an application to
triangular billiards. Invent. Math. 97 (3) (1989), 553–583; erratum ibid 103 (2) (1991),
447. 405, 406
Chapter 9
Deformations of Kleinian groups
Albert Marden
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
2 Kleinian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
2.1 The 3-dimensional point of view . . . . . . . . . . . . . . . . . . . . 413
2.2 The classical examples . . . . . . . . . . . . . . . . . . . . . . . . . 418
2.3 The convex core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
2.4 The Hyperbolization Theorem . . . . . . . . . . . . . . . . . . . . . 420
2.5 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3 Boundary deformations of geometrically finite groups . . . . . . . . . . . 422
3.1 The quasiconformal deformation space . . . . . . . . . . . . . . . . . 423
4 Geometric deformations of the groups; the representation variety . . . . . . 425
4.1 Deforming the convex core . . . . . . . . . . . . . . . . . . . . . . . 427
5 The deformation space boundary . . . . . . . . . . . . . . . . . . . . . . . 427
6 Deformation of surface groups . . . . . . . . . . . . . . . . . . . . . . . . 429
6.1 Bers slices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
6.2 Earthquakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
6.3 Pleated surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
6.4 Complex scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
6.5 Complex earthquakes . . . . . . . . . . . . . . . . . . . . . . . . . . 438
7 Complex scaling for general laminations . . . . . . . . . . . . . . . . . . 438
7.1 Sullivan’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
7.2 Earthquake disks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
7.3 Appendix: holomorphic motion . . . . . . . . . . . . . . . . . . . . 442
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
1 Introduction
The goal of this chapter is
• to discuss how and why the deformations of hyperbolic 3-manifolds are controlled
by the deformation of the complex structures of their boundary components;
• to discuss the approach to deforming Riemann surfaces by directly deforming
their hyperbolic structures.
412 Albert Marden
In this exposition, we assume readers are familiar with the classical theory of Teich-
müller spaces of Riemann surfaces, and with the hyperbolic plane H2 and space H3 .
We will include short introductions to Kleinian groups and 3-manifolds, earthquakes,
complex scaling, pleated surfaces, and in an appendix, holomorphic motions.
We thank Lee Mosher for helpful comments.
2 Kleinian groups
We will represent H3 by one of the two common models that best fits the situation at
hand: (i) Upper halfspace (UHS) {(z, t) : z ∈ C, t > 0} with the metric ds = |dx| t , or
2|dx|
(ii) the ball {x ∈ R : |x| < 1} with the metric ds = 1−|x|2 . Here |dx| is the euclidean
3
distance element. These are referred to as the “conformal models” because they rep-
resent the angles correctly. Stereographic projection extends from S2 → C ∪ ∞ to
an isometry of the ball model onto the UHS model. Möbius transformations extend
naturally from C ∪ ∞ ≡ S2 to UHS and form there the full group of orientation pre-
serving isometries. Stereographic projection conjugates this group to the orientation
preserving isometries of the ball model.
A Kleinian group is a discrete group G of (orientation preserving) Möbius trans-
formations acting on hyperbolic 3-space H3 and its boundary at infinity which we will
denote as ∂ H3 ≡ S2 . For current expositions of the theory, see [62], [34], [47], [45].
For the Kleinian groups G employed in this chapter, we will assume that (i) G is
nonelementary in that its limit set (see below) contains at least three points, (ii) G
has no elliptic elements (no torsion) – except as explicitly mentioned otherwise, and
(iii) G is finitely generated.
It formally follows from Theorem 2.3 that (iii) implies G is finitely presented [55].
The ordinary set of G is denoted by (G) ⊂ S2 and the limit set by (G) ⊂ S2 .
The limit set (G) is the closed, perfect set which is the closure of the set of fixed
points of G. It is also the set of accumulation points of the orbit {G(x)} of any
x ∈ H3 ∪ ∂ H3 . Its complementary open set (G) on ∂ H3 , which may be empty, is
the largest open set in ∂ H3 in which G is properly discontinuous. If (G) = S2 , it
has either one, two or infinitely many components each of which is either simply or
infinitely connected.
The Ahlfors conjecture was that, if G is finitely generated, either (G) = S2 , or
it has zero Lebesgue area. This is now a theorem; the coup de grâce was provided by
confirmation of the Tameness Conjecture (see below).
Under our assumptions, M(G) = ((G) ∪ H3 )/G is a 3-manifold with boundary
∂M(G) = (G)/G, if (G) = ∅; π1 (M(G)) ∼ = G. Each component of ∂M(G) is
a Riemann surface; its complex structure is inherited from that of S2 . On the other
hand, the interior of M(G) has a complete hyperbolic structure. While ∂M(G) is “at
∞” in the metric, it is closely related to M(G) by virtue of the hyperbolic geometry
of lines and planes in H3 .
Chapter 9. Deformations of Kleinian groups 413
The modern theory was ushered in by
Theorem 2.1 (Ahlfors’ Finiteness Theorem [2]). If G is finitely generated, ∂M(G)

has at most a finite number of components. Each component is a closed Riemann
surface with at most a finite number of punctures (and, if elliptics are allowed, a finite
number of cone points).
The group G and manifold M(G) are called geometrically finite if G has a finite
sided fundamental polyhedron in H3 ; in terms of M(G) this means that the manifold is
“essentially” compact, that is, compact except for cusps (see below). Automatically,
the interior of a geometrically finite manifold is homeomorphic to the interior of a
compact 3-manifold, possibly with boundary. If G and M(G) are not geometrically
finite, they are called geometrically infinite.
2.1 The 3-dimensional point of view

A simple loop γ ⊂ ∂M(G) is called compressible if it bounds an essential disk
D ⊂ M(G). This is an embedded disk with the properties (i) it is not homotopic,
relative to γ , into ∂M(G), and (ii) D ∩∂M(G) = ∂D. It is the famous Dehn’s Lemma
and Loop Theorem that tells us that γ is compressible if and only if it is homotopic to
a point in M(G) but not in ∂M(G).
The 3-manifold M(G) is said to be boundary incompressible if the inclusion
π1 (∂M(G)) → π1 (M(G)) is injective on each boundary component. Equivalently,
M(G) is boundary incompressible if and only if each component of (G) is simply
connected: There are no compressible loops.
We will also have cause to refer to essential cylinders C embedded in M(G). This
means that (i) C is not homotopic, relative to ∂C, into ∂M(G), and (ii) C ∩ ∂M(G) =
∂C. Thus an annular region in the boundary is not an essential cylinder. Essential
cylinders arise from two disjoint loops on ∂M(G) which are freely homotopic in
M(G) but not in ∂M(G).
A manifold without essential cylinders is called acylindrical.
The role of parabolics. Before continuing with our story we have to come to terms
with the special role played by parabolic transformations. Suppose ζ is the fixed
point of a parabolic element of G. The group of all elements of G that fix ζ is
either conjugate to z → z + 1, or is a free abelian group of rank two conjugate to
z → z + 1, z → z + τ , for some τ with Im τ > 0.
Within H3 there is a certain “universal horoball” at ζ which is preserved by the
parabolic group but mapped disjoint to itself by every other member of G. The same
size horoball works for all groups. In the rank two case, the boundary horosphere
projects to a flat torus within M(G) called a cusp torus. It bounds the projection of the
horoball which is called a solid cusp torus; topologically it is {w : 0 < |w| ≤ 1} × S1 .
In the rank one case, the projection of the horosphere called a cusp cylinder. It
bounds in M(G) the projection of the horoball which is called a solid cusp cylinder:
414 Albert Marden
topologically it is just {w : 0 < |w| ≤ 1} × (−∞, +∞). In the good cases, in

particular in geometrically finite groups, the solid cusp cylinder can be replaced by
a solid pairing tube. This arises from a pair of punctures on ∂M(G), paired by the
following phenomenon. Take small simple loops surrounding each of the punctures.
there is a cylinder C ⊂ M(G), bounded by these two loops, such that C is relative
boundary of solid pairing tube of the form {w : 0 < |w| ≤ 1}×[−1, +1]. Up in (G)
the parabolic fixed point ζ supports a double horodisk: a pair of invariant, mutually
disjoint disks, tangent at ζ ; the pair is bounded by a double horocycle.
Within the corresponding manifold, these two structures are referred to as rank one
cusps and rank two cusps.
For a simple, explicit example consider the group G = z → z + 1; C/G is
a twice punctured sphere. The horocycles {z : Im z = ±1} project to simple loops
about the two punctures. They bound a pairing cylinder C formed by the projection
of a piece of each vertical halfplane based on the horocycles, and a section of the
horosphere t = 1 in UHS.
The Ahlfors Finiteness Theorem can be extended to state that a finitely generated
group also has at most a finite number of conjugacy classes of rank one and rank two
parabolic subgroups, see [27], [29].
Geometrically finite manifolds. We have already introduced the class of geometri-
cally finite groups/manifolds. The central role this class plays in the theory is confirmed
by the following theorem with details provided in [11], [12].
Theorem 2.2 (The Density Theorem). Geometrically finite groups are dense in all
finitely generated groups.
The topology used here is the topology of algebraic convergence to be introduced

in §4 below.
In a geometrically finite manifold all the punctures on ∂M(G) are arranged in pairs,
paired by solid pairing tubes. The puncture pairs are in one-one correspondence with
the conjugacy classes of rank one parabolic subgroups of G. In fact a geometrically
finite manifold is characterized by the following fact: It has a finite number of solid
cusp tori and solid pairing tubes such that when their interiors are removed, there
results a compact manifold, which we have denoted by M0 (G). In the presence of
cusps, the terms boundary incompressibility and acylindricality are often applied not
to M(G) itself but to the compactified M0 (G).
Finite volume manifolds are geometrically finite. In particular, they cannot have
any rank one cusps. Typical examples are most knot complements, see §2.4.
Geometrically infinite manifolds. The basic tool in exploring these is the following:
Theorem 2.3 (Scott Compact Core [56], [55]). Assume G is finitely generated. There
exists a compact, connected submanifold C ⊂ Int(M(G)) with the properties (i) the
inclusion π1 (C) → π1 (M(G)) is an isomorphism, and (ii) each complementary
component of Int(C) in Int(M(G)) is bounded by a single component of ∂C.
The submanifold C = C(G) is called a compact core. The existence of the core,
which was discovered independently by Shalen, was initially used for the purpose of
proving that G is finitely presented.
Each boundary component S of C corresponds to the end ES of M(G) determined
by the complementary component of C bounded by S. The end ES is called geomet-
rically finite it there is a boundary component R of M(G) such that ES ∼ = S × [0, 1)
is bounded by S and R: S is parallel to R. But here we are interested in the geometri-
cally infinite case. The boundary component R that “should be” parallel to S has gone
missing. What is then the topology of ES ? Has the missing boundary component R
left some trace?
When there are parabolics, one often uses instead of the compact core as above, a
different compact core that takes account of the effect of parabolics.
We will now use the notation M0 (G) for the result of removing the (open) solid
cusp tori and solid cusp cylinders from M(G), which may not be geometrically finite.
The relative compact core Crel = Crel (G) [36] is constructed with respect to M0 (G)
to have the following properties: (i) each torus component of ∂M0 (G) is a component
of ∂Crel , (ii) each cylinder component of ∂M0 (G) intersects ∂Crel in a closed annular
region, (iii) the inclusion π1 (Crel ) → π1 (M(G)) is an isomorphism, and (iv) each
complementary component in Int(M0 (G)) is bounded by a single component of ∂Crel .
The fundamental theorem of geometrically infinite manifolds is this:
Theorem 2.4 (Tameness Theorem [1], [13]). For any finitely generated group G, the
interior of M(G) is homeomorphic to the interior of a compact manifold.
In particular each end ES is topologically S × [0, 1). The Tameness Theorem is far
from obvious (think of wild spheres, etc.) and resolves the long standing conjecture
called the Marden conjecture. It has many ramifications in the theory, not least in
providing the final step needed the confirm the Ahlfors conjecture.
The Density Theorem 2.2 was first formulated as a conjecture by Bers for surface
groups but quickly gained notoriety in the trade as applicable to all finitely generated
groups. It was finally proved at the end of a long line of important preliminary results
by many authors. It is a deep result as in its full implication it needs the Tameness
Theorem and the Ending Lamination Theorem. We will not discuss the latter here
but remain content to note the following. Thurston originally discovered, in certain
cases, that each “missing” boundary component of a geometrically infinite manifold
is characterized by an ”ending lamination”, that can be realized as a lamination on the
relevant boundary component S of the core C(G). He conjectured that this situation
was true in general. His vision has now been confirmed in full generality, using Min-
sky’s bilipschitz model of hyperbolic manifolds [48], by Brock, Canary and Minsky
[11], [12]. The work also requires a deep study of the complex of simple curves on
a surface carried out by Masur and Minsky. Rather than itemize all the contributions
here we refer to the cited references for a full account and bibliography. We will say
a bit more on this subject at the end of §5.
416 Albert Marden
Extending boundary mappings. Suppose F : (G) → (H ) is a quasiconformal

mapping that induces an isomorphism ϕ : G → H and f : ∂M(G) → ∂M(H )
is its projection. Does f , or a mapping homotopic to f , have an extension to a
quasiconformal map M(G) → M(H )? This is not true in general but it is true
for geometrically finite manifolds M(G). However, there does not seem to be a
“canonical” way to construct the extension. Instead, the extension can be constructed
by topological means. Assume first that M(G) is compact.
Denote the components of ∂M(G) by {Ri }, and the components over them by
{i,j } ⊂ (G). Denote the stabilizer of i,j by Gi,j . Then Gi,j ∼ = π1 (Ri )/πc (Ri )
where πc (Ri ) ⊂ π1 (Ri ) denotes the normal subgroup generated by free homotopy
classes of compressing loops on Ri . The subgroup πc (Ri ) is also the kernel of the
inclusion map ι : π1 (Ri ) → π1 (M(G)).
Citing [31, Theorem 13.9, Corollary 13.7], f is homotopic on ∂M(G) to a home-
omorphism f1 which has an extension to a homeomorphism between the manifolds
M(G) → M(H ). According to [50], f1 in turn is homotopic to a (quasiconformal)
diffeomorphism f2 . Now f2 is homotopic to f on ∂M(G). Choose the lift F2 of f2
to H3 ∪ (G) so that F2 also induces ϕ and is homotopic to F on (G). Now F2 has
a quasiconformal extension to all S2 .
If there are parabolics we have to replace the manifold by the compact manifold
M0 (G) resulting from the removal of the solid pairing tubes and the solid cusp tori.
The resulting mapping f2 sends the pairing cylinders and cusp tori on M(G) to those
on M(H ). It then needs to be extended inside the solid pairing tubes and cusp tori.
Although it would suffice in the applications to replace (F, f ) by (F2 , f2 ), we will
present a stronger result.
We started above with a ϕ-equivariant quasiconformal map F : (G) → (H ),
We found a ϕ-equivariant map F2 of S2 whose restriction to (G) is homotopic to F .
Using the density of the loxodromic fixed points in (G) one can verify:
Lemma 2.5. F has an extension to a homeomorphism of S2 that satisfies F (ζ ) =

F2 (ζ ) for all ζ ∈ (G).
Set H = F2 −1 F : (G) → (G). The map H is homotopic to the identity

on each component of (G), induces the identity automorphism of G, and is equal
to the identity on (G). Let γz be a shortest geodesic from z to H (z). There is a
constant C1 < ∞ such that Lh (z) = dh (z, H (z)) < C1 for all z ∈ (G) (lift from
the quotient). Here dh ( ·, ·) denotes the shortest hyperbolic distance on (G). From
this it follows if z → ζ ∈ (G) in the spherical metric, then lim γz = ζ uniformly on
γz . That is, there exists a constant C2 such that d(γz , (G)) < C2 for all z ∈ (G).
Here d( ·, ·) denotes spherical distance. It follows that d(w, (G)) < C2 d(z, (G),
for all w ∈ γz and some C2 < ∞ (actually we only need these estimates for z ∈ (G)
near a point ζ ∈ (G).
I am grateful to Vlad Markovic for allowing inclusion of his unpublished result as
follows.
Proposition 2.6 (Markovic). H is quasiconformal on S2 ; hence F itself is the restric-

tion to (G) of an equivariant quasiconformal map of S2 .
Proof. Markovic’s proof is as follows. Set X = (G). From [51] we know that X
has the property of uniform perfectness. That is, for the hyperbolic metric ρ(w)|dw|
in each component of (G) and some constant C3 > 0,
C3 |dw| 2|dw|
< ρ(w)|dw| < .
d(w, X) d(w, X)
Upon integrating over a shortest geodesic γz of hyperbolic length Lh (z) from z to H (z),
we find that C3 d(z, H (z)) ≤ Lh (z) supw∈γz d(w, X) < Lh C2 d(z, X). In other terms,
d(z, H (z)) < C4 d(z, ζ ) for any ζ ∈ X. Now d(H (z), ζ ) ≤ d(H (z), z) + d(z, ζ ).
Consequently for some constant C5 , d(H (z), ζ ) < C5 d(z, ζ ). The same holds if we
replace z by H (z). We conclude that
d(H (z), ζ )
≤ d(z, ζ ) ≤ C5 d(H (z), ζ ).
C5
So the ratio of distances to ζ is uniformly bounded between 0 and ∞ as z → ζ . We
are now in position to apply the geometric definition of quasiconformality to show
that H (ζ ) is quasiconformal at ζ . Since ζ was arbitrarily chosen this proves H is
quasiconformal on (G).
After applying Theorem 2.9 (iii) we end up with the following useful results.
Theorem 2.7. Suppose G is geometrically finite, and F : (G) → (H ) is a qua-

siconformal map inducing an isomorphism ϕ : G → H . Then F is the restriction
of a ϕ-equivariant quasiconformal map F : H3 ∪ S2 → H3 ∪ S2 . It projects to a
quasiconformal map f : M(G) → M(H ).
Corollary 2.8. With the notation of Theorem 2.7, assume H = G and ϕ = id (ϕ =

id if and only if F pointwise fixes (G)). Then f : Int(M(G)) → Int(M(H )) is
homotopic to the identity map.
Corollary 2.8 is significant in the case that ∂M(G) is compressible and f is not
homotopic to the identity as a self-mapping of each component of ∂M(G).
Instead of a mapping between boundary components, we will now start afresh with
a quasiconformal mapping F : S2 → S2 that induces an isomorphism ϕ : G → H
satisfying F (g(z)) = ϕ(g)F (z) for all g ∈ G, z ∈ S2 ; here G and H are not necessarily
geometrically finite. If the restriction of F to (G) is conformal (G) → (H )
then F is Möbius and the two groups are conjugate. Here we are again using the fact
that (G) has zero area. We now have a choice of extension theorems to H3 and
M(G).
418 Albert Marden
Theorem 2.9 (Basic Extension Theorems). Suppose G, H are arbitrary Kleinian

groups and F : S2 → S2 is an equivariant K-quasiconformal mapping that induces
an isomorphism ϕ : G → H . Then:
(i) [18], [67] The map F has an equivariant extension to H3 that is a homeomorphism
which also induces ϕ; its projection f : M(G) → M(H ) is an orientation
preserving homeomorphism.
(ii) [66] The map F has an equivariant (L, a)-quasi-isometric extension for some
L = L(K), a = a(K); its projection f : M(G) → M(H ) is a (L, a)-quasi-
isometric mapping.
(iii) [39, Corollary B.23] The map F has an equivariant extension to a K 3/2 -bilipschitz
C 1 -diffeomorphism of H3 ; its projection f : M(G) → M(H ) is a K 3/2 -bilip-
schitz C 1 -diffeomorphism (and hence a quasiconformal map).
A mapping f of H3 is called (L, a)-quasi-isometric if there exist finite constants

1 ≤ L and a ≥ 0 such that in the hyperbolic metric
1
d(x, y) − a ≤ d(f (x), f (y)) ≤ Ld(x, y) + a.
L
Thus a quasi-isometric map need not be continuous but at long range it is essentially
bilipschitz. It is L-bilipschitz if a = 0. Quasi-isometric maps can be extended to ∂ H3
and the extension is a quasiconformal mapping of S2 .
2.2 The classical examples

Fuchsian and quasi-Fuchsian groups. A Fuchsian group is a discrete group G that
preserves a round disk in S2 , which we may regard as H2 . If R = H2 /G then
M(G) ∼ = R × [0, 1]. One can see the fibering explicitly in the upper halfspace model
by taking the family of euclidean halfplanes in upper halfspace, bounded by the real
axis. This is also a good place to see the pairing of punctures between R and R , the
G-quotient of the exterior of the disk or upper halfplane used to form R.
A quasi-Fuchsian group is a quasiconformal deformation of a Fuchsian group
G: there exists a quasiconformal map f : S2 → S2 that induces an isomorphism
ϕ : G → H by f g(z) = ϕ(g) f (z) for all z ∈ S2 and all g ∈ G. Such f are
characterized by their Beltrami differentials μ = fz /fz with μ(g)g /g = μ for all
g ∈ G and a.e. z ∈ S2 . It is a classical result of Bowen and Sullivan that the limit set
(G) is the quasiconformal image of a circle and has Hausdorff dimension strictly
between one and two, unless it is again a circle. The 3-manifold M(G) retains the
product structure R × [0, 1]. Conversely, if a manifold M(G) has such a product
structure, G is quasi-Fuchsian.
Schottky groups. These are groups generated as follows. Take an even number of
round disks in C with disjoint closures, Arrange them in pairs. For each pair choose
a Möbius transformation Ti that sends the exterior of one disk to the interior of its
partner. The group G = T1 , . . . , Tg is called a Schottky group. More generally,

the disks need only be regions bounded by Jordan curves but one must postulate the
existence of pairing transformations that take the exterior of one Jordan curve to the
interior of its partner. Schottky groups are free groups and, if g ≥ 2, (G) is totally
disconnected of positive Hausdorff dimension.
If G is Schottky, M(G) is a handlebody bounded by the closed surface R =
(G)/G of genus g ≥ 1 (g = 1 arises from a cyclic group for which M(G) is
directly seen to be a solid torus – a bagel). A handlebody of genus g is a compact
3-manifold such that there are g mutually disjoint compressing curves on R bounding
mutually disjoint disks in M(G). If M(G) is cut along the disks, there results a
topological ball. conversely if an M(G) has this structure, G is Schottky – but is
not necessarily generated from pairs of circles. For a full analysis of circle generated
Schottky groups see the recent paper [28].
2.3 The convex core
The convex core C(G) of M(G) is the smallest (hyperbolically) convex submanifold
of M(G) that contains all the geodesics. It may well coincide with M(G).
The convex core is constructed in H3 as the convex hull C ∗ (G) of (G) and then
projected to M(G). The construction is as follows.
Before introducing a group, consider a region ⊂ C without isolated boundary
points, = C. A maximal disk D ⊂ is a round disk that is not contained in
any larger disk. Its closure intersects ∂ in at least two points. We will call the
circle C = ∂D a maximal circle. Erect the hyperbolic plane PC ⊂ H3 (in the upper
halfspace model) rising from C. It divides H 3 into two halfspaces. Denote by H (PC )
the halfspace bordering S2 \ D. The convex hull of the complement in S2 of is
defined to be

C∗ = H (PC ),
C
the intersection being taken over all maximal circles C in .

The relative boundary Dome() = ∂C ∗ ⊂ H3 lies over as the dome lies over
the floor in a domed stadium. It is the union of “flat pieces” and “bending lines”.
Each flat piece is an open hyperbolic polygon, possibly with infinitely many edges,
contained in some plane PC . The components of the complementary closed set are
geodesics with endpoints in ∂(G). They are called bending lines.
An isolated bending line (isolated on both sides) is the common edge of two flat
pieces. There is a well determined (exterior) bending angle 0 < θ < π measured
so that the limiting case of zero bending means there is no bending at all. There is a
bending measure μ defined on transverse segments τ where in our applications we can
take τ as a geodesic segment with endpoints in flat pieces. The measure is constructed
by taking the limit of the bending of finite approximates to the dome [20].
420 Albert Marden
Suppose now that is simply connected. The ambient hyperbolic metric in H3

restricts to a path metric on Dome(), referred to as its hyperbolic metric. A result
of Thurston is that there is an isometry of Dome() onto H2 , uniquely determined up
to postcomposition with a Möbius transformation; it may be regarded as the Riemann
map.
There is a canonical map r(z) : → Dome(), called the nearest point retraction
that relates to its dome: Given z ∈ , consider the family of horospheres in H3
based at z. There is one that meets the dome at exactly one point, which is defined
to be r(z). If there are no isolated bending lines, r is a homeomorphism. If there
is an isolated bending line with bending angle α then there are two planes P1 , P2
that intersect with angle α along . Denote by P1 , P2 the planes orthogonal to P1 , P2
along . The the exterior bending angle α is also the vertex angle of the crescent in
determined by P1 , P2 . This is the crescent which is the preimage of under r. The
retraction r extends to ∂ = ∂ Dome() and fixes every point.
The nearest point retraction r is known to be 2-lipschitz in the respective hyperbolic
metrics [23]. If in addition is euclidean convex, r is also 2-quasiconformal [24].
Finally we will bring in a Kleinian group G. Construct the convex hull C ∗ (G) of
(G) by repeating our construction on each component of (G), if (G) = ∅. Then
C ∗ (G), like (G), is invariant under G. Its projection C(G) to M(G) is called the
convex core of M(G). The projection of r maps ∂M(G) onto the relative boundary
∂C(G) ⊂ Int(M(G)). If ∂M(G) = ∅ then we set C(G) = M(G).
Chris Bishop used Dome() to develop a new way to numerically estimate the
Riemann map of a simply connected → D. He makes use of the crescent system
under a finite approximation to the dome to deform back to D [7].
A manifold is geometrically finite if and only if its convex core has finite volume.
2.4 The Hyperbolization Theorem

In the 60s and early 70s as the theory was being developed, the unsettling question
that increasingly came to the fore was this. Is the class of hyperbolic manifolds a
rather sporadic class within all 3-manifolds, or does it represent a large class? The
answer was provided by Thurston in the mid 70s by bringing in startlingly original
and powerful new tools.
His Hyperbolization Theorem concerns orientable compact 3-manifolds M 3 with
possible boundary with the following two properties:
• Every embedded 2-sphere bounds a ball (irreducibility).
• Every map f of a torus T into M 3 is homotopic to a map of T into ∂M 3 , provided
f∗ : π1 (T ) → π1 (M 3 ) is injective (atoroidality).
It is known every M 3 may be cut along a finite number of spheres and incompressible
tori so that the resulting pieces have these two properties, after capping off the exposed
spheres with balls.
Theorem 2.10 (The Hyperbolization Theorem). There are two cases.

(i) If ∂M 3 = ∅ then the interior of M 3 has a complete, geometrically finite, hyper-
bolic structure.
(ii) If M 3 is a closed manifold that contains an incompressible surface then M 3 has
a hyperbolic structure.
An incompressible surface is one containing no compressing loop bounding a disk

in its complement. There exist such surfaces whenever π1 (M 3 ) is “sufficiently large”,
which is the case if the first homology group contains infinitely many elements.
For Thurston’s exposition see [63] and for complete accounts of the proof see [34],
[52], [53].
It is conjectured that a closed, irreducible manifold M 3 is hyperbolic whenever
π1 (M 3 ) contains no rank two abelian subgroups. Confirmation of this would com-
pletely decide the hyperbolization question.1
There is a sharper form of the theorem that allows specification of rank one
parabolic subgroups – pairs of punctures.
A striking corollary drawn by Thurston is that a knot complement in S3 has a
hyperbolic structure if and only if it is not a torus knot or a satellite knot.
2.5 Uniqueness
The following theorem appeared not long after Ahlfors’ Finiteness Theorem. It was
quite surprising at the time as hyperbolic surfaces of finite area are most definitely not
rigid.
Theorem 2.11 (Mostow’s Rigidity Theorem). If M(G) has finite volume and
ϕ : G → H is an isomorphism to a Kleinian group H , then ϕ is induced by an
isometry M(G) → M(H ).
Mostow originally proved his theorem [49] for closed manifolds. It was extended
to finite volume manifolds independently by [42] (in dimension 3) and [54] (in n ≥ 3
dimensions).
Taking account of boundary components, Mostow’s theorem was generalized to
manifolds with boundary in [42] as follows:
Theorem 2.12. Suppose G is a geometrically finite Kleinian group and F : (G) →

(H ) is a conformal map that induces an isomorphism ϕ : G → H by F g(z) =
ϕ(g) F (z) for all z ∈ (G) and g ∈ G. Then F is the restriction of a Möbius
transformation and M(H ) is isometric to M(G).
The proof uses the fact that F has an extension to H3 that in turn extends to a
quasiconformal map of S2 . This was discussed in §2.1.
1 Added in proof. As a consequence of Grigori Perelman’s recent work, this is now known to be true.
422 Albert Marden
3 Boundary deformations of geometrically finite groups
In this section we will work with a geometrically finite M(G) with nonempty boundary.
Its boundary is the union of a finite number of Riemann surfaces {Ri }, each of which
is a closed surface of genus gi ≥ 0, with ni ≥ 0 punctures. Since each component of
(G) has a hyperbolic metric the same is true of each Ri .
As we have seen elsewhere in this volume, the Teichmüller space T (Ri ) is a
complex manifold of dimension 3gi + ni − 3. This is positive except in the case that
Ri is the triply punctured sphere.
We would like to apply basic Teichmüller theory in succession to each component
of ∂M(G). Namely suppose f is a quasiconformal map of each component Ri to
another Riemann surface. Let μi denote its Beltrami differential on Ri . Provided
each component of (G) is simply connected and hence serves as the universal covers
of the {Ri }, we can lift the {μi } to a Beltrami differential μ on (G) by setting
μ(π −1 )(z) = μi (z), if z ∈ Ri . The collection μ of the lifted {μi } on (G) satisfies
μ(z)ḡ /g = μ(z) for all g ∈ G and a.e. z ∈ S2 ,
Solve the Beltrami equation on S2 , using the fact the limit set has zero area. The
solution F is quasiconformal on S2 and is uniquely determined up to postcomposi-
tion by a Möbius transformation, which can be used to normalize: we may assume
(0, 1, ∞) ∈ (G) and require that F fix each of these points. The map F induces
an isomorphism θ : G → H to another group H , which necessarily geometrically
finite as well. In view of Proposition 2.7 we may assume the restriction of F to
(G) projects to a quasiconformal map f : ∂M(G) → ∂M(H ) that extends to a
quasiconformal map M(G) → M(H ).
On the one hand if we regard ∂M(G) simply as a union of Riemann surfaces with
no particular relation to the interior of the manifold or to deformations of G itself then
we would find that the Teichmüller space of ∂M(G) is simply the abstract product
of the Teichmüller spaces of the individual boundary surfaces. When all components
of (G) are simply connected, they are representations of the universal covers of
the components of ∂M(G). The classical theory is then operative in that nontrivial
deformations of components of ∂M(G) correspond to nontrivial deformations of G.
So all is fine if M(G) is boundary incompressible. Otherwise we have a little
problem.
Consider an infinitely connected component i,j over Ri that is stabilized by
Gi,j . For example i,j could be the ordinary set of a Schottky group. Consider a
compressing loop γ ∗ ⊂ i,j , the lift of a compressing loop γ ∈ Ri ⊂ ∂M(G),
For example γ may be part of a generating set of π1 (Ri ), say a loop about a handle.
It certainly figures in the deformation of the Riemann surface Ri . But since γ ∗
determines the identity element of Gi,j ⊂ G it can play no role at all in the deformation
of G.
Fix a basepoint O ∈ Ri for each i. Consider the normal subgroup πc (Ri ) ⊂ π1 (Ri )
consisting of loops, basepoint O, freely homotopic to compressing loops of Ri . The
quotient group π1 (Ri )/πc (Ri ) is isomorphic to the stabilizer Gi,j of any component
i,j of (G) over Ri . A automorphism h of Ri lifts to a component i,j if and only

if h∗ (πc (Ri )) = τ πc (Ri )τ −1 , for some arc τ .
The way out of the conundrum has already been suggested in our discussion in §2.1
of extending boundary maps, especially Corollary 2.8. We will consider quasiconfor-
mal mappings F : (G) → (H ) that induce isomorphisms θ : G → H in the sense
that F g = θ(g) F for all g ∈ G. We may assume by Proposition 2.7 that F extends
to a quasiconformal mapping of S2 . Denote the projection by f : ∂M(G) → ∂M(H ).
3.1 The quasiconformal deformation space
Define the quasiconformal deformation space of G to be
T(G) = {θ : G → H is an isomorphism induced by a qc map F : S2 → S2 }/ ≡ .
The quasiconformal map F satisfies F g(z) = θ (g) F (z) for all g ∈ G and z ∈ S2 .
Here “≡” means that mappings F that give rise to conjugate isomorphisms are to be
identified.
We had normalized quasiconformal mappings F of S2 by replacing G by a conju-
gate so that (0, 1, ∞) ∈ (G) and then requiring that F fix these points. Equivalently
then, T(G) can be defined in terms of equivalence classes of normalized quasiconfor-
mal mappings F of S2 which induce isomorphisms θ : G → H where the equivalence
relation is: F1 ∼ F2 if and only if the projection f2−1 f1 : ∂M(G) → ∂M(G)
extends to a mapping Int(M(G)) → Int(M(G)) homotopic to the identity.
Alternatively, F1 ∼ F2 (i) if and only if F1 and F2 induce the same isomorphism
θ : G → H , or (ii) if and only if F1 (ζ ) = F2 (ζ ) for all ζ ∈ (G). In the latter case,
if F2−1 F1 is conformal on a component of (G), it is the identity there, since it
extends to the identity on ∂. For further details consult [46].
We have yet to explain the relation of T(G) to the product of the classical Teich-
müller spaces of the components: T (R1 ) × · · · × T (Rk ).
To do this we have to recognize that a stronger equivalence can be imposed:
F1 F2 if and only if f2−1 f1 : ∂M(G) → ∂M(G) is homotopic to id.
It follows that F1 ∼ F2 in the earlier definition. These two equivalences differ only
when ∂M(G) is compressible.
To mirror the difference in the two equivalence relations we introduce the
group X(G) consisting of equivalence classes of quasiconformal automorphisms
h : ∂M(G) → ∂M(G) which extend to maps M(G) → M(G) that are homotopic
to the identity on the interior – Corollary 2.8. Two such maps h1 , h2 are equivalent
h1 h2 if and only if h−12 h1 is homotopic to the identity on ∂M(G); specifically,
h−1
2 h 1 maps each component Ri onto itself and is homotopic on Ri to the identity.
If M(G) is boundary incompressible, X(G) = id.
424 Albert Marden
Denote by Mod0 (Ri ) the group of homotopy classes of quasiconformal mappings

h : Ri → Ri which extend to M(G) to be homotopic in Int(M(G)) to the identity
map of the interior. If the i,j are simply connected, Mod0 (Ri ) = id. Such a map h
fixes the punctures on Ri , and the set of compressing loops. Extend h to all ∂M(G)
by setting it equal to the identity on Rm , m = i. Then h extends to M(G) and (its
restriction to the interior) is homotopic in Int(M(G)) to the identity. In other terms,
h is the projection of a quasiconformal automorphism H of each component i,j
over Ri with the property that H induces the identity automorphism of Stab(i,j ) and
extends continuously to the identity map of ∂i,j . A typical element of Mod0 (Ri )
arises from a Dehn twist h about a compressing curve γ of some Ri ; γ bounds a disk
D within M(G). Thicken D to D and extend h, first to D so that it is the identity on
each face of D , and then to all M(G).
Consequently the group X(G) splits into a direct product
X(G) = Mod0 (R1 ) × Mod0 (R2 ) × · · · × Mod0 (Rk ).
The group Mod0 (Ri ) acts without fixed points on T (Ri ). For suppose, for example,
that h ∈ Mod0 (Ri ) fixes the origin (Ri , id) in T (Ri ). Then h is homotopic to a
conformal map h0 : Ri → Ri . Now h and then h0 lift to automorphisms h∗ and h∗0
of i,j over Ri ; we can choose h∗ to be homotopic in i,j to h∗0 . We know that h∗
extends continuously to ∂i,j and fixes every point. The same is therefore true of h∗0 .
But h∗0 is the identity since it is a conformal automorphism. Therefore h∗ is homotopic
in i,j to the identity and h is homotopic in Ri to the identity.
Finally classical results using a projection from the space of Beltrami differentials
with respect to G on (G) imply that T(G) is a complex analytic manifold.
As a consequence of our discussion, we arrive at a 3D interpretation of the following
result proved earlier in terms of Fuchsian equivalents of the surfaces {Ri } by Bers [3],
Maskit [46], and Kra [35]:
Theorem 3.1. Suppose G is geometrically finite. Denote the components of ∂M(G)

by {Ri }. Then
T(G) = T (R1 )/ Mod0 (R1 ) × · · · × T (Rk )/ Mod0 (Rk ).
Here Mod0 (Ri ) is the fixed point free subgroup of biholomorphic automorphisms of
T (Ri ) generated by automorphisms of Ri that extend to be homotopic to the identity
in the interior of M(G).
The product T (R1 )×· · ·×T (Rk ) is the universal cover of T(G); they are identical
if and only if ∂M(G) is incompressible. The spaces T(G) and T (Ri )/ Mod0 (Ri ) are
complex analytic manifolds; T(G) has dimension
m

(3gi + ni − 3),
i=1
where gi is the genus of the i’th component of ∂M(G) and ni is the number of its
punctures.
If M(G) is geometrically infinite and boundary incompressible Theorem 3.1 still

holds; in this case Mod0 (Ri ) = id. Thus if (G) is connected and simply connected (a
singly degenerate group in ∂B(G) – see §6.1) then T(G) has dimension (3g + n − 3).
In contrast if G1 is Fuchsian with H2 /G1 conformal to (G)/G then T(G1 ) has
dimension (6g + 2n − 6). If all components Ri are triply punctured spheres, G is
quasiconformally rigid, geometrically finite or not.
Remark 3.2. Suppose all components i are simply connected but that there may be
torsion in Gi . Then in addition to the punctures on each component Ri ⊂ ∂M(G)
there will be bi ≥ 0 “branch” or “cone” points. In this case the dimension count will
be
(3gi + bi + ni − 3).
For it is an interesting fact that T (Ri ) is biholomorphically equivalent to T (Ri ) where
Ri is the result of removing the cone points. That is, the dimension is the same whether
you have bi + ni punctures, or ni punctures and bi cone points.
The basis for the equivalence is the following fact [41], [5]: A homeomorphism
f : Ri → Ri lifts to a homeomorphism f ∗ of i which induces the identity automor-
phism of Gi if and only if f is homotopic in Ri to the identity map. This follows from
the fact that γ is freely homotopic in Ri to f (γ ) for all simple loops γ ⊂ Ri .
There is an analogue of Theorem 3.1 that covers the case that G has elliptics; see
[46], [35] for details.
4 Geometric deformations of the groups;

the representation variety
Assume G is geometrically finite. What we will call its representation variety (actually
it is the quotient of an open subset of an affine algebraic variety) consists of all its type
preserving homomorphisms onto nonelementary subgroups of PSL(2, C):
R(G) = {type preserving homomorphisms of G to nonelementary groups}/ ≡
A homomorphism ϕ is called type preserving if ϕ(g) is parabolic whenever g ∈ G is so.
Here “≡” means that two homomorphisms ϕ, ϕ1 are identified if ϕ1 (G) = T ϕ(G)T −1 ,
for some Möbius transformation T ; we do not wish to distinguish between conjugate
groups. In this general context elementary groups are those with the property that any
two elements of infinite order have a common fixed point.
We are assuming the targets are nonelementary because it is natural to do so in
the context of the theory of Kleinian groups, and also because, with this assumption,
R(G) is a Hausdorff space. The largest class of groups left out are purely elliptic
groups, that is, groups conjugate to a group of rotations of S2 .
The group G has a finite presentation. We can imagine changing the entries in
the generating matrices in such a way that the new matrices continue to satisfy the
426 Albert Marden
relation equations. Doing so will give a homomorphic image. The image group H is
not likely to be discrete, or even finitely presented, unless we remain close enough to
the initial G.
The natural topology in R(G) is given by the topology of algebraic conver-
gence. A sequence {T1,k , . . . , TN,k } of N -generator groups converges, necessarily
to T1 , . . . , TN , if and only if limk→∞ Ti,k = Ti is a Möbius transformation for all i.
For our purposes the most important part of R(G) is its discreteness locus. This
is the following closed subset:
Rdisc (G) = {θ ∈ R(G) : θ : G → H is an isomorphism to a discrete group H }.
It is closed because of Jørgensen’s theorem [32] that the algebraic limit H of a sequence
of isomorphisms {θn : G → Gn } to nonelementary, discrete groups is likewise discrete
and nonelementary; moreover θ = lim θn : G → H is an isomorphism.
Theorem 4.1. T(G) is the connected component of Int(Rdisc (G)) that contains the
identity.
Theorem 4.1 is a direct consequence of the fact [42] that T(G) is open in R(G)
for a geometrically finite group G. This is because the quasiconformal deformations
of G are “strongly stable” in Bers’ terminology: Given ε > 0 and a quasiconformal
image H = ϕ(G), there is a neighborhood N of H in R(G) such that for each group
H = ρ(G) ∈ N , there is a (1 + ε)-quasiconformal map of S2 that induces the
isomorphism ρϕ −1 : H → G → H . Geometrically infinite groups in Rdisc (G) are
not similarly stable.
In particular, the dimension of R(G) is the same as that of T(G).
Now the image H of an isomorphism θ ∈ Int(Rdisc (G)) is geometrically finite
and is a quasiconformal deformation of G. But the isomorphism θ itself may not be
induced by a quasiconformal map of S2 . Thus the interior of the discreteness locus
may have many components only one of which is T(G).
In fact each component D of Int(Rdisc (G)) is itself the quasiconformal deformation
space of any fixed θ : G → G ∈ D.
The definitive study of the components of Rdisc (G) is carried out in [15] (there,
the notation AH(G) is used for what here is denoted Rdisc (G)). In brief, the number
of components of the interior is determined by the number of different homotopy
equivalences between M(G) and the target manifolds, or equivalently, the number of
essentially different isomorphisms between G and the target groups. In the case G is
without parabolics, and ∂M(G) = ∅, there are only a finite number of manifolds which
are homotopy equivalent but not homeomorphic to M(G) [61]. If G is a Fuchsian
closed surface group, then Int(Rdisc (G)) = T(G).
4.1 Deforming the convex core

Assume for simplicity that M(G) is compact, boundary incompressible, and acylindri-
cal. Recall that corresponding to each component Ri of ∂M(G) is a relative boundary
component Ci of the convex core C(G) parallel to it. Each Ci is a simply connected,
embedded, pleated surface (see §§6.2–6.3) with bending lamination (i , μi ); (, μ)
denotes the vector of individual laminations. Conversely we have the following special
case of a more general result of Bonahon and Otal.
Theorem 4.2 ([10]). Suppose M(G) compact, boundary incompressible and acylin-
drical. Assume we are given a measured lamination (, μ) on ∂M(G). There exists
a new geometrically finite hyperbolic structure M(Gμ ) on M(G) whose convex core
boundary ∂C(Gμ ) has the bending lamination (, μ) if and only if on each closed
leaf γ of , 0 < μ(γ ) < π .
If is the union of a finite number of closed leaves, then Gμ is uniquely determined
by (, μ) up to Möbius equivalence.
Consider the case that the bending lamination consists of a maximal system of
(3gi − 3) simple closed geodesics on the convex hull boundary C(Gμ ); here gi
is the genus of the i’th component
of ∂M(G). Choi and Series [17] show that the
complex lengths of these (3gi − 3)-geodesics serve as parameters for the local
deformations of Gμ in its representation variety R(Gμ ): The lift to H3 of a leaf
determines an element T ∈ Gμ . The complex length of is log k + iθ mod (2π i) if
T : z → keiθ z, k > 1, in normalized form.
As yet, the theory of bending laminations is not complete.
5 The deformation space boundary

The closure T(G) of T(G) in R(G) lies in Rdisc (G). It is an interesting fact [34,
Theorem 8.44], [60] (see also [40, appendix]) that T(G) is the interior of T(G). That
is, T(G) is not like an open ball with a slit to the boundary removed.
Denote the boundary of T(G) in Rdisc (G) by ∂T(G). The study of the boundary
is perhaps the most fascinating aspect of the theory; as we approach a boundary point
from the interior, some sort of degeneration of the manifold must occur. The topology
of T(G) is interesting as well.
In general, T(G) is not compact and not a manifold [40]. However it is compact
if M(G) is acylindrical (Thurston’s Compactness Theorem [64]). For when there are
essential cylinders, the cylinder can be squeezed to a point – the cyclic fundamental
group of the cylinder becomes a cyclic parabolic group. This process destroys G. For
a simple example just squeeze to a point a simple geodesic not parallel to a boundary
component on H2 /F where F is Fuchsian and watch what happens to F .
428 Albert Marden
The topology we have been using in R(G) is the topology of algebraic convergence:
a sequence of N-generator groups converges to a group, if their generators converge
and generate the group. The algebraic limit of a sequence {Gn } of nonelementary
N-generator discrete groups is also discrete and nonelementary [32].
But there is a fundamentally different notion of convergence, first discovered by
Troels Jørgensen. It is akin to “tangential convergence”, thinking of a sequence in
the unit disk converging tangentially to the unit circle, whereas algebraic convergence
is akin to radial convergence. It relates to the convergence of the quotient manifolds
rather than convergence of the generators of the fundamental group.
A sequence of N-generator groups {Gn } is said to converge geometrically to a
group H if and only if the following hold:
(i) If lim gn = h exists, gn ∈ Gn , then h ∈ H .
(ii) If h ∈ H then h = lim gn , gn ∈ Gn .
The group H is called the geometric limit of {Gn }. As was first pointed out by
Jørgensen, H is not necessarily finitely generated!
The basic fact needed to explore the boundary is this:
Theorem 5.1 ([33]). Suppose the sequence {Gn } of N -generator, nonelementary

Kleinian groups converges algebraically to G. Then there is a geometrically con-
verging subsequence {Gk }. The limit H of any geometrically converging subsequence
contains G; consequently M(H ) is a covering manifold of M(G).
Furthermore, there is a sequence of surjective homomorphisms {ψk : H → Gk }
such that lim ψk (h) = h for all h ∈ H .
The following result, part (i) of which builds on prior work of a number of re-
searchers, gives considerable insight into when a geometric limit of N-generator
groups differs from the algebraic. We have formulated it to take account of the Tame-
ness Theorem.
Theorem 5.2. Suppose θn : G → Gn is a sequence of isomorphisms converging

algebraically to H . The sequence also converges geometrically to H under one of the
following situations:
(i) [26] If (H ) = ∅, and G has no “new parabolics”.
(ii) [14] If (G) = ∅.
In each case lim (Gn ) = (H ) in Hausdorff convergence.
No new parabolics means that h ∈ H is parabolic only when θn θ −1 (h) is parabolic

for all large indices. It is not required that the algebraic limit be geometrically finite,
which it is not, in general.
There is a countable number of special points on ∂T(G) called cusps. A cusp is a
geometrically finite boundary group. It comes from pinching simple loops on ∂M(G).
In §6.1 we will discuss this in more detail. A maximal cusp is a cusp H with ∂M(H )
a union of triply punctured spheres. Maximal cusps are rigid.
One of the deepest and most fascinating topics in the theory is to describe the
nature of (i) the geometrically infinite boundary groups, and (ii) the geometric limits
at the boundary. Fundamental in this investigation is the Tameness Theorem 2.4 which
says that each end of Int(M(G)) is homeomorphic to a product S × [0, 1) where S
is a finitely punctured compact hyperbolic surface. An end of a geometrically finite
manifold is a neighborhood of a Riemann surface boundary component; for this case
the conclusion of the tameness theorem has long been known. It is for geometrically
infinite ends that the theorem is not at all obvious.
As already mentioned in §2.1, Thurston discovered the trace of the “missing bound-
ary component” of each infinite end: The end ES is characterized by a sequence of
geodesics {γn } that exit it. As a consequence of the Tameness theorem, the geodesics
{γn } are homotopic to simple geodesics {γn∗ } on S such that the sequence {γn∗ } con-
verges to a geodesic lamination called the ending lamination. When S, which repre-
sents the end, is incompressible the ending lamination is uniquely determined, oth-
erwise what is determined uniquely is a certain equivalence class of laminations.
Thurston conjectured that homeomorphism f : M(G) → M(H ) between two hy-
perbolic manifolds is homotopic to an isometry if and only if corresponding ending
laminations are the same, for the infinite ends, and the restriction of f to the boundary
is homotopic to a conformal map ∂M(G) → ∂M(H ) for the others. The recent
confirmation of this Ending Lamination Conjecture was a tour de force due to Yair
Minsky [48] together with Brock, Canary, and Masur, see [11], [12] and the references
there.
On the other hand, Soma [58] completely described the topological structure of
the geometric limits on the boundary of quasi-Fuchsian space of a closed surface.
A similar analysis should work in the presence of parabolics and more generally for
geometric limits at boundary cusps on the deformation space of any geometrically finite
group (personal communication). Soma showed that “anything goes” as topologically
allowed by the situation. At each cusp there is in general a huge range of different
geometric limits, including many that are infinitely generated. It seems likely that the
methods of Minsky et al will give additional insight into the geometric structure of the
geometric limits.
6 Deformation of surface groups
In the analytic theory, perhaps the most useful realization of Teichmüller space T (R)
is as the Bers embedding onto a certain open subset of a Banach space of quadratic
differentials on R. Yet viewing T (R) in this way hides a lot of additional structure that
goes with identifying the images in the Banach space with a space of quasi-Fuchsian
groups.
430 Albert Marden
Fix a hyperbolic Riemann surface R = H2 /G. In this section we will present three
methods for deforming the Fuchsian group G and surface R within R(G). The first is
analytic in character, using quasiconformal mappings and then projective structures.
In this approach, T (R) will appear as a “Bers slice” in the quasiconformal deformation
space T(G). The second and third approaches are geometric; the hyperbolic structure
on R will be deformed by earthquaking or more generally, by complex scaling.
6.1 Bers slices

We will obtain our “slice” by the method of simultaneous uniformization discovered
by Bers.
Fix a Fuchsian group G acting in the upper halfplane (UHP) and of course in LHP
as well. The anticonformal involution J ∗ (z) = z induces an anticonformal involution
J : R top = UHP/G ↔ Rbot = LHP/G. The involution J extends to a anticonformal
involution of M(G) interchanging the two boundary components and inducing the
identity automorphism of π1 (M(G)) ≡ G.
Suppose ϕ : G → H ∈ T(G) is induced by the quasiconformal map F of S2 .
Denote its projection by f : ∂M(G) → ∂M(H ). The f -images of the top and bottom
components of ∂M(G) are called the top and bottom boundary components of ∂M(H ).
The quasiconformal group H carries a natural anti-quasiconformal involution J ∗ that
exchanges the two components of (H ) while pointwise fixing (H ). Its projection
J to ∂M(H ) (which extends to an involution of all M(H )) interchanges the two
boundary components while inducing the identity automorphism of π1 (M(H )).
The natural involution J ∗ and its projection J , which are uniquely determined up
to homotopy, will be denoted by JH ∗ and JH when we want to indicate the group.
Theorem 6.1 (Simultaneous uniformization [3]). Suppose we are given two Riemann
surfaces (S top , Sbot ) of finite hyperbolic area together with an anti-quasiconformal
involution J : Sbot ↔ S top . There exists a quasi-Fuchsian group H , uniquely deter-
mined up to Möbius equivalence, such that the top boundary component of M(H ) is
conformally equivalent to S top , the bottom to Sbot , and J is homotopic to the natural
involution JH .
The result H of simultaneous uniformization however is not a “marked” group,

its relation to the basepoint G is not specified. A marking is specified by choosing a
normalized quasiconformal map F of S2 that (i) projects to a map f of ∂ top M(R) →
∂ top M(H ), and ∂bot M(G) → ∂bot M(H ), and (ii) f satisfies f JG = JH f , up to
homotopy. Via f , a given marking on ∂bot M(G) determines a marking on Sbot and an
isomorphism ϕ : G → H ∈ T(G).
Denote the marked quasi-Fuchsian group H = ((S top , Sbot ; JH ), f ). These groups
fill out the quasi-Fuchsian space T(G).
The Bers slice based on R = Rbot = LHP/G is defined as the set of points
B(R ) = B(G) = {H = ((S, Rbot ; J ), f )}/ ≡ ⊂ T(G).
That is B(G) consists of the normalized quasiconformal deformations F of G which

are conformal on LHP, and which project to a quasiconformal map f of R top =
UHP/G but a conformal map f on Rbot .
The slice is an analytic submanifold of T(G) based on the origin G. Actually there
is a Bers slice based on any point of T(G), and even based on any point of ∂T(G).
Each of these slices is biholomorphically equivalent to the Teichmüller space T (R top ).
The action of the mapping class group on the slice is as follows. If α is an auto-
morphism of Rbot ,
α : ((S, Rbot ; J ), f ) → ((S, Rbot ; J α), f ).
The action of α does not change the conformal type of the bounding surfaces. What it
changes is the topological relationship of the top to the bottom, the marking of the top
relative to that of the bottom. As a result the quasi-Fuchsian group H will change to a
new group H1 in the slice, unless α represents a symmetry of H in that H1 is conjugate
to H – that is α fixes the corresponding Teichmüller point. The automorphism α is a
biholomorphic automorphism of B(G).
Now the compactness of the space of normalized conformal mappings of LHP im-
plies that the closure B(G) in the representation variety R(G) is compact. Therefore
the Bers slice has a compact boundary ∂B(R ) which lies in Rdisc (G). Each boundary
point corresponds to a Kleinian group H with ∂bot M(H ) conformally equivalent to
Rbot , but the top boundary has been somehow degenerated.
The boundary, called the Bers or analytic boundary, is very interesting; it is analo-
gous to the boundary R ∪ ∞ of the Teichmüller space of the torus. There is a certain
countable dense set [38] of distinguished points on it called cusps. A cusp corresponds
to a geometrically finite group H where R top has become “pinched”. Thus if R top is a
closed surface of genus 2, the top component of ∂M(H ) will be one twice-punctured
torus, or two once-punctured tori, or four thrice-punctured spheres, etc. Moreover
the augmented space B(R) ∪ {cusps} can be topologized by designating “horocyclic”
neighborhoods of the cusps. Then its quotient with respect to the mapping class group
is a compactification of moduli space. For an exposition see [43].
The other boundary points H correspond to groups with geometrically infinite top
ends. If H has no new parabolics, then (H ) is connected and simply connected.
The bottom surface (H )/H is conformally equivalent to Rbot while the top surface
has gone missing. Such a group is called singly degenerate.
For closed surfaces, the slice B(G) is itself the intersection with T(G) of a prop-
erly embedded analytic submanifold B∗ (G) ⊂ R(G) of the representation variety,
called the extended Bers slice [30]. It is obtained by constructing all complex pro-
jective structures on Rbot . This means consider all Schwarzian differential equations
{F ; z} = φ on LHP where φ is the lift of a holomorphic quadratic differential on Rbot .
There is a solution, uniquely determined up to postcomposition with a Möbius, which
determines a homomorphism ϕ of G to a nonelementary subgroup of PSL(2, C). In
addition ϕ has the special property that it lifts to a homomorphism between corre-
sponding matrix groups in SL(2, C). The totality of all the homomorphisms make up
432 Albert Marden
the extended slice; if the norm of φ not too large, the solution F will be a conformal
map of LHP, thus determining a point in B(R ).
David Dumas has computed spectacular pictures of B∗ (G) ∩ Int(Rdisc )(G) show-
ing the Bers slice surrounded by an archipelago of islands. He did this in the quasi-
Fuchsian space of a once-punctured torus (www.math.brown.edu/dumas).
6.2 Earthquakes
A geodesic lamination ⊂ H2 is a closed set of mutually disjoint (hyperbolic) lines
referred to as leaves. However two leaves may have a common end point.
If is invariant under a Fuchsian group G, it projects to a lamination on the surface
R = H2 /G.
To introduce Thurston’s theory of earthquakes [65] consider the upper halfplane
model and a line which we may take to be the positive imaginary axis. Denote left
and right quadrants determined by by A and B; A and B have orientations inherited
from C. From the point of view of A, a left earthquake with fracture line is a
discontinuous map which fixes A pointwise, and in B is an isometry moving B to
the left with respect to A, that is, it moves B in the positive direction with respect to
the positive orientation of ∂A. In B it therefore has the form z → kz, k > 1. It is
uniquely determined once the displacement log k along is dictated.
On the other hand if we require that B be fixed, the left earthquake along moves
A to the left from the point of view of a person standing in B. Namely in A it has the
form z → k −1 z.
Next suppose we have a lamination with a finite number of leaves. Fix a gap σ – a
component of H2 \ – as the base of operations. Suppose μ is a positive transverse
measure: that is, to each leaf of the lamination is assigned a positive number as atomic
measure. Normalize the earthquake to be the identity on σ . A transverse geodesic
based in σ will cross a number of leaves. Carry out a sequence of left earthquakes in
sequence along the various leaves, using the displacement assigned by μ.
Formal definition of earthquake. Suppose ⊂ H2 is a geodesic lamination. A
left earthquake is a possibly discontinuous injective and surjective map E : H2 → H2
which is an isometry on each leaf of and on each gap. Given two gaps and/or
leaves X = Y , a line is said to be weakly separating if any path from a point of X
to a point of Y intersects . Let EX , EY denote the respective isometric restrictions
−1
of E. We require that (i) the comparison isometry EX EY be a loxodromic Möbius
transformation T , (ii) the axis of T weakly separates X and Y , and (iii) T translates
Y to the left, when viewed from X. The requirement (iii) means that the direction of
translation along agrees with the orientation induced from X ⊂ H2 \ . The case that
one of X, Y is a line in the boundary of the other is exceptional in that the comparison
map is the identity.
The earthquake maps to another lamination . The inverse of a left earthquake
is a right one.
If has a finite number of leaves, left earthquakes are constructed as illustrated

earlier. Thurston proves that these finite earthquakes are dense in all left earthquakes,
in the topology of uniform convergence on compact sets.
A left earthquake between two Riemann surfaces is an injective, surjective map
which lifts to a left earthquake of H2 . In particular is invariant under the deck
transformations. However if one or more leaves of project to simple geodesics, lifts
are determined only up to Dehn twists along the geodesics. To avoid this ambiguity one
can associate the earthquake with the homotopy type of a homeomorphism between
the surfaces. A more common way is to start with both an invariant lamination in H2 ,
and an invariant transverse measure (more of this below).
Theorem 6.2 (Earthquake Theorem [65]). Every continuous orientation preserving

map ∂ H2 → ∂ H2 is the boundary values of a left earthquake E of H2 . The lamination
is uniquely determined. On , E is uniquely determined except along those leaves
on which it is discontinuous. For each such , there is a range of choices of translations
ranging between the limiting values of E on the two sides; all the choices have same
image in H2 .
Suppose Ri = H2 /Gi , i = 1, 2, are arbitrary Riemann surfaces with possible
boundary contours ∂Ri coming from discontinuity of the action of Gi on maximal
open intervals on ∂ H2 . Assume h : R1 → R2 is an (orientation preserving) homeo-
morphism which extends to a continuous map ∂R1 → ∂R2 . Then the boundary values
on ∂ H2 of a lift of h are the boundary values of a left earthquake of H2 . It projects
back to a left earthquake E : R1 → R2 . E has the same uniqueness indicated above.
This is a very general theorem. The second statement (which includes the first)
follows from the first as lifts of h extend to continuous maps of ∂ H2 .
Associated to any left earthquake is a nonnegative Borel measure μ transverse
to . Two earthquakes corresponding to the same (, μ) have isometric images. The
measure is constructed by a process akin to Riemann integration (see [20]) based on
the example of a finite number of leaves given above. If is invariant under a Fuchsian
group, μ will be as well.
Normally one only works with the restricted class of locally bounded earthquakes.
These are the class of earthquakes whose transverse measures have the property that
for some K < ∞, μ(τ ) < K for all transverse geodesic segments τ of unit length. If
this is the case for (, μ) invariant under a group G, the leaves on H2 /G cannot end
at a puncture, nor can they spiral in to a closed geodesic.
The boundary values on ∂ H2 of locally bounded earthquakes are quasisymmetric
homeomorphisms, which means their boundary values have quasiconformal exten-
sions to H2 (which are equivariant if (, μ) is invariant under deck transformations).
In the other direction, the boundary values of a quasiconformal mapping H2 → H2 ,
for example the boundary values of a lift of a quasiconformal map between surfaces,
is also the boundary values of a locally bounded left earthquake as described in the
Earthquake Theorem. For the details, consult [65].
434 Albert Marden
The bottom line is that if we have Riemann surfaces R, S of finite hyperbolic area,
given a quasiconformal map f : R → S, there is a geodesic lamination and a locally
bounded left earthquake E along it that also sends R to S. The homotopy class of f
is essentially the same as that of E. Conversely every locally bounded earthquake is
essentially homotopic to a quasiconformal map.
Earthquakes map one Fuchsian group G to another. Thus they give rise to the real
Teichmüller space of deformations of Fuchsian groups. In §6.4 we will show how to
complexify the technique so as to get a complex analytic structure.
6.3 Pleated surfaces
We have spoken of the structure of a convex hull boundary component, especially the
dome over a simply connected region. Now consider the reverse process. That is,
given a measured lamination (, μ) in H2 , can we construct a generalized surface in
H3 whose bending measure is μ?
Let’s start with the simplest cases. Take the equatorial plane H2 (the unit disk)
in the ball model and fix a diameter . Bend H2 along with exterior bending angle
0 < θ < π. Here θ = 0 corresponds to no bending at all. The other extreme θ = π
corresponds to two situations: (i) folding H2 in half along , or (more commonly)
(ii) pushing out to ∞ to become a single point ξ thereby forcing H2 in the limit to
become two hyperbolic planes in H3 whose boundaries are tangent at ξ .
To normalize the direction of bending, bend so that the result lies in the upper half
of the ball. The resulting “pleated surface” S bounds on one side a convex region
whose floor on ∂ H3 is bounded by two circular arcs with interior bending angle π − θ.
The dome has only one bending line.
The construction is easily generalized to a finite system of ordered, mutually dis-
joint hyperbolic lines, possibly with common endpoints, 1 , . . . , k ⊂ H2 . Assign an
exterior bending angle 0 < θi < π to each line. Then systematically bend the plane
H2 . For example we may assume that first bend along 1 results in P1 = P constructed
above. Then in P1 locate the copy of 2 , say it lies to the right of 1 . Then bend the
halfplane in P1 lying to the right of 1 along 2 with exterior angle θ2 . And so on for all
the lines. We end up with what is called a pleated surface Pk which is not necessarily
embedded in H3 – it may well have self intersections. It has k bending lines, the
images of the {i }. In any case there is a hyperbolic isometry ϒ : H2 → Pk – such
that ϒ −1 is just unbending. The finite measured lamination is carried to the bending
lines and bending measure on Pk .
Given a geodesic lamination ⊂ H2 and a transverse measure μ with values in
(−∞, +∞), a corresponding pleated surface can be constructed by using finite ap-
proximations [20]. The construction is such that if (, μ) is invariant under a Fuchsian
group G, then the construction of pleated surface automatically determines a defor-
mation of G to a homomorphic image H . This is a group of Möbius transformations
acting in H3 that map the pleated surface onto itself in a manner reflecting the action
of G in H2 , but H is unlikely to be discrete and the pleated surface may not be locally
convex or even locally embedded.
Formally, a pleated surface is determined by a pleating map f : R → M(G) of a
hyperbolic surface R with the properties
(i) f takes any rectifiable path in R to a path in M(G) of the same length;
(ii) every z ∈ R lies in an open geodesic arc which f maps to a geodesic arc in
M(G);
(iii) f sends cusps to cusps: the homomorphism f∗ : π1 (R) → π1 (M(G)) ≡ G
sends parabolics to parabolics.
Assumption (i) can replaced by (i ): geodesic paths in S are sent to rectifiable paths
of the same length in M(G). This apparently stronger definition is equivalent to (i)
[16, I.5.2.6]. The pleated surface is called incompressible if f∗ : π1 (R) → π1 (M(G))
is injective.
The pleating locus is the set ⊂ S consisting of those points z ∈ R with the
following property. There is one and only one open geodesic arc (up to inclusion)
through z which f maps onto a geodesic arc in M(G). The pleating locus is a
closed subset of R and is in fact a geodesic lamination. The image f () is often
referred to as the pleating locus as well, or as the bending lines. The map f is an
isometry of the complementary gaps onto in general infinitely sided polygons in M(G).
Given such a general pleated surface, there is likely to be a great deal of positive
and negative bending. Yet by associating a transverse segment τ to the set of positive
endpoints on ∂ H3 of the oriented leaves through τ and then to a continuum in ∂ H3 ,
it is possible to construct a kind of bending measure called an “R-valued transverse
cocycle”. This real valued measure, which is now only finitely additive, and the
pleating locus ⊂ R characterize the pleated surface. It is very interesting that T (R)
is filled out by the images under a rather explicit finite dimensional space of these
cocycles. For the presentation of Bonahon’s theory of shear cocycles see [9], [8].
6.4 Complex scaling

This section is based on the exposition [22] where the reader will find additional
details.
A crescent is a region bounded by two circular arcs with vertex angles 0 < α < π.
It is conformally equivalent to the wedge Wα = {z ∈ C : 0 < arg z < α}. The
complex scaling map with parameter t ∈ C is defined on the wedge Wα to be,
f+ (z) = f+ (reiθ ) = retθ eiθ = reu+i(1+v)θ , t = u + iv, z = reiθ ∈ Wα .
In parallel with f+ (z) define
f− (z) = e−tθ z, z ∈ Wα .
We list the elementary properties for f+ ; the map f− has complementary properties.
436 Albert Marden
• For all −1 < Im(t) ≤ +1, the map f is a Kt -quasiconformal homeomorphism,

and for all −1 < Im(t) < ∞, the map f is Kt -quasiregular where
1 + |κ(t)| t z
Kt = , κ(t) = · . (6.1)
1 − |κ(t)| 2i − t z
• The boundary values of f on each edge of Wα is the restriction of a Möbius

transformation.
• f commutes with any Möbius transformation that maps Wα onto itself.
For larger angles, we must think of the image of Wα as spread over S2 .
Important special cases for f+ are:
(i) Re(t) = u = 0. Then f (z) = eivθ z. This is a smoothed version of “grafting”.
The wedge Wα of angle α is mapped onto the wedge W(1+v)α .
(ii) t = i. Then f (z) = re2iθ is called the angle doubling map. It is 2-quasiconformal.
(iii) v = Im(t) = 0. Then f (z) = euθ z. This is a smoothed version of an earthquake.
The wedge Wα is mapped onto itself, but it is stretched (or contracted if u < 0)
continuously from its right edge to its left.
Next, suppose we have a pair of symmetric wedges with respect to R,
W+ = {z : 0 ≤ β < arg z < α + β < π }, W− = {z : −α − β < arg z < −β ≤ 0}.
Define f+ in W+ by conjugating W+ to Wα via the rotation z → e−iβ z. Define f−
in W− by conjugating it via the rotation z → eiβ z to the action of f− in the reflected
image of Wα in R.
Interpolating by the identity and a loxodromic transformation, we end up with the
following continuous map Et (z) : C → C:
• Et (z) = f+ (z), if z ∈ W+ ,
• Et (z) = f− (z), if z ∈ W− ,
• Et (z) = z, for z in the component of C \ (W− ∪ W+ ) containing the positive real
axis,
• Et (z) = etα z for z in the component of C \ (W− ∪ W+ ) containing the negative
real axis.
Thus E0 (z) = z while the limiting case E−i collapses the crescents (vertex angles
→ 0).
The mapping Et is the basis of the angle scaling method.
The application is to a discrete G-invariant geodesic lamination ⊂ H2 , where
G is Fuchsian; we can also take G = id. Typically we take to be the set of lifts of a
finite set of mutually disjoint simple closed geodesics on R = H2 /G. In this case we
can find mutually disjoint tubular neighborhoods of the geodesics on R. The totality
of lifts to H2 of these neighborhoods forms a system of mutually disjoint crescents
about the corresponding geodesics. We can take a maximal system of crescents in the
sense each boundary component of one crescent is tangent to a boundary component

of another.
More generally we do not need to take the crescents to be centered on its associated
geodesic or even to contain it. All we need is a set of crescents with the following
three properties, expressed in the disk model D:
• Corresponding to each leaf is exactly one crescent whose vertices are the
endpoints of on ∂ D.
• The crescent interiors are mutually disjoint.
• The set of crescents is invariant under the action of G.
The reflection in ∂ D gives a set of crescents in the exterior Dext of D with the same
vertex angles and associated with the reflection of .
Denote by α the vertex angle of the crescent associated with the leaf
With the given data consider the measured lamination
(, μ) = {, α )}, 0 < α < π, ∈ ,
and its reflection in ∂ D. Apply the angle scaling map Et to the infinite discrete system
crescents in D ∪ Dext . The map Et is normalized by choosing a component
of pairs of
C of D \ {crescents} and requiring Et to be the identity on C . Concentrating on

the case that Im(t) > −1 we arrive at the following result. The term “holomorphic
motion” will be explained in §7.3.
Theorem 6.3 ([22]). Suppose (, μ) is a G-invariant, discrete, measured lamination

in D associated with a system of crescent pairs in D ∪ Dext as described above. There
exists a G-equivariant complex scaling mapping Et (z) = Et (z; , μ) for every t ∈ C.
It has the following properties.
(i) For {−1 < Im(t) < +1}, Et : S2 → S2 is a Kt -quasiconformal homeomor-
phism. Furthermore {Et } is a holomorphic motion of S2 .
(ii) For {−1 < Im(t)}, Et : D → S2 is a Kt -quasiregular developing mapping. For
each z ∈ D, Et (z) is defined and holomorphic in t.
(iii) When t = i, the angle doubling map Ei : D → i = Ei (D) is a 2-quasiconformal
homeomorphism which has a continuous extension to ∂ D. It collapses the cres-
cents in Dext .
The scaling map Et can be decomposed as follows:

Et = Eiv Eu , t = u + iv.
The map Eu : S2 → S2 is a homeomorphism, its actions in D and the exterior Dext
being symmetric. Its action in D is as a smoothed out earthquake sending to a new
lamination u ; we write Eu : (, μ) → (u , μ) with the understanding that each leaf
of u takes its measure from its preimage. Equally Eu sends a crescent system for
(, μ) to one for (u , μ). The values of Eu on ∂ D are exactly the boundary values
of a classical earthquake. By reflection, Eu acts in Dext as well.
438 Albert Marden
In contrast, the action of Eiv on u can be described as bending. We will describe

this action on D, there is a similar action on Dext . When −1 < v < +1, Eiv : D →
Eiv (D) = v is a quasiconformal homeomorphism sending D to a new region v
and the crescent C associated with the leaf ⊂ u of vertex angle α to a crescent
C in v of vertex angle (1 + v)α . There is an isometry Eiv ∗ : D → Dome( )
v

that sends the leaf ⊂ u to the bending line of exterior bending angle vα. To
complete the picture r −1 ( ) ⊂ v is a crescent C ∗ ⊂ v of vertex angle vα , where
r : v → Dome(v ) is the nearest point retraction. The two crescents C , C ∗ in v
have the same vertices. There is a similar interpretation in terms of the pleated surface
generated by (, μ) for all −1 < v.
Thus, if −1 < v < +1, then = Et (D) is simply connected and Et ∗ = Eiv ∗ E
u
∗
maps D onto Dome(v ). The maps Et and Et are related by the fact that they have
the same boundary values on ∂ D.
In short, given the crescent system and −1 < v < +1, Et ∗ (D) = Eiv ∗ E (D) is the
u
pleated surface Dome(v ) in H3 . The interpretation can be extended to −1 < v < ∞
by using the pleated surface determined from u with bending angles given by {α}.
We note that complex scaling can be done equally well in any quasidisk, given a
system of crescents there.
6.5 Complex earthquakes

The range of complex scaling is S2 . In contrast, starting with (, μ) ⊂ H2 and
t = u + iv ∈ C, we showed in [20] how to construct what we called a “quakebend”. A
quakebend consists of an earthquake in D determined by the measure (, uμ) followed
by bending according to (, vμ). In current terminology a quakebend is better called
a complex earthquake and denoted by CE t : D → H3 .
We are ultimately interested in exploring the set
X = {t ∈ C : the image CE t (D) is the dome over a quasidisk}.
The set X is a kind of 1-dimensional slice through T(G). We will use the generalization
of complex scaling to show how to build the floor under such a dome as a mapping
from D.
7 Complex scaling for general laminations

The goal of this section is to generalize complex scaling from discrete laminations to
general measured laminations. For a general measured lamination in D typically has
no isolated leaves, and therefore no associated crescent systems. We will first focus
on generalizing the angle scaling construction to general measured laminations on D
without group actions. After explaining this we will bring group actions back in.
Normalize the measured lamination by requiring that

μ = sup μ(τ ) = 1,
τ
where τ ranges over all transverse open geodesic segments of unit length.
We begin by constructing a map t that extends the restriction of CE t to S1 = ∂ D
to a map of S2 . The result is displayed below as Proposition 7.1 and presented without
proof. The statement uses the notion “holomorphic motion”. For the convenience of
the reader unfamiliar with this important tool, we have appended an explanation in
§7.3.
Proposition 7.1 ([23], [25]). There exists a simply connected, strip-like region T ⊃ R,
symmetric about R, such that for t ∈ T , t = u + iv the following hold.
(i) The complex earthquake CE t : D → H3 is a quasiisometry.
(ii) CE t extends to ∂ D; its restriction to ∂ D is a holomorphic motion of S1 = ∂ D
in S2 .
(iii) The restriction of CE t to S1 is itself the restriction to S1 of a holomorphic motion
t : S2 → S2 which is determined by (, tμ).
(iv) The restriction of t to D is a quasiconformal map onto its image t .
(v) The restriction of t to S1 is injective (from (ii)), thus mapping S1 to a Jordan
curve bounding t . The bending measure of Dome(t ) is vt · μ∗ where μ∗ is
the transfer of μ to the image uμ of under the earthquake map CE uμ .
(vi) If (, μ) is invariant under the group G, the holomorphic motion t can be
chosen so that there is a homomorphism ϕt : G → Ht with
t (g(z)) = ϕt (g) t (z) for all g ∈ G and z ∈ D.
Next we fix a point t0 = iv0 ∈ T and set 0 = t0 (D). We will now use
the quasidisk 0 as our base for scaling. The image of the complex earthquake
CE t0 : D → H3 is Dome(0 ).
To set the stage, assume first that Dome(0 ) has a discrete set of bending lines.
The inverse image of the bending lines under the nearest point retraction r is a set of
associated crescents in 0 . Let Et denote the scaling map of 0 with respect to this
crescent system and its vertex angles. Normalize so it is the identity in a prescribed
gap. By Theorem 6.3, for each t ∈ C, Et extends to a continuous map of the closure
0 → S2 . Set
t = Ei(t−t0 )/t0 , t ∈ T , so t0 = id, 0 = E−i .
Then
t t0 (ζ ) = t (ζ ) ζ ∈ ∂ D, t ∈ T . (7.1)
We will construct for all nonnegative measured laminations (, μ) a mapping t
that generalizes scaling on 0 and continues to satisfy Equation (7.1). But for t , its
parameter t will be allowed to range over C.
440 Albert Marden
After a long chain of reasoning, the bottom line is this:
Theorem 7.2 ([25]). Let (, μ) be a nonnegative measured lamination normalized

by μ = 1. Fix t0 ∈ T and set 0 = t0 (D). The map t : 0 → S2 , t ∈ C, has
the following properties:
(i) t0 = id.
(ii) For each z ∈ 0 , t (z) depends holomorphically on t ∈ C.
(iii) For each t ∈ T , t (z) can be continuously extended to ∂0 so that Equation (7.1)
holds. In particular, the restriction 0 : ∂0 → ∂ D is inverse to t0 : ∂ D →
∂0 .
(iv) For t ∈ T ∩ UHP, t : 0 → S2 is injective, and t (0 ) = t (D).
(v) When t ∈ UHP, t : 0 → S2 is a locally injective quasiregular mapping.
In short, we have constructed a holomorphic family of generalized scaling maps

t (z) : C ×0 → S2 which are related to the boundary values of complex earthquakes
CE t : ∂ D → ∂ Dome(t (0 )), for t ∈ T .
7.1 Sullivan’s theorem

We will digress in order to describe the application of Theorem 7.2 to Sullivan’s
amazing theorem on the metric relation of a simply connected region to its dome. This
states that there is a universal K, such that any simply connected region ⊂ C, =
C, has the following property. There is a K-quasiconformal map : → Dome()
that extends continuously to the identity map ∂ = ∂ Dome(). Moreover if is
invariant under a Möbius group so is . For the history and a proof see [25].
It was initially conjectured by Thurston that K = 2, but we now know that K > 2,
even if one does not require equivariance, [21].
The proof of Sullivan’s theorem given in [25] proceeds as follows. First consider
locally injective quasiregular (developing) mappings g : D → S2 . Quasiregular maps
can be factored g = h f where f is a quasiconformal (homeomorphism) and h
is locally injective (because g is) and holomorphic on the image of f . Via g, the
complex structure on S2 can be pulled back to D, getting a new complex structure
Cg on D and a quasiconformal mapping ĝ : D → Cg . Now bringing in a Riemann
mapping R : Cg → D, we end up with a quasiconformal map R ĝ : D → D. We
can normalize R so that ±1, i ∈ ∂ D are fixed by the extension qs(g) to ∂ D of R ĝ.
The conclusion is that the correspondence g → qs(g) identifies each locally injective
quasiregular map to a point in the universal Teichmüller space UT – recall that UT
can be taken as the space of normalized quasisymmetric maps of S1 .
We will apply our construction to two maps: (i) to the locally injective quasiregular
map
gt = f t0 : D → 0 → S2 , t ∈ UHP .
and (ii) to the quasiconformal map

ft = ,tμ : D → t ∈ S2 , t ∈T.
Consider then the two maps
F : t ∈ T → qs(ft ) ∈ UT , G : t ∈ UHP → qs(gt ) ∈ UT .
At the end of a sequence of arguments [25] one establishes that
Theorem 7.3.
• F is holomorphic in t,
• G is holomorphic in t,
• F = G on their common domain T ∩ UHP.
In short, F has a holomorphic extension to the simply connected T ∪ UHP.

Theorem 7.3 yields a new and shorter proof of Sullivan’s Theorem along with a
greatly improved estimate for K. It was with the idea of getting an improved estimate
for K that led us to enlarge the domain for t as much as possible. The estimate we
found is that K ≤ 13.88 as compared to K ≤ 82.8 in [20]. These estimates are for the
equivariant case: if is invariant under a Möbius group, the map : → Dome()
must be as well. If equivariance is not required to hold, Chris Bishop [6] showed that
the estimate could be improved even further to 7.8.
7.2 Earthquake disks

We finally turn to the matter of earthquake disks. Given a measured lamination
(, μ) ⊂ H2 , and t ∈ C, consider the pleated surface Pt given by the complex
earthquake CE t . The bending measure of Pt is Im(t) · μ∗ , where as usual μ∗ is the
carryforth of μ under the Re(t) · μ earthquake of .
Set
X = {t ∈ C : CE t (D) is the dome of a simply connected region ⊂ C}.
Theorem 7.4 (The Disk Theorem [25]). The set X = X,μ has the following prop-
erties:
• X is symmetric about the real t-axis.
• X ⊂ C is closed.
• X ∩ {Im(t) > 0} = {t ∈ UHP : t : 0 → S2 is injective}.
• C \ X has no bounded components; if X0 ⊂ X is a component, C \ X0 has no
bounded components either.
• Every component of the interior of X is simply connected.
442 Albert Marden
Note that X is invariant under translation t → t + 2π i. There are numerous open

questions about X, in particular, whether it contains isolated points and whether it is
locally connected.
Suppose that (, μ) is invariant under a Fuchsian group G with R = D/G a
Riemann surface of finite area. If t ∈ X then the floor t under the dome Pt is
invariant under a quasi-Fuchsian group. Consequently there is a holomorphic map of
the interior D : Int(X) → T (R).
If R is a once punctured torus, McMullen proved [40] that X is biholomorphically
equivalent to T (R). It was in this paper that the term earthquake disk was coined for
a properly embedded image, in a Teichmüller space, of the unit disk under a complex
earthquake map. In the paper, bending in 3D is replaced by its analogue “grafting” in
2D (see §6.4).
7.3 Appendix: holomorphic motion

Let B ⊂ S2 be an arbitrary set containing at least three points. Let {fλ (z)} denote
a family of functions B → S2 with parameter λ ∈ D. The family {fλ } is called a
holomorphic motion of B if it has the following three properties:
• For each fixed λ ∈ D, the map fλ : z ∈ B → fλ (z) ∈ S2 is injective.
• For each fixed z ∈ B, the map λ → fλ (z) is a holomorphic map D → S2 .
• For each z ∈ B, f0 (z) = z.
What makes this notion significant is the following λ-Lemma as originally formu-
lated by Mañé–Sad–Sullivan with the final statement in Sladkowski [57], see also [19].
Theorem 7.5 (λ-Lemma). Let G be a group of Möbius transformations which preserve

a subset B ⊂ S2 . Suppose {fλ } is a holomorphic motion of B. Suppose further that
for each λ ∈ D, fλ is G-equivariant: fλ induces an isomorphism to a group of Möbius
transformations G → Gλ that preserve fλ (B). Then
(i) fλ (z) is jointly continuous in λ ∈ D and z ∈ B;
(ii) for fixed λ ∈ D, fλ (z) is the restriction to B of a Kλ -quasiconformal, G-equi-
variant mapping fλ ∗ : S2 → S2 , where
1 + |λ|
Kλ = ;
1 − |λ|
(iii) {fλ ∗ } is a holomorphic motion of S2 , λ ∈ D.
Note that continuity in z is not assumed, it is a conclusion. In particular if B = D,

then fλ (D) is a quasidisk.
In our application, λ = t ∈ T , where T is equivalent to the unit disk. We set
B = ∂ D, for then CE t is injective on B for t ∈ T , Proposition 7.1(ii).
References
[1] I. Agol, Tameness of hyperbolic 3-manifolds. Preprint, 2004; arXiv:math.GT/0405568.

415
[2] L. V. Ahlfors, Finitely generated Kleinian groups. Amer. J. Math. 86 (1964) 413–429;
correction ibid 87 (1965), 759. 413
[3] L. Bers, Simultaneous uniformization. Bull. Amer. Math. Soc. 66 (1960), 94–97. 424, 430
[4] L. Bers, Spaces of Kleinian groups. In Several Complex Variables, I, Lecture Notes in
Math. 155, Springer-Verlag, Berlin 1970, 9–34.
[5] L. Bers and L. Greenberg, Isomorphisms between Teichmüller spaces. In Advances in
the Theory of Riemann Surfaces, Ann. of Math. Stud. 66, Princeton University Press,
Princeton, N.J., 1971. 425
[6] C. J. Bishop, An explicit constant for Sullivan’s convex hull theorem. In In the tradition
of Ahlfors and Bers, III, Contemp. Math. 355, Amer. Math. Soc., Providence, RI, 2004,
41–69. 441
[7] C. J. Bishop, A fast approximation to the Riemann map. Preprint. 420
[8] F. Bonahon, Shearing hyperbolic surfaces, bending pleated surfaces, and Thurston’s sym-
pletic form. Ann. Fac. Sci. Toulouse 5 (1996), 233–297. 435
[9] F. Bonahon, Geodesic laminations on surfaces. In Laminations and Foliations in Dynam-
ics, Geometry, and Topology (M. Lyubich et al., eds.), Contemp. Math. 269, Amer. Math.
Soc., Providence, RI, 2001, 1–37. 435
[10] F. Bonahon and J.-P. Otal, Lamination mesurées de plissage des variétés hyperboliques
de dimension 3. Ann. of Math. 160 (2004), 1013–1055. 427
[11] J. F. Brock, R. D. Canary, and Y. Minsky, Classification of Kleinian surface groups II: The
ending lamination conjecture. Preprint, 2004; arXiv:math.GT/0412006. 414, 415, 429
[12] J. F. Brock, R. D. Canary, and Y. Minsky, The classification of finitely-generated Kleinian
groups. In preparation. 414, 415, 429
[13] D. Calegari and D. Gabai, Shrinkwrapping and the taming of hyperbolic 3-manifolds.
J. Amer. Math. Soc. 19 (2006), 385–446. 415
[14] R. D. Canary, A covering theorem for hyperbolic 3-manifolds and its applications. Topol-
ogy 35 (1996), 751–778. 428
[15] R. D. Canary and D. McCullough, Homotopy Equivalence of 3-Manifolds and Deforma-
tion Theory of Finitely Generated Kleinian groups. Mem. Amer. Math. Soc. 172 (812),
(2004). 426
[16] R. D. Canary, D. B. A Epstein, P. Green, Notes on notes of Thurston. In Fundamentals of
Hyperbolic Geometry: Selected Expositions (R. D. Canary, D. B. A. Epstein, A. Marden,
eds.), London Math. Soc. Lecture Note Ser. 328, Cambridge University Press, Cambridge
2006, 1–115. 435
[17] Y. Choi and C. Series, Lengths are coordinates for convex structures. J. Differential Geom.
73 (1) (2006), 75–117. 427
444 Albert Marden
[18] A. Douady and C. J. Earle, Conformally natural extensions of homeomorphisms of the

circle. Acta Math. 157 (1986), 23–48. 418
[19] C. J. Earle, I. Kra, and S. L. Krushkal, Holomorphic motions and Teichmüller spaces.
Trans. Amer. Math. Soc. 343 (1994), 927–948. 442
[20] D. B. A. Epstein and A. Marden, Convex hulls in hyperbolic space, a theorem of Sullivan,
and measured pleated surfaces. In Fundamentals of Hyperbolic Geometry: Selected Ex-
positions (R. D. Canary, D. B. A. Epstein, A. Marden, eds.), London Math. Soc. Lecture
Note Ser. 328, Cambridge University Press, Cambridge 2005, 117–266. 419, 433, 434,
438, 441
[21] D. B. A. Epstein and V. Markovic, The logarithmic spiral: A counterexample to the K = 2
conjecture. Ann. of Math. 161 (2005), 925–957. 440
[22] D. B. A. Epstein, A. Marden, and V. Markovic, Complex angle scaling. In Kleinian Groups
and Hyperbolic 3-Manifolds (Y. Komari, V. Markovic, C. Series, eds.), London Math. Soc.
Lecture Note Ser. 299, Cambridge University Press, Cambridge 2003, 343–362. 435, 437
[23] D. B. A. Epstein, A. Marden, and V. Markovic, Quasiconformal homeomorphisms and the
convex hull boundary. Ann. of Math. 159 (2004), 305–336. 420, 439
[24] D. B. A. Epstein, A. Marden, and V. Markovic, Convex regions in the plane and their
domes. Proc. London Math. Soc. (3) 92 (3) (2006), 624–654. 420
[25] D. B. A. Epstein, A. Marden, and V. Markovic, Complex earthquakes and deformations
of the unit disk. Preprint, 2004. 439, 440, 441
[26] R. A. Evans, Weakly type-preserving sequences and strong convergence. Geom. Dedicata
108 (2004), 71–92. 428
[27] M. Feighn and D. McCullough, Finiteness conditions for 3-manifolds with boundary.
Amer. J. Math. 109 (1987), 1155–1169. 414
[28] J. Gilman and P. Waterman, Classical two-parabolic T -Schottky groups. J. Anal. Math.
98 (2006), 1–42. 419
[29] M. Feighn and G. Mess, Conjugacy classes of finite subgroups in Kleinian groups. Amer.
J. Math. 113 (1991), 179–188. 414
[30] D. Gallo, M. Kapovich, and A. Marden, The monodromy groups of Schwarzian equations
on closed Riemann surfaces. Ann. of Math. 151 (2000), 625–704. 431
[31] J. Hempel 3-Manifolds. Ann. of Math. Studies 86, Princeton University Press, Princeton,
N.J., 1976. 416
[32] T. Jørgensen and P. Klein, Algebraic convergence of finitely generated Kleinian groups.
Quart. J. Math. 33 (1982), 325–332. 426, 428
[33] T. Jørgensen and A. Marden, Algebraic and geometric convergence of Kleinian groups.
Math. Scand. 66 (1990), 47–72. 428
[34] M. Kapovich, Hyperbolic Manifolds and Discrete Groups. Progr. Math. 183, Birkhäuser,
Boston, MA, 2001. 412, 421, 427
[35] I. Kra, On spaces of Kleinian groups. Comment. Math. Helv. 47 (1972), 53–69. 424, 425
[36] D. McCullough, Compact submanifolds of 3-manifolds with boundary. Quart. J. Math.
37 (1986), 299–307. 415
[37] D. McCullough and A. Miller, Homeomorphisms of 3-manifolds with compressible

boundary. Mem. Amer. Math. Soc. 61 (344) (1986).
[38] C. McMullen, Cusps are dense. Ann. of Math. 13 (1991), 217–247. 431
[39] C. McMullen, Renormalization and 3-Manifolds which Fiber over the Circle. Ann. of
Math. Stud. 142, Princeton University Press, Princeton, N.J., 1996. 418
[40] C. McMullen, Complex earthquakes and Teichmüller theory, J. Amer. Math. Soc. 11
(1998), 283–320. 427, 442
[41] A. Marden, On homotopic mappings of Riemann surfaces. Ann. of Math. 90 (1969), 1–8.
425
[42] A. Marden, The geometry of finitely generated kleinian groups, Ann. of Math. 99 (1974),
383–462. 421, 426
[43] A. Marden, Geometric relations between homeomorphic Riemann surfaces, Bull. Amer.
Math. Soc. 3 (1980), 1007–1017. 431
[44] A. Marden, A proof of Ahlfors Finiteness Theorem. In Spaces of Kleinian Groups (Y. Min-
sky, M. Sakuma, C. Series, eds.), London Math. Soc. Lecture Note Ser. 329, Cambridge
University Press, Cambridge 2005, 247–257.
[45] A. Marden, Outer Circles: An Introduction to Hyperbolic 3-Manifolds. Cambridge Uni-
versity Press, Cambridge 2007, to appear. 412
[46] B. Maskit, Self-maps of Kleinian groups. Amer. J. Math. 93 (1971), 840–856. 423, 424,
425
[47] K. Matsuzaki and M. Taniguchi, Hyperbolic Manifolds and Kleinian Groups. Oxford
Math. Monogr., Oxford University Press, New York 1998. 412
[48] Y. Minsky, Classification of Kleinian surface groups I: models and bounds. Preprint, 2004;
arXiv:math.GT/0302208. 415, 429
[49] G. D. Mostow, Strong Rigidity of Locally Symmetric Spaces, Ann. of Math. Stud. 78,
Princeton University Press, Princeton, N.J., 1973. 421
[50] J.R Munkres, Obstructions to smothing piecewise-differentiable homeomorphisms. Ann.
of Math. 72 (1960), 521–554. 416
[51] Ch. Pommerenke, Uniformly perfect sets and Fuchsian groups. Analysis 4 (1984), 299–
321. 417
[52] J.-P. Otal, Thurston’s hyperbolization of Haken manifolds. In Surveys of Differential Ge-
ometry, Vol. III, International Press, Boston, MA, 1998, 77–194. 421
[53] J.-P. Otal, The Hyperbolization Theorem for Fibered 3-Manifolds, SMF/AMS Texts
Monogr. 7, Amer. Math. Soc., Providence, RI; Soc. Math. France, Paris, 2001. 421
[54] G. Prasad, Strong rigidity of Q-rank 1 lattices. Invent. Math. 21 (1973), 255–286. 421
[55] G. P. Scott, Finitely generated 3-manifolds are finitely presented, J. London Math. Soc. 6
(1973), 437–440. 412, 414
[56] G. P. Scott, Compact submanifolds of 3-manifolds. J. London Math. Soc. 7 (1973),
246–250. 414
[57] Z. Slodkowski, Holomorphic motions and polynomial hulls. Proc. Amer. Math. Soc. 111
(1991), 347–355; Abstracts Amer. Math. Soc. 13 (1992), 259. 442
446 Albert Marden
[58] T. Soma, Geometric limits of quasifuchsian groups. Preprint. 429

[59] D. Sullivan, On the ergodic theory at infinity of an arbitrary discrete group of hyperbolic
motions. In Riemann Surfaces and Related Topics (I. Kra, B. Maskit, eds.), Ann. of Math.
Stud. 97, Princeton University Press, Princeton, N.J., 1981, 465–496.
[60] D. Sullivan, Quasiconformal homeomorphisms and dynamics II: Structural stability im-
plies hyperbolicity for Kleinian groups. Acta Math. 155 (1987), 243–260. 427
[61] G. A. Swarup, Two finiteness properties in 3-manifolds. Bull. London Math. Soc. 12 (1980)
296–302. 426
[62] W. P. Thurston, The Geometry and Topology of Three-Manifolds. Princeton Univiversity
Lecture Notes, 1979; http://msri.org/publications/books/gt3m/. 412
[63] W. P. Thurston, Three dimensional manifolds, Kleinian groups and hyperbolic geometry.
Bull. Amer. Math. Soc. 6 (1982), 357–381. 421
[64] W. P. Thurston, Hyperbolic structures on 3-manifolds I: Deformation of acylindrical mani-
folds. Ann. of Math. 124 (1986), 203–246. 427
[65] W. P. Thurston, Earthquakes in two-dimensional hyperbolic geometry. In Fundamentals of
Hyperbolic Geometry: Selected Expositions, (R. D. Canary, D. B. A. Epstein, A. Marden,
eds.), London Math. Soc. Lecture Note Ser. 328, Cambridge University Press, Cambridge
2005, 267–289. 432, 433
[66] P. Tukia, Quasiconformal extension of quasisymmetric mappings compatable with a Mö-
bius group. Acta Math. 154 (1985), 153–193. 418
[67] P. Tukia, personal communication. 418
Chapter 10
Geometry of the complex of curves

and of Teichmüller space
Ursula Hamenstädt
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
2 Train tracks and geodesic laminations . . . . . . . . . . . . . . . . . . . . 449
3 Hyperbolicity of the complex of curves . . . . . . . . . . . . . . . . . . . 454
4 Geometry of Teichmüller space . . . . . . . . . . . . . . . . . . . . . . . 462
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
1 Introduction
Consider a compact oriented surface S of genus g ≥ 0 from which m ≥ 0 points,
so-called punctures, have been deleted. We assume that S is non-exceptional, i.e. that
3g − 3 + m ≥ 2; this rules out a sphere with at most four punctures and a torus with
at most one puncture.
In [11], Harvey associates to such a surface the following simplicial complex.
Definition 1.1. The complex of curves C(S) for the surface S is the simplicial complex
whose vertices are the free homotopy classes of essential simple closed curves on S
and whose simplices are spanned by collections of such curves which can be realized
disjointly.
Here we mean by an essential simple closed curve a simple closed curve which is
not contractible nor homotopic into a puncture. Since 3g − 3 + m is the number of
curves in a pants decomposition of S, i.e. a maximal collection of disjoint mutually not
freely homotopic essential simple closed curves which decompose S into 2g − 2 + m
open subsurfaces homeomorphic to a thrice punctured sphere, the dimension of C(S)
equals 3g − 4 + m.
In the sequel we restrict our attention to the one-skeleton of the complex of curves
which is usually called the curve graph; by abuse of notation, we denote it again
by C(S). Since 3g − 3 + m ≥ 2 by assumption, C(S) is a nontrivial graph which
moreover is connected [11]. However, this graph is locally infinite. Namely, for every
simple closed curve α on S the surface S − α which we obtain by cutting S open along
448 Ursula Hamenstädt
α contains at least one connected component of Euler characteristic at most −2, and
such a component contains infinitely many distinct free homotopy classes of simple
closed curves which viewed as curves in S are disjoint from α.
Providing each edge in C(S) with the standard euclidean metric of diameter 1
equips the curve graph with the structure of a geodesic metric space. Since C(S)
is not locally finite, this metric space (C(S), d) is not locally compact. Masur and
Minsky [19] showed that nevertheless its geometry can be understood quite explicitly.
Namely, C(S) is hyperbolic of infinite diameter. Here for some δ > 0 a geodesic metric
space is called δ-hyperbolic in the sense of Gromov if it satisfies the δ-thin triangle
condition: For every geodesic triangle with sides a, b, c the side c is contained in the
δ-neighborhood of a ∪ b. Later Bowditch [2] gave a simplified proof of the result
of Masur and Minsky which can also be used to compute explicit bounds for the
hyperbolicity constant δ.
Since the Euler characteristic of S is negative, the surface S admits a complete hy-
perbolic metric of finite volume. The group of diffeomorphisms of S which are isotopic
to the identity acts on the space of such metrics. The quotient space under this action
is the Teichmüller space Tg,m for S of all marked isometry classes of complete hyper-
bolic metrics on S of finite volume, or, equivalently, the space of all marked complex
structures on S of finite type. The Teichmüller space can be equipped with a natural
topology, and with this topology it is homeomorphic to R6g−6+2m . The mapping class
group Mg,m of all isotopy classes of orientation preserving diffeomorphisms of S acts
properly discontinuously as a group of diffeomorphisms of Teichmüller space preserv-
ing a complete Finsler metric, the so-called Teichmüller metric. The quotient orbifold
is the moduli space Mod(S) of S of all isometry classes of complete hyperbolic metrics
of finite volume on S (for all this see [12]).
The significance of the curve graph for the geometry of Teichmüller space comes
from the obvious fact that the mapping class group acts on C(S) as a group of sim-
plicial isometries. Even more is true: If S is not a twice punctured torus or a closed
surface of genus 2, then the extended mapping class group of isotopy classes of all
diffeomorphisms of S coincides precisely with the group of simplicial isometries of
C(S); for a closed surface of genus 2, the group of simplicial isometries of C(S) is the
quotient of the extended mapping class group under the hyperelliptic involution which
acts trivially on C(S) (see [13] for an overview on this and related results). Moreover,
there is a natural map : Tg,m → C(S) which is coarsely Mg,m -equivariant and
coarsely Lipschitz with respect to the Teichmüller metric on Tg,m . By this we mean
that there is a number a > 1 such that d((φh), φ(h)) ≤ a for all h ∈ Tg,m and
all φ ∈ Mg,m and that moreover d(h, h ) ≤ a dT (h, h ) + a for all h, h ∈ Tg,m
where dT denotes the distance function on Tg,m induced by the Teichmüller metric
(see Section 4).
As a consequence, the geometry of C(S) is related to the large-scale geometry
of the Teichmüller space and the mapping class group. We discuss this relation in
Section 4. In Section 3 we give a proof of the hyperbolicity of the curve graph using
train tracks and splitting sequences of train tracks in a consistent way as the main tool.
Chapter 10. Geometry of the complex of curves and of Teichmüller space 449
Section 2 introduces train tracks, geodesic laminations and quadratic differentials and
summarizes some of their properties.
This work was partially supported by SFB 611.
2 Train tracks and geodesic laminations

Let S be a nonexceptional surface of finite type and choose a complete hyperbolic
metric on S of finite volume. With respect to this metric, every essential free homo-
topy class of loops can be represented by a closed geodesic which is unique up to
parametrization. This geodesic is simple, i.e. without self-intersection, if and only if
the free homotopy class has a simple representative (see [4]). In other words, there is
a one-to-one correspondence between vertices of the curve graph and simple closed
geodesics on S. Moreover, there is a fixed compact subset S0 of S containing all
simple closed geodesics.
The Hausdorff distance between two closed bounded subsets A, B of a metric
space X is defined to be the infimum of all numbers ε > 0 such that A is contained in
the ε-neighborhood of B and B is contained in the ε-neighborhood of A. This defines
indeed a distance and hence a topology on the space of closed bounded subsets of X;
this topology is called the Hausdorff topology. If X is compact then the space of
closed subsets of X is compact as well. In particular, for the distance on S induced
by a complete hyperbolic metric of finite volume, the space of closed subsets of the
compact set S0 ⊂ S is compact with respect to the Hausdorff topology.
The collection of all simple closed geodesics on S is not a closed set with respect
to the Hausdorff topology, but a point in its closure can be described as follows.
Definition 2.1. A geodesic lamination for a complete hyperbolic structure of finite

volume on S is a compact subset of S which is foliated into simple geodesics.
Thus every simple closed geodesic is a geodesic lamination which consists of a

single leaf. The space of geodesic laminations on S equipped with the Hausdorff
topology is compact, and it contains the closure of the set of simple closed geodesics
as a proper subset. Note that every lamination in this closure is necessarily connected.
To describe the structure of the space of geodesic laminations more explicitly we
introduce some more terminology.
Definition 2.2. A geodesic lamination λ is called minimal if each of its half-leaves is

dense in λ. A geodesic lamination λ is maximal if all its complementary components
are ideal triangles or once punctured monogons. A geodesic lamination is called
complete if it is maximal and can be approximated in the Hausdorff topology for
compact subsets of S by simple closed geodesics.
As an example, a simple closed geodesic is a minimal geodesic lamination. A

minimal geodesic lamination with more than one leaf has uncountably many leaves.
Every minimal geodesic lamination can be approximated in the Hausdorff topology

by simple closed geodesics [5]. Moreover, a minimal geodesic lamination λ is a
sublamination of a complete geodesic lamination [9], i.e. there is a complete geodesic
lamination which contains λ as a closed subset. In particular, every simple closed
geodesic on S is a sublamination of a complete geodesic lamination. Every geodesic
lamination λ is a disjoint union of finitely many minimal components and a finite
number of isolated leaves. Each of the isolated leaves of λ either is an isolated closed
geodesic and hence a minimal component, or it spirals about one or two minimal
components [1], [5], [24]. This means that the set of accumulation points of an
isolated half-leaf of λ is a minimal component of λ.
Geodesic laminations which are disjoint unions of minimal components can be
equipped with the following additional structure.
Definition 2.3. A measured geodesic lamination is a geodesic lamination together

with a translation invariant transverse measure.
A transverse measure for a geodesic lamination λ assigns to every smooth compact

arc c on S with endpoints in the complement of λ and which intersects λ transversely a
finite Borel measure on c supported in c ∩ λ. These measures transform in the natural
way under homotopies of c by smooth arcs transverse to λ which move the endpoints
of the arc c within fixed complementary components of λ. The support of the measure
is the smallest sublamination ν of λ such that the measure on any such arc c which does
not intersect ν is trivial. This support is necessarily a union of minimal components
of λ. An example for a measured geodesic lamination is a weighted simple closed
geodesic which consists of a simple closed geodesic α and a positive weight a > 0.
The measure disposed on a transverse arc c is then the sum of the Dirac masses on the
intersection points between c and α multiplied with the weight a.
The space ML of measured geodesic laminations on S can naturally be equipped
with the weak∗ -topology. This topology restricts to the weak∗ -topology on the space
of measures on a given arc c which is transverse to each lamination from an open
subset of lamination space. The natural action of the group (0, ∞) by scaling is
continuous with respect to this topology, and the quotient is the space PML of
projective measured geodesic laminations. This space is homeomorphic to a sphere
of dimension 6g − 7 + 2m (see [5], [6], [25]).
The intersection number i(γ , δ) between two simple closed curves γ , δ ∈ C(S)
equals the minimal number of intersection points between representatives of the free
homotopy classes of γ , δ. This intersection function extends to a continuous pairing
i : ML × ML → [0, ∞), called the intersection form.
Measured geodesic laminations are intimately related to more classical objects
associated to Riemann surfaces, namely holomorphic quadratic differentials. A holo-
morphic quadratic differential q on a Riemann surface S assigns to each complex
coordinate z an expression of the form q(z)dz2 where q(z) is a holomorphic function
on the domain of the coordinate system, and q(z)(dz/dw)2 = q(w) for overlapping
coordinates z, w. We require that q has at most a simple pole at each puncture of S. If q
does not vanish identically, then its zeros are isolated and independent of the choice of
a complex coordinate. If p ∈ S is not a zero for q then there is a coordinate z near p,
unique up to multiplication with ±1, such that p corresponds to the origin and that
q(z) ≡ 1. Writing z = x + iy for this coordinate, the euclidean metric dx 2 + dy 2
is uniquely determined by q. The arcs parallel to the x-axis (or y-axis, respectively)
define a foliation Fh (or Fv ) on the set of regular points of q called the horizontal
(or vertical) foliation. The vertical length |dy| defines a transverse measure for the
horizontal foliation, and the horizontal length |dx| defines a transverse measure for
the vertical foliation. The foliations Fh , Fv have singularities of the same type at the
zeros of q and at the punctures of S (see [27] for more on quadratic differentials and
measured foliations).
There is a one-to-one correspondence between measured geodesic laminations and
(equivalence classes of) measured foliations on S (see [16] for a precise statement).
The pair of measured foliations defined by a quadratic differential q corresponds under
this identification to a pair of measured geodesic laminations λ = μ ∈ ML which
jointly fill up S. This means that for every η ∈ ML we have i(λ, η) + i(μ, η) > 0.
Vice versa, every pair of measured geodesic laminations λ = μ ∈ ML which
jointly fill up S defines a unique complex structure of finite type on S together with
a holomorphic quadratic differential q(λ, μ) (see [14] and the references given there)
whose area, i.e. the area of the singular euclidean metric defined by q(λ, μ), equals
i(λ, μ). If α, β are simple multi-curves on S, which means that α and β consist of
collections c = c1 ∪ · · · ∪ c ⊂ C(S) of free homotopy classes of simple closed curves
which can be realized disjointly, and if α, β jointly fill up S, then for all a > 0, b > 0
the quadratic differential q(aα, bβ) defined by the measured geodesic laminations aα,
bβ can explicitly be constructed as follows. Choose smooth representatives of α, β,
again denoted by α, β, which intersect transversely in precisely i(α, β) points; for
example, the geodesic representatives of α, β with respect to any complete hyperbolic
metric on S of finite volume have this property. For each intersection point between α,
β choose a closed rectangle in S with piecewise smooth boundary containing this point
in its interior and which does not contain any other intersection point between α, β. We
allow that some of the vertices of such a rectangle are punctures of S. These rectangles
can be chosen in such a way that they provide S with the structure of a cubical complex:
The boundary of each component D of S − α − β is a polygon with an even number of
sides which are subarcs of α, β in alternating order. If D does not contain a puncture,
then its boundary has at least four sides. Thus we can construct the rectangles in such
a way that their union is all of S and that the intersection between any two distinct
such rectangles either is a common side or a common vertex. Each rectangle from
this cubical complex has two sides which are “parallel” to α and two sides “parallel”
to β (see Section 4 of [14] for a detailed discussion of this construction).
Equip each rectangle with an euclidean metric such that the sides parallel to α are
of length b, the sides parallel to β are of length a and such that the metrics on two
rectangles coincide on a common boundary arc. These metrics define a piecewise
euclidean metric on S with a singularity of cone angle kπ ≥ 3π in the interior of each
disc component of S −α −β whose boundary consists of 2k ≥ 6 sides. The metric also

has a singularity of cone angle π at each puncture of S which is contained in a punctured
disc component with two sides. Since there are precisely i(α, β) rectangles, the area of
this singular euclidean metric on S equals abi(α, β). The line segments of this singular
euclidean metric which are parallel to α and β define singular foliations Fα , Fβ on S
with transverse measures induced by the singular metric. The metric defines a complex
structure on S and a quadratic differential q(aα, bβ) which is holomorphic for this
structure and whose horizontal and vertical foliations are just Fα , Fβ , with transverse
measures determined by the weights a and b. The assignment which associates to
a > 0 the Riemann surface structure determined by q(aα, β) is up to parametrization
the geodesic in the Teichmüller space with respect to the Teichmüller metric whose
cotangent bundle contains the differentials q(aα, β) (compare [12], [14]).
A geodesic segment for a quadratic differential q is a path in S not containing any
singularities in its interior and which is a geodesic in the local euclidean structure
defined by q. A closed geodesic is composed of a finite number of such geodesic
segments which meet at singular points of q and make an angle at least π on either
side. Every essential closed curve c on S is freely homotopic to a closed geodesic with
respect to q, and the length of such a geodesic η is the infimum of the q-lengths of any
curve freely homotopic to η (compare [26] for a detailed discussion of the technical
difficulties caused by the punctures of S) and will be called the q-length of our closed
curve c. If q = q(λ, ν) for λ, ν ∈ ML then this q-length is bounded from above by
2i(λ, c) + 2i(μ, c) (see [26]).
Thurston invented a way to understand the structure of the space of geodesic lami-
nations by squeezing almost parallel strands of such a lamination to a simple arc and
analyzing the resulting graph. The structure of such a graph is as follows.
Definition 2.4. A train track on the surface S is an embedded 1-complex τ ⊂ S

whose edges (called branches) are smooth arcs with well-defined tangent vectors at
the endpoints. At any vertex (called a switch) the incident edges are mutually tangent.
Through each switch there is a path of class C 1 which is embedded in τ and contains the
switch in its interior. In particular, the branches which are incident on a fixed switch are
divided into “incoming” and “outgoing” branches according to their inward pointing
tangent at the switch. Each closed curve component of τ has a unique bivalent switch,
and all other switches are at least trivalent. The complementary regions of the train
track have negative Euler characteristic, which means that they are different from discs
with 0, 1 or 2 cusps at the boundary and different from annuli and once-punctured discs
with no cusps at the boundary. A train track is called generic if all switches are at
most trivalent.
In the sequel we only consider generic train tracks. For such a train track τ , every
complementary component is a bordered subsurface of S whose boundary consists of
a finite number of arcs of class C 1 which come together at a finite number of cusps.
Moreover, for every switch of τ there is precisely one complementary component
containing the switch in its closure which has a cusp at the switch. A detailed account
on train tracks can be found in [25] and [23].
A geodesic lamination or a train track λ is carried by a train track τ if there is a
map F : S → S of class C 1 which is isotopic to the identity and maps λ to τ in such a
way that the restriction of its differential dF to every tangent line of λ is non-singular.
Note that this makes sense since a train track has a tangent line everywhere.
If c is a simple closed curve carried by τ with carrying map F : c → τ then c
defines a counting measure μc on τ . This counting measure is the non-negative
weight function on the branches of τ which associates to an open branch b of τ the
number of connected components of F −1 (b). A counting measure is an example for
a transverse measure on τ which is defined to be a nonnegative weight function μ on
the branches of τ satisfying the switch condition: for every switch s of τ , the sum of
the weights over all incoming branches at s is required to coincide with the sum of the
weights over all outgoing branches at s. The set V (τ ) of all transverse measures on τ
is a closed convex cone in a linear space and hence topologically it is a closed cell.
More generally, every measured geodesic lamination λ on S which is carried by τ
via a carrying map F : λ → τ defines a transverse measure on τ by assigning to a
branch b the total mass of the pre-image of b under F ; the resulting weight function is
independent of the particular choice of F . Moreover, every transverse measure for τ
can be obtained in this way (see [25]).
Definition 2.5. A train track is called recurrent if it admits a transverse measure which
is positive on every branch. A train track τ is called transversely recurrent if every
branch b of τ is intersected by an embedded simple closed curve c = c(b) ⊂ S which
intersects τ transversely and is such that S −τ −c does not contain an embedded bigon,
i.e. a disc with two corners at the boundary. A recurrent and transversely recurrent
train track is called birecurrent. A generic transversely recurrent train track which
carries a complete geodesic lamination is called complete.
For every recurrent train track τ , measures which are positive on every branch
define the interior of the convex cone V (τ ) of all transverse measures. A complete
train track is birecurrent [9].
A half-branch b̃ in a generic train track τ incident on a switch v is called large if
the switch v is trivalent and if every arc ρ : (−ε, ε) → τ of class C 1 which passes
through v meets the interior of b̃. A branch b in τ is called large if each of its two
half-branches is large; in this case b is necessarily incident on two distinct switches
(for all this, see [25]).
There is a simple way to modify a transversely recurrent train track τ to another
transversely recurrent train track. Namely, if e is a large branch of τ then we can
perform a right or left split of τ at e as shown in Figure A below. The split τ of a train
track τ is carried by τ . If τ is complete and if the complete geodesic lamination λ is
carried by τ , then for every large branch e of τ there is a unique choice of a right or
left split of τ at e with the property that the split track τ carries λ, and τ is complete.
In particular, a complete train track τ can always be split at any large branch e to a
complete train track τ ; however there may be a choice of a right or left split at e such
that the resulting train track is not complete any more (compare p. 120 in [25]).
right split
a d
e
b c
left split
Figure A
In the sequel we denote by T T the collection of all isotopy classes of complete

train tracks on S. A sequence (τi ) ⊂ T T of complete train tracks is called a splitting
sequence if τi+1 can be obtained from τi by a single split at some large branch e.
3 Hyperbolicity of the complex of curves

In this section we present a proof of hyperbolicity of the curve graph using the main
strategy of Masur and Minsky [19] and Bowditch [2] in a modified form. The first step
consists in guessing a family of uniform quasi-geodesics in the curve graph connecting
any two points. Here a p-quasi-geodesic for some p > 1 is a curve c : [a, b] → C(S)
which satisfies
d(c(s), c(t))/p − p ≤ |s − t| ≤ pd(c(s), c(t)) + p for all s, t ∈ [a, b].
Note that a quasi-geodesic does not have to be continuous. In a hyperbolic geodesic
metric space, every p-quasi-geodesic is contained in a fixed tubular neighborhood of
any geodesic joining the same endpoints, so the δ-thin triangle condition also holds for
triangles whose sides are uniform quasi-geodesics [3]. As a consequence, for every
triangle in a hyperbolic geodesic metric space with uniform quasi-geodesic sides there
is a “midpoint” whose distance to each side of the triangle is bounded from above by a
universal constant. The second step of the proof consists in finding such a midpoint for
triangles whose sides are curves of the distinguished curve family. This is then used
in a third step to establish the δ-thin triangle condition for the distinguished family of
curves and derive from this hyperbolicity of C(S). By abuse of notation, in the sequel
we simply write α ∈ C(S) if α is a free homotopy class of an essential simple closed
curve on S, i.e. if α is a vertex of C(S).
We begin with defining a map from the set T T of complete train tracks on S into
C(S). For this we call a transverse measure μ for a complete train track τ a vertex cycle
[19] if μ spans an extreme ray in the convex cone V (τ ) of all transverse measures on τ .
Up to scaling, every vertex cycle μ is a counting measure of a simple closed

curve c which is carried by τ [19]. Namely, the switch conditions are a family of
linear equations with integer coefficients for the transverse measures on τ . Thus an
extreme ray is spanned by a nonnegative rational solution which can be scaled to a
nonnegative integral solution. From every integral transverse measure μ for τ we
can construct a unique simple weighted multi-curve, i.e. a simple multi-curve together
with a family of weights for each of its components, which is carried by τ and whose
counting measure coincides with μ as follows. For each branch b of τ draw μ(b)
disjoint arcs parallel to b. By the switch condition, the endpoints of these arcs can
be connected near the switches in a unique way so that the resulting family of arcs
does not have self-intersections. Let c be the simple multi-curve consisting of the free
homotopy classes of the connected components of the resulting curve c̃. To each such
homotopy class associate the number of components of c̃ in this class as a weight.
The resulting simple weighted multi-curve is carried by τ , and its counting measure
equals μ. Thus if there are at least two components of c̃ which are not freely homotopic
then the weighted counting measures of these components determine a decomposition
of μ into transverse measures for τ which are not multiples of μ. This is impossible
if μ is a vertex cycle. Hence c consists of a single component and up to scaling, μ is
the counting measure of a simple closed curve on S.
A simple closed curve which is carried by τ , with carrying map F : c → τ , defines
a vertex cycle for τ only if F (c) passes through every branch of τ at most twice, with
different orientation (Lemma 2.2 of [7]). In particular, the counting measure μc of a
simple closed curve c which defines a vertex cycle for τ satisfies μc (b) ≤ 2 for every
branch b of τ .
In the sequel we mean by a vertex cycle of a complete train track τ an integral
transverse measure on τ which is the counting measure of a simple closed curve c on S
carried by τ and which spans an extreme ray of V (τ ); we also use the notion vertex
cycle for the simple closed curve c. Since the number of branches of a complete train
track on S only depends on the topological type of S, the number of vertex cycles for
a complete train track on S is bounded by a universal constant (see [19] and [7]).
The following observation of Penner and Harer [25] is essential for all what follows.
3g−3+m
Denote by MC(S) the space of all simple multi-curves on S. Let P = i=1 γi ∈
MC(S) be a pants decomposition for S, i.e. a simple multi-curve with the maximal
number of components. Then there is a special family of complete train tracks with
the property that each pants curve γi admits a closed neighborhood A diffeomorphic
to an annulus and such that τ ∩ A is diffeomorphic to a standard twist connector
depicted in Figure B. Such a train track clearly carries each pants curve from the
pants decomposition P as a vertex cycle; we call it adapted to P . For every complete
geodesic lamination λ there is a train track τ adapted to P which carries λ ([25], see
also [9], [7]).
Since every simple multi-curve is a subset of a pants decomposition of S, we can
conclude.
Figure B
Lemma 3.1 ([10]). For every pair (α, β) ∈ MC(S) × MC(S) there is a splitting
sequence (τi )0≤i≤m ⊂ T T of complete train tracks with the property that τ0 is adapted
to a pants decomposition Pα ⊃ α and that each component of β is a vertex cycle for τm .
We call a splitting sequence as in the lemma an α → β-splitting sequence. Note

that such a sequence is by no means unique.
The distance in C(S) between two simple closed curves α, β is bounded from
above by i(α, β) + 1 (Lemma 1.1 of [2] and Lemma 2.1 of [19]). In particular, there
is a number D0 > 0 with the following property. Let τ, τ ∈ T T and assume that
τ is obtained from τ by at most one split. Then the distance in C(S) between any
vertex cycle of τ and any vertex cycle of τ is at most D0 (see [19] and the discussion
following Corollary 2.3 in [7]).
Define a map : T T → C(S) by assigning to a train track τ ∈ T T a vertex cycle
(τ ) for τ . By our above discussion, for any two choices , of such a map we have
d((τ ), (τ )) ≤ D0 for all τ ∈ T T . Images under the map of splitting sequences
then define a family of curves in C(S) which connect any pair of points in a D0 -dense
subset of C(S) × C(S), equipped with the product metric. As a consequence, we can
use such images of splitting sequences as our guesses for uniform quasi-geodesics.
It turns out that up to parametrization, these curves are indeed p-quasi-geodesics in
C(S) for a universal number p > 0 only depending on the topological type of the
surface S ([20], see also [7]).
To explain this fact we use the following construction of Bowditch [2]. For multi-
curves α, β ∈ MC(S) which jointly fill up S, i.e. which cut S into components which
are homeomorphic to discs and once punctured discs, and for a number a > 0 let
q(aα, β/ai(α, β)) be the area one quadratic differential whose horizontal foliation
corresponds to the measured geodesic lamination aα and whose vertical measured
foliation corresponds to the measured geodesic lamination β/ai(α, β). For r > 0
define
La (α, β, r) = {γ ∈ C(S) | max{ai(γ , α), i(γ , β)/ai(α, β)} ≤ r}.
Then La (α, β, r) is contained in the set of all simple closed curves on S whose
q(aα, β/ai(α, β))-length does not exceed 2r. Note that we have La (α, β, r) =
L1/ai(α,β) (β, α, r) for all r > 0, moreover α ∈ La (α, β, r) for every component
α of α and every sufficiently large a > 0, and β ∈ La (α, β, r) for every component
β of β and every sufficiently small a > 0. Thus for fixed r > 0 we can think of a
suitably chosen assignment which associates to a number s > 0 a point in Ls (α, β, r)
as a curve in C(S) connecting a component of β to a component of α (provided, of
course, that the sets Ls (α, β, r) are non-empty). Lemma 2.5 of [7] links such curves
to splitting sequences.
Lemma 3.2 ([7]). There is a number k0 ≥ 1 with the following property. Let P be a
pants decomposition of S, let α ∈ MC(S) be such that α and P jointly fill up S and
let (τi )0≤i≤m ⊂ T T be a P → α-splitting sequence. Then there is a non-decreasing
surjective function κ : (0, ∞) → {0, . . . , m} such that κ(s) = 0 for all sufficiently
small s > 0, κ(s) = m for all sufficiently large s > 0 and that for all s ∈ (0, ∞) there
is a vertex cycle of τκ(s) which is contained in Ls (α, P , k0 ).
Since for every multi-curve α ∈ MC(S) and every pants decomposition P of S

there is a P → α-splitting sequence, we conclude that for every k ≥ k0 and every
s > 0 the set Ls (α, P , k) is non-empty. To obtain a control of the size of these sets,
Bowditch [2] uses the following observation (Lemma 4.1 in [2]) whose first part was
earlier shown by Masur and Minsky (Lemma 5.1 of [19]).
Lemma 3.3 ([2]). There is a number k1 ≥ k0 with the following property. For all
α, β ∈ MC(S) which jointly fill up S and every a ∈ (0, ∞) there is a δ ∈ La (α, β, k1 )
such that for every γ ∈ MC(S) we have i(δ, γ ) ≤ k1 max{ai(α, γ ), i(γ , β)/ai(α, β)}.
In particular, for every R > 0, for all α, β ∈ MC(S) and for every a > 0 the diameter
of the set La (α, β, R) is not bigger than 2k1 R + 1.
Proof. In [19], [2] it is shown that there is a number ν > 0 only depending on the
topological type of S and there is an embedded essential annulus in S whose width
with respect to the piecewise euclidean metric defined by the quadratic differential
q(aα, β/ai(α, β)) is at least ν. This means that the distance between the boundary
circles of the annulus is at least ν. Assuming the existence of such an annulus, let δ
be its core-curve. Then for every simple closed curve γ on S and for every essential
intersection of γ with δ there is a subarc of γ which crosses through this annulus and
hence whose length is at least ν; moreover, different subarcs of γ corresponding to
different essential intersections between γ and δ are disjoint. Thus the length with
respect to the singular euclidean metric on S of any simple closed curve γ on S is
at least νi(γ , δ). On the other hand, by construction the minimal length with respect
to this metric of a curve in the free homotopy class of γ is bounded from above by
2 max{ai(α, γ ), i(β, γ )/ai(α, β)} and therefore the core curve δ of the annulus has
the properties stated in the first part of our lemma (see [2]).
The second part of the lemma is immediate from the first. Namely, let α, β ∈
MC(S) and let a > 0. Choose δ ∈ La (α, β, k1 ) which satisfies the properties stated
in the first part of the lemma. If γ ∈ La (α, β, R) for some R > 0 then we have
i(γ , δ) ≤ k1 R and hence d(γ , δ) ≤ k1 R + 1.
As an immediate consequence of Lemma 3.2 and Lemma 3.3 we observe that

there is a universal number D1 > 0 with the following property. Let P be a pants
decomposition for S, let β ∈ MC(S) and let (τi )0≤i≤m ⊂ T T be any P → β-splitting
sequence;
then the Hausdorff distance in C(S) between the sets {(τi ) | 0 ≤ i ≤ m}
and a>0 La (β, P , k1 ) is at most D1 /16. If c > 0 and if j ≤ m is such that there is a
vertex cycle γ for τ (j ) which is contained in Lc (β, P , k1 ) then the splitting sequence
i )0≤i≤j is a P → γ -splitting
(τ sequence and hence the Hausdorff distance between
L
a>0 a (γ , P , k 1 ) and a≥c a (β, P , k1 ) is at most D1 /8. Moreover, for every
L
β ∈ C(S) and every simple multi-curve Q containing β as a component the Hausdorff-
distance between the sets a>0 La (β, P , k1 ) and a>0 La (Q, P , k1 ) is at most D1 /8.
Thus if Q, Q are pants decompositions for S containing a common curve β ∈ C(S)
then the Hausdorff distance between a La (Q, P , k1 ) and a La (Q , P , k1 ) is at
most D1 /4.
On the other hand, for multi-curves P , Q ∈ MC(S) we have

La (P , Q, k1 ) = La (Q, P , k1 ).
a>0 a>0
Therefore from two applications of our above consideration we obtain the following.
Let α, β ∈ C(S) and let P , P , Q, Q be
any pants decompositions for S containing α,
β; then the Hausdorff distance between a>0 La (P , Q, k1 ) and a>0 La (P , Q , k1 )
is not bigger than D1 /2. By our choice of D1 this implies that the Hausdorff distance
between the images under of any α → β- or β → α-splitting sequences is bounded
from above by D1 .
Now let α, β, γ ∈ C(S) be such that their pairwise distance in C(S) is at least 3;
then any two of these curves jointly fill up S. Choose pants decompositions Pα ,
Pβ , Pγ containing α, β, γ . Then there are unique numbers a, b, c > 0 such that
abi(Pα , Pβ ) = bci(Pβ , Pγ ) = aci(Pγ , Pα ) = 1. By construction, we have
La (Pα , Pβ , k1 ) = Lb (Pβ , Pα , k1 ), Lb (Pβ , Pγ , k1 ) = Lc (Pγ , Pβ , k1 )
and Lc (Pγ , Pα , k1 ) = La (Pα , Pγ , k1 ). Choose a point δ ∈ La (Pα , Pβ , k1 ) such that
for every ζ in MC(S) we have
i(δ, ζ ) ≤ k1 max{ai(Pα , ζ ), i(ζ, Pβ )/ai(Pα , Pβ )};
such a point exists by Lemma 3.3. Applying this inequality to ζ = Pγ yields
ci(δ, Pγ ) ≤ k1 . For ζ = Pβ we obtain
i(δ, Pβ )/ci(Pγ , Pβ ) ≤ ai(Pα , Pβ )/ci(Pγ , Pβ ) = 1,
and for ζ = Pα we obtain
i(δ, Pα )/ci(Pγ , Pα ) ≤ 1/aci(Pγ , Pα ) = 1.
Therefore we have δ ∈ Lc (Pγ , Pβ , k1 ) ∩ La (Pα , Pβ , k1 ) ∩ Lc (Pγ , Pα , k1 ). Together

with Lemma 3.2 and our above remark we conclude that there is a universal constant
D2 > 0 such that the distance between φ(α, β, γ ) = δ and the image under of any
α → β-splitting sequence, any α → γ -splitting sequence and any γ → β-splitting
sequence is bounded from above by D2 .
We use the map φ to derive the δ-thin triangle condition for triangles whose sides
are images under the map of splitting sequences in T T .
Lemma 3.4. There is a number D3 > 0 with the following property. Let α, β, γ ∈
C(S) and let a, b, c be the image under of a β → γ , γ → α, α → β-splitting
sequence. Then the D3 -neighborhood of a ∪ b contains c.
Proof. Let α, β, γ ∈ C(S) and assume that d(β, γ ) ≤ p for some p > 0. Let
(τi )0≤i≤m be an α → β-splitting sequence and let (ηj )0≤j ≤ be an α → γ -splitting
sequence; if D1 > 0 is as above then the Hausdorff distance between {(τi ) | 0 ≤
i ≤ m} and {(ηj ) | 0 ≤ j ≤ } is at most 2pD1 . Namely, we observed that the
Hausdorff distance between the image under of any two α → β-splitting sequences
is bounded from above by D1 . Moreover, if d(β, γ ) = 1 then β ∪ γ ∈ MC(S)
and hence there is an α → β-splitting sequence which also is an α → γ -splitting
sequence. Thus the statement of the corollary holds for p = 1, and the general case
follows from a successive application of this fact for the points on a geodesic in C(S)
connecting β to γ .
Now let α, β, γ ∈ C(S) be arbitrary points whose pairwise distance is at least 3.
Let again (τi )0≤i≤m be an α → β-splitting sequence. By the definition of φ and the
choice of the constant D2 > 0 above there is some i0 ≤ m such that the distance
between (τi0 ) and φ(α, β, γ ) is at most D2 . Let (ηj )0≤j ≤ be any α → φ(α, β, γ )-
splitting sequence. By our above consideration, the Hausdorff distance between
{(τi ) | 0 ≤ i ≤ i0 } and {(ηj ) | 0 ≤ j ≤ } is at most 2D1 D2 . Similarly,
let (ζj )0≤j ≤n be an α → γ -splitting sequence. Then there is some j0 > 0 such that
d((ζj0 ), φ(α, β, γ )) ≤ D2 . By our above argument, the Hausdorff distance between
the sets {(τi ) | 0 ≤ i ≤ i0 } and {(ζj ) | 0 ≤ j ≤ j0 } is at most 4D1 D2 .
As a consequence, there are numbers a(α, β) > 0 and a(α, γ ) > 0 such that
φ(α, β, γ ) ∈ La(α,β) (β, α, k1 ) ∩ La(α,γ
) (γ , α, k1 ) and that the Hausdorff distance
between a≥a(α,β) La (β, α, k1 ) and a≥a(α,γ ) La (γ , α, k1 ) is at most 6D1 D2 . The
same argument, applied to a β → α-splitting sequence and a β → γ -splitting se-
quence,
shows that a≤a(α,β) La (α, β, k1 ) is contained in the 6D1 D2 -neighborhood
of a La (β, γ , k1 ). Then {(τi ) | 0 ≤ i ≤ m} is contained in the 12D1 D2 -
neighborhood of the union of the image under of a γ → α-splitting sequence
and a β → γ -splitting sequence. This shows the lemma.
Hyperbolicity of the curve graph now follows from Lemma 3.4 and the following
criterion.
Proposition 3.5. Let (X, d) be a geodesic metric space. Assume that there is a number
D > 0 and for every pair of points x, y ∈ X there is an arc η(x, y) : [0, 1] → X
connecting η(x, y)(0) = x to η(x, y)(1) = y so that the following conditions are
satisfied.
(1) If d(x, y) ≤ 1 then the diameter of η(x, y)[0, 1] is at most D.
(2) For x, y ∈ X and 0 ≤ s ≤ t ≤ 1, the Hausdorff distance between η(x, y)[s, t]
and η(η(x, y)(s), η(x, y)(t))[0, 1] is at most D.
(3) For any x, y, z ∈ X the set η(x, y)[0, 1] is contained in the D-neighborhood of
η(x, z)[0, 1] ∪ η(z, y)[0, 1].
Then (X, d) is δ-hyperbolic for a number δ > 0 only depending on D.
Proof. Let (X, d) be a geodesic metric space. Assume that there is a number D > 0
and there is a family of paths η(x, y) : [0, 1] → X, one for every pair of points
x, y ∈ X, which satisfy the hypotheses in the statement of the proposition. To show
hyperbolicity for X it is then enough to show the existence of a constant κ > 0
such that for all x, y ∈ X and every geodesic ν : [0, ] → X connecting x to y, the
Hausdorff-distance between ν[0, ] and η(x, y)[0, 1] is at most κ. Namely, if this is
the case then for every geodesic triangle with sides a, b, c the side a is contained in
the 3κ + D-neighborhood of b ∪ c.
To show the existence of such a constant κ > 0, let x, y ∈ X and let c : [0, 2k ] → X
be any path of length (c) = 2k parametrized by arc length connecting x to y.
Write η1 = η(c(0), c(2k−1 )) and write η2 = η(c(2k−1 ), c(2k )). By our assump-
tion, the D-neighborhood of η1 ∪ η2 contains η(c(0), c(2k )). Repeat this construction
with the points c(2k−2 ), c(3 · 2k−2 ) and the arcs η1 , η2 . Inductively we conclude
that the path η(c(0), c(2k )) is contained in the (log2 (c))D-neighborhood of a path
c̃ : [0, 2k ] → C(S) whose restriction to each interval [m − 1, m] (m ≤ 2k ) equals
up to parametrization the arc η(c(m − 1), c(m)). Since d(c(m − 1), c(m)) ≤ 1, by
assumption the diameter of each of the sets η(c(m − 1), c(m))[0, 1] is bounded from
above by D and therefore the arc η(c(0), c(2k )) is contained in the (log2 (c))D + D-
neighborhood of c[0, 2k ].
Now let c : [0, k] → X be a geodesic connecting c(0) = x to c(k) = y which
is parametrized by arc length. Let t > 0 be such that η(x, y)(t) has maximal dis-
tance, say χ , to c[0, k]. Choose some s > 0 such that d(c(s), η(x, y)(t)) = χ and
let t1 < t < t2 be such that d(η(x, y)(t), η(x, y)(tu )) = 2χ (u = 1, 2). In the
case that there is no t1 ∈ [0, t) (or t2 ∈ (t, 1]) with d(η(x, y)(t), η(x, y)(t1 )) ≥ 2χ
(or d(η(x, y)(t), η(x, y)(t2 )) ≥ 2χ) we choose t1 = 0 (or t2 = 1). By our choice
of χ, there are numbers su ∈ [0, k] such that d(c(su ), η(x, y)(tu )) ≤ χ (u = 1, 2).
Then the distance between c(s1 ) and c(s2 ) is at most 6χ . Compose the subarc
c[s1 , s2 ] of c with a geodesic connecting η(x, y)(t1 ) to c(s1 ) and a geodesic con-
necting c(s2 ) to η(x, y)(t2 ). We obtain a curve ν of length at most 8χ . By our
above observation, the (log2 (8χ))D + D-neighborhood of this curve contains the arc
η(η(x, y)(t1 ), η(x, y)(t2 )). However, the Hausdorff distance between η(x, y)[t1 , t2 ]
and η(η(x, y)(t1 ), η(x, t)(t2 )) is at most D and therefore the (log2 (8χ ))D + 2D-
neighborhood of the arc ν contains η(x, y)[t1 , t2 ]. But the distance between η(t) and
our curve ν equals χ by construction and hence we have χ ≤ (log2 (8χ ))D + 2D. In
other words, χ is bounded from above by a universal constant κ1 > 0, and η(x, y) is
contained in the κ1 -neighborhood of the geodesic c.
A similar argument also shows that the 3κ1 -neighborhood of η(x, y) contains
c[0, k]. Namely, by the above consideration, for every t ≤ 1 the set A(t) =
{s ∈ [0, k] | d(c(s), η(x, y)(t)) ≤ κ1 } is a non-empty closed subset of [0, k]. The
diameter of the sets A(t) is bounded from above by 2κ1 . Assume to the contrary that
c[0, k] is not contained in the 3κ1 -neighborhood of η(x, y). Then there is a subinter-
val [a1 , a2 ] ⊂ [0, k] of length a2 − a1 ≥ 4κ1 such that d(c(s), η(x, y)[0, 1]) > κ1
for every s ∈ (a1 , a2 ). Since d(c(s), c(t)) = |s − t| for all s, t we conclude
that for every t ∈ [0, 1] the set A(t) either is entirely contained in [0, a1 ] or it
is entirely contained in [a2 , k]. Define C1 = {t ∈ [0, 1] | A(t) ⊂ [0, a1 ]} and
C2 = {t ∈ [0, 1] | A(t) ⊂ [a2 , 1]}. Then the sets C1 , C2 are disjoint and their union
equals [0, 1]; moreover, we have 0 ∈ C1 and 1 ∈ C2 . On the other hand, the sets Ci
are closed. Namely, let (ti ) ⊂ C1 be a sequence converging to some t ∈ [0, 1]. Let
si ∈ A(ti ) and assume after passing to a subsequence that si → s ∈ [0, a1 ]. Now
κ1 ≥ d(c(si ), η(x, y)(ti )) → d(c(s), η(x, y)(t)) and therefore s ∈ A(t) and hence
t ∈ C1 . However, [0, 1] is connected and hence we arrive at a contradiction. In other
words, the geodesic c is contained in the 3κ1 -neighborhood of η(x, y). This completes
the proof of the proposition.
As an immediate corollary, we obtain the following result.
Theorem 3.6 ([19], [2]). The curve graph is hyperbolic.
Proof. Let : T T → C(S) be as before. For α, β ∈ C(S) choose an α → β-splitting

sequence (τi )0≤i≤m . Define an arc η(α, β) : [0,1] → C(S) by requiring that for 1 ≤
i i+1
i ≤ m the restriction of η(α, β) to the interval m+2 , m+2 is a geodesic connecting
m−1
(τi−1 ) to (τi ) and that the restriction of η(α, β) to 0, m+2 1
(or m−2 , 1 ) is a
geodesic connecting α to (τ0 ) (or (τm ) to β).
We claim that this family of arcs satisfy the assumptions in Proposition 3.5. Namely,
we observed before that for all α, β ∈ C(S), the Hausdorff distance between the image
under of any two α → β-splitting sequences is bounded from above by a universal
constant. Now if (τi )0≤i≤m is any splitting sequence, then for all 0 ≤ k ≤ ≤ m
the sequence (τi )k≤i≤ is a (τk ) → (τ )-splitting sequence and hence our curve
system satisfies the second condition in Proposition 3.5.
Moreover, curves α, β ∈ C(S) with d(α, β) = 1 can be realized disjointly, and
α ∪ β is a multi-curve. For such a pair of curves we can choose a constant α → β-
splitting sequence; hence our curve system also satisfies the first condition stated in
Proposition 3.5. Finally, the third condition was shown to hold in Lemma 3.4.
Now hyperbolicity of the curve graph follows from Proposition 3.5.
A curve c : [0, m] → C(S) is called an unparametrized p-quasi-geodesic for some

p > 1 if there is a homeomorphism ρ : [0, u] → [0, m] for some u > 0 such that
d(c(ρ(s)), c(ρ(t)))/p − p ≤ |s − t| ≤ pd(c(ρ(s)), c(ρ(t))) + p
for all s, t ∈ [0, u]. We define a map c : {0, . . . , m} → C(S) to be an unparametrized
q-quasi-geodesic if this is the case for the curve c̃ whose restriction to each interval
[i, i + 1) coincides with c(i). The following observation is immediate from Proposi-
tion 3.5 and its proof.
Corollary 3.7 ([20], [7]). There is a number p > 0 such that the image under of
an arbitrary splitting sequence is an unparametrized p-quasi-geodesic.
Proof. By Proposition 3.5, Theorem 3.6 and their proofs, there is a universal number
D > 0 with the property that for every splitting sequence (τi )0≤i≤m and every geodesic
c : [0, s] → C(S) connecting c(0) = (τ0 ) to c(s) = (τm ), the Hausdorff distance
between the sets {(τi ) | 0 ≤ i ≤ m} and c[0, s] is at most D. From this the corollary
is immediate.
4 Geometry of Teichmüller space

In this section, we relate the geometry of the curve graph to the geometry of Teichmüller
space equipped with the Teichmüller metric. For this we first define a map : Tg,m →
C(S) as follows. By a well-known result of Bers (see [4]) there is a number χ > 0
only depending on the topological type of S such that for every complete hyperbolic
metric on S of finite volume there is a pants decomposition P for S which consists of
simple closed geodesics of length at most χ. Since the distance between any two points
α, β ∈ C(S) is bounded from above by i(α, β) + 1, the collar lemma for hyperbolic
surfaces (see [4]) implies that the diameter in C(S) of the set of simple closed curves
whose length with respect to the fixed metric is at most χ is bounded from above by
a universal constant D > 0. Define : Tg,m → C(S) by assigning to a finite volume
hyperbolic metric h on S a simple closed curve (h) whose h-length is at most χ .
Then for any two maps , with this property and every h ∈ Tg,m the distance
in C(S) between (h) and (h) is at most D. Moreover, the map is coarsely
equivariant with respect to the action of the mapping class group Mg,m on Tg,m and
C(S): For every h ∈ Tg,m and every φ ∈ Mg,m we have d((φ(h)), φ((h))) ≤ D.
The following result is due to Masur and Minsky (Theorem 2.6 and Theorem 2.3
of [19]). For its formulation, let dT be the distance function on Tg,m induced by the
Teichmüller metric.
Theorem 4.1 ([19]). (1) There is a number a > 0 such that

d(h, h ) ≤ a dT (h, h ) + a
for all h, h ∈ Tg,m .
(2) There is a number p̃ > 0 with the following property. Let γ : (−∞, ∞) → Tg,m
be any Teichmüller geodesic; then the assignment t → (γ (t)) is an unparametrized
p̃-quasi-geodesic in C(S).
Proof. Let γ : (−∞, ∞) → Tg,m be any Teichmüller geodesic parametrized by arc
length. Then the cotangent of γ at t = 0 is a quadratic differential q of area one defined
by a pair (λ, μ) ∈ ML × ML of measured geodesic laminations which jointly fill
up S. The cotangent of γ at t is given by the quadratic differential q(t) defined by the
pair (et λ, e−t μ). For k1 > 0 as in Lemma 3.3 and t ∈ R let ζ (t) ∈ C(S) be a curve
whose q(t)-length is at most 2k1 . For every β ∈ [0, 1] the q(t + β)-length of ζ (t) is
bounded from above by 2ek1 and therefore by Lemma 3.3, for every t the distance in
C(S) between ζ (t) and ζ (t +β) is bounded from above by a universal constant k2 > 0.
In particular, the assignment t → ζ (t) satisfies d(ζ (s), ζ (t)) ≤ k2 |s − t| + k2 . Hence
for the proof of our lemma, we only have to show that there is a constant k3 > 0 such
that for every h ∈ Tg,m and every holomorphic quadratic differential q of area one
for h, the distance between (h) and a curve on S whose q-length is bounded from
above by 2k1 is uniformly bounded.
Thus let h be a complete hyperbolic metric of finite volume and let q be a holo-
morphic quadratic differential for h of area one. By the collar lemma of hyperbolic
geometry, a simple closed geodesic c for h whose length is bounded from above by χ is
the core curve of an embedded annulus A whose modulus is bounded from below by a
universal constant ε > 0; we refer to [27] for a definition of the modulus of an annulus
and its properties. Then the extremal length of the core curve of A is bounded from
above by a universal constant m > 0. Now the √ area of q equals one and therefore the
q-length of the core curve c does not exceed m by the definition of extremal length
(see e.g. [22]). In other words, the q-length of the curve (h) is uniformly bounded
which together with Lemma 3.3 implies our claim. The theorem follows.
There are also Teichmüller geodesics in Teichmüller space which are mapped by
to parametrized quasi-geodesics in C(S). For their characterization, denote for
ε > 0 by Tg,mε the subset of Teichmüller space consisting of all marked hyperbolic
ε
metrics for which the length of the shortest closed geodesic is at least ε. The set Tg,m
is invariant under the action of the mapping class group and projects to a compact
subset of moduli space. Moreover, every compact subset of moduli space is contained
ε for some ε > 0.
in the projection of Tg,m
Cobounded Teichmüller geodesics, i.e. Teichmüller geodesics which project to
a compact subset of moduli space, relate the geometry of Teichmüller space to the
geometry of the curve graph. We have
Proposition 4.2 ([8]). The image under of a Teichmüller geodesic γ : R → Tg,m
is a parametrized quasi-geodesic in C(S) if and only if there is some ε > 0 such that
ε .
γ (R) ⊂ Tg,m
Minsky [21] discovered earlier that the Teichmüller metric near a cobounded
geodesic line has properties similar to properties of a hyperbolic geodesic metric
space. Namely, for a Teichmüller geodesic γ : R → Tg,m ε the map which associates
to a point h ∈ Tg,m a point on γ (R) which minimizes the Teichmüller distance is

coarsely Lipschitz and contracts distances in a way which is similar to the contraction
property of the closest point projection from a δ-hyperbolic geodesic metric space to
any of its bi-infinite geodesics.
A hyperbolic geodesic metric space X admits a Gromov boundary which is defined
as follows. Fix a point p ∈ X and for two points x, y ∈ X define the Gromov product
(x, y)p = 21 (d(x, p) + d(y, p) − d(x, y)). Call a sequence (xi ) ⊂ X admissible if
(xi , xj )p → ∞ (i, j → ∞). We define two admissible sequences (xi ), (yi ) ⊂ X to be
equivalent if (xi , yi )p → ∞. Since X is hyperbolic, this defines indeed an equivalence
relation (see [3]). The Gromov boundary ∂X of X is the set of equivalence classes of
admissible sequences (xi ) ⊂ X. It carries a natural Hausdorff topology. For the curve
graph, the Gromov boundary was determined by Klarreich [15] (see also [7]).
For the formulation of Klarreich’s result, we say that a minimal geodesic lamination
λ fills up S if every simple closed geodesic on S intersects λ transversely, i.e. if every
complementary component of λ is an ideal polygon or a once punctured ideal polygon
with geodesic boundary [5]. For any minimal geodesic lamination λ which fills up S,
the number of geodesic laminations μ which contain λ as a sublamination is bounded
by a universal constant only depending on the topological type of the surface S.
Namely, each such lamination μ can be obtained from λ by successively subdividing
complementary components P of λ which are different from an ideal triangle or a
once punctured monogon by adding a simple geodesic line which either connects two
non-adjacent cusps or goes around a puncture. Note that every leaf of μ which is not
contained in λ is necessarily isolated in μ.
Recall that the space L of geodesic laminations on S equipped with the restriction
of the Hausdorff topology for compact subsets of S is compact and metrizable. It
contains the set B of minimal geodesic laminations which fill up S as a subset which
is neither closed nor dense. We define on B a new topology which is coarser than
the restriction of the Hausdorff topology as follows. Say that a sequence (λi ) ⊂ L
converges in the coarse Hausdorff topology to a minimal geodesic lamination μ which
fills up S if every accumulation point of (λi ) with respect to the Hausdorff topology
contains μ as a sublamination. Define a subset A of B to be closed if and only if
for every sequence (λi ) ⊂ A which converges in the coarse Hausdorff topology to a
lamination λ ∈ B we have λ ∈ A. We call the resulting topology on B the coarse
Hausdorff topology. The space B is not locally compact. Using this terminology,
Klarreich’s result [15] can be formulated as follows.
Theorem 4.3 ([15], [7]). (1) There is a natural homeomorphism of B equipped

with the coarse Hausdorff topology onto the Gromov boundary ∂C(S) of the complex
of curves C(S) for S.
(2) For μ ∈ B a sequence (ci ) ⊂ C(S) is admissible and defines the point
(μ) ∈ ∂C(S) if and only if (ci ) converges in the coarse Hausdorff topology to μ.
Recall that every Teichmüller geodesic in Tg,m is uniquely determined by a pair

of projective measured laminations which jointly fill up S. The following corollary is
immediate from Theorem 4.1 and Theorem 4.3 with p̃ > 0 as in Theorem 4.1.
Corollary 4.4. Let λ, μ ∈ PML be such that their supports are minimal and fill
up S. Then the image under of the unique Teichmüller geodesic in Tg,m determined
by λ and μ is a bi-infinite unparametrized p̃-quasi-geodesic in C(S), and every bi-
infinite unparametrized p̃-quasi-geodesic in C(S) is contained in a uniformly bounded
neighborhood of a curve of this form.
Our above discussion also gives information on images under of a convergent

sequence of geodesic lines in Teichmüller space. Namely, if (γi ) is such a sequence of
Teichmüller geodesic lines converging to a Teichmüller geodesic which is determined
by a pair of projective measured geodesic laminations (α, β) so that the support of α
does not fill up S then there is a curve ζ ∈ C(S), a number m > 0 and a sequence
j (i) → ∞ such that d((γi [0, j (i)]), ζ ) ≤ m.
On the other hand, the image under of “most” Teichmüller geodesics are un-
parametrized quasi-geodesics of infinite diameter which are not parametrized quasi-
geodesics. For example, let λ ∈ PML be a projective measured geodesic lamination
whose support λ0 is minimal and fills up S but is not uniquely ergodic. This means
that the dimension of the space of transverse measures supported in λ0 is at least 2.
Let γ : [0, ∞) → Tg,m be a Teichmüller geodesic ray determined by a quadratic dif-
ferential whose horizontal foliation corresponds to λ. By a result of Masur [17], the
projection of γ to moduli space eventually leaves every compact set. On the other
hand, since λ0 is minimal and fills up S the points γ (t) converge as t → ∞ to λ viewed
as a point in the Thurston boundary of the Thurston compactification of Teichmüller
space [18] (compare also [6] for the construction of the Thurston compactification).
By the definition of the map , the projective measured geodesic laminations defined
by the curves γ (t) converge as t → ∞ to λ and therefore the curves γ (t) converge
in the coarse Hausdorff topology to λ0 . By Theorem 4.3, this implies that the diameter
of γ [0, ∞) is infinite.
References
[1] F. Bonahon, Geodesic laminations on surfaces. In Laminations and foliations in dynamics,

geometry and topology (Stony Brook, NY, 1998), Contemp. Math. 269, Amer. Math. Soc.,
Providence, RI, 2001, 1–37. 450
[2] B. Bowditch, Intersection numbers and the hyperbolicity of the curve complex. J. Reine
Angew. Math. 598 (2006), 105–129. 448, 454, 456, 457, 461
[3] M. Bridson, A. Haefliger, Metric spaces of non-positive curvature, Grundlehren Math.
Wiss. 319, Springer-Verlag, Berlin 1999. 454, 464
[4] P. Buser, Geometry and spectra of compact Riemann surfaces. Progr. Math. 106,
Birkhäuser, Boston, MA, 1992. 449, 462
[5] R. Canary, D. Epstein, P. Green, Notes on notes of Thurston. In Analytical and geometric
aspects of hyperbolic space (D. Epstein, ed.), London Math. Soc. Lecture Note Ser. 111,
Cambridge University Press, Cambridge 1987, 3–92. 450, 464
Astérisque 66–67 (1979). 450, 465
[7] U. Hamenstädt, Train tracks and the Gromov boundary of the complex of curves. In Spaces
of Kleinian groups (Y. Minsky, M. Sakuma, C. Series, eds.), London Math. Soc. Lecture
Note Ser. 329, Cambridge University Press, Cambridge 2006, 187–207. 455, 456, 457,
462, 464
[8] U. Hamenstädt, Word hyperbolic extensions of surface groups. Preprint, 2005; arXiv:math.
GT/0505244. 463
[9] U. Hamenstädt, Geometry of the mapping class groups I: Boundary amenability. Preprint,
2006; arXiv:math.GR/0510116. 450, 453, 455
[10] U. Hamenstädt, Geometry of the mapping class groups II: (Quasi)-geodesics. Preprint,
2006; arXiv:math.GR/0511349. 456
[11] W. J. Harvey, Boundary structure of the modular group. In Riemann Surfaces and Related
topics: Proceedings of the 1978 Stony Brook Conference (I. Kra and B. Maskit, eds.), Ann.
of Math. Stud. 97, Princeton, Princeton, N.J., 1981, 245–251. 447
[12] Y. Imayoshi, M. Taniguchi, An introduction to Teichmüller spaces. Springer-Verlag, Tokyo
1992. 448, 452
[13] N. V. Ivanov, Mapping class groups. Chapter 12 in Handbook of geometric topology
(R. J. Daverman and R. B. Sher, eds.), Elsevier Science, Amsterdam 2002, 523–633. 448
[14] S. Kerckhoff, Lines of minima in Teichmüller space. Duke Math. J. 65 (1992), 187–213.
451, 452
[15] E. Klarreich, The boundary at infinity of the curve complex and the relative Teichmüller
space. Unpublished manuscript, 1999. 464
[16] G. Levitt, Foliations and laminations on hyperbolic surfaces. Topology 22 (1983), 119–135.
451
[17] H. Masur, Interval exchange transformations and measured foliations. Ann. of Math. 155
(1982), 169–200. 465
[18] H. Masur, Two boundaries of Teichmüller space. Duke Math. J. 49 (1982), 183–190. 465
[19] H. Masur, Y. Minsky, Geometry of the complex of curves I: Hyperbolicity. Invent. Math.
138 (1999), 103–149. 448, 454, 455, 456, 457, 461, 462
[20] H. Masur, Y. Minsky, Quasiconvexity in the curve complex. In In the tradition of Ahlfors
and Bers, III, Contemp. Math. 355, Amer. Math. Soc., Providence, RI, 2004, 309–320.
456, 462
[21] Y. Minsky, Quasi-projections in Teichmüller space. J. Reine Angew. Math. 473 (1996),
121–136. 463
[22] Y. Minsky, Extremal length estimates and product regions in Teichmüller space. Duke
Math. J. 83 (1996), 249–286. 463
[23] L. Mosher, Train track expansions of measured foliations. Unpublished manuscript. 453
[24] J. P. Otal, Le Théorème d’hyperbolisation pour les variétés fibrées de dimension 3.
Astérisque 235 (1996). 450
[25] R. C. Penner, J. L. Harer, Combinatorics of train tracks. Ann. Math. Studies 125, Princeton
University Press, Princeton, Princeton, N.J., 1992. 450, 453, 454, 455
[26] K. Rafi, A characterization of short curves of a Teichmüller geodesic. Geom. Topol. 9
(2005), 179–202. 452
1984. 451, 463
Part C
Surfaces with singularities and discrete

Riemann surfaces
Chapter 11
Parameters for generalized Teichmüller spaces
Charalampos Charitos and Ioannis Papadoperakis
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
2 The Teichmüller space of a surface . . . . . . . . . . . . . . . . . . . . . 472
3 Hyperbolic surfaces with a conical singularity . . . . . . . . . . . . . . . . 476
3.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
3.2 The geometry of a hyperbolic pair of pants with a conical singularity . 483
3.3 The Teichmüller space of a hyperbolic pair of pants with a conical
singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
4 The Teichmüller space of an ideal 2-dimensional simplicial complex . . . . 493
4.1 Definitions and preliminaries . . . . . . . . . . . . . . . . . . . . . . 493
4.2 The completeness of ideal hyperbolic structures . . . . . . . . . . . . 497
4.3 Parametrization of the Teichmüller space . . . . . . . . . . . . . . . . 497
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
1 Introduction
In this chapter, we consider spaces of hyperbolic metrics on a space X which is
homeomorphic to either a surface or a 2-dimensional simplicial complex with its
vertices deleted.1 In the case where X is a surface, the hyperbolic metrics have
conical singularities. In the case where X is a 2-dimensional simplicial complex with
its vertices deleted, by a hyperbolic metric, we mean a length metric in which every
2-dimensional simplex with its vertices deleted is isometric to an ideal triangle. In
each case, the Teichmüller space T (X) of X is defined as the quotient space of the
set of hyperbolic metrics on X by the group of homeomorphisms of X which are
homotopic to the identity.
The first section concerns hyperbolic metrics on a surface. We consider in this sec-
tion surfaces of finite type and we recall Thurston’s parametrization of the Teichmüller
space by the lengths of finitely many simple closed geodesics.
In the second section we consider hyperbolic surfaces with conical singularities.
We parametrize the Teichmüller space of X in the case where X is a hyperbolic
1 Note that in this chapter, a 2-dimensional simplicial complex will be defined as a topological space equipped
with a certain decomposition by simplices; this is not the usual definition of a simplicial complex.
472 Charalampos Charitos and Ioannis Papadoperakis
pair of pants with a single singularity. This parametrization allows us to show that the
Teichmüller space of a hyperbolic pair of pants with a single singularity is contractible.
In the third section we consider a class of 2-dimensional simplicial complexes X
which admit a decomposition into topological triangles such that every two triangles
have at most one edge in common. We define the Teichmüller space T (X) of X and,
as in the classical case presented in Section 2, we find a finite number of essentially
simple closed geodesics whose lengths parametrize T (X).
We would like to thank G. Tsapogas for his helpful remarks and corrections.
2 The Teichmüller space of a surface

Consider a compact connected oriented differentiable surface X of genus g ≥ 0 with
k ≥ 0 points removed, and m compact boundary components (m ≥ 0). We assume
that 2g − 2 + m + k ≥ 1. Throughout this section “a hyperbolic structure on X”
will always mean a complete, hyperbolic Riemannian metric on X which makes the
boundary ∂X geodesic and the volume of X finite. The condition 2g − 2 + m + k ≥ 1
guarantees that such a structure always exists on X.
Definition 2.1. The Teichmüller space T (X) of X is the quotient of the set H(X) of
hyperbolic structures on X by Diff 0 X, where Diff 0 X is the group of diffeomorphisms
of X which are homotopic to the identity by a homotopy ht that respects the boundary
for each value of the parameter t.
If h is a hyperbolic structure on X, then, by abuse of language, we also denote by h

the equivalence class of h in T (X).
Given an element h ∈ T (X) we shall describe briefly a finite number of simple
closed geodesics whose lengths determine h. This well-known result (see for example
[9] for the compact case) will be used in an essential way in Section 4.
The main object of this section is an exposition of the following well-known the-
orem. The next sections contain an exposition of results with a similar flavour for
surfaces with singularities.
Theorem 2.2. For each element h ∈ T (X) there exists a finite number of simple
closed geodesics in X whose lengths determine h.
The idea for the proof of Theorem 2.2 is as follows:

Consider a hyperbolic structure h on X. Since X is of finite volume, it has a finite
number of boundary components and finitely many cusps (see Theorem 9.8.6 in [16]).
The next two properties are well known (see for example Paragraph 1.6 in [4]):
(P1) Let b be a simple closed curve in X. If b is not contractible to a point and if b is
not homotopic to a non-contractible closed curve contained entirely in a cusp,
then there is a unique simple closed geodesic c homotopic to b.
Chapter 11. Parameters for generalized Teichmüller spaces 473
(P2) Let b1 , b2 be two curves satisfying (P1) and, in addition, b1 ∩ b2 = ∅. If

b1 is not homotopic to b2 , then the corresponding simple geodesics c1 and c2
respectively, satisfy c1 ∩ c2 = ∅.
Based on Properties (P1) and (P2) we may decompose X into surfaces Pk ,
k = 1, . . . , n of genus 0, with geodesic boundary. Each Pk is homeomorphic to
either a disc with two holes i.e. a disc with two open discs removed or, an annulus
minus one point or, a disc minus two points. Furthermore, in the case where X is
not a torus with one cusp, we may assume that each boundary component of each Pk
is adjacent to exactly two distinct surfaces of the decomposition. This geometrical
decomposition can be easily established; we first find simple closed curves which give
a topological decomposition of X into surfaces with the above mentioned properties.
We then replace each such curve by the corresponding simple closed geodesic in its
free homotopy class. In the case where X is a torus with a cusp we just cut X along a
simple closed geodesic c and we obtain a surface homeomorphic to an annulus with
one puncture.
A surface Pk , k ∈ {1, 2, . . . , n} having
three boundary components (i.e. homeomorphic to a disc with two holes) will
be called hyperbolic pair of pants;
two boundary components (i.e. homeomorphic to an annulus minus one point)
will be called cusped annulus;
one boundary component (i.e. homeomorphic to a disc minus two points) will
be called double cusped disc.
In Figure 1, an example of a decomposition of a hyperbolic surface with g = 2,
k = 3 and m = 2 into surfaces of genus 0 is shown.
Figure 1
Let Pk , k ∈ {1, 2, . . . , n} be a surface of the decomposition of X. The restriction

of the hyperbolic structure h on Pk will be denoted by h|Pk . If Pk is a pair of pants,
we denote by ∂i Pk , i = 1, 2, 3, the boundary components of Pk and by lh (∂i Pk ) the
length of ∂i Pk . If Pk has a cusp, then, by abuse of language, we denote by ∂i Pk
either a boundary component of Pk or a cusp of Pk . In the case ∂i Pk is a cusp we set
lh (∂i Pk ) = 0.
If T (Pk ) denotes the Teichmüller space of Pk and if ρ is an element of T (Pk ) we
consider the map
L : T (Pk ) → [0, ∞)3
defined by
L(ρ) = (lρ (∂1 P ), lρ (∂2 P ), lρ (∂3 P )).
It is well known that the map L is a homeomorphism (see Exposé 3, Theorem 5 of [9]
for a proof in the case where Pk is a pair of pants). This implies that the lengths lh (ci )
of the geodesics ci determine the geometry of each Pk . In other words, the lengths
lh (ci ) parametrize the induced hyperbolic structure h|Pk ∈ T (Pk ).
Now, the surfaces Pk , k ∈ {1, 2, . . . , n} are the building-blocks of X in the sense
that every complete hyperbolic surface X of finite volume can be constructed by
considering a finite number of surfaces of genus 0, each being either a hyperbolic pair
of pants or, a cusped annulus or, a double cusped disc, and gluing them along their
boundaries.
Therefore, we consider the following situation: Let P1 , P2 be two hyperbolic
surfaces each being isometric to either a hyperbolic pair of pants or a cusped annulus
or a double cusped disc. Let h1 ∈ T (P1 ) and h2 ∈ T (P2 ) and assume that P1 , P2 have
boundary components c1 , c2 respectively such that lh1 (c1 ) = lh2 (c2 ) = l = 0. This
last condition permits us to glue P1 and P2 , by identifying c1 and c2 via an isometry,
in order to form a surface, say P .
We denote by c the geodesic which is the image of c1 and c2 in P by the natural
inclusion map. Obviously, there is a degree of freedom when the gluing is performed:
any two gluings differ by an amount of rotation t ∈ R (see Figure 2). The parameter t
is called the twist parameter along c and it represents a signed distance measured in
multiples of the geodesic length l. (The twist parameter takes values in R rather than
in S1 since we want to distinguish isometric hyperbolic structures on P which do not
represent the same element in T (P ).)
... ...
... ...
... ...
... ...
. .
..... .....
... ...
... ... t
... ...
..... .....
Figure 2
More precisely, assume that a surface P is obtained by gluing P1 with P2 and

let h be the resulting hyperbolic structure on P . Let a be a simple closed geodesic
of P which intersects c in two essential points and thus a is not isotopic to a curve
disjoint from c. Denote by a the geodesic of P which is isotopic to a curve obtained
by performing a right Dehn twist on a about c; such geodesics a and a are shown in
Figure 3.
a
a
c
.... .... .... ....

.................................... .................................... .................................... ....................................
Figure 3
We consider a collar of c in P , homeomorphic to c × [0, 1] such that c = c × {0}.

Then, for each t ∈ R, there exists a diffeomorphism φt of c × [0, 1] onto itself such
that
(1) φt is the identity in a neighborhood of c × {1};
(2) φt is an isometry in a neighborhood of c × {0};
(3) the lift of φt to the universal covering R × [0, 1] is a translation of the form
x
→ x + lt on R × {0}.
We define a new hyperbolic structure ht on P as follows: ht is equal to φt∗ (h) on
c×[0, 1], where φt∗ denotes the push-forward and ht = h outside c×[0, 1]. Obviously
h0 = h. We shall be saying that ht results from h by performing a rotation of angle t
along c.
Denote by at (respectively at ) the geodesic of P with respect to the hyperbolic
structure ht , which is freely homotopic to a (resp. to a ). If we define the function
f : R → R by f (t) = lht (at ) then we can prove that f is strictly convex and takes
its minimum at zero (see Exposé 7, Proposition 2, p. 130 in [9]). Similarly for the
function f (t) = lht (at ), and we may verify easily that f (t + 1) = f (t) for all t.
Finally, by using the strict convexity of f and f , as well as the relation f (t +1) =
f (t), we deduce easily that if f (t) = f (s) and f (t) = f (s) then t = s. Therefore
the lengths lht (at ) and lht (at ) determine the twist parameter t and hence the hyperbolic
structure ht .
As explained above, to each gluing of pair of pants P1 and P2 along boundary
components c1 and c2 , there corresponds a twist parameter t and geodesics a and a .
If X is not a torus with one cusp then each geodesic ci of X belongs to two distinct
pair of pants, say Pk and Pm , so we may assign to each ci a twist parameter ti as
well as geodesics ai and ai . From the discussion above it follows that each element
h ∈ T (X) is determined by either the parameters {lh (ci ), ti }i=1,2,...,n or the parameters
{lh (ci ), lh (ai ), lh (ai )}i=1,2,...,n . The first family of parameters are known as Fenchel–
Nielsen parameters.
If X is a cusped torus we consider a cusped annulus by cutting X along a simple
closed geodesic c. Let a be a simple closed geodesic which intersects c essentially in
two points and let a be the geodesic which is isotopic to the curve a twisted along c.
In a similar way we may prove that each element h ∈ T (X) is determined by the
parameters {lh (c), lh (a), lh (a )}.
Remark. If X is a closed surface of genus g ≥ 2, U. Hamenstädt in [12] parametrized

T (X) by the lengths of 6g − 5 simple closed geodesics. These curves are defined by
decomposing X into 2g − 2 pair of pants with the aid of 3g − 3 geodesics {ci } and it
is proven that the 3g − 3 twist parameters along ci defined above, are determined by
the lengths of 3g − 2 simple closed geodesics which intersect each ci once.
3 Hyperbolic surfaces with a conical singularity

This section is based on results published by the authors in [8]. We consider compact
hyperbolic surfaces X with a finite number of conical singularities {s1 , . . . , sn }. Every
point p ∈ Int(X) − {s1 , . . . , sn } has a neighborhood isometric to an open subset of
the hyperbolic plane H2 and the angle at each conical singularity (see Definition 3.1
below) is greater than 2π. Such surfaces X will be called hyperbolic surfaces with
conical singularities (h.s.c.s.).
In the first paragraph we present examples of h.s.c.s. and describe general properties
of them, resulting from the fact that h.s.c.s. are spaces of curvature less than or equal
to −1 (see Paragraph 2.1 below for a precise definition). In this way, we shall see that
the geometry of h.s.c.s. is quite different from the geometry of hyperbolic surfaces
without singularities.
In the second paragraph we consider a pair of pants with a single singularity. We
introduce two systems of real parameters which determine the geometry of a pair of
pants with a single singularity. These parameters are of two kinds: those depending
on the position of the conical singularity and those that do not.
In the third paragraph we state the main result of this section which asserts that the
Teichmüller space T (X, s) of a hyperbolic pair of pants X with a single singularity s,
is contractible.
3.1 Preliminaries
We summarize here some standard definitions and statements that we shall need in the
sequel (for more details see, for example, [15]). A geodesic segment (resp. geodesic
ray, geodesic line) in a metric space (Y, | · |) is an isometric map f : I = [a, b] → Y
(resp. I = [0, ∞), I = R). A local geodesic in Y is a path f : I → Y such that
for each t ∈ I , there exists an interval I (t) ⊂ I which is a neighborhood of t in
I , such that the restriction of f to I (t) is a geodesic. A closed geodesic is a local
geodesic g : (−∞, +∞) → Y which is a periodic map. The metric space Y is said
to be geodesic if for each two points x, y there is a geodesic segment f : [a, b] → Y
with f (a) = x, f (b) = y. We write [x, y] for a geodesic segment joining x with y.
A triangle T = (xyz) in a geodesic metric space Y is a triple x, y, z of points

in Y which are called the vertices of T , together with a triple [x, y], [x, z], [y, z] of
geodesic segments joining them. These geodesic segments are called the sides of T .
Let H2 be the hyperbolic plane. If T is a triangle in Y , then a comparison triangle
for T in H2 is a triangle T ∗ whose sides have the same length as the sides of T , together
with a map f from the disjoint union of the sides of T onto the sides of T ∗ , such that
f is consistent on the vertices on T and such that the restriction of f to each side of
T is an isometry. The map f is called a comparison map for T . We note that the
comparison triangle T ∗ together with its associated comparison map f are unique up
to composition by an isometry of H2 .
A triangle T in Y is said to satisfy the CAT(−1)-inequality if its associated com-
parison map is distance non-decreasing, that is, for every x, y in T ,
|x − y|Y ≤ |f (x) − f (y)|H2 .
A geodesic metric space Y is said to be a CAT(−1)-space if every triangle in Y
satisfies the CAT(−1)-inequality. If a metric space Y satisfies locally the CAT(−1)-in-
equality, i.e. if every point in Y has an open neighborhood which is a CAT(−1)-space
with respect to the induced metric, then we say that Y is of curvature less than or
equal to −1.
Assuming that the geodesic metric space Y is locally compact and complete it
follows that every bounded subset of Y is relatively compact i.e. Y is proper. If we
further assume that Y satisfies the CAT(−1)-inequality, then Y is a hyperbolic space
in the sense of Gromov (see for example [5] Chapter 1, Theorem 5.1). Now, since Y is
geodesic proper and hyperbolic in the sense of Gromov, the (visual) boundary bd(Y )
of Y can be defined as the space of equivalence classes of asymptotic geodesic rays
starting at a fixed point [5]. On the other hand, it follows from the CAT(−1)-inequality,
that for every two points x, y in Y ∪ bd(Y ) there is a unique (up to parametrization)
geodesic joining them.
Finally, if Y is of curvature less than or equal to −1 then its universal covering Ỹ
satisfies the CAT(−1)-inequality. This follows from a theorem of Cartan–Hadamard
(see for example [3], Theorem 4.1). If furthermore Y is compact then for every
closed curve a of Y which is not nullhomotopic, there is a unique geodesic in the free
homotopy class of a.
We give now the precise definition of hyperbolic surfaces with conical singularities.
Definition 3.1. Let X be a compact orientable surface with or without boundary

equipped with a geodesic metric distance ρ( ·, ·). We say that X has a hyperbolic
structure with finitely many conical singularities = {s1 , . . . , sn }, if there exists a
triangulation D of X by geodesic triangles such that:
(1) Each triangle of D, with the induced metric, is isometric to a triangle in the
hyperbolic space H2 .
(2) Each si , i = 1, 2, . . . , n is a vertex of the triangulation D.
(3) If v is a vertex of D and if we denote by θD (v) the sum of all angles formed by
the edges of D which have v as vertex, then
θD (v) = 2π if v ∈ Int(X) − and θD (v) = π if v ∈ ∂X − ,
θD (v) > 2π if si ∈ Int(X) ∩ and θD (v) > π if si ∈ ∂X ∩
where ∂X is the boundary of X.
The metric ρ( ·, ·) is called a hyperbolic structure on X and X is called a hyperbolic
surface with conical singularities (h.s.c.s.). The angle θD (si ) will be called the opening
of the conical singularity si with respect to D.
We note here that the definition is independent of the triangulation D, i.e., if D

is another triangulation of X which satisfies conditions (1) and (2) above, and if v is
a vertex of D , then
θD (v) = 2π if v ∈ Int(X) − and θD (v) = π if v ∈ ∂X − ,
θD (si ) = θD (si ) for each i = 1, 2, . . . , n.
Since the opening of a conical singularity is independent of the triangulation, we
may talk about the opening θ(si ) of a conical singularity without reference to the
triangulation considered. In what follows we shall denote by H (X) the space of all
hyperbolic structures on X equipped with the compact-open topology. In [13], [14]
and [18], h.s.c.s. are defined in an equivalent way by using conformal metrics.
Let now X be a compact h.s.c.s. Then X is a space of curvature less than or equal
to −1. This follows from a theorem of Gromov (see for example [11], 4.2.D or [1],
Chapter 10, Theorem 15) which asserts, in dimension 2, that a locally finite polyhedron
K obtained by gluing together geodesic triangles of the hyperbolic space H2 , is of
curvature less than or equal to −1, if and only if for each x ∈ K the direction space
at x satisfies the CAT(1)-inequality, equivalently θ (v) ≥ 2π for each vertex v ∈ K.
If we denote by X̃ the universal covering of X, then X̃ is simply connected,
geodesic, complete and locally compact. Therefore X̃ satisfies the CAT(−1)-inequality
globally, as remarked above.
If X is a closed h.s.c.s. then X̃ is homeomorphic to R2 since X cannot be homeo-
morphic to S2 . On the other hand X = X̃/G where G is a discrete group of isometries
of X̃ which acts freely and cocompactly on X̃. It is well known that G is a hyperbolic
group in the sense of Gromov and bd(X̃) = bd(G) (see for example Theorem 4.1 of
[5]). But since G is a surface group, bd(G) = S1 . From this we easily deduce that
the compactification X̃ ∪ bd(X̃) of X̃ is homeomorphic to the closed unit 2-disc D.
Definition 3.2. A polygon P = (x1 x2 . . . xn ) in X̃ is a subset of X̃ homeomorphic to

the closed unit 2-disc D such that the subarc of ∂P between two consecutive points
xi , xi+1 is a geodesic segment of X̃. This geodesic segment [xi , xi+1 ] will be called
an edge of P . The points xi will be called vertices of P .
We say that a polygon P of X̃ is realizable in H2 if there is a polygon in H2 which
is isometric to P .
We next present examples of h.s.c.s.
Example 1. We consider a regular hyperbolic octagon P in H2 such that the sum of

the angles of P is greater than 2π. By gluing together the sides of P , as indicated in
Figure 4, we obtain a hyperbolic surface of genus 2 with one conical singularity.
a4
a3 a1
a4 a2
a3 a1
a2
Figure 4
Similarly we can construct a hyperbolic structure with one conical singularity on

any closed surface of genus g > 2.
Example 2. Let P be a surface homeomorphic to a disc with two holes, equipped

with a hyperbolic structure with one conical singularity p in its interior and such that
the boundary ∂P = c1 ∪ c2 ∪ c3 consists of closed geodesics (without singularities),
i.e. P is a hyperbolic pair of pants with one conical singularity in its interior. Let di ,
i = 1, 2, 3 be geodesic arcs realizing the distance between p and ci respectively. For
each i = 1, 2, 3 we consider the geodesic loop ai with base point p (i.e. ai is a local
geodesic arc whose initial and terminal point is p) which is freely homotopic to ci
and we may assume, as one can see from the construction below, that ai ∩ aj = {p}
for i = j (see Figure 5 (a)). We denote by l(ci ), l(di ), l(ai ) the lengths of ci , di , ai
respectively. We cut P along the arcs di , i = 1, 2, 3. Then we obtain a connected
surface Q which can be decomposed into one hyperbolic triangle whose sides have
lengths l(ai ) and three quadrilaterals, each with two right angles. The sides of these
quadrilaterals are shown in Figure 5 (b). Therefore, by following the inverse procedure,
we consider a hyperbolic triangle T = (xyz) and three hyperbolic quadrilaterals Q1 ,
Q2 , Q3 , each one with two right angles, which can be glued together to form T , as
shown in Figure 5 (b). Denote by (x, T ) the angle of T at x and by (x, Qi ) the
angle of Qi at x and we define (x) = (x, Q1 ) + (x, Q2 ) + (x, T ). Similarly
we define (y) and (z). We can assume that (x) + (y) + (z) > 2π . Now,
by identifying the equal sides of each quadrilateral Qi , we obtain a hyperbolic pair of
pants with one conical singularity.
c1 c2 d2
c1 d1 x
c2
d1 d2 a1 a2
a1 a2 d1 d2
p a3 y a3 z
... d3 .. d3 d3
...... .
.......................................
c3
c3
(a) (b)
Figure 5
Finally by gluing by isometries along their boundaries 2g − 2 hyperbolic pair of

pants, each one with a single singularity, we obtain a closed h.s.c.s. of genus g with
2g − 2 conical singularities.
Let now s1 , . . . , sn be conical singularities of X and for i = 1, . . . , n let θi be the

opening of si . Then we have the following proposition which connects the number of
singularities with their opening and the genus of X.
n
Proposition 3.3. If X is a closed h.s.c.s. of genus g, then i=1 (θi −2π) < (4g −4)π.
Proof. We consider a triangulation D of X by hyperbolic triangles such that the

points si are vertices of D. Let T , E, V be the number of triangles, edges and vertices
of D respectively. Then the Euler characteristic is V − E + T = 2 − 2g. On the
other hand, 3T = 2E, since each triangle has three edges and each edge is adjacent
to exactly two triangles. The sum of the angles of all triangles is smaller than T π.
Therefore, if we take the sum of the angles around each vertex, we have
n

θi + (V − n)2π < T π.
i=1
Thus,
n

(θi − 2π ) + 2nπ + 2(V − n)π < (V + 4g − 4),
i=1
and so
n

(θi − 2π) < (4g − 4)π. 2
i=1
We deduce the following
Corollary 3.4. Let X be a h.s.c.s. of genus 0 withgeodesic boundary and let m be the
number of boundary components, m ≥ 3. Then ni=1 (θi − 2π) < (2m − 4)π .
Proof. By taking the double of X we obtain a closed h.s.c.s. of genus g = m − 1.

Now the corollary follows by Proposition 3.3.
We exhibit a few properties of h.s.c.s. in order to further understand their geometry.
We begin with an example which shows that if γ is a simple closed non-contractible
curve on a h.s.c.s. X then the unique closed geodesic which exists in the free homotopy
class of γ is not necessarily simple, even if X has a single singularity.
Example 3. We glue eight hyperbolic triangles Ti = (sxi xi+1 ), i = 1, . . . , 7 and

T8 = (sx8 x1 ) as it is shown in Figure 6 and denote by P = (x1 . . . x8 ) the resulting
space. The triangles Ti are chosen so that
• each Ti is isosceles;
• Ti is isometric to Tj for all i, j ;
• the sum of the angles over all vertices x1 , . . . , x8 of P is equal to 2π;
• there is a conical singularity of angle 5π at the center s of P .
x1 a1 m1 x2
a4 a2
m 4 m2
x8 d1
x3
a3 d3
m 3 s m 1
d1 a1
x7 x4
m4 d3 m 2
a4 a2
x6 m3 a3 x5
Figure 6
So P is a regular hyperbolic octagon. If we glue the sides a1 , a2 , a3 , a4 of P ,

as indicated in Figure 4 we obtain a closed surface X of genus 2 with one conical
singularity at s. Let m1 , m 1 , m3 , m 3 be the midpoints of sides [x1 , x2 ], [x3 , x4 ],
[x5 , x6 ], [x7 , x8 ] respectively. We consider the union γ = [m1 , s] ∪ [s, m 3 ] ∪ [m3 , s] ∪
[s, m 1 ] of the oriented geodesic segments (see Figure 6). Obviously, γ is a closed
geodesic in X which has a double point. On the other hand, there is a simple closed
curve on X which is freely homotopic to γ .
Assume that X is a compact h.s.c.s. which has only one conical singularity. In this
case we may find simple closed geodesics on X which cut X into a surface of genus 0.
More precisely we have:
Proposition 3.5. Let X be a closed h.s.c.s. of genus g with a single conical singu-
larity s1 . Then there are g simple closed geodesics of X which do not contain s1 and
which cut X into a surface N of genus 0.
Proof. Let X = X̃/G, where X̃ is the universal covering of X and G is a discrete group
of isometries of X̃ which act freely on X̃. If p : X̃ → X is the covering projection
we consider a point s ∈ p −1 (s1 ). We consider also the Dirichlet polygon D(s) for
G, with center s (see for example [16], p. 243). For h ∈ G, if Hh (s) = {x ∈ X̃ :
dist(x, s) < dist(x, h(s))}, then D(s) = {Hh (s) : h = 1 in G}. By standard
methods we may prove that D(s) = (x1 x2 . . . x4g ) is a 4g-polygon in X̃ with sides
ai , ai , i = 1, . . . , 2g. The surface X is obtained by gluing in an appropriate way the
sides of D(s) (for example the case g = 2 is shown in Figure 6).
Now let di (resp. di ), i = 1, 3, . . . , 2g −1, be the geodesic segments in D(s) which
join the midpoints mi , mi+1 of ai and ai+1 (resp. the midpoints points m i , m i+1 of ai
and ai+1 ). Obviously [s, m ] and [s, m ] are perpendicular to a and a respectively.
i i i i
The quadrilateral (sm1 x2 m2 ) has two right angles and, therefore, the angle at s of
[s, m1 ] and [s, m2 ] is less than π . Thus, the geodesic segment d1 does not contain s.
Similarly, we may prove that di , di do not contain s. Now, we can easily see that di , di
fit together so that the angles formed at mi , mi+1 , m i , m i+1 are equal to π . Therefore,
if we set ci = di ∪ di , i = 1, 3, . . . , 2g − 1, we deduce that p(ci ) are simple closed
geodesics in X which do not contain s1 . Furthermore if we cut X along p(ci ) we
obtain a h.s.c.s. N of genus 0 with geodesic boundary and one conical singularity in
its interior.
Now the question that arises naturally is whether the surface N of Proposition 3.5
can be decomposed into pairs of pants by cutting along simple closed geodesics.
We have the following proposition:
Proposition 3.6. Assume that N has m boundary components, m ≥ 5. If θ (s1 ) ≥

2(m − 2)π then N cannot be decomposed into pairs of pants by cutting it along simple
closed geodesics.
Proof. From Corollary 3.4 we know that θ(s1 ) < 2(m − 1)π .
If the proposition were not true then we would be able to find a simple closed
geodesic in N which does not contain s1 . But in this case s1 is contained in the interior
of a subsurface N with m − 1 boundary components. Therefore we would have that
θ (s1 ) < 2(m − 2)π , which contradicts our hypothesis.
We consider now the special case where X is homeomorphic to a closed surface

of genus two. Then we have
Proposition 3.7. A closed surface of genus two can be decomposed into two hyperbolic
pair of pants by cutting it along three simple closed geodesics.
Proof. From Proposition 3.5 we may find two simple closed geodesics which cut X
into a surface N of genus 0. Now let P = D(s) = (x1 . . . x8 ) be the Dirichlet polygon
in X̃ considered in Proposition 3.5.
We consider the piecewise geodesic of γ = [x3 , s] ∪ [s, x7 ] in P . We show that γ
projects to a closed geodesic in N and thus γ separates N into two hyperbolic pairs
of pants. In fact, the two angles formed by the geodesic segments [x3 , s], [s, x7 ] at s
are equal, and greater than π. On the other hand, by construction it follows that
P (x3 ; s, x2 )+P (x4 ; x5 , x3 )+P (x1 ; x2 , x8 )+P (x6 ; x7 , x5 )+P (x7 ; x8 , s) = π
and
P (x3 ; s, x4 )+P (x2 ; x1 , x3 )+P (x5 ; x4 , x6 )+P (x8 ; x7 , x1 )+P (x7 ; x6 , s) = π.
From these equalities we deduce immediately that the simple closed curve γ is a
geodesic.
3.2 The geometry of a hyperbolic pair of pants with a conical singularity

In this section we assume that X is a h.s.c.s. which is homeomorphic to a disc with
two holes. We assume further that X has only one conical singularity s. We call such
a surface X a hyperbolic pair of pants with one singularity. The aim of this section
is to find a collection of real parameters with the smallest possible cardinality that
determine the geometry of X. These parameters are of two kinds: those depending
on the position of the conical singularity and those which do not.
The disc with two holes X is the simplest compact planar surface which admits an
hyperbolic structure with conical singularities and geodesic boundary. If X has more
than one conical singularity, it is an interesting but harder problem to find parameters
which determine the geometry of X.
Let ∂X = c1 ∪ c2 ∪ c3 and we denote by ρ( ·, ·) the distance on X. Let di be
the geodesic arc that realizes the distance ρ(s, ci ) between the singularity s and the
curve ci . We denote by li the length of ci and by ri the length of di . Obviously the
parameters ri depend on the position of s. We have the following proposition:
Proposition 3.8. Let li be the lengths of the boundary components ci and ri be the
lengths of the geodesic arcs di between s and ci for i = 1, 2, 3.
1) If s ∈ Int(X) then the real parameters li , ri , i = 1, 2, 3 determine at most four
non-isometric hyperbolic structures on X.
2) If s ∈ ∂X then the real parameters li , ri , i = 1, 2, 3 determine at most three
non-isometric hyperbolic structures on X.
Proof. First we assume that s ∈ Int(X) (see Figure 7 (a)). We cut X along d1 ∪ d2 ∪ d3
to obtain a surface denoted by P . Let p : X̃ → X be the covering projection. We may
c3 c2
d3 d2 d3 d2
s
... d1 .. ...
...... .
...................................... ...... ...
c1 ......................................
s
(a) (b)
Figure 7
realize P as a polygon, say P = (qps1 xys2 zws3 ), isometrically embedded in X̃, where
s1 , s2 , s3 ∈ p−1 (s). P with the induced geometry satisfies the CAT(−1)-inequality.
We consider the following cases:
Case A0 : Every geodesic segment [si , sj ], i, j ∈ {1, 2, 3}, intersects ∂P only at its
endpoints. Then the triangle T = (s1 s2 s3 ) of P is non-degenerate and therefore it
has, by CAT(−1) inequality, all its angles smaller than π . The quadrilateral (s1 xys2 )
is realizable in H2 and therefore is uniquely determined up to isometry of H2 , by the
right angles at x and y and by the lengths r2 = length([s1 , x]) = length([s2 , y]) and
l2 = length([x, y]). Therefore the length of [s1 , s2 ] is determined by the parameters
r2 and l2 .
Similarly the quadrilaterals (s1 pqs3 ) and (s2 zws3 ) are realizable in H2 and the
lengths of [s2 , s3 ], [s1 , s3 ] are known. From this analysis we can easily construct X
in the case A0 as follows: The triangle T = (s1 s2 s3 ) and the quadrilaterals Q1 =
(s1 xys2 ), Q2 = (s2 zws3 ), Q3 = (s3 qps1 ) are realized as polygons of H2 and they
are unique up to isometry. By gluing these polygons along their common sides we get
the surface X.
If we denote by P (si ; sj , sk ) the angle in P formed by the geodesic segments
[si , sj ] and [si , sk ] at si , then we define the set A0 of all hyperbolic structures on X
such that P (si ; sj , sk ) < π for every i, j, k ∈ {1, 2, 3}. Obviously to each such
hyperbolic structure we may correspond a unique configuration of H2 , say of type
A0 , which consists of a triangle T and three quadrilaterals Qi of H2 , glued as in
Figure 8 (a). Conversely, to each configuration of type A0 , we may correspond a
unique hyperbolic structure on X.
Finally we remark that in the case A0 , the whole polygon P is realizable in H2 .
Case A0,1 : We assume that P (s1 ; s2 , s3 ) = π. In this case the triangle T = (s1 s2 s3 )
is degenerate and coincides with the geodesic segment [s2 , s3 ] (see Figure 8 (b)). We
y z s3
s2 q w
x p
w s1
x z
s1 s3
y s2
p q
(a) (b)
Figure 8
define by A0,1 the set of all hyperbolic structures on X such that P (s1 ; s2 , s3 ) = π.
Therefore, to each hyperbolic structure in A0,1 corresponds a unique configuration
of H2 , say of type A0,1 , which consists of three quadrilateral Qi of H2 , glued as in
Figure 8 (b). Conversely to each configuration of type A0,1 , we may assign a unique
hyperbolic structure on X.
Similarly to the Case A0,1 we define the cases A0,2 and A0,3 when
P (s2 ; s1 , s3 ) = π and P (s3 ; s1 , s2 ) = π
respectively. Analogously, we define the subsets A0,2 and A0,3 of hyperbolic structures
on X.
Case A1 : The geodesic in P which joins the points s2 , s3 passes through s1 and
P (s1 ; s2 , s3 ) > π . This implies that [s2 , s3 ] = [s2 , s1 ] ∪ [s1 , s3 ]. Now, if we
consider the pentagon R = (s1 s2 zws3 ) it then follows that R (s1 ; s2 , s3 ) > π and R
is realizable in H2 .
The quadrilaterals (s3 pqs1 ) and (s1 xys2 ) are uniquely determined up to isometry
of H2 , similarly to the case A0 . So the lengths of [s1 , s2 ] and [s1 , s3 ] are determined.
Therefore the pentagon R can be constructed in H2 in a unique way, since we know
all its edges and that the angles at z, w are right, and that the angle at s1 is greater
than π . Thus, under the assumption P (s1 ; s2 , s3 ) > π , P is uniquely determined,
and the surface X as well.
We define by A1 the set of all hyperbolic structures on X such that P (s1 ; s2 , s3 ) >
π . Obviously to each hyperbolic structure in A1 we may assign a unique configuration,
say of type A1 , which consists of a pentagon R and two quadrilaterals Q2 , Q3 of H2 ,
glued as in Figure 9. Conversely, to each configuration of type A1 , we may assign
a unique hyperbolic structure on X. The triangle T = (s1 s2 s3 ) appearing in the
configuration of Figure 9 is called imaginary, because it does not actually exists in X.
We remark that P is not necessarily realizable as a polygon of H2 ; this is the reason

that in Figure 9 in H2 , the quadrilaterals Q2 and Q3 intersect each other.
s3
q
x
y
p w
s1
s2
Figure 9
Similarly to the case A1 , we consider the cases A2 and A3 where we have

P (s2 ; s1 , s3 ) > π or P (s3 ; s1 , s2 ) > π,
respectively. In these cases we also define the corresponding sets A2 and A3 and we
set
A0 = A0 ∪ A0,1 ∪ A0,2 ∪ A0,3 and Ak = Ak ∪ A0,k , k = 1, 2, 3.
We assume in the following that s ∈ ∂X (see Figure 7 (b)).
We also assume, without loss of generality, that s ∈ c1 . Then r1 = 0. If we cut
and open X along d1 ∪ d2 then we obtain a 7-gon P = (s1 xys2 s3 qp). We first claim
that P (s1 ; s2 , s3 ) < π . Indeed, we denote by γi the geodesic loop in X based at s
which is freely homotopic to the boundary ci for i = 1, 2, 3. If s ∈ c1 then c1 = γ1 .
Moreover if P (s1 ; s2 , s3 ) ≥ π then it follows that γ1 is the juxtaposition of γ2 and γ3 ,
which is impossible. Now we distinguish the following cases:
Case ∂1 B0 : If P (s2 ; s1 , s3 ) < π and P (s3 ; s1 , s2 ) < π then we define the corre-
sponding subset ∂1 B0 of hyperbolic structures on X. At every hyperbolic structure
in ∂1 B0 we may assign a unique configuration in H2 , say of type ∂1 B0 , (see Figure
10 (a)) and conversely. Obviously the parameters ri and li determine, up to isometry,
each element of ∂1 B0 .
Case ∂1 B0,2 : P (s2 ; s1 , s3 ) = π (see Figure 10 (b)) and we define the corresponding
subset ∂1 B0,2 of hyperbolic structures on X.
Case ∂1 B0,3 : P (s3 ; s1 , s2 ) = π and we define also the subset ∂1 B0,3 of hyperbolic
structures on X.
We set ∂1 B0 = ∂1 B0 ∪ ∂1 B0,2 ∪ ∂1 B0,3 .
s3
q
s3 s2
q
s2
y
y p
p s1 x
x s1
(a) (b)
Figure 10
Case ∂1 B2 : P (s2 ; s1 , s3 ) > π (see Figure 11) and we define the corresponding
subset ∂1 B2 of hyperbolic structures on X. We set ∂1 B2 = ∂1 B2 ∪ ∂1 B0,2 .
s3
q
y
s2
x
p
s1
Figure 11
Case ∂1 B3 : P (s3 ; s1 , s2 ) > π and we also define the subset ∂1 B3 of hyperbolic

structures on X. We set ∂1 B3 = ∂1 B3 ∪ ∂1 B0,3 .
If s ∈ c2 we define sequentially the sets ∂2 B0 , ∂2 B0,1 , ∂2 B0,3 , ∂2 B1 , ∂2 B3 and if
s ∈ c3 we define the sets ∂3 B0 , ∂3 B0,1 , ∂3 B0,2 , ∂3 B1 , ∂3 B2 . In the sequel we similarly
define the sets ∂i Bj , i = 2, 3, j = 0, 1, 2, 3 and i = j .
Finally we set
A0 = A0 ∪ ∂1 B0 ∪ ∂2 B0 ∪ ∂3 B0 , Ak = Ak ∪ ∂i Bk ∪ ∂j Bk , k = 1, 2, 3,
i, j ∈ {1, 2, 3} − {k}.
From the analysis above we deduce that each hyperbolic structure is uniquely
determined, up to isometry, from the parameters li and ri .
On the other hand, given any six parameters li , ri , i = 1, 2, 3 which determine
a hyperbolic structure of type A0 , we cannot deduce that the same li , ri determine a
hyperbolic structure in Ak , k = 1, 2, 3. For example in the case A1 , the lengths of

[si , sj ], which are determined by li , ri , must satisfy certain compatibility conditions so
that s1 ∈ Int(Q2 ). Therefore, if s ∈ Int(X), li , ri determine at most four non-isometric
hyperbolic structures, while if s ∈ ∂X, the parameters li , ri determine at most three
non-isometric hyperbolic structures.
Remark. Let X be a hyperbolic surface with one singularity s, which is homeomorphic

to a disc with n holes, n ≥ 3. We denote again by ci the boundary components of X and
by di the geodesic segments from s to ci . If we set li = length(ci ) and ri = length(di )
then the parameters li , ri do not characterize X in the sense of Proposition 3.8. This
follows from the fact that if we cut and open X along d1 ∪ · · · ∪ dn then instead of a
triangle (s1 s2 s3 ) (see for example the case A0 in Proposition 3.8) we get a hyperbolic
n-gon which, of course, is not determined uniquely by the length of its edges.
In what follows, we introduce a set of real parameters which are defined inde-
pendently of the position of the singularity s. We shall prove that these parameters
determine, up to isometry, the surface X.
Let ∂X = c1 ∪ c2 ∪ c3 and let di be a geodesic segment which realizes the distance
between ci and ci+1(mod 3) , i = 1, 2, 3. It is not hard to prove, by using the CAT(−1)
inequality, that di is unique for each i. Assume d1 ∩ d2 ∩ d3 = {s}. Then if we cut and
open X along d = d1 ∪ d2 ∪ d3 we obtain two connected components P , Q. Each P ,
Q is a polygon isometrically embedded in X̃. We denote by ai the subarc P ∩ ci of ci
and by bi the subarc Q ∩ ci . Set li = length(ci ), ri = length(di ) and ki = length(ai ).
We remark that ri , i = 1, 2, 3 are the distances between the sides of the polygon P .
Definition 3.9. The six non-negative parameters ri , ki , i = 1, 2, 3 will be called the

distance parameters of X.
We have the following theorem:
Theorem 3.10. The parameters li , ri , ki , i = 1, 2, 3 determine a unique, up to

isometry, hyperbolic pair of pants X with one singularity.
Proof. To prove the theorem we distinguish the following cases:

Case I: k1 · k2 · k3 = 0. This implies that we can consider P and Q as hexagons
isometrically embedded in X̃. From the study of the parameters li , ri , ki we can
determine whether s ∈ P and s ∈ / Q or s ∈ Q and s ∈ / P . Indeed, if s ∈ P and
s ∈/ Q then Q is a right-angled hexagon of H2 . Therefore, if we set ki = li − ki ,
the parameters ki , ri satisfy certain compatibility conditions (see for example [2],
Section 7.19, Theorem 7.19.2). Conversely, if ki , ri satisfy these conditions we deduce
that s ∈ P and s ∈ / Q. Similarly we may study the case s ∈ Q and s ∈ / P . In
conclusion the study of parameters li , ri , ki allows one to tell whether s ∈ P and
s∈/ Q or s ∈ Q and s ∈ / P or s ∈ P ∩ Q.
We assume now, without loss the generality, that s ∈ P and s ∈ / Q (similarly we

treat the case s ∈ Q and s ∈
/ P ). Hence P is an hexagon in X̃ which has all its angles
equal to π2 and Q is a right-angled hexagon in H2 . Therefore we deduce that P , Q,
and hence, the pair of pants X, is determined uniquely by the parameters li , ri , ki .
We assume finally that s ∈ P ∩ Q. In this case s is a common vertex of P and
Q. Then the angles (s, P ) and (s, Q) are not necessarily right but are both smaller
than 2π and so P and Q are realizable as 6-gons in H2 . Therefore, P , Q and hence
X are uniquely determined by the parameters li , ri , ki and the theorem is proven in
Case I.
Case II: k1 · k2 = 0, k3 = 0. If the singularity s lies in the interior of X, then the
geodesic segments d2 and d3 have a common part in X (see Figure 12 (a)). Now the
polygon P in X̃ is a 5-gon (x1 x2 sx3 x4 ) which is realizable in H2 . In P we know that the
angles at xi for i = 1, 2, 3, 4 are right and we also know the distances |x1 −x2 |P = k1 ,
|x3 − x4 |P = k2 , |x1 − x4 |P = r1 . Therefore P is unique, up to isometry, and the
distances |s − x2 |P and |s − x3 |P can be expressed as a function of k1 , k2 , r1 (see
Figure (12b)). This analysis determines whether s lies in the component c3 of ∂X or,
not. Indeed, when s ∈ c3 then |s − x2 |P = r2 and |s − x3 |P = r3 . Therefore, if r2 , r3
satisfy certain compatibility conditions with k1 , k2 , r1 then there exists a 5-gon in H2
with four right angles and whose sides have lengths k1 , k2 , r1 , r2 , r3 . The existence
of a such 5-gon implies that s ∈ c3 .
c1 k1 x2 y2
x1 s1 k1
y1 y3
r1 P s k Q2 r1
s Q1
3
x4 y4
y6 s2
.
c2 .................................. ..... ..
.............................. c3 k2 x3 k2
y5
(a) (b) (c)
Figure 12
Therefore we distinguish two subcases:

First we assume that the relations between the parameters k1 , k2 , r1 , r2 , r3 impose
that s ∈ Int(X) (see Figure 12). We have seen that P is realizable in H2 and it is unique.
Q = (y1 s1 y2 y3 y4 y5 s2 y6 ) is also realizable in H2 . Indeed, the geodesic segment
[s1 , s2 ] is contained in Q and it divides Q into two sub-polygons Q1 = (y1 s1 s2 y6 ),
Q2 = (s1 y2 y3 y4 y5 s2 ), see Figure 12(c).
For Q1 we know: |y1 − y6 |Q1 = k3 , (y1 , Q1 ) = (y6 , Q1 ) = π2 , |y1 − s1 |Q1 =
r3 − |x2 − s|P , |y6 − s2 |Q1 = r2 − |x3 − s|P .
Now the lengths |x2 −s|P , |x3 −s|P are uniquely determined from the construction
of P . Therefore Q1 is uniquely determined up to isometry. Analogously, we show
that Q2 is uniquely determined. Finally we conclude that X is uniquely determined

by the parameters li , ri , ki .
Second we assume that the relations between the parameters k1 , k2 , r1 , r2 , r3
impose that s ∈ c3 ⊂ ∂X (see Figure 13). Then P is a 5-gon and Q a 6-gon realizable
in H2 and we proceed as above.
y3
k1 x2 r1 k2
y4 y2
x1
r1 P k1 Q r2
s
y1
x4
r3 s2
....... .
.............................
..... ..
.............................. k2 x3 s1 l3
s
(a) (b) (c)
Figure 13
Finally we treat the cases k2 · k3 = 0, k1 = 0 and k1 · k3 = 0, k2 = 0 similarly.

The case ki = kj = 0 and km = 0 with i, j, m ∈ {1, 2, 3}, cannot occur.
Case III: k1 = k2 = k3 = 0. In this case we deduce that d1 ∩ d2 ∩ d3 = {s} and,
therefore, we fall into Proposition 3.8. We distinguish two subcases:
Subcase IIIa : s ∈ Int(X). Then the angles formed by the geodesic segments [s, x1 ],
[s, x2 ], [s, x3 ] at s are all greater than or equal to π . Thus the parameters li , ri
determine a hyperbolic structure of type A0 (see Proposition 3.8), since in all other
types of hyperbolic structures, we can verify easily, that the above mentioned angles
around s are smaller than π .
Subcase IIIb : s ∈ ∂X. Let xi = ci ∩ (d1 ∪ d2 ∪ d3 ). Then from the relation that relates
the parameters ri we can deduce in which component of ∂X the singularity s belongs.
For example if r2 = r1 + r3 then s ∈ c1 . So we assume without loss of generality,
that s ∈ c1 . Then the angle formed by the geodesics d1 , d3 at s is greater than or equal
to π. Therefore the parameters li , ri determine a unique hyperbolic structure of type
∂1 B0 since in all other types of hyperbolic structures, the angle at s formed by the
segments d1 and d3 is smaller than π .
3.3 The Teichmüller space of a hyperbolic pair of pants

with a conical singularity
In this section we also assume that X is a hyperbolic pair of pants with one conical
singularity, say s. We shall define the Teichmüller space T (X, s) of X and a parameter
space B for T (X, s).
We fix an orientation on X and let Homeo+ (X, ∂) be the set of homeomorphisms
of X which preserve the orientation and each boundary component of X. It is well
known that each element of Homeo+ (X, ∂) is isotopic to the identity (see Exposé 2
in [9]). The space Homeo+ (X, ∂) acts on H (X) as follows: If f ∈ Homeo+ (X, ∂)
and ρ ∈ H(X) then (f, ρ) → f ∗ ρ where f ∗ ρ(x, y) = ρ(f (x), f (y)).
Definition 3.11. We define the Teichmüller space T (X, s) of X with a conical singu-
larity s as the quotient H (X)/ Homeo+ (X, ∂).
Obviously T (X, s) consists of all hyperbolic structures which belong to one of the
sets Ak , k = 0, 1, 2, 3 defined in Proposition 3.8.
To build a parameter space for T (X, s) which will allow us to understand the
topology of T (X, s) we proceed as follows:
We consider a convex subset C0 of R6 consisting of all (a, b, c, r1 , r2 , r3 ) ∈ R6 :
a, b, c > 0 (1)
a ≤b+c (2)
b ≤a+c (3)
c ≤a+b (4)
r1 , r2 , r3 ≥ 0 (5)
r1 + r2 > 0 (6)
r2 + r3 > 0 (7)
r1 + r3 > 0. (8)
Obviously C0 is a non-bounded convex subset of R6 . Let ∂a C0 (resp. ∂b C0 , ∂c C0 )
be the subset of C0 obtained by replacing inequality (2) (resp. (3), (4)) above with the
equality a = b + c. The sets ∂a C0 , ∂b C0 , ∂c C0 are pairwise disjoint subsets of the
boundary ∂C0 of C0 . On the other hand, we define the subset ∂1 C0 (resp. ∂2 C0 , ∂3 C0 )
of C0 which is defined by the equation r1 = 0 (resp. r2 = 0, r3 = 0). The sets ∂1 C0 ,
∂2 C0 , ∂3 C0 are also pairwise disjoint subsets of ∂C0 . However each ∂i C0 intersects
∂a C0 and let ∂i,a C0 = ∂a C0 ∩∂i C0 . Obviously each ∂i,a C0 is homeomorphic to R4 and
is a convex connected component of the boundary ∂(∂a C0 ) of ∂a C0 . More precisely
we have ∂(∂a C0 ) = ∂1,a C0 ∪ ∂2,a C0 ∪ ∂3,a C0 .
On the other hand, for each i, the set ∂i,a C0 is a convex connected component
of the boundary ∂(∂i C0 ) of ∂i C0 . So if we define, in a similar fashion, the subsets
∂i,b C0 = ∂b C0 ∩ ∂i C0 and ∂i,c C0 = ∂c C0 ∩ ∂i C0 , then we have that ∂(∂i C0 ) =
∂i,a C0 ∪ ∂i,b C0 ∪ ∂i,c C0 .
Now we consider three copies of C0 , say C1 , C2 , C3 . On the disjoint union
C0 ∪ C1 ∪ C2 ∪ C3 we identify C0 with C1 along ∂a C, C0 with C2 along ∂b C and
C0 with C3 along ∂c C. Let C be the resulting space. From the convexity of the
sets ∂a C, ∂b C and ∂c C it follows that C is a non-compact manifold embedded in R6 ,
with three boundary components. The interior of C is homeomorphic to R6 and each
boundary component ∂i C of C, i = 1, 2, 3, is homeomorphic to R5 . We remark that
the restriction of ∂i C on every Ci (considered as subset of C) is defined by the relation
ri = 0.
Now we construct an embedding : T (X, s) → C. To do this, we consider

a hyperbolic structure h ∈ T (X, s). As remarked above, the hyperbolic structure
h belongs to a set Ak , k = 0, 1, 2, 3. Therefore, to each h ∈ T (X, s) we may
correspond to h a unique configuration of H2 , as it is described in Figures 8–11. Let
a = |s2 − s3 |, b = |s1 − s3 |, c = |s1 − s2 |, where (s1 s2 s3 ) is the triangle appearing
in each configuration which determines h and (r1 , r2 , r3 ) are the distances defined in
Proposition 3.8; notice here that (s1 s2 s3 ) may be an imaginary triangle.
Now, we define a 1-1 mapping
0 : A0 → C0
such that to each h ∈ A0 we assign the unique parameters (a, b, c, r1 , r2 , r3 ) of C0

which determine h. Obviously we have the relations (∗) below:
0 (A0,1 ) ⊂ ∂a C0 , 0 (A0,2 ) ⊂ ∂b C0 , 0 (A0,3 ) ⊂ ∂c C0
(∗)
0 (∂i B0 ) ⊂ ∂i C0 , i = 1, 2, 3
We define similarly 1 − 1 mappings
k : Ak → Ck , k = 1, 2, 3
which satisfy the relations (∗∗) below:
1 (A0,1 ) ⊂ ∂a C0 , 2 (A0,2 ) ⊂ ∂b C0 , 3 (A0,3 ) ⊂ ∂c C0
(∗∗)
k (∂i Bk ) ⊂ ∂i Ck , k ∈ {1, 2, 3}, i ∈ {1, 2, 3} − {k}.
From the relations (∗) and (∗∗) above we have that 0 | is equal to k | ,
A0 ∩A0 A0 ∩A
so 0 and k fit properly and give a mapping : T (X, s) → C which is 1-1. We set
(T (X, s)) = B.
Proposition 3.12. The mapping is continuous and open, thus : T (X, s) → B is

a homeomorphism.
Proof. We first we show that B is an open subset of C. Indeed, let

(a, b, c, r1 , r2 , r3 ) = (h)
for some h ∈ T (X, s). As explained in Proposition 3.8, a hyperbolic pair of pants
X equipped with the hyperbolic structure h, can be constructed by gluing either a
triangle and three quadrilaterals of H2 or a 5-gon and two quadrilaterals of H2 . The
lengths of the sides of these polygons take values in the set {a, b, c, r1 , r2 , r3 }. More
precisely if (a, b, c, r1 , r2 , r3 ) ∈ Ck then (a, b, c, r1 , r2 , r3 ) determine a configuration
of type Ak and thus we construct (X, h) i.e. the hyperbolic structure h on X. It
is straighforward to see that if we slightly change the lengths of the sides of the
polygons that we glued in order to construct (X, h), then with the new polygons we
may construct a new hyperbolic structure h on X. Therefore there is an ε > 0 such
that if (a , b , c , r1 , r2 , r3 ) ∈ C and it is ε-near to (a, b, c, r1 , r2 , r3 ) then there is

h ∈ T (X, s) such that (h ) = (a , b , c , r1 , r2 , r3 ).
Now, to finish the proof of the proposition it is sufficient to show that the mapping
−1 : B → T (X, s) is continuous. But it is obvious that there is an ε > 0 such
that if (a , b , c , r1 , r2 , r3 ) ∈ B is ε-near to (a, b, c, r1 , r2 , r3 ) ∈ B then for every
x, y ∈ X the values h(x, y) and h (x, y) are arbitrary close, where h, h are the images
of (a, b, c, r1 , r2 , r3 ) and (a , b , c , r1 , r2 , r3 ) by −1 .
The following corollary follows immediately from the above proposition.
Corollary 3.13. T (X, s) is homeomorphic to a non-compact submanifold of R6 of

dimension 6, with three boundary components.
We state now the main theorem of this chapter:
Theorem 3.14. T (X, s) as well as each boundary component of ∂(T (X, s)) are
contractible spaces.
For the proof of this theorem, which is quite long and technical, we construct step
by step a homotopy H : T (X, s) × [0, 1] → T (X, s) such that H|T (X,s)×{0} = id and
H|T (X,s)×{1} = constant map. For details, see [8].
4 The Teichmüller space of an ideal 2-dimensional

simplicial complex
In this section we give the definition of 2-dimensional ideal simplicial complexes X
which is slightly different from the corresponding definition of [6], we consider the
Teichmüller space T (X) of X and we introduce the notion of essentially simple closed
curves in X.
The main result is:
For each element h ∈ T (X) there exists a finite number of essentially simple closed
geodesics in X whose lengths determine h.
4.1 Definitions and preliminaries

A 2-dimensional simplicial complex is a topological space X equipped with two (finite
or infinite) sets C and F that satisfy the following properties:
(1) Each element T ∈ C is a topological triangle, that is, a topological space
homeomorphic to a 2-dimensional closed disc with three distinguished distinct points
on its boundary ∂T . The distinguished points are called the vertices of T . The closures
in ∂T of the connected components of the complement of the vertices are called the
edges of T .
(2) Each element f ∈ F is a homeomorphism f : A → B, where A and B are

distinct edges of two distinct triangles T and T belonging to the collection C. For
each edge A of a triangle T ∈ C, there exists an f ∈ F and an edge B of some
triangle T ∈ C, with f sending A to B. The elements of F are called gluing maps.
(3) As a topological space, the simplicial complex X is the quotient of the disjoint
union of the elements of C by the equivalence relation which identifies any two points
x and y belonging respectively to edges A and B of triangles T and T in C, which are
related by a map f : A → B in F satisfying f (x) = y. Let π : C → X denote the
quotient map. We make the assumption that π is injective on each edge of a triangle
in C. The image by π of a triangle T ∈ C is called a face of X, the image by π of an
edge of T is called an edge of X and the image by π of a vertex of T is called a vertex
of X.
(4) Every two faces T1 , T2 of X with T1 ∩ T2 = ∅ intersect in either a single edge
of X or a single vertex of X.
A face T of X with vertices x, y, z is denoted by (xyz).
(5) The space X is arcwise connected.
(6) The space X is locally finite, which means that each edge and each vertex of X
is contained in finitely many faces.
We shall say that a 2-dimensional simplicial complex is finite if the number of its
faces is finite.
Let X be a 2-dimensional complex and let be the set of edges of X. There is a
subset ⊂ , whose elements, called singular edges, are those edges of X whose
inverse image by π consists of three or more distinct edges of triangles of C. An edge
of X which is in − is called a non-singular edge. The set of vertices of X will
be denoted by S.
We shall be interested in metrics on X = X−S which are obtained in the following
manner: for each topological triangle T ∈ C, we delete the vertices of T and we equip
this triangle with its vertices deleted with a metric which makes it isometric to an ideal
hyperbolic triangle. In this way each edge of X with its vertices deleted is equipped
with an induced metric and becomes isometric to the infinite line. We now take each
map f : A → B in F to be an isometry, with respect to these metrics. In this way, each
face of X, with its vertices deleted, becomes isometric to an ideal hyperbolic triangle.
The length lX (γ ) of a path γ : I → X is now defined as the sum of the lengths of the
components of the intersection of γ with the faces of X. Let h : X × X → R be the
function defined in the following way: for (x, y) ∈ X × X , h(x, y) is equal to the
infimum of the set of lengths of paths in X joining these points.
From Proposition 1.2 of [6] it follows that the function h : X × X → R is a
metric on X and the metric space (X , h) is a length space. Any such metric h on
X will be called an ideal hyperbolic metric on X. Furthermore, if h is complete, h
is called an ideal hyperbolic structure on X. We also say that the space X equipped
with an ideal hyperbolic structure h is an ideal 2-dimensional simplicial complex. The
faces of X are ideal hyperbolic triangles and the edges of X are the edges of these
ideal triangles. In Proposition 4.3 below we shall show that there always exist ideal
hyperbolic structures on X.
In what follows, we shall always assume that the metric h on X is complete.
Although the metric h is defined on X , we shall be speaking about a metric on X.
By the term geodesic (resp. local geodesic) in X we mean a geodesic (resp. a local
geodesic) in X with respect to the metric h, see Paragraph 2.1 for the definitions.
Every ideal hyperbolic structure h on X is a geodesic metric since X with h is a
complete and locally compact length space (see [10], Chapter 1, Theorem 1.10). As
in Paragraph 2.1, we write [x, y] for a geodesic segment joining the points x, y ∈ X .
The space X is a local CAT(−1)-space. This follows from the fact that each point
in X has a neighborhood which is isometric to the union of a finite number of half-
discs in H2 whose diameters are of equal length, the half discs being glued along their
diameters, see Corollary 1.5 of [6].
The universal covering X̃ of X inherits the structure of a 2-dimensional simplicial
complex. Obviously, if we consider an ideal hyperbolic structure h on X then h can
be lifted, via the covering projection p : X̃ → X, to an ideal hyperbolic structure on
X̃, i.e. to a complete length metric h̃ on X̃ − p −1 (S).
Definition 4.1. Let X be a 2-dimensional simplicial complex, C the set of its triangles
and F the associated set of gluing maps. Two ideal hyperbolic structures h and h
on X are said to be equivalent if there exists a homeomorphism F : X → X which
induces the identity map on the edges of X and which satisfies F ∗ (h) = h , where
by F ∗ (h) we denote the pull-back of the metric h via F . We denote by T (X) the
set of equivalence classes of ideal hyperbolic structures on X and we call this set the
Teichmüller space of X.
If h is an ideal hyperbolic structure on X, we shall also denote by h its equivalence
class in T (X).
Let T be an ideal hyperbolic triangle. Then T has a distinguished point which is

the barycentre of T . Furthermore, each edge of T is equipped with a distinguished
point, namely, the foot of the perpendicular drawn from the barycentre of T to that
edge. We shall call this point the centre of the edge.
There are several ways of describing the topology of T (X), and we shall use here
the shift parameters. Let X be a 2-dimensional simplicial complex equipped with an
ideal hyperbolic structure h and let S be the set of vertices of X. In order to describe
the shift parameters, we start by choosing once and for all an orientation on each edge
of X. If T , T are two faces of X with T ∩ T = ∅ then, by hypothesis, they have
only one edge in common, say E. We shall use the same letters T , T to indicate the
corresponding ideal hyperbolic triangles in X . We define the quantity xh (T , T ) (or
xh (T , T , E), if we want to specify the edge E) as the algebraic distance on E from
the distinguished point associated to T to the distinguished point associated to T ,
the sign being specified by the chosen orientation on E. We call xh (T , T ) the shift
parameter on the ordered pair (T , T ).
Let B be the set of ordered pairs (T , T ) where T , T are triangles of X. The shift
parameters define a map : T (X) → RB , by the formula
(h)(T , T ) = xh (T , T ).
The map is clearly injective, since each element of T X is determined by the
associated set of gluing maps, which are isometries between edges of ideal hyperbolic
triangles.
Equip B with the discrete topology, RB with the product topology and T (X) with
the topology induced from the embedding : T (X) → RB .
Note finally that, for reasons explained in [7], the map : T (X) → RB is not
onto.
Remark 4.2. In the definition of a simplicial complex above, we do not identify

two edges belonging to a single triangle in C and every two distinct triangles have at
most one edge in common. This results from condition (4). Example 4 below shows
that without this condition the lengths of essentially simple closed geodesics cannot
determine a hyperbolic structure in T (X).
Example 4. We consider a space Z consisting of two ideal triangles T , T whose

edges are matched as shown in Figure 14. Let E be the edge of Z equipped with
an orientation, see Figure 14. Obviously the ideal triangles T , T have more than
one common edge. It not difficult to show that T (Z) may be parametrized by the
E
T T
Figure 14
shift parameter xh (T , T , E). Indeed, the space Z has two cusps and it is obtained
by performing three gluings along the sides of T and T . So the dimension of T (Z)
is 1 (see Theorem 3.3 in [6]). Now, in order to determine xh (T , T , E) we need the
lengths of two closed geodesics; one which determines |xh (T , T , E)| and a second
one which determines the sign of xh (T , T , E). But, it is easy to see that there exists
only one simple closed curve in X which is not nullhomotopic. Therefore, there is
only one essentially simple closed geodesic in X (see Definition 4.4 below).
4.2 The completeness of ideal hyperbolic structures

Let v be a vertex of X. The link L(v) of v is a simplicial graph embedded in X, which is
defined by taking one vertex on each edge of X ending on v and then joining two such
vertices by an edge contained in a face of X if and only if these vertices correspond
to edges which are on the boundary of that face. We note that if X is locally finite,
the graph L(v) is finite and it is easy to see that the link L(v) is connected. We note
also that if V is a small enough closed ball in X centered at v (with respect to some
simplicial metric on X) then V has a natural structure of a geometric cone of the
form v · L(v), that is, the topological quotient space [0, 1] × L(v)/R, where R is the
equivalence relation (0, x) ∼ (0, x ) for all x, x ∈ L(v).
Let us fix a hyperbolic metric h on X. If v is a vertex of X, then v will be a
vertex (at infinity) of finitely many ideal triangles of X and we call v a cusp of X .
Obviously, a “neighborhood” of the cusp v in X is homeomorphic to v · L(v) − {v}.
We can associate a natural foliation on such a neighborhood of v, which is well defined
up to restriction to a smaller neighborhood. The definition is as follows. Consider
an ideal triangle in X = X − S, having v as one of its ideal vertices. Consider a
foliation of a horoball neighborhood of v in the ideal triangle, whose leaves are pieces
of horocycles which are centered at the ideal vertex v. Considering now all ideal
triangles of X admitting v, as a vertex, we can glue together these foliations, and
define a foliation of a neighborhood of the cusp v in X . (Note that the foliation is
singular with singular locus contained in the singular edges of X). Such a foliation is
called a horocyclic foliation around v. From Proposition 3.4.18 in [17] it follows that:
the metric space (X , h) is complete if and only if, for each cusp v of X and for each
horocyclic foliation of a neighborhood of v, the restriction of the foliation to a smaller
neighborhood of v is a product foliation on a space of the form L(v) × [0, ∞), the
leaves of which are the fibers L(v) × {t}. Proposition 3.4.18 in [17] concerns the case
where X is a hyperbolic surface, but the case considered here follows in the same
way.
Now, if for every two adjacent ideal triangles T , T of X , xh (T , T ) = 0, then h
is obviously a complete metric i.e. an ideal hyperbolic structure. Therefore, we im-
mediately deduce the following:
Proposition 4.3. Every 2-dimensional simplicial complex X possesses an ideal hy-

perbolic structure.
4.3 Parametrization of the Teichmüller space

In this paragraph we shall find real parameters for the Teichmüller space T (X) which
are lengths of closed geodesics in X. These geodesics are not necessarily simple but
each one contains in its free homotopy class a simple closed curve without back and
forth.
For the rest of this paragraph we fix an ideal hyperbolic structure h on X.

We say that a closed curve c : [0, 1] → X has a back and forth if there are
t1 , t2 ∈ [0, 1], t1 < t2 , and an ideal triangle T of X such that c(t1 ), c(t2 ) belong to
the same edge of T and c([t1 , t2 ]) ⊂ T .
Now, we give the following definition
Definition 4.4. Let γ be a closed geodesic in X. We say that γ is essentially simple

if there exists a homotopy ht : [0,1] → X sending γ to a simple closed curve c such
that the closed curve ht does not have back and forth for each value of the parameter t.
In Figure 15 (a) the geodesic γ1 , which has no back and forth, can be homotoped
to a simple curve c1 without back and forth for all times t, while in Figure 15 (b) the
geodesic γ2 , which has no back and forth too, can be homotoped to a simple curve c2
but the new curve c2 has back and forth. So γ1 is an essentially simple geodesic but
not γ2 .
γ1 c1
γ1 c1
(a)
γ2
c2
γ2 c2
(b)
Figure 15
It is easy to see that if γ is an essentially simple closed geodesic, then the images
ht ([0, 1]) remain in the same ideal triangles of X as γ , for all t. Therefore, if there
exists a surface S embedded in X with γ ⊂ S, for example this always happens if γ

does not intersect a singular edge of X , then γ is a simple closed geodesic. Indeed,
if γ is not simple then we can find a simple closed curve c which is homotopic to γ
in S. But this is impossible.
The main theorem of this section is the following:
Theorem 4.5. The element h of T (X) can be determined by the lengths of a finite
number of essentially simple closed geodesics in X.
Proof. Let T = (abc) and T = (acd) be two faces of X which share a common
edge. The corresponding ideal triangles of X are also denoted by T and T . The edge
of X joining the cusps a and c is denoted by E(a, c).
In order to prove the theorem it suffices to find a finite number of essentially simple
closed geodesics of X whose lengths determine the shift parameter xh (T , T ).
We may assume, without loss of generality, that each edge of X is adjacent to
at most three ideal triangles. Indeed, we denote by kE the number of ideal triangles
which are adjacent to an edge E and we call kE the index of E. Assume E is an
edge of index kE ≥ 4. Then there is an ideal simplicial complex Y (not necessarily
connected) and two edges E1 , E2 of Y such that
• kE1 < kE , kE2 < kE and kE1 + kE2 = kE ,
• there exists an isometry E1 → E2 such that X is obtained from Y by identifying
E1 with E2 via this isometry,
• T , T belong to the same connected component of Y and share the common
edge E(a, c).
By repeated application of this procedure, we obtain an ideal simplicial complex
Z which contains T and T glued along E(a, c) and the index of edges of Z is at most
three.
Figure 16 displays how the edge E of X gives rise to two edges E1 , E2 in Y with
kE1 = 2 and kE2 = 3.
E1
E
E2
Figure 16
Since the metric space (X , h) is complete, there is a neighborhood V of the cusp a

(resp. of the cusps b, c, d) in X which is homeomorphic to L(a)×[0, ∞) and foliated
by a horocyclic foliation whose leaves are the fibers L(a) × {t}.
We pick an arbitrary leaf, say Ha , of this horocyclic foliation around a. Ha forms
a closed graph whose edges are the pieces of horocycles lying in ideal triangles and
the vertices are the intersection points of Ha with the edges of X abutting at a. Let
ab = Ha ∩ E(a, b), ac = Ha ∩ E(a, c), ad = Ha ∩ E(a, d) and denote by H [ab , ac ]
(resp. H [ac , ad ]) the subarc of Ha contained in (abc) (resp. in (acd)) with endpoints
ab , ac (resp. ac , ad ), see Figure 17. With respect to the order of inclusion, we consider
a
ab ad
ac
b d
Figure 17
a minimal subgraph Ha0 of Ha which contains H [ab , ac ] and H [ac , ad ]. We have the
following claim:
Claim 1. The graph Ha0 is homeomorphic to either a circle or the configuration shown
in Figure 18 (a); in the latter case the graph will be called of type G1 .
G1 x G2
y
ac ad
ab
(a) (b)
Figure 18
Proof of Claim 1. There exists an arc γ1 : [0, 1] → Ha such that

(i) γ1 is 1-1 for each t ∈ (0, 1);
(ii) γ1 (0) = ab and γ1 (1) ∈ H [ab , ac ] ∪ H [ac , ad ];
(iii) γ1 ([0, 1]) ∩ (H [ab , ac ] ∪ H [ac , ad ]) = {γ1 (0), γ1 (1)}.
Obviously, γ1 (1) ∈ {ab , ac , ad }. Therefore, either γ1 (1) = {ad } or, γ1 (1) = {ac }
or, γ1 (1) = {ab }. We denote these cases by (I1 ), (I2 ), (I3 ) respectively.
Similarly, there exists an arc γ2 : [0, 1] → Ha such that
(i) γ2 is 1-1 for each t ∈ (0, 1);
(ii) γ2 (0) = ad and γ1 (1) ∈ H [ab , ac ] ∪ H [ac , ad ];
(iii) γ2 ([0, 1]) ∩ (H [ab , ac ] ∪ H [ac , ad ]) = {γ2 (0), γ2 (1)}.
Obviously, γ2 (1) ∈ {ab , ac , ad }. Therefore, either γ2 (1) = {ab } or, γ2 (1) = {ac }
or, γ2 (1) = {ad }. We denote these cases by (II1 ), (II2 ), (II3 ) respectively.
By combining these cases for the arcs γ1 and γ2 , and given that each vertex of Ha0
is of index 3, we conclude the proof of the claim. More precisely: In the case (I1 ) or
(II1 ) we may choose Ha0 to be homeomorphic to S1 . In the cases (I3 ) and (II3 ) or, (I3 )
and (II2 ) or, (I2 ) and (II3 ) we may choose Ha0 to be homeomorphic to G1 .
Now, we consider a closed curve r0 : [0, 1] → Ha0 such that
(1) r0 is locally 1-1 and onto;
(2) there are points t1 < t2 in [0, 1] such that r0 (t1 ) = ab , r0 (t2 ) = ad and
r0 ([t1 , t2 ]) ∩ Ha0 = H [ab , ac ] ∪ H [ac , ad ];
(3) the number of times that r0 passes through each vertex of Ha0 is minimum
among all curves r : [0, 1] → Ha0 satisfying properties (1) and (2) above.
For example, in Figure 18 (a) we consider Ha0 to be of type G1 . If we choose a
curve r0 : [0, 1] → Ha0 such that r0 (0) = ab , r0 (t1 ) = ac , r0 (t2 ) = ad , r0 (t3 ) = x,
r0 (t4 ) = ad , r0 (t5 ) = ac , r0 (t6 ) = y, r0 (1) = ab , for 0 < t1 < t2 < t3 < t4 < t5 <
t6 < 1, then we can see that r0 satisfies the requirements (1), (2), (3) above.
Now, for the curve r0 , we consider a partition 0 = t1 < t2 < · · · < tn = 1 of [0, 1]
such that r0 ([ti , ti+1 ]) ⊂ Ti for all i = 1, . . . , n − 1„ where Ti is an ideal triangle of
X and r0 (ti ) ∈ Ei , r0 (ti+1 ) ∈ Ei+1 , where Ei and Ei+1 are edges of Ti . We shall
say that the family {Ti } covers the curve r0 and we remark that the triangles Ti are
not necessarily distinct. By taking the triangles Ti with an orientation on their sides
induced from the orientation of the edges of X , we glue them successively and we
construct a surface Y0 . More precisely, for each i = 1, . . . , n − 1, we glue Ti with
Ti+1 along their common side Ei+1 with shift parameter equal to xh (Ti , Ti+1 ) and Tn
with T1 along En+1 with shift parameter equal to xh (Tn , T1 ). In this way we build a
new surface Y0 with a cusp (Figure 19). All edges Ei of Y0 are ending at this cusp,
which is denoted again by a. Notice that (abc) and (acd) appear as ideal triangles
of Y0 . The surface Y0 is called the development of r0 by the ideal triangles {Ti }. The
edge E(a, b) of Y0 is adjacent to (abc) and to another ideal triangle of Y0 , say (abe).
Obviously, (abe) is one of the ideal triangles Ti which cover r0 , so (abe) is also a face
of X .
For the cusp b of X we work similarly considering the triangles (abc), (abe)
instead of the triangles (abc) and (acd). This means that, we consider a horocyclic
foliation around b, we choose a leaf Hb of it, as well as a minimal subgraph Hb0 of Hb
containing H [ba , bc ] and H [ba , be ], where H [ba , bc ] = Hb ∩ (abc) and H [ba , be ] =
e e
a a
b d b
d
c c
Y0 z Y1
w
e e
a a
b c d b d
c
z z
u u
Y2 Y3
Figure 19
Hb ∩(abe). Similarly to r0 , we consider a curve r1 : [0, 1] → Hb0 satisfying properties

(1), (2), (3) above and which is covered by a finite family of ideal triangles, say {Ti1 },
of X . We “develop” further the surface Y0 by gluing to Y0 all the triangles Ti1 . So,
Y1 contains Y0 and Y1 also contains the development of r1 by the triangles {Ti1 }. In
this way we build a surface Y1 with two cusps a and b (Figure 19). Notice that every
two adjacent ideal triangles of Y1 have the same shift parameter as ideal triangles of
Y1 and as ideal triangles of X . The edge E(b, c) of Y1 is adjacent to (abc) and to
another ideal triangle of Y1 , say (bcz).
We consider now the cusp c of X and a leaf Hc around c as before. We set
H [cz , cb ] = Hc ∩ (zcb), H [cb , ca ] = Hc ∩ (cba), H [cd , ca ] = Hc ∩ (acd) and
we choose a minimal subgraph of Hc0 of Hc containing H [cz , cb ]), H [cb , ca ] and
H [cd , ca ]. We have the following claim:
Claim 2. The graph Hc0 is homeomorphic to either a circle or to a graph of type G1 or

to a graph of type G2 (see Figure 18).
The proof of Claim 2 is similar to the proof that of Claim 1. By combining all
possible ways of going from point cd to point cz in Hc we form a subgraph which, if
it is not a circle, it is either of type G1 or of type G2 .
Now, we consider a closed curve r2 : [0, 1] → Hc0 which satisfies properties (1)
and (3) above, while property (2) is replaced by the following property (2 ).
(2 ) There are points t1 < t2 < t3 < t4 in [0, 1] such that r2 (t1 ) = cz , r2 (t2 ) = cb ,
r2 (t3 ) = ca , r2 (t4 ) = cd and r2 ((ti , ti+1 ))∩{cz , cb , cd , ca } = ∅ for each i = 1, 2, 3, 4.
As before, there exists a family {Ti2 } of ideal triangles of X which cover r2 . So,
we build a surface Y2 which contains Y1 and also contains the development of r2 by
the ideal triangles {Ti2 }. Therefore Y2 is a new surface which contains the cusps a, b, c
(Figure 19). The edges E(c, d) and E(a, d) of Y2 which belong to the triangle (acd)
are adjacent to two ideal triangles of Y2 , say (cdu) and (adw) respectively.
Finally, by considering the cusp d of X we repeat the same procedure with d in
place of c. That is, we consider the three pieces of horocycle H [du , dc ] = Hd ∩ (duc),
H [dc , da ] = Hd ∩ (dca), H [da , dw ] = Hd ∩ (daw), where Hd is a leaf of the
horocyclic foliation around d, we next consider the minimal subgraph Hd0 of Hd
which contains H [du , dc ], H [dc , da ], H [da , dw ] and so on. In this way we construct
a surface Y3 which contains the cusps a, b, c, d (Figure 19).
From the construction of Y3 it follows that there exists a natural projection π : Y3 →
X which is a local isometry in the sense that for every y ∈ Y3 there exists a neighbor-
hood U of y in Y3 which is mapped isometrically by π onto a subset of X . It is imme-
diate to see that for any curve ρ : [0, 1] → π(Y3 ) there exists a lifting r : [0, 1] → Y3
of ρ, i.e. π r = ρ.
Now we shall prove that the shift parameters of all pairs of ideal triangles which
are glued in Y3 , and hence xh (T , T ), are determined by the lengths of simple closed
geodesics of Y3 which are projected to essentially simple geodesics in X .
We need the following
Claim 3. In a neighborhood of the cusp a (resp. of b, c, d) of Y3 we may find a simple
closed curve ra (resp. rb , rc , rd ) (see Figure 20) such that the following holds:
ra
ad
a
ba b ac da
rb b d rd
bc dc
ca
cb cd
c
rc
Figure 20
• ra , rb , rc , rd are projected onto a simple closed curve of X .

• If we set ab = ra ∩ E(a, b), ac = ra ∩ E(a, c), ad = ra ∩ E(a, d) and define,
similarly, the points ba , bc ∈ rb , ca , cb , cd ∈ rc and da , dc ∈ rd , then the
projection π : Y3 → X is 1-1 on the graph G = ra ∪ rb ∪ rc ∪ rd ∪ E[ab , ba ] ∪

E[bc , cb ] ∪ E[ac , ca ] ∪ E[ad , da ] ∪ E[dc , cd ], where E[ab , ba ] ⊂ E(a, b) is the
arc with endpoints ab , ba . Similarly the arcs E[bc , cb ], E[ac , ca ], E[ad , da ],
E[dc , cd ] in the corresponding edges of Y3 are defined.
Proof of Claim 3. If the graph Hx0 for x ∈ {a, b, c, d} is homeomorphic to a circle, then
there are neighborhoods of the cusp x in Y3 and X respectively, which are isometric
via π . Therefore, we may choose an arbitrary simple closed curve rx in a neighborhood
of x in Y3 .
If the graph Hx0 is of type G1 for x ∈ {a, b, c, d} or, if Hx0 is of type G2 for
x ∈ {c, d}, then we choose a simple closed curve ρx in a neighborhood of the cusp
x in X , and we denote by rx a lifting of ρx in Y3 . Obviously rx is a simple closed
curve. Furthermore, we must choose ρx , and hence rx , properly so that the projection
π : Y3 → X restricted on the graph G is 1-1.
Without loss of generality, we assume that Hx0 is of type G1 for x = a and Hx0 is of
type G2 for x = c. We consider points ab , ac , ad placed on Ha0 and points ca , cb , cd ,
placed on Hc0 . Let ρa and ρc be simple closed curves contained in a neighborhood of
cusps a and c of X respectively.
For y ∈ {b, c, d}, if the edge E(a, y) intersects ρa in more than one point then we
define the points ay = E(a, y) ∩ ρa so that E[ay , y) ∩ ρa = ay , where E[ay , y) is the
subset of E(a, y) from ay to y which is homeomorphic to the interval [0, +∞). We
set ay = π −1 (ay ) ∩ ra . Similarly we define cy = E(c, y) ∩ ρc and cy = π −1 (cy ) ∩ rc
where y ∈ {a, b, d}. Now, with this choice of curves and points, the graph G is defined
and we may easily verify that π : G → X is 1-1. In Figures 21 and 22 we indicate
how to choose the curves ρa and ρc as well as the points ay and cy in two typical
cases.
ac ad
ab Ha0
ac ad
ρa
ab
Figure 21
ca
cd
cb cz
Hc0
cd
ca cz
cb ρc
Figure 22
Now, all the simple closed geodesics in Y3 which are used to determine the shift
parameter χh (T , T ), can be homotoped to simple closed curves which do not have
back and forth and which lie in an arbitrarily small neighborhood U of graph G.
A number of such geodesics are drawn in bold in Figure 23. On the other hand,
a d
b c
Figure 23
these geodesics project to essentially simple closed geodesics in X because they are
homotopic in Y3 to simple closed curves lying in U. This proves the theorem.
References
[1] E. Ghys et P. de La Harpe (eds.), Sur les groupes hyperboliques d’après M. Gromov. Progr.
Math. 83, Birkhäuser, Boston, MA, 1990. 478
[2] A. F. Beardon, The geometry of discrete groups. Grad. Texts in Math. 91, Springer-Verlag,
New York 1983. 488
[3] M. Bridson, A. Haefliger, Metric spaces of non-positive curvature. Grundlehren Math. Wiss.
319, Springer-Verlag, Berlin 1999. 477
[4] P. Buser, Geometry and spectra of compact Riemann surfaces. Progr. Math. 106, Birkhäuser,
Boston, MA, 1992. 472
[5] M. Coornaert, T. Delzant, A. Papadopoulos, Géométrie et théorie des groupes. Lecture
Notes in Math. 1441, Springer-Verlag, Berlin 1990. 477, 478
[6] C. Charitos, A. Papadopoulos, The geometry of ideal 2-dimensional simplicial complexes.
Glasgow Math. J. 43 (2001), 39–66. 493, 494, 495, 496
[7] C. Charitos, A. Papadopoulos, Hyperbolic structures and measured foliations on 2-
dimensional complexes. Monatsh. Math. 139 (2003), 1–17. 496
[8] C. Charitos, I. Papadoperakis, On the geometry of hyperbolic surfaces with a conical sin-
gularity. Ann. Global Anal. Geom. 23 (2003), 323–357. 476, 493
Astérisque 66–67 (1979). 472, 474, 475, 491
[10] M. Gromov, Structures métriques pour les variétés riemanniennes. Rédigé par J. La-
fontaine et P. Pansu, Fernand Nathan, Paris 1981. 495
[11] M. Gromov, Hyperbolic groups. In Essays in group theory (S. Gersten, ed.), Math. Sci.
[12] U. Hamenstädt, Parametrizations of Teichmüller space and its Thurston boundary. In Geo-
metric analysis and nonlinear partial differential equations, Springer-Verlag, Berlin 2003,
81–88. 476
[13] R. McOwen, Point singularities and conformal metrics on Riemann Surfaces. Proc. Amer.
Math. Soc. 103 (1) (1988), 222–224. 478
[14] R. McOwen, Conformal metrics with singularities and finite negative total curvature on
Riemann surfaces. In Geometry and nonlinear partial differential equations, Contemp.
Math. 127, Amer. Math. Soc., Providence, RI, 1992, 101–107. 478
[15] F. Paulin, Constructions of hyperbolic groups via hyperbolization of polyhedra. In Group
theory from a geometrical viewpoint ((E. Ghys, A. Haefliger, A. Verjovsky, eds.), ICTP
(Trieste, Italy, 1990), World Scientific Publishing Co., Inc., River Edge, NJ, 1991. 476
[16] J. G. Ratcliffe, Foundations of hyperbolic manifolds, Grad. Texts Math., Springer-Verlag,
1994. 472, 482
[17] W. P. Thurston, Three-dimensional geometry and topology. Princeton Math. Ser. 35, Prince-
ton University Press, Princeton, NJ, 1997. 497
[18] M. Troyanov, Prescribing curvature on compact surfaces with conical singularities. Trans.
Amer. Math. Soc. 324 (1991), 793–821. 478
Chapter 12
On the moduli space of singular euclidean surfaces
Marc Troyanov
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508
2 Piecewise flat surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
2.1 Euclidean triangulation on a surface . . . . . . . . . . . . . . . . . . 510
2.2 The universal branched cover of a piecewise flat surface . . . . . . . . 511
2.3 The developing map of a piecewise flat surface . . . . . . . . . . . . 512
2.4 The holonomy of a piecewise flat surface . . . . . . . . . . . . . . . 514
2.5 The developing map near a singularity . . . . . . . . . . . . . . . . . 515
2.6 Geometric equivalence of euclidean triangulations . . . . . . . . . . . 516
2.7 Flat metrics with conical singularities . . . . . . . . . . . . . . . . . 516
2.8 Relation with Riemann surfaces . . . . . . . . . . . . . . . . . . . . 517
3 Punctured surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
3.1 Punctured surfaces and their fundamental groups . . . . . . . . . . . 518
3.2 Uniformization of a punctured Riemann surface . . . . . . . . . . . . 519
3.3 Some groups of diffeomorphisms of a punctured surface . . . . . . . 521
3.4 Outer automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . 521
3.5 Lifting the group Diff 0 ( ) on U . . . . . . . . . . . . . . . . . . . . 522
4 The representation variety of a finitely generated group in SE(2) . . . . . . 524
4.1 On the cohomology of groups . . . . . . . . . . . . . . . . . . . . . 524
4.2 Abelian representations . . . . . . . . . . . . . . . . . . . . . . . . . 526
4.3 Representations in SE(2) . . . . . . . . . . . . . . . . . . . . . . . . 527
4.4 Conjugation by similarities . . . . . . . . . . . . . . . . . . . . . . . 529
5 Deformation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
5.1 The moduli and Teichmüller spaces . . . . . . . . . . . . . . . . . . 530
5.2 The deformation space of piecewise flat metrics . . . . . . . . . . . . 532
5.3 Revisiting the developing map and the holonomy . . . . . . . . . . . 532
6 The main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
6.1 The case of the sphere . . . . . . . . . . . . . . . . . . . . . . . . . 537
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
508 Marc Troyanov
1 Introduction
The Teichmüller space of a punctured surface is the space of hyperbolic metrics with
cusps up to isotopy on that surface, however, it can also be seen as the space of flat
metrics with conical singularities of prescribed angles at the punctures up to isotopy
and rescaling. The aim of the present chapter is to use this fact and show how the theory
of piecewise flat surfaces and their deformations leads to new geometric structures in
Teichmüller theory.
In the first section, which is rather elementary, we describe the geometry of piece-
wise flat surfaces. The second section describes the topology of punctured surfaces
and their diffeomorphism groups. In the third section we discuss the representation
space of a finitely generated group π into the group SE(2) of rigid motions in the
euclidean plane. In the last two sections, we apply the previous results to construct a
new geometric structure on the Riemann moduli space Mg,n of a surface of genus g
with n punctures. More specifically, we show that this moduli space is a good orbifold1
which admits a family of geometric structures locally modeled on the homogeneous
spaces = T2g × CP2g+n−3 .
We now discuss our main result. We first define a punctured surface g,n of type
(g, n) to be a fixed connected closed orientable surface S of genus g together with a
distinguished set of n pairwise distinct points p1 , p2 , . . . , pn ∈ g,n . The Teichmüller
space Tg,n of g,n is the set of conformal structures on g,n modulo isotopies fixing
the punctures (see Section 5 for a precise definition). This space is a real analytic
variety in a natural way; if 2g − 2 + n > 0, then it is isomorphic to R6g−6+2n . The
group of isotopy classes of orientation-preserving diffeomorphisms of g,n fixing the
punctures is called the pure mapping class group and denoted by PModg,n . It acts in
a natural way on the Teichmüller space Tg,n .
We are now in a position to state the main result:
Theorem. Given a punctured surface g,n such that 2g +n−2 > 0, we can construct
a group homomorphism
: PModg,n → G = Aut(T2g ) × PGL2g+n−2 (C)
and a -equivariant local homeomorphism
H : Tg,n → = T2g × CP2g+n−3 .
To say that H is -equivariant means that H (Aμ) = (A) · H (μ) for all A ∈
PModg,n and μ ∈ Tg,n .
nThe pair (H, ) depends on n parameters β1 , β2 , . . . , βn ∈ (−1, ∞) such that
j =1 βj = 2g − 2 and no βi ∈ Z.
1 Recall that an orbifold is a space which is locally the quotient of a manifold by a finite group. A good
orbifold is globally the quotient of a manifold by a group acting properly and discontinuously (but in general not
freely).
Chapter 12. On the moduli space of singular euclidean surfaces 509
The moduli space Mg,n of g,n is the set of conformal structures on g,n modulo
diffeomorphisms fixing the punctures. It is the quotient of the Teichmüller space by
the pure mapping class group of g,n ; in other words Mg,n is a good orbifold whose
universal cover is Tg,n and fundamental group is PModg,n . In the geometric language
of (G, X)-strucures on manifolds and orbifolds (see [7], [9], [14], [29]), this theorem
says that we have constructed a family of geometric structures on the orbifold Mg,n
which is modeled on the homogeneous space = T2g × CP2g+n−3 . This family is
parametrized by the βi ’s.
The composition of the map H in the theorem with the projection on the torus
T gives us a map ρ : Tg,n → T2g called the character map. It was proved by
2g
W. A. Veech that this map is a real analytic submersion. Its fibers describe a foliation
whose leaves carry a geometric structure locally modelled on the complex projective
space CP2g+n−3 , see [35] for proofs of these facts and a discussion of other related
geometric structures on Tg,n .
The proof of this theorem is based on the following strategy: we first show that the
Teichmüller space can be seen as a deformation space of flat metrics on g,n having
conical singularities of prescribed angles at the punctures. We associate to such a
metric a homomorphism, called the holonomy of the metric, from the fundamental
group of the surface to the group SE(2) of direct isometries of the euclidean plane.
We then show that such a homomorphism can be seen as a point in the variety . In
brief, H : Tg,n → maps the isotopy class of a singular flat metric to its holonomy
representation.
In the special case of the punctured sphere, a stronger form of this theorem has
been obtained by Deligne and Mostow [10] using some techniques from algebraic
geometry and by Thurston [30] using an approach closer to ours.
To conclude this introduction, let us stress that the importance of piecewise flat
metrics in Teichmüller theory is illustrated by the large number of papers dedicated
to this subject. In addition to the work of Veech and Thurston already quoted, let
us mention the contributions of Rivin [27], Bowditch [5], Epstein and Penner [11]
to name a few. Piecewise euclidean metrics also appear in quantum gravity and in
topological quantum field theory, see [2], [8] and the references therein. Although the
present chapter starts with elementary considerations, the reader ought not to consider
it as a global survey of this vast subject.
Acknowledgments. I would like to thank Babak Modami and François Fillastre for
having carefully read the manuscript and for their comments. Finally, this exposition
is dedicated to the memory of Michel Matthey.
510 Marc Troyanov
2 Piecewise flat surfaces

2.1 Euclidean triangulation on a surface
A piecewise flat surface is a metric space obtained by properly gluing a stock of
euclidean triangles in such a way that whenever two triangles meet along an edge,
they are glued by an isometry along that edge. More precisely:
Definition 2.1. A euclidean triangulation of a surface is a set of pairs T =

{(Tα , fα )}α∈A where each Tα is a compact subset of and fα : Tα → R2 is a homeo-
morphism onto a non degenerate triangle fα (Tα ) in the euclidean plane R2 . A subset e
of Tα is an edge if fα (e) is an edge of fα (Tα ) and a point p of Tα is a vertex if its
image under fα is a vertex of fα (Tα ).
The eucliden triangulation T is subject to the following conditions:

(i) The triangles cover the surface: = α Tα .
(ii) If α = β, then the intersection Tα ∩ Tβ is either empty, or an edge or a vertex.
(iii) If Tα ∩ Tβ = ∅, then there is an element gαβ ∈ E(2) (the group of isometries of
the euclidean plane) such that fα = gαβ fβ on that intersection.
An element (Tα , fα ) ∈ T is called a triangle or a 2-simplex of the triangulations,

we often just denote it by Tα . The vertices and edges are called 0- and 1-simplices
respectively.
Two euclidean triangulations T = {(Tα , fα )}α∈A and T = {(Tα , fα )}α∈A of the
same surface are considered to be equal if they have the same simplices and, for
any α ∈ A, there is an isometry gα ∈ E(2) such that fα = gα fα .
Definition 2.2. A piecewise flat surface (, T ) is a surface together with a euclidean
triangulation.
A piecewise flat surface (, T ) comes with a number of additional structures. In

particular there is a well defined area measure which coincides with the 2-dimensional
Lebesgue measure on each euclidean triangle T . We can also define the length (c)
of an arbitrary curve c : [0, 1] → by the following axioms:
(i) if c is contained in a triangle T of T , then (c) is the euclidean length.
(ii) is additive: if c is the concatenation of two curves c1 c2 , then (c) = (c1 ) +
(c2 ).
The piecewise flat surface is thus a length space (see [6] for this notion). If the surface
is connected, then it is also a metric space for the distance given by the infimum of
the lengths of all curves joining two given points.
There is one more structure, called the singularity order and which is defined as
the angular excess at the vertices counted in number of turns. It tells us how singular
each vertex is compared to an ordinary point; the precise definition is the following:
Definition 2.3. The vertex p ∈ is said to be a conical point of total angle θ if

k

θ= ϕj
j =1
where ϕ1 , . . . , ϕk are the angles of all the triangles in T which are incident to p. The
singularity order β(p) of a vertex p is the angular excess at p measured in number of
turns:
θ
β(p) = − 1.
2π
We extend the function β to all points of by setting β(x) = 0 if x ∈ is not a
vertex. The point x is then termed singular if β(x) = 0 (i.e. if θ = 2π) and regular
otherwise.
Proposition 2.4 (Gauss–Bonnet formula). For any euclidean triangulation on a com-

pact surface without boundary , we have

χ() + β(x) = 0, (2.1)
x∈
where χ () is the Euler characteristic of the surface.
The proof is a direct counting argument based on the definition of the Euler char-
acteristic and the fact that the three internal angles of a euclidean triangle add up to π ,
see [31].
2.2 The universal branched cover of a piecewise flat surface

If (, T ) is a piecewise flat surface, we denote by = \ {p1 , . . . , pn } the open
surface obtained by removing the singular vertices p1 , . . . , pn ∈ .
Definition 2.5. A path c : [0, 1] → is admissible if it has finitely many intersections

with the edges of the triangulation and if c(s) ∈ for any 0 < s < 1. A homotopy ct
is an admissible homotopy if s → ct (s) ∈ is an admissible path for any 0 ≤ t ≤ 1.
Let us choose a fixed triangle T0 ∈ T and call it home (or the base triangle). We
also choose a base point x0 in the interior of T0 .
Definition 2.6. The universal branched cover of (, T ) is the euclidean two-dimen-
sional complex Tˆ obtained as follows: a k-simplex σ̂ of Tˆ , where k = 0, 1 or 2, is a
pair (σ, [c]) where σ is a k-simplex of T and [c] is an admissible homotopy class of
paths joining T0 to a point in σ .
The universal branched cover Tˆ is a simplicial complex (which is not locally finite)
and there is an obvious simplicial map Tˆ → T sending (σ, [c]) to σ .
512 Marc Troyanov
We denote by ˆ the geometric realization of Tˆ . This is a triangulated topological

space and it comes with a continuous surjective map P : ˆ → sending each simplex
ˆ
of T homeomorphically onto the corresponding simplex in T . We turn ˆ into a
metric space (in fact a length space) by requiring P to be an isometry on each simplex
(concretely, we give to each simplex σ̂ = (σ, [c]) in Tˆ the geometry of the euclidean
simplex σ in T ).
Another way to understand ˆ is the following: let
˜ be the universal cover of .
It is naturally a length space (in fact a flat Riemannian surface) and ˆ is its metric
completion.
2.3 The developing map of a piecewise flat surface

Definition 2.7. An edge of the piecewise flat surface (, T ) is said to be interior if
it is not contained in the boundary of . The hinge of an interior edge e is the unique
pair of triangles T1 , T2 ∈ T which are incident with e.
Given an interior edge e with hinge (T1 , T2 ) and an isometry f1 : T1 → R2 , there

exists a unique isometry f2 : T2 → R2 such that f1 (T1 ) and f2 (T2 ) have disjoint
interiors and f1 (e) = f2 (e). By juxtaposing these maps, we obtain a map
fe = f1 ∪ f2 : T1 ∪ T2 → R2
which is an isometry of the hinge onto a quadrilateral in the euclidean plane. The
map fe just described is called an unfolding of the hinge. One also says that f2 is the
continuation of f1 across the edge e.
The notions of hinge, unfolding and continuation of an isometry across an edge
ˆ
are similarly defined on the universal branched cover .
Proposition 2.8. Let (, T ) be a piecewise flat surface with home triangle T0 and
choose an isometry f0 from T0 onto a triangle in R2 . Then there exists a unique map
f: ˆ → R2 such that f coincides with f0 on T0 and f maps every hinge onto a
quadrilateral in R2 .
Proof. Let x̂ be a point in . ˆ This point belongs to a simplex σ̂ = (σ, [c]) in Tˆ .

Choose an admissible arc c connecting the base point x0 ∈ T0 to σ . Because c is
admissible, it crosses only finitely many edges e1 , e2 , . . . , em in that order (repetitions
may occur). We associate to the path c a sequence of triangles T1 , T2 , . . . , Tm ∈ T
by requiring that (T0 , T1 ) be the hinge of e1 , (T1 , T2 ) be the hinge of e2 and so on.
We then define fj : Tj → R2 to be the continuation of fj −1 across the edge ej (for
1 ≤ j ≤ m) and we finally set f (x̂) = fm (P (x̂). The point f (x̂) ∈ R2 only depends
on the homotopy class [c] and not on the representative path c.
ˆ → R2 maps every hinge onto a quadri-
It is clear from the construction that f :
lateral in R . Since f extends f0 , the proof is complete.
2 2
ˆ → R2 is the developing map of the piecewise flat

Definition 2.9. The map f :
surface.
Figure 1. The developing map.
ˆ → R2 is another developing map, then, clearly, f = g f where

If f :
g : R → R2 is the unique isometry of the plane such that g(f (T0 )) = f (T0 ).
2
When is the boundary surface of a convex polyhedron in R3 , the developing

map is a very concrete operation. It is obtained by first placing the initial face (home)
somewhere on the plane and then rolling without slipping the polyhedron, face af-
ter face, following an admissible path. Observe in particular that we can move our
polyhedron toward any point in the plane. This is a general fact:
Proposition 2.10. Let be a compact piecewise flat surface without boundary. Then
ˆ → R2 is surjective.
any developing map f :
Proof. Observe first that f is a closed map (because it is an isometry on each triangle).
Suppose that R2 \ f ()ˆ = ∅, then this set is open and we can find a point y ∈ R2
which lies on the boundary of f (). ˆ Because f is closed, we can find x̂ ∈ ˆ with
f (x̂) = y. Let x = P (x̂) ∈ . This point cannot be in the interior of any triangle of
the triangulation, thus x lies on an edge e. Moving slightly the point y if necessary,
we can assume that x lies in the interior e (i.e. that x is not a vertex).
Since has no boundary, e is an interior edge; the developing map f sends the
hinge of e onto a quadrilateral Q in R2 . The interior of e is sent in the interior of
ˆ This contradicts the point y lying on the boundary of f ().
Q ⊂ f (). ˆ 2
514 Marc Troyanov
2.4 The holonomy of a piecewise flat surface

The set of all admissible homotopy classes in the piecewise flat surface (, T ) which
start and end at the base point x0 form a group π with respect to the concatenation.
This group coincides with the fundamental group π1 ( , x0 ).
If [a] ∈ π and (σ, [b]) ∈ T˜ , then (σ, [ba]) is well defined, and this gives a
simplicial action of π on T˜ .
Corresponding to this simplicial action, there is an action of π on ˆ by isometries;
ˆ
the orbits space of this action coincides with the surface itself. i.e. = /π.
ˆ → R is a developing map of (, T ) and γ = [c] ∈ π , then there is a
If f : 2
unique isometry g : R2 → R2 such that the g(f (T0 )) = f (T0 , [c]). We denote this
isometry by g = ϕ(γ ).
Proposition 2.11. The map ϕ : π → E(2) (the group of isometries of the euclidean
plane) is a group homomorphism.
Proof. This easily follows from the construction of the developing map. 2
Definition 2.12. The homomorphism ϕ : π → E(2) is called the holonomy associated

to the developing map f .
Proposition 2.13. If is compact without boundary, then the group H = ϕ(π ) ⊂

E(2) has no bounded orbit (in particular it has no fixed point).
An obvious but important consequence is the fact that H is not conjugate to a

subgroup of O(2).
Proof. Suppose that there is a point y ∈ R2 such that H · y is bounded. Since the
developing map is surjective, there exists a point x̂ ∈ ˆ such that f (x̂) = y. Observe
that H · y = f (π · x̂). Any point in the surface can be connected to x = P (x̂) by
a path of length at most D = diam(), hence any point in ˆ can be connected to a
point in the orbit π · x̂ by a path of length at most D.
Since f preserves the length of all paths, it follows that any point in the image
ˆ can be connected to a point in the orbit H · y by a path of length at most D. The
f ()
last assertion contradicts the surjectivity of f . 2
Recall that the developing map of a piecewise flat surface is not unique, it depends
on the choice of an isometry of the home triangle into R2 . However the holonomy is
well defined up to conjugacy:
Proposition 2.14. Let f, f : ˆ → R2 be two developing maps of the piecewise flat

surface (, T ), and let ϕ, ϕ : π → E(2) be the corresponding holonomies. Then
ϕ (γ ) = gϕ(γ )g −1 where g ∈ E(2) is the unique isometry such that f = g f .
Proof. The holonomy is defined by the condition f (T0 , γ ) = ϕ(γ )(f (T0 )), hence
ϕ (γ )(f (T0 )) = f (T0 , γ ) = g f (T0 , γ ) = g ϕ(γ )(f (T0 ))
= g ϕ(γ ) g −1 (f (T0 )). 2
2.5 The developing map near a singularity

The previous notions can be clearly visualized if one restricts one’s attention to a
simply connected region ⊂ which is a union of triangles and which contains
exactly one singular vertex p of order β = β(p) = 0.
Suppose that the base point x0 sits in and choose a loop c in = \ {p}, based
at x0 and surrounding the point p once (so that [c] is a generator of π1 ( , x0 ) ∼ = Z).
Choose a connected component ˆ of the inverse image P −1 () ⊂ ˆ and still
denote by P the (restriction of the) projection P : ˆ → .
ˆ
We want to describe the geometry of , of the map P as well as the developing
map and holonomy restricted to . ˆ
It is enough to consider the case where is the “star” of the vertex p, i.e. the union
of all triangles incident with p (if is a larger region, the other triangles will simply
appear as an appendix glued to the star of p).
The space ˆ is the geometric realization of a simplicial complex whose simplices
are simplices in together with an admissible homotopy class of curve connecting
the base point to the given simplex.
Let us denote by T1 , T2 , . . . , Tk the list of all triangles (i.e. 2-simplices) incident
with p and assume that x0 ∈ T1 . Assume also that Ti has a common edge with Ti+1
and Tk has a common edge with T1 . Then a triangle in ˆ is given by a pair (Ti , [a])
where [a] is the homotopy class of a curve a in from x0 to Ti . This homotopy class
is parametrized by a single integer d ∈ Z (the degree of a) which counts the number
of times a turns around the point p. In other words, ˆ is an infinite strip made out
of countably many copies of each triangle T1 , T2 , . . . , Tk each indexed by the degree
d∈Z
ˆ =
T1,d ∪ T2,d ∪ · · · ∪ Tk,d .
d∈Z
To develop , ˆ start with an isometry f1 from T1 to a triangle in the euclidean plane

and continue this isometry by unfolding each hinge in . ˆ The developing map f then
clearly satisfies
f (Ti,d ) = R d f (Ti,0 )
where R is a rotation of angle θ (= the sum of the angles at p of the triangles
T1 , T2 , . . . , Tk ) around the point q = f1 (p). The rotation R ∈ E(2) is clearly the
holonomy of the generator [c] of π1 ( , x0 ).
We collect in the next proposition, some of the conclusions of the previous discus-
sion:
516 Marc Troyanov
ˆ contains exactly one

Proposition 2.15. (1) The inverse image P −1 (p) of p in
point p̂.
(2) The holonomy ϕ(c) of [c] is a rotation of angle θ = 2π(β + 1).
(3) If β is not an integer, then q = f (p̂) is the unique fixed point of the rotation ϕ(c).
2
2.6 Geometric equivalence of euclidean triangulations

Let (, T ) be a piecewise flat surface. Choose a triangle Tα0 ∈ T and a point q in
the interior of an edge of Tα0 . If one connects the point q to the opposite vertex in Tα0
by a euclidean segment, one obtains two subtriangles Tα 0 , Tα0 whose union is Tα0 .
If one replaces the triangle Tα0 with Tα 0 and Tα0 in the triangulation Tq , one obtains
a new triangulation Tq .
Definition 2.16. a) The triangulation Tq is said to be obtained from T by an elementary

subdivision.
b) The geometric equivalence is the equivalence relation on the set of euclidean
triangulations on a surface which is generated by elementary subdivisions.
In other words, two euclidean triangulations T1 , T2 on the surface are geo-

metrically equivalent if there is a common subdivison T which is also a euclidean
triangulation.
Proposition 2.17. The area measure dA, the length structure , the singularity or-
der β, the developing map and holonomy are invariants of this equivalence relation.
Proof. The statement is obvious for dA, and β. Observe now that if T is a triangle
of T and T , T is an elementary subdivision of T , then the pair (T , T ) is the hinge
of their common edge e ⊂ T . Observe also that if f : T → R2 is an isometry, then f
is an unfolding of that hinge.
This argument shows that the developing map remains unchanged when subdivising
the triangulation. Since the developing map is invariant, so is the holonomy. 2
2.7 Flat metrics with conical singularities

If (, T ) is a piecewise flat surface, then carries a well defined Riemannian metric
m; this metric is flat (i.e. it has no curvature) and in the neighbourhood of a conical
singularity of total angle θ , we can introduce polar coordinates (r, ϕ), where r ≥ 0 is
the distance to p and ϕ ∈ R/(θ Z) is the angular variable (it is defined modulo θ). In
these coordinates, the metric reads
m = dr 2 + r 2 dϕ 2 .
A calculation shows that this metric can be written as

mβ = |z|2β |dz|2 , (2.2)
where z = iϕ β+1
β+1 (re )
1
(see [31]).
Definition 2.18. A flat surface with conical singularities (, m) is a surface together
with a singular Riemannian metric m which is isometric to the metric mβ in (2.2) in
the neighbourhood of every point p ∈ , where β = β(p) ∈ (−1, ∞).
One says that β(p) is the singularity order of p and p is a conical singularity if
β(p) = 0.
The singular points form a discrete set and the formal sum (with discrete
support) β(p) p is called the divisor of the singular metric m. One sometimes also
says that m represents this divisor.
Proposition 2.19. Any compact flat surface with conical singularities (, m) can be
geodesically triangulated. The resulting triangulation is a euclidean triangulation on
and the associated length structure coincides with the length in the metric m. 2
A proof can be found in [31] and in [30]. See also [18] and [27] for further
discussions on triangulations of piecewise flat surfaces.
Proposition 2.20. Two euclidean triangulations T , T on a compact surface are

geometrically equivalent if and only if they give rise to the same flat surface with
conical singularities m on .
Proof. It is clear from Proposition 2.17, that two triangulations which are geometri-
cally equivalent give rise to the same singular flat metric. Conversely, suppose that the
triangulations T and T define the same metric, then each triangle of T is decomposed
by T in a finite number of polygonal regions. We may then further decompose these
polygons in euclidean triangles, and we thus obtain a new euclidean triangulation of
which is a subdivision of both T and T . 2
2.8 Relation with Riemann surfaces

If (, m) is an oriented flat surface with conical singularities, then it is covered by
charts {(Uj , zj )} such that the metric m takes the form (2.2) in each Uj . The transition
from one such coordinate zj to another one is given by a conformal transformation.
Thus is a Riemann surface with a holomorphic atlas given by {(Uj , zj )}.
Remark 2.21. The reader should observe here that the conical singularities are invis-
ible from the conformal viewpoint. This is a consequence of the formula (2.2) which
shows that the singular metric is conformal to a smooth metric. It can also be seen
as a consequence of the theorem of removability of singularities of locally bounded
meromorphic functions.
518 Marc Troyanov
In the converse direction, we can start with a closed Riemann surface with a divisor
and ask whether there is a conformal flat metric representing this divisor. The answer
is positive and the following theorem classifies all compact euclidean surfaces with
conical singularities.
Theorem 2.22. Let be a compact connected Riemann surface without boundary. Fix
n distinct points p1 , p2 , . . . , pn ∈ and n real numbers β1 , β2 , . . . , βn ∈ (−1, ∞).
There exists a conformal flat metric m on having a conical singularity of order βj
at pj (j = 1, . . . , n) if and only if the Gauss–Bonnet condition χ (S) + jn=1 βj = 0
holds. This metric is unique up to homothety. 2
See [31], a shorter proof can be found in [33, §IV].
Remark 2.23. A careful examination of the proof shows that the metric m depends
continuously on all the parameters: The conformal structure, the points pj and the
orders βj .
There is a similar theorem for the case of hyperbolic metrics with conical singular-
ities, see [15], [23], [26], [32]. There are also various other extensions (non constant
curvature, non orientable surfaces, non compact surfaces, and surfaces with boundary,
see [17], [32], [33]). The case of spherical metric is more delicate, see [12], [34] for
a study of spherical metric with three conical singularities on the 2-sphere.
Theorem 2.24. Given a compact oriented surface , there are natural bijections
between the following three sets:
(1) The set of geometric equivalence classes of euclidean triangulations on up to
homothety;
(2) the set of flat metrics m on with conical singularities up to homothety;

(3) the set of conformal structures on together with a finite real divisor i βi pi
such that βi > −1 and the Gauss–Bonnet condition (2.1) is satisfied.
Proof. Theorem 2.22 says precisely that there is a bijection between sets (2) and (3).
Proposition 2.20 shows that there is a natural injection from (1) to (2), this injection
is surjective by Proposition 2.19. 2
3 Punctured surfaces
3.1 Punctured surfaces and their fundamental groups
Definition 3.1. We define a punctured surface g,n to be an oriented, closed connected
surface of genus g together with a distinguished set of n pairwise distinct points
p1 , p2 , . . . , pn ∈ g,n .
The points p1 , . . . , pn are considered to be special places (with some geometric

significance) on the surface. We call them the punctures and we denote by g,n the
surface obtained by removing them:

g,n = g,n \ {p1 , p2 , . . . , pn }.
The connected sum of two punctured surfaces is defined by removing a disk containing
no puncture in each surface and then gluing them along their boundary. The resulting
surface is again a punctured surface. In fact we have
g1 ,n1 # g2 ,n2 = g1 +g2 ,n1 +n2 ,
where the symbol # means the connected sum. In particular
g,n = g,0 # 0,n . (3.1)
is given
We easily deduce from this observation that the Euler characteristic of g,n
by

χ(g,n ) = 2 − 2g − n. (3.2)

If n > 0, then g,n can be homotopically retracted onto a bouquet of 2g + n − 1
circles and the fundamental group πg,n of g,n is thus a free group on 2g + n − 1
generators.
Note that πg,n also admits the following presentation with 2g + n generators and
one relation:

πg,n = a1 , . . . , ag , b1 , . . . , bg , c1 , . . . , cn | [ai , bi ] = cj , (3.3)

this presentation is a consequence of the identity (3.1) and Van Kampen’s theorem.
3.2 Uniformization of a punctured Riemann surface

Let us fix a conformal structure [m] on g,n . Assuming that 2 − 2g − n < 0, the
uniformization theorem states that ( , m) is conformally equivalent to U/ where
U = {z ∈ C | Re(z) > 0} is the upper-half plane, and ⊂ PSL2 (R) is a Fuchsian
group of the first kind2 isomorphic to πg,n .
The isomorphism πg,n → is compatible with the punctures in the sense that the
generator ci is sent to a parabolic element of and the generators ai , bi are sent to
hyperbolic elements (here, the letters ai , bi , cj refer to the presentation (3.3)).
Let us denote by ϒ ⊂ ∂ U = R ∪ {∞} the set cusp points of , i.e. the set of fixed
points of all parabolic elements in . Following [28, page 10], we define a topology
on Û = U ∪ ϒ as follows: for a point z ∈ U, the family of hyperbolic disks D(z, ρ)
is a fundamental system of neighborhoods of z. For a point y ∈ ϒ the family of
horodisks centered at y is a fundamental system of neighborhoods of y. With this
topology, Û is a Hausdorff space and acts by homeomorphisms. The space is not
2 Recall that a Fuchsian group is a discrete subgroup of PSL (R), it is of the first kind if there is a fundamental
2
domain D ⊂ U of finite hyperbolic area.
520 Marc Troyanov
locally compact and ϒ is topologically a discrete space. Standard arguments from

hyperbolic geometry (see e.g. [28]) show that the projection map P : U → U/ =
extends to a surjective continuous map
P̂ : Û → (3.4)
where Û = U ∪ ϒ. This extension maps ϒ to the punctures {pi } ⊂ .
Remarks. 1) The previous considerations show that there exists a unique metric m−1
on of constant curvature −1 which is complete, has finite volume and belongs to
the conformal structure [m]. This metric has a cusp at each puncture pi , it is the unique
metric such that P ∗ m−1 is the Poincaré metric on U; its existence can also be proved
by directly solving the prescribed curvature equation (see [16], [17]).
2) We know from Theorem 2.22 that the conformal class [m] also contains a metric
m0 on , unique up to homothety, which is flat and has a conical singularity of order
βj at pj (j = 1, . . . , n) provided (2.1) holds.
This flat metric lifts as a flat conformal metric m̃0 = P ∗ (m0 ) on U. For this metric,
U is not complete and its completion is given by Û = U. The map P̂ : Û → is thus
a concrete model of the universal branched covering introduced earlier.
We identify the set ϒ as a subset of as follows: we first fix a base point z̃0 ∈ U
and let z0 = P (z̃0 ) ∈ . For y ∈ ϒ, let us denote by γ̃y the hyperbolic ray in U
starting at z̃0 and asymptotic to the point y, and let γy = P (γ̃y ), this is a path joining
z0 to a puncture pi = P (y). Now let Di ⊂ be a small disk around pi containing
no other puncture, and let γy = γy \ Di .
We now define cy ∈ π1 ( , z0 ) to be the homotopy class of the path obtained by
following γy , then ∂Di (in the positive direction) and then (γy )−1 .
Di
pi
z0
Figure 2. The homotopy class cy .
Recall that we have a canonical isomorphism, ∼

= π1 ( , z0 ) = πg,n , we have
thus constructed a map
ϒ → ,
(3.5)
y → cy .
It is clear that cy ∈ is a parabolic element fixing y, in particular, the map ϒ →
is injective.
3.3 Some groups of diffeomorphisms of a punctured surface

Given a punctured surface = g,n , we define Diff g,n to be the group of diffeo-
momorphisms h : → which leaves the set {p1 , . . . , pn } of punctures invari-
ant. We also introduce the following subgroups: Diff + g,n ⊂ Diff g,n is the subgroup
of orientation preserving diffeomomorphisms, PDiff g,n is the subgroup of pure dif-
feomomorphisms, i.e. diffeomomorphisms fixing each puncture pi individually and
PDiff + +
g,n = PDiff g,n ∩ Diff g,n .
Every element h ∈ Diff +g,n permutes the punctures and we have an exact sequence
1 → PDiff + +
g,n → Diff g,n → Sym(n) → 1.
where Sym(n) is the permutation group of {p1 , . . . , pn }.
We also define Diff 0g,n ⊂ PDiff +
g,n to be the group of diffeomorphisms which are
isotopic to the identity through an isotopy fixing the punctures. The quotient
Modg,n = π0 (Diff + +
g,n ) = Diff g,n / Diff g,n ,
0
is called the mapping class group or the modular group of the punctured surface g,n ,
and
PModg,n = π0 (PDiff + +
g,n ) = PDiff g,n / Diff g,n ,
0
is the pure mapping class group.

These groups have been intensely studied since the pioneer work of Dehn and
Nielsen. We refer to [4], [19], [24], [36] among many other papers for more informa-
tion.
3.4 Outer automorphisms

The mapping class group is related to the group of outer automorphisms of the funda-
mental group of . Let us recall this purely algebraic notion: If π is an arbitrary group,
we denote by Aut(π) the group of all its automorphisms and by Inn(π ) ⊂ Aut(π ) the
subgroup of inner automorphisms (i.e. conjugations γ → αγ α −1 ). This is a normal
subgroup.
Definition 3.2. The group of outer automorphisms of π is the quotient

Out(π) = Aut(π )/ Inn(π ).
An outer automorphism is thus an automorphism of π defined up to conjugacy.
Lemma 3.3. There is a naturally defined group homomorphism

Modg,n → Out(πg,n ).
Proof. This homomorphism is defined as follows. Let h ∈ Diff( ) be an arbitrary

diffeomorphism and fix a base point ∗ and a path δ in connecting ∗ to h(∗). If γ
522 Marc Troyanov
is a loop in based at ∗, then we set

hδ (γ ) = δ −1 (h γ )δ.
This defines an automorphism hδ,# ∈ Aut(πg,n ).
If δ is another path connecting ∗ to h(∗), then hδ,# and hδ ,# are conjugate by
δ δ . The outer automorphism h# ∈ Out(πg,n ) is thus well defined independently
−1
of the choice of the path δ and it is clear that if h is homotopic to the identity, then it
acts trivially on πg,n , i.e. we have defined a map Modg,n → Out(πg,n ). It is routine
to check that it is a group homomorphism. 2
Introducing the group POut(πg,n ) ⊂ Out(πg,n ) of all outer automorphisms pre-
serving the conjugacy class of each ci (i = 1, . . . , n) in the presentation (3.3), we
have the following deep result:
Theorem 3.4. If g > 0 and n > 0, then the homomorphism defined in the previous
lemma induces an isomorphism
∼
: PModg,n −−→ POut(πg,n ). (3.6)
2
This is the so-called Dehn–Nielsen–Baer Theorem, see [19], [36] for a proof.
3.5 Lifting the group Diff 0 ( ) on U

Using the notations of Section 3.2, one writes the universal branched covering of g,n
as P̂ : Û → , where Û = U ∪ ϒ (we still assume 2 − 2g − n < 0).
We denote by Diff + (U) the group of orientation preserving diffeomorphisms of
U = ˜ and we define the normalizer N () and the centralizer C() of in Diff + (U)
by
N () = {h ∈ Diff + (U) | h = h}.
and
C() = {h ∈ Diff + (U) | h γ = γ h for all γ ∈ }.
Observe that C() = ker(ψ), where ψ : N () → Aut() is defined by ψ(h) : γ →
hγ h−1 .
The center of is the intersection Z() = ∩ C(); it is the largest abelian
subgroup of .
Lemma 3.5. Let be an arbitrary Fuchsian group, then Z() is trivial unless is
cyclic.
Proof. This follows from classical Fuchsian group theory. Indeed, it is well known
that if γ1 , γ2 are non-trivial elements in PSL2 (R), then they commute if and only if
they have the same fixed points (see e.g. [20, Theorem 2.3.2]). So if Z() contains a
non-trivial element γ0 , then any γ ∈ \ {id} must have the same fixed points as γ0
and it follows from [20, theorem 2.3.5]) that is cyclic. 2
Recall the projection P : U → = U/ . For any element h ∈ N(), we define
P∗ h : → by P∗ h(x) = P (h(x̃)) where x̃ ∈ U is an arbitrary point in P −1 (x).
This map is well-defined, because the condition h = h means precisely that h maps
-orbits in U to -orbits, and it is clearly a diffeomorphism. We thus have defined a
map
P∗ : N () → Diff + ( ),
and it is obviously a group homomorphism.
Proposition 3.6. is a normal subgroup in N() and P∗ defines an isomorphism

from N ()/ to Diff + ( ).
Proof. It is obvious that ⊂ N () is normal and that P∗ () = {id}. In particular
P∗ factors through a well defined homomorphism N()/ → Diff + ( ). This
homomorphism is surjective since every diffeomorphism of lifts to the universal
cover U of .
Suppose now that P∗ h = id. Then h(x) ∈ · x for all x ∈ U . This means that
there exists a map U → , x → γx such that h(x) = γx x for all x ∈ U . Since h
is continuous, so is this map, but this implies that x → γx is constant because is
a discrete group. It follows that h ∈ and we have shown that P∗ : N()/ →
Diff + ( ) is also injective. 2
Lemma 3.7. P∗ maps C() isomorphically onto Diff 0 ( ).
Proof. Suppose that P∗ h ∈ Diff 0 ( ). Then there exists an isotopy ht ∈ N() such
that h0 = id and h1 = h. Hence ψ(ht ) ∈ Aut() is constant by continuity. Because
ψ(h0 ) = ψ(id) ∈ Aut() is the trivial element, we have h ∈ ker ψ = C().
In the reverse direction, we use an argument going back to Nielsen: Suppose that
h ∈ ker ψ = C() and define ht (x) ∈ U to be the point on the hyperbolic segment
[x, h(x)] such that d(x, ht (x)) = td(x, h(x)) (where d is the hyperbolic distance
in U). Since h ∈ ker ψ and preserves the hyperbolic distance in U, the segment
[γ x, h(γ x)] coincides with [γ x, γ h(x)] for any x ∈ U and any γ ∈ . Therefore
we have ht (γ x) = γ ht (x), i.e. ht ∈ C() ⊂ N(). The path P∗ ht ∈ Diff( ) is an
isotopy from P∗ h to the identity and we conclude that P∗ h ∈ Diff 0 ( ).
We have proved that P∗−1 (Diff 0 ( )) = C(). It is now clear that P∗ : C() →
Diff 0 ( ) is an isomorphism since its kernel is C() ∩ = Z() = {id}. 2
Corollary 3.8. P∗ induces an isomorphism from N ()/( × C()) to the modular

group Modg,n .
524 Marc Troyanov
4 The representation variety of a finitely

generated group in SE(2)
Given a finitely generated group π and an algebraic Lie group G, it is easy to see
that the set Hom(π, G) is an algebraic set. The group G itself acts on Hom(π, G) by
conjugation: g · ϕ(γ ) = g −1 ϕ(γ )g. The quotient space is called the representation
variety of π in G and denoted by
R(π, G) = Hom(π, G)/G.
This variety plays an important role in the study of geometric structures on manifolds,
see e.g. [14].
The discussion in Section 2.4 shows that an element of the representation variety
R(π, E(2)) is associated to any piecewise flat surface (, T ) (where π = π1 ( , x0 )).
In the present section, we investigate the structure of R(π, E(2)) (in fact, for conve-
nience, we shall restrict ourself to the subgroup SE(2) ⊂ E(2) of orientation preserving
isometries of the euclidean plane, this is a subgroup of index 2).
4.1 On the cohomology of groups

We will need some elementary results from group cohomology; here we recall a few
basic definitions and facts.
Let π be an arbitrary group and A be a π -module, i.e. an abelian group with a
representation ρ : π → Aut(A).
Definition 4.1. (1) A 1-cocycle in A is a map σ : π → A such that

σ (γ1 γ2 ) = σ (γ1 ) + ρ(γ1 ) · σ (γ2 )
for any γ1 , γ2 ∈ π . The set of 1-cocycles is an abelian group denoted by Z 1 (π, A).
(2) The 1-cocycle σ ∈ Z 1 (π, A) is a 1- coboundary if it can be written as
σ = δa (γ ) = ρ(γ ) · a − a
for some element a ∈ A. The set of 1-coboundaries is a subgroup of Z 1 (π, A) denoted
by B 1 (π, A).
(3) The quotient
H 1 (π, A) = Z 1 (π, A)/B 1 (π, A)
is the first cohomology group of π with values in A.
Example. Let us compute the first cohomology group when A = k is a field and π is
a finitely generated group. We denote by kρ the π -module k with the representation
ρ : π → Aut(k).
Assume first that the representation ρ : π → Aut(k) is a scalar representation,

i.e. ρ : π → k ∗ ⊂ Aut(k) and that π = Fs = a1 , a2 , . . . , as is a free group on s
generators.
Since π is free, the homomorphism ρ : π → k ∗ is completely determined by the
vector r = (ρ(a1 ), ρ(a2 ), . . . , ρ(as )) ∈ (k ∗ )s . Likewise, a cocycle is given by the
vector
t = (τ (a1 ), τ (a2 ), . . . , τ (as )) ∈ k s .
There is no restriction on the vector t ∈ k s (again because π is free) and thus
Z 1 (π, kρ ) ∼
= ks . (4.1)
An element σ ∈ Z 1 (π, kρ ) is a coboundary if σ = u(id − ρ) for some u ∈ k, thus
B 1 (π, kρ ) ∼
= k · (1 − ρ(a1 ), 1 − ρ(a2 ), . . . , 1 − ρ(as )) ∈ k s .
Let us choose a linear form μ : k s → k such that μ ≡ 0 if ρ is trivial and
μ(1 − ρ(a1 ), 1 − ρ(a2 ), . . . , 1 − ρ(as )) = 0
else. It is easy to check that
B 1 (π, kρ ) ⊕ ker μ = k s = Z 1 (π, kρ )
in k s and we thus obtain the following
Proposition 4.2. For any free group on s generators, we have

ks if ρ is trivial,
H 1 (π, kρ ) = Z 1 (π, kρ )/B 1 (π, kρ ) = ker μ ∼
= s−1
k otherwise. 2
General case. Let us compute the first cohomology group when A = k is a field and
π is a finitely presented group with presentation
π = S | R.
Here S = {a1 , a2 , . . . , as } ⊂ π denotes a finite set generating the group and R =
{r1 , r2 , . . . , rm } ⊂ F (S) (the free group on S) is a finite set of words in S defining
all the relations among the elements of S. We denote by kρ the π -module k with the
representation ρ : π → Aut(k).
For any relation r = ai1 ai2 . . . aip ∈ R, we introduce the linear form λr : k s → k
defined by
p
λr (σ ) = ρ(aiν ) σ (aiμ ) (4.2)
μ=1 ν<μ
and we define : k s → k m , by
(σ ) = (λr1 (σ ), . . . , λrm (σ )). (4.3)
526 Marc Troyanov
Lemma 4.3. The space of 1-cocycles in kρ is given by

Z 1 (π, kρ ) = ker = ker λr ⊂ k s .
r∈R
Proof. If σ ∈ Z 1 (π, kρ ) and r = ai1 ai2 . . . aip ∈ R, then we deduce from the cocycle
relation that
0 = σ (r) = σ (ai1 ai2 . . . aip ) = σ (ai1 ) + ρ(ai1 )σ (ai2 . . . aip )
= σ (ai1 ) + ρ(ai1 )σ (ai2 ) + ρ(aii1 )ρ(ai2 )σ (ai3 . . . aip )
p

= ρ(aiν ) σ (aiμ ). 2
μ=1 ν<μ
On the other hand, since any 1-coboundary in kρ is a multiple of ρ − 1, we have

B 1 (π, kρ ) = k · (ρ − 1) ⊂ k s .
We have proved the following
Proposition 4.4. The first cohomology group of the finitely presented group π =
S | R with value in kρ is given by
H 1 (π, kρ ) = ker /(k(ρ − 1)). 2
In particular, if π has exactly one non trivial relation, then

s−1 if ρ is trivial,
H (π, k) =
1 ∼ k
k s−2 otherwise,
where s = Card(S) is the number of generators.
4.2 Abelian representations

Representations of a finitely presented group π in an abelian Lie group are easy to
describe:
Lemma 4.5. If G is an abelian group, then R(π, G) = Hom(π, G). This set is itself
an abelian topological group.
Proof. Since there are no non trivial inner automorphisms in an abelian group, it is
clear that R(π, G) = Hom(π, G).
We endow Hom(π, G) with the compact open topology and we define a product
on this space by
(ϕ1 ϕ2 )(γ ) = ϕ1 (γ )ϕ2 (γ )
for ϕ1 , ϕ2 ∈ Hom(π, G) and γ ∈ π . The following calculation shows that Hom(π, G)

is a group for this multiplication:
(ϕ1 ϕ2 )(γ1 γ2 ) = ϕ1 (γ1 γ2 )ϕ2 (γ1 γ2 )
= ϕ1 (γ1 )ϕ1 (γ2 )ϕ2 (γ1 )ϕ2 (γ2 )
= ϕ1 (γ1 )ϕ2 (γ1 )ϕ1 (γ2 )ϕ2 (γ2 )
= (ϕ1 ϕ2 )(γ1 )(ϕ1 ϕ2 )(γ2 ).
The identity e in Hom(π, G) is the trivial representation. Observe finally that this
group is abelian since ϕ1 (γ )ϕ2 (γ ) = ϕ2 (γ )ϕ1 (γ ). 2
Recall that the abelianized group of π is the abelian group

Ab(π) = π/[π, π].
Another useful remark is that if G is abelian, then
Hom(π × π , G) = Hom(π , G) × Hom(π , G)
for any groups π , π .
Assume now that π is a finitely generated group. Ab(π ) is then an abelian group
of finite type, hence
Ab(π) = π/[π, π ] = Zr ⊕ F
where F is a finite abelian group (the torsion) and r ∈ N is the abelian rank of π .
We obviously have Hom(π, G) = Hom(Ab(π ), G) and it is clear that all repre-
sentation varieties of a finitely generated group π in an abelian Lie group G can be
deduced from the following special cases:
(1) Hom(Z, R) = R;
(2) Hom(Z, U (1)) = U (1);
(3) Hom(Z/mZ, R) = 0;
(4) Hom(Z/mZ, U (1)) = {z ∈ C | zm = 1}.
For instance, if π is the free group on s generators, then Ab(π ) = Zs . Thus
Hom(π, R) = Rs and Hom(π, U (1)) = Ts .
Another simple example, with torsion, is the group π = a, b, c | [a, b] = cm .
We have Ab(π ) = Z2 ⊕ Z/mZ, therefore Hom(π , R) = R2 and
Hom(π , U (1)) = T2 ⊕ {e2kiπ/m | m = 0, 1, . . . , m − 1}.
4.3 Representations in SE(2)

We denote by SE(2) = Iso+ (R2 ) the group of orientation preserving isometries of the
euclidean plane.
528 Marc Troyanov
We may identify the euclidean plane with the complex line C: any g ∈ SE(2) can
then be written as g(z) = u · z + v where u ∈ U (1) ⊂ C∗ and v ∈ C. We thus identify
SE(2) with the subgroup of GL2 (C) consisting of matrices of the form

uv u, v ∈ C, |u| = 1 .
SE(2) = 0 1
In particular SE(2) is a semidirect product U (1)C and any representation ϕ ∈

Hom(π, SE(2)) can be written as

ρϕ τϕ
ϕ= (4.4)
0 1
where ρϕ : π → U (1) and τϕ : π → C. Observe the following:
Lemma 4.6. The map ρϕ : π → U (1) is a group homomorphism. It only depends on

the conjugacy class of ϕ.
The proof is elementary.
Definition 4.7. The homomorphism ρϕ : π → U (1) is the character of the represen-

tation class ϕ ∈ Hom(π, SE(2)).
Remark. In the literature on group representations, the character χϕ : π → K of a

representation ϕ ∈ GLn (K) is classically defined to be the trace of the representation.
The two notions of characters are equivalent as shown by the formula
χϕ = Tr ϕ = 1 + ρϕ .
Any homomorphism ρ ∈ Hom(π, U (1)) defines a structure of π -module on C.

We will denote by Cρ this π -module, and we have:
Proposition 4.8. Given any pair of maps ρ : π → U (1) and τ : π → C, the

map ϕ : π → SE(2) given by (4.4) is a group homomorphism if and only if
ρ ∈ Hom(π, U (1)) and τ ∈ Z 1 (π, Cρ ).
Proof. Suppose that ϕ : π → SE(2) is given by (4.4). Then we have

ρ(γ1 γ2 ) τ (γ1 γ2 )
ϕ(γ1 γ2 ) =
0 1
and

ρ(γ1 ) τ (γ1 ) ρ(γ2 ) τ (γ2 )
ϕ(γ1 )ϕ(γ2 ) =
0 1 0 1

ρ(γ1 )ρ(γ2 ) τ (γ1 ) + ρ(γ1 )τ (γ2 )
= .
0 1
It follows that ϕ is a group homomorphism (i.e. ϕ(γ1 γ2 ) = ϕ(γ1 )ϕ(γ2 )) if and only if
ρ(γ1 γ2 ) = ρ(γ1 )ρ(γ2 )
and
τ (γ1 γ2 ) = τ (γ1 ) + ρ(γ1 )τ (γ2 ).
In other words ϕ is a homomorphism if and only if ρ : π → U (1) is a homomorphism
and τ is a 1-cocycle in the corresponding π -module Cρ . 2
This proposition says that the map from Hom(π, SE(2)) to the set
{(ρ, τ ) | ρ ∈ Hom(π, U (1)) and τ ∈ Z 1 (π, Cρ )}
given by ϕ → (ρϕ , τϕ ), is a bijection. In particular we have
Corollary 4.9. If π is the free group on s generators, then

Hom(π, SE(2)) Ts × Cs .
Proof. This follows from equation (4.1) and the fact that Hom(π, U (1)) = Ts . 2
4.4 Conjugation by similarities

Recall that a similarity in the plane is the composition of an isometry with a homothety.
Once we identify the euclidean plane with the complex line C, any similarity
g ∈ Sim(2) can be written as g(z) = a · z + b where a ∈ C∗ and b ∈ C. We thus
identify Sim(2) with the following subgroup of GL2 (C):

a b
Sim(2) = a, b ∈ C, a = 0 .
0 1
In particular we have
Sim(2) = R+ SE(2) = C∗ C.
Definition 4.10. Two representations ϕ1 , ϕ2 : π → SE(2) are similar if they are

conjugate modulo a similarity.
Proposition 4.11. Given a homomorphism ρ : π → U (1) and two cocycles τ1 , τ2 ∈

Z 1 (π, Cρ ), then the representations

ρ τ1 ρ τ2
ϕ1 = and ϕ2 = (4.5)
0 1 0 1
are similar if and only if there exists a complex number a ∈ C∗ such that
τ2 = aτ1 ∈ H 1 (π, Cρ ).
530 Marc Troyanov
Proof. The homomorphisms ϕ1 and ϕ2 are similar if and only if there exists

a b
g= ∈ Sim(2)
0 1
such that ϕ2 = gϕ1 g −1 , i.e.

ρ τ2 a b ρ τ1 1/a −b/a
=
0 1 0 1 0 1 0 1

ρ aτ1 + b − ρb
= .
0 1
This shows that
τ2 − aτ1 = b (1 − ρ) ∈ B 1 (π, Cρ ). 2
For any homomorphism ϕ : π → SE(2) and any λ ∈ R+ , we can define a new

homomorphism λ · ϕ : π → SE(2) by

ρ λτ
λ·ϕ = ,
0 1
This formula defines an action of the multiplicative group R+ on R(π, SE(2)), and
we denote the quotient by
SR(π, SE(2)) = R(π, SE(2))/R+ .
It follows directly from the definition that
SR(π, SE(2)) = Hom(π, SE(2))/ Sim(2).
where Sim(2) acts by conjugation on Hom(π, SE(2)).
Let us also define
SRreg = {[ϕ] = [ρ, τ ] ∈ SR(π, SE(2)) | ρϕ = id and τ = 0.}
Corollary 4.12. If π is a free group on s generators, then

SRreg (Ts \ {id}) × CPs−2 .
Proof. This is an immediate consequence of the previous results, in particular Propo-

sition 4.2 and 4.11. 2
5 Deformation theory
5.1 The moduli and Teichmüller spaces
The moduli space of g,n is the quotient of the space of conformal structures on g,n
by the pure diffeomorphism group. Let us be more specific: recall first that a conformal
structure is an equivalence class of smooth Riemannian metric m on = g,n , where

two Riemannian metrics m1 , m2 are equivalent if and only if there exists a function
u : → R such that
m2 = e2u m1 .
We denote by Met() the space of all smooth Riemannian metrics on endowed
with its natural C ∞ topology and by
Conf() = Met()/C ∞ ()
the space of conformal structures. We then define the moduli space of g,n to be the
quotient
Mg,n = Conf()/ PDiff + g,n .
A point μ ∈ Mg,n is concretely represented by a Riemannian metric m on , and
two Riemannian metrics m1 , m2 represent the same modulus point μ if and only if
there exists a smooth function u : → R and a diffeomorphism h ∈ PDiff + g,n such
∗
that m2 = e h (m1 ).
2u
A remark about the smoothness: By definition a point μ in the moduli space

is represented by a smooth metric. In particular, the punctures play no role in the
definition of the spaces Met() and Conf() (but they do in the definition of the
moduli space Mg,n ). However, since only the conformal class of the metric matters,
one may also represent μ by a singular metric m as long as this metric is conformally
equivalent to a smooth one. In particular we can (and will) represent a point in Mg,n
by a metric having conical singularities at the punctures of g,n , see Remark 2.21
The moduli space is a complicated object, and it is useful to also introduce the
simpler space obtained by considering isotopy classes of conformal structures on
g,n instead of isomorphism classes: this is the Teichmüller space defined as
Tg,n = Conf(g,n )/ Diff 0g,n .
Let us list some of the basic facts about these spaces:
(1) The Teichmüller space Tg,n is a real analytic variety in a natural way. If 3g−3+n >
0, then it is isomorphic to R6g−6+2n . This space has also a natural complex
structure.
(2) The pure mapping class group PMod+
g,n acts properly and discontinuously on Tg,n .
(3) The moduli space is the quotient Mg,n = Tg,n / PMod+
g,n . It is thus a good orbifold
of dimension 6g − 6 + 2n with fundamental group π1 (Mg,n ) = PModg,n .
(4) There exists a torsion free subgroup M0 ⊂ PMod+ g,n of finite index acting freely
on Tg,n . The quotient map Tg,n /M0 is a non singular analytic manifold which is
a finite cover of the orbifold Mg,n .
Statement (1) is explained in any textbook on Teichmüller theory such as [1].
Statement (2) was first proved by S. Kravetz [22], see also [1]. (3) is a consequence
of (1) and (2) and the last statement is discussed in [19, §5.4].
532 Marc Troyanov
5.2 The deformation space of piecewise flat metrics

Let us denote by Eg,n the set of all flat metrics on g,n with possible conical singular-
ities at the punctures (it is not empty since we have assumed 2g + n − 2 > 0). To any
flat metric m ∈ Eg,n , we associate the following basic invariants: Its conformal class
[m] ∈ Conf(), its area A = A(m) > 0 and the order βi > −1 of m at the point pi .
Theorem 5.1. The map

Eg,n → Conf() × Rn+ ,
m → ([m], (s1 , . . . , sn )),
where si = A (1 + βi ) > 0, is a bijection.
Proof. This is just a reformulation of Theorem 2.22. 2
Definition 5.2. We will endow the set Eg,n with the topology for which this map is a
homeomorphism.
A metric m2 ∈ Eg,n is said to be a deformation of the metric m1 ∈ Eg,n if the

two metrics differ by a homothety and an isotopy fixing the punctures, i.e. if there
exists h ∈ PDiff 0g,n and λ > 0 such that m2 = λ h∗ (m1 ). We denote by DE g,n the
deformation space of flat metrics on g,n with possible conical singularities at the
punctures:
DE g,n = Eg,n /(R+ × PDiff 0g,n ).
Corollary 5.3. This space is homeomorphic to R6g+3n−7 . In fact we have the following
canonical identification:
DE g,n = Tg,n × ,
where Tg,n is the Teichmüller space and ⊂ Rn is defined by

= β = (β1 , . . . βn ) ∈ Rn | βi > −1 and i βi = 2g − 2 .
the space of
Let us fix an element β = (β1 , . . . βn ) ∈ and denote by Eg,n (β)
singular flat metrics with a conical singularity of order βi at pi (i = 1, . . . , n). We
also introduce the corresponding deformation space: DE g,n (β) = Eg,n (β)/(
R+ ×
PDiff g,n ). The previous corollary gives us the identification
= Tg,n .
DE g,n (β)
5.3 Revisiting the developing map and the holonomy

Consider the punctured surface g,n = Û/ as in Section 3.2, and fix a flat metric
m0 with conical singularity of order βj at pj (j = 1, . . . , n). If f0 is a germ of an
isometry near a point z̃0 , to the euclidean plane (identified with C), then we obtain a
map f : U → C by analytic continuation from f0 . This map is a local isometry for
the metric m0 on U and the canonical metric on C (indeed, the set of points where
a map f between two flat surfaces is an isometry is easily seen to be both open and
closed). The map f extends by continuity to Û. The resulting map f : Û → C is the
developing map which we already met in Section 2.3. The associated holonomy is the
unique homomorphism ϕ : → SE(2) such that
f (γ u) = ϕ(γ )f (u).
Theorem 5.4. The following properties of the developing map and its holonomy are
satisfied:
(1) f : Û → C is surjective;
(2) f (γ u) = ϕ(γ )f (u) for all γ ∈ ;
(3) for any y ∈ P −1 (pi ) ⊂ ϒ, the isometry ϕ(cy ) ∈ SE(2) is a rotation of angle
θi = 2π(βi + 1);
(4) if βi is not an integer, then f (y) is the unique fixed point of ϕ(cy ).
Proof. The first assertion has been proved in Proposition 2.10, the second is the
definition of the holonomy and the last two assertions are contained in Proposi-
tion 2.15. 2
Corollary 5.5. If βi ∈ Z for any i = 1, . . . , n, then the restriction of f to the set ϒ

is determined by the holonomy.
Proof. Fix y ∈ ϒ and let cy ∈ be the corresponding group element given by the
map (3.5). Then f (y) ∈ C is the fixed point of the rotation ϕ(cy ) ∈ SE(2). This fixed
point is given explicitly by
τy
f (y) = , (5.1)
(1 − ρy )
where ρy ∈ U (1) is the rotation part and τy ∈ C is the translation part of ϕ(cy ). 2
Theorem 5.6. (A) There is a well-defined map

hol : DE g,n → SR(πg,n , SE(2)),
such that hol([m]) is the conjugacy class of the holonomy homomorphism
ϕm : πg,n → SE(2).
(B) The map hol : DE g,n → SR(πg,n , SE(2)) is continuous.
(C) There are natural actions of PModg,n on DE g,n and of POut(πg,n ) on
SR(π, SE(2)), and the map hol is -equivariant where is the Dehn–Nielsen–Baer
isomorphism (3.6).
(D) The map hol is locally injective.
534 Marc Troyanov
Remarks. 1.) The map hol : DE g,n → SR(πg,n , SE(2)) is called the holonomy
mapping.
2.) A more elaborate investigation would show that the holonomy mapping is in
fact real analytic, see [35]. The proof below is perhaps not optimal from the point of
view of rigour, but we have tried to emphasize the geometric point of view.
Proof of Theorem 5.6. (A) To any flat metric m on g,n with conical singularities at
the punctures, we have associated a holonomy homomorphism ϕm : πg,n → SE(2)
which depends on the choice of a developing map fm , but changing the developing
map does not affect the conjugacy class of ϕm (see Proposition 2.14). On the other
hand, it is clear that if two flat metrics m, m on g,n are similar, the associated
holonomies ϕm , ϕm are also similar. In short, to any deformation class of flat metric
[m] ∈ DE g,n with conical singularities on g,n we associate a well defined element
[ϕm ] = hol(m) ∈ SR(πg,n , SE(2)).
(B) The developing map fm is not uniquely associated to a flat metric m, but it
is well defined modulo SE(2) (two developing maps for the same metric differ by
postcomposition with an isometry). The SE(2) orbit of the developing map fm varies
continuously with the metric m and therefore it is also the case for the associated
holonomy class. Hence the map hol : DE g,n → SR(πg,n , SE(2)) is continuous.
(C) Any diffeomorphism h of g,n fixing the punctures acts on Eg,n by pulling
back the metric (m → h∗ m). If h is isotopic to the identity, it acts trivially on DE g,n ;
we thus have a well defined action of PModg,n on DE g,n .
Similarly, any automorphism of πg,n acts on Hom(πg,n , SE(2)), and inner auto-
morphisms act trivially on the two representation spaces R(πg,n , SE(2)) and
SR(πg,n , SE(2)). We thus have a natural action of POut(πg,n ) on these spaces. It is
clear from the construction of the isomorphism : PModg,n −− ∼→ POut(πg,n ) (see
the proof of Lemma 3.3) that the map hol is equivariant.
(D) To prove the local injectivity of hol, we consider two nearby flat metrics m, m
with conical singularities on g,n = Û/ and we assume that they have the same
holonomy ϕ. Since the holonomy around a conical singularity pi is a rotation of angle
θi = 2π(βi + 1), it is clear that both metrics m and m have the same singularity order
(the holonomy only controls the cone angle modulo 2π , but since m and m are nearby
metrics, they actually have equal cone angles).
It follows that both metrics are isometric near the singularities: we can thus find
an isotopy h1 of the surface such that m = h∗1 m near the singularities. Hence we
can simply assume without loss of generality that m = m near the singularities; it is
therefore possible to divide the surface in n + 1 parts
g,n = D ∪ E1 ∪ · · · ∪ En ,
where D ⊂ is a compact region and Ei is a neighbourhood of the puncture pi
such that m = m on Ei . We also assume that the Ei are pairwise disjoint disks. We
denote by Êi = P −1 (Ei ) ⊂ Û and D̂ = P −1 (D) ⊂ Û the lifts of Ei and D on the
universal branched cover P : Û → g,n . The set Ê = ∪i Êi ⊂ Û is a neighbourhood

of ϒ = P −1 ({punctures}).
Let fm and fm be the developing maps of m and m . By Corollary 5.5, the two
maps coincide on ϒ. Because the two metrics coincide on Ei , the map fm and fm
coincide up to a rotation on each component of Êi ; we can thus find a second isotopy
h2 of , which is a rotation near the punctures and is the identity on D and such that
fm ĥ2 = fm on Ê.
Replacing m with h∗2 m , we can thus assume that both developing maps coincide
on Ê.
To any point x ∈ U, we associate the set
(m, m , x) = fm−1 (fm (x)) ⊂ Û.
Since fm is a local diffeomorphism, (m, m , x) is a discrete set. It varies continuously
with m, m .
Claim. If m is close enough to m , then for any point x ∈ U , there exists a unique
point y = Q(x) ∈ (m, m , x) which is the nearest point for the hyperbolic distance.
The map x → Q(x) is -equivariant.
Indeed, if x ∈ Ê, then the claim is clear: since fm (x) = fm (x), we have Q(x) = x.
For any point, the claim is clear if m = m (and in this case Q(x) = x). For points in
D̂, and m close to m the claim follows from the compactness of D and the discreteness
and continuity of (x, m, m ).
For t ∈ [0, 1], we denote by Qt (x) the point on the hyperbolic segment [x, Q(x)]
such that dH (x, Qt (x)) = tdH (x, Q(x)) (observe that if x ∈ Ê, then Qt (x) = x for
any t). This is a -equivariant isotopy of U from the identity to Q. It extends as the
identity on ϒ.
We now define an isotopy ht : → by ht (x) = P (Qt (P −1 (x))). It is a well
defined isotopy such that h∗1 m = m, since we clearly have fm = fm Q.
We thus have proved that two metrics with the same holonomy are isotopic provided
they are close enough. In other words, the map hol is locally injective. 2
6 The main theorem

We are now in position to prove the main result. First recall the statement:
Theorem 6.1. Given a punctured surface g,n such that 2g + n − 2 > 0 and β ∈
such that no βi is an integer, there is a well defined group homomorphism
: PModg,n → G = Aut(T2g ) × PGL2g+n−2 C,
and a -equivariant local homeomorphism
H : Tg,n → = T2g × CP2g+n−3 .
536 Marc Troyanov
The theorem says that Mg,n = Tg,n / PModg,n is a good orbifold with a (G, )-
structure.
Proof. The group homomorphism is given by the Dehn–Nielsen–Baer isomorphism

and the map H is essentially given by the holonomy mapping of the previous theorem.
We divide the proof of the theorem in five steps:
Recall that holonomy splits in a rotation part ρm : π → U (1) (the character) and
a translation part τm . The character depends only on the conjugacy class of ϕm .
Step 1. τm is not identically zero.
Indeed, if τm ≡ 0, then the holonomy group ϕm (πg,n ) is a pure rotation group in
the plane. This is impossible by Proposition 2.13.
Step 2. There is a canonical isomorphism
Hom(πg,n , U (1)) Hom(πg,0 , U (1)) × Hom(π0,n , U (1)),
ρm → (ρ , ρ ).
Furthermore ρ ∈ Hom(π0,n , U (1)) is given by
ρ (ci ) = eθi ,
where ci is the homotopy class of a loop traveling once around the puncture pi and
θi = 2π(βi + 1) is the total angle at the cone point pi .
This splitting easily follows from the identity (3.1) and the fact that U (1) is abelian.
Let us now fix an element β = (β1 , . . . βn ) ∈ and set

SRβ (πg,n , SE(2)) = ϕ ∈ SR(πg,n , SE(2)) | ρ (ci ) = eθi , i = 1, . . . , n
and
reg
SR = SRreg ∩ SRβ .
β
Step 3. If at least one βi is not an integer, then we have

reg
SR (πg,n , SE(2)) = T2g × CP2g+n−3 .
β
reg
Indeed, it follows from Step 2 and the results of Section 4 that any ϕ ∈ SR
β
is characterized by ρ ∈ Hom(πg,0 , U (1)) T2g and the projective class of τ ∈
H 1 (πg,n , Cρ ) C2g+n−2 (because πg,n is isomorphic to the free group on s =
2g + n − 1 generators).
Step 4. The group POut(πg,n ) acts naturally on and thus we have a natural homo-
morphism : PModg,n → G = Aut(T2g ) × PGL2g+n−2 C.
This is clear from Step 3 and Part (C) of Theorem 5.6.
Step 5. The map H given by the composition:
∼
Tg,n −− −hol
→ DE g,n (β)
reg ∼
−→ SR −−→
β
is well defined, continuous, locally injective and -equivariant.

Indeed, the fact that no βi is integer, taken together with Step 1, implies that
hol : DE g,n → SR(π, SE(2)) maps DE g,n (β) into SRreg . The map H : Tg,n →
β
is therefore well defined. It follows from Theorem 5.6 that H is continuous, locally
injective and -equivariant.
It remains only to show that H is a local homeomorphism, but since Tg,n and
are both manifolds of dimension 6g − 6 + 2n, the conclusion follows from Brouwer’s
theorem on invariance of dimension. 2
6.1 The case of the sphere

Suppose that g = 0, i.e. is a sphere, choose n numbers (n ≥ 2) β1 , β2 , . . . , βn such
that 2 + i βi = 0, and denote by M the space of flat metrics on S 2 having n conical
singularities of order β1 , β2 , . . . , βn .
Such a metric m ∈ M can be uniformized as follows: identify with C ∪ ∞, and
write m as
n

m=C· |z − pi |2βi |dz|2 ,
i=1
where p1 , p2 , . . . , pn is the set of conical singularities and C is a positive constant
representing a dilation factor. It is easy to see that M is homeomorphic to the quotient
{(p1 , p2 , . . . , pn ) ∈ (C ∪ ∞)n : pi = pj if i = j }/ PSL2 (C),
M is thus a complex manifold of dimension n − 3, its fundamental group is the pure
braid group PBn .
Applying the main theorem, we obtain a representation
: PMod0;n = PBn → PGLn−2 (C)
and a equivariant, local homeomorphism
H : T0,n → CPn−3 .
In fact, a finer analysis shows that the image of is contained in PU(1, n − 3) ⊂
PGLn−2 (C). Furthermore, when the orders satisfy some arithmetical conditions, the
image of is a lattice in PU(1, n − 3):

Theorem 6.2. Assume that −1 < β1 , β2 , . . . , βn < 0, i βi = −2 and suppose that
βi + βj > −1 ⇒ (1 + βj + βi )−1 ∈ N, (6.1)
then (PBn ) is a lattice in PU(1, n − 3).
These lattices are quotients of the braid group. Some of them are non arithmetic.
This theorem was first proved by Schwartz (1873) for n = 4 and by Picard (1888)
for n = 5 in their study of the monodromy of the hypergeometric equations. It has
been generalized for any n by P. Deligne and G. Mostow in 1986, see [10]. These
538 Marc Troyanov
authors use the cohomology with coefficients in flat vector bundle on an algebraic
curve.
In the paper [30], W. Thurston obtained the same result by studying a deformation
space of piecewise flat triangulations on the sphere (this nice paper is a 1987 preprint
of W. Thurston, which has been rewritten and appeared in electronic form in 1998).
It is worthwhile to quote also the related papers [3], [13], [25], [21].
Our approach can be seen as a bridge between the approach of Thurston and that
of Deligne–Mostow.
Observe that the moduli space M = T0,n / PBn carries a complex hyperbolic metric
(depending upon the choice of the βi ’s). It is not complete as a Riemannian manifold
and it carries a natural completion M. Thurston shows that M is a complex hyperbolic
manifold with singularities of conical type. This cone-manifold has finite volume.
Furthermore, when the βi ’s satisfy the condition (6.1), then M is an orbifold. It is
thus possible to construct complete complex hyperbolic orbifolds M of finite volume.
References
[1] W. Abikoff, W. Abikoff, The real analytic theory of Teichmüller space. Lecture Notes in
Math. 820, Springer, Berlin 1980. 531
[2] J. Ambjorn, M. Carfora, A. Marzuoli, The geometry of dynamical triangulations. Lecture
Notes in Phys. New Ser. m Monogr. 50, Springer-Verlag, Berlin, 1997. 509
[3] C. Bavard and E. Ghys, Polygones du plan et polyèdres hyperboliques. Geom. Dedicata
43 (2) (1992) 207–224. 538
[4] J. S. Birman, Braids, links and mapping class groups. Ann. of Math. Stud. 82, Princeton
University Press, Princeton, N.J.; University of Tokyo Press, Tokyo 1974. 521
[5] B. H. Bowditch. Singular euclidean structures on surfaces. J. London Math. Soc. 44 (1991),
553–565. 509
[6] D. Burago, Yu. Burago and S. Ivanov, A course in metric geometry. Grad. Stud. Math. 33,
Amer. Math. Soc., Providence, RI, 2001 510
[7] R. D. Canary, D. B. A. Epstein, and P. Green, Notes on notes of Thurston. In Analytical
and geometric aspects of hyperbolic space (Coventry/Durham, 1984), London Math. Soc.
Lecture Note Ser. 111, Cambridge University Press, Cambridge 1987, 3–92. 509
[8] M. Carfora, C. Dappiaggi, and A. Marzuoli, The modular geometry of random Regge
triangulations. Classical Quantum Gravity 19 (20) (2002), 5195–5220. 509
[9] S. Choi, Geometric structures on orbifolds and holonomy representations. Geom. Dedicata
104 (2004), 161–199. 509
[10] P. Deligne and G. Mostow, Monodromy of hypergeometric functions and non lattice
integral monodromy. Inst. Hautes Études Sci. Publ. Math. 63 (1986), 5–89. 509, 537
[11] D. B. A. Epstein and R. C. Penner, Euclidean decompositions of noncompact hyperbolic
manifolds. J. Differential Geom. 27 (1) (1988), 67–80. 509
[12] A. Eremenko, Metrics of positive curvature with conic singularities on the sphere. Proc.
Amer. Math. Soc. 132 (11) (2004), 3349–3355 518
[13] F. Fillastre, From spaces of polygons to spaces of polyhedra following Bavard, Ghys and
Thurston. Paper on arXiv:math.MG/0308187 538
[14] W. Goldman, Geometric structures on manifolds and varieties of representations. In Ge-
ometry of group representations (Boulder, CO, 1987), Contemp. Math. 74, Amer. Math.
Soc., Providence, RI, 1988, 169–198. 509, 524
[15] M. Heins, On a class of conformal metrics. Nagoya Math. J. 21 (1962), 1–60. 518
[16] D. Hulin and M. Troyanov, Sur la courbure des surfaces ouvertes. C. R. Acad. Sci. Paris
310 (1990) 203-206. 520
[17] D. Hulin and M. Troyanov, Prescribing curvature on open surfaces. Math. Ann. 293 (1992),
277–315 . 518, 520
[18] C. Indermitte, T. M. Liebling, H. Clémençon and Marc Troyanov, Voronoi Diagrams on
Piecewise flat surfaces and an application to biological growth. Theoret. Comput. Sci. 263
(2001) 263–274. 517
[19] N. V. Ivanov, Mapping class groups. In Handbook of geometric topology (R. J. Daverman
and R. B. Sher, eds.) North-Holland, Amsterdam 2002, 523–633. 521, 522, 531
[20] S. Katok, Fuchsian groups. Chicago Lectures in Math., University of Chicago Press,
Chicago, IL, 1992. 522, 523
[21] S. Kojima, Complex hyperbolic cone structures on the configuration spaces. Rend. Istit.
Mat. Univ. Trieste 32 (2001), suppl. 1 149–163. 538
groups. Ann. Acad. Sci. Fenn. Ser. A1 278, (1959) 531
[23] R. McOwen, Point singularities and conformal metrics on Riemann surfaces. Proc. Amer.
Math. Soc. 103 (1988), 222–224. 518
[24] S. Morita, Introduction to mapping class groups of surfaces and related groups. In Hand-
book of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing House,
Zurich 2007, 353–386. 521
[25] G. Mostow, Braids, hypergeometric functions, and lattices. Bull. Amer. Math. Soc. 16 (2)
(1987), 225–246. 538
[26] É. Picard, De l’intégration de l’équation u = eu sur une surface de Riemann fermée.
J. Reine Angew. Math., 130 (1905), 243–258. 518
[27] I. Rivin, Combinatorial optimization in geometry. Adv. in Appl. Math. (1) 31 (2003),
242–271. 509, 517
[28] G. Shimura, Introduction to the arithmetic theory of automorphic functions. Publications
of the Mathematical Society of Japan 11, Iwanami Shoten Publishers, Tokyo; Princeton
University Press, Princeton, N.J., 1971. 519, 520
[29] W. Thurston, Three-dimensional geometry and topology. Vol. 1. Princeton University
Press, Princeton, NJ, 1997 509
[30] W. Thurston, Shapes of polyhedra and triangulations of the sphere, The Epstein birthday
schrift, Geom. Topol. Monogr. 1, Geom. Topol. Publ., Coventry 1998, 511–549 (elec-
tronic). 509, 517, 538
540 Marc Troyanov
[31] M. Troyanov, Les surfaces euclidiennes à singularités coniques. Enseign. Math. (2) 32
(1986), 79–94. 511, 517, 518
[32] M. Troyanov, Prescribing curvature on compact surfaces with conical singularities. Trans.
Amer. Math. Soc. 324 (1991), 793–821. 518
[33] M. Troyanov, Surfaces Riemanniennes à singularités simples. Proc. Sympos. Pure Math.
54 (Part 2) (1993), 619–628. 518
[34] M. Umehara and K. Yamada, Metrics of constant curvature 1 with three conical singular-
ities on 2-sphere. Illinois J. Math. 44 (2000), 72–94. 518
[35] W. A. Veech, Flat surfaces. Amer. J. Math. 115 (3) (1993), 589–689. 509, 534
[36] H. Zieschang, E. Vogt, H.-D. Coldeway, Surfaces and planar discontinuous groups. Lec-
ture Notes in Math. 835, Springer-Verlag, Berlin 1980. 521, 522
Chapter 13
Discrete Riemann surfaces
Christian Mercat
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
2 Discrete Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 542
2.1 Discrete Hodge theory . . . . . . . . . . . . . . . . . . . . . . . . . 542
2.2 Wedge product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
2.3 Energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
3 Period matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
3.1 Intersection number, on and on ♦diamondsuit . . . . . . . . . . . 548
3.2 Canonical dissection, fundamental polygon . . . . . . . . . . . . . . 548
3.3 Bilinear relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
3.4 Basis of harmonic forms, basis of holomorphic forms . . . . . . . . . 552
3.5 Period matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
4 Criticality and integrable system . . . . . . . . . . . . . . . . . . . . . . . 556
4.1 Criticality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 556
4.2 Period matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
4.3 Genus one case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
4.4 ∂¯ operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
4.5 Geometrical interpretation . . . . . . . . . . . . . . . . . . . . . . . 561
4.6 Quasiconformal maps . . . . . . . . . . . . . . . . . . . . . . . . . 562
4.7 Discrete exponential . . . . . . . . . . . . . . . . . . . . . . . . . . 563
4.8 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564
4.9 Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
4.10 Continuous limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
4.11 Cross-ratio preserving maps . . . . . . . . . . . . . . . . . . . . . . 569
4.12 Bäcklund transformation . . . . . . . . . . . . . . . . . . . . . . . . 571
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
1 Introduction
Riemann surfaces theory was a major achievement of XIXth century mathematics,
setting the framework where modern complex analysis bloomed. Nowadays, surfaces
are intensively used in computer science for numeric computations, ranging from
542 Christian Mercat
visualization to pattern recognition and approximation of partial differential equations.

A lot of these computations involve, at the continuous level, analytic functions. But
very few algorithms care about this specificity, although analytic functions form a
relatively small vector space among the space of functions, problems are usually
crudely discretized in a way that does not take advantage of analyticity.
The theory of discrete Riemann surfaces aims at filling this gap and setting the
theoretical framework in which the notion of discrete analyticity is set on solid grounds.
Most of the results in this chapter are a straightforward application of the continuous
theory [7], [14] together with the results in [10], [11], [12], to which we refer for details.
We define the discrete period matrix, which is twice as large as in the continuous case:
the periods of a holomorphic form on the graph and on its dual are in general different,
but the continuous limit theorem, given a refining sequence of critical maps, ensures
that they converge to the same value. The last section deals with criticality, defining the
framework in which the continuous limit theorem takes place, and integrability, defined
as a 3D-consistency of the equations ruling not only the linear discrete holomorphy
but a quadratic version as well.
2 Discrete Riemann surfaces

2.1 Discrete Hodge theory
We recall in this section basic definitions and results from [11] where the notion
of discrete Riemann surfaces was defined. We are interested in discrete oriented
surfaces given by a cellular decomposition ♦ of dimension two, where all faces are
quadrilaterals (a quad-graph [8], [2], [1]). We will assume that ♦ is bipartite with
black vertices connected to white vertices. The diagonals of ♦ connect vertices of
y
x x
Figure 1. The vertices and diagonals of a quadrilateral define a pair of dual edges.
Chapter 13. Discrete Riemann surfaces 543
the same type and define, up to homotopy and away from the boundary, two dual
cellular decompositions and ∗ : The edges in 1∗ are dual to edges in 1 , faces in
vn
v1
e∗
F∗
v
e
v2
1. The vertex F ∗ dual to a face F . 2. Dual edges. 3. The face v ∗ dual to a vertex v.
Figure 2. Duality.
2∗ are dual to vertices in 0 and vice-versa. This correspondance shows that there
is no lack of generality in considering quad-graphs. Their disjoint union is denoted
the double = ∗ . A discrete conformal structure on is a real positive
function ρ on the unoriented edges satisfying ρ(e∗ ) = 1/ρ(e). It defines a genuine
Riemann surface structure on the discrete surface: Choose a length δ and realize each
quadrilateral by a lozenge of side δ, whose diagonals have a length ratio given by ρ.
Gluing them together provides a flat Riemannian metric with conic singularities at the
vertices, hence a conformal structure [16]. Such a discrete conformal structure leads
to a straightforward discrete version of the Cauchy–Riemann equation. A function on
the vertices of ♦ is discrete holomorphic iff for every quadrilateral (x, y, x , y ) ∈ ♦2 ,

f (y ) − f (y) = i ρ(x, x ) f (x ) − f (x) . (2.1)
We recall elements of de-Rham cohomology, doubled in our context: The complex

of chains C() = C0 () ⊕ C1 () ⊕ C2 () is the vector space (over R the field of
reals) spanned by vertices, edges and faces. It is equipped with a boundary operator
∂ : Ck () → Ck−1 (), null on vertices and fulfilling ∂ 2 = 0. The kernel Ker ∂ =:
Z• () of the boundary operator is the set of closed chains or cycles. The image of the
boundary operator are the exact chains. The dual spaces of forms are called cochains,
C k () := Hom(Ck (), R). Coupling is denoted by functional and integral notation.
The dual of the boundary operator is called the coboundary operator d : C k () →
C k+1 (), defined by Stokes formula:

df := f ∂(x, x ) = f (x ) − f (x), dα := α.
(x,x ) F ∂F
A cocycle is a closed cochain and we note α ∈ Z k ().

These spaces are equipped with the canonical scalar product, weighed according
to ρ on edges and averaged on the graph and its dual:

(α, β)R := 21 ρ(e) α β . (2.2)
e∈1 e e
Duality of complexes allows us to define a Hodge operator ∗ on forms by
∗ : C k () → C 2−k ();

C 0 () f
→ ∗f : ∗f := f (F ∗ ),
F
(2.3)
C () α
→ ∗α : ∗α := −ρ(e∗ )
1
α,
e e∗

C 2 () ω
→ ∗ω : (∗ω)(x) := ω.
x∗
It fulfills ∗2 = (− IdC k )k . The endomorphism

:= −d ∗ d ∗ − ∗ d ∗ d is the usual
discrete Laplacian: It decomposes onto the two independent graphs and its formula on
a function at a vertex x ∈ 0 with neighbours x1 , . . . , xV ∈ 0 is the usual weighted
averaged difference:
V

(
(f )) (x) = ρ(x, xk ) (f (x) − f (xk )) .
k=1
The space of harmonic forms is defined as its kernel ∂∗ = Ker

.
The Hodge star and the Laplacian are real operators. Since ∗2 = − Id on functions,
it is natural to consider them on complexified cochains, equipped with the hermitian
version of the scalar product (2.2). The discrete holomorphic forms are special com-
plex harmonic forms: a 1-form
α ∈ C 1 () is holomorphic iff dα = 0 and ∗ α = −iα, (2.4)
that is to say if it is closed and of type (1, 0). Let d , resp. d the compositions of the
exterior derivative with the projection on the space of (1, 0), resp. (0, 1)-forms,
d := π(1,0) d : C 0 () → C 1 (), d := d π(1,0) : C 1 () → C 2 ()
and similarly for d and π(0,1) .

Definition (2.4) is equivalent to d α = 0. We will note α ∈ 1 (). A function
f : 0 → C is holomorphic iff df is holomorphic, which is equivalent to (2.1) and
we note f ∈ 0 ().
In the compact case, − ∗ d ∗ is the adjoint d ∗ of the coboundary operator d and the
Hodge theorem orthogonally decomposes forms into exact, coexact and harmonic,
C k () = Im d ⊕⊥ Im d ∗ ⊕⊥ Ker
,
harmonic forms are the closed and co-closed ones, and harmonic 1-form are the
orthogonal sum of holomorphic and anti-holomorphic ones:
Ker
= Ker d ∩ Ker d ∗ = Ker d ⊕⊥ Ker d .
2.2 Wedge product

We consider similarly C • (♦) cochains on the graph ♦. We construct a wedge product
on ♦ such that
• the canonical weighted hermitian scalar product reads as expected

(α, β) = α ∧ ∗β̄,
• and the coboundary operator d♦ on ♦, is a derivation for this product ∧ : C k (♦)×

C l (♦) → C k+l (♦).
It is defined by the following formulae, for f, g ∈ C 0 (♦), α, β ∈ C 1 (♦) and ω ∈
C 2 (♦):
(f · g)(x) := f (x) · g(x) for x ∈ ♦0 , (2.5)

f (x) + f (y)
f · α := α for (x, y) ∈ ♦1 , (2.6)
2
(x,y) (x,y)

4
α ∧ β := 1
4 α β− α β, (2.7)
(x1 ,x2 ,x3 ,x4 ) k=1(x ,x ) (x ,x ) (xk+1 ,xk ) (xk ,xk−1 )
k−1 k k k+1

f (x1 )+f (x2 )+f (x3 )+f (x4 )
f · ω := ω for (x1 , x2 , x3 , x4 ) ∈ ♦2 . (2.8)
4
(x1 ,x2 ,x3 ,x4 ) (x1 ,x2 ,x3 ,x4 )
A form on ♦ can be averaged into a form on : This map A from C • (♦) to C • ()
is the identity for functions and defined by the following formulae for 1 and 2-forms:

A(α♦ ) := 21
+ + + α♦ , (2.9)
(x,x ) (x,y) (y,x ) (x,y ) (y ,x )
d

A(ω♦ ) := 1
2 ω♦ , (2.10)
x∗ k=1 (xk ,yk ,x,yk−1 )
where notations are made clear in Figure 3. The map A is neither injective nor
surjective in the non simply-connected case, so we can neither define a Hodge star
on ♦ nor a wedge product on . Its kernel on 1-forms is the line spanned by the
coboundary of the bi-constant: Ker(A) = Span(d♦ ε), where ε takes the value +1
on the vertices of and −1 on the vertices of ∗ . But d A = A d♦ so it carries
x1
y yd
y1 xd
x2
x x x
y2
y
(2.9) (2.10)
Figure 3. Notations.
cocycles on ♦ to cocycles on . Its image are these cocycles of verifying that their
holonomies along cycles of only depend on their homology on the combinatorial
surface.
Given a 1-cocycle μ ∈ Z 1 () with such a property, a corresponding 1-cocycle
ν ∈ Z 1 (♦) is built in the following way: Choose an edge (x0 , y0 ) ∈ ♦1 ; for an edge
(x, y) ∈ ♦1 with x and x0 on the same leaf of , choose two paths λx,x0 and λy0 ,y on
the double graph , from x to x0 and y0 to y respectively, and define

ν := μ+ μ− μ (2.11)
(x,y) λx,x0 λy0 ,y [γ ]
where [γ ] = [λx,x0 + (x0 , y0 ) + λy0 ,y + (y, x)] is the class of the full cycle in the
homology of the surface. Changing the base points change μ by a multiple of d♦ ε.
It follows in the compact case that the dimensions of the harmonic forms on ♦ (the
kernel of
A) modulo dε, as well as the harmonic forms on with same holonomies on
the graph and on its dual, are twice the genus of the surface, as expected. Unfortunately,
the space Im A = H ⊥ ⊕ Im d is not stable by the Hodge star ∗. We could nevertheless
define holomorphic 1-forms on ♦ but their dimension would be much smaller than in
the continuous theory, namely the genus of the surface. Criticality, defined in Section 4
provides conditions which ensure that the space ∗ Im A is “close” to Im A.
We construct an heterogeneous wedge product for 1-forms: with α, β ∈ C 1 (),
define α ∧ β ∈ C 1 (♦) by

α ∧ β := 21 α β+ α β . (2.12)
(x,y,x ,y ) (x,x ) (y,y ) (y,y ) (x ,x)
It verifies A(α♦ ) ∧ A(β♦ ) = α♦ ∧ β♦ , the first wedge product being between

1-forms on and the second between forms on ♦. The usual scalar product on
compactly supported forms on reads as expected:

(α, β) = 2
1
ρ(e) α β̄ = α ∧ ∗β̄. (2.13)
e∈1 e e ♦2
2.3 Energies
The L2 norm of the 1-form df , called the Dirichlet energy of the function f , is the
mean of the usual Dirichlet energies on each independent graph:
2
ED (f ) := 1 df 2 = 1 (df, df ) = 1
2 2 4 ρ(x, x ) f (x ) − f (x) (2.14)
(x,x )∈1
ED (f | ) + ED (f | ∗ )
= .
2
Harmonic maps minimize this energy among functions fulfilling certain boundary
conditions.
The conformal energy of a map measures its conformality defect, it is null on
holomorphic functions:
EC (f ) := 41 df − i ∗ df 2 . (2.15)
It is related to the Dirichlet energy through the same formula as in the continuous case:
EC (f ) = 4 (df − i ∗ df, df − i ∗ df )
1
= 4 df + 4 −i ∗ df + 2 Re(df,

1 2 1 2 1
−i ∗ df )

= 21 df 2 + 1
2 Im df ∧ df
♦2
= ED (f ) − A(f ) (2.16)
where the area of the image of the application f in the complex plane has the same
formula

i
A(f ) = 2 df ∧ df (2.17)
♦2

as in the continuous case since, for a face (x, y, x , y ) ∈ ♦2 , the algebraic area of the
oriented quadrilateral f (x), f (x ), f (y), f (y ) is given by

df ∧ df = i Im (f (x ) − f (x))(f (y ) − f (y))
(x,y,x ,y )

= −2iA f (x), f (x ), f (y), f (y ) .
3 Period matrix
We use the convention of Farkas and Kra [7], chapter III, to which we refer for details.
Consider (♦, ρ) a discrete compact Riemann surface.
3.1 Intersection number, on and on ♦
For a given simple (real) cycle C ∈ Z1 (), we construct a harmonic 1-form ηC such
that A ηC counts the algebraic number of times A contains an edge dual to an edge of
C: It is the solution of a Neumann problem on the surface cut open along C (see [10]
for details). It is the minimum of the Dirichlet energy, which is convex on the space
of closed 1-forms having the desired holonomies. It follows from standard homology
technique that ηC depends only on the homology class of C (all the cycles which
differ from C by an exact cycle ∂A) and can be extended linearly to all cycles as
η• : H1 () → C 1 (); it fulfills, for a closed form θ ,

θ= ηC ∧ θ, (3.1)
C ♦
and a basis of the homology provides a dual basis of harmonic forms on . Beware
that if the cycle C ∈ Z1 () is purely on , then this form ηC | = 0 is null on .
The intersection number between two cycles A, B ∈ Z1 () is defined as

A · B := ηA ∧ ηB . (3.2)
♦
It is obviously linear and antisymmetric, it is an integer number for integer cycles.

Let us stress again that the intersection of a cycle on with another cycle on is
always null. A cycle C ∈ Z1 (♦) defines a pair of cycles on each graph C ∈ Z1 (),
C ∗ ∈ Z1 ( ∗ ) which are homologous to C on the surface, composed of portions of
the boundary of the faces on dual to the vertices of C. They are uniquely defined if
we require that they lie “to the left” of C as shown in Figure 4. By the procedure (2.11)
applied to ηC + ηC∗ , we construct a 1-cocycle ηC ∈ Z 1 (♦) unique up to dε, and
since dε ∧ θ = 0 for all θ, Equation (3.2) defines an intersection number on Z1 (♦).
Unlike the intersection number on , this one has all the usual expected properties.
In particular Equation (3.2) holds for A, B ∈ Z1 (♦).
3.2 Canonical dissection, fundamental polygon
The complex ♦ being connected, consider a maximal tree T ⊂ ♦1 , that is to say T is

simply connected and every edge added to T forms a cycle. A canonical dissection or
cut-system ℵ of the genus g discrete Riemann surface ♦ is given by a set of oriented
edges (ek )1≤k≤2g such that the cycles ℵ ⊂ (T ∪ ek ) form a basis of the homology
Figure 4. A path C on ♦ defines a pair of paths C and C ∗ on its left.
group H1 (♦) verifying, for 1 ≤ k, ≤ g

ℵk · ℵ = 0, ℵk+g · ℵ+g = 0, ℵk · ℵ+g = δk, . (3.3)
They actually form a basis of the fundamental group π1 (♦) and the defining relation
among them is (noted multiplicatively)
g

ℵk ℵk+g ℵ−1 −1
k ℵk+g = 1. (3.4)
k=1
The construction of such a basis is standard and we shall not repeat the procedure.
What is less standard is the interpretation of Equation (3.4) in terms of the boundary
of a fundamental domain, discretization introduces some subtleties (that can safely be
skipped in first instance). We end up with the familiar 2g × 2g intersection numbers
matrix on ♦.
Considering T ∪ ek as a rooted graph, we can prune it of all its pending branches,
leaving a simple closed loop ℵ− k , attached to the origin O by a simple path λk (see
Figure 5), yielding the cycle ℵk . These three cycles are deformation retract of one
another, ℵ− k ⊂ ℵk ⊂ T ∪ ek hence are equal in homology.
In the continuous case, a basis of the homology can be realized by 2g simple arcs,
transverse to one another and meeting only at the base point. It defines an isometric
1. 2. 3.
Figure 5. 1. A maximal rooted tree in a quadrilateral decomposition of the torus. 2. An
additional edge defines a rooted cycle ℵ1 , pruned of its dangling trees. 3. Its un-rooted version,
the simple loop ℵ−
1.
model of the surface as a fundamental domain homeomorphic to a disc and bordered

by 4g arcs to identify pairwise. In the discrete case, by definition, the set ♦ \ ℵ of the
cellular complex minus the edges taking part into the cycles basis is homeomorphic
to a disc hence the surface is realized as a polygonal fundamental domain M whose
boundary edges are identified pairwise.
But it is sometimes impossible to choose a basis of the homology verifying (3.3)
by simple discrete cycles which are transverse to one another. For instance, if the path
λk is not empty, the cycle ℵk is not even simple. Moreover, some edges may belong
to several cycles. In this case, the edges on the boundary of this fundamental polygon
can not be assigned a unique element of the basis or its inverse, and therefore can not
be grouped into only 4g continuous paths to identify pairwise but more than 4g.
In fact, the information contained into the basis ℵ is more than simply this polygon,
the set of edges composing the concatenated cycle
(ℵ1 , ℵg+1 , ℵ−1 −1 −1 −1

1 , ℵg+1 , ℵ2 , . . . , ℵg , ℵ2g ) (3.5)
encodes a cellular complex M+ which is not a combinatorial surface and consists of

the fundamental polygon M plus some dangling trees, corresponding to the edges
which belong to more than one cycle or participate more than once in a cycle (the
paths λk ), as exemplified in Figure 6. By construction, the edge ek belongs to the
cycle ℵk only, hence these trees are in fact without branches, simple paths whose
only leaf is the base point O. To retrieve the surface, the edges of this structure M+
are identified group-wise, an edge participating k times in cycles will have [k/2] + 2
representatives to identify together, two on the fundamental polygon and the rest as
edges of dangling trees.
Eliminating repetition, that is to say looking at (3.5) not as a sequence of edges
but as a simplified cycle (or a simplified word in edges), thins M+ into M, pruning
away the dangling paths. The fundamental polygon boundary loses its structure as 4g
arcs to be identified pairwise, in general a basis cycle will be disconnected around the
fundamental domain and a given edge can not be assigned to a particular cycle.
1. 2. 3.
Figure 6. Three different fundamental polygons of a decomposition of the torus (g = 1) by three

quadrilaterals: 1. The standard fundamental domain where the 4g paths are not adapted to ♦.
2. M+ is composed of edges of ♦ composing 4g arcs (which may have portions in common) to
identify pairwise, each edge corresponds to an element of the basis ℵ or its inverse, except for
edges of “dangling trees” which are associated with two such elements. 3. M is composed of
edges of ♦ composing more than 4g arcs to identify pairwise, there is no correspondence with
a basis of cycles.
This peculiarity gives a more complex yet well-defined meaning to the contour
integral formula for a 1-form θ defined on the boundary edges of M+ ,
2g

θ= θ+ θ. (3.6)
∂M k=1 ℵk ℵ−1
k
∗
This basis gives rise to cycles ℵ and ℵ whose homology classes form a basis of
the group for each respective graph, that we compose into ℵ defined by
∗
ℵ
k = ℵk , ℵ
k+g = ℵk , (3.7)
∗
ℵ
k+2g = ℵk+g , ℵ
k+3g = ℵk+g ,
for 1 ≤ k ≤ g so that while the intersection numbers matrix on ♦ is given by the
2g × 2g matrix

0 I
(ℵk · ℵ )k, = , (3.8)
−I 0
the intersection numbers matrix on is the 4g × 4g matrix with the same structure
∗ ∗
⎛ ⎞
0 0 I 0
(ℵ
k · ℵ )k, =⎜
⎜ 0 0 0 I⎟⎟ ∗ . (3.9)
⎝−I 0 0 0⎠ ∗
0 −I 0 0
3.3 Bilinear relations

Proposition 3.1. Given a canonical dissection ℵ, for two closed forms θ, θ ∈ Z 1 (♦),
g

θ ∧ θ = θ θ − θ θ ; (3.10)
♦ j =1 ℵj ℵj +g ℵj +g ℵj
for two closed forms θ, θ ∈ Z 1 (),

2g
θ ∧ θ = θ θ − θ θ . (3.11)
♦ j =1 ℵj ℵj+2g ℵj+2g ℵj
Proof. Each side is bilinear and depends only on the cohomology classes of the forms.
Decompose the forms onto the cohomology basis (αk ) (see below). On , use Equa-
tion (3.15) for the LHS and the duality property Equation (3.14) for the RHS. On ♦,
use their counterparts.
Notice that for a harmonic form θ ∈ H 1 (), the form ∗θ is closed as well, therefore
its norm is given by its boundary values on the fundamental polygon:

2g
θ ∈ H () ⇒ θ =
1 2
θ ∗θ̄ − θ ∗θ̄ . (3.12)
j =1 ℵj ℵj +2g ℵj +2g ℵj
3.4 Basis of harmonic forms, basis of holomorphic forms

We define α , the basis of real harmonic 1-forms, dual to the homology basis ℵ , as
described in Section 3.1,
αk := ηℵ and
k+2g

(3.13)
αk+2g := −ηℵ for 1 ≤ k ≤ 2g
k
which verify

α = δk, ,
ℵ
k
(3.14)
α+2g = δk, ,
ℵ
k+2g
and dually, the intersection matrix elements are given by

ℵ
k · ℵ
= αk ∧ α = (αk , − ∗ α ). (3.15)
♦
On ♦, the elements αk♦ := ηℵk+g and αk+g

♦
:= −ηℵk for 1 ≤ k ≤ g, defined up to
dε, verify A(αk♦ ) = αk + αk+g
, A(α ♦ ) = α
k+g

k+2g + αk+3g and form a basis of the
cohomology on ♦ dual to ℵ as well,

αk♦ := ηℵ♦ and
k+g
(3.16)
♦
αk+g := −ηℵ♦ for 1 ≤ k ≤ g,
k
they fulfill the first identity in Equation (3.15) but the second is meaningless in general
since ∗ can not be defined on ♦. We will drop the mention when no confusion is
possible.
Proposition 3.2. The matrix of inner products on ,

+ ℵk+2g ∗α , 1 ≤ k ≤ 2g, A D
(αk , α )k, = αk ∧ ∗ᾱ = =: (3.17)
♦ − ℵk−2g ∗α , 2g < k ≤ 4g. B C
is a real symmetric positive definite matrix.
Proof. It is real because the forms are real, and symmetric because the scalar prod-
uct (2.13) is skew symmetric. Definition Equation (3.13) and Equation (3.1) lead to
the integral formulae. Positivity follows from the bilinear relation Equation (3.11):
4g 4g
for θ = k=1 ξk αk , with ξk ∈ C, k=1 |ξk |2 > 0,

2g

θ 2 = θ ∗θ̄ − θ ∗θ̄
j =1 ℵj ℵ2g+j ℵ2g+j ℵ2j

4g
2g
(3.18)
= ξk ξ̄ αk ∗α − αk ∗α
k,=1 j =1 ℵj ℵ2g+j ℵ2g+j ℵ2j

4g
= ξk ξ̄ (αk , α ) > 0.
k,=1
The form αk is supported by only one of the two graphs or ∗ , the form ∗αk is
supported by the other one, and the wedge product θ ∧ θ = 0 is null for two 1-forms
supported by the same graph. Therefore the matrices A and C are g ×g-block diagonal
and B is anti-diagonal.

A 0 0 B ∗ , C ∗ 0
A= , B= , C= . (3.19)
0 A ∗ B, ∗ 0 0 C
The matrices of intersection numbers (3.9) and of inner products differ only by the
Hodge star ∗. Because ∗ preserves harmonic forms and the inner product, we get its
matrix representation in the basis α,

−D A
∗= (3.20)
−C B
and because ∗2 = −1,

B2 − C · A + I = 0 (3.21)
A · B = tB · A (3.22)
C · t B = B · C. (3.23)
On ♦, while the Hodge star ∗ can not be defined, we can obviously consider the
following positive scalar product on the classes of closed forms modulo dε, to which
the set (αk♦ ) belong:

(α ♦ , β ♦ ) := A(α ♦ ), A(β ♦ )
⎛ ⎞⎛ ⎞ ⎛ β̄
⎞
t (x ,y ) α ∗ +ρ−ρ ∗ −ρ−ρ ∗ −ρ+ρ ∗
⎜ ⎜
+ρ+ρ (x ,y )
⎟ ⎟
⎜ (y ,x ) α ⎟ ⎜⎜+ρ−ρ ∗ +ρ+ρ ∗ −ρ+ρ ∗ −ρ−ρ ∗ ⎟ ⎜ (y ,x ) β̄ ⎟
⎟
= ⎜ ⎟·⎝ · ⎜ ⎟.
⎝ (x ,y ) α ⎠ −ρ−ρ ∗ −ρ+ρ ∗ +ρ+ρ ∗ +ρ−ρ ∗ ⎠ ⎝(x ,y ) β̄ ⎠
(x,y,x ,y )∈♦2
(y ,x ) α
−ρ+ρ ∗ −ρ−ρ ∗ +ρ−ρ ∗ +ρ+ρ ∗
ρ=ρ(x,x ), (y ,x ) β̄
ρ ∗ =ρ(y,y )
and it yields
tB

A + A ∗ + t B ∗
∗
(αk♦ , α♦ )k, = , (3.24)
B ∗ + B ∗ C + C ∗
which, in general, can not be understood as the periods of a set of forms on ♦ along
the basis ℵ.
Let us decompose the space of harmonic forms into two orthogonal supplements,
H 1 () = H1 ⊕⊥ H⊥1 (3.25)
where the first vector space are the harmonic forms whose holonomies on one graph
are equal to their holonomies on the dual, that is to say
H1 := Vect(αk + αk+g , 1 ≤ k ≤ g or 2g < k ≤ 3g). (3.26)
Definition (3.13) and Equation (3.1) imply that
H⊥1 = Vect(∗αk − ∗αk+g , 1 ≤ k ≤ g or 2g < k ≤ 3g). (3.27)
These elements in the basis (αk + αk+g , αk − αk+g ), for 1 ≤ k ≤ g and 2g < k ≤ 3g,
are represented by the following invertible matrix:
⎛ ⎞
I 0 t B ∗ − t B ∗ A − A ∗
⎜0 I C − C ∗ B ∗ − B ∗ ⎟
⎜ ⎟. (3.28)
⎝0 0 t B ∗ + t B ∗ A + A ∗ ⎠
0 0 C + C ∗ B ∗ + B ∗
It implies in particular that the lower right g × g block is invertible, therefore so is
Equation (3.24).
3.5 Period matrix

Proposition 3.3. The matrix = C −1 · (i − B) is the period matrix of the basis of
holomorphic forms

2g
−1
ζk := (i − ∗) Ck, α+2g (3.29)
=1
in the canonical dissection ℵ, that is to say

δk, for 1 ≤ k ≤ 2g,
ζ = (3.30)
ℵk k−2g, for 2g < k ≤ 4g,
and is symmetric, with a positive definite imaginary part.
The proof is essentially the same as in the continuous case [7] and we include it
for completeness.
Proof of Proposition 3.3. Let ωj := αj + i ∗ αj for 1 ≤ j ≤ 4g. These holomorphic

forms fulfill
Pk,j := 21 (ωk , ωj ) = (αk , αj ) + i (αk , − ∗ αj )

−i ℵj +2g ωk , 1 ≤ j ≤ 2g, (3.31)
=
i ℵj −2g ωk , 2g < j ≤ 4g.
P is the period matrix of the forms (ω) in the homology basis ℵ. The first 2g forms
(ωj )1≤j ≤2g are a basis of holomorphic forms. It has the right dimension and they are
linearly independent:

2g
(λj + iμj )(αj + i ∗ αj )
j =1
2g

2g
2g
= (λj + μk Bj,k ) αj + μk Cj,k α2g+j (3.32)
j =1 k=1 k=1
2g

2g
2g
+i (μj + λk Bj,k ) αj + λk Cj,k α2g+j
j =1 k=1 k=1
is null, for λ, μ ∈ R only when λ = μ = 0 because C is positive definite. Similarly for

the last 2g forms. The change of basis i C −1 on them provides the basis of holomorphic
forms (ζ ). The last 2g rows of P is the 2g × 4g matrix (B − i I, C) hence the periods
of (ζ ) in ℵ are given by (I, ).
The first identity in Equation (3.30) uniquely defines the basis ζ and a holomorphic
1-form is completely determined by whether its periods on the first 2g cycles of ℵ, or
their real parts on the whole set.
Notice that because C is g × g block diagonal and B is anti-diagonal, is de-

composed into four g × g blocks, the two diagonal matrices form i C −1 and are pure
imaginary, the other two form −C −1 · B and are real.

i∗ r i C−1
∗ −C−1
∗ · B ∗ ,
= = . (3.33)
r∗ i −C−1 · B, ∗ i C−1
Therefore the holomorphic forms ζk are real on one graph and pure imaginary on
its dual,
1 ≤ k ≤ g ⇒ ζk ∈ CR1 () ⊕ i CR1 ( ∗ )
(3.34)
g < k ≤ 2g ⇒ ζk ∈ CR1 ( ∗ ) ⊕ i CR1 ().
We will call
= r + i∗ (3.35)
the period matrix on the graph the sum of the real periods of ζk , 1 ≤ k ≤ g, on
, with the associated pure imaginary periods on the dual ∗ , and similarly for ζk ,
g < k ≤ 2g, the period matrix on ∗ .
It is natural to ask how close and ∗ are from one another, and whether their
mean can be given an interpretation. Criticality [10], [11], that we will define in the
next Section 4 answers partially the issue.
4 Criticality and integrable system

The theory of discrete holomorphy can be viewed as the simplest (it is linear) of a
series of integrable theories [1]. We will present its quadratic counterpart, which
leads to another version of discrete analytic functions, based on circle patterns. Along
the way, we will see how discrete exponentials and discrete polynomials emerge due
to integrable systems theory pieces of technology named the Bäcklund or Darboux
transform [1].
4.1 Criticality
Criticality clarifies how purely combinatorial data can be connected to an underlying
geometry and gives a meaning to approximation theorems.
Definition 4.1. A discrete conformal structure (♦, ρ) of a simply connected set is

critical if there exists a discrete holomorphic map Z such that the quadrilateral faces
♦2 can be simultaneously embedded into rhombi by Z in the complex plane. We
call Z a critical map. The non simply connected case is covered using an atlas of
overlapping simply connected patches and critical maps.
A discrete holomorphic function Z only maps vertices to the complex plane. We

extend it linearly to quadrilateral edges and harmonically inside the quadrilateral faces.
Because of the Gauss–Bonnet theorem, it is not possible to globally embed a compact
surface into the plane, therefore we allow for an atlas of local critical maps with a
finite number of fixed local conic singularities. When a continuous limit is taken, their
number, angle and position should not change, and the theorem of isolated singularities
helps us wipe them out as inessential.
It is a simple calculation to check that if Z is a critical map, any discrete holomorphic
function f ∈ (♦) gives rise through (2.6) to a holomorphic 1-form f dZ.
For a holomorphic function f , the equality f dZ ≡ 0 is equivalent to f = λ ε for
some λ ∈ C with ε the bi-constant ε| ≡ +1, ε|∗ ≡ −1.
Following Duffin [5], [6], we introduce the
Definition 4.2. For a holomorphic function f , define on a flat simply connected map
U the holomorphic functions f † , the dual of f , and f , the derivative of f , by the
following formulae:
f † (z) := ε(z) f¯(z), (4.1)
where f¯ denotes the complex conjugate, ε = ±1 is the bi-constant, and
z †
4
f (z) := 2 f dZ + λ ε,
†
(4.2)
δ O
defined up to ε.
It is an immediate calculation [11] to check the following
Proposition 4.3. The derivative f fulfills

d f = f dZ. (4.3)
4.2 Period matrix

Criticality allows to state how close the two period matrices of a refining sequence of
critical maps are:
Theorem 4.1. In the genus one critical case, the period matrices and ∗ are
equal to the period matrix of the underlying surface . For higher genus, given
a refining sequence (♦k , ρk ) of critical maps of , the discrete period matrices k
and ∗k converge to the period matrix .
Proof. The genus one case is postponed to Section 4.3. We proved [11] that, given
a refining sequence of critical maps, any holomorphic function can be approximated
by a sequence of discrete holomorphic functions. Taking the real parts, this implies
as well that any harmonic function can be approximated by discrete harmonic func-
tions. In particular, the discrete solutions fk to a Dirichlet or Neumann problem on
a simply connected set converge to the continuous solution f because the latter can
be approximated by discrete harmonic functions gk and the difference fk − gk being
harmonic and small on the boundary, converges to zero. In particular, each form in the
basis (α♦ ), provides a solution to the Neumann problem Equation (3.16) and a similar
procedure, detailed afterwards, define a converging sequence of forms ζ♦ , yielding
the result.
We can try to replicate the work done on on the graph ♦. A problem is that
A + A and C + C need not be positive definite. Moreover, the Hodge star ∗
does not preserve the space (A(αk♦ )) of harmonic forms with equal holonomies on the
graph and on its dual, so we can not define the analogue of α + i ∗ α on ♦. We first
investigate what happens when we can partially define these analogues:
Assume that for 2g < k ≤ 3g, the holonomies of ∗αk on are equal to the
holonomies of ∗αk+g on ∗ , that is to say C = C ∗ =: 21 C♦ and D ∗ = D ∗ =:
2 D♦ . It implies that the transposes fulfill B = B =: 2 B♦ as well. We can
1 ∗ ∗
1
♦ ♦
then define βk−g ∈ Z 1 (♦) such that A(βk−g ) = ∗αk+g , uniquely up to dε. The last g
t
columns (B♦ , C♦ ) of the matrix of scalar product Equation (3.24) are related to their
periods in the homology basis ℵ♦ in a way similar to Equation (3.17). By the same
reasoning as before, the forms
g

ζk♦ = −1
C♦ k,
α ♦
+g − iβ ♦
+g , 1 ≤ k ≤ g (4.4)
=1
ζ +ζ
verify A(ζk♦ ) = k 2 k+g and have periods on ℵ♦ given by the identity for the first g
cycles and the following g × g matrix, mean of the period matrices on the graph and
on its dual:
−1 + ∗
♦ = C♦ (i − B♦ ) = . (4.5)
2
The same reasoning applies when the periods of the forms ∗αk on the graph and on
its dual are not equal but close to one another. In the context of refining sequences, we
said that the basis (α♦ ), converges to the continuous basis of harmonic forms defined
by the same Neumann problem Equation (3.16). Therefore
C − C ∗ = o(1), B ∗ − B ∗ = o(1). (4.6)
A harmonic form νk+g = o(1) on ∗ can be added to ∗αk+g such that there exists
♦ ♦
βk−g ∈ Z 1 (♦) with A(βk−g ) = ∗αk+g +νk+g , yielding forms ζk♦ , verifying A(ζk♦ ) =
♦
2 (ζk + ζk+g ) + o(1) and whose period matrix is + o(1). Since the periods of αk
1
converge to the same periods as its continuous limit, this period matrix converges to
the period matrix of the surface. Which is the claim of Theorem 4.1.
In the paper [3], R. Costa-Santos and B. McCoy define a period matrix on a special
cellular decomposition of a surface by squares. They do not consider the dual
graph ∗ . Their period matrix is equal to one of the two diagonal blocks of the double
period matrix we construct in this case. They do not have to consider the off-diagonal
blocks because the problem is so symmetric that their period matrix is pure imaginary.
The triangular case is investigated in [4].
4.3 Genus one case

Criticality solves partially the problem of having two different g × g period matrices
instead of one since they converge to one another in a refining sequence. However, on
a genus one critical torus, the situation is simpler: The overall curvature is null and a
critical map is everywhere flat. Therefore the cellular decomposition is the quotient
of a periodic cellular decomposition of the plane by two independent periods. They
can be normalized to (1, τ ). The continuous period matrix is the 1 × 1-matrix τ . A
basis of the two dimensional holomorphic 1-forms is given by the real and imaginary
parts of dZ on
and ∗ respectively,
and the reverse. The discrete period matrix is
Im τ Re τ
the 2 × 2 matrix Re τ Im τ and the period matrices on the graph and on its dual are
both equal to the continuous one.
For illustration purposes, the whole construction, of a basis of harmonic forms,
then projected onto a basis of holomorphic forms, yielding the period matrix, can be
checked explicitly on the critical maps of the genus 1 torus decomposed by square or
triangular/hexagonal lattices:
Consider the critical square (rectangular) lattice decomposition of a torus ♦ =
(Zei θ + Ze−i θ )/(2p ei θ + 2q e−i θ ), with horizontal parameter ρ = tan θ and vertical
parameter its inverse. Its modulus is τ = pq e2 i θ . The two dual graphs and ∗ are
isomorphic. An explicit harmonic form α1 is given by the constant 1/2p on horizontal
and downwards edges of the graph and 0 on all the other edges. Its holonomies are
1 and 0 on the p, resp. q cycles. Considering 1/2q and the dual graph, we construct
∗ ∗
in the same fashion α2 , α1 , α2 . The matrix of inner products is
⎛ q ⎞
p cos 2θ
1 ⎜ ⎟
q
⎜ p cos 2θ ⎟
(αk , α )k, = ⎜ p ⎟ (4.7)
sin 2θ ⎝ cos 2θ q ⎠
p
cos 2θ q
ρ+1/ρ ρ−1/ρ
using 2 = 1/ sin 2θ and = −1/ tan 2θ so that the period matrix is
2

q i sin 2θ cos 2θ
= . (4.8)
p cos 2θ i sin 2θ
Therefore there exists a holomorphic form which has the same periods on the graph
and on its dual, it is the average of the two half forms of Equation (3.30) and its
periods are (1, pq e2 i θ ) along the p, resp. q cycles, yielding the continuous modulus.
This holomorphic form is simply the normalized fundamental form pedZ −i θ .
In the critical triangular/hexagonal lattice, we just point out to the necessary check
by concentrating on a tile of the torus, composed of two triangles, pointing up and
down respectively. We show that there exists an explicit holomorphic form which has
the same shift on the graph and on its dual, along this tile. Let ρ− , ρ\ and ρ/ the three
parameters around a given triangle. Criticality occurs when ρ− ρ\ +ρ\ ρ/ +ρ/ ρ− = 1.
The form which is 1 on the rightwards and South-West edges and 0 elsewhere is
harmonic on the triangular lattice. Its pure imaginary companion on the dual hexagonal
lattice exhibits a shift by i ρ\ in the horizontal direction and i (ρ\ + ρ− ) in the North-
East direction along the tile. Dually, on the hexagonal lattice, the form which is ρ\ ρ−
along the North-East and downwards edges and 1 − ρ\ ρ− along the South-East edges,
is a harmonic form. Its shift in the horizontal direction is 1, in the North-East direction
0, and its pure imaginary companion on the triangular lattice exhibits a shift by i ρ\
in the horizontal direction and i (ρ\ + ρ− ) in the North-East direction along the tile
as before. Hence their sum is a holomorphic form with equal holonomies on the
triangular and hexagonal graphs and the period matrix it computes is the same as the
continuous one. This simply amounts to pointing out that the fundamental form dz
can be explicitly expressed in terms of the discrete conformal data.
4.4 ∂¯ operator
For holomorphic or anti-holomorphic functions, df is, locally on each pair of dual
diagonals, proportional to dZ, resp. dZ, we define ∂ and ∂¯ operators (not to be confused
with the boundary operator on chains) that decompose the exterior derivative into
holomorphic and anti-holomorphic parts yielding

df ∧ df = |∂f |2 + |∂f ¯ |2 dZ ∧ dZ
where the derivatives naturally live on faces:

In the continuous theory, for any derivable function f on the complex plane, the
derivatives ∂ = ddx +i ddy and ∂¯ = ddx −i ddy with respect to z = x +i y and z̄ = x −i y
yield
¯ )(z0 ) + o(|z|).
f (z + z0 ) = f (z0 ) + z(∂f )(z0 ) + z̄(∂f
These derivatives can be seen as a limit of a contour integral over a small loop γ
around z0 :

i ¯ i
(∂f )(z0 ) = lim f d z̄, (∂f )(z0 ) = − lim f dz,
γ →z0 2A(γ ) γ γ →z0 2A(γ ) γ
which leads to the following definitions in the discrete setup:

∂ : C 0 (♦) → C 2 (♦)

i
f
→ ∂f = (x, y, x , y )
→ − f dZ
2A(x, y, x , y ) (x,y,x ,y )
(f (x )−f (x))(ȳ −ȳ)−(x̄ −x̄)(f (y )−f (y))

= (x −x)(ȳ −ȳ)−(x̄ −x̄)(y −y) ,
∂¯ : C 0 (♦) → C 2 (♦)

i
¯ = (x, y, x , y )
→ −
f
→ ∂f f dZ
2A(x, y, x , y ) (x,y,x ,y )
(f (x )−f (x))(y −y)−(x −x)(f (y )−f (y))

= (x −x)(y −y)−(x −x)(y −y) .
¯ ≡ 0 and (with Z(u) noted simply u)

A holomorphic function f verifies ∂f
f (y ) − f (y) f (x ) − f (x)
∂f (x, y, x , y ) = = .
y − y x − x

Notice that the statement f = ∂f dz has no meaning, ∂ is not a derivation endo-
morphism in the space of functions on the vertices of the double.
On the other hand, these differential operators can be extended (see [8]) into op-
erators (the Kasteleyn operator) ∂20 , ∂¯20 : C 2 (♦) → C 0 (♦) simply by transposition,
∂20 = −t ∂02 , leading to endomorphisms of C 0 (♦) ⊕ C 2 (♦). They are such that their
composition, restricted to the vertices ♦0 , gives back the Laplacian:

= 21 (∂ ∂¯ + ∂¯ ∂).
Furthermore, the double derivative ∂20 ∂02 is a well defined endomorphism of

C 0 (♦).
4.5 Geometrical interpretation
Following an idea by Colin de Verdière, generalized by Kenyon, holomorphicity at

criticality can be recognized by eye. Consider the embedding of a graph given by
half the diagonals of a critical map Z. We can embed likewise its dual graph ∗ . They
can actually be represented both at the same time by shrinking every face by a half.
For example the triangular and hexagonal lattices are represented in Figure 7.1. A
discrete holomorphic function is mapping these polygons to similar polygons: Let F
a discrete holomorphic function whose derivative is the discrete holomorphic function
f (this is well defined only up to the bi-constant ε). It maps the polygon x ∗ ∈ 2∗ , dual
to the vertex x ∈ 0 , to a similar polygon centered at F (x), F (x) + f (x)x ∗ where
the multiplication by the complex reiθ encodes the similitude of ratio r and angle θ.
The fact that these polygons still touch at their vertices in the
same manner as in the
reference case Z is precisely the holomorphicity condition ∂x ∗ f dZ = 0.
1. Z 2. Z 2 3. Z 3 4. Z 6
Figure 7. Discrete polynomials in the triangular/hexagonal case
4.6 Quasiconformal maps
For a general discrete function, the derivation with respect to the Z and Z parameters
allow one to define the dilatation coefficient Df : ♦2 → R+ :
¯ |
|∂f | + |∂f
Df := .
¯ |
|∂f | − |∂f
¯ | ≤ |∂f |, we will say that f is quasiconformal. The real
For Df ≥ 1, that is to say |∂f
dilatation can be written in term of the complex dilatation:
¯
∂f (f (x ) − f (x))(y − y) − (x − x)(f (y ) − f (y))
μf = = .
∂f (f (x ) − f (x))(ȳ − ȳ) − (x̄ − x̄)(f (y ) − f (y))
Locally, a discrete harmonic function can be written as a sum of a holomorphic function
and a anti-holomorphic function. Their derivative with respect to Z and Z respectively
give rise to a geometrical interpretation related to Figure 8: The complex dilatation
√
i μf
x∗
x
−−−−→
√
μf F (x)
Figure 8. A quasiconformal map shears a polygon to a non similar shape.
μf (x) can then be defined at a vertex x ∈ and its square root direction is mapped
to the semimajor axis of an ellipse which supports a non similar image of the polygon
x ∗ centered at F (x).
¯ |2 relates the area of the quadrilateral and its image:

The Jacobian J = |∂f |2 − |∂f

df ∧ df = J dZ ∧ dZ.
(x,y,x ,y ) (x,y,x ,y )
4.7 Discrete exponential
Definition 4.4. For a constant λ ∈ C, we define the discrete exponential of parameter λ,

noted exp(:λ: Z), as the solution of
exp(:λ: O) = 1
(4.9)
d exp(:λ: Z) = λ exp(:λ: Z) dZ.
We define its derivatives with respect to the continuous parameter λ:
∂k
Z :k: exp(:λ: Z) := exp(:λ: Z). (4.10)
∂λk
The discrete exponential on the square lattice was defined by Lelong-Ferrand [9],
generalized in [13] and studied independently in [8]. For |λ| = 2/δ, an immediate

check shows that it is a rational fraction in λ at every point: For the vertex x = δ ei θk ,
1+ λδ i θk
2 e
exp(:λ: x) = λδ i θk
(4.11)
k
1− 2 e
where (θk ) are the angles defining (δ ei θk ),

the set of (Z-images of) ♦-edges between
x and the origin. Because the map is critical, Equation (4.11) only depends on the
end points (O, x). It is a generalization of a well known formula, in a slightly better
version,
2 2 n 3 3
λx n λ x 1 + λ2nx λ x
exp(λ x) = 1 + +O = λ x
+O (4.12)
n n 1 − 2n n2
iθ
in the case when the path from the origin to the point x = n1 xn = δ e k is not
restricted to straight equal segments but to a general path of O(|x|/δ) segments of any
directions.
The integration with respect to λ gives an interesting analogue of Z :−k: exp(:λ: Z).
It is defined up to a globally defined discrete holomorphic map. One way to fix it is to
integrate from a given λ0 of modulus 2/δ, which is not a pole of the rational fraction,
along a path that does not cross the circle of radius 2/δ again.
Proposition 4.5. For pointwise multiplication, at every point x ∈ ♦0 ,

exp(:λ: x) · exp(: − λ: x) = 1. (4.13)
The specialization at λ = 0 defines monomials Z :k: := Z :k: exp(:λ: Z)|λ=0 which

fulfill
:k:
Z =k Z :k−1: dZ. (4.14)
The anti-linear duality † maps exponentials to exponentials:

4
exp(:λ:)† = exp(: :). (4.15)
δ 2 λ̄
In particular, exp(:∞:) = 1† = ε is the bi-constant. Because discrete exponential

and discrete monomials do not fulfill
the usual pointwise multiplication relations
exp(λz) exp(μz) = exp (λ + μ)z for λ = −μ, and Z k Z = Z k+ , we denote the
arguments by :λ: and :k: to stress the problem behind the similarity.
Proof of Proposition 4.5. The first assertion is immediate.
The derivation of (4.9) with respect to λ yields

∂k ∂k ∂ k−1
d k exp(:λ: Z) = λ k exp(:λ: Z) + k k−1 exp(:λ: Z) dZ (4.16)
∂λ ∂λ ∂λ
which implies (4.14).
Derivation of exp(:λ:)† gives,
z †

† 4
exp(:λ:) = 2 exp(:λ:) dZ + με
δ O

4 exp(:λ:) − 1 † (4.17)
= + με
δ2 λ
4
= 2 exp(:λ:)† + ν ε
δ λ̄
with μ, ν some constants, so that the initial condition exp(:λ:O)† = 1 at the origin
and the difference equation d exp(:λ:)† = δ 24λ̄ exp(:λ:)† dZ yields the result.
exp(:λ: x)
Note that it is natural to define exp(:λ: (x − x0 )) := exp(:λ: x0 ) as a function of x
with x0 a fixed vertex. It is simply a change of origin. But apart on a lattice where
addition of vertices or multiplication by an integer can be given a meaning as maps
of the lattice, there is no easy way to generalize this construction to other discrete
holomorphic functions such as exp(:λ: (x + n y)) with x, y ∈ ♦0 and n ∈ Z.
4.8 Series
∞ λk Z :k:
The series k=0 k! , wherever it is absolutely convergent, coincides with the ra-
tional fraction (4.11): Its value at the origin is 1 and it fulfills the defining difference
equation (4.9). Using Equation (4.14), a Taylor expansion of exp(:λ: x) at λ = 0 gives
back the same result. We are now interested in the growth rate of the monomials.
Direct analysis gives an estimate of Z :k: :
Proposition 4.6. For x ∈ ♦, at a combinatorial distance d(x, O) of the origin, and

any k ∈ N,
:k: d(x,O)
Z (x)
≤ α+1 δ k
k! α , (4.18)
α−1 2
for any α > 1 arbitrarily close to 1.
∞ λk Z :k:
Corollary 4.7. The series k=0 k! is absolutely convergent for |λ| < 2δ .
Proof of Proposition 4.6. It is proved by double induction, on the degree k and on the
combinatorial distance to the origin.
For k = 0, it is valid for any x since α+1
α−1 = 1 + α−1 > 1, with equality only at
2
the origin.
Consider x ∈ ♦ a neighbor of the origin, Z(x) = δ ei θ , then an immediate
induction gives for k ≥ 1,
i θ k
Z :k: (x) δe
=2 (4.19)
k! 2
which fulfills the condition Equation (4.18) for any k ≥ 1 because α+1 k
α−1 α > 2.
This was done merely for illustration purposes since it is sufficient to check that the
condition holds at the origin, which it obviously does.
Suppose the condition is satisfied for a vertex x up to degree k, and for its neighbor y,
one edge further from the origin, up to degree k − 1. Then,
Z :k: (y) Z :k: (x) Z :k−1: (x) + Z :k−1: (y) Z(y) − Z(x)
= + (4.20)
k! k! (k − 1)! 2
in absolute value fulfills
:k: d(x,O)
Z (y)
≤ α+1 δ k−1 δ α+1 δ
k! α α + 1+
α−1 2 2 α−1 2
d(x,O) k
α+1 δ 2
= α 1+ (4.21)
α−1 2 α−1

α + 1 d(y,O) δ k
= α ,
α−1 2
thus proving the condition for y at degree k. It follows by induction that the condition
holds at any point and any degree.
4.9 Basis
The discrete exponentials form a basis of discrete holomorphic functions on a finite
critical map: given any set of pairwise different reals {λk } of the right dimension, the
associated discrete exponentials will form a basis of the space of discrete holomorphic
functions. See [1] for the formula

1
f (x) = g(λ) exp(: λ : x) dλ (4.22)
2iπ γ
for a certain fixed contour γ in the space of parameters λ, and the definition of g(λ)
as a fixed contour integral in ♦ involving f .
The polynomials however do not form a basis in general: the combinatorial surface
has to fulfill a certain condition called “combinatorial convexity”. A quadrilateral,
when traversed from one side to its opposite, defines a unique chain of quadrilaterals,
that we call a “train-track”. The condition we ask is that two different train-tracks
have different slopes.
On a combinatorially convex set, the discrete polynomials form a basis as well.
4.10 Continuous limit

In a critical map, where quadrilaterals are mapped to rhombi of side δ, we identify a
vertex x with its image Z(x).
The combinatorial distance d♦ (x, O) is related to the modulus |x| through
sin θm |x|
d♦ (x, O) ≤ ≤ d♦ (x, O) (4.23)
4 δ
where θm is the minimum of all rhombi angles. When the rhombi do not flatten, the
combinatorial distance and the modulus (over δ) are equivalent distances.
Lemma 4.8. Let (ABCD) be a four-sided polygon of the Euclidean plane such that
its diagonals are orthogonal and the vertices angles are in [η, 2π − η] with η > 0.
Let (M, M ) be a pair of points on the polygon. There exists a path on (ABCD) from
M to M of minimal length . Then
MM sin η
≥ .
4
It is a straightforward study of a several variables function. If the two points are
on the same side, MM = and sin η ≤ 1. If they are on adjacent sides, the extremal
position with MM fixed is when the triangle MM P , with P the vertex of (ABCD)
η sin η
between them, is isosceles. The angle in P being less than η, MM ≥ sin 2 > 2 .
If the points are on opposite sides, the extremal configuration is given by Figure 9.2.,
sin η
where MM = 4 . 2
A function f : ♦0 → C on the combinatorial surface can be extended to a function
on the image of the critical map in the complex plane fˆ : U → C by stating that
fˆ (Z(x)) = f (x) for the image of a vertex, and extend it linearly on the segments
[Z(x), Z(y)] image of an edge, and harmonically inside each rhombus.
M M
η
P η η
M M
1. M, M on adjacent sides. 2. M, M on opposite sides.
Figure 9. The two extremal positions.
Theorem 4.2. Let (♦k ) a sequence of simply connected critical maps, U the non-
empty intersection of their images in the complex plane and a holomorphic function
f : U → C. If the sequence of minimum angles are bounded away from 0 and the
sequence of rhombi side lengths (δk ) converge to 0, then the function f can be ap-
proximated by a sequence of discrete holomorphic functions fn ∈ (♦k ) converging
to f . The convergence is not only pointwise but C ∞ on the intersection of images.
Conversely a converging sequence of discrete holomorphic functions converges to a
continuous holomorphic function, in particular the discrete polynomials and discrete
exponentials with fixed parameters.
Corollary 4.9. On a Riemann surface, any 1-form can be approximated by a sequence

of discrete holomorphic 1-forms on a refining sequence of critical maps with fixed conic
singularities.
The proof relies on the convergence of polynomials seen as iterated primitives of

the constant function.
Lemma 4.10. Given a sequence of discrete holomorphic functions (fk ) on a refining

sequence of critical
convergingto a holomorphic function f , the sequence
maps
of primitives fk dZ converges to f (z) dz. Moreover, in the compact case, if
the convergence of the functions is of order O(δk2 ), the same property holds for the
primitives.
Proof. Suppose that we are given a sequence of flat vertices Ok ∈ ♦k where the face
containing the fixed flat origin O ∈ U is adjacent to Ok . For a given integer
k, let
F̂k the continuous piecewise harmonic extension of the discrete primitive Ok fk dZ
to U . We want to prove that for any x ∈ U , the following sequence tends to zero
x

(F̂k (x) − F̂k (O)) − f (z) dz . (4.24)

O k∈N
For each integer k consider a vertex xk ∈ ♦0 on the boundary of the face of ♦2
containing x.
We decompose the difference (4.24) into three parts, inside the face containing the
origin O and its neighbor Ok , similarly for x and xk , and purely along the edges of
the graph ♦k itself.
x

F̂k (x) − F̂k (O) − f (z) dz (4.25)

O
xk x

= F̂k (x) − F̂k (xk ) + fk dZ + F̂k (O) − F̂k (Ok ) − f (z) dz ≤
O O
x k x xk
k

≤ F̂k (x) − F̂k (xk ) −
f (z) dz + fk dZ − f (z) dz
xk Ok Ok

Ok

+ F̂k (Ok ) − F̂k (O) − f (z) dz .
O
ξ
On the face of ♦ containing x, the primitive ξ
→ xk f (z) dz is a holomorphic,
hence harmonic function as well as ξ
→ F̂k (ξ ). By the maximum principle, the
ξ
harmonic function ξ
→ F̂k (ξ ) − F̂k (xk ) − xk f (z) dz reaches its maximum on that
face, along its boundary. The difference of the discrete primitive along the boundary
edge (xk , y) ∈ ♦1 at the point ξ = (1 − λ)xk + λ y is equal by definition to
fk (xk ) + fk (y)
F̂k ((1 − λ)xk + λ y) − F̂k (xk ) = λ(y − xk ) . (4.26)
2
The holomorphic f is differentiable with a bounded derivative on U , so averaging the
first order expansions at xk and y, we get
ξ
f (xk ) + f (y)
f (z) dz = λ(y − xk )
xk 2
λ2 f (xk ) + (1 − λ)2 f (y)
+ (y − xk )2 (4.27)
4 3
+ o (ξ − xk ) + o (ξ − y)
3
f (xk ) + f (y)
= λ(y − xk ) + O(δk2 )
2
therefore
x
|F̂k (x) − F̂k (xk ) − f (z) dz| = O(δk2 ). (4.28)
xk
Similarly for the term around the origin.

By definition of fˆk , the 1-form ˆ
xk fk (z) dz along
x edges of the graph ♦ is equal to
the discrete form fk dZ so that Ok fk dZ = Okk fˆk (z) dz on a path along ♦ edges.
Therefore the difference
x xk x

k k
ˆ
f dZ − f (z) dz ≤ f (z) − f (z) dz (4.29)
k k
Ok Ok Ok
is of the same order as the difference |fk (z) − f (z)| times the length (γk ) of a
path on ♦k from Ok to xk . This length is bounded as (γk ) ≤ sin4θm |xk − Ok |.
Since we are interested in the compact case, this length is bounded uniformly and the
difference (4.29) is of the same order as the pointwise difference. We conclude that
the sequence of discrete primitives converges to the continuous primitive and if the
limit for the functions was of order O(δ 2 ), it remains of that order.
The discrete polynomials of degree less than three agree pointwise with their con-
tinuous counterpart, Z :2: (x) = Z(x)2 .
A simple induction then gives the following
Corollary 4.11. The discrete polynomials converge to the continuous ones, the limit
is of order O(δk2 ).
Which implies the main theorem:
Proof of Theorem 4.2. On the simply connected compact set U , a holomorphic func-
tion f can be written, in a local map z as a series,

f (z) = ak zk . (4.30)
k∈N
Therefore, by a diagonal procedure, there exists an increasing integer sequence
(N (n))n∈N such that the sequence of discrete holomorphic polynomials converges
to the continuous series and the convergence is C ∞ ;

N(n)

ak Z :k: → f. (4.31)
n∈N
k=0
4.11 Cross-ratio preserving maps

Once the isometry Z is chosen, holomorphicity of a function f can be written on a
quadrilateral (x, y, x , y ) ∈ ♦2 , writing x = Z(x) for a vertex x ∈ ♦0 , as
f (y ) − f (y) y − y
= (4.32)
f (x ) − f (x) x − x
and f is understood as a diagonal ratio preserving map, and each value at a corner
vertex can be linearly solved in terms of the three others.
A quadratic version is given by the cross-ratio preserving maps: A function f is
said to be quadratic holomorphic iff
(f (y) − f (x)) (f (y ) − f (x )) (y − x) (y − x )
= . (4.33)
(f (x) − f (y )) (f (x ) − f (y)) (x − y ) (x − y)
A rhombic tiling gives rise to two sets of isoradial circle patterns: a set of circles
of common radius δ, whose centers are the vertices of and intersections are the
vertices of ∗ and vice-versa. Two interesting families of cross-ratio preserving maps
are given by circle patterns with the same combinatorics and intersection angles as
one of these two circle patterns.
−−−−−→
Figure 10. A circle pattern with prescribed angles as a cross ratio preserving map.
y q = e−2(θ +θ )

ϕ = e−2ϕ
θ
θ
x x
ϕ
Figure 11. The cross-ratio q is given by the intersection angles.
A change of coordinates helps understanding diagonal ratio preserving maps as a

linearized version of the cross-ratio preserving maps. We will say that the function
w : ♦2 → C solves the Hirota system if, around a face (x, y, x , y ) ∈ ♦2 ,
(y − x)w(x)w(y) + (x − y )w(y )w(x)
(4.34)
+ (y − x )w(x )w(y ) + (x − y)w(y)w(x ) = 0.

This is to be understood as a quadratic version of the Morera theorem f dz = 0 and
w is a half of the derivative of a holomorphic function:
Proposition 4.12. If w solves the Hirota system, then the function f : ♦2 → C defined
up to an additive constant by
f (y) − f (x) = (y − x)w(x)w(y) (4.35)
is quadratic holomorphic.
Proof. The function f is well defined because the associated 1-form is closed by
definition of the Hirota system. The function w disappears in the cross-ratio of f ,
leaving the original cross-ratio.
Conversely, a quadratic holomorphic function defines a solution to the Hirota sys-

tem, unique up to multiplication by λ on , 1/λ on ∗ . Concerning circle patterns
families, w is real on the centers and unitary on the intersections, and encodes the
variation of radius, resp. of direction of the image of the circle:
f (y) − f (x) = r(x)eiθ(y) (y − x). (4.36)
Proposition 4.13. The logarithmic derivative of the Hirota system associated to a

family of cross-ratio preserving maps is a diagonal ratio preserving map.
In other words, for (1 + εg)w to continue solving the Hirota system at first order,
the deformation g must satisfy
g(y ) − g(y) f (y ) − f (y)
= . (4.37)
g(x ) − g(x) f (x ) − f (x)
Proof of Proposition 4.13. The ε contribution of the closeness condition (4.34) for
(1 + εg)w gives
(g(x) + g(y)) f (x) f (y) (y − x) + (g(y) + g(x )) f (y) f (x ) (x − y)
+ (g(x ) + g(y )) w(x ) w(y ) (y − x ) + (g(y ) + g(x)) w(y ) w(x) (x − y ) = 0,
which reads, referring to f :
g(y ) − g(y) f (y ) − f (y)
= . 2
g(x ) − g(x) f (x ) − f (x)
4.12 Bäcklund transformation

The Bäcklund transformation is a way to associate, to a given solution of an integrable
problem, a family of deformed solutions. The two problems under consideration here
are the linear and quadratic holomorphicity constraints on each face. They are given
by a linear, resp. quadratic algebraic relation between the four values of a solution
at the vertices of each face. These relations involve only values supported by the
edges of the rhombus, which are equal on opposite sides, namely the complex label
y − x = x − y.
The Bäcklund transformation is defined by imposing such constraints over new
virtual faces added over each edge, with “vertical edges” labelled by a complex con-
stant λ:
Definition 4.14. Given a linear holomorphic function f ∈ (♦), complex numbers

u, λ ∈ C, its Bäcklund transformation fλ = Bλu (f ) is defined by
fλ (0) = u,
fλ (x) − f (y) λ+x−y (4.38)
= .
fλ (y) − f (x) λ+y−x
Given a quadratic holomorphic function f , complex numbers u, λ ∈ C, its Bäcklund
transformation fλ = Bλu (f ) is defined by
λ (0) = u,
fλ (y) − fλ (x) f (x) − f (y) (y − x)2 (4.39)
= .
fλ (x) − f (x) f (y) − fλ (y) λ2
The right hand sides are the values respectively of the diagonal ratio and cross-ratio
of a parallelogram faces of sides (y − x) and λ seen as “over” the edge (x, y) ∈ ♦1 .
yλ yλ
xλ xλ
y−x+λ
iρ = x−y+λ
λ
q= (y−x)2 y
λ2
x x
Figure 12. The face (xλ , yλ , xλ , yλ ) “over” the face (x, y, x , y ) ∈ ♦2 .
Proposition 4.15. This transformation is well defined in the critical case.
This condition, called three dimensional consistency is an over-determination con-

straint: if the cube “over” the face (x, y, x , y ) ∈ ♦2 is split into two hexagons along
the cycle (y, x , y , yλ , xλ , yλ ), one can see that, given values at these six vertices, the
values at the centers of each hexagons, namely at xλ and at x are overdetermined.
Therefore only certain values at the six vertices are allowed, defined by two alge-
braic relations between them. The compatibility condition is that these two algebraic
relations are equivalent. It is a simple computation to check it is the case for critical
maps.
This transformation verifies
f (O)
Bλ−1 Bλu (f ) = f (4.40)
for any (u, λ). It is an analytic transformation in all the parameters therefore its
derivative is a linear map between the tangent spaces, that is to say between diagonal
xλ xλ
yλ yλ yλ yλ
x xλ
≡
y y
y y
x x
Figure 13. The cube split into two hexagons yielding equivalent compatibility constraints.
ratio preserving maps,

d Bλu (f ) : (f ) → Bλu (f ) . (4.41)
It is not injective and I define the discrete exponential at f as being the direction of
this 1-dimensional kernel. It can be characterized as a derivative with respect to the
initial value at the origin:
∂ v u
expu (:λ:f ) := Bλ−1 Bλ (f ) |v=f (O) ∈ ker d Bλu (f ) (4.42)
∂v

because Bλu Bλv−1 (g) = g for all λ, g and v.
g = Bλu (f )
f = Bλu−1 (g)
Bμu (f )
Bμu −1 (g)
∂ v
expu (:λ:f ) := ∂v Bλ−1 (g)|v=u
Bλv−1 (g)
Figure 14. The discrete exponential expu (:λ:f ) is the kernel of the linear transformation
d Bλu (f ) (here u = f (O)).
As in the discrete exponential case, the value of the Bäcklund transformation at a

given vertex is the image of values at neighbouring vertices by a homography. These
homographies can be encoded as projective operators L(e; λ) ∈ GL2 (C)[λ] lying on
the edges e ∈ ♦2 , called a zero curvature representation:

λ + y − x −2(y − x)(f (x) + f (y))
L((x, y); λ) = for the linear case,
0 λ+x−y
(4.43)
⎛ ⎞
1 −(y − x)w(y)
L((x, y); λ) = ⎝ ⎠ for the Hirota system.
−λ(y − x)/w(x) w(y)/w(x)
(4.44)
Then we define [15] the moving frame : ♦2 → GL2 (C)(λ) by a prescribed value
at the origin and recursively by (y; λ) = L((x, y); λ)(x; λ) and its logarithmic
derivative with respect to λ
d(e; λ) −1
A(e; λ) = (e; λ) (4.45)
dλ
is meromorphic in λ for each edge e. We call f , resp. w, isomonodromic if the
positions and orders of the poles do not depend on the edge e. The two points discrete
Green function (the discrete logarithm) G(O, x) is the inverse of the Laplacian in the
sense that

G(O, •) = δO,• . (4.46)
It can be constructed as the unique isomonodromic solution with some prescribed
data [1], which allows us to give an explicit formula for it, recovering results of
Kenyon [8]: an integral over a loop in the space of discrete exponentials,

1 log 2δ λ
G(O, x) = − 2 exp(:λ: x) dλ (4.47)
8π i C λ
where the integration contour C contains all the possible poles of the rational fraction
exp(:λ: x) but avoids the half line through −x. It is real (negative) on half of the
vertices and imaginary on the others. Because of the logarithm, this imaginary part is
multivalued.
References
[1] A. I. Bobenko, Ch. Mercat, and Y. B. Suris, Linear and nonlinear theories of discrete
analytic functions. Integrable structure and isomonodromic Green’s function. J. Reine
Angew. Math. 583 (2005), 117–161. 542, 556, 566, 574
[2] A. I. Bobenko, Discrete differential geometry. Integrability as consistency. In Discrete
integrable systems, Lecture Notes in Phys. 644, Springer-Verlag, Berlin 2004, 85–110.
542
[3] R. Costa-Santos and B. M. McCoy, Dimers and the critical Ising model on lattices of
genus > 1. Nuclear Phys. B 623 (3) (2002), 439–473. 558
[4] R. Costa-Santos and B. M. McCoy, Finite size corrections for the Ising model on higher
genus triangular lattices. J. Statist. Phys. 112 (2003), 889–920. 559
[5] R. J. Duffin, Basic properties of discrete analytic functions. Duke Math. J. 23 (1956),
335–363. 557
[6] R. J. Duffin, Potential theory on a rhombic lattice. J. Combin. Theory 5 (1968), 258–272.
557
[7] H. M. Farkas and I. Kra, Riemann surfaces. Second edition, Grad. Texts in Math. 71,
Springer-Verlag, New York 1992. 542, 548, 555
[8] R. Kenyon, The Laplacian and Dirac operators on critical planar graphs. Invent. Math.
150 (2) (2002), 409–439. 542, 561, 563, 574
[9] J. Lelong-Ferrand, Représentation conforme et transformations à intégrale de Dirichlet
bornée. Gauthier-Villars, Paris 1955. 563
[10] Ch. Mercat, Holomorphie discrète et modèle d’Ising. PhD thesis, Université Louis
Pasteur, Strasbourg, France, 1998; available at http://www-irma.u-strasbg.fr/annexes/
publications/pdf/98014.pdf 542, 548, 556
[11] Ch. Mercat, Discrete Riemann surfaces and the Ising model. Comm. Math. Phys. 218 (1)
(2001), 177–216. 542, 556, 557
[12] Ch. Mercat, Exponentials form a basis of discrete holomorphic functions on a compact.
Bull. Soc. Math. France 132 (2004), 305–326. 542
[13] Ch. Mercat, Discrete period matrices and related topics. Preprint,2002; arXiv:math-
ph/0111043. 563
[14] D. Mumford, Tata lectures on theta. I. Progr. Math. 28, Birkhäuser, Boston, MA, 1983.
542
[15] F. W. Nijhoff, A. Ramani, B. Grammaticos, and Y. Ohta, On discrete Painlevé equations
associated with the lattice KdV systems and the Painlevé VI equation. Stud. Appl. Math.
106 (3) (2001), 261–314. 574
[16] M. Troyanov, Les surfaces euclidiennes à singularités coniques. Enseign. Math. (2) 32
(1–2) (1986), 79–94. 543
Part D
The quantum theory, 1

Chapter 14
On quantizing Teichmüller and Thurston theories
Leonid O. Chekhov and Robert C. Penner
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 580
2 Classical Teichmüller spaces . . . . . . . . . . . . . . . . . . . . . . . . . 584
2.1 Graph description of Teichmüller spaces . . . . . . . . . . . . . . . . 585
2.1.1 Global coordinates on Teichmüller space . . . . . . . . . . . . 585
2.1.2 Weil–Petersson form . . . . . . . . . . . . . . . . . . . . . . 588
2.1.3 Mapping class group description using graphs . . . . . . . . . 589
2.2 Poisson algebras of geodesic functions . . . . . . . . . . . . . . . . . 591
2.2.1 Multicurves . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
2.2.2 Classical skein relation . . . . . . . . . . . . . . . . . . . . . 591
2.2.3 Poisson brackets for geodesic functions . . . . . . . . . . . . 592
2.2.4 Poisson geodesic algebras for higher genera . . . . . . . . . . 593
3 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
3.1 Quantizing Teichmüller spaces . . . . . . . . . . . . . . . . . . . . . 594
3.2 Geodesic length operators . . . . . . . . . . . . . . . . . . . . . . . 599
3.3 Algebra of quantum geodesics . . . . . . . . . . . . . . . . . . . . . 601
3.4 Quantizing the Nelson–Regge algebras . . . . . . . . . . . . . . . . . 603
3.5 Improving the quantum ordering . . . . . . . . . . . . . . . . . . . . 603
4 Classical Thurston theory of surfaces . . . . . . . . . . . . . . . . . . . . 606
4.1 Measured foliations and Thurston’s boundary . . . . . . . . . . . . . 606
4.2 Train tracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
4.3 Laminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
4.4 Dynamics on train tracks . . . . . . . . . . . . . . . . . . . . . . . . 616
4.5 Decorated measured foliations and freeways . . . . . . . . . . . . . . 618
4.6 Shear coordinates for measured foliations . . . . . . . . . . . . . . . 620
5 On quantizing Thurston theory . . . . . . . . . . . . . . . . . . . . . . . . 622
5.1 Proper length of geodesics . . . . . . . . . . . . . . . . . . . . . . . 622
5.2 Approximating laminations and the main theorem . . . . . . . . . . . 623
5.3 Elements of the proof . . . . . . . . . . . . . . . . . . . . . . . . . . 626
5.3.1 Continued fraction expansion . . . . . . . . . . . . . . . . . . 626
5.3.2 Mapping class group transformations and the unzipping
procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631
580 Leonid O. Chekhov and Robert C. Penner
5.3.3 Asymptotic regime . . . . . . . . . . . . . . . . . . . . . . . 632

5.3.4 Quantum continued fraction expansion . . . . . . . . . . . . . 636
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 638
Appendix A. Combinatorial proof of Theorem 5.1 . . . . . . . . . . . . . . . . 639
Appendix B. Degeneracy of the Poisson structure . . . . . . . . . . . . . . . . 641
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
1 Introduction
One manifestation of Teichmüller space in contemporary mathematical physics is as
the Hilbert space and the algebra of observables for three-dimensional (3D) quantum
gravity since E. Verlinde and H. Verlinde [39] have argued that the classical phase
space of Einstein gravity in a 3D manifold is the Teichmüller space of its boundary.
(Analogously, the classical phase space for 3D Chern–Simons theory is the moduli
space of flat connections on the boundary; this theory was quantized in [12], [13].)
Teichmüller space possesses its canonical (Weil–Petersson) Poisson structure, whose
symmetry group is the mapping class group of orientation-preserving homeomor-
phisms modulo isotopy. The algebra of observables is the collection of geodesic
length functions of geodesic representatives of homotopy classes of essential closed
curves together with its natural mapping class group action.
Given this Poisson structure, one can turn to the problem of quantizing it, thereby
obtaining a variant of the quantum 3D gravity description. According to the correspon-
dence principle: (1) the algebra of observables of the corresponding quantum theory
is the noncommutative deformation of the ∗-algebra of functions on it governed by the
Poisson structure; (2) the Hilbert spaces of the theory are the representation spaces of
these ∗-algebras; and (3) the symmetry group acts on the algebra of observables by au-
tomorphisms. Under the assumption that the quantization of a Poisson manifold exists
and is unique, to solve this problem it suffices to construct a family of ∗-algebras, which
depend on the quantization parameter h̄, and an action of the mapping class group on
this family by outer automorphisms, and to show that the algebra and the action thus
constructed reproduces the classical algebra, the classical action, and the classical
Poisson structure in the limit h̄ → 0. This program has been successfully performed
in [6] and [16], and we describe it in Sections 2 and 3 of this chapter. Actually, the
problem that was solved in these papers differs slightly from the original formulation
because the methods of [6] are suited for describing only open surfaces (surfaces with
nonempty boundary, components of which can, however, reduce to a puncture). The
corresponding Teichmüller space has a degenerate Weil–Petersson Poisson structure,
while the mapping class group acts as symmetry group. We describe the deformation
quantization of the corresponding Teichmüller space, the action of the mapping class
group by outer automorphisms, the representations of the observable algebra, and the
induced action of the mapping class group on the representation space following [6].
Chapter 14. On quantizing Teichmüller and Thurston theories 581
As in [39], the representation space of the observable algebra can also be interpreted
as the space of conformal blocks of the Liouville conformal field theory. This program
is under current development (see [33] and [34]). Our construction can be therefore
interpreted as the construction of the conformal block spaces and the mapping class
group actions for this CFT.
The key point of the quantization procedure is constructing quantum mapping class
group transformations that define in a consistent way the morphisms between quan-
tum ∗-algebras simultaneously preserving the quantum geodesic algebra. The main
mathematical ingredient of the construction is a version of the quantum dilogarithm
by L. D. Faddeev [8]. We interpret the corresponding five-term relation as the only
nontrivial relation in a certain groupoid that has the mapping class group as maximal
subgroup. A similar construction has been made independently and simultaneously
by R. M. Kashaev [16]. The key difference between these two constructions lies in the
dimensions of the Poisson leaves of the two theories: given a graph with v three-valent
vertices, e edges, and f = s faces suitably embedded in a surface Fgs of genus g with s
ideal boundary components, the genus is given by Euler’s formula v −e +f = 2 −2g.
In Kashaev’s approach, there are 2e variables and v + f central elements (at each ver-
tex and at each face), so the Poisson leaf dimension is 2e − v − f = 8g − 8 + 3s,
while in our approach, there are e variables and f central elements, so the Poisson
dimension e − f = 6g − 6 + 2s exactly coincides with the dimension of the Teich-
müller space of Riemann surfaces of genus g with s punctures. The approach of [6] is
thus appropriate for describing 2D topological theories while that of [16] is suitable
for describing Liouville field theory as a lattice theory (for instance, the Liouville field
central charge can be calculated, see [17]).
One of the mathematical tools employed in the quantization [6] is the decorated
Teichmüller theory [25], and the relevant aspects are briefly reviewed in Section 2.
In effect in the classical case, to each edge of a trivalent “fatgraph” (i.e., a graph
plus a cyclic ordering of the half-edges about each vertex) embedded as a “spine”
(i.e., a deformation retract of the surface) F = Fgs is assigned a number Zα , where
α here and below indexes the edges of . The tuple (Zα ) gives global coordinates
on an appropriate Teichmüller space TH = TH (F ) of F (as first studied by Thurston
and later by Fock), where any or all of the “punctures” of Fgs are permitted to be
uniformized instead as circular boundary components (and the subscript H stands for
“holes”); the details are given in Section 2.1. The Weil–Petersson Kähler two-form,
which is known [27] in the (Zα ) coordinates, pulls back to a degenerate two-form, i.e.,
to a degenerate Poisson structure on TH . Furthermore, the action of the mapping class
group MC = MC(F ) of F on TH , which is again known [27] in the (Zα ) coordinates
as well as combinatorially [28], leaves invariant this Poisson structure, and this is the
Poisson manifold TH with MC-action which has been quantized.
More explicitly still, for each edge α, we may associate a pair of Möbius transfor-
mations RZα , LZα depending upon Zα with the following property. For any homotopy
class γ of geodesic in F with corresponding closed edge-path P on , consider the
serial product Pγ of operators RZα , LZα taken in order as one traverses P in some
orientation from some starting point, where one inserts the former (or latter, respec-
tively) if immediately after traversing edge α, then P turns right (or left) in ; thus,
the combinatorial geometry of serially traversing edges of as dictated by P deter-
mines an ordered product Pγ of matrices depending upon (Zα ), and the length lγ of
the geodesic representative of γ for the point of TH determined by (Zα ) is given by
Gγ = 2 cosh(lγ /2) = tr Pγ |. It is the Poisson algebra of these geodesic functions
Gγ , the algebra of observables, which has been quantized.
In the quantum case (after passing to a suitable subspace on which the Poisson
structure is non-degenerate), standard techniques of deformation quantization produce
the appropriate Hilbert space H of the quantum theory, as well as pairs of operators
RZα , LZα on H 2 for each edge α of . Again, to the homotopy class of a geodesic γ in
F or its corresponding closed edge-path P on , we may assign the ordered product
Pγ of these operators as dictated by the combinatorial geometry of P in , whose
trace Gγ = tr Pγ is the “quantum geodesic operator.” The main point is to prove
the invariance under the action of MC, which is intimately connected with functional
properties of the quantum dilogarithm as was mentioned before.
In [6] was proved the existence and uniqueness of an appropriate “proper quan-
tum ordering” of operators that enjoy MC-invariance as well as satisfy the standard
physical requirements. In Section 3.5, we observe that the natural operatorial order-
ing given by the combinatorial geometry of edge-paths in can be used to derive
this physically correct quantum ordering. The improved ordering is required in the
subsequent quantization (discussed below).
In order to explain the further new results in this chapter, we must recall aspects of
Thurston’s seminal work on surface geometry, topology, and dynamics from the 1970–
1980s, which is surveyed in Section 4. Very briefly, Thurston introduced the space
P F 0 = P F 0 (F ) of “projective measured foliations of compact support in F ” as a
boundary for the Teichmüller space T = T (F ), where “Thurston’s compactification”
T = T ∪ P F 0 is a closed ball with boundary sphere P F 0 , where the action of
MC on T extends continuously to the natural action on T . Furthermore, the sphere
P F 0 contains the set of all homotopy classes of geodesics in F as a dense subset,
i.e., P F 0 is an appropriate completion of this set. (Unfortunately, the action of MC
on P F 0 has dense orbits, so the beautiful structure of Thurston’s compactification
does not descend in any reasonable way, with the current state of understanding, to a
useful structure on the level of Riemann’s moduli space.) Thurston also devised an
elegant graphical formalism for understanding P F 0 using “train tracks”, which are
graphs embedded in F with the further structure of a “branched one-submanifold” (cf.
Section 4.2). In effect, a maximal train track gives a chart on the sphere P F 0 , and
furthermore, the combinatorial expression for inclusion of charts effectively captures
the dynamics of the action of diffeomorphisms of F (cf. Section 4.4).
Thus, Thurston theory arises as a natural tool to understand degenerations in Teich-
müller space and dynamics on F , and in light of remarks above, its quantization should
provide a natural tool for studying degenerations of 3D gravity or Liouville conformal
field theory. Many aspects of the survey of classical Thurston surface theory (in
Section 4) are required for our subsequent quantization.
In any case, a mathematically natural problem armed with [6] is to “quantize
Thurston theory”: assign operators Hγ on H to each homotopy class γ of geodesic in
such a way that as γ converges in P F 0 to a projectivized measured foliation [F , μ̄],
then the corresponding operators Hγ converge in an appropriate sense to a well-
defined operator H[F ,μ̄] on H. Upon choosing a spine ⊆ F (i.e., an embedding, up
to homotopy, of a graph whose inclusion is a homotopy equivalence), any homotopy
class of curve γ may be essentially uniquely realized as an edge-path on . Define the
“graph length” g.l.(γ ) to be the total number of edges of traversed by this edge-path
counted with multiplicities.
In the special case of the torus F11 , we have succeeded here (in Section 5) in showing
that the ratio of operators
Hγ = p.l.(γ )/g.l.(γ )
converge weakly to a well-defined operator as γ converges in P F 0 (F11 ), where the
“proper length” is defined by

1 1
p.l.(γ ) = lim tr log 2Tn Pγ ,
n→∞ n 2
with Tn the Chebyshev polynomials. In fact in the classical case, p.l.(γ ) agrees
with half the length of the geodesic homotopic to γ in the Poincaré metric, and in
the quantum case, there is an appropriate operatorial interpretation, both of which
are described in Section 5.1. In particular, for several spines whose corresponding
charts cover the circle P F 0 (F11 ), the analysis involves rather intricate estimates.
This leads to a natural operatorial quantization (in Section 5.3.4) of the standard
simple continued fractions, which are intimately connected with Thurston theory on
F11 (as we shall describe in Section 4.5). To complete the basic theory on the torus,
one would like an intrinsic operatorial description of the circle of unbounded operators
we have constructed, as well as an explication of the mapping class group action on
it, viz., Thurston’s classification of surface automorphisms.
This chapter is organized as follows: Section 2 covers classical Teichmüller theory
and Section 3 the Chekhov–Fock quantization; Section 4 surveys classical Thurston
theory of surfaces, and Section 5 gives our quantization of Thurston’s boundary for the
punctured torus. Excerpts of Sections 2 and 3 are derived from an earlier manuscript
of Chekhov–Fock, and we strive to include further mathematical detail. Section 4
surveys aspects of a large literature on Thurston theory and train tracks, explicitly
covering only what is required in Section 5. Section 5 should be regarded as work in
physics in the sense that some of the formal calculations depend upon manipulations
of asymptotic spectral expansions for which there may be remaining mathematical
issues. Closing remarks in Section 6 discuss the natural extension of these results to
more complicated surfaces as well as other related work. Appendix A includes a novel
proof of the required convergence in the classical case for any surface (in a sense, a
new proof of the existence of Thurston’s compactification), which may yet be useful
in the quantum case. Appendix B contains an analysis of the Casimir operators in

the Poisson algebra in the Z-variables and the appropriate diagonalization of Poisson
structure.
Acknowledgements. We are indebted to L. D. Faddeev, V. V. Fock, M. Lapidus, and
F. Bonahon for useful discussions. The work was partially supported by the RFFI
Grant No. 01-01-00549 (L.Ch.), by the Program Mathematical Methods in Nonlinear
Dynamics (L.Ch.), and by the COBASE Project.
2 Classical Teichmüller spaces

To begin, we shall briefly recall the two related roles played by graphs in Teichmüller
theory as both aspects will be required here.
A fatgraph or ribbon graph is a graph together with a cyclic ordering on the
half-edges incident on each vertex, and we canonically associate to a surface F ()
with boundary obtained by “fattening each edge of the graph into a band” in the natural
way; we shall tacitly require all vertices to have valence at least three unless stated
otherwise, and we shall call a fatgraph cubic if each vertex has valence three. To each
homotopy class of homotopy equivalence ι : → F for some surface F , where ι
respects the orientation, there is a corresponding cell in Tgs × Rs>0 as explained in [29]
in the hyperbolic setting [25] and in the conformal setting [32]. Thus, a homotopy
class of ι : → F is the name of a cell in the canonical cell decomposition. We shall
sometimes suppress the mapping ι : → F and refer to itself as a spine of F , where
is identified with ι() ⊆ F . The cell decomposition is invariant under the action of
the mapping class group (induced by post-composition of ι with homeomorphisms),
and this has been an effective tool for studying Riemann’s moduli space; for instance,
we shall recall here the corresponding presentation of the mapping class groups.
The second role of fatgraphs is exclusive to the hyperbolic setting, namely, fat-
graphs provide a kind of “basis” for geometrically natural global parameterizations
of Teichmüller space. Specifically, fix a homotopy class ι : → F as above, where
we now demand that is cubic, and let E = E() denote the set of edges of . In
several different contexts, one can naturally identify RE >0 with a suitable modification
of an appropriate Teichmüller space of F .
For instance, √for punctured surfaces, recall [25] that the lambda length of a pair
of horocycles is 2 eδ , where δ is the signed hyperbolic distance between the horo-
cycles. Lambda lengths give a global real-analytic parametrization of the decorated
Teichmüller space as the trivial bundle T˜gs = Tgs × Rs>0 over Teichmüller space,
where the fiber over a point is the space of all s-tuples of horocycles in the surface,
one horocycle about each puncture (parameterized by hyperbolic length).
For another example, Thurston’s shear coordinates [36], [2] give global parameters
not only on Teichmüller space (cf. Section 2.1.1) but also on the related space of
measured foliations (cf. Section 4.6).
On the level of Teichmüller space, the two global coordinate systems (lambda
lengths and shear coordinates) are closely related, and we choose to give the exposi-
tion here principally in shear coordinates with the parallel lambda length discussion
relegated to a series of ongoing remarks. On the other hand, certain proofs of identities
involving shear coordinates are easy calculations in lambda lengths.
2.1 Graph description of Teichmüller spaces

2.1.1 Global coordinates on Teichmüller space. In addition to the Teichmüller
space Tgs and decorated Teichmüller space T˜gs , we shall also require the following
modification. Given an open Riemann surface F of finite topological type, a neigh-
borhood of an ideal boundary component is either an annulus or a punctured disk;
in the former case, the ideal boundary component will be called a “true” boundary
component and in the latter will be called a “puncture.” We shall study the latter as a
degeneration of the former with an elaboration

Teich(F ) = Hom π1 (F ), PSL2 (R) / PSL2 (R)
of the usual Teichmüller space, where Hom denotes the space of all discrete faithful
representations with no elliptic elements, i.e., | tr(ρ(γ ))| ≥ 2 for all γ ∈ π1 (F ) for
any representation ρ : π1 (F ) → PSL2 (R).
Assume that ι : → F is a homotopy class of homotopy equivalence and γ ∈
π1 (F ) is conjugate in π1 (F ) to the boundary of a regular neighborhood of an ideal
boundary component of F (). Thus, the ideal boundary component is a puncture if
and only if ρ(γ ) is a parabolic transformation, i.e., | tr(ρ(γ ))| = 2. For any γ ∈ π1 (F )
with | tr(ρ(γ ))| > 2, the underlying free homotopy class of unbased curves contains a
unique hyperbolic geodesic whose length lγ is given by Gγ = 2 cosh(lγ ) = | tr ρ(γ )|,
where Gγ is called the geodesic operator and is constant on the conjugacy class of
γ ∈ π1 (F ). Furthermore, by definition,
Tgs = {[ρ] ∈ Teich(F ) : for all γ ∈ π1 (F ) freely homotopic into
the boundary, we have tr(ρ(γ ))| = 2}
⊆ Teich(F )
Finally, define the space TH (F ) to be the 2s -fold cover of Teich(F ) branched over
Tgs , where the fiber is given by the set of all orientations on the boundary components
of F .
Theorem 2.1. Fix any spine ⊆ F , where is a cubic fatgraph. Then there is a
real-analytic homeomorphsim RE() → TH (F ). The hyperbolic length lγ of a true
boundary component γ is given by lγ = | Zi |, where the sum is over the set of
all edges traversed by γ counted with multiplicity. Furthermore, Zi = 0 if and
only if the corresponding ideal boundary component is a puncture, so Tgs ⊆ TH (F )
is determined by s independent linear constraints.
The theorem is due to Thurston with a systematic study by Fock. We shall not
give a proof here (though there is not a complete proof in the literature so far as we
know), but we shall at least give the construction that defines the homeomorphism
RE() → TH (F ).
The basic idea is to associate to each edge of an appropriate cross ratio. To set
this up, consider the topological surface F + ⊆ F obtained by adjoining a punctured
disk to each true boundary component of F . The fatgraph ι() ⊆ F ⊆ F + is thus
also a spine of F + , and its Poincaré dual in F + is an ideal triangulation of F + (i.e.,
a decomposition into triangles with vertices among the punctures). In the universal
cover of F + , each arc of thus separates two complementary triangles which combine
to give a topological quadrilateral, and the basic idea is to associate to each edge the
cross-ratio of this quadrilateral.
To make this precise and describe the homeomorphism in the theorem, let α =
1, . . . , E = E() index the edges of , and let (Zα ) denote a point of RE . We
associate the Möbius transformation

0 − e Zα /2
XZα = . (2.1)
e −Zα /2 0
to the edge α. To explicate this definition, consider an ideal quadrilateral in the

hyperbolic plane triangulated by a diagonal into two ideal triangles T1 , T2 . We may
conjugate in PSL2 (R) to arrange that the vertices of T1 are 0, −1, ∞ and the vertices
of T2 are 0, −1, t, where 0 < t < ∞, and an appropriate cross ratio of the original
quadrilateral is t. Setting Zα = log t in the formula above, XZα is the Möbius
transformation interchanging 0, ∞ and sending −1 to t, i.e., sending T1 to T2 . Notice
that XZ2 α is the identity in PSL2 (R), so XZα also sends T2 to T1 .
We also introduce the “right” and “left” turn matrices

1 1 0 1
R= , L = R2 = , (2.2)
−1 0 −1 −1
and define the corresponding operators RZ and LZ ,

e −Z/2 − e Z/2
RZ ≡ RXZ = , (2.3)
0 e Z/2

e −Z/2 0
LZ ≡ LXZ = . (2.4)
− e −Z/2 e Z/2
Consider a closed oriented edge-path P in , where we assume that P never con-

secutively traverses an oriented edge followed by its reverse, i.e., there is no “turning
back”. Choosing also an initial base point on P , we may imagine the corresponding
curve serially traversing the oriented edges of with coordinates Z1 , . . . , Zn turn-
ing left or right from Zi to Zi+1 , for i = 1, . . . , n (with the indices mod n so that
Zn+1 = Z1 ). Assign to P the corresponding composition

PZ1 ...Zn = LZn LZn−1 RZn−2 . . . RZ2 LZ1 , (2.5)
where the matrices LZi or RZi are inserted depending on which turn – left or right –
the path takes at the corresponding stage.
Proposition 2.2. There is a one-to-one correspondence between the set of conjugacy

classes of elements of π1 (F ) and free homotopy classes of closed oriented geodesics
in F . For any spine of F , each free homotopy class is uniquely represented by a
cyclically defined closed edge-path P with no turning back, and the length of γ is
determined by
Gγ ≡ 2 cosh(lγ /2) = | tr PZ1 ...Zn |. (2.6)
By construction if P corresponds to a boundary component γ of F (), then the

associated matrix has the form RZ1 RZ2 . . . RZn , or LZ1 LZ2 . . . LZn depending on
the orientation. In this case, because all of the matrices Rx (Lx ) are upper (lower)
triangular, formula (2.6) gives

n

lγ = Zi , (2.7)
i=1
where the sign of this sum gives the orientation of the boundary component, and lγ = 0
corresponds to a puncture. This proves the assertions about boundary lengths.
The Z-coordinates (i.e., log cross ratios) are called (Thurston) shear coordinate
[36], [2] and can alternatively be defined by dropping perpendiculars from each of the
two opposite vertices to the diagonal α of a quadrilateral, and measuring the signed
hyperbolic distance Zα along α between these two projections.
Assume that there is an enumeration of the edges of and that edge α has distinct
endpoints. Given a spine of F , we may produce another spine α of F by contracting
and expanding edge α of , the edge labelled Z in Figure 1, to produce α as in the
figure; the fattening and embedding of α in F is determined from that of in the
natural way. Furthermore, an enumeration of the edges of induces an enumeration
of the edges of α in the natural way, where the vertical edge labelled Z in Figure 1
corresponds to the horizontal edge labelled −Z. We say that α arises from by a
Whitehead move along edge α. We shall also write αβ = (α )β , for any two indices
α, β of edges, to denote the result of first performing a move along α and then along
β; in particular, αα = for any index α.
Proposition 2.3 ([6]). Setting φ(Z) = log(eZ + 1) and adopting the notation of
Figure 1 for shear coordinates of nearby edges, the effect of a Whitehead move is
illustrated in the figure, viz.,
WZ : (A, B, C, D, Z) → (A + φ(Z), B − φ(−Z), C + φ(Z), D − φ(−Z), −Z).
(2.8)
In the various cases where the edges are not distinct and identifying an edge with its
shear coordinate in the obvious notation we have: if A = C, then A = A + 2φ(Z);
if B = D, then B = B − 2φ(−Z); if A = B (or C = D), then A = A + Z (or
C = C + Z); if A = D (or B = C), then A = A + Z (or B = B + Z).
HA
H
B

H
H A + φ(Z) B − φ(−Z)
A
A −Z
- AA
Z
A
A
AA

HH D − φ(−Z) C + φ(Z)

H
H
D C
Figure 1. Whitehead move on shear coordinates.
Sketch of proof. Assume that e is the diagonal of a quadrilateral with consecutive sides
a, b, c, d, where e separates a, b from c, d. Identifying an edge with its lambda length,
the shear coordinate is given by Z = log bd bd
ac , i.e., ac is the required cross-ratio [25].
Furthermore, if f is the lambda length of the other diagonal, then the lambda lengths
satisfy Ptolemy’s relation ef = ac + bd [25], and the transformation laws for shear
coordinates in the proposition are readily derived from this either in the surface for
(2.8) or in the universal cover of the surface in the various cases.
Insofar as hyperbolic lengths of geodesics are well-defined invariants of homotopy

classes of curves in F , these lengths must be invariant under Whitehead moves, so we
have the following
Lemma 2.4. Transformation (2.8) preserves the traces of products over paths (2.6).
2.1.2 Weil–Petersson form. TH (F ) supports its canonical Weil–Petersson Poisson

structure, which has a very simple form in shear coordinates.
Theorem 2.5 ([10]). In the coordinates (Zα ) on any fixed spine, the Weil–Petersson
bracket BWP is given by

3
∂ ∂
BWP = ∧ , (2.9)
v
∂Zvi ∂Zvi+1
i=1
where the sum is taken over all vertices v and vi , i = 1, 2, 3 mod 3, are the labels of
the cyclically ordered half-edges incident on this vertex.
The proof [10] relies on the independence of this form under Whitehead moves
as in [28]. Indeed, the equivalent expression for the Weil–Petersson Kähler two-form
in the punctured case was first given in lambda length coordinates in [27] starting
from Wolpert’s formula [41], and the formula in shear coordinates follows from direct
calculation using the expression Z = log bd ac . (There is more to this geometrically,
however, and one must show [3] that the same expression is the Weil–Petersson form
for surfaces with boundary.)
The set of Casimir functions is described by the following proposition, whose proof
is given in Appendix B.
Proposition
2.6. The center of the Poisson algebra (2.9) is generated by elements of
the form Zα , where the sum is over all edges of in a boundary component of
F () and the sum is taken with multiplicity.
2.1.3 Mapping class group description using graphs. Recall that the mapping class
group MC(F ) of an open surface F is the group of homotopy classes of orientation-
preserving homeomorphisms of F . No special constraints are imposed by the circle
boundary components, i.e., a homeomorphism must fix each boundary component
only setwise, and the homotopies must likewise fix each boundary component only
setwise. Thus, if F has b boundary component circles, p punctures, and genus g,
b+p
then MC(F ) ≈ MC(Fg ), so we generally write MCsg = MC(F ) for any surface of
genus g with s boundary components. In this section, we establish the combinatorial
presentation of MCsg associated with the cell decomposition of decorated Teichmüller
space.
Recall that a cell in the decomposition of T˜gs is described by the homotopy classes
of an embedding ι : → F of a fatgraph as a spine of F . MCsg acts on the set of
homotopy classes of such embeddings by post-composition, and the cell decomposi-
tion of decorated Teichmüller space T˜gs descends to an orbifold cell decomposition of
M̃gs = T˜gs / MC(F ).
The modular groupoid MGsg = MG(F ) is the fundamental path groupoid of M̃gs ,
and MCsg arises as the subgroup of paths based at any point. Specifically, consider
the dual graph Gsg = G(F ) of the codimension-two skeleton of this decomposition
of M̃gs (where there is one vertex for each top-dimensional cell, edges correspond to
Whitehead moves, and two-dimensional cells correspond to pairs of homotopic paths
in the one-skeleton which are homotopic to real endpoints in M̃gs .) The fundamental
path groupoid of Gsg is the modular groupoid, and in particular, MCsg is the stabilizer
in MGsg of any vertex of Gsg .
We may think of a Whitehead move along edge α of fatgraph producing another
fatgraph α as an ordered pair (, α ), i.e., an oriented edge of Gsg . Letting [1 , 2 ]
denote the MCsg -orbit of a pair (1 , 2 ) by the diagonal action of the mapping class
group, the natural composition descends to a well-defined product
[1 , 3 ] = [1 , 2 ][2 , 3 ].
Theorem 2.7 ([28], [27]). The modular groupoid MGsg is generated by Whitehead
moves and relabelings by fatgraph symmetries. A complete list of relations in MGsg
is given by relabelings under fatgraph symmetries together with the two following
relations.
If α and β are two edges with no common endpoints, then
[αβ , α ][α , ] = [αβ , β ][β , ] (Commutativity).
If α and β share exactly one common endpoint, then (see Figure 2 drawn for the
dual graph)
[, α ][α , βα ][βα , αβ ][αβ , β ][β , ] = 1 (Pentagon).
Furthermore, the expression in Theorem 2.5 for the Weil–Petersson form is invariant
under Whitehead moves.
H
C4 B HHB4
B HH
B H
Q
A Q
B X4
A Q B
D4 A Q A4
Y4Q B
A Q B @
I@
H QQB H
E0 HHD0 A A3 B H E3

Y0 HH E 4
B HH
H B H
H
AQQ A X3 B 3Y
A Q A B
A0 A Q C0 B3 A D3
X0Q B
A Q A B
A QQ A B
B0 C3
AA
U
H H
B1 HHA1 D2 C
H 2
HH
X H
H H
1
H H
A A Y2
A A
C1 A
- E2 A
Y E1 X2 B2
A 1 A

A A
D1 A2
Figure 2. Pentagon identity.
Proof. The first parts are immediate consequences of the cell decomposition. Specifi-
cally, by connectivity of T˜gs , any two points can be joined by a smooth path, which we
may put into general position with respect to the codimension-one faces; this proves
the first part. For the second part, a homotopy between edge-paths in M̃gs can likewise
be put into general position with respect to the codimension-two faces; there are two
possibilities for a pair of edges depending upon whether their vertices are disjoint or
not, corresponding respectively to the commutativity and pentagon relations, proving
the second part. The invariance of the expression for the Weil–Petersson form under
Whitehead moves is a direct calculation in lambda lengths [29] or shear coordinates
using (2.8).
2.2 Poisson algebras of geodesic functions

The algebra generated (with multiplication and with the Weil–Petersson Poisson
bracket) by the functions {Gγ } (2.6) was first studied by W. Goldman [14].
2.2.1 Multicurves. In the sequel, disjointly embedded families of geodesics will

play a special role as they constitute a basis for the algebra of observables in both
the classical case considered here and the quantum case discussed in Section 3. The
homotopy class of such a family is called a multiple curve. A multicurve is multiset
based on the set of curves in a multiple curve.
Definition 2.1. Consider the homotopy class of a finite collection C = {γ1 , . . . , γn } of

disjointly embedded (unoriented) simple closed curves γi in a topological surface F ,
where C need not be a mutiple curve. A generalized multicurve (GMC) Ĉ in F is a
multiset based on C; one thus imagines si ≥ 1 parallel copies of components of C, or
in other words, positive integral weights si on each component of C, where si is the
multiplicity of γi in Ĉ. Further, given a hyperbolic structure on F , we associate to Ĉ
the product GĈ = Gsγ11 . . . Gsγnn of geodesic operators (2.6) of all geodesics constituting
a GMC; these operators Poisson commute in the classical case since the components
of C are disjoint. In particular, a GMC containing a contractible component (of length
zero) is twice the GMC with this curve removed.
An edge-path on a spine ⊆ F or its corresponding geodesic γ in F is said to be

graph simple path with respect to if it does not pass more than once through any
edge of . Obviously, the set of graph simple geodesics depends upon and is not
invariant under Whitehead moves. Nevertheless, this notion will be useful in what
follows.
2.2.2 Classical skein relation. The trace relation tr(AB)+tr(AB −1 )−tr A·tr B = 0
for arbitrary 2 × 2 matrices A and B with unit determinant allows one to “disentangle”
any product of geodesic functions, i.e., express it uniquely as a finite linear combination
of GMCs. Introducing the additional factor #G to be the total number of components
in a GMC, we can then uniformly present the classical skein relation as

$ '
@
@ &%
(−1)#G + (−1)#G + (−1)#G' $ = 0.
@
@@ % & (2.10)
2.2.3 Poisson brackets for geodesic functions. Turning attention now to the Pois-
son structure, two geodesic functions Poisson commute if the underlying geodesics
are disjointly embedded. By the Leibnitz rule for the Poisson bracket, it suffices to
consider only “simple” intersections of pairs of geodesics with respective geodesic
functions G1 and G2 of the form
G1 = tr 1 . . . XC1 R 1 XZ1 L1 XA
1
..., (2.11)
G2 = tr 2
. . . XB2 L2 XZ2 R 2 XD
2
..., (2.12)
where the superscripts 1 and 2 pertain to operators and traces in two different matrix
spaces.
The bracket between XC1 and XB2 possesses a simple r-matrix structure
1
{XC1 , XB2 } = (−1)i+j (eii1 ⊗ ejj
2
)XC1 ⊗ XB2 , (2.13)
4
where the “elementary” matrix eij has entry unity in its ith row and j th column and
zero otherwise. Direct calculations then give
1
{G1 , G2 } = (GH − GI ), (2.14)
2
where GI corresponds to the geodesic that is obtained by erasing the edge Z and joining
together the edges “A” and “D” as well as “B” and “C” in a natural way as illustrated
in the middle diagram in (2.10); GH corresponds to the geodesic that passes over the
edge Z twice, so it has the form tr . . . XC RZ RD . . . . . . XB LZ LA . . . as illustrated
in the rightmost diagram in (2.10). These relations were first obtained in [14] in the
continuous parametrization (the classical Turaev–Viro algebra).
Torus example. For the torus, TH (F11 ) has three generators X, Y, Z, where
{X, Y } = {Y, Z} = {Z, X} = 2,
corresponding to the combinatorially unique cubic spine and the Casimir element is
X + Y + Z. The geodesic functions for the three graph simple geodesics are
GX = tr LXY RXZ = e −Y/2−Z/2 + e −Y /2+Z/2 + e Y /2+Z/2 ,
GY = tr RXX LXZ = e −Z/2−X/2 + e −Z/2+X/2 + e Z/2+X/2 , (2.15)
GZ = tr RXY LXX = e −X/2−Y/2 + e −X/2+Y /2 + e X/2+Y /2 .
Introducing the geodesic function
G̃Z = tr LY RZ RX LZ = e −X/2−Y/2−Z + e X/2−Y /2 ( e −Z + e Z +2)+ e X/2+Y /2+Z ,
obtained from GZ by a Whitehead move, we find that {GX , GY } = G̃Z /2 − GZ /2,

and because relation (2.10) implies that GX GY = GZ + G̃Z , we have
1
{GX , GY } = GX GY − GZ , (2.16)
2
plus the cyclic permutations in X, Y , Z, i.e., the classical Poisson algebra closes in
the algebraic span of the geodesic functions {GX , GY , GZ }.
2.2.4 Poisson geodesic algebras for higher genera. In order to generalize the torus
example, we must find a graph on which graph simple geodesics constitute a convenient
algebraic basis. Such a graph is illustrated in Figure 3, where m edges pairwise connect
two horizontal line segments. Graph simple closed geodesics in this picture are those
and only those that pass through exactly two different “vertical” edges, and they are
therefore enumerated by ordered pairs of edges; we denote the corresponding geodesic
functions Gij where i < j . The Poisson algebra for the functions Gij is described by
⎧
⎪
⎪ 0, j < k,
⎪
⎪
⎪
⎪ 0, k < i, j < l,
⎪
⎪
⎨G G − G G , i < k < j < l,
ik j l kj il
{Gij , Gkl } = 1 (2.17)
⎪
⎪ 2 Gij Gj l − Gil , j = k,
⎪
⎪
⎪
⎪ Gil − 21 Gij Gil , i = k, j < l
⎪
⎪
⎩
Gik − 2 Gij Gkj ,
1
j = l, i < k.
The graph in Figure 3 has genus m2 − 1 and two faces (holes) if m is even and
genus (m − 1)/2 and one face (hole) if m is odd. Such geodesic bases for m even
were considered in [21]. The Poisson algebras of geodesics obtained there coincide
exactly with (2.17). These are the so-called soq (m) algebras whose representations
were constructed in [15].
In the mathematical literature, this algebra has also appeared as the Poisson algebra
of the monodromy data (Stokes matrices) of certain matrix differential equations [38]
and on the symplectic groupoid of upper-triangular matrices G [4]. These matrices
we set Gii ≡ 1) and the entries
have entries given by unity on the main diagonal (i.e.,
Gij above it. For m × m-matrices, there are m2 central elements of this algebra

generated by the polynomial invariants fG (λ) ≡ det(G + λGT ) = fi (G)λi . The

total Poisson dimension d of algebra (2.17) is m(m−1)
2 − m2 , and for m = 3, 4, 5, 6, . . .
we have d = 2, 4, 8, 12, . . . . The dimensions of the corresponding Teichmüller spaces
are D = 2, 4, 8, 10, . . . , so we see that the Teichmüller spaces are embedded as the
Poisson leaves in the algebra (2.17).
i ··· j
b b @@@ B B " "

" "
b b " "
b b @@@ B B " "
b @@ B " "
b @ B " "
b @B " "
b
b" "
" "b b
" " B@ b
" " B @ b
" " @ b bb
" " BB @ B @@@
" " b b
" " B B @@@ b b
j ··· i
Figure 3. The special spine for higher genera.
3 Quantization
A quantization of a Poisson manifold, which is equivariant under the action of a discrete
group D, is a family of ∗-algebras Ah̄ depending on a positive real parameter h̄ with
D acting by outer automorphisms and having the following properties:
1. (Flatness) All algebras are isomorphic (noncanonically) as linear spaces.
2. (Correspondence) For h̄ = 0, the algebra is isomorphic as a D-module to the
∗-algebra of complex-valued functions on the Poisson manifold.
3. (Classical limit) The Poisson bracket on A0 given by {a1 , a2 } = limh̄→0 [a1h̄,a2 ]
coincides with the Poisson bracket given by the Poisson structure of the manifold.
3.1 Quantizing Teichmüller spaces

Here we construct a quantization T h̄ (F ) of the Teichmüller space TH (F ) that is
equivariant with respect to the action of the mapping class group D = MC(F ).
Fix a cubic fatgraph as spine of F , and let T h̄ = T h̄ () be the algebra generated
by Zαh̄ , one generator for each unoriented edge α of , with relations
[Zαh̄ , Zβh̄ ] = 2π i h̄{zα , zβ } (3.1)
(cf. (2.9)) and the ∗-structure
(Zαh̄ )∗ = Zαh̄ , (3.2)
where zα and { ·, ·} denotes the respective coordinate functions and the Poisson bracket
on the classical Teichmüller space. Because of (2.9), the righthand side of (3.1) is a
constant taking only five values 0, ±2π i h̄, and ±4π i h̄ depending upon the coinci-
dences of endpoints of edges labelled α and β.

Lemma 3.1. The center Zh of the algebra T h is generated by the sums α∈I Zαh̄ over
all edges α ∈ I surrounding a given boundary component, and the Poisson structure
is non-degenerate on the quotient T h /Zh .
A standard Darboux-type theorem for non-degenerate Poisson structures then gives

the following result.
Corollary 3.2. There is a basis for T h /Zh given by operators pi , qi , for i =

1, . . . , 6g − 6 + 2s, satisfying the standard commutation relations [pi , qj ] = 2π i h̄δij ,
i, j = 1, . . . , 3g − 3 + s.
Not only is the proof of Lemma 3.1 given in Appendix B, but also an algorithm for
diagonalizing this Poisson structure is described there.
Now, define the Hilbert space H to be the set of all L2 functions in the q-variables
and let each q-variable act by multiplication and each corresponding p variable act
by differentiation, pi = 2π i h̄ ∂p∂ i . For different choices of diagonalization of non-
degenerate Poisson structures, these Hilbert spaces are canonically isomorphic.
Torus example. In the case of the bordered torus, we have three generators, X h̄ ,
Y h̄ , and Z h̄ , the commutation relations (3.1) have the form [Xh̄ , Y h̄ ] = [Y h̄ , Z h̄ ] =
[Z h̄ , Xh̄ ] = 4π i h̄, and the single central element is Xh̄ + Y h̄ + Z h̄ . In the Darboux-
type representation, we can identify, e.g., (X h̄ + Y h̄ )/2 with q and (−X h̄ + Y h̄ )/2
with (−2πi h̄)∂/∂q.
On the level of the modular groupoid as a category where all morphisms are invert-
ible and any two objects are related by a morphism, we have constructed one ∗-algebra
per object. In order to describe the D-equivariance we must associate a homomor-
phism of the corresponding ∗-algebras to any morphism in the modular groupoid. For
this, we associate a morphism of algebras to any Whitehead move and must verify that
the relations in Theorem 2.7 are satisfied.
We now define the quantum Whitehead move or flip along an edge of by Equa-
tion (2.8) using the (quantum) function

h̄ π h̄ e−ipz
φ(z) ≡ φ (z) = − dp, (3.3)
2 sinh(πp) sinh(π h̄p)
where the contour goes along the real axis bypassing the origin from above. The
function (3.3) is Faddeev’s generalization [8] of the quantum dilogarithm.
Proposition 3.3. For each unbounded self-adjoint operator Z on H , φ(Z) is a well-

defined unbounded self-adjoint operator on H.
Proof. The function φ h̄ (Z) satisfies the relations (see [6])

φ h̄ (Z) − φ h̄ (−Z) = Z,
2π i h̄
φ h̄ (Z + iπ h̄) − φ h̄ (Z − iπ h̄) = ,
1 + e −Z
2π i
φ h̄ (Z + iπ ) − φ h̄ (Z − iπ ) =
1 + e −Z/h̄
and is meromorphic in the complex plane with the poles at the points {π i(m + nh̄),
m, n ∈ Z+ } and {−π i(m + nh̄), m, n ∈ Z+ }.
Hence the function φ h̄ (Z) is holomorphic in the strip |Im Z| < π min (1, Re h̄) − ε
for any ε > 0, so we need only its asymptotic behavior as Z ∈ R and |Z| → ∞, for
which we have (see, e.g., [18])

φ h̄ (Z) |Z|→∞ = (Z + |Z|)/2 + O(1/|Z|). (3.4)
Therefore, the function φ h̄ (Z) increases as Z goes to plus infinity and represents an
operator in H by the functional calculus [31].
Theorem 3.4. The family of algebras T h̄ = T h̄ () is a quantization of TH (F ) for

any cubic fatgraph spine of F , that is:
1. in the limit h̄ → 0, morphism (2.8) using (3.3) coincides with the classical
morphism (2.8) where φ(Z) = log(1 + e Z );
2. morphism (2.8) using (3.3) is indeed a morphism of ∗-algebras;
3. a flip WZ satisfies WZ2 = I , cf. (2.8), and flips satisfy the commutativity relation;
4. flips satisfy the pentagon relation.1

Furthermore,
1/h̄
5. the morphisms T h̄ () → T 1/h̄ () given by Zαh̄ → Zα commute with mor-
phisms (2.8).
Sketch of proof. Property 1 follows since limh̄→0 φ h̄ (z) = log( e z +1), and Property 3
is obvious.
In order to prove Property 2, we must first verify that [A+φ h̄ (Z), B−φ h̄ (−Z)] = 0
and [A + φ h̄ (Z), D − φ h̄ (−Z)] = −2π i h̄ (since the other relations are obviously
satisfied), which follows from the identity φ h̄ (z) − φ h̄ (−z) = z.
For Property 5, we must verify that the morphism T h̄ () → T 1/h̄ () commutes
with a flip, that is, (A + φ h̄ (Z))/h̄ = A/h̄ + φ h̄ (Z/h̄), (B − φ 1/h̄ (−Z))/h̄ = B/h̄ −
φ h̄ (−Z/h̄), etc., which follows from φ h̄ (z)/h̄ = φ 1/h̄ (z/h̄).
Turning finally to the most nontrivial Property 4, we may reformulate it as follows.
There are seven generators involved in the sequence of flips depicted in Figure 2 for
the dual cell decomposition, which are denoted A, B, C, D, E, X, Y as in the figure.
As a result of a flip, the piece of graph shown in Figure 1 just gets cyclically rotated.
Denote by Ai , Bi , Ci , Di , Ei , Xi , and Yi the algebra elements associated to the edges
of this piece of graph after i flips are performed. From (2.8), (3.3), these elements
1 This result was independently obtained by R. M. Kashaev [16].
evolve as follows:
⎛ ⎞ ⎛ ⎞
Xi+1 Yi − φ h̄ (−Xi )
⎜ ⎟ ⎜ ⎟
⎜ Yi+1 ⎟ ⎜ −Xi ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ Ai+1 ⎟ ⎜ Di ⎟
⎜ ⎟ ⎜ ⎟
⎜ Bi+1 ⎟ −→ ⎜ E ⎟. (3.5)
⎜ ⎟ ⎜ i ⎟
⎜ ⎟ ⎜ ⎟
⎜ Ci+1 ⎟ ⎜ Ai + φ h̄ (Xi ) ⎟
⎜ ⎟ ⎜ ⎟
⎜D ⎟ ⎜B − φ h̄ (−X )⎟
⎝ i+1 ⎠ ⎝ i i ⎠
Ei+1 Ci + φ h̄ (Xi )
We must prove that this operator is periodic with period five.

Assume for a moment that this five-periodicity of Xi has been established. Then
five-periodicity of Yi and other variables follow from simple calculations.
It suffices to prove this five-periodicity for the Xi since the five-periodicity of the
other operators such as Yi then follows from elementary calculations. As to the five-
periodicity of Xi , let us “take logarithms” and introduce four new algebra elements
Ui = e Xi ; Vi = e Yi ; Ũi = e Xi /h̄ ; Ṽi = e Yi /h̄ ,
which satisfy the following commutation relations
Ui Vi = q −2 Vi Ui , Ũi Ṽi = q̃ −2 Ṽi Ũi ,

(3.6)
Ui Ṽi = Ṽi Ui , Vi Ũi = Ũi Vi ,
where
q = e −πi h̄ , q̃ = e −πi/h̄ .
Under the flip, these variables are transformed in an especially simple way,
Ui−1 = Vi−1 , (3.7)

Vi−1 = Ui (1 + qVi ), (3.8)
Ũi−1 = Ṽi−1 , (3.9)
Ṽi−1 = Vi (1 + q̃ Ṽi ). (3.10)
As the first step of the proof, we consider the inverse transformation laws for Xi and Yi :
Xi−1 = −Yi ; Yi−1 = Xi + φ h̄ (Yi ).
Equations (3.7) and (3.9) are obvious. Using the standard formula
1
B+[A,B]
F (z)dz
e A+F (B) = e [A,B] B e A,
we obtain
h̄
Y +2π i h̄ h̄
Vi−1 = e Yi−1 = e Xi +φ (Yi ) = eXi e Y φ (z)dz

π h̄ e−ipz (e−2pπ h̄ − 1)
= Ui exp − dp
2 (−ip) sinh(πp) sinh(π h̄p)
−ip(z−πi h̄)
π h̄ e
= Ui exp dp
i p sinh(πp)
= Ui (1 + q −2 Vi ).
The proof of (3.10) is analogous.
Now, in order to finally prove that Xi is five-periodic, it suffices to verify that both
Ui and Ũi are five-periodic. Indeed, if only the operator Ui is five-periodic, it does
not suffice because the logarithm of an operator is ambiguously defined. However, if
we have two families of operators U and Ũ , which depend continuously on h̄, then,
assuming the existence of an operator X (depending continuously on h̄) such that
U = eX and Ũ = eX/h̄ , then this operator is evidently unique. (It can be found as
lim(m+n/h̄)→0 (U m Ũ n )/(m + n/h̄) for any irrational value of h̄.) The five-periodicity
of sequence (3.8) (and (3.10)) is a direct calculation using (3.6).
The only subtlety remaining is the possibility that some of the edges A, B, C, D, E
coincide. (Note, however, that X and Y must have exactly one common vertex.) If,
say, edges A and C coincide, then the value of the commutator [A, X] is doubled by
definition, and we can then fictitiously split the edge A = C into two half-edges with
the matrices XA/2 assigned to each half-edge using the formula

0 1
XA = XA/2 XA/2 . (3.11)
−1 0
The quantities A/2 = C/2 have the same commutation relations with the rest of
variables as well as the same transformation laws as the quantities A and C before,
so the earlier formulas remain valid if we simply replace there A by A /2 and A by
A/2. The net effect is that commutators with A = C are doubled.
If edges A and B coincide, then we must correct formulas (3.4) using a splitting as
above but demanding A /2 = A/2+X/2. Obviously, [A, X] = 0 in this case (in which
formulas for the quantum ordering below must be also corrected, see Section 3.5).
The formulas in Proposition 2.3 can thus be realized for exponentiated quantities,
although in the current quantum case there will be corrections. For instance, formulas
(3.8) and (3.9) will be different; indeed, letting A = C = X, we calculate that
h̄
eX/2 → eX/2+φ (Z) =: eX/2 (1 + eZ ) :≡ eX/2 + eX/2+Z = eX/2 (1 + e−iπ h̄ eZ ) since
the commutator of X and Z is doubled, where the normal ordering (the Weyl ordering)
: · : is explained in the next section. The transformation for eX itself becomes more
complicated when A = C, namely, eX = (eX/2 )2 → eX (1 + q 3 eZ )(1 + qeZ ).
Corollary 3.5. 1. Let K be an operator acting in the Hilbert space L2 (R) and having
the integral kernel
zx
K(x, z) = F h̄ (z)e− 2π i h̄ , (3.12)
where
h̄ 1 e−ipz
F (z) = exp − dp (3.13)
4 p sinh(πp) sinh(π h̄p)
Then the operator K is unitary up to a multiplicative constant and satisfies the identity
K 5 = const. (3.14)
2. Let h̄ = m/n be a rational number and assume that both m and n are odd.
Introduce a linear operator L(u) acting in the space Cn and depending on one positive
real parameter u through its matrix
L(u)ji = F h̄ (j, u)q −4ij , (3.15)
where
j
−1
F h̄ (j, u) = (1 + u)j/n (1 + q −4k+2 u1/n )−1 .
k=0
Then the following identity holds:
L(u)L(v + uv)L(v + vu−1 + u−1 )L(u−1 v −1 + u−1 )L(v −1 ) = 1. (3.16)
Using this construction, Kashaev [18] constructed the set of eigenfunctions of the
quantum Dehn twist transformation. Namely, Kashaev’s dilogarithm function eb (z) is
1/F h̄ (2z) from (3.13). We return to this discussion when considering sets of quantum
Dehn twists for the torus in Section 5.
3.2 Geodesic length operators

We next embed the algebra of geodesics (2.6) into a suitable completion of the con-
structed algebra T h̄ . For any γ , the geodesic function Gγ (2.6) can be expressed in
terms of shear coordinates on TH :

1
Gγ ≡ tr PZ1 ...Zn = exp mj (γ , α)zα , (3.17)

2
j ∈J α∈E()
where mj (γ , α) are integers and J is a finite set of indices. In order to find the
quantum analogues of these functions, we denote by Tˆ h̄ a completion of the algebra
T h̄ containing exZα for any real x.
For any closed path γ on F , define the quantum geodesic operator Gh̄γ ∈ Tˆ h̄ to be

1

Gh̄γ ≡ ×× tr PZ1 ...Zn ×× ≡ exp mj (γ , α)Zαh̄ + 2π i h̄cjκ (γ , α) , (3.18)
2
j ∈J α∈E()
κ∈{j }
where the quantum ordering ×× · ×× implies that we vary the classical expression (3.17) by
introducing additional integer coefficients cjκ (γ , α), which must be determined from
1
the conditions below. Notice that the operators {Gγh̄ } themselves can be considered
as belonging to the algebra Tˆ h̄ insofar as

1
1

Gγh̄ = exp mj (γ , α)Zαh̄ + 2π icjκ (γ , α) . (3.19)
2h̄
j ∈J α∈E()
In what follows for the notational simplicity, we shall sometimes omit the su-
perscript h̄ from Gh̄ and write it merely G assuming that G is either an operator or a
classical geodesic function depending on the context. We shall concentrate on the case
of quantum functions of the Gh̄ -type; the consideration of the sector G1/h̄ is analogous
and does not lead to new effects (at least at the present stage of understanding), so we
omit it. We also call a quantum geodesic function merely a quantum geodesic (since
implicitly quantum objects admit only functional, not geometrical, descriptions).
We wish to associate an operatorial quantum multicurve QMC to a multiset Ĉ
of quantum geodesics corresponding to disjointly embedded families of nonnegative
integrally weighted geodesics. One ansatz will be that operators corresponding to
disjoint underlying geodesics must commute; this implies that the ordering in which
the quantum geodesics enter the product QMC is immaterial, where the product is
defined as for GMCs. We next formulate the defining properties of quantum geodesics.
1. If closed paths γ and γ do not intersect, then the operators Gh̄γ and Gh̄γ commute.
2. Naturality. The mapping class group MC(F ) (2.8) acts naturally, i.e., for any
{Gh̄γ }, δ ∈ MC(F ) and closed path γ in a spine of F , we have δ(Gh̄γ ) = Gh̄δγ .
3. Geodesic algebra. The product of two quantum geodesics is a linear combination
of QMC’s governed by the (quantum) skein relation [37].
4. Orientation invariance. Quantum traces of direct and inverse geodesic operators
coincide.
5. Exponents of geodesics. A quantum geodesic Gnγ corresponding to the n-fold
concatenation of γ is expressed via Gγ exactly as in the classical case, namely,

Gnγ = 2Tn Gγ /2 , (3.20)
where Tn (x) are Chebyshev’s polynomials.
1
6. Duality. For any γ and γ , the operators Gh̄γ and Gγh̄ commute.
We shall let the standard normal ordering symbol : e a1 e a2 . . . e an : denote the Weyl
ordering e a1 +···+an , i.e.,
1
: e a1 e a2 . . . e an : = 1+(a1 +· · ·+an )+
(a1 +· · ·+an )(a1 +· · ·+an )+· · · (3.21)
2!
for any set of exponents with ai = −aj for i = j . In particular, the Weyl ordering
implies total symmetrization in the subscripts.
Proposition 3.6. For any graph simple geodesic G with respect to any spine , the
coefficients cjκ (γ , α) in (3.18) are identically zero.
Proof. Consider term-by-term the trace of the matrix product for the quantum graph
simple geodesic and expand it in Laurent monomials in e Zi /2 . It is easy to see that
each term e Zi /2 comes either in power +1, or −1 in the corresponding monomial
and there are no equivalent monomials in the sum. This means that in order to have a
Hermitian operator, we must apply the Weyl ordering with no additional q-factors (by
the correspondence principle, each such factor must be again a Laurent monomial in q
standing by the corresponding term, which breaks the self-adjointness unless all such
monomials are unity). Since quantum Whitehead moves must preserve the property
of being Hermitian, if a graph-simple geodesic transforms to another graph-simple
geodesic, then a Weyl-ordered expression transforms to a Weyl-ordered expression,
and only these expressions are self-adjoint.
Torus example. For the torus with one hole, there are three graph simple quantum
geodesics for any spine, which are exactly (2.15) in the Weyl-ordered form. The
quantum geodesics G̃Z obtained from GZ by the flip transformation is
G̃Z = e −X/2−Y/2−Z + e X/2−Y/2−Z + e X/2−Y /2 · 2 cos(π h̄)

(3.22)
+ e X/2−Y/2+Z + e X/2+Y /2+Z .
The product of two graph simple quantum geodesics is
GX GY = e iπ h̄/2 G̃Z + e −iπ h̄/2 GZ . (3.23)
Denoting q ≡ e −iπ h̄ , [A, B]q ≡ q 1/2 AB − q −1/2 BA, and ξ = q − q −1 , we

obtain from (3.23)
[GX , GY ]q = ξ GZ , [GY , GZ ]q = ξ GX , [GZ , GX ]q = ξ GY . (3.24)
This algebra is exactly the soq (3) quantum algebra studied in [15]. There is a unique
central element, the quantum Markov relation
M = GX GY GZ − q 1/2 (G2X + q −2 G2Y + G2Z ). (3.25)
3.3 Algebra of quantum geodesics
Let G1 and G2 correspond to the respective graph simple geodesics with respect to
the same spine having one nontrivial intersection. For G1 and G2 , formula (3.18)
implies, by virtue of Proposition 3.6, the mere Weyl ordering.
After some algebra, we obtain (cf. (2.14))
G1 G2 = e −iπ h̄/2 GZ + e iπ h̄/2 G̃Z , (3.26)

where GZ coincides with the Weyl-ordered GI in the classical case (cf. (2.14)) while
G̃Z contains the quantum correction term
G̃Z = ×× tr 1 tr 2 . . . (eij
1
⊗ ej2i )[XZ1 ⊗ XZ2 ] . . . ××

= : tr 1 tr 2 . . . (eij
1
⊗ ej2i ) XZ1 ⊗ XZ2 + 2(1 − cos π h̄)e11
1
⊗ e22
2
... : .
Here eij
1 ⊗ e2 is the standard r-matrix that permutes the spaces “1” and “2,” and as
ji
a result, the “skein” relation of form (2.14) appears. Locally, this relation has exactly
the form proposed by Turaev [37], i.e., for two graph simple geodesics intersecting at
a single point, we have the defining relation
GZ G̃Z
G1
@
@ = e −iπ h̄/2 +e iπ h̄/2 .
@
G2 @
(3.27)
(The order of crossing lines corresponding to G1 and G2 depends on which quantum
geodesic occupies the first place in the product; the rest of the graph remains unchanged
for all items in (3.27)). Note, however, that if the quantum geodesics G1 and G2
correspond to graph simple geodesics, we may turn the geodesic G̃Z again into the
graph simple geodesic G̃Z by performing the quantum flip with respect to the edge Z.
If we now compare two unambiguously determined expressions: G̃Z , which must
be Weyl ordered, and G̃Z obtained from the geodesic algebra, we find that G̃Z = G̃Z .
This enables us to formulate the main assertion.
Lemma 3.7 ([6]). There exists a unique quantum ordering ×× . . . ×× (3.18), which is
generated by the quantum geodesic algebra (3.27) and consistent with the quantum
mapping class groupoid transformations (2.8), i.e., so that the quantum geodesic
algebra is invariant under the action of the quantum mapping class groupoid.
We can now relax the constraints of graph simplicity of curves: as the quantum
geodesic algebra is quantum mapping class group invariant, having two arbitrary
embedded geodesics with a single intersection, we can transform them using quantum
morphisms to a canonical form of graph simple geodesics and employ the Weyl order.
Relation (3.27) remains valid in both cases.
Let us now address the problem of multiple intersections. Here, we have the
following lemma.
Lemma 3.8 ([6]). If more than one intersection of two QMCs occurs, then the quan-
tum skein relations (3.27) must be applied simultaneously at all intersection points.
This lemma implies the standard Reidemeister moves for curves on a graph, where
the empty loop gives rise to a factor − e −iπ h̄ − e iπ h̄ ; that is, for geodesics inter-
secting generically, apply (3.27) simultaneously at all intersection points to obtain the
Reidemeister moves.
Remark 3.1. The quantum algebra M12 was studied in [6], where the exact corre-
spondence with Kauffman bracket skein quantization of the corresponding Poisson
algebra of geodesics (see [5]) was observed. In [4], this algebra arises as the quantum
deformation algebra of the classical groupoid Poisson relations for the group SL(4).
This algebra is however only one among many quantum Nelson–Regge algebras cor-
responding to Riemann surfaces of higher genera which are described in the next
section.
3.4 Quantizing the Nelson–Regge algebras

The algebra (2.17) was quantized by the deformation quantization method in [21],
[20]. We now explicitly implement the quantization conditions (3.1). It is convenient
to represent the (classical or quantum) elements Gij as chords connecting the points of
the cyclically ordered set of indices i, j . There is then a trichotomy: if two chords do
not intersect, then the corresponding geodesics do not intersect either, and the quantum
geodesics commute (Figure 4 (a)); if two chords have a common endpoint, then the
corresponding geodesics intersect at one point, and the three quantum geodesics Gh̄ij ,
Gjh̄k , Gh̄ki (as depicted in Figure 4 (b)) constitute the quantum subalgebra soq (3); if two
chords intersect at an interior point (as depicted in Figure 4 (c)), then the corresponding
geodesics intersect in two points, and the corresponding quantum geodesics Gh̄ij and
Gh̄kl , i < k < j < l, satisfy the commutation relation
[Gh̄ij , Gh̄kl ] = ξ(Gh̄ik Gjh̄l − Gh̄il Gjh̄k ) (3.28)
with the usual commutator (not the q-commutator) and where again ξ = q − q −1 .
j j
't $ ' t $ i
' k
A t t$
tk A @
A q q
@
i t A q @ q
t
At t @t
t
& % i& % k & %
l j
l (a) (b) (c)
Figure 4. Picture of Nelson–Regge quantum relations.
3.5 Improving the quantum ordering

We now extend the construction of QMCs by taking the products of operatorial matrices
XZ , L, and R along oriented geodesics as before, but we do not apply the trace
operation as in the definition of the geodesic operators.
For the three cases of oriented curves depicted in Figure 5 below, we may apply
the indicated quantum Whitehead move and calculate as follows:
For curve 1, we obtain
XB LX−Z LXA

e −B /2 e −Z/2 e A /2 + e −B /2 e Z/2 e A /2 − e −B /2 e Z/2 e −A /2
=
e B /2 e −Z/2 e A /2 0
1
e A/2−B/2 − e −A/2−B/2− 4 [A,B]
= 1 for B = A (3.29)
e A/2+B/2− 4 [A,B] 0

1 e −B/2 e A/2 − e −B/2 e −A/2
= e − 8 [A,B] for B = A
e B/2 e A/2 0
1
e − 8 [A,B] XB LXA for A = B,
=
XB LXA for A = B.
XC RX−Z LXA

e −C /2 e Z/2 e A /2 − e −C /2 e Z/2 e −A /2
=
e C /2 e Z/2 e A /2 + e C /2 e −Z/2 e A /2 − e C /2 e Z/2 e −A /2
1
e A/2−C/2+Z/2+ 4 [A,B] − e −A/2−C/2+Z/2 − e −A/2−C/2−Z/2
= 1 (3.30)
e A/2+C/2+Z/2 − e −A/2+C/2+Z/2− 4 [A,B]

e −C/2 e Z/2 e A/2 − e −C/2 e Z/2 e −A/2 − e −C/2 e −Z/2 e −A/2
=
e C/2 e Z/2 e A/2 − e C/2 e Z/2 e −A/2
= XC LXZ RXA ,
and if edges A and C coincide, we merely use the same splitting as in (3.11); no
additional factors arise.
XC RX−Z RXD

0 − e −C /2 e Z/2 e −D /2
=
e C /2 e −Z/2 e D /2 − e C /2 e Z/2 e −D /2 − e −C /2 e −Z/2 e −D /2
1
0 − e −C/2−D/2+ 4 [A,B]
= 1 for C = D (3.31)
e C/2+D/2+ 4 [A,B] − e C/2−D/2

1
[A,B] 0 − e −C/2 e −D/2
= e8 for C = D
e C/2 e D/2 − e C/2 e −D/2
1
e 8 [A,B] XC RXD for C = D,
=
XC RXD for C = D.
In these formulas, we have used the identities
1
e A /2+C /2+Z/2 + e A /2+C /2−Z/2 = : e A/2+C/2 ( e Z/2 + e −Z/2 ) :
1 + e −Z (3.32)
= e A/2+C/2+Z/2
and

e −A /2−C /2+Z/2 = : e −A/2−C/2 (1 + e −Z ) e Z/2 :
(3.33)
= e −A/2−C/2+Z/2 + e −A/2−C/2−Z/2 ,
where vertical dots denote the Weyl ordering (3.21) as before implying total sym-
metrization with respect to all the variables {A, B, C, D, Z}.
As we have just observed and amazingly enough, the only thing that changes under
a quantum Whitehead move is the overall factor standing by the product of matrices,
and even this factor can easily be taken into account if included with each left-turn
1
matrix L is an overall factor e − 8 [A,B] , and included with each right-turn matrix R is
1
an overall factor e 8 [A,B] .
The subtlety of potentially multiple intersections between curves corresponds to
the possibility that edges of the graph may coincide. As in the previously useful
convenient fiction of splitting such an edge into two half-edges, the formulas (3.29),
(3.30), and (3.31) remain valid if we replace there A /2 by A /4 and A/2 by A/4 in
case A = C for instance.
We thus have the quantum analogue of Lemma 2.4.
Lemma 3.9. For any oriented non boundary-parallel geodesic γ , taking the finite
(periodic) sequence of matrices with quantum entries in XZi as before but making the
replacement
L̃ = q −1/4 L, R̃ = q 1/4 R, (3.34)
where q = e −iπ h̄ , the resulting product of matrices

Pγ = XZn L̃XZn−1 R̃ . . . XZ3 L̃XZ2 R̃XZ1 R̃ (3.35)
is invariant under quantum Whitehead moves.
Remark 3.2. In the case of a boundary-parallel curve, where a sub-word of the form
X R̃X R̃ transforms to a sub-word of the form X R̃, ˜ we must set the resulting
D −Z D
turn matrices to be
L̃˜ = q −1/2 L, R̃˜ = q 1/2 R,
i.e., they have quantum factors doubled in comparison with L̃ and R̃.
We now address the question of the proper quantum ordering. Using mapping
class group transformations, we can reduce any simple curve either to the form of
a graph simple curve with exactly one left and one right turn (c.f. formulas (2.15)
and Figure 5) if this curve is not boundary-parallel, or to the form X R̃˜ or X L̃˜
D D
for a boundary-parallel curve. In both cases, each term in the corresponding Laurent
polynomial must be self-adjoint, which immediately results in the Weyl ordering by
Proposition 3.6.
1 A B
2 @
@
- 1 A B
@ 2 @ −Z
@ @
@ -
? Z
- @ -
- @
@
@ @
- @ @
3 D C
@
3 @
D C
Figure 5. Three cases of flips for geodesics.
This observation does not suffice to derive the proper quantum ordering in our
quantization of Thurston theory since we cannot consider in a consistent way an
infinite product of matrices corresponding to infinite leaves of a measured foliation.
Our tools for analyzing the Thurston theory in this case will devolve to naturality
of the mapping class group action on the QMC algebra and an operatorial version of
infinite continued fractions in Section 5.3.4. On the other hand, the improved quantum
ordering is used in our analysis of closed geodesics on the torus, i.e., of operatorial
finite continued fraction expansions in Section 5.3.1.
4 Classical Thurston theory of surfaces

Let Fgs denote an oriented smooth surface with s ≥ 0 punctures (so Fgs may be
closed without boundary in this section), with genus g ≥ 0, and with negative Euler
characteristic 2 − 2g − s < 0.
4.1 Measured foliations and Thurston’s boundary

Define a measured foliation on F = Fgs to be a one-dimensional topological foliation
F of F , where in a neighborhood of any p ∈ F , F must restrict (in an appropriate
chart) to the horizontal foliation as in Figure 6 (a) or it must restrict to a foliation with
one n-pronged singularity at p, for n ≥ 3, as illustrated in Figure 6 (b).
(a) (b)
Figure 6. Pictures of foliations.
Furthermore, F comes equipped with a transverse measure μ, which assigns to any

arc a in F that is transverse to F a real number μ(a) ∈ R≥0 , where μ is required to
satisfy:
No holonomy. If a0 , a1 are homotopic through arcs at transverse to F , for 0 ≤ t ≤ 1,
keeping the endpoints of at on the same leaf for all t, then μ(a0 ) = μ(a1 ).
σ
-Additivity. If a is the serial concatenation of transverse arcs a1 , a2 , . . . , then μ(a) =
i≥1 μ(ai ).
In other words, in the neighborhood of a non-singular point, there is a local chart

φ : U → R2 = {(x, y) : x, y ∈ R} so that φ −1 ({y = constant}) are the leaves of the
foliation F ∩U . If two charts Ui and Uj intersect, then the transition function φij are of
the form φij (x, y) = (hij (x, y), cij ± y), where cij is constant. In these coordinates,
the transverse measure is |dy|. In case the transition functions can be chosen with
constant sign φij (x, y) = (hij (x, y), c + y), i.e., if the foliation is “transversely
orientable”, then (away from the singular points) y is the primitive of a closed one-
form on F .
Another canonical construction of a measured foliation on a Riemann surface is
given by taking the leaves of the foliation to be the level sets of a harmonic function,
where the transverse measure is given by integrating the conjugate differential along
transverse arcs. Still another example is given by the homotopy class of a finite
collection of disjointly embedded (weighted) curves, as we shall see.
There is an equivalence relation on measured foliations generated by isotopy and
Whitehead collapse as illustrated in Figure 7, and the set of all equivalence classes
(including the empty measured foliation ∅) is denoted MF = MF (F ), where the
class of (F , μ) is denoted [F , μ]. To naturally topologize MF , we introduce the
discrete set S = S(F ) consisting of all free homotopy classes [c] of simple closed
curves c in F which are neither null homotopic nor puncture-parallel. We shall also
require the set S = S (F ) consisting of homotopy classes of all disjointly embedded
families of curves in F , where each component of the family lies in S and no two
components are homotopic.
@
@
- @
@
@
@
Figure 7. Whitehead collapses on foliations.
One can show [9] that for any [c] ∈ S, there is a representative cF ∈ [c] which
minimizes the μ-transverse measure in its homotopy class; furthermore, given equiv-
alent measured foliations (Fi , μi ), for i = 1, 2, we have μ1 (cF1 ) = μ2 (cF2 ) [9], and
hence there is a well-defined mapping
J : MF → RS≥0 ,

(F , μ) → i(F ,μ) : [c] → μ(cF ) ,
where the empty measured foliation ∅ is identified with 0 = {0}S . This mapping J is
an injection [9] and induces a topology on MF (where a neighborhood of ∅ is home-
omorphic to a cone from ∅ over MF /R>0 , with the natural action by homothety of
R>0 on measures). The function i(F ,μ) is called the (geometric) intersection function
of the measured foliation (F , μ).
Given [c] ∈ S, there is a corresponding measured foliation defined as follows.
Choose a representative c of [c] and collapse F − c onto a spine (making further
choices) to build a foliation Fc of F , whose leaves either lie in the spine or are
homotopic to c, and choose a transverse measure μc on Fc that pulls back under the
collapsing map the counting measure δc on c. It is a classical fact due to Whitehead [9]
that the resulting Whitehead equivalence class [Fc , μc ] is well-defined independent
of any choices. Taking the projective class of this foliation, we may thus regard
S ⊆ P F 0.
Furthermore by construction, i[Fc ,μc ] ([d]), for [d] ∈ S, is just the geometric intersec-
tion number of [c] and [d], i.e., the total number of intersections of representatives c
with d, where c and d intersect minimally. More generally, given a family of curves
c1 , . . . , cn representing a point of S , together with a collection
w1 , . . . , wn ∈ R>0
of “weights”, we may again collapse to a spine of F − {ci }n1 to produce a foliation
F and choose a measure μ on F that pulls back the weighted sum n1 wi δci to get
a well-defined Whitehead equivalence class [F , μ]. As mentioned before, one may
thus associate a measured foliation to a weighted curve family.
Of special interest is the subspace MF 0 = MF 0 (F ) consisting of all measured

foliations (F , μ) of compact support, i.e., any leaf of F with a transverse arc a so that
μ(a) > 0 must be disjoint from a neighborhood of the punctures and no such leaf is
puncture parallel. There is a natural R>0 -action on MF − {∅} and MF 0 − {∅} given
by scaling the transverse measure, and the corresponding quotients
P F = P F (F ) = (MF (F ) − {∅})/R>0 ,
P F 0 = P F 0 (F ) = (MF 0 (F ) − {∅})/R>0
are the spaces of central interest in the sequel. The projective class of (F , μ) and
[F , μ], respectively, will be denoted (F , μ̄) and [F , μ̄].
A point of T = T (Fgs ) = Tgs may be regarded as the class of a hyperbolic metric on
F , i.e., a complete finite-area Riemannian metric on F of constant Gauss curvature −1.
We shall also require the “Yamabe space” Y = Y(F ) of all complete finite-area
Riemannian metrics on F of constant Gauss curvature −x 2 , for some x ∈ R>0 . Y is
canonically homeomorphic to Tgs × R>0 , where (ρ, x) corresponds to the class of the
metric xρ, and we let π : Y → T denote the projection onto the first factor. Define
the map
I : Y → RS≥0 ,
ρ → ρ ( · ),
where ρ ([c]) is the ρ-length of the unique ρ-geodesic in the homotopy class [c].
Thus, if ρ ∈ Y corresponds to (π(ρ), x), then I (ρ) = I (π(ρ), x) = x I (π(ρ), 1).
The basic facts [9] are that I : Y → RS≥0 and J : MF 0 (F ) → RS≥0 are embeddings
with disjoint images, and we may define a completion Y of Y in RS≥0 by setting
Y = I (Y) ∪ J (MF 0 )
and identifying Y with I (Y). Passing to quotients under the homothetic actions of R>0
on Y − {0},
on RS≥0 − {0}, and on MF 0 − {∅}, we obtain Thurston’s compactification
s
T g = Tgs ∪ P F 0 ≈ Y − {0} /R>0
R>0 by P F 0 ≈ (MF 0 − {∅})/R>0 .

of T ≈ (Y − {0})/
Theorem 4.1. 1. ([35], [9], [24]) P F 0 (Fgs ) is naturally a piecewise linear sphere of
s
dimension 6g − 7 + 2s which compactifies Tgs to produce a closed ball T g .
2. ([35], [26]) The action of the mapping class group MCsg on Tgs extends continu-
s
ously to an action on T g , where the action on P F 0 (Fgs ) is the natural one, and there
are explicit piecewise linear formulas for the action of Dehn twist generators.
3. ([35], [9]) Suppose that a sequence of hyperbolic metrics ρi on F tends to
a point [F , μ̄] ∈ P F 0 . In the projectivization of RS≥0 − {0}, the projectivized
length functions ¯ρi of ρi converge to the projectivized intersection function ī(F ,μ̄)
of (F , μ̄).
4. ([35], [1]) The function i(F ,μ) : S → R≥0 extends continuously to the “geomet-
ric intersection pairing” MF 0 × MF 0 → R≥0 , that vanishes on the diagonal and
which is also invariant under MCsg .
5. ([35], [23], [3]) The Weil–Petersson Kähler two-form on Tgs continuously extends
(in the appropriate sense on Yamabe space) to a non-degenerate symplectic form,
called “Thurston’s symplectic form”, on MF 0 (Fgs ) (≈ Y − Y), which is invariant
under MCsg .
Though there is all this beautiful natural structure on Thurston’s boundary, the quo-
tient P F 0 (Fgs )/ MCsg is maximally non-Hausdorff (i.e., its largest Hausdorff quotient
is a singleton) as we shall see, so there is no correspondingly nice Thurston compact-
ification on the level of Riemann’s moduli space.
Torus example. Recall that for the once-punctured torus F = F11 , the Teichmüller
space is T11 ≈ {z ∈ C : |z| < 1}, and the mapping class group is MCsg ≈ PSL2 (Z).
Indeed, the right Dehn twists M and L on the meridian and longitude, respectively,
generate MC11 , and a complete list of relations between them is given by ι = MLM =
LML and ι2 = 1.
A point of P F 0 (F ) is uniquely determined by its “slope”, defined as follows. Fix
two disjointly embedded ideal arcs x, y asymptotic to the puncture p which decompose
F into an ideal quadrilateral, where x ∪ {p} is homotopic to the meridian, and y ∪ {p}
is homotopic to the longitude. Given [F , μ] ∈ MF 0 (F11 ) − {∅}, compact support
guarantees that i[F ,μ] (x) and i[F ,μ] (y) are well-defined and finite, and the ratio |θ| =
i[F ,μ] (y)/ i[F ,μ] (x) ∈ [0, ∞] is therefore projectively well-defined; we further imbue
θ with a sign (when it is finite and non-zero) in the natural way, where the sign is positive
if one (in fact, any) leaf of F immediately after meeting x then meets the copy of y in
the frontier of the ideal quadrilateral which lies to the right (where the orientation of the
ideal quadrilateral is inherited from that of F11 ). It is easy to see that the slope θ is a well-
defined and complete invariant of [F , μ̄] ∈ P F 0 (F11 ), where we regard θ ∈ S 1 in the
1
natural way. Thus, P F 0 (F11 ) ≈ S 1 , and T 1 = T11 ∪P F 0 (F11 ) is a closed ball (which
you should not identify with the Poincaré disk together with its circle at infinity). The
slope θ = p/q is rational if and only if the measured foliation corresponds to the
simple closed curve wrapping p times around the meridian and q times around the
longitude, where p and q are relatively prime integers. Geometrically, deforming
hyperbolic structure to pinch this curve, it is clear from elementary considerations
of hyperbolic geometry that the corresponding geodesic length functions converge
projectively to the geometric intersection number with this curve. Furthermore, the
geometric intersection number of curves with slopes p/q, r/s written in least terms
is given by |ps − qr|.
4.2 Train tracks

A train track τ ⊆ F is a graph (where vertices are called “switches” and edges are
called “branches”) together with the following extra structure:
Smoothness. τ is C 1 away from its switches. Furthermore, for each switch v of τ , there
is a tangent line to τ at v in the tangent plane to F at v so that for each half-branch
whose closure contains v, the one-sided tangent at v lies in .
Non-degeneracy. Vertices of τ are at least trivalent, and for any switch v of τ , there is
an embedding (0, 1) → F with f 21 = v which is C 1 as a map into F .
Geometry. Suppose that C is a component of F − τ , and let D(C) denote the double
of C along the C 1 frontier edges of C so the non-smooth points in the frontier of C
give rise to punctures of D(C). We demand that the Euler characteristic of D(C) be
negative.
The smoothness condition is synonymously called the structure of a “branched one-
submanifold” and leads to the fundamental notion of a graph smoothly supporting a
curve or another train track as we shall see. According to the non-degeneracy condition,
the half-branches incident on a fixed switch decompose canonically into two non-
empty sets of “incoming” and “outgoing” branches. The geometric condition rules
out the following complementary regions: smooth disks (i.e., nullgons), monogons,
bigons, smooth annuli, and once-punctured nullgons, and will be further explained
below.
Let B(τ ) denote the set of branches of τ . A function μ : B(τ ) → R≥0 induces
μ : {half-branches of τ } → R≥0 in the natural way (where μ(b 1 ) = μ(b) if b 1 ⊆ b)
2 2
and satisfies the switch conditions provided that for each switch v of τ , we have

μ(b) = μ(b).
outgoing incoming
half-branches b half-branches b
Such a function satisfying the switch conditions is called a transverse measure on τ ,

and τ itself is said to be recurrent if it supports a positive measure μ with μ(b) > 0
for each branch of τ . In the sequel, train tracks will tacitly be assumed to be recurrent.
Torus example. There is a unique combinatorial type of recurrent trivalent train track
τ in the surface F = F11 , and two embeddings of it as spine are illustrated in Figure 8.
F − τ consists of a single once-punctured bigon. There are two branches of τ so
that μ is uniquely determined by its values on these branches, and the weight on
the remaining branch of τ is given by their sum according to the switch condition.
Notice that this train track is “orientable” in the sense that the graph underlying τ
admits an orientation where incoming points toward outgoing at each vertex. Thus,
fixing an orientation on τ , a measure μ on τ uniquely determines a homology class
in H1 (F11 , R). Equivalently, every measured foliation of compact support on F11 is
transversely orientable.
Construction 4.1. Given a train track τ and a positive measure μ on it, we may
construct a measured foliation of a neighborhood of τ in the following way. For each
branch b of τ , take a rectangle of width μ(b) and length unity foliated by horizontal
leaves. For each switch, place the rectangles of the incoming branches next to one
another and likewise for the outgoing branches, and then finally glue the vertical edges
of all the incoming to all the outgoing rectangles at each switch in the natural way
preserving the transverse measure along the widths of the rectangles by the switch
conditions. This produces from μ a measured foliation of a tie neighborhood of τ ,
where a vertical leaf in any rectangle is called a tie, and the singular ties arise from
the vertical sides of the rectangles. As before by Whitehead’s result, the Whitehead
equivalence class of the resulting measured foliation is well-defined.
If a measured foliation arises in this way from a measure on a train track, then we
say that the train track carries the measured foliation.
B(τ )
Let U (τ ) denote the cone of all measures on τ , i.e., the subspace of R≥0 deter-
mined by the switch conditions. There is again the natural R>0 -action on U (τ ) − {0}

by homothety, and the quotient V (τ ) = (U (τ ) − {0})/R>0 is the polygon of projective
measures on τ . Construction 4.1 thus gives well-defined maps U (τ ) → MF 0 (F and
V (τ ) → P F 0 (F ).
A recurrent train track is maximal if it is not a proper sub track of any recurrent train
track. For general F = Fgs , complementary regions to a maximal train track are either
trigons or once-punctured monogons, but in the special case of the once-punctured
torus, a maximal train track has a single complementary once-punctured bigon.
Theorem 4.2 ([35], [24]). For any maximal recurrent train track τ in F , Construc-
tion 4.1 determines continuous embeddings
U (τ ) → MF 0 (F ) and V (τ ) → P F 0 (F )
onto open sets.
In fact, the geometric condition in the definition of train track precisely guarantees
the injectivity in this theorem. In light of this result, one may regard a maximal train
track in F as indexing a chart on the manifold P F 0 , and we next study the transition
functions of this putative manifold structure.
Torus example. For F = F11 , two embeddings of train tracks as spine are illustrated
in Figure 8, and in fact, every foliation is carried by one of these two train tracks.
The corresponding charts on the circle are also illustrated as well as the two points of
intersection in the closures of these charts.
It is most convenient now to restrict to the “generic” case, where all switches of τ
are trivalent. For each switch of τ , the decomposition of incident half-branches into
incoming/outgoing thus consists of one singleton and one doubleton, and we say a
branch of τ is large if it is a singleton at both its endpoints (which are then necessarily
distinct) as illustrated with the branch labeled e in the left-hand side of Figure 9.
Likewise, if a branch is a doubleton at both its endpoints, then it is called small, while
a branch which is neither small nor large is called half-large.
Figure 8. Charts for torus.
Define the combinatorial splitting of a measured train track (τ, μ) along a large
branch e as illustrated in Figure 9, where we identify an edge with its μ transverse
measure μ(e) for convenience. One imagines separating bands of horizontal leaves in
the rectangle associated to e by excavating along the two “singular leaves” beginning
at the endpoints of the large branch, i.e., beginning at the singular ties. If the measure
μ is so that (either of) the singular leaves starting at an endpoint of e turn left or right,
then the respective split is called a left (case 1) or right (case 3) split, while if the two
singular leaves coincide for e, then the split is called a collision (case 2).
Splitting and smooth isotopy of measured train tracks generates an equivalence
relation on the set of all measured train tracks in F , and we shall let [τ, μ] denote the
equivalence class of the measured train track (τ, μ) and [τ, μ̄] denote the equivalence
class of the projectively measured train track (τ, μ̄).
Theorem 4.3 ([24]). If (τi , μi ) are positively measured train tracks giving rise to
corresponding measured foliations (Fi , μi ) via Construction 4.1, for i = 1, 2, then
[F1 , μ1 ] = [F2 , μ2 ] if and only if [τ1 , μ1 ] = [τ2 , μ2 ].
Thus, the space of all Whitehead equivalence classes of (projectivized) measured foli-
ations is identified with the space of all splitting equivalence classes of (projectivized)
measured train tracks up to isotopy.
There is another aspect to the splitting equivalence relation on the set of all measured
train tracks. In addition to splitting, one considers also shifting along a half-large
a b
∼ b−a if a < b
1m
d c
a b a b
e
∼ if a = b
2m
d c d c
a b
∼ a−b if a > b
3m
d c
Figure 9. Splitting.
branch by pushing two confluent branches of a train track past one another as illustrated
in Figure 10. Shifting plays a role in the later discussion, and a basic result in train
track theory [24] is that if two train tracks are related by shifting, splitting, and smooth
isotopy, then they are also related by splitting and smooth isotopy alone.
a a
b c b c
Figure 10. Shifting.
Let us finally give the idea of the proof that S is dense in MF 0 , as was mentioned
before, by explaining density of S in each chart V (T ) ⊆ P F 0 , for some maximal train
B(τ ) R>0 by a rational
track τ ⊆ F . We may approximate any μ ∈ V (τ ) ⊆ (R≥0 − {0})/
measure μ ∈ (QB(τ ) − {0})/
≥0
R>0 (satisfying the switch conditions). Furthermore
clearing denominators in μ , there are N ∈ Z>0 and ν ∈ (ZB(τ )
≥0 − {0})/R>0 so

that N μ = ν. We may construct an embedded family of curves in F from ν by
arranging ν(b) ≥ 0 tie-transverse strands parallel to b in a tie neighborhood of τ .
By the switch conditions, there are at each vertex exactly as many incoming strands
as outgoing, and there is a unique way to combine strands near vertices to produce a
disjointly embedded family of curves. Let us give each component curve a weight 1/N
and combine any parallel curves while adding their weights to produce our desired
weighted family of disjointly embedded curves. Let [F , μ1 ] denote the corresponding
measured foliation (discussed before); tracing through the constructions, one finds
that [τ, μ ] = 1/N[τ, ν] gives rise to [F , μ1 ]. Letting μ → μ and projectivizing, it
follows easily that families of disjointly embedded curves are dense in V (τ ). With a
little more work [24], one can approximate (in the topology of V (τ )) such disjointly
embedded families with a single curve, and this gives the asserted density of S itself.
Thus, Thurston’s boundary P F 0 (F ) is a completion of the set S. In fact, one
can approximate with a single non-separating curve (provided g = 0); since any two
such curves are equivalent under the action of MCsg , it follows that the action of MCsg
on P F 0 (F ) has a dense orbit, and the maximal non-Hausdorffness of the quotient,
which was mentioned before, is thereby established.
4.3 Laminations
Each basic formulation of the objects presented so far, namely, measured foliations
and measured train tracks, requires passage to the quotient under an appropriate equiv-
alence relation. Thurston has given a more ethereal, elemental, and elegant description
of these objects as “measured geodesic laminations”, where no passage to equivalence
classes is necessary. Here we simply give the definition and a few basic properties
referring the reader to [24] for instance for further details. A “lamination” L in F
is a foliation of a closed subset of F , and a “(transverse) measure” to L is defined
much as before as a σ -additive measure on arcs transverse to L with the analogous
condition of no-holonomy (where the homotopy is through arcs transverse to L with
endpoints disjoint from L). L is a “geodesic lamination” if its leaves are geodesic for
some specified hyperbolic metric. The simplest case of a geodesic lamination is the
geodesic representative of an element of S (F ). (In fact, for different choices of metric,
the spaces of measured geodesic laminations are naturally identified via the circle at
infinity in their universal covers, so we may speak of a geodesic lamination without the
a priori specification of a metric.) A measured geodesic lamination has zero measure
in F , and the intersection of a measured geodesic lamination L with a transverse arc
a in F is a Cantor set together with isolated points corresponding to intersections with
simple geodesic curve components or arc components of L, if any. There is a natural
topology on the set of all measured geodesic laminations in F , which is induced by
the weak topology on the set of all π1 (F )-invariant measures supported on the Möbius
band past infinity. ML(F ) (and ML0 (F )) is the corresponding space of measured
geodesic laminations (and with compact support) and corresponding projectivization
P L(F ) (and P L0 (F )). A basic result in Thurston theory is ML(F ) ≈ MF (F ),
ML0 (F ) ≈ MF 0 (F ), P L(F ) ≈ P F (F ), P L0 (F ) ≈ P F 0 (F ), where train
tracks give suitable charts on any of these piecewise linear manifolds. Furthermore,
the deformation theory due to Thurston, called “earthquaking” cf. [7], which we do
not further discuss here, is most conveniently expressed in the context of laminations.
4.4 Dynamics on train tracks

In this section, we simply recall Thurston’s classification of surface automorphisms
as well as recall several basic facts about “pseudo-Anosov” mappings. In the process,
we develop further basic techniques which will be required in quantization. Since
Thurston’s compactification produces a closed ball upon which the mapping class
group acts continuously, one immediately is led to consider fixed points of this action.
Theorem 4.4 (Thurston’s classification ([35], [9])). Any orientation-preserving home-

omorphism f : F → F is homotopic to a diffeomorphism f : F → F which satisfies
one of the following conditions (and the only overlap is between 1. and 2.).
1. f fixes a unique point of T and is of finite order.
2. f is “reducible” in the sense that f fixes an element of S .
3. f is “pseudo-Anosov” in the sense that there is some λ > 1 together with
two measured foliations (F± , μ± ) , which share singular points and are other-
wise transverse, so that f (F± , μ± ) = λ±1 (F± , μ± ). The projective classes
[F± , μ̄± ] are the unique fixed points of f on T . The invariant λ is called the
dilatation of f or f .
Notice the similarity with the trichotomy elliptic/parabolic/hyperbolic for frac-

tional linear transformations. In the reducible case, one simplifies the dynamics by
cutting F along a representative of the invariant element of S . A pseudo-Anosov map-
ping is the analogue of an Anosov map of the torus in the current context of surfaces
with negative Euler characteristic.
In fact, train tracks provide a powerful tool for analyzing the dynamics of surface
automorphisms owing to the fact that since a train track has a well-defined tangent line
at each point, there is a coherent notion of a train track smoothly “carrying” a curve,
another train track, or a lamination.
Suppose that κ is a smooth curve, a train track, or a measured geodesic lamination.
We say that the train track τ carries κ and write κ < τ , if there is a C 1 map φ : F → F
homotopic to the identity, called the supporting map, so that φ(κ) ⊆ τ , where the
restriction of the differential dφp to the tangent line to κ at p is non-zero for every
p ∈ κ.
We think of φ as squashing together nearly parallel strands of κ. For instance,
any curve arising as before from an integral measure on τ is carried by τ , and more
generally, any curve, train track or lamination κ which lies in a tie-neighborhood of τ
and is transverse to the ties satisfies κ < τ , where the supporting map collapses ties.
For example, if a train track σ arises from τ by splitting and shifting (but no
collapsing), then σ < τ , and we say that σ arises from τ by unzipping.
Suppose that the train track σ is transverse to a tie-neighborhood of τ , say with

supporting map φ : F → F . Let us enumerate the branches bj of τ , for j = 1, . . . , n,
and ai of σ , for i = 1, . . . , m, and choose xj ∈ bj for each j . There is then an
m × n-matrix A = (Aij ) called the incidence matrix, where Aij is the cardinality
of φ −1 (xj ) ∩ ai . It is clear that the incidence matrix A : Rm n
≥0 → R≥0 describes the
inclusion U (σ ) → U (τ ) in the train track coordinates.
We close this section with several basic results about pseudo-Anosov mappings.
Theorem 4.5 ([35], [22]). A homeomorphism f : F → F is a pseudo-Anosov map

if and only if there is a train track τ in F , with each component of F − τ an at most
once punctured polygon, so that τ unzips to f (τ ) with no collisions. Furthermore, the
incidence matrix A of the carrying f (τ ) < τ is Perron–Frobenius, the eigenvector
of A corresponding to the spectral radius λ gives the projective measure μ̄+ , and
likewise the extreme eigenvector of the transpose of A gives μ̄− .
Given a measured train track (τ, μ), consider the foliated neighborhood of τ de-
termined by μ via Construction 4.1. Choose some enumeration of the switches of τ
and serially follow the singular leaves from the switches until the first splitting (ig-
noring shifting), for the first switch, second switch, . . . , last switch, and then begin
anew from the first switch. Suppose there are no collisions, and record the resulting
sequence of right or left splits, so as to produce a semi-infinite word of rights and lefts.
Theorem 4.6 ([22]). The right-left sequence is eventually periodic if and only if the
corresponding measured foliation is fixed by some pseudo-Anosov mapping.
An explicit and simple construction of pseudo-Anosov maps is given by the fol-

lowing result.
Theorem 4.7 ([29]). Suppose that C, D ∈ S admit representative arc families C, D

intersecting minimally which satisfy the condition that each component of F −∪(C∪D)
is an at most once-punctured polygon. Take any composition w of Dehn twists to the
right along elements of C and to the left along elements of D so that the Dehn twist
along each element of C or D occurs at least once in w. Then w is pseudo-Anosov.
Torus example. Consider a generic train track τ in F11 , so τ has one large branch e
and two small branches. The two small branches are canonically linearly ordered by
first taking the branch a to the right and then the branch b to the left at either endpoint
{a,b}
of e, and furthermore U (τ ) ≈ R≥0 , i.e., the measures of the small branches a, b
are unconstrained and uniquely determine the measure on e as well. Given a measure
μ ∈ U (τ ) − {0}, start unzipping (τ, μ) along either singular leaf, i.e., split along
e, to produce another measured train track (τ1 , μ1 ); of course, τ1 is combinatorially
equivalent to τ . For definiteness, suppose that B = μ(b) > μ(a) = A, so the split
is a left split. The edge corresponding to b is the large edge of τ1 , and the two small
edges, in right/left order, have measures (A, B − A). Continue unzipping, i.e., next
split (τ1 , μ1 ) along its large edge to produce (τ2 , μ2 ). Again suppose that B > 2A
for definiteness, so the second split is a left split as well, and the small branches of τ2
have measures (A, B − 2A). Continue unzipping (under the assumption that there are
no collisions) until the first right split, say there are a1 = [ B
A ] > 1 left splits before the
first right split. Perform the right split along the large branch of (τa1 , μa1 ), where the
measures on the small branches are (A, B1 ) = (A, B − m1 A), to produce a train track
whose small edges have measures (A − B1 , B1 ). Continue unzipping and suppose
there are no collisions, i.e., suppose A and B are not rationally related, to produce a
semi-infinite sequence of symbols L (for left splits) and R (for right splits). Let a1
denote the number of L’s that begin this sequence, a2 denote the length of the next
consecutive sequence of R’s, a3 the length of the next consecutive sequence of L’s,
and so on. It follows from the discussion above that the continued fraction expansion
of B/A is given by
1
B/A = a1 + .
1
a2 +
a3 +
..
.
Continued fractions occur in another related
guise as well.
The isomorphism
MC11 ≈ PSL2 (Z) is induced by M → 01 11 and L → −1 1 0 , where M and L are
1
the right Dehn twists on the meridian and longitude respectively. A mapping class in
MC11 is pseudo-Anosov, reducible, periodic if and only if the corresponding fractional
linear transformation is hyperbolic, parabolic, elliptic respectively. Every hyperbolic
element of PSL2 (Z) is conjugate to a product

1 m1 1 0 1 mk 1 0
··· ,
0 1 n1 1 0 1 nk 1
where mi , ni > 0 are unique up to cyclic permutation. Furthermore, m1 , n1 , . . . ,
mk , nk are the partial quotients of the periodic continued fraction expansion of the
dilatation of the corresponding pseudo-Anosov map. It follows from this discussion
that in MC11 , all pseudo-Anosov mappings arise from the previous theorem. Indeed,
the theorem gives the construction of two semi-groups corresponding to right/left or
left/right twisting on meridian/longitude. For each semi-group, it is easy to construct
a train track τ in F11 so that the matrix representation above precisely describes the
action of the corresponding semi-group on the measures of the linearly ordered small
branches of τ ; indeed, these two train tracks are illustrated in Figure 8.
4.5 Decorated measured foliations and freeways

In this section, we recall material from [23] which is required for quantization. If
⊆ F is a cubic fatgraph spine of F , then we may blow-up each vertex of into a
little trigon as illustrated in Figure 11. The resulting object τ = τ has both a natural
branched one-submanifold structure and a fattening, and furthermore, components of
F − τ are either little trigons or once-punctured nullgons. Thus, τ is not a train track,
but it is almost a train track, and is called the freeway associated to . Notice that each
edge of gives rise to a corresponding large branch of τ , and each vertex gives rise
to three small branches. It is easy to see that every measured lamination of compact
support in F is carried by the freeway τ . The frontier of a once-punctured nullgon
component of F − τ is a puncture-parallel curve called a collar curve of F . A small
branch is contained in exactly one collar curve, while a large branch may be contained
in either one or two collar curves.
→ ∼
HH
HH

H

H
H
H HH
Figure 11. Freeway from fatgraph.
A measure on a freeway τ is a function μ ∈ RB(τ ) satisfying the switch conditions,

where we wish to emphasize that the measure is not necessarily nonnegative (as it is
for train tracks). Let U (τ ) denote the vector space of all measures on τ . Notice that
μ ∈ U (τ ) is uniquely determined by its values on the small branches alone, and the
switch conditions are equivalent to the following “coupling equations”
μ(a1 ) + μ(b1 ) = μ(e) = μ(a2 ) + μ(b2 ),
for any large branch e whose closure contains the switches v1 = v2 , where ai , bi
are the small branches incident on vi for i = 1, 2. On the other hand, the values
on the large branches alone also uniquely determine μ, and in fact, these values are
unconstrained by the switch conditions. Indeed, letting ai denote the large branches
incident on a little trigon with opposite small branches αi , for i = 1, 2, 3, we may
uniquely solve for a measure μ on τ , where
1
μ(αi ) = {μ(a1 ) + μ(a2 ) + μ(a3 ) − 2μ(ai )},
2
and so we identify U (τ ) ≈ RLB(τ ) , where LB(τ ) denotes the set of large branches
of τ .
In particular, if μ is a nonnegative measure on τ , then the analogue of Construc-
tion 4.1 in the current context produces a well defined equivalence class of measured
foliations in F , where this measured foliation will typically contain a collection of
puncture-parallel annuli foliated by curves homotopic to collar curves. Deleting these
foliated annuli produces a well-defined (but possibly empty) class in MF 0 (F ). Thus,

a nonnegative measure on τ canonically determines a point of MF 0 (F ) together
with a nonnegative “collar weight”, i.e., the transverse measure of a transverse arc
connecting the boundary components of the corresponding foliated annulus.
In the general case that μ is not necessarily nonnegative, suppose that C is a collar
curve of τ . The switches of τ decompose C into a collection of arcs, each of which
inherits a corresponding real-valued weight from μ. Let {γi }n1 denote the collection of
real numbers associated to the small branches of τ that occur in C. Define the collar
weight of C for μ to be μC = min{γi }n1 . Define a collar weight on F itself to be the
assignment of such a weight to each puncture.
We may modify the original measure μ ∈ U (τ ) by defining μ (b) = μ(b) − μC
if b is contained in the collar curve C for any small branch b of τ . Thus, μ is a
nonnegative measure on μ with identically vanishing collar weights that determines
a corresponding element of MF 0 (F ).
We are led to define the space MF 0 = MF 0 (F ) × Rs of decorated measured
foliation and summarize the previous discussion:
Theorem 4.8 ([23]). The space U (τ ) ≈ RLB(τ ) gives global coordinates on MF 0,

0 → MF 0 , where the fiber over a point
and there is a canonical fiber bundle : MF
is the set Rs of all collar weights on F .
Remark 4.1. The natural action of MCsg is by bundle isomorphisms of . Fur-

thermore, admits a natural MCsg -invariant section σ : MF 0 → MF 0 which is
determined by the condition of identically vanishing collar weights. The restriction
of σ to MF 0 ⊆ MF 0 gives a piecewise-linear embedding of the piecewise-linear
0 ≈ U (τ ) ≈ RLB(τ ) .
manifold MF 0 into the linear manifold (vector space) MF
4.6 Shear coordinates for measured foliations
We now give an equivalent parametrization of measured foliations in terms of “Thur-

ston’s shear coordinates” that are close analogues of Thurston’s shear coordinates Zα
on TH (F ). In fact, we have already encountered these quantities when describing the
splitting procedure train tracks (see Figure 9). There, excavating along two different
singular leaves, we have obtained the “new” edge, which can turn either left or right
(for splittings) or be absent (for collisions).
We assign a corresponding signed quantity (positive for right, negative for left) as
follows. Given a measure μ on the long branches of the freeway τ associated to the
fatgraph spine ⊆ F , define the (Thurston’s foliation-)shear coordinate of the edge
indexed by α to be
1
ζα = (μ(A) − μ(B) + μ(C) − μ(D)),
2
in the notation of Figure 1 for nearby branches. From the very definition, ζα is indepen-
dent of collar weights. Again, Thurston’s foliation-shear coordinates are alternatively
defined in terms of the signed transverse length of the arc between the singular leaves
along Zα , in analogy to the geometric interpretation given before of the shear coordi-
nates on TH .
Note that the shear coordinates ζα are not independent. They are subject to the
restrictions that
ζα = 0 (4.1)
α∈I
for the sum over edges α ∈ I surrounding any given boundary component, and we shall
refer to these conditions as the face conditions for shear coordinates. Thus, the space
of foliation-shear coordinates is of dimension LB(τ ) − n, where we let n denote the
number of boundary components. One sees directly that for any assignment of shear
coordinates, there is a well-defined point of MF 0 realizing them, thereby establishing
a homeomorphism between MF 0 and this sub-vector space RLB(τ )−n ⊆ RLB(τ ) of
shear coordinates on the long branches of τ .
To describe the action of the mapping class group on foliation-shear coordinates,
we shall give the transformation under Whitehead moves, i.e., derive the analogue of
formula (2.8) for measured foliations, which is an elementary calculation using the
formulas for splitting as follows.
Lemma 4.9. Under the Whitehead move in Figure 1, the corresponding foliation-shear
coordinates of the edges A, B, C, D, and Z situated as in Figure 1 are transformed
according to formula (2.8)
MZ : (ζA , ζB , ζC , ζD , ζZ )
→ (ζA + φH (ζZ ), ζB − φH (−ζZ ), ζC + φH (ζZ ), ζD − φH (−ζZ ), −ζZ )
with
φH (ζZ ) = (ζZ + |ζZ |)/2, (4.2)
i.e., φH (x) = x, for x > 0, and zero otherwise. All other shear coordinates on the
graph remain unchanged.
Remark 4.2. Comparing expressions for the classical function φ(x) = log(1 + e x )
and (4.2), one finds that the latter is a projective limit of the former:
1 1 h̄
φH (x) = lim φ(λx) = lim φ (λx), (4.3)
λ→+∞ λ λ→+∞ λ
that is, all three transformations coincide asymptotically in the domain of large absolute
values (or large eigenvalues for the corresponding operators) of Teichmüller space
coordinates {Zα }. We shall actively use this property in Section 5.3.2 when proving
the existence of the quantization of Thurston’s boundary for the punctured torus.
μ(A) μ(B)
@
@
@
R
@ ζZ · @
@ · @
·
· @
@
@ · 1l I
@
@
·
@ ·
@ ? ·
·
· @
6 ·
μ(D) · @ μ(C)
· @
@
@
R · @

@ ·
·
@ · @
@ cut hZ @
I@ @
@
(case ζZ > 0)
Figure 12. Foliation-shear coordinates.
5 On quantizing Thurston theory
5.1 Proper length of geodesics

Definition 5.1. The proper length p.l.(γ ) of a closed curve γ in the classical or
quantum case is constructed from the quantum ordered operator Pγ associated to a
closed oriented edge-path with basepoint (to begin the linearly ordered word Pγ ) as
1
p.l.(γ ) = lim tr log 2Tn (Pγ /2), (5.1)
n→∞ n
where we take the principal branch
of the logarithm and Tn are Chebyshev’s polyno-
mials (cf. (3.20)). Since Tn cosh 2t = cosh nt2 , it follows that p.l.(γ ) agrees with
half the hyperbolic length of γ in the Poincaré metric in the classical case.
More explicitly in the operatorial case, we can determine p.l.(γ ) explicitly in terms
of the spectral expansion of the operator Gγ , which is known exactly. Namely, the basis
of eigenfunctions of Gγ is “doubly reduced” in the sense that each eigenvalue (except 2,
which is singular) with corresponding eigenfunction αS , has the form eS/2 + e −S/2 ,
where S ranges over the entire real axis, and αS has the same eigenvalue as α−S . In
fact, these functions coincide, so there is actually a representation on the positive real
axis, which is nevertheless complete, and is singular at infinity and zero. We may
define the proper length operator to the be one with the same eigenfunctions αS for
S positive (which constitute a basis in the function space) and with eigenvalues to
be |S/2|. This operator p.l.(γ ) is then a well-defined operator on any compactum in
function space.
The proper length of a QMC or GMC Ĉ, again denoted p.l.(Ĉ), is the sum of the
proper lengths of the constituent geodesic length operators (or the sum of half geodesic
lengths calculated in the Poincaré metric in the classical case) weighted by the number
of appearances in the multiset.
5.2 Approximating laminations and the main theorem

Fix once and for all a spine of F with corresponding freeway τ . A measure μ on τ
gives rise to a (possibly empty) measured foliation in F together with a collar weight on
the boundary components of F . Erasing collars yields an underlying measure μ1 ≥ 0
on τ whose support is a sub-train track τ1 ⊆ τ , and the measured train track (τ1 , μ1 )
determines a (possibly empty) measured foliation. Via the canonical embedding of
MF 0 into MF 0 with vanishing collars, we may thus uniquely determine a point of
MF 0 by specifying foliation-shear coordinates on the long branches of τ satisfying the
face conditions 4.1, i.e., MF 0 is naturally identified with a codimension n subspace of
RLB(τ ) . Passing to projective foliations, a point of P F 0 is given by the projectivization
P ζ of a vector of foliation-shear coordinates ζ = (ζi ), where ζ ∈ RLB(τ ) − {0}, and
i indexes the long branches of τ , i.e., the edges of .
β
Definition 5.2. A sequence nβ = (ni }, for β ≥ 1, of integer-valued ni , for i =
1, . . . , LB(τ ), on τ is an approximating sequence for the projectivized measured
β β
foliation P ζ if the face conditions 4.1 hold on nβ and if limβ→∞ ni /nj = ζi /ζj for
all i, j with ζj = 0.
Constructed from n as an integral measure on τ is a GMC Ĉ with integral collar

weights. Just as with decorated measured foliations, components of Ĉ which are
puncture- or boundary-parallel can be erased to produce a corresponding multicurve
to be denoted Ĉn . Ĉn is carried by a sub-train track of τ , and it traverses the long
branch of τ indexed by i some number, say, mi ≥ 0 of times, so mi is the standard train
track coordinate of integral transverse measure. In the usual notation as in Figure 12,
one sees directly that nZ = 21 (mA − mB + mC − mD ). We may also sometimes write
Ĉm for Ĉn
Definition 5.3. A graph length function with respect to the spine is any linear
function

g.l.a (Ĉn ) = g.l.a (Ĉm ) = ai m i . (5.2)

i
In particular, when all ai are unity, the graph length is just the combinatorial length
of Ĉm , i.e., the total number of edges of traversed (with multiplicities) by all the
component curves of Ĉm . When the spine and a are fixed or unimportant, then we
shall write simply g.l.(Ĉn ) or g.l.(Ĉm ) for the graph length.
Any graph length function is evidently additive over disjoint unions of multicurves,
and more generally, is a linear function of m.
We next describe the bordered or punctured torus case in detail. Each multicurve
on the torus is uniquely determined by three nonnegative integers (mX , mY , mZ ) that
satisfy one of the three triangle equalities mi = mj + mk , where {i, j, k} = {X, Y, Z}.
Projectivization allows us to re-scale so that mj and mk are relatively prime using this
degree of freedom, so that the corresponding multicurve has just one component.
As in Figure 8, the space P F 0 (F11 ) of projectivized measured foliations with
compact support is a piecewise-linear circle for r + s = 1, and an alternative family
of charts on this circle is given in Figure 13. The relation ∼ in Figure 13 denotes
the equivalence between different boundary cases between two different charts, and
arrows represent one-simplices in P F (F11 ).
In Figure 13, we use the previous notation GX , etc. (see (2.15)) but in a slightly
different sense. Now, these quantities are (2×2)-matrices, not just geodesic functions,
i.e., we do not evaluate traces in the corresponding formulas. There is thus an ambiguity
in choosing the place in the graph where the matrix products begin. We indicate this
place by drawing the reference cut (the dotted line). Changing the reference cut
when moving along the circle corresponds to passing from one chart to another in the
chart covering of the circle. Of course, choosing the reference cut does not affect the
quantum trace operation.
In order to have a good transition in the boundary cases, for instance, in the upper
case in Figure 13 (and the other cases are similar and omitted), we must ensure that
the corresponding functions for the quantities Gm
(1) m (1)
X GZ and GX G̃Z must coincide
in the limit m → ∞ with each other and with the corresponding quantity calculated
merely for the “short” geodesic function GX . To prove this, given two (2×2)-matrices
GX and GZ corresponding to geodesic curves, we can conjugate them by respective
unitary transformations UX and UZ to diagonal form with real eigenvalues e ±lX /2
and e ±lZ /2 since GX , GZ are hyperbolic. We then have

e mlX /2 0 e lZ /2 0
m
tr GX GZ = tr V V −1 , V = UX−1 UZ
0 e −mlX /2 0 e −lZ /2
(5.3)
and the proper length (5.1) is mlX /2+O(1).2 In order to have a well-defined projective
limit, we shall “kill” the factor m in a consistent way, and this can be achieved by
dividing the result by any graph length function of the curve since g.l.(Gm X GZ ) =
m · g.l.(GX ) + g.l.(GZ ).
In the quantum case, however, the situation is much more involved. Indeed, let us
consider an example of the product, which is of form U m V , as in (5.3), where U =

2 Unless the matrix U −1 G U has the form 0 −r −1 . For this matrix to determine a hyperbolic element,
X Z X r φ
quantity φ must be real
the greater than two. Multiplying this matrix by the diagonal matrix above, we obtain
0 −r −1 e mlX /2
−ml /2 −ml /2 , so for sufficiently large m, the resulting product ceases to be hyperbolic, which is
re X φe X
absurd.
' $
(1, m + 1, m) (1, m + 1, m + 2)
Y X Y X
@
@ · Z @
@ · Z
···
·· ···
··
· ·
@ ··· ···
@··· ∼ @
@···
@
@ @
@
@
@ @
@
X Y X Y
(1) (1)
tr (GX )m GZ tr (GX )m G̃Z
& %
@
R
@
' $ ' $
(m, m + 1, 1) (m, m + 1, 2m + 1)
Y X Y X
@
@ · Z @
@ · Z
···
·· ···
··
· ·
@
@ ··· GX = LY RZ @
@ ···
··· ···
@
@ (1)
GZ = RY LX @
@
@
@ (1) @
@
X Y G̃Z = LY RZ RX LZ = GX GY X Y
(1) (1) m
tr (GZ )m GX here changing the tr (G̃Z ) GX

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·-
··
(m + 1, m, 1)
reference cut (m + 1, m, 2m + 1)
Y X GY = RX LZ Y X
@
@ Z
(2) @
@ Z
GZ = LX RY
@
@·· (2) @
@··
···
@ G̃Z = RX LZ LY RZ = GY GX ···
@
···
···
@ ···
···
@
@
@ @
@
X Y X Y
(2) (2)
tr (GZ )m GY tr (G̃Z )m GY
& % & %
@
I ' $
@
(m + 1, 1, m) (m + 1, 1, m + 2)
Y X Y X
@
@ Z @
@ Z
@
@·· ∼ @
@··
··· ·
··· @
@ ···
··· @
@
··· ··
@
@ @
@
X Y X Y
(2) (2)
tr (GY )m GZ tr (GY )m G̃Z
& %
Figure 13. The circle P L0 (F11 ).

e X/2 , V = e Y/2 , and [X, Y ] = 4π i h̄. Thus, (U m )V = e mX/2+Y /2+mπi h̄/2 while
V (U m ) = e mX/2+Y/2−mπi h̄/2 , and the corresponding logarithms do not coincide as
m → ∞. Moreover, even the Hermiticity condition does not often suffice to determine
the proper length. For instance, given operators U = e αX and V = e iβ∂X , we
2
may calculate that V (U m )V = e mαX +2iβ∂X +mαβ/3 , i.e., the proper length in this
2
case is αX 2 + αβ/3 and depends on the parameter β (of course, this correction is
purely quantum). This illustrates that proving continuity for the boundary transitions
in the quantum case requires more subtle estimates, which we perform in the next
section after deriving recurrence relations for the operators of quantum approximating
multicurves.
We may now formulate our main result on quantizing Thurston theory:
Theorem 5.1. Fix a spine of F11 with corresponding freeway τ . Fix any projectivized
vector P ζ of foliation-shear coordinates on τ and any graph length function g.l. For
any approximating sequence n β to P ζ , the limit
p.l.(Ĉnβ )
lim (5.4)
β→∞ g.l.(Ĉnβ )
exists both in the classical case as a real number and in the quantum case as a weak
operatorial limit.
Because both the numerator and denominator in the limit are additive, this limit
is projectively invariant and defines a continuous function (in the classical case) or a
weakly continuous family of operators (in the quantum case) on the circle P L0 (F1,r s ),
for r + s = 1.
The proof of the previous theorem occupies the remainder of this section. The
continued fraction structure intrinsic to the torus case is used extensively, and various
analogous operatorial recursions are derived and studied. There is a second essentially
combinatorial proof of this result, however, only in the classical case since we have no
means to control the quantum ordering of the procedure. Nevertheless, the structures
discovered are interesting, and we present this second proof in Appendix A, which
depends upon the recursion (5.7) derived later in Lemma 5.2. Indeed, this basic
recursion arises from “Rauzy–Veech–Zorich induction” [30] in the special case of the
torus, which is derived from first principles in the next section.
5.3 Elements of the proof
5.3.1 Continued fraction expansion. In each one-simplex in P L0 (F ), illustrated

as arrows in Figure 13, the approximating multicurve is determined by two nonnegative
integers, m1 and m2 , where we assume that m1 > m2 with m1 and m2 relatively prime.
It is convenient to represent the ratio m2 /m1 as a simple continued fraction:
1
m2 /m1 = . (5.5)
1
a1 +
a2 +
..
.
1
+
1
an−1 +
an
We concentrate on the case mX = m2 , mY = m1 , mZ = mX + mY , with the other

cases following by symmetry, and describe the recurrence procedure for constructing
the corresponding approximating multicurves (i.e., approximating geodesics, since
any multicurve in the torus is just a multiple of a single geodesic).
Referring to Figure 13, it is convenient to visualize this case by drawing a line
2l 1l
• • ··· • • ◦ ··· ◦ ◦ ,
1 2 m 1 − 1 m 1 m1 + 1 m1 + m2 − 1 m1 + m2
(5.6)
which represents (the right side of) the cut over the edge Z (homotopic to the cut
hZ in Figure 12). The geodesic function is constructed as follows. Considering the
multicurve in a neighborhood of the edge Z, we start from the lowest thread that goes
from the edge Z to the edge X (labelled by the circled number one in Figure 12 and
in (5.6)) and moves to the right (this corresponds to the leftmost ◦ in (5.6)). We have
the matrix GY corresponding to passing consecutively through edges Z and X and
come back to the line (5.6), which must be periodically continued, at the point that is
situated to the right by m2 points (in this case, just the leftmost • in (5.6) marked there
by a circled number two). Each time passing the bullet sign, we must set the matrix
GX and passing the ◦ sign we must set the matrix GY .
We thus move along the periodically continued line of circles and bullet signs
in (5.6), at each step jumping m2 points to the right and setting the corresponding
matrices GX or GY . We describe a recurrence procedure that produces from the
continued fraction expansion (5.5) the correct sequence of operators, i.e., the correct
ordered sequence of branches traversed by the corresponding curve. We shall split
the construction procedure into stages. Let us consider the trajectory of the starting
thread (the point m1 + 1 in (5.6)). Each stage terminates as soon as we come closer to
the starting thread 1 than the distance to it at the beginning of the stage (approaching it
from the opposite side compared to the beginning of the stage, see Figure 14 below).
We shall let Li denote the string of matrices for the first i stages, where by convention
we arrange matrices from right to left. We terminate the first stage at the thread that
is closer than m2 to the left of the starting point, i.e., L1 = (GX )a1 GY .
Define L̃0 ≡ GX and introduce L̃i , for i ≥ 1, which is the matrix Li (composed
from the elementary matrices GX and GY ) in which the first two symbols of elementary
matrices must be interchanged: if the first symbol of Li is GY and the second is GX ,
then the first symbol of L̃i is GX and the second is GY ; all other elementary matrices
retain their forms. We illustrate this procedure in Figure 14.3
(Li−2 ) Li−1
-
• • • • • • • • • • • • • • • • • • • • ◦◦◦◦◦◦◦◦◦◦◦◦◦
•
|Li−1 | @
I
- @
@ end of (i − 1)th stage
Li

• • • • • • • • • • ◦◦◦◦◦◦◦◦◦◦◦◦◦
•
|Li |
-
end of i th stage
Li+1 end of (i + 1)th stage
-
• • • • • • • • • • ◦◦◦◦◦◦◦
•
Figure 14. Threads a curve.
One thus sees directly that

L2 = (L1 )a2 −1 L̃0 L1
L3 = (L̃2 )a3 −1 L1 L2
L4 = (L3 )a4 −1 L̃2 L3
L5 = (L̃4 )a5 −1 L3 L4
..
.
and this leads to the following recurrence relation.
Lemma 5.2. Given the simple continued fraction expansion (5.5) of m2 /m1 , the
sequence of matrices Ln associated to the corresponding geodesic is given by the
3 Note that we terminate a stage whenever we come closer to the starting thread; this occurs at each stage on
the opposite side from the previous stage, just as for continued fractions. The appearance of tilded quantities is
explained as follow: each time during the recursion when we start a string “parallel” to some Lk from the right
to the starting thread (in the circled domain), we must follow the same string of branches because, by definition,
there are no threads in the string Lk except the very first thread that appears at a distance closer than |Lk | to
the starting thread. On the other hand, if we start from the left of the starting thread, then we must interchange
exactly the first two appearances of matrices in the resulting string.
following recursion
L2i = (L2i−1 )a2i −1 L̃2i−2 L2i−1

for i ≥ 1, L̃0 = GX , L1 = (GX )a1 GY . (5.7)
L2i+1 = (L̃2i )a2i+1 −1 L2i−1 L2i
Turning to the quantum case, we first show that the proper length operator (5.1)
p.l.( ×× L2i ×× )/g.l.(L2i ) must agree with p.l.( ×× L2i−1 ×× )/g.l.(L2i−1 ) for a2i → ∞, that is,
the operators corresponding to continued fractions of form (5.5) with large coefficient
an must converge to the operator corresponding to the continued fraction terminated
at the (n − 1)st step. To this end, we must analyze the structure of matrix products and
corresponding operators. Indeed, the operators Li and L̃i enjoy elegant commutation
relations as we shall next see.
Notice that for every stage i we have a geodesic corresponding to the matrix Li
(because we can close the corresponding geodesic line without self-intersections), and
we can therefore define the corresponding QMC
Li ≡ ×× tr Li ×× .
The first observation pertains to Li and L̃i : One of the corresponding curves can be
obtained from the other by a parallel shift along the cut hZ illustrated in Figure 12.
Thus, the curves are disjoint and hence are homotopic on the torus, and so tr Li = tr L̃i ,
i.e.,
Li = L̃i for i ≥ 1. (5.8)
Furthermore, the curves Li and Li+1 can be perturbed to have exactly one intersection
as one sees by considering how the corresponding geodesic curves pass through the
cut hZ .4 In the case where i = 2k, we obtain (in the notation of (3.24))
[L2k , L2k+1 ]q = ξ ×× tr L2k L2k+1 ×× , and
a2k+1 −1 × (5.9)
[L2k+1 , L2k ]q = ξ tr L2k−1 (L̃2k )
×
× × ,
where the proper quantum ordering is assumed for the terms in the right-hand sides.
Formulas (5.9) are crucial when proving the continuity. Letting
Im ≡ ×× tr(L2i−1 )m−1 L̃2i−2 L2i−1 ×× ,
L2i−1 ≡ e X /2 + e −X /2 ,
we find
Im−1 L2i−1 = q 1/2 Im + q −1/2 Im−2 , (5.10)
−1/2
L2i−1 Im−1 = q Im + q 1/2
Im−2 , (5.11)
4 One of these curves necessarily has odd subscript 2k + 1, and we can make a small shift of all threads of
this curve to the right from the threads of the second curve L2k (or L2k+2 , depending on the situation). We find
that there are no intersections of threads outside the region between terminating points of L2k+1 and L2k (or
L2k+2 ), and in this domain, when closing the curves, there is produced exactly one intersection point.
where the basis of the recursion is given by I0 ≡ e lY /2 + e −lY /2 and I−1 ≡ e lZ /2 +

e −lZ /2 . From (5.10), we have the exact equalities

Im = q −m/2 I0 e mX /2 + e −mX /2 − q −m/2−1/2 I−1 e (m−1)X /2 + e −(m−1)X /2 ,
(5.12)
or, equivalently,

Im = q m/2 e mX /2 + e −mX /2 I0 − q m/2+1/2 e (m−1)X /2 + e −(m−1)X /2 I−1 .
(5.13)
We wish to present the expression (5.12) or (5.13) in the form e mH1 +H0 , where
H1 , H0 are Hermitian operators independent of m, so the proper length is then just H1
(while H0 introduces quantum corrections that do not affect the proper length limit
but are important for ensuring the proper commutation relations).
Let us choose a compact domain F in the function space L2 (R) such that norms
of all the operators in play are bounded for functions from this domain. Notice that
neither term on the right-hand side in (5.12) is self-adjoint, but if one of these terms
prevails (and it can be only the first term since we have an expression with coefficients
that are all positive in the classical limit of the right-hand side, and hence we must have
a positive left-hand side as well), then we can replace the total sum by this prevailing
term and obtain an approximate equality

q −m/2 I0 e mX /2 + e −mX /2 ∼ q m/2 e mX /2 + e −mX /2 I0
in this limit. By considering the spectral expansion with respect to the eigenfunctions
of the operator X , we immediately conclude that I0 ∼ e −2πi h̄∂/∂|X | in this limit,
and then
H1 = |X |/2, (5.14)
where the modulus has to be understood in terms of the spectral expansion: having a
QMC operator LX which admits a spectral decomposition (see formulas (5.23)–(5.25)
below) in functions |αS , we define the operator |X | by its action on these functions:
|X | |αS = |S| |αS .
A potentially problematic situation is when neither of the terms prevails and their
difference remains finite as m → ∞. This would correspond, as in the classical case
discussed before, to a situation where the corresponding element fails to be hyperbolic.
As we next show, we must obtain “long” curves when m goes to infinity, so this is also
impossible in the quantum case; that is, for GX , GY , GZ , and G̃Z from (2.15), (3.22)
and representing GX = e X /2 + e −X /2 , we must prove the operatorial inequality
GY e |X |/2 > q 1/2 G̃Z . (5.15)
To prove this, first express e X /2 through GX . Taking the positive branch of the
square root, we find from the left-hand side of (5.15) the expression
⎛ ! ⎞
GX GX2
GY ⎝ + − 1⎠,
2 4
and"since GY GX = q 1/2 G̃Z + q −1/2 GZ , we must compare two expressions

G2
GY 4X − 1 and q 1/2 G̃Z − q −1/2 GZ /2. Multiplying by the Hermitian conjugate
on the right in both expressions, we eliminate the square root and arrive at Laurent
polynomial expressions. After some simple algebra, we come to the inequality to be
proved:
1
q G̃Z GZ + q −1 GZ G̃Z > G2Y .
2
To see this, we expand the left-hand side
−X−Z
e + e X−Z + e X+Z + (q + q −1 ) e X
q + q −1 −X−Y −Z
+ e + e X+Y +Z + e Z + 2 e −Z + (q + q −1 )
2
q 2 + q −2 Y +Z q 3 + q −3 Z
+ e + e −Y −Z + e −Y +Z + (q + q −1 ) e −Y + e ,
2 2
while the right-hand side is expressed as
G2Y = e −X−Z + e X−Z + e X+Z + 2 + (q + q −1 )( e X + e −Z ).
Subtracting this expression from the previous one, we obtain that this difference is
q 2 + q −2 2 q + q −1 X+Y +Z
GX + e + e −X−Y −Z − q − q −1 , (5.16)
2 2
and both these terms are positive definite for |q| = 1.
The operatorial inequality (5.15) has therefore been established. This proves that
the limit (5.4) exists and is well defined at rational points of the continued fraction
expansion, and we next turn to the case of infinite continued fraction expansions, i.e.,
infinite sequences of elementary operators.
5.3.2 Mapping class group transformations and the unzipping procedure. We

consider now an infinite continued fraction expansion a1 , a2 , . . . , an , an+1 , . . . ex-
tending the notation of (5.5). As we shall see, there is a corresponding sequence of
unzippings of the freeway τ associated to a spine of F as in Section 4.5 as well as
an associated sequence of mapping class group elements, expressed as Dehn twists,
which reduce an approximating multicurve to one of two possible graph simple curves.
Given the recursive representation (5.7) for the operator of a geodesic curve deter-
mined by a continued fraction expansion (5.5) and applying two (unitary) operators
DX and DY of the modular transformations of the form (3.12) that correspond to the
respective Dehn twists along the corresponding closed curves γX and γY (with the re-
spective geodesic functions GX and GY ), we shall construct the sequence of zipping
or unzipping transformations.
Definition 5.4. An approximating multicurve is determined by two nonnegative inte-

gers m1 and m2 . As in Figure 13, enumerate such a pair as a triple (m1 , m2 , m1 + m2 ).
If m1 > m2 , the action of the Dehn twist DY−1 along γY (an unzipping transformation
– the Dehn twist in the opposite direction) is
DY−1 : (m1 , m2 , m1 + m2 ) → (m1 − m2 , m2 , m1 )
while if m1 < m2 , we apply the unzipping transformation along the curve γX , which
gives
−1
DX : (m1 , m2 , m1 + m2 ) → (m1 , m2 − m1 , m2 ).5
Given a continued fraction expansion (5.5), we construct the sequence of unzipping

transformations
−an −a
D(X D n−1 . . . DY−a3 DX
or Y ) (Y or X)
−a2 −a1
DY , (5.17)
which, when applied to the approximating multicurve (m1 , m2 , m1 + m2 ), reduces it
either to (1, 0, 1) ≡ γY for n even or to (0, 1, 1) ≡ γX for n odd.
Definition 5.5. Equivalently, we can consider the zipping procedure, that is, given a
a an
sequence of Dehn twists DYa1 DX a2 a3
DY . . . DYn−1 DX applied to the curve γY , we obtain
the curve (m1 , m2 , m1 + m2 ).
Considering the sequence (5.17) of quantum Dehn twist operators (3.12) and ex-
ploiting the quantum invariance from Lemma 3.9, we come to the main observation
that having an involved expression for the proper limit (5.1) of a QMC operator con-
structed by the rules described in Lemmas 3.9 and 5.2 in terms of the elementary
operators X, Y, Z, we may perform the sequence (5.17) of unzipping quantum mod-
ular transformations, which reduces this operator to a standard form of the quantum
operator GY or GX expressed through the new operators X (n) , Y (n) , Z (n) related to
the initial operators by this sequence of quantum modular transformations. This is the
operatorial statement of naturality of lengths under the mapping class group action.
It is intuitively natural to imagine that as the geodesic lengths must diverge as
m1 , m2 tend to infinity, we must eventually come to an asymptotic regime where all
quantities X(n) , Y (n) are large in the literal or operatorial sense for all sufficiently
large n. The inexorability of the approach to this asymptotic regime is not obvious
and is described in the next section.
5.3.3 Asymptotic regime. Let us recall the modular transformations for X, Y , and
Z variables:
−1
DX : (X, Y, Z) → (X + 2φ h̄ (Z), −Z, Y − 2φ h̄ (−Z)) (5.18)
and
DY−1 : (X, Y, Z) → (−Z, Y − 2φ h̄ (−Z), X + 2φ h̄ (Z)). (5.19)
5 In terms of the symbolic dynamics of elementary operators G and G , we can present this action as
X Y
−1 −1
follows: DX (GX ) = GX , DX (GX GY ) = GY and DY−1 (GX GY ) = GX , DY−1 (GY ) = GY .
In terms of the quantities U ≡ e X/2 and V ≡ e −Y /2 in the case where X+Y +Z = 0,

we have

−1 U e X/2 + e −X/2−Y U + V U −1 V
DX DX = ≡ , (5.20)
V e −Y/2+X/2 q 1/2 U −1 V

−1 U e X/2+Y /2 q 1/2 U V −1
DY DY = ≡ . (5.21)
V e −Y/2 + e Y /2+X V + U V −1 U
Worth mentioning is that since each operator Zα is Hermitian, each exponential is

positive definite, so we can always write, for instance, that U + V U −1 V > U in
the sense of spectral expansion: f |U + V U −1 V |f > f |U |f for any function
f ∈ L2 (R).
Using now an alternating sequence of transformations (5.20), (5.21), we shall
subsequently show that we attain the asymptotic regime of large positive X (large U )
and large in absolute value negative Y (large V ) starting from every pair of X and Y
lying in a compact domain of the (X, Y )-plane in the classical case or acting within
a compactum of test functions with bounded derivatives in the function space in the
quantum operatorial case.
In this section, we verify that the asymptotic regime is attained for the distinguished
sequence of modular transformations corresponding to the Fibonacci number sequence
−1
(golden mean), namely, for alternating DX and DY−1 . The proof in the general case
is analogous although the structure is more involved, as described in the next section.
−1 −1 −1 −1 −1 −1 n
Given the sequence of transformations DX DY . . . DX DY = DX DY , we
obtain

−1 −1 U
DX DY (5.22)
V
⎛ ⎞
U V −1 U + V
=⎝ ⎠.
V 1/2 (V −1 U q −1/2 + U −1 V q 1/2 )U (V −1 U q −1/2 + U −1 V q 1/2 )V 1/2 + V 1/2 U −1 V 1/2
In the classical case, the asymptotics is already clear from this formula; for the first
entry in (5.22), we have U V −1 U + V = U (V −1 U + U −1 V ) > 2U as the expression
in the parentheses has the form e S + e −S ≥ 2 for any real S. The same logic applies
to the second entry in (5.22), and we deduce that the classical part V (U + U −1 ) has
the same property.
The proof given below in the quantum case is more subtle as it needs a thorough
operatorial analysis. Nevertheless, the estimates turn out to be close to those in the
classical case, which we briefly discuss here: we must prove that a lower bound on
the operatorial spectrum on a compactum in the function space diverges with n. This
is a routine procedure, which uses that the action of operators X and Y in the basis of,
say, normalized Hermitian functions hn has the form

√ √ √
X|hn = 4π h̄ n + 1|hn+1 + n|hn−1
√ i √ √
Y |hn = 4π h̄ n + 1|hn+1 − n|hn−1 .
2
These operators “almost” commute in the domain of large n, which allows the com-
binatorics to be analyzed semiclassically.
In the quantum case, we recall the construction of quantum Dehn twists and their
eigenfunctions from [18]. The generator of the Dehn twist DX has the form
DX = e q1 /2πi h̄ F h̄ (q1 + p1 ),
2
q1 = X/2, p1 = 2π i h̄∂X (5.23)
and because it commutes with the geodesic length operator GX , they share the common
set of eigenfunctions
|αS = e −X
2 /16πi h̄
F h̄ (S + X)F h̄ (−S + X) (5.24)
with the eigenvalues
DX |αS = e S
2 /2πi h̄
GX |αS = 2 cosh(S/2)|αS ; |αS . (5.25)
The functions |αS constitute a complete set of functions in the sense that
αT |αS = δ(S − T )ν −1 (S), ν(S) = 4 sinh(π S) sinh(π h̄S)
and ∞
ν(S)dS |αS αS | = Id
0
We now split the plane of the variables (X, Y ) = (X(0) , Y (0) ) into four sub-domains
−1
and consider the action of the Dehn twists DX and DY−1 in each sub-domain.
Domain I. {X (0) > 0, Y (0) > 0} ∪ {X (0) > 0, Y (0) < 0 and |X(0) | > |Y (0) |}:
−1
DX (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) , X(0) + Y (0) ) ∈ Domain I,
DY−1 (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) + Y (0) , −2X(0) − Y (0) ) ∈ Domain II;
Domain II. {X (0) > 0, Y (0) < 0 and |X(0) | < |Y (0) |} ∪ {X (0) < 0, Y (0) < 0}:
−1
DX (X(0) , Y (0) ) = (X(1) , Y (1) ) = (−X(0) − 2Y (0) , X(0) + Y (0) ) ∈ Domain I,
DY−1 (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) + Y (0) , Y (0) ) ∈ Domain II;
Domain IIIa. {X(0) < 0, Y (0) > 0 and |X(0) | < |Y (0) |}:
−1
DX (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) , X(0) + Y (0) ) ∈ Domain IIIa or IIIb,
DY−1 (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) + Y (0) , −2X(0) − Y (0) ) ∈ Domain I or II;
Domain IIIb. {X (0) < 0, Y (0) > 0 and |X(0) | > |Y (0) |):
−1
DX (X(0) , Y (0) ) = (X(1) , Y (1) ) = (−X(0) − 2Y (0) , X(0) + Y (0) ) ∈ Domain I or II,
DY−1 (X(0) , Y (0) ) = (X(1) , Y (1) ) = (X(0) + Y (0) , Y (0) ) ∈ Domain IIIa or IIIb.
We see that only Domain III is potentially problematic. This regime is however
unstable: absolute values of X and Y variables decrease in this regime and they even-
tually leave the asymptotic regime as soon as we remain in Domain III; immediately
upon leaving this domain, we come to Domains I and II and will never leave this three
quarters of the (X, Y )-plane. The above considerations of U , V just demonstrate
that even if we were initially in Domain III, we come to the nonasymptotic domain
of bounded X and Y and then will leave this compactum moving toward asymptotic
expansions in Domains I and II.
The asymptotic dynamics always takes place in the first three quarters of the (X, Y )-
plane. Nevertheless, even this dynamics is rather involved. The stable regime corre-
sponds to the case where we are in Domain II before applying one or several operators
−1
DX . The application of the first of these operators brings us to Domain I, and upon
−1
subsequent applications of the operators DX we remain in Domain I. Next, if we
were in Domain I, then the very first application of the operator DY−1 brings us to
Domain II, and we then remain in Domain II upon subsequent applications of DY−1 .
We turn now to actual geodesic lengths of curves or proper lengths of operators. If
a sequence of unzipping transformations terminates, this means that we have a graph
simple geodesic, which is either GX if the last transformation was DY−1 or GY if
−1
the last transformation was DX . Considering the corresponding geodesic or proper
lengths, we find that up to exponentially small corrections, the leading contributions
in the above domains are
p.l.(γY ) = X + Y/2 in Domain I, (5.26)

p.l.(γX ) = −Y − X/2 in Domain II (5.27)
(see expressions (2.15)).

Thus, although the transformation laws for the variables X, Y themselves do not
possess the property of linearity with respect to the parameters ai , aj , when applying
−1 a −ai −a
sequences of transformations (DX )i ≡ DX or (DY−1 )aj ≡ DY j , the proper lengths
do possess this property! Namely, starting with variables (X (0) , Y (0) ) lying in the
−ai −a
corresponding domains and applying the sequences of transformations DX or DY j ,
we obtain for the resulting proper lengths the following expressions:

Y (0) X (0)
p.l.(γY (i) ) = − + ai − Y (0) − for X(0) , Y (0) ∈ Domain II (5.28)
2 2
(0) (0)
X Y
p.l.(γX(j ) ) = + aj X (0) + for X(0) , Y (0) ∈ Domain I (5.29)
2 2
Let us now explore the asymptotic formulas (5.26), (5.27) and (5.28), (5.29) first
in the classical case to close this section, relegating the discussion of the quantum case
to the next section.
Assume that we start from the variables (X(0) , Y (0) ) in Domain I and have the
−a
corresponding initial length p.l.(γY (0) ) from (5.26). Applying the transformation DY j ,
we obtain new variables (X (j ) , Y (j ) ) and the new proper length p.l.(γX(j ) ) (5.29)
having form (5.27) in these new variables, which must now lie in Domain II. Note that
explicitly
(X(j ) , Y (j ) ) = (X(0) + Y (0) + (aj − 1)(−2X (0) − Y (0) ), −2X(0) − Y (0) ).
−ai
We then apply the transformation DX to obtain variables (X (j,i) , Y (j,i) ), and the
proper length p.l.(γY (j,i) ) is expressed as in (5.28), where the term multiplied by ai is
(j )
none other than p.l.(γX(j ) ) and the term − Y 2 is exactly p.l.(γY (0) ). We thus find in
the asymptotic regime that the corresponding lengths are related by exactly the same
recurrence relation as for a graph length (the latter of which follows immediately from
(5.7)):
p.l.(γY (j,i) ) = ai p.l.(γX(j ) ) + p.l.(γY (0) ), (5.30)
g.l.(γY (j,i) ) = ai g.l.(γX(j ) ) + g.l.(γY (0) ). (5.31)

It is then easy to conclude that the ratio of these two quantities has a definite limit
as i → ∞ for any sequence of numbers ai . It is a standard estimate: given two
numerical sequences (5.30) and (5.31) and denoting the relative error of their ratio as
εi , i.e., at the ith step, the ratio is S(1 + εi ), where S is constant, for ai+1 > 1 at
the (i + 1)th step, we obtain εi+1 < εi /(ai+1 − 1/2), or if we have two coefficients
ai+1 = ai = 1, then εi+1 < εi−1 /1.5. In general, for εi small enough, we always
have εi+1 < εi . This shows that the relative error decreases exponentially with the
index i.
5.3.4 Quantum continued fraction expansion. Let us turn again to the sequence
(5.17) of unzipping transformations. In order to obtain operatorial expressions, we
consider the unitary operators DX , DY and explicitly indicate the variables in which
these operators are expressed, i.e., we write DX(j ) ≡ DX (X(j ) , Y (j ) ) for the Dehn
twist along γX at the (j + 1)th step. Thus,
(X (n) , Y (n) ) (5.32)
−an −an−1 −a2 −a1 a an
= DX (n−1) DY (n−2) . . . DX (1) DY (0) (X
(0)
, Y (0) )DYa1(0) DX
a2
(1) . . . DYn−1
(n−2) DX (n−1) .
In order to represent such long strings of operators in terms of the original operators
(X (0) , Y (0) ), we invert the dependence, i.e., we remember that, for instance,
−a2 −a1 −a2 a1
DX (1) = DY (0) DX (0) DY (0) ,
etc., which gives

(X (n) , Y (n) ) (5.33)
−a −an an an−1
= DY−a(0)1 DX
−a2
(0) . . . DY (0)n−1 DX (0) , (X
(0)
, Y (0) )DX (0) DY (0)
a2
. . . DX a1
(0) DY (0) .
We shall compute with bases of functions that are convenient in the asymptotic
regime. Let
|fμ,s ≡ e iμ(x−s)
2 /2
μ, s ∈ R. (5.34)
These functions constitute a basis at each μ:
∞
2π 2π
fμ,t |fμ,s = δ(s − t), ds|fμ,s fμ,s | = Id. (5.35)
μ −∞ μ
For two arbitrary real numbers w and γ , we have
2 #
e iwx /2 |fμ,s = e iμws /2 fμ+w,
2
s
1+w/μ
, (5.36)
1 #
e iγ ∂x /2 |fμ,s = √
2
f 1 , (5.37)
,s
1 + γ μ γ +1/μ
and
2π 2π
fμ,s |x|fμ,t = δ (s − t) + sδ(s − t), (5.38)
iμ2 μ
1 2π
fμ,s | ∂x |fμ,t = δ (s − t). (5.39)
i iμ
We now define the dimensionless variable x and set
√ √ 1 ∂
X|fμ,s = 4π h̄x · |fμ,s , Y |fμ,s = 4π h̄ · |fμ,s . (5.40)
i ∂x
The explicit formulas for the operators DX and DY acting on |fμ,s in the asymptotic
regime are
ai
|fμ,s = e i(ai −1)x
2 /2
e −i∂x e ix
2 2 /2
DX |fμ,s , (5.41)
a i(aj −1)∂x2 /2 −ix 2 i∂x2 /2
DYj |fμ,s = e e e |fμ,s . (5.42)
In order to establish the required recurrence relation, we must compare matrix
elements of the three consecutive length operators in the corresponding operatorial
decompositions:
(0) Y
Ast =fμ,s |X + |fμ,t ,
2
(j ) −a X a
Ast =fμ,s DY j | − Y − |DYj fμ,t ,
2
(j,i) −aj −ai Y ai aj
Ast =fμ,s DY DX |X + |DX DY fμ,t .
2
Now, using formulas (5.36)–(5.42), it is straightforward to show that

(j,i) (0) (j )
Ast = Ast + ai Ast , (5.43)
for all s, t, i.e., we again attain the recurrence relation (5.30) in the asymptotic regime,
but now for the matrix elements of the operators of the quantum proper lengths.
Estimates show that the corrections due to both the (operatorial) deviations from
the asymptotic regime and the error parameters εi (as for (5.30), (5.31)) decrease
exponentially with the index i, so the limit (5.4) exists in a weak operatorial sense. We
conclude that ratios (5.4) define a weakly continuous family of operators parameterized
by projective transverse measures on the freeway associated to a spine of the once-
punctured torus. This completes the proof of Theorem 5.1.
6 Conclusion
We hope to have added to the mathematical foundation and general understanding

of the quantization of Teichmüller space and its geometric underpinnings in the first
several sections of this exposition. We also hope that the survey given here of train
tracks and their extensions might be useful.
The quantization of Thurston’s boundary in general seems to be a substantial
project, which we have only just begun here with the quantization of continued frac-
tions. First of all, one would like a better understanding of the operators we have
constructed, for instance, an intrinsic characterization or an explicit calculational
framework for them. At the same time, our current constructions depend upon a
choice of spine, and there would seem to be a more invariant version of the theory,
where the choice of spine is dictated by the combinatorics of the cell decomposition
of Teichmüller space; the calculations in this chapter apply to each such spine (since
there is a combinatorially unique cubic one) for the once-punctured torus.
Second of all, the quantization of Thurston’s boundary for higher-genus or multi-
ply-punctured surfaces may be approachable using the improved quantum ordering.
Namely, in any fixed spine of the surface, there is a fixed finite family of “edge-simple”
closed edge-paths which by definition never twice traverse the same oriented edge. It
is elementary to see that any closed edge-path on may be written non-uniquely as a
concatenation of edge-simple paths, where the particular concatenation depends upon
a starting point. (Edge-simple paths were studied as “canonical curves” on train tracks
in [26]; they contain the extreme points of the polyhedron of projective measures on the
track.) It follows that an arbitrary leaf of a measured foliation carried by a freeway can
be written as a concatenation of paths from this finite collection of edge-simple paths.
The corresponding quantum operatorial statement results from the improved quantum
ordering described here. Thus, whereas the quantization of the once-punctured torus
devolved, in effect, to an analysis of two-letter words, the quantization of Thurston’s
boundary sphere in general may depend upon an analysis of words comprised of letters
which are edge-simple paths.
One appealing long-term goal would be to discover the Thurston classification
already on the operatorial level, for instance, with the dilatation in the pseudo-Anosov
case explicitly computable from the mapping class group operator or from the invariant
projective foliation operator.
Another intriguing aspect involves generalizations of graph length functions inso-
far as the proof of Theorem 5.1 holds taking as graph length any continuous positive
definite function which is homogeneous of degree one. A natural choice of such a
function is induced by the geodesic length of the corresponding geodesic curve taken
for a fixed basepoint in Teichmüller space on its fixed spine, for instance, vanishing
shear coordinates on the usual spine in the once-punctured torus. What sort of regular-
ity (e.g., piecewise smoothness) is achieved in the operators corresponding to points
of Thurston’s boundary under such “gauge fixing”?
Also worth mentioning are very recent advances in the description of quantum
sl(n, R) connections [11], where one finds an improved quantum ordering in a more
complicated higher-dimensional setting.
Appendix A. Combinatorial proof of Theorem 5.1
In this appendix, we give a complementary, combinatorial proof of the classical The-

orem 5.1 using the recurrence relation (5.7). At the present state of understanding, the
proof applies only to the classical case as we cannot control the quantum ordering.
Let us recall the structure of the matrix product (2.5). It is a sequence of matrices
LZ , RZ with different Z. It can be always segregated into clusters of matrices
LZ ≡ LZi+s LZi+s−1 . . . LZi
and
RZ ≡ RZj +k LZj +k−1 . . . LZj .
The periodic extension of expression (5.7) is always an alternating sequence of matrices

LZ and RZ :
PZ1 ,...Zn = . . . (LZ s RZ s−1 ) . . . (LZ · RZ · ) . . . .
First note that it is impossible to have arbitrarily long sequences of only left or right
matrices for a given graph: the maximum length is restricted to be less of equal the
maximum graph length of geodesics around holes. This means that the length of a
single cluster for a given graph is always bounded once the topology is fixed.
One can directly calculate the product (LZ 1 RZ 2 ) for LZ 1 = LZ1 . . . LZm and
RZ 2 = RZm+1 . . . RZm+k :
(LZ 1 RZ 2 ) = As1+ s2+ + B(s1− s2− + S1 s2− + S1 S2 )

(A.1)
+ D(s1+ s2− + s1+ S2 ) + P S1 s2+ ,
where Sj , sj± , for j = 1, 2, are the following coefficient functions:
m m+k
s1± = e ± i=1 Zi /2 , s2± = e ± j =m+1 Zi /2
m

q−1 m m+k

q−1 m+k
S1 = e+ r=1 Zr /2− r=q Zr /2
, S2 = e− r=m+1 Zr /2+ r=q Zr /2
,
q=2 q=m+2
and A, B, D, P are the special (2 × 2)-matrices (“letters”):

+1 0 0 0 0 −1 0 0
A= , B= , D= , P = . (A.2)
−1 0 0 +1 0 +1 −1 0
These letters possess interesting multiplication properties which are summarized in
the next lemma, whose proof is a routine calculation.
Lemma A.1 (The alphabet lemma). The multiplication table of letters (A.2) reads:
A B D P
A A 0 D 0
B P B B P , (A.3)
D A D D A
P P 0 B 0
so the trace of any product of these matrices is either unity or zero. In the product of t
matrices of form (A.1), the only monomials that survive are

(A + D)iα D B jβ (B + P ) . . . (A + D)iρ D B jω (B + P ) and (A + D)t , B t . (A.4)
The main point is that almost all cancellations of letters in long words are due to
the local multiplication rules (A.3). This means that, having a long sequence of letters,
say, LI +N from Lemma 5.2, we can split it into pieces depending on sequences of
letters LI , LI −1 , L̃I , and L̃I −1 , where the index I is also assumed to be big enough.
That is, let LI +N = LI LI −1 L̃I LI . . . LI comprise pN entries LI and L̃I and qN
entries LI −1 and L̃I −1 . We have then the following estimate:6
| log tr LI +N − pN log tr LI − qN log tr LI −1 | < C (qN + pN ), (A.5)
6 This estimate also follows from the properties of long geodesic lines in hyperbolic geometry: for two lines
of large lengths L1 and L2 intersecting at angle α, the length L3 of the third side of the resulting triangle is
L1 + L2 + log((1 − cos α)/2) + O(1/L). This also shows that our estimate is very rough.
where the constant C depends only on the Teichmüller space coordinates Zα and on
the genus and the number of holes of the Riemann surface, and we have also used
(5.8). The ratio of the coefficients is given by the continued fraction
1
qN /pN = , (A.6)
1
aI +1 +
aI +2 +
..
.
1
+
1
aI +N−1 +
aI +N
and also has a definite limit as N → ∞. Now the estimate follows: up to exponential
corrections, p.l.(L) coincides with the log tr L, so for any ε > 0, let us choose the
index I such that ε p.l.(LI −1 )/2 > C and ε p.l.(LI )/2 > C. Thus,
p.l.(L)(LI +N ) pN p.l.(L)(LI ) + qN p.l.(L)(LI −1 )
= (1 + O(ε/2)) , (A.7)
g.l.(LI +N ) pN g.l.(LI ) + qN g.l.(LI −1 )
and because the ratio qN /pN has a definite limit as N → ∞, there exists N0 such that
the relative error of this ratio times the sum of ratios of proper and graph lengths of
LI and LI −1 will not exceed ε/2. Thus, the collective relative error for such fixed I
and for all N > N0 is less than ε, proving the theorem.
Appendix B. Degeneracy of the Poisson structure

We shall explicitly calculate the degeneracy of the Poisson brackets (2.9) for a spe-
cial graph and choose the graph whose “building blocks” are depicted in Figure 15.
Namely, we have a line tree subgraph comprising edges Xi with attached subgraphs
as in Figure 15 (a) and 15 (b). Attaching a subgraph of type a corresponds to adding a
handle (increasing g by unity) while a subgraph of type b corresponds to adding a hole
(increasing s by unity). We shall assume that 2g + 2s > 5 to avoid the once-punctured
torus, which is already handled separately in Section 3.
For the variables A, B, C, D, E in Figure 15 (a) we have the Poisson bracket
(sub)matrix
Ai Bi Ci Di Ei
Ai 0 1 −1 0 0
Bi −1 0 1 1 −1
,
Ci 1 −1 0 1 −1
Di 0 −1 −1 0 2
Ei 0 1 1 −2 0
Di Ei Nj
@
@@
@@
@@
@
@@@
A AA
B@
i @ Ci
@ @ A A
@ A
Ai Aj
Xi Xi+1 Xj Xj +1
(a) (b)
Figure 15. Building blocks.
where the entries are the Poisson brackets between the corresponding variables. Adding
the last row to the next-to-the-last row as well as adding the last column to the next-
to-the-last column, then adding the third row to the second row as well as the third
column to the second column, we obtain the matrix
⎛ ⎞
0 0 −1 0 0
⎜ ⎟
⎜ 0 0 1 0 −2 ⎟
⎜ ⎟
⎜ ⎟
⎜ 1 −1 0 0 −1 ⎟ ,
⎜ ⎟
⎜ 0 0 0 0 2 ⎟
⎝ ⎠
0 2 1 −2 0
which obviously has rank four and can be further reduced (without adding the first
column or row to any other) to the form
⎛ ⎞
0 0 0 0 0
⎜ ⎟
⎜ 0 0 +1 0 0 ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 −1 0 0 0 ⎟.
⎜ ⎟
⎜ 0 0 0 0 +2 ⎟
⎝ ⎠
0 0 0 −2 0
Thus, erasing all columns and rows corresponding to the variables Bi , Ci , Di , and Ei
leaves invariant the rank of the Poisson bracket matrix.
Adjoining the subgraph in Figure 15 (b) creates exactly one degeneracy as the
variable Nj Poisson commutes with everything (as it must be when adding a hole).
It remains only to calculate the rank of the matrix corresponding to a tree graph with
edges Xi and Ai remaining after erasing all B-, C-, D-, E-, and N-variable rows and
columns. The corresponding Poisson bracket matrix has dimension 2g + 2s − 5 > 0
and the simple block-diagonal form
0 1 −1 0 0 · · ·
−1 0 1 0 0
1 −1 0 1 −1
0 −1 0 1 0 0
0 1 −1 0 1 −1 .
0 −1 0 1
..
0 1 −1 0 .
.. ..
. .
Adding each even-index row to its predecessor as well as adding each even-index
column to its predecessor, this reduces to the matrix whose only nonzero elements are
+1 on the main super-diagonal and −1 on the main sub-diagonal. Since this matrix
has corank one, the discussion is complete.
References
[1] F. Bonahon, Bouts des variétés hyperboliques de dimension 3. Ann. of Math. 124 (1986),
441–479. 610
[2] F. Bonahon, Shearing hyperbolic surfaces, bending pleated surfaces and Thurston’s sym-
plectic form. Ann. Fac. Sci. Toulouse Math. (6) 5 (1996), 233–297. 584, 587
[3] F. Bonahon and Y. Sözen, The Weil-Petersson and Thurston symplectic forms. Duke Math
J. 108 (2001), 581–597. 589, 610
[4] A. Bondal, A symplectic groupoid of triangular bilinear forms and the braid groups. Izv.
Ross. Akad. Nauk Ser. Mat. 68 (4) (2004), 19–74; English transl. Izv. Math. 68 (4) (2004),
659–708. 593, 603
[5] D. Bullock and J. H. Przytycki, Multiplicative structure of Kauffman bracket skein module
quantizations. Proc. Am. Math. Soc. 128 (3) (2000), 923–931. 603
[6] L. Chekhov and V. Fock, Talk given at St. Petersburg Meeting on Selected Topics in
Mathematical Physics, LOMI, 26–29 May, 1997; A quantum Techmüller space. Theor.
Math. Phys. 120 (1999), 1245–1259; Quantum mapping class group, pentagon relation,
and geodesics. Proc. Steklov Math. Inst. 226 (1999), 149–163. 580, 581, 582, 583, 587,
595, 602, 603
[7] D. B. A. Epstein and A. Marden, Convex hulls in hyperbolic space, a theorem of Sulli-
van, and measured pleated surfaces. In Analytical and geometric aspects of hyperbolic
space (Coventry/Durham, 1984), London Math. Soc. Lecture Note Ser. 111, Cambridge
University Press, Cambridge 1987, 113–253. 616
[8] L. D. Faddeev, Discrete Heisenberg–Weyl group and modular group. Lett. Math. Phys. 34
(1995), 249–254. 581, 595
Astérisque 66–67 (1979). 608, 609, 616
[10] V. V. Fock, Description of moduli space of projective structures via fat graphs. Preprint,
1994; arXiv:hepth/9312193. 588, 589
[11] V. V. Fock and A. B. Goncharov, Moduli spaces of local systems and higher Teichmüller
theory. Publ. Math. Inst. Hautes Études Sci. 103 (2006), 1–211; Cluster ensembles, quan-
tization and the dilogarithm. Preprint, 2004; arXiv:math.AG/0311245. 639
[12] V. V. Fock and A. A. Rosly, Poisson structures on moduli of flat connections on Riemann
surfaces and the r-matrix. In Moscow Seminar in Mathematical Physics, Amer. Math.
Soc. Transl. Ser. 2 191, Amer. Math. Soc., Providence, RI, 1999, 67–86. 580
[13] V. V. Fock and A. A. Rosly, Flat connections and poluybles. Theoret. and Math. Phys. 95
(1993), 526–535. 580
[14] W. M. Goldman, Invariant functions on Lie groups and Hamiltonian flows of surface group
representations. Invent. Math. 85 (1986), 263–302. 591, 592
[15] M. Havlíček, A. V. Klimyk, and S. Pošta, Representations of the cyclically symmetric
q-deformed algebra soq (3). J. Math. Phys. 40 (4) (1999), 2135–2161. 593, 601
Math. Phys. 43 (2) (1998), 105–115. 580, 581, 596
[17] R. M. Kashaev, Liouville central charge in quantum Teichmüller theory. Proc. Steklov
Math. Inst. 226 (1999), 62–70. 581
[18] R. M. Kashaev, On the spectrum of Dehn twists in quantum Teichmüller theory. In Physics
and Combinatorics (Nagoya 2000), World Sci. Publ., River Edge, N.J., 2001, 63–81. 596,
599, 634
[19] M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy
function. Comm. Math. Phys. 147 (1992), 1–23.
[20] J. E. Nelson and T. Regge, 2 + 1 quantum gravity. Phys. Lett. B 272 (1991), 213–216;
Invariants of 2 + 1 gravity. Comm. Math. Phys. 155 (1993), 561–568. 603
[21] J. E. Nelson and T. Regge, Homotopy groups and (2+1)-dimensional quantum gravity.
Nuclear Phys. B 328 (1989), 190–199; J. E. Nelson, T. Regge, and F. Zertuche, Homotopy
groups and (2 + 1)-dimensional quantum de Sitter gravity. Nuclear Phys. B 339 (1990),
516–532. 593, 603
[22] A. Papadopoulos and R. C. Penner, Enumerating pseudo-Anosov conjugacy classes. Pa-
cific J. Math 142 (1990), 159–173. 617
[23] A. Papadopoulos and R. C. Penner, The Weil-Petersson symplectic structure at Thurston’s
boundary. Trans. Amer. Math. Soc. 335 (1993), 891–904. 610, 618, 620
[24] R. C. Penner and J. L. Harer, Combinatorics of train tracks. Ann. of Math. Stud. 125,
Princeton University Press, Princeton, N.J., 1992. 609, 612, 613, 614, 615
[25] R. C. Penner, The decorated Teichmüller space of Riemann surfaces. Comm. Math. Phys.
113 (1988), 299–339. 581, 584, 588
[26] R. C. Penner, The action of the mapping class group on isotopy classes of curves and arcs
in surfaces. Thesis, Massachusetts Institute of Technology, 1982. 609, 638
[27] R. C. Penner, Weil-Petersson volumes. J. Differential Geom. 35 (1992), 559–608. 581,
589, 590
143–215. 581, 589, 590
[29] R. C. Penner, A construction of pseudo-Anosov homeomorphisms. Proc. Amer. Math. Soc.
104 (1988), 1–19. 584, 591, 617
[30] G. Rauzy, Echanges d’intervalles et transformations induites. Acta Arith. 34 (1979),
325–328. 626
[31] M. Reed and B. Simon, Methods of modern mathematical physics. I. Functional analysis,
second edition, Academic Press, Inc., New York 1980. 596
1984. 584
[33] J. Teschner, From Liouville theory to the quantum geometry of Riemann surfaces. Preprint,
2003; arXiv:hep-th/0308031. 581
[34] J. Teschner, An analog of a modular functor from quantized Teichmüller theory. Handbook
of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing House, Zurich
2007, 685–760. 581
Amer. Math. Soc. 19(1988), 417–431. 609, 610, 612, 616, 617
[36] W. P. Thurston, Minimal stretch maps between hyperbolic surfaces. Preprint, 1984;
arXiv:math.GT/9801039. 584, 587
[37] V. G. Turaev, Skein quantization of Poisson algebras of loops on surfaces. Ann. Sci. École
Norm. Sup. (4) 24 (1991), 635–704. 600, 602
[38] M. Ugaglia, On a Poisson structure on the space of Stokes matrices. Internat. Math. Res.
Notices 1999 (9) (1999), 473–493. 593
[39] E. Verlinde and H. Verlinde, Conformal field theory and geometric quantization. In Su-
perstrings ’89 (Trieste, 1989), World Scientific, River Edge, N.J., 1990, 422–449. 580,
581
[40] O. Ya. Viro, Lectures on combinatorial presentations of manifolds. In Differential Geom-
etry and Topology (Alghero, 1992), World Scientific, River Edge, N.J., 1993, 244–264.
[41] S. Wolpert, On the symplectic geometry of deformations of a hyperbolic surface. Ann. of
Math. 117 (1983), 207–234. 589
Chapter 15
Dual Teichmüller and lamination spaces
Vladimir V. Fock and Alexander B. Goncharov
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
2 Surfaces and triangulations . . . . . . . . . . . . . . . . . . . . . . . . . . 649
3 Laminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652
3.1 Unbounded measured laminations . . . . . . . . . . . . . . . . . . . 652
3.2 Bounded measured laminations . . . . . . . . . . . . . . . . . . . . . 656
4 Teichmüller spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
4.1 Teichmüller space of surfaces with holes T x (S) . . . . . . . . . . . . 660
4.2 Teichmüller space of decorated surfaces T a (S) . . . . . . . . . . . . 664
5 Tropicalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 668
6 Poisson and degenerate symplectic structures . . . . . . . . . . . . . . . . 669
7 Canonical pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
7.1 Additive canonical pairing . . . . . . . . . . . . . . . . . . . . . . . 672
7.2 Multiplicative canonical pairing . . . . . . . . . . . . . . . . . . . . 673
7.3 Proofs of main properties . . . . . . . . . . . . . . . . . . . . . . . . 675
Appendix A. Combinatorial description of D(S) . . . . . . . . . . . . . . . . . 676
Appendix B. Markov numbers . . . . . . . . . . . . . . . . . . . . . . . . . . 679
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
1 Introduction
The Teichmüller space of a surface S is the space of complex structures on S modulo
diffeomorphisms isotopic to the identity. For open surfaces this definition should be
made more precise since we need to specify the behaviour of the complex structure at
the boundary of the surface thus giving different versions of these spaces.
The space of measured laminations for a surface S is defined by William Thurston
[25]. A dense subset of the space of laminations is formed by the ones with finite
number of leaves – just collections of weighted closed curves without intersections
and self intersections. For open surfaces there are different versions of such spaces.
All versions of both Teichmüller and lamination spaces share the following prop-
erties:
648 Vladimir V. Fock and Alexander B. Goncharov
(1) They are topologically trivial.

(2) Isotopy classes of embeddings of surfaces induce canonical maps between these
spaces (forward or backward).
(3) There exist canonical compactifications of Teichmüller spaces by the quotients
of lamination spaces by multiplication by positive real numbers.
(4) There exists a canonical Poisson or degenerate symplectic structure (depending
on a version) on these spaces.
(5) A lamination with integral weights provides a function on Teichmüller space.
Such functions form a basis in the algebra of regular functions on Teichmüller
space with respect to a certain algebraic structure on it.
The original approach to Teichmüller spaces used extensively functional analysis
and was highly nonconstructive. The aim of this chapter is to give a description of all
these spaces for open surfaces using elementary algebra and geometry and make all
the properties obvious.
We also consider generalisations of Riemann surfaces – the so-called ciliated sur-
faces – surfaces with a distinguished set of points on the boundary. Though at the
beginning it requires a little more elaborate definitions, it provides us with more simple
examples of Teichmüller spaces containing most features of the general cases. Namely,
the configurations of points on the boundary of a disk or a cylinder fit perfectly into
the picture.
We included a somehow more technical Section 7 about the canonical pairing
between lamination and Teichmüller spaces in the case of ciliated surfaces.
The present exposition borrows a lot from [4] and Section 11 of [6]. We tried to
restrict ourselves to the purely geometric part of the picture and make it as elementary
as possible. It can be used as an elementary introduction to more algebraic subjects,
such as classical and quantum cluster varieties. To give the reader a further perspective,
let us briefly mention some other sources and related topics.
Explicit coordinate description of the Teichmüller spaces for an open surface goes
back to William Thurston [25] and Robert Penner [21]. This subject was developed
further in [4]. Quantum Teichmüller spaces were constructed in [5] and independently
by Rinat Kashaev in [17]. Jörg Teschner [24] proved that the quantum Teichmüller
space acts on the space of conformal blocks of the Liouville conformal field theory,
as was conjectured in [4], [5]. In [6] the higher Teichmüller spaces were defined,
and it was shown that they parametrise certain discrete, faithful representations of
the fundamental group of the surface in a split real simple Lie group G of higher
rank. These spaces are closely related to the ones studied by Nigel Hitchin [16]. In
particular, for G = SL3 (R) the corresponding higher Teichmüller space turns out
[8] to coincide with the space of real projective structures on S studied by William
Goldman and Suhyong Choi [14], [3]. In [6], [7] it was shown that the A and X
versions of the Teichmüller and lamination spaces can be obtained as the positive real
and tropical points of certain cluster A- and X-varieties. The closely related objects,
cluster algebras, were introduced by Sergey Fomin and Andrei Zelevinsky [11], and
Chapter 15. Dual Teichmüller and lamination spaces 649
studied by them and Arkady Berenstein, Mikhail Gekhtman, Mikhail Shapiro, Alek
Vainshtein ([12] and references therein). As a result of these developments, the cluster
theory was enriched by new examples as well as new features (such as duality, Poisson
structure and quantisation, relations to algebraic K-theory and to the dilogarithm).
In particular the canonical pairings from Section 6, in the special case of a ciliated
disc, can be viewed as the canonical pairings for cluster A and X-varieties of finite
type An , predicted by the general duality conjectures [7].
We are very grateful to Athanase Papadopoulos and Vladimir Turaev, who sug-
gested to write this chapter for the Handbook on Teichmüller spaces. The first author
is grateful to the Fields Institute and Brown University, where this exposition was
mainly written, for hospitality. The authors were supported by the NSF grants DMS-
0400449 (the second author) and CRDF 2622;2660 (the first author). We are very
indebted also to Andrei Levin, Yuri Neretin, Stepan Orevkov, Athanase Papadopou-
los, Robert Penner and especially to Aleksei Rosly for very valuable discussions and
for Guillaume Théret for carefully reading the text and correcting many typos and
mistakes.
2 Surfaces and triangulations

In this section we shall briefly recall basic facts about triangulations of surfaces.
A ciliated surface is a compact oriented surface with boundary and with a finite
set of marked points on the boundary called cilia.
A boundary component without cilia is called a hole.
A triangulation of a ciliated surface is a decomposition of the surface with
contracted holes into triangles such that every vertex of a triangle is either a cilium or
a shrunk hole.
1. g = 0, P = (7) 2. g = 0, P = (7, 0) 3. g = 0, P = (7, 3)
4. g = 1, P = (0)
Figure 1. Examples of triangulations of ciliated surfaces. Internal edges are thick.

The edges of the triangulation belonging to the boundary of the surface are called
external. Other edges are called internal. Denote by F (), E(), E0 (), V () the
set of triangles, edges, external edges and vertices of the triangulation, respectively.
Topologically a ciliated surface is defined by its genus g and a finite collection of
positive integers P = (p1 , . . . , ps ), where s is the number of boundary components
and pi is the number of cilia on the i-th component. There is no canonical order on
components. Denote the number of holes by h, the total number
the set of boundary
of cilia by c = i pi , and the number of internal edges by n.
Observe, that the numbers of faces, vertices and edges, both external and internal,
are determined by the topology:
(1) V () = h + c;
(2) E0 () = c;
(3) E() = 6g − 6 + 3s + 2c;
(4) n = 6g − 6 + 3s + c;
(5) F () = 4g − 4 + 2s + c.
The first two statements are obvious. The others follow from the expression for the
Euler characteristic of the surface,
F () − E() + V () = 2 − 2g + h − s,
and the expression
3F () = 2E() − E0 ()
of the fact that each triangle has three sides.
The topology of the triangulation can be encoded into a skew-symmetric matrix
εαβ , where α, β ∈ E(), defined as

ε αβ = α, i, β,
i∈F ()
where α, i, β is equal to +1 (resp. −1) if both α and β are sides of the triangle i and
α is in the counterclockwise (resp. clockwise) direction from β with respect to their
common vertex. Otherwise α, i, β is equal to zero. The entries of the matrix ε αβ
might have five possible values: 0, ±1, ±2.
The number of triangulations of a given ciliated surface is infinite except for g = 0
and for either P = (k) or P = (k, 0). However one triangulation can be obtained from
another one by a sequence of elementary moves called flips or Whitehead moves. One
triangulation is a flip of another if it is obtained by removing one internal edge and
replacing it by another diagonal of the arising quadrilateral (Figure 2). A flip can be
done with any internal edge unless it belongs to a unique triangle of the triangulation.
Observe that there is a canonical correspondence between the edges of a triangu-
lation and the edges of a flipped one. We shall use the notation α for the new edge
corresponding to the old edge α. However one should be aware that compositions
of a sequence of flips may restore the original triangulations but a nontrivial corre-
spondence between the edges. An example of this phenomenon can be illustrated by
Figure 16 where the composition of five flips does not change the triangulation of the
pentagon, but the canonical correspondence interchanges the diagonals.
@
@
@
@

@
@
@
@
Figure 2. A flip.

It is a simple exercise to check that the matrix εα β encoding the combinatorics
of a triangulation after a flip in the edge γ is given by the formula (unless one of the
triangle of the triangulation is glued to itself):

α β −εαβ if α = γ or β = γ ,
ε = αβ 1 αγ γβ (2.1)
ε + 2 (ε |ε | + |ε αγ |ε γβ ) otherwise.
To give some definitions we shall need to fix the orientation of the holes of the
surface. By that we mean fixing orientation of the boundary components in a way not
necessarily induced by the orientation of the surface. We say that the orientation of a
hole is positive (resp. negative) if it agrees (resp. disagrees) with the orientation of S.
For a given boundary component ρ with even number of cilia introduce a function
ζρ : E0 () → Z/2. First assign numbers to edges of by the following rule. If ρ
is a hole assign 1 to the corresponding vertex. If ρ is a boundary component with
positive even number of cilia assign 1/2 to vertices corresponding to every second
cilium of ρ and −1/2 to the remained vertices corresponding to ρ. Assign then 0 to
the vertices corresponding to other boundary components. The value ζρ (α) is just the
sum of numbers assigned to the ends of the edge α.
The mapping class group D(S) of a ciliated surface S is the group of connected
components of the diffeomorphisms of S preserving the set of cilia and the orientation
of S. In our examples of Figure 1 the mappings class groups are, respectively Z/7Z,
Z/7Z, Z/3Z × Z/7Z and PSL(2, Z). Observe that the mapping class group is finite
if and only if the number of triangulations is finite.
In 7.3 we give a combinatorial description of the mapping class group in terms of
the surface triangulations.
3 Laminations
Taking into account that the reader may be unfamiliar with Thurston’s notion of a
measured lamination [25], we are going to give all definitions here in the form which is
almost equivalent to the original one (the only difference is in the treatment of the holes,
punctures and cilia), but more convenient for us. The construction of coordinates on
the space of laminations we are going to describe is a slight modification of Thurston’s
“train tracks” ([25], Section 9).
It seems worth mentioning here that the definitions of measured laminations are
very similar to the definitions of singular homology groups, and is in a sense an
unoriented version of the latter ones.
There are two different ways to define the notion of measured lamination for
surfaces with boundary, which are analogous to the definition of homology group
with compact and closed support, respectively.
Measured laminations are certain collection of weighted curves on the surface. By
a curve here we mean a curve without self-intersections either closed or connecting
two points on the boundary, disjoint from cilia and considered up to homotopy within
the class of such curves. We call a curve special if it is retractable to a hole or to an
interval in the boundary containing exactly one cilium. We call a curve contractible if
it can be retracted to a point within this class of curves. In particular a nonclosed curve
is contractible if and only if it is retractable to a segment of the boundary containing
no cilia.
3.1 Unbounded measured laminations

Definition. A rational unbounded measured lamination or an X-lamination on a 2-
dimensional ciliated surface is given by orientations of some boundary components
and a homotopy class of a finite collection of non-selfintersecting and pairwise nonin-
tersecting curves with positive rational weights either closed or ending at the boundary
and disjoint from cilia, subject to the following equivalence relations:
(1) A lamination containing a contractible or a special curve is equivalent to the
lamination with this curve removed.
(2) A lamination containing two homotopy equivalent curves of weights u and v,

respectively, is equivalent to the lamination with one of these curves removed
and with the weight u + v on the other.
(3) Oriented boundary components are holes unless they are disjoint from the curves.
The set of all rational unbounded laminations on a given surface S is denoted by
Tx (S, Q). This space has a natural subset, given by collections of curves with integral
weights. This space is denoted by Tx (S, Z). We shall also omit the arguments if they
are clear from the context.
There is an action of the multiplicative group of positive rational numbers on

unbounded laminations Q>0 × Tx (Q) → Tx (Q) given by multiplication of the weights
of all curves by a fixed positive number.
If the surface S is open the set of laminations without closed curves form an open
dense subset of Tx (Q).
To each hole ρ we associate a number rρ equal to the total weight (resp. the negative
of total weight) of curves entering the hole if the orientation of the hole is positive
(resp. negative).
To each boundary component ρ with even number of cilia we associate a number
also denoted by rρ equal to the sum of total weights of curves entering the segments
between cilia taken with alternating signs.
Construction of coordinates. Let us first rotate each boundary component without

cilia infinitely many times in the direction prescribed by the orientation as shown in
Figure 3. Then deform the curves in such a way that they do not cross any edge of the
Figure 3. Twisting curves incident to a boundary component without cilia.
A
α
C
negative
positive
Figure 4. Positive and negative intersection points.
triangulation consecutively in opposite directions. Consider a quadrilateral formed

by two triangles sharing an edge α. Denote its vertices by ABCD in the clockwise
direction with respect to the orientation, starting from a vertex of α. We call an
intersection point of the edge α and a curve of the lamination positive if it belongs to a
segment of a curve connecting AB and CD and negative if it belongs to a segment of
a curve connecting BC and AD. Since the curves of the lamination do not intersect,
having both negative and positive intersection points on the same edge is impossible.
Notice that we do not assign a sign, say, to a curve intersecting BC and CD.
Now assign to each edge α the sum over the positive intersection points of weights
of the respective curves minus the sum over the negative intersection points of weights
of the respective curves. The collection of these numbers, one for each internal edge
of , is the desired set of coordinates.
Observe that although the curves spiralling around the holes without cilia intersect
some edges infinitely many times, only a finite number of these intersection points are
positive or negative, and therefore all numbers on edges obtained is this way are finite.
Now we need to prove that these numbers are global coordinates indeed, i.e., that
the correspondence between rational unbounded measured laminations and rational
numbers on edges is one-to-one. We shall do it by describing an inverse construction.
Note that if we are able to construct a lamination corresponding to the set of
numbers {xα }, we can also do it for the set {uxα } for any rational u ≥ 0. Therefore
we can reduce our task to the case where all numbers on edges are integral.
Take a triangle, parametrise each side of it by R respecting the orientation induced
by the orientation of the triangle, and connect each point with parameter i ∈ 21 + Z≥0 on
one side to the point with parameter −i on the next side in the clockwise direction with
respect to the orientation of the surface. This can be done by curves without mutual
intersections. Now replace every triangle of our triangulation with such triangles with
curves in such a way that the point with parameter i on one side of an edge α is glued
to the point with parameter xα − i on the other side (Figure 5).
Figure 5. Gluing triangles for xα = 2.
Observe that although we have started with an infinite bunch of curves, the resulting
lamination is finite: all these curves glue together into a finite number of connected
components and a possibly infinite number of special curves. Indeed, any connected
component either intersects positively or negatively at least oneedge or is closed.
Since the total number of positive or negative intersection points α∈E()−E0 () |xα |
is finite, the resulting lamination contains no more than this number of connected
components. In particular the number of connected components equals this number

provided all numbers xα are all nonpositive or all nonnegative.
Properties of the coordinates. The constructed coordinates on the space of lamina-

tions correspond to a particular choice of triangulation. If we change the triangulation,
the corresponding coordinates change. The rule for changing the coordinates under a
flip of the triangulation is given by an explicit formula:
⎧
⎪ α
⎨−x if β = α,
β β αβ α
x = x + ε max(0, x ) if εαβ ≥ 0, (3.1)
⎪
⎩ β αβ α αβ
x + ε max(0, −x ) if ε ≤ 0.
If all edges of the quadrilateral taking part in the flip are different, the change of
coordinates can be shown by the graphical rule Figure 6.
@ x1 + max(x0 , 0)
@ x2 − max(−x0 , 0)
@ @
x1 @ x2 @
@ @
@ @
@ ←→ −x 0 @
x0
@ @
@ @
@
4
@
x @ x3 @
@ x4 − max(−x0 , 0)@ x3 + max(x0 , 0)
@ @
Figure 6. Changing coordinates under a flip of the triangulation. Only the changing coordinates
are shown here. The numbers on the other edges remain unchanged.
An X-lamination is integral if all its coordinates are integral.

If ρ is a boundary component with even number of cilia, the canonical map
rρ : Tx → Q is given by rρ = α ζρ (α)x.
Given an edge α a lamination with coordinates xα = 1, xβ = 0 for β = α is just
a single curve homotopy equivalent to the edge α with the ends shifted slightly from
cilia in the counterclockwise direction. A lamination having nonnegative coordinates
is just a union of such curves. It implies that for any lamination without closed curves
there exists a triangulation such that all coordinates of the lamination are nonnegative.
Conversely if a lamination have positive coordinates with respect to a triangulation
then such triangulation is unique.
3.2 Bounded measured laminations

Definition. A rational bounded measured lamination or a rational A-lamination on
a 2-dimensional ciliated surface is a homotopy class of a collection of finite number
of self- and mutually nonintersecting unoriented curves either closed or connecting
two points of the boundary disjoint from cilia with rational weights and subject to the
following conditions and equivalence relations.
(1) Weights of all curves are positive, unless a curve is special.
(2) A lamination containing a curve of weight zero is considered to be equivalent to
the lamination with this curve removed.
(3) A lamination containing a contractible curve is considered to be equivalent to the
lamination with this curve removed.
(4) A lamination containing two homotopy equivalent curves with weights u and v
is equivalent to the lamination with one of these curves removed and with the
weight u + v on the other.
Recall that a contractible curve in this context is either closed or retractable to a
segment of the boundary without cilia.
The set of all rational bounded laminations on a given surface S is denoted by
a
T (S, Q). This space has a natural subset called the set of integral bounded lamina-
tions, consisting of laminations with integral weights. It is denoted by Ta (S, Z). We
shall omit the arguments of Ta if they are clear from the context or if the statement is
valid for any value of the arguments.
The space Ta (S) has a subspace T0a (S) consisting of the laminations such that for
any segment of the boundary between two cilia the total weight of curves ending at it
vanishes.
There is a canonical action of the multiplicative group of positive rational numbers
on bounded lamination Q>0 × Ta (Q) → Ta (Q) given by multiplication of weights of
all curves by a fixed positive number.
For every hole ρ and for every connected component with even number of cilia
there is an action rρ of the additive group of rational numbers rρ : Q × Ta → Ta .
For a hole this group acts by adding a closed loop with a certain weight around the
hole. Similarly, for every boundary component with even number of cilia ρ the group
acts by adding a special curve about every second cilium with a certain weight w and
adding a special curve with weight −w about the remaining cilia of ρ.
For every hole ρ there is a map Aρ : Ta → Q (called a collar map [20], where it
was defined and studied) given by the total weight of the loops surrounding the hole.
There is a canonical map p : Ta (S) → Tx (S) since every bounded lamination can be
considered as an unbounded one. This map obviously commutes with multiplication
by positive numbers and sends integral laminations to integral ones.
Construction of coordinates. Suppose we are given a triangulation of a cili-

ated surface S. We are going to assign, for a given lamination, rational numbers on
edges of and show that these numbers are global coordinates on the space of lami-
nations.
Deform the curves of the lamination in such a way that every curve intersects
every edge at the minimal possible number of points. Assign to each edge α half of
the sum over the intersection points of weights of the respective curves. Collection
of numbers, one for each edge of , is the desired set of coordinates. For the space
T0a (S), the numbers assigned in this way to any external edge vanishes. Thus only
numbers assigned to internal edges serve as coordinates for this space.
Reconstruction. Now we need to prove that these numbers are coordinates indeed,
i,e., that they provide a bijection between the space of bounded measured laminations
and collections of rational numbers on edges. For this purpose we just describe an
inverse construction, which gives a lamination, starting from numbers on edges.
Let us first fix a pair of rationals u and v and introduce the numbers {ãα | α ∈ E()}
by
ãα = uaα + v.
By choosing u and v one can make the numbers {ãα } to be positive integers and to
satisfy the triangle inequality for every triangle of the triangulation with edges α, β, γ :
|ãα − ãβ | ≤ ãγ ≤ ãα + ãβ . (3.2)
Now mark 2ãα points on each edge α and join the points on the sides of each triangle
pairwise by nonintersecting and non-selfintersecting lines connecting points on dif-
ferent sides and passing inside the triangle. One can easily see that once the triangle
inequality is satisfied this can be done in a unique way up to homotopy (Figure 7).
Now take each hole and each cilium and add a closed curve surrounding them with
Figure 7. (ãα , ãβ , ãγ ) = (3, 4, 5).
weight −v. Then divide the weights of all curves by u. The construction is finished.
The proof that the construction is indeed inverse to the previous one is self-evident.
Properties of the coordinates. Just as for the case of X-laminations, the constructed
coordinates on the space of laminations correspond to a particular choice of the trian-
gulation. The rule stating how the coordinates change under a flip of triangulation is
given by the formula:

βδ βδ
max δ|εβδ >0 ε aδ , − δ|εβδ <0 ε aδ − aα if β = α,
aβ = (3.3)
aβ if β = α.
Since any triangulation change is a composition of flips, these rules allow to express
the coordinate change for any triangulation change.
If all edges of the quadrilateral taking part in the flip are different, the coordinates
change can be shown by the graphical rule on Figure 8.
@ @
@ @
a1 @ a2 a1 @ a2
@ @
@ @
@ max(a1 + a3 , a2 + a4 ) − a0 @
a0 ←→
@ @
@ @
@ @
a4 @ a3 a4 @ a3
@ @
@ @
Figure 8. Change of coordinates under a flip of the triangulation. Only the changing coordinates
are shown here, the numbers on the other edges remain unchanged.
A bounded lamination is integral if and only if all coordinates are integral or half
integral and for every triangle with sides α, β, γ the sum aα + aβ + aγ is integral.
If we multiply a lamination by a positive number, the coordinates get multiplied
by the same number.
A bounded lamination is called even if all coordinates are integers. One can easily
check that this condition does not depend on the coordinate system. This set is denoted
a . Observe that an even lamination divided by 2 is not necessarily integral.
by Tev
The canonical map p : Ta → Tx is given by xα = − β εαβ aβ .
The action rρ of the additive group Q corresponding to a boundary component ρ
with even number of cilia is given by
rρ (w) : aα → wζρ (α) + aα .
If ρ is a hole, then the collar map Aρ is given by
Aρ = max(aαt + aβt − aδt ),

t
where t runs over all triangle having ρ as a vertex, αt and βt are the edges of t incident
to ρ and δt is the third edge.
Real lamination spaces. Since the transformation rules for coordinates (8) and (6) are
continuous w.r.t. the standard topology of Qn , the coordinates define a natural topology
on the lamination spaces. One can now define the spaces of real measured laminations
(resp. bounded and unbounded) as a completion of the corresponding spaces of rational
laminations. These spaces are denoted as Ta (R) and Tx (R), respectively. Of course,
we automatically have coordinate systems on these spaces.
Observe that to define real measured laminations it is not enough to replace rational
numbers by real numbers in the definition of the space of laminations unless the surface
S is a disc or a cylinder with one hole. Such definition would not be equivalent to the
one above since a sequence of more and more complicated curves with smaller and
smaller weights may converge to a real measured lamination which is not presentable
by a finite collection of curves.
4 Teichmüller spaces
The Teichmüller space T (S) (resp. moduli space M(S)) of a closed surface S is the
space of complex structures on S modulo diffeomorphisms isotopic to the identity
(resp. modulo all diffeomorphisms). We are going to give two different extensions of
this notion to the case of open surfaces with cilia. But before we recall some basic
facts about relations between complex structures, constant negative curvature metrics
and discrete subgroups of the group PSL(2, R). For more details we recommend the
reviews [1] and [19].
Consider first a vicinity of a boundary component. Topologically it is a cylinder, but
as a complex surface it can be isomorphic either to a cylinder {z ∈ C | 0 < r < |z| < 1}
or to a punctured disk {z ∈ C | 0 < |z| < 1}. The boundary components of the second
kind are called punctures.
According to the Poincaré uniformisation theorem any complex surface S (except
a sphere with less than three holes and a torus without holes) can be represented as
a quotient of the hyperbolic plane H by a discrete subgroup of its automorphism
group (sometimes called the Möbius group) PSL(2, R) of real 2 × 2 matrices with unit
determinant considered up to the factor −1. The group is defined up to conjugation,
is discrete as a subgroup of PSL(2, R), is canonically isomorphic to the fundamental
group of the surface π1 (S) and acts freely on H . The subgroups with these properties
are called Fuchsian groups. In particular, this implies that the Teichmüller space T (S)
for a surface S without cilia is isomorphic to a connected component in the space of
Fuchsian groups considered up to conjugation. The connected component is singled
out by the topology of the quotient of the upper half plane. The group corresponding
to a given complex structure is called the monodromy group.
Recall that an element g of PSL(2, R) is called hyperbolic (resp. parabolic, elliptic)
if it has two distinct real eigenspaces (resp. one one-dimensional real eigenspace, two
conjugate complex eigenspaces). Since eigenspaces are stable points of the action
of g on CP 1 , only parabolic and hyperbolic elements acts on H freely and thus all
elements of must be of these two types.
The quotient of H by the action of the subgroup generated by a parabolic element is
a punctured disk. Thus the only elements of π1 (S) which can be mapped to parabolic
elements are loops surrounding punctures.
On H there exists a unique PSL(2, R)-invariant curvature −1 Riemannian metric
given in the standard coordinates by (Im z)−2 dzd z̄. It induces a metric on S. Since
this metric is of negative curvature, any homotopy class of closed curves contains a
unique geodesic unless a curve surrounds a puncture. In the latter case the length of
the curve can be made arbitrarily small and. In this case we say that the geodesic
length is zero.
Any nontrivial element of has one or two fixed points on RP 1 , which is the
boundary (absolute) of H . The singular set of a ciliated surface S is the subset of
RP 1 consisting of all fixed points of all elements of the monodromy group and
of preimages of all cilia. This set is -invariant by construction and thus its convex
hull in H is also -invariant. The convex core S0 of the surface S is the quotient of
this convex hull by . To describe it geometrically in terms of the original surface
S consider the closed geodesics surrounding holes as well as geodesics connecting
adjacent cilia on the same boundary components. Cut out of the surface the pieces
facing the respective boundary components. These pieces as Riemannian manifolds
are isomorphic either to the positive quadrant of H or its quotients by a cyclic group.
The remaining part of the surface S is just its convex core S0 .
For example, if our surface is a ciliated disk, its convex core is an ideal polygon.
Homotopy classes of closed curves on a surface are in one-to-one correspondence
with the conjugacy classes of its fundamental group. Denote by γ an element of π1 (S)
and by l(γ ) the length of the corresponding geodesic. Then a simple computation
shows that
λ1

l(γ ) = log , (4.1)
λ2
where λ1 and λ2 are the eigenvalues of the monodromy along the loop γ . This
number is obviously well defined, i.e., it does not depend on the choices of particular
representation of π1 (S), of a particular element of π1 (S) representing a given loop,
and of a particular 2 × 2 matrix representing the element of PSL(2, R). This formula
agrees with the convention that the length of the geodesic surrounding a puncture is
zero.
4.1 Teichmüller space of surfaces with holes T x (S)

Definition. The Teichmüller space of a ciliated surface with holes or the Teichmüller
X-space, denoted by T x (S), is the space of complex structures on S together with
an orientation of all holes except the punctures up to the diffeomorphisms homotopy
equivalent to the identity and preserving cilia pointwise.
It is not a priori obvious that this space possesses a natural topology in which it is
connected, but this will be clarified later.
For every hole ρ there exists a map r ρ : T x (S) → R>0 given by the exponential
of the length of the geodesic surrounding this hole (resp. exponential of minus the
length of the geodesics surrounding the hole) if the orientation of the hole agrees
(resp. disagrees) with the orientation of the surface.
For every boundary component with even number of cilia there exists a map denoted
by the same letter r ρ : T x (S) → R>0 defined as follows. Take an horocycle about one
cilium. Then take another one about the next cilium tangent to the first one. Repeat
this procedure until we come back to the original cilium. The value of r ρ is e±l , where
l is the distance between the two horocycles about the first cilium and the sign depends
on whether the final horocycle is inside or outside the original one. The value of r ρ
does not depend on the choice of the original horocycle and may change to its inverse
if we have chosen another starting point.
Construction of coordinates. Let be a triangulation of S. For any point of T x (S)

we are going to describe a rule for assigning a positive real number to each internal
edge of . The collection of these numbers will give us a global parameterisation of
T x (S).
Every edge of connecting two cilia, a cilium and a puncture, or two punctures can
be made geodesic in a canonical way. In order to make geodesic the edges connecting
a cilium or a puncture with a hole first draw a geodesic around each hole and choose
an arbitrary point on it. Then deform the edge to a geodesic ray connecting the cilium
and the chosen point on the geodesic. Observe that the orientation of the hole induces
an orientation of the geodesic. Now move the chosen point along the geodesic in the
direction given by this orientation. The limiting position of the ray is just the desired
one. By construction it is spiralling around the hole. Analogous procedure can be
done for the edges connecting two holes giving a geodesic spiralling around one closed
geodesic at one end and around another one at another end.
We end up by a triangulation of a part (actually of the convex core S0 ) of the surface
S into ideal triangles.
Now lift the triangulation to the upper half plane and consider an edge together
with two adjacent triangles forming an ideal quadrilateral. The quadruple of points of
this quadrilateral have one invariant under the action of PSL(2, R). For our purpose it
is convenient to choose the cross-ratio x by choosing a coordinate on H (sending it to
the upper half plane) equal to 0 and ∞ at the ends of the edge and to −1 at the third
vertex of the quadrilateral (Figure 9). Then the value of the coordinate at the fourth
vertex is the desired parameter x associated to the chosen edge. It is an easy exercise
to show that there are two ways to choose the coordinate and both give the same value
of the cross-ratio.
Reconstruction. Our goal now is to construct a ciliated surface with hyperbolic met-
ric starting from a triangulation of a ciliated surface with real positive numbers {x α }
−1 0 x
Figure 9. Definition of the cross-ratio.
assigned to internal edges. First of all we give a simple prescription for how to re-
store orientations of the boundary components from these data: The orientation of a
boundary component corresponding to a hole γ is just induced from the orientation
of the surface (resp. opposite to the induced one) if the sum x α over all α incident
to γ is positive (resp. negative). If the sum is zero, it means that it is not a hole, but a
puncture.
The construction of the surface itself can be achieved in two equivalent ways. We
shall describe both since one is more transparent from the geometric point of view and
the other is useful for practical computations.
Construction by gluing. We are going to glue the convex core S0 of our surface
S out of ideal hyperbolic triangles. The lengths of the sides of ideal triangles are
infinite and therefore we can glue two triangles in many ways which differ by shifting
one triangle with respect to another along the side. The ways of gluing triangles
can be parameterised by the cross-ratios of four vertices of the obtained quadrilateral
(considered as points of RP 1 ) just as defined by Figure 9.
Now consider the triangulated surface S. Replace every triangle of the triangulation
by an ideal one and glue them together along the edges just as they are glued in S
using numbers assigned to the edges as gluing parameters.
Note, although this is not quite obvious that the resulting surface is not necessarily
complete even if the surface has no cilia. In fact it is the complex core S0 with boundary
components corresponding to holes removed.
Construction of the Fuchsian group. We are now going to construct a discrete
monodromy subgroup of PSL(2, R) starting from a triangulation of S with positive
real numbers on edges. Construct first a graph out of the triangulation in the following
way (Figure 10). Draw a small edge of the graph transversal to every edge of the
triangulation and for every triangle connect the three ends of edges in it pairwise.
Orient the edges of the arising triangle
1 1in
the clockwise direction. Now assign to each
of these edges
the matrix U = −1 0 . Assign to the remaining edges the matrix
(x α )1/2
B(x α ) = 0
−(x α )−1/2 0
, where α is the edge of the triangulation it intersects.
H
HH
A H
HH

AH
* HH
6HU α U
HH HH
U j U H
H
B(x α )H
Y
HH HH

HH U U H?
H A
H
HH
A
HH
Figure 10. Graph for computing monodromy for T x (S). The original triangulation is shown by
the thin lines.
Now to any oriented path on this graph we can associate a matrix by multiplying
consecutively all matrices we meet along it, taking U −1 instead of U each time the
orientation of the path disagrees with the orientation of the edge. (The orientations
of the edges transversal to the edges of the triangulation are not to be taken into
account, since B(x)2 = −1 and therefore B(x) coincides with its inverse in the group
PSL(2, R).) In particular, if we take closed paths starting form a fixed vertex of the
graph, we get a homomorphism of the fundamental group of to the group PSL(2, R).
The image of this homomorphism is the desired group .
The fact that these two constructions are inverse to the above construction of
coordinates is almost obvious, especially for the first one. The only note we would
like to make here is to show where the matrices I and B(x) came from. Consider two
ideal triangles in the upper half plane with vertices at the points −1, 0, ∞ and x, ∞,
0, respectively. Then the Möbius transformation which permutes the vertices of the
first triangle is given by the matrix I , and the one which maps one triangle to another
(respecting the order of vertices given two lines above) is given by B(x).
Properties of the coordinates. The constructed coordinates on Teichmüller space

correspond to a particular choice of triangulation. The rule of coordinate change for
a flip is given by the formula
⎧
⎪ α −1

⎨(x ) if β = α,
β β α ε αβ
x = { x (1 + x ) if εαβ ≥ 0, (4.2)
⎪
⎩ β α −1 ε αβ αβ
x (1 + (x ) ) if ε ≤ 0.
If all edges of the quadrilateral taking part in the flip are different, the coordinate
change can be shown by the graphical rule Figure 11.
The canonical map r ρ : T x → R is given by

rρ = (x α )ζ (α) , (4.3)
α
where the product is taken over all edges of the triangulation .
@ @
@ @
x1 @ x2 x 1 (1 + x 0 ) @ x 2 (1 + (x 0 )−1 )−1
@ @
@ −1 @
@ (x )
0
@
x0 ←→
@ @
@ @
@ @
x @
4 x3
x 4 (1 + (x 0 )−1 )−1@ x 3 (1 + x 0 )
@ @
@ @
Figure 11. Changing of coordinates under a flip of the triangulation. Only the changing coor-
dinates are shown here, the numbers on the other edges remain unchanged.
Integral monodromy. It is obvious form the reconstruction formulae that if all co-
ordinates are equal to 1 then the monodromy group is a subgroup of PSL(2, Z). Of
course the subgroup obtained in this way is a finite index subgroup and is free (and
thus torsion free). It turns out that a kind of converse statement is also true. Namely
for any torsion free finite index subgroup ⊂ PSL(2, Z) there exists a surface S with
a triangulation || which is unique up to the mapping class group action such that is
the monodromy group of a hyperbolic structure on S corresponding to all coordinates
equal to 1. Thus we have an explicit bijection between conjugacy classes of torsion
free finite index subgroups of PSL(2, Z) and triangulated surfaces.
We shall give an explicit construction of the triangulated surface out of a torsion
free discrete subgroup of PSL(2, Z) borrowed from Kontsevich [18].
Observe that the group PSL(2, 0Z)1 is

the free product of the subgroups Z/2Z gener-

ated by the matrix σ = B(1) = −1 0 and Z/3Z generated by the matrix υ = −1 0 .

1 1
Every nontrivial torsion element of PSL(2, Z) is conjugated to either σ or υ. Consider

the finite set F = \ PSL(2, Z). Since is torsion free, the action of the subgroups
Z/3Z and Z/2Z on F is free. For every Z/3Z orbit f, υf, υ 2 f take a triangle with
vertices at the points of this orbit. Then for every Z/3Z orbit f, σf glue together the
triangles f, υf, υ 2 f to the triangle σf, υσf, υ 2 σf along one side sending f to υσf
and σf to τf . We get the desired triangulated surface without boundary and such that
the triangles are numerated by the orbits of τ , edges are the orbits of σ and vertices
are orbits of υσ .
4.2 Teichmüller space of decorated surfaces T a (S)

In this section we are going to reproduce some results of Penner [21].
Before giving a definition of the Teichmüller space T a (S), recall what is a horo-
cycle. A horocycle in the upper half plane H at A ∈ RP 1 is a limit of a circle passing
through a fixed point on H when its hyperbolic centre tends to A. Equivalently a
horocycle at A is an orbit of a subgroup of PSL(2, R) stabilising A. A horocycle at the

point A ∈ R ⊂ RP 1 is a Euclidean circle belonging to the upper half plane tangent
to the real axis at A. A horocycle at ∞ ∈ RP 1 is a horizontal line. Points of one
horocycle are equidistant from another horocycle based at the same point. The set of
all horocyles is isomorphic to the space (R2 − {(0, 0)})/ ± 1. This isomorphism is
equivariant with respect to the PSL(2) action and sends the vector (x, y) ∈ R2 into a
horocycle about the point x/y and of euclidean diameter y −2 .
For a punctured Riemann surface S = H /, where is a discrete subgroup of
PSL(2, R), a horocycle at a cilium or a puncture A is a -invariant set of horocycles
at each preimage of A on RP 1 . By definition the area of a horocycle at the puncture is
the area inside the horocycle of the quotient of H by the subgroup of stabilising the
basepoint. If the horocycle is small enough, its image on S is a small circle surrounding
a puncture or tangent to the boundary at the cilium and orthogonal to any geodesics
coming out of this puncture or cilium. In this case its area coincides with the actual
area inside the image. However the projection of a general horocycle to the surface
may have a relatively complicated topology and its area inside the image differs from
the area of the horocycle.
We say that two horocycles on the ciliated surface S are tangent along the path γ
connecting their base points if the corresponding horocycles at the ends of the lift of
the path to the universal covering H are tangent.
Definition. A decorated ciliated Riemann surface is a ciliated hyperbolic surface

such that every hole is a puncture and with a horocycle chosen at each puncture and
cilium. The Teichmüller space of decorated ciliated surfaces is called the Teichmüller
A-space and is denoted by T a (S).
The space T a (S) has a canonical subspace T0a (S) consisting of decorated surfaces
such that for every path connecting two cilia and retractable to a segment of the
boundary without cilia, the corresponding horocycles are tangent.
Forgetting the horocycles we get a canonical map p : T a → T x .
For every cilium and for every hole ρ there is an action of the multiplicative
group of positive numbers rρ : R>0 × T a → T a given by changing the size of the
corresponding horocycle.
For every hole ρ there is an area map Aρ : T a → R>0 given by the area inside the
corresponding horocycle.
There is a canonical map p : T a → T x given by forgetting horocycles.
Construction of coordinates. Let be a triangulation of S. For any point of T a (S),

we are going to assign a positive real number to each edge of . The collection of
these numbers will give us a global parameterisation of T a (S).
Consider an edge α of the triangulation, deform it into a geodesic and lift it to
the universal covering H . There are the horocycles attached to the ends of the edges.
The number aα which we assign to the edge α is the exponential of half of the length
(resp. of minus the length) of the segment of the lifted edge between the intersection
l
l
aα = el/2 aα = e−l/2
Figure 12. Penner coordinates.
points with the horocycles if the horocycles do not intersect (resp. if the horocycles
do intersect), as shown on Figure 12.
This number can be easily computed algebraically using the correspondence be-
R . xIf x(x

1 , y1 ) and (x2 , y2 ) are vectors, corresponding
tween horocyles and vectors in 2
to the horocycles, then aα = det y11 y22 .

Reconstruction. This is quite similar to that of holed surfaces. There is a canonical

mapping p : T a (S) → T x (S) which just forgets the horocycles and is given explicitly
in coordinates by (4.5). Therefore to reconstruct the surface itself we can apply the
reconstruction procedure for T x (S). To reconstruct the horocycles, consider an ideal
triangle which we have used to glue the surface. On each edge, the a length of the
corresponding geodesics between the horocycles allows us to restore unambiguously
intersection points of the horocycles with the edges.
Observe that given two points A and B on RP 1 and a horocycle about A one can
associate a canonical coordinate z on the upper half plane H , such that z(A) = ∞,
z(B) = 0 and the horocycle is given by the line Im z = 1. Therefore given a lift of an
oriented edge of the triangulation to the upper half plane we get a canonical coordinate
on the latter taking for A the beginning of this edge and for B its end. Changing the
orientation of theedge results
in a Möbius transformation of the coordinate with the
0 a
matrix D(aα ) = −a −1 0α and passing from the side α of a triangle with the edges
α
α, β, γ to the edge β with orientations towards the same vertex results in the Möbius
aγ
1 0
transformation with the matrix F aα aβ = a γa 1 .a
α β
Similarly to the case of the space T x (S) one can give a rule to find the corresponding
Fuchsian group as a monodromy group of a connection on a graph. Construct first a
graph out of the triangulation in the following way (Figure 13). Draw a small edge
of the graph along every edge of the triangulation and for every triangle connect the
six ends of edges to form a hexagon. Orient the edges surrounding every vertex of
the triangulation in the clockwise direction. Assign to each edge of the graph parallel
to the edge α the matrix D(a α ) and
assign to the edge connecting sides α and β of
a
the triangle αβγ the matrix F aα γaβ . Now for any closed path on the surface one can
H
H
aH
γ
H

F aα βaγ HH
Y
HH H H
HH H
H D(a H HH
H
H H
γ)H HH

H
6 H
H
H

α D(aα )
HH
H F
aα
H
HH
aγ aβ

HH
H
HH?
H
HH
*HH D(aβ )

HH
HHH H H
HHH aγ H
H
j

HH F aβ aα
β
H
HH
Figure 13. Graph for reconstructing the monodromy for T a (S).
retract this path to the graph and the monodromy operator can be computed by the
product of the corresponding matrices on the edges.
Moreover this graph with matrices allows to compute the signed distance between
any two horocycles. Indeed, consider the path connecting two corresponding vertices
of the triangulation. Move each endpoint along an edge of the triangulation to the
nearest vertex of the graph and then deform it to pass along the edges of the graph.
Associate to it the product of the matrices along this path. Then the absolute value
of the upper right element of this matrix is equal to the exponential of signed length
between the horocycles. The result does not depend on how we have deformed the
path since the monodromy along the edges surrounding the vertex is strictly lower
triangular.
Properties of the coordinates. Just as for the three cases considered above, the
constructed coordinates correspond to a particular choice of triangulation. The rule
describing how the coordinate changes under a flip of an edge α is given by
⎧
⎪ εβδ −εβδ
⎨ δ|εβδ >0 aδ + δ|εβδ <0 aδ
aβ = if β = α, (4.4)
⎪ aα
⎩a if β = α.
β
If all edges of the quadrilateral taking part in the flip are different, the coordinate
change can be shown by the graphical rule shown on Figure 14.
The canonical map p : T a → T x is given by
αβ
xα = (aβ )−ε . (4.5)
β
@ @
@ @
a1 @ a2 a1 @ a2
@ a1 a3 + a2 a4 @
@ @
@ a0 @
a0 ←→
@ @
@ @
@ @
a4 @ a3 a4 @ a3
@ @
@ @
Figure 14. Changing of coordinates under a flip of the triangulation. Only the changing coor-
dinates are shown here, the numbers on the other edges remain unchanged.
The action rρ of the multiplicative group R>0 corresponding to a hole with even
number of cilia is given by
rρ (w) : aα → wζρ (α) aα .
If ρ is a hole, then the area map Aρ is given by
aδ
t
Aρ = ,
t
a α a
t βt
where t runs over all triangle having γ as a vertex, αt and βt are the edges of t incident
to γ and δt is the third edge.
The subspace T0a is the space where all numbers assigned to external edges are
equal to one.
5 Tropicalisation
In this short section we shall state explicitly in which sense the spaces of laminations
are “tropicalisations” of the corresponding Teichmüller spaces.
Recall that a set F with two binary operations: addition + and multiplication · is
called a semifield if it satisfies the following axioms:
(1) F is an Abelian semigroup with respect to addition.
(2) F is an Abelian group with respect to multiplication.
(3) Addition and multiplication are related by the distributivity law: a · (b + c) =
a · b + a · c.
The simplest examples of a semifield are just positive real or rational numbers with
respect to the usual addition and multiplication. Another class of examples are given
by the sets of real, rational and integral numbers with the usual addition used for
multiplication and usual maximum used for addition. The corresponding semifields
are called real, rational and integral tropical semifields and are denoted by Rt , Qt
and Zt respectively. Every algebraic expression which does not contain substraction
(such expressions are called substraction free) makes sense in a semifield. Consider
for example the polynomial a 3 + 3ab2 + 2. In tropical semifields this expression
can be rewritten using usual operation as max(3a, 2a + b, 0). We say that the second
expression is the tropical analogue of the first one.
Observe that tropical semifields can be considered as certain limits of nontropical
ones. Namely fix a positive real constant and consider the set of real numbers pro-
vided with the addition: (a, b) → log(ea/ + eb/ ) and the multiplication given by
the ordinary addition. This semifield is obviously isomorphic to the semifield of pos-
itive numbers by the isomorphism a → ea/ . In the limit as → 0 the isomorphism
is no longer defined, but the limit of the operations does exist and coincides with the
operations in the tropical semifield.
In particular the tropical analogue of a substraction free rational function of n
variables f (a1 , . . . , an ) is given by
f(a1 , . . . , an ) = lim log f (ea1 / , . . . , ean / ).
→0
One can easily see by comparison of the Sections 3 and 4 that the formulae for the
spaces of laminations are tropical analogues of the respective formulae for Teichmüller
spaces. For example the space of laminations Tx has he same set of coordinate systems
as the Teichmüller space T x , while the transition functions between the coordinate
system (3.1) for the former space are given by tropical analogues of the respective
functions (4.2) for the latter one. Similarly the lamination spaces Ta and T0a are
tropical analogues of the spaces T a and T0a respectively.
In particular one can see that Teichmüller spaces can be canonically compactified
by adding the quotients of the respective real lamination spaces by action of the
multiplicative group of positive real numbers. One can show that it is just a version
of the Thurston compactification [25] for ciliated surfaces.
6 Poisson and degenerate symplectic structures

We are going to give a definition (by explicit formulae) of a Poisson bracket PWP on the
spaces T x (S) and of a degenerate symplectic structure ωWP on T a (S). For surfaces
S without cilia such structures are well known under the name of Weil–Petersson ones
and can be defined as follows.
The spaces of homomorphisms of the fundamental group of a surface S into any
reductive group G considered up to conjugation is known to have a canonical Poisson
structure (see [9], [10] and references therein). Its symplectic leaves are numerated
by conjugacy classes of elements corresponding to loops surrounding holes. The
Teichmüller space T x (S) can be mapped to this space (for G = PSL(2, R)) and if the
surface S has no cilia the corresponding map is a local diffeomorphism in a vicinity
of a generic point. This map induces the Poisson structure PWP .
Another way of defining the same Poisson structure was given by William Goldman
in [15]. For any closed loop one can associate the absolute value of the trace of the
element of PSL(2, R) corresponding to it. Goldman gave an explicit expression for
the Poisson bracket between function of this type as a linear combination of functions
of the same type.
The image of T a (S) in T x (S) under the map p forgetting the horocycles corre-
sponds to the representations of π1 (S) in PSL(2, R) sending all loops surrounding
holes to parabolic elements. It implies that the Poisson structure restricted to the
image is non-degenerate. The inverse image under the map p of the corresponding
symplectic structure defines a degenerate symplectic structure ωWP on T a (S).
The Poisson structure on T x (S) for a ciliated surface S is given by the formula:
∂ ∂
PWP = εαβ x α α ∧ x β β . (6.1)
∂x ∂x
αβ∈E()
The degenerate symplectic structure ωWP on T a (S) for a ciliated surface S is given
by the formula:
daα daβ
ωWP = εαβ ∧ . (6.2)
aα aβ
αβ∈E()
The main property of these Poisson and symplectic structures is that they are
independent of the particular triangulation. This property can be checked explicitly
by substitution of the transformation laws (4.2) (resp. (4.4)) into the expression for
PWP (resp. ωWP ), taking into account the transformation law (2.1) for the matrix εαβ .
Another useful property is the compatibility of the bracket with unramified covering
of surfaces. If S̃ → S is an unramified covering of degree N then there are obvious
maps i x : T x (S) → T x (S̃) and i a : T a (S) → T a (S̃). Then one can easily see
that (i a )∗ ω̃WP = N ωWP , where ω̃WP is the canonical degenerate symplectic form
on T x (S̃). The map i x is not a Poisson map, however it maps symplectic leaves to
symplectic leaves. If i x : L → L̃, where L and L̃ are the symplectic leaves of T x (S)
and T x (S̃) respectively, then (i x )∗ (P̃WP )−1 |L̃ = N(PWP )−1 |L .
One can check that if the surface S has no cilia the formula gives the known
Poisson and symplectic structures. The simplest way to do it is to compute explicitly
the bracket between traces of matrices corresponding to closed loops using both our
definition and Goldman’s one. Due to the compatibility with coverings it is sufficient
to check the coincidence of the brackets for two loops having one intersection point,
where Goldman’s expression is especially simple. We do not give the computation
here since it is straightforward and technical.
The kernel of the symplectic form on the space T0a is spanned by the vector fields
corresponding to the action rρ , where ρ runs through the boundary components with
even number of cilia. The kernel of the Poisson structure on the space T x is spanned by
the differentials of the functions r ρ . In particular if the number of cilia is odd on every
boundary component the symplectic and the Poisson structures are nondegenerate.
The kernel of the symplectic form on the space T a is generated by the orbits of
resizing of all horocycle at a given boundary component by the same amount.
The lamination spaces Tx and Ta also have a canonical Poisson structure PWP and
the degenerate symplectic structure wWP , respectively. They are given by
∂ ∂
PWP = εαβ α ∧ β ,
∂x ∂x
αβ∈E()

wWP = εαβ d aα ∧ d aβ .
αβ∈E()
All the properties of these structures are similar to the respective properties of the
Poisson and symplectic structures on their Teichmüller counterparts.
7 Canonical pairing
Certain curves or collections of curves on a surface define functions on Teichmüller
and lamination spaces. Namely:
1. A closed curve γ on S defines a function (γ ) on T a (S) as well as on T x (S) by the
length of the geodesic w.r.t. the curvature −1 metric in the homotopy class defined
by the curve.
2. If a curve γ on S connects two points on the boundary it defines a function (γ )
on T a (S) in the following way. Move the ends of the curve which do not belong to
holes, in the counterclockwise direction until they hit the cilia. Then take a geodesic
in the same homotopy class and take the exponent of the signed half of the length of
the segment between the corresponding horocycles like in Figure 12.
3. If a collection of curves m = {γi } with rational weights {wi } has the property that
for every segment of the boundary between cilia and for every hole the total weight
of the curves hitting it vanishes, then it defines a function on T x (S) in the following
way. Move first the ends of curves in the counterclockwise direction to hit the cilia.
Lift then the collection of curves to H and choose horocycles at their ends. Then
take the weighted
sum of signed lengths of geodesic segments between the horocycles
(m) = wi (γi ). It is easy to see that this sum does not depend on the choice of
the horocycles due to the vanishing property.
These functions can be considered as functions of two arguments: the first is a
curve or a weighted collection of curves and the second is a point of the appropriate
Teichmüller or lamination space. Our aim now is to extend the first argument to be not
just a single curve, but a point of an appropriate space of laminations. The resulting
functions of two arguments are called pairings.
There are two version or pairings between Teichmüller and lamination spaces –
additive and multiplicative. The main feature of the additive one is that it can be
extended to real laminations as a continuous function. The multiplicative one is defined
for integral laminations only. For every lamination this pairing defines a function on
the Teichmüller space which is a Laurent polynomial in the coordinates with positive
integral coefficients.
7.1 Additive canonical pairing

The additive canonical pairing is a function of two arguments. One argument is a
rational (or, later on, real) lamination, which can be either from T0a or Tx , and the other
is a point of the opposite type Teichmüller space, T x or T0a , respectively.
: Tx × T0a → R,
: T x × T0a → R.
Abusing notation, we shall denote all of them by a single letter .
We will also define the intersection pairing:
I : Tx × T0a → R.
Sometimes it is useful to consider the functions and I as functions of one variable
taking another one as a parameter. In this case we shall put this parameter as a lower
index. For example mx (ma ) := (mx , ma ).
The additive and the intersection pairings are defined as follows.
Definition.
(1) Let γ x ∈ Tx be a single closed curve and ma ∈ T0a . Then (γ x , ma ) is its length
(γ x ) (Rule 1).
(2) Let mx ∈ T x and let γ a ∈ Ta be a single closed curve. Then (mx , γ a ) is equal
to ± a half of its length (γ a ) (Rule 1). The sign is + unless γ a surrounds a
negatively oriented hole.
(3) Let γ x ∈ Tx be a curve connecting two points of the boundary of S and ma ∈ T a .
Then (γ x , ma ) is the signed length (γ x ) given by Rule 2.
(4) Let γ x ∈ Tx and γ a ∈ Ta be two curves. Move the ends of γ a (if any and but
those belonging to holes) counterclockwise to hit cilia. Then I(γ x , γ a ) is a half
of the minimal number of intersection points between γ x and γ a .
(5) Let mx ∈ T x and let ma ∈ T0a be a collection of curves. Then the function (l a )
defined by this Rule 3 is equal to (mx , ma ). (The vanishing property follows
from the definition of T0a .)
(6) Let u, v be positive rational numbers. Let mx1 , mx2 ∈ Tx be two collections of
nonintersecting curves such that no curve from mx1 intersects mx2 , and ma ∈ T a .
Then (umx1 + v mx2 , ma ) = u (mx1 , ma ) + v (mx2 , ma ). If mx ∈ T x and

ma1 , ma2 ∈ T0a be nonintersecting collections of curves such that no curve from
ma1 intersects ma2 then (mx , uma1 + v ma2 ) = u (mx , ma1 ) + v (mx , ma2 ).
Properties of the additive canonical pairing. 1. The additive canonical pairings

as well as the intersection pairing I are continuous functions.
2. (p(ma ), ma ) = (p(ma ), ma ), where ma ∈ T0a and ma ∈ T0a .
3. If a point mx of the space Tx has positive coordinates x1 , . . . , xn and a point
m of the Teichmüller space T0a has coordinates a1 , . . . , an , then (mx , ma ) =

a
α
α x log aα .
4. Let mx ∈ T x be a point of a Teichmüller space T x with coordinates x1 , . . . , xn .
Let C be a positive real number. Denote by (mx )C ∈ T x a point with coordinates
(x1 )C , . . . , (xn )C and by mx ∈ Tx a lamination with coordinates x1 = log x 1 , . . . ,
xn = log x n . Then for any lamination ma ∈ Ta we have
lim ((mx )C , ma )/C = I(mx , ma ).

C→∞
5. Analogously let ma ∈ T0a be a point of Teichmüller space with coordinates

a1 , . . . , an . Let C be a positive real number. Denote by (ma )C ∈ T0a the point in
Teichmüller space with coordinates (a1 )C , . . . , (an )C and by ma ∈ T0a the lamination
with coordinates a1 = log a1 , . . . , an = log an . Then for any lamination mx ∈ Tx we
have
lim (mx , (ma )C )/C = (mx , ma ).
C→∞
For closed surfaces there exists a canonical pairing : T x × T0a → R defined by

Francis Bonahon [2], such that the pairings defined here are its limits. Unfortunately
we do not know any explicit construction of this pairing.
7.2 Multiplicative canonical pairing

The multiplicative canonical pairing is a function of two arguments. One argument is
an integral lamination which can be either in T0a (Z) or Tx (Z), and the other is a point
of the opposite type Teichmüller space, T x or T0a , respectively.
I : Tx (Z) × T0a → R>0 ,
I : T x × T0a (Z) → R>0 .

Abusing notations, we shall denote all of them by a single letter I. Sometimes it
is convenient to consider the function I as a function of one variable taking another
one as a parameter. In this case we shall put this parameter as the lower index. For
example Imx (ma ) := I(mx , ma ).
The multiplicative pairing is defined by the following properties:
Definition.
(1) Let γ x ∈ Tx be a single closed curve with weight k and ma ∈ T a . Then
I(γ x , ma ) is the absolute value of the trace of the monodromy of (γ x )k .
(2) Let mx ∈ Tx and let γ a ∈ Ta be a single closed curve which is not retractable to
a hole. Then I(γ x , ma ) is the absolute value of the trace of the monodromy of
(γ x )k .
(3) Let mx ∈ T x and let γ a ∈ Ta be a single closed curve which is retractable to
a hole and the hole is oriented positively (resp. negatively). Then (γ x , ma )
is equal to the absolute value of the largest (resp. smallest) eigenvalue of the
monodromy of (γ x )k .
(4) Let γ x ∈ Tx be a curve with weight k connecting two points of the boundary of
S and ma ∈ T a . Then I(γ x , ma ) = exp (γ x , ma ).
(5) Let mx ∈ T x and let ma ∈ T0a be a collection of open curves with integral
weights. Then I(mx , ma ) = exp (mx , ma ).
(6) Let u, v be positive integral numbers. Let mx1 , mx2 ∈ Tx be two collections of non-
intersecting curves such that no curve from mx1 intersects or coincides with a curve
from mx2 , and ma ∈ T a . Then I(umx1 + v mx2 , ma ) = (I(mx1 , ma ))u (I(mx2 , ma ))v .
If mx ∈ T x and ma1 , ma2 ∈ T0a are nonintersecting collections of curves such
that no curve from mx1 intersects or coincides with a curve from mx2 , then
I(mx , uma1 + v ma2 ) = (I(mx , ma1 ))u (I(mx , ma2 ))v .
Properties of the multiplicative canonical pairing. 1. Let a1 , . . . , an be coordi-

nates of a lamination ma ∈ T0a (S, Z). Then the function Ima on T x is a Laurent poly-
nomial with positive integral coefficients of (x 1 )1/2 , . . . , (x n )1/2 , where x 1 , . . . , x n
arethe coordinates on T . Its highest term is equal to α (x α )aα and the lowest one
x
is α (x α )−aα . Moreover Ima is a Laurent polynomial of x 1 , . . . , x n if and only if the

lamination ma is even.
2. Let x1 , . . . , xn be coordinates of a lamination mx ∈ Tx (S, Z). Then the function
Ima on T x is a Laurent polynomial with positive integral coefficients of the coordinates
a 1 , . . . , a n on T0a .
3. Let mx ∈ Tx (S, Z) be an integral X-lamination and let ma ∈ T0a be a point of
the Teichmüller A-space. Then limC→∞ log(I(C mx , ma ))/C = (mx , ma ). Anal-
ogously let mx ∈ T x be a point of the Teichmüller X-space and let ma ∈ T0a be an
A-lamination. Then limC→∞ log(I(Cmx , ma ))/C = (mx , ma ).
4. Let ma ∈ T0a be an integral bounded lamination. Then the function Ima on Tx is
the tropical analogue of the function Ima on T x in any coordinate system. Analogously
let mx ∈ Tx be an integral unbounded lamination. Then the function Imx on T0a is the
tropical analogue of the function Imx on T a in any coordinate system.
5. The coordinate functions on the space T a as well as T0a are particular cases
of the pairing, namely the value of the coordinate function aα at the point ma of the
B C
A
B C
Figure 15. Rearranging the intersection points.
Teichmüller space T a is I(mx , ma ), where mx is the edge α of the triangulation with

weight 1.
Conjecturally the converses to Properties 1 and 2 are also true. Namely, the only
functions on T x (resp. T0a ) given by a Laurent polynomial with positive integral
coefficients in every coordinate chart and undecomposable into a sum of two such
polynomials are given by the multiplicative canonical pairing.
7.3 Proofs of main properties

Here we are going to give main ideas of the proofs of the most important properties
of pairings: the continuity of the additive pairing (Property 1) and Properties 1 and 2
of the multiplicative pairing. Other properties are either obvious or follow from the
proof of these three.
Continuity. Let us first prove the continuity of the pairing between Ta and T x . To do
this it suffices to prove it for laminations without curves with negative weights. Indeed,
if we add such a curve to a lamination the length obviously changes continuously. We
are going to show that the length of integral lamination is a convex function of its
coordinates, i.e. that
(mx , ma1 ) + (mx , ma2 ) ≤ (mx , ma1+2 ) (7.1)
where by ma1+2 we have denoted a lamination with coordinates being sums of the
respective coordinates of ma1 and ma2 . Taking into account the homogeneity of ,
one sees that Inequality (7.1) holds for all rational laminations and therefore can be
extended by continuity to all real laminations.
Let us prove Inequality (7.1). Draw both laminations ma1 and ma2 on the surface
and deform them to be geodesic. These laminations in general intersect each other
in a finite number of points. Construct now a new collection of curves out of the
union of ma1 and ma2 in the following way. For every intersection point consider the
triangle ABC of the triangulation it belongs to. Without loss of generality we may
assume that the segments of the curves containing the intersection points connect the
sides AB with BC and AC with BC, respectively. Then remove the neighbourhood
of the intersection point and replace it by a union of two nonintersecting arcs in such
a way that the segments of the resulting curves still connect AB with BC and AC
with BC (Figure 15). The resulting collection of curves considered as a lamination
is just ma1+2 . The total length of the constructed curves can be made as close to
(mx , ma1 ) + (mx , ma2 ) as one desires. However it is not geodesic any longer. When
we deform this collection to the geodesic one its length only decreases, thus giving
the desired inequality (refineq).
Laurent property and positivity. Now we are going to show that the functions Ima
and Imx are given by Laurent polynomials with positive integral coefficients in any
coordinate charts. We shall show it for laminations representable by a single curve
and then extend the result to arbitrary laminations by multiplicativity.
Consider first a single closed curve with weight k as an element of the space Ta .
If M is the monodromy operator around this curve, then by definition Ima = tr(M k ).
According to the construction of the Fuchsian group from Section 3.1 the operator M
is given by the expression
M = B(x i1 )I ±1 B(x i2 ) . . . I ±1 B(x ip )I ±1 ,
where i1 , . . . , ip is the sequence of edges of the triangulation the path intersects and
the signs for I ±1 are chosen according to whether the 0path
turns left or right passing
through a triangle. Multiplying every I ±1 by J = −1 1 from the left and every B
0
by J from the right we do not change the product since J 2 = 1 in PSL(2, R) and
transform the expression for M into M = H (x i1 )E ± H (x i2 ) . . . H (x ip )E ± , where

E + = 01 11 , E − = 11 01 and H (x) = x 0 x −1/2
1/2 0 . Since all coefficients of these
matrices are Laurent monomials with positive coefficients, the trace of any power of
M is a Laurent polynomial with positive coefficients.
Now consider the case of the space T a (S). The reconstruction rule implies that
the traces of the monodromy matrices are Laurent polynomials of the coordinates in
integral powers of the coordinates {aα }. Positivity follows from the existence of the
monomial map p : T a (S) → T x (S).
We leave the proof of the Laurent property and positivity for non closed curves as
an exercise.
Appendix A. Combinatorial description of D(S)

Denote by ||(S) the set of combinatorial types of triangulations of S. For each
element of ||(S) fix a marking, i.e., a numbering of its edges. Denote by (S) the
set of isotopy classes of marked triangulations of S. The presence of the marking
changes the set of triangulations since some of them may have nontrivial symmetry
group. Introducing the marking is a tool to remove this symmetry. (Here and below
the vertical lines | · | indicate the diffeomorphism class.)
The mapping class group D(S) obviously acts freely on the space of marked
triangulations with the space of combinatorial types of triangulations as a quotient,
(S)/D(S) = ||(S).
Recall that a group can be thought of as a category with one object and invert-
ible morphisms. Similarly, a groupoid is a category where all morphisms are iso-
morphisms, and any two objects are isomorphic. Since the automorphism groups of
different objects of a groupoid are isomorphic, we can associate a group to a groupoid,
well defined up to an isomorphism. We are going to construct a groupoid providing the
mapping class group and admitting a simpler description by generators and relations
than the mapping class group itself.
Definition. Let ||(S) be the set of objects. For any two triangulations ||, |1 | ∈
|(S)| let a morphism from || to |1 | be a pair of triangulations of S of types || and
|1 | modulo the diagonal mapping class group action; we denote this morphism by
|, 1 |. For any three triangulations , 1 , 2 , the composition of |, 1 | and |1 , 2 |
is |, 2 |. The described category is called the modular groupoid.
One can easily verify that (1) composition of morphisms is well defined; (2) the
class of the pair of identical triangulations |, | is the identity morphism and the
inverse of the morphism |, 1 | is |1 , |; (3) the group of automorphisms of an
object is the mapping class group D(S).
To give a description of the modular groupoid by generators and relations, we need
to introduce distinguished sets of morphisms called flips and symmetries. Recall that
a morphism |, α | is a flip if the triangulation α is obtained from by removing an
edge α and replacing it by another diagonal of the resulting quadrilateral. We use the
notation α to emphasize the relation of this triangulation to the triangulation . Note
that for a given triangulation , several marked embedded graphs may be denoted by
α because no marking of α is indicated.
To each symmetry σ of a triangulation we assign an automorphism |, σ |.
There is no canonical bijection between the edges of different triangulations even if
a morphism between them is given. However, if two triangulations are related by a flip,
we can introduce such a bijection since all edges of one triangulation coincide with the
edges of the other. We exploit this identification and denote the corresponding edges
of different graphs by the same letter if it is clear which sequence of flips relating these
graphs is considered. To avoid confusion we note that this identification has nothing
to do with the marking.
In this notation, the triangulation α1 ...αn is the graph obtained as a result of con-
secutive flips αn , . . . , α1 of edges of a given graph .
There are three kinds of relations between flips, which are satisfied for any choice
of marking of the triangulations entering the relations.
H
B HH

A Q B
Q
A Q B
A QQBB @
I H
HH B H
H H
Q
A Q A B
A Q A B
A QQ A BB
AU
HH H H
H H
A A

A - A

A A
Figure 16. Pentagon relation.
Proposition. A square of a flip is a symmetry: if |α , | is a flip in an edge α, then

|, α | is also a flip and1
|, α ||α , | = 1. (R.2)
Flips in disjoint edges commute: if α and β are two edges , then
|αβ , α ||α , | = |αβ , β ||β , |. (R.4)
Five consecutive flips in edges α and β having one common vertex is the identity:
for such α and β, the triangulations αβ and βα are related by a flip and
|, α ||α , βα ||βα , αβ ||αβ , β ||β , | = 1. (R.5)
The proofs of Relations R.2 and R.4 are obvious. Relation R.5 is obvious from the
triangulation of a pentagon shown in Figure 16.
Replacing the mapping class group by the modular groupoid, we can express the
latter through generators and relations in a simpler way.
Note that a symmetry can be represented as a ratio of two flips in a given edge
and the modular groupoid is therefore generated by the flips only. We do not describe
relations between flips and graph symmetries in detail because they are quite obvious.
In fact, the symmetry groups of and α act transitively on the set of flips |, α |,
and this action can be considered as relations between flips and symmetries.
The proposition can be proved using direct combinatorial methods of the simplicial
topology. However, we give the main idea of another proof, which is more specific
for dimension 2.
Proof of the proposition. To any connected simplicial complex S we can associate
a groupoid by taking a point in each top-dimensional simplex objects and the ho-
1 The notation R.n indicates the number n of graphs entering this relation.
motopy classes of oriented paths connecting the chosen points as morphisms. The
corresponding group is the fundamental group of the topological space given by the
complex.
To any codimension one simplex we can associate two classes of paths (differing
by orientation and having the identity morphism as their product) connecting adjacent
top-dimensional simplices. It is natural to call them flips. We can associate a relation
between flips to any codimension two simplex. It is obvious that this set of flips
generates the groupoid and that the only relations between the flips are given by
codimension two simplices.
The same is true for an orbifold simplicial complex, where we replace simplices
by quotients of simplices by finite groups. In this case, we must choose one generic
point per each top-dimensional simplex as an object and orbifold homotopy classes of
paths as morphisms. The corresponding group is the orbifold fundamental group of
the orbifold given by the complex. The groupoid is now generated by flips and groups
of top dimensional simplices and still the only nontrivial relations are those given by
codimension two simplices.
Consider now the orbifold simplicial decomposition of moduli space of complex
structures on S by Strebel, Hubbard, Mazur and Penner [21]. The orbifold fundamental
group of the moduli space M is just the mapping class group D(S). Recall that Strebel
orbisimplices are enumerated by ribbon graphs corresponding to S and the dimension
of a simplex is equal to the number of its edges. One can easily see that the groupoid
of the Strebel complex coincides with the modular groupoid. Moreover, the flips of
the former correspond to the flips of the latter. The relations between flips are given
by codimension two cells, which correspond either to graphs with two four-valent
vertices (which produces Relation R.4) or to graphs with one five-valent vertex (which
produces Relation R.5). Relation R.2 holds true for any simplicial complex.
Appendix B. Markov numbers
The set of Markov numbers is a remarkable collection of integers that appeared first
in number theory but is related to lengths of closed geodesics on a holed torus with
hyperbolic structures. In this appendix we shall give a simple description of these
numbers as well as a proof of a few of their properties as an illustration of the technique
developed here.
Consider a torus with one hole T . The space of homotopy classes of simple (i.e.,
without self intersections) unoriented closed paths on it can be parameterised by points
of QP 1 . Indeed, once we have chosen an orientation of the path, we can consider it as
an element of the first homology group of T with compact support. It is also obvious
that any simple (indivisible) class is represented by a unique simple oriented closed
path. Since the first homology group is Z2 , it just gives the desired parameterisation.
Introduce the equiharmonic complex structure on T , i.e. the structure which has
maximal symmetry group Z/3Z. For any closed path γ on T without self-intersections
the numbers Xγ = 23 cosh l(γ ), where l(γ ) are the lengths of the corresponding
geodesic, are called Markov numbers.
The main properties of Markov numbers are the following:
1. Markov numbers are positive integers.
2. Markov numbers include Fibonacci numbers with even indices, 2, 5, 13, 34, 89,
233, . . . .
Call a Markov triple a triple of Markov numbers (X, Y, Z) corresponding to three
geodesics having pairwise one intersection point.
3. Elements of a Markov triple satisfy the Markov equation:
X2 + Y2 + Z2 = 3XYZ. (1)
4. Any integer solution of this equation is a Markov triple.
5. For any Markov triple (X, Y, Z) the triples (Y, Z, X) and (Z, Y − 3XZ, X) are
also Markov triples. Any Markov triple can be obtained from the triple (1, 1, 1) by a
sequence of such transformations.
Since homotopy classes of closed non-selfintersecting curves can be parameterised
by QP 1 , one can choose affine coordinates on QP 1 in such a way that the curves with
coordinates 0, 1 and ∞ have Markov numbers 1. Denote by M(u) the Markov number
corresponding to the curve
p
with 3 p
u ∈ Q.
the coordinate
6. The function ψ q = q arcosh 2 M q , where gcd(p, q) = 1, is extensible
1
to a continuous convex function on

R. 1
x

7. M(x) = M(1 − x) = M x1 = M 1−x = M x−1 = M x−1x

8. For any closed geodesics γ on S there exists a unique geodesics γ going from
the puncture to the puncture which doesn’t intersect γ . Let l(γ ) be the length of the
piece of γ between the intersection points with the horocycle surrounding the region
of area 3. Then el(γ ) = M(γ ).
9. (Markov conjecture). The famous unproven Markov conjecture says that two
Markov numbers M(x) and M(y) are different unless x and y are related by transfor-
mations from Property 7.
Taking into account that the segment [0, 1] is the fundamental domain of the action
of transformations from Property 7, one can reformulate the Markov conjecture as
follows: if M(x) = M(y) and x, y ∈ [0, 1] then x = y.
Proofs of the properties (unfortunately, without the last one and Property 4). There
is only one combinatorial triangulation corresponding to the holed torus. It has one
vertex, three edges and two triangles. This triangulation has obvious Z/3Z symmetry
group cyclically permuting the edges. Let x, y, z be the corresponding coordinates on
the Teichmüller space T x (S).
A closed curve on S can be considered as a bounded lamination if we assign
to it the weight 1. The standard coordinates of such laminations are given by three
nonnegative integers n1 , n2 , n3 . These three numbers have no common factor, because
otherwise the weight of the curve would be greater than 1. On the other hand one
of the numbers should be a sum of the two others since otherwise there would be
a component surrounding the hole. The relation between this parameterisation by
n1 , n2 , n3 and the parameterisation by QP 1 described above is given by

−n2
if n3 = n1 + n2 ,
x = nn21 (2)
n1 if n1 = n2 + n3 or n2 = n3 + n1 .
Denote by Z, X and Y one thirds of the traces of the elements of the Fuchsian
group corresponding to the curves with coordinates (1, 1, 0), (0, 1, 1) and (1, 0, 1)
respectively. They can be easily computed using the explicit formulae for the Fuchsian
group:
1 1/2 1/2
Z= (x y + x 1/2 y −1/2 + x −1/2 y −1/2 ),
3
1
X = (y 1/2 z1/2 + y 1/2 z−1/2 + y −1/2 z−1/2 ), (3)
3
1
Y = (z1/2 x 1/2 + z1/2 x −1/2 + z−1/2 x −1/2 ).
3
Using these expressions we can verify the equality
1
X2 + Y2 + Z2 − 3XYZ = − (xyz − 2 + (xyz)−1 ). (4)
9
The symmetry of the graph obviously cyclically permutes the coordinates and
therefore the numbers Z, X, Y. A flip of an edge acts by Rule (3.3) and it results in
the mapping
(Z, X, Y) → (X, 3YZ − X, Z). (5)

If all three coordinates x, y, z are ones, the corresponding complex surface is just
the equiharmonic punctured torus.
Properties 1, 3, 5, 6 immediately follows from this picture. One can easily check
that M(n) for n ∈ N are just the Fibonacci numbers which gives Property 2. The
property 7 is an immediate consequence of the convexity property of the lamination
length function. The property 4 was proved by Markov himself.
Property 8 stands a little apart from the others since it is related to the spaces T a (S)
and Tx (S) rather than Ta (S) and T x (S) respectively. Consider a coordinate system
U, V, W on T a (S). Let A = Aρ be the area of the region inside the only horocycle
ρ. It easily follows from the expression for the area
(U2 + V2 + W2 ) = UVWA. (6)
The cyclic symmetry of the triangulation acts by cyclic permutation of U, V, W. A
flip of an edge acts by

U2 + W 2
(U, V, W) → W, ,U . (7)
V
On the other hand this transformation law can be rewritten taking into account
Equation (6):
(U, V, W) → (W, UWA − V, U). (8)
This rule coincides with (5) for A = 3.
Now consider the decorated surface with U = V = W = 1. This is the surface
with the area inside the horocycle A = 3. Applying modular transformations we get
obviously the Markov triples, which proves the property 8.
There exists a canonical decomposition (called Farey tessellation) of the upper half
plane H into ideal triangles with vertices in all rational points of its ideal boundary.
The dual graph to this tessellation is the universal three-valent tree. The faces of
this tree are therefore in one-to-one correspondence with rational numbers. On the
pictures below we have drawn a fragment of this tree with corresponding Markov
numbers written on the faces.
1@
@
HH
2 HH
1

HH
HH
Z 5 Z
Z Z
Z Z
29 Z 13 Z
@ @ @ @
169 @ 433 @ 194 @ 34 @
A A A A A A A A
985 A 14701A 37666A 6466A 2897A 7561A 1325A 89 A
L L L L L L L L L L L L L L L L
L L L L L L L L L L L L L L L L
5741 7453378 3276569 8399329 43261 4400489 51641 9077
499393 1278818 48928105 96557 1686049 294685 135137 233
As a concluding remark of this section note that, as it was observed by A. Bondal,

Markov triples are dimensions of elements of distinguished sets of sheaves on CP 2 .
The relations between these two ways of obtaining Markov numbers are completely
unclear and very exciting.
References
Springer-Verlag, Berlin 1980. 659
[2] F. Bonahon, Teichmüller space and geodesic currents. Invent. Math. 92 (1988), 139–162.
673
[3] S. Choi and W. Goldman, The deformation spaces of convex RP2 structures on 2-orbifolds.
The deformation spaces of convex RP2 -structures on 2-orbifolds. Amer. J. Math. 127 (5)
(2005), 1019–1102. 648
[4] V. V. Fock, Dual Teichmüller spaces. Preprint, 1998; arXiv:dg-ga/9702018. 648
[5] V. V. Fock, L. O. Chekhov, Quantum Teichmüller spaces. Teoret. Mat. Fiz. 120 (3) (1999),
511–528; English transl. Theoret. and Math. Phys. 120 (3) (1999), 1245–1259. 648
[6] V. V. Fock and A. B. Goncharov, Moduli spaces of local systems and higher Teichmüller
theory. Publ. Math. Inst. Hautes Études Sci. 103 (2006), 1–211. 648
[7] V.V. Fock, A. B. Goncharov, Cluster ensembles, quantization and the dilogarithm. Preprint,
2004; arXiv:math/0311245. 648, 649
[8] V.V. Fock, A. B. Goncharov, Moduli spaces of convex real projective structures on surfaces.
Preprint, 2006; arXiv:math.DG/0405348. 648
[9] V. V. Fock and A. A. Rosly, Poisson structures on moduli of flat connections on Riemann
Soc. Transl. Ser. 2 191, Amer. Math. Soc., Providence, RI, 1999, 67–86. 669
[10] V. V. Fock and A. A. Rosly, Flat connections and polyubles. Theoret. and Math. Phys. 95
(1993), 526–535. 669
[11] S. Fomin and A. Zelevinsky, Cluster algebras. J. Amer. Math. Soc. 15 (2) (2002), 497–529.
648
[12] S. Fomin and A. Zelevinsky, Cluster algebras: Notes for the CDM-03 conference. In
Current developments in mathematics, 2003, International Press, Somerville, MA, 2003,
1–34. 649
[13] M. Gekhtman, M. Shapiro, A. Vainshtein, Cluster algebras and Weil-Petersson forms.
Duke Math. J. 127 (2) (2005), 291–311.
[14] W. M. Goldman, Convex real projective structures on surfaces. J. Differential Geom. 31
(1990), 795–845. 648
[15] W. M. Goldman, The symplectic nature of fundamental groups of surfaces. Adv. Math. 54
(2) (1984), 200–225. 670
[16] N. J. Hitchin, Lie groups and Teichmüller space. Topology 31 (3) (1992), 449–473. 648
[17] R. M. Kashaev, Quantum dilogarithm and Quantisation of Teichmüller space. Lett. Math.
Phys. 43 (2) (1998), 105–115. 648
[18] M. L. Kontsevich, Intersection theory on the moduli space of curves. Funct. Anal. Appl.
25 (2) (1991), 123–129. 664
[19] S. Krushkal, B. Apanasov and N. Gusevskii, Kleinian groups and uniformization in exam-
ples and problems. Transl. Math. Monogr. 62, Amer. Math. Soc., Providence, R.I., 1986.
659
[20] A. Papadopoulos and R. C. Penner, La forme symplectique de Weil-Petersson et le bord de
Thurston de l’espace de Teichmüller. C. R. Acad. Sci. Paris Sér. I Math. 312 (11) (1991),
871–874. 656
[21] R. C. Penner, The decorated Teichmüller space of Riemann surfaces. Comm. Math. Phys.
113 (1988), 299–339. 648, 664, 679
[22] R. C. Penner, Weil-Peterson volumes in moduli spaces. Preprint, Inst. Mittag-Leffler,

1989.
1984.
[24] J. Teschner, On the relation between quantum Liouville theory and the quantized Teich-
müller spaces. Internat. J. Modern Phys. A 19 (2004), 459–477. 648
[25] W. Thurston, The geometry and topology of three-manifolds. Princeton University Notes,
http://www.msri.org/publications/books/gt3m. 647, 648, 652, 669
Chapter 16
An analog of a modular functor from quantized

Teichmüller theory
Jörg Teschner
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
1.2 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
1.4 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
Part I. Coordinates for the Teichmüller spaces . . . . . . . . . . . . . . . . . . 693
2 The Penner coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
2.1 Triangulations and fat graphs . . . . . . . . . . . . . . . . . . . . . . 693
2.2 Penner coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 695
2.3 Fock coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695
3 The Ptolemy groupoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697
3.1 Groupoids vs. complexes . . . . . . . . . . . . . . . . . . . . . . . . 698
3.2 Change of the triangulation . . . . . . . . . . . . . . . . . . . . . . . 698
3.3 The representation of the mapping class group . . . . . . . . . . . . . 700
4 Teichmüller space as the phase space of a constrained system . . . . . . . 700
4.1 Kashaev’s coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 700
4.1.1 The constraints . . . . . . . . . . . . . . . . . . . . . . . . . 701
4.1.2 Change of fat graph . . . . . . . . . . . . . . . . . . . . . . . 701
4.2 The structure of the Kashaev space Wϕ . . . . . . . . . . . . . . . . . 702
4.2.1 Fock variables vs. Kashaev’s variables . . . . . . . . . . . . . 702
4.2.2 Splitting of Wϕ . . . . . . . . . . . . . . . . . . . . . . . . . 702
5 The Fenchel–Nielsen coordinates . . . . . . . . . . . . . . . . . . . . . . 703
5.1 Definition of the Fenchel–Nielsen coordinates . . . . . . . . . . . . . 704
5.2 Symplectic structure . . . . . . . . . . . . . . . . . . . . . . . . . . 705
5.3 Geodesic lengths from the Penner coordinates . . . . . . . . . . . . . 705
6 Coordinates for surfaces with holes of finite size . . . . . . . . . . . . . . 707
6.1 Useful fat graphs on surfaces with holes of finite size . . . . . . . . . 707
6.2 Kashaev type coordinates . . . . . . . . . . . . . . . . . . . . . . . . 708
Part II. Quantization of the Teichmüller spaces . . . . . . . . . . . . . . . . . . 709
7 Quantization of the Teichmüller spaces . . . . . . . . . . . . . . . . . . . 709
686 Jörg Teschner
7.1 Canonical quantization . . . . . . . . . . . . . . . . . . . . . . . . . 709

7.2 Quantization of the Kashaev space Wϕ . . . . . . . . . . . . . . . . . 711
7.3 Reduction to the quantized Teichmüller spaces . . . . . . . . . . . . 712
8 Representations of the mapping class groups . . . . . . . . . . . . . . . . 714
8.1 Projective unitary representations of groupoids . . . . . . . . . . . . 714
8.2 Representations of symmetries of a groupoid . . . . . . . . . . . . . 715
8.3 The projective representation of the Ptolemy groupoid on K(ϕ) . . . 717
8.4 Reduction to the quantized Teichmüller spaces . . . . . . . . . . . . 717
Part III. A stable modular functor from the quantum Teichmüller spaces . . . . 720
9 The notion of a stable unitary modular functor . . . . . . . . . . . . . . . 721
9.1 Rigged Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . . . 721
9.2 Stable unitary modular functors . . . . . . . . . . . . . . . . . . . . 722
9.3 Representations of the modular groupoid versus modular functors . . 724
10 The modular groupoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726
10.1 Notations and conventions . . . . . . . . . . . . . . . . . . . . . . . 726
10.2 Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726
10.3 Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 728
11 From markings to fat graphs . . . . . . . . . . . . . . . . . . . . . . . . . 731
11.1 The complex Mad () . . . . . . . . . . . . . . . . . . . . . . . . . . 731
11.2 Separated variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 733
11.3 Quantized Teichmüller spaces for surfaces with holes . . . . . . . . . 735
12 Geodesic length operators . . . . . . . . . . . . . . . . . . . . . . . . . . 736
12.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 736
12.2 Construction of the length operators . . . . . . . . . . . . . . . . . . 737
12.3 Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 739
12.4 Relation with the Dehn twist generator . . . . . . . . . . . . . . . . . 740
13 Passage to the length representation . . . . . . . . . . . . . . . . . . . . . 741
13.1 The length representation . . . . . . . . . . . . . . . . . . . . . . . . 741
13.2 Construction of the operator D(σ ) . . . . . . . . . . . . . . . . . . . 742
14 Realization of M() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743
14.1 Two constructions for the generators . . . . . . . . . . . . . . . . . . 743
14.2 The main result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744
14.3 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 746
Appendix A. The special functions eb (x) and sb (x) . . . . . . . . . . . . . . . 747
Appendix B. Operator-theoretical background . . . . . . . . . . . . . . . . . . 748
Appendix C. On the proof of Theorem 2 . . . . . . . . . . . . . . . . . . . . . 749
Appendix D. Proof of Proposition 10 . . . . . . . . . . . . . . . . . . . . . . . 751
Appendix E. On the proof of Theorem 6 . . . . . . . . . . . . . . . . . . . . . 753
E.1 A technical preliminary . . . . . . . . . . . . . . . . . . . . . . . . . 753
E.2 Invariance of length operators . . . . . . . . . . . . . . . . . . . . . 755
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 757
Chapter 16. An analog of a modular functor from quantized Teichmüller theory 687
1 Introduction
The program of the quantization of the Teichmüller spaces T () of Riemann surfaces
which was started in [15], [17] and independently in [30]1 is motivated by certain
problems and conjectures from mathematical physics. One of the main aims of this
program is to construct a one-parameter family of maps

−→ HbT (), OTb (), MTb () , (1.1)
where
√
(i) b is a deformation parameter, related to the traditional h̄ via b = h̄,
(ii) is a two-dimensional topological surface possibly with boundary,
(iii) HbT () is a Hilbert-space (possibly infinite-dimensional),
(iv) OTb () is an algebra of bounded operators on HbT () and,
(v) MTb () is a unitary projective representation of the mapping class group of
on HbT ().

The data OTb (), MTb () are restricted by the requirement that a suitably defined
limit of OTb () for b → 0 should reproduce the commutative algebra of functions on
the Teichmüller space T (), whereas a natural limit b → 0 of the automorphisms
of OTb () which are induced by the representation MTb () should correspond to the
natural action of the mapping class group MC() on T ().
1.1 Motivation
Motivation for studying this problem comes from mathematical physics. A conjecture
of H. Verlinde [53] can be formulated very schematically as the statement that
T
Hb (), MTb () HcL (), MLc () , (1.2)
where
(i) HcL () has a definition in terms of the representation theory of the Virasoro
algebra with central charge c as the so-called space of conformal blocks associated
to ,
and
(ii) MLc () is an action of the mapping class group MC() on HcL () which is
canonically associated to the representation-theoretic definition of HcL ().
Part of the interest in the space HcL () from the side of mathematical physics is
due to the fact that the elements of HcL () represent the basic building blocks in
the so-called Liouville conformal field theory [49]. Deep connections between the
perturbative approach to quantum Liouville theory on the one hand and Teichmüller
1 See [18], [10] for recent generalizations.
688 Jörg Teschner
theory on the other hand have been exhibited by Takhtajan, Zograf and Teo see e.g.
[48] and references therein.
This conjecture may be seen as a non-compact analog of similar relations between
the quantization of moduli spaces of flat connections on Riemann surfaces on the one
hand, and rational conformal field theories on the other hand. For K being a compact
group, the geometric quantization of the moduli space MK () of flat K-connections
on a Riemann surface was performed in [27], [1]. Alternative approaches were based
on more explicit descriptions of the symplectic structure on MK () [19], [2], [3], [8],
[4].2 In either case one may schematically describe one of the main results of these
constructions as an assignment

−→ HkM (), MM k () , (1.3)
where
(i) HkM () is a finite-dimensional vector space,
(ii) MMk () is a projective representation of the mapping class group of on Hk ().
M
Part of the interest in these results was due to the close relations between the represen-
tation MMk () and the Reshetikhin–Turaev invariants of three manifolds [43]. Another
source of interest were the relations to rational conformal field theory, which were
predicted in [56], see [47] for a review of mathematical approaches to the problem
and further references. These relations may, again schematically, be summarized as
the existence of canonical isomorphisms
M
Hk (), MMk ()
ll5 iSSS
SSS
lllll SSS
l ll SSS (1.4)
lll SS)
W ul
o / H RT (), MRT () ,
Hk (), MW k () k k
where
(i) HkW () is the space of conformal blocks in the WZNW-model associated to the
compact group K, which can be defined in terms of the representation theory of
the affine Lie algebra ĝk with level k associated to the Lie algebra g of K,
(ii) MWk () is the natural action of the mapping class group on Hk (), which
W
can be defined by means of the monodromy representation of the Knizhnik–

Zamolodchikov connection,
(iii) HkRT () is the space of invariants in certain tensor products of representations
of the quantum group Uq (g),
k () is the mapping class group representation on Hk () defined by the
(iv) MRT RT
construction of Reshetikhin–Turaev.
2 The equivalence between the different quantization schemes has not been discussed in detail so far. It
boils down to the verification that the monodromy representation of the KZ-connection constructed within the
geometric quantization framework of [27], [1] is equivalent to the mapping class group representation defined in
[2], [3], [8]. It seems that e.g. combining the results of [36] and [6] does the job.
The quantization program (1.1) can be seen as a non-compact analog of (1.3) in the
following sense. In (1.1) the role of the moduli space of flat connections MK () is
taken by the Teichmüller space T (), which can be identified with the component in
the moduli space MG () of flat G = SL(2, R)-connections that has maximal Euler
class [26], [23]. Moreover, the natural symplectic structure on the moduli space of flat
SL(2, R)-connections restricts to the Weil–Petersson symplectic form on T () [22].
A quantization of the Teichmüller space T () may therefore be regarded as providing
a quantization of a topological component in the moduli space MG ().
We expect that any non-compact counterpart of the developments mentioned above
will be mathematically at least as rich as the already known results associated to com-
pact groups K. In particular we expect that certain analogs of the constructions of
Reshetikhin–Turaev and/or Turaev–Viro will capture information on the geometry of
hyperbolic three manifolds, similar and probably related to the appearance of hyper-
bolic volumes in the asymptotic behavior of certain link invariants [29].
1.2 Aims
A major step towards establishing H. Verlinde’s conjecture (1.2) is to show that the
quantization of the Teichmüller spaces (1.1) as initiated in [15], [17], [30] produces an
analog of a modular functor. The basic data of a modular functor are assignments such
as (1.3), which are required to satisfy a natural set of axioms as discussed in Section 9.
One of the most important implications of the axioms of a modular functor are simple
relations between the representations of the mapping class groups associated to and
†c respectively, where †c is the surface that is obtained from by cutting along a
simple closed curve c. These relations imply that the representation MM k () restricts
to – and is generated by – the representations MM ( ) which are associated to those
k
subsurfaces that can be obtained from by cutting along a set of non-intersecting
simple closed curves. This crucial locality property can be seen as the hard core of
the notion of a modular functor.
Within the formalisms introduced in [15], [17], [30]–[32] it is far from obvious that
the quantization of Teichmüller spaces constructed there has such properties. To show
that this is indeed the case is the main problem solved in this chapter. The representa-
tion MTb () constructed and investigated in [30]–[32] is obtained by exploiting the fact
that the mapping class group can be embedded into the so-called Ptolemy groupoid
associated to the transformations between different triangulations of a Riemann sur-
face . A representation of the Ptolemy groupoid is constructed in [30]–[32], which
then canonically induces a projective unitary representation of the mapping class group
MC(). The simplicity of the Ptolemy groupoid, which underlies the elegance of the
constructions in [30]–[32] now turns out to cause a major problem from the point of
view of our aims, since the above-mentioned locality properties implied in the notion
of a modular functor are not transparently realized by the Ptolemy groupoid.
Essentially our task is therefore to go from triangulations to pants decompositions,
which is the type of decomposition of a Riemann surface that is naturally associated
690 Jörg Teschner
to the concept of a modular functor. This requires to construct a change of representa-

tion for HbT () from the one naturally associated to triangulations of [30]–[32] to
another one which is associated to pants decompositions. The main tool for doing this
are the geodesic length operators introduced and studied in [15], [16] and [32], [33],
which are the observables on the quantized Teichmüller spaces that are associated
with the geodesic length functions3 on the classical Teichmüller spaces. The length
operators associated to a maximal set of non-intersecting simple closed curves turn
out to furnish a set of commuting self-adjoint operators, and the simultaneous diago-
nalization of these operators defines the sought-for change of representation.
There is a natural groupoid associated with the transformations between different
pants decompositions. Of particular importance for us will be a certain refinement of
this groupoid which will be called the modular groupoid M(). The modular groupoid
M() has been introduced for the study of rational conformal field theories by Moore
and Seiberg in [38], and it was further studied in particular in [5]. Constructing a
modular functor is essentially equivalent to constructing a tower of representations of
the modular groupoid. Our main aim in the present exposition will be to show that the
quantization of Teichmüller spaces allows one to construct a tower of representations
of the modular groupoid by unitary operators in a natural way.
1.3 Overview
This chapter has three main parts. The first of these parts collects the necessary
results from the “classical” theory of Riemann surfaces. This includes a review of
two types of coordinate systems for the Teichmüller spaces T (), one of which is
associated to triangulations of , the other to pants decompositions. The coordinates
associated to triangulations were first introduced by Penner in [39]. We will also need
to discuss variants of these coordinates due to Fock [15] and Kashaev [30] respectively.
The changes of the underlying triangulation of generate a groupoid, the Ptolemy
groupoid Pt(), which has a useful representation in terms of generators and relations
(Theorem 2).
The coordinates associated to pants decompositions are the classical Fenchel–
Nielsen coordinates, which we review briefly in §5.1. We furthermore explain how
the coordinates of Penner [39] and Kashaev [30], which were originally introduced to
parameterize the Teichmüller spaces of surfaces with punctures only, can be used
to provide coordinates also for the case where the surface has holes represented by
geodesics of finite length.
The material in this part is mostly known, but it is scattered over many places in
the literature, and some basic results were stated in the original references without
a proof. We have therefore tried to give a reasonably self-contained and complete
3 See [57], [58], [59] for some classical work on the symplectic nature of the Fenchel–Nielsen coordinates
which represents important background for our results.
presentation of the relevant material, providing proofs where these are not available
elsewhere.
The second part gives a largely self-contained presentation of the foundations of the
quantization of Teichmüller spaces. Our presentation is heavily inspired by [30]–[32],
but we deviate from these references in some important points. The treatment pre-
sented in this chapter seems to be the first complete and mathematically rigorous
formulation of the quantum theory of the Teichmüller spaces.
The main aims of this chapter are finally achieved in the third part. We begin in
§9 by introducing the notion of a stable unitary modular functor, and by explaining
why having a stable unitary modular functor is equivalent to having a tower of unitary
projective representations of the modular groupoid.
In §10 we will reformulate the main result of [38], [5] concerning the description
of M() in terms of generators and relations in a way that is convenient for us.
Of particular importance for us will be §11, where important first relations between
certain subgroupoids of M() and Pt() are observed.
In §12 we define the geodesic length operators and establish their main proper-
ties. These results are of independent interest since some important properties of the
geodesic length operators had not been proven in full generality before.
A key step in our constructions is taken in §13 by constructing a change of rep-
resentation from the original one to a representation in which the length operators
associated to a pants decomposition are simultaneously diagonalized. An important
feature of this construction is the fact that the unitary operator which describes the
change of representation factorizes into operators associated to the individual three
holed spheres (trinions) which appear in a pants decomposition.
In §14 we construct the corresponding representation of the modular groupoid
M(). The operators which represent M() are constructed out of compositions of
the representatives for the transformations in Pt(). This makes it relatively easy to
verify the relations of M(), but the price to pay is that some crucial locality properties
are more difficult to prove.
1.4 Outlook
In a sequel [51] to this exposition we will calculate the matrix coefficients of the
operators which generate the representation of the modular groupoid explicitly. A
close relation to the modular double DUq (sl(2, R)) of Uq (sl(2, R)) as defined and
studied in [12], [42], [9] will be found.
It should be noted that (HbT (), MTb ()) will not satisfy all the usual axioms of a
modular functor, which require, in particular, that the vector space V() assigned to
each Riemann surface should be finite-dimensional. Most importantly, however, the
assignment → (HbT (), MTb ()) was up to now only constructed for surfaces
which have at least one boundary component.
What will allow us to overcome this unsatisfactory feature are the remarkable
analytic properties that the matrix coefficients of the operators which represent the
692 Jörg Teschner
modular groupoid will be shown to have. It turns out that the mapping class group
representation MTb () assigned to a Riemann surface with a boundary represented by
geodesics of a fixed length depends analytically on the values of these lengths. The
analytic properties of the matrix coefficients will furthermore allow us to “close a hole”
by taking a limit where the length parameter assigned to this boundary component
approaches a certain imaginary value. It will be shown in [51] that the resulting
mapping class group representation is equivalent to the one on the surface which is
obtained by gluing a disc into the relevant boundary component.
Concerning the representation theoretic side of H. Verlinde’s conjecture (1.2) it
should be mentioned that a complete mathematical construction of (HcL (), McL ())
is not available so far, but nontrivial steps in the direction of constructing and describing
(HcL (), MLc ()) precisely have been taken in [50] in the case of surfaces of genus
zero. This includes in particular the derivation of explicit formulae for a set of basic
data which characterize the resulting representation of the braid group uniquely.
The explicit computation of the matrix coefficients of the operators which generate
the representation of the modular groupoid carried out in [51] will therefore allow us
to verify H. Verlinde’s conjecture (1.2) in the case of Riemann surfaces of genus zero.
We furthermore expect that it should be possible to construct non-compact analogs
of (HkRT (), MRT k ()) based the non-compact quantum group DUq (sl(2, R)), and
thereby complete a non-compact analog of the triangle (1.4).
Acknowledgements. The author would like to thank L. Chekhov, V. Fock and espe-
cially R. Kashaev for useful discussions.
We furthermore gratefully acknowledge financial support from the DFG by a
Heisenberg fellowship, as well as the kind hospitality of the Humboldt University
Berlin, the School of Natural Sciences of the IAS Princeton, the Physics Department
of Chicago University and the Caltech, Pasadena.
Part I. Coordinates for the Teichmüller spaces
We will consider two-dimensional surfaces with genus g ≥ 0 and s ≥ 1 boundary

components such that
M ≡ 2g − 2 + s > 0.
On we will consider metrics of constant negative curvature −1. Our main interest
will be the case where the boundary components can be represented by geodesics of
finite length. Such boundary components will also be called holes in the following.
However, to begin with we will focus on the case where the boundary components are
punctures, i.e. holes of vanishing geodesic circumference.
The space of deformations of the metrics of constant negative curvature is called
the Teichmüller space T (). It will be of basic importance for us to have useful
systems of coordinates for T ().
We will consider two classes of coordinate systems which are associated to two
types of graphs drawn on the Riemann surfaces respectively. The first class of coordi-
nates goes back to Penner and is associated to triangulations of the Riemann surface
or the corresponding dual graphs, the so-called fat graphs. We will also describe two
useful variants of the Penner coordinates due to Kashaev and Fock respectively.
The second class of coordinates are the classical Fenchel–Nielsen length-twist
coordinates. One may view them as being associated to a second type of graph on a
surface called marking that determines in particular a decomposition of the surface
into three-holed spheres (trinions).
In the following sections of the first part of this chapter we shall describe these
coordinate systems in some detail, discuss the graphs on that these coordinates
are associated to, as well as the groupoids generated by the transformations between
different choices of these graphs.
2 The Penner coordinates
2.1 Triangulations and fat graphs
Consider a fixed oriented topological surface of genus g with s ≥ 1 punctures. An

ideal triangulation τ of is the isotopy class of a collection of disjointly embedded
arcs in running between the punctures such that τ decomposes into triangles.
There are 2M triangles and 3M edges for any ideal triangulation. As an example we
have drawn a triangulation of the once-punctured torus in Figure 1.
The graph dual to a triangulation is a trivalent fat graph, i.e. a trivalent graph
embedded in the surface with fixed cyclic order of the edges incident to each vertex.
An example for a fat graph on the once-punctured torus is depicted in Figure 2. The
sets of vertices and edges of a fat graph ϕ will be denoted ϕ0 and ϕ1 respectively.
694 Jörg Teschner
0
1
0
1
P 0
1
0
1
111
000
000
111
000
111
000
111
h
Figure 1. Triangulation of the once-punctured torus.
01
10
1
0
0
1
Figure 2. Another representation of the triangulation from Figure 1 and the dual fat graph.
An ideal triangulation is called decorated if the triangles are numbered and if a

corner is marked for each triangle [30]. The decoration of the triangle tv dual to a
vertex v ∈ ϕ0 can be used to fix a numbering convention for the edges eiv , i = 1, 2, 3
which emanate from v as defined in Figure 3.
*
e2v e2v
v
t
e3v
Figure 3. Graphical representation of the vertex v dual to a triangle t. The marked corner defines
a corresponding numbering of the edges that emanate at v.
Remark 1. Decorated ideal triangulations are dual to decorated fat graphs, which
means that the vertices are numbered, and for each vertex v ∈ ϕ0 one has chosen a
distinguished edge ev ∈ ϕ1 . As a convention we will assume that fat graphs always
carry such a decoration unless otherwise stated.
2.2 Penner coordinates

It turns out to be useful to consider a somewhat enlarged object which keeps track
of the choices of horocycles around each of the punctures, the so-called decorated
Teichmüller space T˜ (). T˜ () is defined as a principal Rs+ -bundle over T () by
taking the s-tuple of horocycles around each of the punctures as the fiber over a point of
T (). The ordered s-tuple of hyperbolic lengths of the horocycles gives coordinates
for the fibers.
Given any point P in the decorated Teichmüller space T˜ () and an ideal triangu-
lation τ of , Penner assigns a coordinate value to each of the edges in τ by means of
the following construction. By means of Fuchsian uniformization one may equip the
surface with a unique hyperbolic metric g associated to our chosen point P ∈ T˜ ().
Let τ1 be the set of edges of a triangulation τ . Each edge e in τ1 may be straightened
to a geodesic for the hyperbolic metric g. The coordinate le (P ) is defined as the hyper-
bolic length of the segment of e that lies between the two horocycles surrounding the
punctures that e connects, taken with positive sign if the two horocycles are disjoint,
with negative sign otherwise. We are going to consider the tuple (le )e∈τ1 as a vector
in the vector space Rτ1 of dimension 3M.
Theorem 1 (Penner [39], [40]). (a) For any fixed ideal triangulation τ of , the
function
l : T˜ () → Rτ1 , P → (le (P ))e∈τ1
is a homeomorphism.
(b) The pull-back of the Weil–Petersson two-form ω on T () is given by the
expression

ω=− dle1 (t) ∧ dle2 (t) + dle2 (t) ∧ dle3 (t) + dle3 (t) ∧ dle1 (t) ,
t∈τ2
where the summation is extended over the set τ2 of triangles of τ , and ei (t), i = 1, 2, 3
are the edges bounding the triangle t, labelled in the counter-clockwise sense.
The Teichmüller space T () itself can finally be described as the space of orbits
in T˜ () under the following symmetry. Choose a number d(p) for each puncture p.
Let the action of the symmetry be defined by
le ≡ le + d(p) + d(p ) (2.1)
if the edge e connects the punctures p and p .
2.3 Fock coordinates

There is a useful variant of the Penner coordinates which was introduced by V. Fock
in [15]. In terms of the Penner coordinates one may define the Fock coordinates in
696 Jörg Teschner
a b
e
*
t1 t2
d c
*
Figure 4. Two adjacent triangles and the dual fat graph.
terms of certain cross-ratios. Given a quadrilateral formed by two adjacent triangles

we are going to keep the labelling of the edges introduced in Figure 4. Let
ze = la + lc − lb − ld . (2.2)
The dependence of the Penner coordinates on the choice of horocycles drops out
in the Fock coordinates. However, the variables ze assigned to the 3M edges in a
triangulation are not all independent. To describe the relations that they satisfy it is
convenient to think of the Fock coordinates as being assigned to the edges of the fat
graph dual to the given triangulation. Each closed curve c on is homotopic to a
unique path gc on the fat graph which has minimal length w.r.t. the metric defined
by assigning each edge of ϕ the length one. Such paths will also be called graph
geodesics in the following. The path gc may be described by a sequence of edges
e1c , . . . , enc c ∈ ϕ1 . To a closed curve c let us associate
nc

fϕ,c ≡ zeic . (2.3)
i=1
The definition (2.2) then implies the relations

fϕ,c = 0 (2.4)
for any closed curve c that encircles one puncture only. If one uses the equations (2.4)
to express s of the variables ze in terms of the others one obtains a set of coordinates
for T ().
On the 3M-dimensional space Fϕ that is spanned by the coordinate functions ze (P )
one may define a Poisson bracket WP which is such that
(i) the elements fa , a = 1, . . . , s span the set Cϕ of all c ∈ Fϕ such that
WP (c, w) = 0 for all w ∈ Fϕ , (2.5)
(ii) the Poisson bracket which is induced by WP on the quotient Fϕ /Cϕ coincides
with the Poisson bracket which corresponds to the Weil–Petersson symplectic
form on T ().
There exists a rather simple description for this Poisson-bracket [15]:

WP (ze , ze ) = ne,e , where ne,e ∈ {−2, −1, 0, 1, 2}. (2.6)
The value of ne,e depends on how edges e and e are embedded into a given fat graph.
If e and e don’t have a common vertex at their ends, or if one of e, e starts and ends
at the same vertex then ne,e = 0. In the case that e and e meet at two vertices one
has ne,e = 2 (resp. ne,e = −2) if e is the first edge to the right4 (resp. left) of e
at both vertices, and ne,e = 0 otherwise. In all the remaining cases ne,e = 1 (resp.
ne,e = −1) if e is the first edge to the right (resp. left) of e at the common vertex.
The coordinates ze also have a nice geometrical meaning [15]. In the Fuchsian
uniformization the two triangles that share the common edge e will be mapped into
ideal hyperbolic triangles in the upper half plane. The edges are then represented by
half-circles, and the corners will be at points x1 , . . . , x4 on the real line, see Figure 5.
a
ti e
tj b
d
c
x1 x2 x3 x4
Figure 5. Representation of the triangles ti and tj in the upper half plane.
We then have
(x4 − x1 )(x3 − x2 )
exp(ze ) = . (2.7)
(x4 − x3 )(x2 − x1 )
By means of Möbius transformations xi → ax i +b
cxi +d one may map the corners of one of
the two triangles to −1, 0 and ∞ respectively. The variable ze , being expressed in terms
of the Möbius-invariant cross-ratio therefore parameterizes the different ways of gluing
two ideal hyperbolic triangles along a common edge modulo Möbius transformations.
Given the variables ze one may reconstruct the Riemann surface as represented in
the Fuchsian uniformization by successively mapping ideal hyperbolic triangles into
the upper half-plane, glued along the edges e in the way prescribed by the given
value ze [15].
3 The Ptolemy groupoids

In the previous section we had associated coordinate systems to fat graphs on a surface
: A change of graph will of course induce a change of coordinates. The groupoid
4 W.r.t. to the orientation induced by the embedding of the fat-graph into the surface.
698 Jörg Teschner
generated by the moves between different fat graphs will be the subject of the present
section.
3.1 Groupoids vs. complexes

The groupoids that we will be interested in can be conveniently described as 2-
dimensional connected CW complexes G. The set of vertices G0 of G will be repre-
sented by certain sets of graphs, within this section called fat graphs. The (directed)
edges E ∈ G1 that connect these vertices correspond to the generators (“elementary
moves”) of the groupoid, while the faces F ∈ G2 of G yield the relations.
The groupoid G associated to the 2-dimensional connected CW complexes G will
then simply be the path groupoid of G, which has the vertices in G0 as objects and the
homotopy classes of edge paths between two vertices as morphisms. The homotopy
class of paths leading from vertex V1 ∈ G0 to vertex V2 ∈ G0 will be denoted by
[V2 , V1 ]. Similarly we will sometimes use the notation [WE , VE ] for the element of G
which corresponds to an edge E ∈ G1 .
A path π which represents an element in the homotopy class [W, V ] may be
represented by a chain of edges E ∈ G1 , i.e. an ordered sequence (Eπ,n(π) , . . . , Eπ,1 ),
Eπ,j ∈ G1 for j = 1, . . . , n(π) such that Eπ,j ∈ [Vπ,j +1 , Vπ,j ] for j = 1, . . . ,
n(π ) − 1, and Vπ,1 = V , Vπ,n(π) = W . We will also use the suggestive notation
En En−1 · · · E1 to denote a chain.
3.2 Change of the triangulation

In the case of the Ptolemy groupoids Pt() we will consider a complex P t (), where
the set G0 = P t0 () is defined to be the set of fat graphs on . Let us furthermore
define P t1 () to consist of the following elementary moves.
(i) Permutation (vw). Exchanges the labels of the vertices v and w.
(ii) Rotation ρv . See Figure 6.
*
v v
*
Figure 6. Transformation ρv changes the marked corner of the triangle dual to a vertex v ∈ ϕ0 .
(iii) Flip ωvw . See Figure 7.
Proposition 1 ([39, Proposition 7.1], [32]). The complex P t () is connected, i.e.
for a given surface , any two fat graphs ϕ and ϕ can be connected by a chain of
elementary transformations.
a b a b
v
v e w * e
*
w
d c
* d * c
Figure 7. The flip transformation ωvw changes the diagonal in the quadrilateral formed by the
two adjacent triangles tv and tw .
Validity of the following relations in Pt() can easily be verified pictorially [32].
ρv ρv ρv = id,
ωu1 u2 ωu3 u4 = ωu3 u4 ωu1 u2 , ur = us for r = s,
ωvw ωuw ωuv = ωuv ωvw , (3.1)
(ρv−1× ρw ) ωvw = ωwv (ρv−1
× ρw ),
ωwv ρv ωvw = (vw) (ρv × ρw ).
Theorem 2. The complex P t () is simply connected, i.e. any relation between the
generators (vw), ρv and ωvw of the Ptolemy groupoid is a consequence of the relations
(3.1) together with the relations of the permutation group.
The proof of this theorem is explained in Appendix C.

One of the main virtues of the Penner coordinates is that the corresponding change
of coordinates can be described rather simply.
Lemma 1 (Lemma A.1a of [40]). Let τ be the triangulation obtained by applying the
flip of Figure 7 to a pair of adjacent triangles in a given initial triangulation τ , and
denote e and e the diagonal edge before and after the flip. The coordinates associated
to τ and τ will then agree for each edge that the two triangulations have in common,
and
1 √
λe = (λa λc + λb λd ), λf ≡ 2 exp 21 lf for all f ∈ ϕ1 , (3.2)
λe
where we have labelled the edges according to Figure 7.
The corresponding transformation of the Fock variables is also easy to describe:

e−za = e−za (1 + e−ze ) e−zb = e+zb (1 + e+ze )
ze = −ze (3.3)
e+zd = e+zd (1 + e+ze ) e+zc = e−zc (1 + e−ze )
700 Jörg Teschner
3.3 The representation of the mapping class group

The mapping class group MC() of the topological surface is the group of isotopy
classes of orientation-preserving diffeomorphisms of . An element μ ∈ MC()
of the mapping class group will map a given (decorated) triangulation τ into another
one, μ.τ . The fact that any two triangulations can be mapped into each other by a
composition of the elementary transformations introduced in Section 3.2 therefore
leads to an embedding of the mapping class group into the Ptolemy groupoid:
τ : MC() → Pt(), τ (f ) ≡ [μ.τ, τ ]. (3.4)
More precisely, τ induces a homomorphism MC() → Pt() in the sense that
τ (μ2 μ1 ) = μ1 .τ (μ2 ) τ (μ1 ) for any μ2 , μ1 ∈ MC(),
which embeds MC() injectively into Pt() [41, Theorem 1.3].
4 Teichmüller space as the phase space of a constrained system

As a preparation for the description of the quantum Teichmüller spaces it will be useful
to parameterize the Teichmüller spaces by means of variables assigned to the triangles
instead of the edges of a triangulation [30]. In the following section we shall elaborate
upon the results and constructions in [30], strengthening them somewhat.
4.1 Kashaev’s coordinates

Assume given a fat graph ϕ with set of vertices ϕ0 . For each vertex v ∈ ϕ0 one may
introduce a pair of variables (qv , pv ) according to the following rule. Let us label the
edges that emanate from the vertex v by eiv , i = 1, 2, 3 according to Figure 3. We
will denote the Penner coordinates associated to the edges eiv by liv , i = 1, 2, 3. Let
us then define the pair of variables (qv , pv ) as

(qv , pv ) = l3v − l2v , l1v − l2v . (4.1)
Following Kashaev [30] we will consider the vector space Vϕ R4M obtained by
regarding the variables qv , pv as the components qv (v), pv (v) of vectors v ∈ Vϕ . The
space of linear coordinate functions on Vϕ will be called the Kashaev space Wϕ . On
Wϕ we will consider the Poisson bracket ϕ defined by
ϕ (pv , qw ) = δvw , ϕ (qv , qw ) = 0, ϕ (pv , pw ) = 0. (4.2)
The assignment (4.1) associates a vector v(P ) in a subspace Tϕ ⊂ Vϕ to each point
P ∈ T̃ (). Kashaev has observed that the subspace Tϕ can be characterized by a
suitable set of linear forms hc ∈ Wϕ (“constraints”).
4.1.1 The constraints. To define the linear forms hc let us introduce an embedding
of the first homology H1 (, R) into Wϕ as follows. Each graph geodesic gc which
represents an element [c] ∈ H1 (, R) may be described by an ordered sequence of
vertices vi ∈ ϕ0 , and edges ei ∈ ϕ1 , i = 0, . . . , n, where v0 = vn , e0 = en , and we
assume that vi−1 , vi are connected by the single edge ei . We will define ωi = 1 if the
arcs connecting ei and ei+1 turn around the vertex vi in the counterclockwise sense,
ωi = −1 otherwise. The edges emanating from vi will be numbered eji , j = 1, 2, 3
according to the convention introduced in Figure 3. To each [c] ∈ H1 (, R) we will
assign
⎧
n ⎪
⎨−qvi if {ei , ei+1 } = {e3i , e1i },
hc ≡ ui , ui := ωi pvi if {ei , ei+1 } = {e2i , e3i }, (4.3)
⎪
⎩ i i
i=1 qvi − pvi if {ei , ei+1 } = {e1 , e2 }.
hc is independent of the choice of representative c within the class [c]. Let Cϕ be the
subspace in Wϕ that is spanned by the hc , [c] ∈ H1 (, R).
Lemma 2 ([30]). (i) The mapping H1 (, R) [c] → hc ∈ Cϕ is an isomorphism of

vector spaces.
(ii) The restriction of ϕ to Cϕ coincides with the intersection form I on H1 (, R),
ϕ (hc1 , hc2 ) = I(c1 , c2 ).
(iii) The linear forms hc , [c] ∈ H1 (, R) vanish identically on the subspace Tϕ .
The equations hc (v) = 0, [c] ∈ H1 (, R) characterize the image of T̃ () within
Vϕ . It is useful to recall that H1 (, R) splits as H1 (, R) = H1 (cl , R) ⊕ B(),
where B() is the s − 1-dimensional subspace spanned by the homology classes
associated to the punctures of , and cl is the compact Riemann surface which is
obtained by “filling” the punctures of . The corresponding splitting of Cϕ will be
written as Cϕ = Hϕ ⊕ Bϕ .
4.1.2 Change of fat graph. In order to describe the change of Kashaev variables
induced by a change of fat graph let us, following [30], define the following two
transformations associated to the elementary moves ωvw and ρv respectively.
Av : (qv , pv ) → (pv − qv , −qv ), (4.4)

(Uv , Vv ) → Uv Uw , Uv Vw + Vv ,
Tvw : (4.5)
(Uw , Vw ) → Uw Vv (Uv Vw + Vv )−1 , Vw (Uv Vw + Vv )−1 ,
where we have set Uv ≡ eqv and Vv ≡ epv for all v ∈ ϕ0 .
Lemma 3. The maps Av : Wϕ → Wρv ϕ and Tvw : Wϕ → Wωvw ϕ defined in (4.4)

and (4.5) respectively are canonical, i.e. they preserve the Poisson structure ϕ .
702 Jörg Teschner
The proof is again straightforward. Lemma 3 implies in particular that the mapping
class group acts on Wϕ by canonical transformations.
4.2 The structure of the Kashaev space Wϕ

4.2.1 Fock variables vs. Kashaev’s variables. There is a canonical way to recon-
struct the Fock-variables in terms of Kashaev’s variables which is found by combining
equations (2.2) and (4.1). The result may be formulated as follows. Let v, w ∈ ϕ0 be
the vertices that are connected by the edge e ∈ ϕ1 , and let eiv , i = 1, 2, 3 be the edges
introduced in Figure 3.
⎧
⎪
⎨pv if e = e1v ,
ẑe = ẑe,v + ẑe,w , ẑe,v = −qv if e = e2v , (4.6)
⎪
⎩ v
qv − pv if e = e3 .
The definition (4.6) defines a linear map Iϕ : Fϕ → F̂ϕ ⊂ Wϕ . It will be useful to

describe the properties of this map a bit more precisely.
Lemma 4. (i) ẑe (v(P )) = ze (P ) for all e ∈ ϕ1 , P ∈ T˜ ().

(ii) ϕ (ẑe , ẑf ) = WP (ze , zf ) for all e, f ∈ ϕ1 .
(iii) ϕ (ẑe , hc ) = 0 for all e ∈ ϕ1 , c ∈ H1 (, R).
(iv) fˆc ≡ Iϕ (fc ) = hc for all [c] ∈ B().
Proof. Straightforward verifications.

It is also useful to remark that the transformation of the Fock variables ẑe , e ∈ ϕ1
that is induced by (4.4), (4.5) coincides with (3.3).
4.2.2 Splitting of Wϕ . The linear forms hc ∈ Bϕ turn out to be the Hamiltonian

generators for the symmetry (2.1) [30]. It is therefore natural to consider the subspace
Mϕ ⊂ Wϕ which is spanned by the Hamiltonian vector fields that are generated by
the linear forms hc ∈ Bϕ , as well as Nϕ ≡ Mϕ ⊕ Bϕ .
Proposition 2. There exists a canonical transformation establishing the isomorphism

of Poisson vector spaces Wϕ Tϕ ⊕ Nϕ ⊕ Hϕ , such that
(i) Tϕ T () is the space of linear functions on the Teichmüller space T ();
(ii) the restriction of ϕ to Tϕ coincides with the Poisson bracket induced by the
Weil–Petersson symplectic form.
Proof. As a warmup it may be instructive to count dimensions: We have dim(Wϕ ) =

8g − 8 + 4s and dim(Cϕ ) = dim(H1 (, R)) = 2g + s − 1. In order to deter-
mine dim(Nϕ ) let us choose a canonical basis for H1 (, R), represented by curves
α1 , . . . , αg , β1 , . . . , βg , γ1 , . . . , γs−1 such that the only nontrivial intersection pair-

ings are I(αi , βj ) = δij . Nϕ is spanned by the images of the classes [γ1 ], . . . , [γs−1 ],
together with the Hamiltonian vector fields that they generate. It follows that
dim(Nϕ ) = 2s − 2.
The main point that remains to be demonstrated is the existence of a decomposition
of F̂ϕ as the direct sum
F̂ϕ = Tϕ ⊕ Bϕ such that ϕ (t, h) = 0 for all t ∈ Tϕ , h ∈ Nϕ . (4.7)

Thanks to Lemma 3 we may choose a convenient fat graph to this aim. Let us pick a
basis B() for B() represented by the curves which encircle s −1 of the s punctures.
It is easy to see that we can always construct a fat graph ϕ such that the elements of
B() are represented by singles edges in ϕ1 . These edges end in a unique vertex v(c)
for each c ∈ B(). It is clear that the expression for hc only involves the variables
(qv(c) , pv(c) ) for all c ∈ B(). It follows from (4.2) that the Hamiltonian vector
field generated by hc can likewise be expressed in terms of (qv(c) , pv(c) ) only. The
existence of the sought-for decomposition (4.7) is obvious in this case.
The result is carried over to the general case with the help of Lemma 3. It is clear
that the subspace Tϕ ⊂ F̂ϕ is defined uniquely by the condition (4.7). Tϕ may then
also be described as the quotient of F̂ϕ by the conditions hc = 0 for all [c] ∈ B().
It therefore follows from our discussion in §2.3 that Tϕ is canonically isomorphic to
T (), the vector space of linear functions on T ().
To complete the proof it remains to observe that we have
(i) ϕ (t, h) = 0 for all t ∈ Tϕ , h ∈ Hϕ ,
(ii) ϕ (h, n) = 0 for all h ∈ Hϕ , n ∈ Nϕ .
(i) follows directly from Lemma 4, whereas part (ii) can easily be verified by consid-
ering the fat graph ϕ above.
5 The Fenchel–Nielsen coordinates

We will now be interested in the case of Riemann surfaces with a boundary that is
represented by a collection of s > 0 geodesics. Another standard set of coordinates
for the Teichmüller spaces is associated to the decomposition of a Riemann surface
into trinions (three-holed spheres). We are now going to review the definition of
these coordinates. Different sets of Fenchel–Nielsen coordinates will be associated to
different markings of the Riemann surface in a way which is analogous to the relation
between the Penner coordinates and fat graphs.
Let us denote by S3 the sphere with three holes (trinion). As a concrete model we
may e.g. choose
S3 ≡ {z ∈ C; |z| ≥ ε, |1 − z| ≥ ε, |z| < 1/ε}. (5.1)
704 Jörg Teschner
Any trinion with a smooth boundary is diffeomorphic to S3 .
Definition 1. A marking σ of a surface consists of the following data.

(i) A cut system Cσ , which is a set Cσ = {c1 , . . . , c3g−3+s } of simple non-intersecting
oriented closed curves ci on . Cutting along the curves in C decomposes the
surface into a collection Pσ of trinions.
(ii) A choice of a trivalent graph T with a single vertex vT in each trinion T ∈ Pσ
such that the graphs on the different trinions glue to a connected graph σ on .
(iii) A choice of a distinguished boundary component for each trinion T ∈ Pσ .
These data will be considered up to isotopy.
An example for the graphs T is depicted in Figure 10.
Figure 8. A trinion equipped with a marking graph.
5.1 Definition of the Fenchel–Nielsen coordinates

The basic observation underlying the definition of the Fenchel–Nielsen twist coor-
dinates is the fact that for each triple (l1 , l2 , l3 ) of positive real numbers there is a
unique metric of constant curvature −1 on the three-holed sphere (trinion) such that
the boundary components are geodesics with lengths li , i = 1, 2, 3. A trinion with
its metric of constant curvature −1 will be called hyperbolic trinion. There further-
more exist three distinguished geodesics on each hyperbolic trinion that connect the
boundary components pairwise.
Let us assume that the geodesic c separates two trinions Ta and Tb . Pick boundary
components ca and cb of Ta and Tb respectively by starting at c, following the marking
graphs, and turning left at the vertices. As mentioned above, there exist distinguished
geodesics on Ta and Tb that connect c with ca and cb respectively. Let δc be the signed
geodesic distance between the end-points of these geodesics on c, and let
δc
θc = 2π (5.2)
lc
be the corresponding twist-angle. In a similar way one may define θ in the case that
cutting along c opens a handle.
Given a cut system C = {c1 , . . . , cκ }, κ = 3g − 3 + s, we thereby associate to

each Riemann surface a tuple (l1 , . . . , lκ ; θ1 , . . . , θκ ) of real numbers. It can be
shown (see e.g. [28]) that the Riemann surface is characterized uniquely by the tuple
(l1 , . . . , lκ ; eiθ1 , . . . , eiθκ ) ∈ (R+ )κ ×(S 1 )κ . In order to describe the Teichmüller space
T () of deformations of it suffices to allow for arbitrary real values of the twist
angles θi . Points in T () are then parametrized by tuples (l1 , . . . , lκ ; θ1 , . . . , θκ ) ∈
(R+ )κ × Rκ .
Remark 2. The marking graph σ allows one to distinguish systems of Fenchel–

Nielsen coordinates which are related to each other by Dehn twists, θc = θc + 2π kc ,
k ∈ Z, c ∈ C. To use the markings for the parametrization of different systems
of Fenchel–Nielsen coordinates is then closely analogous to using fat graphs for the
specification of systems of Penner coordinates.
The definition of the Fenchel–Nielsen coordinates does not use the choice of a
distinguished boundary component for each trinion. The latter has been included into
the definition 1 for later convenience only.
5.2 Symplectic structure

Let us furthermore notice that the Weil–Petersson symplectic form becomes particu-
larly simple in terms of the Fenchel–Nielsen coordinates:
Theorem 3 ([58]).
κ
1
ωWP = dτi ∧ dli , τi = li θi . (5.3)
2π
i=1
The content of the theorem may also be paraphrased as follows:

(i) The geodesic length functions associated to non-intersecting closed curves Pois-
son-commute.
(ii) The Hamiltonian flows generated by the geodesic length functions coincide with
the Fenchel–Nielsen twist flows.
5.3 Geodesic lengths from the Penner coordinates

A nice feature of the Fock coordinates is that they lead to a particularly simple way to
reconstruct the Fuchsian group corresponding to the point P in Teichmüller space that
is parametrized by the variables ze (P ), e ∈ ϕ1 . Assume given a graph geodesic gc on
the fat graph homotopic to a simple closed curve c on . Let the edges be labelled ei ,
i = 1, . . . , r according to the order in which they appear on gc , and define σi to be 1
if the path turns left5 at the vertex that connects edges ei and ei+1 , and to be equal to
5 W.r.t. to the orientation induced by the embedding of the fat-graph into the surface.
706 Jörg Teschner
−1 otherwise. The generator X(c) of the Fuchsian group that corresponds to c is then
constructed as follows [15].
Xc = Vσr E(zer ) . . . Vσ1 E(ze1 ), (5.4)
where the matrices E(z) and V are defined respectively by

z

0 +e+ 2 1 1
E(z) = z , V = . (5.5)
−e− 2 0 −1 0
Given the generator Xc of the Fuchsian group one may then calculate the hyperbolic
length of the closed geodesic isotopic to c via

2 cosh 21 lc = |tr(Xc )|. (5.6)
The proof of (5.4) was omitted in [15]. We are therefore now going to explain how
to verify the validity of equation (5.4).
It was remarked in the Section 2.3 that for given values of the coordinates ze one
may construct a uniformized representation of the corresponding Riemann surface by
successively mapping ideal hyperbolic triangles into the upper half plane which are
glued according to the values ze . Iterating this procedure ad infinitum one generates a
tessellation of the upper half plane by ideal hyperbolic triangles. Let us now consider a
generator X(c) of the Fuchsian group, represented on the a upper
half plane by a Möbius
transformation MX(c) , where MX (u) ≡ au+b cu+d if X = b . The element c ∈ π ()
c d 1
may then be represented by an open path on the upper half plane which leads from a
chosen base point u to its image under MX .
But one may equivalently represent the motion along the path by standing still at the
base point and moving the tessellation around by means of Möbius transformations.
More precisely, let us assume that our base point u is located within the ideal hyperbolic
triangle t0 with corners at −1, 0, ∞, and that the path Pc representing our chosen
element c ∈ π1 () crosses the edges ei i = 1, . . . , r in the order of the labelling.
We may assume that the edge e1 connects the points 0 and eze1 . After having crossed
edge e1 one would have left the triangle t0 into the triangle t1 with corners at 0, eze1 , ∞.
The Möbius transformation M1 corresponding to E(ze1 ) brings one back into t0 : It
can be checked that it leaves the set of corners {−1, 0, eze1 , ∞} on the two triangles
glued along e1 invariant, but exchanges the two triangles. To continue along the path
Pc in this fashion we now need to map the next edge e2 to be crossed to the edge going
from 0 to ∞ before we can apply the Möbius transformation corresponding to E(ze2 )
in the same manner as before. This is precisely what the Möbius transformation MVσ1
does: It simply rotates the edges of our fundamental triangle t0 . Moreover, the triangle
t2 that would be reached when leaving t0 through e2 will be mapped by MVσ1 into
the ideal hyperbolic triangle with corners 0, eze2 , ∞. The fact that eze2 is indeed the
position that the corner of t2 is mapped into by MVσ1 follows from the prescription for
gluing t1 and t2 along e2 in terms of ze2 and the fact that MVσ1 preserves cross-ratios.
By continuing in this fashion one generates the Möbius transformation MX(c) that
evidently maps the original tessellation representing the chosen point P in T () into
another one that is equally good as a representation for P . By assumption, MX(c)

represents a closed path on the considered fat graph. This means that the points that
are mapped into each other by MX(c) are to be identified as different representatives
for the same points on the surface corresponding to our point P ∈ T ().
6 Coordinates for surfaces with holes of finite size
Throughout we are mainly interested in the case of Riemann surfaces which have a
boundary ∂ represented by s geodesics of finite length. We therefore need to discuss
how to introduce analogs of the previously described coordinate systems for T ()
for the cases of interest here. When considering surfaces with holes of finite size one
has to choose if one wants to keep the geodesic lengths of the boundary components
variable, or if one wants to consider surfaces which have fixed boundary length
given by the tuple = (l1 , . . . , ls ) ∈ Rs+ . We shall find the first option often more
convenient to work with. Passing to a representation in which the boundary lengths
are fixed will then be almost trivial.
6.1 Useful fat graphs on surfaces with holes of finite size
Riemann surfaces with s holes can always be represented by considering a Riemann

surface e with s pairs of punctures, from which is obtained by cutting e along
the geodesics b1 , . . . , bs that encircle the pairs of punctures. This simple observation
allows us to use the coordinates discussed previously in order to define coordinates for
the Teichmüller spaces of surfaces with s holes. In order to spell out more precisely
how to do this, let us first introduce a convenient class of fat graphs.
Let us consider a pair of punctures (P1 , P2 ). Let c be a geodesic such that cutting
along c produces two connected components one of which is a two-punctured disc
D with punctures P1 and P2 . A given fat graph ϕ e will be said to have standard form
near D if there exists a neighborhood of the disc D in which ϕ e is homotopic to the
fat graph depicted on the left half of Figure 9.
......
*
c P2 P1 c P2 P1
v *
Figure 9. Simple fat graphs in a neighborhood of the disc defined by the geodesic c which
encircles two punctures P1 and P2 .
708 Jörg Teschner
For surfaces e with s pairs of punctures there exist fat graphs ϕ e which are of
standard form in the neighborhood of s − 1 discs Di . The simplest possible form of
a fat graph around the remaining two punctures is indicated on the right of Figure 9.
A fat graph ϕ e on a surface with 2s punctures will be said to have standard form if it
has standard form near s − 1 discs Di , and if it has the form depicted on the right of
Figure 9 in a neighborhood of the remaining disc.
We will finally say that a fat graph ϕ on a Riemann surface with s geodesic
boundaries has standard form if (, ϕ) can be obtained from a pair ( e , ϕ e ) con-
sisting of a 2s-punctured surface and a fat graph ϕ e of standard form by cutting
e along s geodesics b1 , . . . , bs , each of which encircles a pair of punctures. The
embedding → e furthermore induces an embedding MC() → MC( e ) of the
respective mapping class groups. The subgroup of MC( e ) which is generated by the
diffeomorphisms that are supported on ⊂ e preserves the set of fat graphs which
have standard form.
6.2 Kashaev type coordinates

If we only use fat graphs of standard form, it becomes easy to adapt the previously
discussed systems of coordinates to the case of interest in the rest of this chapter. We
may in particular consider the space Wϕ of Kashaev variables associated to the fat
graph ϕ. A subspace Tϕ of Wϕ can again be defined by means of the decomposition
(4.7). It is furthermore convenient to introduce the set ϕ1 which only contains the
edges of ϕ that do not end in boundary components of .
Lemma 5. (i) We have Tϕ Tϕ e .

(ii) The Fock coordinates {ze ; e ∈ ϕ1 } form a set of coordinates for Tϕ .
Proof. Let us first note that the linear form h1 which is via (4.3) associated to puncture
P1 can be expressed in terms of the variables (pv , qv ) associated to the vertex v in
Figure 9 only. This means that both pv and qv are contained in Nϕ e . Instead of the
linear form h2 associated to puncture P2 we may consider hc = h1 + h2 , which can
be expressed exclusively in terms of the variables associated to the vertices contained
in . Part (i) of the lemma follows easily from these observations.
In order to verify part (ii) one may again consider ϕ e . When writing the relations
fc = 0, c ∈ B() in terms of the Fock variables ze , e ∈ ϕ1e one will always find
contributions containing the ze , e ∈ ϕ1e \ ϕ1 . It is then easy to convince oneself that
the relations fˆc = 0 may be used to express the ze , e ∈ ϕ1e \ ϕ1 in terms of the ze ,
e ∈ ϕ1 . After this is done, all relations fc = 0, c ∈ B() are satisfied, the variables
ze , e ∈ ϕ1 are therefore unconstrained.
However, in this case the relation between the Teichmüller space T () and Tϕ is
slightly more complicated. In order to describe this relation let us consider the spaces
Fun(T ()) and Fun(Tϕ ) of smooth functions on T () and Tϕ respectively. These
spaces carry canonical Poisson brackets { . , . }WP and { . , . }ϕ uniquely defined by the
bilinear forms WP and ϕ respectively.
Proposition 3. We may represent Fun(T ()) as the subspace of Fun(Tϕ ) which is

defined by the conditions
{F, li }ϕ = 0, i = 1, . . . , s, F ∈ Fun(Tϕ ), (6.1)
where li is the length function which is associated to the i-th boundary component via
equations (5.4) and (5.6).
It should be noted that the length functions associated to the boundary components
are contained in Fun(T ()).
Part II. Quantization of the Teichmüller spaces

Our aim is to construct certain classes of infinite-dimensional representations of the
mapping class groups MC(). One possible approach to this problem is to “quantize”
Poisson manifolds like the Teichmüller spaces on which MC() acts as a group of
symmetries.
Our construction will proceed in two main steps. Quantization of the Kashaev
spaces Wϕ leads to a rather elegant construction of projective unitary representations of
the mapping class groups [30]. However, these representations turn out to be reducible.
The second step will therefore be to identify distinguished subrepresentations within
the representations coming from the quantization of the Kashaev spaces as the mapping
class group representations which are naturally associated to the quantization of the
Teichmüller spaces. A direct construction of the latter is not known, which is why this
somewhat indirect construction seems to be most efficient at the moment.
7 Quantization of the Teichmüller spaces

7.1 Canonical quantization
Quantization of a Poisson manifold P means “deforming” the space of functions on
P into a one-parameter (h̄) family of noncommutative algebras Ph̄ in such a way that
the deformed product f ∗h̄ g satisfies
f ∗h̄ g = fg + h̄(f, g) + O(h̄2 ), (7.1)

where f g is the ordinary commutative product of functions on P and (f, g) is the
Poisson bracket on P . If the Poisson manifold P has a group G of symmetries it is
710 Jörg Teschner
natural to demand that these symmetries are preserved by quantization in the sense
that any g ∈ G is realized as an automorphism f → ag (f ) of Ph̄ .
Representations of the group G can be constructed by studying representations of
the algebra Ph̄ by operators O(f ) on a Hilbert space H ,
O(f ) · O(g) = O(f ∗h̄ g).
The Hilbert space H will then typically come equipped with a unitary projective repre-
sentation of the group G of symmetries by operators Ug such that the automorphisms
Ag (O(f )) ≡ O(ag (f )) are realized as
Ag (O) = Ug · O · U−1
g .
Quantization is particularly simple if there exist coordinate functions q1 , . . . , qN

and p1 , . . . , pN defined globally on P such that the Poisson bracket takes the form
(pv , qw ) = δvw , (qv , qw ) = 0, (pv , pw ) = 0, (7.2)
for v, w ∈ {1, . . . , N}. One may then define Ph̄ in such a way that the relations
i[pv , qw ] = h̄δvw , [qv , qw ] = 0, [pv , pw ] = 0, (7.3)
hold for [f, g] ≡ f ∗h̄ g − g ∗h̄ f . There exists a standard representation O of these
commutation relations on (dense subspaces of) the Hilbert space H = L2 (RN ) of
square-integrable functions (q), q = (q1 , . . . , qN ), which is generated by pairs of
operators
1
pv ≡ O(pv ), qv ≡ O(qv ), v = 1, . . . , N,
2π h̄
that are defined respectively by
1 ∂
qv (q) ≡ qv (q), pv (q) ≡ (q). (7.4)
2π i ∂qv
This simple example for the quantization of Poisson manifolds is often referred to as
“canonical quantization”.
Remark 3. The representation that is constructed in this way is irreducible in the

following sense. If O is a bounded operator on L2 (RN ) which commutes6 with the
operators pv and qv for all v = 1, . . . , N then O = χ , the operator of multiplication
with the complex number χ.
Following the discussion in Sections 2, it seems natural to define the quantized

Teichmüller spaces as the noncommutative algebra Th̄ () with generators ze and
6 Commutativity [O, A] = 0 with a self-adjoint unbounded operator A is, by convention, understood in the
sense of commutativity with the spectral projections of A. For the reader’s convenience, we have collected the
relevant operator-theoretical results in Appendix B.
relations
i[ze2 , ze1 ] = h̄ WP (ze2 , ze1 ) (7.5)
fϕ,c = 0, for all c ∈ B(). (7.6)
The space H() will be defined as an irreducible representation of the commutation
relations (7.5) which satisfies the additional conditions (7.6).
7.2 Quantization of the Kashaev space Wϕ

For each given fat graph ϕ let us define the Hilbert space K(ϕ) as the space of square
integrable functions (q) of the Kashaev-variables q = (q1 , . . . , q2M ). On K(ϕ) we
shall consider the basic operators pv , qv defined in (7.4) for v = 1, . . . , 2M. The
noncommutative algebras of operators which are generated by the operators qv , pw
with the commutation relations
[pv , qw ] = (2π i)−1 δv,w (7.7)
may be considered as representing quantized algebras of functions on the Kashaev
space Wϕ .
Remark 4. It is worth noting that the decoration of the triangles is used to define the
concrete realization of the space K(ϕ) as a space of square integrable functions.
When quantizing the Kashaev space Wϕ we get more than we ultimately want. In
order to see this, let us introduce quantum analogs of the coordinate functions hc and
ẑe respectively, i.e. self-adjoint operators ĥc and ẑe on H (ϕ) which are defined by
formulae very similar to (4.3) and (4.6) respectively, normalized in such a way that
the following commutation relation hold:
i[ ẑe1 , ẑe2 ] = h̄ WP (ze1 , ze2 ), (7.8)
i[ ĥc1 , ĥc2 ] = h̄ I(c1 , c2 ). (7.9)
[ ẑe , ĥc ] = 0 for all c ∈ H1 (, R). (7.10)
We observe that we do have a representation, henceforth denoted Zϕ , of the algebra
(7.5), but this representation is neither irreducible, nor does it fulfill the additional
relations (7.6). The latter point becomes most clear if one introduces the operators f̂c ,
c ∈ B() associated to the relations (7.6) which are defined by replacing ze → ẑe in
(2.3). We may then observe that
f̂c = hc for c ∈ B(), (7.11)
which is verified in the same way as statement (iv) in Lemma 4.
712 Jörg Teschner
7.3 Reduction to the quantized Teichmüller spaces

In order to see that the representations Zϕ “contain” irreducible representations of
the quantized Teichmüller spaces Th̄ (), let us consider the noncommutative algebra
Zh̄ () with generators ze and the only relations (7.5). One should observe that the
fc generate the center of the algebra Zh̄ . Irreducible unitary representations Zf of this
algebra are parametrized by linear functions f : B() → R. Such representations are
such that the operators fc = Zf (fc ) are realized as the operators of multiplication with
the real numbers f(c), c ∈ B().
The representations can be constructed concretely by forming linear combinations
tk and tk∨ , k = 1, . . . , 3g − 3 + s of the ze which
(i) are mutually linearly independent, and linearly independent of the fc , and
(ii) which satisfy the commutation relations

2π tk , tk∨ = i δkk . (7.12)
Canonical quantization realizes Zf on the Hilbert space Hf (ϕ) L2 (R3g−3+s ) which

consists of square-integrable functions (t), t = (t1 , . . . , t3g−3+s ). The operators
zf,e ≡ Zf (ze ) are then realized as linear combination of the operators
1 ∂
tk f (t) = tk f (t), tk∨ f (t) = (t), fc f (t) = f(c)f (t),
2π i ∂tk f
where k = 1, . . . , 3g − 3 + s and c ∈ B().
Our aim is to describe how the representation Zϕ decomposes into the representa-
tions Zf . In order to do this, let us introduce the representation
⊕ ⊕

Zϕ ≡ df Zf on the space Hz (ϕ) ≡ df Hf (ϕ).
B () B ()
The space Hz (ϕ) is spanned by square-integrable families ≡ (f )f∈B () of func-
tions f ∈ Hf (ϕ) which are associated to the linear functions f : B() → R in the
dual B () Rs−1 up to a set of measure zero. The representatives ze ≡ Zϕ (ze ) are
defined as follows
ze ≡ zf,e f f∈B (). (7.13)
Proposition 4. The decomposition of the representation Zϕ into irreducible represen-

tations of Zh̄ may be written as follows:

⊕
Zϕ df Zf ⊗ 1Hh (ϕ) , (7.14)
B ()
where the space Hh (ϕ) is isomorphic to L2 (Rg ). There exists a unitary operator Iϕ ,
Iϕ : K(ϕ) → Hz (ϕ) ⊗ Hh (ϕ) such that
Iϕ · ẑe · I−1 −1
ϕ = ze ⊗ 1 and Iϕ · ĥc · Iϕ = 1 ⊗ hc , (7.15)
for any e ∈ ϕ1 and c ∈ H1 (cl , R), respectively.
Proof. Let us recall the direct sum decomposition

Wϕ Tϕ ⊕ Nϕ ⊕ Hϕ . (7.16)
To each of the three spaces Tϕ , Nϕ , Hϕ one may choose coordinates which bring the
Poisson bracket to the canonical form (7.2). The corresponding operators
( t̂k , t̂k∨ ), k = 1, . . . , 3g − 3 + s,
∨
( f̂l , f̂l ), l = 1, . . . , s − 1,
∨
( ĥm , ĥm ), m = 1, . . . , g,
can be constructed as linear combinations of the (pv , qv ), v ∈ ϕ1 , in such a way that
the only nontrivial commutation relations are

2π t̂k , t̂k∨ = ib2 δkk , k, k = 1, . . . , 3g − 3 + s,

2π f̂l , f̂l∨ = ib2 δll , l, l = 1, . . . , s − 1, (7.17)
∨

2π ĥm , ĥm = ib δmm , m, m = 1, . . . , g.
2
It will be convenient to form the following vectors with 4M operator-valued compo-

nents:
v = (. . . , qv , . . . , pw , . . . ),
v̂ = (. . . , t̂k , . . . , f̂l , . . . , ĥm , . . . , t̂k∨ , . . . , f̂l∨ , . . . , ĥ∨m , . . . ).
The linear change of variables v̂ = v̂(v) can then be represented by a symplectic
(4M × 4M)-matrix Jϕ ,
v̂ = Jϕ v, Jϕ ∈ Sp(2M, R). (7.18)
On the other hand let us note that Hz (ϕ) ⊗ Hh (ϕ) is canonically isomorphic to
L2 (R2M )via
K : ⊗ ψ → , (t, f, h) ≡ f (t)ψ(h). (7.19)
The corresponding representation of the commutation relations (7.17) on the Hilbert
space L2 (R2M ) is obtained by renaming the operators qv , pv as follows
(. . . , qv , . . . , pw , . . . ) ≡ (. . . , tk , . . . , fl , . . . , hm , . . . , tk∨ , . . . , fl∨ , . . . , h∨m , . . . ).
It follows from the Stone–von Neumann uniqueness theorem for the representation of
the commutation relations (7.17) that these two representation must be related by a
unitary transformation. This transformation may be characterized more precisely as
follows.
Lemma 6. a) To each γ ∈ Sp(2M, R) there exists a unitary operator Jγ on L2 (R2M )

such that
Jγ · v · J−1
γ = γ v. (7.20)
The operators Jγ generate a projective unitary representation of Sp(2M, R).
714 Jörg Teschner
b) The operators Jγ can be represented in the form

Jγ = exp iJγ (v) , (7.21)
where Jγ (v) is a quadratic expression in the operators v.
Proof. Part a) is a classical result of I. E. Segal, [45]. Part b) follows easily from the
observation that the quadratic functions of the operators v generate a representation of
the Lie algebra of Sp(2M, R) which satisfies the infinitesimal version of (7.20), see
e.g. [24] for more details.
One may therefore find an operator Jϕ on L2 (R2M ) which represents the transfor-
mation (7.18) in the sense that
Jϕ · v · J−1
ϕ = v̂ (v) = Jϕ v. (7.22)
The sought-for isomorphism Iϕ can finally be constructed as Iϕ = K−1 · J−1
ϕ .
Remark 5. It is worth noting that the definition of Hf (ϕ) depends only on the com-
binatorial structure of the fat graph ϕ, not on the way it is embedded into the Riemann
surface . It follows that the isomorphism Hf (μ.ϕ) Hf (ϕ), μ ∈ MC() is
canonical.
8 Representations of the mapping class groups

The representations of the mapping class group associated to the quantized Teichmüller
spaces will be obtained by means of a very general construction which produces rep-
resentations of the group G of symmetries of a two-dimensional CW complex G out
of representations of the edge path groupoid of G. We shall first describe this con-
struction, before we discuss how to construct representations of the Ptolemy groupoid
on the quantized Teichmüller spaces. The latter will then induce the sought-for rep-
resentation of MC().
8.1 Projective unitary representations of groupoids

Let us recall that a groupoid G is a category such that all morphisms are invertible.
The objects of G will here be denoted by letters U, V , W, .... Anticipating that the
groupoids G we will be interested in are path groupoids of some topological space we
will use the notation [W, V ] ≡ HomG (V , W ). The elements of [W, V ] will also be
called “paths”.
Definition 2. A unitary projective representation of the groupoid G consists of the

following data:
(i) A Hilbert space H(V ) associated to each object V ∈ Ob(G),

(ii) a map u which associates to each path π ∈ [W, V ] in G a unitary operator
u(π) : H(V ) → H (W ),
(iii) a family of maps ζV , V ∈ Ob(G) which associate to each closed path π ∈ [V , V ]
a number ζV (π) ∈ C with |ζ (π )| = 1.
These data are required to satisfy the relations
a) u(π2 π1 ) = u(π2 ) · u(π1 ), b) u(π −1 ) = u† (π ),
c) ζV (π2 π1 ) = ζV (π2 )ζV (π1 ), d) ζV (π −1 ) = (ζV (π ))∗ , (8.1)
e) u(π) = ζV (π) if π ∈ [V , V ], f) u(id) = 1,
where we use the notation ζV (π) also to denote the operator which multiplies each
vector of HV by the number ζV (π).
The groupoids of interest will be the path groupoids G of two-dimensional CW

complexes G. The set of objects is given by the set of vertices G0 , whereas the set of
morphisms coincides with the set of paths in the complex G. Since each path π may
be represented as a chain Eπ,n(π) · · · Eπ,2 Eπ,1 of edges in G1 it is clear that
a projective unitary representation of the path groupoid G of a two-dimensional CW
complex G is characterized completely by specifying the images u(E) for E ∈ G1 .
Existence of the family of maps ζV such that relation e) is fulfilled represents a rather
nontrivial constraint that the operators u(E), E ∈ G1 have to satisfy. Of course it
suffices to satisfy these constraints for the 2-cells π ∈ G2 .
8.2 Representations of symmetries of a groupoid

The group of symmetries G of a two-dimensional CW complex G is the group of all
invertible mappings

G0 V −→ μ.V ∈ G0 ,
μ:
G1 E −→ μ.E ∈ G1 .
There is an associated action on the paths in the complex G,
[W, V ] π → μ.π ∈ [μ.W, μ.V ],
which is such that
μ(π2 π1 ) = μ(π2 ) μ(π1 ), μ(π −1 ) = (μ(π ))−1 . (8.2)
We will assume that we are given a unitary projective representation of G which is
compatible with the symmetry G in the sense that H (V ) is canonically isomorphic
with H(μ.V ), H(V ) H(μ.V ). We are going to show that the given representation
of the groupoid G canonically induces a representation of its group G of symmetries.
716 Jörg Teschner
Let us fix a base point V ∈ Ob(G) and assume having chosen a path πV (μ) ∈
[μ.V , V ] for each μ ∈ G. Let then R(μ) : H(V ) → H (V ) H(μ.V ) be defined
by
RV (μ) = u(πV (μ)). (8.3)
We are going to assume that the paths πμ1 .V (μ2 ) are the translates of πV (μ2 ) under
μ1 , i.e. that πμ1 .V (μ2 ) = μ1 .πV (μ2 ). It follows that
Rμ1 .V (μ2 ) = RV (μ2 ).
The operators RV (μ) satisfy the relations

RV (μ2 ) · RV (μ1 ) = ϑV (μ2 , μ1 ) RV (μ2 μ1 ), (8.4)

ϑV (μ2 , μ1 ) = ζV πV−1 (μ2 μ1 ) πμ1 .V (μ2 ) πV (μ1 ) . (8.5)
We may next observe that the apparent dependence on the base point V ∈ G0 is
inessential. Let V , W ∈ G0 , and let us pick a path πW,V ∈ [W, V ]. For an operator
OV : H (V ) → H(V ) we define
A[W,V ] (OV ) = u(πW,V ) · OV (μ) · u† (πW,V ). (8.6)

It is easy to convince oneself that AW,V OV (μ) does not depend on the choice of a
path πW,V ∈ [W, V ]. We furthermore have

AW,V RV (μ) = RW (μ). (8.7)
It easily follows that ϑV (μ2 , μ1 ) does not depend on V , i.e. ϑV (μ2 , μ1 ) ≡ ϑ(μ2 , μ1 ).
To summarize: The operators RV (μ) generate a projective unitary representation
RV (G) of G on H,
RV (μ2 ) · RV (μ1 ) = ϑ(μ2 , μ1 ) RV (μ2 μ1 ). (8.8)

The operators A[W,V ] express the unitary equivalence of the representations RV asso-
ciated to the different V ∈ G0 , which allows us to regard

R ≡ (RV )V ∈G0 , (AE )E∈G1
as the representation of G canonically associated to the given representation of the

groupoid G.
Remark 6. There is of course some ambiguity in the construction, coming from the
choice of a representative πV (μ) ∈ [μ.V , V ]. However, it is clearly natural to consider
two representations r, r as equivalent if the generators rV (μ) and rV (μ) differ from
each other just by multiplication with a (possibly μ-dependent) central element. The
cocycle ϑ of the representation r will differ from the cocycle ϑ of r by a coboundary.
8.3 The projective representation of the Ptolemy groupoid on K(ϕ)

Following [32] closely we shall define a projective representation of the Ptolemy
groupoid in terms of the following set of unitary operators on K(ϕ):
πi
e−πi(pv +qv ) e−3πi qv ,
2 2
Av ≡ e 3
where v, w ∈ ϕ0 . (8.9)
Tvw ≡ eb (qv + pw − qw )e−2πi pv qw ,
The special function eb (U ) can be defined in the strip |z| < |cb |, cb ≡ i(b +b−1 )/2
by means of the integral representation

i0+∞
1 dw e−2izw
log eb (z) ≡ . (8.10)
4 w sinh(bw) sinh(b−1 w)
i0−∞
We refer to Appendix A for more details on this remarkable special function. These
operators are unitary for (1 − |b|)b = 0. They satisfy the following relations [32]
Tvw Tuw Tuv = Tuv Tvw , (8.11)
Av Tuv Au = Au Tvu Av , (8.12)
Tvu Au Tuv = ζ Au Av Puv , (8.13)
Au3 = id, (8.14)
where ζ = eπicb /3 , cb ≡ 2i (b + b−1 ). The relations (8.11) to (8.14) allow us to define

2
a projective representation of the Ptolemy groupoid as follows.

• Assume that ωuv ∈ [ϕ , ϕ]. To ωuv let us associate the operator
u(ωuv ) ≡ Tuv : K(ϕ) v → Tuv v ∈ K(ϕ ).
• For each fat graph ϕ and vertices u, v ∈ ϕ0 let us define the following operators
Auϕ : K(ϕ) v → Au v ∈ K(ρu ϕ),
Pϕuv : K(ϕ) v → Puv v ∈ K((uv) ϕ).
It follows immediately from (8.11)–(8.14) that the operators Tuv , Au and Puv can
be used to generate a unitary projective representation of the Ptolemy groupoid in
K(ϕ) L2 (R2M ).
8.4 Reduction to the quantized Teichmüller spaces

Theorem 4. The isomorphism Iϕ maps the operators u(π ) which represent the Ptolemy
groupoid on K(ϕ) to operators of the form u (π ) = Vz (π ) ⊗ Vh (π ), where Vz (π ) ≡
(Vf (π ))f∈B () is a family of unitary operators Vf (π ) on Hf (ϕ).
For each fixed f ∈ B () one may use the operators Vf (π ) to generate a unitary
projective representation of the Ptolemy groupoid.
718 Jörg Teschner
Proof. To begin with, let us note that each path π ∈ ρ ≡ [ϕ , ϕ] canonically defines
a map c → c ≡ A(c) for each c ∈ H1 (, R). This map is defined in an obvious
way for the elementary moves depicted in Figures 6, 7 if we require that c coincides
with c outside the triangles depicted in these figures.
Lemma 7. a) The map c → A(c) preserves the symplectic (intersection) form on

H1 (, R).
b) The operator u(π) maps u(π) · hϕ,c · (u(π ))−1 = hϕ ,A(c) .
Proof. Direct verifications.
Proposition 5. For each path π ∈ [ϕ , ϕ] there exists an operator H(π ) on K(ϕ)

such that the operators V(π) on L2 (R2M ) defined by
V(π) = H(π) · U(π), U(π ) ≡ J−1
ϕ · u(π ) · Jϕ ,
(i) commute with all operators hm , h∨m , m = 1, . . . , g and fl , l = 1, . . . , s − 1,

(ii) generate a unitary projective representation of Pt() on L2 (R2M ).
The operators H(π) can be represented in the form
H(π) ≡ exp(iHπ (hϕ )), (8.15)
∨ ∨
where Hπ (hϕ ) is a quadratic function of the 2g operators h1 , . . . , hg , h1 , . . . , hg
on K(ϕ ).
Proof. The existence of operators H(π) of the form (8.15) which are such that state-
ment (i) of the proposition is verified follows directly from Lemma 6 if one takes
into account that the transformation hϕ ,c → hϕ ,c is represented by an element of
Sp(g, R) according to part a) of Lemma 7.
In order to prove statement (ii) of the proposition, we mainly need to check that
the operators V(π) satisfy the relations of the Ptolemy groupoid. Let us consider a
closed path π ∈ [ϕ, ϕ] which decomposes into a chain of edges as π = πn · · · π1 .
V(π) ≡ V(πn ) . . . V(π1 )
On the one hand one may observe that V(π ) can be factorized as
V(π) = H(π) · U(πn ) . . . U(π1 ) = H(π )ζϕ (π ). (8.16)
The operator H(π) in (8.16) can be represented as follows:

H(π ) = H(πn ) · U(πn ) · H(πn−1 ) · U(πn )† . . . U(π1 ) · H(π1 ) · U(π1 )† ,

where πn−k ≡ πn · · · πn−k . It follows from Lemma 7 together with (8.15) that
U(πj+1 ) · H(πj ) · U(πj+1 )† = exp(iHπj (hϕ )).
Taking into account equation (7.15) we conclude that
K−1 · H(π) · K = 1 ⊗ Hh (π ). (8.17)
On the other hand let us note that V(π ) commutes with all operators hm , h∨m ,
m = 1, . . . , g. Equation (8.16) implies that the same is true for H(π ). However,
the representation of the operators hm , h∨m , m = 1, . . . , g on Hh (ϕ) is irreducible
(see Remark 3). This allows us to conclude that H(π ) = ηϕ (π ) ∈ C, |ηϕ (π )| = 1.
Inserting this into (8.16) proves our claim.
It follows from statement (i) in the proposition that

K−1 · V(π) · K = Vz (π ) ⊗ 1. (8.18)
The task remains to describe the operators Vz (π ) more precisely.
Proposition 6. There exists a family of unitary operators Vf (π ) : Hf (ϕ) → Hf (ϕ ),

f ∈ B () which represent the action of the operator Vz (π ) : Hz (ϕ) → Hz (ϕ ).
Proof. To begin with, let us observe that it follows from (8.9) that the operators
u(E) associated to the edges E = ρv , E = ωvw in Pt 1 () can all be factorized as
u(E) = Q(E)·G(ẑe ), where Q(E) is of the form Q(E) = exp(iQE (v)) for a quadratic
function QE , and G ≡ 1 if E = ρv and G(z) = eb (z) if E = ωvw . It follows that the
corresponding operator V(E) defined in Proposition 5 can be factorized as
V(E) = Q (E) · G(ze ), where Q (E) ≡ H(E) · J−1
ϕ · Q(E) · Jϕ . (8.19)
The operator Q (E) is a product of four operators Jγk , γk ∈ Sp(2M, R) for k =
1, 2, 3, 4. If follows from Lemma 6 that it can be represented in the form

Q (E) = exp iJE (v ) ,
for some expression JE (v ) which is quadratic in v . Note that the operators G(ze ) and
V(E) commute with all operators hm , h∨m , m = 1, . . . , g and fl , l = 1, . . . , s − 1. It
follows that the same is true for Q (E), which implies that JE (v ) ≡ JE (z) depends only
on the vector z ≡ (ze )e∈ϕ1 . V(E) is therefore
of the form
V (E) = exp iJE ( z ) ·G(ze ).
Our claim follows easily, Wf (E) = exp iJE (zf,e ) · G zf,e ) does the job.
Theorem 4 follows by combining Propositions 5 and 6.
We are finally in the position to define more precisely what we will regard as the
quantized Teichmüller spaces. To this aim let us note that the Hilbert spaces H0 (ϕ)
associated to the origin 0 in B () form irreducible representations of the relations
(7.5), (7.6). Functions of the operators z0,e generate the algebras B(H0 (ϕ)) of all
bounded operators on H0 (ϕ), which suggests to interpret B(H0 (ϕ)) as particular rep-
resentations of the quantized algebras Th̄ () of functions on the Teichmüller spaces.
The operators V0 (π) generate a unitary projective representation of the Ptolemy
groupoid which allows us to regard two operators Oϕ2 ∈ B(H0 (ϕ2 )) and Oϕ1 ∈
B(H0 (ϕ1 )) as equivalent, Oϕ2 ∼ Oϕ1 , iff
O2 = V0 (π) · O1 · (V0 (π ))−1 , π ∈ [ϕ2 , ϕ1 ]. (8.20)
720 Jörg Teschner
Let furthermore MCϕ be the unitary projective representation of the mapping class
group MC() which is generated from the operators V0 (π ) by means of the construc-
tion in Section 8.2.
Definition 3. (i) We define the algebra Th̄ () as the algebra generated by the families
O ≡ (Oϕ )ϕ∈P t0 () of bounded operators Oϕ on H0 (ϕ) such that Oϕ2 ∼ Oϕ1 for all
ϕ2 , ϕ1 ∈ P t0 (). The algebra Th̄ () will be called the quantized algebra of functions
on the Teichmüller spaces.
(ii) Let MCh̄ () be the subalgebra of Th̄ () generated by the families MC(μ) ≡
(MCϕ (μ))ϕ∈P t0 () for all μ ∈ MC().
Part III. A stable modular functor from the quantum

Teichmüller spaces
Let us recall that systems of Fenchel–Nielsen coordinates are naturally associated to

markings of the surfaces . The transformations between the different markings of a
Riemann surface generate yet another groupoid, called the stable modular groupoid.
Given that the quantum version of the changes between the Penner coordinates asso-
ciated to different fat graphs was represented by a unitary projective representation
of the Ptolemy groupoid it is natural to expect that the quantization of the Fenchel–
Nielsen coordinates should similarly come with a unitary projective representation of
the modular groupoid.
Pants decompositions have one big advantage over ideal triangulations: The glu-
ing operation allows us to build “larger” surfaces from simple pieces of the same
type, namely hyperbolic surfaces with holes. It is natural to require that the unitary
projective representations of the modular groupoid assigned to surfaces should cor-
respondingly be organized in a “tower-like” fashion: They should allow restriction to,
and should be generated by the representations assigned to the surfaces which are
obtained from by cutting along simple closed curves on .
Our aim in the rest of this chapter will be to show that such a structure can be
constructed from the quantized Teichmüller spaces considered in the previous part.
This is of great importance since
having a tower of projective unitary representations of the stable modular
groupoid is equivalent to having a stable unitary modular functor.
The notion of a stable unitary modular functor will be introduced in the next section.
One main difference to the more conventional (two-dimensional) modular functors
as defined e.g. in [52], [6] is that one restricts attention to Riemann surfaces of
genus g and with n parametrized boundary components which are stable in the sense
that 2g − 2 + n > 0.
We will then explain why having a stable unitary modular functor is equivalent to
having a tower of projective unitary representations of the modular groupoid before we
take up the task to actually construct the latter from the quantization of the Teichmüller
spaces as described previously.
9 The notion of a stable unitary modular functor

Given that the usual definitions of a modular functor take several pages to fully write
them down [52], [6], we shall only briefly explain the most important features. The
missing details will not differ much from the definitions discussed in [52], [6].
9.1 Rigged Riemann surfaces

We will consider compact oriented surfaces with boundary ∂ = β∈A() bβ ,
where A() ≡ π0 (∂) is the set of connected components of ∂. A surface is
called an extended surface if one has chosen orientation-preserving homeomorphisms
pβ : bβ → S 1 for each connected component bβ of the boundary. To be concrete, let
S 1 = {z ∈ C; |z| = 1}. An e-surface of genus g and with n boundary circles is
called stable if 2g − 2 + n > 0.
We will use the terminology rigged Riemann surface, or r-surface for short, for
triples ˆ = (, y, c), where
is a stable extended surface,
y is a Lagrangian subspace of H1 (, R), and
c : A() → L is a coloring of the boundary of by elements of a set L.

ˆ and given β, β ∈ A() such that c(β) = c(β ) we can define
Given an r-surface
ˆ ≡ ββ
a new r-surface ˆ = ( , y , c ) which is obtained from
ˆ by gluing the
boundary components bβ and bβ .
The surface ≡ ββ is defined by identifying all points p ∈ bβ with (pβ−1

pβo )(p) ∈ bβ , where pβo : bβ → S 1 is defined by pβo (p) = −pβ (p). There is
a corresponding projection Pββ : → ≡ ββ which maps bβ , bβ to the
same simple closed curve on .
The Lagrangian subspace y is given by the image of H1 (, R) under the pro-
jection Pββ .
The coloring c is obtained from c by putting c (α ) = c(α) if Pββ (bα ) = bα ∈

∂ .
722 Jörg Teschner
9.2 Stable unitary modular functors

Let L now be a space with a measure dν. A stable modular functor with central charge
is the following collection of data.
Assignment ˆ = (, y, c) → H(, y, c), where
ˆ is an r-surface,
–
– H(, y, c) is a separable Hilbert space.
Mapping class group: Assignment

[f ] −→ U[f ] : H(, y, c) → H ( , yf , cf ) ,
– [f ] is the isotopy class of a homeomorphism f : → ,

– U[f ] is a unitary operator,
– yf is the Lagrangian subspace of H1 ( , R) determined from y via f ,
– cf : A( ) → L: the coloring of boundary components of induced
from c via f .
Disjoint union: There exist unitary operators

∼
G21 : H(2 1 , y2 ⊕ y1 , c2 c1 ) −−→ H (2 , y2 , c2 ) ⊗ H (1 , y1 , c1 ).
Gluing: Let ( , y ) be obtained from (, y) by gluing of two boundary com-

ponents α, β. There then exists a unitary operator

∼
Gαβ : dν(s) H , y, csαβ −− → H ( , y , c ),
L
where the coloring csαβ : A() → L is defined from c via
csαβ (c) = c (Pαβ (c)) if c ∈ A() \ {α, β},
csαβ (c) = s if c ∈ {α, β}.
These data are required to satisfy the following “obvious” consistency and com-
patibility conditions.
Multiplicativity. For all homeomorphisms f : 1 → 2 , g : 2 → 3 there exists
ζ (f, g) ∈ S 1 such thay we have
U[f ] U[g] = ζ (f, g)U[f g] . (9.1)
ζ (f, g) has to satisfy the condition ζ (f, g)ζ (f g, h) = ζ (g, h)ζ (f, g h).
Functoriality. The gluing isomorphisms and the disjoint union isomorphisms are
functorial in .7
7 One is considering the category with objects r-surfaces, and morphisms isotopy classes of homeomorphisms
of r-surfaces, equipped additionally with the gluing and disjoint union operations.
Compatibility. The gluing isomorphisms and the disjoint union isomorphisms are
mutually compatible.
Symmetry of gluing. Gαβ = Gβα .
It would take us several pages to write out all conditions in full detail, we therefore
refer to [52], [6] for more details. However, it seems that the following two “naturality”
requirements represent a key to the understanding of the notion of the modular functor:
Naturality. a) Let f1 : 1 → 1 , f2 : 2 → 2 be r-homeomorphisms. We then
have
G21 · U[f2 f1 ] · G†21 ≡ U[f2 ] ⊗ U[f1 ] .
b) If the r-homeomorphism f : 1 → 2 induces an r-homeomorphism f : 1 →
2 of the surfaces 1 , 2 obtained from 1 , 2 by the gluing construction, we have

Gαβ · dν(s) U[f ] , y, csαβ · G†αβ = U[f ] ( , y , c .
L
These requirements make clear how the mapping class group representations on the
spaces H( , y , c ) restrict to and are generated by the representations assigned to the
surfaces which are obtained from by cutting along simple closed curves on .
Remark 7. The standard definitions of modular functors assume that the Hilbert
spaces H(, y, c) are finite-dimensional. They are therefore not suitable for nonra-
tional conformal field theories. Our definition should be seen as a first step towards
the definition of analogs of modular functors which are associated to nonrational con-
formal field theories in a way similar to the connections between rational conformal
field theories and modular functors mentioned in the introduction.
However, there is one important ingredient of the usual definition that does not have
an obvious counterpart in our framework. In the more standard definitions of modular
functors it is required that there exists a distinguished element s0 in L which has
the property that coloring a boundary component β with s0 is equivalent to “closing”
this boundary component. More precisely, let csβ be a coloring of the boundary
components of an extended surface which assigns s ∈ L to the component with
label β, and let ˆ = ( , y , c ) be the r-surface obtained from ˆ = (, y, c) by
β̌ β̌ β̌ β̌
gluing a disc to bβ . The more standard definitions of modular functors assume or
imply existence of an element s0 in L such that
H (, y, cs0 β ) H(β̌ , yβ̌ , cβ̌ ), U[f ] (, y, cs0 β ) U[f ] (β̌ , yβ̌ , cβ̌ ).
This yields additional relations between the mapping class group representations as-
signed to surfaces with different numbers of boundary components.
In the case of the quantized Teichmüller spaces it ultimately turns out that an
analog of the element s0 in L exists only if one considers the analytic continuation of
H (, y, c) and U[f ] (, y, c) with respect to the boundary labels c(β) ∈ L. This fact,
and the corresponding improvement of our definition of a stable modular functor will
be elaborated upon elsewhere.
724 Jörg Teschner
9.3 Representations of the modular groupoid versus modular functors

It turns out to be very useful to reformulate the notion of a stable unitary modular
functor in terms of generators
and relations as follows. Any surface can be glued
from a surface 0 = p∈σ0 Tp which is a disjoint union of trinions. The different ways
of doing this can be parametrized8 by markings σ . The gluing construction determines
a canonical Lagrangian subspace yσ of H1 ( , R) from the tautological Lagrangian
subspace y0 ≡ H1 (0 , R). By iterating the gluing and disjoint union isomorphisms
one defines unitary operators G(σ, c ) : H ( , yσ , c ) → H(σ, c ), where

H(σ, c ) ≡ dνσ (S) H (Tp , cSσ,p ). (9.2)
L p∈σ0
We have used the following notation:

• The integration is extended over the set L of all colorings S : Cσ c → sc ∈
L equipped with the canonical product measure dνσ (S) obtained from dν by
choosing any numbering of the elements of Cσ .
• cSσ,p is the coloring of the boundary components of Tp which is defined by as-
signing
c (β ) if Pσ,p (β) = β ∈ A( ),
cσ,p (β) =
S
sc if Pσ,p (β) = c ∈ Cσ .
Pσ,p (β) is the embedding Tp → defined from σ by the gluing construction.
Unitary mappings between the different spaces H(σ, c) arise in two ways: First,
one may have different markings σ2 , σ1 such that the Lagrangian subspaces defined
by the gluing construction coincide, yσ1 = yσ2 . In this case one gets unitary operators
Fσ2 σ1 (c) : H(σ1 , c) → H(σ2 , c) from the composition
Fσ2 σ1 (c) ≡ G(σ2 , c) · (G(σ1 , c))† . (9.3)
Secondly,
one has the mappings U[f ] which may map between spaces H (, yσ , c) and
H , (yσ )f , cf . In the case that μ : → represents an element of the mapping
class group it is natural to define operators Vσ μ (c) : H(σ, c) → H (σμ , cμ ) by
Vσ μ (c) = G(σμ , cμ ) · U[μ] · (G(σ, c))† , (9.4)
where σμ is the image of the marking σ under μ.
It turns out – as will be reviewed in the next section – that there exists a set M1 ()
of elementary moves between the different markings σ such that any two markings
σ2 , σ1 can be connected by paths π which are composed out of the elementary moves.
There furthermore exists a set M2 () of relations which makes the resulting two-
dimensional CW complex M() simply connected. The corresponding path groupoid
is called the modular groupoid and denoted by M().
8 Note that markings with the same cut system will yield equivalent representations for . This redundancy
will give a useful book-keeping device when the representation of the mapping class group is considered.
Our construction of the operators Fσ2 σ1 (c), Vσ μ (c) assigns unitary operators to
each of the elementary moves . This yields unitary operators U(π, c) : H(σ1 , c) →
H(σ2 , cπ ) for each path π ∈ [σ2 , σ1 ], where cπ is the coloring of boundary compo-
nents which is obtained from c by tracking the relabeling of boundary components
defined by π . Due to the fact that multiplicativity holds only projectively, cf. eqn.
(9.1), one will find that there exists ζ (π ) ∈ S 1 such that U(π ) = ζ (π ) for all closed
paths, starting and ending at the same marking σ . In other words, a stable unitary
modular functor canonically defines projective unitary representations of the stable
modular groupoid.
Conversely, the definition of a tower of projective unitary representations of the
modular groupoids M() involves the following data.
Assignment (σ, c) → H (σ, c), where H (σ, c) is constructed out of spaces
H(S3 , c3 ) assigned to trinions as in (9.2).
Assignment

π ∈ [σ2 , σ1 ] −→ U(π, c) : H(σ1 , c) → H (σ2 , cπ ) .
It is required that the operators U(π, c) generate unitary projective representations

of the modular groupoids associated to the surfaces . In order to formulate the
additional requirements which turn this collection of representations of M() into a
tower let us note that markings can be glued in a natural way. We will use the notation
αβ σ for the marking obtained by gluing boundary components α and β. Gluing and
disjoint union now become realized in a trivial manner,
H (σ2 σ1 , c2 c1 ) = H (σ2 , c2 ) ⊗ H (σ1 , c1 ), (9.5)

H(αβ σ, c ) = dν(s) H σ, csαβ . (9.6)
L
The data specified above are then required to satisfy the following naturality conditions.
Naturality. a) Let π = π2 π1 ∈ [σ2 , σ2 ] [σ1 , σ1 ]. We then have
U(π2 π1 , c2 c1 ) ≡ U(π2 , c2 ) ⊗ U(π1 , c1 ). (9.7)
b) Given a path π ∈ [σ2 , σ1 ] in M() let αβ π ∈ [αβ σ2 , αβ σ1 ] be the corre-
sponding path in M(αβ ) defined by the gluing construction. We then have

U(αβ π, c ) = dν(s) U(π, csαβ . (9.8)
L
From a tower of projective unitary representations of the modular groupoids one can
reconstruct a stable unitary modular functor as follows. The system of isomorphisms
U(π, c) : H(σ1 , c) → H(σ2 , c) allows us to identify the Hilbert spaces associated to
different markings σ , thereby defining H(, c). The representation of the modular
groupoid defines a representation of the mapping class group on H (, c) via the
construction in Section 8.1. We refer to [6] for more details.
726 Jörg Teschner
It is important to note (see next section for a detailed discussion) that the elements
of the set M1 () of elementary moves only change the markings within subsurfaces of
genus zero with three or four holes, or within subsurfaces of genus one with one hole.
The corresponding operators will be called Moore–Seiberg data. They characterize
a tower of representations of the modular groupoid completely. Let us furthermore
note that the faces/relations that one needs to define a two-dimensional CW complex
M() which has as set of edges M1 () turn out to involve only subsurfaces of genus
zero with three to five holes, and subsurfaces of genus one with one or two holes. This
means that one only needs to verify a finite number of relations to show that a given
set of Moore–Seiberg data defines a stable unitary modular functor.
10 The modular groupoid

The modular groupoid is the groupoid generated from the natural transformations
relating the different markings of a Riemann surface [38], [5]. A complete set of
generators and relations has been determined in [38], [5], [20].
Remark 8. The CW complex M() defined below will be a subcomplex of the

complex denoted Mmax () in [5], since that reference allowed for cut systems that
yield connected components with less than three boundary components, whereas we
will exclusively consider cut systems that yield connected components with exactly
three holes. All other deviations from [5] are due to slightly different conventions in
the definition of the generators.
10.1 Notations and conventions

We have a unique curve c(σ, e) ∈ C associated to each edge e ∈ σ1 of the marking
graph σ . The trinions Tp ∈ Pσ are in one-to-one correspondence with the vertices
p ∈ σ0 of σ .
The choice of a distinguished boundary component cp for each trinion Tp , p ∈ σ0
will be called the decoration of the marking graph σ . The distinguished boundary
component cp will be called outgoing, the other two boundary components of Tp
incoming. Two useful graphical representation for the decoration are depicted in
Figure 10.
10.2 Generators
The set of edges M1 () will be given by elementary moves denoted as (pq), Zp ,
Bp , Fpq and Sp . The indices p, q ∈ σ0 specify the relevant trinions within the pants
decomposition of that is determined by σ . The move (pq) will simply be the
operation in which the labels p and q get exchanged. Graphical representations for
the elementary moves Zp , Bp , Fpq and Sp are given in Figures 11–14.
3 3
p p
*
1 2 1 2
Figure 10. Two representations for the decoration on a marking graph.
3 3
Zp
p p *
*
2 1 2 1
Figure 11. The Z-move.
3 3
Bp
p
* *
2 1 2 1
Figure 12. The B-move.
q p
* Fpq *
p q
* *
3 2 1 3 2 1
Figure 13. The F -move.

728 Jörg Teschner
Sp *
p
*
c p
Figure 14. The S-move.
10.3 Relations
The relations of M() correspond to the faces ∈ M2 (). In the following we

will define a set R of faces which is large enough to make the complex M() simply
connected. A face in M2 () may be characterized by choosing a chain =
E,n( ) · · · E,1 , where E,i ∈ M1 () for i = 1, . . . , n( ). In order to
simplify notation we will generically factorize as = 2 1 and write

2 )−1 = 1 instead of = id.
(
Locality. Let us introduce the notation supp(m) by supp(m) = {p} if m = Zp , Bp ,

Sp , supp(m) = {p, q} if m = (pq), Fpq and supp(m2 m1 ) = supp(m2 ) ∪ supp(m1 ).
We then have
m2 m1 = m1 m2 whenever supp(m1 ) ∩ supp(m2 ) = ∅. (10.1)
We will list the remaining relations ordered by the topological type of the surfaces on
which the relevant graphs can be drawn.
Relations supported on surfaces of genus zero.

g = 0, s = 3: Zp Zp Zp = id. (10.2)
a) Fqp Bp Fpq = (pq) Bq Fpq Bp ,

g = 0, s = 4: b) Fqp Bp−1 Fpq = (pq) Bq−1 Fpq Bp−1 , (10.3)
c) Apq Aqp = (pq).
g = 0, s = 5: Fqr Fpr Fpq = Fpq Fqr . (10.4)

We have used the abbreviation
Apq ≡ Zq−1 Fpq Zq−1 Zp . (10.5)
In Figures 15 and 16 we have given diagrammatic representations for relations (10.3),
b) and (10.4) respectively.
3 2 1
*
q p
*
q * p Bp−1 Fpq q * p
Fpq Bp−1 (pq)
3 2 1 Bp−1 Fqp
p
q
* * q *
* p * p * q
Figure 15. The hexagon relation.
r Fqr q
*
q *
* r p Fpq
p
* * *
p
Fpq q *
r *
*
Fpr p
r
* p r *
q * * q Fqr
* *
Figure 16. The pentagon identity.
Relations supported on surfaces of genus one. In order to write the relations

transparently let us introduce the following composites of the elementary moves.
a) Bp ≡ Zp−1 Bp Zp−1 ,

g = 0, s = 3: (10.6)
b) Tp ≡ Zp−1 Bp Zp Bp ,
g = 0, s = 4: Bqp ≡ Zq−1 Fqp

−1
Bq Fpq
−1
Zq−1 (pq), (10.7)
g = 1, s = 2: Sqp ≡ (Fqp Zq )−1 Sp (Fqp Zq ). (10.8)
It is useful to observe that the move Tp represents the Dehn twist around the boundary
component of the trinion tp numbered by i = 1 in Figure 10. With the help of these
730 Jörg Teschner
definitions we may write the relations supported on surfaces of genus one as follows:
a) Sp2 = Bp ,
g = 1, s = 1: (10.9)
b) Sp Tp Sp = Tp−1 Sp Tp−1 .
−1
g = 1, s = 2: Bqp = Sqp Tq−1 Tp Spq . (10.10)
Relation (10.10) is represented diagrammatically in Figure 17.
Bqp q
* * *
p q *
p
Spq −1
Sqp
q *
p*
Tq−1 Tp p c3
b1 * *q b2
Figure 17. Relation for the two-punctured torus.
Theorem 5. The complex M() is connected and simply connected for any e-sur-
face .
Proof. The theorem follows easily from [5], Theorem 5.1. We noted previously that
our complex M() differs from the complex Mmax () of [5] in having a set of vertices
which corresponds to decompositions of into connected components with exactly
three holes. The edges of M1max () which correspond to the F-move of [5] simply can’t
appear in M1 (). Otherwise the set of edges of M() coincides with the relevant
subset of M1max (), with the exception that our move Bp is a composition of the B-
and the Z-move of [5]. To complete the proof it remains to check that our set of faces
is equivalent to the subset of M2max () which involves only the vertices M0 () of our
smaller complex M(). This is a useful exercise.
Definition 4. Let M () be the complex which has the same set of vertices as M(),
a set of edges given by the moves
(pq), Zp , Bp , Bp , Sp , Tp , Fpq , Apq , Bpq , Spq (10.11)
defined above, as well as faces given by equations (10.1)–(10.10).
11 From markings to fat graphs

A key step in our construction of a stable modular functor from the quantized Teich-
müller spaces will be the definition of a distinguished class of fat graphs ϕσ which
are associated to the elements σ of a certain subset of the set M0 () of all markings
of .
Definition 5. Let Aσ ⊂ Cσ be the set of all curves c which are incoming for both
adjacent trinions. We will say that a marking σ is admissible iff there is no curve
c ∈ Cσ which is outgoing for both adjacent trinions, and if cutting along all curves
c ∈ Aσ yields connected components all of which have genus zero. The set of all
admissible markings will be denoted by M0ad ().
To each admissible marking σ we may naturally associate a fat graph ϕσ on by

the following construction. In order to construct the fat graph ϕσ , it will be useful
to consider a certain refinement of the pants decomposition associated to σ which is
defined as follows. For each curve c ∈ Aσ let Ac be a small annular neighborhood of
c which contains c in its interior, and which is bounded by a pair of curves (c+ , c− )
that are isotopic to c. If c ∈ A() represents a boundary component of we may
similarly consider an annular subset Bc of bounded by c+ ≡ c and another curve
c− isotopic to c. If c ∈ Cσ belongs to neither of these two classes we will simply
set c+ ≡ c ≡ c− . By cutting along all such curves c± we obtain a decomposition of
into trinions Tp , p ∈ σ0 and annuli Ac , c ∈ Aσ , each equipped with a marking
graph. We may then replace the markings on each of these connected components
by fat graphs according to Figures 18–20. The re-gluing of trinions Tp , p ∈ σ0 and
annuli Ac , Bc to recover the surface may then be performed in such a way that the
fat graphs on the connected components glue to a fat graph ϕσ on .
11.1 The complex Mad ()

It is natural to consider the complex Mad () for which the set of vertices M0ad ()
is the subset of M0 () which consists of the admissible markings, and which has a
set of edges M1ad () given by the subset of M1 () that contains those edges which
connect two admissible markings.
Proposition 7. The complex M ad () is connected and simply connected.
Proof. Let us consider two admissible markings σ, σ ∈ M0 (). There exists a path
in M() which connects σ and σ . This path may be represented as a chain
C = E,n( ) · · · E,1 , E,i ∈ Mi () composed out of the moves Zp , Fpq , Bp
and Sp .
For a given marking σ ∈ M0 () let [σ ] be the set of all markings σ which differ
from σ only in the choice of decoration. The moves Zp act transitively on [σ ]. By
732 Jörg Teschner
Figure 18. Substitution for a trinion T .
*
*
c c
Figure 19. Substitution for an annulus A.
*
*
b b
Figure 20. Substitution for a boundary component B.
inserting Zp moves if necessary we may therefore modify C to a chain D which

takes the form
D = Z,n( ) F,n( ) Z,n( )−1 · · · F,1 Z,0 ,
where Z,i , i = 0, . . . , n( ) are chains composed out of Zp -moves only, and the
moves F,n( ) ∈ M1 () connect markings σ̃,i and σ,i , i = 1, . . . , n( ) which
are admissible. We clearly must have [σ,i+1 ] = [σ̃,i ], i = 1, . . . , n( ) − 1 and
[σ,1 ] = [σ ], [σ,n( ) ] = [σ ]. Connectedness of M ad () would follow if the
chains Z,i , i = 0, . . . , n( ) are homotopic to chains Y,i which represent paths in
M ad (). That this indeed the case follows from the following lemma.
Lemma 8. Assume that σa , σb ∈ M0ad () satisfy σb ∈ [σa ]. There then exists a path
ab in Mad () which connects σa and σb .
Sketch of proof. Let us recall that a marking σ is irreducible if it is admissible and if

there are no edges e in Aσ such that cutting along c(e, σ ) yields two disconnected
components. A marking σ which is irreducible has only one outgoing external edge.
With the help of Lemma 17 in Appendix D it is easy to show that for an admissible
graph σ there always exists a sequence of Zp -moves in M1ad () which transforms σ
to a graph σ that is irreducible.
We may and will therefore assume that σa and σb are both irreducible. Using
Lemma 17 again allows us to transform σb to a marking σc which is such that the
outgoing external edges of σa and σc correspond to the same outgoing boundary
component of .
The graphs σa and σc can finally be connected by a chain which is composed out
of moves Bp and Fqp it only. This follows from the connectedness of M(0 ) in the
case of a surface 0 of genus zero [38]. In this way one constructs a sequence of
moves that connects σa to σb
It remains to prove that M ad () is simply connected as well. Let us consider any
closed path in M ad (). The path is contractible in M (). The deformation of
to a trivial path may be performed recursively, face by face. The crucial observation
to be made is the following one.
Lemma 9. The paths which represent the boundaries of the faces of M2 () are
paths in M ad ().
Proof. By direct inspection of the relations (10.2)–(10.10).
Lemma 9 implies that deforming a path ∈ M ad () by contracting a face in
M2 () will produce a path which still represents a path in Mad (). It follows
that is contractible in M ad ().
11.2 Separated variables

We had observed in Section 6 that the Fock variables ze , e ∈ ϕ1 form a set of coordi-
nates for the Teichmüller spaces of surfaces with holes. When considering fat graphs
ϕσ associated to a marking σ it will be useful to replace the Fock variables ze , e ∈ ϕ1
by an alternative set of coordinates (qc , pc ), c ∈ Cσ for Tϕ which satisfy
ϕ (pc1 , pc2 ) = 0 = ϕ (qc1 , qc2 ),
c1 , c2 ∈ Cσ . (11.1)
ϕ (pc1 , qc2 ) = δc1 ,c2 ,
These coordinates are constructed as follows.
For c ∈ Aσ let Ac be an annular neighborhood of c such that the part of ϕσ which
is contained in Ac is isotopic to the model depicted in Figure 21.
Let e1 and e2 be the two edges that are entirely contained in Ac with labelling
defined by Figure 21. Out of ze1 and ze2 we may then define
qc ≡ 21 (ze1 − ze2 ), pc ≡ − 21 (ze1 + ze2 ). (11.2)
In the case c ∈ Cσ \ Aσ let us note that there is a unique trinion Tc for which
the curve c is the outgoing boundary component. The part of ϕσ contained in Tc is
depicted in Figure 22.
734 Jörg Teschner
1
2
A
Figure 21. Annulus Ac and fat graph ϕ on A.
ec
vc
ec,2 *
ec,1
c2 c1
Figure 22. Part of ϕσ contained in Tc .
Let ec,2 and ec,1 be the edges of ϕσ as indicated in Figure 22, and let cε , ε = 1, 2
be the curves which represent the corresponding boundary components of Tp . We
shall then define
qc ≡ yc2 , pc ≡ −yc1 , (11.3)
where
1
ycε ≡ zec,ε + fcε , ε = 1, 2. (11.4)
2
The same construction yields linear combinations (p̂c , q̂c ), c ∈ Cσ of the Kashaev
variables which represent the coordinates (pc , qc ) within Wϕσ . Note that our construc-
tion of the fat graph ϕσ implies that Tc contains a unique vertex vc ∈ ϕσ,0 . It turns
out that the variables q̂c , p̂c have a simple relation to the Kashaev variables qvc , pvc .
Lemma 10. (i) q̂c = qvc + hc2 , p̂c = pvc − hc1 .

(ii) ϕ (pc , qc ) = δc,c , c, c ∈ Cσ .
Proof. In order to prove part (i) let us first look at the vertices v ∈ ϕσ,0 which appear
on the graph geodesic homotopic to c. The contribution to ycε of the edges that
are incident to v is ẑe1 + ẑe2 + ẑe3 if ei , i = 1, 2, 3 are the three edges incident
to v. It then follows from (4.6) that yc does not depend on both pv and qv . What
remains are contributions from the vertex vc , as well as contributions from the vertices
contained in annuli Ac which are determined as follows. Note that the homology
class [c] can be decomposed as a linear combination of classes [c ] with c ∈ Aσ . It is
straightforward to check that each c ∈ Aσ which appears in this decomposition yields
a contribution −hc to fˆc . These contributions sum up to give −hc . What remains
is the contribution from the vertex vc . Part (i) of the lemma now follows easily by
recalling the definition (4.6).
Part (ii) of the lemma is a trivial consequence of part (i).
11.3 Quantized Teichmüller spaces for surfaces with holes

The coordinates introduced in the previous section make it straightforward to modify
the discussion of the quantization of Teichmüller spaces from the case of punctured
Riemann surfaces (Section 7) to the case of Riemann surfaces with holes.
Bearing in mind that the Fock variables ze , e ∈ ϕ1 are unconstrained in the present
case (see Lemma 5) leads us to identify the algebra of functions on the Teichmüller
spaces with the algebras of function of the variables ze , e ∈ ϕ1 . A convenient set
of coordinates is given by the coordinate functions (pc , qc ), c ∈ Cσ . Canonical
quantization of the Teichmüller spaces is therefore straightforward, and leads to an
algebra of operators with generators (pc , qc ), c ∈ Cσ , which is irreducibly represented
(in the sense of Remark 3) on the Hilbert space Hz (σ ) L2 (R3g−3+2s ). We will
normalize the operators (pc , qc ), c ∈ Cσ , such that
(i) [ pc , qc ] = (2π i)−1 δcc , c, c ∈ Cσ ,
(11.5)
(ii) [ pc , pc ] = 0 = [ qc , qc ].
Quantization of the Kashaev space Wϕσ produces a reducible representation of
the algebra (11.5) which is generated by operators (q̂c , p̂c ) associated to the pairs
of variables (q̂c , p̂c ). It also yields operators hc , c ∈ H1 (, R) which represent
the quantization of the Poisson vector space Hϕσ with basis hc , c ∈ H1 (, R). The
algebra generated by the hc , c ∈ H1 (, R) has a center generated by the hc , c ∈ B().
Following the discussion at the beginning of Section 7.3 one constructs a representation
of this algebra on the space

Hh (σ ) ≡ df Hh,f (σ ),
B ()
where Hh,f (σ ) L2 (Rg ) is an irreducible representation of the algebra i[hc2 , hc1 ] =

b2 I(c2 , c1 ) for c2 , c1 ∈ H1 (cl , R).
The following Proposition 8 describes how the quantized Teichmüller spaces are
related to the quantized Kashaev space.
Proposition 8. There exists a unitary operator Iσ ,

Iσ : K(ϕσ ) → Hz (σ ) ⊗ Hh (σ )
such that
Iσ · p̂c · I−1
σ = pc ⊗ 1, Iσ · q̂c · I−1 −1
σ = qc ⊗ 1 and Iσ · hc · Iσ = 1 ⊗ hc ,
for any e ∈ σ1 and c ∈ H1 (, R), respectively.

736 Jörg Teschner
Let us recall that a move m ∈ [τm , σm ], m ∈ M1 () is admissible if both τm

and σm are admissible. Given an element m ∈ M1ad () it is natural to consider the
corresponding fat graphs θm ≡ ϕτm , ϕm ≡ ϕσm on , to pick a path πm ∈ [θm , ϕm ]
and consider the operator u(m) ≡ u(πm ). The reduction to the quantized Teichmüller
spaces proceeds as in Section 8.4. By multiplying the operators Iσ · u(m) · I−1
σ with
suitably chosen operators H(m) one gets operators v(m) which factorize as v(m) =
w(m) ⊗ 1. The resulting operators w(m) : Hz (ϕm ) → Hz (θm ) will then generate a
unitary projective representation of the path groupoid of M ad ().
12 Geodesic length operators

Of fundamental importance for us will be to define and study quantum analogs of the
geodesic length functions on the Teichmüller spaces, the geodesic length operators.
12.1 Overview
When trying to define operators which represent the geodesic length functions one
has to face the following difficulty: The classical expression for Lϕ,c ≡ 2 cosh 21 lc as
ze
given by formula 5.6 is a linear combination of monomials in the variables e± 2 of a
very particular form,

Lϕ,c = Cϕ,c (τ ) ex(τ ) , x(τ ) ≡ τ (e) ze , (12.1)
τ ∈F e∈ϕ1
where the summation is taken over the space F of all maps ϕ1 e → τ (e) ∈ 21 Z.
The coefficients Cϕ,c (τ ) are positive integers, and non-vanishing for a finite number
of τ ∈ F only.
In the quantum case one is interested in the definition of length operators Lϕ,c
which should be representable by expressions similar to (12.1),

b
Lϕ,c = Cϕ,c (τ ) ex(τ ) , x(τ ) ≡ τ (e) ze . (12.2)
τ ∈F e∈ϕ1
The following properties seem to be indispensable if one wants to interpret an

operator of the general form (12.2) as the quantum counterpart of the functions Lϕ,c =
2 cosh 21 lc :
(a) Spectrum. Lϕ,c is self-adjoint. The spectrum of Lϕ,c is simple and equal to
[2, ∞). This is necessary and sufficient for the existence of an operator lϕ,c – the
geodesic length operator – such that Lϕ,c = 2 cosh 21 lc .
(b) Commutativity.
Lϕ,c , Lϕ,c = 0 if c ∩ c = ∅.
(c) Mapping class group invariance.

aμ (Lϕ,c ) = Lμ.ϕ,c , aμ ≡ a[μ.ϕ,ϕ] for all μ ∈ MC().
b (τ ) which appear in (12.2) should be de-
(d) Classical limit. The coefficients Cϕ,c
formations of the classical coefficients Cϕ,c (τ ) in the sense that
b
lim Cϕ,c (τ ) = Cϕ,c (τ ).
b→0
Length operators were first defined and studied in the pioneering work [16]. It was
observed in [16] that the necessary deformation of the coefficients Cϕ,c b (τ ) is indeed
nontrivial in general. However, a full proof that the length operators introduced in [16]
fulfil the requirements (a) and (c) does not seem to be available yet. We will therefore
present an alternative approach to this problem, which will allow us to define length
operators that satisfy (a)-(d) in full generality.
12.2 Construction of the length operators

Our construction of the length operators will proceed in two steps. First, we will
construct length operators Lσ,c in the case that the fat graph ϕ under consideration
equals ϕσ . This will facilitate the verification of the properties (a)-(d) formulated
above. In order to define the length operators Lϕ,c in the general case we shall then
simply pick any marking σ such that the given curve c is contained in the cut system
Cσ , and define
Lϕ,c ≡ a[ϕ,ϕσ ] (Lσ,c ). (12.3)
Independence of this construction from the choice of σ will follow from Theorem 6
below. Definition (12.3) reduces the proof of properties (a), (b) to the proof of the
corresponding statements for the length operators Lσ,c which will be given below.
Property (c) follows from a[μ.ϕ,ϕ] a[ϕ,ϕσ ] = a[μ.ϕ,ϕσ ] .
In order to prepare for our construction of length operators it is useful us recall the
construction of the fat graph ϕσ in Section 11.
Definition 6. Case c ∈ Aσ . Let A be an annular neighborhood of c such that the part

of ϕσ which is contained in A is isotopic to the model depicted in Figure 21. We will
then define
Lϕ,c ≡ e−2πbqc + 2 cosh 2π bpc . (12.4)
Case c ∈/ Aσ . If a curve c is not contained in Aσ , it is necessarily the outgoing
boundary component of a trinion Tp (cf. Figure 18). Let cε , ε = 1, 2 be the curves
which represent the incoming boundary components of Tp as indicated in Figure 22.
Given that Lσ ;ci , i = 1, 2 are already defined we will define Lσ ;c by
Lσ,c = 2 cosh(yc2 + yc1 ) + e−yc2 Lσ,c1 + eyc1 Lσ,c2 + eyc1 −yc2 , (12.5)
where ycε , ε = 1, 2 are defined as yc2 = 2πbqc , yc1 = −2π bpc .
738 Jörg Teschner
It is easy to see that this recursively defines length operators for all remaining
c ∈ Cσ .
Remark 9. Let us note that (11.5) implies that [ycε , Lσ,cε ] = 0 for ε, ε ∈ {1, 2}. We
therefore do not have an issue of operator ordering in (12.5).
Proposition 9. The length operators Lσ,c , Lσ,c associated to different curves c, c ∈

Cσ commute with each other.
Proof. Let us recall that cutting the surface along all of the curves c ∈ Aσ , yields a
set of connected components which all have genus zero. For a given curve c ∈ Cσ \Aσ
let c be the connected component which contains c. Cutting c along c produces
two connected components. The component which has c as its outgoing boundary
component will be denoted c . It follows from Definition 6 that Lσ,c is an operator
function of the operators pd and qd , where d ∈ Cσ is contained in c . The claim
therefore follows immediately from (11.5) if c and c are disjoint.
Otherwise we have the situation that one of c , c , say c is a subsurface of the
other. The crucial point to observe is that the resulting expression for Lσ,c depends
on the variables pd and qd associated to the subsurface c exclusively via Lσ,c . The
claim therefore again follows from (11.5).
The following theorem expresses the consistency of our definition with the auto-
morphisms induced by a change of the marking σ .
Theorem 6. For a given curve c let σi , i = 1, 2 be markings such that c is contained

in both cut systems Cσ1 and Cσ2 . We then have
a[ϕσ2 ,ϕσ1 ] (Lσ1 ,c ) = Lσ2 ,c. (12.6)
On the proof of Theorem 6: The description of the modular groupoid in terms of

generators and relations given in Section 10 reduces the proof of Theorem 6 to the case
that σ2 and σ1 are connected by one of the elementary moves m defined in Section 10.2.
In order to reduce the proof of Theorem 6 to a finite number of verifications one would
need to have simple standard choices for the paths πm ∈ [ϕσ2 , ϕσ1 ] for all elementary
moves m. Existence of such standard paths πm turns out to be nontrivial, though. The
task to find suitable paths πm is particularly simple for a subclass of moves m which
is defined as follows.
Definition 7. (i) Let π ∈ [ϕ , ϕ] be a path in the complex P t() which is described

by a sequence Sπ ≡ (ϕ ≡ ϕn , . . . , ϕ1 ≡ ϕ) of fat graphs such that consecutive
elements of Sπ are connected by edges in P t 1 (). We will say that π ∈ [ϕ , ϕ] is
realized locally in a subsurface → if the restrictions of ϕi to \ coincide
for all i = 1, . . . , n.
(ii) We will say that the move m = [τm , σm ] ∈ M1 () can be realized locally if
there exists a path π ∈ [θm , ϕm ] in the complex P t() that is realized locally in m
in the sense of (i).
For moves m = [τm , σm ] ∈ M1 () that can be realized locally we may choose
essentially the same path πm ∈ [θm , ϕm ] for all surfaces into which m can be
embedded. It is then crucial to observe the following fact
Proposition 10. If m ∈ M1 () is admissible, but can not be realized locally, there
always exists a path m which is (i) homotopic to m within M ad (), and (ii) takes
the form
m = Ym m Ym−1 , (12.7)
where Ym is a chain composed out of Zp -moves and Fpq -moves which can all be
realized locally.
The proof of Proposition 10 is given in Appendix D. It therefore suffices to prove

Theorem 6 in the case that σ2 and σ1 are connected by any elementary move m that
can be realized locally. This amounts to a finite number of verifications which can be
carried out by straightforward, but tedious calculations. Some details are given in the
Appendix E.
12.3 Spectrum
Theorem 7. The spectrum of Lσ,c is simple and equal to [2, ∞).
Proof. To begin with, let us consider the following simple model for the length oper-
ators:
L ≡ 2 cosh 2π bp + e−2πbq , (12.8)
where p, q are operators on L2 (R) which satisfy the commutation relations [p, q] =
(2πi)−1 .
A basic fact is that L is self-adjoint. Indeed, being a sum of two positive self-adjoint
operators L is self-adjoint on the intersection of the domains of the summands. The
main spectral properties of this operator are summarized in the following proposition.
Proposition 11 ([33]). (i) We have Spec(L) = (2, ∞).

(ii) The spectrum of L in L2 (R) is simple.
Validity of Theorem 7 in the case c ∈ Aσ is a direct consequence of Proposition 11.

It remains to treat the case c ∈
/ Aσ . We will keep the notations introduced in Defini-
tion 6. The main ingredient will be an operator Cσ,c : H(σ ) → H(σ ) which maps all
st ,
length operators Lσ,c to the simple standard form Lσ,c
st
Lσ,c = 2 cosh 2π bpc + e−2πbqc (12.9)
in the sense that the following commutation relations are satisfied:
Cσ,c · Lσ,c = Lσ,c
st
· Cσ,c . (12.10)
740 Jörg Teschner
Definition 8. Let the unitary operator Cσ,c be defined as

sb (s1 − pc ) 2πi s2 qc
C−1
σ,c = eb (qc − s2 ) e , (12.11)
s b (s 1 + p c )
where
sε = (2π b)−1 arcosh 21 Lσ,cε ε = 1, 2,
and the special function sb (x) is a close relative of eb (x) defined in the Appendix A.
Lemma 11. The unitary operator Cσ,p satisfies (12.10).
Proof. Proposition 11 is proven by means of a direct calculation using the explicit

form of Lσ,p given in (12.5) and the functional equations (A.3) and (A.9).
The proof of Theorem 7 is thereby reduced to Proposition 11.
12.4 Relation with the Dehn twist generator

To round off the picture we shall now discuss, following [32], the relation between
the length operators Lϕ,c and the operator Dϕ,c which represents the Dehn twist Dc
along c.
A closed curve c will be called a curve of simple type if the connected components
of \ c all have more than one boundary component. It is not hard to see (using the
construction in Section 11, for example) that for curves of simple type there always
exists a fat graph ϕ and an annular neighborhood Ac of c in which ϕ takes the form
depicted in Figure 23.
v *
*
w
Figure 23. Annulus Ac and fat graph ϕ on A.
Lemma 12. Let be a surface with genus g and s holes. The pure mapping class
group MC() is generated by the Dehn twists along geodesics of simple type.
Proof. For g ≥ 2 it is known that the Dehn twists along non-separating closed curves
suffice to generate the mapping class group MC() [21]. Closed curves which are
non-separating are always of simple type.
For g = 0 any closed curve is of simple type. In the remaining case g = 1 one
may note that the only closed curve c which is not of simple type is the one which
separates a one-holed torus from the rest of . It is then well-known that the Dehn
twists along a- and b-cycles of the one-holed torus generate the Dehn-twist along c.
If we supplement these generators by the Dehn twists along the remaining closed
curves (which are all of simple type), we get a complete system of generators for
MC() [21].
We may therefore use a fat graph which in an annular neighborhood of c takes the
form depicted in Figure 23. It is easy to see that the action of the Dehn twist Dc on
the fat graph ϕ can be undone by a single flip ωvw . As the representative Fϕ,c for the
Dehn twist Dc on K(ϕ) we may therefore choose Fϕ,c = Tvw .
The operator hϕ,c associated to the homology cycle c is hϕ,c ≡ 21 (pv + qw ). It is
not very difficult to verify that the operator
Dϕ,c = ζ −6 exp(2π i h2ϕ,c ) Fϕ,c . (12.12)
commutes with all hc , c ∈ H1 (, R). The prefactor ζ −6 , ζ = eπib /3 was inserted to
2
define a convenient normalization. We then have the following result.
Proposition 12 ([33]). Dc coincides with the following function of the length opera-
tor lc :

l2c
Dc = ζ −6 exp i .
8π b2
This should be compared with the classical result that the geodesic length functions
are the Hamiltonian generators of the Fenchel–Nielsen twist flow, which reproduces
the Dehn twist for a twist angle of 2π .
13 Passage to the length representation

13.1 The length representation
Our aim is to define a representation for H() which is such that the length operators
associated to a cut system are all realized as multiplication operators. For a surface
and a marking σ on let
HL (σ ) ≡ Ksp⊗σ1 , (13.1)
where Ksp L2 (R+ , η) and η is the spectral measure of the operator lA = 2 arcosh 21 LA .
The numbering of the edges associated to σ defines canonical isomorphisms
HL (σ ) HL () ≡ L2 (L, dηL ), (13.2)
3g−3+2s
where L R+ and dηL = dη(l1 ) ∧ · · · ∧ dη(l3g−3+2s ) is the corresponding
product measure. For e ∈ σ1 and f : R+ → C let us define the multiplication operator
me [f ] by
me [f ] (λσ ) ≡ f (le )(λσ ). (13.3)
742 Jörg Teschner
The operator me [f ] will represent a bounded operator me [f ] : HL (σ ) → HL (σ ) iff

f ∈ L∞ (R+ , dη).
For the rest of this section let us consider an admissible marking σ and the associ-
ated fat graph ϕσ . Definition 6 yields a family of mutually commuting length operators
Lσ,c associated to the curves c which constitute the cut system Cσ . It follows from the
spectral theorem9 for the family {Le ; e ∈ σ1 } of self-adjoint operators that there exists
a unitary operator
D(σ ) : Hz (σ ) −→ HL (σ )
that diagonalizes the set of length operators Lσ,e , e ∈ σ1 in the sense that
D(σ ) · Lσ,e = me [2 cosh 21 l] · D(σ ). (13.4)
Our aim for the rest of this section will be to give a recursive construction for such
an operator D(σ ) in terms of operators Cp which are associated to the vertices p ∈ σ0
of σ .
13.2 Construction of the operator D(σ )

The main ingredient will be an operator Cσ : H (σ ) → H(σ ) which maps all length
operators Lσ,c to the simple standard form Lcst ,
st
Lσ,c = e−2πbqc + 2 cosh 2π bpc . (13.5)
The operators Cσ,c that were defined in Definition 8 solve this task locally for every
curve c ∈ Cσ \ Aσ . If the trinion Tp has cp as the outgoing boundary component we
will define
Cσ,p ≡ Cσ,cp . (13.6)
We are now ready to define
Cσ ≡ Cσ,p . (13.7)
p∈σ0
Let us note that we do not have to specify the order in which the operators Cσ,p appear
thanks to the following lemma.
Lemma 13. We have Cσ,p Cσ,q = Cσ,q Cσ,p .
Proof. This follows from Proposition 9 and Definition 6, keeping in mind (11.5).
In order to construct the sought-for operator D(σ ) it now remains to map the
st to multiplication operators. Proposition 11 ensures existence of
length operators Lσ,e
an operator
dσ,e : L2 (R) → Ksp such that dσ,e · Lest = me [2 cosh 21 l] · dσ,e .
9 See Appendix B for the precise statement.
Out of Eσ we may finally define the operator D(σ ) as

D(σ ) ≡ dσ · Cσ , dσ ≡ dσ,e , (13.8)
e∈σ1
where Iσ is the operator introduced in Proposition 4. It is straightforward to verify

that the operator D(σ ) indeed satisfies the desired property (13.4).
14 Realization of M()
14.1 Two constructions for the generators
Our aim is to define operators U(m) associated to the edges m ∈ M1 (). We will
give two constructions for these operators, each of which makes certain properties
manifest. The proof of the equivalence of these two constructions will be the main
difficulty that we will have to deal with.
The first construction. Let us recall that a move m ∈ [τm , σm ], m ∈ M1 () is

admissible if both τm and σm are admissible. Given an element m ∈ M1ad () it is
natural to consider the corresponding fat graphs θm ≡ ϕτm , ϕm ≡ ϕσm on , to pick a
path πm ∈ [θm , ϕm ] and define operators Ũ(m) : HL (σm ) → HL (τm ) as
Ũ(m) ≡ D(τm ) · u(πm ) · D(σm )† . (14.1)
The second construction. We note that for all m ∈ M1 () the markings τm and
σm will coincide outside of a subsurface m → . Let us therefore consider the
restrictions τm and σm of τm and σm to m respectively. Admissibility of τm and σm
is obvious for m ∈ M1 (), allowing us to use the first construction in order to define
an operator
U (m) : HL (σm

) → HL (τm )
Out of U (m) we may then construct the sought-for operator U(m) by acting with
U (m) non-trivially only on those tensor factors of HL (σ ) = Ksp⊗σ1 which correspond
to the subsurface m → . More precisely, let Em ⊂ σ1 be the set of edges in
σ1 which have nontrivial intersection with m . Out of U (m) let us then define the
operator U(m) : K(σm ) → K(τm ) by applying definition (B.5) to the case O ≡ U (m)
and J ≡ Em .
Comparison. The crucial difference between Ũ(m) and U(m) is that the latter is
manifestly acting locally in HL (σ ), in the sense that it acts only on the tensor factor
of HL (σ ) which corresponds to the subsurface m . This is not obvious in the case of
Ũ(m).
744 Jörg Teschner
The length operators Lσ,c associated to the boundary components c ∈ A() form
a commutative family of operators. The joint spectral decomposition for this family
of operators leads us to represent HL () as

HL () dη (c) HL (, c), (14.2)
L
where the integration is extended over the set L Rs+ of all colorings c of the boundary
by elements of R+ . It follows from Theorem 6 that the operators U(m) and Ũ(m)
commute (up to permutations of the boundary components) with the length operators
Lσ,c , c ∈ A(). Within the representation (14.2) one may therefore10 represent the
operators U(m)and Ũ(m) by families of operators (U(m, c))c∈L and (Ũ(m, c))c∈L .
14.2 The main result

The following theorem is the main result of this chapter.
Theorem 8. The operators U(m, c), c ∈ L, generate a tower of projective unitary

representations of the modular groupoids M().
The proof of Theorem 8 will take up the rest of this section.

To begin with, let us note that the necessary structure (9.2) of the Hilbert spaces
follows trivially from our definition of the length representation in Section 13, where
in the present case we simply have H(S3 , c3 ) C. It is furthermore clear that the
operators Ũ(m, c), c ∈ L generate a unitary projective representation of the modular
groupoid M() for each surface within the considered class. Let us finally note
that the naturality properties formulated in Section 9.3 clearly hold for the operators
U(m, c), c ∈ L. Our main task is therefore to show that Ũ(m, c) = U(m, c), as will be
established in Proposition 13 below.
Proposition 13. For all m ∈ M1ad () there exists a path πm ∈ [θm , ϕm ] such that we
have
Ũ(m, c) = U(m, c).
Proof. When we compare the respective definitions of U(m) and Ũ(m), we observe
that there are two main discrepancies that we need to deal with. First, it is not always
true that the path πm can be realized locally in the sense of Definition 7. It may
therefore not be clear a priori why there should exist a simple relation between U(m)
and Ũ(m).
Second, we may observe that the definition of the operators U(m) and Ũ(m) involves
products of operators Cσ,p , where the set of vertices p that the product is extended
over is generically much smaller in the case of U(m). This means that most of the
10According to Proposition 14 in Appendix B.
factors Cσ,p must ultimately cancel each other in the expression for Ũ(m). The first
step will be to prove Proposition 13 in the case that m can be realized locally.
Lemma 14. We have

Ũ(m) = U(m) (14.3)
whenever m = [τm , σm ] ∈ M1 () can be realized locally.
Proof. It follows from (12.11) and (11.5) that Cσ,p can be represented as a function
of the following operators

Cσ,p = Cσ,p pvp , qvp ; Lc2 , Lc1 ,
where vp is the vertex of ϕσ contained in the trinion Tp and cε ≡ cε (p), ε = 1, 2 are
the curves which represent the outgoing and incoming boundary components of Tp
respectively. Let us recall that the length operators Lτm ,cε and Lσm ,cε ε = 1, 2 satisfy
(12.6). These observations imply that
u(πm ) · Cσm ,p = Cτm ,p · u(πm ), (14.4)
whenever the operators pσ,p and qσ,p commute with u(πm ). Our task is therefore to
determine the set of all p ∈ σ0 for which this is the case. The condition that m can
be realized locally implies that (14.4) will hold unless p is located within m . The
claim now follows straightforwardly from these observations.
In order to treat the general case let us recall that Proposition 10 implies that πm may
be chosen as πm = ym π̂m ym −1 , where y is uniquely defined by the factorization
m
of Ym into elementary moves that can be realized locally, and π̂m is the (fixed) path
which was chosen to represent m in the case that m can be realized locally. This leads
to the following representation for Ũ(m):
Ũ(m) = D(τm ) · u(πm ) · D(σm )†
−1
(14.5)
= D(τm ) · u(ym ) · u(π̂m ) · u(ym ) · D(σm )† .
By using Lemma 14 one may deduce from (14.5) that the following holds:
Ũ(m) = Ũ(Ym ) · U(m) · Ũ(Ym−1 ) = U(Ym ) · U(m) · U(Ym )† . (14.6)
It remains to observe that
Lemma 15. We have

U(m1 ) · U(m2 ) · U(m1 )† = U(m2 )
whenever supp(m1 ) ∩ supp(m2 ) = ∅.
Recall that we had defined the notation supp(m) by supp(m) = {p} if m =

Zp , Bp , Sp and supp(m) = {p, q} if m = (pq), Fpq .
746 Jörg Teschner
Proof. Let us factorize HL () as HL () = Ksp⊗β1 ⊗ Ksp⊗σ1 \β1 , where β1 ⊂ σ1 is the
set of all edges which end in boundary components of . The representation (14.2)
may be rewritten as
⊕
HL () dηs (c) HL (, c), (14.7)
Rs+
where HL (, c) Ksp⊗σ1 \β1 for all c ∈ Rs+ . Let U(m, c), c ∈ Rs(m)
+ be the unitary
operators on HL (m , c) which represent the operators U(m) in the representation
(14.7). Within this representation it becomes almost trivial to complete the proof of
Proposition 13. Let β21 = β2 ∪ β1 , where βj ⊂ σ1 , j = 1, 2 are the sets of edges
which correspond to boundary components of mj . Let δj ⊂ σ1 , j = 1, 2 be the
sets of edges that are fully contained in the interior of mj respectively. Let finally
σ1 = σ1 \ (β21 ∪ δ2 ∪ δ1 ) We may then factorize HL () in the following way:

HL () = Ksp⊗β21 ⊗ Ksp⊗δ2 ⊗ Ksp⊗δ1 ⊗ Ksp⊗σ1 , (14.8)
where we define Ksp⊗δ = C if δ = ∅. This may be rewritten as

HL () dμ(c21 ) HL (, c21 ), (14.9)
s
R+21
⊗σ
where HL (, c21 ) Ksp⊗δ2 ⊗ Ksp⊗δ1 ⊗ Ksp 1 for all c21 ∈ Rs+21 , s21 = card(β21 ).
In the representation (14.9) we may represent U(mj ), j = 1, 2 by families of op-
erators U(mj , c21 ) which take the form U(m2 , c21 ) U2 (m2 , c21 ) ⊗ id ⊗ id and
U(m1 , c21 ) id ⊗ U1 (m1 , c21 ) ⊗ id respectively. The lemma follows easily from
these observations.
It follows from the lemma that U(Ym ) · U(m) · U(Ym )† = U(m), which completes
the proof of Proposition 13.
Taken together our results show that the quantization of Teichmüller spaces yields
a tower of unitary projective representations of the modular groupoid in the sense of
§9.3.
14.3 Concluding remarks

Let us recall that the mapping class group is a group of symmetries for both com-
plexes P t () and M(). Having projective unitary representations of the associated
groupoids Pt() and MC() will therefore induce corresponding representations of
the mapping class group by means of the construction in §8.2. It follows quite easily
from Proposition 7 that these two representations are equivalent.
Finally it is clearly of interest to calculate the phases in the relations u(π ) =
ζ ν( ) .
Conjecture 1. There exists a definition for the operators U(m), m ∈ M1 () such
that the phases ζ ν̂( ) which appear in the relations u(π ) = ζ ν( ) are trivial for all
but one ∈ M2 (), which can be chosen as the relation (10.9),a). The phase which
appears in the relation (10.9),a) is given as
πi
ζ ν( ) = e 2 cL , cL ≡ 1 + 6(b + b−1 )2 .
We now believe to have a proof of this conjecture. Details will appear elsewhere.
Our conjecture is also strongly supported by the calculation in [31] which establishes
a similar result for the realization of the pure mapping class group on the quantized
Teichmüller spaces.
Appendix A. The special functions eb (x) and sb (x)
The function sb (x) may be defined with the help of the following integral representa-
tion.
∞

1 dt sin 2xt x
log sb (x) = − . (A.1)
i t 2 sinh bt sinh b−1 t t
0
This function, or close relatives of it like

πi 2 πi
2 x e− 24 (2−Q ) sb (x),
2
eb (x) = e (A.2)
have appeared in the literature under various names like “Quantum Dilogarithm” [13],
“Hyperbolic G-function” [44], “Quantum Exponential Function” [60] and “Double
Sine Function”, we refer to the appendix of [34] for a useful collection of properties
of sb (x) and further references. The most important properties for our purposes are

(i) Functional equation: sb x − i b2 = 2 cosh π bx sb x + i b2 . (A.3)
(ii) Analyticity: sb (x) is meromorphic,
poles: x = cb + i(nb − mb−1 ), n, m ∈ Z≥0 , (A.4)
−1 ≥0
zeros: x = −cb − i(nb − mb ), n, m ∈ Z .
(iii) Self-duality: sb (x) = s1/b (x). (A.5)
(iv) Inversion relation: sb (x)sb (−x) = 1. (A.6)
(v) Unitarity: sb (x) = 1/sb (x̄). (A.7)
− π12i (1−4cb2 ) −1
(vi) Residue: resx=cb sb (x) = e (2π i) . (A.8)
748 Jörg Teschner
The function eb (x) clearly has very similar properties as sb (x). We list the properties
that get modified compared to sb (x) below.

(i) Functional equation: eb x − i b2 = (1 + e2πbx ) eb x + i b2 . (A.9)
πi
(iv) Inversion relation: eb (x)eb (−x) = eπix e− 6 (1+2cb ) .
2 2
(A.10)
−1
(vi) Residue: resx=cb sb (x) = (2π i) . (A.11)
Among the most remarkable properties satisfied by the function eb (x) is the so-
called pentagonal relation (A.12) which underlies the validity of the pentagonal relation
(8.11),
eb (p) · eb (q) = eb (q) · eb (p + q) · eb (p), (A.12)
Relation (A.12) is valid if q and p represent [p, q] = (2π i)−1 on L2 (R). Equa-
(A.12) in turn is equivalent to the following property of the function Eb (x) ≡
tion
eb − 2πx b :
Eb (U) · Eb (V) = Eb (U + V), (A.13)
where U = e2πbq , V = e2πbp . Proofs of (A.12) and (A.13) can be found in [60], [14],
[9], [54].
Appendix B. Operator-theoretical background

Let H be a separable Hilbert space. The algebra of bounded operators in H will
be denoted by B(H ). We will only need some of the most basic notions and results
from functional analysis and the theory of operator algebras as summarized e.g. in
[55, Chapter 14]. For the reader’s convenience and to fix some notations we shall
formulate the results that we need below.
Theorem 9 ([55, Theorem 14.8.14]). Let C be a commutative von Neumann subal-

gebra of B(H), where H is a separable Hilbert space. Then there exists a compact,
separable Hausdorff space X, a Radon measure μ on X, a measurable family of
Hilbert spaces {Hx }x∈X , and a unitary bijection

U : H → HC ≡ dμ(s) Hs (B.1)
X
such that
U · C · U† = {mf ; f ∈ L∞ (x, dμ)}, (B.2)
where mf is the multiplication operator defined by
mf : HC {x }x∈X → {f (x)x }x∈X ∈ HC . (B.3)
If C is a commutative von Neumann subalgebra of B(H ) we will call an operator

O on H C-decomposable if there exists a family of bounded operators Ox , defined on
Hx for μ-almost every x ∈ X, such that

U · O · U† = dμ(s) Ox . (B.4)
X
Let DC be the algebra of all C-decomposable operators.
Proposition 14 ([55, Proposition 14.8.8]). DC is the commutant of C within B(H ).
The unbounded operators that we will have to deal with will all be self-adjoint. We
will freely use standard functional calculus for self-adjoint operators. When we say
that two unbounded self-adjoint operators commute,
[ A, B ] = 0,
we will always mean commutativity of the spectral projections. Let F = {Aı ; ı ∈ I}
be a family of commuting self-adjoint operators defined on dense domains in a sep-
arable Hilbert space H. Standard functional calculus for commuting self-adjoint
operators associates to F a commutative von Neumann subalgebra CF of B(H).
Theorem 9 applied to CF yields the existence of a common spectral decomposition
for the family F , where X represents the one-point compactification of the spectrum
of F .
Tensor product notation. For a given finite set I we will often use the notation H ⊗I
instead of H ⊗card(I) , where card(I) is the number of elements in I. In order to avoid
fixing a numbering of the elements of I we shall find it useful to employ the following
“leg-numbering” notation. To a given a subset J ⊂ I we may associate the canonical
permutation operators
PJ : H ⊗I → H ⊗J ⊗ H ⊗I\J .
To an operator O ∈ B(H ⊗J ) we may then associate an operator OJ ∈ B(H ⊗I ) via
OJ ≡ P−1
J · (O ⊗ id) · PJ . (B.5)
If J = {i1 , i2 , . . . } we will sometimes write Oi1 i2 ... instead of OJ . We will also
abbreviate OJ1 ∪J2 ∪... to OJ1 J2 ... .
Appendix C. On the proof of Theorem 2

A similar statement is known [17] for the closely related groupoid Pt() whose el-
ements are the moves [ϕ̃2 , ϕ̃1 ] between fat graphs ϕ̃2 , ϕ̃1 which do not have the
decoration introduced in Section 3.1, but which have a numbering of the edges in-
stead.

Theorem 2 . The groupoid Pt() t() which
is the path groupoid of the complex P
has vertices ϕ̃ and the following generators and relations:
750 Jörg Teschner
Generators:
(i) Flips Fe along the edges e ∈ ϕ̃1 (see Figure 7).
(ii) The exchanges (ef ) of the numbers assigned to edges e and f .
Relations:
(a) There is no vertex that both edges e and f are incident to: Ff Fe = Fe Ff .
(b) The edges e and f are incident to the same vertex: Fe Ff Fe Ff Fe = (ef ).
(c) (ef ) Fe = Ff (ef ).
(d) Fe Fe = id.
It was observed in [41], [17], [10] that Theorem 2 follows quite easily from the fact
[39], [35] that the fat graphs ϕ̃ can be used to label cells in certain cell decompositions
[25], [39] of T () × Rs+ . This allows one to associate a path π̂ in T () × Rs+ to

each path π̃ ∈ [ϕ̃2 , ϕ̃1 ] in Pt(). Each closed path π̂ in T () × Rs+ can be deformed
into small circles around the codimension two faces of the cell decompositions from
[25], [39]. The latter are easily identified with the relations listed in Theorem 2 . This
implies that no relations other than those listed in Theorem 2 are needed to contract
a closed path π̃ ∈ [ϕ̃, ϕ̃] to the identity.
It remains to show that Theorem 2 follows from Theorem 2 . To this aim it is impor-

tant to observe that Pt() imbeds into Pt() by means of the following construction:
Given the numbered fat graph ϕ̃ let us define a decorated fat graph ϕ according to
the following rule: Pick any numbering of the vertices of ϕ̃. For each vertex v let us
distinguish among the edges which emanate from v the one with the smallest number
assigned to it. For each move Fe let us choose a lift φe of the form φe = δe ωue ve δe ,
where the vertices ue and ve represent are the ends of e and δe , δe change the decoration
only. It is then crucial to check that the relations of Pt() ensure that the image of

Pt() within Pt() is simply connected. This follows if the images of the relations
listed in Theorem 2 are contractible within Pt(). We have drawn a particular ex-
ample for such an image in Figure 24 below. It is easy to verify that the image of the
corresponding path in Pt() is closed. More generally one needs to consider all rela-
tions obtained by changes of the numbering of the edges. These relations are obtained
by inserting η−1 η between the moves which occur in the image of the relation under
consideration, where η represents the change of decoration induced by a change of
numbering. In a similar way one may convince oneself that all the relations listed
in Theorem 2 are mapped to closed paths in Pt(), which completes the proof that

Pt() embeds into Pt().
It remains to show that each closed path in Pt() is homotopic to a closed path

in the image of Pt(). Let us consider a closed path π ∈ [ϕ, ϕ] ∈ Pt(). Write
π = ωn · · · ω1 , where each ωk is of the form ωk = δk ωuk vk δk , with δk , δk being
composed out of moves ρw and permutations only. Choose any numbering for the
edges of ϕ and denote the resulting numbered fat graph by ϕ̃. The path π then defines

a path π̃ in Pt() by means of the following construction: Substitute the moves ωr
for k = 1, . . . , n by the corresponding flips along the edges which connect uk and vk ,
and then multiply the result by the necessary permutations of the numbers assigned to
F1
(12) F1 *
1 2 *
*
*
* * *
1 2
2 1
* *
F2
* * F2
1 1
* *
2 2
F1
* *
Figure 24. Image of the pentagon in Pt().
the edges of ϕ̃. π̃ has the form π̃ = n · · · 1 , where k ∈ [ϕ̃k+1 , ϕ̃k ] factorize
into a single flip Fk times a change of numbering Pk , k = Pk Fk .
The path π̃ is mapped to a path π in Pt() by means of the construction above.
π has the form π = φn · · · φ1 with φk ∈ [ϕk+1 , ϕk ]. On the other hand let us note

−1
that the path π = φn · · · φ1 with φk ≡ ηk+1 ωk ηk is clearly homotopic to π
for all changes of decorations ηk which satisfy ηn+1 = η1 . For suitable choice of the
ηk one gets φk ∈ [ϕk+1 , ϕk ]. It is then easy to see that φk and φk are homotopic.
Appendix D. Proof of Proposition 10

One may easily convince oneself that the moves Zp and Bp can generically not be
realized locally. There are simple sufficient criteria for a move m to be realizable
locally.
Lemma 16. (i) A move m = [τm , σm ] ∈ M1 () can always be realized locally if all
but one of the boundary components of m are contained in {ce ; e ∈ Aσ }.
(ii) The moves Bp , Bp , Tp can be realized locally if the curve cp,1 which represents
the boundary component of tp assigned number 1 in Figure 12 is contained in Aσ .
(iii) The moves Fpq , Apq , Sp , Spq and Bpq can always be realized locally.
Proof. Straightforward verifications.

752 Jörg Teschner
Proposition 15 (= Proposition 10). If m ∈ M1 () is admissible, but can not be

realized locally, there always exists a path m which is (i) homotopic to m within
M1ad (), and (ii) takes the form
m = Ym m Ym−1 , (D.1)
where Ym is a chain composed out of Zp -moves and Fpq -moves which can all be
realized locally.
Proof. The claim will follow easily from another auxiliary result that we will formulate
as a separate lemma. As a preparation, let us recall that cutting a surface along the
curves c(σ, e), e ∈ Aσ produces a surface † , the connected components of which all
have genus zero. Let us call a marking σ irreducible if the set of connected components
of † has only one element. It is easy to see that a marking σ is irreducible iff it has
precisely one outgoing external edge.
Lemma 17. Let σ be a marking on a surface that is irreducible. For any chosen
boundary component b of there exists a chain of Zp -moves and Fpq -moves that
(i) preserves admissibility in each step, (ii) consists only of moves that can be realized
locally, and (iii) transforms σ to a irreducible marking σ whose outgoing external
edge ends in the chosen component b.
Sketch of proof. In order to check the following arguments it may be useful to think
of an irreducible marking as being obtained from the corresponding marking σ † on
the surface † of genus zero by identifying the appropriate boundary circles. To each
incoming boundary component of † there corresponds a unique element of Aσ .
The following claim is easy to verify. By means of Fpq -moves one may transform
σ to a marking σ̃ which has the following two properties. First, for each vertex p ∈ σ̃0
at least one of the edges that are incident at p is contained in Aσ . Second, there is no
edge e ∈ σ1 which connects a vertex p to itself. These two properties insure that for
each p ∈ σ̃0 either Zp or Zp−1 preserves the admissibility of the marking. It is then
not very hard to construct a sequence of Zp -moves that (i) can all be realized locally
and (ii) which transform σ̃ to a marking σ̃ whose outgoing external edge ends in the
chosen boundary component b.
By means of a chain of Fpq -moves one may finally transform σ̃ back to a marking

σ that differs from the original marking only by the desired change of decoration.
End of proof of Lemma 15. Lemma 16 allows us to restrict attention to the cases
m = Zp , Bp , Bp , Tp . We need to transform the original marking to another one
which has the property that both incoming boundary components of the trinion tp are
contained in Aσ . Given a curve c ∈ Cσ \ Aσ there is a unique subsurface c →
with marking σc such that (i) c is the unique outgoing curve in the boundary of c ,
and (ii) the incoming boundary curves of c are contained in Aσ . In order to infer
the existence of the chain Yp it clearly suffices to apply Lemma 17 to the subsurfaces
c1 and c2 respectively, where c1 and c2 are the incoming boundary components of
the trinion with label p.
Lemma 15 implies that πm may be chosen as πm = ym π̂m ym −1 , where y is

m
uniquely defined by the factorization of Ym into elementary moves that can be realized
locally, and π̂m is the (fixed) path which was chosen to represent m in the case that m
can be realized locally.
Appendix E. On the proof of Theorem 6
E.1 A technical preliminary

To begin with, let us present a technical result that facilitates the explicit computations.
The operator functions we will be interested in are of the form

O ≡ O {zı ; ı ∈ ϕ1 } = Cτ ex(τ ) , x(τ ) ≡ τ (e) ze , (E.1)
τ ∈F e∈ϕ1
where the summation is taken over the space F of all maps ϕ1 e → τ (e) ∈ 21 Z,
and the coefficients Cτ are assumed to be non-vanishing for a finite number of τ ∈ F
only, in which case we assume Cτ ∈ R+ . These operators are densely defined and
positive self-adjoint due to the self-adjointness of x(τ ). The cone generated by operator
functions of the form (E.1) will be denoted C+ (ϕ).
Proposition 16. Let ϕ, ϕ be two fat graphs that are related by ρ ≡ ωvw ∈ P t(),
and let us adopt the labelling of the relevant edges given by Figure 25. For each
O ∈ C+ (ϕ) one has
−1
aρ (O) ≡ u(π) · O · u(π)−1 = Eb (ze ) · O · Eb (ze ) , (E.2)

where Eb (z) ≡ eb − 2πb z
, and O is related to O via

O ≡ O za , zb + ze , zc , zd + ze , −ze , zı ; ı ∈ ϕ1 \ {a , b , c , d , e } if
(E.3)
O = O za , zb , zc , zd , ze , zı ; ı ∈ ϕ1 \ {a, b, c, d, e} .
Proof. Given the explicit expression for Tvw we only have to verify that
e−2πi pv qw · O · e−2πi pv qw = O ,
with O and O being related by (E.3). Keeping in mind (4.6) we notice that the
operators zf may be represented in the form zf = zf,v + zf,w , where v, w ∈ ϕ0 are
the vertices connected by the edge f ∈ ϕ0 . Let us label the vertices in ϕ0 and ϕ0 such
that the edges a and d are incident to the vertices va and vd besides to the vertex v
respectively, and similarly for the vertices vb , vc and w. This leads to the following
754 Jörg Teschner
a b a b
v
v
e
w * e
*
w
d c
* d * c
Figure 25. Labelling of the edges which are relevant for the description of a flip.
representations for the relevant Fock variables:

ze = 2π b(qv − pv + pw ), ze = 2π b(qw − qv − pw ),
za = 2π bpv + za,va , za = 2π bpv + za ,va ,
zb = 2π b(qw − pw ) + zb,vb , zb = 2π b(qv − pv ) + zb ,vb , (E.4)
zc = 2π b(−qw ) + zc,vc , zc = 2π b(−qw ) + zc ,vc ,
zd = 2π b(−qv ) + zd,vd , zd = 2π bpw + zd ,vd .
To complete the proof of Proposition 16 is now the matter of a straightforward calcu-
lation.
Remark 10. This yields in particular the formulae

1 1 1 1
e−za = e− 2 za (1 + e−ze )e− 2 za , e−zb = e+ 2 zb (1 + e+ze )e+ 2 zb ,
1 1
ze = −ze , 1 1
e+zd = e+ 2 zd (1 + e+ze )e+ 2 za , e+zc = e− 2 zc (1 + e−ze )e− 2 zc .
It is quite obvious that these transformations reduce to their classical counterparts (3.3)
in the limit b → 0.
As an example let us consider the monomials Mnnba nndc ≡ e 2 (na za +nb zb +nc zc +nd zd ) ,
1
where n! , ! ∈ {a, b, c, d} are restricted by the requirement that 2N ≡ na +nc −nb −nd
must be even. One then finds certain simplifications on the right hand side of (E.2):
−1 nb nd Eb (ze + π ib2 N) nb nd
Eb (ze ) · Mnnba nndc · Eb (ze ) = Mna nc Mna nc
Eb (ze − π ib2 N)

N−1 (E.5)
2
n n 2m −ze n n
= Mnba ndc (1 + q e ) Mnba ndc .
m=− N−1
2
E.2 Invariance of length operators

In the main text we have explained how to reduce the proof of Theorem 6 to the
following lemma:
Lemma 18. Assume that m ∈ M1 () can be realized locally. We then have the
relation
u(πm ) · Lσm ,c = Lτm ,c · u(πm ) (E.6)
for all curves c such that c ∈ Cσm and c ∈ Cτm .
On the proof of Lemma 18. It is easy to see that our recursive Definition 6 of the length
operators Lσ,c reduces the task of proving Lemma 18 to the curves that represent a
boundary component of m . Let us write Lσ,e ≡ Lσ,c(σ,e) .
In the cases m = Fpq , Sp , Tp , Zp we will need to verify Lemma 18 by direct cal-
culations. Let us begin with the case m = Fpq . The claim is again trivial for the length
operators associated to the incoming boundary components of m . Let us therefore
focus on the length operators Lσm ,f , Lτm ,f assigned to the outgoing boundary com-
ponent of m . Definition 6 yields the following expressions for the length operators
Lσm ,f and Lτm ,f respectively.
Lσm ,f = ez21 +y3 +y2 +y1 + ez21 −y3 +y2 +y1 + e−z21 −y3 −y2 −y1
+ ey1 −y2 −y3 + ey1 +y2 −y3 + e−y1 −y2 −y3
+ ez21 +y2 +y1 L3 + ey1 −y3 L2 + e−y2 −y3 L1 ,
(E.7)
Lτm ,f = e+z32 +y3 +y2 +y1 + e−z32 −y3 −y2 −y1 + e−z32 −y3 −y2 +y1

+ ey1 +y2 +y3 + ey1 −y2 −y3 + ey1 +y2 −y3

+ ey2 +y1 L3 + ey1 −y3 L2 + e−z32 −y2 −y3 L1 .
We have used the abbreviations yι ≡ yϕσ ,eι , yι ≡ yϕτ ,eι , Lι ≡ Lσ,eι and Lι ≡ Lτ,eι
for ι ∈ {1, 2, 3}, as well as zι ≡ zϕσ ,eι , zι ≡ zϕτ ,eι for ι ∈ {32, 21}. The labelling of
the edges is the one introduced in Figure 26. According to Proposition 16 we need to
f f
e21 Fpq e32

e1
e3
e1 e3
e2 e2
Figure 26. Labelling of the edges of σm for m = Fpq .

756 Jörg Teschner
calculate

Eb (z32 ) · ey3 +y2 +y1 +z32 + ey1 −y2 −y3

+ ey2 +y1 L3 + ey1 −y3 L2 + e−z32 −y2 −y3 L3
1 1 (E.8)
+ e 2 (y1 +y2 −y3 ) (1 + e−z32 )e 2 (y1 +y2 −y3 )
1 1
+ e− 2 (y1 +y2 +y3 +z32 ) (1 + e−z32 )e− 2 (y1 +y2 +y3 +z32 ) · Eb (z32 )† .
We finally need to apply equation (E.5). The terms in the first line of (E.8) have N = 1,
those in the second line N = 0, and all other terms have N = −1. Straightforward
application of equation (E.5) shows that the expression given in (E.8) equals Lτm ,f , as
claimed.
The next case we will consider is m = Sp . It clearly suffices to restrict attention to
the case that m = 1,1 , the one-holed torus. On m let us consider the fat graphs
ϕi , i = 1, 2, 3 depicted in Figure 27.
a c c
c b
b b
d d d
a a a
(1) (2) (3)
Figure 27. Fat graphs ϕi on the one-holed torus.
We have ϕ1 = ϕ̃m and ϕ = θ respectively. The sequence of fat graph (ϕ , ϕ , ϕ )

3 m 3 2 1
defines the path πm ∈ [θm , ϕ̃m
] which we will use. The only relevant curve is the curve
β which represents the boundary of m . Definition 6 yields the following expressions

for the length operators Lϕ1 ,β and Lϕ3 ,β respectively:
Lϕ1 ,β = ezd −zc + 2 cosh(zc + zd + za + zb )
+ ezd (1 + ezb + ezb +za ) + e−zc (1 + e−za + e−za −zb ),
(E.9)
Lϕ3 ,β = ezc −za + 2 cosh(za + zd + zc + zb )
+ ezc (1 + ezb + ezb +zd ) + e−za (1 + e−zd + e−zd −zb ).
The labelling of the relevant edges is the one introduced in Figure 27. With the help
of Proposition 16 one may calculate

Lϕ2 ,β ≡ a[ϕ2 ,ϕ1 ] Lϕ1 ;β = ezc −za + 2 cosh(za + zd + zc + zb )
+ ezc (1 + ezd + ezd +zb ) + e−za (1 + e−zb + e−zb −zb ).
(E.10)

One may then apply Proposition 16 once more to calculate a[ϕ3 ,ϕ2 ] Lϕ2 ,β . The result
is Lϕ3 ,β as claimed.
We have furthermore verified by direct calculations that Lemma 18 holds in the

cases m = Tp , Zp , Bp , Bp respectively. These calculations proceed along very similar
lines as in the previous two cases, which motivates us to omit the details.
In the remaining cases m = Apq , m = Bpq and m = Spq one may use the
factorization of the chains πm which follows from the definitions (10.5), (10.7) and
(10.8) respectively in order to reduce the proof to the cases where the result is already
established. This completes our discussion of the proof of Lemma 18.
References
[1] S. Axelrod, S. Della Pietra, and E. Witten, Geometric quantization of Chern-Simons gauge
theory. J. Differential Geom. 33 (1991), 787–902. 688
[2] A. Yu. Alekseev, H. Grosse, and V. Schomerus, Combinatorial quantization of the Hamil-
tonian Chern-Simons theory. I; II. Comm. Math. Phys. 172 (1995), 317–358; ibid. 174
(1996), 561–604. 688
[3] A. Yu. Alekseev and V. Schomerus, Representation theory of Chern-Simons observables.
Duke Math. J. 85 (1996), 447–510. 688
[4] J. E. Andersen, J. Mattes, and N. Reshetikhin, Quantization of the algebra of chord dia-
grams. Math. Proc. Cambridge Philos. Soc. 124 (1998), 451–467. 688
[5] B. Bakalov and A. Kirillov, Jr., On the Lego-Teichmüller game. Transform. Groups 5
(2000), 207–244. 690, 691, 726, 730
[6] B. Bakalov and A. Kirillov, Jr., Lectures on tensor categories and modular functors. Univ.
Lecture Ser. 21, Amer. Math. Soc., Providence, RI, 2001. 688, 720, 721, 723, 725
[7] E. W. Barnes, The theory of the double gamma function. Philos. Trans. Roy. Soc. London
Ser. A 196 (1901), 265–387.
[8] E. Buffenoir and P. Roche, Link invariants and combinatorial quantization of Hamiltonian
Chern-Simons theory. Comm. Math. Phys. 181 (1996), 331–365. 688
[9] A. G. Bytsko and J. Teschner, R-operator, co-product and Haar-measure for the modular
double of Uq (sl(2, R)). Comm. Math. Phys. 240 (2003), 171–196. 691, 748
[10] L. O. Chekhov and R. C. Penner, On quantizing Teichmüller and Thurston theories.
In Handbook of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing
House, Zurich 2007, 579–645. 687, 750
[11] M. Dehn, Die Gruppe der Abbildungsklassen. Acta Math. 69 (1938), 135–206.
[12] L. D. Faddeev, Modular double of a quantum group. In Conférence Moshé Flato 1999,
Vol. I (Dijon), Math. Phys. Stud. 21, Kluwer Acad. Publ., Dordrecht 2000, 149–156. 691
[13] L. D. Faddeev and R. M. Kashaev, Quantum dilogarithm. Modern Phys. Lett. A 9 (1994),
427–434. 747
[14] L. D. Faddeev, R. M. Kashaev, and A.Yu. Volkov, Strongly coupled quantum discrete Liou-
ville theory. I. Algebraic approach and duality. Comm. Math. Phys. 219 (2001), 199–219.
748
758 Jörg Teschner
[15] V. V. Fock, Dual Teichmüller spaces. Preprint, 1998; arXiv:hep-th/9702018. 687, 689,
690, 695, 697, 706
[16] V. V. Fock and L. O. Chekhov, Quantum mapping class group, pentagon relation, and
geodesics. Tr. Mat. Inst. Steklova 226 (1999), 163–179; English transl. Proc. Steklov Inst.
Math. 226 (1999), 149–163. 690, 737
[17] V. V. Fock and L. O. Chekhov, Quantum Teichmüller spaces. Teoret. Mat. Fiz. 120 (1999),
511–528; English transl. Theoret. and Math. Phys. 120 (1999), 1245–1259. 687, 689,
749, 750
[18] V. V. Fock and A. B. Goncharov, Cluster ensembles, quantization and the dilogarithm.
Preprint, 2007; arXiv:math.AG/0311245. 687
[19] V. V. Fock and A. A. Rosly, Poisson structure on moduli of flat connections on Riemann
Soc. Transl. (2) 191, Amer. Math. Soc., Providence, RI, 1999, 67–86. 688
[20] L. Funar and R. Gelca, On the groupoid of transformations of rigid structures on surfaces.
J. Math. Sci. Univ. Tokyo 6 (1999), 599–646. 726
[21] S. Gervais, Presentation and central extensions of mapping class groups. Trans. Amer.
Math. Soc. 348 (1996), 3097–3132. 740, 741
(1984), 200–225. 689
[23] W. M. Goldman, Topological components of spaces of representations. Invent. Math. 93
(1988), 557–607. 689
[24] V. Guillemin and S. Sternberg, Symplectic techniques in physics. 2nd ed., Cambridge
University Press, Cambridge 1990. 714
[25] J. L. Harer, The virtual cohomological dimension of the mapping class group of an ori-
entable surface. Invent. Math. 84 (1986), 157–176. 750
[26] N. J. Hitchin, The self-duality equations on a Riemann surface. Proc. London Math. Soc.
(3) 55 (1987), 59–126. 689
[27] N. J. Hitchin, Flat connections and geometric quantization. Comm. Math. Phys. 131 (1990),
347–380. 688
[28] Y. Imayoshi and M. a. Taniguchi, An introduction to Teichmüller spaces. Springer-Verlag,
Tokyo 1992. 705
[29] R. M. Kashaev, The hyperbolic volume of knots from the quantum dilogarithm. Lett. Math.
Phys. 39 (1997), 269–275. 689
Math. Phys. 43 (1998), 105–115. 687, 689, 690, 691, 694, 700, 701, 702, 709
[31] R. M. Kashaev, Liouville central charge in quantum Teichmüller theory. Tr. Mat. Inst.
Steklova 226 (1999), 72–81; English transl. Proc. Steklov Inst. Math. 226 (1999), 63–71.
747
[32] R. M. Kashaev, On the spectrum of Dehn twists in quantum Teichmüller theory. In Physics
and combinatorics (Nagoya, 2000), World Scientific, Singapore 2001, 63–81. 689, 690,
691, 698, 699, 717, 740
[33] R. Kashaev, The quantum dilogarithm and Dehn twists in quantum Teichmüller theory. In
Integrable structures of exactly solvable two-dimensional models of quantum field theory
(Kiev, 2000), NATO Sci. Ser. II Math. Phys. Chem. 35, Kluwer Acad. Publ., Dordrecht
2001, 211–221. 690, 739, 741
[34] S. Kharchev, D. Lebedev, and M. Semenov-Tian-Shansky, Unitary representations of

Uq (sl(2, R)), the modular double and the multiparticle q-deformed Toda chains. Comm.
Math. Phys. 225 (2002), 573–609. 747
[35] M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy
function. Comm. Math. Phys. 147 (1992), 1–23. 750
[36] Y. Laszlo, Hitchin’s and WZW connections are the same. J. Differential Geom. 49 (1998),
547–576. 688
[37] F. Luo, A presentation of the mapping class groups. Math. Res. Lett. 4 (1997), 735–739.
[38] G. Moore and N. Seiberg, Classical and quantum conformal field theory. Comm. Math.
Phys. 123 (1989), 177–254. 690, 691, 726, 733
113 (1987), 299–339. 690, 695, 698, 750
[40] R. C. Penner, Weil-Petersson volumes. J. Differential Geom. 35 (1992), 559–608. 695,
699
143–215. 700, 750
[42] B. Ponsot and J. Teschner, Clebsch-Gordan and Racah-Wigner coefficients for a continu-
ous series of representations of Uq (sl(2, R)). Comm. Math. Phys. 224 (2001), 613–655.
691
[43] N. Reshetikhin and V. G. Turaev, Invariants of 3-manifolds via link polynomials and
quantum groups. Invent. Math. 103 (1991), 547–597. 688
[44] S. N. M. Ruijsenaars, First order analytic difference equations and integrable quantum
systems. J. Math. Phys. 38 (1997), 1069–1146. 747
[45] I. E. Segal, Transforms for operators and symplectic automorphisms over a locally compact
abelian group. Math. Scand. 13 (1963), 31–43. 714
[46] T. Shintani, On a Kronecker limit formula for real quadratic fields. J. Fac. Sci. Univ. Tokyo
Sect. IA Math. 24 (1977), 167–199.
[47] C. Sorger, La formule de Verlinde. Astérisque 237 (1996), Sém. Bourbaki, Vol. 1994/95,
Exp. No. 794, 87–114. 688
[48] Takhtajan, L. A., Teo, L.-P., Quantum Liouville theory in the background field formalism.
I. Compact Riemann surfaces. Comm. Math. Phys. 268 (2006), 135–197. 688
[49] J. Teschner, Liouville theory revisited. Classical Quantum Gravity 18 (2001), R153–R222.
687
[50] J. Teschner, A lecture on the Liouville vertex operators. Internat. J. Modern Phys. A 19
(2004), Suppl. May, 436–458. 692
[51] J. Teschner, An analog of a modular functor from quantized Teichmüller theory, II. In
preparation. 691, 692
[52] V. G. Turaev, Quantum invariants of knots and 3-manifolds. De Gruyter Stud. Math. 18,
Walter de Gruyte, Berlin 1994. 720, 721, 723
[53] H. Verlinde, Conformal field theory, two-dimensional quantum gravity and quantization
of Teichmüller space. Nuclear Phys. B 337 (1990), 652–680. 687
[54] A.Yu.Volkov, Noncommutative hypergeometry. Comm. Math. Phys. 258 (2005), 257–273.
748
760 Jörg Teschner
[55] N. R. Wallach, Real reductive groups. II. Pure Appl. Math. 132, Academic Press, Boston
1992. 748, 749
[56] E. Witten, Quantum field theory and the Jones polynomial. Comm. Math. Phys. 121 (1989),
351–399. 688
[57] S. A. Wolpert, On the symplectic geometry of deformations of a hyperbolic surface. Ann.
of Math. (2) 117 (1983), 207–234. 690
[58] S. A. Wolpert, On the Weil-Petersson geometry of the moduli space of curves. Amer. J.
Math. 107 (1985), 969–997. 690, 705
25 (1987), 275–296. 690
[60] S. L. Woronowicz, Quantum exponential function. Rev. Math. Phys. 12 (2000), 873–920.
747, 748
Chapter 17
On quantum moduli space of flat

PSL2 (R)-connections on a punctured surface
Rinat M. Kashaev
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761
2 The decorated moduli space of flat PSL2 (R)-connections . . . . . . . . . . 763
2.1 Subgroups of PSL2 (R) . . . . . . . . . . . . . . . . . . . . . . . . . 763
2.2 The moduli space of flat connections . . . . . . . . . . . . . . . . . 763
2.3 A first definition of the decorated moduli space . . . . . . . . . . . . 764
2.4 A second definition of the decorated moduli space . . . . . . . . . . . 765
2.5 Equivalence of the two definitions . . . . . . . . . . . . . . . . . . . 767
2.6 Coordinates for the decorated moduli space . . . . . . . . . . . . . . 767
2.7 Reconstruction of a flat connection from the coordinates . . . . . . . 770
2.8 The transition functions . . . . . . . . . . . . . . . . . . . . . . . . . 771
2.9 Goldman’s symplectic structure . . . . . . . . . . . . . . . . . . . . 772
3 Towards the quantum theory . . . . . . . . . . . . . . . . . . . . . . . . . 772
3.1 Quantization of an ideal triangle . . . . . . . . . . . . . . . . . . . . 773
3.2 Quantization of the flip transformation . . . . . . . . . . . . . . . . . 775
3.3 A solution by Woronowicz and Zakrzewski . . . . . . . . . . . . . . 777
3.4 Spectrum of the length operator . . . . . . . . . . . . . . . . . . . . 780
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 781
1 Introduction
Quantum theory of the Teichmüller space of a punctured surface developed recently in
the works [1], [2], [4], [8] is an example of a topological quantum field theory with infi-
nite dimensional quantum state space. This theory, for example, is connected with the
quantum theory of the Liouville equation [16]. From the purely mathematical view-
point it leads to an interesting infinite dimensional projective unitary representation
of the mapping class group of the underlying surface.
Teichmüller space is a connected component of a bigger space, namely the moduli
space of PSL2 (R) flat connections on the surface, or equivalently the space of conju-
gacy classes of representations of the fundamental group of the surface in the group
762 Rinat M. Kashaev
PSL2 (R). This space is symplectic with Goldman’s symplectic form which restricts to
the Weil–Petersson symplectic form on the Teichmüller component. It is thus natural
to ask if one can extend the quantum Teichmüller theory to a quantum theory of the
moduli space of flat connections. Physically this is the question of quantizing the
Chern–Simons gauge theory with a non-compact gauge group which in our case is the
group PSL2 (R) .
Unlike Teichmüller space, the other connected components of the moduli space
are not topologically trivial [6]. This is why quantization of the whole moduli space is
not expected to be a straightforward generalization of the quantum Teichmüller space.
Since the quantization procedure is not yet a functor (probably with the exception
of the deformation quantization), the canonical quantization in physics is sensitive to
a choice of coordinates. In the case of the Teichmüller space a convenient system of
coordinates is given by real analytic coordinates, for example, the Penner coordinates
coming from the decorated Teichmüller space, in which both the Weil–Petersson sym-
plectic structure and the action of the mapping class group are described by explicit
rational expressions. This is why the first step in the quantization program of the
moduli space of flat connections could be an appropriate choice of coordinates.
This chapter is essentially a review of some of the results related to the quantization
problem of the moduli space of irreducible flat PSL2 (R)-connections on a punctured
surface with parabolicity conditions around the punctures.
Section 2 is devoted to the decorated moduli space of flat connections which gen-
eralizes Penner’s decorated Teichmüller space. In addition to the constructions of the
work [11], here we also give another geometrical definition for the decorated moduli
space which adapts to our context Penner’s original definition of the decorated Teich-
müller space. It is worth noticing that the definition of [11], being more algebraic in
nature, remains valid for the group PSL2 (F), where F is an arbitrary field.
In Section 3, we first show the relevance of the works of Woronowicz and Za-
krzewski [17], [18] to the quantization problem of the moduli space. Then, we con-
sider a simple example of an annulus triangulated into two ideal tetrahedra, and derive
a formula for the trace of the holonomy associated to the only non-contractible simple
loop in the annulus. Quantization of this trace function leads to a certain difference
operator in the Hilbert space of square integrable functions on the real line. The spec-
trum of this operator appears to be very much sensitive to the sign variable which
distinguishes between different components of the moduli space.
Acknowledgement. I would like to thank V. Fock for pointing out a few inaccurate
statements in the initial version of this exposition and for making a number of valuable
comments. This work is supported in part by the Swiss National Science Foundation.
Chapter 17. On quantum moduli space of flat PSL2 (R)-connections 763
2 The decorated moduli space of flat PSL2 (R)-connections
2.1 Subgroups of PSL2 (R)

Fix a one-parameter subgroup U ⊂ PSL2 (R) = SL2 (R)/{±1} of parabolic elements
corresponding to upper triangular unipotent matrices, i.e. matrices with trace equal to
±2. To avoid confusion with the term “parabolic subgroup” in the algebraic group
context, in this chapter subgroups conjugate to U will be called unipotent subgroups.
The normalizer of U , B = N (U ), is a Borel subgroup corresponding to upper
triangular matrices. In the exact sequence of group homomorphisms
1→U →B→T →1
the group T = B/U is identified with the Cartan subgroup represented by diagonal
matrices. The Bruhat decomposition of PSL2 (R) with respect to B consists of only
two cells
PSL2 (R) = Bθ B B
where θ ∈ N (T ) ⊂ G is a fixed representative of the only nontrivial element of the
Weyl group N (T )/T . We choose 0 it1 in a unique way by the condition θ = 1 so that
2
it is represented by the matrix −1 0 .

In the standard action of PSL2 (R) by linear fractional transformations in the upper
half of the complex plane, the Cartan subgroup is characterized by the condition that it
fixes 0 and ∞, while the Borel subgroup is the stabilizer of ∞. The element θ is given
by inversion with respect to the unit circle followed by complex conjugation and sign
change. Elements in the two Bruhat cells are distinguished depending on whether ∞
is moved or not.
2.2 The moduli space of flat connections

Consider a punctured surface of finite type = g,s which is the complement of
a finite set of points V = {P1 , . . . , Ps } ⊂ , called punctures, in a closed oriented
surface of genus g:
= − V.
We assume the following restrictions on g and s:
s > 0, κ := −χ() = 2g − 2 + s > 0.
Denote by G the group PSL2 (R). Let M be the space of gauge classes of irreducible flat
G-connections on with parabolic holonomies around the punctures. Connections
representing points of M will be called flat connections and the space M itself will
be called the moduli space of flat connections. There is a one-to-one correspondence
between elements of M and conjugacy classes of irreducible representations in G
of the fundamental group of the surface with the parabolicity conditions around the
punctures. The Teichmüller space T is a connected component of M given by faithful

representations corresponding to complete hyperbolic structures on .
The decorated Teichmüller space of Penner [14] is a (trivial) principal Rs>0 -bundle
πT : T˜ → T , where each fiber is given by collections of s horocycles, one about
each puncture. To generalize this construction to the case of the moduli space M we
first define a “connection horocycle” at a puncture associated to a flat connection and
which corresponds to a usual geometric horocycle at the level of gauge equivalence
classes.
2.3 A first definition of the decorated moduli space

With any path a in , starting at some x ∈ and ending at some puncture P , we
associate the homotopy class (a) of loops based at x which go along a towards P ,
then go around P along a small circle in the counterclockwise direction with respect
to the orientation of the surface, and then return back to x along a −1 .
Denote by H the two-dimensional hyperbolic plane for which the group G is the
group of orientation preserving isometries. Consider a flat connection f on and a
pair (a, h) where a is a path with initial point x ∈ and terminal point a puncture P ,
and h is a horocycle in H centered at the only fixed point of the f -holonomy along the
loop (a). We say that two such pairs (a1 , h1 ) and (a2 , h2 ) are equivalent if a1 and a2
have the same terminal point and h1 = f12 h2 , where f12 is the f -parallel transport
along any path in homotopic in ∪ {P } to the path a1 · a2−1 . This is an equivalence
relation which does not depend on the choice of the path defining the element f12 , since
different choices give parallel transports differing only by elements of the stabilizer
subgroups of the horocycles. Namely, let fi be the f -holonomies along the loops
(ai ) for i = 1, 2. One has fi hi = hi , i = 1, 2. If f12 is the f -parallel transport
along a fixed path homotopic to a1 · a2−1 in ∪ {P }, then any other choice for this path
gives an f -parallel transport of the form f12 = f m f = f f m for some integer m.
1 12 12 2

Thus, one has f12 h2 = f12 h2 .
A connection horocycle at the puncture P associated to the flat connection f (also
called simply an f -horocycle) is an equivalence class of such pairs.
Gauge transformations of connections naturally act on connection horocycles.
Let g : → G correspond to a gauge transformation f → f given by f (p) =
g(x)f (p)g(y)−1 for any path p in connecting points x and y with f (p), f (p)
being the f - and f -parallel transports along p respectively. Then, this gauge trans-
formation acts on an f -horocycle represented by a pair (a, h) to give an f -horocycle
represented by the pair (a, g(x)h) where x ∈ is the initial point of a.
We define the decorated moduli space M̃ as the collection of gauge classes of
multiples (f, H1 , . . . , Hs ), where f is a flat connection and Hi is an f -horocycle at
the puncture Pi for all 1 ≤ i ≤ s.
Proposition 1. The natural projection π : M̃ → M given by π([f, H1 , . . . , Hs )]) =

[f ] is a principal Rs>0 -fibration.
Proof. A free and transitive action of the group R>0 on f -horocycles at a given
puncture can be seen as follows.
Let a be a path in with initial point x ∈ and terminal point a puncture P . Let
Ua ⊂ G be the stabilizer unipotent subgroup of the f -holonomy along the loop (a)
with respect to the adjoint action. Any horocycle h representing an f -horocycle at P
is invariant with respect to the action of Ua which, in fact, coincides with the stabilizer
group of h. The normalizer subgroup N (Ua ) ⊂ G of Ua acts transitively on the
space of all horocycles representing f -horocycles at P . In this way we obtain a free
and transitive action of the quotient group N(Ua )/Ua on the set of all f -horocycles
at P . On the other hand, the group N (Ua )/Ua is identified with the Cartan subgroup
conjugate to the subgroup of diagonal matrices. The latter is isomorphic to R>0 .
2.4 A second definition of the decorated moduli space
Here, following the paper [11], we give a definition of the decorated moduli space in
terms of flat graph connections.
Given an immersed graph ⊂ , a flat graph G-connection on is an iso-
morphism class of flat G-connections on with fixed parallel transports along the
edges of . A graph gauge transformation of a flat graph connection is a usual gauge
transformation modulo gauge transformations relating different representatives of flat
graph connections.
Equivalently, one can define a flat graph G-connection as a representation in G of
the edge-path groupoid of , graph gauge transformations being equivalence trans-
formations of representations.
Suppose that an immersed graph ⊂ is the one-dimensional skeleton of a cell
complex homotopically equivalent to . Then, there is a one-to-one correspondence
between equivalence classes of flat graph G-connections on (with respect to graph
gauge transformations) and moduli of flat connections on . A typical example of a flat
graph G-connection is given by a representation of the fundamental group π1 (, x)
in G, where the immersed graph is given by a single vertex x and the homotopy
classes of simple loops in based at x as edges. Graph gauge transformations in this
case correspond to overall conjugations by elements of G.
We shall confine ourselves to special immersed graphs on associated with ar-
bitrary collections of ideal arcs. Recall that an ideal arc on is a nontrivial isotopy
class of a simple path running between punctures. Let A be the set of ideal arcs.
Define a pairing
I : A × A → Z≥0 , I (e, f ) = min(|a ∩ b| | a ∈ e, b ∈ f )
where |a ∩ b| denotes the number of intersection points of a and b.
ˆ ⊂ be the complement of a disjoint union of small open disks centered at
Let
ˆ is a compact surface homotopically equivalent to
the punctures. The subsurface
with the boundary given by a disjoint union of s circles {L(P )}P ∈V associated with
the punctures

ˆ =
∂ L(P ).
P ∈V
To each collection of ideal arcs α ⊂ A let us choose a collection of representatives

α̃ = {ae | e ∈ α} so that for any ideal arcs e and f in α the arcs ae ∩ ˆ and af ∩ ˆ
are simple paths with |ae ∩ af | = I (e, f ). Such a choice is always possible. It can be
given, for example, by a collection of geodesics with respect to a complete hyperbolic
structure on . The set (α) = ( ˆ ∩ α̃) ∪ ∂
ˆ is an immersed graph in with vertices
∂ ˆ ∩ α̃ and two types of edges: “long” edges corresponding to ideal arcs, and “short”
edges corresponding to segments of the boundary components of ˆ between vertices.
If α ⊂ β, we identify (α) with the corresponding subgraph of (β). Thus, for
any α ⊂ A , (α) is a subgraph of (A ).
Let a and b be two paths having one and the same initial point in and with the
terminal points at some punctures. Let m ∈ M be represented by a flat connection f .
The homotopy class of the path a −1 · b is called m-admissible if the f -holonomies
along the loops (a) and (b) (recall that they are parabolic) do not commute.
It is clear that this notion depends only on the homotopy class [a −1 · b] and the
moduli m = [f ], but not on the choices of their representatives. It is straightforward
to see that the homotopy class [a −1 ·b] is m-admissible if and only if the f -holonomies
along the loops (a) and (b) have distinct fixed points. As a particular case of this
definition, we have a notion of an m-admissible ideal arc.
For α ⊂ A define an open subset M(α) ⊂ M by the condition that for any
m ∈ M(α) the set α is the maximal family of m-admissible ideal arcs. To avoid trivial
cases with M(α) = ∅, we also define
W := {α ⊂ A | M(α) = ∅ } .
It is clear that
M= M(α).
α∈W
We construct a fiber bundle π : M̃ → M as the disjoint union of fiber bundles

π : M̃(α) → M(α) for all α ∈ W . The fiber π −1 (m) over a point m ∈ M(α) consists
of flat graph connections on (α) which represent m and whose parallel transport
operators belong to a fixed unipotent subgroup U for all short edges, and to the subset
θ T , for all long edges, where T is the group N(U )/U identified with a subgroup of
N (U ) through a section of the canonical projection N(U ) → N(U )/U . Let us show
that M̃(α) is a principal T s -bundle over M(α).
Our strategy is to show that any flat connection f , representing a given point
m ∈ M(α), is equivalent to any other flat graph connection in π −1 (m).
As every vertex of (α) belongs to a boundary component of ,ˆ the f -holonomies
around boundary components, based at vertices of (α), are parabolic. Thus, one can
always replace f by such an equivalent connection that all associated holonomies
around the boundary components, based at vertices, belong to one and the same
unipotent subgroup U ⊂ G. This automatically makes the parallel transport operators

along short edges belong to B = N (U ). The parallel transport operators along long
edges belong to the big Bruhat cell BθB due to the condition of admissibility. Thus,
making an appropriate graph gauge transformation with values in B, we can make
the parallel transport operators along all long edges belong to the subset θ T . The
remaining freedom in the graph gauge transformations consists of arbitrary T valued
functions on vertices of (α), which can be used to eliminate the T -parts of the parallel
transport operators along all short edges, thus obtaining elements of π −1 (m). This still
leaves unfixed the graph gauge transformations, given by T valued functions taking
one and the same value at all vertices associated with one boundary component. This
residual gauge group is isomorphic to T s , which acts freely and transitively on the
space π −1 (m). Thus, the union M̃(α) = m∈M(α) π −1 (m) is a principal T s -bundle
over M(α).
2.5 Equivalence of the two definitions
To establish equivalence of the two definitions of the decorated moduli space it is

enough to show that for a given point m ∈ M(α) of the moduli space there exists
a one-to-one correspondence between the collection of gauge classes of connection
horocycles around the punctures on one hand and flat graph connections on (α), on
the other.
Let f be a flat graph connection on (α) representing a point m̃ ∈ M̃(α) of the
decorated moduli space in the sense of the second definition. Then we define an f -
horocycle Hi at the puncture Pi to be represented by a pair (a, h0 ), where a is a path
in the complement − , ˆ connecting a vertex of the graph (α) on the boundary
component L(Pi ) to the puncture Pi , and h0 is the horocycle in the upper half complex
plane given by the equation z = 1. This definition is consistent in the sense that it
defines a unique connection horocycle for each puncture. Thus, the gauge class of the
collection (f, H1 , . . . , Hs ) gives a point in the decorated moduli space in the sense of
the first definition.
Conversely, for each point m̃ of the decorated moduli space given by the first
definition with π(m̃) ∈ M(α) there exists a unique representative (f, H1 , . . . , Hs )
such that the f -holonomies along the edges of (α) correspond to a graph connection
in M̃(α) in the sense of the second definition and all f -horocycles Hi are represented
by the horocycle z = 1 in the upper half complex plane.
2.6 Coordinates for the decorated moduli space
Recall that an ideal triangulation of is a maximal family of pairwise non-intersecting

ideal arcs. The set of ideal triangulations is denoted by . It is a countable infinite
set.
With each ideal triangulation τ we can associate an open dense subset Mτ of

M defined as those moduli of flat connections for which τ is an admissible ideal
triangulation, i.e. with all ideal arcs admissible. This set is given as the following
union
Mτ = M(α), W (τ ) = {α ∈ W | τ ⊂ α}.
α∈W (τ )
Following the paper [11] we describe some of the properties of the subsets Mτ .
Theorem 1 ([11]). Let , M, {Mτ }τ ∈ be as above. Then

(i) the collection {Mτ }τ ∈ is a covering for M;
(ii) there exists a finite sub-covering {Mτ }τ ∈ , ⊂ , || < ∞.
Part (i) of this theorem is equivalent to existence of an m-admissible ideal trian-

gulation for any m ∈ M. Part (ii) implies that there exists a positive integer valued
topological invariant given by the minimal size of finite sub-coverings. One can esti-
mate from above this number by

23κ 4κ
.
2κ + 1 2κ
This is a very rough estimation. For example, in the case of a sphere with four
punctures = 0,4 this formula gives 27 7 = 896 while one can show that there
exists a covering of size 7.
Theorem 2 ([11]). Let , π : M̃ → M, {Mτ }τ ∈ be as above. Then,

(i) for each τ ∈ one has π −1 (Mτ ) = ε∈{±1}2κ M̃ε (τ ) with each M̃ε (τ ) being
a principal Rs>0 -bundle homeomorphic to the complement of ψτ,ε−1 (0) for certain
rational mapping ψτ,ε : R3κ>0 → R (see the end of Section 2.7 for details);

(ii) if σ : {±1} → Z is defined by σ (ε) = 21 2κ
2κ
i=1 εi , then for −κ ≤ k ≤ κ,

the sets M̃k = τ ∈ ε∈σ −1 (k) M̃ε (τ ) are principal Rs>0 -bundles which are
disjoint for different k;
(iii) there exist principal bundle isomorphisms between M̃±κ and the decorated Teich-
müller space T˜ of Penner.
Theorems 1 (i) and 2 (ii) imply that

M= Mk , Mk = M̃k /Rs>0 .
−κ≤k≤κ
Part (iii) of Theorem 2 implies that M±κ , being homeomorphic to Teichmüller

space, are open cells of dimension 3κ − s. The total number of components is 2κ + 1
which formally at s = 0 coincides with the number of connected components of the
moduli space of flat G connections on a closed surface [6]. In our case, however, the
components Mk can be empty or not connected. For example, in the case κ = 1,

M0 = ∅. The non-connectedness of Mk originates from the fact that there are two
distinct conjugacy classes C± of parabolic elements in PSL2 (R) distinguished by the
sign of the off diagonal matrix element when represented by an upper triangular matrix
with trace equal to 2. For example, consider a sphere with four punctures. One can
show that there are two points m, m ∈ M0 such that the holonomies around two fixed
punctures are in one and the same conjugacy class for m and in distinct conjugacy
classes for m . Thus, it is clear that m and m cannot be in one and the same connected
component of the moduli space M. However, taking into account the formal agreement
of the invariant k with Goldman’s relative Euler class, it is natural to expect that for
fixed conjugacy types of the holonomies around the punctures nonempty Mk ’s are
connected.
Part (i) of Theorem 2 extends to M Penner’s coordinate charts for Teichmüller
space. Coordinatization is given by assigning positive real numbers to edges and
signs to faces of an ideal triangulation. The geometrical meaning of the positive real
numbers at edges is identical to the meaning of the Penner coordinates in the decorated
Teichmüller space but adapted to our definitions of the decorated moduli space.
Given m̃ ∈ M̃, m = π(m̃), let a, b be two paths starting at x ∈ and ending
at two punctures P and Q (possibly coinciding) such that the path a −1 · b represents
an m-admissible ideal arc e. Choose a flat connection f representing m and two
horocycles p, q in hyperbolic plane so that the pairs (a, p) and (b, q) represent the
two f -horocycles at P and Q respectively. The m-admissibility of e means that the
horocycles p and q are centered at distinct points u and v of the circle at infinity
of hyperbolic plane. Thus, following Penner, we calculate the signed hyperbolic
distance δ between p and q along the geodesic from u to v taken with positive sign
if p ∩ q = ∅ and with negative sign if p ∩ q = ∅, and we assign to the ideal arc e
the real positive number exp(δ/2). It is clear that this number is independent of the
choices we made: the flat connection f which represents m ∈ M or the pairs (a, p)
and (b, q) representing the two f -horocycles.
To interpret the sign variables associated with ideal triangles, let three paths a, b, c
starting from some x ∈ and ending at three punctures P , Q, R be such that the
paths a −1 · b, b−1 · c, c−1 · a represent three m-admissible ideal arcs which thus
specify an oriented ideal triangle t in with the orientation induced from that of
. On the other hand, the fixed points of the f -holonomies along the loops (a),
(b), (c) are distinct and specify an oriented ideal triangle t in hyperbolic plane with
the orientation induced from that of the hyperbolic plane. Taking into account the
natural correspondence between the vertices of t and t we compare their orientations
and assign to t a positive sign if the two orientations agree and a negative sign if
they disagree. In particular, the part (iii) of Theorem 2 implies that the Teichmüller
component of the moduli space corresponds to putting all signs to one and the same
value, and that corresponds to the fact that the orientations of all ideal triangles in
and the orientations of the corresponding ideal triangles in H simultaneously agree or
disagree.
2.7 Reconstruction of a flat connection from the coordinates

Now, let us show how one can reconstruct a flat connection together with an s-tuple
of connection horocycles starting from a collection of positive real numbers on edges
and signs on faces of an ideal triangulation of .
A point m ∈ M is represented by infinitely many different flat connections on ,
and first we reconstruct a flat graph connection on a two dimensional cell complex
associated to an ideal triangulation.
As any ideal triangulation τ of is a subset of ideal arcs, we can associate to
τ the immersed graph (τ ) ⊂ as in Section 2.4. Due to the definition of an
ideal triangulation, any pair of edges of (τ ) intersect at most at their endpoints,
and thus the graph (τ ) as a subset of is a one-dimensional cell complex. The
connected components of its complement in ˆ are given by hexagons which are in
one-to-one correspondence with ideal triangles of τ . Thus, we obtain a cell complex
decomposition of ˆ associated with τ .
Each short edge of (τ ) belongs to the boundary of a unique hexagon, and the
boundary of each hexagon consists of three short and three long edges in alternating
order. In addition, each short edge has a distinguished orientation induced from that
of the surface . In other words, the short edges can be identified with (oriented)
corners and long edges with sides of ideal triangles of τ .
Given an ideal triangulation τ of , a set of positive real numbers associated with
edges of τ , and a set of signs associated with triangles of τ , we construct a flat graph
G-connection on (τ ) as follows.
To each long edge with corresponding positive real number a we associate a parallel
transport given by the matrix

0 a −1
.
−a 0
Notice, that the square of this matrix is the negative unit matrix which corresponds to
the unit element in the group G and thus the orientation of the edge is irrelevant.
To each short edge with distinguished orientation we associate a parallel transport
given by the matrix

1 −εa/(bc)
0 1
where ε is the sign associated to the unique ideal triangle which has this short edge as
its corner, a being the positive real number associated with this triangle’s side opposite
to the corner under consideration and b, c being the numbers associated with the other
two sides of the triangle.
Let f be a flat connection on representing the flat graph connection we have
just constructed. Each vertex x of (τ ) belongs to the closure of a unique small disk
in the complement − ˆ and thus there is a path ax in this disk having this vertex
as its initial point and the puncture P in its center as its terminal point. This path is
in fact the intersection of the disk with a path representing the ideal arc associated to
x. We define an f -horocycle at P as the equivalence class of the pair (ax , h0 ) where

h0 ⊂ H is the horocycle represented in the upper half complex plane by the horizontal
line z = 1.
Finally, it remains to note that the coordinates should satisfy non-triviality con-
ditions for holonomies around the punctures. Each such holonomy, being a product
of parallel transports along short edges around a puncture, is automatically parabolic
provided the upper off-diagonal element is non-zero. Thus, all the s non-triviality con-
ditions can be given by a single inequality by taking the product of upper off-diagonal
elements of all s holonomies. This is how the rational mapping ψτ,ε in Theorem 2(i)
is constructed.
2.8 The transition functions
Recall that a flip transformation is an elementary transformation from one ideal tri-
angulation of to another obtained by taking an ideal quadrilateral composed of two
distinct ideal triangles, and replacing its diagonal by the opposite diagonal. It is well
known fact that any two ideal triangulations of are related by a finite sequence
of flip transformations [7], [13]. Thus, to describe the transition functions between
coordinate charts associated with two ideal triangulations it is enough to describe the
transition function in the case of a single flip.
Let m̃ ∈ M̃, m = π(m̃), and τ, τ be two m-admissible ideal triangulations related
by a single flip transformation. Then, the geometrical interpretation of the coordinates
given in Section 2.6 implies that the positive real numbers and the signs associated with
ideal arcs common to τ and τ and ideal triangles are identical. Thus, we only need
to describe the formulas relating the coordinates within the quadrilateral associated
with the flip transformation. A way to obtain these formulas is to compare the parallel
transports in this quadrilateral with respect to the two coordinate systems. The result
is as follows.
Let a, b, c, d be the sides of the quadrilateral and e, f its diagonals, with e in τ and
f in τ and the triangles t1 = (a, b, e) and t2 = (c, d, e) in τ and t3 = (d, a, f ) and
t4 = (b, c, f ) in τ , see Figure 1. Then, denoting the positive real numbers associated
r r
c @b c @b
r e@r r f @r
−→
@ @
d @r a d @r a
Figure 1. The flip transformation.
with the edges by the same symbols a, b, . . . and the signs of triangles t1 , t2 , . . . by
ε1 , ε2 , . . . we have the following equations
ε3 ef = ε1 bd + ε2 ac, ε4 ef = ε2 bd + ε1 ac (2.1)
which imply that

ε1 ε2 = ε3 ε4 .
In the case of coinciding signs these equations reduce to the Ptolemy formula of [14]
generating the transition functions for the decorated Teichmüller space.
2.9 Goldman’s symplectic structure

In this parameterization one has an explicit description of the Goldman’s symplectic
structure [5]. The pull-back of the symplectic form to M̃ is given by the same formula
as that of Penner in [15] for the pull-back of the Weil–Petersson form. Namely, if ω
denotes the Goldman’s symplectic form on M, and τ is a given ideal triangulation
then, up to an overall multiplicative factor, one has
da ∧ db db ∧ dc dc ∧ da
π ∗ω = + +
ab bc ca
with the summation over all ideal triangles of τ with side coordinate functions a, b, c
with the cyclic order induced from the orientation of . The drawback of this form
is that it is not a symplectic form, and this prevents us to immediately proceed to
quantization. There are at least two possibilities to overcome this obstacle.
One possibility is to consider only gauge invariant combinations of the coordinate
functions which correspond to coordinates in the moduli space M. For example,
for each quadrilateral with sides a, b, c, d with the cyclic order induced from the
orientation of , the ratio ac/(bd) is gauge invariant. In this way we come to the
shearing coordinates of Bonahon–Thurston and Chekhov–Fock.
Another possibility is to consider the projectivization of M̃, P M̃ = M̃/R>0 , where
the group R>0 is the diagonal subgroup of the structure group Rs>0 , and then to realize
it as a sub-manifold in a larger space with a symplectic structure in such a way that the
moduli space M is obtained by a Hamiltonian reduction. In the case of Teichmüller
space this approach has been materialized in the work [8], where the larger space is
given by the product space P T˜ × H 1 (; R) of even dimension
dim P T˜ + dim H 1 (; R) = (3κ − 1) + (κ + 1) = 4κ
and which has a mapping class group invariant symplectic structure. A work on an
adaptation of this approach to the case of the decorated moduli space is in progress.
In the next section we consider only one aspect of this approach, namely, the flip
transformation and its quantization.
3 Towards the quantum theory

The principal ingredient of the quantum Teichmüller theory is the quantum operator
which realizes the flip transformation, and which among other things satisfies the
pentagonal equation. In the case of the moduli space the flip transformation acts also
on sign variables, so a part of the problem of quantization is the problem of finding
the associated solution of the pentagonal equation. Here we show that the solution of
Woronowicz and Zakrzewski associated with the quantum ‘ax + b’ group [18] gives
rise to a viable quantum theory of the moduli space.
3.1 Quantization of an ideal triangle

The starting point is the following observation. The contribution to the pull-back of
the Goldman’s symplectic form coming from an ideal triangle with side variables a,
b, c can be written in the form
da ∧ db db ∧ dc dc ∧ da du ∧ dv
+ + = (3.1)
ab bc ca uv
where
u = a/c, v = b/c. (3.2)
The definition of the variables u, v is not unique as it does not respect the cyclic
symmetry of the side variables. One can fix this ambiguity by choosing a distinguished
corner in the triangle, see Figure 2. In this way, taking also into account the sign
r
c @b
r @
∗r
a
Figure 2. An ideal triangle with a distinguished corner.
variables, we associate to each ideal triangle with a distinguished corner a triple of

variables (u, v, ε) taking values in R2>0 × {±1}.
In quantum theory, motivated by equation (3.1), we associate to each ideal triangle
with a distinguished corner a unitary C-algebra A, called the quantum triangle alge-
bra, which contains a triple of elements (u, v, e) where a pair of invertible elements
u, v (quantizations of the variables u, v) satisfies the “quantum plane” commutation
relation
uv = q 2 vu
with q ∈ C − 0 being a quantization parameter, while element e (a quantization of the
sign variable ε) satisfies the relations
e2 = 1, ue = eu, ve = ev.
The minimal algebra satisfying these conditions is the algebra generated by these three
elements, but for the reasons which will become clear later on, we shall choose for
actual quantum triangle algebra a bigger algebra. In quantum Teichmüller theory the
sign operator is trivial and one takes for A the algebra End(L2 (R)) of linear operators
in the Hilbert space L2 (R) of square integrable functions f (x) on the real axis with
respect to the standard Lebesgue measure with the realizations
u = e2πbq , v = e2πbp .
Here b ∈ C, p and q are operators satisfying the Heisenberg commutation relation

pq − qp = (2π i)−1 , (3.3)
so the initial quantization parameter takes the form q = eiπb . Using the fact that the
2
classical variables u, v are real and positive it is assumed that b ∈ R>0 and p, q are
self-adjoint operators realized as differentiation and multiplication operators:
1 df (x)
(qf )(x) = xf (x), (pf )(x) = . (3.4)
2π i dx
To take into account the sign degrees of freedom, we take for the quantum triangle
algebra A the algebra End(L2 (R) ⊗ C2 ) with the realizations

1 0
u = e2πbq ⊗ idC2 , v = e2πbp ⊗ idC2 , e = idL2 (R) ⊗ . (3.5)
0 −1
This algebra has the property that if an element x ∈ A commutes both with p ⊗ idC2
and q ⊗ idC2 then there exists unique x ∈ End(C2 ) such that x = idL2 (R) ⊗ x.
In order to simplify the notation, in what follows we shall identify elements of the
algebras End(L2 (R)) and End(C2 ) with their images in A by the canonical inclusions.
To complete the quantization of a triangle we have to restore the broken cyclic
symmetry of the side variables (a, b, c). This means that we have to construct an
inner automorphism of the quantum triangle algebra corresponding to a change of
the distinguished corner. Namely, consider the transformation drawn in Figure 3.
Classically, the new variables (û, v̂, ε̂) corresponding to the right hand side of Figure 3
r r
c @b −→ c ∗@ b
r @
∗r r @r
a a
Figure 3. A change of the distinguished corner of an ideal triangle.
are given by the formula

(û, v̂, ε̂) = (b/a, c/a, ε) = (v/u, 1/u, ε)
where we have used the definitions (3.2). The quantum version of this transformation
has the form
(û, v̂, ê) = (q u−1 v, u−1 , e), (3.6)
which corresponds to
(q̂, p̂) = (p − q, −q).
Now, there exists a unique up to multiplicative factor unitary operator A ∈ End(L2 (R))
which realizes the latter transformation by conjugation:
A(q, p)A−1 = (p − q, −q).
Explicitly we have
A = e−iπ/3 ei3π q eiπ(p+q)
2 2
(3.7)
where the normalization constant is chosen so that
A3 = 1. (3.8)
Thus, the operator A considered as an element of A realizes by conjugation the trans-
formation (3.6).
3.2 Quantization of the flip transformation

Recall that if a triangulated quadrilateral has side variables a, b, c, d and a diagonal
variable e so that it is decomposed into two triangles t1 = (a, b, e) and t2 = (c, d, e),
then the flip transformation replaces the diagonal variable e by another diagonal vari-
able f and the triangles t1 , t2 by triangles t3 = (d, a, f ) and t4 = (b, c, f ), see Fig-
ure 1, with the side and the sign variables transforming according to equations (2.1).
The choice of the distinguished corners for the four triangles as in Figure 4 corresponds
to the following identifications:
(u1 , v1 ) = (a/e, b/e), (u2 , v2 ) = (e/d, c/d),
(u3 , v3 ) = (a/d, f/d), (u4 , v4 ) = (b/f, c/f ),
so that the triple (ui , vi , εi ) is associated with the triangle ti for 1 ≤ i ≤ 4. The flip
r r
c ∗@b c ∗ t@ b
4
r t2 e t1@ ∗ r −→ r f @ ∗r
@ @ t3
d@r a d@r a
Figure 4. A choice of distinguished corners for the triangles t1 , . . . , t4 .
transformation implies the following relations:

u3 = u1 u2 , v2 = v3 v4 , v3 u4 = v1 u2 , ε3 v3 = ε1 v1 + ε2 v2 u1 , ε1 ε2 = ε3 ε4
(3.9)
which are consistent with the equation
dui ∧ dvi dui ∧ dvi
= .
ui vi ui vi
i=1,2 i=3,4
In quantum theory, to each ideal triangle of a given ideal triangulation we associate

a copy of the quantum triangle algebra A in such a way that copies corresponding to
different ideal triangles mutually commute. Thus, in the case of the quadrilateral under
consideration, the quantum algebras associated to the two triangulations are given by
the tensor product algebra A ⊗ A with tensor components corresponding to quantum
triangle algebras associated with the triangles t1 and t2 for one triangulation, and with
the triangles t3 and t4 for another triangulation. Here, in the case of L2 -spaces the
tensor product is defined by
End(L2 (X)) ⊗ End(L2 (Y )) := End(L2 (X × Y )).
The quantum flip transformation by definition is an isomorphism between these two

quadrilateral algebras. We assume that the same formulae (3.9) are valid on the
quantum level. The quantum flip transformation operator T ∈ A ⊗ A is defined as a
unitary operator realizing the quantum flip transformation by conjugation,
T(ui , vi , ei )T−1 = (ui+2 , vi+2 , ei+2 ), i ∈ {1, 2},
where
(u1 , v1 , e1 ) = (u ⊗ 1, v ⊗ 1, e ⊗ 1), (u2 , v2 , e2 ) = (1 ⊗ u, 1 ⊗ v, 1 ⊗ e).
Equivalently, we have the following system of linear equations:
Tq ⊗ 1 = (q ⊗ 1 + 1 ⊗ q)T (3.10)
T(p ⊗ 1 + 1 ⊗ p) = 1 ⊗ pT (3.11)
T(p ⊗ 1 + 1 ⊗ q) = (p ⊗ 1 + 1 ⊗ q)T (3.12)
Te 2πbp
e ⊗ 1 = (e2πbp e ⊗ 1 + e2πbq ⊗ e2πbp e)T (3.13)
Te ⊗ e = e ⊗ eT. (3.14)
Proposition 2. Any solution of the system of equations (3.10)–(3.14) has the form
T = e2π ip⊗q F (3.15)
where
F = b (e⊗e, r ⊗r, q⊗1+1⊗(p−q))+e⊗1b (e⊗e, r ⊗r, q⊗1+1⊗(p−q)) (3.16)
with b , b : {±1}2 × C0≤≤b → C being functions which are analytic in {±1}2 ×
C0<<b and for any x ∈ R satisfying one and the same functional equation
b (ε, −ρ, x + ib) = (1 + εqe2πbx )b (ε, ρ, x) (3.17)
and r ∈ A is any element satisfying the relations
r2 = 1, rq − qr = rp − pr = re + er = 0. (3.18)
Proof. If we define F = e−2π ip⊗q T, then the system (3.10)–(3.14) simplifies to

Fq ⊗ 1 = q ⊗ 1F
F(p ⊗ 1 + 1 ⊗ p) = (p ⊗ 1 + 1 ⊗ p)F
F(p ⊗ 1 + 1 ⊗ q) = (p ⊗ 1 + 1 ⊗ q)F
Fe 2πbp
e ⊗ 1 = e2πbp e ⊗ 1(1 + qe2πb(q⊗1+1⊗(p−q)) e ⊗ e)F
Fe ⊗ e = e ⊗ eF
which implies that F must be of the form given by formula (3.16) with b , b satisfying
equation (3.17). The analyticity properties of b , b follow from the fact that a proper
definition of conjugation by a positive operator a in a Hilbert space (in our case such
positive operator is of the form e2πbp ⊗ 1) is the analytic continuation to t = −i of
conjugation by the associated family of unitary operators {a it | t ∈ R}.
3.3 A solution by Woronowicz and Zakrzewski

The functional equation (3.17) can be solved by using Faddeev’s non-compact quantum
dilogarithm function ϕb (z) [3] defined by

e−i2zw dw
ϕb (z) := exp −1
(3.19)
i0+R 4 sinh(wb) sinh(wb )w
in the strip |z| < (b + b−1 )/2 and analytically continued to the whole complex plane
through the following pair of functional equations:
±1
ϕb (z − ib±1 /2) = ϕb (z + ib±1 /2) 1 + e2πb z . (3.20)
It is a meromorphic function with poles given by the set
P := {ib(m + 1/2) + ib−1 (n + 1/2) | m, n ∈ Z≥0 } (3.21)
and zeros given by the set −P . It satisfies the “inversion relation”
2 +b−2 )/24
ϕb (z)ϕb (−z) = ϕb (0)2 eiπz , ϕb (0) = eiπ(b
2
(3.22)
and has the “unitarity property” with respect to complex conjugation
ϕb (z) = 1/ϕb (z). (3.23)
Proposition 3. For any α ∈ {±1} the function

1 + iαρeπz/b (1 − ε)/2
b (ε, ρ, z) = (3.24)
ϕb (z − ib−1 (1 − ε)/4)
has the following properties:
(i) it is analytic in the domain −b/2 < (z);
(ii) it solves the functional equation (3.17);
(iii) it satisfies the “inversion relation”

e−iπz
2
−2
b (ε, ρ, z)b (ε, ρ, −z) = 1 + ε + iαeiπb /4 (1 − ε)ρ ; (3.25)
2ϕb (0) 2
(iv) it has the “unitarity property” with respect to complex conjugation

b (ε, ρ, z) = 1/b (ε, ρ, z). (3.26)
Proof. (i) The function 1/ϕb (z) is analytic in the domain −(b + b−1 )/2 < (z) while
the function 1/ϕb (z − ib−1 /2) is so in the domain −b/2 < (z). The intersection of
these two domains is the domain of analyticity of the function b (ε, ρ, z).
(ii) Using the functional equation (3.20) with a plus sign, we have
b (ε, −ρ, x + ib) ϕb (x − ib−1 (1 − ε)/4) −1
= −1
= 1 + e2πb(x+ib/2−ib (1−ε)/4)
b (ε, ρ, x) ϕb (x − ib (1 − ε)/4 + ib)
= 1 + qe2πbx e−iπ(1−ε)/2 = 1 + εqe2πbx .
(iii) The case where ε = 1 is equivalent to the inversion relation (3.22). For
ε = −1, by using the functional equation (3.20) with a minus sign and the inversion
relation (3.22), we have
(1 + iαρeπz/b )(1 + iαρe−πz/b )
b (−1, ρ, z)b (−1, ρ, −z) =
ϕb (z − ib−1 /2)ϕb (−z − ib−1 /2)
(1 − iαρe−πz/b )(1 + iαρe−πz/b )
= iαρeπ z/b
ϕb (z − ib−1 /2)ϕb (−z − ib−1 /2)
iαρeπz/b (1 + e−2πz/b )
=
ϕb (z − ib−1 /2)ϕb (−z − ib−1 /2)
iαρeπz/b
=
ϕb (z − ib−1 /2)ϕb (−z + ib−1 /2)
−2
iαρeπz/b iαρeiπb /4
= −1 /2)2
= .
ϕb (0)2 eiπ(z−ib ϕb (0)2 eiπz
2
(iv) The case where ε = 1 is equivalent to equation (3.23). For ε = −1, by using
(3.23) and (3.20) with a minus sign, we have
ϕb (z + ib−1 /2)
b (−1, ρ, z)b (−1, ρ, z) = (1 − iαρeπz/b )(1 + iαρeπz/b )
ϕb (z − ib−1 /2)
1 + e2πz/b
= = 1.
1 + e2πz/b 2
The solution (3.24) is found by Woronowicz [17]. Namely, if Fh̄ (r, ρ) is Woronow-
icz’s quantum exponential function (see formulae (1.18), (1.19) of [17]), then
b (ε, ρ, z) = F2πb2 (εe2πbz , αρ(1 − ε)/2).
This solution, together with b = 0, leads to a unitary operator (3.15) when r is self-
adjoint. A deep result of Woronowicz and Zakrzewski (Theorem 2.1 of [18]) implies
the following
Theorem 3 ([18]). Let b be such that

−2
ieiπb /4 = α = ±1 (3.27)
and let b (ε, ρ, z) be defined by equation (3.24). Then, the unitary operator
T = e2π ip⊗q b (e ⊗ e, r ⊗ r, q ⊗ 1 + 1 ⊗ (p − q)) (3.28)
satisfies the following pentagonal equation in A⊗3 :
T12 T13 T23 = T23 T12 . (3.29)
√
Condition (3.27) implies that b = 1/ 4k + 2 for some k ∈ Z≥0 , and thus q =
eiπ/(4k+2) is a root of unity.
The solution (3.28) of the pentagonal equation has an additional property corre-
sponding to the symmetry with respect to changing the distinguished corners of ideal
triangles.
Proposition 4. Let b and T be as in Theorem 3 and let A ∈ A be defined by equa-

tion (3.7). Then, the following equation holds in A⊗2 :
TA ⊗ 1PT = ζ A ⊗ A (3.30)
where P ∈ A⊗2 is the permutation operator and ζ = e −iπ(b+b−1 )2 /12 .
Proof. Equation (3.30) is equivalent to the identity

e2π ip⊗q e2π i(p−q)⊗p b (e ⊗ e, r ⊗ r, p ⊗1−1⊗ q)b (e ⊗ e, r ⊗ r, 1⊗q − p ⊗1) = ζ A ⊗1P
which, by using equations (3.25), (3.27), simplifies to
1
e2π ip⊗q e2π i(p−q)⊗p e−iπ(p⊗1−1⊗q) ϕb (0)−2 (−1)mn (e⊗e)m (r⊗r)n = ζ A⊗1P.
2
2
m,n∈{0,1}
By definition of the algebra A, the permutation operator P has the form

P = σL2 (R2 ) σC4
where for any vector space V the operator σV ⊗2 ∈ End(V ⊗2 ) is the corresponding
permutation operator defined by σV ⊗2 (x ⊗ y) = y ⊗ x for any x, y ∈ V . One can
verify by an explicit and straightforward calculation that
1
σC4 = (−1)mn (e ⊗ e)m (r ⊗ r)n
2
m,n∈{0,1}
and
σL2 (R2 ) = eiπ/6 A−1 ⊗ 1e2π ip⊗q e2π i(p−q)⊗p e−iπ(p⊗1−1⊗q) .
2
Thus, equation (3.30) is true provided ζ = ϕb (0)−2 e−iπ/6 .
Equations (3.8), (3.29), (3.30) permit to construct a projective representation of

the groupoid of decorated ideal triangulations of the surface in the space A⊗2κ [9].
In particular, we obtain a unitary projective representation of the mapping class group
of .
3.4 Spectrum of the length operator

The purpose of this section is to illustrate how different the quantum theory of the
whole moduli space could be compared to that of the Teichmüller space.
Let us consider an annulus given by two ideal triangles with side coordinate func-
tions a, b, c and d, b, c with boundary edges given by a and d, and sign variables ε1
and ε2 . The contribution to the pull-back of the Goldman’s symplectic form coming
from these two triangles can be written in the form
dx ∧ dy
xy
where x = b/c and y = ad/(bc) are two gauge invariant combinations of the coor-
dinate functions. The same gauge invariant combinations of the coordinate functions
enter the trace of the holonomy corresponding to the only generator of the fundamental
group of this annulus:
Lε = x + x −1 + εy
where ε = ε1 ε2 . As log(x) and log(y) are Poisson conjugate variables, the quantum
version of this trace function leads to the following two operators
L± = 2 cosh(2πbq) ± e2πbp
corresponding to two possible values ε = ±1. Here again operators p and q are
self-adjoint operators in a Hilbert space satisfying the Heisenberg commutation re-
lation (3.3). In particular, in the representation (3.4) the exponential operator e2πbp
becomes a finite shift operator of the form
(e2πbp f )(x) = f (x − ib).
To make sense of such pure imaginary shifts, one has to assume that the domain
D(e2π bp ) of this operator consists of functions f (z) defined in the strip −b ≤ z ≤ 0,
analytic in its interior, and with square integrable restrictions to the boundary compo-
nents z = 0 and z = −b.
In quantum Teichmüller theory all sign variables take one and the same value
and thus one deals with only the operator L+ , while L− appears in other connected
components of M. The principal difference between these two non-bounded operators
is that L+ is self-adjoint while L− is only symmetric but not self-adjoint.
Theorem 4 ([10]). The spectrum Sp(L+ ) of the operator L+ consists of only a con-
tinuous part which is given by a subset of positive real numbers:
Sp(L+ ) = Spc (L+ ) = R≥2 .
In the Teichmüller component the absolute value of the trace of the holonomy hγ
along a closed contour γ is related to the length lγ of the geodesic isotopic to γ through
the formula
|Tr(hγ )| = 2 cosh(lγ /2)
Thus, the above result means that the quantum spectrum Sp(lγ ) of the length operator
coincides with the set R≥0 of its possible classical values.
To solve the spectral problem of L− one has to study its self-adjoint extensions.
Theorem 5 ([12]). For any θ ∈ [0, 1[ such that θ + b−2 /4 = ∈ Z, the operator
L− admits a self-adjoint extension Lθ , L− ⊂ Lθ ⊂ L−∗ , with the spectrum Sp(L )
θ
consisting of a continuous part
Spc (Lθ ) = R≥2
and a point part

Spp (Lθ ) = −2 cosh 2π b θ − Z< .
Appearance of a point (discrete) part in the spectrum is an interesting property

which presumably reflects the non-trivial topology of the moduli space.
References
[1] L. O. Chekhov, V. V. Fock, A quantum Teichmüller space. Theoret. and Math. Phys. 120
(1999), 1245–1259. 761
[2] L. O. Chekhov and R. C. Penner, On quantizing Teichmüller and Thurston theories.
In Handbook of Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing
House, Zurich 2007, 579–645. 761
[3] L. D. Faddeev, Discrete Heisenberg–Weyl group and modular group. Lett. Math. Phys. 34
(1995), 249–254. 777
[4] V. V. Fock, A. B. Goncharov, Dual Teichmüller and lamination spaces. In Handbook of
Teichmüller theory (A. Papadopoulos, ed.), Volume I, EMS Publishing House, Zurich
2007, 647–684. 761
(1984), 200–225. 772
[6] W. M. Goldman, Topological components of spaces of representations. Invent. Math. 93
(1988), 557–607. 762, 768
[7] J. Harer, The virtual cohomological dimension of the mapping class group of an oriented
surface. Invent. Math. 84 (1986), 157–176. 771
Math. Phys. 43 (1998), 105–115. 761, 772
[9] R. M. Kashaev, Liouville central charge in quantum Teichmüller theory. Proc. Steklov
Inst. Math. 226 (1999), 63–71. 780
[10] R. M. Kashaev, The quantum dilogarithm and Dehn twists in quantum Teichmüller theory.
In Integrable structures of exactly solvable two-dimensional models of quantum field the-
ory (Kiev, 2000), NATO Sci. Ser. II Math. Phys. Chem. 35, Kluwer Academic Publishers,
Dordrecht 2001, 211–221. 781
[11] R. M. Kashaev, Coordinates for the moduli space of flat PSL(2, R)-connections. Math.
Res. Lett. 12 (1) (2005), 23–36. 762, 765, 768
[12] R. M. Kashaev, On selfadjoint extensions of some difference operator. St. Petersburg Math.
J. 17 (2006), 157–167. 781
[13] L. Mosher, Mapping class groups are automatic. Ann. of Math. 142 (1995), 303–384. 771
113 (1987), 299–339. 764, 772
[15] R. C. Penner, Weil–Petersson volumes. J. Differential Geom. 35 (1992), 559–608. 772
[16] J. Teschner, Liouville theory revisited. Classical Quantum Gravity 18 (23) (2001),
R153–R222. 761
[17] S. L. Woronowicz, Quantum exponential function. Rev. Math. Phys. 12 (6) (2000),
873–920. 762, 778
[18] S. L. Woronowicz, S. Zakrzewski, Quantum ‘ax + b’ group. Rev. Math. Phys. 14 (7–8)
(2002), 797–828. 762, 773, 779
List of Contributors
Charalampos Charitos, Laboratory of Mathematics, Agricultural University of Athens,

75 Iera Odos, Athens, Greece; email: bakis@aua.gr
Leonid O. Chekhov, Steklov Mathematical Institute, Gubkina 8, 117966, GSP–1,
Moscow, Russia; email: chekhov@mi.ras.ru.
Georgios D. Daskalopoulos, Department of Mathematics, Brown University, Provi-
dence, RI 02912, U.S.A.; email: daskal@math.brown.edu
Vladimir V. Fock, Institute for Theoretical and Experimental Physics, B.Cheremush-
kinskaya 25, 117259 Moscow, Russia; email: fock@math.brown.edu
Alexander B. Goncharov, Brown University, Box 1917, Providence, RI 02906, U.S.A.;
email: sasha@math.brown.edu
Ursula Hamenstädt, Mathematisches Institut, Universität Bonn, Beringstraße 1, 53115
Bonn, Germany; email: ursula@math.uni-bonn.de
William J. Harvey, Department of Mathematics, King’s College London, Strand, Lon-
don WC2R 2LS, United Kingdom; email: bill.harvey@kcl.ac.uk
Frank Herrlich, Mathematisches Institut II, Universität Karlsruhe, 76128 Karlsruhe,
Germany; email: herrlich@math.uni-karlsruhe.de
Rinat M. Kashaev, Université de Genève, Section de mathématiques, 2-4, rue du
Lièvre, CP 64, 1211 Genève 4, Suisse; email: Rinat.Kashaev@math.unige.ch
Albert Marden, School of Mathematics, University of Minnesota, Minneapolis,
MN 55455, U.S.A.; email: am@umn.edu
Christian Mercat, Institut de Mathématiques et de Modélisation de Montpellier (I3M
UMR CNRS 5149), Université Montpellier II – CC 51, 34095 Montpellier cedex 5,
France; email: mercat@math.univ-montp2.fr
Shigeyuki Morita, Graduate School of Mathematical Sciences, University of Tokyo,
Komaba, Tokyo 153-8914, Japan; email: morita@ms.u-tokyo.ac.jp
Lee Mosher, Department of Mathematics and Computer Science, Rutgers University,
Newark, NJ 07102, U.S.A.; email: mosher@andromeda.rutgers.edu
Jean-Pierre Otal, Laboratoire Painlevé, Université des Sciences et Techniques de
Lille 1, 59655 Villeneuve d’Ascq cedex, France; email: otal@picard.ups-tlse.fr
Ioannis Papadoperakis, Laboratory of Mathematics, Agricultural University of Athens,
75 Iera Odos, Athens, Greece; email: papadoperakis@aua.gr
Athanase Papadopoulos, Institut de Recherche Mathématique Avancée, CNRS et Uni-
versité Louis Pasteur, 7 rue René Descartes, 67084 Strasbourg cedex, France; email:
papadopoulos@math.u-strasbg.fr
784 List of Contributors
Robert C. Penner, Departments of Mathematics and Physics/Astronomy, University of

Southern California, Los Angeles, CA 90089, U.S.A.; email: rpenner@math.usc.edu
Gabriela Schmithüsen, Mathematisches Institut II, Universität Karlsruhe, 76128 Karls-
ruhe, Germany; email: schmithuesen@math.uni-karlsruhe.de
Jörg Teschner, Institut für theoretische Physik, Freie Universität Berlin, Arnimallee 14,
14195 Berlin, Germany, and Max Planck Institut für Gravitationsphysik (Albert Ein-
stein Institut), Am Mühlenberg 1, 14476 Potsdam, Germany; teschner@mail.desy.de
Guillaume Théret, Institut de Recherche Mathématique Avancée, CNRS et Univer-
sité Louis Pasteur, 7 rue René Descartes, 67084 Strasbourg cedex, France, email:
theret@math.u-strasbg.fr
Marc Troyanov, École Polytechnique Féderale de Lausanne (EPFL), Institut de Géomé-
trie, Algèbre et Topologie, 1015 Lausanne, Switzerland; email: marc.troyanov@epfl.ch
Richard A. Wentworth Department of Mathematics, Johns Hopkins University, Balti-
more, MD 21218, U.S.A.; email: wentworth@jhu.edu
Index
absolute Galois group, 285 Bochner formula, 50
accidental parabolic element, 320 boundary, 477
acylindrical manifold, 413 Bers, 431
admissible complex, 96 Teichmüller, 138
admissible homotopy, 511 Thurston, 193
admissible path, 511 visual, 141
affine deformation, 302, 305 boundary incompressible, 413
affine diffeomorphism, 311 BP-map, 368
affine group, 312 Braungardt’s construction, 298, 315
Ahlfors conjecture, 412 BSCC-map, 368
Ahlfors Finiteness Theorem, 413
Ahlfors–Bers map, 228 canonical dissection, 548
algebraic convergence of groups, 426 cataclysm coordinates, 155
cellular decomposition, 543
anti-stretch line, 164
character, 528
anti-stretch ray, 164
character variety, 38, 72
approximating sequence, 623
cilium, 649
Arakelov–Paršin theorem, 94
circle pattern, 570
arational measured foliation, 140, 391
coarse Hausdorff topology, 464
asymmetric metric, 165
coboundary, 524
asymptotic rays, 141
cocycle, 524
augmented Teichmüller space, 298, 314,
collar lemma, 79
318
collar map, 656, 658
automorphism groups of free groups, 361
collar weight, 620
collision, 613
Bäcklund transformation, 571 combination theorems, 407
balancing condition, 97 compact core of a manifold, 415
Beltrami coefficient, 39, 224, 258 relative, 415
Beltrami differential, 40, 226, 309 comparison map, 477
Beltrami equation, 40, 224, 259 comparison triangle, 477
Beltrami harmonic differential, 45 complete geodesic lamination, 148, 449
Belyi function, 255 complete train track, 453
clean, 255 completion of a geodesic lamination, 148
Bers boundary, 431 complex earthquake, 438
Bers fiber-space theorem, 265 complex of curves, 401, 447
Bers slice, 430, 431 complex scaling, 435
extended, 431 conformal blocks, 687, 688
Bers universal Teichmüller space, 213 conformal energy, 547
bilinear relations, 552 conformal field theory, 688, 690, 723
786
conformal structure, 120, 123 decorated moduli space, 762

discrete, 543 decorated Teichmüller space, 210
conformal structure with boundary, 124 decorated Teichmüller space of the
conformal structures solenoid, 217
equivalent, 123 decorated universal Teichmüller space,
continuous limit, 566 212
convergence in the pullback sense, 74 deformation map, 298, 319
convergent ray, 139 deformation of Kleinian groups, 425
convex cocompact, 88 deformation of surface groups, 429
convex cocompact subgroup, 400 deformation space boundary, 427
convex core, 660 degenerate hyperbolic pair of pants, 144
convex core of a manifold, 419 degenerate right-angled hexagon, 145
Sullivan’s Theorem on the, 441 degree of a complex bundle, 70
Corlette–Donaldson theorem, 64, 71 Dehn twist, 363, 705, 740
covering with cusps, 316 Dehn–Nielsen theorem, 358
universal, 316 Deligne–Mumford compactification, 297
critical graph, 119, 131 δ-hyperbolic, 448
critical trajectory, 57, 266 ∂¯ operator, 560
criticality, 556 dessin, 255
cross ratio, 342, 696, 697 balanced, 256
cross ratio preserving map (discrete), 569 clean, 256
curvature less than or equal to, 477 Galois, 256
curve graph, 447 developing map, 513
cusp, 142, 265 diffeotopy group, 355
on ∂T(G), 429 dilatation, 39, 126, 301
on Bers boundary, 431 Dirichlet energy, 547
rank one, 414 discrete area, 547
double horodisk, 414 discrete Cauchy–Riemann equation, 543
solid cusp cylinder, 413 discrete conformal structure, 543
solid pairing tube, 414 discrete derivative, 557
rank two, 414 discrete exponential, 563
solid cusp torus, 413 discrete Green function, 574
cusp points, 519 discrete Hodge star, 544
cusped annulus, 473 discrete holomorphic function, 543
cut system, 341 discrete Laplacian, 544
cylinder discrete polynomial, 564
core curve of, 335 discrete quasiconformal map, 562
horizontal, 323 discrete Riemann surface, 542
maximal, 331 discrete series, 564
discrete wedge product, 546
decent neighbourhood, 317 Disk Theorem, 441
decorated measured foliation, 620 distance parameters, 488
787
distinguished oriented edge of a Fenchel–Nielsen flow, 195

tesselation, 211 Fermat curves, 272
distinguished oriented edge of filling curves, 88
Farey tesselation, 209 filling foliations, 89
distinguished parameter of a fine moduli space for Riemann surfaces
quadratic differential, 130 with Schottky structure, 338
domain of discontinuity, 399 flat connection, 763
double annulus, 326 flat graph connection, 765
double cusped disc, 473 flat structure, 301
doubly cusped region, 317, 320 flip transformation, 650, 771
duality, 543 flux homomorphism, 379
Fock coordinates, 695, 699, 702
earthquake, 432 foliation
earthquake disk, 441
horizontal, 40, 130, 136
Earthquake Theorem, 433
measured, 116
Eells–Sampson theorem, 49
partial, 118
Eilenberg–MacLane space, 358
vertical, 40, 130, 136
energy, 47
freeway, 619
first variation of, 48
Fricke space, 38
second variation of, 48
F -structure, 266
energy and holomorphic maps, 48, 93
Fuchsian deformation, 225
energy density, 47, 60
Fuchsian group, 38, 224, 659
energy extremal problem, 47
deformation of, 40
energy minimizer, 47
Fuchsian uniformization, 695, 697, 705
equivalent conformal structures, 123
equivalent hyperbolic structures, 142
Gardiner’s formula, 69
essential curve, 118, 389
generalized hyperbolic pair of pants, 145
essentially simple, 498
geodesic, 136, 476, 495
euclidean triangulation, 510
essentially simple, 498
extending boundary maps, 416
geodesic current, 230
extremal length, 128
geodesic Hölder distribution, 232
extremal length of a homotopy class
of curves, 128 geodesic lamination, 147, 449
extremal length of a measured complete, 148
foliation class, 129 completion of, 148
extremal map, 39, 267 maximally stretched, 166
measured, 150
family of Riemann surfaces, 337 geodesic length, 660, 704, 705
Farb–Kaimanovich–Masur theorem, 95 geodesic length operators, 690, 691, 736
Farey tesselation, 208 geodesic line, 136, 476
fat graph, 584 geodesic metric space, 136
fat graphs, 693, 698, 701, 707, 731 geodesic operator, 585
Fenchel–Nielsen coordinates, 146, 475, geodesic ray, 136, 303, 309, 322, 476
690, 693, 703, 720 geodesic segment, 476
788
geometric convergence of groups, 428 Higgs bundle and Teichmüller’s theorem,

geometric intersection, 133 73
geometrically finite group/manifold, Higgs bundle and the Milnor–Wood
414 inequality, 73
geometrically infinite group/manifold, hinge, 512
414 Hirota system, 570
Gerstenhaber–Rauch conjecture, 56 Hitchin map, 72
gluing maps, 494 Hitchin–Simpson theorem, 70
Goldman’s symplectic structure, 772 holomorphic form
graph length, 623 discrete, 544
graph simple path, 591 holomorphic function
Gromov boundary, 464 discrete, 543
Gromov hyperbolic, 391, 448 holomorphic motion, 239, 442
holomorphic quadratic differential, 450
Gromov–Hausdorff convergence, 74
admissible, 302
group
critical point of, 300, 302
handlebody, 344
regular point of, 300, 302
Hecke, 270
holonomy, 514
Möbius, 209
homeotopy group, 359
mapping class, 81, 123, 253, 260, 294, Hopf differential, 51
312, 339 horizontal foliation, 130, 136
modular, 209 horocycle, 664
group cohomology, 524 horocycle topology, 315
groupoid, 677 horocyclic area map, 665, 668
horocyclic foliation, 497
h.s.c.s., 478 horocyclic measured foliation, 152
horocyclic measured geodesic lamination,
handlebody group, 344
153
harmonic map, 47, 61
horogeodesic path, 158
combinatorial, 98
minimal, 159
equivariant, 63
Hubbard–Masur theorem, 58
existence of, 49, 64
hyperbolic metric space, 392
order of, 62 hyperbolic pair of pants, 144, 473
uniqueness of, 51 degenerate, 144
Hausdorff distance, 449 generalized, 145
Hausdorff topology, 449 hyperbolic structure, 142
Hecke group, 270 ideal, 494
height of a measured foliation with hyperbolic surface with conical
one cylinder, 119 singularities, 478
Higgs bundle, 69 Hyperbolization Theorem, 421
moduli space of, 71
polystable, 70 IA automorphism group, 362
stability of, 70 ideal boundary, 64
789
ideal hyperbolic metric, 494 stretch, 164

ideal hyperbolic structure, 494 line of minima, 200
ideal triangulation, 146, 210 Liouville map, 231
intersection form, 450 Liouville measure, 231
intersection graph, 297 Liouville theory, 687
intersection number, 548 local quasiconformal dilatation, 126
irreducible subgroup, 397 locally ringed space, 315
Ishihara’s theorem, 49 lower central series, 371
isothermal coordinates, 38
isotrivial, 93 Malcev completion, 372
map
Jenkins–Strebel differential, 58, 278, 322 Hitchin, 72
Jenkins–Strebel geodesic, 137 quasiconformal, 39, 126, 562
Jenkins–Strebel ray, 298, 323 quasisymmetric, 213, 259
end point of, 299, 328 stretch, 164
Jenkins–Strebel rays Teichmüller, 134
modularly equivalent, 138 mapping class
similar, 137 as outer automorphism, 81
Jenkins–Strebel structure, 131 as Weil–Petersson isometry, 83
Johnson homomorphism, 371 finite order, 393
Johnson subgroup, 369 pseudo-Anosov, 83, 392, 394
Kashaev coordinates, 700–702, 708 pseudoperiodic, 84
Kleinian group, 412 reducible, 83, 392, 394
Knizhnik–Zamolodchikov connection, semisimple, 84
688 Thurston classification, 83, 392
Korevaar–Schoen limits, 75 mapping class group, 81, 123, 253, 260,
294, 312, 339, 357, 387, 448, 521,
lambda length, 206 589, 651, 676, 687, 700, 702
λ-Lemma, 442 classification of subgroups, 91
lamination, 652 superrigidity of, 95
bounded, 656 mapping class group representation, 687
even, 658 marked Riemann surface, 294
integral, 655, 656 marking, 38, 703, 720, 731
real, 659 Markov numbers, 679
unbounded, 652 maximal dilatation, 301
lattice in PSL2 (R), 296 maximal geodesic lamination, 449
Laurent property, 674 measure-equivalence, 119
leaf, 147 measured foliation, 57, 116, 390, 606
length functional, 143 arational, 140, 391
length space, 56 extremal length of a, 68
limit set, 224, 398 horocyclic, 152
line intersection number, 59
anti-stretch, 164 projective equivalence, 58
790
totally transverse, 154 coarse, 295

uniquely ergodic, 140 compactification of, 297, 318
measured foliation class, 117 fine, 295
extremal length of a, 129 moduli space of flat connections, 763
measured foliation with one cylinder, 119 moduli space of metric graphs, 378
height of a, 119 moduli space of Riemann surfaces, 377
measured geodesic lamination, 150, moduli spaces, 688
450 modulus, 124, 463
horocyclic, 153 modulus of a cylinder, 128
length of a, 151 modulus of a homotopy class of curves,
minimal, 194 128
projective, 151 monodromy group, 659
topological, 191 monotonicity formula for energy minimiz-
totally transverse, 194 ers, 61
uniquely ergodic, 194 Morgan–Shalen compactification, 75
metric Mostow rigidity, 421
asymmetric, 165 moving frame, 574
Teichmüller, 43, 133, 261, 267, multiple curve, 591
294, 301 Mumford–Mahler compactness, 82
Weil–Petersson, 78
metric graph, 377 nearest point retraction, 420
minimal action, 67 Nielsen realization problem, 85, 87
minimal geodesic lamination, 449 nilpotent quotient, 371
minimal measured geodesic nonpositively curved space, 54, 56
lamination, 194 nonrefraction, 80
mirror Veech group, 314 norm of a quadratic differential, 130
Möbius group, 209
modular curve Cg , 253 origami, 296
modular double, 691 outer automorphism groups of free groups,
modular family of Riemann 361
surfaces Cg,n , 264 outer automorphisms, 521
modular functor, 689–691, 720 Outer Space, 376
modular group, 209
commensurability , 274 pairing
modular group additive, 672
Teichmüller (see also mapping class intersection, 672
group), 261 multiplicative, 673
modular groupoid, 589, 690, 720, pants decomposition, 447, 689, 690, 720
726, 743 parabolic element, 413
modularly equivalent Jenkins–Strebel partial foliation, 118
rays, 138 patchwork of double annuli, 326
moduli space, 253, 262, 294, 531, 689 patchwork of rectangles, 323
boundary of, 297 peakless function, 189
791
Penner coordinates, 690, 693, 695, quantum flip transformation, 776

699, 705 quantum geodesic, 599
pentagon relation, 678 quantum multicurve, 600
period matrix, 555 quantum ordering, 600
piecewise flat surface, 510 quantum triangle algebra, 773
pinched, 213 quantum Whitehead move, 595
ping pong quasi-Fuchsian deformation, 225
hyperbolic, 403 quasi-Fuchsian group, 418
topological, 395 quasi-Fuchsian representation, 88
pleated surface, 434 quasi-geodesic, 454
Poincaré metric, 38 quasiconformal deformation space, 423
polygon, 478 quasiconformal dilatation, 126
realizable, 478 local, 126
polygonal pattern, 561 quasiconformal homeomorphism, 39,
polystable, 70 126, 224, 259
precisely invariant, 316 Beltrami coefficient of, 39
pro-finite completion, 215 dilatation of, 39
product foliation structure, 129 extremal, 39
projective foliation class, 117 quasiconformal map
projective measured geodesic discrete, 562
lamination, 151 quasisymmetric homeomorphism, 213,
proper action, 65 259
proper length, 622
proper metric space, 136 R-space, 316
pseudo-Anosov homeomorphism, rational conformal field theory, 688, 690,
257, 392 723
Ptolemy groupoid, 689, 697, 717 ray, 64
Ptolemy transformation, 208 anti-stretch, 164
punctured solenoid, 214 convergent, 139
punctured surface, 508, 518 stretch, 164
pure mapping class group, 521 rays
asymptotic, 141
quadratic differential, 40, 130, 389 recurrent train track, 453
initial, 42 reducible mapping class, 394
Jenkins–Strebel, 58 reducible subgroup, 397
Jenkins–Strebel type, 278 reductive, 71
terminal, 42 regular B-group, 320
quadrilateral, 124 regular point, 62
modulus of a, 124 representation variety, 425, 524
quantization, 594 discreteness locus of, 426
quantization of Teichmüller spaces, representations of groupoids, 714
687, 689, 709 representations of mapping class
quantum dilogarithm function, 777 groups, 687
792
Reshetikhin–Turaev invariants, 688 small action, 67

ribbon graph, 584 spine, 584
Riemann moduli space, 82 splitting sequence, 454
Deligne–Mumford compactification of, square-tiled surface, 296
82 stability, 70
Riemann surface, 120 stable Riemann surface, 297, 317, 341
discrete, 542
Schottky covering of, 341
Riemann surfaces
stable Riemann surfaces
universal family of, 294
Riemann’s extension theorem, 316 family of, 321
rigged Riemann surface, 721 standard generators, 339
right-angled hexagon stretch line, 164
degenerate, 145 stretch map, 164
rigidity of hyperbolic manifolds, 421 stretch ray, 164
structure sheaf, 315
saddle connection, 331 stump, 151
Schottky basis, 336 sublamination, 450
Schottky covering, 299, 336 sufficiently large, 91
generalized, 300 surface
Schottky group, 299, 336, 418
ciliated, 649
Schottky space, 299, 337, 339
decorated, 665
extended, 343
switch conditions, 611
partial compactification of, 341
Schottky structure, 338, 342 symplectic basis, 366
Schottky subgroup, 402 symplectic class, 369
Schottky uniformization, 299, 336 symplectic homomorphism, 343
self-duality equations, 70 symplectic topology, 378
semifield, 668 symplectomorphism group, 356
semisimple mapping class, 84 systole function, 200
semisimple representation, 65
shear coordinate, 587, 620
shift parameter, 146, 495 Tameness Theorem, 415
shifting, 613 Teichmüller boundary, 138, 267
Siegel modular group, 367
Teichmüller compactification, 139
similar Jenkins–Strebel rays, 137
Teichmüller curve, 296, 311, 314, 346
similar representations, 529
simultaneous uniformization, 430 boundary points of, 334
singular edge, 494 cusps of, 297, 299
singular point, 62 Teichmüller deformation, 295, 300, 301
singular set, 92, 96, 660 Teichmüller deformation of
singularity order, 517 punctured Riemann surface, 302
Skora’s theorem, 67 Teichmüller disc, 265, 295, 303, 304, 404
slope, 70 boundary points of, 299, 330, 334
793
global stabilizer of, 312 Torelli group, 367

image in Schottky space of, 347 totally transverse measured foliation, 154
stabilizer of, 296 totally transverse measured geodesic lam-
Teichmüller embedding, 295, 303, 346 ination, 194
extension of, 299, 332 train track, 452, 611
Teichmüller existence theorem, 41, 57 trajectory, 322
Teichmüller extremal problem, 39 critical, 322
Teichmüller geodesic, 295, 391 horizontal, 322
Teichmüller line, 135 regular, 322
horizontal foliation of a, 136 vertical, 322
vertical foliation of a, 136 translation length
Teichmüller map, 41, 134 Weil–Petersson, 84
Teichmüller metric, 43, 133, 261, translation length of a representation, 63
267, 294, 301, 448 translation surface, 269
Teichmüller modular group, 261, 294 transverse measure, 450, 453, 607, 611
Teichmüller ray, 135 transversely recurrent train track, 453
Teichmüller space, 42, 44, 123, 143, tree, 57
209, 294, 357, 376, 389, 448, folding of, 67
531, 659 minimal action on a, 67
A-space, 664 morphism of, 67
X-space, 660 small action on a, 67
augmented, 314 tree dual to a foliation, 58
cotangent space of, 45 tree of projective lines, 342
partial compactification of, 314, 318 triangle, 477
tangent space of, 45 imaginary, 485
Teichmüller space is Stein, 90 triangulation, 649
Teichmüller space of decorated surfaces, tropical analogue, 669
665 tropicalization, 668
Teichmüller structure, 295, 320 twist parameter, 474
universal, 339 2-dimensional simplicial complex, 493
Teichmüller theorem, 43, 53, 73, 87
uniformization theorem, 38
Teichmüller uniqueness theorem, 41, 55 unipotent subgroup, 763
tension field, 47 uniquely ergodic measured foliation, 140
tesselation, 211 uniquely ergodic measured geodesic
three dimensional consistency, 573 lamination, 194
Thurston boundary, 76, 193, 267 universal branched cover, 511
Thurston compactification, 76, 669 universal family, 294, 319
Tits alternative, 397 universal family over T¯g,n , 320
topological cylinder, 128 universal Teichmüller space, 212
modulus of, 128 unparametrized quasi-geodesic, 462
topological measured geodesic
lamination, 191 Veech curve, 272, 273
topological triangle, 124 Veech group, 296, 311, 314
794
parabolic elements of, 335 Weil–Petersson metric, 78

Veech surface, 311 completion of, 79
vertex, 494 curvature of, 78
vertex cycle, 454 geodesic convexity of, 88
vertical foliation, 130, 136 geodesics of, 83
visual boundary, 141 incompleteness of, 78
volume preserving diffeomorphism isometry group of, 83
group, 356 Weyl ordering, 600
vortex equations, 73 Whitehead collapse, 607
Whitehead equivalence, 116
weakly hyperbolic, 401 Whitehead move, 116, 587
Weil–Petersson form, 588, 669,
689, 705, 762 Zygmund spaces, 243

Handbook of Teichmuller Theory

Uploaded by

Copyright:

Available Formats

Handbook of Teichmuller Theory

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Handbook of Teichmuller Theory

Uploaded by

Copyright:

Available Formats

titelei11_papadopoulos 30.4.

2007 10:00 Uhr Seite 1

IRMA Lectures in Mathematics and Theoretical Physics 11

Institut de Recherche Mathématique Avancée

IRMA Lectures in Mathematics and Theoretical Physics

Previously published in this series:

1 Deformation Quantization, Gilles Halbout (Ed.)

Volumes 1–5 are available from Walter de Gruyter (www.degruyter.de)

© 2007 European Mathematical Society

European Mathematical Society Publishing House

Phone: +41 (0)44 632 34 36

Typeset using the author’s TE X files: I. Zimmermann, Freiburg

Strasbourg, April 2007 Athanase Papadopoulos

Introduction to Teichmüller theory, old and new

Part A. The metric and the analytic theory, 1

Chapter 2. On Teichmüller’s metric and Thurston’s asymmetric metric

Chapter 3. Surfaces, circles, and solenoids

Chapter 4. About the embedding of Teichmüller space in the space

Chapter 5. Teichmüller spaces, triangle groups and Grothendieck dessins

Chapter 6. On the boundary of Teichmüller disks in Teichmüller

Part B. The group theory, 1

Chapter 8. Geometric survey of subgroups of mapping class groups

Chapter 9. Deformations of Kleinian groups

Chapter 10. Geometry of the complex of curves and of Teichmüller space

Part C. Surfaces with singularities and discrete Riemann surfaces

Chapter 12. On the moduli space of singular euclidean surfaces

Chapter 13. Discrete Riemann surfaces

Part D. The quantum theory, 1

Chapter 15. Dual Teichmüller and lamination spaces

Chapter 16. An analog of a modular functor from quantized Teichmüller theory

Chapter 17. On quantum moduli space of flat PSL2 (R)-connections

List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783

1 An overview of the structure of Teichmüller space

1.1 Complex structure

1.2 Metric structures

Weil–Petersson metric, by establishing a quasi-isometry between this metric and the

1.3 Symplectic and Poisson structures

1.4 Boundary structures

Thurston’s boundary. Thurston’s boundary of Teichmüller space is induced from

The visual boundaries. Different visual boundaries of Teichmüller space can be

1.5 The harmonic maps approach

1.6 The group theory

group of the Hatcher–Thurston complex of a compact oriented surface of positive

group whose action on S 2 = ∂ H3 preserves a round disk. A quasi-Fuchsian group

1.7 Representation theory

1.8 Dessins d’enfants

geometry, although Grothendieck claimed on several occasions (in particular in his

1.9 Physics and quantum theory

2 An overview of the content of this volume

2.1 The metric and the analytic theory, 1

equivalence class of horizontal foliation, and a harmonic map interpretation of the

Chapter 3, written by Robert Penner, contains an exposition of a generalized notion

A decoration of a hyperbolic surface with cusps is the choice of a closed horocycle

In Chapter 5, William Harvey considers the complex theory of moduli space of

2.2 The group theory, 1

Chapter 8 by Lee Mosher expounds on the dynamics of subgroups of the mapping

Chapter 9 by Albert Marden concerns the deformation theory of Kleinian groups,

2.3 Surfaces with singularities and discrete Riemann surfaces

In Chapter 12, Charalampos Charitos and Ioannis Papadoperakis consider gen-

• H3 = hyperbolic 3-space SL(2, C)/ SU(2);

For any qc map f : S → R we associate μf ∈ B1 (S). If S = H/ a Beltrami

ϕ(z0 ) = 0 ϕ(z0 ) = 0, k = 1 ϕ(z0 ) = 0, k = 2

Proposition 2.17. If f : → N is harmonic with energy E(f ) and N has nonpositive

ϕ(z) = fz , fz = 41 |fx |2 − |fy |2 − 2i fx , fy . (2.34)

Theorem 2.34. Let f : → T be a harmonic map to an R-tree. Then x ∈ is

Theorem 2.35 (Sun [180]). Let f : → T be a harmonic map to an R-tree. Then