Tangent Bundle

SMOOTH MANIFOLDS
AND FIBRE BUNDLES

WITH APPLICATIONS TO
THEORETICAL PHYSICS
SMOOTH MANIFOLDS
AND FIBRE BUNDLES
WITH APPLICATIONS TO
THEORETICAL PHYSICS
Steinar Johannesen
Oslo and Akershus University College of Applied Sciences
Norway
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2017 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed on acid-free paper

Version Date: 20160817
International Standard Book Number-13: 978-1-4987-9671-2 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Library of Congress Cataloging‑in‑Publication Data
Names: Johannesen, Steinar.

Title: Smooth manifolds and fibre bundles with applications to theoretical
physics / Steinar Johannesen.
Description: Boca Raton : CRC Press, [2016] | Includes bibliographical
references and index.
Identifiers: LCCN 2016030311| ISBN 9781498796712 (hardback) | ISBN
9781498796729 (e-book)
Subjects: LCSH: Manifolds (Mathematics) | Fiber bundles (Mathematics) |
Differential equations. | Lie groups.
Classification: LCC QA613.2 .J64 2016 | DDC 514/.72--dc23
LC record available at https://lccn.loc.gov/2016030311
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at

http://www.crcpress.com
Contents
Preface ix
1 INTRODUCTION 1
Space curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Curves on surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 SMOOTH MANIFOLDS AND VECTOR BUNDLES 9

Smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Partitions of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
The rank of a map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
The tangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3 VECTOR FIELDS AND DIFFERENTIAL EQUATIONS 47

Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Integral curves and local flows . . . . . . . . . . . . . . . . . . . . . . . 51
Global flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4 TENSORS 69
Dual vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Tensor bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
The Lie derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Distributions and integral manifolds . . . . . . . . . . . . . . . . . . . . 107
5 DIFFERENTIAL FORMS 117

Exterior forms on a vector space . . . . . . . . . . . . . . . . . . . . . . 117
Differential forms on a manifold . . . . . . . . . . . . . . . . . . . . . . 126
Exterior differentiation of forms . . . . . . . . . . . . . . . . . . . . . . 146
Wedge product of vector-valued forms . . . . . . . . . . . . . . . . . . . 152
Interior product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
The Lie derivative of forms . . . . . . . . . . . . . . . . . . . . . . . . . 157
v
vi Contents
6 INTEGRATION ON MANIFOLDS 163

Manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Exact forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Integration of differential forms . . . . . . . . . . . . . . . . . . . . . . 176
7 METRIC AND SYMPLECTIC STRUCTURES 181

Covariant tensors of degree 2 . . . . . . . . . . . . . . . . . . . . . . . . 181
Pseudo-Riemannian manifolds . . . . . . . . . . . . . . . . . . . . . . . 185
The Hodge star operator . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Time dependent vector fields . . . . . . . . . . . . . . . . . . . . . . . . 212
Symplectic manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Lagrangian systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Conservative systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Time dependent systems . . . . . . . . . . . . . . . . . . . . . . . . . . 241
8 LIE GROUPS 247

Lie groups and their Lie algebras . . . . . . . . . . . . . . . . . . . . . . 247
Group representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Lie subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
The exponential map . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Closed subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Matrix groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
The algebra of quaternions . . . . . . . . . . . . . . . . . . . . . . . . . 284
Left invariant forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
9 GROUP ACTIONS 293

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
The adjoint representation . . . . . . . . . . . . . . . . . . . . . . . . . 295
The groups SO(3) and S3 . . . . . . . . . . . . . . . . . . . . . . . . . 299
The Lorentz group and Sl(2, C) . . . . . . . . . . . . . . . . . . . . . . 303
Semidirect products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Affine spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Infinitesimal group actions . . . . . . . . . . . . . . . . . . . . . . . . . 318
Hamiltonian systems with symmetry . . . . . . . . . . . . . . . . . . . . 322
Lagrangian systems with symmetry . . . . . . . . . . . . . . . . . . . . 326
Gravitational central fields . . . . . . . . . . . . . . . . . . . . . . . . . 328
10 FIBRE BUNDLES 331

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
Induced bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Principal fibre bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Associated bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Contents vii
Tensorial forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

Covariant derivative of forms on principal bundles . . . . . . . . . . . . 365
The curvature form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Horizontal lifts of vector fields . . . . . . . . . . . . . . . . . . . . . . . 367
Local sections and trivializations . . . . . . . . . . . . . . . . . . . . . . 369
Horizontal lifts of curves . . . . . . . . . . . . . . . . . . . . . . . . . . 377
Parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
Forms in associated bundles . . . . . . . . . . . . . . . . . . . . . . . . 382
Covariant derivative of sections in associated vector bundles . . . . . . . 386
Covariant derivative of tensor fields . . . . . . . . . . . . . . . . . . . . 389
Covariant derivative of sections along smooth maps . . . . . . . . . . . . 390
Linear connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Koszul connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Structure equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
Metrical connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
The Schwarzschild - de Sitter spacetime . . . . . . . . . . . . . . . . . . 448
Affine transformations and Killing vector fields . . . . . . . . . . . . . . 462
Conformal transformations . . . . . . . . . . . . . . . . . . . . . . . . . 471
11 ISOMETRIC IMMERSIONS AND THE SECOND FUNDAMENTAL

FORM 475
Connections in reduced subbundles . . . . . . . . . . . . . . . . . . . . 475
The normal bundle and the bundle of adapted orthonormal frames . . . . 476
The second fundamental form . . . . . . . . . . . . . . . . . . . . . . . 482
The shape tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
The shape operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
The formulae of Gauss and Weingarten . . . . . . . . . . . . . . . . . . 491
Strain and vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
The equations of Gauss, Ricci and Codazzi . . . . . . . . . . . . . . . . 501
Pseudo-Riemannian hypersurfaces . . . . . . . . . . . . . . . . . . . . . 507
The Robertson–Walker spacetime . . . . . . . . . . . . . . . . . . . . . 511
The Friedmann cosmological models . . . . . . . . . . . . . . . . . . . 514
12 JET BUNDLES 525

Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
Affine bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
Derivations and the Frölicher-Nijenhuis bracket . . . . . . . . . . . . . . 535
First order jet bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
Holonomic tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . 552
Contact cotangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . 558
Jet fields and connections . . . . . . . . . . . . . . . . . . . . . . . . . . 560
Equivariant jet fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562
Second order jet bundles . . . . . . . . . . . . . . . . . . . . . . . . . . 567
Prolongation of vector fields . . . . . . . . . . . . . . . . . . . . . . . . 577
viii Contents
Calculus of variations . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
A PRELIMINARIES 591
Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
The permutation group . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
Group actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
Categories and functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
Homotopy theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
Topological groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621
Topological vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . 623
Bibliography 629
Index 633
Preface
The intention of this book is to provide the necessary mathematical background from
the theory of smooth manifolds, fibre bundles, Lie groups and differential geometry
for students interested in classical mechanics and the general theory of relativity, and
for students interested in the wide range of applications in diverse areas of mathemat-
ics. Although these mathematical subjects become more and more important in the
recent developments of theoretical physics, it is hard to find books in physics with
a self-contained presentation of the mathematical foundation needed in the physical
theory. The purpose of this book is to fill this gap. This book can as well be used
as a textbook in a pure mathematics course in differential geometry, giving a self-
contained and systematic approach based on principal fibre bundles and jet bundles,
and also containing the necessary background theory of smooth manifolds and Lie
groups. It is assumed that the reader has a good knowledge of basic analysis, linear
algebra and point set topology (see [13], [14], [20] and [22]).
The first chapter is an introduction to differential geometry for curves and sur-
faces in the 3-dimensional Euclidean space R3 , in order to obtain a good intuition of
important concepts in differential geometry such as curvature and torsion. A leading
goal in the next chapters will be to generalize these concepts to spaces of higher di-
mension, with 4-dimensional spacetimes used in the theory of relativity as important
examples.
In Chapters 2 to 6 we start by presenting the fundamental theory of smooth man-
ifolds, including the tangent bundle, global flows for vector fields, tensor bundles,
differential forms, Frobenius’ theorem for distributions, integration on manifolds and
Stokes’ theorem.
Chapter 7 is devoted to metrical structures in vector bundles and pseudo-
Riemannian manifolds. Symplectic manifolds are also introduced, and they are used
to describe Hamiltonian and Lagrangian systems, including the Hamilton-Jacobi the-
orem for time-dependent systems.
Chapters 8 and 9 contain the theory of Lie groups and group actions, which is
needed for the last part of the book. They include the relationship between a Lie
group and its Lie algebra, the exponential map, the closed subgroup theorem, the
Lorentz group, the adjoint representation, semidirect products and affine spaces. Fi-
nally, some applications to Hamiltonian systems with symmetry, including Noether’s
theorem, are given.
Chapters 10, 11 and 12 are a crucial part of the book, describing fibre bundles
and differential geometry which is the foundation for the general theory of relativ-
ity. The main topic is the theory of Ehresmann connections in principal fibre bun-
ix
x Preface
dles, which is specialized to linear connections and metrical connections in pseudo-

Riemannian manifolds. The Schwarzschild metric describing black holes and the
Robertson-Walker metric for homogeneous and isotropic universe models are in-
cluded as important examples. In Chapter 11 we use Ehresmann connections to de-
scribe isometric immersions and the second fundamental form, and in Chapter 12 we
introduce jet bundles and give an alternative definition of connections described as
jet fields.
Finally, Chapter 13 is an appendix which contains some necessary topics which
are used elsewhere in the book. A main subject is the theory of covering spaces and
homotopy theory which is used in the theory of Lie groups in Chapter 8.
Steinar Johannesen
Acknowledgment In May 2013 Steinar Johannesen had a stroke and he was

not able to take care of the final publication of this book on which he spent many
years writing. His legal guardian, Erik Andersen, will heartly thank two of Steinar
Johannesen’s colleagues, professor Jan O. Kleppe and professor Øyvind Grøn, for
assisting in the process of having this book published. He also thanks Dr. Johannes
Kleppe for having contributed so much in preparing the final manuscript of the book.
Erik Andersen
Ole Moes vei 7 E
N-1165 Oslo
Chapter 1
INTRODUCTION
In this chapter we will investigate some important results in the differential geometry
of curves and surfaces in the 3-dimensional Euclidean space R3 , in order to obtain a
good intuition of geometric concepts such as curvature and torsion. These concepts
and results will be generalized to higher dimensional manifolds of various types in
the next chapters.
SPACE CURVES
1.1 Arc length Let c : I → R3 be a smooth curve in R3 defined on an interval I. If
I is not an open interval, this means that c is the restriction of a smooth curve defined
on an open interval containing I. We say that the curve c is regular if c′ (t) 6= 0 for
every t ∈ I. If φ : J → I is a diffeomorphism, then c ◦ φ : J → R3 is also a regular
curve called a reparametrization of c.
Given a regular curve γ : I → R3 , we define its arc length function s : I → R with
initial point t0 ∈ I by Z t
s(t) = kγ ′ (u)k du
t0
for t ∈ I. The curve α = γ ◦ s−1 : J → R3 defined on the interval J = s(I) is a

reparametrization of γ , called a parametrization by arc length , satisfying kα ′ (t)k= 1
for t ∈ J.
1.2 The Frenet frame Let c : I → R3 be a smooth curve parameterized by arc

length. The vector-valued function t = c′ : I → R3 is called the unit tangent vector
field of c. Since
t(s) · t(s) = 1
for s ∈ I, it follows that
t′ (s) · t(s) = 0
for s ∈ I. The vector-valued function t′ : I → R3 is called the curvature vector field of
c, and the function κ : I → R defined by
κ (s) = kt′ (s)k
1
2 SMOOTH MANIFOLDS AND FIBRE BUNDLES
for s ∈ I which measures how fast the unit tangent vector is changing, is called the
curvature function of c. Assuming that κ > 0, we now have a vector-valued function
n : I → R3 , called the principal normal vector field of c, defined by
n(s) = t′ (s)/ kt′ (s)k
for s ∈ I so that
t′ = κ n. (1)
We also have a vector-valued function b : I → R3 defined by
b(s) = t(s) × n(s)
for s ∈ I, called the binormal vector field of c. The vector fields t, n and b form an
orthonormal frame field along the curve c, called the Frenet frame field of c. The
plane through c(s) spanned by n(s) and b(s) is called the normal plane at c(s) for
s ∈ I. The planes through c(s) spanned by t(s) and n(s), and by t(s) and b(s), are
called the osculating plane and the rectifying plane at c(s), respectively.
Since
n(s) · t(s) = 0 and n(s) · n(s) = 1
for s ∈ I, it follows that
n′ (s) · t(s) = −n(s) · t′ (s) = −κ and n′ (s) · n(s) = 0
for s ∈ I. Introducing the torsion function τ : I → R of c defined by
τ (s) = n′ (s) · b(s)
for s ∈ I, we therefore have that
n′ = −κ t + τ b. (2)
In the same way we see that
b′ (s) · t(s) = −b(s) · t′ (s) = 0 , b′ (s) · n(s) = −τ (s) and b′ (s) · b(s) = 0
for s ∈ I, which implies that

b′ = −τ n. (3)
Since b is perpendicular to the osculating plane, we see that the torsion τ measures
the rate at which the curve deviates from being a plane curve.
Equations (1), (2) and (3) are called the Frenet formulae . They express the rate
of change of the Frenet frame in terms of the frame itself. Since the Frenet frame
field is adapted to the curve, with the vector field t tangent to the curve and the vector
fields n and b orthogonal to the curve, it gives information about the shape of the
curve in terms of the curvature κ and the torsion τ .
INTRODUCTION 3
CURVES ON SURFACES
1.3 The Gauss’ formulae Consider a surface in R3 parameterized by a smooth

map r : U → R3 defined on an open subset U of R2 . Using the coordinates x = (x1 , x2 )
in U, the surface and the map r are said to be regular if the vector fields
e1 = ∂ r1 and e2 = ∂ r2
∂x ∂x
are everywhere linearly independent so that they span a tangent plane at each point
on the surface with a unit normal vector
e1 × e2
u= .
ke1 × e2 k
The derivative of e j with respect to ek , which is given by
∂ej ∂ 2r
e jk = k = ,
∂x ∂ xk ∂ x j
can be decomposed in a component in the tangent plane and a component along the
unit normal u by the Gauss’ formula
2
i
e jk = ∑ Γ jk ei + a jk u (1)
i=1
where a jk = e jk · u for 1 ≤ j, k ≤ 2 . The coefficients Γ jk

i are called the Christoffel
symbols of the second kind and will be determined in 1.5. Since e jk = e k j , we see
i and a are symmetric in the indices j and k .
that Γ jk jk
1.4 The metric on a surface Let γ : I → U be a regular curve in U, and consider

the corresponding curve α = r ◦ γ : I → R3 on the surface. Since
2
dα dγ i
dt = ∑ dt ei
i=1
it follows that
2
dα 2 dγ j dγ k
dt = ∑ g jk dt dt ,
j,k=1
where
g jk = e j · ek (1)
for 1 ≤ j, k ≤ 2. The dot product defines a metric in R3 , and g jk are the components
of the induced metric in each tangent plane with respect to the basis consisting of e1
and e2 . The induced metric g is a symmetric bilinear function in each tangent plane,
given by
g (e j , ek ) = g jk
for 1 ≤ j, k ≤ 2. Using the coefficients a jk in Gauss’ formulae we also define the
shape tensor II on the surface, which is a symmetric bilinear function in each tangent
plane given by
II (e j , ek ) = a jk
for 1 ≤ j, k ≤ 2. Classically g was denoted by I , and I and II were called the first
and second fundamental form on the surface. We will follow the modern use of the
notion “form” which is reserved for tensors which are skew symmetric multilinear
functions in each tangent plane.
1.5 The Christoffel symbols By taking the derivative of Equation (1) in 1.4 with
respect to el and using the Gauss’ formulae we obtain
2
∂ g jk i i
l = e j l · ek + e j · e kl = ∑ (gik Γ j l + gi j Γk l ) = [ jl , k ] + [kl , j ], (1)
∂x i=1
where
2
[ jk, l ] = ∑ gil Γ jki
i=1
for 1 ≤ j, k, l ≤ 2 is called Christoffel symbols of the first kind. Permuting the indices
in Equation (1) and using that [ jk, l ] is symmetric in the indices j and k, we have that
[ jk, l ] = 12 {( [l j, k ] + [k j, l ] ) + ( [lk, j ] + [ jk, l ] )

∂ g lk ∂ gl j ∂ g jk
− ( [ j l , k ] + [kl , j ] ) } = 21 j + k − l ,
∂x ∂x ∂x
and
2
i
Γ jk = ∑ gil [ jk, l ]
l=1
where (gi j ) is the inverse of the matrix (gi j ) .
1.6 The Weingarten formulae The curvature of the surface is determined by the
rate of change of the unit normal vector u. Since u · u = 1, we have that u k · u = 0 for
k = 1, 2 , where
u k = ∂ uk
∂x
is the derivative of u with respect to ek . Hence u k lies in the tangent plane and is
given by the Weingarten formula
2
u k = − ∑ a ki ei (1)
i=1
INTRODUCTION 5
for k = 1, 2 . From Gauss’ formulae and the equations u · e j = 0 it follows that

2
a jk = e jk · u = − e j · u k = ∑ gi j a ki
i=1
for 1 ≤ j, k ≤ 2 , so that
2
a ki = ∑ gi j a jk (2)
j=1
for 1 ≤ i, k ≤ 2 . The coefficients a ki in the Weingarten formulae are said to be obtained

from the coefficients a jk in Gauss’ formulae by raising the first index.
We now define the shape operator S on the surface, which is a linear operator in
each tangent plane given by
2
S(ek ) = ∑ a ki ei (3)
i=1
for k = 1, 2 . Given arbitrary tangent vector fields
2
ti = ∑ vki ek
k=1
for i = 1, 2, we have that

2 2 2
II (t1 , t2 ) = ∑ a jk vk1 v2j = ∑ ∑ gi j a ki vk1 v2j
j,k=1 j,k=1 i=1
2 2
j
= ∑ gi j ∑ a ki vk1 v2 = g (S(t1 ), t2 ) .
i, j=1 k=1
As the shape tensor II is symmetric, it follows that the shape operator S is self-adjoint
in each tangent plane.
1.7 The Darboux frame Let s : I → R be the arc length function of the curve α
defined in 1.4 with initial point t0 ∈ I, and let β = γ ◦ s−1 be a reparametrization of γ
so that c = α ◦ s−1 = r ◦ β is a parametrization of α by arc length. From the relation
2
dβ i
t = dc
dt = ∑ dt ei (1)
i=1
and Gauss’ formulae it follows that

2 2
d2β i dβ j dβ k
t′ = κ n = ∑ e+ ∑ dt dt e jk
dt 2 i
i=1 j,k=1
2 2 2
d2 β i i dβ j dβ k dβ j dβ k
=∑ dt 2
+ ∑ Γ jk dt dt ei + ∑ a jk dt dt u .
i=1 j,k=1 j,k=1
As the first term after the last equality is orthogonal to both u and t, it must be parallel
to
e = u×t .
The vector fields t, e and u form an orthonormal frame field along the curve c, called
the Darboux frame field of c, which is adapted to the surface as well as to the curve.
Now we have that
t′ = κ n = κg e + κn u, (2)
where κg = κ n · e = κ u · b and κn = κ n · u are called the geodesic and normal
curvatures of the curve c. They are given by
2 2
d2 β i i dβ j dβ k
κg e = ∑ dt 2
+ ∑ Γ jk dt dt ei (3)
i=1 j,k=1
and
2
dβ j dβ k
κn = ∑ a jk dt dt = II (t, t). (4)
j,k=1
The curves c with vanishing geodesic curvature satisfy the differential equations
2
d2 β i i dβ j dβ k
dt 2
+ ∑ Γ jk dt dt = 0
j,k=1
for i = 1, 2 and are called geodesics. These are the curves on the surface which corre-
spond to straight lines in a plane. Introducing the geodesic torsion τg = e′ · u of the
curve c, we obtain the formulae
e′ = −κg t + τg u and u′ = −κn t − τg e
which together with formula (2) express the rate of change of the Darboux frame in
terms of the frame itself.
1.8 Gaussian curvature From Equation (4) in 1.7 we see that the normal cur-
vature κn of a curve c parameterized by arc length through a point p on the surface
only depends on its unit tangent vector t at this point which is given by Equation (1)
in 1.7. The value of
κn (t) = II (t, t) = g (S(t), t)
for all unit vectors t in the tangent plane at the point p gives information about the
curvature of the surface at p. Since the set of unit tangent vectors is compact, there is
a maximal and a minimal value of the normal curvature. They are called the principal
curvatures, and the corresponding unit tangent vectors define the principal directions.
If the maximal and minimal values are equal, all directions are called principal.
As the shape operator S is self-adjoint, there is an orthonormal basis {t1 , t2 } in
INTRODUCTION 7
the tangent plane at p consisting of eigenvectors for S with eigenvalues λ1 and λ2 ,

respectively. Since
κn (at1 + bt2) = a2 λ1 + b2 λ2 = λ1 + b2 (λ2 − λ1 )
for every (a, b) ∈ R2 with kat1 + bt2k2 = a2 + b2 = 1 , these eigenvalues are the prin-
cipal curvatures. Their product
K = λ1 λ2
is called the Gaussian curvature of the surface at p. By (2) and (3) in 1.6 we see
that the matrix of S with respect to the basis {e1 , e2 } is B −1A , where A = (ai j ) and
B = (gi j ) . We therefore have that
det A
K = det (B −1A) = det B .
Chapter 2
SMOOTH MANIFOLDS AND VECTOR
BUNDLES
The purpose of this chapter is to generalize the differential calculus in Euclidean

spaces to a kind of topological spaces called manifolds which are only locally Eu-
clidean. The idea is to use local coordinate systems around each point which overlap
smoothly. We will investigate how tangent vectors at a point transform under coordi-
nate changes. Using this we will make equivalence classes forming a tangent space
at each point of the manifold. Together they constitute the tangent bundle which is
our first example of a vector bundle.
SMOOTH MANIFOLDS
2.1 Definition A topological space M is said to be locally Euclidean if it is

Hausdorff and if each point p in M has an open neighbourhood U homeomorphic to
an open set in Rn for some integer n ≥ 0. A map x : U → Rn sending U homeomor-
phically onto an open set x(U) ⊂ Rn , is called a coordinate map at p. Its component
functions xi : U → R, i = 1, ..., n, are called coordinate functions , and U is called a co-
ordinate neighbourhood of p. The pair (x,U) is called a local coordinate system or
a local chart .
If (x,U) and (y,V ) are two local charts on M, the open sets x(U ∩V ) and y(U ∩
V ) are mapped homeomorphically onto each other by the maps y ◦ x−1 and x ◦ y−1 ,
which are called the coordinate transformations . We say that the local charts (x,U)
and (y,V ) are C∞ -related if the coordinate transformations y ◦ x−1 and x ◦ y−1 are C∞ .
An atlas on M is a family of C∞ -related local charts A = {(xα ,Uα ) | α ∈ A}
where {Uα |α ∈ A} is an open cover of M.
A smooth structure on M is an atlas D which is maximal in the sense that if
(x,U) is a local chart on M which is C∞ -related to every chart in D, then (x,U) ∈ D.
A locally Euclidean space M which is second countable, is called a topological
manifold . If it also has a smooth structure, it is called a smooth manifold .
2.2 Proposition If A is an atlas on a locally Euclidean space M, then there is a

unique smooth structure D on M containing A .
9
PROOF : Let D be the set of all charts on M which are C∞ -related to every chart in
A . In order to prove that D is an atlas, we must prove that y ◦ x−1 is C∞ for any pair
of local charts (x,U) and (y,V ) in D. Let a be a point in the domain x(U ∩ V ) of
y ◦ x−1 , and let (xα ,Uα ) be a local chart in A around x−1 (a). Then y ◦ x−1 is C∞ at
a, since y ◦ x−1 |x(U∩V ∩Uα ) is the composition of the C∞ coordinate transformations
xα ◦ x−1 and y ◦ x−1 −1 is C∞ at every point in x(U ∩ V ), and it is clear
α . Hence y ◦ x
that D is the unique maximal atlas containing A .
2.3 Definition A topological space X is called a Lindelöf space if every open

cover of X contains a countable subcover.
2.4 Proposition If a topological space X is second countable, then it is a Lindelöf

space.
PROOF : Let O be an open cover of X, and let B be a countable basis for its topology.
Then this topology also has a countable basis {Oi | i ∈ N} consisting of those sets in
B which are contained in at least one open set in O. If for each positive integer i we
choose an open set Ui in O such that Oi ⊂ Ui , then {Ui | i∈N} is a countable subcover
of O.
2.5 Corollary In a second countable space X there cannot exist an uncountable

family of disjoint nonempty open sets.
S
PROOF : Suppose there is such a family {Uα |α ∈ A} in X. Then U = α ∈A Uα is a
second countable subspace of X which is not a Lindelöf space, since {Uα |α ∈ A} is
an open cover of U with no countable subcover. We thus obtain a contradiction to
Proposition 2.4.
2.6 Proposition A locally Euclidean space M with an atlas A is second countable

if and only if A contains a countable atlas.
PROOF : Let {(xi ,Ui ) | i ∈ N} be a countable atlas contained in A , and let {Oi j | j ∈
N} be a countable basis for the topology on xi (Ui ) for each i. Then the family
{x−1
i (Oi j ) | (i, j) ∈ N × N} is a countable basis for the topology on M which is there-
fore second countable.
The converse statement follows from Proposition 2.4.
2.7 Remarks The dimension n of the Euclidean space Rn having an open subset
homeomorphic with a coordinate neighbourhood of a point p in a smooth manifold
M, is uniquely determined by p. For if x : U → Rn and y : V → Rm are two coor-
dinate maps at p, then the derivative D(y ◦ x−1 )(x(p)) : Rn → Rm of the coordinate
transformation y ◦ x−1 is an isomorphism with inverse D(x ◦ y−1 )(y(p)), and hence
n = m.
SMOOTH MANIFOLDS AND VECTOR BUNDLES 11
This result is also true more generally for topological manifolds, where the coor-
dinate transformation y ◦ x−1 is a homeomorphism between the nonempty open sets
x(U ∩V ) and y(U ∩V ) in Rn and Rm , respectively. The proof then depends on a deep
result from algebraic topology by Brouwer saying that two nonempty, open subsets
of Rn and Rm can only be homeomorphic when n = m (topological invariance of
dimension, cf. [23]).
n is called the dimension of M at p and is denoted dim p (M). We say that the
manifold M is n-dimensional and write dim(M) = n or simply M n if it has dimension
n at all points p.
A manifold M being locally Euclidean, is also locally connected, i.e., each point
has a basis of connected neighbourhoods. Hence the connected components are open
subsets of M and are therefore also manifolds. A manifold M can therefore be written
as a disjoint union of connected manifolds which are both open and closed in M.
2.8 Proposition A connected manifold has the same dimension at all points.
PROOF : Suppose M is connected, and let f : M → Z be the function defined by
f (p) = dim p (M). Since f is continuous, f (M) is also connected and hence consists
of only one number n since Z has the discrete topology.
2.9 Examples
(a) The Euclidean space Rn is a smooth manifold with atlas {(id, Rn )}, where
id : Rn → Rn is the identity map.
(b) An n-dimensional vector space V is a smooth manifold with the topology de-
fined in Proposition 13.117 in the appendix, and with atlas {(x,V )} where
x : V → Rn is any linear isomorphism.
(c) Let M be an open subset of a smooth manifold N with smooth structure D.
Then M is also a smooth manifold with smooth structure D |M = {(x,U) ∈
D |U ⊂ M}.
(d) An open interval I is a smooth manifold with atlas {(r, I)}, where r : I → R is
the inclusion map. We call (r, I) the standard local chart on I.
(e) The n-sphere is the set Sn = {x ∈Rn+1 | kxk = 1}. Sn is a smooth manifold with
an atlas consisting of 2n + 2 local charts ( fi ,Ui ) and (gi ,Vi ) for i = 1, ..., n + 1,
where Ui = {x ∈ Sn |xi > 0} and Vi = {x ∈ Sn |xi < 0} are open half spheres,
and fi : Ui → Rn and gi : Vi → Rn are homeomorphisms onto the open ball
D n = {y ∈ Rn | k y k < 1} given by
fi (x) = gi (x) = (x1 , ..., xi−1 , xi+1 , ..., xn+1 ).
Their inverses are given by

1
fi−1 (y) = (y1 , ..., yi−1 , (1− k y k2 ) 2 , yi , ..., yn ) and
1
g−1 1
i (y) = (y , ..., y
i−1
, − (1− k y k2) 2 , yi , ..., yn )
from which we see that the coordinate transformations f j ◦ fi−1 , g j ◦ g−1

i ,
g j ◦ fi−1 and f j ◦ g−1
i are C ∞.
The smooth structure on Sn can also be defined by an atlas consisting of only

two local charts (φ1 , Sn \ {p}) and (φ2 , Sn \ {a}), where p = (0, 0, ..., 1) and
a = (0, 0, ..., −1) are the north and south pole, and where φ1 : Sn \ {p} → Rn
and φ2 : Sn \ {a} → Rn are the stereographic projections given by
φ1 (x) = (1 − xn+1)−1 (x1 , ..., xn ) and

φ2 (x) = (1 + xn+1)−1 (x1 , ..., xn ).
To find the inverse of φ1 , we must solve the equation y = φ1 (x) with respect to
x. Using k x k = 1, we see that
1 − (xn+1 )2 n+1
k y k2 = n+1 2
= 1 + xn+1 , so that
(1 − x ) 1−x
2 2
1+ k y k2 = and 1 − xn+1 = .
1 − xn+1 1 + kyk2
Hence the inverse of φ1 is given by

!
2y1 2yn kyk2 − 1
φ1−1 (y) = , ..., , ,
1 + kyk2 1 + kyk2 kyk2 + 1
and similarly
!
2y1 2yn 1 − kyk2
φ2−1 (y) = , ..., , .
1 + kyk2 1 + kyk2 1 + kyk2
From this we see that

y
φ2 ◦ φ1−1 (y) = φ1 ◦ φ2−1 (y) =
kyk2
which proves that the coordinate transformations φ2 ◦ φ1−1 and φ1 ◦ φ2−1 are C∞ .
We also see that the two atlases defined above are C∞ -related and hence give
the same differeniable structure on Sn .
(f) The product M × N of two smooth manifolds M and N is again a smooth man-
ifold. If {(xα ,Uα ) | α ∈ A} and {(yβ ,Vβ ) | β ∈ B} are atlases for M and N, re-
spectively, then {(xα × yβ ,Uα × Vβ ) | (α , β ) ∈ A × B} is an atlas for M × N. In
the same way we can form the product manifold M1 × · · · × Mn with n factors.
(g) The n-dimensional torus Tn = S1 × · · · × S1 is the product manifold of n circles.
SMOOTH MAPS
2.10 Definition Let f : M → N be a continuous map between two smooth man-

ifolds M and N. We say that f is differentiable of class Ck at a point p in M, for
0 ≤ k ≤ ∞, if y ◦ f ◦ x−1 is Ck in an open neighbourhood of x(p) for local charts
(x,U) and (y,V ) around p and f (p), respectively. If this is true for one pair of local
charts, it is true for every other pair since the local charts around a point in a smooth
manifold are C∞ -related.
If f is differentiable of class Ck at every point in an open subset O of M, we say
that f is differentiable of class Ck on O, and if O = M we just say that f is differen-
tiable of class Ck . This is equivalent to saying that y ◦ f ◦ x−1 is Ck for all local charts
(x,U) and (y,V ) on M and N, respectively. We will be interested almost exclusively
in the C∞ -case, and by a smooth map we mean a map which is differentiable of class
C∞ .
A smooth map f is called a diffeomorphism if it is bijective and has a smooth
inverse. We say that f is a local diffeomorphism at a point p in M if it is a diffeomor-
phism from an open neighbourhood of p onto an open neighbourhood of f (p).
2.11 Remark For a function f : M → R we use the standard atlas {(id, R)} on
R. Hence f is smooth at a point p in M if f ◦ x−1 : x(U) → R is C∞ in an open
neighbourhood of x(p) for a local chart (x,U) around p. f is smooth if f ◦ x−1 is C∞
for all local charts (x,U) on M.
2.12 Definition Let V be an open neighbourhood of a point p in M n , and let

f : V → R be a continuous function which is smooth at p. If (x,U) is a local chart
around p, we define the i-th partial derivative of f at p with respect to (x,U) to be
∂f
(p) = Di ( f ◦ x−1 )(x(p)) ,
∂ xi
i.e., the i-th partial derivative of f ◦ x−1 at x(p). In the same way we define the m-th
order partial derivative of f at p with respect to (x,U) to be
∂m f
(p) = Dik11 · · · Dikrr ( f ◦ x−1 )(x(p))
(∂ xi1 )k1··· (∂ xir )kr
for (i1 , ..., ir ) ∈ Inr and non-negative integers k1 , ..., kr with m = ∑ri=1 ki .
An n-tuple k = (k1 , ... , kn ) of non-negative integers is called a multi-index . Its
length and factorial are defined by
n n
|k| = ∑ ki and k ! = ∏ ki ! ,
i=1 i=1
and we set
∂ |k| f ∂ |k| f
k (p) = 1 k1 (p) .
∂x (∂ x ) ··· (∂ xn )kn
We let 1i denote the multi-index where all the components are equal to zero except
the i-th component which is 1.
2.13 Remark If M = Rn and the local chart is (id, Rn ), we have
∂f ∂m f
(p) = Di f (p) and (p) = Dik11 · · · Dikrr f (p)
∂ xi (∂ xi1 )k1··· (∂ xir )kr
in accordance with the usual notation for partial derivatives of functions defined on
an open subset of Rn .
2.14 Proposition Let f and g be functions which are smooth at a point p in M,

and let (x,U) be a local chart around p. Let a ∈ R. Then
∂ ( f +g) ∂f ∂g
(1) (p) = i (p) + i (p)
∂ xi ∂x ∂x
∂ (a f ) ∂f
(2) (p) = a i (p)
∂ xi ∂x
∂ ( f g) ∂f ∂g
(3) (p) = i (p) g(p) + f (p) i (p)
∂ xi ∂x ∂x
PROOF : Follows from the definition and the corresponding rules for partial deriva-
tives of functions defined on an open subset of a Euclidean space.
2.15 Definition Let p be a point in the smooth manifold M, and let F p be the al-
gebra of all smooth functions f : V → R , each defined on some open neighbourhood
V of p. Then a linear functional l on F p is called a local derivation at p if it satisfies
the relation l( f g) = l( f ) g(p) + f (p) l(g) for every f , g ∈ F p .
2.16 Example If (x,U) is a local chart around a point p in the smooth manifold

M, then it follows from Proposition 2.14 that the operator l = ∂ i on F p taking f
∂x
p
∂f
to i (p) is a local derivation at p. It is called the i-th partial derivation at p with
∂x
respect to (x,U).
PARTITIONS OF UNITY
2.17 Definition A family {Uα |α ∈ A} of sets in a topological space X is said to
be locally finite if each point p in X has a neighbourhood V such that V ∩Uα 6= 0/ for
only a finite number of indices α ∈ A.
If {Uα |α ∈ A} and {Vβ |β ∈ B} are two covers of X, we say that {Uα |α ∈ A} is a
refinement of {Vβ |β ∈ B} if for every α ∈ A, there is a β ∈ B such that Uα ⊂ Vβ .
2.18 Definition By a partition of unity on a smooth manifold M we mean a

family {φα |α ∈ A} of nonnegative smooth functions on M such that their supports
form a locally finite cover of M and we have that
∑ φα (p) = 1
α ∈A
for every p ∈ M.
We say that a partition of unity {φα |α ∈A} is subordinate to a cover {Vα |α ∈A}
of M if supp (φα ) ⊂ Vα for every α ∈ A.
2.19 Lemma If {Fα |α

S
∈ A} is a locally finite family of closed sets in a topo-
logical space X, then F = α ∈A Fα is closed.
PROOF : If p is a point in F c , there is an open neighbourhood V of p and a finite
subset B of the index set A such that V ∩ Fα = 0/ for every α ∈ A − B. This implies
T
that V ∩ α ∈B Fαc is an open neighbourhood of p contained in F c , showing that F c is
open and hence that F is closed in X.
2.20 Lemma If a topological space X is locally compact and second countable,

then it is covered by a sequence {Gi }∞
i=1 of open sets with compact closure such that
Gi ⊂ Gi+1 for all i.
PROOF : Since X is locally compact, we know that each point in X has an open
neighbourhood with compact closure, and these neighbourhoods form an open cover
of X which by Proposition 2.4 has a countable subcover {Ui }∞
i=1 . Let G1 = U1 , and
suppose that
Gk = U1 ∪ · · · ∪Uik .
Then there is a smallest integer ik+1 greater than ik such that
Gk ⊂ U1 ∪ · · · ∪Uik+1 ,
and we define
Gk+1 = U1 ∪ · · · ∪Uik+1 .
Since ik ≥ k, we have that Gk ⊃ Uk which shows that {Gi }∞ i=1 covers X, and the other
properties of {Gi }∞
i=1 in the proposition are clearly satisfied.
2.21 Lemma Let O be an open cover of a smooth manifold M. Then there

is a countable atlas {(xi ,Ui )|i ∈ N} on M such that {Ui |i ∈ N} is a locally finite
refinement of O with xi (Ui ) = B2 (0) for each i ∈ N , and the family {xi −1 (B1 (0))|i ∈
N} is an open cover of M.
PROOF : Let {G j }∞j=1 be a sequence of open sets in M as given by Lemma 2.20, and
let G−1 = G0 = 0./ For each integer j ≥ 0 and each point p ∈ G j+1 − G j there is an
open set Vp in O containing p and a local chart (x p ,U p ) around p such that
U p ⊂ (G j+1 − G j−1 ) ∩Vp .
We may as well assume that x p (U p ) = B2 (0), if necessary by shrinking U p so that

x p (U p ) becomes an open ball centered at x p (p), and then combining x p with a trans-
lation and a scaling mapping this open ball onto B2 (0). We let Wp = x p −1 (B1 (0)).
Then {Wp | p ∈ G j+1 − G j } is an open cover of G j+1 − G j which contains a finite
subcover since G j+1 − G j is compact. We thus obtain a sequence {pi }∞ i=1 in M such
that {Wpi | i ∈ N} is an open cover of M, and the family {U pi | i ∈ N} is locally finite
since G j ∩U p = 0/ for p ∈ Gk+1 − Gk when k ≥ j + 1. If we set xi = x pi and Ui = U pi ,
the family {(xi ,Ui ) | i ∈ N} is hence an atlas on M with the desired properties.
2.22 Lemma The non-negative real function f : R → R defined by

−1/x
e for x > 0
f (x) =
0 for x ≤ 0
is smooth on R, and f (x) > 0 for x > 0.

PROOF : We must show that f is smooth at 0. We see by induction that
f (n) (x) = e−1/x x−2n Pn (x)
for x > 0, where Pn (x) is a polynomial of degree ≤ n. Indeed, this is clearly true for
n = 0, and assuming that it is true for n we have that
f (n+1) (x) = e−1/x x−2(n+1)Pn+1 (x)
where
Pn+1(x) = (1 − 2nx)Pn(x) + x2 Pn′ (x).
From this it follows by induction that f (n) (0) = 0 which is clearly true for n = 0.
Assuming that it is true for n we have that
f (n) (x)− f (n) (0)

lim = lim e−1/x x−2n−1 Pn (x) = 0
x→0+ x x→0+
since, by using the change of variable x = 1/t, we have that

m
lim e−1/x x−m = lim t t = 0
x→0+ t→∞ e
for every non-negative integer m.

The last equality follows from the fact that the exponential function et grows
faster that any power of t as t → ∞. This can be seen from the series expansion
∞ n
et = ∑ tn!
n=0
which implies that

m+1
et ≥ t
(m+1)!
so that
m (m+1)!
0≤ tt ≤
e t
for t > 0.
2.23 Theorem Let O be an open cover of a smooth manifold M. Then there is a

partition of unity {φα |α ∈ A} on M subordinate to O.
PROOF : Let {(xi ,Ui ) | i ∈ N} be a countable atlas on M satisfying the properties
given in Lemma 2.21, and for each i ∈ N, let hi : M → R be the non-negative function
defined by
f (1− k xi (p) k2 ) for p ∈ Ui
hi (p) =
0 for p ∈ M − Ui
where f is the function defined in Lemma 2.22. We see that each hi is smooth on
M with supp (hi ) ⊂ Ui , and that hi (p) > 0 if and only if p ∈ xi −1 (B1 (0)). Since the
family {xi −1 (B1 (0))|i ∈ N} is a locally finite cover of M, we have a well-defined
smooth function
∞
h = ∑ hi
i=1
on M with h(p) > 0 for p ∈ M. The family {ψi |i ∈ N} where ψi = hi /h is hence a

partition of unity on M such that {supp (ψi )|i ∈ N} is a locally finite refinement of
O.
If O = {Vα |α ∈ A}, we have a map µ : N → A so that supp (ψi ) ⊂ Vµ (i) for all
i ∈ N. If we define φα = ∑µ (i)=α ψi for α ∈ A, it follows from Lemma 2.19 that
{φα |α ∈ A} is a partition of unity on M subordinate to O.
2.24 Corollary If F is a closed and O an open subset of a smooth manifold M

with F ⊂ O, then there exists a smooth function f : M → R with 0 ≤ f (p) ≤ 1 for all
p ∈ M such that f (p) = 1 if p ∈ F and supp ( f ) ⊂ O.
PROOF : By Theorem 2.23 there is a partition of unity { f , g} on M subordinate to
the open cover {O, M − F}. Then supp ( f ) ⊂ O, and since supp (g) ⊂ M − F and
f + g = 1 it follows that f (p) = 1 if p ∈ F. Hence f has the desired properties.
2.25 Corollary Let M be a smooth manifold and W a finite dimensional vector

space. If g : O → W is a smooth map defined on an open subset O of M, and if F is a
closed subset of M with F ⊂ O, then there exists a smooth map f : M → W defined
on M which coincides with g on F.
PROOF : By Corollary 2.24 there exists a smooth function h : M → R with h(q) = 1
for q ∈ F and supp (h) ⊂ O, and we let

h(q) g(q) for q ∈ O
f (q) = .
0 for q ∈/O
THE RANK OF A MAP

2.26 Definition Let f : M n → N m be a map which is smooth at a point p
in M, and let (x,U) and (y,V ) be local charts around p and f (p),
respectively.

∂ (yi ◦ f )
We define the rank of f at p to be the rank of the m × n-matrix (p) .
∂xj
This does not depend on the local charts (x,U) or (y,V ). For let (x′ ,U ′ ) and
(y′ ,V ′ ) be another pair of local charts around p and f (p), respectively. Then
we have that y′ ◦ f ◦ x′ −1 = (y′ ◦ y−1 ) ◦ (y ◦ f ◦ x−1 ) ◦ (x ◦ x′ −1 ), and the derivatives
D(x ◦ x′ −1 )(x′ (p)) and D(y′ ◦ y−1 )(y( f (p))) of the coordinate transformations are
isomorphisms.
A point p in M is called a critical point of f : M n → N m if the rank of f at p is
smaller than min(n, m). Otherwise p is called a regular point of f .
If p is a critical point of f , the value f (p) is called a critical value of f . Other
points in N are called regular values of f .
2.27 Remark If f has rank k atp, then f has rank ≥ k in some neighbourhood
∂ (yi ◦ f )
of p, for some k × k submatrix of j has non-zero determinant at p, and
∂x
hence by continuity in a neighbourhood of p.
In particular, if f : M n → N m has rank k = m or k = n at p, then f has rank k in
some neighbourhood of p.
2.28 The Rank Theorem

(1) If f : M n → N m has rank k at p, then there are local charts (x,U) and (y,V )
around p and f (p), respectively, and a C∞ map h : x(U) → Rm−k such that
y ◦ f ◦ x−1 (a1 , ..., an ) = (a1 , ..., ak , h1 (a), ..., hm−k (a)) for a ∈ x(U).
(2) If f : M n → N m has rank k in a neighbourhood of p, then there are local charts

(x,U) and (y,V ) around p and f (p), respectively, such that
y ◦ f ◦ x−1 (a1 , ..., an ) = (a1 , ..., ak , 0, ..., 0) for a ∈ x(U).
PROOF OF (1) : Let (u,W ) and (y,V ) be local charts around p and f (p), respectively.
i j
By a permutation functions u and y we may assume that the k × k-
of the coordinate
∂ (yα ◦ f )
submatrix A = (p) , where 1 ≤ α , β ≤ k, has non-zero determinant.
∂ uβ
Let f1 : u(W ) → Rk and f2 : u(W ) → Rm−k be the component maps of y ◦ f ◦ u−1 :
u(W ) → Rk × Rm−k , i.e. y ◦ f ◦ u−1 (a, b) = ( f1 (a, b), f2 (a, b)) for (a, b) ∈ u(W ) ⊂
Rk × Rn−k . Then

−1 D1 f1 (u(p)) D2 f1 (u(p))
D(y ◦ f ◦ u )(u(p)) =
D1 f2 (u(p)) D2 f2 (u(p))
where D1 f1 (u(p)) = A is non-singular.
Let φ : u(W ) → Rk × Rn−k be the C∞ -map defined by φ (a, b) = ( f1 (a, b), b) for
(a, b) ∈ u(W ) ⊂ Rk × Rn−k . Then

D1 f1 (u(p)) D2 f1 (u(p))
Dφ (u(p)) =
0 In−k
is non-singular, and by the inverse function theorem, φ is a local diffeomorphism on
an open neighbourhood O of u(p).
We let (x,U) be the local chart around p with U = u−1 (O) and x = φ ◦ u|U .
Then if φ (a, b) = (c, d) ∈ x(U), we have y ◦ f ◦ x−1 (c, d) = y ◦ f ◦ u−1 (a, b) =
( f1 (a, b), f2 (a, b)) = (c, h(c, d)) where h = f2 ◦ φ −1 : x(U) → Rm−k is C∞ .
PROOF OF (2) : Let (x,U) and (v,V ) be local charts around p and f (p), respectively,
so that v ◦ f ◦ x−1 : x(U) → Rk × Rm−k has the form in (1), i.e., v ◦ f ◦ x−1 (a, b) =
(a, h(a, b)) for (a, b) ∈ x(U) ⊂ Rk × Rn−k .
By choosing smaller coordinate neighbourhoods U and V , we may assume that f has
rank k in U, and that x(U) = O1 ×O2 and v(V ) = O3 ×O4 for open sets O1 ⊂ O3 ⊂ Rk ,
O2 ⊂ Rn−k and O4 ⊂ Rm−k . From (1) we see that O3 may be replaced by O1 . Since

−1 Ik 0
D(v ◦ f ◦ x ) = ,
D 1 h D2 h
we have that D2 h = 0 in x(U), and hence there is a C∞ map h : O1 → Rm−k so that
h(a, b) = h(a) for (a, b) ∈ O1 × O2 .
Let φ : O1 × Rm−k → O1 × Rm−k be the C∞ -map defined by φ (a, b) = (a, b − h(a)).
φ is a diffeomorphism with inverse ψ : O1 × Rm−k → O1 × Rm−k given by ψ (c, d) =
(c, d + h(c)).
We let (y,V ) be the local chart around f (p) with y = i ◦ φ ◦ v, where i : O1 × Rm−k →
Rk × Rm−k is the inclusion map. Then y ◦ f ◦ x−1 (a, b) = φ ◦ v ◦ f ◦ x−1 (a, b) =
φ (a, h(a)) = (a, 0) for (a, b) ∈ x(U).
2.29 Proposition Let f : M → N be a smooth map from a connected manifold M

to a manifold N, and suppose that f has rank 0 on M. Then f is a constant map.
PROOF : If p is a point in f (M), then f −1 (p) is a nonempty closed subset of M by
the continuity of f . We want to show that it is also open. Since M is connected, this
implies that f −1 (p) = M so that f is constant on M.
Let n and m be the dimensions of M and N, respectively, and let q be an arbitrary
point in f −1 (p). By the second part of the rank theorem, there are local charts (x,U)
and (y,V ) around q and p, respectively, such that
y ◦ f ◦ x−1 (a1 , ..., an ) = (0, ..., 0)
for a ∈ x(U). This shows that U ⊂ f −1 (p) and completes the proof that f −1 (p) is
open in M.
SUBMANIFOLDS
2.30 Definition A smooth map f : M n → N m is called an immersion at a point p
in M if it has rank n at p, and it is called a submersion at p if it has rank m at p. f
is called an immersion (submersion ) if it is an immersion (submersion) at all points
p in M. An immersion f : M → N which is a homeomorphism onto its image f (M)
endowed with the subspace topology, is called an embedding.
Let M and N be smooth manifolds of dimensions n and m, respectively, with
M ⊂ N. M is called an immersed submanifold of N if the inclusion map i : M → N is
an immersion. It is called a submanifold if the inclusion map is an embedding. The
number m − n is called the codimension of M in N, and M is called a hypersurface of
N if the codimension is 1.
2.31 Remark Let f : M → A be a bijection from a smooth manifold M to a

set A. Then A can be made into a smooth manifold in a natural way so that f is
a diffeomorphism. First define a topology on A so that f is a homeomorphism by
defining a subset O of A to be open iff f −1 (O) is open in M. If D denotes the smooth
structure on M, we define the smooth structure on A to be {(x ◦ f −1 , f (U)) | (x,U) ∈
D}.
If f : M → N is a one-to-one immersion, the image f (M) can be made into a
smooth manifold in the manner just described. f (M) is then an immersed subman-
ifold of N. If f : M → N is an embedding, f (M) will have the subspace topology
inherited from N and hence is a submanifold of N.
2.32 Examples
(a) The map f : R → R2 defined by f (t) = (2 sin(t), sin(2t)) is an immersion which
is not one-to-one. Its image is the Geronos lemniscate L given by the equation
x4 − 4x2 + 4y2 = 0.
(b) The maps f1 = f |<−π ,π > and f2 = f |<0,2π > are one-to-one immersions which
are not embeddings. They make the image L into an immersed submanifold of
R2 in two different ways.
(c) The map f3 = f |<− π , π > is an embedding which makes an open subset of L
2 2
into a submanifold of R2 .
(d) An open subset M of a smooth manifold N, with the smooth structure defined
in Example 2.9 (c), is a submanifold of N.
2.33 Remarks (b) of the above example shows that given a subset A of a smooth
manifold M, there is not in general a unique manifold structure (i.e., topology and
smooth structure) on A which makes it an immersed submanifold of M. However,
we will see in Corollary 2.40 that there is at most one such smooth structure for a
given topology on A.
(c) is an example of the next proposition which says that an immersion is locally
an embedding.
2.34 Proposition A smooth map f : M n → N m is an immersion if and only if one

of the following equivalent assertions are satisfied :
(1) Every point p in M has is an open neighbourhood U such that f |U is an embed-
ding.
(2) For each point p in M, there exists local charts (x,U) around p in M and (y,V )
around f (p) in N such that f (U) = {q ∈ V |yn+1 (q) = ... = ym (q) = 0} and
x1 (q) = y1 ◦ f (q) , . . . , xn (q) = yn ◦ f (q) for q ∈U.
PROOF : Assume that f : M n → N m is an immersion, and let p be a point in
M. By the rank theorem there are local charts (x,U) around p in M and (v,W )
around f (p) in N such that the map v ◦ f ◦ x−1 : x(U) → Rn × Rm−n is given by
v ◦ f ◦ x−1 (a) = (a, 0) for a ∈ x(U), in particular f (U) ⊂ W . Then f |U is an embed-
ding since f |U = i ◦ v−1 ◦ (v ◦ f ◦ x−1 ) ◦ x, where v ◦ f ◦ x−1 is an embedding, x and
v−1 are diffeomorphisms and i : W → N is the inclusion map. This proves (1).
In order to prove (2), we let (y,V ) be the local chart around f (p) in N with
V = v−1 (v(W ) ∩ (x(U) × Rm−n )) and y = v|V . Then f (U) ⊂ V , and we have that
y ◦ f (U) = y ◦ f ◦ x−1 (x(U)) = y(V ) ∩ (Rn × {0}) and y ◦ f (q) = y ◦ f ◦ x−1 (x(q)) =
(x(q), 0) for q ∈U.
Conversely, assume that (2) is satisfied. Then y ◦ f ◦ x−1 (a) = (a, 0) for a∈x(U), and
hence f is an immersion at p. The same obviously holds if (1) is satisfied. Since this
is true for all points p in M, we have that f is an immersion.
2.35 Corollary Let M and N be smooth manifolds of dimensions n and m, re-

spectively, with M ⊂ N. Then M is an immersed submanifold of N if and only if one
of the following equivalent assertions are satisfied :
(1) Every point p in M has is an open neighbourhood U which is a submanifold of
N.
(2) For each point p in M, there exists local charts (x,U) and (y,V ) around p in
M and N, respectively, such that U = {q ∈V |yn+1 (q) = ... = ym (q) = 0} and
x1 (q) = y1 (q) , . . . , xn (q) = yn (q) for q ∈U.
PROOF : Follows from Proposition 2.34 applied to the inclusion map i : M → N.
2.36 Proposition Let M be a subset of a smooth manifold N m . Then M is a

submanifold of N of dimension n if and only if the following assertion is satisfied:
For each point p in M, there exists a local chart (y,V ) around p in N having the
submanifold property y(V ∩ M) = y(V ) ∩ (Rn × {0}).
PROOF : Assume that M is a submanifold of N of dimension n, and let p be a point in
M. Let (x,U) and (y,V ) be local charts around p in M and N, respectively, satisfying
assertion (2) of Corollary 2.35. Since M is a subspace of N, we have that U = M ∩ O
for some open set O in N. Replacing V by V ∩ O, we see that the assertion above is
satisfied.
Conversely, assume that M is a subset of N satisfying the assertion above. We give
M the subspace topology, and we define a smooth structure on M as follows. Let
i : Rn → Rn × Rm−n and p : Rn × Rm−n → Rn be the natural injection and projection
defined by i(a) = (a, 0) and p(a, b) = a. A local chart (y,V ) in N having the sub-
manifold property y(V ∩ M) = y(V ) ∩ (Rn × {0}) gives rise to a local chart (x,U)
on M with coordinate neighbourhood U = V ∩ M and coordinate map x = p ◦ y|U . If
(x′ ,U ′ ) is another local chart on M obtained in the same way from (y′ ,V ′ ), then (x,U)
and (x′ ,U ′ ) are C∞ -related since x′ ◦ x−1 = p ◦ y′ ◦ y−1 ◦ i. We thus obtain a smooth
structure on M which by Corollary 2.35 makes it a submanifold of N.
2.37 Proposition Let M be an immersed submanifold of N, and let f : P → N be

a smooth map with f (P) ⊂ M. If f is continuous considered as a map into M, then it
is also smooth as a map into M.
PROOF : Let p be a point in P, and let n and m be the dimensions of M and N. By
Corollary 2.35 there are local charts (x,U) and (y,V ) around f (p) in M and N, respec-
tively, such that U = {q∈V |yn+1 (q) = ... = ym (q) = 0} and x1 (q) = y1 (q), ..., xn (q) =
yn (q) for q ∈U.
Since f is continuous as a map into M, f −1 (U) is an open subset of P containing
p. Hence there is a local chart (z,W ) around p in P with W ⊂ f −1 (U). We have
that x j ◦ f ◦ z−1 = y j ◦ f ◦ z−1 for j = 1, ..., n, and since f : P → N is smooth, these
functions are C∞ at z(p). This proves that f is smooth at p as a map into M, and since
p was arbitrary, this completes the proof.
2.38 Corollary Let M be a submanifold of N, and let f : P → N be a smooth map

with f (P) ⊂ M. Then f is also smooth as a map into M.
PROOF : Since M has the subspace topology, f is continuous as a map into M. The
result then follows from the proposition.
2.39 Remark If M is only an immersed submanifold of N, the assumption that

f is continuous as a map into M is necessary. Consider for instance the immersed
submanifold L of R2 defined in Example 2.32 with manifold structure given by the
immersion f2 . Then f3 : <− π2 , π2 > → R2 is a smooth map which is discontinuous
considered as a map into L.
2.40 Corollary Let A be a subset of a smooth manifold N, and fix a topology

on A. Then there is at most one smooth structure on A which makes it an immersed
submanifold of N.
PROOF : Let P and M be equal as topological spaces and let each have a smooth
structure which makes it an immersed submanifold of N. Then id : P → M is a home-
omorphism, and hence by Proposition 2.37 a diffeomorphism proving that P and M
must have the same smooth structure.
2.41 Proposition Let f : M n → N m be smooth, and let q be a point in N. If f has

constant rank k in a neighbourhood of f −1 (q), then f −1 (q) is a closed submanifold
of M of dimension n − k.
PROOF : Let p be a point in f −1 (q). By the rank theorem there are local charts (u,U)
around p in M and (y,V ) around q in N such that the map y ◦ f ◦ u−1 : u(U) →
Rk × Rm−k is given by y ◦ f ◦ u−1 (a, b) = (a, 0) for (a, b) ∈ u(U) ⊂ Rk × Rn−k . If
u(p) = (a0 , b0 ), then u(U ∩ f −1 (q)) = u(U) ∩ ({a0 } × Rn−k ).
Let r : Rk × Rn−k → Rn−k × Rk be the map defined by r(a, b) = (b, a − a0 ). Then
the local chart (x,U) where x = r ◦ u has the submanifold property x(U ∩ f −1 (q)) =
x(U)∩(Rn−k ×{0}). Since p was an arbitrary point in f −1 (q), this proves that f −1 (q)
is a closed submanifold of M of dimension n − k.
2.42 Example The smooth map f : Rn+1 → R defined by f (x) = k x k2 has rank
n in Rn+1 \ {0}. Hence the n-sphere Sn = f −1 (1) is a closed submanifold of Rn+1 of
dimension n.
This manifold structure coincides with the one defined in Example 2.9 (e). To
see this, we note that id ◦ fi−1 and id ◦ g−1 i have rank n on D. Hence the inclusion
map i : Sn → Rn+1 is an embedding when we use the atlas consisting of the local
charts ( fi ,Ui ) and (gi ,Vi ) on Sn , and the standard atlas consisting of the local chart
(id, Rn+1 ) on Rn+1 . The manifold structure making Sn a submanifold of Rn+1 is
uniquely determined by Corollary 2.40.
2.43 Proposition Let f : M n → N m be a submersion. Then f is an open mapping.

PROOF : Let W be an open subset of M, and suppose that q ∈ f (W ). Choose a point

p ∈ W with f (p) = q. By the rank theorem there are local charts (x,U) and (y,V )
around p and q, respectively, with U ⊂ W such that the map y ◦ f ◦ x−1 : x(U) → Rm
is given by y ◦ f ◦ x−1 (a, b) = a for (a, b) ∈ x(U) ⊂ Rm × Rn−m . As the projection
pr1 : Rm × Rn−m → Rm on the first factor is an open mapping, it follows that f (U) =
y−1 ◦ pr1 ◦ x(U) is an open neighbourhood of q contained in f (W ). Hence f (W ) is
open in N for every open subset W of M, which shows that f is an open mapping.
VECTOR BUNDLES
2.44 Definition Let M be a smooth manifold. The product manifold M × Rn is
called an n-dimensional product bundle over M. The projection pr1 : M × Rn → M
on the first factor is a surjective smooth map called the projection of the bundle. For a
point p in M, the inverse image pr1 −1 (p) = {p} × Rn is called the fibre over p. It has
a natural vector space structure such that the projection pr2 : {p} × Rn → Rn on the
second factor is a linear isomorphism, i.e., with operations (p, u) + (p, v) = (p, u + v)
and k (p, u) = (p, k u) for vectors u and v in Rn and real numbers k.
The product bundle is an example of a more general bundle called a vector bundle
which we define next. A vector bundle is a smooth manifold which looks locally like
a product bundle.
2.45 Definition Let M be a smooth manifold. A smooth manifold E is called an

n-dimensional vector bundle over M if the following three conditions are satisfied :
(i) There is a surjective smooth map π : E → M which is called the projection of
the total space E onto the base space M.
(ii) For each point p ∈ M, π −1 (p) is an n-dimensional vector space called the fibre
over p.
(iii) For each point p ∈ M there is an open neighbourhood U around p and a diffeo-
morphism t : π −1 (U) → U × Rn such that the diagram
t ✲
π −1 (U) U × Rn
❅
πU ❅ pr1
❘
❅ ✠
U
is commutative, and t is a linear isomorphism on each fibre. The pair

(t, π −1 (U)) is called a local trivialization around p.
We often use the projection π : E → M instead of the total space E to denote the
bundle in order to indicate more of the bundle structure in the notation. The fibre
π −1 (p) is also denoted by E p .
A family of local trivializations {(tα , π −1 (Uα ))|α ∈ A}, where {Uα |α ∈ A} is an
open cover of M, is called a trivializing cover of E.
2.46 Remark If (t, π −1 (U)) is a local trivialization, we let the map t p : E p → Rn

be the linear isomorphism defined by t p = pr2 ◦ t| E p for each p ∈ U. We have that
t(v) = (p,t p (v)) when v ∈ E p .
If E = {e1 , ..., en } is the standard basis for Rn , then B p = {t p−1 (e1 ), ...,t p−1 (en )}
is a basis for the fibre π −1 (p).
2.47 Proposition If π : E → M is an n-dimensional vector bundle and p is a

point in M, then there is a local trivialization (t, π −1 (U)) around p where U is the
coordinate neighbourhood of a local chart (x,U) on M.
PROOF : If (t1 , π −1 (U1 )) is a local trivialization and (x2 ,U2 ) a local chart around p,
let U = U1 ∩U2 and x = x2 |U , and let t : π −1 (U) → U × Rn be the diffeomorphism
induced by t1 .
2.48 Proposition Let π : E → M be a map from a set E to a smooth manifold M,

and let {Uα |α ∈ A} be an open cover of M. Suppose that for each α ∈ A, there is a
bijection tα : π −1 (Uα ) → Uα × Rn such that the diagram
tα ✲
π −1 (Uα ) Uα × Rn
❅
π Uα ❅ pr1
❘
❅ ✠
Uα
is commutative. Moreover, suppose that for each pair α , β ∈ A, the map tβ ◦ tα−1 is a
diffeomorphism from (Uα ∩ Uβ ) × Rn onto itself which is a linear isomorphism on
each fibre {p} × Rn for p ∈Uα ∩Uβ .
Then there is a unique topology and smooth structure on E and a unique vec-
tor space structure on each fibre π −1 (p) for p ∈ M such that π : E → M is an n-
dimensional vector bundle and (tα , π −1 (Uα )) is a local trivialization for each α ∈ A.
PROOF : We first define a topology τ on E such that the projection π is continuous
and the bijections tα : π −1 (Uα ) → Uα × Rn are homeomorphisms for every α ∈ A.
The continuity of π implies that each π −1 (Uα ) must be open in τ , and hence a set O
in E must belong to τ if and only if O∩ π −1 (Uα )∈ τ for every α ∈A. The requirement
that the bijections tα are homeomorphisms now defines τ uniquely as the collection
of sets O in E such that tα (O ∩ π −1 (Uα )) is open in Uα × Rn for every α ∈ A. This
collection τ is clearly a topology on E, and we must prove that it satisfies the above
requirements.
We first show that π −1 (Uα ) is open in τ for each α ∈ A. For every β ∈ A we have
that tβ (π −1 (Uα ) ∩ π −1 (Uβ )) = (Uα ∩ Uβ ) × Rn which is open in Uβ × Rn , thereby
proving the assertion.
We next show that tα is a homeomorphism for every α ∈ A. If O ⊂ π −1 (Uα )
belongs to τ , we have that tα (O ∩ π −1 (Uα )) is open in Uα × Rn , which proves that
tα is an open map. To show that it is continuous, let O be an open set in Uα × Rn . For
every β ∈ A we have that tβ (tα−1 (O) ∩ π −1 (Uβ )) = (tβ ◦ tα−1 ) (O ∩ (Uα ∩Uβ ) × Rn )
which is open in Uβ × Rn , and hence tα−1 (O) ∈ τ showing that tα is continuous.
From this it also follows that the projection π is continuous since π Uα = pr1 ◦ tα
is the composition of the continuous maps tα and pr1 in the above commutative
diagram for each α ∈ A.
Now let {(xβ ,Vβ ) | β ∈ B} be an atlas on M such that {Vβ |β ∈ B} is a refinement
of the open cover {Uα |α ∈ A}. We then have a map φ : B → A such that Vβ ⊂ Uφ (β )
for every β ∈ B. By the assumptions in the proposition we now see that the family
B = {((xβ × id) ◦ tφ (β ) , π −1 (Vβ ))| β ∈B} is an atlas on E. To show that E is a smooth
manifold, we must show that it is Hausdorff and second countable. By Proposition
2.6 we know that M has an atlas of the above form where B is countable, thus showing
that E is second countable.
To see that the E is Hausdorff, let p1 and p2 be two different points in E. If p1
and p2 lie in different fibres, they can be separated by the neighbourhoods π −1 (W1 )
and π −1 (W2 ), where W1 and W2 are disjoint neighbourhoods of π (p1 ) and π (p2 ) in
M. If p1 and p2 lie in the same fibre π −1 (q) with q∈Uα , they can be separated by the
neighbourhoods tα−1 (O1 ) and tα−1 (O2 ), where O1 and O2 are disjoint neighbourhoods
of tα (p1 ) and tα (p2 ) in Uα × Rn .
This shows that E is a smooth manifold with a unique smooth structure such that
the bijections tα : π −1 (Uα ) → Uα × Rn are diffeomorphisms for every α ∈A. We see
that the projection π is smooth since π Uα = pr1 ◦ tα is the composition of the smooth
maps tα and pr1 .
We finally show that there is a unique vector space structure on each fibre π −1 (p)
for p ∈M such that (tα , π −1 (Uα )) is a local trivialization for each α ∈A. If tα is to be
a linear isomorphism on the fibre π −1 (p) where p ∈Uα , we must define addition and
scalar multiplication on this fibre by
u + v = tα−1 (tα (u) + tα (v)) and k u = tα−1 (k tα (u))
for elements u and v in π −1 (p) and real numbers k. If p ∈ Uβ for another index
β ∈ A, we have that tβ = (tβ ◦ tα−1 ) ◦ tα is also a linear isomorphism since it is the
composition of the linear isomorphisms tα and tβ ◦ tα−1 . Hence each fibre has a well-
defined vector space structure such that (tα , π −1 (Uα )) is a local trivialization for each
α ∈ A, and this completes the construction of the vector bundle π : E → M.
2.49 Definition Let π1 : E1 → M1 and π2 : E2 → M2 be two vector bundles. A

bundle map from π1 to π2 is a pair ( f˜, f ) of smooth maps f˜ : E1 → E2 and f : M1 →
M2 such that the diagram
f˜
E1 ✲ E
2
π1 π2
❄ ❄
f✲
M1 M2
is commutative, and f˜ is linear on each fibre. We let f˜p denote the induced linear
map f˜p : π1−1 (p) → π2−1 ( f (p)) between the fibres for each p ∈ M1 . If f˜ and f are
diffeomorphisms, ( f˜, f ) is called an equivalence .
2.50 Definition Let π1 : E1 → M and π2 : E2 → M be two vector bundles over the

same base space M. A bundle map from π1 to π2 over M is a smooth map f : E1 → E2
such that the diagram
f✲
E1 E2
❙ ✓
π1 ❙ ✓ π2
✇
❙ ✴
✓
M
is commutative, and f is linear on each fibre. We let f p denote the induced linear map
f p : π1−1 (p) → π2−1 (p) between the fibres for each p ∈ M. If f is a diffeomorphism,
it is called an equivalence over M .
2.51 Definition A vector bundle π : E → M is said to be trivial if it is equivalent

over M with a product bundle.
2.52 Definition Let π : E → M be an n-dimensional vector bundle, and let V be

an open subset of M and i : V → M be the inclusion mapping. Then a smooth map
s : V → E with π ◦ s = i is called a section of π on V , or a lifting of i to E. The set
of all sections of π on V is denoted by Γ(V ; E).
If (t, π −1 (U)) is a local trivialization of π with U ⊂ V and s : V → E is a map
with π ◦ s = i, then it follows from the commutative diagram in Definition 2.45 that
pr1 ◦ t ◦ s = idU . Hence s|U is completely determined by the map h : U → Rn given
by (t ◦ s)(p) = (p, h(p)) for p ∈ U. h is called the local representation of s on U. We
see that s is smooth on U if and only if the local representation h is smooth.
2.53 Proposition If π : E → M is a vector bundle and V is an open subset of

M, then the set Γ(V ; E) of sections of π on V is a module over the ring F (V ) of

C∞ -functions on V with operations defined by
(s1 + s2 )(p) = s1 (p) + s2 (p) and ( f s)(p) = f (p) s(p)
for sections s1 , s2 and s and C∞ -functions f on V . The zero element in Γ(V ; E) is the
zero section ζ : V → E on V defined by ζ (p) = 0 for p ∈ V .
PROOF : By considering local representations we see that s1 + s2 , f s and ζ are sec-
tions of π on V , and using the vector space structure on each fibre we see that the
axioms for a module are satisfied.
2.54 Proposition Let π : E → M be an n-dimensional vector bundle and s : M →

E be a map with π ◦ s = idM , and let E = {e1 , ..., en } be the standard basis for Rn .
Then s is a section of π on M if and only if the map a : U → Rn defined by
n
s(p) = ∑ ai (p) t p−1 (ei )
i=1
for p ∈ U is smooth for every local trivialization (t, π −1 (U)) in a trivializing cover
of E .
PROOF : Since
(t ◦ s)(p) = (p, a(p))
for p ∈U, a is the local representation of s on U as defined in Definition 2.52.
2.55 Remark Definition 2.52 and Proposition 2.54 can be generalized in the
following way. Let π : E → M be an n-dimensional vector bundle, and let f : N → M
be a smooth map from a smooth manifold N . Then a smooth map F : N → E with
π ◦ F = f is called a section of π along f , or a lifting of f to E. The set of all liftings
of f to E is denoted by Γ( f ; E) .
If (t, π −1 (U)) is a local trivialization of π and F : N → E is a map with π ◦ F = f ,
then it follows from the commutative diagram in Definition 2.45 that pr1 ◦ t ◦ F = fU .
Hence F| f −1 (U) is completely determined by the map a : f −1 (U) → Rn , called the
local representation of F on U , given by
(t ◦ F)(p) = ( f (p), a(p))
for p ∈ f −1 (U) . If E = {e1 , ..., en } is the standard basis for Rn , this relation is also
equivalent to
n
F(p) = ∑ ai (p) t −1
f (p) (ei )
i=1
for p ∈ f −1 (U) .
We see that F is smooth on f −1 (U) if and only if the local represen-
tation a is smooth.
2.56 Proposition Let π : E → M be an n-dimensional vector bundle over a

smooth manifold M.
(1) If s1 : V → E is a section of π on an open subset V of M, and if W is a closed

subset of M with W ⊂ V , then there exists a section s : M → E of π on M which
coincides with s1 on W .
(2) If v ∈ E p for a point p ∈ M, then there exists a section s : M → E of π on M
with s(p) = v.
PROOF OF (1) : By Corollary 2.24 there exists a smooth function f : M → R with
f (q) = 1 for q ∈ W and supp ( f ) ⊂ V , and we let

f (q) s1 (q) for q ∈ V
s(q) = .
0 for q ∈
/V
PROOF OF (2) : Let (t, π −1 (U)) be a local trivialization around p, and let E =
{e1 , ..., en } be the standard basis for Rn . If
n
v = ∑ ai t p−1 (ei ) ,
i=1
we have a section s1 : U → E of π on U given by

n
s1 (q) = ∑ ai tq−1 (ei )
i=1
for q ∈ U. Using part (1) there therefore exists a section s : M → E of π on M which

coincides with s1 on the closed subset {p} of M with {p} ⊂ U so that s(p) = v.
2.57 Proposition Let π1 : E1 → M1 and π2 : E2 → M2 be two vector bundles

of dimension n, and let ( f˜, f ) be a bundle map from π1 to π2 where f˜ induces a
linear isomorphism between the fibres π1−1 (p) and π2−1 ( f (p)) for every p ∈ M1 . If
s2 is a section of π2 on an open subset V2 of M2 and V1 = f −1 (V2 ), then the map
s1 : V1 → E1 defined by
s1 (p) = f˜p−1 ◦ s2 ◦ f (p)
for p ∈ V1 is a section of π1 on V1 .
s1 is called the pull-back of s2 by f˜ and is denoted by f˜ ∗ (s2 ).

PROOF : Let p be a point in V1 , and let (t 1 , π1 −1 (U1 )) and (t 2 , π2 −1 (U2 )) be local
trivializations of π1 and π2 around p and f (p), respectively, with U2 ⊂ V2 and U1 ⊂
f −1 (U2 ). Then the map t 2 ◦ f˜ ◦ t −1 : U1 × Rn → U2 × Rn is smooth and induces a
1
−1
linear isomorphism t 2, f (q)◦ f˜q ◦ t 1,q between the fibres {q} × Rn and { f (q)} × Rn
for every q ∈ U1 .
Now t i ◦ si (q) = (q, hi (q)) for i = 1, 2, where the local representations hi : Ui →
Rn of si are given by
h1 (q) = t 1,q ◦ f˜q−1 ◦ s2 ( f (q))
and
h2 (q) = t 2,q ◦ s2 (q) .
Combining this it follows that
−1 −1
h1 (q) = (t 2, f (q) ◦ f˜q ◦ t 1,q ) ◦ h2 ( f (q))
for q ∈ U1 . If E is the standard basis for Rn , we have that

−1 −1 −1 −1
mEE { (t 2, f (q) ◦ f˜q ◦ t 1,q ) } = mEE (t 2, f (q) ◦ f˜q ◦ t 1,q ) .
which shows that s1 is smooth in a neighbourhood of every point p in V1 and hence

is a section of π1 on V1 .
2.58 Proposition An n-dimensional vector bundle π : E → M is trivial if and only

if there are n sections s1 , ..., sn of π on M which are everywhere linearly independent,
i.e., s1 (p), ..., sn (p) ∈ π −1 (p) are linearly independent for every p ∈ M.
PROOF : Suppose first that E is trivial, and let f : E → M × Rn be an equivalence
over M. Then the sections si : M → E defined by si (p) = f −1 (p, ei ) for p ∈ M and
i = 1, ..., n, where ei is the i-th standard basis vector in Rn , clearly are everywhere
linearly independent.
Conversely, suppose there are n sections s1 , ..., sn of π on M which are every-
where linearly independent, and let f : E → M × Rn be the map defined by
n
f ( ∑ ai si (p)) = (p, a)
i=1
for p ∈ M and a ∈ Rn . To show that f is an equivalence over M, let (t, π −1 (U)) be

a local trivialization in E, and let hi : U → Rn be the local representation of si on U
for i = 1, ..., n so that (t ◦ si )(p) = (p, hi (p)) for p ∈ U. Writing the vectors in Rn
as column vectors, we then have a smooth map h : U → Gl (n, R), where h(p) is the
matrix with h1 (p), ..., hn (p) as column vectors for p ∈ U. Combining this, we have
that
n n
t ◦ f −1 (p, a) = t ( ∑ ai si (p)) = (p , ∑ ai hi (p)) = (p, h(p) a)
i=1 i=1
and hence also

f ◦ t −1 (p, b) = (p, h(p)−1 b)
for every p ∈ U and a, b ∈ Rn . Since this holds for every local trivialization
(t, π −1 (U)) in E, it follows that f is an equivalence over M.
2.59 Corollary Let π : E → M be an n-dimensional vector bundle, and let

s1 , ..., sn be n sections of π on an open subset U of M which are everywhere lin-
early independent, i.e., s1 (p), ..., sn (p) ∈ π −1 (p) are linearly independent for every
p ∈ U. Then we have a unique local trivialization (t, π −1 (U)) in E associated with
s1 , ..., sn such that t p (si (p)) = ei for p ∈ U and i = 1, ..., n, where ei is the i-th standard
basis vector in Rn .
A map s : U → E, with π (s(p)) = p for every p ∈ U, is a section of π on U if and
only if the map a : U → Rn defined by
n
s(p) = ∑ ai (p) si (p)
i=1
for p ∈ U is smooth.
PROOF : The existence and uniqueness of the local trivialization (t, π −1 (U)) follows
from the last part of the proof of Proposition 2.58, replacing π with the bundle π | U .
The last part of the corollary follows from the relation
(t ◦ s)(p) = (p, a(p))
which holds for p ∈U, and which shows that a is the local representation of s on U
as defined in Definition 2.52.
2.60 Definition Let π : E → M and π ′ : E ′ → M be two vector bundles over M

of dimensions n and k, respectively. Then π ′ is called a subbundle of π if E ′ is a
submanifold of E and the inclusion map i : E ′ → E is a bundle map over M.
2.61 Remark The inclusion map i being a bundle map is equivalent to the asser-
tion that π ′ = π |E ′ and that each fibre π ′−1 (p) in E ′ is a k-dimensional subspace of
the fibre π −1 (p) in E for all p ∈ M.
2.62 Proposition Suppose that π : E → M is an n-dimensional vector bundle,

and that E ′ is a subset of E such that π −1 (p) ∩ E ′ is a k-dimensional subspace of the
fibre π −1 (p) for every p ∈ M. Then E ′ is a k-dimensional subbundle of E if and only
if one of the following equivalent assertions are satisfied :
(1) For each point p in M there is an open neighbourhood U around p and k sec-
tions s1 , ..., sk of π on U such that s1 (q), ..., sk (q) is a basis for π −1 (q) ∩ E ′ for
all q ∈ U.
(2) For each point p in M, there exists a local trivialization (t, π −1 (U)) around
p in E having the subbundle property t (π −1 (U) ∩ E ′ ) = U × Rk × {0} ⊂
U × Rk × Rn−k .
PROOF : Suppose first that E ′ is a k-dimensional subbundle of E with projection
π′ = π |E ′ , and let (t ′ , π ′ −1 (U)) be a local trivialization around p in E ′ . Then the
sections s j : U → E defined by s j (q) = i ◦ t ′ −1 (q, e j ) for j = 1, ..., k, where e j is the
j-th standard basis vector in Rk and i : E ′ → E is the inclusion map, clearly satisfy
assertion (1).
We next prove that assertion (1) implies (2). If p is a point in M, let s j : U1 → E for
j = 1, ..., k be the sections given in (1) defined on an open neighbourhood U1 around

p. Let (t2 , π −1 (U2 )) be a local trivialization around p in E, and put U0 = U1 ∩U2 . If
h j : U0 → Rn is the local representation of s j as defined in Definition 2.52, we have
a smooth map f : U0 → L(Rk , Rn ) given by
k
f (q)(a) = ∑ a j h j (q)
j=1
for q ∈ U0 and a ∈ Rk , where f (p) is of rank k. If π1 : Rk × Rn−k → Rk and π2 :

Rk × Rn−k → Rn−k are the projections, there is a permutation σ in Sn such that the
standard matrix of π1 ◦ πσ ◦ f (p) is non-singular, where πσ : Rn → Rn is the linear
isomorphism obtained by permuting the coordinates in Rn with σ as defined in 13.8
in the appendix.
Now let A r (q) be the standard matrix of πr ◦ πσ ◦ f (q) for q ∈ U0 and r = 1, 2.
Then A1 is non-singular in an open neighbourhood U of p contained in U0 . If g(q) :
Rn → Rn is the linear isomorphism with standard matrix
−1
A1 (q) 0 A1 (q) −1 0
=
A2 (q) In−k −A2 (q) A1 (q) −1 In−k
for q ∈ U, and i1 : Rk → Rk × Rn−k is the linear map defined by i1 (a) = (a, 0) for
a ∈ Rk , then the map g : U → L(Rn , Rn ) is smooth, and we have that
πσ ◦ f (q) = g(q) −1 ◦ i1
so that
g(q) ◦ πσ ◦ f (q) = i1 .
Hence the local trivialization (t, π −1 (U)) around p given by t(v) = (id U ×
g (π (v)) ◦ πσ ) ◦ t2 (v) for v ∈ π −1 (U) satisfies the subbundle property in (2). Indeed
we have that
t2 (π −1 (q) ∩ E ′ ) = {q} × f (q) (Rk )
which implies that
t (π −1 (q) ∩ E ′ ) = {q} × g (q) ◦ πσ ◦ f (q) (Rk ) = {q} × Rk × {0}
for q ∈ U.
Finally we assume that assertion (2) is satisfied and show that E ′ is a subbundle
of E. It only remains to show that E ′ is a k-dimensional submanifold of E. Let v be a
point in E ′ , and let (t1 , π −1 (U1 )) be a local trivialization in E having the subbundle
property and (x2 ,U2 ) a local chart on M around π (v). Then ((x2 × id) ◦ t1 , π −1 (U1 ∩
U2 )) is a local chart around v on E having the submanifold property, and the result
hence follows from Proposition 2.36.
2.63 Remark The sections s1 , ..., sk in assertion (1) of the proposition are said to
form a local basis for E ′ on U.
A section s of π : E → M on an open subset U of M is said to belong to or lie in

the subbundle E ′ of E, and we write s ∈ E ′ , if s(q) ∈ E ′ for all q ∈ U.
If s′ is a section of π ′ : E ′ → M on U, then s = i ◦ s′ is a section of π : E → M on
U which belongs to E ′ . Conversely, each section s ∈ E ′ is obtained in this way from
a unique section s′ .
2.64 Proposition Let π1 : E1 → M and π2 : E2 → M be subbundles of the vector

bundle π : E → M so that
Ep = E 1 p ⊕ E 2 p
for every p ∈ M, and let f1 : E → E and f2 : E → E be maps which are projections
on E 1 p and E 2 p in each fibre E p , i.e.,
v = f1 (v) + f2 (v) where f1 (v) ∈ E 1 p and f2 (v) ∈ E 2 p
for v ∈ E p . Then f1 and f2 are bundle maps over M.

PROOF : Let {s1 , ..., sn } and {sn+1 , ..., sn+m } be local bases for E1 and E2 , respec-
tively, on an open neighbourhood U of a point p0 ∈ M. Then {s1 , ..., sn+m } is a
local basis for E on U, and by Corollary 2.59 there is a unique local trivialization
(t, π −1 (U)) in E such that t p (si (p)) = ei for p ∈ U and i = 1, ..., n + m , where ei is
the i-th standard basis vector in Rn+m . Now we have that
t ◦ f1 ◦ t −1 (p, a, b) = (p, a, 0) and t ◦ f2 ◦ t −1 (p, a, b) = (p, 0, b)
for (p, a, b) ∈ U × Rn × Rm , showing that f1 and f2 are bundle maps over M.
2.65 Proposition Let π : E → M and π ′ : E ′ → M be two vector bundles over

a smooth manifold M of dimensions n and m, respectively, and let f : E → E ′ be a
bundle map over M so that dim ker( f p ) = k for every p ∈ M. Then
[ [
ker( f ) = ker( f p ) and im( f ) = im( f p )
p∈M p∈M
are subbundles of E and E ′ , respectively, of dimensions k and n − k, with projections

π ′′ : ker( f ) → M and π ′′′ : im( f ) → M sending the sets ker( f p ) and im( f p ) to p for
each p ∈ M.
PROOF : We first show that ker( f ) is a k-dimensional subbundle of E. Given a point
p on M, we choose local trivializations (t, π −1 (U)) and (t ′ , π ′ −1 (U ′ )) around p in
E and E ′ , respectively. Let i1 : Rk → Rk × Rn−k and i2 : Rn−k → Rk × Rn−k be the
linear maps defined by i1 (a) = (a, 0) and i2 (b) = (0, b) for a ∈ Rk and b ∈ Rn−k , and
let π2 : Rk × Rn−k → Rn−k and π2′ : Rm−n+k × Rn−k → Rn−k be the projections on the
second factor. Then there are permutations σ and σ ′ in Sn and Sm , respectively, such
that the standard matrix of π2′ ◦ πσ ′ ◦ t p′ ◦ f p ◦ t p−1 ◦ πσ−1 ◦ i2 is non-singular, where
πσ : Rn → Rn and πσ ′ : Rm → Rm are the linear isomorphisms obtained by permuting
the coordinates in Rn and Rm with σ and σ ′ as defined in 13.8 in the appendix.
Now let A r (q) be the standard matrix of π2′ ◦ πσ ′ ◦ tq′ ◦ fq ◦ tq−1 ◦ πσ−1 ◦ ir for
q ∈ U ∩ U ′ and r = 1, 2. Then A2 is non-singular in an open neighbourhood V of
p contained in U ∩ U ′ . If g(q) : Rn → Rn is the linear isomorphism with standard
matrix
Ik 0
A1 (q) A2 (q)
for q ∈ V , then the map g : V → L(Rn , Rn ) is smooth, and we have that
π2′ ◦ πσ ′ ◦ tq′ ◦ fq ◦ tq−1 ◦ πσ−1 = π2 ◦ g(q)
so that
π2′ ◦ πσ ′ ◦ tq′ ◦ fq = π2 ◦ g(q) ◦ πσ ◦ tq .
This shows that
ker( fq ) = ker(πσ ′ ◦ tq′ ◦ fq ) ⊂ ker(π2′ ◦ πσ ′ ◦ tq′ ◦ fq )

= (g(q) ◦ πσ ◦ tq )−1 ( Rk × {0})
for q ∈ V . As dim ker( fq ) = k , the above inclusion is actually an equality.

Thus we have a local trivialization (t ′′ , π −1 (V )) around p given by t ′′ (v) = (id V ×
g (π (v)) ◦ πσ ) ◦ t(v) for v ∈ π −1 (V ) which satisfies
t ′′ (π −1 (q) ∩ ker( f )) = {q} × Rk × {0}
for every q ∈ V , completing the proof that ker( f ) is a k-dimensional subbundle of E.

To see that im( f ) is a subbundle of E ′ , let s1 , ..., sn be a local basis for E on
an open neighbourdood U of a point p on M. Then the sections f ◦ s1 , ..., f ◦ sn in
E ′ span im( f ) on U. Hence n − k of these sections form a local basis for im( f )
on an open neighbourhood V of p contained in U, showing that im( f ) is an n − k-
dimensional subbundle of E ′ .
THE TANGENT BUNDLE

2.66 Let p be a point on a smooth manifold M of dimension n, and let c : I → M
be a smooth curve on M passing through p, where I is an open interval containing
t0 , and c(t0 ) = p. By the tangent vector to the curve c at p we mean an object which
is represented by the vector (x ◦ c)′ (t0 ) in a local chart (x,U) around p. If (y,V ) is
another local chart around p, we have that
(y ◦ c)′ (t0 ) = D(y ◦ x−1 )(x(p)) (x ◦ c)′ (t0 ) .
Note that an arbitrary vector v ∈ Rn is the representation in the local chart (x,U) of a
curve c defined by c(t) = x−1 (x(p) + (t − t0 ) v). In order to obtain the tangent vectors
to M at p, we therefore consider the set of all pairs (x, v) where (x,U) is a local chart
around p and v ∈ Rn , and we say that two pairs (x, v) and (y, w) are equivalent, and
write (x, v) ∼ p (y, w), if
w = D(y ◦ x−1 )(x(p)) v .
By the chain rule, this is an equivalence relation, and the equivalence class of (x, v)
is denoted by [x, v] p and is called a tangent vector to M at p. The set of all tangent
vectors at p is called the tangent space Tp M to the manifold M at p, and the set
[
TM = Tp M
p∈M
with the projection π : T M → M defined by π ([x, v] p ) = p is called the tangent bun-

dle of M. For each local chart (x,U) on M, we have a bijection tx : π −1 (U) → U × Rn
defined by tx ([x, v]q ) = (q, v).
2.67 Proposition The tangent bundle π : T M → M of a smooth manifold M of

dimension n is an n-dimentional vector bundle with (tx , π −1 (U)) as a local trivializa-
tion for each local chart (x,U) on M.
PROOF : The maps tx are clearly bijections satisfying the commutative diagram of
Proposition 2.48. Moreover, if (x,U) and (y,V ) are local charts on M, we have that
ty ◦ tx−1 (q, v) = ty ([x, v]q ) = ty ([y, D(y ◦ x−1 )(x(q)) v]q )

= (q, D(y ◦ x−1 )(x(q)) v)
for (q, v) ∈ (U ∩V ) × Rn which shows that
(y × id) ◦ ty ◦ tx−1 ◦ (x−1 × id) (a, v) = (y ◦ x−1 (a), D(y ◦ x−1 )(a) v)
for (a, v)∈x(U ∩V )× Rn . Hence the map ty ◦ tx−1 is a diffeomorphism from (U ∩V )×

Rn onto itself which is a linear isomorphism on each fibre {q} × Rn for q∈U ∩V . By
Proposition 2.48 it follows that π : T M → M is a vector bundle with (tx , π −1 (U)) as
a local trivialization for each local chart (x,U) on M.
2.68 Remark We say that (tx , π −1 (U)), where tx ([x, v]q ) = (q, v) for q ∈ U and
v ∈ Rn , is the local trivialization in the tangent bundle π : T M → M associated with
the local chart (x,U) on M.
2.69 Definition Let c : I → M be a smooth curve on M defined on an open

interval I. The tangent vector to the curve c at the point c(t) for t ∈ I is the vector
[x, (x ◦ c)′ (t)]c(t) in Tc(t) M for any local chart (x,U) around c(t). This vector is well
defined by the discussion in 2.66, i.e., if (x,U) and (y,V ) are two local charts around
c(t) we have that [x, (x ◦ c)′ (t)]c(t) = [y, (y ◦ c)′ (t)]c(t) .
The tangent vector to c at the point c(t) is also denoted simply by c′ (t), and we
obtain a smooth curve c′ : I → T M on the tangent bundle of M which is a lifting of
c, i.e., with π ◦ c′ = c. Indeed, using the local charts (id, R) and (x,U) on R and M,
we have that
(x × id) ◦ tx ◦ c′ ◦ id (t) = (x ◦ c (t), (x ◦ c)′ (t))
for t ∈ c−1 (U) which shows that c′ is smooth.
Given a smooth map h : J → I from an open interval J, the curve γ = c ◦ h is
called a reparametrization of c. We have that
γ ′ (t) = [ x, (x ◦ γ )′ (t) ] γ (t) = [ x, h′ (t) (x ◦ c)′ (h(t)) ] c(h(t)) = h′ (t) c′ (h(t))
for t ∈ J.
2.70 If f : M → N is a smooth map between the manifolds M n and N m , we want

to define a map f∗ : T M → T N so that ( f∗ , f ) is a bundle map between the tangent
bundles. Let p be a point on M, and let (x,U) and (y,V ) be local charts around p
and f (p). If [x, v] p is the tangent vector to a curve c at p with c(t0 ) = p, we define
f∗ ([x, v] p ) to be the tangent vector of the curve f ◦ c at the point f (p), i.e.,
f∗ (c′ (t0 )) = ( f ◦ c)′ (t0 ) (1)
so that
f∗ ([x, (x ◦ c)′ (t0 )] p ) = [y, (y ◦ f ◦ c)′ (t0 )] f (p) = [y, D(y ◦ f ◦ x−1 )(x(p)) (x ◦ c)′ (t0 )] f (p)
which implies
f∗ ([x, v] p ) = [y, D(y ◦ f ◦ x−1 )(x(p)) v] f (p) . (2)
This shows in particular that f∗ ([x, v] p ) is well defined and does not depend on the
curve c having tangent vector [x, v] p . The induced map f∗ p : Tp M → T f (p) N between
the tangent spaces is given by the commutative diagram
f∗ p
Tp M ✲ T N
f (p)
tx,p ty, f (p)

❄ ❄
Rn ✲ Rm
D(y ◦ f ◦ x−1 )(x(p))
2.71 Proposition For each smooth map f : M → N between the manifolds M n

and N m , ( f∗ , f ) is a bundle map between their tangent bundles.
PROOF : The pair ( f∗ , f ) clearly satisfies the commutative diagram of Definition
2.49. We must show that f∗ is smooth on T M and is linear on each fibre. For lo-
cal charts (x,U) and (y,V ) around p and f (p), respectively, we have
ty ◦ f∗ ◦ tx−1 (q, v) = ty ◦ f∗ ([x, v]q ) = ty ( [y, D(y ◦ f ◦ x−1 )(x(q)) v] f (q) )

= ( f (q), D(y ◦ f ◦ x−1 )(x(q)) v )
for (q, v) ∈ (U ∩ f −1 (V )) × Rn which shows that
(y × id) ◦ ty ◦ f∗ ◦ tx−1 ◦ (x−1 × id) (a, v) = (y ◦ f ◦ x−1 (a), D(y ◦ f ◦ x−1 )(a) v)
for (a, v)∈x(U ∩ f −1 (V )) × Rn , and this completes the proof of the proposition.
2.72 Proposition Let M n , N m and Pr be smooth manifolds.

(1) If the maps f : M → N and g : N → P are smooth, we have that (g ◦ f )∗ =
g∗ ◦ f ∗ .
(2) If id : M → M is the identity on M, then id∗ : T M → T M is the identity on T M.
(3) If f : M → N is a diffeomorphism, then so is f∗ : T M → T N, and we have that
( f∗ )−1 = ( f −1 )∗ .
PROOF : (1) If [x, v] p is the tangent vector to the curve c at p with c(t0 ) = p, then
both (g ◦ f )∗ ([x, v] p ) and (g∗ ◦ f∗ )([x, v] p ) are the tangent vector to the curve g ◦ f ◦ c
at g( f (p)).
(2) If [x, v] p is the tangent vector to the curve c at p with c(t0 ) = p, then id∗ ([x, v] p )
is also the tangent vector to the same curve c at p.
(3) By (1) we have that f∗ ◦ ( f −1 )∗ = ( f ◦ f −1 )∗ = id∗ and ( f −1 )∗ ◦ f∗ =
( f −1 ◦ f )∗ = id∗ . Hence (2) shows that f∗ is a diffeomorphism with ( f∗ )−1 =
( f −1 )∗ .
2.73 Remark Assertion (1) in the proposition is still true for a map g defined
only on an open submanifold V of N if we identify T ( f −1 (V )) with T M| f −1 (V ) and
TV with T N|V , as the proof still holds for p ∈ M ∩ f −1 (V ).
2.74 Proposition Let M and N be smooth manifolds of dimensions n and m,

respectively. Then the tangent space of the product manifold M × N at a point (p, q)
can be written as a direct sum
T(p,q) (M × N) = iq ∗ (Tp M) ⊕ i p ∗ (Tq N) , (1)
where iq : M → M × N and i p : N → M × N are the embeddings defined by iq (p ′ ) =

(p ′ , q) and i p (q′ ) = (p, q′ ) for p ′ ∈ M and q′ ∈ N . If (z 1 ,U) and (z 2 ,V ) are local
charts around p and q, and if z = z 1 × z 2 , then
[ z, v ] (p,q) = iq ∗ ( [ z 1 , v1 ] p ) + i p ∗ ( [ z 2 , v2 ]q ) (2)
for v = (v1 , v2 ) ∈ Rn × Rm .
Given a smooth map f : M × N → P , we have that
f∗ ( [ z, v ] (p,q) ) = fq ∗ ( [ z 1 , v1 ] p ) + f p ∗ ( [ z 2 , v2 ]q ) , (3)
where fq : M → P and f p : N → P are the maps defined by fq (p ′ ) = f (p ′ , q) and

f p (q′ ) = f (p, q′ ) for p ′ ∈ M and q′ ∈ N .
If g : Q → M × N is a smooth map with component maps g 1 : Q → M and
g 2 : Q → N , and if (x,W ) is a local chart around a point q on Q, then
g ∗ ( [ x, v ] q ) = ig 2 (q) ∗ ◦ g 1 ∗ ( [ x, v ]q ) + ig 1 (q) ∗ ◦ g 2 ∗ ( [ x, v ]q ) (4)
for v ∈ Rr where r = dim (Q ) .

If h : Q → P is the composed map given by h = f ◦ g , we have that
h ∗ ( [ x, v ] q ) = fg 2 (q) ∗ ◦ g 1 ∗ ( [ x, v ]q ) + fg 1 (q) ∗ ◦ g 2 ∗ ( [ x, v ]q ) (5)
for v ∈ Rr .
PROOF : The first two formulae follow from
[ z, v ] (p,q) = [ z, (v1 , 0) ] (p,q) + [ z, (0, v2 ) ] (p,q)

= [ z, D (z ◦ iq ◦ z−1 −1
1 ) (z 1 (p)) v1 ] iq (p) + [ z, D (z ◦ i p ◦ z 2 ) (z 2 (p)) v2 ] i p (q)
= iq ∗ ( [ z 1 , v1 ] p ) + i p ∗ ( [ z 2 , v2 ]q ).
By applying f∗ on both sides of formula (2) and using that fq = f ◦ iq and f p =

f ◦ i p , we obtain formula (3). Using formula (2) we also have that
g ∗ ( [ x, v ] q ) = [ z, D (z ◦ g ◦ x−1 ) (x(q)) v ] g(q)

= ig 2 (q) ∗ ( [ z 1 , D (z 1 ◦ g 1 ◦ x−1 ) (x(q)) v ] g1 (q) )
+ ig 1 (q) ∗ ( [ z 2 , D (z 2 ◦ g 2 ◦ x−1 ) (x(q)) v ] g2 (q) )
= ig 2 (q) ∗ ◦ g 1 ∗ ( [ x, v ]q ) + ig 1 (q) ∗ ◦ g 2 ∗ ( [ x, v ]q )
which completes the proof of formula (4). Finally, formula (5) is obtained by com-
bining formula (3) and (4) and using Proposition 2.72 (1).
2.75 Remark To avoid confusion when the manifolds M and N have elements in
common, we denote the maps iq , i p , fq and f p by iq1 , i p2 , fq1 and f p2 in this case.
In particular, if f : N × N → P is a smooth map, and h : N → P is the map given
by h(q) = f (q, q) for q ∈ N, then formula (5) reduces to
h ∗ ( [ x, v ] q ) = fq1∗ ( [ x, v ]q ) + fq2∗ ( [ x, v ]q ) , (6)
where fq1 : N → P and fq2 : N → P are the maps defined by fq1 (p) = f (p, q) and
fq2 (p) = f (q, p) for p ∈ N .
2.76 Proposition Let f : M → N be a smooth map. Then its graph G( f ) =

{(p, f (p)) | p ∈ M} is a submanifold of M × N , and we have that
T(p, f (p)) G( f ) = { i f (p) ∗ (v) + i p ∗ ◦ f ∗ (v) | v ∈ Tp M } (1)
for each p ∈ M . If π1 : M × N → M is the projection on the first factor, then π1 | G( f )

is a diffeomorphism.
PROOF : Let F : M → M × N be the map defined by F(p) = (p, f (p)) for p ∈ M . If
(x,U) and (y,V ) are local charts around p and f (p), respectively, we have that
(x × y) ◦ F ◦ x−1 (a) = (a, y ◦ f ◦ x−1 (a))
for a ∈ x(U ∩ f −1 (V )) , showing that F is an immersion. Since π1 ◦ F = idM where

both F and π1 are continuous, it follows that F is a homeomorphism onto its image
G( f ) endowed with the subspace topology. Hence G( f ) is a submanifold of M × N ,
and π1 | G( f ) is a diffeomorphism. Formula (1) follows from Proposition 2.74 which
implies that
F∗ (v) = i f (p) ∗ (v) + i p ∗ ◦ f ∗ (v)
for v ∈ Tp M .
2.77 We will now see that a tangent vector to a smooth manifold M n at a point
p has an alternative description as a local derivation on the algebra F p of all smooth
functions, each defined on some open neighbourhood of p in M.
Let [x, v] p be the tangent vector to a curve c at p with c(t0 ) = p, and let f : V → R
be a smooth function defined on an open neighbourhood V of p in M. We say that
( f ◦ c)′ (t0 ) is the derivative of f along c at p. This derivative actually depends only
on f and the tangent vector [x, v] p to c at p since we have that
[id, ( f ◦ c)′ (t0 )] f (p) = f∗ ([x, v] p )
so that
( f ◦ c)′ (t0 ) = (tid, f (p) ◦ f∗ )([x, v] p ) ,
where tid, f (p) : T f (p) R → R is the isomorphism defined in Remark 2.46. Hence
( f ◦ c)′ (t0 ) is also called the derivative of f at p with respect to [x, v] p . Using the
curve c defined by c(t) = x−1 (x(p) + (t − t0 ) v), we see that
f ◦ x−1 (x(p) + hv) − f ◦ x−1 (x(p))

( f ◦ c)′ (t0 ) = lim .
h→0 h
By the chain rule we now have that

n n
∂f
( f ◦ c)′ (t0 ) = ∑ Di ( f ◦ x−1 )(x(p)) (xi ◦ c)′ (t0 ) = ∑ vi (p) .
i=1 i=1
∂ xi
Hence we may think of the tangent vector [x, v] p as an operator

n
l = ∑ v ∂ i
i
i=1
∂x
p
on the algebra F p . By Example 2.16, l is a local derivation at p. We will see that in

this way we obtain a very useful description of the tangent space Tp M as the set of
all local derivations at p. We first need two lemmas.
2.78 Lemma Let l be a local derivation at a point p in the smooth manifold M.

Then we have that
(1) l( f ) only depends on the local behavior of the function f in F p . More precisely,
this means that if two smooth functions fi : Vi → R which are defined on open
neighbourhoods Vi of p for i = 1, 2, coincide on an open neighbourhood W of
p contained in V1 ∩V2 , then l( f1 ) = l( f2 ).
(2) l( f ) = 0 for each constant function f in F p .
PROOF : (1) Considering the difference f2 − f1 , it is clearly enough to prove that
l( f ) = 0 for each smooth function f : V → R defined on an open neighbourhood V
of p and vanishing on an open neighbourhoodW of p contained in V . To show this, let
h : M → R be a smooth function with h = 1 on M −W and h(p) = 0. Such a function h
exists by Corollary 2.24. Then we have that l( f ) = l(h f ) = l(h) f (p) + h(p) l( f ) = 0,
which proves the assertion in (1).
(2) Suppose first that g : V → R has the value 1 in the open neighbourhood V of p.
Then we have that l(g) = l(g · g) = l(g) 1 + 1 l(g), which shows that l(g) = 0. If f
has the value c in V , we have that l( f ) = c l(g) = 0 by the linearity of l.
2.79 Lemma Let f : U → R be a C∞ -function defined on a convex open neigh-

bourhood U of a point a in Rn . Then there are C∞ -functions gi : U → R for i = 1, ..., n
such that
n
(1) f (u) = f (a) + ∑ (ui − ai ) gi (u) for u ∈U,
i=1
(2) gi (a) = Di f (a).

PROOF : Let u ∈U, and define a C∞ -function hu : [0, 1] → R by hu (t) = f (a + t(u −
a)). Then we have that
Z 1
f (u) − f (a) = hu (1) − hu(0) = h′u (t) dt
0
Z 1 n n
= ∑ Di f (a + t(u − a)) (ui − ai) dt = ∑ (ui − ai) gi (u) ,
0 i=1 i=1
Z 1
where gi (u) = Di f (a + t(u − a)) dt for i = 1, ..., n.
0
2.80 Proposition The set Tp′ M of all local derivations at p is an n-dimensional

vector space. If (x,U) is a local chart around p, then the set { ∂ 1 , ..., ∂∂xn } of par-
∂x
p p
tial derivations at p with respect to (x,U) is a basis for Tp′ M, and any local derivation
l at p can be written as
n
l = ∑ l(xi ) ∂ i .
i=1
∂x
p
PROOF : Let f : V → R be a smooth function defined on an open neighbourhoodV of

p in M, and let (x,U) be a local chart around p. Choose a convex open neighbourhood
W of x(p) contained in x(U ∩V ). By Lemma 2.79 there are C∞ -functions gi : W → R
for i = 1, ..., n such that
n
(1) ( f ◦ x−1 )(u) = ( f ◦ x−1 )(x(p)) + ∑ (ui − xi (p)) gi (u) for u ∈W , and
i=1
∂f
(2) gi (x(p)) = Di ( f ◦ x−1 )(x(p)) = (p).
∂ xi
When u = x(q), assertion (1) implies that
n
f (q) = f (p) + ∑ (xi (q) − xi (p)) (gi ◦ x)(q) for q ∈ x−1 (W ),
i=1
and applying the derivation l on both sides gives

n n
∂f
l( f ) = ∑ [ l(xi ) (gi ◦ x)(p) + 0 l(gi ◦ x) ] = ∑ l(xi ) (p).
i=1 i=1
∂ xi

∂ ∂

This shows that { 1 , ..., ∂ xn } spans the vector space Tp′ M, and that any local
∂x
p p
derivation l at p can be written as

n
l = ∑ l(xi ) ∂ i .
i=1
∂x
p

To show that { ∂ 1 , ..., ∂∂xn } is linearly independent, assume that
∂x
p p

n
∑ vi ∂∂xi = 0.
i=1 p

By applying both sides to the functions x j and using that ∂ i (x j ) = δi j , we see that
∂x
p
v j = 0 for j = 1, ..., n.
[
2.81 We now let T ′ M = Tp′ M and define the projection π ′ : T ′ M → M by
! p∈M
n
π′ ∑ vi ∂∂xi = p. We have a bijective map eM : T M → T ′ M defined by
i=1 p

n
eM ([x, v] p ) = ∑ v ∂ i .
i
i=1
∂x
p
which is a linear isomorphism on each fibre Tp M.

If (tx , π −1 (U)) are the local trivializations of the bundle Tp M defined in Proposi-
tion 2.67 for each local chart (x,U) on M, then the bijections tx′ : π ′ −1 (U) → U × Rn
defined by tx′ = tx ◦ e−1 M satisfy the conditions of Proposition 2.48. Hence there is
a unique topology and smooth structure on T ′ M such that π ′ : T ′ M → M is an n-
dimensional vector bundle and (tx′ , π ′ −1 (U)) is a local trivialization for each local
chart (x,U) on M. We have that eM is an equivalence over M between the bundles π
−1
and π ′ since eM |π ′ −1 (U) = tx′ ◦ tx .
For each smooth map f : M → N we now define a map f# : T ′ M → T ′ N so that
the diagram
f∗✲
TM TN
eM eN
❄ ❄
f# ✲
T ′M T ′N
is commutative. Since f# = eN ◦ f∗ ◦ e−1 M we see that ( f # , f ) is a bundle map, and we

want to give a more explicit description of f# . If l = eM ([x, v] p ) is a local derivation
in T ′ M, it follows from the diagram that f# (l) = eN ( f∗ ([x, v] p )). Now let g : V → R
be a smooth function defined on an open neighbourhood V of f (p) in N. From the
discussion in 2.77 we then see that by applying the isomorphism tid,g( f (p)) to both
sides of the relation
g∗ ( f∗ ([x, v] p )) = (g ◦ f )∗ ([x, v] p )
we have that
f# (l)(g) = l(g ◦ f ) .
We will from now on identify the bundle π ′ : T ′ M → M with π : T M → M and
call both the tangent bundle of M. Hence we drop the apostophes and write
f∗ instead
n
of f# , and we speak of both [x, v] p and the local derivation ∑ vi ∂∂xi as a tangent
i=1 p
vector to M at p.
2.82 Remark If (tx , π −1 (U)) is the local trivialization in the tangent bundle
π : T M → M associated with a local chart (x,U) around a point p in M n , and if

E = {e1 , ..., en } is the standard basis for Rn , then we have that

∂

−1
= tx,p (ei ) for i = 1, ... , n .
∂ xi
p
2.83 Remark Let c : I → M be a smooth curve on M defined on an open interval

chart on I where r : I → R is the inclusion map.
I, and let (r, I) be the standard local

Then the partial derivation ∂ 1 with respect to (r, I) at a point t ∈ I is denoted
∂r
t
d

simply by dr , and we have that
t
!
d

c′ (t) = c∗ dr .
t
Indeed, if (x,U) is a local chart around c(t), it follows from Definition 2.69 that
c′ (t) = [ x, D(x ◦ c ◦ r−1 )(r(t)) e1 ] c(t) = c∗ ( [ r, e1 ] t ) .
2.84 Lemma Let V be a vector space of dimension n. Then the tangent space TpV
at any point p ∈ V may be identified with V by means of the linear isomorphism ω p :
TpV → V given by ω p = x−1 ◦ tx,p for any linear isomorphism x : V → Rn , ω p being
independent of the choice of x. We will refer to this as the canonical identification.
If E = {e1 , ..., en } is the standard basis for Rn and B = {v1 , ..., vn } is a basis for
V so that x(vi ) = ei for i = 1, ..., n, then B ∗ = {x1 , ..., xn } is the dual basis of B, and

n n
ω p ( ∑ a ∂ j ) = ∑ a j v j .
j
j=1
∂x j=1
p
If l ∈ TpV , then
xi ◦ ω p (l ) = l (xi )
for i = 1, ..., n, and
λ ◦ ω p (l ) = l (λ )
for any linear functional λ on V .
The map ω = (idV × x−1 ) ◦ tx : TV → V × V is an equivalence over V so that

n
ω ( ∑ a ∂ j ) = (p, x−1 (a))
j
j=1
∂x
p
and
n
ω −1
(p, q) = ∑ x (q) ∂ j j
j=1
∂x
p
for p, q ∈ V and a ∈ Rn .
Let c : I → V be a smooth curve on V defined on an open interval I. Then we have
that
n
c(t) = ∑ (xi ◦ c)(t) vi
i=1
and
n
ωc(t) ◦ c′ (t) = ∑ (xi ◦ c)′ (t) vi = lim 1 [ c(t + h) − c(t) ]
i=1 h→0 h
for t ∈ I. If the curve c lies in a subspace W of V , then ωc(t) ◦ c′ (t) ∈ W .

If F : V → W is a linear map between the finite dimensional vector spaces V and
W , and if the tangent spaces TpV and TF(p)W are canonically identified with V and W
by means of the linear isomorphisms ω p : TpV → V and ωF(p) : TF(p)W → W obtained
as above from the linear isomorphisms x : V → Rn and y : W → Rm , respectively, then
ωF(p) ◦ F∗ p ◦ ω p−1 = F .
PROOF : We have that (x,V ) is a local chart on V , and that tx,p : TpV → Rn is a linear
isomorphism. Hence ω p is also a linear isomorphism which is independent of the
choice of x. Indeed, if y : V → Rn is another linear isomorphism, then it follows from
2.70 that
−1
ty,p ◦ tx,p = D (y ◦ x−1 )(x(p)) = y ◦ x−1 ,
so that
y−1 ◦ ty,p = x−1 ◦ tx,p .

n
To prove the next part of the lemma, let l = ∑ a ∂ j . Then j
j=1
∂x
p
n
ω p (l ) = x−1 (a) = ∑ a j vj ,
j=1
and
xi ◦ ω p (l ) = ai = l (xi )
for i = 1, ..., n. If
n
λ= ∑ bj xj ,
j=1
then
n n n
λ ◦ ω p (l ) = ∑ b j x j ◦ ω p (l ) = ∑ b j l (x j ) = l ( ∑ b j x j ) = l (λ ) .
j=1 j=1 j=1
Using Definition 2.69 we have that

n
ωc(t) ◦ c′ (t) = x−1 ◦ tx,c(t) ◦ c′ (t) = x−1 (x ◦ c)′ (t) = ∑ (xi ◦ c)′ (t) vi
i=1
which also implies that
ωc(t) ◦ c′ (t) = x−1 ( lim 1 [ (x ◦ c)(t + h) − (x ◦ c)(t) ] ) = lim 1 [ c(t + h) − c(t) ]

h→0 h h→0 h
for t ∈ I since x is a linear homeomorphism.

If the curve c lies in a subspace W of V , then we may assume that the basis B is
an extension of a basis C = {v1 , ..., vm } for W so that x ◦ c (I) ⊂ x(W ) = Rm × {0},
which shows that ωc(t) ◦ c′ (t) ∈ W .
The last part of the lemma follows from the commutative diagram in 2.70 which
shows that
ωF(p) ◦ F∗ p ◦ ω p−1 = y−1 ◦ ty,F(p) ◦ F∗ p ◦ tx,p

−1
◦ x=
y−1 ◦ D (y ◦ F ◦ x−1 )(x(p)) ◦ x = y−1 ◦ y ◦ F ◦ x−1 ◦ x = F.
Chapter 3
VECTOR FIELDS AND DIFFERENTIAL
EQUATIONS
In this chapter we develop the theory of integral curves and flows of vector fields on
a smooth manifold, and prove the important result that a smooth vector field has a
smooth flow. We first establish the necessary facts about existence and uniqueness of
solutions of differential equations.
VECTOR FIELDS
3.1 Definition By a vector field on an open subset V of a smooth manifold M
we mean a section of its tangent bundle π : T M → M on V . A section of π along a
smooth map f : N → M from a smooth manifold N is called a vector field along f .
If X : M → T M is a map with π ◦ X = idM and p is a point in M, then X(p) is
often denoted by X p . If f : V → R belongs to the algebra F (V ) of C∞ -functions
on V and we think of X p as a local derivation at p, we can define a new function
XV ( f ) : V → R by letting X operate on f at each point p ∈ V , i.e.,
XV ( f )(p) = X p ( f ) .
3.2 Proposition Let π : T M → M be the tangent bundle of a smooth manifold

M n and X : M → T M be a map with π ◦ X = idM . Then X is a vector field on M if
and only if one of the following equivalent assertions are satisfied :
(1) For any local chart (x,U) on M the map a : U → Rn defined by

n
X(p) = ∑ a (p) ∂ i
i
i=1
∂x
p
for p ∈ U is smooth.
(2) Whenever V is open in M and f ∈ F (V ) , then XV ( f ) ∈ F (V ) .
PROOF OF (1) : Follows from Proposition 2.54 and Remark 2.82.
47
PROOF OF (2) : Assume first that X is a vector field on M, and let f ∈ F (V ) for an
open set V in M. If (x,U) is a local chart on M with U ⊂ V and a : U → Rn is the
smooth map defined in (1), we have that
n
∂f
XV ( f )| U = ∑ ai
i=1
∂ xi
showing that XV ( f )| U ∈ F (U). Since this is true for every coordinate neighbourhood
U contained in V , we have that XV ( f ) ∈ F (V ).
Conversely, assume that X satisfies (2) and let (x,U) be a local chart on M. Then
the map a : U → Rn defined in (1) is given by XU (xi ) = ai for i = 1 , ..., n and hence
is smooth, showing that X is a vector field on M.
3.3 Example If (x,U) is a local chart on the smooth manifold M n , then we have
for each i = 1 , ..., n a vector field Xi = ∂ i called the i’th partial derivation on U
∂x
given by

Xi (p) = ∂ i .
∂x
p
If X is an arbitrary vector field on M, we have that
n
X| U = ∑ ai ∂ i
i=1
∂x
where a : U → Rn is the local representation of X on U as defined in

Proposition 3.2 (1) and Definition 2.52.
3.4 Example Let V be a vector space of dimension n, and let ω : TV → V × V
be the equivalence over V obtained from the canonical identification of the tangent
space TpV with V for each p ∈ V as described in Lemma 2.84. Then we have a vector
field P = ω −1 ◦ (idV , idV ) on V , called the position vector field , given by
n
P = ∑ xi ∂ i
i=1
∂x
for any linear isomorphism x : V → Rn .

3.5 Definition Let A be an algebra. Then a linear map F : A → A is called a
derivation of A if it satisfies the relation F( f g) = F( f ) g + f F(g) for every f , g ∈
A.
3.6 Example If X is a vector field on an open subset V of a smooth manifold

M, then the map XV : F (V ) → F (V ) defined in Definition 3.1 is a derivation of the
algebra F (V ) of C∞ -functions on V . Indeed, we have that
XV ( f g)(p) = X p ( f g) = X p ( f ) g(p) + f (p) X p (g) = [ XV ( f ) g + f XV (g) ] (p)
for every p ∈ V and f , g ∈ F (V ).
VECTOR FIELDS AND DIFFERENTIAL EQUATIONS 49
3.7 Lemma Let M be a smooth manifold, and let F : F (M) → F (M) be a

derivation of the algebra F (M). If f1 , f2 ∈ F (M) are smooth functions with f1 |V =
f2 |V for an open subset V of M, then we have that F( f1 )|V = F( f2 )|V .
PROOF : Considering the difference f2 − f1 , it is clearly enough to prove that
F( f )|V = 0 for each smooth function f ∈ F (M) with f |V = 0. To show this, let
p ∈ V and choose a smooth function h ∈ F (M) with h = 1 on M − V and h(p) = 0.
Such a function h exists by Corollary 2.24. Then we have that
F( f )(p) = F(h f )(p) = F(h)(p) f (p) + h(p) F( f )(p) = 0
which completes the proof since p was an arbitrary point in V .
3.8 Proposition Let M n be a smooth manifold, and let F : F (M) → F (M) be a

derivation of F (M). Then there is a unique vector field X on M such that F = XM .
PROOF : We first note that if f ∈ F p is a smooth function defined on an open neigh-
bourhood V of a point p on M, then there exists a smooth function g ∈ F (M) which
coincides with f on an open neighbourhood W of p with W ⊂ V . Indeed, there is
a smooth function h : M → R with h(p) = 1 on W and supp (h) ⊂ V , and we may
extend h f to a smooth function g on M by defining it to be zero outside V .
Then by Lemma 2.78, we must define
X p ( f ) = F(g)(p) ,
and it follows from Lemma 3.7 that X p ( f ) does not depend on the choice of W and
g. Furthermore, we see the X p is a local derivation at p since F is a derivation on
F (M).
It only remains to show that X is smooth on M. By Proposition 3.2 (1) we must
show that the functions ai : U → R defined by
ai (p) = X p (xi )
for p ∈ U and i = 1 , ..., n, are smooth for every local chart (x,U) on M. Fix p in U,
and let W be an open neighbourhood of p with W ⊂ U. In the same way as above we
extend each function h xi on U to a smooth function yi on M so that they coincide on
W . Then we have that
ai (q) = F(yi ) (q)
for q ∈ W showing that ai is smooth at p. Since p was an arbitrary point in U, this
completes the proof that X is smooth.
3.9 Remark Because of Proposition 3.8 we will not distinguish between the vec-
tor field X and the map XM , and a vector field on M can be thought of as a derivation
of the algebra F (M) of C∞ -functions on M.
3.10 Definition Let f : M → N be a smooth map, and let X and Y be vector fields
on M and N, respectively. We say that X and Y are f -related if the diagram
f∗✲
TM TN
✻ ✻
X Y
f
M ✲ N
is commutative, i.e., if f∗ (X p ) = Y f (p) for every p ∈ M.
3.11 Proposition If f : M → N is a smooth map, then the vector fields X and Y

on M and N, respectively, are f -related if and only if
YN (g) ◦ f = XM (g ◦ f )
for every smooth function g : N → R.

PROOF : Follows from Definition 3.10 since
YN (g) ( f (p)) = Y f (p) (g)
and
XM (g ◦ f ) (p) = X p (g ◦ f ) = f∗ (X p ) (g)
for every p ∈ M.
3.12 Remark If f : M → N is a smooth map and X is a vector field on M, there

may be no vector field Y on N which is f -related to X. Neither is there always a
vector field X on M which is f -related to a given vector field Y on N.
However, in one important special case such f -related vector fields always exist.
3.13 Definition Let f : M → N be a diffeomorphism.

(1) If X is a vector field on M, there is a unique vector field Y on N which is
f -related to X given by
Y = f∗ ◦ X ◦ f −1 .
Y is denoted by f∗ (X) and is called the push-forward of X by f .
(2) If Y is a vector field on N, there is a unique vector field X on M which is
f -related to Y given by
X = ( f∗ )−1 ◦ Y ◦ f .
X is denoted by f ∗ (Y ) and is called the pull-back of Y by f . Note that f ∗ (Y ) =
( f −1 )∗ (Y ).
3.14 Proposition Let f : M n → N m be an immersion, and let Y be a vector field

on N with Y f (p) ∈ f∗ (Tp M) for each p ∈ M. Then there is a unique vector field X on
M which is f -related to Y .
PROOF : For each p ∈ M there is a unique tangent vector X p ∈ Tp M with f∗ (X p ) =
Y f (p) . To show that the map X : M → T M is smooth, let p be an arbitrary point on
M. By the rank theorem there are local charts (x,U) and (y,V ) around p and f (p),
respectively, such that
y ◦ f ◦ x−1 (a1 , ..., an ) = (a1 , ..., an , 0, ..., 0)
for a ∈ x(U), which implies that

!
∂

f∗ = ∂ i
∂ xi ∂y
q f (q)
for q ∈ U and i = 1, ... , n.

By Proposition 3.2 (1) the vector field Y has a local representation b : V → Rm
which is smooth and is given by
m
Y | V = ∑ bi ∂ i .
i=1
∂y
Hence if
n
X| U = ∑ ai ∂ i ,
i=1
∂x
we have that
n
Y f (q) = f∗ (Xq ) = ∑ ai (q) ∂ i
i=1
∂y
f (q)
so that ai (q)= bi ( f (q))

for q ∈ U and i = 1, ... , n. This shows that the local repre-
sentation a : U → Rn for X on U is smooth and hence that X is a vector field on M
which is f -related to Y .
INTEGRAL CURVES AND LOCAL FLOWS

3.15 Definition Let X be a vector field on a smooth manifold M n . A smooth
curve γ : I → M defined on an open interval I is called an integral curve for X if
γ ′ (t) = X(γ (t)) (1)
for t ∈ I. If I contains 0 and γ (0) = p0 , the point p0 is called the starting point or
initial condition of γ .
The integral curve γ is called maximal if it has no extension to an integral curve

for X on any larger open interval.
By a local flow for X at a point p0 on M we mean a map
α : I × U0 → M ,
where I is an open interval containing 0, and U0 is an open neighbourhood of p0 ,

such that the curve α p : I → M defined by α p (t) = α (t, p) is an integral curve for X
with initial condition p for each point p in U0 .
3.16 Remark If (x,U) is a local chart around p0 , the problem of finding an

integral curve γ : I → U for X|U with initial condition p0 can be transformed to a
similar problem on x(U). The coordinate map x may be written as x = i ◦ x̃, where
x̃ : U → x(U) is a diffeomorphism and i : x(U) → Rn is the inclusion map. Suppose
that
n
X| U = ∑ ai ∂ i ,
i=1
∂x
and let f = a ◦ x−1 , c = x̃ ◦ γ and x0 = x(p0 ). By applying tx to both sides of (1) in

Definition 3.15 and using Definition 2.69 and the proof of Proposition 3.2 (1), we
obtain
(γ (t), (x ◦ γ )′ (t)) = (γ (t), a(γ (t)))
which is equivalent to
c′ (t) = f (c(t))
for t ∈ I, and we have that c(0) = x0 . We say that f is a vector field on x(U), and c is
called an integral curve for f with initial condition x0 .
If α : I × U0 → U is a local flow for X|U at p0 , where I is an open interval
containing 0 and U0 is an open subset of U containing p0 , then the map
β : I × x(U0 ) → x(U)
defined by β = x̃ ◦ α ◦ (id × x−1 ) is called a local flow for f at x0 . Indeed, for each
point u = x(p) in x(U0 ), we have that βu = x̃ ◦ α p is an integral curve for f with initial
condition u if and only if α p is an integral curve for X with initial condition p, so the
problem of finding a local flow for X|U on U is transformed to a similar problem on
x(U).
3.17 Definition Let U be an open subset of a finite dimensional normed vector

space E. By a time-dependent vector field on U we mean a continuous map f :
J × U → E where J is an open interval containing 0. A curve c : I → U of class C1
defined on an open subinterval I of J is called an integral curve for f if
c′ (t) = f (t, c(t)) (1)
for t ∈ I. We see by induction that if the map f is smooth, so is c. If I contains 0 and

c(0) = x0 , the point x0 is called the starting point or initial condition of c.
The integral curve c is called maximal if it has no extension to an integral curve

for f on any larger open subinterval of J.
3.18 Remark A continuous curve c : I → U is an integral curve for f with

c(t0 ) = x0 if and only if it satisfies the integral equation
Zt
c(t) = x0 + f (u, c(u)) du
t0
3.19 Definition Let f : J × U → E be a time-dependent vector field, and let x0

be a point in U. By a local flow for f at x0 we mean a map
α : I × U0 → U ,
where I is an open subinterval of J containing 0, and U0 is an open subset of U
containing x0 , such that the curve αx : I → U defined by αx (t) = α (t, x) is an integral
curve for f with initial condition x, i.e., such that
D1 α (t, x) = f (t, α (t, x)) and α (0, x) = x
for all t ∈ I and x ∈ U0 .
3.20 The contraction lemma Let M be a nonempty complete metric space with
distance function d, and let f : M → M be a contraction, i.e., there is a constant C
with 0 ≤ C < 1 such that
d ( f (x), f (y)) ≤ C d (x, y)
for all x, y ∈ M. Then f has a unique fixed point, i.e., a point x such that f (x) = x. If
x0 is an arbitrary point in M, we have that x = limn→∞ f n (x0 ).
PROOF : If x and y are two fixed points for f , we have that
d (x, y) = d ( f (x), f (y)) ≤ C d (x, y)
which implies that d (x, y) = 0 since C < 1. This shows that x = y and proves the
uniqueness of the fixed point for f .
To show existence, let x0 be an arbitrary point in M, and let xn+1 = f (xn ) for
n = 0, 1 , .... By induction we have that
d (xi , xi+1 ) ≤ Ci d (x0 , x1 ) ,
so if n < m, we have
m−1 m−1 n m
−C
d (xn , xm ) ≤ ∑ d (xi, xi+1 ) ≤ ( ∑ Ci ) d (x0, x1 ) = C1 −C d (x0 , x1 ) .
i=n i=n
As {Cn } is a Cauchy sequence, so is {xn }. It therefore converges to a point x in M

since M is complete. Continuity of f then shows that
f (x) = lim f (xn ) = lim xn+1 = x
n→∞ n→∞
which completes the proof that f has a unique fixed point.

3.21 Definition A mapping f : U → E defined on a subset U of a finite dimen-

sional normed vector space E is said to be Lipschitz if there is a constant K such
that
k f (x) − f (y) k ≤ K k x − y k
for all x, y ∈ U. We call K a Lipschitz constant for f on U. We say that f is locally
Lipschitz if it is Lipschitz in a neighbourhood of each point in U.
If J is an open interval, then a mapping f : J × U → E is said to be Lipschitz on
U uniformly with respect to J if there is a constant K such that
k f (t, x) − f (t, y) k ≤ K k x − y k
for all x, y ∈ U and t ∈ J.
We say that f is locally Lipschitz on U uniformly with respect to J if for each
point (t0 , x0 ) in J × U there is an open subinterval J0 of J containing t0 and a neigh-
bourhood U0 of x0 in U such that f is Lipschitz on U0 uniformly with respect to J0 .
3.22 Proposition If f1 , f2 : X → Y are two continuous maps between the topo-

logical spaces X and Y , where Y is Hausdorff, then the set D = {x ∈ X | f1 (x) = f2 (x)}
is closed in X.
PROOF : We have that D = f −1 (∆), where f : X → Y × Y is the continuous map
defined by f (x) = ( f1 (x), f2 (x)) for x ∈ X, and ∆ ⊂ Y × Y is the diagonal which is
closed since Y is Hausdorff.
3.23 Proposition Let f : J × U → E be a time-dependent vector field which is

locally Lipschitz on U uniformly with respect to J, and let (t0 , x0 ) be a point in J ×U.
Choose an open subinterval J0 of J containing t0 and a positive real number a such
that the closed ball Ba (x0 ) is contained in U and there are positive constants K and L
such that
(1) k f (t, x) − f (t, y) k ≤ K k x − y k for x, y ∈ Ba (x0 ) and t ∈ J0 ,

(2) k f (t, x) k ≤ L for x ∈ Ba (x0 ) and t ∈ J0 .
If Ib ⊂ J0 is the open interval (t0 − b,t0 + b), where b is a positive real number with
b < 1/K and b ≤ a/L, then there is a unique integral curve α : Ib → U for f defined
on Ib with α (t0 ) = x0 .
If αi : Ii → U where i = 1, 2 are two integral curves for f with α1 (s) = α2 (s) for
some real number s in I1 ∩ I2 , then α1 and α2 are equal on I1 ∩ I2 .
PROOF : Let M be the set of all continuous curves
α : Ib → Ba (x0 ) .
Then M is a nonempty compete metric space with the usual supremum metric given
by
d (α , β ) = sup k α (t) − β (t) k .
t∈Ib
We define the map S : M → M by

Z t
S(α )(t) = x0 + f (u, α (u)) du .
t0
Indeed we have that

Z t

k S(α )(t) − x0 k = f (u, α (u)) du

< bL ≤ a
t 0
for every t ∈ Ib which shows that S(α ) is a continuous curve in Ba (x0 ) and hence
belongs to M. Since
Z t

d (S(α ), S(β )) ≤ sup
t [ f (u, α (u)) − f (u, β (u)) ] du ≤ bK d (α , β )

t∈Ib 0
and bK < 1 , we have that S is a contraction and hence has a unique fixed point
c by the contraction lemma. If i : Ba (x0 ) → U is the inclusion map, it follows by
Remark 3.18 that the curve α = i ◦ c is the unique integral curve for f defined on Ib
with α (t0 ) = x0 and α (Ib ) ⊂ Ba (x0 ). This shows the existence of an integral curve
α : Ib → U for f with α (t0 ) = x0 , and uniqueness will follow from the last part of
the proposition.
Let I = I1 ∩I2 and I0 = {t ∈ I|α1 (t) = α2 (t)}. Then I0 is closed in I by Proposition
3.22. Since I is connected, it will follow that I0 = I if we can show that I0 is also open
and nonempty. If t1 ∈ I0 and α1 (t1 ) = α2 (t1 ) = x1 , choose Ib as in the first part of the
proposition with (t1 , x1 ) instead of (t0 , x0 ), and let in addition b be so small that Ib ⊂ I
and αi (Ib ) ⊂ Ba (x1 ) for i = 1, 2. Then it follows from the first part of the proof that
α1 and α2 are equal on Ib so that I0 is open. Since I0 contains s, it is also nonempty,
and this completes the proof that the curves α1 and α2 are equal on the intersection
of their domains.
3.24 Corollary Let f : J × U → E be a time-dependent vector field which is

locally Lipschitz on U uniformly with respect to J, and let (t0 , x0 ) be a point in J ×U.
Then there is a unique maximal integral curve α : I → U for f defined on an open
subinterval I of J containing t0 with α (t0 ) = x0 .
PROOF : Let I be the set of all open subintervals I ′ of J containing t0 such S
that
there exists an integral curve α ′ : I ′ → U for f with α ′ (t0 ) = x0 , and put I = I ′ ∈I I ′ .
Define the curve α : I → U as follows. If t ∈ I, choose an interval I ′ in I containing
t and set α (t) = α ′ (t). It follows by Proposition 3.23 that I is nonempty and that α
is well defined and uniquely determined by the above conditions.
3.25 Example Let f : (−2, 2) × R → R be the time-dependent vector field

defined by f (t, x) = tx2 . By separation of variables we see that the maximal in-
tegral curve for f with initial condition 2 is the curve α : (−1, 1) → R given by
α (t) = 2/(1 − t 2).
3.26 Remark In Example 3.25 we see that the maximal integral curve is defined
on a smaller time interval that the time-dependent vector field. The reason is that
α (t) → ∞ when t → −1+ and t → 1− , and we see that f (t, α (t)) is not bounded.
Another case when this can happen is when the integral curve tends to leave the
open set U where the time-dependent vector field is defined. For instance will the
domain of the maximal integral curve be smaller if we remove from U a point lying
on the curve.
The next proposition shows that apart from these two cases, the maximal integral
curve is always defined on the same time interval as the time-dependent vector field.
3.27 Proposition Let f : (a, b) × U → E be a time-dependent vector field which

is locally Lipschitz on U uniformly with respect to (a, b), and let α : (a0 , b0 ) → U
be a maximal integral curve for f . Assume that
(1) there exists an ε > 0 such that α ((b0 − ε , b0 )) ⊂ U ,

(2) there exists a B such that k f (t, α (t)) k ≤ B for t ∈ (b0 − ε , b0 ) .
Then we have that b0 = b, and a similar result holds for the left endpoints.
PROOF : By Remark 3.18 it follows from (2) that
Z t
1

k α (t1 ) − α (t2 ) k = f (u, α (u)) du
t2
≤ B |t1 − t2 |
for t1 ,t2 ∈ (b0 − ε , b0 ). Hence limt→b0 − α (t) exists and equals a point x0 in U by
hypothesis (1). If b0 < b, it follows from Proposition 3.23 that there is an integral
curve β for f defined on some open interval (b0 − ε1 , b0 + ε1 ) contained in (a, b)
with β (b0 ) = x0 . Since α ′ = β ′ on (b0 − ε1 , b0 ), we have that α and β only differ
by a constant on this interval. As their limit when t → b0 − are equal, this constant
must be zero. Hence α may be extended to an integral curve for f on a larger open
subinterval of (a, b), and this contradiction shows that b0 = b.
3.28 Theorem Let f : J × U → E be a time-dependent vector field which is

locally Lipschitz on U uniformly with respect to J, and let x0 be a point in U. Choose
an open subinterval J0 of J containing 0 and a positive real number a such that the
closed ball B2a (x0 ) is contained in U and there are positive constants K and L such
that
(1) k f (t, x) − f (t, y) k ≤ K k x − y k for x, y ∈ B2a (x0 ) and t ∈ J0 ,
(2) k f (t, x) k ≤ L for x ∈ B2a (x0 ) and t ∈ J0 .

If Ib ⊂ J0 is the open interval (−b, b), where b is a positive real number with b < 1/K
and b ≤ a/L, then there is a unique map α : Ib × Ba (x0 ) → U which is a local flow
for f at x0 , and α is Lipschitz on Ib × Ba (x0 ).
PROOF : If x is a point in Ba (x0 ), then Ba (x) ⊂ B2a (x0 ), so by Proposition 3.23 there
is a unique integral curve αx : Ib → U for f defined on Ib with αx (t0 ) = x. Hence the
map α : Ib × Ba (x0 ) → U defined by α (t, x) = αx (t) is the unique local flow for f
defined on Ib × Ba (x0 ).
It remains to show that α is Lipschitz on Ib ×Ba (x0 ). If x, y ∈ Ba (x0 ), we have that
sup k α (t, x) − α (t, y) k

t∈Ib
Z t

≤ k x − y k + sup
[ f (u, α (u, x)) − f (u, α (u, y)) ] du

t∈Ib 0
≤ k x − y k + bK sup k α (t, x) − α (t, y) k

t∈Ib
so that
sup k α (t, x) − α (t, y) k ≤ 1 −1bK k x − y k ,
t∈I
b
which shows that α (t, x) is Lipschitz in x uniformly in t. If s,t ∈ Ib and x ∈ Ba (x0 ),

we have that
Z t

k α (s, x) − α (t, x) k = f (u, α (u, x)) du

≤ L ks−t k ,
s
showing that α (t, x) is also Lipschitz in t uniformly in x. Hence α (t, x) is Lipschitz

in t and x jointly.
3.29 Remark We want to prove that if a time-dependent vector field f : J ×U →

E is of class Ck , then it has a local flow α at each point x0 in U which is also of class
Ck . The hard part is to prove that α is C1 in the second variable x. Assuming this
for the moment, we see from Definition 3.19 by differentiating with respect to x, that
D2 α = λ must satisfy the equation
D1 λ (t, x) = D2 f (t, α (t, x)) ◦ λ (t, x) with λ (0, x) = idE ,
which is a time-dependent linear differenial equation depending on x as a parameter.
In the next proposition we show that such an equation has a unique solution λ
which is continuous. Using this we will show in Theorem 3.33 that D2 α exists and
equals λ and, by an induction argument, that the local flow α is in fact Ck . We first
need a lemma.
3.30 Lemma (Gronwall’s inequality) Let f , g : [a, b) → R be continuous and

non-negative. Suppose that A ≥ 0 and that for t ∈ [a, b) we have
Zt
f (t) ≤ A + f (u) g(u) du . (1)
a
Then it follows that Z t
f (t) ≤ A exp ( g(u) du) (2)
a
for t ∈ [a, b). In particular, if A = 0, then f (t) = 0 for t ∈ [a, b).
PROOF : We first suppose that A > 0, and let h : [a, b) → R be the function defined
by Zt
h(t) = A + f (u) g(u) du
a
appearing on the right side of (1). Then h(t) > 0 and h′ (t) = f (t) g(t) ≤ h(t) g(t) so
that h′ (t) / h(t) ≤ g(t) for t ∈ [a, b). Integration gives
Z t
log h(t) − logA ≤ g(u) du
a
and hence Z t
h(t) ≤ A exp ( g(u) du)
a
for t ∈ [a, b). Since f (t) ≤ h(t), this shows (2) in the case when A > 0.
If (1) is satisfied for A = 0, then (1) and hence (2) is satisfied for every A > 0
which shows that f (t) = 0 for t ∈ [a, b).
3.31 Corollary If b < a, then Lemma 3.30 is still valid if we replace the interval
[a, b) by (b, a] and take absolute values of the integrals.
PROOF : Let f , g : [−a, −b) → R be the functions defined by f (t) = f (−t) and
g(t) = g(−t), and suppose that
Zt
f (t) ≤ A + | f (u) g(u) du |
a
for t ∈ (b, a]. Making the substitution s = −u in the integral, this implies that
Z −t
f (−t) ≤ A + f (s) g(s) ds .
−a
By Lemma 1.61 we now have that

Z −t
f (−t) ≤ A exp ( g(s) ds)
−a
so that Z t
f (t) ≤ A exp ( | g(u) du | )
a
for t ∈ (b, a].
3.32 Proposition Let J be an open interval containing 0, E and F be finite

dimensional normed vector spaces, and V be an open subset of F. Let
g : J × V → L(E, E)
be a continuous map. Then there is a unique map
λ : J × V → L(E, E)
such that
D1 λ (t, x) = g(t, x) ◦ λ (t, x) and λ (0, x) = id | E
for all t ∈ J and x ∈ V . Furthermore, this λ is continuous.
PROOF : Choose a point x in V , and let
f : J × L(E, E) → L(E, E)
be the time-dependent vector field defined by f (t, v) = g(t, x) ◦ v for t ∈ J and v ∈

L(E, E). From the inequality
k f (t, v1 ) − f (t, v2 ) k ≤ k g(t, x) k k v1 − v2 k
we see that f is locally Lipschitz on L(E, E) uniformly with respect to J, since

g(t, x) is bounded on J0 × {x} for any open interval J0 with compact closure
J0 ⊂ J.
Let λx : I → L(E, E) be the maximal integral curve for f with λx (0) = id | E . We
will show that I = J using Proposition 3.27. Let J = (a, b) and I = (a0 , b0 ), and
assume that b0 < b. Hypothesis (1) in Proposition 3.27 is clearly satisfied for ε = b0
as the open set U now is the whole space L(E, E). Since g is continuous, there is a
constant C such that k g(t, x) k ≤ C for t ∈ [0, b0 ]. From the integral equation
Z t
λx (t) = id | E + g(u, x) ◦ λx (u) du (1)
0
we therefore have that

Zt
k λx (t) k ≤ 1 + C k λx (u) k du
0
for every t ∈ [0, b0 ). By Gronwall’s inequality it follows that
k λx (t) k ≤ eCt ≤ eC b0
so that
k f (t, λx (t)) k ≤ k g(t, x) k k λx(t) k ≤ CeC b0
for t ∈ (0, b0 ). This shows that hypothesis (2) in Proposition 3.27 is also satisfied
with B = CeC b0 . The assumption b0 < b hence leeds to a contradiction. A similar
argument using Corollary 3.31 applies to a and a0 , and this completes the proof that
I = J.
We therefore have a unique map λ : J × V → L(E, E), defined by λ (t, x) = λx (t)
for t ∈ J and x ∈ V , which satisfies the linear differetial equation and the initial value
condition of the proposition. It only remains to show that λ is continuous. We know
from Definition 3.17 that λ is continuous in the first variable t for each fixed value
of x.
Let (t0 , x0 ) be a point in J × V , and let ε > 0. Choose an open inter-
val J0 containing 0 and t0 with compact closure J0 ⊂ J and a compact neigh-
bourhood V0 of x0 in V . Then there are positive constants C and K such that
k λ (t, x0 ) k ≤ C for t ∈ J0 and k g(t, x) k ≤ K for (t, x) ∈ J0 × V0 . Let l be the

length of the interval J0 , and let r = ε K e−Kl . Then there is an open subinterval J1 of
J0 containing t0 and an open neighbourhood V1 of x0 contained in V0 such that
k λ (t, x0 ) − λ (t0, x0 ) k < Kr (2)
for t ∈ J1 and
k g(t, x) − g(t, x0) k < Cr
for (t, x) ∈ J0 × V1 .
If x is an arbitrary point in V1 and ψ : J0 → R is the function defined by
ψ (t) = k λ (t, x) − λ (t, x0) k, it follows from the integral Equation (1) that
Z t
ψ (t) ≤ k g(u, x) ◦ λ (u, x) − g(u, x0) ◦ λ (u, x0 ) k du
0
Z t
≤ k g(u, x) k k λ (u, x) − λ (u, x0) k du
0
Z t
+ k g(u, x) − g(u, x0) k k λ (u, x0 ) k du
0
Z t Z t
≤K ψ (u) du + r t = K (ψ (u) + Kr ) du
0 0
so that Z t
ψ (t) + Kr ≤ Kr + K (ψ (u) + Kr ) du
0
for t ≥ 0 in J0 . By Gronwall’s inequality we therefore have that
ψ (t) + Kr ≤ Kr eKt ≤ Kr eKl ,
and hence
k λ (t, x) − λ (t, x0) k ≤ Kr (eKl − 1) (3)
for t ≥ 0 in J0 . A similar argument using Corollary 3.31 gives the same result for
t ≤ 0 in J0 . From (2) and (3) it now follows that
k λ (t, x) − λ (t0, x0 ) k ≤ k λ (t, x) − λ (t, x0) k + k λ (t, x0 ) − λ (t0 , x0 ) k < ε
for (t, x) ∈ J1 × V1 which shows that λ is continuous at (t0 , x0 ).
3.33 Theorem Let f : J × U → E be a time-dependent vector field of class Ck

for an integer k ≥ 1, and let x0 be a point in U. Then there are positive real numbers
a and b such that the local flow α : Ib × Ba (x0 ) → U in Theorem 3.28 is of class Ck .
PROOF : We first show that if f is of class C1 , then so is the local flow α : Ib ×
Ba (x0 ) → U in Theorem 3.28. We allready know that α is Lipschitz so that there are
positive constants A and B with
k α (t1 , x1 ) − α (t2 , x2 ) k ≤ A kt1 − t2 k +B k x1 − x2 k (1)
for t1 ,t2 ∈ Ib and x1 , x2 ∈ Ba (x0 ).

Let g : Ib × Ba (x0 ) → L(E, E) be the continuous map given by g(t, x) =
D2 f (t, α (t, x)), and let λ : Ib × Ba (x0 ) → L(E, E) be the unique solution of the time-
dependent linear differenial equation
D1 λ (t, x) = g(t, x) ◦ λ (t, x) with λ (0, x) = idE . (2)
We claim that D2 α exists and equals λ which is continuous by Proposition 3.32.

Let x be a point in Ba (x0 ), and choose positive real numbers c and d such that
Bd (x) ⊂ Ba (x0 ) and c < b. If θ : Ib × Bd (x) → E is the map defined by
θ (t, h) = α (t, x + h) − α (t, x) ,
we have that
θ (t, h) − λ (t, x) h
Z t
= [ f (u, α (u, x + h)) − f (u, α (u, x)) ] du − [ λ (t, x) − idE ] h
0
Z t
= [ f (u, α (u, x + h)) − f (u, α (u, x)) ] du
0
Zt (3)
− D2 f (u, α (u, x)) λ (u, x) h du
0
Z t
= [ f (u, α (u, x + h)) − f (u, α (u, x)) − D2 f (u, α (u, x)) θ (u, h) ] du
0
Zt
+ D2 f (u, α (u, x)) [ θ (u, h) − λ (u, x) h ] du
0
for t ∈ (−c, c) and h ∈ Bd (x).

By the continuity of g there is a constant M such that
k D2 f (u, α (u, x)) k ≤ M (4)
for every u ∈ [−c, c].

In order to estimate the first integral in the last expression in (3), we use the mean
value theorem which implies that
k f (u, α (u, x + h)) − f (u, α (u, x)) − D2 f (u, α (u, x)) θ (u, h) k
(5)
≤ k θ (u, h) k sup k D2 f (u, y) − D2 f (u, α (u, x)) k,
where the sup is taken over all y on the line segment between α (u, x) and α (u, x + h).
By (1) we have that k θ (u, h) k ≤ B k h k.
Given ε > 0, let k = ε e−Mc . For each u ∈ [−c, c] there is an open ball Bu ⊂ U of
radius δu centered at α (u, x) and an open interval Ju ⊂ Ib of radius ru centered at u
such that ru < δu / 2A and
k
k D2 f (s, z) − D2 f (u, α (u, x)) k < 2B
for all (s, z) ∈ Ju × Bu , by the continuity of D2 f at the point (u, α (u, x)). Since the
interval [−c, c] is compact, it is covered by a finite number of intervals Jui for i =
1, ..., n. Let
δ = min {δui / 2B|1 ≤ i ≤ n} ∪ {d} ,
and suppose that k h k < δ . If u ∈ [−c, c], then u ∈ Jui for some i. Hence it follows
from (1) that both α (u, x) and α (u, x + h) and hence every y on the line segment
between α (u, x) and α (u, x + h) belong to Bui . From this it follows that
k
k D2 f (u, y) − D2 f (ui , α (ui , x)) k < 2B
and
k
k D2 f (u, α (u, x)) − D2 f (ui , α (ui , x)) k < 2B
so that
k D2 f (u, y) − D2 f (u, α (u, x)) k < Bk (6)
Combining (6) with (3), (4) and (5), we obtain
Z t
k θ (t, h) − λ (t, x) h k ≤ k k h k + M k θ (u, h) − λ (u, x) h k du
0
for t ∈ [0, c). Gronwall’s inequality hence implies that
k θ (t, h) − λ (t, x) h k ≤ ε k h k
for t ∈ [0, c), and a similar argument using Corollary 3.31 gives the same result for
t ∈ (−c, 0]. Since for each point (t, x) in Ib × Ba (x0 ) we may always choose a positive
real number c with |t| < c < b, this completes the proof that D2 α exists and equals
λ so that the local flow α is of class C1 in the second variable x. From the equation
D1 α (t, x) = f (t, α (t, x)) we see that α is also of class C1 in the first variable t.
This shows the theorem in the case k = 1. Now assume inductively that it is true
for k − 1, and let f be of class Ck . We have to show that D1 α and D2 α are of class
Ck−1 . Let
G : Ib × Ba (x0 ) × L(E, E) → E × L(E, E)
be the time-dependent vector field on Ba (x0 ) × L(E, E) given by G(t, x, w) =
(0, g(t, x) ◦ w), and let
Λ : Ib × Ba (x0 ) × L(E, E) → Ba (x0 ) × L(E, E)
be the map defined by Λ(t, x, w) = (x, λ (t, x) ◦ w). By (2) we have that
D1 Λ(t, x, w) = (0 , g(t, x) ◦ λ (t, x) ◦ w) = G(t, x, λ (t, x) ◦ w) = G(t, Λ(t, x, w))

with
Λ(0, x, w) = (x, id|E ◦ w) = (x, w)
showing that Λ is a local flow for G. Since G is of class Ck−1 , it follows by the
induction hypothesis that there are positive real numbers a′ and b′ and some open
neighbourhood V of id|E in L(E, E) such that Λ is of class Ck−1 on Ib′ × Ba′ (x0 ) ×V .
Since λ (t, x) = pr2 ◦ Λ(t, x, id|E ), where pr2 : E × L(E, E) → L(E, E) is the projec-
tion on the second factor, we have that D2 α = λ is of class Ck−1 on Ib′ × Ba′ (x0 ).
From the equation D1 α (t, x) = f (t, α (t, x)) we see that this is also the case for D1 α ,
and this completes the induction step.
3.34 Remark It follows from the equation D1 α (t, x) = f (t, α (t, x)) that the local
flow α is in fact Ck+1 in the first variable t.
Since a and b depend on k as well as on x0 , we cannot yet conclude that the
theorem is true when k = ∞. However, this will follow from the following Remark
3.35 and Theorem 3.39.
3.35 Remark Let U be an open subset of a finite dimensional normed vector

space E. By a (time-independent) vector field on U we mean a continuous map
f : U → E . With f we may associate a time-dependent vector field g : R × U → E
on U defined by g(t, x) = f (x). By a local flow for f at a point x0 in U we mean a
local flow for g at x0 , i.e., a map
α : I × U0 → U ,
where I is an open interval containing 0, and U0 is an open subset of U containing x0 ,

such that
D1 α (t, x) = f (α (t, x)) and α (0, x) = x
for all t ∈ I and x ∈ U0 . The time-dependent vector fields are thus seen to include the
time-independent ones as a special case. We see that f is of class Ck on U if and only
if g is of class Ck on R ×U, and f is locally Lipschitz on U if and only if g is locally
Lipschitz on U uniformly with respect to R.
On the other hand, with a general time-dependent vector field g : J × U → E on
U, where J is an open interval containing 0, we may associate a time-independent
vector field h : J × U → R × E on J × U defined by h(t, x) = (1, g(t, x)). Let
β : I × I0 × U0 → J × U
be a local flow for h, where I and I0 are open subintervals of J containing 0, and U0
is an open subset of U containing x0 , so that
D1 β (t, s, x) = h(β (t, s, x)) and β (0, s, x) = (s, x)
for all t ∈ I and (s, x) ∈ I0 ×U0 . If β1 and β2 are the components of β , it follows that
β1 (t, s, x) = s + t, and hence we have that
D1 β2 (t, s, x) = g(s + t, β2(t, s, x)) and β2 (0, s, x) = x .

Then the map

α : I × U0 → U
defined by α (t, x) = β2 (t, 0, x) satisfies the equation
D1 α (t, x) = g(t, α (t, x)) with α (0, x) = x
showing that α is a local flow for g. We see that g is of class Ck on J × U if and
only if h is of class Ck , and if β is of class Ck , so is α . Hence the study of time-
dependent vector fields of class Ck is reduced to the study of time-independent ones
which we will consider from now on. On the other hand, there is no natural condition
on h corresponding to the condition that g is locally Lipschitz on U uniformly with
respect to J, which is precisely what was needed in the beginning of the proof of
Proposition 3.32, and this is the reason why we have studied time-dependent vector
fields explicitly in the preceding theory.
3.36 Proposition If X is a vector field on a smooth manifold M and p is a point

on M, then there is an integral curve γ : I → M for X with initial condition p.
If γi : Ii → M where i = 1, 2 are two integral curves for X with with γ1 (s) = γ2 (s)
for some real number s in I1 ∩ I2 , then γ1 and γ2 are equal on I1 ∩ I2 .
PROOF : The first part follows from Remark 3.16, Proposition 3.23 and Definition
3.17.
To prove the last part, let I = I1 ∩ I2 and I0 = {t ∈ I|γ1 (t) = γ2 (t)}. Then I0 is
closed in I by Proposition 3.22. Since I is connected, it will follow that I0 = I if we
can show that I0 is also open and nonempty. If t1 ∈ I0 and γ1 (t1 ) = γ2 (t1 ) = x1 , let
(x,U) be a local chart around x1 . By Proposition 3.23 we have that x ◦ γ1 and x ◦ γ2
and hence γ1 and γ2 are equal on γ1 −1 (U) ∩ γ2 −1 (U) so that I0 is open. Since I0
contains s, it is also nonempty, and this completes the proof that the curves γ1 and γ2
are equal on the intersection of their domains.
3.37 Corollary If X is a vector field on a smooth manifold M and p is a point

on M, then there is a unique maximal integral curve γ : I → M for X with initial
condition p.
PROOF : Proved in the same way as Corollary 3.24.
GLOBAL FLOWS
3.38 Definition Let X be a vector field on a smooth manifold M. For each point
p on M we denote by γ p : I(p) → M the maximal integral curve for X with initial
condition p. The set
D(X) = {(t, p) ∈ R × M|t ∈ I(p)}
is called the domain of the flow for X, and the (global) flow for X is the map γ :
D(X) → M defined by γ (t, p) = γ p (t) for p ∈ M and t ∈ I(p).
3.39 Theorem Let X be a vector field on a smooth manifold M. Then D(X) is
an open subset of R × M containing {0} × M, and the flow γ : D(X) → M for X is
smooth.
PROOF : It follows immediately from Definition 3.38 that {0} × M ⊂ D(X). Given
an integer k ≥ 1, we will show that D(X) is open and that the flow γ is of class
Ck on D(X). For each point p on M we let I * (p) be the set of points t in I(p)
for which there is a real number b > 0 and an open neighbourhood U of p such
that (t − b,t + b) × U ⊂ D(X) and γ is of class Ck on this product. Then I * (p) is
clearly open in I(p) and contains 0 by Theorem 3.33 and Remark 3.16. Since I(p) is
connected, it will follow that I * (p) = I(p) if we can show that I * (p) is also closed
in I(p).
Let s ∈ I(p) be a point in the closure of I * (p). By Theorem 3.33 and Remark
3.16 there is an open neighbourhood V of γ (s, p), a real number a > 0 and a unique
local flow β : (−a, a) ×V → M for X which is of class Ck . Since the integral curve γ p
is continuous, we have that γ p−1 (V ) is an open neighbourhood of s. Let t1 ∈ I * (p) ∩
γ p−1 (V ) ∩ (s − a, s + a), and choose a real number b > 0 and an open neighbourhood
U of p such that (t1 − b,t1 + b) ×U ⊂ D(X) and γ is of class Ck on this product. By
choosing smaller b and U if necessary, we may assume that γ ((t1 − b,t1 + b) ×U) ⊂
V.
Now let α : (t1 − a,t1 + a) × U → M be the map defined by
α (t, q) = β (t − t1, γ (t1 , q)).
Then α is of class Ck , and we have that α (t1 , q) = β (0, γ (t1 , q)) = γ (t1 , q) for every
q ∈ U. For each such q we therefore have that αq is an integral curve for X defined
on the interval (t1 − a,t1 + a) with αq (t1 ) = γq (t1 ). Since γq is maximal, it follows
that (t1 − a,t1 + a) ×U ⊂ D(X) and that γ coincides with α and hence is of class Ck
on this product. Since s ∈(t1 − a,t1 + a), this implies that s ∈ I * (p) which completes
the proof of the theorem.
3.40 Proposition Let X be a vector field on a smooth manifold M, and let γ be

its flow. If p is a point on M and t belongs to I(p), then
I(γ (t, p)) = I(p) − t ,
and we have that
γ (s, γ (t, p)) = γ (s + t, p)
for all s ∈ I(p) − t.
PROOF : The curves γ1 : I(γ (t, p)) → M and γ2 : I(p) − t → M defined by
γ1 (s) = γ (s, γ (t, p)) and γ2 (s) = γ (s + t, p)
are both maximal integral curves for X with initial condition γ (t, p) and hence are
equal.
3.41 Definition If X is a vector field on a smooth manifold M with global flow

γ : D(X) → M, and if t ∈ R, we define
Dt (X) = {p ∈ M|(t, p) ∈ D(X)}
and let γt : Dt (X) → M be the map defined by γt (p) = γ (t, p).
3.42 Proposition For each s,t ∈ R we have that

(1) Dt (X) is open in M.
(2) If p ∈ Dt (X) and γt (p) ∈ Ds (X), then p ∈ Ds+t (X) and γs (γt (p)) = γs+t (p). In
particular the domain of γs ◦ γt is contained in Ds+t (X), and it equals Ds+t (X)
if s and t both have the same sign.
(3) γt (Dt (X)) = D−t (X), and γt is a diffeomorphism onto its image with inverse
γ−t .
PROOF : (1) Follows from Theorem 3.39 since Dt (X) = it−1 (D(X)), where it : M →
R × M is the continuous map defined by it (p) = (t, p).
(2) By Definition 3.41 and 3.38 the assertions p ∈ Dt (X) and γt (p) ∈ Ds (X) are
equivalent to t ∈ I(p) and s ∈ I(γ (t, p)), respectively. Hence by Proposition 3.40 we
have that s+t ∈ I(p) and γ (s, γ (t, p)) = γ (s+t, p) which is equivalent to p ∈ Ds+t (X)
and γs (γt (p)) = γs+t (p). This proves the first part of (2) and implies that the domain
of γs ◦ γt is contained in Ds+t (X).
Now assume that s and t both have the same sign, and let p ∈ Ds+t (X) which
is equivalent to s + t ∈ I(p). Since I(p) is an open interval around 0, it follows that
t ∈ I(p) and s ∈ I(p) − t = I(γ (t, p)) which again is equivalent to p ∈ Dt (X) and
γt (p) ∈ Ds (X), showing that p is in the domain of γs ◦ γt .
(3) Let p ∈ Dt (X) which is equivalent to t ∈ I(p) by Definition 3.41 and 3.38. Since
I(p) also contains 0, we have that −t ∈ I(p) − t = I(γ (t, p)) which again is equivalent
to γt (p) ∈ D−t (X). By taking s = −t in part (2), it follows that γ−t (γt (p)) = p for
every p ∈ Dt (X). Replacing t with −t we also have that γt (γ−t (p)) = p for every
p ∈ D−t (X), and this completes the proof of (3).
3.43 Remark The domain of γs ◦ γt is generally not equal to Ds+t (X) in part (2)
of the preceding proposition if s and t have opposite sign. For instance, when s = −t,
the domain of γs ◦ γt is Dt (X), whereas Ds+t (X) = M.
3.44 Definition A vector field X on a smooth manifold M is said to be complete if

D(X) = R × M, where D(X) is the domain of the flow for X as defined in Definition
3.38. This means that the maximal integral curve γ p for X with initial condition p is
defined on (−∞, ∞) for every p ∈ M.
In this case the maps γt in Definition 3.41 are defined on M for every t ∈ R, and
the family {γt |t ∈ R} is a group of diffeomorphisms on M parametrized by the real
numbers called the 1-parameter group of X. When X is not complete, this family is
not a group, but is referred to as the local 1-parameter group of X.
3.45 Proposition Let X be a vector field on a smooth manifold M, and let α : I →

M be a maximal integral curve for X. If α ((a, b)) is contained in a compact subset of
M for each finite open subinterval (a, b) of I, we have that I = R.
PROOF : It suffices to show that a ∈ I and b ∈ I for each such finite open subinterval
(a, b) of I. Let {tn } be a sequence in (a, b) which converges to b. Since α ((a, b))
is contained in a compact set, there is a subsequence {α (tni )} of {α (tn )} which
converges to some point p on M. As the domain of the flow γ : D(X) → M for X is
an open subset of R × M containing {0} × M, there is a real number δ > 0 and an
open neighbourhood U of p such that (−δ , δ ) × U ⊂ D(X). Choose i so large that
tni ∈ (b − δ , b) and α (tni ) ∈ U, and let β : (tni − δ , tni + δ ) → M be the curve defined
by β (t) = γ (t − tni , α (tni )). Then β is an integral curve for X with β (tni ) = α (tni ),
and since α is maximal, it follows that b ∈ (a,tni + δ ) ⊂ I. A similar argument holds
for a.
3.46 Corollary Let X be a vector field with compact support on a smooth mani-
fold M. Then X is complete.
PROOF : We must show that I(p) = R for each point p on M. Let K be the support
of X, and let γ be its flow. Suppose first that γ p (s) = q ∈
/ K for some real number
s ∈ I(p). Then it follows from Proposition 3.36 that γ p must coincide on I(p) with
the constant curve c : R → M defined by c(t) = q for all t, and since γ p is maximal,
we have that I(p) = R.
If, on the other hand, the above assumption is not satisfied, we have that
γ p (I(p)) ⊂ K, in which case I(p) = R follows from Proposition 3.45.
3.47 Corollary A vector field on a compact smooth manifold M is complete.

PROOF : Follows immediately from Corollary 3.46 since each such vector field has
compact support.
3.48 Proposition Let f : M → N be a smooth map, and let X and Y be vector fields
on M and N, respectively, which are f -related. If c : I → M is an integral curve for
X with initial condition q, then f ◦ c is an integral curve for Y with initial condition
f (q).
PROOF : If c is an integral curve for X, then
c ′ (t) = X (c (t))
for t ∈ I. From this it follows that
( f ◦ c) ′ (t) = f∗ ◦ c ′ (t) = f∗ ◦ X ◦ c (t) = Y ◦ f ◦ c (t) = Y ( f ◦ c (t))

which shows that f ◦ c is an integral curve for Y . If c has initial condition c(0) = q,
then f ◦ c has initial condition ( f ◦ c)(0) = f (q).
3.49 Proposition Let f : M → N be a diffeomorphism, and let X and Y be vector

fields on M and N, respectively, which are f -related. Then c : I → M is an integral
curve for X with initial condition q if and only if f ◦ c is an integral curve for Y with
initial condition f (q).
If γ : D (X) → M is the flow for X, then the flow for Y is given by β : D (Y ) → N
where
D (Y ) = (id × f ) (D (X)) and β = f ◦ γ ◦ (id × f −1 ) ,
and we have that
D t (Y ) = f (D t (X)) and β t = f ◦ γ t ◦ f −1
for t ∈ R .
PROOF : The first part of the proposition follows from Proposition 3.48 applied to f
and f −1 .
Now, let γq : I(q) → M and β p : J(p) → N be the maximal integral curves for X
and Y , respectively, with initial conditions q and p. By the first part of the proposition,
we have that
J( f (q)) = I(q) and β f (q) = f ◦ γq (1)
for q ∈ M.
From the first relation in (1) it follows that (t, f (q)) ∈ D (Y ) if and only if (t, q) ∈
D (X), which shows that D (Y ) = (id × f ) (D (X)). For a fixed t ∈ R, we have that
f (q) ∈ D t (Y ) if and only if q ∈ D t (X) so that D t (Y ) = f (D t (X)).
From the second relation in (1) we have that β (t, p) = f ◦ γ (t, f −1 (p)) for
(t, p) ∈ D (Y ), which implies that β = f ◦ γ ◦ (id × f −1 ). For a fixed t ∈ R, we have
that β t (p) = f ◦ γ t ( f −1 (p)) for each p ∈ D t (Y ) so that β t = f ◦ γ t ◦ f −1 .
3.50 Corollary Let f : M → M be a diffeomorphism, and let X be a vector field

on M with flow γ . Then f∗ (X) = X if and only if γ t ◦ f = f ◦ γ t for every t ∈ R.
PROOF : If β is the flow for f∗ (X), it follows from Proposition 3.49 that β t =
f ◦ γ t ◦ f −1 for every t ∈ R. Hence it is enough to show that f∗ (X) = X if and only
if β = γ . Assuming that β = γ , we have that
f∗ (X)(q) = f∗ (X)(β q (0)) = β q′ (0) = γ q′ (0) = X (γ q (0)) = X (q)
for every q ∈ M which shows that f∗ (X) = X, and the converse statement is obviously
true.
Chapter 4
TENSORS
With the tangent bundle of a smooth manifold described in Chapter 2 at our disposal,
we can now form new vector bundles by replacing the tangent space at each point
with new vector spaces of various types, gluing them together in a smooth way. The
sections in these new vector bundles will be covector fields and various types of
tensor fields, including differential forms which will be investigated more fully in the
next chapter. We also describe the Lie derivative of tensor fields, and the fundamental
integrability theorems of Frobenius for integral manifolds of distributions.
DUAL VECTOR BUNDLES
4.1 If V is a vector space, we let V ∗ denote the dual space L(V, R) consisting
of all linear functionals on V . If F : V → W is a linear map, we have a linear map
F ∗ : W ∗ → V ∗ defined by F ∗ (λ ) = λ ◦ F so that the diagram
F✲
V W
❙ ✓
F ∗ (λ ) ❙ ✓ λ
✇ ✴
❙ ✓
R
is commutative.
4.2 Proposition Let U, V and W be vector spaces.

(1) If F : U → V and G : V → W are linear maps, we have that (G ◦ F)∗ = F ∗ ◦ G∗ .
(2) If id : V → V is the identity on V , then id ∗ : V ∗ → V ∗ is the identity on V ∗ .
(3) If F : V → W is an isomorphism, so is F ∗ : W ∗ → V ∗ with (F ∗ )−1 = (F −1 )∗ .
PROOF : Follows immediately from the definition.
69
4.3 Proposition If {v1 , ..., vn } is a basis for a vector space V , we have a basis
{v∗1 , ..., v∗n } for the dual space V ∗ consisting of the linear functionals v∗i defined by
v∗i (v j ) = δi, j for j = 1, ..., n. If λ is a linear functional in V ∗ , we have that
n
λ = ∑ λ (vi ) v∗i .
i=1
{v∗1 , ..., v∗n } is called the dual basis of {v1 , ..., vn }.

PROOF : If v = ∑ni=1 ai vi , we have that
n n
λ (v) = ∑ ai λ (vi ) = ∑ λ (vi ) v∗i (v)
i=1 i=1
which shows the last assertion of the proposition and also that the set {v∗1 , ..., v∗n }
spans V ∗ . To show that it is linearly independent, assume that ∑ni=1 ai v∗i = 0. By
applying both sides to v j , we see that a j = 0 for j = 1, ..., n.
4.4 Proposition For each vector space V of dimension n we have a natural iso-
morphism iV : V → V ∗∗ given by iV (v)(λ ) = λ (v) for v ∈ V and λ ∈ V ∗ , i.e., iV is an
isomorphism such that the diagram
F
V ✲ W
iV iW
❄ ❄
F ∗∗✲
V ∗∗ W ∗∗
is commutative for every linear map F : V → W .

PROOF : Let {v1 , ..., vn } be a basis for V with dual basis {v∗1 , ..., v∗n }, and let v =
∑ni=1 ai vi be a vector in ker(iV ). Then
0 = iV (v)(v∗j ) = v∗j (v) = a j
for j = 1, ..., n so that v = 0, which shows that iV is injective. Since dimV ∗∗ = n by

Proposition 4.3, it follows that iV is an isomorphism.
To show that it is natural, let F : V → W be a linear map, and let v ∈ V and λ ∈ V ∗ .
Then we have that
F ∗∗ (iV (v)) (λ ) = iV (v) (F ∗ (λ )) = F ∗ (λ )(v) = λ (F(v)) = iW (F(v)) (λ )
showing that the diagram is commutative.

TENSORS 71
4.5 Remark We will often identify V ∗∗ with V by means of the natural isomor-
phism iV , and we may consider a vector v ∈ V to be a functional on V ∗ mapping
λ ∈ V ∗ onto λ (v).
4.6 Proposition Let F : V → W be a linear map, and let B and C be bases for V
and W , respectively, with dual bases B ∗ and C ∗ . Then we have that
∗
mCB∗ (F ∗ ) = mB t
C (F) .
PROOF : Let B = {v1 , ..., vn } and C = {w1 , ..., wm }. If mB

C (F) = A, we have that
m
F(v j ) = ∑ Ai j wi
i=1
for j = 1, ..., n. From this it follows that
F ∗ (w∗j ) (vi ) = (w∗j ◦ F)(vi ) = A ji
which shows that

n
F ∗ (w∗j ) = ∑ A ji v∗i
i=1
for j = 1, ..., m, and this completes the proof of the proposition.
4.7 Now let π : E → M be an n-dimensional vector bundle over the smooth

manifold M, and let E p = π −1 (p) be the fibre over p for each p ∈ M. We want to
define a new bundle [
E∗ = E p∗ ,
p∈M
called the dual bundle of E , by replacing each fibre E p by its dual space. We have a
projection π ∗ : E ∗ → M defined by π ∗ (λ ) = p for λ ∈ E p∗ .
4.8 Proposition π ∗ : E ∗ → M is an n-dimensional vector bundle over M.

PROOF : Let E = {e1 , ..., en } be the standard basis for Rn , and let E ∗ = {e1 , ..., en }
be its dual basis. We then have an isomorphism r : (Rn )∗ → Rn defined by r(ei ) = ei
for i = 1, ..., n.
Let {(tα , π −1 (Uα )) | α ∈A} be a trivializing cover of E. For each α ∈A we have a
bijection tα′ : π ∗ −1 (Uα ) → Uα × Rn defined by tα′ (λ ) = (p,tα′ ,p (λ )) for λ ∈ π ∗ −1 (p),
where tα′ ,p = r ◦ (tα−1 ∗
,p ) . We will show that these bijections satisfy the conditions of
Proposition 2.48.
For each pair of indices α , β ∈ A, we have that tα ◦ tβ−1 is a diffeomorphism from
(Uα ∩Uβ ) × Rn onto itself given by
(tα ◦ tβ−1 )(p, v) = (p, (tα ,p ◦ tβ−1

,p )(v)) ,
where tα ,p ◦ tβ−1
,p
: Rn → Rn is a linear isomorphism for each p ∈Uα ∩Uβ . Since
tβ′ ,p ◦ tα′ −1 −1 ∗
,p = r ◦ (tα ,p ◦ tβ ,p ) ◦ r
−1
,
∗
it follows from Proposition 4.6, using the fact that mEE (r) and mEE ∗ (r−1 ) are both the
identity matrix, that
mEE (tβ′ ,p ◦ tα′ −1 E −1 t
,p ) = mE (tα ,p ◦ tβ ,p ) .
Hence tβ′ ◦ tα′ −1 is also a diffeomorphism from (Uα ∩ Uβ ) × Rn onto itself which is
a linear isomorphism on each fibre. By Proposition 2.48 there is therefore a unique
topology and smooth structure on E ∗ such that π ∗ : E ∗ → M is an n-dimensional
vector bundle over M with (tα′ , π ∗ −1 (Uα )) as local trivializations.
4.9 Remark We say that (t ′ , π ∗ −1 (U)), where t p′ = r ◦ (t p−1 )∗ for p ∈ U, is the lo-
cal trivialization in the dual bundle π ∗ : E ∗ → M corresponding to the local trivializa-
tion (t, π −1 (U)) in the n-dimensional vector bundle π : E → M. Here r : (Rn )∗ → Rn
is the linear isomorphism defined by r(ei ) = ei for i = 1, ..., n, where E = {e1 , ..., en }
is the standard basis for Rn and E ∗ = {e1 , ..., en } is its dual basis.
4.10 Proposition Let π : E → M be an n-dimensional vector bundle over the

smooth manifold M. Then the bundle map iE : E → E ∗∗ over M given by iE (v)(λ ) =
λ (v) for v ∈ E p and λ ∈ E p∗ where p ∈ M, is an equivalence.
PROOF : We know from Proposition 4.4 that the induced map iE p : E p → E p∗∗ between
the fibres is a natural isomorphism for each p ∈ M. It only remains to prove that iE is
a diffeomorphism.
It is enough to show that t ′′ ◦ iE = t for each local trivialization (t, π −1 (U)) in a
trivializing cover of E, where (t ′′ , π ∗∗ −1 (U)) is the corresponding local trivialization
for E ∗∗ defined in Remark 4.9. We have that t p′ = r ◦ (t p −1 )∗ so that t p′′ = r ◦ (t p′ −1 )∗ =
r ◦ (r−1 )∗ ◦ t p ∗∗ for p ∈ M.
We first prove that
(r−1 )∗ ◦ i Rn = r−1 .
Let E = {e1 , ..., en } be the standard basis for Rn and E ∗ = {e1 , ..., en } be its dual
basis. Then we have that
(r−1 )∗ (i Rn (ei )) (e j ) = i Rn (ei ) (r−1 (e j )) = i Rn (ei ) (e j ) = e j (ei ) = δi j
for i, j = 1, ..., n which implies that
(r−1 )∗ (i Rn (ei )) = ei
for i = 1, ..., n , thereby proving the assertion.
From this it now follows that
t p′′ ◦ iE p = r ◦ (r−1 )∗ ◦ t p ∗∗ ◦ iE p = r ◦ (r−1 )∗ ◦ i Rn ◦ t p = t p
which completes the proof of the proposition.
TENSORS 73
4.11 Proposition Let (t, π −1 (U)) be a local trivialization for an n-dimensional

vector bundle π : E → M, and let (t ′ , π ∗ −1 (U)) be the corresponding local trivializa-
tion for its dual bundle π ∗ : E ∗ → M. Furthermore, let E = {e1 , ..., en } be the standard
basis for Rn and E ∗ = {e1 , ..., en } be its dual basis. If p is a point in U, then
−1 −1
B p∗ = {t p′ (e1 ), ...,t p′ (en ) } = { e1 ◦ t p , ..., en ◦ t p }
is a basis for the fibre π ∗ −1(p) which is dual to the basis B p for π −1 (p) given in
Remark 2.46. Moreover
B p∗∗ = { iE p (t p−1 (e1 )), ..., iE p (t p−1 (en )) } = { e1 ◦ t p′ , ..., en ◦ t p′ }
is a basis for the fibre π ∗∗ −1(p) in the double dual bundle π ∗∗ : E ∗∗ → M which may
be identified with B p according to Remark 4.5.
PROOF : Since t p′ = r ◦ (t p−1 )∗ , we have that
−1
t p′ (ei ) = t p∗ ◦ r−1 (ei ) = t p∗ (ei ) = ei ◦ t p ,
and
(ei ◦ t p )(t p−1 (e j )) = ei (e j ) = δi j
which shows that B p∗ is the dual basis of B p .
Using the notation in the proof of Proposition 4.10, we have that
−1
iE p (t p−1 (ei )) = t p′′ (ei ) = ei ◦ t p′
which shows the last part of the proposition.
4.12 Proposition Let π : E → M be an n-dimensional vector bundle and s : M →

E ∗ be a map with π ∗ ◦ s = idM , and let E ∗ = {e1 , ..., en } be the dual of the standard
basis for Rn . Then s is a section of π ∗ on M if and only if the map a : U → Rn defined
by
n
s(p) = ∑ ai (p) ei ◦ t p
i=1
for p ∈ U is smooth for every local trivialization (t, π −1 (U)) in a trivializing cover
of E.
PROOF : If (t ′ , π ∗ −1 (U)) is the corresponding local trivialization for the dual bundle
E∗ as defined in Remark 4.9, it follows that
t p′ (ei ◦ t p ) = r ◦ (t p−1 )∗ ◦ t p∗ (ei ) = r (ei ) = ei
so that
(t ′ ◦ s)(p) = (p, a(p))
for p ∈ U. Hence a is the local representation of s on U as defined in Definition
2.52.
4.13 If M is a smooth manifold of dimension n, the dual bundle π ∗ : T ∗ M → M

of the tangent bundle π : T M → M is called the cotangent bundle over M. The fibre
Tp∗ M over a point p in M is called the cotangent space of M at p.
If f : V → R is a smooth function defined on an open neighbourhood of a point p
in M, we define the differential of f at p to be the element d f (p) in Tp∗ M defined by
d f (p)(l) = l( f )
for each local derivation l in Tp M. By the discussion in 2.77 we know that this is the
derivative of f at p with respect to the tangent vector l.
It follows from Lemma 2.78 (1) that d f (p) only depends on the local behavior of
f around p, i.e., if fi : Vi → R for i = 1, 2 are two smooth functions defined on open
neighbourhoods Vi of p which coincide on an open neighbourhood of p contained in
V1 ∩V2 , then d f1 (p) = d f2 (p).
If (x,U) is a local chart around p, we have that
!
∂

dxi (p) = ∂ j (xi ) = δi j
∂xj ∂x
p p
which shows that the set {dx1 (p), ..., dxn (p)} of differentials of the coordinate func-
tions of the local chart (x,U) at p is a basis for Tp∗ M which is dual to the basis

{ ∂ 1 , ..., ∂∂xn } for Tp M of partial derivations at p with respect to (x,U).
∂x
p p
4.14 Proposition Let c : I → M be a smooth curve on a smooth manifold M

defined on an open interval I containing t, and let f : V → R be a smooth function
defined on an open neighbourhood V of c(t). Then we have that
( f ◦ c)′ (t) = d f (c(t)) (c′ (t)) .
PROOF : Follows from the discussion in 2.77 and 4.13.
4.15 Proposition If (tx , π −1 (U)) is the local trivialization for the tangent bundle
π : T M → M corresponding to a local chart (x,U) around a point p in M n as defined
in Remark 2.68, and if E ∗ = {e1 , ..., en } is the dual of the standard basis for Rn , then
we have that
dxi (p) = ei ◦ tx,p for i = 1, ... , n .
PROOF : Follows from Remark 2.82 and Proposition 4.11.
4.16 Definition By a covector field or a 1-form on an open subset V of a smooth

manifold M we mean a section of its cotangent bundle π ∗ : T ∗ M → M on V .
TENSORS 75
4.17 Proposition Let π ∗ : T ∗ M → M be the cotangent bundle of a smooth mani-

fold M n and ω : M → T ∗ M be a map with π ∗ ◦ ω = idM . Then ω is a 1-form on M
if and only if the map a : U → Rn defined by
n
ω (p) = ∑ ai (p) dxi (p)
i=1
for p ∈ U is smooth for every local chart (x,U) on M .

PROOF : Follows from Propositions 4.12 and 4.15.
4.18 Proposition If f : V → R is a smooth function defined on an open subset V

of a smooth manifold M n , then its differential d f : V → T ∗ M is a 1-form on V . For
any local chart (x,U) on M with U ⊂ V we have that
n
∂f i
d f |U = ∑ i dx .
i=1
∂ x
PROOF : We have that

!
n n
∂f
d f (p) = ∑ d f (p) ∂ j dxi (p) = ∑ i (p) dxi (p)
i=1
∂x ∂x i=1
p
for p ∈ U.
4.19 Definition Let S be a subset of a finite dimensional vector space V . Then

the annihilator of S is the subspace A(S) of V ∗ defined by
A(S) = { λ ∈ V ∗ |λ (S) = {0}} .
4.20 Proposition Let V1 and V2 be two subspaces of a finite dimensional vector

space V with V = V1 ⊕ V2 . Then we have that
V ∗ = A(V1 ) ⊕ A(V2) .
If π1 : V → V and π2 : V → V are the projections on V1 and V2 , then π2∗ : V ∗ → V ∗

and π1∗ : V ∗ → V ∗ are the projections on A(V1 ) and A(V2 ), respectively.
λ = λ ◦ π2 + λ ◦ π1
where λ ◦ π2 ∈ A(V1 ) and λ ◦ π1 ∈ A(V2 ) for every λ ∈ V ∗ . Since we also have that
A(V1 ) ∩ A(V2) = {0} ,
this completes the proof of the proposition.

4.21 Proposition Let W be a subspace of a finite dimensional vector space V .

Then we have that
dim W + dim A(W ) = dim V .
PROOF : Choose a basis C = {w1 , ... , wr } for W , and extend it to a basis
B = {w1 , ... , wn } for V with dual basis B ∗ = {w∗1 , ... , w∗n } . We contend that
{w∗r+1 , ... , w∗n } is a basis for A(W ) .
It is clearly a linearly independent subset of A(W ) , so we only need to show
that it spans A(W ) . This follows from the fact that every element λ ∈ A(W ) can be
written as λ = ∑ni=1 ai w∗i , where a j = λ (w j ) = 0 for j = 1, ... , r , thus showing that
dim A(W ) = n − r .
4.22 Proposition Let π ′ : E ′ → M be a k-dimensional subbundle of an n-

dimensional vector bundle π : E → M. Then
[
A(E ′ ) = A(E p′ )
p∈M
is an (n − k)-dimensional subbundle of the dual bundle π ∗ : E ∗ → M, with projection

π ′′ : A(E ′ ) → M sending each set A(E p′ ) to p.
PROOF : Given a point p in M, we choose a local trivialization (t, π −1 (U)) around p

in E having the subbundle property
t (π −1 (U) ∩ E ′ ) = U × Rk × {0} ⊂ U × Rk × Rn−k .
Let (t ′ , π ∗ −1 (U)) be the corresponding local trivialization in the dual bundle π ∗ :

E ∗ → M described in Remark 4.9, where t p′ = r ◦ (t p−1 )∗ for p ∈ U. In the same way
as in the proof of Proposition 4.21 we see that
t p′ (A(E p′ )) = r(A( Rk × {0})) = {0} × Rn−k
for p ∈ U. If α : Rk ×Rn−k → Rn−k ×Rk is the linear isomorphism given by α (a, b) =

(b, a) for a ∈ Rk and b ∈ Rn−k , we therefore have that (α ◦ t ′ , π ∗ −1 (U)) is a local
trivialization around p in E ∗ with the subbundle property
α ◦ t ′ (π ∗ −1 (U) ∩ A(E ′ )) = U × Rn−k × {0} ⊂ U × Rn−k × Rk .
TENSOR BUNDLES
4.23 Let V1 , ...,Vk be vector spaces. Their tensor product V1 ⊗ · · ·⊗Vk is defined
to be a vector space V with a multilinear map φ : V1 × · · · × Vk → V satisfying the
TENSORS 77
following universal property. If F : V1 × · · · × Vk → W is any multilinear map into

some vector space W , then there is a unique linear map F∗ : V → W which makes the
diagram
φ V
✟✯
✟
✟
V1 × · · · × Vk F∗
❍ ❄
❍❥
❍
F W
commutative. It follows from the definition that the tensor product is uniquely defined
in the sense that if V ′ is another vector space satisfying the above condition, then
there is a unique linear isomorphism between V and V ′ .
If vi ∈ Vi for i = 1, ..., k, we denote the element φ (v1 , ..., vk ) by v1 ⊗ · · · ⊗ vk .
Given a linear map Fi : Vi → Vi′ for each i = 1, ..., k, we have a unique linear map
F1 ⊗ · · · ⊗ Fk : V1 ⊗ · · · ⊗ Vk → V1′ ⊗ · · · ⊗ Vk′ such that the diagram
φ ✲
V1 × · · · × Vk V1 ⊗ · · · ⊗ Vk
F1 × · · · × Fk F1 ⊗ · · · ⊗ Fk
❄ ❄
V1′ × · · · × Vk′ ✲ V′ ⊗ ···⊗V′
1 k
φ′
is commutative since φ ′ ◦ (F1 × · · · × Fk ) is multilinear, and we have that
F1 ⊗ · · · ⊗ Fk (v1 ⊗ · · · ⊗ vk ) = F1 (v1 ) ⊗ · · · ⊗ Fk (vk ) .
It remains to prove the existence of V . We first consider the tensor product of the
dual spaces V1∗ , ...,Vk∗ .
4.24 If V1 , ...,Vk ,W are vector spaces, we let Lk (V1 , ...,Vk ; W ) be the space of
all multilinear maps F : V1 × .. × Vk → W . The space Lk (V1 , ...,Vk ; R) consisting of
all multilinear functionals T : V1 × .. × Vk → R is also denoted by T (V1 , ...,Vk ). If
k = 1, then T (V ) = V ∗ is just the dual space of V . We want to prove in general
that T (V1 , ...,Vk ) = V1∗ ⊗ · · · ⊗ Vk∗ so that V1 ⊗ · · · ⊗ Vk = T (V1∗ , ...,Vk∗ ) when we
identify each Vi with its double dual Vi∗∗ as usual.
The components F j ∈ T (V1 , ...,Vk ) of a multilinear map F ∈ Lk (V1 , ...,Vk ; W )
with respect to a basis C = {w1 , ..., wm } for W are defined by
m
F(v1 , ..., vk ) = ∑ F j (v1 , ..., vk ) w j
j=1
for (v1 , ..., vk ) ∈ V1 × .. × Vk .

4.25 If T ∈ T (V1 , ...,Vk ) and S ∈ T (Vk+1 , ...,Vk+l ), we define their tensor prod-
uct T ⊗ S ∈ T (V1 , ...,Vk+l ) by
(T ⊗ S)(v1 , ..., vk , vk+1 , ..., vk+l ) = T (v1 , ..., vk )S(vk+1 , ..., vk+l )
We see that ⊗ is bilinear, i.e., we have that

(1) (T1 + T2 ) ⊗ S = T1 ⊗ S + T2 ⊗ S and (k T ) ⊗ S = k (T ⊗ S)
(2) T ⊗ (S1 + S2) = T ⊗ S1 + T ⊗ S2 and T ⊗ (k S) = k (T ⊗ S)
for all multilinear functionals T, T1 , T2 ∈T (V1 , ...,Vk ) and S, S1 , S2 ∈T (Vk+1 , ...,Vk+l )
and every real number k.
Since (U ⊗ T ) ⊗ S = U ⊗ (T ⊗ S), we also have a well-defined tensor product
with an arbitrary number of factors. In particular we have a multilinear map
φ : V1∗ × · · · × Vk∗ → T (V1 , ...,Vk )
defined by
φ (λ1 , ..., λk ) = λ1 ⊗ · · · ⊗ λk
where
(λ1 ⊗ · · · ⊗ λk )(v1 , ..., vk ) = λ1 (v1 ) · · · λk (vk ) .
4.26 Proposition If Br = {vr1 , ..., vrnr } is a basis for the vector space Vr for r =
1 , ..., k, then
D = {v1i1∗ ⊗ · · · ⊗ vkik∗ | (i1 , ..., ik ) ∈ In1 × · · · × Ink } ,
where In = {1 , ..., n}, is a basis for T (V1 , ...,Vk ). If T ∈ T (V1 , ...,Vk ) we have that
T= ∑ T (v1i1 , ..., vkik ) v1i1∗ ⊗ · · · ⊗ vkik∗ .

i1 ,...,ik
PROOF : If T ∈ T (V1 , ...,Vk ) we have that
∑ T (v1i1 , ..., vkik ) v1i1∗ ⊗ · · · ⊗ vkik∗ (v1j1 , ..., vkjk ) = T (v1j1 , ..., vkjk )
i1 ,...,ik
for 1 ≤ jr ≤ nr and 1 ≤ r ≤ k, which shows the last assertion of the proposition and
also that D spans T (V1 , ...,Vk ). To show that it is linearly independent, assume that
∑ ai1 ,...,ik v1i1∗ ⊗ · · · ⊗ vkik∗ = 0 .

i1 ,...,ik
By applying both sides to (v1j1 , ..., vkjk ), we see that a j1 ,..., jk = 0 for 1 ≤ jr ≤ nr and
1 ≤ r ≤ k.
TENSORS 79
4.27 Remark We can order the index set In1 × · · · × Ink for the basis D lexico-
graphically so that (i1 , ..., ik ) < ( j1 , ..., jk ) if there is an integer r ∈Ik such that is = js
for 1 ≤ s < r and ir < jr . We then have a unique bijection b : In1 × · · · × Ink → In
which is strictly increasing and where n = ∏kr=1 nk . It is explicitly given by
k k
b (i1 , ..., ik ) = ∑ (ir − 1) ∏ nj + 1 .
r=1 j=r+1
We let T (B1 , ..., Bk ) be the basis {v1 , ..., vn } for T (V1 , ...,Vk ) obtained by reindex-
ing D so that
v b (i1 ,...,ik ) = v1i1∗ ⊗ · · · ⊗ vkik∗
for (i1 , ..., ik ) ∈ In1 × · · · × Ink .
4.28 Proposition If V1 , ...,Vk are vector spaces, the tensor product V1∗ ⊗ · · · ⊗
Vk∗ equals T (V1 , ...,Vk ) with the multilinear map φ : V1∗ × · · · × Vk∗ → T (V1 , ...,Vk )
defined in 4.25.
PROOF : Let Br = {vr1 , ..., vrnr } be a basis for the vector space Vr for r = 1 , ..., k, and
let F : V1 × · · · × Vk∗ → W be a multilinear map into some vector space W . Then we
∗
have a linear map F∗ : T (V1 , ...,Vk ) → W defined by
F∗ (v1i1∗ ⊗ · · · ⊗ vkik∗ ) = F(v1i1∗ , ..., vkik∗ ) (1)
for (i1 , ..., ik ) ∈ In1 × · · · × Ink . If λr = ∑ni=1

r
ari vri ∗ is an arbitrary element in Vr∗ for
r = 1 , ..., k, we have that
F∗ (λ1 ⊗ · · · ⊗ λk ) = ∑ a1i1 · · · akik F∗ (v1i1∗ ⊗ · · · ⊗ vkik∗ )

i1 ,...,ik
= ∑ a1i1 · · · akik F(v1i1∗ , ..., vkik∗ ) = F(λ1 , ..., λk )

i1 ,...,ik
which shows that the diagram in 4.23 is commutative.

To prove uniqueness, we see that a linear map F∗ making the diagram commuta-
tive clearly must satisfy (1), and this determines F∗ uniquely on T (V1 , ...,Vk ).
4.29 Corollary Let V1 , ...,Vk be vector spaces, and let φ ′ : V1∗∗ × · · · × Vk∗∗ →
T (V1∗ , ...,Vk∗ ) be the multilinear map defined in 4.25. Then the tensor product
V1 ⊗ · · · ⊗ Vk equals T (V1∗ , ...,Vk∗ ) with the multilinear map φ : V1 × · · · × Vk →
T (V1∗ , ...,Vk∗ ) given by φ = φ ′ ◦ (iV1 × · · · × iVk ).
We have that iV1 ⊗ · · · ⊗ iVk is the identity on T (V1∗ , ...,Vk∗ ), and if vi ∈ Vi for
i = 1, ..., k, then v1 ⊗ · · · ⊗ vk is the element in T (V1∗ , ...,Vk∗ ) given by
v1 ⊗ · · · ⊗ vk (λ1 , ..., λk ) = λ1 (v1 ) · · · λk (vk )
for λi ∈ Vi∗ where i = 1, ..., k.

PROOF : Let F : V1 × · · · × Vk → W be a multilinear map into some vector space

W . Then we have a multilinear map F ◦ (iV1 × · · · × iVk ) −1 : V1∗∗ × · · · × Vk∗∗ → W ,
so by Proposition 4.28 there is a unique linear map F∗ : T (V1∗ , ...,Vk∗ ) → W with
F ◦ (iV1 × · · · × iVk ) −1 = F∗ ◦ φ ′ . This implies that F = F∗ ◦ φ and completes the proof
of the first part of the corollary. The last part follows from 4.23 and Remark 4.5.
4.30 Corollary If Br = {vr1 , ..., vrnr } is a basis for the vector space Vr for r =
1 , ..., k, then
D = {v1i1 ⊗ · · · ⊗ vkik | (i1 , ..., ik ) ∈ In1 × · · · × Ink } ,
where In = {1 , ..., n}, is a basis for V1 ⊗ · · · ⊗ Vk . If T ∈ T (V1∗ , ...,Vk∗ ) we have that
T= ∑ T (v1i1∗ , ..., vkik∗ ) v1i1 ⊗ · · · ⊗ vkik .

i1 ,...,ik
The tensor product V1 ⊗ · · ·⊗Vk is generated by the set φ (V1 × · · ·×Vk ) consisting
of elements of the form v1 ⊗ · · · ⊗ vk , where vi ∈ Vi for i = 1, ..., k, which are called
simple . If Fi : Vi → Vi′ is a linear map for i = 1, ..., k, we have that
F1 ⊗ · · · ⊗ Fk (V1 ⊗ · · · ⊗ Vk ) = F1 (V1 ) ⊗ · · · ⊗ Fk (Vk ) .
PROOF : Follows from 4.23, Proposition 4.26 and Corollary 4.29.
4.31 Remark By 4.23 and 4.25 we have a bilinear map
µ : T (V1 , ...,Vk ) × T (Vk+1 , ...,Vk+l ) → T (V1 , ...,Vk+l )
defined by
µ (T, S) = T ⊗ S
which induces a linear map
µ∗ : T (V1 , ...,Vk ) ⊗ T (Vk+1 , ...,Vk+l ) → T (V1 , ...,Vk+l ) .
If the vector spaces V1 , ...,Vk+l are finite dimensional, it follow from Proposition 4.26
and 4.28 that µ∗ is a linear isomorphism since it is a surjective linear map between
spaces of the same dimension.
4.32 If Fi : Vi → Wi is a linear map for i = 1 , ..., k, we have a linear map
(F1 , ..., Fk )∗ : T (W1 , ...,Wk ) → T (V1 , ...,Vk )
defined by
(F1 , ..., Fk )∗ (T ) = T ◦ (F1 × · · · × Fk ) ,
i.e.,
(F1 , ..., Fk )∗ (T ) (v1 , ..., vk ) = T (F1 (v1 ), ..., Fk (vk )) .
TENSORS 81
In particular we have that
(F1 , ..., Fk )∗ (λ1 ⊗ · · · ⊗ λk ) = F1∗ (λ1 ) ⊗ · · · ⊗ Fk∗ (λk ) .
for λi ∈ Vi∗ where i = 1, ..., k.
4.33 Proposition
(1) If Fi : Ui → Vi and Gi : Vi → Wi are linear maps for i = 1 , ..., k, we have that
(G1 ◦ F1 , ..., Gk ◦ Fk )∗ = (F1 , ..., Fk )∗ ◦ (G1 , ..., Gk )∗ .
(2) If id i : Vi → Vi is the identity on Vi for i = 1 , ..., k, then (id 1 , ..., id k )∗ :
T (V1 , ...,Vk ) → T (V1 , ...,Vk ) is the identity on T (V1 , ...,Vk ).
(3) If Fi : Vi → Wi are isomorphisms for i = 1 , ..., k, then so is (F1 , ..., Fk )∗ :

T (W1 , ...,Wk ) → T (V1 , ...,Vk ) with (F1 , ..., Fk )∗ −1 = (F1−1 , ..., Fk−1 )∗ .
(4) If Fi : Vi → Wi are linear maps for i = 1 , ..., k + l, we have that (F1 , ..., Fk+l )∗
(T ⊗ S) = (F1 , ..., Fk )∗ (T ) ⊗ (Fk+1 , ..., Fk+l )∗ (S) for every T ∈ T (W1 , ...,Wk )
and S ∈ T (Wk+1 , ...,Wk+l ).
PROOF : Follows immediately from the definition.
4.34 Proposition If Fi : Vi → Wi is a linear map for i = 1 , ..., k, then the linear

map
F1 ⊗ · · · ⊗ Fk : V1 ⊗ · · · ⊗ Vk → W1 ⊗ · · · ⊗ Wk
defined in 4.23 is given by
F1 ⊗ · · · ⊗ Fk = (F1∗ , ..., Fk∗ )∗ .

(F1∗ , ..., Fk∗ )∗ (v1 ⊗ · · · ⊗ vk )(λ1 , ..., λk ) = (v1 ⊗ · · · ⊗ vk )(F1∗ (λ1 ), ..., Fk∗ (λk ))
= λ1 (F1 (v1 )) · · · λk (Fk (vk )) = (F1 (v1 ) ⊗ · · · ⊗ Fk (vk ))(λ1 , ..., λk )
for vi ∈ Vi and λi ∈ Wi∗ where i = 1, ..., k.
4.35 Corollary
(1) If Fi : Ui → Vi and Gi : Vi → Wi are linear maps for i = 1 , ..., k, we have that
(G1 ◦ F1 ) ⊗ · · · ⊗ (Gk ◦ Fk ) = (G1 ⊗ · · · ⊗ Gk ) ◦ (F1 ⊗ · · · ⊗ Fk ) .
(2) If id i : Vi → Vi is the identity on Vi for i = 1 , ..., k, then id 1 ⊗ · · · ⊗ id k : V1 ⊗
· · · ⊗ Vk → V1 ⊗ · · · ⊗ Vk is the identity on V1 ⊗ · · · ⊗ Vk .
(3) If Fi : Vi → Wi are isomorphisms for i = 1 , ..., k, then so is F1 ⊗ · · · ⊗ Fk : V1 ⊗
· · · ⊗ Vk → W1 ⊗ · · · ⊗ Wk with (F1 ⊗ · · · ⊗ Fk ) −1 = F1−1 ⊗ · · · ⊗ Fk−1 .
PROOF : Follows from Propositions 4.2 and 4.33 or immediately from the definition
in 4.23.
4.36 Definition Let Ar be an mr × nr - matrix for r = 1 , ..., k. The tensor prod-

uct A1 ⊗ · · · ⊗ Ak is the m × n - matrix A defined by
Ai j = A1i1 j1 · · · Akik jk
where m = ∏kr=1 mr and n = ∏kr=1 nr , and where i = b (i1 , ..., ik ) and j =

b ( j1 , ..., jk ) for the bijections b : In1 × · · · × Ink → In and b′ : Im1 × · · · × Imk → Im
′
defined in Remark 4.27.
4.37 Proposition Let Fr : Vr → Wr be a linear map for r = 1 , ..., k, and let

Br and Cr be bases for Vr and Wr , respectively. Let B = T (B1 , ..., Bk ) and
C = T (C1 , ..., Ck ) be the bases for T (V1 , ...,Vk ) and T (W1 , ...,Wk ) defined in Re-
mark 4.27. Then we have that
B
mCB ((F1 , ..., Fk )∗ ) = mB 1 t k t
C1 (F1 ) ⊗ · · · ⊗ mCk (Fk ) .
PROOF : Let Br = {vr1 , ..., vrnr } and Cr = {wr1 , ..., wrmr }. If mB r r

Cr (F) = A , we have that
mr
Fr (vrj ) = ∑ Arij wri
i=1
for 1 ≤ j ≤ nr and 1 ≤ r ≤ k. From this it follows that
(F1 , ..., Fk )∗ (w1j1∗ ⊗ · · · ⊗ wkjk∗ ) (v1i1 , ..., vkik ) =
((w1j1∗ ◦ F1 ) ⊗ · · · ⊗ (wkjk∗ ◦ Fk )) (v1i1 , ..., vkik ) = A1j1 i1 · · · Akjk ik
which shows that
(F1 , ..., Fk )∗ (w1j1∗ ⊗ · · · ⊗ wkjk∗ ) = ∑ A1j1 i1 · · · Akjk ik v1i1∗ ⊗ · · · ⊗ vkik∗

i1 ,...,ik
for ( j1 , ..., jk ) ∈ Im1 × · · · × Imk . If B = {v1 , ..., vn } and C = {w1 , ..., wm }, where
n = ∏kr=1 nr and m = ∏kr=1 mr , and if (A1 )t ⊗ · · · ⊗ (Ak )t = A, it follows from
Remark 4.27 and Definition 4.36 that
n
(F1 , ..., Fk )∗ (w j ) = ∑ Ai j vi
i=1
4.38 Now let π r : E r → M be an nr -dimensional vector bundle over the smooth

TENSORS 83
manifold M for r = 1 , ..., k, and let E pr = π r −1 (p) be the fibre over p for each p ∈ M.
We want to define a new bundle
[
T (E 1 , ..., E k ) = T (E p1 , ..., E pk )
p∈M
with projection π ′ : T (E 1 , ..., E k ) → M defined by π ′ (ω ) = p for ω ∈T (E p1 , ..., E pk ).
4.39 Proposition π ′ : T (E 1 , ... , E k ) → M is an n-dimensional vector bundle over

M, where n = ∏kr=1 nr .
PROOF : Let Er = {er1 , ... , ernr } be the standard basis for Rnr for r = 1 , ... , k, and
let B = T (E1 , ... , Ek ) be the basis for T (Rn1 , ... , Rnk ) defined in Remark 4.27. If
B = {v1 , ... , vn } and if E = {e1 , ... , en } is the standard basis for Rn , we have a linear
isomorphism r′ : T (Rn1 , ..., Rnk ) → Rn defined by r′ (vi ) = ei for i = 1, ... , n.
Let {Uα |α ∈ A} be an open cover of M such that for each α ∈ A, there are local
trivializations (tαr , π r −1 (Uα )) in E r for r = 1 , ... , k. Then we have for each α ∈ A a
bijection tα′ : π ′ −1 (Uα ) → Uα × Rn defined by tα′ (ω ) = (p,tα′ ,p (ω )) for ω ∈ π ′ −1 (p),
∗
where tα′ ,p = r′ ◦ (tα1 ,p−1 , ... ,tαk ,p−1 ) . We will show that these bijections satisfy the
conditions of Proposition 2.48.
For each pair of indices α , β ∈ A, we have that tαr ◦ tβr −1 is a diffeomorphism
from (Uα ∩Uβ ) × Rnr onto itself given by
−1
(tαr ◦ tβr ) (p, v) = (p , (tαr ,p ◦ tβr ,p−1 ) (v))
for r = 1 , ... , k, where tαr ,p ◦ tβr ,p−1 : Rnr → Rnr is a linear isomorphism for each p ∈
Uα ∩Uβ . Since
−1 ∗
tβ′ ,p ◦ t ′ α ,p = r′ ◦ (tα1 ,p ◦ tβ1 ,p−1 , ... ,tαk ,p ◦ tβk ,p−1 ) ◦ r′ −1 ,
it follows from Proposition 4.37, using the fact that mB ′ E ′ −1 ) are both
E (r ) and mB (r
the identity matrix, that
E1 1 E
mEE (tβ′ ,p ◦ tα′ −1 1 −1 t k k k −1 t
,p ) = mE1 (tα ,p ◦ tβ ,p ) ⊗ · · · ⊗ mEk (tα ,p ◦ tβ ,p ) .
Hence tβ′ ◦ tα′ −1 is also a diffeomorphism from (Uα ∩ Uβ ) × Rn onto itself which is
a linear isomorphism on each fibre. By Proposition 2.48 there is therefore a unique
topology and smooth structure on T (E 1 , ... , E k ) such that π ′ : T (E 1 , ... , E k ) → M
is an n-dimensional vector bundle over M with (tα′ , π ′ −1 (Uα )) as local trivializations.
∗
4.40 Remark We say that (t ′ , π ′ −1 (U)), where t p′ = r′ ◦ (t p1 −1 , ...,t pk −1 ) for
p ∈ U, is the local trivialization in the tensor bundle π ′ : T (E 1 , ..., E k ) → M cor-
responding to the local trivializations (t r , π r −1 (U)) in the nr -dimensional vector
bundles π r : E r → M for r = 1 , ..., k. Here r′ : T (Rn1 , ..., Rnk ) → Rn is the linear

isomorphism defined by
i1 ik
r′ ( e(1) ⊗ · · · ⊗ e(k) ) = ei ,
nr
for (i1 , ..., ik ) ∈ In1 × · · · × Ink , where Er∗ = {e(r)
1
, ..., e(r) } is the dual of the standard
basis in R for r = 1 , ..., k and E = {e1 , ... , en } is the standard basis for Rn , and
n r
where i = b (i1 , ..., ik ) for the bijection b : In1 × · · · × Ink → In defined in Remark 4.27
with n = ∏kr=1 nr .
4.41 Definition Let π r : E r → M be an nr -dimensional vector bundle over the

smooth manifold M for r = 1 , ..., k. If {Uα |α ∈ A} is an open cover of M such that
for each α ∈ A, there are local trivializations (tαr , π r −1 (Uα )) in E r for r = 1 , ..., k,
then the family {(tαr , π r −1 (Uα ))|(r, α ) ∈ Ik × A} is called a trivializing cover of the
bundles E 1 , ..., E k over M.
4.42 Definition If π r : E r → M is an nr -dimensional vector bundle over the

smooth manifold M for r = 1 , ..., k, we define their tensor product by
E 1 ⊗ · · · ⊗ E k = T (E 1 ∗ , ..., E k ∗ ) ,
with projection π ′ : E 1 ⊗ · · · ⊗ E k → M sending each set E p1 ⊗ · · · ⊗ E pk to p . By

Proposition 4.8 and 4.39 this is an n-dimensional vector bundle over M , where
n = ∏kr=1 nr .
4.43 Proposition Let π1r : E1r → M and π2r : E2r → M be two nr -dimensional
vector bundles for each r = 1 , ... , k , and let f r be bundle maps over M from π1r to π2r .
Then we have a bundle map ( f 1 , ... , f k ) ∗ over M from π2′ : T (E21 , ... , E2k ) → M to
π1′ : T (E11 , ... , E1k ) → M given by
( f 1 , ... , f k ) ∗p = ( f 1, p , ... , f k, p ) ∗
i.e.,
( f 1 , ... , f k ) ∗ (T ) (v1 , ... , vk ) = T ( f 1 (v1 ), ... , f k (vk ))
for all T ∈ π2′ −1 (p) and vr ∈ π1r −1 (p) where p ∈ M and r = 1, ... , k .
PROOF : The proof is similar to the proof of Proposition 4.39, and we use the same
notation. Choose a trivializing cover {(ti,r α , πir −1 (Uα ))|(r, α )∈Ik × A} of the bundles
Ei1 , ... , Eik over M for i = 1, 2 , and let (ti,′ α , πi ′ −1 (Uα )) be the corresponding local
trivializations in the tensor bundles πi ′ : T (Ei1 , ... , Eik ) → M.
For each index α ∈ A, we have that t2, r ◦ f ◦ t r −1 is a smooth map from U ×
α r 1,α α
R r to Uα × R r given by
n n
r r −1 r r −1
(t2,α ◦ f r ◦ t1,α ) (p, v) = (p, (t2,α ,p ◦ f r ◦ t1,α ,p ) (v))
TENSORS 85
r −1
α ,p ◦ f r ◦ t1,α ,p : R → R is a linear map for each p ∈ Uα .
r
for r = 1 , ... , k , where t2, nr nr
Since
′ ∗ ′ −1
t1,α ,p ◦ ( f 1 , ... , f k ) ◦ t 2,α ,p
1 1 −1 k k −1 −1 ∗
= r ◦ (t2,α ,p ◦ f 1 ◦ t1,α ,p , ... ,t2,α ,p ◦ f k ◦ t1,α ,p ) ◦ r ,
it follows from Proposition 4.37, using the fact that mB E −1

E (r) and mB (r ) are both the
′ ∗ ′ −1
mEE (t1,α ,p ◦ ( f 1 , ... , f k ) ◦ t 2,α ,p )
E
= mEE11 (t2,
1 1 −1 t k k k −1 t
α ,p ◦ f 1 ◦ t1,α ,p ) ⊗ · · · ⊗ mEk (t2,α ,p ◦ f k ◦ t1,α ,p ) .
Hence t1,′ ◦ ( f , ... , f ) ∗ ◦ t ′ −1 is a smooth map from U × Rn to U × Rn which

α 1 k 2,α α α
is linear on each fibre, thus showing that ( f 1 , ... , f k ) ∗ is a bundle map over M .
4.44 Corollary Let π1r : E1r → M and π2r : E2r → M be two nr -dimensional vector
bundles for each r = 1 , ... , k , and let f r be bundle maps over M from π1r to π2r .
Then we have a bundle map f 1 ⊗ · · · ⊗ f k over M from π1′ : E11 ⊗ · · · ⊗ E1k → M to
π2′ : E21 ⊗ · · · ⊗ E2k → M given by
( f 1 ⊗ · · · ⊗ f k ) p = f 1, p ⊗ · · · ⊗ f k, p
i.e.,
f 1 ⊗ · · · ⊗ f k ( v1 ⊗ · · · ⊗ vk ) = f 1 (v1 ) ⊗ · · · ⊗ f k (vk )
r −1
for all vr ∈ π1 (p) where p ∈ M and r = 1, ... , k .
4.45 Proposition Let s : M → T (E 1 , ..., E k ) be a map with π ′ ◦ s = idM , and let

1 , ..., e nr } be the dual of the standard basis in Rnr for r = 1 , ..., k . Then s is
Er∗ = {e(r) (r)
a section of π ′ on M if and only if the functions Ai1 ,...,ik : U → R defined by
i1 i
s(p) = ∑ Ai1 ,...,ik (p) (e(1) ◦ t p1 ) ⊗ · · · ⊗ (e(k)
k
◦ t pk )
i1 ,...,ik
for p ∈ U are smooth for every local trivialization (t r , π r −1 (U)) in E r belonging to a

trivializing cover of the bundles E 1 , ..., E k over M.
PROOF : If (t ′ , π ′ −1 (U)) is the corresponding local trivialization for the tensor bun-
dle T (E 1 , ..., E k ), it follows that
i1 i
t p′ ((e(1) ◦ t p1 ) ⊗ · · · ⊗ (e(k)
k
◦ t pk )) = ei ,
where i = b (i1 , ..., ik ) for the bijection b : In1 × · · · × Ink → In defined in Remark 4.27
with n = ∏kr=1 nr . If a : U → Rn is the map defined by ai = Ai1 ,...,ik , we have that
(t ′ ◦ s)(p) = (p, a(p))
for p ∈ U, which shows that a is the local representation of s on U as defined in
Definition 2.52.
4.46 For each section T of π ′ on M we have a map
TM : Γ(M; E 1 ) × · · · × Γ(M; E k ) → F (M)

defined by
TM (s1 , ... , sk ) (p) = T (p) (s1 (p) , ... , sk (p))
for each point p on M and for sections sr ∈ Γ(M; E r ) where r = 1, ... , k. Indeed, if
(x,U) is a local chart on M, and if Ai1 ,...,ik and ari are the components of the local
representations of T and sr on U for r = 1, ... , k as defined in Propositions 4.45 and
2.54, then
TM (s1 , ... , sk )|U = ∑ Ai1 ,...,ik a1i1 · · · akik
i1 ,...,ik
which shows that TM (s1 , ... , sk ) is a smooth function on M.

We see that TM is multilinear over F (M). The next proposition shows that each
such multilinear map arises from a unique section of π ′ on M in this way.
4.47 Proposition Let M be a smooth manifold, and let
F : Γ(M; E 1 ) × · · · × Γ(M; E k ) → F (M)
be a map which is multilinear over F (M). Then there is a unique section T of π ′ on

M such that F = TM .
PROOF : If vi ∈ E pi , we choose a section si on M with si (p) = vi for i = 1, ... , k using
Proposition 2.56 (2). Then we must define
T (p)(v1 , ... , vk ) = F (s1 , ... , sk )(p). (1)
For T (p) to be well defined, we need to show that T (p)(v1 , ... , vk ) does not
depend on the choice of s1 , ... , sk . It clearly suffices to show that
F (s1 , ... , sk )(p) = 0 (2)
if sr (p) = 0 for some r. Then it will follow that sr in (1) may be replaced by any
other section of π r on M with the same value at p without changing the value of
F (s1 , ... , sk ) (p).
Suppose that sr (p) = 0, and choose a local chart (x,U) around p and a smooth
function f : M → R with f (p) = 1 and supp ( f ) ⊂ U. If
nr
sr | U = ∑ bi ηir ,
i=1
where ηir :U → Er is the section given by ηir (p) = t pr −1 (eri ) for p ∈ U and i =
1, ... , nr , we extend the functions f bi and sections f ηir on U to smooth functions gi

and sections θi on M, respectively, by defining them to be zero outside U. Then we
have that
nr
f 2 sr = ∑ gi θi
i=1
TENSORS 87
so that
nr
f 2 F (s1 , ... , sr , ... , sk ) = F (s1 , ... , f 2 sr , ... , sk ) = ∑ gi F (s1 , ... , θi , ... , sk ) .
i=1
Evaluating at p and using that f (p) = 1 and gi (p) = 0 for i = 1, ... , nr , we obtain (2),
and this compeletes the proof that T (p) is well defined by formula (1).
It only remains to show that T is smooth on M. By Proposition 4.45 we must
show that the functions Ai1 ,...,ik : U → R defined by
Ai1 ,...,ik (p) = T (p)(ηi11 (p), ... , ηikk (p))
for p ∈ U, are smooth for every local chart (x,U) on M. Fix p in U, and let V be an
open neighbourhood of p with V ⊂ U. By Proposition 2.56 (1) there are sections θir
of π r on M which coincide with ηir on V for i = 1, ... , nr and r = 1, ... , k. Then we
have that
Ai1 ,...,ik (q) = F (θi11 , ... , θikk )(q)
for q ∈ V , showing that Ai1 ,...,ik is smooth at p. Since p was an arbitrary point in U,
this completes the proof that T is smooth.
4.48 Remark Because of Proposition 4.47 we will not distinguish between the
section T and the map TM , and a section in Γ(M; T (E 1 , ... , E k )) can be thought of as
an operation on k sections in Γ(M; E 1 ) , ... , Γ(M; E k ), respectively, yielding a smooth
funnction on M.
4.49 Proposition Let π1r : E1r → M1 and π2r : E2r → M2 be two vector bundles of
dimensions nr1 and nr2 for each r = 1, ... , k, and let ( f˜r , f ) be bundle maps from π1r
to π2r . If s2 is a section of π2′ : T (E21 , ... , E2k ) → M2 on an open subset V2 of M2 and
V1 = f −1 (V2 ), then the map s1 : V1 → T (E11 , ... , E1k ) defined by
s1 (p)(v1 , ..., vk ) = s2 ( f (p))( f˜1 (v1 ), ... , f˜k (vk ))
for p ∈ V1 and vr ∈ π1r −1 (p) where r = 1, ... , k , is a section of π1′ : T (E11 , ... , E1k ) →
M1 on V1 .
s1 is called the pull-back of s2 by ( f˜1 , ... , f˜k ) and is denoted by ( f˜1 , ... , f˜k ) ∗ (s2 ).
We will give two alternative proofs.
FIRST PROOF : Let p be a point in V1 , and let (t r1 , π1r −1 (U1 )) and (t r2 , π2r −1 (U2 )) be
local trivializations of π1r and π2r around p and f (p), respectively, with U2 ⊂ V2 and
r r
U1 ⊂ f −1 (U2 ). Then the maps t r2 ◦ f˜r ◦ t r1 −1 : U1 × Rn1 → U2 × Rn2 are smooth and
linear on each fibre.
Now let (t ′i , πi′ −1 (Ui )) be the local trivialization of πi′ defined in Remark 4.40
for i = 1, 2. We have that t ′i : πi′ −1 (Ui ) → Ui × Rni is given by t ′i (ω ) = (q,t ′i,q (ω ))
∗
for q ∈ Ui and ω ∈ πi′ −1 (q), where t ′i,q = ri ◦ (t 1i,q−1 , ... , t ki,q−1 ) and ni = ∏kr=1 nri .
Hence t ′i ◦ si (q) = (q, hi (q)) where the local representations hi : Ui → Rni of si are
given by
h1 (q) = r1 ◦ ( f˜1 ◦ t 11,q−1 , ... , f˜k ◦ t k1,q−1 ) ∗ ◦ s2 ( f (q))
and
h2 (q) = r2 ◦ (t 12,q−1 , ... , t 12,q−1 ) ∗ ◦ s2 (q) .
Combining this we have that
h1 (q) = r1 ◦ (t 12, f (q) ◦ f˜1 ◦ t 11,q−1 , ... , t k2, f (q) ◦ f˜k ◦ t k1,q−1 ) ∗ ◦ r2−1 ◦ h2 ( f (q))
r
for q ∈ U1 . If Eir and Bi are the standard basis for Rni and Rni , respectively, for
r = 1, ... , k and i = 1, 2, we have that
mB 2 1 ˜ 1 −1 k ˜ k −1 ∗ −1
B1 (r1 ◦ (t 2, f (q) ◦ f 1 ◦ t 1,q , ... , t 2, f (q) ◦ f k ◦ t 1,q ) ◦ r2 )
E 1 E k
= mE11 (t 12, f (q) ◦ f˜1 ◦ t 11,q−1 ) t ⊗ · · · ⊗ mE1k (t k2, f (q) ◦ f˜k ◦ t 1,q
k −1 t
)
2 2
which shows that s1 is smooth in a neighbourhood of every point p in V1 and hence

is a section of π1′ on V1 .
SECOND PROOF : The Proposition can also be proved using Proposition 4.45
(i,r) (i,r) r
and the standard basis Eir = {e 1 , ... , e nr } in Rni with dual basis (Eir )∗ =
i
nr
1 , ... , e i } for r = 1, ..., k and i = 1, 2.
{e (i,r) (i,r)
Again, let p be a point in V1 , and let (t r1 , π1r −1 (U1 )) and (t r2 , π2r −1 (U2 )) be local
trivializations of π1r and π2r around p and f (p), respectively, with U2 ⊂ V2 and U1 ⊂
r r
f −1 (U2 ). Since the maps t r2 ◦ f˜r ◦ t r1 −1 : U1 × Rn1 → U2 × Rn2 are smooth and linear
on each fibre, there are smooth functions a rji : U1 → R such that
nr2
(1,r) (2,r)
t r2, f (q) ◦ f˜r ◦ t r1,q−1 ( e i ) = ∑ a rji (q) e j
j=1
for q ∈ U1 and i = 1, ... , nr1 .

By Proposition 4.45, there are smooth functions A j1 ,..., jk : U2 → R defined by
j j
s2 (q′ ) = ∑ A j1 ,..., jk (q′ ) (e (2,1)
1
◦ t 12,q′ ) ⊗ · · · ⊗ (e (2,k)
k
◦ t k2,q′ )
j1 ,..., jk
for q′ ∈ U2 . From this it follows that

i1 i
s1 (q) = ∑ Bi1 ,...,ik (q) (e (1,1) ◦ t 11,q ) ⊗ · · · ⊗ (e (1,k)
k
◦ t k1,q )
i1 ,...,ik
for q ∈ U1 , where
(1,1) (1,k)
Bi1 ,...,ik (q) = s1 (q) (t 11,q−1 ( e i1 ) , ... ,t k1,q−1 ( e ik ))
(1,1) (1,k)
= s2 ( f (q)) ( f˜1 ◦ t 11,q−1 ( e i1 ) , ... , f˜k ◦ t k1,q−1 ( e ik ))
= ∑ A j1 ,..., jk ( f (q)) a1j1 i1 (q) · · · akjk ik (q).
j1 ,..., jk
TENSORS 89
Since the functions Bi1 ,...,ik : U1 → R are smooth, it follows that s1 is smooth in a
neighbourhood of every point p in V1 and hence is a section of π1′ on V1 .
4.50 Proposition Let π1 : E1 → M1 and π2 : E2 → M2 be two vector bundles of

dimension n, and let ( f˜, f ) be a bundle map from π1 to π2 where f˜ induces a linear
isomorphism between the fibres π1−1(p) and π2−1 ( f (p)) for every p ∈ M1 . Then there
is a bundle map (g̃, f ) between the dual bundles π1∗ : E1∗ → M1 and π2∗ : E2∗ → M2
which is given by
g̃ p = ( f˜p−1 ) ∗
i.e.,
g̃(λ ) = λ ◦ f˜p−1
for p ∈ M1 and λ ∈ π1−1 (p) .
PROOF : Let q0 ∈ M1 , and choose local trivializations (t 1 , π1−1 (U1 )) and (t 2 ,

−1
π2 (U2 )) of π1 and π2 around q0 and f (q0 ), respectively, with U1 ⊂ f −1 (U2 ). Let
(t1′ , π1 ∗ −1 (U1 )) and (t2′ , π2 ∗ −1 (U2 )) be the corresponding local trivializations in the
n ∗
dual bundles, and let E be the standard basis for R with dual basis E .
Then t 2 ◦ f˜ ◦ t 1−1 : U1 × Rn → U2 × Rn is a smooth map given by
(t2 ◦ f˜ ◦ t1−1 ) (q, v) = (p, (t2,p ◦ f˜q ◦ t1,q

−1
) (v)) ,
−1
where t2,p ◦ f˜q ◦ t1,q : Rn → Rn is a linear isomorphism for each q ∈ U1 and where
p = f (q). Since
′ −1
t2,p ◦ g̃ q ◦ t ′ 1,q = r ◦ { (t2,p ◦ f˜q ◦ t1,q
−1 −1 ∗
) } ◦ r−1 ,
∗
′ −1
mEE (t2,p ◦ g̃ q ◦ t ′ 1,q ) = { mEE (t2,p ◦ f˜q ◦ t1,q
−1 −1 t
) } .
Hence t2′ ◦ g̃ ◦ t ′ 1−1 is a smooth map from U1 × Rn to U2 × Rn which is linear on

each fibre, thus showing that (g̃, f ) is bundle map.
4.51 Definition Let π r : E r → M be an nr -dimensional vector bundle over the

smooth manifold M for r = 1, ..., k + l. If ω and η are sections of the bundles
T (E 1 , ..., E k ) and T (E k+1 , ..., E k+l ), respectively, on an open subset V of M, we
define their tensor product ω ⊗ η to be the section of T (E 1 , ..., E k+l ) on V defined
by
(ω ⊗ η )(p) = ω (p) ⊗ η (p)
for p ∈ V . If l = 0, then η is simply a smooth function f on V , and f ⊗ ω and ω ⊗ f
simply mean f ω .
4.52 Definition If V is a vector space, the space
T kl (V ) = T (V , ... ,V ,V ∗ , ... ,V ∗ ) = V ∗ ⊗ · · · ⊗ V ∗ ⊗V ⊗ · · · ⊗ V
| {z } | {z } | {z } | {z }
k l k l
is called the tensor space of type (kl ) on V , or the space of tensors on V which are
covariant of degree k and contravariant of degree l. We let T 00 (V ) = R. The spaces
T k0 (V ) and T 0k (V ) of purely covariant and contravariant tensors of degree k are
usually denoted simply by T k (V ) and Tk (V ), respectively. If B is a basis for V with
dual basis B ∗ , then the basis
T (B , ... , B , B ∗ , ... , B ∗ )
| {z } | {z }
k l
for T kl (V ) defined in Remark 4.27 is denoted by T kl (B). We let T 00 (B) = {1}.

If F : V → W is a linear map, we write F ∗ : T k (W ) → T k (V ) instead of
(F, ..., F)∗ : T k (W ) → T k (V ) . By Proposition 4.33 (4) we have that F ∗ (T ⊗ S) =
F ∗ (T ) ⊗ F ∗ (S) for every covariant tensor T and S on W . If F : V → W is a linear
isomorphism, we also denote the map
( F , ... , F , (F −1 )∗ , ... , (F −1 )∗ )∗ : T kl (W ) → T kl (V )
| {z } | {z }
k l
simply by F ∗ : T kl (W ) → T kl (V ), so that
F ∗ (T )(v1 , ... , vk , λ 1 , ... , λ l ) = T (F(v1 ), ... , F(vk ), λ 1 ◦ F −1 , ... , λ l ◦ F −1 )
for T ∈ T kl (W ), v1 , ... , vk ∈ V and λ 1 , ... , λ l ∈ V ∗ . When k = 0 and l = 1, the vector

spaces T1 (V ) and T1 (W ) may be identified with V and W as in Remark 4.5, and the
map F ∗ : T1 (W ) → T1 (V ) is then identified with F −1 : W → V by Proposition 4.4.
If π : E → M is a vector bundle over the smooth manifold M, we let
T kl (E) = T (E , ... , E , E ∗ , ... , E ∗ )

| {z } | {z }
k l
and denote its projection by πlk . The bundles T k0 (E) and T 0k (E) are usually denoted
simply by T k (E) and Tk (E) with projections π k and πk , respectively.
4.53 Proposition For each local trivialization (t, π −1 (U)) in the n-dimensional
vector bundle π : E → M, we have a corresponding local trivialization (t ′ , (πlk ) −1 (U))
in the tensor bundle πlk : T kl (E) → M, where t p′ = r′′ ◦ (t p−1 )∗ for p ∈ U. Here r′′ :
T kl (Rn ) → Rm is the linear isomorphism defined by
r′′ ( e i1 ⊗ · · · ⊗ e ik ⊗ e j 1 ⊗ · · · ⊗ e j l ) = ees ,
for (i1 , ..., ik , j1 , ..., jl ) ∈ Ink+l , where E = {e 1 , ..., e n } is the standard basis for Rn with
TENSORS 91
dual basis E ∗ = {e 1 , ..., e n } and B = {e e1 , ... , eem } is the standard basis for Rm , and
where s = b (i1 , ..., ik , j1 , ..., jl ) for the bijection b : Ink+l → Im defined in Remark 4.27
with m = nk+l .
We have that t ′ = (id U × r′′ ) ◦ τ , where ( id U × r′′ ,U × T kl (Rn )) is a local
trivialization for the vector bundle pr1 : U × T kl (Rn ) → U , and τ : (πlk ) −1 (U) →
U × T kl (Rn ) is the equivalence over U given by
τ (T ) = (p, (t p−1 )∗ (T ))
for T ∈ T kl (E p ) where p ∈ U.
PROOF : By Remarks 4.9 and 4.40 we have that
t p′ = r′ ◦ ( t p−1 , ... ,t p−1 , t p∗ ◦ r−1 , ... ,t p∗ ◦ r−1 ) ∗ = r′′ ◦ (t p−1 ) ∗

| {z } | {z }
k l
for p ∈ U, where
r′′ = r′ ◦ ( id Rn , ... , id Rn , r−1 , ... , r−1 ) ∗ : T kl (Rn ) → Rm

| {z } | {z }
k l
is given by
r′′ ( e i1 ⊗ · · · ⊗ e ik ⊗ e j 1 ⊗ · · · ⊗ e j l ) = r′ ( e i1 ⊗ · · · ⊗ e ik ⊗ e j 1 ⊗ · · · ⊗ e j l ) = ees
for (i1 , ..., ik , j1 , ..., jl ) ∈ Ink+l . Indeed, we have that
(id Rn , ... , id Rn , r−1 , ... , r−1 ) ∗ ( e i1 ⊗ · · · ⊗ e ik ⊗ e j 1 ⊗ · · · ⊗ e j l )

= e i1 ⊗ · · · ⊗ e ik ⊗ e j 1 ⊗ · · · ⊗ e j l
as both sides have the same effect on e r1 ⊗ · · · ⊗ e rk ⊗ e s 1 ⊗ · · · ⊗ e s l for

(r1 , ..., rk , s1 , ..., sl ) ∈ Ink+l .
4.54 Remark If V1 , ...,Vk are vector spaces and σ ∈ Sk is a permutation, we have

a linear isomorphism
φ : T (Vσ (1) , ...,Vσ (k) ) → T (V1 , ...,Vk )
given by
φ (T ) (v1 , ..., vk ) = T (vσ (1) , ..., vσ (k) )
for every T ∈ T (Vσ (1) , ...,Vσ (k) ) and vi ∈ Vi for i = 1, ... , k. We will not distinguish be-
tween the isomorphic spaces obtained from T (V1 , ...,Vk ) by interchanging the spaces
V1 , ...,Vk .
In particular, if V is a vector space, every tensor product with k factors V ∗ and
l factors V in any order is identified with T kl (V ) with the canonical order given in
Definition 4.52 placing all the factors V ∗ to the left of V . Hence if T ∈ T kl11 (V ) and
+k2
S ∈ T kl22 (V ), their tensor product T ⊗ S ∈ T kl11+l2
(V ) is given by
(T ⊗ S) (v 1 , ... , v k1 +k2 , λ 1 , ... , λ l1 +l2 )

= T (v 1 , ... , v k1 , λ 1 , ... , λ l1 ) S(v k1 +1 , ... , v k1 +k2 , λ l1 +1 , ... , λ l1 +l2 )
where vi ∈ V for i = 1, ... , k1 + k2 and λ j ∈ V ∗ for j = 1, ... , l1 + l2 . If F : V → W is

a linear isomorphism, we have that F ∗ (T ⊗ S) = F ∗ (T ) ⊗ F ∗ (S).
4.55 Proposition Let ( f˜, f ) be a bundle map from a vector bundle π1 : E1 → M1

to a vector bundle π2 : E2 → M2 , and let s2 be a section of (π2 )kl : T kl (E2 ) → M2
on M2 . If l > 0 , we suppose that f˜ induces a linear isomorphism between the fibres
π1−1 (p) and π2−1 ( f (p)) for every p ∈ M1 . Then the map s1 : M1 → T kl (E1 ) defined
by
s1 (p) = ( f˜p )∗ s2 ( f (p))
for p ∈ M1 , i.e.,
s1 (p)(v1 , ... , vk , λ 1 , ... , λ l ) = s2 ( f (p))( f˜p (v1 ), ... , f˜p (vk ) , λ 1 ◦ f˜p−1 , ... , λ l ◦ f˜p−1 )
for all v1 , ..., vk ∈ π1 −1 (p) and λ1 , ..., λl ∈ π1∗ −1 (p) where p ∈ M1 , is a section of
(π1 )kl : T kl (E1 ) → M1 on M1 .

4.56 Remark When k = 0 and l = 1, we identify T1 (Er ) = Er∗∗ with Er by means

of the equivalence iEr : Er → Er ∗∗ defined in Proposition 4.10 for r = 1, 2. If s2 is a
section of π2 : E2 → M2 on M2 , we have that
[ iE1 ◦ f˜ ∗ (s2 ) ] (p) (λ ) = iE1 ◦ f˜p−1 ◦ s2 ◦ f (p) (λ ) = λ ◦ f˜p−1 ◦ s2 ( f (p))

= (iE2 ◦ s2 ) ( f (p)) (λ ◦ f˜p−1 ) = f˜ ∗ (iE2 ◦ s2 ) (p) (λ )
for λ ∈ π1∗ −1 (p) where p ∈ M1 , which shows that the above definition of the pull-
back of s2 is consistent with the one in Proposition 2.57.
4.57 Definition If M is a smooth manifold of dimension n, the bundle Tlk (M) =

T kl (T M) with projection πlk is called the tensor bundle of type (kl ) over M. A section
of this bundle on an open subset V of M is called a tensor field of type (kl ) on
V . The set of all tensor fields of type (kl ) on V is denoted by T kl (V ). In particular
T 00 (V ) = F (V ) is the set of C∞ -functions on V .
The bundles T0k (M) and Tk0 (M) of purely covariant and contravariant tensors of
degree k are usually denoted simply by T k (M) and Tk (M) with projections π k and πk .
TENSORS 93
Their sections on an open subset V of M are denoted by T k (V ) and Tk (V ) and are

called covariant and contravariant tensor fields of degree k on V , respectively.
4.58 Proposition Let πlk : Tlk (M) → M be the tensor bundle of type (kl ) over a
smooth manifold M n and T : M → Tlk (M) be a map with πlk ◦ T = idM . Then T is a
j ,..., j
tensor field of type (kl ) on M if and only if the functions Ai11,...,ikl : U → R defined by

T (p) = ∑
j ,..., j
Ai11,...,ikl (p) dxi1 (p) ⊗ · · · ⊗ dxik (p) ⊗ ∂ ⊗ ··· ⊗ ∂j
i1 ,...,ik
∂ x j1 ∂x l
p p
j1 ,..., jl

for p ∈ U are smooth for every local chart (x,U) on M. Here each ∂ jr for r =
∂x
p
1, ..., l is considered as a functional on Tp∗ M as described in Remark 4.5.
PROOF : Follows from Remark 2.82 and Propositions 4.11, 4.15 and 4.45.
4.59 For each tensor field T of type (kl ) on M we have a multilinear map
TM : T1 (M) × · · · × T1 (M) × T 1 (M) × · · · × T 1 (M) → F (M)

| {z } | {z }
k l
defined by
TM (X1 , ... , Xk , λ1 , ... , λl ) (p) = T (p) (X1 (p) , ... , Xk (p) , λ1 (p) , ... , λl (p))
for each point p on M and for vector fields X1 , ... , Xk ∈ T1 (M) and 1-forms
λ1 , ... , λl ∈ T 1 (M) as described in 4.46. Conversely, each such multilinear map
arises from a unique tensor field in this way by Proposition 4.47. Hence we will
not distinguish between the tensor field T and the map TM , and a tensor field of type
(kl ) on M can be thought of as an operation on k vector fields and l 1-forms yielding
a smooth funnction on M.
4.60 If f : M → N is a diffeomorphism and T is a tensor field of type (kl ) on N,

then we denote its pullback ( f∗ ) ∗ (T ) simply by f ∗ (T ). If N is an open submanifold
of a manifold P and T is a tensor field on P, we also denote f ∗ (T |N ) simply by f ∗ (T ).
We use the same notation when f : M → N is a smooth map and T is a covariant
tensor field of degree k on an open subset V of N. In particular, when k = 0 then T is
simply a smooth function on V , and f ∗ (T ) = T ◦ f .
It follows from Proposition 4.49 that if T is covariant of degree k on V , then
f ∗ (T ) is a covariant tensor field of degree k on the open subset f −1 (V ) of M, but we
also want to write out the second proof of Proposition 4.49 explicitely in this case
using Proposition 4.58.
4.61 Proposition If f : M → N is a smooth map and T is a covariant tensor field

of degree k on an open subset V of N, then the map f ∗ (T ) : f −1 (V ) → T k (M) defined

by
f ∗ (T ) (p) = ( f∗ p )∗ T ( f (p))
i.e.,
f ∗ (T ) (p) (v1 , ... , vk ) = T ( f (p)) ( f∗ (v1 ) , ... , f∗ (vk ))
for p ∈ f −1 (V ) and v1 , ... , vk ∈Tp M, is a covariant tensor field of degree k on f −1 (V ).
PROOF : Let p be a point in f −1 (V ), and let (x,U) and (y,W ) be local charts around
p and f (p), respectively, with W ⊂ V and U ⊂ f −1 (W ). Then there are smooth
functions A j1 ,... , jk : W → R defined by
T (q′ ) = ∑ A j1 ,..., jk (q′ ) dy j1 (q′ ) ⊗ · · · ⊗ dy jk (q′ )

j1 ,... , jk
for q′ ∈ W . From this it follows that
f ∗ (T )(q) = ∑ Bi1 ,... ,ik (q) dxi1 (q) ⊗ · · · ⊗ dxik (q)

i1 ,... ,ik
for q ∈ U, where
!

∗
Bi1 ,...,ik (q) = f (T )(q) , ∂
... , ∂
∂ xi1 ∂x k
i
! q q!!
∂
∂

= T ( f (q)) f∗ i1 , ... , f∗ ik
∂x ∂x
q q
!! !!
∂
∂

j1 jk
= ∑ A j1 ,..., jk ( f (q)) dy ( f (q)) f∗ i1 · · · dy ( f (q)) f∗ ik
j1 ,..., jk
∂x ∂x
q q
!! !!
∂
∂

= ∑ A j1 ,..., jk ( f (q)) f∗ (y j1 ) · · · f∗ (y jk )
j ,..., j
∂ xi1 ∂ xik
1 k q q
∂ (y j1 ◦ f ) ∂ (y jk ◦ f )
= ∑ A j1 ,..., jk ( f (q)) i1 (q) · · · (q) .
j1 ,..., jk
∂x ∂ xik
Since the functions Bi1 ,...,ik : U → R are smooth, it follows that f ∗ (T ) is smooth in
a neighbourhood of every point p in f −1 (V ) and hence is a covariant tensor field of
degree k on f −1 (V ).
4.62 Proposition Let f : M → N be a diffeomorphism, and let T be a tensor field

of type (kl ) on N. If (x,U) and (y,V ) are local charts on M and N, respectively, and if

T (p) = ∑ Asr11 ,...,s r1 rk ∂
,...,rk (p) dy (p) ⊗ · · · ⊗ dy (p) ⊗ ∂ ys1
l ⊗ ···⊗ s
∂
r1 ,...,rk
∂y
l
s1 ,...,sl p p
TENSORS 95
for p ∈ V , then

∗
f (T )(q) = ∑
j ,..., j
Bi11,...,ikl (q) dxi1 (q) ⊗ · · · ⊗ dxik (q) ⊗ ∂ ⊗ ··· ⊗ ∂j
i1 ,...,ik
∂ x j1 ∂x
l
q q
j1 ,..., jl
where
j ,..., j ∂ (yr1 ◦ f ) ∂ (yrk ◦ f )
Bi11,...,ikl (q) = ∑ Asr11 ,...,s
r ,...,r
,...,rk ( f (q))
l
∂x i1 (q) · · ·
∂ xik
(q)
1 k
s1 ,...,sl
∂ (x j1 ◦ f −1 ) ∂ (x jk ◦ f −1 )
∂y s1 ( f (q)) · · · ∂ ysk
( f (q))
for q ∈ U ∩ f −1 (V ).
PROOF : Since
!! !
∂
∂
∂ (yr ◦ f )
dyr ( f (q)) f∗ = f∗ (yr ) = (q)
∂ xi ∂ xi ∂ xi
q q
and
! !
∂
∂
∂ (x j ◦ f −1 )
j −1 −1
dx (q) ◦ ( f )∗ ∂ ys
= (f )∗ ∂ ys
(x j ) = ∂ ys
( f (q)) ,
f (q) f (q)
it follows from Proposition 4.58 that

!

j ,..., j
Bi11,...,ikl (q) = ∗
f (T )(q) ∂ , ... , i , dx l (q) , ... , dx l (q)
∂ j j

∂ xi1 ∂x k
q q
! ! !

= T ( f (q)) f∗ ∂ , ..., f∗ ∂
, dx jl (q) ◦ ( f −1 )∗ , ..., dx jl (q) ◦ ( f −1 )∗
∂ xi1 ∂ xik
q q
∂ (yr1 ◦ f ) ∂ (yrk ◦ f )
= ∑ Asr11 ,...,s
,...,rk ( f (q))
l
i1 (q) · · · (q)
r1 ,...,rk ∂x ∂ xik
s1 ,...,sl
∂ (x j1 ◦ f −1 ) ∂ (x jk ◦ f −1 )
∂ y s1
( f (q)) · · · ∂ y sk
( f (q)).
CONTRACTION
4.63 Definition If A = (ai j ) is an n × n-matrix, we define the trace of A to be
n
tr (A) = ∑ aii ,
i=1
i.e., the trace is the sum of the diagonal elements.
4.64 Proposition If A and B are n × n-matrices, then
tr (AB) = tr (BA) .
PROOF : If A = (ai j ) and B = (bi j ), then

n n n n
tr (AB) = ∑ ∑ aik bki = ∑ ∑ bki aik = tr (BA) .
i=1 k=1 k=1 i=1
4.65 Corollary If A and B are n × n-matrices and B is invertible, then
tr (BAB−1 ) = tr (A) .
PROOF : It follows from Proposition 4.64 that
tr (BAB−1 ) = tr (B−1 BA) = tr (A) .
4.66 Definition If V is an n-dimensional vector space and F : V → V is a linear

map, we define the trace of F to be
tr (F) = tr (mB
B (F))
where B is a basis for V . It follows from Corollary 4.65 that tr (F) is well defined
and does not depend on the choice of B. Indeed, if C is another basis for V , we have
that
tr (mCC (F)) = tr (mB B C B
C (id) mB (F) mB (id)) = tr (mB (F)) .
4.67 If V is an n-dimensional vector space, we have a linear isomorphism
φ : L(V,V ) → T 11 (V )
given by φ (F) = T where

T (v, λ ) = λ (F(v))
for v ∈ V , λ∈ V∗ and F ∈ L(V,V ).
If B = {v1 , ..., vn } is a basis for V with dual basis B ∗ = {v∗1 , ..., v∗n }, then
φ −1 (T ) = F is given by
n n
F(v) = ∑ v∗i (F(v)) vi = ∑ T (v, v∗i ) vi
i=1 i=1
for v ∈ V and T ∈ T 11 (V ). Now if
T = ∑ Ti j v∗i ⊗ v j ,
i, j
TENSORS 97
it follows in particular that

n n
F(v j ) = ∑ T (v j , v∗i ) vi = ∑ T ji vi
i=1 i=1
so that mB i
B (F) = A where Ai j = T j .
4.68 Example The tensor δ ∈ T 11 (V ) corresponding to the identity map id ∈

L(V,V ) is called the Kronecker delta on the vector space V . It is given by
δ (v, λ ) = λ (v)
for v ∈ V and λ ∈ V ∗ . If B = {v1 , ..., vn } is any basis for V with dual basis B ∗ =
{v∗1 , ..., v∗n }, then
δ = ∑ v∗i ⊗ vi .
i
If M is a smooth manifold, we have a tensor field T of type (11 ) on M given by

T (p) = δ p for every p ∈ M, where δ p is the Kronecker delta on Tp M. We have that
T |U = ∑ dxi ⊗ ∂ i
i
∂x
for every local chart (x,U) on M, showing that T is in fact a tensor field. It is called
the Kronecker delta tensor on M and is also denoted by δ .
4.69 Definition We have a linear map
C : T 11 (V ) → R ,
called a contraction , defined by C(T ) = tr (F) where F = φ −1 (T ).
4.70 Proposition If B = {v1 , ..., vn } is a basis for V with dual basis B ∗ =

{v∗1 , ..., v∗n }, and if
j
T = ∑ Ti v∗i ⊗ v j ,
i, j
then
C(T ) = ∑ Tii .
i
PROOF : Follows from 4.67.
4.71 Definition We have a linear map
Csr : T kl (V ) → T k−1
l−1 (V ) ,
called a contraction in the r-th covariant and the s-th contravariant index, which
is defined in the following way. Let T ∈ T kl (V ), and fix v1 , ... , vk−1 ∈ V and
λ1 , ... , λl−1 ∈ V ∗ . Then
Csr (T ) (v1 , ... , vk−1 , λ1 , ... , λl−1 ) = C (S) ,
where S ∈ T 11 (V ) is the tensor defined by
S(v, λ ) = T (v1 , ... , vr−1 , v, vr , ... , vk−1 , λ1 , ... , λs−1 , λ , λs , ... , λl−1 )
for v ∈ V and λ ∈ V ∗ .
4.72 Proposition If B = {v1 , ..., vn } is a basis for V with dual basis B ∗ =

{v∗1 , ..., v∗n }, and if
j ,..., j
T= ∑ A i11,...,ikl v∗i1 ⊗ · · · ⊗ v∗ik ⊗ v j1 ⊗ · · · ⊗ v j l ,
i1 ,...,ik
j1 ,..., j l
then
j ,..., j
l−1 ∗
Csr (T ) = ∑ B i11,...,ik−1 vi1 ⊗ · · · ⊗ v∗ik−1 ⊗ v j1 ⊗ · · · ⊗ v j l−1
i ,...,i
1 k−1
j1 ,..., j l−1
where
j ,..., j j ,..., j ,m, j ,..., j
B i11,...,ik−1
l−1
= ∑ A i11,...,ir−1
s−1 s l−1
,m,ir ,...,ik−1 .
m
PROOF : Follows from Definition 4.71 and Proposition 4.70.
4.73 Definition Let T be a tensor field of type (kl ) on a smooth manifold M. Then
we have a tensor field S of type (k−1
l−1 ) on M given by
S(p) = Csr (T (p))

j ,..., j j ,..., j
for every p ∈ M. If (x,U) is a local chart on M, and if Ai11,...,ikl and Bi11,...,ik−1
l−1
are the
components of the local representations of T and S on U, respectively, then it follows
from Proposition 4.72 that
j ,..., j j ,..., j ,m, j ,..., j
B i11,...,ik−1
l−1
= ∑ A i11,...,ir−1
s−1 s l−1
,m,ir ,...,ik−1 ,
m
showing that S is in fact a tensor field. It is called the contraction of T in the r-th
covariant and the s-th contravariant index, and it is also denoted by Csr (T ).
4.74 Proposition Let C : T kl (V ) → T k−l 0 (V ) where k ≥ l be the l-fold contrac-

tion in the i-th contravariant and the i-th covariant index for 0 ≤ i ≤ l. Then we have
that
C(T ⊗ λ 1 ⊗ · · · ⊗ λ k ) = T (λ 1 , ... , λ l ) λ l+1 ⊗ · · · ⊗ λ k
for every T ∈ Tl (V ) and λ 1 , ... , λ k ∈ T 1 (V ).
TENSORS 99
PROOF : Let B = {v1 , ..., vn } be a basis for V with dual basis B ∗ = {v∗1 , ..., v∗n }, and
suppose that
T = ∑ A j1 ,..., jl v j1 ⊗ · · · ⊗ v jl
j1 ,..., jl
and
λ r = ∑ arir v∗ir
ir
for r = 1, ..., k. Then it follows that
T ⊗ λ 1 ⊗ ···⊗ λ k = ∑ A j1 ,..., jl a1i1 · · · akik v∗i1 ⊗ · · · ⊗ v∗ik ⊗ v j1 ⊗ · · · ⊗ v jl

i ,...,i
1 k
j1 ,..., jl
so that
C(T ⊗ λ 1 ⊗ · · · ⊗ λ k )
= ∑ Ai1 ,...,il a1i1 · · · alil ∑ al+1 k ∗ ∗
il+1 · · · aik vil+1 ⊗ · · · ⊗ vik
i1 ,...,il il+1 ,...,ik
= T (λ 1 , ... , λ l ) λ l+1 ⊗ · · · ⊗ λ k .
THE LIE DERIVATIVE

4.75 Definition Let X be a vector field on a smooth manifold M with global flow
γ : D(X) → M, and let T be a tensor field of type (kl ) on M.
If p is a point on M, then γ t : Dt (X) → M is a diffeomorphism from the open
neighbourhood Dt (X) of p onto the open neighbourhood D−t (X) of γ t (p) with in-
verse γ −t for each t ∈ I(p), and we define the Lie derivative of T at p with respect
to X to be
d ∗

LX T (p) = lim 1 [ γ t∗ T (p) − T (p) ] = dt γ t T (p) ,
t→0 t
0
k
using the vector space topology on T l (Tp M) defined in Proposition 13.117 in the
appendix.
4.76 Proposition If c : I(p) → T kl (Tp M) is the curve defined by c(t) = γ t∗ T (p)

and we canonically identify Tc(0) T kl (Tp M) with T kl (Tp M) as in Lemma 2.84, then
LX T (p) = c′ (0) , and the Lie derivative LX T is a well-defined tensor field of type (kl )
on M.
Let (x,U) be a local chart on M, and let
n
X | U = ∑ ai ∂ i
i=1
∂x
and
Ai11,...,ik l dxi1 ⊗ · · · ⊗ dxik ⊗ ∂ j1 ⊗ · · · ⊗ ∂ j l .
j ,..., j
T |U = ∑
i ,...,i ∂x ∂x
1 k
j1 ,..., j l
Then we have that

Bi11,...,ik l dxi1 ⊗ · · · ⊗ dxik ⊗ ∂ j1 ⊗ · · · ⊗ ∂ j l
j ,..., j
LX T | U = ∑ ∂x ∂x
i1 ,...,ik
j1 ,..., j l
where
j ,... , j l
j ,..., j
n ∂ Ai 1,...,i l n
j ,... , j , j, jm+1 ,..., j l ∂ a jm
i 1
Bi11,...,ik l = ∑a ∂x i
k
− ∑∑ Ai11,...,ik m−1
∂xj
i=1 m=1 j=1
k n
j ,..., j ∂ ai
+ ∑ ∑ Ai11,... ,im−1
l
,i,im+1 ,...,ik ∂ xim
.
m=1 i=1
PROOF : If p 0 is a point in U, we may choose an open neighbourhood W of p 0

contained in U and a an open interval I containing 0 such that I × W ⊂ D(X) and
γ (I ×W ) ⊂ U. By Proposition 3.42 (3) we know that γ t : W → γ t (W ) is a diffeomor-
phism for each t ∈ I with inverse γ −t . We must show that the curve c : I → T kl (Tp M)
defined by c(t) = γ t∗ T (p) is smooth so that LX T (p) is a well-defined tensor in
T kl (Tp M). By Proposition 4.62 we have that

γt∗ (T )(q) = ∑
j ,..., j
Ct i11,...,ik l (q) dxi1 (q) ⊗ · · · ⊗ dxik (q) ⊗ ∂ ⊗ ···⊗ j
∂
i1 ,...,ik
∂ x j1 ∂x l
q q
j1 ,..., j l
where
j ,..., j ∂ (xr1 ◦ γt ) ∂ (xrk ◦ γt )
Ct i11,...,ik l (q) = ∑ Ars11 ,...,s
r ,...,r
,...,rk (γt (q))
l
∂x i1 (q) · · ·
∂ xik
(q)
1 k
s1 ,...,s l
(1)
∂ (x j1 ◦ γ−t ) ∂ (x j l ◦ γ−t )
· ∂ x s1
(γt (q)) · · · ∂ xs l
(γt (q))
for q ∈ W .
We now transform the problem to open sets in Rn using the local chart (x,U).
The coordinate map x can be written as x = i ◦ x̃, where x̃ : U → x(U) is a diffeomor-
phism and i : x(U) → Rn is the inclusion map. Consider the functions f i = ai ◦ x−1
j ,... , j j ,... , j j ,... , j j ,... , j
and Fi11,...,ik l = Ai11,...,ik l ◦ x−1 on x(U) and Ht i11,...,ik l = Ct i11,...,ik l ◦ x−1 on x(W ) repre-
senting X, T and γ t∗ T , and let
α : I × x(W ) → x(U)
be the local flow for f given by α = x̃ ◦ γ ◦ (id × x−1 ) , where γ is the local flow for
X on I × W . Then
D1 α (t, u) = f (α (t, u)) and α (0, u) = u (2)
TENSORS 101
for all t ∈ I and u ∈ x(W ), and we have that αt = x̃ ◦ γt ◦ x−1 . If u = x(q), it follows
from (1) that
j ,..., j ,...,s l
Ht i11,...,ik l (u) = ∑ Frs11,...,r k
(α (t, u)) Di1 +1 α r1 (t, u) · · · Dik +1 α rk (t, u)
r1 ,...,rk
s1 ,...,s l
(3)
j1 jl
· Ds1 +1 α (−t, α (t, u)) · · · Ds l +1 α (−t, α (t, u))
for u ∈ x(W ), which shows that the curve c is smooth and that the Lie derivative LX T
is a well-defined tensor field of type (kl ) on M.
Using (2) we can now find the derivative of each term in (3) with respect to t at
t = 0. First note that
Di+1 α r (0, u) = Di ur = δir and Ds+1,i+1 α r (0, u) = 0 .
Hence we have that

d
n n s ,... ,s
∂ Ar11 ,...,rkl
s1 ,...,s l ,...,s l
dt (Fr1 ,...,rk (α (t, u))) = ∑ Di Frs11,...,r k
(u) f i (u) = ∑ ai (q) ∂ xi
(q) ,
0 i=1 i=1

d
∂ ar
r r r
dt (Di+1 α (t, u)) = Di+1 D1 α (0, u) = Di f (u) = ∂ xi (q) ,
0
and

d (Ds+1 α j (−t, α (t, u))) = −Ds+1 D1 α j (0, u) = −Ds f j (u) = − ∂ asj (q) .
dt ∂x
0
j ,..., j
Inserting this in (3), we obtain the formula for Bi11,...,ik l in the proposition.
4.77 Remark If f is a smooth function on M, the formula for LX f | U reduces to

n
∂f
LX f | U = ∑ ai ∂ xi ,
i=1
which shows that LX f = X( f ).

If Y is a vector field on M with
n
Y |U = ∑ bj ∂j ,
j=1
∂x
we have that
j j
LX Y | U = ∑ ai ∂ b i − b i ∂ a i ∂
j .
i, j
∂x ∂x ∂x
4.78 Remark Let B = {v1 , ... , vn } be a basis for Tp M with dual basis B ∗ =
{v1 , ... , vn }, and let c : I → M be an integral curve for X with initial condition p given
by c(t) = γ t (p). If vi (t) = γ t ∗ (vi ) and vi (t) = vi ◦ γ −t ∗ for i = 1, ... , n and t ∈ I, then
Bt = {v1 (t), ... , vn (t)} is a basis for Tc(t) M with dual basis Bt∗ = {v1 (t), ... , vn (t)},
since (γ t ) ∗ p : Tp M → Tc(t) M is a linear isomorphism. We say that vi (t) and vi (t) are
obtained from vi and vi , respectively, by Lie transport along the integral curve c. If
j ,..., j
T (c(t)) = ∑ C p i11,...,ik l (t) vi1 (t) ⊗ · · · ⊗ vik (t) ⊗ v j1 (t) ⊗ · · · ⊗ v j l (t) ,
i1 ,...,ik
j1 ,..., j l
then
γ t∗ T (p) (vi1 , ... , vik , v j1 , ... , v j l )

= T (γ t (p)) (γ t ∗ (vi1 ) , ... , γ t ∗ (vik ), v j1 ◦ γ −t ∗ , ... , v j l ◦ γ −t ∗ )
j ,..., j
= C p i11,...,ik l (t)
so that
j ,..., j
γ t∗ T (p) = ∑ C p i11,...,ik l (t) vi1 ⊗ · · · ⊗ vik ⊗ v j1 ⊗ · · · ⊗ v j l .
i1 ,...,ik
j1 ,..., j l
Hence we have that

d
j1 ,..., j l i1 ik
LX T (p) = ∑
i1 ,...,ik
dt C p i1 ,...,ik v ⊗ · · · ⊗ v ⊗ v j1 ⊗ · · · ⊗ v j l .
0
j1 ,..., j l
Let (x,U) be a local chart around p, and choose an open neighbourhood W of p

open interval I containing 0 such that I ×W ⊂ D(X) and γ (I ×
contained in U and an

W ) ⊂ U. If vi = ∂ i for i = 1, ... , n and q ∈ W , then the functions C i11,...,ik l : I ×W →
j ,..., j
∂x
q
j1 ,..., j l j ,..., j j ,..., j
R defined by C i1 ,...,ik (t, q) = Cq i11,...,ik l (t) are smooth since Cq i11,...,ik l (t) coincides
j ,..., j
with Ct i11,...,ik l (q) in the proof of Proposition 4.76.
4.79 Proposition Let X be a vector field on the smooth manifold M.

(1) LX is linear over R.
(2) If T ∈ T kl11 (M) and S ∈ T kl22 (M), then LX (T ⊗ S) = (LX T ) ⊗ S + T ⊗ (LX S).
(3) LX δ = 0, where δ ∈ T 11 (M) is the Kronecker delta tensor.

(4) LX commutes with contractions, i.e., if T ∈ T kl (M), and if 1 ≤ r ≤ k and 1 ≤
s ≤ l, then LX (Csr T ) = Csr (LX T ).
TENSORS 103
(5) If T ∈ T kl (M), and if X1 , ... , Xk ∈ T1 (M) and λ1 , ... , λl ∈ T 1 (M), then
LX [ T (X1 , ... , Xk , λ1 , ... , λl ) ] = (LX T ) (X1 , ... , Xk , λ1 , ... , λl )

k
+ ∑ T (X1 , ... , LX Xi , ... , Xk , λ1 , ... , λl )
i=1
l
+ ∑ T (X1 , ... , Xk , λ1 , ... , LX λi , ... , λl ) .
i=1
PROOF : Point (1) to (4) follow directly from Remark 4.78. Point (5) follows from
(2) and (4) since T (X1 , ... , Xk , λ1 , ... , λl ) can be obtained from T ⊗ X1 ⊗ · · · ⊗ Xk ⊗
λ1 ⊗ · · · ⊗ λl by applying contractions repeatedly.
4.80 Definition A Lie algebra is a vector space A with a bilinear map [ , ] :

A × A → A , called the bracket product or just bracket, satisfying
(1) [ X,Y ] = − [Y, X ] ,
(2) [ X, [Y, Z ] ] + [Y, [ Z, X ] ] + [ Z, [ X,Y ] ] = 0
for every X,Y, Z ∈ A .
4.81 Remark A bilinear map satisfying (1) is called skew symmetric , and the
identity (2) is called the Jacobi identity .
A bilinear map satisfies (1) if and only if it is alternating , i.e., if
(3) [ X, X ] = 0 for every X ∈ A .
Indeed, it follows from (3) that
0 = [ X + Y, X + Y ] = [ X, X ] + [ X,Y ] + [Y, X ] + [Y,Y ] = 0 + [ X,Y ] + [Y, X ] + 0
which implies (1). Conversely, if (1) is satisfied, we have that [ X, X ] is equal to its
own inverse and hence must be 0 for every X ∈ A , thus implying (3).
4.82 Proposition Every associative algebra A is a Lie algebra, where the bracket
[ X,Y ] = XY − Y X is the commutator product.
PROOF : We immediately see that the commutator product is bilinear and skew sym-
metric, and it only remains to prove the Jacobi identity. If X,Y, Z ∈ A , we have that
[ X, [Y, Z ] ] + [Y, [ Z, X ] ] + [ Z, [ X,Y ] ] = { X (Y Z − ZY ) − (Y Z − ZY ) X }

+ {Y (ZX − X Z) − (ZX − X Z)Y }
+ { Z (XY − Y X ) − (XY − Y X ) Z } = 0
since the product of every permutation of X, Y and Z occurs once with a plus and
once with a minus in the formula.
4.83 Remark As special cases we have the Lie algebra End (V ) of endomor-
phisms of a real or complex vector space V , and the Lie algebras gl (n, R) and
gl (n, C) of real or complex n × n-matrices, all considered as vector spaces over R.
4.84 Definition A subspace A of a Lie algebra B which is closed under the

bracket product [ , ] , is called a Lie subalgebra of B. It is called an ideal in B if
[ X,Y ] ∈ A for every X ∈ B and Y ∈ A .
A linear map φ : A → B between the Lie algebras A and B is called a Lie
algebra homomorphism if
φ ([ X,Y ]) = [ φ (X), φ (Y ) ]
for every X,Y ∈ A . If φ is also bijective, it is called a Lie algebra isomorphism. A
Lie algebra isomorphism of a Lie algebra with itself is called a Lie algebra auto-
morphism. The linear map φ : A → B is called a Lie algebra anti-homomorphism
if
φ ([ X,Y ]) = −[ φ (X), φ (Y ) ]
for every X,Y ∈ A .
If B = End (V ) for a vector space V , or if B = gl (n, R), then a Lie algebra
homomorphism φ : A → B is called a representation of the Lie algebra A .
4.85 Proposition Let A be an algebra. Then the set D (A ) of derivations of A

is a Lie subalgebra of End (A ).
PROOF : If X,Y ∈ D (A ), we have that
[ X,Y ] ( f g) = X {Y ( f ) g + f Y (g) } − Y { X( f ) g + f X(g) }

= { XY ( f ) g + Y ( f ) X(g) + X( f ) Y (g) + f XY (g) }
− {Y X( f ) g + X( f ) Y (g) + Y ( f ) X(g) + f Y X(g) }
= [ X,Y ] ( f ) g + f [ X,Y ] (g)
for every f , g ∈ A , which shows that [ X,Y ] is also a derivation of A .
4.86 Proposition If X and Y are vector fields on a smooth manifold M n , then

their commutator product [ X,Y ] , defined by
[ X,Y ] ( f ) = X(Y ( f )) − Y (X( f ))
for f ∈ F (M), is a vector field on M, and we have that [ X,Y ] = LX Y .
PROOF : The assertion that [ X,Y ] is a vector field follows by Remark 3.9 and Propo-
sition 4.85. To prove the last assertion, let (x,U) be a local chart on M, and let
n n
X | U = ∑ ai ∂ i and Y |U = ∑ bj ∂j .
i=1
∂x j=1
∂x
TENSORS 105
Then we have that

!
n n j 2
∂f i ∂b ∂ f i j ∂ f
X(Y ( f )) | U = ∑ a ∂ ii
∑ b ∂x j j
= ∑ a i j +a b i j ,
i=1
∂x j=1 i, j
∂x ∂x ∂x ∂x
and we similarly obtain

j ∂f 2
i j ∂ f
Y (X( f )) | U = ∑ bi ∂ ai j +b a
i, j
∂x ∂x i
∂x ∂x j
so that
j j ∂f
i ∂b i ∂a
[ X,Y ] ( f ) | U = ∑ a
∂xi −b i
∂x j .
∂x
i, j
Hence it follows from Remark 4.77 that [ X,Y ] = LX Y .
4.87 Proposition Let M be a smooth manifold. If X,Y ∈ T 1 (M) and f , g ∈ F (M),

then
[ f X, gY ] = f g [ X,Y ] + f X(g)Y − g Y ( f ) X .
PROOF : By Propositions 4.79 (2) and 4.86 we have that
[ X, gY ] = LX (gY ) = (LX g)Y + g (LX Y ) = g [ X,Y ] + X(g)Y
which also implies
[ f X,Y ] = − [Y, f X ] = − ( f [Y, X ] + Y ( f ) X ) = f [ X,Y ] − Y ( f ) X .
Combining this, we have that
[ f X, gY ] = g [ f X,Y ] + f X(g)Y = f g [ X,Y ] + f X(g)Y − g Y( f ) X .
4.88 Proposition Let f : M → N be a smooth map, and let Xi and Yi be vector

fields on M and N, respectively, which are f -related for i = 1, 2. Then the commutator
products [ X1 , X2 ] and [Y1 ,Y2 ] are also f -related.
PROOF : By Proposition 3.11 we have that
Yi (g) ◦ f = Xi (g ◦ f )
for g ∈ F (N) and i = 1, 2, which implies that
[Y1 ,Y2 ] (g) ◦ f = Y1 (Y2 (g)) ◦ f − Y2 (Y1 (g)) ◦ f

= X1 (Y2 (g) ◦ f ) − X2 (Y1 (g) ◦ f )
= X1 (X2 (g ◦ f )) − X2 (X1 (g ◦ f )) = [ X1 , X2 ] (g ◦ f ).
4.89 Proposition Let f : M → N be a diffomorphism. If X is a vector field and T

a tensor field of type (kl ) on N, then
f ∗ (LX T ) = L f ∗ (X) f ∗ (T ) .
PROOF : Let γ and β be the global flows for f ∗ (X) and X, respectively, and let q ∈ M
and p = f (q). If c : I(q) → T kl (Tq M) and b : J(p) → T kl (Tp N) are the curves defined
by
c(t) = γ t∗ f ∗ (T )(q) and b(t) = β t∗ T (p) ,
and we canonically identify Tc(0) T kl (Tq M) with T kl (Tq M) and Tb(0) T kl (Tp N) with
T kl (Tp N) as in Lemma 2.84, then
L f ∗ (X) f ∗ (T )(q) = c′ (0) and LX T (p) = b′ (0) .
Now β t = f ◦ γ t ◦ f −1 by Proposition 3.49, so that
c(t) = f ∗ ( f ◦ γ t ◦ f −1 )∗ (T )(q) = ( f∗ q )∗ (β t∗ T )(p) = ( f∗ q )∗ ◦ b(t)
where ( f∗ q )∗ : T kl (Tp N) → T kl (Tq M) is the linear map defined in Definition 4.52.

Hence it follows from Lemma 2.84 that
L f ∗ (X) f ∗ (T )(q) = c′ (0) = ( f∗ q )∗ ◦ b′ (0) = f ∗ (LX T )(q) .
4.90 Proposition Let X be a vector field on a smooth manifold M with global

flow γ : D(X) → M, and let T be a tensor field of type (kl ) on M. If p is a point on M
and c : I(p) → T kl (Tp M) is the curve defined by c (t) = γ t∗ T (p), then we have that
c ′ (t0 ) = γ t∗0 (LX T )(p)
for every t0 ∈ I(p) if we canonically identify Tc(t0 ) T kl (Tp M) with T kl (Tp M) as in

Lemma 2.84.
PROOF : Let c 0 : I(p) − t0 → T kl (Tp M) be the curve defined by c 0 (s) = c (s + t0 ) =

γ s+t0 T (p) so that c ′ (t0 ) = c′0 (0). Consider the diffeomorphism γ t0 : Dt0 (X) →
∗
D−t0 (X), and denote the restriction of X to D−t0 (X) by X0 . Let J(q) be the domain of
the maximal integral curve for X0 with initial condition q. If α and β are the flows for
X0 and γ t∗0 (X0 ), respectively, it follows from Proposition 3.49 that α s = γ t0 ◦ β s ◦ γ −t0
for every s ∈ R so that α s ◦ γ t0 = γ t0 ◦ β s . Since α s is the restriction of γ s to Ds (X0 ), it
follows by Proposition 3.42 (2) that α s ◦ γ t0 is the restriction of γ s+t0 to γ −t0 (Ds (X0 )).
Hence we have that
c 0 (s) = β s∗ (γ t∗0 T ) (p)
for every s ∈ J(γ t0 (p)) so that
c′0 (0) = Lγ ∗t (X) γ t∗0 (T ) (p) = γ t∗0 (LX T ) (p) .

0
TENSORS 107
4.91 Proposition Let X and Y be vector fields on a smooth manifold M with

global flows α and β , respectively, such that [X,Y ] = 0 on M. Let p be a point on M,
and suppose there are open intervals I and J containing 0 such that both expressions
α (t, β (s, p)) and β (s, α (t, p)) are defined for (s,t) ∈ I × J. Then we have that
α (t, β (s, p)) = β (s, α (t, p))
for every (s,t) ∈ I × J.

PROOF : Fix t ∈ J, and consider the curves c i : I → M for i = 1, 2 defined by c 1 (s) =
α (t, β (s, p)) and c 2 (s) = β (s, α (t, p)) for s ∈ I. Then c2 is an integral curve for Y
with initial condition α (t, p), and we will show that this is also the case for c 1 .
For each s ∈ I we have that c ′1 (s) = α t ∗ (Y (β (s, p))) = α t ∗ (Y (q)) where
q = β (s, p) ∈ D t (X). Since LX Y = [X,Y ] = 0, it follows from Proposition 4.90 that
Y (q) = α t∗ Y (q) = α −t ∗ ◦ Y ◦ α t (q) so that c ′1 (s) = Y (α t (q)) = Y (c 1 (s)). Hence c 1
is also an integral curve for Y with initial condition α (t, p), which completes the
proof of the proposition.
4.92 Proposition Let X and Y be vector fields on a smooth manifold M with

global flows α and β , respectively. Let p be a point on M, and suppose there are
open intervals I and J containing 0 such that
α (t, β (s, p)) = β (s, α (t, p))
for (s,t) ∈ I × J. Then we have that [X,Y ](p) = 0.

PROOF : Keeping t ∈ J fixed and differentiating both sides of the equation with re-
spect to s at s = 0, we obtain
α t ∗ (Y (p)) = Y (α t (p)) .
Hence Y (p) = α −t ∗ ◦ Y ◦ α t (p) = α t∗ Y (p) for t ∈ J so that

d
[X,Y ](p) = LX Y (p) = dt α t∗ Y (p) = 0 .

0
DISTRIBUTIONS AND INTEGRAL MANIFOLDS
4.93 Proposition Let X be a vector field on a smooth manifold M n with X(p) 6= 0

for a point p on M. Then there exists a local chart (x,U) around p such that X = ∂ 1
∂x
on U.
PROOF : Let γ : D(X) → M be the global flow for X, and choose an open neighbour-
hood W of p and an open interval I containing 0 such that I × W ⊂ D(X). Let (z,V )
be a local chart around p with V ⊂ W and z(p) = 0.
Then we have that X(p) = [ z, v1 ] p for a vector v1 6= 0 in Rn , and we extend {v1 }
to a basis B = {v1 , ... , vn } for Rn . If E = {e1 , ... , en } is the standard basis for Rn , we
have a linear isomorphism φ : Rn → Rn given by φ (vi ) = ei for i = 1, ... , n. Hence if
(y,V ) is the local chart with y = φ ◦ z, we have that
X(p) = [ z, v1 ] p = [ y, D(y ◦ z−1 )(z(p)) v1 ] p = [ y, φ (v1 ) ] p = [ y, e1 ] p .
Now let O be an open neighbourhood of 0 in Rn−1 with {0} × O ⊂ y(V ), and let
θ : I × O → M be the map defined by
θ (a1 , a2 , ... , an ) = γ (a1 , y−1 (0, a2 , ... , an )) .
Using the local chart (t, I × O) on I × O, where t : I × O → R × Rn−1 is the inclusion

map, we have that
θ∗ ([t, e1 ] 0 ) = X(p) = [ y, e1 ] p
and
θ∗ ([t, ei ] 0 ) = [ y, D (y ◦ θ ◦ t −1 ) (0) ei ] p = [ y, ei ] p
for 2 ≤ i ≤ n, showing that θ∗ 0 is non-singular. By the inverse function theorem, θ
is a local diffeomorphism at 0 with a smooth inverse x : U → Rn defined on an open
neighbourhood U of p. Since
X(θ (a)) = θ∗ ([t, e1 ] a ) = [ x, D (x ◦ θ ◦ t −1 ) (a) e1 ] θ (a) = [ x, e1 ] θ (a)
for a ∈ x(U), it follows that X = ∂ 1 on U.

∂x
4.94 Definition By a distribution ∆ on a smooth manifold M n we mean a sub-

bundle of its tangent bundle π : T M → M. We say that ∆ is involutive or integrable if
for any pair of vector fields X,Y ∈ ∆ defined on some open subset U of M, we have
that [X,Y ] ∈ ∆ .
An immersed submanifold N of M is called an integral manifold of ∆ if
i∗ (Tp N) = ∆ p for every p ∈ N, where i : N → M is the inclusion map. A connected
integral manifold of ∆ is called maximal if it is not contained in any larger connected
integral manifold of ∆.
4.95 Proposition Let ∆ be a distribution on a smooth manifold M, and sup-

pose that there is an integral manifold of ∆ through every point on M. Then ∆ is
integrable.
PROOF : Let X and Y be two vector fields on some open subset U of M lying in
∆ , and let p ∈ U. If N is an integral manifold of ∆ passing through p, it follows
from Proposition 3.14 that there are unique vector fields X ′ and Y ′ on U ∩ N which
TENSORS 109
are i-related to X and Y , respectively, where i : U ∩ N → U is the inclusion map. By

Proposition 4.88 it follows that [ X ′ ,Y ′ ] and [ X,Y ] are i-related so that
i∗ [ X ′ ,Y ′ ] p = [ X,Y ] p .
Since [ X ′ ,Y ′ ] p ∈ Tp N, it follows that [ X,Y ] p ∈ ∆ p for every p ∈ U, thus showing

that [ X,Y ] also lies in ∆ .
4.96 Proposition A distribution ∆ on a smooth manifold M is integrable if and

only if given a local basis {X1 , ... , Xk } for ∆ on an open subset U of M, there are
C∞ -functions crij : U → R such that
k
[ Xi , X j ] = ∑ crij Xr
r=1
for i, j = 1, ... , k.
PROOF : If ∆ is integrable, we know that [ Xi , X j ] belongs to ∆ and hence may be

written as a linear combination of X1 , ... , Xk for i, j = 1, ... , k, with coefficients crij
which are C∞ -functions on U by Corollaries 2.38 and 2.59.
Conversely, suppose that {X1 , ... , Xk } is a local basis for ∆ on U satisfying the
conditions of the proposition, and let X and Y be two vector fields on U belonging to
∆ . Then there are C∞ -functions ai : U → R and b j : U → R such that
k k
X = ∑ ai Xi and Y= ∑ bi X j
i=1 j=1
so that
[ X,Y ] = ∑ [ ai Xi , b j X j ] = ∑ { ai b j ∑ crij Xr + ai Xi (b j ) X j − b j X j (ai ) Xi }

ij ij r
by Proposition 4.87. This shows that [ X,Y ] also belongs to ∆ so that ∆ is integrable.
4.97 Proposition Let ∆ be a k-dimensional distribution on a smooth manifold

M n . Then the following three assertions are equivalent for a local chart (x,U) on M :
(1) The local trivialization (tx , π −1 (U)) associated with (x,U) has the subbundle
property tx (π −1 (U) ∩ ∆ ) = U × Rk × {0} ⊂ U × Rk × Rn−k .
(2) { ∂ 1 , ... , ∂ k } is a local basis for ∆ on U.

∂x ∂x
(3) For every (a0 , b0 ) ∈ x(U) ⊂ Rk × Rn−k , the set x−1 (Rk × {b0}) is an integral
manifold for ∆ called a slice of U, and each slice is a submanifold of M.
PROOF : The equivalence of (1) and (2) follows from Remark 2.82.
To show the equivalence of (2) and (3), let S = x−1 (Rk × {b0}) and let t : Rk ×
Rn−k → Rk × Rn−k be the translation defined by t(a, b) = (a, b − b0). Then the local
chart (u,U) , where u = t ◦ x , has the submanifold property u(U ∩ S) = u(U) ∩ (Rk ×
{0}).
We have a local chart (y, S) on S with coordinate map y = p ◦ x|S , where p :
Rk × Rn−k → Rk is the projection on the first factor. If i : S → M is the inclusion
mapping, we have that
!
∂
n
∂ (xl ◦ i)
i∗ =∑ (q) ∂ l = ∂j
∂yj l=1
∂ y j
∂x ∂x
q q q

∂

∂
for q ∈ S and j = 1, ... , k. This shows that { 1 , ... , k } is a basis for i∗ (Tq S)
∂x ∂x
q q
and completes the proof that (2) and (3) are equivalent.
4.98 Frobenius’ Integrability Theorem (local version) Let ∆ be a k-

dimensional integrable distribution on a smooth manifold M n . Then, for each point
p ∈ M, there is a local chart (x,U) around p satisfying the assertions of Proposition
4.97.
PROOF : We will prove the theorem by induction on k. When k = 1, assertion (2) in
Proposition 4.97 follows immediately from Proposition 4.93.
Now assume that the theorem is true for distributions of dimension k − 1 , and
let {X1 , ... , Xk } be a local basis for ∆ on an open subset V of M containing p. By
Proposition 4.93 there is a local chart (y,W ) around p with W ⊂ V such that
X1 = ∂ 1
∂y
on W , and we may assume that y(W ) = I × O for an open interval I and an open
subset O of Rn−1 . We define a new local basis {Y1 , ... ,Yk } for ∆ on W by
Y1 = X1 ,
Y j = X j − X j (y1 ) X1 for j = 2, ... , k .
Since ∆ is integrable, there are C∞ -functions crij : W → R such that
k
[Yi ,Y j ] = ∑ crij Yr (1)
r=1
for i, j = 1, ... , k, and we have that
Y j (y1 ) = 0 for j = 2, ... , k .

TENSORS 111
By applying both sides of (1) to y1 , we see that this implies

c1i j = 0 for i, j = 1, ... , k .
Now define N = y−1 ({t0 } × O) , where y(p) = (t0 , a0 ) ∈ I × O, and let i : N → W

be the inclusion map. By Proposition 3.14 there is for each j = 1, ... , k − 1 a unique
vector field Z j on N which is i-related to Y j+1 , and it follows from (1) with c1i j = 0
and Proposition 4.96 that the distribution ∆′ on N with local basis {Z1 , ... , Zk−1 } is
integrable. By the induction hypothesis there is therefore a local chart (z,U ′ ) around
p on N such that { ∂ 1 , ... , ∂
} is a local basis for ∆′ on U ′ .
∂z ∂ zk−1
Let pr2 : I × O → O be the projection on the second factor, and let ρ : O →
{t0 } × O be the diffeomorphism given by ρ (a) = (t0 , a). Furthermore, let π : W → N
be the submersion defined by i ◦ π = y−1 ◦ ρ ◦ pr2 ◦ y.
Consider the local chart (x,U) around p on M with coordinate neighbour-
hood U = π −1 (U ′ ) and coordinate map x : U → R × Rn−1 defined by x = { id ×
(z ◦ y−1 ◦ ρ )} ◦ y, i.e.,
1
j y for j = 1
x = j−1
z ◦ π for j = 2, ... , n
so that x ◦ y−1 (t, a) = (t, z ◦ y−1 (t0 , a)). Then we have that x(U) = I × z(U ′ ), and
n j
Y1 = ∂ 1 = ∑ ∂∂ yx1 ∂
= ∂1 on U. (2)
∂y j=1
∂xj ∂x
We want to show that

Y j (xr ) = 0 on U (3)
for j = 1, ... , k and r = k + 1, ... , n which implies that { ∂ 1 , ... , ∂ k } is a local ba-
∂x ∂x
sis for ∆ on U, thus completing the proof of the theorem. The case j = 1 follows
immediately from (2), and by (1) with c1i j = 0 and (2) we have that
k
∂
(Y j (xr )) = Y1 (Y j (xr )) = [Y1 ,Y j ] (xr ) = ∑ c1l j Yl (xr )
∂ x1 l=2
and
Y j (xr ) ◦ i = Z j−1 (xr ◦ i) = Z j−1 (zr−1 ) = 0
for j = 2, ... , k and r = k + 1, ... , n.
Hence for each a ∈ z(U ′ ) and r = k + 1, ... , n, the curve γ : I → Rk−1 defined by
γ j (t) = Y j+1 (xr ) ◦ x−1 (t, a)
for j = 1, ... , k − 1, is an integral curve for the time-independent vector field f :
Rk−1 → Rk−1 given by
k−1
f j (u) = ∑ c1l+1j+1 ul ,
l=1
and we have that γ (t0 ) = 0. The uniqueness part of Proposition 3.23 and Remark 3.35
therefore implies that γ is identically zero, and this completes the proof of (3).
4.99 Proposition Let ∆ be a k-dimensional distribution on a smooth manifold

M n , and let (x,U) be a local chart on M satisfying the assertions of Proposition
4.97. If N is a connected integral manifold of ∆ contained in U, then N is an open
submanifold of a slice of U.
PROOF : Let i : N → M be the inclusion map, and let p : Rk × Rn−k → Rn−k be the
projection on the second factor. For every q ∈ N and every smooth function g defined
on some open neighbourhood of (p ◦ x)(q) in Rn−k , we have that
!
∂

(p ◦ x) ∗ (g) = ∂ i (g ◦ p ◦ x) = Di (g ◦ p)(x(q)) = 0
∂ xi ∂x
q q
for i = 1, ... , k, which implies that (p ◦ x ◦ i) ∗ = 0 . Hence it follows by Proposition

2.29 that p ◦ x ◦ i is constant, thus showing that N is contained in a slice S of U.
Since S is a submanifold of M, it follows from Corollary 2.38 that the inclusion
map i′ : N → S is smooth. As N and S are both integral manifolds of ∆ , we have that
i′∗ q is bijective for each q ∈ N which shows that N is an open submanifold of S.
4.100 Proposition Let ∆ be a k-dimensional integrable distribution on a smooth

manifold M n , and let N1 and N2 be two integral manifolds of ∆ passing through a
point q on M. Then there is an integral manifold N of ∆ passing through q which is
an open submanifold of both N1 and N2 .
PROOF : Let (x,U) be a local chart around q on M satisfying the assertions of Propo-
sition 4.97. Since the inclusion map ir : Nr → M is continuous, Nr ∩ U is an open
subset of Nr for r = 1, 2.
Let Cr be the connected component of Nr ∩ U containing q (in the topology in-
dused by Nr ). Since Nr is locally connected, Cr is an open submanifold of Nr and
hence is a connected integral manifold of ∆ contained in U. By Proposition 4.99,
both C1 and C2 are therefore open submanifolds of the slice S of U passing through
q. Hence N = C1 ∩ C2 is an integral manifold of ∆ satisfying the assertions in the
proposition.
4.101 Corollary Let ∆ be a k-dimensional integrable distribution on a smooth

manifold M n , and let N1 and N2 be two integral manifolds of ∆ . Then N1 ∩ N2 is open
in both N1 and N2 , and N1 and N2 induce the same topology on N1 ∩ N2 .
PROOF : The first assertion follows immediately from the proposition, since for each
q ∈ N1 ∩ N2 there is a set N with q ∈ N ⊂ N1 ∩ N2 which is open in both N1 and N2 .
To prove the last assertion, let T1 and T2 be the topologies on N1 ∩ N2 induced by
N1 and N2 , respectively. If O ∈ T1 , then O is an open submanifold of N1 and hence
is an integral manifold of ∆ . By the first part of the corollary, we therefore have that
O = O ∩ N2 is open in N2 . Hence T1 ⊂ T2 , and the opposite inclusion follows in the
same way.
TENSORS 113
4.102 Proposition Let N be an integral manifold of a k-dimensional integrable

distribution ∆ on M n , and let f : P → M be a smooth map with f (P) ⊂ N. Then f is
also smooth as a map into N.
PROOF : Let q ∈ P, and choose a local chart (x,U) around f (q) on M satisfying
the assertions of Proposition 4.97. Let C1 and C2 be the connected components of
N ∩U containing f (q) in the topologies induced by N and M, respectively. Since the
inclusion map i : N → M is continuous, we have that C1 ⊂ C2 . We will show that we
in fact have C1 = C2 .
Since N is second countable and locally connected, it follows from Corollary 2.5
that N ∩ U has only a countable number of connected components in the topology
induced by N, each contained in a slice of U by Proposition 4.99. If p : Rk × Rn−k →
Rn−k is the projection on the second factor, we therefore have that p ◦ x (C2 ) is a
countable connected subset of Rn−k . Hence p ◦ x (C2 ) must consist of a single point,
showing that C2 is contained in the slice S of U passing through f (q). By Corollary
4.101 we know that N and M induce the same topology on N ∩ S, which implies that
C1 = C2 .
Since P is locally connected, the connected component D of f −1 (U) containing
q is open in P, and we have that f (D) ⊂ C1 which is an open submanifold of S
by Proposition 4.99. Since S is a submanifold of M, it follows from Corollary 2.38
that f |D is smooth considered as a map into C1 , thus showing that f is smooth at q
considered as a map into N.
4.103 Proposition Let X be a topological space which is connected and locally

connected, and suppose there is a countable open cover {Ui |i∈N} of X such that each
connected component of each Ui is second countable. Then X is second countable.
PROOF : We assume that the space X is nonempty, otherwise the proposition is triv-
S
ially true. Let Ci be the family of connected components in Ui , and let C = ∞ i=1 Ci .
Since X is locally connected, the sets in Ci are open and disjoint. Hence it follows
from Corollary 2.5 that given a set V ∈ C , there is only a contable number of sets
V ′ ∈ C with V ∩V ′ 6= 0./
We now inductively define a sequence of families Ai ⊂ C as follows. Fix a
set U ∈ C , and let A 1 = {U}. For each integer i ≥ 2, let A i = {V ∈ C |V ∩ V ′ 6=
0/ for some V ′ ∈ A i−1 }. It follows by induction that each family Ai is countable.
S S
Hence the set Y = ∞ i=1 V ∈Ai V is nonempty, open and second countable. We want
to show that Y is also closed in X. Since X is connected, this implies that X = Y
which completes the proof that X is second countable.
Let p ∈ Y , and choose a set V ∈ C containing p. Since V ∩ Y 6= 0, / there is a
positive integer i and a set V ′ ∈ A i such that V ∩ V ′ 6= 0.
/ Then V ∈ A i+1 which
shows that p ∈ Y and hence that Y is a closed subset of X.
4.104 Frobenius’ Integrability Theorem (global version) Let ∆ be a k-

dimensional integrable distribution on a smooth manifold M n . Then, for each point
p ∈ M, there is a unique maximal connected integral manifold N p of ∆ passing
through p. Every connected integral manifold of ∆ passing through p is an open

submanifold of N p .
PROOF : Let T be the family of all subsets of M which are unions of integral mani-
folds of ∆ . Then T is a new topology on M which is finer than its original topology
T0 . Indeed, T is clearly closed under unions, and 0/ is obtained by taking a union
with an empty indexing set. By induction using Proposition 4.100, we see that T is
also closed under finite intersections, and it contains M by Theorem 4.98.
To see that T is finer than T0 , let U be an arbitrary open set in T0 . For each
point q ∈ U, there is an integral manifold N of ∆ passing through q by Theorem 4.98.
Since the inclusion map i : N → M is continuous, N ∩ U is an open submanifold of
N. Hence N ∩U is an integral manifold of ∆ passing through q and contained in U,
showing that U is open in T which is therefore finer than T0 . In particular, it follows
that T is a Hausdorff topology.
We next show that each integral manifold N of ∆ is a subspace of (M, T ). Each
open set U in N is an integral manifold of ∆ and hence belongs to T . Conversely,
suppose that U is a subset of N belonging to T , and let q be an arbitrary point in U.
Then, by the definition of T , there is an integral manifold N1 of ∆ passing through
q and contained in U. By Proposition 4.100 there is an integral manifold N2 of ∆
passing through q which is an open submanifold of both N and N1 , and this shows U
is open in N.
Since the integral manifolds of ∆ form an open cover of the Hausdorff space
(M, T ) consisting of subspaces which are locally Euclidean, it follows that (M, T )
is also locally Euclidean. The family A consisting of all local charts on every in-
tegral manifold of ∆ , is an atlas on (M, T ). To see that y ◦ x−1 is C∞ for any pair
of local charts (x,U) and (y,V ) in A , let q be a point in U ∩ V and suppose that
(x,U) and (y,V ) are local charts on the integral manifolds N1 and N2 , respectively.
By Proposition 4.100 there is an integral manifold N of ∆ passing through q which is
an open submanifold of both N1 and N2 , and we choose a local chart (z,W ) around q
on N. Then y ◦ x−1 is C∞ at x(q), since y ◦ x−1 |x(U∩V ∩W ) is the composition of the C∞
coordinate transformations z ◦ x−1 and y ◦ z−1 . Hence y ◦ x−1 is C∞ at every point in
x(U ∩V ), thus showing that A is an atlas on M. We let D be the unique differetiable
structure on M containing A .
Now let {Nα | α ∈ A} be the family of connected components in (M, T ). Since
(M, T ) is locally connected, each Nα is open in T , and D |Nα = {(x,U) ∈ D |U ⊂
Nα } is a smooth structure on Nα . We want to show that each Nα is a maximal con-
nected integral manifold of ∆ , which will complete the proof of the theorem. It only
remains to show that Nα is second countable.
By Propositions 4.100 and 2.6 there is a countable atlas {(xi ,Ui ) | i ∈ N} on
(M, T0 ) consisting of local charts satisfying the assertions of Proposition 4.97. For
every positive integer i, it follows by Proposition 4.99 that each connected compo-
nent of Nα ∩Ui (in the topology induced by Nα ) is an open submanifold of a slice of
Ui and is therefore second countable. Hence it follows from Proposition 4.103 that
Nα is second countable.
TENSORS 115
4.105 Remark The partition {Nα | α ∈ A} of M into maximal connected integral

manifolds of ∆ is called a foliation of M, and each maximal connected integral
manifold of ∆ is called a folium or leaf of the foliation.
4.106 Proposition Let N be a maximal connected integral manifold of a k-

dimensional integrable distribution ∆ on M n , and let (x,U) be a local chart on M
satisfying the assertions of Proposition 4.97 with x(U) = O1 × O2 for connected
open sets O1 ⊂ Rk and O2 ⊂ Rn−k . Then the connected components of N ∩U (in the
topology induced by N) are slices of U.
PROOF : By Proposition 4.99 it follows that a connected component C of N ∩ U is
an open submanifold of a slice S of U. Since S is a connected integral manifold of ∆ ,
it follows from Theorem 4.104 that S ⊂ N ∩U so that S = C.
4.107 Proposition Let f : M1 → M2 be a diffeomorphism, and let ∆ 1 and ∆ 2 be

distributions on M1 and M2 , respectively, such that
f∗ (∆ 1 p ) = ∆ 2 f (p)
for every p ∈ M1 . Suppose that N1 is an integral manifold of ∆1 and that N2 = f (N1 )
with inclusion maps i1 : N1 → M1 and i2 : N2 → M2 , and let g : N1 → N2 be the bijec-
tive map induced by f making the diagram
f✲
M1 M2
✻ ✻
i1 i2
g✲
N1 N2
commutative. If N2 is endowed with the manifold structure making g a diffeomor-

phism (see Remark 2.31), then N2 is an integral manifold of ∆ 2 .
PROOF : We have that N2 is an immersed submanifold of M2 such that
i2 ∗ (T f (p) N2 ) = i2 ∗ ◦ g∗ (Tp N1 ) = f∗ ◦ i1 ∗ (Tp N1 ) = f∗ (∆ 1 p ) = ∆ 2 f (p)
for each p ∈ N1 .
4.108 Proposition Let ∆ 1 and ∆ 2 be integrable distributions of dimensions k

and n − k on a smooth manifold M n so that
Tp M = ∆ 1 p ⊕ ∆ 2 p
for every p ∈ M . Then, for each point p ∈ M, there is a local chart (x,U) around
p such that { ∂ 1 , ... , ∂ k } and { ∂
, ... , ∂ n } are local bases for ∆ 1 and ∆ 2 ,
∂x ∂x ∂ xk+1 ∂x
respectively, on U.
PROOF : By Frobenius’ integrability theorem there are local charts (y,V ) and (z,W )
around p such that { ∂ 1 , ... , ∂ k } and { ∂
, ... , ∂ n } are local bases for ∆ 1 and
∂y ∂y ∂ zk+1 ∂z
∆ 2 on V and W , respectively.

Since Tp M = ∆ 1 p ⊕ ∆ 2 p , we have that { ∂ 1 , ... , ∂ k , ∂k+1 , ... , ∂∂zn }
∂y ∂y ∂z
p p p p
is a basis for Tp M . Hence there are real numbers ar j such that

k n
∂ = ∑ ar j ∂ + ∑ ar j ∂
∂yj ∂ yr ∂ zr
p r=1 p r=k+1 p
for j = k +1, ... , n . By applying both sides of this formula to the coordinate functions
yi where i = k + 1, ... , n , we see that
n
∂ yi
δi j = ∑ ar j ∂ zr (p)
r=k+1

∂ yi
for i, j = k + 1, ... , n , which shows that the matrix (p) , where k + 1 ≤ i, j ≤ n ,
∂zj
is non-singular. By the inverse function theorem there is therefore a local chart (x,U)
around p on M with U ⊂ V ∩W and coordinate functions
i
z |U for i = 1, ... , k
xi = .
yi |U for i = k + 1, ... , n
Since
∂ k i
= ∑ ∂ zj ∂ i
∂yj ∂ y ∂x
U i=1
for j = 1, ... , k and

n
∂ ∂ yi ∂
j = ∑
∂ z U i=k+1 ∂ z j ∂ xi
for j = k + 1, ... , n , we see the local chart (x,U) has the desired properties.
Chapter 5
DIFFERENTIAL FORMS
EXTERIOR FORMS ON A VECTOR SPACE

5.1 Definition Let V be a vector space, and let V k denote the cartesian product
V × · · · × V with k factors. A multilinear map T : V k → W into a vector space W
is said to be skew symmetric if it changes sign whenever any pair of variables is
interchanged, i.e., if
T (v1 , ..., vi , ..., v j , ..., vk ) = −T (v1 , ..., v j , ..., vi , ..., vk )
for all v1 , ..., vk ∈ V .
5.2 Remark A multilinear map T : V k → W is skew symmetric if and only if it

is alternating , which means that T (v1 , ..., vk ) = 0 whenever vi = v j for some i 6= j.
Indeed, if T is alternating, we have that
0 = T (v1 , ..., vi + v j , ..., vi + v j , ..., vk )

= T (v1 , ..., vi , ..., vi , ..., vk ) + T (v1 , ..., vi , ..., v j , ..., vk )
+ T (v1 , ..., v j , ..., vi , ..., vk ) + T (v1 , ..., v j , ..., v j , ..., vk )
= 0 + T (v1 , ..., vi , ..., v j , ..., vk ) + T (v1 , ..., v j , ..., vi , ..., vk ) + 0
showing that T is skew symmetric. Conversely, if T is skew symmetric, we have that

T (v1 , ..., v , ..., v , ..., vk ) is equal to its own additive inverse and hence must be 0.
5.3 Definition The skew symmetric tensors in T k (V ) are called exterior forms
of degree k or simply exterior k-forms on V . They form a subspace of T k (V ) which
we denote by Λk (V ). We let Λ0 (V ) = R.
More generally, we let T k (V ;W ) be the space Lk (V, ...,V ;W ) of all multilinear
maps F : V k → W into a vector space W , and Λk (V ;W ) be the subspace consisting
of all skew symmetric multilinear maps into W . The elements in Λk (V ;W ) are called
vector-valued exterior k-forms on V with values in W . We let Λ0 (V ;W ) = W and
T kl (V,W ) denote the space
T (V , ...,V ,W ∗ , ...,W ∗ ) .
| {z } | {z }
k l
117
If B is a basis for V , and C is a basis for W with dual basis C ∗ , then the basis
T (B , ... , B , C ∗ , ... , C ∗ )
| {z } | {z }
k l
for T kl (V,W ) defined in Remark 4.27 is denoted by T kl (B, C ).

If G : U → V is a linear map, then G∗ : T k (V ) → T k (U) maps Λk (V ) into
Λk (U), and we also use the notation G∗ for the induced map G∗ : Λk (V ) → Λk (U) .
More generally, we have a linear map G∗ : T k (V ;W ) → T k (U;W ) defined by
G∗ (F) = F ◦ Gk ,
i.e.,
G∗ (F) (v1 , ..., vk ) = F (G(v1 ), ..., G(vk ))
for v1 , ..., vk ∈ U, which maps Λk (V ;W ) into Λk (U;W ). We also use the notation G∗
for the induced map G∗ : Λk (V ;W ) → Λk (U;W ) , and we have that
λ ◦ G∗ (F) = G∗ (λ ◦ F)
for every functional λ ∈ W ∗ .
5.4 Proposition A multilinear map T : V k → W is skew symmetric if and only if
T σ = ε (σ ) T (1)
for every permutation σ ∈ Sk , where ε (σ ) denotes the sign of σ .

PROOF : By Definition 5.1 we know that T is skew symmetric if and only if
T τ = ε (τ ) T
for every transposition τ , and this is certainly the case if (1) is satisfied for every
permutation σ ∈ Sk .
Suppose conversely that T is skew symmetric, and let σ be an arbitrary permuta-
tion in Sk . We know by Proposition 13.7 in the appendix that σ may be expressed as
a product of transpositions
σ = τ1 · · · τm .
Then (1) follows by induction on the number m of factors in σ . Assuming it is true
for σ ′ = τ1 · · · τm−1 , we have that
′ ′ ′
T σ = T σ τm = (T τm )σ = ε (τm ) T σ = ε (τm ) ε (σ ′ ) T = ε (σ ) T

DIFFERENTIAL FORMS 119
5.5 Corollary A tensor T ∈ T k (V ) belongs to Λk (V ) if and only if
T σ = ε (σ ) T (1)
for every permutation σ ∈ Sk .
5.6 Now consider the linear map A : T k (V ;W ) → T k (V ;W ) defined by

1
ε (σ ) T σ
k! σ∑
A (T ) =
∈S k
i.e.,
1
k! σ∑
A (T )(v1 , ..., vk ) = ε (σ ) T (vσ (1) , ..., vσ (k) )
∈S k
which is called the alternation operator.
5.7 Proposition If T ∈ T k (V ;W ) and σ ∈ Sk , then A (T σ ) = A (T ) σ =

ε (σ ) A (T ) .
1 1
A (T σ ) = ∑ ε (τ ) (T σ ) τ = ε (σ ) ε (τσ ) T τσ = ε (σ ) A (T )
k! τ ∈S k! τ∑
∈S
k k
and
1 1
A (T ) σ = ε (τ ) (T τ ) σ = ε (σ ) ∑ ε (σ τ ) T σ τ = ε (σ ) A (T )
k! τ∑
∈S k! τ ∈S
k k
since the maps Rσ : Sk → Sk and Lσ : Sk → Sk given by Rσ (τ ) = τσ and Lσ (τ ) = σ τ

are bijections and ε is a homomorphism.
5.8 Proposition
(1) If T ∈ T k (V ;W ), then A (T ) ∈ Λk (V ;W ) .
(2) If ω ∈ Λk (V ;W ), then A (ω ) = ω .
(3) If T ∈ T k (V ;W ), then A (A (T )) = A (T ) .
PROOF : (1) Follows from Proposition 5.4 and 5.7.
(2) If ω ∈ Λk (V ;W ), then
1 1
∑ ε (σ ) ω σ =
k! σ∑
A (ω ) = ε (σ ) ε (σ ) ω = ω
k! σ ∈S ∈S
k k
by Proposition 5.4.
(3) Follows from (1) and (2).
5.9 Proposition If F : U → V is a linear map, and if AU and AV are the alterna-

tions on T k (U;W ) and T k (V ;W ), respectively, then
F ∗ ◦ AV = AU ◦ F ∗ .
PROOF : Since F ∗(T σ ) = F ∗(T ) σ for every T ∈ T k (V ;W ) and σ ∈ Sk by 13.9 in the

appendix, the proposition follows from the linearity of F ∗ : T k (V ;W ) → T k (U;W ).
5.10 Proposition If T ∈ T k (V ) and S ∈ T l (V ), then

(1) A (T ⊗ S) = (−1)kl A (S ⊗ T ) ,
(2) A (A (T ) ⊗ S) = A (T ⊗ S) = A (T ⊗ A (S)) .
PROOF : (1) Let τ ∈ Sk+l be the permutation defined by

r + k for 1 ≤ r ≤ l
τ (r) = .
r − l for l + 1 ≤ r ≤ l + k
Since τ may be obtained by moving each of the last k numbers {l + 1 , ..., l + k} in

succession l positions forward so that they become the first k numbers, we see that τ
may be written as a product of kl transpositions and therefore has sign ε (τ ) = (−1)kl .
We now have that
T ⊗ S (v1 , ..., vk , vk+1 , ..., vk+l ) = S ⊗ T (vk+1 , ..., vk+l , v1 , ..., vk )

= S ⊗ T (vτ (1) , ..., vτ (l) , vτ (l+1) , ..., vτ (l+k) ) .
Hence T ⊗ S = (S ⊗ T )τ which by Proposition 5.7 implies that
A (T ⊗ S) = (−1)kl A (S ⊗ T ) .
(2) We have an injective homomorphism φ : Sk → Sk+l given by

σ (r) for 1 ≤ r ≤ k
φ (σ )(r) = ,
r for k + 1 ≤ r ≤ k + l
i.e., φ (σ ) acts as σ on {1 , ..., k} and leaves {k + 1 , ..., k + l} fixed. Then we have

that T σ ⊗ S = (T ⊗ S) φ (σ ) and ε (φ (σ )) = ε (σ ) since φ maps transpositions in Sk to
transpositions in Sk+l and hence is sign preserving. By Proposition 5.7 we have that
A (T σ ⊗ S) = ε (σ ) A (T ⊗ S), and from this it follows that
1 1
ε (σ ) A (T σ ⊗ S) =
k! σ∑ ∑ A (T ⊗ S) = A (T ⊗ S)
A (A (T ) ⊗ S) =
∈S k! σ ∈S
k k
since A is linear and ⊗ is bilinear.

The last equality in (2) is proved in the same way using the injective homomor-
phism ψ : Sl → Sk+l given by

r for 1 ≤ r ≤ k
ψ (σ )(r) = .
σ (r − k) + k for k + 1 ≤ r ≤ k + l
Alternatively the last equality follows from the first by (1) as follows
A (T ⊗ A (S)) = (−1)kl A (A (S) ⊗ T) = (−1)kl A (S ⊗ T ) = A (T ⊗ S) .
5.11 Definition If ω ∈ Λk (V ) and η ∈ Λl (V ), we define their wedge product or

exterior product ω ∧ η ∈ Λk+l (V ) by
(k + l)!
ω ∧η = A (ω ⊗ η ) .
k! l!
5.12 Proposition
(1) ∧ is bilinear.
(2) If F : V → W is a linear map, then F ∗ (ω ∧ η ) = F ∗ (ω ) ∧ F ∗ (η ) for every
exterior form ω and η on W .
(3) If ω ∈ Λk (V ) and η ∈ Λl (V ), then ω ∧ η = (−1)kl η ∧ ω .
(4) If ω ∈ Λk (V ), η ∈ Λl (V ) and θ ∈ Λm (V ), then
(k + l + m)!
(ω ∧ η ) ∧ θ = ω ∧ (η ∧ θ ) = A (ω ⊗ η ⊗ θ ) .
k! l! m!
PROOF : (1) Follows from the fact that ⊗ is bilinear and the alternation A is linear.
(2) Follows from the corresponding relation for ⊗ and Proposition 5.9.
(3) Follows from Proposition 5.10 (1).
(4) By Proposition 5.10 (2) we have that
(k + l + m)! (k + l)! (k + l + m)!
(ω ∧ η ) ∧ θ = A ( A (ω ⊗ η ) ⊗ θ ) = A (ω ⊗ η ⊗ θ ) ,
(k + l)! m! k! l! k! l! m!
and a similar calculation gives

(k + l + m)!
ω ∧ (η ∧ θ ) = A (ω ⊗ η ⊗ θ ) .
k! l! m!
5.13 Proposition
(1) If ω1 , ..., ωk ∈ V ∗ , then
ω1 ∧ · · · ∧ ωk = k! A (ω1 ⊗ · · · ⊗ ωk ) = ∑ ε (σ ) ωσ (1) ⊗ · · · ⊗ ωσ (k) .

σ ∈Sk
(2) The map φ : V ∗ k → Λk (V ) defined by φ (ω1 , ..., ωk ) = ω1 ∧ · · · ∧ ωk is multilin-

ear and alternating.
PROOF : (1) The first equality is proved by induction on k. It is clearly true when
k = 1. Assuming it is true for k − 1, we have that
k!
ω1 ∧ · · · ∧ ωk = A ((ω1 ∧ · · · ∧ ωk−1 ) ⊗ ωk ) = k! A (ω1 ⊗ · · · ⊗ ωk )
(k − 1)! 1!
by Proposition 5.10 (2).

The last equality follows from the definition of A using that
−1
(∗) ωσ (1) ⊗ · · · ⊗ ωσ (k) = ( ω1 ⊗ · · · ⊗ ωk ) σ
and ε (σ ) = ε (σ −1 ) for σ ∈ Sk , and that the map I : Sk → Sk given by I(σ ) = σ −1 is

a bijection.
(2) φ is multilinear by Proposition 5.12 (1). By applying A to both sides of (∗) and
using Proposition 5.7 and the first equality in (1) we have that
ωσ (1) ∧ · · · ∧ ωσ (k) = ε (σ ) ω1 ∧ · · · ∧ ωk .
Hence φ σ = ε (σ ) φ which shows that φ is alternating.
5.14 Proposition If B = {v1 , ..., vn } is a basis for the vector space V , then
D = {v∗i1 ∧ · · · ∧ v∗ik | 1 ≤ i1 < ... < ik ≤ n}
is a basis for Λk (V ) when k > 0. If ω ∈ Λk (V ) we have that
ω= ∑ ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik .

i1 <...<ik
In particular dim Λk (V ) = (nk ) when 0 ≤ k ≤ n, and Λk (V ) = {0} for k > n.

ω= ∑ ω (vi1 , ..., vik ) v∗i1 ⊗ · · · ⊗ v∗ik ,

i1 ,...,ik
and by applying A to both sides it follows by Proposition 5.8 (2) and 5.13 (1) that
1
ω= ∑ ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik .
k! i1 ,...,ik
As ω is alternating, we have that the coefficient ω (vi1 , ..., vik ) is 0 if the indices
i1 , ..., ik are not all different. If they are different and σ ∈ Sk , we have that
ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik = ω (viσ (1) , ..., viσ (k) ) v∗iσ (1) ∧ · · · ∧ v∗iσ (k)
since both ω and the map φ from Proposition 5.13 (2) are alternating. Since there are
k! such rearrangements of the indices i1 , ..., ik , we have that
ω= ∑ ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik

i1 <...<ik
which shows the last assertion of the proposition and also that D spans Λk (V ). To
show that it is linearly independent, assume that
∑ ai1 ,...,ik v∗i1 ∧ · · · ∧ v∗ik = 0 .

i1 <...<ik
By applying both sides to (v j1 , ..., v jk ) where 1 ≤ j1 < ... < jk ≤ n, we see that
a j1 ,..., jk = 0.
5.15 Let V and W be vector spaces, and let C = {w1 , ..., wm } be a basis for W
with dual basis C ∗ = {w∗1 , ..., w∗m }. Then we have a linear isomorphism
φ : T k (V ;W ) → T k1 (V,W )
defined by
φ (F)(v1 , ..., vk , λ ) = λ (F(v1 , ..., vk ))
for F ∈ T (V ;W ), v1 , ..., vk ∈ V and λ ∈ W ∗ , with inverse
k
ψ : T k1 (V,W ) → T k (V ;W )
given by
m
ψ (T )(v1 , ..., vk ) = ∑ T (v1 , ..., vk , w∗i ) wi
i=1
k
for T ∈ T 1 (V,W ) and v1 , ..., vk ∈ V . Indeed, we have that
φ (ψ (T ))(v1 , ..., vk , w∗j ) = w∗j (ψ (T )(v1 , ..., vk )) = T (v1 , ..., vk , w∗j )
for j = 1, ..., m and

m
ψ (φ (F))(v1 , ..., vk ) = ∑ φ (F)(v1 , ..., vk , w∗i ) wi
i=1
m
= ∑ w∗i (F(v1 , ..., vk )) wi = F(v1 , ..., vk )
i=1
which shows that ψ = φ −1 . In particular, it follows that ψ does not depend on the
choice of the basis C .
If the space V is also finite dimensional, then it follows from Remarks 4.5 and
4.31 that we have a linear isomorphism
µ∗ : T k (V ) ⊗ W → T k1 (V,W )
induced by the bilinear map
µ : T k (V ) × W → T k1 (V,W )
defined by
µ (T, w) = T ⊗ w
k
for T ∈ T (V ) and w ∈ W , i.e.,
µ (T, w)(v1 , ..., vk , λ ) = T (v1 , ..., vk ) λ (w)
for v1 , ..., vk ∈ V and λ ∈ W ∗ . The composed map
ψ ◦ µ : T k (V ) × W → T k (V ;W )
is given by
ψ ◦ µ (T, w)(v1 , ..., vk ) = T (v1 , ..., vk ) w
k
for T ∈ T (V ), w ∈ W and v1 , ..., vk ∈ V .
Hence if A is the alternation operator, we have that
ψ ◦ µ∗ ◦ (A , id )∗ = A ◦ ψ ◦ µ∗
since
ψ ◦ µ∗ ◦ (A , id )∗ (T ⊗ w)(v1 , ..., vk ) = ψ ◦ µ (A (T ), w)(v1 , ..., vk )
1
k! σ∑
= A (T )(v1 , ..., vk ) w = ε (σ ) T (vσ (1) , ..., vσ (k) ) w
∈S k
1
k! σ∑
= ε (σ ) ψ ◦ µ (T, w)(vσ (1) , ..., vσ (k) )
∈S k
= A (ψ ◦ µ (T, w))(v1 , ..., vk ) = A ◦ ψ ◦ µ∗ (T ⊗ w)(v1 , ..., vk ) ,
and so we have a commutative diagram

ψ ◦ µ∗ ✲
T k (V ) ⊗ W T k (V ;W )
(A , id )∗ A
❄ ❄
T k (V ) ⊗ W ✲ T k (V ;W )
ψ ◦ µ∗
By Proposition 5.8 it follows that the linear isomorphism ψ ◦ µ∗ maps the subspace
Λ k (V ) ⊗ W of T k (V ) ⊗ W onto Λ k (V ;W ) . We let Λ k1 (V,W ) denote the image
µ∗ (Λk (V ) ⊗W ) consisting of all tensors in T k1 (V,W ) which are skew symmetric in
the first k variables.
5.16 Proposition If B = {v1 , ..., vn } and C = {w1 , ..., wm } are bases for the
vector spaces V and W , respectively, then
D = {(v∗i1 ∧ · · · ∧ v∗ik ) ⊗ w j | 1 ≤ i1 < ... < ik ≤ n, 1 ≤ j ≤ m}
is a basis for Λ k1 (V,W ) when k > 0. If ω ∈ Λ k1 (V,W ) we have that
ω= ∑ ω (vi1 , ..., vik , w∗j ) (v∗i1 ∧ · · · ∧ v∗ik ) ⊗ w j .

i1 <...<ik , j
In particular dim Λ k1 (V,W ) = m (nk ) when 0 ≤ k ≤ n, and Λ k1 (V,W ) = {0} for k > n.
PROOF : By 5.15 and Proposition 5.14 it follows that D is a basis for Λ k1 (V,W ) when
k > 0, as well as the last part of the proposition. If
j
ω= ∑ a i1 , ...,ik (v∗i1 ∧ · · · ∧ v∗ik ) ⊗ w j ,
i1 <...<ik , j
then
ω (vr1 , ..., vrk , w∗s ) = a sr1 , ...,rk
for 1 ≤ r1 < ... < rk ≤ n and 1 ≤ j ≤ m which yields the formula for ω .
5.17 Proposition Let {v1 , ... , vn } be a basis for the vector space V , and let ω ∈
Λn (V ). If w j = ∑ni=1 ai j vi for j = 1, ... , n, we have that
ω (w1 , ... , wn ) = det(ai j ) ω (v1 , ... , vn )
PROOF : Let η ∈ T n (Rn ) be the covariant tensor on Rn defined by

n n
η ((a11 , ... , an1 ), ... , (a1n , ... , ann )) = ω ( ∑ ai1 vi , ... , ∑ ain vi )
i=1 i=1
Then η ∈ Λn (Rn ) so that η = c det for some c ∈ R, and we have that
c = η (e1 , ... , en ) = ω (v1 , ... , vn ) .
5.18 Corollary If {v1 , ... , vn } and {w1 , ... , wn } are two bases for a vector space
V with w j = ∑ni=1 ai j vi for j = 1, ... , n, then we have that
v∗1 ∧ · · · ∧ v∗n = det(ai j ) w∗1 ∧ · · · ∧ w∗n
PROOF : Using Propositions 5.14 and 5.17 we have that
v∗1 ∧ · · · ∧ v∗n = v∗1 ∧ · · · ∧ v∗n (w1 , ...wn ) w∗1 ∧ · · · ∧ w∗n = det(ai j ) w∗1 ∧ · · · ∧ w∗n .
DIFFERENTIAL FORMS ON A MANIFOLD
5.19 Now let π : E → M be an n-dimensional vector bundle over the smooth

manifold M, and let E p = π −1 (p) be the fibre over p for each p ∈ M. We want to
define a new bundle [
Λk (E) = Λk (E p )
p∈M
of exterior k-forms on each fibre E p . We have a projection π ′ : Λk (E) → M defined

by π ′ (ω ) = p for ω ∈ Λk (E p ).
5.20 Proposition π ′ : Λk (E) → M is a subbundle of dimension (nk ) of the vector

bundle π k : T k (E) → M.
PROOF : If (t, π −1 (U)) is a local trivialization for E, it follows from Propositions
4.45 and 5.13 (1) that there are sections si1 ,...,ik : U → T k (E) defined by
si1 ,...,ik (p) = (ei1 ◦ t p ) ∧ · · · ∧ (eik ◦ t p )
for p ∈ U and 1 ≤ i1 < ... < ik ≤ n, and {si1 ,...,ik (p) | 1 ≤ i1 < ... < ik ≤ n} is a basis
for Λk (E p ) by Propositions 4.11 and 5.14. The result hence follows from Proposition
2.62.
5.21 Definition Let π : E → M be an n-dimensional vector bundle over the

smooth manifold M. If ω and η are sections of the bundles Λk (E) and Λl (E),
respectively, on an open subset V of M, we define their wedge product ω ∧ η to

be the section of Λk+l (E) on V defined by
(ω ∧ η )(p) = ω (p) ∧ η (p)
for p ∈ V . If l = 0, then η is simply a smooth function f on V , and f ∧ ω and ω ∧ f
simply mean f ω .
A section of the bundle π : Λk (T M) → M defined on an open subset V of M is
called a differential form of degree k or simply a k-form on V . The set of all k-forms
on V is denoted by Ωk (V ). In particular Ω0 (V ) = F (V ) is the set of C∞ -functions on
V . We let Ω(V ) be the direct sum of Ωk (V ) for k = 0, ..., n.
5.22 Corollary If M is a smooth manifold and ω : M → Λ k (T M) is a map with

π ◦ ω = idM , then ω is a k-form on M if and only if the functions ωi1 ,...,ik : U → R
defined by
ω (p) = ∑ ωi1 ,...,ik (p) dxi1 (p) ∧ · · · ∧ dxik (p)
i1 <...<ik
for p ∈ U are smooth for every local chart (x,U) on M .

PROOF : Follows from Propositions 4.58 and 5.13 (1).
5.23 If π : E → M and πe : Ee → M are vector bundles over the smooth manifold

M of dimensions n and m, respectively, we let
e = T ( E , ..., E , Ee ∗ )
T k1 (E, E)
| {z }
k
and denote its projection by π1k . We want to define a subbundle

[
e =
Λ k1 (E, E) Λ k1 (E p , Eep )
p∈M
of tensors which are skew symmetric in the first k variables on each fibre E p . We
e → M defined by π ′′ (ω ) = p for ω ∈ Λ k1 (E p , Eep ).
have a projection π ′′ : Λ k1 (E, E)
5.24 Proposition e → M is a subbundle of dimension m (n ) of the

π ′′ : Λ k1 (E, E) k
k k e
vector bundle π1 : T 1 (E, E) → M.
PROOF : If (t, π −1 (U)) and (e e are local trivializations for E and E,

t , πe−1 (U)) e it fol-
lows from Propositions 4.11, 4.45 and 5.13 (1) that there are sections si1 ,...,ik , j :
U∩U e → T k1 (E, E)
e defined by
t p−1 (e j )
si1 ,...,ik , j (p) = ((ei1 ◦ t p ) ∧ · · · ∧ (eik ◦ t p )) ⊗ e
for p ∈ U ∩ U,e 1 ≤ i1 < ... < ik ≤ n and 1 ≤ j ≤ m, and {si ,...,i , j (p) | 1 ≤ i1 < ... <
1 k
ik ≤ n, 1 ≤ j ≤ m} is a basis for Λ k1 (E p , Eep ) by Propositions 4.11 and 5.16. The result

hence follows from Propositions 2.62.
5.25 We now define the bundle

[
e =
T k (E ; E) T k (E p ; Eep )
p∈M
with Eep -valued covariant tensors of degree k in each fibre E p . We have a projection
e → M given by π k (ω ) = p for ω ∈ T k (E p ; Eep ), and a map
π k : T k (E ; E)
e → T k1 (E, E)
φ : T k (E ; E) e
defined by
φ (F)(v1 , ..., vk , λ ) = λ (F(v1 , ..., vk ))
for F ∈ T k (E p ; Eep ), v1 , ..., vk ∈ E p and λ ∈ Eep∗ . From 5.15 we know that φ is a linear
isomorphism on each fibre with π1k ◦ φ = π k , and we give T k (E ; E) e the manifold
structure such that φ is an equivalence over M. We also define the bundle
[
e =
Λ k (E ; E) Λ k (E p ; Eep )
p∈M
with Eep -valued exterior k-forms in each fibre E p , which is a subbundle of T k (E ; E)

e
since
e = φ −1 (Λ k1 (E, E))
Λ k (E ; E) e .
e → M induced by π k is given by π ′ (ω ) = p for ω ∈
The projection π ′ : Λ k (E ; E)
k e
Λ (E p ; E p ).
5.26 Let π : E → M be an n-dimensional vector bundle over the smooth manifold

M. If W is a vector space of dimension m, we let πe : EW → M be the vector bundle
EW = M × W where πe is the projection on the first factor. If C = {w1 , ..., wm } is a
t , πe−1 (M)) for EW given by
basis for W , we have a local trivialization (e
m
e
t (p, w) = (p, a) for p ∈ M and w = ∑ aj wj
j=1
so that the projection ρ p : {p} ×W → W on the second factor is a linear isomorphism

for each fibre {p} × W. From this we also obtain the linear isomorphisms
ρ ′′p : Λ k1 (E p , {p} × W ) → Λ k1 (E p ,W )
induced by
(id, ... , id, ρ ∗p ) ∗ : T k1 (E p , {p} × W ) → T k1 (E p ,W )
and
ρ ′p : Λ k (E p ; {p} × W ) → Λ k (E p ;W )
given by
ρ ′p (F) = ρ p ◦ F
for F ∈ Λ k (E p ; {p} × W ). If
φep : Λ k (E p ; {p} × W ) → Λ k1 (E p , {p} × W )
is the linear isomorphism defined as in 5.25 and
φ p : Λ k (E p ;W ) → Λ k1 (E p ,W )
is the corresponding linear isomorphism defined by
φ p (G)(v1 , ..., vk , λ ) = λ (G(v1 , ..., vk ))
for G ∈ Λ k (E p ;W ), v1 , ..., vk ∈ E p and λ ∈ W ∗ , we have a commutative diagram
φep
Λ k (E p ; {p} × W ) ✲ Λ k (E , {p} × W )
1 p
ρ ′p ρ ′′p
❄ ❄
Λ k (E p ;W ) ✲ Λ k1 (E p ,W )
φp
since
φ p ◦ ρ ′p (F)(v1 , ... , vk , λ ) = λ ( ρ ′p (F)(v1 , ... , vk )) = λ ◦ ρ p ◦ F (v1 , ... , vk )
= φep (F)(v1 , ... , vk , ρ ∗p (λ )) = ρ ′′p ◦ φep (F)(v1 , ... , vk , λ )
for F ∈ Λ k (E p ; {p} × W ), v1 , ..., vk ∈ E p and λ ∈ W ∗ .

Using the vector spaces Λ k (E p ;W ) and Λ k1 (E p ,W ) where p ∈ M as fibres, we
can now construct the vector bundles
π ′ : Λ k (E ;W ) → M and π ′′ : Λ k1 (E,W ) → M
such that all the maps in the corresponding commutative diagram
φe
Λ k (E ; EW ) ✲ Λ k1 (E , EW )
ρ′ ρ ′′
❄ ❄
k
Λ (E ;W ) ✲ Λ k1 (E ,W )
φ
are equivalences over M.

5.27 Let π : E → M be an n-dimensional vector bundle over the smooth manifold

M. If W is a vector space of dimension m, we let πe : EW → M be the vector bundle
EW = M × W where πe is the projection on the first factor. If C = {w1 , ..., wm } is a
t , πe−1 (M)) for EW given by
basis for W , we have a local trivialization (e
m
e
t (p, w) = (p, a) for p ∈ M and w = ∑ aj wj
j=1
so that the projection ρ p : {p} ×W → W on the second factor is a linear isomorphism

for each fibre {p} × W. From this we also obtain the linear isomorphisms
ρ ′′p : Λ 11 (W , E p ) → Λ 11 ({p} × W , E p )
induced by
(ρ p , id ) ∗ : T 11 (W , E p ) → T 11 ({p} × W , E p )
and
ρ ′p : Λ 1 (W ; E p ) → Λ 1 ({p} × W ; E p )
given by
ρ ′p (F) = F ◦ ρ p
for F ∈ Λ 1 (W ; E p ). If
φep : Λ 1 ({p} × W ; E p ) → Λ 11 ({p} × W , E p )
is the linear isomorphism defined as in 5.25 and
φ p : Λ 1 (W ; E p ) → Λ 11 (W , E p )
is the corresponding linear isomorphism defined by
φ p (G)(v, λ ) = λ (G(v))
for G ∈ Λ 1 (W ; E p ), v ∈ W and λ ∈ E p∗ , we have a commutative diagram
φp
Λ 1 (W ; E p ) ✲ Λ 11 (W , E p )
ρ ′p ρ ′′p
❄ ❄
1
Λ ({p} × W ; E p ) ✲ Λ 1 ({p} × W , E )
1 p
φep
since
φep ◦ ρ ′p (F)(v, λ ) = λ ( ρ ′p (F)(v)) = λ ◦ F ◦ ρ p (v)

= φ p (F)(ρ p (v), λ ) = ρ ′′p ◦ φ p (F)(v, λ )
for F ∈ Λ 1 (W ; E p ), v ∈ {p} × W and λ ∈ E p∗ .

Using the vector spaces Λ 1 (W ; E p ) and Λ 11 (W , E p ) where p ∈ M as fibres, we
can now construct the vector bundles
π ′ : Λ 1 (W ; E ) → M and π ′′ : Λ 11 (W, E ) → M
such that all the maps in the corresponding commutative diagram
φ
Λ 1 (W ; E ) ✲ Λ 11 (W , E )
ρ′ ρ ′′
❄ ❄
1
Λ (EW ; E ) ✲ Λ 11 (EW , E )
φe
are equivalences over M.
5.28 Proposition Let π : E → M and πe : Ee → M be vector bundles over the

smooth manifold M of dimensions n and m, respectively, and let E = {e1 , ..., en }
and F = { f1 , ..., fm } be the standard bases for Rn and Rm with dual bases E ∗ =
{e1 , ..., en } and F ∗ = { f 1 , ..., f m }. If (t, π −1 (U)) and (e
t , πe −1 (U)) are local trivial-
izations in E and E, e then we have a local trivialization ((id U × β ) ◦ τ , (π k ) −1 (U))
in the vector bundle π k : T k (E ; E) e → M, where τ : (π k ) −1 (U) → U × T k (Rn ; Rm )
is the equivalence over U defined by
τ (F) = (p, et p ◦ F ◦ (t p−1 ) k )
for p ∈ U and F ∈ T k (E p ; Eep ) , and β : T k (Rn ; Rm ) → Rs is the linear isomorphism

given by
β b(i 1 , ... , i k , j) (G) = f j ◦ G(e i 1 , ... , e i k )
for G ∈ T k (Rn ; Rm ) and (i1 , ..., ik , j) ∈ Ink ×Im , where b : Ink ×Im → Is is the bijection
defined in Remark 4.27 with s = nk m. If
e → T k1 (E, E)
φ : T k (E ; E) e
is the equivalence over M defined in 5.25 and (t ′ , π ′′ −1 (U)) is the local trivialization
e defined in Remark 4.40, then we have that
in T k1 (E, E)
(id U × β ) ◦ τ = t ′ ◦ φ .
PROOF : Let (t ′′ , πe ∗ −1 (U)) be the local trivialization in the dual bundle πe ∗ : Ee ∗ →

t , πe −1 (U)), and let G = {g1, ..., g s } be the standard basis for

M corresponding to (e
R . Then we have that
s
(t p−1 , ... ,t p−1 , t p′′ −1 ) ∗ ◦ φ (F)(e i 1 , ... , e i k , f j )
= φ (F)(t p−1 (e i 1 ), ... , t p−1 (e i k ), f j ◦ e

tp)
t p ◦ F ◦ (t p−1 ) k (e i 1 , ... , e i k )
= fj◦e
for (i1 , ..., ik , j) ∈ Ink × Im , so that
(t p−1 , ... ,t p−1 , t p′′ −1 ) ∗ ◦ φ (F)
= ∑ t p ◦ F ◦ (t p−1 ) k (e i 1 , ... , e i k ) e i1 ⊗ · · · ⊗ e ik ⊗ f j
fj◦e
i1 ,...,ik , j
for p ∈ U. Hence it follows that
t p′ ◦ φ (F) = ∑ t p ◦ F ◦ (t p−1 ) k (e i 1 , ... , e i k ) g b(i 1 , ... , i k , j) = β ◦ τ (F)

fj◦e
i1 ,...,ik , j
for p ∈ U and F ∈ T k (E p ; Eep ) , which completes the proof of the proposition.
5.29 Corollary Let π : E → M be a vector bundle over the smooth manifold M

of dimension m, and let W be an n-dimensional vector space. Let E = {e1 , ..., en }
and F = { f1 , ..., fm } be the standard bases for Rn and Rm with dual bases E ∗ =
{e1 , ..., en } and F ∗ = { f 1 , ..., f m }. If (t, π −1(U)) is a local trivialization in E and
x : W → Rn is a linear isomorphism, then we have a local trivialization ((id U ×
β ) ◦ τ , π ′ −1 (U)) in the vector bundle π ′ : Λ 1 (W ; E) → M, where τ : π ′ −1 (U) →
U × L(Rn , Rm ) is the equivalence over U defined by
τ (F) = (p, t p ◦ F ◦ x−1 )
for p ∈ U and F ∈ Λ 1 (W ; E p ), and β : L(Rn , Rm ) → R n m is the linear isomorphism

given by
β b(i, j) (G) = f j ◦ G(ei )
for G ∈ L(Rn , Rm ), i = 1, ... , n and j = 1, ... , m, where b : In × Im → I n m is the
bijection defined in Remark 4.27. If
ρ ′ : Λ 1 (W ; E ) → Λ 1 (EW ; E ) and φe : Λ 1 (EW ; E ) → Λ 11 (EW , E )
are the equivalences over M defined in 5.27, using the basis C = {x−1 (e1 ), ..., x−1 (en )}
for W , and if (t ′ , π ′′′ −1 (U)) is the local trivialization in Λ 11 (EW , E) = T (EW , E ∗ ) de-
fined in Remark 4.40, then we have that
(id U × β ) ◦ τ = t ′ ◦ φe ◦ ρ ′ .
PROOF : Let τe : π ′′ −1 (U) → U × L(Rn , Rm ) be the equivalence over U defined as in

Proposition 5.28 for the vector bundle π ′′ : Λ 1 (EW ; E) → M , using the local trivial-
t , πe−1 (M)) for EW given in 5.27. Then we have that e
ization (e t p ◦ ρ p−1 = x for p ∈ M ,
which shows that
τe ◦ ρ ′ (F) = (p, t p ◦ F ◦ ρ p ◦ et p−1 ) = (p, t p ◦ F ◦ x−1 ) = τ (F)
for p ∈ U and F ∈ Λ 1 (W ; E p ) . The corollary now follows from Proposition 5.28.

smooth manifold M of dimensions n and m, and let g : E → Ee be a bundle map over
e → M, called
M. Then we have a section s on M in the vector bundle π ′ : Λ 1 (E ; E)
the section determined by the bundle map g, defined by
s(p) = g p
for p ∈ M. Conversely, each section s of π ′ on M is determined by a unique bundle
map g given by
g(v) = s(p)(v)
for p ∈ M and v ∈ E p .
PROOF : Let ((id U × β ) ◦ τ , (π k ) −1 (U)) be the local trivialization in the vector
bundle π′ e → M given in Proposition 5.28, obtained from local trivial-
: Λ 1 (E ; E)
−1
izations (t, π (U)) and (e t , πe −1 (U)) in E and E,e and let pr2 : U × Rm → Rm and
′
pr2 : U × L(R , R ) → L(R , R ) be the projections on the second factor. Then we
n m n m
see that s is smooth on U since

β b(i, j)
◦ pr2′ ◦ τ ◦ s(p) = f j ◦ pr2 ◦ e
t ◦ g ◦ t −1 (p, ei )
for p ∈ U, i = 1, ... , n and j = 1, ... , m, showing the first part of the proposition.
To prove the last part, let s be a section of π ′ on M, and let B : L(Rn , Rm ) × Rn →
R be the bilinear map given by B(G, a) = G(a) for G ∈ L(Rn , Rm ) and a ∈ Rn . Then
m
we see that g is smooth on π −1 (U) since

t ◦ g ◦ t −1 = B ◦ {( pr2′ ◦ τ ◦ s) × idRn } .
pr2 ◦ e

smooth manifold M of dimensions n and m. Then we have an equivalence
e →T
ψ : T k (E ; Λ 1(E ; E)) k+1 e
(E ; E)
e → M and π k+1 :
over M between the vector bundles πe k : T k (E ; Λ 1(E ; E))
T k+1 e
(E ; E) → M which is given by
ψ (F)(v1 , ... , vk+1 ) = F(v1 , ... , vk )(vk+1 )
for F ∈ T (E p ; Λ (E p ; Eep )) and v1 , ... , vk+1 ∈ E p where p ∈ M.
k 1
PROOF : Consider first the linear isomorphism φ : T k (Rn ; L(Rn , Rm )) → T k+1

(Rn ;
Rm ) given by
φ (G)(a1 , ... , ak+1 ) = G(a1 , ... , ak )(ak+1 )
for G ∈ T k (Rn ; L(Rn , Rm )) and a1 , ... , ak+1 ∈ Rn . Let r = nm and s = nk+1 m , and
let E = {e1 , ..., en } , F = { f1 , ..., fm } and G = {g1 , ..., gr } be the standard bases for
Rn , Rm and Rr , respectively, with dual bases E ∗ = {e1 , ..., en } , F ∗ = { f 1 , ..., f m }
and G ∗ = {g1 , ..., gr }. Let β 1 : L(Rn , Rm ) → Rr , βe k : T k (Rn ; Rr ) → Rs and β k+1 :
k+1
T (Rn ; Rm ) → Rs be the linear isomorphisms given by
b (i, j) b (i , ... ,i k , l )
β1 1 (G1 ) = f j ◦ G1 (e i ) , βe k 2 1 (G 2 ) = g l ◦ G 2 (e i1 , ... , e ik )
and
b (i , ... ,i k+1 , j)
3 1
β k+1 (G 3 ) = f j ◦ G 3 (e i1 , ... , e ik+1 )
for i, i1 , ... , ik+1 ∈ In , j ∈ Im and l ∈ Ir , where b1 : In × Im → Ir , b2 : Ink × Ir → Is and
b3 : Ink+1 × Im → Is are the bijections defined in Remark 4.27. Then we have that
b (i , ... ,i k+1 , j)
3 1
β k+1 (φ (G)) = f j ◦ G(e i1 , ... , e ik )(e i k+1 )
b (i k+1 , j)
= β1 1 ◦ G(e i1 , ... , e ik ) = g b1 (ik+1 , j) ◦ (β 1 ◦ G)(e i1 , ... , e ik )
b (i , ... ,i k , b1 (i k+1 , j)) b (i , ... ,i k+1 , j)
= βek 2 1 (β 1 ◦ G) = βe k 3 1 (β 1 ◦ G)
which shows that

β k+1 ◦ φ (G) = βe k ( β 1 ◦ G) (1)
for G ∈ T k (Rn ; L(Rn , Rm )) .
Now let (t, π −1 (U)) and (e e By
t , πe −1 (U)) be local trivializations in E and E.
h −1
Proposition 5.28 we have a local trivialization (t h , (π ) (U)) in the vector bun-
e → M , where t h,p = β h ◦ τ h,p and
dle π h : T h (E ; E)
t p ◦ F ◦ {t p−1 } h
τ h,p (F) = e
for p ∈ U and F ∈ T h (E p ; Eep ) with h = 1, k + 1 .

We also have a local trivialization (e t k , (πe k ) −1 (U)) in the vector bundle πe k :
e → M , where e
T k (E ; Λ 1 (E ; E)) t k,p = βek ◦ τek,p and
τek,p (F) = t1,p ◦ F ◦ {t p−1 } k
for p ∈ U and F ∈ T k (E p ; Λ 1(E p ; Eep )) .

Combining this, it follows that
t p ◦ F (t p−1 (a1 ), ... ,t p−1 (ak )) ◦ t p−1 (ak+1 )

τ k+1,p ◦ ψ (F)(a1 , ... , ak+1 ) = e
= τ 1,p ◦ F ◦ {t p−1 } k (a1 , ... , ak )(ak+1 ) = φ (τ 1,p ◦ F ◦ {t p−1 } k )(a1 , ... , ak+1 )
for a1 , ... , ak+1 ∈ Rn , showing that
τ k+1,p ◦ ψ (F) = φ (τ 1,p ◦ F ◦ {t p−1 } k ) (2)
for p ∈ U and F ∈ T k (E p ; Λ 1 (E p ; Eep )) .

From (1) and (2) it now follows that
t k+1,p ◦ ψ (F) = β k+1 ◦ φ (τ 1,p ◦ F ◦ {t p−1 } k ) = βe k (t 1,p ◦ F ◦ {t p−1 } k ) = e

t k,p (F)
so that
t k −1 = id U × Rs ,
t k+1 ◦ ψ ◦ e
thus showing that ψ is an equivalence.
5.32 Proposition Let π1 : E1 → M, π2 : E2 → M and πe : Ee → M be vector

bundles over the smooth manifold M of dimensions n1 , n2 and m. Then we have an
equivalence
e → Λ 1(E2 ; Λ 1(E1 ; E))
ψ : Λ 1(E1 ; Λ 1(E2 ; E)) e
e → M and πb2 : Λ 1(E2 ;
over M between the vector bundles πb1 : Λ 1(E1 ; Λ 1(E2 ; E))
1 e → M which is given by
Λ (E1 ; E))
ψ (F)(v2 )(v1 ) = F(v1 )(v2 )
for F ∈ Λ 1(E1,p ; Λ 1(E2,p ; Eep )) , v1 ∈ E1,p and v2 ∈ E2,p where p ∈ M.

PROOF : Consider first the isomorphism φ : L(Rn1 , L(Rn2 , Rm )) → L(Rn2 , L(Rn1 , Rm ))
given by
φ (G)(a2 )(a1 ) = G(a1 )(a2 )
for G ∈ L(Rn1 , L(Rn2 , Rm )) ,a1 ∈ Rn1 and a2 ∈ Rn2 . Let ri = ni m and s = n1 n2 m ,
and let Ei = {e1 , ..., eni } , F = { f1 , ..., fm } and Gi = {gi1 , ..., giri } be the standard
i i
ni
bases for Rni , Rm and Rri , respectively, with dual bases Ei∗ = {e(i) 1
, ..., e(i) }, F∗ =
1 , ..., g ri } for i = 1, 2. Let β : L(Rni , Rm ) → Rri and
{ f 1 , ..., f m } and Gi∗ = {g(i) (i) i
βe i : L(Rni , Rr3−i ) → Rs be the linear isomorphisms given by

bi ( j, k) e
b ( j, l )
βi (G) = f k ◦ G(e ij ) and βei i l
(G) = g(i) ◦ G(e ij )
for j ∈ Ini , k ∈ Im and l ∈ Ir3−i , where bi : Ini × Im → Iri and e

bi : Ini × Ir3−i → Is are the
bijections defined in Remark 4.27. We also have a permutation σ of Is defined by
σ =b
b1 ◦ ρ ◦ b
b2−1 ,
where ρ : In2 × In1 × Im → In1 × In2 × Im is the bijection given by
ρ ( j2 , j1 , k) = ( j1 , j2 , k)
for ( j2 , j1 , k) ∈ In2 × In1 × Im , and b

bi : Ini × In3−i × Im → Is is the bijection defined in
Remark 4.27 for i = 1, 2. Now we have that

b
b ( j , j , k) e
b ( j , b ( j , k)) b ( j , k)
βe2 2 2 1 (β1 ◦ φ (G)) = βe2 2 2 1 1 (β1 ◦ φ (G)) = β1 1 1 ◦ φ (G)(e2j2 )
b2 ( j2 , k)
= f k ◦ φ (G)(e2j2 )(e1j1 ) = f k ◦ G(e1j1 )(e2j2 ) = β2 ◦ G(e11 )
j
e
b1 ( j1 , b2 ( j2 , k)) b
b1 ( j1 , j2 , k) σ◦b
b2 ( j2 , j1 , k)
= βe1 (β2 ◦ G) = βe1 (β2 ◦ G) = βe1 (β2 ◦ G)
which shows that

βe 2 ( β1 ◦ φ (G)) = πσ ◦ βe 1 ( β 2 ◦ G) (1)
for G ∈ L(Rn1 , L(Rn2 , Rm )) .
Now let (ti , πi −1 (U)) and (e e By
t , πe −1 (U)) be local trivializations in Ei and E.
e −1
Proposition 5.28 we have a local trivialization (t i , πei (U)) in the vector bundle
e → M , where e
πei : Λ 1(Ei ; E) t i,p = β i ◦ τ i,p and
−1
τ i,p (F) = e
t p ◦ F ◦ ti,p
for p ∈ U and F ∈ Λ 1(Ei,p ; Eep ) with i = 1, 2 .

We also have a local trivialization (b t i , πbi−1 (U)) in the vector bundle πbi :
e → M , where b
Λ 1(Ei ; Λ 1(E3−i ; E)) t i,p = βei ◦ τei,p and
−1
τei,p (F) = e
t3−i,p ◦ F ◦ ti,p
for p ∈ U and F ∈ Λ 1(Ei,p ; Λ 1(E3−i,p ; Eep )) .

−1 −1 −1
τ 1,p ◦ ψ (F) ◦ t2,p (a2 )(a1 ) = e
t p ◦ ψ (F)(t2,p (a2 ))(t1,p (a1 ))
−1 −1 −1
=e
t p ◦ F (t1,p (a1 ))(t2,p (a2 )) = τ 2,p ◦ F ◦ t1,p (a1 )(a2 )
−1
= φ (τ 2,p ◦ F ◦ t1,p )(a2 )(a1 )
for a1 ∈ Rn1 and a2 ∈ Rn2 , showing that

−1 −1
τ 1,p ◦ ψ (F) ◦ t2,p = φ (τ 2,p ◦ F ◦ t1,p ) (2)
for p ∈ U and F ∈ Λ 1(E1,p ; Λ 1(E2,p ; Eep )) .

From (1) and (2) it now follows that
t 2,p ◦ ψ (F) = βe2 (e

b −1
t1,p ◦ ψ (F) ◦ t2,p ) = βe 2 ( β1 ◦ φ (τ 2,p ◦ F ◦ t1,p
−1
))
= πσ ◦ βe 1 ( β 2 ◦ τ 2,p ◦ F ◦ t1,p
−1
) = πσ ◦ βe1 (e −1
t2,p ◦ F ◦ t1,p ) = πσ ◦ b
t 1,p (F)
so that
b t 1 −1 = id U × πσ ,
t2 ◦ ψ ◦ b
thus showing that ψ is an equivalence.
5.33 Proposition Let ( f˜i , f ) be a bundle map from a vector bundle π1i : E1i → M1
to a vector bundle π2i : E2i → M2 for i = 1, 2, and suppose that f˜2 induces a linear
isomorphism between the fibres (π12 ) −1 (p) and (π22 ) −1 ( f (p)) for every p ∈ M1 . If s2
is a section of π2′ : Λ k (E21 ; E22 ) → M2 on an open subset V2 of M2 and V1 = f −1 (V2 ),
then the map s1 : V1 → Λ k (E11 ; E12 ) defined by
−1
s1 (p)(v1 , ..., vk ) = f˜2,p ◦ s2 ( f (p))( f˜1 (v1 ), ... , f˜1 (vk ))
for all v1 , ..., vk ∈ (π11 ) −1 (p) where p ∈ V1 , is a section of π1′ : Λ k (E11 ; E12 ) → M1
on V1 .
s1 is called the pull-back of s2 by ( f˜1 , f˜2 ) and is denoted by ( f˜1 , f˜2 ) ∗ (s2 ).
PROOF : Follows from Proposition 4.49 since
φ 1 ◦ s1 = ( f˜1 , ... , f˜1 , g̃)∗ (φ 2 ◦ s2 ) ,
where φ i : Λ k (Ei1 ; Ei2 ) → Λ k1 (Ei1 , Ei2 ) is the equivalence over Mi defined in 5.25 for
i = 1, 2 , and (g̃, f ) is the bundle map between the dual bundles (π12 ) ∗ : (E12 ) ∗ → M1
and (π22 ) ∗ : (E22 ) ∗ → M2 defined in Proposition 4.50. Indeed, we have that
φ 1 ◦ s1 (p)(v1 , ..., vk , λ ) = λ ◦ s1 (p)(v1 , ..., vk )

−1
= λ ◦ f˜2,p ◦ s2 ( f (p))( f˜1 (v1 ), ... , f˜1 (vk ))
= φ 2 ◦ s2 ( f (p))( f˜1 (v1 ), ... , f˜1 (vk ), g̃(λ ))
for v1 , ..., vk ∈ (π11 ) −1 (p) and λ ∈ (π12 ) ∗ −1 (p).
5.34 Corollary Let ( f˜, f ) be a bundle map from a vector bundle π1 : E1 → M1

to a vector bundle π2 : E2 → M2 , and let s2 be a section of π2′ : Λ k (E2 ;W ) → M2 on
M2 . Then the map s1 : M1 → Λ k (E1 ;W ) defined by
s1 (p) = ( f˜p )∗ s2 ( f (p))
for p ∈ M1 , i.e.,
s1 (p)(v1 , ... , vk ) = s2 ( f (p))( f˜p (v1 ), ... , f˜p (vk ))
for all v1 , ..., vk ∈ π1 −1 (p) where p ∈ M1 , is a section of π1′ : Λ k (E1 ;W ) → M1 on M1 .

ρ ′1−1 ◦ s1 = ( f˜, g̃) ∗ (ρ ′2−1 ◦ s2 ) ,

where ρ ′i : Λ k (Ei ; Mi ×W ) → Λ k (Ei ;W ) is the equivalence over Mi defined in 5.26

for i = 1, 2 and g̃ = f × id : M1 × W → M2 × W . Indeed, we have that
ρ ′1−1 ◦ s1 (p)(v1 , ... , vk ) = (p, s1 (p)(v1 , ... , vk ))

= g̃ p−1 ( f (p), s2 ( f (p))( f˜p (v1 ), ... , f˜p (vk )))
= g̃ p−1 ◦ (ρ ′2−1 ◦ s2 )( f (p))( f˜p (v1 ), ... , f˜p (vk ))
for p ∈ M1 and v1 , ..., vk ∈ E1,p .
5.35 Let π : E → M and πe : Ee → M be vector bundles over the smooth manifold

e → M. Then we
M , and let T be a section on M in the vector bundle π k : T k (E ; E)
have a map
e
TM : Γ(M; E) × · · · × Γ(M; E) → Γ(M; E)
| {z }
k
defined by
TM (s1 , ... , sk ) (p) = T (p) (s1 (p), ... , sk (p))
e →
for each point p ∈ M and sections s1 , ... , sk ∈ Γ(M; E) . Indeed, let φ : T k (E ; E)
e and i e : Ee → Ee ∗∗ be the equivalences over M defined in 5.25 and Propo-
T k1 (E, E) E
sition 4.10. For each choice of s1 , ... , sk we have a linear map B : Γ(M; Ee ∗ ) → F (M)
given by
B(s)(p) = φ ◦ T (p) (s1 (p), ... , sk (p), s(p))
for p ∈ M and s ∈ Γ(M; Ee ∗ ) , which satisfies
B(s)(p) = s(p) ( TM (s1 , ... , sk )(p)) = iEe ◦ TM (s1 , ... , sk )(p) (s(p)) .
Hence iEe ◦ TM (s1 , ... , sk ) is the unique section S in Γ(M; Ee ∗∗ ) with B = SM which is
given by Proposition 4.47.
We see that TM is multilinear over F (M). The next proposition shows that each
such multilinear map arises from a unique section of π k in this way.
5.36 Proposition Let π : E → M and πe : Ee → M be vector bundles over a smooth

manifold M , and let
e
F : Γ(M; E) × · · · × Γ(M; E) → Γ(M; E)
| {z }
k
be a map which is multilinear over F (M). Then there is a unique section T on M in

e → M such that F = TM .
the vector bundle π k : T k (E ; E)
PROOF : Let
G : Γ(M; E) × · · · × Γ(M; E) × Γ(M; Ee ∗ ) → F (M)

| {z }
k
be the multilinear map given by
G(s1 , ... , sk , s) (p) = s(p) (F(s1 , ... , sk )(p))
for p ∈ M , s1 , ... , sk ∈ Γ(M; E) and s ∈ Γ(M; Ee ∗ ) . Then we have that F = TM for a

section T of π k if and only if G = (φ ◦ T ) M , where φ : T k (E ; E)e → T k1 (E, E)
e is
the equivalence over M defined in 5.25. The result therefore follows from Proposition
4.47 applied to multilinear map G.
e can be thought of as
section T and the map TM , and a section in Γ(M; T k (E ; E))
e The map TM is
an operation on k sections in Γ(M; E) yielding a section in Γ(M; E).
k e
skew symmetric if and only if T is a section in Γ(M; Λ (E ; E)).
5.38 Let π : E → M be a vector bundle over a smooth manifold M , and let W

be a finite dimensional vector space. For each section ω on M in the vector bundle
π ′ : Λ k (E ;W ) → M we have a map
ωM : Γ(M; E) × · · · × Γ(M; E) → F (M;W )

| {z }
k
defined by
ωM (s1 , ... , sk ) (p) = ω (p) (s1 (p), ... , sk (p))
for each point p ∈ M and sections s1 , ... , sk ∈ Γ(M; E) . Indeed, if ρ ′k : Λ k (E ; EW ) →
Λ k (E ;W ) is the equivalence ρ ′ over M defined in 5.26, we have that
−1
ωM = ρ ′0 ◦ (ρ ′k ◦ ω )M .
We see that ωM is multilinear and skew symmetric over F (M). The next proposition
shows that each such multilinear skew symmetric map arises from a unique section
of π ′ in this way.
5.39 Proposition Let π : E → M be a vector bundle over a smooth manifold M

and W be a finite dimensional vector space, and let
F : Γ(M; E) × · · · × Γ(M; E) → F (M;W )

| {z }
k
be a map which is multilinear and skew symmetric over F (M). Then there is a
unique section ω on M in the vector bundle π ′ : Λ k (E ;W ) → M such that F = ωM .
PROOF : Apply Proposition 5.36 to the map ρ ′0 −1 ◦ F, using the notation in 5.38.
section ω and the map ωM , and a section in Γ(M; Λ k (E ;W )) can be thought of as
an operation on k sections in Γ(M; E) yielding a vector-valued smooth function in
F (M;W ).
5.41 A section of the bundle π ′ : Λ k (T M ;W ) → M defined on an open subset

V of the smooth manifold M is called a vector-valued differential form of degree k
or simply a vector-valued k-form on V with values in W . The set of all W -valued k-
forms on V is denoted by Ω k (V ;W ). In particular Ω 0 (V ;W ) = F (V ;W ) = C∞ (V,W )
is the set of C∞ -maps from V to W .
More generally, if π : E → M is a vector bundle over M, then a section of the
bundle π k : T k (T M ; E) → M defined on an open subset V of M is called a bundle-
valued covariant tensor field of degree k on V with values in E, and the set of all
E-valued covariant tensor fields on V of degree k is denoted by T k (V ; E).
A section of the subbundle π ′ : Λ k (T M ; E) → M defined on an open subset V of
M is called a bundle-valued k-form on V with values in E, and the set of all E-valued
k-forms on V is denoted by Ω k (V ; E). In particular Ω 0 (V ; E) = Γ(V ; E) is the set of
all sections of π on V .
5.42 Proposition Let M be a smooth manifold and ω : M → Λ k (T M;W ) be a

map with π ′ ◦ ω = idM . Then ω is a vector-valued k-form on M with values in W if
and only if one of the following equivalent assertions are satisfied:
(1) Given a basis C = {w1 , ..., wm } for W , the functions ω ij1 ,...,ik : U → R defined
by
φ ◦ ω (p) = ∑ ω ij1 ,...,ik (p) (dxi1 (p) ∧ · · · ∧ dxik (p)) ⊗ w j
i1 <...<ik , j
for p ∈ U are smooth for every local chart (x,U) on M.

(2) The components ω j of ω with respect to the basis C = {w1 , ..., wm } for W
defined by
m
ω (p)(v1 , ..., vk ) = ∑ ω j (p)(v1 , ..., vk ) w j
j=1
for p ∈ M and v1 , ..., vk ∈ Tp M, are real-valued differential forms on M.
(3) For every functional λ ∈ W ∗ , the map λ · ω : M → Λ k (T M) defined by
(λ · ω ) (p) = λ ◦ (ω (p))
for p ∈ M, is a real-valued differential form on M.

PROOF : (1) Follows from Propositions 4.15, 4.45 and 5.13 (1).
(2) Using the maps ψ and µ defined in 5.15, the components ω j of ω are given by
the formula
m
ω (p) = ∑ ψ ◦ µ (ω j (p), w j )
j=1
m
φ ◦ ω (p) = ∑ ω j (p) ⊗ w j .
j=1
Comparing this with the formula for φ ◦ ω (p) in (1), we hence obtain that
ω j (p) = ∑ ω ij1 ,...,ik (p) dxi1 (p) ∧ · · · ∧ dxik (p)

i1 <...<ik
for p ∈ U and j = 1, ..., m, and so (2) is equivalent to (1) by Corollary 5.22.

(3) (3) is equivalent to (2) since w∗j · ω = ω j for j = 1, ..., m, and if
m
λ= ∑ c j w∗j ,
j=1
then
m
λ ·ω = ∑ cj ω j .
j=1
5.43 Remark When k = 0 , then ω (p) ∈ W may be thought of as a map from

{0} to W mapping 0 to ω (p), so that λ ◦ ( ω (p)) simply means λ ( ω (p)) .
5.44 Proposition Let π : E → M be a vector bundle over a smooth manifold M

and ω : M → Λ k (T M; E) be a map with π ′ ◦ ω = idM . Then ω is a bundle-valued
k-form on M with values in E if and only if one of the following equivalent assertions
are satisfied :
(1) The functions ω ij1 ,...,ik : U ∩ V → R defined by
φ ◦ ω (p) = ∑ ω ij1 ,...,ik (p) (dxi1 (p) ∧ · · · ∧ dxik (p)) ⊗ s j (p)

i1 <...<ik , j
for p ∈ U ∩ V are smooth for any local basis B = {s1 , ..., sm } for E on some
open subset V of M and every local chart (x,U) on M.
(2) The components ω j of ω with respect to any local basis B = {s1 , ..., sm } for E
on some open subset V of M defined by
m
ω (p)(v1 , ..., vk ) = ∑ ω j (p)(v1 , ..., vk ) s j (p)
j=1
for p ∈ V and v1 , ..., vk ∈ Tp M, are real-valued differential forms on V .

(3) For any section s of the dual bundle π ∗ : E ∗ → M on some open subset V of M,
the map s · ω : V → Λ k (T M) defined by
(s · ω ) (p) = s(p) ◦ ω (p)
for p ∈ V , is a real-valued differential form on V .

PROOF : (1) Follows from Corollary 2.59 and Propositions 4.11, 4.15, 4.45 and 5.13
(1).
(2) Using the maps ψ and µ defined in 5.15, the components ω j of ω are given by
the formula
m
ω (p) = ∑ ψ ◦ µ (ω j (p), s j (p))
j=1
m
φ ◦ ω (p) = ∑ ω j (p) ⊗ s j (p) .
j=1
Comparing this with the formula for φ ◦ ω (p) in (1), we hence obtain that
ω j (p) = ∑ ω ij1 ,...,ik (p) dxi1 (p) ∧ · · · ∧ dxik (p)

i1 <...<ik
for p ∈ U ∩ V and j = 1, ..., m, and so (2) is equivalent to (1) by Corollary 5.22.

(3) (3) is equivalent to (2) since s j∗ · ω = ω j for j = 1, ..., m, where B ∗ = {s1∗ , ..., sm∗ }
is the dual local basis for E ∗ on V , and if
m
s= ∑ c j s j∗ ,
j=1
then
m
s·ω = ∑ cj ω j .
j=1
5.45 Example Given a smooth manifold M, we have a bundle-valued 1-form

ε ∈ Ω 1 (M; T M) on M with values in the tangent bundle defined by
ε (p) = id Tp M
for p ∈ M, i.e.,
ε (p)(v) = v
for p ∈ M and v ∈ Tp M, so that φ ◦ ε = δ is the Kronecker delta tensor on M defined

in Example 4.68.
5.46 Proposition Let G : W1 → W2 be a linear map between the finite dimen-

sional vector spaces W1 and W2 , and let ω ∈ Ω k (M;W1 ) be a vector-valued k-form
on a smooth manifold M with values in W1 . Then the map G · ω : M → Λ k (T M;W2 )
defined by
(G · ω ) (p) = G ◦ (ω (p))
for p ∈ M, is a vector-valued k-form on M with values in W2 , and we have that
λ · (G · ω ) = (λ ◦ G ) · ω
for every functional λ ∈ W2∗ .

PROOF : Follows from Proposition 5.42 (3).
5.47 Proposition Let f : M → N be a smooth map and ω ∈ Ω k (N;W ) be a

vector-valued k-form on N with values in a finite dimensional vector space W . Then
the map f ∗ ( ω ) : M → Λ k (T M;W ) defined by
f ∗ ( ω ) (p) (v1 , ... , vk ) = ω ( f (p)) ( f∗ (v1 ) , ... , f∗ (vk ))
for p ∈ M and v1 , ... , vk ∈ Tp M, is a vector-valued k-form on M with values in W , and

we have that
λ · f ∗ ( ω ) = f ∗ (λ · ω )
5.48 Proposition Let f : M → N be a smooth map, and let G : W1 → W2 be a

linear map between the finite dimensional vector spaces W1 and W2 . If ω ∈ Ω k (N;W1 )
is a vector-valued k-form on the smooth manifold N with values in W1 , then we have
that
f ∗ (G · ω ) = G · f ∗ ( ω ) .
PROOF : For every functional λ ∈ W2∗ we have that
λ · f ∗ (G· ω ) = f ∗ (λ ·(G· ω )) = f ∗ ((λ ◦ G )· ω ) = (λ ◦ G ) · f ∗ ( ω ) = λ ·(G · f ∗ ( ω ))
by Propositions 5.47 and 5.46.
5.49 Let M be a smooth manifold and W be a finite dimensional vector space.

If f : V → W is a vector-valued smooth function defined on an open neighbourhood
V of a point p ∈ M, and if l ∈ Tp M, we define l( f ) to be the derivative f∗ (l) of f
at p with respect to the tangent vector l, canonically identifying T f (p)W with W by
means of the linear isomorphism ω f (p) : T f (p)W → W given in Lemma 2.84. This is
analogous to the definition of l( f ) for real-valued functions f given in 2.77. For any
linear functional λ on W , we have that
λ ◦ l( f ) = λ ◦ ω f (p) ◦ f∗ (l) = f∗ (l)(λ ) = l(λ ◦ f ) .
If C = {w1 , ..., wm } is a basis for W and x : W → Rm is the linear isomorphism

given by
m
x( ∑ aj wj ) = a
j=1
for a ∈ Rm , then in particular we have that
x j ◦ l( f ) = l(x j ◦ f )
for j = 1, ... , m. Hence l( f ) for a W -valued function f is defined by means of its

components with respect to any basis C for W , where l acts in the usual way as a
local derivation on each component function.
If X is a vector field on M and f ∈ F (M;W ), we let X( f ) ∈ F (M;W ) be the
vector-valued function defined by
X( f )(p) = X p ( f )
for p ∈ M. We have that

λ ◦ X( f ) = X(λ ◦ f )
for every linear functional λ on W .
In particular, if f : V → C is a complex-valued function defined on V with smooth
real and complex part, we define
l( f ) = l(Re f ) + i l(Im f )
and
X( f ) = X(Re f ) + i X(Im f ) .
5.50 Proposition Let M be a smooth manifold and W be a finite dimensional

vector space. If f : U → R and g : V → W are, respectively, a real-valued and a
vector-valued smooth function defined on open neighbourhoods U and V of a point
p ∈ M, and if l ∈ Tp M, then we have that
l( f g) = l( f ) g(p) + f (p) l(g)
PROOF : For every functional λ ∈ W ∗ we have that
λ ◦ l( f g) = l( f λ ◦ g) = l( f ) λ ◦ g(p) + f (p) l(λ ◦ g) = λ ( l( f ) g(p) + f (p) l(g)) .

5.51 Let f : V → W be a vector-valued smooth function defined on an open neigh-

bourhood V of a point p in a smooth manifold M with values in a finite dimensional
vector space W . In the same way as in 4.13 we define the differential of f at p to be
the element d f (p) in Λ 1 (Tp M;W ) defined by
d f (p)(l) = l( f )
for each local derivation l in Tp M. By the discussion in 5.49 we know that this is the
derivative of f at p with respect to the tangent vector l, so that
d f (p) = f ∗ p
when we canonically identify T f (p)W with W as in Lemma 2.84. For every functional
λ ∈ W ∗ we have that
λ ◦ d f (p)(l) = λ ◦ l( f ) = l(λ ◦ f ) = d(λ ◦ f )(p)(l)
so that
λ · d f = d(λ ◦ f ) .
5.52 Proposition Let ω be an n-form on a smooth manifold M n , and let (x,U)

and (y,V ) be two local charts around a point p on M. If
ω | U = f dx1 ∧ · · · ∧ dxn and ω | V = g dy1 ∧ · · · ∧ dyn ,
then
∂ yi
f = g det on U ∩V
∂xj
PROOF : Since
∂
n
∂ yi

= ∑ j (q) ∂ i
∂xj i=1
∂ x ∂y
q q
for j = 1, ...n and q ∈ U ∩V , it follows from Corollary 5.18 that
i
1 ∂y
n
dy ∧ · · · ∧ dy = det j dx1 ∧ · · · ∧ dxn
∂x
on U ∩V which completes the proof of the proposition.
5.53 Proposition Let f : M → N be a smooth map, and let ω and η be differential

forms on N. Then we have that
f ∗ (ω ∧ η ) = f ∗ ω ∧ f ∗ η .
PROOF : For each point p on M, it follows from Definition 5.21 and Proposition 5.12
(2) that
[ f ∗ ( ω ∧ η ) ] (p) = ( f∗ p )∗ [ (ω ∧ η ) ( f (p)) ] = ( f∗ p )∗ [ ω ( f (p)) ∧ η ( f (p)) ]
= ( f∗ p )∗ [ ω ( f (p)) ] ∧ ( f∗ p )∗ [ η ( f (p)) ] = [ f ∗ ω ] (p) ∧ [ f ∗ η ] (p)
= [ f ∗ ω ∧ f ∗ η ] (p) .
EXTERIOR DIFFERENTIATION OF FORMS

5.54 We want to extend the differential d : Ω0 (M) → Ω1 (M) of functions on a
smooth manifold M, as defined in 4.13 and Proposition 4.18, to a map d : Ωk (M) →
Ωk+1 (M) called the exterior derivative of arbitrary k-forms on M for k ≥ 0. We first
prove a local result.
5.55 Proposition Let U be a coordinate neighbourhood in a smooth manifold M n .

Then there is a unique family of maps dU : Ωk (U) → Ωk+1 (U) for k ≥ 0 satisfying
the following four conditions
(1) if f ∈ Ω0 (U), then dU ( f ) = d f which is the ordinary differential of f ,
(2) dU (ω + η ) = dU ω + dU η for every ω , η ∈ Ωk (U) ,
(3) dU (ω ∧ η ) = (dU ω ) ∧ η + (−1)k ω ∧ (dU η ) whenever ω ∈ Ωk (U) and η ∈
Ωl (U) ,
(4) dU ◦ dU = 0 .
If x : U → Rn is a coordinate map and the form ω ∈ Ωk (U), where k > 0, is given by
ω = ∑ ω i1 ,...,ik dxi1 ∧ · · · ∧ dxik ,

i1 <...<ik
then
dU (ω ) = ∑ d ω i1 ,...,ik ∧ dxi1 ∧ · · · ∧ dxik
i1 <...<ik
n ∂ ω i1 ,...,ik
= ∑ ∑ dxi ∧ dxi1 ∧ · · · ∧ dxik .
i1 <...<ik i=1
∂ xi
PROOF : We first prove uniqueness of dU . If ω is the k-form given in the last part of
the proposition, it follows from (1) and (3) that
dU (ω i1 ,...,ik dxi1 ∧ · · · ∧ dxik ) = d ω i1 ,...,ik ∧ dxi1 ∧ · · · ∧ dxik

+ (−1)0 ω i1 ,...,ik ∧ dU (dxi1 ∧ · · · ∧ dxik ) ,
and we will prove by induction that dU (dxi1 ∧ · · · ∧ dxik ) = 0. In fact, for k = 1 we

have that dU (dxi1 ) = dU (dU (xi1 )) = 0 by (1) and (4). If the formula holds for k − 1,
it follows from (3) that
dU (dxi1 ∧ · · · ∧ dxik ) = dU (dxi1 ) ∧ dxi2 ∧ · · · ∧ dxik

+ (−1)1 dxi1 ∧ dU (dxi2 ∧ · · · ∧ dxik ) = 0 .
The uniqueness of dU as well as the formula for dU (ω ) now follows from (2).
To prove existence, we must show that the map dU given by (1) and the last part
of the proposition really satisfies all four conditions. We immediately see that (2) is
satisfied. Note that we have
dU ( f dxi1 ∧ · · · ∧ dxik ) = d f ∧ dxi1 ∧ · · · ∧ dxik
for arbitrary indices i1 , ..., ik in In . If some of the indices are equal, then both sides
are equal to zero. Otherwise there is a permutation σ ∈ Sk such that iσ (1) < ... < iσ (k) ,
and by Propositions 5.4 and 5.13 (2) we have
dU ( f dxi1 ∧ · · · ∧ dxik ) = dU (ε (σ ) f dxiσ (1) ∧ · · · ∧ dxiσ (k) )

= ε (σ ) d f ∧ dxiσ (1) ∧ · · · ∧ dxiσ (k) = d f ∧ dxi1 ∧ · · · ∧ dxik .
In wiew of (2), it is enough to prove (3) when ω = f dxi1 ∧ · · · ∧ dxik and η = g dx j1 ∧

· · · ∧ dx jl , in which case we have
dU (ω ∧ η ) = (g d f + f dg) ∧ dxi1 ∧ · · · ∧ dxik ∧ dx j1 ∧ · · · ∧ dx jl

= (d f ∧ dxi1 ∧ · · · ∧ dxik ) ∧ (g dx j1 ∧ · · · ∧ dx jl )
+ (−1)k ( f dxi1 ∧ · · · ∧ dxik ) ∧ (dg ∧ dx j1 ∧ · · · ∧ dx jl )
= (dU ω ) ∧ η + (−1)k ω ∧ (dU η ).
Similarly, it is enough to prove (4) for the case where ω = f dxi1 ∧ · · · ∧ dxik . We then
have that
n
∂f
dU (dU ω ) = dU ( ∑ dxi ∧ dxi1 ∧ · · · ∧ dxik )
i=1
∂ xi
n n
∂2 f j i i1 ik
=∑ ∑ j i dx ∧ dx ∧ dx ∧ · · · ∧ dx
i=1 j=1
∂ x ∂ x
where the terms with i = j vanish since dxi ∧ dxi = 0, and the other terms
∂2 f ∂2 f
cancel in pairs since = j i and dxi ∧ dx j = − dx j ∧ dxi .
∂ xi ∂ x j ∂x ∂x
The arguments above also hold for 0-forms if we make the usual convention that
a wedge product with no factors is replaced by the constant function 1 on U.
5.56 Remark Let ω be the k-form given in the last part of the proposition,
where x : U → Rm is any smooth map defined on an open subset U of M with m
components. Then dU (ω ) is still uniquely determined by the first part of the given
formula, provided that dU satisfies the four conditions in the proposition. In fact the
first part of the proof is still valid for the given ω even if (x,U) is not a coordinate
system.
5.57 Proposition If V and U are coordinate neighbourhoods in a smooth mani-

fold M with V ⊂ U, then
dV (ω |V ) = (dU ω )|V
for every ω ∈ Ωk (U) where k ≥ 0.
PROOF : The case k = 0 follows from 4.13, for if f ∈ C∞ (U) and p ∈ V we have that
d ( f |V )(p) = d f (p) since f |V and f coincide in a neighbourhood of p. This shows
that d ( f |V ) = (d f )|V .
If k > 0 it is enough to prove the assertion when ω = f dxi1 ∧ · · · ∧ dxik for a
coordinate map x : U → Rn . Then x|V is a coordinate map on V , and using the first
part of the proof we have that
dV (ω |V ) = dV ( f |V d (x|V )i1 ∧ · · · ∧ d (x|V )ik )

= d ( f |V ) ∧ d (x|V )i1 ∧ · · · ∧ d (x|V )ik ) = (dU ω )|V .
5.58 Proposition Let M n be a smooth manifold. Then there is a unique family of

maps d : Ωk (M) → Ωk+1 (M) for k ≥ 0 satisfying the following four conditions
(1) if f ∈ Ω0 (M), then d f is the ordinary differential of f ,
(2) d (ω + η ) = d ω + d η for every ω , η ∈ Ωk (M) ,
(3) d (ω ∧ η ) = (d ω )∧ η + (−1)k ω ∧(d η ) whenever ω ∈ Ωk (M) and η ∈ Ωl (M),

(4) d ◦ d = 0 .
If (x,U) is a local chart on M and the form ω ∈ Ωk (M), where k > 0, is given locally
by
ω |U = ∑ ω i1 ,...,ik dxi1 ∧ · · · ∧ dxik ,

i1 <...<ik
then
(d ω )|U = ∑ d ω i1 ,...,ik ∧ dxi1 ∧ · · · ∧ dxik

i1 <...<ik
n ∂ ω i1 ,...,ik
= ∑ ∑ dxi ∧ dxi1 ∧ · · · ∧ dxik .
i1 <...<ik i=1
∂ xi
PROOF : We first prove existence of d. If ω ∈ Ωk (M) and p is a point in M, we

choose a coordinate neighbourhood U of p in M and define
d ω (p) = dU (ω |U )(p) .
Then d is well defined, for if V is another coordinate neighbourhood of p it follows

from Proposition 5.57 that
dU (ω |U )(p) = dU∩V (ω |U∩V )(p) = dV (ω |V )(p) .
We have to show that d satisfies the four conditions of the proposition.

To prove (1), note that the first part of the proof of Proposition 5.57 also holds
for every f ∈ C∞ (M). Hence
dU ( f |U ) = d ( f |U ) = (d f )|U
where d now denotes the ordinary differential of the functions f |U and f , and this
completes the proof of (1).
To prove (2) we use that the same condition is satisfied for d|U so that
d (ω + η ) (p) = dU (ω |U + η |U )(p) = dU (ω |U )(p) + dU (η |U )(p)
= d ω (p) + d η (p) ,
and the last two conditions follow in the same way.
Finally, the formula for (d ω )|U follows from Proposition 5.55 since
(d ω )|U = dU (ω |U )
by the definition of d.
We now prove that if ω ∈ Ωk (M) and p is a point in M, then d ω (p) is uniquely
determined by the four conditions in the proposition. We first show that d ω (p) only
depends on the local behavior of the k-form ω , i.e., if η is another k-form on M
which coincides with ω on an open neighbourhood V of p, then d ω (p) = d η (p).
To see this let W be an open neighbourhood of p with W ⊂ V , and choose a smooth
function h : M → R with h = 1 on W and supp (h) ⊂ V . Such a function h exists by
Corollary 2.24. Then h ω = h η , and by (1) and (3) we have that
d (hω )(p) = dh(p) ∧ ω (p) + h(p) ∧ d ω (p) = d ω (p)
so that
d ω (p) = d (hω )(p) = d (hη )(p) = d η (p) .
The uniqueness of d ω (p) follows from (1) in the case when k = 0. If k > 0 we
choose a local chart (x,U) around p and suppose that ω |U has the local expression
given in the last part of the proposition. If V is an open neighbourhood of p with V ⊂
U, we may again choose a smooth function g : M → R with g = 1 on V and supp(g) ⊂
U. We now extend the functions g ω i1 ,...,ik and g xi on U to smooth functions η i1 ,...,ik
and yi on M, respectively, by defining them to be zero outside U, and we let
η = ∑ η i1 ,...,ik dyi1 ∧ · · · ∧ dyik .

i1 <...<ik
Then η is a k-form on M which coincides with ω on V , and by Remark 5.56 we see

that d η is uniquely determined by the four conditions in the proposition and is given
by
d η = ∑ d η i1 ,...,ik ∧ dyi1 ∧ · · · ∧ dyik .
i1 <...<ik
Combining this we now see that d ω (p) is uniquely determined and is given by
d ω (p) = ∑ d ω i1 ,...,ik (p) ∧ dxi1 (p) ∧ · · · ∧ dxik (p)
i1 <...<ik
thus proving the uniqueness of d.
5.59 Proposition Let f : M → N be a smooth map, and let ω be a k-form on N.

Then we have that
f ∗ (d ω ) = d ( f ∗ ω ) .
In particular, if g : N → R is a smooth function on N, then
f ∗ (dg) = d (g ◦ f ) .
PROOF : If l is a local derivation at a point p on M, it follows from 2.81, 4.13 and
Proposition 4.61 that
( f ∗ (dg)) (p) (l) = dg ( f (p)) ( f∗ l) = ( f∗ l) (g) = l (g ◦ f ) = d (g ◦ f ) (p) (l)
which proves the last relation in our proposition.
Using this, we can now show the first relation. Let (y,V ) be a local chart on N,
and let U be a coordinate neighbourhood contained in f −1 (V ). If
ω |V = ∑ ω j1 ,..., jk dy j1 ∧ · · · ∧ dy jk ,
j1 <...< jk
it follows by induction from Proposition 5.53 and Remark 5.56 that

f ∗ (d ω ) |U = ∑ d (ω j1 ,..., jk ◦ f |U ) ∧ d(y j1
◦ f |U ) ∧ · · · ∧ d(y jk ◦ f |U )
j1 <...< jk
= d |U ( f ∗ ω |U ) = d ( f ∗ ω ) |U .
5.60 Definition The exterior derivative d ω of a vector-valued k-form ω ∈

Ω k (M;W ) is defined by means of the components with respect to a basis C =
{w1 , ..., wm } for W , i.e., d ω ∈ Ω k+1 (M;W ) is the vector-valued (k + 1)-form defined
by
(d ω ) j = d (ω j )
for j = 1, ..., m.
5.61 Proposition If ω ∈ Ω k (M;W ) is a vector-valued k-form on M, then

λ · d ω = d (λ · ω )
for every functional λ ∈ W ∗ . In particular, the exterior derivative d ω does not depend
on the choice of basis C for W in Definition 5.60.
PROOF : If C ∗ = {w∗1 , ..., w∗m } is the dual of the basis C and

m
λ= ∑ c j w∗j ,
j=1
then
m m m
λ ·dω = ∑ c j (d ω ) j = ∑ c j d (ω j ) = d ( ∑ c j ω j ) = d (λ · ω ) .
j=1 j=1 j=1
5.62 Proposition Let G : W1 → W2 be a linear map between the finite dimen-

sional vector spaces W1 and W2 , and let ω ∈ Ω k (M;W1 ) be a vector-valued k-form
on a smooth manifold M with values in W1 . Then we have that
d (G · ω ) = G · d ω .
PROOF : For every functional λ ∈ W2∗ we have that
λ · d (G · ω ) = d (λ · (G · ω )) = d ((λ ◦ G ) · ω ) = (λ ◦ G ) · d ω = λ · (G · d ω )
by Propositions 5.61 and 5.46.
5.63 Proposition Let f : M → N be a smooth map, and let ω ∈ Ω k (N;W ) be a

vector-valued k-form on N with values in a finite dimensional vector space W . Then
we have that
f ∗ (d ω ) = d ( f ∗ ω ) .
In particular, if g : N → W is a smooth vector-valued function on N, then
f ∗ (dg) = d (g ◦ f ) .
PROOF : For every functional λ ∈ W ∗ we have that
λ · f ∗ (d ω ) = f ∗ (λ · d ω ) = f ∗ (d(λ · ω )) = d( f ∗ (λ · ω )) = d(λ · f ∗ ω ) = λ · d( f ∗ ω )
by Propositions 5.47, 5.61 and 5.59.
5.64 Proposition Let M n be a smooth manifold and W be a finite dimensional

vector space. Then the family of maps d : Ωk (M;W ) → Ωk+1 (M;W ) for k ≥ 0 satis-
fies the following conditions:
(1) if f ∈ Ω0 (M;W ), then d f is the ordinary differential of f defined in 5.51,
(2) d (ω + η ) = d ω + d η for every ω , η ∈ Ωk (M;W ) ,
(3) d ◦ d = 0 .
PROOF : Condition (1) follows from 5.51. Using Propositions 5.58 (2) and (4) we
have that
λ · d (ω + η ) = d (λ · ω + λ · η ) = d (λ · ω ) + d (λ · η ) = λ · (d ω + d η )
and
λ · d(d ω ) = d(λ · d ω ) = d(d (λ · ω )) = 0
for every ω , η ∈ Ωk (M;W ) and every functional λ ∈ W ∗ , thereby showing (2) and
(3).
WEDGE PRODUCT OF VECTOR-VALUED FORMS

5.65 Let W1 , W2 and W be finite dimensional vector spaces, and let
ν : W1 × W2 → W
be a bilinear map. If T ∈ T k (V ;W1 ) and S ∈ T l (V ;W2 ) for a vector space V , we

define their tensor product T ⊗ν S ∈ T k+l (V ;W ) with respect to ν by
( T ⊗ν S)(v1 , ..., vk , vk+1 , ..., vk+l ) = ν ( T (v1 , ..., vk ), S(vk+1 , ..., vk+l )) .
If ω ∈ Λk (V ;W1 ) and η ∈ Λl (V ;W2 ), their wedge product ω ∧ ν η ∈ Λk+l (V ;W )

with respect to ν is defined by
(k + l)!
ω ∧ν η = A (ω ⊗ ν η) .
k! l!
We will often omit the reference to ν in the notation if it is clear from the context.
Since ν is bilinear and A is linear, it follows that ⊗ ν and ∧ ν are bilinear. If
F : U → V is a linear map from a vector space U, we see that
F ∗ ( T ⊗ ν S) = F ∗ ( T ) ⊗ ν F ∗ (S) ,
and hence
F ∗ (ω ∧ ν η ) = F ∗ (ω ) ∧ ν F ∗ (η )
by Proposition 5.9.
5.66 Proposition If ω i and η j are the components of ω ∈ Λk (V ;W1 ) and η ∈

Λl (V ;W2 ) with respect to the bases B = {e1 , ... , er } and C = { f1 , ... , fs } for W1
and W2 , respectively, and if gi j = ν (ei , f j ) for i = 1, ... , r and j = 1, ... , s, then
( ω ∧ ν η )(v1 , ..., vk+l ) = ∑ ( ω i ∧ η j )(v1 , ..., vk+l ) gi j (1)

ij
for v1 , ..., vk+l ∈ V , and

λ ◦ ( ω ∧ ν η ) = ∑ λ (gi j ) ω i ∧ η j (2)
ij
for every linear functional λ ∈ W ∗ .

PROOF : Since ν : W1 × W2 → W is bilinear, we have that
( ω ⊗ ν η ) (v1 , ... , vk , vk+1 , ... , vk+l )
= ν ( ω (v1 , ... , vk ), η (vk+1 , ... , vk+l ))
= ∑ ω i (v1 , ... , vk ) η j (vk+1 , ... , vk+l ) ν (ei , f j )

ij
= ∑ ( ω i ⊗ η j ) (v1 , ... , vk , vk+1 , ... , vk+l ) gi j

ij
for every v1 , ..., vk+l ∈ V . By appying the alternation operator A on both sides, we
obtain formula (1), and formula (2) is obtained from (1) by applying λ on both
sides.
5.67 Proposition Let V , W1 and W be finite dimensional vector spaces, and let
ν : W1 × W1 → W be a bilinear map so that
ν (u, v) = (−1) n ν (v, u)
for u, v ∈ W1 , where n is a fixed integer. If ω ∈ Λk (V ;W1 ) and η ∈ Λl (V ;W1 ) , then
ω ∧ η = (−1) kl + n η ∧ ω .
5.68 Proposition If ω ∈ Ω k (V ;W1 ) and η ∈ Ω l (V ;W2 ) are vector-valued differ-

ential forms on an open subset V of a smooth manifold M, we have a vector-valued
differential form ω ∧ ν η ∈ Ωk+l (V ;W ) , called the wedge product of ω and η with
respect to ν , defined by
( ω ∧ ν η )(p) = ω (p) ∧ ν η (p)
for p ∈ V . If ω i j
and η are the components of ω and η with respect to the bases B =
{e1 , ... , er } and C = { f1 , ... , fs } for W1 and W2 , respectively, and if gi j = ν (ei , f j )
for i = 1, ... , r and j = 1, ... , s, then
( ω ∧ ν η )(p)(v1 , ..., vk+l ) = ∑ ( ω i ∧ η j )(p)(v1 , ..., vk+l ) gi j (1)
ij
for p ∈ V and v1 , ..., vk+l ∈ Tp M, and

λ · ( ω ∧ ν η ) = ∑ λ (gi j ) ω i ∧ η j (2)
ij
for every linear functional λ ∈ W ∗ .

5.69 Proposition Let V be an open subset of a smooth manifold M. If W1 , W2

and W are finite dimensional vector spaces and ν : W1 × W2 → W is a bilinear map,
then
d (ω ∧ ν η ) = (d ω ) ∧ ν η + (−1)k ω ∧ ν (d η )
whenever ω ∈ Ω k (V ;W1 ) and η ∈ Ω l (V ;W2 ).
PROOF : Follows from the same relation for real-valued forms given in proposition
5.58 (3), together with Definition 5.60 and Propositions 5.61 and 5.68.
5.70 Proposition Let f : M → N be a smooth map. If W1 , W2 and W are finite

dimensional vector spaces and ν : W1 × W2 → W is a bilinear map, then
f ∗ (ω ∧ ν η ) = f ∗ ω ∧ ν f ∗ η
whenever ω ∈ Ω k (N;W1 ) and η ∈ Ω l (N;W2 ).

PROOF : For each point p on M, it follows from 5.65 and Proposition 5.68 that
[ f ∗ ( ω ∧ ν η ) ] (p) = ( f∗ p ) ∗ [ ( ω ∧ ν η ) ( f (p)) ] = ( f∗ p ) ∗ [ ω ( f (p)) ∧ ν η ( f (p)) ]
= ( f∗ p ) ∗ [ ω ( f (p)) ] ∧ ν ( f∗ p ) ∗ [ η ( f (p)) ] = [ f ∗ ω ] (p) ∧ ν [ f ∗ η ] (p)
= [ f ∗ ω ∧ ν f ∗ η ] (p) .
INTERIOR PRODUCT
5.71 Definition If V is a vector space, we define the interior product of a vector
v ∈ V and a covariant tensor T ∈ T k (V ) to be the covariant tensor iv T ∈ T k−1 (V )
given by
iv T (v1 , ... , vk−1 ) = T (v, v1 , ... , vk−1 )
for v1 , ... , vk−1 ∈ V . We let iv T = 0 when k = 0. If T is alternating, so is iv T .
We define the kernel of a k-form ω ∈ Λk (V ) by
ker (ω ) = {v ∈ V | iv ω = 0} ,
which equals the kernel of the linear map φ : V → Λ k−1 (V ) given by φ (v) = iv ω for
v ∈ V.

smooth manifold M, and let s : M → T k (E) be a section in the vector bundle

π k : T k (E) → M. Then we have a bundle map φ : E → T k−1 (E) over M given
by
φ (v) = iv s(p)
for p ∈ M and v ∈ E p .
PROOF : Let (t, π −1 (U)) be a local trivialization in the vector bundle π : E → M, and
let τ :(π k ) −1 (U)
→ U × T k (Rn ) and τ ′ : (π k−1 ) −1 (U) → U × T k−1 (Rn ) be the
corresponding equivalences over U defined in Proposition 4.53. Then
τ ◦ s(p) = (p, h(p))
for p ∈ U, where h : U → T k (Rn ) is the smooth map given by h(p) = (t p−1 )∗ s(p).
From this it follows that
τ ′ ◦ φ ◦ t −1 (p, a) = (p, B(a, h(p)))
for (p, a) ∈ U ×Rn , where B : Rn ×T k (Rn ) → T k−1 (Rn ) is the bilinear map defined
by B(a, T ) = ia T . Indeed, we have that
(t p−1 )∗ {it −1
p (a)
s(p)}(a1 , ... , ak−1 ) = h(p)(a, a1 , ... , ak−1 ) = ia h(p)(a1 , ... , ak−1 )
for p ∈ U and a, a1 , ... , ak−1 ∈ Rn .
5.73 Remark If ω : M → Λk (E) is a section in the vector bundle π ′ : Λ k (E) → M,

then the bundle map obtained from ω as described in Proposition 5.72 also induces
a bundle map φ : E → Λk−1 (E) over M .
Suppose that there is an integer r such that dim ker ω (p) = dim ker (φ p ) = r for
every p ∈ M . Then it follows from Proposition 2.65 that ker(φ ) is an r-dimensional
subbundle of E which we also call the kernel of ω and denote by ker(ω ) .
5.74 Proposition If v ∈ V and λ1 , ... , λk ∈ V ∗ , we have that

k
iv ( λ1 ∧ · · · ∧ λk ) = ∑ (−1)i+1 iv ( λi ) λ1 ∧ · · · ∧ λi−1 ∧ λi+1 ∧ · · · ∧ λk .
i=1
PROOF : If v1 , ... , vk ∈ V , it follows from Proposition 5.13 (1) that

iv1 ( λ 1 ∧ · · · ∧ λ k )(v2 , ... , vk ) = ( λ 1 ∧ · · · ∧ λ k )(v1 , ... , vk )

λ 1 (v1 ) λ 1 (v2 ) ··· λ 1 (vk )

λ 2 (v1 ) λ 2 (v2 ) ··· λ 2 (vk )

= ∑ ε (σ ) λ σ (1) (v1 ) · · · λ σ (k) (vk ) = .. .. ..
σ ∈Sk
. . .

λ k (v1 ) λ k (v2 ) ··· λ k (vk )
k
= ∑ (−1)i+1 λ i (v1 ) (λ 1 ∧ · · · ∧ λ i−1 ∧ λ i+1 ∧ · · · ∧ λ k )(v2 , ... , vk )
i=1
k
= ∑ (−1)i+1 iv1 ( λ i )(λ 1 ∧ · · · ∧ λ i−1 ∧ λ i+1 ∧ · · · ∧ λ k ) (v2 , ... , vk )
i=1
where we have expanded the determinant along the first column.
5.75 Proposition If ω ∈ Λk (V ) and η ∈ Λl (V ), we have that
iv (ω ∧ η ) = (iv ω ) ∧ η + (−1)k ω ∧ (iv η ) .
PROOF : If ω = ω1 ∧· · · ∧ ωk and η = η1 ∧· · · ∧ ηl , where ω1 , ... , ωk , η1 , ... , ηl ∈ V ∗ ,

k
iv (ω ∧ η ) = ∑ (−1)i+1 iv ( ωi ) ω1 ∧ · · · ∧ ωi−1 ∧ ωi+1 ∧ · · · ∧ ωk ∧ η
i=1
l
+ ∑ (−1)k+ j+1 iv ( η j ) ω ∧ η1 ∧ · · · ∧ η j−1 ∧ η j+1 ∧ · · · ∧ ηl
j=1
= (iv ω ) ∧ η + (−1)k ω ∧ (iv η ) .
The general result now follows since every form ω ∈ Λk (V ) and η ∈ Λl (V ) can be
expressed as a linear combination of forms of the above type.
5.76 Proposition If F : V → W is a linear map, and if v ∈ V and T ∈ T k (W ),

then we have that
iv F ∗ (T ) = F ∗ (iF(v) T ) .
iv F ∗ (T ) (v1 , ... , vk−1 ) = F ∗ (T ) (v, v1 , ... , vk−1 ) = T (F(v), F(v1 ), ... , F(vk−1 ))
= iF(v) T (F(v1 ), ... , F(vk−1 )) = F ∗ (iF(v) T ) (v1 , ... , vk−1 )
for v1 , ... , vk−1 ∈ V .

5.77 Proposition If X is a vector field and ω a k-form on a smooth manifold M,

we have a (k − 1)-form iX ω on M, called the interior product of X and ω , defined
by
(iX ω )(p) = iX(p) ω (p)
for each p ∈ M.
PROOF : We have that iX ω = φ ◦ X, where φ : T M → Λk−1 (T M) is the bundle map
over M obtained from ω as described in Remark 5.73.
THE LIE DERIVATIVE OF FORMS

5.78 Definition If T ∈ T k (M) is a covariant tensor field of degree k on a smooth
manifold M, and if σ ∈ Sk is a permutation, we define T σ by T σ (p) = T (p) σ for
every p ∈ M.
We define a map A : T k (M) → T k (M), called the alternation operator on
k
T (M), by
A (T )(p) = A (T (p))
for each p ∈ M. We see that A is linear over F (M).
5.79 Proposition If f : M → N is a smooth map, and if AM and AN are the

alternations on T k (M) and T k (N), respectively, then
f ∗ ( AN (T )) = AM ( f ∗ (T ))
for every T ∈ T k (N).

PROOF : For each point p on M, it follows from Definition 5.78 and Proposition 5.9
that
[ f ∗ ( AN (T )) ] (p) = ( f∗ p )∗ [ AN (T ) ( f (p)) ] = ( f∗ p )∗ [ A Tf (p) N (T ( f (p))) ]
= A Tp M [ ( f∗ p )∗ (T ( f (p))) ] = A Tp M [ f ∗ (T )(p) ] = [ AM ( f ∗ (T )) ] (p)
where A Tp M and A Tf (p) N are the alternations on T k (Tp M) and T k (T f (p) N) , re-
spectively.
5.80 Proposition If X is a vector field on the smooth manifold M, then the Lie
derivative LX commutes with the alternation operator A on T k (M).
PROOF : Let γ be the global flow for X, and let T ∈ T k (M) and p ∈ M. If c : I(p) →
k
T l (Tp M)and b : I(p) → T kl (Tp M) are the curves defined by
c(t) = γ t∗ (A (T ))(p) and b(t) = γ t∗ T (p) ,
and we canonically identify Tc(0) T kl (Tp M) and Tb(0) T kl (Tp M) with T kl (Tp M) as in
Lemma 2.84, then
LX (A (T ))(p) = c′ (0) and LX T (p) = b′ (0) .
Since c(t) = A ◦ b(t) for t ∈ I(p) by Proposition 5.79, it now follows from Lemma
2.84 that
LX ( A (T ))(p) = c′ (0) = A ◦ b′ (0) = A (LX T )(p) .

(1) If ω is a k-form on M, so is LX ω .
(2) If ω ∈ Λk (V ) and η ∈ Λl (V ), then LX (ω ∧ η ) = (LX ω ) ∧ η + ω ∧ (LX η ).
PROOF : (1) It follows from Propositions 5.8 and 5.80 that
LX ω = LX (A (ω )) = A (LX ω )
which shows that LX ω is a k-form on M.

(2) Follows from Propositions 4.79 (2) and 5.80.
5.82 Proposition If X is a vector field and ω a k-form on the smooth manifold

M, then we have that
LX (d ω ) = d (LX ω ) .
PROOF : Let p0 be a point on M n and (x,U) be a local chart around p0 , and choose
an open neighbourhood W of p0 contained in U and a real number ε > 0 such that
< −ε , ε > × W ⊂ D(X) and γ (< −ε , ε > × W ) ⊂ U. Let cq : I → M be an integral
curve for X with initial condition q ∈ W , and let vi (t) be obtained from vi = dxi (q)
for i = 1, ... , n by Lie transport along cq as defined in Remark 4.78. Then we have
that
∗
vi (t) = dxi (q) ◦ γ −t ∗ = γ −t dxi (q) = d (xi ◦ γ −t ) (γ t (q)) = d yi (p) ,
where p = cq (t) and (y, γ t (W )) is a local chart on M with y = x ◦ γ −t . Then if
ω (cq (t)) = ∑ Cq i1 ,...,i (t) vi1 (t) ∧ · · · ∧ vik (t)

k
i1 ,...,ik
= ∑ Dt i1 ,...,ik (p) d yi1 (p) ∧ ... ∧ d yik (p)

i1 ,...,ik
where Dt i1 ,...,ik = Ct i1 ,...,ik ◦ γ −t , we have that

n
d ω (cq (t)) = ∑ ∑ ∂ i Dt i1 ,...,ik dyi (p) ∧ dyi1 (p) ∧ · · · ∧ dyik (p)
i1 ,...,ik i=1
∂y
p

n
= ∑ ∑ ∂ i Ct i1 ,...,ik vi (t) ∧ vi1 (t) ∧ · · · ∧ vik (t)
i1 ,...,ik i=1
∂x
q
so that

n
d ∂

i i1 ik
LX (d ω ) (q) = ∑ ∑ dt ∂ xi C i1 ,...,ik dx (q) ∧ dx (q) ∧ · · · ∧ dx (q)
i1 ,...,ik i=1 0 q
= d (LX ω ) (q)
for every q ∈ W .
5.83 Proposition The Lie derivative of a k-form ω with respect to a vector field
X on a smooth manifold M is given by the Cartan formula
LX ω = d (iX ω ) + iX (d ω ) .
PROOF : Follows since both LX and d ◦ iX + iX ◦ d are derivations on Ωk (U) for each
open subset U of M, which commute with d and restriction to open subsets, and have
the same effect on functions. Indeed, whenever ω ∈ Ωk (U) and η ∈ Ωl (U) , it fol-
lows from Propositions 5.55 and 5.75 that
(d ◦ iX ) (ω ∧ η ) = d [ (iX ω ) ∧ η + (−1)k ω ∧ (iX η ) ]
= (d ◦ iX ) (ω ) ∧ η + (−1)k−1 (iX ω ) ∧ (d η )
+ (−1)k (d ω ) ∧ (iX η ) + ω ∧ (d ◦ iX ) (η )
and
(iX ◦ d ) (ω ∧ η ) = iX [ (d ω ) ∧ η + (−1)k ω ∧ (d η ) ]
= (iX ◦ d ) (ω ) ∧ η + (−1)k+1 (d ω ) ∧ (iX η )

+ (−1)k (iX ω ) ∧ (d η ) + ω ∧ (iX ◦ d ) (η )
so that
(d ◦ iX + iX ◦ d ) (ω ∧ η ) = (d ◦ iX + iX ◦ d ) (ω ) ∧ η + ω ∧ (d ◦ iX + iX ◦ d ) (η ) .
Moreover, we have that
(d ◦ iX + iX ◦ d ) ◦ d = d ◦ iX ◦ d = d ◦ (d ◦ iX + iX ◦ d )
and
LX ( f ) = X( f ) = d f (X) = iX (d f ) = (d ◦ iX + iX ◦ d )( f ) .
The result hence follows from the expressions for ω in local coordinates.
5.84 Proposition Let ω be a k-form on the smooth manifold M, and let

X1 , ... , Xk+1 be vector fields on M. Then we have that
k+1
d ω (X1 , ... , Xk+1 ) = ∑ (−1)i+1 Xi (ω (X1 , ... , X̂i , ... , Xk+1 ))
i=1
+ ∑ (−1)i+ j ω ( [ Xi , X j ], X1 , ... , X̂i , ... , X̂ j , ... , Xk+1 ) ,

1≤i< j≤k+1
where a circumflex over a term means that it is omitted.

PROOF : We prove the formula by induction on k. It is clearly true for k = 0 since
d ω (X1 ) = X1 (ω ) when ω is a smooth function on M. Assuming that the formula is
true for k − 1, it follows from Propositions 4.79 (5) and 5.83 that
d ω (X1 , ... , Xk+1 ) = iX1 (d ω ) (X2 , ... , Xk+1 )
= LX1 ω (X2 , ... , Xk+1 ) − d (iX1 ω ) (X2 , ... , Xk+1 )
k+1
= X1 (ω (X2 , ... , Xk+1 )) − ∑ ω ( X2 , ... , [ X1 , X j ], ... , Xk+1 )
j=2
k+1
− ∑ (−1)i Xi (ω (X1 , X2 , ... , X̂i , ... , Xk+1 ))
i=2
− ∑ (−1)i+ j ω ( X1 , [ Xi , X j ], X2 , ... , X̂i , ... , X̂ j , ... , Xk+1 )

2≤i< j≤k+1
k+1
= ∑ (−1)i+1 Xi (ω (X1 , ... , X̂i , ... , Xk+1 ))
i=1
+ ∑ (−1)i+ j ω ( [ Xi , X j ], X1 , ... , X̂i , ... , X̂ j , ... , Xk+1 ) .

1≤i< j≤k+1
5.85 Remark Proposition 5.84 is also valid for vector-valued k-forms ω on M

with values in a finite dimensional vector space W . To see this, we simply apply
Proposition 5.84 to λ · ω for an arbitrary linear functional λ on W , and use 5.49 and
Proposition 5.61.
5.86 Remark The case k = 1 of Proposition 5.84 and Remark 5.85 is particularly
useful. If ω is a real-valued or a vector-valued 1-form on the smooth manifold M, and

if X and Y are vector fields on M, then we have that
d ω ( X,Y ) = X ( ω (Y )) − Y ( ω ( X)) − ω ( [ X ,Y ] ) .
5.87 Proposition Let ω be a k-form on a smooth manifold M, and suppose that

there is an integer r such that dim ker ω (p) = r for every p ∈ M . Then ker(ω ) is an
r-dimensional distribution on M. A vector field X on M belongs to ker(ω ) if and only
if iX ω = 0 . If X and Y are vector fields belonging to ker(ω ), then we have that
iX iY d ω = i [X,Y ] ω , (1)
and the distribution ker(ω ) is integrable when ω is a closed form.

PROOF : The first part of the proposition follows from Remark 5.73, and formula (1)
follows from Proposition 5.84.
5.88 Corollary Let ω be a nowhere vanishing 1-form on a smooth manifold M n .

Then ker(ω ) is an (n − 1)-dimensional distribution on M. It is integrable if and only
if i∗ d ω = 0 , where i : ker(ω ) → T M is the inclusion map.

Chapter 6
INTEGRATION ON MANIFOLDS
MANIFOLDS WITH BOUNDARY

6.1 Let A be any subset of Rn such that O ⊂ A ⊂ O for an open set O in Rn . A map
f : A → Rm is said to be smooth at a point p ∈ A if there is an open neighbourhood
V1 of p in Rn and a smooth map f1 : V1 → Rm such that f1 | A ∩V1 = f | A ∩V1 .
The derivative D f (p) is defined to be D f1 (p). We see that it is well defined, for
if f2 : V2 → Rm is another smooth map defined on some open neighbourhood V2 of p
with f2 | A ∩V2 = f | A ∩V2 , then D f1 (p) = D f2 (p). Indeed, this is clearly satisfied for
p ∈ O, and if p ∈ A − O we can choose a sequence {pk } in O ∩V1 ∩V2 converging to
p so that
D f1 (p) = lim D f1 (pk ) = lim D f2 (pk ) = D f2 (p) .
k→∞ k→∞
f is said to be smooth on A if it is smooth at every point in A.
6.2 Proposition Let A be a subset of Rn such that O ⊂ A ⊂ O for an open set O

in Rn . Then a map f : A → Rm is smooth on A if and only if it can be extended to a
smooth map f˜ : V → Rm defined on some open set V in Rn containing A.
PROOF : If f has such an extension f˜, it follows directly from the definition that f is
smooth on A.
Conversely, suppose that f is smooth on A, and choose for each point p in A
an open neighbourhood Vp ofS p in Rn and a smooth map g p : Vp → Rm such that
g p | A ∩Vp = f | A ∩Vp . Let V = p∈A Vp , and let {φ p |p ∈ A} be a partition of unity on
V subordinate to the open cover {Vp |p ∈ A}. For each point p in A, let f p be the
extension of the map φ p g p to V defining it to be zero outside Vp . Then f˜ = ∑ p∈A f p
is a smooth map which is an extension of f to the open set V containing A.
6.3 Let Hn = {x ∈ Rn |xn ≥ 0} denote the upper half space in Rn , and let
IntHn = {x ∈ Rn |xn > 0} and ∂ Hn = {x ∈ Rn |xn = 0} be its interior and its bound-
ary, respectively.
A second countable Hausdorff space M is called a topological manifold with
boundary if each point p in M has an open neighbourhood U homeomorphic to an
open set in Hn for some integer n ≥ 0. A map x : U → Hn sending U homeomorphi-
cally onto an open subset x(U) of Hn , is called a coordinate map at p, and the pair
163
(x,U) is called a local coordinate system or a local chart . We can now define an
atlas and a smooth structure on M in the same way as in Definition 2.1. If M has a
smooth structure, it is called a smooth manifold with boundary.
A point p in M n is called a boundary point if there is a local chart (x,U) around
p with x(p) ∈ ∂ Hn . By the inverse function theorem it then follows that y(p) ∈
∂ Hn for every local chart (y,V ) around p. The set of boundary points in M is called
the boundary of M and is denoted by ∂ M. Its complement M − ∂ M is called the
interior of M and is denoted by IntM.
If (x,U) is a local chart on M, we have that x (U ∩ ∂ M) = x(U) ∩ ∂ Hn and x (U ∩
IntM) = x(U) ∩ IntHn . Hence Int M is open in M, and it follows from Example 2.9
(c) and Proposition 2.36 that Int M and ∂ M are smooth submanifolds of M without
boundary of dimensions n and n − 1, respectively. For each local chart (x,U) on
M with x(U) ∩ ∂ Hn 6= 0, / we have an induced local chart (x̃, Ũ) on ∂ M defined by
Ũ = U ∩ ∂ M and x̃ = p ◦ x|Ũ where p : Rn−1 × R → Rn−1 is the projection on the
first factor. These charts form an induced atlas on ∂ M.
6.4 Let M n be a smooth manifold with boundary, and let U be an open neigh-
bourhood of a point p in M. Then a map x : U → Rn , sending U homeomorphically
onto some subset x(U) of Rn , is called a generalized coordinate map at p if there
is open set O in Rn with O ⊂ x(U) ⊂ O, and if y ◦ x−1 and x ◦ y−1 are C∞ for every
local chart (y,V ) in the smooth structure on M.
Note that the domain x(U ∩V ) of y ◦ x−1 is an open subset of x(U) so that x(U ∩
V ) = x(U) ∩W for some open set W in Rn , and we have that O ∩W ⊂ x(U) ∩W ⊂
O ∩W . To show the last inclusion, let Q be any open neighbourhood of a point q ∈
x(U) ∩ W . Then Q ∩ W is also an open neighbourhood of q so that Q ∩ W ∩ O 6= 0/
which shows that q ∈ O ∩W . Hence it is meaningful to require that y ◦ x−1 is C∞ in
the sense defined in 6.1. The same is true for x ◦ y−1 which is defined on the open
subset y(U ∩V ) of the half space Hn .
The pair (x,U) is called a generalized local chart around p. If f : V → R is a
smooth function defined on some open neighbourhood V of p, we define the i-th
partial derivative of f at p with respect to (x,U) to be
∂f
(p) = Di ( f ◦ x−1 )(x(p)) ,
∂ xi
i.e., the i-th partial derivative of f ◦ x−1 at x(p).
EXACT FORMS
6.5 A k-form ω on a smooth manifold M is said to be closed if d ω = 0 . It is
called exact if there is a (k − 1)-form η on M such that ω = d η . Since d ◦ d = 0 ,
INTEGRATION ON MANIFOLDS 165
every exact form is closed, but the converse statement is not true in general. We are
going to show that it is in fact true for smoothly contractible manifolds, and therefore
also true locally on an arbitrary smooth manifold.
6.6 Lemma Let V = V1 ⊕V2 where dimV1 = n and dimV2 = 1, and let πi : V → V
be the projection with range Vi for i = 1, 2 . If λ ∈ Λ1 (V ) is a non-zero 1-form on V
with λ = π2∗ (λ ), then every k-form ω ∈ Λk (V ) can be written uniquely as
ω = α + (λ ∧ β )
where α ∈ Λk (V ) and β ∈ Λk−1 (V ) are forms with α = π1∗ (α ) and β = π1∗ (β ).

If v is a basis vector for V2 with λ (v) = 1, then α and β are given by α = π1∗ (ω )
and β (u1 , ..., uk−1 ) = ω (v , π1 (u1 ) , ..., π1 (uk−1 )) for u1 , ..., uk−1 ∈ V .
PROOF : The last part of the lemma follows immediately from the first, and it only
remains to show the existence of the forms α and β . Let B1 = {v1 , ..., vn } be a
basis for V1 , and let v0 = v. Then B = {v0 , ..., vn }is a basis for V with dual basis
B ∗ = {v∗0 , ..., v∗n } where λ = v∗0 . By Proposition 5.14 we have that
ω= ∑ ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik

0≤i1 <...<ik ≤n
which shows that

ω = α + (λ ∧ β )
where
α= ∑ ω (vi1 , ..., vik ) v∗i1 ∧ · · · ∧ v∗ik
1≤i1 <...<ik ≤n
and
β= ∑ ω (v0 , vi1 , ..., vik−1 ) v∗i1 ∧ · · · ∧ v∗ik−1 ,
1≤i1 <...<ik−1 ≤n
and we see that α = π1∗ (α ) and β = π1∗ (β ).
6.7 Remark The condition ω = π ∗j (ω ) for a form ω ∈ Λr (V ) in the above lemma,

where j = 1, 2, is equivalent to the condition that
ω (v1 , ..., vr ) = 0 if some vi ∈ V3− j .
6.8 Proposition Let M n be a smooth manifold and I = [0, 1], and let it : M →
M × I be the map defined by it (p) = (p,t) for p ∈ M and t ∈ I. Then there is a family
of linear maps
h : Ωk (M × I) → Ωk−1 (M)
for k > 0 satisfying
d ◦ h + h ◦ d = i1∗ − i0∗ .
PROOF : Let ρ1 : M × I → M and ρ2 : M × I → I be the projections on the first and

second factor, and let i p : I → M × I be the map defined by i p (t) = (p,t) for t ∈ I and
p ∈ M. Then the tangent space of M × I at a point (p,t) can be written as a direct
sum
T(p,t) (M × I) = it ∗ (Tp M) ⊕ i p ∗ (Tt I) .
The summands are the ranges of the projections π1 ∗ (p,t) and π2 ∗ (p,t), where π1 =
it ◦ ρ1 and π2 = i p ◦ ρ2 . Let s : M × I → R be the function defined by s = i ◦ ρ2
where i : I → R is the inclusion map. Then we have a non-zero 1-form ds(p,t) on
T(p,t) (M × I) with ds(p,t) = π2∗ (ds(p,t)) since s ◦ π2 = i ◦ ρ2 ◦ i p ◦ ρ2 = i ◦ ρ2 = s.
By Lemma 6.6 it now follows that each k-form ω ∈ Ωk (M × I) can be written
uniquely as
ω = α + ( ds ∧ β )
where α ∈ Ωk (M × I) and β ∈ Ωk−1 (M × I) are forms with α (p,t) = π1∗ (α (p,t))
and β (p,t) = π1∗ (β (p,t)) for all (p,t) ∈ M × I, and we define
Z 1
(hω )(p)(v1 , ..., vk−1 ) = (it∗ β )(p)(v1 , ..., vk−1 ) dt
0
for p ∈ M and v1 , ..., vk−1 ∈ Tp M.

Having given an invariant definition of h, we can now prove the formula in the
proposition locally using a coordinate system. Let (x,U) be a local chart in M, and
let x = x ◦ ρ1 . Then (x1 , ..., xn , s) is a generalized coordinate map, in the sense given
in 6.4, defined on the open coordinate neighbourhood U × I in M × I.
If
ω | U×I = ∑ α i1 ,...,ik d x i1 ∧ ... ∧ d x ik

i1 <...<ik
+ ∑ β j1 ,..., jk−1 d s ∧ d x j1 ∧ ... ∧ d x j k−1 ,

j1 <...< jk−1
then
Z 1

(hω )(p) = ∑ β j1 ,..., jk−1 (p,t) dt d x j1 (p) ∧ ... ∧ d x j k−1 (p)
j1 <...< jk−1 0
for p ∈ U, so that
d (hω )(p)
n Z 1

∂ β j1 ,..., jk−1
= ∑ ∑ (p,t) dt d x j (p) ∧ d x j1 (p) ∧ ... ∧ d x j k−1 (p) .
j1 <...< jk−1 j=1 0 ∂xj
Furthermore, we have that

n ∂ α i1 ,...,ik
(d ω ) | U×I = ∑ ∑ d x i ∧ d x i1 ∧ ... ∧ d x ik
i1 <...<ik i=1
∂ xi
∂ α i1 ,...,ik
+ ∑ ∂s d s ∧ d x i1 ∧ ... ∧ d x ik
i1 <...<ik
n ∂ β j1 ,..., jk−1
+ ∑ ∑ d x j ∧ d s ∧ d x j1 ∧ ... ∧ d x j k−1 ,
j1 <...< jk−1 j=1
∂xj
which implies
h(d ω )(p)
Z 1

∂ α i1 ,...,ik
= ∑ ∂s (p,t) dt d x i1 (p) ∧ ... ∧ d x ik (p)
i1 <...<ik 0
n Z 1

∂ β j1 ,..., jk−1
− ∑ ∑ (p,t) dt d x j (p) ∧ d x j1 (p) ∧ ... ∧ d x j k−1 (p) .
j1 <...< jk−1 j=1 0 ∂xj
d (hω )(p) + h(d ω )(p)

Z 1

∂ α i1 ,...,ik
= ∑ ∂s (p,t) dt d x i1 (p) ∧ ... ∧ d x ik (p)
i1 <...<ik 0
= ∑ [ α i1 ,...,ik (p, 1) − α i1 ,...,ik (p, 0) ] d x i1 (p) ∧ ... ∧ d x ik (p)

i1 <...<ik
= i1∗ ω (p) − i0∗ ω (p)
6.9 Poincaré Lemma If M is a smoothly contractible manifold, then every closed

form ω on M is exact.
PROOF : Let F : M × I → M be a smooth contraction of M to a point p ∈ M , and
let it : M → M × I be the map defined by it (q) = (q,t) for q ∈ M and t ∈ I. Then
F ◦ i0 = id |M is the identity map, and F ◦ i1 the constant map on M with value p .
Hence it follows from Proposition 6.8 that
ω = (F ◦ i0 ) ∗ ω − (F ◦ i1 ) ∗ ω = −(d ◦ h + h ◦ d ) F ∗ ω = d(−h F ∗ ω )
since d (F ∗ ω ) = F ∗ (d ω ) = 0 .
6.10 Proposition Let (x,U) be a local chart around a point p on a smooth man-
ifold M where x(U) is star-shaped with respect to x(p) . Then U is smoothly con-
tractible to p .
PROOF : We have a smooth contraction F : U × I → U of U to p defined by
F(q,t) = x−1 (t x(p) + (1 − t) x(q))
for q ∈ U and t ∈ I .
ORIENTATION
6.11 Definition We say that two ordered bases (v1 , ... , vn ) and (w1 , ... , wn ) for a
vector space V are equally oriented , and write (v1 , ... , vn ) ∼ (w1 , ... , wn ), if
n
w j = ∑ a i j vi for j = 1, ... , n (1)
i=1
with det (ai j ) > 0. If det (ai j ) < 0, the bases are said to be oppositely oriented .
We see that ∼ is an equivalence relation in the family of ordered bases for V ,
and the equivalence class of (v1 , ... , vn ) is denoted by [ v1 , ... , vn ] and is called an
orientation of V . Since each ordered basis in V is equivalent to either (v1 , ... , vn ) or
(−v1 , ... , vn ), we see that V has exactly two orientations. If µ denotes one orientation
of V , the other is denoted by −µ . A vector space V with an orientation µ is called an
oriented vector space and is denoted by (V, µ ) if we want to indicate the orientation
explicitly. An ordered basis (v1 , ... , vn ) in (V, µ ) is said to be positively oriented if
[ v1 , ... , vn ] = µ and negatively oriented if [ v1 , ... , vn ] = −µ .
Now let (V, µ ) and (W, ν ) be two oriented vector spaces of dimension n, and
let F : V → W be a linear isomorphism. By applying F to both sides of (1), we see
that (v1 , ... , vn ) ∼ (w1 , ... , wn ) implies that (F(v1 ), ... , F(vn )) ∼ (F(w1 ), ... , F(wn )).
From this and the same argument for F −1 we see that {(F(v1 ), ... , F(vn )) | (v1 , ... , vn )
∈ µ } is an orientation of W which we denote by F(µ ). We say that F is orientation
preserving if F(µ ) = ν and orientation reversing if F(µ ) = −ν .
6.12 Example The standard orientation of Rn is [ e1 , ... , en ] where (e1 , ... , en ) is

the standard basis.
If M is a smooth manifold, we can introduce the standard orientation on the
product bundle M × Rn by choosing the orientation [ (p, e1 ), ... , (p, en ) ] on the fibre
{p} × Rn for each point p in M.
6.13 Definition Let π : E → M be an n-dimensional vector bundle, and choose

an orientation µ p on each fibre π −1 (p). The family µ = { µ p |p ∈ M} is said to be an
orientation of E if for each point p in M, there is an open neighbourhood U around
p and n sections s1 , ..., sn of π on U such that [ s1 (q), ..., sn (q) ] = µq for all q ∈ U.
The vector bundle E is said to be orientiable if it has an orientation, and non-
orientiable otherwise. A vector bundle E with an orientation µ is called an oriented
vector bundle and is denoted by (E, µ ) if we want to indicate the orientation explic-
itly in the notation.
6.14 Proposition Let V be an n-dimensional vector space, and let ω ∈ Λn (V ) be

a non-zero n-form on V . Then there is a unique orientation µ of V such that
[ v1 , ... , vn ] = µ if and only if ω (v1 , ... , vn ) > 0 .
PROOF : Follows directly from Proposition 5.17.
6.15 Proposition Let π : E → M be an n-dimensional vector bundle, and let ω

be a nowhere vanishing section of the bundle π ′ : Λn (E) → M on M. Then there is a
unique orientation µ of E such that
[ v1 , ... , vn ] = µ p if and only if ω (p) (v1 , ... , vn ) > 0
for every p ∈ M.
In the last two propositions µ is called the orientation determined by ω , and ω is
called a volume element compatible with the orientation µ .
PROOF : Given a point p on M, let (t, π −1 (U)) be a local trivialization around p with
U connected. Then we have a nowhere vanishing smooth function f : U → R defined
by
f (q) = ω (q) (tq−1 (e1 ), ...,tq−1 (en ))
for q ∈ U, which must be either positive or negative on U, and we can define n
sections s1 , ..., sn of π on U by s1 (q) = sgn ( f (q))tq−1 (e1 ) and si (q) = tq−1 (ei ) for
i = 2, ..., n so that [ s1 (q), ..., sn (q) ] = µq for all q ∈ U. This completes the proof that
µ is an orientation of E.
6.16 Proposition If µ is an orientation of the n-dimensional vector bundle π :

E → M, then there is a nowhere vanishing section ω of the bundle π ′ : Λn (E) → M
on M which is compatible with µ .
PROOF : Let {Uα | α ∈ A} be an open cover of M such that for each α ∈ A, there
are n sections sα ,1 , ..., sα ,n of π on Uα with [ sα ,1 (p), ..., sα ,n (p) ] = µ p for every p ∈
Uα . Let sα∗ ,1 , ..., s∗α ,n be the sections of the dual bundle π ∗ : E ∗ → M on Uα so that
{s∗α ,1 (p), ..., s∗α ,n (p)} is the dual basis of {sα ,1 (p), ..., sα ,n (p)} for p ∈ Uα .
To see that they are sections of π ∗ , let E = {e1 , ..., en } be the standard basis
for Rn and E ∗ = {e1 , ..., en } be its dual basis, and let (tα , π −1 (Uα )) be the local
trivialization in E associated with the sections sα ,1 , ..., sα ,n . For each p ∈ Uα and

i = 1, ..., n, we have that sα ,i (p) = tα−1,p (ei ) so that s∗α ,i (p) = ei ◦ tα ,p by Proposition
4.11. From Proposition 4.12 it therefore follows that s∗α ,i is a section of π ∗ on Uα for
i = 1, ..., n.
Now let {φα |α ∈ A} be a partition of unity on M subordinate to the open cover
{Uα |α ∈ A}, and for each α ∈ A, let ωα be the extension of the section φα s∗α ,1 ∧
. . . ∧ s∗α ,n to M defining it to be zero outside Uα . Then ω = ∑α ∈A ωα is a nowhere
vanishing section of π ′ : Λn (E) → M which is compatible with µ .
6.17 Corollary The n-dimensional vector bundle π : E → M is orientiable if and

only if one of the following equivalent assertions are satisfied :
(1) The bundle π ′ : Λn (E) → M has a nowhere vanishing section.
(2) The bundle π ′ : Λn (E) → M is trivial.
PROOF : (1) follows from Proposition 6.15 and 6.16, and (2) from Proposition 2.58.
6.18 Definition A smooth manifold M is said to be orientiable if its tangent

bundle is orientiable, and non-orientiable otherwise. An orientation of T M is also
called an orientation of M.
A smooth manifold M with an orientation µ is called an oriented smooth man-
ifold and is denoted by (M, µ ) if we want to indicate the orientation explicitly. A
local chart (x,U) on M is said to be positively oriented with respect to µ if

[ ∂ 1 , ..., ∂∂xn ] = µ p
∂x
p p
for every p ∈ U. It is said to be negatively oriented with respect to µ if

[ ∂ 1 , ..., ∂∂xn ] = −µ p
∂x
p p
for every p ∈ U.
6.19 Remark If (tx , π −1 (U)) is the local trivialization in the tangent bundle
assosiated with the local chart (x,U) on M, then it follows by Remark 2.82 that (x,U)
is positively (negatively) oriented if and only if the equivalence tx : π −1 (U) → U × Rn
is orientation preserving (orientation reversing) on each fibre when U × Rn is given
the standard orientation.
6.20 Proposition If (M n , µ ) is an oriented smooth manifold, then a local chart

(x,U) on M with U connected is either positively or negatively oriented.
PROOF : If ω is a nowhere vanishing n-form on M which is compatible with the

orientation µ , we have a nowhere vanishing smooth function f : U → R defined by

f (q) = ω (q) ( ∂ 1 , ..., ∂∂xn )
∂x
q q
for q ∈ U, which must be either positive or negative on U since U is connected. Hence

the result follows from Definition 6.18.
6.21 Remark If the local chart (x,U) on M n is negatively oriented and φ : Rn →

Rn is the reflection given by
φ (a1 , a2 , ..., an ) = (−a1 , a2 , ..., an ) ,
then (φ ◦ x,U) is a local chart on M which is positively oriented.
6.22 Proposition A smooth manifold M n is orientable if and only if it has an

atlas A such that i
∂y
det j > 0 on U ∩V
∂x
for every pair of local charts (x,U) and (y,V ) in A .
PROOF : Assume first that M is orientable, and let µ be an orientation of M. Then it
follows from Proposition 6.20 and Remark 6.21 that M has an atlas A consisting of
positively oriented charts. If (x,U) and (y,V ) are two local charts in A , the inequality
in the proposition follows from the relations

∂
n
∂ yi ∂

j = ∑ j (p)
∂x i=1
∂ x ∂ yi
p p
for j = 1, ..., n and p ∈ U ∩V .

Conversely, assume that M has an atlas A satisfying the property in the propo-
sition, and let p be a point on M. To define the orientation µ p on the tangent space
Tp M, we choose a local chart (x,U) around p belonging to A and let

µ p = [ ∂ 1 , ..., ∂∂xn ] .
∂x
p p
It follows from the inequality in the proposition the µ p is well defined and does not
depend on the choice of local chart (x,U).
The family µ = {µ p |p ∈ M} is an orientation of M since for each point p on
M, there is a local chart (x,U) around p belonging to A , and the partial derivations
X j = ∂ j for j = 1, ..., n are n vector fields on U such that [ X1 (q), ..., Xn (q) ] = µq
∂x
for all q ∈ U.
6.23 Remark An atlas satisfying the property of Proposition 6.22 is called an

oriented atlas on M. Such an atlas A uniquely determines an orientation µ of M
such that every local chart in A is positively oriented with respect to µ .
6.24 Proposition Let M n be an orientable smooth manifold, and let (x,U) and
i
∂y
(y,V ) be local charts on M with U and V connected. Then det j cannot change
∂x
sign on U ∩V .
PROOF : Let µ be an orientation of M. Then it follows from Proposition 6.20 that
each of the local charts (x,U) and (y,V ) is either positively or negatively oriented.
From the relations
∂
n
∂ yi

= ∑ j (p) ∂ i
∂xj i=1
∂ x ∂y
p p

∂ yi
for j = 1, ..., n and p ∈ U ∩V , we see that det is positive on U ∩V if the local
∂xj
charts (x,U) and (y,V ) are equally oriented, and it is negative on U ∩ V if they are
oppositely oriented.
6.25 Lemma Let M n be a smooth manifold with boundary. If p ∈ ∂ M and (x,U)

and (y,V ) are two local charts around p, we have that
∂ yn ∂ yn
(p) = 0 for j = 1, ..., n − 1 and ∂ xn
(p) > 0 .
∂xj
PROOF : Since
x (U ∩V ∩ ∂ M) = x(U ∩V ) ∩ ∂ Hn
and
y (U ∩V ∩ ∂ M) = y(U ∩V ) ∩ ∂ Hn ,
we have that
(y ◦ x−1 ) (x(U ∩V ) ∩ ∂ Hn ) = y(U ∩V ) ∩ ∂ Hn
which implies the first assertion in the lemma.
The second assertion follows from the inclusion
(y ◦ x−1 ) (x(U ∩V )) = y(U ∩V ) ⊂ Hn
and the fact that

∂ yi
det (p) 6= 0 .
∂xj
6.26 Proposition If M n is an orientable smooth manifold with boundary, then ∂ M

is an orientable smooth submanifold of M n without boundary of dimension n − 1.
PROOF : By 6.3 we only need to show that ∂ M is orientable. Let A be an oriented

atlas on M, and let (x,U) and (y,V ) be two local charts in A intersecting ∂ M. If
e and (ỹ, Ve ) are the induced local charts on ∂ M, it follows by Lemma 6.25 that
(x̃, U)
∂ yn
det D (y ◦ x−1 ) = detD (ỹ ◦ x̃−1 ) · ∂ xn
n
on Ue ∩V e where ∂ yn > 0. Hence detD (y ◦ x−1 ) > 0 on U ∩ V implies that
∂x
e ∩ Ve , showing that the induced atlas on ∂ M is also ori-
det D (ỹ ◦ x̃−1 ) > 0 on U
ented.
6.27 Let M n be a smooth manifold with boundary, and let

p ∈ ∂ M. If (x,U) is a
n
local chart around p, we say that the tangent vector ∑ vi ∂ i is inward pointing if
i=1
∂x
p
vn > 0 and outward pointing if vn < 0.
n
This does not depend on the choice of local chart, for if ∑ wi ∂ i is the repre-
∂y i=1 p
sentation of the tangent vector with respect to another local chart (y,V ) around p, we
have that
w = D(y ◦ x−1 )(x(p)) v
so that
∂ yn
wn = ∂ xn (p) vn
∂ yn
by Lemma 6.25 where ∂ xn (p) > 0. Hence wn > 0 (resp., wn < 0) if and only if vn > 0
(resp., vn < 0).
6.28 Proposition Let (M n , µ ) be an oriented smooth manifold with boundary,

and let i : ∂ M → M be the inclusion map. Then we have an induced orientation ∂ µ
on ∂ M such that given an ordered basis (v1 , ... , vn−1 ) in Tp (∂ M), we have that
[ v1 , ... , vn−1 ] = (∂ µ ) p if and only if [ u, i∗ v1 , ... , i∗ vn−1 ] = µ p
for any outward pointing vector u ∈ Tp M.

If A is an atlas on M consisting of positively oriented local charts with respect
to µ , then the orientation determined by the induced atlas on ∂ M is (−1)n ∂ µ .
PROOF : We first note that (u, i∗ v1 , ... , i∗ vn−1 ) is an ordered basis for Tp M since
u∈/ i∗ Tp (∂ M). If u′ is another outward pointing vector in Tp M, then
n−1
u′ = a u + ∑ ai i∗ vi ,
i=1
where a > 0. Indeed, if (tx , π −1 (U)) is the local trivialization assosiated with a local
chart (x,U) around p, then tx,p (u′ ) n = atx,p (u) n where tx,p (u) n > 0 and tx,p (u′ ) n > 0
since u and u′ are outward pointing. Hence
[ u′ , i∗ v1 , ... , i∗ vn−1 ] = [ u, i∗ v1 , ... , i∗ vn−1 ]
which shows that (∂ µ ) p does not depend on the choice of u.

If (v1 , ... , vn−1 ) and (w1 , ... , wn−1 ) are two ordered bases for Tp (∂ M), we
clearly have that (v1 , ... , vn−1 ) ∼ (w1 , ... , wn−1 ) if and only if (u, i∗ v1 , ... , i∗ vn−1 )
∼ (u, i∗ w1 , ... , i∗ wn−1 ). For suppose that
n−1
wj = ∑ A i j vi for j = 1, ... , n − 1,
i=1
and that
n−1
u = B 00 u + ∑ B i0 i∗ vi and
i=1
n−1
i∗ w j = B 0 j u + ∑ B i j i∗ vi for j = 1, ... , n − 1.
i=1
Then
1 0
B=
0 A
which implies that detB = det A , showing that (∂ µ ) p is well defined by the conditions
in the proposition.
Now let (x,U) be a local chart around p ∈ ∂ M which is positively oriented with

e be the induced local chart on ∂ M. Then − ∂ n is an
respect to µ , and let (x̃, U) ∂x
p
outward pointing vector in Tp M, and we have that
! !

[ − ∂∂xn , i∗ ∂
, ... , i∗ ∂
] = [ − ∂∂xn , ∂ 1 , ... , ∂n−1 ]
∂ x̃1 ∂ x̃n−1 ∂x ∂x
p p p p p p

= (−1)n [ ∂ 1 , ... , ∂∂xn ] = (−1)n µ p
∂x
p p
so that

[ ∂ 1 , ... , ∂n−1 ] = (−1)n (∂ µ ) p .
∂ x̃ ∂ x̃
p p
This shows the last part of the proposition and also that ∂ µ is an orientation of ∂ M.
6.29 Let (N, ν ) and (M, µ ) be two oriented smooth manifolds where N n−1 is an
immersed submanifold of M n , and let i : N → M be the inclusion map and p ∈ N .
Then a tangent vector u ∈ Tp M − i∗ (Tp N) is said to be positive if given an ordered

basis (v1 , ... , vn−1 ) in Tp N , we have that
[ v1 , ... , vn−1 ] = ν p if and only if [ u, i∗ v1 , ... , i∗ vn−1 ] = µ p .
A tangent vector u ∈ Tp M − i∗ (Tp N) is said to be negative if −u is positive.
The sign of u is well defined and does not depend on the choice of ordered ba-
sis (v1 , ... , vn−1 ) , for let (w1 , ... , wn−1 ) be another ordered basis in Tp N. Then we
clearly have that (v1 , ... , vn−1 ) ∼ (w1 , ... , wn−1 ) if and only if (u, i∗ v1 , ... , i∗ vn−1 )
∼ (u, i∗ w1 , ... , i∗ wn−1 ) .
Two tangent vectors u, u′ ∈ Tp M − i∗ (Tp N) have the same sign if
[ u, i∗ v1 , ... , i∗ vn−1 ] = [ u′ , i∗ v1 , ... , i∗ vn−1 ]
for an ordered basis (v1 , ... , vn−1 ) in Tp N , which is satisfied if u′ = a u + w where
a > 0 and w ∈ i∗ (Tp N) . They have opposite sign if a < 0 .
A vector field X : N → T M along the inclusion map i : N → M is said to be
positive if X(p) is a positive vector for every p ∈ N .
6.30 Remark In the case where (M, µ ) is an oriented smooth manifold with
boundary and (N, ν ) is the oriented manifold (∂ M, ∂ µ ) , the positive (negative) tan-
gent vectors are the outward (inward) pointing vectors.
6.31 Proposition Let N n−1 be an immersed submanifold of an oriented smooth

manifold (M n , µ ), and let X : N → T M be a vector field along the inclusion map i :
N → M with X(p) ∈ Tp M − i∗ (Tp N) for every p ∈ N . Then we have a unique induced
orientation ν of N such that X is positive, i.e., given an ordered basis (v1 , ... , vn−1 ) in
the tangent space Tp N at an arbitrary point p ∈ N, we have that
[ v1 , ... , vn−1 ] = ν p if and only if [ X(p), i∗ v1 , ... , i∗ vn−1 ] = µ p .
PROOF : If (v1 , ... , vn−1 ) and (w1 , ... , wn−1 ) are two ordered bases in Tp N , we
clearly have that (v1 , ... , vn−1 ) ∼ (w1 , ... , wn−1 ) if and only if (X(p), i∗ v1 , ... ,
i∗ vn−1 ) ∼ (X(p), i∗ w1 , ... , i∗ wn−1 ) , which shows that ν p is well defined by the con-
ditions in the proposition.
Now let ω be a volume element on M compatible with µ , and let s1 , ..., sn−1 be a
local basis for T N on an open neighbourhood U of a point p ∈ N such that
[ s1 (p), ... , sn−1 (p) ] = ν p
which implies that
ω (p)(X(p), i∗ ◦ s1 (p), ... , i∗ ◦ sn−1 (p)) > 0 .
By continuity there is therefore a neighbourhood V of p contained in U so that
ω (q)(X(q), i∗ ◦ s1 (q), ... , i∗ ◦ sn−1 (q)) > 0
for q ∈ V , which implies that
[ s1 (q), ... , sn−1 (q) ] = νq
for q ∈ V . This shows that ν is an orientation of N .
INTEGRATION OF DIFFERENTIAL FORMS
6.32 We want to define the integral of an n-form ω with compact support on

an oriented smooth manifold (M n , µ ). Suppose first that there is a positively oriented
local chart (x,U) on M with supp (ω ) ⊂ U. Then we have a smooth function f : U →
R such that
ω | U = f dx1 ∧ ... ∧ dxn
with supp ( f ) = supp (ω ), and we define
Z Z
ω= f ◦ x−1
M x(U)
where the last integral is the Riemann integral over any rectangle containing the
support of f ◦ x−1 . We extend the Zintegrand to Rn by defining it to be zero outside
x(U). The next lemma shows that ω is well defined and does not depend on the
M
choice of local chart (x,U) containing supp (ω ).
6.33 Lemma Let (x,U) and (y,V ) be positively oriented local charts on M with
supp (ω ) ⊂ U ∩V . If
ω | U = f dx1 ∧ ... ∧ dxn and ω | V = g dy1 ∧ ... ∧ dyn ,
then Z Z
f ◦ x−1 = g ◦ y−1 .
x(U) y(V )
PROOF : Since supp ( f ) and supp (g) are both contained in U ∩ V , we may replace
x(U) and y(V ) in the above integrals by x(U ∩ V ) and y(U ∩ V ), respectively. By
Proposition 5.52 we have that
f ◦ x−1 = (g ◦ y−1 ) ◦ (y ◦ x−1 ) det D(y ◦ x−1 )
on x(U ∩ V ), and the coordinate transformation y ◦ x−1 is a diffeomorphism map-

ping x(U ∩ V ) onto y(U ∩ V ). Since the local charts (x,U) and (y,V ) are positively
oriented, we have that detD (y ◦ x−1 ) > 0 on x(U ∩V ). Hence
Z Z Z
f ◦ x−1 = (g ◦ y−1 ) ◦ (y ◦ x−1 ) | detD (y ◦ x−1 ) | = g ◦ y−1
x(U∩V ) x(U∩V ) y(U∩V )
by the change of variable formula for multiple integrals.
6.34 Lemma Let ω and (x,U) be as above, and let A = {(xα ,Uα ) | α ∈A} be an
atlas on M consisting of positively oriented local charts. Let {φα |α ∈A} be a partition
of unity on M subordinate to the open cover {Uα |α ∈ A}. Then we have that
Z Z
M
ω= ∑ φα ω
α ∈A M
where only a finite number of terms in the sum are non-zero.

PROOF : Each φα ω is an n-form on M with compact support contained in U, and
since supp (ω ) is compact, φα ω is non-zero only for a finite number of indices α ∈ A.
If
ω | U = f dx1 ∧ ... ∧ dxn ,
we have that
Z Z Z Z
ω= f ◦ x−1 = ∑ (φα f ) ◦ x−1 = ∑ φα ω .
M x(U) α ∈A x(U) α ∈A M
6.35 We now want to define the integral of an n-form ω with compact support on
an oriented smooth manifold (M n , µ ) in the case where supp (ω ) is not necessarily
contained in a single coordinate neighbourhood. Let A = {(xα ,Uα ) | α ∈ A} be an
atlas on M consisting of positively oriented local charts with respect to µ , and let
{φα |α ∈ A} be a partition of unity on M subordinate to the open cover {Uα |α ∈ A}.
Then we define Z Z
ω= ∑ φα ω .
M α ∈A M
Since supp (ω ) is compact, Zonly a finite number of terms in the sum are non-zero.
The next lemma shows that ω is well defined and does not depend on the choice
M
of atlas and partition of unity on M.
6.36 Lemma Suppose we have two atlases A = {(xα ,Uα ) | α ∈ A} and

B = {(yβ ,Vβ ) | β ∈ B} on M consisting of positively oriented local charts, and let
{φα |α ∈ A} and {ψβ |β ∈ B} be partitions of unity on M subordinate to the open
covers {Uα |α ∈ A} and {Vβ |β ∈ B}, respectively. Then we have that
Z Z
∑ M
φα ω = ∑ ψβ ω .
α ∈A β ∈B M
PROOF : By Lemma 6.34 we have that

Z Z
M
φα ω = ∑ ψβ φα ω
β ∈B M
for each α ∈ A, and Z Z
M
ψβ ω = ∑ φα ψβ ω
α ∈A M
for each β ∈ B. Hence it follows that

Z Z Z Z
∑ φα ω = ∑ ∑ ψβ φα ω = ∑ ∑ φα ψβ ω = ∑ ψβ ω .
α ∈A M α ∈A β ∈B M β ∈B α ∈A M β ∈B M
6.37 Remark If there is a positively oriented

Z local chart (x,U) on M with
supp (ω ) ⊂ U, we now have two definitions of ω given in 6.32 and 6.35. How-
M
ever, by Lemma 6.34, these two definitions coincide.
6.38 Stokes’ Theorem Let M n be an oriented smooth manifold with boundary,

and let i : ∂ M → M be the inclusion map. If ω is an (n − 1) - form on M with compact
support, then Z Z
dω = i ∗ω .
M ∂M
PROOF : Let µ be the orientation of M, and let ∂ µ be the induced orientation on ∂ M

given in Proposition 6.28. Suppose first that there is a positively oriented local chart
(x,U) on M with supp (ω ) ⊂ U. If
n
ω | U = ∑ (−1)i−1 fi dx1 ∧ ... ∧ dxi−1 ∧ dxi+1 ∧ ... ∧ dxn ,
i=1
we have that !
n
∂f
d ω |U = ∑ ∂ xii dx1 ∧ ... ∧ dxn .
i=1
Choose a closed cube Cr = {a ∈ Rn ||ai | ≤ r for 1 ≤ i ≤ n} containing the support of

fi ◦ x−1 for i = 1, ... , n. We now distinguish two types of local charts (x,U).
Consider first the case where x(U) ∩ ∂ Hn = 0, / and let gi be the extension of
fi ◦ x−1 to Rn defining it to be zero outside x(U) for i = 1, ... , n. Then we have that
Z n Z r Z r
dω = ∑ ··· Di gi dx1 ... dxn
M i=1 −r −r
n Z r Z r
=∑ ··· [ gi (x1 , ... , xi−1 , r, xi+1 , ... , xn )
i=1 −r −r
− gi (x1 , ... , xi−1 , − r, xi+1 , ... , xn ) ] dx1 ... dxi−1 dxi+1 ... dxn = 0 .
Since i ∗ ω = 0, this completes the proof of Stokes’ theorem in this case.

Consider next the case where x(U) ∩ ∂ Hn 6= 0, e be the induced local
/ and let (x̃, U)
−1
chart on ∂ M. Let gi be the extension of fi ◦ x to H , defining it to be zero outside
n
x(U) for i = 1, ... , n. Then we have that

Z n Z rZ r Z r
dω = ∑ ··· Di gi dx1 ... dxn
M i=1 0 −r −r
Z r Z r
=− ··· gn (x1 , ... , xn−1 , 0) dx1 ... dxn−1 .
−r −r
Since i ∗ dxn = d (xn ◦ i) = 0, it follows that
(i ∗ ω ) | Ue = (−1)n−1 ( fn ◦ i) d x̃1 ∧ ... ∧ d x̃n−1 .
e on ∂ M is positively (resp., negatively)

By Proposition 6.28, the local chart (x̃, U)
oriented with respect to ∂ µ when n is even (resp., odd).
Z Hence using this chart, we
must include an extra factor (−1)n in the formula for i ∗ ω , and we obtain
∂M
Z Z r Z r
i∗ω = − ··· gn (x1 , ... , xn−1 , 0) dx1 ... dxn−1
∂M −r −r
which completes the proof of Stokes’ theorem in the case where x(U) ∩ ∂ Hn 6= 0. /
Now consider the general case where supp(ω ) is not necessarily contained in
a single coordinate neighbourhood. Let A = {(xα ,Uα ) | α ∈ A} be an atlas on M
consisting of positively oriented local charts with respect to µ , and let {φα |α ∈A} be
a partition of unity on M subordinate to the open cover {Uα |α ∈ A}. By the first part
of the proof we have that
Z Z
d (φα ω ) = i ∗ (φα ω )
M ∂M
for every α ∈ A, and the integrals are non-zero only for a finite number of indices α
since supp (ω ) is compact. Hence we have that
Z Z Z Z
dω = ∑ d (φα ω ) = ∑ i ∗ (φα ω ) = i ∗ω
M α ∈A M α ∈A ∂M ∂M
which completes the proof of Stokes’ theorem in the general case.

Chapter 7
METRIC AND SYMPLECTIC
STRUCTURES
COVARIANT TENSORS OF DEGREE 2

7.1 Definition Let V be a finite dimensional vector space. A bilinear functional
g : V × V → R is said to be non-degenerate if
g(v, w) = 0 for every w ∈ V implies that v = 0 ,
and it is said to be positive definite (resp., negative definite ) if it satisfies the stronger
condition
g(v, v) > 0 (resp. g(v, v) < 0) for every v 6= 0 .
The transpose gt of g is the bilinear functional gt : V × V → R defined by
gt (v, w) = g(w, v)
for v, w ∈ V . We say that g is symmetric if gt = g and skew symmetric if gt = −g.

A bilinear functional g ∈ T 2 (V ) which is symmetric and non-degenerate, is
called a metric on V . If it is also positive definite, it is called a Riemannian met-
ric or an inner product on V . A metric is said to be definite if it is either positive or
negative definite, otherwise it is said to be indefinite. We also use <v, w> as an alter-
native notation for g(v, w) when v, w ∈ V . If V and W are vector spaces with metrics
g and h, respectively, then a linear map F : V → W is called a linear isometry if
h(F(v), F(w)) = g(v, w)
for all v, w ∈ V .
A bilinear functional g ∈ T 2 (V ) which is skew symmetric and non-degenerate,
is called a symplectic form on V .
7.2 If V is a vector space, we have a linear isomorphism
φ : T 2 (V ) → L(V,V ∗ )
given by φ (g) = G, where

G(v)(w) = g (v, w)
181
for v, w ∈ V and g ∈ T 2 (V ). The inverse map
φ −1 : L(V,V ∗ ) → T 2 (V )
is given by φ −1 (G) = g, where
g(v, w) = G(v)(w)
for v, w ∈ V and G ∈ L(V,V ∗ ).

If V is finite dimensional and B = {v1 , ..., vn } is a basis for V with dual basis
B ∗ = {v∗1 , ..., v∗n }, then
g = ∑ g i j v∗i ⊗ v∗j
i, j
where g i j = g(vi , v j ) for i, j = 1, ..., n. These numbers are called the components of g
with respect to the basis B, and the matrix (g i j ) is also denoted by mB (g) . We see
that mB (g t ) = mB (g) t .
Now if φ (g) = G, we have that
n n
G(v j ) = ∑ G(v j )(vi ) v∗i = ∑ g j i v∗i
i=1 i=1
so that mB t
B ∗ (G) = (g i j ) . Hence if
n
v = ∑ a i vi ,
i=1
then
n n
G(v) = ∑ ai v∗i where ai = ∑ gji aj .
i=1 j=1
We say that G(v) is obtained from v by lowering indices , and the map G is denoted
by g ♭ . We have that g ♭ (v) = i v g . The linear functional g ♭ (v) is also denoted by v ♭ .
By the rank of the bilinear map g we mean the rank of the matrix (g i j ) , which
is equal to dim G(V ) and therefore independent of the choice of the basis B. We
immediately see that g and its transpose gt always have the same rank.
We see that g is non-degenerate if and only if φ (g) = G is a linear isomorphism
so that the matrix (gi j ) is invertible, i.e., det(gi j ) 6= 0. The inverse of (gi j ) is denoted
by (gi j ). If
n
λ = ∑ ai v∗i ,
i=1
then
n n
G −1 (λ ) = ∑ ai vi where ai = ∑ gji aj .
i=1 j=1
We say that G −1 (λ )
is obtained from λ by raising indices , and the map G −1 is
denoted by g . We have that i g ♯ (λ ) g = λ . The vector g ♯ (λ ) is also denoted by λ ♯ .
♯
METRIC AND SYMPLECTIC STRUCTURES 183
More generally, we have linear isomorphisms
L sr : T kl (V ) → T k+1 r k k−1
l−1 (V ) and R s : T l (V ) → T l+1 (V ) ,
also called lowering and raising of indices, respectively, which are defined by
L sr (T ) (w1 , ... , wk+1 , λ 1 , ... , λ l−1 )
= T (w1 , ... , wr−1 , wr+1 , ... , wk+1 , λ 1 , ... , λ s−1 , G(wr ), λ s , ... , λ l−1 )
and
R sr (T ) (w1 , ... , wk−1 , λ 1 , ... , λ l+1 )
= T (w1 , ... , wr−1 , G −1 (λ s ), wr , ... , wk−1 , λ 1 , ... , λ s−1 , λ s+1 , ... , λ l+1 )
for w1 , ... , wk+1 ∈ V and λ 1 , ... , λ l+1 ∈ V ∗ .

Finally, we see that g is symmetric if and only if the matrix (gi j ) is symmetric,
i.e., we have that gi j = g ji for i, j = 1, ..., n.
7.3 Proposition Let B = {v1 , ..., vn } is a basis for a vector space V with dual
basis B ∗ = {v∗1 , ..., v∗n } , and let g be a metric on V and T ∈ T kl (V ) be a tensor so that
j ,..., j
g = ∑ g i j v∗i ⊗ v∗j and T = ∑ A i11 ,...,ikl v∗i1 ⊗ · · · ⊗ v∗ik ⊗ v j1 ⊗ · · · ⊗ v j l .
i, j i1 ,...,ik
j1 ,..., j l
Then the tensors L sr (T ) and R sr (T ) obtained by lowering and raising indices in T are
given by
j ,..., j l−1 ∗
L sr (T ) = ∑ B i11 ,...,ik+1 vi1 ⊗ · · · ⊗ v∗ik+1 ⊗ v j1 ⊗ · · · ⊗ v j l−1
i ,...,i
1 k+1
j1 ,..., j l−1
and
j ,..., j l+1 ∗
R sr (T ) = ∑ C i11 ,...,ik−1 vi1 ⊗ · · · ⊗ v∗ik−1 ⊗ v j1 ⊗ · · · ⊗ v j l+1 ,
i ,...,i
1 k−1
j1 ,..., j l+1
where
j ,..., j j ,..., j ,m, j ,..., j
B i11,...,ik+1
l−1
= ∑ g ir m A i11,...,ir−1
s−1 s l−1
,ir+1 ,...,ik+1
m
and
j ,..., j j ,..., j ,j ,..., j
C i11,...,ik−1
l+1
= ∑ g js m A i11,...,ir−1
s−1 s+1 l+1
,m,ir ,...,ik−1 .
m

7.4 If ψ : T 11 (V ) → T 1 (V ;V ) is the linear isomorphism defined in 5.15 and

2
T ∈ T (V ) , then we have that
T (v, w) = g(v, ψ ◦ R 11 (T ) (w)) = gt (ψ ◦ R 12 (T ) (v), w)
for every v, w ∈ V . Indeed, if B = {v1 , ..., vn } is a basis for V with dual basis B ∗ =
{v∗1 , ..., v∗n }, then C = {(gt ) ♯ (v∗1 ), ..., (gt ) ♯ (v∗n )} is also a basis for V with dual basis
C ∗ = {g ♭ (v1 ), ..., g ♭ (vn )} since
g ♭ (vi ) {(gt ) ♯ (v∗j )} = g (vi , (gt ) ♯ (v∗j )) = gt ((gt ) ♯ (v∗j ), vi ) = v∗j (vi ) = δi, j .
Hence we have that

n n
ψ ◦ R 11 (T ) (w) = ∑ R 11 (T ) (w, g ♭ (v j )) (gt ) ♯ (v∗j ) = ∑ T (v j , w) (gt ) ♯ (v∗j )
j=1 j=1
and
n n
ψ ◦ R 12 (T ) (v) = ∑ R 12 (T ) (v, g ♭ (v j )) (gt ) ♯ (v∗j ) = ∑ T (v, v j ) (gt ) ♯ (v∗j )
j=1 j=1
which completes the proof of the assertion.
7.5 Definition A covariant tensor field g ∈ T 2 (M) of degree 2 on a smooth

manifold M is said to be non-degenerate if g(p) is non-degenerate for every p ∈ M.
Let T be a tensor field of type (kl ) on M . Then we have tensor fields S1 and S2 on M
of type (k+1 k−1
l−1 ) and (l+1 ) , respectively, given by
S1 (p) = L sr (T (p)) and S2 (p) = R sr (T (p))
j ,..., j j ,..., j
for every p ∈ M. If (x,U) is a local chart on M, and if g i j , Ai11,...,ikl , Bi11,...,ik+1
l−1
and
j ,..., j
Ci11,...,ik−1
l+1
are the components of the local representations of g, T , S1 and S2 on U,
then it follows from Proposition 7.3 that
j ,..., j j ,..., j ,m, j ,..., j
B i11,...,ik+1
l−1
= ∑ g ir m A i11,...,ir−1
s−1 s l−1
,ir+1 ,...,ik+1
m
and
j ,..., j j ,..., j ,j ,..., j
C i11,...,ik−1
l+1
= ∑ g js m A i11,...,ir−1
s−1 s+1 l+1
,m,ir ,...,ik−1 ,
m
showing that S1 and S2 are in fact tensor fields. We say that they are obtained by
lowering and raising of indices in T , and they are denoted by L sr (T ) and R sr (T ),
respectively. If X is a vector field on M, then the covector field L 11 (X) is usually
denoted by g ♭ (X) or X ♭ . Similarly, if ω is a covector field on M, then the vector field
R 11 (ω ) is usually denoted by g ♯ (ω ) or ω ♯ .
PSEUDO-RIEMANNIAN MANIFOLDS
7.6 Now consider the linear map S : T k (V ) → T k (V ) defined by
1
Tσ
k! σ∑
S (T ) =
∈Sk
i.e.,
1
k! σ∑
S (T )(v1 , ..., vk ) = T (vσ (1) , ..., vσ (k) )
∈S k
which is called the symmetrization operator. If α , β ∈ V ∗ , we define their symmetric

product αβ by
αβ = S (α ⊗ β ) = 1 (α ⊗ β + β ⊗ α ) .
2
In particular, we have that αα = α ⊗ α which is usually denoted by α 2 .
If g : V × V → R is a symmetric bilinear functional, and if B = {v1 , ..., vn } is a
basis for V with dual basis B ∗ = {v∗1 , ..., v∗n }, then
g= ∑ g i j v∗i ⊗ v∗j = ∑ 21 ( g i j + g j i ) v∗i ⊗ v∗j

i, j i, j
= 1 ∑ g i j v∗i ⊗ v∗j + 12 ∑ g i j v∗j ⊗ v∗i = ∑ g i j v∗i v∗j .

2 i, j i, j i, j
7.7 Definition Let V be a finite dimensional vector space with a metric g. Then
two vectors v, w ∈ V are said to be orthogonal , and we write v ⊥ w, if g(v, w) = 0. If
S is a subset of V , then v ⊥ S means that v ⊥ w for every w ∈ S. We have a subspace
S⊥ of V , called the orthogonal space of S, which is given by
S⊥ = {v ∈ V |v ⊥ S } = g ♯ (A(S)) ,
where A(S) is the annihilator of S defined in Definition 4.19. When S = {w}, then S⊥
is denoted simply by w⊥ and is called the orthogonal space of w.
The norm of a vector v ∈ V is defined by kvk= |g(v, v)| 1/2 . A non-zero vector
v is called a unit vector if kvk = 1 and a null vector if kvk = 0 . A set of mutually
orthogonal vectors is called orthogonal , and it is called orthonormal if all the vectors
are unit vectors.
If W is a subspace of V , then the restriction g| W ×W is a symmetric bilinear
functional on W × W which is denoted simply by g| W . We say that W is a non-
degenerate subspace of V if g| W is non-degenerate.
7.8 Proposition Let W be a subspace of a finite dimensional vector space V with

a metric g. Then we have that
(1) dim W + dim W ⊥ = dim V

(2) (W ⊥ ) ⊥ = W
PROOF : (1) Follows from Proposition 4.21.
(2) Follows since W ⊂ (W ⊥ ) ⊥ , where
dim W = dim (W ⊥ ) ⊥ = dim V − dim W ⊥
according to (1).
7.9 Proposition A subspace W of a finite dimensional vector space V with a

metric g is non-degenerate if and only if
V = W ⊕W ⊥ .
We have that W is non-degenerate if and only if W ⊥ is non-degenerate.
PROOF : From Definition 7.7 we see that W is non-degenerate if and only if W ∩
W ⊥ = {0}. The first part of the proposition therefore follows from Proposition 7.8
(1) using the formula
dim (W + W ⊥ ) + dim (W ∩W ⊥ ) = dim W + dim W ⊥ = dim V .
The last part of the proposition follows from the first part and Proposition 7.8 (2).
7.10 Proposition An n-dimensional vector space V with a metric g always has

an orthonormal basis.
PROOF : We prove the proposition by induction on n. It is obviously true for n = 0,
and we will prove it for n > 0 assuming that it is true for every vector space V of
dimension n − 1.
There is a vector v ∈ V with a = g(v, v) 6= 0. Otherwise, if g(v, v) = 0 for every
v ∈ V , we would have that
g(v, w) = 1 [ g(v + w, v + w) − g(v, v) − g(w, w) ] = 0

2
for every v, w ∈ V , contradicting the fact that g is non-degenerate.
We let vn = |a| −1/2 v so that g(vn , vn ) = ±1 , and consider the subspace
W = {w ∈ V |g(vn , w) = 0}
which is of dimension n − 1 since it is the kernel of the non-zero functional F : V → R
defined by F(v) = g(vn , v).
Now g is non-degenerate on W , for suppose that u ∈ W with g(u, w) = 0 for
every w ∈ W . Since g(u, vn ) = 0, it follows that g(u, v) = 0 for every v ∈ V which
implies that u = 0. By the induction hypothesis there is therefore an orthonormal
basis {v1 , ..., vn−1 } for W , and B = {v1 , ..., vn } is a basis for V with the desired
properties.
7.11 Remark Let B = {v1 , ..., vn } be an orthonormal basis for a vector space
V with a metric g. Then the subspaces Vr = L(v1 , ..., vr ) are non-degenerate for r =
1, ... , n. Indeed, if v ∈ Vr with g(v, w) = 0 for every w ∈ Vr , then v = ∑ri=1 ai vi where
a j = g(v, v j )/g(v j , v j ) = 0 for j = 1, ... , r .
7.12 Proposition Let g be a metric on an n-dimensional vector space V , and

let k be the number of vectors vi in an orthonormal basis B = {v1 , ..., vn } for which
g(vi , vi ) = −1. Then k does not depend on the choice of orthonormal basis B, and
it is called the index of the metric g. It is also called the index of the vector space V
and is denoted by ind V .
PROOF : Since g is negative definite on the subspace {0} of V , we may choose a
subspace W of V of maximal dimension such that g is negative definite on W . We
will prove that k is the dimension of W .
We may assume that the vectors in B are ordered in such a way that g(vi , vi ) = −1
for 1 ≤ i ≤ k and g(vi , vi ) = 1 for k + 1 ≤ i ≤ n. Then V = V1 ⊕ V2 , where V1 and V2
are the subspaces of V spanned by {v1 , ..., vk } and {vk+1 , ..., vn }, respectively, and g
is negative definite on V1 and positive definite on V2 . Indeed, if
k
0 6= v = ∑ ai vi
i=1
we have that
k k
g(v, v) = ∑ g(vi , v j ) ai a j = − ∑ (ai )2 < 0 ,
i, j=1 i=1
showing that g is negative definite on V1 , and a similar argument applies to V2 . In

particular, it follows that dim (W ) ≥ dim (V1 ) = k.
To show that dim (W ) ≤ k, we consider the linear map F : V → V1 defined by
F(v) = v1 for v = v1 + v2 with v1 ∈ V1 and v2 ∈ V2 , and we let G = F|W : W →
V1 . Then ker (G) ⊂ W ∩ V2 which implies that ker (G) = {0}, since g is negative
definite on W and positive definite on V2 , and this completes the proof that dim (W ) ≤
dim (V1 ) = k.
7.13 Remark Let B = {v1 , ..., vn } be an orthonormal basis for an n-dimensional

vector space V with a metric g of index k. Then the n-tuple ε = (ε1 , ... , εn ) ∈
{−1, 1}n , where εi = g(vi , vi ) for i = 1, ... , n , is called the signature of B . An n-
tuple ε ∈ {−1, 1}n having k components which are −1 and n − k components which
are 1, is called a signature compatible with g.
Unless otherwise specified, we will assume that the vectors in an orthonormal
basis B are ordered so that

−1 for 1 ≤ i ≤ k
εi = .
1 for k + 1 ≤ i ≤ n
We call this ε the canonical signature for the metric g.

The metric g is positive definite, negative definite or indefinite when k = 0, k = n

or 0 < k < n, respectively. If the index of g is 1 and n ≥ 2, g is called a Lorentz metric.
7.14 Proposition Let W be a non-degenerate subspace of a finite dimensional

vector space V with a metric g. Then we have that
ind V = ind W + ind W ⊥ .
PROOF : By Propositions 7.9 and 7.10 there are orthonormal bases B = {v1 , ..., vr }
and C = {vr+1 , ..., vn } for W and W ⊥ where the number of vectors vi for which
g(vi , vi ) = −1 are ind W and ind W ⊥ , respectively. The proposition now follows
since {v1 , ..., vn } is an orthonormal basis for V .
7.15 Definition Given an n-tuple ε = (ε1 , ... , εn ) ∈ {−1, 1}n , there is a unique
metric g εn in Rn such that the standard basis E = {e1 , ..., en } is orthonormal with
signature ε . It is given by
n
g nε (a, b) = ∑ εi ai bi
i=1
for a, b ∈ Rn . The vector space Rn equipped with this metric is denoted by Rnε and is
called the pseudo-Euclidean space of dimension n and signature ε . If the index of the
metric g nε is k and ε is its canonical signature, then g nε and Rnε are also denoted by
g nk and Rnk , respectively. The matrix S = m E (g nε ) is the n × n diagonal matrix with
diagonal elements S ii = εi for i = 1, ... , n, and it is called the signature matrix for
the metric g nε . We have that
g nε (a, b) = at S b
for column vectors a and b in Rn . The space Rn1 where n ≥ 2 is called the n-
dimensional Minkowski space , and the space Rn equipped with the metric g n0 is
called the n-dimensional Euclidean space.
7.16 Definition By a fibre metric in a vector bundle π : E → M over a smooth

manifold M we mean a section g in the vector bundle π 2 : T 2 (E) → M so that g(p)
is a metric in the fibre E p for each p ∈ M. If the metric g(p) is Riemannian for each
p ∈ M, then g is called a Riemannian fibre metric in π : E → M.
If Fi : N → E for i = 1, 2 are liftings of a smooth map f : N → M from a smooth
manifold N, we often use < F1 , F2 > as an alternative notation for the function h :
N → R defined by h(p) = g( f (p)) (F1 (p), F2 (p)) for p ∈ N.
By a metric on a smooth manifold M we mean a fibre metric in its tangent bundle
π : T M → M, i.e., a covariant tensor field g ∈ T 2 (M) on M of degree 2 so that g(p) is
a metric in the tangent space Tp M for each p ∈ M. A manifold with a metric is called
a pseudo-Riemannian manifold. A vector field X along a smooth map f : N → M
from a smooth manifold N with kX(p)k = 1 for every p ∈ N is called a unit vector
field along f . If N is an open subset of M and f is the inclusion map, it is called a
unit vector field on N.
An immersion or an embedding f : M1 → M2 between two pseudo-Riemannian

manifolds M1 and M2 with metrics g1 and g2 is said to be isometric if f ∗ g2 = g1 . If
f : M1 → M2 is also a diffeomorphism, it is called an isometry . An (immersed) sub-
manifold M1 of a pseudo-Riemannian manifold M2 with metric g2 is called a pseudo-
Riemannian (immersed) submanifold if i∗ g2 is a metric on M1 , where i : M1 → M2 is
the inclusion map. If the codimension of M1 is 1, M1 is called a pseudo-Riemannian
hypersurface of M2 .
If the metric g(p) is Riemannian (resp., Lorentz) for each p ∈ M, then g is called
a Riemannian (resp. Lorentz) metric on M, and M is called a Riemannian (resp.
Lorentz) manifold.
7.17 Proposition Every vector bundle π : E → M has a Riemannian fibre metric.
PROOF : Let {(tα , π −1 (Uα )) | α ∈A} be a trivializing cover of E, and let {φα |α ∈A}
be a partition of unity on M subordinate to the open cover {Uα |α ∈ A}. For each
α ∈ A, we let gα be the section in π 2 : T 2 (E) → M defined by

φα (p) tα∗ ,p g n0 for p ∈ Uα
gα (p) = ,
0 for p ∈
/ Uα
where n is the dimension of the vector bundle and g n0 is the Riemannian metric in
Rn defined in Definition 7.15. Then g = ∑α ∈A gα is a Riemannian fibre metric in
π : E → M.
7.18 Proposition A vector bundle π : E → M with a fibre metric g over a smooth

manifold M n is equivalent over M to its dual bundle π ∗ : E ∗ → M, with the equiva-
lence e : E → E ∗ given by
e(v)(w) = g(p)(v, w)
for p ∈ M and v, w ∈ E p .
PROOF : Let E = {e1 , ..., en } be the standard basis for Rn , and let E ∗ = {e1 , ..., en }
be its dual basis. For every local trivialization (t, π −1 (U)) in E, the functions gi j :
U → R defined by
g(p) = ∑ gi j (p) (e i ◦ t p ) ⊗ (e j ◦ t p )
i, j
for p ∈ U, are smooth by Proposition 4.45. If (t ′ , π ∗ −1 (U)), where t p′ = r ◦ (t p−1 )∗

for p ∈ U, is the corresponding local trivialization in the dual bundle π ∗ : E ∗ → M
defined in Remark 4.9, we have that
(t p−1 )∗ ◦ e p ◦ t p−1 (ei )(e j ) = g(p)(t p−1 (ei ),t p−1 (e j )) = gi j (p) ,
so that
t p′ ◦ e p ◦ t p−1 (ei ) = ∑ gi j (p) e j
j
for p ∈ U and i = 1, ... , n. Hence it follows that

t ′ ◦ e ◦ t −1 (p, v) = (p ,t p′ ◦ e p ◦ t p−1 (v))
for p ∈ U and v ∈ Rn , where
mEE (t p′ ◦ e p ◦ t p−1 ) = (gi j (p)) ,
showing that e is an equivalence.
7.19 Definition If α and β are 1-forms on an open subset V of a smooth manifold

M, we define their symmetric product αβ to be the covariant tensor field on V of
degree 2 defined by
(αβ )(p) = α (p) β (p)
for p ∈ V .
7.20 Example Consider the smooth manifold M = R2 . A point p ∈ M with

cartesian coordinates (x, y) in the open set
V = {(x, y) ∈ R2 |y 6= 0 or x > 0}
can also be described by polar coordinates (r, φ ) in the open set
U = {(r, φ ) ∈ R2 |r > 0 and − π < φ < π } .
The connection between these coordinates is given by
x = r cos φ
y = r sin φ
and
p
r= x2 + y2

 cot−1 (x/y) − π for y < 0
φ= 0 for y = 0 and x > 0

cot−1 (x/y) for y > 0
As well as being real numbers, we can also think of x, y and r, φ as functions of p ,

so that they are the coordinate functions of local charts (id, M) and (ψ ,V ) on M. The
meaning will always be clear from the context. Here ψ = i ◦ ρ −1 , where ρ : U → V
is the diffeomorphism given by
ρ (r, φ ) = (r cos φ , r sin φ ) ,
and i : U → R2 is the inclusion map. We have that

cos φ −r sin φ
Dρ (r, φ ) =
sin φ r cos φ
so that
det Dρ (r, φ ) = r > 0
for (r, φ ) ∈ U . Proposition 2.80 now implies that
∂ ∂y
= ∂∂ xr ∂ + ∂ r ∂ = cos φ ∂ + sin φ ∂
∂r ∂x ∂y ∂x ∂y
and
∂ ∂y
= ∂∂φx ∂ + ∂ φ ∂ = − r sin φ ∂ + r cos φ ∂ ,
∂φ ∂x ∂y ∂x ∂y
and by Proposition 4.18 we also have that
dx = ∂∂ xr dr + ∂∂φx d φ = cos φ dr − r sin φ d φ

and
∂y ∂y
dy = ∂ r dr + ∂ φ d φ = sin φ dr + r cos φ d φ
on V . The standard metric on R2 is given by
g = dx2 + dy2
in cartesian coordinates. Using polar coordinates, we therefore have that
g|V = dr2 + r2 d φ 2 .

cartesian coordinates (x, y, z) in the open set
V = {(x, y, z) ∈ R3 |y 6= 0 or x > 0}
can also be described by spherical coordinates (r, θ , φ ) in the open set
U = {(r, θ , φ ) ∈ R3 |r > 0 , 0 < θ < π and − π < φ < π } .
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ
and
p
r = x2 + y2 + z2
p
θ = cos−1 ( z / x2 + y2 + z2 )

 cot−1 (x/y) − π for y < 0
φ= 0 for y = 0 and x > 0

As well as being real numbers, we can also think of x, y, z and r, θ , φ as functions

of p , so that they are the coordinate functions of local charts (id, M) and (ψ ,V ) on
M. The meaning will always be clear from the context. Here ψ = i ◦ ρ −1 , where
ρ : U → V is the diffeomorphism given by
ρ (r, θ , φ ) = (r sin θ cos φ , r sin θ sin φ , r cos θ ) ,

 
sin θ cos φ r cos θ cos φ −r sin θ sin φ
Dρ (r, θ , φ ) =  sin θ sin φ r cos θ sin φ r sin θ cos φ 
cos θ −r sin θ 0
so that
det Dρ (r, θ , φ ) = (cos θ )(r2 cos θ sin θ ) + (r sin θ )(r sin2 θ ) = r2 sin θ > 0
for (r, θ , φ ) ∈ U . Proposition 2.80 now implies that
∂ ∂y
= ∂∂ xr ∂ + ∂ r ∂ + ∂∂ rz ∂ = sin θ cos φ ∂ + sin θ sin φ ∂ + cos θ ∂ ,
∂r ∂x ∂y ∂z ∂x ∂y ∂z
∂ ∂y
= ∂∂θx ∂ + ∂ θ ∂ + ∂∂θz ∂ = r cos θ cos φ ∂ + r cos θ sin φ ∂ − r sin θ ∂
∂θ ∂x ∂y ∂z ∂x ∂y ∂z
and
∂ ∂y
= ∂∂φx ∂ + ∂ φ ∂ + ∂∂φz ∂ = − r sin θ sin φ ∂ + r sin θ cos φ ∂ ,
∂φ ∂x ∂y ∂z ∂x ∂y
and by Proposition 4.18 we also have that
dx = ∂∂ xr dr + ∂∂θx d θ + ∂∂φx d φ = sin θ cos φ dr + r cos θ cos φ d θ − r sin θ sin φ d φ ,
∂y ∂y ∂y
dy = ∂ r dr + ∂ θ d θ + ∂ φ d φ = sin θ sin φ dr + r cos θ sin φ d θ + r sin θ cos φ d φ
and
dz = ∂∂ rz dr + ∂∂θz d θ + ∂∂φz d φ = cos θ dr − r sin θ d θ
g = dx2 + dy2 + dz2
in cartesian coordinates. Using spherical coordinates, we therefore have that
g|V = dr2 + r2 (d θ 2 + sin2 θ d φ 2 ) .

cartesian coordinates (w, x, y, z) in the open set
V = {(w, x, y, z) ∈ R4 |y 6= 0 or x > 0}
can also be described by hyperspherical coordinates (r, χ , θ , φ ) in the open set
U = {(r, χ , θ , φ ) ∈ R4 |r > 0 , 0 < χ < π , 0 < θ < π and − π < φ < π } .
w = r cos χ
x = r sin χ sin θ cos φ
y = r sin χ sin θ sin φ
z = r sin χ cos θ
and
p
r= w2 + x2 + y2 + z2
p
χ = cos−1 ( w / w2 + x2 + y2 + z2 )
p
θ = cos−1 ( z / x2 + y2 + z2 )

 cot−1 (x/y) − π for y < 0
φ= 0 for y = 0 and x > 0

As well as being real numbers, we can also think of w, x, y, z and r, χ , θ , φ as func-

tions of p , so that they are the coordinate functions of local charts (id, M) and (ψ ,V )
on M. The meaning will always be clear from the context. Here ψ = i ◦ ρ −1 , where
ρ (r, χ , θ , φ ) = (r cos χ , r sin χ sin θ cos φ , r sin χ sin θ sin φ , r sin χ cos θ ) ,

Dρ (r, χ , θ , φ ) =
 
cos χ −r sin χ 0 0
sin χ sin θ cos φ r cos χ sin θ cos φ r sin χ cos θ cos φ −r sin χ sin θ sin φ 
 
 sin χ sin θ sin φ r cos χ sin θ sin φ r sin χ cos θ sin φ r sin χ sin θ cos φ 
sin χ cos θ r cos χ cos θ −r sin χ sin θ 0
so that
det Dρ (r, χ , θ , φ ) = (cos χ )(r3 cos χ sin2 χ sin θ ) − (−r sin χ )(r2 sin3 χ sin θ )
= r3 sin2 χ sin θ > 0
for (r, χ , θ , φ ) ∈ U . Proposition 4.18 now implies that
dw = cos χ dr − r sin χ d χ ,
dx = sin χ sin θ cos φ dr + r cos χ sin θ cos φ d χ

+ r sin χ cos θ cos φ d θ − r sin χ sin θ sin φ d φ ,
dy = sin χ sin θ sin φ dr + r cos χ sin θ sin φ d χ

+ r sin χ cos θ sin φ d θ + r sin χ sin θ cos φ d φ
and
dz = sin χ cos θ dr + r cos χ cos θ d χ − r sin χ sin θ d θ
g = dw2 + dx2 + dy2 + dz2
in cartesian coordinates. Using hyperspherical coordinates, we therefore have that
g|V = dr2 + r2 (d χ 2 + sin2 χ (d θ 2 + sin2 θ d φ 2 )) .
By Proposition 2.41 we know that the 3-sphere
S3a = {(w, x, y, z) ∈ R4 |w2 + x2 + y2 + z2 = a2 }
with radius a is a closed submanifold of the Euclidean space R4 . Let i : S3a → R4 be

the inclusion map and p : R × R3 → R3 be the projection on the last factor, and let
Va = V ∩ S3a . Then (p ◦ ψ ◦ i,Va ) is a local chart on S3a with coordinate functions
χ ◦ i, θ ◦ i and φ ◦ i which we also denote simply by χ , θ and φ . The pull-back to Va
of the standard metric g on R4 is given by
i∗ g|Va = a2 (d χ 2 + sin2 χ (d θ 2 + sin2 θ d φ 2 )) .

7.23 Example Consider the smooth manifold

p
M = {(w, x, y, z) ∈ R4 |w > x2 + y2 + z2 } .
A point p ∈ M with cartesian coordinates (w, x, y, z) in the open set
V = {(w, x, y, z) ∈ M|y 6= 0 or x > 0}
can also be described by pseudospherical coordinates (r, χ , θ , φ ) in the open set
U = {(r, χ , θ , φ ) ∈ R4 |r > 0 , χ > 0 , 0 < θ < π and − π < φ < π } .
w = r cosh χ
x = r sinh χ sin θ cos φ
y = r sinh χ sin θ sin φ
z = r sinh χ cos θ
and
p
r= w2 − x2 − y2 − z2
p
χ = cosh−1 ( w / w2 − x2 − y2 − z2 )
p
θ = cos−1 ( z / x2 + y2 + z2 )

 cot−1 (x/y) − π for y < 0
φ= 0 for y = 0 and x > 0

As well as being real numbers, we can also think of w, x, y, z and r, χ , θ , φ as func-

tions of p , so that they are the coordinate functions of local charts (id, M) and (ψ ,V )
on M. The meaning will always be clear from the context. Here ψ = i ◦ ρ −1 , where
ρ (r, χ , θ , φ ) = (r cosh χ , r sinh χ sin θ cos φ , r sinh χ sin θ sin φ , r sinh χ cos θ ) ,
Dρ (r, χ , θ , φ ) =
 
cosh χ r sinh χ 0 0
sinh χ sin θ cos φ r cosh χ sin θ cos φ r sinh χ cos θ cos φ −r sinh χ sin θ sin φ 
 
 sinh χ sin θ sin φ r cosh χ sin θ sin φ r sinh χ cos θ sin φ r sinh χ sin θ cos φ 
sinh χ cos θ r cosh χ cos θ −r sinh χ sin θ 0
so that
det Dρ (r, χ , θ , φ ) = (cosh χ )(r3 cosh χ sinh2 χ sin θ ) − (r sinh χ )(r2 sinh3 χ sin θ )
= r3 sinh2 χ sin θ > 0
for (r, χ , θ , φ ) ∈ U . Proposition 4.18 now implies that
dw = cosh χ dr + r sinh χ d χ ,
dx = sinh χ sin θ cos φ dr + r cosh χ sin θ cos φ d χ

+ r sinh χ cos θ cos φ d θ − r sinh χ sin θ sin φ d φ ,
dy = sinh χ sin θ sin φ dr + r cosh χ sin θ sin φ d χ

+ r sinh χ cos θ sin φ d θ + r sinh χ sin θ cos φ d φ
and
dz = sinh χ cos θ dr + r cosh χ cos θ d χ − r sinh χ sin θ d θ
on V . The Lorentz metric on R41 is given by
g 41 = − dw2 + dx2 + dy2 + dz2
in cartesian coordinates. Using pseudospherical coordinates, we therefore have that
g 41 |V = − dr2 + r2 (d χ 2 + sinh2 χ (d θ 2 + sin2 θ d φ 2 )) .
By Proposition 2.41 we know that the 3-hyperboloid
H3a = {(w, x, y, z) ∈ R4 |w2 − x2 − y2 − z2 = a2 }
where a > 0 is a closed submanifold of the Minkowski space R41 . Let i : H3a → R4
be the inclusion map and p : R × R3 → R3 be the projection on the last factor, and
let Va = V ∩ H3a . Then (p ◦ ψ ◦ i,Va ) is a local chart on the upper sheet of H3a with
coordinate functions χ ◦ i, θ ◦ i and φ ◦ i which we also denote simply by χ , θ and
φ . The pull-back to Va of the Lorentz metric g 41 on R41 is given by
i∗ g 41 |Va = a2 (d χ 2 + sinh2 χ (d θ 2 + sin2 θ d φ 2 )) .
THE HODGE STAR OPERATOR

7.24 Let V be a finite dimensional vector space with a metric g. Identifying V ∗∗
with V as in Remark 4.5 by means of the natural isomorphism iV : V → V ∗∗ given
by iV (v)(λ ) = λ (v) for v ∈ V and λ ∈ V ∗ , we have an isomorphism H = iV ◦ G −1 :
V ∗ → V ∗∗ . Hence we obtain a non-degenerate bilinear functional h : V ∗ × V ∗ → R

given by
h(λ , µ ) = H(λ )(µ ) = µ (G −1 (λ ))
for λ , µ ∈ V ∗ . If B = {v1 , ..., vn } is a basis for V with dual basis B ∗ = {v∗1 , ..., v∗n },
then the components of h with respect to B ∗ are given by
n
h(v∗i , v∗j ) = v∗j ( ∑ g ik vk ) = g i j
k=1
for i, j = 1, ..., n, so that

h = ∑ g i j vi ⊗ v j .
i, j
If g is symmetric, so is h, and therefore h is a metric on V ∗ which is said to be induced

by the metric g on V . We have that
h(G(v), G(w)) = G(w)(v) = g(w, v) = g(v, w)
for v, w ∈ V , which shows that G : V → V ∗ is a linear isometry with respect to the

metrics g and h. Hence if g is positive definite (resp., negative definite), the same is
true for h.
The metric h may be extended to a symmetric bilinear functional h k : T k (V ) ×
T (V ) → R so that
k
1
h k ( λ 1 ⊗ · · · ⊗ λ k , µ 1 ⊗ · · · ⊗ µ k) = h ( λ 1, µ 1) · · · h ( λ k, µ k) (1)
k!
for λ 1 , ... , λ k , µ 1 , ... , µ k ∈ V ∗ . Since this implies that
1
h k ( λ 1 ⊗ · · · ⊗ λ k , µ 1 ⊗ · · · ⊗ µ k) = µ ⊗ · · · ⊗ µ k ( G −1 (λ 1 ), ... , G −1 (λ k )) ,
k! 1
(2)
it follows that
r
hk(λ 1 ⊗ ··· ⊗ λ k , ∑ ci µ i1 ⊗ · · · ⊗ µ ik ) = 0
i=1
for every λ 1 , ... , λ k ∈V∗ if and only if

r
∑ ci µ i1 ⊗ · · · ⊗ µ ik = 0 ,
i=1
so that h k is well defined and non-degenerate.

From (2) and Proposition 4.74 it also follows that
1
h k ( λ 1 ⊗ · · · ⊗ λ k , µ 1 ⊗ · · · ⊗ µ k) = C(R( µ 1 ⊗ · · · ⊗ µ k ) ⊗ λ 1 ⊗ · · · ⊗ λ k ) ,
k!
where C : T kk (V ) → R is the k-fold contraction in the i-th contravariant and the i-th
covariant index for 0 ≤ i ≤ k, and R : T k0 (V ) → T 0k (V ) is the linear isomorphism

given by
R(T )(λ 1 , ... , λ k ) = T ( G −1 (λ 1 ), ... , G −1 (λ k ))
for T ∈ T k0 (V ) and λ 1 , ... , λ k ∈ V ∗ , raising all the indices of T as defined above.
Hence we have that
1
h k ( T, S) = C(R( T ) ⊗ S)
k!
for T, S ∈ T k (V ) . If
T= ∑ A i1 ,...,ik v∗i1 ⊗ · · · ⊗ v∗ik
i1 ,...,ik
then
R(T ) = ∑ A i1 ,...,ik vi1 ⊗ · · · ⊗ vik where A i1 ,...,ik = ∑ gi1 j1 · · · gik jk A j1 ,..., jk .

i1 ,...,ik j1 ,..., jk
Hence, if
S= ∑ B j1 ,..., jk v∗j1 ⊗ · · · ⊗ v∗jk ,
j1 ,..., jk
we have that
1
h k ( T, S) = ∑ A i1 ,...,ik B i1 ,...,ik .
k! i1 ,...,ik
We will show that the restriction e

h k : Λk (V )×Λk (V ) → R of hk is non-degenerate,
k
and therefore is a metric on Λ (V ). By (1) and Proposition 5.13 (1) we have that
1
h k ( λ 1 ∧ · · · ∧ λ k , µ 1 ⊗ · · · ⊗ µ k) =
k! ∑ ε (σ ) h ( λ σ (1) , µ 1 ) · · · h ( λ σ (k) , µ k )
σ ∈Sk

h ( λ 1, µ 1) h ( λ 1, µ 2) ··· h ( λ 1, µ k )

1 h ( λ 2 , µ 1 ) h ( λ 2 , µ 2 ) ··· h ( λ 2, µ k )

= .. .. ..
k! . . .

h ( λ k, µ ) h ( λ k, µ ) ··· h ( λ k, µ k)
1 2
for λ 1 , ... , λ k , µ 1 , ... , µ k ∈ V ∗ . Since the last expression is skew symmetric as a

function of µ 1 , ..., µ k , it follows from Proposition 5.8 (2) that

h(λ 1 , µ 1 ) h(λ 1 , µ 2 ) · · · h(λ 1 , µ k )

h(λ 2 , µ 1 ) h(λ 2 , µ 2 ) · · · h(λ 2 , µ k )
e
h k (λ 1 ∧ · · · ∧ λ k , µ 1 ∧ · · · ∧ µ k ) = .. .. .. . (3)
. . .

h(λ k , µ ) h(λ k , µ ) · · · h(λ k , µ )
1 2 k
Now let B = {v1 , ..., vn } be an orthonormal basis for V with dual basis B ∗ =
{v∗1 , ..., v∗n }. Then B ∗ is an orthonormal basis for V ∗ , since G is a linear isometry
and G(vi ) = g(vi , vi ) v∗i for i = 1, ..., n. From (3) it follows that
D = {v∗i1 ∧ · · · ∧ v∗ik | 1 ≤ i1 < ... < ik ≤ n}
is an othonormal basis for Λk (V ), which completes the proof that e

h k is non-
degenerate and therefore is a metric on Λk (V ).
7.25 Proposition Let (V, µ ) be an n-dimensional oriented vector space with a

metric g. Then there is a unique volume element ω in V compatible with µ such that
ω = v∗1 ∧ · · · ∧ v∗n for any positively oriented orthonormal basis (v1 , ..., vn ) in V . ω is
called the metric volume element in V compatible with µ .
If (w1 , ..., wn ) is any positively oriented basis in V , we have that
1/2
ω = |g | w∗1 ∧ · · · ∧ w∗n
where g = det (g(wi , w j )) .

PROOF : Let {v1 , ... , vn } and {w1 , ... , wn } be two positively oriented bases for V with
w j = ∑ni=1 ai j vi for j = 1, ... , n. Then we have that
g(wi , w j ) = ∑ ari as j g(vr , vs )

rs
for i, j = 1, ... , n, which means that
C = At BA
for the matrices A = (ai j ), B = (g(vi , v j )) and C = (g(wi , w j )). Since the bases are
equally oriented, we have that det(A) > 0 , and
v∗1 ∧ · · · ∧ v∗n = det(A) w∗1 ∧ · · · ∧ w∗n
by Corollary 5.18.
Suppose now that both bases are orthonormal. Then det(B) = det(C) = (−1)k ,
where k is the index of the metric g, and it follows that det(A) = 1 which proves the
first part of the proposition.
If on the other hand only the basis {v1 , ... , vn } is assumed to be orthonormal, then
we obtain det(A) = | det(C) | 1/2 which proves the last part.
7.26 The antisymmetric Levi–Civita symbol is defined by


 1 if (i1 , ... , in ) is an even permutation
εi1 ... in = −1 if (i1 , ... , in ) is an odd permutation

0 otherwise
Let V be an n-dimensional oriented vector space with a metric g of index k. If

(v1 , ..., vn ) is a positively oriented basis in V , it follows from Proposition 5.13 (1)
and 7.25 that the metric volume element ω in V is given by
ω= ∑ ai1 ,...,in v∗i1 ⊗ · · · ⊗ v∗in

i1 ,...,in
with components
1/2
ai1 ... in = | g | εi1 ... in ,
where g = det (g(vi , v j )) . Raising the indices we have that
−1/2
ai1 ... in = | g | ε i1 ... in
where ε i1 ... in = (−1)k εi1 ... in .
7.27 Let (V, µ ) be an n-dimensional oriented vector space with a metric g, and
let ε be the metric volume element in V compatible with µ . If 0 ≤ k ≤ n, we have a
linear isomorphism R : T k0 (V ) → T 0k (V ) given by
R(T )(λ 1 , ... , λ k ) = T (g ♯ (λ 1 ), ... , g ♯ (λ k ))
for T ∈ T k0 (V ) and λ 1 , ... , λ k ∈ V ∗ , raising all the indices of T as defined in 7.24. We

see that R maps Λk (V ) onto Λk (V ∗ ). Furthermore, let C : T nk (V ) → T n−k 0 (V ) be the
k-fold contraction in the i-th contravariant and the i-th covariant index for 0 ≤ i ≤ k.
We can now define a linear map
∗ : Λk (V ) → Λn−k (V ) ,
called the Hodge star operator , by

1
∗ω = C(R( ω ) ⊗ ε )
k!
for ω ∈ Λk (V ). Let B = {v1 , ..., vn } be a basis for V with dual basis B ∗ = {v∗1 , ..., v∗n },
and suppose that
ω = ∑ ai1 ,...,ik v∗i1 ∧ · · · ∧ v∗ik
i1 <...<ik
and
ε = c1,...,n v∗1 ∧ · · · ∧ v∗n .
It will be useful to express the forms ω and ε in terms of tensor products of the basis
vectors from B ∗ , instead of wedge products. By Proposition 4.26 we have that
ω= ∑ ai1 ,...,ik v∗i1 ⊗ · · · ⊗ v∗ik

i1 ,...,ik
and
ε= ∑ ci1 ,...,in v∗i1 ⊗ · · · ⊗ v∗in ,
i1 ,...,in
where the new coefficients ai1 ,...,ik = ω (vi1 , ..., vik ) and ci1 ,...,in = ε (vi1 , ..., vin ) are
skew symmetric in their indices, and agree with the old ones when the indices are
strictly increasing. Now we have that
R(ω ) = ∑ a i1 ,...,ik vi1 ⊗ · · · ⊗ vik where a i1 ,...,ik = ∑ gi1 j1 · · · gik jk a j1 ,..., jk ,

i1 ,...,ik j1 ,..., jk
so that
∗ω = ∑ b j1 ,..., jn−k v∗j1 ⊗ · · · ⊗ v∗jn−k
j1 ,..., jn−k
with
1
b j1 ,..., jn−k = ∑ ai1 ,...,ik ci1 ,...,ik , j1 ,..., jn−k .
k! i1 ,...,ik
As the coefficients ai1 ,...,ik

and b j1 ,..., jn−k are also skew symmetric in their indices, it
follows from Proposition 5.13 (1) that
∗ω = ∑ b j1 ,..., jn−k v∗j1 ∧ · · · ∧ v∗jn−k

j1 <...< jn−k
where
b j1 ,..., jn−k = ∑ ai1 ,...,ik ci1 ,...,ik , j1 ,..., jn−k .
i1 <...<ik
n
In particular, if ω = g ♭ (v) for a vector v = ∑ ai vi , then
i=1
!
n
∗ω = ∑ i
∑ a c i, j 1 ,..., j n−1 v∗j 1∧ · · · ∧ v∗j n−1
j 1 <...< j n−1 i=1
n
= ∑ (−1) i+1 c 1,...,n ai v∗1∧ · · · ∧ v∗i−1 ∧ v∗i+1 ∧ · · · ∧ v∗n
i=1
n
= c 1,...,n ∑ (−1) i+1 i v (v∗i ) v∗1∧ · · · ∧ v∗i−1 ∧ v∗i+1 ∧ · · · ∧ v∗n = i v (ε )
i=1
by Proposition 5.74.
If C = {w1 , ..., wn } is an orthonormal basis for V with dual basis C ∗ =
{w∗1 , ..., w∗n } , we have that
∗( w∗i1 ∧ · · · ∧ w∗ik ) = g(wi1 , wi1 ) · · · g(wik , wik ) w∗j1 ∧ · · · ∧ w∗jn−k (1)
if
w∗i1 ∧ · · · ∧ w∗ik ∧ w∗j1 ∧ · · · ∧ w∗jn−k = ε . (2)
Indeed, if
wi r for 1 ≤ r ≤ k
vr = ,
w j r−k for k + 1 ≤ r ≤ n
then B = {v1 , ..., vn } is a positively oriented orthonormal basis for V , and we have
that
∗( v∗1 ∧ · · · ∧ v∗k ) = b k+1,...,n v∗k+1 ∧ · · · ∧ v∗n
where
b k+1,...,n = a1,...,k = g(v1 , v1 ) · · · g(vk , vk ) .
In particular, we have that
∗1 = ε and ∗ ε = (−1) s
where s is the index of the metric g.

By Proposition 5.12 (3) it follows from (2) that
w∗j1 ∧ · · · ∧ w∗jn−k ∧ (−1) k(n−k) w∗i1 ∧ · · · ∧ w∗ik = ε ,
so that (1) implies that
∗∗( w∗i1 ∧ · · · ∧ w∗ik ) = (−1) k(n−k)+s w∗i1 ∧ · · · ∧ w∗ik .
Hence we have that

∗∗ ω = (−1) k(n−k)+s ω
for ω ∈ Λk (V ).
From (1) it follows that
( w∗i1 ∧ · · · ∧ w∗ik ) ∧ ∗( w∗i1 ∧ · · · ∧ w∗ik ) = < w∗i1 ∧ · · · ∧ w∗ik , w∗i1 ∧ · · · ∧ w∗ik > ε ,
and we have that
( w∗r1 ∧ · · · ∧ w∗rk ) ∧ ∗( w∗i1 ∧ · · · ∧ w∗ik ) = 0
and
< w∗r1 ∧ · · · ∧ w∗rk , w∗i1 ∧ · · · ∧ w∗ik > = 0
when {r1 , ..., rk } 6= {i1 , ..., ik }. Hence it follows from Proposition 5.14 that
ω ∧ ∗ η = < ω,η > ε
for ω , η ∈ Λk (V ).
7.28 Definition Let M n be an oriented pseudo-Riemannian manifold with a met-

ric g of index s. Then we have a linear map
∗ : Ω k (M) → Ω n−k (M) ,
called the Hodge star operator , defined by
(∗ ω )(p) = ∗ ω (p)
for p ∈ M , ω ∈ Ω k (M) and k = 0, ..., n. We let Ω k (M) = {0} and ∗ : Ω k (M) →

Ω n−k (M) be the zero map when k < 0 or k > n.
We now define the codifferential operator δ : Ω k (M) → Ω k−1 (M) by
δ = (−1) n(k+1)+s+1 ∗ d ∗ ,
and the Laplace–Beltrami operator ∆ : Ω k (M) → Ω k (M) by
∆ = δ d+d δ
for k = 0, ..., n.
7.29 Let M n be a compact oriented pseudo-Riemannian manifold. Then we have

a metric on the vector space Ω k (M) of k-forms on M defined by
Z
< ω,η > = ω ∧ ∗η
M
for ω , η ∈ Ω k (M) and k = 0, ..., n. The metric may also be extended to Ω(M) =
n
⊕ k=0 Ω k (M) by requiring the various Ω k (M) to be orthogonal.
7.30 Proposition Let M n be a compact oriented pseudo-Riemannian manifold

with a metric g of index s. Then δ is the adjoint of d , and ∆ is self-adjoint on Ω(M) ,
i.e.,
< d ω,η > = < ω,δ η >
and
< ∆ω,η > = < ω,∆η >
for ω , η ∈ Ω(M) .
PROOF : To prove the first formula, we only need to consider the case where ω ∈
Ω k−1 (M) and η ∈ Ω k (M) for k = 1, ..., n. By Proposition 5.58 we have that
d ( ω ∧ ∗ η ) = d ω ∧ ∗ η + (−1) k−1 ω ∧ d ∗ η .
Since d ∗ η ∈ Ω n−k+1 (M) , it follows from 7.27 that
(−1) k−1 d ∗ η = (−1) k−1 (−1) (n−k+1)(k−1) + s ∗∗ d ∗ η
= (−1) n(k−1) − k(k−1) + s ∗∗ d ∗ η = (−1) n(k+1) + s ∗∗ d ∗ η

= − ∗ δη .
Combining this and using Stokes’ theorem, we conclude that

Z Z
0= d (ω ∧ ∗ η ) = (d ω ∧ ∗ η − ω ∧ ∗ δη ) = < d ω , η > − < ω , δ η >
M M
which completes the proof of the first formula in the proposition. The second formula
follows immediately from the first.
7.31 Let f : M1 → M2 be a smooth map between two n-dimensional smooth

manifolds M1 and M2 with volume elements ω1 and ω2 . Then the smooth function
J : M1 → R defined by
f ∗ ω2 = J ω1
is called the Jacobian function of f . For each point p ∈ M1 , the number |J(p)| mea-
sures the rate of change of volume near p under the mapping f . The map f is said to
be volume preserving if J = 1 so that f ∗ ω2 = ω1 .
7.32 Let M n be a smooth manifold with a volume element ω . Then a vector field
X on M with global flow γ : D(X) → M is called incompressible if the diffeomor-
phism γ t : Dt (X) → D−t (X) is volume preserving for each t ∈ R. By Proposition
4.90 this is equivalent to the assertion that
LX ω = 0 .
In order to measure the deviation of a vector field X on M from being incompressible,

we define the ω -divergence div ω X of X with respect to the volume element ω by
L X ω = (div ω X ) ω .
If (x,U) is a local chart on M and

n
ω | U = h dx1∧ ... ∧ dxn and X | U = ∑ ai ∂ i ,
i=1
∂x
then
n
(div ω X ) | U = 1
h i=1
∑ ∂∂xi (hai )
since
n
(LX ω )|U = X(h)dx1 ∧ ... ∧ dxn + ∑ hdx1 ∧ ... ∧ LX (dxi ) ∧ ... ∧ dxn
i=1
n n
∂h ∂ ai
= ∑ ai i dx1 ∧ ... ∧ dxn + ∑ h i dx1 ∧ ... ∧ dxn
i=1 ∂ x i=1 ∂ x
n
= ∑ ∂∂xi (hai )dx1 ∧ ... ∧ dxn
i=1
by Proposition 4.76 and 4.79 (2).
7.33 Proposition Let (M n , µ ) be an oriented pseudo-Riemannian manifold with

a metric g, and let ε be the metric volume element on M compatible with µ . If X is a
vector field on M, we have that
div ε X = ∗ d ∗(X ♭ ) = − δ (X ♭ ) .
In this case div ε X is usually denoted simply by div X .


L X ε = d (i X ε ) = d ∗(X ♭ )
by 7.27 and Proposition 5.83.
7.34 Definition Let M n be a pseudo-Riemannian manifold with a metric g. Then

we define the gradient of a smooth function f : M → R by
grad f = (d f ) ♯ .
If (x,U) is a local chart on M, we have that

n
∂ f i
d f |U = ∑ ∂ x i dx
i=1
so that
∂ f ∂
grad f | U = ∑ gij ∂ x j ∂ xi
.
i, j
7.35 Remark Let M n be an oriented pseudo-Riemannian manifold with a metric

g, and let f : M → R be a smooth function. Then we have that
∆ f = δ (d f ) = − div(grad f ) .
Hence the Laplace-Beltrami operator has the opposite sign of the ordinary Laplacian
on functions
∇ 2 f = div(grad f ) .
7.36 Let (M, µ ) be a 3-dimensional oriented Riemannian manifold with a metric

g, and let ε be the metric volume element on M compatible with µ . Then we define
the curl of a vector field X on M by
curl X = {∗ d (X ♭ )} ♯
so that we have a commutative diagram
grad curl div

F (M) ✲ T (M) ✲ T (M) ✲ F (M)
1 1
φ1 φ2 φ3 φ4
❄ ❄ ❄ ❄
d ✲ 1 d ✲ d ✲
Ω 0 (M) Ω (M) Ω 2 (M) 3
Ω (M)
where φ 1 : F (M) → Ω 0 (M) , φ 2 : T1 (M) → Ω 1 (M) , φ 3 : T1 (M) → Ω 2 (M) and

φ 4 : F (M) → Ω 3 (M) are the linear isomorphisms defined by
φ 1 ( f ) = f , φ 2 (X) = X ♭ , φ 3 (X) = ∗(X ♭ ) and φ 4 ( f ) = ∗ f
for f ∈ F (M) and X ∈ T1 (M) . Since d ◦ d = 0 , it follows in particular that
curl(grad f ) = 0 and div(curl X ) = 0
for every f ∈ F (M) and X ∈ T1 (M) . By 7.27 we also have that
i curlX ε = d (X ♭ ) .
7.37 Example Consider the smooth manifold M = R3 with its standard orien-
tation µ so that the local chart (id, M) is positively oriented, and with its standard
metric
g = dx2 + dy2 + dz2
in cartesian coordinates x, y and z which are the coordinate functions of id. Then the
metrical volume element is given by
ε = dx ∧ dy ∧ dz .
For a smooth function f : M → R we have that

∂f ∂f ∂f
d f = ∂ x dx + ∂ y dy + ∂ z dz .
Since (g i j ) is the identity matrix, it follows that
∂f ∂f ∂f
grad f = (d f ) ♯ = ∂ x ∂ + ∂ y ∂ + ∂ z ∂ .
∂x ∂y ∂z
Let X be a vector field on M with
X = ax ∂ + ay ∂ + az ∂
∂x ∂y ∂z
so that
X ♭ = ax dx + ay dy + az dz .
Then we have that

♭ ∂ az ∂ ay ∂ ax ∂ az ∂ ay ∂ ax
d (X ) = ∂ y − ∂ z dy ∧ dz + ∂ z − ∂ x dz ∧ dx + ∂ x − ∂ y dx ∧ dy
which implies that

∂ az ∂ ay ∂ ax ∂ az ∂ ay ∂ ax
∗ d (X ♭ ) = ∂y − ∂z dx + ∂ z − ∂ x dy + ∂x − ∂y dz ,
thus showing that

♭ ♯ ∂ az ∂ ay ∂ ∂a ∂a ∂ ∂ ay ∂ ax ∂
curl X = {∗ d (X )} = ∂y − ∂z + ∂ zx − ∂ xz + ∂x − ∂y .
∂x ∂y ∂z
Now identifying the tangent space of R3 at an arbtrary point with R3 as described in

Lemma 2.84, we obtain the usual expressions

∂f ∂f ∂f
grad f = ∂ x , ∂ y , ∂ z
and

i j k

∂ az ∂ ay ∂ ax ∂ az ∂ ay ∂ ax
curl X = = ∂ ∂ ∂ .
∂y − ∂z , ∂z − ∂x , ∂x − ∂y ∂x ∂y ∂ z

ax ay az
By 7.32 we have that

∂a ∂ ay ∂a
div X = ∂ xx + ∂ y + ∂ zz
and
2
∂ 2 ax ∂ ay ∂ 2 az
∇ 2f = 2 + + 2 .
∂x ∂ y2 ∂z
7.38 Example Let (M 3 , µ ) be an oriented Riemannian manifold with a metric

g, and suppose that (ψ , O) is a positively oriented local chart on M with coordinate
functions u, v and w so that
g | O = 12 du2 + 12 dv2 + 12 dw2 ,

U V W
where U, V and W are positive smooth functions on O. Then the vector fields
eu = U ∂ , ev = V ∂ and ew = W ∂
∂u ∂v ∂w
form an orthonormal local basis B for T M on O, with dual basis B ∗ consisting of
eu = 1 du , ev = 1 dv and ew = 1 dw .
U V W
The metrical volume element is given by
1
ε = eu ∧ ev ∧ ew = du ∧ dv ∧ dw .
UVW
For a smooth function f : O → R we have that
∂f ∂f ∂f ∂f ∂f ∂f
d f = ∂ u du + ∂ v dv + ∂ w dw = U ∂ u eu + V ∂ v ev + W ∂ w ew .
Since mB (g) is the identity matrix at each point, it follows that

∂f ∂f ∂f
grad f = (d f ) ♯ = U ∂ u eu + V ∂ v ev + W ∂ w ew .
Let X be a vector field on O with
X = a u eu + a v ev + a w ew = a u U ∂ + a v V ∂ + a w W ∂
∂u ∂v ∂w
so that
au a a
X ♭ = a u eu + a v ev + a w ew = du + v dv + w dw .
U V W
Then we have that
h i h i
d (X ♭ ) = ∂∂v w − ∂∂w v dv ∧ dw + ∂∂w u − ∂∂u w dw ∧ du
a a a a
W V U W
h i h i
+ ∂∂u v − ∂∂v u du ∧ dv = VW ∂∂v w − ∂∂w v ev ∧ ew
a a a a
V U W V
h i h i
+ WU ∂∂w u − ∂∂u w ew ∧ eu + UV ∂∂u v − ∂∂v u eu ∧ ev
a a a a
U W V U
which implies that

h i h i
∗ d (X ♭ ) = VW ∂∂v w − ∂∂w v eu + WU ∂∂w u − ∂∂u w ev
a a a a
W V U W
h i
+ UV ∂∂u v − ∂∂v u ew ,
a a
V U
thus showing that

h i
curl X = {∗ d (X ♭ )} ♯ = VW ∂∂v w − ∂∂w v eu
a a
W V
h i h i
+ WU ∂∂w u − ∂∂u w ev + UV ∂∂u v − ∂∂v u ew .
a a a a
U W V U
By 7.32 we have that

h i
div X = UVW ∂∂u + ∂∂v + ∂∂w
au av aw
VW WU UV
h i
∂f ∂f ∂f
∇ 2 f = div(grad f ) = UVW ∂∂u U ∂ u + ∂∂v V ∂ v + ∂∂w W ∂ w .
VW WU UV
7.39 Example We are going to apply the formulae in Example 7.38 to the ori-
ented Riemannian manifold M = R3 and the spherical coordinate system (ψ , O) with
coordinate functions r, θ and φ . By Example 7.21 this coordinate system is positively

oriented with respect to the standard orientation in R3 , and we have that
g|O = dr2 + r2 d θ 2 + r2 sin2 θ d φ 2
so that
U = 1 , V = 1r and W = r sin 1
θ .
Given a smooth function f : O → R and a vector field X on O with
X = a r er + a θ eθ + a φ eφ ,
we have that
∂f ∂f ∂f
grad f = ∂ r er + 1r ∂ θ eθ + r sin 1
θ ∂ φ eφ ,
h i
1 ∂ ∂ aθ
curl X = r sin θ ∂θ ( sin θ a φ ) − ∂ φ er
h i h i
1 ∂ ar ∂ 1 ∂ ∂ ar
+ r sin θ ∂φ − sin θ ∂r (r a φ ) e θ + r ∂r (r a θ ) − ∂ θ eφ ,
∂ aφ
div X = 12 ∂∂r (r2 ar ) + r sin
1 ∂ 1
θ ∂ θ (sin θ aθ ) + r sin θ ∂ φ and
r

∂f ∂f ∂2 f
∇ 2 f = 12 ∂∂r r2 ∂ r + 2 1 ∂
sin θ ∂θ + 2 2
1
.
r ∂θ
r sin θ r sin θ ∂ φ 2
7.40 Proposition Let (M, µ ) be an oriented pseudo-Riemannian manifold with a

metric g, and let Pn−1 be an immersed pseudo-Riemannian submanifold of M n with
orientation ν and metric i∗ (g), where i : P → M is the inclusion map. Then there is a
unique vector field N : P → T M along i, called the positive unit normal field on P,
so that N(p) is a positive unit normal vector to i∗ (Tp P) in Tp M for each p ∈ P.
PROOF : The map N : P → T M is clearly uniquely defined, so we only need to show
that it is smooth. Let ω and η be volume elements on M and P compatible with
the orientations µ and ν , respectively. Now let p 0 ∈ P, and choose an orthonor-
mal basis C = {v1 , ..., vn } for Tp 0 M so that v j = i∗ (w j ) for j = 1, ... , n − 1 , where
B = {w1 , ..., wn−1 } is an orthonormal basis for Tp 0 P . We can also assume that
ω (p 0 )(vn , v1 , ..., vn−1 ) > 0 and η (p 0 )(w1 , ..., wn−1 ) > 0 so that N(p 0 ) = vn .
By Corollary 2.35 there are local charts (x,U) and (y,V ) around p 0 on P and M,
respectively, such that U = {q ∈V |yn (q) = 0} and x1 (q) = y1 (q) , . . . , xn−1 (q) =
yn−1 (q) for q ∈U. Hence y ◦ i = j 1 ◦ x , where j 1 : Rn−1 → Rn−1 × R is the linear
map given by j 1 (a) = (a, 0) for a ∈ Rn−1 . This implies that
−1 −1
ty,p ◦ i∗ p = j 1 ◦ tx,p and i∗ p ◦ tx,p = ty,p ◦ j1
for p ∈ U by the commutative diagram in 2.70, where (tx , π ′ −1 (U)) and (ty , π −1 (V ))
are the local trivializations in the tangent bundles π ′ : T P → P and π : T M → M

associated with the local charts (x,U) and (y,V ), respectively.
We have vector fields Y1′ , ... ,Yn′ along i|U where {Y1′ (p), ... ,Yn′ (p)} is a basis for
Tp M for each p ∈ U and Yi′ (p 0 ) = vi for i = 1, ... , n , given by
Yi′ (p) = ty,p

−1
◦ ty,p 0 (vi )
for p ∈ U and i = 1, ... , n. We also have that Y j′ = i∗ ◦ X j′ for j = 1, ... , n − 1 , where
Xi′ (p) = tx,p

−1
◦ tx,p 0 (wi )
for p ∈ U and i = 1, ... , n − 1, so that {X1′ (p), ... , Xn−1

′
(p)} is a basis for Tp P for
each p ∈ U and Xi′ (p 0 ) = wi for i = 1, ... , n − 1 .
Let W be an open neighbourhood
of p 0 on P contained in U so that the i × i -
′ ′
matrix g(p)(Yα (p),Yβ (p)) , where 1 ≤ α , β ≤ i, is non-singular for each p ∈ W
and 1 ≤ i ≤ n. Then the subspace Vi (p) of Tp M spanned by { Y1′ (p), ... ,Yi′ (p) } is
non-degenerate for p ∈ W and 1 ≤ i ≤ n. Using this, we are going to construct vector
fields Y1 , ... ,Yn along i|W so that {Y1 (p), ... ,Yn (p)} is an orthonormal basis for Tp M
for each p ∈ W and Yi (p 0 ) = vi for i = 1, ... , n .
By applying the Gram-Schmidt process to { Y1′ (p), ... ,Yn′ (p) } , we obtain an
orthogonal basis { Y1′′ (p), ... ,Yn′′ (p) } for Tp M with Yi′′ (p 0 ) = vi for i = 1, ... , n ,
given by
Y1′′ (p) = Y1′ (p)
and
i−1 g(p) (Y ′ (p) ,Y ′′ (p))
Yi′′ (p) = Yi′ (p) − Y j′′ (p)
i j
∑ ′′ ′′
j=1 g(p) (Y j (p) ,Y j (p))
for p ∈ W and i = 2, ... , n. Indeed, assuming inductively that { Y1′′ (p), ... ,
′′ (p) } is an orthogonal basis for V ′′ / Vi−1 (p) ⊥
Yi−1 i−1 (p) , we have that Y j (p) ∈
so that g(p)(Y j′′ (p),Y j′′ (p)) 6= 0 for j = 1, ... , i − 1 since Vi−1 (p) is a non-
degenerate subspace of Tp M , thus showing that Yi′′ (p) is well defined. Further-
more, g(p)(Yi′′ (p),Yr′′ (p)) = 0 for r = 1, ... , i − 1 which shows that { Y1′′ (p), ... ,
Yi′′ (p) } is an orthogonal basis for Vi (p) . Hence we have an orthonormal basis
{ Y1 (p), ... ,Yn (p) } for Tp M with Yi (p 0 ) = vi for i = 1, ... , n , given by
Yi (p) = Yi′′ (p) / kYi′′ (p)k
for p ∈ W and i = 1, ... , n. We also have that Y j = i∗ ◦ X j for j = 1, ... , n − 1 , where

the orthonormal basis { X1 (p), ... , Xn−1 (p) } for Tp P is obtained in the same way by
using the Gram-Schmidt process on { X1′ (p), ... , Xn−1′ (p) }.
Now there is an open neighbourhood O of p 0 contained in W so that

ω (p)(Yn (p),Y1 (p), ...,Yn−1 (p)) > 0 and η (p)(X1 (p), ..., Xn−1 (p)) > 0 for p ∈ O.
This implies that N(p) = Yn (p) for p ∈ O, proving that N is smooth at p 0 . As p 0
was an arbitrary point on P, this shows that N is a vector field along i, thus complet-
ing the proof of the proposition.
7.41 Proposition Let (M, µ ) be a 1-dimensional oriented Riemannian manifold.

Then there is a unique vector field τ on M , called the positive unit tangent field , so
that (τ (p)) is a positively oriented orthonormal basis in Tp M for each p ∈ M.
PROOF : The map τ : M → T M is clearly uniquely defined, and it is also smooth. For
if (x,U) is a positively oriented local chart on M, then we have that
∂
τ |U = ∂x / k ∂∂x k .
7.42 Proposition Let P n−1 be a Riemannian immersed submanifold of an ori-

ented Riemannian manifold (M n , µ ) , and let i : P → M be the inclusion map. If
N : P → T M is the positive unit normal field on P with respect to an orientation ν of
P and ω ∈ Ω 1 (M) is a 1-form on M , we have that
i ∗ (∗ ω ) = ( ω ◦ i)(N) ε
where ε is the metric volume element on P compatible with ν .

PROOF : Let B = (v2 , ..., vn ) be a positively oriented orthonormal basis for the tan-
gent space Tp P at a point p ∈ P. Then C = (N p , i ∗ v2 , ..., i ∗ vn ) is a positively oriented
orthonormal basis for Tp M , and we let C ∗ = (ω1 , ..., ωn ) be the dual basis in Tp∗ M .
Using the dual basis B ∗ = ((i ∗ p ) ∗ ω2 , ..., (i ∗ p ) ∗ ωn ) of B in Tp∗ P , we have that
ε (p) = (i ∗ p ) ∗ ( ω2 ∧ · · · ∧ ωn ) .
Since (i ∗ p ) ∗ ( ω1 ) = 0, this implies that
(i ∗ p ) ∗ (∗ ω k ) = δ 1k ε (p) = ω k (N p ) ε (p)
for k = 1, ..., n, which shows the formula in the proposition.
7.43 Gauss’ Divergence Theorem Let (M n , µ ) be an oriented Riemannian man-

ifold with boundary, and let ε and eε be the metric volume elements on M and ∂ M
compatible with µ and ∂ µ , respectively. Let N : ∂ M → T M be the outward unit nor-
mal field on ∂ M, and i : ∂ M → M be the inclusion map. If X is a vector field on M
with compact support, then we have that
Z Z
( div X ) ε = < X ◦ i, N > e
ε.
M ∂M
PROOF : By Theorem 6.38 and Propositions 7.33 and 7.42 we have that
Z Z Z Z
( div X ) ε = ∗d ∗(X ♭ ) ε = d ∗(X ♭ ) = i ∗ (∗X ♭ )
M M M ∂M
Z Z
= (X ◦ i) ♭ (N) e
ε= < X ◦ i, N > e
ε.
∂M ∂M
7.44 The Classical Stokes’ Theorem Let M 2 be a compact manifold with bound-
ary in R3 with its standard metric <, > and orientation, and let N : M → T R3 be the
positive unit normal field on M with respect to an orientation µ of M. Let dA and
dS be the metric volume elements on M and ∂ M compatible with µ and ∂ µ , respec-
tively, and let i : M → R3 and j : ∂ M → R3 be the inclusion maps. If T : ∂ M → T R3
is the vector field along j defined by T = j ∗ ◦ τ , where τ is the positive unit tangent
field on ∂ M , and X is a vector field on an open subset U of R3 containing M, then
we have that Z Z
< curl X ◦ i, N > dA = < X ◦ j, T > dS .
M ∂M
PROOF : By Theorem 6.38 and Proposition 7.42 we have that

Z Z Z
< curl X ◦ i, N > dA = (∗ d (X ♭ ) ◦ i)(N) dA = i ∗ (∗∗ d (X ♭ ))
M M M
Z Z Z Z
= i ∗ (d (X ♭ )) = d (i ∗ ( X ♭ )) = j ∗( X ♭) = j ∗ ( X ♭ )(τ ) dS
M M ∂M ∂M
Z Z
= (X ◦ j) ♭ (T ) dS = < X ◦ j, T > dS.
∂M ∂M
TIME DEPENDENT VECTOR FIELDS
7.45 Definition Let π : T M → M be the tangent bundle of a smooth manifold M,

and let V be an open subset of M. By a time-dependent vector field on V we mean a
smooth map X : J × V → T M such that π ◦ X = π2 , where J is an open interval and
π2 : J × V → M is the map given by π2 (t, p) = p for (t, p) ∈ J × V . This means that
the map X t : V → T M defined by X t (p) = X(t, p) for p ∈ V , is a vector field on V for
each t ∈ J.
With X we may associate a (time independent) vector field Xe : J ×V → T (J ×M) ,
called the suspension of X, given by
!

e p) = i p ∗
X(t, d + i t ∗ ◦ X(t, p)
dr
t
for (t, p) ∈ J × V , where i p : J → J × M and it : M → J × M are the embeddings

defined by i p (t ′ ) = (t ′ , p) and it (p′ ) = (t, p′ ) for t ′ ∈ J and p′ ∈ V , and (r, J) is
the standard local chart on J where r : J → R is the inclusion map. We see that
π2 ∗ ◦ Xe = X .
7.46 Definition Let X be a time-dependent vector field on a smooth manifold

M over the open time interval J. A smooth curve γ : I → M defined on an open

subinterval I of J is called an integral curve for X if
γ ′ (t) = X(t, γ (t)) (1)
for t ∈ I. If I contains 0 and γ (0) = p0 , the point p0 is called the starting point or
initial condition of γ . The integral curve γ is called maximal if it has no extension to
an integral curve for X on any larger open interval.
7.47 Proposition Let X be a time-dependent vector field on a smooth manifold

M over the open time interval J, and let Xe : J × M → T (J ×M) be its suspension.
Then a smooth curve γ : I → M defined on an open subinterval I of J containing s is
an integral curve for X with γ (s) = p if and only if the curve β : I − s → J × M given
by
β (t − s) = (t, γ (t))
for t ∈ I, is an integral curve for Xe with initial condition (s, p).

PROOF : Follows from Remark 2.83 and Proposition 2.74 which implies that
!
d

β ′ (t − s) = iγ (t) ∗ dr + i t ∗ ◦ γ ′ (t)
t
and !

d
e β (t − s)) = iγ (t) ∗
X( + i t ∗ ◦ X(t, γ (t))
dr
t
for t ∈ I.

M over the open time interval J. For each point (s, p) on J × M we denote by γ(s,p) :
I(s, p) → M the maximal integral curve for X with γ(s,p) (s) = p. The set
D(X) = {(t, s, p) ∈ J × J × M|t ∈ I(s, p)}
is called the domain of the time-dependent flow for X, and the (global) time-
dependent flow for X is the map γ : D(X) → M defined by γ (t, s, p) = γ(s,p) (t) for
(s, p) ∈ J × M and t ∈ I(s, p).
7.49 Corollary Let X be a time-dependent vector field on a smooth manifold M

over the open time interval J, and let Xe : J × M → T (J ×M) be its suspension. If
γ : D(X) → M is the global time-dependent flow for X and β : D(X) e → J × M the
global flow for X,e and if Λ : R × R × M → R × R × M is the diffeomorphism given
by Λ (t, s, p) = (t − s, s, p) for t, s ∈ R and p ∈ M, then
e
Λ ( D(X)) = D(X)
and
β ◦ Λ (t, s, p) = (t, γ (t, s, p))
for (t, s, p) ∈ D(X) . In particular, D(X) is an open subset of J × J × M containing
∆J × M, where ∆J ⊂ J × J is the diagonal, and the time-dependent flow γ for X is
smooth. If π2 : J × M → M is the projection on the second factor, we have that
π2 ◦ β ◦ Λ = γ .
7.50 Definition If X is a time-dependent vector field on a smooth manifold M

over the open time interval J with global time-dependent flow γ : D(X) → M, and if
t, s ∈ J, we define
Dt,s (X) = {p ∈ M|(t, s, p) ∈ D(X)}
and let γt,s : Dt,s (X) → M be the map defined by γt,s (p) = γ (t, s, p).
7.51 Corollary Let X be a time-dependent vector field on a smooth manifold M

over the open time interval J, and let Xe : J × M → T (J ×M) be its suspension. If
γ : D(X) → M is the global time-dependent flow for X and β : D(X) e → J × M the
e and if i s : M → J × M is the embedding given by i s (p) = (s, p) for
global flow for X,
p ∈ M, then
e ∩ ( {s} × M )
i s ( Dt,s (X)) = D t−s (X)
and
β t−s ◦ i s = i t ◦ γt,s
for t, s ∈ J. If π2 : J × M → M is the projection on the second factor, we have that
π2 ◦ β t−s ◦ i s = γt,s .

M over the open time interval J. For each s,t, u ∈ J we have that
(1) Dt,s (X) is open in M.
(2) If p ∈ Dt,s (X) and γt,s (p) ∈ Du,t (X), then p ∈ Du,s (X) and γu,t (γt,s (p)) = γu,s (p).
In particular the domain of γu,t ◦ γt,s is contained in Du,s (X), and it equals
Du,s (X) if t is between s and u.
(3) γt,s (Dt,s (X)) = Ds,t (X), and γt,s is a diffeomorphism onto its image with inverse
γs,t .
PROOF : Follows from Corollary 7.51 and Proposition 3.42.

M over an open time interval J containing 0. For each point p on M we denote by
α p : I(p) → M the maximal integral curve for X with initial condition p. The set
D ′ (X) = {(t, p) ∈ J × M|t ∈ I(p)}
is called the domain of the flow for X, and the (global) flow for X is the map α :
D ′ (X) → M defined by α (t, p) = α p (t) for p ∈ M and t ∈ I(p).
7.54 Remark If γ : D(X) → M is the global time-dependent flow for X and

i : J × M → J × J × M is the embedding given by i (t, p) = (t, 0, p) for t ∈ J and
p ∈ M, then
i ( D ′ (X)) = D(X) ∩ ( J × {0} × M ) and α = γ ◦ i .
In particular, D ′ (X) is an open subset of J × M containing {0} × M, and the flow α

for X is smooth.
If β is the global flow for the suspension Xe : J × M → T (J ×M) of X and Λ :
R × R × M → R × R × M is the diffeomorphism defined in Corollary 7.49, then
i = Λ ◦ i so that
e ∩ ( J × {0} × M )
i ( D ′ (X)) = D(X)
and
β ◦ i (t, p) = (t, α (t, p))
for (t, p) ∈ D ′ (X) . If π2 : J × M → M is the projection on the second factor, we have
that
π2 ◦ β ◦ i = α .
7.55 Definition Let X be a time-dependent vector field on a smooth manifold M

over an open time interval J containing 0. If α : D ′ (X) → M is the global flow for X,
and if t ∈ J, we define
Dt (X) = {p ∈ M|(t, p) ∈ D ′ (X)}
and let αt : Dt (X) → M be the map defined by αt (p) = α (t, p).
7.56 Remark If γ : D(X) → M is the global time-dependent flow for X, then
D t (X) = D t,0 (X) and α t = γ t,0
for t ∈ J. Hence, if β is the global flow for the suspension Xe : J × M → T (J ×M) of

X and i s : M → J × M is the embedding given by i s (p) = (s, p) for s ∈ J and p ∈ M,
then
e ∩ ( {0} × M )
i 0 ( Dt (X)) = D t (X)
and
β t ◦ i 0 = i t ◦ αt
for t ∈ J. If π2 : J × M → M is the projection on the second factor, we have that
π2 ◦ β t ◦ i 0 = αt .
7.57 Definition Let πlk : Tlk (M) → M be the tensor bundle of type (kl ) over a
smooth manifold M. By a time-dependent tensor field of type (kl ) on M we mean a
smooth map T : J × M → Tlk (M) such that πlk ◦ T = π2 , where J is an open interval
containing 0, and π2 : J × M → M is the projection on the second factor. This means
that the map Tt : M → Tlk (M) defined by Tt (p) = T (t, p) for p ∈ M, is a tensor field
of type (kl ) on M for each t ∈ J.
7.58 Definition Let X be a time-dependent vector field on a smooth manifold M

over the open time interval J with global time-dependent flow γ : D(X) → M, and let
T be a covariant tensor field of degree k on M.
If p is a point on M and s ∈ J, then we define the Lie derivative of T at (s, p)
with respect to X to be

LX T (s, p) = lim 1 [ γ t,s
∗ d
T (p) − T (p) ] = dt γ t,s
∗
T (p) ,
t→s t−s
s
using the vector space topology on T k (Tp M) defined in Proposition 13.117 in the
appendix.
7.59 Proposition Let X be a time-dependent vector field on a smooth manifold M

over the open time interval J with suspension Xe : J × M → T (J ×M) , and let T be a
covariant tensor field of degree k on M. Then the Lie derivative LX T is a well-defined
covariant time-dependent tensor field of degree k on M given by
(LX T ) s = i s∗ ( LXe ( π2∗ T ))
for s ∈ J, where i s : M → J × M is the embedding defined by i s (p) = (s, p) for p ∈ M,

and π2 : J × M → M is the projection on the second factor.
If γ : D(X) → M is the global time-dependent flow for X and c : I(s, p) →
∗ T (p), then we have that
T k (Tp M) is the curve defined by c (t) = γ t,s
c ′ (t) = γ t,s
∗
(LX T ) t (p)
for every t ∈ I(s, p) if we canonically identify Tc(t) T k (Tp M) with T k (Tp M) as in

Lemma 2.84.
PROOF : Let β : D(X) e → J × M be the global flow for X,

e and let b : I(s, p) − s →
T k (T(s,p) (J × M)) be the curve defined by
∗
b(t − s) = β t−s ( π2∗ T )(s, p)
for t ∈ I(s, p). Since γt,s = π2 ◦ β t−s ◦ i s by Corollary 7.51, it follows that
∗
c(t) = i s∗ ( β t−s ( π2∗ T ))(p) = (i s∗ p ) ∗ ( β t−s
∗
( π2∗ T ))(s, p) = (i s∗ p ) ∗ ◦ b(t − s)
for t ∈ I(s, p), where (i s∗ p ) ∗ : T k (T(s,p) (J × M)) → T k (Tp M) is the linear map
Defined in Definition 4.52. This shows that the curve c is smooth, and if we
canonically identify Tc(s) T k (Tp M) with T k (Tp M) and Tb(0) T k (T(s,p) (J × M)) with
T k (T(s,p) (J × M)) as in Lemma 2.84, we have that
(LX T ) s (p) = c′ (s) = (i s∗ p ) ∗ ◦ b′ (0) = i s∗ ( LXe ( π2∗ T ))(p) .
Since
b′ (t − s) = β t−s
∗
( LXe ( π2∗ T ))(s, p)
and β t−s ◦ i s = i t ◦ γt,s by Proposition 4.90 and Corollary 7.51, it also follows that
c′ (t) = (i s∗ p) ∗ ◦ b′ (t − s) = i s∗ (β t−s
∗
( LXe ( π2∗ T )))(p)
∗
= γ t,s (i t∗ ( LXe ( π2∗ T )))(p) = γ t,s
∗
(LX T ) t (p).

M over the open time interval J, and let ω be a k-form on M. Then we have a time-
dependent (k − 1)-form iX ω on M, called the interior product of X and ω , defined
by
(iX ω )(t, p) = iX(t,p) ω (p)
for each p ∈ M and t ∈ J. We see that
(iX ω ) t = i Xt ω
for each t ∈ J.

M over the open time interval J with suspension Xe : J × M → T (J ×M) , and let ω
be a k-form on M. Then we have that
i Xt ω = i t∗ ( iXe ( π2∗ ω ))
for t ∈ J, where i t : M → J × M is the embedding defined by i t (p) = (t, p) for p ∈ M,

and π2 : J × M → M is the projection on the second factor.
PROOF : Using Proposition 5.76 we have that
i t∗ ( iXe ( π2∗ ω ))(p) = (i t ∗ p ) ∗ ( iXe ( π2∗ ω ))(t, p)

= (i t ∗ p ) ∗ iX(t,p)
e ( π2 ∗ (t,p) ) ∗ ω (p) = (i t ∗ p ) ∗ ( π2 ∗ (t,p) ) ∗ i π2 ∗ ◦ X(t,p)
e ω (p)
= ((π2 ◦ i t ) ∗ p ) ∗ i X(t,p) ω (p) = ( i Xt ω )(p)
for every p ∈ M.

M over the open time interval J, and let ω be a k-form on M . Then we have that
(LX ω ) t = d ( i Xt ω ) + i Xt (d ω )
for t ∈ J.
PROOF : If X e : J × M → T (J ×M) is the suspension of X, it follows from Proposi-

tions 7.59, 7.61, 5.83 and 5.59 that
(LX ω ) t = i t∗ ( LXe ( π2∗ ω )) = i t∗ { d ( iXe ( π2∗ ω )) + iXe ( d ( π2∗ ω )) }

= d ( i t∗ ( iXe ( π2∗ ω ))) + i t∗ ( iXe ( π2∗ (d ω ))) = d ( i Xt ω ) + i Xt (d ω )
for t ∈ J.
SYMPLECTIC MANIFOLDS
7.63 Definition A non-degenerate, closed 2-form ω ∈ Ω 2 (M) is called a sym-
plectic form on M. A smooth manifold M with a symplectic form ω is called a
symplectic manifold, and it is also denoted by (M, ω ) . If (M1 , ω1 ) and (M2 , ω2 ) are
two sympectic manifolds, then a smooth map f : M1 → M2 is said to be symplectic
or to be a canonical transformation if f ∗ ω2 = ω1 .
7.64 Proposition Let (M1 , ω1 ) and (M2 , ω2 ) be sympectic manifolds, and let
π1 : M1 × M2 → M1 and π2 : M1 × M2 → M2 be the projections on the first and second
factor. Then
Ω = π1∗ ω1 − π2∗ ω2
is a symplectic form on M1 × M2 . A smooth map f : M1 → M2 with graph G( f ) is
symplectic if and only if
i∗ Ω = 0 ,
where i : G( f ) → M1 × M2 is the inclusion map.
PROOF : As
Ω (p, q) ( iq ∗ (v1 ) + i p ∗ (v2 ), iq ∗ (w1 ) + i p ∗ (w2 ))
= ω1 (p)(v1 , w1 ) − ω2 (q)(v2 , w2 )
for v1 , w1 ∈ Tp M1 and v2 , w2 ∈ Tq M2 , the 2-form Ω is non-degenerate. By Proposition
5.59 it is also closed, and Ω is therefore a symplectic form on M1 × M2 . The last part
of the proposition follows since
i ∗ Ω (p, f (p)) ( i f (p) ∗ (v) + i p ∗ ◦ f ∗ (v), i f (p) ∗ (w) + i p ∗ ◦ f ∗ (w))
= ( ω1 − f ∗ ω2 )(p)(v, w)
for v, w ∈ Tp M1 .
7.65 Proposition Let V be an n-dimensional vector space, and let ω ∈ Λ2 (V ) be

a skew symmetric bilinear functional of rank r. Then r = 2 m for an integer m, and
there is a basis B = {v1 , ..., vn } for V such that
 
0 Im 0
mB (ω ) =  − Im 0 0  ,
0 0 0
where Im is the m × m identity matrix. If B ∗ = {v∗1 , ..., v∗n } is the dual basis of B, then
we have that
m
ω = ∑ v∗i ∧ v∗i+m .
i=1
PROOF : We prove the first part of the propsition by induction on n. It is true for
n = 0 and n = 1 since ω = 0 in this case, so that m = 0 and we can let B be any basis
for V . We will now prove it for n > 1 assuming that it is true for every vector space
V of dimension less than n.
The assertion is obviously true if ω = 0. Otherwise, there are vectors vm , v2m ∈ V
with a = ω (vm , v2m ) 6= 0. Dividing vm by a, we may assume that ω (vm , v2m ) = 1. Let
V1 = L(vm , v2m ) be the subspace of V spanned by vm and v2m , and let
W1 = {w ∈ V | ω (w, v) = 0 for every v ∈ V1 } .
Then we have that V = V1 ⊕ W1 . Indeed, if v ∈ V , then
v − ω (v, v2m ) vm + ω (v, vm ) v2m ∈ W1
which shows that V = V1 + W1 , and we clearly have that V1 ∩ W1 = {0} .
The restriction ω1 : W1 × W1 → R of ω is a skew symmetric bilinear func-
tional of rank r − 2. By the induction hypothesis there is therefore a basis B1 =
{v1 , ..., vm−1 , vm+1 , ..., v2m−1 , v2m+1 , ..., vn } for W1 with
 
0 Im−1 0
mB1 (ω1 ) =  − Im−1 0 0 ,
0 0 0
and B = {v1 , ..., vn } is then a basis for V satisfying the first part of the proposition.
To prove the last part of the proposition, we use that
v∗i ∧ v∗i+m (v j , vk ) = v∗i (v j ) v∗i+m (vk ) − v∗i+m (v j ) v∗i (vk ) = δi, j δi+m,k − δi+m, j δi,k
for i = 1, ..., m and j, k = 1, ..., n , which follows from Proposition 5.13 (1). This
implies that
m
∗ ∗ δ −δ for 1 ≤ j, k ≤ 2m
∑ vi ∧ vi+m (v j , vk ) = 0 j+m,k k+m, j otherwise
i=1
which completes the proof of the last formula in the proposition.
7.66 Corollary Let V be an n-dimensional vector space, and let ω ∈ Λ2 (V ) be a

skew symmetric bilinear functional. Then ω is a symplectic form on V if and only if
n = 2k for an integer k, and the k-th exterior power ω k = ω ∧ · · · ∧ ω 6= 0 .
PROOF : If r = 2 m is the rank of ω , and B = {v1 , ..., vn } is the basis for V given in
Proposition 7.65, then we have that
m
ωm = ∑ v∗i1 ∧ v∗i1 +m ∧ · · · ∧ v∗im ∧ v∗im +m
i1 ,...,im =1
= ∑ v∗σ (1) ∧ vσ∗ (1)+m ∧ · · · ∧ vσ∗ (m) ∧ v∗σ (m)+m

σ ∈Sm
= (−1) m (m−1) / 2 ∑ v∗σ (1) ∧ · · · ∧ v∗σ (m) ∧ v∗σ (1)+m ∧ · · · ∧ vσ∗ (m)+m
σ ∈Sm
= m ! (−1) m (m−1) / 2 v∗1 ∧ · · · ∧ v∗2m 6= 0.
Each term in the third sum is obtained by moving successively the basis vectors
vσ∗ (2) , ..., v∗σ (m) in the dual basis leftwords and collecting them on the left hand side
of the exterior product, resulting in 1 + ... + (m − 1) = m (m − 1)/ 2 transpositions
as each basis vector v∗σ ( j+1) is moved successively to the left of the j basis vectors
v∗σ ( j)+m , ..., v∗σ (1)+m for j = 1, ..., m − 1.
Now, if ω is symplectic, then r = n and the last assertion in the proposition is
satisfied with k = m. On the other hand, if ω is not symplectic, then r < n. In this
case, either n is odd, or n = 2k for an integer k > m so that ω k = 0 which follows
from the above formula for ω m .
7.67 Proposition Let ω ∈ Ω 2 (M) be a 2-form on a smooth manifold M n . Then

ω is non-degenerate if and only if n = 2k for an integer k, and the k-th exterior power
ω k = ω ∧ · · · ∧ ω is a nowhere vanishing n-form on M. In particular, a manifold M
with a non-degenerate 2-form ω must be orientable.
7.68 Darboux’ Theorem Let ω ∈ Ω 2 (M) be a non-degenerate 2-form on a

2k
smooth manifold M . Then d ω = 0 if and only if there is a local chart (x,U) around
each point p ∈ M with x(p) = 0 and
k
ω |U = ∑ dxi ∧ dxi+k .
i=1
PROOF : Let (y,V ) be a local chart around p with y(p) = 0, and suppose that the
form ω is closed and is given locally by
ω |V = ∑ ω i j dyi ∧ dy j .
i< j
We let η 0 = ω |V and η 1 be the 2-form with constant local representation on V given

by
η 1 (q) = ∑ ω i j (p) dyi (q) ∧ dy j (q)
i< j
for q ∈ V , and we let η be the time-dependent 2-form on V defined on an open time

interval J containing [ 0, 1 ] by
e
ηt = η0 +t η
for t ∈ J, where η e = η1−η0.

Since η t (p) = ω (p) which is non-degenerate for t ∈ J, there is for each such t an
open subinterval It of J containing t and an open neighbourhood Ut of p contained
in V such that η is non-degenerate on It × Ut . As the interval [ 0, 1 ] is compact, it is
S
covered by a finite number of intervals I for i = 1, ..., n. Let I ′ = ni=1 Iti , and choose
Tn ti
an open ball Br (0) contained in y( i=1 Uti ) . By the Poincaré lemma, there is a 1-
form θ on U ′ = y−1 (Br (0)) such that η e = d θ , and we may assume that θ (p) = 0 .
Now, since η is non-degenerate on I ′ ×U ′ , we have a time-dependent vector field
X on U ′ over the time interval I ′ , where Xt = − (η t ) ♯ (θ ) is obtained from − θ by
raising indices for each t ∈ I ′ . Let α : D ′ (X) → M be the flow for X. As Xt (p) = 0 for
t ∈ I ′ , we have that α (t, p) = p for every t ∈ I ′ which shows that I ′ × {p} ⊂ D ′ (X) .
Using that D ′ (X) is open and the interval [ 0, 1 ] is compact, it follows in the same
way as above that there is an open subinterval I of I ′ containing [ 0, 1 ] and an open
neighbourhood U of p contained in U ′ with I × U ⊂ D ′ (X) .
Fix a q ∈ U, and consider the curve c : I → Λ2 (Tq M) given by c(t) = α t∗ η t (q)
for t ∈ I. If we canonically identify T c(t) Λ2 (Tq M) with Λ2 (Tq M) as in Lemma 2.84
and use Remark 2.75 and Propositions 7.59 and 7.62, we have that
c′ (t) = α t∗ (LX η t ) t (q) + α t∗ η

e (q) = α t∗ d ( i Xt η t ) (q) + α t∗ η
e (q)
= α t∗ (− d θ + η
e ) (q) = 0
for t ∈ I. If (z,U) is the local chart around p with z = y ◦ α 1 | U , we therefore have

that
ω (q) = α ∗1 η 1 (q) = ∑ ω i j (p) dzi (q) ∧ dz j (q)
i< j
for q ∈ U by Propositions 5.53 and 5.59. Hence it follows from Proposition 4.15 that
ω (q) = tz,∗q β
for q ∈ U, where (tz , π −1 (U)) is the local trivialization for the tangent bundle π :
T M → M corresponding to the local chart (z,U), and β ∈ Λ2 (R2k ) is the symplectic
2-form on R2k given by
β = ∑ ω i j (p) ei ∧ e j ,
i< j
where E = {e1 , ..., e2k } is the standard basis for R2k and E ∗ = {e1 , ..., e2k } is the dual
basis.
By Proposition 7.65 there is a basis B = {v1 , ..., v2k } for R2k with dual basis
∗
B = {v1 , ..., v2k } so that
k
β = ∑ vi ∧ vi+k .
i=1
Now let x = Λ ◦ z , where Λ : R2k → R2k is the linear isomorphism with Λ(vi ) = ei
for i = 1, ..., 2k. By Proposition 4.6 it follows that Λ∗ (ei ) = vi for i = 1, ..., 2k, so that
!
k
β = Λ∗ ∑ ei ∧ ei+k .
i=1
Since tx, q = Λ ◦ tz, q by the commutative diagram in 2.70, this finally shows that
!
k k
ω (q) = tx,∗ q ∑ ei ∧ ei+k = ∑ dxi (q) ∧ dxi+k (q)
i=1 i=1
for q ∈ U, which completes the proof of the “only if” part of the theorem. The “if”
part is immediately seen to be true.
7.69 Definition If (M, ω ) is a symplectic manifold, then a local chart (x,U)

having the properties described in Darboux’ theorem, is called a symplectic chart on
M. Its coordinate map x and coordinate functions xi are said to be canonical.
7.70 Proposition Let π : T ∗ M → M be the cotangent bundle of a smooth mani-

fold M n . Then we have a 1-form θ on T ∗ M defined by
θ (v) = v ◦ π∗ v
for v ∈ T ∗ M, called the canonical 1-form or the Cartan form on T ∗ M.

If (x,U) is a local chart on M and (tx′ , π −1 (U)) is the corresponding local trivial-
ization in the cotangent bundle defined in Remark 4.9, we obtain a local chart (z,W )
on T ∗ M where W = π −1 (U) and z = (x × id) ◦ tx′ so that
!
n
z ∑ ai dxi (u) = (x(u), a)
i=1
for u ∈ U and a ∈ Rn . Denoting the two component maps of z : W → Rn × Rn by q

and p , we have that
n
θ | W = ∑ p i dqi .
i=1
The 2-form ω = − d θ is a symplectic form on T ∗ M which is given locally by
n
ω | W = ∑ dqi ∧ d p i .
i=1
It is called the canonical 2-form on T ∗ M.

PROOF : Let (tx , π ′ −1 (U)) and (tz , π ′′ −1 (W )) be the local trivializations for the tan-
gent bundles π ′ : T M → M and π ′′ : T (T ∗ M) → T ∗ M corresponding to the lo-
cal charts (x,U) and (z,W ), and let E = {e1 , ..., en } and F = { f1 , ..., f2n } be the
standard bases for Rn and R2n , respectively, with dual bases E ∗ = {e1 , ..., en } and
F ∗ = { f 1 , ..., f 2n }.
n
Now let v = ∑ ai dxi (u) be a point in W . By Proposition 4.15 we have that
i=1
n
∗
v = tx,u (λ ) where λ = ∑ ai ei .
i=1
Since
x ◦ π ◦ z−1 = pr1 | x(U)× Rn ,
it follows that D(x ◦ π ◦ z−1 )(z(v)) = pr1 , where pr1 : Rn × Rn → Rn is the projection
on the first factor. By 2.70 we therefore have a commutative diagram
tz,v
Tv (T ∗ M) ✲ Rn × Rn
π∗ v pr1
❄ ❄
Tu M ✲ Rn
tx,u
❅
v❅ λ
❅❅
❘ ✠
R
Using that
2n n n
λ ◦ pr1 = ∑ λ ◦ pr1 ( fi ) f i = ∑ λ (ei ) f i = ∑ ai f i ,
i=1 i=1 i=1
we finally have that

n n
∗
θ (v) = tz,v (λ ◦ pr1 ) = ∑ ai ( f i ◦ tz,v ) = ∑ z n+i (v) dzi (v) .
i=1 i=1
As in the proof of Corollary 7.66 we see that
ω n | V = n ! (−1) n (n−1) / 2 dz1 ∧ · · · ∧ dz2n
which shows that ω n is a nowhere vanishing 2n-form on T ∗ M. Hence it follows from

Proposition 7.67 that the 2-form ω is symplectic.
7.71 Proposition Let π : T ∗ M → M be the cotangent bundle of a smooth man-

ifold M, and let θ and ω be the canonical 1- and 2-forms on T ∗ M. Then we have
that
β ∗ θ = β and β ∗ ω = − d β
for every 1-form β on M.
β ∗ θ (u)(v) = θ (β (u))(β ∗ u (v)) = β (u) ◦ π∗ β (u)(β ∗ u (v))
= β (u) ◦ (π ◦ β ) ∗ u (v) = β (u)(v)
for every u ∈ M and v ∈ Tu M. The last formula in the proposition follows from the
first one since ω = − d θ .

ifold M, and let ω be the canonical 2-form on T ∗ M . If (x,U) is a local chart on M,
and (z,W ) is the corresponding local chart on T ∗ M defined in Proposition 7.70, then
( )
∂

B= , ..., ∂∂pn
∂ p1
v v
is a basis for ker π ∗ v for each v ∈ W , and ω (v)(w1 , w2 ) = 0 for every w1 , w2 ∈ ker π ∗ v .
If β is a closed 1-form on M , then
(ω ◦ β )(u)(w1 , w2 ) = (ω ◦ β )(u)((β ◦ π ) ∗ (w1 ), w2 )+ (ω ◦ β )(u)(w1 , (β ◦ π ) ∗ (w2 ))
for u ∈ M and w1 , w2 ∈ Tβ (u) (T ∗ M) .

PROOF : Since !
∂
∂ (xi ◦ π )
π∗ (xi ) = (v) = δ ik
∂ zk ∂ zk
v
for i = 1, ..., n and k = 1, ..., 2n , we have that
!
2n n
π∗ ∑ ai ∂∂zi = ∑ ai ∂ i
∂x
i=1 v i=1 u
for u ∈ U, v ∈ Tu∗ M and a ∈ R2n , which proves the first part of the proposition.
Using that w − (β ◦ π ) ∗ (w) ∈ ker π ∗ β (u) when u ∈ M and w ∈ Tβ (u) (T ∗ M), and
that β ∗ ω = 0 when β is closed by Proposition 7.71, we now obtain that
0 = ( ω ◦ β )(u)(w1 − (β ◦ π ) ∗ (w1 ), w2 − (β ◦ π ) ∗ (w2 ))

= ( ω ◦ β )(u)(w1 , w2 ) − ( ω ◦ β )(u)((β ◦ π ) ∗ (w1 ), w2 )
− ( ω ◦ β )(u)(w1 , (β ◦ π ) ∗ (w2 ))
which shows the last part of the proposition.
HAMILTONIAN SYSTEMS
7.73 Definition Let (M, ω ) be a symplectic manifold, and let H ∈ F (M) be
a smooth function on M. Then the vector field XH = ω ♯ (dH) obtained by raising
indices in the 1-form dH, is called the Hamiltonian vector field with energy function
H. The function H is also called the Hamiltonian .
7.74 Proposition Let XH be a Hamiltonian vector field with energy function H

on a symplectic manifold (M, ω ) of dimension 2n, and let (x,U) be a symplectic
chart on M. Denoting the two component maps of x : U → Rn × Rn by q and p so
that
n
ω | U = ∑ dqi ∧ d p i ,
i=1
we have that
n
∂H ∂ ∂H ∂
XH | U = ∑ ∂ pi ∂ qi − ∂ qi ∂ pi .
i=1
A smooth curve γ : I → U is an integral curve for XH if and only if it satisfies the
Hamiltonian equations
(q i ◦ γ ) ′ (t) = ∂∂ H ′ ∂H
p i (γ (t)) and (p i ◦ γ ) (t) = − ∂ q i (γ (t))
for t ∈ I and i = 1, ..., n.

PROOF : Using the basis

( )
∂

B= , ..., ∂∂qn , ∂ 1 , ..., ∂∂pn
∂ q1 ∂p
u u u u
for Tu M with dual basis

B ∗ = dq1 (u), ..., dqn (u), d p 1 (u), ..., d p n (u)
where u ∈ U, we have that

0 In
mB ( ω (u)) =
− In 0
so that
∗ ♯ t −1 0 In
mB
B ( ω (u) ) = { mB ( ω (u)) } = ,
− In 0
where In is the n × n identity matrix. The proposition now follows from the formula
n
∂H i ∂H
dH | U = ∑ i dq + ∂ p d p i .
i=1
∂q i
7.75 Definition Let (M, ω ) be a symplectic manifold. To each smooth function

f ∈ F (M) we associate a vector field X f = ω ♯ (d f ) obtained by raising indices in the
1-form d f . If f , g ∈ F (M), their Poisson bracket { f , g} is the differeniable function
on M defined by
{ f , g} = ω (X f , Xg ) .
7.76 Proposition Let (M, ω ) be a symplectic manifold, and consider two smooth
functions f , g ∈ F (M) . If γ : I → M is an integral curve for Xg , then we have that
( f ◦ γ ) ′ (t) = { f , g} ◦ γ (t)
for t ∈ I . In particular, the function f ◦ γ is constant on I if and only if { f , g} vanishes
on γ (I) .
( f ◦ γ ) ′ (t) = d f (γ (t))(γ ′ (t)) = d f (γ (t))(Xg (γ (t)))
= ω (γ (t))(X f (γ (t)), Xg (γ (t))) = { f , g} ◦ γ (t)
for t ∈ I .
7.77 Corollary Let XH be a Hamiltonian vector field with energy function H on

a symplectic manifold (M, ω ), and let γ : I → M be an integral curve for XH . Then
the function H ◦ γ is constant on I.
PROOF : Follows from Proposition 7.76 since {H, H} vanishes on M .
LAGRANGIAN SYSTEMS
7.78 Definition Let π : E → M and πe : Ee → M be two vector bundles over a

smooth manifold M. A smooth map f : E → Ee is said to be fibre preserving if the
diagram
f✲
E Ee
❙ ✓
π ❙ ✓ πe
✇
❙ ✴
✓
M
is commutative. We let f p denote the induced smooth map f p : π −1 (p) → πe −1 (p)

between the fibres for each p ∈ M.
7.79 Proposition Let π : E → M and πe : Ee → M be two vector bundles over

a smooth manifold M of dimensions n and m, respectively, and let f : E → Ee be a
smooth map which is fibre preserving. Then we have a smooth and fibre preserving
map
e ,
F( f ) : E → Λ 1 (E ; E)
called the fibre derivative of f , which is defined by
F( f )(v) = ( f p ) ∗ v
for p ∈ M and v ∈ E p , where we canonically identify Tv E p with E p and T f (v) Eep with
Eep as in Lemma 2.84.
PROOF : Let (t, π −1 (U)) and (e e Then

t , πe −1 (U)) be local trivializations in E and E.
we have a smooth map h : U × R → R so that
n m
t ◦ f ◦ t −1 (p, a) = (p, h(p, a))

e
for (p, a) ∈ U × Rn . For each p ∈ U and v ∈ E p we have a commutative diagram

( f p)∗ v ✲ T Ee
Tv E p f (v) p
ωv e f (v)
ω
❄ ❄
Ep ✲ Eep
F( f )(v)
where ωv = t p−1 ◦ t t p , v and ω t p−1 ◦ t et p , f (v) are the canonical identifications

e f (v) = e
given in Lemma 2.84. If τ : π ′ −1 (U) → U × L(Rn , Rm ) is the equivalence over U
defined in Proposition 5.28 and t(v) = (p, a), it therefore follows from the commuta-
tive diagram in 2.70 that
τ ◦ F( f ) ◦ t −1 (p, a) = τ ◦ F( f )(v) = (p, e

t p ◦ F( f )(v) ◦ t p−1 )
= (p, e e f (v) ◦ ( f p ) ∗ v ◦ ω −1
tp ◦ ω −1
v ◦ t p ) = (p, t e
−1
t p , f (v) ◦ ( f p ) ∗ v ◦ t t p , v )
t p ◦ f p ◦ t p−1 )(t p (v))) = (p, D2 h(p, a))

= (p, D(e
which shows that the fibre derivative F( f ) is smooth and fibre preserving.

smooth manifold M, and let f : E → W be a smooth map into an m-dimensional
vector space W . We let f p = f |E p denote the restriction of f to the fibre E p for each
p ∈ M. Then we have a smooth and fibre preserving map
F( f ) : E → Λ 1 (E ;W ) ,
called the fibre derivative of f , which is defined by
F( f )(v) = ( f p ) ∗ v
for p ∈ M and v ∈ E p , where we canonically identify Tv E p with E p and T f (v) W with

W as in Lemma 2.84.
PROOF : Let π e : EW → M be the vector bundle EW = M × W defined in 5.26, and
consider the smooth and fibre preserving map fe : E → EW given by
fe(v) = (p, f (v))
for p ∈ M and v ∈ E p . For each p ∈ M we have that
ρ p ◦ fep = f p ,
where ρ p : {p} × W → W is the projection on the second factor. Since ρ p is a linear

isomorphism, it follows from Lemma 2.84 that
ρ p ◦ F( fe)(v) = F( f )(v)
for p ∈ M and v ∈ E p . If ρ ′ : Λ 1 (E ; EW ) → Λ 1 (E ;W ) is the equivalence over M

defined in 5.26, we therefore have that
ρ ′ ◦ F( fe) = F( f ) ,
which shows that the fibre derivative F( f ) is smooth and fibre preserving.
7.81 Proposition Let π : T M → M and π ∗ : T ∗ M → M be the tangent and cotan-

gent bundle of a smooth manifold M n , and let L : T M → R be a smooth function. If
(x,U) is a local chart on M, and if (tx , π −1 (U)) and (tx′ , π ′−1 (U)) are the correspond-
ing local trivializations in the tangent and cotangent bundle defined in Remark 2.68
and 4.9, we obtain local charts (y,V ) and (z,W ) on T M and T ∗ M, respectively, where
V = π −1 (U) , W = π ∗ −1(U) , y = (x × id) ◦ tx and z = (x × id) ◦ tx′ so that
!
n
y ∑ ai ∂ i = (x(u), a)
i=1
∂x
u
and !
n
i
z ∑ ai dx (u) = (x(u), a)
i=1
for u ∈ U and a ∈ Rn . Denoting the two component maps of y : V → Rn ×Rn by q and

q̇ , and the two component maps of z : W → Rn × Rn by q and p , the fibre derivative
F(L) : T M → T ∗ M , which is called the Legendre transformation corresponding to
the Lagrangian function L , is given by
qi ◦ F(L) = qi and pi ◦ F(L) = ∂ Li

∂ q̇
for i = 1, ..., n. The function pi ◦ F(L) is called the conjugate momentum of the coor-
dinate function qi on V . We have that

In 0
D(z ◦ F(L) ◦ y−1 )(y(v)) = ,
B(v) A(v)
where In is the n × n identity matrix and

∂ 2L ∂ 2L
B(v) = i j (v) and A(v) = (v)
∂ q̇ ∂ q i
∂ q̇ ∂ q̇ j
for v ∈ V . In particular, we see that F(L) is a local diffeomorphism in V if and only

if the matrix A(v) is non-singular for every v ∈ V .
PROOF : For each u ∈ U and v ∈ Tu M we have a commutative diagram
(Lu ) ∗ v ✲ T R
Tv (Tu M) L(v)
ωv e L(v)
ω
❄ ❄
Tu M ✲ R
F(L)(v)
−1 eL(v) = id −1 ◦ t id , L(v) are the canonical identifications

where ωv = tx,u ◦ t tx,u , v and ω
given in Lemma 2.84. Let r : (Rn )∗ → Rn be the linear isomorphism defined by
r(ei ) = ei for i = 1, ..., n, where E = {e1 , ..., en } is the standard basis for Rn and
E ∗ = {e1 , ..., en } is its dual basis. By Remark 4.9 we have that
−1 ∗
z ◦ F(L)(v) = (x(u), r ◦ (tx,u ) ◦ F(L)(v))
where
−1 ∗ −1 e L(v) ◦ (Lu ) ∗ v ◦ ω −1 −1
(tx,u ) ◦ F(L)(v) = F(L)(v) ◦ tx,u =ω v ◦ tx,u
= t id , L(v) ◦ (Lu ) ∗ v ◦ t −1 −1
tx,u , v = D(Lu ◦ tx,u )(tx,u (v)) .
Hence it follows that

zi ◦ F(L)(v) = yi (v)
and
−1
zn+i ◦ F(L)(v) = Di (Lu ◦ tx,u )(tx,u (v)) = ∂n+i
L
(v)
∂y
for i = 1, ..., n. Indeed, if F = { f1 , ..., f2n } is the standard basis for R2n , the final
term in the last formula equals c′ (0), where c : R → R is the curve defined by
c(t) = L ◦ y−1 (y(v) + t fn+i )
for t ∈ R. If tx (v) = (u, a), the last equality in the formula follows since
c(t) = L ◦ y−1 (x(u), a + tei ) = L ◦ tx−1 (u, a + tei) = Lu ◦ tx,u

−1
(a + tei) .
7.82 Proposition Let π : T M → M and π ∗ : T ∗ M → M be the tangent and

cotangent bundle of a smooth manifold M n , and let L : T M → R be a smooth function.
If θ and ω are the canonical 1- and 2-forms on T ∗ M, we let
θL = F(L)∗ θ and ωL = F(L)∗ ω

be the pull-backs to T M by the fibre derivative of L. We have that ωL = − d θL , and

ωL is called the Lagrange 2-form corresponding to the Lagrangian function L.
If (x,U) is a local chart on M, and (y,V ) is the corresponding local chart on T M
defined in Proposition 7.81, we have that
θL | V = ∑ ∂ Li dqi
i
∂ q̇
and
∂ 2L i j ∂ 2L
ωL | V = ∑ i j dq ∧ dq + dqi ∧ d q̇ j .
ij
∂ q̇ ∂ q ∂ q̇i ∂ q̇ j
The matrix of ωL (v) with respect to the basis

( )
∂

B= , ..., ∂∂qn , ∂ 1 , ..., ∂∂q̇n
∂ q1 ∂ q̇
v v v v
for Tv (T M) is therefore

B(v) − B(v) t A(v)
mB ( ωL (v)) =
−A(v) 0
for v ∈ V , where A(v) and B(v) are the matrices defined in Proposition 7.81.
PROOF : By Propositions 7.70 and 7.81 we have that
θL | V = ∑ {p i ◦ F(L)} d {qi ◦ F(L)} = ∑ ∂ Li dqi ,

i i
∂ q̇
and using that

∂L ∂ 2L j ∂ 2L
d =∑ j dq + d q̇ j ,
∂ q̇i j
i
∂ q̇ ∂ q ∂ q̇i ∂ q̇ j
we obtain the formula for ωL | V .
7.83 Definition Let M be a smooth manifold, and let L : T M → R be a smooth

function. Then L is said to be a regular Lagrangian if the fibre derivative F(L) is a
local diffeomorphism. It is called a hyperregular Lagrangian if F(L) is a diffeomor-
phism.
7.84 Remark We see that the Lagrange 2-form ωL is symplectic if and only if L
is a regular Lagrangian.
7.85 Definition Let M be a smooth manifold, and let L : T M → R be a regular

Lagrangian. Then we define the action A : T M → R by
A(v) = F(L)(v)(v)
for v ∈ T M, and the energy E : T M → R by E = A − L . The vector field XE =

(ωL ) ♯ (dE) obtained by raising indices in the 1-form dE, is called the Lagrangian
vector field for L.
7.86 Remark If (x,U) is a local chart on M, and (y,V ) and (z,W ) are the corre-
sponding local charts on T M and T ∗ M defined in Proposition 7.81, we have that
A| V = ∑ q̇i ∂ Li ,
i
∂ q̇
since
λ (v) = ∑ q̇i (v) p i (λ )
i
for v ∈ Tu M and λ ∈ Tu∗ M when u ∈ U.
7.87 Proposition Let M be a smooth manifold, and let XE be the Lagrangian

vector field for a regular Lagrangian L : T M → R . If γ : I → M is an integral curve
for XE , then the function E ◦ γ is constant on I.
(E ◦ γ ) ′ (t) = dE(γ (t))(γ ′ (t)) = dE(γ (t))(XE (γ (t)))
= ωL (γ (t))(XE (γ (t)), XE (γ (t))) = 0
for t ∈ I .
7.88 Definition Let π : T M → M be the tangent bundle of a smooth manifold M.

Then a vector field X on T M satisfying
π∗ ◦ X = idT M
is called a second order equation on M.
7.89 Proposition Let π : T M → M be the tangent bundle of a smooth manifold

M. Then a vector field X on T M is a second order equation on M if and only if
(π ◦ γ ) ′ = γ
for every integral curve γ : I → T M of X.

PROOF : If γ : I → T M is an integral curve of X, then
(π ◦ γ ) ′ (t) = π∗ ◦ γ ′ (t) = π∗ ◦ X(γ (t))
for t ∈ I by 2.70. The proposition therefore follows from Proposition 3.36.

7.90 Proposition Let (x,U) be a local chart on a smooth manifold M n , and let
(y,V ) be the corresponding local chart on T M defined in Proposition 7.81. Further-
more, let XE be a Lagrangian vector field for a regular Lagrangian L : T M → R . Then
XE is a second order equation on M, and a smooth curve γ : I → V is an integral curve
for XE if and only if it satisfies the Lagrange’s equations
′
∂L
(q i ◦ γ ) ′ (t) = q̇ i (γ (t)) and i ◦ γ (t) = ∂ Li (γ (t))
∂ q̇ ∂q
for t ∈ I and i = 1, ..., n.

PROOF : Using the basis
( )
∂

B= , ..., ∂∂qn , ∂ 1 , ..., ∂∂q̇n
∂ q1 ∂ q̇
v v v v
for Tv (T M) with dual basis

B ∗ = dq1 (v), ..., dqn (v), d q̇ 1 (v), ..., d q̇ n (v)
where v ∈ V , it follows from Proposition 7.82 that

B ♭ B(v) t − B(v) −A(v)
mB∗ ( ωL (v) ) =
A(v) 0
so that
!
∗ ♯
0 A(v) −1
mB
B ( ωL (v) ) = .
−A(v) −1 A(v) −1 {B(v) t − B(v) } A(v) −1
By Remark 7.86 we have that
E | V = ∑ q̇ j ∂ Lj − L| V ,
j
∂ q̇
so that
dE | V = ∑ (ai dqi + bi d q̇ i )
i
where
∂ 2L
ai = ∂ Ei = ∑ q̇ j − ∂L
∂q j
∂ qi ∂ q̇ j ∂ qi
and
∂ 2L
bi = ∂ Ei = ∑ q̇ j .
∂ q̇ j
∂ q̇i ∂ q̇ j
Writing a, b, e and q̇ as column vectors, where ei = ∂ Li , this means that

∂q
a = B t q̇ − e and b = A q̇ .
Now we have that

n
XE | V = (ωL ) ♯ (dE)| V = ∑ ci ∂ i + d i ∂ i
i=1
∂q ∂ q̇
where
c = q̇ and Ad = −(B t q̇ − e) + (B t − B) q̇ = e − B q̇ .
Hence γ is an integral curve for XE if and only if
(q i ◦ γ ) ′ (t) = q̇ i (γ (t))
and
′
∂L ∂L
◦ γ (t) = d (γ (t)) (γ ′ (t))
∂ q̇ i ∂ q̇i

∂ 2L ′ ∂ 2L
=∑ i
j
j (γ (t)) dq (γ (t)) (γ (t)) + (γ (t)) d q̇ j (γ (t)) (γ ′ (t))
j
∂ q̇ ∂ q ∂ q̇i ∂ q̇ j

∂ 2L 2
=∑ (γ (t)) (q j ◦ γ ) ′ (t) + ∂i L j (γ (t)) (q̇ j ◦ γ ) ′ (t)
j
∂ q̇i ∂ q j ∂ q̇ ∂ q̇

∂ 2L 2
=∑ (γ (t)) q̇ j (γ (t)) + ∂i L j (γ (t)) (q̇ j ◦ γ ) ′ (t) = ∂ Li (γ (t))
j
∂ q̇i ∂ q j ∂ q̇ ∂ q̇ ∂q
for t ∈ I and i = 1, ..., n.

n
To show that XE is a second order equation on M, let v = ∑ wi ∂ i . Then we
i=1
∂x
u
have that !
∂
∂ (xi ◦ π )
π∗ (xi ) = (v) = δ ik
∂ yk ∂ yk
v
for i = 1, ..., n and k = 1, ..., 2n , which implies that

n
π∗ ◦ XE (v) = ∑ q̇i (v) ∂ i = v .
∂x
i=1 u
7.91 Remark If the regular Lagrangian L is independent of the coordinate q i ,

i
then q is called a cyclic coordinate . From the Lagrange’s equations it follows that its
conjugate momentum pi ◦ F(L) = ∂ Li is a constant of the motion , i.e., it is constant
∂ q̇
on γ (I) for every curve γ : I → V which is an integral curve for XE .
7.92 Proposition Let M be a smooth manifold, and let XE be the Lagrangian

vector field for a hyperregular Lagrangian L : T M → R . Let XH be the Hamiltonian
vector field with energy function H = E ◦ F(L) −1 , where E is the energy of L. Then
XE and XH are F(L) -related, so that the diagram
F(L) ∗
T ( T M) ✲ T ( T ∗ M)
✻ ✻
XE XH
TM ✲ T ∗M
F(L)
❅
❅
E ❅ H
❘
❅ ✠
R
is commutative. If θ is the canonical 1-form on T ∗ M and θL = F(L) ∗ θ , then
θ (XH ) = A ◦ F(L) −1 and θL (XE ) = A
where A is the action of L.
PROOF : Let ω be the canonical 2-form on T ∗ M, and let ωL = F(L) ∗ ω be the La-
grange 2-form on T M. Then we have that
ω (F(L)(v)) ( F(L) ∗ ◦ XE (v), F(L) ∗ (w)) = ωL (v) ( XE (v), w)
= dE(v)(w) = d(H ◦ F(L))(v)(w) = F(L) ∗ dH(v)(w)
= dH(F(L)(v)) ( F (L) ∗ (w)) = ω (F(L)(v)) ( XH ◦ F(L)(v)), F(L) ∗ (w))
for every v ∈ T M and w ∈ Tv (T M). Since F(L) ∗ v : Tv (T M) → TF(L)(v) (T ∗ M) is a

linear isomorphism for each v ∈ T M, this implies that
F(L) ∗ ◦ XE = XH ◦ F(L) ,
showing that XE and XH are F(L) -related.

Now let v ∈ T M and w = F(L)(v) , and let π and π ′ denote the projections in the
tangent and cotangent bundle of M, respectively. Then we have that
θ (XH )(w) = θ (w)(XH (w)) = w ◦ π∗′ ◦ XH ◦ F(L)(v)
= w ◦ π∗′ ◦ F(L) ∗ ◦ XE (v) = w ◦ π∗ ◦ XE (v) = w(v) = A(v)

and
θL (XE )(v) = θL (v)(XE (v)) = θ (w) (F(L) ∗ ◦ XE (v))
= θ (w) (XH ◦ F(L)(v)) = θ (XH )(w)
7.93 Proposition Let π : T M → M and π ∗ : T ∗ M → M be the tangent and

cotangent bundle of a smooth manifold M n , and let H : T ∗ M → R be a smooth func-
tion. Given a local chart (x,U) on M, we let (y,V ) and (z,W ) be the corresponding
local charts on T M and T ∗ M defined in Proposition 7.81. Then the fibre derivative
F(H) : T ∗ M → T M , obtained by identifying T ∗∗ M with T M as usual by the equiva-
lence iT M : T M → T ∗∗ M defined in Proposition 4.10, is given by
qi ◦ F(H) = qi and q̇i ◦ F(H) = ∂∂ H

pi
for i = 1, ..., n. We have that

−1 In 0
D(y ◦ F(H) ◦ z )(z(v)) = ,
B(v) A(v)
where In is the n × n identity matrix and

∂ 2H ∂ 2H
B(v) = j (v) and A(v) = ∂p ∂p (v)
∂ pi ∂ q i j
for v ∈ W . In particular, we see that F(H) is a local diffeomorphism in W if and only

if the matrix A(v) is non-singular for every v ∈ W .
PROOF : For each u ∈ U and v ∈ Tu∗ M we have a commutative diagram
(Hu ) ∗ v ✲
Tv (Tu∗ M) TH(v) R
ωv e H(v)
ω
❄ ❄
Tu∗ M ✲ R
iT M ◦ F(H)(v)
where ωv = tx,u ′−1 ◦ t ′ eH(v) = id −1 ◦ t id , H(v) are the canonical identifications

tx,u , v and ω
given in Lemma 2.84. Let r : (Rn )∗ → Rn be the linear isomorphism defined by
r(ei ) = ei for i = 1, ..., n, where E = {e1 , ..., en } is the standard basis for Rn and
E ∗ = {e1 , ..., en } is its dual basis. By Remark 4.9 we have that
′ −1
y ◦ F(H)(v) = (x(u), tx,u ◦ F(H)(v)) = (x(u), r ◦ (tx,u )∗ ◦ iT M ◦ F(H)(v))
where
−1
′ −1
(tx,u )∗ ◦ iT M ◦ F(H)(v) = iT M ◦ F(H)(v) ◦ tx,u
′
−1
e H(v) ◦ (Hu ) ∗ v ◦ ω −1
=ω ′ −1
v ◦ tx,u = t id , H(v) ◦ (Hu ) ∗ v ◦ t t ′ , v
x,u
′ −1 ′
= D(Hu ◦ tx,u )(tx,u (v)) .

yi ◦ F(H)(v) = zi (v)
and
′−1
yn+i ◦ F(H)(v) = Di (Hu ◦ tx,u ′
)(tx,u (v)) = ∂ n+i
H
(v)
∂z
for i = 1, ..., n. Indeed, if F = { f1 , ..., f2n } is the standard basis for R2n , the final
term in the last formula equals c′ (0), where c : R → R is the curve defined by
c(t) = H ◦ z−1 (z(v) + t fn+i )
for t ∈ R. If tx′ (v) = (u, a), the last equality in the formula follows since
−1 −1
c(t) = H ◦ z−1 (x(u), a + tei ) = H ◦ tx′ (u, a + tei) = Hu ◦ tx,u
′
(a + tei) .
7.94 Definition Let M be a smooth manifold, and let H : T ∗ M → R be a smooth

function. Then we define the action G : T ∗ M → R of H by
G = θ (XH ) ,
where θ is the canonical 1-form on T ∗ M, and XH is the Hamiltonian vector field on

T ∗ M with energy function H.
7.95 Remark If (x,U) is a local chart on M, and (z,W ) is the corresponding local
chart on T ∗ M defined in Proposition 7.81, we have that
G| W = ∑ pi ∂∂ H
pi ,
i
by Proposition 7.70 and 7.74.
7.96 Definition Let M be a smooth manifold, and let H : T ∗ M → R be a smooth

function. Then H is said to be a regular Hamiltonian if the fibre derivative F(H) is
a local diffeomorphism. It is called a hyperregular Hamiltonian if F(H) is a diffeo-
morphism.
7.97 Proposition Let M n be a smooth manifold, and let H : T ∗ M → R be a

hyperregular Hamiltonian with action G. Then the function L : T M → R given by
L = A − E , where
A = G ◦ F(H)−1 and E = H ◦ F(H)−1 ,
is a hyperregular Lagrangian, and we have that
F(L) = F(H)−1 .
PROOF : Let (x,U) be a local chart on M, and let (y,V ) and (z,W ) be the correspond-
ing local charts on T M and T ∗ M defined in Proposition 7.81. By Proposition 7.93
we have that
qi ◦ F(H)−1 = qi and ∂∂ H −1
p ◦ F(H) = q̇
i
i
for i = 1, ..., n. Hence it follows from Remark 7.95 that
L| V = ∑ {p j ◦ F(H)−1 } q̇ j − H | W ◦ F(H)−1 ,
j
and using the chain rule on the last term, we obtain
∂ {p j ◦ F(H)−1 } j
pi ◦ F(L) = ∂ Li = ∑ i q̇ + {pi ◦ F(H)−1 }
∂ q̇ j
∂ q̇
∂ {p j ◦ F(H)−1 }
− ∑ { ∂∂ pH ◦ F(H)−1 } = pi ◦ F(H)−1
j j ∂ q̇i
for i = 1, ..., n. The proposition now follows since we also have that
qi ◦ F(L) = qi = qi ◦ F(H)−1
for i = 1, ..., n.
7.98 Remark Since
A = G ◦ F(H)−1 = θ (XH ) ◦ F(L) = θ (XE ◦ F(L)−1 ) ◦ F(L) ,
it follows from Proposition 7.92 that A is the action of L, and hence that E = A − L
is the energy of L. Therefore H = E ◦ F(L)−1 can be recovered from L in the way
described in Proposition 7.92.
7.99 Proposition Let M n be a smooth manifold, and let L : T M → R be a hy-

perregular Lagrangian with energy E. Then the function H : T ∗ M → R given by
H = E ◦ F(L) −1 is a hyperregular Hamiltonian, and we have that
F(H) = F(L)−1 .
PROOF : Let (x,U) be a local chart on M, and let (y,V ) and (z,W ) be the correspond-
ing local charts on T M and T ∗ M defined in Proposition 7.81. By the same proposition
we also have that
∂L
qi ◦ F(L)−1 = qi and ◦ F(L)−1 = pi
∂ q̇i
for i = 1, ..., n. Hence it follows from Remark 7.86 that
H |W = ∑ {q̇ j ◦ F(L)−1 } p j − L| V ◦ F(L)−1 ,
j
and using the chain rule on the last term, we obtain
∂ {q̇ j ◦ F(L)−1 }
q̇i ◦ F(H) = ∂∂ H
pi = ∑ ∂ pi p j + {q̇i ◦ F(L)−1 }
j
∂ {q̇ j ◦ F(L)−1 }
− ∑ { ∂ Lj ◦ F(L)−1 } ∂p = q̇i ◦ F(L)−1
j
∂ q̇ i
for i = 1, ..., n. The proposition now follows since we also have that
qi ◦ F(H) = qi = qi ◦ F(L)−1
for i = 1, ..., n.
7.100 Remark By Proposition 7.92 and Definition 7.94 we have that

H = E ◦ F(L)−1 = (A − L) ◦ F(L)−1 = G − L ◦ F(H) ,
so that
L = G ◦ F(H)−1 − H ◦ F(H)−1 .
Hence L can be recovered from H in the way described in Proposition 7.97.
CONSERVATIVE SYSTEMS
7.101 Proposition Let M be a pseudo-Riemannian manifold with a metric g, and
let V ∈ F (M) be a smooth function on M. Then the map L : T M → R defined by
L = T − V ◦ π , where
T (v) = 12 g(u)(v, v)
for u ∈ M and v ∈ Tu M, is a hyperregular Lagrangian whose fibre derivative F(L) :
T M → T ∗ M is given by
F(L) u = g(u) ♭
for u ∈ M. The action and energy of L are given by A = 2T and E = T + V ◦ π .
PROOF : Follows from Remark 2.75 and Lemma 2.84.
7.102 Example Consider a system of n particles with masses m1 , ... , mn moving

in R3 . Then the state of the system is specified by a pair (q, v) ∈ R3n × R3n , where
(q 3k−2 , q 3k−1 , q 3k ) is the position and (v 3k−2 , v 3k−1 , v 3k ) the velocity of the kth parti-
cle for k = 1, ... , n. The point q moves in the configuration space R3n , and R3n × R3n
is called the state space of the system. Let E = {e1 , ..., e3n } be the standard basis for
R3n with dual basis E ∗ = {e1 , ..., e3n }, where ei is the ith component of the identity
map id : R3n → R3n for i = 1, ... , 3n. By Lemma 2.84 the state space may be iden-
tified with the total space in the tangent bundle π : T R3n → R3n by means of the
equivalence ω : T R3n → R3n × R3n over R3n given by

3n
ω ∑ v ∂∂e i
i
= (q, v)
i=1 q
for (q, v) ∈ R3n ×R3n . As well as being real numbers, we also let q1 , ... , q3n , v1 , ... , v3n
denote the coordinate functions of the local chart (ω , T R3n ) on T R3n .
We have a metric g on R3n defined by
3n
g(q)(v, w) = ∑ Mi vi wi
i=1
for q ∈ R3n and v, w ∈ Tq R3n , where M3k−2 = M3k−1 = M3k = mk for k = 1, ... , n. Let
(F3k−2 , F3k−1 , F3k ) be the force acting on the kth particle for k = 1, ... , n. Assuming
that the forces are conservative, there is a potential energy function V ∈ F (R3n ) such
that
Fi = − ∂ Vi ◦ π
∂q
for i = 1, ... , 3n.
By Proposition 7.101 we have a hyperregular Lagrangian L : T R3n → R defined
by L = T − V ◦ π , where T : T R3n → R is the kinetic energy function given by
3n
T ◦ ω −1 (q, v) = ∑ 1
2 Mi (vi )2
i=1
for (q, v) ∈ R3n × R3n . The energy of L, called the total energy function, is therefore
E = T + V ◦ π , i.e.,
3n
E ◦ ω −1 (q, v) = ∑ 12 Mi (vi )2 + V (q)
i=1
for (q, v) ∈ R3n × R3n . The Lagrange’s equations for a curve γ : I → T R3n now take
the form
(q i ◦ γ ) ′ (t) = (v i ◦ γ )(t) and Mi (vi ◦ γ ) ′ (t) = − ∂ Vi ◦ π ◦ γ (t)

∂q
for t ∈ I and i = 1, ..., 3n. Combining these equations, we obtain Newton’s second
law
Mi (qi ◦ γ ) ′′ (t) = Fi ◦ γ (t)
for t ∈ I and i = 1, ..., 3n.
The system can also be described in the phase space R3n × (R3n )∗ , which may
be identified with the total space in the cotangent bundle π : T ∗ R3n → R3n by means
of the equivalence ω e : T ∗ R3n → R3n × (R3n )∗ over R3n given by ω eq = (ωq∗ )−1 for
3n
q ∈ R , i.e.,
3n 3n
ωe ( ∑ p i de i (q) ) = q, ∑ pi ei
i=1 i=1
for (q, p) ∈ R ×R . As well as being real numbers, we also let q1 , ... , q3n , p1 , ... , p3n
3n 3n
denote the coordinate functions of the local chart ((id × r) ◦ ωe , T ∗ R3n ) on T ∗ R3n ,
n ∗
where r : (R ) → R is the linear isomorphism defined by r(e ) = ei for i = 1, ..., 3n.
n i
The Legendre transformation F(L) : T R3n → T ∗ R3n is given by

3n 3n
−1 i i i
e ◦ F(L) ◦ ω
ω q, ∑ v ei = q, ∑ Mi v e
i=1 i=1
for (q, v) ∈ R3n × R3n, with inverse

3n 3n
ω ◦ F(L)−1 ◦ ω e −1 q, ∑ pi ei = q, ∑ pi ei
i=1 i=1 Mi
for (q, p) ∈ R3n × R3n . By Proposition 7.92 the Hamiltonian of the system is H =
E ◦ F(L) −1 , i.e.,
3n 3n
−1 i p2
H◦ω e q, ∑ pi e = ∑ i + V (q)
i=1 i=1 2Mi
for (q, p) ∈ R3n × R3n . The Hamilton equations for a curve γ : I → T ∗ R3n therefore
take the form
(q i ◦ γ ) ′ (t) = 1 (p i ◦ γ )(t) and (pi ◦ γ ) ′ (t) = − ∂ Vi ◦ π ◦ γ (t)

Mi ∂q
for t ∈ I and i = 1, ..., 3n. Combining these equations, we again obtain Newton’s
second law.
TIME DEPENDENT SYSTEMS

7.103 Theorem Let ω ∈ Ω 2 (M) be a closed 2-form of rank 2k on a smooth
manifold M n . Then there is a local chart (x,U) around each point p ∈ M with x(p) = 0
and
k
i=1
PROOF : By Proposition 5.87 we know that ker(ω ) is an integrable distribution

on M of dimension n − 2k . Hence there is a local chart (y,V ) around p so that
∂
{ 2k+1
, ... , ∂ n } is a local basis for ker(ω ) on V by Frobenius’ integrability theo-
∂y ∂y
rem. We may assume that y(p) = 0 and that y(V ) = O1 × O2 for connected open sets
O1 ⊂ R2k and O2 ⊂ Rn−2k . Then the form ω is given locally by
ω |V = ∑ a i j dyi ∧ dy j ,
1≤i< j≤2k
where the skew symmetric matrix ( ω i j (q)) , given by ω i j (q) = a i j (q) for 1 ≤ i <
j ≤ 2k , is of rank 2k for every q ∈ V . Using that ω is closed, we have that
n
∂ ai j
d ω |V = ∑ ∑ dyr ∧ dyi ∧ dy j = 0
1≤i< j≤2k r=1 ∂ yr
which implies that

∂ ai j
=0 (1)
∂ yr
for 1 ≤ i < j ≤ 2k and r = 2k + 1, ... , n .
Now define N = y−1 ( O1 × {0}) , and let α : N → V be the inclusion map. Then
∗
α ( ω ) is a symplectic form on N given by
α ∗(ω ) = ∑ b i j dui ∧ du j ,
1≤i< j≤2k
where b i j = a i j ◦ α and (u, N) is the local chart on N defined by ui = yi ◦ α for

i = 1, ... , 2k . Let pr1 : R2k × Rn−2k → R2k be the projection on the first factor, and
let π : V → N be the submersion defined by π = u−1 ◦ pr1 ◦ y . Then u ◦ π = pr1 ◦ y
so that ui ◦ π = yi for i = 1, ... , 2k . From (1) it also follows that b i j ◦ π = a i j which
implies that ω |V = (α ◦ π )∗ ( ω ) .
By Darboux’ theorem there is now a local chart (z,W ) around p on N with z(p) =
0 and
k
α ∗ ( ω )|W = ∑ dzi ∧ dzi+k .
i=1
Using this, we obtain the local chart (x,U) around p on M with coordinate neigh-
bourhood U = π −1 (W ) and coordinate map x : U → R2k × Rn−2k defined by x =
{(z ◦ u−1 ) × id } ◦ y , i.e.,
j
z ◦ π for j = 1, ... , 2k
xj = ,
yj for j = 2k + 1, ... , n
which satisfies the conditions x(p) = 0 and

k
i=1
7.104 Theorem Let π : T ∗ M → M be the cotangent bundle of a connected

n
smooth manifold M , and let XH be a Hamiltonian vector field with energy function
H on T ∗ M, and S : M → R be a smooth function on M . Then the following two
assertions are equivalent :
(1) If γ : I → M is an integral curve for the vector field π∗ ◦ XH ◦ dS on M , then
dS ◦ γ is an integral curve for XH .
(2) S satisfies the Hamilton-Jacobi equation H ◦ dS = E , where E is a constant.
PROOF : Since dS and XH are sections of the vector bundles π : T ∗ M → M and
π ′′ : T ( T ∗ M) → T ∗ M , respectively, it follows from the commutative diagram
π∗ ✲
T ( T ∗ M) TM
π ′′ π′
❄ ❄
✲
T ∗M π M
that π∗ ◦ XH ◦ dS is a section of π ′ : T M → M .
Let α : I → T ∗ M be the smooth curve on T ∗ M given by α = dS ◦ γ . If γ : I → M
is an integral curve for π∗ ◦ XH ◦ dS , we have that
α ′ (t) = (dS) ∗ ◦ γ ′ (t) = (dS ◦ π ) ∗ ◦ XH (α (t))
for t ∈ I . From Proposition 7.72 it follows that
ω (α (t))((dS ◦ π ) ∗ ◦ XH (α (t)), w) = ω (α (t))(XH (α (t)), w)
− ω (α (t))(XH (α (t)), (dS ◦ π ) ∗ (w))
for t ∈ I and w ∈ Tα (t) (T ∗ M) , where
ω (α (t))(XH (α (t)), (dS ◦ π ) ∗ (w)) = dH(α (t)) ((dS ◦ π ) ∗ (w))
= (dS) ∗ dH(γ (t))(π ∗ (w)) = d(H ◦ dS)(γ (t))(π ∗ (w)) .

Hence we conclude that
(dS ◦ π ) ∗ ◦ XH (α (t)) = XH (α (t))
if and only if
d(H ◦ dS)(γ (t)) = 0
for t ∈ I , showing that the assertions (1) and (2) are equivalent.
7.105 Definition An odd dimensional smooth manifold M with a closed 2-form

ω of maximal rank is called a contact manifold , and it is also denoted by (M, ω ) .
7.106 Definition Let (M, ω ) be a symplectic manifold, and let H ∈ F (R × M) be

a smooth function on R × M. Then the time dependent vector field XH on M over the
time interval R defined by (XH ) s = XH s for s ∈ R, is called the Hamiltonian vector
field with energy function H.
7.107 Proposition Let (M, ω ) be a symplectic manifold, and let t : R × M → R

and π2 : R × M → M be the projections on the first and second factor. If H ∈ F (R ×
M) is a smooth function on R × M, we have a contact manifold (R × M, ωH ) , where
ωH = π2∗ ω + dH ∧ dt .
The suspension XeH of the Hamiltonian vector field with energy function H is a local
basis for ker ωH .
PROOF : We see that
! ! !
d

d
ωH (s, p) i p ∗ a dr + i s ∗ (v) , i p ∗ b dr + i s ∗ (w)
s s
= ω (p)(v, w) − a dHs (p)(w) + b dHs (p)(v)
= ω (p)(v − a XH (s, p), w) + b ω (p)( XH (s, p), v) = 0
for every b ∈ R and w ∈ Tp M if and only if v = a XH (s, p) . This shows that

!

XeH (s, x) = i p ∗ dr
d
+ i s ∗ ◦ XH (s, p) forms a basis for ker ωH (s, p) for each
s
(s, p) ∈ R × M .
7.108 Theorem (Hamilton-Jacobi) Let π : T ∗ M → M be the cotangent bundle

of a smooth manifold M n , and let XH be a time dependent Hamiltonian vector field
on T ∗ M over the open time interval J with energy function H , and W : J × M → R
be a smooth function on J × M . Then the following two assertions are equivalent :
(1) If γ : I → M is an integral curve for the time dependent vector field Y on M

given by Ys = π∗ ◦ XHs ◦ dWs for s ∈ J , then the curve α : I → T ∗ M defined by
α (s) = dWs (γ (s)) for s ∈ I is an integral curve for XH .
(2) W satisfies the Hamilton-Jacobi equation

Hs ◦ dWs + ∂∂Wt = const
s
on M for each s ∈ J .
PROOF : Let (r, J) be the standard local chart on J where r : J → R is the inclusion
map, and let (x,U) be a local chart on M and (z,W ) be the corresponding local chart
on T ∗ M defined in Proposition 7.70. We denote by t and y the two component maps
of the coordinate map r × x : J × U → R × Rn on J × M.
Now let F : J × M → T ∗ M be the smooth map defined by F(s, u) = dWs (u) for
s ∈ J and u ∈ M , which is given locally by
n
∂W
F(s, u) = ∑ (s, u) dxi (u)
i=1
∂ yi
so that
!
d
2n
∂ (zi ◦ F) 2 n
Fu ∗ dr = ∑ ∂t (s, u) ∂ i = ∑ ∂ W i (s, u) ∂ i
i=1
∂z i=1
∂t ∂y ∂p
s F(s,u) F(s,u)
for s ∈ J and u ∈ U . Using the local expression for ω given in Proposition 7.74, we
see that
! !
d
n
∂ 2W
ω (F(s, u)) Fu ∗ dr ,w =− ∑ (s, u) dqi (F(s, u)) (w)
i=1
∂ t ∂ yi
s

n
= − ∑ ∂ i ∂∂Wt dxi (u) (π∗ (w)) = − d ∂∂Wt (u) (π∗ (w))
∂x i=1 s s
u
for each w ∈ TF(s,u) (T ∗ M) where s ∈ J and u ∈ U .

If γ : I → M is an integral curve for Y , we have that
! !
d
d

α ′ (s) = Fγ (s) ∗ dr + (dWs ) ∗ (γ ′ (s)) = Fγ (s) ∗ dr + (dWs ◦ π ) ∗ (XHs (α (s)))
s s
for s ∈ I . From Proposition 7.72 it follows that
ω (α (s))((dWs ◦ π ) ∗ (XHs (α (s))), w) = ω (α (s))(XHs (α (s)), w)
− ω (α (s))(XHs (α (s)), (dWs ◦ π ) ∗ (w))

for s ∈ I and w ∈ Tα (s) (T ∗ M) , where
ω (α (s))(XHs (α (s)), (dWs ◦ π ) ∗ (w)) = dHs (α (s)) ((dWs ◦ π ) ∗ (w))
= (dWs ) ∗ dHs (γ (s))(π ∗ (w)) = d(Hs ◦ dWs )(γ (s))(π ∗ (w)) .
Hence we conclude that

!

d
Fγ (s) ∗ dr + (dWs ◦ π ) ∗ (XHs (α (s))) = XHs (α (s))
s
if and only if
d Hs ◦ dWs + ∂∂Wt (γ (s)) = 0
s
for each s ∈ I , showing that the assertions (1) and (2) are equivalent.
Chapter 8
LIE GROUPS
LIE GROUPS AND THEIR LIE ALGEBRAS

8.1 Definition A Lie group G is a group which is at the same time a smooth
manifold such that the maps µ : G × G → G and ν : G → G given by the group
operations µ (g, h) = gh and ν (g) = g−1 are smooth. The unit element of G is denoted
by e.
8.2 Remark Smoothness of µ and ν is equivalent to assuming that the map

ρ : G × G → G given by ρ (g, h) = gh−1 is smooth.
8.3 If G is a Lie group and g ∈ G, we define the left and right translations
Lg : G → G and Rg : G → G by
Lg (h) = gh and Rg (h) = hg
for h ∈ G. Both Lg and Rg are diffeomorphisms with inverses Lg−1 and Rg−1 , respec-
tively. In particular, it follows that a Lie group has the same dimension at all points.
A vector field X on G is called left invariant if Lg ∗ (X) = X for every g ∈ G, which
means that X is Lg -related to itself for every g ∈ G. It follows from Propositions 4.86
and 4.88 that the left invariant vector fields form a Lie subalgebra of the Lie algebra
of all vector fields on G.
8.4 Proposition If G is a Lie group and X ∈ Te G, then there is a unique left

invariant vector field X on G with Xee = X which is given by
e
Xeg = Lg ∗ (X)
for g ∈ G. It is called the left invariant vector field determined by X.

PROOF : Since Lgh = Lg ◦ Lh , we have that
Xegh = Lg ∗ (Xeh )
for every g, h ∈ G. To complete the proof, we only need to show that Xe is smooth in
247
some open neighbourhood V of e. Since Lg maps V diffeomorphically onto an open

neighbourhood Lg (V ) of g, and we have that
Xe |Lg (V ) = Lg ∗ (Xe |V ) ,
it then follows that Xe is smooth at every point g ∈ G and therefore is a left invariant
vector field on G with Xee = X.
Let (z,U) be a local chart around e, and let n be the dimension of G. As G is
a Lie group, the map µ : G × G → G given by µ (g, h) = gh is smooth. Choose an
open neighbourhood V of e so that µ (V × V ) ⊂ U, and let x = z|V . Then the map
f : x(V ) × x(V ) → Rn defined by f = z ◦ µ ◦ (x−1 × x−1 ) is smooth, and we have that
f (a, b) = z ◦ Lx−1 (a) ◦ x−1 (b)
for a, b ∈ x(V ).
Now let (tz , π −1 (U)) be the local trivialization in the tangent bundle π : T G → G
associated with the local chart (z,U) on G. If X = [x, v] e , we have that
Xe ◦ x−1 (a) = Lx−1 (a) ∗ (X) = [ z, D (z ◦ Lx−1 (a) ◦ x−1 )(x(e)) v ] x−1 (a)
= [ z, D2 f (a, x(e)) v ] x−1 (a)
so that
(z × id) ◦ tz ◦ Xe ◦ x−1 (a) = (a, D2 f (a, x(e)) v)
for a ∈ x(V ), thus showing that Xe is smooth in V .
8.5 We thus have a linear isomorphism φ from the tangent space Te G to the Lie
e and we introduce
algebra of all left invariant vector fields on G given by φ (X) = X,
a bracket product in Te G which makes φ a Lie algebra isomorphism by defining
e Ye ] (e)
[ X,Y ] = [ X,
for X,Y ∈ Te G. The vector space Te G with this bracket product is called the Lie
algebra of G and is denoted by L (G) or g.
8.6 Definition A smooth homomorphism φ : G → H between the Lie groups

G and H is called a Lie group homomorphism . If φ is also a diffeomorphism, it is
called a Lie group isomorphism. A Lie group isomorphism of a Lie group with itself
is called a Lie group automorphism.
If G is a Lie group, then a Lie group homomorphism φ : R → G is called a one-
parameter subgroup of G.
8.7 Proposition Let φ : G → H be a Lie group homomorphism, and suppose that

X ∈ Te G and X ′ = φ∗ (X). Then the left invariant vector fields Xe and Xe′ on G and H
determined by X and X ′ , respectively, are φ -related.
LIE GROUPS 249
PROOF : For each g ∈ G, we have that
φ ◦ Lg = Lφ (g) ◦ φ
so that
φ∗ (Xeg ) = φ∗ Lg ∗ (X) = Lφ (g) ∗ (X ′ ) = Xe′ φ (g) .
8.8 Proposition If φ : G → H is a Lie group homomorphism, then φ∗ e : Te G →

Te H is a Lie algebra homomorphism.
PROOF : Let X,Y ∈ Te G, and suppose that X ′ = φ∗ (X) and Y ′ = φ∗ (Y ). Then it fol-
lows from Propositions 8.7 and 4.88 that [ X, e Ye ] and [ Xe′ , Ye′ ] are φ -related, which
implies that
φ∗ ( [ X,Y ] ) = [ φ∗ (X) , φ∗ (Y ) ] .
8.9 Remark We will usually denote φ∗ e simply by φ∗ : g → h.
8.10 Examples
(a) Let gl (n, R) be the Lie algebra of real n × n-matrices with bracket product
[ A, B ] = AB − BA as defined in Remark 4.83. It is a vector space of dimension
n2 , and we let y i j : gl (n, R) → R be the linear funtional assigning to each ma-
2
trix A its i j-th entry A i j . Then we have a linear isomorphism z : gl (n, R) → Rn
given by z b ( j, i) = y i j for 1 ≤ i, j ≤ n, where b : In × In → In2 is the bijection
defined in Remark 4.27. We give gl (n, R) the manifold structure defined in
Example 2.9 (b) so that z is a coordinate map. If E = {e1 , ... , en2 } is the stan-
2
dard basis for Rn , we have a corresponding basis C = {E ji |1 ≤ i, j ≤ n} for
gl (n, R) , where E ji is the n × n-matrix with all the entries equal to zero except
the i j-th entry which is 1. We have that z(E ji ) = e b ( j, i) for 1 ≤ i, j ≤ n, and
E ji Esr = δ j r Esi
for 1 ≤ i, j, r, s ≤ n.
Now let Gl (n, R) be the group of non-singular real n × n-matrices. As the
determinant function det : gl (n, R) → R is given by
det (A) = ∑ ε (σ ) A σ (1) 1 · · · A σ (n) n

σ ∈Sn
which is a polynomial map and therefore is continuous, it follows that

Gl (n, R) = det −1 ( R − {0}) is an open submanifold of gl (n, R) . If α :
Gl (n, R) → gl (n, R) is the inclusion map, then (z ◦ α , Gl (n, R)) is a local chart
on Gl (n, R) with coordinate functions x i j = y i j ◦ α . Moreover, since
n
(AB) i j = ∑ A ik B k j
k=1
and
(A−1 ) i j = (−1) i+ j det (A ji )/det (A) ,
where A i j is the matrix obtained from A by deleting the i-th row and the
j-th column, it follows that x i j (AB−1 ) is a rational function of x kl (A) and
x kl (B) with non-zero denominator. Hence the group operations in Gl (n, R)
are smooth so that Gl (n, R) is a Lie group called the general linear group .
We will show that the Lie algebra g of Gl (n, R) may be identified with
gl (n, R) using Lemma 2.84. We have that g = Te Gl (n, R) = Te gl (n, R) which
we canonically identify with gl (n, R) by means of the linear isomorphism
ω : g → gl (n, R) given by ω = z−1 ◦ tz,e . It only remains to show that
ω ( [ X,Y ] ) = [ ω (X) , ω (Y ) ] (1)
for X,Y ∈ g , so that ω is a Lie algebra isomorphism.
As usual we let Xe and Ye be the left invariant vector fields on Gl (n, R) deter-
mined by X and Y . By Lemmas 2.84 and 2.78 (1) we have that
ω (X) i j = X (x i j )
for every X ∈ g and 1 ≤ i, j ≤ n , so that
ω ( [ X,Y ] ) i j = [ X,Y ] (x i j ) = X ( Ye (x i j )) − Y ( Xe (x i j )) .
To compute the last expression, we use that

n
x i j ◦ LA (B) = x i j (AB) = ∑ x ik (A) x k j (B)
k=1
for A, B ∈ Gl (n, R) , so that

n
Ye (x i j )(A) = YeA (x i j ) = LA ∗ (Y )(x i j ) = Y (x i j ◦ LA ) = ∑ x ik (A) Y (x k j )
k=1
and hence
n n
X ( Ye (x i j )) = ∑ X (x ik ) Y (x k j ) = ∑ ω (X) ik ω (Y ) k j = (ω (X) ω (Y )) i j
k=1 k=1
which implies (1).

LIE GROUPS 251
(b) Let V be a finite dimensional real vector space and End (V ) be the Lie algebra
of linear endomorphisms of V with bracket product [ F, G ] = F ◦ G − G ◦ F. If
B = {v1 , ..., vn } is a basis for V , we have a Lie algebra isomorphism
φ : End (V ) → gl (n, R)
given by
φ (F) = mB
B (F)
for F ∈ End(V ) . We give End (V ) the manifold structure defined in Example

2.9 (b) so that z̃ = z ◦ φ is a coordinate map and hence φ is a diffeomorphism.
Corresponding to the basis C for gl (n, R) defined in (a), we have a basis D =
{F ji |1 ≤ i, j ≤ n} for End (V ) , where F ji : V → V is the linear map given by
F ji (vk ) = δ j k vi for k = 1, ... , n, so that φ (F ji ) = E ji for 1 ≤ i, j ≤ n. Hence we
have that
F ji ◦ Fsr = δ j r Fsi
for 1 ≤ i, j, r, s ≤ n.
Now let Aut (V ) be the group of non-singular endomorphisms of V , where the
group product is composition of endomorphisms. Then
Aut (V ) = φ −1 (Gl (n, R))
is an open submanifold of End (V ) and hence a Lie group since
G ◦ F −1 = φ −1 ( φ (G) φ (F)−1 )
for F, G ∈ Aut (V ) , and the map
ψ : Aut (V ) → Gl (n, R)
induced by φ is a Lie group isomorphism.

We will show that the Lie algebra h of Aut (V ) may be identified with End (V )
using Lemma 2.84. We have that h = Te Aut (V ) = Te End (V ) which we canon-
ically identify with End (V ) by means of the linear isomorphism ω e:h→
End (V ) given by ω e = z̃−1 ◦ tz̃,e . Since D(z ◦ ψ ◦ z̃−1 )(z̃(e)) = id , it follows
from the commutative diagram in 2.70 that tz̃,e = tz,e ◦ ψ∗ which shows that
ωe = φ −1 ◦ ω ◦ ψ∗ is a Lie algebra isomorphism. We also have a commutative
diagram
ψ∗✲
h g
e
ω ω
❄ ❄
φ✲
End (V ) gl (n, R)
which shows that ψ∗ is identified with φ when we identify the Lie algebras of
Aut (V ) and Gl (n, R) with End (V ) and gl (n, R) in the way described above.
We often denote End (V ) and Aut (V ) by gl (V ) and Gl (V ), respectively. When
V = Rn , we always use the standard basis E = {e1 , ..., en } to define the isomor-
phisms φ and ψ and the basis D.
(c) Let gl (n, C) be the Lie algebra of complex n × n-matrices with bracket prod-
uct [ A, B ] = AB − BA as defined in Remark 4.83. It is a real vector space of
dimension 2 n2 , and we let y i j : gl (n, C) → C be the complex linear funtional
assigning to each matrix A its i j-th entry A i j . Then we have a linear isomor-
2
phism z : gl (n, C) → R 2 n given by z b (1, j, i) = Re y i j and z b (2, j, i) = Im y i j
for 1 ≤ i, j ≤ n, where b : I2 × In × In → I 2 n2 is the bijection defined in Re-
mark 4.27. We give gl (n, C) the manifold structure defined in Example 2.9 (b)
so that z is a coordinate map. If E = {e1 , ... , e 2 n2 } is the standard
p basis for
2 2
R , we have a corresponding basis C = {E j |1 ≤ i, j ≤ n} ∪ { −1 E ji |1 ≤
n i
i, j ≤ n} for gl (n, C) , where E ji is the n × n-matrix with all the entries equal
to zero except the i j-th entry which is 1. We have that z(E ji ) = e b (1, j, i) and
p
z( −1 E ji ) = e b (2, j, i) for 1 ≤ i, j ≤ n.
Now let Gl (n, C) be the group of non-singular complex n × n-matrices. As the
determinant function det : gl (n, C) → C is given by
det (A) = ∑ ε (σ ) A σ (1) 1 · · · A σ (n) n

σ ∈Sn
which is a polynomial map and therefore is continuous, it follows that

Gl (n, C) = det −1 ( C − {0}) is an open submanifold of gl (n, C) . If α :
Gl (n, C) → gl (n, C) is the inclusion map, then (z ◦ α , Gl (n, C)) is a local
chart on Gl (n, C) whose coordinate functions are the real and imaginary parts
of x i j = y i j ◦ α . In the same way as in (a) we see that the group operations in
Gl (n, C) are smooth so that Gl (n, C) is a Lie group called the complex gen-
eral linear group . Using 5.49 we also see that the Lie algebra of Gl (n, C) may
be identified with gl (n, C) in the same way as in (a).
(d) Let V be a finite dimensional complex vector space and End (V ) be the Lie
algebra of linear endomorphisms of V with bracket product [ F, G ] = F ◦ G −
G ◦ F. If B = {v1 , ..., vn } is a basis for V , we have a Lie algebra isomorphism
φ : End (V ) → gl (n, C)
given by
φ (F) = mB
B (F)
for F ∈ End(V ) . We give End (V ) the manifold structure defined in Example

2.9 (b) so that z̃ = z ◦ φ is a coordinate map and hence φ is a diffeomorphism.
Corresponding to thep basis C for gl (n, C) defined in (c), we have a basis D =
{F ji |1 ≤ i, j ≤ n} ∪ { −1 F ji |1 ≤ i, j ≤ n} for End (V ) , where F ji : V → V is
LIE GROUPS 253
the linear map given by F ji (vk ) = δ j k vi for k = 1, ... , n, so that φ (F ji ) = E ji

for 1 ≤ i, j ≤ n.
Now let Aut (V ) be the group of non-singular endomorphisms of V , where the
group product is composition of endomorphisms. Then
Aut (V ) = φ −1 (Gl (n, C))
is an open submanifold of End (V ) and hence a Lie group since
G ◦ F −1 = φ −1 ( φ (G) φ (F)−1 )
for F, G ∈ Aut (V ) , and the map
ψ : Aut (V ) → Gl (n, C)
induced by φ is a Lie group isomorphism.
In the same way as in (b) we see that the Lie algebra of Aut (V ) may be iden-
tified with End (V ) using Lemma 2.84, and ψ∗ is then identified with φ . We
often denote End (V ) and Aut (V ) by gl (V ) and Gl (V ), respectively. When
V = C n , we always use the standard basis E = {e1 , ..., en } to define the isomor-
phisms φ and ψ and the basis D.
GROUP REPRESENTATIONS
8.11 Definition A Lie group homomorphism φ : G → H, where H = Aut (V ) for
a vector space V or H = Gl (n, R), is called a representation of the Lie group G. The
representation φ is said to be faithful if it is injective.
We say that the representations φ i : G → Aut (V i ) for i = 1, 2 are equivalent if
there is a linear isomorphism S : V 1 → V 2 so that
S ◦ φ 1 (g) = φ 2 (g) ◦ S
for every g ∈ G. Two representations φ i : G → Gl (n, R) for i = 1, 2 are said to be
equivalent if there is a matrix A ∈ Gl (n, R) so that
A φ 1 (g) = φ 2 (g) A
for every g ∈ G.
8.12 Proposition Let ψ : G → Aut (V ) be a representation of a Lie Group G on

a finite dimensional vector space V . Then we also have a representation ψ ∗ : G →
Aut (V ∗ ) of G on the dual space V ∗ , called the contragradient representation , which
is defined by
ψ ∗ (g) = ψ (g−1 ) ∗
for g ∈ G .
PROOF : We immediately see that ψ ∗ is a group homomorphism, so it only remains

to show that it is smooth. This follows since
i◦ ψ ∗ = α ◦ ψ ◦ φ ,
where i : Aut (V ∗ ) → End (V ∗ ) is the inclusion map, φ : G → G is the smooth map

given by φ (g) = g−1 for g ∈ G , and α : End (V ) → End (V ∗ ) is the linear map given
by α (F) = F ∗ for F ∈ End (V ) .
8.13 Proposition If V is a finite dimensional vector space, we have a representa-

tion ρ kl : Gl (V ) → Aut (T kl (V )) of the Lie group Gl (V ) on the vector space T kl (V )
given by
ρ kl (F) = (F −1) ∗
for F ∈ Gl (V ) as defined in Definition 4.52, i.e.,
ρ kl (F)(T )(v1 , ... , vk , λ 1 , ... , λ l ) = T (F −1 (v1 ), ... , F −1 (vk ), λ 1 ◦ F, ... , λ l ◦ F)

for T ∈ T kl (V ) , v1 , ... , vk ∈ V and λ 1 , ... , λ l ∈ V ∗ . In particular, we have that
ρ 00 (F) = id R for every F ∈ Gl (V ).
PROOF : We immediately see that ρ is a group homomorphism, so it only remains
to show that it is smooth. If B is a basis for V and C = T kl (B) is the corresponding
basis for T kl (V ), it follows from Propositions 4.6 and 4.37 that
mCC ((F −1) ∗ ) = { mB ( F) −1 } t ⊗ · · · ⊗ { mB

B ( F)
−1 t
} ⊗ mB (F) ⊗ · · · ⊗ mB
B (F) .
| B {z } |B {z }
k l
Hence if ψ B : Gl (V ) → Gl (n, R) and ψ C : Aut (T kl (V )) → Gl (nk+l , R) , where

n = dim(V ), are the Lie group isomorphisms defined in Example 8.10 (b), we have
that
ψ C ◦ ρ kl ◦ ψ −1
B (A) = {A
−1 t
} ⊗ · · · ⊗ {A −1 } t ⊗ A ⊗ · · · ⊗ A
| {z } | {z }
k l
for A ∈ Gl (n, R), thus showing that ρ is smooth and completing the proof of the
proposition.
LIE SUBGROUPS
8.14 Definition Let G be a Lie group. A Lie group H is called a Lie subgroup of
G if it is a subgroup and an immersed submanifold of G.
8.15 Proposition If G is a Lie group, then the connected component H of G

containing e is an open Lie subgroup of G.
LIE GROUPS 255
PROOF : By Remark 2.7 we know that H is an open submanifold of G. Since the

map ρ : G × G → G given by ρ (g, h) = gh−1 is smooth, it follows that ρ (H × H) is
a connected set containing e. Hence we have that ρ (H × H) ⊂ H, which shows that
H is a subgroup of G.
To prove that H is a Lie subgroup of G, we must show that the map ω : H × H →
H given by ω (g, h) = gh−1 is smooth. Since i ◦ ω = ρ ◦ (i × i) is smooth, where
i : H → G is the inclusion map, this follows from Corollary 2.38.
8.16 Proposition If H is a Lie subgroup of G, then h is a Lie subalgebra of g.

PROOF : Since H is a Lie subgroup of G, the inclusion map i : H → G is a Lie
group homomorphism. Hence we may use Proposition 8.8 to conclude that h is a Lie
subalgebra of g.
8.17 Proposition Let G be a Lie group, and let h be a Lie subalgebra of g. Then
we have an integrable distribution ∆ on G given by
e
∆ g = {X(g)|X ∈ h}
for g ∈ G. If H is a Lie subgroup of G with Lie algebra h, then H is an integral

manifold for ∆.
PROOF : We first show that ∆ is an integrable distribution on G. If {X1 , ... , Xk } is a
f1 , ... , Xek form a local basis for ∆ on G,
basis for h, then the left invariant vector fields X
showing that ∆ is a distribution on G. Since h is a Lie algebra, there are real numbers
crij such that
k
[ Xi , X j ] = ∑ crij Xr (1)
r=1
for i, j = 1, ... , k. By applying Lg ∗ to both sides of (1) for each g ∈ G, we obtain
k
[ Xei , Xej ] = ∑ crij Xer
r=1
which by Proposition 4.96 implies that the distribution ∆ is integrable.

To prove the last part of the propositioon, let H be a Lie subgroup of G with Lie
algebra h, and let i : H → G be the inclusion map. Since
i∗ (Th H) = i∗ ◦ Lh ∗ (Te H) = Lh ∗ ◦ i∗ (Te H) = Lh ∗ (h) = ∆ h
for every h ∈ H, we see that H is an integral manifold of ∆.
8.18 Theorem Let G be a Lie group, and let h be a Lie subalgebra of g. Then
there is a unique connected Lie subgroup H of G with Lie algebra h.
PROOF : Let ∆ be the integrable distribution on G determined by the Lie subalgebra h

as defined in Proposition 8.17, and let H be the maximal connected integral manifold
for ∆ passing through e. Since
Lg ∗ (∆ h ) = ∆ gh
for every g, h ∈ G, it follows from Proposition 4.107 that Lg−1 (H) is also a maximal
connected integral manifold of ∆ passing through e and therefore coincides with H
for each g ∈ H. Hence we have that g, h ∈ H implies g−1 h ∈ H, which shows that H
is a subgroup of G.
To prove that H is a Lie subgroup of G, we must show that the map ρ : H ×H → H
given by ρ (g, h) = gh−1 is smooth. This follows from Proposition 4.102 as the map
i ◦ ρ : H × H → G is smooth, where i : H → G is the inclusion map.
Finally, it follows from Proposition 8.16 that the Lie algebra of H is ∆ e with
bracket product inherited from g. Hence the Lie algebra of H is h, and this completes
the existence part of the theorem.
To prove the uniqueness part, let H ′ be any connected Lie subgroup of G with
Lie algebra h. By Proposition 8.17 we know that H ′ is a connected integral manifold
of ∆ passing through e. Hence the Frobenius theorem implies that H ′ is an open
submanifold of H. Since both H ′ and H are subgroups of G, H ′ is also a subgroup of
H, and each left coset of H ′ in H is open in H. As H is connected, this implies that
H ′ = H.
8.19 Example We have that Gl (n, R) is a submanifold and a Lie subgroup of

Gl (n, C) since
2
z(Gl (n, R)) = z(Gl (n, C)) ∩ ( Rn × {0}) ,
2 2
where z : gl (n, C) → R n × R n is the linear isomorphism defined in Example
2
8.10 (c). If z̃ : gl (n, R) → R n is the corresponding linear isomorphism defined in
Example 8.10 (a), we have a commutative diagram
φ✲
gl (n, R) gl (n, C)
z̃ z
❄ ❄
2 i ✲ 2 2
Rn Rn × Rn
2 2 2
where φ : gl (n, R) → gl (n, C) is the inclusion map, and i : Rn → Rn × Rn is the
map given by i(a) = (a, 0) .
As described in Example 8.10, the Lie algebras eg and g of Gl (n, R) and Gl (n, C)
may be identified with gl (n, R) and gl (n, C), respectively, by means of the linear
isomorphisms ω g → gl (n, R) and ω : g → gl (n, C) given by ω
e :e e = z̃−1 ◦ tz̃,e and
LIE GROUPS 257
ω = z−1 ◦ tz,e . If ψ : Gl (n, R) → Gl (n, C) is the inclusion map induced by φ , we

have by 2.70 a commutative diagram
ψ ∗✲
e
g g
tz̃,e tz,e
❄ ❄
2 i ✲ 2 2
Rn Rn × Rn
since D(z ◦ ψ ◦ z̃−1 )(z̃(e)) = i . Combining these two diagrams, we obtain the com-
mutative diagram
ψ ∗✲
e
g g
e
ω ω
❄ ❄
φ✲
gl (n, R) gl (n, C)
which shows that ψ ∗ is identified with the inclusion map φ when we identify the Lie
algebras of Gl (n, R) and Gl (n, C) with gl (n, R) and gl (n, C) in the way described
Example 8.10.
COVERINGS
8.20 Definition A covering p : Me → M is said to be smooth if M and M e are

smooth manifolds of the same dimension and the covering projection p is smooth of
maximal rank. The manifold Me is called a covering manifold of M.
8.21 Remark It follows from the rank theorem that a covering p : M e → M is

smooth if and only if, for each evenly covered neighbourhood U in M and each sheet
eα over U, the homeomorphism pα : U
U eα → U induced by p is in fact a diffeomor-
phism.
8.22 Proposition e → M be a covering of a smooth manifold M n with

Let p : M
a connected covering space M e . Then Me is locally Euclidean and second countable,

e such that the covering is smooth.
and there is a unique smooth structure on M
PROOF : We first show that M e is second countable. For each point u ∈ M there is an
open neighbourhood U around u which is evenly covered by p. By eventually choos-
ing a smaller U using Propositions 13.38 and 13.81 in the appendix, we may assume
that U is also connected. Since M is second countable and therefore is a Lindelöf
space, M is covered by a countable family {Ui | i ∈ N } of such neighbourhoods.
Now it follows from Proposition 13.82 in the appendix that the connected com-
ponents of p−1 (Ui ) are the sheets over Ui which are second countable since they
are homeomorphic to Ui . We also have that M e is locally connected since this is true
for the manifold M and p is a local homeomorphism. By applying Proposition 4.103
to the countable open cover { p−1 (Ui ) | i ∈ N } of Me , it therefore follows that M
e is
second countable.
We next show that M e is locally Euclidean. Since p is a local homeomorphism,
e has an open neighbourhood U
each point ũ on M e1 which is mapped homeomorphi-
e1 → U1 is the induced
cally by p onto an open neighbourhood U1 of p (ũ) . If p 1 : U
homeomorphism and (x2 ,U2 ) is a local chart around p (ũ) , then x̃ = x2 ◦ p 1 maps the
e = p −1 (U1 ∩ U2 ) of ũ homeomorphically onto an open set
open neighbourhood U 1
inRn . Since Me is also Hausdorff by Proposition 13.76 in the appendix, we therefore
have that Me is locally Euclidean.
If the covering is smooth, then it follows from the rank theorem that p1 is a
diffeomorphism. Hence (x̃, U)e must be a local chart around ũ in the differetiable
e . As these local charts around each point on M
structure on M e form an atlas on M
e,
the uniqueness of the smooth structure follows from Proposition 2.2.
e ) be another local chart on M
To prove the existence, let (ỹ, V e obtained in the
same way as (x̃, U)e from a homeomorphism q 1 : Ve1 → V1 induced by p and a local
chart (y2 ,V2 ) on M. Then the coordinate transformation ỹ ◦ x̃ −1 = y2 ◦ q 1 ◦ p −1 −1
1 ◦ x2
−1
is the restriction of y2 ◦ x2 to x2 (U1 ∩ U2 ∩ V1 ∩ V2 ), showing that the local charts
e and (ỹ, Ve ) are C∞ -related. Hence we obtain a smooth structure on M
(x̃, U) e such
−1 −1
that the covering is smooth, since y2 ◦ p ◦ x̃ = ỹ ◦ x̃ .
8.23 Corollary Let M be a connected differeniable manifold, and let u ∈ M. If H

e ũ) → (M, u) with
is a subgroup of π1 (M, u) , then there is a smooth covering p : (M,
a connected covering manifold M e such that p∗ π1 (M,
e ũ) = H .
PROOF : Since a connected smooth manifold is pathwise connected, locally path-

wise connected and semi-locally simply connected, the result follows from Theorem
13.104 in the appendix and Proposition 8.22.
LIE GROUPS 259
8.24 Proposition If p : Me → M is a smooth covering, then a lifting f˜ : N → M

e
of a smooth map f : N → M is smooth.
PROOF : Let u ∈ N, and choose an evenly covered neighbourhood U of f (u). Let
Ueα be the sheet over U containing f˜ (u) , and let sα : U → Ueα be the inverse of
the diffeomorphism induced by p and iα : U eα → M e be the inclusion map. Then
eα ) is an open neighbourhood of u, and we have that f˜|V = iα ◦ sα ◦ f |V
V = f˜−1 (U
which shows that f˜ is smooth at u.
8.25 Proposition Let G be a connected Lie group with unit element e, and let p :
e ẽ) → (G, e) be a smooth covering of (G, e) with a connected covering manifold
(G,
Ge . Then there is a unique Lie group structure on G
e with ẽ as unit element such that
the covering projection p is a Lie group homomorphism.
PROOF : Let ρ : (G × G , (e, e)) → (G, e) be the smooth map given by ρ (g, h) =
gh−1 for g, h ∈ G. Then it follows from Proposition 13.109 in the appendix that
e × G,
ρ∗ ◦ (p × p)∗ π1 (G e (ẽ, ẽ)) ⊂ p∗ π1 (G,
e ẽ)
since
ρ∗ ◦ (p × p)∗ ( [ (σe1 , σe2 ) ] ) = [ ρ ◦ (p ◦ σe 1 , p ◦ σe 2 ) ] = [ p ◦ σe 1 ] ∗ [ p ◦ σe2 ] −1

e ẽ)
e 2 ] −1 ) ∈ p∗ π1 (G,
= p∗ ( [ σe1 ] ∗ [ σ
for every pair of closed paths σe1 and σe2 in G e at ẽ .

By the lifting theorem and Proposition 8.24 there is therefore a unique smooth
e×G
map ρe : (G e , (ẽ, ẽ)) → (G,
e ẽ) which is a lifting of ρ ◦ (p × p) , so that the diagram
ρe ✲
e × G,
(G e (ẽ, ẽ)) e ẽ)
(G,
p× p p
❄ ❄
(G × G , (e, e)) ✲ (G, e)
ρ
is commutative.
We now let ge−1 = ρe (ẽ, ge) and gee g, e
h = ρe (e h −1 ) for ge, e e . To prove that this
h∈G
e with ẽ as unit element, we must show that

defines a Lie group structure on G
e,
geẽ = ẽ ge = ge and gege−1 = ge−1 ge = ẽ for every ge ∈ G (1)
and that
ge
(e h) e
k = ge(e
hek) for every ge, e
h, e e.
k∈G (2)
To show the first part of (1), we use the fact that G is a group so that the two
maps µ 1 , µ 2 : (G, e) → (G, e) given by µ 1 (g) = g e and µ 2 (g) = e g and the iden-
tity map id : (G, e) → (G, e) are all equal. Hence the same is true for the maps
e 1, µ
µ e ẽ) → (G,
e 2 : (G, e ẽ) given by µ
e 1 (e e 2 (e
g) = ge ee and µ g) = eege and the identity map
e : (G,
id e ẽ) → (G,e ẽ) , since they are the unique liftings of µ 1 ◦ p , µ 2 ◦ p and id ◦ p ,
respectively. Indeed, we have that
e 1 (e
p◦ µ g) = p ◦ ρe (e
g, ρe (e
e, ee)) = ρ (p (e
g), ρ (e, e)) = µ 1 ◦ p (e
g)
and
e 2 (e
p◦ µ g) = p ◦ ρe (e
e, ρe (e
e, ge)) = ρ (e, ρ (e, p (e
g))) = µ 2 ◦ p (e
g)
for every ge ∈ G e = id ◦ p . This completes the proof of

e , and we also have that p ◦ id
the first part of (1), and the second part of (1) and (2) are proved in the same way.
Finally, we have that
ge
p (e h) = p ◦ ρe (e e, e
g, ρe (e g), ρ (e, p (e
h)) = ρ (p (e g) p (e
h))) = p (e h)
for every ge, e e , which shows that the covering projection p is a Lie group homo-
h∈G
morphism, and this completes the proof of the existence part of the proposition.
To show the uniqueness part, suppose that Ge has any Lie group structure with ee
as unit element such that the covering projection p is a Lie group homomorphism,
and let e
λ :G e×Ge→G e be the smooth map defined by e g ,e
λ (e h) = gee
h −1 for ge, e e.
h∈G
Then we have that
p◦e g ,e
λ (e g) p (e
h) = p (e g,e
h) −1 = ρ ◦ (p × p )(e h)
for every ge, e e , which shows that e

h∈G λ is a lifting of ρ ◦ (p × p) with e
λ (e
e , ee) = ee.
e
Hence λ = ρe by the unique lifting theorem, thus showing that the Lie group structure
e must coincide with the one defined in the first part of the proof and therefore is
on G
uniquely determined by the conditions in the proposition.
8.26 Corollary Let G be a connected Lie group with unit element e, and let H be
e ẽ) → (G, e) such that
a subgroup of π1 (G, e) . Then there is a smooth covering p : (G,
e ẽ) = H , where the covering manifold G
p∗ π1 (G, e is a Lie group with unit element ẽ
and the covering projection p is a Lie group homomorphism.
LIE GROUPS 261
8.27 Proposition Let φ : G → H be a Lie group homomorphism between the

connected Lie groups G and H such that φ∗ : g → h is a Lie algebra isomorphism.
Then φ is a smooth covering projection. If H is simply connected, then φ is a diffeo-
morphism.
PROOF : We have that φ∗ g is an isomorphism for each g ∈ G, since the left
translations Lg and Lφ (g) are diffeomorphisms in G and H, respectively, with
Lφ (g) ◦ φ ◦ Lg−1 = φ . By the inverse function theorem it follows that φ is a local dif-
feomorphism which therefore has a discrete kernel and is open by Proposition 13.86
in the appendix. Since φ (G) contains an open neighbourhood of the unit element e
in H, it follows from Proposition 13.112 in the appendix that φ is a surjection. The
proposition now follows from Propositions 13.111 and 13.99 in the appendix.
8.28 Theorem Let G and H be Lie groups, and let Φ : g → h be a Lie algebra
homomorphism. If G is simply connected, then there is a unique Lie group homo-
morphism φ : G → H with φ∗ = Φ.
PROOF : We consider the graph k of Φ given by k = {(X, Φ (X))|X ∈ g} . As Φ
is a Lie algebra homomorphism, k is a Lie subalgebra of g × h = L (G × H) , so
by Theorem 8.18 there is a unique connected Lie subgroup K of G × H whose Lie
algebra is k.
Now let π1 : G × H → G and π2 : G × H → H be the projections on the first and
second factor, and let ω = π1 |K . Then ω : K → G is a Lie group homomorphism,
and ω∗ : k → g is a Lie algebra isomorphism since
ω∗ (X, Φ (X)) = X
for X ∈ g. By Proposition 8.27 it follows that ω is a diffeomorphism, and we have a

Lie group homomorphism φ : G → H defined by φ = π2 ◦ ω −1 . Since
φ∗ (X) = π2 ∗ (X, Φ (X)) = Φ (X)
for X ∈ g, we have that φ∗ = Φ, showing that φ satisfies the conditions of the theorem.
To show uniqueness of φ , suppose that ψ : G → H is another Lie group homo-
morphism with ψ∗ = Φ. Then the map θ : G → G × H defined by
θ (g) = (g, ψ (g))
for g ∈ G is a Lie group homomorphism which is a one-to-one immersion. By Re-

mark 2.31 its image G ′ is a connected Lie subgroup of G × H with Lie algebra k,
since
θ∗ (X) = (X, Φ (X))
for X ∈ g. Hence we have that G ′ = K by Theorem 8.18, showing that ψ (g) = φ (g)
for every g ∈ G.
THE EXPONENTIAL MAP
8.29 Proposition Let G be a Lie group, and let X ∈ g. Then there is a unique
one-parameter subgroup φX : R → G with φ X′ (0) = X.
The left invariant vector field Xe on G determined by X is complete, with flow
γ : R × G → G given by
γ (t, g) = Lg ◦ φX (t) .
In particular, φX is the maximal integral curve for Xe with initial condition φX (0) = e.
PROOF : We use the local chart (r, R) on the Lie group R, where r : R → R is the
d

identity map. Then the tangent vector dr is a basis for the Lie algebra of R, which
0
we may also identify with R, and we have a unique Lie algebra homomorphism
ΦX : R → g such that !
d

ΦX dr =X .
0
By Theorem 8.28 there is a unique Lie group homomorphism φX : R → G with
φX ∗ = ΦX , and hence the first part of the proposition follows from Remark 2.83
which implies that !

d
φ X′ (0) = φX ∗ dr .
0
To prove the last part of the proposition, let Ls : R → R be the left translation on
R defined by Ls (t) = s + t. Then
!
d
d

Ls ∗ dr = dr
t Ls (t)
d
for every s,t ∈ R, which shows that dr is the left invariant vector field on R deter-

d
d
mined by dr . Hence it follows from Proposition 8.7 that the vector fields dr and
0
Xe are φX -related, so that
!
d

φ X′ (t) = φX ∗ dr = Xe (φX (t))
t
for every t ∈ R. Since Xe is left invariant, the last formula is still true if φX is replaced
by Lg ◦ φX , and this completes the proof of the proposition.
LIE GROUPS 263
8.30 Definition If G is a Lie group, we define the exponential map exp : g → G

by
exp (X) = φX (1)
for X ∈ g.
8.31 Proposition The one-parameter subgroup φX : R → G defined in Proposi-

tion 8.29 is given by
φX (t) = exp (t X)
for t ∈ R.
PROOF : Using the Lie group homomorphism µ t : R → R defined by µ t (s) = t s for
s ∈ R, we obtain a one-parameter subgroup φX ◦ µ t : R → G satisfying
(φX ◦ µ t ) ′ (0) = t φ X′ (0) = t X
by Definition 2.69. Hence we have that
φX ◦ µ t = φ t X ,
which implies that

φX (t) = φ t X (1) = exp (t X) .
8.32 Corollary Let G be a Lie group, and let X ∈ g. Then

(1) exp (t1 + t2 ) X = (exp t1 X ) (exp t2 X ) for t1 ,t2 ∈ R ,
(2) exp (− t X ) = (exp t X ) −1 for t ∈ R .
8.33 Corollary Let G be a Lie group, and let φ : G → G be the smooth map
given by φ (g) = g−1 for g ∈ G. Then we have that
φ ∗ g = − ( R g−1 ◦ L g−1 ) ∗ g
for g ∈ G. In particular, the linear map φ ∗ e : g → g is given by φ ∗ e ( X) = −X for

X ∈ g.
PROOF : By Proposition 8.31 and Corollary 8.32 (2) we have that
φ ◦ Lg ◦ φ X (t) = exp (− t X ) g−1 = R g−1 ◦ φ −X (t)
for t ∈ R . Hence it follows from 2.70 that
φ ∗ ◦ (L g ) ∗ ( X) = (φ ◦ Lg ◦ φ X ) ′ (0) = − (R g−1 ) ∗ ( X)
for every X ∈ g.
8.34 Proposition Let G be a Lie group, and let X,Y ∈ g. Then we have that
[ X,Y ] = 0 if and only if
exp (t X) exp (sY ) = exp (sY ) exp (t X)
for all s,t ∈ R.
PROOF : Let α and β be the flows of the left invariant vector fields Xe and Ye on G
determined by X and Y , respectively. Then we have that
α (t, β (s, e)) = exp (sY ) exp (t X)
and
β (s, α (t, e)) = exp (t X) exp (sY )
for every s,t ∈ R by Propositions 8.29 and 8.31. The result therefore follows from
Propositions 4.91 and 4.92.
8.35 Proposition Let G be a Lie group, and let X,Y ∈ g with [ X,Y ] = 0 . Then
we have that
exp t ( X + Y ) = exp (t X) exp (t Y )
for every t ∈ R.
PROOF : Using Corollary 8.32 (1) and Proposition 8.34 we see that the smooth curve
γ : R → G given by
γ (t) = exp (t X) exp (t Y )
for t ∈ R, is a one-parameter subgroup of G, and we have that γ ′ (0) = X + Y by
Remark 2.75. The result therefore follows from Proposition 8.31.
8.36 Lemma Let V be a vector space of dimension n , and let v ∈ V . If ψ v : R → V

is the curve on V defined by ψ v (t) = t v for t ∈ R , and we canonically identify T0V
with V as in Lemma 2.84, then we have that ψ v′(0) = v.
PROOF : Let x : V → Rn be a linear isomorphism, and let x(v) = a. We identify T0V
with V by means of the linear isomorphism ω 0 : T0V → V given by ω 0 = x−1 ◦ tx,0 .
Now x ◦ ψ v (t) = t a for t ∈ R so that
ψ v′ (0) = [ x, (x ◦ ψ v ) ′ (0) ] ψ v (0) = [ x, a ] 0
which implies that

ω 0 ◦ ψ v′(0) = x−1 (a) = v .
8.37 Proposition A finite dimensional vector space V is an abelian Lie group

with addition as its group operation, and its Lie algebra may be identified with V .
The exponential map exp : V → V is then identified with id V : V → V .
LIE GROUPS 265
PROOF : Since that map ψ v : R → V , defined as in Lemma 8.36 by ψ v (t) = t v for

t ∈ R , is a one-parameter subgroup of V with ψ v′ (0) = v, we have that
exp (v) = ψ v (1) = v
for v ∈ V .
8.38 Proposition The exponential map exp : g → G is smooth, and

exp ∗ 0 : g → g is the identity map. Hence exp maps an open neighbourhood U of
0 ∈ g diffeomorphically onto an open neighbourhood V of e ∈ G .
We let log : V → U denote the inverse of the diffeomorphism induced by exp .
PROOF : Let Y be the vector field on G × g defined by
e
Y (g, X) = (X(g), 0) = (Lg ∗ (X), 0) .
By Proposition 8.29 the flow γ : R × G × g → G × g of Y is given by
γ (t, g, X) = (Lg ◦ φX (t), X) ,
and if π : G × g → G is the projection on the first factor, we have that
exp (X) = π ◦ γ (1, e, X)
which shows that exp is smooth.

Using the curves φX and ψX on G and g, respectively, defined in Proposition 8.29
and Lemma 8.36, we have that
exp ∗ (X) = (exp ◦ ψX ) ′ (0) = φ X′ (0) = X
for every X ∈ g by 2.70 and Proposition 8.31, showing that exp ∗ 0 = id . The last part
of the proposition now follows from the inverse function theorem.
8.39 Remark If x : g → Rn is a linear isomorphism and y = x ◦ log , then (y,V )

is a local chart around e on G, called a canonical chart or a canonical coordinate
system on G. Let E = {e1 , ..., en } be the standard basis for Rn and C = {X1 , ..., Xn }
be a basis for g so that x(Xi ) = ei for i = 1, ..., n. Then the coordinate map y is given
by
n
y( exp ( ∑ ai Xi )) = a
i=1
for a ∈ x(U) .
8.40 Proposition If φ : G → H is a Lie group homomorphism, then
exp ◦ φ ∗ = φ ◦ exp
so that we have a commutative diagram
φ∗✲
g h
exp exp
❄ ❄
φ✲
G H
PROOF : If X ∈ g, then it follows from 2.70 and Proposition 8.29 that φ ◦ φX : R → H

is a one-parameter subgroup of H with
(φ ◦ φX ) ′ (0) = φ ∗ X .
Hence we have that
exp (φ ∗ X) = φ ◦ φX (1) = φ (exp X) .
8.41 Proposition Let H be a Lie subgroup of G, and let X ∈ g. Then X ∈ h if and

only if exp (t X) ∈ H for every t ∈ R.
PROOF : Since H is a Lie subgroup of G, the inclusion map i : H → G is a Lie group
homomorphism, and we have that
exp ◦ i∗ = i ◦ exp
by Proposition 8.40. Hence it follows that exp (t X) ∈ H whenever X ∈ h and t ∈ R.

Assuming conversely that exp (t X) ∈ H for every t ∈ R, it follows by Proposi-
tions 8.17 and 4.102 that the map φ : R → H , defined by
i ◦ φ (t) = exp (t X)
for t ∈ R , is smooth and therefore a one-parameter subgroup of H with i∗ ◦ φ ′ (0) =

X . This shows that X ∈ h.
8.42 Proposition Let H be a subgroup of a Lie group G, and let h be a subspace

of g. Suppose that ψ : U → V is a diffeomorphism from an open neighbourhood U
of 0 ∈ g where exp is one-to-one onto an open neighbourhood V of e ∈ G , and that
exp (U ∩ h ) = ψ (U ∩ h ) = V ∩ H . (1)
Then H is a submanifold and a Lie subgroup of G, h is a Lie subalgebra of g, and h

is the Lie algebra of H.
LIE GROUPS 267
PROOF : Choose a basis B = {X1 , ..., Xk } for h, and extend B to a basis C =

{X1 , ..., Xn } for g. Let x : g → Rn be the linear isomorphism given by x(Xi ) = ei for
i = 1, ..., n , where E = {e1 , ..., en } is the standard basis for Rn , and let y = x ◦ ψ −1
so that
n
y( ψ ( ∑ ai Xi )) = a
i=1
for a ∈ x(U) . Then (y,V ) is a local chart around e on G having the submanifold
property
y(V ∩ H) = y(V ) ∩ (Rk × {0}) .
Hence the same is true for the local chart (y ◦ Lg−1 , gV ) around g for each g ∈ H,
which shows that H is a submanifold and a Lie subgroup of G.
Now it follows from Proposition 8.16 that the Lie algebra L (H) of H is a Lie
subalgebra of g, and it only remains to prove that L (H) = h. If X ∈ L (H), then
exp (t X) ∈ H for every t ∈ R by Proposition 8.41. We choose a t 6= 0 with tX ∈ U
and exp (t X) ∈ V . Then it follows from (1) and the fact that exp is one-to-one on
U that tX ∈ U ∩ h , which shows that X ∈ h. Since L (H) and h are subspaces of g
having the same dimension k with L (H) ⊂ h, we conclude that L (H) = h.
CLOSED SUBGROUPS
8.43 Theorem Let G be a Lie group, and let H be a closed subgroup of G. Then
H is a submanifold and a Lie subgroup of G.
Let h denote the Lie algebra of H and h′ be a subspace of g so that g = h ⊕ h′ ,
and let φ : g → G be the map given by
φ (X + X ′) = exp (X ′ ) exp (X)
for X ∈ h and X ′ ∈ h′ . Then there are open neighbourhoods W and W ′ of 0 in h

and h′ , respectively, and an open neighbourhood V of e in G so that φ induces a
diffeomorphism ψ : U → V having the properties in Proposition 8.42 where U =
W + W ′.
PROOF : We will prove the theorem by showing that the set
h = {X ∈ g|exp (t X) ∈ H for all t ∈ R}
is a subspace of g which together with the map ψ : U → V described above satisfies

formula (1) in Proposition 8.42.
Given a sequence {Xi }∞ i=1 of elements in g which converges to an element
X ∈ g, and a sequence {ti }∞
i=1 of positive real numbers which converges to 0 so that
exp (ti Xi ) ∈ H for all i, we claim that X ∈ h. If t ∈ R, we let ki be the largest integer
≤ t/ti so that
t/ti − 1 < ki ≤ t/ti
for all i. Then the sequence {ki ti Xi }∞
i=1 converges to t X, and
exp (ki ti Xi ) = exp (ti Xi ) ki ∈ H
which shows that exp (t X) ∈ H , since H is closed. As this is true for all t ∈ R, we
have that X ∈ h.
Using this, we can now show that h is a subspace of g. It is clearly closed under
multiplication with scalars, so we only need to show that it is closed under addition.
Suppose that U ′ is an open neighbourhood of 0 ∈ g which is mapped diffeomorphi-
cally by exp onto an open neighbourhood V ′ of e ∈ G . Let Y, Z ∈ h, and consider
the smooth curve γ : R → G given by
γ (t) = exp (t Y ) exp (t Z)
for t ∈ R. We choose an open interval I with 0 ∈ I ⊂ γ −1 (V ′ ) , and let α : I →

U ′ be the smooth curve given by α (t) = log ◦ γ (t) for t ∈ I. Then it follows from
Remark 2.75 and Proposition 8.38 that α ′ (0) = Y + Z , and we have that α (0) = 0
and exp (α (t)) = γ (t) ∈ H for t ∈ I.
Now let {ti }∞i=1 be a sequence of positive real numbers in I which converges to 0,
and let Xi = α (ti )/ti for i = 1, 2, 3.... Then the sequence {Xi }∞
i=1 converges to Y + Z,
and exp (ti Xi ) = exp (α (ti )) ∈ H for all i. This shows that Y + Z ∈ h and completes
the proof that h is a subspace of g.
Choose a basis B = {X1 , ..., Xk } for h, and extend B to a basis C = {X1 , ..., Xn }
for g. Then
g = h ⊕ h′ ,
where h′ = L(Xk+1 , ..., Xn ) is the subspace of g spanned by {Xk+1 , ..., Xn } . Let φ :
g → G be the smooth map given by
φ (X + X ′) = exp (X ′ ) exp (X)
for X ∈ h and X ′ ∈ h′ . Using the curves φX and ψX on G and g, respectively, defined

in Proposition 8.29 and Lemma 8.36, we have that
φ ∗ (Xi ) = (φ ◦ ψXi ) ′ (0) = φ X′i (0) = Xi
for i = 1, ..., n by 2.70 and Proposition 8.31, showing that φ ∗ 0 = id G . Hence there
are open neighbourhoods W and W ′ of 0 in h and h′ , respectively, with W +W ′ ⊂ U ′ ,
such that φ is a diffeomorphism from W + W ′ onto an open neighbourhood V of e.
We claim that W ′ can be chosen so that exp (X ′ ) ∈ / H when X ′ ∈ W ′ − {0}.
′ ∞ ′
Otherwise, there would be a sequence {Xi }i=1 in h − {0} which converges to 0
with exp (Xi′ ) ∈ H for all i. We have a norm on g so that the linear isomorphism
x : g → Rn given by x(Xi ) = ei for i = 1, ..., n is an isometry, where E = {e1 , ..., en }
LIE GROUPS 269
is the standard basis for Rn . Let ti = kXi′ k and Yi′ = Xi′ / ti for i = 1, 2, 3.... As the set
S′ = {Y ′ ∈ h′ |kY ′ k = 1} is compact, there is a subsequence {Yi′r }∞ ′ ∞
r=1 of {Yi }i=1 which
′ ∞
converges to an element Y ∈ S , and {tir }r=1 is a sequence of positive real numbers
which converges to 0 so that exp (tir Yir ) ∈ H for all r. This shows that Y ∈ h , which
is impossible since h ∩ S′ = 0/ .
We now clearly have that φ (U ∩ h ) ⊂ V ∩ H , where U = W + W ′ . To prove
the reverse inclusion, let g ∈ V ∩ H . Then there are elements X ∈ W and X ′ ∈ W ′
with g = exp (X ′ ) exp (X) . Since g, exp (X) ∈ H , we have that exp (X ′ ) ∈ H which
implies that X ′ = 0. Hence it follows that g ∈ φ (U ∩ h ) , which completes the proof
of the theorem.
8.44 Theorem Let H be a closed subgroup of a Lie group G. Then the set
G/H = {gH|g ∈ G} of left cosets has a unique manifold structure so that the natural
projection π : G → G/H is smooth, and given a map f : G/H → M into a smooth
manifold M, then f is smooth if and only if f ◦ π is smooth.
For each g ∈ G there is an open neighbourhood Ug of π (g) in G/H and a smooth
map sg : Ug → G so that π ◦ sg = ĩg , where ĩg : Ug → G/H is the inclusion map.
PROOF : We give G/H the quotient topology, consisting of the sets O ⊂ G/H such
that π −1 (O) are open in G . Then the projection π is continuous, and it is also an
open map since [
π −1 (π (U)) = UH = Uh
h∈H
is open for every open set U in G .

We next show that G/H is Hausdorff. Let ρ : G × G → G be the continuous map
given by ρ (g1 , g2 ) = g−1 / ρ −1 (H) . Given
2 g1 . Then g1 H 6= g2 H if and only if (g1 , g2 ) ∈
two distinct cosets g1 H and g2 H in G/H, there are therefore open neighbourhoodsU1
and U2 of g1 and g2 in G, respectively, so that (U1 ×U2) ∩ ρ −1 (H) = 0/ , as ρ −1 (H) is
closed in G× G. This implies that π (U1 ) and π (U2 ) are disjoint open neighbourhoods
of g1 H and g2 H in G/H.
The space G/H is also second countable. If {Oi | i ∈ N} is a countable basis
for the topology in G, then {π (Oi ) | i ∈ N} is a countable basis for the topology in
G/H. For each g ∈ G we let L̃g : G/H → G/H denote the left translation on G/H
induced by the left translation Lg : G → G on G. It is the homeomorphism defined by
L̃g (g′ H) = gg′ H for g′ ∈ G so that π ◦ Lg = L̃g ◦ π .
Now let ψ : W + W ′ → V be the diffeomorphism given in Theorem 8.43, and
choose open neighbourhoods W0 and W0′ of 0 contained in W and W ′ , respectively,
such that ψ maps W0 +W0′ onto an open neighbourhood Ve of e with Ve−1Ve ⊂ V . We
contend that π ◦ ψ induces a homeomorphism φ : W0′ → π (Ve ) . For each gH ∈ π (Ve )
there are X ∈ W0 and X ′ ∈ W0′ with ψ (X + X ′ ) = g . Using that exp (X) ∈ H it
follows that φ (X ′ ) = gH , thus showing that φ is surjective.
To show that it is injective, assume that φ (X ′ ) = φ (Y ′ ) where X ′ ,Y ′ ∈ W0′ . Then
we have that
exp (−Y ′ ) exp (X ′ ) = h ∈ H ∩V ,
and there is an X ∈ W with exp (X) = h . From this it follows that
ψ (X ′ ) = ψ (X + Y ′ )
which shows that X ′ = Y ′ .

The bijection φ is clearly continuous. It is also open since
φ (O) = φ (W0 + O)
is open for every open set O in W0′ . This completes the proof that φ is a homeo-
morphism.
Let x : h′ → Rn−k be a linear isomorphism, and consider for each g ∈ G the open
neighbourhood Ug = π (Vg ) of π (g) , where Vg = Lg (Ve ) , and the map yg : Ug → Rn−k
given by yg = x ◦ φ −1 ◦ L̃g−1 . We contend that {(yg ,Ug ) | g ∈ G} is an atlas on G/H.
To see this, let π (g1 exp (X1′ )) = π (g2 exp (X2′ )) , where X1′ , X2′ ∈ W0′ , be a point
in Ug1 ∩ Ug2 . Then there is an h ∈ H so that
g−1 ′ ′
2 g1 exp (X1 ) h = exp (X2 ) ,
and we can choose an open neighbourhood W1′ of X1′ contained in W0′ such that
g−1 ′
2 g1 exp (W1 ) h ⊂ Ve .
We have a smooth map α : W1′ → W0′ given by
α (X ′ ) = ρ ◦ ψ −1 (g−1 ′
2 g1 exp (X ) h)
for X ′ ∈ W1′ , where ρ : g → h′ is the projection defined by ρ (X + X ′) = X ′ for X ∈ h

and X ′ ∈ h′ . Since
φ ◦ α (X ′ ) = π ( g−1 ′ ′
2 g1 exp (X )) = L̃g−1 ◦ L̃g1 ◦ φ (X )
2
for every X ′ ∈ W1′ , it follows that
φ −1 ◦ L̃g−1 ◦ L̃g1 ◦ φ | W1′ = α ,

2
which shows that yg2 ◦ y−1

g1 is smooth. Hence G/H is a smooth manifold, and the
projection π : G → G/H is smooth since
φ −1 ◦ L̃g−1 ◦ π (g′ ) = ρ ◦ ψ −1 ◦ Lg−1 (g′ )
for every g ∈ G and g′ ∈ Vg .

For each g ∈ G, we let sg : Ug → G be the smooth map given by
sg = ig ◦ Lg ◦ ψ ◦ φ −1 ◦ L̃g−1 ,
where ig : Vg → G is the inclusion map. Then we have that π ◦ sg = ĩg , which proves
LIE GROUPS 271
the last part of the theorem. Suppose that f : G/H → M is a map into a smooth
manifold M so that f ◦ π is smooth. Then f is also smooth since
f | Ug = f ◦ π ◦ sg
for every g ∈ G.
To prove the uniqueness of the manifold structure, let (G/H) 1 and (G/H) 2 de-
note G/H with two manifold structures satisfying the first part of the theorem. Then
we immediately see that the identity map id : (G/H) 1 → (G/H) 2 and its inverse
are smooth, since π ◦ id = π and π ◦ id −1 = π . This shows that the two manifold
structures must coincide.
8.45 Proposition Let φ : G → H be a Lie group homomorphism. Then ker φ is

a closed submanifold and a Lie subgroup of G with Lie algebra ker φ ∗ .
PROOF : Since ker φ is a closed subgroup of G, it follows from Theorem 8.43 that
it is a closed submanifold and a Lie subgroup of G. Let k be its Lie algebra, and
suppose that X ∈ g. Then it follows from Propositions 8.40 and 8.41 that X ∈ k if and
only if
exp (t φ ∗ (X)) = φ ◦ exp (t X) = e
for every t ∈ R, which is equivalent to φ ∗ (X) = 0 . Hence we have that k = ker φ ∗ .
MATRIX GROUPS
8.46 Let V be a finite dimensional complex vector space, and let B = {v1 , ..., vn }
be a basis for V . Then we have a linear isomorphism ψ : V → C n given by
n
ψ (v) = a when v= ∑ aj vj ,
j=1
where a is written as a column vector. We define a norm in V by letting kvk = kak so

that ψ is an isometric isomorphism, using the norm in C n given by
n 1/2
2
kak = ∑ |aj | .
j=1
Let φ : gl (V ) → gl (n, C) be the linear isomorphism defined by
φ (F) = mB
B (F)
for F ∈ gl (V ) . If φ (F) = A , then ψ ◦ F ◦ ψ −1 is the linear map T ∈ gl (C n ) given

by T (a) = A a for a ∈ C n , so that
kFk = kT k = max {k T (a) k | k a k ≤ 1}

( 2 !1/2 ) (1)
n n n
= max ∑ ∑ Ai j a j ∑ 2
|aj | ≤ 1

i=1 j=1 j=1
by Propositions 13.120 and 13.123 in the appendix. We also define a norm in gl (n, C)
by letting kAk = kFk, so that φ is an isometric isomorphism. If C = {e1 , ..., en } is
the standard basis for C n and A s is the s-th column in A, we see that
|A rs | ≤ kA s k ≤ kAk
for all r and s by taking a = e s in formula (1).
8.47 We have an inner product in C n given by

n
<a, b > = at b = ∑ aj bj
j=1
for column vectors a and b in C n , so that kak = <a, a>1/2 . The adjoint of a matrix
A ∈ gl (n, C) is the unique matrix A∗ ∈ gl (n, C) which satisfies the relation
<A a, b > = <a, A∗ b >
for all a, b ∈ C n . Since
t t
<A a, b > = (A a) t b = at A t b = at A b = <a, A b > ,
t
we have that A∗ = A .
Using the Cauchy Schwartz inequality we see that
kA ak2 = <A a, A a> = <a, A∗ A a> ≤ kA∗ Ak kak2
so that
kAk2 ≤ kA∗ Ak ≤ kA∗ k kAk .
Hence we have that
kAk ≤ kA∗ k ≤ k(A∗ )∗ k = kAk
which implies that kA∗ k = kAk .
A matrix A ∈ gl (n, C) is said to be unitary if it satisfies the relation
<A a, A b > = <a, b >
for all a, b ∈ C n , which is equivalent to assertion that A−1 = A∗ .

LIE GROUPS 273
8.48 Definition An n × n-matrix A is said to be upper triangular if A i j = 0 for

1 ≤ j < i ≤ n , so that all its elements below the diagonal are zero.
8.49 Remark We see that the product AB of two upper triangular n × n-matrices
A and B is an upper triangular n × n-matrix with diagonal elements
(AB) ii = A ii B ii
for i = 1, ..., n .
8.50 Proposition Let V be an n-dimensional complex vector space, and let

T ∈ gl (V ). Then there is a basis B = {v1 , ..., vn } for V so that mB
B (T ) is an upper
triangular matrix.
PROOF : We prove the proposition by induction on n. It is obviously true for n = 1,
and we will prove it for n > 1 assuming that it is true for every complex vector space
V of dimension n − 1.
Let v1 be an eigenvector for T with eigenvalue a 11 , and extend {v1 } to a basis
{v1 , w2 , ...wn } for V . Let V1 = L(v1 ) and W = L(w2 , ...wn ) so that V = V1 ⊕ W , and
consider the linear map S = π ◦ T | W , where π : V → W is the projection on W . As
dim(W ) = n − 1 , there is by the induction hypothesis a basis C = {v2 , ..., vn } for W
so that the matrix mCC (S) is upper triangular, i.e.,
j
S(v j ) = ∑ a i j vi
i=2
for 2 ≤ j ≤ n and complex numbers a i j . Since
T (v j ) = a 1 j v1 + S(v j )
for 2 ≤ j ≤ n and complex numbers a 1 j , this completes the proof of the proposition.
8.51 Corollary If A ∈ gl (n, C) , there is a matrix B ∈ Gl (n, C) so that BAB−1

is upper triangular.
PROOF : Let T ∈ gl (C n ) be the linear map so that mEE (T ) = A , where E is the
standard basis for C n , and choose a basis B for C n so that mB B (T ) is an upper
triangular matrix C. If we let B = mEB (id) , where id : C n → C n is the identity map,
we have that
BAB−1 = mEB (id) mEE (T ) mB B

E (id) = mB (T ) = C .
8.52 Proposition The exponential map exp : gl (n, C) → Gl (n, C) is given by

∞
Ak
exp (A) = ∑ , (1)
k=0 k!
and we have that

det( exp (A)) = e tr (A) (2)
for A ∈ gl (n, C) .
PROOF : By 8.46 and Proposition 13.126 in the appendix we have a continuous group
homomorphism φ : R → Gl (n, C) given by
∞
(tA)k
φ (t) = ∑ k!
k=0
for t ∈ R. Using the local chart (z ◦ α , Gl (n, C)) on Gl (n, C) defined in Example
8.10 (c), we see that
∞
zi (Ak ) t k
zi ◦ φ (t) = ∑
k=0 k!
is a power series with infinite radius of convergence for i = 1, ..., 2 n2 . By differentiat-

ing the series term by term at t = 0, we have that (zi ◦ φ ) ′ (0) = zi (A) for each i. This
shows that φ is a one-parameter subgroup of Gl (n, C) with φ ′ (0) = A, so that
∞
Ak
exp (A) = φ (1) = ∑
k=0 k!
by Definition 8.30.
Using Corollary 8.51 we now choose a B ∈ Gl (n, C) so that BAB−1 = C is an
upper triangular matrix. Then we have that
det( exp (A)) = det( B exp (A) B−1 ) = det( exp (BAB−1 ))
n
= ∏ e Ci i = e tr (C) = e tr (A)
i=1
by Proposition 13.126 in the appendix, Remark 8.49 and Corollary 4.65.
8.53 Remark By Example 8.19 and Proposition 8.40 the exponential map exp :
gl (n, R) → Gl (n, R) is induced by exp : gl (n, C) → Gl (n, C) and satisfies the same
two formulas for every A ∈ gl (n, R) .
8.54 Corollary Let V be a finite dimensional real or complex vector space. Then
the exponential map exp : gl (V ) → Gl (V ) is given by
∞
Fk
exp (F) = ∑ , (1)
k=0 k!
LIE GROUPS 275
and we have that

det( exp (F)) = e tr (F) (2)
for F ∈ gl (V ) .
PROOF : If B = {v1 , ..., vn } is a basis for V , we have an isometric algebra isomor-
phism
φ : gl (V ) → gl (n, K)
where K is either R or C, which is given by
φ (F) = mB
B (F)
for F ∈ gl (V ) . Moreover, φ induces a Lie group isomorphism
ψ : Gl (V ) → Gl (n, K)
so that ψ ∗ = φ when we identify the Lie algebras of Gl (V ) and Gl (n, K) with gl (V )

and gl (n, K) as described in Example 8.10. Using Proposition 8.40 we therefore have
a commutative diagram
φ✲
gl (V ) gl (n, K)
exp exp
❄ ❄
ψ✲
Gl (V ) Gl (n, K)
which implies formula (1) and (2) since

!
∞ ∞
φ (F) k Fk
ψ ◦ exp (F) = exp ◦ φ (F) = ∑ k!
=φ ∑
k=0 k=0 k!
and
det( exp (F)) = det(ψ ◦ exp (F)) = det( exp ◦ φ (F)) = e tr (φ (F)) = e tr (F)
for F ∈ gl (V ) .
8.55 Examples
(a) Consider the subgroup
Sl(n, R) = {A ∈ Gl (n, R)| det(A) = 1 }
of matrices in Gl (n, R) with determinant 1 , called the special linear group, and
the subspace
sl(n, R) = {A ∈ gl (n, R)|tr (A) = 0 }
of matrices in gl (n, R) with trace 0 . If U is an open neighbourhood of 0 ∈

gl (n, R) which is mapped diffeomorphically by exp onto an open neighbour-
hood V of I ∈ Gl (n, R) , we have that
exp (U ∩ sl (n, R) ) = V ∩ Sl(n, R)
by Remark 8.53. Using Proposition 8.42 we therefore conclude that Sl(n, R) is

a submanifold and a closed Lie subgroup of Gl (n, R) , having the Lie subalge-
bra sl(n, R) of gl (n, R) as its Lie algebra.
(b) Consider the subgroup
Sl(n, C) = {A ∈ Gl (n, C)| det(A) = 1 }
of matrices in Gl (n, C) with determinant 1 , called the complex special linear

group, and the subspace
sl(n, C) = {A ∈ gl (n, C)|tr (A) = 0 }
of matrices in gl (n, C) with trace 0 . Let W be an open neighbourhood of

0 ∈ gl (n, C) which is mapped diffeomorphically by exp onto an open neigh-
bourhood of I ∈ Gl (n, C) , and let
U = {A ∈ W ||tr (A) | < 2π }
and V = exp (U) . We contend that
exp (U ∩ sl (n, C) ) = V ∩ Sl(n, C)
which by Proposition 8.42 implies that Sl(n, C) is a submanifold and a closed

Lie subgroup of Gl (n, C) , having the Lie subalgebra sl(n, C) of gl (n, C) as its
Lie algebra. To prove the assertion, suppose that A ∈ U ∩ sl (n, C) . Then we
have that
det( exp (A)) = e tr (A) = 1
which implies that exp (A) ∈ V ∩ Sl(n, C) . Conversely, assuming that A ∈ U
with exp (A) ∈ V ∩ Sl(n, C) , we have that |tr (A) | < 2π and
e tr (A) = det( exp (A)) = 1 .
This shows that tr (A) = 0 so that A ∈ U ∩ sl (n, C) , and completes the proof
of the assertion.
8.56 Proposition Let ψ : gl (n, K) → gl (n, K) , where K is either R or C, be a

bounded linear map with ψ −1 = ψ and ψ (AB) = ψ (A) ψ (B) whenever AB = BA.
Then
G = {A ∈ Gl (n, K)|A−1 = ψ (A) }
LIE GROUPS 277
is a submanifold and a closed Lie subgroup of Gl (n, K) , having the Lie subalgebra
g = {A ∈ gl (n, K)|ψ (A) = −A }
of gl (n, K) as its Lie algebra. There is an open neighbourhood U of 0 ∈ gl (n, K)

which is mapped diffeomorphically by exp onto an open neighbourhood V of I ∈
Gl (n, K) so that
exp (U ∩ g ) = V ∩ G . (1)
Suppose in addition that |tr (ψ (A)) | = |tr (A) | for all A ∈ gl (n, K) in the case where
K = C. Then we have that G ∩ Sl(n, K) is a submanifold and a closed Lie subgroup
of Gl (n, K) with Lie algebra g ∩ sl(n, K) , and the above neighbourhoods U and V
can be chosen so that
exp (U ∩ g ∩ sl(n, K)) = V ∩ G ∩ Sl(n, K) . (2)
PROOF : We first observe that
ψ ◦ exp = exp ◦ ψ
which follows from the relation

!
n n
Ak ψ (A) k
ψ ∑ = ∑
k=0 k! k=0 k!
when n → ∞ . Let W be an open neighbourhood of 0 ∈ gl (n, K) which is mapped

diffeomorphically by exp onto an open neighbourhood of I ∈ Gl (n, K) , and let
U = W ∩ ψ (W ) ∩ (−W ) ∩ (−ψ (W )) (3)
and V = exp (U) . Suppose that A ∈ U ∩ g . Then we have that
ψ ◦ exp (A) = exp ◦ ψ (A) = exp (−A) = exp (A) −1
which implies that exp (A) ∈ V ∩ G . Conversely, assuming that A ∈ U with exp (A) ∈
V ∩ G , we have that ψ (A), −A ∈ U and
exp ◦ ψ (A) = ψ ◦ exp (A) = exp (A) −1 = exp (−A) .
This shows that ψ (A) = −A so that A ∈ U ∩ g , and completes the proof of formula
(1). Formula (2) follows by the same arguments combined with Example 8.55, just
replacing (3) by
U = W ∩ ψ (W ) ∩ (−W ) ∩ (−ψ (W )) ∩ {A ∈ gl (n, C)||tr (A) | < 2π } (3’)
in the case where K = C. The proposition now follows from Proposition 8.42.
8.57 Using Proposition 8.56 we obtain the following closed Lie subgroups of
Gl (n, R) or Gl (n, C) in the relative topology
(1) The unitary group U(n) = {A ∈ Gl (n, C)|A−1 = A∗ }

(2) The orthogonal group O(n) = {A ∈ Gl (n, R)|A−1 = At }
(3) The complex orthogonal group O(n, C) = {A ∈ Gl (n, C)|A−1 = At }
(4) The special unitary group SU(n) = U(n) ∩ Sl(n, C)
(5) The special orthogonal group SO(n) = O(n) ∩ Sl(n, R)
(6) The special complex orthogonal group SO(n, C) = O(n, C) ∩ Sl(n, C)
with Lie algebras consisting of
(1’) skew hermitian matrices u(n) = {A ∈ gl (n, C)|A∗ = −A }

(2’) skew symmetric real matrices o(n) = {A ∈ gl (n, R)|At = −A }
(3’) skew symmetric complex matrices o(n, C) = {A ∈ gl (n, C)|At = −A }
(4’) matrices in u(n) with trace 0 su(n) = u(n) ∩ sl(n, C)
(5’) matrices in o(n) so(n) = o(n)
(6’) matrices in o(n, C) so(n, C) = o(n, C)
respectively.
8.58 Examples
(a) Let S be the signature matrix for the metric g εn in Rn defined in Definition 7.15.
Then we have a closed Lie subgroup
Oε (n) = {A ∈ Gl (n, R)|A−1 = S At S }
of Gl (n, R) , called the pseudo-orthogonal group of signature ε , with the Lie

subalgebra
oε (n) = {A ∈ gl (n, R)|S At S = −A }
of gl (n, R) as its Lie algebra by Proposition 8.56. If the index of the metric g εn
is k and ε is its canonical signature, then Oε (n) and oε (n) are also denoted
by Ok (n) and ok (n) , or by O(k, n − k) and o(k, n − k) , respectively. ok (n)
consists of matrices of the form

a x
xt b
where a ∈ o(k), b ∈ o(n − k) and x is an arbitrary k × (n − k) matrix. Indeed, if

a x
A= ,
y b
LIE GROUPS 279
we have that

−Ik 0 at yt −Ik 0 at −yt
S At S = =
0 In−k xt bt 0 In−k −xt bt
which equals
−a −x
−A =
−y −b
if and only if at = −a, bt = −b and y = xt .
(b) Let V be a finite dimensional real vector space with a metric g of index k. If
B = {v1 , ..., vn } is an orthonormal basis for V with g(vi , vi ) = εi for i = 1, .., n,
the linear isomorphism ψ : V → R n given by
n
ψ (v) = a when v= ∑ aj vj ,
j=1
is an isometry from V to the pseudo-Euclidean space Rεn . Consider the subgroup
Oε (V ) = {F ∈ Gl (V )|F ∗ g = g}
of Gl(V ) consisting of the linear isometries in V . If φ : gl (V ) → gl (n, R) is

the linear isomorphism defined by
φ (F) = mB
B (F)
for F ∈ gl (V ) , we have that Oε (V ) = φ −1 (Oε (n)) . Indeed, if φ (F) = A , then

F ∗ g = g if and only if
g nε (a, b) = g nε (Aa, Ab) = at At SA b = g nε (a,(SAt S)Ab)
for every a, b ∈ Rn , which is equivalent to A−1 = S At S. Hence Oε (V ) is a closed

Lie subgroup of Gl(V ) in the relative topology, with the Lie subalgebra
oε (V ) = φ −1 (oε (n))
of gl (V ) as its Lie algebra. We see that F ∈ oε (V ) if and only if
g(F(v), w) = − g(v, F(w)) (1)
for every v, w ∈ V , which follows from the fact that A = φ (F) ∈ oε (n) if and
only if
g εn (Aa, b) = at At S b = g nε (a,(SAt S) b) = − g nε (a, Ab)
for every a, b ∈ Rn . If ε is the canonical signature of g, then Oε (V ) and oε (V )
are also denoted by Ok (V ) and ok (V ) , respectively. If the index is 0, they are
denoted simply by O(V ) and o(V ) .
8.59 Proposition For each matrix A ∈ SO(3) there is a unit column vector w ∈ R3
with Aw = w . If F : R3 → R3 is the linear map having A as its standard matrix, there
is a positively oriented orthonormal basis B = {u, v, w} in R3 so that
 
cos θ − sin θ 0
mBB (F) =
 sin θ cos θ 0 
0 0 1
where 0 ≤ θ < 2π . The linear map F, also denoted by R w (θ ), is a rotation by an

angle θ about the line through the origin with direction vector w. We have that
R w (θ ) = R −w (2π − θ )
when 0 < θ < 2π , and R w (0) = id for every unit vector w.

PROOF : Since
det(I − A) = det(I − At ) = det(I − A−1)

= det(A−1 ) det(A − I) = − det(I − A)
so that det(I − A) = 0, it follows that 1 is an eigenvalue of A. Let w be an eigenvector

of A with eigenvalue 1 and kwk= 1, and choose vectors u and v so that B = {u, v, w}
is a positively oriented orthonormal basis in R3 . Then it follows from Example
8.58 (b) that  
a1 a2 0
mB
B (F) =
 b1 b2 0 
0 0 1
where
a21 + b21 = 1 , a22 + b22 = 1 and a1 b2 − b1a2 = 1
so that
(a1 − b2)2 + (b1 + a2)2 = (a21 + b21) + (a22 + b22 ) − 2(a1b2 − b1a2 ) = 0 ,
which implies that
a1 = b2 = cos θ and b1 = −a2 = sin θ where 0 ≤ θ < 2π .
8.60 Example The 3 × 3 matrices

     
0 0 0 0 0 1 0 −1 0
L1 =  0 0 −1  , L2 =  0 0 0  and L3 =  1 0 0 
0 1 0 −1 0 0 0 0 0
LIE GROUPS 281
form a basis for the Lie algebra so(3) of skew symmetric matrices, satisfying the
commutation relations
[ L1 , L2 ] = L3 , [ L2 , L3 ] = L1 and [ L3 , L1 ] = L2 .
The corresponding one-parameter subgroups φ Li : R → SO(3) are given by

φ Li (θ ) = exp ( θ Li ) for θ ∈ R and i = 1, 2, 3 , where
   
1 0 0 cos θ 0 sin θ
exp ( θ L1 ) =  0 cos θ − sin θ  , exp ( θ L2 ) =  0 1 0 
0 sin θ cos θ − sin θ 0 cos θ
and  
cos θ − sin θ 0
exp ( θ L3 ) =  sin θ cos θ 0 
0 0 1
are the standard matrices for rotations by the angle θ about the x−, y− and z−axis,
respectively.
8.61 Let g be the Lorentz metric in R4 given by
g(a, b) = at S b
for column vectors a and b in R4 , where

 
−1 0 0 0
 0 1 0 0 
S=  
0 0 1 0 
0 0 0 1
is the signature matrix. The vector space R4 equipped with this metric is called the
Minkowski space. A vector a ∈ R4 is said to be spacelike, lightlike or timelike if
g(a, a) > 0, g(a, a) = 0 or g(a, a) < 0, respectively. We denote the components of a
with respect to the standard basis E = {e0 , e1 , e2 , e3 } for R4 by a0 , a1 , a2 and a3 . The
pseudo-orthogonal group
O 1 (4) = {A ∈ Gl (4, R)|At S A = S }
described in Example 8.58 (a) is called the Lorentz group and is also denoted by L .
If A ∈ L , then det(A) 2 = 1 , and
3
−A200 + ∑ A2r0 = −1
r=1
so that
3
A200 = 1 + ∑ A2r0 ≥ 1 .
r=1
A matrix A ∈ L is called forward-timelike if A00 ≥ 1 and backward-timelike if

A00 ≤ −1. Hence L is the union of four disjoint subsets which are both open and
closed :
L ↑+ = {A ∈ L |A00 ≥ 1, det(A) = 1} , L ↑− = {A ∈ L |A00 ≥ 1, det(A) = −1},

L ↓+ = {A ∈ L |A00 ≤ −1, det(A) = 1}, L ↓− = {A ∈ L |A00 ≤ −1, det(A) = −1}.
For every A ∈ L we have that At ∈ L so that

3 3
∑ A2r0 = ∑ A20r = A200 − 1 (1)
r=1 r=1
If A, B ∈ L are forward-timelike, then so is AB since

3
3 1/2 3 1/2

∑ A0r Br0 ≤ ∑ A20r ∑ B2r0 = (A200 − 1) 1/2 (B200 − 1) 1/2

r=1 r=1 r=1
= {(A00 − 1)(B00 + 1)}1/2 {(A00 + 1)(B00 − 1)}1/2
= {(A00 B00 − 1) 2 − (A00 − B00 ) 2 }1/2 ≤ A00 B00 − 1
which implies that

3
(AB) 00 = ∑ A0r Br0 ≥ 1 .
r=0
Furthermore, if A ∈ L is forward-timelike, then so is A−1 since
(A−1 ) 00 = (S At S) 00 = A00 .
This shows that L ↑+ is a closed subgroup and therefore a Lie subgroup of L , called
the proper Lorentz group. The matrices S, P = −S and −I are the standard matrices
for time inversion, space inversion (or parity transformation ) and total inversion,
respectively, and we have that
↑− ↑+ ↓+ ↑+
L = PL , L = (−I) L and L ↓− = S L ↑+
.
8.62 e : gl (3, R) → gl (4, R)

We have an injective Lie algebra homomorphism α
given by
1 0
e (A) =
α
0 A
for A ∈ gl (3, R) , which induces a Lie group homomorphism α : SO(3) →
L ↑+ and a corresponding Lie algebra homomorphism α∗ : so(3) → o 1 (4) . By
LIE GROUPS 283
Example 8.58 (a) the Lie algebra o 1 (4) consists of matrices of the form
 
0 b1 b2 b3
 b1 0 −a3 a2 
 
 b2 a3 0 −a1 
b 3 −a2 a1 0
for arbitrary real parameters ai and bi . The matrices Ri = α∗ (Li ) for i = 1, 2, 3 are
obtained by choosing ai = 1 and all other parameters equal to 0. Let Bi for i = 1, 2, 3
be the matrices obtained by choosing bi = 1 and all other parameters equal to 0.
These six matrices form a basis for the Lie algebra o 1 (4) , satisfying the commutation
relations
[ Rσ (1) , Rσ (2) ] = ε (σ ) Rσ (3) , [ Bσ (1) , Bσ (2) ] = −ε (σ ) Rσ (3)
and
[ Bσ (1) , Rσ (2) ] = ε (σ ) Bσ (3)
for every permutation σ ∈ S3 .
The corresponding one-parameter subgroups φ Ri : R → L ↑+ and φ Bi : R →
L ↑+ are given by φ Ri = α ◦ φ Li for i = 1, 2, 3 , and by φ Bi (χ ) = exp ( χ Bi ) for
χ ∈ R and i = 1, 2, 3 , where
   
cosh χ sinh χ 0 0 cosh χ 0 sinh χ 0
 sinh χ cosh χ 0 0   0 
exp( χ B1 ) =   , exp( χ B2 ) =  0 1 0 
 0 0 1 0   sinh χ 0 cosh χ 0 
0 0 0 1 0 0 0 1
and  
cosh χ 0 0 sinh χ
 0 1 0 0 
exp ( χ B3 ) = 
 0

0 1 0 
sinh χ 0 0 cosh χ
are the standard matrices for velocity transformations or boosts with velocity param-
eter χ along the x−, y− and z−axis, respectively.
8.63 Remark By formula (1) in 8.61 we see that a matrix A ∈ L ↑+ belongs

to α (SO(3)) if and only if A00 = 1, which is also equivalent to the assertion that
A e0 = e0 .
Hence if A, B ∈ L ↑+ , then A e0 = B e0 if and only if A = BR for a matrix R ∈
α (SO(3)) .
8.64 Example Let J be the skew symmetric 2n × 2n matrix

0 I
J=
−I 0
where I is the n × n identity matrix. Then we have a closed Lie subgroup

Sp (n, C) = {A ∈ Gl (2n, C)|A−1 = J −1At J }
of Gl (2n, C) , called the complex symplectic group , with the Lie subalgebra
sp(n, C) = {A ∈ gl (2n, C)|J −1At J = −A }
of gl (2n, C) as its Lie algebra by Proposition 8.56. sp(n, C) consists of matrices of
the form
a x
y −at
where x and y are symmetric n × n matrices, and a is an arbitrary n × n matrix. Indeed,
if
a x
A= ,
y b
we have that

0 −I at yt 0 I bt −xt
J −1At J = =
I 0 xt bt −I 0 −yt at
which equals
−a −x
−A =
−y −b
if and only if xt = x, yt = y and b = −at .
Consider the non-degenerate skew symmetric bilinear form on C 2n given by
n
<a, b > = at J b = ∑ (ai bn+i − an+i bi )
i=1
2n
for column vectors a and b in C . A matrix A ∈ gl (2n, C) satisfies the relation
<A a, A b > = <a, b >
for all a, b ∈ C 2n if and only if At JA = J, which is equivalent to A ∈ Sp (n, C).
In the case n = 1 we have that
Sp (1, C) = Sl(2, C)
since At JA = det(A) J when A ∈ gl (2, C).
THE ALGEBRA OF QUATERNIONS

8.65 The algebra H of quaternions is the real vector space H = R × R3 with a
product defined by
(a, u)(b, v) = (ab − u · v, av + bu + u × v)
LIE GROUPS 285
for (a, u), (b, v) ∈ H. This product is associative since
{(a, u)(b, v) }(c, w) = (ab − u · v, av + bu + u × v) (c, w)
= ((ab − u · v) c − (av + bu + u × v) · w,
(ab − u · v) w + c (av + bu + u × v) + (av + bu + u × v) × w)
= (abc − a v · w − b u · w − c u · v − u × v · w,
ab w + ac v + bc u + a v × w + b u × w + c u × v
− (v · w) u + (u · w) v − (u · v) w)
= (a (bc − v · w) − u · (bw + cv + v × w),

a (bw + cv + v × w) + (bc − v · w) u + u × (bw + cv + v × w))
= (a, u) (bc − v · w, bw + cv + v × w) = (a, u){(b, v)(c, w) }
for (a, u), (b, v), (c, w) ∈ H, but it is not commutative. We call the components a and u
of the quaternion (a, u) its scalar and vector part, respectively. The quaternions with
zero scalar part form a subspace H 0 = {0} × R3 of H called the subspace of pure
quaternions. The injection ρ : R → H defined by ρ (a) = (a, 0) is an algebra homo-
morphism, and we identify the subalgebra R × {0} of H with R, and the quaternion
(a, 0) with the real number a. If {i, j, k} is the standard basis for R3 , we also denote
the quaternions (0, i), (0, j) and (0, k) simply by i, j and k, so that the quaternion
q = (a, (b, c, d)) can be written as q = a + bi + c j + dk. We have that
i2 = j2 = k2 = −1
and
i j = − j i = k , jk = −k j = i and ki = −ik = j .
The conjugate of a quaternion q = (a, u) is the quaternion q̄ = (a, −u), and we have
that
q q̄ = q̄q = a2 + kuk 2 = kqk 2 ,
where the non-negative real number
q
kqk = a2 + kuk 2
is called the norm of q = (a, u). Hence each quaternion q 6= 0 has a multiplicative
inverse which is given by q−1 = q̄ / kqk 2, showing that H is a division algebra. We
also have that
p q = q̄ p̄ and kp qk = kpkkqk
when p, q ∈ H.
The injection σ : C → H defined by σ (a, b) = (a, (0, 0, b)) is an algebra homo-

morphism, and we identify the subalgebra R × ({0} × {0} × R) of H with C, and the
quaternion (a, (0, 0, b)) = a + bk with the complex number (a, b) = a + bi. Since
a + bi + c j + dk = (a + dk) + j(c + bk) ,
H can be considered to be a 2-dimensional complex vector space with a basis E =

{1, j}, and with multiplication with complex scalars from the right. Using the left
translations Lq : H → H in H defined by Lq (p) = q p for p, q ∈ H, we have an injective
algebra homomorphism φ : H → gl(2, C) given by φ (q) = mEE (Lq ) for q ∈ H, so that

z1 −z̄2
φ (z1 + jz2 ) =
z2 z̄1
for z1 , z2 ∈ C. Indeed, using that jz = z̄ j when z ∈ C, we have that (z1 + jz2 ) j =

−z̄2 + jz̄1 . In particular it follows that

1 0 0 i 0 −1 i 0
φ (1) = , φ (i) = , φ ( j) = and φ (k) = .
0 1 i 0 1 0 0 −i
The last three matrices are skew hermitian with trace 0 , and we denote them by τ1 ,
τ2 and τ3 , respectively. We have that τr = i σr for r = 1, 2, 3 , where

0 1 0 i 1 0
σ1 = , σ2 = and σ3 =
1 0 −i 0 0 −1
are the Pauli spin matrices which are hermitian with trace 0. We also introduce the
hermitian and skew hermitian matrix

1 0 i 0
σ0 = and τ0 =
0 1 0 i
so that τ0 = i σ0 . We have that B = {τ1 , τ2 , τ3 } , C = {τ0 , τ1 , τ2 , τ3 } and D =

{τ1 , τ2 , τ3 , σ1 , σ2 , σ3 } are bases for the Lie algebras su(2) , u(2) and sl(2, C), re-
spectively. Finally, we note that
det (φ (q)) = kqk
when q ∈ H.
8.66 Now consider the module Hn over H with multiplication with quaternionic
scalars from the right. We define the inner product of two quaternionic vectors a, b ∈
Hn by
n
<a, b> = ∑ āi bi .
i=1
This product is additive in a and b, and we have that
<b, a> = <a, b> , <aq, b> = q̄ <a, b> and <a, bq> = <a, b> q
LIE GROUPS 287
when q ∈ H. The non-negative real number

n 1/2
2
kak = ∑ kaik = <a, a>1/2
i=1
is called the length of a. A quaternionic linear map F : Hn → Hn is said to be sym-

plectic if
kF(a)k = kak
for every a ∈ Hn , which is equivalent to the assertion that
<F(a), F(b)> = <a, b>
for every a, b ∈ Hn . Indeed, assuming that F preserves the length of every quater-
nionic vector and using that
ka + bqk 2 = kak 2 + <a, b> q + <a, b> q + kbqk 2 ,
we see that <F(a), F(b)> q and <a, b> q have equal scalar parts for every quaternion
q. By choosing q = 1, i, j and k, this shows that F preserves the inner product.
Let E = {e1 , ... , en } be the standard basis for Hn , where ei is the vector with all
components equal to 0 except the i-th component which is 1. Writing a ∈ Hn as a
column vector, we have that F(a) = Aa, where A = mEE (F) is the quaternionic n × n
matrix given by
n
F(e j ) = ∑ ei Ai j
i=1
for j = 1, ... , n. We see that F is symplectic if and only if Āt A = I, where I is the
quaternionic n × n identity matrix.
Using only complex scalars, we have a complex vector space isomorphism
ψ : Hn → C2n defined by ψ (a) = x where ai = xi + jxn+i for i = 1, ..., n. The in-
ner product of the quaternionic vectors a and b, where ψ (a) = x and ψ (b) = y, is
given by
n n
<a, b> = ∑ āi bi = ∑ (x̄i − x̄n+i j)(yi + j yn+i )
i=1 i=1
2n n
= ∑ x̄i yi + j ∑ (xi yn+i − xn+i yi ) .
i=1 i=1
By 8.47, 8.57 and Example 8.64 the quaternionic linear map F : Hn → Hn is sym-
plectic if and only if the matrix mB −1
B (ψ ◦ F ◦ ψ ), where B is the standard basis in
2n
C , belongs to U(2n) ∩ Sp (n, C) . Indeed, we need to show that a complex linear
map F preserving the inner product in Hn actually is quaternionic linear, i.e., that
F(aq) = F(a)q for every a ∈ Hn and q ∈ H. But this follows since
<F(b), F(aq) − F(a)q> = <F(b), F(aq)> − <F(b), F(a)> q

= <b, aq> − <b, a> q = 0
for every b ∈ Hn . If mEE (F) = a + j b for complex n × n matrices a and b, then

B −1 a −b̄
mB (ψ ◦ F ◦ ψ ) = .
b ā
By Propositions 8.42 and 8.56 we know that
Sp (n) = U(2n) ∩ Sp (n, C)
is a closed Lie subgroup of Gl (2n, C) , called the symplectic group , with the Lie
subalgebra
sp(n) = u(2n) ∩ sp(n, C)
of gl (2n, C) as its Lie algebra. sp(n) consists of matrices of the form

a −b̄
b ā
where a is a skew hermitian and b a symmetric n × n matrix. Indeed, if

a c
A=
b −at
is a matrix in sp(n, C) , where b and c are symmetric as described in Example 8.64,

then we have that ∗
∗ a b̄
A =
c̄ −ā
which equals
−a −c
−A =
−b at
if and only if a∗ = −a and c = −b̄.
In the case n = 1 we have that
Sp (1) = SU(2)
since Sp (1, C) = Sl(2, C) by Example 8.64.
8.67 Example Let H be the Lie algebra of quaternions with bracket product
[ p, q ] = p q − q p as defined in Remark 4.83. It is a real vector space of dimension
4, and the identity map z : H → R4 is a real linear isomorphism. We give H the
manifold structure defined in Example 2.9 (b) so that z is a coordinate map.
Now let H∗ = H − {0} be the group of non-zero quaternions which is an open
submanifold of H . If α : H∗ → H is the inclusion map, then (z ◦ α , H∗ ) is a local
chart on H∗ . By 8.65 the group operations in H∗ are smooth so that H∗ is a Lie group.
We will show that the Lie algebra g of H∗ may be identified with H using
Lemma 2.84. We have that g = Te H∗ = Te H which we canonically identify with
LIE GROUPS 289
H by means of the linear isomorphism ω : g → H given by ω = z−1 ◦ tz,e . It only

remains to show that
ω ( [ X,Y ] ) = [ ω (X), ω (Y ) ] (1)
for X,Y ∈ g , so that ω is a Lie algebra isomorphism.
As usual we let Xe and Ye be the left invariant vector fields on H∗ determined by
X and Y . If h : H → H is the identity map, it follows from 5.49 that
ω (X) = ω ◦ h ∗ (X) = X (h)
for every X ∈ g , so that
ω ( [ X,Y ] ) = [ X,Y ] (h) = X ( Ye (h)) − Y ( Xe (h)) .
To compute the last expression, we will use the real linear maps L p : H → H and
R p : H → H defined by L p (q) = p q and R p (q) = q p for p, q ∈ H. Indentifying the
tangent space Tp H with H by means of the linear isomorphism ω p : Tp H → H given
by ω p = z−1 ◦ tz,p , we have that
Ye (h)(p) = Yep (h) = L p ∗ (Y )(h) = ω p ◦ h ∗ ◦ L p ∗ (Y )

= ω p ◦ L p ∗ (Y ) = L p ◦ ω (Y ) = R ω (Y ) (p)
for p ∈ H∗ . This implies that
X ( Ye (h)) = ω ω (Y ) ◦ R ω (Y ) ∗ (X) = R ω (Y ) ◦ ω (X) = ω (X) ω (Y )
which completes the proof of (1).

The exponential map exp : H → H∗ is given by
∞
qk
exp (q) = ∑ (2)
k=0 k!
for q ∈ H , which is proved in the same way as in Proposition 8.52.

Now consider the injective Lie algebra homomorphism φ : H → gl(2, C) defined
in 8.65, and the induced map ψ : H∗ → Gl(2, C) which is a Lie group homomor-
phism. Identifying the Lie algebra of Gl(2, C) with gl(2, C) as usual by means of
the linear isomorphism ω e =e z−1 ◦ tez,e , where e
z : gl (2, C) → R8 is the linear isomor-
phism defined in Example 8.10 (c), the Lie algebra homomorphism ψ∗ is identified
with φ in the same way as in Example 8.19. Indeed, e z ◦ φ ◦ z−1 = λ : R4 → R8 is
the linear map given by λ (a, b, c, d) = (a, c, −c, a, d, b, b, −d) for (a, b, c, d) ∈ R 4 , so
−1 = D(e
that tez,e ◦ ψ∗ ◦ tz.e z ◦ ψ ◦ z−1 )(z(e)) = λ by the commutative diagram in 2.70.
This implies that ω e ◦ ψ∗ ◦ ω −1 = φ .
Using Proposition 8.40 we therefore have a commutative diagram
φ✲
H gl (2, C)
exp exp
❄ ❄
ψ✲
H∗ Gl (2, C)
where φ ( H 0 ) = sp(1) and ψ ( S3 ) = Sp (1) by 8.66. Hence it follows from Proposi-

tions 8.42 and 8.56 that the group S3 of unit quaternions is a closed Lie subgroup of
H∗ , with the Lie subalgebra H 0 of H consisting of the pure quaternions as its Lie al-
gebra. Furthermore, φ induces a Lie group isomorphism from S3 to Sp (1) = SU(2) ,
which we also denote by φ , and a corresponding Lie algebra isomorphism φ ∗ from
H 0 to sp(1) = su(2) .
LEFT INVARIANT FORMS

8.68 Definition A k-form ω on G is called left invariant if Lg∗ (ω ) = ω for every
g ∈ G, which means that ω (h) = Lg∗ (ω (gh)) for every g, h ∈ G. Left invariant 1-forms
on G are also known as Maurer-Cartan forms.
8.69 Proposition If G is a Lie group and ω ∈ Λk ( g ), then there is a unique left

e on G with ω
invariant k-form ω e (e) = ω which is given by
e (g) = Lg∗−1 (ω )
ω
for g ∈ G. It is called the left invariant k-form determined by ω .

PROOF : Since Lh−1 = L(gh)−1 ◦ Lg , we have that
e (h) = Lg∗ (ω
ω e (gh))
for every g, h ∈ G.
To complete the proof, we only need to show that ω e is smooth. If {X1 , ... , Xn } is
a basis for g, then the left invariant vector fields X f1 , ... , X
fn form a local basis for T G
on G. If {ω 1 , ... , ω n } is the dual local basis for Λ1 (T G) on G, then
e=
ω ∑ ωi1 , ... ,ik ω i1 ∧ · · · ∧ ω ik
i1 <...<ik
where
e (g) ((f
ωi1 , ... ,ik (g) = ω X1 ) g , ... , (Xek ) g ) = ω (X1 , ... , Xk )
LIE GROUPS 291
e is smooth.
for g ∈ G, thus showing that ω
8.70 Proposition A k-form ω on G is left invariant if and only if ω (X1 , ... , Xk )

is a constant function on G for any choice of left invariant vector fields X1 , ... , Xk on
G.
PROOF : If the vector fields X1 , ... , Xk are left invariant, then
Lg∗ (ω )(h) ((X1 ) h , ... , (Xk ) h ) = ω (gh) (Lg ∗ (X1 ) h , ... , Lg ∗ (Xk ) h )
= ω (gh) ((X1 ) gh , ... , (Xk ) gh )
for every g, h ∈ G. Assuming that ω is left invariant, the last assertion of the proposi-
tion follows by choosing h = e.
The converse statement follows from the fact that any tangent vector w ∈ Th G
equals Xeh for the left invariant vector field Xe determined by X = (Lh−1 )∗ (w).
8.71 Definition A vector-valued k-form ω on G with values in a finite dimen-

sional vector space W is called left invariant if Lg∗ (ω ) = ω for every g ∈ G, which
means that ω (h) = Lg∗ (ω (gh)) for every g, h ∈ G. Vector-valued left invariant 1-
forms on G are also known as vector-valued Maurer-Cartan forms.
8.72 Proposition A vector-valued k-form ω on G with values in a finite dimen-

sional vector space W is left invariant if and only if the real-valued k-form λ · ω is
left invariant for every functional λ ∈ W ∗ .
PROOF : Follows from Proposition 5.47.
8.73 Proposition Let G is a Lie group and W be a finite dimensional vector space.
If ω ∈ Λk ( g ;W ), e on G with
then there is a unique left invariant W -valued k-form ω
e (e) = ω which is given by
ω
e (g) = Lg∗−1 (ω )
ω
for g ∈ G. It is called the left invariant W -valued k-form determined by ω .
PROOF : e = λ]
Follows from Propositions 5.42 (3) and 8.72 since λ · ω ◦ ω for every
∗
functional λ ∈ W by Definition 5.3.
8.74 Proposition A vector-valued k-form ω on G with values in a finite di-

mensional vector space W is left invariant if and only if ω (X1 , ... , Xk ) is a constant
vector-valued function on G for any choice of left invariant vector fields X1 , ... , Xk
on G.
PROOF : Follows from Propositions 8.70 and 8.72 since
λ ◦ ω (X1 , ... , Xk ) = (λ · ω )(X1 , ... , Xk )

8.75 Definition By the canonical Maurer–Cartan form or the canonical 1-

form on a Lie group G we mean the unique g-valued left invariant 1-form θ on
G determined by the 1-form id : g → g in Λ1 ( g ; g). It is given by
θ (g)(w) = (Lg−1 ) ∗ (w)
for g ∈ G and w ∈ Tg G.
8.76 Remark By Proposition 8.74 we have that
θ (g)(Xeg ) = X
for every g ∈ G and X ∈ g, where Xe is the left invariant vector field on G determined
by X.
8.77 Theorem (Maurer-Cartan’s structure equation) If θ is the canonical

Maurer-Cartan form on a Lie group G, then we have that
d θ (g)(w1 , w2 ) = −[ θ (g)(w1 ), θ (g)(w2 ) ]
for g ∈ G and w1 , w2 ∈ Tg G.
PROOF : Suppose that w1 , w2 ∈ Tg G , and let Xi = θ (g)(wi ) = (Lg−1 ) ∗ (wi ) so that
e e is a constant
Xi g = Lg ∗ (Xi ) = wi for i = 1, 2. For each X ∈ g we know that θ ( X)
function on G with value X. Hence it follows from Remark 5.86 and Lemma 2.78 (2)
that
d θ (g)(w1 , w2 ) = d θ ( Xe1 , Xe2 )(g)

= Xe1 ( θ ( Xe2 ))(g) − X
e2 ( θ ( Xe1 ))(g) − θ ( [ Xe1 , Xe2 ] )(g)
= −[ X1 , X2 ] = −[ θ (g)(w1 ), θ (g)(w2 ) ] .
8.78 Remark The formula in Theorem 8.77 can also be written in the form
d θ = − 12 θ ∧ θ
where the wedge product is with respect to the bilinear map
ν : g×g → g
given by ν (X,Y ) = [ X,Y ].

Chapter 9
GROUP ACTIONS
INTRODUCTION
9.1 Definition Let M be a smooth manifold and G be a Lie group. By an operation
or an action of G on M on the left we mean an operation µ : G × M → M of G on
M as Defined in Definition 13.12 in the appendix, where µ is smooth. An operation
of G on M on the right is defined similarly as in Remark 13.13 in the appendix.
9.2 Proposition Let G be a Lie group and V a finite dimensional vector space,
and let µ : G × V → V and φ : G → Aut (V ) be maps so that
µ (g, v) = φ (g)(v)
for g ∈ G and v ∈ V . Then µ is an operation of G on V on the left which is linear in
its second argument if and only if φ is a representation.
µ is called the operation determined by φ .
PROOF : Assume that φ is smooth. Then we see that µ is smooth since
µ = B ◦ (φ × idV ) ,
where B : End (V ) × V → V is the bilinear map defined by B(F, v) = F(v) for F ∈
End (V ) and v ∈ V .
Conversely, assume that µ is smooth, and let B = {v1 , ..., vn } be a basis for V
with dual basis B ∗ = {v∗1 , ..., v∗n } . Then φ is smooth since
v∗i ◦ φ (g)(v j ) = v∗i ◦ µ (g, v j )
for g ∈ G and i, j = 1, ... , n.
Since
φ (e)(v) = µ (e, v)
for v ∈ V where e ∈ G is the unit element, and
φ (g) ◦ φ (h)(v) = µ (g, µ (h, v)) and φ (gh)(v) = µ (gh, v)
for g, h ∈ G and v ∈ V , we see that µ is an operation if and only if φ is a representation
of G on V .
293
9.3 Definition By an operation or an action of a Lie group G on a vector bundle

π : E → M on the left we mean a pair (µ e , µ ) of operations µ
e : G × E → E and
µ : G × M → M of G on E and M, respectively, such that the corresponding pair
(L̃g , Lg ) of left translations forms a bundle map for each g ∈ G. This implies that the
projection π : E → M is equivariant with respect to the actions µ e and µ of G on E
and M .
9.4 Proposition Let µ : G × M → M be an operation of a Lie group G on a

smooth manifold M n on the left. Then we have an operation (µ e , µ ) of G on the
tangent bundle π : T M → M such that the left translations Lg and L̃g corresponding
e , respectively, are related by
to µ and µ
L̃g = (Lg) ∗
for g ∈ G.
PROOF : By Proposition 2.71 we know that (L̃g , Lg ) is a bundle map for each g ∈ G,
and it only remains to prove that µe is smooth at every point (h, v) ∈ G × Tq M where
q ∈ M. Let (z,U) be a local chart around µ (h, q) on M, and choose local charts (x,V )
and (y,W ) around h and q on G and M, respectively, so that µ (V × W ) ⊂ U. Then
the map f : x(V ) × y(W ) → Rn defined by f = z ◦ µ ◦ (x × y)−1 is smooth, and we
have that
f (a, b) = z ◦ L x−1 (a) ◦ y−1 (b)
for a ∈ x(V ) and b ∈ y(W ).
If (ty , π −1 (W )) and (tz , π −1 (U)) are the local trivializations in the tangent bundle
associated with the local charts (y,W ) and (z,U) , we have that
e ◦ (id V × ty )−1 (g, p, v) = tz ◦ Lg ∗ ([ y, v ] p )

tz ◦ µ
= tz ( [ z, D(z ◦ Lg ◦ y−1 )(y(p)) v ] Lg (p) ) = (Lg (p), D(z ◦ Lg ◦ y−1 )(y(p)) v )
for g ∈ V and (p, v) ∈ W × Rn . This shows that
e ◦ (id V × ty )−1 ◦ (x × y × id Rn )−1 (a, b, v)

(z × id Rn ) ◦ tz ◦ µ
= (z ◦ Lx−1 (a) ◦ y−1 (b), D(z ◦ Lx−1 (a) ◦ y−1 )(b) v) = ( f (a, b), D2 f (a, b) v)
for (a, b, v) ∈ x(V ) × y(W ) × Rn , which completes the proof of the proposition.
9.5 Corollary Let G be a Lie group acting on the smooth manifold M on the left,
and suppose that p ∈ M is a fixed point, i.e., that g p = p for every g ∈ G. For each
g ∈ G, we let Lg : M → M be the diffeomorphism defined by Lg (q) = gq for q ∈ M.
Then we have a representation φ : G → Aut (Tp M) of G defined by
φ (g) = (Lg) ∗ p
for g ∈ G.
GROUP ACTIONS 295
PROOF : Using that the tangent space Tp M is a submanifold of T M , we obtain an

action of G on Tp M on the left which is linear in its second argument by Proposition
9.4. The result therefore follows from Proposition 9.2.
9.6 Proposition Let (µ e , µ ) be an operation of a Lie group G on a vector bundle

π : E → M of dimension n . Then there is an operation (µ̂ , µ ) of G on the dual bundle
π ∗ : E ∗ → M such that the left translations L̃g and L̂g corresponding to µ e and µ̂ ,
respectively, are related by
(L̂g ) p = (L̃g−1 ) p∗
for each p ∈ M and g ∈ G .
PROOF : Using Proposition 4.50, we only need to prove that µ̂ is smooth at every
point (h, v) ∈ G × Eq∗ where q ∈ M. Choose local trivializations (t 1 , π1−1 (U1 )) and
(t 2 , π2−1 (U2 )) of π around q and hq, respectively, and a coordinate neighbourhood V
around h on G, with µ (V ×U1 ) ⊂ U2 . Let (t1′ , π1 ∗ −1 (U1 )) and (t2′ , π2 ∗ −1 (U2 )) be the
corresponding local trivializations in the dual bundle, and let E be the standard basis
for Rn with dual basis E ∗ .
Then t 2 ◦ µe ◦ (idV × t 1−1 ) : V × U1 × Rn → U2 × Rn is a smooth map given by
e ◦ (idV × t1−1 )) (g, p, v) = (gp, (t2,gp ◦ L̃ g,p ◦ t1,p

(t2 ◦ µ −1
) (v)) ,
−1
where t2,gp ◦ L̃ g,p ◦ t1,p : Rn → Rn is a linear isomorphism for each g ∈ V and p ∈ U1 .
Since
′ −1
t2,gp ◦ L̂ g,p ◦ t ′ 1,p = r ◦ { (t2,gp ◦ L̃ g,p ◦ t1,p
−1 −1 ∗
) } ◦ r−1 ,
∗
′ −1
mEE (t2,gp ◦ L̂ g,p ◦ t ′ 1,p ) = { mEE (t2,gp ◦ L̃ g,p ◦ t1,p
−1 −1 t
) } .
Hence t2′ ◦ µ̂ ◦ (idV × t ′ 1−1 ) is a smooth map from V × U1 × Rn to U2 × Rn , thus

showing that (µ̂ , µ ) is an action of G on π ∗ : E ∗ → M .
THE ADJOINT REPRESENTATION

9.7 Let G be a Lie group. For each g ∈ G, we have an automorphism ig : G → G
defined by
ig (h) = ghg−1
for h ∈ G, called the inner automorphism of G by the element g. Defining Ad g : g → g
by Ad g = ig ∗ , we have a commutative diagram
Ad g ✲
g g
exp exp
❄ ❄
ig ✲
G G
by Proposition 8.40, so that
exp t Ad g (X) = g (exp t X ) g−1
for X ∈ g and t ∈ R .
Now the map a : G × G → G defined by a (g, h) = ghg−1 = ig (h) for g, h ∈ G, is
an operation of G on itself on the left with fixed point e. By Corollary 9.5 we thus
have a representation
Ad : G → Gl (g)
of G defined by Ad (g) = Ad g for g ∈ G, which is called the adjoint representation .
If we define
ad : g → gl (g)
by ad = Ad ∗ and again use Proposition 8.40, we have the commutative diagram
ad
g ✲ gl (g)
exp exp
❄ ❄
Ad ✲
G Gl (g)
The contragradient Ad ∗ : G → Gl (g ∗ ) of the adjoint representation is called the

coadjoint representation .
9.8 Examples
(a) For each non-singular matrix A ∈ Gl (n, R) the inner automorphism iA :
Gl (n, R) → Gl (n, R) is induced by the linear isomorphism eA : gl (n, R) →
gl (n, R) given by
eA (B) = ABA −1
for B ∈ gl (n, R). Hence it follows from Lemma 2.84 that Ad (A) = eA ∗ = eA
when we canonically identify Te gl (n, R) with gl (n, R) as in Example 8.10 (a).
The adjoint representation
Ad : Gl (n, R) → Gl (gl (n, R))

GROUP ACTIONS 297
is therefore given by
Ad (A)(B) = ABA −1
for A ∈ Gl (n, R) and B ∈ gl (n, R).
(b) In the same way we see that the adjoint representation
Ad : Gl (n, C) → Gl (gl (n, C))
is given by
Ad (A)(B) = ABA −1
for A ∈ Gl (n, C) and B ∈ gl (n, C).
(c) We also see that the adjoint representation
Ad : Gl (V ) → Gl (gl (V )) ,
where V is a finite dimensional real or complex vector space, is given by
Ad (F)(G) = F ◦ G ◦ F −1
for F ∈ Gl (V ) and G ∈ gl (V ).
(d) The adjoint representation
Ad : H∗ → Gl ( H)
of the Lie group H∗ of non-zero quaternions described in Example 8.67 is given

by
Ad (p)(q) = p q p −1
for p ∈ H∗ and q ∈ H.
9.9 Proposition Let G be a Lie group. Then we have that
ad (X)(Y ) = [ X,Y ]
for X,Y ∈ g.
PROOF : Let Xe and Ye be the left invariant vector fields on G determined by X and Y .
According to Proposition 4.86 we have that
e Ye ] (e) = L e Ye (e) ,
[ X,Y ] = [ X, X
and the flow γ : R × G → G of the vector field Xe is given by
γ (t, g) = g exp (t X)
for t ∈ R and g ∈ G by Propositions 8.29 and 8.31.

Now let c : R → g be the curve defined by
c(t) = [ γ t∗ Ye ] e = [ R ∗exp (t X) Ye ] e = (R exp (−t X) ) ∗ Yeexp (t X)
= (R exp (−t X) ) ∗ (L exp (t X) ) ∗ Yee = i exp (t X) ∗ Y = Ad (exp (t X)) Y
= Λ ◦ Ad ◦ φX (t)
for t ∈ R, where φX : R → G is the one-parameter subgroup defined in Proposition

8.29, and Λ : End (g) → g is the linear map given by Λ(F) = F( Y ) for F ∈ End (g) .
Using 2.70 and Proposition 4.76 it follows that
LXe Ye (e) = c′ (0) = Λ ◦ (Ad ◦ φX ) ′ (0) = ad (X)(Y )
if we canonically identify Tc(0) g with g as in Lemma 2.84.
9.10 Definition The Lie subalgebra
Z(g) = {X ∈ g|[ X,Y ] = 0 for all Y ∈ g}
of a Lie algebra g, which is the kernel of ad : g → gl (g) , is called the center of g.

We say that g is abelian if Z(g) = g, so that [ X,Y ] = 0 for every X,Y ∈ g .
9.11 Definition The subgroup
Z(G) = {g ∈ G|gh = hg for all h ∈ G}
of a Lie group G is called the center of G.
9.12 Proposition The center Z(G) of a connected Lie group G is a closed sub-
manifold and a Lie subgroup of G with Lie algebra Z(g), and it is the kernel of the
adjoint representation Ad : G → Gl (g) .
PROOF : The first assertion follows from the last by Proposition 8.45. If g ∈ Z(G) ,
we have for each X ∈ g that
exp t Ad g (X) = g (exp t X ) g−1 = exp t X
when t ∈ R , so that Ad g (X) = X which shows that g ∈ ker Ad .

Conversely, suppose that g ∈ ker Ad , and let U be an open neighbourhood of
0 ∈ g which is mapped diffeomorphically by exp onto an open neighbourhood V of
e ∈ G . Then
g (exp X ) g−1 = exp Ad g (X) = exp X
for every X ∈ U , which shows that gh = hg for every h ∈ V . Using that G is con-
nected, it follows from Proposition 13.112 in the appendix that g ∈ Z(G) .
GROUP ACTIONS 299
9.13 Corollary A connected Lie group G is abelian if and only if its Lie algebra
g is abelian.
9.14 Proposition Let φ : G1 → G2 be a Lie group homomorphism. Then we have

that
Ad(φ (g)) ◦ φ∗ = φ∗ ◦ Ad(g)
for every g ∈ G1 .
PROOF : Follows from the relation
iφ (g) ◦ φ = φ ◦ ig
for the inner automorphisms ig and iφ (g) in G1 and G2 .
9.15 Proposition If θ is the canonical Maurer-Cartan form on a Lie group G,

then
R g∗ θ = Ad (g−1 ) · θ
for every g ∈ G.
R g∗ θ (h) (w) = θ (hg) ((R g ) ∗ w) = (Lg−1 ) ∗ ◦ (Lh−1 ) ∗ ◦ (R g ) ∗ (w)
= (Lg−1 ) ∗ ◦ (R g ) ∗ ◦ (Lh−1 ) ∗ (w) = Ad (g−1 ) θ (h)(w)
for g, h ∈ G and w ∈ Th G.
THE GROUPS SO(3) AND S3

9.16 Consider the adjoint representation Ad : S3 → Gl ( H 0 ) which by Example
9.8 (d) and Proposition 9.14 is given by
Ad (q)(p) = q p q −1 = q p q̄
for q ∈ S3 and p ∈ H 0 . Using the basis C = {i, j, k} for H 0 , we also have a repre-
sentation ρe : S3 → Gl (3, R) defined by
ρe (q) = mCC (Ad (q))
for q ∈ S3 . Since kAd (q)(p)k=kpk for p ∈ H 0 when q ∈ S3 , and ρe is a continuous

map from the connected set S3 , it follows that ρe (S3 ) ⊂ SO(3). We want to show
that ρe (S3 ) = SO(3) , and that the homomorphism ρ : S3 → SO(3) induced by ρe is a

double covering of SO(3).
Every unit quaternion q ∈ S3 can be written in the form
q = cos θ2 + sin θ2 t
where t = (0, w) ∈ S3 ∩ H 0 is a pure unit quaternion and 0 ≤ θ < 4π . We contend

that ρe (q) is the standard matrix for a rotation R w (θ ) by the angle θ about the axis
through the origin with direction vector w, as defined in Proposition 8.59. To show
this, we choose pure unit quaternions r = (0, u) and s = (0, v) so that B = {u, v, w}
is a positively oriented orthonormal basis in R3 . Then we have that
r2 = s2 = t 2 = −1
and
rs = −s r = t , s t = −ts = r and tr = −rt = s ,
so that
Ad (q)(r) = cos θ r + sin θ s ,

Ad (q)(s) = − sin θ r + cos θ s ,
Ad (q)(t) = t .
This implies that

 
cos θ − sin θ 0
−1
mB
B(α ◦ Ad (q) ◦ α ) =  sin θ cos θ 0 
0 0 1
where α : R3 → H 0 is the linear isomorphism given by α (ξ ) = (0, ξ ) for ξ ∈ R3 ,

showing that α −1 ◦ Ad (q) ◦ α = R w (θ ). We have that ker ρ = {1, −1} since each
q ∈ ker ρ commutes with every pure quaternion p ∈ H 0 and therefore is real of norm
1. As this is a discrete subgroup of S3 , the result now follows from Proposition 13.111
in the appendix.
Since
cos θ2 + sin θ2 t = exp θ2 t ,
we see that the standard matrix for the rotation R w (θ ) is given by

ρ cos θ2 + sin θ2 t = exp θ ρ ∗ 2t .
By Example 8.60 it follows that the Lie algebra isomorphism ρ ∗ : H 0 → so(3) maps
the basis { i/2, j/2, k/2} for H 0 onto the basis {L1 , L2 , L3 } for so(3) .
9.17 Remark The representation of rotations in R3 by unit quaternions described

in 9.16 is very useful when combining rotations. For instance, to find the composi-
tion R i ( π2 ) ◦ R j ( π2 ) we simply compute
GROUP ACTIONS 301

cos π4 + sin π4 i cos π4 + sin π4 j = √12 (1 + i) · √12 (1 + j) = 21 (1 + i + j + k)
√
= 21 + 2
3
· √13 (i + j + k) = cos π3 + sin π3 · √13 (i + j + k) ,
giving the rotation R v ( 23π ) where v = √1

3
(i + j + k) .
9.18 Remark The bracket product in H 0 is given by

[ (0, u),(0, v) ] = (0,2 u × v)
for u, v ∈ R3 . Hence we have a Lie algebra isomorphism α2 = 12 α : R3 → H 0
from the Lie algebra R3 with bracket product [ u,v ] = u × v , which is given by
α2 (u) = (0, u/2) for u ∈ R3 . By 9.16 we also have a Lie algebra isomorphism
λ = ρ ∗ ◦ α2 : R3 → so(3) which maps the standard basis E = { i, j, k} for R3 onto
the basis {L1 , L2 , L3 } for so(3) . Since
λ (u) = mEE ( α −1 ◦ ad (α2 (u)) ◦ α ) ,
we have that
λ (u) v = α2−1 ( [ α2 (u),α2 (v) ] ) = u × v
for column vectors u, v ∈ R3 .
If A = ρ (q) where q ∈ S3 , it follows from Proposition 9.14 that
Ad (q) = ρ∗−1 ◦ Ad (A) ◦ ρ∗
so that
A = mEE ( α −1 ◦ Ad (q) ◦ α ) = mEE ( λ −1 ◦ Ad (A) ◦ λ ) .
This shows that
Ad (A)(λ (u)) = λ (Au)
for every matrix A ∈ SO(3) and column vector u ∈ R3 .
9.19 Consider the smooth map q : R3 → S3 given by

q(φ , θ , ψ ) = exp φ2 k exp θ2 i exp ψ
2 k
for (φ , θ , ψ ) ∈ R3 , which corresponds to performing in succession a rotation by an

angle ψ about the z-axis, by an angle θ about the x-axis, and finally by an angle φ
about the z-axis. We say that q(φ , θ , ψ ) is the quaternion determined by the Euler
angles φ , θ and ψ . Since p i = i p̄ for every quaternion p, we have that

φ ψ
q(φ , θ , ψ ) = exp 2 k cos θ2 + sin θ2 i exp 2 k

φ +ψ
= exp 2 k cos θ2 + i exp ψ −
2
φ
k sin θ2

φ +ψ
= exp 2 k cos θ2 + j exp ψ − φ2 + π k sin θ2
= a+ jb
for complex numbers

φ +ψ
a = exp 2 i cos θ2 and b = exp ψ − φ2 + π i sin θ2
so that
φ +ψ ψ −φ +π
|a| = cos θ2 , arga = 2 , |b| = sin θ2 and arg b = 2 .
This shows that q maps the set
U = {(φ , θ , ψ ) | 0 ≤ φ < 2π , 0 ≤ θ ≤ π , −2π ≤ ψ < 2π }
onto S3 . Furthermore, q induces a bijection qe between the subsets
{(φ , θ , ψ ) ∈ U | 0 < θ < π } and {a + j b ∈ S3 | ab 6= 0 }
of U and S3 , so that qe(φ , θ , ψ ) = a + j b if and only if
φ = arga − argb + π2 , θ = 2 arg ( |a| + |b| i ) and ψ = arg a + argb − π2 .
By restricting the Euler angles to the open set
U ◦ = {(φ , θ , ψ ) | 0 < φ < 2π , 0 < θ < π , −2π < ψ < 2π } ,
they therefore are the coordinate functions of a local chart on S3 .

Combining q with the double covering projection ρ : S3 → SO(3) defined in 9.16,
we also obtain a smooth map R = ρ ◦ q : R3 → SO(3) given by
R(φ , θ , ψ ) = exp ( φ L3 ) exp ( θ L1 ) exp ( ψ L3 )

  
cos φ − sin φ cos θ sin φ sin θ cos ψ − sin ψ 0
=  sin φ cos φ cos θ − cos φ sin θ   sin ψ cos ψ 0  =
0 sin θ cos θ 0 0 1
 
cos φ cos ψ − sin φ sin ψ cos θ − cos φ sin ψ − sin φ cos ψ cos θ sin φ sin θ
 sin φ cos ψ + cos φ sin ψ cos θ − sin φ sin ψ + cos φ cos ψ cos θ − cos φ sin θ 
sin ψ sin θ cos ψ sin θ cos θ
for (φ , θ , ψ ) ∈ R3 , mapping the set
V = {(φ , θ , ψ ) | 0 ≤ φ < 2π , 0 ≤ θ ≤ π , 0 ≤ ψ < 2π }
onto SO(3). By restricting the Euler angles to the open set
V ◦ = {(φ , θ , ψ ) | 0 < φ < 2π , 0 < θ < π , 0 < ψ < 2π } ,
they are the coordinate functions of a local chart on SO(3).

GROUP ACTIONS 303
9.20 Remark The quaternion q = q(φ + π2 , θ , 0) corresponds to performing

a rotation by an angle θ about the x-axis, followed by a rotation by an angle
φ + π2 about the z-axis. Hence it will transform the vector k to the vector w =
(sin θ cos φ , sin θ sin φ , cos θ ) in agreement with the direct calculation
Ad (q)(k) = qk q −1

= exp φ +2π /2 k exp θ2 i k exp − θ2 i exp − φ +2π /2 k

φ + π /2 φ + π /2
= exp 2 k exp ( θ i) k exp − 2 k
(1)
= exp φ +2π /2 k (cos θ + sin θ i) exp − φ −2π /2 k

= exp π2 k cos θ + exp (φ k) sin θ i
= cos θ k + (cos φ + sin φ k) sin θ i
= sin θ cos φ i + sin θ sin φ j + cos θ k = t
where t = (0, w) . From this it also follows from 9.7 that

exp ψ2 t = exp ◦ Ad (q) ψ2 k = i q ◦ exp ψ2 k = q exp ψ
2 k q−1 .
This shows that a rotation by an angle ψ about the axis through the origin with
direction vector w can be performed as a composition of rotations about the x- and
z-axis in the following way :
(1) First rotate by the angle −(φ + π2 ) about the z-axis so that w is transformed to
the vector (0, − sin θ , cos θ ) in the yz-plane.
(2) Rotate by the angle −θ about the x-axis so that this vector is transformed to the
vector k along the z-axis.
(3) Now perform the rotation by the angle ψ about the z-axis.
(4) Rotate by the angle θ about the x-axis so that k is transformed back to the vector
(0, − sin θ , cos θ ).
(5) Finally rotate by the angle φ + π2 about the z-axis so that this vector is trans-
formed back to w.
THE LORENTZ GROUP AND Sl(2, C)
9.21 Let Re = α ◦ R : R3 → L ↑+ be the composition of R with the injective Lie

group homomorphism α : SO(3) → L ↑+ described in 8.62, and consider the smooth
map Λ : R6 → L ↑+ given by
e φ1 + π , θ1 , 0) exp ( χ B3 ) R(
Λ(φ1 , θ1 , χ , φ2 , θ2 , ψ2 ) = R( e φ2 , θ2 , ψ2 )
2
↑+
for (φ1 , θ1 , χ , φ2 , θ2 , ψ2 ) ∈ R6 . We contend that Λ maps the set W × R ×V onto L ,
where
W = {(φ , θ ) | − π2 ≤ φ < 32π , 0 ≤ θ ≤ π }
and V is the set defined in 9.19.
To show this, suppose that A ∈ L ↑+ , and let
r = (A200 − 1) 1/2 and χ = log (A00 + r)
so that
3
∑ A2r0 = r2 , A200 − r2 = 1 , cosh χ = A00 and sinh χ = r .
r=1
Now choose (φ1 , θ1 ) ∈ W so that
A10 = r sin θ1 cos φ1 , A20 = r sin θ1 sin φ1 and A30 = r cos θ1 ,
using spherical coordinates for the vector (A10 , A20 , A30 ) . Then we have that
e φ1 + π , θ1 , 0) exp ( χ B3 ) e0
R( 2
   
1 0 0 0 cosh χ 0 0 sinh χ 1
 0 − sin φ1 − cos θ1 cos φ1 sin θ1 cos φ1  0 1 0 0   0
=    
0 cos φ1 − cos θ1 sin φ1 sin θ1 sin φ1  0 0 1 0  0
0 0 sin θ1 cos θ1 sinh χ 0 0 cosh χ 0
   

1 0 0 0 A00 A00
 0 − sin φ1 − cos θ1 cos φ1 sin θ1 cos φ1  0   A10 
=
0
 = 
  A20  = A e0 .
cos φ1 − cos θ1 sin φ1 sin θ1 sin φ1  0
0 0 sin θ1 cos θ1 r A30
By 9.19 and Remark 8.63 there are therefore Euler angles (φ2 , θ2 , ψ2 ) ∈ V so that
Λ(φ1 , θ1 , χ , φ2 , θ2 , ψ2 ) = A , thus completing the proof of the assertion. By restrict-
ing the parameters to the open set W ◦ × R∗ × V ◦ , where R∗ = R − {0} and
W ◦ = {(φ , θ ) | − π2 < φ < 3π

2 ,0<θ <π},
they are the coordinate functions of a local chart on L ↑+ .

As the proper Lorentz group L ↑+ is the image of the connected set W × R × V
by the continuous map Λ, it must be connected. Hence it follows from Proposition
13.37 in the appendix that the sets L ↑+ , L ↑− , L ↓+ and L ↓− defined in 8.61 are
the connected components of the Lorentz group L .
GROUP ACTIONS 305
9.22 Let A = φ ◦ q : R3 → SU(2) be the composition of the smooth map q : R3 →

S with the Lie group isomorphism φ : S3 → SU(2) described in 9.19 and Example
3
8.67, and consider the smooth map S : R6 → Sl(2, C) given by
S(φ1 , θ1 , χ , φ2 , θ2 , ψ2 ) = A(φ1 + π2 , θ1 , 0) exp ( χ2 σ3 ) A(φ2 , θ2 , ψ2 )
for (φ1 , θ1 , χ , φ2 , θ2 , ψ2 ) ∈ R6 . We contend that S maps the set W × R × U onto

Sl(2, C) , where W and U are the sets defined in 9.19 and 9.21.
To show this, suppose that B ∈ Sl(2, C) , and let x ∈ R4 be the vector given by
3
i(x0 + x3 ) −x2 + ix1
iBB ∗ = B τ0 B ∗ = ∑ xr τr =
r=0
x2 + ix1 i(x0 − x3 )
satisfying
−x20 + x21 + x22 + x23 = det( iBB ∗ ) = −1 ,
where C = {τ0 , τ1 , τ2 , τ3 } is the basis for u(2) introduced in 8.65. In the same way
as in 9.21 we let
r = (x20 − 1) 1/2 and χ = log (x0 + r)
so that
3
∑ x2r = r2 , x20 − r2 = 1 , cosh χ = x0 and sinh χ = r .
r=1
We choose (φ1 , θ1 ) ∈ W so that
x1 = r sin θ1 cos φ1 , x2 = r sin θ1 sin φ1 and x3 = r cos θ1 ,
using spherical coordinates for the vector (x1 , x2 , x3 ) . Now we have that
i A(φ1 + π2 , θ1 , 0) exp ( χ σ3 ) A(φ1 + π2 , θ1 , 0) ∗

= exp φ +2π /2 τ3 exp θ2 τ1 (cosh χ τ0 + sinh χ τ3 ) exp − θ2 τ1 exp − φ +2π /2 τ3

φ + π /2
= x0 τ0 + r exp 2 τ3 exp θ
2 τ1 τ3 exp − θ2 τ1 exp − φ +2π /2 τ3
3
= x0 τ0 + r (sin θ cos φ τ1 + sin θ sin φ τ2 + cos θ τ3 ) = ∑ xr τr = iBB ∗
r=0
where the third equality is obtained by applying the Lie group isomorphism φ : S3 →
SU(2) to formula (1) in Remark 9.20. This implies that
{A(φ1 + π2 , θ1 , 0) exp ( χ2 σ3 ) }−1 B ∈ SU(2) .
By 9.19 there are therefore Euler angles (φ2 , θ2 , ψ2 ) ∈ U so that S(φ1 , θ1 , χ , φ2 , θ2 , ψ2 )

= B , thus completing the proof of the assertion. By restricting the parameters to the
open set W ◦ × R∗ ×U ◦ , they are the coordinate functions of a local chart on Sl(2, C).
It also follows that Sl(2, C) is connected, as it is the image of the connected set
W × R × U by the continuous map S .
9.23 Consider the representation σb : Sl(2, C) → Gl ( u(2)) defined by
σb (B)(A) = BA B ∗ = (i B) A (i B) ∗
for B ∈ Sl(2, C) and A ∈ u(2) . Using the linear isomorphism β : R4 → u(2) given by
3
i(x0 + x3 ) −x2 + ix1
β (x) = ∑ xr τr = x2 + ix1 i(x0 − x3 )
r=0
for x ∈ R4 , we also have a representation σ

e : Sl(2, C) → Gl (4, R) defined by
σe (B) = mEE ( β −1 ◦ σb (B) ◦ β )
for B ∈ Sl(2, C), where E is the standard basis in R4 .

b (B)(A)) = det(A) for A ∈ u(2) when B ∈ Sl(2, C), and
Since det( σ
det( β (x)) = −x20 + x21 + x22 + x23
for x ∈ R4 , it follows that σ e (Sl(2, C)) ⊂ L↑+ as

e (Sl(2, C)) ⊂ O1 (4) . Actually, σ
σe is a continuous map from the connected set Sl(2, C) . We want to show that
σe (Sl(2, C)) = L↑+ , and that the homomorphism σ : Sl(2, C) → L↑+ induced by σe
is a double covering of L↑+ .
Let B = {u, v, w} be a positively oriented orthonormal basis in R3 , and let r =
β (0, u), s = β (0, v) and t = β (0, w) be the corresponding matrices in su(2) ⊂ u(2) .
Since β coincides with the algebra homomorphism φ : H → gl(2, C) described in
8.65 on {0} × R3, we have that
r2 = s2 = t 2 = −σ0
and
rs = −s r = t , s t = −ts = r and tr = −rt = s .
Now let B = exp ( χ2 b) where b = w1 σ1 + w2 σ2 + w3 σ3 and χ ∈ R . Since t = ib ,
we have that b2 = σ0 so that
i B = cosh χ2 τ0 + sinh χ2 t and (i B) ∗ = −i B .
This implies that
σb (B)(τ0 ) = cosh χ τ0 + sinh χ t ,

σb (B)(r) = r ,
σb (B)(s) = s ,
σb (B)(t) = sinh χ τ0 + cosh χ t ,
GROUP ACTIONS 307
so that  
cosh χ 0 0 sinh χ
 0 1 0 0 
b (B) ◦ β ) = 
mCC ( β −1 ◦ σ 


0 0 1 0
sinh χ 0 0 cosh χ
using the basis C = {e0 , (0, u), (0, v), (0, w)} in R4 . Hence β −1 ◦ σb (B) ◦ β is a ve-
locity transformation or boost with velocity parameter χ along the axis through the
origin with direction vector w. In particular, we have that
σ ◦ exp ( χ2 σ3 ) = exp ( χ B3 )
for χ ∈ R . Since we also have that σ ◦ A = Re , it follows from 9.21 and 9.22 that
σe (Sl(2, C)) = L↑+ . If B ∈ ker σ , then B ∈ SU(2) since σb (B)(τ0 ) = τ0 . Hence it fol-
lows from 9.16 that ker σ = {σ0 , −σ0 } which is a discrete subgroup of Sl(2, C). By
Proposition 13.111 in the appendix we therefore conclude that the homomorphism
σ : Sl(2, C) → L↑+ is a double covering of the proper Lorentz group L↑+ .
Since
σ ∗ | su (2) = α ∗ ◦ ρ ∗ ◦ φ ∗−1
and
σ ◦ exp ( χ2 b) = exp χ σ ∗ b
2
for χ ∈ R , it follows from 8.62, 8.65, 9.16 and Example 8.67 that the Lie algebra
isomorphism σ ∗ : sl(2, C) → o 1 (4) maps the basis { τ1 /2, τ2 /2, τ3 /2, σ1 /2, σ2 /2,
σ3 /2} for sl(2, C) onto the basis {R1 , R2 , R3 , B1 , B2 , B3 } for o 1 (4) .
9.24 Remark The matrix A = A(φ + π2 , θ , 0) in SU(2) corresponds to per-

forming a rotation by an angle θ about the x-axis, followed by a rotation by an
angle φ + π2 about the z-axis. Hence it will transform the vector k to the vec-
tor w = (sin θ cos φ , sin θ sin φ , cos θ ) . By applying the Lie group isomorphism
φ : S3 → SU(2) to formula (1) in Remark 9.20 and dividing by i , we obtain
Ad (A)(σ3 ) = w1 σ1 + w2 σ2 + w3 σ3 = b
so that
χ χ χ χ
exp 2 b = exp ◦ Ad (A) 2 σ3 = i A ◦ exp 2 σ3 = A exp 2 σ3 A−1 .
Hence a boost with velocity parameter χ along the axis through the origin with direc-
tion vector w can be performed as a composition of the five transformations described
in Remark 9.20, where the rotation (3) is replaced by a boost with velocity parameter
χ along the z-axis.
9.25 Remark The physical interpretation of the coordinate x0 is that x0 = ct ,

where t is time and c is the velocity of light in vacuum. The velocity v is given by
tanh χ = v/c ,
where χ is the velocity parameter. From this it follows that
cosh χ = (1 − v2/c2 ) −1/2 = γ and sinh χ = γ v/c .
The boost exp ( χ B3 ) gives the connection between the coordinates (ct, x, y, z) and
(ct ′ , x′ , y′ , z′ ) with respect to two inertial systems S and S ′ with parallel coordinate
axes, where S ′ is moving with constant velocity v with respect to S in the positive
z-direction so that their origins coincide at the time t = t ′ = 0. Then we have that
    ′ 
ct γ 0 0 γ v/c ct
 x   0 1 0 0   x′ 
    
 y = 0 0 1 0   y′ 
z γ v/c 0 0 γ z′
which implies that
x = x′ , y = y′ , z = γ (z′ + vt ′ ) , t = γ (t ′ + z′ v/c2 ) .
SEMIDIRECT PRODUCTS
9.26 Proposition Let Fi : Vi → Vi be a linear map in the finite dimensional vector

space Vi for i = 1, 2, 3, and let B : V1 ×V2 → V3 be a bilinear map. Then we have that
F3 (B(u, v)) = B(F1 (u), v) + B(u, F2(v)) (1)
for (u, v) ∈ V1 × V2 if and only if
exp (t F3 )(B(u, v)) = B(exp (t F1 )(u), exp (t F2 )(v)) (2)
for (u, v) ∈ V1 × V2 and t ∈ R.

PROOF : By 4.23 the relation (1) implies that
t F3 ◦ B ∗ = B ∗ ◦ ( t F1 ⊗ id V2 + id V1 ⊗ t F2 )
for t ∈ R , where B ∗ : V1 ⊗ V2 → V3 is the linear map induced by B . Hence we have

that
(t F3 ) k ◦ B ∗ = B ∗ ◦ ( t F1 ⊗ id V2 + id V1 ⊗ t F2 ) k
for every positive integer k , which implies that
!
n n
( t F3 ) k ( t F1 ⊗ id V2 + id V1 ⊗ t F2 ) k
∑ k!
◦ B∗ = B∗ ◦ ∑ k!
k=0 k=0
GROUP ACTIONS 309
for every positive integer n . By letting n → ∞ , we thus obtain

exp ( t F3 ) ◦ B ∗ = B ∗ ◦ exp ( t F1 ⊗ id V2 + id V1 ⊗ t F2 ) .
Now using the relation
!
n n
( t F1 ⊗ id V2 ) k ( t F1 ) k
∑ k!
= ∑ k!
⊗ id V2
k=0 k=0
and letting n → ∞ , we obtain the formula

exp ( t F1 ⊗ id V2 ) = exp ( t F1 ) ⊗ id V2 ,
and similarly we have that
exp ( id V1 ⊗ t F2 ) = id V1 ⊗ exp ( t F2 ) .
Since
( t F1 ⊗ id V2 ) ◦ ( id V1 ⊗ t F2 ) = ( id V1 ⊗ t F2 ) ◦ ( t F1 ⊗ id V2 ) ,
it therefore follows from Proposition 13.126 in the appendix that
exp ( t F1 ⊗ id V2 + id V1 ⊗ t F2 ) = exp ( t F1 ⊗ id V2 ) ◦ exp ( id V1 ⊗ t F2 )

= { exp ( t F1 ) ⊗ id V2 } ◦ { id V1 ⊗ exp ( t F2 ) } = exp ( t F1 ) ⊗ exp ( t F2 )
so that
exp ( t F3 ) ◦ B ∗ = B ∗ ◦ { exp ( t F1 ) ⊗ exp ( t F2 ) }
Conversely, fix (u, v) ∈ V1 × V2 and assume that (2) is satisfied for every t ∈ R .
Let c : R → V3 be the curve defined by
c(t) = B(exp (t F1 )(u), exp (t F2 )(v))
for t ∈ R . If we canonically identify Tc(0)V3 with V3 as in Lemma 2.84 and use Re-
mark 2.75 and Proposition 8.29 and 8.31, we have that
F3 (B(u, v)) = c′ (0) = B(F1 (u), v) + B(u, F2(v))
which shows that (1) is satisfied.
9.27 Proposition The set of Lie algebra automorphisms

Aut(g) = {φ ∈ Gl (g)|φ ([ X,Y ]) = [ φ (X), φ (Y ) ] for X,Y ∈ g }
in a Lie algebra g is a closed Lie subgroup of Gl (g) in the relative topology with
the Lie subalgebra
∂ (g) = {ψ ∈ gl (g)|ψ ([ X,Y ]) = [ ψ (X),Y ] + [ X, ψ (Y ) ] for X,Y ∈ g }
of gl(g) , consisting of the derivations in g , as its Lie algebra.
PROOF : Since Aut(g) is a closed subgroup of Gl (g) , it follows from Theorem

8.43 that it is a closed Lie subgroup of Gl (g) in the relative topology. Let h be its
Lie algebra.
By Proposition 8.41 we know that an element ψ ∈ gl(g) belongs to h if and
only if exp (t ψ ) ∈ Aut(g) for every t ∈ R. Hence we conclude that h = ∂ (g) by
Proposition 9.26, using the bilinear map B : g × g → g given by B(X,Y ) = [ X,Y ]
for X,Y ∈ g .
9.28 Let g and h be Lie algebras, and let σ : g → ∂ (h) be a Lie algebra homomor-
phism of g into the Lie algebra ∂ (h) of derivations in h. Then we have a Lie algebra
structure on g × h given by
[ (X1 ,Y1 ), (X2 ,Y2 ) ] = ( [ X1 , X2 ] , [Y1 ,Y2 ] + σ (X1 )(Y2 ) − σ (X2 )(Y1 ) )
for X1 , X2 ∈ g and Y1 ,Y2 ∈ h, satisfying the Jacobi identity since
[ (X1 ,Y1 ), [ (X2 ,Y2 ), (X3 ,Y3 ) ] ] + [ (X2 ,Y2 ), [ (X3 ,Y3 ), (X1 ,Y1 ) ] ]
+ [ (X3 ,Y3 ), [ (X1 ,Y1 ), (X2 ,Y2 ) ] ]
= ( 0 , [ Y1 , σ (X2 )(Y3 ) − σ (X3 )(Y2 ) ]

+ σ (X1 ) { [ Y2 ,Y3 ] + σ (X2 )(Y3 ) − σ (X3 )(Y2 ) } − σ ([ X2 , X3 ])(Y1 )
+ [ Y2 , σ (X3 )(Y1 ) − σ (X1 )(Y3 ) ]

+ σ (X2 ) { [ Y3 ,Y1 ] + σ (X3 )(Y1 ) − σ (X1 )(Y3 ) } − σ ([ X3 , X1 ])(Y2 )
+ [ Y3 , σ (X1 )(Y2 ) − σ (X2 )(Y1 ) ]

+ σ (X3 ) { [ Y1 ,Y2 ] + σ (X1 )(Y2 ) − σ (X2 )(Y1 ) } − σ ([ X1 , X2 ])(Y3 ) )
= ( 0 , {− [ Y2 , σ (X1 )(Y3 ) ] − [ σ (X1 )(Y2 ),Y3 ] + σ (X1 )([ Y2 ,Y3 ]) }

+ { σ (X2 ) ◦ σ (X3 ) − σ (X3 ) ◦ σ (X2 ) − σ ([ X2 , X3 ]) } (Y1 )
+ {− [ Y3 , σ (X2 )(Y1 ) ] − [ σ (X2 )(Y3 ),Y1 ] + σ (X2 )([ Y3 ,Y1 ]) }

+ { σ (X3 ) ◦ σ (X1 ) − σ (X1 ) ◦ σ (X3 ) − σ ([ X3 , X1 ]) } (Y2 )
+ {− [ Y1 , σ (X3 )(Y2 ) ] − [ σ (X3 )(Y1 ),Y2 ] + σ (X3 )([ Y1 ,Y2 ]) }

+ { σ (X1 ) ◦ σ (X2 ) − σ (X2 ) ◦ σ (X1 ) − σ ([ X1 , X2 ]) } (Y3 ) ) = ( 0, 0 )
for X1 , X2 , X3 ∈ g and Y1 ,Y2 ,Y3 ∈ h. We see that g × h is a Lie algebra with this Lie
algebra structure. It is called the semidirect product of g and h with respect to σ , and
it is denoted by g ×σ h.
The Lie algebra homomorphisms i 1 : g → g ×σ h and i 2 : h → g ×σ h , given by
GROUP ACTIONS 311
i 1 (X) = (X , 0) and i 2 (Y ) = (0,Y ) for X ∈ g and Y ∈ h , are injections so that i 1 (g)

is a Lie subalgebra and i 2 (h) is an ideal in g ×σ h with
g ×σ h = i 1 (g) ⊕ i 2 (h) .
We also have a Lie algebra homomorphism ψ 1 : g ×σ h → g given by ψ 1 (X,Y ) = X

for X ∈ g and Y ∈ h.
9.29 Proposition Let k be a Lie algebra, and let g be a Lie subalgebra and h be
an ideal in k so that
k=g⊕h.
Then the map σ : g → ∂ (h) given by σ (X)(Y ) = [ X,Y ] for X ∈ g and Y ∈ h , is a
Lie algebra homomorphism of g into the Lie algebra ∂ (h) of derivations in h, and
the map ψ : g ×σ h → k given by ψ (X,Y ) = X + Y for X ∈ g and Y ∈ h , is a Lie
algebra isomorphism.
PROOF : Since h is an ideal in k, we have that σ (X)(h) ⊂ h for every X ∈ g , and
σ (X) is a derivation in h by the Jacobi identity which implies that
σ (X)([Y1 ,Y2 ]) = [ X, [ Y1 ,Y2 ] ] = [[ X,Y1 ],Y2 ] + [ Y1 , [ X,Y2 ] ]

= [ σ (X)(Y1 ),Y2 ] + [ Y1 , σ (X)(Y2 ) ]
for every X ∈ g and Y1 ,Y2 ∈ h . It also implies that
σ ([ X1 , X2 ])(Y ) = [ [ X1 , X2 ],Y ] = [ X1 , [ X2 ,Y ] ] − [ X2 , [ X1 ,Y ] ]
= { σ (X1 ) ◦ σ (X2 ) − σ (X2 ) ◦ σ (X1 )}(Y )
for every X1 , X2 ∈ g and Y ∈ h , which shows that σ : g → ∂ (h) is a Lie algebra

homomorphism. Finally, we see that ψ : g ×σ h → k is a linear isomorphism which
is also a Lie algebra isomorphism since
ψ ( [ (X1 ,Y1 ), (X2 ,Y2 ) ] ) = ψ ( [ X1 , X2 ] , [Y1 ,Y2 ] + σ (X1 )(Y2 ) − σ (X2 )(Y1 ) )
= [ X1 , X2 ] + [ Y1 ,Y2 ] + [ X1 ,Y2 ] + [ Y1 , X2 ] = [ X1 + Y1 , X2 + Y2 ]
= [ ψ (X1 ,Y1 ), ψ (X2 ,Y2 ) ]
for X1 , X2 ∈ g and Y1 ,Y2 ∈ h .
9.30 Let G and H be Lie groups, and let ρ : G → Aut (H) be a homomorphism
of G into the group Aut (H) of Lie group automorphisms of H, so that the map
φ : G× H → H given by φ (g, h) = ρ (g)(h) is smooth. Then we have a group structure
on G × H given by
(g1 , h1 )(g2 , h2 ) = (g1 g2 , h1 ρ (g1 )(h2 ))

for g1 , g2 ∈ G and h1 , h2 ∈ H, satisfying the associative law since
{( g1 , h1 ) ( g2 , h2 ) } ( g3 , h3 ) = (g1 g2 , h1 ρ (g1 )(h2 )) ( g3 , h3 )

= ( (g1 g2 ) g3 , h1 ρ (g1 )(h2 ) ρ (g1 g2 )(h3 )) = ( g1 (g2 g3 ), h1 ρ (g1 )(h2 ρ (g2 )(h3 )))
= ( g1 , h1 ) ( g2 g3 , h2 ρ (g2 )(h3 )) = ( g1 , h1 ) {( g2 , h2 ) ( g3 , h3 ) }
for g1 , g2 , g3 ∈ G and h1 , h2 , h3 ∈ H. The unit element is (eG , eH ), where eG and eH

are the unit elements in G and H, and we have that
(g, h)−1 = (g−1 , ρ (g−1 )(h−1 ))
for g ∈ G and h ∈ H. We see that G × H is a Lie group with this group structure and
the product manifold structure. It is called the semidirect product of G and H with
respect to ρ , and it is denoted by G ×ρ H.
The Lie group homomorphisms i 1 : G → G ×ρ H and i 2 : H → G ×ρ H, given
by i 1 (g) = (g, eH ) and i 2 (h) = (eG , h) for g ∈ G and h ∈ H, are embeddings so that
i 1 (G) and i 2 (H) are closed Lie subgroups of G ×ρ H, and we have that
(g1 , h1 )(g2 , h2 )(g1 , h1 )−1 = (g1 g2 g−1 −1 −1

1 , h1 ρ (g1 )(h2 ) ρ (g1 g2 g1 )(h1 ))
for g1 , g2 ∈ G and h1 , h2 ∈ H. In particular, we have that

i 1 (g) i 2 (h) i 1 (g)−1 = i 2 ◦ ρ (g)(h) (1)
for g ∈ G and h ∈ H. We also have a Lie group homomorphism ψ 1 : G ×ρ H → G
given by ψ 1 (g, h) = g for g ∈ G and h ∈ H.
Let g and h be the Lie algebras of G and H, respectively, and let k be the Lie
algebra of G ×ρ H. By Propositions 8.8 and 9.27 and Corollaries 9.5 and 2.38 we
have a Lie group homomorphism φ : G → Aut (h) given by φ (g) = ρ (g) ∗ for g ∈ G ,
and σ = φ ∗ : g → ∂ (h) is therefore a Lie algebra homomorphism. We want to show
that the map
ψ : g ×σ h → k
given by
ψ (X,Y ) = i 1 ∗ (X) + i 2 ∗ (Y )
for X ∈ g and Y ∈ h , is a Lie algebra isomorphism. By Proposition 2.74 we know
that ψ is a linear isomorphism since
k = i 1 ∗ (g) ⊕ i 2 ∗ (h) ,
where i 1 ∗ : g → k and i 2 ∗ : h → k are injective Lie algebra homomorphisms. Now it
follows from (1) that
i i 1 (g) ◦ i 2 = i 2 ◦ ρ (g)
for g ∈ G, where i i 1 (g) : G ×ρ H → G ×ρ H is the inner automorphism of G ×ρ H by
i 1 (g) as defined in 9.7, and this implies that
Ad i 1 (g) ◦ i 2 ∗ = i 2 ∗ ◦ φ (g)
GROUP ACTIONS 313
so that
Λ 1 ◦ Ad ◦ i 1 = Λ 2 ◦ φ ,
where Λ 1 : gl (k) → L(h, k) and Λ 2 : gl (h) → L(h, k) are the linear maps given by
Λ 1 (F) = F ◦ i 2 ∗ and Λ 2 (G) = i 2 ∗ ◦ G for F ∈ gl (k) and G ∈ gl (h) . Hence we
have that
Λ 1 ◦ ad ◦ i 1 ∗ = Λ 2 ◦ σ ,
which means that
[ i 1 ∗ (X), i 2 ∗ (Y ) ] = i 2 ∗ (σ (X)(Y ))
for X ∈ g and Y ∈ h . This implies that ψ is a Lie algebra isomorphism since
ψ ( [ (X1 ,Y1 ), (X2 ,Y2 ) ] ) = ψ ( [ X1 , X2 ] , [Y1 ,Y2 ] + σ (X1 )(Y2 ) − σ (X2 )(Y1 ) )
= i 1 ∗ ( [ X1 , X2 ] ) + i 2 ∗ ( [Y1 ,Y2 ] + σ (X1 )(Y2 ) − σ (X2 )(Y1 ) )
= [ i 1 ∗ (X1 ), i 1 ∗ (X2 ) ] + [ i 2 ∗ (Y1 ), i 2 ∗ (Y2 ) ] + [ i 1 ∗ (X1 ), i 2 ∗ (Y2 ) ] + [ i 2 ∗ (Y1 ), i 1 ∗ (X2 ) ]
= [ i 1 ∗ (X1 ) + i 2 ∗ (Y1 ), i 1 ∗ (X2 ) + i 2 ∗(Y2 ) ] = [ ψ (X1 ,Y1 ), ψ (X2 ,Y2 ) ]
for X1 , X2 ∈ g and Y1 ,Y2 ∈ h .
AFFINE SPACES
9.31 Definition A set E on which a vector space V acts freely and transitively is
called an affine space , and the action of a vector v ∈ V on a point p ∈ E is denoted
by p + v. We say that E is modelled on the vector space V . Each vector v ∈ V can be
identified with a bijection Tv : E → E given by Tv (p) = p + v for p ∈ E. This Tv is
called the translation on E by v, and V is called the vector space of translations on E.
For each pair of points p, q ∈ E there is a unique vector v ∈ V with p + v = q which
is denoted by q − p .
A map F : E1 → E2 between the affine spaces E1 and E2 is called affine if there
is a linear map T : V1 → V2 between the corresponding vector spaces of translations
so that
F(p + v) = F(p) + T (v) (1)
for every p ∈ E1 and v ∈ V1 . The map T is called the linear part of F. If G : E2 →
E3 is an affine map into the affine space E3 with linear part S : V2 → V3 , then the
composition G ◦ F is again an affine map with linear part S ◦ T .
A subset E1 of an affine space E2 is called an affine subspace of E2 if it is an affine
space having the same vector space V of translations as E2 , so that the inclusion map
i : E1 → E2 is an affine map with the identity map idV : V → V as its linear part. This
means that each translation on E2 by a vector v ∈ V maps E1 onto itself and induces
the translation on E1 by v.
9.32 Proposition Let F : E1 → E2 be an affine map with T : V1 → V2 as its

linear part. Then F is bijective if and only if T is bijective, and the inverse map
F −1 : E2 → E1 is affine with T −1 : V2 → V1 as its linear part.
PROOF : The first part of the propositiom follows from the relation
T (p2 − p1) = F(p2 ) − F(p1 )
for p1 , p2 ∈ E1 . This also implies that
T −1 (q2 − q1) = F −1 (q2 ) − F −1 (q1 )
for q1 , q2 ∈ E2 , completing the proof of the last part of the propsition.
9.33 Example A vector space V acts freely and transitively on itself and is
therefore an affine space. A map F : V1 → V2 between the vector spaces V1 and V2 is
affine if it is the composition of a linear map T : V1 → V2 and a translation Tw on V2
by a vector w ∈ V2 , i.e.,
F = Tw ◦ T
so that
F(v) = T (v) + w
for every v ∈ V1 . This follows from formula (1) in Definition 9.31 by taking p = 0
and w = F(0). Conversely, if F satisfies the above formula, then
F(u + v) = T (u + v) + w = T (u) + T (v) + w = F(u) + T (v)
for every u, v ∈ V1 , showing that F is affine with T as its linear part.
9.34 Example Let E be an affine space modelled on a vector space V , and fix
a point o ∈ E . Then we have a bijection S o : V → E given by S o (v) = o + v for
v ∈ V , which is an affine map with the identity map idV : V → V as its linear part.
The inverse map S o−1 : E → V is given by S o−1 (p) = p − o for p ∈ E , and the vector
p − o is called the position vector of p . There is a unique vector space structure on
E such that So is a linear isomorphism, which is given by
a p + b q = S o (a S o−1 (p) + b S o−1 (q))
i.e.,
a p + b q = o + a(p − o) + b(q − o)
for a, b ∈ R and p, q ∈ E . In this vector space structure o becomes the zero vector.
9.35 Example Let ρ : G → Gl (V ) be a representation of a Lie group G in a

finite dimensional vector space V , and consider the semidirect product G ×ρ V . Let
GA (V ) be the group of invertible affine maps F : V → V given by F(w) = T (w) + v
GROUP ACTIONS 315
for w ∈ V , where T ∈ Gl (V ) and v ∈ V , and where the group product is composition.

Then we have a homomorphism σ : G ×ρ V → GA (V ) given by
σ (g, v)(w) = ρ (g)(w) + v
for (g, v) ∈ G ×ρ V and w ∈ V . Indeed, we have that
σ ((g1 , v1 )(g2 , v2 ))(w) = σ (g1 g2 , ρ (g1 )(v2 ) + v1)(w) = ρ (g1 g2 )(w) + ρ (g1 )(v2 ) + v1
= ρ (g1 )(ρ (g2 )(w) + v2 ) + v1 = σ (g1 , v1 ) ◦ σ (g2 , v2 )(w)
for every (g1 , v1 ), (g2 , v2 ) ∈ G ×ρ V and w ∈ V . In the special case where ρ = id :

Gl (V ) → Gl (V ), σ is a group isomorphism, and we give GA (V ) the manifold struc-
ture so that σ is a Lie group isomorphism.
We now have a homomorphism ψ : GA (V ) → Gl (V × R) which is given by
ψ ◦ σ (T, v)(w,t) = ( T (w) + t v, t)
for (T, v) ∈ Gl (V ) ×id V and (w,t) ∈ V × R . Indeed, we have that
ψ ( σ ( T1 , v1 ) ◦ σ ( T2 , v2 )) (w,t) = ψ ◦ σ (( T1 , v1 )( T2 , v2 )) (w,t)
= ψ ◦ σ ( T1 ◦ T2 , T1 (v2 ) + v1 )(w,t) = ( T1 ◦ T2 (w) + T1 (t v2 ) + t v1 ,t)
= ( T1 ( T2 (w) + t v2 ) + t v1 ,t) = {ψ ◦ σ ( T1 , v1 )} ◦ {ψ ◦ σ ( T2 , v2 )} (w,t)
for every (T1 , v1 ), (T2 , v2 ) ∈ Gl (V ) ×id V and (w,t) ∈ V × R . If α : V → V × R is

the affine map defined by α (w) = (w, 1) for w ∈ V , then
ψ (F) ◦ α = α ◦ F
for every F ∈ GA (V ) . Now since α (V ) = V × {1}, we obtain a map ψ1 : GA (V ) →

GA (V × {1}) which is given by
ψ1 ◦ σ (T, v)(w, 1) = ( T (w) + v, 1)
for (T, v) ∈ Gl (V ) ×id V and w ∈ V . If we let α1 : V → V × {1} be the invertible

affine map induced by α , it follows that
F = α1−1 ◦ ψ1 (F) ◦ α1
for every F ∈ GA (V ) . Given a basis B = {v1 , ..., vn } for V , we have that C =

{(v1 , 0), ..., (vn , 0), (0, 1)} is a basis for V × R. If mB
B (T ) = A and [v]B = a, then

C A a
mC (ψ ◦ σ (T, v)) = .
0 1
Let φ : Gl (V × R) → Gl (n + 1, R) be the Lie group isomorphism given by

φ (S) = mCC (S) for S ∈ Gl (V × R) , and let i 1 : Gl (V ) → Gl (V ) ×id V and i 2 : V →
Gl (V ) ×id V be the Lie group homomorphisms defined as in 9.30 by i 1 (T ) = (T , 0)

and i 2 (v) = (In , v) for T ∈ Gl (V ) and v ∈ V . Then φ ◦ ψ ◦ σ ◦ i 1 (Gl (V )) = G ,
φ ◦ ψ ◦ σ ◦ i 2 (V ) = H and φ ◦ ψ ◦ σ (Gl (V ) ×id V ) = K are the closed Lie sub-
groups of Gl (n + 1, R) given by

A 0 In a
G={ | A ∈ Gl (n, R)} , H = { | a ∈ Rn }
0 1 0 1
and
A a
K={ | A ∈ Gl (n, R) and a ∈ Rn } .
0 1
For every B ∈ gl (n, R) and a ∈ Rn we have that

B 0 exp (tB) 0
exp t =
0 0 0 1
and
0 a In ta
exp t =
0 0 0 1
for t ∈ R , which follows from the relation
 n 
n k ( t B) k
∑ 1 tB 0
=  ∑ k!
0 
k=0 k!
0 0 k=0
0 1
when n → ∞ , and the fact that

k
0 ta
=0
0 0
when k ≥ 2. Hence the Lie algebras of G, H and K are the Lie subalgebras of gl (n +
1, R) given by

B 0 0 a
g={ | B ∈ gl (n, R)} , h = { | a ∈ Rn }
0 0 0 0
and
B a
k = g⊕h = { | B ∈ gl (n, R) and a ∈ Rn } .
0 0
The bracket product in gl (V ) ×id V is defined as in 9.28 and 9.30 by
[ (T1 , v1 ), (T2 , v2 ) ] = (T1 ◦ T2 − T2 ◦ T1 , T1 (v2 ) − T2 (v1 ))
for (T1 , v1 ), (T2 , v2 ) ∈ gl (V ) ×id V .
9.36 Lemma Let E be an affine space modelled on a vector space V of dimension

n. For each point p ∈ E we have a bijective affine map S p : V → E defined by S p (v) =
GROUP ACTIONS 317
p + v for v ∈ V . We give E the manifold structure so that S p is a diffeomorphism,

which is independent of the choice of p. Then the tangent space Tp E at any point
p ∈ E may be identified with V by means of the linear isomorphism ω p : Tp E → V
given by ω p = x−1 ◦ tx′ ,p for any bijective affine map x′ : E → Rn with linear part
x : V → Rn , ω p being independent of the choice of x′ . We will refer to this as the
canonical identification.
If E = {e1 , ..., en } is the standard basis for Rn and B = {v1 , ..., vn } is a basis for
V so that x(vi ) = ei for i = 1, ..., n, then B ∗ = {x1 , ..., xn } is the dual basis of B, and

n n
ω p ( ∑ a ∂′ j ) = ∑ a j v j .
j
j=1
∂x j=1
p
If l ∈ Tp E , then
i
xi ◦ ω p (l ) = l (x′ )
for i = 1, ..., n, and
λ ◦ ω p (l ) = l (λ ◦ S p−1 )
for any linear functional λ on V .
The map ω = (idE × x−1 ) ◦ tx′ : T E → E × V is an equivalence over E so that

n
ω ( ∑ a j ∂′ j ) = (p, x−1 (a))
j=1
∂x
p
and
n
ω −1
(p, v) = ∑ x (v) ∂′ j
j
j=1
∂x
p
for p ∈ E , v ∈ V and a ∈ Rn .
If F : E1 → E2 is an affine map with T : V1 → V2 as its linear part, and if the
tangent spaces Tp E1 and TF(p) E2 are canonically identified with V1 and V2 by means
of the linear isomorphisms ω p : Tp E1 → V1 and ωF(p) : TF(p) E2 → V2 , then we have
that
ωF(p) ◦ F∗ p ◦ ω p−1 = T .
PROOF : Since Sq = S p ◦ Tv when q = S p (v) and Tv : V → V is translation by v, the
manifold structure on E is independent of the choice of p. We have that (x′ , E) is a
local chart on E since x′ ◦ S p (v) = x′ (p) + x(v) for v ∈ V , which implies that x′ =
Tx′ (p) ◦ x ◦ S p−1 where Tx′ (p) : Rn → Rn is translation by x′ (p). As tx′ ,p : TpV → Rn is
a linear isomorphism, the same is true for the map ω p . To show that it is independent
of the choice of x′ , let y′ : E → Rn be another bijective affine map with linear part
y : V → Rn . Then it follows from 2.70 that
−1
ty′ ,p ◦ tx−1 ′
′ ,p = D (y ◦ x
′
)(x′ (p)) = D (y ◦ x−1 )(0) = y ◦ x−1 ,
so that
y−1 ◦ ty′ ,p = x−1 ◦ tx′ ,p .

n
To prove the next part of the lemma, let l = ∑ a ∂′ j . Then j
j=1
∂x
p
n
ω p (l ) = x−1 (a) = ∑ a j vj ,
j=1
and
i
xi ◦ ω p (l ) = ai = l (x′ )
for i = 1, ..., n. If
n
λ= ∑ bj xj ,
j=1
then
n n n
j j
λ ◦ ω p (l ) = ∑ b j x j ◦ ω p (l ) = ∑ b j l (x′ )=l( ∑ b j x′ ) = l (λ ◦ S p−1 ) .
j=1 j=1 j=1
Now assume that the linear isomorphisms ω p : Tp E1 → V1 and ωF(p) : TF(p) E2 →

V2 are obtained from the bijective affine maps x′ : E1 → Rn and y′ : E2 → Rm with
linear parts x : V1 → Rn and y : V2 → Rm , respectively. Then the last part of the lemma
follows from the commutative diagram in 2.70 which shows that
ωF(p) ◦ F∗ p ◦ ω p−1 = y−1 ◦ ty′ ,F(p) ◦ F∗ p ◦ tx−1

′ ,p ◦ x
−1
= y−1 ◦ D (y′ ◦ F ◦ x′ )(x(p)) ◦ x
= y−1 ◦ D (y ◦ SF(p)
−1
◦ F ◦ S p ◦ x−1 )(0) ◦ x
= y−1 ◦ y ◦ T ◦ x−1 ◦ x = T
since
F ◦ S p (v) = F(p + v) = F(p) + T (v) = SF(p) ◦ T (v)
for v ∈ V .
INFINITESIMAL GROUP ACTIONS
9.37 Proposition Let G be a Lie group acting on the smooth manifold M on the
right. For each p ∈ M, we let σ p : G → M be the smooth map defined by σ p (g) = pg
for g ∈ G. Then we have a Lie algebra homomorphism σ : g → T 1 M defined by
σ (X) p = σ p ∗ (X)
GROUP ACTIONS 319
for X ∈ g and p ∈ M, and σ (X) is σ p -related to the left invariant vector field Xe on G
determined by X. The homomorphism σ is called the infinitesimal action of G on
M on the right.
PROOF : We first need to show that σ (X) is smooth on M so that it is a vector field
on M. Let (z,U) be a local chart around a point p on M, and let n be the dimension of
M at p. If ν : M × G → M is the operation of G on M, there is an open neighbourhood
V of p on M and a local chart (y,W ) around e on G so that ν (V × W ) ⊂ U, and we
let x = z|V . Then the map f : x(V ) × y(W ) → Rn defined by f = z ◦ ν ◦ (x−1 × y−1 )
is smooth, and we have that
f (a, b) = z ◦ σx−1 (a) ◦ y−1 (b)
for a ∈ x(V ) and b ∈ y(W ).

Now let (tz , π −1 (U)) be the local trivialization in the tangent bundle π : T M → M
assosiated with the local chart (z,U) on M. If X = [y, v] e , it follows that
σ (X) ◦ x−1 (a) = σx−1 (a) ∗ (X) = [z, D (z ◦ σx−1 (a) ◦ y−1 )(y(e)) v] x−1 (a)
= [z, D2 f (a, y(e)) v] x−1 (a)
so that
(z × id) ◦ tz ◦ σ (X) ◦ x−1 (a) = (a, D2 f (a, y(e)) v)
for a ∈ x(V ), thus showing that σ (X) is smooth in V .
To see that the vector field σ (X) is σ p -related to Xe for every X ∈ g and p ∈ M,
we first note that
σ p ◦ L g = σ pg
for every g ∈ G, since
σ p ◦ L g (h) = σ p (gh) = p (gh) = (pg) h = σ pg (h)
for h ∈ G, and this implies that
σ p ∗ (Xeg ) = σ p ∗ ◦ Lg ∗ (X) = (σ p ◦ Lg ) ∗ (X) = σ pg ∗ (X) = σ (X) σ p (g) .
From this it now follows that σ is a Lie algebra homomorphism. Indeed, we see
immediately from the definition that σ is linear, and if X,Y ∈ g, then [ σ (X), σ (Y ) ]
e Ye ] by Proposition 4.88, so that
is σ p -related to [ X,
e Ye ] e ) = [ σ (X), σ (Y ) ] p
σ ( [ X,Y ] ) p = σ p ∗ ( [ X,
for every p ∈ M.
9.38 Proposition If G is a Lie group acting on the smooth manifold M on the

right, then
(R g ) ∗ σ (X) = σ ( Ad (g−1 ) X )
for every X ∈ g and g ∈ G.
PROOF : Since
R g ◦ σ p (h) = phg = pg g−1 hg = σ pg ◦ ig−1 (h)
for p ∈ M and h ∈ G, we have that
((R g ) ∗ σ (X)) pg = (R g )∗ σ (X) p = (R g ) ∗ ◦ σ p ∗ (X)
= σ pg ∗ ◦ Ad (g−1 ) (X) = σ ( Ad (g−1 ) X ) pg .
right. Then the vector field σ (X) defined in Proposition 9.37 is complete, with flow
γ : R × M → M given by
γ (t, p) = p exp (t X) .
σ (X) p = γ p ′ (0) .
If G acts effectively on M, then σ is injective, and if G acts freely on M, then
σ (X) is nowhere zero when X 6= 0.
PROOF : From Proposition 9.37 we know that σ (X) is σ p -related to the left invariant
vector field Xe on G determined by X, and the one-parameter subgroup φX : R → G
defined in Proposition 8.29 is an integral curve for Xe with initial condition e . Hence
it follows from Proposition 3.48 that γ p = σ p ◦ φX is an integral curve for σ (X) with
initial condition σ p (e) = p , showing that γ is the flow for σ (X) .
Now suppose that G acts effectively on M and that σ (X) = 0 . Then γ (t, p) = p
for every p ∈ M and t ∈ R. Since G acts effectively, this implies that φX (t) = e for
every t ∈ R which shows that X = 0.
Finally, suppose that G acts freely on M and that σ (X) p = 0 for a point p ∈ M.
For each g ∈ G, we let R g : M → M be the diffeomorphism defined by R g (p) = pg
for p ∈ M. Since
γ p (s + t) = R exp (t X) ◦ γ p (s)
for s,t ∈ R, we have that
γ p ′ (t) = R exp (t X) ∗ ◦ γ p ′ (0) = 0
for t ∈ R. This implies that γ p (t) = p and therefore φX (t) = e for every t ∈ R, since
G acts freely on M, thus showing that X = 0.
left. For each p ∈ M, we let σ p : G → M be the smooth map defined by σ p (g) = g p
for g ∈ G. Then we have a Lie algebra anti-homomorphism σ : g → T 1 M defined
by
σ (X) p = σ p ∗ (X)
GROUP ACTIONS 321
for X ∈ g and p ∈ M, satisfying
(L g ) ∗ σ (X) = σ ( Ad (g−1 ) X )
for every X ∈ g and g ∈ G. The vector field σ (X) is complete, with flow γ : R× M →
M given by
γ (t, p) = exp (t X) p .
σ (X) p = γ p ′ (0) .
PROOF : e : M × G → M defined by
The map µ
e (p, g) = g−1 p
µ
e : g → T 1 M be
for p ∈ M and g ∈ G, is an action of G on M on the right. We let σ
the corresponding Lie algebra homomorphism defined in Proposition 9.37. Then we
have that
σ p = σep ◦ φ
for p ∈ M , where φ : G → G is the smooth map given by φ (g) = g−1 for g ∈ G. By
Corollary 8.33 it follows that
σ = −σe ,
and the proposition now follows from Propositions 9.37, 9.38 and 9.39.
9.41 Proposition Let µ : G × M → M and µ ′ : G × M ′ → M ′ be two operations

of a Lie group G on the smooth manifolds M and M ′ on the left, and let σ : g → T 1 M
and σ ′ : g → T 1 M ′ be the corresponding Lie algebra anti-homomorphisms defined in
Proposition 9.40. If f : M → M ′ is a smooth map which is equivariant with respect to
the actions µ and µ ′ , then the vector fields σ (X) and σ ′ (X) are f -related for every
X ∈ g.
f ◦ σ p (g) = f ◦ Lg (p) = L′g ◦ f (p) = σ ′f (p) (g)
for g ∈ G and p ∈ M , which implies that
f∗ ◦ σ (X)(p) = f∗ ◦ σ p ∗ (X) = ( f ◦ σ p ) ∗ (X) = σ ′f (p) ∗ (X) = σ ′ (X) ◦ f (p)
for p ∈ M and X ∈ g .
HAMILTONIAN SYSTEMS WITH SYMMETRY

9.42 Definition An action µ : G × M → M of a Lie group G on a symplectic
manifold (M, ω ) on the left is called symplectic if the map Lg is symplectic for every
g ∈ G.
9.43 Definition Let µ : G × M → M be a symplectic action of a Lie group G on

a symplectic manifold (M, ω ) on the left. Given a smooth map J : M → g∗ , we have
for each ξ ∈ g a smooth function J(ˆ ξ ) : M → R defined by J(
ˆ ξ )(p) = J(p)(ξ ) for
p ∈ M. The map J is called a momentum mapping for µ if
σ (ξ ) = XJ(ˆ ξ ) ,
ˆ ξ),
which means that σ (ξ ) is the Hamiltonian vector field with energy function J(
for every ξ ∈ g.
9.44 Proposition Let J be a momentum mapping for a symplectic action of a Lie

group G on a symplectic manifold (M, ω ) on the left, and let XH be a Hamiltonian
vector field on M with energy function H which is invariant under the action, i.e.,
H(p) = H(gp)
for every p ∈ M and g ∈ G. If c : I → M is an integral curve for XH , then the function
J ◦ c is constant on I.
PROOF : If γ : R × M → M is the flow for σ (ξ ) defined in Proposition 9.40 for a
given ξ ∈ g, we have that H ◦ γ p is constant for each p ∈ M . Hence it follows from
Proposition 7.76 that {H, J( ˆ ξ )} = 0 on M which implies that J( ˆ ξ ) ◦ c is constant on
I . Since this is true for every ξ ∈ g, it follows that the function J ◦ c is constant on
I.
9.45 Definition Let µ : G × M → M be a symplectic action of a Lie group G on

a symplectic manifold (M, ω ) on the left with a momentum mapping J . We say that
J is Ad ∗ -equivariant if it is equivariant with respect to µ and the action of G on g∗
determined by the coadjoint representation. This means that
J ◦ L g = Ad (g−1 ) ∗ ◦ J
for every g ∈ G .
9.46 Proposition Let J be an Ad ∗ -equivariant momentum mapping for a sym-

plectic action of a Lie group G on a symplectic manifold (M, ω ) on the left. Then we
have that
ˆ ξ ), J(
{J( ˆ η ) } = J(
ˆ [ ξ,η ])
for every ξ , η ∈ g .
GROUP ACTIONS 323
PROOF : Let p ∈ M and ξ , η ∈ g. Using that J is Ad ∗ -equivariant, we have that

ˆ ξ ) (exp (t η ) p) = J(exp (t η ) p) (ξ )
J(
= J(p) ◦ Ad (exp (−t η )) (ξ ) = α ◦ Ad ◦ exp (−t η )
for every t ∈ R , where α : End (g) → R is the linear map given by

α (F) = J(p) ◦ F(ξ )
for F ∈ End (g) . Taking the derivative with respect to t at 0, it follows from Proposi-
tions 7.76, 9.9 and 9.40 that
ˆ ξ ), J(
{J( ˆ η ) } (p) = − J(p) ◦ ad (η ) (ξ ) = J(
ˆ [ ξ , η ] ) (p) .
Since this is true for every p ∈ M , we obtain the formula in the proposition for
ξ,η ∈ g.
9.47 Proposition Let (M, ω ) be a symplectic manifold where ω = −d θ for a

1-form θ on M , and let G be a Lie group acting on M on the left so that L∗g (θ ) = θ
for every g ∈ G . Then the map J : M → g∗ given by
J(p)(ξ ) = θ (σ (ξ ))(p)
for p ∈ M and ξ ∈ g , is an Ad ∗ -equivariant momentum mapping for the action. We
have that
ˆ ξ ) = θ ( σ (ξ ))
J(
for every ξ ∈ g .
PROOF : Let ξ ∈ g , and consider the flow γ : R × M → M for σ (ξ ) defined in
Proposition 9.40. Since
γ t∗ θ = L∗exp (t ξ ) θ = θ
for every t ∈ R , it follows that the Lie derivative L σ (ξ ) θ = 0 . Using Proposition
5.83 we now have that
ˆ ξ ) = d(i σ (ξ ) θ ) = − i σ (ξ ) (d θ ) = i σ (ξ ) (ω )
d J(
which implies that
σ (ξ ) = XJ(ˆ ξ ) ,
thus showing that J is a momentum mapping for the action.
J is also Ad ∗ -equivariant, since
J ◦ L g (p)(ξ ) = θ (σ (ξ )) (L g (p)) = θ (L g (p)) ◦ (L g ) ∗ ◦ (L g ) −1

∗ ◦ σ (ξ ) ◦ L g (p)
= L∗g (θ ) {(L g ) ∗ σ (ξ ) } (p) = θ { σ ( Ad (g−1 ) ξ ) } (p) = J(p) ◦ Ad (g−1 ) ξ
= Ad (g−1 ) ∗ ◦ J(p)(ξ )
for every p ∈ M and ξ ∈ g .


ifold M, and let f : M → M be a diffeomorfism. Then we have a diffeomorphism
f ∗ : T ∗ M → T ∗ M defined by
( f ∗) p = ( f ∗ q) ∗
where q = f −1 (p) for p ∈ M, i.e.,
f ∗ (v) = v ◦ f ∗
for v ∈ T ∗ M , so that ( f ∗ , f −1 ) is a bundle map and we have a commutative diagram
∗
f✲
T ∗M T ∗M
π π
❄ ❄
f
M ✛ M
We also have that ( f ∗ ) ∗ θ = θ , where θ is the canonical 1-form on T ∗ M.

PROOF : The assertion that ( f ∗ , f −1 ) is a bundle map follows from Proposition 4.50
applied to the bundle map ( f −1
∗ ,f
−1 ) for the tangent bundle. We also have that
( f ∗ ) ∗ θ (v)(w) = θ ( f ∗ (v)) (( f ∗ ) ∗ (w)) = f ∗ (v) ◦ π∗ ◦ ( f ∗ ) ∗ (w)

= v ◦ ( f ◦ π ◦ f ∗ ) ∗ (w) = v ◦ π ∗ (w) = θ (v)(w)
for v ∈ T ∗ M and w ∈ Tv (T ∗ M) , which shows the last part of the proposition.
9.49 Proposition Let µ : G × M → M be an action of a Lie group G on a smooth

b : G × T ∗ M → T ∗ M defined by
manifold M on the left. Then we have an action µ
b (g, v) = Lg∗−1 (v)
µ
for g ∈ G and v ∈ T ∗ M , which leaves the canonical 1-form θ on T ∗ M invariant,

b satisfies (L̂g ) ∗ θ = θ for every g ∈ G . The Ad ∗ -
i.e., the left translation L̂g of µ
equivariant momentum mapping J : T ∗ M → g∗ for this action defined in Proposition
9.47 is given by
Ju = (σu ∗ e ) ∗
for u ∈ M, i.e.,
ˆ ξ )(v) = v( σ (ξ ) u )
J(
for ξ ∈ g , u ∈ M and v ∈ Tu∗ M , where σ is the Lie algebra anti-homomorphism
associated with the action µ as defined in Proposition 9.40.
GROUP ACTIONS 325
PROOF : Let σ b be the Lie algebra anti-homomorphism associated with the action µ b.
As the projection π ′ : T ∗ M → M is equivariant with respect to the actions µ b and µ ,
b (ξ ) and σ (ξ ) are π ′ -related. This implies that
it follows from Proposition 9.41 that σ
ˆ ξ )(v) = θ (v)( σb (ξ ) v ) = v ◦ π∗′ ◦ σb (ξ )(v) = v ◦ σ (ξ ) ◦ π ′ (v) = v( σ (ξ ) u )
J(
for u ∈ M and v ∈ Tu∗ M .
9.50 Example Let E be an affine space modelled on a finite dimensional vector

space V . For each point q ∈ E we let σq : V → E be the affine map defined by σq (v) =
q+v for v ∈ V , having idV : V → V as its linear part. If the tangent spaces T0V and Tq E
are canonically identified with V by means of the linear isomorphisms ω0 : T0V → V
and ωq : Tq E → V defined in Lemmas 2.84 and 9.36, we have that
ωq ◦ σq ∗ 0 ◦ ω0−1 = idV .
Now let J : T ∗ E → T0∗V be the Ad ∗ -equivariant momentum mapping for the action
of V on E defined in Proposition 9.49. Then we have that Jq = (σq ∗ 0 ) ∗ so that
(ω0∗ )−1 ◦ Jq ◦ ωq∗ = idV ∗ .
Identifying the cotangent spaces Tq∗ E and T0∗V with V ∗ by means of the linear iso-
morphisms ω eq = (ωq∗ )−1 : Tq∗ E → V ∗ and (ω0∗ )−1 : T0∗V → V ∗ , it follows that
J(q, p) = p
for (q, p) ∈ E × V ∗ . This momentum mapping J : E × V ∗ → V ∗ is called the linear

momentum .
9.51 Example Let G be a Lie subgroup of Gl (n, R) , and consider the action
µ : G × Rn → Rn given by
µ (B, q) = Bq
for B ∈ G and q ∈ Rn written as a column vector. For each vector q ∈ Rn we let
σq = ρq |G , where ρq : gl (n, R) → Rn is the linear map given by ρq (B) = Bq for
B ∈ gl (n, R). Now it follows from Lemma 2.84 that σq ∗ e = ρq |g when we canon-
ically identify Tq Rn with Rn and Te gl (n, R) with gl (n, R) as in Example 8.10 (a).
Hence the Ad ∗ -equivariant momentum mapping J : Rn × (Rn ) ∗ → g∗ for the action
µ defined in Proposition 9.49 is given by
J(q, p)(B) = p(Bq)
for (q, p) ∈ Rn × (Rn ) ∗ and B ∈ g.
9.52 Example Let J3 : R3 × (R3 ) ∗ → so(3)∗ be the Ad ∗ -equivariant momen-

tum mapping for the action µ : SO(3) × R3 → R3 of SO(3) on the 3-dimensional
Euclidean space R3 as described in Example 9.51. Using the Euclidean metric g 30
which is the dot product in R3 as defined in Definition 7.15, we may identify (R3 ) ∗
with R3 by means of the linear isomorphism e : R3 → (R3 ) ∗ given by
e(u)(v) = u · v
for u, v ∈ R3 . By Remark 9.18 we may also identify the Lie algebra so(3) with R3
by means of the Lie algebra isomorphism λ : R3 → so(3) given by
λ (u) v = u × v
for column vectors u, v ∈ R3 . With these identifications we obtain a momentum map-

ping J = e−1 ◦ λ ∗ ◦ J3 ◦ (idR3 × e) : R3 × R3 → R3 given by
e ◦ J(q, p)(u) = J3 (q, e(p))(λ (u)) = e(p)(λ (u) q)

= p · (u × q) = (q × p) · u = e(q × p)(u)
for q, p, u ∈ R3 so that
J(q, p) = q × p
for (q, p) ∈ R3 × R3 . This momentum mapping is called the angular momentum.
LAGRANGIAN SYSTEMS WITH SYMMETRY

9.53 Noether’s theorem Let µ : G × M → M be an action of a Lie group G on
a smooth manifold M on the left, and let (µ e , µ ) be the corresponding action of G on
the tangent bundle π : T M → M defined in Proposition 9.4. Let XE be a Lagrangian
vector field for a regular Lagrangian L : T M → R which is invariant under the action,
i.e.,
L(v) = L(gv)
for every v ∈ T M and g ∈ G. Then the action µ e leaves the Lagrangian 1-form θL on
e satisfies (L̃g ) ∗ θL = θL for every g ∈ G .
T M invariant, i.e., the left translation L̃g of µ
The Ad ∗ -equivariant momentum mapping J : T M → g∗ for this action defined in
Proposition 9.47 is given by
Ju = (σu ∗ e ) ∗ ◦ F(L) u
for u ∈ M, i.e.,
ˆ ξ )(v) = F(L)(v)( σ (ξ ) u )
J(
for ξ ∈ g , u ∈ M and v ∈ Tu M , where σ is the Lie algebra anti-homomorphism
associated with the action µ as defined in Proposition 9.40. If c : I → T M is an
integral curve for XE , then the function J ◦ c is constant on I.
GROUP ACTIONS 327
PROOF : b , µ ) be the action of G on the cotangent bundle π ′ : T ∗ M → M defined

Let (µ
in Proposition 9.49. We first show that the fibre derivative F(L) : T M → T ∗ M is
equivariant with respect to the actions µ e and µb of G on T M and T ∗ M, i.e., that
L̂g ◦ F(L) = F(L) ◦ L̃g

for g ∈ G. By taking the derivative at v ∈ Tu M on both sides of
Lu ◦ L̃g = Lu ,
identifying Tgv (Tu M) and Tg (Tu M) with Tu M and TL(v) R with R in the usual way and
using that L̃g is linear on Tu M, we obtain
F(L)(gv) ◦ L̃g = F(L)(v) (1)
which implies that
L̂g ◦ F(L)(v) = F(L)(v) ◦ L̃g−1 = F(L) ◦ L̃g (v)
for v ∈ T M and g ∈ G .
Using this, we now obtain that
(L̃g ) ∗ θL = (L̃g ) ∗ F(L) ∗ θ = (F(L) ◦ L̃g ) ∗ θ

= (L̂g ◦ F(L)) ∗ θ = F(L) ∗ (L̂g ) ∗ θ = F(L) ∗ θ = θL
for every g ∈ G . Since σe (ξ ) and σ (ξ ) are π -related, we also have that

ˆ ξ )(v) = θL (v)( σe (ξ ) v ) = θ (F(L)(v)) ( F(L) ∗ ◦ σe (ξ )(v))
J(
= F(L)(v) ◦ π∗′ ◦ F(L) ∗ ◦ σe (ξ )(v) = F(L)(v) ◦ π∗ ◦ σe (ξ )(v)
= F(L)(v) ◦ σ (ξ ) ◦ π (v) = F(L)(v)( σ (ξ ) u )
for u ∈ M and v ∈ Tu M . From (1) we see that the action A : T M → R and hence also
the energy E = A − L is invariant under the operation µe , since
A(v) = F(L)(v)(v) = F(L)(gv)(gv) = A(gv)

for every v ∈ T M and g ∈ G. The last part of the theorem therefore follows from
Proposition 9.44.
9.54 Example Consider a conservative systen consisting of a single particle of

mass m moving in R3 , with potential energy function V and Lagrangian L given by
L(q, v) = 1
2 mkvk 2 −V (q)
for (q, v) ∈ R3 × R3 as described in Example 7.102. Let µ : SO(3) × R3 → R3 be the

action of SO(3) on R3 defined by
µ (B, q) = Bq
for B ∈ SO(3) and q ∈ R3 written as a column vector. Using the same notation as in
Example 9.52, the Ad ∗ -equivariant momentum mapping for this action described in
Theorem 9.53 is J = e−1 ◦ λ ∗ ◦ J3 ◦ F(L) : R3 × R3 → R3 , which is given by
J(q, v) = q × mv
for (q, v) ∈ R3 × R3 . This momentum mapping is called the angular momentum of

the system. If the potential energy function V is invariant under the action µ , i.e.,
V (q) = V (Bq)
for every q ∈ R3 and B ∈ SO(3), then J is a constant of the motion by Noether’s

theorem.
GRAVITATIONAL CENTRAL FIELDS

9.55 Consider a conservative system consisting of a single particle of mass m
moving in R3 as described in Example 9.54, where the potential energy of the particle
only depends on its distance from the center of force at the origin. Then the angular
momentum J is a constant of the motion, and the particle is therefore moving in a
plane M through the origin having J as its normal vector. Using cylinder coordinates
in R3 with the z-axis along J, so that (r, φ ) are polar coordinates on M in a coordinate
neighbourhood N, the Lagrangian of the system is given by
L(v) = 12 m(ṙ2 + r2 φ̇ 2 ) − V (r)
for tangent vectors v ∈ T N with coordinates (r, φ , ṙ, φ˙) as described in Example 7.20
and Proposition 7.81. We assume that V is the potential energy of a gravitational
force from a mass M at the origin, so that
V (r) = −γ Mm
r
where γ is Newton’s gravitational constant.
Since φ is a cyclic coordinate, its conjugete momentum pφ ◦ F(L) = mr2 φ̇ is
a constant of the motion, which is equal to the length l of the angular momentum
J. The motion of the particle in N is given by an integral curve γ : I → T N for the
Lagrangian vector field XE . From the Lagrange’s equations it follows that
(r ◦ γ ) ′ (t) = ṙ ◦ γ (t) and (φ ◦ γ ) ′ (t) = φ̇ ◦ γ (t)
so that
m {r ◦ γ (t)} 2 (φ ◦ γ ) ′ (t) = l (1)
for t ∈ I. Assuming that J 6= 0, we see that the function φ ◦ γ is a strictly increasing
GROUP ACTIONS 329
diffeomorphism onto an open interval I ′ . The orbit of the particle is described by the
function ρ = (r ◦ γ ) ◦ (φ ◦ γ )−1 : I ′ → R . Since
r ◦ γ (t) = ρ (s) where s = φ ◦ γ (t), (2)

it follows from (1) that
ρ ′ (s)
(r ◦ γ ) ′ (t) = ρ ′ (s) (φ ◦ γ ) ′ (t) = l 2
= − l u ′ (s) (3)
m ρ (s) m
where we have introduced the function u = 1 : I ′ → R .

ρ
By Proposition 7.87 the total energy E is a constant of the motion so that
E = 12 m(r ◦ γ ) ′ (t)2 + U(r ◦ γ (t)) (4)
for t ∈ I, where
l2
U(r) = − γ Mm
2m r2 r
is called the effective potential energy . It has an absolute minimum at the point
(r0 , E0 ) where
l2 γ Mm
r0 = and E0 = − . (5)
γ Mm2 2r0
Combining (2), (3) and (4), we now obtain the equation
2 2
E = l u ′ (s) 2 + l u(s) 2 − γ Mm u(s)
2m 2m
which can be written in the form
(r0 u′ (s))2 + (r0 u(s) − 1)2 = 1 − E

E0
for s ∈ I ′ . Its solution is given by
r0 u(s) − 1 = ε cos(s − s0 )
so that
r0
ρ (s) =
1+ ε cos(s−s0 )
1/2
for s ∈ I ′ , where s0 is an integration constant and ε = 1− E . Hence the particle
E0
is moving along a conic section with one focus at the origin and with eccentricity ε .
The cases E = E0 , E0 < E < 0, E = 0 and E > 0 correspond to a circle, an ellipse, a
parabola and a hyperbola with eccentricities ε satisfying ε = 0, 0 < ε < 1, ε = 1 and
ε > 1, respectively.
By a rotation of the coordinate system we may assume that s0 = 0, so that the
polar equation of the conic section is
r0
r= ,
1+ ε cos φ
which can be also written in the form
r = r0 − ε r cos φ . (6)
Introducing cartesian coordinates x = r cos φ and y = r sin φ in the plane M, we see

that when ε > 0, the eccentricity ε is the ratio of the distances of the particle from
r
the focus at the origin and the line x = 0 , which is called the directrice .
ε
In the case 0 ≤ ε < 1 we obtain Kepler’s first law , saying that the planets are
moving in elliptical orbits with the sun at one focus. The other focus is the point
(−2c, 0), where c is called the linear eccentricity . We consider a circle as a special
case of an ellipse where the two foci coincide. The minimal and maximal values of r
are
r r
r1 = 0 and r2 = 0 .
1−ε 1+ε
Hence the semimajor axis a and the linear eccentricity c are given by
r0 r E
a = 12 (r1 + r2 ) = = 0 0 and c = 21 (r2 − r1 ) = aε , (7)
1−ε 2 E
and the semiminor axis is

p p p
b= a2 − c2 = a 1 − ε 2 = ar0 . (8)
Equation (6) can be written in cartesian form as

p
a x2 + y2 = b2 − cx (9)
so that
p 2
2a − x2 + y2 = 4a2 − 4(b2 − cx) + (x2 + y2) = (x + 2c)2 + y2 ,
showing that the sum of the distances of the point (x, y) from the two foci is 2a, in
agreement with the usual definition of an ellipse. By Equation (9) we also have that
a2 (x2 + y2 ) = b2 (a2 − c2) − 2 b2cx + (a2 − b2) x2 = a2 (x2 + b2) − b2(x + c)2
so that
(x+c)2 y2
2
+ 2 =1,
a b
which is the cartesian equation of the ellipse. From (5), (7) and (8) we finally obtain
γ Mm γ Mm2 b2
E =− and l 2 = .
2a a
We see that the energy E does not depend on the semiminor axis b. Furthermore, the
length l of the angular momentum for a given energy E is largest for a circular orbit.
Chapter 10
FIBRE BUNDLES
INTRODUCTION
10.1 Definition Let M and F be smooth manifolds, and let G be a Lie group which
acts on F effectively on the left. A smooth manifold E is called a fibre bundle over
M with fibre F and structure group G if the following three conditions are satisfied :
(i) There is a surjective smooth map π : E → M which is called the projection
of the total space E onto the base space M. For each point p ∈ M the inverse
image π −1 (p) is called the fibre over p.
(ii) We have a family of G-related local trivializations D = {(tα , π −1 (Uα ))|α ∈A},
in the sense defined below, such that {Uα |α ∈ A} is an open cover of M. By a
local trivialization we mean a pair (t, π −1 (U)), where U is an open subset of
M and t : π −1 (U) → U × F is a diffeomorphism such that the diagram
t ✲
π −1 (U) U ×F
❅
πU ❅ pr1
❘
❅ ✠
U
is commutative. Two local trivializations (t, π −1 (U)) and (s, π −1 (V )) are said
to be G-related if there is a smooth map φ : U ∩ V → G, called the transition
map , so that
s ◦ t −1 (p, v) = (p, φ (p) v)
for every p ∈ U ∩V and v ∈ F.
(iii) The family D is maximal in the sense that if (t, π −1 (U)) is a local trivialization
which is G-related to every local trivialization in D, then (t, π −1 (U)) ∈ D.
331
bundle in order to indicate more of the bundle structure in the notation. F is also
called the typical fibre to distinguish it from the fibre π −1(p) over p.
A family D satisfying (ii) and (iii) is called a fibre bundle structure on E, whereas a
family of local trivializations satisfying only (ii) is called a trivializing cover of E.
10.2 Proposition If A is a trivializing cover of a fibre bundle π : E → M, then

there is a unique fibre bundle structure D on E containing A .
PROOF : Let D be the set of all local trivializations in E which are G-related to
every local trivialization in A = {(tα , π −1 (Uα ))|α ∈ A}. In order to prove that D
is a fibre bundle structure, we must show that any pair of local local trivializations
(t, π −1 (U)) and (s, π −1 (V )) in D are G-related. For each α ∈ A we have smooth
maps φα : V ∩Uα → G and ψα : U ∩Uα → G so that
s ◦ tα−1 (p, v) = (p, φα (p) v) and tα ◦ t −1 (q, v) = (q, ψα (q) v)
for every p ∈ V ∩Uα , q ∈ U ∩Uα and v ∈ F. From this it follows that
s ◦ t −1 (p, v) = (p, φα (p) ψα (p) v)
for p ∈ U ∩V ∩Uα and v ∈ F. In particular, we have that
φα (p) ψα (p) = φβ (p) ψβ (p)
whenever p ∈ U ∩V ∩Uα ∩Uβ . This completes the proof that the local trivializations
(t, π −1 (U)) and (s, π −1 (V )) are G-related with transition map φ : U ∩V → G given
by
φ (p) = φα (p) ψα (p)
for p ∈ U ∩ V ∩ Uα , thus showing that D is the unique fibre bundle structure on E
containing A .
10.3 Proposition The projection π is a submersion, and the fibre E p = π −1 (p)

over p is a closed submanifold of E for each point p ∈ M. If (t, π −1 (U)) is a local
trivialization around p, then the map t p : E p → F defined by t p = pr2 ◦ t|E p is a
diffeomorphism. Hence each fibre E p is diffeomorphic to the typical fibre F.
If i p : E p → E is the inclusion map and u ∈ E p , then Vu = i p ∗ (Tu E p ) is a sub-
space of Tu E which is called the vertical subspace at u. The vectors in Vu are called
vertical tangent vectors at u. We have that Vu = ker π∗ u .
PROOF : From the diagram in Definition 10.1 (ii) we see that π is a submersion.
Hence E p is a closed submanifold of E by Proposition 2.41. In the same way we also
have that {p} × F = pr1 −1 (p) is a closed submanifold of U × F.
By Corollary 2.38, the map s p : E p → {p} × F induced by t is therefore a diffeo-
morphism, and so is the map pr2 : {p} × F → F, thus showing that t p = pr2 ◦ s p is a
diffeomorphism.
FIBRE BUNDLES 333
Since π ◦ i p (u) = p for every u ∈ E p , we have that π∗ u ◦ i p ∗ u = (π ◦ i p )∗ u = 0

which implies that Vu ⊂ ker π∗ u . Equality now follows since
dim u (E p ) = dim t p (u) (F) = dim u (E) − dim p (M) = dim u (ker π∗ u ) .
10.4 Remark Let (t, π −1 (U)) and (s, π −1 (V )) be two G-related local trivializa-
tions in the fibre bundle π : E → M with transition map φ : U ∩V → G. Then we have
that
s p ◦ t p−1 (v) = φ (p) v
for every p ∈ U ∩ V and v ∈ F, i.e., the diffeomorphism s p ◦ t p−1 : F → F coincides
with the operation of the group element φ (p) ∈ G on the fibre F for every p ∈ U ∩V .
Since G acts effectively on F, we know that φ (p) is uniquely determined by s p ◦ t p−1 .
Moreover, we have that
s p (u) = φ (p)t p (u)
when p ∈ U ∩ V and u ∈ E p . Indeed, the last formula it obtained from the first by
letting v = t p (u).
10.5 Proposition Let M and F be smooth manifolds, and let G be a Lie group
which acts on F effectively on the left. Let π : E → M be a map from a set E, and let
{Uα |α ∈ A} be an open cover of M. Suppose that for each α ∈ A, there is a bijection
tα : π −1 (Uα ) → Uα × F such that the diagram
tα ✲
π −1 (Uα ) Uα × F
❅
π Uα ❅ pr1
❘
❅ ✠
Uα
is commutative. Moreover, suppose that for each pair α , β ∈A, there is a smooth map
φ β α : Uα ∩Uβ → G so that
tβ ◦ tα−1 (p, v) = (p, φ β α (p) v)
for every p ∈ Uα ∩Uβ and v ∈ F.

Then there is a unique topology, smooth structure and fibre bundle structure D
on E such that π : E → M is a fibre bundle over M with fibre F and structure group
G, and such that (tα , π −1 (Uα )) is a local trivialization for each α ∈ A.
PROOF : We first define a topology τ on E so that the projection π is continuous and
the bijections tα : π −1 (Uα ) → Uα × F are homeomorphisms for every α ∈ A. The
continuity of π implies that each π −1 (Uα ) must be open in τ , and hence a set O in
E must belong to τ if and only if O ∩ π −1 (Uα ) ∈ τ for every α ∈ A. The requirement
of sets O in E such that tα (O ∩ π −1 (Uα )) is open in Uα × F for every α ∈ A. This

requirements.
that tβ (π −1 (Uα ) ∩ π −1 (Uβ )) = (Uα ∩ Uβ ) × F which is open in Uβ × F, thereby
belongs to τ , we have that tα (O ∩ π −1 (Uα )) is open in Uα × F, which proves that
tα is an open map. To show that it is continuous, let O be an open set in Uα × F. For
every β ∈ A we have that tβ (tα−1 (O) ∩ π −1 (Uβ )) = (tβ ◦ tα−1 ) (O ∩ (Uα ∩ Uβ ) × F)
which is open in Uβ × F, and hence tα−1 (O) ∈ τ showing that tα is continuous.
is the composition of the continuous maps tα and pr1 in the above commutative
diagram for each α ∈ A.
Now let {(y γ ,Wγ )|γ ∈C} be an atlas on F, and let {(xβ ,Vβ )|β ∈ B} be an atlas
on M such that {Vβ |β ∈ B} is a refinement of the open cover {Uα |α ∈ A}. We then
have a map ρ : B → A such that Vβ ⊂ Uρ (β ) for every β ∈ B. By the assumptions
in the proposition we now see that the family B = {((xβ × y γ ) ◦ tρ (β ) , tρ−1
(β )
(Vβ ×
Wγ ))|(β , γ ) ∈ B ×C } is an atlas on E. To show that E is a smooth manifold, we must
show that it is Hausdorff and second countable. By Proposition 2.6 we know that M
and F have atlases of the above form where B and C are countable, thus showing that
E is second countable.
To see that E is Hausdorff, let p1 and p2 be two different points in E. If p1 and
p2 lie in different fibres, they can be separated by the neighbourhoods π −1 (W1 ) and
π −1 (W2 ), where W1 and W2 are disjoint neighbourhoods of π (p1 ) and π (p2 ) in M.
If p1 and p2 lie in the same fibre π −1 (q) with q ∈Uα , they can be separated by the
of tα (p1 ) and tα (p2 ) in Uα × F.
the bijections tα : π −1 (Uα ) → Uα × F are diffeomorphisms for every α ∈ A. We see
maps tα and pr1 .
By the assumptions in the proposition the family A = {(tα , π −1 (Uα ))|α ∈ A}
is a trivializing cover of E which is contained in a unique fibre bundle structure D
according to Proposition 10.2, and this completes the construction of the fibre bundle
π : E → M having (tα , π −1 (Uα )) as a local trivialization for each α ∈ A.
10.6 Definition Let π1 : E1 → M1 and π2 : E2 → M2 be two fibre bundles with

the same fibre F and structure group G. A bundle map from π1 to π2 is a pair ( f˜, f )
of smooth maps f˜ : E1 → E2 and f : M1 → M2 such that the diagram
FIBRE BUNDLES 335
f˜ ✲
E1 E2
π1 π2
❄ ❄
f✲
M1 M2
is commutative, and for every pair of local trivializations (t, π −1 (U)) and (s, π −1 (V ))
in E1 and E2 , respectively, there is a smooth map φ : U ∩ f −1 (V ) → G so that
s ◦ f˜ ◦ t −1 (p, v) = ( f (p), φ (p) v) (1)
for every p ∈ U ∩ f −1 (V ) and v ∈ F. We let f˜p denote the induced smooth map
f˜p : π1−1 (p) → π2−1 ( f (p)) between the fibres for each p ∈ M1 .
If f˜ and f are diffeomorphisms, ( f˜, f ) is called an equivalence . In that case it
follows from (1) that
t ◦ f˜ −1 ◦ s −1 (q, v) = ( f −1 (q), {φ ◦ f −1 (q)} −1 v)
for every q ∈ V ∩ f (U) and v ∈ F, so that ( f˜ −1 , f −1 ) is also a bundle map.
10.7 Definition Let π1 : E1 → M and π2 : E2 → M be two fibre bundles over the

same base space M with the same fibre F and structure group G. A bundle map from
π1 to π2 over M is a smooth map f : E1 → E2 such that ( f , idM ) is a bundle map. If
f is a diffeomorphism, it is called an equivalence over M.
10.8 Definition Let M be a smooth manifold and G be a Lie Group. By a system

of transition maps corresponding to an open cover {Uα |α ∈ A} of M with values in
G we mean a family {φ β α |α , β ∈ A} of smooth maps φ β α : Uα ∩Uβ → G satisfying
the consistency condition
φ γ β (p) φ β α (p) = φ γ α (p) (1)
for every triple of indices α , β , γ ∈ A and every p ∈ Uα ∩ Uβ ∩ Uγ .
10.9 Remark If we put α = β = γ in (1), we obtain
φ α α (p) = e (2)
for every α ∈ A and every p ∈ Uα . Now putting α = γ in (1) and using (2), we also
have that
φ α β (p) = φ β α (p) −1 (3)
for every pair of indices α , β ∈ A and every p ∈ Uα ∩ Uβ .
10.10 Proposition Let M and F be smooth manifolds, and let G be a Lie group
which acts on F effectively on the left. Given a system {φ β α |α , β ∈ A} of transition
maps corresponding to an open cover {Uα |α ∈ A} of M with values in G, there is a
fibre bundle π : E → M with fibre F, structure group G and trivializing cover A =
{(tα , π −1 (Uα ))|α ∈ A} , where each pair of local trivializations (tα , π −1 (Uα )) and
(tβ , π −1 (Uβ )) in A are G-related with transition map φ β α : Uα ∩Uβ → G.
PROOF : We first define the set
[
X= Uα × F × {α } ,
α ∈A
and we say that two elements (p, v, α ) and (q, w, β ) in X are equivalent, and write
(p, v, α ) ∼ (q, w, β ) , if
p = q and φ β α (p) v = w .
It follows from (1) in Definition 10.8 and (2) and (3) in Remark 10.9 that ∼ is an
equivalence relation, and we denote the equivalence class of (p, v, α ) by [ p, v, α ] and
let E = X/ ∼ be the quotient set. We now have a map π : E → M defined by
π ([ p, v, α ]) = p ,
and for each α ∈ A we have a bijection tα : π −1 (Uα ) → Uα × F defined by
tα ([ p, v, α ]) = (p, v)
such that the diagram in Proposition 10.5 is commutative. Moreover, we have that
tβ ◦ tα−1 (p, v) = tβ ([ p, v, α ]) = tβ ([ p, φ β α (p) v, β ]) = (p, φ β α (p) v)
for every p ∈ Uα ∩Uβ and v ∈ F, and the proposition therefore follows from Propo-
sition 10.5.
10.11 Definition Let π : E → M be a fiber bundle, and let V be an open subset

of M and i : V → M be the inclusion mapping. Then a smooth map s : V → E with
π ◦ s = i is called a section of π on V , or a lifting of i to E.
pr1 ◦ t ◦ s = idU . Hence s|U is completely determined by the map h : U → F given by
(t ◦ s)(p) = (p, h(p)) for p ∈ U. h is called the local representation of s on U. We
INDUCED BUNDLES
10.12 Let π : E → M be a fibre bundle over M with fibre F and structure group
G. If f : N → M is a smooth map from a smooth manifold N, we define the set
Ee = {(p, u) ∈ N × E| f (p) = π (u)}
FIBRE BUNDLES 337
and let πe : Ee → N and fe : Ee → E be the restrictions to Ee of the projections pr1 :

N × E → N and pr2 : N × E → E on the first and second factor. Then we have a
commutative diagram
fe✲
Ee E
πe π
❄ ❄
f✲
N M
since f ◦ πe (p, u) = f (p) = π (u) = π ◦ fe(p, u) for (p, u) ∈ E.

e
Now let A = {(tα , π −1 (Uα ))|α ∈ A} be a trivializing cover of E. For each α ∈ A
we have an open subset U eα = f −1 (Uα ) of N and a bijection e eα ) → U
tα : πe−1 (U eα × F
defined by
etα (p, u) = (p,tα , f (p) (u)) (1)
for p ∈ U eα and u ∈ E f (p) . Then {U

eα |α ∈ A} is a open cover of N, and if φ β α :
Uα ∩ Uβ → G is the transition map between the local trivializations (tα , π −1 (Uα ))
and (tβ , π −1 (Uβ )), we have that
e tα−1
tβ ,p ◦ e −1
,p (v) = tβ , f (p) ◦ tα , f (p) (v) = φ β α ◦ f (p) v (2)
so that
tα−1 (p, v) = (p, φ β α ◦ f (p) v)
tβ ◦ e
e
for every p ∈ U eα ∩ Ueβ and v ∈ F. By Proposition 10.5 there is therefore a unique

topology, smooth structure and fibre bundle structure on Ee such that πe : Ee → N is a
fibre bundle over N with fibre F and structure group G, and such that (e eα )) is
tα , πe−1 (U
a local trivialization for each α ∈A. It is called the local trivialization in Ee associated
with the local trivialization (tα , π −1 (Uα )) in E.
By (1) we see that
etα ,p = tα , f (p) ◦ fep
eα , so that
for p ∈ U
tα ◦ fe ◦ e
tα−1 (p, v) = ( f (p), v) (3)
eα and v ∈ F. Hence it follows from Proposition 10.3 that ( fe, f ) is a bundle
for p ∈ U
map which induces a diffeomorphism between the fibres Eep and E f (p) for each p ∈ N.
Since πe and fe are the components of the inclusion map i : Ee → N × E, we also
have that
tα−1 (p, v) = (p, f (p), v)
(id × tα ) ◦ i ◦ e
eα and v ∈ F. Using this, we will show that Ee is a closed submanifold of
for p ∈ U
eα ) and tα (u0 ) = ( f (p0 ), v0 ) ∈ Uα × F, we choose local
N × E. If (p0 , u0 ) ∈ πe−1 (U
charts (e e e ⊂U
x, V ), (x,V ) and (y,W ) around p0 , f (p0 ) and v0 , respectively, with V eα
e
and f (V ) ⊂ V . Then we have that
(e
x × (x × y) ◦ tα )) ◦ i ◦ ((e tα ) −1 (a, ξ ) = (a, x ◦ f ◦ xe−1 (a), ξ )
x × y) ◦ e
for a ∈ xe(Ve ) and ξ ∈ y(W ).
Now let r, n and m be the dimensions of N, M and F, respectively, and let φ :
xe(Ve ) × Rn × Rm → xe(Ve ) × Rm × Rn be the C∞ -map defined by φ (a, b, ξ ) = (a, ξ , b −
x ◦ f ◦ xe−1 (a)). φ is a diffeomorphism with inverse ψ : xe(Ve ) × Rm × Rn → xe(Ve ) ×
Rn × Rm given by ψ (a, ξ , b) = (a, b + x ◦ f ◦ xe−1 (a), ξ ).
If j : xe(Ve ) × Rn × Rm → Rr × Rn × Rm is the inclusion map, we therefore have
a local chart (Ve × tα−1 (V × W ), j ◦ φ ◦ (e x × (x × y) ◦ tα )) around (p0 , u0 ) in N × E
having the submanifold property, showing that Ee is a submanifold of N × E. It is
closed as Ee = ( f × π )−1 (∆), where ∆ ⊂ M × M is the diagonal which is closed since
M is Hausdorff.
The fibre bundle πe : Ee → N is called the pull-back of π : E → M by f or the fibre
bundle induced from π : E → M by f , and it is denoted by f ∗ (E) with projection
f ∗ (π ). The bundle map ( fe, f ) is called the canonical bundle map of the induced
bundle.
A smooth map F : N → E with π ◦ F = f is called a lifting of f or a section
of π along f . The set of all liftings of f is denoted by Γ( f ; E) . For each lifting F
of f there is a unique section Fe : N → Ee of πe with fe◦ Fe = F , which is given by
e
F(p) = (p, F(p)) for p ∈ N. Conversely, if s : N → Ee is a section of πe , then s = Fe
where F = fe◦ s . Fe is called the section in the induced fibre bundle determined by
the lifting F.
If π : E → M is a vector bundle, so is the induced bundle πe : Ee → N. Indeed,
it follows from (2) that e tα−1
tβ ,p ◦ e e
,p is a linear isomorphism whenever p ∈ Uα ∩ Uβ .
e
Hence there is a well-defined vector space structure on each fibre Eep so that e tα ,p is
a linear isomorphism for any α ∈ A with p ∈ U eα . From (3) it follows that ( fe, f ) is a
bundle map which induces a linear isomorphism between the fibres Eep and E f (p) for
each p ∈ N. The vector space structure on Eep is therefore given by
a1 (p, v1 ) + a2 (p, v2 ) = (p, a1 v1 + a2 v2 )
for v1 , v2 ∈ E f (p) and a1 , a2 ∈ R.
If N is an immersed submanifold of M with inclusion map i : N → M, we consider
the subset E|N = π −1(N) of E, and let πN : E|N → N be the map induced by the
projection π : E → M. We give E|N the fibre bundle structure so that the bijection
e : Ee → E|N given by e(p, u) = u for (p, u) ∈ E, e is an equivalence over N. The fibre
bundle πN : E|N → N is called the restriction of E to N, and i ′ = ei ◦ e−1 : E|N → E is
the inclusion map. Given a trivializing cover A = {(tα , π −1 (Uα ))|α ∈ A} of E, we
let Ueα = Uα ∩ N and tα′ : π −1 (U eα ) → U
eα × F be the map induced by tα : π −1 (Uα ) →
Uα × F so that e eα ))|α ∈ A} is a
tα ◦ e−1 = tα′ for α ∈ A . Then A ′ = {(tα′ , π −1 (U
trivializing cover of E|N .
FIBRE BUNDLES 339
10.13 Proposition If π : E → M is a vector bundle and f : N → M is a smooth

map from a smooth manifold N, then the set Γ( f ; E) of liftings of f is a module over
the ring F (N) of C∞ -functions on N with operations defined by
(F1 + F2)(p) = F1 (p) + F2(p) and (gF)(p) = g(p) F(p)
for liftings F1 , F2 and F and C∞ -functions g on N.

PROOF : Follows from Proposition 2.53 by considering the sections Fe1 , Fe2 and Fe in
the induced vector bundle πe : Ee → N determined by the liftings F1 , F2 and F, since
F1 + F2 = fe◦ (Fe1 + Fe2 ) and gF = fe◦ (gF)

e
where ( fe, f ) is the canonical bundle map of πe : Ee → N.
10.14 Proposition Let π : E → M be an n-dimensional vector bundle, and let

F1 , ..., Fn be n liftings of a smooth map f : N → M from a smooth manifold N which
are everywhere linearly independent, i.e., F1 (p), ..., Fn (p) ∈ π −1 ( f (p)) are linearly
independent for every p ∈ N. Then a map F : N → E with π ◦ F = f is a lifting of f
if and only if the map a : N → Rn defined by
n
F(p) = ∑ ai (p) Fi (p)
i=1
for p ∈ N is smooth.
PROOF : Let F ei be the section in the induced vector bundle πe : Ee → N determined by

the lifting Fi for i = 1, ... , n, and let Fe : N → Ee be the map given by F(p)
e = (p, F(p))
for p ∈ N. Then the proposition follows from Corollary 2.59 since
n
e
F(p) = ∑ ai (p) Fei (p)
i=1
for p ∈ N.
PRINCIPAL FIBRE BUNDLES

10.15 Definition Let M be a smooth manifold and G be a Lie group. Then a
principal fibre bundle P over M with structure group G is a fibre bundle π : P → M
where the fibre F coincides with the structure group G which is acting on F by
left translation. The principal bundle P is also denoted by P (M, G) and is called a
principal G-bundle over M.
10.16 If π : P → M is a principal G-bundle, we may define an operation of G

on P on the right in the following way. Let u ∈ Pp and g ∈ G, and choose a local
trivialization (t, π −1 (U)) in P with p ∈ U. If t(u) = (p, h), we define ug ∈ Pp so that
t(ug) = (p, hg), i.e.,
π (ug) = π (u) and t p (ug) = t p (u) g when p = π (u) ∈ U .
We must show that the operation of G does not depend on the local trivialization
(t, π −1 (U)). So let (s, π −1 (V )) be another local trivialization with p ∈ V which is
G-related to (t, π −1 (U)) with transition map φ : U ∩V → G. Then we have that
s p (ug) = φ (p) t p (ug) = φ (p) [t p (u) g ] = [ φ (p)t p (u) ] g = s p (u) g
since Lφ (p) commutes with Rg , showing that we have a well-defined map ν : P× G →

P with ν (u, g) = ug for u ∈ P and g ∈ G .
Now we have that
u (g1 g2 ) = (u g1 ) g2 and ue = u
for every u ∈ P and g1 , g2 ∈ G, since
t p (u (g1 g2 )) = t p (u) (g1 g2 ) = (t p (u) g1 ) g2 = t p (u g1 ) g2 = t p ((u g1 ) g2 )
and
t p (ue) = t p (u) e = t p (u)
when u ∈ Pp . As the map
t ◦ ν ◦ (t −1 × id ) : U × G × G → U × G
is given by
t ◦ ν ◦ (t −1 × id ) (p, h, g) = (p, hg)
for p ∈ U and h, g ∈ G , it follows that ν is smooth and hence is an operation of G
on P on the right.
We see that G acts freely on P, and that the orbits of G are the fibres of P. Indeed,
if ug = u for a u ∈ Pp , then t p (u) g = t p (u) which implies that g = e. Moreover,
given u1 , u2 ∈ Pp there is a g ∈ G with t p (u1 ) g = t p (u2 ) so that u1 g = u2 .
Using these results, we can now give an alternative definition of a principal fibre
bundle.
10.17 Definition Let M be a smooth manifold and G be a Lie group. A smooth

manifold P is called a principal fibre bundle over M with structure group G if the
following three conditions are satisfied :
(i) G acts freely on P on the right.
FIBRE BUNDLES 341
(ii) There is a surjective smooth map π : P → M which is called the projection of

the total space P onto the base space M satisfying
π (ug) = π (u) for every u ∈ P and g ∈ G .
For each point p ∈ M the inverse image π −1 (p) is called the fibre over p.
(iii) For each point p ∈ M there is an open neighbourhood U around p and a dif-
feomorphism t : π −1 (U) → U × G of the form t(u) = (π (u), ψ (u)) , where
ψ : π −1 (U) → G is a map satisfying
ψ (ug) = ψ (u) g for every u ∈ π −1 (U) and g ∈ G .
The pair (t, π −1 (U)) is called a local trivialization around p.
10.18 Remark It follows from 10.16 that a principal G-bundle π : P → M as

defined in Definition 10.15 clearly satisfies Definition 10.17.
Conversely, suppose that π : P → M is a principal G-bundle as defined in Def-
inition 10.17. Then we have a family A = {(tα , π −1 (Uα ))|α ∈ A} of local triv-
ializations such that {Uα |α ∈ A} is an open cover of M, and where each diffeo-
morphism tα : π −1 (Uα ) → Uα × G is of the form tα (u) = (π (u), ψα (u)) where
ψα : π −1 (Uα ) → G is a map satisfying
ψα (ug) = ψα (u) g for every u ∈ π −1 (Uα ) and g ∈ G .
If u ∈ π −1 (Uα ∩Uβ ), we have that
ψβ (ug) ψα (ug)−1 = ψβ (u) ψα (u)−1
for every g ∈ G, which shows that the map ψ β α : π −1 (Uα ∩Uβ ) → G given by
ψ β α (u) = ψβ (u) ψα (u)−1
is constant on each fibre. Hence it follows that the local trivializations (tα , π −1 (Uα ))
and (tβ , π −1 (Uβ )) are G-related with transition map φβ α : Uα ∩Uβ → G given by
φβ α (p) = ψ β α ( tα−1 (p, e))
for p ∈ Uα ∩Uβ . This shows that A is a trivializing cover of E which is contained in

a unique fibre bundle structure D by Proposition 10.2.
10.19 Definition Let πi : Pi → Mi be a principal Gi -bundle for i = 1, 2. A ho-

momorphism from π1 to π2 is a triple ( f˜, f , ψ ) of smooth maps f˜ : P1 → P2 and
f : M1 → M2 and a Lie group homomorphism ψ : G1 → G2 such that
π2 ◦ f˜ = f ◦ π1 (1)
and
f˜ ◦ R g = R ψ (g) ◦ f˜ (2)
for every g ∈ G1 . We let f˜p denote the induced smooth map f˜p : π1−1 (p) → π2−1 ( f (p))
between the fibres for each p ∈ M1 .
If f˜, f and ψ are inclusion maps which are immersions, then π1 : P1 → M1 is
called a subbundle of π2 : P2 → M2 . If in addition M2 = M1 and f = id, then the
homomorphism ( f˜, f , ψ ) is called a reduction of the structure group and π1 : P1 →
M1 is called a reduced subbundle of π2 : P2 → M2 . If f˜, f and ψ are diffeomorphisms,
then ( f˜, f , ψ ) is called an isomorphism. An isomorphism of a principal fibre bundle
with itself is called an automorphism.
10.20 Remark A homomorphism ( f˜, f , ψ ) is actually determined by the pair

( f , ψ ) satisfying (2), since f (p) = π2 ◦ f˜(u) for any u ∈ P1 with π1 (u) = p. This is
˜
well defined, for if we choose another u′ ∈ P1 with π1 (u′ ) = p, then u′ = ug for a
g ∈ G, so that
π2 ◦ f˜(u′ ) = π2 ( f˜(u) ψ (g)) = π2 ◦ f˜(u) .
If (t, π −1 (V )) is a local trivialization in P and s is the section on V given by s(p) =
t −1 (p, e) for p ∈ V , then f |V = π2 ◦ f˜ ◦ s which shows that f is smooth.
10.21 Example By a frame in a vector space V we mean an ordered basis B =

(v1 , ..., vn ) for V . With each frame B in V we associate a linear isomorphism F :
Rn → V given by F(ei ) = vi for i = 1, ..., n, where E = (e1 , ..., en ) is the standard
basis for Rn , so that
n
F(a) = ∑ ai vi
i=1
for a ∈ R . Conversely, each linear isomorphism F : Rn → V is obtained from the
n
unique frame (F(e1 ), ..., F(en )) in V . We let L (V ) denote the set of all linear iso-
morphisms F : Rn → V .
If π : E → M is an n-dimensional vector bundle, we define L p (E) = L (E p ) for
p ∈ M. We want to prove that
[
L (E) = L p (E)
p∈M
is a principal fibre bundle over M called the frame bundle of E, with structure group
Gl (Rn ) and projection π ′ : L (E) → M sending each set L p (E) to p.
Let {(tα , π −1 (Uα )) | α ∈ A} be a trivializing cover of E. For each α ∈ A we have
a bijection φα : π ′ −1 (Uα ) → Uα × Gl (Rn ) defined by
φ α (u) = (p , tα ,p ◦ u)
for p ∈ Uα and u ∈ L p (E). For each pair of indices α , β ∈ A, we have that
φ α ◦ φβ−1 (p, F) = (p , φ α β (p) ◦ F)

FIBRE BUNDLES 343
for (p, F) ∈ (Uα ∩Uβ ) × Gl (Rn ) , where φ α β : Uα ∩Uβ → Gl (Rn ) is the differen-
tiable map given by
φ α β (p) = tα ,p ◦ tβ−1
,p
for p ∈ Uα ∩Uβ . By Proposition 10.5 it follows that π ′ : L (E) → M is a principal

fibre bundle over M with structure group Gl (Rn ) and with (φα , π ′ −1 (Uα )) as a local
trivialization for each α ∈ A.
A map s : V → L (E) defined on an open subset V of M is a section of π ′ if and
only if the maps s j : V → E defined by s j (p) = s(p)(e j ) for p ∈ V and j = 1, ... , n
are sections of π : E → M forming a local basis for E. Indeed, for each α ∈ A we
have that
ψ (tα ,p ◦ s(p)) i j = ei ◦ tα ,p ◦ s(p)(e j ) = ei ◦ tα ,p ( s j (p)) = [ tα ,p ( s j (p)) ]i
for p ∈ V ∩ Uα and i, j = 1, ... , n, which shows that s is smooth on V if and only if

s j is smooth on V for j = 1, ... , n. The section s is also written as s = (s1 , ... , sn ) .
We see that L (E) is an open submanifold of the total space Λ 1 (Rn ; E) of the
vector bundle π ′′ : Λ 1 (Rn ; E) → M defined in 5.27. Hence a section s of π ′ may be
identified with the section i ◦ s of π ′′ , where i : L (E) → Λ 1 (Rn ; E) is the inclusion
map.
Let ( fˆ, f ) be a bundle map between the n-dimensional vector bundles π 1 : E 1 →
M 1 and π 2 : E 2 → M 2 , where fˆ : E 1 → E 2 induces a linear isomorphism between
the fibres π1−1 (p) and π2−1 ( f (p)) for every p ∈ M 1 . Then we have a homomorphism
( f˜, f , id) between the frame bundles π 1′ : L (E 1 ) → M 1 and π ′2 : L (E 2 ) → M 2 ,
where f˜ : L (E 1 ) → L (E 2 ) is the map given by
f˜(u) = fˆp ◦ u
for p ∈ M 1 and u ∈ L p (E 1 ) , and id : Gl (Rn ) → Gl (Rn ) is the identity map. To

see that f˜ is smooth, let (t i , π −1 (U i )) be a local trivialization in π i : E i → M i and
i
(φ i , π i′ −1 (U i )) be the corresponding local trivialization in π i′ : L (E i ) → M i for i =
1, 2 . Then we have that
φ 2 ◦ f˜ ◦ φ 1−1 (p, F) = (p , φ (p) ◦ F)
for (p, F) ∈ (U 1 ∩ f −1 (U 2 )) × Gl (Rn ) , where φ : U 1 ∩ f −1 (U 2 ) → Gl (Rn ) is the

smooth map given by
−1
φ (p) = t 2, f (p) ◦ fˆp ◦ t 1,p
for p ∈ U 1 ∩ f −1 (U 2 ) .
10.22 Remark We say that (φ , π ′ −1 (U)), where φ p (u) = t p ◦ u for p ∈ U and u ∈

L p (E), is the local trivialization in the frame bundle π ′ : L (E) → M corresponding
to the local trivialization (t, π −1 (U)) in the vector bundle π : E → M.
The frame bundle L (T M) of the tangent bundle of M is denoted simply by
L (M) and is called the frame bundle of M, and L p (T M) is denoted by L p (M) for
p ∈ M. If (tx , π −1 (U)) is the local trivialization in the tangent bundle associated with
a local chart (x,U) on M, then the corresponding local trivialization (φ , π ′ −1 (U)) in
the frame bundle is denoted by (φ x , π ′ −1 (U)) . If X1 , ... , Xn are vector fields on an
open subset V of M forming a local basis for T M, then the section s = (X1 , ... , Xn )
in π ′ : L (M) → M is called a frame field on V .
10.23 Example If V is an n-dimensional vector space, we let A (V ) be the set

of invertible affine maps w : Rn → V given by w(b) = u(b) + v for b ∈ Rn , where
u ∈ L (V ) and v ∈ V . We have a bijection σ V : L (V ) × V → A (V ) given by
σ V (u, v)(b) = u(b) + v
for b ∈ Rn . The Lie group GA (Rn ) defined in Example 9.35 operates on A (V ) on

the right by composition, and we have that
σ V (u, v) ◦ σn (T, a) (b) = u( T (b) + a) + v = u ◦ T (b) + u(a) + v

= σ V (u ◦ T , u(a) + v)(b) = σ V ((u, v) (T, a ))(b)
where σn : Gl (Rn ) ×id Rn → GA (Rn ) is the Lie group isomorphism given by
σn (T, a)(b) = T (b) + a
for b ∈ Rn , and we define an operation of the Lie group Gl (Rn )×id Rn on L (V )×V
on the right by
(u, v) (T, a ) = (u ◦ T , u(a) + v )
for (u, v) ∈ L (V ) × V and (T, a) ∈ Gl (Rn ) ×id Rn .
If π : E → M is an n-dimensional vector bundle, we define A p (E) = A (E p ) for
p ∈ M, and let σ p = σ E p : L p (E) × E p → A p (E) be the bijection defined above.
We want to prove that [
A (E) = A p (E)
p∈M
is a principal fibre bundle over M called the affine frame bundle of E, with structure
group GA (Rn ) and projection π ′′ : A (E) → M sending each set A p (E) to p.
Let {(tα , π −1 (Uα )) | α ∈ A} be a trivializing cover of E where each Uα is the
coordinate neighbourhood for a local chart (xα ,Uα ) on M. For each α ∈ A we have
a bijection φα : π ′′ −1 (U) → U × GA (Rn ) defined by
φ α ◦ σ p (u, v) = (p , σn (tα ,p ◦ u, tα ,p (v) + xα (p)))
for p ∈ Uα and (u, v) ∈ L p (E) × E p . For each pair of indices α , β ∈ A , we have that
φ α ◦ φβ−1 (p , σn (T, a )) = (p , σn (tα ,p ◦ tβ−1 −1

,p ◦ T , tα ,p ◦ tβ ,p (a − xβ (p)) + xα (p) ))
= (p , φ α β (p) ◦ σn (T, a ))
FIBRE BUNDLES 345
for p ∈ Uα ∩Uβ and (T, a) ∈ Gl (Rn ) ×id Rn , where φ α β : Uα ∩Uβ → GA (Rn ) is

the smooth map given by
φ α β (p) = σn (tα ,p ◦ tα−1,p , xα (p) − tα ,p ◦ tβ−1
,p (xβ (p)) )
for p ∈ Uα ∩Uβ . By Proposition 10.5 it follows that π ′′ : A (E) → M is a principal

fibre bundle over M with structure group GA (Rn ) and with (φα , π ′′ −1 (Uα )) as a
local trivialization for each α ∈ A.
The structure group GA (Rn ) operates on A (E) on the right by composition
since
φ α ,p (σ p (u, v) σn (T, a )) = σn (tα ,p ◦ u, tα ,p (v) + xα (p)) ◦ σn (T, a )

= σn ( (tα ,p ◦ u) ◦ T , tα ,p ◦ u(a) + tα ,p(v) + xα (p) )
= σn (tα ,p ◦ (u ◦ T ), tα ,p ( u(a) + v) + xα (p)) = φ α ,p ◦ σ p ((u, v) (T, a ))
= φ α ,p (σ p (u, v) ◦ σn (T, a ))
for (u, v) ∈ L p (E) × E p and (T, a) ∈ Gl (Rn ) ×id Rn .
10.24 Remark We say that (φ , π ′ −1 (U)) , where φ p ◦ σ p (u, v) = σn (t p ◦ u,

t p (v) + x(p)) for p ∈ U and (u, v) ∈ L p (E) × E p , is the local trivialization in
the affine frame bundle π ′ : A (E) → M corresponding to the local trivialization
(t, π −1 (U)) in the vector bundle π : E → M , where U is the coordinate neighbour-
hood for a local chart (x,U) on M.
The affine frame bundle A (T M) of the tangent bundle of M is denoted simply
by A (M) and is called the affine frame bundle of M, and A p (T M) is denoted by
A p (M) for p ∈ M. If (tx , π −1 (U)) is the local trivialization in the tangent bundle
assosiated with a local chart (x,U) on M, then the corresponding local trivialization
(φ , π ′ −1 (U)) in the affine frame bundle is denoted by (φ x , π ′ −1 (U)) .
10.25 Example Given an n-dimensional vector bundle π : E → M, we let πe :
E (E) → L (E) be the vector bundle induced from π : E → M by the projection
π ′ : L (E) → M in the frame bundle of E, as shown in the commutative diagram
πe ′ ✲
E (E) E
πe π
❄ ❄
π ′✲
L (E) M
where [
E (E) = L p (E) × E p .
p∈M
If (t, π −1 (U)) is a local trivialization in the vector bundle π : E → M , where U is the

coordinate neighbourhood for a local chart (x,U) on M, then the corresponding local
trivialization (e e in the vector bundle πe : E (E) → L (E) is given by
t , πe−1 (U))
e
t (u, v) = (u,t p (v))
for p ∈ U and (u, v) ∈ L p (E) × E p , where Ue = π ′ −1 (U).

′
We also have that πe : E (E) → E is a principal fibre bundle with structure group
e in this bundle is given by
t′ , πe−1 (U))
Gl (Rn ) . The corresponding local trivialization (e
t′ (u, v) = (v,t p ◦ u)
e
for p ∈ U and (u, v) ∈ L p (E) × E p.

We have a smooth map ψ : E (E) → Rn defined by
ψ (u, v) = u−1 (v)
for p ∈ M and (u, v) ∈ L p (E) × E p . Indeed, if (φ′ , π ′ −1 (U)) is the local trivializa-
tion in the frame bundle π ′ : L (E) → M corresponding to the local trivialization
(t, π −1 (U)) in π : E → M , we have that
ψ ◦ et −1 ◦ (φ′ × id Rn ) −1 (p, T, a) = ψ ( t p−1 ◦ T , t p−1 (a) ) = T −1 (a)
for p ∈ U , T ∈ Gl (Rn ) and a ∈ Rn .

The map σ : E (E) → A (E) given by σ (u, v) = σ p (u, v) for p ∈ M and (u, v) ∈
L p (E) × E p is a diffeomorphism, since
φ ◦ σ ◦ et −1 ◦ (φ′ × id Rn ) −1 (p, T, a) = (p, σn ( T , a + x(p)))
for p ∈ U, a ∈ Rn and T ∈ Gl (Rn ) , so that
(x × σn−1) ◦ φ α ◦ σ ◦ e
t −1 ◦ ((x × id Gl (Rn ) ) ◦ φ′ × id Rn ) −1 (b, T, a) = (b, T , a + b)
for a, b ∈ Rn and T ∈ Gl (Rn ) .

If π : T M → M is the tangent bundle over M , then E (T M) is denoted simply
by E (M) .
10.26 Proposition If π : P → M is a principal G-bundle and u ∈ P, then the map

σu : G → P defined by σu (g) = ug for g ∈ G, is an embedding whose image is the
fibre Pp where p = π (u). Moreover, the image σ u ∗ (g) = Vu is the vertical subspace
at u, and these subspaces form a distribution V on P.
If X ∈ g , then the vector field σ (X) on P defined by σ (X) u = σu ∗ (X) for
u ∈ P, belongs to V and is called the fundamental vector field determined by X. If
{X1 , ..., Xm } is a basis for g, then {σ (X1 ), ..., σ (Xm )} is a local basis for V on P.
PROOF : From 10.16 we have that σu = i p ◦ ρ u , where i p : Pp → P is the inclusion
map and ρ u : G → Pp is the diffeomorphism given by
ρ u = t p−1 ◦ L t p (u) ,
FIBRE BUNDLES 347
i.e.,
ρ u (g) = t p−1 (t p (u) g)
for g ∈ G. Hence it follows that σu is an embedding whose image is Pp , and we have
that
σ u ∗ (g) = i p ∗ ◦ ρ u ∗ (g) = i p ∗ (Tu Pp ) = Vu .
The last part of the proposition now follows from Proposition 2.62 and 9.37.
10.27 Remark By Proposition 4.95 and 10.3 we know that the distribution V on
P is integrable having the connected components of the fibres of π : P → M as its
maximal connected integral manifolds.
ASSOCIATED BUNDLES
10.28 Let π : P → M be a principal G-bundle, and let ψ : G → G′ be a Lie group
homomorphism into a Lie group G′ which acts on a smooth manifold F effectively
on the left. We are going to construct a fibre bundle π ′ : E → M with fibre F and
structure group G′ , called the fibre bundle associated with P. Consider the action of
G on the product manifold X = P × F on the right defined by
(u, v) g = (u g, ψ (g−1 ) v) ,
and let E = P ×ψ F be the set of orbits in X. The orbit of an element (u, v) ∈ X is

denoted by [ u, v] = (u, v) G. We have a canonical projection p ′ : X → E given by
p ′ (u, v) = [ u, v] , and a map π ′ : E → M defined by π ′ ([ u, v]) = π (u) so that the
diagram
p ′✲
X E
pr1 π′
❄ ❄
π✲
P M
is commutative, where pr1 : X → P is the projection on the first factor.

Now let A = {(tα , π −1 (Uα ))|α ∈ A} be a trivializing cover of P. For each α ∈ A
we have a surjective map τα : π −1 (Uα ) × F → Uα × F defined by
τα (u, v) = (p, ψ ◦ tα ,p (u) v)

for (u, v) ∈ π −1 (Uα ) × F , where p = π (u) . Since
τα (u′ , v′ ) = τα (u, v) ⇔ π (u′ ) = p and ψ ◦ tα ,p (u′ ) v′ = ψ ◦ tα ,p (u) v

⇔ u′ = u g for a g ∈ G and ψ ◦ tα ,p (u′ ) v′ = ψ ◦ tα ,p (u′ ) (ψ (g−1 ) v)
⇔ u′ = u g and v′ = ψ (g−1 ) v for a g ∈ G ⇔ [ u′ , v′ ] = [ u, v] ,
we therefore have a well-defined bijection t ′ α : π ′ −1 (Uα ) → Uα × F with t ′ α ◦ p ′ =

τα , given by
t ′ α ([ u, v]) = (p, ψ ◦ tα ,p (u) v) (1)
for [ u, v] ∈ π ′ −1 (Uα ) where p = π (u) . If sα : Uα → P is the section of π defined by
sα (p) = tα−1 (p, e)
for p ∈ Uα , then
t ′ α ([ sα (p), v]) = (p, v) (2)
for (p, v) ∈ Uα × F. Now if (tα , π −1 (Uα ))
and (tβ , π −1 (Uβ ))
are two local trivializa-
tions in A which are G-related with transition map φ β α : Uα ∩ Uβ → G, we have
that
−1
t ′ β ◦ t ′ α (p, v) = (p, ψ ◦ tβ ,p (sα (p)) v) = (p, ψ ◦ tβ ,p ◦ tα−1,p (e) v) = (p, ψ ◦ φ β α (p) v)
for every p ∈ Uα ∩ Uβ and v ∈ F. By Proposition 10.5 there is therefore a unique

topology, smooth structure and fibre bundle structure on E such that π ′ : E → N is a
fibre bundle over M with fibre F and structure group G′ , and such that (t ′ α , π ′ −1 (Uα ))
is a local trivialization for each α ∈ A. It is called the local trivialization in E associ-
ated with the local trivialization (tα , π −1 (Uα )) in P.
The canonical projection p ′ : X → E is smooth as we see from the commutative
diagram
p ′ π ′ −1 (Uα )
π −1 (Uα ) × F ✲ π ′ −1 (U )
α
❍
❍❍
tα × id F τα❍❍ t ′α
❄ ❍❍
❥ ❄
Uα × G × F ✲ U ×F
α
id Uα × µ
where µ : G × F → F is the group action of G on F defined by µ (g, v) = ψ (g) v .

Actually, p ′ : X → E is a principal G-bundle over E. To see this, let φ : G ×
F → F × G be the diffeomorphism defined by φ (g, v) = (ψ (g) v, g) with inverse
φ −1 (v, g) = (g, ψ (g−1 ) v) .
FIBRE BUNDLES 349
φ✲
G×F F ×G
❅
µ ❅ pr1
❘
❅ ✠
F
Then we have local trivializations (π −1 (Uα ) ×F , ρα ) in X, where ρα : π −1 (Uα ) ×

F → π ′ −1 (Uα ) × G is the diffeomorphism given by the commutative diagram
ρα ✲ π ′ −1 (U ) × G
π −1 (Uα ) × F α
tα × id F t ′ α × id G
❄ ❄
id Uα × φ
Uα × G × F ✲ U ×F ×G
α
so that
ρα (u, v) = (t ′ α × idG )−1 (τα ,p (u, v),tα ,p (u)) = (p ′ (u, v),tα ,p (u))
where p = π (u) , and
ρα−1 (w, g) = (tα × idF )−1 (p, g, ψ (g−1 ) t ′ α ,p (w)) = (tα−1,p (g), ψ (g−1 ) t ′ α ,p (w))
where p = π ′ (w) . Using this, we see that
ρβ ◦ ρα−1 (w, g) = (w,tβ ,p ◦ tα−1,p (g)) = (w, φ β α (p) g)
for every w ∈ π ′ −1 (Uα ∩Uβ ) and g ∈ G.

For each u ∈ P we have a diffeomorphism µ u : F → E p from the typical fibre F
to the fibre E p where p = π (u) , given by
µ u (v) = [ u, v]
for v ∈ F. This follows from (1) which implies that
t ′ α ,p ◦ µ u (v) = ψ (tα ,p (u)) v
so that
−1
µ u = t ′ α ,p ◦ L ψ ( tα ,p (u)) ,
showing that µ u is the composition of the diffeomorphisms L ψ ( tα ,p (u)) : F → F and
t ′ α−1
,p : F → E p .
For each v ∈ F we have a smooth map ν v : P → E given by
ν v (u) = [ u, v]
for u ∈ P, so that the diagram
νv ✲
P E
❙ ✓ ′
π ❙ ✓ π
✇
❙ ✴
✓
M
is commutative. This follows from the commutative diagram in the beginning of this
section and the relation
tα (u) = (p,tα ,p (u))
for u ∈ π −1 (Uα ) where π (u) = p , which together with (1) implies that
t ′ α ◦ ν v ◦ tα−1 (p, g) = (p, ψ (g) v)
for p ∈ Uα and g ∈ G.
Since
[ u g, v] = [ u, ψ (g) v] ,
it follows that
µ ug = µ u ◦ L ψ (g)
and
ν ψ (g) v = ν v ◦ R g
for u ∈ P, v ∈ F and g ∈ G.
10.29 Examples
(a) Let π : P → M be a principal G-bundle, and let π ′ : E → M be the fibre bundle
associated with P with fibre G obtained from the Lie group homomorphism
id : G → G, using that G is acting effectively on itself on the left by left transla-
tion. We contend that π ′ : E → M is equivalent over M to π : P → M, with the
equivalence e : E → P given by
e ([ u, g ]) = ug
for u ∈ P and g ∈ G. This is clearly well defined, for if h ∈ G, then
e ([ uh, id (h−1) g ]) = uhh−1 g = ug = e ([ u, g ]) .
To show that e is an equivalence, let (t, π −1 (U)) be a local trivialization in P,

FIBRE BUNDLES 351
and let (t ′ , π ′ −1 (U)) be the associated local trivialization in E defined as in

10.28 by
t ′ ([ u, g]) = (p,t p (u) g) = (p,t p (ug)) = t ◦ e ([ u, g ])
for u ∈ π −1 (p) and g ∈ G where p ∈ U , which shows that

−1
t ◦ e ◦ t′ = id U × G .
The inverse e−1 : P → E is given by
e−1 (u) = [ u, e ]
for u ∈ P.
Using the equivalence e, we may therefore identify the fibre bundle E associ-
ated with P with fibre G with the principal fibre bundle π : P → M itself, and the
diffeomorphism µ u : G → E p defined in 10.28 is then identified with the dif-
feomorphism e p ◦ µ u : G → Pp which is induced by the embedding σu : G → P
defined in Proposition 10.26, since
e p ◦ µ u (g) = ug = σu (g)
for each u ∈ P and g ∈ G.

(b) Let π : P → M be a principal G-bundle, and let ψ : G → Aut (W ) be a repre-
sentation of G on an m-dimensional vector space W . Then Aut (W ) acts effec-
tively on W on the left, and we will show that the fibre bundle π ′ : E → M
with fibre W associated with P is an m-dimensional vector bundle over M.
Using the same notation as in 10.28, we consider two local trivializations
(tα , π −1 (Uα )) and (tβ , π −1 (Uβ )) for P which are G-related with transition map
φ β α : Uα ∩Uβ → G. Then t ′ β ◦ t ′ −1 α is a diffeomorphism from (Uα ∩Uβ ) × W
onto itself which is a linear isomorphism on each fibre {p}×W for p ∈ Uα ∩Uβ ,
since
−1
t ′ β ,p ◦ t ′ α ,p = ψ ◦ φ β α (p) ∈ Aut (W ) .
Hence there is a unique vector space structure on each fibre E p such that
π ′ : E → M is a vector bundle having ((id × x) ◦ t ′ α , π ′ −1 (Uα )) as a local trivi-
alization for each α ∈ A, where x : W → Rm is any linear isomorphism. Since
t ′ α ,p ([ u, v]) = ψ ◦ tα ,p (u)(v) ,
where ψ ◦ tα ,p (u) ∈ Aut (W ), we have that

−1
a1 [ u, v1 ] + a2 [ u, v2 ] = t ′ α ,p (a1 t ′ α ,p ([ u, v1 ]) + a2 t ′ α ,p ([ u, v2 ]))
= [ u, a1 v1 + a2v2 ]
for u ∈ π −1 (p), v1 , v2 ∈ W and a1 , a2 ∈ R.

From this it also follows that (p ′ , π ) is a bundle map between the vector bun-
dles pr1 : P × W → P and π ′ : E → M which is a linear isomorphism on each
fibre. As usual pr1 : P × W → P is given the vector bundle structure with
(id × x, pr1−1 (P)) as a local trivialization, i.e., with product manifold structure
and vector space operations
a1 ( u, v1 ) + a2 ( u, v2 ) = ( u, a1 v1 + a2v2 )
on each fibre {u} × W so that the projection ρu : {u} × W → W on the second

factor is a linear isomorphism and
t ′ α ,p ◦ p ′ ◦ ρu−1 = ψ ◦ tα ,p (u) ∈ Aut (W )
when π (u) = p ∈ Uα , thus showing the above assertions about (p ′ , π ). It also

follows that the map µ u = p u′ ◦ ρu−1 : W → E p where p = π (u) , given by
µ u (v) = [ u, v]
for v ∈ W , is a linear isomorphism for each u ∈ P.

(c) Let π : P → M be a principal G-bundle, and let π ′ : E → M be the associated
vector bundle obtained from a representation ψ : G → Aut (W ) of G on an m-
dimensional vector space W as described in (b). Let ρ : G → Aut (gl (W )) be
the representation of the structure group G on the finite dimensional vector
space gl (W ) given by
ρ (g)(F) = ψ (g) ◦ F
for g ∈ G and F ∈ gl (W ). Then it follows from (b) that the fibre bundle
π ′′ : E ′ → M with fibre gl (W ) associated with P is a vector bundle which
we contend is equivalent over M to the vector bundle π ′′′ : Λ 1 (W ; E ) → M
defined in 5.27, with the equivalence e : E ′ → Λ 1 (W ; E ) given by
e ([ u, F ]) = µ u ◦ F
for u ∈ P and F ∈ gl (W ). This is clearly well defined, for if g ∈ G, then
e ([ u g, ρ (g−1 ) F ]) = µ u g ◦ ρ (g−1 ) (F) = µ u ◦ ψ (g) ◦ ψ (g−1 ) ◦ F

= µ u ◦ F = e ([ u, F ]) .
To show that e is an equivalence, let (t , π −1 (U)) be a local trivialization in P,
and let (t ′ , π ′ −1 (U)) and (t ′′ , π ′′ −1 (U)) be the corresponding local trivializa-
tions in E and E ′ , respectively, defined as in 10.28 by
t ′ ([ u, v]) = (p, ψ (t p (u)) (v))
and
t ′′ ([ u, F ]) = (p, ρ (t p (u)) (F))
for [ u, v] ∈ π ′ −1 (U) and [ u, F ] ∈ π ′′ −1 (U) where p = π (u) . Given a linear iso-
morphism x : W → Rm , we obtain an equivalence ex : U ×gl (W ) → U ×gl (Rm )
FIBRE BUNDLES 353
defined by ex (p, F) = (p, x ◦ F ◦ x−1 ) for (p, F) ∈ U × gl (W ) . Combining this,

we now have that
t ′′ ([ u, F ]) = (p, ψ (t p (u)) ◦ F) = (p,t p′ ◦ µ u ◦ F) = e−1

x ◦ τ x ◦ e ([ u, F ])
where τ x : π ′′′ −1 (U) → U × gl (Rm ) is the equivalence over U defined in Corol-

lary 5.29, showing that
−1
e−1
x ◦ τx ◦ e ◦ t
′′
= id U × gl (W ) .
Using the equivalence e, we may therefore identify the fibre bundle E ′ associ-
ated with P with fibre gl (W ) with the vector bundle π ′′′ : Λ 1 (W ; E ) → M ,
and the linear isomorphism µ ′u : gl (W ) → E p′ defined in (b) is then identified
with the linear isomorphism µ ′′u = e p ◦ µ ′u : gl (W ) → Λ 1 (W ; E p ) given by
µ ′′u (F) = µ u ◦ F
for u ∈ Pp and F ∈ gl (W ) where p ∈ M .

We have a smooth map ν = e ◦ ν′id W : P → Λ 1 (W ; E ) given by
ν (u) = µ u
for u ∈ P. From 10.28 we see that
t ′′ ◦ ν′id W ◦ t −1 = idU × ψ
so that
e−1
x ◦ τx ◦ ν ◦ t
−1
= idU × ψ
for every local trivialization (t , π −1 (U)) in P.
(d) Let π : P → M be a principal G-bundle, and let ρ : G → Aut (W ) be the repre-
sentation of G on a finite dimensional vector space W given by
ρ (g) = id W
for g ∈ G . Then it follows from (b) that the fibre bundle π ′ : E → M with fibre
W associated with P is a vector bundle which we contend is equivalent over M
to the trivial vector bundle πe : EW → M defined in 5.26, with the equivalence
e : E ′ → EW given by
e ([ u, v ]) = (π (u), v)
for u ∈ P and v ∈ W . This is clearly well defined, for if g ∈ G , then
e ([ ug, ρ (g−1 ) v ]) = (π (ug), v) = (π (u), v) = e ([ u, v ]) .
To show that e is an equivalence, let (t , π −1 (U)) be a local trivialization in P,

and let (t ′ , π ′ −1 (U)) be the corresponding local trivialization in E defined as in

10.28 by
t ′ ([ u, v]) = (p, ρ (t p (u)) (v))
for [ u, v] ∈ π ′ −1 (U) where p = π (u) . Combining this, we now have that
t ′ ([ u, v ]) = (p, v) = e ([ u, v ]) ,
showing that
−1
id U × W ◦ e ◦ t ′ = id U × W .
Using the equivalence e, we may therefore identify the fibre bundle E associ-
ated with P with fibre W with the vector bundle πe : EW → M , and the linear
isomorphism µ u : W → E p defined in (b) is then identified with the linear iso-
morphism µ ′u = e p ◦ µ u : W → {p} × W given by
µ ′u (v) = (p, v)
for u ∈ Pp and v ∈ W where p ∈ M .

(e) Let π ′ : L (E) → M be the frame bundle of an n-dimensional vector bundle
π : E → M. Then id : Gl (Rn ) → Aut (Rn ) is a representation of the structure
group Gl (Rn ) on the finite dimensional vector space Rn , so by (b) the fibre
bundle π ′′ : E ′ → M with fibre Rn associated with L (E) is a vector bundle
which we contend is equivalent over M to the vector bundle π : E → M, with
the equivalence e : E ′ → E given by
e ([ u, ξ ]) = u( ξ )
for u ∈ L (E) and ξ ∈ Rn . This is clearly well defined, for if G ∈ Gl (Rn ), then
e ([ u ◦ G, id (G−1 ) ξ ]) = u ◦ G ◦ G−1 ( ξ ) = u( ξ ) = e ([ u, ξ ]) .
To show that e is an equivalence, let (t , π −1 (U)) be a local trivialization in

the vector bundle π : E → M, and let (φ , π ′ −1 (U)) and (t ′ , π ′′ −1 (U)) be the
corresponding local trivializations in L (E) and E ′ , respectively, defined as in
10.28 and Example 10.21 by
φ (u) = (p , t p ◦ u)
and
t ′ ([ u, ξ ]) = (p, φ p (u) ξ )
for u ∈ L p (E) and ξ ∈ Rn where p ∈ U . Combining this, we have that
t ′ ([ u, ξ ]) = (p, t p ◦ u ( ξ )) = t ◦ e ([ u, ξ ])
which shows that

t ◦ e ◦ t ′ −1 = id U × Rn .
FIBRE BUNDLES 355
ated with L (E) with fibre Rn with the vector bundle π : E → M, and the linear
isomorphism µ u : Rn → E p′ defined in (b) is then identified with u : Rn → E p
since
ep ◦ µ u = u
for each u ∈ L (E).
(f) Let π ′ : L (E) → M be the frame bundle of an n-dimensional vector bundle
π : E → M. By Proposition 8.13 we have a representation ρ kl : Gl (Rn ) →
Aut (T kl (Rn )) of the structure group Gl (Rn ) on the finite dimensional vector
space T kl (Rn ) given by
ρ kl (F) = (F −1) ∗
for F ∈ Gl (Rn ), i.e.,
ρ kl (F)(T )(v1 , ... , vk , λ1 , ... , λl ) = T (F −1 (v1 ), ... , F −1 (vk ), λ1 ◦ F, ... , λl ◦ F)
for T ∈ T kl (Rn ) , v1 , ... , vk ∈ Rn and λ 1 , ... , λ l ∈ (Rn )∗ . By (b) the fibre

bundle π ′′ : E ′ → M with fibre T kl (Rn ) associated with L (E) is therefore
a vector bundle which we contend is equivalent over M to the tensor bundle
πlk : Tlk (E) → M, with the equivalence e : E ′ → Tlk (E) given by
e ([ u, T ]) = (u −1) ∗ (T )
for u ∈ L (E) and T ∈ T kl (Rn ), i.e.
e ([ u, T ])(v1 , ... , vk , λ 1 , ... , λ l ) = T (u−1 (v1 ), ... , u−1 (vk ), λ 1 ◦ u, ... , λ l ◦ u)
for u ∈ L p (E) , T ∈ T kl (Rn ) , v1 , ... , vk ∈ E p and λ 1 , ... , λ l ∈ E p∗ where p ∈ M.

This is clearly well defined, for if G ∈ Gl (Rn ), then
e ([ u ◦ G, ρ kl (G−1 ) T ]) = (u−1 ) ∗ ◦ (G−1 ) ∗ ◦ G ∗ (T ) = (u−1) ∗ (T ) = e ([ u, T ]).

φ (u) = (p , t p ◦ u)
and
t ′ ([ u, T ]) = (p, ρ kl (φ p (u)) (T ))
for u ∈ L p (E) and T ∈ T kl (Rn ) where p ∈ U . Combining this, we have that
t ′ ([ u, T ]) = (p, (t p−1 ) ∗ ◦ (u−1 ) ∗ (T )) = τ ◦ e ([ u, T ])

where τ : (πlk ) −1 (U) → U × T kl (Rn ) is the equivalence over U defined in

Proposition 4.53, showing that
τ ◦ e ◦ t ′ −1 = id U × T k (Rn ) .
l
ated with L (E) with fibre T kl (Rn ) with the tensor bundle πlk : Tlk (E) → M,
and the linear isomorphism µ u : T kl (Rn ) → E p′ defined in (b) is then identified
with (u −1) ∗ : T kl (Rn ) → T kl (E p ) since
e p ◦ µ u = (u −1) ∗
for u ∈ L p (E).
(g) Let π ′ : L (E) → M be the frame bundle of an n-dimensional vector bundle π :
E → M. Then Ad : Gl (Rn ) → Aut (gl (Rn )) is a representation of the structure
group Gl (Rn ) on the finite dimensional vector space gl (Rn ), so by (b) the fibre
bundle π ′′ : E ′ → M with fibre gl (Rn ) associated with L (E) is a vector bundle
which we contend is equivalent over M to the vector bundle π ′′′ : Λ 1 (E ; E ) →
M , with the equivalence e : E ′ → Λ 1 (E ; E ) given by
e ([ u, F ]) = u ◦ F ◦ u−1
for u ∈ L (E) and F ∈ gl (Rn ). This is clearly well defined, for if G ∈ Gl (Rn ),
then
e([u ◦ G, Ad(G−1 )F]) = (u ◦ G) ◦ (G−1 ◦ F ◦ G) ◦ (G−1 ◦ u−1 )

= u ◦ F ◦ u−1 = e([u, F]).

φ (u) = (p , t p ◦ u)
and
t ′ ([ u, F ]) = (p, Ad (φ p (u)) (F))
for u ∈ L p (E) and F ∈ gl (Rn ) where p ∈ U . Combining this, we have that
t ′ ([ u, F ]) = (p, t p ◦ u ◦ F ◦ u−1 ◦ t p−1 ) = τ ◦ e ([ u, F ])
where τ : π ′′′ −1 (U) → U × gl (Rn ) is the equivalence over U defined in Propo-

sition 5.28, showing that
τ ◦ e ◦ t ′ −1 = id U × gl (Rn ) .
FIBRE BUNDLES 357
Using the equivalence e, we may therefore identify the fibre bundle E ′ asso-
ciated with L (E) with fibre gl (Rn ) with the vector bundle π ′′′ : Λ 1 (E ; E )
→ M , and the linear isomorphism µ u : gl (Rn ) → E p′ defined in (b) is then
identified with the linear isomorphism µ ′u = e p ◦ µ u : gl (Rn ) → Λ 1 (E p ; E p )
given by
µ ′u (F) = u ◦ F ◦ u−1
for u ∈ L p (E) and F ∈ gl (Rn ) where p ∈ M .
(h) Let π ′ : L (E) → M be the frame bundle of an n-dimensional vector bundle
π : E → M, and let ρ : Gl (Rn ) → Aut (L (Rm , Rn )) be the representation of
the structure group Gl (Rn ) on the finite dimensional vector space L (Rm , Rn )
given by
ρ (G)(F) = G ◦ F
for G ∈ Gl (Rn ) and F ∈ L (Rm , Rn ) . Then it follows from (b) that the fibre
bundle π ′′ : E ′ → M with fibre L (Rm , Rn ) associated with L (E) is a vec-
tor bundle which we contend is equivalent over M to the vector bundle π ′′′ :
Λ 1 (Rm ; E ) → M defined in 5.27, with the equivalence e : E ′ → Λ 1 (Rm ; E )
given by
e ([ u, F ]) = u ◦ F
for u ∈ L (E) and F ∈ L (Rm , Rn ) . This is clearly well defined, for if G ∈
Gl (Rn ), then
e ([ u ◦ G, ρ (G−1 ) F ]) = (u ◦ G) ◦ (G−1 ◦ F) = u ◦ F = e ([ u, F ]) .

φ (u) = (p , t p ◦ u)
and
t ′ ([ u, F ]) = (p, ρ (φ p (u)) (F))
for u ∈ L p (E) and F ∈ L (Rm , Rn ) where p ∈ U . Combining this, we have that
t ′ ([ u, F ]) = (p, t p ◦ u ◦ F ) = τ ◦ e ([ u, F ])
where τ : π ′′′ −1 (U) → U × L (Rm , Rn ) is the equivalence over U defined in

Corollary 5.29, showing that
τ ◦ e ◦ t ′ −1 = id U × L (Rm ,Rn ) .
ated with L (E) with fibre L (Rm , Rn ) with the vector bundle π ′′′ : Λ 1 (Rm ; E )
→ M , and the linear isomorphism µ u : L (Rm , Rn ) → E p′ defined in (b) is then

identified with the linear isomorphism µ ′u = e p ◦ µ u : L (Rm , Rn ) → Λ 1 (Rm ;
E p ) given by
µ ′u (F) = u ◦ F
for u ∈ L p (E) and F ∈ L (Rm , Rn ) where p ∈ M .
(i) Let π ′ : L (E) → M be the frame bundle of an n-dimensional vector bundle
π : E → M, and let πe : Ee → M be the associated vector bundle obtained from
a representation ψ : Gl (Rn ) → Aut (W ) of the structure group Gl (Rn ) on an
m-dimensional vector space W . Then we have a representation ρ k : Gl (Rn ) →
Aut(T k (Rn ;W )) of the structure group Gl (Rn ) on the finite dimensional vec-
tor space T k (Rn ;W ) given by
ρ k (G)(F) = ψ (G) ◦ F ◦ {G −1 } k
for G ∈ Gl (Rn ) and F ∈ T k (Rn ;W ) . By (b) the fibre bundle π ′′ : E ′ → M with

fibre T k (Rn ;W ) associated with L (E) is a vector bundle which we contend
e → M, with the equivalence
is equivalent over M to the bundle π k : T k (E ; E)
e given by
e : E ′ → T k (E ; E)
e u ◦ F ◦ {u −1 } k
e ([ u, F ]) = µ
for u ∈ L p (E) and F ∈ T k (Rn ;W ) where p ∈ M . This is clearly well defined,

for if G ∈ Gl (Rn ), then
e ([ u ◦ G, ρ k (G −1 ) F ]) = {µ
e u ◦ ψ (G) } ◦ {ψ (G) −1 ◦ F ◦ {G } k } ◦
{{G −1 } k ◦ {u −1 } k } = e ([ u, F ]) .
the vector bundle π : E → M, and let (φ , π ′ −1 (U)) be the corresponding local
trivialization in L (E) defined as in Example 10.21 by
φ (u) = (p , t p ◦ u)
for u ∈ L p (E) where p ∈ U . Given a linear isomorphism x : W → Rm ,

we let z : T k (Rn ;W ) → T k (Rn ; Rm ) be the linear isomorphism defined by
z(F) = x ◦ F for F ∈ T k (Rn ;W ). Furthermore, let ((id × x) ◦ e t , πe −1(U)) and
−1
′
(t , π ′′ (U)) be the local trivializations in Ee and E , respectively, defined as in
′
10.28 by
e
t ([ u, v ]) = (p, ψ (φ p (u)) (v))
and
t ′ ([ u, F ]) = (p, ρ k (φ p (u)) (F))
for u ∈ L p (E) , v ∈ W and F ∈ T k (Rn ;W ) where p ∈ U . Combining this, we
have that
FIBRE BUNDLES 359
(id × z) ◦ t ′ ([ u, F ]) = (p, x ◦ ψ (φ p (u)) ◦ F ◦ {φ p (u) −1 } k )
= (p, x ◦ e e u ◦ F ◦ {u −1 } k ◦ {t p−1 } k ) = τ ◦ e ([ u, F ])
tp ◦ µ
where τ : π ′′ −1 (U) → U × T k (Rn ; Rm ) is the equivalence over U defined in

Proposition 5.28, showing that
τ ◦ e ◦ t ′ −1 ◦ (id × z)−1 = id U × T k (Rn ;Rm ) .
Using the equivalence e, we may therefore identify the fibre bundle E ′ as-
sociated with L (M) with fibre T k (Rn ;W ) with the vector bundle π k :
e → M , and the linear isomorphism µ u : T k (Rn ;W ) → E p′ defined in
T k (E ; E)
(b) is then identified with the linear isomorphism µ ′u = e p ◦ µ u : T k (Rn ;W ) →
T k (E p ; Eep ) given by
µ ′u (F) = µ
e u ◦ F ◦ {u −1 } k
for u ∈ L p (M) and F ∈ T k (Rn ;W ) where p ∈ M .
CONNECTIONS
10.30 Proposition If π : P → M is a principal G-bundle, then
(R g ) ∗ σ (X) = σ ( Ad (g−1 ) X )
for every X ∈ g and g ∈ G.

PROOF : Since
R g ◦ σu (h) = uhg = ug g−1 hg = σug ◦ ig−1 (h)
for u ∈ P and h ∈ G, we have that
((R g ) ∗ σ (X)) ug = (R g )∗ σ (X) u = (R g ) ∗ ◦ σu ∗ (X)
= σug ∗ ◦ Ad (g−1 ) (X) = σ ( Ad (g−1 ) X ) ug .
10.31 Corollary If π : P → M is a principal G-bundle, then
V ug = (R g )∗ V u
for every u ∈ P and g ∈ G.

(R g ) ∗ ◦ σu ∗ = σug ∗ ◦ Ad (g−1 )
so that
(R g ) ∗ V u = (R g ) ∗ ◦ σ u ∗ (g ) = σ ug ∗ (g ) = V ug .
10.32 Definition A (principal) connection in a principal G-bundle π : P → M is

a distribution H on P such that
(i) Tu P = H u ⊕ Vu for every u ∈ P,
(ii) H ug = (R g )∗ H u for every u ∈ P and g ∈ G.
The subspace H u of Tu P is called the horizontal subspace at u, and the vectors in
H u are called horizontal tangent vectors at u.
10.33 If H is a connection in the principal G-bundle π : P → M and u ∈ P, we let

hu : Tu P → Tu P and vu : Tu P → Tu P be the projections on the horizontal and vertical
subspace of Tu P uniquely determined by
w = hu (w) + vu (w) where hu (w) ∈ H u and vu (w) ∈ Vu
for w ∈ Tu P. The vectors hu (w) and vu (w) are called the horizontal and vertical
components of w.
Now the vertical subspace Vu is isomorphic to the Lie algebra g by Proposition
10.26. If τu : Vu → g is the inverse of the linear isomorphism from g to Vu induced
by σu ∗ : T G → T P , we therefore have a g-valued 1-form ωu = τu ◦ vu on Tu P so
that ωu ◦ σu ∗ = id G . We will prove that the map ω : P → Λ 1 (T P ; g) defined by
ω (u) = ωu for u ∈ P, is a g-valued 1-form on P called the connection form of the
given connection H. We first need a definition.
10.34 Definition Let π : P → M be a principal G-bundle. Then a g-valued 1-form

ω on P is called a connection form if it satisfies
(i) ω (u) (σ (X) u ) = X for every u ∈ P and X ∈ g,
(ii) R g∗ ω = Ad (g−1 ) · ω for every g ∈ G.
10.35 Proposition If H is a connection in the principal G-bundle π : P → M,

then the map ω : P → Λ 1 (T P ; g) defined in 10.33 is a connection form on P with
H u = ker ω (u) for u ∈ P.
PROOF : Formula (i) in the definition is just a reformulation of the formula
ωu ◦ σu ∗ = id G established in 10.33.
We next show that ω is a g-valued 1-form on P. Let C = {X1 , ..., Xm } be a
basis for g, and let u ∈ P. Choose a local basis {Ym+1 , ...,Ym+n } for H on an open
FIBRE BUNDLES 361
neighbourhood U of u, and let Y j = σ (X j )|U for j = 1, ..., m. Then {Y1 , ...,Ym+n }

is a local basis for T P on U, and we let {ω 1 , ..., ω m+n } be the dual local basis for
T ∗P on U. In fact, it follows from Corollary 2.59 and Proposition 4.11 that the maps
ω j : U → T ∗P defined by ω j (w) (Yi w ) = δi j for 1 ≤ i, j ≤ m + n and w ∈ U are
sections in T ∗P .
Now the components of ω with respect to the basis C for g are ω j for j =
1, ..., m. Indeed, we have that

X j for 1 ≤ j ≤ m
ω (u)(Y j u ) = .
0 for m + 1 ≤ j ≤ m + n
So if w ∈ Tu P is written as
m+n
w= ∑ ω j (u)(w)Y j u ,
j=1
then
m
ω (u)(w) = ∑ ω j (u)(w) X j
j=1
showing the above assertion and completing the proof that ω is a g-valued 1-form on
P.
To prove formula (ii) in the definition, suppose first that w ∈ H u . Then (R g )∗ w ∈
H ug which implies that
R g∗ ω (u) (w) = ω (ug) ((R g )∗ w) = 0 = Ad (g−1 ) ω (u)(w) .
If on the other hand w ∈ V u , then w = σ (X) u for an X ∈ g. By Proposition 10.30 we

have that
(R g )∗ w = σ ( Ad (g−1 ) X ) ug
so that
R g∗ ω (u) (w) = ω (ug) ((R g )∗ w) = Ad (g−1 ) X = Ad (g−1 ) ω (u)(w) ,
and this completes the proof of formula (ii) since Tu P = H u ⊕Vu . The last part of the
proposition follows immediately from the definition of ω .
10.36 Proposition Let ω be a connection form in the principal G-bundle π : P →

M. Then there is a unique connection H in P whose connection form is ω , which is
given by H u = ker ω (u) for u ∈ P .
PROOF : We first show formula (i) in Definition 10.32. By Definition 10.34 (i) and
Proposition 10.26 we have that
w − σ (ω (u) (w)) u ∈ H u and σ (ω (u) (w)) u ∈ Vu
for every w ∈ Tu P , which shows that Tu P = H u + Vu . If w ∈ H u ∩ Vu , then w =

σ (X) u for an X ∈ g , where X = ω (u) (w) = 0 so that w = 0 . Hence we have that

Tu P = H u ⊕ Vu with projections hu : Tu P → Tu P and vu : Tu P → Tu P on H u and Vu
given by
hu (w) = w − σ (ω (u) (w)) u and vu (w) = σ (ω (u) (w)) u (1)
for w ∈ Tu P.
We next show that H is a distribution in P. Let ω j for j = 1, ..., m be the com-
ponents of ω with respect to a basis C = {X1 , ..., Xm } for g, and let u0 ∈ P. If
{Y1 , ...,Ym+n } is a local basis for T P on an open neighbourhood U of u0 , then the
vector fields {Z1 , ..., Zm+n } defined by
m m
Zi u = hu (Yi u ) = Yi u − σ ( ∑ ω j (u) (Yi u ) X j ) u = Yi u − ∑ ω j (u) (Yi u) σ (X j ) u
j=1 j=1
for i = 1, ..., m + n and u ∈ U , span H on U. Hence n of the vector fields Zi form a

local basis for H on an open neighbourhood W of u0 contained in U, showing that H
is a distribution in P.
To show formula (ii) in Definition 10.32, let w ∈ H u . Then we have that
ω (ug) ((R g )∗ w) = R g∗ ω (u) (w) = Ad (g−1 ) ω (u)(w) = 0
so that (R g )∗ w ∈ H ug . Since (R g )∗ is injective and dim (H ug ) = dim (H u ) , it fol-

lows that H ug = (R g )∗ H u , which completes the proof that H is a connection in P.
From the last formula in (1) and 10.33 we have that
ω (u) (w) = τu ◦ vu (w)
for w ∈ Tu P , showing that ω is the connection form of H.
10.37 Let π : P → M be a principal G-bundle, and let f : N → M be a smooth

map from a smooth manifold N. Then the induced bundle πe : Pe → N defined in 10.12
is also a principal G-bundle. The action of G on Pe on the right is given by
(p, u) g = (p, ug) (1)
for (p, u) ∈ Pe and g ∈ G, since
e
tα ,p ((p, u) g) = e
tα ,p (p, u) g = tα , f (p) (u) g = tα , f (p) (ug) = e
tα ,p (p, ug)
eα , u ∈ Pf (p) and g ∈ G by 10.12 and 10.16.

for p ∈ U
By applying the map fe : Pe → P defined in 10.12 on both sides of (1), we obtain
fe((p, u) g) = ug = fe(p, u) g
which implies that

fe◦ σe(p,u) = σu (2)
FIBRE BUNDLES 363
e (p,u) : G → Pe and σu : G → P are the maps defined by

where σ
σe (p,u) (g) = (p, u) g and σu (g) = u g .
We also have that

fe◦ Reg = Rg ◦ fe (3)
where Reg and Rg are the right translations by g on Pe and P, respectively.
Given a connection form ω on P, we will show that fe∗ ( ω ) is a connection form
e If X ∈ g, and if σ (X) and σe (X) are the fundamental vector fields determined
on P.
e respectively, it follows from (2) that
by X on P and P,
fe∗ σ
e (X) (p,u) = fe∗ σe (p,u) ∗ (X) = σu ∗ (X) = σ (X) u
so that
fe∗ ( ω )(p, u) (σe (X) (p,u) ) = ω (u) (σ (X) u ) = X .
By (3) and Proposition 5.48 we also have that
Re g∗ fe∗ ( ω ) = fe∗ (R g∗ ω ) = fe∗ (Ad (g−1 ) · ω ) = Ad (g−1 ) · fe∗ ( ω )
which completes the proof that fe∗ ( ω ) is a connection form on P.

e
e e
If H and H are the connections in P and P, respectively, with connection forms
ω and fe∗ ( ω ) , we have that
fe∗ ( ω ) (p, u) = ω (u) ◦ fe∗ (p,u)
which implies that

e (p,u) = ( fe∗ (p,u) ) −1 (H u )
H
for p ∈ N and u ∈ Pf (p) .
TENSORIAL FORMS
10.38 Definition Let π : P → M be a principal G-bundle and ρ : G → Aut (W ) a
representation of G on a finite dimensional vector space W . Then a W -valued k-form
ω ∈ Ω k (P;W ) is said to be pseudo-tensorial of type (ρ ,W ) if
R g∗ ω = ρ (g−1 ) · ω
for every g ∈ G. It is said to be horizontal if ω (u) (w1 , ..., wk ) = 0 whenever u ∈ P

and at least one wi ∈ Tu P is vertical. We say that ω is tensorial of type (ρ ,W ) if it is
both horizontal and pseudo-tensorial of type (ρ ,W ). A vector field X on P is said to
be vertical if Xu is a vertical vector in Tu P for every u ∈ P.
If H is a connection in P, then a vector-valued k-form ω on P is said to be vertical
if ω (u) (w1 , ..., wk ) = 0 whenever u ∈ P and at least one wi ∈ Tu P is horizontal. A

vector field X on P is said to be horizontal if Xu is a horizontal vector in Tu P for every
u ∈ P.
10.39 Proposition Let π : P → M be a principal G-bundle, and let ρi : G →

Aut (Wi ) be a representation of G on a finite dimensional vector space Wi for i = 1, 2 .
If F : W1 → W2 is a linear map satisfying
F ◦ ρ 1 (g) = ρ 2 (g) ◦ F
for every g ∈ G, and ω ∈ Ω k (P;W1 ) is a vector-valued k-form on P which is pseudo-

tensorial (tensorial) of type (ρ 1 ,W1 ) , then the form F · ω is pseudo-tensorial (tenso-
rial) of type (ρ 2 ,W2 ) .
PROOF : If ω is pseudo-tensorial of type (ρ1 ,W1 ) , then it follows from Proposition
5.48 that
R g∗ (F · ω ) = F · R g∗ ( ω ) = F · (ρ 1 (g−1 ) · ω ) = ρ 2 (g−1 ) · (F · ω )
for every g ∈ G, which shows that F · ω is pseudo-tensorial of type (ρ 2 ,W2 ) . If ω is

horizontal, so is F · ω .
10.40 Proposition Let H be a connection in the principal G-bundle π : P →

M. Then the maps h : T P → T P and v : T P → T P , where h(w) and v(w) are the
horizontal and vertical components of w for each w ∈ T P, are bundle maps over P
called the horizontal and vertical projection determined by H. We have that
(R g )∗ ◦ h = h ◦ (R g )∗ and (R g )∗ ◦ v = v ◦ (R g )∗
for every g ∈ G.
PROOF : By Proposition 2.64 we know that h and v are bundle maps over P. The
last part of the proposition follows from the relation
(R g )∗ (w) = (R g )∗ ◦ h (w) + (R g )∗ ◦ v (w)
where
(R g )∗ ◦ h (w) ∈ H ug and (R g )∗ ◦ v (w) ∈ V ug
for every w ∈ Tu P by Definition 10.32 and Corollary 10.31.
10.41 Proposition Let H be a connection in the principal G-bundle π : P → M,

and let h : T P → T P and v : T P → T P be the horizontal and vertical projection
determined by H. Then a W -valued k-form ω ∈ Ω k (P;W ) is horizontal if and only
if h∗ (ω ) = ω and vertical if and only if v∗ (ω ) = ω .
FIBRE BUNDLES 365
PROOF : Suppose that w1 , ... , wk ∈ Tu P, and let wi,1 = h(wi ) and wi,2 = v(wi ) for
i = 1, ... , k. Then we have that
2
ω (u)(w1 , ... , wk ) = ∑ ω (u)(w1,i1 , ... , wk,ik ) .
i1 ,...,ik =1
If ω is horizontal, it follows that
ω (u)(w1 , ... , wk ) = ω (u)(h(w1 ), ... , h(wk )) = h∗ ω (u)(w1 , ... , wk ) ,
and a similar argument applies when ω is vertical. The converse statements follow
directly from the definition.
COVARIANT DERIVATIVE OF FORMS ON PRINCIPAL

BUNDLES
10.42 Definition Let H be a connection in the principal G-bundle π : P → M
with horizontal projection h : T P → T P . Then the exterior covariant derivative of
a vector-valued k-form ω ∈ Ω k (P;W ) is the vector-valued (k + 1)-form D ω ∈
Ω k+1 (P;W ) defined by
D ω = h∗ (d ω ) ,
i.e.,
D ω (u) (w1 , ..., wk+1 ) = d ω (u) (h (w1 ), ..., h (wk+1 ))
for u ∈ P and w1 , ..., wk+1 ∈ Tu P.
10.43 Proposition Let π : P → M be a principal G-bundle with a connection H,

and let ρ : G → Aut (W ) be a representation of G on a finite dimensional vector space
W . If ω ∈ Ω k (P;W ) is a pseudo-tensorial k-form of type (ρ ,W ), then its exterior
covariant derivative D ω is a tensorial k + 1-form of type (ρ ,W ).
PROOF : If u ∈ P and w1 , ... , wk+1 ∈ Tu P, then
R g∗ (D ω ) (u) (w1 , ... , wk+1 ) = (D ω ) (ug) ((R g )∗ w1 , ... , (R g )∗ wk+1 )
= (d ω ) (ug) (h ◦ (R g )∗ w1 , ... , h ◦ (R g )∗ wk+1 )
= (d ω ) (ug) ((R g )∗ ◦ h (w1 ) , ... , (R g )∗ ◦ h (wk+1 ) )
= R g∗ (d ω ) (u) (h (w1 ), ... , h (wk+1 ) ) = d (R g∗ ω ) (u) (h (w1 ), ... , h (wk+1 ) )
= d (ρ (g−1 ) · ω ) (u) (h (w1 ), ... , h (wk+1 ) )
= ρ (g−1 ) · d ω (u) (h (w1 ), ... , h (wk+1 ) ) = ρ (g−1 ) · D ω (u) (w1 , ... , wk+1 ) ,
thereby proving that D ω is pseudo-tensorial of type (ρ ,W ).

If at least one wi ∈ Vu , then h (wi ) = 0 so that D ω (u) (w1 , ... , wk+1 ) = 0 , which
shows that D ω is horizontal.
THE CURVATURE FORM

10.44 Definition Let π : P → M be a principal G-bundle with a connection H,
and let ω be the connection form of H. Then the g-valued 2-form Ω = D ω is called
the curvature form of H. The connection H is said to be flat if Ω = 0 .
10.45 Remark From Proposition 10.35 it follows that the connection form ω is
a vertical pseudo-tensorial 1-form of type (Ad, g), and hence the curvature form Ω is
a tensorial 2-form of type (Ad, g).
10.46 Proposition Let π : P → M be a principal G-bundle over a 1-dimensional

smooth manifold M. Then every connection H in P is flat.
PROOF : If Ω is the curvature form of H, we must show that Ω(u)(w1 , w2 ) = 0 for
every u ∈ P and w1 , w2 ∈ Tu P. Since M is 1-dimensional, we know that dim(Hu ) = 1.
If {w} is a basis for Hu and h(wi ) = ci w for i = 1, 2 , we have that
Ω(u)(w1 , w2 ) = c1 c2 d ω (u)(w, w) = 0
since d ω (u) is skew symmetric.

and let ω and Ω be the connection form and the curvature form of H. Then we have
that
Ω(X,Y ) = d ω (h ◦ X, h ◦ Y ) = − ω ([ h ◦ X , h ◦ Y ] )
for any pair of vector fields X,Y on an open subset U of P.
PROOF : Follows from Remark 5.86 since ω (h ◦ X) = 0 and ω (h ◦ Y ) = 0 .
10.48 Corollary Let H be a connection in the principal G-bundle π : P → M, and

let Ω be the curvature form of H. Then H is an integrable distribution on P if and
only if Ω = 0.
PROOF : If H is integrable, then [ h ◦ X , h ◦ Y ] ∈ H which implies that Ω(X,Y ) = 0
for every pair of vector fields X,Y on P.
Conversely, assume that Ω = 0. For any pair of vector fields X,Y ∈ H de-
fined on some open subset U of P, it follows from Proposition 10.47 that [ X ,Y ] =
[ h ◦ X , h ◦ Y ] ∈ H , which shows that H is an integrable distribution on P.
FIBRE BUNDLES 367
HORIZONTAL LIFTS OF VECTOR FIELDS
10.49 Proposition Let π : P → M be a principal G-bundle with a connection

H, and let i : H → T P be the inclusion map which is a bundle map over P. Then
(π∗ ◦ i, π ) is a bundle map from H to T M such that α = π∗ ◦ i induces a linear
isomorphism between the fibres H u and Tp M for each u ∈ P, where p = π (u). If X is
a vector field on M, then the pull-back α ∗ (X) is a section of the bundle π ′ : H → P ,
and X ∗ = i ◦ α ∗ (X) is a horizontal vector field on P which is π -related to X and is
called the horizontal lift of X.
PROOF : If w ∈ H u with π ∗ u (w) = 0, then w ∈ H u ∩ Vu = {0} which shows that
α u is injective. Since dim (H u ) = dim (Tu P) − dim (Vu ) = dim (Tp M) by Definition
10.32 (i) and Proposition 10.3, it follows that α u is a linear isomorphism. The last
part of the propostion now follows from Proposition 2.57.
10.50 Proposition If π : P → M is a principal G-bundle with a connection H and

X ∗ is the horizontal lift of a vector field X on M, then
(R g )∗ X ∗ = X ∗
for every g ∈ G. Conversely, if Y is a horizontal vector field on P satisfying (R g )∗ Y =

Y for every g ∈ G, then Y is the horizontal lift of a unique vector field X on M.
PROOF : Since π ◦ R g = π , it follows that
π∗ ◦ (R g )∗ = π∗
for every g ∈ G, and by Definition 10.32 and Corollary 2.38 we have a bundle map
(rg , R g ) from the vector bundle π ′ : H → P to itself, where rg : H → H is the map
induced by (R g )∗ : T P → T P so that
i ◦ rg = (R g )∗ ◦ i
for every g ∈ G, where i : H → T P is the inclusion map.

Combining this, it follows that the bundle map (α , π ) from the vector bundle
π ′ : H → P to the tangent bundle πM : T M → M , where α = π∗ ◦ i, satisfies
α ◦ rg = π∗ ◦ i ◦ rg = π∗ ◦ (R g )∗ ◦ i = π∗ ◦ i = α
for every g ∈ G, which means that
α ug ◦ (rg ) u = α u
and therefore
(rg ) u ◦ α u−1 = α ug
−1
for every u ∈ P and g ∈ G.

From the definition of the pull-back α ∗ (X), we now have that
α ∗ (X) u = α u−1 (X p ) and α ∗ (X) ug = α ug
−1
(X p )
where p = π (u) = π (ug), so that
(rg ) u α ∗ (X) u = α ∗ (X) ug
for every u ∈ P and g ∈ G. Hence it follows that
rg ◦ α ∗ (X) = α ∗ (X) ◦ R g
which implies that
(R g )∗ ◦ X ∗ = (R g )∗ ◦ i ◦ α ∗ (X) = i ◦ rg ◦ α ∗ (X) = i ◦ α ∗ (X) ◦ R g = X ∗ ◦ R g
so that
(R g )∗ X ∗ = (R g )∗ ◦ X ∗ ◦ R −1
g =X
∗
for every g ∈ G, showing the first part of the proposition.

To prove that last part, suppose that Y is a horizontal vector field on P satisfying
(R g )∗ Y = Y for every g ∈ G. Then we have that
π∗ ◦ Y ◦ R g = π∗ ◦ (R g )∗ ◦ Y = π∗ ◦ Y
for every g ∈ G, which shows that π∗ ◦ Y is constant on each fibre Pp . Hence we have
a well-defined map X : M → T M satisfying π∗ ◦ Y = X ◦ π given by
X(p) = π∗ ◦ Y (u) ,
where u is any point in P with π (u) = p. If (t, π −1 (U)) is a local trivialization in P,
then
X(p) = π∗ ◦ Y ◦ t −1 (p, e)
for p ∈ U, and this completes the proof that X is a vector field on M which is π -related
to Y .
10.51 Proposition Let H be a connection in the principal G-bundle π : P → M

with horizontal projection h : T P → T P . If X ∗ and Y ∗ are the horizontal lifts of the
vector fields X and Y on M, then
(1) ( X + Y )∗ = X ∗ + Y ∗ ,
(2) ( f X)∗ = ( f ◦ π ) X ∗ for every smooth function f : M → R ,
(3) h ◦ [ X ∗ ,Y ∗ ] = [ X,Y ] ∗ .
PROOF : (1) X ∗ + Y ∗ is a horizontal vector field on P which is π -related to X + Y
since
π∗ ( X ∗u + Y ∗u ) = π∗ ( X ∗u ) + π∗ (Y u∗ ) = X π (u) + Y π (u)
for every u ∈ P.
FIBRE BUNDLES 369
(2) ( f ◦ π ) X ∗ is a horizontal vector field on P which is π -related to f X since
π∗ ( ( f ◦ π )(u) X ∗u ) = ( f ◦ π )(u) π∗ ( X ∗u ) = f (π (u)) X π (u)
for every u ∈ P.
(3) By Propositions 4.88 and 10.3 we know that [ X ∗ ,Y ∗ ] is π -related to [ X,Y ] ,
and the vertical component of [ X ∗ ,Y ∗ ] u lies in the kernel of π∗ u for every u ∈ P.
Hence we have that
π∗ ( h [ X ∗ ,Y ∗ ] u ) = π∗ ( [ X ∗ ,Y ∗ ] u ) = [ X,Y ] π (u)
for u ∈ P, which shows that h ◦ [ X ∗ ,Y ∗ ] is a horizontal vector field on P which is

π -related to [ X,Y ] .
LOCAL SECTIONS AND TRIVIALIZATIONS

10.52 Proposition Let s : V → P be a section in a principal G-bundle π : P → M
on an open subset V of M. Then the map ρ : V × G → π −1 (V ) defined by
ρ (p, g) = s(p) g
for p ∈ V and g ∈ G is a diffeomorphism, and we have that
ρ ∗ (ig ∗ (v) + i p ∗ (w)) = (R g ) ∗ ◦ s ∗ (v) + σ ( (Lg−1 ) ∗ w ) ρ (p,g) (1)
for p ∈ V , g ∈ G , v ∈ Tp M and w ∈ Tg G , where ig : V → V × G and i p : G → V × G

are the embeddings defined by ig (p ′ ) = (p ′ , g) and i p (g ′ ) = (p, g ′ ) for p ′ ∈ V and
g′ ∈ G.
Moreover, (ρ −1 , π −1 (V )) is a local trivialization in P, called the local trivializa-
tion associated with the section s, and
s(p) = ρ (p, e)
for p ∈ V .
Conversely, any local trivialization (t, π −1 (V )) in P is associated with a unique
section s on V which is given by
s(p) = t −1 (p, e)
for p ∈ V .
If sk : Vk → P for k = 1, 2 are two sections in P with associated local trivializa-
tions (tk , π −1 (Vk )) which are G-related with transition map φ : V1 ∩V2 → G, then
s1 (p) = s2 (p) φ (p)

for p ∈ V1 ∩V2 . If λ : V1 ∩V2 → V2 × G is the map defined by
λ (p) = (p, φ (p))
for p ∈ V1 ∩V2 , then

λ∗ (v) = iφ (p) ∗ (v) + i p ∗ ◦ φ∗ (v) (2)
and
s 1∗ (v) = (R φ (p) ) ∗ ◦ s 2∗ (v) + σ ( (L φ (p) −1 ) ∗ ◦ φ ∗ (v) ) s 1 (p) (3)
for p ∈ V1 ∩V2 and v ∈ Tp M .
PROOF : Let p ∈ V and g ∈ G . Since
ρg = R g ◦ s and ρ p = σρ (p,g) ◦ Lg−1 ,
we have that
ρg ∗ (v) = (R g ) ∗ ◦ s ∗ (v) and ρ p ∗ (w) = σ ( (Lg−1 ) ∗ w ) ρ (p,g)
for v ∈ Tp M and w ∈ Tg G , which proves formula (1).

We next show that ρ is a diffeomorphism, and that (ρ −1 , π −1 (V )) is a local triv-
ialization in P. Since π ◦ ρ = pr1 and G acts freely on P having the fibres of π as
orbits, we see that ρ is a bijection. Indeed, if u ∈ π −1 (V ) and π (u) = p, then both u
and s(p) lie in the fibre π −1 (p), and there is a unique g ∈ G with u = s(p) g = ρ (p, g).
If we denote this g by ψ (u), we obtain a map ψ : π −1 (V ) → G so that
ρ −1 (u) = (π (u), ψ (u))
for u ∈ π −1 (V ). As π ◦ Rg ◦ s = idV for each g ∈ G, we have that
v − (R g ) ∗ ◦ s ∗ ◦ π ∗ (v) ∈ ker π∗ u = Vu = σ u ∗ ( g ) = σ u ∗ ◦ (Lg−1 ) ∗ ( Tg G )
for every v ∈ Tu P where u = ρ (p, g), showing that ρ∗ u surjective. Since dim u (P) =
dim p (M) + dim g (G) , it follows from the inverse function theorem that ρ and ρ −1
are diffeomorphisms. For every u = ρ (p, g) ∈ π −1 (V ) and h ∈ G we have that uh =
ρ (p, gh) so that
ψ (uh) = gh = ψ (u) h ,
thus completing the proof that (ρ −1 , π −1 (V )) is a local trivialization in P.
Now, if (t, π −1 (V )) is any local trivialization in P, then the map s : V → P given
by
s(p) = t −1 (p, e)
for p ∈ V is clearly a section on V . From 10.16 it follows that
s(p) g = t −1 (p, g)
for p ∈ V and g ∈ G, which shows that (t, π −1 (V )) is the local trivialization associated
with s.
FIBRE BUNDLES 371
Finally, if (tk , π −1 (Vk )) for k = 1, 2 are two G-related local trivializations in P

with transition map φ : V1 ∩V2 → G associated with the sections sk , then
s1 (p) = t1−1 (p, e) = t2−1 ◦ t2 ◦ t1−1 (p, e) = t2−1 (p, φ (p) e) = s2 (p) φ (p)
for p ∈ V1 ∩V2 , and (2) follows from formula (4) in Proposition 2.74. Now we have
that
s1 (p) = ρ2 ◦ λ (p)
for p ∈ V1 ∩ V2 , where ρ2 : V2 × G → π −1(V2 ) is the map defined by ρ2 (p, g) =
s2 (p)g , and formula (3) therefore follows from (1) and (2).
10.53 Remark For each local chart (x,U) on the manifold M n , the local trivial-
ization (φx , π ′ −1 (U)) in the frame bundle π ′ : L (M) → M defined in Example 10.21
is associated with the frame field

∂ ∂
s= 1 , ... , n
∂x ∂x
on U by Remark 2.82.
10.54 Corollary Suppose that s1 = (X1 , ... , Xn ) and s2 = (X1′ , ... , Xn′ ) are frame
fields on open subsets V1 and V2 of a smooth manifold M, and that the associated local
trivializations (t1 , π −1 (V1 )) and (t2 , π −1 (V2 )) in the frame bundle π : L (M) → M are
Gl (Rn )-related with transition map φ : V1 ∩V2 → Gl (Rn ). Then we have that
n
X j (p) = ∑ φ (p) ij Xi′ (p) (1)
i=1
for p ∈ V1 ∩ V2 and 1 ≤ j ≤ n, where the components of φ (p) are with respect to

the basis D = {F ji |1 ≤ i, j ≤ n} for gl (Rn ) defined in Example 8.10 (b), where
F ji : Rn → Rn is the linear map given by F ji (ek ) = δ j k ei for k = 1, ... , n.
PROOF : By Example 10.21 and Proposition 10.52 we have that
!
Xk (p) = s 1 (p)(ek ) = s2 (p) ◦ φ (p)(ek ) = s2 (p) ∑ φ (p) ij F ji (ek )
ij
!
= s2 (p) ∑ φ (p) ik ei = ∑ φ (p) ik Xi′ (p)
i t
for p ∈ V1 ∩V2 and 1 ≤ k ≤ n.

on an open subset V of M, and let ρ : V × G → π −1 (V ) be the map defined by
ρ (p, g) = s(p) g
for p ∈ V and g ∈ G. If ω is a tensorial k-form on π −1 (V ) of type (ψ ,W ), where

ψ : G → Aut (W ) is a representation of G on a finite dimensional vector space W ,
e = s∗ (ω ) is the pull-back of ω by s, then
and if ω
ω (ρ (p, g))(ρg ∗ (v1 ) + ρ p ∗ (w1 ), ... , ρg ∗ (vk ) + ρ p ∗ (wk )) = ψ (g−1 ) ω

e (p)(v1 , ... , vk )
(1)
for p ∈ V , g ∈ G , v1 , ... , vk ∈ Tp M and w1 , ... , wk ∈ Tg G , where ρg = ρ ◦ ig and
ρp = ρ ◦ ip .
Conversely, given any W -valued k-form ω e on V , there is a unique tensorial k-form
ω on π −1 (V ) of type (ψ ,W ) with ω e = s∗ (ω ), which is given by (1).
tions (tk , π −1 (Vk )) which are G-related with transition map φ : V1 ∩ V2 → G, and if
ωe k = s k∗ (ω ) for a tensorial k-form ω on π −1 (V1 ∩V2 ) of type (ψ ,W ), then
e 1 (p)(v1 , ... , vk ) = ψ (φ (p) −1 ) ω

ω e 2 (p)(v1 , ... , vk ) (2)
for p ∈ V1 ∩V2 and v1 , ... , vk ∈ Tp M .

PROOF : By formula (1) in Proposition 10.52 we have that
ω (ρ (p, g)) ( ρg ∗ (v1 ), ... , ρg ∗ (vk )) = R g∗ ω (s(p))(s ∗ (v1 ), ... , s ∗ (vk ))
= ψ (g−1 ) ω (s(p))(s ∗ (v1 ), ... , s ∗ (vk )) = ψ (g−1 ) ω

e (p)(v1 , ... , vk )
and
ρ p ∗ (wi ) ∈ Vρ (p,g) for i = 1, ... , k
which implies (1).
Now let ωe be an arbitrary W -valued k-form on V , and let ω : π −1 (V ) →
k
Λ (T P ; W ) be the map defined by (1). To show that ω is smooth, we first consider
the map η : V × G → Λ k ( T (V × G) ; W ) given by
η (p, g) (ig ∗ (v1 ) + i p ∗ (w1 ), ... , ig ∗ (vk ) + i p ∗ (wk )) = ψ (g−1 ) ω

e (p)(v1 , ... , vk ) (3)
for p ∈ V , g ∈ G , v1 , ... , vk ∈ Tp M and w1 , ... , wk ∈ Tg G . If α : V × G → V is the

projection on the first factor and α̃ : V × G × W → V × W is the map given by
α̃ (p, g, ξ ) = (p, ψ (g) ξ )
for p ∈ V , g ∈ G and ξ ∈ W , we have a bundle map (α̃ , α ) between the vector bundles
π1 : V × G ×W → V × G and π2 : V ×W → V which is a linear isomorphism on each
fibre. Then it follows from Propositions 5.33 that η is a W -valued k-form on V × G
since
η = ρ ′1 ◦ (α ∗ , α̃ ) ∗ (ρ ′2−1 ◦ ω
e) ,
where ρ ′1 : Λ k (T (V × G) ;V × G × W ) → Λ k (T (V × G) ;W ) and ρ ′2 : Λ k (TV ; V ×
W ) → Λ k (TV ;W ) are the equivalences over V × G and V , respectively, defined in
5.26. Hence it follows that ω = (ρ −1 )∗ η is a W -valued k-form on π −1 (V ) .
FIBRE BUNDLES 373
e = s∗ (ω ).
It only remains to show that ω is tensorial of type (ψ ,W ) and that ω
Since
σρ (p,g) ∗ ◦ (L g−1 ) ∗ : Tg G → Vρ (p,g)
is a linear isomorphism by Proposition 10.26, it follows from (1) that ω is horizontal.
Indeed, if ρg ∗ (vr ) + ρ p ∗ (wr ) ∈ Vρ (p,g) for some r, then vr = 0 by formula (1) in
Proposition 10.52 so that ω e (p)(v1 , ... , vk ) = 0 . Furthermore, we have that
R h ◦ ρ (p, g) = ρ (p, gh)
which implies that
R h ◦ ρg = ρgh and R h ◦ ρ p = ρ p ◦ R h
so that
R h∗ ω (ρ (p, g)) ( ρg ∗ (v1 ) + ρ p ∗ (w1 ), ... , ρg ∗ (vk ) + ρ p ∗ (wk ))

= ω (ρ (p, gh)) ( ρgh ∗ (v1 ) + ρ p ∗ ◦ (R h ) ∗ (w1 ), ... , ρgh ∗ (vk ) + ρ p ∗ ◦ (R h ) ∗ (wk ))
= ψ ((gh)−1 ) ω
e (p)(v1 , ... , vk ) = ψ (h−1 ) ψ (g−1 ) ω
e (p)(v1 , ... , vk )
= ψ (h−1 ) ω (ρ (p, g)) ( ρg ∗ (v1 ) + ρ p ∗ (w1 ), ... , ρg ∗ (vk ) + ρ p ∗ (wk )) .
This shows that ω is a tensorial k-form of type (ψ ,W ) on π −1 (V ), and since

s(p) = ρ (p, e) = ρe (p)
for p ∈ V , we have that
s∗ (ω ) (p)(v1 , ... , vk ) = ω (ρ (p, e)) ( ρe ∗ (v1 ), ... , ρe ∗ (vk )) = ω
e (p)(v1 , ... , vk ) .
s1 (p) = ρ2 ◦ λ (p)
for p ∈ V1 ∩V2 , where λ is the map defined in Proposition 10.52 and ρ 2 : V2 × G →
π −1 (V2 ) is given by ρ 2 (p, g) = s2 (p)g . Hence it follows from (1) that
e 1 (p)(v1 , ... , vk ) = ω (s1 (p)) (s1 ∗ (v1 ), ... , s1 ∗ (vk ))

ω
= ω (ρ 2 (p, φ (p))) ( ρ 2 φ (p) ∗ (v1 ) + ρ 2 p ∗ ◦ φ∗ (v1 ), ... , ρ 2 φ (p) ∗ (vk ) + ρ 2 p ∗ ◦ φ∗ (vk ))
= ψ (φ (p) −1 ) ω
e 2 (p)(v1 , ... , vk )
for p ∈ V1 ∩V2 and v1 , ... , vk ∈ Tp M , which completes the proof of formula (2).

on an open subset V of M, and let ρ : V × G → π −1 (V ) be the map defined by
ρ (p, g) = s(p) g
e = s∗ (ω ) is the
for p ∈ V and g ∈ G. If ω is a connection form on π −1 (V ), and if ω
pull-back of ω by s, then
ω (ρ (p, g)) ( ρg ∗ (v) + ρ p ∗ (w)) = Ad (g−1 ) ω

e (p)(v) + θ (g)(w) (1)
for p ∈ V , g ∈ G, v ∈ Tp M and w ∈ Tg G, where ρg = ρ ◦ ig and ρ p = ρ ◦ i p , and

where θ is the canonical Maurer-Cartan form on G.
Conversely, given any g-valued 1-form ω e on V , there is a unique connection form
−1 e ∗
ω on π (V ) with ω = s (ω ), which is given by (1).
tions (tk , π −1 (Vk )) which are G-related with transition map φ : V1 ∩ V2 → G, and if
ωe k = s k∗ (ω ) for a connection form ω on π −1(V1 ∩V2 ), then
ω e 2 (p)(v) + θe (p)(v)
e 1 (p)(v) = Ad (φ (p) −1 ) ω (2)
for p ∈ V1 ∩ V2 and v ∈ Tp M , where θe = φ ∗ (θ ) is the pull-back of the canonical

Maurer-Cartan form θ on G by φ .
PROOF : By formula (1) in Proposition 10.52 we have that
ω (ρ (p, g)) ( ρg ∗ (v)) = R g∗ ω (s(p))(s ∗ (v))
= Ad (g−1 ) ω (s(p))(s ∗ (v)) = Ad (g−1 ) ω

e (p)(v)
and
ω (ρ (p, g)) (ρ p ∗ (w)) = (Lg−1 ) ∗ (w) = θ (g)(w)
which implies (1).
Now let ωe be an arbitrary g-valued 1-form on V , and let ω : π −1 (V ) →
1
Λ (T P ; g) be the map defined by (1). To show that ω is smooth, we first consider
the map η : V × G → Λ 1 ( T (V × G) ; g) given by
η (p, g) (ig ∗ (v) + i p ∗ (w)) = Ad (g−1 ) ω

e (p)(v) + θ (g)(w) (3)
for p ∈ V , g ∈ G, v ∈ Tp M and w ∈ Tg G. If α : V × G → V and γ : V × G → G are

the projections on the first and second factor, and α̃ : V × G × g → V × g is the map
given by
α̃ (p, g, X) = (p, Ad (g) X)
for p ∈ V , g ∈ G and X ∈ g, we have a bundle map (α̃ , α ) between the vector bundles
π1 : V × G × g → V × G and π2 : V × g → V which is a linear isomorphism on each
fibre. Then it follows from Propositions 5.33 and 5.47 that η is a g-valued 1-form
on V × G since
η = ρ ′1 ◦ (α ∗ , α̃ ) ∗ (ρ ′2−1 ◦ ω
e ) + γ ∗ (θ ) ,
where ρ ′1 : Λ k (T (V × G) ;V × G × g) → Λ k (T (V × G) ; g) and ρ ′2 : Λ k (TV ; V ×
g) → Λ k (TV ; g) are the equivalences over V × G and V , respectively, defined in
5.26. Hence it follows that ω = (ρ −1 )∗ η is a g-valued 1-form on π −1 (V ) .
FIBRE BUNDLES 375
e = s∗ (ω ).
It only remains to show that ω is a connection form on π −1 (V ) with ω
If h ∈ G, we have that
σρ (p,g) (h) = ρ (p, gh) = ρ p ◦ L g (h)
which implies that
σ (X) ρ (p,g) = σρ (p,g) ∗ (X) = ρ p ∗ ◦ (L g ) ∗ (X)
so that
ω (ρ (p, g)) (σ (X) ρ (p,g)) = θ (g) ((L g ) ∗ (X)) = X
by Definition 8.75. Furthermore, we have that
R h ◦ ρ (p, g) = ρ (p, gh)
which implies that
R h ◦ ρg = ρgh and R h ◦ ρ p = ρ p ◦ R h
so that
R h∗ ω (ρ (p, g)) ( ρg ∗ (v) + ρ p ∗ (w))

= ω (ρ (p, gh)) ( ρgh ∗ (v) + ρ p ∗ ◦ (R h ) ∗ (w))
= Ad ((gh)−1 ) ω
e (p)(v) + θ (gh) ((R h ) ∗ (w))
e (p)(v) + R h∗ θ (g)(w)
= Ad ((gh)−1 ) ω
= Ad (h−1 ) {Ad (g−1 ) ω
e (p)(v) + θ (g)(w)}
= Ad (h−1 ) ω (ρ (p, g)) ( ρg ∗ (v) + ρ p ∗ (w))
by Proposition 9.15. This shows that ω is a connection form on π −1 (V ), and since
s(p) = ρ (p, e) = ρe (p)
for p ∈ V , we have that
s∗ (ω ) (p)(v) = ω (ρ (p, e)) ( ρe ∗ (v)) = ω

e (p)(v) .

s1 (p) = ρ2 ◦ λ (p)
for p ∈ V1 ∩V2 , where λ is the map defined in Proposition 10.52 and ρ 2 : V2 × G →

π −1 (V2 ) is given by ρ 2 (p, g) = s2 (p)g . Hence it follows from (1) that
e 1 (p)(v) = ω (s1 (p)) (s1 ∗ (v))

ω
= ω (ρ 2 (p, φ (p))) ( ρ 2 φ (p) ∗ (v) + ρ 2 p ∗ ◦ φ∗ (v))
= Ad (φ (p) −1 ) ω
e 2 (p)(v) + θ (φ (p))(φ∗ (v))
e 2 (p)(v) + θe (p)(v)
= Ad (φ (p) −1 ) ω
for p ∈ V1 ∩V2 and v ∈ Tp M , which completes the proof of formula (2).
10.57 Definition If π : P → M is a principal G-bundle with a connection H, then

a smooth map F : N → P from a smooth manifold N is said to be horizontal at a
point p ∈ N if F∗ (Tp N) ⊂ HF(p) . We say that F is horizontal if it is horizontal at
every point p ∈ N.
10.58 Remark If ω is the connection form of H, then the map F : N → P is

horizontal at p ∈ N if and only if F ∗ ( ω )(p) = 0 , since H F(p) = ker ω (F(p)) .
If F is a lifting of a smooth map f : N → M , then F is horizontal with respect to
ω if and only if the section Fe : N → Ee in the induced fibre bundle determined by F
is horizontal with respect to fe∗ ( ω ) , since F ∗ ( ω ) = Fe ∗ ( fe∗ ( ω )) .

and let ω be the connection form of H. If s : V → P is a horizontal section of π on
an open subset V of M, and ρ : V × G → π −1 (V ) is the map defined by
ρ (p, g) = s(p) g
for p ∈ V and g ∈ G, then
ω (ρ (p, g)) ( ρg ∗ (v) + ρ p ∗ (w)) = θ (g)(w) (1)
for p ∈ V , g ∈ G , v ∈ Tp M and w ∈ Tg G , where ρg = ρ ◦ ig and ρ p = ρ ◦ i p , and

where θ is the canonical Maurer-Cartan form on G. Moreover, we have that Hρ (p,g) =
ρg ∗ (Tp M) for (p, g) ∈ V × G.
tions (tk , π −1 (Vk )) which are G-related with transition map φ : V1 ∩V2 → G, then
s1 (p) = s2 (p) φ (p)
for p ∈ V1 ∩V2 . Assuming that s2 is horizontal on V1 ∩V2 , we have that s1 is horizontal

on V1 ∩ V2 if and only if φ is constant on each connected component of V1 ∩ V2 . In
particular, if the sections s1 and s2 are horizontal on V1 ∩V2 and s1 (p 0 ) = s2 (p 0 ) for
a point p 0 in V1 ∩ V2 , then s1 and s2 coincide on the connected component of p 0 in
V1 ∩V2 .
FIBRE BUNDLES 377
PROOF : Formula (1) follows from Proposition 10.56 since s∗ (ω ) = 0, and it implies
that
Hρ (p,g) = ker ω (ρ (p, g)) = ρg ∗ (Tp M)
for (p, g) ∈ V × G, thus proving the first part of the proposition.
By formula (2) in Proposition 10.56 we have that
s 1∗ (ω )(p)(v) = φ ∗(θ ) (p)(v) = θ (φ (p))(φ ∗ (v)) = (Lφ (p)−1 ) ∗ ◦ φ ∗ (v)
for p ∈ V1 ∩V2 and v ∈ Tp M , which shows that s1 is horizontal on V1 ∩V2 if and only
if φ ∗ = 0. The proposition now follows from Proposition 2.29 and Proposition 13.38
in the appendix.
HORIZONTAL LIFTS OF CURVES

and let β : I → P be a horizontal lifting of a smooth curve γ : I → M defined on an
open interval I. Then a smooth curve α : I → P is a horizontal lifting of γ if and only
if
α (t) = β (t) g
for t ∈ I where g ∈ G.
PROOF : Let πe : Pe → I be the principal G-bundle induced by γ with canonical bundle
e : I → Pe and βe : I → Pe are the sections of πe determined by the liftings
map (γe, γ ). If α
α and β of γ , respectively, it follows from Remark 10.58 and Proposition 10.59 that
α is horizontal if and only if there is a group element g ∈ G such that
e (t) = βe(t) g
α
for t ∈ I. By applying γe on both sides and using formula (3) in 10.37, we see that
this is equivalent to the assertion that
α (t) = β (t) g
for t ∈ I.
10.61 Proposition Let H be a flat connection in a principal G-bundle π : P → M n ,

and let u0 ∈ P and p 0 = π (u0 ). Then there is a horizontal section s : V → P of π
defined on some open neighbourhood V of p 0 with s(p 0 ) = u0 .
PROOF : By Corollary 10.48 we know that H is an integrable distribution on P. Let
N be an integral manifold of H passing through u0 . If i : N → P is the inclusion map,
it follows from Proposition 10.49 that π ◦ i : N → M is an immersion of rank n. By
the rank theorem it maps an open neighbourhood U of u0 in N diffeomorpically onto

an open neighbourhood V of p 0 in M. If σ : V → U is the inverse of the map induced
by π ◦ i and s = i ◦ σ , we have that s(p 0 ) = u0 and s∗ (Tp M) = i∗ (Tσ (p) N) = Hs(p)
for every p ∈ V .

form ω , and let γ : I → M be a smooth curve defined on an open interval I. If t0 ∈ I
and u0 ∈ π −1 (γ (t0 )), then there is a unique horizontal lifting α : I → P of γ with
α (t0 ) = u0 .
PROOF : Let πe : Pe → I be the principal G-bundle induced by γ with canonical bundle
map (γe, γ ). By 10.37 and Proposition 10.46 we know that γe ∗ ( ω ) is a connection
form on Pe corresponding to a flat connection since the base manifold I in the induced
bundle is 1-dimensional.
Let J be the set of all open subintervals J ′ of I containing t0 such that there
S
exists a horizontal section s′ : J ′ → Pe of πe with s′ (t0 ) = (t0 , u0 ), and put J = J′ ∈J J ′ .
Define the horizontal section s : J → Pe as follows. If t ∈ J, choose an interval J ′ in
J containing t and set s(t) = s′ (t). It follows by Propositions 10.61 and 10.59 that
J is nonempty and that s is well defined.
Now we contend that J = I. Otherwise, if an endpoint t1 of J is contained in I,
there is by Propositions 10.61 a horizontal section s′′ : J ′′ → Pe of πe defined on an
open subinterval J ′′ of I containing t1 . If t2 ∈ J ∩ J ′′ and s(t2 ) = s′′ (t2 ) g , we have a
horizontal section s′ : J ∪ J ′′ → Pe defined by

′ s(t) for t ∈ J
s (t) =
s′′ (t) g for t ∈ J ′′
contradicting the assumption that t1 is an endpoint of J. This shows that there is a

unique horizontal section s : I → Pe of πe with s(t0 ) = (t0 , u0 ), and the result now
follows from 10.12 and Remark 10.58 with α = γe ◦ s .
10.63 Definition Let M be a smooth manifold, and let p 0 , p1 ∈ M. By a curve

in M from p 0 to p1 we mean a continuous map γ : [ a, b] → M with γ (a) = p 0 and
γ (b) = p1 . The curve γ is said to be smooth if it has an extension to a smooth curve
γ 0 : I → M defined on an open interval I containing [ a, b]. It is called piecewise
smooth if there is a partition a = t0 < ... < tr = b of [ a, b] such that γ |[ ti ,ti+1 ] is a
smooth curve for each i = 0, ... , r − 1.
We define the inverse curve γ −1 : [ a, b] → M from p1 to p 0 by γ −1 (t) = γ (a +
b − t) for t ∈ [ a, b], and we see that γ −1 is smooth or piecewise smooth if and only
if the same is true for γ .
10.64 Proposition If two points p 0 and p1 on a smooth manifold M can be joined

by a continuous curve c : [ a, b] → M, then they can also be joined by a piecewise
smooth curve γ : [ a, b] → M.
FIBRE BUNDLES 379
PROOF : Let A = {(xα ,Uα ) | α ∈ A} be an atlas on M where xα (Uα ) is convex for

every α ∈ A, and let ε be the Lebesgue number of the covering {c−1 (Uα ) | α ∈ A}
of [ a, b]. If a = t0 < ... < tr = b is a partition of [ a, b] with mesh less that 2 ε , then
c([ti , ti+1 ]) is contained in a coordinate neighbourhood Uαi for each i = 0, ... , r − 1,
and we define

−1 ti+1 − t t − ti
γ (t) = xαi xαi ◦ c (ti ) + xαi ◦ c (ti+1 )
ti+1 − ti ti+1 − ti
for t ∈ [ti , ti+1 ].
10.65 Definition Let M be a pseudo-Riemannian manifold. The arc length of a

piecewise smooth curve γ : [ a, b] → M is defined by
Z b
L(γ ) = kγ ′ (t)k dt .
a
If γ : I → M is a smooth curve defined on an open or closed interval I with kγ ′ (t)k> 0

for t ∈ I, we define its arc length function s : I → R with initial point t0 ∈ I by
Z t
s(t) = kγ ′ (u)k du
t0
for t ∈ I. The curve α = γ ◦ s−1 : J → M defined on the interval J = s(I) is a

reparametrization of γ , called a parametrization by arc length , satisfying kα ′ (t)k= 1
for t ∈ J.
10.66 Definition Let π : P → M be a principal G-bundle with a connection H.

Then a smooth curve γ : [ a, b] → P is said to be horizontal if it has an extension to
a smooth curve γ 0 : I → P defined on an open interval I containing [ a, b] which is
horizontal on [ a, b]. A piecewise smooth curve γ : [ a, b] → P is horizontal if there
is a partition a = t0 < ... < tr = b of [ a, b] such that γ |[ ti ,ti+1 ] is a horizontal smooth
curve for each i = 0, ... , r − 1.

and let β : [ a, b] → P be a horizontal lifting of a piecewise smooth curve γ : [ a, b] →
M. Then a piecewise smooth curve α : [ a, b] → P is a horizontal lifting of γ if and
only if
α (t) = β (t) g (1)
for t ∈ [ a, b] where g ∈ G.
PROOF : Let a = t0 < ... < tr = b be a partition of [ a, b] such that α | [ ti ,ti+1 ] and
β |[ ti ,ti+1 ] are smooth liftings of γ |[ ti ,ti+1 ] for i = 0, ... , r − 1. By Proposition 10.60 we
know that α is horizontal on <ti , ti+1 > if and only if
α (t) = β (t) gi (2)

for t ∈< ti , ti+1 > where gi ∈ G. If α is horizontal, formula (2) actually holds for
t ∈ [ti , ti+1 ] by the continuity of α and β , and it follows by induction that gi = g0 for
i = 1, ... , r − 1 . Hence formula (1) is true for t ∈ [ a, b] with g = g0 .
Conversely, assuming that formula (1) is true we have that α is horizontal on
[ti , ti+1 ] by the continuity of αi∗ (ω ) , where α i : Ii → P is an extension of α |[ ti ,ti+1 ]
to a smooth curve defined on an open interval Ii containing [ti , ti+1 ], and ω is the
connection form of H.

and let γ : [ a, b] → M be a piecewise smooth curve in M. If u0 ∈ π −1 (γ (a)), then there
is a unique piecewise smooth horizontal lifting α : [ a, b] → P of γ with α (a) = u0 .
PROOF : Let a = t0 < ... < tr = b be a partition of [ a, b] such that γ | [ ti ,ti+1 ] is a
smooth curve for each i = 0, ... , r − 1. If γ i : Ii → M is an extension of γ |[ ti ,ti+1 ] to a
smooth curve defined on an open interval Ii containing [ti , ti+1 ], then it follows from
Proposition 10.62 that there are horizontal liftings α i : Ii → P of γ i for i = 0, ... , r − 1
with α0 (t0 ) = u0 and αi (ti ) = αi−1 (ti ) for i = 1, ... , r − 1. Then we have a piecewise
smooth horizontal lifting α : [ a, b] → P of γ with α (a) = u0 defined by α (t) = αi (t)
for t ∈ [ti , ti+1 ], thus completing the existence part of the proposition. The uniqueness
part follows from Proposition 10.67.
PARALLEL TRANSPORT

and let π ′ : E → M be the fibre bundle associated with P with fibre F obtained from a
Lie group homomorphism ψ : G → G′ into a Lie group G′ which acts on the smooth
manifold F effectively on the left. For each piecewise smooth curve γ : [ a, b] → M
in M from p 0 to p1 we have a map
−1 −1
τγ : π ′ (p 0 ) → π ′ (p1 ) ,
called parallel transport or parallel translation along γ , defined for each u ∈

π −1 (p 0 ) and v ∈ F by
τγ ([ u, v]) = [ β (b), v]
where β : [ a, b] → P is the horizontal lifting of γ with β (a) = u.
10.70 Remark We must show that τγ is well defined, i.e., that
τγ ([ ug, ψ (g−1 ) v]) = τγ ([ u, v])
for every g ∈ G. If β : [ a, b] → P is the horizontal lifting of γ with β (a) = u, then it

FIBRE BUNDLES 381
follows from Proposition 10.67 that α = Rg ◦ β is the unique horizontal lifting of γ

with α (a) = Rg (u) so that
τγ ([ ug, ψ (g−1 ) v]) = [ α (b), ψ (g−1 ) v] = [ β (b) g, ψ (g−1 ) v] = [ β (b), v] = τγ ([ u, v]) .

We see that τγ is a diffeomorphism since
τγ = µ β (b) ◦ µ −1
β (a)
,
for any horizontal lift β of γ , where µ β (a) : F → E p 0 and µ β (b) : F → E p 1 are

the diffeomorphisms defined in 10.28. As β −1 is the horizontal lifting of γ −1 with
β −1 (a) = β (b) and β −1 (b) = β (a) = u , we have that
τγ −1 ◦ τγ ([ u, v]) = [ β −1 (b), v] = [ u, v]
so that
τγ−1 = τγ −1 .
In the case where ψ = id G and F = G which is acting on itself by left translation,
the fibre bundle π ′ : E → M may be identified with the principal fibre bundle π : P →
M as in Example 10.29 (a). The parallel transport τγ : π ′ −1 (p 0 ) → π ′ −1 (p1 ) along
γ in E is then identified with the parallel transport
τeγ = e p 1 ◦ τγ ◦ e−1 −1 −1
p 0 : π (p 0 ) → π (p1 )
along γ in P which is given by

τeγ (u) = β (b) ,
where β : [ a, b] → P is the horizontal lifting of γ with β (a) = u.

and let π ′ : E → M be the fibre bundle associated with P with fibre F obtained from a
Lie group homomorphism ψ : G → G′ into a Lie group G′ which acts on the smooth
manifold F effectively on the left. For each smooth curve γ : I → M in M defined on
an open interval I and each t0 ,t1 ∈ I we have a map
−1 −1
τ tt01 : π ′ (γ (t0 )) → π ′ (γ (t1 )) ,
called parallel transport or parallel translation along γ from t0 to t1 , defined for
each u ∈ π −1(γ (t0 )) and v ∈ F by
τ tt01 ([ u, v]) = [ β (t1 ), v]
where β : I → P is the horizontal lifting of γ with β (t0 ) = u.
10.72 Remark By Propositions 10.60 and 10.62 it follows in the same way as in
t
Remark 10.70 that τ t01 is a well-defined diffeomorphism with inverse
( τ tt01 ) −1 = τ tt10 ,
and we have that

τ tt01 = µ β (t ) ◦ µ −1
β (t
1 0)
for any horizontal lift β of γ , where µ β (t )

: F → Eγ (t 0 ) and µ β (t ) : F → Eγ (t1 ) are
0 1
the diffeomorphisms defined in 10.28. It also follows from the definition that
τ tt12 ◦ τ tt01 = τ tt02
for all t0 ,t1 ,t2 ∈ I.
In the case where ψ = id G and F = G which is acting on itself by left translation,
the fibre bundle π ′ : E → M may be identified with the principal fibre bundle π : P →
M as in Example 10.29 (a). The parallel transport τ t01 : π ′ −1 (γ (t0 )) → π ′ −1 (γ (t1 ))
t
along γ in E from t0 to t1 is then identified with the parallel transport

τe tt01 = e γ (t1 ) ◦ τ tt01 ◦ e −1 −1 −1
γ (t0 ) : π (γ (t0 )) → π (γ (t1 ))
along γ in P from t0 to t1 which is given by

τe tt01 (u) = β (t1 ) ,
where β : I → P is the horizontal lifting of γ with β (t0 ) = u.
FORMS IN ASSOCIATED BUNDLES

10.73 Proposition Let π : P → M be a principal G-bundle, and let π ′ : E → M
be the associated vector bundle obtained from a representation ψ : G → Aut (W ) of
G on a finite dimensional vector space W as defined in Example 10.29 (b). For each
u ∈ P , let µ u : W → E p where π (u) = p be the linear isomorphism given by
µ u (v) = [ u, v]
for v ∈ W . If ω is a bundle-valued k-form on M with values in E, then we have a
vector-valued k-form ω on P with values in W given by
ω (u)(w1 , ... , wk ) = µ u−1 ◦ ω (π (u))(π∗ (w1 ), ... , π∗ (wk )) (1)
for u ∈ P and w1 , ... , wk ∈ Tu P, which is tensorial of type (ψ ,W ). Conversely, given a
tensorial W -valued k-form η on P of type (ψ ,W ), there is a unique E-valued k-form
ω on M with η = ω , which is given by
ω (p)(v1 , ... , vk ) = µ u ◦ η (u)(w1 , ... , wk ) (2)
for p ∈ M and v1 , ... , vk ∈ Tp M, where u ∈ P and w1 , ... , wk ∈ Tu P are chosen so that
π (u) = p and π∗ (wi ) = vi for i = 1, ... , k. If s : V → P is a section of π : P → M on
an open subset V of M, and η e = s∗ (η ) is the pull-back of η by s, then
e (p)
ω (p) = µ s(p) ◦ η
for p ∈ V .
FIBRE BUNDLES 383
PROOF : Using the bundle map (p ′ , π ) between the vector bundles pr1 : P ×W → P
and π ′ : E → M defined in Example 10.29 (b), which is a linear isomorphism on each
fibre, it follows from Proposition 5.33 that the pull-back (π∗ , p ′ ) ∗ (ω ) is a bundle-
valued k-form on P with values in P × W . Now we have that
ω = ρ ′ ◦ (π∗ , p ′ ) ∗ (ω ) ,
where ρ ′ is the equivalence over P between the vector bundles Λ k (T P ; P ×W ) → P

and Λ k (T P ;W ) → P defined as in 5.26 by
ρ ′ (α ) = ρu ◦ α
for α ∈ Λ k (Tu P ; {u} ×W ) , where ρu : {u} ×W → W is the projection on the second

factor so that
−1
µ u−1 = ρu ◦ p u′
for each u ∈ P. Hence it follows that ω is a vector-valued k-form on P with values in
W which is clearly horizontal since Vu = ker π∗ u by Proposition 10.3.
To show that it is tensorial of type (ψ ,W ), we use that
µ u (v) = [ u, v] = [ u g, ψ (g−1 ) v ] = µ ug ◦ ψ (g−1 )(v)
for u ∈ P, v ∈ W and g ∈ G so that
µ u = µ ug ◦ ψ (g−1 ). (3)
Therefore we have that

−1
µ ug = ψ (g−1 ) ◦ µ u−1
which implies that
R g∗ ω (u) (w1 , ... , wk ) = ω (ug) ((R g )∗ w1 , ... , (R g )∗ wk )

−1 ◦ ω (π (u))(π (w )), ... , π (w )) = ψ (g−1 ) · ω (u) (w , ... , w )
µ ug ∗ 1 ∗ k 1 k
for u ∈ P and w1 , ... , wk ∈ Tu P.

Conversely, assume that η is a W -valued k-form on P which is tensorial of type
(ψ ,W ), and let ω : M → Λ k (T M ; E ) be the map defined by (2). We must first show
that ω is well defined and does not depend on the choice of u and w1 , ... , wk . If wi is
replaced by another element w′i ∈ Tu P with π∗ (w′i ) = vi , then wi − w′i ∈ ker π∗ u = Vu
which shows that ω (p)(v1 , ... , vk ) is unchanged since η is horizontal. Moreover, if
u is replaced by another element u′ ∈ π −1 (p), then u′ = ug for a g ∈ G. Using (3) we
therefore have that
µ u ◦ η (u)(w1 , ... , wk ) = µ ug ◦ ψ (g−1 ) ◦ η (u)(w1 , ... , wk )

= µ ug ◦ R g∗ η (u) (w1 , ... , wk ) = µ ug ◦ η (ug) ((R g )∗ w1 , ... , (R g )∗ wk )
= µ u′ ◦ η (u′ )(w1 , ... , wk )
since η is tensorial of type (ψ ,W ), thus completing the proof that ω is well defined.
To show that ω is an E-valued k-form on M, let s : V → P be a section in P de-
fined on an open subset V of M with associated local trivialization (t, π −1 (V )), and
let (t ′ , π ′ −1 (V )) be the associated local trivialization in E. Then t ′ is an equivalence
over V between the vector bundles π V′ : π ′ −1 (V ) → V and pr1 : V × W → V , and it
follows from formula (2) in 10.28 that
ω (p)(v1 , ... , vk ) = [ s(p), η (s(p))(s∗ (v1 ), ... , s∗ (vk ))]

= [ s(p), s∗ (η )(p)(v1 , ... , vk )] = t ′ −1 ( p, s∗ (η )(p)(v1 , ... , vk ))
for p ∈ V and v1 , ... , vk ∈ Tp M . Hence we have that
ω |V = (id V ∗ ,t ′ ) ∗ (ρ ′ −1 ◦ s∗ (η ))
where ρ ′ : Λ k (T M ;V × W ) → Λ k (T M ;W ) is the equivalence over V defined in

5.26, and the result therefore follows by Proposition 5.33.
10.74 Proposition Let π : P → M be a principal G-bundle, and let π ′ : E → M

be the associated vector bundle obtained from a representation ψ : G → Aut (W ) of
G on a finite dimensional vector space W . If ω is a bundle-valued k-form on M with
values in E and Xi∗ is the horizontal lift of the vector field Xi on M for i = 1, ... , k,
then we have that
ω ( X1∗ , ... , Xk∗ ) = ω ( X1 , ... , Xk ) .
PROOF : For every u ∈ P we have that
ω ( X1∗ , ... , Xk∗ )(u) = ω (u)( X1,u

∗ , ... , X ∗ )
k,u
= µ u−1 ◦ ω (π (u))( π∗ (X1,u

∗ ), ... , π (X ∗ ))
∗ k,u
= µ u−1 ◦ ω (π (u)) ( X1,π (u) , ... , Xk,π (u) )
= µ u−1 ◦ ω ( X1 , ... , Xk )(π (u)) = ω ( X1 , ... , Xk )(u) .

and let π ′ : E → M be the associated vector bundle obtained from a representation
ψ : G → Aut (W ) of G on a finite dimensional vector space W . Then the exterior
covariant derivative of a bundle-valued k-form ω ∈ Ω k (M; E) is the bundle-valued
(k + 1)-form D ω ∈ Ω k+1 (M; E) which is obtained from the vector-valued k + 1-form
D ω ∈ Ω k+1 (P;W ) as in Proposition 10.73, i.e.,
Dω = Dω
or
D ω (p)(v1 , ... , vk+1 ) = µu ◦ d ω (u)(w1 , ... , wk+1 )
FIBRE BUNDLES 385
for p ∈ M and v1 , ... , vk+1 ∈ Tp M, where u ∈ P and w1 , ... , wk+1 ∈ Hu are chosen so
that π (u) = p and π∗ (wi ) = vi for i = 1, ... , k + 1.
10.76 Let π : P → M be a principal G-bundle, and let π ′ i : Ei → M be the as-

sociated vector bundle obtained from a representation ψi : G → Aut (Wi ) of G on a
finite dimensional vector space Wi for i = 1, 2, 3. Suppose that ν : W1 × W2 → W3 is
a bilinear map such that
ψ3 (g) ◦ ν = ν ◦ (ψ1 (g) × ψ2(g))
for every g ∈ G. Then we have a bilinear map ν p : E1,p × E2,p → E3,p for each p ∈ M
defined by
ν p ( [ u, v1 ], [ u, v2 ]) = [ u, ν ( v1 , v2 )]
for u ∈ π −1 (p), v1 ∈ W1 and v2 ∈ W2 . It is well defined since
ν p ( [ u g, ψ1 (g−1 ) v1 ], [ u g, ψ2 (g−1 ) v2 ] ) = [ u g, ν ( ψ1 (g−1 ) v1 , ψ2 (g−1 ) v2 )]

= [ u g, ψ3 (g−1 ) ν (v1 , v2 )] = [ u, ν ( v1 , v2 )]
for every g ∈ G. If µ i,u : Wi → Ei,p where π (u) = p is the linear isomorphism defined
by µ i,u (v) = [ u, v] for v ∈ Wi , we have that
µ 3,u ◦ ν = ν p ◦ (µ 1,u × µ 2,u ) (1)
for each u ∈ P.
Now if ω ∈ Ω k (V ; E1 ) and η ∈ Ω l (V ; E2 ) are bundle-valued differential forms
on an open subset V of M, we have a bundle-valued differential form ω ∧ ν η ∈
Ωk+l (V ; E3 ) , called the wedge product of ω and η with respect to ν , defined by
ω ∧ν η = ω ∧ν η
or
( ω ∧ ν η )(p)(v1 , ... , vk+l ) = µ 3,u ◦ ( ω ∧ ν η )(u)(w1 , ... , wk+1 ) (2)
for p ∈ V and v1 , ... , vk+l ∈ Tp M, where u ∈ P and w1 , ... , wk+l ∈ Tu P are chosen so
that π (u) = p and π∗ (wi ) = vi for i = 1, ... , k + 1. Using (1) we have that
µ 3,u ◦ ( ω (u) ∧ ν η (u)) = µ 1,u ◦ ω (u) ∧ ν p µ 2,u ◦ η (u)
which together with (2) shows that
( ω ∧ ν η )(p) = ω (p) ∧ ν p η (p)
for p ∈ V . Using Proposition 5.69 and Definition 10.75 of the exterior covariant
derivative, we have that
D (ω ∧ ν η ) = (D ω ) ∧ ν η + (−1)k ω ∧ ν (D η )
We will often omit the reference to ν in the notation if it is clear from the context.
COVARIANT DERIVATIVE OF SECTIONS IN ASSOCIATED

VECTOR BUNDLES
ψ : G → Aut (W ) of G on a finite dimensional vector space W . If s ∈ Γ(M; E) is a
section of π ′ , then the vector-valued function s ∈ F (P ;W ) is given by
s (u) = µ u−1 ◦ s (π (u))
for u ∈ P. The exterior covariant derivative of s is the bundle-valued 1-form Ds ∈

Ω 1 (M; E) given by
Ds (p)(v) = µu ◦ d s (u)(w) = µu ◦ w( s )
for p ∈ M and v ∈ Tp M, where u ∈ P and w ∈ Hu are chosen so that π (u) = p and

π∗ (w) = v, and w is considered to act componentwise as a local derivation at u as
defined in 5.49. Ds is usually denoted by ∇s .
The element Ds (p)(v) ∈ E p , also denoted by ∇v s , is called the covariant deriva-
tive of s at p with respect to the tangent vector v ∈ Tp M. If X is a vector field on M,
then the section Ds (X) ∈ Γ(M; E) , which is denoted by ∇X s , is called the covariant
derivative of s with respect to X. It is given by
∇X s (p) = ∇Xp s = µu ◦ Xu∗ ( s )
for p ∈ M, where u ∈ P is chosen so that π (u) = p , and X ∗ is the horizontal lifting

of X.

ψ : G → Aut (W ) of G on a finite dimensional vector space W . Suppose that v ∈ Tp M
is the tangent vector to a smooth curve γ : I → M at p, where I is an open interval
containing t0 and γ (t0 ) = p. If s ∈ Γ(M; E) is a section of π ′ , then the covariant
derivative of s at p with respect to v is given by

∇v s = lim 1 [ τ tt00 +h s(γ (t0 + h)) − s(p) ] = dt
d τ tt s(γ (t)) ,
h→0 h 0
t0
where τ tt0 : π ′ −1 (γ (t)) → π ′ −1 (p) is the parallel transport in E along γ from t to

t0 , and the limit is taken with respect to the vector space topology on E p defined in
Proposition 13.117 in the appendix.
PROOF : Choose a u ∈ π −1 (p) and a w ∈ Hu with π∗ (w) = v, and let α : I → P be
FIBRE BUNDLES 387
the horizontal lift of γ with α (t0 ) = u. Then we also have that α ′ (t0 ) = w, since
α ′ (t0 ) ∈ Hu and π∗ (α ′ (t0 )) = γ ′ (t0 ) = v. By Remark 10.72 we now have that
τ tt0 = ( τ tt0 ) −1 = µ u ◦ µ −1
α (t)
so that
τ tt0 s(γ (t)) = µu ◦ s (α (t))
for t ∈ I. Hence it follows from 5.49 and Lemma 2.84 that

d t ′
dt τ t0 s(γ (t)) = µu ◦ s ∗ (α (t0 )) = µu ◦ s ∗ (w) = µu ◦ w( s ) = ∇v s
t0
10.79 Remark Let π : P → M be a principal G-bundle with a connection H,

ψ : G → Aut (W ) of G on a finite dimensional vector space W . Suppose that v ∈ Tp M
is the tangent vector to a smooth curve γ : I → M at p, where I is an open interval
containing 0 and γ (0) = p, and let s ∈ Γ(M; E) be a section of π ′ .
If B = {v1 , ... , vn } is a basis for E p , we let vi (t) = τ t0 (vi ) for i = 1, ... , n and
t ∈ I be the vectors obtained by parallel transport in E along the curve γ from 0 to
t. Then Bt = {v1 (t), ... , vn (t)} is a basis for Eγ (t) , since τ t0 : E p → Eγ (t) is a linear
isomorphism. If
n
s(γ (t)) = ∑ ci (t) vi (t) ,
i=1
then
n
τ t0 s(γ (t)) = ∑ ci (t) vi
i=1
so that ci are the components with respect to B of the smooth curve c : I → E p given
by c(t) = τ t0 s(γ (t)) = µu ◦ s (α (t)). Hence we have that
n
∇v s = ∑ ci′ (0) vi
i=1
by Lemma 2.84.

ψ : G → Aut (W ) of G on a finite dimensional vector space W . Then we have that
(1) ∇X1 +X2 s = ∇X1 s + ∇X2 s
(2) ∇X (s1 + s2 ) = ∇X s1 + ∇X s2
(3) ∇ f X s = f ∇X s
(4) ∇X ( f s) = X( f ) s + f ∇X s
for all vector fields X, X1 , X2 ∈ T1 (M), sections s, s1 , s2 ∈ Γ(M; E) and smooth func-
tions f ∈ F (M).
PROOF : We see that (1), (2) and (3) follow immediately from Definition 10.77 and
Proposition 10.51. To prove (4), we use that
f s (u) = µ u−1 ( f (π (u)) s (π (u))) = f (π (u)) µ u−1 (s (π (u))) = f ◦ π (u) s (u)
for u ∈ P, which implies that

f s = ( f ◦ π) s .
Since
Xu∗ ( f ◦ π ) = π∗ (Xu∗ )( f ) = X p ( f ) = X( f )(p)
where π (u) = p, it follows from Proposition 5.50 that
Xu∗ ( f s ) = X( f )(p) s(u) + f (p) Xu∗ ( s ) .
By applying µu on both sides, we now obtain
∇X ( f s)(p) = X( f )(p) s(p) + f (p) ∇X s (p)

and let W a finite dimensional vector space. Consider the associated vector bundle
obtained from the representation ρ : G → Aut (W ) given by ρ (g) = id W for g ∈ G ,
which may be identified with the trivial vector bundle πe : EW → M as described
in Example 10.29 (d). Then each vector-valued smooth function f : M → W on M
corresponds to a section s = (idM , f ) of πe , and we have that
∇v s = (p, v( f ))
for every p ∈ M and v ∈ Tp M.

PROOF : Since
s (u) = µ u−1 ◦ s (π (u)) = µ u−1 (π (u), f ◦ π (u)) = f ◦ π (u)
for u ∈ P , we have that

s= f ◦π .
Hence it follows from 5.51 and Proposition 5.63 that
∇v s = µu ◦ d s (u)(w) = µu ◦ π ∗ (d f )(u)(w) = µu ◦ d f (p)(v) = (p, v( f )) ,
where u ∈ P and w ∈ Hu are chosen so that π (u) = p and π∗ (w) = v.

FIBRE BUNDLES 389
COVARIANT DERIVATIVE OF TENSOR FIELDS
10.82 Remark If π : L (M) → M is the frame bundle over a smooth manifold

M n with a connection H, then the associated vector bundle obtained from the rep-
resentation ρ kl : Gl (Rn ) → Aut (T kl (Rn )) may be identified with the tensor bundle
πlk : Tlk (M) → M as in Example 10.29 (f). Suppose that v ∈ Tp M is the tangent vec-
tor to a smooth curve γ : I → M at p, where I is an open interval containing 0 and
γ (0) = p, and let T ∈ T kl (M) be a tensor field of type (kl ) on M. Choose a u ∈ L p (M),
and let α : I → L (M) be the horizontal lift of γ with α (0) = u.
Now let B = {v1 , ... , vn } be a basis for Tp M with dual basis B ∗ = {v1 , ... , vn },
and let vi (t) = τ t0 (vi ) = α (t) ◦ u−1 (vi ) and vi (t) = τ t0 (vi ) = vi ◦ u ◦ α (t)−1 for i =
1, ... , n and t ∈ I be the vectors and covectors obtained by parallel transport in T M
and T ∗ M along the curve γ from 0 to t. Then Bt = {v1 (t), ... , vn (t)} is a basis for
Tγ (t) M with dual basis Bt∗ = {v1 (t), ... , vn (t)}, since τ t0 : Tp M → Tγ (t) M is a linear
isomorphism.
Since τ t0 = (u ◦ α (t)−1 ) ∗ in the tensor bundle Tlk (M), it follows by induction
from Remark 4.54 that
vi1 (t) ⊗ · · · ⊗ vik (t) ⊗ v j1 (t) ⊗ · · · ⊗ v j l (t) = τ t0 (vi1 ⊗ · · · ⊗ vik ⊗ v j1 ⊗ · · · ⊗ v j l )
for (i1 , ..., ik , j1 , ..., j l ) ∈ Ink+l and t ∈ I, which shows that the parallel transport in
Tlk (M) of the basis vectors in T kl (B) are obtained by taking the tensor product of
the parallel transported basis vectors from B and B ∗ . If
j ,..., j
T (γ (t)) = ∑ C p i11,...,ik l (t) vi1 (t) ⊗ · · · ⊗ vik (t) ⊗ v j1 (t) ⊗ · · · ⊗ v j l (t) ,
i1 ,...,ik
j1 ,..., j l
then
j ,..., j
τ t0 T (γ (t)) = ∑ C p i11,...,ik l (t) vi1 ⊗ · · · ⊗ vik ⊗ v j1 ⊗ · · · ⊗ v j l
i ,...,i
1 k
j1 ,..., j l
j ,..., j
so that C p i11,...,ik l are the components with respect to T kl (B) of the smooth curve
c : I → T kl (Tp M) given by c(t) = τ t0 T (γ (t)) = (u−1 ) ∗ ◦ T (α (t)). Hence we have
that
d
j1 ,..., j l i1 ik
∇v T = ∑
i ,...,i
dt C p i1 ,...,ik v ⊗ · · · ⊗ v ⊗ v j1 ⊗ · · · ⊗ v j l
1 k 0
j1 ,..., j l
by Lemma 2.84.

(1) If f is a smooth function on M, then ∇X f = X( f ).
(2) If T ∈ T kl11 (M) and S ∈ T kl22 (M), then ∇X (T ⊗ S) = (∇X T ) ⊗ S + T ⊗ (∇X S).
(3) ∇X δ = 0, where δ ∈ T 11 (M) is the Kronecker delta tensor.

(4) ∇X commutes with contractions, i.e., if T ∈ T kl (M), and if 1 ≤ r ≤ k and
1 ≤ s ≤ l, then ∇X (Csr T ) = Csr (∇X T ).
(5) If T ∈ T kl (M), and if X1 , ... , Xk ∈ T1 (M) and λ1 , ... , λl ∈ T 1 (M), then
∇X [ T (X1 , ... , Xk , λ1 , ... , λl ) ] = (∇X T ) (X1 , ... , Xk , λ1 , ... , λl )

k
+ ∑ T (X1 , ... , ∇X Xi , ... , Xk , λ1 , ... , λl )
i=1
l
+ ∑ T (X1 , ... , Xk , λ1 , ... , ∇X λi , ... , λl ) .
i=1
PROOF : Let f be a smooth function on M which can be considered to be a tensor

field of type (00 ) . If p ∈ M and B is a basis for Tp M, we have that T 00 (Tp M) = R
with basis T 00 (B) = {1}. Suppose that X p is the tangent vector to a smooth curve
γ : I → M at p, where I is an open interval containing 0 and γ (0) = p. If f (γ (t)) = c(t)
for t ∈ I, then it follows from 2.77 and Remark 10.82 that ∇Xp f = c′ (0) = X p ( f ),
Point (2) to (4) follow directly from Remark 10.82. Point (5) follows from (2) and
(4) since T (X1 , ... , Xk , λ1 , ... , λl ) can be obtained from T ⊗ X1 ⊗ · · · ⊗ Xk ⊗ λ1 ⊗ · · ·
⊗λl by applying contractions repeatedly.
COVARIANT DERIVATIVE OF SECTIONS ALONG SMOOTH

MAPS
10.84 Let π : P → M be a principal G-bundle, and let π ′ : E → M be the fibre
bundle associated with P with fibre F obtained from a Lie group homomorphism
ψ : G → G′ into a Lie group G′ which acts on the smooth manifold F effectively
on the left. Consider a smooth map f : N → M from a smooth manifold N, and let
πe : Pe → N and πe ′ : Ee → N be fibre bundles induced from π : P → M and π ′ : E → M
by f . We contend that the fibre bundle π ′′ : E ′ → N with fibre F associated with
πe : Pe → N is equivalent over N to the induced fibre bundle πe ′ : Ee → N, with the
equivalence e : E ′ → Ee given by
e ([ (p, u), v ]) = (p, [ u, v ])
for p ∈ N, u ∈ P and v ∈ F where f (p) = π (u). This is clearly well defined, for if
g ∈ G, then
FIBRE BUNDLES 391
e([(p, u)g, ψ (g−1 )v]) = [(p, ug), ψ (g−1 )v] = (p, [ug, ψ (g−1 )v])
= (p, [u, v]) = e([(p, u), v ]).
To show that e is an equivalence, let (t ′ , π ′ −1 (U)) be the local trivialization in E

associated with a local trivialization (t, π −1 (U)) in P, defined as in 10.28 by
t ′ ([ u, v ]) = (q, ψ ◦ tq (u) v)
for q ∈ U, u ∈ Pq and v ∈ F. Let (e e and (t˜ ′ , πe ′ −1 (U)),

t , πe−1 (U)) e where U
e = f −1 (U),
e e
be the corresponding local trivializations in P and E, respectively, defined as in 10.12
by
t˜(p, u) = (p,t f (p) (u))
and
t˜ ′ (p, [ u, v ]) = (p,t ′f (p) ([ u, v ])) = (p, ψ ◦ t f (p) (u) v)
e u ∈ Pf (p) and v ∈ F. Now if (t ′′ , π ′′ −1 (U)) is the local trivialization in E ′
for p ∈ U,
e we have that
t , πe−1 (U)),
associated with (e
t ′′ ([ (p, u), v ]) = (p, ψ ◦ t˜p (p, u) v) = (p, ψ ◦ t f (p) (u) v)
e u ∈ Pf (p) and v ∈ F, which shows that
for p ∈ U,
t˜ ′ ◦ e ◦ t ′′ −1 = id Ue × F .
Using the equivalence e, we may therefore identify the fibre bundle E ′ associated
with Pe with fibre F with the fibre bundle Ee induced from E by f , and the diffeomor-
phism µ ′(p,u) : F → E p′ defined in 10.28 is then identified with the diffeomorphism
e (p,u) = e p ◦ µ ′
µ (p,u): F → Eep given by
e (p,u) (v) = (p, µ u (v))

µ
for p ∈ N , u ∈ Pf (p) and v ∈ F, where µ u : F → E f (p) is the corresponding diffeomor-

phism for the fibre bundle E. If ( fe ′ , f ) is the canonical bundle map of the induced
bundle πe ′ : Ee → N, we have that
fep′ ◦ µ
e (p,u) = µ u
10.85 Let π : P → M be a principal G-bundle with a connection H, and let

π ′ : E → M be the associated vector bundle obtained from a representation ψ : G →
Aut (W ) of G on a finite dimensional vector space W . Consider a smooth map f :
N → M from a smooth manifold N, and let πe : Pe → N and πe ′ : Ee → N be the fibre
bundles induced from π : P → M and π ′ : E → M by f , respectively, with canonical
bundle maps ( fe, f ) and ( fe′ , f ) .
If F : N → E is a lifting of f and v ∈ Tp N is a tangent vector at a point p ∈ N, we

define the covariant derivative of F at p with respect to v by
∇v F = fe′ ( ∇v Fe ) ,
where Fe is the section determined by F in the induced vector bundle πe ′ : Ee → N,

e
which is given by F(p) = (p, F(p)) for p ∈ N. If X is a vector field on N, we define
∇X F (p) = ∇Xp F
for p ∈ N, so that
∇X F = fe′ ◦ ∇X Fe .
Hence ∇X F : N → E is the lifting of f corresponding to the section ∇X Fe in the
induced vector bundle πe ′ : Ee → N.
Now let F ∈ F (Pe ;W ) be the vector-valued function given by
F (p, u) = µ u−1 ◦ F (p)
for p ∈ N and u ∈ Pf (p) . As µ u = fep′ ◦ µ

e (p,u) and F = fe′ ◦ Fe , this means that
−1
e (p,u)
F (p, u) = µ e ( πe ′ (p, u))
◦ F
e Using Definition 10.77 we therefore have that

for (p, u) ∈ P.
∇v Fe = µ
e (p,u) ◦ w( F )
so that
∇v F = µ u ◦ w( F )
e (p,u) is chosen so that πe∗ (w) = v,
for p ∈ N and v ∈ Tp N, where u ∈ Pf (p) and w ∈ H
and where w is considered to act componentwise as a local derivation at (p, u) as
defined in 5.49. We recall from 10.37 that
e (p,u) = ( fe∗ (p,u) ) −1 (H u ) .
H
If X ∗ is the horizontal lifting of X, we have that
∇X F (p) = µ u ◦ Xu∗ ( F )
10.86 Remark If F : I → E is a lifting of a smooth curve γ : I → M on M defined

on an open interval I, and we use the standard local chart (r, I) on I where r : I → R
DF
is the inclusion map, then ∇ d F is also denoted by and is called the covariant
dr dr
derivative of F along γ .
FIBRE BUNDLES 393

ψ : G → Aut (W ) of G on a finite dimensional vector space W . If f : N → M is a
smooth map from a smooth manifold N and s : M → E is a section of π ′ on M, then
we have that
∇v (s ◦ f ) = ∇ f∗ (v) s
for every v ∈ T N.
PROOF : Let πe : Pe → N be the fibre bundle induced from π : P → M by f with
canonical bundle map ( fe, f ). If F = s ◦ f , then the map F ∈ F (Pe ;W ) is given by
F (p, u) = µ u−1 ◦ s ◦ f (p) = µ u−1 ◦ s ◦ π (u) = s (u) = s ◦ fe(p, u)

e (p,u) is chosen so that πe∗ (w) = v, then fe∗ (w) ∈ H u and
e If w ∈ H
for (p, u) ∈ P.
π∗ ◦ fe∗ (w) = f∗ ◦ πe∗ (w) = f∗ (v) .
Hence we have that
∇ f∗ (v) s = µu ◦ fe∗ (w)( s ) = µu ◦ w( s ◦ fe) = ∇v (s ◦ f )

ψ : G → Aut (W ) of G on a finite dimensional vector space W . If f : N → M is a
smooth map from a smooth manifold N, then we have that
(1) ∇X1 +X2 F = ∇X1 F + ∇X2 F
(2) ∇X (F1 + F2) = ∇X F1 + ∇X F2
(3) ∇gX F = g ∇X F
(4) ∇X (gF) = X(g) F + g ∇X F
for all vector fields X, X1 , X2 ∈ T1 (N), liftings F, F1 , F2 ∈ Γ( f ; E) of f and smooth
functions g ∈ F (N).
PROOF : e Fe1 and
Follows from 10.85 and Proposition 10.80 applied to the sections F,
e ′ e ′
F2 in the vector bundle πe : E → N induced from π : E → M by f .
10.89 Lemma Let π : P → M be a principal G-bundle with a connection H, and

let π ′ : E → M be the associated vector bundle obtained from a representation ψ :
G → Aut (W ) of G on a finite dimensional vector space W . Suppose that f : N → M
is a smooth map from a smooth manifold N, and that X ∈ T1 (N) is a vector field on
N. If F1 , F2 ∈ Γ( f ; E) are liftings of f with F1 | V = F2 | V for an open subset V of N,
then we have that (∇X F1 )| V = (∇X F2 )| V .
PROOF : Considering the difference F2 − F1 , it is clearly enough to prove that

(∇X F)| V = 0 for each lifting F ∈ Γ( f ; E) of f with F | V = 0. To show this, let
p ∈ V and choose a smooth function h ∈ F (N) with h = 1 on N − V and h(p) = 0.
Such a function h exists by Corollary 2.24. Then we have that
∇X F(p) = ∇X (hF)(p) = X(h)(p) F(p) + h(p) ∇X F(p) = 0

ψ : G → Aut (W ) of G on a finite dimensional vector space W . Suppose that F : I → E
is a lifting of the smooth curve γ : I → M defined on an open interval I. Using the
standard local chart (r, I) on I where r : I → R is the inclusion map, the covariant
derivative of F along γ at t0 ∈ I is given by

d t

(t0 ) = lim 1 [ τ t00 F (t0 + h) − F(t0 ) ] = dt
DF t +h
dr h→0 h τ t0 F (t) ,
t0
where τ tt 0 : π ′ −1 (γ (t)) → π ′ −1 (γ (t 0 )) is the parallel transport in E along γ from t to

t 0 , and the limit is taken with respect to the vector space topology on Eγ (t 0 ) defined
in Proposition 13.117 in the appendix.
PROOF : Let πe : Pe → I and πe ′ : Ee → I be the fibre bundles induced from π : P → M
and π′: E → M by γ , and let (γe ′ , γ ) be the canonical bundle map of πe ′ : Ee → I. If
e
F is the section in πe ′ : Ee → I determined by F which is a lifting of the identity map
β = idI : I → I, it follows from Proposition 10.78 that

t
∇d d
Fe = dt τe t Fe (β (t)), (1)
dr
0
t0 t0
t
where τe t 0 : πe ′ −1 (β (t)) → πe ′ −1 (β (t 0 )) is the parallel transport in Ee along β from t
to t 0 .
Now let α : I → P be a horizontal lift of γ . Then the section α e : I → Pe in πe : Pe → I
determined by α , which is given by α e (t) = (t, α (t)) for t ∈ I, is a horizontal lift of β
by Remark 10.58. The proposition therefore follows by applying γe ′ on both sides of
(1), since
γe ′ ◦ τe tt 0 ◦ Fe (β (t)) = γe ′ ◦ µ
e (t e −1 ◦ Fe (t)
◦ µ
0 ,α (t 0 )) (t,α (t))
= µ α (t ◦ µ −1
α (t)
◦ F(t) = τ tt 0 ◦ F(t)
0)
for t ∈ I by 10.85 and Remark 10.72.

FIBRE BUNDLES 395

ψ : G → Aut (W ) of G on a finite dimensional vector space W . Consider a smooth
curve γ : I → M on M defined on an open interval I, and let (r, I) be the standard
local chart on I where r : I → R is the inclusion map. Then a lifting F : I → E of γ
is said to be parallel along γ if the covariant derivative DF/ d r = 0 on I.

and let π ′ : E → M be the associated vector bundle obtained from a representation ψ :
G → Aut (W ) of G on a finite dimensional vector space W . Then a lifting F : I → E
of a smooth curve γ : I → M defined on an open interval I is parallel along γ if and
only if the following assertion is satisfied :
For each pair of points t0 ,t1 ∈ I we have that
τ tt10 F (t 0 ) = F (t1 ) ,
where τ t 01 : π ′ −1 (γ (t 0 )) → π ′ −1 (γ (t 1 )) is the parallel transport in E along γ from t 0

t
to t 1 .
PROOF : If c : I → E γ (t 1 ) is the smooth curve given by c(t) = τ tt 1 F (t) for t ∈ I, then


d s

′ t d s t DF
c (t) = ds τ t 1 F (s) = τ t 1 ds τ t F (s) = τ t 1 d r (t)
t t
for t ∈ I. Hence the curve c is constant if and only if DF/ d r = 0 on I.
LINEAR CONNECTIONS
10.93 Proposition Let π : L (M) → M be the frame bundle over a smooth man-
ifold M n . Then we have an Rn -valued 1-form θ on L (M) defined by
θ (u) = u−1 ◦ π∗ u
for u ∈ L (M), called the canonical form or the dual form on L (M), which is
tensorial of type (id, Rn ) with ker θ (u) = Vu for every u ∈ L (M). We have that
θ = ε , where ε ∈ Ω 1 (M; T M) is the bundle-valued 1-form on M defined in Example
5.45.
10.94 Remark If s = (X1 , ... , Xn ) is a frame field on an open subset V of M, then

the 1-forms s∗ θ i for i = 1, ... , n form the dual local basis for Λ 1 (T M) on V . Indeed,
we have that
s∗ θ (p)(v) = θ (s(p))(s∗ (v)) = s(p)−1 (v)
for p ∈ V and v ∈ Tp M, showing that
s∗ θ i (p)(X j (p)) = ei ◦ s∗ θ (p) ◦ s(p)(e j ) = δi j
for p ∈ V and i, j = 1, ... , n.
10.95 Definition A connection H in the frame bundle π : L (M) → M over a

smooth manifold M is called a linear connection on M.
10.96 Proposition Let H be a linear connection on a smooth manifold M n , and

let ξ ∈ Rn . Then we have a horizontal vector field B(ξ ) on L (M), called the basic
vector field determined by ξ , where B(ξ )u is the unique vector in Hu such that
π∗ (B(ξ ) u ) = u(ξ ) (1)
for u ∈ L (M), i.e.,

B(ξ ) u = α u−1 ◦ u(ξ ) (2)
where (α , π ) is the bundle map from H to T M defined in Proposition 10.49.
If θ and ω is the dual form and the connection form of H, respectively, then the
map B : Rn → T 1 L (M) is uniquely determined by the fact that
θ (u) (B(ξ ) u ) = ξ and ω (u) (B(ξ ) u ) = 0 (3)
for every u ∈ L (M) and ξ ∈ Rn .

PROOF : The last formula in (3) is equivalent to the assertion that B(ξ ) u ∈ Hu , and
the first formula in (3) is obtained from (1) by applying the linear isomorphism u−1
on both sides.
It only remains to prove that B(ξ ) is a vector field on L (M). If {Y1 , ...,Yn }
is a local basis for H on an open subset V of L (M), we must show that the map
a : V → Rn defined by
n
B(ξ ) u = ∑ ai (u)Yi u
i=1
for u ∈ V , is smooth. By applying θ (u) on both sides, it follows from (3) that
n
ξ = ∑ ai (u) hi (u)
i=1
where
hi (u) = θ (u)(Yi u ) = u−1 ◦ αu (Yi u )
FIBRE BUNDLES 397
for i = 1, ... , n. Since u and αu are linear isomorphisms, these vectors form a basis
for Rn . We therefore have a smooth map h : V → Gl (n, R), where h(u) is the matrix
with h1 (u), ..., hn (u) as column vectors for u ∈ V . Writing the vectors ξ and a(u) as
column vectors as well, the above equation can be written as
ξ = h(u) a(u)
so that
a(u) = h(u)−1 ξ
for u ∈ V . This shows that a is smooth and hence that B(ξ ) is a vector field on
L (M).
10.97 Proposition Given a linear connection H on a smooth manifold M n , then
(R F ) ∗ B(ξ ) = B(F −1 (ξ )) (1)
for every ξ ∈ Rn and F ∈ Gl (Rn ). Moreover, B(ξ ) is nowhere zero when ξ 6= 0, and
if {ξ1 , ... , ξn } is a basis for Rn , then {B(ξ1 ), ... , B(ξn )} is a local basis for H on
L (M).
PROOF : For each u ∈ L (M) it follows from Definition 10.32 (ii) that (R F ) ∗ B(ξ ) u
and B(F −1 (ξ )) u ◦ F are vectors in H u ◦ F such that
π∗ ((R F ) ∗ B(ξ ) u ) = π∗ (B(ξ )u ) = u(ξ ) = u ◦ F (F −1 (ξ )) = π∗ (B(F −1 (ξ )) u ◦ F )
since π ◦ R F = π . Hence
(R F ) ∗ ◦ B(ξ ) = B(F −1 (ξ )) ◦ RF

The last part of the Proposition follows from formula (2) in Proposition 10.96
since α u−1 ◦ u is a linear isomorphism for every u ∈ L (M).
10.98 Definition Let H be a linear connection on a smooth manifold M n , and

let θ be the dual form on L (M). Then the Rn -valued 2-form Θ = D θ is called the
torsion form of H on L (M).
10.99 Remark From Proposition 10.43 and 10.93 it follows that the torsion form
Θ is a tensorial 2-form of type (id, Rn ).
10.100 Definition Let H be a linear connection on a smooth manifold M n , and

let π ′ : T M → M be the tangent bundle. Then we have a bundle-valued 2-form T ∈
Ω 2 (M; T M), called the torsion form on M, defined by T = Θ , where Θ is the torsion
form of H on L (M), i.e.,
T (p)(v1 , v2 ) = u ◦ Θ(u)(w1 , w2 )
for p ∈ M and v1 , v2 ∈ Tp M, where u ∈ L (M) and w1 , w2 ∈ Tu L (M) are chosen so

that π (u) = p and π∗ (wi ) = vi for i = 1, 2. By Proposition 10.93 we have that T = D ε ,
where ε ∈ Ω 1 (M; T M) is the bundle-valued 1-form on M defined in Example 5.45.
10.101 Proposition Suppose that H is a linear connection on a smooth manifold

M n , and that s = (X1 , ... , Xn ) is a frame field on an open subset V of M and E =
{e1 , ... , en } is the standard basis for Rn . Let Θ and T be the torsion forms of H on
L (M) and M, respectively, and Θ e = s∗ (Θ) be the pull-back of Θ by s. Then we have
that
n
T (X j , Xk ) = ∑ T j ik Xi ,
i=1
where the smooth functions T jik : V → R, called the components of T with respect to
the frame field s = (X1 , ... , Xn ), are defined by
e i (X j , Xk )
T j ik = Θ
e i are the components of Θ

for 1 ≤ i, j, k ≤ n, where Θ e with respect to the basis E . If
e i ∗ i
θ = s θ for i = 1, ... , n are the pull-backs of the components of the dual form θ by
s, which form a dual local basis on V to the frame field s = (X1 , ... , Xn ) as described
in Remark 10.94, then we have that
ei =
Θ ∑ T jik θe j ∧ θe k
j<k
for 1 ≤ i ≤ n.
PROOF : By Proposition 10.73 and Example 10.29 (e) we have that
n
e j , Xk ) = ∑ Θ
T (X j , Xk ) = s · Θ(X e i (X j , Xk ) Xi
i=1
for 1 ≤ j, k ≤ n.
10.102 Proposition If T is the torsion form of a linear connection H on a smooth

manifold M, then we have that
T (X,Y ) = ∇X Y − ∇Y X − [ X , Y ]
for every vector field X and Y on M.

PROOF : Since θ = ε and X = ε (X), where ε ∈ Ω 1 (M; T M) is the bundle-valued
1-form on M defined in Example 5.45, it follows from Proposition 10.74 that X =
θ (X ∗ ) where X ∗ is the horizontal lift of X, and similarly Y = θ ( Y ∗ ). Hence we have
that
∇X Y (p) = u ◦ Xu∗ ( θ ( Y ∗ ))
FIBRE BUNDLES 399
for p ∈ M, where u ∈ P is chosen so that π (u) = p . By Remark 5.86 it therefore

follows that
T ( X,Y )(p) = T (p)( X p ,Yp ) = u ◦ d θ (u)( Xu∗ ,Yu∗ )

= u ◦ Xu∗ ( θ ( Y ∗ )) − u ◦ Yu∗ ( θ ( X ∗ )) − u ◦ θ (u)( [ X ∗ ,Y ∗ ] u )
= ∇X Y (p) − ∇Y X (p) − [ X , Y ] p ,
since
u ◦ θ (u)( [ X ∗ ,Y ∗ ] u ) = π∗ ( [ X ∗ ,Y ∗ ] u ) = [ X , Y ] p

and let π ′ : P ×Ad g → M be the associated vector bundle obtained from the adjoint
representation Ad : G → Aut (g) of G on the Lie algebra g. Then we have a bundle-
valued 2-form R ∈ Ω 2 (M; P ×Ad g), called the curvature form on M, defined by
R = Ω , where Ω is the curvature form of H on P. i.e.,
R(p)(v1 , v2 ) = µ u ◦ Ω(u)(w1 , w2 )
for p ∈ M and v1 , v2 ∈ Tp M, where u ∈ P and w1 , w2 ∈ Tu P are chosen so that π (u) = p

and π∗ (wi ) = vi for i = 1, 2.
10.104 Remark If π ′ : L (E) → M is the frame bundle of an n-dimensional

vector bundle π : E → M with a connection H, then the associated vector bundle
obtained from the adjoint representation Ad : Gl (Rn ) → Aut (gl (Rn )) may be iden-
tified with the vector bundle π ′′ : Λ 1 (E ; E ) → M as in Example 10.29 (g). We may
therefore assume that the curvature form R ∈ Ω 2 (M; Λ 1 (E ; E )), and we have that
R(p)(v1 , v2 ) = u ◦ Ω(u)(w1 , w2 ) ◦ u−1
for p ∈ M and v1 , v2 ∈ Tp M, where u ∈ L p (E) and w1 , w2 ∈ Tu L (E) are chosen so

that π∗′ (wi ) = vi for i = 1, 2. If X and Y are vector fields on M and Z is a section of
π on M, then R(X,Y ) is a section of π ′′ on M, and hence R(X,Y ) Z is again a section
of π on M.
10.105 Proposition Let π ′ : L (E) → M be the frame bundle of an n-dimensional

vector bundle π : E → M, and consider the associated vector bundle obtained from
the adjoint representation Ad : Gl (Rn ) → Aut (gl (Rn )) which may be identified with
the vector bundle π ′′ : Λ 1 (E ; E ) → M as in Example 10.29 (g). Suppose that ω is
a bundle-valued k-form on M with values in Λ 1 (E ; E ), and that η = ω . Let E =
{e1 , ... , en } be the standard basis for Rn and D = {Fji |1 ≤ i, j ≤ n} be the basis for
gl (Rn ) defined in Example 8.10 (b), where Fji : Rn → Rn is the linear map given by
Fji (el ) = δ j l ei for l = 1, ... , n. If s = (s1 , ... , sn ) is a section of π ′ : L (E) → M on

an open subset V of M, and η e = s∗ (η ) is the pull-back of η by s, then we have that
n
e ji (p)(v1 , ... , vk ) s i (p)
ω (p)(v1 , ... , vk ) s j (p) = ∑ η
i=1
e ji are the components of η

for p ∈ V , v1 , ... , vk ∈ Tp M and j = 1, ... , n, where η e with
respect to the basis D.
PROOF : By Proposition 10.73 and Example 10.29 (g) we know that
e (p)(v1 , ... , vk )
ω (p)(v1 , ... , vk ) ◦ s(p) = s(p) ◦ η
for p ∈ V and v1 , ... , vk ∈ Tp M. Hence it follows that
ω (p)(v1 , ... , vk ) s l (p) = ω (p)(v1 , ... , vk ) ◦ s(p)(el )
e ji (p)(v1 , ... , vk ) s(p) ◦ Fji (el )

e (p)(v1 , ... , vk )(el ) = ∑ η
= s(p) ◦ η
ij
n n
e li (p)(v1 , ... , vk ) s(p)(ei ) = ∑ η
=∑η e li (p)(v1 , ... , vk ) s i (p)
i=1 i=1
for p ∈ V , v1 , ... , vk ∈ Tp M and l = 1, ... , n.
10.106 Corollary Suppose that H is a linear connection on a smooth manifold

M n , and that s = (X1 , ... , Xn ) is a frame field on an open subset V of M. Let E =
Fji (ek ) = δ j k ei for k = 1, ... , n. If Ω and R are the curvature forms of H on L (M)
and M, respectively, and Ω e = s∗ (Ω) is the pull-back of Ω by s, then we have that
n
R(Xk , Xl ) X j = ∑ R ji kl Xi ,
i=1
where the smooth functions R ji kl : V → R, called the components of R with respect

to the frame field s = (X1 , ... , Xn ), are defined by
e i (Xk , Xl )
R ji kl = Ω j
e i are the components of Ω

for 1 ≤ i, j, k, l ≤ n, where Ω e with respect to the basis D. If
j
θe = s θ for i = 1, ... , n are the pull-backs of the components of the dual form θ by
i ∗ i
e ji =
Ω ∑ R ji kl θe k ∧ θe l
k<l
for 1 ≤ i, j ≤ n.
FIBRE BUNDLES 401
10.107 Proposition If R is the curvature form defined in Remark 10.104, then

we have that
R(X,Y ) Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇ [ X ,Y ] Z
for every vector field X and Y on M , and every section Z of π on M.
PROOF : Choose u ∈ L p (E) for a p ∈ M, and let X ∗ and Y ∗ be the horizontal lifts of
X and Y , respectively. By Proposition 10.47 we have that
Ω (u)(Xu∗ ,Yu∗ ) = − ω (u)([ X ∗ ,Y ∗ ]u ) = − ω (u)(v ( [ X ∗ ,Y ∗ ]u )) ,
and it follows from Proposition 10.26 that the vertical component
v ( [ X ∗ ,Y ∗ ]u ) = σ (F) u
for a unique F ∈ gl (Rn ). Hence we have that
Ω (u)(Xu∗ ,Yu∗ ) = − F
so that
R(X,Y ) Z (p) = u ◦ Ω (u)(Xu∗ ,Yu∗ ) ◦ Z (u) = − u ◦ F ◦ Z (u). (1)
On the other hand we have that
∇Y Z = DZ (Y ) = DZ (Y ∗ ) = dZ (Y ∗ ) = Y ∗ (Z )
so that
∇X ∇Y Z (p) − ∇Y ∇X Z (p) − ∇ [ X ,Y ] Z (p)
= u ◦ Xu∗ (Y ∗ (Z )) − u ◦ Yu∗ ( X ∗ (Z )) − u ◦ h ( [ X ∗ ,Y ∗ ] u )(Z ) (2)
= u ◦ v ( [ X ∗ ,Y ∗ ] u )(Z ) = u ◦ σ (F) u (Z ).
By Proposition 9.39 the integral curve γ u : R → L (E) for the fundamental vector
field σ (F) with initial condition u is given by
γ u (t) = u ◦ exp (t F)
for t ∈ R. Using that Z is tensorial, we define a curve c : R → Rn by
c(t) = Z ◦ γ u (t) = Z (u ◦ exp (t F)) = exp (−t F) ◦ Z (u) = Λu ◦ φ −F (t)

for t ∈ R , where φ −F : R → Gl (Rn ) is the one-parameter subgroup defined in Propo-
sition 8.29, and Λu : gl (Rn ) → Rn is the linear map given by
Λu (G) = G ◦ Z (u)
for G ∈ gl (Rn ) . By 5.49 and 2.70 it follows that
σ (F) u (Z ) = c′ (0) = Λu ◦ φ −F
′
(0) = −F ◦ Z (u) (3)
if we canonically identify Tc(0) (Tu L (E)) with Tu L (E) as in Lemma 2.84. The
proposition now follows from formula (1), (2) and (3).
10.108 Let R be the curvature form of a linear connection H on a smooth

manifold M . By 5.25 , Proposition 5.31 and Remark 4.54 we have an equivalence
φ : T 2 (T M ; Λ 1 (T M ; T M)) → T 31 (M) over M defined by
φ (F)(v1 , v2 , v3 , λ ) = λ ◦ F(v2 , v3 )(v1 )
for p ∈ M , F ∈ T 2 (Tp M ; Λ 1 (Tp M ; Tp M)) , v1 , v2 , v3 ∈ Tp M and λ ∈ Tp∗ M . The

contraction Ric = C 21 (φ ◦ R) ∈ T 2 (M) is called the Ricci curvature tensor of H.
We see that
Ric(p)(v1 , v2 ) = tr (G)
for p ∈ M and v1 , v2 ∈ Tp M , where G : Tp M → Tp M is the linear map given by
G(v) = R(p)(v, v2 )(v1 )
for v ∈ Tp M . If s = (X1 , ... , Xn ) is a frame field on an open subset V of M , it follows

from Corollary 10.106 that
n
Ric(X j , Xl ) = ∑ R jkkl .
k=1
10.109 Theorem Let π : P → M be a principal G-bundle with a connection H,

G → Aut (W ) of G on a finite dimensional vector space W . Let ν : P → Λ 1 (W ; E )
be the smooth map defined in Example 10.29 (c) , which is given by
ν (u)(v) = µ u (v) = [ u, v]
e = s∗ (ω )
for u ∈ P and v ∈ W . If s is a section of π on an open subset V of M, and ω
is the pull-back by s of the connection form ω of H, then we have that
e )(p)(v)
∇(ν ◦ s)(p)(v) = (ν ◦ s)(p) ◦ (ψ ∗ · ω
for p ∈ V and v ∈ Tp M.
PROOF : If ρ : V × G → π −1 (V ) is the diffeomorphism defined by
ρ (p, g) = s(p) g
for p ∈ V and g ∈ G , then
ν ◦ s ◦ ρ (p, g) = µ −1 −1 −1 −1
s(p)g ◦ ν (s(p)) = ψ (g ) ◦ µ s(p) ◦ µ s(p) = ψ (g ) .
Now let ie : V → V × G and i p : G → V × G be the embeddings defined by

ie (p ′ ) = (p ′ , e) and i p (g ′ ) = (p, g ′ ) for p ′ ∈ V and g ′ ∈ G , and let ρ p = ρ ◦ i p .
For each p ∈ V and v ∈ Tp M there is by Proposition 10.26 a unique X ∈ g with
s ∗ (v) + ρ p ∗ (X) ∈ Hs(p) so that
e (p)(v) + X .
0 = ω (s(p)) (s ∗ (v) + ρ p ∗ (X)) = ω
FIBRE BUNDLES 403
Hence it follows from 5.51, Proposition 5.63 and Corollary 8.33 that
d ν ◦ s (s(p))(s ∗ (v) + ρ p ∗ (X)) = ρ ∗ d ν ◦ s (p, e)( i e ∗ (v) + i p ∗ (X))

e )(p)(v)
= d ( ν ◦ s ◦ ρ ) (p, e)( i e ∗ (v) + i p ∗ (X)) = − ψ ∗ (X) = (ψ ∗ · ω
so that
∇(ν ◦ s)(p)(v) = µs(p) ◦ d ν ◦ s(s(p))(s∗ (v) + ρ p∗(X))
e )(p)(v).
= (ν ◦ s)(p) ◦ (ψ∗ · ω
10.110 Proposition Suppose that π ′ : L (E) → M is the frame bundle of an n-

dimensional vector bundle π : E → M with a connection H, and that s = (s1 , ... , sn )
is a section of π ′ on an open subset V of M. Let E = {e1 , ... , en } be the standard basis
for Rn and D = {Fji |1 ≤ i, j ≤ n} be the basis for gl (Rn ) defined in Example 8.10
(b), where Fji : Rn → Rn is the linear map given by Fji (ek ) = δ j k ei for k = 1, ... , n.
If ω is the connection form of H and ω e = s∗ (ω ) is the pull-back of ω by s, then we
have that
n
e ji (p)(v) s i (p)
∇s j (p)(v) = ∑ ω
i=1
e ji are the components of ω
for p ∈ V , v ∈ Tp M and j = 1, ... , n, where ω e with respect
to the basis D.
PROOF : For each k we have that s k = Λk · s , where Λk : gl (Rn ) → Rn is the linear
map given by Λk (F) = F(ek ) for F ∈ gl (Rn ). Hence it follows by Proposition 5.62
and Theorem 10.109 that
e ij (p)(v)s(p) ◦ Fji (ek )
e (p)(v)(ek ) = ∑ ω
∇sk (p)(v) = ∇s(p)(v)(ek ) = s(p) ◦ ω
ij
n n
eki (p)(v)s(p)(ei ) = ∑ ω
= ∑ω eki (p)(v)si (p)
i=1 i=1
for p ∈ V , v ∈ Tp M and k = 1, ... , n.

M n , and that s = (X1 , ... , Xn ) is a frame field on an open subset V of M. Let E =
e = s∗ (ω ) is
Fji (ek ) = δ j k ei for k = 1, ... , n. If ω is the connection form of H and ω
the pull-back of ω by s, then we have that
n
∇Xk X j = ∑ Γki j Xi ,
i=1
where the smooth functions Γki j : V → R, called the connection components or the
connection coefficients of H with respect to the frame field s = (X1 , ... , Xn ), are
defined by
e ji (Xk )
Γki j = ω
for 1 ≤ i, j, k ≤ n, where ω e with respect to the basis D. If
e i ∗ i
θ = s θ for i = 1, ... , n are the pull-backs of the components of the dual form θ by
n
e ji =
ω ∑ Γki j θe k
k=1
for 1 ≤ i, j ≤ n.
10.112 Remark Let M n be a smooth manifold with a linear connection H. By the

connection components or the connection coefficients Γki j of H with respect to a local
chart (x,U) on M we mean the connection coefficients with respect to the coordinate
frame field
∂ ∂
s= 1 , ... , n .
∂x ∂x
In the same way we define the components T j ik and R ji kl of the torsion and curvature
forms T and R of H with respect to the local chart (x,U).
10.113 Proposition Let H be a linear connection on a smooth manifold M n ,

i
and let Γk j be the connection coefficients of H with respect to a frame field s =
(X1 , ... , Xn ) on an open subset V of M. If X and Y are vector fields on V with
n n
X = ∑ ai Xi and Y = ∑ b i Xi ,
i=1 i=1
then we have that !

n n
j i
∇X Y = ∑ ∑ a b ;j Xi
i=1 j=1
where
n
b i ; j = X j (b i ) + ∑ b k Γ jk
i
k=1
for 1 ≤ i, j ≤ n.
PROOF : By Proposition 10.80 and Corollary 10.111 we have that
∇X Y = ∑ a j ∇X j (b i Xi ) = ∑ a j { X j (b i ) Xi + b i ∇X j Xi }
ij ij
!
j i j k j i k i
= ∑ a X j (b ) Xi + ∑ a b ∇X j Xk = ∑ a X j (b ) + ∑ b Γ jk Xi .
ij kj ij k
FIBRE BUNDLES 405
10.114 Corollary Let H be a linear connection on a smooth manifold M n , and

let Γki j be the connection coefficients of H with respect to a local chart (x,U) on M.
If X and Y are vector fields on U with
n n
X = ∑ ai ∂ i and Y = ∑ b i ∂ i ,
i=1
∂x i=1
∂x
then we have that !

n n
j i ∂
∇X Y = ∑ ∑ a b ;j
i=1 j=1
∂ xi
where
i n
b i ; j = ∂ b j + ∑ b k Γ jk
i
∂x k=1
for 1 ≤ i, j ≤ n.
10.115 Proposition Suppose that π ′ : L (E) → M is the frame bundle of an n-

dimensional vector bundle π : E → M with a connection H, and that s = (s1 , ... , sn )
is a section of π ′ on an open subset V of M with a dual local basis (α 1 , ... , α n )
in the dual bundle π ∗ : E ∗ → M . Let E = {e1 , ... , en } be the standard basis for Rn
with dual basis E ∗ = {e1 , ... , en }, and let D = {Fji |1 ≤ i, j ≤ n} be the basis for
e = s∗ (ω ) is
Fji (ek ) = δ j k ei for k = 1, ... , n. If ω is the connection form of H and ω
the pull-back of ω by s, then we have that
n
∇ α i (p)(v) = − ∑ ω
e ji (p)(v) α j (p)
j=1

for p ∈ V , v ∈ Tp M and i = 1, ... , n, where ω e with respect
to the basis D.
PROOF : From Example 10.29 (f) we know that the dual bundle π ∗ : E ∗ → M is
associated to the frame bundle L (E) by means of the representation ψ : Gl (Rn ) →
Aut ((Rn ) ∗ ) given by
ψ (F) = (F −1) ∗
for F ∈ Gl (Rn ), i.e.,
ψ (F)(λ ) = λ ◦ F −1
for λ ∈ (Rn ) ∗ . By Corollary 8.33 it follows that ψ∗ : gl (Rn ) → End ((Rn ) ∗ ) is given
by
ψ∗ (F) = −F ∗
for F ∈ gl (Rn ) , so that
n n
ψ∗ (Fji )(ek ) = − ek ◦ Fji = − ∑ ek ◦ Fji (el ) el = − ∑ δ j l ek (ei ) el = − δ i k e j
l=1 l=1
for k = 1, ... , n. We also have a smooth map ν : L (E) → Λ 1 ((Rn ) ∗ ; E ∗ ) given by
ν (u) = (u−1 ) ∗
for u ∈ L (E) .
For each k we now have that α k = Λ k · ν ◦ s , where Λ k : gl ((Rn ) ∗ ) → (Rn ) ∗
is the linear map given by Λ k (F) = F(ek ) for F ∈ gl ((Rn ) ∗ ) . Hence it follows by
Proposition 5.62 and Theorem 10.109 that
∇ α k (p)(v) = ∇(ν ◦ s)(p)(v)(ek ) = (ν ◦ s)(p) ◦ (ψ ∗ · ω

e )(p)(v)(ek )
n
e ji (p)(v) (ν ◦ s)(p) ◦ ψ ∗ (Fji )(ek ) = − ∑ ω
=∑ω e jk (p)(v) (ν ◦ s)(p)(e j )
ij j=1
n
e jk (p)(v) α j (p)
=−∑ ω
i=1
for p ∈ V , v ∈ Tp M and k = 1, ... , n.

M n , and that s = (X1 , ... , Xn ) is a frame field on an open subset V of M with a dual
local basis (α 1 , ... , α n ) in the cotangent bundle π ∗ : T ∗ M → M . Let E = {e1 , ... , en }
be the standard basis for Rn and D = {Fji |1 ≤ i, j ≤ n} be the basis for gl (Rn )
defined in Example 8.10 (b), where Fji : Rn → Rn is the linear map given by Fji (ek ) =
δ j k ei for k = 1, ... , n. If ω is the connection form of H and ω e = s∗ (ω ) is the pull-
back of ω by s, then we have that
n
∇Xk α i = − ∑ Γki j α j ,
j=1
where Γki j are the connection coefficients of H with respect to the frame field s =
(X1 , ... , Xn ) given by
Γki j = ω
e ji (Xk )
for 1 ≤ i, j, k ≤ n, where ω e with respect to the basis D.

i
(X1 , ... , Xn ) on an open subset V of M. Let (α 1 , ... , α n ) be the dual local basis in the
cotangent bundle π ∗ : T ∗ M → M . If X is a vector field and α a covector field on V
with
n n
X = ∑ ai Xi and α = ∑ b i α i ,
i=1 i=1
FIBRE BUNDLES 407
then we have that !

n n
j
∇X α = ∑ ∑ a b i;j αi
i=1 j=1
where
n
b i ; j = X j (b i ) − ∑ b k Γ jki
k=1
for 1 ≤ i, j ≤ n.
PROOF : By Proposition 10.80 and Corollary 10.116 we have that
∇X α = ∑ a j ∇X j (b i α i ) = ∑ a j { X j (b i ) α i + b i ∇X j α i }
ij ij
!
j i j k j
= ∑ a X j (b i ) α + ∑ a b k ∇X j α = ∑ a X j (b i ) − ∑ b k Γ jki αi .
ij kj ij k

If X is a vector field and α a covector field on U with
n n
X = ∑ ai ∂ i and α = ∑ b i dxi ,
i=1
∂x i=1
then we have that !

n n
j
∇X α = ∑ ∑ a b i;j dxi
i=1 j=1
where
n
∂ bi
b i;j = − ∑ b k Γ jki
∂xj k=1
for 1 ≤ i, j ≤ n.

i
(X1 , ... , Xn ) on an open subset V of M. Let (α 1 , ... , α n ) be the dual local basis in the
cotangent bundle π ∗ : T ∗ M → M . If X is a vector field and T a tensor field of type
( kl ) on V with
n
X = ∑ ai Xi
i=1
and
j ,..., j
T= ∑ B i11 ,...,ikl α i1 ⊗ · · · ⊗ α ik ⊗ X j1 ⊗ · · · ⊗ X j l ,
i1 ,...,ik
j1 ,..., j l
then we have that

!
n
h j ,..., j
∇X T = ∑ ∑ a B i11 ,...,ikl; h α i1 ⊗ · · · ⊗ α ik ⊗ X j1 ⊗ · · · ⊗ X j l
i1 ,...,ik h=1
j1 ,..., j l
where
l n
j ,..., j j ,... , j j ,... , j , j, jm+1 ,..., j l j
B i11 ,...,ikl; h = Xh (B i11 ,..., ikl ) + ∑ ∑ B i11 ,...,ikm−1 Γ hmj
m=1 j=1
k n
j ,..., j i
− ∑ ∑ B i11 ,... ,im−1
l
,i,im+1 ,...,ik Γ h im .
m=1 i=1
PROOF : Using that

j ,..., j
B i11 ,...,ikl; h = ∇Xh T Xi1 , ... , Xik , α j1 , ... , α j l ,
the result follows from Proposition 10.83 (5) and Corollaries 10.111 and 10.116.

If X is a vector field and T a tensor field of type ( kl ) on U with
n
X = ∑ ai ∂ i
i=1
∂x
and
B i11 ,...,ikl dxi1 ⊗ · · · ⊗ dxik ⊗ ∂ j1 ⊗ · · · ⊗ ∂ j l ,
j ,..., j
T= ∑ ∂x ∂x
i1 ,...,ik
j1 ,..., j l
then we have that

!
n
dxi1 ⊗ · · · ⊗ dxik ⊗ ∂ j1 ⊗ · · · ⊗ ∂ j l
h j ,..., j
∇X T = ∑
i1 ,...,ik
∑ a B i11 ,...,ikl; h
∂x ∂x
h=1
j1 ,..., j l
where
j ,... , j
j ,..., j
∂ B i 1 ,..., i l l n
j ,... , j , j, jm+1 ,..., j l j
1
B i11 ,...,ikl; h =
∂ xh
k
+ ∑ ∑ B i11 ,...,ikm−1 Γ hmj
m=1 j=1
k n
j ,..., j i
− ∑ ∑ B i11 ,... ,im−1
l
,i,im+1 ,...,ik Γ h im .
m=1 i=1
FIBRE BUNDLES 409

i be the components
G → Aut (W ) of G on a finite dimensional vector space W . Let Γ jk
of H with respect to a frame field s = (X1 , ... , Xn ) on an open subset V of M. If
F : I → T M is a lifting of a smooth curve γ : I → M lying in V, with
n n
F = ∑ ai {Xi ◦ γ } and γ ′ = ∑ bi {Xi ◦ γ }
i=1 i=1
for smooth functions ai : I → R and bi : I → R , and we use the standard local chart
(r, I) on I where r : I → R is the inclusion map, then we have that
!
n n
DF
=∑ ai ′ + ∑ i
{Γ jk ◦ γ } b j ak {Xi ◦ γ } .
dr i=1 j,k=1
PROOF : Let Fi : I → T M be the lifting of γ given by Fi = Xi ◦ γ for i = 1, ... , n. Then

it follows from Remark 2.83 and Propositions 10.87 and 10.88 that
!
n n n
DFk
(t) = ∇γ ′ (t) Xk = ∑ b j (t) ∇Fj (t) Xk = ∑ ∑ b j (t) Γ jk
i
(γ (t)) Xi (γ (t))
dr j=1 i=1 j=1
for t ∈ I and k = 1, ... , n, which implies that

!
n n n
DF k′ k DFk i′ i j k
dr
= ∑ {a Fk + a
dr
}=∑ a + ∑ {Γ jk ◦ γ} b a {Xi ◦ γ } .
k=1 i=1 j,k=1
10.122 Proposition Let H be a linear connection on a smooth manifold M n , and

let Γki j be the connection coefficients of H with respect to a local chart (x,U) on
M. Then the components T j ik and R ji kl of the torsion and curvature forms of H with
respect to (x,U) are given by
(1) T jik = Γ ji k − Γki j
∂ Γlij ∂ Γki j n
(2) R ji kl = − + ∑ ( Γlhj Γki h − Γkhj Γlih )
∂ xk ∂ xl h=1
PROOF : (1) By Proposition 10.102 we have that

n
∂ ∂ ∂ ∂
T
∂xj
,
∂ xk
= ∇∂
∂ xk
−∇∂
∂xj
= ∑ ( Γ ji k − Γki j ) ∂∂xi
∂xj ∂ xk i=1
(2) By Propositions 10.107 and 10.80 we have that

R ∂k , ∂ l ∂ j = ∇ ∂ ∇ ∂ ∂ j − ∇ ∂ ∇ ∂ ∂
∂x ∂x ∂x ∂x ∂xj
∂ xk ∂ xl ∂ xl ∂ xk
! !
n n
∂ ∂
= ∇∂ ∑ Γlhj −∇∂ ∑ Γkhj
∂ xh ∂ xh
∂ xk h=1 ∂ xl h=1
( ! !)
n ∂ Γlij n ∂ Γki j n
∂
= ∑ +∑ Γlhj Γki h − +∑ Γkhj Γlih
i=1 ∂x k
h=1 ∂x l
h=1
∂ xi
10.123 Proposition Suppose that H is a linear connection on a smooth manifold

M n , and that sk : Vk → L (M) for k = 1, 2 are two sections in the frame bundle
π : L (M) → M with associated local trivializations (tk , π −1 (Vk )) which are Gl (Rn )-
related with transition map φ : V1 ∩V2 → Gl (Rn ). If Γki j : V1 → R and Γ′ ki j : V2 → R
are the connection coefficients of H with respect to the frame fields s1 = (X1 , ... , Xn )
and s2 = (X1′ , ... , Xn′ ), respectively, then
r
Γki j (p) = ∑ Γ′ts (p) φ (p) tk φ (p) sj {φ (p)−1 } ir + ∑ Xk (φ rj )(p) {φ (p)−1 } ir (1)
rst r
for p ∈ V1 ∩V2 and 1 ≤ i, j, k ≤ n, where the components in gl (Rn ) are with respect
to the basis D = {F ji |1 ≤ i, j ≤ n} defined in Example 8.10 (b), where F ji : Rn → Rn
is the linear map given by F ji (ek ) = δ j k ei for k = 1, ... , n.
Conversely, assume that we have a family of frame fields defined on sets which
form an open cover of M, and for each frame field a family { Γki j , 1 ≤ i, j, k ≤ n } of
smooth functions satisfying the transformation rule given in formula (1). Then there
is a unique linear connection H on M having these Γki j as connection coefficients.
PROOF : e k = s k∗ (ω ) for k = 1, 2 , then it

If ω is the connection form of H and ω
follows from 5.51, Lemma 2.84 and Example 9.8 (c) that formula (2) in Proposition
10.56 is equivalent to
e 1 (p)(v) = φ (p) −1 ◦ ω
ω e 2 (p)(v) ◦ φ (p) + φ (p) −1 ◦ d φ (p)(v) (2)
for p ∈ V1 ∩V2 and v ∈ Tp M . So if we let v = Xk (p) and use that F ji ◦ Fsr = δ j r Fsi ,
we obtain the equivalent formula
e 1 ) ji (Xk )(p) =
(ω ∑ {φ (p)−1 } ir (ωe 2 )sr (Xk )(p) φ (p) sj
rs
−1
+ ∑ {φ (p) } ir Xk (φ rj )(p)
r
for p ∈ V1 ∩V2 and 1 ≤ i, j, k ≤ n. Since we have that

Xk (p) = ∑ φ (p) tk Xt′ (p)
t
FIBRE BUNDLES 411
for p ∈ V1 ∩V2 and 1 ≤ k ≤ n by Corollary 10.54, we therefore see that formula (1)
in our proposition is equivalent to formula (2) in Proposition 10.56.
10.124 Corollary Let H be a linear connection on a smooth manifold M n . If

i
Γk j : U → R and Γ k j : U ′ → R are the connection coefficients of H with respect to
i ′
the local charts (x,U) and (x′ ,U ′ ) on M , then

r
r ∂ x′ t ∂ x′ s ∂ xi ∂ 2 x′ ∂ xi
Γki j = ∑ Γ′ts k
∂x ∂x j ∂ x′ r + ∑ k j ∂ x′ r (1)
rst r ∂x ∂x
on U ∩U ′ for 1 ≤ i, j, k ≤ n.
Conversely, assume that we for each local chart in an atlas on M have a family
{ Γki j , 1 ≤ i, j, k ≤ n } of smooth functions satisfying the transformation rule given
in formula (1). Then there is a unique linear connection H on M having these Γki j as
connection coefficients.
PROOF : To obtain formula (1), apply Proposition 10.123 to the coordinate frame
fields
∂ ∂ ∂ ∂
s1 = 1 , ... , n and s2 = ′1
, ... , ′n
∂x ∂x ∂x ∂x
−1
where φ (p) = tx′ ,p ◦ tx,p = D(x′ ◦ x−1 )(x(p)) for p ∈ U ∩ U ′ .
KOSZUL CONNECTIONS
10.125 Definition A Koszul connection on a smooth manifold M is a map ∇ :
T1 (M) × T1 (M) → T1 (M) , where ∇ ( X,Y ) is denoted by ∇X Y , satisfying
(1) ∇X1 +X2 Y = ∇X1 Y + ∇X2 Y
(2) ∇X (Y1 + Y2 ) = ∇X Y1 + ∇X Y2
(3) ∇ f X Y = f ∇X Y
(4) ∇X ( fY ) = X( f ) Y + f ∇X Y
for all vector fields X, X1 , X2 ,Y,Y1 ,Y2 ∈ T1 (M) and smooth functions f ∈ F (M).
10.126 Remark Given a Koszul connection ∇ and a vector field Y on a smooth

manifold M, the map F : T1 (M) → T1 (M) defined by F(X) = ∇X Y for X ∈ T1 (M)
is linear over F (M). By Proposition 5.36 there is therefore a unique bundle-valued
1-form ∇ Y ∈ Ω 1 (M; T M) with F = (∇ Y ) M . Denoting ∇ Y (p)(v) by ∇v Y for p ∈ M
and v ∈ Tp M , we therefore have that
(∇X Y ) p = ∇Xp Y
for p ∈ M. Hence the value of ∇X Y at a point p ∈ M only depends on the value of X

at p and, as we shall see in the next lemma, on the local behavior of Y around p.
10.127 Lemma Let ∇ be a Koszul connection on a smooth manifold M, and let
X be a vector field on M. If Y1 ,Y2 are vector fields on M with Y1 | V = Y2 | V for an
open subset V of M, then we have that (∇X Y1 )| V = (∇X Y2 )| V .
PROOF : Considering the difference Y2 − Y1 , it is clearly enough to prove that
(∇X Y )| V = 0 for each vector field Y on M with Y | V = 0. To show this, let p ∈ V
and choose a smooth function h ∈ F (M) with h = 1 on M −V and h(p) = 0. Such a
function h exists by Corollary 2.24. Then we have that
∇X Y (p) = ∇X (hY )(p) = X(h)(p) Y (p) + h(p) ∇X Y (p) = 0
10.128 Proposition A Koszul connection ∇ on a smooth manifold M induces a

unique Koszul connection ∇V on each open subset V of M so that
(∇X Y )| V = (∇V ) (X | V ) (Y | V )
for every vector field X and Y on M. The connection ∇V is usually denoted simply
by ∇.
PROOF : Let X e and Ye be vector fields on V . Given a point p ∈ V , we choose vector
fields X and Y on M which coincide with Xe and Ye , respectively, on an open neigh-
bourhood U of p with U ⊂ V using Proposition 2.56 (1), and we define
(∇V ) Xe Ye (p) = ∇X Y (p) .
By Remark 10.126 and Lemma 10.127 we see that ∇V is well defined, being indepen-
dent of the choice of the vector fields X and Y . Using the same choice for every point
p in the open set U defined above, we also see that ∇V is a Koszul connection.
10.129 Definition Let ∇ be a Koszul connection on a smooth manifold M, and

let s = (X1 , ... , Xn ) be a frame field on an open subset V of M. Then the smooth
functions Γki j : V → R defined by
n
∇Xk X j = ∑ Γki j Xi
i=1
for 1 ≤ j, k ≤ n, are called the components or the coefficients of ∇ with respect to

the frame field s = (X1 , ... , Xn ).
10.130 Proposition Let ∇ be a Koszul connection on a smooth manifold M, and
let Γki j be the components of ∇ with respect to a frame field s = (X1 , ... , Xn ) on an
open subset V of M. If X and Y are vector fields on V with
n n
X = ∑ ai Xi and Y = ∑ bi Xi ,
i=1 i=1
FIBRE BUNDLES 413
then we have that !

n n
j i
∇X Y = ∑ ∑ a b ;j Xi
i=1 j=1
where
n
bi ; j = X j (bi ) + ∑ bk Γ jk
i
k=1
for 1 ≤ i, j ≤ n.
PROOF : By Definitions 10.125 and 10.129 we have that
∇X Y = ∑ a j ∇X j (bi Xi ) = ∑ a j { X j (bi ) Xi + bi ∇X j Xi }
ij ij
!
j i j k j i k i
= ∑ a X j (b ) Xi + ∑ a b ∇X j Xk = ∑ a X j (b ) + ∑ b Γ jk Xi .
ij kj ij k
10.131 Proposition Suppose that ∇ is a Koszul connection on a smooth manifold

M n , and that sk : Vk → L (M) for k = 1, 2 are two sections in the frame bundle
π : L (M) → M with associated local trivializations (tk , π −1 (Vk )) which are Gl (Rn )-
related with transition map φ : V1 ∩V2 → Gl (Rn ). If Γki j : V1 → R and Γ′ ki j : V2 → R
are the components of ∇ with respect to the frame fields s1 = (X1 , ... , Xn ) and s2 =
(X1′ , ... , Xn′ ), respectively, then
r
Γki j (p) = ∑ Γ′ts (p) φ (p) tk φ (p) sj {φ (p)−1 } ir + ∑ Xk (φ rj )(p) {φ (p)−1 } ir (1)
rst r
for p ∈ V1 ∩V2 and 1 ≤ i, j, k ≤ n, where the components in gl (Rn ) are with respect
to the basis D = {F ji |1 ≤ i, j ≤ n} defined in Example 8.10 (b), where F ji : Rn → Rn
is the linear map given by F ji (ek ) = δ j k ei for k = 1, ... , n.
PROOF : By Corollary 10.54 we have that
n n
X j (p) = ∑ φ (p) sj Xs′ (p) and Xk (p) = ∑ φ (p) tk Xt′ (p)
s=1 t=1
for p ∈ V1 ∩V2 and 1 ≤ j, k ≤ n, so that

r
∇Xk X j (p) = ∑ φ (p) tk Xt′ (φ rj )(p) + ∑ φ (p) sj Γ′ ts (p) Xr′ (p)
rt s

r
=∑ ∑ Γ′ts (p) φ (p) tk φ (p) sj + Xk (φ )(p) Xr′ (p)r
j
r st
by Proposition 10.130. Combining this with

n
Xr′ (p) = ∑ {φ (p)−1 } ir Xi (p) ,
i=1
we obtain formula (1).
10.132 Corollary The covariant derivative ∇ of a linear connection H on a

smooth manifold M is a Koszul connection on M. Conversely, given a Koszul con-
nection ∇ on M, there is a unique linear connection H on M having ∇ as its covariant
derivative.
PROOF : The first assertion follows from Proposition 10.80, and the second assertion
from Propositions 10.123 and 10.131.
STRUCTURE EQUATIONS
10.133 Lemma Let H be a linear connection on a smooth manifold M n . If F ∈
gl (Rn ) and ξ ∈ Rn , we have that
[ σ (F), B(ξ ) ] = B(F(ξ )) .
PROOF : According to Proposition 4.86 we have that [ σ (F), B(ξ ) ] = L σ (F) B(ξ ),
and the flow γ : R × L (M) → L (M) of the fundamental vector field σ (F) is given
by
γ (t, u) = u ◦ exp (t F)
for t ∈ R and u ∈ L (M) by Proposition 9.39.
Now fix u ∈ L (M) , and let c : R → Tu L (M) be the curve defined by
c(t) = [ γ t∗ B(ξ ) ] u = [ R ∗exp (t F) B(ξ ) ] u = [ (R exp (−t F) ) ∗ B(ξ ) ] u

= B(exp (t F) ξ ) u = Λu ◦ φF (t)
for t ∈ R , where φF : R → Gl (Rn ) is the one-parameter subgroup defined in Propo-

sition 8.29, and Λu : gl (Rn ) → Tu L (M) is the linear map given by
Λu (G) = B(G(ξ )) u
for G ∈ gl (Rn ) . Using Proposition 4.76 it follows that
[ L σ (F) B(ξ ) ] u = c′ (0) = Λu ◦ φ F′ (0) = B(F(ξ )) u
if we canonically identify Tc(0) (Tu L (M)) with Tu L (M) as in Lemma 2.84.

FIBRE BUNDLES 415
10.134 Lemma Let π : P → M be a principal G-bundle with a connection H. If

X ∈ g and Y is a horizontal vector field on P, then the vector field [ σ (X),Y ] is also
horizontal.
PROOF : According to Proposition 4.86 we have that [ σ (X),Y ] = L σ (X) Y , and the
flow γ : R × P → P of the fundamental vector field σ (X) is given by
γ (t, u) = u exp (t X)
for t ∈ R and u ∈ P by Proposition 9.39.

Now fix u ∈ P, and let c : R → Tu P be the curve defined by
c(t) = [ γ t∗ Y ] u = [ R ∗exp (t X) Y ] u = (R exp (−t X) ) ∗ Y u exp (t X)
for t ∈ R. By Definition 10.32 (ii) we know that c actually is a curve which lies in
Hu , and therefore
[ L σ (X) Y ] u = c′ (0) ∈ Hu
by Proposition 4.76 if we canonically identify Tc(0) (Tu P) with Tu P as in Lemma
2.84.
10.135 Theorem (The first structure equation) Let H be a linear connection

on a smooth manifold M n , and let ω , θ and Θ be the connection form, the dual form
and the torsion form of H. Then we have that
d θ (u)(w1 , w2 ) = −{ ω (u)(w1)(θ (u)(w2 )) − ω (u)(w2 )(θ (u)(w1 ))} + Θ(u)(w1 , w2 )
for every u ∈ L (M) and w1 , w2 ∈ Tu L (M).

PROOF : If w1 , w2 ∈ Hu , then ω (u)(wi ) = 0 and h(wi ) = wi for i = 1, 2. The structure
equation therefore follows from the definition of the torsion form Θ in this case.
Suppose next that w1 , w2 ∈ Vu . Then the right-hand side of the structure equation
is 0 since h(wi ) = 0 and θ (u)(wi ) = 0 for i = 1, 2. To prove that the left-hand side is
also 0 , let Fi = ω (u)(wi ) so that σ (Fi ) u = wi for i = 1, 2. Since σ (F) is a vertical
vector field on L (M) so that θ ( σ (F))(u) = 0 for every u ∈ L (M) and F ∈ gl (Rn ) ,
and σ : gl (Rn ) → T 1 L (M) is a Lie algebra homomorphism, it follows from Re-
mark 5.86 and Lemma 2.78 (2) that
d θ (u)(w1 , w2 ) = d θ ( σ (F1 ) , σ (F2 ))(u)

= σ (F1 ) ( θ ( σ (F2 )))(u) − σ (F2 ) ( θ ( σ (F1 )))(u) − θ ( [ σ (F1 ), σ (F2 ) ] )(u) = 0
from which we obtain the structure equation.

Finally, suppose that w1 ∈ Vu and w2 ∈ Hu , and let F = ω (u)(w1 ) and ξ =
θ (u)(w2 ) so that σ (F) u = w1 and B(ξ ) u = w2 . Since θ ( B(ξ )) and θ ( σ (F)) are
constant functions on L (M) with values ξ and 0, respectively, it follows from Re-
mark 5.86 and Lemmas 2.78 (2) and 10.133 that
d θ (u)(w1 , w2 ) = d θ ( σ (F), B(ξ ))(u)

= σ (F) ( θ ( B(ξ )))(u) − B(ξ ) ( θ ( σ (F)))(u) − θ ( [ σ (F), B(ξ ) ] )(u)
= − θ (u) (B(F(ξ )) u ) = −F(ξ ) = − ω (u)(w1 ) ( θ (u)(w2 )) .
From this we obtain the structure equation also in this case since ω (u)(w2 ) = 0,
θ (u)(w1 ) = 0 and h(w1 ) = 0 so that Θ(u)(w1 , w2 ) = 0. Since both sides of the struc-
ture equation are bilinear and skew symmetric as a function of w1 and w2 , the theorem
follows from these three cases.
10.136 Remark The first structure equation can also be written in the form
dθ = − ω ∧ θ + Θ
ν : gl (Rn ) × Rn → Rn
given by ν (F, ξ ) = F(ξ ).
10.137 Corollary Let H be a linear connection on a smooth manifold M n , and let

ω , θ and Θ be the connection form, the dual form and the torsion form of H. Then
we have that
n
d θ i = − ∑ ω ji ∧ θ j + Θ i
j=1
for i = 1, ..., n , where θ i and Θ i are the components of θ and Θ with respect to the
standard basis E = {e1 , ... , en } for Rn , and ω ji are the components of ω with respect
to the basis D = {Fji |1 ≤ i, j ≤ n} for gl (Rn ) defined in Example 8.10 (b), where
Fji : Rn → Rn is the linear map given by Fji (ek ) = δ j k ei for k = 1, ... , n.
PROOF : Since ν (Fji , ek ) = δ j k ei for 1 ≤ i, j, k ≤ n , it follows from Proposition 5.68

that
(ω ∧ θ )(u)(w1 , w2 ) = ∑ (ω ji ∧ θ j )(u)(w1 , w2 ) ei
ij
10.138 Theorem (The second structure equation) Let π : P → M be a principal

G-bundle with a connection H, and let ω and Ω be the connection form and the
curvature form of H. Then we have that
d ω (u)(w1 , w2 ) = −[ ω (u)(w1 ), ω (u)(w2 ) ] + Ω(u)(w1 , w2 )
for every u ∈ P and w1 , w2 ∈ Tu P.

FIBRE BUNDLES 417
PROOF : If w1 , w2 ∈ Hu , then ω (u)(wi ) = 0 and h(wi ) = wi for i = 1, 2. The structure

equation therefore follows from the definition of the curvature form Ω in this case.
Suppose next that w1 , w2 ∈ Vu , and let Xi = ω (u)(wi ) so that σ (Xi ) u = wi for
i = 1, 2. Since ω ( σ (Xi )) is a constant function on P with value Xi for i = 1, 2, and
σ : g → T 1 P is a Lie algebra homomorphism, it follows from Remark 5.86 and
Lemma 2.78 (2) that
d ω (u)(w1 , w2 ) = d ω ( σ (X1 ), σ (X2 ))(u)

= σ (X1 ) ( ω ( σ (X2 )))(u) − σ (X2 ) ( ω ( σ (X1 )))(u) − ω ( [ σ (X1 ), σ (X2 ) ] )(u)
= − ω (u) (σ ([ X1 , X2 ] ) u ) = −[ X1 , X2 ] = −[ ω (u)(w1 ), ω (u)(w2 ) ] .
From this we obtain the structure equation since h(wi ) = 0 for i = 1, 2 so that
Ω(u)(w1 , w2 ) = 0.
Finally, suppose that w1 ∈ Vu and w2 ∈ Hu . Then the right-hand side of the struc-
ture equation is 0 since h(w1 ) = 0 and ω (u)(w2 ) = 0. To prove that the left-hand
side is also 0 , let X = ω (u)(w1 ) so that σ (X) u = w1 , and choose a horizontal vector
field Y on P with Yu = w2 using Proposition 2.56 (2). Since ω ( σ (X)) is a constant
function on P with value X and the vector field [ σ (X),Y ] is horizontal by Lemma
10.134, it follows from Remark 5.86 and Lemma 2.78 (2) that
d ω (u)(w1 , w2 ) = d ω ( σ (X),Y )(u)

= σ (X) ( ω (Y ))(u) − Y ( ω ( σ (X)))(u) − ω ( [ σ (X),Y ] )(u) = 0
which proves the structure equation also in this case. Since both sides of the structure
equation are bilinear and skew symmetric as a function of w1 and w2 , the theorem
follows from these three cases.
10.139 Remark The formula in Theorem 10.138 is also called the structure
equation of Elie Cartan, and it can be written in the form
d ω = − 21 ω ∧ ω + Ω
ν : g×g → g
given by ν (X,Y ) = [ X,Y ].

let ω and Ω be the connection form and the curvatore form of H. Then we have that
n
d ω ji = − ∑ ωki ∧ ω jk + Ω ji
k=1
for i, j = 1, ..., n , where ω ji and Ω ji are the components of ω and Ω with respect
to the basis D = {Fji |1 ≤ i, j ≤ n} for gl (Rn ) defined in Example 8.10 (b), where
Fji : Rn → Rn is the linear map given by Fji (ek ) = δ j k ei for k = 1, ... , n.
PROOF : Since
ν (Fji , F lk ) = δ j k F li − δl i Fjk
for 1 ≤ i, j, k, l ≤ n by Example 8.10 (b), it follows from Proposition 5.68 that
(ω ∧ ω )(u)(w1 , w2 ) = ∑ (ωki ∧ ω lk )(u)(w1 , w2 ) F li

ikl
−∑ (ω jl ∧ ω lk )(u)(w1 , w2 ) F jk = 2 ∑ (ωki ∧ ω jk )(u)(w1 , w2 ) Fji

jkl i jk
10.141 Theorem (Bianchi’s 1st identity) Let H be a linear connection on a

smooth manifold M n , and let Ω, θ and Θ be the curvature form, the dual form and
the torsion form of H. Then we have that
DΘ = Ω ∧ θ .
PROOF : If we apply d on both sides of the first structure equation
dθ = − ω ∧ θ + Θ
and use Propositions 5.64 (3) and 5.69, we obtain
0 = −(d ω ) ∧ θ + ω ∧ (d θ ) + dΘ .
If h : T L (M) → T L (M) is the horizontal projection determined by H, we therefore

have that
DΘ = h∗ (dΘ) = h∗ (d ω ) ∧ h∗ (θ ) − h∗ (ω ) ∧ h∗ (d θ ) = Ω ∧ θ
by 5.65 and Proposition 10.41, since h∗ (d ω ) = Dω = Ω by the definition of Ω, and

h∗ (θ ) = θ and h∗ (ω ) = 0 as θ is horizontal and ω is vertical.
10.142 Theorem (Bianchi’s 2nd identity) Let H be a connection in the principal

G-bundle π : P → M, and let Ω be the curvature form of H. Then we have that
DΩ = 0 .
PROOF : If we apply d on both sides of the second structure equation
d ω = − 12 ω ∧ ω + Ω
FIBRE BUNDLES 419
and use Propositions 5.64 (3) and 5.69, we obtain
0 = − 21 (d ω ) ∧ ω + 12 ω ∧ (d ω ) + dΩ .
If h : T P → T P is the horizontal projection determined by H, we therefore have that
DΩ = h∗ (dΩ) = 21 h∗ (d ω ) ∧ h∗ (ω ) − 12 h∗ (ω ) ∧ h∗ (d ω ) = 0
by 5.65 since h∗ (ω ) = 0 as ω is vertical.
10.143 Proposition Let π : L (M) → M be the frame bundle over a smooth

manifold M n with a connection H, and let π ′ : E → M be the associated vector bundle
obtained from a representation ψ : Gl (Rn ) → Aut (W ) of the structure group Gl (Rn )
on a finite dimensional vector space W .
If ω ∈ Ω k (M; E) is a bundle-valued k-form on M with values in E, we let η ∈
k
Ω (L (M);W ) be the corresponding vector-valued k-form on L (M) with values in
W . Considering ω as a section in the vector bundle π k : T k (T M; E) → M, we have
that
∇v ω ( v1 , ... , vk ) = µ u 0 ◦ w( η (B(ξ1 ), ... , B(ξk )))
for p ∈ M and v, v1 , ... , vk ∈ Tp M , where u 0 ∈ L (M) , w ∈ Hu 0 and ξ1 , ... , ξk ∈ Rn
are chosen so that π (u 0 ) = p , π∗ (w) = v and u 0 (ξi ) = vi for i = 1, ..., k.
PROOF : By Definition 10.77 and Example 10.29 (i) the section ω has a correspond-
e on L (M) with values in T k (Rn ;W ) which is given
ing vector-valued function η
by
e (u) = µ −1
η u ◦ ω (π (u)) ◦ u
k
for u ∈ L (M) . If Λ : T k (Rn ;W ) → W is the linear map defined by
Λ(F) = F(ξ1 , ... , ξk )
for F ∈ T k (Rn ;W ) , we therefore have that
(Λ · η e (u)(ξ1 , ... , ξk ) = µ −1
e )(u) = η u ◦ ω (π (u))(u(ξ1 ), ... , u(ξk ))
= η (u)(B(ξ1 ) u , ... , B(ξk ) u ) = η (B(ξ1 ), ... , B(ξk )) (u)
for u ∈ L (M) . Now we have that
e (u 0 )(w) ◦ {u −1
∇v ω = µ u 0 ◦ d η k
0 } ,
so using Proposition 5.62 we obtain
e (u 0 )(w) ( ξ1 , ... , ξk ) = µ u 0 ◦ (Λ · d η
∇v ω ( v1 , ... , vk ) = µ u 0 ◦ d η e )(u 0 )(w)
e )(u 0 )(w) = µ u 0 ◦ w(Λ · η
= µ u 0 ◦ d(Λ · η e ) = µ u 0 ◦ w( η (B(ξ1 ), ... , B(ξk ))) .

manifold M n with a connection H, and let T be the torsion form of H on M. Then we
have that
T (p)(v1 , v2 ) = −π∗ ([ B(ξ1 ), B(ξ2 ) ] u )
for p ∈ M and v1 , v2 ∈ Tp M , where u ∈ L (M) and ξ1 , ξ2 ∈ Rn are chosen so that
π (u) = p and u(ξi ) = vi for i = 1, 2.
PROOF : Using Remark 5.86, Lemma 2.78 (2) and Proposition 10.93, we have that
T (p)(v1 , v2 ) = u ◦ d θ ( B(ξ1 ), B(ξ2 ))(u)

= u ◦ B(ξ1 ) u ( θ (B(ξ2 ))) − u ◦ B(ξ2 ) u ( θ (B(ξ1 ))) − u ◦ θ (u)( [ B(ξ1 ), B(ξ2 ) ] u )
= −π∗ ([ B(ξ1 ), B(ξ2 ) ] u ) ,
since θ (B(ξi ))(u) = ξi for u ∈ L (M) and i = 1, 2 by Proposition 10.96.

manifold M n with a connection H, and let π ′ : E → M be the associated vector bundle
obtained from a representation ψ : Gl (Rn ) → Aut (W ) of the structure group Gl (Rn )
on a finite dimensional vector space W . Let T be the torsion form of H on M .
If ω ∈ Ω 2 (M; E) is a bundle-valued 2-form on M with values in E, we let
η ∈ Ω 2 (L (M);W ) be the corresponding vector-valued 2-form on L (M) with val-
ues in W . Considering ω as a section in the vector bundle π 2 : T 2 (T M; E) → M,
we have that
µ u ◦ Dη (u)(w1 , w2 , w3 ) =
∑ {( ∇vσ (1) ω ) ( vσ (2) , vσ (3) ) + ω (p)( T (p)(vσ (1) , vσ (2) ), vσ (3) ) }

σ ∈A3
for p ∈ M and v1 , v2 , v3 ∈ Tp M , where u ∈ L (M) and w1 , w2 , w3 ∈ Hu are chosen so

that π (u) = p and π∗ (wi ) = vi for i = 1, 2, 3 , and A3 denotes the alternating group of
three elements.
PROOF : Choose ξ1 , ξ2 , ξ3 ∈ Rn so that u(ξi ) = vi for i = 1, 2, 3. Using Remark 5.85
and Propositions 10.143 and 10.144, we then have that
FIBRE BUNDLES 421
µ u ◦ Dη (u)(w1 , w2 , w3 ) = µ u ◦ d η (B(ξ1 ), B(ξ2 ), B(ξ3 ))(u)
= ∑ { µ u ◦ B(ξσ (1) ) u (η (B(ξσ (2) ), B(ξσ (3) )))

σ ∈A3
− µ u ◦ η (u)( [ B(ξσ (1) ), B(ξσ (2) ) ] u , B(ξσ (3) ) u ) }
= ∑ {( ∇vσ (1) ω ) ( vσ (2) , vσ (3) ) + ω (p)( T (p)(vσ (1) , vσ (2) ), vσ (3) ) } .

σ ∈A3
10.146 Corollary Let R and T be the curvature form and the torsion form of a
linear connection H on a smooth manifold M . Then Bianchi’s 1st and 2nd identity
can be written in the form
(1) ∑ R(p)(vσ (1) , vσ (2) ) vσ (3) =

σ ∈A3
∑ {( ∇vσ (1) T ) ( vσ (2) , vσ (3) ) + T (p)( T (p)(vσ (1) , vσ (2) ), vσ (3) ) } , and
σ ∈A3
(2) ∑ {( ∇vσ (1) R ) ( vσ (2) , vσ (3) ) + R(p)( T (p)(vσ (1) , vσ (2) ), vσ (3) ) } = 0
σ ∈A3
respectively, for every p ∈ M and v1 , v2 , v3 ∈ Tp M , where A3 denotes the alternating

group of three elements.
PROOF : Choose u ∈ L (M) and w1 , w2 , w3 ∈ Hu so that π (u) = p and π∗ (wi ) = vi
for i = 1, 2, 3. Then we have that
(2 + 1)!
u ◦ ( Ω ∧ θ )(u)(w1 , w2 , w3 ) = u ◦ A ( Ω ⊗ θ )(u)(w1 , w2 , w3 )
2! 1!
1
= ∑ ε (σ ) {u ◦ Ω(u)(wσ (1) , wσ (2) ) ◦ u−1 } {u ◦ θ (u)(wσ (3) ) }
2! 1!
σ ∈S3
= ∑ {u ◦ Ω(u)(wσ (1) , wσ (2) ) ◦ u−1 } π∗ (wσ (3) ) = ∑ R(p)(vσ (1) , vσ (2) ) vσ (3)
σ ∈A3 σ ∈A3
The corollary now follows from Theorems 10.141 and 10.142 and Proposition
10.145.
GEODESICS
10.147 Definition Let M be a smooth manifold with a linear connection H. Then

a smooth curve γ : I → M defined on an open interval I is called a geodesic on M
if γ ′ is parallel along γ . The geodesic γ is called maximal if it has no extension to a
geodesic on any larger open interval.

manifold M n with a connection H, and let γ : I → M be a smooth curve defined
on an open interval I containing t0 with a horizontal lifting β : I → L (M). Then
γ is a geodesic if and only if β is an integral curve for the basic vector field B(ξ )
determined by the vector ξ ∈ Rn with β (t0 )(ξ ) = γ ′ (t0 ).
PROOF : The proposition follows from Definition 10.71 and Proposition 10.96 which
implies that
π∗ (β ′ (t)) = γ ′ (t) and π∗ (B(ξ ) β (t) ) = β (t)(ξ ) = τ tt0 (γ ′ (t0 ))
for every t ∈ I.
10.149 Proposition Let M n be a smooth manifold with a linear connection H,

and let v ∈ Tp M for a p ∈ M. Then there is a unique maximal geodesic γ : I → M
with γ (0) = p and γ ′ (0) = v. Denoting this geodesic by γv : I(v) → M, we have that
I(t v) = t −1 I(v) when t ∈ R − {0} and I(0) = R , and γ t v (s) = γ v (ts) for s ∈ I(tv).
PROOF : Let π : L (M) → M be the frame bundle over M, choose a u ∈ π −1 (p) and
let ξ ∈ Rn be the vector with u(ξ ) = v. If β : I → L (M) is the unique maximal
integral curve for the basic vector field B(ξ ) with initial condition u, then it follows
from Proposition 10.148 that γ = π ◦ β is the unique maximal geodesic on M with
γ (0) = p and γ ′ (0) = v.
To prove the second part of the proposition, let µ t : R → R be the smooth map
defined by µ t (s) = ts for s ∈ R. Then the horizontal lift β ◦ µ t of γ v ◦ µ t is the
maximal integral curve for the basic vector field B(t ξ ) with initial condition u, since
(β ◦ µ t ) ′ (s) = t β ′ (ts) = t B(ξ ) (β (ts)) = B(t ξ ) (β ◦ µ t (s))
by Definition 2.69. As u(t ξ ) = t v, we conclude that γ t v = γ v ◦ µ t .
10.150 Using the above notation, we let D be the subset of T M given by D =

{v ∈ T M|1 ∈ I(v)} , and we define the exponential map exp : D → M by
exp (v) = γv (1)

FIBRE BUNDLES 423
for v ∈ D. For each p ∈ M, we let D p = D ∩ Tp M and call the restriction exp p =

exp | D p : D p → M the exponential map at p.
10.151 Remark By Proposition 10.149 the maximal geodesic γv : I(v) → M is

given by
γ v (t) = γ t v (1) = exp (t v)
for t ∈ I(v), which is equivalent to the conditions 1 ∈ I(t v) or t v ∈ D.
10.152 Let πe : E (M) → L (M) be the vector bundle induced from the tangent
bundle π : T M → M by the projection π ′ : L (M) → M in the frame bundle, as shown
e = πe ′ −1 (D). Then we have
in the commutative diagram in Example 10.25, and let D
e → L (M) defined by
a map E : D
E(u, v) = β u,v (1)
e where β u,v is the unique horizontal lifting of the maximal geodesic γv

for (u, v) ∈ D,
with β u,v (0) = u. From Proposition 10.148 we know that β u,v is an integral curve for
the basic vector field B(u−1 (v)), and we have a commutative diagram
πe ′ ✲
e
D D
E exp
❄ ❄
π ′✲
L (M) M
10.153 Proposition Let M n be a smooth manifold with a linear connection H.

Then E : De → L (M) and exp : D → M are smooth maps defined on the open
e and D of E (M) and T M , respectively, and D p is an open star- shaped
subsets D
neighbourhood of 0 ∈ Tp M for each p ∈ M.
PROOF : We first show that D e is an open subset of E (M) and that E is smooth. Let
Y be the vector field on L (M) × E (M) defined by
Y (w, u, v) = (B(u−1(v)) w , 0) = (α w−1 ◦ w ◦ u−1(v), 0)
for w ∈ L (M) and (u, v) ∈ E (M) , where (α , π ) is the bundle map from H to T M
defined in Proposition 10.49. The flow γ : D(Y ) → L (M) × E (M) for Y is given by
γ (t, w, u, v) = (β (u−1(v))(t, w), u, v) ,

where β (ξ ) is the flow for the basic vector field B(ξ ) for each ξ ∈ Rn . Hence we
have that
De = (πe , id E (M) )−1 (D 1 (Y ))
and
E(u, v) = π1 ◦ γ1 (u, u, v)
for (u, v) ∈ De , where π1 : L (M) × E (M) → L (M) is the projection on the first
factor, thus completing the proof that E is smooth with an open domain D. e
As the projection πe ′ : E (M) → T M is an open mapping onto T M by Propositions
10.3 and 2.43, it follows that D = πe ′ (De ) is open in T M. For each frame field s : V →
L (M) defined on an open subset V of M we have that
exp (v) = π ′ ◦ E(s ◦ π (v), v)
for v ∈ D ∩ π −1 (V ) , which shows that exp is smooth.
10.154 Proposition For each p ∈ M we have that exp p ∗ 0 : Tp M → Tp M is the

identity map. Hence exp p maps an open star-shaped neighbourhood U p of 0 ∈ Tp M
diffeomorphically onto an open neighbourhood Vp of p ∈ M .
We let log p : Vp → U p denote the inverse of the diffeomorphism induced by exp p .
PROOF : Using the curves γ v and ψv on M and Tp M, respectively, defined in Propo-
sition 10.149 and Lemma 8.36, we have that
exp p ∗ (v) = (exp ◦ ψv ) ′ (0) = γ v′ (0) = v
for every v ∈ Tp M by 2.70 and Proposition 10.151, showing that exp p ∗ 0 = id . The
last part of the proposition now follows from the inverse function theorem.
10.155 Remark If x : Tp M → Rn is a linear isomorphism and y = x ◦ log p , then

(y,Vp ) is a local chart around p on M, called a normal chart or a normal coordinate
system on M. Let E = {e1 , ..., en } be the standard basis for Rn and C = {v1 , ..., vn }
be a basis for Tp M so that x(vi ) = ei for i = 1, ..., n. Then the coordinate map y is
given by
n
y( exp p ( ∑ ai vi )) = a
i=1
for a ∈ x(U p ) .
10.156 By 10.12 we know that the bundle map (πe ′ , π ′ ) between the vector
bundles πe : E (M) → L (M) and π : T M → M induces a linear isomorphism πeu′ :
Eu (M) → Tp M for each p ∈ M and u ∈ L p (M). We have that s u = E ◦ πeu′ −1 ◦ log p :
Vp → L (M) is a frame field where s u (q) is obtained from u by parallel transport
FIBRE BUNDLES 425
along the geodesic γ log p (q) for each q ∈ Vp . Indeed, if v = log p(q) and β u,v is the
unique horizontal lifting of the maximal geodesic γv with β u,v (0) = u, then
−1
s u (q) = E ◦ πeu′ (v) = E(u, v) = β u,v (1) = τ 01 (u)
where τ 01 denotes parallel transport in L (M) along the geodesic γv from 0 to 1, and
π ′ ◦ s u (q) = π ′ ◦ β u,v (1) = γv (1) = exp p (v) = q .
10.157 Proposition Let M be a smooth manifold with a linear connection H.

Then a smooth curve γ : I → M defined on an open interval I is a geodesic on M if
and only if for each local chart (x,U) on M, we have that
n
(xi ◦ γ ) ′′ (t) + ∑ i
Γ jk (γ (t)) (x j ◦ γ ) ′ (t) (xk ◦ γ ) ′ (t) = 0
j,k=1
for t ∈ I ∩ γ −1 (U) and i = 1, ... , n, where Γ jki are the components of H with respect

∂ ∂
to the coordinate frame field s = , ... , n . These are called the geodesic
∂ x1 ∂x
equations.

n
γ ′ (t) = ∑ (xi ◦ γ ) ′ (t) ∂ i .
i=1
∂x
γ (t)
10.158 Proposition Let M n be a smooth manifold with a linear connection H, and

i be the components of H with respect to a normal coordinate system (y,V )
let Γ jk p
around a point p on M as defined in Remark 10.155. Then we have that
i
Γ jk (p) + Γki j (p) = 0
for 1 ≤ i, j, k ≤ n. If the torsion T (p) = 0, then
i
Γ jk (p) = 0
for 1 ≤ i, j, k ≤ n.
PROOF : Assume that y = x ◦ log p where x : Tp M → Rn is a linear isomorphism,
and that x(v) = a for a tangent vector v ∈ Tp M. Then
y ◦ γ v (t) = y ◦ exp p (t v) = t a
for t ∈ I(v), so that
n
i
∑ Γ jk (p) a j ak = 0
j,k=1
for i = 1, ... , n by Proposition 10.157. By choosing a = e j we see that Γ ji j (p) = 0

for 1 ≤ i, j ≤ n. The formula in the proposition is now obtained by choosing a =
e j + ek .
METRICAL CONNECTIONS
10.159 Definition Let π : P → M be a principal G-bundle with a connection
H, and let g be a fibre metric in the associated vector bundle π ′ : E → M obtained
from a representation ψ : G → Aut (W ) of G on a finite dimensional vector space
W . Then H is said to be a metrical connection or a connection compatible with
g if parallel transport in E preserves the fibre metric, i.e., if the parallel transport
τ tt 01 : π ′ −1 (γ (t 0 )) → π ′ −1 (γ (t 1 )) in E from t 0 to t 1 along any smooth curve γ : I → M
on M, where t 0 and t 1 are arbitrary points in the open interval I, is a linear isometry.
By Propositions 2.56 (2) and 10.92 this is equivalent to saying that the function
c : I → R defined by c(t) = g(γ (t)) (F1 (t), F2 (t)) for t ∈ I, is constant for every pair
of parallel liftings F1 and F2 along γ . The function c is usually denoted by <F1 , F2 > .

H, and let g be a fibre metric in the associated vector bundle π ′ : E → M obtained
from a representation ψ : G → Aut (W ) of G on a finite dimensional vector space W .
Then the connection H is compatible with g if and only if the following condition is
satisfied:
If F1 : I → E and F2 : I → E are liftings of an arbitrary smooth curve γ : I → M on
M defined on an open interval I, and if (r, I) is the standard local chart on I where
r : I → R is the inclusion map, then
D E D E
d
< F1 , F2 > = DF1 , F2 + F1 , DF2 . (1)
dr dr dr
PROOF : If F1 and F2 are parallel along γ so that DF 1 / d r = 0 and DF 2 / d r = 0 on

I, then it follows from (1) that < F1 , F2 > is a constant function on I. Hence (1)
implies that H is compatible with g, which completes the proof of the if part of the
proposition.
Suppose conversely that the connection H is compatible with g, and consider a
smooth curve γ : I → M on M defined on an open interval I. Choose an orthonormal
basis B t 0 = {v1 , ..., vn } in the fibre E γ (t 0 ) with

−1 for 1 ≤ i ≤ r
<vi , vi > = εi =
1 for r + 1 ≤ i ≤ n
for a point t 0 ∈ I, and let Vi : I → E be the lifting of γ obtained from vi by parallel

transport along γ for i = 1, ... , n. We know that Vi is smooth since
t
Vi (t) = τ t 0 (vi ) = µ α (t) ◦ µ −1 (v ) = ν wi ◦ α (t)
α (t ) i
0
for t ∈ I, where α : I → P is a horizontal lifting of γ , and ν wi : P → E with wi =

t
µ −1 (v ) ∈ W is the smooth map defined in 10.28. Since τ t 0 is an isometry, we
α (t ) i
0
FIBRE BUNDLES 427
have that B t = {V1 (t), ...,Vn (t)} is an orthonormal basis in the fibre E γ (t) with <
Vi (t),Vi (t) > = εi for i = 1, ... , n and t ∈ I.
Now if F1 : I → E and F2 : I → E are two liftings of γ , then there are smooth maps
a : I → Rn and b : I → Rn so that
n n
F1 = ∑ ai Vi and F2 = ∑ bi Vi
i=1 i=1
by Proposition 10.14. As DV i / d r = 0 on I for i = 1, ... , n, it follows from Proposition

10.88 that
n n
DF1 d ai DF2 d bi
dr
= ∑ Vi and = ∑ Vi .
i=1 d r dr i=1 d r
Using that the liftings V1 , ... ,Vn are everywhere orthonormal, we therefore have that
n n
d d d ai d bi
dr
< F1 , F2 > =
dr
∑ εi ai bi = ∑ εi ( dr
bi + ai
dr
)
i=1 i=1
D E D E
DF1
= , F2 + F1 , DF2
dr dr
which completes the proof of the only if part of the proposition.
10.161 Lemma Let π : P → M be a principal G-bundle with a connection H, and

let g be a fibre metric in the associated vector bundle π ′ : E → M obtained from a
representation ψ : G → Aut (W ) of G on a finite dimensional vector space W . Suppose
that
X < s1 , s2 > = < ∇X s1 , s2 > + < s1 ,∇X s2 > (1)
for all vector fields X ∈ T1 (M) and sections s1 , s2 ∈ Γ(M; E). If f : N → M is a
smooth map from a smooth manifold N, then we have that
Y < F1 , F2 > = < ∇ Y F1 , F2 > + < F1 ,∇ Y F2 > (2)
for all vector fields Y ∈ T1 (N) and liftings F1 , F2 of f .

PROOF : Fix a point p ∈ N, and let σ1 , ... , σn be a local basis for E on an open
neighbourhood U of f (p). If F1 : N → E and F2 : N → E are liftings of f , then there
are smooth maps α : f −1 (U) → Rn and β : f −1 (U) → Rn so that
n n
F1 (q) = ∑ αi (q) σi ◦ f (q) and F2 (q) = ∑ β j (q) σ j ◦ f (q)
i=1 j=1
for q ∈ f −1 (U) by Proposition 10.14.

Choose an open neighbourhood V of f (p) with V ⊂ U. Then it follows by Propo-
sition 2.56 (1) and Corollary 2.25 that there are sections s1 , ... , sn of π ′ on M and
smooth maps a : N → Rn and b : N → Rn which coincide with σ1 , ... , σn and α , β
on V and f −1 (V ), respectively. Using Propositions 10.87 and 10.88 we therefore

have that
n
∇ Yp F1 = ∑ { Yp (ai ) si ( f (p)) + ai(p) ∇ f∗ (Yp ) si } ,
i=1
and we have a similar formula for ∇Yp F2 . Combining these and using (1) it follows
that
< ∇ Yp F1 , F2 (p) > + < F1(p) ,∇ Yp F2 >
= ∑ { Yp (ai ) b j (p) + ai (p) Yp (b j ) } < si ( f (p)), s j ( f (p)) >

i, j
+ ∑ ai (p) b j (p) { < ∇ f∗ (Yp ) si , s j ( f (p)) > + < si ( f (p)),∇ f∗ (Yp) s j > }
i, j
= ∑ { Yp ( ai b j ) < si ( f (p)), s j ( f (p)) > + ai (p) b j (p) f∗ (Yp ) < si , s j > }

i, j
= ∑ Yp ( ai b j <si ◦ f , s j ◦ f > ) = Yp < F1 , F2 >

i, j
which completes the proof of (2), since the point p ∈ N was arbtrary.

and let g be a fibre metric in the associated vector bundle π ′ : E → M obtained from
a representation ψ : G → Aut (W ) of G on a finite dimensional vector space W . Then
the connection H is compatible with g if and only if
X < s1 , s2 > = < ∇X s1 , s2 > + < s1 ,∇X s2 > (1)
for all vector fields X ∈ T1 (M) and sections s1 , s2 ∈ Γ(M; E).
PROOF : Suppose that H is compatible with g, and let p ∈ M and γ : I → M be a
smooth curve defined on an open interval I containing 0 with γ (0) = p and γ ′ (0) = X p .
If (r, I) is the standard local chart on I where r : I → R is the inclusion map, then it
follows from 2.77 and Proposition 10.87 that
d
X p < s1 , s2 > = dr < s1 ◦ γ , s2 ◦ γ >
0
and
D (si ◦ γ )
∇ si = ∇ (si ◦ γ ) =
Xp γ ∗ ( dr
d
0
) si = ∇ drd 0
dr
(0)
for i = 1, 2. Formula (1) hence follows at the arbitrary point p ∈ M by applying

Proposition 10.160 to the liftings s1 ◦ γ and s2 ◦ γ of γ .
Conversely, assuming that (1) is satisfied, it follows from Proposition 10.160 and
Lemma 10.161, applied to an arbitrary smooth curve γ : I → M defined on an open
d on I, that H is compatible with g.
interval I and the vector field dr
FIBRE BUNDLES 429
10.163 Proposition Let X and Y be vector fields and α a 1-form on a pseudo-

Riemannian manifold M with a metric g. Then we have that
(∇X Y ) ♭ = ∇X Y ♭ and (∇X α ) ♯ = ∇X α ♯
when ∇ is the covariant derivative corresponding to a linear connection H on M

compatible with g.
PROOF : By Propositions 10.162 and 10.83 we have that
(∇X Y ) ♭ (Z) = < ∇X Y , Z > = X <Y , Z > − <Y , ∇X Z >
= ∇X (Y ♭ (Z)) − Y ♭ (∇X Z) = (∇X Y ♭ )(Z)
for every vector field Z on M, which completes the proof of the first formula. The sec-
ond formula is obtained from the first with Y = α ♯ by applying g ♯ on both sides.
10.164 Corollary Let X be a vector field and T a tensor field of type (kl ) on a
pseudo-Riemannian manifold M with a metric g. Then we have that
L sr (∇X T ) = ∇X L sr (T ) and R sr (∇X T ) = ∇X R sr (T )
when ∇ is the covariant derivative corresponding to a linear connection H on M

compatible with g.
10.165 Theorem Let M be a pseudo-Riemannian manifold with a metric g, and

let T ∈ Ω 2 (M; T M) be a bundle-valued 2-form on M with values in the tangent
bundle. Then there is a unique linear connection H on M with torsion form T which
is compatible with g. It is given by
1
< ∇X Y , Z > = { X < Z,Y > + Y < Z , X > − Z < X ,Y >
2
+ < X , [ Z ,Y ] > + <Y , [ Z , X ] > + < Z , [ X ,Y ] > (1)
+ < X , T ( Z,Y ) > + <Y , T ( Z , X ) > + < Z , T ( X ,Y ) > }
for every vector field X,Y, Z ∈ T1 (M). In the case when T = 0 , H is called the Levi–
Civita connection on M.
PROOF : If H is compatible with g, then it follows from Proposition 10.162 that
Z < X ,Y > = < ∇Z X ,Y > + < X ,∇Z Y > (2)

for each vector field X,Y, Z ∈ T1 (M). By a permutation of the vector fields X, Y and
Z, we also have that
X < Z,Y > = < ∇X Z,Y > + < Z, ∇X Y > (3)
and
Y < Z , X > = < ∇Y Z , X > + < Z , ∇Y X >, (4)
from which we obtain
2 < ∇X Y , Z > − X < Z,Y > − Y < Z , X > + Z < X ,Y >
= < X, ∇Z Y − ∇Y Z > + <Y , ∇Z X − ∇X Z > + < Z, ∇X Y − ∇Y X > .
Using Proposition 10.102 we therefore obtain formula (1) which shows the unique-
ness of ∇ and therefore of H.
To show the existence of H, we fix vector fields X and Y on M, and let ω :
T1 (M) → F (M) be the map where ω (Z) is the right-hand side of (1) for every
vector field Z ∈ T1 (M). Then ω is linear over F (M). Indeed, if f ∈ F (M), we have
that
1
ω ( f Z) = { X < f Z,Y > + Y < f Z , X > − f Z < X ,Y >
2
+ < X , [ f Z ,Y ] > + <Y , [ f Z , X ] > + < f Z , [ X ,Y ] >
+ < X , T ( f Z,Y ) > + <Y , T ( f Z , X ) > + < f Z , T ( X ,Y ) > } (5)

1
= f ω (Z) + { X( f ) < Z,Y > + Y ( f ) < Z , X >
2
− Y ( f ) < X, Z > − X( f ) <Y , Z > } = f ω (Z).
Hence ω may be considered to be a 1-form on M, and by Proposition 7.18 there is a

unique vector field ∇X Y on M so that ω (Z) = < ∇X Y , Z > for every Z ∈ T1 (M).
Next we need to show that ∇ satisfies the conditions (1) – (4) for a Koszul con-
nection given in Definition 10.125. The first two conditions follow immediately from
formula (1). To prove the last two conditions, let f ∈ F (M). Then we have that
1
<∇fX Y,Z > = { f X < Z,Y > + Y < Z , f X > − Z < f X ,Y >
2
+ < f X , [ Z ,Y ] > + <Y , [ Z , f X ] > + < Z , [ f X ,Y ] >
+ < f X , T ( Z,Y ) > + <Y , T ( Z , f X ) > + < Z , T ( f X ,Y ) > } (6)

1
= < f ∇X Y , Z > + { Y ( f ) < Z, X > − Z( f ) < X ,Y >
2
+ Z( f ) <Y , X > − Y ( f ) < Z, X > } = < f ∇X Y , Z >
FIBRE BUNDLES 431
and
1
< ∇X ( f Y ), Z > = { X < Z, f Y > + f Y < Z , X > − Z < X , f Y >
2
+ <X ,[ Z, f Y ]> + < f Y,[ Z,X ]> + <Z,[ X , f Y ]>
+ < X , T ( Z, f Y ) > + < f Y , T ( Z , X ) > + < Z , T ( X , f Y ) > } (7)

1
= < f ∇X Y , Z > + { X( f ) < Z,Y > − Z( f ) < X ,Y >
2
+ Z( f ) < X ,Y > + X( f ) < Z,Y > } = < X( f ) Y + f ∇X Y , Z >
for every Z ∈ T1 (M), which completes the proof that ∇ is a Koszul connection and
hence gives rize to a unique connection H by Corollary 10.132.
Finally, we must show that H has torsion form T and is compatible with the
metric g. This follows from (1) which implies that
< ∇X Y − ∇Y X − [ X ,Y ], Z > = < T ( X ,Y ), Z >
and
< ∇X Y , Z > + <Y , ∇X Z > = X <Y , Z >
for every X,Y, Z ∈ T1 (M).
10.166 Proposition Let M n be a pseudo-Riemannian manifold with a metric

g, and let (x,U) be a local chart on M . Then the components of the Levi-Civita
connection with respect to the coordinate frame field

∂ ∂
s= 1 , ... , n
∂x ∂x
are given by
n
i
Γ jk = ∑ gil [ jk, l ]
l=1
where
∂ g lk ∂ g l j ∂gjk
[ jk, l ] = 12 + k − l .
∂xj ∂x ∂x
The terms [ jk, l ] and Γ ji k are called Christoffel symbols of the first and second kind,
respectively.
PROOF : Inserting the vector fields X = ∂ j , Y = ∂ k and Z = ∂ l into formula (1)

∂x ∂x ∂x
in Theorem 10.165 and using Corollary 10.111, we obtain
n
i 1 ∂ g lk ∂ gl j ∂ g j k
[ jk, l ] = ∑ gil Γ jk = 2 j + k − l .
i=1
∂x ∂x ∂x
10.167 Proposition Let π : T M → M be the tangent bundle of a pseudo-

Riemannian manifold M n with a metric g, and let L : T M → R be the hyperregular
Lagrangian defined by
L(v) = 21 g(u)(v, v)
for u ∈ M and v ∈ Tu M as described in Proposition 7.101. Then a smooth curve
γ : I → M is a geodesic on M if and only if the curve γ ′ : I → T M is an integral curve
for the Lagrangian vector field XE of L.
PROOF : Let (x,U) be a local chart on M, and let (y,V ) be the corresponding local
chart on T M defined in Proposition 7.81. We denote the two component maps of
y : V → Rn × Rn by q and q̇ , so that q = x ◦ π . Since
n
1
L|V = 2 ∑ (gi j ◦ π ) q̇ i q̇ j ,
i, j=1
the Lagrange’s equations for the curve α = γ ′ take the form
(x i ◦ γ ) ′ (t) = (q̇ i ◦ α )(t)
and
n n ′ n
∂ g jk
∑ (gi j ◦ γ )(q̇ j ◦ α ) (t) = 12 ∑ j k
i (γ (t)) (q̇ ◦ α )(t) (q̇ ◦ α )(t)
j=1 j,k=1
∂ x
for t ∈ I ∩ γ −1 (U) and i = 1, ..., n. By Definition 2.69 the first equation is obviously
true. The second equation can also be written in the form
n
∑ gi j (γ (t))(x j ◦ γ ) ′′ (t)
j=1
n
∂ gi j ∂gjk
= −∑ (γ (t)) − 12 (γ (t)) (x j ◦ γ ) ′ (t) (x k ◦ γ ) ′ (t)
j,k=1
∂ xk ∂ xi
n
∂ g ik ∂ gi j ∂gjk
= − ∑ 12 (γ (t)) + k (γ (t)) − (γ (t)) (x j ◦ γ ) ′ (t) (x k ◦ γ ) ′ (t) ,
j,k=1
∂xj ∂x ∂ xi
which is equivalent to the geodesic equation

n
(xi ◦ γ ) ′′ (t) + ∑ i
Γ jk (γ (t)) (x j ◦ γ ) ′ (t) (xk ◦ γ ) ′ (t) = 0
j,k=1
for t ∈ I ∩ γ −1 (U) and i = 1, ... , n.
10.168 Proposition Let πi : Pi → Mi be a principle Gi -bundle for i = 1, 2, and let

FIBRE BUNDLES 433
( f˜, f , ψ ) be a homomorphism from π1 to π2 where f : M1 → M2 is a diffeomorphism.

If H1 is a connection in P1 , then there is a unique connection H2 in P2 so that
f˜∗ (H 1, u ) = H 2, f˜(u)
for every u ∈ P1 . If ωi and Ωi are the connection form and the curvature form of Hi
on Pi , respectively, for i = 1, 2, then
f˜ ∗ ω 2 = ψ ∗ · ω 1 and f˜ ∗ Ω 2 = ψ ∗ · Ω 1 .
PROOF : We first prove the uniqueness of the connection H2 . Let v ∈ P2 , and choose
a u ∈ π1−1 ( f −1 ◦ π2 (v)) . Since π2 ( f˜(u)) = f (π1 (u)) = π2 (v) , there is a g ∈ G2 with
v = f˜(u) g . Hence we have that
H 2, v = (R g ) ∗ H 2, f˜(u) = (R g ) ∗ ◦ f˜∗ (H 1, u )
which completes the proof of the uniqueness of H2 .

To prove existence of the connection H2 , let s1 : V1 → P1 be a section in π1 :
P1 → M1 defined on an open subset V1 of M1 . Then s2 = f˜ ◦ s1 ◦ f −1 is a section in
π2 : P2 → M2 on the open subset V2 = f (V1 ) of M2 . Let ω e 1 = s∗1 (ω 1 ) be the pull-back
e 2 be the g 2 -valued 1-form on V2 given by ω
of ω 1 by s1 , and let ω e 2 = ψ ∗ · ( f −1 ) ∗ ω
e1
so that f˜ ∗ ω
e 2 = ψ∗· ωe 1 by Proposition 5.48. By Proposition 10.56 there is a unique
connection form η 2 on π2 −1 (V2 ) with ω e 2 = s2∗ ( η 2 ) . We want to show that f˜ ∗ η 2 =
ψ ∗ · η 1 , where η 1 is the restriction of ω 1 to π1 −1 (V1 ) .
If ρr : Vr × Gr → πr −1 (Vr ) is the map defined by
ρ r (p, g) = s r (p) g
for p ∈ Vr and g ∈ Gr , and θr is the canonical Maurer-Cartan form on Gr for r = 1, 2,

then we have that
η r ( ρ r (p, g)) ( ρ r,g ∗ (v) + ρ r,p ∗ (w)) = Ad (g−1 ) ω

e r (p)(v) + θ r (g)(w)
for p ∈ V r , g ∈ G r , v ∈ Tp M r and w ∈ Tg G r . Furthermore, we have that
f˜ ◦ ρ 1 (p, g) = f˜(s 1 (p) g) = f˜ ◦ s 1 (p) ψ (g) = s 2 ◦ f (p) ψ (g) = ρ 2 ( f (p), ψ (g))
so that
f˜ ◦ ρ 1, g = ρ 2, ψ (g) ◦ f and f˜ ◦ ρ 1, p = ρ 2, f (p) ◦ ψ
for p ∈ V 1 and g ∈ G 1 , and we also have that ψ ∗ θ 2 = ψ ∗ · θ 1 since
ψ ∗ θ 2 (g)(w) = θ 2 (ψ (g))(ψ ∗ (w)) = (Lψ (g)−1 ) ∗ ◦ ψ ∗ (w)

= (Lψ (g)−1 ◦ ψ ) ∗ (w) = (ψ ◦ Lg−1 ) ∗ (w) = ψ ∗ θ 1 (g)(w)
for g ∈ G 1 and w ∈ Tg G 1 . Using this and Proposition 9.14, it follows that

f˜ ∗ η 2 ( ρ 1(p, g)) ( ρ 1, g ∗ (v) + ρ 1, p ∗ (w))

= η 2 ( f˜ ◦ ρ 1 (p, g)) ( f˜∗ ◦ ρ 1, g ∗ (v) + f˜∗ ◦ ρ 1, p ∗ (w))
= η 2 ( ρ 2 ( f (p), ψ (g))) ( ρ 2, ψ (g) ∗ ◦ f ∗ (v) + ρ 2, f (p) ∗ ◦ ψ ∗ (w))
= Ad (ψ (g)−1 ) ω
e 2 ( f (p))( f ∗ (v)) + θ 2 (ψ (g))(ψ ∗ (w))
= Ad (ψ (g)−1 ) f ∗ ω
e 2 (p)(v) + ψ ∗ θ 2 (g)(w)
= Ad (ψ (g)−1 ) ψ ∗ ω
e 1 (p)(v) + ψ ∗ θ 1 (g)(w)
= ψ ∗ {Ad (g−1 ) ω
e 1 (p)(v) + θ 1 (g)(w)}
= ψ ∗ η 1 ( ρ 1(p, g)) ( ρ 1, g ∗ (v) + ρ 1, p ∗ (w))
for p ∈ V 1 , g ∈ G 1 , v ∈ Tp M 1 and w ∈ Tg G 1 , which completes the proof that

f˜ ∗ η 2 = ψ ∗ · η 1 .
Now let H e r be the connection on πr −1 (Vr ) with connection form η r for r = 1, 2 .
Then we have that f˜∗ ( H e 1, u ) ⊂ He ˜ for u ∈ π1 −1 (V1 ) . If ir : H er → T Pr is the
2, f (u)
inclusion map and αr = πr ∗ ◦ ir , then it follows from Proposition 10.49 that f˜∗ u
induces a linear isomorphism α −1 e 1, u and H
◦ f ∗ π1 (u) ◦ α 1, u between H e ˜ so
2, f˜(u) 2, f (u)
e 1, u ) = H
that in fact f˜∗ ( H e ˜ for each u ∈ π1 −1 (V1 ) .
2, f (u)
Suppose that H e 2′ is the connection on π2 −1 (V2′ ) obtained in the same way as
above from another section s1′ : V1′ → P1 in π1 : P1 → M1 where V2′ = f (V1′ ) . Then
it follows from the uniqueness part of the proposition that H e 2 and He 2′ coincide on
−1 ′
π2 (V2 ∩ V2 ) . Hence there is a connection H2 in P2 with a connection form ω2
satisfying the properties in the proposition.
Using Propositions 5.62 and 5.63, we also have that f˜ ∗ (d ω 2 ) = ψ ∗ · d ω 1 .
Hence it follows by the second structure equation and the fact that ψ ∗ is a Lie al-
gebra homomorphism that
f˜ ∗ Ω 2 (u)(w1 , w2 ) = Ω 2 ( f˜(u))( f˜∗ (w1 ), f˜∗ (w2 ))

= d ω 2 ( f˜(u))( f˜∗ (w1 ), f˜∗ (w2 )) + [ ω 2 ( f˜(u))( f˜∗ (w1 )), ω 2 ( f˜(u))( f˜∗ (w2 )) ]
= f˜ ∗ (d ω 2 )(u)(w1 , w2 ) + [ f˜ ∗ ω 2 (u)(w1 ), f˜ ∗ ω 2 (u)(w2 ) ]
= ψ ∗ {d ω 1 (u)(w1 , w2 ) + [ ω 1 (u)(w1 ), ω 1 (u)(w2 ) ]} = ψ ∗ Ω 1 (u)(w1 , w2 )
for u ∈ P1 and w1 , w2 ∈ Tu P1 , which shows that f˜ ∗ Ω 2 = ψ ∗ · Ω 1 and completes

the proof of the proposition.
10.169 Definition Let π1 : P1 → M be a reduced subbundle of a principal fibre

bundle π2 : P2 → M. Then a connection H2 in P2 is said to be reducible to a connection
H1 in P1 if H2 is obtained from H1 in the way described in Proposition 10.168.
FIBRE BUNDLES 435
10.170 Proposition Let π : E → M be an n-dimensional vector bundle with

a fibre metric g of index k, and let ε be a signature compatible with g. Then we
have a reduced subbundle π ′′ : Oε (E) → M of the frame bundle π ′ : L (E) → M,
called the orthonormal frame bundle of E of signature ε , with fibre Oε ,p (E) = {u ∈
L p (E)|u∗ g(p) = g εn } for p ∈ M and structure group Oε (Rn ). If an ε is not specified,
we will assume that ε is the canonical signature for g, and the bundle Oε (E) is
denoted simply by O (E) and is called the orthonormal frame bundle of E.
PROOF : Let p 0 ∈ M, and choose an orthonormal basis B = {v1 , ..., vn } for E p 0
with signature ε . If (t, π −1 (U)) is a local trivialization in the vector bundle π : E →
M, where U is an open neighbourhood of p 0 on M, then we have a section s′ =
(s′1 , ... , s′n ) of π ′ : L (E) → M on U with s′i (p 0 ) = vi for i = 1, ... , n , given by
s′i (p) = tx,p
−1
◦ tx,p 0 (vi )
for p ∈ U and i = 1, ... , n. Let V bean open connected neighbourhood
of p 0 con-
tained in U so that the i × i -matrix g(p)( sα′ (p), s′β (p)) , where 1 ≤ α , β ≤ i, is
non-singular for each p ∈ V and 1 ≤ i ≤ n. Then the subspace Vi (p) of E p spanned
by {s′1 (p), ... , s′i (p) } is non-degenerate for p ∈ V and 1 ≤ i ≤ n. Using this, we are
going to constuct a section s = (s1 , ... , sn ) of π ′′ : Oε (E) → M on V with si (p 0 ) = vi
for i = 1, ... , n.
By applying the Gram-Schmidt process to {s′1 (p), ... , s′n (p) }, we obtain an or-
thogonal basis {s′′1 (p), ... , s′′n (p) } for E p with s′′i (p 0 ) = vi for i = 1, ... , n , given by
s′′1 (p) = s′1 (p)
and
i−1 g(p) ( s′ (p) ,s′′ (p))
s′′i (p) = s′i (p) − s′′j (p)
i j
∑ ′′ ′′
j=1 g(p) (s j (p) ,s j (p))
for p ∈ V and i = 2, ... , n. Indeed, assuming inductively that {s′′1 (p), ... ,
s′′i−1 (p) } is an orthogonal basis for Vi−1 (p) , we have that s′′j (p) ∈ / Vi−1 (p) ⊥
so that g(p)(s′′j (p), s′′j (p)) 6= 0 for j = 1, ... , i − 1 since Vi−1 (p) is a non-
degenerate subspace of E p , thus showing that s′′i (p) is well defined. Furthermore,
g(p)(s′′i (p), s′′r (p)) = 0 for r = 1, ... , i − 1 which shows that {s′′1 (p), ... , s′′i (p) } is an
orthogonal basis for Vi (p) . Hence we have an orthonormal basis {s1 (p), ... , sn (p) }
for E p with si (p 0 ) = vi for i = 1, ... , n , given by
si (p) = s′′i (p) / ks′′i (p)k
for p ∈ V and i = 1, ... , n.
Since the map fi : V → R given by fi (p) = g(p)(si (p), si (p)) for p ∈ V is smooth
and V is connected, we have that fi (p) = g(p 0 )(vi , vi ) = εi for p ∈ V and i = 1, ... , n.
This implies that
s(p)∗ g(p)(a, b) = g(p)(s(p) a, s(p) b)

!
n n
= g(p) ∑ ai si (p) , ∑ b j s j (p) = g εn (a, b)
i=1 j=1
for every a, b ∈ Rn , which shows that s(p) ∈ Oε ,p (E) for p ∈ V . Hence the lo-
cal trivialization (ρ −1 , π ′ −1 (V )) in the frame bundle π ′ : L (E) → M, where ρ :
V × Gl(Rn ) → π ′ −1 (V ) is the diffeomorphism given by
ρ (p, G) = s(p) ◦ G
for p ∈ V and G ∈ Gl(Rn ), induces a bijection ts : π ′′ −1 (V ) → V × Oε (Rn ). Indeed,

since ρ (p, G) ∗ g(p) = G ∗ g εn for p ∈ V , we have that ρ (p, G) ∈ π ′′ −1 (V ) for p ∈ V
if and only if G ∈ Oε (Rn ). If s̃ = (s̃1 , ... , s̃n ) is another section of π ′′ : Oε (E) → M
of index ε on some open set V e , and if φ : V ∩ V e → Gl(Rn ) is the transition map
between the corresponding local trivializations in L (E), then φ (p) ∈ Oε (Rn ) for p ∈
V ∩ Ve , since s(p) = s̃(p) ◦ φ (p) by Proposition 10.52 which implies that φ (p) ∗ g nε =
φ (p) ∗ s̃(p) ∗ g(p) = s(p) ∗ g(p) = g εn for p ∈ V ∩ Ve . Hence φ induces a smooth map
ψ : V ∩ Ve → Oε (Rn ) so that
ts̃ ◦ ts−1 (p, G) = (p, ψ (p) ◦ G)
for every p ∈ V ∩ V e and G ∈ Oε (Rn ). By Proposition 10.5 there is therefore a

unique topology, smooth structure and fibre bundle structure on Oε (E) such that
π ′′ : Oε (E) → M is a principal fibre bundle over M with structure group Oε (Rn ),
and such that (ts , π ′′ −1 (V )) is a local trivialization for each section s = (s1 , ... , sn )
of π ′′ : Oε (E) → M of index ε on some open set V . We see that the inclusion map
i : Oε (E) → L (E) is an immersion, and hence π ′′ : Oε (E) → M is a reduced sub-
bundle of the frame bundle π ′ : L (E) → M.
10.171 Remark If M n is a pseudo-Riemannian manifold and ε is a signature

compatible with its metric g, then the orthonormal frame bundle Oε (T M) of the
tangent bundle of M is denoted simply by Oε (M) and is called the orthonormal
frame bundle of M of signature ε . It is a reduced subbundle of the frame bundle
L (M) of M . If ε is the canonical signature for g, the bundle Oε (M) is denoted
simply by O (M) and is called the orthonormal frame bundle of M. The fibres of
these orthonormal frame bundles are denoted by Oε ,p (M) and O p (M), respectively,
for p ∈ M .
10.172 Proposition Let π : E → M be an n-dimensional vector bundle with a fibre

metric g of index k, and let ε be a signature compatible with g. Then a connection H
in L (E) is a metrical connection if and only if it is reducible to a connection H ′ in
the orthonormal frame bundle Oε (E) of E of signature ε .
PROOF : Suppose that H is reducible to a connection H ′ in Oε (E), and let γ : I → M
be a smooth curve on M defined on an open interval I. If t0 ,t1 ∈ I and u ∈ Oε ,γ (t0 ) (E),
then the horizontal lifting β : I → Oε (E) of γ in Oε (E) with β (t0 ) = u is also hori-
zontal in L (E). Hence if v0 , v1 ∈ Eγ (t0 ) and ξi = u−1 (vi ) for i = 1, 2, we have that
FIBRE BUNDLES 437
g(γ (t0 )) (v0 , v1 ) = u∗ g(γ (t0 )) (ξ0 , ξ1 ) = g nε (ξ0 , ξ1 )

= β (t1 )∗ g(γ (t1 )) (ξ0 , ξ1 ) = g(γ (t1 )) (τ t01 (v0 ), τ t01 (v1 ))
t t
t
which shows that the parallel transport τ t01 is an isometry.
Now assume conversely that H is a metrical connection, and let v ∈ Hu for a u ∈
Oε ,p (E) . Choose a smooth curve γ : I → M defined on an open interval I containing
t0 with γ (t0 ) = p and γ ′ (t0 ) = π∗ (v), where π : L (E) → M is the projection in the
frame bundle, and let β : I → L (E) be the horizontal lifting of γ with β (t0 ) = u.
Then we have that β ′ (t0 ) = v, and
β (t)∗ g(γ (t)) (ξ0 , ξ1 ) = β (t0 )∗ g(γ (t0 )) (ξ0 , ξ1 ) = g nε (ξ0 , ξ1 )
for ξ0 , ξ1 ∈ Rn , which shows that β (t) ∈ Oε (E) for all t ∈ I. Hence we have that
Hu ⊂ Tu Oε (E) for every u ∈ Oε (E) , showing that H is reducible to a connection in
Oε (E).
10.173 Proposition Let M n be a pseudo-Riemannian manifold with a metric

g of index r, and let s : V → O (M) be a section in the orthonormal frame bundle
π : O (M) → M. Let θe = s∗ θ be the pull-back of the dual form θ by s, and let Θe
n n
be an R -valued 2-form on V . Then there is a unique or (R )-valued 1-form ω
e on V
which satisfies the structure equation
d θe = − ω
e ∧ θe + Θ.
e (1)
PROOF : Let θ e i and Θ e i be the components of θe and Θ e with respect to the standard
basis E = {e1 , ... , en } for Rn , and let ωe ji be the components of ω e with respect to
i n
the basis D = {Fj |1 ≤ i, j ≤ n} for gl (R ) defined in Example 8.10 (b), where
Fji : Rn → Rn is the linear map given by Fji (ek ) = δ j k ei for k = 1, ... , n. By Remark
10.94 we know that θe i for i = 1, ... , n form a dual local basis on V to the orthonormal
frame field s = (X1 , ... , Xn ). Hence there are unique smooth functions aijk , bijk and
cijk so that
e i = ∑ ai θe k
ω j jk (2)
k
for i, j = 1, ..., n, and
d θe i = 12 ∑ bijk θe j∧ θe k and ei= 1

Θ 2 ∑ cijk θe j ∧ θe k
jk jk
for i = 1, ..., n, where
bijk = −bki j and cijk = −cki j . (3)
e is or (Rn )-valued, we have that

Since the form ω
si j aikj = −aijk , (4)

where si j = sgn (i − r − 1/2)( j − r − 1/2). We have that
si j s jk ski = 1 (5)
for every triple of indices i, j and k. Indeed, we can partition the index set {1, ..., n}
in two disjoint classes {1, ..., r} and {r + 1, ..., n}. Then two indices i and j belong to
the same class if and only if si j = 1. Now there are two possibilities. Either all three
indices i, j and k belong to the same class, in which case all the factors in the product
are 1, or two indices can belong to one class and the third to the other class, in which
case two of the factors in the product are −1 and the third is 1. Using the structure
equation, we now obtain
1 e i = ∑ ai θe j∧ θe k + 1
2 ∑ bijk θe j∧ θe k = d θe i = − ∑ e ji ∧ θe j + Θ
ω jk 2 ∑ cijk θe j∧ θe k
jk j jk jk
so that
aijk − aki j = bijk − cijk (6)
which together with (4) implies that
aijk + ski aki j = bijk − cijk . (7)
By interchanging i, j and k cyclically, and multiplying the next equation by si j and

the last by −ski using (5), we obtain
j j j
si j aki + aijk = si j (bki − cki ) (8)
and
−ski aikj − si j akij = −ski (bikj − cikj ). (9)
Adding these three equations and dividing by 2, we have that
aijk = 12 (bijk + si j bkij − ski bikj ) − 12 (cijk + si j ckij − ski cikj ) (10)
which shows the uniqueness of the form ω e.

To prove existence, suppose that ω e is the form whose components are defined by
(2), where the coefficients aijk are the smooth functions defined by (10). Using (3)
and (5) we have that
si j aik = 21 (si j bik + bki j − ski bkji ) − 21 (si j cik + cki j − ski ckji ) = −aijk
j j j
and
aki j = 12 (bki j + ski bkji − si j bikj ) − 12 (cki j + ski ckji − si j cikj ) = aijk − (bijk − cijk )
e is or (Rn )-valued, and

which shows that formula (4) and (6) are satisfied. Hence ω
we have that
FIBRE BUNDLES 439
e ji ∧ θe j = ∑ aijk θe j ∧ θe k =
−∑ ω ∑ (aijk − aki j ) θe j∧ θe k
j jk j<k
= ∑ bijk θe j∧ θe k − ∑ cijk θe j∧ θe k = d θe i − Θe i
j<k j<k
e satisfies the structure Equation (1).

for i = 1, ..., n , which shows that ω
10.174 Proposition Let M n be a pseudo-Riemannian manifold with a metric g

of index r, and let sk : Vk → O (M) for k = 1, 2 be two sections in the orthonormal
frame bundle π : O (M) → M with associated local trivializations (tk , π −1 (Vk )) which
are Or (Rn )-related with transition map φ : V1 ∩ V2 → Or (Rn ). Let θe = ĩ∗ (θ ) be the
pullback of the dual form θ on L (M), where ĩ : O (M) → L (M) is the inclusion
map, and let Θ be an Rn -valued 2-form on O (M) which is tensorial of type (id, Rn ).
Then the unique or (Rn )-valued 1-forms ωek on Vk which satisfy the structure equation
d θek = − ω
ek ∧ θek + Θ
ek (1)
for k = 1, 2, where θek = s k∗ (θe) and e k = s ∗ (Θ),
Θ k are related by the relation
e 1 (p)(v) = φ (p) −1 ◦ ω
ω e 2 (p)(v) ◦ φ (p) + φ (p) −1 ◦ d(i ◦ φ )(p)(v) (2)
for p ∈ V1 ∩V2 and v ∈ Tp M , where i : Or (Rn ) → gl(Rn ) is the inclusion map.
e 1 defined
PROOF : Because of the uniqueness, it is enough to prove that the form ω
by formula (2) satisfies the structure Equation (1) for p ∈ V1 ∩ V2 and v ∈ Tp M ,
e 2.
supposing that this is true for ω
If α : Or (Rn ) → Or (Rn ) is the map given by α (g) = g−1 and ψ = i ◦ α ◦ φ , it
follows by Proposition 10.55 that
θe1 = ψ ∧ θe2 and e1 = ψ ∧ Θ
Θ e2 .
Using Proposition 5.69 we therefore have that

d θe1 = d ψ ∧ θe2 + ψ ∧ d θe2 = d ψ ∧ θe2 − ψ ∧ (ω
e2 ∧ θe2 ) + ψ ∧ Θ
e2 = − ω
e1 ∧ θe1 + Θ
e1
since
d ψ (p)(v1 ) = ψ ∗ p (v1 ) = α ∗ φ (p) ◦ d(i ◦ φ ) (p)(v1 )
= − φ (p) −1 ◦ d(i ◦ φ )(p)(v1 ) ◦ φ (p) −1
by 5.51 and Corollary 8.33, which implies that
d ψ (p)(v1 ) (θe2 (p)(v2 )) − ψ (p) ◦ ω

e2 (p)(v1 ) (θe2 (p)(v2 ))
= − φ (p) −1 ◦ d(i ◦ φ )(p)(v1 ) ◦ φ (p) −1 (θe2 (p)(v2 ))

e2 (p)(v1 ) ◦ φ (p) ◦ φ (p) −1 (θe2 (p)(v2 ))
− φ (p) −1 ◦ ω
e1 (p)(v1 ) (θe1 (p)(v2 ))
=−ω
for p ∈ V1 ∩V2 and v1 , v2 ∈ Tp M .
10.175 Let M n be a pseudo-Riemannian manifold with a metric g, and let R

be the curvature form of the Levi-Civita connection on M. Using the equivalence
φ : T 2 (T M ; Λ 1 (T M ; T M)) → T 31 (M) over M defined in 10.108 and lowering the
contravariant index, we obtain a covariant tensor field R = L 11 (φ ◦ R) ∈ T 4 (M) on
M of degree 4, called the covariant curvature tensor on M, which is given by
R(p)(v1 , v2 , v3 , v4 ) = φ ◦ R(p)(v2 , v3 , v4 , g(p) ♭ (v1 )) = < R(p)(v3 , v4 ) v2 , v1 >
for p ∈ M and v1 , v2 , v3 , v4 ∈ Tp M . If R ji kl are the components of the curvature form

R with respect to a frame field s = (X1 , ... , Xn ) on an open subset V of M, and if g i j
and Ri j kl are the components of the local representations of g and R on V , then
n
Ri j kl = ∑ g ir R jrkl
r=1
for 1 ≤ i, j, k, l ≤ n.
10.176 Proposition Let T ∈ T 4 (V ) be a covariant tensor of degree 4 on a vector

space V satisfying the relations
(1) T (v1 , v2 , v3 , v4 ) = −T (v2 , v1 , v3 , v4 ) ,
(2) T (v1 , v2 , v3 , v4 ) = −T (v1 , v2 , v4 , v3 ) and
(3) T (v1 , v2 , v3 , v4 ) + T (v1 , v3 , v4 , v2 ) + T (v1 , v4 , v2 , v3 ) = 0
for every v1 , v2 , v3 , v4 ∈ V . Then T also satisfies the relation
(4) T (v1 , v2 , v3 , v4 ) = T (v3 , v4 , v1 , v2 )
for v1 , v2 , v3 , v4 ∈ V .
A tensor T with these properties is called curvaturelike.
PROOF : Using the first three relations we have that
2 T (v1 , v2 , v3 , v4 ) = T (v1 , v2 , v3 , v4 ) − T (v2 , v1 , v3 , v4 )
= − {T (v1 , v3 , v4 , v2 ) + T (v1 , v4 , v2 , v3 ) }
+ {T (v2 , v3 , v4 , v1 ) + T (v2 , v4 , v1 , v3 ) }
= − {T (v3 , v1 , v2 , v4 ) + T (v3 , v2 , v4 , v1 ) }
+ {T (v4 , v1 , v2 , v3 ) + T (v4 , v2 , v3 , v1 ) }
= T (v3 , v4 , v1 , v2 ) − T (v4 , v3 , v1 , v2 ) = 2 T (v3 , v4 , v1 , v2 )
for v1 , v2 , v3 , v4 ∈ V .
FIBRE BUNDLES 441
10.177 Proposition Let T ∈ T 4 (V ) be a curvaturelike tensor on a vector space

V satisfying the relation
T (v1 , v2 , v1 , v2 ) = 0
for every v1 , v2 ∈ V . Then we have that T = 0.
PROOF : Combining the above relation for T with relation (4) in Proposition 10.176,
we have that
0 = T (v1 + v3 , v2 , v1 + v3 , v2 )
= 0 + T (v1 , v2 , v3 , v2 ) + T (v3 , v2 , v1 , v2 ) + 0
= 2 T (v1 , v2 , v3 , v2 )
which shows that

T (v1 , v2 , v3 , v2 ) = 0
for every v1 , v2 , v3 ∈ V . Now combining this with relation (2) in Proposition 10.176,
we see that
0 = T (v1 , v2 + v4 , v3 , v2 + v4 )
= 0 + T (v1 , v2 , v3 , v4 ) + T (v1 , v4 , v3 , v2 ) + 0
= T (v1 , v2 , v3 , v4 ) − T (v1 , v4 , v2 , v3 )
which implies that

T (v1 , v2 , v3 , v4 ) = T (v1 , v4 , v2 , v3 )
for v1 , v2 , v3 , v4 ∈ V . This shows that T (v1 , v2 , v3 , v4 ) is unchanged by a cyclic per-
mutation of the vectors v2 , v3 and v4 . Using relation (3) in Proposition 10.176 we
therefore conclude that
T (v1 , v2 , v3 , v4 ) = 0
for every v1 , v2 , v3 , v4 ∈ V .
10.178 Proposition Let T ∈ T 4 (V ) be a curvaturelike tensor on a vector space

V , and let {v1 , v2 } and {w1 , w2 } are two pairs of vectors in V with w j = ∑2i=1 ai j vi
for j = 1, 2. Then we have that
T (w1 , w2 , w1 , w2 ) = det(ai j ) 2 T (v1 , v2 , v1 , v2 ) .
PROOF : Follows from relation (1) and (2) in Proposition 10.176.
10.179 Example Let V be a finite dimensional vector space with a metric g. Then
the tensor Tg ∈ T 4 (V ) defined by
Tg (v1 , v2 , v3 , v4 ) = g(v1 , v3 ) g(v2 , v4 ) − g(v1 , v4 ) g(v2 , v3 )
for v1 , v2 , v3 , v4 ∈ V is curvaturelike. We see immediately that it satisfies the first two

relations in Proposition 10.176, and relation (3) follows since
Tg (v1 , v2 , v3 , v4 ) + Tg(v1 , v3 , v4 , v2 ) + Tg(v1 , v4 , v2 , v3 )
= g(v1 , v3 ) g(v2 , v4 ) − g(v1 , v4 ) g(v2 , v3 )

+ g(v1 , v4 ) g(v3 , v2 ) − g(v1 , v2 ) g(v3 , v4 )
+ g(v1 , v2 ) g(v4 , v3 ) − g(v1 , v3 ) g(v4 , v2 ) = 0
for every v1 , v2 , v3 , v4 ∈ V . We let Qg : V × V → R be the function defined by
Qg (v1 , v2 ) = Tg (v1 , v2 , v1 , v2 ) = g(v1 , v1 ) g(v2 , v2 ) − g(v1 , v2 )2
for v1 , v2 ∈ V . From 7.2 it follows that a 2-dimensional subspace W of V with a basis

{v1 , v2 } is non-degenerate if and only if Qg (v1 , v2 ) 6= 0.
10.180 Proposition Let M be a pseudo-Riemannian manifold with a metric g, and

let R be the covariant curvature tensor on M. Then R(p) is a curvaturelike tensor on
Tp M for every p ∈ M .
PROOF : We must show that R(p) satisfies relation (1) – (3) in Proposition 10.176
for every p ∈ M .
(1) Let ε be a signature compatible with g. By Proposition 10.172 we know
that the Levi-Civita connection H is reducible to a connection H ′ in the orthonor-
mal frame bundle π ′ : Oε (M) → M. If Ω and Ω ′ are the curvature forms of H
and H ′ on L (M) and Oε (M), respectively, it follows from Proposition 10.168 that
i ∗ Ω = ψ ∗ · Ω ′ , where i : Oε (M) → L (M) and ψ : Oε (Rn ) → Gl(Rn ) are the inclu-
sion maps. Choosing u ∈ Oε ,p (M) and w3 , w4 ∈ Tu Oε (M) so that π∗′ (wr ) = vr for
r = 3, 4, we now have that
< R(p)(v3 , v4 ) v2 , v1 > = g(p)(u ◦ i ∗ Ω(u)(w3 , w4 ) ◦ u−1 (v2 ), v1 )

= g εn ( Ω ′ (u)(w3 , w4 ) ◦ u−1 (v2 ), u−1 (v1 ))
= −g nε ( Ω ′ (u)(w3 , w4 ) ◦ u−1 (v1 ), u−1 (v2 ))
= −g(p)(u ◦ i ∗ Ω(u)(w3 , w4 ) ◦ u−1 (v1 ), v2 ) = −< R(p)(v3 , v4 ) v1 , v2 >
by Remark 10.104 and Example 8.58 (b).

(2) Follows since the curvature form R on M satisfies the relation
R(p)(v3 , v4 ) = −R(p)(v4 , v3 )
for p ∈ M and v3 , v4 ∈ Tp M .
(3) Follows from Corollary 10.146 (1).
10.181 Definition Let M be a pseudo-Riemannian manifold with a metric g,

FIBRE BUNDLES 443
and let R be the covariant curvature tensor on M. A 2-dimensional subspace of the

tangent space Tp M at a point p ∈ M is called a tangent plane to M at p . For each
non-degenerate tangent plane W at p , the number
R(p)(v1 , v2 , v1 , v2 )
K(W ) = ,
Qg(p) (v1 , v2 )
where {v1 , v2 } is a basis for W and Qg(p) : Tp M × Tp M → R is the function defined

in Example 10.179, is called the sectional curvature of W . If dim p (M) = 2 , then
K(Tp M) is called the Gaussian curvature at p.
10.182 Remark It follows from Proposition 10.178 and Example 10.179 that
the sectional curature K(W ) is well defined, being independent of the choice of basis
{v1 , v2 } for W , and with Qg(p) (v1 , v2 ) 6= 0 when W is non-degenerate. In the next
proposition we will show that the covariant curvature tensor R is completely deter-
mined by the sectional curvature K on M. We first need a lemma.
10.183 Lemma Let V be a finite dimensional vector space with a metric g. Then
the set
B = {(v1 , v2 ) ∈ V × V |Qg (v1 , v2 ) 6= 0}
is dense in V × V .
PROOF : Consider two vectors u1 , u2 ∈ V , and let O1 and O2 be open neighbourhoods
around u1 and u2 , respectively. We must show that (O1 × O2 ) ∩ B 6= 0. /
We first show that there are vectors v1 ∈ O1 and v2 ∈ O2 which are linearly inde-
pendent. Let {e1 , e2 } be a basis for a 2-dimensional subspace W of V containing u1
and u2 , and suppose that u j = ∑2i=1 ai j ei for j = 1, 2. Then there is a real number t so
that vi = ui + tei ∈ Oi for i = 1, 2, and

a11 + t a12 a11 a12
= + (a11 + a22 ) t + t 2 6= 0
a21 a22 + t a21 a22
which implies that v1 and v2 are linearly independent.

If Qg (v1 , v2 ) 6= 0, then (v1 , v2 ) ∈ (O1 × O2 ) ∩ B . On the other hand, if
Qg (v1 , v2 ) = 0, the subspace W is degenerate, so the metric g must be indefinite. We
contend that there exists a vector v3 ∈ V with Qg (v1 , v3 ) < 0. Indeed, if g(v1 , v1 ) = 0
we choose a v3 ∈ V with g(v1 , v3 ) 6= 0, and if g(v1 , v1 ) 6= 0 we choose a v3 ∈ V so
that g(v3 , v3 ) and g(v1 , v1 ) have opposite sign. Now there is a real number t so that
v2 + t v3 ∈ O2 and
Qg (v1 , v2 + t v3 ) = 2 t Tg (v1 , v2 , v1 , v3 ) + t 2 Qg (v1 , v3 ) 6= 0 ,
which implies that (v1 , v2 + t v3 ) ∈ (O1 × O2 ) ∩ B.

10.184 Proposition Let M be a pseudo-Riemannian manifold with a metric g,

and let R and K be the covariant curvature tensor and the sectional curvature on M.
Let T ∈ T 4 (Tp M) be a curvaturelike tensor in the tangent space Tp M, and suppose
that
T (v1 , v2 , v1 , v2 )
K(W ) =
Qg(p) (v1 , v2 )
for each non-degenerate tangent plane W at p , where {v1 , v2 } is a basis for W . Then
we have that T = R(p).
PROOF : Follows from Proposition 10.177 and Lemma 10.183 applied to the curva-
turelike tensor T − R(p).
10.185 Let M n be a pseudo-Riemannian manifold with a metric g, and let Ric

be the Ricci curvature tensor of the Levi-Civita connection on M. If p ∈ M and B =
{w1 , ..., wn } is an orthonormal basis for Tp M with signature ε , we have that
n n
Ric(p)(v1 , v2 ) = ∑ εi < R(p)(wi , v2 ) v1 , wi > = ∑ εi R(p)(wi , v1 , wi , v2 )
i=1 i=1
for v1 , v2 ∈ Tp M. Using Proposition 10.180 (4) we see that

Ric(p)(v1 , v2 ) = Ric(p)(v2 , v1 )
for p ∈ M and v1 , v2 ∈ Tp M, showing that the Ricci tensor of the Levi-Civita connec-
tion is symmetric.
Taking the contraction of Ric after raising the first index, we obtain a smooth func-
tion S = C11 (R 11 (Ric)) ∈ F (M) on M, called the scalar curvature on M. If R ji kl are
the components of the curvature form R with respect to a frame field s = (X1 , ... , Xn )
on an open subset V of M, and if g i j and Ric i j are the components of the local
representations of g and Ric on V , then
S = ∑ g i j Ric i j = ∑ g i j R ikk j .
i, j i, j, k
10.186 Proposition Let M n be a pseudo-Riemannian manifold with a metric g,

and let R be the curvature form of the Levi-Civita connection H on M. If R ji kl are the
components of R with respect to a frame field s = (X1 , ... , Xn ) on an open subset V
of M, then Bianchi’s 2nd identity can be written in the form
R ji kl ; m + R ji l m ; k + R ji mk ; l = 0
for 1 ≤ i, j, k, l, m ≤ n.
PROOF : Let (α 1 , ... , α n ) be the dual local basis of the frame field s. Using the equiv-
alence φ : T 2 (T M ; Λ 1 (T M ; T M)) → T 31 (M) over M defined in 10.108, we have
that
φ ◦ R = ∑ R ji kl α j ⊗ α k ⊗ α l ⊗ Xi ,
i, j,k,l
FIBRE BUNDLES 445
so that
∇Xm (φ ◦ R) = ∑ R ji kl ; m α j ⊗ α k ⊗ α l ⊗ Xi
i, j,k,l
for 1 ≤ m ≤ n by Proposition 10.119. To see that
∇Xm (φ ◦ R) = φ ◦ ∇Xm R ,
we consider the frame bundle π : L (M) → M over M and the vector-valued functions
φ ◦ R and R on L (M). For each p ∈ M and u ∈ L p (M) we have a commutative
diagram
ψ✲
T 2 (Rn ; gl (Rn )) T 31 (Rn )
bu
µ eu
µ
❄ ❄
φ p✲
T 2 (Tp M ; Λ 1 (Tp M ; Tp M)) T 31 (Tp M)
where ψ : T 2 (Rn ; gl (Rn )) → T 31 (Rn ) is the linear isomorphism defined by
ψ (G)(a1 , a2 , a3 , λ ) = λ ◦ G(a2 , a3 )(a1 )
for G ∈ T 2 (Rn ; gl (Rn )) , a1 , a2 , a3 ∈ Rn and λ ∈ (Rn ) ∗ , and µ eu are the linear

bu and µ
isomorphisms defined in Example 10.29 (i) and (f). By Definition 10.77 it follows
that
φ ◦ R (u) = µ e u−1 ◦ φ ◦ R (π (u)) = ψ ◦ µ b u−1 ◦ R (π (u)) = ψ ◦ R (u)
for u ∈ L (M), so that
eu ◦ d φ ◦ R (u)(w) = µ
∇v (φ ◦ R) = µ eu ◦ ψ ◦ d R (u)(w)
bu ◦ d R (u)(w) = φ (∇v R)
=φ ◦µ
for p ∈ M and v ∈ Tp M, where u ∈ L (M) and w ∈ Hu are chosen so that π (u) = p

and π∗ (w) = v. The proposition now follows from Corollary 10.146 (2).
10.187 Remark Contracting Bianchi’s 2nd identity with respect to the indices i
and m leads to
∑ g m r R r j kl ; m − Ric j l ; k + Ric j k ; l = ∑ R jmkl ; m + ∑ R jml m ; k + ∑ R jmmk ; l = 0

m,r m m m
for 1 ≤ j, k, l ≤ n. Raising the index j and contracting with l then gives
∑ Ric mk; m − S ; k + ∑ Ric lk ; l = −∑ g m r R rl kl ; m − S ; k + ∑ Ric lk ; l = 0

m l m,r,l l
which implies that

∑ ( Ric lk − 12 S δ kl ) ; l = 0
l
for 1 ≤ k ≤ n. Lowering the index l gives the covariant form
∑ ( Ric k l − 12 S g k l ) ; l = 0
l
for 1 ≤ k ≤ n. The tensor G = Ric − 12 S g is called the Einstein curvature tensor.

Using this we can now write down the fundamental equation of general relativity,
the Einstein field equation
G + Λg = κ T ,
where Λ is the cosmological constant and T is the energy-momentum tensor. The
fundamental constant κ , called the Einstein constant, is given by
8πγ
κ= ,
c4
where γ = 6.673 · 10−11 N m2/ kg2 is Newton’s gravitational constant and c = 2.998 ·
10 8 m/ s is the speed of light in vacuum.
In both special and general relativity one often replaces the SI units with natural
units where c = 1, so that time is measured in meters where 1s = 2.998 · 10 8 m. In
general relativity one also uses geometrized units where in addition γ = 1, so that
κ = 8π . Since γ/c2 = 7.425 · 10−28 m/ kg , this means that we are also measuring
mass in meters where 1 kg = 7.425 · 10−28 m.
The fundamental idea of general relativity is that, contrary to Newton’s theory
of gravity, there are no gravitational forces. Instead the energy-momentum tensor
generates a curvature in spacetime, where free particles move along geodesics.
The spacetime of events is a 4-dimensional Lorentz manifold, and the path of
a particle through spacetime is called its world line . The world lines of material
particles are timelike, while photons follow null world lines. We imagine that each
material particle is equipped with a clock measuring its proper time , which in natural
units is the arc length of its world line from some specified starting point. The world
line is usually parametrized by proper time.
The energy-momentum tensor for a perfect fluid is given by
T = (ρ + p) u♭⊗ u♭ + p g ,
where ρ and p are the energy density and pressure measured by an observer co-
moving with the fluid. They are functions on the spacetime manifold M, and u is a
timelike unit vector field on M called the flow vector field of the fluid.
FIBRE BUNDLES 447
10.188 Example By Proposition 2.41 we know that the sphere

S2a = {(x, y, z) ∈ R3 |x2 + y2 + z2 = a2 }
with radius a > 0 is a closed submanifold of the Euclidean space R3 . Let ia : S2a → R3
be the inclusion map and p : R × R2 → R2 be the projection on the last factor. If
(ψ ,V ) is the spherical coordinate system on R3 defined in Example 7.21 with coor-
dinate functions r, θ and φ , and if Va = V ∩ S2a and ψa = p ◦ ψ ◦ ia , then (ψa ,Va ) is
a local chart on S2a with coordinate functions θ ◦ ia and φ ◦ ia which we also denote
simply by θ and φ . Let ga = i∗a g be the pull-back to S2a of the standard metric g on
R3 . Then we have that
ga |Va = a2 d θ 2 + a2 sin2 θ d φ 2 .
We see that
X1 = 1 ∂ and X2 = 1 ∂
a ∂θ a sin θ ∂ φ
form an orthonormal frame field on Va with dual local basis
θe 1 = a d θ and θe 2 = a sin θ d φ .
Since
d θe 1 = 0
d θe 2 = a cos θ d θ ∧ d φ = a1 cot θ θe 1∧ θe 2 ,
the non-zero coefficients bijk and aijk are

2
b12 2
= − b21 = 1a cot θ
and
1
a22 2
= − a12 = 12 (b22
1 2
+ b21 2
− b12 ) = − 1a cot θ .
This shows that
e 12 = − 1a cot θ θe 2 = − cos θ d φ
e 21 = − ω
ω
so that
e 12 = − Ω
Ω e 12 = d ω
e 21 + ω
e 11 ∧ ω
e 21 + ω e 22 = sin θ d θ ∧ d φ = 12 θe 1∧ θe 2 .
e 21 ∧ ω
a
We have a diffeomorphism σa : R3 − {0} → (0, ∞) × Sa2 defined by
σa (v) = ( k v k , av/ k v k )
for v ∈ R − {0}, with inverse ρa : (0, ∞) × Sa2 → R3 − {0} given by
3
ρa (r, u) = (r/a) u
for (r, u) ∈ (0, ∞) × Sa2 , so that
σa (V ) = (0, ∞) × Va and p ◦ ψ = ψa ◦ σa,2 .
THE SCHWARZSCHILD – DE SITTER SPACETIME
10.189 We will now find the vacuum solution of Einstein’s field equation with
cosmological constant Λ for a spacetime which is static and spherically symmetric.
Let M be a Lorentz manifold with a metric g so that
g|N = − e2α dt 2 + e2β dr 2 + r 2 ( d θ 2 + sin2 θ d φ 2 )
in natural units for a local chart (ψ , N) on M with coordinate functions t, r, θ and

φ , where α = α̃ ◦ r and β = β̃ ◦ r for real functions α̃ and β̃ . The partial derivatives
∂α ∂β ′ ′ ′ ′
∂ r and ∂ r are denoted by α and β , respectively, and are given by α = α̃ ◦ r and
β ′ = β̃ ′ ◦ r. We see that
X1 = e−α ∂ , X2 = e−β ∂ , X3 = 1 ∂ and X4 = 1 ∂

∂t ∂r r ∂θ r sin θ ∂ φ
form an orthonormal frame field on N with dual local basis
θe 1 = eα dt , θe 2 = eβ dr , θe 3 = r d θ and θe 4 = r sin θ d φ .
Since
d θe 1 = − α ′ e−β θe 1∧ θe 2
d θe 2 = 0
d θe 3 = 1r e−β θe 2∧ θe 3
d θe 4 = 1r e−β θe 2∧ θe 4 + 1r cot θ θe 3∧ θe 4 ,
b112 = − b121 = − α ′ e−β
bi2i = − bii2 = 1r e−β for i = 3, 4
b434 = − b443 = 1r cot θ
and
a121 = a211 = 12 (b121 − b211 − b112) = α ′ e−β
a2ii = − ai2i = 12 (b2ii + bii2 − bi2i ) = − 1r e−β for i = 3, 4
a344 = − a434 = 12 (b344 + b443 − b434) = − 1r cot θ

FIBRE BUNDLES 449
e are
This shows that the non-zero components of ω
e 12 = α ′ e−β θe 1 = α ′ eα −β dt
e 21 = ω
ω
e 23 = − 1r e−β θe 3 = − e−β d θ
e 32 = − ω
ω
e 24 = − 1r e−β θe 4 = − e−β sin θ d φ

e 42 = − ω
ω
e 34 = − 1r cot θ θe 4 = − cos θ d φ
e 43 = − ω
ω
so that
e1=Ω
Ω e 2 = dω
e 21 = − eα −β (α ′′ + α ′ 2 − α ′ β ′ ) dt ∧ dr
2 1
= − e−2β (α ′′ + α ′ 2 − α ′ β ′ ) θe 1∧ θe 2
e1=Ω
Ω 3
e3=ω
1 e 32 = − 1r α ′ e−2β θe 1∧ θe 3
e 21 ∧ ω
e 41 = Ω
Ω e 14 = ω e 42 = − 1r α ′ e−2β θe 1∧ θe 4
e 21 ∧ ω
e2=−Ω
Ω 3
e 3 = dω
2 e 32 + ω e 34 = β ′ e−β d r ∧ d θ = 1r β ′ e−2β θe 2∧ θe 3
e 42 ∧ ω
e 42 = − Ω
Ω e 42 = d ω
e 42 + ω e 43 = β ′ e−β sin θ d r ∧ d φ
e 32 ∧ ω
− e−β cos θ d θ ∧ d φ + e−β cos θ d θ ∧ d φ = 1r β ′ e−2β θe 2∧ θe 4
e3=−Ω
Ω e 43 = d ω
e 43 + ω e 42 = sin θ d θ ∧ d φ − e−2β sin θ d θ ∧ d φ
e 23 ∧ ω
4
= 12 ( 1 − e−2β ) θe 3∧ θe 4
r
By Corollary 10.106 the non-zero components of the curvature form R are
R2112 = −R2121 = R1212 = −R1221 = − e−2β (α ′′ + α ′ 2 − α ′ β ′ )
Ri11i = −Ri1i1 = R1i 1i = −R1i i1 = − 1r α ′ e−2β for i = 3, 4
Ri22i = −Ri2i2 = −R2i 2i = R2i i2 = 1r β ′ e−2β for i = 3, 4
R4334 = −R4343 = −R3434 = R3443 = 12 ( 1 − e−2β )

r
Using the components of R with equal contravariant and second covariant index, we
find the non-zero components of the Ricci tensor
Ric 11 = R1221 + R1331 + R1441 = e−2β (α ′′ + α ′ 2 − α ′ β ′ ) + 2r α ′ e−2β
Ric 22 = R2112 + R2332 + R2442 = − e−2β (α ′′ + α ′ 2 − α ′ β ′ ) + 2r β ′ e−2β
Ric 33 = R3113 + R3223 + R3443 = 1r (β ′ − α ′ ) e−2β + 12 ( 1 − e−2β )

r
Ric 44 = R4114 + R4224 + R4334 = 1r (β ′ − α ′ ) e−2β + 12 ( 1 − e−2β )
r
and the scalar curvature
S = −Ric 11 + Ric 22 + Ric 33 + Ric 44
= − 2 e−2β (α ′′ + α ′ 2 − α ′ β ′ ) + 4r (β ′ − α ′ ) e−2β + 22 ( 1 − e−2β )

r
From these we obtain the non-zero components of the Einstein tensor
G 11 = Ric 11 + 12 S = 2r β ′ e−2β + 12 ( 1 − e−2β ) = 12 {r( 1 − e−2β )} ′

r r
1 2 ′ −2β 1 −2β
G 22 = Ric 22 − 2 S = r α e − 2 (1−e )
r
= 2r (α + β ) ′ e−2β − 12 {r( 1 − e−2β )} ′
r
G ii = Ric ii − 12 S = e−2β (α ′′ + α ′ 2 − α ′ β ′ ) − 1r (β ′ − α ′ ) e−2β
n o n o
= e−2β (α + β )′′ + (α + β )′ α ′ − 2β ′ + 1r − e−2β (β ′′ − 2β ′ 2 ) + 2r β ′ e−2β
n o
= e−2β (α + β )′′ + (α + β )′ α ′ − 2β ′ + 1r − 1 {r(1 − e−2β )}′′ for i = 3, 4
2r
We see that the vacuum field equation G + Λg = 0 is equivalent to the conditions
α + β = C and r( 1 − e−2β ) = Rs + Λ r3
3
for constants C and Rs . Changing the constant C simply corresponds to a rescaling of
the time coordinate t. Hence we can choose C = 0 so that
R
e2α = e−2β = 1 − rs − Λ r2
3
and
−1
R R
g|N = − 1 − rs − Λ r2 dt 2 + 1 − rs − Λ r2 dr 2 + r 2 (d θ 2 + sin2 θ d φ 2 ), (1)
3 3
where g is called the Schwarzschild–de Sitter metric. The value of the constant Rs ,
called the Schwarzschild radius , will be determined in Section 10.190. If Λ = 0, the
metric g is called the Schwarzschild metric, and t, r, θ and φ are called Schwarzschild
coordinates . If Rs = 0, the metric g is called the de Sitter metric when Λ > 0 and the
anti de Sitter metric when Λ < 0, and t, r, θ and φ are called Eddington coordinates .
The coordinate expression

R R −1 2
g|N = − 1 − rs dt 2 + 1 − rs dr + r 2 ( d θ 2 + sin2 θ d φ 2 ), (2)
of the Schwarzschild metric is not defined for r = 0 and r = Rs . Hence the

Schwarzschild spacetime M has two connected components M 1 = R ×(Rs , ∞) × S2
and M 2 = R ×(0, Rs ) × S2 . The metric on M is given by

R R −1 2
g = − 1 − rs dt 2 + 1 − rs dr + r 2g1 ,
FIBRE BUNDLES 451
where g1 is the metric on the 2-sphere S2 of radius 1 described in Example 10.188.

The coordinate neighbourhood N has the connected components N 1 = R ×(Rs , ∞)
×V1 and N 2 = R ×(0, Rs) × V1 , where (ψ1 ,V1 ) is the local chart on S2 with coordi-
nate functions θ and φ .
The value r = 0 corresponds to a physical singularity in the Schwarzschild space-
time where the curvature tensors diverge. On the other hand, at r = Rs the curvature
tensors are finite and well behaved. This value of r is therefore only a coordinate
singularity which can be removed by choosing new coordinates. Physically it corre-
sponds to a horizon in the spacetime, separating the components M1 and M2 which
are called the exterior Schwarzschild spacetime and the black hole, as will be ex-
plained in Section 10.191.
10.190 Timelike geodesics in the Schwarzschild spacetime The Lagrangian

of a test particle of unit mass moving in the coordinate neighbourhood N in the
Schwarzschild spacetime M is given by
n o
R R −1 2
L(v) = 12
2 2
− 1 − rs t˙ 2 + 1 − rs ṙ + r 2 θ̇ + r 2 sin2 θ φ̇ (1)
for tangent vectors v ∈ T N with coordinates (t, r, θ , φ , t˙, ṙ, θ̇ , φ˙) as described in
10.189 and Proposition 7.81. Since t and φ are cyclic coordinates, their conjugete
momenta

R
p t ◦ F(L) = − 1 − rs t˙ and p φ ◦ F(L) = r 2 sin2 θ φ̇
are constants of the motion. If γ : I → T N is an integral curve for the Lagrangian

vector field XE of L, we denote the constant fuctions p t ◦ F(L) ◦ γ and p φ ◦ F(L) ◦ γ
simply by p t and p φ . As well as being real numbers and coordinate functions of
the local chart (ψ , N), we also let t, r, θ , φ , t˙, ṙ, θ̇ and φ̇ denote the components
of the map (ψ × id) ◦ tψ ◦ γ as functions of the proper time τ . Then it follows from
Lagrange’s equations that
p φ2 cot θ
∂L
= ∂ L ◦ γ = r 2 sin θ cos θ φ̇ =
d 2 d 2
d τ (r θ̇ ) = d τ ◦γ .
∂ θ̇ ∂θ r2 sin2 θ
Multiplying by 2 r 2 θ̇ we therefore have that

d 2 2 2
d τ {(r θ̇ ) + ( p φ cot θ ) } = 0. (2)
By the spherical symmertry there is no loss of generality assuming that θ (τ 0 ) = π /2

and θ̇ (τ 0 ) = 0 at some time τ 0 . Then it follows from (2) that θ (τ ) = π /2 and θ̇ (τ ) =
0 for every τ ∈ I, showing that the particle is moving in the equatorial plane.
As the path of the test particle is parameterized by proper time, it now follows
from (1) that
p 2φ
R −1 2 R −1 2
− 1 − rs p t + 1 − rs ṙ + 2 = −1
r
so that
p 2φ
R
ṙ 2 + 1 − rs 1 + 2 = p 2t .
r
This implies that
1
2 ṙ 2 + V (r) = E (3)
where
Rs p 2φ Rs p 2φ
V (r) = − + 2− and E = 21 ( p 2t − 1) .
2r 2r 2 r3
The first two terms in the formula for V (r) constitute the effective Newtonian poten-
tial for the motion of a test particle of unit mass in a gravitational central field from
a mass M at the origin as described in 9.55, where
Rs = 2 γ M .
The third term is a relativstic correction, and Equation (3) can be solved by numerical
methods. It was a major breakthrough in the general theory of relativity when this
relativistic term was used to predict a perihelion precession of 43 arc seconds per
century for the orbit of the planet Mercury, which was exactly the part of the observed
precession which could not be accounted for by Newton’s theory.
10.191 Eddington–Finkelstein coordinates In this section we will see how the

coordinate singularity r = Rs in the local expression for the Schwarzschild metric
given in (2) in 10.189 can be removed by introducing new coordinates. We first define
a new radial coordinate function

er = r + Rs log r − 1 ,
Rs
which is called a tortoise coordinate since e

r → −∞ when r → Rs . As

R −1
r = 1 − rs
de dr ,
the Schwarzschild metric then takes the form

R
g|N j = 1 − rs (−dt 2 + de r 2 ) + r 2 ( d θ 2 + sin2 θ d φ 2 ) (1)
for j = 1, 2. Let γ : I → N1 be the geodesic for a radially moving light ray in the
exterior Schwarzschild spacetime. Then we have that
r) ◦ γ = const
(t + e
for ingoing light and

r) ◦ γ = const
(t − e
for outgoing light.
r we obtain a local chart (ψ i e , N)
Introducing a new coordinate function v = t + e
FIBRE BUNDLES 453
on M with coordinate functions v, r, θ and φ , called ingoing Eddington–Finkelstein

coordinates, satisfying
r 2 = −dv (dv − 2 de
−dt 2 + de r)
so that
R
g|N = − 1 − rs dv 2 + 2 dv dr + r 2 ( d θ 2 + sin2 θ d φ 2 ). (2)
Since det(g i j )|N = −r4 sin2 θ , this coordinate expression of the Schwarzschild met-
ric is well defined for all r > 0. Hence the coordinate singularity at r = Rs has been
removed, and the Lorentz manifold M and the coordinate neighbourhood N can be
replaced by E = R ×(0, ∞) × S2 and V = R ×(0, ∞) × V1 , with the metric

R
g i e = − 1 − rs dv 2 + 2 dv dr + r 2 g1 .
Now we have that

v ◦ γ = const
r) ◦ γ = const
(v − 2 e
for outgoing light. This shows that the surface r = Rs acts as a one-way membrane,
letting the light rays cross only from the outside to the inside. Since the light cannot
escape from the region r < Rs , it is called a black hole , and the surface r = Rs is
called a horizon .
We can also introduce a coordinate function u = t − e r and obtain a local chart
(ψ o e , N) on M with coordinate functions u, r, θ and φ , called outgoing Eddington–
Finkelstein coordinates, satisfying
r 2 = −du (du + 2 de
−dt 2 + de r)
so that

R
g|N = − 1 − rs du 2 − 2 du dr + r 2 ( d θ 2 + sin2 θ d φ 2 ). (3)
The Lorentz manifold M and the coordinate neighbourhood N can again be replaced
by E and V , but this time with the metric

R
g o e = − 1 − rs du 2 − 2 du dr + r 2 g1 .
Now we have that

r) ◦ γ = const
(u + 2 e
u ◦ γ = const
for outgoing light. Hence the light rays can cross the surface r = Rs only from the
inside to the outside, and the region r < Rs is called a white hole .
The Lorentz manifolds (M 1 , g|M 1 ) , (M 1 , g i e |M 1 ) and (M 1 , g o e |M 1 ) are iso-

metric isomorphic, as are the Lorentz manifolds (M 2 , g|M 2 ) , (M 2 , g i e |M 2 ) and
(M 2 , g o e |M 2 ) . But the Lorentz manifolds (E, g i e ) and (E, g o e ) are not isometric iso-
morphic, in agreement with the fact that a black hole and a white hole have different
physical interpretations. However, in the next section we will see that both (E, g i e )
and (E, g o e ) can be isometrically embedded in a larger Lorentz manifold.
10.192 Kruskal-Szekeres coordinates Using the coordinate functions
v = t +e
r , u = t −e
r
described in 10.191, and fixing functions ε1 : M → {−1, 1} and ε2 : M → {−1, 1}

which are constant on M1 and M2 with

ε1 ε2 = sgn r − 1 ,
Rs
we can define new coordinate functions

ve = ε1 exp v , ue = − ε2 exp − u
2Rs 2Rs
and
T = 12 (e
v + ue) , R = 12 (e
v − ue)
so that

r e
r r r
R2 − T 2 = − ve ue = sgn −1 exp = −1 exp . (1)
Rs Rs Rs Rs
The transformation from the Schwarzschild spacetime M is given by
1/2
T = ε1 r − 1 exp r sinh t
Rs 2Rs 2Rs
1/2
r r t
R = ε1 −1 exp cosh
Rs 2Rs 2Rs
when r > Rs , mapping M1 onto
K2−ε1 = {(T, R) ∈ R2 ||T | < ε1 R } × S2 ,
and by
1/2
T = ε1 1 − r exp r cosh t
Rs 2Rs 2Rs
1/2
R = ε1 1 − r exp r
sinh t
Rs 2Rs 2Rs
when 0 < r < Rs , mapping M2 onto
p
K3−ε1 = {(T, R) ∈ R2 ||R| < ε1 T < R2 + 1 } × S2 .
FIBRE BUNDLES 455
The inverse transformation is given by (1) and by

tanh t =T
2Rs R
when r > Rs , and
t
tanh =R
2Rs T
when 0 < r < Rs . Now using that
−dT 2 + dR2 = − dev d ue = 1 2 (− ve ue) (− dv du)

4Rs

r Rs r
= 3
1 − r exp r 2 ),
(−dt 2 + de
4Rs Rs
we have that

4R 3
g|N = r s exp − r (−dT 2 + dR2 ) + r 2 ( d θ 2 + sin2 θ d φ 2 ) (2)
Rs
where r is given implicitely by (1). As this coordinate expression of the
Schwarzschild metric is well defined for all r > 0, and using all possible alternatives
for ε1 , we obtain a new Lorentz manifold
K = {(T, R) ∈ R2 |T 2 − R2 < 1} × S2
and a local chart (ψ k ,W ) on K with coordinate neighbourhood
W = {(T, R) ∈ R2 |T 2 − R2 < 1} × V1
and coordinate functions T , R, θ and φ , called Kruskal–Szekeres coordinates . The

metric g k on K is

4R 3
g k = r s exp − r (−dT 2 + dR2) + r 2 g1 .
Rs
To find the transformations from the Eddington–Finkelstein extensions, we use
that

ue = − ε2 exp er exp − v = − ε1 r − 1 exp r exp − v
Rs 2Rs Rs Rs 2Rs
and

e
r u r r u
ve = ε1 exp exp = ε2 −1 exp exp ,
Rs 2Rs Rs Rs 2Rs
which imply that
n o
ε1 v r r
T= exp − −1 exp exp − v
2 2Rs Rs Rs 2Rs
n o
ε1 v r r
R= exp + −1 exp exp − v
2 2Rs Rs Rs 2Rs
and
n o
ε2 r r u
T= −1 exp exp − exp − u
2 Rs Rs 2Rs 2Rs
n o
ε2 r r u
R= −1 exp exp + exp − u .
2 Rs Rs 2Rs 2Rs
The inverse transformations are given by (1) and by

ε1 exp v = R + T , ε2 exp − u = R−T .
2Rs 2Rs
Choosing ε1 = 1 and ε2 = 1 in these formulae, we see that the regions
K12 = {((T, R), u) ∈ K | R > −T } and K14 = {((T, R), u) ∈ K | R > T }
are isometric to the ingoing and outgoing Eddington–Finkelstein extension of K1 , re-
spectively. Hence K2 is isometric to the black hole, and K4 is isometric to the white
hole. The regions K1 and K3 are both isometric to the external Schwarzschild space-
time. But these two regions cannot be connected by any timelike or null curve. An
object entering the black hole K2 cannot avoid the singularity at T 2 − R2 = 1 where
r = 0.
10.193 Outgoing and ingoing coordinates in the de Sitter spacetime The

coordinate expression
−1
g|N = − 1 − Λ r2 dt 2 + 1 − Λ r2 dr 2 + r 2 ( d θ 2 + sin2 θ d φ 2 ), (1)
3 3
of the de Sitter metric, obtained from (1) in 10.189 with Rs = 0 and Λ > 0, is not
defined for r = Rd = (3/Λ)1/2 . Hence the de Sitter spacetime M has two connected
components M 1 = R ×(0, Rd ) × S2 and M 2 = R ×(Rd , ∞) × S2 . The metric on M
is given by
−1
g = − 1 − Λ r2 dt 2 + 1 − Λ r2 dr 2 + r 2 g1 ,
3 3
where g1 is the metric on the 2-sphere S2 of radius 1 described in Example 10.188.
The coordinate neighbourhood N has the connected components N 1 = R ×(0, Rd )
×V1 and N 2 = R ×(Rd , ∞) × V1 , where (ψ1 ,V1 ) is the local chart on S2 with coor-
dinate functions θ and φ .
In this section we will see how the coordinate singularity r = Rd in the local
expression for the de Sitter metric given in (1) can be removed by introducing new
coordinates. We first define a new radial coordinate function
R +r
R
r = d log d
e ,
2 Rd − r
r → ∞ when r → Rd . As
which is called a tortoise coordinate since e
−1
r = 1 − Λ r2
de dr ,
3
FIBRE BUNDLES 457
the de Sitter metric then takes the form

g|N j = 1 − Λ r2 (−dt 2 + der 2 ) + r 2 ( d θ 2 + sin2 θ d φ 2 ) (2)
3
for j = 1, 2. Let γ : I → N1 be the geodesic for a radially moving light ray in the
interior de Sitter spacetime. Then we have that
r) ◦ γ = const
(t − e
for outgoing light and

r) ◦ γ = const
(t + e
for ingoing light.
Introducing a new coordinate function v = t − e r we obtain a local chart (ψ o e , N)
on M with coordinate functions v, r, θ and φ , called outgoing coordinates, satisfying
r 2 = −dv (dv + 2 de
−dt 2 + de r)
so that

g|N = − 1 − Λ r2 dv 2 − 2 dv dr + r 2 ( d θ 2 + sin2 θ d φ 2 ). (3)
3
Since det(g i j )|N = −r4 sin2 θ , this coordinate expression of the de Sitter metric is
well defined for all r > 0. Hence the coordinate singularity at r = Rd has been re-
moved, and the Lorentz manifold M and the coordinate neighbourhood N can be
replaced by E = R ×(0, ∞) × S2 and V = R ×(0, ∞) × V1 , with the metric

g o e = − 1 − Λ r2 dv 2 − 2 dv dr + r 2 g1 .
3
Now we have that
v ◦ γ = const
r) ◦ γ = const
(v + 2 e
for ingoing light. This shows that the surface r = Rd acts as a one-way membrane,
letting the light rays cross only from the inside to the outside.
We can also introduce a coordinate function u = t + e r and obtain a local chart
(ψ i e , N) on M with coordinate functions u, r, θ and φ , called ingoing coordinates,
satisfying
r 2 = −du (du − 2 de
−dt 2 + de r)
so that

g|N = − 1 − Λ r2 du 2 + 2 du dr + r 2 ( d θ 2 + sin2 θ d φ 2 ). (4)
3
The Lorentz manifold M and the coordinate neighbourhood N can again be replaced
by E and V , but this time with the metric

g i e = − 1 − Λ r2 du 2 + 2 du dr + r 2 g1 .
3
Now we have that

r) ◦ γ = const
(u − 2 e
u ◦ γ = const
for ingoing light. Hence the light rays can cross the surface r = Rd only from the
outside to the inside.
The Lorentz manifolds (M 1 , g|M 1 ) , (M 1 , g o e |M 1 ) and (M 1 , g i e |M 1 ) are iso-
metric isomorphic, as are the Lorentz manifolds (M 2 , g|M 2 ) , (M 2 , g o e |M 2 ) and
(M 2 , g i e |M 2 ) . But the Lorentz manifolds (E, g o e ) and (E, g i e ) are not isometric iso-
morphic. However, in the next section we will see that both (E, g o e ) and (E, g i e ) can
be isometrically embedded in a larger Lorentz manifold which is analogous to the
Kruskal–Szekeres extension of the Schwarzschild spacetime.
10.194 Blau–Guendelman–Guth coordinates Using the coordinate functions
v = t −e
r , u = t +e
r
described in 10.193, and fixing functions ε1 : M → {−1, 1} and ε2 : M → {−1, 1}

which are constant on M1 and M2 with
ε1 ε2 = sgn(Rd − r) ,
we can define new coordinate functions

ve = ε1 exp v , ue = − ε2 exp − u
Rd Rd
and
T = 12 (e
v + ue) , R = 12 (e
v − ue)
so that R −r
R2 − T 2 = − ve ue = sgn(Rd − r) exp − 2er = d . (1)
Rd Rd + r
The transformation from the de Sitter spacetime M is given by
R − r 1/2
d t
T = ε1 sinh
Rd + r Rd
R − r 1/2
d t
R = ε1 cosh
Rd + r Rd
when 0 < r < Rd , mapping M1 onto

p
B2−ε1 = {(T, R) ∈ R2 ||T | < ε1 R < T 2 + 1 } × S2 ,
and by
FIBRE BUNDLES 459
r − R 1/2
d t
T = ε1 cosh
r + Rd Rd
r − R 1/2
d t
R = ε1 sinh
r + Rd Rd
when r > Rd , mapping M2 onto

p
B3−ε1 = {(T, R) ∈ R2 ||R| < ε1 T < R2 + 1 } × S2 .
The inverse transformation is given by
r 1 − (R2 − T 2 )
= , (2)
Rd 1 + (R2 − T 2 )
and by
t
tanh =T
Rd R
when 0 < r < Rd , and
t
tanh =R
Rd T
when r > Rd . Now using that
R −r
v d ue = 12 (− ve ue) (− dv du) = 12
−dT 2 + dR2 = − de d
r 2) ,
(−dt 2 + de
Rd Rd Rd + r
we have that
g|N = (Rd + r)2 (−dT 2 + dR2 ) + r 2 ( d θ 2 + sin2 θ d φ 2 ) (3)
where r is given by (2). As this coordinate expression of the de Sitter metric is well
defined for all r > 0, and using all possible alternatives for ε1 , we obtain a new
Lorentz manifold
B = {(T, R) ∈ R2 | |R2 − T 2 | < 1} × S2
and a local chart (ψ b ,W ) on B with coordinate neighbourhood
W = {(T, R) ∈ R2 | |R2 − T 2 | < 1} × V1
and coordinate functions T , R, θ and φ , called Blau–Guendelman–Guth (BGG) co-

ordinates. The metric g b on B is
g b = (Rd + r)2 (−dT 2 + dR2) + r 2 g1 .
To find the transformations from the extensions (E, g o e ) and (E, g i e ) of (M, g),
we use that
R −r
ue = − ε2 exp − 2er exp − v = − ε1 d exp − v
Rd Rd Rd + r Rd
and R −r
ve = ε1 exp − 2er exp u = ε2 d exp u ,
Rd Rd Rd + r Rd
which imply that
n R −r o
ε1 v
T= exp − d
exp − v
2 Rd Rd + r Rd
n R −r o
ε1 v
R= exp + d
exp − v
2 Rd Rd + r Rd
and
n R − r o
ε2 u
T= d
exp − exp − u
2 Rd + r Rd Rd
n R − r o
ε2 u
R= d
exp + exp − u .
2 Rd + r Rd Rd
The inverse transformations are given by (2) and by

ε1 exp v = R + T , ε2 exp − u = R − T .
Rd Rd
Choosing ε1 = 1 and ε2 = 1 in these formulae, we see that the regions
B12 = {((T, R), u) ∈ B | R > −T } and B14 = {((T, R), u) ∈ B | R > T }
are isometric to the outgoing and ingoing extension of B1 , respectively. The regions
B1 and B3 are both isometric to the internal de Sitter spacetime, but these two regions
cannot be connected by any timelike or null curve.
By (2) we also have that
2Rd
Rd + r =
1 + (R2 − T 2 )
so that
(Rd + r)2 (−dT 2 + dR2)

= d (Rd + r) [1 + (R2 − T 2 )] dr + (Rd + r )2 (dR2 − dT 2 )
= [1 + (R2 − T 2 )] dr 2 + (Rd + r) d (R2 − T 2 ) dr + (Rd + r)2 (dR2 − dT 2 )
= −{T dr + (Rd + r) dT }2 + {R dr + (Rd + r) dR }2 + dr2
= −{d (Rd + r) T }2 + {d (Rd + r) R}2 + dr2
and
FIBRE BUNDLES 461
−{(Rd + r) T }2 + {(Rd + r) R }2 + r2

4(R2 −T 2 ) [1 − (R2 − T 2 )]2
= R2d + = R2d .
[1 + (R2 − T 2 )]2 [1 + (R2 − T 2 )]2
Hence there is an isometric embedding α : B → R51 of B as a 4-pseudosphere
S4Rd = {(v, w, x, y, z) ∈ R5 | − v2 + w2 + x2 + y2 + z2 = R2d }
with radius Rd in the pseudo-Euclidean space R51 given by

2 [1 − (R2 − T 2 )]2
α ((T, R), u) = Rd (T, R) , u (4)
[1 + (R2 − T 2 )]2 [1 + (R2 − T 2 )]2
for ((T, R), u) ∈ B . On the coordinate domain W we have that
v ◦ α |W = (Rd + r) T
w ◦ α |W = (Rd + r) R
x ◦ α |W = r sin θ cos φ (5)
y ◦ α |W = r sin θ sin φ
z ◦ α |W = r cos θ
where r is given by (2), so that

2Rd T
v ◦ α |W =
1 + (R2 − T 2 )
2Rd R
w ◦ α |W =
1 + (R2 − T 2 )
1 − (R2 − T 2 )
x ◦ α |W = Rd sin θ cos φ
1 + (R2 − T 2 )
1 − (R2 − T 2 )
y ◦ α |W = Rd sin θ sin φ
1 + (R2 − T 2 )
1 − (R2 − T 2 )
z ◦ α |W = Rd cos θ
1 + (R2 − T 2 )
in BGG coordinates. On the open submanifold
U = {(T, R) ∈ R2 | 0 < |R2 − T 2 | < 1} × V1
we also have that

(
ε1 (R2d − r2 )1/2 sinh(t/Rd ) when 0 < r < Rd
v ◦ α |U =
ε1 (r2 − R2d )1/2 cosh(t/Rd ) when r > Rd
(
ε1 (R2d − r2 )1/2 cosh(t/Rd ) when 0 < r < Rd
w ◦ α |U =
ε1 (r2 − R2d )1/2 sinh(t/Rd ) when r > Rd
x ◦ α |U = r sin θ cos φ
y ◦ α |U = r sin θ sin φ
z ◦ α |U = r cos θ
in Eddington coordinates.
AFFINE TRANSFORMATIONS AND KILLING VECTOR FIELDS

10.195 Definition Let M and N be smooth manifolds with linear connections H
e respectively. Then a smooth map f : M → N is called affine if
and H,
τe tt10 ◦ f ∗ γ (t 0 ) = f ∗ γ (t 1 ) ◦ τ tt10
for every smooth curve γ : I → M defined on an open interval I and for each pair of
points t0 ,t1 ∈ I, where τ tt 01 and τe tt 01 are the parallel translations from t 0 to t 1 in T M
and T N along γ and f ◦ γ , respectively.
10.196 Remark If f : M → N is an affine map and F : I → T M is a lifting of a

smooth curve γ : I → M which is parallel along γ , then it follows from Proposition
10.92 that the lifting f∗ ◦ F of f ◦ γ is parallel along f ◦ γ . Applying this to the lifting
γ ′ of γ , it follows that f ◦ γ is a geodesic on N whenever γ is a geodesic on M . It also
follows that
exp ◦ f∗ | D = f ◦ exp .
Indeed, let v ∈ D p and γv : I(v) → M be the unique maximal geodesic on M with
γ (0) = p and γ ′ (0) = v. Then f ◦ γv is a geodesic on N with f ◦ γv (0) = f (p) and
( f ◦ γv )′ (0) = f∗ (v) , and we have that
exp ◦ f∗ (v) = f ◦ γv (1) = f ◦ exp (v) .
10.197 Proposition Let M and N be smooth manifolds with linear connections

H and H ′ , respectively, and let f : M → N be an affine map. If X, Y and Z are vector
fields on M which are f -related to the vector fields X ′ , Y ′ and Z ′ on N, then we have
that
FIBRE BUNDLES 463
(1) ∇X Y and ∇ ′X ′ Y ′ are f -related, where ∇ and ∇ ′ are the covariant derivatives of
H and H ′ on M and N, respectively.
(2) T (X,Y ) and T ′ (X ′ ,Y ′ ) are f -related, where T and T ′ are the torsion tensors of
H and H ′ on M and N, respectively.
(3) R(X,Y ) Z and R ′ (X ′ ,Y ′ ) Z ′ are f -related, where R and R ′ are the curvature
tensors of H and H ′ on M and N, respectively.
PROOF : (1) Let γ : I → M be a smooth curve defined on an open interval I contain-
t
ing t 0 with γ (t 0 ) = p and γ ′ (t 0 ) = X p , and let τ tt 0 and τe t 0 be the parallel translations
from t to t 0 in T M and T N along γ and f ◦ γ , respectively. Then we have that
t t
f ∗ p ◦ τ tt 0 Y (γ (t)) = τe t 0 ◦ f ∗ γ (t) Y (γ (t)) = τe t 0 Y ′ ( f ◦ γ (t))
for t ∈ I . Since ( f ◦ γ ) ′ (t 0 ) = f ∗ (X p ) = X ′f (p) , it follows from Proposition 10.78 that
f ∗ ( ∇Xp Y ) = ∇ ′X ′ Y ′ ,
f (p)
showing that ∇X Y and ∇ ′ X ′ Y ′ are f -related.

(2) Follows from (1) and Propositions 4.88 and 10.102.
(3) Follows from (1) and Propositions 4.88 and 10.107.
10.198 Let f : M → N be a diffeomorphism between the n-dimensional smooth

manifolds M and N. Then we have an isomorphism ( f˜, f , id) between the frame
bundles π : L (M) → M and π ′ : L (N) → N obtained from the equivalence ( f ∗ , f )
between the tangent bundles πe : T M → M and πe′ : T N → N as described in Example
10.21, where the map f˜ : L (M) → L (N) is said to be induced by f and is given by
f˜(u) = f ∗ p ◦ u
for p ∈ M and u ∈ L p (M) .

If θ and θ ′ are the dual forms on L (M) and L (N) , we have that
f˜ ∗(θ ′ ) = θ (1)
since
f˜ ∗(θ ′ )(u) = θ ′ ( f˜(u)) ◦ f˜∗ u = f˜(u) −1 ◦ π ∗′ f˜(u) ◦ f˜∗ u
= ( f ∗ p ◦ u) −1 ◦ ( f ◦ π ) ∗ u = u−1 ◦ π ∗ u = θ (u)
for p ∈ M and u ∈ L p (M) . Moreover, f˜ : L (M) → L (N) is the unique diffeomor-

phism with π ◦ f˜ = f ◦ π satisfying (1). Indeed, if g̃ : L (M) → L (N) is a diffeo-
morphism with π ◦ g̃ = f ◦ π and g̃ ∗(θ ′ ) = θ , then
u−1 ◦ π ∗ u = θ (u) = g̃ ∗(θ ′ )(u) = θ ′ (g̃(u)) ◦ g̃ ∗ u
= g̃(u) −1 ◦ π ∗′ g̃(u) ◦ g̃ ∗ u = g̃(u) −1 ◦ f ∗ p ◦ π ∗ u
which implies that

g̃(u) = f ∗ p ◦ u = f˜(u)
for p ∈ M and u ∈ L p (M) .
The fundamental vector fields σ (X) and σ ′ (X) on L (M) and L (N) are f˜-
related for each X ∈ gl (Rn ) , since
σ ′f˜(u) (F) = f˜(u) ◦ F = R F ◦ f˜(u) = f˜ ◦ R F (u) = f˜(u ◦ F ) = f˜ ◦ σu (F)
for u ∈ L (M) and F ∈ Gl (Rn ) , which implies that

σ ′ (X) f˜(u) = σ ′f˜(u) ∗ (X) = f˜∗ ◦ σu ∗ (X) = f˜∗ ( σ (X) u )
for u ∈ L (M) and X ∈ gl (Rn ) .

10.199 Proposition Let f : M → N be a diffeomorphism between the n-
dimensional smooth manifolds M and N with linear connections H and H ′ , respec-
tively, and let f˜ : L (M) → L (N) be the map induced by f . Then the following
assertions are equivalent :
(1) f is affine.
(2) f˜∗ (Hu ) = Hf′˜(u) for every u ∈ L (M) .
(3) f˜∗ ( ω ′ ) = ω , where ω and ω ′ are the connection forms of H and H ′ .

(4) If X and Y are vector fields on M which are f -related to the vector fields X ′ and
Y ′ on N , and if ∇ and ∇ ′ are the covariant derivatives of H and H ′ on M and
N, respectively, then ∇X Y and ∇ ′X ′ Y ′ are f -related.
(5) The basic vector fields B(ξ ) and B ′ (ξ ) on L (M) and L (N) are f˜-related for
each ξ ∈ Rn .
PROOF : We first show that (3) implies (2). Assuming that f˜ ∗ ( ω ′ ) = ω , we have
that ω ′ ( f˜(u)) ◦ f˜∗ u = ω (u) which implies that f˜∗ (Hu ) ⊂ Hf′˜(u) for every u ∈ L (M) .
Now since Hu and Hf′˜(u) both have dimension n by Proposition 10.49, we thus obtain
(2).
We next show that (2) implies (1). Assuming that β : I → L (M) is a horizontal
lift of a smooth curve γ : I → M in M defined on an open interval I, it follows from
t t
(2) that f˜ ◦ β is a horizontal lift of f ◦ γ . Hence if τ t 01 and τe t 01 are the parallel trans-
lations from t 0 to t 1 in T M and T N along γ and f ◦ γ , we have that
τe tt10 = ( f˜ ◦ β )(t 1 ) ◦ ( f˜ ◦ β )(t 0 ) −1 = { f ∗ γ (t 1 ) ◦ β (t 1 ) } ◦ { f ∗ γ (t 0 ) ◦ β (t 0 ) } −1
= f ∗ γ (t 1 ) ◦ τ tt10 ◦ f ∗−1γ (t ,
0)
showing that (1) is true.

FIBRE BUNDLES 465
From Proposition 10.197 we know that (1) implies (4), so it only remains to show
that (4) implies (3) in order to prove that the first four assertions are equivalent. By
Proposition 10.168 there is a linear connection H ′′ on N with a connection form ω ′′
satisfying f˜∗ ( ω ′′ ) = ω . From the first part of the proof we know that the corre-
sponding covariant derivative ∇′′ satisfies (4) with H ′ and ∇′ replaced with H ′′ and
∇′′ . Assuming that (4) is also true for H ′ and ∇′ , it follows that ∇′ and ∇′′ are two
coinciding Koszul connections on N . By Corollary 10.132 we therefore have that
ω ′ = ω ′′ , which shows that (3) is true.
Finally, we will show that (5) is equivalent to (2). We first note that
π ∗′ ◦ f˜∗ (B(ξ ) u ) = f ∗ ◦ π ∗ (B(ξ ) u ) = f ∗ ◦ u(ξ ) = f˜(u)(ξ )
for u ∈ L (M) and ξ ∈ Rn . Assuming (2), we also know that B(ξ ) u ∈ Hu implies
that f˜∗ (B(ξ ) u ) ∈ Hf′˜(u) . Hence it follows from Proposition 10.96 that
f˜∗ (B(ξ ) u ) = B ′ (ξ ) f˜(u)
for every u ∈ L (M) and ξ ∈ Rn , which shows that (5) is true.

Conversely, it follows from the last equality that (5) implies (2), since
Hu = {B(ξ ) u |ξ ∈ Rn } and Hf′˜(u) = {B ′ (ξ ) f˜(u) |ξ ∈ Rn }
by formula (2) in Proposition 10.96.
10.200 Proposition Let π : L (M) → M be the frame bundle of a smooth man-

ifold M, and let X be a vector field on M with global flow γ : D(X) → M. Then we
have a map γe : B → L (M) defined on the open subset B = (id × π )−1 (D(X)) of
R × L (M) which is given by
γe(t, u) = γet (u)
for (t, u) ∈ B, where γet : π −1 (Dt (X)) → π −1 (D−t (X)) is the map induced by the
diffeomorphism γ t : Dt (X) → D−t (X) as described in 10.198.
The curve γeu : I(p) → L (M) defined by γeu (t) = γe(t, u) for t ∈ I(p) is smooth
for each p ∈ M and u ∈ L p (M) , and we have a vector field Xe on L (M) with global
e ′
flow γe which is given by X(u) = γe u (0) for u ∈ L (M) . Xe is called the natural lift of
X , and it is the unique vector field on L (M) satisfying
(1) (R F ) ∗ Xe = Xe for every F ∈ Gl (Rn ) .
(2) L Xe θ = 0 .
(3) Xe and X are π -related.

Conversely, given a vector field Xe on L (M) satisfying assertion (1) and (2), there is
a unique vector field X on M which satisfies assertion (3).
PROOF : We first show that Xe is smooth on L (M) . Given a point p on M, we

let (y,V ) be a local chart around p , and choose an open neighbourhood U of p
and an open interval I containing 0 so that γ (I × U) ⊂ V . Now let x = y|U , and
let (φ x , π −1 (U)) and (φ y , π −1 (V )) be the corresponding local trivializations in the
frame bundle L (M) as defined in Remark 10.22. Furthermore, let (β , Gl (Rn )) be
the local chart on Gl (Rn ) given by
b(i, j)
β (G) = e j ◦ G(ei )
for G ∈ Gl (Rn ) and i, j = 1, ... , n , where b : In × In → I n2 is the bijection defined
in Remark 4.27 and E = {e1 , ..., en } is the standard basis for Rn with dual basis
E ∗ = {e1 , ..., en } . Then we obtain local charts (e
x, π −1 (U)) and (e
y, π −1 (V )) on L (M)
where xe = (x × β ) ◦ φ x and ye = (y × β ) ◦ φ y , and we have that
e
X(u) y ◦ γeu ) ′ (0) ] u
= [ ye, (e
so that
e
t ye ◦ X(u) = (u, h(u))
y ◦ γeu ) ′ (0) for u ∈ π −1 (U) . Now we have that
where h(u) = (e
φ y ◦ γeu (t) = (γ t (p),t y,γ t (p) ◦ γ t ∗ p ◦ t −1

x,p ◦ t x,p ◦ u)
= (γ t (p), D(y ◦ γ t ◦ x−1 )(x(p)) ◦ φ x,p (u))
= (γ (t, p), D2 (y ◦ γ ◦ (id × x)−1 )(t, x(p)) ◦ φ x,p (u))
for (t, u) ∈ I × π −1 (U) where π (u) = p . This implies that γeu is smooth in I for
u ∈ π −1 (U) , and that
h ◦ xe −1 (a, b) = (D1 (y ◦ γ ◦ (id × x)−1 )(0, a),

β (D1 D2 (y ◦ γ ◦ (id × x)−1 )(0, a) ◦ β −1 (b)))
2
for (a, b) ∈ xe(π −1 (U)) ⊂ Rn × Rn , thus showing that Xe is smooth.
Using that (γet , γ t , id) is a homomorphism, we see that Xe satisfies (3) since
π ◦ γeu (t) = π ◦ γet (u) = γ t ◦ π (u) = γ π (u) (t)
when t ∈ I(π (u)) , which implies that
e
π∗ ◦ X(u) = (π ◦ γeu ) ′ (0) = γ′ π (u) (0) = X ◦ π (u)
for u ∈ L (M) .
We next prove that γe is the global flow for Xe . From Proposition 3.42 (2) we know
that
γ s ◦ γ t (p) = γ s+t (p)
for p ∈ Dt (X) ∩ γt−1 (Ds (X)) , which implies that
γeγeu (t) (s) = γe(s, γe(t, u)) = γes ◦ γet (u) = γes+t (u) = γe(s + t, u) = γeu (s + t)
FIBRE BUNDLES 467
for u ∈ π −1 (Dt (X)) ∩ γet −1 (π −1 (Ds (X))) when s ∈ I(p) − t . By taking the derivative
with respect to s at s = 0 we see that
e γeu (t)) = γeu′ (t)
X(
when t ∈ I(p) for each p ∈ M and u ∈ L p (M) . Hence it follows from Proposition
3.48 that each γeu is a maximal integral curve for Xe with initial condition u, thus
showing that γe is the global flow for Xe .
Since (γet , γ t , id) is a homomorphism, it follows from Corollary 3.50 that Xe satis-
fies (1) since
R F ◦ γet = γet ◦ R F
for every t ∈ R and F ∈ Gl (Rn ) . Xe also satisfies (2) since
γet∗ (θ )(u) = θ (u)
for every u ∈ L (M) and t ∈ I(π (u)) by formula (1) in 10.198.

To show the uniqueness of Xe , let Xb be a vector field on L (M) which is π -
related to X and satisfies L Xb θ = 0 and (R F ) ∗ Xb = Xb for every F ∈ Gl (Rn ) . Let
b → L (M) be its global flow, and let γ̄ : A → M be the restriction of γ to
γb : D(X)
b . By Proposition 3.49 the condition (R F ) ∗ Xb = Xb implies that
A = (id × π )(D(X))
b = (id × R F )(D(X))
D(X) b
for every F ∈ Gl (Rn ) , which shows that D(X) b = (id × π )−1(A ) . As Xb and X are
π -related, it follows from Proposition 3.48 that π ◦ γbu is an integral curve for X with
b . This
initial condition π (u), so that π ◦ γbu (t) = γ̄ π (u) (t) for u ∈ L (M) and t ∈ I(u)
implies that
π ◦ γbt = γ̄ t ◦ π
for t ∈ R . Finally, the condition L Xb θ = 0 implies that
γbt∗ (θ )(u) = θ (u)

b by Proposition 4.90. Now it follows from 10.198 that
for every t ∈ R and u ∈ Dt (X)
γb = γe|D (X) b e
b which shows that X = X .
To prove the last part of the proposition, consider a vector field Xe on L (M)
satisfying assertion (1) and (2), and let
e
X(p) = π∗ ◦ X(u)
for p ∈ M , where u ∈ L (M) is chosen so that π (u) = p . To see that X is well defined,
let u1 , u2 ∈ L (M) with π (u1 ) = π (u2 ), and choose F ∈ Gl (Rn ) with R F (u1 ) = u2 .
Then it follows from (1) that
e 2 ) = π∗ ◦ Xe ◦ R F (u1 ) = π∗ ◦ (R F ) ∗ ◦ X(u
π∗ ◦ X(u e 1)
e 1 ) = π∗ ◦ X(u
= (π ◦ R F ) ∗ ◦ X(u e 1) .
If s : U → L (M) is a frame field on an open subset U of M, we have that

X |U = π∗ ◦ Xe ◦ s which shows that X is smooth and therefore is a vector field on
M satisfying π∗ ◦ Xe = X ◦ π .
10.201 Definition Let M be a smooth manifold with a linear connection H . Then

a vector field X on M with global flow γ : D(X) → M is called an infinitesimal affine
transformation of M if the diffeomorphism γ t : Dt (X) → D−t (X) is affine for each
t ∈ R.
10.202 Proposition Let X be a vector field on the smooth manifold M n with a

linear connection H , and let Xe be the natural lift of X . Then the following assertions
are equivalent :
(1) X is an infinitesimal affine transformation of M .
(2) L Xe ω = 0 , where ω is the connection form of H .
e B(ξ )] = 0 for every ξ ∈ Rn , where B(ξ ) is the basic vector field determined
(3) [ X,
by ξ .
PROOF : e → L (M) be the global flow for Xe . Then we have that (1) and
Let γe : D(X)
(2) are equivalent, since they are both equivalent to the assertion that
γet∗ (ω )(u) = ω (u)
for every u ∈ L (M) and t ∈ I(u) by Propositions 4.90 and 10.199. We also have that
(1) and (3) are equivalent, since they are both equivalent to the assertion that
γet∗ (B(ξ )) u = B(ξ ) u
for every ξ ∈ Rn , u ∈ L (M) and t ∈ I(u) by Propositions 4.86, 4.90 and 10.199.
10.203 Let f : M → N be a diffeomorphism between the n-dimensional pseudo-

Riemannian manifolds M and N with metrics g and g′ of index k. If f˜ : L (M) →
L (N) is the map induced by f , we have that
f˜(u) ∗ g′ ( f (p)) = u ∗ ( f ∗ g′ )(p)
for every p ∈ M and u ∈ L p (M) . Hence f is an isometry if and only if
f˜ (Oε (M)) ⊂ Oε (N)
where ε is any signature compatible with g.
10.204 Proposition Let f : M → N be an isometry between the pseudo-

Riemannian manifolds M and N with Levi-Civita connections H and H ′ . Then the
map f is affine.
FIBRE BUNDLES 469
PROOF : Let H ′′ be the connection on N obtained from H in the way described in

Proposition 10.168, so that
f˜∗ (H u ) = H′′f˜(u)
for every u ∈ L (M). Then it follows from Propositions 10.199 and 10.197 that f
is affine with respect to the connections H and H ′′ , and that H ′′ has no torsion. By
Proposition 10.172 we know that H is reducible to a connection in Oε (M) . As f
is an isometry, the same must be true for H ′′ by 10.203. This implies that H ′′ must
coincide with the Levi–Civita connection H ′ on N .
10.205 Definition Let M be a pseudo-Riemannian manifold with a metric g . Then

a vector field X on M with global flow γ : D(X) → M is called an infinitesimal isom-
etry or a Killing vector field on M if the diffeomorphism γ t : Dt (X) → D−t (X) is
an isometry for each t ∈ R. By Proposition 10.204 every infinitesimal isometry is an
infinitesimal affine transformation of M with respect to the Levi–Civita connection.
10.206 Proposition Let X be a vector field on the pseudo-Riemannian manifold

M n with a metric g , and let Xe be the natural lift of X . Then the following assertions
are equivalent:
(1) X is an infinitesimal isometry.
(2) Xe |Oε (M) belongs to T Oε (M) .
(3) L X g = 0 .
(4) X(g(Y, Z)) = g([ X,Y ], Z ) + g(Y, [ X, Z ]) for every vector field Y and Z on M.
(5) g(∇Y X, Z ) + g(∇Z X,Y ) = 0 for every vector field Y and Z on M.
PROOF : Let γ : D(X) → M be the global flow for X . Then we have that (1) and (2)
are equivalent, since they are both equivalent to the assertion that
γ̃ t (Oε ( Dt (X))) ⊂ Oε ( D−t (X))
for every t ∈ R by 10.203.

We also have that (1) and (3) are equivalent, since they are both equivalent to the
assertion that
γ t∗ (g)(p) = g(p)
for every p ∈ M and t ∈ I(p) by Proposition 4.90.
Furthermore, we have that (3) and (4) are equivalent, since
X(g(Y, Z)) = LX (g(Y, Z)) = (LX g)(Y, Z) + g(LX Y, Z ) + g(Y, LX Z )

= (LX g)(Y, Z) + g([ X,Y ], Z ) + g(Y, [ X, Z ])
by Remark 4.77 and Propositions 4.79 (5) and 4.86.

Using that the Levi–Civita connection has no torsion so that
T (X,Y ) = ∇X Y − ∇Y X − [ X , Y ] = 0
for every vector field Y , we finally see that (4) and (5) are equivalent since
g(∇Y X, Z ) + g(∇Z X,Y ) = g(∇X Y − [ X,Y ], Z ) + g(Y, ∇X Z − [ X, Z ])

= {g(∇X Y, Z ) + g(Y, ∇X Z )} − {g([ X,Y ], Z ) + g(Y, [ X, Z ])}
= X(g(Y, Z)) − {g([ X,Y ], Z ) + g(Y, [ X, Z ])}
10.207 Let X be a vector field on the pseudo-Riemannian manifold M n with a

metric g , and let (x,U) be a local chart on M. If
n
X |U = ∑ ξ i ∂ i
i=1
∂x
and
g | U = ∑ g i j dxi ⊗ dx j ,
ij
then it follows from Propositions 4.76 and 10.166 and Corollary 10.118 that
LX g | U = ∑ B i j dxi ⊗ dx j
ij
where
n
∂gij ∂ξ k ∂ξ k
Bij = ∑ ξk k + gkj i + g ki
k=1
∂x ∂x ∂xj
n
∂ (g k j ξ k ) ∂ (g ki ξ k ) ∂ gk j ∂g ∂gij
= ∑ + − ξk + kij − k
k=1
∂xi ∂xj ∂xi ∂x ∂x
n n
∂ (g k j ξ k ) ∂ (g ki ξ k ) k l
= ∑ ∂xi
+
∂xj
− 2 ξ ∑ g kl Γ i j
k=1 l=1
n
∂ξj ∂ξi
=
∂xi
+
∂xj
− 2 ∑ ξl Γ il j = ξ j ;i + ξ i ; j .
l=1
Hence X is a Killing vector field if and only if
ξ j ;i + ξ i ; j = 0
for every local chart (x,U) and 1 ≤ i, j ≤ n.

FIBRE BUNDLES 471
CONFORMAL TRANSFORMATIONS
10.208 Definition A diffeomorphism f : M → N between two pseudo-
Riemannian manifolds M and N with metrics g and g′ is said to be conformal if
f ∗ (g′ ) = e 2 σ g
for a smooth function σ ∈ F (M) . Two metrics g and ge on a smooth manifold M

are said to be conformally related if the identity map id : M → M is conformal with
respect to g and ge, i.e., if
ge = e 2 σ g
for a smooth function σ ∈ F (M) .
10.209 Proposition Let g and ge be two metrics on a smooth manifold M which

are conformally related so that
ge = e 2 σ g
for a smooth function σ ∈ F (M), and let R and Re be the corresponding curvature
forms on M. If
U = grad σ
and B ∈ Ω 1 (M; T M) is the bundle-valued 1-form on M defined by
1
B(X) = −X(σ ) U + ∇X U + 2 U(σ ) X
for vector fields X on M , then we have that

e
R(X,Y ) Z = R(X,Y ) Z + {g(X, Z ) B(Y ) − g(Y, Z ) B(X)}
(1)
+ {g(B(X), Z ) Y − g(B(Y ), Z ) X }
for every vector field X, Y and Z on M .
PROOF : Let H and H e be the Levi–Civita connections of g and ge with covariant
e
derivatives ∇ and ∇, respectively, and let K ∈ T 2 (M; T M) be the bundle-valued
covariant tensor field of degree 2 on M defined by
e X Y − ∇X Y
K(X,Y ) = ∇
for vector fields X and Y on M . We will first show that
K(X,Y ) = X(σ ) Y + Y (σ ) X − g(X,Y ) U. (2)
As the connections H and H e have no torsion, the tensor field K is symmetric by

Proposition 10.102. Since H and H e are metrical connections compatible with g and
ge, respectively, it follows from Proposition 10.162 that
e X Y, Z) + ge(Y, ∇
X( ge(Y, Z)) = ge( ∇ e X Z) = e 2 σ {g( ∇
e X Y, Z) + g(Y, ∇
e X Z)}
and
X( ge(Y, Z)) = X(e 2 σ g(Y, Z)) = 2 X(σ ) e 2 σ g(Y, Z)

+ e 2 σ {g(∇X Y, Z) + g(Y, ∇X Z)}
which implies that
g(K(X,Y ), Z) + g(Y, K(X, Z)) = 2 X(σ ) g(Y, Z) (3)
for each vector field X,Y, Z ∈ T1 (M). By a permutation of the vector fields X, Y and
Z, we also have that
g(K(Y, X), Z) + g(X, K(Y, Z)) = 2 Y(σ ) g(X, Z) (4)
and
g(K(Z, X),Y ) + g(X, K(Z,Y )) = 2 Z(σ ) g(X,Y ) (5)
from which we obtain
g(K(X,Y ), Z) = X(σ ) g(Y, Z) + Y (σ ) g(X, Z) − Z(σ ) g(X,Y ). (6)
Now using that

Z(σ ) = d σ (Z) = g(U, Z) ,
formula (6) implies that
g(K(X,Y ) − X(σ ) Y − Y (σ ) X + g(X,Y ) U , Z ) = 0
for every vector field Z, which completes the proof of formula (2). By Proposition
10.107 we have that
e
R(X,Y eX ∇
)Z = ∇ eY Z − ∇
eY ∇
eX Z − ∇
e (7)
[ X ,Y ] Z
where
eX ∇
∇ e Y Z = ∇X (∇Y Z + K(Y, Z)) + K(X, ∇Y Z + K(Y, Z)) (8)
and
e
∇ [ X ,Y ] Z = ∇ [ X ,Y ] Z + K([ X , Y ], Z). (9)
Formula (2) implies that
∇X K(Y, Z) + K(X, ∇Y Z) = X(Y (σ )) Z + Y (σ ) ∇X Z

+ X(Z(σ )) Y + Z(σ ) ∇X Y − X(g(Y, Z)) U − g(Y, Z) ∇X U
+ X(σ ) ∇Y Z + (∇Y Z)(σ ) X − g(X, ∇Y Z) U
= {X(σ ) ∇Y Z + Y (σ ) ∇X Z } − {g(X, ∇Y Z) + g(Y, ∇X Z)} U

+ {g(U, ∇X Z) Y + g(U, ∇Y Z) X } + {X(Y(σ )) Z + Z(σ ) ∇X Y
− g(∇X Y, Z) U } − g(Y, Z) B 1 (X) + g(B 1(X),Z) Y
FIBRE BUNDLES 473
where
B 1 (X) = ∇X U ,
so that
{∇X K(Y, Z) + K(X, ∇Y Z)} − {∇Y K(X, Z) + K(Y, ∇X Z)}
− K([ X , Y ], Z) = {g(X, Z ) B 1 (Y ) − g(Y, Z ) B 1 (X)} (10)
+ {g(B 1 (X), Z ) Y − g(B 1(Y ), Z ) X }.
We also have that
K(X, K(Y, Z)) = X(σ ) {Y (σ ) Z + Z(σ ) Y − g(Y, Z) U }

+ {Y (σ ) Z(σ ) + Z(σ ) Y (σ ) − g(Y, Z) U(σ )} X
− g(X ,Y (σ ) Z + Z(σ ) Y − g(Y, Z) U) U
= X(σ )Y (σ ) Z + Z(σ ) {X(σ ) Y + Y (σ ) X }

− {X(σ ) g(Y, Z) + Y (σ ) g(X, Z)} U − Z(σ ) g(X,Y ) U
− g(Y, Z) B 2 (X) − g(B 2(Y ),Z) X
where
1
B 2 (X) = −X(σ ) U + 2 U(σ ) X ,
so that
K(X, K(Y, Z)) − K(Y, K(X, Z)) = {g(X, Z ) B 2 (Y ) − g(Y, Z ) B 2 (X)}
(11)
+ {g(B 2 (X), Z ) Y − g(B 2(Y ), Z ) X }.
Since B = B 1 + B 2 , formula (1) in the proposition now follows from formula

(7) – (11).
Chapter 11
ISOMETRIC IMMERSIONS AND THE
SECOND FUNDAMENTAL FORM
CONNECTIONS IN REDUCED SUBBUNDLES

11.1 Proposition Let π : P → M be a principal G-bundle with a connection form
ω , and let H be a Lie subgroup of G so that
g = h ⊕W
for a subspace W of g with Ad(H)(W ) ⊂ W . If π ′ : Q → M is a reduced subbundle of

π : P → M with structure group H, and if i : Q → P is the inclusion map and ρ : g → h
is the projection on h, then ρ · i∗ (ω ) is a connection form on Q.
Suppose that η ∈ Ω k (P; g) is a g-valued k-form on P which is tensorial of type
(Ad, g) , and let ρ ′ : g → W be the projection on W and ψ : H → Aut (W ) be the
representation of H on W obtained from Ad . Then ρ ′ · i∗ (η ) is a W -valued k-form
on Q which is tensorial of type (ψ ,W ) .
PROOF : Let α : H → G be the inclusion map, and let u ∈ Q. Then we have that
i ◦ σu (h) = i ◦ Rh (u) = Rα (h) ◦ i(u) = σi(u) ◦ α (h)
for h ∈ H by Definition 10.19, which implies that
i∗ σ (X) u = i∗ σu ∗ (X) = σi(u) ∗ α∗ (X) = σ ( α∗ (X)) i(u)
for X ∈ h. Hence we have that
ρ · i∗ (ω ) (u) (σ (X) u ) = ρ ◦ ω (i(u)) (σ (α∗ (X)) i(u) ) = ρ ◦ α∗ (X) = X
which shows (i) in Definition 10.34. Using Proposition 5.48 we also have that
Rh∗ (ρ · i∗ (ω )) = ρ · Rh∗ i∗ (ω ) = ρ · i∗ Rα∗ (h) (ω )
= (ρ ◦ Ad (α (h)−1 )) · i∗ (ω ) = Ad (h−1 ) · (ρ · i∗ (ω )) ,
since Ad (α (h−1 ))(W ) ⊂ W and
Ad (α (h−1 )) ◦ α∗ (X) = α∗ ◦ Ad (h−1 )(X)
475
for X ∈ h by Proposition 9.14, thereby showing (ii) in Definition 10.34.

Rh∗ (ρ ′ · i∗ (η )) = ρ ′ · Rh∗ i∗ (η ) = ρ ′ · i∗ Rα∗ (h) (η )
= (ρ ′ ◦ Ad (α (h)−1 )) · i∗ (η ) = ψ (h−1 ) · (ρ ′ · i∗ (η )) ,
which shows that ρ ′ · i∗ (η ) is pseudo-tensorial of type (ψ ,W ) . It is also horizon-

tal, since π ◦ i = π ′ implies that i∗ (ker π ′∗ u ) ⊂ ker π ∗ i(u) for u ∈ Q. Hence i∗ maps
vertical vectors in Tu Q into vertical vectors in Ti(u) P by Proposition 10.3.
THE NORMAL BUNDLE AND THE BUNDLE OF ADAPTED

ORTHONORMAL FRAMES
11.2 Let N m and M n be pseudo-Riemannian manifolds with metrics g and g′ of

index r and k, respectively, and let f : M → N be an isometric immersion. Then it fol-
lows from Proposition 7.9 that f∗ (Tp M) and f∗ (Tp M) ⊥ are non-degenerate subspaces
of T f (p) N with
T f (p) N = f∗ (Tp M) ⊕ f∗ (Tp M) ⊥
for every p ∈ M. Let πeM : f ∗ (T N) → M be the pullback of the tangent bundle πe :
T N → N by f with canonical bundle map ( fe, f ) . The pullback ge = fe ∗ (g) of the
metric g is a fibre metric in f ∗ (T N) of index r. There is also a bundle map ei1 : T M →
f ∗ (T N) over M from the tangent bundle πe1 : T M → M so that fe◦ ei1 = f∗ , which is
given by ei1 (v) = (πe1 , f∗ )(v) = (p, f∗ (v)) for p ∈ M and v ∈ Tp M .
We want to define a subbundle πe 2 : N (M) → M of πeM : f ∗ (T N) → M of dimen-
sion d = m − n, called the normal bundle of M in N, with fibre
N p (M) = ei1 (Tp M) ⊥
for p ∈ M. If ei 2 : N (M) → f ∗ (T N) is the inclusion map, then g′′ = ei 2∗ (e g) is a

fibre metric in N (M) of index l = r − k. Let ( fe1 , f ) and ( fe2 , f ) be the bundle maps
from πe1 : T M → M and πe 2 : N (M) → M to πe : T N → N, respectively, where fer =
fe◦ ei r for r = 1, 2 so that fe1 = f∗ and fe2 = fe|N (M) . We also have bundle maps
η 1 : f ∗ (T N) → T M and η 2 : f ∗ (T N) → N (M) over M so that
v = ei1 ◦ η 1 (v) + ei 2 ◦ η 2 (v)
for v ∈ f ∗ (T N) . For each p ∈ M it follows that
v = fe1 ◦ η
e 1,p (v) + e e 2,p (v)
f2◦ η
ISOMETRIC IMMERSIONS AND THE SECOND FUNDAMENTAL FORM 477
for v ∈ T f (p) N , where ηe 1,p : T f (p) N → Tp M and η e 2,p : T f (p) N → N p (M) are the linear
maps defined by η e r,p = η r,p ◦ fe−1p for r = 1, 2 .
Now let π : Oε (N) → N and πM : Oε ( f ∗ (T N)) → M be the orthonormal frame
bundles of N and f ∗ (T N), respectively, with signature ε given by

−1 for 1 ≤ i ≤ k and n + 1 ≤ i ≤ n + l
εi = .
1 for k + 1 ≤ i ≤ n and n + l + 1 ≤ i ≤ m
We have a bundle map ( f ′ , f ) from πM : Oε ( f ∗ (T N)) → M to π : Oε (N) → N ,

where f ′ (u) = fep ◦ u for p ∈ M and u ∈ Oε ,p ( f ∗ (T N)) .
The orthonormal frame bundle πM : Oε ( f ∗ (T N)) → M may also be identified
with the pullback bundle πM ′ : f ∗ (O (N)) → M of π : O (N) → N by f by means of
ε ε
the equivalence e : f (Oε (N)) → Oε ( f ∗ (T N)) over M given by
∗
e(p, u)(a) = (p, u(a))
for p ∈ M , u ∈ Oε , f (p) (N) and a ∈ Rm , since
e(p, u) ∗ ge(p) = ( fep ◦ e(p, u)) ∗ g( f (p)) = u∗ g( f (p))
for p ∈ M and u ∈ Oε , f (p) (N) . We have that ( f ′ ◦ e, f ) is the canonical bundle map
′ : f ∗ (O (N)) → M .
of πM ε
We want to define a reduced subbundle π ′ : O (N, M) → M of πM : Oε ( f ∗ (T N))
→ M, called the bundle of adapted orthonormal frames, with fibre
O p (N, M) = {u ∈ Oε ,p ( f ∗ (T N))|u( Rn × {0}) = ei1 (Tp M)}
for p ∈ M and structure group Ok (Rn )×Ol (Rd ) . Let (i ′ , idM , φ ) be the corresponding
homomorphism, where i ′ : O (N, M) → Oε ( f ∗ (T N)) and φ : Ok (Rn ) × Ol (Rd ) →
Oε (Rm ) are the inclusion maps. Furthermore, let j 1 : Rn → Rn × Rd and j 2 : Rd →
Rn × Rd be the linear maps given by j 1 (a) = (a, 0) and j 2 (b) = (0, b) for a ∈ Rn and
b ∈ Rd , and let k 1 : Rn × Rd → Rn and k 2 : Rn × Rd → Rd be the projections on the
first and second factor.
11.3 Proposition The normal bundle πe 2 : N (M) → M is a subbundle of πeM :

f ∗ (T N) → M, and the bundle π ′ : O (N, M) → M of adapted orthonormal frames is
a reduced subbundle of πM : Oε ( f ∗ (T N)) → M.
PROOF : Let p 0 ∈ M, and choose an orthonormal basis C = {v1 , ..., vm } for T f (p 0 ) N
with signature ε so that vi = f∗ (wi ) for i = 1, ... , n , where B = {w1 , ..., wn } is an or-
thonormal basis for Tp 0 M . By Proposition 2.34 there are local charts (x,U) and (y,V )
around p 0 and f (p 0 ) on M and N, respectively, such that f (U) = {q ∈V |yn+1 (q) =
... = ym (q) = 0} and x1 (q) = y1 ◦ f (q) , . . . , xn (q) = yn ◦ f (q) for q ∈ U. Hence
y ◦ f = j 1 ◦ x , which implies that
−1
ty, f (p) ◦ f∗ p = j 1 ◦ tx,p and f∗ p ◦ tx,p = ty,−1f (p) ◦ j 1
for p ∈ U by the commutative diagram in 2.70, where (tx , πe1−1 (U)) and (ty , πe−1 (V ))
are the local trivializations in the tangent bundles πe1 : T M → M and πe : T N → N
associated with the local charts (x,U) and (y,V ), respectively.
We have an adapted frame field (Y1 , ... ,Ym ) on U with Yi (p 0 ) = (p 0 , vi ) for i =
1, ... , m , given by
Yi (p) = fep−1 ◦ ty,−1f (p) ◦ ty, f (p 0 ) (vi )
for p ∈ U and i = 1, ... , m, so that
−1
Yi (p) = ei1 ◦ tx,p ◦ tx,p 0 (wi ) ∈ e
i1 (Tp M)
for p ∈ U and i = 1, ... , n. Let W be an open connected neighbourhood

of p 0 con-
tained in U so that the i × i -matrix ge(p)(Yα (p),Yβ (p)) , where 1 ≤ α , β ≤ i, is
non-singular for each p ∈ W and 1 ≤ i ≤ m. Then the subspace Vi (p) of f ∗ (T N) p
spanned by { Y1 (p), ... ,Yi (p) } is non-degenerate for p ∈ W and 1 ≤ i ≤ m. Using
this, we are going to constuct an adapted orthonormal frame field (X1 , ... , Xm ) on W
with Xi (p 0 ) = (p 0 , vi ) for i = 1, ... , m.
By applying the Gram-Schmidt process to { Y1 (p), ... ,Ym (p) }, we obtain an
orthogonal basis { Z1 (p), ... , Zm (p) } for f ∗ (T N) p with Zi (p 0 ) = (p 0 , vi ) for i =
1, ... , m , given by
Z1 (p) = Y1 (p)
and
i−1
ge(p) (Yi (p) ,Z j (p))
Zi (p) = Yi (p) − ∑ Z j (p)
j=1 ge(p) (Z j (p) ,Z j (p))
for p ∈ W and i = 2, ... , m. Indeed, assuming inductively that { Z1 (p), ... , Zi−1 (p) }
is an orthogonal basis for Vi−1 (p) , we have that Z j (p) ∈ / Vi−1 (p) ⊥ so that
ge(p)(Z j (p), Z j (p)) 6= 0 for j = 1, ... , i − 1 since Vi−1 (p) is a non-degenerate
subspace of f ∗ (T N) p , thus showing that Zi (p) is well defined. Furthermore,
ge(p)(Zi (p), Zr (p)) = 0 for r = 1, ... , i − 1 which shows that { Z1 (p), ... , Zi (p) }
is an orthogonal basis for Vi (p) . Hence we have an adapted orthonormal frame field
s = (X1 , ... , Xm ) on W with Xi (p 0 ) = (p 0 , vi ) for i = 1, ... , m, given by
Xi (p) = Zi (p) / kZi (p)k
for p ∈ W and i = 1, ... , m. As {Xn+1 (p), ... , Xm (p)} is a basis for N p (M) for p ∈
W , it follows from Proposition 2.62 that the normal bundle πe 2 : N (M) → M is a
subbundle of πeM : f ∗ (T N) → M.
Since the map hi : W → R given by hi (p) = ge(p)(Xi (p), Xi (p)) for p ∈ W is
smooth and W is connected, we have that hi (p) = g( f (p 0 ))(vi , vi ) = εi for p ∈ W
and i = 1, ... , m. This implies that
s(p)∗ ge(p)(a, b) = ge(p)(s(p) a, s(p) b)

!
m m
= ge(p) ∑ ai Xi (p) , ∑ b j X j (p) = g εm (a, b)
i=1 j=1
for every a, b ∈ Rm , which shows that s(p) ∈ O p (N, M) for p ∈ W . Hence the
−1
local trivialization (ρ −1 , πM (W )) in the bundle πM : Oε ( f ∗ (T N)) → M, where
m −1
ρ : W × Oε (R ) → πM (W ) is the diffeomorphism given by
ρ (p, G) = s(p) ◦ G
for p ∈ W and G ∈ Oε (Rm ), induces a bijection ts : π ′ −1 (W ) → W × Ok (Rn ) ×

Ol (Rd ) . Indeed, since ρ (p, G)−1 ( ei1 (Tp M)) = G−1 ( Rn × {0}) for p ∈ W , we have
that ρ (p, G) ∈ π ′ −1 (W ) for p ∈ W if and only if G ∈ Ok (Rn ) × Ol (Rd ) . If s′ =
(X1′ , ... , Xm′ ) is another adapted orthonormal frame field on some open set W ′ , and
if φ : W ∩W ′ → Oε (Rm ) is the transition map between the corresponding local triv-
ializations in Oε ( f ∗ (T N)), then φ (p) ∈ Ok (Rn ) × Ol (Rd ) for p ∈ W ∩ W ′ , since
s(p) = s′ (p) ◦ φ (p) by Proposition 10.52 which implies that φ (p)( Rn × {0}) =
φ (p) ◦ s(p)−1 ( ei1 (Tp M)) = s′ (p)−1 ( ei1 (Tp M)) = Rn × {0} for p ∈ W ∩W ′ . Hence φ
induces a smooth map ψ : W ∩W ′ → Ok (Rn ) × Ol (Rd ) so that
ts′ ◦ ts−1 (p, G) = (p, ψ (p) ◦ G)
for every p ∈ W ∩ W ′ and G ∈ Ok (Rn ) × Ol (Rd ). By Proposition 10.5 there there-

fore is a unique topology, smooth structure and fibre bundle structure on O (N, M)
such that π ′ : O (N, M) → M is a principal fibre bundle over M with structure
group Ok (Rn ) × Ol (Rd ), and such that (ts , π ′ −1 (W )) is a local trivialization for
each adapted orthonormal frame field s = (X1 , ... , Xm ) on some open set W . We
see that the inclusion map i ′ : O (N, M) → Oε ( f ∗ (T N)) is an immersion, and hence
π ′ : O (N, M) → M is a reduced subbundle of the bundle πM : Oε ( f ∗ (T N)) → M.
11.4 Proposition There are homomorphisms (λ 1 , idM , φ 1 ) and (λ 2 , idM , φ 2 )

from the bundle π ′ : O (N, M) → M of adapted orthonormal frames to the or-
thonormal frame bundles π1 : O (M) → M and π2 : O (N (M)) → M, where λ 1 :
O (N, M) → O (M) and λ 2 : O (N, M) → O (N (M)) are the maps defined by
ei r,p ◦ λ r (u) = u ◦ j r
for p ∈ M , u ∈ O p (N, M) and r = 1, 2 , and where φ 1 : Ok (Rn ) × Ol (Rd ) → Ok (Rn )

and φ 2 : Ok (Rn ) × Ol (Rd ) → Ol (Rd ) are the projections on the first and second fac-
tor.
PROOF : Using that
j r ◦ φ r (G) = G ◦ j r
for G ∈ Ok (Rn ) × O l (Rd ) and r = 1, 2 , we see that
ei r,p ◦ λ r (u) ◦ φ r (G) = u ◦ j r ◦ φ r (G) = u ◦ G ◦ j r
which implies that

λ r (u ◦ G) = λ r (u) ◦ φ r (G) (1)
for p ∈ M , u ∈ O p (N, M) and G ∈ Ok (Rn ) × Ol (Rd ). Hence we have that
π r ◦ λ r = idM ◦ π ′
and
λ r ◦ R G = R φ r (G) ◦ λ r
for G ∈ Ok (Rn ) × Ol (Rd ) , and it only remains to prove that λ r is differentible.
Let s = (X1 , ... , Xm ) be an adapted orthonormal frame field on an open set W in
M. By Proposition 3.14 there is a unique orthonormal frame field s 1 = (X1′ , ... , Xn′ )
on W in O (M) with ei1 ◦ X j′ = X j for j = 1, ... , n, and we let s 2 = (Xn+1 , ... , Xm ) so
that s r = λ r ◦ s for r = 1, 2 . If (ts , π ′−1 (W )), (ts 1 , π −1 −1
1 (W )) and (ts 2 , π 2 (W )) are the
local trivializations in O (N, M), O (M) and O (N (M)) associated with the sections
s, s 1 and s 2 , it follows by replacing u with s(p) in (1) that
λ r ◦ ts−1 (p, G) = ts−1

r
(p, φ r (G))
for p ∈ W and G ∈ Ok (Rn ) × Ol (Rd ). Hence we have that
ts r ◦ λ r ◦ ts−1 = idW × φ r ,
showing that λ r is smooth and completing the proof that ( λ r , idM , φ r ) is a homomor-
phism for r = 1, 2 .
11.5 Remark The principal fibre bundles and homorphisms described in 11.2
and in Propositions 11.3 and 11.4 can be summarized in the commutative diagrams
′
i✲ ′
f✲
O (N, M) Oε ( f ∗ (T N)) Oε (N)
π′ πM π
❄ ❄ ❄
id M f
M ✲ M ✲ N
and
O (M) ✛λ 1 O (N, M)
λ✲
2
O (N (M))
π1 π′ π2
❄ ❄ ❄
M ✛ id M M
id M ✲ M
11.6 Let N m and M n be pseudo-Riemannian manifolds, and let f : M → N be an

isometric immersion. Let exp : D → N be the exponential map in the tangent bundle
πe : T N → N, and consider the smooth map φ = exp ◦ fe2 : D e → N defined on the
open set De = fe (D) in N (M) containing ζ (M), where ζ : M → N (M) is the zero
−1
2
section in the normal bundle πe 2 : N (M) → M and ( fe2 , f ) is the bundle map from
πe 2 : N (M) → M to πe : T N → N defined in 11.2. Using that
φ◦ζ = f ,
we see that φ∗ ζ (p) : Tζ (p) N (M) → T f (p) N is the linear isomorphism given by
φ∗ ζ (p) (ζ∗ (v1 ) + i p ∗(v2 )) = f∗ (v1 ) + e

f 2 ◦ ωζ (p) (v2 )
for p ∈ M, v1 ∈ Tp M and v2 ∈ Tζ (p) N p (M) , where i p : N p (M) → N (M) is the inclu-

sion map and ωζ (p) : Tζ (p) N p (M) → N p (M) is the canonical identification defined
in Lemma 2.84.
11.7 Theorem Let N m and M n be pseudo-Riemannian manifolds, and let

f : M → N be an isometric embedding. Then the map φ defined in 11.6 induces a
diffeomorphism from an open subset W of N (M) containing ζ (M) to an open sub-
set U of N containing f (M).
This diffeomorphism is called a tubular map , and the sets W and U are called tubular
neighbourhoods for f in N (M) and N, respectively. When f is the inclusion map,
they are called tubular neighbourhoods of M in N (M) and N.
PROOF : By the inverse function theorem φ induces a diffeomorphism φ p : Wp → Vp′′
from an open neighbourhood Wp of ζ (p) in N (M) to an open neighbourhood Vp′′ of
f (p) in N for each p ∈ M, and we may assume that
φ p (Wp ∩ ζ (M)) = Vp′′ ∩ f (M). (1)
Indeed, since φ p (Wp ∩ ζ (M)) ⊂ Vp′′ ∩ f (M) is an open set in f (M) in the subspace
topology, we have that φ p (Wp ∩ ζ (M)) = f (M) ∩ O for an open set O in N. Replacing
Vp′′ and Wp by Vp′′ ∩ O and φ p−1 (Vp′′ ∩ O), respectively, we see that condition (1) is
satisfied.
Let {Vp′ |p ∈ M} and {Vp |p ∈ M} be locally finite open covers of the open sub-
S
manifold V ′′ = ′′
p∈M Vp of N with Vp ⊂ Vp′ ⊂ Vp′′ for p ∈ M, and let
V = {u ∈ V ′′ |φ p−1 (u) = φq−1 (u) whenever u ∈ Vp ∩Vq for any p, q ∈ M} .
For each p ∈ M there is an open neighbourhood U p′ of f (p) in N and a finite set of

points p1 , ... , pr in M so that U p′ ∩Vq = 0/ for every q ∈ M − {p1 , ... , pr }. If
k
\ r
\ c
f (p) ∈ Vpi ∩ Vpi ,
i=1 i=k+1
then it follows from (1) that

k
\
ζ (p) ∈ φ p−1
i
(Vp′ i )
i=1
which shows that

k r
\ \ c
Up = φ φ p−1
i
(V ′
pi ) ∩ V pi ∩ U p′
i=1 i=k+1
S
is an open neighbourhood of f (p) contained in V . Thus U = p∈M U p and W =
S −1
p∈M φ p (U p ) satisfy the conditions in the theorem, U being an open subset of N
with f (M) ⊂ U ⊂ V .
THE SECOND FUNDAMENTAL FORM

11.8 We now want to apply Proposition 11.1 to the bundle π ′ : O (N, M) → M of
adapted orthonormal frames, which is a reduced subbundle of πM : Oε ( f ∗ (T N)) →
M, in order to obtain a connection form on O (N, M) from the Levi–Civita connection
on Oε (N). We have that Ok (Rn ) × Ol (Rd ) is a Lie subgroup of Oε (Rm ) so that
oε (Rm ) = i1 ∗ ok (Rn ) ⊕ i2 ∗ ol (Rd ) ⊕ k ,
where i1 : Ok (Rn ) → Ok (Rn ) × Ol (Rd ) and i2 : Ol (Rd ) → Ok (Rn ) × Ol (Rd ) are the
Lie group homomorphisms given by i1 (F) = F × id Rd and i2 (G) = id Rn × G for
F ∈ Ok (Rn ) and G ∈ Ol (Rd ) , and
k = {F ∈ oε (Rm )|F( Rn × {0}) ⊂ {0} × Rd and F( {0} × Rd ) ⊂ Rn × {0}}
is a subspace of oε (Rm ) which satisfies
Ad (Ok (Rn ) × Ol (Rd ))(k) ⊂ k .
Let ρ 1 : oε (Rm ) → i1 ∗ ok (Rn ) + i2 ∗ ol (Rd ) , ρ 2 : oε (Rm ) → i1 ∗ ok (Rn ) + i2 ∗ ol (Rd )

and ρ 3 : oε (Rm ) → k be the projections on i1 ∗ ok (Rn ) , i2 ∗ ol (Rd ) and k, respec-
tively, and let ρ = ρ 1 + ρ 2 . We have linear maps ρ 21 : k → L (Rn , Rd ) and ρ 12 :
k → L (Rd , Rn ) given by ρ 21 (F) = k 2 ◦ F ◦ j 1 and ρ 12 (F) = k 1 ◦ F ◦ j 2 for F ∈ k ,
and we let j 3 : k → oε (Rm ) be the inclusion map and ψ : Ok (Rn ) × Ol (Rd ) → Aut(k)
be the representation obtained from Ad.
If θ 1 and θ are the dual forms on O (M) and Oε (N), respectively, we have that
j 1 · λ 1∗ (θ 1 ) = ( f ′ ◦ i ′ ) ∗ (θ ) ,
since
j 1 ◦ λ 1∗ (θ 1 )(u)(v) = j 1 ◦ θ 1 ( λ 1 (u))( λ 1 ∗ (v)) = j 1 ◦ λ 1 (u) −1 ◦ (π1 ◦ λ 1 ) ∗ (v)
= u−1 ◦ ei 1 ◦ π∗′ (v) = ( fep ◦ u)−1 ◦ f∗ ◦ π∗′ (v) = ( f ′ ◦ i ′ )(u)−1 ◦ π∗ ◦ ( f ′ ◦ i ′ ) ∗ (v)

= ( f ′ ◦ i ′ )∗ (θ )(u)(v)
for p ∈ M , u ∈ O p (N, M) and v ∈ Tu O (N, M) . It also follows that
( f ′ ◦ i ′ ) ∗ (θ )(u) = u−1 ◦ ei 1 ◦ π∗′ u
for u ∈ O (N, M) .
Let ω be the Levi–Civita connection form on Oε (N) satisfying the first structure
equation
dθ = − ω ∧ θ .
If ωM = f ′ ∗ ( ω ) is the pull-back of ω to Oε ( f ∗ (T N)), and θ ′ = ( f ′ ◦ i ′ ) ∗ (θ ) and
ω ′ = ( f ′ ◦ i ′ ) ∗ ( ω ) are the pull-backs of θ and ω to O (N, M), it follows from Propo-
sitions 5.63 and 5.70 that
dθ ′ = − ω ′ ∧ θ ′ . (1)
By 10.37 and Proposition 11.1 we know that ρ · ω ′ is a connection form on O (N, M),
and we let ω 0 , ω 1 and ω 2 be the connection forms on Oε ( f ∗ (T N)), O (M) and
O (N (M)) obtained from ρ · ω ′ in the way described in Proposition 10.168 so that
∗
i ′ (ω 0) = (φ∗ ◦ ρ ) · ω ′ and λ r∗ ( ω r ) = (φ r ∗ ◦ ρ ) · ω ′ ,
and hence
ir ∗ · λ r∗ ( ω r ) = ρ r · ω ′ (2)
for r = 1, 2 . Since
ρ 2 · ω ′ (u)( Rn × {0}) = {0} × {0} and ρ 3 · ω ′ (u)( Rn × {0}) ⊂ {0} × Rd
for u ∈ O (N, M) , it follows from (1) that
dθ 1 = − ω 1 ∧ θ 1
showing that ω 1 is the Levi–Civita connection form on O (M) , and
τ∧ θ ′ = 0 (3)
where τ = ρ 3 · ω ′ is a k-valued 1-form on O (N, M) , called the second fundamental

form on O (N, M) , which is tensorial of type (ψ , k) . This follows from Proposition
11.1 since
j 3 · τ = ω ′ − (φ∗ ◦ ρ) · ω ′ = i ′ ∗ (τM) ,
where τ M = ωM − ω 0 is the difference of two connection forms on Oε ( f ∗ (T N)) and
therefore is tensorial of type (Ad , oε (Rm )) .
11.9 By Example 10.29 (b) the fibre bundle π ′′ : E → M with fibre k associated
with O (N, M) is a vector bundle which we contend is equivalent over M to the vector
bundle π ′′′ : Q → M with fibre
Q p = {F ∈ oε ( f ∗ (T N) p )|F( ei1 (Tp M)) ⊂ N p (M) and

F( N p (M)) ⊂ ei1 (Tp M)}
for p ∈ M, with the equivalence e : E → Q given by
e ([ u, F ]) = u ◦ F ◦ u−1
for u ∈ O (N, M) and F ∈ k. This is clearly well defined, for if G ∈ Ok (Rn ) × Ol (Rd ),
then
e ([ u G, ψ (G−1 ) F ]) = (u ◦ G) ◦ (G−1 ◦ F ◦ G) ◦ (G−1 ◦ u−1 )

= u ◦ F ◦ u−1 = e ([ u, F ]) .
To show that e is an equivalence, let (ts , π ′ −1 (W )) be the local trivialization in
O (N, M) assosiated with an adapted orthonormal frame field s on some open set W
in M, and let (t s′ , π ′′ −1 (W )) be the corresponding local trivialization in E defined as
in 10.28 by
t s′ ([ u, F ]) = (p, ψ (ts,p (u)) (F))
for u ∈ O p (N, M) and F ∈ k where p ∈ W . By (2) in 10.28 we also have a local
−1
trivialization (e
ts , πeM (W )) in the vector bundle πeM : f ∗ (T N) → M given by
e
ts (s(p)(a)) = (p, a)
for p ∈ W and a ∈ Rm . Combining this, we have that
t s′ ([ u, F ]) = (p, s(p)−1 ◦ u ◦ F ◦ u−1 ◦ s(p)) = τ s ◦ e ([ u, F ])
where τ s : π ′′′ −1 (W ) → W × k is the equivalence over W induced by the equivalence

τs′ : πM′ −1 (W ) → W × gl (Rm ) defined in Proposition 5.28 for the vector bundle π ′ :
M
−1
Λ 1 ( f ∗ (T N) ; f ∗ (T N)) → M using the local trivialization (e
ts , πeM (W )) in the vector
bundle πeM : f ∗ (T N) → M . Hence we have that
−1
τ s ◦ e ◦ t s′ = id W × k .
Using the equivalence e, we may therefore identify the fibre bundle E associated
with O (N, M) with fibre k with the vector bundle π ′′′ : Q → M , and the linear iso-
morphism µ u : k → E p defined in Example 10.29 (b) is then identified with the linear
isomorphism µ ′u = e p ◦ µ u : k → Q p given by
µ ′u (F) = u ◦ F ◦ u−1
for u ∈ O p (N, M) and F ∈ k where p ∈ M .

Now we have a unique bundle-valued 1-form α ∈ Ω 1 (M; Q) with τ = α , called

the second fundamental form on M, defined by
α (p)(v) = u ◦ τ (u)(w) ◦ u−1
for p ∈ M and v ∈ Tp M, where u ∈ O p (N, M) and w ∈ Tu O (N, M) is chosen so that

π∗′ (w) = v. We also have that
α (p)(v1 )(ei1 (v2 )) = u ◦ τ (u)(w1 ) ( θ ′ (u)(w2 ))
for p ∈ M and v1 , v2 ∈ Tp M, where u ∈ O p (N, M) and wr ∈ Tu O (N, M) is chosen so

that π∗′ (wr ) = vr for r = 1, 2 . Hence it follows from (3) in 11.8 that
α (p)(v1 )(ei1 (v2 )) = α (p)(v2 )(ei1 (v1 ))
since
α (p)(v1 )(ei1 (v2 )) − α (p)(v2 )(ei1 (v1 )) = u ◦ (τ ∧ θ ′ )(u)(w1 , w2 ) = 0 .
11.10 Let E = {e1 , ... , em } be the standard basis for Rm and D = {Fji |1 ≤ i, j ≤
m} be the basis for gl (Rm ) defined in Example 8.10 (b), where Fji : Rm → Rm is the
linear map given by Fji (er ) = δ j r ei for r = 1, ... , m. If s = (Z1 , ... , Zm ) is an adapted
orthonormal frame field on an open set V in M, and τe = s∗ (τ ) is the pull-back of τ
by s, it follows from Proposition 10.105 that
m
α (p)(v) Z i (p) = ∑ τeir (p)(v) Z r (p)
r=n+1
for p ∈ V , v ∈ Tp M and i = 1, ... , n, and

n
α (p)(v) Z r (p) = ∑ τeri (p)(v) Z i (p)
i=1
for p ∈ V , v ∈ Tp M and r = n + 1, ... , m, where τe ji are the components of τe with

respect to the basis D. Since the form τe is oε (Rm ) -valued, we have that
s ir τeri = −τeir
for i = 1, ... , n and r = n + 1, ... , m , where s ir = sgn (i − k − 1/2)(r − n − l − 1/2) .

By Proposition 3.14 there is a unique orthonormal frame field s 1 = ( Y1 , ... ,Yn )
on V in O (M) with ei1 ◦ Y j = Z j for j = 1, ... , n, so that s 1 = λ 1 ◦ s . Let θe j =
s ∗ (θ ′ j ) = ( f ′ ◦ i ′ ◦ s) ∗ (θ j ) = s 1∗ (θ 1j ) for j = 1, ... , n be the pull-backs of the non-
zero components of the 1-form θ ′ by s, which form a dual local basis on V to the
frame field s 1 as described in Remark 10.94. Then we have that
n
τe rj = ∑ A irj θe i
i=1
for 1 ≤ j ≤ n and n + 1 ≤ r ≤ m, where the smooth functions A irj : V → R are defined

by
A irj = τe rj ( Yi )
for 1 ≤ i, j ≤ n and n + 1 ≤ r ≤ m . Using formula (3) in 11.8 we have that
τe∧ θe = 0
so that
n
0= ∑ τe jr ∧ θe j = ∑ (A irj − A rj i ) θe i ∧ θe j
j=1 i< j
for n + 1 ≤ r ≤ m . This implies that
A irj = A rj i
for 1 ≤ i, j ≤ n and n + 1 ≤ r ≤ m .
THE SHAPE TENSOR
11.11 Consider the representation ψ h : Ok (Rn )×Ol (Rd ) → Aut(T h (Rn ; Rd )) of

the structure group Ok (Rn ) × Ol (Rd ) in the bundle O (N, M) of adapted orthonormal
frames on the finite dimensional vector space T h (Rn ; Rd ) given by
ψ h (G)(T ) = φ 2 (G) ◦ T ◦ {φ 1 (G) −1 } h
for G ∈ Ok (Rn ) × Ol (Rd ) and T ∈ T h (Rn ; Rd ) . By Example 10.29 (b) the fibre bun-
dle π ′′ : E → M with fibre T h (Rn ; Rd ) associated with O (N, M) is a vector bundle
which we contend is equivalent over M to the bundle π h : T h (T M ; N (M)) → M,
with the equivalence e : E → T h (T M ; N (M)) given by
e ([ u, T ]) = λ 2 (u) ◦ T ◦ {λ 1 (u) −1 } h
for u ∈ O p (N, M) and T ∈ T h (Rn ; Rd ) where p ∈ M. This is clearly well defined,

for if G ∈ Ok (Rn ) × Ol (Rd ), then
e ([ u ◦ G, ψ h (G −1 ) T ]) = {λ 2 (u) ◦ φ 2 (G) } ◦ {φ 2 (G) −1 ◦ T ◦ {φ 1 (G) } h } ◦
{{φ 1 (G) −1 } h ◦ {λ 1 (u) −1 } h } = e ([ u, T ]) .

in 10.28 by
t s′ ([ u, T ]) = (p, ψ h (ts,p (u)) (T ))
for u ∈ O p (N, M) and T ∈ T h (Rn ; Rd ) where p ∈ W . Since
λ r (u) = λ r (s(p)) ◦ φ r (s(p) −1 ◦ u)
for u ∈ O p (N, M) and r = 1, 2 , it follows that
φ r (s(p) −1 ◦ u) = s r (p) −1 ◦ λ r (u)
where s r = λ r ◦ s. Combining this, we have that
t s′ ([ u, T ]) = (p, s 2 (p) −1 ◦ λ 2 (u) ◦ T ◦ {λ 1 (u) −1 } h ◦ {s 1 (p) } h ) = τ s ◦ e ([ u, T ]) ,
where τ s : (π h ) −1 (W ) → W × T h (Rn ; Rd ) is the equivalence over W defined in

Proposition 5.28, using the local trivializations (e ts 1 , πe−1 e e−1
1 (W )) and (ts 2 , π 2 (W )) in
e
the vector bundles πe1 : T M → M and π 2 : N (M) → M given as in formula (2) in
10.28 by
e
ts r (s r (p)(ar )) = (p, ar )
for p ∈ W , a1 ∈ Rn and a2 ∈ Rd . Hence we have that
−1
τ s ◦ e ◦ t s′ = id W × T h (Rn ;Rd ) .
with O (N, M) with fibre T h (Rn ; Rd ) with the vector bundle π h : T h (T M ; N (M))
→ M , and the linear isomorphism µ u : T h (Rn ; Rd ) → E p defined in Example
10.29 (b) is then identified with the linear isomorphism µ ′u = e p ◦ µ u :
T h (Rn ; Rd ) → T h (Tp M ; N p (M)) given by
µ ′u (T ) = λ 2 (u) ◦ T ◦ {λ 1 (u) −1 } h
for u ∈ O p (N, M) and T ∈ T h (Rn ; Rd ) where p ∈ M .

If φ p : T h (Tp M ; N p (M)) → T h1 (Tp M, N p (M)) and φ : T h (Rn ; Rd ) →
h
T 1 (Rn , Rd ) are the linear isomorphisms defined in 5.15, we have that
(φ p ◦ µ ′u )(T )(v1 , ... , vh , λ ) = φ (T )(λ 1 (u) −1 (v1 ), ... , λ 1 (u) −1 (vh ), λ ◦ λ 2 (u))
for v1 , ... , vh ∈ Tp M , λ ∈ N p∗ (M) , u ∈ O p (N, M) and T ∈ T h (Rn ; Rd ) .

Suppose that v ∈ Tp M is the tangent vector to a smooth curve γ : I → M at p,
where I is an open interval containing t0 and γ (t0 ) = p. If T is a section in the bundle
π h : T h (T M ; N (M)) → M , then the covariant derivative of T at p with respect to
v is given by
∇v T = lim 1 [ (τ 2 ) tt00 +h ◦ T (γ (t0 + h)) ◦ {(τ 1 ) tt00 +h } h − T (p) ]

h→0 h

d

= dt (τ 2 ) tt0 ◦ T (γ (t)) ◦ {(τ 1 ) tt0 } h ,
t0
where (τ 1 ) tt0 : π1 −1 (p) → π1 −1 (γ (t)) is the parallel transport in T M along γ from t0

to t , and (τ 2 ) tt0 : π2 −1 (γ (t)) → π2 −1 (p) is the parallel transport in N (M) along
γ from t to t0 . The limit is taken with respect to the vector space topology on
T h (Tp M ; N p (M)) defined in Proposition 13.117 in the appendix.
11.12 Since the representation ψ 1 : Ok (Rn ) × Ol (Rd ) → Aut(L (Rn , Rd )) defined

in 11.11 satisfies
ρ 21 ◦ ψ (G) = ψ 1 (G) ◦ ρ 21
for every G ∈ Ok (Rn ) × Ol (Rd ) , it follows from Proposition 10.39 that the vector-
valued 1-form ρ 21 · τ on O (N, M) is tensorial of type (ψ 1 , L (Rn , Rd )) . Hence we
have a unique bundle-valued 1-form α 21 ∈ Ω 1 (M; Λ 1 (T M ; N (M))) with ρ 21 · τ =
α 21 , which is given by
α 21 (p)(v) = λ 2 (u) ◦ k 2 ◦ τ (u)(w) ◦ j 1 ◦ λ 1 (u) −1 = η 2 ◦ α (p)(v) ◦ ei 1

π∗′ (w) = v.
Using Proposition 5.31, we also have a symmetric covariant tensor field II ∈
T 2 (M ; N (M)) of degree 2 on M with values in the normal bundle πe 2 : N (M) → M,
called the shape tensor , which is defined by
II(p)(v1 , v2 ) = α 21 (p)(v1 )(v2 )
for p ∈ M and v1 , v2 ∈ Tp M .
11.13 Suppose that v ∈ Tp M is the tangent vector to a smooth curve γ : I → M

at p, where I is an open interval containing 0 and γ (0) = p, and let T be a section in
the bundle π 2 : T 2 (T M ; N (M)) → M . Choose a u ∈ O p (N, M) , and let α : I →
O (N, M) be the horizontal lift of γ with α (0) = u.
Now let B = {v1 , ... , vn } and C = {w1 , ... , wd } be bases for Tp M and
N p (M), respectively, with dual bases B ∗ = {v1 , ... , vn } and C ∗ = {w1 , ... , wd }.
We let vi (t) = (τ 1 ) t0 (vi ) = λ 1 (α (t)) ◦ λ 1 (u) −1 (vi ) and vi (t) = (τ 1 ) t0 (vi ) = vi ◦
λ 1 (u) ◦ λ 1 (α (t)) −1 for i = 1, ... , n and t ∈ I be the vectors and covectors obtained
by parallel transport in T M and T ∗ M along the curve γ from 0 to t. Similarly,
we let wi (t) = (τ 2 ) t0 (wi ) = λ 2 (α (t)) ◦ λ 2 (u) −1 (wi ) and wi (t) = (τ 2 ) t0 (wi ) =
wi ◦ λ 2 (u) ◦ λ 2 (α (t)) −1 for i = 1, ... , d and t ∈ I be the vectors and covectors ob-
tained by parallel transport in N (M) and N ∗ (M). Then Bt = {v1 (t), ... , vn (t)} is a
basis for Tγ (t) M with dual basis Bt∗ = {v1 (t), ... , vn (t)} , since (τ 1 ) t0 : Tp M → Tγ (t) M
is a linear isomorphism. Similarly, Ct = {w1 (t), ... , wd (t)} is a basis for Nγ (t) (M)
with dual basis Ct∗ = {w1 (t), ... , wd (t)} .
If τ t0 denotes parallel transport along γ from 0 to t in T 21 (T M, N (M)) , it fol-
lows from 11.11 that
vi1 (t) ⊗ vi2 (t) ⊗ w j (t) = τ t0 (vi1 ⊗ vi2 ⊗ w j )

for (i1 , i2 , j) ∈ In2 × Id and t ∈ I, which shows that the parallel transport in
T 21 (T M, N (M)) of the basis vectors in T 21 (B, C ) are obtained by taking the tensor
product of the parallel transported basis vectors from B and C ∗ . If
j
φ ◦ T (γ (t)) = ∑ C p i1 , i2 (t) vi1 (t) ⊗ vi2 (t) ⊗ w j (t) ,
i1 , i2 , j
then
φ ◦ τ t0 T (γ (t)) = ∑ C p ij1 , i2 (t) vi1 ⊗ vi2 ⊗ w j
i1 , i2 , j
where now denotes parallel transport in T 2 (T M ; N (M)) , so that C p ij1 , i2 are

τ t0
the components with respect to T 21 (B, C ) of the smooth curve c : I → T 21 (Tp M,
N p (M)) given by c(t) = φ ◦ τ t0 T (γ (t)) . Hence we have that

d
j i1 i2
φ ◦ ∇v T = ∑ dt C p i1 , i2 v ⊗ v ⊗ w j (1)
i1 , i2 , j 0
by Lemma 2.84.
If X, Y and Z are vector fields on M , we have the relation
∇X2p [ T (Y, Z) ] = ( ∇Xp T ) (Yp , Z p ) + T (p) ( ∇X1p Y, Z p ) + T (p) (Yp , ∇X1p Z). (2)
for p ∈ M. Indeed, writing
Yγ (t) = ∑ A pk (t) vk (t) and Zγ (t) = ∑ B pk (t) vk (t) ,

k k
we obtain
!
T (γ (t))(Yγ (t) , Zγ (t) ) = ∑ ∑ C p ij1 , i2 (t) A pi1 (t) B pi2 (t) w j (t) .
j i1 , i2
Hence formula (2) follows from (1) and Remark 10.79.
THE SHAPE OPERATOR

11.14 Consider the representation ψ ′1 : Ok (Rn ) × Ol (Rd ) → Aut(L (Rd , Rn )) of
the structure group Ok (Rn ) × Ol (Rd ) in the bundle O (N, M) of adapted orthonormal
frames on the finite dimensional vector space L (Rd , Rn ) given by
ψ ′1(G)(F) = φ 1 (G) ◦ F ◦ φ 2 (G) −1
for G ∈ Ok (Rn )× Ol (Rd ) and F ∈ L (Rd , Rn ) . By Example 10.29 (b) the fibre bundle
π ′′ : E → M with fibre L (Rd , Rn ) associated with O (N, M) is a vector bundle which
we contend is equivalent over M to the bundle πb 1 : Λ 1 (N (M) ; T M) → M, with the

equivalence e : E → Λ 1 (N (M) ; T M) given by
e ([ u, F ]) = λ 1 (u) ◦ F ◦ λ 2 (u) −1
for u ∈ O p (N, M) and F ∈ L (Rd , Rn ) where p ∈ M. This is clearly well defined, for
if G ∈ Ok (Rn ) × Ol (Rd ), then
e ([ u ◦ G, ψ ′1(G −1 ) F ]) = {λ 1 (u) ◦ φ 1 (G) } ◦ {φ 1 (G) −1 ◦ F ◦ φ 2 (G) } ◦
{φ 2 (G) −1 ◦ λ 2 (u) −1 } = e ([ u, F ]) .
in 10.28 by
t s′ ([ u, F ]) = (p, ψ ′1(ts,p (u)) (F))
for u ∈ O p (N, M) and F ∈ L (Rd , Rn ) where p ∈ W . Since
λ r (u) = λ r (s(p)) ◦ φ r (s(p) −1 ◦ u)
for u ∈ O p (N, M) and r = 1, 2 , it follows that
φ r (s(p) −1 ◦ u) = s r (p) −1 ◦ λ r (u)
where s r = λ r ◦ s. Combining this, we have that
t s′ ([ u, F ]) = (p, s 1 (p) −1 ◦ λ 1 (u) ◦ F ◦ λ 2 (u) −1 ◦ s 2 (p)) = τ s ◦ e ([ u, F ]) ,
where τ s : (πb 1 ) −1 (W ) → W × L (Rd , Rn ) is the equivalence over W defined in Propo-

sition 5.28, using the local trivializations (e ts 1 , πe−1 e e−1
1 (W )) and (ts 2 , π 2 (W )) in the vec-
tor bundles πe1 : T M → M and πe 2 : N (M) → M given as in formula (2) in 10.28
by
e
ts r (s r (p)(ar )) = (p, ar )
for p ∈ W , a1 ∈ Rn and a2 ∈ Rd . Hence we have that
−1
τ s ◦ e ◦ t s′ = id W × L (Rd ,Rn ) .
with O (N, M) with fibre L (Rd , Rn ) with the vector bundle πb 1 : Λ 1 (N (M) ; T M)
→ M , and the linear isomorphism µ u : L (Rd , Rn ) → E p defined in Example
10.29 (b) is then identified with the linear isomorphism µ ′u = e p ◦ µ u : L (Rd , Rn ) →
Λ 1 (N p (M) ; Tp M) given by
µ ′u (F) = λ 1 (u) ◦ F ◦ λ 2 (u) −1
for u ∈ O p (N, M) and F ∈ L (Rd , Rn ) where p ∈ M .

11.15 Since the representation ψ ′1 : Ok (Rn ) × Ol (Rd ) → Aut(L (Rd , Rn )) defined

in 11.14 satisfies
ρ 12 ◦ ψ (G) = ψ ′1(G) ◦ ρ 12
for every G ∈ Ok (Rn ) × Ol (Rd ) , it follows from Proposition 10.39 that the vector-
valued 1-form ρ 12 · τ on O (N, M) is tensorial of type (ψ ′1, L (Rd , Rn )) . Hence we
have a unique bundle-valued 1-form α 12 ∈ Ω 1 (M; Λ 1 (N (M) ; T M)) with ρ 12 · τ =
α 12 , which is given by
α 12 (p)(v) = λ 1 (u) ◦ k 1 ◦ τ (u)(w) ◦ j 2 ◦ λ 2 (u) −1 = η 1 ◦ α (p)(v) ◦ ei 2

π∗′ (w) = v.
Using Proposition 5.32, we also have a section S in the vector bundle πb :
Λ 1 (N (M) ; Λ 1 (T M; T M)) → M defined by
S(p)(v1 )(v2 ) = − α 12 (p)(v2 )(v1 )
for p ∈ M, v1 ∈ N p (M) and v2 ∈ Tp M. By Example 8.58 (b) we have the relation
g′ (p)(S(p)(v1 )(v2 ), v3 ) = g′′ (p)(II(p)(v2 , v3 ), v1 )
for p ∈ M , v1 ∈ N p (M) and v2 , v3 ∈ Tp M. Since the shape tensor II is symmetric,

it follows that S(p)(v) is a self-adjoint linear operator on Tp M for every p ∈ M and
v ∈ N p (M). If Z is a section in the normal bundle πe 2 : N (M) → M on an open subset
V of M, then S(Z) ∈ Ω 1 (V ; T M)) is a bundle-valued 1-form on V with values in T M.
It is also denoted by SZ and is called the shape operator determined by Z.
THE FORMULAE OF GAUSS AND WEINGARTEN

11.16 Let ∇ ′ and ∇ be the covariant derivatives corresponding to the connection
forms ωM and ω on Oε ( f ∗ (T N)) and Oε (N), respectively, and let Z be a section in
πeM : f ∗ (T N) → M. Then we have that
∇v ( fe◦ Z) = fe( ∇v′ Z)
when v ∈ Tp M . If Y1 and Y2 are sections in πe1 : T M → M and πe 2 : N (M) → M,

respectively, then
e r,p ( ∇v ( fer ◦ Yr )) = η r ( ∇v′ ( ei r ◦ Yr ))
η
for r = 1, 2 .
Suppose that Z is the pull-back of a vector field Z ′ on N as defined in Proposition
2.57. Then it follows from Proposition 10.87 that
∇ f∗ (v) Z ′ = fe( ∇v′ Z)

when v ∈ Tp M , and if X and X ′ are vector fields on M and N, respectively, which are
f -related, then
∇X ′ Z ′ ◦ f = fe ◦ ∇X′ Z .
11.17 Theorem Let ∇ ′ and ∇ 0 be the covariant derivatives corresponding to

the connection forms ωM and ω 0 on Oε ( f ∗ (T N)), and let Z be a section in πeM :
f ∗ (T N) → M. If v ∈ Tp M , then
∇v′ Z = ∇v0 Z + α (p)(v) Z p .
PROOF : By Definition 10.77 we know that
∇v′ Z − ∇v0 Z = u ◦ (w − w 0 )(Z) ,
where u ∈ Oε ,p ( f ∗ (T N)), w ∈ ker ωM (u) and w 0 ∈ ker ω 0 (u), and where Z :

Oε ( f ∗ (T N)) → Rm is the smooth map defined by
Z(u) = u −1 ◦ Z(πM (u))
for u ∈ Oε ( f ∗ (T N)) . If σu : Oε (Rm ) → Oε ( f ∗ (T N)) is defined as in Proposition

10.26 by σu (G) = u ◦ G for G ∈ Oε (Rm ), we have that
Z ◦ σu (G) = Z(u ◦ G) = G −1 ◦ Z(u) = Λ u ◦ φ (G)
for G ∈ Oε (Rm ) , where φ : Oε (Rm ) → Oε (Rm ) is the smooth map given by φ (G) =
G −1 for G ∈ Oε (Rm ) , and Λu : gl (Rm ) → Rm is the linear map given by
Λu (F) = F ◦ Z (u)
for F ∈ gl (Rm ) . As w − w 0 ∈ Vu , we also have that
w − w 0 = σu ∗ ◦ ωM (u)(w − w 0 ) = − σu ∗ ◦ ωM (u)(w 0 )
= − σu ∗ ◦ (ωM − ω 0 )(u)(w 0 ) = − σu ∗ ◦ τ M (u)(w 0 )
so that
(w − w 0 )(Z) = − τ M (u)(w 0 )(Z ◦ σu ) = − Λ u ◦ φ ∗ ◦ τ M (u)(w 0 )
= Λ u ◦ τ M (u)(w 0 ) = τ M (u)(w 0 ) ◦ Z (u)
by Corollary 8.33. This implies that
∇v′ Z − ∇v0 Z = u ◦ τ M (u)(w 0 ) ◦ u−1 ◦ Z(p) = α (p)(v) Z p ,
where the last equality follows by choosing u in O p (N, M) .

11.18 Proposition Let ∇ 0 , ∇ 1 and ∇ 2 be the covariant derivatives corresponding

to the connection forms ρ · ω ′ , ω 1 and ω 2 , and let Y1 and Y2 be sections in πe1 : T M →
M and πe 2 : N (M) → M, respectively. If v ∈ Tp M , then
∇v0 ( ei r ◦ Yr ) = ei r ( ∇vr Yr )
for r = 1, 2 , and ∇ 0 is also the covariant derivative corresponding to the connection

form ω 0 on Oε ( f ∗ (T N)), as defined in Theorem 11.17.
PROOF : If Zr = ei r ◦ Yr , then Z r = j r ◦ Yr ◦ λ r since
Z r (u) = u−1 ◦ ei r ◦ Yr ( π ′ (u)) = j r ◦ λ r (u) −1 ◦ Yr ( π ′ (u)) = j r ◦ Yr ◦ λ r (u)
for u ∈ O (N, M) and r = 1, 2 . Now choose a u ∈ O p (N, M) and a w ∈ ker ρ · ω ′ (u) ,

and let ur = λ r (u) and wr = λ r ∗ (w) for r = 1, 2 . Then we have that
ω r (ur )(wr ) = λ r∗ ( ω r )(u)(w) = (φ r ∗ ◦ ρ ) · ω ′ (u)(w) = 0
so that
∇v0 Zr = u ◦ dZ r (u)(w) = u ◦ d ( j r ◦ Yr ◦ λ r )(u)(w)
= u ◦ j r ◦ λ r∗ (d Yr )(u)(w) = ei r ◦ ur ◦ d Yr (ur )(wr ) = ei r ( ∇vr Yr ) ,
11.19 Corollary Let ∇, ∇ 1 and ∇ 2 be the covariant derivatives corresponding to

the metrical connections without torsion in T N , T M and N (M) , and let Y1 and Y2
be sections in πe1 : T M → M and πe 2 : N (M) → M, respectively. If v ∈ Tp M , then
e r,p ( ∇v ( fer ◦ Yr )) = ∇vr Yr

η
and
e 3−r,p ( ∇v ( fer ◦ Yr )) = α (p)(v)( ei r ◦ Yr (p))
ei 3−r ◦ η
for r = 1, 2 .
11.20 Remark Writing ∇v ( fer ◦ Yr ) as a sum of the two orthogonal components

given in the corollary is called the Gauss’ formula when r = 1 and the Weingarten
formula when r = 2 .
11.21 Let N m and M 1 be pseudo-Riemannian manifolds with metrics g and g′ of

index r and k, respectively, and let f : M → N be an isometric immersion and ue be a
unit vector field on M. The index k of g′ can be either 0 or 1, and we have that
< ue(p), ue(p) > = ε
for p ∈ M, where ε = 1 − 2k.

Let ∇ ′ and ∇ 0 be the covariant derivatives corresponding to the connection forms

ωM and ω 0 on Oε ( f ∗ (T N)), and let Z be a section in πeM : f ∗ (T N) → M. We have
that Z = Z1 + Z2 where Zn = ein ◦ η n ◦ Z for n = 1, 2, and we let u = ei 1 ◦ ue. Since
α (p)(v)(Z1 (p)) = ei 2 ◦ η 2 ( ∇v′ Z1 ) = ei 2 ◦ η 2 ( ∇v′ ( ε <Z , u > u))

= ε <Z(p), u(p) > ∇v′ u
and
α (p)(v)(Z2 (p)) = ei 1 ◦ η 1 ( ∇v′ Z2 ) = ε <∇v′ Z2 , u(p) > u(p)
= − ε <Z(p),∇v′ u > u(p) ,
it follows from Theorem 11.17 that
∇v0 Z = ∇v′ Z + ε <Z(p),∇v′ u > u(p) − ε <Z(p), u(p) > ∇v′ u
for p ∈ M and v ∈ Tp M. When N is a Lorenz manifold, ∇v0 Z is called the Fermi

derivative of Z at p with respect to v. The corresponding parallel transport is called
Fermi–Walker transport .
STRAIN AND VORTICITY
11.22 Proposition Let X be a vector field on a pseudo-Riemannian manifold M n ,

and let F be a vector field along an integral curve c : I → M for X obtained by Lie
transport. Then F satisfies Jacobi’s equation
DF
(t) = ∇F(t) X
dr
for t ∈ I, using the standard local chart (r, I) on I where r : I → R is the inclusion
map.
PROOF : Let (x,U) be a local chart around a point c(t0 ) = p0 on the curve c, and
suppose that
n
F(t) = ∑ ai (t) ∂ i
i=1
∂x
c(t)
for t ∈ c−1 (U) and

n
X(p) = ∑ bi (p) ∂ i
i=1
∂x
p
for p ∈ U. If γ : D(X) → M is the global flow for X, we have that c(t) = γ p 0 (t − t0 )

and F(t) = γt−t0 ∗ F(t0 ) for t ∈ I. This implies that
n ∂ (xi ◦ γt−t0 )
ai (t) = F(t)(xi ) = F(t0 )(xi ◦ γt−t0 ) = ∑ a j (t0 ) (p 0 )
j=1
∂xj
for t ∈ c−1 (U), so that

n n
∂ 2 (xi ◦ γ ) i
ai ′ (t0 ) = ∑ a j (t0 ) j (0, p 0 ) = ∑ a j (t0 ) ∂ bj (p 0 ) .
j=1
∂x ∂t j=1
∂x
By Proposition 10.121 and Corollary 10.114 we therefore have that

!
DF
n n
∂

(t0 ) = ∑ ai ′ (t0 ) + ∑ i
Γ jk (p 0 ) b j (p 0 ) ak (t0 ) i
dr i=1 j,k=1
∂x
p0
!!
n n
∂ bi
n
∂

=∑ ∑ a j (t0 ) ∂xj
(p 0 ) + ∑ bk (p 0 ) Γ jk
i
(p 0 )
∂x i = ∇F(t0 ) X .
i=1 j=1 k=1 p0
11.23 Let N m and the open interval I be pseudo-Riemannian manifolds with

metrics g and g′ of index 1, and let c : I → N be an isometric embedding which
is an integral curve for a unit vector field u on N. Let ζ : I → N (I) be the zero
section in the normal bundle πe 2 : N (I) → I. Then the map φ defined in 11.6 induces
a diffeomorphism, also denoted by φ , between tubular neighbourhoods W and U
for c in N (I) and N as described in Theorem 11.7. Disregarding the connected
components of W and U not containing ζ (I) and c(I), respectively, we may assume
that W and U are connected.
Suppose that h =kπe2 ∗ ◦ φ ∗−1 ◦ ukis a non-zero function on U, and let ũ = (1/h) u|U
and û = φ ∗ ũ. While the integral curves of u are parametrized by arc length, we will
show that ũ corresponds to a reparametrization of these curves so that a vector field
F along c obtained by Lie transport with respect to ũ and being orthogonal to c at a
t0 ∈ I, will remain orthogonal to c at all t ∈ I. By taking the Fermi derivative of the
section obtained from F in the normal bundle, we will derive expressions for the rate
of stretching (strain) and the rate of rotation (vorticity) in the space orthogonal to u
at each point on c with respect to a Fermi–Walker transported basis.
If the dimension of N is m = 4, the flow of a timelike unit vector field u can be
thought of as a family of world lines of reference particles or observers constituting
a reference frame R. The vector field u is called the four-velocity field of these ob-
servers, and the 3-dimensional subspace of the tangent space Tp N at a point p ∈ N
orthogonal to u(p) is called the simultaneity space or the rest space of the observer
at p.
Using the standard local chart (r, I) on I, where r : I → R is the inclusion map,
d
we see that dr is a unit vector field on I since it is c-related to u by means of the
isometric embedding c. As we also have that kπe2 ∗ ◦ û k= 1 on W , this implies that
d
πe2 ∗ ◦ û = a dr e2
◦π
where a : W → R is a smooth function with |a| = 1. Furthermore, h ◦ c = 1 so that

d
πe2 ∗ ◦ û ◦ ζ = c−1
∗ ◦ u ◦ c = dr .
Since W is connected it follows that a = 1, showing that the vector field û on W is

d
πe 2 -related to the vector field dr on I.
If γ : D(û) → W is the global flow for û, then it follows from Proposition 3.48
d
that πe 2 ◦ γ p is an integral curve for dr with initial condition πe 2 (p), i.e.,
πe 2 ◦ γ p (t) = πe 2 (p) + t
when p ∈ W and t ∈ I(p). Hence
πe2 ◦ γt−t0 = Tt−t0 ◦ πe2

which implies that
πe2 ∗ ◦ γt−t0 ∗ = Tt−t0 ∗ ◦ πe2 ∗ (1)
for t,t0 ∈ I, where Tt−t0 : I − t + t0 → I is translation by t − t0 .
Now let F be a vector field along c obtained by Lie transport with respect to the
vector field ũ, and with F(t0 ) ∈ c∗ (Tt0 I) ⊥ for a t0 ∈ I. Then Fb = φ ∗−1 ◦ F is a vec-
tor field along ζ obtained by Lie transport with respect to the vector field û so that
b = γt−t ∗ F(t
F(t) b 0 ) for t ∈ I, and F(t
b 0 ) ∈ Vζ (t ) = ker πe2 ∗ ζ (t ) is a vertical tangent
0 0 0
b
vector. From (1) it follows that F(t) ∈ Vζ (t) is also a vertical tangent vector so that
F(t) ∈ c∗ (Tt I) ⊥ for every t ∈ I. Hence F = ce2 ◦ Y for a section Y in the normal bundle
d
πe 2 : N (I) → I, and the Fermi derivative of Y at t with respect to v = dr is given by
t
∇v2 Y = ηe 2,t ( ∇v F ) = η
e 2,t ( ∇F(t) ue)

= ηe 2,t F(t) 1 u + 1 ∇F(t) u = η e 2,t ( ∇F(t) u)
h h ◦ c(t)
for t ∈ I, where ∇ is the covariant derivative corresponding to the Levi–Civita con-

nection on N and ηe 2,t : Tc(t) N → N t (I) is the linear map defined in 11.2.
Let ν : T N → T N be the bundle map over N given by
e 2,t
ν p = c̃2,t ◦ η
for p = c(t) ∈ N, where c is an integral curve for u through p as described above. We
see that ν p : Tp N → Tp N is the orthogonal projection on the fibre
ker u♭ (p) = u(p) ⊥ = c∗ (Tt I) ⊥ = c̃2 (N t (I))

in the distribution ker u♭ on N. If (x, O) is a local chart on N and

m
u|O = ∑ ui ∂ i , (2)
i=1
∂x
we have that

∂

∂ ∂
m
∂

k
νp i = i + i , u(p) u(p) = ∑ ν i (p) k
∂x ∂x ∂x k=1
∂x
p p p p
for p ∈ O, where
ν ik = δ ik + uk ui
for i, k = 1, ... , m.
By composing the bundle-valued 1-form ∇u ∈ Ω 1 (N ; T N) with the equivalence
φ : T 1 (T N ; T N) → T 11 (T N, T N) over N defined in 5.25 and lowering the con-
travariant index, we obtain a covariant tensor field τe = L 11 (φ ◦ ∇u) ∈ T 2 (N) on N
of degree 2, which is given by
τe(p)(v1 , v2 ) = φ ◦ ∇u(p)(v2 , g(p) ♭ (v1 )) = < v1 ,∇v2 u > = ∇v2 u♭ (v1 )
for p ∈ N and v1 , v2 ∈ Tp N . The last equality follows from Proposition 10.163. Using
Corollary 10.118 we therefore have that
m
τe|O = ∑ u i ; j dxi ⊗ dx j .
i, j=1
The pull-back tensor field τ = ν ∗ τe is given by
τ (p)(v1 , v2 ) = < ν (p)(v1 ),∇ν (p) (v2 ) u > = < v1 ,∇ν (p) (v2 ) u >
for p ∈ N and v1 , v2 ∈ Tp N.
Now let θe and ω e be the symmetric and antisymmetric parts of τe, and let θ = ν ∗ θe
∗ e
and ω = ν ω be their pull-backs by ν . The symmetric tensor field θ is called the
rate of strain , and the 2-form ω the vorticity or the rate of rotation . We have that
m
τ |O = ∑ τi j dxi ⊗ dx j
i, j=1
where
m m
τi j = ∑ u k ; l ν ik ν jl = ∑ u i ; l ν jl
k,l=1 l=1
for i, j = 1, ... , m. The corresponding components of θ |O and ω |O are

m
1
θi j = 2 ∑ {u k ; l + u l ; k } ν ik ν jl (3)
k,l=1
and
m
1
ωij = 2 ∑ {u k, l − u l , k } ν ik ν jl (4)
k,l=1
∂ uk
for i, j = 1, ... , m, where u k, l = . From (2) we also have that
∂ xl
m
u♭ |O = ∑ u i dxi
i=1
so that
m m
du♭ |O = ∑ u i, j dx j ∧ dxi = − ∑ (u i, j − u j, i ) dxi ⊗ dx j = −2 ω
e |O .
i, j=1 i, j=1
This implies that

j ∗ du♭ = −2 j ∗ ω
where j : ker(u♭ ) → T N is the inclusion map, since ν ◦ j = j. By Corollary 5.88
it follows that the distribution ker(u♭ ) is integrable if and only if the vorticity ω
vanishes.
Given an orientation on N, we have an induced orientation on ker(u♭ ) so that
an ordered basis (v2 , ..., vm ) in ker(u♭ (p)) is positively oriented if and only if
(u(p), v2 , ..., vm ) is a positively oriented basis in Tp N for each p ∈ N. Then the metric
volume element in ker(u♭ ) is η = i u η̃ , where η̃ is the metric volume element in N.
Indeed, if the positively oriented basis (v2 , ..., vm ) in ker(u♭ (p)) is orthonormal, then
η̃ (p) = u(p) ∗∧ v∗2 ∧ · · · ∧ v∗m and η (p) = v∗2 ∧ · · · ∧ v∗m .
If the dimension of N is m = 4, the vorticity vector field Ω on N is defined by
Ω(p) = −(∗ j ∗ ω (p)) ♯
so that
i Ω(p) η (p) = − j ∗ ω (p) = 1
2 j ∗ du♭ (p)
for p ∈ N. If
4 4
Ω(p) = ∑ Ω i vi , j ∗ ω (p) = ∑ ω jk v∗j ⊗ v∗k
i=2 j,k=2
and
4
η (p) = ∑ η i jk v∗i ⊗ v∗j ⊗ v∗k
i, j,k=2
for a basis {v2 , ..., v4 } in ker(u♭ (p)) with dual basis {v∗2 , ..., v∗4 }, then
4
∑ Ω i η i jk = ω k j (5)
i=2
for 2 ≤ j, k ≤ 4.
11.24 Let N be a Lorentz manifold with a stationary metric g so that

4 2 4
g|O = − e2α dx1 − ∑ wi dxi + ∑ h i j dxi dx j
i=2 i, j=2
in natural units for a local chart (x, O) on N, where α = α̃ ◦ (x2 , x3 , x4 ), wi =

w̃i ◦ (x2 , x3 , x4 ) and h i j = h̃ i j ◦ (x2 , x3 , x4 ) for real-valued functions α̃ , w̃i and h̃ i j on
R3 . Let u be a unit vector field on N so that
u|O = e−α ∂ 1 ,
∂x
which means that the coordinate system (x, O) is comoving in the reference frame
R defined by u. Hence each reference particle in R has a world line c which is an
integral curve for u satisfying xi ◦ c = constant for i = 2, 3, 4. Since
u1 = e−α and ui = 0 for i = 2, 3, 4
and
u1 = −eα and ui = eα wi for i = 2, 3, 4,
the non-zero components of ν are
ν ik = δ 1k w i + δ ik for i = 2, 3, 4 and k = 1, 2, 3, 4.
From (4) in 11.23 it follows that the non-zero components of ω are given by
2 ω i j = (u 1, 1 − u 1, 1) wi w j + (u 1, j − u j, 1 ) wi + (u i, 1 − u 1, i ) w j + (u i, j − u j, i )
= −eα (α, j wi − α,i w j ) + (eα wi ), j − (eα w j ),i = eα (wi, j − w j,i )
for i, j = 2, 3, 4.
Now fix an arbitrary point p ∈ O, and introduce a local chart (x̃, O) on N where
4
x̃1 = eα (p) x1 − ∑ wi (p) xi and x̃i = xi for i = 2, 3, 4,
i=2
so that
4
g(p) = − d x̃1 (p) 2 + ∑ h i j (p) d x̃i (p) d x̃ j (p) .
i, j=2
Assuming that N has an orientation such that the local chart (x, O) is positively ori-
ented, it follows from 7.26 that the metric volume element in ker(u♭ (p)) is given
by
4
η (p) = ∑ η i jk (p) d x̃i (p) ⊗ d x̃ j (p) ⊗ d x̃k (p)
i, j,k=2
where
η i jk (p) = det(h rs (p)) 1/2 ε i jk
for i, j, k = 2, 3, 4. If the vorticity vector Ω(p) is given by

4 4
Ω(p) = ∑ Ω (p) ∂ i = ∑ Ω i (p)
i ∂
+ wi (p) ∂ 1 ,
i=2
∂ x̃ i=2
∂ xi ∂x
p p p
then it follows from (5) in 11.23 that

4
det(h rs (p)) 1/2 ∑ Ω i (p) ε i jk = ω k j (p)
i=2
for every p ∈ O, so that

4
det(h rs ) 1/2 ∑ Ω i ε i jk = ω k j
i=2
for 2 ≤ j, k ≤ 4. Multiplying by 1
2 η i jk = 1
2 det(h rs ) −1/2 ε i jk and summing over
j, k = 2, 3, 4, we obtain
4 4
Ωi = 1
2 det(h rs ) −1/2 ∑ ε i jk ωk j = 1
2 eα det(h rs ) −1/2 ∑ ε i jk wk, j
j,k=2 j,k=2
for i = 2, 3, 4, and
4 1/2
kΩ|O k = ∑ h i j Ωi Ω j .
i, j=2
11.25 Example Let N be an oriented Lorentz manifold with a stationary and

cylindrically symmetric metric g so that
g|O = − f dt 2 + 2k dt d φ + l d φ 2 + eµ (dr2 + dz2 )

2 2
= − f dt − k d φ + D d φ 2 + eµ (dr2 + dz2 )
f f
in natural units for a positively oriented local chart (ψ , O) on N with coordinate func-
tions t, r, φ and z, where f = f˜ ◦ r, k = k̃ ◦ r, l = l˜ ◦ r and µ = µ̃ ◦ r for real functions
f˜, k̃, l˜ and µ̃ , and where D2 = f l + k2 > 0 and f > 0 . Then we have that
1
p 2µ D2 −1/2 k ∂ e−µ f

k ∂
Ω|O = 2 f e =
f f ,r ∂ z 2D f ,r ∂ z
and
e−µ /2 f k
kΩ|O k = .
2D f ,r
THE EQUATIONS OF GAUSS, RICCI AND CODAZZI
11.26 We now consider the second structure equation
d ω = − 12 ω ∧ ω + Ω ,
where Ω is the curvature form on Oε (N) , and the wedge product is with respect to
the bilinear map
ν : oε (Rm ) × oε (Rm ) → oε (Rm )
given by ν (X,Y ) = [ X,Y ]. If Ω ′ = ( f ′ ◦ i ′ ) ∗ (Ω) is the pull-back of Ω to O (N, M),
it follows from Propositions 5.63 and 5.70 that
d ω ′ = − 12 ω ′ ∧ ω ′ + Ω ′ . (1)
For each u ∈ O (N, M) we have that
(ρ 1 · ω ′ ) ∧ (ρ 2 · ω ′ )(u) = 0,
(ρ · ω ′ ) ∧ τ (u) ⊂ k,
(ρ 1 · ω ′ ) ∧ (ρ 1 · ω ′ )(u) ⊂ i1 ∗ ok (Rn ),
(ρ 2 · ω ′ ) ∧ (ρ 2 · ω ′ )(u) ⊂ i2 ∗ ol (Rd ) and
τ ∧ τ (u) ⊂ i1 ∗ ok (Rn ) + i2 ∗ ol (Rd ).
Hence (1) implies that
d (ρ r · ω ′ ) = − 12 (ρ r · ω ′ )∧ (ρ r · ω ′ ) − 12 ρ r · ( τ ∧ τ ) + ρ r · Ω ′ (2)
for r = 1, 2 , and
d τ = − (ρ · ω ′ )∧ τ + ρ 3 · Ω ′ . (3)
If Ω 1 and Ω 2 are the curvature forms of ω 1 and ω 2 on O (M) and O (N (M)),
respectively, we have that
d ω r = − 21 ω r ∧ ω r + Ω r
so that
d (ρ r · ω ′ ) = − 21 (ρ r · ω ′ )∧ (ρ r · ω ′ ) + ir ∗ · λ r∗ (Ω r )
for r = 1, 2 by (2) in 11.8. Combining this equation with (2) we find that
ir ∗ · λ r∗ (Ω r ) = − 12 ρ r · ( τ ∧ τ ) + ρ r · Ω ′ (4)
for r = 1, 2 .
11.27 Consider the representations ρ1 : Ol (Rd ) → Aut(gl (Rd )) and ρ2 , ρ3 :

Ol (R ) → Aut(L (Rn , Rd )) of the structure group Ol (Rd ) in the orthonormal frame
d
bundle π2 : O (N (M)) → M given by

ρ1 (G)(F1 ) = G ◦ F1 , ρ2 (G)(F2 ) = F2 and ρ3 (G)(F3 ) = G ◦ F3
for G ∈ Ol (Rd ) , F1 ∈ gl (Rd ) and F2 , F3 ∈ L (Rn , Rd ) . By Example 10.29 (d)
and (h) the corresponding associated vector bundles may be identified with π22 :
Λ 1 (Rd ; N (M)) → M , πe21 : EL (Rn ,Rd ) → M and π21 : Λ 1 (Rn ; N (M)) → M . Let
κ : gl (Rd ) × L (Rn , Rd ) → L (Rn , Rd )

be the bilinear map given by
κ (F1 , F2 ) = F1 ◦ F2
for F1 ∈ gl (Rd ) and F2 ∈ L (Rn , Rd ) , which satisfies
ρ3 (G) ◦ κ = κ ◦ (ρ1 (G) × ρ2 (G))
for every G ∈ Ol (Rd ) , and let ∇ 22 and ∇ 21 be the covariant derivatives in π22 :
Λ 1 (Rd ; N (M)) → M and π21 : Λ 1 (Rn ; N (M)) → M corresponding to the con-
nection form ω 2 on O (N (M)) . Given a section s ∈ Γ(M ; Λ 1 (Rd ; N (M))) and a
vector-valued function f ∈ F (M ; L (Rn , Rd )) , it follows from 10.76 and Proposition
10.81 that
∇ v21 (s ∧ κ (idM , f )) = (∇ v22 s) ◦ f (p) + s(p) ◦ v ( f )
for every p ∈ M and v ∈ Tp M .
We also have a representation ρ : Ok (Rn ) → Aut(gl (Rn )) of the structure group
Ok (Rn ) in the orthonormal frame bundle π1 : O (M) → M given by
ρ (G)(F) = G ◦ F
for G ∈ Ok (Rn ) and F ∈ gl (Rn ) . By Example 10.29 (h) the corresponding associ-
ated vector bundle may be identified with π11 : Λ 1 (Rn ; T M) → M , and we let ∇ 11
be the covariant derivative in this bundle corresponding to the connection form ω 1
on O (M) . The smooth maps ν 1 : O (M) → Λ 1 (Rn ; T M) and ν 2 : O (N (M)) →
Λ 1 (Rd ; N (M)) defined in Example 10.29 (c) are the inclusion maps.
11.28 Proposition Let R ′ and R be the curvature tensors in f ∗ (T N) and N, re-

spectively, and let ( fb, f ) be the bundle map from πb ′ : Λ 1 ( f ∗ (T N) ; f ∗ (T N)) → M
to πb : Λ 1 (T N ; T N ) → N given by
fb(F) = fep ◦ F ◦ fep−1
for p ∈ M and F ∈ Λ 1 ( f ∗ (T N) p ; f ∗ (T N) p ) . Then R ′ = ( f∗ , fb) ∗ (R) is the pull-back

of R by ( f∗ , fb) as defined in Proposition 5.33, i.e.,
R ′ (p)(v1 , v2 ) = fep−1 ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ◦ fep

for p ∈ M and v1 , v2 ∈ Tp M.
PROOF : Choose u ∈ O p (N, M) and vectors w1 , w2 ∈ Tu O (N, M) so that π∗′ (wi ) = vi

for i = 1, 2 . Then we have that
R ′ (p)(v1 , v2 ) = u ◦ Ω ′(u)(w1 , w2 ) ◦ u−1
= fep−1 ◦ ( fep ◦ u) ◦ Ω ′(u)(w1 , w2 ) ◦ ( fep ◦ u)−1 ◦ fep
= fep−1 ◦ ( f ′ ◦ i ′ )(u) ◦ Ω( f ′ ◦ i ′ (u)) (( f ′ ◦ i ′ ) ∗ (w1 ), ( f ′ ◦ i ′ ) ∗ (w2 ))
◦ ( f ′ ◦ i ′ )(u)−1 ◦ fep
= fep ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ◦ fep ,

since π ( f ′ ◦ i ′ (u)) = f (π ′ (u)) = f (p) and π ∗ (( f ′ ◦ i ′ ) ∗ (wi )) = f∗ (π∗′ (wi )) = f∗ (vi )
for i = 1, 2 .
11.29 Theorem Let N and M be pseudo-Riemannian manifolds, and let f : M →

N be an isometric immersion. If R , R 1 and R 2 are the curvature tensors in N, M and
N (M), respectively, and α is the second fundamental form, then we have that
e r,p ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ◦ fer = R r(p)(v1 , v2 )
η
+ η r,p ◦ α (p)(v1 ) ◦ α (p)(v2 ) ◦ ei r − η r,p ◦ α (p)(v2 ) ◦ α (p)(v1 ) ◦ ei r
for every p ∈ M and v1 , v2 ∈ Tp M .
FIRST PROOF : The theorem follows from formula (4) in 11.26 and Proposition
11.28. Choose u ∈ O p (N, M) and vectors w1 , w2 ∈ Tu O (N, M) so that π∗′ (wi ) = vi
for i = 1, 2 . Then we have that
η r,p ◦ u ◦ ir ∗ · λ r∗ (Ω r )(u)(w1 , w2 ) ◦ u−1 ◦ ei r
= λ r (u) ◦ k r ◦ ir ∗ · λ r∗ (Ω r )(u)(w1 , w2 ) ◦ j r ◦ λ r (u) −1
= λ r (u) ◦ Ω r (λ r (u))(λ r ∗ (w1 ), λ r ∗ (w2 )) ◦ λ r (u) −1
= R r(p)(v1 , v2 )
since π r (λ r (u)) = π ′ (u) = p and π r ∗ (λ r ∗ (wi )) = π∗′ (wi ) = vi for i = 1, 2 . Further-
more, we have that
η r,p ◦ u ◦ ρ r · Ω ′(u)(w1 , w2 ) ◦ u−1 ◦ ei r = η r,p ◦ R ′ (p)(v1 , v2 ) ◦ ei r
e r,p ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ◦ fer
=η
and
1 −1 e
2 η r,p ◦ u ◦ ρ r · ( τ ∧ τ )(u)(w1 , w2 ) ◦ u ◦ i r
= η r,p ◦ u ◦ [ τ (u)(w1 ), τ (u)(w2 ) ] ◦ u−1 ◦ ei r
= η r,p ◦ {u ◦ τ (u)(w1 ) ◦ u−1 } ◦ {u ◦ τ (u)(w2 ) ◦ u−1 } ◦ ei r
− η r,p ◦ {u ◦ τ (u)(w2 ) ◦ u−1 } ◦ {u ◦ τ (u)(w1 ) ◦ u−1 } ◦ ei r
= η r,p ◦ α (p)(v1 ) ◦ α (p)(v2 ) ◦ ei r − η r,p ◦ α (p)(v2 ) ◦ α (p)(v1 ) ◦ ei r
which completes the proof of the theorem.

SECOND PROOF : Let X and Y be vector fields on M with X p = v1 and Yp = v2 , and
let Z r = η r ◦ Z ′ for the section Z ′ = fe∗ (Z) in πeM : f ∗ (T N) → M which is the pull-
back of a vector field Z on N as defined in Proposition 2.57. Then it follows from
Propositions 10.107, 11.28 and Corollary 11.19 that
e r,p ◦ R( f (p))( f∗ (X p ), f∗ (Yp )) ◦ fer (Z r,p )

η
= η r ◦ R ′ (p)(v1 , v2 ) ◦ ei r (Z r,p )
= { ∇Xrp ∇Yr Z r + η r ◦ α (p)(X p ) ◦ α (p)(Yp ) ◦ ei r (Z r,p ) }
− { ∇Yrp ∇Xr Z r + η r ◦ α (p)(Yp ) ◦ α (p)(X p ) ◦ ei r (Z r,p ) } − ∇ [rX ,Y ](p) Z r
= R r(p)(X p ,Yp )(Z r,p ) + η r ◦ α (p)(X p ) ◦ α (p)(Yp ) ◦ ei r (Z r,p )

− η r ◦ α (p)(Yp ) ◦ α (p)(X p ) ◦ ei r (Z r,p ).
11.30 Corollary Let N and M be pseudo-Riemannian manifolds, and let f : M →

N be an isometric immersion. If R , R 1 and R 2 are the curvature tensors in N, M and
N (M), respectively, and α is the second fundamental form, then we have that
< R( f (p))( f∗ (v1 ), f∗ (v2 )) ( fer (v3 )), fer (v4 ) > = < R r(p)(v1 , v2 )(v3 ), v4 >
+ < α (p)(v1 )(ei r (v3 )), α (p)(v2 )(ei r (v4 )) >
− < α (p)(v2 )(ei r (v3 )), α (p)(v1 )(ei r (v4 )) >
e r,p(T f (p) N).

for every p ∈ M , v1 , v2 ∈ Tp M and v3 , v4 ∈ η
PROOF : By Theorem 11.29 we have that
< R( f (p))( f∗ (v1 ), f∗ (v2 )) ( fer (v3 )), fer (v4 ) >
e r,p ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ( fer (v3 )), v4 >
= <η
= < R r(p)(v1 , v2 )(v3 ), v4 >
+ < α (p)(v1 ) ◦ α (p)(v2 ) (ei r (v3 )), ei r (v4 ) >
− < α (p)(v2 ) ◦ α (p)(v1 ) (ei r (v3 )), ei r (v4 ) >
= < R r(p)(v1 , v2 )(v3 ), v4 >
+ < α (p)(v1 )(ei r (v3 )), α (p)(v2 )(ei r (v4 )) >
− < α (p)(v2 )(ei r (v3 )), α (p)(v1 )(ei r (v4 )) >
where the last equality follows from formula (1) in Example 8.58 (b).
11.31 Remark The formula in Theorem 11.29 or its corollary is called Gauss’
equation when r = 1 and the Ricci equation when r = 2. Gauss’ equation can also
be written in the following two forms:

N be an isometric immersion. If R and R 1 are the curvature tensors in N and M,
respectively, and II is the shape tensor, then we have that
< R( f (p))( f∗ (v1 ), f∗ (v2 )) ( f∗ (v3 )), f∗ (v4 ) > = < R 1(p)(v1 , v2 )(v3 ), v4 >
+ < II(p)(v1 , v3 ), II(p)(v2 , v4 ) > − < II(p)(v2 , v3 ), II(p)(v1 , v4 ) >
for every p ∈ M and v1 , v2 , v3 , v4 ∈ Tp M.

N be an isometric immersion. If K and K 1 are the sectional curvatures in N and M,
respectively, and II is the shape tensor, then we have that
K( f∗ (W )) = K 1 (W )
< II(p)(v1 , v1 ), II(p)(v2 , v2 ) > − < II(p)(v1 , v2 ), II(p)(v1 , v2 ) >
−
< v1 , v1 > < v2 , v2 > − < v1 , v2 > 2
for every p ∈ M and every non-degenerate tangent plane W to M at p , where {v1 , v2 }
is a basis for W .
11.34 Theorem (The Codazzi Equation) Let N and M be pseudo-Riemannian

manifolds, and let f : M → N be an isometric immersion. If R is the curvature tensor
in N and II is the shape tensor, we have that
e 2,p ◦ R( f (p))( f∗ (v1 ), f∗ (v2 )) ( f∗ (v3 )) = ( ∇v1 II) ( v2 , v3 ) − ( ∇v2 II) ( v1 , v3 )
η
for every p ∈ M and v1 , v2 , v3 ∈ Tp M .

FIRST PROOF : Let s be an adapted orthonormal frame field on M, and let s r = λ r ◦ s
for r = 1, 2 . Choose a3 ∈ Rn so that s 1 (p)(a3 ) = v3 , and let X , Y and Z be vector
fields on M with X p = v1 , Yp = v2 and Z(q) = s 1 (q)(a3 ) for q ∈ M . By formula (3)
in 11.26 we have that
e 2,p ◦ R( f (p))( f∗ (X p ), f∗ (Yp )) ◦ f∗ (v3 )
η
= η 2 ◦ R ′ (p)(X p , Yp ) ◦ ei 1 (v3 )
e ′ )(p)( X p ,Yp ) ◦ s(p) −1 ◦ ei 1 (v3 )
= η 2 ◦ s(p) ◦ (ρ 3 · Ω
e ′ )(p)( X p ,Yp ) ◦ j 1 (a3 )
= s 2 (p) ◦ k 2 ◦ (ρ 3 · Ω
= s 2 (p) ◦ k 2 ◦ d τe ( X,Y )(p) ◦ j 1 (a3 )
e ′ )∧ τe } (p)( X p ,Yp ) ◦ j 1 (a3 ),
+ s 2 (p) ◦ k 2 ◦ {(ρ · ω
where
d τe ( X,Y )(p) = X p ( τe (Y )) − Yp ( τe ( X)) − τe ( [ X ,Y ] )(p)

= { X p ( τe (Y )) − τe(p) ( ∇X1p Y ) } − {Yp ( τe ( X)) − τe(p) ( ∇Y1p X) }
and
e ′ )∧ τe } (p)( X p ,Yp ) ◦ j 1
k 2 ◦ {(ρ · ω
e ′ )(p)( X p ) ◦ τe (p)(Yp ) − τe (p)(Yp ) ◦ (ρ 1 · ω
= k 2 ◦ { (ρ 2 · ω e ′ )(p)( X p ) } ◦ j 1
e ′ )(p)(Yp ) ◦ τe (p)( X p ) − τe (p)( X p ) ◦ (ρ 1 · ω
− k 2 ◦ { (ρ 2 · ω e ′ )(p)(Yp ) } ◦ j 1 .
The second part after the final equality sign in each of the last two formulas follows
from the first part by interchanging X and Y . Concerning the first parts, we have that
e ′ )(p)( X p ) ◦ τe (p)(Yp ) + X p ( τe (Y )) } ◦ j 1 (a3 )

s 2 (p) ◦ k 2 ◦ {(ρ 2 · ω
e 2 (p)( X p ) ◦ (ρ 21 · τe)(Y ) (p) + s 2 (p) ◦ X p ((ρ 21 · τe) (Y )) } (a3 )
= {s 2 (p) ◦ ω
= {∇X22p (ν 2 ◦ s 2 ) ◦ (ρ 21 · τe) (Y )(p) + (ν 2 ◦ s 2 )(p) ◦ X p ((ρ 21 · τe) (Y )) } (a3 )
= ∇X21p {(ν 2 ◦ s 2 ) ∧ κ (idM , (ρ 21 · τe) (Y )) } (a3 ) = ∇X2p ( II(Y, Z)),
since
{(ν 2 ◦ s 2 ) ∧ κ (idM , (ρ 21 · τe) (Y )) } (q) (a3 ) = s 2 (q) ◦ (ρ 21 · τe) (Y )(q) (a3 )

= η 2 ◦ s(q) ◦ τe (Y )(q) ◦ j 1 (a3 ) = η 2 ◦ α (Y )(q) ◦ ei 1 ◦ s 1 (q)(a3 ) = II(Y, Z)(q)
for q ∈ M . Furthermore, we have that
s 2 (p) ◦ k 2 ◦ τe(p) ( ∇X1p Y ) ◦ j 1 (a3 ) = η 2 ◦ s(p) ◦ τe(p) ( ∇X1p Y ) ◦ j 1 (a3 )
= η 2 ◦ α (p)( ∇X1p Y ) ◦ ei 1 ◦ s 1 (p)(a3 ) = II(p)( ∇X1p Y , Z p )
and
e ′ )(p)( X p ) ◦ j 1 (a3 )
s 2 (p) ◦ k 2 ◦ τe (p)(Yp ) ◦ (ρ 1 · ω
= η 2 ◦ s(p) ◦ τe (p)(Yp ) ◦ s(p) −1 ◦ ei 1 ◦ s 1 (p) ◦ ω
e 1 (p)( X p )(a3 )
= η 2 ◦ α (p)(Yp ) ◦ ei 1 ◦ ∇X11p (ν 1 ◦ s 1 )(a3 ) = II(p)( Yp ,∇X1p Z ).
The theorem now follows from formula (2) in 11.13, as the final term in the last three
formulas combine to (∇Xp II)(Yp , Z p ) .
SECOND PROOF : Let X , Y and Z be vector fields on M with X p = v1 , Yp = v2 and
Z p = v3 . Then it follows from Proposition 10.107, Corollary 11.19 and formula (2)
in 11.13 that
e 2,p ◦ R( f (p))( f∗ (X p ), f∗ (Yp )) ◦ f∗ (Z p )

η
= η 2 ◦ R ′(p)(X p , Yp ) ◦ ei 1 (Z p )
= { ∇X2p [ α (Y )(ei 1 ◦ Z) ] + η 2 ◦ α (p)(X p )( ei 1 ( ∇Y1p Z )) }
− { ∇Y2p [ α (X)(ei 1 ◦ Z) ] + η 2 ◦ α (p)(Yp )( ei 1 ( ∇X1p Z )) }
− η 2 ◦ α (p)(∇X1p Y − ∇Y1p X)(ei 1 (Z p ))
= {∇X2p [ II (Y, Z) ] − II(p) ( ∇X1p Y, Z p ) − II(p) (Yp ,∇X1p Z) }
− {∇Y2p [ II (X, Z) ] − II(p) ( ∇Y1p X, Z p ) − II(p) ( X p ,∇Y1p Z) }

= ( ∇Xp II) (Yp , Z p ) − ( ∇Yp II) ( X p , Z p ).
PSEUDO-RIEMANNIAN HYPERSURFACES
11.35 Let M n be a pseudo-Riemannian hypersurface of a pseudo-Riemannian
manifold N n+1 , with metrics g′ and g of index k and r, respectively. Then the normal
bundle πe 2 : N (M) → M is 1-dimensional with index l = r − k which can be 0 or 1.
The number ε = 1 − 2 l is called the sign of the hypersurface M. A section s of πe 2 on
an open subset V of M is called a normal field to M on V . It is called a unit normal
field if ks(p)k= 1 for every p ∈ V .
11.36 Proposition Let f : N → R be a smooth function on a pseudo-Riemannian

manifold N n+1 with a metric g, and let a ∈ R be a point in the image of f . If
<grad f , grad f > is either positive or negative in a neighbourhood of f −1 (a), then
M = f −1 (a) is a closed pseudo-Riemannian hypersurface of N with sign equal to the
sign of <grad f , grad f >, and with U = ( grad f /kgrad f k)|M as a unit normal field.
PROOF : Since
d f = (grad f ) ♭ 6= 0
in a neighbourhood of f −1 (a), it follows from Proposition 2.41 that M is a closed
hypersurface of N. If i : M → N is the inclusion map, then
<grad f (p), i∗ (v) > = d f (p)(i∗ (v)) = i∗(d f )(p)(v) = d( f ◦ i)(p)(v) = 0
for p ∈ M and v ∈ Tp M, which shows that grad f (p) ∈ / i∗ (Tp M). Hence it follows that
i∗ g is a metric on M, and that (grad f )|M is a normal field to M.
11.37 Let V be a vector space of dimension n with a metric g. Then we have a

fibre metric h in TV defined by
h(p) = ω p∗ g
for p ∈ V , where ω p : TpV → V is the canonical identification given by ω p =

x−1 ◦ tx,p for any linear isomorphism x : V → Rn , as described in Lemma 2.84. Let
E = {e1 , ..., en } be the standard basis for Rn and B = {v1 , ..., vn } be a basis for V so
that x(vi ) = ei for i = 1, ..., n. Then B ∗ = {x1 , ..., xn } is the dual basis of B. If
g = ∑ g i j xi ⊗ x j ,
i, j
then
h = ∑ g i j dxi ⊗ dx j
i, j
since !
∂

h(p) , ∂ = g(vi, v j ) = g i j
∂ xi ∂ x j
p p
for p ∈ V . By Proposition 10.166 it follows that all the Christoffel symbols are zero,
so that the Levi–Civita connection on V is flat.
Now let f : V → R be the smooth function defined by
f (p) = g(p, p)
for p ∈ V . Then we have that

f = < P, P >
where
n
P = ∑ xi ∂ i
i=1
∂x
is the position vector field on V defined in Example 3.4, since
f (p) = g(p, p) = h(p)(ω p−1 (p), ω p−1 (p)) = h(p)(P(p), P(p))
for p ∈ V . By Corollary 10.114 and Proposition 10.166 we have that
∇X P = X
so that
<grad f , X > = X( f ) = X < P, P > = 2 < P, ∇X P > = 2 < P, X >
for every vector field X on V . This shows that
grad f = 2 P
and
<grad f , grad f > = 4 f .
If a > 0 and ε = ±1, it follows from Proposition 11.36 that Qa,ε = f −1 (ε a2 ) is
a closed pseudo-Riemannian hypersurface of V with sign ε and unit normal field
U = P ′/a where P ′ = P|Qa,ε . The shape tensor is given by
II(X,Y ) = − ε <X , Y > U / a
for vector fields X and Y on Qa,ε , since
<II(X,Y ), U > = < η 2 ◦ ∇X′ ( ei 1 ◦ Y ), U > = <∇X′ ( ei 1 ◦ Y ), P ′ >/ a

= − <ei 1 ◦ Y, ∇X′ P ′ >/ a = − <ei 1 ◦ Y, ei 1 ◦ X >/ a = − <X , Y >/ a
by 11.16 and Corollaries 11.19. Since V is flat, it follows from Corollary 11.33 that
the sectional curvature K of Qa,ε is given by
K(W ) = ε /a2
for every p ∈ Qa,ε and every non-degenerate tangent plane W at p when n ≥ 3.
11.38 Definition Let fk : Rn+1

k → R be the smooth function on the pseudo-
n+1
Euclidean space Rk defined by
fk (p) = g n+1
k (p, p)
for p ∈ Rn+1
k , and let a > 0. Then the set
Snk (a) = fk−1 (a2 ) = {p ∈ Rn+1

k | g n+1 2
k (p, p) = a }
is called the pseudosphere with radius a of dimension n and index k. It is a closed

pseudo-Riemannian hypersurface of Rn+1
k with sign 1. The set
−1
Hnk (a) = fk+1 (−a2 ) = {p ∈ Rn+1 n+1 2
k+1 | g k+1 (p, p) = −a }
is called the pseudohyperboloid with radius a of dimension n and index k. It is a

closed pseudo-Riemannian hypersurface of Rn+1 k+1 with sign −1. The sets S0 (a) and
n
H0 (a) will also be denoted by Sa and Ha , respectively.

n n n
11.39 Consider a vector space V of dimension n with a metric g, and let M be

a pseudo-Riemannian hypersurface in V with sign ε . Let Q1,ε be the hypersurface
in V defined in 11.37, and let ei : M → V and bi : Q1,ε → V be the inclusion maps.
By Lemma 2.84 we have an equivalence ω : TV → V × V over V obtained from the
canonical identification of the tangent space TpV with V for each p ∈ V . It is given
by ω = (idV × x−1 ) ◦ tx for any linear isomorphism x : V → Rn . If Z is a unit normal
field on an open subset U of M, we have a smooth map φ : U → Q1,ε , called the

Gauss map determined by Z, defined by
bi ◦ φ = pr2 ◦ ω ◦ Z
where pr2 : V × V → V is the projection on the second factor. Since

ωφ−1 −1
(p) ◦ ω p (Z p ) = ωφ (p) (φ (p)) = Pφ (p)
which is a unit normal to Q1,ε at φ (p) , it follows that ωφ−1(p) ◦ ω p : TpV → Tφ (p) V
induces a linear isomorphism τ p : Tp M → Tφ (p) Q1,ε so that
bi ∗ φ (p) ◦ τ p = ω −1 ◦ ω p ◦ ei ∗ p
φ (p)
for p ∈ U. Now let

n
v = ∑ a ∂ i i
i=1
∂x
p
be a vector in Tp M. Using that
n
Z = ∑ (x i ◦ φ ) ∂ i ,
i=1
∂x
it follows from Corollaries 11.19 and 10.114 that

n n
i
′ j ∂ (x ◦ φ ) ∂
−SZ (p)(v) = η 1 ( ∇v Z) = η 1 ∑ ∑ a (p) i
i=1
j
j=1 ∂x ∂x
p
= η 1 ◦ ω p−1 ◦ ωφ (p) ◦ ( bi ◦ φ ) ∗ p (v) = τ p−1 ◦ φ ∗ p (v),

which shows that
−SZ (p) = τ p−1 ◦ φ ∗ p
for p ∈ U.
11.40 Consider a vector space V of dimension n with a Riemannian metric g,

and let M be a hypersurface in V , and SZ be the shape operator determined by a unit
normal field Z on an open subset U of M. By 11.15 we know that SZ (p) is a self-
adjoint linear operator on Tp M for each point p ∈ U. Hence there is an orthonormal
basis {v1 , ... , vn−1 } in Tp M consisting of eigenvectors for SZ (p) with eigenvalues
λ 1 , ... , λ n−1 , respectively. By Corollary 11.33 it follows that the sectional curvature
of the tangent plane W to M at p, spanned by vi and v j where i 6= j, is given by
K(W ) = < II(p)(vi , vi ), II(p)(v j , v j ) > − < II(p)(vi , v j ), II(p)(vi , v j ) >
= < SZ (p)(vi ), vi > < SZ (p)(v j ), v j > − < SZ (p)(vi ), v j >2 = λ i λ j .
The eigenvalues λ 1 , ... , λ n−1 are called the principal curvatures at p. If they are
different, the corresponding eigenvectors define the principal directions at p. When
V = R3 with the standard metric g, then K(Tp M) is the Gaussian curvature at p
defined in Definition 10.181, in agreement with our Definition in 1.8.
THE ROBERTSON–WALKER SPACETIME

11.41 Let M be a Lorentz manifold with a metric g so that
g|V = − dt 2 + a 2 ( d χ 2 + r 2 ( d θ 2 + sin2 θ d φ 2 ))
for a local chart (ψ ,V ) on M with coordinate functions t, χ , θ and φ , where a = ã ◦ t
and r = r̃ ◦ χ for real functions ã and r̃. This is the general form of a metric in a homo-
geneous and isotropic universe model using natural units. The local chart is assumed
to be a comoving coordinate system so that each test particle has constant spatial
coordinates. The coordinate t, called the cosmic time , is the proper time for each test
particle. The partial derivatives ∂∂at and ∂∂χr are denoted by ȧ and r′ , respectively, and
are given by ȧ = ã ′ ◦ t and r′ = r̃ ′ ◦ χ . We see that
X1 = ∂ , X2 = 1 ∂ , X3 = 1 ∂
and X4 = 1 ∂
∂t a ∂χ ar ∂θ a r sin θ ∂ φ
form an orthonormal frame field on V with dual local basis
θe 1 = dt , θe 2 = a d χ , θe 3 = a r d θ and θe 4 = a r sin θ d φ .
Since
d θe 1 = 0
d θe 2 = aȧ θe 1∧ θe 2
′
d θe 3 = aȧ θe 1∧ θe 3 + ar r θe 2∧ θe 3
′
d θe 4 = aȧ θe 1∧ θe 4 + ar r θe 2∧ θe 4 + a1r cot θ θe 3∧ θe 4 ,
bi1i = − bii1 = aȧ for i = 2, 3, 4

′
bi2i = − bii2 = ar r for i = 3, 4
b434 = − b443 = a1r cot θ

and
a1ii = ai1i = 12 (b1ii − bii1 + bi1i) = ȧa for i = 2, 3, 4

′
a2ii = − ai2i = 21 (b2ii + bii2 − bi2i) = − ar r for i = 3, 4
a344 = − a434 = 21 (b344 + b443 − b434) = − a1r cot θ

This shows that
e 12 = ȧa θe 2 = ȧ d χ
e 21 = ω
ω
e 13 = ȧa θe 3 = ȧ r d θ
e 31 = ω
ω
e 14 = ȧa θe 4 = ȧ r sin θ d φ
e 41 = ω
ω
′
e 23 = − ar r θe 3 = − r′ d θ
e 32 = − ω
ω
′
e 24 = − ar r θe 4 = − r′ sin θ d φ
e 42 = − ω
ω
e 34 = − a1r cot θ θe 4 = − cos θ d φ

e 43 = − ω
ω
so that
e1=Ω
Ω e 2 = dω
e 21 + ω
e 31 ∧ ω
e 23 + ω
e 41 ∧ ω
e 24
2 1
= ä dt ∧ d χ = aä θe 1∧ θe 2
e1=Ω
Ω e 3 = dω
e 31 + ω
e 21 ∧ ω
e 32 + ω
e 41 ∧ ω
e 34
3 1
= ä r dt ∧ d θ + ȧ r′ d χ ∧ d θ − ȧ r′ d χ ∧ d θ = äa θe 1∧ θe 3
e1=Ω
Ω e 4 = dω
e 41 + ω
e 21 ∧ ω
e 42 + ω
e 31 ∧ ω
e 43
4 1
= ä r sin θ dt ∧ d φ + ȧ r′ sin θ d χ ∧ d φ + ȧ r cos θ d θ ∧ d φ

− ȧ r′ sin θ d χ ∧ d φ − ȧ r cos θ d θ ∧ d φ = äa θe 1∧ θe 4
e 23 = − Ω
Ω e 3 = dω
e 32 + ω
e 12 ∧ ω
e 31 + ω
e 42 ∧ ω e 34
2

ȧ2 r′′
′′ 2
= − r d χ ∧ d θ + ȧ r d χ ∧ d θ = 2
− 2 θe 2∧ θe 3
a ra
e2=−Ω
Ω e 4 = dω
e 42 + ω
e 12 ∧ ω
e 41 + ω
e 32 ∧ ω
e 43
4 2
= − r′′ sin θ d χ ∧ d φ − r′ cos θ d θ ∧ d φ + ȧ2 r sin θ d χ ∧ d φ

2 ′′
+ r′ cos θ d θ ∧ d φ = ȧ2 − r 2 θe 2∧ θe 4
a ra
e3=−Ω
Ω e 4 = dω
e 43 + ω
e 13 ∧ ω
e 41 + ω
e 23 ∧ ω
e 42
4 3
= sin θ d θ ∧ d φ + ȧ2 r2 sin θ d θ ∧ d φ − (r′ ) 2 sin θ d θ ∧ d φ

ȧ2 1 (r′ ) 2
= 2
+ 2 2 − 2 2 θe 3∧ θe 4
a r a r a
′ ′
ȧ2 ′′
= − r2 + 12 1
+ r
θe 3∧ θe 4
a2 ra a r2 r
′ 2 ′
ȧ2 ′′
= − r2 − 12 r 1
− r θe 3∧ θe 4
a 2 ra 2a r′ r2 r
Because of the isotropy we must have that

′ 2
1 r
2
− =k
r r
for a constant k, so that
r′ 2 = 1 − kr2. (1)
Introducing the local chart (ρ ,V ) on M with coordinate functions t, r, θ and φ , we
now have that

2 2 dr 2 2 2 2 2
g|V = − dt + a + r ( d θ + sin θ d φ )
1−kr2
which is the Robertson–Walker form of the metric. Rescaling r and a by the factors
|k|1/2 and |k|−1/2 , respectively, when k 6= 0, we may assume that k has the values
1, 0 or −1, and the solution of the differential Equation (1) with initial conditions
r̃(0) = 0 and r̃′ (0) ≥ 0 is given by

 sin χ for k = 1 , 0<χ <π
r= χ for k = 0 , 0<χ <∞ .

sinh χ for k = −1 , 0 < χ < ∞
The induced metric ge on the simultaneity space Vt0 = ψ −1 ({t0 } × R3 ) at the time
t = t0 is given by
ge = ae(t0 ) 2 ( d χ 2 + r 2 ( d θ 2 + sin2 θ d φ 2 ))
When k = 0 we see from Example 7.21 that this is the Euclidean metric on R3 after a
rescaling of the radial coordinate χ by the factor ae(t0 ). When k = 1 and k = −1 it is
the metric of the 3-sphere S3ã(t ) with radius ae(t0 ) in the Euclidean space R4 , and of
0
the upper sheet of the 3-hyperboloid H3ã(t ) in the Minkowski space R41 , respectively,
0
as we see from Examples 7.22 and 7.23. By 11.37 it follows that the sectional cur-
vature of Vt0 , called the spatial curvature at the time t0 , is K = k/ ae(t0 ) 2 . The three
cases k = 1, k = 0 and k = −1 with positive, zero and negative spatial curvatures are
called closed, flat and open universe models, respectively.
THE FRIEDMANN COSMOLOGICAL MODELS
11.42 The dynamics of the Robertson–Walker spacetime is determined by the

scale factor ae(t) which can be found by solving Einstein’s field equation. By Corol-
lary 10.106 the non-zero components of the curvature form R are
Ri11i = −Ri1i1 = R1i 1i = −R1i i1 = aä for i = 2, 3, 4

2
Ri jji = −Ri ji j = ȧ +k
2
for i 6= j ∈ {2, 3, 4}
a
Using the components of R with equal contravariant and second covariant index, we
find the non-zero components of the Ricci tensor
Ric 11 = − 3aä
2
Ric i i = äa + 2 ȧ +k
2
for i = 2, 3, 4
a
and the scalar curvature
2
S = −Ric 11 + Ric 22 + Ric 33 + Ric 44 = 6aä + 6 ȧ +k
2
a
From these we obtain the non-zero components of the Einstein tensor
2
G 11 = Ric 11 + 12 S = 3 ȧ +k
2a
2
G i i = Ric i i − 21 S = −2 äa − ȧ +k for i = 2, 3, 4
a2
We assume that the universe is a perfect fluid with energy-momentum tensor
given by
T = (ρ + p) u♭⊗ u♭ + p g ,
as described in Remark 10.187. Since the universe is homogeneous and isotropic,
the energy density ρ and the pressure p can only be functions of time, and the spatial
components of the flow vector field u must be zero. This means that ρ = ρ̃ ◦ t and
p = p̃ ◦ t for real functions ρ̃ and p̃ , and u = X1 so that u♭ = − θe 1 . Hence the non-
zero components of the energy-momentum tensor are
T11 = ρ and Ti i = p for i = 2, 3, 4.

From the Einstein field equation we now obtain the Friedmann equations
2 2
3 ȧ +k
2
= κ ρ + Λ and − 2 äa − ȧ +k
2
= κ p−Λ .
a a
Inserting the first Friedmann equation into the second we obtain

ä κ Λ
a = − 6 (ρ + 3 p) + 3 . (1)
Multipying the first Friedmann equation with a2/ 3 , taking the derivative with respect
to t and dividing by 2a ȧ yields
ä κ d 2 Λ
a = 6a da (ρ a ) + 3 . (2)
Now subtracting Equation (1) from (2) and multipying with 6a2/ κ , we have that
d
a da (ρ a2 ) + (ρ + 3 p) a2 = 0
which implies that

∂ 3 ∂ 3
∂ t (ρ a ) + p ∂ t (a ) = 0. (3)
Since a3 and ρ a3 are proportional to the volume V and the energy U of any fluid
element, this equation means that
dU + p dV = 0
which is the first law of thermodynamics for an adiabatic process. This is in agree-
ment with the homogeneity and isotropy of our universe model which does not allow
any temperature gradient or heat flow.
We now assume that the perfect fluid obeys the barotropic equation of state
p = wρ
where w is a constant. For dust, radiation and Lorentz invariant vacuum energy the
values of w are 0 , 1/ 3 and −1, respectively. By dust we mean a set of collisionless,
nonrelativistic particles with vanishing pressure, and by radiation we mean either
electromagnetic radiation or massive particles with relative velocities close to the
speed of light. Multiplying Equation (3) with (a3 ) w we have that
∂ 3 (w+1)
∂t { ρ a }=0,
which implies that

−3 (w+1)
ρ = ρ 0 aa (4)
0
for constants ρ 0 and a 0 . We see that ρ 0 is the energy density of the fluid at the time
t 0 when ae(t 0 ) = a 0 . We let t 0 be the present time, and ρ 0 and a 0 be the present values
of the energy density and the scale factor. If ρ m0 , ρ r0 and ρ v0 are the present energy
densities of non-interacting dust, radiation and Lorentz invariant vacuum energy, we
have that −3 −4
ρ = ρ m0 aa + ρ r0 aa + ρ v0 . (5)
0 0
−3
The energy density of dust falls off as a due to the decrease in number density of
particles as the universe expands. The energy density of radiation falls off faster as
a−4 because of the additional loss of energy by a factor a−1 due to the redshift of each
photon. This is also the case for massive relativistic particles which behave like a gas
of photons, as their rest energy can be considered to be negligible compaired to their
total energy. From the Friedmann equations we see that the cosmological constant
Λ can also be interpreted as Lorentz invariant vacuum energy, and can therefore be
removed from the equations by including the term Λ/κ in ρ v0 .
To find the time evolution of the scale factor a we now use the first Friedmann
equation with Λ = 0, which implies that
2
ȧ
a = κ ρ − k2 . (6)
3 a
If ρ satisfies Equation (4) with w > −1/ 3 , the fate of the universe is determined by k.
When k = 0 or k = −1, it will expand forever, but when k = 1 it will stop expanding
and recollapse to a Big Crunch. If w < −1/ 3 , the universe will expand forever for
all values of k.
Using the metric g we see that the instantaneous physical distance l from an
observer at χ = 0 to a galaxy with radial coordinate χ = χ0 is given by l = a χ0 . The
velocity of the galaxy relative to the observer due to the expansion or contraction of
the universe is
dl ȧ
dt = a a χ0 = H l, (7)
where H = ȧ/a is called the Hubble parameter . Its present value, called the Hubble
constant , is approximately H0 = 20 (km/s)/ Mly. Equation (7) is called Hubble’s
law , stating that the velocity of a galaxy is proportional to its distance. The time
tH = 1/H is called the Hubble age of the universe. It coincides with the age t of
the universe if ȧ is constant, which means that the velocity of each fluid particle
comoving with the expansion of the universe relative to an observer at χ = 0 is
constant.
The energy density ρc in a flat universe model, called the critical energy density ,
is given by
H2 = κ ρc
3
which is obtained from Equation (6) with k = 0. Its present value is approximately
ρ c 0 = 2 · 10−26 kg/ m3 . Introducing the relative energy density Ω = ρ /ρ c of a uni-
verse model, we see from Equation (6) that its spatial curvature is
k
= H 2 (Ω − 1) .
a2
Hence the universe model is closed if Ω > 1 , flat if Ω = 1 and open if Ω < 1 . The
curvature radius is r
1 k
a=
H Ω−1
when Ω 6= 1 . We denote the present relative energy densities of non-interacting dust,
radiation and vacuum by Ω m0 , Ω r 0 and Ω v0 , respectively.
11.43 Universe models with dust and radiation We introduce a new time co-
ordinate η , called the parametric time , so that η = η̃ ◦ t for a real function η̃ with
η̃ ′ = 1/ ã and η̃ (0) = 0. Then we have that
∂a
∂ η = a ȧ .
In a universe model with dust and radiation it follows from (5) and (6) in 11.42 that
2
∂a
∂η = κ a30 ( ρ m0 a + ρ r 0 a 0 ) − k a2 ,
3
which can be written as

∂a 2
∂η = 2 α a + β 2 − k a2 (1)
where
1/2
α = κ ρ m0 a30 = 1 Ω m0 H 02 a30 and β = κ 1/2
ρ r0 a20 = Ω r 0 H 0 a20 .
6 2 3
When k 6= 0, Equation (1) can also be written as
2
∂a α 2
∂η = γ − k a − (2)
k
2
where γ = α + β 2 . Now let ε = |γ | when γ 6= 0 , and let ε = 1 when γ = 0 . Intro-
k
ducing the function b = ε −1/2 a − α , we have that
k

∂b 2
∂η = σ − k b2 (3)
where σ = γ / ε . When k = 1, the solution of this differential equation is

b = sin(η + c)
so that
a = α + ε 1/2 sin(η + c) ,
where c is an integration constant. The initial conditions ae(0) = 0 and ae′ (0) ≥ 0
imply that
sin c = − ε −1/2 α and cos c = ε −1/2 β ,
which give the solution
a = α (1 − cos η ) + β sin η
t = α (η − sin η ) + β (1 − cos η )
When k = −1, the solution of the differential Equation (3) is

b = 1 eη +c − σ e−(η +c)
2
so that
n o
1 c
a = −α + ε 1/2 e − σ e−c cosh η + 1 ec + σ e−c sinh η ,
2 2
where c is an integration constant. The initial conditions ae(0) = 0 and ae′ (0) ≥ 0
imply that
1 c 1 c
e − σ e−c = ε −1/2 α and e + σ e−c = ε −1/2 β ,
2 2
which give the solution
a = α (cosh η − 1) + β sinh η
t = α (sinh η − η ) + β (cosh η − 1)
When k = 0, the solution of Equation (1) with initial condition ae(0) = 0 is
a = 1 α η2 + β η
2
t = 1 α η3 + 1 β η2
6 2
11.44 Vacuum dominated universe models We now consider universe models

with Lorentz invariant vacuum energy corresponding to a cosmological constant Λ >
0. It follows from (5) and (6) in 11.42 that
ȧ2 = κ ρ v0 a2 − k ,
3
ȧ2 = ω 2 a2 − k (1)
where 1/2
κ 1/2
ω= ρ v0 = Ω v0 H 0 = (Λ/3)1/2 = 1/Rd . (2)
3
A solution in this case is
 1

 cosh ω t for k = 1 , −∞ < t < ∞
 ω
1 ωt
a= ω e for k = 0 , −∞ < t < ∞ .


 1
ω sinh ω t for k = −1 , 0<t <∞
Note that the initial condition ae(0) = 0 can only be applied when k = −1. When
k = 0 and k = 1 the universe is infinitely old without a Big Bang singularity.
Using the comoving coordinate system (ψ ,V ) described in 11.41 with coordinate

functions t, χ , θ and φ we have that
g|V = − dt 2 + a 2 ( d χ 2 + r 2 ( d θ 2 + sin2 θ d φ 2 )), (3)
where r is the function on V given by

 sin χ for k = 1 , 0<χ <π
r= χ for k = 0 , 0<χ <∞

sinh χ for k = −1 , 0 < χ < ∞
satisfying the relation
r′ 2 = 1 − kr2 . (4)
For each k we are going to find an isometric embedding α : M → R51 of M as
either the entire or a part of the 4-pseudosphere
S41 (Rd ) = {(v, w, x, y, z) ∈ R5 | − v2 + w2 + x2 + y2 + z2 = R2d }
with radius Rd in the pseudo-Euclidean space R51 . This also shows that the spacetime
M actually is the same as either the entire or a part of the extended de Sitter spacetime
described in 10.189 and 10.194.
We first consider the cases where k 6= 0 . Using (1), (4) and the relations ä = ω 2 a
and r′′ = −kr we have that
−dt 2 + a 2 d χ 2 = 1 [ ȧ2 (r′ 2 + kr2 ) − ω 2a2 ] dt 2 + a2(r′ 2 + kr2 ) d χ 2

k
= 1 (ȧ r′ dt − k a r d χ ) 2 − 1 ω 2 a2 dt 2 + (ȧ rdt + a r′ d χ ) 2

k k
= 1 2 [ d(ω ar′ ) 2 − d ȧ 2 ] + d(ar) 2

kω
and
1
[ (ω ar′ ) 2 − ȧ 2 ] + (ar) 2
k ω2
= 1 2 [ ω 2 a2 (1 − kr2 ) − (ω 2 a2 − k) ] + (ar) 2 = 12 = R2d .

kω ω
Hence there is an isometric embedding α : M → R51 given by

ȧ/ ω when k = 1
v ◦ α |V =
ar′ when k = −1
′
ar when k = 1
w ◦ α |V =
ȧ/ ω when k = −1 (5)
x ◦ α |V = ar sin θ cos φ
y ◦ α |V = ar sin θ sin φ
z ◦ α |V = ar cos θ
on the coordinate domain V .
When k = 1 we have that Ω v0 > 1 so that Rd = (3/Λ)1/2 < 1/H 0 according to

(2). The isometric embedding α : M → R51 is given by
α (t, χ , u) = Rd ( sinh(t/Rd ), cosh(t/Rd ) cos χ , cosh(t/Rd ) sin χ u) (6)
for (t, χ , u) ∈ R × [ 0, π ] × S2, mapping M onto the entire pseudosphere S41 (Rd ). On
the coordinate domain V we have that
v ◦ α |V = Rd sinh(t/Rd )
w ◦ α |V = Rd cosh(t/Rd ) cos χ
x ◦ α |V = Rd cosh(t/Rd ) sin χ sin θ cos φ
y ◦ α |V = Rd cosh(t/Rd ) sin χ sin θ sin φ
z ◦ α |V = Rd cosh(t/Rd ) sin χ cos θ
where −∞ < t < ∞ , 0 < χ < π , 0 < θ < π and −π < φ < π . We can also verify
directly that α is an isometry in this case. Using Proposition 4.18 we have that
α ∗ dv|V = cosh(t/Rd ) dt
α ∗ dw|V = sinh(t/Rd ) cos χ dt − Rd cosh(t/Rd ) sin χ d χ
α ∗ dx|V = sinh(t/Rd ) sin χ sin θ cos φ dt + Rd cosh(t/Rd ) cos χ sin θ cos φ d χ
+ Rd cosh(t/Rd ) sin χ cos θ cos φ d θ − Rd cosh(t/Rd ) sin χ sin θ sin φ d φ
α ∗ dy|V = sinh(t/Rd ) sin χ sin θ sin φ dt + Rd cosh(t/Rd ) cos χ sin θ sin φ d χ
+ Rd cosh(t/Rd ) sin χ cos θ sin φ d θ + Rd cosh(t/Rd ) sin χ sin θ cos φ d φ
α ∗ dz|V = sinh(t/Rd ) sin χ cos θ dt + Rd cosh(t/Rd ) cos χ cos θ d χ
− Rd cosh(t/Rd ) sin χ sin θ d θ .
The Lorentz metric on R51 is given by
g 51 = − dv2 + dw2 + dx2 + dy2 + dz2
in cartesian coordinates. Hence we have that
α ∗ g 51 |V = − dt 2 + R2d cosh2 (t/Rd ) [ d χ 2 + sin2 χ (d θ 2 + sin2 θ d φ 2 ) ] .
The simultaneity space Vt0 at the time t = t0 is the intersection between the pseudo-
sphere S41 (Rd ) and the hyperplane v = Rd sinh(t0 /Rd ). This intersection is a 3-sphere
S3a 0 with radius a 0 = Rd cosh(t0 /Rd ) in the Euclidean space R4 as described in 11.41
and Example 7.22. The radius initially shrinks until it reaches a minimal value Rd at
t0 = 0, and then it increases, describing a “bouncing universe” without a Big Bang.
When k = −1 we have that Ω v0 < 1 so that Rd = (3/Λ)1/2 > 1/H 0 according to
α (t, χ , u) = Rd ( sinh(t/Rd ) cosh χ , cosh(t/Rd ), sinh(t/Rd ) sinh χ u) (7)

for (t, χ , u) ∈ (0, ∞) × [ 0, ∞) × S2, mapping M onto the open submanifold
{(v, w, x, y, z) ∈ S41 (Rd ) | v > 0, w > Rd }
of the pseudosphere. On the coordinate domain V we have that
v ◦ α |V = Rd sinh(t/Rd ) cosh χ
w ◦ α |V = Rd cosh(t/Rd )
x ◦ α |V = Rd sinh(t/Rd ) sinh χ sin θ cos φ
y ◦ α |V = Rd sinh(t/Rd ) sinh χ sin θ sin φ
z ◦ α |V = Rd sinh(t/Rd ) sinh χ cos θ
where 0 < t < ∞ , 0 < χ < ∞ , 0 < θ < π and −π < φ < π . In this case we can also
verify directly that α is an isometry since
α ∗ dv|V = cosh(t/Rd ) cosh χ dt + Rd sinh(t/Rd ) sinh χ d χ

α ∗ dw|V = sinh(t/Rd ) dt
α ∗ dx|V = cosh(t/Rd ) sinh χ sin θ cos φ dt + Rd sinh(t/Rd ) cosh χ sin θ cos φ d χ
+ Rd sinh(t/Rd ) sinh χ cos θ cos φ d θ − Rd sinh(t/Rd ) sinh χ sin θ sin φ d φ
α ∗ dy|V = cosh(t/Rd ) sinh χ sin θ sin φ dt + Rd sinh(t/Rd ) cosh χ sin θ sin φ d χ
+ Rd sinh(t/Rd ) sinh χ cos θ sin φ d θ + Rd sinh(t/Rd ) sinh χ sin θ cos φ d φ
α ∗ dz|V = cosh(t/Rd ) sinh χ cos θ dt + Rd sinh(t/Rd ) cosh χ cos θ d χ
− Rd sinh(t/Rd ) sinh χ sin θ d θ .
Hence we have that
α ∗ g 51 |V = − dt 2 + R2d sinh2 (t/Rd ) [ d χ 2 + sinh2 χ (d θ 2 + sin2 θ d φ 2 ) ] .
sphere S41 (Rd ) and the hyperplane w = Rd cosh(t0 /Rd ). This intersection is the upper
sheet of the 3-hyperboloid H3a 0 with radius a 0 = Rd sinh(t0 /Rd ) in the Minkowski
space R41 as described in 11.41 and Example 7.23.
We finally consider the case where k = 0 . Using the relations ȧ = ω a and r′ = 1
we have that
−dt 2 + a 2 d χ 2
= ȧ dt (−ȧ−2 · ȧ dt − ȧ r2 dt − 2 a r d χ ) + ( ȧ2 r2 dt 2 + 2 a ȧ a r dt d χ + a2 d χ 2 )
= 12 d ȧ d( ȧ −1 − ȧr2 ) + ( ȧ r dt + a d χ ) 2
ω
= 1 2 {−d [ ȧ − ( ȧ −1 − ȧr2 ) ] 2 + d [ ȧ + ( ȧ −1 − ȧr2 ) ] 2 } + d(a r) 2

4ω
and
1
{−[ ȧ − ( ȧ −1 − ȧr2 ) ] 2 + [ ȧ + ( ȧ −1 − ȧr2 ) ] 2 } + (a r) 2
4 ω2
= 12 ȧ ( ȧ −1 − ȧr2 ) + (a r) 2 = 12 = R2d .
ω ω
Hence there is an isometric embedding α : M → R51 given by
v ◦ α |V = 1 [ ȧ − ( ȧ −1 − ȧr2 ) ]
2ω
w ◦ α |V = 1 [ ȧ + ( ȧ −1 − ȧr2 ) ]
2ω
(8)
x ◦ α |V = ar sin θ cos φ
y ◦ α |V = ar sin θ sin φ
z ◦ α |V = ar cos θ
(v + w) ◦ α |V = ȧ/ ω
(v − w) ◦ α |V = ( ȧr2 − ȧ −1)/ ω
on the coordinate domain V .

When k = 0 we have that Ω v0 = 1 so that Rd = (3/Λ)1/2 = 1/H 0 according to
α (t, χ , u) = Rd ( sinh(t/Rd ) + 21 χ 2 et/Rd , cosh(t/Rd ) − 12 χ 2 et/Rd , et/Rd χ u) (9)
for (t, χ , u) ∈ R × [ 0, ∞) × S2, mapping M onto the open submanifold
{(v, w, x, y, z) ∈ S41 (Rd ) | v + w > 0}
of the pseudosphere. On the coordinate domain V we have that
v ◦ α |V = Rd sinh(t/Rd ) + 12 Rd χ 2 et/Rd
w ◦ α |V = Rd cosh(t/Rd ) − 12 Rd χ 2 et/Rd
x ◦ α |V = Rd et/Rd χ sin θ cos φ
y ◦ α |V = Rd et/Rd χ sin θ sin φ
z ◦ α |V = Rd et/Rd χ cos θ
so that
(v + w) ◦ α |V = Rd et/Rd
(v − w) ◦ α |V = Rd χ 2 et/Rd − Rd e−t/Rd
where −∞ < t < ∞ , 0 < χ < ∞ , 0 < θ < π and −π < φ < π . We can also verify
directly that α is an isometry. Using Proposition 4.18 we see that
α ∗ dv|V = [ cosh(t/Rd ) + 21 χ 2 et/Rd ] dt + Rd χ et/Rd d χ

α ∗ dw|V = [ sinh(t/Rd ) − 12 χ 2 et/Rd ] dt − Rd χ et/Rd d χ
α ∗ dx|V = et/Rd χ sin θ cos φ dt + Rd et/Rd sin θ cos φ d χ
+ Rd et/Rd χ cos θ cos φ d θ − Rd et/Rd χ sin θ sin φ d φ
α ∗ dy|V = et/Rd χ sin θ sin φ dt + Rd et/Rd sin θ sin φ d χ

+ Rd et/Rd χ cos θ sin φ d θ + Rd et/Rd χ sin θ cos φ d φ
α ∗ dz|V = et/Rd χ cos θ dt + Rd et/Rd cos θ d χ − Rd et/Rd χ sin θ d θ .
Hence we have that
α ∗ g 51 |V = − dt 2 + R2d e2t/Rd [ d χ 2 + χ 2 (d θ 2 + sin2 θ d φ 2 ) ] .
sphere S41 (Rd ) and the hyperplane v + w = Rd et0 /Rd .
11.45 Flat universe models with vacuum energy and either dust or radiation
We now consider flat universe models with energy density
−s
ρ = ρ v0 + (ρ c 0 − ρ v0 ) aa ,
0
consisting of Lorentz invariant vacuum energy, and of dust when s = 3 or radiation

when s = 4. The constants ρ c 0 , ρ v0 and a 0 are the present values of the critical energy
density, the vacuum energy density and the scale factor. We see that the present value
of the total energy density is ρ c 0 in agreement with the assumption that the universe
is flat. As usual we denote the present relative energy density of vacuum by Ω v0 .
Now it follows from (6) in 11.42 that
as−2 ȧ2 = κ { ρ v0 as + (ρ c 0 − ρ v0) as0 } ,

3
2
s
2 as−2 ȧ2 = ω 2 (as + As) (1)
where
1/2
κ
ω = 2s = s Ω v0 H 0
1/2
ρ v0 and
3 2
1/s 1/s
ρ c 0 − ρ v0 1 − Ω v0
A= ρ v0 a0 = a0 .
Ω v0
s/2
Introducing the function b = a , Equation (1) takes the form
A
ḃ 2 = ω 2 (b2 + 1) (2)
having the solution

b = sinh( ω t + c) ,
where c is an integration constant. Hence the solution of Equation (1) with initial
condition ae(0) = 0 is
a = A sinh 2/s ( ω t) .
The Hubble parameter and the critical energy density are
H = aȧ = 2s ω coth( ω t)
and
ρ c = κ3 H 2 = ρ v0 coth2 ( ω t) ,
and the relative energy density of vacuum is therefore
Ω v = tanh2 ( ω t) .
The ratio of the age t of the universe to the Hubble age tH is given by
√
2 arctanh Ωv
t/tH = H t = 2s ω t coth( ω t) = √ .
s Ωv
At the present time t = t 0 we obtain

p a s/2
0
ω t 0 = arctanh Ω v0 = arcsinh .
A
Using the observed value Ω v0 = 0.7 and s = 3 , we therefore have that
A/a 0 = 0.75 and ω t 0 = 1.2 .
Hence
a/a 0 = 0.75 sinh 2/3 (1.2 t/t 0 )
is a realistic expression for the expansion of the universe, and H 0 t 0 is approximately
equal to 1. This means that the present age t 0 of the universe is approximately equal
to its Hubble age tH 0 as a result of a period of decelerated expansion followed by a
period of accelerated expansion.
Chapter 12
JET BUNDLES
BUNDLES
12.1 Definition Let M and F be smooth manifolds. A smooth manifold E is
called a bundle over M with fibre F if the following two conditions are satisfied:
(i) There is a surjective smooth map π : E → M which is called the projection
of the total space E onto the base space M. For each point p ∈ M the inverse
image π −1 (p) is called the fibre over p.
(ii) For each point p ∈ M there is an open neighbourhood U around p and a diffeo-
morphism t : π −1 (U) → U × F such that the diagram
t ✲
π −1 (U) U ×F
❅
πU ❅ pr1
❘
❅ ✠
U
is commutative. The pair (t, π −1(U)) is called a local trivialization around p.
bundle in order to indicate more of the bundle structure in the notation. F is also
called the typical fibre to distinguish it from the fibre π −1(p) over p.
A family of local trivializations {(tα , π −1 (Uα ))|α ∈ A}, where {Uα |α ∈ A} is an
open cover of M, is called a trivializing cover of E.
12.2 Proposition If π : E → M is a bundle with fibre F and p is a point in M,

then there is a local trivialization (t, π −1 (U)) around p where U is the coordinate
neighbourhood of a local chart (x,U) on M.
525
PROOF : If (t1 , π −1 (U1 )) is a local trivialization and (x2 ,U2 ) a local chart around p,
let U = U1 ∩ U2 and x = x2 |U , and let t : π −1 (U) → U × F be the diffeomorphism
induced by t1 .
12.3 Proposition The projection π is a submersion, and the fibre E p = π −1 (p)

over p is a closed submanifold of E for each point p ∈ M. If (t, π −1 (U)) is a local
trivialization around p, then the map t p : E p → F defined by t p = pr2 ◦ t|E p is a
diffeomorphism. Hence each fibre E p is diffeomorphic to the typical fibre F.
If i p : E p → E is the inclusion map and u ∈ E p , then Vu = i p ∗ (Tu E p ) is a sub-
space of Tu E which is called the vertical subspace at u. The vectors in Vu are called
vertical tangent vectors at u. We have that Vu = ker π∗ u for each u ∈ E, and these
subspaces form a distribution V on E . The subbundle π ′ : V → E of π ′′ : T E → E is
called the vertical tangent bundle of π : E → M .
Suppose that (t, π −1 (U)) is a local trivialization around a point u0 in E with
t(u 0 ) = (p 0 , v 0 ) , and that (x, O) and (y, Q) are local charts on M n and F m around
p 0 and v 0 , respectively, with O ⊂ U . Then (z,W ) , where W = t −1 (O × Q) and z =
(x × y) ◦ t , is a local chart around u 0 on E , and
n o
∂ ∂
n+1 , ... , n+m
∂z
∂z
is a local basis for V on W .
Let pr2 : Rn × Rm → Rm be the projection on the second factor. Then we have a
local trivialization (t ′ , π ′ −1 (W )) in the vertical tangent bundle π ′ : V → E , where
t ′ = (idW × pr2 ) ◦ tz |π ′ −1 (W )
so that !
m
t ′
∑ ak ∂ z∂n+k = (u, a)
k=1 u
for u ∈ W and a ∈ Rm .
PROOF : From the diagram in Definition 12.1 (ii) we see that π is a submersion.
Hence E p is a closed submanifold of E by Proposition 2.41. In the same way we also
have that {p} × F = pr1 −1 (p) is a closed submanifold of U × F.
By Corollary 2.38, the map s p : E p → {p} × F induced by t is therefore a diffeo-
morphism, and so is the map pr2 : {p} × F → F, thus showing that t p = pr2 ◦ s p is a
diffeomorphism.
Since π ◦ i p (u) = p for every u ∈ E p , we have that π∗ u ◦ i p ∗ u = (π ◦ i p )∗ u = 0
which implies that Vu ⊂ ker π∗ u . Equality now follows since
dim u (E p ) = dim t p (u) (F) = dim u (E) − dim p (M) = dim u (ker π∗ u ) .
To prove the last assertions in the proposition, let j p : Q → O × Q be the embed-
ding defined by j p (v) = (p, v) for p ∈ O and v ∈ Q . Then we have that
! !

t∗ ∂ = jp ∗ ∂
∂ zn+k ∂ yk
u v
JET BUNDLES 527
for k = 1, ... , m , where t(u) = (p, v) . Indeed, we have that

!
∂
∂ ( f ◦ t)
t∗ (f) = (u) = Dn+k ( f ◦ t ◦ z−1 )(z(u))
∂ zn+k ∂ zn+k
u
= Dn+k ( f ◦ (x × y)−1 )(x(p), y(v)) = Dk ( f ◦ j p ◦ y−1 )(y(v))

!
∂ ( f ◦ jp) ∂

= (v) = j p ∗ (f)
∂ yk ∂ yk
v
for every smooth function f defined on an open neighbourhood of (p, v) in O × Q ,

completing the proof of the last part of the proposition.
12.4 Remark The coordinate functions zi for i = 1, 2, ... , n and zn+α for α =
1, 2, ... , m on E defined in the proposition are also denoted by xi and yα . It will be
clear from the context whether xi and yα denote coordinate functions on E or on M
and F, respectively.
12.5 Remark Let π : T M → M be the tangent bundle of a smooth manifold

M n . For each local chart (x,U) on M we obtain an induced local chart (z,W ) on T M
where W = π −1 (U) and z = (x × id) ◦ tx , so that
!
n
∂

z ∑ ai ∂ xi
= (x(u), a)
i=1 u
for u ∈ U and a ∈ Rn . The coordinate functions zi and zn+i for i = 1, 2, ... , n on T M

are also denoted by xi and ẋi . It will be clear from the context whether xi denote
coordinate functions on T M or on M.
12.6 Proposition Let M and F be smooth manifolds, and let π : E → M be a map

from a set E. Let {Uα |α ∈ A} be an open cover of M. Suppose that for each α ∈ A,
there is a bijection tα : π −1 (Uα ) → Uα × F such that the diagram
tα ✲
π −1 (Uα ) Uα × F
❅
π Uα ❅ pr1
❘
❅ ✠
Uα
is commutative. Moreover, suppose that for each pair α , β ∈ A, the map tβ ◦ tα−1 is a
diffeomorphism from (Uα ∩Uβ ) × F onto itself.
Then there is a unique topology and smooth structure on E such that π : E → M

is a bundle over M with fibre F, and such that (tα , π −1 (Uα )) is a local trivialization
for each α ∈ A.
PROOF : We first define a topology τ on E so that the projection π is continuous and
the bijections tα : π −1 (Uα ) → Uα × F are homeomorphisms for every α ∈ A. The
continuity of π implies that each π −1 (Uα ) must be open in τ , and hence a set O in
E must belong to τ if and only if O ∩ π −1 (Uα ) ∈ τ for every α ∈ A. The requirement
of sets O in E such that tα (O ∩ π −1 (Uα )) is open in Uα × F for every α ∈ A. This
requirements.
that tβ (π −1 (Uα ) ∩ π −1 (Uβ )) = (Uα ∩ Uβ ) × F which is open in Uβ × F, thereby
belongs to τ , we have that tα (O ∩ π −1 (Uα )) is open in Uα × F, which proves that
tα is an open map. To show that it is continuous, let O be an open set in Uα × F. For
every β ∈ A we have that tβ (tα−1 (O) ∩ π −1 (Uβ )) = (tβ ◦ tα−1 ) (O ∩ (Uα ∩ Uβ ) × F)
which is open in Uβ × F, and hence tα−1 (O) ∈ τ showing that tα is continuous.
is the composition of the continuos maps tα and pr1 in the above commutative dia-
gram for each α ∈ A.
Now let {(y γ ,Wγ )|γ ∈C} be an atlas on F, and let {(xβ ,Vβ )|β ∈ B} be an atlas
on M such that {Vβ |β ∈ B} is a refinement of the open cover {Uα |α ∈ A}. We then
have a map ρ : B → A such that Vβ ⊂ Uρ (β ) for every β ∈ B. By the assumptions
in the proposition we now see that the family B = {((xβ × y γ ) ◦ tρ (β ) , tρ−1 (β )
(Vβ ×
Wγ ))|(β , γ ) ∈ B ×C } is an atlas on E. To show that E is a smooth manifold, we must
show that it is Hausdorff and second countable. By Proposition 2.6 we know that M
and F have atlases of the above form where B and C are countable, thus showing that
E is second countable.
To see that the E is Hausdorff, let p1 and p2 be two different points in E. If p1
and p2 lie in different fibres, they can be separated by the neighbourhoods π −1 (W1 )
and π −1 (W2 ), where W1 and W2 are disjoint neighbourhoods of π (p1 ) and π (p2 ) in
M. If p1 and p2 lie in the same fibre π −1 (q) with q∈Uα , they can be separated by the
of tα (p1 ) and tα (p2 ) in Uα × F.
the bijections tα : π −1 (Uα ) → Uα × F are diffeomorphisms for every α ∈ A. We see
maps tα and pr1 . This completes the construction of the bundle π : E → M over M
with fibre F, having (tα , π −1 (Uα )) as a local trivialization for each α ∈ A.
12.7 Definition Let π1 : E1 → M1 and π2 : E2 → M2 be two bundles. A bundle

map from π1 to π2 is a pair ( f˜, f ) of smooth maps f˜ : E1 → E2 and f : M1 → M2
JET BUNDLES 529
such that the diagram
f˜ ✲
E1 E2
π1 π2
❄ ❄
f✲
M1 M2
is commutative. We let f˜p denote the induced smooth map f˜p : π1−1(p) → π2−1 ( f (p))
between the fibres for each p ∈ M1 . If f˜ and f are diffeomorphisms, ( f˜, f ) is called
an equivalence .
12.8 Definition Let π1 : E1 → M and π2 : E2 → M be two bundles over the same

base space M. A bundle map from π1 to π2 over M is a smooth map f : E1 → E2 such
that ( f , idM ) is a bundle map. If f is a diffeomorphism, it is called an equivalence
over M.
12.9 Definition A bundle π : E → M with fibre F is said to be trivial if it is

equivalent over M with the bundle pr1 : M × F → M , where pr1 is the projection on
the first factor.
12.10 Definition Let π : E → M be a bundle over M with fibre F, and let V be

an open subset of M and i : V → M be the inclusion mapping. Then a smooth map
s : V → E with π ◦ s = i is called a section of π on V , or a lifting of i to E. The
set of all sections of π on V is denoted by Γ(V ; E) or ΓV (E). Given a point p on M,
we let Γ p (E) denote the set of all sections s : V → E , each defined on some open
neighbourhood V of p.
pr1 ◦ t ◦ s = idU . Hence s|U is completely determined by the map h : U → F given
by (t ◦ s)(p) = (p, h(p)) for p ∈ U. h is called the local representation of s on U. We
12.11 Let π : E → M be a bundle over M with fibre F. If f : N → M is a smooth

map from a smooth manifold N, we define the set
Ee = {(p, u) ∈ N × E| f (p) = π (u)}
and let πe : Ee → N and fe : Ee → E be the restrictions to Ee of the projections pr1 :

N × E → N and pr2 : N × E → E on the first and second factor. Then we have a
commutative diagram
fe✲
Ee E
πe π
❄ ❄
f✲
N M
since f ◦ πe (p, u) = f (p) = π (u) = π ◦ fe(p, u) for (p, u) ∈ E.

e
−1
Now let A = {(tα , π (Uα ))|α ∈ A} be a trivializing cover of E. For each α ∈ A
we have an open subset U eα = f −1 (Uα ) of N and a bijection e eα ) → U
tα : πe−1 (U eα × F
defined by
etα (p, u) = (p,tα , f (p) (u)) (1)
for p ∈ Ueα and u ∈ E f (p) . Then {U
eα |α ∈ A} is a open cover of N, and we will show
that the map etβ ◦ e−1
tα is a diffeomorphism from (U eα ∩ U
eβ ) × F onto itself for each
pair of indices α , β ∈ A . Let φ β α : (Uα ∩Uβ ) × F → F be the second component of
the smooth map tβ ◦ tα−1 so that
tβ ◦ tα−1 (q, v) = (q, φ β α (q, v))
for q ∈ Uα ∩Uβ and v ∈ F. Then we have that
e tα−1
tβ ,p ◦ e −1
,p (v) = tβ , f (p) ◦ tα , f (p) (v) = φ β α ( f (p), v) (2)
so that
e tα−1 (p, v) = (p, φ β α ◦ ( f × idF ) (p, v))
tβ ◦ e
for every p ∈ U eα ∩ U eβ and v ∈ F. Since this is also true with the indices α and β
interchanged, we have shown the above assertion for e tα−1 . By Proposition 12.6
tβ ◦ e
there is therefore a unique topology and smooth structure on Ee such that πe : Ee → N
is a bundle over N with fibre F, and such that (e eα )) is a local trivialization
tα , πe−1 (U
e
for each α ∈ A. It is called the local trivialization in E associated with the local
trivialization (tα , π −1 (Uα )) in E.
By (1) we see that
tα ,p = tα , f (p) ◦ fep
e
eα , so that
for p ∈ U
tα ◦ fe ◦ e
tα−1 (p, v) = ( f (p), v) (3)
eα and v ∈ F. Hence it follows from Proposition 12.3 that ( fe, f ) is a bundle
for p ∈ U
map which induces a diffeomorphism between the fibres Eep and E f (p) for each p ∈ N.
Since πe and fe are the components of the inclusion map i : Ee → N × E, we also
have that
tα−1 (p, v) = (p, f (p), v)
(id × tα ) ◦ i ◦ e
JET BUNDLES 531
for p ∈ Ueα and v ∈ F. Using this, we will show that Ee is a closed submanifold of
eα ) and tα (u0 ) = ( f (p0 ), v0 ) ∈ Uα × F, we choose local
N × E. If (p0 , u0 ) ∈ πe−1 (U
charts (e e e ⊂U
x, V ), (x,V ) and (y,W ) around p0 , f (p0 ) and v0 , respectively, with V eα
e
and f (V ) ⊂ V . Then we have that
(e
x × (x × y) ◦ tα )) ◦ i ◦ ((e tα ) −1 (a, ξ ) = (a, x ◦ f ◦ xe−1 (a), ξ )
x × y) ◦ e
for a ∈ xe(Ve ) and ξ ∈ y(W ).

Now let r, n and m be the dimensions of N, M and F, respectively, and let φ :
xe(Ve ) × Rn × Rm → xe(Ve ) × Rm × Rn be the C∞ -map defined by φ (a, b, ξ ) = (a, ξ , b −
x ◦ f ◦ xe−1 (a)). φ is a diffeomorphism with inverse ψ : xe(Ve ) × Rm × Rn → xe(Ve ) ×
Rn × Rm given by ψ (a, ξ , b) = (a, b + x ◦ f ◦ xe−1 (a), ξ ).
If j : xe(Ve ) × Rn × Rm → Rr × Rn × Rm is the inclusion map, we therefore have
a local chart (Ve × tα−1 (V × W ), j ◦ φ ◦ (e
x × (x × y) ◦ tα )) around (p0 , u0 ) in N × E
having the submanifold property, showing that Ee is a submanifold of N × E. It is
closed as Ee = ( f × π )−1 (∆), where ∆ ⊂ M × M is the diagonal which is closed since
M is Hausdorff.
The bundle πe : Ee → N is called the pull-back of π : E → M by f or the bundle
induced from π : E → M by f , and it is denoted by f ∗ (E) with projection f ∗ (π ).
A smooth map F : N → E with π ◦ F = f is called a lifting of f . The set of all
liftings of f is denoted by Γ( f ; E) . For each lifting F of f there is a unique section
Fe : N → Ee of πe with fe◦ Fe = F , which is given by F(p) e = (p, F(p)) for p ∈ N.
Conversely, if s : N → Ee is a section of πe , then s = Fe where F = fe◦ s . Fe is called
the section in the induced bundle determined by the lifting F.
If N is an immersed submanifold of M with inclusion map i : N → M, we consider
the subset E|N = π −1(N) of E, and let πN : E|N → N be the map induced by the
projection π : E → M. We give E|N the topology and manifold structure so that the
bijection e : Ee → E|N given by e(p, u) = u for (p, u) ∈ E, e is an equivalence over N.
The bundle πN : E|N → N is called the restriction of E to N, and i ′ = ei ◦ e−1 : E|N → E
is the inclusion map. Given a trivializing cover A = {(tα , π −1 (Uα ))|α ∈ A} of E, we
let Ueα = Uα ∩ N and tα′ : π −1 (U
eα ) → Ueα × F be the map induced by tα : π −1 (Uα ) →
Uα × F so that e eα ))|α ∈ A} is a
tα ◦ e−1 = tα′ for α ∈ A . Then A ′ = {(tα′ , π −1 (U
trivializing cover of E|N .
12.12 Definition Let f : N → M be a smooth map and π : T M → M be the tangent

bundle of M. Then a smooth map X : N → T M which is a lifting of f is called a vector
field along f .
Let πe : f ∗ (T M) → N be the pullback of the tangent bundle by f . Then a form
ω ∈ Ω k (N; f ∗ (T M)) is called a bundle-valued k-form along f .
12.13 Let π : E → M be a bundle over M with fibre F, and let π ′′ : T E → E be

the tangent bundle of E and πe ′ : π ∗ (T M) → E be the pullback of the tangent bundle
π ′ : T M → M of M. From the commutative diagram
π∗ ✲
TE TM
π ′′ π′
❄ ❄
π ✲
E M
we see that the map (π ′′ , π ∗ ) : T E → E × T M induces a bundle map ρ : T E →

π ∗ (T M) over E .
12.14 Proposition Let π : E → M be a bundle over M n with fibre F m . Then

the pullback πe : π (Λ k (T M )) → E of the bundle π ′ : Λ k (T M ) → M can be
′ ∗
identified with a subbundle of π ′′ : Λ k (T E ) → E by means of the bundle map

α : π ∗ (Λ k (T M )) → Λ k (T E ) over E defined by
α (u, ω ) = ω ◦ (π∗ u ) k
for p ∈ M , u ∈ E p and ω ∈ Λ k (Tp M ) , which is an embedding. We have that
αu = ρu∗
for u ∈ E, where ρ is the bundle map over E defined in 12.13. If (z,W ) is a local
chart around a point u 0 on E as defined in Proposition 12.3, then
{dzi1 ∧ · · · ∧ dzik | 1 ≤ i1 < ... < ik ≤ n}
is a local basis for α (π ∗ (Λ k (T M ))) on W . A k-form ω on E belongs to α (π ∗ (Λ k (
T M ))) if and only if ω (u) (w1 , ..., wk ) = 0 whenever u ∈ E and at least one wi ∈ Tu E
is vertical. A form on E satisfying this condition is said to be horizontal over M. The
set of all horizontal k-forms on E is denoted by Ω k0 (E).
α (u, dxi1 (π (u)) ∧ · · · ∧ dxik (π (u))) = π ∗ (dxi1 ∧ · · · ∧ dxik )(u)

= d(xi1 ◦ π )(u) ∧ · · · ∧ d(xik ◦ π )(u) = dzi1 (u) ∧ · · · ∧ dzik (u)
for u ∈ W and 1 ≤ i1 < ... < ik ≤ n , which shows the first part of the propo-
sition. It only remains to prove that every horizontal k-form ω on E belongs to
α (π ∗ (Λ k (T M ))) . Let u ∈ E p and define η ∈ Λ k (Tp M ) by
η (v1 , ... , vn ) = ω (u)(w1 , ... , wk )
for v1 , ... , vk ∈ Tp M, where w1 , ... , wk ∈ Tu E are chosen so that π∗ (wi ) = vi for i =
1, ... , k. We must show that η is well defined and does not depend on the choice of
w1 , ... , wk . If wi is replaced by another element w′i ∈ Tu E with π∗ (w′i ) = vi , then wi −
w′i ∈ ker π∗ u = Vu which shows that η (v1 , ... , vk ) is unchanged since ω is horizontal.
As ω (u) = α (u, η ), it follows that ω belongs to α (π ∗ (Λ k (T M ))) .
JET BUNDLES 533
12.15 Remark In the case k = 1, the pullback πe ′ : π ∗ (T ∗ M) → E of the cotangent

bundle π ′ : T ∗ M → M of M is called the cotangent bundle of E horizontal over M .
We have that
α (π ∗ (T ∗ M) u ) = A(Vu )
for u ∈ E .
AFFINE BUNDLES
12.16 Definition A bundle π ′ : A → M with fibre Rn is said to be affine if the
following two conditions are satisfied :
(i) There is a vector bundle π : E → M so that A p is an affine space modelled on
E p for every p ∈ M .
(ii) For each point p ∈ M there are local trivializations (t ′ , π ′ −1 (U)) and
(t, π −1 (U)) around p in A and E, respectively, so that tq′ is an affine map with
linear part tq for every q ∈ U .
We say that π ′ : A → M is modelled on the vector bundle π : E → M, and that
(t ′ , π ′ −1 (U)) is modelled on the local trivialsation (t, π −1 (U)) in E .
12.17 Definition Let π1′ : A1 → M1 and π2′ : A2 → M2 be two affine bundles

modelled on the vector bundles π1 : E1 → M1 and π2 : E2 → M2 , respectively. Then
a bundle map ( f˜, f ) from π1′ to π2′ is said to be affine if there is a bundle map (g̃, f )
from π1 to π1 so that f˜p is an affine map with linear part g̃ p for every p ∈ M. We say
that (g̃, f ) is the linear part of ( f˜, f ) . If f˜ and f are diffeomorphisms, it is called an
affine equivalence .
In particular, if π1′ : A1 → M and π2′ : A2 → M are two affine bundles over the
same base space M , then an affine bundle map from π1′ to π2′ over M is a smooth map
f : A1 → A2 such that (idM , f ) is an affine bundle map. If f is a diffeomorphism, it is
called an affine equivalence over M.
12.18 Proposition Let π ′ : A → M be an affine bundle modelled on the vector

bundle π : E → M , and let s : M → A be a section of π ′ . Then we have an affine
equivalence S : E → A over M with idE : E → E as its linear part, given by
S(v) = s ◦ π (v) + v
for v ∈ E . Its inverse is given by
S −1 (u) = u − s ◦ π ′ (u)
for u ∈ A . There is a unique vector bundle structure on A so that S is an equivalence
of vector bundles, and we have that s is its zero section.
PROOF : Let (t ′ , π ′ −1 (U)) be a local trivialization in A is modelled on a local trivial-

ization (t, π −1 (U)) in E , and let h : U → Rn be the local representation of s on U so
that
t ′ ◦ s(p) = (p, h(p))
for p ∈ U . Then we have that
t ′ ◦ S ◦ t −1 (p, a) = (p, h(p) + a)
for p ∈ U and a ∈ Rn , showing the first part of the proposition.

As in Example 9.34 there is now a unique vector space structure on each fibre A p
so that S p : E p → A p is a linear isomorphism. It is given by
a u + b v = s(p) + a(u − s(p)) + b(v − s(p))
for a, b ∈ R and u, v ∈ A p . This shows that π ′ : A → M is a vector bundle satisfying

the assertions in the proposition, with (t ◦ S −1 , π ′ −1 (U)) as a local trivialization for
each local trivialization (t, π −1 (U)) in E .
12.19 Proposition Let π ′ : A → M be an affine bundle modelled on the m-

dimensional vector bundle π : E → M over M n , and let π ′′ : V → A be the vertical
tangent bundle of π ′ : A → M and πe : Ee → A be the bundle induced from π : E → M
by π ′ . Then we have an equivalence η : V → Ee over A given by
η ◦ i p ∗ (w) = (u, ωu (w))
for p ∈ M , u ∈ A p and w ∈ Tu (A p ) , where i p : A p → A is the inclusion map and ωu :

Tu (A p ) → E p is the canonical identification defined in Lemma 9.36. If (t ′ , π ′ −1 (U))
is a local trivialization around p in A modelled on the local trivialization (t, π −1 (U))
in E , we have that
ωu = t p−1 ◦ t t ′p , u .
We also have a canonical bundle map (η e , π ′ ) from π ′′ : V → A to π : E → M where
′
e = πe ◦ η , which induces a linear isomorphism between the fibres Vu and Eπ ′ (u) for
η
each u ∈ A.
PROOF : Let (e e and (t ′′ , π ′′ −1 (U))
t , πe−1 (U)) e = π ′ −1 (U) , be the local trivial-
e , where U
e
izations in E and V , respectively, obtained from the local trivializations (t, π −1 (U))
and (t ′ , π ′ −1 (U)) in E and A as described in 12.11 and Proposition 12.3. Then we
have that
t ◦ η = t ′′
e
since
tu ◦ η u ◦ i p ∗ (w) = t p ◦ ωu (w) = t t ′p , u (w) = tu′′ ◦ i p ∗ (w)
e
for p ∈ U , u ∈ A p and w ∈ Tu (A p ) , showing that η is an equivalence over A . The last

part of the proposition follows from 10.12.
JET BUNDLES 535
DERIVATIONS AND THE FRÖLICHER–NIJENHUIS BRACKET

12.20 Definition Given a linear map θ ∈ Λ1 (V ;W ) between the finite dimen-
sional vector spaces V and W , we define the interior product of a vector-valued
r-form α ∈ Λr (V ;W ) on V with values in W and a k-form ω ∈ Λk (W ) on W to be the
(r + k − 1)-form iα , θ ω ∈ Λr+k−1 (V ) on V given by
iα , θ ω (v1 , ... , vr+k−1 ) =

1 (1)
r! (k−1)! σ ∈S
∑ ε (σ ) ω (α (vσ (1) , ..., vσ (r) ), θ (vσ (r+1) ) , ..., θ (vσ (r+k−1) ))
r+k−1
for v1 , ... , vr+k−1 ∈ V . We let iα ω = 0 when k = 0. Note that

(r+k−1)!
iα , θ = A ◦ Iα , θ
r! (k−1)!
where A is the alternation operator in T r+k−1 (V ) and Iα , θ : T k (W ) → T r+k−1 (V )

is the linear map given by
Iα , θ ω = ω ◦ (α × θ k−1 )
for ω ∈ T k (W ), i.e.,
Iα , θ ω (v1 , ... , vr+k−1 ) = ω (α (v1 , ..., vr ), θ (vr+1 ) , ..., θ (vr+k−1 ))
(2)
= θ ∗ (iα (v1 ,...,vr ) ω )(vr+1 , ... , vr+k−1 )
for ω ∈ T k (W ) and v1 , ... , vr+k−1 ∈ V .
12.21 Proposition Let ( f˜, f ) be a bundle map between the vector bundles π ′ :
E ′ → N and π : E → M of dimensions m and n, respectively, and let πe : Ee → N be the
pullback of π : E → M by f . Let s : M → T k (E) be a section in the vector bundle
π k : T k (E) → M. Then we have a bundle map
e →T
ψ : T r (E ′ ; E) r+k−1
(E ′ )
e → N and π ′ r+k−1 :
over N between the vector bundles π ′′ : T r (E ′ ; E)
T r+k−1 ′
(E ) → N , which is given by
ψ (α ) = I α , f˜p s ( f (p))
for p ∈ N and α ∈ T r (E p′ ; Eep ).
PROOF : Let (t ′ , π −1 (V )) and (t, π −1 (U)) be local trivializations in the vector

bundles π ′ : E ′ → N and π : E → M, respectively, with f (V ) ⊂ U, and let τ :
(π k ) −1 (U) → U × T k (Rn ) and τ ′ : (π ′ r+k−1 ) −1 (V ) → V × T r+k−1 (Rm ) be the
corresponding equivalences over U and V defined in Proposition 4.53. Let τ ′′ :

(π ′′ ) −1 (V ) → V × T r (Rm ; Rn ) be the equivalence over V defined in Proposition
5.28. Then we have that
τ ◦ s(q) = (q, h(q))
for q ∈ U, where h : U → T k (Rn ) is the smooth map given by h(q) = (tq−1 )∗ s(q).
We also have that
−1
t ◦ f˜ ◦ t ′ (p, v) = ( f (p), F(p)(v))
for (p, v) ∈ V × Rm , where F : V → T 1 (Rm ; Rn ) is the smooth map given by F(p) =
t f (p) ◦ f˜p ◦ t p′ −1 for p ∈ V . From this it follows that
−1
τ ′ ◦ ψ ◦ τ ′′ (p, S) = (p, h( f (p)) ◦ {S × F(p) k−1 })
for (p, S) ∈ V × T r (Rm ; Rn ), since

−1
τ ′ ◦ ψ ◦ τ ′′ (p, S)(a1 , ... , ar+k−1 )
−1 ∗
= (t p′ ) {It −1 ◦ S ◦ (t ′ )r , f˜p s ( f (p))}(a1 , ... , ar+k−1 )
f (p) p
= (t f−1 ∗
(p) ) {s ( f (p))}(S(a1 , ..., ar ),
−1 −1
t f (p) ◦ f˜p ◦ t p′ (ar+1 ) , ...,t f (p) ◦ f˜p ◦ t p′ (ar+k−1 ))
= h( f (p))(S(a1 , ..., ar ), F(p)(ar+1 ) , ..., F(p)(ar+k−1 ))
for a1 , ... , ar+k−1 ∈ Rm .
12.22 Remark Let ω : M → Λk (E) be a section in the vector bundle π ′ : Λ k (E) →

M. Then the bundle map ψ obtained from ω as described in Proposition 12.21 also
gives rise to a bundle map
e → Λ r+k−1 (E ′ )
φ : Λ r (E ′ ; E)
over N given by
φ (α ) = i α , f˜p ω ( f (p))
for p ∈ N and α ∈ Λ r (E p′ ; Eep ).
12.23 Proposition If θ ∈ Λ1 (V ;W ) , α ∈ Λr (V ;W ) and λ1 , ... , λk ∈ W ∗ , we have

that
k
iα , θ ( λ1 ∧ · · · ∧ λk ) = ∑ (−1) i+1 iα , θ ( λ i ) ∧ θ ∗ (λ 1 ∧ · · · ∧ λ i−1 ∧ λ i+1 ∧ · · · ∧ λ k ) .
i=1
JET BUNDLES 537
PROOF : If v1 , ... , vr+k−1 ∈ V , it follows from Proposition 5.13 (1) that
Iα , θ ( λ 1 ∧ · · · ∧ λ k )(v1 , ... , vr+k−1 )
= ( λ 1 ∧ · · · ∧ λ k )(α (v1 , ..., vr ), θ (vr+1 ) , ..., θ (vr+k−1 ))
= ∑ ε (σ ) λ σ (1) (α (v1 , ..., vr )) λ σ (2) ( θ (vr+1 )) · · · λ σ (k) ( θ (vr+k−1 ))

σ ∈Sk

λ 1 (α (v1 , ..., vr )) λ 1 ( θ (vr+1 )) · · · λ 1 ( θ (vr+k−1 ))

λ 2 (α (v1 , ..., vr )) λ 2 ( θ (vr+1 )) · · · λ 2 ( θ (vr+k−1 ))

= .. .. ..
. . .

λ k (α (v1 , ..., vr )) λ k ( θ (vr+1 )) · · · λ k ( θ (vr+k−1 ))
k
= ∑ (−1) i+1 λ i (α (v1 , ..., vr ))
i=1
(λ 1 ∧ · · · ∧ λ i−1 ∧ λ i+1 ∧ · · · ∧ λ k )( θ (vr+1 ), ... , θ (vr+k−1 ))

k
= ∑ (−1) i+1 Iα , θ ( λ i ) ⊗ θ ∗ (λ 1 ∧ · · · ∧ λ i−1 ∧ λ i+1 ∧ · · · ∧ λ k )(v1 , ... , vr+k−1 )
i=1
where we have expanded the determinant along the first column. The result now
follows by applying the alternation operator on both sides and using Definition
5.11.
12.24 Proposition If θ ∈ Λ1 (V ;W ) , α ∈ Λr (V ;W ) , ω ∈ Λk (W ) and η ∈ Λl (W ) ,

we have that
iα , θ ( ω ∧ η ) = iα , θ ( ω ) ∧ θ ∗ (η ) + (−1) k (r−1) θ ∗ ( ω ) ∧ iα , θ (η ) .
PROOF : If ω = ω1 ∧ · · · ∧ ωk and η = η1 ∧ · · · ∧ ηl , where ω1 , ... , ωk , η1 , ... , ηl

∈ W ∗ , it follows from Proposition 12.23 that
k
iα , θ ( ω ∧ η ) = ∑ (−1) i+1 iα , θ ( ωi ) ∧ θ ∗ ( ω1 ∧ · · · ∧ ωi−1 ∧ ωi+1 ∧ · · · ∧ ωk ∧ η )
i=1
l
+ ∑ (−1) k+ j+1 iα , θ ( η j ) ∧ θ ∗ ( ω ∧ η1 ∧ · · · ∧ η j−1 ∧ η j+1 ∧ · · · ∧ ηl )
j=1
= iα , θ ( ω ) ∧ θ ∗ (η ) + (−1) k (r−1) θ ∗ ( ω ) ∧ iα , θ (η ) .
The general result now follows since every form ω ∈ Λk (W ) and η ∈ Λl (W ) can be
expressed as a linear combination of forms of the above type.
12.25 Proposition Let
F1✲
V1 V2
θ1 θ2
❄ ❄
W1 ✲ W
2
F2
be a commutative diagram of vector spaces and linear maps, and suppose that α1 ∈
Λr (V1 ;W1 ) and α2 ∈ Λr (V2 ;W2 ) are vector-valued r-forms satisfying
F2 ◦ α1 = α2 ◦ F1 r .
Then it follows that

i α1 , θ1 F2∗ ( ω ) = F1∗ ( i α2 , θ2 ω )
for ω ∈ Λk (W2 ) .
I α1 , θ1 F2∗ ( ω ) = I α1 , θ1 ( ω ◦ F2 k ) = ω ◦ F2 k ◦ (α1 × θ1k−1 )
= ω ◦ (α2 × θ2k−1 ) ◦ F1 r+k−1 = F1∗ ( ω ◦ (α2 × θ2k−1 )) = F1∗ ( I α2 , θ2 ω ) ,
and the result now follows from Proposition 5.9.
12.26 Proposition Let f : N → M be a smooth map. If α ∈ Ω r (N; f ∗ ( T M)) is

a bundle-valued r-form along f and ω a k-form on M, we have an (r + k − 1)-form
iα ω on N, called the interior product of α and ω , defined by
(iα ω )(p) = i α (p), f∗ p ω ( f (p))
for each p ∈ N .
PROOF : We have that iα ω = φ ◦ α , where φ : Λ r (T N; f ∗ (T M)) → Λ r+k−1 (T N) is
the bundle map over N obtained from ω as described in Remark 12.22.
12.27 Let V and W be vector spaces, and let B = {u1 , ..., un } be a basis for V
with dual basis B ∗ = {u∗1 , ..., u∗n }. Then we have a linear isomorphism
φe : L(W ∗, Λ k (V )) → Λ k1 (V,W )
defined by
φe (G)(v1 , ..., vk , λ ) = G(λ )(v1 , ..., vk )
JET BUNDLES 539
for G ∈ L(W ∗, Λ k (V )), v1 , ..., vk ∈ V and λ ∈ W ∗ , with inverse
e : Λ k1 (V,W ) → L(W ∗, Λ k (V ))
ψ
given by
e (ω )(λ ) =
ψ ∑ ω (ui1 , ..., uik , λ ) u∗i1 ∧ · · · ∧ u∗ik
i1 <...<ik
for ω ∈ Λ k1 (V,W ) and λ ∈ W ∗ . Indeed, we have that
φe(ψ
e (ω ))(u j1 , ..., u jk , λ ) = ψ
e (ω )(λ )(u j1 , ..., u jk ) = ω (u j1 , ..., u jk , λ )
for 1 ≤ j1 < ... < jk ≤ n and

e (φe(G))(λ ) =
ψ ∑ φe(G)(ui1 , ..., uik , λ ) u∗i1 ∧ · · · ∧ u∗ik
i1 <...<ik
= ∑ G(λ )(ui1 , ..., uik ) u∗i1 ∧ · · · ∧ u∗ik = G(λ )

i1 <...<ik
which shows that ψ e = φe−1 . In particular, it follows that ψ

e does not depend on the
choice of the basis B.
If the space W is also finite dimensional, then it follows from 5.15 that we have
a linear isomorphism
µ∗ : Λ k (V ) ⊗ W → Λ k1 (V,W )
induced by the bilinear map
µ : Λ k (V ) × W → Λ k1 (V,W )
defined by
µ (ω , w) = ω ⊗ w
for ω ∈ Λ k (V ) and w ∈ W , i.e.,
µ (ω , w)(v1 , ..., vk , λ ) = ω (v1 , ..., vk ) λ (w)
for v1 , ..., vk ∈ V and λ ∈ W ∗ . The composed map
e ◦ µ : Λ k (V ) × W → L(W ∗, Λ k (V ))
ψ
is given by
e ◦ µ (ω , w)(λ ) = λ (w) ω
ψ
for ω ∈ Λ k (V ), w ∈ W and λ ∈ W ∗ .
Now let φ : Λ k (V ;W ) → Λ k1 (V,W ) and ψ : Λ k1 (V,W ) → Λ k (V ;W ) be the linear
isomorphisms induced by the isomorphisms φ : T k (V ;W ) → T k1 (V,W ) and ψ :
T k1 (V,W ) → T k (V ;W ) defined in 5.15. Then we have a linear isomorphism
e ◦ φ : Λ k (V ;W ) → L(W ∗, Λ k (V ))
ψ
which is given by
e ◦ φ (ω )(λ ) = λ ◦ ω
ψ
k
for ω ∈ Λ (V ;W ) and λ ∈ W ∗. Its inverse is the isomorphism
ψ ◦ φe : L(W ∗, Λ k (V )) → Λ k (V ;W )
given by
m
ψ ◦ φe (G)(v1 , ..., vk ) = ∑ G(w∗i )(v1 , ..., vk ) wi
i=1
for G ∈ L(W ∗, Λ k (V )) and v1 , ..., vk ∈ V , where C = {w1 , ..., wm } is an arbitrary basis

for W with dual basis C ∗ = {w∗1 , ..., w∗m }.
12.28 Definition Let f : N → M be a smooth map between the smooth manifolds

N and M. An R-linear map D : Ω(M) → Ω(N) is called a derivation along f of degree
r if it satisfies the following two properties:
(i) If ω ∈ Ω k (M), then D ω ∈ Ω k+r (N) ,
(ii) D (ω ∧ η ) = (D ω ) ∧ f ∗ (η ) + (−1) kr f ∗ ( ω ) ∧ (D η ) whenever ω ∈ Ω k (M)
and η ∈ Ω l (M) (Leibniz’ rule).
The derivation D is said to be of type i∗ if D f = 0 for every f ∈ Ω 0 (M) , and of type
d∗ if
D ◦ d = (−1) kr d ◦ D .
12.29 Proposition Let f : N → M be a smooth map between the smooth man-

ifolds N and M , and let α ∈ Ω r (N; f ∗ ( T M)) be a bundle-valued r-form along f .
Then we have a derivation iα : Ω(M) → Ω(N) along f of type i∗ and degree (r − 1) ,
where iα ω is the interior product of α and ω for each k-form ω on M.
Conversely, for every derivation D : Ω(M) → Ω(N) along f of type i∗ and degree
(r − 1) , there is a unique bundle-valued r-form α along f so that D = iα . The map
D 1 : Ω 1 (M) → Ω r (N) induced by D is linear over F (M) , and for each such map
D 1 there is a unique bundle-valued r-form α along f so that D 1 (ω ) = iα ω for
ω ∈ Ω 1 (M) .
PROOF : The first part of the proposition follows from Propositions 12.24 and 12.26.
To prove the last part, we first note that
D 1 (g ω ) = (Dg) ∧ f ∗ (ω ) + f ∗ (g) D 1 ω = f ∗ (g) D 1 ω
for every g ∈ Ω 0 (M) and ω ∈ Ω 1 (M) , showing that D 1 is linear over F (M) .
From this it follows that given a point p ∈ N , D 1 ω (p) only depends on the
value of ω at f (p) , i.e., D 1 ω 1 (p) = D 1 ω 2 (p) whenever ω 1 ( f (p)) = ω 2 ( f (p)).
Considering the difference ω 2 − ω 1 it is clearly enough to prove that D 1 ω (p) = 0 if
JET BUNDLES 541
ω ( f (p)) = 0. To show this, we choose a local chart (x,U) around f (p) and a smooth
function h : M → R with h( f (p)) = 1 and supp (h) ⊂ U. If
n
ω | U = ∑ bi dxi ,
i=1
we extend the functions h bi and 1-forms h dxi on U to smooth functions gi and 1-

forms θi on M, respectively, by defining them to be zero outside U. Then we have
that
n
h2 ω = ∑ gi θi
i=1
so that
n
(h ◦ f ) 2 D 1 ω = D 1 (h2 ω ) = ∑ (g i ◦ f ) D 1 θi .
i=1
Evaluating at p and using that h( f (p)) = 1 and gi ( f (p)) = 0 for i = 1, ... , n, we

conclude that D 1 ω (p) = 0 in the case when ω ( f (p)) = 0.
Now assume that D = iα for a bundle-valued r-form α along f . Then we have
that
D 1 ω (p) = iα ω (p) = ω ( f (p)) ◦ α (p)
for ω ∈ Ω 1 (M) and p ∈ N . Let C p = {w1 , ..., wn } be a basis for T f (p) M with dual
basis C p∗ = {w∗1 , ..., w∗n }, and choose 1-forms ω 1 , ... , ω n on M with ω i ( f (p)) = w∗i
for i = 1, ... , n using Proposition 2.56 (2). Then
n
α (p)(v1 , ..., vr ) = ∑ D 1 ω i (p)(v1 , ..., vr ) wi (1)
i=1
for v1 , ..., vr ∈ Tp N , which shows the uniqueness of α .

To show existence, we now define α by formula (1) for each p ∈ N. From 12.27
and the above arguments we know that it is well defined. It only remains to show
that α is smooth on N. Fix p ∈ N, and choose a local chart (x,U) around f (p) on M
and an open neighbourhood V of f (p) with V ⊂ U. By Proposition 2.56 (1) there are
1-forms ω i on M which coincide with dxi on V for i = 1, ... , n. Then we have that

n
α (q)(v1 , ..., vr ) = ∑ D 1 ω (q)(v1 , ..., vr ) ∂ i
i
i=1
∂x
f (q)
for q ∈ f −1 (V ) and v1 , ..., vr ∈ Tq N , showing that α is smooth at p by Proposition

5.44 (2). As p was an arbitrary point in N, this completes the proof that α is smooth.
Now since D and iα coincide on Ω 0 (M) and Ω 1 (M) , they coincide on Ω(M) by
Leibniz’ rule.
12.30 Proposition Let f : N → M be a smooth map between the smooth man-

ifolds N and M , and let α ∈ Ω r (N; f ∗ ( T M)) be a bundle-valued r-form along f .
Then we have a derivation dα : Ω(M) → Ω(N) along f of type d∗ and degree r

defined by
dα = iα ◦ d + (−1) r d ◦ iα
Conversely, for every derivation D : Ω(M) → Ω(N) along f of type d∗ and degree
r , there is a unique bundle-valued r-form α along f so that D = dα .
PROOF : We see that dα satisfies Leibniz’ rule since
dα (ω ∧ η ) = iα {(d ω ) ∧ η + (−1) k ω ∧ (d η ) }
+ (−1) r d {(iα ω ) ∧ f ∗ (η ) + (−1) k(r−1) f ∗ ( ω ) ∧ (iα η ) }
= iα (d ω ) ∧ f ∗ (η ) + (−1) (k+1)(r−1) f ∗ (d ω ) ∧ (iα η )
+ (−1) k {(iα ω ) ∧ f ∗ (d η ) + (−1) k(r−1) f ∗ ( ω ) ∧ iα (d η ) }

+ (−1) r {d (iα ω ) ∧ f ∗ (η ) + (−1) r+k−1 (iα ω ) ∧ f ∗ (d η )
+ (−1) k(r−1) { f ∗ (d ω ) ∧ (iα η ) + (−1) k f ∗ ( ω ) ∧ d (iα η ) } }
= {iα (d ω ) + (−1) r d (iα ω ) } ∧ f ∗ (η )

+ (−1) kr f ∗ ( ω ) ∧ {iα (d η ) + (−1) r d (iα η ) }
+ {(−1) (k+1)(r−1) + (−1) r (−1) k(r−1) } f ∗ (d ω ) ∧ (iα η )

+ {(−1) k + (−1) r (−1) r+k−1 } (iα ω ) ∧ f ∗ (d η )
= (dα ω ) ∧ f ∗ (η ) + (−1) kr f ∗ ( ω ) ∧ (dα η )
whenever ω ∈ Ω k (M) and η ∈ Ω l (M) . Since dα is clearly R-linear, it is a derivation

along f of degree r which is of type d∗ since
dα ◦ d = (−1) r d ◦ iα ◦ d = (−1) r d ◦ dα ,
thus proving the first part of the proposition.

To prove the last part, let D : Ω(M) → Ω(N) be a derivation along f of type d∗
and degree r . Given vector fields X1 , ... , Xr ∈ T1 (N) , we have a map
F : F (M) → F (N)
defined by
F(g) = (Dg)N (X1 , ... , Xr )
for g ∈ F (M), satisfying
F(g1 g2 ) = f ∗ (g2 ) F(g1 ) + f ∗ (g1 ) F(g2 )

JET BUNDLES 543
for g1 , g2 ∈ F (M). Then F = XN for a vector field X along f , and we obtain a map
G : T1 (N) × · · · × T1 (N) → Γ(N; f ∗ (T M))

| {z }
r
which is multilinear and skew symmetric over F (N). Hence there is a unique bundle-
valued r-form α ∈ Ω r (N; f ∗ (T M)) along f such that G = αN . We have that
d α g(p) = iα (dg)(p) = dg( f (p)) ◦ α (p)
for p ∈ N, so that
(d α g)N (X1 , ... , Xr )(p) = G(X1 , ... , Xr )(p)(g) = (Dg)N (X1 , ... , Xr )(p)
for g ∈ F (M) , X1 , ... , Xr ∈ T1 (N) and p ∈ N, showing that D = dα .
12.31 Proposition Let f : N → M be a smooth map between the smooth mani-

folds N and M . Then every derivation D : Ω(M) → Ω(N) along f can be written as
a sum
D = iD + dD ,
where i D and d D are derivations along f of type i∗ and type d∗ , respectively.
PROOF : Let d D be the derivation along f of type d∗ so that d D g = Dg for g ∈ Ω 0 (M) ,
and let i D = D − d D .
12.32 Consider the commutative diagram
g1
N1 ✲ N
2
f1 f2
❄ ❄
M1 ✲ M
2
g2
of smooth manifolds and mappings. Two bundle-valued r-forms α1 ∈ Ω r (N1 ;

f1∗ (T M1 )) and α2 ∈ Ω r (N2 ; f2∗ (T M2 )) along f1 and f2 , respectively, are said to be
(g1 , g2 )-related if
g 2∗ f1(p) ◦ α1 (p) = α 2 (g1 (p)) ◦ (g 1∗ p) r
for p ∈ N1 , i.e.,
g2∗ ◦ α1 (p)(v1 , ... , vr ) = α2 (g1 (p))(g1∗ (v1 ), ... , g1∗ (vr ))
for p ∈ N1 and v1 , ... , vr ∈ Tp N1 . By Proposition 12.25 this implies that
iα1 g∗2 ( ω ) = g∗1 (iα2 ω )

for ω ∈ Ω k (M2 ) , since
(iα1 g∗2 ( ω ))(p) = i α1 (p), f1 ∗ p g∗2 ( ω ) ( f1 (p)) = i α1 (p), f1 ∗ p g∗2∗ f1(p) ( ω (g 2 ( f1 (p))))
= g∗1∗ p(i α2 (g1 (p)), f2 ∗ g ω ( f2 (g1 (p)))) = g∗1∗ p(iα2 ω (g1 (p))) = g∗1 (iα2 ω )(p)
1 (p)
for every p ∈ N1 . It also follows that

dα1 g∗2 ( ω ) = g∗1 (dα2 ω )
for ω ∈ Ω k (M2 ) . Note that α2 is uniquely determined by α1 when g1 is a surjective
submersion.
Assuming that f1 and g1 are surjective submersions, let α1 ∈ Ω r (N1 ; f1∗ (T M1 ))
and α2 ∈ Ω r (N2 ; f2∗ (T M2 )) be (g1 , g2 )-related bundle-valued r-forms along f1
and f2 , respectively, and let β1 ∈ Ω s (N1 ; g∗1 (T N2 )) and β2 ∈ Ω s (M1 ; g∗2 (T M2 ))
be ( f1 , f2 )-related bundle-valued s-forms along g1 and g2 . Consider the map D :
Ω(M2 ) → Ω(N1 ) given by
D = d α1 ◦ d β2 − (−1)rs d β1 ◦ d α2 .
We see that D satisfies Leibniz’ rule since
D (ω ∧ η ) = dα1 {(dβ2 ω ) ∧ g∗2 (η ) + (−1) ks g∗2 (ω ) ∧ (dβ2 η ) }
− (−1) rs dβ1 {(dα2 ω ) ∧ f2∗ (η ) + (−1) kr f2∗ ( ω ) ∧ (dα2 η ) }
= dα1 (dβ2 ω ) ∧ (g2 ◦ f1 )∗ (η ) + (−1) (k+s)r f1∗ (dβ2 ω ) ∧ dα1 (g∗2 ( η ))
+ (−1) ks {dα1 (g∗2 ( ω )) ∧ f1∗ (dβ2 η ) + (−1) kr (g2 ◦ f1 )∗ ( ω ) ∧ dα1 (dβ2 η ) }
− (−1) rs {dβ1 (dα2 ω ) ∧ ( f2 ◦ g1 )∗ (η ) + (−1) (k+r)s g∗1 (dα2 ω ) ∧ dβ1 ( f2∗ ( η ))
+ (−1) kr {dβ1 ( f2∗ ( ω )) ∧ g∗1 (dα2 η ) + (−1) ks ( f2 ◦ g1 )∗ ( ω ) ∧ dβ1 (dα2 η ) } }
= {dα1 (dβ2 ω ) − (−1) rs dβ1 (dα2 ω ) } ∧ (g2 ◦ f1 )∗ (η )
+ (−1) k(r+s) (g2 ◦ f1 )∗ ( ω ) ∧ {dα1 (dβ2 η ) − (−1) rs dβ1 (dα2 η ) }
+ {(−1) (k+s)r − (−1) rs (−1) kr } f1∗ (dβ2 ω ) ∧ g∗1 (dα2 η )
+ {(−1) ks − (−1) rs (−1) (k+r)s } g∗1 (dα2 ω ) ∧ f1∗ (dβ2 η )
= (D ω ) ∧ (g2 ◦ f1 )∗ (η ) + (−1) k(r+s) (g2 ◦ f1 )∗ ( ω ) ∧ (D η )
whenever ω ∈ Ω k (M2 ) and η ∈ Ω l (M2 ) . Since D is clearly R-linear, it is a derivation

along g2 ◦ f1 of degree r + s which is of type d∗ .
By Proposition 12.30 there is now a unique bundle-valued (r + s)-form [ α1 , β1 ] ∈
Ω r+s (N1 ; (g2 ◦ f1 )∗ (T M2 )) along g2 ◦ f1 , called the Frölicher–Nijenhuis bracket of
JET BUNDLES 545
α1 and β1 , so that
D = d [ α1 , β1 ] .
12.33 Proposition Let α , β ∈ Ω 1 (M; T M) be bundle-valued 1-forms on M . Then

we have that
[ α , β ](X,Y ) = β ( α ([ X,Y ])) + [ α (X), β (Y ) ] − β ([ α (X),Y ]) − β ([ X, α (Y ) ])

+ α ( β ([ X,Y ])) + [ β (X), α (Y ) ] − α ([ β (X),Y ]) − α ([ X, β (Y ) ])
for every pair of vector fields X and Y on M .

PROOF : For each f ∈ F (M) we have that
[ α , β ](X,Y )( f ) = d f ([ α , β ](X,Y )) = i [ α , β ] (d f )(X,Y ) = d [ α , β ] ( f )(X,Y )

= d α ◦ d β ( f )(X,Y ) + d β ◦ d α ( f )(X,Y ) ,
and Remark 5.86 implies that
d α ◦ d β ( f )(X,Y ) = (iα ◦ d − d ◦ iα ) ◦ i β (d f )(X,Y )
= d (d f · β )(α (X),Y ) + d (d f · β )(X, α (Y )) − d (d f · β · α )(X,Y )
= { α (X)(d f · β (Y )) − Y (d f · β · α (X)) − d f · β ([ α (X),Y ]) }

+ {X(d f · β · α (Y )) − α (Y )(d f · β (X)) − d f · β ([ X, α (Y ) ]) }
− {X(d f · β · α (Y )) − Y (d f · β · α (X)) − d f · β · α ([ X,Y ]) }
= β ( α ([ X,Y ]))( f ) + α (X) β (Y )( f ) − α (Y ) β (X)( f )

− β ([ α (X),Y ])( f ) − β ([ X, α (Y ) ])( f )
Combining this we obtain the formula in the proposition.
12.34 Corollary Let α ∈ Ω 1 (M; T M) be a bundle-valued 1-form on M . Then

we have that
1
2 [ α , α ](X,Y ) = α ( α ([ X,Y ])) + [ α (X), α (Y ) ] − α ([ α (X),Y ]) − α ([ X, α (Y ) ])
for every pair of vector fields X and Y on M .

FIRST ORDER JET BUNDLES
12.35 Definition Let π : E → M be a bundle over M with fibre F, and let p be

a point on M . We say that two sections φ , ψ ∈ Γ p (E) have contact of order 0 at p
if φ (p) = ψ (p) . They are said to have kth order contact at p for k ≥ 1 if φ ∗ and
ψ ∗ have (k − 1) st order contact at every point in Tp M . This defines an equivalence
relation in Γ p (E) for each k ≥ 0 , and the equivalence class containing φ is called the
k-jet of φ at p and is denoted by j kp φ . The points p and φ (p) are called the source
and target of j kp φ , respectively, and the set of equivalence classes in Γ p (E) is called
the k-jet space of E at p and is denoted by J pk E . We define the k-jet manifold of E to
be the disjoint union [
J kE = J pk E .
p∈M
The 0-jet manifold J 0 E is identified with E. The maps πk : J k E → M and πk,0 :

J k E → E defined by πk ( j kp φ ) = p and πk,0 ( j kp φ ) = φ (p), respectively, for p ∈ M
and φ ∈ Γ p (E), are called the source and target projections. If 0 ≤ l ≤ k , the l-jet
projection πk, l : J k E → J l E is the map defined by πk, l ( j kp φ ) = j lp φ . We have that
πk,m = πl,m ◦ πk, l when 0 ≤ m ≤ l ≤ k , and πk = π ◦ πk,0 when k ≥ 0 .
12.36 Proposition Let π : E → M be a bundle over M n with fibre F m , and let

πe ′ : π ∗ (T M) → E be the pullback of the tangent bundle π ′ : T M → M by π , and
π ′′ : V → E be the vertical bundle of π . Then π1,0 : J 1 E → E is an affine bundle
over E modelled on the vector bundle π ′′′ : Λ 1 (π ∗ (T M) ;V ) → E. A section of π1,0
is called a jet field on E.
PROOF : Consider a local chart (z,W ) around a point u 0 on E as described in Propo-

sition 12.3, and let z1 : W → Rn and z2 : W → Rm be the component maps of
−1
z : W → Rn × Rm . Then we have a bijection τz : π1,0 (W ) → W × L(Rn , Rm ) which is
given by
τz ( j 1p φ ) = (φ (p), D(z2 ◦ φ ◦ x−1 )(x(p)))
for p ∈ O and φ ∈ Γ p (E) with φ (p) = u ∈ W . We see that τz is well defined since
D(z2 ◦ φ ◦ x−1 )(x(p)) = pr2 ◦ tz,u ◦ φ ∗ p ◦ tx,p

−1
,
where pr2 : Rn × Rm → Rm is the projection on the second factor.

Suppose that (e e ) is another local chart on E obtained in the same way from a
z, W
e in E and local charts (e
local trivialization (t , π −1 (U))
e e and (e
x, O) e on M and F,
y, Q)
e e
respectively, with O ⊂ U . Then we have that
τ z̃ ◦ τz−1 (u, G) =
(1)
z2 ◦ z−1 )(z(u)) + D2 (e
u, {D1 (e z2 ◦ z−1 )(z(u)) ◦ G } ◦ D(x ◦ xe−1 )(e
z1 (u))
JET BUNDLES 547
for u ∈ W ∩ W e and G ∈ L(Rn , Rm ) . Hence it follows from Proposition 12.6

1
that π1,0 : J E → E is a bundle over E with fibre R n m , and with ((id W × β )
−1
◦ τz , π1,0 (W )) as a local trivialization for each local chart (z,W ) on E of the above
form, using the linear isomorphism β : L(Rn , Rm ) → R n m defined in Corollary 5.29.
By Proposition 5.28 we have a local trivialization ((id W × β ) ◦ τz′ , π ′′′ −1 (W )) in
the vector bundle π ′′′ : Λ 1 (π ∗ (T M) ;V ) → E associated with the local chart (z,W )
on E , where τz′ : π ′′′ −1 (W ) → W × L(Rn , Rm ) is the equivalence over W defined by
τz′ (F) = (u, pr2 ◦ tz,u ◦ F ◦ tx,p

−1
)
for u ∈ W , F ∈ Λ 1 (π ∗ (T M) ;V ) u and p = π (u) . We also have that
τ ′z̃ ◦ τz′ −1 (u, G) = (u, D2 (e

z2 ◦ z−1 )(z(u)) ◦ G ◦ D(x ◦ xe−1 )(e
z1 (u))) (2)
for u ∈ W ∩ We and G ∈ L(Rn , Rm ) . Using this, we can define an operation of the

vector space Λ 1 (π ∗ (T M) ;V ) u on the fibre J 1 E u by
′
τz,u ( j 1p φ + F) = τz,u ( j 1p φ ) + τz,u (F) . (3)
In fact, we have that j 1p φ + F = j 1p ψ for a section ψ ∈ Γ p (E) given by
ψ = z −1 ◦ {z ◦ φ + tz,u ◦ F ◦ tx,p
−1 −1
◦ Tx(p) ◦ x} ,
where Tx(p) : Rn → Rn is translation by x(p) , since this implies that
z2 ◦ ψ ◦ x−1 = z2 ◦ φ ◦ x−1 + pr2 ◦ tz,u ◦ F ◦ tx,p

−1 −1
◦ Tx(p) .
′ (F)
It follows from (1) with G = τz,u ( j 1p φ + F) and G = τz,u ( j 1p φ ) , (2) with G = τz,u
and (3) that we also have
τ z̃,u ( j 1p φ + F) = τ z̃,u ( j 1p φ ) + τ ′z̃,u (F) ,
showing that the operation on J 1 E u given by (3) is well defined and does not depend
on the choice of local chart (z,W ) .
12.37 Remark From each local chart (z,W ) on E with coordinate functions xi and
−1
yα defined in Remark 12.4, we obtain an induced local chart ((z × β ) ◦ τz , π1,0 (W ))
1 i α α
on J E with coordinate functions x , y and y i defined by
∂ (yα ◦ φ )
xi ( j 1p φ ) = xi (p) , yα ( j 1p φ ) = yα ◦ φ (p) and yαi ( j 1p φ ) = (p)
∂ xi
−1
for j 1p φ ∈ π1,0 (W ) , i = 1, 2, ... , n and α = 1, 2, ... , m. It will be clear from the con-
text whether x and yα denote coordinate functions on J 1 E , on E or on M and F
i
respectively. The functions yαi are called derivative coordinates .

12.38 Proposition Let π : E → M be a bundle over M n with fibre F m . Then we

have a canonical embedding λ : J 1 E → Λ 1 (π ∗ (T M) ; T E) given by
λ ( j 1p φ ) = φ ∗ p
for p ∈ M and φ ∈ Γ p (E) , which is an affine bundle map over E. Its linear part is the
map α : Λ 1 (π ∗ (T M) ;V ) → Λ 1 (π ∗ (T M) ; T E) given by
α (F) = iu ◦ F
for u ∈ E and F ∈ Λ 1 (π ∗ (T M) ;V ) u , where i : V → T E is the inclusion map.
PROOF : Let (z,W ) be a local chart around a point u0 in E as described in Proposition
−1
12.3, and let τz : π1,0 (W ) → W × L(Rn , Rm ) be the equivalence over W defined in
the proof of Proposition 12.36. Furthemore, let τz′ : π ′ −1 (W ) → W × L(Rn , Rm ) and
τz′′ : π ′′ −1 (W ) → W × L(Rn , Rn+m ) be the equivalences over W in the vector bundles
π ′ : Λ 1 (π ∗ (T M) ;V ) → E and π ′′ : Λ 1 (π ∗ (T M) ; T E) → E defined in Proposition
5.28. Then we have that
τ ′′z,u ◦ λ u ◦ τ −1
z,u (G)(a) = (a, G(a))
and
−1
τ ′′z,u ◦ α u ◦ τ ′z,u (G)(a) = (0, G(a))
for u ∈ W , G ∈ L(Rn , Rm ) and a ∈ Rn , showing that
−1
λ u ◦ τ −1 ′ −1 ′ ′
z,u (G + G ) = λ u ◦ τ z,u (G) + α u ◦ τ z,u (G )
for G, G ′ ∈ L(Rn , Rm ) , which completes the proof of the proposition.
12.39 Proposition If π : E → M is a trivial bundle over M with fibre F, then

π1,0 : J 1 E → E is a vector bundle over E .
PROOF : Let t : E → M × F be a trivialization over M , and let pr1 : M × F → M and
pr2 : M × F → F be the projections on the first and second factor. Then we have a
section s : E → J 1 E of π1,0 given by s(u) = j 1p φ u for u ∈ E , where p = π (u) and
φ u : M → E is the section of π defined by
φ u (q) = t −1 (q, pr2 ◦ t(u))
for q ∈ M . Note that φ u (p) = u when p ∈ M and u ∈ E p since p = π (u) = pr1 ◦ t(u) ,
which shows that π1,0 ◦ s = idE .
In order to prove that s is smooth, consider a local chart (z,W ) around a point
u 0 on E as described in Proposition 12.3. Using the same notation as in the proof of
Proposition 12.36, we have that
z2 ◦ φu (q) = z2 (u)
for u ∈ W and q ∈ M , which implies that
τz ◦ s(u) = (u, 0)
for u ∈ W . The result now follows from Proposition 12.18.
JET BUNDLES 549

1
π1 : J E → M is also a bundle.
PROOF : Let (t, π −1 (U)) be a local trivialization in π : E → M , where U is the
coordinate neighbourhood of a local chart (x,U) on M , and consider the triv-
ial bundle π ′ : Rn × F → Rn . Since π1′ = π ′ ◦ π1,0
′ is a submersion, we have that
J 01 (Rn × F) = π1′ −1 (0) is a closed submanifold of J 1 (Rn × F) .

Now let t ′ : π1−1 (U) → U × J 01 (Rn × F) be the map given by
t ′ ( j 1p φ ) = (p, j 01 ψ )
for p ∈ U and φ ∈ Γ p (E) , where ψ ∈ Γ0 (Rn × F) is the section defined by

−1
ψ = (Tx(p) × idF ) ◦ (x × idF ) ◦ t ◦ φ ◦ x−1 ◦ Tx(p) ,
using the translation Tx(p) : Rn → Rn by x(p) . We want to show that t ′ is a diffeomor-

phism. Consider the local chart (z,W ) in E obtained from (t, π −1 (U)) , (x,U) and
a local chart (y, Q) on F as described in Proposition 12.3, so that W = t −1 (U × Q)
and z = (x × y) ◦ t . We also obtain the local chart (e z , Rn × Q) in Rn × F, where
e
z = idR × y . With the same notation as in the proof of Proposition 12.36, we have
n
that
z2 ◦ ψ = z2 ◦ φ ◦ x−1 ◦ Tx(p) ,
e
which shows that
z2 ◦ ψ (0) = z2 ◦ φ (p)
e
and
z2 ◦ ψ )(0) = D(z2 ◦ φ ◦ x−1 )(x(p)) .
D(e
−1
Using the local charts ((z × β ) ◦ τz , π1,0 (W )) and (η , O) in J 1 E and J 01 (Rn × F),
′ −1 ({0} × Q) and η = (e
respectively, where O = π1,0 z2 × β ) ◦ τ z̃ |O , we now have that
(x × η ) ◦ t ′ ( j 1p φ ) = (x(p), e
z2 ◦ ψ (0), β ◦ D(e
z2 ◦ ψ )(0))
= (z ◦ φ (p), β ◦ D(z2 ◦ φ ◦ x−1 )(x(p))) = (z × β ) ◦ τ z ( j 1p φ ) .
This completes the proof that π1 : J 1 E → M is a bundle, with (t ′ , π1−1 (U)) as a lo-
cal trivialization for each local trivialization (t, π −1 (U)) in π : E → M of the above
form.
12.41 Corollary The 1-jet bundle π1 : J 1 (Rn × F) → Rn of the trivial bundle

π : Rn × F → Rn is trivial, having a trivialization t : J 1 (Rn × F) → Rn × J 01 (Rn × F)
over Rn given by
t( j a1 φ ) = (a, j 01 ψ )
for a ∈ Rn and φ ∈ Γa (Rn × F) , where ψ ∈ Γ0 (Rn × F) is the section defined by
ψ = (Ta −1 × idF ) ◦ φ ◦ Ta .
PROOF : Follows from the proof of Proposition 12.40, using the identity as a local
trivialization over Rn in π : Rn × F → Rn , and as a coordinate map on Rn .
12.42 Proposition Let ( f˜, f ) be a bundle map between the bundles π : E → M

and πe : Ee → M, e where f : M → M e is a diffeomorphism. Then we have a smooth map
j 1 f˜ : J 1 E → J 1 Ee , called the 1-jet prolongation of f˜ , defined by
j 1 f˜ ( j 1p φ ) = j 1f (p) ( f˜ ◦ φ ◦ f −1 )
for p ∈ M and φ ∈ Γ p (E) . We have that ( j 1 f˜, f˜) is an affine bundle map be-
tween the affine bundles π1,0 : J 1 E → E and πe1,0 : J 1 Ee → Ee . It is modelled on
the bundle map (λ , f˜) between the vector bundles π ′′′ : Λ 1 (π ∗ (T M) ;V ) → E
e ; Ve ) → Ee , where λ (F) = f˜∗u ◦ F ◦ f ∗−1
and πe′′′ : Λ 1 (π ∗ (T M) p for u ∈ E , F ∈
1 ∗ 1 ˜
Λ (π (T M) ;V ) u and p = π (u) . We also have that ( j f , f ) is a bundle map between
the bundles π1 : J 1 E → M and πe1 : J 1 Ee → Me.
PROOF : Let u 0 be a point on E , and consider local charts (z,W ) and (e e ) around
z, W
˜
u 0 and f (u 0 ), respectively, of the form described in Proposition 12.3. Using the same
notation as in the proof of Proposition 12.36, we have that
τ z̃ ◦ j 1 f˜ ◦ τz−1 (u, G) = ( f˜(u), {D1 (e

z2 ◦ f˜ ◦ z−1 )(z(u))
(1)
+ D2 (ez2 ◦ f˜ ◦ z−1 )(z(u)) ◦ G } ◦ D(x ◦ f −1 ◦ xe−1 )(e z1 ◦ f˜ (u)))
and
τ ′z̃ ◦ λ ◦ τz′ −1 (u, G) =
(2)
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ G ◦ D(x ◦ f −1 ◦ xe−1 )(e
( f˜(u), D2 (e z1 ◦ f˜ (u)))
e ) and G ∈ L(Rn , Rm ) . It now follows from (1) with G = τz,u ( j 1p φ +
for u ∈ W ∩ f˜ −1 (W
′ (F) that we have
F) and G = τz,u ( j 1p φ ) , and (2) with G = τz,u
j 1 f˜ ( j 1p φ + F) = j 1 f˜ ( j 1p φ ) + λ (F)
e ) , j 1p φ ∈ J 1 E u and F ∈ Λ 1 (π ∗ (T M) ;V ) u , showing that j 1 f˜ is an

for u ∈ W ∩ f˜ −1 (W
affine map with linear part λ on each fibre. We see that ( j 1 f˜, f˜) is a bundle map since
πe1,0 ◦ j 1 f˜ ( j 1p φ ) = πe1,0 ( j 1f (p) ( f˜ ◦ φ ◦ f −1 )) = { f˜ ◦ φ ◦ f −1 } ◦ f (p)
= f˜ ◦ φ ( p) = f˜ ◦ π1,0 ( j 1p φ )
for p ∈ M and φ ∈ Γ p (E) , which shows that
πe1,0 ◦ j 1 f˜ = f˜ ◦ π1,0 .
From this it also follows that ( j 1 f˜, f ) is a bundle map since
πe1 ◦ j 1 f˜ = πe ◦ πe1,0 ◦ j 1 f˜ = πe ◦ f˜ ◦ π1,0 = f ◦ π ◦ π1,0 = f ◦ π1 .

JET BUNDLES 551
12.43 Proposition
(1) Suppose that ( f˜, f ) is a bundle map from the bundle π : E → M to the bundle
π ′ : E ′ → M ′ , and that (g̃, g) is a bundle map from the bundle π ′ : E ′ → M ′ to the
bundle π ′′ : E ′′ → M ′′ , where f : M → M ′ and g : M ′ → M ′′ are diffeomorphisms.
Then we have that
j 1(g̃ ◦ f˜) = j 1g̃ ◦ j 1 f˜ .
(2) Let (idE , idM ) be the bundle map from the bundle π : E → M to itself, where
idM : M → M and idE : E → E are the identity maps on M and E, respectively.
Then we have that
j 1(idE ) = idJ 1 E .
(3) If ( f˜, f ) is an equivalence from the bundle π : E → M to the bundle π ′ : E ′ → M ′ ,

then j 1 f˜ : J 1 E → J 1 E ′ is a diffeomorphism, and we have that
( j 1 f˜) −1 = j 1( f˜ −1 ) .
PROOF : (1) We have that
j 1(g̃ ◦ f˜) ( j 1p φ ) = j 1g( f (p)) (g̃ ◦ f˜ ◦ φ ◦ f −1 ◦ g−1 )
= j 1g̃ ( j 1f (p) ( f˜ ◦ φ ◦ f −1 )) = j 1g̃ ◦ j 1 f˜ ( j 1p φ )
for p ∈ M and φ ∈ Γ p (E) .

(2) We have that
j 1(idE ) ( j 1p φ ) = j 1idM (p) (idE ◦ φ ◦ idM−1 ) = j 1p φ
for p ∈ M and φ ∈ Γ p (E) .

(3) By (1) and (2) we have that
j 1 f˜ ◦ j 1( f˜ −1 ) = j 1( f˜ ◦ f˜ −1 ) = j 1(idE ) = idJ 1 E
and
j 1( f˜ −1 ) ◦ j 1 f˜ = j 1( f˜ −1 ◦ f˜) = j 1(idE ) = idJ 1 E .
12.44 Remark A section φ : M → E in the bundle π : E → M is a bundle map

over M between the bundles idM : M → M and π : E → M. Identifying J 1M with M,
the 1-jet prolongation j 1φ : M → J 1 E of φ defined in Proposition 12.42 is given by
j 1φ (p) = j 1p φ
for p ∈ M .
Let ( f˜, f ) be a bundle map between the bundles π : E → M and πe : Ee → M, e
where f : M → M e is a diffeomorphism. If φ : M → E is a section of π , then the
e → Ee of πe is called the push-forward of φ by ( f˜, f ), and it is
section f˜ ◦ φ ◦ f −1 : M
˜
denoted by f ∗ (φ ). By Proposition 12.42 we have that
j 1 f˜ ◦ j 1φ = j 1 ( f˜∗ (φ )) ◦ f
which implies that

( j 1 f˜) ∗ ( j 1φ ) = j 1 ( f˜∗ (φ )) .
HOLONOMIC TANGENT VECTORS

12.45 Lemma Given a section φ ∈ Γ p (E) with φ (p) = u in the bundle π : E → M
over M with fibre F, the tangent space Tu E can be written as a direct sum
Tu E = Hu ⊕ Vu
where Hu = φ ∗ p (Tp M) and Vu = ker π ∗ u . The projections hu : Tu E → Tu E and vu :

Tu E → Tu E on the subspaces Hu and Vu are given by hu = φ ∗ p ◦ π ∗ u and vu = id Tu E −
hu .
PROOF : For each w ∈ Tu E we have that
w = φ ∗ p ◦ π ∗ u (w) + {w − φ ∗ p ◦ π ∗ u (w)}
where φ ∗ p ◦ π ∗ u (w) ∈ Hu and w − φ ∗ p ◦ π ∗ u (w) ∈ Vu .

Suppose that w ∈ Hu ∩ Vu . Since w ∈ Hu , we have that w = φ ∗ p (v) for a vector
v ∈ Tp M. Now using that w ∈ Vu , it follows that v = π ∗ u (w) = 0 which shows that
w = 0.

πe ′ : π1,0∗ (T E) → J 1 E be the pullback of the tangent bundle π ′ : T E → E by the
target projection π1,0 : J 1 E → E. Then we have two subbundles πe ′′ : H e → J 1 E and

πe ′′′ : Ve → J 1 E of πe ′ so that
∗
π1,0 eσ ⊕ V
(T E) σ = H eσ
e σ = {σ } × φ ∗(Tp M) and V
for σ ∈ J 1 E , where H eσ = {σ } × Vu when σ = j 1p φ and
φ (p) = u . They are called the bundles of holonomic and vertical tangent vectors,
respectively. The maps e ∗ (T E) → π ∗ (T E) and v
h : π1,0 1,0
∗ (T E) → π ∗ (T E) which
e : π1,0 1,0
are projections on Heσ and V eσ in each fibre π (T E) σ , are bundle maps over J 1 E
∗
1,0
called the holonomic and vertical projection.
JET BUNDLES 553
PROOF : Using the same notation as in the proof of Proposition 12.36, we have that
pr1 ◦ tz,u = tx,p ◦ π ∗ u
where pr1 : Rn × Rm → Rn is the projection on the first factor and u = π (p) .

e)→W
We have a local trivialization τez : πe ′−1 (W e × Rn × Rm in the vector bundle
e ′ ∗ 1 e = π −1 (W )
π : π1,0 (T E) → J E associated with the local chart (z,W ) on E , where W 1,0
and
τez ( j 1p φ , w) = ( j 1p φ , tx,p ◦ π ∗ u (w), pr2 ◦ tz,u (w − φ ∗ p ◦ π ∗ u (w)))
= ( j 1p φ , pr1 ◦ tz,u (w), { pr2 − τz,u ( j 1p φ ) ◦ pr1 } ◦ tz,u (w))
for p ∈ O , φ ∈ Γ p (E) with φ (p) = u ∈ W and w ∈ Tu E , since
(τz,u × id Rn × Rm ) ◦ τez ◦ (τz,u × tz,u) −1 (G, a, b) = (G, a, b − G(a))
for G ∈ L(Rn , Rm ) and (a, b) ∈ Rn × Rm . Using that w − φ ∗ p ◦ π ∗ u (w) ∈ Vu so that
pr1 ◦ tz,u (w − φ ∗ p ◦ π ∗ u (w)) = 0 ,
we now have that

e ) ∩ H)
τez (πe ′−1 (W e =W
e × Rn × {0}
and
e)∩ V
τez (πe ′−1 (W e) = W
e × {0} × Rm ,
e → J 1 E and πe ′′′ : Ve → J 1 E are subbundles of πe ′ :

completing the proof that πe ′′ : H
∗ 1
π1,0 (T E) → J E satisfying the assertions in the proposition. The last part of the
proposition now follows from Proposition 2.64.
∗ (T E) where σ = j 1 φ , φ (p) = u and

12.47 Remark Suppose that ( σ , w) ∈ π1,0 p
w ∈ Tu E is the tangent vector given by

n m
w = ∑ a ∂ i + ∑ b α ∂ α
i
∂x ∂y
i=1 u α =1 u
in the local coordinates described in Remark 12.4. Then we have that

!!
n
∂

e
h ( σ , w) = ( σ , φ ∗ p ◦ π ∗ u (w)) = σ , ∑ ai φ ∗ p i
i=1
∂x
p
( )!
n
∂
m
∂

α
= σ , ∑ ai i + ∑ y i (σ ) α
∂x ∂y
i=1 u α =1 u
and
( ) !
m n
ve( σ , w) = ( σ , w − φ ∗ p ◦ π ∗ u (w)) = σ, ∑ α
b − ∑ ai yαi (σ ) ∂ .
∂ yα
α =1 i=1 u
12.48 Definition Let π : E → M be a bundle over M n with fibre F m . A vector

field X : J 1 E → T E along π1,0 is called a total derivative if the section Xe : J 1 E →
∗ (T E) determined by X belongs to H,
π1,0 e i.e., if
X( j 1p φ ) ∈ φ ∗ (Tp M)
for p ∈ M and φ ∈ Γ p (E) . Let (z,W ) be a local chart on E with coordinate functions
−1
described in Remark 12.4. The vector field Xi : π1,0 (W ) → T E along π1,0 defined by
!
∂
m
Xi (σ ) = φ ∗ = ∂ i + ∑ yαi (σ ) ∂ α
∂ xi ∂x α =1
∂y
p u u
−1
for σ ∈ π1,0 (W ) where σ = j 1p φ and φ (p) = u , is called the i-th coordinate total
d
derivative and is denoted by dx i for i = 1, ... , n . The tangent vector Xi (σ ) in Tu E is

d

denoted by dxi , and we have that
σ
!
d
∂
∂( f ◦ φ)
( f ) = ∂ i ( f ◦ φ ) =
df
(σ ) = dxi ( f ) = φ∗ (p)
dxi ∂ xi ∂x ∂ xi
σ p p
when σ = j 1p φ , φ (p) = u and f ∈ Fu (E) , so that
df ∂( f ◦ φ)
◦ j 1φ = .
dxi ∂ xi
12.49 Let π : E → M be a bundle over M n with fibre F m , and let π ′ : T E → E

and π1′ : T J 1 E → J 1 E be the tangent bundles of E and J 1 E . By 12.13 the map
(π1′ , π1,0 ∗ ) : T J 1 E → J 1 E × T E induces a bundle map ρ 1 : T J 1 E → π1,0
∗ (T E) over
1
J E , as we see from the commutative diagram
π1,0 ∗✲
T J 1E TE
π1′ π′
❄ ❄
π1,0 ✲
J 1E E
JET BUNDLES 555
−1
Using the local charts (z,W ) and ((z × β ) ◦ τz , π1,0 (W )) on E and J 1 E with coordi-
nate functions described in Remarks 12.4 and 12.37, we have that
! ! ! !
∂

ρ1 = σ , ∂ i , ρ1 ∂ = σ , ∂ α
∂ xi ∂x ∂ yα ∂y
σ u σ u
!
∂

and ρ1 ∂ yαi = (σ , 0)
σ
−1
for σ ∈ π1,0 (W ) where π1,0 (σ ) = u .
e ∈ Ω 1 (J 1 E; π1,0
Let ω ∗ (T E)) be the bundle-valued 1-form determined by the com-
posed bundle map b h=e ∗ (T E) over J 1 E as described in Proposi-

h ◦ ρ 1 : T J 1 E → π1,0
tion 5.30. Then it follows from Proposition 5.44 and Remark 12.47 that
n
e | π −1 (W ) = ∑ dxi ⊗ d i
φ1 ◦ ω (1)
1,0 dx
i=1
where φ 1 : Λ 1 (T J 1 E ; π1,0∗ (T E)) → T 1 (T J 1 E, π ∗ (T E)) is the equivalence over

1 1,0
1
J E defined as in 5.25.
′ :V
Let π1,0 1 1
1,0 → J E be the vertical tangent bundle of the bundle π1,0 : J E → E .
The composed map vb = ve◦ ρ 1 : T J 1 E → π1,0 ∗ (T E) is a bundle map over J 1 E with
dim ker(b 1
vσ ) = n(1 + m) for every σ ∈ J E. By Proposition 2.65 we therefore have
that C = ker(b v) is a distribution on J 1 E, called the Cartan distribution , which is
given by Cσ = ρ 1−1 (H eσ ) = ( j 1φ ) ∗ (Tp M) ⊕ V1,0 σ for σ ∈ J 1 E where σ = j 1p φ .
By Proposition 12.36 we know that π1,0 : J 1 E → E is an affine bundle mod-
elled on the vector bundle π ′′′ : Λ 1 (π ∗ (T M) ;V ) → E over E , where π ′′ : V → E
is the vertical tangent bundle of π : E → M . According to Proposition 12.19 we
therefore have an equivalence η : V1,0 → π1,0 ∗ (Λ 1 (π ∗ (T M) ;V )) over J 1 E from the
vertical tangent bundle π1,0 ′ :V 1 1

1,0 → J E of π1,0 : J E → E to the vector bundle
πe : π1,0 (Λ (π (T M) ;V )) → J E induced from π : Λ 1 (π ∗ (T M) ;V ) → E by π1,0 .
′′′ ∗ 1 ∗ 1 ′′′
∗ (V )) → T 1 (π ∗ (T M), π ∗ (V )) is the equivalence over J 1 E
If φ 2 : Λ 1 (π1∗ (T M) ; π1,0 1 1 1,0
defined as in 5.25, we have that
! !

φ2◦ η ∂ = (σ , dxi (p)) ⊗ σ , ∂ α (2)
∂ yαi ∂y
σ u
−1
for i = 1, 2, ... , n , α = 1, 2, ... , m and σ ∈ π1,0 (W ) , where σ = j 1p φ and φ (p) = u .
Indeed we have that
ye αi = β b(i,α ) ◦ τz,u
so that !

t β ◦ τz,u , σ ∂ = e b(i,α )
∂ ye αi
σ
by Remark 2.82, where ye αi = yαi |π −1 (u) , β : L(Rn , Rm ) → R n m is the linear isomor-

1,0
phism defined in Corollary 5.29, b : In × Im → I n m is the bijection defined in Remark
4.27, and E = {e1 , ..., enm } is the standard basis for Rnm . Hence it follows that
!

η ∂ = (σ , F iα ) ,
∂ yαi
σ
where F iα : Tp M → Vu is the linear map given by

F iα ∂ = δi j ∂
∂xj ∂ yα
p u
for j = 1, ... , n, which implies formula (2).

Let ω ∈ Ω 1 (M) be a 1-form on M . Then we have a bundle-valued 1-form S ω ∈
1
Ω (J 1 E ;V1,0 ) on J 1 E given by
S ω (σ )(ξ ) = {φ 2 ◦ η } −1 {(σ , ω (p)) ⊗ vb( ξ )}
for σ ∈ J 1 E and ξ ∈ Tσ J 1 E where π1 (σ ) = p . If

n
ω | U = ∑ ai dxi ,
i=1
then it follows from Remark 12.47 that

! !

dyαi (σ ) ◦ S ω (σ ) ∂ j = ω (p) ∂ i (σ , dyα (u)) ◦ vb ∂ j
∂x ∂x ∂x
σ p σ
= −ai (p) yαj (σ )
and
! !

dyαi (σ ) ◦ S ω (σ ) ∂ = ω (p) ∂ (σ , dyα (u)) ◦ vb ∂
∂ yβ ∂x i
∂ yβ
σ p σ
= ai (p) δαβ ,
showing that
n
φ 3 ◦ S ω | π −1 (W ) = ∑ (ai ◦ π1 ) α
dy − ∑ yαj dx j
⊗ ∂ ∂yα (3)
1,0 i
i, α j=1
where φ 3 : Λ 1 (T J 1 E ; V1,0) → T 11 (T J 1 E,V1,0 ) is the equivalence over J 1 E defined

as in 5.25.
Now let θ ∈ Ω 1 (J 1 E) be a 1-form on J 1 E , and consider the map Dθ : Ω 1 (M) →
1
Ω (J 1 E) given by
Dθ (ω ) = iS ω θ
JET BUNDLES 557
for ω ∈ Ω 1 (M) . We see that it is linear over F (M) since
Dθ ( f ω ) = iπ1∗ ( f ) S ω θ = π1∗ ( f ) Dθ (ω )
for f ∈ F (M) . By Proposition 12.29 there is therefore a unique bundle-valued 1-

form Sθ ∈ Ω 1 (J 1 E ; π1∗ (T M)) along π1 so that
D θ (ω ) = iS θ ω
for ω ∈ Ω 1 (M) . It is given by

n n
S θ (σ )(ξ ) = ∑ D θ (ω )(σ )(ξ ) ∂ i = ∑ θ (σ ) ◦ S ω i (σ )(ξ ) ∂ i
i
i=1
∂x i=1
∂x
p p
for σ ∈ J 1 E and ξ ∈ Tσ J 1 E where π1 (σ ) = p , and where ω 1 , ... , ω n are 1-forms on

M with ω i (p) = dxi (p) for i = 1, ... , n. If
θ | π −1 (W ) = ∑ ai dxi + ∑ bα dyα + ∑ cαi dyαi ,

1,0
i α i, α
then
n m n
∂
φ 4 ◦ S θ | π −1 (W ) = ∑ ∑ cαi dy α − ∑ yαj dx j ⊗ ◦ π1 (4)
1,0
i=1 α =1 j=1
∂ xi
where φ 4 : Λ 1 (T J 1 E ; π1∗ (T M)) → T 11 (T J 1 E, π1∗ (T M)) is the equivalence over J 1 E

defined as in 5.25.
Finally, suppose that we have an n-form ε ∈ Ω n (M) on M , and consider the map
D ε : Ω 1 (J 1 E) → Ω n (J 1 E) given by
D ε (θ ) = iS θ ε
for θ ∈ Ω 1 (J 1 E) . We see that it is linear over F (J 1 E) since
D ε ( f θ ) = i f S θ ε = f D ε (θ )
for f ∈ F (J 1 E) , and that D ε (θ ) = 0 when θ belongs to A(V1,0 ) . By Proposition

12.29 there is therefore a unique bundle-valued n-form Sε ∈ Ω n (J 1 E ;V1,0 ) on J 1 E
so that
D ε (θ ) = iS ε θ
for θ ∈ Ω 1 (J 1 E) . It is given by

S ε (σ )(ξ1 , ... , ξn ) = ∑ D ε (θ iα )(σ )(ξ1 , ... , ξn ) ∂ ∂yα
i, α i
σ

∂

= ∑ iS θ α ε (σ )(ξ1 , ... , ξn ) ∂ yα
i, α i i
σ
for σ ∈ J 1 E and ξ1 , ... , ξn ∈ Tσ J 1 E , where θ iα are 1-forms on J 1 E with θ iα (σ ) =

dyαi (σ ) for i = 1, ... , n and α = 1, ... , m. If
ε | U = a dx1 ∧ ... ∧ dxn ,
then it follows from Proposition 12.23 that

n
iS θ α ε | π −1 (W ) =
i 1,0
∑ (−1) k+1 iS θ iα (dxk ) ∧ π1∗(a dx1 ∧ · · · ∧ dxk−1 ∧ dxk+1 ∧ · · · ∧ dxn )
k=1
n
= (a ◦ π1 ) dx1 ∧ · · · ∧ dxi−1 ∧ dy α − ∑ yαj dx j ∧ dxi+1 ∧ · · · ∧ dxn
j=1
so that
φ 5 ◦ S ε | π −1 (W ) = (5)
1,0
n
∑ (a ◦ π1) dx ∧ · · · ∧ dx ∧ dy − ∑ y j dx ∧ dx ∧ · · · ∧ dx ⊗ ∂∂yα
1 i−1 α α j i+1 n
i
i, α j=1
where φ 5 : Λ n (T J 1 E ;V1,0 ) → Λ n1 (T J 1 E ,V1,0 ) is the equivalence over J 1 E defined

as in 5.25. The bundle-valued n-form Sε will be important in our study of the calculus
of variations in 12.90.
CONTACT COTANGENT VECTORS
12.50 Lemma Given a section φ ∈ Γ p (E) with φ (p) = u in the bundle π : E → M

over M with fibre F, the cotangent space Tu∗ E can be written as a direct sum
Tu∗ E = A(Hu ) ⊕ A(Vu )
where A(Hu ) = ker(φ ∗ p ) ∗ and A(Vu ) = α (π ∗ (T ∗ M) u ) .

PROOF : We have that A(Hu ) = ker(φ ∗ p ) ∗ since
λ (Hu ) = (φ ∗ p ) ∗ (λ )(Tp M)
for λ ∈ Tu∗ E . The proposition therefore follows from Proposition 4.20 and Remark
12.15.

πe ′ : π1,0
∗ (T ∗ E) → J 1 E be the pullback of the cotangent bundle π ′ : T ∗ E → E by the
JET BUNDLES 559
e → J 1E
target projection π1,0 : J 1 E → E. Then we have two subbundles πe ′′ : A(H)
and πe ′′′ : A(Ve ) → J 1 E of πe ′ so that
∗
π1,0 e σ ⊕ A(Ve ) σ
(T ∗ E) σ = A(H)
e σ = {σ } × ker(φ ∗ p ) ∗ and A(Ve ) σ = {σ } × α (π ∗(T ∗ M) u )

for σ ∈ J 1 E , where A(H)
when σ = j 1p φ and φ (p) = u . They are called the bundles of contact and hori-
zontal cotangent vectors, respectively. The maps ve ∗ : π1,0∗ (T ∗ E) → π ∗ (T ∗ E) and
1,0
e ∗ ∗ ∗ ∗ ∗
h : π (T E) → π (T E) which are projections on A(H) e σ and A(Ve ) σ in each fibre
1,0 1,0
∗ (T ∗ E) , are bundle maps over J 1 E called the contact and horizontal projection .
π1,0 σ
PROOF : Follows from Propositions 2.64, 4.22, 12.46 and 12.50.
12.52 Remark Suppose that ( σ , λ ) ∈ π1,0∗ (T ∗ E) where σ = j 1 φ , φ (p) = u and

p
∗
λ ∈ Tu E is the cotangent vector given by
n m
λ = ∑ ai dxi (u) + ∑ b α dy α (u)
i=1 α =1
in the local coordinates described in Remark 12.4. Then it follows from Remark
12.47 that
ve ∗ ( σ , λ ) = ( σ , λ ◦ vu )
! ! !
m n
= σ, ∑ λ ◦ vu ∂ dy α (u) + ∑ λ ◦ vu ∂ i
dx (u)
∂ yα ∂ xi
α =1 u i=1 u
! ! !
m n m
= σ, ∑ λ ∂ dy α (u) − ∑ λ ∑ yαi (σ ) ∂ dx (u)
i
∂ yα ∂ yα
α =1 u i=1 α =1 u
( )!
m n
α
= σ, ∑ bα dy (u) − ∑ yαi (σ ) i
dx (u)
α =1 i=1
and
e
h ∗ ( σ , λ ) = ( σ , λ ◦ hu )
! ! !
n
∂
m
∂

= σ , ∑ λ ◦ hu i dxi (u) + ∑ λ ◦ hu dy α (u)
∂x ∂ yα
i=1 u α =1 u
! !
n
∂
m
∂

α i
= σ,∑ λ i + ∑ y i (σ ) dx (u)
∂x ∂ yα
i=1 u α =1 u
( ) !
n m
= σ,∑ ai + ∑ b α yαi (σ ) i
dx (u) .
i=1 α =1
12.53 Remark By Proposition 12.14 the vector bundle πe ′ : π1,0 ∗ (T ∗ E) → J 1 E

′ ∗ 1
can be identified with a subbundle of the cotangent bundle π : T J E → J 1 E by
∗ (T ∗ E) → T ∗ J 1 E over J 1 E defined by
means of the bundle map α 1 : π1,0
α 1 (σ , λ ) = λ ◦ π1,0 ∗ σ
for σ ∈ J 1 E and λ ∈ Tu∗ E where u = π1,0 (σ ) , which is an embedding. We have that
α 1 (σ , dxi (u)) = dxi (σ ) and α 1 (σ , dy α (u)) = dy α (σ ) .
By Remark 12.52 we see that

( )
n n
dy 1 − ∑ y1i dxi , ... , dy m − ∑ ymi dxi and {dx1 , ... , dxn }
i=1 i=1
e → J 1 E and π ′′′ : α 1 (A(Ve )) → J 1 E,

are local bases for the subbundles π ′′ : α 1 (A(H))
respectively.
JET FIELDS AND CONNECTIONS

12.54 Definition A connection in a bundle π : E → M is a distribution H on E
such that
Tu E = H u ⊕ Vu
for every u ∈ E. The subspace H u of Tu E is called the horizontal subspace at u, and
the vectors in H u are called horizontal tangent vectors at u.
By Proposition 2.64 the maps h : T E → T E and v : T E → T E which are projec-
tions on H u and Vu in each fibre Tu E, are bundle maps over E called the horizontal
and vertical projection determined by H. The bundle-valued 1-form ω ∈ Ω 1 (E; T E)
JET BUNDLES 561
determined by the horizontal projection h as described in Proposition 5.30, is called

the connection form of H.
12.55 Let π : E → M n be a bundle with fibre F m , and let Γ : E → J 1 E be a jet

field on E. Then we have a commutative diagram
e
Γ ve πe1,0 ✲
TE ✲ ∗ (T E)
π1,0 ✲ π ∗ (T E) TE
1,0
π′ πe′ πe′ π′
❄ ❄ ❄ ❄
Γ✲ idJ 1 E ✲ π1,0 ✲
1 1
E J E J E E
where Γ e = ( Γ ◦ π ′ , idT E ) is the map defined by Γ(w)

e = ( Γ(u), w) for u ∈ E and
w ∈ Tu E . We see that v = πe1,0 ◦ ve◦ Γ e : T E → T E is a bundle map over E with
dim ker(vu ) = n for every u ∈ E. By Proposition 2.65 we therefore have that H =
ker(v) is a connection in E, called the connection determined by the jet field Γ, which
is given by Hu = φ ∗ (Tp M) for u ∈ E where Γ(u) = j 1p φ .
Every connection H in E is determined by a unique jet field Γ. To define Γ(u) for
a point u ∈ E, let (z,W ) be a local chart around u as described in Proposition 12.3, and
let hz : W × Rn+m → W × Rn+m be the bundle map over W given by hz = tz ◦ h ◦ tz−1 ,
where h is the horizontal projection determined by H . Now we define Γ(u) = j 1p φ
for the section φ ∈ Γ p (E) given by
φ = z−1 ◦ Tz(u) ◦ hz,u ◦ i1 ◦ Tx(p)

−1
◦x
where π (u) = p , Tx(p) and Tz(u) are translations by x(p) and z(u) in Rn and Rn+m ,
respectively, and i1 : Rn → Rn × Rm is the injection given by i1 (a) = (a, 0) for a ∈ Rn .
−1
If τz : π1,0 (W ) → W × L(Rn , Rm ) is the equivalence over W defined in the proof of
Proposition 12.36, we have that
τz ◦ Γ(u) = (φ (p), D(z2 ◦ φ ◦ x−1 )(x(p)) = (u, pr2 ◦ hz,u ◦ i1 )
where pr2 : Rn × Rm → Rm is the projection on the second factor, showing that Γ is

a jet field on E. We also have that
−1
tz,u ◦ φ ∗ p ◦ tx,p = D(z ◦ φ ◦ x−1 )(x(p)) = hz,u ◦ i1 ,
−1 −1
φ ∗ (Tp M) = tz,u ◦ hz,u (Rn × {0}) = hu ◦ tz,u (Rn × {0}) = Hu
since
−1 −1 −1
Tu E = tz,u (Rn × {0}) ⊕ tz,u ({0} × Rm) = tz,u (Rn × {0}) ⊕ Vu .
This shows that H is the connection determined by the jet field Γ, and that Γ is
uniquely determined by H .
12.56 Remark If ω is the connection form of H , then it follows from Proposition

5.44 and Remark 12.47 that
( )
n m
∂
φ 1 ◦ ω |W = ∑ dxi ⊗ + ∑ (yαi ◦ Γ) ∂ α
i=1
∂ xi α =1
∂y
in the local coordinates described in Remarks 12.4 and 12.37, where φ 1 :

Λ 1 (T E ; T E) → T 11 (E) is the equivalence over E defined as in 5.25.
12.57 Definition Let H be a connection in a bundle π : E → M , and let

ω ∈ Ω 1 (E; T E) be the connection form of H. Then the bundle-valued 2-form
Ω ∈ Ω 2 (E; T E) given by
Ω = − 21 [ ω , ω ]
is called the curvature form of H. The connection H is said to be flat if Ω = 0 .
12.58 Proposition Let H be a connection with curvature form Ω in a bundle

π : E → M , and let h and v be the horizontal and vertical projection determined by
H. Then we have that
Ω(X,Y ) = − v ◦ [ h ◦ X, h ◦ Y ]
for every pair of vector fields X and Y on E.
PROOF : Since V is an integrable distribution on E, we have that
Ω(X, v ◦ Y ) = ω ([ ω (X), v ◦ Y ]) − ω ([ X, v ◦ Y ]) = − ω ([ v ◦ X, v ◦ Y ]) = 0
by Corollary 12.34. This implies that
Ω(X,Y ) = Ω(h ◦ X, h ◦ Y ) = ω ([ h ◦ X, h ◦ Y ]) − [ h ◦ X, h ◦ Y ]
= − v ◦ [ h ◦ X, h ◦ Y ] .
EQUIVARIANT JET FIELDS

12.59 Definition By an operation or an action of a Lie group G on a bundle
π : E → M on the right we mean a pair (µ e , µ ) of operations µ
e : E × G → E and
µ : M × G → M of G on E and M, respectively, such that the corresponding pair
(Reg , Rg ) of right translations forms a bundle map for each g ∈ G. This implies that
JET BUNDLES 563
e and µ of G on
the projection π : E → M is equivariant with respect to the actions µ
E and M .
12.60 Proposition Let π : P → M be a principal G-bundle over M n , where µ :

P × G → P is the action of G m on P on the right. Then we have an operation (µe, µ)
of G on the jet bundle π1,0 : J 1 P → P such that the right translations Rg and Reg
e , respectively, are related by
corresponding to µ and µ
Reg = j 1Rg
for g ∈ G.
PROOF : By Proposition 12.42 we know that (R eg , Rg ) is a bundle map for each g ∈
G, and it only remains to prove that µ e is smooth at every point (σ , h) ∈ J 1 P × G.
Suppose that (t, π −1 (U)) is a local trivialization around the point u′ = π1,0 (σ ) in P
with t(u′ ) = (q, h′ ) . Let (x, O) be a local chart around q on M with O ⊂ U , and let
(y′′ , Q′′ ), (y′ , Q′ ) and (y, Q) be local charts around h′ h , h′ and h on G with Q′ Q ⊂
Q′′ . Then (z′ ,W ′ ) and (z,W ) , where W ′ = t −1 (O × Q′′ ) , W = t −1 (O × Q′ ) , z′ =
(x × y′′ ) ◦ t and z = (x × y′ ) ◦ t , are local charts around u′ h and u′ on P. The map
f : z(W ) × y(Q) → Rn+m defined by f = z′ ◦ µ ◦ (z × y)−1 is smooth, and we have
that
f (a, b) = z′ ◦ R y−1 (b) ◦ z−1 (a)
for a ∈ z(W ) and b ∈ y(Q).
−1 −1
If τz′ : π1,0 (W ′ ) → W ′ × L(Rn , Rm ) and τz : π1,0 (W ) → W × L(Rn , Rm ) are the
equivalences over W and W associated with the local charts (z′ ,W ′ ) and (z,W ), re-
′
spectively, we have that
e ◦ (τz × id Q )−1 (u, G, g) = τz′ ◦ Reg ( j 1p φ ) = τz′ ( j 1p (Rg ◦ φ ))

τz′ ◦ µ
= (Rg ◦ φ (p), D(z′2 ◦ Rg ◦ z−1 )(z(u)) ◦ D(z ◦ φ ◦ x−1 )(x(p)))
= (Rg (u), D(z′2 ◦ Rg ◦ z−1 )(z(u)) ◦ (id Rn , G))
for g ∈ Q and (u, G) ∈ W × L(Rn , Rm ) where τz−1 (u, G) = j 1p φ . This shows that
(z′ × id) ◦ τz′ ◦ µ

e ◦ (τz × id Q )−1 ◦ (z × id × y)−1 (a, G, b)
= (z′ ◦ Ry−1 (b) ◦ z−1 (a), D(z′2 ◦ Ry−1 (b) ◦ z−1 )(a) ◦ (id Rn , G))
= ( f (a, b), D1 f (a, b) ◦ (id Rn , G))
for (a, G, b) ∈ z(W ) × L(Rn , Rm ) × y(Q) , which completes the proof of the proposi-
tion.
12.61 Proposition Let π : P → M be a principal G-bundle over a smooth manifold

M n , and let H be the connection determined by a jet field Γ : P → J 1 P on P. Then Γ
is equivariant with respect to the actions of G on P and J 1 P described in Proposition

12.60 if and only if
H ug = (R g )∗ H u
for every u ∈ P and g ∈ G. A connection H satisfying this condition is called a prin-
cipal connection .
PROOF : Let u ∈ P and g ∈ G, and suppose that Γ(u) = j 1p φ and Γ(ug) = j 1p ψ so
that Hu = φ ∗ (Tp M) and Hug = ψ ∗ (Tp M). Then the condition in the proposition is
equivalent to the assertion that
ψ ∗ p = (R g ◦ φ ) ∗ p ,
which is also equivalent to
Γ ◦ Rg (u) = j 1p ψ = j 1p (R g ◦ φ ) = j 1Rg ◦ Γ(u) .
As u ∈ P and g ∈ G was arbitrary, this completes the proof of the proposition.
12.62 Proposition Let π : P → M be a principal G-bundle, and let π ′ : E →

M be the fibre bundle associated with P with fibre F obtained from a Lie group
homomorphism ψ : G → G′ into a Lie group G′ which acts on the smooth manifold
F effectively on the left. If Γ is an equivariant jet field on P, then there is a unique jet
field Γ′ on E so that the diagram
j 1 ν v✲
J 1P J 1E
✻ ✻′
Γ Γ
νv ✲
P E
is commutative for every v ∈ F, where ν v : P → E is the smooth map defined in

10.28. The connections H and H ′ determined by Γ and Γ′ are related by
Hw′ = ν v ∗ (Hu )
for u ∈ P and v ∈ F, where w = ν v (u).

PROOF : Let w ∈ E p with p ∈ M . For each u ∈ Pp there is a unique v ∈ F so that
ν v (u) = w. If ν v′ (u′ ) = w with u′ ∈ Pp and v′ ∈ F, then u′ = u g and v′ = ψ (g−1 ) v
for a g ∈ G. Using that the jet field Γ is equivariant, we have that
j 1 ν v′ ◦ Γ(u′ ) = j 1 ν v′ ◦ Γ ◦ Rg (u) = j 1 ν v′ ◦ j 1Rg ◦ Γ(u) = j 1 ν v ◦ Γ(u) ,

JET BUNDLES 565
showing that Γ′ (w) is well defined for every w ∈ E. We also have that
′
π1,0 ◦ Γ′ ◦ ν v = π1,0
′
◦ j 1 ν v ◦ Γ = ν v ◦ π1,0 ◦ Γ = ν v ,
so it only remains to show that Γ′ is smooth.

Let (t, π −1 (U)) be a local trivialization in P, where U is the coordinate neigh-
bourhood of a local chart (x,U) around a point p 0 ∈ M , and let (t ′ , π ′ −1 ( U)) be the
local trivialization in E associated with (t, π −1 (U)) as described in 10.28. We then
have that
t ′ ◦ ν v ◦ t −1 (p, g) = (p, µ (g, v))
for p ∈ U, g ∈ G and v ∈ F, where µ : G × F → F is the group action of G on F
defined by µ (g, v) = ψ (g) v for g ∈ G and v ∈ F. Given a point v 0 ∈ F, we choose
local charts (y, Q) and (y′ , Q′ ) around e and v0 on G and F, respectively, obtaining lo-
cal charts (z,W ) and (z′ ,W ′ ) on P and E where W = t −1 (U × Q), W ′ = t ′ −1 (U × Q′ ),
z = (x × y) ◦ t and z′ = (x × y′ ) ◦ t ′ . Using the same notation as in the proof of Propo-
sition 12.42, the above formula implies that
−1
τz′ ,2 ◦ Γ′ ◦ t ′ (p, v) = τz′ ,2 ◦ Γ′ ◦ ν v ◦ t −1 (p, e) = τz′ ,2 ◦ j 1 ν v ◦ Γ ◦ t −1 (p, e)
= {τz′ ,2 ◦ j 1 ν v ◦ τz−1 } ◦ {τz ◦ Γ ◦ t −1 }(p, e) = D1 (z′2 ◦ ν v ◦ z−1 )(x(p), y(e))
+ D2 (z′2 ◦ ν v ◦ z−1 )(x(p), y(e)) ◦ (τz,2 ◦ Γ ◦ t −1 (p, e))
= D1 (y′ ◦ µ ◦ (y × y′)−1 )(y(e), y′ (v)) ◦ (τz,2 ◦ Γ ◦ t −1 (p, e))
for p ∈ U and v ∈ Q′ , showing that Γ′ is a jet field on E.

Now suppose that w = ν v (u) where u ∈ P and v ∈ F, and that Γ(u) = j 1p φ . Then
it follows that
Γ′ (w) = Γ′ ◦ ν v (u) = j 1 ν v ◦ Γ(u) = j 1p (ν v ◦ φ )
so that
Hw′ = (ν v ◦ φ ) ∗ (Tp M) = ν v ∗ (Hu ) ,
thus proving that last assertion in the proposition.
12.63 Theorem Let π : P → M be a principal G-bundle, and let π ′ : E → M be

the associated vector bundle obtained from a representation ψ : G → Aut (W ) of G
on a finite dimensional vector space W . Let H ′ be the connection in E obtained from
a principal connection H in P as described in Proposition 12.62, and let v′ : T E →
T E be the vertical projection determined by H ′ . Using the canonical bundle map
e , π ′ ) from the vertical tangent bundle π ′′ : V → E to the vector bundle π ′ : E → M
(η
described in Proposition 12.19, we obtain a bundle map (K, π ′ ) from the tangent
bundle π ′′ : T E → E to π ′ : E → M where
e ◦ v′ .
K= η
The covariant derivative of a section s ∈ Γ(M; E) of π ′ is given by

∇s (p)(v) = K ◦ s∗ (v)
for p ∈ M and v ∈ Tp M.
PROOF : We use the same notation as in 10.28 and Definition 10.77. Let (t, π −1 (U))
be a local trivialization around p in P, and let (t ′ , π ′ −1 (U)) be the local trivialization
in E associated with (t, π −1 (U)). Now let u = t −1 (p, e) ∈ P so that π (u) = p. Then
there is a unique w ∈ Hu with π∗ (w) = v, and a unique ξ ∈ W with ν ξ (u) = s(p).
Since
π′ ◦ s ◦ π = π = π′ ◦ νξ ,
s∗ (v) = ν ξ ∗ (w) + {s∗ ◦ π∗ (w) − ν ξ ∗ (w) }
′
where ν ξ ∗ (w) ∈ Hs(p) and s∗ ◦ π∗ (w) − ν ξ ∗ (w) ∈ Vs(p) , so that
v′ ◦ s∗ (v) = s∗ ◦ π∗ (w) − ν ξ ∗ (w) .

From (1) in 10.28 it follows that
t ′ ◦ s ◦ π ◦ t −1 (p′ , g′ ) = (p′ , ψ (g′ ) µu−1 ′ ′ ′
′ ◦ s ◦ π (u )) = (p , ψ (g ) s ◦ t
−1 ′ ′
(p , g ))
and
t ′ ◦ ν ξ ◦ t −1 (p′ , g′ ) = t ′ ([ u′ , ξ ]) = (p′ , ψ (g′ ) ξ )
for (p′ , g′ ) ∈ U × G, where u′ = t −1 (p′ , g′ ).
Now let ie : U → U × G , i p : G → U × G , i′ξ : U → U ×W and i′p : W → U ×W
be the embeddings defined as in Proposition 2.74. Then there is a unique X ∈ g with
t∗ (w) = ie ∗ (v) + i p ∗ (X). Hence we have that
t∗′ ◦ s∗ ◦ π∗ (w) = i′ξ ∗ (v) + i′p ∗ ◦ s ∗ (w) + i′p ∗ ◦ ωξ−1 ◦ ψ∗ (X) ξ
and
t∗′ ◦ ν ξ ∗ (w) = iξ′ ∗ (v) + i′p ∗ ◦ ωξ−1 ◦ ψ∗ (X) ξ ,
which implies that
t p′ ∗ ◦ v′ ◦ s∗ (v) = s ∗ (w) .
Using the commutative diagram
t p′ ∗ s(p)
Ts(p) E p ✲ TW
ξ
ωs(p) ωξ
❄ ❄
t p′
Ep ✲ W
JET BUNDLES 567
we conclude that
K ◦ s∗ (v) = ωs(p) ◦ v′ ◦ s∗ (v) = t p′−1 ◦ ωξ ◦ s ∗ (w) = µu ◦ d s (u)(w) = ∇s (p)(v) .
SECOND ORDER JET BUNDLES

12.64 Proposition Let π : E → M be a bundle over M with fibre F. Then π∗ :
T E → T M is a bundle with fibre T F.
PROOF : Suppose that (t, π −1 (U)) is a local trivialization in E , and let pr1 : U ×
F → U and pr2 : U × F → F be the projections on the first and second factor. Then
(((pr1 ◦ t) ∗ , (pr2 ◦ t) ∗ ), π∗−1 (T M|U )) is a local trivialization in T E .
12.65 Remark We say that (t 1 , π∗−1 (T M|U )) , where t 1 = ((pr1 ◦ t) ∗ , (pr2 ◦ t) ∗ ) :

π∗−1 (T M|U ) → T M|U × T F , is the local trivialization in the bundle π∗ : T E → T M
associated with the local trivialization (t, π −1 (U)) in π : E → M . Here pr1 : U × F →
U and pr2 : U × F → F are the projections on the open set U in the base manifold M
and the fibre F.
Let (x, O) and (y, Q) be local charts on M n and F m , respectively, with O ⊂ U.
Then (z,W ) , where W = t −1 (O × Q) and z = (x × y) ◦ t , is a local chart on E , and
we let z1 : W → Rn and z2 : W → Rm be the component maps of z : W → Rn × Rm .
We also obtain local charts (x1 , T M|O ) and (y1 , T F|Q ) on T M and T F, where
x1 = (x × id Rn ) ◦ tx and y1 = (y × id Rm ) ◦ ty . In the same way as above, we have a
local chart (z1 , T E|W ) on T E, where z1 = (x1 × y1 ) ◦ t 1 , and we let z11 : T E|W → R2n
and z12 : T E|W → R2m be the component maps of z1 : T E|W → R2n × R2m .
If φ : Oe → E is a section in π : E → M , we have that
ty ◦ (pr2 ◦ t ◦ φ ) ∗ ◦ tx−1 (p, ξ ) = (pr2 ◦ t ◦ φ (p), D(z2 ◦ φ ◦ x−1 )(x(p)) ξ )
e ∩ O and ξ ∈ Rn , which implies that

for p ∈ O
−1
z12 ◦ φ ∗ ◦ x1 (a, ξ ) = (z2 ◦ φ ◦ x−1 (a), D(z2 ◦ φ ◦ x−1 )(a) ξ )
for a ∈ x(O e ∩ O) and ξ ∈ Rn .

Denoting the component maps of z12 : T E|W → Rm × Rm by z12,1 : T E|W → Rm
and z12,2 : T E|W → Rm , it follows that
−1
D1 (z12,2 ◦ φ ∗ ◦ x1 ) (a, ξ )(ζ ) = D2 (z2 ◦ φ ◦ x−1 )(a) (ξ , ζ )
e ∩ O) and ξ , ζ ∈ Rn .
for a ∈ x(O

πe ′ : π1∗ (T M) → J 1 E be the pullback of the tangent bundle π ′ : T M → M by π1 ,
and πe ′′ : π1,0
∗ (V ) → J 1 E be the pullback of the vertical bundle π ′′ : V → E by π .
1,0
Then π2,1 : J 2 E → J 1 E is an affine bundle over J 1 E modelled on the vector bundle
π ′′′ : T s 2 (π1∗ (T M) ; π1,0
∗ (V )) → J 1 E .
PROOF : Consider a local chart (z,W ) around a point u 0 on E as described in Re-

−1 −1
mark 12.65, and let W1 = π1,0 (W ) . Then we have a bijection τz2 : π2,1 (W1 ) →
2
W1 × T s (R ; R ) which is given by
n m
τz2 ( j 2p φ ) = ( j 1p φ , D2 (z2 ◦ φ ◦ x−1 )(x(p)))
for p ∈ O and φ ∈ Γ p (E) with φ (p) = u ∈ W . It follows from Remark 12.65 that τz2
is well defined since
−1
D(z12 ◦ φ ∗ ◦ x1 )(x1 (v)) = pr21 ◦ tz1 ,w ◦ (φ ∗ ) ∗ v ◦ tx−1
1 ,v
for v ∈ Tp M , where φ ∗ (v) = w and pr21 : R2n × R2m → R2m is the projection on the
second factor.
Suppose that (e e ) is another local chart on E obtained in the same way from a
z, W
e in E and local charts (e
local trivialization (t , π −1 (U))
e e and (e
x, O) e on M and F,
y, Q)
e e
respectively, with O ⊂ U . Then we have that
−1
τ 2z̃ ◦ τz2 z2 ◦ z−1 )(z(u)) + D2 (e
(σ , S) = (σ , {D1 (e z2 ◦ z−1 )(z(u)) ◦ G }
◦ D2 (x ◦ xe−1 )(e z2 ◦ z−1 )(z(u))
z1 (u)) + {D11 (e
z2 ◦ z−1 )(z(u)) ◦ (idRn × G) + D21 (e
+ D12 (e z2 ◦ z−1 )(z(u)) ◦ (G × idRn ) (1)
z2 ◦ z−1 )(z(u)) ◦ (G × G) + D2 (e
+ D22 (e z2 ◦ z−1 )(z(u)) ◦ S }
◦ (D(x ◦ xe−1 )(e
z1 (u)) × D(x ◦ xe−1 )(e
z1 (u))))
for σ ∈ W1 ∩ We1 and S ∈ T s 2 (Rn ; Rm ) , where τz (σ ) = (u, G) . Hence it follows from

Proposition 12.6 that π2,1 : J 2 E → J 1 E is a bundle over J 1 E with fibre R n(n+1) m/2,
−1
with ((id W1 × βs ) ◦ τz2 , π2,1 (W1 )) as a local trivialization for each local chart (z,W )
on E of the above form, using a fixed linear isomorphism βs : T s 2 (Rn ; Rm ) →
R n(n+1) m/2 defined in the same way as β in the proof of Proposition 5.31.
By Proposition 5.28 we have a local trivialization ((id W1 × βs ) ◦ τez2 , π ′′′ −1 ( W1 ))
in the vector bundle π ′′′ : T s 2 (π1∗ (T M) ; π1,0
∗
(V )) → J 1 E associated with the local
−1
chart (z,W ) on E , where τez2 : π ′′′ (W1 ) → W1 × T s 2 (Rn ; Rm ) is the equivalence
over W1 defined by
−1 −1
τez2 (T ) = (σ , pr2 ◦ tz,u ◦ T ◦ (tx,p × tx,p ))
for σ ∈ W1 and T ∈ T s 2 (π1∗ (T M) ; π1,0

∗ (V )) , where u = π (σ ) and p = π (u) . We
σ 1,0
also have that
−1
z2 ◦ z−1 )(z(u)) ◦ S
τez̃2 ◦ τez2 (σ , S) = (σ , D2 (e
(2)
◦ (D(x ◦ xe−1 )(e z1 (u)) × D(x ◦ xe−1 )(e
z1 (u))))
JET BUNDLES 569
e1 and S ∈ T s 2 (Rn ; Rm ) . Using this, we can define an operation of the
for σ ∈ W1 ∩ W
vector space T s 2 (π1∗ (T M) ; π1,0
∗ (V )) on the fibre J 2 E by
σ σ
τz,2 σ ( j 2p φ + T ) = τz,2 σ ( j 2p φ ) + τez,2 σ (T ) . (3)
It follows from (1) with S = τz,2 σ ( j 2p φ + T ) and S = τz,2 σ ( j 2p φ ) , (2) with S = τez,2 σ (T )
and (3) that we also have
τ 2z̃, σ ( j 2p φ + T ) = τ 2z̃, σ ( j 2p φ ) + τez̃,2 σ (T ) ,
showing that the operation on J 2 E σ given by (3) is well defined and does not depend
on the choice of local chart (z,W ) .
12.67 Remark From each local chart (z,W ) on E with coordinate functions xi
and yα defined in Remark 12.4, we obtain an induced local chart ((((z × β ) ◦ τz ) ×
−1
βs ) ◦ τz2 , π2,0 (W )) on J 2 E with coordinate functions xi , yα , yαi and yαi j defined by
∂ (yα ◦ φ )
xi ( j 2p φ ) = xi (p) , yα ( j 2p φ ) = yα ◦ φ (p) , yαi ( j 2p φ ) = (p)
∂ xi
∂ (yα ◦ φ )
and yαi j ( j 2p φ ) = (p)
∂ x j ∂ xi
−1
for j 2p φ ∈ π2,0 (W ) , i, j = 1, 2, ... , n and α = 1, 2, ... , m. It will be clear from the
context whether xi and yα denote coordinate functions on J 2 E , on J 1 E , on E or on
M and F, respectively, and whether yαi denote coordinate functions on J 2 E or on
J 1 E . The functions yαi and yαi j are called derivative coordinates .

π2 : J 2 E → M is also a bundle.
PROOF : Let (t, π −1 (U)) be a local trivialization in π : E → M , where U is the
coordinate neighbourhood of a local chart (x,U) on M , and consider the trivial
bundle π ′ : Rn × F → Rn . Since π2′ = π1′ ◦ π2,1′ is a submersion, we have that
J 02 (Rn × F) = π2′ −1 (0) is a closed submanifold of J 2 (Rn × F) .
Now let t ′′ : π2−1 (U) → U × J 02 (Rn × F) be the map given by
t ′′ ( j 2p φ ) = (p, j 02 ψ )
for p ∈ U and φ ∈ Γ p (E) , where ψ ∈ Γ0 (Rn × F) is the section defined by

−1
ψ = (Tx(p) × idF ) ◦ (x × idF ) ◦ t ◦ φ ◦ x−1 ◦ Tx(p) ,
using the translation Tx(p) : Rn → Rn by x(p) . We want to show that t ′′ is a diffeomor-

phism. Consider the local chart (z,W ) in E obtained from (t, π −1 (U)) , (x,U) and a
local chart (y, Q) on F as described in Proposition 12.3, so that W = t −1 (U × Q)
and z = (x × y) ◦ t . We also obtain the local chart (e z , Rn × Q) in Rn × F, where

e
z = idRn × y . With the same notation as in the proof of Proposition 12.66, we have
that
z2 ◦ ψ = z2 ◦ φ ◦ x−1 ◦ Tx(p) ,
e
which shows that
z2 ◦ ψ )(0) = D2 (z2 ◦ φ ◦ x−1 )(x(p)) .
D2 (e
Using the diffeomorphism t ′ : π1−1 (U) → U × J 01 (Rn × F) defined in the proof of
Proposition 12.40 and the diffeomorphism η : π2,0′ −1 ({0} × Q) → π ′ −1 ({0} × Q) ×
1,0
2 2
T s (R ; R ) induced by τ z̃ , we now have that
n m
(idM × η ) ◦ t ′′ ( j 2p φ ) = (p, j 01 ψ , D2 (e
z2 ◦ ψ )(0))
= (t ′ ( j 1p φ ), D2 (z2 ◦ φ ◦ x−1 )(x(p))) = (t ′ × id) ◦ τ 2z ( j 2p φ ) .
This completes the proof that π2 : J 2 E → M is a bundle, with (t ′′ , π2−1 (U)) as a lo-
cal trivialization for each local trivialization (t, π −1 (U)) in π : E → M of the above
form.
12.69 Proposition Let ( f˜, f ) be a bundle map between the bundles π : E → M

and πe : Ee → M, e where f : M → M e is a diffeomorphism. Then we have a smooth map
j 2 f˜ : J 2 E → J 2 Ee , called the 2-jet prolongation of f˜ , defined by
j 2 f˜ ( j 2p φ ) = j 2f (p) ( f˜ ◦ φ ◦ f −1 )
for p ∈ M and φ ∈ Γ p (E) . We have that ( j 2 f˜, j 1 f˜) is an affine bundle map between
the affine bundles π2,1 : J 2 E → J 1 E and πe2,1 : J 2 Ee → J 1 Ee . It is modelled on the
bundle map (λ , j 1 f˜) between the vector bundles π ′′′ : T s 2 (π1∗ (T M) ; π1,0
∗ (V )) → J 1 E
e ; π ∗ (Ve )) → J 1 Ee , where λ (T ) = f˜∗u ◦ T ◦ ( f ∗−1

and πe′′′ : T s 2 (π1∗ (T M) −1
p × f ∗ p ) for
1,0
σ ∈ J 1 E , T ∈ T s 2 (π1∗ (T M) ; π1,0∗ (V )) , u = π (σ ) and p = π (σ ) .
σ 1,0 1
We also have that ( j f˜, f ) is a bundle map between the bundles π2 : J 2 E → M
2
and πe2 : J 2 Ee → M e.
PROOF : Let u 0 be a point on E , and consider local charts (z,W ) and (e e ) around
z, W
u 0 and f˜(u 0 ), respectively, of the form described in Proposition 12.3. Using the same
notation as in the proof of Proposition 12.66, we have that
−1
τ 2z̃ ◦ j 2 f˜ ◦ τz2 z2 ◦ f˜ ◦ z−1 )(z(u))
(σ , S) = ( j 1 f˜(σ ), {D1 (e
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ G } ◦ D2 (x ◦ f −1 ◦ xe−1 )(e
+ D2 (e z1 ◦ f˜ (u))
+ {D11 (ez2 ◦ f˜ ◦ z−1 )(z(u)) + D12 (e
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ (idRn × G)
(1)
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ (G × idRn )
+ D21 (e
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ (G × G) + D2 (e
+ D22 (e z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ S }
◦ (D(x ◦ f −1 ◦ xe−1 )(e
z1 ◦ f˜ (u)) × D(x ◦ f −1 ◦ xe−1 )(e
z1 ◦ f˜ (u))))
JET BUNDLES 571
and
−1
z2 ◦ f˜ ◦ z−1 )(z(u)) ◦ S
τez̃2 ◦ λ ◦ τez2 (σ , S) = ( j 1 f˜(σ ), D2 (e
(2)
◦ (D(x ◦ f −1 ◦ xe−1 )(e
z1 ◦ f˜ (u)) × D(x ◦ f −1 ◦ xe−1 )(e z1 ◦ f˜ (u))))
−1 −1 ˜ −1 e
for σ ∈ π1,0 (W ) ∩ π1,0 ( f (W )) and S ∈ T s 2 (Rn ; Rm ) , where τz (σ ) = (u, G) . It
now follows from (1) with S = τz,2 σ ( j 2p φ + T ) and S = τz,2 σ ( j 2p φ ) , and (2) with S =
τez,2 σ (T ) that we have
j 2 f˜ ( j 2p φ + T ) = j 2 f˜ ( j 2p φ ) + λ (T )
−1 −1 ˜ −1 e
for σ ∈ π1,0 (W ) ∩ π1,0 ( f (W )) , j 2p φ ∈ J 2 E σ and T ∈ T s 2 (π1∗ (T M) ; π1,0
∗ (V )) ,
σ
showing that j 2 f˜ is an affine map with linear part λ on each fibre. We see that
( j 2 f˜, j 1 f˜) is a bundle map since
πe2,1 ◦ j 2 f˜ ( j 2p φ ) = πe2,1 ( j 2f (p) ( f˜ ◦ φ ◦ f −1 )) = j 1f (p) ( f˜ ◦ φ ◦ f −1 )
= j 1 f˜ ( j 1p φ ) = j 1 f˜ ◦ π2,1 ( j 2p φ )
for p ∈ M and φ ∈ Γ p (E) , which shows that
πe2,1 ◦ j 2 f˜ = j 1 f˜ ◦ π2,1 .
From this it also follows that ( j 2 f˜, f ) is a bundle map since
πe2 ◦ j 2 f˜ = πe1 ◦ πe2,1 ◦ j 2 f˜ = πe1 ◦ j 1 f˜ ◦ π2,1 = f ◦ π1 ◦ π2,1 = f ◦ π2 .
12.70 Remark A section φ : M → E in the bundle π : E → M is a bundle map

over M between the bundles idM : M → M and π : E → M. Identifying J 2M with M,
the 2-jet prolongation j 2φ : M → J 2 E of φ defined in Proposition 12.69 is given by
j 2φ (p) = j 2p φ
for p ∈ M .
Let ( f˜, f ) be a bundle map between the bundles π : E → M and πe : Ee → M,
e where
e
f : M → M is a diffeomorphism. As in Remarks 12.44 we then have that
j 2 f˜ ◦ j 2φ = j 2 ( f˜∗ (φ )) ◦ f
which implies that
( j 2 f˜) ∗ ( j 2φ ) = j 2 ( f˜∗ (φ ))
for every section φ : M → E of π .
12.71 Proposition Let π : E → M be a bundle over M n with fibre F m . Then there

is a bundle map i1,1 : J 2 E → J 1 (J 1 E) over J 1 E between the bundles π2,1 : J 2 E →
J 1 E and (π1 ) 1,0 : J 1 (J 1 E) → J 1 E , given by
i1,1 ( j 2p φ ) = j 1p ( j 1φ )
for p ∈ M and φ ∈ Γ p (E) , which is an embedding.
PROOF : Consider a local chart (z,W ) around a point u 0 on E as described in Propo-

−1
sition 12.3, and let (z̃,W1 ) be the induced local chart on J 1 E where W1 = π1,0 (W )
2 −1 2
and z̃ = (z × β ) ◦ τz . Using the equivalences τz : π2,1 (W1 ) → W1 × T s (R ; R ) and
n m
−1
τz̃ : (π1 ) 1,0 (W1 ) → W1 × L(Rn , Rm × Rnm ) over W1 , we have that
(τz × id) ◦ τz2 ( j 2p φ ) = (φ (p) , D(z2 ◦ φ ◦ x−1 )(a) , D2 (z2 ◦ φ ◦ x−1 )(a))
and
(τz × id) ◦ τz̃ ◦ i1,1 ( j 2p φ ) = (φ (p) , D(z2 ◦ φ ◦ x−1 )(a) , D( z̃2 ◦ j 1φ ◦ x−1 )(a))
where
z̃2 ◦ j 1φ ◦ x−1 (a) = (z2 ◦ φ ◦ x−1 (a) , β ◦ D(z2 ◦ φ ◦ x−1 )(a))
for a = x(p) ∈ x(O) and φ ∈ Γ p (E) with φ (p) ∈ W . Let βe : T s 2 (Rn ; Rm ) →
L(Rn , Rnm ) be the linear injection defined by
βe(T )(a) = β (Ta )
for T ∈ T s 2 (Rn ; Rm ) and a ∈ Rn , where Ta ∈ L(Rn , Rm ) is the linear map given by

Ta (b) = T (a, b) for b ∈ Rn . Then we have that
(τz × id) ◦ τz̃ ◦ i1,1 ◦ (τz2 ) −1 ◦ (τz × id) −1 (u, G, T ) = (u, G, G, βe(T ))
for u ∈ E , G ∈ L(Rn , Rm ) and T ∈ T s 2 (Rn ; Rm ) , which shows that i1,1 is a well-

defined embedding and a bundle map over J 1 E .
12.72 Remark Using the coordinate functions xi , yα and yαi on J 1 E defined in

Remark 12.37, we denote the coordinate functions on J 1 (J 1 E) by xi , yα , yαi , yα; j and
yαi ; j . Note that the coordinate functions yαj and yα; j are different on J 1 (J 1 E) , but that
they coincide on i1,1 (J 2 E) since
∂ (yα ◦ j 1 φ ) ∂ (yα ◦ φ )
yα; j (i1,1 ( j 2p φ )) = yα; j ( j 1p ( j 1φ )) = j (p) = (p) = yαj ( j 2p φ ) .
∂x ∂xj
The same is true for the coordinate functions yαi ; j and yαj ;i when i 6= j , since
∂ (yαi ◦ j 1 φ ) ∂ (yα ◦ φ )
yαi ; j (i1,1 ( j 2p φ )) = yαi ; j ( j 1p ( j 1φ )) = j (p) = (p) = yαi j ( j 2p φ ) .
∂x ∂ x j ∂ xi
The elements of J 1 (J 1 E) are called repeated jets , and those belonging to i1,1 (J 2 E)
are called holonomic.
12.73 Proposition Let π : E → M be a bundle over M n with fibre F m . Then there

is a bundle map (D 1 , π1,0 ) , called the Spencer map , between the bundles (π1 ) 1,0 :
J 1 (J 1 E) → J 1 E and π ′ : Λ 1 (π ∗ (T M) ;V ) → E , where
D 1 = j 1 (π1,0 ) − (π1) 1,0 .

JET BUNDLES 573
We also have a bundle πb2,1 : Jb2 E → J 1 E , called the semi-holonomic 2-jet bundle ,
where Jb2 E is the submanifold of J 1 (J 1 E) given by Jb2 E σ = ker D 1 σ for σ ∈ J 1 E ,
and πb2,1 is the restriction of (π1 ) 1,0 to Jb2 E .
PROOF : By Proposition 12.36 we know that π1,0 : J 1 E → E is an affine bundle over
E modelled on the vector bundle π ′ : Λ 1 (π ∗ (T M) ;V ) → E. To show that D 1 is well
defined, we need to prove that j 1 (π1,0 )(σ ) and (π1 ) 1,0 (σ ) are in the same fibre of
J 1 E over E for each σ ∈ J 1 (J 1 E). Let σ = j 1p ψ where ψ ∈ Γ p (J 1 E). Then we have
that
π1,0 ◦ j 1 (π1,0 )( j 1p ψ ) = π1,0 ( j 1p (π1,0 ◦ ψ )) = π1,0 ◦ ψ (p)
and
π1,0 ◦ (π1 ) 1,0 ( j 1p ψ ) = π1,0 (ψ (p)) = π1,0 ◦ ψ (p) ,
showing that D 1 ( j 1p ψ ) is a well-defined element in the fibre of Λ 1 (π ∗ (T M) ;V ) over
π1,0 ◦ ψ (p).
Using the same notation as in the proof of Proposition 12.71, the diffeomorphism
−1
(τz × id) ◦ τz̃ : (π1 ) 1,0 (W1 ) → W × L(Rn , Rm ) × L(Rn , Rm ) × L(Rn , Rnm ) is given by
(τz × id) ◦ τz̃ = (π1,0 ◦ (π1 ) 1,0 , pr2 ◦ τz ◦ (π1 ) 1,0 , pr2 ◦ τz ◦ j 1 (π1,0 ), pr3 ◦ τz̃ ) ,
where pr2 : W × L(Rn , Rm ) → L(Rn , Rm ) and pr3 : W1 × L(Rn , Rm ) × L(Rn , Rnm ) →

L(Rn , Rnm ) are the projections on the last factor. If τz′ : π ′ −1 (W ) → W × L(Rn , Rm )
is the equivalence over W described in the proof of Proposition 12.36, we have that
τz′ ◦ D 1 ◦ τz̃−1 ◦ (τz × id)−1 (u, F, G, H) = (u, G − F)
for u ∈ E , F, G ∈ L(Rn , Rm ) and H ∈ L(Rn , Rnm ), showing that D 1 is a bundle map.

We also obtain an equivalence Λ ◦ (τz × id) ◦ τz̃ |πb −1 (W ) over W1 , where Λ :
2,1 1
W × L(Rn , Rm ) × L(Rn , Rm ) × L(Rn , Rnm ) → W × L(Rn , Rm ) × L(Rn , Rnm ) is the
projection given by
Λ(u, F, G, H) = (u, F, H) ,
completing the proof that πb2,1 : Jb2 E → J 1 E is a bundle.
12.74 Remark Using the coordinate functions on J 1 (J 1 E) described in Remark

12.72, we see that Jb2 E is the submanifold of J 1 (J 1 E) where yαj = yα; j for j = 1, ... , n.
From this it follows that
i1,1 (J 2 E) ⊂ Jb2 E ⊂ J 1 (J 1 E) .
We use xi , yα , yαi and yαi ; j as coordinate functions on Jb2 E .

πe ′ : π2,1
∗ (T J 1 E) → J 2 E be the pullback of the tangent bundle π ′ : T J 1 E → J 1 E by
e → J 2E
the 1-jet projection π2,1 : J 2 E → J 1 E. Then we have two subbundles πe ′′ : H
e → J 2 E of πe ′ so that
and πe ′′′ : V
∗
π2,1 eσ ⊕ V
(T J 1 E) σ = H eσ
for σ ∈ J 2 E , where He σ = {σ } × ( j 1 φ ) ∗ (Tp M) and Veσ = {σ } ×V1 ν when σ = j 2p φ

and π2,1 (σ ) = ν . The maps e ∗ (T J 1 E) → π ∗ (T J 1 E) and v
h : π2,1 2,1
∗ (T J 1 E) →
e : π1,0
∗ (T J 1 E) which are projections on H
π1,0 eσ and V eσ in each fibre π ∗ (T J 1 E) σ , are
2,1
2
bundle maps over J E .
PROOF : e → J 2E
Apply Proposition 12.46 to the bundle π1 : J 1 E → M, and let πe ′′ : H
e → J 2 E be the bundles induced from the subbundles of holonomic and
and πe ′′′ : V
vertical tangent vectors in πe1′ : (π1 ) 1,0
∗ (T J 1 E) → J 1 (J 1 E) by the embedding i
1,1 :
2 1 1
J E → J (J E) , using that (π1 ) 1,0 ◦ i1,1 = π2,1 .
12.76 Remark Suppose that ( σ , w) ∈ π2,1 ∗ (T J 1 E) where σ = j 2 φ , π (σ ) = ν

p 2,1
and w ∈ Tν J 1 E is the tangent vector given by

w = ∑ a ∂ i + ∑ b α ∂ α + ∑ cαi ∂ ∂yα
i
∂x α
∂y i
i ν ν i, α ν
in the local coordinates described in Remark 12.37. Then we have that

!!
n
∂

e
h ( σ , w) = ( σ , ( j 1 φ ) ∗ p ◦ π1 ∗ ν (w)) = i
σ , ∑ a (j φ)∗p 1
i=1
∂ xi
p
( )!
n
∂

= σ , ∑ ai α ∂ α ∂
i + ∑ y i (σ ) ∂ y α + ∑ y i j (σ ) ∂ y α
i=1
∂x α i, α j
ν ν ν
and
ve( σ , w) = ( σ , w − ( j 1 φ ) ∗ p ◦ π1 ∗ ν (w))
( ) ( ) !
n
∂
n
= σ ,∑ bα − ∑ ai yαi (σ ) α +∑ cαi − ∑ ai yαi j (σ ) ∂ ∂yα .
α
∂y i
i=1 ν i, α i=1 ν
12.77 Definition Let π : E → M be a bundle over M n with fibre F m . A vector

field X : J 2 E → T J 1 E along π2,1 is called a total derivative if the section Xe : J 2 E →
∗ (T J 1 E) determined by X belongs to H,
π2,1 e i.e., if
X( j 2p φ ) ∈ ( j 1 φ ) ∗ (Tp M)
JET BUNDLES 575
−1
for p ∈ M and φ ∈ Γ p (E) . Let (z,W ) and ((z × β ) ◦ τz , π1,0 (W )) be local charts on E
1
and J E with coordinate functions described in Remarks 12.4 and 12.37. The vector
−1
field Xi : π2,0 (W ) → T J 1 E along π2,1 defined by
!
∂

1
Xi (σ ) = ( j φ ) ∗ = ∂ i + ∑ yαi (σ ) ∂ α + ∑ yαi j (σ ) ∂ ∂yα
∂ xi ∂x α
∂y
j, α j
p ν ν ν
−1
for σ ∈ π2,0 (W ) where σ = j 2p φ and π2,1 (σ ) = ν , is called the i-th coordinate total
d 1
derivative and is denoted by dx i for i = 1, ... , n . The tangent vector Xi (σ ) in Tν J E

d

is denoted by dxi , and we have that
σ
!
d
∂
∂ (g ◦ j 1 φ )
(g) = ∂ i (g ◦ j 1 φ ) =
dg
(σ ) = dx
(g) = ( j 1 φ ) ∗ (p)
dxi i
∂ xi ∂x ∂ xi
σ p p
when σ = j 2p φ , π2,1 (σ ) = ν and g ∈ Fν (J 1 E) . In particular, it follows by Definition

12.48 that
d df ∂ 2( f ◦ φ )
i j (σ ) = (p)
dx dx ∂ xi ∂ x j
when σ = j 2p φ , φ (p) = u and f ∈ Fu (E) .
12.78 Let π : E → M be a bundle over M n with fibre F m , and let π1′ : T J 1 E → J 1 E

′
and π2 : T J 2 E → J 2 E be the tangent bundles of J 1 E and J 2 E . By 12.13 the map
(π2′ , π2,1 ∗ ) : T J 2 E → J 2 E × T J 1 E induces a bundle map ρ 2 : T J 2 E → π2,1∗ (T J 1 E)
2
over J E , as we see from the commutative diagram
π2,1 ∗✲
T J 2E T J 1E
π2′ π1′
❄ ❄
π2,1 ✲
J 2E J 1E
−1
Using the local charts (z,W ) , ((z × β ) ◦ τz , π1,0 (W )) and ((((z × β ) ◦ τz ) × βs ) ◦ τz2 ,
−1 1 2
π2,0 (W )) on E , J E and J E with coordinate functions described in Remarks 12.37
and 12.67, we have that
! ! ! !

∂

∂

ρ2 i = σ, i , ρ2 ∂ = σ , ∂ α ,
∂x ∂x ∂ yα ∂y
σ ν σ ν
! ! !

ρ2 ∂ ∂
= σ , ∂ yα and ρ 2 ∂ = (σ , 0)
∂ yαi i ∂ yαi j
σ ν σ
−1
for σ ∈ π2,0 (W ) where π2,1 (σ ) = ν .
e ∈ Ω 1 (J 2 E ; π2,1
Let ω ∗ (T J 1 E)) be the bundle-valued 1-form determined by the
composed bundle map b h=e ∗ (T J 1 E) over J 2 E as described in

h ◦ ρ 2 : T J 2 E → π2,1
Proposition 5.30. Then it follows from Proposition 5.44 and Remark 12.76 that
n
e | π −1 (W ) = ∑ dxi ⊗ d i
φ1 ◦ ω (1)
2,0 dx
i=1
∗ (T J 1 E)) → T 1 (T J 2 E, π ∗ (T J 1 E)) is the equivalence

where φ 1 : Λ 1 (T J 2 E ; π2,1 1 2,1
2
over J E defined as in 5.25.
Now let d ωe : Ω(J 1 E) → Ω(J 2 E) be the derivation along π2,1 of type d∗ and de-
gree 1 obtained from ω e as described in Proposition 12.30. If σ = j 2p φ , we have that
(i ωe dyα )(σ ) = dyα ( j 1p φ ) ◦ ω

e (σ ) = dyα ( j 1p φ ) ◦ ( j 1 φ ) ∗ p ◦ π2 ∗ σ
n n
∂ (yα ◦ φ )
= d(yα ◦ j 1 φ )(p) ◦ π2 ∗ σ = ∑ i (p) dxi (p) ◦ π2 ∗ σ = ∑ yαi (σ ) dxi (σ )
i=1
∂ x i=1
so that
n
d ωe dyα = − d ◦ i ωe dyα = ∑ dxi ∧ dyαi .
i=1

n
d ωe dyαi = ∑ dx j ∧ dyαi j and d ωe dxi = 0 .
j=1
If f ∈ F (J 1 E) , then
n
df
d ωe f = ∑ dxi dxi
i=1
since
n n
∂ ( f ◦ j 1φ ) df
(d ωe f )(σ ) = (i ωe d f )(σ ) = ∑ i (p) dxi (p) ◦ π2 ∗ σ = ∑ dxi (σ ) dxi (σ ) .
i=1
∂ x i=1
Consider the commutative diagram

JET BUNDLES 577
j 2φ ✲
M J 2E
idM π2,1
❄ ❄
M ✲ J 1E
j 1φ
and let ε ∈ Ω 1 (M; T M) be the bundle-valued 1-form defined in Example 5.45 corre-
sponding to the Kronecker delta tensor on M. If σ = j 2p φ , we have that
e (σ ) ◦ ( j 2 φ ) ∗ p = ( j 1 φ ) ∗ p ◦ π2 ∗ σ ◦ ( j 2 φ ) ∗ p = ( j 1 φ ) ∗ p ◦ ε (p) ,
ω
showing that the bundle-valued 1-forms ε and ω e along idM and π2,1 are ( j 2 φ , j 1 φ )-
related as described in 12.32. Hence it follows that
d ◦ ( j 1 φ ) ∗ = ( j 2 φ ) ∗ ◦ d ωe . (2)
The derivation d ωe will be important in our study of the calculus of variations in

12.90.
PROLONGATION OF VECTOR FIELDS

12.79 Let π : E → M be a bundle over M n with fibre F m , and let π ′′ : V → E
′′
and π1 : V1 → J 1 E be the vertical tangent bundles of the bundles π : E → M and
π1 : J 1 E → M. We will construct a map i1 : J 1V → V1 which is an equivalence
over M between the 1-jet bundle π1′ : J 1V → M of the bundle π ′ : V → M where
π ′ = π ◦ π ′′ , and the bundle π1′′′ : V1 → M where π1′′′ = π1 ◦ π1′′ .
Let (t, π −1 (U)) be a local trivialization around a point u on E with t(u) = (p, v) ,
where U is the coordinate neighbourhood of a local chart (x,U) around p on M
with x(p) = a , and consider the local chart (z,W ) on E obtained from (t, π −1 (U)) ,
(x,U) and a local chart (y, Q) around v on F as described in Proposition 12.3, so that
W = t −1 (U × Q) and z = (x × y) ◦ t .
The bundle π ′ : V → M has fibre T F, and we have a local trivialization
(t , π ′ −1 (U)) on V with t ′ = (πU′ , (pr2′ ◦ t)∗ ◦ i) , where pr2′ : U × F → F is the pro-
′
jection on the second factor and i : V → T E is the inclusion map. From this we also
obtain a local chart (z′ ,W ′ ) on V with W ′ = t ′ −1 (U × T F|Q ) and z′ = (x × y′ ) ◦ t ′ , us-
ing the local chart (y′ , T F|Q ) on T F where y′ = (y × id) ◦ ty . We let β : L(Rn , Rm ) →
R n m and β ′ : L(Rn , R2m ) → R 2nm be linear isomorphisms defined as in Corollary
5.29.
To define the map i1 : J 1V → V1 , we consider a smooth map α : I × O → E sat-
isfying π ◦ α = pr2 , where pr2 : I × O → M is the projection on the second factor,
and where O is an open neighbourhood of p on M and I is an open interval contain-

ing 0 so that α (I × O) ⊂ W . Let α e : I × x(O) → Rm be the smooth map given by
e −1
α = z2 ◦ α ◦ (id × x) . Now let αq : I → E and αt : O → E be the smooth maps
defined by αq (t) = αt (q) = α (t, q) for t ∈ I and q ∈ O, and consider the section
ψ : O → V of π ′ given by ψ (q) = αq′ (0) for q ∈ O. Using the local charts defined
above, we have that
z′2 ◦ ψ ◦ x−1 (b) = (y × id) ◦ ty ◦ (pr2′ ◦ t)∗ ◦ ψ (q)

= (y × id) ◦ ty ◦ (pr2′ ◦ t ◦ αq ) ′ (0) = (z2 ◦ αq (0), (z2 ◦ αq ) ′ (0))
e (0, b), D1 α
= (α e (0, b))
for b = x(q) ∈ x(O). As in the proof of Proposition 12.40 it follows that
(z′2 × β ′ ) ◦ τ z′ ( j 1p ψ ) = (z′2 ◦ ψ ◦ x−1 (a), β ′ ◦ D(z′2 ◦ ψ ◦ x−1 )(a))
= (α e (0, a), β ′ (D2 α

e (0, a), D1 α e (0, a), D2 D1 α
e (0, a))) .
We now let c : I → J 1 E be the curve given by c(t) = j 1p αt for t ∈ I, and define

i1 ( j 1p ψ ) = c′ (0). Using the local chart (e e ) on J p1 E with W
z, W e = J p1 E ∩ π −1 (W ) and
1,0
z = (z2 × β ) ◦ τ z | We , we have that
e
z ◦ c(t) = (z2 ◦ αt ◦ x−1 (a), β ◦ D(z2 ◦ αt ◦ x−1 )(a))

e
for t ∈ I, so that
z × id) ◦ tez ◦ i1 ( j 1p ψ ) = (e
(e z ◦ c) ′ (0))
z ◦ c(0), (e
= (α e (0, a), D1 α
e (0, a), β ◦ D2 α e (0, a), β ◦ D1 D2 α
e (0, a)) .
z × id) ◦ tez ◦ i1 = (z′2 × β ′ ) ◦ τ z′ | We ′ ,

j1 ◦ (e (1)
where W ′ −1 (W ′ ) and j : Rm × R n m × Rm × R n m → Rm × Rm × R 2nm is

e ′ = J p1V ∩ π 1,0 1
the linear isomorphism given by
j1 (a, b, c, d) = (a, c, β ′ ◦ (β × β ) −1(b, d))
for a, c ∈ Rm and b, d ∈ Rn m , showing that the map i1 : J 1V → V1 is a well-defined

equivalence over M. It satisfies the relation
′
πe1,0 ◦ i1 = π1,0 (2)
where πe1,0 : V1 → V is the smooth map induced by π1,0 ∗ : T J 1 E → T E , since
π1,0 ∗ ◦ i1 ( j 1p ψ ) = π1,0 ∗ ◦ c′ (0) = (π1,0 ◦ c) ′ (0) = α p′ (0) = ψ (p) = π1,0

′
( j 1p ψ ) .
JET BUNDLES 579
We also have that
(id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ) ∗ ◦ i1 = pr2′′′ ◦ τ z′ (3)
where pr2′′ : W × L(Rn , Rm ) → L(Rn , Rm ) and pr2′′′ : W ′ × L(Rn , R2m ) → L(Rn , R2m )
are the projections on the second factor, since
pr2′′ ◦ τz ◦ c(t) = pr2′′ ◦ τz ( j 1p αt ) = D(z2 ◦ αt ◦ x−1 )(a) = D2 α

e (t, a)
for t ∈ I so that
(id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ) ∗ ◦ i1 ( j 1p ψ ) = (id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ◦ c) ′ (0)
e (0, a)) = D(z′2 ◦ ψ ◦ x −1 )(a) = pr2′′′ ◦ τ z′ ( j 1p ψ ) .

e (0, a), D1 D2 α
= (D2 α

π1′′ ◦ i1 ◦ j 1ψ = j 1 (π ′′ ◦ ψ ) (4)
since
π1′′ ◦ i1 ( j 1p ψ ) = π1′′ ◦ c′ (0) = c(0) = j 1p α 0 = j 1p (π ′′ ◦ ψ ) .
12.80 Remark Using the local coordinates described in Remarks 12.4, 12.5 and
12.37, we have coordinate functions xi , yα and ẏα on V which give rise to the coor-
dinate functions xi , yα , ẏα , yαi and ẏαi on J 1V for i = 1, 2, ... , n and α = 1, 2, ... , m.
We also have coordinate functions xi , yα and yαi on J 1 E from which we obtain the
coordinate functions xi , yα , yαi , ẏα and ẏαi on V1 . By formula (1) in 12.79 we have
that
xi ◦ i1 = xi , yα ◦ i1 = yα , yαi ◦ i1 = yαi , ẏα ◦ i1 = ẏα and ẏαi ◦ i1 = ẏαi
for i = 1, 2, ... , n and α = 1, 2, ... , m.
12.81 Let X be a vertical vector field on E , i.e., a section on E in the vertical

tangent bundle π ′′ : V → E , using the same notation as in 12.79. Regarding X as a
bundle map over M between the bundles π : E → M and π ′ : V → M , we obtain a
vertical vector field X 1 on J 1 E , called the 1-jet prolongation of X , defined by
X 1 = i1 ◦ j 1X .
Indeed, it follows from formula (4) in 12.79 that
π1′′ ◦ X 1 ( j 1p φ ) = π1′′ ◦ i1 ( j 1p (X ◦ φ )) = j 1p φ
for p ∈ M and φ ∈ Γ p (E) , showing that X 1 is a section on J 1 E in the vertical tangent

bundle π1′′ : V1 → J 1 E . Using formula (2) in 12.79 we also have that
′
πe1,0 ◦ X 1 = πe1,0 ◦ i1 ◦ j 1X = π1,0 ◦ j 1X = X ◦ π1,0 .

π ′′ : T E → E be the tangent bundle of E and π ′ : T E → M be the bundle where
π ′ = π ◦ π ′′ . Then the map r1 : J 1 T E → T J 1 E defined by
r1 ( j 1p ψ ) = i1 ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ )) + ( j 1φ ) ∗ ◦ π∗ ◦ ψ (p)
for p ∈ M and ψ ∈ Γ p (T E) where φ = π ′′ ◦ ψ , is a bundle map over T E from the

′ : J 1 T E → T E of π ′ to the bundle π
1-jet bundle π1,0 1
1,0 ∗ : T J E → T E . We have that
π1′′ ◦ r1 ◦ j 1ψ = j 1φ
where π1′′ : T J 1 E → J 1 E is the projection in the tangent bundle of J 1 E .

PROOF : We first note that the term i1 ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ )) is well defined since
π ′′ ◦ φ ∗ ◦ π∗ ◦ ψ = φ ◦ π ◦ π ′′ ◦ ψ = φ = π ′′ ◦ ψ
and
π∗ ◦ ( ψ − φ ∗ ◦ π∗ ◦ ψ ) = 0 .
From this and formula (4) in 12.79 it also follows that
π1′′ ◦ i1 ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ )) = j 1p ( π ′′ ◦ {ψ − φ ∗ ◦ π∗ ◦ ψ }) = j 1p φ
and
π1′′ ◦ ( j 1φ ) ∗ ◦ π∗ ◦ ψ (p) = j 1φ ◦ π ◦ π ′′ ◦ ψ (p) = j 1p φ ,
showing that r1 ( j 1p ψ ) is well defined. It also shows the last part of the proposition.
Using formula (2) in 12.79 we now have that
′
π1,0 ∗ ◦ r1 = π1,0
since
′
π1,0 ∗ ◦ r1 ( j 1p ψ ) = π1,0 ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ )) + (π1,0 ◦ j 1φ ) ∗ ◦ π∗ ◦ ψ (p)
′
= (ψ − φ ∗ ◦ π∗ ◦ ψ )(p) + φ ∗ ◦ π∗ ◦ ψ (p) = ψ (p) = π1,0 ( j 1p ψ ) .
Formula (3) in 12.79 implies that
(id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ) ∗ ◦ r1 ( j 1p ψ )
= pr2′′′ ◦ τ z′ ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ ))
+ (id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ◦ j 1φ ◦ π ) ∗ ◦ ψ (p)
= D(z′2 ◦ { ψ − φ ∗ ◦ π∗ ◦ ψ } ◦ x−1 )(a)

+ (id × β ) −1 ◦ tβ ◦ {D(z2 ◦ φ ◦ x−1 ) ◦ x ◦ π } ∗ ◦ ψ (p) .
JET BUNDLES 581
Let O be an open neighbourhood of p on M so that φ (O) ⊂ W , and let pr2 :

W × Rn+m → Rn+m be the projection on the second factor with component maps
pr21 : W × Rn+m → Rn and pr22 : W × Rn+m → Rm . Then we have that
z′2 ◦ ( ψ − φ ∗ ◦ π∗ ◦ ψ )(q) = (y × id) ◦ ty ◦ (pr2′ ◦ t) ∗ ◦ ( ψ − φ ∗ ◦ π∗ ◦ ψ )(q)
= (y ◦ pr2′ ◦ t ◦ φ (q), {D(y ◦ pr2′ ◦ t ◦ z−1 )(z ◦ φ (q))

− D(y ◦ pr2′ ◦ t ◦ φ ◦ x−1 )(x(q)) ◦ D(x ◦ π ◦ z−1 )(z ◦ φ (q)) } ◦ pr2 ◦ tz ◦ ψ (q))
= (z2 ◦ φ (q), {pr22 − D(z2 ◦ φ ◦ x−1 )(x(q)) ◦ pr21 } ◦ tz ◦ ψ (q))
for q ∈ O , which implies that
D(z′2 ◦ { ψ − φ ∗ ◦ π∗ ◦ ψ } ◦ x−1 )(a)

= (D(z2 ◦ φ ◦ x−1 )(a), D(pr22 ◦ tz ◦ ψ ◦ x−1 )(a)
− D(z2 ◦ φ ◦ x−1 )(a) ◦ D(pr21 ◦ tz ◦ ψ ◦ x−1 )(a)
− i pr21 ◦ tz ◦ ψ ◦ x−1 (a) D 2 (z2 ◦ φ ◦ x−1 )(a)) .
We also have that
(id × β ) −1 ◦ tβ ◦ {D(z2 ◦ φ ◦ x−1 ) ◦ x ◦ π } ∗ ◦ ψ (p) = (D(z2 ◦ φ ◦ x−1 )(a),
β −1 ◦ D( β ◦ D(z2 ◦ φ ◦ x−1 ) ◦ x ◦ π ◦ z−1 )(z ◦ φ (p)) ◦ pr2 ◦ tz ◦ ψ (p))

= (D(z2 ◦ φ ◦ x−1 )(a), i pr21 ◦ tz ◦ ψ ◦ x−1 (a) D 2 (z2 ◦ φ ◦ x−1 )(a)) .
Combining this, we see that the second derivatives cancel, and we obtain that
(id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ) ∗ ◦ r1 ( j 1p ψ )
= (D(z2 ◦ φ ◦ x−1 )(a), D(pr22 ◦ tz ◦ ψ ◦ x−1 )(a)

− D(z2 ◦ φ ◦ x−1 )(a) ◦ D(pr21 ◦ tz ◦ ψ ◦ x−1 )(a)) .
The bundle π ′ : T E → M has fibre Rn × T F, and we have a local trivialization

(t ′′ , π ′ −1 (U)) on T E with t ′′ = (πU′ , pr2′′′′ ◦ tx ◦ π∗ , (pr2′ ◦ t) ∗ ) , where pr2′′′′ : Rn ×
Rn → Rn is the projection on the second factor. From this we also obtain a local
chart (z′′ ,W ′′ ) on T E with W ′′ = t ′′ −1 (U × Rn × T F|Q ) and z′′ = (x × id × y′ ) ◦ t ′′ ,
using the local chart (y′ , T F|Q ) on T F where y′ = (y × id) ◦ ty . We now have that
τz′′ ( j 1p ψ ) = (ψ (p), D(z′′2 ◦ ψ ◦ x−1 )(a))
= (ψ (p), D(pr2′′′′ ◦ tx ◦ π∗ ◦ ψ ◦ x−1 )(a),

D(y′ ◦ (pr2′ ◦ t) ∗ ◦ ψ ◦ x−1 )(a))
= (ψ (p), D(pr21 ◦ tz ◦ ψ ◦ x−1 )(a),
D(z2 ◦ φ ◦ x−1 )(a), D(pr22 ◦ tz ◦ ψ ◦ x−1 )(a))
which shows that
(id × β ) −1 ◦ tβ ◦ (pr2′′ ◦ τz ) ∗ ◦ r1 ◦ τz−1

′′ (u, G1 , G2 , G3 ) = (G2 , G3 − G2 ◦ G1 )
for u ∈ W ′′ , G1 ∈ L(Rn , Rn ) and G2 , G3 ∈ L(Rn , Rm ).
12.83 Remark Let π : E → M be a bundle over M n with fibre F m , and consider

the tangent bundle π ′′ : T E → E of E and the bundle π ′ : T E → M where π ′ = π ◦ π ′′ .
Let r1 : J 1 T E → T J 1 E be the bundle map over T E defined in Proposition 12.82, and
suppose that ψ ∈ ΓU (T E) is a section of π ′ on an open subset U of M , given in the
local coordinates described in Remark 12.4 by

n m
ψ (p) = ∑ a (p) ∂ i + ∑ b α (p) ∂ α
i
∂x ∂y
i=1 φ (p) α =1 φ (p)
for p ∈ U where φ = π ′′ ◦ ψ . Using also the local coordinates described in Remarks

12.5 and 12.37, we have that
′
ẋi ◦ r1 ( j 1p ψ ) = ẋi ◦ π1,0 ∗ ◦ r1 ( j 1p ψ ) = ẋi ◦ π1,0 ( j 1p ψ ) = ẋi ◦ ψ (p) = ai (p)
and
ẏα ◦ r1 ( j 1p ψ ) = ẏα ◦ π1,0 ∗ ◦ r1 ( j 1p ψ ) = ẏα ◦ π1,0

′
( j 1p ψ ) = ẏα ◦ ψ (p) = bα (p)
for p ∈ U. We also have that
ẏαi ◦ r1 ( j 1p ψ ) = ẏαi ◦ i1 ( j 1p ( ψ − φ ∗ ◦ π∗ ◦ ψ )) + ẏαi ◦ ( j 1φ ) ∗ ◦ π∗ ◦ ψ (p)

= ∂ i { ẏα ◦ ψ − (π∗ ◦ ψ ) (yα ◦ φ )} + (π∗ ◦ ψ )(p) (yαi ◦ j 1φ )
∂x
p

α n
∂ (yα ◦ φ )
n ∂ (yα ◦ φ )
= ∂ b i (p) − ∂ i ∑ a ∂xj j
+ ∑ a (p) ∂ i
j
∂x ∂x j=1 j=1
∂x ∂xj
p p
n n
α j ∂ (yα ◦ ψ )
= ∂ b i (p) − ∑ ∂ a i (p) j (p) = ẏαi − ∑ ẋ ij yαj ( j 1p ψ )
∂x j=1
∂x ∂x j=1
JET BUNDLES 583
which implies that

n
r1 ( j 1p ψ ) = ∑ ai (p) ∂ i + ∑ b α (p) ∂ α + ∑ ẏαi − ∑ ẋ ij yαj ( j 1p ψ ) ∂ ∂yα
∂x 1 ∂y i
i α
jpφ 1 i, α jpφ j=1 j 1p φ
for p ∈ U.
12.84 Let X be a vector field on E , i.e., a section on E in the tangent bundle

π ′′ : T E → E , using the same notation as in Proposition 12.82. Regarding X as a
bundle map over M between the bundles π : E → M and π ′ : T E → M , we obtain a
vector field X 1 on J 1 E , called the 1-jet prolongation of X , defined by
X 1 = r1 ◦ j 1X .
Indeed, it follows from Proposition 12.82 that
π1′′ ◦ X 1 ( j 1p φ ) = π1′′ ◦ r1 ( j 1p (X ◦ φ )) = j 1p φ
for p ∈ M and φ ∈ Γ p (E) , showing that X 1 is a section on J 1 E in the tangent bundle

π1′′ : T J 1 E → J 1 E . We also have that
′
π1,0 ∗ ◦ X 1 = π1,0 ∗ ◦ r1 ◦ j 1X = π1,0 ◦ j 1X = X ◦ π1,0 ,
which shows that the vector fields X 1 and X are π1,0 - related.
12.85 Remark Suppose that the vector field X on E is given by

n m
X | W = ∑ ai ∂ i + ∑ b α ∂α
i=1
∂x α =1
∂y
on an open subset W of E, using the local coordinates described in Remark 12.4. By

Definition 12.48 and Remark 12.83 we have that
n
α α α j α
ẏ i ◦ X ( j p φ ) = ẏ i ◦ r1 ( j p (X ◦ φ )) = ẏ i − ∑ ẋ i y j ( j 1p (X ◦ φ ))
1 1 1
j=1
n n
∂ (bα ◦ φ ) ∂ (a j ◦ φ ) ∂ (yα ◦ φ ) d bα j
= i (p) − ∑ i (p) (p) = − ∑ yαj da ( j 1p φ )
∂x j=1
∂x ∂xj dxi
j=1
dxi
which shows that

n
d bα j
X | π −1 (W ) = ∑ (a ◦ π1,0 ) ∂ i + ∑ (b α ◦ π1,0 ) ∂ α + ∑
1 i
dxi
− ∑ yαj da
dxi
∂
∂ yαi .
1,0
i
∂x α
∂y i, α j=1
12.86 Let π : E → M be a bundle over M n with fibre F m , and let π ′′ : V → E

′′
and π2 : V2 → J 2 E be the vertical tangent bundles of the bundles π : E → M and
π2 : J 2 E → M. We will construct a map i2 : J 2V → V2 which is an equivalence
over M between the 2-jet bundle π2′ : J 2V → M of the bundle π ′ : V → M where
π ′ = π ◦ π ′′ , and the bundle π2′′′ : V2 → M where π2′′′ = π2 ◦ π2′′ .
Let (t, π −1 (U)) be a local trivialization around a point u on E with t(u) = (p, v) ,
where U is the coordinate neighbourhood of a local chart (x,U) around p on M
with x(p) = a , and consider the local chart (z,W ) on E obtained from (t, π −1 (U)) ,
(x,U) and a local chart (y, Q) around v on F as described in Proposition 12.3, so that
W = t −1 (U × Q) and z = (x × y) ◦ t .
The bundle π ′ : V → M has fibre T F, and we have a local trivialization
(t , π ′ −1 (U)) on V with t ′ = (πU′ , (pr2′ ◦ t)∗ ◦ i) , where pr2′ : U × F → F is the
′
projection on the second factor and i : V → T E is the inclusion map. From this
we also obtain a local chart (z′ ,W ′ ) on V with W ′ = t ′ −1 (U × T F|Q ) and z′ =
(x × y′ ) ◦ t ′ , using the local chart (y′ , T F|Q ) on T F where y′ = (y × id) ◦ ty . We let
β : L(Rn , Rm ) → R n m , β ′ : L(Rn , R2m ) → R 2nm , βs : T s 2 (Rn ; Rm ) → R n(n+1) m/2
and βs′ : T s 2 (Rn ; R2m ) → R n(n+1) m be linear isomorphisms defined as in the proof
of Proposition 5.31.
To define the map i2 : J 2V → V2 , we consider a smooth map α : I × O → E sat-
isfying π ◦ α = pr2 , where pr2 : I × O → M is the projection on the second factor,
and where O is an open neighbourhood of p on M and I is an open interval contain-
ing 0 so that α (I × O) ⊂ W . Let α e : I × x(O) → Rm be the smooth map given by
−1
e = z2 ◦ α ◦ (id × x) . Now let αq : I → E and αt : O → E be the smooth maps
α
defined by αq (t) = αt (q) = α (t, q) for t ∈ I and q ∈ O, and consider the section
ψ : O → V of π ′ given by ψ (q) = αq′ (0) for q ∈ O. Using the local charts defined
above, we have that
z′2 ◦ ψ ◦ x−1 (b) = (y × id) ◦ ty ◦ (pr2′ ◦ t)∗ ◦ ψ (q)

= (y × id) ◦ ty ◦ (pr2′ ◦ t ◦ αq ) ′ (0) = (z2 ◦ αq (0), (z2 ◦ αq ) ′ (0))
e (0, b), D1 α
= (α e (0, b))
for b = x(q) ∈ x(O). As in the proof of Proposition 12.66 it follows that
(((z′2 × β ′ ) ◦ τ z′ ) × βs′ ) ◦ τ z2′ ( j 2p ψ )
= (z′2 ◦ ψ ◦ x−1 (a), β ′ ◦ D(z′2 ◦ ψ ◦ x−1 )(a), βs′ ◦ D2 (z′2 ◦ ψ ◦ x−1 )(a))
e (0, a), β ′ (D2 α
e (0, a), D1 α
= (α e (0, a)), βs′ (D22 α
e (0, a), D2 D1 α e (0, a), D22 D1 α
e (0, a))) .
We now let c : I → J 2 E be the curve given by c(t) = j 2p αt for t ∈ I, and define

i2 ( j 2p ψ ) = c′ (0). Using the local chart (e e ) on J p2 E with W
z, W e = J p2 E ∩ π −1 (W ) and
2,0
z = (((z2 × β ) ◦ τ z ) × βs ) ◦ τ z2 | We , we have that
e
z ◦ c(t) = (z2 ◦ αt ◦ x−1 (a), β ◦ D(z2 ◦ αt ◦ x−1 )(a), βs ◦ D2 (z2 ◦ αt ◦ x−1 )(a))
e
JET BUNDLES 585
for t ∈ I, so that
z × id) ◦ tez ◦ i1 ( j 2p ψ ) = (e
(e z ◦ c) ′ (0))
z ◦ c(0), (e
= (α e (0, a), βs ◦ D22 α

e (0, a), β ◦ D2 α e (0, a),
D1 α e (0, a), βs ◦ D1 D22 α
e (0, a), β ◦ D1 D2 α e (0, a)) .
z × id) ◦ tez ◦ i2 = (((z′2 × β ′ ) ◦ τ z′ ) × βs′ ) ◦ τ z2′ | We ′ ,

j2 ◦ (e (1)
where W ′ −1 (W ′ ) and j : Rm × R n m × R n(n+1) m/2 × Rm × R n m ×

e ′ = J p2V ∩ π 2,0 2
R n(n+1) m/2 → Rm × Rm × R 2nm × R n(n+1) m is the linear isomorphism given by
j2 (a, b, c, d, e, f ) = (a, d , β ′ ◦ (β × β ) −1 (b, e), βs′ ◦ (βs × βs ) −1 (c, f ))
for a, d ∈ Rm , b, e ∈ Rn m and c, f ∈ R n(n+1) m/2, showing that the map i2 : J 2V → V2

is a well-defined equivalence over M. It satisfies the relation
′
πe2,1 ◦ i2 = i1 ◦ π2,1 (2)
where i1 : J 1V → V1 is the equivalence over M defined in 12.79 and πe2,1 : V2 → V1 is

the smooth map induced by π2,1 ∗ : T J 2 E → T J 1 E , since
′
π2,1 ∗ ◦ i2 ( j 2p ψ ) = π2,1 ∗ ◦ c′ (0) = (π2,1 ◦ c) ′ (0) = i1 ( j 1p ψ ) = i1 ◦ π2,1 ( j 2p ψ ) .
We also have that
(id × βs ) −1 ◦ tβs ◦ (pr2′′ ◦ τz2 ) ∗ ◦ i2 = pr2′′′ ◦ τ 2z′ (3)
where pr2′′ : W × T s 2 (Rn ; Rm ) → T s 2 (Rn ; Rm ) and pr2′′′ : W ′ × T s 2 (Rn ; R2m ) →

T s 2 (Rn ; R2m ) are the projections on the second factor, since
pr2′′ ◦ τz2 ◦ c(t) = pr2′′ ◦ τz2 ( j 2p αt ) = D2 (z2 ◦ αt ◦ x−1 )(a) = D22 α

e (t, a)
for t ∈ I so that
(id × βs ) −1 ◦ tβs ◦ (pr2′′ ◦ τz2 ) ∗ ◦ i2 ( j 2p ψ ) = (id × βs ) −1 ◦ tβs ◦ (pr2′′ ◦ τz2 ◦ c) ′ (0)
= (D22 α e (0, a)) = D2 (z′2 ◦ ψ ◦ x −1 )(a) = pr2′′′ ◦ τ 2z′ ( j 2p ψ ) .

e (0, a), D1 D22 α

π2′′ ◦ i2 ◦ j 2ψ = j 2 (π ′′ ◦ ψ ) (4)
since
π2′′ ◦ i2 ( j 2p ψ ) = π2′′ ◦ c′ (0) = c(0) = j 2p α 0 = j 2p (π ′′ ◦ ψ ) .
12.87 Remark Using the local coordinates described in Remarks 12.4, 12.5 and
12.67, we have coordinate functions xi , yα and ẏα on V which give rise to the co-
ordinate functions xi , yα , ẏα , yαi , ẏαi , yαi j and ẏαi j on J 2V for i, j = 1, 2, ... , n and
α = 1, 2, ... , m. We also have coordinate functions xi , yα , yαi and yαi j on J 2 E from
which we obtain the coordinate functions xi , yα , yαi , yαi j , ẏα , ẏαi and ẏαi j on V2 . By
formula (1) in 12.86 we have that
xi ◦ i2 = xi , yα ◦ i2 = yα , yαi ◦ i2 = yαi , yαi j ◦ i2 = yαi j ,

ẏα ◦ i2 = ẏα , ẏαi ◦ i2 = ẏαi and ẏαi j ◦ i2 = ẏαi j
for i, j = 1, 2, ... , n and α = 1, 2, ... , m.
12.88 Let X be a vertical vector field on E , i.e., a section on E in the vertical

tangent bundle π ′′ : V → E , using the same notation as in 12.86. Regarding X as a
bundle map over M between the bundles π : E → M and π ′ : V → M , we obtain a
vertical vector field X 2 on J 2 E , called the 2-jet prolongation of X , defined by
X 2 = i2 ◦ j 2X .
Indeed, it follows from formula (4) in 12.86 that
π2′′ ◦ X 2 ( j 2p φ ) = π2′′ ◦ i2 ( j 2p (X ◦ φ )) = j 2p φ
for p ∈ M and φ ∈ Γ p (E) , showing that X 2 is a section on J 2 E in the vertical tangent

bundle π2′′ : V2 → J 2 E . Using formula (2) in 12.86 we have that
′
πe2,1 ◦ X 2 = πe2,1 ◦ i2 ◦ j 2X = i1 ◦ π2,1 ◦ j 2X = i1 ◦ j 1X ◦ π2,1 = X 1 ◦ π2,1 .
The vector fields X 2 and X 1 can also be considered as bundle-valued 0-forms along
idJ 2 E and idJ 1 E which are (π2,1 , π2,1 )-related as described in 12.32. Hence it follows
that
∗ ∗
iX 2 ◦ π2,1 = π2,1 ◦ iX 1 .
12.89 Remark Suppose that the vertical vector field X on E is given by

m
X |W = ∑ b α ∂α
α =1
∂y
on an open subset W of E, using the local coordinates described in Remarks 12.4. By

Definition 12.48 and Remark 12.87 we have that
ẏα ◦ X 2 ( j 2p φ ) = ẏα ( j 2p (X ◦ φ )) = b α (φ (p)) ,
∂ (bα ◦ φ ) bα 1
ẏαi ◦ X 2 ( j 2p φ ) = ẏαi ( j 2p (X ◦ φ )) = i (p) = ddxi ( j pφ )
∂x
JET BUNDLES 587
and

∂ 2 (bα ◦ φ ) d d bα
ẏαi j ◦ X 2 ( j 2p φ ) = ẏαi j ( j 2p (X ◦ φ )) = (p) = dx ( j 2p φ ) ,
∂ x j ∂ xi j dxi
which shows that

d bα d bα
X | π −1 (W ) = ∑ (b ◦ π2,0 ) ∂ α + ∑
2 α
dxi
◦ π2,1
∂
∂ yαi + ∑ d
dx j dxi
∂
∂ yαi j .
2,0
α
∂y i, α i j, α
CALCULUS OF VARIATIONS
12.90 Let M n be a smooth manifold with a volume element ε , and let π : E → M
be a bundle over M with fibre F m . Then a smooth function L : J 1 E → R is called
a Lagrangian function for π , and the horizontal n-form L π1∗ ε on J 1 E is the corre-
−1
sponding Lagrangian . Let (x,U) , (z,W ) and ((z × β ) ◦ τz , π1,0 (W )) be local charts
1
on M , E and J E with coordinate functions described in Remarks 12.4 and 12.37. If
ε | U = a dx1 ∧ ... ∧ dxn ,

−1
the function L = (a ◦ π1 ) L|π −1 (W ) on π1,0 (W ) is called a Lagrangian density . The
1,0
Cartan form of L is the n-form on J 1 E given by
Θ L = L π1∗ ε + d S ε L ,
where Sε ∈ Ω n (J 1 E ;V1,0 ) is the bundle-valued n-form on J 1 E obtained from ε as

described in 12.49. Since d S ε L = iS ε dL, we have that
Θ L | π −1 (W ) = L π1∗ ε | π −1 (W )
1,0 1,0
n
∂L α
(1)
+∑ 1 i−1
∂ yαi dx ∧ · · · ∧ dx ∧ dy − ∑ yαj dx j
∧ dxi+1 ∧ · · · ∧ dxn .
i, α j=1
The Euler–Lagrange form of L is the (n + 1)-form on J 2 E defined by

∗
E L = π2,1 (dL ∧ π1∗ ε ) + d ωe Θ L ,
where d ωe : Ω(J 1 E) → Ω(J 2 E) is the derivation along π2,1 of type d∗ and degree 1
described in 12.78. Since

d ∂L
d ωe Θ L | π −1 (W ) = − ∑ dx i ∂ yαi dy α ∧ dx1 ∧ · · · ∧ dxn
2,0
i, α

∂L α 1 n
−∑ ∂ yαi ◦ π2,1 dy i ∧ dx ∧ · · · ∧ dx ,
i, α
it follows that
m n
∂L d ∂L
E L | π −1 (W ) = ∑ ∂ y α ◦ π2,1 − ∑ dxi ∂ yαi dy α ∧ dx1 ∧ · · · ∧ dxn . (2)
2,0
α =1 i=1
A section φ ∈ ΓU (E) of π on the open subset U of M is called an extremal of L

if it satisfies the following assertion :
For each compact submanifold N n of U with boundary, and each vertical vector field
X on E vanishing on π −1 (∂ N) with global flow γ : D(X) → E , we have that
Z
d

dt ( j 1 (γ t ◦ φ )) ∗ L ε = 0. (3)
N
0
The map α = γ ◦ (id × φ ) : B → E , defined on the open subset B = (id ×

φ )−1 (D(X)) of R × U, is called the variation of φ induced by X. Since
(X ◦ φ )(p) = α p′ (0)
for p ∈ U, it follows from 12.79 and 12.81 that

d 1

d 1
X 1 ( j 1p φ ) = i1 ◦ j 1X( j 1p φ ) = i1 ( j 1p (X ◦ φ )) = dt j p αt = dt j p (γ t ◦ φ )
0 0
so that

d

1 ∗ ∗
( j φ ) (dX 1 L)(p) = (iX 1 dL)( j 1p φ ) = dL(X 1
)( j 1p φ ) = dt 1
( j (γ t ◦ φ )) L(p)
0
for p ∈ U by Proposition 4.14. Condition (3) is therefore equivalent to the condition

Z
( j 1 φ ) ∗ (dX 1 L) ε = 0. (4)
N
12.91 Theorem Let M n be a smooth manifold with a volume element ε , and let
π : E → M be a bundle over M with fibre F m . Let L ∈ C∞ (J 1 E) be a Lagrangian
−1
function for π , and let (x,U) , (z,W ) and ((z × β ) ◦ τz , π1,0 (W )) be local charts on
1
M , E and J E with coordinate functions described in Remarks 12.4 and 12.37. Then
a section φ ∈ ΓU (E) of π with φ (U) ⊂ W is an extremal of L if it satisfies the Euler–
Lagrange equations
n
2 ∗ ∂L d ∂L
( j φ) ∂ y α ◦ π2,1 − ∑ dxi ∂ yα =0
i
i=1
−1
on U for α = 1, ... , m , where L is the Lagrangian density on π1,0 (W ).
JET BUNDLES 589
PROOF : Let N n be a compact submanifold of U with boundary, and let X be a verti-

cal vector field on E vanishing on π −1(∂ N) . If
m
X |W = ∑ b α ∂α ,
α =1
∂y
we have that
iX 1 Θ L | π −1 (W ) = ∑ (−1) i+1 (b α ◦ π1,0 ) ∂∂ L 1 i−1 i+1

yαi dx ∧ · · · ∧ dx ∧ dx ∧ · · · ∧ dx
n
1,0
i, α
so that

d bα ∂L α d ∂L
d ωe iX 1 Θ L | π −1 (W ) = ∑ dxi
◦ π2,1 ∂ yαi ◦ π2,1 + (b ◦ π2,0 ) dxi ∂ yαi
2,0
i, α
dx1 ∧ · · · ∧ dxn = − iX 2 d ωe Θ L | π −1 (W ) .
2,0
As X vanishes on π −1 (∂ N) , it follows from Stoke’s theorem that

Z Z
0= ( j 1 φ ◦ iN) ∗ iX 1 Θ L = d ( j 1 φ ) ∗ iX 1 Θ L
∂N N
Z Z
= ( j 2 φ ) ∗ d ωe iX 1 Θ L = − ( j 2 φ ) ∗ iX 2 d ωe Θ L .
N N
where iN : ∂ N → N is the inclusion map. Now using that

Z Z Z
( j 1 φ ) ∗ (dX 1 L) ε = ( j 1 φ ) ∗ iX 1 (dL ∧ π1∗ ε ) = ( j 2 φ ) ∗ iX 2 π2,1
∗
(dL ∧ π1∗ ε ) ,
N N N
we see that condition (4) in 12.90 is equivalent to the condition

Z
( j 2 φ ) ∗ iX 2 E L = 0 .
N
Since
m n
α ∂L d ∂L
iX 2 E L | π −1 (W ) = ∑ (b ◦ π2,0 ) ∂ y α ◦ π2,1 − ∑ dxi ∂ yαi dx1 ∧ · · · ∧ dxn ,
2,0
α =1 i=1
this completes the proof of the theorem.

Appendix A
PRELIMINARIES
MAPS
13.1 Let f : A → B be a map, and let C and D be sets with C ⊂ A and f (C) ⊂
D ⊂ B . Then a map g : C → D is said to be induced by f if g(x) = f (x) for all x ∈ C,
which means that we have a commutative diagram
f✲
A B
✻ ✻
i1 i2
g
C ✲ D
where i1 : C → A and i2 : D → B are the inclusion maps. If D = B , g is called the

restriction of f to C and is denoted by f |C . We denote g by fD when C = f −1 (D).
13.2 Given two maps f : A → B and g : C → D , we define their composition

g ◦ f : f −1 (B∩C) → D by g ◦ f (a) = g( f (a)) for a ∈ f −1 (B∩C) . We do not exclude
the case where f −1 (B ∩ C) = 0/ . If the map f : A → B is injective, we denote by
f −1 : f (A) → A the inverse of the bijective map f f (A) defined by f −1 ( f (a)) = a for
a ∈ A.
THE PERMUTATION GROUP

13.3 Definition If In = {1 , ..., n}, a bijective map σ : In → In is called a permu-
tation of In . This set of permutations is a group Sn called the permutation group of
n elements, where the group product is composition of mappings. We often write σ τ
instead of σ ◦ τ for two permutations σ and τ .
591
13.4 Remark We have an injective homomorphism φ : Sn−1 → Sn given by

φ (σ )(k) = σ (k) for k ∈ In−1 and φ (σ )(n) = n, so that Sn−1 may be identified with
the subgroup of Sn consisting of all permutations leaving n fixed.
13.5 Definition A transposition τ in Sn is a permutation which interchanges

two numbers in In and leaves all the other numbers fixed, i.e., there are two different
numbers i, j ∈ In such that τ (i) = j, τ ( j) = i and τ (k) = k for k ∈ In − {i, j}.
13.6 Remark We have that τ −1 = τ and τ 2 = id for every transposition τ .
13.7 Proposition Every permutation in Sn can be expressed as a product of

transpositions.
PROOF : We prove the proposition by induction on n. We have that S1 only consists

of the identity map which can be considered as a product of transpositions where the
number of factors is zero.
Assume next the the assertion in the proposition is true for Sn−1, and let σ be
a permutation in Sn . If σ (n) = k is different from n, we let τ be the transposition
in Sn which interchanges n and k, otherwise we let τ be the identity. Then we have
that τσ (n) = n, so that τσ = φ (σ ′ ) for a permutation of σ ′ ∈ Sn−1 , where φ is the
injective homomorphism defined in Remark 13.4. By the induction hypothesis there
exist transpositions τ1′ , ..., τk′ in Sn−1 such that σ ′ = τ1′ · · · τk′ . Then we have that σ =
τ −1 τ1 · · · τk where τi = φ (τi′ ) is a transposition in Sn for i = 1 , ..., k, and this completes
the proof of the proposition.
13.8 Let A × · · · × A = An be the set of n-tuples of elements in a set A, and let σ

be a permutation in Sn . Then we have a map πσ : An → An defined by
πσ (x1 , ..., xn ) = (xσ (1) , ..., xσ (n) ). (1)
The n-tuple (x1 , ..., xn ) is a map x : In → A with x(i) = xi , and we have that πσ (x) =
x ◦ σ . From this it follows that
(πσ ◦ πτ )(x) = (x ◦ τ ) ◦ σ = x ◦ (τσ ) = πτσ (x)
so that
πσ ◦ πτ = πτσ
for all τ , σ ∈ Sn .
If h : C → A is a map from a set C into A, we define hn : Cn → An by
hn (x1 , ..., xn ) = (h(x1 ) , ..., h(xn )) ,
and we let πσ′ : Cn → Cn be the map on Cn corresponding to πσ and which is also

PRELIMINARIES 593
defined by (1). If the n-tuple (x1 , ..., xn ) is again interpreted as a map x : In → C with
x(i) = xi , we have that hn (x) = h ◦ x. From this it follows that
(πσ ◦ hn )(x) = (h ◦ x) ◦ σ = h ◦ (x ◦ σ ) = (hn ◦ πσ′ )(x)
so that
πσ ◦ hn = hn ◦ πσ′
for every σ ∈ Sn .
13.9 Given a map f : An → B and a permutation σ in Sn , we define the map

f : A → B by f σ = f ◦ πσ so that
σ n
f σ (x1 , ..., xn ) = f (xσ (1) , ..., xσ (n) ) .
If τ and σ are two permutations in Sn , we have that
f σ τ = f ◦ πσ τ = f ◦ πτ ◦ πσ = ( f τ )σ .
Given a map h : C → A as in 13.8, we have that
f σ ◦ hn = f ◦ πσ ◦ hn = f ◦ hn ◦ πσ′ = ( f ◦ hn )σ .
The map f ◦ hn is called the pull-back of f by h and is usually denoted by h∗ ( f ). The

last relation may then be written as
h∗ ( f σ ) = h∗ ( f )σ .
If B is a ring, we may define the sum and product of two function f : An → B and
g : An → B as usual. We then have that
( f + g)σ = f σ + gσ and ( f g)σ = f σ gσ
for any permutation σ in Sn . We also have that (− f )σ = − f σ .
13.10 Theorem There is a unique homomorphism ε : Sn → {1, −1} such that

ε (τ ) = −1 for every transposition τ .
ε (σ ) is called the sign of the permutation σ . We say that σ is even if ε (σ ) = 1 and
odd if ε (σ ) = −1. The kernel An of ε , consisting of the even permutations, is called
the alternating group of n elements.
PROOF : If σ is a permutation in Sn , we know by Proposition 13.7 that it may be
expressed as a product of transpositions
σ = τ1 · · · τm .
From the conditions in the theorem it then follows that ε (σ ) = (−1)m , and this proves
the uniqueness of ε .
To prove existence, we must show that the sign (−1)m is independent of the
expression of σ as a product of transpositions, i.e., if σ = τ1′ · · · τk′ is another such
expression, then (−1)k = (−1)m .
Let S = {(i, j)∈In2 | i < j}, and consider the polynomials ∆i j : Zn → Z defined by
∆i j (x1 , ..., xn ) = x j − xi for (i, j) ∈ S. Let
∆= ∏ ∆i j .
(i, j)∈S
If τ is a transposition in Sn interchanging the integers r and s in In , where r < s, we

have that
∆τ = ∏ ∆τi j
(i, j)∈S
where ∆τrs
= −∆rs . We see that the factors ∆i j which do not have r or s as an index
remain unchanged when we apply τ . The other factors may be grouped in pairs of
the following three types
1) ∆rk ∆sk where s < k ,
2) ∆rk ∆ks where r < k < s ,
3) ∆kr ∆ks where k < r ,
which are also all unchanged when we apply τ .

Hence we have that ∆τ = −∆ for every transposition τ . Using the two expressions
for σ given above, we see that
∆σ = (−1)k ∆ = (−1)m ∆
which shows that (−1)k = (−1)m and completes the proof of the theorem.
13.11 Remark We see that if a permutation is expressed as a product of trans-

positions, then the number of factors is either always odd or always even. In the
first case it is an odd permutation with sign −1, and in the second case it is an even
permutation with sign 1.
GROUP ACTIONS
13.12 Definition Let S be a set and G a group with identity element e. By an
operation or an action of G on S on the left we mean a map µ : G × S → S , where
µ (g, s) is denoted by gs for g ∈ G and s ∈ S, satisfying
(gh) s = g (hs) and es = s

PRELIMINARIES 595
for every g, h ∈ G and s ∈ S. If there is such an operation µ , we say that G operates

or acts on S on the left, or that S is a G - set.
For each g ∈ G, we let L g : S → S be the map defined by L g (s) = gs for s ∈ S.
It is a bijective map also called a permutation of S, and we obtain a homomorphism
g 7→ L g from G into the group of permutations of S called a representation of G.
If this representation is injective, it is also said to be faithful, and G is said to act
effectively on S. This means that the only element g ∈ G with gs = s for all s ∈ S, is
the identity element e.
If we have the stronger condition that gs = s for some s ∈ S implies g = e, then
G is said to act without fixed point or freely on S.
We say that G acts transitively on S if there for any pair of elements s,t ∈ S always
exists a group element g ∈ G with gs = t .
If S and S′ are two G - sets, then a map f : S → S′ is called a G - map, or it is said
to be equivariant with respect to the two actions of G , if
f (gs) = g f (s)
for every g ∈ G and s ∈ S.
13.13 Remark Similarly, we define an operation of G on S on the right to be a

map ν : S × G → S , where ν (s, g) is denoted by sg for g ∈ G and s ∈ S, satisfying
s (gh) = (sg) h and se = s
for every g, h ∈ G and s ∈ S. For each g ∈ G, we let R g : S → S be the bijective map

defined by R g (s) = sg for s ∈ S.
13.14 Example A group G operates on itself on the left by left translation. If

g ∈ G, we define the left translation Lg : G → G by Lg (h) = gh for h ∈ G, and we
have an operation µ : G × G → G given by µ (g, h) = gh = Lg (h) .
Similarly, if H is a subgroup of G, then G operates by left translation on the set
G/H = {gH|g ∈ G} of left cosets. Indeed, if g1 ∈ G, then Lg1 (g2 H) = (g1 g2 )H .
The canonical projection π : G → G/H defined by π (g) = gH is a G - map.
13.15 Definition Let G be a group which operates on a set S on the left, and
let s ∈ S. Then Gs = {gs|g ∈ G} is a subset of S called the orbit of s under G, and
Gs = {g ∈ G|gs = s} is a subgroup of G called the isotropy group of s in G.
13.16 Proposition If G is a group which operates on a set S on the left, then two
orbits under G are either disjoint or equal. The orbits therefore form a partition of the
set S.
PROOF : If two orbits Gs1 and Gs2 have an element s in common, then s = g1 s1 =
g2 s2 for group elements g1 , g2 ∈ G. Hence we have that Gs = G gi si = Gsi for
i = 1, 2.
13.17 Proposition If G is a group which operates on a set S on the left, then

Gg s = g Gs g−1 for every g ∈ G and s ∈ S.
PROOF : If h ∈ Gs , then (ghg−1)(gs) = (gh)s = g(hs) = gs , which implies that
g Gs g−1 ⊂ Gg s . By replacing g and s with g−1 and gs , respectively, in the last
inclusion, we have that g−1 Gg s g ⊂ Gs so that Gg s ⊂ g Gs g−1 .
13.18 Proposition Let G be a group which operates on a set S on the left, and let
s ∈ S. If H = Gs is the isotropy group of s in G, and π : G → G/H is the canonical
projection, then the map φ : G → S given by φ (g) = gs induces an injective G - map
φ : G/H → S with φ ◦ π = φ . The image of φ is the orbit Gs .
PROOF : We have that φ (g1 ) = φ (g2 ) ⇔ g1 s = g2 s ⇔ (g−1 2 g1 ) s = s ⇔
g−1
2 g1 ∈ H ⇔ g1 H = g2 H . Hence we have a well-defined injective map φ : G/H →
S given by
φ (gH) = gs
for g ∈ G, which satisfies φ ◦ π = φ .

Since both π and φ are G - maps and π is surjective, it follows that φ is also a
G - map. Indeed, we have that φ (g1 π (g2 )) = φ (π (g1 g2 )) = g1 φ (π (g2 )) for every
g1 , g2 ∈ G.
CATEGORIES AND FUNCTORS

13.19 Definition By a category C we mean a class of objects Ob(C ) such that
for each ordered pair of objects X,Y ∈ Ob (C ) we have a set Mor(X,Y ), called the
set of morphisms from X to Y , and for each ordered triple of objects X,Y, Z ∈ Ob (C )
a map
Mor (Y, Z) × Mor(X,Y ) → Mor (X, Z) ,
called the law of composition , sending a pair of morphisms f ∈ Mor (X,Y ) and g ∈
Mor (Y, Z) to a morphism g ◦ f ∈ Mor (X, Z), satisfying the following three axioms:
(i) The sets Mor (X,Y ) and Mor (X ′ ,Y ′ ) are disjoint unless X = X ′ and Y = Y ′ .
(ii) The law of composition is associative, i.e., if f ∈ Mor (X,Y ), g ∈ Mor (Y, Z)
and h ∈ Mor (Z,W ), then (h ◦ g) ◦ f = h ◦ (g ◦ f ).
(iii) For every object Y there is a morphism id Y ∈ Mor (Y,Y ), called the identity
morphism of Y , such that id Y ◦ f = f for every f ∈ Mor (X,Y ) and g ◦ id Y = g
for every g ∈ Mor (Y, Z).
PRELIMINARIES 597
A morphism f ∈ Mor (X,Y ) is also denoted by f : X → Y .
13.20 Definition A morphism f : X → Y is called an isomorphism if there is a

morphism g : Y → X such that g ◦ f = idX and f ◦ g = id Y . The morphism g is called
the inverse of f and is denoted by f −1 . If X = Y , f is also called an automorphism.
13.21 Definition By a covariant functor F from a category C to a category D

we mean a rule which to each object X in C associates an object F(X) in D , and to
each morphism f : X → Y in C associates a morphism F( f ) : F(X) → F(Y ) in D ,
satisfying the following two conditions:
(i) If f : X → Y and g : Y → Z are two morphisms in C , then F(g ◦ f ) =
F(g) ◦ F( f ) .
(ii) For each object X in C we have that F(idX ) = id F(X) .
If we with f : X → Y associate a morphism F( f ) : F(Y ) → F(X) in D going in
the opposite direction, and replace the formula for composition in (i) by F(g ◦ f ) =
F( f ) ◦ F(g) , then F is called a contravariant functor from C to D .
A functor F from C to D is often denoted by F : C → D or by the transforma-
tion rule X 7→ F(X) and f 7→ F( f ) for each object X and morphism f in C .
13.22 Remark A functor F : C → D transforms isomorphisms in C to isomor-

phisms in D .
13.23 Definition If F : C → D and G : D → E are two functors, we define

their composed functor G ◦ F : C → E by G ◦ F (X) = G (F (X)) and G ◦ F ( f ) =
G (F ( f )) for each object X and morphism f in C . The composition G ◦ F is covari-
ant if F and G are of the same variance, otherwise it is contravariant.
13.24 Definition A category C is called a subcategory of the category D if there

is a covariant functor I : C → D , called the inclusion functor , given by I(X) = X
and I( f ) = f for each object X and morphism f in C .
This means that Ob (C ) ⊂ Ob (D) , and that Mor C (X,Y ) ⊂ Mor D (X,Y ) for
every pair X,Y ∈ Ob (C ). Moreover, for each pair of morphisms f : X → Y and
g : Y → Z in C , their composition in C coincides with their composition in D , and
for each object X in C , the identity morphism in Mor C (X, X) coincides with the
identity morphism in Mor D (X, X).
If in addition Mor C (X,Y ) = Mor D (X,Y ) for every pair X,Y ∈ Ob (C ), C is
called a full subcategory of D .
CONNECTIVITY
13.25 Definition A topological space X is said to be connected if it is not a

disjoint union of two nonempty open subsets. A subset A of X is called connected if
it is connected considered as a subspace of X.
A topological space X is said to be locally connected if every point in X has a
basis of connected neighbourhoods.
13.26 Proposition A topological space X is connected if and only if one of the

following equivalent conditions is satisfied:
(1) X it is not a disjoint union of two nonempty closed subsets.
(2) X and 0/ are the only subsets of X which are both open and closed.
PROOF : If two complimentary sets in X are open, then they are also closed, and
vice versa, so we may replace open by closed in Definition 13.25. This gives the
equivalent condition (1).
The existence of nonempty open complementary sets U and U c in X is equivalent
to the existence of a set U in X different from X and 0/ which is both open and closed.
This gives the condition (2).
13.27 Example The closed interval [ a, b ] in R is connected. To prove this, sup-

pose that [ a, b ] is the disjoint union of the nonempty closed subsets A and B. Without
loss of generality, we may assume that b ∈ B . If c = sup A , then c ∈ A since A is
closed. Hence we have that c < b and < c, b ] ⊂ B , and this implies that c ∈ B since
B is closed, contradicting the fact that A and B are disjoint.
13.28 Proposition If f : X → Y is a continuous map from a connected topological

space X, then the image f (X) is also connected.
PROOF : If f (X) is the disjoint union of the nonempty subsets U and V which
are open relative to f (X) , then X = f −1 (U) ∪ f −1 (V ) is also a disjoint union of
nonempty open sets, contradicting the fact that X is connected.
13.29 Proposition If {Aλ |S λ ∈ Λ} is a family of connected subsets of X whose

intersection is nonempty, then Aλ is connected.
T S
PROOF : Let x0 be a point in Aλ , and let A = Aλ . Suppose that A is the disjoint
union of two subsets U and V which are open relative to A, and that U is the one
containing x0 . Then Aλ ∩U and Aλ ∩V are open in Aλ , and since x0 ∈ Aλ ∩U and Aλ
is connected, it follows that Aλ ∩V = 0/ for every λ ∈ Λ. Hence we have that V = 0/
which shows that A is connected.
PRELIMINARIES 599
13.30 Corollary A topological space X is connected if and only if each pair of

points in X belongs to a connected subset of X.
PROOF : If X is connected, then the last assertion is obviously true if we let the
connected subset be X itself.
Conversely, suppose that the last assertion in the corollary holds, and fix a point
x0 in X. For each point x ∈ X there is then a connected subset Ax of X containing x0
and x. Since X is the union of the family {Ax |x ∈ X} whose intersection contains x0 ,
it follows from Proposition 13.29 that X is connected.
13.31 Example A set A ⊂ Rn is said to be convex if x, y ∈ A implies that x +

t (y − x) ∈ A for 0 ≤ t ≤ 1. By Corollary 13.30 every convex set A in Rn is connected
since the line segment {x + t (y − x)|0 ≤ t ≤ 1} is the image of the continuous map
σ : [ 0, 1 ] → A from the connected set [ 0, 1 ], defined by σ (t) = t x + (1 − t) x for
t ∈ [ 0, 1 ]. In particular, the open and closed balls D n = {y ∈ Rn | k y k < 1} and
E n = {y ∈ Rn | k y k ≤ 1} in Rn are connected.
13.32 Example The n-sphere Sn = {x ∈ Rn+1 | kxk = 1} is connected when

n ≥ 1 by Proposition 13.29, since it is the union of the closed half spheres S 1 =
{x ∈ Sn |xn+1 ≥ 0} and S 2 = {x ∈ Sn |xn+1 ≤ 0} with nonempty intersection S 1 ∩
S 2 = {x ∈ Sn |xn+1 = 0}. These half spheres are connected as they are the images
of the continuous maps f1 : E n → Rn+1 and f2 : E n → Rn+1 from the closed ball
E n = {y ∈ Rn | k y k ≤ 1}, given by
1 1
f1 (y) = (y, (1− k y k2 ) 2 ) and f2 (y) = (y, − (1− k y k2 ) 2 )
for y ∈ E n .
13.33 Proposition If X and Y are two connected topological spaces, then their
product X × Y is also connected.
PROOF : Let (x1 , y1 ) and (x2 , y2 ) be two arbitrary points in X × Y . Then the sets
X × {y1 } and {x2 } × Y are connected, since they are the images of the continuous
maps f : X → X × Y and g : Y → X × Y given by f (x) = (x, y1 ) and g(y) = (x2 , y),
respectively. These two sets have the point (x2 , y1 ) in common, so their union is
connected by Proposition 13.29. Hence the points (x1 , y1 ) and (x2 , y2 ) belong to the
connected subset ( X × {y1 } ) ∪ ( {x2 } × Y ) of X × Y , and the proposition follows
from Corollary 13.30.
13.34 Proposition Let {Xλ |λ ∈ Λ} be a family of connected topological spaces.

Then the product
X= ∏ Xλ
λ ∈Λ
is connected.
PROOF : If the indexing set Λ is finite, the proposition follows by induction from
Proposition 13.33. In the general case, suppose that X is a disjoint union of two
nonempty open sets U and V , and let x ∈ U and y ∈ V . Choose an open neighbour-
hood
W = ∏ Wλ
λ ∈Λ
of y contained in V , where each Wλ is an open neighbourhood of yλ , and where
Wλ = Xλ for all indices λ except those contained in a finite subset Λ0 of Λ. Let z be
the point in X defined by

yλ for λ ∈ Λ0
zλ = .
xλ for λ ∈ Λ − Λ0
By the first part of the proof, we know that the space
X0 = ∏ Xλ
λ ∈Λ0
is connected, and hence the same is true for the subspace
Z= ∏ Zλ
λ ∈Λ
of X defined by
Xλ for λ ∈ Λ0
Zλ = ,
{xλ } for λ ∈ Λ − Λ0
as Z is the image of the continuous map f : X0 → X given by

uλ for λ ∈ Λ0
f (u)λ = .
xλ for λ ∈ Λ − Λ0
This however contradicts the fact that Z is the disjoint union of the sets Z ∩U and
Z ∩ V which are open in Z and nonempty since x ∈ Z ∩ U and z ∈ Z ∩ V . Hence we
conclude that X must be connected.
13.35 Proposition If A is a connected subset of a topological space X, and if

A ⊂ B ⊂ A, then B is also connected.
PROOF : Suppose that B is the disjoint union of the subsets U1 and U2 which are
open relative to B. Then the sets Vi = A ∩Ui for i = 1, 2 are open in A. If Ui contains
a point x, then Ui is a neighbourhood of x in B and must contain a point from A, as
A is dense in B. Hence Ui 6= 0/ implies that Vi 6= 0/ . Since A is connected, it therefore
follows that either U1 or U2 must be empty, showing that B is also connected.
13.36 Let x be a point in a topological space X. Then the set {x} is connected, and
the union of every connected subsets of X containing x is connected by Proposition
13.29. This is the largest connected subset containing x, and it is called the connected
PRELIMINARIES 601
component or just the component of x in X. By Proposition 13.35 it follows that it is

closed.
The connected components of two points x and y in X are either disjoint or equal
by Proposition 13.29. Hence the connected components form a partition of X into
closed connected subsets.
13.37 Proposition If A is a nonempty subset of a topological space X which is

both open and closed, then A is a union of connected components in X. If A is also
connected, then it is a connected component.
PROOF : Let A be a nonempty subset of X which is both open and closed. If C is a
connected component in X containing a point x ∈ A, then C ∩ A is a nonempty subset
of C which is both open and closed relative to C. Since C is connected, it follows
that C ∩ A = C so that C ⊂ A, and this completes the proof of the first assertion in the
proposition.
If A is connected, then we also have A ⊂ C, which shows that A coincides with C
and therefore is a connected component.
13.38 Proposition A topological space X is locally connected if and only if the

connected components of open subsets of X are open. If X is locally connected, then
every point in X has a basis of open connected neighbourhoods.
PROOF : Suppose that X is locally connected, and let C be a connected component
of an open set U in X. Then each point x ∈ C has a connected neighbourhood V
contained in U. As C is the largest connected subset of U containing x, we have that
V ⊂ C, showing that C is a neighborhood of x and therefore is open.
Conversely, suppose that the connected components of open subsets of X are
open, and let U be an open neighbourhood of a point x ∈ X. Then the connected
component C of U containing x is an open connected neighbourhood of x with C ⊂
U. This shows that X is locally connected, and it also shows the last part of the
proposition.
13.39 Corollary In a locally connected topological space X, the connected com-

ponents are open as well as closed.
PROOF : Follows from 13.36 and Proposition 13.38.
HOMOTOPY THEORY
13.40 In this chapter we let I denote the closed unit interval [0, 1] and ∂ I denote
its boundary {0, 1}.
13.41 Definition A pair (X, A) consisting of a topological space X and a subspace

A is called a topological pair . By a continuous map f : (X, A) → (Y, B) between two
topological pairs we mean a continuous map f : X → Y such that f (A) ⊂ B. The
set of all continuous maps from (X, A) to (Y, B) is denoted by C(X, A;Y, B). We thus
obtain a category T 2 of topological pairs and continuous maps.
The topological pair (X, 0)/ is identified with X, and the category of topological
spaces and continuous maps is a full subcategory of T 2 denoted by T .
If A = {x0 } for a point x0 ∈ X, then the topological pair (X, A) is called a pointed
topological space with base point x0 , and it is usually denoted simply by (X, x0 ). The
category of pointed topological spaces and continuous maps is a full subcategory of
T 2 denoted by T0 .
The cartesian product between a topological pair (X, A) and the unit interval I is
defined by (X, A) × I = (X × I, A × I).
13.42 Definition Let f , g : (X, A) → (Y, B) be two continuous maps between the
topological pairs (X, A) and (Y, B), and let X ′ be a subset of X. Then we say that f
is homotopic to g relative to X ′ , and we write f ≃ g rel X ′ , if there is a continuous
map F : (X, A) × I → (Y, B) such that
F(x, 0) = f (x) for x ∈ X,

F(x, 1) = g(x) for x ∈ X,
F(x,t) = f (x) = g(x) for x ∈ X ′ , t ∈ I .
Such a map F is called a homotopy from f to g relative to X ′ , and we write

F : f ≃ g rel X ′ . In particular, it is necessary that f|X ′ = g|X ′ . If X ′ = 0,
/ we omit
/ and write simply f ≃ g and F : f ≃ g .
the phrase "relative to 0",
13.43 Proposition Let {Ak }nk=1 be a finite closed covering of a topological space
X. Then a map f : X → Y into a topological space Y is continuous if and only if f |Ak
is continuous for every k.
PROOF : The only if part follows since f | Ak = f ◦ ik , where ik : Ak → X is the inclu-
sion map which is continuous for every k.
Conversely, suppose that f |Ak is continuous for every k, and let V be a closed
subset of Y . Then f −1 (V ) ∩ Ak = ( f |Ak ) −1 (V ) is closed in Ak , and therefore also
closed in X as Ak is closed. Since
n
[
f −1 (V ) = f −1 (V ) ∩ Ak ,
k=1
it follows that f −1 (V ) is closed in X, thus showing that f is continuous.
13.44 Proposition Let (X, A) and (Y, B) be topological pairs, and let X ′ ⊂ X.
Then homotopy relative to X ′ is an equivalence relation in C(X, A;Y, B).
PRELIMINARIES 603
PROOF : If f ∈ C(X, A;Y, B), we define the constant homotopy F : f ≃ f rel X ′

by F(x,t) = f (x) for x ∈ X and t ∈ I. This shows that homotopy relative to X ′ is
reflexive .
To prove that it is symmetric , let F : f ≃ g rel X ′ be any homotopy relative to X ′ .
We then have an inverse homotopy G : g ≃ f rel X ′ defined by G(x,t) = F(x, 1 − t)
for x ∈ X and t ∈ I.
Finally, we show that homotopy relative to X ′ is transitive . If F : f ≃ g rel X ′
and G : g ≃ h rel X ′ are two homotopies relative to X ′ , we define their combined
homotopy H : f ≃ h rel X ′ by

F(x, 2t) for x ∈ X , t ∈ [0, 1/2]
H(x,t) =
G(x, 2t − 1) for x ∈ X , t ∈ [1/2, 1]
Since F(x, 1) = g(x) = G(x, 0) for every x ∈ X, H is well defined, and it follows from
Proposition 13.43 that H is continuous.
13.45 Remark The equivalence class of a continuous map f ∈ C(X, A;Y, B) with
respect to the above equivalence relation is called the homotopy class of f relative
to X ′ and is denoted by [ f ]X ′ . If X ′ = 0,
/ we omit the reference to X ′ .
13.46 Proposition Let f0 , f1 : (X, A) → (Y, B) and g0 , g1 : (Y, B) → (Z,C) be

continuous maps, and let X ′ and Y ′ be subsets of X and Y , respectively, with f0 (X ′ ) ⊂
Y ′ . If f0 ≃ f1 rel X ′ and g0 ≃ g1 rel Y ′ , then g0 ◦ f0 ≃ g1 ◦ f1 rel X ′ .
PROOF : Let F : f0 ≃ f1 rel X ′ and G : g0 ≃ g1 rel Y ′ be homotopies. From these
we obtain the homotopies
g0 ◦ F : g0 ◦ f0 ≃ g0 ◦ f1 rel X ′
and
G ◦ ( f1 × id) : g0 ◦ f1 ≃ g1 ◦ f1 rel X ′ ,
so that g0 ◦ f0 ≃ g1 ◦ f1 rel X ′ by Proposition 13.44.
13.47 Definition If f ∈ C(X, A;Y, B) and g ∈ C(Y, B; Z,C), we define the compo-
sition of the hopotopy classes [ f ] and [g] by [g] ◦ [ f ] = [g ◦ f ]. This is well defined
by Proposition 13.46, and we thus obtain a category h2 of topological pairs and
homotopy classes of continuous maps called the homotopy category of topological
pairs . We have a canonical functor F : T 2 → h2 given by F((X, A)) = (X, A) and
F( f ) = [ f ].
As in Definition 13.41, we have the full subcategories h and h0 of h2 , where the
objects are topological spaces and pointed topological spaces, respectively.
13.48 Definition A continuous map f : (X, A) → (Y, B) is called a homotopy

equivalence if the homotopy class [ f ] is an isomorphism in h2 , i.e., if there is a
continuous map g : (Y, B) → (X, A) such that g ◦ f ≃ idX and f ◦ g ≃ idY . The map
g is called a homotopy inverse of f .
Two topological pairs (X, A) and (Y, B) are said to be homotopically equivalent or
of the same homotopy type if there exists a homotopy equivalence f : (X, A) →
(Y, B) .
13.49 Definition A continuous map f : (X, A) → (Y, B) is said to be null ho-

motopic if it is homotopic to a constant map. In particular, if the identity map
idX : (X, A) → (X, A) is null homotopic, then (X, A) is said to be contractible . If
c : (X, A) → (X, A) is the constant map mapping X to a point x0 ∈ A, then a homo-
topy F : idX ≃ c is called a contraction of (X, A) to x0 .
13.50 Example A set V ⊂ Rn is said to be star-shaped with respect to a point

x0 ∈ V if x ∈ V implies that x0 + t (x − x0 ) ∈ V for 0 ≤ t ≤ 1. Every star-shaped set V
in Rn is contractible since the map F : V × I → V defined by F(x,t) = t x0 + (1 −t) x
is a contraction of V to x0 . In particular, the Euclidean space Rn itself and every
convex subset of Rn is star-shaped and therefore contractible.
More generally, we have that every topological pair (V, A), where V and A are
star-shaped with respect to a point x0 ∈ A, is contractible since F is a contraction of
(V, A) to x0 .
13.51 Proposition A topological space X is contractible if and only if it has the

same homotopy type as a one-point space.
PROOF : Suppose first that X is contractible, and let F be a contraction of X to a
point x0 ∈ X. If f : X → {x0 } is the canonical map and g : {x0 } → X the inclusion
map, we have that F : idX ≃ g ◦ f and f ◦ g = id {x0 } , showing that f is a homotopy
equivalence from X to {x0 } .
Conversely, assume that X is of the same homotopy type as a one-point space
{x0 } , and let f : X → {x0 } be a homotopy equivalence with a homotopy inverse
g : {x0 } → X . Then g ◦ f : X → X is a constant map, and since idX ≃ g ◦ f , it follows
that X is contractible.
13.52 Definition Let X be a topological space, and let x0 , x1 ∈ X. By a path in X

from x0 to x1 we mean a continuous map σ : I → X with σ (0) = x0 and σ (1) = x1 .
The points x0 and x1 are called the initial and final point of σ , respectively. If x0 = x1 ,
σ is called a closed path or a loop at x0 .
If σ maps I to a single point x0 ∈ X, σ is called the constant path at x0 and is
denoted by <x0 >. Given a path σ from x0 to x1 , we define the inverse path σ −1 from
x1 to x0 by σ −1 (t) = σ (1 − t). For two paths σ and τ with σ (1) = τ (0), we define
the combined path σ ∗ τ by

σ (2t) for t ∈ [0, 1/2]
σ ∗ τ (t) =
τ (2t − 1) for t ∈ [1/2, 1]
13.53 We say that two points x0 and x1 in a topological space X are equivalent,
and write x0 ∼ x1 , if there is a path in X from x0 to x1 . It follows from Definition
PRELIMINARIES 605
13.52 that ∼ is an equivalence relation, and the equivalence class of a point x ∈ X is

called the path component of x and consists of all points which can be connected to
x by a path in X. The set of all path components in X is denoted by π0 (X).
If f : X → Y is a continuous map and σ is a path in X from x0 to x1 , then f ◦ σ
is a path in Y from f (x0 ) to f (x1 ). Hence x0 ∼ x1 implies that f (x0 ) ∼ f (x1 ), and we
obtain an induced map f∗ : π0 (X) → π0 (Y ) mapping the path component of a point
x in X to the path component of f (x) in Y .
If f , g : X → Y are homotopic, then f∗ = g∗ . Indeed, if F : f ≃ g is a homotopy,
we have for each point x ∈ X a path τ in Y from f (x) to g(x) given by τ (t) = F(x,t)
for t ∈ I, showing that f (x) and g(x) belong to the same path component in Y .
We thus obtain a covariant functor π0 : h → S from the homotopy category h
of topological spaces to the category S of sets, assigning to a topological space X
the set π0 (X) of path components in X, and to a homotopy class [ f ] the map f∗ .
13.54 Definition A topological space X is said to be pathwise connected if every

pair of points in X can be connected with a path. A subset A of X is called pathwise
connected if it is pathwise connected considered as a subspace of X, i.e., if every pair
of points in A can be connected with a path lying entirely in A.
A topological space X is said to be locally pathwise connected if every point in
X has a basis of pathwise connected neighbourhoods.
13.55 Proposition The path component of a point x in a topological space X is

the largest pathwise connected subset of X containing x.
PROOF : Let C be the path component of x. Then we have that A ⊂ C for every path-
wise connected subset A of X containing x, since any point y ∈ A can be connected
to x with a path in A which is certainly also a path in X.
It only remains to show that C itself is pathwise connected. If y ∈ C, there is a path
σ in X from x to y, and we contend that σ actually is a path in C. In fact, for every
t ∈ I, we have a path σt in X from x to σ (t) given by σt (s) = σ (st) for s ∈ I, showing
that σ (t) ∈ C. This proves that every pair of points in C can be connected with a path
lying entirely in C, thereby completing the proof that C is pathwise connected.
13.56 Proposition A topological space X is locally pathwise connected if and

only if the path components of open subsets of X are open. If X is locally pathwise
connected, then every point in X has a basis of open pathwise connected neighbour-
hoods.
PROOF : Suppose that X is locally pathwise connected, and let C be a path com-
ponent of an open set U in X. Then each point x ∈ C has a pathwise connected
neighbourhood V contained in U. As C is the largest pathwise connected subset of U
containing x, we have that V ⊂ C, showing that C is a neighborhood of x and therefore
is open.
Conversely, suppose that the path components of open subsets of X are open, and
let U be an open neighbourhood of a point x ∈ X. Then the path component C of U
containing x is an open pathwise connected neighbourhood of x with C ⊂ U. This
shows that X is locally pathwise connected, and it also shows the last part of the
proposition.
13.57 Corollary In a locally pathwise connected space X, the path components

are open as well as closed.
PROOF : By 13.53 and Proposition 13.56 we know that the path components are open
sets forming a partition of X. Hence they are also closed in X.
13.58 Proposition A pathwise connected space X is connected.

PROOF : If X is pathwise connected, then every pair of points x0 and x1 in X can
be connected by a path σ : I → X. Using Proposition 13.28 we have that σ (I) is a
connected subset of X containing x0 and x1 . Hence the result follows from Corollary
13.30.
13.59 Corollary A locally pathwise connected space X is locally connected.
13.60 Corollary Each connected component in a topological space X is a union

of path components in X.
PROOF : Let C be a connected component in X. Then it follows from Propositions
13.55 and 13.58 that the path component in X of a point x ∈ C is connected, and
therefore is contained in C.
13.61 Proposition In a locally pathwise connected space X the path components

coincide with the connected components.
PROOF : By Proposition 13.55 and 13.58 and Corollary 13.57 we know that each
path component in X is a nonempty connected set which is both open and closed,
and therefore is a connected component in X by Proposition 13.37.
13.62 Corollary A locally pathwise connected space X is pathwise connected if

and only if it is connected.
13.63 Proposition Topological spaces of the same homotopy type always have
the same number of path components. In particular, every contractible space is path-
wise connected.
PROOF : If f : X → Y is a homotopy equivalence, then the induced map f∗ : π0 (X) →
π0 (Y ) is a bijection by 13.53 and Remark 13.22.
13.64 Proposition If f : X → Y is continuous and X is pathwise connected, then

f (X) is also pathwise connected.
PRELIMINARIES 607
13.65 Definition We say that two paths σ and τ in a topological space X are
homotopic, and write σ ≃ τ , if they are homotopic relative to ∂ I. In particular, we
must have that σ (0) = τ (0) and σ (1) = τ (1). A homotopy F from σ to τ relative to
∂ I is called simply a homotopy from σ to τ , and it is denoted by F : σ ≃ τ .
The homotopy class of σ relative to ∂ I is denoted by [σ ] and is called the path
class of σ . The points σ (0) and σ (1) are called the initial and final point of [σ ],
respectively.
13.66 Proposition Let σ0 and σ1 be homotopic paths in X from x0 to x1 , and τ0

and τ1 be homotopic paths in X from x1 to x2 . Then the combined paths σ0 ∗ τ0 and
σ1 ∗ τ1 are homotopic.
PROOF : If F : σ0 ≃ σ1 and G : τ0 ≃ τ1 are homotopies, we define a homotopy
H : σ0 ∗ τ0 ≃ σ1 ∗ τ1 by

F(2s,t) for s ∈ [0, 1/2]
H(s,t) =
G(2s − 1,t) for s ∈ [1/2, 1]
Since F(1,t) = x1 = G(0,t) for every t ∈ I, H is well defined, and it follows from
Proposition 13.43 that H is continuous.
13.67 Definition If [σ ] is a path class from x0 to x1 and [τ ] a path class from x1

to x2 , we define the combined path class [σ ] ∗ [τ ] from x0 to x2 by [σ ] ∗ [τ ] = [σ ∗ τ ] .
13.68 Proposition For each topological space X we have a category P (X),

called the path category of X, where the objects are the points in X, the morphisms
from x1 to x0 are the path classes from x0 to x1 , and the law of composition is the
combination of path classes as defined above. The identity morphism of x0 is the path
class of the constant path at x0 .
PROOF : We first show that the law of composition is associative. If ρ , σ and τ are
paths in X with ρ (1) = σ (0) and σ (1) = τ (0), we have a homotopy F : (ρ ∗ σ ) ∗ τ ≃
ρ ∗ (σ ∗ τ ) defined by
 ρ σ τ
4s 1+t



ρ 1+t for 0 ≤ s ≤ 4 ✄ ✄

 ✄ ✄
F(s,t) = σ (4s − t − 1) for 1+t 4 ≤s≤ 4
2+t ✄ ✄


 ✄ ✄

 τ 4s−t−2 ✄ ✄
for 2+t 4 ≤s≤1
2−t ✄ ✄
ρ σ τ
showing that ( [ρ ] ∗ [σ ] ) ∗ [τ ] = [ρ ] ∗ ( [σ ] ∗ [τ ] ) .
Furthermore, for any path σ in X from x0 to x1 , we have the homotopies

G : <x0> ∗ σ ≃ σ and H : σ ∗ <x1 > ≃ σ defined by
σ
 ❆
 σ (0) for 0 ≤ s ≤ 1−t
2
❆
G(s,t) = ❆
 σ 2s+t−1 1−t ❆
1+t for 2 ≤s≤1 ❆
❆
<x0> σ
and
σ
 ✁
2s

 σ 1+t for 0 ≤ s ≤ 1+t
2 ✁
H(s,t) = ✁

 σ (1) 1+t ✁
for 2 ≤s≤1 ✁
✁
σ <x1 >
which shows that [<x0>] ∗ [σ ] = [σ ] and [σ ] ∗ [<x1>] = [σ ] and completes the proof
of the proposition.
13.69 Proposition Every morphism in P (X) is an isomorphism.

PROOF : If σ is a path in X from x0 to x1 , we have a homotopy F : σ ∗ σ −1 ≃ <x0>
defined by
 <x0>

 σ (0) for 0 ≤ s ≤ 2t

 ✁❆

 1 ✁ ❆

 σ (2s − t) t
for 2 ≤s≤ 2 ✁ ❆
F(s,t) = ✁ ❆



 σ (2 − 2s − t) for 12 ≤ s ≤ 1 − 2t ✁ ❆



 ✁ ❆
σ (0) for 1 − 2t ≤ s ≤ 1 σ σ −1
which shows that [σ ] ∗ [σ −1 ] = [<x0>] . By replacing σ with σ −1 in this formula, we

also have that [σ −1 ] ∗ [σ ] = [<x1 >] . Hence [σ ] is an isomorphism in P (X) with
inverse [σ ]−1 = [σ −1 ] .
13.70 If X is a topological space and x0 ∈ X, it follows from Proposition 13.68

and 13.69 that the path classes of the closed paths at x0 form a group, where the
group operation is combination of path classes as defined in Definition 13.67. This
group is called the fundamental group of X at x0 , and it is denoted by π1 (X, x0 ).
If f : (X, x0 ) → (Y, y0 ) is a continuous map, we have a homomorphism f∗ :
π1 (X, x0 ) → π1 (Y, y0 ) defined by f∗ ([σ ]) = [ f ◦ σ ]. It follows from Proposition 13.46
PRELIMINARIES 609
that f∗ is well defined, and that f∗ = g∗ for two homotopic maps f , g : (X, x0 ) →
(Y, y0 ) .
We thus obtain a covariant functor π1 : h0 → g , called the fundamental group
functor , from the homotopy category h0 of pointed topological spaces to the cate-
gory g of groups, assigning to a pointed topological space (X, x0 ) its fundamental
group π1 (X, x0 ), and to a homotopy class [ f ] the homomorphism f∗ .
In particular, every contractible pointed topological space has a trivial fundamen-
tal group.
13.71 To describe how the fundamental group π1 (X, x0 ) depends on the base
point x0 , we consider the covariant functor F : P (X) → g from the path category
P (X) to the category g of groups, which assigns to each point x0 the fundamental
group π1 (X, x0 ) and to to each morphism [σ ] : x0 → x1 the homomorphism
h[σ ] : π1 (X, x0 ) → π1 (X, x1 )
defined by h[σ ] ([ω ]) = [σ ] ∗ [ω ] ∗ [σ ]−1. From Proposition 13.69 it follows that every
h[σ ] is an isomorphism so that π1 (X, x0 ) ∼ = π1 (X, x1 ) whenever x0 and x1 are in the
same path component in X.
This isomorphism is determined up to an inner automorphism of the fundamental
group. More precisely, if [σ ], [τ ] : x0 → x1 are two morphisms from x0 to x1 , and if
[θ ] = [τ ] ∗ [σ ]−1 ∈ π1 (X, x1 ), we have that
h[τ ] ([ω ]) = [θ ] ∗ h[σ ] ([ω ]) ∗ [θ ]−1
which shows that h[σ ] and h[τ ] are conjugate by the inner automorphism of π1 (X, x1 )
defined by [θ ].
13.72 Definition A topological space X is said to be simply connected if it is

pathwise connected and has a trivial fundamental group at every point.
13.73 Example Every star-shaped subset V of Rn is simply connected. In fact,

V is pathwise connected, and if V is star-shaped with respect to x0 , then (V, x0 ) is
a contractible pointed topological space as described in Example 13.50. The result
therefore follows from 13.70 and 13.71.
13.74 Proposition A topological space X is simply connected if and only if the

following assertion is satisfied:
For each pair of points x0 and x1 in X, there is a unique path class from x0 to x1 .
PROOF : Suppose first that X is simply connected, and let x0 , x1 ∈ X. As X is path-
wise connected, there is a path σ from x0 to x1 . If τ is another such path, then
[σ ] ∗ [τ ]−1 ∈ π1 (X, x0 ) so that [σ ] ∗ [τ ]−1 = [<x0>] since π1 (X, x0 ) is trivial. Hence
it follows that [σ ] = [τ ].
Conversely, suppose that the above assertion is satisfied. Then X is pathwise con-
nected, and π1 (X, x0 ) is trivial for every x0 ∈ X, thus showing that X is simply con-
nected.
13.75 Proposition If f , g : X → Y are homotopic maps with f (x0 ) = y0 and

g(x0 ) = y1 , then there is a path σ in Y from y0 to y1 such that
f∗ = h [σ ] ◦ g∗ : π1 (X, x0 ) → π1 (Y, y0 ) .
PROOF : Let F : f ≃ g be a homotopy, and let σ be the path from y0 to y1 in Y

defined by σ (t) = F(x0 ,t). We want to show that
[ f ◦ ω ] = [σ ] ∗ [g ◦ ω ] ∗ [σ ]−1
for every closed path ω at x0 . Let H : I × I → Y be the map defined by H = F ◦ (ω ×

id) , and consider the paths α0 , α1 , β0 and β1 in I × I given by αi (s) = (s, i) and
βi (t) = (i,t) for i = 0, 1 . Since I × I is star-shaped and therefore simply connected,
it follows from Proposition 13.74 that α0 ≃ β0 ∗ α1 ∗ β1−1 so that
f ◦ ω = H ◦ α0 ≃ (H ◦ β0 ) ∗ (H ◦ α1 ) ∗ (H ◦ β1−1 ) = σ ∗ (g ◦ ω ) ∗ σ −1 ,
13.76 Proposition If f : X → Y is a homotopy equivalence with f (x0 ) = y0 ,

then the induced map f∗ : π1 (X, x0 ) → π1 (Y, y0 ) is an isomorphism.
PROOF : Let g : Y → X be a homotopy inverse of f , and let g(y0 ) = x1 and f (x1 ) = y1 .
Then f induces the maps f∗i : π1 (X, x i ) → π1 (Y, y i ) for i = 0, 1, where f∗0 = f∗ is the
induced map referred to in the proposition.
Since g ◦ f ≃ idX and f ◦ g ≃ idY , we have by Proposition 13.75 a path σ in X
from x1 to x0 with g∗ ◦ f∗0 = h [σ ] , and a path τ in Y from y1 to y0 with f∗1 ◦ g∗ = h [τ ] ,
so that the diagram
f∗0 ✲ π (Y, y )
π1 (X, x0 ) 1 0
✟
h [σ ] ✟✟ h [τ ]
✟
❄ ✙✟ g∗
✟ ❄
π1 (X, x1 ) ✲ π (Y, y )
1 1
f∗1
is commutative. As h [σ ] and h [τ ] are both isomorphisms, it follows that g∗ is both

surjective and injective. Hence we conclude that g∗ , and therefore also f∗ , is an iso-
morphism.
PRELIMINARIES 611
13.77 Corollary Every contractible space X is simply connected.

13.78 Proposition Let {Xλ |λ ∈ Λ} be a family of topological spaces, and let
X= ∏ Xλ
λ ∈Λ
be the product space. We choose a base point xλ in Xλ for each λ ∈ Λ , and use
x = {xλ |λ ∈ Λ} as a base point in X. Then the projections p λ : X → Xλ induce a
group isomorphism
p∗ : π1 (X, x) → ∏ π1 (Xλ , xλ )
λ ∈Λ
given by p ∗ λ = p λ ∗ for λ ∈ Λ.
PROOF : From 13.70 it follows that p∗ is a homomorphism, and it is also surjective
since
p ∗ ( [ { σλ |λ ∈ Λ } ] ) = { [ σλ ]|λ ∈ Λ }
for every closed path σλ in Xλ at xλ for λ ∈ Λ.
To show that it is injective, let σ = { σλ |λ ∈ Λ } and τ = { τλ |λ ∈ Λ } be two
closed paths in X at x with p ∗ ( [ σ ] ) = p ∗ ( [ τ ] ). Then we have a homotopy Fλ :
σλ ≃ τλ for each λ ∈ Λ , from which we obtain a homotopy F : σ ≃ τ given by
F = { Fλ |λ ∈ Λ } . This shows that [ σ ] = [ τ ] and completes the proof that p∗ is
injective.
COVERINGS
13.79 Definition Let p : Xe → X be a continuous map. Then an open subset U

of X is said to be evenly covered by p if p−1 (U) is a disjoint union of open subsets
Ueα of Xe , each of which is mapped homeomorphically onto U by p. The sets U eα are
called sheets over U.
A continuous map p : Xe → X is called a covering of X if every point x ∈ X has
an open neighbourhood which is evenly covered by p. The spaces Xe and X are called
the covering space and the base space of the covering p, respectively, and the map p
is also called the covering projection. For each point x ∈ X, the set p−1 (x) is called
the fibre over x.
13.80 Remark It follows immediately from the definition that a covering projec-
tion is surjective, and that each fibre is a discrete subspace of the covering space.
Indeed, if p : Xe → X is a covering projection and U is an evenly covered neigh-

bourhood of x ∈ X, then each sheet U eα over U is an open subset of Xe containing
exactly one point from the fibre p−1 (x), and every point in the fibre is contained in a
sheet.
13.81 Proposition If U is an open set in X which is evenly covered by a con-

tinuous map p : Xe → X , then every open subset of U is also evenly covered by p.
PROOF : Suppose that [
p−1 (U) = eα
U
α ∈A
is a disjoint union of open subsets Ueα of Xe which are mapped homeomorphically

onto U by p. If V is an open subset of U, we let
Veα = p−1 (V ) ∩ U
eα = (p| e )−1 (V )
Uα
for every α ∈ A. Then each Veα is an open subset of Xe which is mapped homeomor-
phically onto V by p, and we have a disjoint union
[
p−1 (V ) = Veα
α ∈A
which shows that V is evenly covered by p.
13.82 Proposition Let U be a nonempty connected open subset of X which is

evenly covered by a continuous map p : Xe → X . Then the sheets over U are the
connected components of p−1 (U).
PROOF : Since the sheets over U are open sets forming a partition of p−1 (U), they
are also closed in p−1 (U). Furthermore, each sheet is nonempty and connected since
it is homeomorphic to U. The result therefore follows from Proposition 13.37.
13.83 Proposition Let p : Xe → X be a covering projection. If the base space X

is Hausdorff, then so is the covering space Xe .
PROOF : Let x̃ 1 and x̃ 2 be two different points in Xe . If x̃ 1 and x̃ 2 lie in different

fibres they can be separated by the neighbouhoods p−1 (U1 ) and p−1 (U2 ) , where
U1 and U2 are disjoint neighbourhoods of p (x̃ 1 ) and p (x̃ 2 ) .
If x̃ 1 and x̃ 2 lie in the same fibre over a point x ∈ X , and U is an evenly covered
neighbourhood of x, then x̃ 1 and x̃ 2 lie in different sheets over U which are disjoint
open neighbourhoods of x̃ 1 and x̃ 2 in Xe .
PRELIMINARIES 613
13.84 Definition A continuous map f : X → Y is called a local homeomor-

phism if each point x ∈ X has an open neighbourhood which is mapped homeomor-
phically by f onto an open neighbourhood of f (x).
13.85 Proposition A covering projection is a local homeomorphism.
PROOF : Let p : Xe → X be a covering projection and x̃ ∈ X,

e and choose an evenly
covered neighbourhoodU of p (x̃). Then the sheet U eα over U containing x̃ is an open
neighbourhood of x̃ which is mapped homeomorphically onto U by p, thus showing
that p is a local homeomorphism.
13.86 Proposition A local homeomorphism is an open map.

PROOF : If f : X → Y is mapping an open neighbourhood U of a point x ∈ X home-
omorphically onto an open neighbourhood f (U) of f (x), then this is so also for all
open neighbourhoods of x contained in U. The proposition now follows since the
collection of these neighbourhoods for all x ∈ X is a basis for the topology on X.
13.87 Definition Let Y and B be topological spaces, and let p : Y → B be a

continuous map. Given a continuous map f : X → B from a topological space X, a
continuous map g : X → Y is called a lifting of f with respect to p if p ◦ g = f . If it
is clear from the situation, we omit the reference to p.
13.88 The Unique Lifting Theorem Let p : Xe → X be a covering projection,

and let f : Y → X be a continuous map from a connected space Y . If f˜1 , f˜2 : Y → Xe
are two liftings of f such that f˜1 (y0 ) = f˜2 (y0 ) for a point y0 ∈ Y , then f˜1 = f˜2 .
PROOF : We will prove that the sets Y1 = { y| f˜1 (y) = f˜2 (y) } and Y2 = { y| f˜1 (y) 6=
f˜2 (y) } are both open in Y . Since Y is connected and Y = Y1 ∪Y2 with y0 ∈ Y1 , it then
follows that Y1 = Y so that f˜1 = f˜2 .
First let y ∈ Y1 , and choose an evenly covered neighbourhood U of f (y). Let
e
Uα be the sheet over U containing f˜1 (y) = f˜2 (y) , and let sα : U → U eα be the
inverse of the homeomorphism induced by p and iα : U eα → Xe be the inclusion map.
−1
Then V = f˜ (U eα ) ∩ f˜ (U
−1 eα ) is an open neighbourhood of y, and we have that
1 2
f˜1 |V = iα ◦ sα ◦ f |V = f˜2 |V so that V ⊂ Y1 , thus showing that Y1 is open.
Next let y ∈ Y2 , and again choose an evenly covered neighbourhoodU of f (y). Let
eα and U
U eβ be the disjoint sheets over U containing f˜1 (y) and f˜2 (y), respectively.
eα ) ∩ f˜−1 (U
Then V = f˜1−1 (U eβ ) is an open neighbourhood of y, and we have that
2
f˜1 (V ) ∩ f˜2 (V ) = 0/ so that V ⊂ Y2 . Hence we have shown that Y2 also is open.
13.89 The Homotopy Lifting Theorem Let p : Xe → X be a covering projection,

and let f˜ : Y → Xe be a lifting of a continuous map f : Y → X . Then a homotopy
F : Y × I → X from f can be lifted to a homotopy Fe : Y × I → Xe from f˜ . If F is a
homotopy relative to a subset Y ′ of Y , then Fe is also a homotopy relative to Y ′ .
PROOF : We first fix a point y ∈ Y . For each t ∈ I there are open neighbourhoods
Ut of y in Y and Vt of t in I such that F(Ut × Vt ) is contained in an evenly covered
neighbourhood of F(y,t) in S
X. Since I is compact,
T
there is a finite number of points
t0 , ... ,tn ∈ I such that I = ni=0 Vti , and Ny = ni=0 Uti is an open neighbourhood of y.
Let 0 = k0 < k1 < ... < kr = 1 be a partition of I such that each interval [ ki , ki+1 ]
is contained in at least one of the sets Vt0 , ... ,Vtn . By induction on i we will define
continuous maps Fei : Ny × [ 0, ki ] → Xe for i = 0, ... , r so that p ◦ Fei = F |N × [ 0,k ] and
y i
Fei (u, 0) = f˜(u) for every u ∈ Ny .

The map Fe0 is uniquely defined by the last equation. Supposing next that Fei is
already defined, we want to define Fei+1 . By the first part of the proof we know that
F(Ut × [ ki , ki+1 ]) is contained in an evenly covered neighbourhood U in X. For each
sheet U eα over U we let sα : U → U eα be the inverse of the homeomorphism induced
by p, and we let Qα = {u ∈ Ny |Fei (u, ki ) ∈ U
eα } . The sets Qα form a disjoint open
cover of Ny . We now define

Fei (u,t) for u ∈ Ny , t ∈ [0, ki ]
Fei+1 (u,t) =
sα ◦ F(u,t) for u ∈ Qα , t ∈ [ki , ki+1 ]
Since Fei (u, ki ) = sα ◦ F(u, ki ) for every u ∈ Qα , the map Fei+1 is well defined, and
it follows from Proposition 13.43 that it is continuous. Furthermore, we have that
p ◦ Fei+1 = F |Ny × [ 0,ki+1 ] and Fei+1 (u, 0) = f˜(u) for every u ∈ Ny , which completes
the induction step.
For every y ∈ Y we now have an open neighbourhood Ny of y and a continuous
map Fey : Ny × I → Xe so that p ◦ Fey = F |Ny ×I and Fey (u, 0) = f˜(u) for every u ∈ Ny . If
u ∈ Ny ∩Nz , then Fey |{u}×I and Fez |{u}×I are both liftings of F |{u}×I , and we have that
Fey (u, 0) = fe(u) = Fez (u, 0) . Since {u} × I is connected, it follows from the unique
lifting theorem that Fey |{u}×I = Fez |{u}×I . This shows that Fey and Fez coincide on
(Ny ∩Nz )×I , and we thus obtain a continuous map Fe : Y ×I → Xe so that Fe|Ny ×I = Fey
for every y ∈ Y . We have that p ◦ Fe = F and F(y,
e 0) = f˜(y) for every y ∈ Y , showing
that Fe is a homotopy from f˜ which is a lifting of F .
If F is a homotopy relative to Y ′ and y ∈ Y ′ , then Fe ({y} × I) is a connected
subset of the fibre over f (y) containing Fe (y, 0) = f˜(y) . Since the fibre is a discrete
PRELIMINARIES 615
topological space, it follows that Fe (y,t) = f˜(y) for all t ∈ I, which shows that Fe is
also a homotopy relative to Y ′ .
13.90 Remark Let p : Xe → X be a covering projection, and let f , g : Y → X

be two continuous maps which are homotopic. Then it follows from the homotopy
lifting theorem that f has a lifting if and only if the same is true for g. Hence the ex-
istence of liftings with respect to a covering projection is a problem in the homotopy
category.
13.91 Corollary Let p : (X,e x̃0 ) → (X, x0 ) be a covering projection. Then a path
σ in X from x0 can be lifted to a unique path σ̃ in Xe from x̃0 .
PROOF : Let Y = {y0 } be a topological space consisting of a single point y0 , and
let α : I → Y × I and β : Y × I → I be the canonical homeomorphisms defined by
α (t) = (y0 ,t) and β (y0 ,t) = t for t ∈ I. Furthermore, let f : Y → X and f˜ : Y → Xe
be the continuous maps given by f (y0 ) = x0 and f˜(y0 ) = x̃0 .
Then f˜ is a lifting of f , and F = σ ◦ β is a homotopy from f which can be
lifted to a homotopy Fe from f˜ by the homotopy lifting theorem. We now have that
σ̃ = Fe ◦ α is a path in Xe from x̃0 which is a lifting of σ , and this completes the
existence part of the corollary. The uniqueness of σ̃ follows from the unique lifting
theorem.
13.92 Proposition Let p : (X, e x̃0 ) → (X, x0 ) be a covering projection, and let σ e
and τe be paths in Xe from x̃0 . Then we have that σ e ≃ τe if and only if p ◦ σe ≃ p ◦ τe .
PROOF : e ≃ τe , it follows from Proposition 13.46 that p ◦ σe ≃ p ◦ τe .

If σ
Conversely, suppose that p ◦ σe ≃ p ◦ τe . By the homotopy lifting theorem we
e ≃ p ◦ τe can be lifted to a homotopy Fe from σ
know that a homotopy F : p ◦ σ e to a
e . Since ω
path ω e and τe are both paths from x̃0 which are liftings of p ◦ τe , it follows
e = τe , showing that σe ≃ τe .
from Corollary 13.91 that ω
13.93 Corollary e x̃0 ) → (X, x0 ) is a covering projection, then the induced

If p : (X,
e x̃0 ) → π1 (X, x0 ) is injective.
homomorphism p∗ : π1 (X,
13.94 Let p : Xe → X be a covering projection, and let σ be a path in X from x0

to x1 . For each point x̃0 ∈ p−1 (x0 ) , there is by Corollary 13.91 a unique lifting of σ
to a path σex̃0 in Xe from x̃0 , and it follows from Proposition 13.92 that σex̃0 (1) only
depends on the path class of σ . Hence we have a map
f [σ ] : p−1 (x0 ) → p−1 (x1 )
defined by f [σ ] (x̃0 ) = σex̃0 (1) , which is called the translation over σ .

We thus obtain a contravariant functor F : P (X) → S from the path category
P (X) to the category S of sets, which assigns to each point x0 the fibre p−1 (x0 )
and to to each morphism [σ ] : x1 → x0 the translation f [σ ] : p−1 (x0 ) → p−1 (x1 ) .
From Proposition 13.69 it follows that every f [σ ] is a bijection. So if X is pathwise
connected, every fibre has the same number of points, called the multiplicity of the
covering p. A covering is also said to be n - fold if it has multiplicity n.
13.95 Proposition Let p : (X, e x̃0 ) → (X, x0 ) be a covering projection, and let σ
e
and τe be paths in Xe from x̃0 which are liftings of the paths σ and τ in X from x0 ,
e (1) = τe(1) if and only if σ (1) = τ (1) and
respectively. Then we have that σ
e x̃0 ).
[ σ ∗ τ −1 ] ∈ p∗ π1 (X, (1)
PROOF : If σe (1) = τe(1) , then σ (1) = τ (1) , and σ e ∗ τe−1 is a well-defined closed
e x̃0 ) .
path at x̃0 so that [ σ ∗ τ −1 ] = p∗ [ σe ∗ τe−1 ] ∈ p∗ π1 (X,
Conversely, suppose that σ (1) = τ (1) and that (1) is satisfied. Then there exists
a path class [ ω e x̃0 ) with
e ] ∈ π1 (X,
[ σ ] ∗ [ τ ] −1 = [ p ◦ ω
e]
from which it follows that
e ] ∗ [τ ] = [ p ◦ ω
[ p ◦ σe ] = [ σ ] = [ p ◦ ω e ] ∗ [ p ◦ τe] = [ p ◦ (ω
e ∗ τe)] .
e ∗ τe are paths in Xe from x̃0 with p ◦ σe ≃ p ◦ (ω

This shows that σe and ω e ∗ τe) , so
that σe ≃ ω
e ∗ τe by Proposition 13.92. In particular, we have that σe (1) = τe(1) , which
completes the proof of the proposition.
13.96 Definition Two coverings p 1 : (Xe 1 , x̃ 1 ) → (X, x) and p 2 : (Xe 2 , x̃ 2 ) →

(X, x) of the pointed topological space (X, x) are said to be equivalent if there is a
homeomorphism φ : (Xe 1 , x̃ 1 ) → (Xe 2 , x̃ 2 ) with p 2 ◦ φ = p 1 .
13.97 Remark Let X be a topological space which is pathwise connected and

e x̃0 ) → (X, x0 ) be a covering with a path-
locally pathwise connected, and let p : (X,
e If Σ is the set of all paths in X from x0 and σ0 is
wise connected covering space X.
PRELIMINARIES 617
e x̃0 ) defined
the constant path at x0 , then we have a surjective map φ : (Σ, σ0 ) → (X,
e (1) , where σe is the unique path in Xe from x̃0 which is a lifting of σ .
by φ (σ ) = σ
By Proposition 13.95 it follows that φ (σ ) = φ (τ ) if and only if σ (1) = τ (1)
e x̃0 ) , in which case we write σ ∼ τ . Then ∼ is clearly
and [ σ ∗ τ −1 ] ∈ p∗ π1 (X,
an equivalence relation in Σ , and if π : Σ → Σ / ∼ is the canonical projection, we
have an induced bijection φ : Σ / ∼ → Xe so that φ ◦ π = φ . If < σ > denotes the
equivalence class of σ ∈ Σ and p ′ = p ◦ φ , we have that p ′ (< σ >) = σ (1) .
There is a unique topology τ on Σ / ∼ such that φ is a homeomorphism. If
σ ∈ Σ , there is a basis for neighbourhoods of σe (1) in Xe consisting of open path-
wise connected neighbourhoods U e which are mapped homeomorphically by p onto
open pathwise connected neighbourhoods U of σ (1) in X , since p is a local home-
−1 e
omorphism. Then φ (U) consists of all equivalence classes in Σ / ∼ having a
representative of the form σ ∗ σ 1 , where σ 1 is a path in U from σ (1) . The sets
−1
φ (U) e , also denoted by < σ ,U >, form a basis for neighbourhoods of < σ > in
the topology τ . Using this topology, we have that p ′ : ( Σ / ∼, < σ0 > ) → (X, x0 ) is
e x̃0 ) → (X, x0 ) .
a covering of (X, x0 ) which is equivalent to p : (X,
e x̃0 ) and the projection p can be
This shows that the pointed topological space (X,
described up to equivalence by means of (X, x0 ) and the subgroup H = p∗ π1 (X, e x̃0 )
e x̃0 ) → (X, x0 ) belongs to H.
of π1 (X, x0 ) . We also say that the covering p : (X,
13.98 e x̃0 ) → (X, x0 ) be a covering projection, and consider the opera-

Let p : (X,
tion of the fundamental group π1 (X, x0 ) on the fibre p−1 (x0 ) given by
[σ ] x̃ = f [σ ]−1 (x̃)
for [σ ] ∈ π1 (X, x0 ) and x̃ ∈ p−1 (x0 ) , where f [σ ]−1 is translation over the path σ −1
as defined in 13.94.
The isotropy group of x̃0 is p∗ π1 (X, e x̃0 ) , and if the covering space Xe is pathwise
connected, the orbit of x̃0 is the entire fibre p−1 (x0 ) . Then it follows from Proposi-
tion 13.18 that the map φ : π1 (X, x0 ) → p−1 (x0 ) given by φ ([σ ]) = [σ ] x̃0 induces a
e x̃0 ) → p−1 (x0 ) .
bijection φ : π1 (X, x0 ) / p∗ π1 (X,
13.99 Proposition A covering projection p : Xe → X with a pathwise connected

covering space Xe is a homeomorphism if and only if p∗ π1 (X,
e x̃0 ) = π1 (X, p(x̃0 )) for
some point x̃0 ∈ Xe .
PROOF : If p satisfies the last assertion of the proposition, then it follows from 13.94
and 13.98 that p is injective. Since every covering projection is surjective, continuous
and open by Remark 13.80, Definition 13.79 and Propositions 13.85 and 13.86, it
follows that p is a homeomorphism.
Assuming conversely that p is a homeomorphism, then the last assertion of the
proposition follows from 13.70 for any point x̃0 ∈ Xe .
13.100 The Lifting Theorem Let p : (X, e x̃0 ) → (X, x0 ) be a covering projection,
and let Y be a topological space which is pathwise connected and locally pathwise
connected. Then a continuous map f : (Y, y0 ) → (X, x0 ) can be lifted to a map f˜ :
e x̃0 ) if and only if f∗ π1 (Y, y0 ) ⊂ p∗ π1 (X,
(Y, y0 ) → (X, e x̃0 ) .
PROOF : If there is such a base point preserving lifting f˜ with p ◦ f˜ = f , then we

obtain p∗ ◦ f˜∗ = f∗ by using the canonical functor into the homotopy category of
pointed topological spaces followed by the fundamental group functor, and this im-
e x̃0 ) .
plies that f∗ π1 (Y, y0 ) ⊂ p∗ π1 (X,
Conversely, supposing that the last inclusion is satisfied, we must show the exis-
tence of a lifting f˜ of f preserving base points. Let y ∈ Y , and choose a path σ in Y
from y0 to y, using the fact that Y is pathwise connected. Then f ◦ σ is a path in X
from x0 which by Corollary 13.91 can be lifted to a unique path σ e in Xe from x̃0 , and
we define f˜(y) = σ
e (1) .
For f˜ to be well defined, we must show that f˜(y) does not depend on the choice
of the path σ . So let τ be another path in Y from y0 to y, and let τe be the unique path
in Xe from x̃0 which is a lifting of f ◦ τ . Then f ◦ σ (1) = f (y) = f ◦ τ (1) and
e x̃0 ) ,
[( f ◦ σ ) ∗ ( f ◦ τ ) −1 ] = f∗ ( [ σ ∗ τ −1 ] ) ∈ f∗ π1 (Y, y0 ) ⊂ p∗ π1 (X,
so that σe (1) = τe(1) by Proposition 13.95.

It follows immediately from the definition that p ◦ f˜ = f and f˜(y0 ) = x̃0 , so it
e be an arbitrary neighbourhood of
only remains to show that f˜ is continuous. Let U
e Since p is a local homeomorphism, U
f˜(y) in X. e contains an open neighbourhood
e of f˜(y) which is mapped homeomorphically by p onto an open neighborhood V
V
e be the inverse homeomorphism and i : Ve → Xe be the
of f (y). We let s : V → V
inclusion map. Since Y is locally pathwise connected, there is a pathwise connected
neighbourhood W of y with f (W ) ⊂ V , and we will show that f˜(W ) ⊂ Ve .
Let y′ ∈ W , and choose a path ρ in W from y to y′ . Then ρe = i ◦ s ◦ f ◦ ρ is the
unique path in Xe from f˜(y) which is a lifting of f ◦ ρ . Now σ ∗ ρ is a path in Y from
y0 to y′ , and σe ∗ ρe is the unique path in Xe from x̃0 which is a lifting of f ◦ (σ ∗ ρ ).
PRELIMINARIES 619
Hence we have that f˜ (y′ ) = σe ∗ ρe (1) = ρe (1) ∈ Ve , which shows that f˜(W ) ⊂ Ve ⊂ U
e
and completes the proof that f˜ is continuous.
13.101 Corollary Let X be a topological space which is pathwise connected

and locally pathwise connected. Then two coverings p 1 : (Xe 1 , x̃ 1 ) → (X, x) and
p 2 : (Xe 2 , x̃ 2 ) → (X, x) with pathwise connected covering spaces Xe 1 and Xe 2 are
equivalent if p 1 ∗ π1 ( Xe 1 , x̃ 1 ) = p 2 ∗ π1 ( Xe 2 , x̃ 2 ) .
13.102 Definition A topological space X is said to be semi-locally simply con-

nected if every point x0 ∈ X has a neighbourhood V such that i∗ : π1 (V, x0 ) →
π1 (X, x0 ) is trivial, where i : (V, x0 ) → (X, x0 ) is the inclusion map. This means
that every closed path σ in V at x0 is null homotopic in X. A neighbourhood V of x0
having this property and which in addition is open and pathwise connected, is called
a pathwise simple neighbourhood of x0 .
13.103 Proposition If X is a topological space which is locally pathwise con-

nected and semi-locally simply connected, then every point in X has a basis of path-
wise simple neighbourhoods.
PROOF : Let U ⊂ V be two neighbourhoods of x0 , and suppose that every closed
path in V at x0 is null homotopic in X. Then this holds in particular for every closed
path in U at x0 . The proposition hence follows from Proposition 13.56.
13.104 Theorem Let X be a topological space which is pathwise connected,

locally pathwise connected and semi-locally simply connected. If x 0 ∈ X and H is a
e x̃ 0 ) → (X, x 0 ) with a pathwise
subgroup of π1 (X, x 0 ) , then there is a covering p : (X,
e x̃ 0 ) = H .
connected covering space Xe such that p∗ π1 (X,
PROOF : Let Σ be the set of paths in X from x 0 , and let σ 0 be the constant path
at x 0 . We say that two paths σ and τ in Σ are equivalent, and write σ ∼ τ , if
σ (1) = τ (1) and [ σ ∗ τ −1 ] ∈ H . Then ∼ is an equivalence relation in Σ , and we
denote the equivalence class of σ by < σ > . Let Xe = Σ / ∼ be the quotient set and
xe0 = < σ 0 > be the equivalence class of the constant path at x 0 . We define the map
e x̃ 0 ) → (X, x 0 ) by p (< σ >) = σ (1) .
p : (X,
If σ ∈ Σ and U is an open neighbourhood of σ (1) , we let < σ ,U > be the set of
all equivalence classes in Xe having a representative of the form σ ∗ σ 1 , where σ 1
is a path in U from σ (1) . We will show that the sets < σ ,U > form a basis for a
topology τ on Xe .
If < τ > ∈ < σ ,U >, then τ = σ ∗ σ 1 for a path σ 1 in U from σ (1) . For every
path τ 1 in U from τ (1) we then have that τ ∗ τ 1 = σ ∗ (σ 1 ∗ τ 1 ) , showing that
< τ ,U > ⊂ < σ ,U > . Since σ = τ ∗ σ 1 −1 , it also follows that < σ > ∈ < τ ,U > so
that < σ ,U > ⊂ < τ ,U > , thus showing that < τ ,U > = < σ ,U > .
Therefore, if < ω > ∈ < σ ,U > ∩ < τ ,V >, we have that < ω ,U ∩V > ⊂ < σ ,U >
∩ < τ ,V > , which shows that the sets < σ ,U > form a basis for a topology on Xe .
Giving Xe this topology τ , the map p is continuous since < σ ,U > is an open
neighbourhood of < σ > with p (< σ ,U >) ⊂ U for each open neighbourhood U
of p (< σ >) . The map p is also open since p (< σ ,U >) is the path component
of U containing p (< σ >), which is open since X is locally pathwise connected.
In particular, we have that p (< σ ,U >) = U for each open pathwise connected
neighbourhood U of p (< σ >) .
We next show that p maps < σ ,U > homeomorphically onto U when U is a
pathwise simple neighbourhood of p (< σ >) . We already know that p is continuous
and open and that p maps < σ ,U > onto U, so it only remains to prove that the
restriction of p to < σ ,U > is injective. Let σ 1 and σ 2 be two paths in U from
σ (1) such that p (< σ ∗ σ 1 >) = p (< σ ∗ σ 2 >) . Then σ 1 (1) = σ 2 (1) so that
σ 1 ∗ σ 2 −1 is a closed path in U at σ (1) which must be null homotopic since the
neighbourhood U is semi-locally simply connected. Hence it follows that
[(σ ∗ σ 1 ) ∗ (σ ∗ σ 2) −1 ] = [ σ ] ∗ [ σ 1 ∗ σ 2 −1 ] ∗ [ σ −1 ] = [ σ 0 ] ∈ H
which implies that < σ ∗ σ 1 > = < σ ∗ σ 2 > .

Using this, we can now show that every pathwise simple neighbourhood U of a
point x ∈ X is evenly covered by p. We know that p−1 (U) is the union of the open sets
< σ ,U > where < σ > ∈ p−1 (U), and these sets are either equal or disjoint. Indeed,
if < σ ,U > and < τ ,U > are not disjoint and contain a common element < ω >, we
have that < ω ,U > = < σ ,U > and < ω ,U > = < τ ,U > showing that they are equal.
Hence U is evenly covered by p, and it follows from Proposition 13.103 that p is a
covering projection.
e x̃ 0 ) = H and that Xe is pathwise con-
Finally, we need to verify that p∗ π1 (X,
e : I → Xe be the map defined by σ
nected. If σ ∈ Σ , we let σ e (t) = < σ t > , where
σ t ∈ Σ is the path given by σ t (s) = σ (s t) for s ∈ I. We want to show that σe is the
unique path in Xe from x̃ 0 which is a lifting of σ , and that σ
e (1) = < σ > . The last
assertion follows immediately from the definition of σe . Since σe (0) = < σ 0 > = x̃ 0
and p ◦ σe (t) = σ t (1) = σ (t) for t ∈ I, it only remains to show that σe is continuous,
as the uniqueness follows from Corollary 13.91.
e (t 0 ) ∈ < τ ,U > , then < σ t 0 ,U > = < τ ,U > and σ (t 0 ) = p ◦ σ
If σ e (t 0 ) ∈ U .
Choosing a convex open subset J of I containing t 0 with σ (J) ⊂ U , we contend
e (J) ⊂ < τ ,U > . If fact, if t ∈ J and ω is the path in U defined by ω (s) =
that σ
σ ((1 − s) t 0 + s t) for s ∈ I, we have a homotopy H : σ t ≃ σ t 0 ∗ ω given by

σ (s t + u s (2 t0 − t)) for s ∈ [0, 1/2]
H(s, u) =
σ (s t + u (1 − s)(2 t0 − t)) for s ∈ [1/2, 1]
PRELIMINARIES 621
so that
σe (t) = < σ t > = < σ t 0 ∗ ω > ∈ < σ t 0 ,U > = < τ ,U > ,
thus completing the proof that σ e is continuous.

From this it now follows that p∗ π1 (X, e x̃ 0 ) = H , since [ σ ] ∈ H ⇔
σ ∼ σ 0 ⇔ < σ > = x̃ 0 ⇔ σe is a closed path in Xe at x̃ 0 ⇔ [ σe ] ∈ π1 (X, e x̃ 0 )
e x̃ 0 ) , where we have used Corollary 13.93 to obtain
⇔ [ σ ] = p∗ ( [ σe ] ) ∈ p∗ π1 (X,
the last equivalence. It also follows that Xe is pathwise connected, since there is a
path σe in Xe connecting x̃ 0 to < σ > for every element < σ > in Xe .
TOPOLOGICAL GROUPS
13.105 Definition A topological group G is a group which is at the same time
a topological space such that the maps µ : G × G → G and ν : G → G given by the
group operations µ (g, h) = gh and ν (g) = g−1 are continuous. The unit element of
G is denoted by e.
13.106 Remark Continuity of µ and ν is equivalent to assuming that the map

ρ : G × G → G given by ρ (g, h) = gh−1 is continuous.
13.107 Definition If G is a topological group, we let Ω (G, e) denote the set of

closed paths in G at e. We define the product of two paths ω , θ ∈ Ω (G, e) to be the
path ω θ given by ω θ (t) = ω (t) θ (t) for t ∈ I. This defines a group structure on
Ω (G, e). The unit element is the constant path <e> at e.
13.108 Proposition Let G be a topological group, and let ω 0 , ω 1 , θ 0 , θ 1 ∈

Ω (G, e). If ω 0 ≃ ω 1 and θ 0 ≃ θ 1 , then ω 0 θ 0 ≃ ω 1 θ 1 .
PROOF : If F : ω 0 ≃ ω 1 and G : θ 0 ≃ θ 1 are homotopies, we have a homotopy
H : ω 0 θ 0 ≃ ω 1 θ 1 given by H(s,t) = F(s,t) G(s,t) for s,t ∈ I.
13.109 Proposition If G is a topological group, the canonical projection φ :

Ω (G, e) → π1 (G, e) given by φ (ω ) = [ ω ] is a group homomorphism.
PROOF : If ω , θ ∈ Ω (G, e), we have that ω ≃ ω ∗ <e> and θ ≃ <e> ∗ θ . Hence it
follows from Proposition 13.108 that
ω θ ≃ ( ω ∗ <e>) (<e> ∗ θ ) = ω ∗ θ
which shows that [ ω θ ] = [ ω ] ∗ [ θ ] .

13.110 Proposition If G is a topological group, then π1 (G, e) is abelian.

PROOF : If ω , θ ∈ Ω (G, e), we have that ω ∗ <e> ≃ ω ≃ <e> ∗ ω and <e> ∗ θ ≃
θ ≃ θ ∗ <e> . Hence it follows from Proposition 13.108 that
ω ∗ θ = ( ω ∗ <e>) (<e> ∗ θ ) ≃ (<e> ∗ ω ) ( θ ∗ <e>) = θ ∗ ω .
13.111 Proposition Let φ : G → H be a continuous homomorphism between the

topological groups G and H. Then φ is a covering projection if and only if it is an
open surjection with a discrete kernel.
PROOF : If φ is a covering projection, then it follows from Remark 13.80 and Propo-
sitions 13.85 and 13.86 that φ is an open surjection with a discrete kernel.
Conversely, suppose that φ satisfies these properties, and let K = ker φ and W be
an open neighbourhood of e in G with W ∩ K = {e}. Since the map ρ : G × G → G
given by ρ (g, h) = g −1 h is continuous, there is an open neighbourhood V of e in G
such that V −1 V ⊂ W . Then U = φ (V ) is an open neighbourhood of e in H which we
claim is evenly covered by φ .
We first show that [
φ −1 (U) = Vk ,
k∈K
and that φ maps V k onto U for each k ∈ K. The last assertion as well as the inclusion
⊃ follows from the fact that
φ (V k) = {φ (vk)|v ∈ V } = {φ (v)|v ∈ V } = φ (V ) = U
for every k ∈ K . To prove the inclusion ⊂ , let g ∈ φ −1 (U), and choose a v ∈ V such
that φ (v) = φ (g) . Then k = v−1 g ∈ K and g = vk ∈ V k .
We next show that φ |V k is injective for each k ∈ K . Suppose that φ (v1 k) =
φ (v2 k) where v1 , v2 ∈ V . Then φ (v1 ) = φ (v2 ) so that
v−1
2 v1 ∈ V
−1
V ∩ K ⊂ W ∩ K = {e} ,
which implies that v1 = v2 .
Finally, we show that the sets V k are disjoint for k ∈ K. If g ∈ V k1 ∩V k2 where
k1 , k2 ∈ K , then g = v1 k1 = v2 k2 for v1 , v2 ∈ V so that
v−1 −1
2 v1 = k2 k1 ∈ V
−1
V ∩ K ⊂ W ∩ K = {e} ,
which implies that k1 = k2 . This shows that V k1 ∩V k2 = 0/ when k1 6= k2 and com-
pletes the proof that U is evenly covered by φ .
Now let h ∈ H . As φ is a surjection, there is a g ∈ G with φ (g) = h . Since the
left translations Lg and Lh in G and H, respectively, are homeomorphisms such that
Lh ◦ φ ◦ Lg−1 = φ , it follows that h U is an open neighbourhood of h which is evenly
covered by φ . Indeed, the inverse image φ −1 (h U) is the disjoint union of the open
sets g V k = Lg (V k) which are mapped homeomorphically onto h U = Lh (U) by φ
for k ∈ K. This completes the proof that φ is a covering projection.
PRELIMINARIES 623
13.112 Proposition Let U be an open neighbourhood of the unit element e in a

connected topological group G. Then U generates G, i.e., we have that
∞
[
G= Un ,
n=1
where U n denotes the set of all n-fold products of elements in U.

PROOF : Let V = U ∩ U −1 , where U −1 = {g−1 |g ∈ U}. Then V is an open neigh-
bourhood of e contained in U with V = V −1 , and the set
∞
[ ∞
[
H= Vn ⊂ Un
n=1 n=1
is a subgroup of G which is open, since hV ⊂ H for every h ∈ H. From this it follows

that each left coset of H in G is open. As G is connected, this implies that H = G,
TOPOLOGICAL VECTOR SPACES
13.113 Definition A topological vector space V is a vector space which is at the

same time a topological space such that the maps µ : V ×V → V and ρ : R ×V → V
given by the vector space operations µ (u, v) = u + v and ρ (k, v) = kv are continuous.
13.114 Definition A subset C of a vector space V is said to be convex if x, y ∈ C

implies that x + t (y − x) ∈ C for 0 ≤ t ≤ 1. A set B ⊂ V is said to be balanced if
x ∈ B and |t| ≤ 1 implies that t x ∈ B .
13.115 Proposition A topological vector space V has a basis of balanced open

neighbourhoods of 0.
PROOF : Let U be a neighbourhood of 0 in V . By the continuity of scalar multipli-
cation there is a real number r > 0 and an open neighbourhood W of 0 such that
(−r, r)W ⊂ U. We have that (−r, r)W is a balanced open neighbourhood of 0 since
[
(−r, r)W = tW .
0 < |t| < r
13.116 Definition If M is a Banach space, we let Br (x0 ) and Br (x0 ) denote,

respectively, the open and closed ball of radius r centered at x0 , i.e.,
Br (x0 ) = {x ∈ M| k x − x0 k < r} and Br (x0 ) = {x ∈ M| k x − x0 k ≤ r} .
We let Sr (x0 ) denote the sphere of radius r centered at x0 , i.e.,
Sr (x0 ) = {x ∈ M| k x − x0 k = r} .
13.117 Proposition A finite dimensional vector space V has a unique Hausdorff

topology compatible with the vector space operations. With this topology, every lin-
ear isomorphism x : V → Rn , where n = dim (V ), is a homeomorphism.
PROOF : If x : V → Rn is a linear isomorphism, then there is a unique topology T
on V so that x is a homeomorphism, consisting of the subsets O of V such that x(O)
is open in Rn . This is clearly a Hausdorff topology on V compatible with the vector
space operations. Is does not depend on the choice of x, for if y : V → Rn is another
linear isomorphism, then y = (y ◦ x−1 ) ◦ x is the composition of the homeomorphisms
x and y ◦ x−1 . This completes the proof of the existence part of the proposition.
To show the uniqueness part, suppose that V is given a Hausdorff topology T ′
compatible with the vector space operations, and let x : V → Rn be any linear isomor-
phism. Let E = {e1 , ..., en } be the standard basis for Rn and B = {v1 , ..., vn } be a
basis for V so that x(vi ) = ei for i = 1, ..., n. Then we have that
n
x−1 (a) = ∑ a j vj
j=1
for a ∈ Rn , showing that x−1 : Rn → V is continuous.

We can also prove that x is continuous. Let ε > 0. As the sphere Sε (0) in Rn is
compact, it follows that x−1 (Sε (0)) is compact and therefore closed in V since V is
Hausdorff. By Proposition 13.115 there is a balanced open neighbourhood U of 0 in
V with U ∩ x−1 (Sε (0)) = 0.
/ Then x(U) is a balanced open neighbourhood of 0 in Rn
such that x(U) ∩ Sε (0) = 0,
/ which shows that x(U) ⊂ Bε (0). From this it now follows
that x is continuous, since a − b ∈ U implies that x(a) − x(b) = x(a − b) ∈ Bε (0),
thus showing that T ′ = T and completing the proof of the uniqueness part of the
proposition.
13.118 Definition A linear map F : X → Y between the normed spaces X and Y

is said to be bounded if there is a real number M ≥ 0 such that
kF(v)k ≤ M kvk (1)
for every v ∈ X.
13.119 Proposition Let F : X → Y be a linear map between the normed spaces

X and Y . Then the following assertions are equivalent
PRELIMINARIES 625
(1) F is continuous at 0.
(2) F is uniformly continuous.
(3) F is bounded.
PROOF : The equivalence of (1) and (2) follows since kF(v) − F(w)k = kF(v − w)k
by the linearity of F.
Suppose that F is bounded, and choose a real number M > 0 so that formula (1)
in Definition 13.118 is satisfied for every v ∈ X. Then F is continuous at 0 since
kF(v)k < ε whenever kvk < ε /M.
Suppose conversely that F is continuous at 0, and choose a δ > 0 so that
kF(v)k < 1 when kvk < δ . Then formula (1) in Definition 13.118 is satisfied with
any M > 1/δ , since kF(v/Mkvk)k < 1 for all v 6= 0. The formula is clearly also
satisfied for v = 0, thus showing that F is bounded.
13.120 Proposition Let X and Y be normed spaces. Then the set L(X,Y ) of
bounded linear maps from X to Y is a normed space with the norm defined by
kFk = sup {kF(v)k | kvk ≤ 1} (1)
for F ∈ L(X,Y ). We have that
kF(v)k ≤ kFk kvk (2)
for every F ∈ L(X,Y ) and v ∈ X, so that kFk is the least real number M ≥ 0 satisfying
inequality (1) in Definition 13.118 for every v ∈ X. If Y is a Banach space, so is
L(X,Y ).
PROOF : If F, G ∈ L(X,Y ), then F + G ∈ L(X,Y ) and
kF + Gk ≤ kFk + kGk
by the triangle inequality in Y which implies that
k(F + G)(v)k = kF(v) + G(v)k ≤ kF(v)k + kG(v)k ≤ kFk + kGk
for every v ∈ X with kvk ≤ 1. Furthermore, if F ∈ L(X,Y ) and a is a scalar, we have

that aF ∈ L(X,Y ) and
kaFk = |a|kFk
since
k(aF)(v)k = kaF(v)k = |a|kF(v)k
for every v ∈ X with kvk ≤ 1. Finally, if F 6= 0, then F(v) 6= 0 for some v ∈ X with
kvk ≤ 1, thus showing that kFk > 0. This completes the proof that L(X,Y ) is a vector
space, and that (1) defines a norm in L(X,Y ).
Inequality (2) is clearly true when v = 0, and it follows from

1
kF(v) k = F ( v ) ≤ kFk
kvk kvk
when v 6= 0.
Now suppose that Y is a Banach space, and let {Fi }∞ i=1 be a Cauchy sequence in
L(X,Y ). Since
kFi (v) − Fj (v)k ≤ kFi − Fj k kvk
for every positive integer i and j, it follows that {Fi (v)}∞
i=1 is a Cauchy sequence in
Y which must converge to a vector F(v) ∈ Y for every v ∈ X. As each Fi is linear, the
same is true for F.
For each ε > 0 there is an N so that kFi − Fj k < ε for i, j ≥ N. Hence we have
that
kFi (v) − Fj (v)k ≤ ε kvk
for v ∈ X and i, j ≥ N, and letting j → ∞ we obtain
kFi (v) − F(v)k ≤ ε kvk
for v ∈ X and i ≥ N. This shows that Fi − F ∈ L(X,Y ) so that F ∈ L(X,Y ), and

kFi − Fk ≤ ε for i ≥ N which shows that {Fi }∞
i=1 converges to F in the norm of
L(X,Y ).
13.121 Proposition Let X,Y and Z be normed spaces. Then the composition
of the bounded linear maps F ∈ L(X,Y ) and G ∈ L(Y, Z) is a bounded linear map
G ◦ F ∈ L(X, Z) with
kG ◦ Fk ≤ kGkkFk .
PROOF : Using inequality (2) in Proposition 13.120 we have that
kG ◦ F(v)k ≤ kGkkF(v)k ≤ kGkkFkkvk
for every v ∈ X.
13.122 Corollary Let X be a Banach space. Then L(X, X) is a Banach algebra.

13.123 Proposition Let ψ : X → Y be an isometric isomorphism between the

normed spaces X and Y . Then the map φ : L(X, X) → L(Y,Y ) given by φ (F) =
ψ ◦ F ◦ ψ −1 for F ∈ L(X, X), is an isometric isomorphism.
kφ (F)k = sup {k ψ ◦ F ◦ ψ −1 (w)k | w ∈ Y and kwk ≤ 1}

= sup {kF ◦ ψ −1 (w)k | w ∈ Y and k ψ −1 (w)k ≤ 1}
= sup {kF(v)k | v ∈ X and kvk ≤ 1} = kFk
for every F ∈ L(X, X).

PRELIMINARIES 627
13.124 Lemma Let A be an element in a Banach algebra X with kAk < 1. Then
I − A is invertible.
PROOF : The elements Sn = ∑nk=0 Ak form a Cauchy sequence in X since
m
m
k
kSm − Snk = ∑ A ≤ ∑ kAkk
k=n+1 k=n+1
when n < m , and ∑∞ k

k=0 kAk is a convergent geometric series. As X is complete, the
sequence {Sn }∞
n=0 must therefore converge to an element B ∈ X, and we have that
Sn (I − A) = (I − A) Sn = I − An+1
for each n. Letting n → ∞ and using that kAn+1 k ≤ kAkn+1 , we thus obtain
B (I − A) = (I − A) B = I ,
showing that I − A is invertible with inverse B.
13.125 Proposition The set U of invertible elements in a Banach algebra X is

open.
PROOF : If A ∈ U, then the map LA : X → X given by LA (B) = AB for B ∈ X, is a
homeomorphism with inverse LA−1 , mapping U onto U. Hence the open ball B1 (I) =
{C ∈ X |kC − Ik < 1} is mapped by LA onto an open neighbourhood of A which is
contained in U by Lemma 13.124.
13.126 Proposition Let X be a Banach algebra, and let U be the set of invertible
elements in X. Then we have a continuous map exp : X → U given by
∞ k
exp (A) = ∑ Ak! (1)
k=0
for A ∈ X. If A, B ∈ X with AB = BA , then
exp (A + B) = exp (A) exp (B), (2)
and we have that exp (0) = I and exp (−A) = exp (A) −1 for A ∈ X . If A ∈ X and
B ∈ U, then
B exp (A) B−1 = exp (BAB−1 ). (3)
PROOF : We see that the series in (1) converges since
m
k m k
∑ A ≤ ∑ kAk

k=n+1 k! k=n+1 k!
when n < m , using the fact that

∞
kAkk
ekAk = ∑
k=0 k!
is a convergent series for all A ∈ X. The map exp is also continuous, since
∞ ∞
n k k k
exp (A) − ∑ A = ∑ A ≤ ∑ a
k! k! k!
k=0 k=n+1 k=n+1
for every n when kAk ≤ a, so that the series in (1) is uniformly convergent on every
bounded subset of X.
If A, B ∈ X with AB = BA , we have the usual Binomial formula
k
(A + B)k = ∑ ( kr ) Ak−r B r
r=0
which implies that

k
(A + B)k Ak−r Br
= ∑
k! r=0 (k − r)! r!
so that
! !
2n n n
(A + B)k Ak Bk
∑ k!
− ∑ ∑
k=0 k=0 k! k=0 k!
! ! ! !
2n 2n 2n 2n
kAkk kBkk kAkk kBkk
≤ ∑ k!
∑ k!
+ ∑ k!
∑ k!
k=n+1 k=0 k=0 k=n+1
for every n. As the last expression converges to 0 when n → ∞ , this completes the
proof of formula (2).
Finally, formula (3) follows from the relation
!
n n
Ak (BAB−1 ) k
B ∑ B−1 = ∑
k=0 k! k=0 k!
when n → ∞ .
Bibliography
[1] Abraham, R. and Marsden, J. Foundations of Mechanics, Second edition.

Benjamin-Cummings, New York. (1978)
[2] Apostol, T.M. Mathematical Analysis, Second edition. Addison Wesley, Read-
ing, MA. (1974)
[3] Bishop, R.L. and Crittenden, R.J. Geometry of Manifolds. Chelsea Publishing,
Providence, RI. (2001)
[4] Chevalley, C. Theory of Lie Groups. Princeton University Press, Princeton, NJ.
(1946)
[5] Curtis, W.D. and Miller, F.R. Differential Manifolds and Theoretical Physics.
Academic Press, Orlando, FL. (1985)
[6] Frankel, T. Gravitational Curvature. Freeman, San Francisco, CA. (1979)
[7] Goldstein, H. Classical Mechanics, Second edition. Addison Wesley, Reading,

MA. (1980)
[8] Grøn, Ø. and Hervik, S. Einstein’s General Theory of Relativity. Springer, New
York. (2007)
[9] Kobayashi, S. and Nomizu, K. Foundations of Differential Geometry, Volume

I. Interscience, New York. (1963)
[10] Kobayashi, S. and Nomizu, K. Foundations of Differential Geometry, Volume

II. Interscience, New York. (1969)
[11] Lang, S. Algebra. Addison Wesley, Reading, MA. (1965)
[12] Lang, S. Differential Manifolds. Springer-Verlag, New York. (1985)
[13] Lang, S. Linear Algebra. Addison Wesley, Reading, MA. (1966)
629
630 Bibliography
[14] Lang, S. Real Analysis. Addison Wesley, Reading, MA. (1969)
[15] Mangiarotti, L. and Sardanashvily, G. Connections in Classical and Quantum

Field Theory. World Scientific, Singapore. (2000)
[16] Matsushima, Y. Smooth Manifolds. Marcel Dekker, New York. (1972)
[17] Miller, W. Symmetry Groups and Their Applications. Academic Press, New
York. (1972)
[18] Misner, C.W., Thorn, K.S. and Wheeler, J.A. Gravitation. Freeman, San Fran-
cisco, CA. (1973)
[19] O’Neill, B. Semi-Riemannian Geometry. Academic Press, San Diego, CA.

(1983)
[20] Rudin, W. Principles of Mathematical Analysis, Second edition. McGraw-Hill,

New York. (1964)
[21] Sauders, D.J. The Geometry of Jet Bundles. Cambridge University Press, Cam-
bridge. (1989)
[22] Schubert, H. Topology. MacDonald & Co, London. (1968)
[23] Spanier, E.H. Algebraic Topology. McGraw-Hill, New York. (1966)
[24] Spivak, M. Differential Geometry, Volume I, Second edition. Publish or Perish,

Houston, TX. (1979)
[25] Spivak, M. Differential Geometry, Volume II, Second edition. Publish or Perish,
Houston, TX. (1979)
[26] Spivak, M. Differential Geometry, Volume III, Second edition. Publish or Per-
ish, Houston, TX. (1979)
[27] Steenrod, N. The Topology of Fibre Bundles. Princeton University Press,

Princeton, NJ. (1951)
[28] Varadarajan, V.S. Lie Groups, Lie Algebras and Their Representations.
Prentice-Hall, Englewood Cliffs, NJ. (1974)
Bibliography 631
[29] Warner, F.W. Foundations of Smooth Manifolds and Lie Groups. Scott, Fores-
man and Company, Glenview, IL. (1971)
[30] Westenholz, C. von Differential Forms in Mathematicel Physics. North-

Holland Publishing Company, Amsterdam. (1981)
Index
action, 231, 237 black hole, 453

Ad ∗ -equivariant, 322 Blau–Guendelman–Guth coordinates,
adapted orthonormal frame, 477 459
adjoint matrix, 272 boost, 283, 307
adjoint representation, 296 boundary, 163, 164
affine bundle, 533 boundary point, 164
affine bundle map, 533 bounded linear map, 624
linear part, 533 bracket product, 103
over a manifold, 533 bundle, 525
affine equivalence, 533 base space, 525
over a manifold, 533 fibre, 525
affine frame bundle, 344 projection, 525
affine map, 313, 462 total space, 525
linear part, 313 trivial, 529
affine space, 313 bundle map, 27, 334, 528
modelled on a vector space, 313 over a manifold, 27, 335, 529
affine subspace, 313
alternating, 103, 117 C ∞ -related, 9
alternating group, 593 canonical 1-form, 222, 292
alternation, 119, 157 canonical 2-form, 223
angular momentum, 326, 328 canonical bundle map, 338
annihilator, 75 canonical chart, 265
anti de Sitter metric, 450 canonical coordinate functions, 222
arc length, 379 canonical coordinate map, 222
arc length function, 1, 379 canonical coordinate system, 265
associated bundle, 347 canonical form, 395
atlas, 9 canonical Maurer–Cartan form, 292
automorphism., 597 canonical signature, 187
canonical transformation, 218
backward-timelike matrix, 282 Cartan distribution, 555
balanced set, 623 Cartan form, 222, 587
ball, 599, 624 Cartan formula, 159
basic vector field, 396 cartesian product, 602
BGG coordinates, 459 category, 596
Bianchi’s 1st identity, 418, 421 center, 298
Bianchi’s 2nd identity, 418, 421 Christoffel symbols, 3, 4, 431
binormal vector field, 2 closed path, 604
633
634 Index
closed universe model, 513 horizontal over the base manifold,

coadjoint representation, 296 533
codifferential operator, 203 cotangent space, 74
codimension, 20 covariant curvature tensor, 440
combined homotopy, 603 covariant derivative of lifting, 392
combined path, 604 covariant derivative of section, 386
combined path class, 607 covariant functor, 597
comoving coordinate system, 499, 511 covector field, 74
complex general linear group, 252 covering, 611
complex orthogonal group, 278 base space, 611
complex special linear group, 276 belonging to a subgroup, 617
complex symplectic group, 284 covering projection, 611
component, 601 covering space, 611
composed functor, 597 fibre, 611
composition of maps, 591 multiplicity, 616
configuration space, 240 n -fold, 616
conformal, 471 covering manifold, 257
conformally related metrics, 471 critical energy density, 516
conjugate momentum, 229 critical point, 18
connected, 598 critical value, 18
connected component, 601 curl, 205
connection, 360, 560 curvature form, 366, 399, 562
connection coefficients, 404, 412 curvature function, 2
connection components, 404, 412 curvature vector field, 1
connection form, 360, 561 curvaturelike tensor, 440
constant homotopy, 603 cyclic coordinate, 234
constant of the motion, 234
constant path, 604 Darboux frame field, 6
contact cotangent vector, 559 Darboux’ theorem, 221
contact manifold, 244 de Sitter metric, 450, 456, 519
contact projection, 559 definite metric, 181
continuous map, 602 derivation, 48, 540
contractible, 604 of type d∗ , 540
contraction, 97, 98, 604 of type i∗ , 540
contragradient representation, 253 derivative coordinates, 547, 569
contravariant functor, 597 diffeomorphism, 13
convex set, 599, 623 differentiable of class Ck , 13
coordinate functions, 9 differential form, 127
coordinate map, 9, 163 bundle-valued, 140
coordinate neighbourhood, 9 along a smooth map, 531
coordinate total derivative, 554, 575 closed, 164
coordinate transformations, 9 exact, 164
cosmic time, 511 vector-valued, 140
cosmological constant, 446 differential of a function, 74
cotangent bundle, 74 vector-valued, 145
Index 635
dimension of manifold, 11 Fermi–Walker transport, 494

directrice, 330 fibre bundle, 331
distribution, 108 base space, 331
integrable, 108 fibre, 331
involutive, 108 fibre bundle structure, 332
divergence, 204 local trivialization, 331, 525
dual basis, 70 projection, 331
dual bundle, 71 structure group, 331
dual form, 395 total space, 331
trivializing cover, 332, 525
eccentricity, 330 typical fibre, 332, 525
Eddington coordinates, 450 fibre derivative, 227, 228
Eddington–Finkelstein coordinates fibre metric, 188
ingoing, 453 fibre preserving, 227
outgoing, 453 flat connection, 366, 562
effective potential energy, 329 flat universe model, 513
Einstein constant„ 446 flow vector field, 446
Einstein curvature tensor, 446 foliation, 115
Einstein field equation, 446 folium, 115
embedding, 20 leaf, 115
energy, 232 forward-timelike matrix, 282
energy-momentum tensor, 446 four-velocity field, 495
equally oriented, 168 Frölicher–Nijenhuis bracket, 544
equivalence, 27, 335, 529 frame, 342
over a manifold, 27, 335, 529 frame bundle, 342
equivalent coverings, 616 frame field, 344
equivalent representations, 253 Frenet formulae, 2
equivariant, 595 Frenet frame field, 2
Euclidean space, 188 Friedmann equations, 515
Euler angles, 301 Frobenius’ integrability theorem
Euler–Lagrange equations, 588 global version, 113
Euler–Lagrange form, 587 local version, 110
even permutation, 593 full subcategory, 597
evenly covered, 611 fundamental group, 608
event, 446 fundamental group functor, 609
exponential map, 263, 422 fundamental vector field, 346
exterior covariant derivative, 365, 384,
386 G-map, 595
exterior derivative, 146, 150 G-related, 331
exterior forms, 117 Gauss map, 510
vector-valued, 117 Gauss’ divergence theorem, 211
exterior product, 121 Gauss’ equation, 505
Gauss’ formula, 3, 493
faithful representation, 253, 595 Gaussian curvature, 7, 443
Fermi derivative, 494 general linear group, 250
636 Index
generalized coordinate map, 164 horizontal projection, 364, 559, 560

generalized local chart, 164 horizontal subspace, 360, 560
geodesic, 6, 422 horizontal tangent vector, 360, 560
maximal, 422 horizontal vector field, 364
geodesic curvature, 6 Hubble age, 516
geodesic equations., 425 Hubble constant, 516
geodesic torsion, 6 Hubble parameter, 516
geometrized units, 446 Hubble’s law, 516
Geronos lemniscate, 21 hyperregular Hamiltonian, 237
global flow, 65, 215 hyperregular Lagrangian, 231
domain, 65, 215 hypersurface, 20, 507
global time-dependent flow, 213 sign, 507
domain, 213
gradient, 205 identity morphism, 596
group action immersed submanifold, 20
effective, 595 immersion, 20
free, 595 inclusion functor, 597
on the left, 293, 594 incompressible vector field, 204
on the right, 293, 595 indefinite metric, 181
transitive, 595 index of a metric, 187
without fixed point, 595 induced bundle, 338, 531
induced map, 591
Hamilton–Jacobi equation, 243, 245 infinitesimal action, 319
Hamiltonian, 225 infinitesimal affine transformation, 468
Hamiltonian equations, 225 infinitesimal isometry, 469
Hamiltonian vector field, 225, 244 inner automorphism, 295
Hodge star operator, 200, 202 inner product, 181
holonomic projection, 552 integral curve, 51, 52, 213
holonomic tangent vector, 552 initial condition, 51, 52, 213
homotopic maps relative to a subset, 602 maximal, 52, 53, 213
homotopic paths, 607 starting point, 51, 52, 213
homotopically equivalent, 604 integral manifold, 108
homotopy, 602 maximal, 108
homotopy category of topological pairs, interior, 163, 164
603 interior product, 154, 157, 217, 535, 538
homotopy class, 603 inverse curve, 378
homotopy equivalence, 603 inverse homotopy, 603
homotopy inverse, 603 inverse of a morphism, 597
homotopy type, 604 inverse path, 604
horizon, 453 inward pointing, 173
horizontal component, 360 isometric embedding, 189
horizontal cotangent vector, 559 isometric immersion, 189
horizontal form, 363, 532 isometry, 189
horizontal lift, 367 isomorphism, 597
horizontal map, 376 isotropy group, 595
Index 637
Jacobi identity, 103 Lie derivative, 99, 216

Jacobi’s equation, 494 Lie group, 247
Jacobian function, 204 automorphism, 248
jet, 546 homomorphism, 248
source, 546 isomorphism, 248
target, 546 Lie algebra of, 248
jet field, 546 Lie subgroup, 254
jet manifold, 546 one-parameter subgroup, 248
source projection, 546 representation, 253
target projection, 546 Lie transport, 102
jet space, 546 lifting, 28, 338, 529, 531, 613
lightlike vector, 281
k-th order contact, 546 Lindelöf space, 10
Kepler’s first law, 330 linear connection, 396
kernel, 154, 155 linear eccentricity, 330
Killing vector field, 469 linear isometry, 181
Koszul connection, 411 linear momentum, 325
Kronecker delta, 97 Lipschitz, 54
Kronecker delta tensor, 97 uniformly, 54
Kruskal–Szekeres coordinates, 455 Lipschitz constant, 54
local basis, 32
Lagrange 2-form, 231 local chart, 9, 164
Lagrange’s equations, 233 local coordinate system, 9, 164
Lagrangian, 587 local derivation, 14
Lagrangian density, 587 local diffeomorphism, 13
Lagrangian function, 229, 587 local flow, 52, 53
Lagrangian vector field, 232 local homeomorphism, 613
Laplace–Beltrami operator, 203 local representation, 336
law of composition, 596 locally connected, 598
associative, 596 locally Euclidean, 9
left invariant form, 290 locally finite, 15
vector-valued, 291 locally Lipschitz, 54
left translation, 247 uniformly, 54
Legendre transformation, 229 locally pathwise connected, 605
Levi–Civita connection, 429 loop, 604
Levi–Civita symbol, 199 Lorentz group, 281
Lie algebra, 103 proper, 282
abelian, 298 Lorentz manifold, 189
anti-homomorphism, 104 Lorentz metric, 188
automorphism, 104 lowering indices, 182–184
homomorphism, 104
ideal, 104 Maurer–Cartan form, 290
isomorphism, 104 vector-valued, 291
representation, 104 metric, 181, 188
subalgebra, 104 metric volume element, 199
638 Index
metrical connection, 426 oriented vector space, 168

Minkowski space, 188, 281 orientiable
momentum mapping, 322 manifold, 170
morphism, 596 vector bundle, 169
multi-index, 13 orthogonal, 185
factorial, 13 orthogonal group, 278
length, 13 orthogonal space of a set, 185
orthogonal space of a vector, 185
natural lift, 465 orthonormal, 185
natural units, 446 orthonormal frame bundle, 435, 436
negative definite, 181 osculating plane, 2
negative vector, 175 outward pointing, 173
negatively oriented
local chart, 170 parallel along a curve, 395
ordered basis, 168 parallel translation, 380, 381
Newton’s gravitational constant, 446 parallel transport, 380, 381
Newton’s second law, 241 parametric time, 517
Noether’s theorem, 326 parametrization by arc length, 1, 379
non-degenerate, 181, 184, 185 parity transformation, 282
non-orientiable partial derivation, 14, 48
manifold, 170 partial derivative, 13, 164
vector bundle, 169 of order k, 13
norm, 185 partition of unity, 15
normal bundle, 476 path, 604
normal chart, 424 final point, 604
normal coordinate system, 424 initial point, 604
normal curvature, 6 path category, 607
normal field, 507 path class, 607
normal plane, 2 final point, 607
null homotopic, 604 initial point, 607
null vector, 185 path component, 605
pathwise connected, 605
odd permutation, 593 pathwise simple, 619
open universe model, 513 Pauli spin matrices, 286
oppositely oriented, 168 perfect fluid, 446
orbit, 595 permutation, 591, 595
orientation permutation group, 591
of smooth manifold, 170 phase space, 241
of vector bundle, 169 piecewise smooth curve, 378
of vector space, 168 Poincaré lemma, 167
orientation preserving, 168 pointed topological space, 602
orientation reversing, 168 base point, 602
oriented atlas, 172 Poisson bracket, 226
oriented smooth manifold, 170 position vector, 314
oriented vector bundle, 169 position vector field, 48
Index 639
positive definite, 181 vector part, 285

positive unit normal field, 209
positive unit tangent field, 211 raising indices, 182–184
positive vector, 175 rank, 18, 182
positively oriented rate of rotation, 497
local chart, 170 rate of strain, 497
ordered basis, 168 rectifying plane, 2
principal G-bundle, 339 reduced subbundle, 342
principal connection, 360, 564 reducible connection, 434
principal curvature, 6, 510 reduction of the structure group, 342
principal direction, 6 reference frame, 495
principal directions, 510 refinement, 15
principal fibre bundle, 339, 340 reflexive, 603
automorphism, 342 regular curve, 1
base space, 341 regular Hamiltonian, 237
fibre, 341 regular Lagrangian, 231
homomorphism, 341 regular point, 18
isomorphism, 342 regular surface, 3
local trivialization, 341 regular value, 18
projection, 341 relative energy density, 516
structure group, 339, 340 reparametrization of curve, 1, 36
total space, 341 repeated jets, 572
principal normal vector field, 2 representation, 595
product bundle, 24 rest space, 495
fibre, 24 restriction of bundle, 338, 531
projection, 24 restriction of map, 591
product of paths, 621 Ricci curvature tensor, 402
prolongation of a bundle map, 550, 570 Ricci equation, 505
prolongation of a vector field, 579, 583, Riemannian manifold, 189
586 Riemannian metric, 181
proper time, 446 right translation, 247
pseudo-Euclidean, 188
pseudo-orthogonal group, 278 scalar curvature, 444
pseudo-Riemannian Schwarzschild coordinates, 450
hypersurface, 189 Schwarzschild metric, 450
manifold, 188 Schwarzschild radius, 450
submanifold, 189 Schwarzschild–de Sitter metric, 450
pseudo-tensorial form, 363 second fundamental form, 483, 485
pseudohyperboloid, 509 second order equation, 232
pseudosphere, 509 section, 27, 336, 529
pull-back, 29, 87, 92, 137, 338, 531, 593 along a smooth map, 28, 338
belonging to subbundle, 33
quaternion, 284 determined by a bundle map, 133
pure, 285 local representation, 27, 28, 529
scalar part, 285 push-forward, 552
640 Index
sectional curvature, 443 subordinate, 15

semi-holonomic 2-jet bundle, 573 suspension, 212
semi-locally simply connected, 619 symmetric, 181, 603
semidirect product, 310, 312 symmetric product, 185, 190
semimajor axis, 330 symmetrization, 185
semiminor axis, 330 symplectic action, 322
shape operator, 5, 491 symplectic chart, 222
shape tensor, 4, 488 symplectic form, 181, 218
sheet, 611 symplectic group, 288
sign of a permutation, 593 symplectic manifold, 218
signature, 187 symplectic map, 218
signature matrix, 188 system of transition maps, 335
simply connected, 609
simultaneity space, 495 tangent bundle, 35
skew hermitian, 278 tangent plane, 443
skew symmetric, 103, 117, 181, 278 tangent space, 35
slice, 109 canonical identification, 43, 317
smooth covering, 257 tangent vector, 35
smooth manifold, 9 tensor bundle, 92
with boundary, 164 tensor field, 92
smooth map, 13 bundle-valued, 140
smooth structure, 9 tensor product, 76, 78, 82, 84, 89, 152
space inversion, 282 simple element, 80
spacelike vector, 281 tensor space, 90
spacetime, 446 contravariant, 90
spatial curvature, 513 covariant, 90
special complex orthogonal group, 278 tensorial form, 363
special linear group, 275 time inversion, 282
special orthogonal group, 278 time-dependent tensor field, 216
special unitary group, 278 time-dependent vector field, 52, 212
speed of light, 446 timelike vector, 281
Spencer map, 572 topological group, 621
standard orientation, 168 topological manifold, 9
star-shaped, 604 with boundary, 163
state space, 240 topological pair, 602
Stokes’ theorem, 178, 212 topological vector space, 623
structure equation torsion form, 397
first, 415 torsion function, 2
second, 416 tortoise coordinate, 452, 456
subbundle, 31, 342 total derivative, 554, 574
subbundle property, 31 total inversion, 282
subcategory, 597 trace, 95, 96
submanifold, 20 transition map, 331
submanifold property, 22 transitive, 603
submersion, 20 translation, 313, 616
Index 641
transpose, 181 f -related, 50

transposition, 592 along a smooth map, 47, 531
trivializing cover, 84 complete, 66
tubular map, 481 left invariant, 247
tubular neighbourhoods, 481 local 1-parameter group, 67
pull-back, 50
unit normal field, 507 push-forward, 50
unit normal vector, 3 velocity transformation, 283, 307
unit tangent vector field, 1 vertical component, 360
unit vector, 185 vertical form, 364
unit vector field, 188 vertical projection, 364, 560
along a smooth map, 188 vertical projection., 552
unitary group, 278 vertical subspace, 332, 526
unitary matrix, 272 vertical tangent bundle, 526
upper half space, 163 vertical tangent vector, 332, 526, 552
upper triangular matrix, 273 vertical vector field, 363
volume element, 169
vector bundle, 24 volume preserving map, 204
base space, 24 vorticity, 497
fibre, 24 vorticity vector field, 498
local trivialization, 25
projection, 24 wedge product, 121, 127, 152, 153, 385
total space, 24 Weingarten formula, 4, 493
trivial, 27 white hole, 453
trivializing cover, 25 world line, 446
vector field, 47, 52, 63
1-parameter group, 67 zero section, 28

Tangent Bundle

Uploaded by

Copyright:

Available Formats

Tangent Bundle

Uploaded by

Document Information

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

Tangent Bundle

Uploaded by

Copyright:

Available Formats

SMOOTH MANIFOLDS

AND FIBRE BUNDLES

© 2017 by Taylor & Francis Group, LLC

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-4987-9671-2 (Hardback)

Library of Congress Cataloging‑in‑Publication Data

Names: Johannesen, Steinar.

Visit the Taylor & Francis Web site at

and the CRC Press Web site at

2 SMOOTH MANIFOLDS AND VECTOR BUNDLES 9

3 VECTOR FIELDS AND DIFFERENTIAL EQUATIONS 47

5 DIFFERENTIAL FORMS 117

6 INTEGRATION ON MANIFOLDS 163

7 METRIC AND SYMPLECTIC STRUCTURES 181

8 LIE GROUPS 247

9 GROUP ACTIONS 293

10 FIBRE BUNDLES 331

Tensorial forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

11 ISOMETRIC IMMERSIONS AND THE SECOND FUNDAMENTAL

12 JET BUNDLES 525

Calculus of variations . . . . . . . . . . . . . . . . . . . . . . . . . . . 587

dles, which is specialized to linear connections and metrical connections in pseudo-

Acknowledgment In May 2013 Steinar Johannesen had a stroke and he was

for t ∈ I. The curve α = γ ◦ s−1 : J → R3 defined on the interval J = s(I) is a

1.2 The Frenet frame Let c : I → R3 be a smooth curve parameterized by arc

κ (s) = kt′ (s)k

n(s) = t′ (s)/ kt′ (s)k

b(s) = t(s) × n(s)

n′ (s) · t(s) = −n(s) · t′ (s) = −κ and n′ (s) · n(s) = 0

for s ∈ I. Introducing the torsion function τ : I → R of c defined by

τ (s) = n′ (s) · b(s)

for s ∈ I, we therefore have that

In the same way we see that

for s ∈ I, which implies that

1.3 The Gauss’ formulae Consider a surface in R3 parameterized by a smooth

The derivative of e j with respect to ek , which is given by

where a jk = e jk · u for 1 ≤ j, k ≤ 2 . The coefficients Γ jk

1.4 The metric on a surface Let γ : I → U be a regular curve in U, and consider

[ jk, l ] = 12 {( [l j, k ] + [k j, l ] ) + ( [lk, j ] + [ jk, l ] )

where (gi j ) is the inverse of the matrix (gi j ) .

for k = 1, 2 . From Gauss’ formulae and the equations u · e j = 0 it follows that

for 1 ≤ i, k ≤ 2 . The coefficients a ki in the Weingarten formulae are said to be obtained

for i = 1, 2, we have that

and Gauss’ formulae it follows that

e′ = −κg t + τg u and u′ = −κn t − τg e

the tangent plane at p consisting of eigenvectors for S with eigenvalues λ1 and λ2 ,

κn (at1 + bt2) = a2 λ1 + b2 λ2 = λ1 + b2 (λ2 − λ1 )

The purpose of this chapter is to generalize the differential calculus in Euclidean

2.1 Definition A topological space M is said to be locally Euclidean if it is

2.2 Proposition If A is an atlas on a locally Euclidean space M, then there is a

2.3 Definition A topological space X is called a Lindelöf space if every open

2.4 Proposition If a topological space X is second countable, then it is a Lindelöf

2.5 Corollary In a second countable space X there cannot exist an uncountable

2.6 Proposition A locally Euclidean space M with an atlas A is second countable

fi (x) = gi (x) = (x1 , ..., xi−1 , xi+1 , ..., xn+1 ).

Their inverses are given by

from which we see that the coordinate transformations f j ◦ fi−1 , g j ◦ g−1

The smooth structure on Sn can also be defined by an atlas consisting of only

φ1 (x) = (1 − xn+1)−1 (x1 , ..., xn ) and