T. Wedhorn. Manifolds, Sheaves, and Cohomology

Springer Studium Mathematik – Master
Torsten Wedhorn
Manifolds,
Sheaves,
and Cohomology
Series editors
M. Aigner, Freie Universität Berlin, Berlin, Germany
H. Faßbender, Technische Universität Braunschweig, Braunschweig, Germany
B. Gentz, Universität Bielefeld, Bielefeld, Germany
D. Grieser, Universität Oldenburg, Oldenburg, Germany
P. Gritzmann, Technische Universität München, Garching, Germany
J. Kramer, Humboldt-Universität zu Berlin, Berlin, Germany
V. Mehrmann, Technische Universität Berlin, Berlin, Germany
G. Wüstholz, ETH Zürich, Zürich, Switzerland
The series „Springer Studium Mathematik“ is aimed at students of all areas of mathemat-
ics, as well as those studying other subjects involving mathematics, and anyone working in
the field of applied mathematics or in teaching. The series is designed for Bachelor’s and
Master’s courses in mathematics, and depending on the courses offered by universities,
the books can also be made available in English.
More information about this series at http://www.springer.com/series/13893

Torsten Wedhorn
Manifolds, Sheaves,
and Cohomology
Torsten Wedhorn
Technische Universität Darmstadt
Darmstadt, Germany

ISBN 978-3-658-10632-4 ISBN 978-3-658-10633-1 (eBook)
DOI 10.1007/978-3-658-10633-1
Library of Congress Control Number: 2016947452
Mathematics Subject Classification (2010): 14-0, 18-01
Springer Spektrum
© Springer Fachmedien Wiesbaden 2016
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made.
Editor: Ulrike Schmickler-Hirzebruch
Printed on acid-free paper
This Springer Spektrum imprint is published by Springer Nature

The registered company is Springer Fachmedien Wiesbaden
Preface
The language of geometry has changed drastically in the last decades. New and fun-
damental ideas such as the language of categories, sheaves, and cohomology are now
indispensable in many incarnations of geometry, such as the theory of complex analytic
spaces, algebraic geometry, or non-archimedean geometry. This book is intended as an
introduction to these ideas illustrating them by example with the most ubiquitous branch
of geometry, the theory of manifolds.
In its contemporary form, a “geometric object” is usually defined as an “object” that
“locally” “looks like” a “standard geometric object”. Depending on the geometry that one
is interested in, there will be very different “standard geometric objects” as the basic build-
ing blocks. For the theory of (finite-dimensional) manifolds one chooses open subsets of
finite-dimensional R- or C-vector spaces together with their “differentiable structure”.
To make the notion of a geometric object precise, one proceeds in general as follows.
First one introduces the language of categories yielding the notions of objects and the
precise meaning of “looks like” as being isomorphic in that category. Next one has to
find a (maybe very large) category C that contains the “standard geometric objects” as a
subcategory and in which it makes sense to use the word “locally”. Then finally one can
give the precise definition of a geometric object as an object of C that is locally isomorphic
to an object in the subcategory of standard geometric objects.
In this textbook we choose for such a category C the category of ringed spaces, al-
though other choices such as ringed topoi might even be more natural. But for the sake
of an introduction, ringed spaces seem to be the most accessible choice and they are also
adequate for many geometric theories such as differential geometry, complex geometry,
or the theory of schemes in algebraic geometry. Moreover a good grasp of ringed spaces
will help a great deal in understanding more abstract concepts such as ringed topoi.
For the theory of manifolds the basic building blocks are open subsets X of finite-
dimensional R- or C-vector spaces together with the collection of their “functions”, where
a “function” will be an ˛-fold continuously differentiable function for ˛ 2 N0 [ f1g
or an analytic function (also called C ! -function) on some open subspace of X. Then a
continuous map between standard geometric objects is a C ˛ -map (˛ 2 N0 [f1; !g) if and
only if the composition with f sends a C ˛ -function to a C ˛ -function. This allows one to
view such standard geometric objects and their structure preserving maps as a subcategory
v
vi Preface
of (locally) ringed spaces over R or C. Hence one obtains the notion of a geometric object
by the general procedure explained above. These are called premanifolds.
A manifold will then be a premanifold whose underlying topological space has cer-
tain good properties (being Hausdorff and second countable). Let me briefly digress on
this choice of terminology. First of all I follow the classical terminology. But an even
more compelling reason to restrict the class of manifolds by asking for these topologi-
cal properties is a multitude of techniques and results where these topological properties
are indispensable hypotheses (such as embedding results or the theory of integration on
manifold). In this textbook this is not that visible as many of such results are not covered
here. Hence we will more often encounter premanifolds than manifolds and it would have
been tempting to change terminology, if only to get rid of the annoying “pre-” everywhere.
But I decided against this to remind the reader that the contents of this book are only the
very beginning of a journey into the wondrous world of differential geometry – a world in
which very often manifolds and not premanifolds are the central objects.
A fundamental idea in modern mathematics is the notion of a sheaf. Sheaves are needed
to define the notion of a ringed space but their usefulness goes far beyond this. Sheaves
embody the principle for passing from local to global situations – a central topic in math-
ematics. In the theory of smooth manifolds one can often avoid the use of sheaves as there
is another powerful tool for local-global constructions, namely partitions of unity. Their
existence corresponds to the fact that the sheaf of smooth functions is soft (see Chap. 9).
But this is particular to the case of real C ˛ -manifolds with ˛ 1. In all other geometries
mentioned before the theory of sheaves is indispensable. Hence sheaves will be a central
topic of this book.
Together with sheaves and manifolds (as ringed spaces), the third main topic is the co-
homology of sheaves. It is the main tool to make use of sheaves for local-global problems.
As a rule, it allows one to consider the obstruction for the passage from local to global
objects as an element in an algebraic cohomology object of a sheaf, usually a group.
Moreover, the formalism of cohomology also yields a powerful tool to calculate such ob-
structions. In addition, many interesting objects (such as fiber bundles) are classified by
the cohomology of certain sheaves.
It is not the goal of this book – and also would have been clearly beyond my abilities –
to give a new quick and streamlined introduction to differential geometry using clever
arguments to obtain deep results with minimal technical effort. Quite the contrary, the
focus of this book is on the technical methods necessary to work with modern theories
of geometry. As a principle I tried to explain these techniques in their “correct general-
ity” (which is certainly a very subjective notion) to provide a reliable point of departure
towards geometry.
There are some instances where I deviate from this principle, either because of lack
of space or because I think that the natural generality and abstractness would seriously
conceal the underlying simple idea. This includes the following subjects:
Preface vii
1. One might argue that the natural framework for sheaves are sheaves on an arbitrary site
(i.e., a category endowed with a Grothendieck topology) or even general topos theory.
But in my opinion this would have seriously hampered the accessibility of the theory.
2. Instead of working with manifolds modeled on open subsets of Rn or C n one might
argue that it is more natural (and not much more difficult) to model them on open
subsets of arbitrary Banach spaces. I decided against this because the idea of the book
is to demonstrate general techniques used in geometry in the most accessible example:
finite-dimensional real and complex manifolds.
3. In the chapter on cohomology I do not use derived categories, although I tried to for-
mulate the theory in such a way that a reader familiar with the notion of a derived
category can easily transfer the results into this language1 .
Moreover, there are several serious omissions due to lack of space, among them Lie alge-
bras, manifolds with corners (or, more generally, singular spaces), and integration theory –
just to name a few. The educated reader will find many more such omissions.
Prerequisits
The reader should have knowledge of basic algebraic notions such as groups, rings and
vector spaces, basic analytic and topological notions such as differentialbility in several
variables and metric spaces.
Further prerequisites are assembled in five appendix chapters. It is assumed that the
reader knows some but not all of the results here. Therefore many proofs and many exam-
ples are given in the appendices. These appendices give (Chap. 12) a complete if rather
brisk treatment of basic concepts of point set topology, (Chap. 13) a quick introduction
to the language of categories focused on examples, (Chap. 14) some basic definitions and
results of abstract algebra, (Chap. 15) those notions of homological algebra necessary to
cope with the beginning of cohomology theory, and (Chap. 16) a reminder on the notion of
differential and analytic functions on open subsets of finite-dimensional R- and C-vector
spaces.
Outline of contents
The main body of the text starts with two preliminary chapters. The first chapter (Chap. 1)
introduces more advanced concepts from point set topology. The main notions of the first
three sections are paracompact and normal spaces, covering important techniques like
Urysohn’s theorem, the Tietze extension theorem and the Shrinking Lemma for paracom-
pact Hausdorff spaces. The last two sections of this chapter focus on separated and proper
maps. In Chap. 2 basic notions of algebraic topology used in the sequel are introduced.
Here we restrict the contents to those absolutely necessary (but with complete proofs) and
ignore all progress made in the last decades.
1
Of course, for most readers that are familiar with derived categories the cohomology chapter will
not contain that many new results anyway.
viii Preface
In the third chapter (Chap. 3) we introduce the first main topic of the book: sheaves. We
introduce two (equivalent) definitions of sheaves. The first one is that of a rule attaching
to every open set of a topological space a set of so-called sections such that these sections
can be glued from local to global objects. This is also the definition that generalizes from
topological spaces to more abstract geometric objects such as sites. There is another point
of view of sheaves that works just fine for sheaves on topological spaces, namely étalé
spaces. It is proved that both concepts are equivalent and explained that some construc-
tions for sheaves are more accessible via the first definition (such as direct images) and
some are more accessible via étalé spaces (such as inverse images).
In Chap. 4 we introduce the class of geometric objects that will be studied in this book:
manifolds. As explained above, we start by defining the very general category of ringed
spaces over a fixed ring. Then we explain how to consider open subsets of real or complex
finite-dimensional vector spaces as ringed spaces over R or C. This yields our class of
standard geometric objects. Then a premanifold is by definition a ringed space that is
locally isomorphic to such a standard geometric object.
The central topic of Chap. 5 is that of linearization. We start by linearizing manifolds
by introducing tangent spaces. Then the derivative of a morphism at a point is simply
the induced map on tangent spaces and we can think of it as a pointwise linearization of
the morphism. Next we study morphisms of (pre)manifolds that can even be locally lin-
earized. These are precisely the morphisms whose derivatives have a locally constant rank.
Examples are immersions, submersions, and locally constant maps. These linearization
techniques are then used in the remaining sections of the chapter to study submanifolds,
their intersections (or, more generally, fiber products of manifolds), and quotients of man-
ifolds by an equivalence relation.
Chapter 6 introduces the symmetry groups in the theory of manifolds, the Lie groups. It
focuses on actions of Lie groups on manifolds. The construction of quotients of manifolds
in Chap. 5 then yields the existence of quotients for proper free actions of Lie groups.
In Chap. 7 we start to study local-global problems by introducing the first cohomology
of a sheaf of (not necessarily abelian) groups via the language of torsors or, equivalently,
by the language of Čech cocycles.
This is used in Chap. 8 to classify fiber bundles that are important examples of locally
but not globally trivial objects. We start the chapter by introducing the general notion of
a morphism that looks locally like a given morphisms p. Very often it is useful to restrict
the classes of local isomorphisms. This yields the notion of a twist of p with a structure
sheaf that is a subsheaf of the sheaf of all automorphisms of p. Specializing to the case
that p is a projection and the structure sheaf is given by the faithful action of a Lie group
G on the fiber then yields the notion of a fiber bundle with structure group G. Specializing
further we obtain the important notions of G-principal bundles and vector bundles. We
explain that vector bundles can also be described as certain modules over the sheaf of
functions of the manifold. In the last two sections we study the most important examples
of vector bundles on a manifold, the tangent bundle and the bundles of differential forms.
In particular we will obtain the de Rham complex of a manifold.
Preface ix
As mentioned above, for real C ˛ -manifolds with ˛ 1 it is often possible to use other
techniques than sheaves for solving local-global problems. The sheaf-theoretic reason for
this lies in the softness of the sheaf of C ˛ -functions. This notion is introduced in Chap. 9.
We deduce from the softness of the structure sheaf the existence of arbitrary fine partitions
of unity. In the last section we show that the first cohomology of a soft sheaf is trivial and
deduce immediately some local-global principles. For real C ˛ -manifolds with ˛ 1
all these results can also be obtained via arguments with a partition of unity. But these
examples illustrate how to use the triviality of certain cohomology classes also in cases
where partitions of unity are not available, for instance for certain complex manifolds.
After giving the rather ad hoc definition of the first cohomology in Chap. 7 (that had the
advantage also to work for sheaves of not necessarily abelian groups) we now introduce
cohomology in arbitrary degree in Chap. 10. After a quick motivation on how to do this,
it becomes clear that it is more natural not to work with a single sheaf but with a whole
complex of sheaves. This is carried out in the first three sections. Applying the whole
formalism of cohomology to the de Rham complex we obtain de Rham’s theorem relating
de Rham cohomology and cohomology of constant sheaves in Sect. 10.4. We conclude the
chapter by proving an other important result: the theorem of proper base change, either in
the case of arbitrary topological spaces and proper separated maps (giving the theorem its
name) or for metrizable spaces and closed maps.
In the last chapter (Chap. 11) we focus on the cohomology of constant sheaves. We
show that for locally contractible spaces this is the same as singular cohomology, which
is quickly introduced in Sect. 11.1. In particular we obtain the corollary that we can
describe the de Rham cohomology on a manifold also via singular cohomology. Then we
use the proper base change theorem to prove the main result of this chapter, the homotopy
invariance of the cohomology of locally constant sheaves for arbitrary topological spaces.
We conclude with some quick applications.
As with almost every mathematical text, this book contains a myriad of tiny exercises
in the form of statements where the reader has to make some straightforward checks to
convince herself (or himself) that the statement is correct. Beyond this, all chapters and
appendices end with a group of problems. Some of the problems are sketching further
important results that were omitted from the main text due to lack of space, and the reader
should feel encouraged to use these problems as a motivation to study further literature on
the topic.
Acknowledgments
This book grew out of a lecture I gave for third year bachelor students in Paderborn and I
am grateful for their motivation to get a grip on difficult and abstract notions. Moreover,
I thank all people who helped to improve the text by making comments on a preliminary
version of the text, in particular Benjamin Schwarz and Joachim Hilgert. Special thanks
go to Christoph Schabarum for TeXing (and improving) some of the passages of the book,
to Jean-Stefan Koskivirta for designing lots of exercises for my class of which almost all
x Preface
of them can now be found in the problem sections, and to Joachim Hilgert for giving me
access to his collection of figures.
Paderborn
January, 2016 Torsten Wedhorn
Standard Notation
For a set I and elements i; j 2 I we use the “Kronecker delta”

8
<1; if i D j I
ıij WD
:0; if i ¤ j :
We denote by N the set f1; 2; : : : g of natural numbers (without 0), N0 WD N [ f0g. Z

denotes the ring of integers, Q the field of rational numbers, R (respectively C) the field
of real (respectively complex) numbers endowed with the metric given by the standard
absolute value. The symbol K always either denotes R or C.
To unify the notation for n-fold continuously differentiable, smooth, and analytic maps
we define
b 0 WD N0 [ f1; !g;
N b WD N [ f1; !g:
N
We extend the usual total order on N0 to N b 0 by requesting n < 1 < ! for all n 2 N0 .
We also define 1 ˙ ˛ WD 1 and ! ˙ ˛ WD ! for all ˛ 2 N0 .
A set I is said to be countable if there exists an injective map I ,! N. In particular,
all finite sets are countable.
By a monoid we mean a set M endowed with an associative binary operation M M !
M , .m; m0 / 7! mm0 , having an identity element eM . A monoid homomorphism is a map
'W M ! N such that '.mm0 / D '.m/'.m0 / for all m; m0 2 M and '.eM / D eN . A
monoid M is called commutative if mm0 D m0 m for all m; m0 2 M .
All rings are assumed to have a unit, usually denoted by 1, and ring homomorphisms
are assumed to preserve the unit. All fields are assumed to be commutative.
If R is a ring, by an R-module we mean a left R-module if not stated otherwise (for a
reminder on some notions about modules see Appendix Section 14.1). If R is commuta-
tive, we will not distinguish between left R-modules and right R-modules.
In a metric space .X; d / we denote for x0 2 X and r 2 R>0 the open and closed ball
by
Br .x0 / WD f x 2 X I d.x; x0 / < r g ;

Br .x0 / WD f x 2 X I d.x; x0 / r g :
xi
xii Standard Notation
Let V be a finite-dimensional K-vector space. If not otherwise stated, we endow a subset

X of V always with the topology induced by some norm on V . As all norms on V are
equivalent, this topology does not depend on the choice of the norm. See Appendix 12 for
a reminder on topological spaces.
Contents
1 Topological Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Countability Properties for Topological Spaces . . . . . . . . . . . . . . 1
1.2 Paracompact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Normal Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Separated Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Proper Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 Algebraic Topological Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . 21

2.1 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Path Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Fundamental Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Covering Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Fundamental Groups and Coverings of Topological Groups . . . . . . 34
2.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1 Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Stalks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Sheaves Attached to Presheaves . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Sheaves and Étalé Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Direct and Inverse Images of Sheaves . . . . . . . . . . . . . . . . . . . . 57
3.6 Limits and Colimits of Sheaves . . . . . . . . . . . . . . . . . . . . . . . . 63
3.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.1 Ringed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Premanifolds and Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Examples of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 Topological Properties of (Pre)Manifolds . . . . . . . . . . . . . . . . . . 84
xiii
xiv Contents
4.5 Basic Constructions and Further Examples of Manifolds . . . . . . . . 85

4.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5 Linearization of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.1 Tangent Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.2 Local Properties of Morphisms, Local Isomorphisms, Immersions,
Submersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3 Morphisms of Locally Constant Rank . . . . . . . . . . . . . . . . . . . . 101
5.4 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.5 Fiber Products of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.6 Quotients of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

6.1 Definition and Examples of Lie Groups . . . . . . . . . . . . . . . . . . . 123
6.2 Lie Group Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.3 Quotients by Lie Group Actions . . . . . . . . . . . . . . . . . . . . . . . 128
6.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7 Torsors and Non-abelian Čech Cohomology . . . . . . . . . . . . . . . . . . . 139

7.1 Torsors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2 Non-abelian Čech Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 142
7.3 First Term Sequence of Cohomology . . . . . . . . . . . . . . . . . . . . 147
7.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
8 Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
8.1 Twists, Fiber Bundles, and Principal Bundles . . . . . . . . . . . . . . . 154
8.2 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.3 OX -Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.4 Vector Bundles and Finite Locally Free Modules . . . . . . . . . . . . . 173
8.5 Tangent Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
8.6 Differential Forms and De Rham Complex . . . . . . . . . . . . . . . . . 180
8.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
9 Soft Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

9.1 Definition and Examples of Soft Sheaves . . . . . . . . . . . . . . . . . . 193
9.2 Softness of Sheaves of Differentiable Functions . . . . . . . . . . . . . . 198
9.3 Triviality of H 1 for Soft Sheaves and Applications . . . . . . . . . . . . 200
9.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
10 Cohomology of Complexes of Sheaves . . . . . . . . . . . . . . . . . . . . . . . 205

10.1 Strategy for the Definition of Cohomology of Sheaves . . . . . . . . . . 206
10.2 Definition of Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Contents xv
10.3 Acyclic Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

10.4 Applications: Theorems of De Rham and of Mittag-Leffler . . . . . . . 219
10.5 Cohomology and Inverse and Direct Image . . . . . . . . . . . . . . . . . 222
10.6 Proper Base Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
11 Cohomology of Constant Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . 233

11.1 Singular Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
11.2 Cohomology and Singular Cohomology . . . . . . . . . . . . . . . . . . . 238
11.3 Homotopy Invariance of Cohomology . . . . . . . . . . . . . . . . . . . . 240
11.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
11.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
12 Appendix A: Basic Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

12.1 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
12.2 Continuous, Open, and Closed Maps . . . . . . . . . . . . . . . . . . . . . 247
12.3 Closure and Interior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
12.4 Construction of Topological Spaces . . . . . . . . . . . . . . . . . . . . . 250
12.5 Local Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
12.6 Hausdorff Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
12.7 Connected Spaces and Locally Constant Maps . . . . . . . . . . . . . . . 257
12.8 Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
12.9 Topological Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
12.10 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
13 Appendix B: The Language of Categories . . . . . . . . . . . . . . . . . . . . 271

13.1 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.2 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
13.3 Digression: Ordered Sets, Transfinite Induction, and Zorn’s Lemma . 276
13.4 Limits and Colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
13.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
14 Appendix C: Basic Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

14.1 The Category of Modules over a Ring . . . . . . . . . . . . . . . . . . . . 291
14.2 Multilinear Algebra and Tensor Products . . . . . . . . . . . . . . . . . . 297
14.3 Tensor Algebra, Exterior Algebra, Symmetric Algebra . . . . . . . . . . 303
14.4 Modules over Commutative Rings . . . . . . . . . . . . . . . . . . . . . . 308
14.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
15 Appendix D: Homological Algebra . . . . . . . . . . . . . . . . . . . . . . . . . 317

15.1 Homotopy Category of Modules . . . . . . . . . . . . . . . . . . . . . . . 317
15.2 Diagram Chases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
15.3 Injective Modules and K-injective Complexes . . . . . . . . . . . . . . . 323
xvi Contents
15.4 Abelian Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

15.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
16 Appendix E: Local Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331

16.1 Differentiable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
16.2 Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
16.3 Higher Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
16.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Topological Preliminaries
1
In this chapter we prove some results on topological spaces that will be needed later and
go beyond the basic topological results and notions assembled in the Appendix Chap. 12.
The chapter consists of two independent parts.
In the first part (Sects. 1.1–1.3) we introduce, after a quick review of countability
properties, paracompact spaces. This is one of the central topological notions in this
book. We show that the following classes of topological spaces are paracompact: Metriz-
able spaces (Proposition 1.13) and locally compact, second countable Hausdorff spaces
(Proposition 1.10), see also Remark 1.14. Then we show that paracompact Hausdorff
spaces are normal (Proposition 1.18). Hence Urysohn’s separation theorem, the Tietze
extension theorem (Theorem 1.15), and the shrinking lemma (Proposition 1.20, Corol-
lary 1.21) are available for paracompact spaces.
The second part (Sects. 1.4 and 1.5) introduces relative versions of Hausdorff spaces
and compact spaces: separated maps and proper maps.
1.1 Countability Properties for Topological Spaces
Recall that we call a set M countable if there exists an injective map M ! N (equiva-
lently, M D ; or there exists a surjective map N ! M ). Hence any finite set is countable.
Every subset of a countable set is again countable. Countable unions and finite products
of countable sets are again countable.
Definition 1.1 (Countability properties). Let X be a topological space.
1. X is called first countable if every point of x has a countable neighborhood basis.

2. X is called second countable if the topology has a countable basis.
3. X is called a Lindelöf space if every open covering of X has a countable subcovering.
© Springer Fachmedien Wiesbaden 2016 1

T. Wedhorn, Manifolds, Sheaves, and Cohomology, Springer Studium Mathematik – Master,
DOI 10.1007/978-3-658-10633-1_1
2 1 Topological Preliminaries
4. X is called separable if it contains a countable dense subset.

5. X is called -compact if it is the union of countably many compact subspaces.
Example 1.2. Let X be a topological space.
1. Every metrizable space X is first countable. Let d be a metric inducing the given
topology on X. For x 2 X the sets
n o
B1=n .x/ WD f y 2 X I d.x; y/ < 1=n g I n 2 N
n o
B1=n .x/ WD f y 2 X I d.x; y/ 1=n g I n 2 N
both form countable neighborhood bases of x.

2. The space Rn is second countable and -compact for all n 0: Choose a norm on Rn
yielding a metric d . Then a countable basis is given by
˚
B1=m .x/ I x 2 Qn ; m 2 N ;
and Rn is the union of the countably many closed balls BN .0/, N 2 N.
3. Let X be an uncountable set endowed with the discrete topology. Then X is first
countable but not second countable.
Remark 1.3.
1. Every subspace Z of a second countable space X is second countable: If BX a count-

able basis of the topology of X, then f B \ Z I B 2 BX g is a countable basis for Z.
In particular, every subspace of Rn is second countable.
2. Let f W X ! Y be a surjective open continuous map. If B is a basis of the topology of
X, then f f .U / I U 2 B g is a basis of the topology of Y . In particular we see that if
X is second countable, then Y is second countable.
Q
3. Let .Xn /n be a countable family of second countable spaces. Then n Xn is second
countable: If Bn is a countable basis for Xn , then
( )
Y
Un I Un 2 Bn , Un D Xn for all but finitely many n
n
Q
is a countable basis for n Xn .
4. Suppose that X has a countable open covering .Un /n such that the subspace Un is
second countable for all n. Then X is second countable: If Bn is a countable basis for
S
Un , then n Bn is a countable basis for X.
1.1 Countability Properties for Topological Spaces 3
Proposition 1.4. Let X be a topological space.
1. If X is second countable, then X is first countable, separable, and a Lindelöf space.

2. Suppose that X is a metrizable Lindelöf space. Then X is second countable.
Problem 1.4 shows that a separable metrizable space is also second countable.
Proof. 1. Let B D f Bn I n 2 N g be a countable basis for X. Then for every point x 2 X

the set f B 2 B I x 2 B g is a countable neighborhood basis of x. For every non-empty
B 2 B choose xB 2 B and set Q WD f xB I B 2 B g. Then Q is dense in X because it
meets every open subset of X. Hence X is first countable and separable.
It remains to show that X is Lindelöf. Let .Ui /i 2I be an open covering of X. Let M
be the set of n 2 N such that there exists in 2 I with Bn Uin . As .Ui /i is an open
covering and B is a basis, .Bn /n2M is an open covering. Hence .Uin /n2M is a countable
subcovering of .Ui /i .
˚ 2. Fix n 2 N. As X is Lindelöf, we can choose a countable subcovering Un of
B1=n .x/ I x 2 X . Then the union of all Un is a countable basis for X.
Proposition 1.5. Let G be a topological group.
1. Suppose that G is connected and that the neutral element of G has a compact neigh-
borhood C . Then G is -compact.
2. Suppose that G is -compact. Then there exists for every neighborhood U of the
S
neutral element e of G a sequence .gn /n in G such that G D n gn U .
3. If G is -compact and first countable, then G is second countable.
Proof. 1. As G is connected, G is generated by C (Appendix ˚ Corollary 12.59).

Re-
placing C by the compact subset C [ C 1 (with C 1 WD g 1 I g 2 C ), we see
S
G D n2N C n with C n D f g1 g2 gn I gi 2 C g. But C n is compact, as the im-
age of the compact n-fold product C C under the n-fold multiplication map
G G ! G.
S
2. Replacing U by its interior we may assume that U is open. Let G D n2N Kn
with Kn G compact subspace. For each n the open subsets kU for k 2 Kn cover
S
the compact set Kn . Therefore there exist kn;1 ; : : : ; kn;rn 2 Kn with Kn i kn;i U
S S
and hence G n2N 1i rn kn;i U . Renumbering the countably many kn;i yields the
desired result.
3. Let .Un /n2N be a fundamental system of neighborhoods of e. We first claim that for
all n 2 N there exists m 2 N such that Um1 Um Un . Indeed, consider the continuous
map ˛W G G ! G, .g; h/ 7! g 1 h. Then ˛ 1 .Un / is a neighborhood of .e; e/ in G G.
Hence there exist m; m0 2 N such that Um Um0 ˛ 1 .Un /. Replacing m and m0 both
by some r 2 N with Ur Um \ Um0 we may assume that m D m0 . This proves the claim.
S
By 2. there exist for all n 2 N sequences .gn;k /k in G such that G D k2N gn;k Un .
We show that f gn;k Un I k; n 2 N g is a basis of the topology of G. In fact, let V be an
open subset of G and g 2 V . Then there exists n 2 N such that gUn V . By our claim
above we may choose m 2 N such that Um1 Um Un . If we choose k 2 N such that
g 2 gm;k Um , then gm;k Um gUm1 Um gUn V .
Combining 1. and 3. of Proposition 1.5 we obtain:
Corollary 1.6. A connected locally compact first countable topological group is second
countable.
1.2 Paracompact Spaces
Definition 1.7 (Refinement). Let X be a topological space and let U D .Ui /i 2I be

a covering. A covering .Vj /j 2J is called a refinement of U if for all j 2 J there exists
i 2 I with Vj Ui . In other words, there exists a map ˛W J ! I such that Vj U˛.j /
for all j 2 J .
A refinement .Vj /j of U is called open, if Vj is open in X for all j 2 J .
Any subcovering is a refinement.
Definition 1.8 (Paracompact spaces). A topological space X is called paracompact if

every open covering has an open refinement that is locally finite.
We do not suppose that paracompact spaces are Hausdorff (as it is often done, for
instance in [BouGT1]).
Example 1.9.
1. Every compact space is paracompact.
2. Every discrete topological space X is paracompact: The covering .fxg/x2X is a locally
finite open refinement of every open covering.
Proposition 1.10. Let X be a locally compact and second countable Hausdorff space.
Then X has the following properties:
S
1. There exists a sequence .Kn /n of compact subspaces of X such that X D n Kn and
ı
Kn KnC1 for all n 2 N. In particular X is -compact.
1.2 Paracompact Spaces 5
2. Every open covering of X has a countable locally finite open refinement .Vl /l2N such
that Vl is compact for all l. In particular X is paracompact.
Proof. We first construct a countable base B such that B is compact for all B 2 B. Let
B0 be a countable base of X, Cx a compact neighborhood of x 2 X (automatically closed
in X because X is Hausdorff). Define
B WD f V \ W I V 2 B0 and W 2 B0 with W Cx for some x 2 X g :

S
Then B is a countable basis because X D x Cx . The closure of each set B 2 B is
a closed subset of Cx for some x and hence compact. Write B D fB1 ; B2 ; : : : g.
Proof of 1. We construct compact subspaces Kn of X and i.n/ 2 N0 with i.n/ >
ı
S
i.n 1/ such that B1 [ [ Bi.n/ Kn KnC1 for all n 2 N0 (then n2N Kn
S
i 2N Bi D X).
Define Kn and i.n/ inductively. Set K0 WD ; and i.0/ WD 0. Now let n 1. As
S
Kn1 is compact, we find i.n/ 2 N with i.n/ > i.n 1/ and Kn1 1j i.n/ Bj . Set
S
Kn WD 1j i.n/ Bj . Then Kn is compact and
[
Kn1 Bj Knı :
1j i.n/
Proof of 2. Let U D .Ui /i be an open covering of X. Let W D .Wj /j 2J be the

refinement of U consisting of those W 2 B with W Ui for some i. Fix n 2 N and set
ı
Wj;n WD Wj \ .KnC2 n Kn1 /. Then
[
Wj;n KnC1 n Knı :
„ ƒ‚ …
j 2J
compact
S
Let J.n/ J be finite such that j 2J.n/ Wj;n KnC1 n Knı . Let V D .Vl /l be the
open covering consisting of Wn;j for n 2 N and j 2 J.n/ (it is a covering because
S
n Kn D X). Then V is countable and a refinement of U by construction. We have
Wn;j KnC2 and hence Vl is compact for all l. Finally, V is locally finite: For x 2 X let
ı
n 2 N with x 2 Kn . Then KnC1 is an open neighborhood of x, which intersects only the
finitely many Wj;m with j 2 J.m/ and m n C 1 (by the definition of Wj;m ).
Remark 1.11. Let X be a paracompact space. Then every closed subspace Y of X is

paracompact. Indeed, the proof is the same as for compact spaces. Let U be an open
S
covering of Y . Hence U D .Ui \Y /i 2I for open subsets Ui of X with Y i Ui . Adding
X nY to .Ui /i , we obtain an open covering of X. This has a locally finite refinement .Vj /j
because X is paracompact. Then .Vj \ Y /j is a locally finite refinement of U.
Open subspaces of paracompact spaces are not paracompact in general ([BouGT1]

Chap. I, §9, Exercise 11). Hence we introduce the following terminology.
Remark and Definition 1.12 (Hereditarily paracompact). A topological space X is

called hereditarily paracompact if the following equivalent conditions are satisfied:
(i) Every open subspace of X is paracompact.

(ii) Every subspace of X is paracompact.
Indeed, suppose that every open subspace is paracompact and let Y X be an arbitrary
subspace. An open covering of Y is of the form .Y \ Ui /i , where .Ui /i 2I is a family of
S
open subsets with Y U WD i Ui . As U is paracompact, there exists a locally finite
open refinement .Vj /j of the covering .Ui /i of U . Then .Y \ Vj /j is a locally finite open
refinement of .Y \ Ui /i .
Proposition 1.13. Every metrizable topological space is hereditarily paracompact.
There exist hereditarily paracompact spaces that are not metrizable (Problem 1.13).
Proof. As every subspace of a metrizable space is again metrizable via the restricted met-
ric, it suffices to show that every metric space .X; d / is paracompact. Let .Ui /i 2I be an
open covering. Choose a well ordering on I (Appendix Proposition 13.27). For n 2 N
define for all i 2 I open subsets Vi;n of X by induction on n 2 N. Let Xi;n be the set of
x 2 X with
S
(a) x 2 Ui n j <i Uj ,
(b) B32n .x/ Ui ,
(c) x … Vj;m for m < n and for all j 2 I .
S
Now define Vi;n WD x2Xi;n B2n .x/. We claim that .Vi;n /i 2I;n2N is a locally finite open
refinement of .Ui /i .
S
Clearly we have Vi;n Ui by (b). To see that i;n Vi;n D X, let x 2 X. Let i 2 I be
the smallest element such that x 2 Ui and let n 2 N with B32n .x/ Ui . If x 2 Xi;n ,
then x 2 Vi;n . Otherwise (c) does not hold, hence there exists j 2 I and m < n such that
x 2 Vj;m .
It remains to show that .Vi;n / is locally finite. Let x 2 X and let i0 2 I be the smallest
element such that x 2 Vi0 ;m for some m 2 N. Choose p 2 N such that B2p .x/ 2 Vi0 ;m .
We claim that B2mp .x/ meets only finitely many of the Vi;n . More precisely, we claim:
(i) If n m C p, then B2mp .x/ \ Vi;n D ; for all i 2 I .

(ii) If n < m C p, then there is at most one i 2 I with B2mp .x/ \ Vi;n ¤ ;.
Let us show (i). Let i 2 I and n m C p. Since n > m, (c) implies that for y 2 Xi;n
one has y … Vi0 ;m . As B2p .x/ 2 Vi0 ;m we find d.x; y/ 2p for all y 2 Xi;n . Hence
1.3 Normal Spaces 7
B2mp \ B2n .y/ D ; for all y 2 Xi;n because n p C 1 and m C p p C 1. This

proves (i).
To show (ii) let n < m C p. Suppose we have y 2 Vi;n and z 2 Vj;n with i < j . It
suffices to show that d.y; z/ > 2mpC1 . By definition of the Vi;n we find yQ 2 Xi;n and
zQ 2 Xj;n with y 2 B2n .y/Q and z 2 B2n .Qz /. We have B32n .y/
Q Ui by (b) and zQ … Ui
by (a). Hence d.y;Q zQ / 3 2n and hence d.y; z/ > 2n 2mpC1 .
Remark 1.14. One can also show that locally compact, second countable Hausdorff
spaces are metrizable (e.g., [Br1] Chap. 1, Theorem 12.12).
1.3 Normal Spaces
Theorem 1.15 (Urysohn’s theorem/Tietze extension theorem). Let X be a topological

space. Then the following assertions are equivalent:
(i) For any two closed subsets A; B X with A \ B D ; there exists a continuous
function f W X ! Œ0; 1 such that f .a/ D 0 for all a 2 A and f .b/ D 1 for all
b 2 B.
(ii) For all disjoint closed sets A and B of X there exist open disjoint sets U and V such
that A U and B V (see Fig. 1.1).
(iii) For every closed subset A and every neighborhood W of A there exists an open
neighborhood U of A such that UN W .
(iv) For every closed subspace A of X and continuous map f W A ! R there exists a con-
tinuous function fQW X ! R such that fQjA D f .
(v) For every closed subspace A of X and continuous map f W A ! Œ1; 1 there exists
a continuous function fQW X ! Œ1; 1 such that fQjA D f .
Figure 1.1 Separation of two

closed subsets A und B
V
A B
U
Of course, one can replace in (i) and in (v) the compact interval by any space home-
omorphic to it, for example any other compact interval in R or the extended real line
R D R [ f˙1g (Appendix Example 12.9).
Proof. We prove
(i) ) (ii) ) (iii) ) (i) ) (v) ) (iv) ) (i).
We start with the easy implications. To see “(i) ) (ii)” take U D f x 2 X I f .x/ < 1=2 g
and V D f x 2 X I f .x/ > 1=2 g. To see “(ii) ) (iii)”, we may assume that W is an
open neighborhood of A in X. Let B WD X n W . Then (ii) implies that there exists an
open neighborhood U of A such that UN \ B D ; and hence UN W .
Next we show “(iv) ) (i)”. Let A and B disjoint closed subsets of X. Then f W A [
B ! R with f .A/ D f0g and f .B/ D f1g is continuous. Hence (iv) implies that f can be
extended to a continuous map fQW X ! R. Then g WD min.max.fQ; 0/; 1/ is a continuous
map on X with values in Œ0; 1, equal to 0 on A and equal to 1 on B.
“(iii) ) (i)”. Let A and B be disjoint closed subsets of X and denote by D Q the
set of dyadic rational numbers in Œ0; 1. First, we will construct open subsets Uq X for
every q 2 D with the following properties:
1. For each q 2 D, A Uq .
2. U1 D X and for each q 2 D with q < 1, B \ Uq D ;.
3. For q; p 2 D with q < p, UN q Up .
Set U1 WD X. Then X n B is a neighborhood of A and by (iii) we find an open subset

U0 X such that A U0 and UN 0 X n B.
We will continue this construction by recursion. Let m 2 N. Assume that Uq has been
constructed for q D a=2n 2 Œ0; 1 for all n < m and a D 0; : : : ; 2n . For p D b=2m 2 Œ0; 1
with 2 − b, set q1 WD ..b C 1/=2/=2m1 2 Œ0; 1 and q2 WD ..b 1/=2/=2m1 2 Œ0; 1 so
that p D .q1 C q2 /=2. Therefore Uq1 and Uq2 have already been constructed and we have
UN q1 Uq2 . By (iii) we find an open subset Up X such that UN q1 Up and UNp Uq2 .
˚
We now define f W X ! Œ0; 1 by f .x/ WD inf q 2 D I x 2 Uq , which is well-
defined because U1 D X. Then f .a/ D 0 for a 2 A because A U0 and f .b/ D 1 for
b 2 B because B U1 and B \ Uq D ; for q 2 D with q < 1. It is left to show that
f is continuous. Notice that the sets Œ0; c/ Œ0; 1 for c 2 .0; 1 and .c; 1 Œ0; 1 for
c 2 Œ0; 1/ generate the topology on Œ0; 1. Hence it suffices to show that the preimage of
sets of this form are open in X.
Let c 2 .0; 1 and x 2 X with f .x/ 2 Œ0; c/. Since D Œ0; 1 is dense, we can choose
q 2 D with f .x/ < q < c. By definition of f there exists p 2 D with p < q and
x 2 Up so that we have x 2 Uq . For y 2 Uq we obviously have f .y/ q < c so that
x 2 Uq f 1 .Œ0; c// is an open neighborhood of x.
Let c 2 Œ0; 1/, x 2 X with f .x/ 2 .c; 1 and choose q 2 D with c < q < f .x/. Then
there exists p 2 D with q < p and x … Up so that x … UN q because UN q Uq . Hence,
1.3 Normal Spaces 9
X n UN q is an open neighborhood of x. Furthermore, for y 2 X n UN q we have y … Up for

p q because Up UN q so that c < q < f .y/ and X n UN q f 1 ..c; 1/.
“(v) ) (iv)”. Suppose first that f 0. By choosing a homeomorphism Œ1; 1 Š
Œ0; 1 R it follows from (v) that we find a continuous extension g1 of f with values
in Œ0; 1. Then B WD g11 .1/ is closed and A \ B D ;. Define a continuous map
hW A [ B ! R0 by hjA WD f and hjB D 0 and let g2 be a continuous extension of h with
values in Œ0; 1. Then fQ WD min.g1 ; g2 / is a continuous extension of f to X with values
in Œ0; 1/.
For a general f extend max.f; 0/ and max.f; 0/ to continuous functions g1 and g2 ,
respectively, on X with values in Œ0; 1/. Then set fQ WD g1 g2 .
“(i) ) (v)”. Let A X be a closed subset and f W A ! Œ1; 1 a continuous map.
We find f 1 .Œ1; 1=3/ A and f 1 .Œ1=3; 1/ A to be disjoint and closed in A
and therefore also closed in X. Due to (i) with target interval Œ1=3; 1=3 instead of
Œ0; 1 there exists a continuous function gW X ! Œ1=3; 1=3 with g.x/ D 1=3 for x
with 1 f .x/ 1=3 and g.x/ D 1=3 for x with 1=3 f .x/ 1. This implies
jf .x/ g.x/j 2=3 for x 2 A. We are now going to construct a sequence .gn /n2N0 of
continuous maps gn W X ! Œ1; 1 with the following properties:
1. jgn .x/j .1=3/.2=3/n for x 2 X.

P
2. j nmD0 gm .x/ f .x/j .2=3/nC1 for x 2 A.
For n D 0 set g0 WD g. Let n 2 N and assume that gm has already been constructed for
Pn1
m < n. Then .3=2/n . mD0 gm f / maps into Œ1; 1 by assumption. Hence, the above
Pn1
construction yields a continuous map g 0 W X ! Œ1=3; 1=3 with j.3=2/n . mD0 gm .x/
f .x// g 0 .x/j 2=3 for x 2 A. Set gn WD .2=3/n g 0 so that jgn .x/j D .2=3/n jg 0 .x/j
P
.1=3/.2=3/n for x 2 X. Using the inductive hypothesis, we have j nmD0 gm .x/f .x/j D
Pn1
j mD0 gm .x/ f .x/ .2=3/n g 0 .x/j .2=3/nC1 for x 2 A.
P
Now jgn j .1=3/.2=3/n for n 2 N0 and .1=3/ 1 m
mD0 .2=3/ is an absolute convergent
P1
series with limit 1. Hence, mD0 gn converges absolutely and uniformly to a continuous
P
map gW X ! Œ1; 1. For x 2 A we have j nmD0 gm .x/f .x/j .2=3/nC1 and therefore
P1
mD0 gm .x/ D f .x/ so that g jA D f .
Definition 1.16 (Normal spaces). A topological space X is called normal if it satisfies

the equivalent conditions of Theorem 1.15.
Let X be a normal space. Every closed subspace of X is normal (use (ii) of Theo-
rem 1.15).
Very often some further separation axioms are included in the definition of “normal”.
We will state these properties always explicitly. For instance we will use the following
property.
Definition 1.17. A topological space X is called T1 -space if fxg is closed in X for all
x 2 X.
Every Hausdorff space is a T1 -space. Conversely, if X is a normal T1 -space, then X is

Hausdorff.
Let .Ui /i be an open covering of a topological space X. As being a closed subset is
a local property (Appendix Proposition 12.32), the space X is a T1 -space if and only if
each Ui is a T1 -space.
Proposition 1.18. Every paracompact T1 -space is normal and Hausdorff.
There exist normal spaces that are not paracompact ([BouGT2] IX, §4, Exercise 27).
Proof. As every normal T1 -space is Hausdorff, it suffices to show that X is normal. Let
X be a paracompact space, A; B X be closed with A \ B D ;.
Claim: Let Y; Z X be closed with Y \ Z D ;. If for all y 2 Y there exists
y 2 Vy X open and Z Wy X open with Vy \ Wy D ;, then there exists an open
neighborhood T of Y and an open neighborhood U of Z with U \ T D ;.
Let us first show that the claim implies that X is normal. As X is a T1 -space, we may
apply the claim to Y D A and Z D fxg for some x 2 B. Hence there exists an open
neighborhood T of A and U of x with T \ U D ;. Then we can apply our claim again to
Z D A and Y D B to see that there exist open neighborhoods UQ of A and VQ of B with
UQ \ VQ D ;. Hence X is normal.
Now let us show the claim. As X is paracompact, we find a locally finite open re-
finement .Ti /i of the open cover of X formed of X n Y and of the Vy for y 2 Y . Let
IY WD f i 2 I I Y \ Ti ¤ ; g. Being a refinement implies that for i 2 IY one has
9 yi 2 Y W Ti Vyi : (*)
S
Let T WD i 2IY Ti . Then T X open and Y T .
It remains to find U . For all z 2 Z there exists z 2 Sz X open such that
Sz intersects only finitely many Ti because .Ti /i is locally finite. In particular Jz WD
T
f i 2 IY I Ti \ Sz ¤ ; g is finite. Set Uz WD Sz \ i 2Jz Wyi . Then Uz is an open neigh-
borhood of z. For i 2 Jz we have Ti Vyi by (*) and hence Wyi \Ti D ;. For i 2 IY nJz
S
one has Sz \ Ti D ;. Hence Uz \ T D ;. Set U WD z2Z Uz .
Proposition 1.19. Every metrizable topological space X is normal.
By Propositions 1.18 and 1.13 there are the following implications:
“metrizable” ) “hereditarily paracompact and Hausdorff” ) “normal and Hausdorff”:
As the proofs were somewhat delicate, we give also a simple direct proof.
1.3 Normal Spaces 11
Proof. Let d be a metric on X inducing its topology. For Y X closed, the map X 2 R,
x 7! d.x; Y / WD infy2Y d.x; y/, is continuous with d.x; Y / D 0 if and only if x 2 Y .
Let A; B X be closed subsets with A \ B D ;. Then
d.x; A/
f W X ! Œ0; 1; f .x/ WD
d.x; A/ C d.x; B/
is a continuous function with f .a/ D 0 for all a 2 A and f .b/ D 1 for all b 2 B.
We conclude this section with two versions of an important technical tool, the shrinking
lemma.
Proposition 1.20 (Shrinking lemma). Let X be a normal space and let .Ui /i 2I be an
open covering such that for all x 2 X there exist only finitely many i with x 2 Ui . Then
there exists an open covering .Vi /i 2I such that Vi Ui for all i.
The hypothesis on .Ui /i is of course satisfied if .Ui /i is locally finite.
Proof. We choose a well ordering on I (Appendix Proposition 13.27) and show by trans-
finite recursion (Appendix Proposition 13.26) that there exists a family .Vi /i 2I of open
S S
subsets such that for all i one has Vi Ui and X D j <i Vj [ j i Uj .
Let i 2 I and assume that we have already constructed Vj for j < i. Set
! !!
[ [
Ai WD Ui n Vj [ Uj :
j <i j >i
Then Ai \ Uj is closed in Uj for all j i and Ai \ Vj is closed in Vj for all j < i.

Hence Ai is closed in X because being closed is a local property (Appendix Proposi-
tion 12.32). As X is normal, there exists an open neighborhood Vi of Ai with Vi Ui
S S
(Theorem 1.15 (iii)). Clearly we have X D j i Vj [ j >i Uj .
S
It remains to show that X D i 2I Vi . By the assumption on .Ui /i we find for all
S S
x 2 X an ix 2 I such that x … Uj for j ix . Now x 2 X D j <ix Vj [ j ix Uj and
S
hence x 2 j <ix Vj .
If X is a paracompact Hausdorff space, we can omit the hypothesis on .Ui /i .
Corollary 1.21 (Shrinking lemma). Let X be a paracompact Hausdorff space and let
.Ui /i 2I be an open covering of X.
1. There exists an open covering .Tj /j 2J such that .Tj /j 2J is a locally finite refinement
of .Ui /i .
2. There exists an open covering .Vi /i 2I such that Vi Ui for all i 2 I .
Proof. As X is paracompact there exists a locally finite open covering .Wj /j 2J and a map
˛W J ! I such that Wj U˛.j / for all j 2 J . As X is normal (Proposition 1.18) we can
apply Proposition 1.20 to .Wj /j . Hence we find an open covering .Tj /j 2J with Tj Wj
for all j . In particular .Tj /j 2J is locally finite again.
S
For all i 2 I set Vi WD j 2˛1 .i / Tj . Then .Vi /i 2I is an open covering of X. As .Tj /j
S
is locally finite, j 2˛1 .i / Tj is closed and hence
[ [
Vi Tj Wj Ui :
j 2˛1 .i / j 2˛1 .i /
1.4 Separated Maps
We now come to a relative version of Hausdorff for continuous maps, so-called “separated
maps”. To characterize separated maps (Proposition 1.25) we first introduce the following
notion for subspaces.
Definition 1.22 (Relatively Hausdorff subspace). Let X be a topological space. A sub-

space A of X is called relatively Hausdorff if any two distinct points in A have disjoint
neighborhoods in X.
If A is a relatively Hausdorff subspace of a topological space X, then A is Hausdorff.

The converse does not necessarily hold (Problem 1.16). If X is Hausdorff, then any sub-
space of X is relatively Hausdorff.
Proposition 1.23. Let A be a relatively Hausdorff subspace of X and let C and C 0 be

compact disjoint subspaces of A. Then there exist disjoint open neighborhoods U of C
and U 0 of C 0 in X.
Proof. Suppose first that C 0 D fx 0 g for some x 0 2 X. As C is compact and A relatively

Hausdorff in X we find a finite open covering .U1 ; : : : ; Un / of C in X and for each i D
1; : : : ; n an open neighborhood Ui0 of x 0 in X such that Ui0 \ Ui D ; for all i. We set
T S
U 0 D i Ui0 and U D i Ui .
Now let C 0 be arbitrarily compact. As we have just seen we find for each point x 0 2 C 0
an open disjoint neighborhoods Ux0 0 of x 0 and Ux 0 of C in X. As C 0 is compact, there
exists a finite subset F C 0 such that .Ux0 0 /x 0 2F is a covering of C 0 . Now we set U WD
T 0
S 0
x 0 2F Ux 0 and U D x 0 2F Ux 0 .
For the definition of separated maps recall that for a continuous map f W X ! Y of
topological spaces the diagonal of f (Appendix Example 12.29 5) is the map
f W X ! X Y X WD f .x; x 0 / 2 X X I f .x/ D f .x 0 / g ; x 7! .x; x/:

1.4 Separated Maps 13
Definition 1.24 (Separated maps). A continuous map f W X ! Y is called separated if

the diagonal map f W X ! X Y X is a closed topological embedding (or, equivalently
by Appendix Example 12.29 5, if f .X/ is closed in X Y X).
Every injective continuous map is separated by Appendix Example 12.29 5. This also
follows from (ii) of the following characterization of separated maps.
Proposition 1.25. For a continuous map f W X ! Y of topological spaces the following

assertions are equivalent:
(i) The map f is separated.

(ii) All fibers of f are relatively Hausdorff subspaces in X.
(iii) For every relatively Hausdorff subspace B of Y the inverse image f 1 .B/ is rela-
tively Hausdorff in X.
Proposition 1.25 (ii) shows in particular that the property “separated” is local on the
target.
Proof. “(i) ) (ii)”. Let y 2 Y and x1 ; x2 2 f 1 .y/ with x1 ¤ x2 . Then .x1 ; x2 /

is in .X Y X/ n f .X/. Because f .X/ X Y X is closed there exists an open
neighborhood Ui of xi in X (i D 1; 2) such that U1 U2 \ X Y X is contained in
.X Y X/ n f .X/. Hence U1 \ U2 D ;.
“(ii) ) (i)”. Let .x1 ; x2 / 2 .X Y X/ n f .X/ and y WD f .x1 / D f .x2 /. As
f 1 .y/ is relatively Hausdorff, there exist neighborhoods U1 of x1 and U2 of x2 such that
U1 \ U2 D ;. Therefore W WD .U1 U2 / \ .X Y X/ is a neighborhood of .x1 ; x2 / in
X Y X and we have W \ f .X/ D ;. Thus .X Y X/ n f .X/ is open in X Y X.
“(ii) ) (iii)”. Let x1 ; x2 2 f 1 .B/ with x1 ¤ x2 and y1 WD f .x1 /; y2 WD f .x2 /. Sup-
pose first that y1 ¤ y2 . Since B Y is relatively Hausdorff, there exist neighborhoods
V1 of y1 and V2 of y2 in Y such that V1 \ V2 D ;. Then f 1 .V1 / and x2 2 f 1 .V2 / are
disjoint neighborhoods of x1 and x2 in X. Now suppose that y1 D y2 . Then x1 and x2 are
points of the same fiber of f . In this case the assertion follows from (ii).
“(iii) ) (ii)”. Clearly every one-point subspace is relatively Hausdorff.
Corollary 1.26. Let f W X ! Y , gW Y ! Z be continuous maps.
1. If f and g are separated, g ı f is separated.

2. If g ı f is separated, f is separated.
Proof. Assertion 1 follows immediately from Proposition 1.25 (iii).

To show 2 let y be a point of Y . Then f 1 .y/ is a subspace of .g ı f /1 .g.y//, which
is relatively Hausdorff in X. Hence f 1 .y/ is relatively Hausdorff in X.
Choosing for Z a point we obtain the following result.
Corollary 1.27. Let f W X ! Y be a continuous map.
1. If Y is Hausdorff and f is separated, then X is Hausdorff.

2. If X is Hausdorff, then f is separated.
1.5 Proper Maps
There is also a relative version for continuous maps of the compactness notion. These
are the proper maps. We first prove a characterization for compact spaces. We start with
a lemma.
Lemma 1.28. Let X and Y be topological spaces, A X and B Y compact subsets

and W X Y open with A B W . Then there exist open neighborhoods U of A in
X and V of B in Y such that U V W .
Proof. For every .a; b/ 2 A B there exist open subsets U.a;b/ X and V.a;b/ Y such
that .a; b/ 2 U.a;b/ V.a;b/ W . Fixing b 2 B yields an open covering .U.a;b/ /a2A of the
compact set A so that there exist a1 ; : : : ; an with A U.a1 ;b/ [ : : : : : : [ U.an ;b/ . We define
open sets Ub WD U.a1 ;b/ [: : :[U.an ;b/ and Vb WD V.a1 ;b/ \: : :\V.an ;b/ and observe A Ub ,
b 2 Vb . Furthermore Ub Vb is contained in W . Now .Vb /b2B is an open covering of
the compact set B so that there exist b1 ; : : : ; bn with B Vb1 [ : : : [ Vbn . For the open
subsets U WD Ub1 \ : : : \ Ubn and V WD Vb1 [ : : : [ Vbn we find A U , B V and
U V W as before.
Proposition 1.29. A topological space X is compact if and only if the projection ZX !
Z is closed for every topological space Z.
Proof. For every topological space Z denote by p1 W Z X ! Z the projection. Suppose

that X is not compact. Then there exists an open covering .Ui /i 2I of X that has no finite
subcovering. In particular I is infinite. Set
Z WD f J I I J finite and non-emptyg [ fI g:
Endow Z with the topology generated by the sets VK WD f J 2 Z I K J g for every

finite subset K I . For finite subsets K; K 0 I we have VK \ VK 0 D VK[K 0 and
V; D Z so that these sets form a basis of the topology of Z (Appendix Remark 12.5).
We will show that p1 W Z X ! Z is not closed. Consider the subset
n [ o
A WD .J; x/ 2 Z X I x 2 X n Ui
i 2J
1.5 Proper Maps 15
of the product space Z X. If .J; x/ … A there exists j 2 J with x 2 Uj so that

S
.J; x/ 2 Vfj g Uj . Also for every .J 0 ; x 0 / 2 Vfj g Uj we have x 0 2 Uj i 2J 0 Ui ,
hence the open subset Vfj g Uj of Z X is contained in .Z X/ n A. Therefore A is
closed.
˚ S S
Now p1 .A/ D J 2 Z I X n i 2J Ui ¤ ; D Z n fI g because i 2J Ui ¤ X for
S
every finite subset J I and i 2I Ui D X by assumption. If Z n fI g was closed, fI g
would be open in Z and therefore an element of the basis of topology given above. But
this is absurd because I is infinite. Hence p1 .A/ is not closed proving that p1 W Z X ! Z
is not closed.
Conversely, assume that X is compact. Let Z be a topological space, let A Z X
be a closed subset, and let z 2 Z n p1 .A/. Then p11 .z/ D fzg X .Z X/ n A and X
and fzg are compact. Therefore we may apply Lemma 1.28 to find an open neighborhood
U of z in Z with U X .Z X/ n A. Then U Z n p1 .A/. Because z was arbitrary
we see that Z n p1 .A/ is open.
We can now formulate the main result of this section and define the notion of a proper
map.
Definition and Theorem 1.30 (Proper maps). A continuous map f W X ! Y between

topological spaces is called proper if it satisfies the following equivalent properties:
(i) For every topological space Z the map idZ f W Z X ! Z Y is closed.

(ii) The map f is closed and for every compact subspace V of Y the preimage f 1 .V /
is a compact subspace of X.
(iii) The map f is closed and f 1 .y/ is compact for all y 2 Y .
(iv) For every continuous map gW Y 0 ! Y , the projection Y 0 Y X ! Y 0 is closed.
Sometimes a continuous map is defined to be proper if inverse images of compact

subspaces are again compact. In general this is strictly weaker than our definition (Prob-
lem 1.21). For locally compact Hausdorff spaces both definitions are equivalent (Prob-
lem 1.20).
Proof. “(i) ) (ii)”. Taking for Z the topological space consisting of a single point shows
that (i) implies that f is closed.
Let V Y be compact and let Z be a topological space. As Z X ! Z Y is
closed, the restriction Z f 1 .V / ! Z V is also closed. Due to Proposition 1.29
the projection Z V ! Z is closed because V is compact and therefore the composition
Z f 1 .V / ! Z V ! Z is closed. This composition is the projection Z f 1 .V / !
Z. As Z was arbitrary, Proposition 1.29 shows that f 1 .V / is compact.
“(ii) ) (iii)”. A topological space with a single point is compact.
“(iii) ) (iv)”. Let gW Y 0 ! Y be a continuous map and f 0 W Y 0 Y X ! Y 0 the
projection. First we note that Im.f 0 / D f y 0 2 Y 0 I g.y 0 / 2 Im.f / g D g 1 .f .X// is
closed in Y 0 because f is closed by assumption. Let A Y 0 Y X be a closed subset and

y 0 2 Y 0 n f 0 .A/. We have to find a neighborhood of y 0 in Y 0 that does not meet f 0 .A/.
Now we have two different cases.
If y 0 … Im.f 0 /, then Y 0 nIm.f 0 / is an open neighborhood of y 0 contained in Y 0 nf 0 .A/.
If y 0 2 Im.f 0 /, we have .f 0 /1 .y 0 / D fy 0 g f 1 .g.y 0 // .Y 0 Y X/ n A. Since
Y Y X Y 0 X is a subspace, there exists A0 Y 0 X closed with A D A0 \
0
.Y 0 Y X/. By hypothesis, fy 0 g f 1 .g.y 0 // is compact and we may apply Lemma 1.28

to fy 0 g f 1 .g.y 0 // .Y 0 X/ n A0 . This yields V 0 Y 0 and U X open with fy 0 g
f 1 .g.y 0 // V 0 U and V 0 U .Y 0 X/ n A0 . Then the set V 0 \ g 1 .Y n f .X n U //
is open in Y 0 since f is closed, it contains y 0 , and has an empty intersection with f 0 .A/.
Thus f 0 .A/ is closed.
“(iv) ) (i)”. Let Z be a topological space and gW Z Y ! Y the projection. Then the
projection .Z Y / Y X ! Z Y is closed by (iv) and it is equal to idZ f W Z X !
Z Y.
Example 1.31. Characterization (iii) shows that an injective continuous map is proper if
and only if it is a closed embedding.
Proposition 1.32. Let f W X ! Y and gW Y ! Z be continuous maps.
1. If f and g are proper, then g ı f is proper.

2. Suppose that g is separated and that g ı f is proper. Then f is proper.
3. Let g ı f be proper and f surjective. Then g is proper.
Proof. Assertion 1 is clear by Theorem 1.30 (ii).

Let us show 2. By Appendix Example 12.29 4, f W X 7! X Z Y , x 7! .x; f .x// is an
embedding. Its image is the inverse image of the diagonal Y Z Y under f idY , and the
diagonal is closed in Y Z Y because g is separated. Hence f is a closed embedding and
hence proper (Example 1.31). Now f is the composition of f followed by the projection
X Z Y ! Y , which is proper by Theorem 1.30 (iv) because g ı f is proper. Hence f is
proper by 2.
To show 3 let z 2 Z. As f is surjective, it induces a surjective map f 1 .g 1 .z// !
g .z/. As g ı f is proper, f 1 .g 1 .z// is compact. Hence g 1 .z/ is compact (Appendix
1
Proposition 12.51). Now let B Y be closed. As f is surjective, we have g.B/ D

g.f .f 1 .B///, which is closed because g ı f is closed. Therefore g is closed.
We obtain a new proof of Appendix Proposition 12.52.
Corollary 1.33. Let f W X ! Y be a continuous map. Suppose that X is compact and

that Y is Hausdorff. Then f is proper and the subspace f .X/ of Y is closed in Y and
compact.
1.6 Problems 17
Proof. Applying Proposition 1.32 2 to the special case that Z consists of a single point
shows that f is proper. Hence f .X/ is closed in Y . Being the image of a compact space,
f .X/ is compact (Appendix Proposition 12.51 2).
Proposition 1.34. Let .fi W Xi ! Yi /i 2I be a finite family of proper continuous maps.

Then the product map .xi /i 7! .fi .xi //i is proper.
One can show that the product of an arbitrary family of proper maps is again proper
([BouGT1], I, §10.2, Corollary 3 of Theorem 1).
Proof. By induction we may assume that I D f1; 2g. Let Z be a topological space. Then
f1 f2 idZ is the composition of the proper maps f1 idX1 idZ and idY2 f1 idZ .
Hence it is proper by Proposition 1.32.
1.6 Problems
Problem 1.1. Let X be a Lindelöf space and let f W X ! Y be a surjective continuous

map. Show that Y is again a Lindelöf space.
Problem 1.2. Show that the product of a Lindelöf space and a compact space is again
a Lindelöf space.
Problem 1.3. Let X be a topological space.
1. Suppose that X is a Lindelöf space. Show that every closed subspace of X is again
a Lindelöf space.
2. Show that every subspace of X is Lindelöf if and only if every open subspace of X is
Lindelöf. Such spaces are called hereditarily Lindelöf spaces.
Problem 1.4. Let X be a metrizable space. Show that the following assertions are equiv-
alent:
(i) X is second countable.

(ii) X is a Lindelöf space.
(iii) X is separable.
Problem 1.5. A topological space X is called regular if X is a T1 -space and if for all
x 2 X and closed subsets A X with x … A there exist disjoint open neighborhoods of
x and of A.
1. Show that every Hausdorff normal space is regular.

2. Show that a T1 -space X is regular if and only if the closed neighborhoods of any x 2 X
form a fundamental system of neighborhoods of x.
3. Show that every subspace of a regular space is again regular.
4. Show that in a regular Lindelöf space X every open covering has a countable locally
finite open refinement. In particular X is paracompact.
Problem 1.6. Show that every locally compact Hausdorff space is regular (Problem 1.5).
Problem 1.7. Show that every compact Hausdorff space is normal.
Problem 1.8. Let X be a locally compact Hausdorff space. Show that X is paracompact
if and only if X is the sum of a family of -compact spaces.
Problem 1.9. Let X be a topological space. A subset A of X is called meager if it is

a union of countably many nowhere dense subsets (Appendix Problem 12.35). Show that
the following assertions are equivalent:
(i) Every intersection of countably many open dense subsets of X is dense in X.

(ii) Every countable union of closed subsets without interior points has no interior point
in X.
(iii) The complement of every meager subset is dense in X.
If X satisfies these conditions, then X is called a Baire space.
Problem 1.10. Show that every locally compact Hausdorff space is a Baire space (Prob-
lem 1.9).
Hint: Let .Un /n2N be a family of open dense subsets, V X open. Construct by induc-
tion open subsets Vn with V1 D V , VnC1 Vn \ Un , and Vn compact for n 2 (use
T T
Problem 1.6). Show that V \ n Un n Vn ¤ ;.
Problem 1.11. Let X ¤ ; be a locally compact Hausdorff space and let .An /n be a count-
S
able family of closed sets such that X D n An . Show that there exists an n 2 N such
that Aın ¤ ;.
Hint: Problem 1.10.
Problem 1.12. Let X be a paracompact space and let Y be a compact space. Show that
X Y is paracompact.
Remark: The product of two paracompact spaces is in general not paracompact (Prob-
lem 1.13).
1.6 Problems 19
Problem 1.13. Let X be the topological space whose underlying set is the set R of real
numbers and whose topology is generated by B WD f .a; b I a; b 2 R; a < b g.
1. Show that B is a basis of the topology, that the closed intervals are closed in this topol-
ogy, and that the topology is finer then the left order topology (Appendix Problem 13.3)
on R.
2. Show that every open subset of X is a countable union of intervals of the form .a; b
and .a; b/ that are pairwise disjoint. Deduce that X is normal and Hausdorff.
3. Show that X is hereditarily Lindelöf (Problem 1.3). Deduce that X is hereditarily
paracompact (use Problem 1.5).
4. Show that X X is not normal.
5. Deduce that X X is not paracompact and that X is not metrizable.
6. Show that every compact subspace of X is countable. Deduce that X is not -compact.
Problem 1.14. Show that every Hausdorff locally compact topological group is paracom-
pact.
Problem 1.15. Let .X; / be a totally ordered set and endow X with the order topology
(Appendix Problem 12.24). Show that X is normal.
Problem 1.16. Let X be a set endowed with the particular point topology with respect to
some p 2 X (Appendix Problem 12.7). Show that any subspace A of X n fpg is discrete
(in particular Hausdorff) but that A is not relatively Hausdorff in X if A consists of more
than one point.
Problem 1.17. Let f W X ! Y , gW Y ! Z be continuous maps. Suppose that g ı f is

proper.
1. Let f be surjective. Show that g is proper.

2. Let g be injective. Show that f is proper.
Problem 1.18. Let f W X ! Y be a continuous map. Let .Bi /i be a family of subsets

of Y and suppose: (a) that the interiors cover Y or (b) that .Bi /i is a locally finite closed
covering. Show that f is proper if and only if its restriction f 1 .Ai / ! Ai is proper for
all i.
Hint: Use Appendix Problem 12.13.
Problem 1.19. Let f W X ! Y be a surjective map of topological spaces such that X

carries the inverse image topology of Y . Show that f is proper and open.
Problem 1.20. Let X and Y be Hausdorff spaces and suppose that Y is locally compact.
Show that a continuous map f W X ! Y is proper if and only if for every compact subspace
C of Y the inverse image f 1 .C / is compact.
Hint: Problem 1.18.
Problem 1.21. Let .Xi /i 2I be a non-countable family of discrete topological spaces each
Q
having at least two points. Endow that product set i 2I Xi with the topology generated
Q
by the subsets i 2I Mi with Mi Xi for all i and Mi D Xi for all but countably many
i 2 I.
1. Show that in X every countable intersection of open subsets is again open. Deduce
that no point of X has a countable fundamental system of neighborhoods.
2. Show that X is regular and Hausdorff.
3. Show that any compact subspace of X is finite.
4. Let X 0 be the set X endowed with the discrete topology and let f W X 0 ! X be the
identity. Show that f is continuous and that f 1 .C / is compact for every compact
subspace C of X. Show that f is not proper.
Algebraic Topological Preliminaries
2
In this chapter we briefly introduce some elementary notions and results on homotopy,
fundamental groups, and covering spaces that are used throughout the book.
2.1 Homotopy
We start with the notion of homotopy that is a central notion in topology.
Definition 2.1 (Homotopy). Two continuous maps f; gW X ! Y of topological spaces

are called homotopic if there exists a continuous map H W X Œ0; 1 ! Y such that
H.x; 0/ D f .x/ and H.x; 1/ D g.x/ for all x 2 X. Then H is called a homotopy
between f and g. We then write f ' g or H W f ' g.
One should think of H W X Œ0; 1 ! Y as a continuously varying family of maps

H t W X ! Y , x 7! H.x; t/ parametrized by t 2 Œ0; 1 interpolating between f D H0 and
g D H1 .
Remark and Definition 2.2. Let X and Y be topological spaces. Then the homotopy
relation ' is an equivalence relation on the set of all continuous maps X ! Y :
1. Clearly, one has f ' f via the homotopy H W X Œ0; 1 ! Y , .x; t/ 7! f .x/.
2. Given a homotopy H W f ' g, then H W g ' f via the inverse homotopy H W .x; t/ 7!
H.x; 1 t/.
3. Let H W f ' g and KW g ' h be given. Then H KW f ' h via the product homotopy
8
<H.x; 2t/; 0 t 1=2I
.H K/.x; t/ WD
:K.x; 2t 1/; 1=2 t 1:

DOI 10.1007/978-3-658-10633-1_2
22 2 Algebraic Topological Preliminaries
The equivalence class of f W X ! Y is denoted by Œf and called the homotopy class of

f . We denote by ŒX; Y the set of homotopy classes of continuous maps X ! Y .
A continuous map X ! Y is called null homotopic if it is homotopic to a constant
map.
Remark and Definition 2.3 (Homotopy category). We define the homotopy category
(h-Top) as follows:
(a) Objects are topological spaces.

(b) For two objects X and Y we define Hom(h-Top) .X; Y / WD ŒX; Y , the set of homotopy
classes of continuous maps X ! Y .
(c) The identity of a topological space in (h-Top) is the homotopy class of idX .
(d) For Œf 2 ŒX; Y and Œg 2 ŒY; Z we define the composition by Œg ı Œf WD Œg ı f .
This is well defined. Indeed, let f; f 0 W X ! Y and g; g 0 W Y ! Z be continuous
maps, and let H W f ' f 0 and KW g ' g 0 be homotopies. Then g ı f ' g 0 ı f 0 via
the homotopy .x; t/ 7! K.H.x; t/; t/.
A continuous map f W X ! Y whose homotopy class is an isomorphism in (h-Top) is

called a homotopy equivalence. This means that there exists a continuous map gW Y ! X
such that g ı f ' idX and f ı g ' idY . Topological spaces X and Y are called homotopy
equivalent if they are isomorphic in (h-Top).
Example 2.4. Let n 1 be an integer and let jj jj be the Euclidean norm on Rn . Then the
.n 1/-dimensional sphere
S n1 WD f x 2 Rn I jjxjj D 1 g
is homotopy equivalent to Rn n f0g: The inclusion iW S n1 ! Rn n f0g and the map
pW Rn n f0g ! S n1 , x 7! jjxjj
x
are mutually inverse homotopy equivalences. Indeed
p ı i D idS n1 and i ı p ' idRn nf0g via the homotopy
.Rn n f0g/ Œ0; 1 ! Rn n f0g; .x; t/ 7! .1 t/x=jjxjj C tx:
This is in particular an example of homotopy equivalent spaces that are not homeomorphic
(S n1 is compact, Rn n f0g not).
Definition 2.5 ((Locally) contractible spaces). A topological space X is called con-

tractible, if it is homotopy equivalent to a point (i.e., to the unique topological space
consisting of a single point). It is called locally contractible if every point has a funda-
mental system of open contractible neighborhoods.
In other words, X is contractible if and only if there exists a point x0 2 X such that the
constant map X ! X, x 7! x0 and idX are homotopy equivalent, i.e., if and only if idX is
null homotopic.
2.2 Paths 23
Example 2.6. Let V be a normed real or complex vector space.
1. Recall that a subset S of V is called star-shaped with star center s0 2 S if for all
s 2 S the line segment f ts C .1 t/s0 I 0 t 1 g from s to s0 is contained in S.
Then the subspace S of V is contractible: H W .s; t/ 7! ts C .1 t/s0 is a homotopy
between the constant map s 7! s0 and idS .
2. Every open ball in V is star-shaped. As the open balls form a basis of the topology on
V , we see that every open subspace of V is locally contractible.
2.2 Paths
We quickly introduce some notation for paths. In this section, let a; b 2 R with a < b.
Definition 2.7 (Paths). Let X be a topological space.
1. A continuous map W Œa; b ! X is called path in X. The point .a/ is called the start
point, .b/ is called the end point of . We say that is a path from .a/ to .b/. We
also set:
fg WD .Œa; b/ X:
2. A path W Œa; b ! X is called closed or a loop if .a/ D .b/.
Let W Œa; b ! X be a map. Let 'W Œ0; 1 ! Œa; b, '.t/ D a C .b a/t. Then is
continuous if and only if ı ' is continuous and fg D f ı 'g. This allows us usually to
assume that Œa; b D Œ0; 1.
Definition 2.8. Let X be a topological space.
1. Let W Œ0; 1 ! X be a path. Define
W Œ0; 1 ! X; .t/ D .1 t/
the inverse path.

2. Let ; ıW Œ0; 1 ! X be two paths with .1/ D ı.0/. Define the product or the con-
catenation of paths by
8
<.2t/; 0 t 1=2I
ıW Œ0; 1 ! X; t 7!
:ı.2t 1/; 1=2 t 1;
i.e., ı is the path where one first “walks along and then along ı each time with
double velocity”.
2.3 Path Connected Spaces
Let X be a topological space. We write x y if there exists a path with start point x
and end point y. Then is an equivalence relation on X as shown by the construction in
Definition 2.8.
Definition 2.9. An equivalence class with respect to is called a path component of

X. The set of path components of X is denoted by 0 .X/. The space X is called path
connected if there exists a single path component, i.e., 0 .X/ consists of a single point.
A topological space X is called locally path connected if every point of X has funda-
mental system of path connected neighborhoods.
The empty space is not path connected by definition.
Remark 2.10. Let f W X ! Y be a surjective continuous map. If X is path connected,

then Y is path connected.
Indeed, let y; y 0 2 Y . Choose x; x 0 2 X with f .x/ D y and f .x 0 / D y 0 and a path
in X connecting x and x 0 . Then f ı is a path connecting y and y 0 .
Example 2.11.
1. A discrete space X is locally path connected, but it is not path connected if it contains
more than one point.
2. Conversely, there are path connected topological spaces that are not locally path con-
nected (Problem 2.7).
3. Every open subspace of of a normed R-vector space is locally path connected.
Proposition 2.12. Every path connected topological space X is connected.
Proof. Assume that X is not connected. Then X is the disjoint union of two non-empty
open subsets U and V . Choose x 2 U and y 2 V and let W Œ0; 1 ! X be a path with
.0/ D x and .1/ D y. Then Œ0; 1 is the disjoint union of the open non-empty sub-
sets 1 .U / and 1 .V /. This is a contradiction, because Œ0; 1 is connected (Appendix
Proposition 12.43).
In general there exist connected spaces that are not path connected (Problems 2.6
and 2.7). But for locally path connected spaces this cannot happen. More precisely we
have:
Proposition 2.13. Let X be a topological space such that every point has a path con-
nected neighborhood. Then a subset of X is a path component if and only if it is a con-
nected component. All path components are open and closed in X, i.e., X is the sum of its
path components.
2.4 Fundamental Group 25
Proof. By Proposition 2.12 every path component A of X is connected (and hence con-
tained in a connected component). Let x 2 X and let A be the path component of X
containing x. By hypothesis there exists a path connected neighborhood V of x. Then
A [ V is path connected and hence contained in A. This shows that A is open in X. As
the complement of A is a union of path components, A is also closed in X.
Proposition 2.12 shows that every connected component B of X is a union of path
components. Each of theses path components is open and closed in X and hence in B. As
B is connected, B can only contain a single path component.
Remark 2.14. Suppose that X is a locally path connected space. Then Proposition 2.13
holds also for every open subspace of X. This shows in particular that every point of X
has a fundamental system of open path connected neighborhoods.
Remark 2.15. Let f W X ! Y be a continuous map. If x; x 0 2 X are connected by

some path , then f .x/ and f .x 0 / are connected by the path f ı . Hence f induces
a map 0 .f /W 0 .X/ ! 0 .Y /. We obtain a functor 0 W (Top) ! (Sets). Moreover, if
f; gW X ! Y are homotopic continuous maps, then 0 .f / D 0 .g/ and we even obtain
a functor
0 W (h-Top) ! (Sets):
Hence every homotopy equivalence f W X ! Y induces a bijection 0 .X/ ! 0 .Y /. In
particular, we see that every (locally) contractible space is (locally) path connected.
2.4 Fundamental Group
Definition 2.16. Let X and Y be topological spaces, let A X be a subspace, and let
f; gW X ! Y be continuous maps. A homotopy H W f ' g is said to be relative to A if
H.a; t/ D f .a/ D g.a/ for all a 2 A, t 2 Œ0; 1. We write H W f ' g .rel A/ in this case.
The existence of such a homotopy implies in particular that f jA D g jA . The same

constructions as in Remark 2.2 show that “homotopy relative to A” is an equivalence
relation on the set of all continuous maps X ! Y .
Definition 2.17. Let X be a topological space and let ; ıW Œ0; 1 ! X be two paths with
the same start and end points. A homotopy of paths between and ı is a homotopy
H W ' ı .rel f0; 1g/, i.e., H is a homotopy that leaves the start point and the end point
fixed.
Remark 2.18. Let ; 1 ; 2 ; 3 ; ı1 ; ı2 W Œ0; 1 ! X be paths in X.
1. Let 'W Œ0; 1 ! Œ0; 1 be continuous with '.0/ D 0 and '.1/ D 1. Then ' ı '
.rel f0; 1g/.
2. Set x0 WD .0/, x1 WD .1/, and let "i be the constant path t 7! xi . Then
"0 ' ' "1 .rel f0; 1g/:
3. Assume i .1/ D ıi .0/ for i D 1; 2. Then 1 ' 2 .rel f0; 1g/ and ı1 ' ı2 .rel f0; 1g/
imply that 1 ı1 ' 2 ı2 .rel f0; 1g/.
4. If one has 1 ' 2 .rel f0; 1g/, then 1 2 is null-homotopic.
5. Assume 1 .1/ D 2 .0/ and 2 .1/ D 3 .0/. Then
1 .2 3 / ' .1 2 / 3 .rel f0; 1g/:

Proof. 1. A homotopy ı ' ' is given by .s; t/ 7! .ts C .1 t/'.s//.
2.,5. One has "0 D ı ' with '.t/ D 0 for 0 t 12 and '.t/ D 2t 1 for
1
2
t 1. Hence "0 ' .rel f0; 1g/ by 1. Similarly one shows ' "1 .rel f0; 1g/
and 5.
3. If H W 1 ' 2 .rel f0; 1g/ and GW ı1 ' ı2 .rel f0; 1g/, then 1 ı1 ' 2 ı2
.rel f0; 1g/. via the homotopy .s; t/ 7! H.2s; t/ for 0 s 12 and .s; t/ 7! G.2s 1; t/
for 12 s 1.
4. One has 1 ' 2 .rel f0; 1g/ if and only if 1 ' 2 .rel f0; 1g/. Hence we
may assume that 1 D 2 by 3. Then .s; t/ 7! 1 .2s.1 t// for 0 s 12 and
.s; t/ 7! 1 .2.1 s/.1 t// for 12 s 1 is a homotopy between and the constant
path with value .0/.
Definition 2.19. Let X be a topological space, x 2 X. Define the fundamental group of

X with base point x by
1 .X; x/ WD f W Œ0; 1 ! X I path with .0/ D .1/ D xg =.' .rel f0; 1g/;
i.e., 1 .X; x/ is the set of homotopy classes Œ of closed paths starting (and ending) in
x. Define a multiplication on 1 .X; x/ by
ŒŒı WD Œ ı:
By Remark 2.18 this is well defined and yields a group structure on 1 .X; x/. The neutral
element is the constant path with value x and the inverse of Œ 2 1 .X; x/ is Œ .
In the sequel we will often write instead of Œ for elements in 1 .X; x/.
Remark 2.20. Let (Toppt) be the category whose objects are pointed topological spaces,
i.e., pairs .X; x/ consisting of a topological space X and a point x 2 X. Morphisms
.X; x/ ! .Y; y/ are continuous maps f W X ! Y with f .x/ D y. Composition in
(Toppt) is given by composition of maps.
If f W .X; x/ ! .Y; y/ is a morphism in (Toppt), then 7! f ı defines a group
homomorphism 1 .f /W 1 .X; x/ ! 1 .Y; y/. It is easy to check that we obtain a functor
1 W (Toppt) ! (Grp):
2.4 Fundamental Group 27
Remark 2.21. Let X be a topological space, x0 ; x1 2 X be points and let be a path in

X with start point x0 and end point x1 . Then
W 1 .X; x0 / ! 1 .X; x1 /; 7!
is an isomorphism of groups: Remark 2.18 shows that is well defined and that for
; 0 2 1 .X; x0 / one has
./ . 0 / D 0 ' 0 ' 0 D . 0 /:
Hence is a group homomorphism. An inverse is given by W ı 7! ı .

In particular we see that for a path connected space X the fundamental group 1 .X; x0 /
is up to isomorphism independent of the chosen point x0 2 X.
Definition 2.22. A topological space X is called simply connected if X is path connected

and if 1 .X; x/ D 1 for all (equivalently, for one) x 2 X.
In other words, X is simply connected if and only if every closed path in X is homo-
topic relative f0; 1g to a given constant path.
Remark 2.23. Let X and Y be topological spaces, x0 2 X, y0 2 Y , let p and q be the

projections from X Y to X and Y , respectively. Then
.1 .p/; 1 .q//W 1 .X Y; .x0 ; y0 // ! 1 .X; x0 / 1 .Y; y0 /
is an isomorphism of groups. An inverse is given by 1 .i/1 .j /, where iW X ! X Y ,

x 7! .x; y0 / and j W Y ! X Y , y 7! .x0 ; y/.
We conclude this section by showing that homotopic maps induce isomorphic maps on
fundamental groups.
Proposition 2.24. Let f0 ; f1 W X ! Y be continuous maps and let H W f0 ' f1 be a ho-

motopy. Let x 2 X and define yi WD fi .x/ for i D 0; 1. Then .t/ WD H.x; t/ is a path
from y0 to y1 and the following diagram commutes:
1 .Y; y0 /
1 .f0 /
1 .X; x/ Š
1 .f1 /
1 .Y; y1 / :
Proof. Let Œ 2 1 .X; x/. We have to show that
.f0 ı / ' .f1 ı / .rel f0; 1g/:
Let hW Œ0; 1 Œ0; 1 ! Y , .s; t/ 7! H..s/; t/. Let a; b; c; d W Œ0; 1 ! Œ0; 1 Œ0; 1,
a.t/ D .t; 0/, b.t/ D .1; t/, c.t/ D .0; t/, d.t/ D .t; 1/ be the sides of the square. Then
.f0 ı/ D hı.ab/ and .f1 ı/ D hı.c d /. Clearly there is a homotopy ab ' c d
in Œ0; 1Œ0; 1 relative f0; 1g (e.g., a linear homotopy .s; t/ 7! .1t/.ab/.s/Ct.cd /.s/)
and this homotopy induces the desired homotopy by composition.
Corollary 2.25. Let f W X ! Y be a homotopy equivalence.
1. Let x 2 X. Then 1 .f /W 1 .X; x/ ! 1 .Y; f .x// is a group isomorphism.

2. The topological space X is simply connected if and only if Y is simply connected.
Proof. Let gW Y ! X be a homotopy inverse map of f . As g ı f (respectively f ı g)

is homotopic to the identity, Proposition 2.24 implies that 1 .f / is injective (respectively
surjective). This shows (1).
Assertion (2) follows from (1) and because X is path connected if and only if Y is path
connected (Remark 2.15).
Corollary 2.26. Every contractible space is simply connected.
Altogether we obtain the following implications for various notions of connectivity

(where the first two notions have been only defined for subspaces of normed vector
spaces):
“convex” ) “star shaped” ) “contractible” ) “simply connected”

) “path connected” ) “connected”:
Each
˚ of these implications
is a proper implication: The union ˚ of the coordinate axes
.x; y/ 2 R2 I xy D 0 is star shaped but not convex, R2 n .x; y/ I y D x 2 ; x 0
is not star shaped but contractible (Problem 2.8), Rn n f0g is simply connected for n 3
(Problem 2.15) but not contractible (see Corollary 11.21 below), R2 nf0g is path connected
(Problem 2.5) but not simply connected (see Corollary 2.44 below), and Problems 2.6
and 2.7 give examples of connected spaces that are not path connected.
2.5 Covering Spaces 29
2.5 Covering Spaces
Definition 2.27 (Covering space). Let X be a topological space.
1. A continuous map pW XQ ! X is called a covering space of X or a covering map if for

all x 2 X there exists x 2 U X open such that
a
p 1 .U / D UQ i sum of topological spaces, I ¤ ; some set (2.1)
i 2I
for open sets UQ i X such that p j UQi W UQ i ! U is a homeomorphism for all i 2 I (see
Fig. 2.1).
2. A covering map pW XQ ! X is called a universal covering map if XQ is simply con-
nected.
3. Define the category (CovSp(X)) of covering spaces of X as follows. Objects are cov-
ering spaces pW XQ ! X. A morphism between two covering spaces p1 W XQ 1 ! X and
p2 W XQ 2 ! X is a continuous map ˛W XQ 1 ! XQ2 such that p2 ı ˛ D p1 .
For a justification of the notion of a “universal covering” see Remark 2.38 below.
A covering map p is always a surjective map: For every U as in (2.1) and for every
y 2 U we have a bijection p 1 .fyg/ $ I and I ¤ ; by definition. The cardinality of the
set p 1 .x/ is locally constant in x. If it is constant (e.g., if X is connected), we call the
cardinality of a fiber the degree of the covering p.
Sometimes we consider pointed covering spaces pW .X; Q x/Q ! .X; x/ of pointed topo-
logical spaces. By this we mean a covering map pW XQ ! X with p.x/ Q D x. One has the
obvious notion of a morphism of pointed covering spaces.
Remark 2.28. Let X be a topological space. Let E ¤ ; be a set considered as a discrete

topological space. Then the projection X E ! X is a covering map. Covering maps
that are isomorphic to a covering map of this form are called trivial covering maps.
A continuous map pW XQ ! X is a covering map if and only if it is locally on X a trivial
covering map.
Figure 2.1 Triviality of the

~
covering over U Ui
x U X
Example 2.29.
1. The function pW R ! S 1 D f z 2 C I jzj D 1 g, x 7! e 2ix is a universal covering.

Indeed, R is simply connected because it is convex. The map is a covering: For j D
`
1; 1 let Uj D S 1 nfj g. Then U1 [U1 D S 1 . One has p 1 .U1 / D n2Z .n; nC1/ and
`
p j.n;nC1/ is a homeomorphism .n; n C 1/ ! U1 ; similarly for p 1 .U1 / D n2Z .n
1
2
; n C 12 /.
2. The function expW C ! C is a universal covering (use (1) and the polar decomposi-
tion of complex numbers).
3. For n 2, the map f W C ! C, z 7! z n is surjective, but there exists no open
neighborhood U of 0 such that #f 1 .fz0 g/ D #f 1 .f0g/ D 1 for all z0 2 U . Hence
f is not a covering. However its restriction to C n f0g is a covering of degree n
(Problem 2.17).
Remark 2.30. The fibers of a covering map pW XQ ! X are relatively Hausdorff, hence
every covering map is separated (Proposition 1.25). In particular, if X is Hausdorff, then
XQ is Hausdorff (Corollary 1.27).
Let pW XQ ! X and let f W Z ! X be a continuous map. A continuous map fQW Z ! XQ

such that p ı fQ D f is called a lifting of f along p.
Proposition 2.31 (Uniqueness of liftings). For i D 1; 2 let fQi W Z ! XQ be liftings of

f along a covering map pW XQ ! X. Suppose that Z is connected and that there exists
z0 2 Z with fQ1 .z0 / D fQ2 .z0 /. Then fQ1 D fQ2 .
Proof. As p ı fQi D f , .fQ1 ; fQ2 / yields a map fQW Z ! XQ X X.

Q Let p XQ X XQ
be the diagonal. As p is a separated by Remark 2.30 and a local homeomorphism, p
is closed andn open in XQ X XQ (Proposition
o 1.25 and Appendix Remark 12.37). Hence
Q 1 Q Q
f .p / D z 2 Z I f1 .z/ D f2 .z/ is closed and open in Z. It contains z0 and hence
is equal to Z because Z is connected.
The proof shows that the same assertion remains true for separated local homeomor-
phisms p.
Proposition 2.32 (Lifting homotopies). Let pW XQ ! X be a covering map. Let Z

be a topological space, let H W Z Œ0; 1 ! X be a homotopy and let fQW Z ! XQ be
a continuous map such that p.fQ.z// D H.z; 0/. Then there exists a unique homotopy
HQ W Z Œ0; 1 ! XQ such that p ı HQ D H and such that fQ.z/ D HQ .z; 0/ for all z 2 Z.
The proof will show that if H is a homotopy relative to a subset A Z, then HQ is also
a homotopy relative to A.
Proof. If HQ exists, it is uniquely determined (apply Proposition 2.31 to Hz .t/ WD H.z; t/

for all z 2 Z). Hence it suffices to construct HQ locally, i.e., to show that for all z 2 Z
there exists an open neighborhood W of z in Z and HQ W W Œ0; 1 ! XQ such that p ı HQ D
H jW Œ0;1 and such that fQ.z/ D HQ .z; 0/ for all z 2 W .
Let .U˛ /˛ be an open covering of X such that for all ˛, p 1 .U˛ / is the sum of open
subspaces each of which is mapped homeomorphically to U˛ . As H is continuous, every
point .z; t/ has a neighborhood of the form W t .a t ; b t / in Z Œ0; 1 such that H.W t
.a t ; b t // U˛ . As Œ0; 1 is compact, we find 0 D t0 < t1 < tr D 1 and a neighborhood
W of z such that for all i D 1; : : : ; r, H.W Œti 1 ; ti / is contained in some U˛i .
By induction we may assume that HQ jW Œ0;ti 1 has already been constructed. As p
is a covering map, there exists an open subspace UQ ˛i XQ containing HQ .z; ti 1 / such
that pi WD p j UQ ˛ W UQ ˛i ! U˛i is a homeomorphism. After shrinking W we may assume
i
that HQ .W fti 1 g/ UQ ˛i . Now we can extend HQ to W Œ0; ti by HQ jW Œti 1 ;ti WD
pi1 ı H jW Œti 1 ;ti .
Taking Z to be a point yields the path lifting property for covering spaces.
Corollary 2.33 (Lifting of paths). Let pW XQ ! X be a covering map. Let W Œ0; 1 ! X

be a path and let xQ 0 2 p 1 ..0//. Then there exists a unique path Q W Œ0; 1 ! XQ with
p ı Q D and .0/
Q D xQ 0 .
Corollary 2.34. Let pW XQ ! X be a covering map. For every topological space Z,

composition with p yields an injective map
Q ,! Hom(h-Top) .Z; X/:

Hom(h-Top) .Z; X/
In other words, a covering map is an epimorphism in the category (h-Top). The image
Q ! Hom(h-Top) .Z; X/ can be described in terms of the fundamental
of Hom(h-Top) .Z; X/
group.
Proposition 2.35. Let pW XQ ! X be a covering map, xQ 0 2 X, Q x0 WD p.xQ 0 /. Let

f W .Z; z0 / ! .X; x0 / be a morphism of pointed topological spaces, where Z is path
connected and locally path connected.
Then there exists a morphism fQW .Z; z0 / ! .XQ ; xQ 0 / of pointed topological spaces with
p ı fQ D f if and only if 1 .f /.1 .Z; z0 // 1 .p/.1 .X; Q xQ 0 //.
The morphism fQ is unique by Proposition 2.31 if it exists.
Proof. If there exists a morphism fQW .Z; z0 / ! .XQ ; xQ 0 / of pointed topological spaces
with pı fQ D f , then 1 .f / D .1 .p/ı1 .fQ// and therefore clearly 1 .f /.1 .Z; z0 //
Q xQ 0 //.
1 .p/.1 .X;
Now assume 1 .f /.1 .Z; z0 // 1 .p/.1 .X; Q xQ 0 //. Let z 2 Z and let be a path
from z0 to z in Z, which exists because Z is path connected. Then f ı becomes a path
from x0 to f .z/ in X so that Corollary 2.33 yields a unique path Q in XQ starting at xQ 0 such
that p ı Q D f ı . Set fQ.z/ D .1/ Q so that .p ı fQ/.z/ D .p ı Q /.1/ D .f ı /.1/ D
f ..1// D f .z/. We will show that this is well defined (independent of the choice of )
and defines a continuous map.
Let ı be another path from z0 to z in Z and let ıQ be the unique path in XQ starting at
xQ 0 such that p ı ıQ D f ı ı. Furthermore set WD ı , which is a closed path in Z
starting at z0 , and let Q be the unique path in XQ starting at xQ 0 such that p ı Q D f ı .
By assumption we have 1 .f /.Œ / 2 1 .p/.1 .X; Q xQ 0 // so that there exists a closed path
0 Q 0
Q in X starting at xQ 0 with 1 .p/.ŒQ / D 1 .f /.Œ /. Therefore there exists a homotopy
of paths H W .p ı Q 0 / ' .f ı / .rel f0; 1g/ and applying Proposition 2.32 yields a unique
homotopy of paths HQ W Q 00 ' Q .rel f0; 1g/ for a path Q 00 in XQ such that p ı HQ D H .
This implies p ı Q 00 D p ı Q 0 and Q 00 .0/ D .0/
Q D xQ 0 D Q 0 .0/ so that Q 00 D Q 0 by the
uniqueness of lifts. In particular Q has to be a closed path and due to the uniqueness of lifts
Q
we have Q D Q ıQ and therefore ı.1/ D .1/.
Q Hence, fQW Z ! XQ is a well-defined map.
Q
Now let z 2 Z, xQ WD f .z/ and x WD p.x/ Q D f .z/. By definition there exist x 2 U
X open and xQ 2 UQ XQ such that p j UQ W UQ ! U is a homeomorphism. Since Z is locally
path connected, there exists z 2 V f 1 .U / open and connected. It is sufficient to show
that fQ.V / UQ so that fQjV D .p j UQ /1 ı f jV has to be continuous. Let v 2 V , let be
a path from z to v in V and ı a path from z0 to z in Z so that ı becomes a path from z0 to
v. Furthermore let ıQ be a path from xQ 0 to xQ with p ı ıQ D f ıı and set Q WD .p j UQ /1 ıf ı.
Then .ı Q /.0/
Q Q /
D xQ 0 and pı.ı Q D .pıı/.pıQ Q D .f ıı/.f ı/ D f ı.ı/ so that ı
/ Q Q
Q Q
lifts f ı.ı/. Therefore f .v/ D .ı /.1/ Q D ..p j UQ / ıf ı/.1/ D .p j UQ / .f .v// 2 UQ .
1 1
Corollary 2.36. The map fQ always exists if Z is in addition simply connected.
Corollary 2.37. Let X be path connected and locally path connected, x 2 X, and for
i D 1; 2 let pi W .XQ i ; xQ i / ! .X; x/ be covering maps such that XQ i is path connected. Set
Ci WD 1 .pi /.1 .XQi ; xQ i // 1 .X; x/. Then there exists a unique morphism of pointed
covering spaces ˛W .XQ 1 ; xQ 1 / ! .XQ2 ; xQ 2 / if and only if C1 C2 .
If C1 D C2 , then ˛ is necessarily an isomorphism of covering spaces. By symmetry,

there also exists a unique morphism of pointed covering spaces ˇW .XQ2 ; xQ 2 / ! .XQ1 ; xQ 1 /
and by uniqueness we have ˛ ı ˇ D idXQ2 and ˇ ı ˛ D idXQ 1 .
Remark 2.38. Let X be path connected and locally path connected, x 2 X. If pW XQ !

X is a universal covering map, xQ 2 p 1 .x/, then there exists for each covering map
qW .YQ ; y/
Q ! .X; x/ a unique morphism of pointed covering spaces .X; Q x/
Q ! .YQ ; y/Q
(apply Corollary 2.37 to the path component of YQ containing y). Q In other words, a pointed
Q x/
universal covering .X; Q is an initial object in the category of pointed coverings of .X; x/.
Remark 2.39. Let X be path connected and locally path connected, x 2 X. Corol-
lary 2.37 shows that the map
8 9
ˆ ( )
<isomorphism classes of pointed> =
subgroups
path connected covering spaces ! ;
:̂ >
; of 1 .X; x/ (2.2)
Q Q ! .X; x/
pW .X; x/
Q x/
.pW .X; Q ! .X; x// 7! 1 .p/.1 .X;Q x//
Q
is injective.
Let X be in addition semilocally simply connected (Problem 2.19, e.g., if X has an
open covering by simply connected subspaces), then one can show that (2.2) is in fact
bijective (Problem 2.20). In particular there exists a universal covering, corresponding to
the trivial subgroup.
The map (2.2) can also be extended to a functor on the category of (not necessarily path
connected) covering spaces as follows.
Remark 2.40. Let X be path connected and locally path connected and let pW XQ ! X
be a covering map. Fix x 2 X, and let F WD p 1 .x/ be the fiber. Then F is a discrete
Q Define a right action
subspace of X.
F 1 .X; x/ ! F
as follows. Let xQ 2 F and be a closed path representing an element in 1 .X; x/.

By Corollary 2.36 there exists a unique lifting to a path Q in XQ with start point x.Q Let
xQ Œ WD .1/
Q (this depends only on the homotopy class of by Proposition 2.32). Let
p 0 W XQ 0 ! X be a second covering map and let gW XQ ! XQ 0 be a morphism of covering
spaces. Then the induced map on fibers gx W p 1 .x/ ! p 01 .x/ is equivariant with respect
to the 1 .X; x/-action. We obtain a functor
˚x W (CovSp(X)) ! (Sets-1 .X; x/); p 7! p 1 .x/: (2.3)
Here (Sets-G) denotes the category of sets together with a right action by the group G.
The action of 1 .X; x/ on p 1 .x/ is transitive if and only if XQ is path connected.
Indeed, the condition is clearly necessary. Conversely suppose that XQ is path connected.
Let yQ 2 p 1 .x/ and choose a path Q in XQ from xQ to y.
Q Then yQ D xQ Œp ı Q .
Q x/
If pW .X; Q ! .X; x/ is a pointed path connected covering space, then the stabilizer
xQ of xQ 2 p 1 .x/ in the group 1 .X; x/ consists of those homotopy classes of paths
whose lift to XQ with start point xQ has also end point x. Q x//
Q Hence xQ D 1 .p/.1 .X; Q
and we see that the functor (2.3) indeed generalizes (2.2) by attaching to a subgroup of
1 .X; x/ the transitive 1 .X; x/-set of left cosets n1 .X; x/.
If X has a universal covering space, then one can show that the functor ˚x is an equiv-
alence of categories (Problem 2.21). Here we prove only the following easy very special
case.
Proposition 2.41. Let X be a simply connected locally path connected space. Then every
covering map of X is isomorphic to a trivial covering, i.e., to pr2 W F X ! X for
a discrete topological space F .
Proof. Fix x0 2 X. Let pW XQ ! X be a covering map and set F WD p 1 .x0 /. By

Corollary 2.36 there exists for every z 2 F a unique continuous map iz W X ! XQ such
that p ı iz D idX and iz .x0 / D z. As p is separated (Remark 2.30), iz is proper (Proposi-
tion 1.32 2) and hence a closed embedding. As p is a local homeomorphism, iz is an open
embedding (Appendix Remark 12.37 3). Therefore iz .X/ is open and closed in XQ and
hence a connected component isomorphic to X (recall that connected components and
path components are the same because XQ is locally path connected [Proposition 2.13]).
Let xQ 2 X,Q choose a path from p.x/ Q to x0 , and let Q be the lift in XQ with start point
Q Then the end point of Q is a point z 2 F and hence xQ and z are in the same path
x.
`
component, i.e., xQ 2 iz .X/. This shows that XQ D z2F iz .X/ Š F X.
2.6 Fundamental Groups and Coverings of Topological Groups
We now study the example of topological groups. We first show that their fundamental
groups are always abelian. This relies on the following simple remark.
Remark 2.42. Let X be a topological space, e 2 X, and let mW X X ! X be a con-

tinuous map such that m.x; e/ D m.e; x/ D x for all x 2 X. By functoriality we obtain
a homomorphism of groups
1 .m/W 1 .X; e/ 1 .X; e/ D 1 .X X; .e; e// ! 1 .X; e/; .; ı/ 7! ı;
where the equality is by Remark 2.23.

We claim that 1 .m/ coincides with group multiplication in 1 .X; e/. Indeed, let " be
the homotopy class of the constant path with image e. Then one has
ı ' . "/ ." ı/ D . "/ ." ı/ ' ı:
As a group G is abelian if and only if the multiplication G G ! G is a homomor-

phism of groups, one sees in particular that 1 .X; e/ is an abelian group, because 1 .m/
is a homomorphism of groups.
2.6 Fundamental Groups and Coverings of Topological Groups 35
Proposition 2.43. Let G be a topological group with neutral element e 2 G and let
pW GQ ! G be a covering map, where GQ is a path connected and locally path connected
topological space. Fix eQ 2 p 1 .e/.
1. There exists a unique group structure on GQ that makes GQ into a topological group with
neutral element eQ such that p is a homomorphism of groups.
2. The topological group GQ is abelian if and only if G is abelian.
3. One has an exact sequence of abelian groups
1 .p/ @eQ
Q e/
1 ! 1 .G; Q ! 1 .G; e/ ! Ker.p/ ! 1: (2.4)
In particular we see that @eQ yields a group isomorphism 1 .G; e/ ! Ker.p/ if GQ is

simply connected.
Proof. (1) and (2). Denote by mW G G ! G the group law on G. By Proposition 2.35
Q GQ GQ ! GQ
we can lift m ı .p p/ along p to a (necessarily unique) continuous map mW
Q e;
with m. Q e/
Q D eQ if
.1 .m/ ı 1 .p p//.1 .GQ G;

Q .e;
Q e/// Q
Q 1 .p/.1 .G//:
But Remark 2.42 shows that .1 .m/ı1 .p p//.Œ1 ; 2 / D 1 .p/.Œ1 2 /. Therefore m Q
exists. The uniqueness of liftings implies that m Q is associative and that m
Q is commutative
if m is commutative. The same argument shows that G ! G, g 7! g 1 can be lifted to
a map GQ ! GQ mapping eQ to eQ and defining an inverse for the multiplication m. Q
As m Q is a lifting of mı.p p/, p is a surjective homomorphism of groups. In particular
G is abelian if GQ is abelian.
(3). The homomorphism 1 .p/ is injective by Proposition 2.32. By Remark 2.40 we
have a transitive 1 .G; e/-action on Ker.p/, which yields a surjective map @eQ W 1 .G; e/ !
Ker.p/, Œ 7! eQ Œ with @1 Q D Im.1 .p//. The alternative description of the group
eQ .e/
law of 1 .G; e/ in Remark 2.42 shows that @eQ is a group homomorphism.
By Example 2.29 we deduce:
Corollary 2.44. One has 1 .S 1 ; 1/ D 1 .C n f0g; 1/ Š Z.
The first equality is given by the inclusion S 1 ,! C n f0g, which is a homotopy equiva-
lence by Example 2.4. Going through the construction of @eQ one sees that this isomorphism
is given by attaching to n 2 Z the closed path t 7! e 2i nt . The spheres S n for n 2 are
simply connected (Problem 2.15).
2.7 Problems
Problem 2.1. Let Y be a contractible space. Show that any two continuous maps X ! Y
are homotopic.
Problem 2.2. Let I be an index set.
1. Let fi ; gi W Xi ! Yi be continuous maps of topological spaces. Show that if fi ' gi

Q Q Q Q
for all i 2 I , then the maps i fi and i gi from i Xi to i Yi are homotopic.
2. Let X and Y be topological spaces, let Y be contractible. Show that the projection
X Y ! X is a homotopy equivalence.
Problem 2.3. Show that the inclusions O.n/ ,! GLn .R/ and U.n/ ,! GLn .C/ are
homotopy equivalences.
Hint: Use the polar decomposition and Problem 2.2.
Problem 2.4. Show that the union of the coordinate axes .R f0g/ [ .f0g R/ in R2 is
not homeomorphic to R.
Problem 2.5. Let n 2 be an integer, let U Rn be an open subspace, and let F U

be a finite subset. Show that if U is path connected, then U n F is path connected.
Problem 2.6. Let X be an infinite countable set endowed with the cofinite topology (Ap-
pendix Problem 12.16). Show that every non-empty open subspace is connected and that
no open subspace is path connected.
Problem 2.7. Let X be the subspace
f .1=n; y/ I n 2 N; y 2 Œ0; 1 g [ f .0; y/ I y 2 Œ0; 1 g [ f .x; 0/ I x 2 Œ0; 1 g
of R2 , and let X 0 WD X n f0; 0g.
1. Show that X is path connected but not locally path connected.

2. Show that X 0 is connected but not path connected.
˚
Problem 2.8. Show that R2 n .x; y/ I y D x 2 ; x 0 is not star shaped but con-
tractible.
2.7 Problems 37
Problem 2.9. Let X be a topological space. The fundamental groupoid ˘.X/ is the cate-
gory whose objects are the points of X. For x; y 2 X the set of morphisms Hom˘.X / .x; y/
is the set of paths from x to y modulo homotopy relative f0; 1g. Composition is defined
by Œ ı Œı WD ı .
1. Show that ˘.X/ is a groupoid (Appendix Problem 13.7). For x 2 X show that the
map Aut˘.X / .x/ ! 1 .X; x/, Œ 7! Œ is an isomorphism of groups.
2. Show that every continuous map f W X ! Y induces a functor ˘.f /W ˘.X/ ! ˘.Y /.
3. Show that if f is a homotopy equivalence, then ˘.f / is an equivalence of categories.
Problem 2.10. Every group G can be considered as a category with a single object whose
set of endomorphisms is the group G. Composition is defined by group multiplication
(convince yourself that this is really a category and even a groupoid [Appendix Prob-
lem 13.7]).
Let X be a path connected, let ˘.X/ be its fundamental groupoid (Problem 2.9). Fix
a point x 2 X and consider the group 1 .X; x/ as a groupoid. Show that the natural
inclusion functor 1 .X; x/ ! ˘.X/ is an equivalence of categories.
Problem 2.11. State and prove a similar result as Remark 2.23 for the fundamental
groupoid of X Y (Problem 2.9).
Problem 2.12. Let X be a topological space and fix x 2 X. Consider the map
( ) ( )
W S 1 ! X continuous W Œ0; 1 ! X closed path
! ; 7! ı e;
.1/ D x .0/ D x
where eW Œ0; 1 ! S 1 is the map t 7! e 2i t .
1. Show that this map induces an isomorphism Œ.S 1 ; 1/; .X; x/ ! 1 .X; x/, where the
left-hand side denotes the set of continuous maps f W S 1 ! X with f .1/ D x up to
homotopy relative f1g.
2. Let be a closed path in X with start point x and let be the corresponding continuous
map S 1 ! X. Show that is null-homotopic if and only if can be extended to
a continuous map D WD f z 2 C I jzj 1 g ! X.
3. Let G be a path connected topological group with neutral element e. Show that the
map “forgetting the base point” yields a bijective map Œ.S 1 ; 1/; .G; e/ ! ŒS 1 ; G.
Problem 2.13. Let n 1 be an integer. A piece-wise affine path in a subspace A of Rn is

a path W Œ0; 1 ! A such that there exist 0 D t0 < t1 < < tr D 1 and v1 ; : : : ; vr 2 Rn
such that .t/ D .ti 1 / C .t ti 1 /vi for all t with ti 1 t ti .
Let U Rn be an open subspace. Show that every path in U is homotopic relative
f0; 1g in U to a piecewise affine path in U .
Hint: Use that every homotopy of paths is uniformly continuous.
Problem 2.14. Let n 1 be an integer and let 0 ¤ v 2 Rn . Show that Rn n R0 v is

contractible. Deduce that S n1 n fx0 g is contractible for any x0 2 S n1 .
Problem 2.15. Let n 3 be an integer. Show that Rn n f0g and S n1 are simply con-
nected.
Hint: After replacing a closed path in Rn n f0g by a piecewise affine path (Prob-
lem 2.13), show that there exists 0 ¤ v 2 Rn such that R0 v 2 Rn n fg. Then use
Problem 2.14.
Problem 2.16. Show that a covering map of finite degree is proper. Show that a covering
map of degree 1 is a homeomorphism.
Problem 2.17. Let n 1 be an integer. Show that C n f0g ! C n f0g, z 7! z n is

a covering map of degree n.
1. Show that every morphism in (CovSp(X)) is again a covering map.

2. Let pW XQ ! X be a covering map with XQ connected. Show that A WD Aut(CovSp(X )) .p/
Q i.e., each xQ 2 XQ has an open neighborhood U such
acts properly discontinuous on X,
that U \ ˛.U / D ; for all ˛ ¤ idXQ .
Problem 2.19. A topological space X is called semilocally simply connected if every

point x 2 X has an open neighborhood U such that the map 1 .U; x/ ! 1 .X; x/,
induced by the inclusion, is trivial.
1. Show that if X has an open covering by simply connected subspaces, then X is semilo-
cally simply connected.
2. Suppose that X has a universal covering. Show that X is semilocally simply con-
nected.
S
3. For n 2 N let Cn WD @B 1 .. n1 ; 0// R2 and set X WD n Cn , considered as a subspace
n
of R2 . Show that X is not semilocally simply connected.
Problem 2.20. Let X be a path connected, locally path connected, semilocally simply
connected (Problem 2.19) space. The goal of this problem is to show that the map (2.2) is
surjective (and hence bijective). In particular X has a universal covering space.
Fix x0 2 X. Let XQ be the set of homotopy classes (relative f0; 1g) of paths in X starting
at x0 and let pW XQ ! X be the map Œ 7! .1/. Let xQ 0 2 XQ be the homotopy class of the
constant path with value x0 . Fix a subgroup C of 1 .X; x0 /.
1. Show that there exists a left action 1 .X; x0 / XQ ! XQ given by Œ Œ WD Œ .
Let XQC WD C nXQ be the set of C -orbits in XQ and xQ C;0 be the image of xQ 0 in XQ C . Show
that p induces a map pC W XQ C ! X.
2.7 Problems 39
2. Let U be the set of open path connected subspaces U X such that 1 .U; x/ !
1 .X; x/ is trivial for one (equivalently, for all) x 2 U . Show that U is a basis of the
topology of X.
3. For U 2 U and for a path in X with start point x0 and end point in U set UQ Œ WD
f Œ I path in U with .0/ D .1/g. Show that the UQ Œ form a basis for a topol-
ogy on X. Q Endow XQ C with the quotient topology.
4. Show that pC W .XQ C ; xQ C;0 / ! .X; x0 / is a pointed path connected covering.
5. Show that 1 .pC /.1 .XQC ; xQ C;0 // D C . Deduce that XQ is simply connected.
Problem 2.21. Let X be a topological space and a point x. We want to show that the
functor ˚x (2.3) is an equivalence of categories if X has a universal covering space (see
Problem 2.20 for an existence criterion). Fix a pointed universal covering W .X;Q x/
Q !
.X; x/.
1. Let Œ 2 1 .X; x/ and let Q be a lift of in XQ with start point x.

Q Show that there
exists a unique automorphism ˛ of the covering space such that ˛.x/ Q D .1/.
Q Show
that this construction yields a well-defined group isomorphism 1 .X; x/ ! Aut./.
Q
In particular, we obtain a left action of 1 .X; x/ on X.
2. For every set F (considered as a discrete topological space) with right 1 .X; x/-action
let .F / WD F 1 .X;x/ XQ be the quotient space of F XQ by the 1 .X; x/ right action
pr2
Q Show that F XQ ! XQ ! induces a continuous map
Q 7! .f ; 1 x/.
.f; x/
pF W .F / ! X and that pF is a covering map.
3. Show that for every morphism ˛W F1 ! F2 of sets with right 1 .X; x/-action, the map
˛ idXQ induces a morphism of covering spaces .˛/W .F1 / ! .F2 /. Show that
one obtains a functor
W (Sets-1 .X; x/) ! (CovSp(X)):
4. Show that the functors ˚x and are mutually inverse equivalences of categories.
Problem 2.22. Let 'W G ! H be a continuous homomorphism of topological groups.

Show that ' is a covering map if and only if ' is open surjective and Ker.'/ is a discrete
subgroup of G.
Problem 2.23. Let z0 2 C and let hW C n fz0 g ! S 1 , z 7! zz0

jzz0 j
.
1. Show that h is a homotopy equivalence and induces isomorphisms
Wz0 W ŒS 1 ; C n fz0 g ! ŒS 1 ; S 1 ! Œ.S 1 ; 1/; .S 1 ; 1/

! HomZ .1 .S 1 ; 1/; 1 .S 1 ; 1// D Z
(use Problem 2.12). For W S 1 ! C n fz0 g, Wz0 . / is called the winding number of
around z0 . We also write Wz0 ./ if is the closed path corresponding to .
2. Show that for a continuous map W S 1 ! C, the map C n .S 1 / ! Z, z 7! Wz . / is

locally constant.
3. Let W Œ0; 1 ! C n f0g be a closed path and let W
Q Œ0; 1 ! C be a lifting along the
covering C ! C n f0g, z 7! exp.2 iz/. Show that W0 ./ D .1/ Q .0/.
Q
4. Let W Œ0; 1 ! Cnfz0 g be a piecewise continuously differentiable closed path. Deduce
from 3 that Z
1 dz
Wz0 ./ D :
2 i z z0

Sheaves
3
In this chapter we introduce one of the central notions in this book: sheaves. They are the
main tool to keep track systematically of locally defined data attached to an open subset
of a topological space. Giving such data are compatible with restriction to smaller open
subsets yields the notions of a presheaf. A sheaf is then a presheaf where such local data,
which are given on an open covering .Ui /i and are compatible with restriction to pairwise
intersections Ui \ Uj , can be glued to a global datum. One basic example is to attach to
an open set U the R-algebra of continuous functions U ! R: For an open covering .Ui /i
of X giving a continuous function on X is the same as giving continuous functions fi on
all Ui such that fi jUi \Uj D fj jUi \Uj for all i; j .
After defining presheaves and sheaves we make precise in the second section the notion
of “defined in some unspecified neighborhood of a point” and introduce stalks of sheaves
and germs of sections of sheaves. Section 3.3 explains how to make a presheaf into a sheaf.
In Sect. 3.4 we will see that one may view sheaves on a topological space X also as local
homeomorphisms E ! X, called étalé spaces. This second point of view for sheaves
will come in handy in Sect. 3.5 where we define for continuous maps the direct image and
inverse image of sheaves. The direct image can be easily defined in terms of sheaves. The
inverse image is more easily understood in terms of étalé spaces. In the final section we
construct limits and colimits of sheaves showing that the category of sheaves is complete
and cocomplete.
Notation: In this chapter X always denotes a topological space.
3.1 Presheaves and Sheaves
Definition 3.1 (Presheaf). A presheaf F on a topological space X consists of the follow-

ing data:

DOI 10.1007/978-3-658-10633-1_3
42 3 Sheaves
(a) For every open set U of X a set F.U /.

(b) For each pair of open sets U V of X a map resVU W F.V / ! F.U /, called a restric-
tion map,
such that the following conditions hold:
1. resUU D idF.U / for every open set U X.

2. For U V W open sets of X, resW U D resU ı resV .
V W
Let F1 and F2 be presheaves on X. A morphism of presheaves 'W F1 ! F2 is a family

of maps 'U W F1 .U / ! F2 .U / (for all U X open) such that for all pairs of open sets
U V in X the following diagram commutes:
'V
F1 .V / F2 .V /
resVU resVU
'U
F1 .U / F2 .U /:
Composition of morphisms ' and of presheaves is defined in the obvious way: ' ı WD
.'U ı U /U . We obtain the category (PSh(X)) of presheaves on X.
If U V are open sets of X and s 2 F.V / we will usually write s jU instead of resVU .s/.
The elements of F.U / are called sections of F over U . Very often we will also write
.U; F/ instead of F.U /.
Remark 3.2. We can also describe presheaves as follows. Let .OpenX / be the category
whose objects are the open sets of X and, for two open sets U; V X, Hom.U; V /
consists of the inclusion map U ! V if U V and is empty otherwise. In other words,
.OpenX / is the category attached to the set of open subsets of X partially ordered by
inclusion (Appendix Remark 13.32). Composition of morphisms is the composition of
the inclusion maps. Then a presheaf is the same as a contravariant functor F from the
category .OpenX / to the category (Sets) of sets. A morphism of presheaves is the same as
a morphism of functors.
Remark and Definition 3.3. By replacing (Sets) in Remark 3.2 by some other category C
(e.g., the category of abelian groups, the category of rings, or the category of R-algebras,
R a fixed ring) we obtain the notion of a presheaf F with values in C (e.g., a presheaf of
groups, a presheaf of rings, a presheaf of A-algebras for some commutative ring A). This
signifies that F.U / is an object in C for every open subset U of X and that the restriction
maps are morphisms in C .
A morphism F1 ! F2 of presheaves with values in C is then simply a morphism of
functors.
3.1 Presheaves and Sheaves 43
If F is a presheaf, one should think of F.U / as some kind of maps on open subsets U of
X with a way to restrict to smaller open subsets. In fact most of the presheaves that we
will encounter will be presheaves of functions in the following sense.
Example 3.4 (Presheaf of functions). Let E be a set. For each open subset U X let
MapE .U / WD MapX IE .U / be the set of all maps U ! E. For an open subset V U
we define resUV W MapE .U / ! MapE .V / as the usual restriction of maps. Then MapX IE is
a presheaf on X.
More generally, a family F of subsets F.U / MapE .U /, where U runs through the
open subsets of X, is called a presheaf of E-valued functions on X, if it is stable under
restriction, i.e., for all open sets V U and all f 2 F.U / one has f jV 2 F.V /. Then
F together with the restriction maps is a presheaf of sets. If E is a group (respectively an
R-module for some ring R, respectively an A-algebra for some commutative ring A), then
F is a presheaf of groups (respective of R-modules, respective of A-algebras).
Example 3.5. Let X be a topological space.
1. Let Y be a topological space. For U X open set
CX IY .U / WD f f W U ! Y I f continuousg :
Then CX IY is a presheaf of Y -valued functions on X.

If Y D K (K D R or K D C), then CX IK .U / is a commutative K-algebra and CX IK is
a presheaf of K-algebras.
2. Let V and W be finite-dimensional R-vector spaces and let X be an open subspace of
b 0 . For U X open set
V . Let ˛ 2 N
CX˛ IW .U / WD f f W U ! W I f is a R-C ˛ -mapg :
Then CX˛ IW is a presheaf of functions on X. It is a presheaf of R-vector spaces. If

W D R we simply write CX˛ . This is a presheaf of R-algebras.
3. Let V and W be finite-dimensional C-vector spaces. Let X be an open subspace of V .
For U X open set
OX IW .U / WD OXholIW .U / WD f f W U ! W I f holomorphicg :
Then OX IW (with the usual restriction maps) is a presheaf of C-vector spaces. For
W D C we simply write OX or OXhol . This is a sheaf of C-algebras.
4. Let E be a set and set F.U / WD E for all U X open. Define all restriction maps to
be the identity idE . Then F is a presheaf the constant presheaf with value E. We can
also view this as the sheaf of constant E-valued functions on X.
44 3 Sheaves
Example 3.6. Let ˛ 2 N. b For X D R and U R open let dU W C˛ .U / ! C˛1 .U / be

R R
the derivative f 7! f 0 . Then .dU /U is a morphism of presheaves of R-vector spaces (but
not of R-algebras because usually .fg/0 ¤ f 0 g 0 ).
Definition 3.7 (Sheaf). Let X be a topological space. The presheaf F on X is called a

S
sheaf , if for all open sets U in X and every open covering U D i 2I Ui the following
condition holds:
(Sh) Given si 2 F.Ui / for all i 2 I such that si jUi \Uj D sj jUi \Uj for all i; j 2 I . Then
there exists a unique s 2 F.U / such that s jUi D si for all i 2 I .
A morphism of sheaves is a morphism of presheaves.
We obtain the category of sheaves on the topological space X, which we denote by

.Sh.X//.
In the same way we can define the notion of a sheaf of groups, a sheaf of rings, a sheaf of
R-modules (R a fixed ring), or a sheaf of A-algebras (A a fixed commutative ring). Instead
of a (pre)sheaf of abelian groups (equivalently, of Z-modules) we sometimes speak just
of an abelian (pre)sheaf .
The uniqueness assertion in Condition (Sh) can be also phrased as follows.
Remark 3.8. Let X be a topological space, F a sheaf on X, let U X be open and let
S
U D i Ui be an open covering. Let s; s 0 2 F.U / such that s jUi D s 0 jUi for all i 2 I .
Then s D s 0 .
Remark 3.9. If F is a sheaf on X, F.;/ is a set consisting of one element (use the
covering of the empty set U D ; with empty index set I D ; in Condition (Sh)).
In particular, if the only open subsets of X are X and ; (for instance if X consists of
one point), then a sheaf F on X is already uniquely determined by F.X/ and sometimes
we identify F with F.X/.
Remark 3.10 (Sheaf of functions). Let E be a set and let F be a presheaf of E-valued
functions on X (Example 3.4). Then F is a sheaf if and only if the following condition
holds:
(ShF) For every open subset U of X, for every open covering .Ui /i of U and for every
map f W U ! E such that f jUi 2 F.Ui / for all i one has f 2 F.U /.
Remark 3.11 (Restriction of a sheaf to an open subspace). Let F be a presheaf on

a topological space X and let U be an open subspace of X. Then every open subset V
of U is also open in X and we define a presheaf FjU on U by FjU .V / WD F.V / with the
same restriction maps as for F. If F is a sheaf, then FjU is a sheaf.
3.1 Presheaves and Sheaves 45
Example 3.12. Notations of Example 3.5. The presheaves CX IY , CX˛ IW , and OXhol are
sheaves.
The presheaf of constant functions with values in some set is in general not a sheaf: If
U1 and U2 are disjoint non-empty open subsets and if f1 W U1 ! E and f2 W U2 ! E are
constant maps that take different values, there exists no constant map f on U D U1 [ U2
whose restriction to Ui is fi for i D 1; 2. The problem is that “constant” cannot be
checked locally, i.e., does not satisfy Condition (ShF). If one takes instead the sheaf of
locally constant functions with values in some set E, then this is a sheaf.
To define a sheaf it suffices to give its values on a basis of the topology. For instance if
an open covering .Ui /i of X is given, it suffices to give the value of a sheaf only on those
open subsets that are contained in one of the Ui s.
Remark 3.13. Let X be a topological space and let F be a sheaf on X. Let B a basis of
the topology on X. If we know the value F.U / of a sheaf on every element U of B, we
can use the sheaf property to determine F.V / on an arbitrary open V of X. We simply
cover V by elements of B. Here is a more systematic way of saying this:
8 9
ˆ
< Y >
=
F.V / D .sU / 2 F.U / I for U 0 U both in B: sU jU 0 D sU 0 (3.1)
:̂ U 2B
>
;
U V
D lim F.U /: (3.2)

U 2B
U V
Using this observation, we see that it suffices to define a sheaf on a basis B of open sets
of the topology of a topological space X: Consider B as a full subcategory of .OpenX /.
Then a presheaf on B is a functor FW Bopp ! (Sets). Every such presheaf F on B can be
extended to a presheaf F0 on X by using (3.2) as a definition, i.e., for V open in X,
F0 .V / D lim F.U /: (3.3)

U 2B
U V
The restriction maps are then given by the projections via the description (3.1). A mor-
phism of presheaves on B is again defined as a morphism of functors.
Then the presheaf F0 on X is a sheaf if and only if F satisfies the condition .Sh/ of
Definition 3.7 for every U 2 B and for every open covering .Ui /i of U with Ui 2 B for
all i. In this case, we say that F is a sheaf on B.
Attaching to a sheaf F on B the sheaf F0 on X is clearly functorial in F and we obtain
an equivalence between the category of sheaves on B and the category of sheaves on X.
A similar remark holds for sheaves of groups, of R-modules (R some ring), or of A-

algebras (A some commutative ring).
46 3 Sheaves
3.2 Stalks
Let X be a topological space, F be a presheaf on X, considered as a functor FW .OpenX /opp

! (Sets), and let x 2 X be a point. Let
U.x/ WD f U X open I x 2 U g
be the set of open neighborhoods of x, ordered by inclusion. We consider U.x/ as a full

subcategory of Open.X/. By restricting F to U.x/ we obtain a functor FW U.x/opp !
(Sets). Note that the category U.x/opp is filtered (Appendix Definition 13.37): For any
two neighborhoods U1 and U2 of x there exists a neighborhood V of x with V U1 \ U2
(e.g., V D U1 \ U2 ).
Definition and Remark 3.14.
1. The colimit
Fx WD colim F (3.4)
U.x/
is called the stalk of F in x. More concretely (Appendix Example 13.33), one has
Fx D f .U; s/ I U open neighborhood of x, s 2 F.U /g = ; (3.5)
where two pairs .U1 ; s1 / and .U2 ; s2 / are equivalent, if there exists an open neighbor-
hood V of x with V U1 \ U2 such that s1jV D s2jV (note that V might be smaller
than U1 \ U2 ).
2. For each open neighborhood U of x we have a canonical map
F.U / ! Fx ; s 7! sx ; (3.6)
which sends s 2 F.U / to the class of .U; s/ in Fx . We call sx the germ of s in x.

3. If 'W F ! G is a morphism of presheaves on X, we have an induced map
'x WD colim 'U W Fx ! Gx

U 3x
of the stalks in x. It sends the equivalence class of .U; s/ in Fx to the class of

.U; 'U .s// in Gx . We obtain a functor F 7! Fx from the category of presheaves
on X to the category of sets. For every x 2 X and every open neighborhood one has
a commutative diagram
s7!sx
F.U / Fx
'U 'x
t 7!tx
G.U / Gx :
If F is a presheaf of functions one should think of the stalk Fx as the set of functions
defined in some unspecified open neighborhood of x.
3.2 Stalks 47
Remark 3.15. If F is a presheaf on X with values in C , where C is any category in which

filtered colimits exist (for instance the category of groups, of R-modules [R ring], or of A-
algebras [A commutative ring]), then the stalk Fx is an object in C and we obtain a functor
F 7! Fx from the category of presheaves on X with values in C to the category C .
Let us make this more precise for a sheaf G of groups. The underlying set of Gx is
given by (3.5). The group law on Gx is defined as follows. Let g; h 2 Gx be represented
by .U; s/ and .V; t/. Choose an open neighborhood W of x with W U \ V . Then
.U; s/ .W; s jW / and .V; t/ .W; t jW / and the product gh is the equivalence class of
.W; .s jW /.t jW //.
In the same way addition and multiplication (respectively scalar multiplication) is de-
fined on the stalk for a sheaf of rings (respective of R-modules).
Example 3.16. Let X D K and OK be the sheaf of analytic functions on X. Fix z0 2

K. Then two analytic functions f1 and f2 defined in open neighborhoods U1 and U2 ,
respectively, of z0 agree on some open neighborhood V U1 \ U2 if and only if they
have same Taylor expansion around z0 . Therefore
( )
X
OK;z0 D an .z z0 / power series with radius of convergence > 0 :
n
n0
We have a similar description of the stalk of the sheaf of analytic functions on any open
subset X of Km for some m 2 N.
Example 3.17 (Stalk of the sheaf of continuous functions). Let X be a topological

space, let CX D CX IR be the sheaf of continuous R-valued functions on X, and let x 2 X.
Then
CX;x D f .U; f / I x 2 U X open, f W U ! R continuousg = ;
where .U; f / .V; g/ if there exists x 2 W U \ V open such that f jW D g jW .
In particular we see that for x 2 V U open and f W U ! R continuous one has
.U; f / .V; f jV /. As CX is a sheaf of R-algebras, CX;x is an R-algebra.
If the germ s 2 CX;x of a continuous function at x is represented by .U; f /, then
s.x/ WD f .x/ 2 R is independent of the choice of the representative .U; f /. We obtain
an R-algebra homomorphism
evx W CX;x ! R; s 7! s.x/;
which is surjective because CX;x contains in particular the germs of all constant functions.
Let mx WD ker.evx / D f s 2 CX;x I s.x/ D 0 g. Then mx is a maximal ideal because
CX;x =mx Š R is a field. Let s 2 CX;x n mx be represented by .U; f /. Then f .x/ ¤ 0. By
shrinking U we may assume that f .y/ ¤ 0 for all y 2 Y because f is continuous. Hence
1=f exists and s is a unit in CX;x . Therefore the complement of mx consists of units of
CX;x . This shows that CX;x is a local ring with maximal ideal mx .
48 3 Sheaves
The same argument shows that for every open subspace X of a finite-dimensional R-
vector space (respectively a finite-dimensional C-vector space) the stalk CX;x
˛
(respectively
OX;x ) is a local ring with residue field R (respectively C).
hol
The following result will be used very often.
Proposition 3.18. Let X be a topological space, F and G presheaves on X, and let

'; W F ! G be two morphisms of presheaves.
1. Assume that F is a sheaf. Then the induced maps on stalks 'x W Fx ! Gx are injective
for all x 2 X if and only of 'U W F.U / ! G.U / is injective for all open subsets U X.
2. If F and G are both sheaves, the maps 'x are bijective for all x 2 X if and only if 'U
is bijective for all open subsets U X.
3. If F and G are both sheaves, the morphisms ' and are equal if and only if 'x D x
for all x 2 X.
Proof. For U X open consider the map

Y
F.U / ! Fx ; s 7! .sx /x2U :
x2U
We claim that this map is injective if F is a sheaf. Indeed let s; t 2 F.U / such that sx D tx
for all x 2 U . Then for all x 2 U there exists an open neighborhood Vx U of x such
S
that s jVx D t jVx . Clearly, U D x2U Vx and therefore s D t by Remark 3.8.
Using the commutative diagram
Q
F.U / x2U Fx
Q
'U x 'x
Q
G.U / x2U Gx ;
we see that (3) and the necessity of the condition in (1) are implied by the above claim.
Moreover, a filtered colimit of injective maps is always injective again. Indeed, this
follows either from abstract nonsense (Appendix Example 13.50) or we can argue directly.
Let s0 ; t0 2 Fx such that 'x .s0 / D 'x .t0 /. Let s0 be represented by .s; U / and t0 by .t; V /.
By shrinking U and V we may assume U D V . As
'U .s/x D 'x .s0 / D 'x .t0 / D 'U .t/x ;
there exists an open neighborhood x 2 W U such that
'W .s jW / D 'U .s/jW D 'U .t/jW D 'W .t jW /:

3.2 Stalks 49
As 'W is injective, we find s jW D t jW and hence s0 D sx D tx D t0 . Therefore the

condition in (1) is also sufficient.
Hence we are done if we show that the bijectivity of 'x for all x 2 U implies the
surjectivity of 'U . Let t 2 G.U /. For all x 2 U we choose an open neighborhood
U x of x in U and s x 2 F.U x / such that .'U x .s x //x D tx . Then there exists an open
neighborhood V x U x of x with 'V x .s x jV x / D t jV x . Then .V x /x2U is an open covering
of U and for x; y 2 U
'V x \V y .s x jV x \V y / D t jV x \V y D 'V x \V y .s y jV x \V y /:
As we already know that 'V x \V y is injective, this shows s x jV x \V y D s y jV x \V y and the

sheaf condition (Sh) ensures that we find s 2 F.U / such that s jV x D s x for all x 2 U .
Clearly, we have 'U .s/x D tx for all x 2 U and hence 'U .s/ D t.
Definition 3.19. We call a morphism 'W F ! G of sheaves injective (respectively surjec-

tive, respectivelybijective) if 'x W Fx ! Gx is injective (respectively surjective, respectively
bijective) for all x 2 X.
Problem 3.7 relates these notions to the categorical notions of monomorphism (respec-
tively epimorphism, respectively isomorphism).
Remark 3.20. Let 'W F ! G be a morphism of sheaves.
1. Then ' is injective (respectively bijective) if and only if 'U W F.U / ! G.U / is injective
(respectively bijective) for all open U X (Proposition 3.18).
2. The morphism ' is surjective if and only if for all open subsets U X and every
S
t 2 G.U / there exist an open covering U D i Ui (depending on t) and sections
si 2 F.Ui / such that 'Ui .si / D t jUi , i.e., locally we can find a preimage of t. But the
surjectivity of ' does not imply that 'U W F.U / ! G.U / is surjective for all open sets
U of X as Example 3.22 shows.
Similarly, as we defined the notions of injectivity or surjectivity “stalkwise”, we also

define the notion of an exact sequence of sheaves of groups “stalkwise”. Recall that
'
a sequence G ! H ! K of homomorphisms of groups is called exact if Im.'/ D
Ker. /.
'
Definition 3.21. A sequence F ! G ! H of homomorphisms of sheaves of groups
on a topological space is called exact if for all x 2 X the induced sequence on stalks
'x x
Fx ! Gx ! Hx is an exact sequence of groups. A sequence ! Fi 1 ! Fi !
Fi C1 ! : : : is called exact if Fi 1 ! Fi ! Fi C1 is exact for all i.
A homomorphism 'W F ! G of sheaves of groups is injective (respectively surjective)

' '
if and only if 0 ! F ! G (respectively F ! G ! 0) is an exact sequence.
50 3 Sheaves
Example 3.22. Let OX be the sheaf of holomorphic functions on an open subset X of C.

For every open subspace U X and f 2 OX .U / we let DU .f / WD f 0 be the derivative.
We obtain a morphism DW OX ! OX of sheaves of C-vector spaces. Then D is surjective,
because locally every holomorphic function has a primitive.
But there exist open subsets U of X and functions f on U that have no primitive., for
instance U D Br .z0 / n fz0 g contained in X and f W z 7! 1=.z z0 /. More precisely, by
complex analysis we know that DU is surjective if and only if every connected component
of U is simply connected. The sufficiency of this condition will be also an immediate
application of cohomological methods developed later (Example 7.6).
We obtain an exact sequence of sheaves of C-vector spaces
D
0 ! CX ! OX ! OX ! 0;
where CX denotes the sheaf of locally constant C-valued functions on X and where U is
the inclusion for all U X open.
Remark 3.23. Let F and G be sheaves on a topological space X and let .Ui /i be an open
covering of X. A morphism of sheaves 'W F ! G is injective (respectively surjective,
respectively bijective) if and only if its restriction ' jUi W FjUi ! GjUi to morphisms of
sheaves on Ui is injective (respectively surjective, respectively bijective) for all i because
these notions are defined via the stalks.
But note that the existence of the morphism ' is crucial. There exist sheaves F and G
such that FjUi is isomorphic to GjUi for all i and such that F and G are not isomorphic (see
Example 3.42).
Definition 3.24. Let F be a sheaf of abelian groups (using additive notation) on a topolog-
ical space X, U X open and s 2 F.U / a section. The support of s is the set supp.s/ WD
f x 2 U I sx ¤ 0 g. The support of F is the set supp.F/ WD f x 2 X I Fx ¤ 0 g.
Lemma 3.25. Let F be a sheaf of abelian groups on a topological space, U X open,

and s 2 F.U / a section. Then supp.s/ is closed in U .
Proof. For x 2 U n supp.s/ we have sx D 0 so that there exists V U open with

s jV D 0. This implies that sx 0 D 0 for every x 0 2 V and therefore V U n supp.s/.
Hence U n supp.s/ is open.
Example 3.26. Let X be a topological space. Let CX D CX IR be the sheaf of continuous

R-valued functions on X. Let U X be open and s 2 CX .U / a continuous function
U ! R. In the proof of Lemma 3.25 we have just seen that U n supp.s/ is the interior of
f x 2 U I s.x/ D 0 g. Therefore
supp.s/ D f x 2 U I s.x/ ¤ 0 g (closure in U ).

3.3 Sheaves Attached to Presheaves 51
3.3 Sheaves Attached to Presheaves
There is a functorial way to attach to a presheaf a sheaf.
Proposition and Definition 3.27 (Sheafification). Let F be a presheaf on a topological

space X. Then there exists a pair .F; Q F /, where FQ is a sheaf on X and F W F ! FQ
is a morphism of presheaves, such that the following holds: If G is a sheaf on X and
'W F ! G is a morphism of presheaves, then there exists a unique morphism of sheaves
Q FQ ! G with 'Q ı F D '. The pair .F;
'W Q F / is unique up to unique isomorphism.
Moreover, the following properties hold:
1. For all x 2 X the map on stalks F;x W Fx ! FQ x is bijective.

2. For every presheaf G on X and every morphism of presheaves 'W F ! G there exists
a unique morphism 'WQ FQ ! GQ making the diagram
FQ
F
F (3.7)
' 'Q
G
G GQ
commutative. In particular, F 7! FQ is a functor from the category of presheaves on X

to the category of sheaves on X.
The sheaf FQ is called the sheaf associated to F or the sheafification of F.
We can reformulate the first part of Proposition 3.27 by saying that the sheafification
functor is a left adjoint functor to the inclusion functor of the category of sheaves into the
category of presheaves.
Proof. For U X open, elements of F.U Q / are by definition families of elements in the
stalks of F, which locally give rise to sections of F. More precisely, we define
(
Y
Q
F.U / WD .sx / 2 Fx I 8x 2 U : 9 an open neighborhood W U of x,
x2U
)
and t 2 F.W /: 8w 2 W : sw D tw :
For U V the restriction map F.V Q / ! F.U Q / is induced by the natural projection
Q Q
Q
x2V Fx ! x2U Fx . Then it is easy to check that F is a sheaf. For U X open, we
define F;U W F.U / ! F.U / by s 7! .sx /x2U . The definition of FQ shows that, for x 2 X,
Q
FQ x D Fx and that F;x is the identity.
52 3 Sheaves
Q /
Now let G be a presheaf on X and let 'W F ! G be a morphism. Sending .sx /x 2 F.U
Q / defines a morphism FQ ! G.
to .'x .sx //x 2 G.U Q By Proposition 3.18 (3) this is the
unique morphism making the diagram (3.7) commutative.
If we assume in addition that G is a sheaf, then the morphism of sheaves G W G ! G,Q
which is bijective on stalks, is an isomorphism by Proposition 3.18 (2). Composing the
morphism FQ ! GQ with 1 Q FQ ! G. Finally, the uniqueness of
G , we obtain the morphism 'W
Q
.F; F / is a formal consequence.
Remark 3.28. If F is a presheaf of (abelian) groups, of rings, of R-modules, or of R-

algebras, its associated sheaf is a sheaf of (abelian) groups, of rings, of R-modules, or of
R-algebras.
Remark 3.29. From Proposition 3.18 (2), we get the following characterization of the
sheafification. Let F be a presheaf and G be a sheaf. Then G is isomorphic to the sheafi-
fication of F if and only and if there exists a morphism W F ! G such that x is bijective
for all x 2 X.
Example 3.30. Let E be a set and let F be a presheaf of functions with values in E. Then
its sheafification is given by
Q / D ff W U ! E j 9 open covering .Ui /i of U

F.U
such that f jUi 2 F.Ui / for all ig:
Indeed, this is a sheaf by Remark 3.10 and the inclusions F.U / ,! F.UQ / for U open
define a morphism of presheaves F ! FQ that is bijective on stalks. Hence we can apply
Remark 3.29.
Example 3.31. Let E be a set and denote by EX the sheaf of locally constant func-
tions, i.e., EX .U / D ff W U ! E locally constant mapg. This is the sheafification of the
presheaf of constant functions with values in E. The sheaf EX is called the constant sheaf
with value E.
If E is a group, then the multiplication in E makes EX into a sheaf of groups. A similar
remark applies if E is an R-module (R some ring) or an A-algebra (A some commutative
ring).
3.4 Sheaves and Étalé Spaces 53
3.4 Sheaves and Étalé Spaces
We now give a different description of sheaves via so-called étalé spaces.
Definition 3.32 (Étalé space). Let X be a topological space.
1. A pair .E; /, where E is a topological space and W E ! X is a local homeomor-

phism (Appendix Definition 12.36), is called an étalé space over X.
2. Let .E1 ; 1 / and .E2 ; 2 / be étalé spaces over X. A morphism f W .E1 ; 1 / ! .E2 ; 2 /
of étalé spaces is a continuous map f W E1 ! E2 such that 1 D 2 ı f .
Denote by (Ét/X) the category of étalé spaces over X.
One can visualize an étalé space E as a puff pastry lying over the base space X (see
Fig. 3.1).
Definition and Remark 3.33. Let .E; / be an étalé space over X. For U X open,
a section of E over U is a continuous map sW U ! E with ı s D idU . The fiber of E
over x 2 X is the set Ex WD 1 .x/. A morphism of étalé spaces f W .E1 ; 1 / ! .E2 ; 2 /
gives rise to maps fx W .E1 /x ! .E2 /x for x 2 X.
We will now show that the notion of an étalé spaces over X is equivalent to the notion
of a sheaf (of sets) over X. More precisely, we will construct functors
F W (Ét/X) ! (Sh(X)); GW .PSh.X// ! (Ét/X)
and we will show that G ı F is isomorphic to the identity functor and that F ı G is isomor-
phic to the sheafification functor. In particular, F and the restriction of G to (Sh(X)) yield
an equivalence of the categories (Ét/X) and .Sh.X//.
Figure 3.1 Étalé space
X
54 3 Sheaves
Construction 3.34. Let .E; / be an étalé space over X. Define a presheaf E of E-valued
functions by
E.U / WD fsW U ! E sectiong D f sW U ! E I s continuous, ı s D idU g :
It is a sheaf of E-valued functions because being continuous and being a section of is

local on X (Remark 3.10).
Let f W .E1 ; 1 / ! .E2 ; 2 / be a morphism of étalé spaces and E1 , E2 the correspond-
ing sheaves. For U X open and s 2 E1 .U / we have 2 ı .f ı s/ D 1 ı s D idU and
because f and s were continuous we have f ı s 2 E2 .U /. Now we are able to define the
morphism fOW E1 ! E2 of sheaves by fOU .s/ D f ı s for U X open and s 2 E1 .U /. We
obtain a functor denoted by
FW (Ét/X) ! .Sh.X//:
The Lemma 3.35 gives a description of stalks and of the topology of an étalé space.
Lemma 3.35. Let .E; / be an étalé space over X and E D F..E; // the associated
sheaf on X.
1. The maps x W Ex ! Ex ; s 7! s.x/ are bijective for every x 2 X.

2. The topology of E is the finest topology such that every s 2 E.U /, for any U X
open, is continuous (direct image topology).
Proof. 1. For x 2 X, the stalk Ex is the set of equivalence classes of pairs .U; s/, where
U is an open neighborhood of x and sW U ! E is a section of . Here .U; s/ and .V; t/
are equivalent if there exists x 2 W U \ V open such that s jW D t jW .
For e 2 E, x WD .e/ (i.e., e 2 Ex ) there exists an open neighborhood e 2 V
E such that jV is a homeomorphism onto its open image. Then . jV /1 is obviously
a section of with . jV /1 .x/ D e. This shows that x is surjective for every x 2 X.
Let sW U ! E, s 0 W U 0 ! E be sections with s.x/ D s.x 0 / DW e for a point x 2 U \ U 0 .
Then there exists e 2 V E open such that jV W V ! .V / is a homeomorphism
onto its open image. We set W WD .V / \ U \ U 0 and replace V by j 1 V .W /. Now
0
jV W V ! W is a homeomorphism again and x 2 W U \U is open. Then s jW ı jV D
idV D s 0 jW ı jV and therefore s jW D s 0 jW . Hence x is injective for every x 2 X.
2. Recall that in the direct image topology a subset W of E is open if and only if
s 1 .W / X is open for every U X open and s 2 E.U /. This is true for the topology
of E because every section is a homeomorphism onto its open image and we have already
seen in the first part of the proof that E admits an open cover by images of sections.
Now we attach conversely to every presheaf an étalé space as follows.

3.4 Sheaves and Étalé Spaces 55
`
Construction 3.36. Let E be a presheaf on X. Define E WD x2X Ex as a set, the map
W E ! X, Ex 3 e 7! x, and define the maps fU;s W U ! E, x 7! sx for U X open,
s 2 E.U /. Endow E with the finest topology such that all maps fU;s are continuous.
To show that .E; / is an étalé space, let e 2 E and x WD .e/. Then e 2 Ex so that
there exist x 2 U X open and s 2 E.U / with sx D e. Now ı fU;s D idU and
with V WD fU;s .U / this also implies fU;s ı jV D idV so that jV and fU;s are inverse to
each other. Furthermore .fU 0 ;s 0 /1 .V / D fy 2 U 0 \ V I sy0 D sy g X is open for every
U 0 X open, s 0 2 F.U / so that e 2 V E is open. Therefore .E; / is an étalé space
over X.
Let fOW E1 ! E2 be a morphism of presheaves and let .E1 ; 1 /, .E2 ; 2 / be the étalé
spaces corresponding to E1 and E2 , respectively. Define the map
f W E1 ! E2 ; .E1 /x 3 e 7! fOx .e/ 2 .E2 /x
so that 1 D 2 ı f . Then .f ı fU;s /.x/ D f .sx / D fOx .sx / D fO.s/x D fU;fO.s/ .x/ and
therefore f ı fU;s D fU;fO.s/ is continuous for U X open, s 2 E.U /. By definition of
the topology of E1 , f is continuous and therefore a morphism of étalé spaces. We obtain
a functor
GW .PSh.X// ! (Ét/X):
Lemma 3.37. There is a natural isomorphism from F ı G to the sheafification.
Proof. Let E be a presheaf on X, .E; / D G.E/ and E0 D F..E; //. By construction

fU;s W U ! E; x 7! sx is a section of .E; / and therefore an element of E0 .U / for U X
open, s 2 E.U /. We define a morphism of presheaves W E ! E0 by E.U / ! E0 .U /; s 7!
fU;s .
Let x 2 X be an arbitrary point. By construction Ex D Ex and due to Lemma 3.35
there is a bijective map x W E0x ! Ex D Ex ; sQ 7! sQ .x/. For U X open, s 2 E.U / we
have x . x .sx // D x ..fU;s /x / D fU;s .x/ D sx so that x ı x D idEx and x is bijective
for every x 2 X. It is straightforward to check that defines a natural transformation
and due to Proposition 3.27 we attain an isomorphism from E0 to the sheafification of E in
a natural way making it a natural isomorphism.
Lemma 3.38. There is a natural isomorphism G ı F to the identity functor.
Proof. Let .E; / be an étalé space over X, E D F..E; // and .E 0 ; 0 / D G.E/. By
F
Lemma 3.35 we have a bijection x W Ex ! Ex and by construction E 0 D x2X Ex . This
defines a bijective map W E 0 ! E with ı D 0 . For U X open and s 2 E.U /
we have .fU;s .x// D .sx / D s.x/ for every x 2 X so that ı fU;s D s. The topology
of E 0 is the finest such that fU;s W U ! E 0 is continuous for every s; U and the topology
56 3 Sheaves
of E is the finest such that sW U ! E is continuous for every s; U . This implies that is
a homeomorphism:
W E is open , s 1 .W / X is open for every sW U ! E

1 1
, fU;s . .W // X is open for every sW U ! E
1
, .W / E 0 is open:
It is straightforward to check that this isomorphism is natural.
As the sheafification of sheaf is the sheaf itself we deduce from Lemma 3.37 and
Lemma 3.38:
Proposition 3.39. Let X be a topological space X. The functors F and G yield an equiv-
alence between the category (Ét/X) of étalé spaces over X and the category (Sh(X)) of
sheaves on X.
Example 3.40 (Étalé Spaces of constant sheaves). Let E be a set that we also consider
as a discrete topological space. Let EX be the constant sheaf with values in E on a topo-
logical space X (Example 3.31). Then the corresponding étalé space is .X E; pr1 /
because for U X open the sections of pr1 over U are just the maps x 7! .x; s.x//,
where sW U ! E is locally constant.
Note that the map pr1 W X E ! X is a trivial covering map. More generally, every
covering map is an étalé space that is locally on X a trivial covering map (Remark 2.28).
Hence we obtain the following result.
Definition and Proposition 3.41 (Locally constant sheaves). A sheaf F on X is called

locally constant if there exists an open covering .Ui /i of X such that FjUi is a constant
sheaf.
The equivalence of (Sh(X)) and (Ét/X) yields an equivalence between the full sub-
category of locally constant sheaves on X and the category of covering spaces of X
(Definition 2.27).
Example 3.42. Let L be the sheaf of complex logarithms L on C n f0g,
L.U / WD f lW U ! C holomorphic I exp ı l D idU g ; U C n f0g open:
For every simply connected open subspace U of C n f0g the choice of a logarithm l0 on
U yields an isomorphism of sheaves of abelian groups .2 iZ/U Š LjU : One attaches to
a locally constant function t with values in 2 iZ on an open subset V of U the logarithm
l0jV C t. Hence L is a locally constant sheaf of abelian groups. But it is not constant
3.5 Direct and Inverse Images of Sheaves 57
because L.C n f0g/ D ;. The associated étalé space to L is the covering map expW C !
C n f0g.
Suppose that X is path connected, locally path connected, and semilocally simply con-
nected (Problem 2.19). This will for instance be the case if X is the underlying topological
space of a connected premanifold (see Sect. 4.2). Then the functor ˚x (2.3) is for all
x 2 X an equivalence of categories (Problem 2.20 and Problem 2.21). Hence we see that
the above equivalence also yields an equivalence between the category of locally constant
sheaves and the category of sets endowed with a right 1 .X; x/-action.
There is an analogous equivalence for sheaves of R-modules (R some ring), see Prob-
lem 3.14.
3.5 Direct and Inverse Images of Sheaves
In this section f W X ! Y denotes a continuous map of topological spaces. We will now

see how to use f in order to attach to a sheaf on X a sheaf on Y (direct image) and to
a sheaf on Y a sheaf an X (inverse image). We start with the direct image.
Definition 3.43 (Direct image of a sheaf). Let f W X ! Y be a continuous map. Let F

be a presheaf on X. We define a presheaf f F on Y by (for V Y open)
.f F/.V / D F.f 1 .V //
the restriction maps given by the restriction maps for F. Then f F is called the direct
image of F under f . Whenever 'W F1 ! F2 is a morphism of presheaves, the family of
maps f .'/V WD 'f 1 .V / for V Y open is a morphism f .'/W f F1 ! f F2 . We obtain
a functor f from the category of presheaves on X to the category of presheaves on Y .
Example 3.44 (Direct image of constant sheaves). Let pW X ! Y be a continuous map.

Let E be a set and let EX and EY be the sheaf of locally constant E-valued functions
on X and Y respectively. For V Y open and for locally constant map gW V ! E the
composition g ı pW p 1 .V / ! E is locally constant. Hence we obtain a morphism of
sheaves
'W EY ! p EX :
Now suppose that p is surjective, that Y has the quotient topology of X, and that p has
connected fibers (for instance if pW Y I ! Y is the projection for a connected space
I ). For V Y open a locally constant map hW p 1 .V / ! E is the same as a continuous
map if we endow E with the discrete topology. The restriction of h to the fibers of p is
constant and hence by the universal property of the quotient topology there exists a unique
continuous map gW V ! E such that g ı p D h. Hence we see that ' is an isomorphism
in this case.
58 3 Sheaves
The following properties are immediate.
Remark 3.45.
1. If F is a sheaf on X, f F is a sheaf on Y . Therefore f also defines a functor

f W .Sh.X// ! .Sh.Y //.
2. If gW Y ! Z is a second continuous map, there exists an identity g .f F/ D .gıf / F
that is functorial in F.
We now define the inverse image of a sheaf.
Definition 3.46 (Inverse image of a sheaf). Let f W X ! Y be a continuous map and let
G be a presheaf on Y . Define a presheaf f C G on X by
U 7! lim G.V /; (3.8)

!
V f .U /;
V Y open
the restriction maps being induced by the restriction maps of G. Let f 1 G be the sheafifi-
cation of f C G. We call f 1 G the inverse image of G under f .
If f is the inclusion of a subspace X of Y , we also write GjX instead of f 1 G and we
write G.X/ WD .f 1 .G//.X/.
Note that even if G is a sheaf, f C G is not a sheaf in general.
Remark 3.47. Let f W X ! Y be a continuous map of topological spaces. Let G be

a presheaf on Y . The construction of f C G and hence of f 1 G is functorial in G. Therefore
we obtain a functor
f 1 W (PSh(Y )) ! (Sh(X)):
Remark 3.48. Let f W X ! Y be an open continuous map and G a presheaf on Y . Then

for U X open one has f C G.U / D G.f .U //. In this case, if G is a sheaf, f C G is
a sheaf and hence f C G D f 1 G.
In particular if f is the inclusion of an open subspace U D X of Y . Then for every
sheaf G on Y and V U open
.GjU /.V / D G.V /:

Direct image and inverse image are functors that are adjoint to each other. More precisely:
Proposition 3.49. Let f W X ! Y be a continuous map of topological spaces, let F be

a sheaf on X and let G be a presheaf on Y . Then there is a bijection
Hom.Sh.X // .f 1 G; F/ $ Hom.PreSh.Y // .G; f F/;

' 7! ' [ ; (3.9)
]
7 ;
which is functorial in F and G.
Proof. Let 'W f 1 G ! F be a morphism of sheaves on X, and let t 2 G.V /, V Y

open. Since f .f 1 .V // V , we have a map G.V / ! f C G.f 1 .V //, and we define
'V[ .t/ as the image of t under the map
'f 1 .V /
G.V / ! f C G.f 1 .V // ! f 1 G.f 1 .V // ! F.f 1 .V // D f F.V /:
Conversely, let W G ! f F be a morphism of sheaves on Y . To define the morphism

]
it suffices to define a morphism of presheaves f C G ! F, which we call again ] .
Let U be open in X, and s 2 f C G.U /. If V is some open neighborhood of f .U /, U is
contained in f 1 .V /. Let V be such a neighborhood such that there exists sV 2 G.V /
representing s. Then V .sV / 2 f F.V / D F.f 1 .V //. Let U .s/ 2 F.U / be the
]
restriction of the section V .sV / to U .

Clearly, these two maps are inverse to each other. Moreover, it is straightforward –
albeit quite cumbersome – to check that the constructed maps are functorial in F and G.

Example 3.50. Let X be a topological space, let iW A ,! X be the inclusion of a closed

subspace and let F be a sheaf on A. As X n A is open in X, we have
8
<F ; if x 2 AI
x
i .F/x D (3.10)
:f g; if x 2 X n A:
Via (3.9) there corresponds to idi F a morphism of sheaves i 1 .i F/ ! F. Looking at

stalks, (3.10) and (3.12) shows that this is an isomorphism.
Remark 3.51 (Inverse image and composition). Let f W X ! Y , gW Y ! Z be contin-

uous maps.
1. Let iW (Sh(Z)) ! .PreSh.Z// be the inclusion functor. By Proposition 3.49 the functor
f W (Sh(X)) ! (Sh(Y )) is right adjoint to f 1 W (Sh(Y )) ! (Sh(X)), the functor i ı
60 3 Sheaves
g W (Sh(Y )) ! .PreSh.Z// is right adjoint to g 1 W .PreSh.Z// ! (Sh(Y )), and the

functor i ı .g ı f / is right adjoint to .g ı f /1 . Hence the isomorphism of functors
i ı .g ı f / Š .i ı g / ı f (Remark 3.45) yields an isomorphism of their left adjoints
.g ı f /1 Š f 1 ı g 1 : (3.11)
2. If x is a point of X and iW fxg ! X is the inclusion, the definition (3.8) shows that
i 1 F D Fx
for every presheaf F on X (more precisely: i 1 .F/.fxg/ D Fx ).

3. In particular, (3.11) yields for each presheaf G on Y a functorial isomorphism
.f 1 G/x Š Gf .x/ : (3.12)
The identification (3.11) and (3.12) will also follow immediately from the description
of the inverse image in terms of étalé spaces (Proposition 3.55).
Remark 3.52. We will almost never use the concrete description of f 1 G in the sequel.
Very often we are given f , F, and G as in the Proposition 3.49, and a morphism of sheaves
W G ! f F. Then usually it will be sufficient to understand for each x 2 X the map
(3.12)
x W Gf .x/ D
]
.f 1 G/x ! Fx
induced by ] W f 1 G ! F on stalks. The proof of Proposition 3.49 shows that we can

describe this map in terms of as follows. For every open neighborhood V Y of f .x/,
we have maps
G.V / ! F.f 1 .V // ! Fx ;
V
]
and taking the colimit over all V we obtain the map x W Gf .x/ ! Fx .
Remark 3.53. Note that if F is a sheaf of rings (or of R-modules, or of A-algebras) on

X, f F is a sheaf on Y with values in the same category. A similar statements holds for
the inverse image. Finally, Proposition 3.49 holds (with the same proof) if we consider
morphisms of sheaves of rings (or of R-modules, etc.).
We have already seen that there is a natural correspondence between sheaves and étalé
space and that it is possible to describe the sheafification of a presheaf in terms of associ-
ated étalé spaces. We will now show that the formation of the inverse image of a presheaf
has a simple description in terms of étalé spaces: The corresponding étalé space is given
by the fiber product. Hence let us consider continuous maps f W X 0 ! X and W E ! X
of topological spaces. We form the fiber product (Appendix Definition 12.28) and obtain
the following commutative diagram, where gW E X X 0 ! E and 0 W E X X 0 ! X 0 are

the projections:
0
E X X 0 X0
g f

E X:
For x 0 2 X 0 the map g induces a homeomorphism 01 .x 0 / ! 1 .f .x 0 //. The inverse

is given by e 7! .e; x 0 /.
Lemma 3.54. Suppose that has one of the following properties:
1. homeomorphism,
2. open topological embedding, or
3. local homeomorphism.
Then 0 has the same property.
Proof. Assertion 1 is clear because 7! 0 is functorial: If $ is a continuous inverse

of , then $ 0 W x 0 7! .$ .x/; x 0 / is a continuous inverse of 0 . If induces a homeomor-
phism E ! U for U X open, then 0 induces a homeomorphism E 0 ! f 1 .U / by
1. This shows 2. Finally, if there exists an open covering .Wi /i of E such that jUi is an
open embedding for all i, then .g 1 .Wi //i is an open covering of E 0 and 0 j g1 .Wi / is an
open embedding by 2. This proves 3.
The fiber product construction above yields a functor (Ét/X) ! (Ét/X 0 ) by sending a mor-
phism f W E1 ! E2 of étalé spaces over X to the map E1 X X 0 ! E2 X X 0 induced by
f idX 0 .
Proposition 3.55 (Inverse image via étalé spaces). Let f W X ! Y be a continuous map
of topological spaces, E a presheaf on Y and .E; / the étalé space over Y associated
to E. The functor that sends E to the sheaf associated to the étalé space .E Y X; 0 / is
naturally isomorphic to the inverse image functor f 1 .
Proof. Denote by E0 WD F.E; / and F WD F.E Y X; 0 / the associated sheaves on Y

and X respectively.
For U X we have F.U / D f sW U ! E Y X I s section of 0 g and
f C E0 .U / D f .V; s/ I V Y open with f .U / V , s 2 E0 .V /g = ;
where .V; s/ .V 0 ; s 0 / if there exists W V 0 \ V open with f .U / W such that

s jW D s jW . Now let ŒV; s 2 f C E0 .U /. Then ı .s ı f /jU D f ı i and we obtain
0
62 3 Sheaves
a continuous map sQ W U ! E Y X; x 7! .s.f .x//; x/. Then sQ is a section of 0 over U ,

i.e., sQ 2 F.U /. Notice that the construction of sQ does only depend on the values of s on
f .U / and is hence independent of the choice of .V; s/ within its equivalence class. We
get a morphism of presheaves !W f C E0 ! F by f C E0 .U / ! F.U /; s 7! sQ .
As E0 is the sheafification of E (Lemma 3.37) there is a morphism of presheaves
W E ! E0 that is bijective on stalks. We claim that the morphism of presheaves ! ı
f C . /W f C E ! F is also bijective on stalks. Let x 2 X, y WD f .x/, x 2 U X, V Y
open with f .U / V and s 2 E.V /. By Lemma 3.35, the map x W Fx ! .E Y X/x ,
s 7! s.x/ is bijective. Furthermore the map fx0 W .E Y X/x ! Ey ; .e; x/ 7! e is bijective
as well and by construction we have Ey D Ey , so that fx0 ı x W Fx ! Ey is bijective. Now
Q /x
.! ı f C . //x .sy / D !y ..fV;s /y / D .fV;s
and
.fx0 ı Q
x /..fV;s /x /
Q .x// D f 0 ..fV;s .y/; x// D fV;s .y/ D sy :
D fx0 .fV;s x
Therefore .! ı f C . //x and .fx0 ı x / are inverse to each other.

Since f 1 E is the sheafification of f C E and ! ı f C . / is bijective on stalks, it induces
an isomorphism f 1 E ! F. By construction it is functorial.
Example 3.56 (Inverse image of constant sheaves). Let f W X ! Y be a continuous

map. Let E be a set, let EY be the sheaf of locally constant E-valued functions on Y . The
corresponding étalé space is the projection E Y ! Y , where we consider E as a discrete
topological space. Then the projection .E Y / Y X ! E X is a homeomorphism
compatible with the projections to X (an inverse is given by .e; x/ 7! ..e; f .x//; x/).
Hence f 1 EY D EX .
Remark and Definition 3.57 (Pullback of sections). Let f W X ! Y be a continuous

map, let G be a sheaf on Y , and let t 2 G.V /, V Y open. Let W G ! Y be the étalé
space corresponding to G and consider t as a continuous section tW V ! G of . Then
f 1 .t/W f 1 .V / ! G Y X; x 7! .t.f .x//; x/
is a continuous section of G Y X ! X. Hence we obtain a pullback map
f 1 W G.V / 7! .f 1 G/.f 1 .V //; (3.13)
which is functorial in G and compatible with restrictions to smaller open subsets of Y .
Example 3.58. Let X be a topological space, let Z be a subspace of X, and denote by

iW Z ! X the inclusion. If E is a sheaf on X with corresponding étalé space W E ! X,
then the étalé space corresponding to FjZ is the usual restriction j 1 .Z/ W 1 .Z/ !
3.6 Limits and Colimits of Sheaves 63
Z. The pullback i 1 .s/ of a continuous section sW X ! E of over X is simply the

restriction s jZ and we usually write s jZ instead of i 1 .s/.
An application is a simple proof of the following fact (Proposition 3.59), which general-
izes the sheaf property from open coverings to more general coverings.
Proposition 3.59. Let .Ai /i 2I be a covering of a topological space X satisfying one of

the following hypotheses:
1. The interiors Aıi of Ai cover X.

2. The covering .Ai /i is locally finite and Ai is closed in X for all i.
Let E be a sheaf on X and let si 2 E.Ai / be sections such that si jAi \Aj D sj jAi \Aj for all
i; j 2 I . Then there exists a unique section s 2 E.X/ such that s jAi D si for all i 2 I .
Proof. Let W E ! X be the étalé space corresponding to E. Then si is a continuous

section of over Ai . By hypothesis, there exists a unique map sW X ! E with s jAi D si .
Hence s is a section of . Moreover, s is continuous by Appendix Corollary 12.34 and
hence defines an element of E.X/.
3.6 Limits and Colimits of Sheaves
We fix a topological space X. We will describe limits and colimits of sheaves and show in
particular that the category (Sh(X)) is complete and cocomplete. Let I be a small category
and let I 3 i 7! Fi be an I -diagram of sheaves on X.
Remark 3.60 (Limits of sheaves). Using the explicit construction of limits in the cate-
gory of sets (Appendix Example 13.33), one sees that U 7! limi Fi .U /, U X open,
defines a sheaf of sets. It is a limit in the category of presheaves on X and in particular in
the category of all sheaves on X.
Q
As special cases we obtain for a family of sheaves .Fi /i the product of sheaves i Fi
Q
given by U 7! i Fi .U /.
Remark 3.61 (Limits and stalks). Let x 2 X be a point. The maps Fi .U / ! .Fi /x yield
maps limi Fi .U / ! limi .Fi /x . Taking the (filtered) colimit over the open neighborhoods
of x we obtain a map
.lim Fi /x ! lim.Fi /x : (3.14)
i i
As filtered colimits commute with finite limits (Appendix Proposition 13.39), we deduce
that (3.14) is an isomorphism if I is finite. In general, (3.14) is not bijective (Prob-
lem 3.15).
64 3 Sheaves
Remark 3.62 (Colimits of sheaves). The map U 7! colimi Fi .U /, U X open, is

a colimit of the I -diagram i 7! Fi in the category of presheaves on X. Hence the universal
property of the sheafification (Proposition 3.27) shows that the sheafification of U 7!
colimi Fi .U / is a colimit in the category of sheaves on X.
Remark 3.63 (Colimits and stalks). Let x 2 X be a point. The maps Fi .U / !

.Fi /x yield maps colimi Fi .U / ! colimi .Fi /x . As colimits commute with each other
(Appendix Remark 13.36) and sheafification does not change the stalks, we obtain an
isomorphism of sets
.colim Fi /x ! colim.Fi /x : (3.15)

i i
Remark 3.64. Let f W X ! Y be a continuous map of topological spaces. Let I be a

(small) category.
1. Let i 7! Fi be an I -diagram of sheaves on X. As the direct image functor

f W (Sh(X)) ! (Sh(Y )) has a left adjoint (namely the inverse image functor f 1 ), it
commutes with arbitrary limits (Appendix Proposition 13.47) and hence we obtain an
isomorphism f .limi Fi / ! limi f Fi of sheaves on Y . In particular, f is left exact
(Appendix Definition 13.48).
2. Dually, let i 7! Gi be an I -diagram of sheaves on Y . Then as f 1 has a right ad-
joint, we have an isomorphism colimi f 1 Gi ! f 1 .colimi Gi / of sheaves on X. In
particular, f 1 is right exact (Appendix Definition 13.48).
Remark 3.65 (Limits and colimits of sheaves in other categories). Instead of sheaves
with values in the category of sets we can also consider sheaves with values in the category
of groups or the category of R-modules (R a fixed ring). These categories are complete
and cocomplete and the forgetful functor to the category of sets preserves limits. Hence
the same constructions as above yield limits or colimits of sheaves with values in these
categories.
We can also consider the category of commutative rings or the category of R-algebras
(R a fixed commutative ring). These categories are complete and in them exist arbitrary
filtered colimits. Again the forgetful functor to the category of sets preserves limits. Hence
the constructions above yield limits and filtered colimits of sheaves in these categories.
3.7 Problems
Problem 3.1. Let F be the presheaf of bounded continuous functions on Rn (n 2 N) with

values in R. Show that F is not a sheaf. Show that its sheafification is the sheaf CRn IR of
continuous R-valued functions.
3.7 Problems 65
Problem 3.2. Let X be a topological space and let BX be the Borel- -algebra of X (i.e.,
the -algebra generated by the open subsets of X). For U X open let MX .U / be the set
of measurable functions with value in R. Show that MX is a presheaf of functions, which
is a sheaf if X is a Lindelöf space.
Problem 3.3. For U Rn open, n 2 N, let L1 .U / be the L1 -space of R-valued

Lebesgue integrable functions U ! R modulo the space of functions f such that jf j D 0
almost everywhere.
1. Show that U 7! L1 .U / with the usual restriction map is a presheaf of R-vector spaces
on Rn , which is not a sheaf.
2. Show that its sheafification is the sheaf U 7! L1loc .U /, where L1locR.U / is the quotient
of the space of Lebesgue measurable functions f W U ! R with K jf j dx < 1 for
each compact subspace K U by the subspace of functions f such that jf j D 0
almost everywhere.
Problem 3.4. Let X be a topological space and let F be a presheaf. A subpresheaf of F

is a presheaf G such that G.U / F.U / for all U X open such that for all open subsets
V U X the restriction maps G.U / ! G.V / are induced by the restriction maps for
F.
Suppose that F is a sheaf. Show that a subpresheaf G of F is a sheaf if and only if for
every open subset U of X, for every open covering .Ui /i of U and for every s 2 F.U /
with s jUi 2 G.Ui / one has s 2 G.U /.
Problem 3.5. Let X be a topological space and define a presheaf ˝ on X by ˝.U / WD

f V U I V X openg. For U 0 U the restriction ˝.U / ! ˝.U 0 / is defined by
V 7! V \ U 0 .
1. Show that ˝ is a sheaf.

2. Show that ˝ classifies subsheaves in the following sense. Let F be a sheaf on X and
let ˚W F ! ˝ be a morphism of sheaves. Then U 7! f s 2 F.U / I ˚U .s/ D U g is
a subsheaf (Problem 3.4) G˚ of F and one obtains a bijection, functorial in F,
Hom(Sh(X )) .F; ˝/ $ fsubsheaves of Fg; ˚ 7! G˚ :
Problem 3.6. Let X be a topological space, x 2 X and denote by ix W fxg ! X the

inclusion. For every set E (considered as a sheaf on fxg) we call .ix / .E/ the skyscraper
sheaf in x with value E.
1. Show that E 7! .ix / .E/ defines a functor (Sets) ! .Sh.X//, which is right adjoint
to the functor .Sh.X// ! (Sets) that sends a sheaf F to its stalk Fx in x.
2. Show that for the stalks of the skyscraper sheaf one has .ix / .E/y D E for y 2 fxg
and that .ix / .E/y is a singleton for y … fxg.
66 3 Sheaves
Problem 3.7. Let 'W F ! G be a morphism of sheaves on a topological space X. Show

that the following assertions are equivalent:
(i) ' is a monomorphism (respectively an epimorphism, respectively an isomorphism)

in the category of sheaves on X.
(ii) ' is injective (respectively surjective, respectively bijective).
Hint: To show “' epimorphism ) ' surjective” use Problem 3.6.
Problem 3.8. Let X be a topological space. Show that the functor from the category of
sets to the category of sheaves on X that sends a set E to the constant sheaf EX is left
adjoint to the functor F 7! F.X/.
Problem 3.9. Let G be a sheaf, let f W X ! Y be a continuous map of topological spaces,

and let U X be an open subset. Show that .f 1 G/.U / can be described as the set
Q
of s D .sx /x2U 2 x2U Gf .x/ such that for all x 2 U the following condition holds:
There exist x 2 W U open, V Y open with f .W / V , and t 2 G.V / such that
tf .w/ D sf .w/ for all w 2 W .
Problem 3.10. Let f W X ! Y be a continuous map. Show that the functor f C W

(PSh(Y )) ! (PSh(X)) is left adjoint to the functor f W (PSh(X)) ! (PSh(Y )).
Problem 3.11. Let f W X ! Y be a continuous map.
1. Let G be a sheaf on Y and let s 2 G.Y /. Show that supp.f 1 .s// D f 1 .supp.s//.
2. Let F be a sheaf on X. Then a section sQ 2 .f F/.Y / is the same as a section s 2 F.X/.
Show that supp.Qs / D f .supp s/.
Problem 3.12. Let X be a topological space. One says that an abelian sheaf F on X
satisfies the principle of unique continuation if every section of F over any open subset of
X has open support.
1. Let U X be open and connected and s; t 2 F.U /. Suppose that sx D tx for one
point x 2 U . Show that s D t.
2. Let gW Z ! X be a continuous map. Show that g 1 F also satisfies the principle of
unique continuation.
3. Show that every locally constant sheaf satisfies the principle of unique continuation.
4. Let X be an open subset of a finite-dimensional K-vector space. Show that the sheaf
of analytic functions on X satisfies the principle of unique continuation.
Hint: Appendix Problem 16.3.
3.7 Problems 67
Problem 3.13. Let X be a topological space, let F be an abelian sheaf on X, and let
.E; / be the corresponding étalé space over X. Show that F satisfies the principle of
unique continuation (Problem 3.12) if and only if is separated.
Problem 3.14. Let X be a path connected, locally path connected, semilocally simply
connected (Problem 2.19) space so that there exists a universal covering of X (Prob-
lem 2.20). Let R be a ring. A local system of R-modules is a locally constant sheaf of
R-modules. Choose x0 2 X. Show that the following categories are equivalent:
(i) The category of local systems of R-modules on X.

(ii) The category of functors ˘.X/ ! (R-Mod) (here ˘.X/ is the fundamental groupoid
of X, see Problem 2.9).
(iii) The category of representations of 1 .X; x0 / on R-modules.
Problem 3.15. Let .Fi /i 2I be a family of sheaves on X and let x 2 X.

Q Q
1. Show that the map (3.14) . i 2I Fi /x ! i 2I Fi;x is injective.
2. Let X D R, x D 0 and CR be the sheaf of continuous R-valued functions on R. For
n 2 N let fn 2 CR;0 be the germ of the function x 7! 1=.x .1=n//. Show that
Q Q
.fn /n2N is not in the image of . n CR /0 ! n CR;0 .
Problem 3.16. Let f W X ! Y be a continuous map and let F be a sheaf on X. Define

the proper direct image fŠ F as the sheaf on Y with (V Y open)
˚
.fŠ F/.V / WD s 2 F.f 1 .V // I f W supp.s/ ! V is proper :
1. Show that fŠ F is a subsheaf of f F and that F 7! fŠ F yields a functor fŠ W (Sh(X)) !

(Sh(Y )). Show that f D fŠ if f is proper.
Hint: Problem 1.18.
2. Let gW Y ! Z be a second continuous map. Show that the identification .g ı f / D
g ı f of functors (Sh(X)) ! (Sh(Z)) induces .g ı f /Š D gŠ ı fŠ .
Hint: Use Problem 3.11.
Problem 3.17. Let Z X be a locally closed subspace of a topological space X, let

iW Z ! X be the inclusion, and let iŠ W (Sh(Z)) ! (Sh(X)) be the proper direct image
functor (Problem 3.16).
1. Let F be a sheaf on Z. Show that for x 2 X one has

8
<F ; if x 2 ZI
x
iŠ .F/x D
:f g; if x 2 X n Z:
Show that iŠ yields an equivalence between (Sh(Z)) and the full subcategory of
(Sh(X)) of sheaves F such that Fx D f g for all x 2 X n Z. The quasi-inverse functor
is induced by i 1 . Deduce that iŠ W (Sh(Z)) ! (Sh(X)) is left adjoint to i 1 .
68 3 Sheaves
2. For a sheaf F on X define for U X open
FZ .U / D f s 2 F.U / I supp.s/ Z g :
Show that FZ is a subsheaf of F and show that F 7! i Š F WD i 1 FZ yields a functor

(Sh(X)) ! (Sh(Z)). Show that i Š D i 1 if Z is open in X.
3. Show that i Š is right adjoint to iŠ .
Problem 3.18. Let X be a topological space, U X open, Z WD X n U . Let F be an

abelian sheaf on X. Let j W U ! X and iW Z ! X be the inclusions. Show that there is
an exact sequence of abelian sheaves
0 ! jŠ j 1 F ! F ! i i 1 F ! 0:
Hint: Problem 3.17.

Manifolds
4
In this chapter we define a real (respectively complex) premanifold as a space together

with a sheaf of functions that locally resembles Rn (respectively C n ) for some n to-
gether with its sheaf of C ˛ -functions (˛ 2 N0 [ f1; !g fixed) (respective of holomorphic
functions). Hence we first define the abstract notion of a space together with a sheaf of
R-algebras for some ring R. These are called R-ringed spaces. Then a real (respectively
complex) premanifold will be a special case of an R-ringed (respectively a C-ringed)
space. A (real or complex) manifold is defined to be a premanifold whose underlying
topological space is Hausdorff and second countable. This implies several other nice
topological properties. In the last section we explain simple constructions for how to get
new (pre)manifolds from old ones. The most important constructions will be the product
and the gluing of (pre)manifolds.
4.1 Ringed Spaces
Ringed spaces formalize the idea of giving a geometric object by specifying its underlying
topological space and the “functions” on all open subsets of this space. This is motivated
by the observation that to say that a continuous map F W U ! Rn (U Rm open) is a C ˛ -
map if and only if for every C ˛ -function f W Rn ! R the composition f ı F W U ! R is
a C ˛ -map (in fact it suffices to take for f the projections to the coordinates). These func-
tions usually form some commutative algebra over some ring1 . For instance the R-valued
C ˛ -functions on an open subset of Rm form a commutative R-algebra. An immediate
observation is that because of continuity the germs of such functions f at a point x either
satisfy f .x/ D 0 or 1=f exists in some open neighborhood of x. In other words, the stalk
1
Notable exceptions are for instance algebras of differential operators, which are almost never com-
mutative.

DOI 10.1007/978-3-658-10633-1_4
70 4 Manifolds
of the sheaf of functions in x is a local ring whose maximal ideal consists of germs f with
f .x/ D 0. This leads to the notion of a locally ringed space.
The notion of a (locally) ringed space also gives us the flexibility to consider very
general notions of “functions”: Any sheaf of rings will do. It is not necessarily a sheaf of
functions. This is very useful in other geometric theories such as in the theory of complex
analytic spaces or in algebraic geometry. But in this book, for almost all ringed spaces the
underlying sheaf of rings will be a sheaf of functions.
We fix a commutative ring R. In the sequel, R will usually be the field R of real numbers
or the field C of complex numbers. But occasionally we will consider other cases, for
example R D Z.
Definition 4.1 ((Locally) ringed spaces).
1. An R-ringed space is a pair .X; OX /, where X is a topological space and where OX is

a sheaf of commutative R-algebras on X. The sheaf of rings OX is called the structure
sheaf of .X; OX /.
2. A locally R-ringed space is an R-ringed space .X; OX / such that the stalk OX;x is
a local ring for all x 2 X. We then denote by mx the maximal ideal of OX;x and by
.x/ WD OX;x =mx its residue field.
As every ring has a unique structure as Z-algebra, we simply say (locally) ringed space
instead of (locally) Z-ringed space. Usually we will denote a (locally) R-ringed space
.X; OX / simply by X.
b 0 WD
Our principle example will be sheaves of real-valued C ˛ -functions (with ˛ 2 N
N0 [ f1; !g) or holomorphic functions. Basic notions and properties of such functions
are recalled in Appendix 16.
Example 4.2. Recall that K denotes the field of real or of complex numbers. Let X be
b 0 . We denote by C˛ the
an open subset of a finite-dimensional K-vector space. Let ˛ 2 N X
˛
sheaf of C -functions, i.e.,
CX˛ .U / D f f W U ! K I f is C ˛ -functiong ; U X open.
Then CX˛ is a sheaf of K-algebras.

For ˛ D 0 this is the sheaf of continuous K-valued functions on X. If K D C and
˛ 1, then CX˛ is the sheaf of holomorphic functions on X and we will denote this sheaf
usually by OX or OXhol . If K D R and 1 ˛ 1, then CX˛ is the sheaf of R-valued
˛-fold continuously differentiable functions. If K D R and ˛ D !, then CX! is the sheaf
of R-valued real analytic functions on X.
4.1 Ringed Spaces 71
The same argument as for sheaves of continuous functions (Remark 3.17) yields the
following observation: For all x 2 X the stalk CX;x
˛
is a local ring. In particular .X; CX˛ / is
a locally K-ringed space. The maximal ideal mx of CX;x
˛
consists of germs of C ˛ -functions
with f .x/ D 0. The evaluation map f 7! f .x/ induces an isomorphism of K-algebras
.x/ WD CX;x
˛
=mx ! K, where we consider K as K-algebra via idK .
Definition and Remark 4.3 (Morphisms of (locally) ringed spaces). Let X D .X; OX /
and Y D .Y; OY / be R-ringed spaces. A morphism of R-ringed spaces X ! Y is a pair
.f; f [ /, where f W X ! Y is a continuous map of the underlying topological spaces and
where f [ W OY ! f OX is a homomorphism of sheaves of R-algebras on Y .
The datum of f [ is equivalent to the datum of a homomorphism of sheaves of R-
algebras f ] W f 1 OY ! OX on X by Proposition 3.49. Usually we simply write f instead
of .f; f ] / or .f; f [ /.
Morphisms of locally ringed spaces have to satisfy an additional property. To state this
property, observe that a morphism f W X ! Y of R-ringed spaces induces morphisms on
the stalks as follows. Let x 2 X. Using the identification .f 1 OY /x D OY;f .x/ established
in (3.12), we get a homomorphism of R-algebrasF17@Fx , stalk of a sheaf
fx WD .f ] /x W OY;f .x/ ! OX;x :
By Remark 3.52 there is the following more explicit description of this homomorphism:
For U an open neighborhood of f .x/ one has a map
fU[
OY .U / ! OX .f 1 .U // ! OX;x :
These maps induce the map an stalks fx W OY;f .x/ D colim OY .U / ! OX;x .
Now let X and Y be locally R-ringed spaces. We define a morphism of locally R-
ringed spaces X ! Y to be a morphism .f; f [ / of ringed spaces such that the homomor-
] ]
phism of local rings fx W OY;f .x/ ! OX;x is local (i.e., fx .mf .x/ / mx ).
In general there exist locally ringed spaces and morphisms of ringed spaces between them
that are not morphisms of locally ringed spaces (Problem 4.1). For spaces with functions
of C ˛ -functions such as the premanifolds defined below in Definition 4.13 we will see that
every morphism of ringed spaces is automatically a morphism of locally ringed spaces (see
Example 4.5).
Remark 4.4. The composition of morphisms of (locally) R-ringed spaces is defined in

the obvious way using the compatibility of direct images with composition (i.e., .g ı
f / D g ı f , see Remark 3.45 (2)). We obtain the category of (locally) R-ringed
spaces.
In general, f [ (or f ] ) is an additional datum for a morphism. For instance it might

happen that f is the identity but f [ is not an isomorphism of sheaves (Problem 4.5). We
72 4 Manifolds
will usually encounter the simpler case that the structure sheaf is a sheaf of functions on
open subsets of X and that f [ is given by composition with f . The following special case
and its globalization (Proposition 4.18) is the main example.
Example 4.5 (Principal example). Let X V and Y W be open subsets of finite-

b 0.
dimensional K-vector spaces V and W . Let ˛ 2 N
Every C ˛ -map f W X ! Y defines by composition a morphism of locally K-ringed
spaces .f; f [ /W .X; CX˛ / ! .Y; C˛Y / by
fU[ W C˛Y .U / ! f .CX˛ /.U / D CX˛ .f 1 .U //; t 7! t ı f
for U Y open.
The induced map on stalks fx W C˛Y;f .x/ ! CX;x ˛
is then also given by composing a K-
˛
valued C -function t, defined in some neighborhood of f .x/, with f , which yields a
K-valued C ˛ -function t ı f defined in some neighborhood of x.
Conversely, let .f; f [ /W .X; CX˛ / ! .Y; C˛Y / be any morphism of K-ringed spaces. We
claim:
1. .f; f [ / is automatically a morphism of locally K-ringed spaces.

2. For all U Y open the R-algebra homomorphism fU[ W C˛Y .U / ! CX˛ .f 1 .U // is au-
tomatically given by the map t 7! t ı f . Note that then f is a C ˛ -map (choose a basis
of W ; considering for t projections to the coordinates shows that each component of
f is a C ˛ -map).
]
To show 1 let x 2 X. Set ' WD fx , B WD CX;x ˛
, and A WD C˛Y;f .x/ . Then 'W A ! B is
a homomorphism of local K-algebras such that A=mA D K and B=mB D K. We claim
that ' is automatically local, equivalently that ' 1 .mB / is a maximal ideal of A. Indeed,
' induces an injective homomorphism of K-algebras
A=' 1 .mB / ,! B=mB D K:
As a homomorphism of K-algebras, it is automatically surjective, hence A=' 1 .mB / Š K

is a field and hence ' 1 .mB / is a maximal ideal.
Let us show 2. Let U Y be open and x 2 f 1 .U /. Consider the commutative
diagram of K-algebra homomorphisms
fU[
C˛Y .U / CX˛ .f 1 .U //
t 7!tf .x/ s7!sx
]
fx
C˛Y;f .x/ CX;x
˛
evf .x/ Wt 7!t .f .x// evx Ws7!s.x/
K K:
4.1 Ringed Spaces 73
The evaluation maps are surjective. Hence there exists a homomorphism of K-algebras
]
W K ! K making the lower rectangle commutative if and only if one has fx .ker.evf .x/ //
]
ker.evx /. But this latter condition is satisfied because fx is local by 1. Moreover, as
a homomorphism of K-algebras, one must have D idK . Therefore we find fU[ .t/.x/ D
t.f .x//, which shows (b).
Remark 4.6. A morphism f W X ! Y of R-ringed spaces is an isomorphism in the

category of R-ringed spaces if and only if f is a homeomorphism and fx W OY;f .x/ ! OX;x
is an isomorphism of R-algebras for all x 2 X.
Indeed, .f; f [ / is an isomorphism if and only if f is a homeomorphism and f [ is an
isomorphism of sheaves of rings. And the description of fx in Remark 4.3 shows that if
f is a homeomorphism, then f [ is an isomorphism if and only if fx is an isomorphism
for all x 2 X.
Next we define the notion of an open embedding of ringed spaces and of a local iso-
morphism of ringed spaces.
Remark and Definition 4.7. Let X be a (locally) R-ringed space and let U X be
open. Then .U; OX jU / is a (locally) R-ringed space, which we usually denote simply by
U . Such a (locally) R-ringed space is called an open subspace of X. There is an inclusion
morphism iW U ! X of (locally) R-ringed spaces, where the continuous map iW U ! X is
the inclusion of the underlying topological spaces and where i [ is given by the restriction
OX .V / ! i .OX jU /.V / D OX .U \ V / for all V X open. Then i ] W i 1 OX ! OX jU is
the identity. In particular ix is the identity for all x 2 U .
For a morphism f W X ! Y of (locally) R-ringed spaces we denote by f jU W U ! Y
the composition f ı i of morphisms of (locally) ringed spaces.
Finally, a morphism j W .Z; OZ / ! .X; OX / of (locally) R-ringed spaces is called
open embedding if U WD j.Z/ is open in X and j induces an isomorphism .Z; OZ / !
.U; OX jU /.
Definition 4.8. A morphism f W X ! Y of (locally) R-ringed spaces is called a local

isomorphism if there exists an open covering .Ui /i of X such that f jUi W Ui ! Y is an
open embedding for all i.
In other words, a morphism f is a local isomorphism if there exists an open covering

.Ui /i of X and for all i 2 I an open subspace Vi of Y such that f induces for all i an
isomorphism Ui ! Vi of ringed spaces.
Remark 4.9. Remark 4.6 shows that a morphism f W X ! Y of R-ringed spaces is a local
isomorphism if and only if f is a local homeomorphism and fx W OY;f .x/ ! OX;x is an
isomorphism.
74 4 Manifolds
Example 4.10. Let X V and Y W be open subsets of finite-dimensional K-vector

spaces V and W , let ˛ 2 N, b and let f W X ! Y be a C ˛ -map, which we consider as
a morphism of locally K-ringed spaces .X; CX˛ / ! .Y; C˛Y / (Example 4.5). Then the
inverse function theorem (Appendix 16.16) means that f is a local isomorphism if and
and only if for all x 2 X the derivative Df .x/W V ! W is a K-linear isomorphism.
For ˛ 1 it will follow from Proposition 5.11 that Df .x/ is an isomorphism if fx is an

isomorphism.
We conclude our short introduction to ringed spaces by generalizing the gluing proce-
dure from topological spaces to (locally) ringed spaces. Suppose we have given a family
.Ui /i 2I of locally R-ringed spaces, for all i; j 2 I an open subset Uij Ui (considered
as open R-ringed subspace of Ui ), and for all i; j 2 I an isomorphism 'j i W Uij ! Uj i of
locally R-ringed spaces such that
(a) Ui i D Ui for all i 2 I ,

(b) the cocycle condition holds: 'kj ı 'j i D 'ki on Uij \ Ui k , i; j; k 2 I .
In the cocycle condition we implicitly assume that 'j i .Uij \ Ui k / Uj k , such that the
composition is meaningful. The cocycle condition implies for i D j D k that 'i i D idUi ,
for i D k that 'ij1 D 'j i , and that 'j i induces an isomorphism Uij \ Ui k ! Uj i \ Uj k .
We call such a datum a gluing datum of locally ringed spaces.
Proposition 4.11. There exists a locally R-ringed space X together with morphisms
i W Ui ! X, such that:
(a) For all i the map i is an open embedding.

(b) j ı 'j i D i on Uij for all i; j .
S
(c) X D i i .Ui /.
(d) i .Ui / \ j .Uj / D i .Uij / D j .Uj i / for all i; j 2 I .
Furthermore, .X; . i /i / has the following universal property (which determines it

uniquely up to unique isomorphism): If Z is a locally R-ringed space, and for all
i 2 I , i W Ui ! Z is a morphism of locally R-ringed spaces, such that j ı 'j i D i on
Uij for all i; j 2 I , then there exists a unique morphism W X ! Z with ı i D i for
all i 2 I .
In other words, the R-ringed space X is a colimit in the category of R-ringed spaces.
Proof. By forgetting the structure sheaves in the gluing datum, we obtain a gluing da-
tum of topological spaces, which we glue to a topological space X (Appendix Propo-
sition 12.27). It is endowed with open topological embeddings i W Ui ! X, we have
4.2 Premanifolds and Manifolds 75
S
X D i i .Ui /, and a subset U X is open if and only if for all i the preimage i1 .U /
is open in Ui .
To obtain an R-locally ringed space, we have to “glue” the structure sheaves on the
Ui so as to define a sheaf OX of R-algebras on X. The sheaf OX is uniquely determined
by its sections (and the corresponding restriction maps) on a basis of the topology (Re-
mark 3.13). It is thus sufficient to define it on those open subsets U X that are contained
in one of the i .Ui /, and to check that this is well defined and satisfies the sheaf axioms.
For each such U , we fix an i with U i .Ui /, and we set OX .U / D OUi . i1 .U //. If
U Ui \ Uj , then we identify the rings OUi . i1 .U // and OUj . j1 .U // with OUij .U /
via 'j i . This allows us to define restriction maps. We obtain a sheaf OX of R-algebras
on X that is independent of our choices. Since all the Ui are locally R-ringed spaces, the
same is true for X.
Furthermore, with this definition the i are morphisms of locally R-ringed spaces; they
identify Ui with . i .Ui /; OX j i .Ui / /.
The universal property follows from the universal property of the construction of the
underlying topological space (Appendix Proposition 12.27), which yields a continuous
map W X ! Z, and that it suffices to give a morphism of sheaves ] W 1 OZ ! OX for
those open sets that are contained in one of the i .Ui / (Remark 3.13).
Example 4.12 (Disjoint union). As a (trivial) special case of this construction we have
the disjoint union of locally R-ringed spaces. We simply let Uij D ; for all i, j , so the
underlying topological space is indeed the sum of the topological spaces of the Ui . The
structure sheaf is the unique sheaf of rings whose restriction to Ui is OUi . We denote the
`
disjoint union by i 2I Ui and call it the sum or coproduct. The universal property of
the glued space means that this is indeed a coproduct in the category of locally R-ringed
spaces.
Proposition 4.11 and Example 4.12 are also valid in the category of arbitrary R-ringed
spaces (with the same proof).
4.2 Premanifolds and Manifolds
b 0 D N0 [ f1; !g. We now define premanifolds as locally K-

In this section we fix ˛ 2 N
ringed spaces that are locally isomorphic to an open subspace of Km endowed with either
(for K D R) the sheaf of R-valued C ˛ -functions or (for K D C) the sheaf of holomorphic
functions.
Definition 4.13.
1. A locally R-ringed space .M; OM / is called (real) C ˛ -premanifold if there exists an

S
open covering M D i 2I Ui such that for all i 2 I there exist m 2 N0 , an open
76 4 Manifolds
subspace Y of Rm , and an isomorphism of locally R-ringed spaces .Ui ; OM jUi / !

.Y; C˛Y / (called chart). Here m and Y may depend on i. In this case the structure sheaf
is denoted by CM ˛
and is called the sheaf of C ˛ -functions on M .
1
We call a real C -premanifold a smooth premanifold and a real C ! -premanifold a
real analytic premanifold.
2. A morphism of C ˛ -premanifolds .M; OM / ! .N; ON / is defined as a morphism of
locally R-ringed spaces. Such a morphism is also called a C ˛ -map. Again, C 1 -maps
are also called smooth, and C ! -maps are called real analytic maps.
We obtain the category of C ˛ -premanifolds. A (local) isomorphism in the category of

C ˛ -premanifolds is called a (local) C ˛ -diffeomorphism.
Definition 4.14.
1. A C-ringed space .M; OM / is called complex premanifold if there exists an open

S
covering M D i 2I Ui such that for all i 2 I there exist m 2 N0 , an open sub-
space Y of C m (both dependent on i) and an isomorphism of locally C-ringed spaces
.Ui ; OM jUi / ! .Y; Ohol
Y / (again called chart).
2. A morphism of complex premanifolds .M; OM / ! .N; ON / is defined as a morphism
of locally C-ringed spaces. Such a morphism is also called a holomorphic map.
Again we obtain the category of complex premanifolds. A (local) isomorphism in the

category of complex premanifolds is called a (locally) biholomorphic map.
Definition 4.15. A (real) C ˛ -manifold (respectively a complex manifold) is a C ˛ -

premanifold (respectively a complex premanifold) whose underlying topological space is
Hausdorff and second countable. A morphism of manifolds is a morphism of premani-
folds.
Terminology: To ease the handling of the different types of (pre)manifolds we will use
the following terminology. A (pre)manifold is
O 0 ; in this case
either a real C ˛ -(pre)manifold where ˛ will be always an element in N
we set K WD R. This will be called the real case or that the (pre)manifold is of real
type;
or a complex (pre)manifold. In this case we set K WD C. This will be called the
complex case or that the (pre)manifold is of complex type.
If not otherwise stated, the type of the premanifold is always the same within some
statement. For instance, if we speak of a morphism f of (pre)manifolds, then both
(pre)manifolds are real C ˛ and f is a C ˛ -map (with the same ˛ everywhere) or both
are complex (pre)manifolds and f is holomorphic.
All categories of real C ˛ -(pre)manifolds and complex (pre)manifolds will sometimes

be generically denoted by (PMfd) respectively (Mfd).
If M is a premanifold of arbitrary type, we denote its structure sheaf by OM . A chart
is usually denoted by .U; x/ with U M open, xW U ! Km an open embedding. Its
components are denoted by x i W U ! K, i D 1; : : : ; m. We denote by x 1 W x.U / ! U
the inverse isomorphism of manifolds. For real C ˛ -premanifolds M the structure sheaf
will also be denoted by CM˛
.
Remark and Definition 4.16 (Open submanifolds). Let M be a premanifold and let
U M be an open subspace (Definition 4.7). Then .U; OM jU / is a premanifold, called
an open subpremanifold of M . If M is a manifold, then .U; OM jU / is a manifold, called
open submanifold of M , because every subspace of a Hausdorff (respectively a second
countable) space is again Hausdorff (respectively second countable).
Description of the Structure Sheaf and of Morphisms of Premanifolds

We will now see that the structure sheaf of a premanifold is the sheaf of K-valued mor-
phisms of premanifolds and any morphism between premanifolds is given by the underly-
ing map of topological spaces.
Let M and N be premanifolds. We denote by OM IN the sheaf on M of N -valued
morphisms, i.e., for an open subset U of M , OM IN .U / is the set of all morphisms U ! N
of premanifolds. The restriction maps are given by the restriction of morphisms of locally
K-ringed spaces to open subspaces (Remark 4.7).
Let M be a premanifold. We claim that we have for its structure sheaf OM D OM IK .
More precisely, for every open subspace U of M , a morphism tW U ! K of premanifolds
yields an element tK[
.idK / 2 OM .U /, where idK 2 OK .K/ is the identity.
Proposition 4.17. The map t 7! tK

[
.idK / defines an isomorphism of sheaves K-algebras
W OM IK ! OM :
Proof. Indeed, is clearly a morphism of sheaves of K-algebras. To see that is an

isomorphism, we may work locally (Remark 3.23) and hence we may assume that M has
a global chart xW M ! U Km . By composing with the isomorphism x 1 , we may
assume that M Km is open. In this case, is an isomorphism by definition.
Next we show that morphisms between premanifolds M and N are given be the under-
lying continuous maps.
Proposition 4.18. A morphism of premanifolds f W M ! N is the same as a continuous

map f W M ! N satisfying the following condition:
t 2 ON .V / ) t ı f 2 OM .f 1 .V // for all V N open. (*)

78 4 Manifolds
Proof. Let f W M ! N be a morphism of premanifolds. We have to show that f [ is

given by (*). As the equality of two morphisms of sheaves can be checked on stalks
(Proposition 3.18), we have to see that fx W ON;f .x/ ! OM;x is given for all x 2 X by
composing a germ of a K-valued function at f .x/ with f . This can again be done locally
on M and N and hence we may assume that M Km and N Kn are open. In this case
we have seen in Example 4.5 that fx has the desired form.
Conversely, let f W M ! N be a continuous map satisfying (*), then t 7! t ı f defines
by Proposition 4.17 a morphism f [ W ON ! f OM of sheaves of K-algebras and hence
a morphism of premanifolds.
Dimension of Premanifolds
Let M be a premanifold and let p 2 M . An integer m 0 is called dimension of M in p
(denoted dimp .M /) if there exists an open neighborhood U of p and a chart xW U ! Y ,
where Y Km is open.
The dimension is uniquely determined. Let yW V ! Z Kn be a second chart and
consider the change of charts
'W x jU \V ı y 1 jy.U \V / W y.U \ V / ! x.U \ V /;
which is an isomorphism of manifolds of an open subset of Kn onto an open subset of Km .

We claim that the existence of such an isomorphism implies m D n. Indeed in the complex
case and in the real case for ˛ 1 this is clear because we may form the derivative of ' in
a point and obtain a K-linear isomorphism Kn ! Km . In the real case with ˛ D 0 this is
more difficult but a standard result from algebraic topology (for instance [tDi] (10.3.8)).
The function
M ! N0 ; p 7! dimp .M /
is by definition locally constant. If it is constant D m (for instance if M is connected),
then we say that M has dimension m and write m D dim M .
Example 4.19 (Premanifolds of dimension 0). Let M be a set and consider it as a dis-
crete topological space. Let OM be the sheaf of all K-valued maps on M , i.e., OM .U / is
the K-algebra of all maps U ! K for every subset U M (automatically open). Then
O 0 for K D R (respectively a complex manifold
.M; OM / is a C ˛ -premanifold for all ˛ 2 N
for K D C) of dimension 0.
Conversely, every 0-dimensional premanifold is of this form. Such a premanifold is
automatically Hausdorff. It is second countable if and only if M is countable as a set.
Let M be any premanifold and let Md be the underlying set of M endowed with the
discrete topology and consider Md as a 0-dimensional premanifold (of the same type as
M ) as explained above. Then idM W Md ! M is a morphism of premanifolds.
Premanifold Structure Defined by an Atlas

An important alternative (and more classical) way to construct premanifolds is by way of
an atlas as follows.
Figure 4.1 Charts of a pre-

manifold
M Ui
Uj
xi xj
Let M be a topological space. Suppose that there exists an open covering M D

S
i 2I Ui and for all i 2 I homeomorphisms xi W Ui ! Yi with Yi K open satisfy-
ni
ing the following condition. For all i; j 2 I the homeomorphism (change of charts)
xi1 xj
xi .Ui \ Uj / ! Ui \ Uj ! xj .Ui \ Uj /
is a C ˛ -diffeomorphism (in the real case) respectively a biholomorphic map (in the com-
plex case), see Fig. 4.1. Such a family .Ui ; xi /i is called an atlas of M . It defines the
structure of a premanifold on M as follows. For V M open define OM .V / as the set of
maps f W V ! K such that f jUi \V ı xi1 W xi .Ui \ V / ! K is C ˛ respectively holomor-
phic for all i. This defines a sheaf of K-algebras on M and we obtain a K-ringed space
.M; OM /.
To see that this is a premanifold, fix i0 2 I . For V Ui0 open let f W V ! K
be a map such that f ı xi1 0
W xi0 .Ui0 \ V / ! K is a function (C ˛ respectively holo-
morphic). Then f 2 OM .V / because the change of charts between i and i0 are
C ˛ -diffeomorphisms (respectively biholomorphic). Therefore xi0 yields an isomorphism
.Ui0 ; OM jUi0 / ! .Yi0 ; OYi0 /, where OYi0 is the sheaf of C ˛ respectively holomorphic func-
tions on Yi0 . Hence .M; OM / is a ringed space that is locally isomorphic to a manifold.
Hence it is a premanifold.
0
Conversely, if .M; OM / is a premanifold, we may choose a family of charts
0
S
.xi W .Ui ; OM jUi / ! .Yi ; OYi //i 2I with Yi Kni open and with i 2I Ui D M . We
obtain an atlas .Ui ; xi /i of M defining a premanifold structure OM as above. For V M
open, OM .V / is the K-algebra of maps f W V ! K such that f jV \Ui is a morphism
80 4 Manifolds
of premanifolds V \ Ui ! K. Hence f is a morphism of premanifolds. This shows

0
OM D OM by Proposition 4.17.
Altogether we see that for a topological space M the structure of a premanifold on
M can be either given by defining the structure sheaf or by specifying an atlas for M .
Different atlases giving rise to the same structure sheaf are called equivalent.
Example 4.20. If U Km is an open subspace, then .U; idU / is an atlas defining on

U the structure of a real C ˛ -manifold for all ˛ (if K D R) or a complex manifold (if
K D C).
Similarly one has by Proposition 4.18 the following description of morphisms of pre-
manifolds.
Remark 4.21 (Description of morphisms via an atlas). Let M and N be premanifolds

and let f W M ! N be a continuous map of the underlying topological spaces. Then the
following assertions are equivalent:
(i) The continuous map f is a morphism of premanifolds (Proposition 4.18).

(ii) For every chart .V; y/ of N and for every chart .U; x/ of M the composition
x 1 f y
f.U;x/;.V;y/ W x.f 1 .V / \ U / ! f 1 .V / \ U ! V ! y.V /
is a C ˛ -map (respectively a holomorphic map).

(iii) There exists an atlas .Ui ; xi /i of M and an atlas .Vj ; yj /j of N such that the com-
position f.Ui ;xi /;.Vj ;yj / is a C ˛ -map (respectively a holomorphic map) for all i and
j.
Definition and Remark 4.22 (Charts of morphisms). Let M and N be premanifolds,

let F W M ! N be a continuous map and let p 2 M . A chart of F at p consists of
open neighborhoods F .p/ 2 V N and p 2 U M and a commutative diagram of
continuous map
F jU
U V
x Š Š y
FQ
UQ VQ ;
where xW U ! UQ Km and yW V ! VQ Kn are charts, m D dimp .M /, n D

dimF .q/.N /.
Such a chart always exists: Choose a chart .V; y/ at F .p/ and a chart .U; x/ at p with
U F 1 .V / and define FQ WD y ı F jU ı x 1 .
In addition one may assume:
1. x.p/ D 0 and y.F .p// D 0 (by composing x and y with a translation).

2. UQ D B1 .0/ and VQ D B1 .0/ for any chosen norms on Km and Kn (first shrink VQ , then
UQ so that we may assume UQ D Bı .0/ and VQ D B" .0/, then compose x and y with
multiplication by 1=ı and 1=", respectively).
In Theorem 5.22 we will give an important criterion for a morphism F to have a chart
.U; x; V; y; FQ / where FQ is the restriction of a linear map.
Changing of Structure
Remark 4.23 (Weakening of structure). Let ˛; ˇ 2 N b 0 with ˛ ˇ and let M be a real

C -premanifold. Then every atlas of M is also an atlas of a C ˛ -structure on M because
ˇ
every C ˇ -change of charts is also a C ˛ -diffeomorphism. We obtain on M the structure

of a C ˛ -premanifold M˛ independent of the choice of the atlas. This structure is called
the underlying C ˛ -structure of M . If M is a manifold, so is M˛ because the underlying
topological spaces are the same.
If M is an open submanifold of a finite-dimensional real vector space V , then OM˛
is the sheaf of R-valued C ˛ -functions on M (take the atlas consisting of a single chart
M ,! V Š Km ).
Every morphism of C ˇ -premanifolds F W M ! N is also a morphism F˛ W M˛ ! N˛
of C ˛ -premanifolds and we obtain a faithful functor M 7! M˛ from the category of
ˇ
C ˇ -premanifolds to the category of C ˛ -manifolds. The inclusion W CM ! CM ˛
˛
defines
a morphism .idM ; /W M˛ ! M of locally R-ringed spaces, called canonical, which is
functorial in M .
b 0 and let M 0 be a real C ˇ0 -premanifold. Suppose ˛ minfˇ; ˇ 0 g. Then
Let ˇ 0 2 N
a C ˛ -map M ! M 0 is defined to be a continuous map M ! M 0 that is a morphism of
C ˛ -premanifolds M˛ ! M˛0 .
Conversely, one can show that for ˛ 2 N b and for every real C ˛ -premanifold M there
exists for all ˇ ˛ the structure of a C -premanifold MQ on M such that MQ ˛ D M
ˇ
([KoPu]). If M is a manifold, then any two such C ˇ -manifolds are isomorphic ([Whi2]
and [Gra]). Both the existence and the uniqueness result are wrong for ˛ D 0 ([Mil],
[Ker]). The uniqueness result is also wrong if M is not second countable or not Hausdorff
b with ˛ < ˇ; see [KoPu] and Problem 4.11).
(for all ˛; ˇ 2 N
Remark 4.24 (Complex manifolds as real manifolds). Let M be a complex premani-

fold. By identifying C m with R2m we may consider every chart xW U ! C m of M as
a chart xR W U ! R2m . As every complex analytic map between open subsets of C m is in
particular real analytic, we see that for an atlas .Ui ; xi /i of M we obtain the structure of
82 4 Manifolds
a real analytic premanifold MR defined by the atlas .Ui ; .xi /R /i . This structure is inde-
pendent of the choice of atlas and we call MR the underlying real analytic premanifold.
If M is a manifold, so is MR because the underlying topological spaces are the same.
Every holomorphic map F W M ! N between complex premanifolds is a real analytic
map FR W MR ! NR and we obtain a faithful functor M 7! MR from the category
of complex premanifolds to the category of real analytic premanifolds. The inclusions
OM ! OMR IC (here we consider C as a real analytic manifold) and OMR ! OMR IC
define functorial morphisms
MR .M; OMR IC / ! M
of locally R-ringed spaces (for the left arrow) and locally C-ringed spaces (for the right
arrow), which we call canonical.
A map f W M ! N between complex premanifolds is called a real C ˛ -map (˛ 2 N b0)
if it is a C ˛ -map MR ! NR (Remark 4.23).
4.3 Examples of Manifolds
Example 4.25 (1-dimensional real torus). Let T 1 WD R=Z be the 1-dimensional real
torus. Endow T 1 with the quotient topology with respect to the projection W R ! T 1 .
N T 1 ! S 1 D f z 2 C I jzj D 1 g, e.t/
Then eW N WD exp.2 it/ is a homeomorphism, in
particular T 1 is compact, Hausdorff, and second countable.
Define the structure of a real analytic manifold on T 1 as follows:
1. We may use an atlas. Set U1 WD f x C Z 2 R=Z I 0 < x < 1 g and U2 WD f x C Z 2

R=Z I 1=2 < x < 1=2 g. Then U1 and U2 are open in T 1 and U1 [ U2 D T 1 . Define
˚1 W U1 ! .0; 1/ R; x C Z 7! x;
˚2 W U2 ! .1=2; 1=2/ R; x C Z 7! x:
Then ˚1 and ˚2 are homeomorphisms and the change of charts

˚2 ı˚11
.0; 1=2/ D ˚1 .U1 \ U2 / ! ˚2 .U1 \ U2 / D .0; 1=2/; x 7! x
is a real analytic diffeomorphism.

2. Alternatively, we can directly define the structure sheaf on T 1 . For V T 1 open
define a sheaf of R-algebras
˚
C!T 1 .V / WD f W V ! R I f ı W 1 .V / ! R is C ! :
We obtain a R-ringed space T 1 that is a real analytic manifold. This can be checked
locally on T 1 . Cover T 1 by open subsets V such that there exists U R with U WD
4.3 Examples of Manifolds 83
jU W U ! V a homeomorphism (any open V ¨ T 1 has this property). We claim that

jU is an isomorphism of R-ringed spaces .U; C!U / ! .V; C!T 1 jV /. We have to show
that for all W V open and f W W ! R map:
f 2 C!T 1 .W / W, f ı j 1 .W / 2 C!R . 1 .W //
, f ı .U j 1 .W / / 2 C!U .U1 .W //:
U „ ƒ‚ …
DU \W
`
But 1 .W / D n2Z .U \ W / C n. Hence this equivalence is clear.
The projection W R ! T 1 is a morphism of manifolds: Either one can use Proposi-

tion 4.18 and (2) or the description of morphisms via an atlas (Remark 4.21) using the
atlas .Ui ; ˚i /i D1;2 defined above.
Example 4.26 (Projective space). Recall that for a field k the projective n-space is de-
fined as follows:
˚
P n .k/ D .k nC1 n f0g/= D L k nC1 I L 1-dim. sub-v.s.
v 7! hvi
where v v 0 if and only if there exists 2 k such that v 0 D v. For .x0 ; : : : ; xn / 2

k nC1
n f0g we denote by .x0 W : : : W xn / its equivalence class in P n .k/.
There is the following alternative description. Let
˚
P WD g D .aij /0i;j n 2 GLnC1 .k/ I a10 D a20 D D an0 D 0
and let L0 D he0 i D h.1; 0; : : : ; 0/i 2 P n .k/. Then
GLnC1 .k/=P ! P n .k/; g 7! g.L0 /
is bijective.
Now let k D K. Endow P n .K/ with quotient topologies in three ways by surjective
maps:
(i) KnC1 n˚f0g ! P n .K/, .x0 ; : : : ; xn / 7! .x0 W : : : W xn /.

P
(ii) S n WD x 2 KnC1 I jjxjj2 WD niD0 jxi j2 D 1 ! P n .K/,
.x0 ; : : : ; xn / 7! .x0 W : : : W xn /.
(iii) GLnC1 .K/ ! P n .K/, g 7! g.L0 /.
84 4 Manifolds
Then we have a commutative diagram of continuous maps
i Wx7!x .a00 ;:::;an0 / 7 .aij /Wp

Sn KnC1 n f0g GLnC1 .K/
x=jjxjj2 7 xWr
2
1 3
P .K/;
n
where the right horizontal map p is open (being the restriction of a projection map). Hence
all three topologies on P n .K/ defined above are equal. Indeed, let U P n .K/ be a subset.
Then 11 .U / D i 1 .21 .U // is open if and only if 21 .U / D r 1 .11 .U // is open.
Hence the quotient topologies of S n and of KnC1 n f0g are equal. Furthermore 21 .U / D
p.31 .U // is open if and only if 31 .U / D p 1 .21 .U // is open. Therefore the quotient
topologies of KnC1 n f0g and of GLnC1 .K/ are equal.
As P is closed in GLn .K/, Description (iii) shows that P n .K/ is Hausdorff (Appendix
Proposition 12.58). By Description (ii), P n .K/ is compact because S n is compact (Ap-
pendix Proposition 12.52).
We define the structure of a real analytic R-manifold on P n .R/ (respectively a complex
manifold on P n .C/) by an atlas .Ui ; ˚i /0i n with
˚i W Ui WD f .x0 W : : : W xn / 2 P n .K/ I xi ¤ 0 g ! Kn ;

x0 xi 1 xi C1 xn
.x0 W : : : W xn / 7! ;:::; ; ;:::; :
xi xi xi xi
˚
Then ˚i .Ui \ Uj / D .y0 ; : : : ; yi 1 ; yi C1 ; : : : ; yn / I yj ¤ 0 and for 0 i ¤ j n the
change of charts is given by
˚i .Ui \ Uj / ! Ui \ Uj
! ˚j .Ui \ Uj /;
.y0 ; : : : ; yi 1 ; yi C1 ; : : : ; yn / 7! .y0 W : : : W yi 1 W 1 W yi C1 W : : : W yn /

y0 1 yj 1 yj C1 yn
7! ;:::; ;:::; ; ;::: W ;
yj yj yj yj yj
„ƒ‚…
i
which is a real bianalytic (respectively a biholomorphic) map.
4.4 Topological Properties of (Pre)Manifolds
Remark 4.27. Let M be a premanifold.
1. Locally, M is homeomorphic to an open convex subspace of Rn . Hence M is locally

contractible (Definition 2.5), first countable, locally compact, and a T1 -space.
4.5 Basic Constructions and Further Examples of Manifolds 85
2. As M is locally contractible, it is in particular locally path connected. Hence in M

connected components and path components are equal and all connected components
are open and closed in M (Proposition 2.13).
3. If M has a countable atlas, then M is second countable (Remark 1.3 4). Therefore
every Hausdorff premanifold with a countable atlas is a manifold.
As a premanifold also has an open covering consisting of simply connected subspaces,

it is semilocally simply connected (Problem 2.19) and hence each connected premanifold
has a universal covering space (Problem 2.20). We will not use this fact in the sequel.
Proposition 4.31 then shows that the universal cover has a canonical structure of a pre-
manifold.
Theorem 4.28. Every manifold M has the following properties:
1. M is paracompact and Lindelöf. More precisely, for every open covering .Ui /i of M
there exists a countable locally finite refinement .Vn /n such that for all n the closure
Vn is compact and contained in some Ui .
S
2. There exists a sequence .Cn /n2N of compact subspaces of M such that M D n Cn
ı
and Cn CnC1 for all n. In particular M is -compact.
3. M is normal.
Moreover, Remark 1.14 (for which we did not give a proof) implies that every manifold
is metrizable. We will not use this fact in the sequel.
Proof. All assertions of 1 hold by Proposition 1.10 except for the property that Vn is con-
tained in some Ui for all n. This last property follows from the shrinking lemma 1.21.
Assertion 2 also follows from Proposition 1.10. Finally, as M is paracompact and Haus-
dorff, M is normal by Proposition 1.18.
4.5 Basic Constructions and Further Examples of Manifolds
Products
Let M and N be premanifolds with atlases .Ui ; ˚i /i of M and .Vj ; j /j of N . Then
.Ui Vj ; ˚i j / is an atlas on the product space M N making M N into a premanifold.
In particular for x 2 M and y 2 N one has dim.x;y/ .M N / D dimx .M / C dimy .N /.
If M and N are manifolds, then M N is a manifold because the product of two
Hausdorff (respectively second countable) spaces is again Hausdorff (respectively second
countable).
86 4 Manifolds
Figure 4.2 Möbius band
Example 4.29 (Real tori). In Example 4.25 we endowed T 1 WD R=Z with the structure
of a 1-dimensional real analytic manifold. Hence we obtain for all n 1 the structure of
an n-dimensional real analytic manifold on the n-dimensional real torus T n WD Rn =Zn D
.T 1 /n .
Gluing
Suppose that ..Ui /i 2I ; .Uij /i;j ; .'ij /i;j / is a gluing datum of locally K-ringed spaces such
that the Ui are all premanifolds. Then the glued locally K-ringed space (Proposition 4.11)
is locally isomorphic to a premanifold. Hence it is a premanifold.
Example 4.30 (Möbius band). If I D f1; 2g consists of two elements, then the non-
redundant data of such a gluing datum consist of two premanifolds U1 and U2 , open
subpremanifolds U12 U1 , U21 U2 and an isomorphism 'W U12 ! U21 .
For instance, identify S 1 D f z 2 C I jzj D 1 g. Then the (open) Möbius band is glued
from the two rectangular strips U1 WD .S 1 n f1g/ .1; 1/ and U2 WD .S 1 n f1g/
.1; 1/˚ along the open submanifolds
U12 WD U21 WD U1 \ U2 D W .1; 1/ with
W D z 2 S 1 I Im.z/ ¤ 0 using the real analytic isomorphism
8
<.z; t/; if Im.z/ > 0I
U12 3 .z; t/ 7! 2 U21
:.z; t/; if Im.z/ < 0;
(see Fig. 4.2). We obtain a real analytic manifold.
Coverings of Manifolds
Proposition 4.31. Let M be a premanifold, let MQ be a topological space and let f W MQ !

M be a local homeomorphism. Then there exists a unique structure of a premanifold on
MQ such that f is a local isomorphism of premanifolds.
This proposition can in particular be applied if f is a covering map, for instance if MQ

is a universal covering of a connected premanifold M .
4.6 Problems 87
Proof. Existence: Define OMQ WD f 1 OM and let f ] WD idf 1 OM W f 1 OM ! OMQ . Then

.f; f ] /W MQ ! M is a morphism of K-ringed spaces such that f is a local homeomor-
phism and such that fx is an isomorphism. Hence it is a local isomorphism of K-ringed
spaces (Remark 4.9). Clearly, any K-ringed space that is locally isomorphic to a preman-
ifold is itself a premanifold (of the same type). Hence MQ is a premanifold.
Uniqueness: If O0 is another sheaf of K-algebras defining a manifold structure on MQ
such that f is a local isomorphism, then idMQ is a local isomorphism and a homeomor-
phism and hence an isomorphism.
Corollary 4.32. We keep the notation of Proposition 4.31. If M is a manifold and f is

a covering map with countable fibers, then MQ is a manifold.
One can show that if MQ is a path connected topological space, then the fibers of any
covering map W MQ ! M of a manifold M are automatically countable: By [LeeJo]
Prop. 1.9 the fundamental group 1 .M; p/ is countable for every p 2 M and by Re-
mark 2.40 there is a surjection 1 .M; p/ ! 1 .p/ if MQ is path connected.
Proof. Every covering map is separated (Remark 2.30), hence MQ is Hausdorff by Corol-
lary 1.27. As M is a Lindelöf space (Theorem 4.28), we find a countable open covering
`
.Un /n by charts such that ˚ 1 .Un / D l2Ln UQ n;l and ˚W UQ n;l ! Un is a homeomor-
phism for all n and l. By hypothesis, Ln is countable. The countably many UQ n;l s form an
open covering of MQ . As every Un is homeomorphic to some open subset of Rm , Un and
hence UQ n;l is second countable for all n and all l 2 Ln . Therefore MQ is second countable
(Remark 1.3 4).
4.6 Problems
Problem 4.1. Let R be a local integral domain that is not a field and let K be its field
of fractions. Show that the inclusion R ! K is not a local homomorphism. Use this
observation to construct locally ringed spaces whose underlying topological spaces consist
of a single point and a morphism of ringed spaces between them that is not a morphism of
locally ringed spaces.
Problem 4.2. Show that R ! R, x 7! x, induces an automorphism of the real analytic
manifold T 1 .
Problem 4.3. Let ˛ 2 N,b m; n 2 N, and let F W RnC1 n f0g ! RmC1 n f0g be a morphism
of C -manifolds that is homogeneous of degree d 2 Z (i.e., F . x/ D d F .x/ for
˛
all x 2 RnC1 n f0g and 2 R ). Show that F induces a morphism of C ˛ -manifolds

P n .R/ ! P m .R/.
Remark: See also Problem 6.4 for a generalization.
88 4 Manifolds
Problem 4.4. Let P be the topological space consisting of a single point * and define
a sheaf of K-algebras on P by OP .P / D K.
1. Show that P is a final object in the category of K-ringed spaces and in the category of
locally K-ringed spaces.
2. Let X be a locally K-ringed space. Show that f 7! f . / yields a bijection between
the set of morphisms of locally K-ringed spaces P ! X and the set of x 2 X such
that .x/ D K.
Note that P WD .P; OP / is the 0-dimensional manifold consisting of a single point.
Problem 4.5. Let KŒ" be the K-algebra KŒX=.X 2 /, where " denotes the image of X
in KŒ". Let P Œ" be the K-ringed space whose underlying topological space consists of
a single point with OP Œ" .P Œ"/ D KŒ". Let X be a locally K-ringed space.
1. Show that KŒ" is a 2-dimensional K-vector space with basis 1 and " and that the
multiplication is given by .a1 C b1 "/.a2 C b2 "/ D a1 a2 C .a1 b2 C a2 b1 /". Deduce that
KŒ" is a local ring with maximal ideal K" WD f a" I a 2 K g and that P Œ" is a locally
K-ringed space.
2. Show that i W KŒ" ! K, a C b" 7! a, is a K-algebra homomorphism that yields
a morphism of locally K-ringed spaces iW P ! P Œ" (with P as in Problem 4.4).
3. Let tW P Œ" ! X be a morphism of locally K-ringed spaces, x WD t. /. Show that
tx W OX;x ! KŒ" induces a K-linear map t W mx =m2x ! K" Š K (where the isomor-
phism is the map a" 7! a).
4. Show that the construction in 3 yields a bijection between the set of morphisms of
locally K-ringed spaces P Œ" ! X and the set
˚
.x; / I x 2 X with .x/ D K, 2 .mx =m2x /_ :
Problem 4.6. Let .X; OX / be a locally ringed space.
1. Let U X be an open and closed subset. Show that there exists a unique section
eU 2 .X; OX / such that eU jV D 1 for all open subsets V of U and eU jV D 0 for all
open subsets V of X n U . Show that U 7! eU yields a bijection
OC.X/ $ Idem. .X; OX // (*)
from the set of open and closed subsets of X to the set of idempotent elements of the
ring .X; OX / (e 2 .X; OX / is called idempotent if e 2 D e).
2. Show that eU eU 0 D eU \U 0 for U; U 0 2 OC.X/.
3. Show that X is connected if and only if there exists no idempotent element e 2
.X; OX / with e ¤ 0; 1.
4.6 Problems 89
4. Now suppose that X is locally connected (Appendix Problem 12.22). Show that the
bijection (*) induces a bijection between connected components of X and indecom-
posable idempotents (an idempotent is called indecomposable if it cannot be written
as a sum of non-zero idempotents).
Problem 4.7. Let n > 1 be an odd integer and endow the topological space R via the
two atlases .R; idR / and .R; ˚/ with ˚W R ! R, x 7! x n , in two ways with the structure
of a smooth manifold. Show that these two structures are not equal but that the resulting
manifolds are isomorphic.
Problem 4.8. Show that a premanifold is Hausdorff if and only if for all p; q 2 M with
p ¤ q there exists U M open with p; q 2 U and f 2 OM .U / such that f .p/ ¤ f .q/.
Problem 4.9. Let M be a Hausdorff premanifold. Show that the following assertions are
equivalent:
(i) The connected components of M are second countable.

(ii) The connected components of M are -compact.
(iii) M is paracompact.
Problem 4.10. Show that a Hausdorff premanifold is a manifold if and only if it is para-
compact and has countably many connected components.
Hint: Problem 4.9.
Problem 4.11. Let U1 D U2 D R (considered as 1-dimensional real analytic manifold)

and glue U1 and U2 along idRnf0g to a premanifold X.
1. Show that X is a second countable connected real analytic premanifold that is not
Hausdorff. Let ˚i WD idR W Ui ! R. Then A D .Ui ; ˚i /i D1;2 is an atlas of X. Sketch
X.
2. For ˛; ˇ 2 N b with ˛ < ˇ let ' D '˛ˇ W R ! R be a C ˛ -diffeomorphism with
'˛ .0/ D 0 and such that ' jRnf0g is a C ! -diffeomorphism but that ' is not a C ˇ -
map in 0. Show that .U1 ; ˚1 / and .U2 ; ' ı ˚2 / form an atlas of the structure of a real
analytic premanifold X ' on the topological space X. Show that the underlying C ˛ -
premanifolds X˛ and .X ' /˛ are isomorphic but that Xˇ and .X ' /ˇ are not isomorphic.
ˇ
3. Give an example of a map '˛ with the above properties.
N WD ˝Œ0; 1/ endowed
Problem 4.12. Let .˝; / be an infinite well-ordered set and let L
with the lexicographic order (i.e., .!1 ; x1 / .!2 ; x2 / , !1 < !2 or [!1 D !2 and
x1 x2 ]). Endow LN with the order topology (Appendix Problem 12.24). Let !0 2 ˝ be
N n f.!0 ; 0/g.
the smallest element and define the open long ray as the subspace L WD L
90 4 Manifolds
1. Show that L is a connected normal Hausdorff space.

2. Let ˝ WD N. Show that L is homeomorphic to R>0 .
3. Suppose that ˝ is uncountable. Show that L is not second countable.
4. Endow L with the structure of a 1-dimensional real analytic premanifold.
Problem 4.13. Let M1 , M2 , and N be premanifolds. Show that a map F W N ! M1 M2

is a morphism of premanifolds if and only if each component N ! Mi of F is a morphism
of premanifolds.
Problem 4.14. Let M be a connected premanifold. Show that there exists a universal
cover space W MQ ! M and that there exists a unique premanifold structure on MQ such
that is a local isomorphism of premanifolds.
Hint: Problem 2.20.
Problem 4.15. Let M be a connected complex premanifold.
1. Show that every non-constant holomorphic function f W M ! C is open.

Hint: Appendix Problem 16.4.
2. Let M be compact. Show that every holomorphic function f W M ! C is constant.
Problem 4.16. Let N be a premanifold, let W M 0 ! M be a surjective local isomor-

phism of premanifolds (for instance a covering map) and let F W M ! N be a map. Show
that F is a morphism of premanifolds if and only if F ı is a morphism of premanifolds.
Linearization of Manifolds
5
It is a standard and very useful technique to approximate geometric objects and maps by
linear objects and maps. A principal (trivial) example in basic analysis is approximating
an open subspace U of Kn at a point by the linear space Kn itself by visualizing Kn as
a tangent space to U at the given point. Another (less trivial) example is the derivative
of a differentiable map f on U in a point that is the best approximation of f nearby this
point by a linear map. In the first section we extend this technique to premanifolds and
morphisms of premanifolds. Thus we first define the tangent space of a premanifold at
a point, which will be a K-vector space that we visualize as the “best linear approxima-
tion” of the premanifold at the given point. Then we define the derivative in a point x of
a morphism f . This will be a linear map from tangent space at x to the tangent space at
f .x/.
In the next section we study local properties of morphisms of premanifolds, in par-
ticular those given by properties of the derivative. The main result of Sect. 5.3 then will
show that those morphisms whose derivative have locally constant rank as linear maps can
themselves be linearized locally by choosing an appropriate chart.
These techniques allow us in the following sections to study submanifolds, fiber prod-
ucts of manifolds (in particular intersection of submanifolds), and quotient manifolds.
Notation: From now on a premanifold will always be either a real C ˛ -premanifold
with ˛ 2 N b or a complex premanifold. Hence we no longer consider the case of a C 0 -
premanifold.
5.1 Tangent Spaces
We will now define the notion of the tangent space of a premanifold in a point. There are
several ways to think of tangent vectors at a point p on the premanifolds M . We start
by considering them as “derivatives c 0 .0/ of curves” cW I ! M (I K a “small” open
neighborhood of 0) with c.0/ D p. Of course the notion of the derivative of the curve c
DOI 10.1007/978-3-658-10633-1_5
92 5 Linearization of Manifolds
on M does not exist yet (it is one of the goals of this section to define this notion!). But
if it existed, we would expect it to satisfy formal properties of the derivative such as the
chain rule
“Df .p/ ı c 0 .0/ D .f ı c/0 .0/” (*)
for every K-valued morphism f defined in some neighborhood of p. Also by taking for
f the components of a chart at p we would expect that “c 0 .0/” is uniquely determined by
“Df .p/ ı c 0 .0/” if f runs through all such functions.
Now, c 0 .0/ may not yet be defined, but the right-hand side of (*) perfectly makes sense
as f ıc is a map from the open subset I of K to K for which a derivative is already defined.
Hence we simply define a tangent vector as an equivalence class of curves cW I ! M with
c.0/ D p, where such curves c1 and c2 are called equivalent if .f ı c1 /0 .0/ D .f ı c2 /0 .0/
for all K-valued morphisms f defined in an open neighborhood of p.
The set of these tangent vectors will be the tangent space of M at p. We then endow
this set with the structure of a K-vector space by “transport of structure” via a chart.
Hence let us first explain what we mean by “transport of structure”. Given a set T and
a bijective map ˛W T ! V , where V is a K-vector space, we define an addition and a scalar
multiplication on T by t1 C t2 WD ˛ 1 .˛.t1 / C ˛.t2 // and t WD ˛ 1 . ˛.t//. In other
words, we endow T with the unique structure of a K-vector space such that ˛ becomes a
K-linear isomorphism. If 'W V ! W is a K-linear isomorphism, then the structures on T
defined by ˛ and by ' ı ˛ coincide.
After this preparation we can now define tangent spaces via the procedure motivated
above.
Definition and Remark 5.1 (Tangent space). Let M be a premanifold, p 2 M a point,

m WD dimp .M /.
1. Define a set
Tp .M / WD f.I; c/ j I K open with 0 2 I;

cW I ! M morphism with c.0/ D pg=
where .I1 ; c1 / .I2 ; c2 / if and only if for every germ f 2 OM;p one has .f ıc1 /0 .0/ D
.f ı c2 /0 .0/.
2. Let .U; x/ be a chart at p (i.e., p 2 U ), then the map
Tp .x/W Tp .M / ! Km ; .I; c/ 7! .x ı c jc 1 .U / /0 .0/
yields a well-defined bijection (an inverse map is given by sending a 2 Km to the

equivalence class of .I; ca /, where ca .t/ WD x 1 .x.p/ C ta/, restricted to a suffi-
ciently small neighborhood I of 0). We use this bijection to endow Tp .M / with the
structure of a K-vector space via transport of structure as explained above. This is
5.1 Tangent Spaces 93
TF(p)(N)
Tp (M)
p F(p)
Tp(F)
Figure 5.1 Tangent map in a point p
independent of the choice of the chart as every change of charts has as derivative a
K-linear isomorphism Km ! Km . As m D dimp .M / we find
dimK .Tp .M // D dimp .M /: (5.1)
This definition of tangent vectors makes it easy to define the derivative of a morphism
F : It is simply F applied to a tangent vector.
Remark and Definition 5.2 (Derivative). Let F W M ! N be a morphism of premani-

folds. Then we define the derivative of F at p or the tangent map of F at p (see Fig. 5.1)
by
dF .p/ WD Tp .F /W Tp .M / ! TF .p/ .N /; Œ.I; c/ 7! Œ.I; F ı c/:
To see that this is a well-defined K-linear map let us first consider the case that M and N
are open submanifolds of Km and Kn , respectively. As charts at p and F .p/ we choose
the inclusions M ,! Km and N ,! Kn . We obtain isomorphisms
Km ! Tp .M /; a 7! Œ.I; ca /;
TF .p/ .N / ! K ;
n
Œ.J; c/ 7! c 0 .0/;
where ca .t/ D p C ta and I is small enough such that p C ta 2 M for all t 2 I . Hence
Tp .F / F .p C ta/ F .p/
Km ! Tp .M / ! TF .p/ .N / ! Kn ; a 7! lim ;
t !0 t
in other words, this composition is the derivative of F at p. This shows that Tp .F / is
a well-defined K-linear map in this case.
In general, for all charts .U; x/ at p and .V; y/ at F .p/ the chain rule (Appendix
Proposition 16.5) shows
TF .p/ .y/Œ.I; F ı c/ D .y ı F ı c/0 .0/ D D.y ı F ı x 1 /.x.p//.x ı c/0 .0/

D D.y ı F ı x 1 /TF .p/ .y/Œ.I; c/:
The arguments above show in particular that we have the following description of the
tangent map.
Remark 5.3. Let F W M ! N be a morphism of premanifolds. Given a chart .U; x; V;

y; FQ / for F (Definition 4.22) we obtain for all q 2 U a commutative diagram of K-linear
maps
Tq .F /
Tq .M / TF .q/ .N /
Tq .x/ Š Š TF .q/ .y/ (5.2)
Tx.q/ .FQ /
Km D Tx.q/ .UQ / Ty.F .q//.VQ / D Kn ;
and Tx.q/ .FQ / is given by the Jacobi matrix of FQ in x.q/ (Appendix Remark 16.18).
Remark 5.4 (Chain rule). Let F W M ! N and GW N ! Q be morphisms of preman-

ifolds. Then the definition of the derivative implies immediately the following elegant
version of the chain rule:
Tp .G ı F / D TF .p/ .G/ ı Tp .F / “Chain rule”: (5.3)
It might be surprising that the chain rule seems to be a triviality. The difficulty is hidden
in the fact that our definition of derivative of a morphism of premanifolds is well defined.
This was proved in Remark 5.2 and used the classical chain rule.
We can express the chain rule also as follows. Let (PMfd*) be the category of preman-
ifolds M together with a point p 2 M . Morphisms .M; p/ ! .N; q/ are morphisms of
premanifolds F W M ! N such that F .p/ D q. Then
.M; p/ 7! Tp .M /; .F W .M; p/ ! .N; q// 7! .Tp .F /W Tp .M / ! Tq .N //
is a (covariant) functor (PMfd*) ! (K-Vec). In particular, if F is an isomorphism of

premanifolds, then Tp .F / is an isomorphism of K-vector spaces.
Remark 5.5.
1. Let M be a premanifold, U M open (hence U is premanifold), p 2 U . Then

Tp .U / D Tp .M /.
2. Let M , N be premanifolds, p 2 M , q 2 N . If .U; x/ is a chart of M at p and .V; y/
is a chart of N at q, then .U V; x y/ is a chart of M N at .p; q/. Therefore
Remark 5.1 2 shows
T.p;q/ .M N / D Tp .M / Tq .N /:
A similar statement holds for the derivative of a product of morphisms of premanifolds.

Tangent vectors as derivations

We will now see that we can also consider tangent vectors more algebraically as deriva-
tions. Let us start with the general definition of a derivation.
Definition 5.6. Let A be a commutative K-algebra and M an A-module. An element of
DerK .A; M / WD f W A ! M j K-linear map,

.ab/ D a .b/ C b .a/ for all a; b 2 Ag
is called a K-derivation on A with values in M . Then DerK .A; M / is a K-subspace (even

an A-submodule) of the A-module of all maps A ! M .
Of course the same definition works just fine if we replace K with an arbitrary commu-
tative ring but we will not need this generality.
Remark 5.7 (Tangent vectors as derivations). Let M be a premanifold, p 2 M a point,

n WD dimp .M /. Then D Œ.I; c/ 2 Tp .M / defines a map
W OM;p 7! K; f 7! .f ı c/0 .0/:
Then is K-linear and .fg/ D f .p/ .g/ C g.p/ .f /. In other words, is a K-

derivation on OM;p with values in K, where we consider K as an OM;p -module via the
scalar multiplication f a D f .p/a for f 2 OM;p and a 2 K. We obtain a K-linear map
W Tp .M / ,! DerK .OM;p ; K/; (5.4)
which is injective by the definition of Tp .M /. We use this map to consider Tp .M / as a

K-subspace of DerK .OM;p ; K/.
If F W M ! N is a morphism of manifolds, then the K-algebra homomorphism
Fp W ON;F .p/ ! OM;p , g 7! g ı F , induces a K-linear map
F W DerK .OM;p ; K/ ! DerK .ON;F .p/ ; K/; 7! ı Fp :
The diagram
Tp .F /
Tp .M / TF .p/ .N /

F
DerK .OM;p ; K/ DerK .ON;F .p/ ; K/
commutes. Indeed, let D Œ.I; c/ 2 Tp .M / identified with the derivation f 7! .f ı

c/0 .0/. Then
F .f 7! .f ı c/0 .0// D .g 7! .g ı F ı c/0 .0// 2 DerK .ON;F .p/ ; K/;

which is identified with Œ.I; F ı c/ D Tp .F /. /.
If we choose a chart we obtain the following coordinate description of the tangent

space.
Remark 5.8. Let m D dimp .M / and .U; x/ be a chart at p with coordinate function
x i W U ! K. Then Tp .x/W Tp .M / ! Tx.p/ .x.U // D Km is a K-linear isomorphism with
coordinates dx i D Tp .x i /W Tp .M / ! K. In particular, .dx i /1i m is a K-basis of the
dual space Tp .M /_ .
On Km we have the partial derivatives in direction of the standard basis vectors ei .
We use the chart to make these into derivations on OM;p . More precisely, we define for
i D 1; : : : ; m a derivation
@./
@i WD .p/W OM;p ! K;
@x i
@f @ d
@i .f / WD .p/ WD .f ı x 1 /.x.p// D j t D0 f .x 1 .x.p/ C tei //:
@x i @xi dt
In other words, let cei W I ! U , cei .t/ WD x 1 .x.p/ C tei / (with 0 2 I K so small such
that x.p/ C tei 2 x.U / for all t 2 I ). Then
@i D Œ.I; cei / 2 Tp .M /:
Clearly dx j .@i / D ıij (Kronecker delta). Hence .@i /1i m is linearly independent and
therefore a basis of Tp .M / because dim Tp .M / D m. It is the basis dual to the basis
.dx i /i of Tp .M /_ .
@f
For U open neighborhood of p and f 2 OM .U / we set @x i .p/ WD @i .fp /.
We give a further description of the space of derivations. This is the following purely
algebraic fact applied to the K-algebra OM;p and its maximal ideal mp .
Remark 5.9. Let k be a field, let A be k-algebra that is a local ring with maximal ideal
m and residue field k, i.e., the composition k ! A ! A=m is an isomorphism that we
use to identify k with A=m. We claim that 7! jm induces an isomorphism of k-vector
spaces
Derk .A; k/ ! Homk .m=m2 ; k/ D .m=m2 /_ : (5.5)
Indeed for a; b 2 m we have .ab/ D a .b/ C b .a/ D 0 and hence jm2 D 0. Hence
induces a homomorphism of abelian groups m=m2 ! k, which is k-linear again because
of the Leibniz rule.
To see that this map is bijective, observe that we have a k-linear isomorphism A !
k ˚ m, a 7! .a; N a a/,
N where aN is the image of a in A=m D k. In other words, the
exact sequence 0 ! m ! A ! k ! 0 splits via the given k-algebra structure k ! A
(Appendix Definition 14.5). If tW m=m2 ! k is a k-linear map we obtain a k-derivation

W A D k ˚ m ! k by . ; a/ D t.a mod m2 /. This defines an inverse map to (5.5).
Altogether we obtain K-linear maps

Tp .M / ! DerK .OM;p ; K/ ! .mp =mp2 /_ (5.6)
for every premanifold M and every p 2 M . The first map is injective and the second
map is an isomorphism. After choosing a chart .U; x/ at p, the image of can be de-
scribed as the derivations that are linear combinations of @1 ; : : : ; @m , m D dimp .M /. In
particular dimK .mp =mp2 / dimp .M /. For real C ˛ -premanifolds with ˛ < 1, the map
is not an isomorphism (in fact, mp =mp2 is an infinite-dimensional K-vector space in this
case, see Problem 5.3). But for C 1 -premanifolds and (real or complex) analytic preman-
ifolds, we will now show that mp =mp2 has the same dimension as Tp .M / and hence is an
isomorphism. Indeed, by choosing a chart at p it suffices to prove the following lemma.
Lemma 5.10. Let X D Km with coordinate functions x i W X ! K. For K D R let OX be

the sheaf of C 1 -functions or of analytic functions on X. For K D C let OX be the sheaf
of holomorphic functions on X. Let a D .a1 ; : : : ; am / 2 Km and let ma be the maximal
ideal of OX;a . Then ma is generated by the germs of x i ai , i D 1; : : : ; m, and their
images in ma =m2a form a K-basis.
Proof. We may assume that a D 0. Let f 2 m WD m0 , i.e., f is C ˛ -function with

˛ 2 f1; !g defined on some neighborhood U of 0 with f .0/ D 0. By shrinking U , we
may assume that U is convex. Set
Z1
fi .u/ WD @i f .tu/ dt for u 2 U:
0
Pn
Then the fundamental theorem of calculus implies that f D i
i D1 fi x (here we use
f .0/ D 0). Moreover, as integration and partial differentiation commute, fi is again
a C ˛ -function (here we need ˛ 1; for ˛ < 1, we could only conclude that fi is a
C ˛1 -function). Therefore we see that the ideal m is generated by the coordinate functions
x 1 ; : : : ; x m . Hence their images in m=m2 form a generating system. As we already know
that dimK .m=m2 / m, these images form a K-basis.
Summarizing we see that we have the following description of the tangent space.
Proposition 5.11. For real C ˛ -premanifolds M with ˛ 1 or for complex premanifolds

M and for all p 2 M one has K-linear isomorphisms
Tp .M / ! DerK .OM;p ; K/ ! .mp =mp2 /_ :

Finally, Problem 5.5 gives a further description of the tangent space of a premanifold
M as in Proposition 5.11: Let P Œ" be the locally K-ringed space whose underlying topo-
logical space is a single point * with OP Œ" .P Œ"/ D KŒ" WD KŒT =.T 2 /, where " denotes
the image of T in KŒT =.T 2 /. Then one has for p 2 M
( )
tW P Œ" ! M morphism of locally
Tp .M / D : (5.7)
K-ringed spaces with t. / D p
It is also possible to describe the K-vector space structure on Tp .M / via the right-hand
side (Problem 5.5). Moreover, it is not difficult to see that the map induced by a morphism
of premanifolds F W M ! N on tangent spaces is given by
Tp .M / 3 .tW P Œ" ! M / 7! F ı t 2 TF .p/ .M /:
We conclude the section by explaining that the tangent space does not change when the
structure is weakened.
Remark 5.12. Let M be a real C ˇ -premanifold, let ˛ 2 N b with ˛ ˇ, and let M W M !

M˛ be the canonical morphism of locally R-ringed spaces, where M˛ is the underlying
C ˛ -premanifold (Remark 4.23). Then [ D ] W OM ! OM˛ is the inclusion and the re-
striction of derivations defines for all p 2 M a K-linear map M W DerK .OM˛ ;p ; K/ !
DerK .OM;p ; K/, which is functorial in M , i.e., for every morphism of C ˇ -premanifolds
F W M ! N the diagram
.F˛ /
DerK .OM˛ ;p ; K/ DerK .ON˛ ;F .p/ ; K/
M N
F
DerK .OM;p ; K/ DerK .ON;F .p/ ; K/
commutes. The functoriality shows in particular that if we choose a chart .U; x/ at p, then
the i-th partial derivative @x@ i for M˛ along the i-th coordinate (i D 1; : : : ; dimp .M /) is
sent to the i-th partial derivative for M . As these derivations form a basis of the tangent
spaces, one sees that M induces an isomorphism
Tp .M˛ / ! Tp .M /; (5.8)
which we use in the sequel to identify these tangent spaces. The functoriality of M also
shows that for a morphism of C ˇ -premanifolds F W M ! N the derivatives of F and of
F˛ are the same via the identification (5.8).
5.2 Local Properties of Morphisms, Local Isomorphisms, Immersions, Submersions 99
Remark 5.13. Let M be a complex premanifold, let MR be the underlying real analytic
manifold, and let MRIC WD .M; OMR IC / ! MR and MRIC ! M be the canonical
morphisms (Remark 4.24). Let p 2 M and let mp , mC;p and mR;p be the maximal ideal
of OM;p , OMR IC;p and CM!
R ;p
, respectively. Then mC;p D mR;p ˝R C and hence we
obtain a functorial isomorphism of C-vector spaces
Tp .MRIC / WD .mC;p =m2C;p /_ D .mR;p =m2R;p /_ ˝R C D Tp .MR / ˝R :C (*)
On the other hand, the inclusion OM ,! OMR IC sends mp to mC;p and hence induces
a functorial C-linear map W Tp .MRIC / ! Tp .M /. Let
W Tp .MR / ! Tp .M /
be the composition of the R-linear injective map Tp .MR / ! Tp .MRIC /, 7! ˝ 1,

followed by . Then is a functorial R-linear map. We claim that is an isomorphism.
Indeed, let .U; zW U ! C m / be a chart of M and p with z.p/ D 0. Let zpi be the image
of the i-the coordinate function in mp =mp2 , and let xpi and ypi be the image of Re.z i /
and Im.z i / in mR;p =m2R;p . Lemma 5.10 shows that .zp1 ; : : : ; zpm / is a C-basis of mp =mp2
and .xp1 ; yp1 ; : : : ; xpm ; ypm / is an R-basis of mR;p =m2R;p . Then .zp1 ; zNp1 ; : : : ; zpm ; zNpm / is an
R-basis of mp =mp2 . As the standard real analytic coordinate change .z i ; zN i / 7! .x i ; y i /
is an isomorphism, we see that the dual map of is an isomorphism. Hence is an
isomorphism.
In the sequel we will use to identify Tp .MR / with the underlying real vector space
of the C-vector space Tp .M /. The functoriality of shows that if F W M ! N is a holo-
morphic map, then via this identification of tangent spaces the derivatives of F and of FR
coincide.
5.2 Local Properties of Morphisms, Local Isomorphisms, Immersions,

Submersions
Local properties may be checked via charts.
Remark 5.14 (Chart principle). If P is a property of a continuous map F between pre-

manifolds M and N that is local on source and target (Appendix Sect. 12.5) and that
is stable under composition with isomorphisms of manifolds, then a continuous map
F W M ! N has the property P if and only if for every p 2 M there exists a chart
.U; x; V; y; FQ / of F at p such that FQ has the property P.
Examples are the properties P of being open or being a morphism of premanifolds.

Important applications of the chart principle are the following two results.
Proposition 5.15. Let F W M ! N be a morphism of premanifolds. Then F is locally

constant if and only if Tp .F / D 0 for all p 2 M .
Proof. Both properties (“locally constant” and “Tp .F / D 0 for all p”) are local on source
and target and stable under composition with isomorphisms (for the second property this
follows from the chain rule, Remark 5.4). Hence by choosing charts for F we may as-
sume that M and N are open in some K-vector space. Then the claim follows from the
analogous result of local analysis (Appendix Proposition 16.6).
The chart principle also yields immediately the following theorem of inverse functions.
Theorem 5.16. Let F W M ! N be a morphism of premanifolds, let p 2 M . Then F is

a local isomorphism at p if and only if Tp .F / is an isomorphism of K-vector spaces.
Proof. Again by choosing charts for F we may assume that M and N are open in some K-
vector space, and the claim follows from the analogous result of local analysis (Appendix
Theorem 16.16).
Definition 5.17. Let F W M ! N be a morphism of premanifolds, let p 2 M and let

Tp .F /W Tp .M / ! TF .p/ .N / be the induced tangent map.
1. rkp .F / WD rk.Tp .F // is called the rank of F at p.

2. F is called an immersion at p if Tp .F / is injective. F is called an immersion if it is an
immersion at every p 2 M .
3. F is called an submersion at p if Tp .F / is surjective. F is called an submersion if it
is a submersion at every p 2 M .
Remark 5.18.
1. The chain rule (5.3) implies that the composition of two submersions (respective of two
immersions) is again a submersion (respectively an immersion). Moreover if F ı G is
a submersion and G is surjective (respectively if F ı G is an immersion), then F is
a submersion (respectively then G is an immersion).
2. By Theorem 5.16 a morphism of premanifolds is a local isomorphism if and only if it
is a submersion and an immersion.
Proposition
˚ W M ! N be a morphism of premanifolds and fix r 2 N0 . Then
5.19. Let F
p 2 M I rkp .F / r is open in M .
Proof. Again the chart principle allows us to assume that M Km , N Kn open.

As F is a C 1 -map, the map M 3 p 7! Tp .F / 2 HomK .Km ; Kn / is continu-
ous, and f A 2 HomK .Km ; Kn / I rk.A/ n g is open because via the identification
HomK .Km ; Kn / D Mnm .K/ its complement is the closed set of all matrices whose
all r-minors vanish.
5.3 Morphisms of Locally Constant Rank 101
Corollary 5.20. Let F W M ! N be a morphism of premanifolds.
1. The set U WD f p 2 M I F is an immersion at pg is open in M , and rkp .F / D

dimp .M / for all p 2 U .
2. The set V WD f p 2 M I F is a submersion at pg is open in M , and rkp .F / D
dimF .p/ .N / for all p 2 V .
5.3 Morphisms of Locally Constant Rank
In this section we will show that morphisms of locally constant rank are locally linear and
deduce some properties of such morphisms.
Corollary 5.20 shows that for immersions and for submersions F the map U 3 p 7!
rkp .F / is locally constant. In other words, immersions and submersions are examples of
morphisms of locally constant rank in the following sense.
Definition 5.21. Let F W M ! N be a morphism of premanifolds and let p 2 M . Then

F is called of locally constant rank at p if there exists p 2 W M open such that
W ! N0 , q 7! rkq .F / is constant. The morphism F is called of locally constant rank if
it is of locally constant rank at all p 2 M .
Clearly, the set of points in M , where F is locally of constant rank, is open in M . One
can show that it is also dense in M (Problem 5.6).
The main result about morphisms of locally constant rank shows that they have partic-
ularly simple charts.
Theorem 5.22 (Rank theorem). Let F W M ! N be a morphism of premanifolds and

let p 2 M , m WD dimp .M /, n WD dimF .p/ .N /, k 2 N0 . Assume that there exists
p 2 W M open such that rkq .F / D k 2 N0 for all q 2 W . Then there exists a chart
of F at p
F jU
M U V N
x Š Š y
FQ
Km UQ VQ Kn ;
such that FQ .x1 ; : : : ; xm / D .x1 ; : : : ; xk ; 0; : : : ; 0/ for all .x1 ; : : : ; xm / 2 UQ .

Proof. Let .U; x; V; y; FQ / be a chart of F at p. By replacing U by U \ W we can assume

U W . Due to Remark 5.3, we obtain a commutative diagram of K-vector spaces
Tq .F /
Tq .M / TF .q/ .N /
Tq .x/ Š Š TF .q/ .y/
Tx.q/ .FQ /
Tx.q/.UQ / Ty.F .q//.VQ /;
for all q 2 U W . In particular, the rank of FQ at x.q/ 2 UQ has to be k for every q 2 U

so that UQ 3 qQ 7! rkqQ FQ is constant (with value k) because x.U / D UQ . Therefore we
may assume p 2 M Km , N Kn open and F W M ! N is a morphism of manifolds
of constant rank. Composing x with a permutation of coordinates we may also rearrange
coordinates.
Denote all elements q 2 M Km as q D .x; y/ where x 2 Kk ; y 2 Kmk and set

@F @Fi
.x; y/ WD .x; y/ 2 Mnk .K/;
@x @xj 1i n;1j k

@F @Fi
.x; y/ WD .x; y/ 2 Mn.mk/ .K/
@y @yj 1i n;1j mk
for all morphisms of manifolds F between open subsets of Km and Kn . Moreover,

let F 0 W M ! Kk and F 00 W M ! Knk be the morphisms of manifolds with F .q/ D
0
.F 0 .q/; F 00 .q// for q 2 M . As JF .p/ has rank k, we may assume that @F
@x
.p/ is invertible
after rearranging the coordinates. With this notation we have for the Jacobi matrix
!
@F 0 @F 0
@x .x; y/ @y .x; y/
JF .x; y/ D @F 00 @F 00
2 Mnm .K/;
@x
.x; y/ @y
.x; y/
where the upper-left matrix is an invertible square matrix of size k for .x; y/ D p.
Q M ! Km , .x; y/ 7! .F 0 .x; y/; y/. Then the Jacobi matrix of 'Q has for all
Define 'W
q 2 M the form !
@F 0
.q/
J'Q .q/ D @x
0 Imk
so that J'Q .p/ 2 GLm .K/. By the inverse function theorem (Appendix Theorem 16.16)
there exists an open neighborhood U of p such that ' WD 'Q jU W U ! '.UQ / DW UQ Km
is an isomorphism of manifolds. By shrinking U we may assume that UQ is connected and
of the form UQ D U 0 U 00 , where U 0 Kk , U 00 Kmk are open subsets. Then U 0 and
U 00 are connected (Appendix Proposition 12.41).
5.3 Morphisms of Locally Constant Rank 103
For .x; y/ 2 UQ we have F .' 1 .x; y// D .x; g.x; y// where gW UQ ! Knk is defined
as F 00 ı ' 1 . Hence we find
!
Ik 0
JF ı' 1 .x; y/ D @g @g :
@x
.x; y/ @y
.x; y/
We have JF ı' 1 .x; y/ D JF .' 1 .x; y//J' 1 .x; y/ by the chain rule. As J' 1 .x; y/ is
invertible, rk.JF ı' 1 .x; y// D rk.JF .' 1 .x; y/// is equal to k on UQ . Thus we must have
@g 0 00 00
@y .x; y/ D 0 for every .x; y/ 2 U U and since U is connected, g does only depend
on x (Appendix Proposition 16.6).
Finally define V WD U 0 Knk and W V ! V , .u; v/ ! .u; v g.u//. This map is
an isomorphism of manifolds and for .x; y/ 2 UQ we find
FQ .x; y/ WD .F .' 1 .x; y/// D .x; g.x// D .x; g.x/ g.x//

D .x1 ; : : : ; xk ; 0; : : : ; 0/
as desired.
Corollary 5.23. Let F W M ! N be a morphism of premanifolds, let p 2 M , m WD

dimp .M /, n WD dimF .p/ .N /. Assume that F is an immersion (respectively a submersion)
at p. Then m n (respectively m n) and there exists a chart .U; x; V; y; FQ / of F
at p such that FQ .x1 ; : : : ; xm / D .x1 ; : : : ; xm ; 0; : : : ; 0/ (respectively FQ .x1 ; : : : ; xm / D
.x1 ; : : : ; xn /).
Proof. If F is an immersion at p, then there exists an open neighborhood U of p such

that F is an immersion at q for all q 2 U and such that U 3 q 7! rkq .F / is constant
(Corollary 5.20). Now we can apply the rank theorem.
The same argument shows the claim if F is a submersion at p.
Corollary 5.24. Let F W M ! N be a submersion of premanifolds. Then F is an open

map.
Proof. By the chart principle and by Corollary 5.23 we may assume that F is a projection
map .x1 ; : : : ; xm / 7! .x1 ; : : : ; xn / with n m. This is an open map.
Proposition 5.25. Let F W M ! N be a morphism of premanifolds and let Z be a third

premanifold.
1. Suppose that F is an immersion. Then a continuous map GW Z ! M is a morphism

of premanifolds if and only if F ı G is a morphism of premanifolds.
2. Suppose that F is a surjective submersion. Then a map GW N ! Z is a morphism of
premanifolds if and only if G ı F is a morphism of premanifolds.
The hypothesis in 1 that G is continuous cannot be omitted (Problem 5.11), see how-
ever Problem 5.12.
Proof. Let us show 1. The property of a continuous map being a morphism of preman-
ifolds is local on source and target. Hence by the chart principle we may assume that
Z Kz , M Km and N Kn are open. As F is an immersion, we may also assume
by Corollary 5.23, that F is of the form .x1 ; : : : ; xm / 7! .x1 ; : : : ; xm ; 0; : : : ; 0/. Hence
it remains to show that a continuous map GW Z ! Km is a C ˛ -map if and and only if
Z ! Kn , z 7! .G.z/; 0; : : : ; 0/ is a C ˛ -map. But this is clear.
A similar argument shows Assertion (2) once we have seen that G is continuous. But
for V Z open, we have G 1 .V / D F .F 1 .G 1 .V /// because F is surjective and this
is open in N because F is open (Corollary 5.24).
5.4 Submanifolds
For every subspace S of a premanifold on M we define a function on S to be a map

S ! K that is locally the restriction of a morphism U ! K for some U M open. We
obtain the sheaf OS of functions on S, and S is be defined to be subpremanifold if .S; OS /
is a premanifold. Then the main result (Theorem 5.34) will be that a subpremanifold is
always isomorphic to an embedding in the following sense.
Definition 5.26. A morphism W M ! N of premanifolds is called an embedding if is

a topological embedding (Appendix Definition 12.19) and an immersion.
Example 5.27. Any morphism W M ! N of premanifolds that is a section of a morphism

W N ! M (i.e., ı D idM ) is an embedding: It is a topological embedding by Appendix
Remark 12.20 4 and for any p 2 M one has T.p/ ./ ı Tp ./ D Tp .idM / D idTp .M / , which
shows that is an immersion.
In particular for every morphism of premanifolds F W M ! N its graph morphism
F W M ! M N , m 7! .m; F .m//, is an embedding. It is a section of the projection
M N ! M.
Remark 5.28. An embedding is an injective immersion. The converse is not true. We

give two examples:
1. 'W .=4; =2/ ! R2 , t 7! .sin 2t/. sin.t/; cos.t//. This is shown in Fig. 5.2.
2. Endow Q with the discrete topology (which makes Q into a 0-dimensional smooth
manifold). Then the inclusion Q ,! R is an injective immersion (here R is endowed
with the usual structure of a 1-dimensional smooth manifold).
5.4 Submanifolds 105
0,75
0,5
0,25
-1,25 -1 -0,75 -0,5 -0,25 0 0,25 0,5 0,75 1 1,25
-0,25
-0,5
-0,75
Figure 5.2 Locally around the origin the image of ' is not homeomorphic to an open interval
Remark 5.29. The composition GıF of two embeddings F and G is again an embedding
because G ıF is a topological embedding by Appendix Remark 12.20 3 and an immersion
by the chain rule.
Definition and Remark 5.30 (Submanifolds). Let .M; OM / be a premanifold.
1. Let S M be a subspace. A function on S (C ˛ -function if M is real C ˛ -manifold;

holomorphic function if M is complex manifold) is a map f W S ! K such that for all
s 2 S there exists s 2 U M open and g 2 OM .U / such that g jU \S D f jU \S .
Define a sheaf OS of K-valued functions by
OS .V / WD ff W V ! K function on V g; V S open:
Then .S; OS / is a K-locally ringed space (all stalks OS;s are local K-algebras with
residue field K by the same argument as in Remark 3.17).
2. A (sub)premanifold of M is a subspace S of M such that .S; OS / is a (pre)manifold.
Remark 5.31. Let M be a premanifold, S M a closed subspace and iW S ! M the

inclusion. Let IS be the sheaf of functions vanishing on S, i.e., for U M open we
˚
define IS .U / WD f 2 OM .U / I f jS D 0 . Then we have an exact sequence of sheaves
of abelian groups

0 ! IS ! OM ! i OS ! 0; (5.9)
where is the inclusion and is given by OM .U / 3 g 7! g jU \S 2 i OS .U /. To show

the exactness of (5.9) we have to show that the induced sequence of stalks is exact. Fix
x 2 X. As X n S is open in X, .i OS /x D 0 and .IS /x D OX;x for x 2 X n S. Hence we
may assume that x 2 S. Then f 2 .i OS /x D OS;x is represented by a function .U; f /,
where U is a neighborhood of x and where f W U \S ! K is the restriction of a morphism
gW U ! K. Hence we see that x W OM;x ! .i OS /x is surjective. Elements of its kernel
are represented by morphisms gW U ! K (U some neighborhood of x depending on g)
whose restriction to U \ S is zero, i.e., the kernel of x is IS;x .
Once we have the general formalism of OM -modules available (Sect. 8.3), one checks
immediately that IS is an ideal of OM and that (5.9) is an exact sequence of OM -modules.
The next goal is to show that a subspace of a premanifold is a subpremanifold if and

only if it is the image of an embedding. Such subpremanifolds are also often called regular
submanifolds. There are useful more general classes of submanifolds (see Problem 5.16
or Problem 5.18), which will not be used here.
We start with some easy remarks.
Remark 5.32. Let M be a premanifold, S M a subspace.
1. If S is open in M , then .S; OS / is the open subpremanifold of M defined in Defini-

tion 4.16.
2. If S is a discrete subspace of M , then S is a 0-dimensional subpremanifold of M .
3. Let ˚W M ! N be an isomorphism of premanifolds. Then S is a subpremanifold of
M if and only if ˚.S/ is a subpremanifold of N .
S
4. Let Ui M be open for i 2 I such that S i Ui . Then S is a subpremanifold of
M if and only if S \ Ui is a subpremanifold of Ui for all i 2 I .
As an immersion looks locally as the inclusion of a subspace into a vector space, we

study this case first.
Example 5.33. Let M Km be open and W Km be a K-subspace. Let us show that

S WD W \ M is a submanifold of M of dimension dim.W /.
Clearly, S W is open. We may assume that S ¤ ;. After applying a K-linear
automorphism of Km (which is a bianalytic isomorphism) we may also assume that W D
f .x1 ; : : : ; xk ; 0; : : : ; 0/ 2 Km I xi 2 K g for some 0 k m.
Now there are two structures of a K-ringed space on S. One as an open submanifold
of W , call its structure sheaf C˛S . Then .S; C˛S / is a C ˛ -manifold. The other is the sheaf
OS defined in Definition 5.30.
We claim that OS D C˛S . Let V S be open. We have to show that a function

f W V ! K is C ˛ if and only if for every p 2 V there exists p 2 U Rm open and a
C ˛ -function gW U ! K such that g jU \V D f jU \V .
Clearly we have OS .V / C˛S .V /. Conversely, let f W V ! K be a C ˛ -function. Let
U WD f .x1 ; : : : ; xk ; tkC1 ; : : : ; tm / 2 Km I x D .x1 ; : : : ; xk / 2 V g \ M:
Then U M is open and U \ S D V . Define
gW U ! K; g.x1 ; : : : ; xk ; tkC1 ; : : : ; tm / WD f .x1 ; : : : ; xk /:
Then g 2 CM˛ .U / with g jV D f . Therefore f 2 OS .V /.
Theorem 5.34. Let M be a premanifold.
1. Let S M be a subpremanifold. Then S M is locally closed and the inclusion

W S ,! M is an embedding.
2. Conversely, let iW N ! M be an embedding. Then S WD i.N / is locally closed in M ,
S is a subpremanifold and iW N ! S is an isomorphism of premanifolds.
Proof. 2. By Corollary 5.20 we may assume that p 7! rkp .i/ is constant and that N is
of constant dimension n. Let p 2 N . By the rank theorem there exists p 2 V D Vp N
open and a commutative diagram
i jV
N V U M
y Š Š x
{QW.x1 ;:::;xn /7!.x1 ;:::;xn ;0;:::;0/
Kn VQ UQ Km ;
where U D Up open in M .
As i is a homeomorphism N ! S, i.V / is open in S, say i.V / D U 0 \S with U 0 M
open. Replacing U by U \ U 0 and UQ by x.U \ U 0 /, we can assume that U \ S D i.V /.
Similarly we may assume {Q.VQ / D UQ \ .Kn f0g/, hence {Q.VQ / UQ closed and therefore
i.V / closed in U . Hence U is an open neighborhood of i.p/ such that U \ S is locally
closed in U . As p was arbitrary, this shows that S is locally closed.
Moreover, i.V / is a submanifold of U because {Q.VQ / is a submanifold of UQ . As the Up
for p 2 N cover S, this implies that S is a subpremanifold of M .
Finally, the chart shows that for each Up the map iW i 1 .Up \ S/ D Vp ! Up \ S is an
isomorphism of manifolds. Hence iW N ! S is an isomorphism of premanifolds because
being an isomorphism can be checked locally on the target.
1. The map is a morphism of premanifolds: Let U M be open, f 2 OM .U /. We
have to show that f ı .j1 .U / / D f jU \S 2 OS .U \ S/. This holds by definition of OS .
The map is an embedding: we have to show that is an immersion. Let p 2 S. By

definition of OS the homomorphism of K-algebras
p] W OM;p ! OS;p ; Œ.f; U / 7! Œ.f jU \S ; U \ S/
is surjective. Hence the K-linear map
W DerK .OS;p ; K/ ! DerK .OM;p ; K/
is injective. Therefore Tp ./ is injective by Remark 5.7.

Hence is an embedding and hence S is locally closed in M by 2.
The proof of Theorem 5.34 2 shows that there is also the following more classical
characterization of a subpremanifold.
Remark 5.35. A subspace S of M is a subpremanifold if and only if for all p 2 S there

exists a chart xW U ! Km of M at p such that x.U \ S/ D x.U / \ L, where L is
a subspace of Km (here x can be chosen such that L D Kd f0g with d D dimp .S/).
This description in particular shows:
Corollary 5.36. Let M be a premanifold and let S be a subpremanifold. Then S is open

in M if and only if dimp .S/ D dimp .M / for all p 2 S.
Corollary 5.37. Let M be a (pre)manifold, S a sub(pre)manifold of M , and let T be

a sub(pre)manifold of S. Then T is a sub(pre)manifold of M .
Proof. Let iW S ! M and j W T ! S be the inclusions. By Theorem 5.34 1 these are

embeddings. Hence i ı j is an embedding and therefore T is a subpremanifold of M by
Theorem 5.34 2.
Proposition 5.38. Let F W M ! N be a morphism of premanifolds, let S M and

T N be subpremanifolds such that F .S/ T . Then the induced map F jS W S ! T is
a morphism of premanifolds.
Proof. Let iW S ! M and j W T ! N be the inclusions. Then F ı i D j ı F jS . Hence

j ı F jS is a morphism. As F jS is continuous and j is an immersion, this implies that F jS
is a morphism by Proposition 5.25.
If F W M ! N is a morphism of premanifolds, then a fiber of F is in general not

a subpremanifold of M . For instance the union of the coordinate axes in K2 is the fiber
F 1 .0/ of the analytic map K2 ! K, .x; y/ 7! xy, and this is not a submanifold of
K2 (see Problem 5.13). The following result and its corollaries give criteria under which
a fiber of a morphism of premanifolds is a subpremanifold.
Proposition 5.39. Let F W M ! N be a morphism of premanifolds, let q 2 N , and

assume that the rank of F is locally constant on an open neighborhood of F 1 .q/. Then
F 1 .q/ is a subpremanifold of M . For p 2 F 1 .q/ one has
Tp .F 1 .q// D ker.Tp .F /W Tp .M / ! Tq .N //; (5.10)
in particular dimp .F 1 .q// D dimp .M / rkp .F /.
Proof. By the rank theorem 5.22 we can choose for all p 2 F 1 .q/ a chart .U; x; V; y; FQ /
of F at p such that x.p/ D 0, the rank of F on U is constant, and FQ is given by
Km UQ 3 .x1 ; : : : ; xm / 7! .x1 ; : : : ; xk ; 0 : : : ; 0/ 2 VQ Kn :
By Remark 5.32 4 it suffices to show that U \F 1 .q/ is a submanifold. By Remark 5.32 3

we may assume that F D FQ . But then F 1 .q/ D FQ 1 .0/ D UQ \ W with W D
f.0; : : : ; 0; xkC1 ; : : : ; xm / 2 Km g and the claim follows from Example 5.33.
Corollary 5.40. Let F W M ! N be a morphism of premanifolds, let q 2 N , and assume

that F is a submersion at p for all p 2 F 1 .q/. Then F 1 .q/ is a submanifold of M with
dimp .F 1 .q// D dimp .M / dimq .N / for all p 2 F 1 .q/.
A point q 2 N such that F is a submersion at p for all p 2 F 1 .q/ is called a regular

value. Note that all q 2 N n f .M / are regular values.
Proof. By Corollary 5.20 there exists an open neighborhood U of F 1 .q/ such that F jU
is a submersion and in particular F jU has locally constant rank. Hence we may apply
Proposition 5.39.
Corollary 5.41. Let U Km be open and let F W U ! K be a C ˛ -function. Let q 2 K

such that for all p 2 F 1 .q/ there exists 1 i m with @x
@F
i
.p/ ¤ 0. Then M WD F 1 .q/
is a C -submanifold of U (and hence of K ) of dimension m 1 with
˛ m
( )
X
m
@F
Tp .M / D y2K I m
yi .p/ D 0 :
i D1
@xi
Example 5.42. Let Q be a non-degenerate quadratic form on Rm . Then
S WD f x 2 Rm I Q.x/ D 1 g
is an analytic submanifold of Rm : By the classification of real quadratic forms we may

assume that Q.x/ D x12 C C xk2 xkC1
2
xm
2
after a linear change of coordinates.
Figure 5.3 Quadrik
As a polynomial function, Q is real analytic. For all p D .p1 ; : : : ; pm / 2 Rm we have

@i Q.p/ ¤ 0 if pi ¤ 0. Hence we can apply Corollary 5.41 to F D Q and q D 1. Fig. 5.3
shows the set
˚
x 2 R3 I x12 C x22 x32 D 1 :
In particular we see that the n-sphere
˚
S n WD .x1 ; : : : ; xnC1 / 2 RnC1 I x12 C C xnC1
2
D1
is a real analytic submanifold of RnC1 with

( )
X
nC1
Tp .S / D
n
.y1 ; : : : ; ynC1 / 2 R nC1
I yi xi D 0 ; p D .x1 ; : : : ; xnC1 / 2 S n :
i D1
We conclude this section by stating several important results that we will neither prove
nor use in the sequel.
Remark and Definition 5.43. The first is Sard’s Theorem, which is the fact that for
a morphism F W M ! N of manifolds “most” values are regular. More precisely, let N
be a second countable premanifold. Then every atlas has a countable subatlas (Proposi-
tion 1.4 1). We say that a subset Z N is a zero set if there exists a countable atlas
.Un ; ˚n /n2N of N such that ˚n .Z \ Un / is contained in a subset of Km that has Lebesgue
measure 0. Then Z has this property for all countable atlases (as the image of a set of mea-
sure zero under a C 1 -diffeomorphism is again a set of measure zero by the transformation
formula).
We can now state the theorem of Sard (see [Ste] II, §3 for a proof for real preman-
ifolds, the case of complex premanifolds can be reduced to the real analytic case by
Remark 5.13).
Theorem 5.44 (Theorem of Sard). Let M and N be second countable premanifolds and
let F W M ! N be a morphism of premanifolds. Suppose that M and N are either complex
premanifolds or that M and N are real C ˛ -premanifolds such that ˛ maxfdimF .p/ N
dimp M C 1; 1g for all p 2 M . Then the set of non-regular values of F is a subset of
measure zero in N .
Remark 5.45. Sometimes it can be useful to know whether a premanifold M can be

embedded into the standard affine space KN for some N . As subspaces of Hausdorff (re-
spectively second countable) spaces are again Hausdorff (respectively second countable)
a necessary criterion certainly is that M is a manifold. We list the following results.
Suppose that M is a connected real C ˛ -manifold of dimension n 2 N0 with ˛ 2 N. b
Then there exists a closed embedding iW M ,! RN ([Whi1] IV, Theorem 1A for 1 ˛
1 with N D 2n; the cases 2 ˛ 1 with N D 2n C 1 follow quite easily from Sard’s
Theorem, see [Ste] II Theorem 4.4; for ˛ D ! this is shown in [Gra] Theorem 3).
Assume that M is a complex manifold. Then there do not exist holomorphic embed-
dings into C N in general. For instance, if M is compact and connected, then for all N
every holomorphic map M ! C N is constant (Problem 4.15). Complex manifolds M
such that there exists N 2 N and a closed holomorphic embedding iW M ! C N are
called Stein manifolds1 . Every non-compact connected complex manifold of dimension 1
is a Stein manifold ([GuRo] Chap. IX, Sect. B, Theorem 10).
If M is a compact complex manifold, one can ask whether there exists an embedding
of M into some other nice “standard space”, for instance an embedding M ,! P N .C/
(automatically closed) for some N 2 N. But a result of Chow ([GuRo] Chap. V, Sec. D,
Theorem 7) shows that if there exists such an embedding, then M is projective, i.e., one
can find homogeneous polynomials f1 ; : : : ; fr 2 CŒT0 ; : : : ; TN such that
˚
M D z D .z0 W W zN / 2 P N .C/ I f1 .z/ D D fr .z/ D 0 :
This shows that compact manifolds that are embeddable in projective space are rather
special. For instance complex tori T (i.e., T D V =, where V is a finite-dimensional C-
vector space of dimension g and D Zv1 ˚ ˚Zv2g for R-linearly independent vectors
v1 ; : : : ; v2g 2 V ) are almost never embeddable into projective space if dim.T / > 1 (more
1
This is not the usual definition of a Stein manifold, and the fact that this property is equivalent
to the usual definition is a deep theorem by Remmert ([GuRo] Chap. VII, C, Theorem 13); it also
shows that one can always choose N D 2n C 1 if M is of (complex) dimension n 2 N.
precisely the locus of projective complex tori in the g 2 -dimensional space classifying all
complex tori has dimension g.g C 1/=2).
5.5 Fiber Products of Manifolds
In Corollary 5.40 we saw a criterion for the fiber of a submersion to be a submanifold. This
is a special case of the following existence criterion of fiber products of premanifolds. Let
F W M ! S and GW N ! S be morphisms of premanifolds and define a set
M S N WD f .p; q/ 2 M N I F .p/ D G.q/ g : (5.11)
We endow M S N with the subspace topology of M N . In other words, M S N is the

fiber product of F and G in the category of topological spaces (Appendix Example 13.46).
If G is the inclusion of a subspace T of S, then M S T D F 1 .T / is the inverse image.
In particular, if T consists of a single point s 2 S, then M S fsg D F 1 .s/ is the fiber in
s.
The goal in this section is to give a criterion when M S N is a subpremanifold of
M N . In this case we call M S N the fiber product premanifold of F and G. It follows
then from Proposition 5.38 that M S N is in fact a categorical fiber product (Appendix
Definition 13.43) in the category of premanifolds. It is even a fiber product in the category
of all K-ringed spaces (Problem 5.20).
Definition 5.46. Let F W M ! S and GW N ! S be morphisms of premanifolds and let

z D .p; q/ 2 M N with F .p/ D G.q/ DW s. Then F and G are called transversal at z
if Tp .F / C Tq .G/W Tp .M / ˚ Tq .N / ! Ts .S/ is surjective.
Theorem 5.47. Let F W M ! S and GW N ! S be morphisms of premanifolds, let

Z WD M S N and suppose that F and G are transversal at all points z of Z. Then
Z is a subpremanifold of M N and for all z D .p; q/ 2 Z we have
˚
Tz Z D . ; / 2 Tp M Tq N I Tp .F /. / D Tq .G/./ : (5.12)
If G is the inclusion of a point s 2 S, then F is transversal to G in all points of

Z D F 1 .s/ if and only if F is a submersion in all points of Z. Hence in this special case
the theorem is just Corollary 5.40. In the proof we will reduce to this case.
If G is the inclusion of a submanifold of S, then we identify Z with F 1 .N / and (5.12)
can be written as
Tp .F 1 .N // D Tp .F /1 .TF .p/ .N //; p 2 F 1 .N /: (5.13)

5.5 Fiber Products of Manifolds 113
Proof. (i). We start with the following remark. Let .Ui /i be an open covering of S
and set Zi WD F 1 .Ui / Ui G 1 .Ui /. Then .Zi /i is an open covering of Z and Z is
a subpremanifold of M N if and only if Zi is a subpremanifold of F 1 .Ui / G 1 .Ui /.
As also the transversality condition is local on S, we may prove the theorem locally on S.
(ii). Now we show that we may assume that G is the embedding of a submanifold
of S. Indeed, we may write Z D .F G/1 .S /, where S S S is the diagonal
(a submanifold by Example 5.27). Moreover, we claim that F and G are transversal in
.p; q/ 2 Z if and only if F G is transversal to the diagonal of S S in .p; q/. Indeed,
set P WD Tp .F /.Tp M /, Q WD Tq .G/.Tq N /, T WD Ts S, where s D F .p/ D G.q/.
Then the claim follows from the easy linear algebra fact that P C Q D T if and only if
.P ˚ Q/ C T D T ˚ T , where T is the diagonal in T ˚ T . Finally
˚
. ; / 2 Tp M Tq N I Tp .F /. / D Tq .G/./
˚
Š . ; ; / 2 .Tp M Tq N / Ts S I .Tp .F /. /; Tq .G/.// D .; / ;
which shows that (5.12) holds if and only if the corresponding description of the tangent
space of the fiber product of F G and S holds. Hence from now on we may assume
that G is a submanifold of S.
(iii). We now reduce to the case that G is the embedding of a single point of s. By
(i) we may work locally on S. Hence by Remark 5.35 we may assume there exists an
open embedding ˚W S ! Ks and a subspace L of Ks such that ˚.N / D ˚.S/ \ L.
Let f W S ! W WD Ks =L be the composition of ˚ with the projection Ks ! W . Then
f 1 .0/ D N and hence F 1 .N / D .f ı F /1 .0/. For p 2 F 1 .N / the morphism F is
transversal to N ! G if and only if Tp .˚ ı F /.Tp M / C L D Ks , i.e., if and only if f ı F
is a submersion in p. Hence we may replace F by f , S by W , and N by f0g.
(iv). Now we apply Corollary 5.40. We obtain that F 1 .N / is a subpremanifold of M
and hence that the graph of F j F 1 .N / is a subpremanifold of M N . By (5.10) we find
Tp .F 1 .N // D ker.Tp .f /W Tp .M / ! W /
D Tp .˚ ı F /1 .L/ D Tp .F /1 .TF .p/ .N //:
Remark 5.48. Let F W M ! S and GW N ! S be morphisms of premanifolds and let

Z WD M S N . The reduction steps in the proof of Theorem 5.47 show that the open-
ness of a set of points where a morphism is a submersion (Corollary 5.20) implies that
f z 2 Z I F and G are transversal in zg is open in Z.
Remark 5.49. Let F W M ! S and GW N ! S be morphisms of premanifolds, let Z WD

M S N , let W Z ! M and $ W Z ! N be the projections, and suppose that F and
G are transversal at all points z of Z. Then (5.12) implies that for z D .p; q/ 2 Z,
s WD F .p/ D G.q/ the commutative diagram
Tz ./
Tz Z Tp M
Tz .$/ Tp .F /
Tq .G/
Tq .N / Ts S
induces isomorphisms
Ker.Tz .// ! Ker.Tq .G//;

Coker.Tz .// ! Coker.Tq .G//
(and similarly for Tz .'/ and Tp .F /).
As submersions are clearly transversal to all other morphisms, Theorem 5.47 and Re-
mark 5.49 imply Corollary 5.50.
Corollary 5.50. Let F W M ! S and GW N ! S be morphisms of premanifolds and

suppose that F is a submersion. Then M S N is a subpremanifold of M N and the
projection M S N ! N is a submersion.
5.6 Quotients of Manifolds
Let M be a premanifold and let R M M be an equivalence relation, i.e., R is a subset

of M M such that for m; m0 ; m00 2 R one has
(a) .m; m0 / 2 R, .m0 ; m00 / 2 R ) .m; m00 / 2 R,

(b) .m; m/ 2 R,
(c) .m; m0 / 2 R ) .m0 ; m/ 2 R.
We denote by M=R the set of equivalence classes. Let pW M ! M=R be the projection
and endow M=R with the quotient topology, i.e., V M=R is open if and only if p 1 .V /
is open in M .
We study the question when M=R can be endowed with the structure of a premanifold.
Proposition 5.51. Let N be a topological space, M a premanifold and pW M ! N

a surjective map. Then there exists at most one structure of a premanifold on N such that
p is a submersion.
5.6 Quotients of Manifolds 115
Proof. Let N1 and N2 be premanifolds with underlying topological space N such that the
morphisms p1 W M ! N1 and p2 W M ! N2 with underlying map p are submersions. Now
idW N1 ! N2 is a continuous map and we clearly have id ıp1 D p2 as maps of topological
spaces. By Proposition 5.25 2 idW N1 ! N2 is already a morphism of premanifolds. By
the same argument idW N2 ! N1 is a morphism of premanifolds. Therefore N1 and N2
carry the same premanifold structure.
If there exists on N the structure of a premanifold such that p is a submersion, then

Proposition 5.25 2 shows that the structure sheaf of N can be described by
˚
ON .V / WD f W V ! K map I f ı p jp1 .V / 2 OM .p 1 .V // :
Remark and Definition 5.52 (Quotient manifold). Let M be a premanifold and let
R M M be an equivalence relation. Then Proposition 5.51 shows that there exists
at most one structure of a premanifold on M=R such that the projection pW M ! M=R is
a submersion. In this case the premanifold M=R is called the quotient premanifold of M
by R and the equivalence relation R is called regular.
Theorem 5.53. Let M be a premanifold and let R M M be an equivalence relation.

Then the following assertions are equivalent:
(i) There exists the structure of a premanifold on M=R such that the projection pW M !
M=R is a submersion.
(ii) R is a submanifold of M M and the restriction pr1 W R ! M of the first projection
M M ! M is a submersion.
If these assertions hold, M=R is Hausdorff if and only if R is closed in M M .
Proof. “(i) ) (ii)”. Assume that M=R is a premanifold in such a way that the projection
pW M ! M=R is a submersion. Then we have R D M M=R M as topological spaces.
Hence, Corollary 5.50 implies (ii).
“(ii) ) (i)”. Conversely, let R be a submanifold of M M such that pr1 W R ! M
is a submersion. Notice that this implies that pr2 W R ! M is also a submersion because
W R ! R; .x; y/ 7! .y; x/ is an isomorphism with pr2 D pr1 ı .
Idea of proof. The idea to construct the premanifold structure on M=R is rather simple:
If we already had constructed this structure and knew that p is a submersion, then locally
on an open subspace U of M the morphism p would have a section s (because locally
a submersion looks like a surjective linear map that clearly has a section). Then the image
of s is a submanifold N of U isomorphic to p.U / and every point of U is equivalent to
precisely one point of N . Hence we will construct U and N and then endow p.U / with
the premanifold structure of N .
To do this, we again think backwards: If we had already constructed M=R, then for
x0 2 M the tangent map Tx0 .p/ would be surjective and hence the restriction to a subspace
L0 of Tx0 .M / is an isomorphism. This would be a complement to the subspace L that we

expect to become the kernel of Tx0 .p/ (see (*) below). If N then is a subpremanifold of
M in a neighborhood of x0 with Tx0 .N / D L0 , then N is locally isomorphic to M=R by
the theorem of inverse functions 5.16.
Step 1: p is open. For U M open, we have
p 1 .p.U // D f x 2 M I 9 y 2 U W .x; y/ 2 R g D pr1 ..M U / \ R/:
Since the submersion pr1 W R ! M is open (Corollary 5.24), p 1 .p.U // M is open so

that p.U / X=R is open by definition of the quotient topology.
Step 2: Locality on M . Next we show that we can define the premanifold structure
on M=R locally on M . Assume that there exists an open covering .Ui /i 2I of M and
the structure of a premanifold on p.Ui / such that the projection p jUi W Ui ! p.Ui / is
a submersion for i 2 I . Then p.Ui \ Uj / is an open subset of p.Ui / and of p.Uj / for
i; j 2 I so that there are two structures of a premanifold on p.Ui \ Uj / making it into
an open submanifold of p.Ui / or p.Uj / respectively. By assumption we find the map
p jUi \Uj W Ui \ Uj ! p.Ui \ Uj / to be a submersion for both structures on p.Ui \ Uj / so
that both premanifolds structures on p.Ui \ Uj / agree by Proposition 5.51. Hence, there
exists a unique structure on M=R compatible with the given structure on p.Ui / for i 2 I
because .p.Ui //i 2I is an open covering of M=R.
After these preparations the main point of the proof is the following.
Main construction. For all x0 2 M there exists an open neighborhood U M of x0 ,
a submanifold N of U containing x0 and a morphism rW U ! N of premanifolds such
that for all x 2 U , r.x/ is the unique point of N equivalent to x.
Step 3: “Main construction ) (i)”. Let us first show that the construction yields
a premanifold structure on p.U / such that pW U ! p.U / is a submersion (then we have
shown (i) by Step 2). As the inclusion of N into U is a right inverse of rW U ! N , we
find that r is a submersion. The continuous map pW N ! p.U / is bijective and open,
hence a homeomorphism. Thus, transporting the premanifold structure of N to p.U / via
pW N ! p.U / makes pW U ! p.U / a submersion because we have p jU D p jN ı r.
Step 4: Construction of U , N , and r. Define
L WD f 2 Tx0 .M / I . ; 0/ 2 Tx0 ;x0 .R/ g (*)
and choose a submanifold NQ of M such that x0 2 NQ and L0 WD Tx0 .NQ / is a complemen-

tary vector space of L in Tx0 .M /. Because pr1 W R ! M is a submersion, we find NQ M
and R to be transversal in M M and therefore
˙ WD .NQ M / \ R
is a submanifold of M M by Theorem 5.47.

Next we claim that pr2 W ˙ ! M is a local isomorphism at .x0 ; x0 / 2 ˙ . It suffices
to show that T.x0 ;x0 / .pr2 /W T.x0 ;x0 / .˙ / ! Tx0 .M / is an isomorphism (Theorem 5.16). For
5.6 Quotients of Manifolds 117
the surjectivity let 2 Tx0 .M /. There exists 2 Tx0 .M / such that . ; / 2 Tx0 ;x0 .R/
0
because pr2 W R ! M is a submersion. Choose 1 2 L, 2 2 L with 1 C 2 D so that
. 2 ; / D . ; / . 1 ; 0/ 2 Tx0 ;x0 .R/ and . 2 ; / 2 Tx0 .NQ / Tx0 .M /:
Thus . 2 ; / 2 Tx0 ;x0 .˙ / so that T.x0 ;x0 / .pr2 / is surjective. For . ; / 2 ker.T.x0 ;x0 / .pr2 //
we have D 0 and . ; 0/ 2 Tx0 ;x0 .˙ /. Hence, we must have . ; 0/ 2 Tx0 ;x0 .R/ and
2 Tx0 .NQ / and therefore 2 L \ L0 D 0 so that T.x0 ;x0 / .pr2 / is injective.
By definition there exist open neighborhoods W; V M of x0 such that pr2 W ˙ \.W
W / ! V is an isomorphism of premanifolds. In particular, its inverse has to be of the
form
V ! ˙ \ .W W /; x 7! .r.x/; x/
for a morphism rW V ! NQ \ W and we must have V W . Moreover, for x 2 NQ \ V

we have .x; x/ 2 ˙ \ .W W / so that r.x/ D x because pr2 W ˙ \ .W W / ! V is
bijective.
Set U WD V \ r 1 .NQ \ V / and N WD NQ \ U . Then NQ \ V U because r j NQ \V D id.
Hence r.U / U and we obtain a morphism rW U ! N . We have x0 2 U and hence
x0 D r.x0 / 2 N . For x 2 U we have .r.x/; x/ 2 R so that r.x/ is equivalent to x.
Conversely, suppose that x 2 U and y 2 N are equivalent. We have y; x 2 V W ,
.y; x/ 2 R and y 2 NQ so that .y; x/ 2 ˙ \ .W W /. Because pr2 W ˙ \ .W W / ! V
is bijective, we must have .y; x/ D .r.x/; x/.
Step 5: “M=R Hausdorff , R M M closed”. By Step 1, p p is open
and surjective. Hence the diagonal M=R is closed in M=R M=R if and only if .p
p/1 .M=R / D R is closed in M M .
Proposition 5.54. Let F W M ! N be a morphism of premanifolds. Let R and S be

regular equivalence relations on M and N respectively. Suppose that F maps equivalence
classes with respect to R to equivalence classes with respect to S. Then the induced map
FN W M=R ! N=S is a morphism of premanifolds.
Proof. Let M W M ! M=R and N W N ! N=S the projections. Then FN ıM D N ıF .

Hence FN ı M is a morphism. As M is a surjective submersion, this implies that FN is
a morphism (Proposition 5.25).
5.7 Problems
Problem 5.1. Let M be a real C ˛ -premanifold with ˛ 1 or a complex manifold, let

p 2 M , and f 2 OM;p .
1. Choose a representative .U; fQ/ of the germ f and a chart .U; ˚/ at p and define the
vanishing order of f at p as
˚
ordp .f / WD inf r 2 N0 I D r .f ı ˚ 1 /.˚.p// ¤ 0 :
Show that ordp .f / does not depend on the choice of U , fQ, or ˚ and that
n o
ordp .f / D sup r 2 N0 I f 2 mpr :
2. Let .V; jj jj/ be a normed K-vector space of finite dimension and suppose that M V
is open. Show that

r
ordp .f / D sup r 2 R0 I lim jjx pjj jf .x/j D 0
M 3x!p
and that ordp .f / is the supremum of all r 2 N0 such that the Taylor expansion up to
degree r of f at p is zero.
3. Let M be a real analytic or a complex premanifold. Show that ordp .f / D 1 if and
T
only if f D 0 and hence r0 mpr D 0.
Problem 5.2. Let p 2 Rn . Show that attaching to f 2 C1 Rn ;p its Taylor series in p (in
variables t1 ; : : : ; tn ) induces an isomorphism of R-algebras
, !
\
C1
Rn ;p mpr ! RŒŒt1 ; : : : ; tn ;
r0
where the right-hand side denotes the R-algebra of formal power series in t1 ; : : : ; tn with
values in R.
Hint: The injectivity is Problem 5.1.
Problem 5.3. Let M be a real C ˛ -premanifold with 1 ˛ < 1, let p 2 M and assume
that dimp .M / > 0. We wish to show that mp =mp2 is an infinite-dimensional R-vector
space.
1. Show that it suffices to consider the case M D R and p D 0, which we assume from
now on.
5.7 Problems 119
2. For f 2 m WD m0 define

o.f / WD ord0 .f / WD sup ˇ 0 I lim jxjˇ f .x/ D 0 :
x!0
Show that ˚o.f / > ˛ C 1 or o.f / 2 N.

3. Show that jxjˇ I ˛ < ˇ < ˛ C 1 is linearly independent in m=m2 .
Problem 5.4. Let M and N be real C ˛ -premanifolds with ˛ 1 or let M and N be

complex premanifolds. Let f W M ! N be a morphism of premanifolds. Show that f is
a local isomorphism if and only if fx W OY;f .x/ ! OX;x is an isomorphism for all x 2 X.
Problem 5.5. For every K-vector space V define the K-algebra KŒV as follows. As a
K-vector space we set KŒV WD K ˚ V . Define a multiplication on KŒV by
. ; v/ . 0 ; v 0 / WD . 0
; v0 C 0
v/:
1. Show that these definitions make KŒV into a commutative local K-algebra whose
maximal ideal is V . Show that V 2 D 0 and that KŒK Š KŒT =.T 2 / DW KŒ".
2. Show that for every K-linear map uW V ! W of K-vector spaces the map KŒV !
KŒW , . ; v/ 7! . ; u.v//, is a homomorphism of K-algebras and that one obtains
a functor V 7! KŒV from the category of K-vector spaces to the category of K-
algebras.
3. Let P ŒV be the locally K-ringed space consisting of a singe point * and with
OP ŒV .P ŒV / D KŒV . Let M be a real C ˛ -premanifold with ˛ 1 or a com-
plex manifold. Show that the identification (5.7) yields for all r 2 N an identification
( )
tW P ŒV ! M morphism of locally
HomK .V; Tp .M // D : (*)
K-ringed spaces with t. / D p
Show that addition K K ! K and for 2 K scalar multiplication K ! K, ˛ 7! ˛

induce via (*) the K-vector space structure on Tp .M / defined in Definition 5.1.
Problem 5.6. Let X be a topological space and let W X ! N0 be a map such that
f x 2 X I .x/ r g is open in X for all r 2 N0 and such that there exists an open
covering .Ui /i such that jUi is bounded for all i. Show that
˚
x 2 X I 9 neighborhood W of x such that jW is constant
is open and dense in X.

Deduce that if F W M ! N is a morphism of premanifolds, then the set of points of M
where F is of locally constant rank is open and dense in M .
Problem 5.7. Show that the underlying real analytic manifold of P 1 .C/ is isomorphic to
the 2-sphere S 2 .
˚
Problem 5.8. Let X WD .x; jxj/ 2 R2 I x 2 R . Is X a C 1 -submanifold of R2 ?
Problem 5.9. Let M be a compact premanifold of constant dimension n. Show that M

cannot be embedded into Kn .
Problem 5.10. Let n 2 N and consider detW Mn .K/ ! K. Show that det is a morphism
of real analytic (for K D R) manifolds respectively of complex manifolds (for K D C).
Determine rkp .det/ for all p 2 Mn .K/.
Problem 5.11. Show that F W .0; 2/ ! R2 , t 7! .sin3 t; sin t cos t/ is an injective im-
mersion of real manifolds. Define a map
8
ˆ
<t C ; if t < I
ˆ
GW .0; 2/ ! .0; 2/; t 7! 0; if t D I
ˆ
:̂
t ; if t > :
Show that F ı G is a morphism of real manifolds but that G is not continuous.
Problem 5.12. Let F W M ! N be an embedding of premanifolds, let Z be a premani-

fold, and let GW Z ! M be a map. Show that G is a morphism of premanifolds if and
only if F ı G is a morphism of premanifolds.
˚
Problem 5.13. Show that S WD .x; y/ 2 R2 I xy D 0 is not a real C 1 -submanifold of
R2 .
Hint: What would be T0 .S/?
Problem 5.14. Let F be an injective (respectively bijective) morphism of premanifolds

that is locally of constant rank. Show that F is an immersion (respectively an isomor-
phism).
Problem 5.15. Let V and W be finite-dimensional K-vector spaces, n WD dimK .V /

and m WD dimK .W / and let r be an integer with 0 r minfm; ng. Show that
f f 2 HomK .V; W / I rk.f / D r g is a analytic submanifold of HomK .V; W / of dimen-
sion r.m C n r/.
Problem 5.16. Let M be a premanifold. A pair .N; i/, where N is a premanifold and
iW N ! M is an injective immersion is called immersed subpremanifold of M .
1. Show that every submanifold together with the inclusion is an immersed subpremani-
fold but that the converse does not hold in general.
2. Let .N; i/ be an immersed subpremanifold and suppose that N is compact and M is
Hausdorff. Show that i.N / is a closed subpremanifold of M .
5.7 Problems 121
Problem 5.17. Consider the real analytic map W R ! R2 , t 7! .sin3 t; sin t cos t/ and
let N1 WD .; / and N2 WD .0; 2/ considered as open submanifolds of R. Show that
.N1 ; jN1 / and .N2 ; jN2 / are immersed subpremanifolds of M , that .N1 / D .N2 / but
that there exists no isomorphism of manifolds 'W N1 ! N2 such that jN1 D jN2 ı '.
Problem 5.18. Let M be a premanifold.
1. Show that for an immersed premanifold .N; i/ (Problem 5.16) the following assertions
are equivalent:
(i) For each premanifold Z a map Z ! N is a morphism of premanifolds if and only
if i ı f W Z ! M is a morphism of premanifolds.
(ii) For each premanifold Z a map Z ! N is continuous if and only if i ıf W Z ! M
is a morphism of premanifolds.
In this case .N; i/ is called initial subpremanifold of M .
2. Show that every subpremanifold together with the inclusion is an initial subpremani-
fold.
Hint: Problem 5.12.
3. Let .N1 ; i1 / and .N2 ; i2 / be initial subpremanifolds of M with i1 .N1 / D i2 .N2 /. Show
that there exists a unique isomorphism F W N1 ! N2 such that i1 D i2 ı F (compare to
Problem 5.17).
Problem 5.19. Let r 2 R n Q and R ! R=Z, x 7! xN the canonical map. Show that
iW R ! R=Z R=Z, x 7! .x; N rx/ is an injective immersion with dense image. Deduce
that i is not an embedding of manifolds.
Problem 5.20. Show that the fiber product constructed in Theorem 5.47 is a fiber product
in the category of all K-ringed spaces.
Problem 5.21. Let G be a finite group of automorphisms of a premanifold M and let

M G WD f p 2 M I g.p/ D p for all g 2 Gg.
1. Show that for all p 2 M G there exists a system of local coordinate at p with respect
to which G acts linearly.
2. Show that M G is a submanifold of M and that Tp .M G / D Tp .M /G for p 2 M G .
Problem 5.22. Let F W M ! N be a submersion of premanifolds. Show that the equiv-

alence relation p q W, F .p/ D F .q/ is regular and that M= Š F .M / as open
premanifolds of N .
Lie Groups
6
A Lie group is a premanifold together with a group structure such that multiplication
and inversion are morphisms of premanifolds. They play an important role as symme-
try groups of premanifolds and this is also the main focus of this chapter. Hence after
defining Lie groups and Lie subgroups and studying their topology we will consider Lie
groups acting on premanifolds in Sect. 6.2. The main result is the existence of quotients
of premanifolds by a proper and free Lie group action in the last section.
6.1 Definition and Examples of Lie Groups
Definition 6.1. A real Lie group (respectively a complex Lie group) is a real analytic
premanifold (respectively a complex premanifold) G endowed with a group structure such
that
mG W G G ! G; .g; h/ 7! gh;
iG W G ! G; g 7! g 1
are real analytic maps (respectively holomorphic maps). As usual, we speak simply of a
Lie group if we consider the real and the complex case simultaneously.
A homomorphism of Lie groups is a morphism of premanifolds that is a group homo-
morphism.
We obtain the categories of real and of complex Lie groups.

In the real case we could also have defined Lie groups and their homomorphisms via
C ˛ -premanifolds with some fixed ˛ 2 N0 [ f1g. But one would obtain the same notion:
One can show that weakening the structure from analytic to C ˛ (Remark 4.23) yields an
equivalence from the category of real analytic Lie groups to the category of real C ˛ -Lie
groups ([BouLie1] III, §8.1).
DOI 10.1007/978-3-658-10633-1_6
124 6 Lie Groups
Remark and Definition 6.2. The group law on a complex Lie group G defines on the
underlying real analytic premanifold GR the structure of a real Lie group. We obtain
a functor G 7! GR from the category of complex Lie groups to the category of real Lie
groups.
One can show that this functor has a left adjoint functor H 7! HC , and HC is called
the complexification of the real Lie group H ([HiNe] 15.1.4).
The following result in particular shows that the analyticity of the multiplication al-
ready implies that the inversion is analytic.
Proposition 6.3. Let G be a real analytic premanifold or a complex premanifold endowed

with a group structure such that mG W G G ! G, .g; h/ 7! gh is a morphism.
1. Then G is a Lie group.

2. For g 2 G the maps lg W G ! G, h 7! gh and rg W G ! G, h 7! hg are isomorphisms
of premanifolds.
3. The map iG W G ! G, g 7! g 1 , is an isomorphism of premanifolds.
Proof. 2. Both maps lg and rg are bijective morphisms and their inverse, namely lg1 and
rg1 respectively, is again a morphism.
1. Let W G G ! G G, .g; h/ 7! .g; gh/. For all .g; h/ 2 G G, T.g;h/ . / has
the form !
idTg G 0
1 2
;
T.g;h/ .mG / T.g;h/ .mG /
where T i mG denotes the i-th partial derivative of mG . As T.g;h/

2
.mG / D Th .lg / is an
isomorphism by 2, T.g;h/ . / is an isomorphism. Therefore is a local isomorphism
(Theorem 5.16) and hence an isomorphism because is bijective. For g 2 G we have
1 .g; e/ D .g; g 1 /. Hence iG is a morphism.
3. The morphism iG has as inverse again a morphism of premanifolds, namely itself.

Example 6.4. Let n 2 N0 .
1. Let V be a finite-dimensional R- (respectively C-) vector space. Then .V; C/ is a real

(respectively complex) Lie group.
2. GLn .R/ is real Lie group, GLn .C/ is a complex Lie group: multiplication of matrices
is polynomial and in particular analytic, we conclude by Proposition 6.3 (it also follows
directly from Cramer’s rule, that inversion of matrices is also polynomial).
More generally, if V is a finite-dimensional R- (respectively C-) vector space, then
GL.V / is a real (respectively complex) Lie group.
6.1 Definition and Examples of Lie Groups 125
A Lie group has the following topological properties.
Proposition 6.5. Let G be a Lie group.
1. G is Hausdorff and paracompact. All connected components of G are open and closed
in G and have the same dimension.
2. The underlying premanifold of G is a manifold if and only if G has countably many
connected components.
Proof. Every premanifold is a T1 -space. In particular feg is closed in G, where e 2 G is

the neutral element. Therefore G is Hausdorff by Appendix Proposition 12.58.
2. If G has uncountably many connected components, G cannot be second countable.
Conversely, suppose that G has countably many connected components. As countable
unions of second countable spaces are again second countable, it suffices to show that
a connected Lie group is second countable. This follows from Corollary 1.6.
1. As G is locally connected, its connected components are open and closed (Propo-
sition 2.13). If G 1 is a connected component, then left multiplication with any g 2 G 1
yields an isomorphism G 0 ! G 1 , where G 0 is the identity component of G. In particular
dim.G 1 / D dim.G 0 /. By 2, G 0 is a manifold and hence paracompact (Theorem 4.28).
Therefore G is the topological sum of its paracompact connected components and hence
paracompact.
Definition 6.6. Let G be a Lie group. A Lie subgroup of G is a subgroup of G that is

a subpremanifold.
Similarly, as there are more general notions of submanifolds, there are also more gen-
eral notions of Lie subgroups.
Proposition 6.7. Let H be a Lie subgroup of G.
1. H is a Lie group.
2. H is closed in G.
Proof. 1. By Proposition 6.3 it suffices to show that the multiplication H H ! H is

a morphism. This follows from Proposition 5.38 applied to the multiplication morphism
mG .
2. As a submanifold, H is locally closed in G. Every locally closed topological
subgroup is closed (Appendix Proposition 12.58).
It can be shown that if G is a real Lie group and H is any closed subgroup, then H
is a real Lie subgroup of G (e.g., [BouLie1] III, §8.2, Theorem 2). The analogous result
does not hold for complex Lie groups (for instance, R C is a closed subgroup of the
complex Lie group .C; C/ but it is not a complex Lie group).
126 6 Lie Groups
6.2 Lie Group Actions
We now consider the important case that a Lie group acts on a premanifold. Let us first
recall some notions of a group G acting on a set X. Then X is called a G-set and the
action is a map G X ! X, usually simply denoted by .g; x/ 7! gx, that satisfies
1x D x and g.hx/ D .gh/x for all g; h 2 G and x 2 X.
1. The action is called transitive (respectively simply transitive) if for all x; y 2 X there
exists a (respectively there exists a unique) g 2 G with gx D y.
2. For x 2 X the subgroup Gx WD StabG .x/ WD f g 2 G I gx D x g of G is called the
stabilizer of x in G. The subset Gx WD f gx I g 2 G g is called the G-orbit of x. The
map G 7! X, g 7! gx, induces a bijection G=Gx ! Gx of sets. The set of G-orbits
in X is denoted by X=G.
3. The action is called free if for all x 2 X its stabilizer Gx is trivial.
4. The action is called faithful if 1 2 G is the only element g 2 G such that x 7! gx is
T
the identity for all x 2 X (in other words x2X Gx D f1g).
If X and Y are G-sets, a map f W X ! Y is called G-equivariant of morphism of G-sets,

if f .gx/ D gf .x/ for all x 2 X and g 2 G. Then G-sets and G-equivariant maps
together with the usual composition form a category.
Definition 6.8. Let G be a real (respectively complex) Lie group. A C ˛ -G-premanifold

(respectively a complex G-premanifold) is a real C ˛ -premanifold (respectively a complex
premanifold) M together with a C ˛ -map (respectively a holomorphic map) G M ! M ,
which defines an action of the group G on the set M . In the real case we consider the
analytic premanifold G as a C ˛ -premanifold (Remark 4.23).
If we consider the cases of a real Lie group and a real C ˛ -premanifold (˛ 2 Nb fixed)
and the complex case simultaneously, we simply speak of a G-premanifold or of the Lie
group G acting on the premanifold M .
A morphism of G-premanifolds is a G-equivariant morphism of premanifolds.
We obtain the category of G-premanifolds.
Proposition 6.9. Let G be a Lie group, let M and N be G-premanifolds, and let F W M !
N be a morphism of G-premanifolds. For all p 2 M and g 2 G one has
dimp .M / D dimgp .M /; rkp .F / D rkgp .F /:
Moreover, F is a submersion (respectively immersion) at p if and only if F is a submersion

(respectively immersion) at gp.
6.2 Lie Group Actions 127
Proof. The map gM W M ! M , m 7! gm is an isomorphism of premanifolds with inverse

.g 1 /M . Set q WD F .p/ 2 N . Then F .gp/ D gq and all assertions follow from the
commutative diagram
Tp .F /
Tp .M / Tq .N /
Tp .gM / Tq .gN /
Tgp .F /
Tgp .M / Tgq .N /;
where the vertical arrows are isomorphisms.
Corollary 6.10. Notation as in Proposition 6.9. Assume that the action of G on M is

transitive. Then F is of constant rank and for all q 2 N the fiber F 1 .q/ is a subpreman-
ifold of M .
Proof. By Proposition 6.9 the map M 3 p 7! rkp .F / is constant. The second assertion
follows from Proposition 5.39.
Example 6.11. Let ˇW Kn Kn ! K be a bilinear form with matrix B 2 Mn .K/ with

respect to the standard basis of Kn . Let
G WD f g 2 GLn .K/ I ˇ.gv; gw/ D ˇ.v; w/ 8 v; w 2 Kn g

˚
D g 2 GLn .K/ I t gBg D B :

Ip 0
This is a closed subgroup of GLn .K/. For instance G D O.p; q/ for B D 0 Iq and
K D R, or B D Sp2m .K/ for n D 2m and B D 0 Im
Im 0 .
Then G D F 1 .B/, where
F W GLn .K/ ! Mn .K/; F .A/ D t ABA:
Let GLn .K/ act transitively on GLn .K/ from the left via .g; A/ 7! Ag 1 and let it act
on Mn .K/ by .g; X/ 7! t g 1 Xg 1 for g 2 GLn .K/, X 2 Mn .K/. Then F is GLn .K/-
equivariant:
F .g A/ D F .Ag 1 / D t g 1t ABAg 1 D g F .A/:
Hence G is a submanifold of GLn .K/ by Corollary 6.10 and hence a Lie subgroup of
GLn .K/.
Corollary 6.12. Let G be a Lie group and let M be a G-premanifold. Then for all p 2 M
the stabilizer Gp of p is a Lie subgroup of G.
Proof. The morphism aW G ! M , g 7! gp, is G-equivariant if we let G act on itself via

left multiplication. Hence Gp D a1 .p/ is a Lie subgroup by Corollary 6.10.
128 6 Lie Groups
6.3 Quotients by Lie Group Actions
Our next goal is a criterion when the quotient of a premanifold by a Lie group action is
again a premanifold. For this we first define the notion of a proper action.
Proper Actions
Definition 6.13. Let G be a topological group, X a topological space and suppose that G
acts on X via a continuous map G X ! X. Then we say that G acts properly on X or
that the action of G on X is proper if the map
W G X ! X X; .g; x/ 7! .x; gx/
is a proper map (Definition 1.30).
Note that the condition does not say that the action map G X ! X is proper.
Remark 6.14. The topological group G acts properly on X if and only if is closed and
for all x 2 X the stabilizer Gx WD f g 2 G I gx D x g is a compact subgroup of G.
Indeed, by definition of properness, G acts properly on X if and only if is closed
and for all .x; y/ 2 X X the fiber 1 .x; y/ D Gx;y fxg is compact, where Gx;y D
f g 2 G I gx D y g. But if Gx;y ¤ ; we can choose h 2 Gx;y and Gx ! Gx;y , g 7! hg,
is a homeomorphism with inverse k 7! h1 k.
We refer to Problem 6.2 for other characterizations of proper action.
Example 6.15. Let G be a topological group and H G be a closed subgroup. Suppose

that G acts properly on a topological space X. Then Proposition 1.32 shows that H acts
properly on X because the closed embedding H X ,! G X is proper (Example 1.31)
and separated.
Consider in particular the case that G acts on itself by left multiplication. Then G
G ! G G, .g; g 0 / 7! .g 0 ; gg 0 / is a homeomorphism and in particular proper. Hence
the action of H on G by left multiplication is proper.
Actions of compact groups are always proper, more precisely:
Proposition 6.16. Let G be a compact Hausdorff topological group acting on a Hausdorff

space X. Then this action is proper and the canonical map W X ! X=G is proper (X=G
endowed with the quotient topology of X).
Proof. As G is compact, the projection pr2 W G X ! X is proper (Proposition 1.29). The

map aW G X ! X, .g; x/ 7! gx is the composition of the homeomorphism G X !
6.3 Quotients by Lie Group Actions 129
G X, .g; x/ 7! .g; gx/ (an inverse is given by .g; x/ 7! .g; g 1 x/) followed by pr2 .
Hence a is proper. The map W .g; x/ 7! .x; gx/ is the composition of the diagonal
W G X ! .G X/ .G X/ followed by the product of pr2 and a. The diagonal is
a closed embedding and in particular proper because G and X are Hausdorff. Hence we
see that is proper because products and compositions of proper maps are again proper
(Proposition 1.34 and Proposition 1.32).
It remains to show that is proper. Let A X be closed, then 1 ..A// D GA
is the image of G A under the action map a. As we have seen that a is proper, GA is
closed in X and hence .A/ is closed in X=G by the definition of the quotient topology.
Hence we see that is closed. The fibers of are the G-orbits in X, which are compact
by Lemma 6.17. Hence is proper (Theorem 1.30).
Lemma 6.17. Let G be a compact topological group acting on a Hausdorff topological

space. Then for all x 2 X the G-orbit Gx is compact and closed in X. The map aW G !
X, g 7! gx induces a homeomorphism G=Gx ! Gx.
Proof. The stabilizer Gx D a1 .x/ is closed in G. Hence the quotient G=Gx is Hausdorff
(Appendix Proposition 12.58). As image of the compact space G it is also compact.
Therefore the bijective continuous map G=Gx ! Gx is a homeomorphism (Appendix
Corollary 12.53). Hence Gx is compact and thus closed in X.
Remark 6.18. Let G be a topological group acting continuously on a topological space

X. Then the action of G on X is proper (respectively free) if and only if the map W G
X ! X X, .g; x/ 7! .x; gx/ is proper (respectively injective). Therefore the action is
proper and free if and only if is closed topological embedding.
Existence of Quotients by Lie Group Actions

We can now formulate the main result of this section.
Theorem 6.19. Let G be a Lie group and let M be a G-premanifold. Suppose that the
action is free and proper. Then there is a unique structure of a premanifold on the set of
orbits M=G such that the projection W M ! M=G is a submersion. Its topology is the
quotient topology of M . It is Hausdorff.
Proof. The uniqueness follows from Proposition 5.51. If such a premanifold structure
exists on M=G, then is surjective and open (Corollary 5.24) and hence the topology
of M=G is the quotient topology. It remains to show the existence of the premanifold
structure. We use Theorem 5.53. Hence we have to show:
1. R WD f .p; gp/ I p 2 M; g 2 G g is a closed submanifold of M M .

2. pr1 W R ! M is a submersion.
130 6 Lie Groups
Now R is the image of W G M ! M M , .g; p/ 7! .p; gp/ and is a closed

topological embedding by Remark 6.18. Hence for (1) it suffices to show that is an
immersion. For g 2 G and p 2 M the derivative T.g;p/ ./ is of the form
!
0 idTp .M /
;
Tg .ap /
where ap W G ! M , g 7! gp. As the action is free, ap is injective and by equivariance, ap

is constant rank. Hence ap is an immersion and therefore T.g;p/ ./ is injective.
Let us show (2). As yields an isomorphism G M ! R, we have a commutative
diagram
GM
.g;p/7!gp

pr2
R M (6.1)
pr2
pr1
M
M=R;
where the square is a fiber product diagram and where is an isomorphism. As pr2 W G
M ! M is a submersion, pr1 W R ! M is a submersion.
The proof shows that to get a premanifold structure on M=G such that M ! M=G
is a submersion it suffices to assume that W G M ! M M , .g; p/ 7! .p; gp/ is an
embedding (instead of the properness of the action). In this case the premanifold M=G is
Hausdorff if and only if is a closed embedding.
Example 6.20. Let V be a finite-dimensional K-vector space. The action of K on V

by scalar multiplication is not proper as the stabilizer of 0 2 V is not compact. But its
restriction to V n f0g is a proper and free action. Its quotient is the projective space P .V /
of lines in V .
Remark 6.21. We keep the notation and hypotheses of Theorem 6.19. Let p 2 M and
ap W G ! M , g 7! gp. Then the sequence
Te .ap / Tp ./
0 ! Te .G/ ! Tp M ! T.p/ .M=G/ ! 0 (6.2)
is exact. In particular one has dim.p/ .M=G/ D dimp .M / dim G. If dim.G/ D 0, then
is a local isomorphism.
Indeed, in the proof of Theorem 6.19 we have seen that ap is an immersion, hence
Te .ap / is injective. As is a submersion, Tp ./ is surjective. To show exactness in the
middle, we apply Remark 5.49 to (6.1) and obtain
ker.Tp .// Š ker.T.p;ep/ .pr1 // Š ker.T.e;p/ .pr2 W G M ! M // Š Te .G/:

Figure 6.1 Klein bottle
Example 6.22. Let M D S 1 S 1 and consider the action by G D Z=2Z for which the
non-trivial element acts by .z; w/ 7! .1=z; w/. This is a free action and it is proper by
Proposition 6.16. The quotient manifold M=G is a 2-dimensional real analytic manifold
called the Klein bottle (see Fig. 6.1, note that there is a self intersection as the Klein bottle
cannot be embedded smoothly into R3 ).
Corollary 6.23. Let G be a Lie group and let H be a Lie subgroup.
1. There exists on G=H a unique structure of an analytic Hausdorff premanifold such

that the projection W G ! G=H is a submersion.
2. If G has only countably many connected components, then G=H is an analytic mani-
fold.
3. The dimension of G=H is constant and dim.G=H / D dim.G/ dim.H /.
4. If H is normal in G, then the induced group law on G=H makes G=H into a Lie group.
Proof. The Lie group H acts properly (Example 6.15) and freely on G. Hence 1 follows
from Theorem 6.19. Remark 6.21 implies 3, and 4 follows from Proposition 5.54 applied
to the multiplication morphism of G.
Finally, if G has countably many connected components, then G is second countable
(Proposition 6.5). Therefore G=H is second countable because G ! G=H is surjective
and open (Remark 1.3 2). Hence G=H is a manifold.
Corollary 6.24. Let 'W G ! G 0 be a morphism of Lie groups and let H 0 G 0 be a Lie
subgroup. Then ' 1 .H 0 / is a Lie subgroup of G. In particular Ker.'/ is a Lie subgroup
of G.
132 6 Lie Groups
Proof. Let F be the composition of ' followed by the projection G 0 ! G 0 =H 0 and let
eN 0 2 G 0 =H 0 be the image in the neutral element of G 0 . Let g 2 G act on G by left
multiplication and on G 0 =H 0 by left multiplication with '.g/. Then F is G-equivariant
and the action of G on G is transitive. Hence we can apply Corollary 6.10 to see that the
subgroup F 1 .e 0 / D ' 1 .H 0 / is a subpremanifold.
Example 6.25. Let n 1. Then SLn .K/ D Ker.detW GLn .K/ ! K / is a Lie subgroup
of GLn .K/.
Example 6.26 (Flag manifold). Let E D Kn and let n D .n1 ; : : : ; nr / be a tuple of

P
integers ni 2 N with i ni D n. For j D 0; : : : ; r set mj WD n1 C C nj and let
Flagn .E/ be the set of flags of type n of E, i.e., flags of subspaces
0 D W0 W1 Wr D E
with dim Wi D mi . Then GLn .K/ acts on Flagn .E/ via .g; .Wi /i / 7! .g.Wi //i . By
basis extension this action is transitive. Let Ei be the subspace generated by the standard
vectors e1 ; : : : ; emi . Then the stabilizer Ln of the flag .Ei /i in GLn .K/ is the Lie subgroup
of matrices of the form 0 1
A1
B : C
B : C
B 0 A2 : : :: C
B : C
B : :: :: C
@ : : : A
0 0 Ar
with Ai 2 GLni .K/. Hence we may apply Corollary 6.23 to endow Flagn .E/ D
GLn .K/=Ln with the structure of an analytic manifold.
Let 0 d n, then Grassd .E/ WD Flag.d;nd / .E/ is the Grassmann manifold of
d -dimensional subspaces of E.
In Example 4.26 we endowed projective space P n .K/ with the structure of an analytic
manifold. As a set we have P n .K/ D Grass1 .KnC1 / and it is not difficult to check directly
the manifold structure defined in Example 4.26 and in Example 6.26 are the same. This
can also be deduced from the following result.
Proposition 6.27. Let G be a Lie group, let M be a G-premanifold of type C ˛ with ˛ 2

b Suppose that the G-action on M is transitive and choose p 2 M . Let H WD StabG .p/
N.
be its stabilizer. Suppose that G has only countably many connected components. Then
aW G ! M , g 7! gp induces an isomorphism of manifolds
N G=H ! M;
aW gH 7! gp:
Proof. The group H is a Lie subgroup by Corollary 6.12. Clearly, aN is bijective. It is

a morphism of premanifolds by Proposition 5.54. It suffices to show that aN is a submer-
sion (looking at charts [Corollary 5.23] we see that every injective submersion is a local
isomorphism and hence every bijective submersion is an isomorphism).
As G acts transitively on G=H and on M , there exist m; k; r 2 N0 such that
dimp .M / D m, dimq .G=H / D k and rkq .a/ N D r for all p 2 M , q 2 G=H .
Assume that r < m (in particular m > 0). As G=H is a second countable (Corol-
lary 6.23 2), G=H is Lindelöf (Proposition 1.4). Hence there exist countably many charts
S
.Un ; xn ; Vn ; yn ; aQ n /, n 2 N, of aN such that n Un D G=H and such that aQ n is given
by .z1 ; : : : ; zk / 7! .z1 ; : : : ; zr ; 0; : : : ; 0/. Hence Im.aQ n / is contained in a zero set (Re-
mark 5.43) of Km and hence Im.a/ N has measure zero. But this is a contradiction to the
surjectivity of a. N
The hypothesis that G has only countably many connected components is necessary.
Let G D R be endowed with the discrete topology making G into a 0-dimensional real Lie
group. Let G act on M D R (with the usual structure of a 1-dimensional real manifold)
via .g; p/ 7! g C p. Then G=Gp Š G is discrete for p 2 M but M is not.
If the underlying topological space of G is second countable, then G has only countably
many connected components. This is for instance the case if G is a closed subgroup of
GLn .K/.
Example 6.28. Consider H WD f z 2 C I Im.z/ > 0 g as an open real analytic submani-

fold of C D R2 . Let the real Lie group G WD SL2 .R/ act on H by Möbius transformation
! !
a b az C b
; z 7! :
c d cz C d
This makes H into a SL2 .R/-manifold. The action is transitive and

( ! )
a b ai C b
StabG .i/ D 2 SL2 .R/ I Di
c d ci C d
( ! )
a b
D 2 SL2 .R/ I a D d; b D c
c d
D SO.2/:
Hence Proposition 6.27 shows that H D SL2 .R/= SO.2/.
To see how many connected components a Lie group precisely has, the following re-
mark is often helpful.
134 6 Lie Groups
Remark 6.29. Let G be a Lie group. Its identity component G 0 is an open and closed
normal subgroup, in particular it is a Lie subgroup with dim.G 0 / D dim.G/. Hence if we
consider the abstract group 0 .G/ as a 0-dimensional Lie group, then the canonical map
$G W G ! G=G 0 D 0 .G/ is a surjective submersion.
Any homomorphism of Lie groups 'W H ! G induces a unique group homomorphism
0 .'/W 0 .H / ! 0 .G/ making the diagram of Lie group homomorphisms
'
H G
$H $G
0 .'/
0 .H / 0 .G/
commutative. We obtain a functor 0 from the category of Lie groups to the category of
groups (identified with the category of 0-dimensional Lie groups).
Let H be a Lie subgroup of G and let be the image of 0 .H / ! 0 .G/. As
G ! G=H is open and surjective, it induces a bijective map
0 .G/= ! 0 .G=H /: (6.3)
In particular we see that G is connected if G=H and H are connected.
Example 6.30. The action of the real Lie group SO.n/ on S n1 by matrix multiplication
is analytic. The stabilizer of .1; 0; : : : ; 0/ 2 S n1 is SO.n 1/ and we deduce S n1 Š
SO.n/= SO.n 1/. Arguing by induction on n starting with SO.1/ D 1 we deduce via
Remark 6.29 that SO.n/ is connected for all n. As SO.n/ is of index 2 in O.n/ we also
see that O.n/ has two connected components and that O.n/0 D SO.n/.
Let B GLn .R/ be the subgroup of upper triangular matrices. As an analytic man-
ifold, B is isomorphic to Rn.n1/=2 . In particular, B is connected. By Gram-Schmidt
orthogonalization, multiplication of matrices yields an isomorphism B O.n/ ! GLn .R/
of real analytic manifolds. Hence we deduce that GLn .R/ has two connected components.
Its identity component is the subgroup of matrices with positive determinant.
6.4 Problems
Problem 6.1. Let G be a topological group that acts properly on a topological space X.
Let Y be a G-invariant subspace. Show that the action of G on Y is proper.
Problem 6.2. Let G be a topological group acting continuously on a topological space X.
1. Suppose that G acts freely on X and let R WD f .gx; x/ I g 2 G; x 2 X g X X.

Show that $ W R ! G, .gx; x/ 7! g is well defined. Show that G acts properly on X
if and only if R is closed in X X and $ is continuous.
6.4 Problems 135
2. Suppose that G and X are Hausdorff and that G is locally compact. Show that G
acts properly on X if and only if for all x; y 2 X there exist neighborhoods U of x
and V of y such that f g 2 G I gU \ V ¤ ; g is relatively compact in G (Appendix
Problem 12.31).
Problem 6.3. Let G be a finite group and let M be a G-manifold such that the action of
G on M is free. Show that there exists on M=G a unique structure of a premanifold such
that W M ! M=G is a submersion. Show that is a local isomorphism.
Problem 6.4. Let G and H be Lie groups acting properly and freely on premanifolds M
and N , respectively. Let 'W G ! H be a homomorphism of Lie groups and F W M ! N
a morphism of premanifolds such that F .gm/ D '.g/F .m/. Show that there exists
a unique morphism FN W M=G ! N=H such that the following diagram commutes:
F
M N
FN
M=G N=H :
Problem 6.5. Let M D .1; 1/ S 1 (considered as a real analytic manifold) and let
G D Z=2Z act on M such that the non-trivial element acts by .t; z/ 7! .t; z/. Show
that this action is free and proper and that M=G is isomorphic to the Möbius band.
Problem 6.6. Show that the quotient manifold R=Z and the submanifold S 1 of R2 are
isomorphic real analytic manifolds.
Problem 6.7. Let 'W G ! G 0 be a homomorphism of Lie groups. Show that the following
(i) ' is a submersion (respectively a local isomorphism).

(ii) T1 .'/W T1 .G/ ! T1 .G 0 / is surjective (respectively bijective).
(iii) ' jG 0 W G 0 ! G 00 is surjective (respectively a covering map), where H 0 denotes the
connected component of 1 in a topological group H .
Problem 6.8. Let G be a connected Lie group. Show that there exists a universal cover
space W GQ ! G and that for all eQ 2 1 .e/ there exists a unique Lie group structure
on GQ such that eQ is the neutral element of GQ and such that is a local isomorphism of
premanifolds and a group homomorphism. Show that .G; Q / is unique up to a unique
isomorphism of Lie groups.
Hint: Problem 4.14.
136 6 Lie Groups
Problem 6.9. Let G be a Hausdorff locally compact second countable topologically

group that acts continuously and transitively on a Hausdorff Baire space X (Problem 1.9).
Let x 2 X and Gx D f g 2 G I gx D x g. Show that the map G X, g 7! gx, is open
and that the induced map G=Gx ! X is a homeomorphism.
Hint: Problem 1.11.
Problem 6.10. Let G be a Lie group with countable many connected components, let M
be a G-premanifold, and let p 2 M . Let Gp M be the orbit of p.
1. Show that the following assertions are equivalent:

(i) Gp M is locally closed.
(ii) Gp is a submanifold of M .
Hint: If Gp M is locally closed, use Problem 6.9 and Problem 1.10 to see that
G ! Gp, g 7! gp, is open.
2. Suppose that Gp is locally closed in M . Show that G ! M , g 7! gp, induces a
G-equivariant isomorphism of premanifolds G= StabG .p/ ! Gp.
3. Suppose that the action of G on M is proper. Show that 1 and 2 hold for each orbit.
4. Let r 2 R. Show that R S 1 S 1 , .x; .z1 ; z2 // 7! .e 2ix z1 ; e 2i rx z2 /, defines an
analytic action of .R; C/ on S 1 S 1 . Show that the equivalent assertions of 1 hold if
and only if r 2 Q.
Hint: Problem 5.19
Problem 6.11. For n 2 show that there are isomorphisms of real analytic manifolds
S 2nC1 Š U.n/=U.n1/ Š SU.n/= SU.n1/. Deduce that U.n/ and SU.n/ are connected
for all n 1. Show that GLn .C/ is connected.
Problem 6.12. Let E and F be finite-dimensional K-vector spaces with d WD dim.F /

dim.E/. Let I.F; E/ be the set of injective K-linear maps F ! E and define on I.E; F /
an equivalence relation by u u0 W, 9 v 2 GL.F /W u0 D u ı v. Show that I.F; E/
is an open submanifold of HomK .F; E/, that is a regular equivalence relation and that
I.F; E/= Š Grassd .F /.
Problem 6.13. Use the notations of Example 6.26.
1. Show that the Plücker embedding
Grassd .E/ ! P .d .E// WD Grass1 .d .E//; F 7! d .F /
is a closed embedding of analytic manifolds.

2. Show that
Y
r1
Flagn .E/ ! Grassmi .E/; .W1 Wr1 / 7! .Wi /1i r1
i D1
6.4 Problems 137
is a closed embedding of analytic manifolds.

3. Deduce that Grassd .E/ and, more generally, Flagn .E/ are compact.
Problem 6.14. Let .V; h ; i/ be a symplectic K-space, i.e., V is a finite-dimensional

K-vector space with a non-degenerate alternating bilinear form h ; i. Show that the sym-
plectic group
Sp.V; h ; i/ WD f g 2 GL.V / I hgv; gwi D hv; wi for all v; w 2 V g
is a closed Lie subgroup of GL.V / and endow the set of Lagrangians of .V; h ; i/ (i.e.,
the set of K-subspaces U of V with U ? D U ) with the structure of an analytic quotient
manifold of Sp.V; h ; i/.
Torsors and Non-abelian Čech Cohomology
7
Very often one is confronted with construction problems where it is easier to make the
construction locally. Then the question arises whether these local constructions “glue” to
a global object. For instance, on premanifolds it is often easier to construct objects if the
premanifold is isomorphic to an open ball in Km , which is locally always the case. In this
chapter we study a general technique to deal with such gluing problems – at least if the
difference of two possible local objects is given by a sheaf of groups. For instance two
primitives of a K-valued function always differ by a locally constant K-valued function
and these form a sheaf of groups.
7.1 Torsors
The precise mathematical notion that captures “local objects always differing by a sheaf
of groups” is that of a torsor. The idea is to encode all local objects into a sheaf on which
a sheaf of groups then acts simply transitively.
Definition 7.1 (Torsor). Let X be a topological space, let F and G be sheaves of sets on
X and let A be a sheaf of groups on X.
1. An A-sheaf on X is a sheaf of sets F on X together with a morphisms of sheaves of

sets
aW A F ! F
such that for every open subset U X the map
aU W A.U / F.U / ! F.U /
is an action of the group A.U / on the set F.U /.

DOI 10.1007/978-3-658-10633-1_7
140 7 Torsors and Non-abelian Čech Cohomology
A morphism F ! G of A-sheaves is a morphism 'W F ! G of sheaves such that for all

U X the map
'U W F.U / ! G.U /
is A.U /-equivariant. We obtain the category of A-sheaves on X.
2. An A-sheaf T on X is called an A-pseudotorsor if for all U X open the action
A.U / T .U / ! F.U / is simply transitive.
3. An A-pseudotorsor T on X is called an A-torsor if there exists an open covering
S
X D i Ui such that T .Ui / ¤ ; for all i.
We obtain the full subcategory of A-torsors of the category of A-sheaves. We denote
this category by .Tors.A//.
Example 7.2. Let M C be open and let f W M ! C be a holomorphic function. Define

a sheaf Primf on M by
˚
Primf .U / WD F W U ! C holomorphic I F 0 D f jU
for U M open. Then any two primitives of f jU differ by a locally constant func-
tion. In other words, the constant sheaf CM acts on it (recall: CM .U / D f hW U !
C I h locally constantg) by addition:
CM .U / Primf .U / ! Primf .U /; .h; F / 7! h C F:
This makes Primf into a CM -pseudotorsor because two primitives of a holomorphic func-
tions only differ by a locally constant function. Now the local existence of primitives just
means that the CM -pseudotorsor Primf is in fact a torsor. Here this is the case because we
may find a covering of M by convex open subsets Ui and Primf .Ui / ¤ ; for all i 2 I .
Next we come to the question of global existence of objects on a space X. If the local
objects are encoded by a torsor T on X, then this is equivalent to the question of whether
T .X/ is non-empty.
Recall that a pointed set is a pair .H; x/ consisting of a set H and an element x 2 H .
A morphism of pointed sets .H; x/ ! .H 0 ; x 0 / is a map 'W H ! H 0 with '.x/ D x 0 .
We obtain the category of pointed sets. Every group G gives rise to a pointed set .G; e/,
where e is the neutral element.
Definition 7.3. Let X be a topological space and let A be a sheaf of groups on X.
1. Let A act on itself by left multiplication. This makes A into an A-torsor. An A-torsor
that is isomorphic to A is called a trivial A-torsor.
2. Define H 1 .X; A/ as the set of isomorphism classes of A-torsors on X. It is a pointed
set; the distinguished element is the isomorphism class of trivial torsors. It is called
the (first) cohomology set of A.
7.1 Torsors 141
3. Let T be an A-torsor over X, and let U D .Ui /i 2I be an open covering of X. Then

we say that U trivializes T if T .Ui / ¤ ; for all i 2 I . We denote by H 1 .U; A/ the
isomorphism classes of A-torsors over X trivialized by U. Then
[
H 1 .X; A/ D H 1 .U; A/:
U
Proposition 7.4. Let X be a topological space and let A be a sheaf of groups on X.
1. Every morphism in the category .Tors.A// is an isomorphism.

2. An A-torsor T over X is trivial if and only if T.X/ ¤ ;.
Proof. (1). Let T and S be A-torsors on X and let 'W T ! S be an A-equivariant

S
morphism of sheaves on X. If X D i Ui is an open covering and ' jUi is an isomorphism
for all i 2 I , then ' is an isomorphism (Proposition 3.18). Choosing a covering .Ui /i that
trivializes S and T and replacing X by one of the Ui we may assume that T .X/ ¤ ; ¤
S.X/. Then T .U / ¤ ; ¤ S.U / for all U X open. Hence it suffices to show: Let
A be a group acting simply transitively on two non-empty sets X and Y . Then every A-
equivariant map 'W X ! Y is bijective. Let x 2 X, y WD '.x/. Then we have a diagram
with bijective diagonal maps
A
g7!gx g7!gy
'
X Y;
which commutes because ' is G-equivariant. Hence ' is bijective.

(2). If T Š A, then T .X/ ¤ ; because A.X/ is a group, in particular it is non-empty.
Conversely, assume there exists s 2 T .X/. Then
A ! F; A.U / 3 g 7! g .s jU / 2 T .U /; U X open
is an isomorphism of A-torsors.
The first important result says that for constant group sheaves on simply connected
spaces H 1 is trivial:
Proposition 7.5. Let X be a topological space. Let A be a locally constant sheaf of

groups on X. Suppose that X is simply connected and locally path connected. Then A is
a constant sheaf of groups and H 1 .X; A/ D 0.
Proof. As any A-torsor T is a locally isomorphic to A (Proposition 7.4), it is a locally

constant sheaf on X. Hence it suffices to show that any locally constant sheaf F on X
is constant. Let pW E ! X be the étalé space corresponding to F. Then p is a covering
map (Example 3.41). As X is simply connected and locally path connected, this covering
map has to be trivial (Proposition 2.41), i.e., E D X F , where F is some fiber of p,
considered as a discrete topological space. This shows that F is a constant sheaf.
Using Example 7.2 we obtain the following criterion for the existence of global primi-
tives.
Example 7.6. Let M C be open and let f W M ! C be a holomorphic function. Then

the torsor Primf (Example 7.2) is trivial if and only if there exists F W M ! C such that
F0 D f .
If H 1 .M; CM / D 0, then every holomorphic function on M has a primitive. Hence we
can deduce from Proposition 7.5 the classical result that if M is simply connected, then
every holomorphic function on M has a primitive.
Note that to get this result we used that Primf is a torsor (and not only a pseudotorsor),
i.e., that locally a primitive always exists. Hence the cohomological formalism allowed us
to extend a local existence result to a global one.
7.2 Non-abelian Čech Cohomology
Notation: Let X be a topological space and A a sheaf of groups on X.

We may also give a more concrete description of H 1 .X; A/ in terms of cocycles,
which is often advantageous for concrete calculations. To ease the notation, for two sec-
tions s 2 .U; A/ and t 2 .V; A/ we will often write st 2 .U \ V; A/ instead of
s jU \V t jU \V , and s D t instead of s jU \V D t jU \V .
Definition and Remark 7.7 (Čech cohomology). Fix an open covering U D .Ui /i 2I
of X. For i1 ; : : : ; ir 2 I define Ui1 :::ir WD Ui1 \ \ Uir .
1. A Čech 1-cocycle of A on U is a tuple D .gij /i;j 2I , where gij 2 A.Uij / such that
the cocycle condition
gkj gj i D gki 2 A.Uij k / (7.1)
holds for all i; j; k. Choosing i D j D k we see that gi i D 1; choosing k D i we see
that gij D gj1i for all i; j 2 I . The set of all Čech 1-cocycles of A on U is denoted
by ZL .U; A/.
1
2. Two Čech 1-cocycles and 0 on U are called cohomologous if there exist hi 2 A.Ui /
for all i such that we have
hi gij D gij0 hj 2 A.Uij /

7.2 Non-abelian Čech Cohomology 143
for all i; j 2 I . This is easily checked to be an equivalence relation on ZL 1 .U; A/. The
equivalence classes are called Čech cohomology classes, and the set of cohomology
classes of Čech 1-cocycles on U is called the (first) Čech cohomology of A on U and
is denoted by HL 1 .U; A/. This is a pointed set in which the distinguished element is
the cohomology class of the cocycle .eij / with eij D 1 for all i; j . It is denoted by 1.
Then a Čech 1-cocycle .gij / 2 ZL 1 .U; A/ is cohomologous to 1 if there exist hi 2 A.Ui /

such that gij D h1
i hj .
Definition and Remark 7.8 (Refinement of Čech cohomology). Let V D .Vj /j 2J be

a second open covering of X. A morphism of open coverings V ! U is a map W J ! I
such that Vj U .j / for all j 2 J (then V is a refinement of U). We obtain the category
of open coverings (Cov(X)) of X.
If .gi i 0 / is a Čech 1-cocycle on U, then the tuple .g/jj 0 D g .j / .j 0 /jVj \Vj 0 is a Čech
1-cocycle on V . It is easy to see that this construction induces a map
W HL 1 .U; A/ ! HL 1 .V ; A/:
Therefore we obtain a functor U ! HL 1 .U; A/ from (Cov(X))opp to the category of

pointed sets. We leave it to the reader to make the (easy) check that the category
(Cov(X))opp is filtered (Problem 7.1). The pointed set
HL 1 .X; A/ WD colim HL 1 .U; A/;

U
is called the (first) Čech cohomology of A on X. Note that the explicit description of
a filtered colimit of sets in Appendix Example 13.38 shows that the map from HL 1 .U; A/
to the colimit in the category of sets the distinguished points are all sent to the same point
e. Hence the colimit of the HL 1 .U; A/ in the category of pointed sets is the colimit in the
category of sets endowed with the point e.
Remark 7.9 (Abelian Čech cohomology). If A is a sheaf of abelian groups, the set
of Čech 1-cocycles of A on U forms an abelian group with respect to componentwise
multiplication .gij /.gij0 / WD .gij gij0 /. The equivalence relation of being cohomologous
is compatible with the group structure and therefore HL 1 .U; A/ is an abelian group. For
every refinement V of U the map HL 1 .U; A/ ! HL 1 .V ; A/ is a homomorphism of abelian
groups. Therefore HL 1 .X; A/ is an abelian group.
Example 7.10 (Čech cohomology of the circle). Let U D .U0 ; U1 / be an open covering
of X consisting of two open subsets. Then
ZL 1 .U; A/ ! A.U0 \ U1 /;
and a Čech 1-cocycle of A on U is given by an element g D g01 2 A.U0 \U1 /. Moreover,

g; g 0 2 A.U0 \ U1 / are cohomologous if and and only if there exist hi 2 A.Ui /, i D 0; 1,
such that g 0 D h0 gh1 .
Let X D S 1 D f z 2 C I jzj D 1 g, A be an abelian group, A D AS 1 the associated
constant sheaf, U0 D S 1 n f1g, U1 D S 1 n f1g. Then
A.U0 / D A.U1 / D A
because
˚ U0 and U1 are connected. Moreover U0 \ U1 D WC t W with W˙ WD
z 2 S 1 I ˙Im.z/ > 0 . Hence
A.U0 \ U1 / ! A A; f 7! .f jWC ; f jW /:
The restriction map A.Ui / 7! A.U0 \ U1 / is given by A 3 a 7! .a; a/ 2 A2 . Hence we

see:
g D .aC ; a / 2 A2 and g 0 D .aC

0
; a0 / cohomologous
0
, 9 b0 ; b1 2 AW aC D b0 C a C C b1 ; a0 D b0 C a C b1
0
, 9 b 2 AW aC D aC C b; a0 D a C b
0
, aC a0 D aC a :
In other words H 1 .U; AS 1 / Š A.

` Q
Remark 7.11. Let X D i Xi . Then HL 1 .X; A/ D i 2I HL 1 .Xi ; AjXi /.
Remark 7.12 (Čech cohomology and torsors). We will now construct an isomorphism
of pointed sets
H 1 .X; A/ Š HL 1 .X; A/: (7.2)
Let T be an A-torsor and let U D .Ui /i 2I be an open covering of X that trivializes T , i.e.,
T .Ui / ¤ ; for all i. Set Uij D Ui \ Uj for all i; j 2 I . Choose elements ti 2 T .Ui /. As
A acts simply transitively, there exists a unique element gij 2 A.Uij / such that gij tj D ti .
We have gkj gj i ti D tk D gki ti and thus gkj gj i D gki . Hence .gij /ij is a Čech 1-cocycle.
For a different choice of elements ti we obtain a cohomologous 1-cocycle. Hence this
construction yields a morphism of pointed sets
cA;U W H 1 .U; A/ ! HL 1 .U; A/: (7.3)

7.2 Non-abelian Čech Cohomology 145
If W V ! U is a morphism of open coverings, one easily checks that
cA;U
H 1 .U; A/ HL 1 .U; A/
(7.4)
cA;V
H 1 .V ; A/ HL 1 .V ; A/
is a commutative diagram. By taking inductive limits we obtain a map of pointed sets
cA W H 1 .X; A/ ! HL 1 .X; A/: (7.5)
Proposition 7.13. The maps cA;U are isomorphisms of pointed sets. In particular, cA is
an isomorphism.
Proof. We define an inverse of cA;U as follows. Let .gij / be a representative of a class in

HL 1 .U; A/. For V X open we set
( )
Y
T .V / D .ti / 2 A.Ui \ V / I ti tj1 D gij : (7.6)
i
Endowed with the obvious restriction maps, T is a sheaf. We define an A-action on T

via g .ti /i D .ti g 1 /i . For a fixed k 2 I and for V Uk the map A.V / ! T .V /,
g 7! .gi k g 1 /i , defines an isomorphism of AjUk -sheaves AjUk ! T jUk whose inverse is
given by .ti / 7! tk1 . Thus T is an A-torsor that is trivialized by U. If .gij / is replaced
by a cohomologous cocycle .gij0 / D .hi gij hj1 / with an associated A-torsor T 0 , then
.ti / 7! .hi ti /i defines an isomorphism T ! T 0 of A-torsors.
Remark 7.14. The commutative diagram (7.4) then also shows that the maps W HL 1
.U; A/ ! HL 1 .V ; A/ are injective and depend only on U and V (but not on the choice
of a morphism W V ! U). In particular HL 1 .U; A/ ! HL 1 .X; A/ is injective.
Corollary 7.15. If A and A0 are two sheaves of groups on a topological space X, then
H 1 .X; A A0 / D H 1 .X; A/ H 1 .X; A0 /.
Proof. This is clear for Čech cohomology.
For concrete calculations of HL 1 .X; A/ it suffices very often to calculate HL 1 .U; A/ for
a sufficiently nice open covering U by the following result.
Proposition 7.16. Let U D .Ui /i 2I be an open covering of X such that HL 1 .Ui ; A/ D 0

for all i 2 I . Then HL 1 .U; A/ ! HL 1 .X; A/ is an isomorphism.
Proof. It suffices to show: Let V D .Vj /j 2J be a refinement of U and let W J ! I be

a morphism of open coverings. Then W HL 1 .U; A/ ! HL 1 .V ; A/ is bijective.
It is injective by Remark (7.14). Hence it remains to show that is surjective. Let
Ui \ V WD .Ui \Vj /j 2J . This is an open covering of Ui . As H .Ui \ V ; A/ ! HL 1 .Ui ; A/
L 1
is injective (Remark 7.14), we find
HL 1 .Ui \ V ; A/ D 1: (*)
Let D .jj 0 /j;j 0 2J 2 ZL 1 .V ; A/ and define .i / 2 ZL 1 .Ui \ V ; A/ by

.i /
jj 0 WD jj 0 jUi \Vj \Vj 0 :
.i /
By (*) there exist gj 2 A.Ui \ Vj / such that
1
.i / .i /
jj 0 D gj gj 0
and hence for all i; i 0 2 I

1 1
.i 0 / .i / .i 0 / .i /
gj gj D gj 0 gj 0 2 A.Ui \ Ui 0 \ Vj \ Vj 0 /: (**)
Now define 2 ZL 1 .U; A/ by

1
.i 0 / .i /
i 0 i jVj WD gj gj :
1
. .j //
This is possible because of (**). Moreover set hj WD gj 2 A.U .j / \ Vj /. Then
we have on Vj \ Vj 0 D U .j / \ U .j 0 / \ Vj \ Vj 0 :
1
. .j // . .j 0 //
hj . .//jj 0 D hj .j / .j 0 /D hj gj gj
1 1
. .j 0 // . .j 0 // . .j 0 //
D gj D gj gj 0 hj 0
D jj 0 hj 0 ;
and hence ./ is cohomologous to .
Corollary 7.17. Let X be a locally path connected space and let G be a group. Let
U D .Ui /i be an open covering of X such that Ui is simply connected for all i. Then
H 1 .U; GX / D H 1 .X; GX /.
Proof. We can apply Proposition 7.16 by Proposition 7.5.

7.3 First Term Sequence of Cohomology 147
Example 7.18. Let X D S 1 , let A be an abelian group, and let U D .U0 ; U1 / WD

.S 1 n f1g; S 1 n f1g/. Then Ui is homeomorphic to the open interval .0; 1/ and hence
simply connected. Hence Corollary 7.17 implies that HL 1 .S 1 ; AS 1 / D HL 1 .U; AS 1 / Š A
by Example 7.10.
7.3 First Term Sequence of Cohomology
An important tool to calculate cohomology pointed sets is by relating them to already-

known cohomology pointed sets using long exact cohomology sequences. This is particu-
larly useful if the cohomology sets carry a group structure but can be also used in general.
Hence we explain first what is meant by an exact sequence of pointed sets.
Definition 7.19. A sequence of morphisms of pointed sets

'i 1 'i 'i C1 :::
: : : ! .Hi 1 ; ei 1 / ! .Hi ; ei / ! .Hi C1 ; ei C1 / !
is called exact if for all i
Ker.'i C1 / WD f h 2 Hi I 'i C1 .h/ D ei C1 g D Im.'i /:
Note that if 'W .H; e/ ! .H 0 ; e 0 / is a morphism of pointed sets, then Ker.'/ D feg
does not imply that ' is injective in general.
Remark 7.20. Let 'W A ! A0 be a homomorphism of sheaves of groups. Sending a Čech

cocycle .gij / of A to .'.gij // defines a homomorphism of pointed sets
HL 1 .'/W HL 1 .X; A/ ! HL 1 .X; A0 /: (7.7)
We obtain a functor HL 1 .X; / from the category of sheaves of groups on X to the category
of pointed sets.
We can also define H 1 .'/W H 1 .X; A/ ! H 1 .X; A0 / via torsors as follows. Let T be
an A-torsor. We let A act on A0 T via
.a; .a0 ; t// 7! .a0 '.a1 /; at/
for a 2 A.U /, a0 2 A0 .U /, t 2 T .U /, U X open. Let A0 A T be the sheafification of

the presheaf that sends U X open to the set of A.U /-orbits in A0 T . We let A0 act on
A0 A T from the left by multiplication on the first factor. This makes A0 A T into an
A0 -torsor (see Problem 7.3) and T 7! A0 A T defines the map H 1 .'/ of pointed sets.
Remark 7.21. Let X be a topological space and let

1 ! A0 ! A ! A00 ! 1 (7.8)
be an exact sequence of sheaves of groups on X. Define a connecting map
ıW A00 .X/ ! HL 1 .X; A0 / (7.9)
as follows. For g 00 2 A00 .X/ let U D .Ui /i be an open covering of X such that there
exist gi 2 A.Ui / whose image in A00 .Ui / is g 00 jUi . For all i; j let gij0 2 A0 .Uij / be the
unique element that is mapped to gi gj1 2 A.Ui \ Uj /. Then .gij0 / is a Čech cocycle on
U. A different choice of elements gi yields a cohomologous cocycle. Therefore its class
ı.g 00 / in HL 1 .X; A/ is well defined. It is clear that ı is a morphism of pointed sets.
One can define ı also via torsors as a map ıW A00 .X/ ! H 1 .X; A0 / as follows. Let
g 2 A.X 00 /. Define a sheaf of sets Tg00 on X by
00
˚
Tg00 .U / WD t 2 A.U / I .t/ D g 00 jU : (7.10)
Then A0 .U / acts on Tg00 .U / by .g 0 ; t/ 7! .g 0 /t and this makes Tg00 into an A0 -torsor.
Then ı.g 00 / is the isomorphism class of Tg00 .
Proposition 7.22. The following sequence of pointed sets is exact:

X X
1 ! A0 .X/ ! A.X/ ! A00 .X/
ı HL 1 ./ HL 1 ./ (7.11)
! HL 1 .X; A0 / ! HL 1 .X; A/ ! HL 1 .X; A00 /:
Moreover, one has the following assertions.
1. Assume that A0 is a subgroup sheaf of the center of A (in particular it is a sheaf of

abelian groups). Then ı is a homomorphism of groups, and HL 1 ./ induces an injection
of the group Coker.ı/ into the pointed set HL 1 .X; A/.
2. Assume that A0 , A, and A00 are abelian sheaves. Then the sequence (7.11) is an exact
sequence of abelian groups.
The same sequence exists with H 1 instead of HL 1 and it is an instructive (and not too
difficult) exercise to check all of the above assertions in the language of torsors.
Proof. X injective. Proposition 3.18.

Im.X / D Ker.X /. “” is clear. Let g 2 Ker.X /. Then gx 2 Ker.x / D Im.x /
for all x 2 X. Hence there exists x 2 U x X open and g 0x 2 A0 .U x / such that
.g 0x / D g jU x . For x; y 2 X one has
.g 0x jU x \U y / D g jU x \U y D .g 0y jU x \U y /
7.4 Problems 149
and hence g 0x jU x \U y D g 0y jU x \U y because W A0 .U x \ U y / ! A.U x \ U y / is injective.

Hence there exists g 0 2 A0 .X/ such that .g 0 / D g.
Im.X / Ker.ı/. Let g 00 D X .g/ 2 Im.X /. Then we may choose gi D g jUi in the
definition of ı and hence gij0 D 1 and therefore g 00 2 Ker.ı/.
Im.X / Ker.ı/. Let g 00 2 Ker.ı/, i.e., we find an open covering .Ui /i of X,
gi 2 A.Ui / and h0i 2 A0 .Ui / such that
.gi / D g 00 jUi and .h01 0 1 0 1 0 1

i hj / D gi gj , 1 D .hi /gi gj .hj /
for all i; j 2 I . Replacing gi by .h0i /gi we may assume that gi D gj 2 A.Ui \ Uj /.

Hence there exists g 2 A.X/ such that g jUi D gi . Then .g/jUi D g 00 jUi and hence
.g/ D g 00 .
Im.ı/ D Ker.HL 1 .//. Let .Ui /i be an open covering and let D .gij0 /i;j 2
HL 1 .U; A0 /. Then:
2 Im.ı/ , HL 1 ././ D .gi gj1 /i;j with gi 2 A.Ui /

, 1 ./ is cohomologous to 1:
Im.HL 1 .// D Ker.HL 1 .//. This follows from Im.Ui / D Ker.Ui / for every open
subset Ui of X.
Assertions (1) and (2) are easy to check.
Remark 7.23. One can define the first terms of the cohomology sequence (7.11) in the
following more general situation. Let A0 be a sheaf of subgroups of a sheaf of groups
A and let A=A0 be the sheafification of the presheaf of left cosets U 7! A.U /=A0 .U /.
Then .A=A0 /.X/ is a pointed set (the distinguished element is the image of 1 2 A.X/ in
.A=A0 /.X/). The same construction of ı and the same arguments as above show that one
still has an exact sequence of pointed sets
1 ! A0 .X/ ! A.X/ ! .A=A0 /.X/

ı (7.12)
! HL 1 .X; A0 / ! HL 1 .X; A/:
7.4 Problems
Problem 7.1. Let X be a topological space. Show that the category (Cov(X))opp is fil-
tered.
Problem 7.2. Let X be a topological space, G a sheaf of groups acting on a sheaf of sets
T . The action is called transitive (respectively free) if the morphism G T ! T T ,
.g; t/ 7! .t; gt/ for g 2 G.U /, t 2 T .U /, U X open, is a surjective (respectively
injective) morphism of sheaves.
1. Show that G acts freely on T if and only if for all U X open and for all t 2 T .U /
the stabilizer of t in G.U / is trivial.
2. Show that T is a pseudotorsor if and only if the G-action is free and transitive.
3. Let e be the final sheaf, i.e., e.U / is the set consisting of a single element for all
U X open. Show that e is a final object in (Sh(X)) and that a G-pseudotorsor T is
a torsor if and only if T ! e is an epimorphism.
Problem 7.3. Let X be a topological space and G be a sheaf of groups on X. Let T

(respectively T 0 ) be a sheaf of sets on X that is endowed with a right action T G ! T
(respectively a left action G T 0 ! T 0 ) of G. Define a left G-action of T T 0 by
g .t; t 0 / WD .tg 1 ; gt 0 / for g 2 G.U /, t 2 T .U /, t 0 2 T 0 .U /, U X open. The
contracted product T G T 0 is defined as the sheaf associated with the presheaf whose
value on U is the set of G.U /-orbits of T .U / T 0 .U /.
Let 'W G ! G 0 be a homomorphism of sheaves of groups on X and let Gr0 be the sheaf
G 0 with the right G-action given by g 0 g WD g 0 '.g/ on local sections. Let G 0 act on
Gr0 G T from the left on Gr0 G T via the first factor.
1. Show that for every sheaf E on X one has Hom(Sh(X )) .E; T G T 0 / D Hom(Sh(X )) .E; T
T 0 /G , where . /G denotes G-invariants.
2. Show that T 7! Gr0 G T defines a functor from the category of sheaves with G-action
to the category of sheaves with G 0 -action. Show that this functor is left adjoint to the
functor T 0 7! T 0' , where T 0' is the sheaf T 0 with G-action .g; t 0 / 7! '.g/t 0 .
3. Suppose that G acts transitively (respectively freely) on T (Problem 7.2). Show that G 0
acts transitively (respectively freely) on Gr0 G T . Deduce that if T is a G-torsor, then
Gr0 G T is a G 0 -torsor defining a map H 1 .'/W H 1 .X; G/ ! H 1 .X; G 0 / of pointed
sets.
4. Show that the following diagram is commutative:
H 1 .'/
H 1 .X; G/ H 1 .X; G 0 /
cG cG 0
HL 1 .'/
HL 1 .X; G/ HL 1 .X; G 0 /:
Problem 7.4. Let M C be open, f W M ! C be a holomorphic map and let Lf .U /

be the set of logarithms of f , i.e., the set of holomorphic maps gW U ! C such that
exp ıg D f jU for U M open. Endow Lf with the structure of a ZM -pseudotorsor,
which is a torsor if and only if f .z/ ¤ 0 for all z 2 M . Show that a holomorphic map
f W M ! C has a logarithm if M is simply connected.
Problem 7.5. Let X be a topological space, x 2 X, G a group, and let G be the skyscraper
sheaf in x with values G (Problem 3.6). Show that H 1 .X; G/ D 1.
7.4 Problems 151
Problem 7.6. Let n 1 and let A be an abelian group.
1. Show H 1 .P 1 .R/; AP 1 .R/ / D A and H 1 .P n .R/; AP n .R/ / D f a 2 A I 2a D 0 g for

n 2.
2. Show that H 1 .P n .C/; AP n .C/ / D 0.
Problem 7.7. Let X D P 1 .C/, U D .U0 ; U1 / with U0 D P 1 .C/ n f.1 W 0/g and U1 D
P 1 .C/ n f.0 W 1/g, and let OP 1 .C/ the sheaf of holomorphic functions on P 1 .C/. Show
that H 1 .U; OP 1 .C/ / D 0. In Theorem 10.22 we will see that H 1 .C; OC / D 0. Deduce
that H 1 .P 1 .C/; OP 1 .C/ / D 0.
Hint: Use the theory of Laurent series.
Problem 7.8. Let X C be open and let OX be the sheaf of holomorphic functions. Let
f W X ! C be a holomorphic function.
1. Show that the following sequence of sheaves of abelian groups is exact:

dlog
1 ! .C /X ! OX ! OX ! 0;
where dlog.g/ WD g 0 =g.

2. Show that if X is simply connected, then there exists a holomorphic map gW X ! C
with g 0 =g D f .
3. Show that the sheaf of holomorphic solutions of the differential equation u0 D f u is
a locally constant sheaf of 1-dimensional C-vector spaces. Show that the differential
equation has a solution without zeros if X is simply connected.
Bundles
8
We will now study morphisms of premanifolds W X ! B that look locally on B like

a fixed morphism pW Z ! B. We then call a twist of p. Important special cases are
fiber bundles, where pW B F ! B is a projection. Moreover, one often endows twists
with an additional datum that restricts the changes between different local isomorphisms
of p and to a fixed subsheaf of the sheaf all automorphisms of p. For fiber bundles
this subsheaf will usually be given by a faithful action of a Lie group G on the fiber F .
An important special case of a fiber bundle is the notion of a principal bundle for a Lie
group G. They are in a way the universal fiber bundles for a given structure group G
(Remark 8.22).
Most of the results in Sect. 8.1 hold verbatim if one replaces “premanifold” with “-
topological space” and “Lie group” with “topological group”. Notable exceptions are
Remark 8.10 and Example 8.21 as for topological groups the projection G ! G=H does
not have local sections in general.
In the next section we study a very important special case of fiber bundles: vector
bundles. This is the main topic of this chapter. There are two ways to look at vector
bundles. The first point of view is that of a fiber bundle where the typical fiber is a vector
space and where the local coordinate changes are linear. This is explained in Sect. 8.2.
The second point of view is to consider vector bundles as locally free modules over the
structure sheaf. Hence we introduce such modules in Sect. 8.3 and prove in Sect. 8.4 that
both point of views are equivalent.
Among the most important vector bundles is the tangent bundle whose fiber at a point
p of a premanifold M is the tangent space Tp .M /. This will be described as a geometric
vector bundle as well as a locally free module. The dual of the tangent bundle is the
vector bundle of differential 1-forms and by taking exterior powers we obtain differential
forms of arbitrary degree. We conclude the chapter by defining the de Rham complex of
a premanifold.

DOI 10.1007/978-3-658-10633-1_8
154 8 Bundles
8.1 Twists, Fiber Bundles, and Principal Bundles
Twists will be premanifolds endowed with a morphism to a fixed premanifold B, the “base
premanifold”. Hence let us introduce the following general terminology.
Definition 8.1. Let C be a category and let B be an object of C . Define the category of
B-objects of C , denoted by C=B :
(a) Objects in C=B are morphisms pW X ! B in C , usually denoted by .X; p/.

(b) A morphism .X; p/ ! .Y; q/ in C=B is a morphism W X ! Y in C such that q ıf D
p. It is called a B-morphism.
Composition is given by composition in C .
In the sequel we will take for C the category of premanifolds (real C ˛ for a fixed ˛ or
complex). We fix a premanifold B. Then objects in C=B are called B-premanifolds.
Twists
Let pW Z ! B be a morphism of premanifolds, in other words, .Z; p/ is a B-premanifold.
Let
Aut.p/ D AutB .Z/
be the sheaf of groups on B that attaches to U B the group of automorphism of U -
premanifolds ˛W p 1 .U / ! p 1 .U /.
Definition and Remark 8.2 (Twists). We fix a morphism pW Z ! B and a sheaf G of

subgroups of AutB .Z/.
1. For a morphism of premanifolds W X ! B a G-twisting atlas of p for is a family

.Ui ; hi /i 2I where .Ui /i is an open covering of B and where hi is an isomorphism of
Ui -premanifolds
hi W 1 .Ui / ! p 1 .Ui / (8.1)
such that for all i; j 2 I we have
gij WD hi j 1 .Uij / ı hj1 jp1 .Uij / 2 G.Uij /; (8.2)
where as usual Uij WD Ui \ Uj .

Note the similarity to the notion of an atlas of a premanifold, where one also has
a family of charts covering the premanifold such that the chart changes have certain
properties (namely being C ˛ -maps).
2. Given such a G-twisting atlas .Ui ; hi /i 2I and a refinement .Vj /j 2J of .Ui /i defined by
W J ! I such that Vj U .j / we let kj W 1 .Vj / ! p 1 .Vj / be the restriction of
8.1 Twists, Fiber Bundles, and Principal Bundles 155
h .j / . Then .Vj ; kj /j 2J is again a G-twisting atlas of p for , called a refinement of

.Ui ; hi /i 2I .
Two G-twisting atlases .Ui ; hi /i 2I and .Ui0 ; h0i /i 2I 0 are called equivalent if there ex-
ists refinements .Vj ; kj /j 2J and .Vj ; kj0 /j 2J of .Ui ; hi /i 2I and .Ui0 ; h0i /i 2I 0 respectively
(with the same underlying open covering .Vj /j ) such that kj ı .kj0 /1 2 G.Vj /.
3. A twist of p with structure sheaf G is a morphism of premanifold W X ! B together
with an equivalence class of G-twisting atlases of p for .
4. Let D .W X ! B; ŒUi ; hi i 2I / and 0 D . 0 W X 0 ! B; ŒUi ; h0i i 2I / be twists of p
with structure sheaf G (after replacing G-twisting atlases by equivalent atlases we can
assume that the underlying open covering of both atlases is the same). An isomorphism
! 0 of twists of p with structure sheaf G is an isomorphism of B-manifolds
F W X ! X 0 such that for all i 2 I there exists gi 2 G.Ui / making the following
diagram commutative:
F
1 .Ui / 01 .Ui /
hi h0i
gi
p 1 .Ui / p 1 .Ui /:
Remark and Definition 8.3 (Twists with full structure sheaf). If G D Aut.p/, then
any two G-twisting atlases are equivalent and it suffices to ask only for the existence of
an Aut.p/-twisting atlas. In other words, a twist of p with Aut.p/-structure is simply
a morphism W X ! B of premanifolds such that there exists an open covering .Ui /i
of B and isomorphism of Ui -premanifolds 1 .Ui / ! p 1 .Ui /. In this case we simply
speak of a twist of p.
In general the equivalence class of a G-twisting atlas is a datum containing additional

information: There exist non-isomorphic twists of p with structure sheaf G such that the
underlying B-manifolds are isomorphic (Problem 8.1).
Remark 8.4 (Twists and Čech cohomology). Let pW Z ! B be a morphism of preman-

ifolds and G a sheaf of subgroups of AutB .Z/. Let D .W X ! B; ŒUi ; hi i 2I / be a twist
of p with structure sheaf G. Then the gij defined in (8.2). Satisfy for all i; j; k 2 I the
cocycle condition
gkj gj i D gki 2 G.Uij k /: (8.3)
We obtain an element .gij /ij 2 ZL 1 .U; G/, where U WD .Ui /i . Let U . / be its class in
HL 1 .U; G/. Refining .Ui ; hi /i by a refinement V replaces U . / by its image in HL 1 .V ; G/.
If one chooses an equivalent G-twisting atlas, then .gij /ij is replaced by a cohomologous
cocycle. Similarly, any isomorphism of twists replaces .gij /ij by a cohomologous cocy-
cle. Hence the image . / of U . / in HL 1 .B; G/ depends only on the isomorphism class
of the twist of p with structure sheaf G.
156 8 Bundles
Conversely, the twist (or rather a B-premanifold isomorphic to W X ! B) can be

glued from the family .p 1 .Ui //i along the p 1 .Uij / using the automorphisms gij , see
Proposition 4.11. Using the notation there, the isomorphisms hi are then given by the
inverse of the maps i considered as isomorphisms onto their image.
Altogether, we obtain the following classification result of twists.
Theorem 8.5. The map constructed in Remark 8.4 yields a bijection of pointed sets
( )
isomorphism classes of twists of p
W ! HL 1 .B; G/: (8.4)
with structure sheaf G
The trivial class in HL 1 .B; G/ corresponds to the isomorphism class of the trivial twist
pW Z ! B endowed with the equivalence class of the trivial G-twisting atlas .B; idZ /.
Via the isomorphism HL 1 .B; G/ Š H 1 .B; G/ (Proposition 7.13), every isomorphism

class of a twist with structure sheaf G also corresponds to the isomorphism class of a
G-torsor. This can even be made into an equivalence of categories; see Problem 8.2 for
details.
Remark 8.6. One can work more generally with a not necessarily injective homomor-
phism ˛W G ! Aut.p/. In this case one should add the .gij /i;j as an additional datum to a
G-twisting atlas. Using the obvious notion of “refinement” for such a G-twisting atlas one
obtains the notion of equivalence for such twisting atlases and hence a generalization of
a twist. The constructions in Remark 8.4 then still yield a bijective map
( )
isomorphism classes of twists of p
! HL 1 .B; G/:
with structure sheaf G
We leave details to the reader.
Remark 8.7 (Pullback of twists). Let pW Z ! B be a submersion and let rW B 0 ! B be

a morphism of premanifolds. We set Z 0 WD B 0 B Z. As p is a submersion, we obtain by
Corollary 5.50 a commutative diagram
Z0 Z
p0 p
r
B0 B
of premanifolds, where r .p/ WD p 0 is the projection. Moreover p 0 is again a submersion.

Let W X ! B be a twist of p. As looks locally on B like p, is also a submersion,

X 0 WD B 0 B X is a premanifold, and the projection r ./ WD 0 W X 0 ! B 0 is a submer-
sion. If .Ui ; hi /i is a twisting atlas of p for , then .f 1 .Ui /; idB 0 hi /i is a twisting atlas
of p 0 for 0 . In particular, 0 is a twist of p 0 , called the pullback of via r.
Sheaves of Groups defined by Lie Groups

The most important special case of a twist is a fiber bundle where the structure sheaf is
given by the action of a Lie group on the fiber. Hence let us first study how Lie groups
give rise to sheaves of groups on a premanifold B.
Definition and Remark 8.8. Let M be a premanifold. Let OBIM be the sheaf of mor-
phisms from B to M , i.e., for U B open we let OBIM .U / be the set of morphisms
of premanifolds U ! M . Then OBIM is a sheaf of M -valued functions on B. If
'W M ! N is a morphism of premanifolds, composition with ' yields a morphism of
sheaves 'B W OBIM ! OBIN . We obtain a functor from the category of premanifolds to the
category of sheaves of sets on B.
If M D G is a Lie group, then the group structure on G endows OBIG .U / with a group
structure by pointwise multiplication making OBIG into a sheaf of groups. For a homomor-
phism 'W G ! H of Lie groups, 'B is a morphism of sheaves of groups OBIG ! OBIH .
We obtain a functor from the category of Lie groups to the category of sheaves of groups
on B.
Example 8.9.
1. If G is an abstract group considered as a 0-dimensional Lie group, then OBIG is the

constant group sheaf GB on B attached to G.
2. Let G D .K; C/ considered as a 1-dimensional Lie group. Then OBIK is the underly-
ing sheaf of abelian groups of the structure sheaf OB on B.
3. Let G D GLn .K/. Then GLn;B WD OBIGLn .K/ is the group sheaf that attaches to
U B open the group GLn .OB .U //. In particular OB WD OBIK is the sheaf of
K-valued nowhere vanishing morphisms.
To calculate HL 1 .B; OBIG / it is useful to have exact cohomology sequences. To use this
tool, we show that every short exact sequence of Lie groups yields an exact sequence of
sheaves on any premanifold.
Remark 8.10. Let W H ,! G be the inclusion of a Lie subgroup into a Lie group G
and let G=H be the quotient manifold (Corollary 6.23). Then B identifies OBIH with
a sheaf of subgroups of OBIG . We claim that OBIG=H is the sheaf OBIG =OBIH of left cosets
(Remark 7.23).
Indeed, composition with the projection W G ! G=H yields a morphism of sheaves
B W OBIG ! OBIG=H and for two morphisms g; g 0 W U ! G, U B open, one has
158 8 Bundles
ı g D ı g 0 if and only if .g 1 g 0 /.b/ 2 H for all b 2 U . Hence B induces an

injective morphism of sheaves OBIG =OBIH ! OBIG=H . As is a surjective submersion,
it has locally on G=H sections that are morphisms of premanifolds. This shows that B is
surjective. Hence OBIG =OBIH ! OBIG=H is also surjective.
We obtain by Remark 7.23 a cohomology sequence
1 ! Hom(PMfd) .B; H / ! Hom(PMfd) .B; G/ ! Hom(PMfd) .B; G=H /

ı (8.5)
! HL 1 .B; OBIH / ! HL 1 .B; OBIG /:
If H is normal in G, then G=H is a Lie group and (8.5) can be extended by HL 1 .X; OBIG=H /
on the right-hand side (Proposition 7.22). If G is an abelian Lie group, this sequence is an
exact sequence of abelian groups.
Example 8.11. We consider the complex exponential sequence

exp.2i /
0 ! Z ! C ! C ! 1;
which we consider as an exact sequence of complex Lie groups. Hence we obtain by

Remark 8.10 for every complex premanifold B an exact sequence of sheaves of abelian
groups
f 7!exp.2if /
0 ! ZB ! OB ! OB ! 1;
yielding a cohomology sequence
0 ! Z0 .B/ ! OB .B/ ! OB .B/

(8.6)
! HL 1 .B; ZB / ! HL 1 .B; OB / ! HL 1 .B; OB /:
Hence we deduce that every non-vanishing holomorphic function on B has a logarithm

if and only if H 1 .B; ZB / ! H 1 .B; OB / is injective. This is for instance the case if
H 1 .B; ZB / D 0 and hence if B is simply connected (Proposition 7.5).
Moreover, by (8.15) below we will see that HL 1 .B; OB / classifies (complex holomor-
phic) line bundles on B. To use the cohomological formalism to study HL 1 .B; OB / it
would be desirable to extend (8.6) to the right by defining higher cohomology groups
H p .B; / for p 2. This will be the topic of Chap. 10.
Fiber Bundles
We now define fiber bundles over a fixed premanifold B as twists of the projection B
F ! B, where F is some fixed premanifold. The structure sheaf will be given by a faith-
ful action of a Lie group G on F .
Remark and Definition 8.12 (Fiber bundles). Let us fix a premanifold F , a Lie group
G, and a faithful action of G on F by a morphism of premanifolds G F ! F . Let
pW B F ! B be the projection. Then
OBIG ! Aut.p/; g 7! .b; f / 7! .b; g.b/f / (8.7)
(g 2 OBIG .U /, .b; f / 2 U F , U B open) is a homomorphism of sheaves of groups

over B. It is injective because the G-action on F is faithful. We use it to consider OBIG as
a sheaf of subgroups of Aut.p/.
A twist W X ! B of pW B F ! B (without specifying any structure group, Re-
mark 8.3) is called a fiber bundle with basis B and typical fiber F . A twist .W X !
B; ŒUi ; hi i 2I / of pW B F ! B with structure group sheaf OBIG is called a fiber bundle
with basis B, typical fiber F , and structure group G. A OBIG -twisting atlas for a fiber
bundle with basis B and typical fiber F is called a G-bundle atlas. An equivalence class
of G-bundle atlases is called a G-bundle structure. In other words, a fiber bundle with
basis B, typical fiber F , and structure group G is the same as a fiber bundle with basis B
and typical fiber F together with the datum of a G-bundle structure.
An isomorphism of twists of p with structure group sheaf OBIG is called an isomor-
phism of fiber bundles with basis B, typical fiber F , and structure group G. We obtain
a category in which all morphisms are isomorphisms, denoted by .FIBF;G .B//. We denote
by FibF;G .B/ the set of isomorphisms classes in .FIBF;G .B//.
By Theorem 8.5 we have a bijection
FibF;G .B/ ! HL 1 .B; OBIG /: (8.8)
The trivial cocycle in HL 1 .B; OBIG / corresponds to the isomorphism class of the trivial
fiber bundle pW B F ! B endowed with the equivalence class of the trivial G-bundle
atlas .B; idBF /.
Very often, one writes also simply HL 1 .B; G/ instead of HL 1 .B; OBIG /.
Remark 8.13. If the action of G on F is not faithful, the homomorphism (8.7) is not
injective and one has to fix the cocycle defining the fiber bundle as an extra datum (Re-
mark 8.6).
Example 8.14. Let F be a discrete space, considered as a 0-dimensional premanifold,

and consider the group Aut.F / of all permutations of F as a 0-dimensional Lie group.
Then the action of Aut.F / on F is faithful and a fiber bundle with basis B, typical fiber
F , and structure group Aut.F / is the same as a covering space pW BQ ! B of B such that
p 1 .b/ Š F for all b 2 B (Remark 8.3).
Example 8.15 (Möbius band). Let B D S 1 D f z 2 C I jzj D 1 g, F D .1; 1/ and let

G WD f˙1g, considered as a 0-dimensional real Lie group, act on F via multiplication.
160 8 Bundles
By Example 7.18 we have HL 1 .S 1 ; GS 1 / Š HL 1 .U; GS 1 / D ZL 1 .U; GS 1 /= where U D

.U0 ; U1 / WD S 1 n f1g; S 1 n f1g/. Moreover ZL 1 .U; GS 1 / D GS 1 .U0 \ U1 / D G G and
."C ; " / ."0C ; "0 / if "C " D "0C "0 .
Therefore there are precisely two isomorphism classes of fiber bundles with typical
fiber .1; 1/ and structure group f˙1g over S 1 . The fiber bundle corresponding to the
trivial class is B F .
The non-trivial isomorphism class of the fiber bundle is given by the cocycle .C1; 1/ 2
ZL 1 .U; GS 1 / (or
˚ by the cohomologous cocycle .1; C1/). If we write U0 \U1 D WC [W
with W˙ WD z 2 S 1 I ˙Im.z/ > 0 , then it is obtained from gluing U0 F and U1 F
along .U0 \ U1 / F via the automorphism
8
<.b; x/; if x 2 WC I
.U0 \ U1 / F 3 .b; x/ 7!
:.b; x/; if x 2 W :
We obtain the Möbius band (Example 4.30).
Fiber bundles “inherit” the following properties from the trivial bundle B F ! B.
Proposition 8.16. Let F ¤ ; be a premanifold and let W X ! B be a fiber bundle with

basis B and typical fiber F .
1. Then is a surjective submersion.

2. If F is Hausdorff, then is separated. If B and F are manifolds, then X is a manifold.
Proof. 1. Working locally on B we can assume that X D B F and that is the

projection. This shows that is surjective submersion.
2. To see that is separated, we can again assume the is the projection B F ! B.
This is separated because F is Hausdorff. If B is also Hausdorff, then X is Hausdorff
by Corollary 1.27. If B is second countable, then B is a Lindelöf space (Theorem 4.28).
Hence we can assume there exists a countable open covering .Un /n of B such that for all
n one has isomorphisms hn W 1 .Un / ! Un F . As Un F is second countable, X has
a countable covering by second countable open subspaces. Hence X is second countable.

Remark and Definition 8.17 (Pullback of fiber bundles). Let F ¤ ; be a premanifold

and let G be a Lie group acting faithfully on F by a morphism of premanifolds G F !
F . Let rW B 0 ! B be a morphism of premanifolds. Then the projection B 0 B .B F / !
B 0 F is an isomorphism with inverse .b 0 ; f / 7! .b 0 ; r.b 0 /; f /. Hence if W X ! B is
a fiber bundle with typical fiber F , then its pullback 0 WD r ./W X 0 WD B 0 B X ! B 0 is
a fiber bundle with typical fiber F over B 0 (Remark 8.7). Note that the projection X 0 ! X
induces isomorphisms of the fibers 01 .b 0 / ! 1 .r.b 0 //.
Now let .Ui ; hi /i be a G-bundle atlas. We set
h0i WD idr 1 .Ui / hi Wr 1 .Ui / Ui 1 .Ui / D 01 r 1 .Ui /

! r 1 .Ui / Ui .Ui F / D r 1 .Ui / F:
Then .r 1 .Ui /; h0i /i is a G-bundle atlas for 0 . It is immediate that this construction
sends equivalent G-bundle atlases for to equivalent G-bundle atlases for 0 . Therefore
we have constructed for every fiber bundle D .; ŒUi ; hi i / with typical fiber F and
structure group G over B a fiber bundle r . / with typical fiber F and structure group G
over B 0 .
If f W 1 ! 2 is an isomorphism of fiber bundles with typical fiber F and structure
group G over B, then r .f / WD idB 0 f W B 0 B X1 ! B 0 B X2 is an isomorphism of
fiber bundles with typical fiber F and structure group G over B 0 . We obtain a functor
r W .FIBF;G .B// ! .FIBF;G .B 0 // (8.9)
called pullback under r.

If .gij W Ui \ Uj ! G/ij is the cocycle of OBIG given by a G-bundle atlas .Ui ; hi /i
of , then .gij ı rW r 1 .Ui \ Uj / ! G/ij is the cocycle of OB 0 IG given by the G-bundle
atlas .r 1 .Ui /; h0i /i of r . /. The map .gij /ij 7! .gij ı r/ij yields a morphism of pointed
sets r W HL 1 .B; OBIG / ! HL 1 .B 0 ; OB 0 IG /. It is a homomorphism of abelian groups if G is
abelian. We obtain a commutative diagram
FibF;G .B/ HL 1 .B; OBIG /
r r (8.10)
FibF;G .B 0 / HL 1 .B 0 ; OB 0 IG /:
Definition and Remark 8.18 (Morphisms of fiber bundles). Let F ¤ ; be a preman-

ifold and let G be a Lie group acting faithfully on F by a morphism of premanifolds
G F ! F . Let rW B 0 ! B be a morphism of premanifolds and let D .X !
B; ŒUi ; hi i / (respectively 0 D .X 0 ! B 0 ; ŒUi0 ; h0i i /) be fiber bundles with typical fiber
F and structure group G over B (respectively B 0 ). A morphism 0 ! of fiber bundles
with typical fiber F and structure group G along r is an isomorphism RW 0 ! r . / of
fiber bundles with typical fiber F and structure group G over B 0 . Sometimes one writes
R pr2
.Or ; r/ for such a morphism, where rO is the composition X 0 ! B 0 B X ! X.
Principal Bundles
An important special case of a fiber bundle is a principal bundle.
162 8 Bundles
Definition 8.19 (Principal bundle). Let G be a Lie group and consider the faithful action
on itself by left multiplication. A fiber bundle with basis B, typical fiber G, and structure
group G is called a principal G-bundle over B.
We denote by PrincG .B/ the set of isomorphism classes of principal G-bundles. By

(8.8) we have a bijection
PrincG .B/ ! HL 1 .B; OBIG /: (8.11)
There is a different way to describe the G-bundle structure for a principal G-bundle.
Proposition 8.20. Let G be a Lie group. The following data are equivalent:
(a) A principal G-bundle .W X ! B; ŒUi ; hi i 2I / over B.

(b) A morphism of premanifolds W X ! B and a right action of G on X by a morphism
of premanifolds such that there exists an open covering .Ui /i 2I of B and for all i 2 I
an isomorphism of Ui -premanifolds hi W 1 .Ui / ! Ui G such that
h1 0 1 0
i .b; gg / D hi .b; g/g ; b 2 Ui ; g; g 0 2 G: (8.12)
Moreover, if these data are given, then the right action of G on X preserves the fibers
X.b/ WD 1 .b/ and the induced action of G on X.b/ is simply transitive for all b 2 B.
Proof. Let there be given a datum as in (b). Then W X ! B is a fiber bundle with basis
B and typical fiber G. Let us show the last assertion. The G-right action preserves the
fibers because of (8.12) and because the hi preserve fibers. For x 2 X and g 0 2 G we
choose i 2 I with b WD .x/ 2 Ui . Let hi .x/ D .b; g/. Then xg 0 D h1 0
i .b; gg / by
(8.12) and hence .xg 0/ D b. Moreover, the hi induce for all b an isomorphism
hi;b W X.b/ ! fbg G D G
and (8.12) shows that hi;b is G-equivariant if we endow the right-hand side with the (sim-
ply transitive) action of G by right multiplication. Hence the right action of G on X.b/ is
simply transitive.
Let us now construct a G-bundle structure from a datum as in (b). For i; j 2 I ,
1
b 2 Ui \ Uj the map hi;b ı hj;b W G ! G is a G-equivariant isomorphism. Hence it is
1
necessarily of the from g 7! gi;j;b g for some gi;j;b 2 G. Therefore b 7! hi;b ı hj;b defines
a morphism Ui \ Uj ! G, b 7! gi;j;b , for i; j 2 I , and .Ui ; hi /i is a G-bundle atlas for
. A different choice of .Ui ; hi /i as in (b) yields equivalent G-bundle atlases. We obtain
a well-defined G-bundle structure on .
Conversely, let .W X ! B; ŒUi ; hi i 2I / be a principal G-bundle. For b 2 B choose
i 2 I with b 2 Ui . Then hi induces an isomorphism of the fibers hi;b W X.b/ ! fbg G D
G, which we use to define a right action on X.b/ by transport of structure, i.e., the right
action is given by
.x; g/ 7! h1
i;b .hi;b .x/g/; x 2 X.b/; g 2 G:
For every j 2 I with b 2 Uj there exists gi;j;b 2 G such that hi;b .x/ D gi;j;b hj;b .x/ and
hence we obtain the same right G-action on X.b/ if we use hj for the definition. Therefore
the G-action is independent of the choice of i and we obtain a right G-action on X. The
same argument shows that passing to an equivalent OBIG -twisting atlas does not change
the G-action on X. To see that this action is a morphism of premanifolds we may work
locally on B. Then over Ui this follows because hi is an isomorphism of premanifolds.

Example 8.21. Let G be a Lie group, let H be a Lie subgroup of G, let G=H be the
quotient (Corollary 6.23), and let W G ! G=H be the projection. Then the action of H
on G by right multiplication makes into a principal H -bundle over G=H .
Indeed, as is a surjective submersion, it looks locally as a projection (Corollary 5.23)
and we find an open covering .Ui /i of G=H and for all i sections si W G=H ! G of .
Then
hi W 1 .Ui / ! Ui H; g 7! .g/; si ..g//1 g
is an Ui -isomorphism with inverse morphism .b; h/ 7! si .b/h. Clearly, h1
i satisfies
(8.12).
Remark 8.22 (Bundles associated with principal bundles). Let G be a Lie group and
let F be a G-premanifold.
Suppose that G acts faithfully on F . Then we obtain from (8.11) and (8.8) isomor-
phisms of pointed sets
PrincG .B/ ! HL 1 .B; OBIG / ! FibF;G .B/: (8.13)
This isomorphism can be described by the following construction of fiber bundles from
a principal bundle. Let W X ! B be a principal G-bundle over B whose G-bundle
structure is given by a right action X G ! X (Proposition 8.20). Then
.x; f / g WD .xg; g 1 f /
defines a right G-action on X F by a morphism of premanifolds. Let X G F be the

set of G-orbits in X F endowed with the quotient topology. Then
Q X G F ! B;
W Œx; f 7! .x/
is a well-defined continuous map.

164 8 Bundles
Let us define the structure of a fiber bundle with typical fiber F and structure group G
Q Let .Ui ; hi W 1 .Ui / ! Ui G/ be a G-bundle atlas for the principal bundle and
on .
let .gij / be the attached cocycle. Consider the continuous maps
kQi W Ui F ! Q 1 .Ui /; .b; f / 7! h1

i .b; 1/; f :
By definition of the quotient topology, Q 1 .Ui / is open in X. Using that G acts simply
transitively on the fibers of and that the hi are G-equivariant, one checks that kQi is
bijective. We have

kQj1 ı kQi .b; f / D .b; gj i f /; b 2 Ui \ Uj ; f 2 F:
Therefore kQj1 ı kQi is over Ui \ Uj an isomorphism of premanifolds. In particular, kQi is

a local isomorphism and hence an isomorphism because kQi is bijective. As kQj1 ı kQi is an
isomorphism of premanifolds over Ui \ Uj , there is a unique structure of a premanifold
on X G F such that kQi is an isomorphism onto the open submanifold Q 1 .Ui /. Hence
Q X ! B is a fiber bundle over B with typical fiber F . Setting hQ i WD kQi1 , .Ui ; hQ i / is a
W
G-bundle atlas for , Q which is given by the cocycle .gij /i;j .
Note that for the construction of X G F it was not necessary to assume that G acts
faithfully on F (Remark 8.13). For instance we could have taken G acting trivially on
the one point manifold F D f g. Then X G F D X=G is the set of G-orbits in X.
Moreover, WQ X=G ! B is a fiber bundle with typical fiber f g over B, hence Q is an
isomorphism of premanifolds.
The construction above yields in particular the following functoriality in G for G-
principal bundles.
Remark 8.23 (Functoriality for principal bundles). Let 'W G ! H be a homomor-

phism of Lie groups. Then G acts on H from the left by .g; h/ 7! '.g/h. This action is
faithful if and only if ' is injective. Let D .W X ! B; X G ! X/ be a principal G-
bundle over B and set '. / WD X G H . By Remark 8.22 this is a fiber bundle over B with
typical fiber H . Then H acts on it from the right by the second factor Œx; h h0 7! Œx; hh0
making '. / into a principal H -bundle. We obtain a commutative diagram
PrincG .B/ HL 1 .B; OBIG /
7!'. / HL 1 .'B /
PrincH .B/ HL 1 .B; OBIH /:

8.2 Vector Bundles 165
8.2 Vector Bundles
Among the most important examples of fiber bundles are vector bundles. We continue to
denote by B a premanifold.
Definition 8.24 (Vector bundle). Let V be a finite-dimensional K-vector space. The Lie
group GL.V / acts faithfully on V by left multiplication. A fiber bundle with basis B,
with typical fiber V , and structure group GL.V / is called a vector bundle with typical fiber
V over B. We say that has rank n if dimK .V / D n. A vector bundle of rank 1 is called
a line bundle.
If B is a real C ˛ -premanifold we may also choose for V to be a finite-dimensional

complex vector space V and consider GLC .V / as a real Lie group. In this case we speak
of a complex C ˛ -vector bundle on B of complex rank dimC .V /.
If B is a complex premanifold, we may also consider a real vector bundle in the cat-
b on the underlying real C ˛ premanifold of B. In this
egory of C ˛ -premanifolds (˛ 2 N)
˛
case we speak of a C -vector bundle on B.
In the sequel, if we speak simply of a vector bundle over a premanifold B, we will
either mean
(a) vector bundles with a real vector space as fiber over a real C ˛ -premanifold B (also
called real C ˛ -vector bundle), or
(b) vector bundles with a complex vector space as fiber over a complex premanifold B
(also called holomorphic vector bundles).
Remark 8.25. By (8.8) we have a bijection

( )
isomorphism classes of vector bundles
! HL 1 .B; GLn;B / (8.14)
of rank n with basis B
and in particular
( )
isomorphism classes of line bundles
Pic.B/ WD ! HL 1 B; OB : (8.15)
with basis B
As OB D GL1;B is a sheaf of abelian groups, Pic.B/ has the structure of an abelian group.
It is called the Picard group of B.
Example 8.26. The map eW R f˙1g ! R , .x; "/ 7! " exp.x/ is an isomorphism of
real Lie groups. Its inverse is given by z 7! .log.jzj/; sgn.z//. Let B be a real preman-
ifold. Applying the functor G 7! OBIG to the isomorphism e (Remark 8.8) we obtain
an isomorphism OB .Z=2Z/B ! OB of sheaves of abelian groups on B (identifying
166 8 Bundles
the groups f˙1g and Z=2Z). Therefore by (8.15) we obtain an isomorphism of abelian
groups
Pic.B/ ! H 1 .B; OB / H 1 .B; .Z=2Z/B /: (8.16)
We will see that for real C ˛ -manifolds B one has H 1 .B; OB / D 0: For ˛ 1 we will
prove this in Corollary 9.15; for real analytic manifolds see Remark 10.21. Therefore we
see
Pic.B/ Š H 1 .B; .Z=2Z/B / (8.17)
for real manifolds B.
The vector bundle structure can also be described as follows.
Proposition 8.27. Let V be a finite-dimensional K-vector space. The following data are
equivalent:
(a) A vector bundle .W E ! B; ŒUi ; hi i 2I / with typical fiber V over B.

(b) A morphism W E ! B of premanifolds and for each b 2 B the structure of a K-
vector space on E.b/ WD 1 .b/ such that there exists an open covering .Ui /i of B
and isomorphisms of Ui -premanifolds hi W 1 .Ui / ! Ui V such that
hi;b WD hi jE.b/ W E.b/ ! fbg V D V
is an isomorphism of K-vector spaces.
In the sequel we will denote vector bundles usually by .E; / and it is understood that
they are endowed with both of the above structures. In particular the fibers E.b/ will
always be endowed with the structure of a K-vector space.
Proof. The proof is similar to the proof of Proposition 8.20. Given a datum as in (b),
is a fiber bundle over B with typical fiber F . Moreover, we find that for i; j 2 I the map
1
b 7! hi;b ı hj;b is a morphism Ui \ Uj ! GL.V /. Hence .Ui ; hi /i is a OBIGL.V / -bundle
atlas for . Any two choices of .Ui ; hi /i 2I as in (b) define equivalent OBIGL.V / -bundle
atlases for and hence a well-defined OBIGL.V / -bundle structure on .
Conversely, let .Ui ; hi /i 2I be a OBIGL.V / -bundle atlas for . For b 2 B choose i 2 I
with b 2 Ui . Then hi induces a bijection hi;b W E.b/ ! fbg V D V , which we use to
define a structure of K-vector space on E.b/ by transport of structure. For any i 0 2 I
with b 2 Ui 0 , hi;b and hi 0 ;b differ only by a linear automorphism of V and hence define the
same vector space structure on Eb . Therefore the vector space structure is independent of
the choice of i. The same argument shows that passing to an equivalent OBIGL.V / -twisting
atlas does not change the vector space structure on E.b/.
8.3 OX -Modules 167
Although for fiber bundles (over a fixed basis) we defined only the notion of an isomor-
phism, the alternative description in Proposition 8.27 allows us to define a more general
notion of morphism for vector bundles. First let us agree that we mean by a vector bundle
`
over B (without specifying a typical fiber or a rank) a decomposition B D i Bi of B
into open and closed subpremanifolds and for all i a finite-dimensional K-vector space Vi
and a vector bundle over Bi with typical fiber Vi . Then for a vector bundle .E; / over B
its rank is a locally constant function B ! N0 , b 7! rkb .E/.
Definition 8.28. Let .E; / and .E 0 ; 0 / be vector bundles over B (possibly of differ-
ent rank and in particular with different typical fiber). A morphism of vector bundles
.E; / ! .E 0 ; 0 / is a B-morphism ˚W E ! E 0 of premanifolds such that for all p 2 B
the maps ˚.p/ WD ˚ jE.b/ W E.b/ ! E 0 .b/ are K-linear.
The composition of two morphisms of vector bundles is again a morphism. We obtain

the category of vector bundles over B, which we denote by .Vec.B//.
8.3 OX -Modules
Vector bundles over a premanifold B can also be described as finite locally free modules
over the structure sheaf OB . Let us first explain what we mean by this. In this section let
.X; OX / be an arbitrary ringed space.
The Category of OX -Modules
Definition 8.29. An OX -(pre-)module is a (pre-)sheaf F on X together with two mor-

phisms (addition and scalar multiplication) of sheaves
F F ! F; .s; s 0 / 7! s C s 0 for s; s 0 2 F.U /, U X open;

OX F ! F; .a; s/ 7! as for a 2 OX .U /, s 2 F.U /, U X open
such that addition and scalar multiplication by OX .U / define on F.U / the structure of an
OX .U /-module.
If F1 and F2 are OX -premodules, a morphism of sheaves wW F1 ! F2 is called a
homomorphism of OX -premodules, if for each open subset U X the map F1 .U / !
F2 .U / is an OX .U /-module homomorphism, i.e.,
wU .s C s 0 / D wU .s/ C wU .s 0 /;
wU .as/ D awU .s/
for all s; s 0 2 F.U / and a 2 OX .U /. A homomorphism of OX -modules is a homomor-

phism of OX -premodules.
168 8 Bundles
The composition of two homomorphisms of OX -modules is again a homomorphism of

OX -modules. We obtain the category of OX -modules, which we denote by (OX -Mod). The
OX -module F such that F.U / D f0g for all open sets U X is called the zero module
and simply denoted by 0.
Example 8.30.
1. Let X be a topological space and let ZX be the constant sheaf of rings on X with value
Z. As Z-modules are simply abelian groups, a ZX -module is simply a sheaf of abelian
groups on X.
2. Let A be a commutative ring. Let X be a space that consists of a single point and let
OX be the sheaf of rings with OX .X/ D A. Then an OX -module F is just an A-module
M (by attaching M D F.X/ to F).
Remark 8.31. Let F be an OX -module and x 2 X. The OX .U /-module structures on

F.U /, where U is an open neighborhood of x, induce on the stalk Fx an OX;x -module
structure. If OX;x is a local ring with maximal ideal mx and residue field .x/ WD
OX;x =mx , then we call the .x/-vector space F.x/ WD Fx =mx Fx the fiber of F in x.
If wW F ! F0 is a homomorphism of OX -modules, the induced map on stalks wx W Fx !
0
Fx is a homomorphism of OX;x -modules. If OX;x is local, it also induces a .x/-linear map
w.x/W F.x/ ! F0 .x/ on fibers.
Remark 8.32. As the sheafification commutes with formation of finite products, addition
and scalar multiplication on an OX -premodule F induce by functoriality an addition and
a scalar multiplication on the sheafification Fa of F. Hence Fa is an OX -module.
Definition 8.33. An OX -submodule of F is an OX -module G such that G.U / is a subset

of F.U / for all open sets U X and such that the inclusions U W G.U / ,! F.U / form
a homomorphism W G ! F of OX -modules.
The OX -submodules of OX are called ideals of OX .
Example 8.34. Let M be a premanifold and let S M be a subset. Then the sheaf IS
of functions vanishing on S (Remark 5.31) is an ideal of OM .
Remark 8.35. Let F and F0 be OX -modules. If w1 and w2 are two homomorphisms

F ! F0 , we can define their sum in the obvious way:
.w1 C w2 /U WD w1;U C w2;U W F.U / ! F0 .U /:
Let a 2 .X; OX / and wW F ! F0 be a homomorphism. Then we define a homo-

morphism awW F ! F0 by .aw/U WD .ajU /wU . In this way we endow the set of
homomorphisms F ! F0 of OX -modules with the structure of a .X; OX /-module. We
8.3 OX -Modules 169
denote this .X; OX /-module by HomOX .F; F0 /. The composition of homomorphisms is

.X; OX /-bilinear. In particular, the additive group structures on the sets HomOX .F; F0 /
make (OX -Mod) into a preadditive category (Appendix Definition 15.21).
Remark 8.36 (Construction with OX -modules). As a general principle, functorial con-

struction for A-modules, where A is a commutative ring, can be generalized to construc-
tions for OX -modules. Given a functor on usual A-modules, we apply it to all sets of
sections F.U /. We obtain an OX -premodule. If this is not already a sheaf, we sheafify to
obtain an OX -module.
For instance let us consider limits and colimits of OM -modules. Let I be a small
category and let FW I ! (OX -Mod), i 7! Fi be an I -diagram of OX -modules.
Remark 8.37 (Limits of OX -modules). The explicit description of limits of R-modules

shows that the OX -premodule U 7! limI Fi .U / (taking the limit in the category of
OX .U /-modules) is in fact an OX -module denoted by limI Fi and called the limit of the
diagram F. As for sheaves (Remark 3.61) one obtains for all x 2 X a functorial morphism
.lim Fi /x ! lim.Fi /x ; (8.18)

I I
which is an isomorphism if I is finite. This is a homomorphism of OX;x -modules.

Taking for I a category whose only morphisms are the identities, we obtain as a special
Q
case the product of the family .Fi /i 2I with I WD Ob.I /. It is denoted by i 2I Fi . The
Q
universal property for products implies that composition with projections prj W i Fi !
Fj yields for every OX -module G a .X; OX /-linear bijection
!
Y Y
Hom(OX -Mod) G; Fi ! Hom(OX -Mod) .G; Fi /: (8.19)
i 2I i 2I
Another special case is the kernel of a homomorphism wW F ! G of OX -modules as the

limit of the diagram
w
F G: (8.20)
0
It is denoted by Ker.w/ and it is the sheaf U 7! Ker.wU W F.U / ! G.U //. By (8.18) we
have Ker.w/x D Ker.wx /. In particular, w is an injective homomorphism of sheaves if
and only if Ker.w/ D 0.
Remark 8.38 (Colimits of OX -modules). Consider the OX -premodule U 7! colimI

Fi .U /, where we take the colimit in the category of OX .U /-modules. It is in general
not a sheaf. Its sheafification is an OX -module denoted by colimI Fi and called the colimit
170 8 Bundles
of the diagram F. The same argument as for sheaves (Remark 3.63) shows that one obtains
for all x 2 X a functorial isomorphism of OX;x -modules
.colim Fi /x ! colim.Fi /x : (8.21)

I I
Again taking for I a category whose only morphisms are the identities, we obtain as
a special case the direct sum of the family .Fi /i 2I with I WD Ob.I /. It is denoted by
L
i 2I Fi . If Fi D F for all i, we also write F . The universal property for direct sums
.I /
of modules (Appendix Sect. (14.3)) and the universal property of the sheafification yield
for every OX -module G an isomorphism of .X; OX /-modules
!
M Y
Hom(OX -Mod) Fi ; G ! Hom(OX -Mod) .Fi ; G/: (8.22)
i 2I i 2I
L Q
If I is finite, then i 2I Fi D i 2I Fi .
Another special case is the cokernel of a homomorphism wW F ! G of OX -modules
as the colimit of the diagram (8.20). It is denoted by Coker.w/ and it is the sheaf
associated with the presheaf U 7! Coker.wU W F.U / ! G.U //. By (8.21) we have
Coker.w/x D Coker.wx /. In particular, w is a surjective homomorphism of sheaves if
and only if Coker.w/ D 0.
Let F be an OX -submodule of G and let iW F ! G be the inclusion. Then
G=F WD Coker.i/
is called the quotient of G by F. It is the OX -module associated with the OX -premodule

U 7! G.U /=F.U /.
For a homomorphism wW F ! G of OX -modules we can also define its image
w.F/ WD Im.w/ WD Ker.G ! Coker.w//
and its coimage

Coim.w/ WD Coker.Ker.w/ ! F/:
It is the sheaf associated with the presheaf U 7! Im.wU W F.U / ! G.U // respective to
U 7! Coim.wU /. As kernel and cokernel are compatible with the formation of stalks, we
also have Im.w/x D Im.wx / and Coim.w/x D Coim.wx /. As wx induces for all x 2 X
an isomorphism Coim.wx / ! Im.wx / of OX;x -modules, we deduce that w induces an
isomorphism Coim.w/ ! Im.w/. In particular we see:
Proposition 8.39. The category of OX -modules is an abelian category.
Therefore, we also have the notion of exact sequences of OX -modules: A sequence

wi 1 wi
! Fi 1 ! Fi ! Fi C1 ! : : : of homomorphisms of OX -modules is called exact if
Ker.wi / D Im.wi 1 / for all i.
8.3 OX -Modules 171
Finite Locally Free OX -modules

As for modules over rings, every family of global sections of an OX -module F yields
a homomorphism of a free OX -module to F. More precisely:
Remark and Definition 8.40 (Bases of OX -modules). Let F be an OX -module and let
s WD .si /i 2I be a family of sections si 2 F.X/. Then one obtains a homomorphism of
OX -modules X
.I /
ws W OX ! F; .fi /i 2I 7! fi .si jU /; (8.23)
i 2I
for U X open and fi 2 OX .U /. As for modules over rings, s is called a generating
system (respectively a basis) if ws is surjective (respectively bijective).
Definition 8.41. An OX -module F is called finite locally free if there exists an open cov-
ering .Ui /i of X such that for all i the OUi -module FjUi is isomorphic to OnUi for some
n 2 N0 (depending on i).
The full subcategory of (OX -Mod) whose objects are the finite locally free OX -modules
is denoted by .FLF.X//.
If F is a finite locally free OX -module, its stalk Fx is a finitely generated free OX;x -
module for all x 2 X. We set rkx .F/ WD rkOX;x .Fx / and obtain a locally constant function
X ! N0 ; x 7! rkx .F/;
called the rank of F.
Next we define the OX -module of all homomorphisms between two OX -modules. In
particular we obtain the dual of an OX -module. The general principle is again the one
sketched in Remark 8.36.
Definition and Remark 8.42. Let .X; OX / be a ringed space and let F and G be two
OX -modules. The presheaf
U 7! HomOX jU .FjU ; GjU /
with the obvious restriction maps is a sheaf. The right-hand side is a .U; OX /-module.
Therefore this sheaf has the structure of an OX -module, and we denote this OX -module by
HomOX .F; G/. Moreover:
1. By (8.19) and (8.22) one has for every family .Fi /i 2I of OX -modules and every OX -
module G isomorphisms of OX -modules
!
Y Y
HomOX G; Fi ! HomOX .G; Fi /;
i 2I i 2I
! (8.24)
M Y
HomOX Fi ; G ! HomOX .Fi ; G/:
i 2I i 2I
172 8 Bundles
2. As for modules (Appendix Example 14.7), evaluation on 1 yields an identification
HomOX .OX ; F/ D F: (8.25)
3. If F and G are finite locally free, then HomOX .F; G/ is finite locally free and
rkx .HomOX .F; G// D rkx .F/ rkx .G/ for all x 2 X.
Indeed, let .Ui /i be an open covering such that FjUi Š OnUi and GjUi Š Om
Ui for all i.
Then
HomOX .F; G/jUi D HomOUi .FjUi ; GjUi /

Š HomOUi .OnUi ; Om
Ui /
(8.24)
D HomOUi .OUi ; OUi /nm
(8.25)
D Onm
Ui :
4. For all x 2 X, sending a homomorphism of OX -modules to its germ in x yields a ho-

momorphism of OX;x -modules
HomOX .F; G/x ! HomOX;x .Fx ; Gx /: (8.26)
This is an isomorphism if F and G are finite locally free (see also Problem 8.5). Indeed,
by replacing X by a sufficiently small neighborhood of x we may assume that F and
G are free. Then by (8.24) we may assume that F D G D OX . Then this is clear by
(8.25).
The surjectivity of (8.26) means that every OX;x -linear map w0 W Fx ! Gx is induced
by a homomorphism of OU -modules wU W FjU ! GjU for some open neighborhood U
of x. The injectivity of (8.26) means that any two such homomorphisms wU and wU 0
become equal after restricting to an open neighborhood of x contained in U \ U 0 .
Definition 8.43 (Dual OX -module). Let .X; OX / be a ringed space and let F be an OX -
module. The OX -module
F_ WD HomOX .F; OX / (8.27)
is called the dual OX -module of F.
Remark 8.44. If F D OXn , then F_ Š OXn by Remark 8.42 2 and 1. Hence, if F is a finite
locally free OX -module, then its dual F_ is again locally free, and rkx .F_ / D rkx .F/ for
all x 2 X.
8.4 Vector Bundles and Finite Locally Free Modules 173
8.4 Vector Bundles and Finite Locally Free Modules
Now let M be a premanifold. We will show that the category of vector bundles over M
and the category of finite locally free OM -modules are equivalent. We start by defining
a functor
ModW .Vec.M // ! .FLF.M //: (8.28)
Let W E ! M be a vector bundle. In particular, all fibers E.p/ WD 1 .p/ for p 2 M

are endowed with the structure of a K-vector space. We attach an OM -module E WD
Mod.E; / to .E; / as follows. For U M open set
E.U / WD f sW U ! E morphism of premanifold I ı s D idU g :
For s; s 0 2 E.U / define s C s 0 2 E.U / by .s C s 0 /.p/ WD s.p/ C s 0 .p/ (addition inside

the K-vector space E.p/). For f 2 OM .U /, s 2 E.U / define f s 2 E.U / by .f s/.p/ WD
f .p/s.p/ (scalar multiplication inside E.p/). This makes E into an OM -module.
Let ˚W .E; / ! .E 0 ; 0 / be a morphism of vector bundles. Set E WD Mod.E; /,
E0 WD Mod.E 0 ; 0 /. For U M open define
'U W E.U / ! E0 .U /; s 7! ˚ j 1 .U / ı s:
As ˚ is linear on fibers, 'U is a homomorphism of OM .U /-modules. Therefore

Mod.˚/ WD .'U /U is a homomorphism of OM -modules. We obtain a functor Mod
from the category of vector bundles over M to the category of OM -modules.
Let E D Mod.E; / for a vector bundle .E; /. Let us show that E is finite locally
free. Let W M be open such that 1 .W / D W Kn , for some n 2 N. For U W
one then has
E.U / Š f sW U ! U Kn morphism I s.p/ D .p; v/ for some v 2 Kn g

D Hom(PMfd) .U; Kn /
D OM .U /n
and hence EjW Š .OM jW /n . This concludes the construction of the functor (8.28).
Note that we have rkp .E/ D rkp .E/ for p 2 M .
Proposition 8.45. The functor Mod from the category of vector bundles over M to the
category of finite locally free OM -modules is an equivalence of categories.
Proof. Construction of a functor GW .FLF.M // ! .Vec.M //. Let E be a finite locally

free OM -module. Let p 2 M . Recall that OM;p is a local K-algebra with maximal ideal
mp and residue class field OM;p =mp D K. The stalk Ep is a free OM;p -module of rank
rkp .E/. Let E.p/ WD Ep =mp Ep be the fiber of E. It is a finite-dimensional K-vector space
174 8 Bundles
of dimension rkp .E/. For p 2 U M open and s 2 E.U / we denote by s.p/ the image
of sp 2 Ep in E.p/. We define the set
a
E WD E.p/
p2M
and the map W E ! M such that 1 .p/ D E.p/ for p 2 M .

Let U M be open such that EjU Š .OM jU /n . In other words, there exists a basis
e D .e1 ; : : : ; en / of EjU (Remark 8.40). In particular .e1 .p/; : : : ; en .p// is a K-basis of
E.p/. Hence we get a bijection
a X
n
˚U;e W U Kn ! 1 .U / D E.p/; .p; / 7! i ei .p/:
p2U i D1
This bijection endows 1 .U / with the structure of a premanifold. For a different choice
UQ and eQ the “change of charts” ˚U;e
1
ı ˚UQ ;eQ is an automorphism of .U \ UQ / Kn given
by a matrix in GLn .OM .U \ UQ //. This makes .E; / into a vector bundle over M .
A morphism wW E ! E0 between finite locally free OM -modules induces K-linear
maps w.p/W E.p/ ! E0 .p/ for all p 2 M (Remark 8.31) and hence a map E ! E 0 on
the attached vector bundles. Covering M by open subsets U as above, one sees that this
map is a morphism of vector bundles. Hence we obtain a functor G from the category of
finite locally free OM -modules to the category of vector bundles over M .
G and Mod are quasi-inverse. Let .E; / be a vector bundle over M and let E D
Mod.E; /. For p 2 M let s 2 Ep , i.e., s is the equivalence class of a pair .U; sQ /, where
p 2 U M and sQ W U ! E is a section of . The homomorphism of OM;p -modules
Ep ! E.p/; s 7! s.p/
is surjective and its kernel is mp Ep . Hence we obtain an isomorphism of K-vector spaces

E.p/ Š E.p/ D .G ı Mod/.E; /.p/ and hence a bijective map E ! .G ı Mod/.E; /.
Looking at charts ˚U;e as above one sees that this is locally an isomorphism of vector
bundles. Hence it is an isomorphism of vector bundles. It is clear that this isomorphism is
functorial in .E; /.
Conversely, let E be a finite locally free OM -module and let .E; / D G.E/. Let
U M be open, s 2 E.U /. Then s.p/ 2 E.p/ D E.p/ and we obtain a map sW U ! E,
p 7! s.p/ such that ı s D idU . Looking at charts, it is easy to see that s is a morphism
of manifolds and hence an element of Mod.E; /.U /. We obtain maps E.U / ! .Mod ı
G/.E/.U / and it is immediate that these maps define an isomorphism of OM -modules,
which is functorial in E.
Let E be a finite locally free OM -module and let .E; / be the corresponding vector
bundle. Then the proof shows that the K-vector spaces E.p/ and E=mp Ep are identified.
8.4 Vector Bundles and Finite Locally Free Modules 175
Definition and Remark 8.46 (Dual and direct sum of vector bundles). Let M be a pre-
manifold, let .E; / and .E 0 ; 0 / be vector bundles over M and let E and E0 , respectively,
be the corresponding finite locally free OM -modules.
1. The vector bundle .E _ ; _ / corresponding to the dual OM -module E_ is called the

dual of .E; /. For each point p 2 M the fiber of E _ in p is given by the dual space
of E.p/:
(8.26)
E _ .p/ D Ep_ =mp Ep_ D .Ep =mp Ep /_ D E.p/_ :
2. The vector bundle corresponding to E ˚ E0 is called the direct sum of .E; / and
.E 0 ; 0 /. It is denoted by .E ˚ E 0 ; ˚ 0 /. For p 2 M one has .E ˚ E 0 /.p/ D
E.p/ ˚ E 0 .p/.
Definition 8.47. Let .X; OX / be a ringed space, let E be an OX -module, and let F be an
OX -submodule of E. Then F is called a direct summand of E if there exists a submodule
G of E such that F ˚ G D E. The submodule F is called a local direct summand if there
exists an open covering .Ui /i of X such that FjUi is a direct summand of EjUi for all i 2 I .
Remark 8.48. Let F be a submodule of an OX -module E. Let iW F ! E be the in-

clusion and let pW E ! E=F be the projection. As for modules over a ring (Appendix
Remark 14.5) one sees that the following assertions are equivalent:
(i) The submodule F is a direct summand of E.

(ii) There exists a homomorphism of OX -modules rW E ! F such that r ı i D idE .
(iii) There exists a homomorphism of OX -modules sW E=F ! E such that p ı s D idE=F .
Local direct summands of a finite locally free module E correspond to subbundles of

the vector bundle attached to E:
Proposition and Definition 8.49. Let M be a premanifold, E and F finite locally free
OM -modules and let E and F be the corresponding vector bundles over M . Let wW E ! F
be a homomorphism of OM -modules and let $ W E ! F be the corresponding morphism
of vector bundles.
1. The map $ is surjective if and only if the morphism of sheaves w is surjective. In this
case we say that F is a quotient bundle of E.
2. The map $ is injective if and only if the morphism of sheaves w is injective and w.E/
is a local direct summand of F. In this case we say that E is a subbundle of F .
176 8 Bundles
Proof. The map $ is surjective (respectively injective) if and only if for all p 2 M the
induced linear map on the fiber $ .p/W E.p/ ! F .p/ is surjective (respectively injective).
We show (1). Let mp be the maximal ideal of OM;p and let w.p/ be the K-linear map
E.p/ WD Ep =mp Ep ! F.p/ WD Fp =mp Fp induced by w. In the proof of Proposition 8.45
we have seen that E.p/ D E.p/, F .p/ D F.p/ and w.p/ D $ .p/. The stalk Ep is
a finitely generated free OM;p -module. Hence the first assertion follows immediately from
Nakayama’s lemma (Appendix 14.38).
Let us show (2). Suppose that w is injective and that w.E/ is a local direct summand
of F. Let p 2 M . To see that $ .p/ is injective we may pass to a sufficiently small neigh-
borhood of p and can assume that there exists a homomorphism of OX -modules rW E ! F
such that r ıw D idE (Remark 8.48). Then r induces a K-linear map r.p/W E.p/ ! F .p/
with r.p/ ı w.p/ D id. As w.p/ D $ .p/, we see that $ .p/ is injective.
Conversely, assume that $ .p/ D w.p/ is injective for all p 2 M . Let rN0 W F.p/ !
E.p/ be a K-linear map with rN0 ı w.p/ D idE.p/ . Let .f1 ; : : : ; fn / be an OM;p -basis of
rN0
Fp . For all i let eNi be the image of fi under Fp ! F.p/ ! E.p/ and let ei 2 Ep
be an element whose image in E.p/ is eNi . Sending fi to ei defines an OM;p -linear map
r 0 W Fp ! Ep . Then r 0 ı wp is an endomorphism of the finitely generated free OM;p -module
Ep , which is modulo mp the identity. Hence it is bijective (Appendix Proposition 14.39).
Hence if we set r0 WD .r 0 ı wp /1 ı r 0 we have r0 ı wp D idEp .
As (8.26) is surjective for homomorphisms F ! E, we find an open neighborhood U
of p and a homomorphism rW FjU ! EjU of OU -modules such that rp D r0 . As (8.26) is
injective for homomorphisms E ! E, we have r ıw jU D idEjU after possibly shrinking U .
As p was arbitrary, this shows that w is injective and that w.E/ is a local direct summand
of F by Remark 8.48.
The proof shows that if $ .p/ is injective for a single point p, then there exists an open
neighborhood U of p such that E jU is a subbundle of F jU .
The local-global principle explained in Chap. 7 yields a cohomological criterion for
a local direct summand to be a direct summand and hence for subbundles to be global
direct summands:
Remark 8.50 (Splitting of submodules). Let .X; OX / be a ringed space. Let E be a local
direct summand of an OX -module F. Let pW F ! F=E be the projection and let S be
the sheaf of splittings of p, i.e., for U X open, S.U / is the set of homomorphisms
of OU -modules rW .F=E/jU ! FjU with p jU ı r D id. For any r; r 0 2 S.U / one has
p jU ı .r r 0 / D 0, i.e., r r 0 takes values in E. Hence the sheaf
A WD HomOM .F=E; E/;
considered as a sheaf of abelian groups, acts simply transitively on S by .w; s/ 7! w C s,

w 2 A.U /, s 2 S.U /, U X open. This makes S into a pseudo-A-torsor. By hypothesis
there exists on open covering .Ui /i with S.Ui / ¤ ; for all i. Hence S is an A-torsor. Its
8.5 Tangent Bundle 177
class
.E/ 2 H 1 .X; HomOM .F=E; E//
is trivial if and only if S.X/ ¤ ;, i.e., if and only if E is a (global) direct summand of F.
Definition and Remark 8.51 (Split subbundles). We keep the notation of Defini-
tion 8.49. We call the vector bundle E a split subbundle of F if w is injective and w.E/ is
a direct summand of F, i.e., if F is the direct sum of the image of $ and a subbundle of
F.
If E is a subbundle of F , then Remark 8.50 shows that there exists a class .E/ 2
H 1 .M; HomOM .F=E; E// that vanishes if and only if E is a split subbundle of F .
8.5 Tangent Bundle
In this and the following section we will assume that we are either in the case of real C 1 -
(pre)manifolds (real smooth case) or in the case of real C ! -(pre)manifolds with K D R
(real analytic case) or in the case of complex (pre)manifolds with K D C (complex case).
Definition and Remark 8.52. Let M be a premanifold. We define the tangent bundle
TM as a set a
TM WD Tp .M /:
p2M
Let D M W TM ! M be the map such that 1 .p/ D Tp .M / for all p.

Definition of an atlas of TM : Let xW U ! UQ Km be a chart of M . For p 2 U we
obtain an isomorphism Tp .x/W Tp .M / ! Km . Let V WD 1 .U / TM and define as
a chart for TM
yW V ! VQ WD UQ Km K2m ; Tp .M / 3 7! .x.p/; Tp .x/. //:
If we define for a second chart .U 0 ; x 0 / of M a chart .V 0 ; y 0 / of TM in the same way, the

change of charts y 0 ı y 1 is given by
y.V \ V 0 / D x.U \ U 0 / Km ! x 0 .U \ U 0 / Km D y 0 .V \ V 0 /;
(*)
Q v/ 7! .x 0 ı x 1 /.p/;
.p; Q Jx 0 ıx 1 .p/.v/
Q ;
which is again an isomorphism of premanifolds. Hence if .U; x/ runs through an atlas of

M the .V; y/ form an atlas of TM . We obtain on TM the structure of a premanifold1 .
1
If M was a C ˛ -premanifold with ˛ < 1, then the second component of (*) would be only C ˛1
and hence TM would be only C ˛1 -premanifold. Taking this added complication into account, most
results of Sect. 8.5 also hold for real C ˛ -premanifolds with ˛ < 1.
178 8 Bundles
Clearly, .TM ; / is a vector bundle over M and we have rkp .TM / D dimp .M /. We
denote by TM the finite locally free OM -module corresponding to TM (Proposition 8.45).
If M is a manifold, then TM is a manifold (Proposition 8.16).
Definition and Remark 8.53. Let F W M ! N be a morphism of premanifolds. Define
TF W TM ! TN ; Tp .M / 3 7! Tp .F /. / 2 TF .p/ .N /:
Then the diagram

TF
TM TN
M N
F
M N
is commutative, and TF is K-linear on fibers. We claim that TF is a morphism of premani-
folds. Indeed, this can be checked locally on M and N and we may assume that M Km
and N Kn are open. Then TF is given by
TF W TM D M Km ! N Kn D TN ; .p; x/ 7! .F .p/; DF .p//;
which is a morphism of premanifolds2 .

For morphisms F W M ! N and GW N ! P the chain rule takes the following elegant
form:
TGıF D TG ı TF : (8.29)
Sections of the OM -module TM are called vector fields:
Definition 8.54. Let M be a premanifold. For U M open, elements of
TM .U / D f sW U ! TM I ı s D idU g
are called vector fields over U . For a vector field X we sometimes write Xp instead of
X.p/ for p 2 U .
Example 8.55. Let V be a finite-dimensional K-vector space and let M V be a sub-

manifold. For every p 2 M we have Tp .M / Tp .V / D V and hence we may view the
tangent bundle TM as the submanifold of V V
˚
TM D .p; v/ 2 M V I v 2 Tp .M / :
Hence a vector field X is the same as a morphism F W M ! V of manifolds with F .p/ 2

Tp .M /. The correspondence is given by X.p/ D .p; F .p// for all p 2 M .
2
Again, TF would be only a morphism of C ˛1 -premanifolds for ˛ < 1.
8.5 Tangent Bundle 179
As a special case let M D U V D Kn be open. Then TU D U Kn and a vector

field X corresponds to a morphism F W U ! Kn .
Our next goal is to give a concrete description of TM by derivations generalizing the

description of a single tangent space by derivations (Proposition 5.11).
Definition and Remark 8.56. Let .X; OX / be a K-ringed space.
1. A K-derivation of OX is a K-linear homomorphism DW OX ! OX of sheaves such that

DU .fg/ D fDU .g/ C gDU .f / for all U X open, f; g 2 OX .U /. Denote by
DerK .OX / the K-vector space of K-derivations of OX . It is an OX .X/-module via
.g D/U .f / WD g jU DU .f / for D 2 DerK .OX /, g 2 OX .X/, f 2 OX .U /:
2. Define an OX -module via
Der K .OX /.U / WD DerK .OX jU /:
Proposition 8.57. Let M be a premanifold.
1. Let X be a vector field over U . Then the Lie derivative of X
LX W OM .U / ! OM .U /; f 7! .p 7! Tp .f /.Xp //
is a K-derivation of OM .U /.
2. The morphism of sheaves
LW TM ! Der K .OM /;
(8.30)
TM .U / 3 X 7! .LX jV /V U open 2 Der K .OM /.U /
is an isomorphism of OM -modules.
Proof. Assertion (1) follows from the product rule.

Let us show that L is OM -linear. For X; Y 2 TM .U / and g 2 OM .U / we have
LgX CY .f /.p/ D Tp .f /..gX C Y /.p// D Tp .f /.g.p/X.p/ C Y.p//

D g.p/Tp .f /.X.p// C Tp .f /.Y.p// D .g LX C LY /.f /.p/:
Let us construct an inverse map. Let .DV /V U open 2 Der K .OM /.U /. For p 2 U recall
that Tp .M / D DerK .OM;p ; K/ (Proposition 5.11). Define
Xp 2 DerK .OM;p ; K/ by Xp .ŒV; f / WD DV .f /.p/:
Using a chart it is easy to check that U 3 p 7! Xp 2 Tp .M / is a vector field on U and

that this defines an inverse map to L.
180 8 Bundles
Remark 8.58. Let M be a premanifold. Let .U; x/ be a chart with coordinate functions
x 1 ; : : : ; x m . Then
@ @f
W OM jU ! OM jU ; OM .V / 3 f 7! 2 OM .V /; V U open
@x i @x i
is an element of Der K .OM /.U / and . @x@ 1 ; : : : ; @x@m / is a basis of the free OU -module
Der K .OM /jU by Remark 5.8.
8.6 Differential Forms and De Rham Complex
Recall that in this section we only consider real C ˛ -premanifolds with ˛ 1 and com-
plex premanifolds. Differential forms are by definition sections of exterior powers of the
dual of the OM -module TM . Hence we introduce first the notion of “exterior powers” for
general OX -modules using the construction principle of Remark 8.36. We refer to Ap-
pendix Sect. 14.3 for the notion of the exterior power of a module over a commutative
ring.
Definition and Remark 8.59 (Exterior powers of OX -modules). Let .X; OX / be

a ringed space, let F be an OX -module, and let r 0 be an integer. The OX -module
attached to the OX -premodule
^r
U 7! F.U /
.U;OX /
is denoted by rOX F or simply F. It is called the r-th exterior power of F.

r
As the formation of exterior powers is functorial for modules over a ring, we see that
r F is a covariant functor in F.
Remark and Definition 8.60 (Exterior powers of finite locally free modules). Let F
be a finite locally free OX -module of constant rank n 2 N0 . Then r F is a locally free
OX -module of rank nr (Appendix Proposition 14.22).
In particular, det.F/ WD n F is a locally free OX -module of rank 1 that we call the de-
terminant of F. If u is an endomorphism of the OX -module OXn , then u is given by a matrix
A 2 Mn .OX .X//. By Appendix Proposition 14.22, n .u/ is given by the multiplication
with det.A/.
The construction in Appendix Remark 14.24 for modules yields a homomorphism of
OX -modules
r .F_ / ! .r F/_ : (8.31)
We claim that this is an isomorphism if F is finite locally free. Indeed, this can be checked
locally on X. Hence we can assume F D OXn and the claim follows because (14.24) is an
isomorphism for finitely generated free modules (Appendix Remark 14.24).
8.6 Differential Forms and De Rham Complex 181
Definition 8.61. Let M be a premanifold. Define locally free OM -modules
^
i
_
1
˝M WD .TM / ; i
˝M WD 1
.˝M /; i 0:
0
In particular ˝M D OM . The OM -module ˝M i
is called the sheaf of i-differential forms.
For every U M open we call an element of ˝M i
.U / an i-form on U . If M is a real
1
C -premanifold (respectively a real analytic premanifold, respectively a complex pre-
i
manifold), then elements of ˝M .U / are called smooth i-forms (respectively real analytic
i-forms, respectively holomorphic i-forms).
V
Remark 8.62. For every i 0 we have ˝M
i
D Hom. i TM ; OM / by Remark 8.60 and
hence for U M open
i
˝M .U / D f ˛W TU TU ! OU I ˛ alternating i-multilinear formg :
If M is a premanifold of dimension d 2 N0 , then rk ˝M

1
D rk TM D d and hence
!
d
r
rk ˝M D (8.32)
r
r
by Remark 8.60. In particular, ˝M D 0 for r > d .
Remark and Definition 8.63. Let M be a premanifold. We define
d W OM ! ˝M
1
(8.33)
as follows. For f 2 OM .U / let df W TM jU D Der K .OU ; OU / ! OU be the morphism that

sends @ 2 DerK .OV ; OV / to @.f jV / for V U open. Then d is a morphism of sheaves
of K-vector spaces (but not of OM -modules!) and
d.fg/ D f dg C gdf; f; g 2 OM .U /, U M open:
For p 2 M one can think of .df /.p/ 2 ˝M 1

.p/ D Tp .M /_ as the infinitesimal
variation of f at p in an unspecified direction, which gives after evaluation on a tangent
vector the derivative of f at p in that tangent direction.
182 8 Bundles
Remark 8.64 (Local description of differential forms). Let M be a premanifold, let

.U; x/ be a chart with coordinate functions x 1 ; : : : ; x m W U ! K (and hence dimp .M / D
m for p 2 M ). Then dx i 2 ˝M 1
.U / and .dx 1 ; : : : ; dx m / is a basis of ˝M
1
jU . This basis
is dual to the basis . @x i /i of TM .U / (Remark 8.58). For f 2 OM .U / one has
@
X m
@f
df D i
dx i
i D1
@x
(evaluate both sides on @x@j to see the identity).

Let r 1 be an integer. Then ˝M r
jU is a free OU -module and the
dx i1 ^ ^ dx ir ; 1 i1 < < ir m
form a basis (Appendix Proposition 14.22).
Proposition 8.65. Let M be a premanifold. There exists a unique family of morphisms

i C1
.d W ˝M
i
! ˝M /i 0 such that:
(a) d is a morphism of sheaves of K-vector spaces.

(b) d W ˝M
0
D OM ! ˝M 1
is the morphism defined in Remark 8.63.
(c) For U M open, ! 2 ˝M r
.U /, 2 ˝M
s
.U / one has
d.! ^ / D d! ^ C .1/r ! ^ d:
i 1 i C1
(d) d ı d D 0W ˝M ! ˝M for all i 1.
V 1
Proof. As ˝M 1
generates ˝M as OM -algebra, there exists at most one such family.
Because of this uniqueness, it suffices to prove the existence locally. Hence we may
assume that M Km is open. Let x 1 ; : : : ; x m W M ! K be the coordinate functions.
Then ˝M r
is a free OM -module and the dx i1 ^ ^ dx ir for 1 i1 < < ir m form
r
a basis of ˝M . Conditions (a)–(d) imply that we must define
d.f dx i1 ^ ^ dx ir / D df ^ dx i1 ^ ^ dx ir : (8.34)
Then (a) and (b) are clearly satisfied. To check (c) and (d) is a straightforward computation
(for (d) one uses that taking partial derivations in different directions commute with each
other by Appendix Proposition 16.19).
8.6 Differential Forms and De Rham Complex 183
Example 8.66 (Gradient, curl, and divergence). Gradient, curl, and divergence are all
special cases of the above construction. Let U Km be open. Then a section of TU
over an V U , i.e., a vector field X over V , can be considered as a morphism of
manifolds XW V ! Km (Example 8.55). Moreover, ˝U1 is a free OU -module with ba-
sis .dx 1 ; : : : ; dx m /, where x i W U ! K is the i-th coordinate function. Hence the choice
of coordinates allows us to identify
OU D ˝U0 ! ˝Um ;
f 7! f .dx 1 ^ ^ dx m /;
and
TU ! ˝U1 ! ˝Um1 ;
X D .f1 ; : : : ; fm / 7!
X
m
fi dx i 7!
X
m
b
fi .1/i 1 .dx 1 ^ ^ dx i ^ ^ dx m /;
i D1 i D1
where as usual .b/ denotes an omission of that term.

The identification of 1-forms and vector fields identifies d W ˝M 0
! ˝M 1
, f 7!
Pm @f @f @f
i D1 @x i dx i
with the gradient f !
7 . @x 1
; : : : ; @x m /.
The identification of vector fields and .m 1/-forms and of m-forms and functions
identifies d W ˝Um1 ! ˝Um given by
!
X m
b
fi .1/i 1 dx 1 ^ ^ dx i ^ ^ dx m 7!
X m
@f
i
.dx 1 ^ ^ dx m /
i D1 i D1
@x
P
with the divergence X D .f1 ; : : : ; fm / 7! m @f
i D1 @x i .
Finally, if m D 3, then d W ˝U ! ˝U is given by sending the 1-form f1 dx 1 C f2 dx 2 C
1 2
f3 dx 3 to

@f2 @f1 @f1 @f3 @f3 @f2
dx ^ dx
1 2
dx ^ dx C
1 3
dx 2 ^ dx 3 :
@x 1 @x 2 @x 3 @x 1 @x 2 @x 3
Hence it is identified with the curl.
Definition 8.67. Let M be a premanifold. The complex

d d d
OM ! ˝M
1
! ˝M
2
! : : : (8.35)
is called the de Rham complex of M . Note that although the ˝M r

are OM -modules, the
maps d are only morphism of sheaves of K-vector spaces.
For U M an r-form ! 2 ˝M r
.U / is called closed (respectively exact) if d! D 0
(respectively if there exists 2 ˝M .U / such that d D !). As d ı d D 0, every exact
r1
form is closed.
184 8 Bundles
Theorem 8.68. (Poincaré lemma) Let M WD f z 2 Km I 8 i D 1; : : : ; mW jzi j < 1 g

Km . Then for every r 2 N every closed form ! 2 ˝M
r
.M / is exact.
We do not give a proof of this result here but refer to [AmEs3] XI Corollary 3.12 for
the real C 1 -case. The explicit formula given there also shows the Poincare lemma in the
real analytic case. For the complex case we refer to [Wel] II, Example 2.13.
Corollary 8.69. Let M be a premanifold. Then the de Rham complex is an exact sequence
of sheaves of K-vector spaces.
Proof. The exactness of the de Rham complex means that locally every closed form is
exact. Hence we may assume that M is as in the Poincaré lemma 8.68.
As the kernel of d W OM ! ˝M 1
is the sheaf of locally constant K-valued functions,
Corollary 8.69 means (by Appendix Example 15.3) that we have a quasi-isomorphism of
KM -modules
qi s
KM ! ˝M : (8.36)
In Chap. 10 we will define the cohomology of a complex of sheaves and then (8.36) will
imply that KM and ˝M have the same cohomology.
Example 8.70. Let us give an immediate application of the Poincaré lemma to vector
fields. Let U Km be open. Identifying morphisms of manifolds XW U ! Km , X D
.f1 ; : : : ; fm / with 1-forms !X over U , we see by Example 8.66 that d.!X / D 0 if and
only if
@fi @fj
j
D 8 i; j D 1; : : : ; m: (*)
@x @x i
Hence from the exact de Rham complex we obtain an exact sequence of sheaves of K-
vector spaces
grad
0 ! KU ! OU ! I ! 0; (**)
where for V U open I.V / denotes the K-vector space of those vector fields .f1 ; : : : ;
fm /W V ! Km satisfying (*) and where grad is the morphism sending a function on V to
its gradient.
Hence the cohomology sequence obtained from (**) shows that every vector field
.f1 ; : : : ; fm / on U satisfying (*) is the gradient of a morphism U ! K if and only if
H 1 .U; KU / ! H 1 .U; OU / is injective. This is for instance the case if U is simply con-
nected because then H 1 .U; KU / D 0 by Proposition 7.5.
8.7 Problems 185
8.7 Problems
Problem 8.1. Let the group G WD f˙1g act on S 1 by a rotation by , i.e., .1/ e i WD
e i. C/ D e i . Show that there are precisely two isomorphism classes X and Y of fiber
bundles with basis B WD S 1 , typical fiber S 1 , and structure group f˙1g. Show that X and
Y are isomorphic as B-manifolds.
Problem 8.2. Let pW Z ! B be a morphism of premanifolds, G Aut.p/ a sheaf of

subgroups. Let D .W X ! B; ŒUi ; hi i / be a twist of p with structure sheaf G. Denote
the trivial twist .p; ŒB; idZ / by 0 .
1. Show that for U B open jU WD . j 1 .U / ; ŒUi \ U; hi j 1 .Ui \U / / is a twist of

p jp1 .U / W p 1 .U / ! U with structure sheaf GjU .
2. For U B open let T .U / be the set of isomorphisms jU ! 0jU of twists of p jp1 .U /
with structure sheaf GjU . Let G act on T via composition. Show that T is a G-torsor.
3. Show that 7! T yields an equivalence of categories from the category of twists of
p with structure sheaf G (morphisms in this category are isomorphisms of twists) and
the category .Tors.G//. Show that this equivalence induces on isomorphism classes
the composition of (8.4) with the isomorphism HL 1 .B; G/ ! H 1 .B; G/ (7.5).
Problem 8.3. Let B and Z be premanifolds, let A be the group of automorphisms of the
premanifold Z, and let pW Z ! B be the constant morphism with image fbg for some
b 2 B. Show that Aut.p/ is the skyscraper sheaf in b with value A and deduce that there
are no non-trivial twists of p.
Hint: Problem 7.5.
Problem 8.4. Show that for a sequence of OX -modules 0 ! F0 ! F ! F00 (respectively

F0 ! F ! F00 ! 0) the following assertions are equivalent:
(i) The sequence is exact.

(ii) The sequence 0 ! HomOX .G; F0 / ! HomOX .G; F/ ! HomOX .G; F00 / (respectively
0 ! HomOX .F00 ; G/ ! HomOX .F; G/ ! HomOX .F0 ; G/) of .X; OX /-modules is
exact for every OX -module G.
(iii) The sequence 0 ! HomOX .G; F0 / ! HomOX .G; F/ ! HomOX .G; F00 / (respectively
0 ! HomOX .F00 ; G/ ! HomOX .F; G/ ! HomOX .F0 ; G/) of OX -modules is exact for
every OX -module G.
Problem 8.5. Let .X; OX / be a ringed space. An OX -module F is called of finite type
(respectively of finite presentation) if there exists an open covering .Ui /i of X and for all
i an exact sequence of OUi -modules OnUi ! FjUi ! 0 for some n 2 N0 (respectively
OmUi ! OUi ! F jUi ! 0 for some m; n 2 N0 ).
n
186 8 Bundles
1. Show that every finite locally free OX -module is of finite presentation.

2. Show that (8.26) is an isomorphism for every OX -module G and for every OX -module
F of finite presentation.
Hint: Use Problem 8.4 and the five lemma to reduce to F D OX .
3. Let F and G be OX -modules of finite presentation. Let x 2 X and let w0 W Fx ! Gx be
an isomorphism of OX;x -modules. Show that there exists an open neighborhood U of
x and an isomorphism wW FjU ! GjU of OU -modules such that wx D w0 .
Problem 8.6. Let .X; OX / be a ringed space and let F be an OX -module of finite type
(Problem 8.5).
1. Let x 2 X and s1 ; : : : ; sn 2 F.X/ such that the germs .s1 /x ; : : : ; .sn /x generate the
OX;x -module Fx . Show that there exists an open neighborhood U of x such that
.s1 /y ; : : : ; .sn /y generate the OX;y -module Fy for all y 2 U .
2. Show that for every r 2 N0 the subset
f x 2 X I Fx can be generated as OX;x -module by r elementsg
is open in X.
3. Deduce that Supp.F/ WD f x 2 X I Fx ¤ 0 g is closed in X.
Problem 8.7. Let F and G be two OX -modules.
1. Show that
U 7! F.U / ˝OX .U / G.U /
defines an OX -premodule. Its sheafification is an OX -module, which is called the ten-

sor product of F and G and denoted by F ˝OX G.
2. Let W F G ! F ˝OX G be the sheafification of the morphism of presheaves F.U /
G.U / ! F.U / ˝OX .U / G.U /, .f; g/ 7! f ˝ g. Show that for an OX -module H com-
position with yields a OX .X/-linear bijection from the set of OX -linear morphisms
F ˝OX G ! H with the set of OX -bilinear maps F G ! H.
3. Show that for all x 2 X there exists a functorial isomorphism of OX;x -modules .F˝OX
G/x ! Fx ˝OX;x Gx .
4. Show that if F and G are finite locally free OX -modules, then F ˝OX G is finite locally
free and rkx .F ˝OX G/ D rkx .F/ rkx .G/ for all x 2 X.
5. Suppose that .X; OX / is a premanifold M and that F and G are finite locally free OM -
modules. Let F and G be the corresponding vector bundles and define F ˝ G to
be the vector bundle corresponding to F ˝OX G. Show that for all p 2 M one has
.F ˝ G/.p/ D F .p/ ˝K G.p/ for the fibers.
8.7 Problems 187
Problem 8.8. Let F, G and H be OX -modules. See Problem 8.7 for the definition of
tensor products.
1. Show that there exists a functorial isomorphism of OX -modules
HomOX .F ˝OX G; H/ ! HomOX .F; HomOX .G; H//: (8.37)
2. Show that there exists a functorial homomorphism of OX -modules
HomOX .F; G/ ˝OX H ! HomOX .F; G ˝OX H/; (8.38)
which is an isomorphism if F or H is finite locally free.
Problem 8.9. Let .X; OX / be a ringed space. Denote by Pic.X/ the set of isomorphism
classes of finite locally free OX -modules of rank 1. Show that .ŒL; ŒM/ 7! ŒL ˝OX M
defines the structure of an abelian group on Pic.X/ such that for ŒL 2 Pic.X/ one has
ŒL1 D ŒL_ .
Show that if .X; OX / D B is a premanifold, then this group structure on Pic.B/ coin-
cides with the group structure defined in Remark 8.25.
Hint: Problem 8.8.
Problem 8.10. Let .X; OX / be a ringed space, let E, F, H be OX -modules. Assume that
H is finite locally free.
1. Show that there are functorial isomorphisms of .X; OX /-modules
HomOX .E ˝OX F; H/ ! HomOX E; F_ ˝OX H ! HomOX F; E_ ˝OX H :
An OX -linear homomorphism ˇW E ˝OX F ! H is called a pairing. If H D OX , then

ˇ is called a bilinear form. A pairing is called perfect if the corresponding OX -module
homomorphisms sˇ W E ! F_ ˝OX H and rˇ W F ! E_ ˝OX H are both isomorphisms.
2. Show that if E and F are finite locally free, then rˇ is an isomorphism if and only if sˇ
is an isomorphism.
Problem 8.11. Let .X; OX / be a ringed space, n 1 an integer, and let F be a locally
free OX -module of rank n. Show that the wedge product yields for all 1 r n a perfect
pairing (Problem 8.10)
r F ˝OX nr F ! n F D det.F/:
In particular we obtain an isomorphism of OX -modules
r F ! .nr F/_ ˝OX det.F/:

188 8 Bundles
Problem 8.12.
1. Let A be a commutative ring, M a finitely generated free A-module and let !W M
M ! A be a symplectic pairing (i.e., ! is an A-bilinear map, !.m; m/ D 0 for all
m 2 M , and the induced map M ! M _ is an isomorphism of A-modules). Show
that the rank of M is even, say 2n, and that there exists a basis of M such that ! is
given with respect to the basis by the matrix
!
0 In
J WD 2 M2n .A/;
In 0
where In is the .n n/-identity matrix.

2. Let .X; OX / be a ringed space, let E be a finite locally free OX -module, and let
!W E ˝OX E ! OX be an alternating bilinear form (Problem 8.10). It is called
symplectic if ! is perfect.
Show that if ! is a symplectic form on E, then there exists for all x 2 X an open
neighborhood U of x and integer n 0 and an isomorphism of OU -modules EjU !
U , which identifies ! with the alternating form on the free module OU given by the
O2n 2n
matrix J .
Remark: If .X; OX / is a real manifold and E is the tangent bundle of X, then this result
is called the Theorem of Darboux.
Problem 8.13. Let V D K2n and !0 be the symplectic form on V given by the matrix J
(notation as in Problem 8.12). Let Sp.V; !0 / be the Lie group of K-linear automorphisms
of V preserving !0 and let W Sp.V; !0 / ! GL.V / be the inclusion. Let M be a pre-
manifold. Show that there exists an equivalence of the category of pairs .E; !/, where
E is a vector bundle of rank 2n and ! is a symplectic form on E and where morphisms
are isomorphisms of vector bundles respecting the symplectic forms, and the category
.Tors.OM ISp.V;!0 / //. In particular one obtains a bijection between isomorphism classes of
pairs .E; !/ as above and H 1 .M; OM ISp.V;!0 / / D H 1 .M; Sp.V; !0 //.
Show that H 1 ./ corresponds to the map that sends the isomorphism class of .E; !/ to
the isomorphism class of E.
Hint: Problem 8.12.
Problem 8.14. Let n 2 N. Show that for every premanifold M the category .Tors
.OM ISLn .K/ // is equivalent to the category of pairs .E; ı/, where E is a finite locally
free OM -module of rank n and ıW n E ! OM is an isomorphism of OM -modules and
where morphisms are isomorphisms of finite locally free modules preserving ı. In par-
ticular one obtains a bijection between isomorphism classes of pairs .E; ı/ as above and
H 1 .M; OM ISLn .K/ /.
Let L be the line bundle corresponding to n E. Show that the datum of an isomorphism
ı as above is equivalent to the datum of a global section sW M ! L such that s.p/ ¤ 0
for all p 2 M .
8.7 Problems 189
Problem 8.15. Let G be a Lie group, let H be a Lie subgroup. For every premanifold B
let B W PrincH .B/ ! PrincG .B/ be the map of pointed sets deduced by functoriality from
the inclusion H ! G. Let 2 PrincG .B/ be given by W X ! B together with a right
G-action.
1. Consider G=H as a G-premanifold and denote by =H the fiber bundle over B with
typical fiber G=H and structure group G associated with via (8.13). Show that the
quotient of X=H (with respect to the induced right action of H ) exists, that induces
a morphism of premanifolds =H W X=H ! B, and that the underlying B-premanifold
of =H is given by =H .
2. Let s 2 HomB .B; X=H /, i.e., sW B ! X=H is a morphism of premanifolds such
that =H ı s D idB . Let s 2 PrincH .B/ be the pullback of the principal H -bundle
X ! X=H by s. Show that s 7! s yields a bijection between HomB .B; X=H / and
B1 . /.
Problem 8.16. Notation of Problem 8.15. Let 2 PrincH .G=H / be the class of the
principal H -bundle G ! G=H . Show that G=H . / is trivial.
Problem 8.17. Let rW B 0 ! B be a morphism of premanifolds.
1. Show that the pullback of fiber bundles yields a functor r from the category of vector
bundles over B to the category of vector bundles over B 0 . Show that for b 0 2 B 0 and
for a vector bundle E over B one has rkb 0 .r .E// D rkr.b 0 / .E/.
2. Show that if F is a subbundle (respectively a quotient bundle) of a vector bundle E on
B, then r .F / is a subbundle (respectively a quotient bundle) of r .E/.
Problem 8.18. Let .f; f ] /W .X; OX / ! .Y; OY / be a morphism of ringed spaces and let
E be an OY -module.
1. Consider OX as a f 1 OY -module via f ] . Show that f 1 E is a f 1 OY -module and

define f .E/ WD OX ˝f 1 OY f 1 E endowed with scalar multiplication by OX via the
first factor. Show that E 7! f E defines a functor from the category of OY -modules
to the category of OX -modules, called pullback of OY -modules.
2. Show that f .E/ is a finite locally free OX -module if E is a finite locally free OY -
module. In this case one has rkx .f E/ D rkf .x/ .E/ for all x 2 X.
3. Let f W M ! N be a morphism of premanifolds. Show that the pullback of finite
locally free ON -modules corresponds to the pullback of vector bundles (Problem 8.17)
via the equivalence (8.28).
190 8 Bundles
Problem 8.19. Recall that if A is a commutative ring and L is an A-module, then a Lie
bracket on L is an A-bilinear map L L ! L, .m; n/ 7! Œm; n such that Œm; m D 0 for
all m 2 L and such that the Jacobi identity ŒŒm; n; ` C ŒŒn; `; m C ŒŒ`; m; n D 0 holds
for all m; n; ` 2 L. A pair .L; Œ; / consisting of an A-module L and a Lie bracket on L is
called a Lie algebra over A. A homomorphism .L1 ; Œ; / ! .L2 ; Œ; / of Lie algebras over
A is an A-linear map uW L1 ! L2 such that Œu.m/; u.n/ D u.Œm; n/ for all m; n 2 L1 .
Let .X; OX / be a K-ringed space. Show that ŒD1 ; D2 WD D1 ı D2 D2 ı D1 defines
on DerK .OX / the structure of a Lie algebra over the ring OX .X/.
In particular if M is a real C ˛ -premanifold with ˛ 1 or a complex premanifold,
then via (8.30), the OM .M /-module of vector fields over M is endowed with the structure
of a Lie algebra.
Problem 8.20. Let M be a real C ˛ -premanifold with ˛ 1 or a complex premanifold.

Show that there exists a functorial bijection between the set of morphisms of locally K-
ringed spaces P Œ" ! M (Problem 4.5) and the tangent bundle TM .
Problem 8.21. Let M and N be real C ˛ -premanifolds with ˛ 1 or complex preman-

ifolds. Show that TM N D TM TN .
Problem 8.22. Let G be a (real or complex) Lie group with multiplication mG W G

G ! G. Show that TmG W TGG D TG TG ! TG defines on TG the structure of a Lie
group with identity element 0 2 Te .G/. Show that the canonical projection TG ! G is
a homomorphism of Lie groups with kernel Te .G/.
Problem 8.23. Let G be a (real or complex) Lie group. For g 2 G let `g W G ! G,

h 7! gh. Show that
˚W G Te .G/ ! TG ; .g; x/ 7! Te .`g /.x/ (8.39)
p
is an isomorphism of vector bundles. Deduce that TG and ˝G , p 0, are free OG -
modules. Is ˚ a homomorphism of Lie groups if we endow TG with the Lie group
structure defined in Problem 8.22?
Problem 8.24. Notation as in Problem 8.23. Let TG .G/ be the OG .G/-module of global
vector fields on G.
1. Show that G TG .G/ ! TG .G/, .g; X/ 7! T .`g / ı X ı `1 g , defines an action of

the group G on TG .G/ by K-linear automorphisms. Let Lie.G/ be the K-subspace of
TG .G/ consisting of X 2 TG .G/ such that X D T .`g / ı X ı `1 g for all g 2 G.
2. Show that Lie.G/ is K-Lie subalgebra of the Lie algebra TG .G/ (Problem 8.19) and
that this construction yields a functor Lie from the category of real (respectively com-
plex) Lie groups to the category of Lie algebras over R (respectively over C).
8.7 Problems 191
3. Show that the restriction of ˚ 1 (Problem 8.23) to Lie.G/ followed by the projec-
tion to Te .G/ yields an isomorphism of K-vector spaces Lie.G/ ! Te .G/, which is
functorial in G.
4. Let G be a Lie group and let G 0 be its identity component. Show that Lie.G 0 / Š
Lie.G/.
5. Deduce from Problem 6.7 that a homomorphism of connected Lie groups 'W G ! G 0
is a covering map if and only if Lie.'/W Lie.G/ ! Lie.G 0 / is an isomorphism.
Remark: One can show that the functor Lie induces an equivalence between the category
of simply connected Lie groups and the category of finite-dimensional K-Lie algebras
(e.g., [HiNe] 9.4.11 and 9.5.9) and deduce that two Lie groups have isomorphic Lie alge-
bras if and only if their universal covers (Problem 6.8) are isomorphic.
Problem 8.25. Let .X; OX / be a locally ringed space and let E be a finite locally free
OX -module. Show that every local direct summand of E is finite locally free.
Hint: Use Problem 8.5 and Appendix Problem 14.31.
Problem 8.26. Let D WD .nz /z2C 2 ZC such that f z 2 C I nz ¤ 0 g is discrete and

closed in C (such a tuple is called a divisor on C). For U C open let LD .U / be the
set of meromorphic functions f on U such that ordz .f / nz for all z 2 U . Show
that LD is via restriction of meromorphic functions a sheaf on C. Show that addition
of meromorphic functions and multiplication with holomorphic functions makes LD into
a locally free OC -module of rank 1.
Problem 8.27. Notation of Definition 8.49.
1. Suppose that F is a quotient bundle of E. Show that wW E ! F has locally on M

a section and deduce that Ker.w/ is a local direct summand of E and hence corresponds
to a subbundle of E, denoted by Ker.$ / and called the Kernel of $ . Show that for the
fibers one has Ker.$ /.p/ D Ker.$ .p// for p 2 M .
Hint: Problem 8.25.
2. Let E be a subbundle of F . Show that F=E is a finite locally free OM -module. The
corresponding vector bundle is denoted by F=E and called the quotient of F by the
subbundle E. Show that for the fibers one has .F=E/.p/ D F .p/=E.p/ for p 2 M .
Problem 8.28. Let E be a finite-dimensional K-vector space and d an integer with

0 d dimK .E/. Let M WD Grassd .E/. For a point p 2 M let Up E be the
`
corresponding K-subspace of dimension d . Show that U WD p2M Up is a subbundle
of rank d of the trivial vector bundle M E called the tautological vector bundle on
Grassd .E/.
192 8 Bundles
Problem 8.29. Notation as in Problem 8.28. Let U be the OGrassd .E/ -module correspond-
ing to the tautological vector bundle on Grassd .E/. It is a local direct summand of
E WD OGrassd .E/ ˝K E. Show that TGrassd .E/ Š HomOGrassd .E/ .E=U; U/. In particular
Tp .Grassd .E// D HomK .E=Up ; Up / for all p 2 Grassd .E/.
Problem 8.30. Let M be a real C ˛ -premanifold with ˛ 1 or a complex premanifold.

p
For p 0 and U M open identify ˝M .U / with the set of p-multilinear alternating
p
morphisms TU ! OU .
p
1. Let U M be open, X a vector field over U , and ! 2 ˝M .U /, p 1. Define
p1
iX .!/ 2 ˝M .U / by
iX .!/. 1 ; : : : ; p1 / WD !.X; 1 ; : : : ; p1 /
for i 2 TM .V /, V U open. For f 2 ˝M 0

.U / D OM .U / define iX .f / WD 0. Show
p p1
that this defines a homomorphism iX W ˝M ! ˝M of OM -modules.
p q
2. Show for p; q 0 and for ! 2 ˝M .U /, 2 ˝M .U / one has
iX .! ^ / D iX .!/ ^ C .1/p ! ^ iX ./:
3. For p 0 define the Lie derivative
p p
LX W ˝M ! ˝M ; LX WD d ı iX C iX ı d:
Show that for p D 0 this is the map defined in Proposition 8.57.

4. Show that for differential forms ! and one has
d ı LX D LX ı d;
LX .! ^ / D LX .!/ ^ C ! ^ LX ./:
Soft Sheaves
9
In this chapter we study soft sheaves. The class of soft sheaves is on one hand large enough
to include all kinds of interesting sheaves on real C ˛ -manifolds if ˛ 1, for instance
the structure sheaf. On the other hand the cohomology of soft sheaves vanishes, and this
yields immediately several local-global principles. The chapter starts with the definition
and general properties of soft sheaves. In particular we will see that every global section
has an arbitrary fine partition (Proposition 9.9). In Sect. 9.2 we will show that the structure
sheaf of a real C ˛ -manifold for ˛ 1 is a soft sheaf. In the last section we show that
the first cohomology vanishes for any soft sheaf of groups on a paracompact Hausdorff
space. We obtain several immediate corollaries for real C ˛ -manifolds with ˛ 1: The
fact that sections of vector bundles over closed sets can always be extended to a global
section (and in particular a smooth version of Tietze’s extension theorem), a description
of the Picard group, or the fact that every subbundle of a vector bundle is split.
9.1 Definition and Examples of Soft Sheaves
Let X be a topological space, let Z X be a subspace, and let iW Z ! X be the inclusion.

Let F be a sheaf on X. Recall that we defined F.Z/ as .i 1 F/.Z/, in other words
[
F.Z/ D .si ; Ui /i 2I I Ui X open with Z Ui ;
i
si 2 F.Ui / with .si /z D .si 0 /z (9.1)

for all i; i 0 and all z 2 Z \ Ui \ Ui 0 =
with .Ui ; si /i 2I .Vj ; tj /j 2J if .si /z D .tj /z for all i 2 I , j 2 J and for all z 2
Ui \ Vj \ Z.

DOI 10.1007/978-3-658-10633-1_9
194 9 Soft Sheaves
Our first goal is to show that under certain hypotheses, F.Z/ has a simpler description.
For every open neighborhood U of Z we have a restriction map
F.U / ! F.Z/; s 7! s jZ WD Œ.U; s/
and hence obtain a map
rZ WD rZX WD colim rZU W colim F.U / ! F.Z/: (9.2)

U ZU
Here the left-hand side has the simple description
colim F.U / D f .U; s/ I U X open with Z U , s 2 F.U /g = (9.3)

ZU
with .U; s/ .V; t/ if there exists Z W U \ V open such that s jW D t jW .

The map rZ is always injective. Let .U; s/ and .V; t/ represent elements in F0 .Z/ with
sz D tz for all z 2 Z. Then there exists Z W X open with s jW D t jW . We have the
following criteria for rZ to be bijective.
Proposition 9.1. The map rZ (9.2) is an isomorphism for every sheaf F on X if one of
the following hypotheses is satisfied:
1. The space X is Hausdorff paracompact and Z X is closed.

2. The subspace Z has a fundamental system of Hausdorff paracompact open neighbor-
hoods.
3. The subspace Z is compact and a relatively Hausdorff subspace of X (Definition 1.22).
Hypothesis 2 is in particular satisfied for every subspace Z if X is Hausdorff and

hereditarily paracompact, for instance if X is a paracompact Hausdorff premanifold or if
X is metrizable (Proposition 1.13).
Proof. Let s 2 F.Z/ be represented by .Ui ; si /i 2I .

(i). Let us assume that Hypothesis 1 or Hypothesis 2 is satisfied. In this case we first
show that we may assume that .Ui /i is a locally finite open covering of X and that X is
paracompact Hausdorff. If 1 is satisfied, we first add U0 WD X n Z to the covering (with
s0 2 F.U0 / arbitrary) to obtain a covering of X and then replace it by a locally finite re-
finement. If 2 is satisfied, we replace X by an open Hausdorff paracompact neighborhood
S
X 0 of Z contained in i Ui and replace .Ui \ X 0 /i by a locally finite refinement. In both
cases there exists an open covering .Vi /i of X with Vi Ui by Corollary 1.21. Clearly
.Vi /i is again locally finite.
9.1 Definition and Examples of Soft Sheaves 195
(ii). We now conclude the proof if Hypothesis 1 or Hypothesis 2 is satisfied. Note that
˚
for all i; j 2 I the set Wij WD x 2 Ui \ Uj I .si /x D .sj /x is open in X. Define
˚
W WD x 2 X I 8 i; j 2 I W x 2 Vi \ Vj ) .si /x D .sj /x
\ c c
D Vi [ Vj [ Wij ;
i;j 2I
where . /c denotes the complement in X. Then Z W . It suffices to show that W is

open in X. Indeed, then si jVi \Vj \W D sj jVi \Vj \W for all i; j 2 I implies that there exists
sQ 2 F.W / such that sQ jVi \W D si jVi \W for all i; hence .W; sQ / represents a preimage of s.
To see that˚ W is open we choosefor all y 2 X an open neighborhood Wy of y such
that I.y/ WD i 2 I I Wy \ Vi ¤ ; is finite. For y 2 W one has
˚
W \ Wy D x 2 Wy I 8 i; j 2 I.y/W x 2 Vi \ Vj ) .si /x D .sj /x ;
which is open in X. Hence every point y of W contains an open neighborhood contained

in W , namely W \ Wy . This shows that W is open in X.
(iii). Now suppose that Hypothesis 3 is satisfied. By Appendix Proposition 12.54,
every z 2 Z has a compact neighborhood Kz of z in Z contained in Uiz for some iz 2 I .
Since Z is compact, finitely many Kz1 ; : : : ; Kzn cover K. Set Kr WD Kzr , Ur WD Uizr and
sr WD sizr for r D 1; : : : ; n.
For n D 1 we are done. Now s1jU1 \U2 and s2jU1 \U2 are two sections whose restriction to
\U2
K1 \ K2 is equal to s jK1 \K2 . Hence the injectivity of the map rKU11\K 2
(9.2) yields an open
neighborhood V of K1 \ K2 in U1 \ U2 with s1jV D s2jV . As Z is relatively Hausdorff
in X and Ki n V is compact we find by Proposition 1.23 disjoint open neighborhoods
Ui0 Ui of Ki n V for i D 1; 2. The three sections s1jU10 , s2jU20 , and s1jV D s2jV glue to
a section on U10 [ U20 [ V , which extends s jK1 [K2 . We conclude by induction.
Definition 9.2. Let X be a paracompact Hausdorff space. A sheaf F on X is called soft

if for every closed subspace A of X the restriction map F.X/ ! F.A/ is surjective.
Even if one of the hypotheses of Proposition 9.1 is satisfied, the notion of sections over
a subspace is somewhat subtle. Consider the following example.
Example 9.3. Let X be a paracompact Hausdorff space and let CX be the sheaf of con-
tinuous R-valued functions on X. We claim that CX is soft.
Let A X be closed and let s 2 CX .A/. We want to extend s to a section over X.
As paracompact Hausdorff spaces are normal (Proposition 1.18), it would be tempting
simply to apply Tietze’s extension theorem (Theorem 1.15 (iii)) to conclude that s can be
extended from A to X. But there is the following technical point to observe: The restric-
tion CX jA is usually not the sheaf of continuous functions on A. For instance if A D fxg
for some x 2 X, then CX jA .A/ D CX;x and CX jA is not the sheaf of continuous functions
196 9 Soft Sheaves
fxg ! R. More generally CX .A/ consists by Proposition 9.1 of germs of continuous

functions defined in some open neighborhood of A as in (9.3). Hence there is only a map
W CX .A/ ! CA .A/, which sends a representative .U; s/ (U an open neighborhood of
A, s a continuous function on A) to the continuous function s jA on A. The composition

CX .X/ ! CX .A/ ! CA .A/ is surjective by Tietze’s extension theorem but it is not
immediate that CX .X/ ! CX .A/ is surjective.
Nevertheless we can use Theorem 1.15 as follows. Let U be an open neighborhood
of A, sQ 0 2 CX .U / extending s. Applying Theorem 1.15 (iii) twice we find a closed
neighborhood B of A and a closed neighborhood C of B with C U . By Tietze’s
extension theorem (Theorem 1.15 (iv)) we find a continuous function f W X ! R with
f .b/ D 1 for all b 2 B and f .x/ D 0 for all x 2 X n C ı . Then the germ of f in every
point of A is 1 and hence sQ WD f jU sQ 0 also extends s. It can be extended to a continuous
function on X by 0.
Example 9.4. Let X be a connected paracompact Hausdorff space consisting of more

than one point. Then the constant sheaf ZX is not soft. Take x; y 2 X, x ¤ y, A WD
fx; yg. As X is Hausdorff, there exist open neighborhoods U 3 x and V 3 y with
U \ V D ;. The locally constant function f with value 0 on U and value 1 on V defines
a section of ZX .A/ that cannot be extended to X because every locally constant function
on X is constant.
Remark 9.5. Let X be a paracompact Hausdorff space, let A X be a closed subspace,

and let F be a soft sheaf on X. Then FjA is a soft sheaf on A: If A is closed and B A is
closed, then B is closed in X. Therefore the composition of restrictions F.X/ ! F.A/ !
F.B/ is surjective and hence F.A/ ! F.B/ is surjective.
Proposition 9.6. Let X be a paracompact Hausdorff space and F be a sheaf on X. Sup-

pose that there exists a family .Zj /j 2J of subspaces such that the interior Zjı cover X
and such that for all j and every closed subspace A of X with A Zj the restriction
F.Zj / ! F.A/ is surjective. Then F is soft.
Proof. We first remark that one can replace .Zj /j by a refinement. Let .Yi /i 2I be a refine-
ment of .Zj /j such that .Yiı /i is an open covering. For all i 2 I let j 2 J with Yi Zj . If
A is a closed subspace of X with A Yi , then the surjectivity of F.Zj / ! F.Yi / ! F.A/
implies the surjectivity of F.Yi / ! F.A/.
Hence we may first replace .Zj /j by the open covering .Zjı /j and then by a locally
finite open covering .Ui /i because X is paracompact. By the shrinking lemma (Corol-
lary 1.21) we find an open covering .Vi /i 2I with Vi Ui for all i. As a subset of Vi
is closed in X if and only if it is closed in Vi , the restriction Fj Vi is soft for all i by
Remark 9.5.
Let A X be a closed subspace, s 2 F.A/. We want to extend s to X. For J I let
S
CJ WD A [ i 2J Vi . This is a closed subset of X because .Vi /i is locally finite (Appendix
9.1 Definition and Examples of Soft Sheaves 197
Corollary 12.33). We have CI D X. Define

˚
E WD .J; tJ / I J I , tJ 2 F.CJ / with tJ jA D s :
Then .;; s/ 2 E . Define a partial order on E by .J; t/ .J 0 ; t 0 / if J J 0 and t 0 jCJ D t.

If F E is a totally ordered subset, then F has an upper bound: Set
[
JQ WD J
.J;tJ /2F
and let tQ 2 F.CJQ / be the unique section such that tQjFJ D tJ (exists by Proposition 3.59).
Therefore we may apply Zorn’s lemma. Let .J; tJ / be a maximal element of E . It suffices
to show that J D I .
Assume that there exists i 2 I n J . As Fj Vi is soft, there exists ti 2 F.Vi / such that
ti j Vi \CJ D tJ j Vi \CJ . Let tQ 2 F.Vi [ CJ / be the element that restricts to ti and to tJ . Then
.J [ fig; tQ/ 2 E with .J; tJ / < .J [ fig; tQ/. This is a contradiction.
Corollary 9.7. Let X be a paracompact Hausdorff space and F be a sheaf on X. Then

F is soft if and only if every point x 2 X has a closed neighborhood N such that FjN is
soft.
Proof. The condition is necessary by Remark 9.5 and sufficient by Proposition 9.6.
Proposition 9.8. Let X be a paracompact Hausdorff space and let OX be a sheaf of (not
necessarily commutative) rings on X. Suppose that OX is soft. Then every left and every
right OX -module is soft.
Proof. It suffices to show the assertion for left modules. Let A X be closed and
s 2 F.A/. By Proposition 9.1 there exists U X open with A U and sQ 2 F.U /
extending s. As X is normal (Proposition 1.18), we find by Theorem 1.15 (iii) a closed
neighborhood C of A contained in U . Then A and @C are disjoint closed subsets. As OX
is soft, O.X/ ! OX .A [ @C / is surjective and we find u 2 OX .X/ with ux D 1 for all
x 2 A and ux D 0 for all x 2 @C . Then sQ jC ujC extends s and can be extended to X by
zero.
Proposition 9.9 (Soft sheaves and partitions of sections). Let X be a paracompact

Hausdorff space, let A be a soft sheaf of abelian groups on X and let s 2 A.X/. Then for
every open covering .Ui /i 2I of X there exists a family .si /i 2I of si 2 A.X/ such that:
(a) supp.si / Ui for all i 2 I . ˚

(b) For all x 2 X there exists x 2 U X open such that i 2 I I si jU ¤ 0 is finite.
P
(c) For all x 2 X one has sx D i 2I .si /x (a finite sum by (b)).
P
Then .si /i is called a partition of s subordinate to .Ui /i and we write s D i 2I si .
198 9 Soft Sheaves
Proof. (i). Assume first that .Ui /i is locally finite. By the shrinking lemma there exist
S
closed subsets Si with Si Ui and X D i Si .
S
For J I set SJ WD i 2J Si . Then SJ is closed because .Si /i is locally finite
(Appendix Corollary 12.33). Let E be the set of pairs .J; .si /i 2J / with J I , si 2 A.X/
P
with supp.si / Ui for all i 2 J and such that . i 2J si /jSJ D s jSJ . Then .;; ;/ 2 E , in
particular E ¤ ;.
The set E is partially ordered by .J; .si /i 2J / .K; .ti /i 2K / if J K and si D ti for
all i 2 J . Clearly, every totally ordered subset of E has an upper bound. Hence E has
a maximal element .J; .si /i 2J / by Zorn’s lemma. It suffices to show that J D I .
Assume there exists ˛ 2 I n J . There exists a unique section t˛ 2 A..X n U˛ / [ .SJ [
S˛ // such that
X
t˛ D 0 on X n U˛ and t˛ D s si on SJ [ S˛
i 2J
P
because s i 2J si D 0 on .X n U˛ / \ .SJ [ S˛ / SJ . As A is soft, there exists
an extension s˛ 2 A.X/ of t˛ . Hence .J [ f˛g; .si /i 2J [f˛g / 2 E is strictly larger than
.J; .si /i 2J /. This is a contradiction.
(ii). In general, let .Vj /j 2J be a locally finite refinement of .Ui /i such that Vj
U.j / for a map W J ! I . Let .tj /j be a partition of s subordinate to .Vj /. Define
P
si WD j 21 .i / tj , which makes sense because .Vj /j 2J is locally finite. Then .si /i 2I is
a partition of s subordinate to .Ui /i .
In fact, the existence of such partitions characterizes soft sheaves (Problem 9.5).
9.2 Softness of Sheaves of Differentiable Functions
The next goal is to show that the structure sheaf of a real C ˛ -manifold with ˛ 1 is soft.
This is essentially equivalent to the fact that there exist arbitrary fine partitions of 1 (see
Corollary 9.13 and Problem 9.5). Hence it is not surprising that we start with a lemma
that is also a standard step in the construction of smooth partitions of unity.
Lemma 9.10. Let m 2 N0 , U Rm open and K U compact. Then there exists a

C 1 -function 'W Rm ! R with 0 ' 1 and ' jK D 1, supp.'/ U .
Proof. For p 2 Rm , r > 0 set Cr .p/ WD f x 2 Rm I jjx pjj1 < r g. Then .Cr .rp//p2Zm
is a locally finite open cover of Rm .
Let 8
<exp 1 ; x 2 .1; 1/I
W R ! R; x 7! 1x 2
:0; x 2 R n .1; 1/:
9.2 Softness of Sheaves of Differentiable Functions 199
Then is a C 1 -function with supp./ D Œ1; 1. Moreover, define for " > 0 and p 2 Rm
Ym x
i
p" W Rm ! R; x 7! pi :
i D1 "
This is also a C 1 -function and we have supp.p" / D C" ."p/ and p" jC" ."p/ > 0. Hence,
P
" W Rm ! R, x 7! p2Zm p" .x/ (a finite sum for every x 2 Rm ) is a C 1 -function with
P
" > 0. Hence 'p" WD p" = " is a well-defined C 1 -function and p2Zm 'p" .x/ D 1 for
every x 2 Rm and supp.'p" / D supp.p" /.
n U is compact, there exists o" > 0 such that C2" .p/ U for
As K
P
every p 2 K. Set
P" WD p 2 Zm I supp.'p" / \ K ¤ ; and define 'W Rm ! R, x 7! p2P" 'p" .x/. Since
P P
p2Zm 'p .x/ D 1 for every x 2 R , the definition of P" yields '.x/ D p2P" 'p" .x/ D 1
" m
S
for every x 2 K. Also, we have supp.'/ p2P" supp.'p" / U because supp.'p" / D
C" .p/ C2" .p/ U for every p 2 K.
Theorem 9.11. Let M be a real paracompact Hausdorff C ˛ -premanifold with ˛ 2 N0 [

f1g. Then CM
˛
is a soft sheaf.
Proof. Every point p 2 M has a chart .U; x/ at p and a compact neighborhood p 2 C

˛
U . By Corollary 9.7 it suffices to show that x .CM jC / is a soft sheaf on the compact subset
x.C / of R , where m D dimp .M /. Then every closed subset of x.C / is compact. Hence
m
it suffices to show that every C ˛ -function f defined on some open neighborhood U of

a compact subset K Rm can be extended to a C ˛ -function on Rm after possibly shrink-
ing U . As K is compact, there exists " > 0 such that K 0 WD f x 2 Rm I d.x; K/ " g
U (d some metric on Rm defining the topology of Rm ).
Due to Lemma 9.10 there exists a C ˛ -function gW Rm ! R such that g jK 0 D 1 and
such that supp.g/ U (we may even assume that g 0). Let
8
<.fg/.x/; x 2 UI
hW Rm ! R; x 7!
:0; x 2 Rn n U:
Then h is the desired extension because h D f on the open neighborhood f x 2

Rm I d.x; K/ < " g of K.
Corollary 9.12. Let M be a real paracompact Hausdorff C 1 -premanifold. Then the

tangent sheaf TM and the sheaves of differential forms ˝M
r
, r 0, are soft sheaves.
Proof. Theorem 9.11 and Proposition 9.8.
Finally, as we have for every soft sheaves and every global section arbitrary fine parti-
tions, we obtain in particular the existence of partitions of unity.
200 9 Soft Sheaves
Corollary 9.13 (Partition of unity). Let M be a paracompact Hausdorff C ˛ -premanifold.

For every open covering .Ui /i there exist fi 2 CM
˛
.M / such that:
(a) supp.fi / Ui and .supp fi /i 2I is locally finite.

P
(b) i 2I fi D 1.
Proof. Theorem 9.11 and Proposition 9.9.
9.3 Triviality of H 1 for Soft Sheaves and Applications
In the previous section we have seen that many interesting sheaves are soft. Now we show
that the first cohomology vanishes for soft sheaves. This will allow us to prove several
global results that a priori hold only locally.
Theorem 9.14. Let X be a paracompact Hausdorff space and let G be a soft sheaf of
groups on X. Then H 1 .X; G/ D 0.
Once we have defined higher cohomology groups in Chap. 10, we will show that
H p .X; A/ D 0 for all p 1 and for every soft sheaf A of abelian groups (Proposi-
tion 10.17).
Proof. Let T be a G-torsor. We have to show that T .X/ ¤ ; because then T is trivial by
Proposition 7.4. Let .Ui /i be an open covering of X such that there exist ti 2 T .Ui / for
all i. As X is paracompact, we may assume that .Ui /i is locally finite. By the shrinking
S
lemma we find Si Ui such that Si is closed in X and such that i Si D X.
S
For J I set SJ WD i 2J Si (note that SJ is closed in X because .Si /i is locally
finite) and define
E WD f .t; J / I J I; t 2 T .SJ / g :
Then . ; ;/ 2 E and hence E ¤ ;. It is an ordered set with partial order .t; J / .t 0 ; J 0 /

if J J 0 and t D t 0 jSJ . As sections over the members of a locally finite closed covering
can be glued (Proposition 3.59), every totally ordered subset of E has an upper bound.
Hence there exists a maximal element .t; J / in E by Zorn’s lemma. We claim that J D I
(then T .X/ ¤ ;).
Assume that there exists j 2 I n J . Then there exists a unique g 2 G.SJ \ Si / such
that t jSJ \Si D gti jSJ \Si . As G is soft, there exists gQ 2 G.X/, which extends g. Replacing
ti by gQ jSi ti we may assume that t jSJ \Si D ti jSJ \Si and hence we may extend t to Si [ SJ ;
a contradiction to the maximality of .t; J /.
We now apply Theorem 9.14 to real C ˛ -manifolds for ˛ 1 using the fact that their
structure sheaf is soft. Note that all the results below apply slightly more generally to real
paracompact Hausdorff C ˛ -premanifolds with ˛ 1.
9.3 Triviality of H 1 for Soft Sheaves and Applications 201
Corollary 9.15. Let M be a real C ˛ -manifold with ˛ 1 and let E be a CM

˛
-module.
Then H .M; E/ D 0.
1
Proof. The sheaf E is soft by Theorem 9.11 and Proposition 9.8. Hence H 1 .M; E/ D 0
by Theorem 9.14.
Corollary 9.16 (Smooth version of Tietze’s extension theorem). Let M be a real C ˛ -

manifold with ˛ 1, S M a closed subspace. Then for every f 2 OS .S/ (i.e., for
every map f W S ! R such that for every point s 2 S there exist an open neighborhood
U and g 2 CM˛
.U / with g jU \S D f jU \S ) there exists fQ 2 CM
˛
.M / such that f D fQjS .
In particular, any C ˛ -function S ! R on a closed submanifold S of M can be ex-

tended to M .
Proof. Let IS be the sheaf of C ˛ -functions vanishing on S (Remark 5.31) and consider
the exact sequence (5.9) of OM -modules. The cohomology sequence (Proposition 7.22)
yields an exact sequence CM˛
.M / ! OS .S/ ! H 1 .M; IS /. As IS is an ideal in CM ˛
,
H .M; IS / D 0 by Corollary 9.15. Therefore CM .M / ! OS .S/ is surjective.
1 ˛

More generally, a similar argument shows that we can always extend sections of vector
bundles from closed subspaces to the whole manifold:
Corollary 9.17. Let M be a real paracompact Hausdorff C ˛ -premanifold with ˛ 2 N0 [

f1g and let .E; p/ be any vector bundle over M . Then any section s of E over a closed
subspace S of M can be extended to a section of E over M .
Here by a section of E over S we mean a map sW S ! E with p ı s D idS and such for
all p 2 S there exists an open neighborhood U of p in M and a C ˛ -section t of E over
U such that t jU \S D s jU \S .
Proof. Again let IS be the sheaf of C ˛ -functions vanishing on S and let E be the sheaf
of sections of E that is a locally free CM
˛
-module. Then the global section s of the sheaf
E=IS E can be lifted to a global section of E because H 1 .M; IS E/ D 0 by Corollary 9.15.

Corollary 9.18. Let M be a real C ˛ -manifold with ˛ 1. Then composition with sign

OM ! f˙1gM Š .Z=2Z/M induces on cohomology an isomorphism
wW Pic.M / ! H 1 .M; .Z=2Z/M /:
In particular Pic.M / D 1 if M is simply connected.
For a line bundle L over M the cohomology class w.L/ 2 H 1 .M; .Z=2Z/M / is called
the (first) Stiefel-Whitney class of L.
202 9 Soft Sheaves
Proof. The first assertion follows from (8.16) and H 1 .M; CM

˛
/ D 0. The second assertion
follows from Proposition 7.5.
Example 9.19. Let M WD S 1 considered as a real C ˛ -manifold with ˛ 1. Then

H 1 .S 1 ; .Z=2Z/S 1 / Š Z=2Z (Example 7.10). Hence Pic.S 1 / Š Z=2Z has exactly two
elements. As in Example 8.15 one sees that the two line bundles are the trivial line bundle
and the Möbius bundle (which is the line bundle constructed as the Möbius band (Exam-
ple 4.30) but with typical fiber R instead of .1; 1/).
Corollary 9.20. Let M be a real C ˛ -manifold with ˛ 1 and let E be a vector bundle
over M . Then every subbundle of E is split.
Proof. Remark 8.51 and Corollary 9.15.
Corollary 9.21. Let M be a real C ˛ -manifold with ˛ 1, let E and F be vector

bundles over M , and let $ W E ! F be a surjective morphism of vector bundles. Then for
all U M open the map on sections E.U / ! F .U / is surjective.
Proof. Let wW E ! F be the surjective homomorphism of finite locally free OM -

modules corresponding to $ (Proposition 8.49). Then E.U / ! F .U / is identified
with wU W E.U / ! F.U /. The exact sequence of C˛U -modules 0 ! Ker.w/jU !
EjU ! FjU ! 0 yields a cohomology sequence E.U / ! F.U / ! H 1 .U; Ker.w// and
H 1 .U; Ker.w// D 0 by Corollary 9.15.
9.4 Problems
Problem 9.1. Let X be a hereditarily paracompact Hausdorff space and let A X be

locally closed. Show that for every soft sheaf F on X its restriction FjA is a soft sheaf on
A.
Problem 9.2. Let M be a real analytic or a complex premanifold of dimension > 0.

Show that OM is not soft.
Problem 9.3. Let X be a paracompact Hausdorff space and let OX be a sheaf of (not
necessarily commutative) rings on X. Show that OX is soft if and only if every point of x
has a neighborhood U such that for all disjoint closed subspaces A, B of U there exists
a section s 2 OX .U / such that s jA D 1 and s jB D 0.
Problem 9.4. Let X be a paracompact Hausdorff space. Show that every colimit of soft
abelian sheaves is again soft.
9.4 Problems 203
Problem 9.5. Let X be a paracompact Hausdorff space and let A be a sheaf of abelian
groups on X. Show that the following assertions are equivalent:
(i) A is soft.
(ii) For every closed subspace A of X, every s 2 A.A/ and for every locally finite open
covering .Ui /i of A in X there exist si 2 A.X/ for all i with supp.si / Ui and
P
sx D i 2I .si /x for all x 2 A.
Suppose that A is a sheaf of (not necessarily commutative) rings. Show that A is soft if
and only if (ii) holds for A D X and s D 1.
Problem 9.6. Let X be a paracompact Hausdorff space. A sheaf of abelian groups A on

X is called fine if the sheaf HomZX .A; A/ is soft.
1. Show that a sheaf of abelian groups A on X is fine if and only if for every closed subset
A of X and for every neighborhood U of A there exists an endomorphism of A that is
1 on A and 0 outside U .
2. Show that every fine sheaf is soft.
3. Let OX be a soft sheaf of (not necessarily commutative) rings. Show that every OX -
module is fine.
4. Let A be a fine sheaf of abelian groups on X. Show that for every sheaf M of abelian
groups on X the tensor product (Problem 8.7) A ˝ZX M is fine.
Problem 9.7. Let M be a real C ˛ -manifold with ˛ 1 and let p 2 M . Show that
forming the germ OM .M / ! OM;p is surjective.
Problem 9.8. Let M be a real C ˛ -manifold with ˛ 1. Show that there exists a proper
morphism f W M ! R.
Problem 9.9. A commutative monoid M (written additively) is called integral if for all
m 2 M the map M ! M , n 7! m C n is injective. The integral monoids together with
homomorphisms of monoids form a category and one has the notion of a sheaf of integral
monoids.
1. Show that M is integral if and only if the canonical homomorphism M ! M gp (Ap-

pendix Problem 13.22) is injective.
2. Generalize Proposition 9.9 to soft sheaves A of integral monoids.
3. Let M˚ be a real paracompact Hausdorff C ˛ -premanifold with ˛ 1. Show that
U 7! f 2 CM ˛
.U / I f .p/ 0 8 p 2 U is a soft sheaf of integral monoids. Deduce
that in Corollary 9.13 one can assume that fi 0 for all i.
204 9 Soft Sheaves
Problem 9.10. Let U be a tautological line bundle on P 1 .R/ D Grass1 .R2 / (Prob-
lem 8.28) and let pW S 1 ! P 1 .R/ be the canonical map. Show that the pullback p .U /
(Problem 8.17) is isomorphic to the Möbius bundle.
Problem 9.11. Let X be a paracompact Hausdorff space, let F be a presheaf of sets on

X and FQ its sheafification. Suppose that for every locally finite open covering .Ui /i of X
and for all si 2 F.Ui / with si jUi \Uj D sj jUi \Uj for all i, j there exists s 2 F.X/ such that
s jUi D si for all i. Show that F.X/ ! F.X/Q is surjective.
Problem 9.12. Let be a topological space. For every topological space X let S .X/
be the set of continuous maps ! X. Let A be an abelian group and let S .X; A/ be
the abelian group of maps S .X/ ! A, called the group of singular -cochains of X
with values in A. If U V X are open subspaces, one has an injection S .U / ,!
S .V / and hence restriction maps S .V; A/ ! S .U; A/ making U 7! S .U; A/ into
a presheaf of abelian groups. Let U 7! S .U; A/ be its sheafification.
1. Deduce from Problem 9.11 that S .X; A/ ! S .X; A/ is surjective.

2. Show that S .X; A/ is a fine sheaf (Problem 9.6).
Problem 9.13. Consider the smooth manifold P n .R/, n 1. Show that Pic.P n .R// Š
Z=2Z.
Hint: Problem 7.6.
Cohomology of Complexes of Sheaves
10
In this chapter we generalize for sheaves of abelian groups (or more generally, for OX -
modules, where .X; OX / is a ringed space) cohomology groups in degree 1 to higher
degrees in such a way that we can extend the exact sequence of cohomology. The idea
(motivated in Sect. 10.1) is to replace an OX -module by a quasi-isomorphic complex
of such OX -modules whose cohomology we expect to be trivial (namely injective OX -
modules), then to apply the functor of global sections to this complex to obtain a complex
of .X; OX /-modules, and finally take the cohomology of this complex. Of course, once
one follows the strategy of first replacing an OX -module by a complex, it is only natural to
start with an arbitrary complex of OX -modules1 . This will yield the (hyper-)cohomology
H p .X; F / of a complex F defined in Sect. 10.2. Much more generally, one may view
these constructions as a special case of the right derivation of a left exact functor between
abelian categories, which is here applied to the functor of taking global sections of an
OX -module.
In Sect. 10.3 we will study classes of acyclic OX -modules F (i.e., H p .X; F/ D 0 for
all p 1). Injective OX -modules are acyclic, almost by definition. We will introduce
flabby sheaves as a technical tool and we will also prove that soft sheaves are acyclic.
The cohomological techniques developed in the first three sections will allow us in
Sect. 10.4 to define de Rham cohomology and to prove the de Rham isomorphism. We
can also compute some cohomology groups of sheaves of holomorphic functions and get
as a corollary the theorem of Mittag-Leffler from complex analysis in one variable.
After studying some functoriality properties of cohomology with respect to inverse
and direct images in Sect. 10.5 we formulate and prove in the last section the proper base
change theorem for bounded below complexes of abelian sheaves on arbitrary topological
spaces.
1
Even more natural would be to work systematically in a category of complexes, where quasi-
isomorphisms are defined to be isomorphisms, i.e., in the derived category. But the required amount
of homological algebra is beyond the scope of this book.

DOI 10.1007/978-3-658-10633-1_10
206 10 Cohomology of Complexes of Sheaves
Notation: In this chapter .X; OX / will denote a ringed space and we set R WD OX .X/.
If K is a commutative ring such that .X; OX / is K-ringed, then R is a K-algebra.
Recall that the category of OX -modules and the category of R-modules are abelian
(Appendix Definition 15.25).
10.1 Strategy for the Definition of Cohomology of Sheaves
Consider the additive functor (Appendix Definition 15.21)
.X; /W (OX -Mod) ! (R-Mod); F ! .X; F/ WD F.X/:
It is left exact, i.e., it commutes with finite limits (Remark 8.37). Equivalently (by Ap-
pendix Remark 15.27), for every exact sequence
0 ! F0 ! F ! F00
of OX -modules the induced sequence
0 ! .X; F0 / ! .X; F/ ! .X; F00 /
is exact.
Remark and Definition 10.1 (Goal of the construction of cohomology). We would like
to construct a so-called ı-functor extending .X; /, i.e.,
(a) a family of additive functors H n .X; /W (OX -Mod) ! (R-Mod) for n 0 such that
H 0 .X; / D .X; /,
(b) for every exact sequence 0 ! F0 ! F ! F00 ! 0 of OX -modules a family of
homomorphisms of R-modules ıW H n .X; F00 / ! H nC1 .X; F0 /, n 0,
satisfying the following conditions:
1. For every exact sequence 0 ! F0 ! F ! F00 ! 0 of OX -modules the sequence
0 ! H 0 .X; F0 / ! H 0 .X; F/ ! H 0 .X; F00 /

ı
! H 1 .X; F0 / ! H 1 .X; F/ ! H 1 .X; F00 /
ı
! H 2 .X; F0 / ! H 2 .X; F/ ! : : :
is exact.
10.1 Strategy for the Definition of Cohomology of Sheaves 207
2. For every morphism .0 ! F0 ! F ! F00 ! 0/ ! .0 ! G0 ! G ! G00 ! 0/ of

exact sequences of OX -modules the diagram
ı
H n .X; F00 / H nC1 .X; F0 /
ı
H n .X; G00 / H nC1 .X; G0 /
is commutative.
Moreover, to characterize the cohomology up to unique isomorphism, we would like

this ı-functor .H n .X; : : : /; ı/n to be universal in the following sense: If .F n ; ı/n0 is
any ı-functor extending .X; / then there is a unique family of morphisms of functors
' n W H n.X; / ! F n compatible with ı such that ' 0 D id.
Clearly, a universal ı-functor extending .X; / is unique up to unique isomorphism
(once we have shown that it exists). It is then called the right derived functor of .X; /.
Remark and Definition 10.2 (Strategy of the construction of cohomology). To get an

idea of how to construct such a universal ı-functor extending .X; / we think backwards.
Assume that we have already constructed the right derived functors H n .X; / of .X; /.
An OX -module I is called -acyclic if H n .X; I/ D 0 for all n > 0.
Now let F be an arbitrary OX -module. Assume:
There exists an exact sequence of OX -modules

0 ! F ! I0 ! I1 ! : : : (~)
such that In is -acyclic for all n 0:
In other words, F is quasi-isomorphic to the complex I of -acyclic OX -modules.

Claim: The cohomology modules can be calculated as follows:
H n .X; F/ D H n 0 ! .X; I0 / ! .X; I1 / ! .X; I2 / ! : : : ; (10.1)
where the right-hand side denotes the cohomology of a complex (Appendix Defini-
tion 15.2), i.e., the right-hand side is equal to
Rn WD Ker. .X; In / ! .X; InC1 //= Im. .X; In1 / ! .X; In //:
Let us believe this claim for a moment. Then this suggests a way of defining right derived
functors H n .X; : : : /:
1. Find a class I of OX -modules that we expect to be always -acyclic once we have

defined .H n .X; : : : /; ı/n0 .
2. Show that (~) is possible for every OX -module F with In in I .

3. Then define H n .X; F/ via (10.1).
As exact sequences of the form 0 ! I ! F ! F00 ! 0 with I injective OX -module

always split, it is plausible to take all injective OX -modules as this class. Then (~) will be
always possible by Lemma 10.3 below.
Let us show the claim. Denote the right-hand side of (10.1) by Rn . Let n D 0.
Applying the left exact functor .X; / to the exact sequence 0 ! F ! I0 ! I1 we
get the exact sequence
0 ! .X; F/ ! .X; I0 / ! .X; I1 /
and hence
H 0 .X; F/ D .X; F/ D Ker. .X; I0 / ! .X; I1 // D R0 :
Assume now that n 1. Set I1 WD K0 WD F and define for p 1
Kp WD Coker.Ip2 ! Ip1 / Š Im.Ip1 ! Ip / D Ker.Ip ! IpC1 /;
where the isomorphism is induced by Ip1 ! Ip . From the exact sequence 0 ! Kp !

Ip ! IpC1 we get
H 0 .X; Kp / D Ker.H 0 .X; Ip / ! H 0 .X; IpC1 //: (*)
The short exact sequence
0 ! Kp ! Ip ! KpC1 ! 0
yields a long exact sequence
0 ! H 0 .X; Kp / ! H 0 .X; Ip / ! H 0 .X; KpC1 /

! : : :
! H n .X; Kp / ! H n .X; Ip / ! H n .X; KpC1 /
„ ƒ‚ …
D0
! H nC1 .X; Kp / ! H nC1 .X; Ip / ! : : :

„ ƒ‚ …
D0
and hence isomorphisms H n .X; KpC1 / ! H nC1 .X; Kp / for n 1. As F D K0 we see

by induction that
H n .X; F/ D H n .X; K0 / Š Š H 1 .X; Kn1 /

. /
Š H 0 .X; Kn /= Im.H 0 .X; In1 / ! H 0 .X; Kn // D Rn :
10.2 Definition of Cohomology 209
10.2 Definition of Cohomology
The strategy outlined above suggests how to define cohomology of an OX -module F as

follows:
qi s
1. Choose a quasi-isomorphism F ! I , where I is a complex consisting of injective
OX -module Ip for all p with Ip D 0 for p < 0.
2. Apply the functor .X; / to I and take the cohomology of complexes of !
.X; Ip / ! .X; IpC1 / ! : : : .
But if we replace a single OX -module F by a complex, why not start with a whole complex
of OX -modules F ? Then it will turn out that the crucial property of I is its K-injectivity
(Appendix Definition 15.18). Hence the starting point of the construction of cohomology
of arbitrary complexes of OX -modules is the following result, which ensures that step (1)
is always possible.
Lemma 10.3. The following abelian categories have injective and K-injective resolutions
(Appendix Definition 15.28):
1. The abelian category of R-modules for a ring R.

2. The abelian category of OX -modules for a ringed space .X; OX /.
We will not prove this lemma here but refer to [Spa] Theorem 4.5, [Stacks] Tag 079P,
or [AJS] Theorem 5.4.
Lemma 10.3 in particular shows that if F is bounded below, then we find a quasi-
qi s
isomorphism F ! I , where I is bounded below and consists of injective OX -modules
(in particular I is K-injective by Appendix Proposition 15.19). This is the only case
that we will apply in the rest of this book (and we refer to Appendix Problem 15.14
and Problem 10.1 for some hints on the existence of such injective resolutions). Hence
in the sequel we usually give proofs only in this case or even only in the case where
F is concentrated in degree 0 and refer to the literature for the general arguments from
homological algebra.
Now we define the sheaf cohomology of a complex of OX -modules as follows.
Definition and Theorem 10.4. Let .X; OX / be a ringed space and let R WD .X; OX /.
qi s
Let F be a complex of OX -modules. Choose a quasi-isomorphism F ! I , where I is
a K-injective complex of OX -modules and define for n 2 Z the n-th cohomology of F by
H n .X; F / WD H n . ! .X; Ip1 / ! .X; Ip / ! .X; IpC1 / ! : : : /:
1. This defines for all n 2 Z additive functors
H n .X; /W K.OX / ! (R-Mod)

such that for an OX -module F (considered as a complex concentrated in degree 0) one

has a functorial isomorphism
H 0 .X; F/ ! .X; F/:
If the complex F is bounded below, say there exists a 2 Z such that Fp D 0 for all
p < a, then H n .X; F / D 0 for all n < a.
2. For every short exact sequence of complexes of OX -modules
u v
0 ! F ! G ! H ! 0 (10.2)
up vp
(i.e., 0 ! Fp ! Gp ! Hp ! 0 is an exact sequence of OX -modules for all p 2 Z)
there are connecting homomorphisms of R-modules ıW H n.X; H / ! H nC1 .X; F /
making the sequence
:::
ı H n .X;u/ H n .X;v/
! H n .X; F / ! H n .X; G / ! H n .X; H /
ı H nC1 .X;u/ H nC1 .X;v/ (10.3)
! H nC1 .X; F / ! H nC1 .X; G / ! H nC1 .X; H /
ı
! : : :
exact and this long exact cohomology sequence is functorial for morphisms of short
exact sequences of complexes.
If Z X is any subspace, we simply write H n .Z; F / instead of H n .Z; F jZ /.

One should not confuse cohomology as a complex H p .F / defined as the OX -module
Ker.Fp ! FpC1 /= Im.Fp1 ! Fp / and the cohomology H n .X; F / just defined, which
is an R-module. To distinguish these two notions, H n .X; F / is sometimes also called the
hypercohomology of the complex, but we will not use this terminology.
For a sheaf of abelian groups F, which is the same as a ZX -module, we now have for
the moment two definitions of H 1 .X; F/: the definition given here and the definition via
torsors (or Čech cocycles) given in Chap. 7. We will see in Proposition 10.16 that these
definitions coincide.
Before proving Theorem 10.4 we make the following remark.
Remark 10.5. Let uW F ! G be a morphism of complexes of OX -modules. Let

.X; F / be the complex of R-modules
: : : ! .X; Fp / ! .X; FpC1 / ! : : :
and let .X; u/W .X; F / ! .X; G / be the morphism of complexes of R-modules
given by .X; u/p WD .X; up /W .X; Fp / ! .X; Gp /. This defines a functor
(Com(OX )) ! (Com(R)). Applying to a homotopy u ' v of homomorphisms of
complexes (Appendix Definition 15.4 and Remark 15.26) the functor .X; / yields a ho-
motopy .X; u/ ' .X; v/. Hence .X; / induces a functor of homotopy categories
(Appendix Definition 15.6 and Remark 15.26)
.X; /W K.OX / ! K.R/:
In particular, H n . .X; u// D H n . .X; v// for homotopic morphisms u and v of com-
plexes of OX -modules for all n 2 Z.
Proof (Proof of Theorem 10.4). (i). We first show that H n .X; F / depends up to unique
qi s
isomorphism not on the choice of the K-injective resolution. Let sW F ! I and
qi s
tW F ! J be two K-injective resolutions. Applying Appendix Lemma 15.20 to u D
idW F ! F we obtain a unique isomorphism wW I ! J in K.OX / such that w ı s D t
and hence a unique isomorphism
H n . .w//W H n . .X; I // ! H n . .X; J //
by Remark 10.5.
(ii). H n .X; / defines an additive functor H n .X; /W K.OX / ! (R-Mod), again by
Appendix Lemma 15.20 and Remark 10.5.
(iii). Assume that Fp D 0 for all p < a. Then we may choose a K-injective res-
qi s
olution F ! I , where Ip D 0 for all p < a (Lemma 10.3). Hence H n .X; F / D
H n . .X; I // D 0 for all n < a.
qi s
(iv). Let F be an OX -module. Then a K-injective resolution F ! I with Ip D 0 for
p < 0 is an exact sequence
: : : ! 0 ! F ! I0 ! I1 ! : : : :
As .X; / is left exact, we obtain
.X; F/ D Ker. .X; I0 / ! .X; I1 // D H 0 . .X; I //:
(v). Proof of (2): If F , G , and H are bounded below: We claim that there exists
a commutative diagram in (Com(OX ))
0 F G H 0
(*)
0 I J K 0;
such that:
1. The complexes in the lower row are bounded below and consist of injective OX -
modules and the vertical arrows are quasi-isomorphisms.
2. The lower row is an exact sequence of complexes.

qi s
Let us briefly indicate the argument: By Lemma 10.3 we can find F ! I as desired.
Forming the componentwise pushout in the abelian category (OX -Mod), we obtain a di-
agram as desired, but where J and K are arbitrary bounded below complexes. As I
consists componentwise of injective objects, the sequence 0 ! I ! J ! K ! 0 is
componentwise split. Hence we can assume that (10.2) is componentwise split and that the
components of A are injective (see also Appendix Problem 15.13). Then choose an injec-
qi s
tive resolution H ! K such that Hp ! Kp is injective for all p. Set Jp WD Ap ˚ Kp .
Then it is easy to see that there are homomorphisms ı p W Kp ! ApC1 such that
!
p
p dA ıp
dJ WD p
0 dK
makes J into a complex and such that if we define Gp ! Jp as the sum of idAp and
Hp ! Kp , then (*) commutes.
As Ip is injective, the exact sequences 0 ! Ip ! Jp ! Kp ! 0 even split. Hence this
sequence stays exact after applying .X; /. We obtain an exact sequence of complexes
of R-modules
0 ! .I / ! .J / ! .K / ! 0:
Now we can apply Appendix Lemma 15.14 to obtain (10.3).
(vi). For the proof of (2) in general we refer to [Stacks] Tag 0152.
Remark 10.6. Let F and G be complexes of OX -modules. Then every quasi-iso-

morphism uW F ! G yields an isomorphism
H n .X; F / ! H n .X; G /:
Indeed, if iW G ! I is a K-injective resolution of G , then i ı u is a K-injective resolution

of F .
To achieve all goals formulated in Remark 10.1 we still have to show that if we restrict
our construction of cohomology to a single OX -module, then we obtain a universal ı-
functor. To make sense of this we will always consider an OX -module F as the complex
of OX -modules
! 0 ! 0 ! F ! 0 ! 0 ! :::;
where F sits in degree 0 of the complex. Then we write H n .X; F/ for its cohomology.
Corollary 10.7. Restricting the functors H n .X; : : : / to the category of OX -modules con-
sidered as complexes concentrated in degree 0 the family .H n .X; /; ı/n0 is a universal
ı-functor extending .X; /. Moreover, every injective OX -module I is -acyclic (i.e.,
H n .X; I/ D 0 for all n > 0).
Proof. Theorem 10.4 immediately implies that .H n .X; /; ı/n0 is a ı-functor that ex-
tends .X; /.
Let I be an injective OX -module. If we consider I as a complex concentrated in degree
0, then idW I ! I is a K-injective resolution. Hence the definition of H n .X; I/ shows that
H n .X; I/ D 0 for all n > 0.
It remains to show the universality of the ı-functor. Let .T n ; ı 0 /n0 be a ı-functor
extending .X; /. We have to show that for all n 0 and for all OX -modules F there exist
unique homomorphisms 'Fn W H n .X; F/ ! T n .F/ compatible with ı and ı 0 and functorial
in F such that 'F0 D id .X;F/ . We proceed by induction on n.
The assertion is clear for n D 0. Let n > 0. Choose an injective homomorphism
iW F ! I with I an injective OX -module and set Q WD Coker.i/. We obtain a short exact
sequence of OX -modules
0!F!I!Q!0
and hence by induction hypothesis a commutative diagram
ı
::: H n1 .X; I/ H n1 .X; Q/ H n .X; F/ H n .X; I/ D 0
'In1 n1
'Q
ı0
::: T n1 .I/ T n1 .Q/ T n .F/
with exact rows. Hence ı is surjective and Ker.ı/ Ker.ı 0 ı 'Q n1
/. Therefore there
exists a unique 'F W H .X; F/ ! T .F/ making the diagram commutative. It is straight
n n n
forward – albeit a bit cumbersome – to check that 'Fn is functorial in F.
For concrete calculations working with injective resolutions is often not advisable because
injective modules tend to be quite large (e.g., non-zero injective abelian groups are never
finitely generated Z-modules by Appendix Problem 15.14). But in fact the strategy of
Remark 10.2 required only to choose a resolution by a complex of -acyclic modules:
qi s
Proposition 10.8. Let F be a bounded below complex of OX -modules and let F ! A
be a quasi-isomorphism, where A is a bounded below complex of OX -modules such that
An is -acyclic for all n 2 Z. Then
H n .X; F / D H n . ! .X; Ap / ! .X; ApC1 / ! : : : /:
In the next section we will see several examples of -acyclic OX -modules.
Proof. If F is concentrated in degree 0 and An D 0 for all n < 0, we have seen the result
in (10.1). For the general case see [Stacks] Tag 05TA.
All of the results above can be vastly generalized with (almost) verbatim the same proofs.
Instead of the functor .X; /W (OX -Mod) ! ( .X; OX /-Mod) one considers an arbitrary
left exact functor between abelian categories
F W A ! B:
Then all of the above notions, statements, and proofs generalize – if(!) A has injective
and K-injective resolutions. One obtains a universal ı-functor from A to B extending
F , which is called the right derived functor of F and which is denoted by Rn F (in this
notation we would have H n .X; / D Rn .X; /). More generally one obtains functors
Rn F W K.A/ ! B with all the properties above. And one has Rn F .I / D 0 for all injective
objects I in A and all n > 0.
Examples are the following.
1. Let f W .X; OX / ! .Y; OY / be a morphism of ringed spaces. The left exactness of

the functors .f 1 .V /; /, V Y open, shows that the functor f W (OX -Mod) !
(OY -Mod) is left exact. We obtain right derived functors Rn f .
This is in fact a generalization of the cohomology functors H n .X; / above. Taking
for Y the one point space with OY .Y / D .X; OX / one has Rn f D H n .X; / by
identifying the category of OY -modules and the category of R-modules.
2. For any abelian category A and for every object Z in A the functor
A ! (Ab); X 7! HomA .Z; X/
is left exact. Hence if A has injective and K-injective resolutions, then there exist the
right derived functors X 7! ExtnA .Z; X/ WD Rn HomA .Z; X/.
Finally, there exists a dual theory for right exact functors F by replacing all arrows in all
diagrams by an arrow in the opposite direction: The dual notion of an injective object and
a K-injective complex is called a projective object (see Appendix Problem 14.4) and a K-
projective complex, respectively. One obtains left derived functors denoted by Ln F . Note
however that if an abelian category has injective and K-injective (right) resolutions, then
the dual assertion, namely that there always exist projective and K-projective (left) resolu-
tions, might not be true. For instance, the category of R-modules (for any not necessarily
commutative ring R) also has projective and K-projective resolutions ([Spa] Theorem C)
but this is usually not the case for the category of OX -modules for a ringed space .X; OX /.
Often it is possible to circumvent this problem by working solely with acyclic resolutions.
10.3 Acyclic Sheaves 215
10.3 Acyclic Sheaves
We continue to denote by .X; OX / a ringed space. Proposition 10.8 shows that we can
define the right derived functor of a left exact functor F of a bounded below complex also
via a resolution with F -acyclic objects. The following lemma will be the main tool to see
whether a certain class of objects is F -acyclic.
Lemma 10.9. Let F W A ! B be a left exact functor between abelian categories and
suppose that A has injective and K-injective resolutions. Let I Ob.A/ such that the
following conditions are satisfied:
(a) For every object X of A there exists an injective morphism X ,! M with M 2 I .

(b) If I is an injective object of A, then I 2 I .
(c) Let
0 ! X 0 ! X ! X 00 ! 0
be an exact sequence in A with X 0 ; X 2 I . Then one has X 00 2 I and F .X/ !

F .X 00 / is surjective.
Then every object in I is F -acyclic, i.e., Rn F .I / D 0 for all I 2 I and n 1.
Proof. Let M 2 I and choose an exact sequence

d0 d1 d2
0 ! M ! I 0 ! I 1 ! I 2 ! : : :
with I p injective for all p. As I 0 2 I by (b), we can apply (c) to the exact sequence 0 !
M ! I 0 ! Im.d 0 / ! 0. This shows that Im.d 0 / 2 I and that 0 ! F .M / ! F .I 0 / !
F .Im.d 0 // ! 0 is exact. Then induction and the exact sequences 0 ! Im.d p1 / !
I p ! Im.d p / ! 0 show that Im.d p / 2 I for all p and that 0 ! F .Im.d p1 // !
F .I p / ! F .Im.d p // ! 0 is exact. For n 1 we find therefore
Im.F .I n1 / ! F .I n // D F .Im.d n1 // D Ker.F .I n / ! F .I nC1 //
and hence Rn F .M / D 0.
The following class of sheaves will yield -acyclic objects.
Definition 10.10. A sheaf F on a topological space X is called flabby or flasque if the

restriction maps F.X/ ! F.U / are surjective for all open subsets U X.
Flabby sheaves will allow construction of -acyclic resolutions in an easy functorial

way as follows.
Remark 10.11 (Godement resolution). Let F be a sheaf on a topological space. Define

Q
a sheaf FŒ0 on X by FŒ0 .U / WD x2U Fx for U X open, where the restriction maps
are given by the projections. The sheaf FŒ0 is flabby and there is an injective morphism
of sheaves
F W F ,! FŒ0 ; F.U / 3 s 7! .sx /x2U 2 FŒ0 .U /; U X open:
Every morphism 'W F ! G of sheaves induces a morphism of flabby sheaves ' Œ0 W FŒ0 !
Œ0 Q
GŒ0 with 'U D x2U 'x . We obtain a functor . /Œ0 from the category of sheaves to the
full subcategory of flabby sheaves. The morphism F is functorial in F.
Now suppose that F is an OX -module. Then FŒ0 is an OX -module: addition is given
by the addition within the stalks and scalar multiplication of a 2 OX .U / on .sx /x2U 2
FŒ0 .U / by .ax sx /x2U . We obtain a functor . /Œ0 from the category of OX -modules to the
full subcategory of flabby OX -modules. The morphism F is a functorial homomorphism
of OX -modules.
In particular we find for every OX -module a functorial exact sequence of OX -modules

0 ! F ! FŒ0 ! Coker./ ! 0 (10.4)
with FŒ0 flabby. Applying the same argument to Coker./ we obtain inductively a functo-
rial exact sequence of OX -modules

0 ! F ! FŒ0 ! FŒ1 ! : : : ; (10.5)
where FŒp is a flabby OX -module for all p 0. This resolution of F is called Godement
resolution.
Finally, the variant of Appendix Lemma 15.11 for the abelian category of OX -modules
shows that given a 2 Z there exists for every complex F of OX -modules with Fp D 0
qi s
for all p < a a quasi-isomorphism F ! A , where Ap is a flabby OX -module for all p
and Ap D 0 for all p < a.
Proposition 10.12.
1. Let .X; OX / be a ringed space. Then every injective OX -module is flabby.

2. Let X be a paracompact Hausdorff space. Then every flabby sheaf F on X is soft.
Proof. (1). Let I be an injective OX -module. By Remark 10.11 there exists an injective
homomorphism of OX -module iW I ! G where G is flabby. As I is injective, this makes I
into a direct summand of G. Hence I is flabby.
(2). Let A X be closed, s 2 F.A/. By Proposition 9.1 there exists an open
neighborhood U of A and sQ 2 F.U / extending s. As F is flabby, one can extend sQ to X.

10.3 Acyclic Sheaves 217
Proposition 10.13. Let X be a topological space and let F be a flabby sheaf on X.
1. For every open subset U of X the restriction FjU is again flabby.

2. Suppose that X is hereditarily paracompact and Hausdorff. Then for any subspace Z
of X the restriction FjZ is flabby again.
Recall Proposition 1.13, which shows that any metrizable space satisfies the hypotheses
of 2.
Proof. The first assertion is clear. To show 2 let W Z be open and let s 2 F.W /. By
Proposition 9.1 2 we can extend s to a section s 0 over an open neighborhood U of W in
X. As F is flabby, we can extend s 0 to a section sQ 2 F.X/. Then sQ jZ 2 F.Z/ extends s.

Lemma 10.14. Let X be a topological space.
1. Let G be a flabby sheaf of groups. Then every G-torsor is trivial.

2. Let 1 ! G0 ! G ! G00 ! 1 be an exact sequence of sheaves of groups and let G0 and
G be flabby. Then G00 is flabby and G.U / ! G00 .U / is surjective for all U X open.
Proof. 1. The argument is very similar to the proof that torsors for soft sheaves are trivial
(Theorem 9.14) but as we can work with open sets we do not run into any topological
difficulties requiring such tools as the shrinking lemma. Indeed, let T be a G-torsor, let
.Ui /i 2I be an open covering such that there exist ti 2 T .Ui / for all i. For J I set
S
UJ WD i 2J Ui . Then UI D X. Define E WD f .t; J / I J I; t 2 T .UJ / g. Then E ¤ ;
because . ; ;/ 2 E . It is partially ordered by .t; J / .t 0 ; J 0 / if J J 0 and t 0 jUJ D t.
As T is a sheaf, every totally ordered subset of E has an upper bound in E . Hence there
exists a maximal element .t; J / in E by Zorn’s lemma. It suffices to show that J D I .
Assume there exists i 2 I n J and let g 2 G.UJ \ Ui / with t jUJ \Ui D gti jUJ \Ui .
As G is flabby, we can extend g to gQ 2 G.X/. Replacing ti by gQ jUi ti we may assume
t jUJ \Ui D ti jUJ \Ui and hence we can glue t and ti to a section over UJ [fi g. This contradicts
the maximality of .t; J /.
2. As G0 jU is flabby for every open subset U X, H 1 .U; G0 jU / D 1 by 1 (here we use
the cohomology defined in Definition 7.3) and the cohomology sequence attached to the
exact sequence 1 ! G0 jU ! GjU ! G00 jU ! 1 shows that G.U / ! G00 .U / is surjective.
Hence we obtain a commutative diagram
G.X/ G00 .X/
G.U / G00 .U /
with G.X/ ! G.U / ! G00 .U / a composition of surjective maps. Therefore G00 .X/ !
G00 .U / is surjective. This shows that G00 is flabby.
Proposition 10.15. A flabby OX -module is -acyclic.
Proof. This follows from Lemma 10.9 using Remark 10.11 and Lemma 10.14.
Proposition 10.16. Let X be a topological space and let A be a sheaf of abelian groups,
which we may consider a ZX -module. Then H 1 .X; A/ as defined in Definition 10.4 and
H 1 .X; A/ defined via torsors in Definition 7.3 coincide.
The proof will show that there is a functorial isomorphism between these cohomology
groups.
Proof. Denote for the moment the cohomology group defined via torsors in Definition 7.3
by HQ 1 .X; /. By Remark 10.11 there exists a functorial exact sequence of sheaves of
abelian groups 0 ! A ! A0 ! C ! 0 with A0 flabby. Then H 1 .X; A0 / D 0
by Proposition 10.15 and HQ 1 .X; A0 / D 0 by Lemma 10.14 1. Hence the long ex-
act cohomology sequence (10.3) and the long exact sequence for HQ 1 (7.11) imply that
H 1 .X; A/ D HQ 1 .X; A/ D Coker.A0 .X/ ! C.X//.
Proposition 10.17. Let .X; OX / be a ringed space such that X is paracompact and Haus-
dorff. Then every soft OX -module is -acyclic.
Proof. We check the conditions of Lemma 10.9 for the class I of soft OX -modules. By
Remark 10.11 we can embed every OX -module into a flabby OX -module A, which is soft
by Proposition 10.12. The same proposition also shows that every injective OX -module is
soft.
It remains to show Condition (c) of Lemma 10.9. Let 0 ! F0 ! F ! F00 ! 0 be
an exact sequence of OX -modules with F and F0 soft. Then F.X/ ! F00 .X/ is surjective
because H 1 .X; F0 / D 0 by Theorem 9.14. Let A X be closed. Then F0 jA is soft by
Remark 9.5 and therefore the same argument proves that F.A/ ! F00 .A/ is surjective.
Then the commutative diagram
F.X/ F00 .X/
F.A/ F00 .A/
shows that F00 .X/ ! F00 .A/ is surjective. Therefore F00 is soft.
Every complex of OX -modules F is also a complex of sheaves of abelian groups (i.e.,

a complex of ZX -modules), which we call a F for the moment. Similarly, if M is a module
over a ring R, we denote by a M the underlying additive group.
10.4 Applications: Theorems of De Rham and of Mittag-Leffler 219
Proposition 10.18. Let F be a bounded below complex of OX -modules. Then we have
H n .X; a F / D a H n .X; F /:
qi s
Proof. By Remark 10.11 there exists a quasi-isomorphism F ! A of OX -module,
where Ap is flabby for all p. By Proposition 10.8, .X; A / computes the cohomology
of F . By Lemma 10.15 each Ap is also -acyclic if considered as a ZX -module. Hence
it also computes the cohomology of a F .
10.4 Applications: Theorems of De Rham and of Mittag-Leffler
Definition and Remark 10.19 (De Rham Cohomology). Let M be a real C ˛ -pre-
manifold with ˛ 1 or a complex premanifold. Then
d d d
˝M W ! 0 ! ˝M
0
! ˝M
1
! ˝M
2
! : : :
is a bounded below complex of KM -modules that is exact except in degree 0 (Corol-

lary 8.69). For n 0 we call
n
HDR .M / WD H n .M; ˝M /
the n-th de Rham cohomology of M . The quasi-isomorphism (8.36)

qi s
KM ! ˝M
yields by Remark 10.6 for all n 0 an isomorphism
H n .M; KM / ! HDR
n
.M /: (10.6)
p
If ˝M is -acyclic for all p 0, it calculates its own cohomology by Proposition 10.8
and we obtain
fclosed n-forms on M g
HDRn
.M / D : (10.7)
fexact n-forms on M g
If M is of dimension m 2 N0 , this implies in particular HDR
n
.M / D 0 for n > m.
If M is a real C 1 -manifold, then ˝M is soft for all p 0 (Corollary 9.12) and hence
p
-acyclic (Proposition 10.17). Hence we obtain:

Theorem 10.20 (De Rham). Let M be a real C 1 -manifold. Then for all n 0
fclosed n-forms on M g
H n .M; RM / Š HDR
n
.M / Š : (10.8)
Note that this result in particular implies the surprising fact that the question of whether
every closed n-form is exact on M depends only on the underlying topological space of
M . We will even see in Corollary 11.19 below that H n .M; RM / D H n .M 0 ; RM 0 / if M
and M 0 are homotopy equivalent2 .
p
Remark 10.21. One can show that ˝M is also -acyclic for all p 0 if M is a real
analytic manifold (combine [Car] Théoreme 3 together with the fact that for every real
analytic manifold there exists a closed embedding into some RN by [Gra]). Hence one
has the isomorphisms (10.8) also for real analytic manifolds M .
For complex manifolds M one still has that H n .M; CM / D HDR n

.M / by the Poincaré
lemma (Remark 10.19) but (10.7) does not hold in general (see Problem 10.6 for M D
P 1 .C/). Still, there are interesting special cases where one still has an analogue of (10.8).
One of these cases relies on the following result.
Theorem 10.22. Let U C be open and OU be the sheaf of holomorphic functions.

Then H i .U; OU / D 0 for all i > 0.
We use the following standard fact from complex analysis (the “Dolbeault lemma”):
For every C-valued C 1 -function g on U there exists a C-valued C 1 -function f on U
such that @f
@zN
D g. For a proof see [Hoe] Theorem 1.4.4 in the case considered here or
[GuRo] Chap. VI, C, Lemma 1 for a version in an arbitrary complex dimension).
Proof. Let CU1IC be the sheaf of C-valued C 1 -functions on U . Applying the Dolbeault
lemma to open subsets of U we see that @@zN W CU1IC ! CU1IC is surjective. By the Cauchy–
Riemann differential equations we obtain an exact sequence
@
@zN
0 ! OU ! CU1IC ! CU1IC ! 0:
Hence OU has a resolution by the complex

@
@zN
! 0 ! CU1IC ! CU1IC ! 0 ! : : : ;
„ƒ‚… „ƒ‚…
degree 0 degree 1
2
This is formulated in a somewhat too dramatic way: The Poincaré lemma (Theorem 8.68), for
which we did not give a proof, is in the C 1 -case often shown (for instance in the given reference) by
proving that smooth maps between premanifolds that are homotopic via a smooth homotopy induce
the same map on the right-hand side of (10.8). One then concludes that on a smoothly contractible
C 1 -manifold every closed form is exact. Nevertheless it remains surprising that continuity of the
homotopy is already sufficient.
10.4 Applications: Theorems of De Rham and of Mittag-Leffler 221
which consists of soft (and hence -acyclic) sheaves. Hence the cohomology is calculated
by this complex. Clearly this implies H n .U; OU / D 0 for n 2. Moreover, the Dolbeault
lemma (now applied to U ) shows that H 1 .U; OU / D 0.
Corollary 10.23. Let U C be open. Then for all n 0
fclosed holomorphic n-forms on U g

H n .U; CU / Š HDR
n
.U / Š : (10.9)
fexact holomorphic n-forms on U g
As U has dimension 1 as complex manifold, ˝Un D 0 for n > 1. In particular, all terms
in (10.9) are 0 for n > 1.
Proof. By Theorem 10.22, OU D ˝U0 is -acyclic. Denote by z the coordinate on U .

Then f 7! f dz defines an isomorphism OU ! ˝U1 of OU -modules, hence ˝U1 is -
acyclic as well. We conclude by Remark 10.19.
Much more generally, one has for every Stein manifold (Theorem 5.45) X that
H i .X; F/ D 0 for all i > 0 and for all finite locally free OX -modules F (a special
case of “Cartan’s Theorem B”, e.g., [GuRo], Chap. VIII, A, Theorem 14). Hence Corol-
lary 10.23 holds if U is an arbitrary Stein manifold.
Corollary 10.24 (Theorem of Mittag-Leffler). Let U C be open and let S U be

discrete and closed. For each z0 2 S let there be given a finite principal part of a Laurent
series
Xr
ai .z z0 /i (*)
i D1
with ai 2 C and r depending on z0 .

Then there exists a meromorphic function f on U such that f is holomorphic on U n S
and such that for all z0 2 S the principal part of the Laurent series expansion at z0 is given
by (*).
Proof. Let PU be the sheaf on U of functions sW V ! C .N/ , V U open, such that

f z0 2 V I s.z0 / ¤ 0 g is discrete and closed in V . We think of such a function s as a map
P
that attaches to z0 the principal part i 2N ai .z z0 /i if s.z0 / D .a1 ; a2 ; a3 ; : : : /.
Let MU be the sheaf of meromorphic functions on U . We obtain an exact sequence of
sheaves of C-vector spaces
principal part
0 ! OU ! MU ! PU ! 0:
As H 1 .U; OU / D 0 by Theorem 10.22, the map MU .U / ! PU .U / is surjective.

10.5 Cohomology and Inverse and Direct Image
Remark 10.25. Recall that we defined a functor f 1 W (Sh(Y )) ! (Sh(X)) such that
.f 1 G/x D Gf .x/ (10.10)
for every sheaf G on Y and x 2 X. More precisely, f 1 G was defined as the sheafification
of the presheaf on X
.f C G/.U / WD colim G.V /:
V f .U /
If G1 and G2 are sheaves on Y , then f .G1 G2 / D f C .G1 / f C .G2 / because filtered

C
colimits commute with finite limits (Appendix Proposition 13.39). As sheafification com-
mutes with finite limits, the inverse image functor f 1 commutes with finite limits. In
other words, it is left exact.
In particular, if G is a sheaf of abelian groups on Y , then the group law aW G G ! G
defines via functoriality a group law
f 1 .a/
f 1 .G/ f 1 .G/ D f 1 .G G/ ! f 1 .G/:
Hence f 1 yields a functor
f 1 W (ZY -Mod) ! (ZX -Mod):
Here we identify (ZX -Mod) with the category of sheaves of abelian groups on X. We
obtain also functors
f 1 W (Com(ZY )) ! (Com(ZX )); and f 1 W K.ZY / ! K.ZX /
by defining f 1 .G / WD . ! f 1 .Gp / ! f 1 .GpC1 / ! : : : /.
Proposition 10.26. The functor f 1 W (ZY -Mod) ! (ZX -Mod) is exact. In particular, if
uW G ! H is a quasi-isomorphism in (Com(ZY )), then f 1 .u/W f 1 G ! f 1 H is
a quasi-isomorphism.
Proof. It is left exact by Remark 10.25 and right exact because it is left adjoint to the
functor f (Proposition 3.49 and Remark 3.53) and in particular commutes with arbitrary
colimits (Appendix Proposition 13.47).
One can also use (10.10) to prove Proposition 10.26 with less abstract nonsense.
10.5 Cohomology and Inverse and Direct Image 223
Remark and Definition 10.27. Let G be a complex of sheaves of abelian groups on

qi s
Y . Choose a K-injective resolution G ! I . By Proposition 10.26 we obtain a quasi-
qi s qi s
isomorphism f 1 G ! f 1 I . Now let sW f 1 I ! J be a K-injective resolution.
We obtain for all n 2 Z homomorphisms of abelian groups
f 1 W H n .Y; G / D H n . ! .Y; Ip / ! .Y; IpC1 / ! : : : /

! H n . ! .X; f 1 Ip / ! .X; f 1 IpC1 / ! : : : /
s
! H n . ! .X; Jp / ! .X; JpC1 / ! : : : /
D H n .X; f 1 G /;
where the first arrow is given by pullback of sections (Definition 3.57) in every degree. If
G is concentrated in degree 0, then we can assume that Ip D 0 and Jp D 0 for all p < 0.
In particular we see that f 1 is the usual pullback of sections for n D 0 in this case.
Lemma 10.28. Let F be an injective ZX -module. Then f .F/ is an injective ZY -module.
Proof. Let G ! H be an injective homomorphism of ZY -modules. Then f 1 G ! f 1 H

is injective. Therefore the map
Hom.H; f I/ D Hom.f 1 H; I/ ! Hom.f 1 G; I/ D Hom.G; f I/
is surjective.
The proof generalizes immediately to the case where f is an arbitrary functor F W A !

B between abelian categories A and B that has an exact left adjoint functor.
Proposition 10.29. Let f W X ! Y be a continuous map, let F be a sheaf of abelian

groups on X and assume that Rq f .F/ D 0 for all q > 0. Then we get an isomorphism
H p .Y; f F/ ! H p .X; F/ for all p 0.
Proof. Let 0 ! F ! I0 ! I1 ! : : : be an injective resolution. Then Rq f .F/ is by

definition the complex cohomology of f .I /. Hence Rq f .F/ D 0 for all q > 0 implies
that
0 ! f .F/ ! f .I0 / ! f .I1 / ! : : :
is exact. By Lemma 10.28 the abelian sheaves f .Ip / are injective. Hence
H p .Y; f .F// D H p . ! 0 ! .Y; f .I0 // ! .Y; f .I1 // ! : : : /

D H p . ! 0 ! .X; I0 / ! .X; I1 / ! : : : /
D H p .X; F/:
Corollary 10.30. Let X be a topological space, let iW A ,! X be the inclusion of a closed

subspace, and let F be an abelian sheaf on A. Then H p .X; i F/ D H p .A; F/ for all
p 0.
Proof. Then i W (ZA -Mod) ! (ZX -Mod) is exact by (3.10) and therefore Rq i D 0 for all
q > 0. Hence we can apply Proposition 10.29.
Proposition 10.31 (Mayer–Vietoris sequence). Let A and B closed subspaces of a topo-

logical space X. Then for every abelian sheaf F on X there is a long exact sequence
: : : ! H p .A [ B; F/ ! H p .A; F/ ˚ H p .B; F/ ! H p .A \ B; F/
(10.11)
! H pC1 .A [ B; F/ ! : : :
Here we write H p .A; F/ instead of H p .A; FjA /. Similarly for the other terms.
Proof. Let iW A ! X, j W B ! X, kW A \ B ! X, and lW A [ B ! X be the inclusions.

Let G be an abelian sheaf with Gx D 0 for all x 2 X n A. Then G ! i .i 1 G/ is
an isomorphism by (3.10). Hence every morphism of sheaves F ! G factors F !
i .i 1 F/ ! i .i 1 G/ ! G. As i factors through l, this shows that F ! i .i 1 F/ factors
through l .l 1 F/, similarly for j . These factorizations yield a morphism l .l 1 F/ !
i .i 1 F/ ˚ j .j 1 F/. The same argument shows that the composition of F ˚ F ! F,
.s; t/ 7! s t, and of F ! k .k 1 F/ yields a morphism i .i 1 F/ ˚ j .j 1 F/ !
k .k 1 F/. Moreover, the sequence
0 ! l .l 1 F/ ! i .i 1 F/ ˚ j .j 1 F/ ! k .k 1 F/ ! 0
is exact on stalks by (3.10), hence it is exact. It induces the long exact sequence (10.11)
by Corollary 10.30.
10.6 Proper Base Change
In this section we will prove the proper base change theorem. We will consider a commu-
tative diagram of topological spaces
p
W X
q f (10.12)
g
Z Y:
Let F be a sheaf of abelian groups on X. Then there is a functorial morphism of abelian

sheaves
g 1 .f F/ ! q .p 1 F/ (10.13)
10.6 Proper Base Change 225
defined as follows. As the functor g 1 is left adjoint to g it suffices to define a func-

torial morphism f F ! g .q .p 1 F// D f .p .p 1 F//. As p and p 1 are adjoint,
idp1 F corresponds to a morphism F ! p .p 1 F/. Now we apply the functor f to this
morphism.
The universal property of a fiber product yields a continuous map W ! X Y Z,
w 7! .p.w/; q.w// and (10.12) is called cartesian if this map is a homeomorphism.
Assume that this is the case and that g is the inclusion of a point Z D fyg to Y . Then
p is the inclusion f 1 .y/ ! X and q is given by the functor G 7! .f 1 .y/; G/. Hence
(10.13) is the morphism
.f F/y ! f 1 .y/; Fjf 1 .y/ (10.14)
defined as the composition
f .F/y D colim F.f 1 .V //

V Y open
V 3y
.A/
! colim F.U / (10.15)
U X open
U f 1 .y/
.B/
! F.f 1 .y// D f 1 .y/; Fjf 1 .y/ :
Remark 10.32. Suppose that (10.12) is cartesian. If f is proper (respectively separated),

then q is proper (respectively separated).
Indeed, for “proper” this follows from Theorem 1.30 (iv): If Z 0 ! Z is a continuous
map, then Z 0 Z W D Z 0 Y X and hence the projection Z 0 Z W ! Z 0 is closed.
For “separated” we use Proposition 1.25 (ii). Let z 2 Z and let w; w 0 2 q 1 .z/ with
w ¤ w 0 . As (10.12) is cartesian, p induces a homeomorphism q 1 .z/ ! f 1 .g.z//. In
particular p.w/ ¤ p.w 0 /. Hence there exist open disjoint neighborhoods U of p.w/ and
U 0 of p.w 0 / in X. Then p 1 .U / and p 1 .U 0 / are open disjoint neighborhoods of w and
w 0 respectively in W . Hence q 1 .z/ is relatively Hausdorff in W .
Theorem 10.33 (Proper base change). Let a cartesian diagram (10.12) be given and
let F be a bounded below complex of abelian sheaves on X. Suppose that one of the
following hypotheses is satisfied:
1. The map f is separated and proper.

2. The spaces X and W are Hausdorff hereditarily paracompact (e.g., if X and W are
manifolds or – more generally – are metrizable) and the maps f and q are closed.
Then one has a functorial isomorphism
g 1 .Rn f F / ! Rn q p 1 F (10.16)
for all n 0, which is given by (10.13) for n D 0.

For Z D fyg with y 2 Y we obtain the following corollary.
Corollary 10.34. Let f W X ! Y be a continuous map of topological spaces. Let F be

a bounded below complex of sheaves of abelian groups on X and y 2 Y . Denote by
iy W f 1 .y/ ! X the inclusion. Suppose that one of the following hypotheses is satisfied:
1. The map f is proper and separated.

2. The space X is Hausdorff and hereditarily paracompact and f is closed.
Then one has a functorial isomorphism

Rn f .F /y ! H n f 1 .y/; iy1 .F / (10.17)
for all n 0, which is given by (10.15) for n D 0.
For the proof of Theorem 10.33 we will use the following result.
Lemma 10.35. Suppose that the hypotheses of Corollary 10.34 are satisfied and let I be
a flabby abelian sheaf on X. Then Ijf 1 .y/ is -acyclic for all y 2 Y .
Proof. If Hypothesis 2 is satisfied, then Ijf 1 .y/ is flabby by Proposition 10.13 2 and hence
-acyclic by Proposition 10.15. Now suppose that Hypothesis 1 is satisfied. Then f 1 .y/
is compact and relatively Hausdorff in X. We show that Ijf 1 .y/ is soft and hence -
acyclic (Proposition 10.17). Indeed, if A f 1 .y/ is closed then A is compact and
relatively Hausdorff in X. Hence by Proposition 9.1 3 we can extend every section s 2
I.A/ to a section s 0 over an open neighborhood U of A in X. As I is flabby, we can extend
s 0 to a section sQ over X. Its restriction to f 1 .y/ extends s.
Proof (Proof of Theorem 10.33). (i). We first show the claim for n D 0 and for F D F
concentrated in degree 0. We have to show that (10.13) is an isomorphism. This we can
do on stalks. Therefore we may assume that Z D fyg for some y 2 Y . Hence it suffices
to prove Corollary 10.34 for n D 0 and F D F. We show that the morphisms (A) and
(B) in (10.15) are isomorphisms.
Morphism (A) is an isomorphism because of Appendix Proposition 12.11 as both hy-
potheses imply that f is closed. To see that Morphism (B) is an isomorphism we can
apply Proposition 9.1. If Hypothesis 2 is satisfied, then every neighborhood of f 1 .y/ is
Hausdorff paracompact. If Hypothesis 1 is satisfied, then f 1 .y/ is compact because f
is proper and a relatively Hausdorff subspace because f is separated.
(ii). Next we show that for every flabby abelian sheaf I on X the pullback p 1 I is
acyclic for the functor q .
For every z 2 Z we have .p 1 I/j q 1 .z/ D Ijf 1 .g.z/ . Hence Lemma 10.35 shows that
the restrictions of G WD p 1 I to the fibers of q are -acyclic.
10.6 Proper Base Change 227
Let 0 ! G ! J0 ! J1 ! : : : be an injective resolution. We have to show that
0 ! q G ! q J0 ! q J1 ! : : :
is exact. We check this on stalks at all z 2 Z. Now the hypotheses of Theorem 10.33
imply that the hypotheses of Corollary 10.34 are satisfied for the morphism q (for Hypoth-
esis 1 use Remark 10.32). Hence we can apply Corollary 10.34 for the map q and n D 0
and for complexes concentrated in degree 0, which we already proved in (i). Therefore it
suffices to see that
0 ! q 1 .z/; Gj q 1 .z/
(*)
! q 1 .z/; J0 jq 1 .z/ ! q 1 .z/; J1 jq 1 .z/ ! : : :
is exact. As Jp is flabby (Proposition 10.12), Lemma 10.35 shows that Jp j q 1 .z/ is -

acyclic for all p and all z 2 Z. Hence Gjq 1 .z/ ! J jq 1 .z/ computes the cohomology
of Gjq 1 .z/ by Proposition 10.8. As we have already seen that Gjq 1 .z/ is -acyclic, (*) is
exact.
qi s
(iii). We now conclude the proof. Choose an injective resolution F ! I . In
particular I is bounded below and consists of flabby abelian sheaves (Proposition 10.12).
By definition of Rn f we have
Rn f .F / D H n ! f Ip ! f IpC1 ! : : : :
As taking inverse images is exact (Proposition 10.26), we deduce
g 1 .Rn f .F // D H n ! g 1 .f Ip / ! g 1 .f IpC1 / ! : : :
.i / (**)
D H n ! q .p 1 Ip / ! q .p 1 IpC1 / ! : : :
As p 1 is an exact functor, we see that p 1 I is a resolution of p 1 F . It consists of

q -acyclic sheaves by (ii). Hence (**) computes Rn q .p 1 F / by the analogue of Propo-
sition 10.8 for the functor q .
Corollary 10.36. Let X, Y , f be as in Corollary 10.34 and let F be an abelian sheaf on

X. Assume that
H n f 1 .y/; iy1 F D 0 (*)
for all n 1 and y 2 Y . Then
H n .Y; f F/ D H n .X; F/:
Proof. By Corollary 10.34, (*) implies Rn f .F/ D 0 for all n 1. Hence we can apply
Proposition 10.29.
10.7 Problems
Problem 10.1. Let .X; OX / be a ringed space. Show that every bounded below complex
of OX -modules has an injective resolution.
Hint: Let F be an OX -module and for every x 2 X let Fx ,! Ix be an injective OX;x -linear
map to an injective OX;x -module Ix (Appendix Problem 15.14). Show that U 7! I.U / WD
Q
x2U Ix is an injective OX -module and that the canonical homomorphism F ! I of
OX -modules is injective. Conclude by Appendix Problem 15.11.
Problem 10.2. Let A and A0 be abelian categories, let F W A ! A0 be a left exact

functor and let .F n ; ı/n0 be a ı-functor extending F . Show that .F n ; ı/n0 is universal
if for all n 1 and for every object X in A there exists a monomorphism uW X ! Y in
A such that F n .u/ D 0.
Problem 10.3. Let X be a topological space, x 2 X, A be an abelian group, and let A be

the skyscraper sheaf in x with values A (Problem 3.6). Show that H p .X; A/ D 0 for all
p > 0.
Problem 10.4. Let .X; OX / be a ringed space, let F be a bounded below complex of OX -
modules, and let U; V X be open. Show that there exists a Mayer–Vietoris sequence
: : : ! H p .U [ V; F / ! H p .U; F / ˚ H p .V; F / ! H p .U \ V; F /
! H pC1 .U [ V; F / ! : : :
Hint: Show that for every flabby sheaf A of abelian groups on X the sequence
s7!.s j U ;s j V / .s;t /7!st
0 ! A.U [ V / ! A.U / ˚ A.V / ! A.U \ V / ! 0
is exact.
Problem 10.5. Let A be an abelian group, n 1, p 0 and write H p .X; A/ instead of

H p .X; AX /. Let A2 WD f a 2 A I 2a D 0 g. Show that
8
ˆ
ˆ A; if p D 0 or if p D n and n oddI
ˆ
ˆ
ˆ
<A=2A; if p is even and 0 < p nI
H p .P n .R/; A/ Š
ˆA2 ;
ˆ if p is odd and 0 < p < nI
ˆ
ˆ
:̂
0; otherwise
and 8
<A; if p is even and 0 p 2nI
H p .P n .C/; A/ Š
:0; otherwise:
10.7 Problems 229
Problem 10.6. Let X WD P 1 .C/.
1. Show that HDR 0

.X/ Š HDR
2
.X/ Š C and HDR 1
.X/ D 0. Deduce that HDR2
.X/ is not
isomorphic to the quotient of the space of closed holomorphic 2-forms by the space of
exact holomorphic 2-forms.
p
2. Show that H 0 .X; OX / Š H 1 .X; ˝X1 / Š C and H q .X; ˝X / D 0 for all .p; q/ …
f.0; 0/; .1; 1/g.
Hint: Problem 10.4.
Problem 10.7. Generalize the results of Problem 10.6 (2) and show that
8
<C; if 0 p D q nI
p
H q P n .C/; ˝P n .C/ Š
:0; otherwise.
Problem 10.8. Let M be a real C ˛ -manifold with ˛ 1 and let PicC .M / be the group
of complex C ˛ -line bundles on M . Show that the complex exponential sequence induces
an isomorphism of abelian groups PicC .M / ! H 2 .M; ZM /.
Problem 10.9. Let f W X ! Y be a continuous map and let F be a bounded below

complex of abelian sheaves on X. Show that Rn f .F / is the sheaf associated with the
presheaf V 7! H n .f 1 .V /; F / for all n 2 Z.
Problem 10.10. Let M be a real simply connected manifold. Show that every closed
1-form is exact.
Problem 10.11. Show that on the real C 1 -manifold S 1 there exists a non-exact 1-form
! that is unique up to multiplication with 2 R . What is !?
Problem 10.12. Let f W X ! Y . Show that the proper direct image functor fŠ (Prob-
lem 3.16) from the category of abelian sheaves on X to the category of abelian sheaves
on Y is a left exact functor. Its right derived functor by Rn fŠ is called higher direct image
with compact support. If f is the unique map to the space Y consisting of a single point,
then we consider Rn fŠ .F / as an abelian group for every complex F of abelian sheaves.
This abelian group is called the n-th cohomology with compact support and it is denoted
by Hcn .X; F /. We write c .X; / instead of Hc0 .X; /.
Prove for locally compact Hausdorff spaces an analogue of the Mayer–Vietoris se-
quence for closed subspaces (Proposition 10.31) for the cohomology with compact sup-
port.
Problem 10.13. Let X be a locally compact Hausdorff space (respectively a paracom-

pact Hausdorff space), U X open, Z WD X n U . Show that the exact sequence in
Problem 3.18 yields for every abelian sheaf F a long exact sequence of cohomology with
compact support (Problem 10.12)
! Hcn .U; FjU / ! Hcn .X; F/ ! Hcn .Z; FjZ / ! HcnC1 .U; FjU / ! : : :
(respectivey a long exact sequence
! H n .U; FjU / ! H n .X; F/ ! H n .Z; FjZ / ! H nC1 .U; FjU / ! : : : /
Problem 10.14. Let X be a paracompact locally compact Hausdorff space and let F be
an abelian sheaf. Show that the following assertions are equivalent:
(i) F is soft.
(ii) H n .X; jŠ .FjU // D 0 for all n 1 and for every open topological embedding j W U !
X.
(iii) H 1 .X; jŠ .FjU // D 0 for every open topological embedding j W U ! X.
Here jŠ is the proper direct image (Problem 3.16).

Hint: Problem 10.13.
Problem 10.15. Let X be a locally compact Hausdorff space. A sheaf F on X is called

c-soft if the restriction map F.X/ ! F.K/ is surjective for every compact subspace K
of X.
1. Show that for an abelian sheaf F the following assertions are equivalent:
(i) F is c-soft.
(ii) Hcn .X; jŠ .FjU // D 0 (Problem 10.12) for all n 1 and for every open topological
embedding j W U ! X.
(iii) Hc1 .X; jŠ .FjU // D 0 for every open topological embedding j W U ! X.
2. Show that every colimit of c-soft abelian sheaves on X is a c-soft abelian sheaf.
3. Show that the functors Hcn .X; / commute with colimits.
Problem 10.16. Let M be a real C ˛ -manifold with ˛ 1. Show that any soft sheaf
on M is c-soft (Problem 10.15) and deduce that Hcn .M; E/ D 0 for all n 1 and every
CM
˛
-module E.
Problem 10.17. Let U R be open and connected.
1. Show that
f 7!f 0
0 ! RU ! C1U ! C0U ! 0
is an exact sequence of R-vector spaces.
10.7 Problems 231
2. Show that Hcn .U; RU / D R (Problem 10.12) for n D 1 and Hcn .U; RU / D 0 for
n ¤ 1. R
f 7!f 0
Hint: The sequence 0 ! c .U; C1U / ! c .U; C0U / ! R ! 0 is exact, where
R R1
f WD 1 f .x/ dx.
Problem 10.18. Let X be a topological space and A X a closed subspace. For every
sheaf of abelian groups F define
A .X; F/ WD f s 2 .X; F/ I supp.s/ A g :
Show that A .X; / is a left exact functor from the category of abelian sheaves to the
p
category of abelian groups. For every complex A of abelian sheaves HA .X; A / WD
Rp A .X; A / is called the local cohomology group with support in A.
Problem 10.19. Let X be a topological space, A X closed and U WD X n A. Show

that for every bounded below complex of abelian sheaves A one has an exact sequence
p pC1
! HA .X; A / ! H p .X; A / ! H p .U; A / ! HA .X; A / ! : : : ;
p
where HA .X; / denotes the local cohomology with support in A (Problem 10.18).
Problem 10.20. Let A be an abelian category with injective and K-injective resolutions.
qi s
Let M and N be complexes of objects in A and choose a K-injective resolution N !
I . For p 2 Z the abelian group
p
ExtA .M ; N / WD HomK.A/ .M ; I Œp/;
is called the Ext group of M and N . Here the shift Œp is defined in Appendix Prob-
lem 15.1.
qi s
1. Show that if M ! J is a quasi-isomorphism, then
p
ExtA .M ; N / D HomK.A/ .J ; I Œp/:
p
2. Show that the group ExtA .M ; N / does not depend on the choice of the K-injective
p
resolution of N up to unique isomorphism and that M 7! ExtA .M ; N / and N 7!
p
ExtA .M ; N / define functors K.A/opp ! (Ab) and K.A/ ! (Ab) that send quasi-
isomorphisms of complexes to isomorphism of abelian groups.
3. Show that for all objects M and N of A (considered as complexes concentrated in
degree 0) one has Ext0A .M; N / D HomA .M; N /.
4. Show that for p; q 2 Z and for all complexes M , N , P of objects in A and all
qi s qi s qi s
K-injective resolutions M ! J , N ! I , P ! K the composition
HomK.A/ .J ; I Œp/ HomK.A/ .K ; J Œq/ ! HomK.A/ .K ; I Œp C q/;

.v; u/ 7! vŒq ı u
yields a well-defined Z-bilinear map

p q pCq
[W ExtA .M ; N / ExtA .P ; M / ! ExtA .P ; N /; (10.18)
which is called the Yoneda product. Show that the Yoneda product is associative and
L p
makes p ExtA .M ; M / into a ring.
5. Show that for every object M of A the p-th right derived functor of the left exact
p
functor A ! (Ab), X 7! HomA .M; X/, is the functor X 7! ExtA .M; X/.
Problem 10.21. Let .X; OX / be a ringed space. Show that for all OX -modules F
p
and all p 2 Z there is a functorial isomorphism H p .X; F/ ! ExtOX .OX ; F/ (Prob-
L
lem 10.20). Deduce that the Yoneda product (10.18) endows p H .X; OX / with the
p
L
structure of a ring and p H p .X; F/ with the structure of a right module over the ring
L
p H .X; OX /. The scalar multiplication
p
H p .X; F/ H q .X; OX / ! H pCq .X; F/; . ; ˛/ 7! [ ˛
is called the cup product.

L
Problem 10.22. Let M be a real manifold. Show that via the isomorphism p H p .M;
L p
KM / ! p HDR .M / the cup product (Problem 10.21) on the left-hand side corresponds
to the exterior product .!; / 7! ! ^ of differential forms on the right-hand side.
Problem 10.23. Let X be a topological space and let 1 ! A ! G ! H ! 1 (*) be

an exact sequence of sheaves of groups on X such that A is in the center of G. Show that
there exists a functorial morphism of pointed sets @W H 1 .X; H/ ! H 2 .X; A/ such that
@
H 1 .X; G/ ! H 1 .X; H/ ! H 2 .X; A/ is an exact sequence of pointed sets and such that
@ is the usual connecting morphism if (*) is a sequence of sheaves of abelian groups.
Hint: Apply . /Œ0 to (*).
Problem 10.24. Let M be a real manifold. Show that every principal PGLn .R/-bundle
(respectively every principal PGLn .C/-bundle, where we consider PGLn .C/ as a real Lie
group) over M comes from a principal GLn .R/-bundle (respectively a principal GLn .C/-
bundle) by the functoriality of bundles (Remark 8.23) if and only if H 2 .M; .Z=2Z/M / D
0 (respectively H 3 .M; ZM / D 0).
Hint: Use Problem 10.23 and the exponential sequence.
Remark: In Corollary 11.14 we will see H 2 .M; .Z=2Z/M / D H 3 .M; ZM / D 0 if M is
a contractible manifold.
Cohomology of Constant Sheaves
11
In this chapter we focus on cohomology of constant sheaves. We start by defining singular

(co)homology and then show that for a locally contractible space X singular cohomology
with values in a ring R and the cohomology of the constant sheaf RX are equal. We deduce
that RX is acyclic if X is contractible and locally contractible.
Combining this result for X D Œ0; 1 and the proper base change theorem we deduce
homotopy invariance of cohomology of constant sheaves for continuous maps between
arbitrary topological spaces in Sect. 11.3. We conclude the chapter with some easy appli-
cations.
Notation: Let R always be a commutative ring and let X be a topological space.
11.1 Singular Cohomology
In this section we define singular cohomology and show that it vanishes on contractible
spaces.
Definition 11.1. Let n 2 N0 and set Œn WD f0; : : : ; ng. Let ei 2 RŒn D RnC1 be the i-th
standard unit vector for i D 0; : : : ; n.
1. The n-dimensional standard simplex is

( )
X
n
WD Œn WD
n
.t0 ; : : : ; tn / 2 R nC1
I ti D 1; ti 0 :
i D0
Then 0 D f1g R. We identify
1 D f .t0 ; t1 / I 0 t0 D 1 t1 1 g ! Œ0; 1; .t0 ; t1 / 7! t1 :

DOI 10.1007/978-3-658-10633-1_11
234 11 Cohomology of Constant Sheaves
2 is an equilateral triangle, 3 is a tetrahedron.

2. Every weakly increasing map ˛W Œm ! Œn induces an affine map
X
m X
m
.˛/W Œm ! Œn; ti ei 7! ti e˛.i / :
i D0 i D0
Then .˛ ı ˇ/ D .˛/ ı .ˇ/ and .id/ D id.

3. For 0 i n let ıin W Œn 1 ! Œn be the weakly increasing injective map that misses
the value i. Then ıjnC1 ı ıin D ıinC1 ı ıjn1 for i < j (both compositions miss i and j ).
Set
din WD .ıin /W Œn 1 ! Œn:
This maps Œn 1 injectively on the “i-th face of Œn”. For instance
d01 .1/ D .0; 1/ 2 1 d11 .1/ D .1; 0/ 2 1 :
By functoriality one has
djnC1 ı din D dinC1 ı djn1 for i < j : (11.1)
Definition 11.2. Let X be a topological space, n 2 N0 .
1. A continuous map W n ! X is called a (singular) n-simplex in X. The set of n-

simplices in X is denoted by ˙n .X/. The i-th face of is ı din (an (n1)-simplex).
2. Denote by Sn .X/ the free abelian group with basis the set of singular n-simplices in
X. Set Sn .X/ WD 0 for n < 0. An element of x 2 Sn .X/ is called a singular n-chain.
P
We think of x as a formal finite linear combination x D n , n 2 Z.
3. Let G be an abelian group. A singular n-cochain with values in G is a map ˙n .X/ !
G. Equivalently, it is a group homomorphism Sn .X/ ! G. We denote the group of
singular n-cochains with values in G by S n .X; G/ D HomZ .Sn .X/; G/.
4. Define boundary operators for all q 2 Z:
X
q
q
@q W Sq .X/ ! Sq1 .X/; 7! .1/i ı di ;
i D0
@ W S .X; G/ ! S
q q qC1
.X; G/; @ .'/ D .1/qC1 .' ı @qC1 /:
q
Example 11.3.
1. One can identify ˙0 .X/ D X (as a set), S0 .X/ D Z.X / and
S 0 .X; G/ D f'W X ! G mapg:

11.1 Singular Cohomology 235
2. Associating to 2 ˙1 .X/ the path Œ0; 1 ! X, t 7! .t; 1t/ yields an identification

of ˙1 .X/ and the set of paths in X. Hence ' 2 S 1 .X; G/ is a map that attaches to
each path in X an element in G. For 2 ˙1 .X/, considered as a path Œ0; 1 ! X,
one has
@1 . / D .1/ .0/ 2 S0 .X/:
For ' 2 S 0 .X; G/ considered as a map X ! G one has
@0 .'/. / D '.@1 . // D '. .0// '. .1//
for a path in X. In particular
Ker.@0 / D f'W X ! G j '.x/ D '.y/ if x and y are in the same

(11.2)
path-connected componentg:
Lemma 11.4. For all q 2 Z one has @q ı @q1 D 0 and @q ı @q1 D 0. In particular we
obtain complexes
@3 @2 @1
: : : ! S2 .X/ ! S1 .X/ ! S0 .X/ ! 0 ! : : : ; (11.3)
@0 @1 @2
: : : ! 0 ! S .X; G/ ! S .X; G/ ! S .X; G/ ! : : :
0 1 2
(11.4)
Proof. It suffices to show @q ı @q1 D 0. We decompose
X X
q q1
.1/i Cj ı dj ı di
q q1
@q @q1 D
j D0 i D0
P P
into the parts i <j and i j . When we rewrite the first sum using (11.1), the result is
the negative of the second sum.
Definition 11.5. We define for all n 2 Z the singular homology
Ker.@n W Sn .X/ ! Sn1 .X//

Hnsing .X/ WD Hnsing .X; Z/ WD
Im.@nC1 W SnC1 .X/ ! Sn .X//
and the singular cohomology or Betti cohomology with coefficients in G
Ker.@n W S n .X; G/ ! S nC1 .X; G//

n
Hsing .X; G/ WD :
Im.@n1 W S n1 .X; G/ ! S n .X; G//
sing
Hence both the fundamental group and H1 .X; Z/ consist of equivalence classes of
paths. These groups are related as follows (see Problem 11.5 or [tDi] (9.2.1)).
Remark 11.6. Let X be a path connected space, x0 2 X. Attaching to the homotopy class
sing
Œ 2 1 .X; x0 / its class in H1 .X; Z/ yields a well-defined isomorphism of groups
sing
1 .X; x0 /ab ! H1 .X; Z/:
Here for any group G we denote by G ab its maximal abelian quotient, i.e., G ab D
G=ŒG; G, where the commutator subgroup ŒG; G is the subgroup of G generated by all
commutators ghg 1 h1 , which is a normal subgroup of G.
Definition and Remark 11.7. Let R be a commutative ring.
sing
1. We also define Hn .X; R/ by replacing Sn .X/ by the free R-modules Sn .X; R/ WD
R.˙n .X // with basis the set of singular n-simplices in X. Then
@3 @2 @1
: : : ! S2 .X; R/ ! S1 .X; R/ ! S0 .X; R/ ! 0 ! : : : (11.5)
is a complex of R-modules and we obtain the singular homology with values in R,

sing
denoted by Hn .X; R/. Then Sn .X/ D Sn .X; Z/ and Sn .X; R/ D Sn .X/ ˝Z R.
2. If G D R, then (11.4) is a complex of R-modules with S n .X; R/ D Sn .X; R/_ (dual
as R-module) and @q is up to the sign .1/qC1 the dual map of @qC1 . In particular,
n
Hsing .X; R/ is an R-module.
sing
Example 11.8. Let R be a commutative ring. Then H0 .X; R/ Š R.0 .X // , where
0 .X/ is the set of path-connected components of X.
As forming the dual of an R-module is in general only left exact (and not an exact
functor), the relation between the dual of the homology of a complex of modules and the
homology of the complex of the dual modules is in general quite complicated and best
expressed via spectral sequences. We use only the following general fact for complexes of
free R-modules (e.g., [BouA3] Chap. X, §5.6, Cor. 3 and Cor. 4 de Théoreme 3). Variants
of the version given here are sometimes grouped under the universal coefficient theorem
for singular cohomology.
Proposition 11.9. Let n 2 Z and let R be a commutative ring. Assume one of the follow-
ing two conditions:
sing
(a) R is a principal ideal domain and that Hn1 .X; R/ is a free R-module.
sing
(b) Hm .X; R/ is a free R-module for all m n 1.
Then
n
Hsing .X; R/ Š Hnsing .X; R/_ D HomR Hsing
n
.X; R/; R : (11.6)
11.1 Singular Cohomology 237
Corollary 11.10.
.X; K/ Š Hn .X; K/_ .
sing
1. Let K be a field. For all n 2 Z one has Hsing
n
.X; R/ Š Hn .X; R/_

sing
2. Let R be a commutative ring. If n 1 then one has Hsing
n
sing
because Hm .X; R/ is a free R-module for all m < 1 by Example 11.8.
We encourage the reader to prove 1 directly using the fact that every subspace U of a
K-vector space V is a direct summand and hence any linear map on U can be extended
to V .
Definition and Remark 11.11. Let Y be a topological space. Define the cone over Y by
C.Y / WD .Y Œ0; 1/= ;
where .y; t/ .y 0 ; t 0 / if y D y 0 and t D t 0 < 1 or if t D t 0 D 1. Endow C.Y / with the

quotient topology. Then C.n / D nC1 .
Proposition 11.12. Let X be a contractible space and let R be a commutative ring. Then
one has for all n > 0
Hnsing .X; R/ D Hsing
n
.X; R/ D 0:
More generally, one can show that if f; gW X ! Y are homotopic continuous maps
of topological spaces, then they induce homotopic morphisms of complexes S .X; R/ !
S .Y; R/. In particular, they induce the same map on the singular homology. This gives
a connection between the topological notion of homotopy and the algebraic notion of
homotopy between morphism of complexes.
sing
Proof. Set I WD Œ0; 1. By Proposition 11.9 it suffices to show that Hn .X; R/ D 0 for
all n > 0. Let p 2 X and let F W X I ! X be a homotopy of idX and the constant map
X ! X, x 7! p.
Define a morphism of complexes "W S .X; R/ ! S .X; R/ by "n WD 0 for n ¤ 0 and
P P
"0 . x2X nx x/ WD . x nx /p (where we identify 0-simplices in X and points of X). It
suffices to show that there exists a homotopy between the morphism of complexes " and
sing
idS .X;R/ (then they induce the same maps on Hn .X; R/ and hence idH sing .X;R/ D 0 for
n
all n > 0).
Let W n ! X be an n-simplex in X. Then F ı. idI /W n I ! X factors through
C.n / D nC1 and hence induces an .n C 1/-simplex hn . /W n ! X. This defines
homomorphisms hn W Sn .X; R/ ! SnC1 .X; R/. One easily checks .hn /n is the desired
homotopy (note that the numbering for homotopies is reversed because the differential in
the complexes here lowers the degree).
11.2 Cohomology and Singular Cohomology
We now compare singular cohomology with values in a commutative ring R and sheaf
cohomology of the sheaf of locally constant R-valued functions. In the sequel we will
write H n .X; R/ instead of H n .X; RX / for all n 0.
Theorem 11.13. Let X be a locally contractible space (e.g., a premanifold) and let R be
a commutative ring. Then there is for all n 2 Z an isomorphism H n .X; R/ Š Hsing
n
.X; R/
of R-modules.
Proof. (i). For V U X open we have ˙n .V / ˙n .U / and we define a surjective

map
S n .U; R/ ! S n .V; R/; '!7 ' j˙n .V / :
Let SnR be the sheafification of the presheaf of abelian groups U 7! S n .U; R/. Then one
easily checks that for U X open one has
SnR .U / D S n .U; R/=S n .U; R/0 ; (*)
where
˚
S n .U; R/0 D ' 2 S n .U; R/ I 9 covering .Vi /i of U with ' j˙n .Vi / D 0 for all i :
As S n .U; R/ ! S n .V; R/ is surjective for V U , (*) shows in particular that the

sheaves SnR are flabby.
(ii). The map @n yield morphisms of sheaves of abelian groups @n W SnR ! SnC1
R and
we obtain a complex
@0 @1 @2
: : : ! 0 ! S0R ! S1R ! S2R ! : : : (**)
of ZX -modules. As X is locally path connected, a map X ! R is locally constant if and

only if it is constant on path-connected components. Hence (11.2) shows that Ker.@0 / D
RX .
Moreover, the complex (**) is exact in all degrees > 0. Let x 2 X. By hypothesis
on X we can compute stalks of presheaves in a point x by taking colimits only over
open contractible neighborhoods of x. Now Proposition 11.12 shows that for all open
contractible neighborhoods U of x the complex of cochains S .U; R/ is exact in degree
> 0. This implies that SR is exact on stalks and hence exact in degree > 0. Therefore we
have seen that
qi s
RX ! SR
is a flabby resolution of RX and hence

p pC1
H n .X; R/ D H n ! SR .X/ ! SR .X/ ! : : : :
11.2 Cohomology and Singular Cohomology 239
(iii). Let pW S .X; R/ ! SR .X/ be the morphism of complexes induced by (*). It

remains to show that p is an isomorphism in the homotopy category K.R/. This follows
from the “theorem of small chains” ([tDi] Theorem 9.4.5).
Corollary 11.14. Let X be a contractible locally contractible space. Then H n .X; R/ D

0 for all n > 0 and for every commutative ring R.
Proof. Proposition 11.12 and Theorem 11.13.
Corollary 11.15. 1 Let M be a contractible real C 1 -manifold, n 1. Then every closed

n-form on M is exact.
Proof. We have
fclosed n-forms on M g Theorem 10.20

D H n .M; R/
Corollary 11.14
D 0:
By Remark 10.21, Corollary 11.15 also holds for real analytic manifolds.
Let us give a proof of the Weierstraß theorem from complex analysis using Corol-
lary 11.14.
Example 11.16. Let X be a connected 1-dimensional complex manifold (for instance an

open subset of C). Let U X be open. A divisor on U is an element D D .nz /z2U 2 ZU
such that supp.D/ WD f z 2 U I nz ¤ 0 g is a closed and discrete subspace of U . Denote
the set of divisors on U by DX .U /. We obtain an abelian sheaf DX of Z-valued functions
on X with stalks DX;x D Z for all x 2 X.
Let MX be the sheaf of those meromorphic functions f on open subspaces U of X
such that 1=f exists as meromorphic function (i.e., a meromorphic function f on U is in
MX .U / if f jW ¤ 0 for every connected component W of U ). Let z0 2 X. To describe

the stalk MX;z 0
we choose a chart of X at z0 and hence may assume that X C is open.

Then MX;z0 is the abelian group of non-vanishing Laurent series with finite principal parts
P
nm an .z z0 / , am ¤ 0, that converge in U n fz0 g for some open neighborhood U of
n
z0 in C.
For every f 2 MX .U / let divU .f / WD .ordz .f //z2U 2 D.U /, where ordz .f / denotes
the order of vanishing of f at z (equal to k 2 Z<0 if z is a pole of order k for f ). We
obtain a morphism of abelian sheaves divW MX ! DX , which is given on stalks by sending
1
Note that this corollary is something of a circular statement because it uses the Poincaré lemma
for smooth manifolds, which is often proved by showing this corollary first.
P
nm an .z z0 /n with am ¤ 0 to m 2 Z. In particular, it is surjective. Therefore we get
an exact sequence of abelian sheaves
div
1 ! OX ! MX ! DX ! 0:
The long cohomology sequence shows that for every divisor D on X there exists a mero-
morphic function f on X with div.f / D D if and only if Pic.X/ D H 1 .X; OX / !
H 1 .X; MX / is injective.
To analyze Pic.X/ we use the complex exponential sequence (Example 8.11). It yields
an exact sequence
! H 1 .X; OX / ! Pic.X/ ! H 2 .X; Z/ ! H 2 .X; OX / ! : : :
If X C is open, then H i .X; OX / D 0 for all i 1 (Theorem 10.22) and we obtain

an isomorphism Pic.X/ ! H 2 .X; Z/. In particular it follows from Corollary 11.14 that
Pic.X/ D 1 if X is contractible2 . Hence in this case every divisor on X comes from
a meromorphic function.
11.3 Homotopy Invariance of Cohomology
Lemma 11.17. Let X be a topological space and let pW X Œ0; 1 ! X be the projection.
Then the pullback maps (Definition 10.27)
p 1 W H n .X; R/ ! H n .X Œ0; 1; R/
are isomorphism of R-modules for all n 0 and for every commutative ring R.
Proof. As Œ0; 1 is Hausdorff and compact, p is proper and separated (Proposition 10.32).
Moreover we have p .RX Œ0;1 / D RX (Example 3.44).
By Corollary 11.14 we have H n .p 1 .x/; R/ D H n .Œ0; 1; R/ D 0 for all n > 0 and
x 2 X. Therefore the result follows from Corollary 10.36.
Theorem 11.18 (Homotopy Invariance of Cohomology). Let X and Y be topological

spaces, and let f; gW X ! Y be homotopic continuous maps. Then
f 1 D g 1 W H n .Y; R/ ! H n .X; R/
for all n 0 and for every commutative ring R.
2
In fact it can be shown that H 2 .X; Z/ D 1 for every open subspace X of C (see for instance
[Ive] VI, 6.8., considering X as an open subset of P 1 .C/).
11.4 Applications 241
Proof. For t D 0; 1 let i t W X ! X Œ0; 1, i t .x/ D .x; t/. Let pW X Œ0; 1 ! X be the
first projection. Then p ı i t D idX . Hence the composition
p1 i t1
H n .X; R/ ! H n .X Œ0; 1; R/ ! H n .X; R/
is the identity. As p 1 is an isomorphism by Lemma 11.17, we see that i01 D i11 and
that both maps are isomorphisms.
Let F be a homotopy between f and g. Then F ı i0 D f and F ı i1 D g and hence
f 1 D i01 ı F 1 D i11 ı F 1 D g 1 .
Corollary 11.19. Let X and Y be topological spaces and let f W X ! Y and gW Y ! X

be continuous maps such that g ı f is homotopic to idX and such that g ı f is homotopic
to idY . Then f and g induce mutually inverse isomorphisms
H n .X; R/ Š H n .Y; R/
for all n 0 and for every commutative ring R.
11.4 Applications
Example 11.20. Let n 2 N and let R be a commutative ring. Let

˚
S n D x 2 RnC1 I x02 C C xn2 D 1 :
Define S˙n D f x 2 S n I ˙xn 0 g. Each of these half spheres is homeomorphic to the

closed unit ball in Rn and in particular contractible. Moreover, SCn \ Sn D S n1 . Hence
the Mayer–Vietoris sequence (Proposition 10.31) has the form
0 ! H 0 .S n ; R/ ! H 0 SCn ; R ˚ H 0 Sn ; R ! H 0 .S n1 ; R/ ! H 1 .S n ; R/

(11.7)
! 0 ! ! 0 ! H p1 .S n1 ; R/ ! H p .S n ; R/ ! 0 ! : : :
Hence we see 8
<R; for p D 0; 1;
H p .S 1 ; R/ D
:0; otherwise.
Moreover we have H 0 .S n ; R/ D R because S n is connected. Hence (11.7) shows by

induction that for all n 1 one has
8
<R; for p D 0; n;
H p .S n ; R/ D (11.8)
:0; otherwise.
Corollary 11.21. Let n 2. Let x0 2 Rn and X WD Rn n fx0 g. Then

8
<R; for p D 0; n 1;
H p .X; R/ D (11.9)
:0; otherwise,
for every commutative ring R. In particular we see that X is not contractible.
Proof. We may assume that x0 D 0. The inclusion iW S n1 ,! X WD Rn n f0g is a ho-

motopy equivalence (Example 2.4). Hence the claim follows from Example 11.20 via
homotopy invariance (Corollary 11.19).
Corollary 11.22. Let f W Rn ! Rm be a homeomorphism. Then n D m.
Proof. The homeomorphism f induces a homeomorphism Rn n f0g ! Rm n ff .0/g and

hence an isomorphism H p .Rn n f0g; Z/ ! H p .Rm n ff .0/g; Z/ for all p. Now the claim
follows from Corollary 11.21
Corollary 11.23 (Brouwer’s fixed point theorem). For n 2 N let D n WD f x 2

Rn I jjxjj2 1 g be the closed unit ball. Then every continuous map f W D n ! D n has
a fixed point (i.e., there exists x 2 D n with f .x/ D x).
Proof. Assume that there exists a continuous map f W D n ! D n without fixed points.
Then we can construct a continuous map rW D n ! S n1 as follows. Let r.x/ denote the
point, where the half line starting in f .x/ towards x intersects S n1 . Let iW S n1 ! D n
be the inclusion. As r ı i D id, the map
i 1 W H n1 .D n ; Z/ ! H n1 .S n1 ; Z/
is surjective. But for n 2 one has H n1 .D n ; Z/ D 0 because D n is contractible and

H n1 .S n1 ; Z/ D Z by (11.8). For n D 1 one has H 0 .D 1 ; Z/ D Z and H 0 .S 0 ; Z/ Š
Z2 . In both cases we obtain a contradiction!
For n D 1, D 1 is a closed interval and Brouwer’s fixed point theorem is also a standard
application of the intermediate value theorem.
11.5 Problems
In the problem section, R always denotes a commutative ring.
Problem 11.1. What is the chain complex S .X/ for X consisting of a single point?
sing n
What are Hn .X/ and Hsing .X; R/?
11.5 Problems 243
Problem 11.2. Let X be a topological space and let .Xi /i be its family of path compo-
sing L sing
nents. Show that Hn .X/ D i Hn .Xi / for all n 0.
Problem 11.3. Suppose that R ¤ 0. Show that Hsing

0
.X 0 ; R/ ¤ H 0 .X 0 ; R/ with X 0
defined in Problem 2.7.
Problem 11.4. Let R be a commutative ring without Z-torsion (e.g., if R is a Q-algebra).

sing sing
Show that Hn .X; R/ Š Hn .X/ ˝Z R (*).
Give an example such that (*) does not hold for R D Z=2Z.
Hint: R is a flat Z-module (Appendix Problem 14.9). For the example see Problem 7.6.
Problem 11.5. Let X be a topological space, let x0 2 X. Show that 1 .X; x0 /ab Š
sing
H1 .X; Z/. The following steps might be useful:
1. Let ; W Œ0; 1 ! X paths with .1/ D .0/. Let ! 2 ˙2 .X/, .t0 ; t1 ; t2 / 7! .

/.t1 =2 C t2 /. Show that @2 .!/ D C .
2. Let H W 1 Œ0; 1 ! X be a homotopy ' of paths ; W 1 ! X relative to
f0; 1g. Show that H factors through the quotient map 1 Œ0; 1 ! 2 , .t0 ; t1 ; t/ 7!
.t0 ; t1 .1t/; t1 t/ and hence yields an element 2 ˙2 .X/. Show that @2 ./ D cC ,
where c is a constant map.
3. Deduce from (1) and (2) that attaching to a path in X with start and end point x0 its
sing
class in H1 .X; Z/ yields a well-defined group homomorphism hX;x0 W 1 .X; x0 /ab !
sing
H1 .X; Z/, where . /ab denotes the maximal abelian quotient.
4. Suppose that X is path connected. Choose for every x 2 X a path x from x0 to x.
Let 2 ˙1 .X/ and let k. / be the closed path . .0/ / .1/ . Show that the linear
extension kQ of k to Ker.@1 / yields an inverse to hX;x0 .
Hint: To show that kQ maps Im.@2 / to 0 use that 2 is contractible and deduce that for
! 2 ˙2 .X/ one has !2 !0 D !1 if !i WD ! ı di2 is the i-th face of !.
p
Problem 11.6. Show that Hcn .Rn ; R/ D R (Problem 10.12) and Hc .Rn ; R/ D 0 for
p ¤ n.
p
Problem 11.7. What is dimR .HDR .Rn n f0g// for n 1, p 0?
Problem 11.8. Let X be a topological space. Then

sing
bk .X/ WD dimQ Hk .X; Q/ 2 N0 [ f1g; k0
is called the k-th (rational) Betti number of X. Suppose that bk .X/ < 1 for all k and
that bk .X/ ¤ 0 for only finitely many k. Then
X
.X/ WD .1/k bk .X/
k0
is called the Euler characteristic of X. Determine the Betti numbers of S n , of P n .R/ and
of P n .C/ and deduce that
8 8
<2; if n is evenI <1; if n is evenI
.S n / D .P n .R// D
:0; if n is oddI :0; if n is oddI
and .P n .C// D n C 1. Show that there exist no integers n; m 1 such that P n .R/ and
P m .C/ are homotopy equivalent. Show that S k is not homotopically equivalent to any
projective space for k 3. What is with k D 1; 2?
Hint: Problem 10.5.
Problem 11.9. Let n 1. Show that EndZ .H n .S n ; Z// D Z. For every continuous map
f W S n ! S n let deg.f / WD f 1 2 EndZ .H n .S n ; Z// D Z.
1. Show that deg.f / depends only on the homotopy class of f and that deg.f ı g/ D
deg.f / deg.g/ for all continuous maps f; gW S n ! S n .
2. Identify S 1 D f z 2 C I jzj D 1 g. Show that deg.z 7! z k / D k for all k 2 Z.
3. Let f W S n ! S n be induced by an orthogonal matrix S 2 OnC1 .R/. Show that
deg.f / D det.S/.
4. Let f W S n ! S n be without fixed points. Show that f is homotopic to idS n and
deduce deg.f / D .1/nC1 .
5. Show that a continuous map f W S n ! S n is homotopic to a constant map if and only
if it is not surjective. Show that in this case there exist x; y 2 S n with f .x/ D x and
f .y/ D y.
6. Let G be a non-trivial group that acts freely on S n . Show that G Š Z=2Z if n is even.
Problem 11.10. Show the hairy ball theorem. Let X D S n with n even. Show that every
continuous section s of the tangent bundle TX ! X has a zero.
Hint: Consider s as a continuous map sW S n ! RnC1 . If s has no zero then one can assume
that s.x/ 2 S n for all x 2 S n . Then use Problem 11.9.
Appendix A: Basic Topology
12
In this appendix some basic notions in point set topology are recalled. We assume that the
reader has already some familiarity with the concepts in this chapter and in particular with
the notion of a metric space. For the convenience of the reader we included some proofs
such that this chapter might also serve as a rather brisk introduction to point set topology.
For a detailed introduction to this topic we refer to [BouGT1] and [BouGT2].
12.1 Topological Spaces
Definition 12.1 (Topological space). Let X be a set. A topology on X is a set T of

subsets of X such that T is stable under arbitrary unions and under finite intersections.
The pair .X; T /, usually denoted simply by X, is called a topological space and the
subsets in T are called open. A subset of X is called closed if its complement in X is
open.
If T and T 0 are topologies on X, then we say that T is coarser than T 0 or, equivalently,
that T 0 is finer than T if T T 0 .
Note that the properties of a topology T on a set X in particular imply that T contains
the union and the intersection of the empty family of sets in T . Hence every topology
contains the empty set ; and the set X.
If S is any set of subsets of a set X, then the intersection of all topologies on X contain-
ing S is a topology. This is then the smallest topology containing S, called the topology
generated by S.
Example 12.2 (Discrete topology). A simple example of topologies on a set X is the set
of all subsets of X. This topology is called the discrete topology on X. As every subset is
a union of subsets consisting of a single element, a topological space X carries the discrete

DOI 10.1007/978-3-658-10633-1_12
246 12 Appendix A: Basic Topology
topology if and only if the sets fxg are open for all x 2 X. In other words, the discrete
topology is the topology generated by f fxg I x 2 X g.
Example 12.3 (Metric spaces and normed spaces as topological spaces). Let .X; d /
be a metric space. For x0 2 X and r 2 R>0 set
Br .x0 / WD f x 2 X I d.x; x0 / < r g ; Br WD f x 2 X I d.x; x0 / r g :
Recall that a subset U of X is called open if for all x 2 U there exists r 2 R>0 such that
Br .x/ U . Then the open subsets form a topology on X, called the topology induced by
the metric d . In the sequel we endow every metric space with the induced topology.
A topological space X D .X; T / is called metrizable if there exists a metric d on X
such that T is the topology induced by d .
Every norm on a real or complex vector space V yields a metric and hence induces
a topology on V . Equivalent norms induce the same topology.
If V is finite-dimensional, then all norms are equivalent. Hence there exists in this
case a unique topology on V induced by some norm. We will usually endow V with this
topology.
Definition 12.4 (Neighborhoods and basis of a topology). Let X be a topological space.
1. Let Y X be a subset. A subset V of X is called a neighborhood of Y if there exists

an open subset U of X such that Y U V . A fundamental system of neighbor-
hoods of Y is a set S of neighborhoods of Y such that for every neighborhood V of
Y there exists W 2 S with W V . For x 2 X, a neighborhood of fxg (respectively
a fundamental system of neighborhoods of fxg) is also called a neighborhood of x
(respectively a fundamental system of neighborhoods of x).
2. A set B of open subsets of X is called a basis of the topology if every open subset of
X is a union of sets in B.
A subset B of the topology on X is a basis if and only if B contains a fundamental

system of open neighborhoods of every x 2 X.
Remark 12.5. Let X be a set. Let S be a set of subsets of X and let T be the topology
generated by S. Then T consists of arbitrary unions of finite intersections of sets in S. In
particular the set of intersections of all finite families of sets in S is a basis of the topology
T.
In a metric space the open balls Br .x0 /, for r 2 R>0 , x0 2 X, form a basis of the
topology.
12.2 Continuous, Open, and Closed Maps 247
12.2 Continuous, Open, and Closed Maps
Definition 12.6 (Continuous maps, homeomorphisms). Let X and Y be topological

spaces and let f W X ! Y be a map. Let x0 2 X. Then f is called continuous in x0 if for
every neighborhood V of f .x0 / its inverse image f 1 .V / is a neighborhood of x0 .
The map f is called continuous if the following equivalent conditions hold:
(i) The map f is continuous in all x0 2 X.

(ii) For every open subset V Y its inverse image f 1 .V / is open in X.
(iii) For every closed subset B Y its inverse image f 1 .B/ is closed in X.
A continuous map f W X ! Y is called a homeomorphism if there exists a continuous

map gW Y ! X such that g ı f D idX and f ı g D idY .
When we write that f W X ! Y is a continuous map, it will be always tacitly understood

that X and Y are topological spaces.
The composition of continuous maps is continuous. Let S be a set of subsets of Y
generating the topology on Y . Then a map f W X ! Y between topological spaces is
continuous if and only if f 1 .V / is open in X for all V 2 S.
A bijective continuous map f W X ! Y is not necessarily a homeomorphisms because
its inverse map f 1 W Y ! X might not be continuous (see Problem 12.10 for examples).
Example 12.7 (Continuity for metric spaces). For maps between metric spaces, con-
dition (i) for continuity can be expressed by the Weierstraß’ "-ı condition: A map
f W .X; d / ! .X 0 ; d 0 / between metric spaces is continuous if and only if for all " > 0 and
for all x0 2 X there exists ı > 0 such that d.x0 ; x/ < ı implies d 0 .f .x0 /; f .x// < " for
all x 2 X.
Example 12.8 (Continuity of linear maps). Let K be either R or C and let V and W
be finite-dimensional vector spaces over K endowed with the unique topology given by
a norm. Then it is a standard result from basic analysis that every linear map uW V ! W
is continuous. In particular, every bijective linear map is a homeomorphism.
More generally and more precisely, let V1 ; : : : ; Vr and W be finite-dimensional vector
spaces over K and let ˛W V1 Vr ! W be an r-multilinear map (i.e., ˛ is linear in
each component Vi ). Then there exist norms on Vi and W such that
jj˛.v1 ; : : : ; vr /jj jjv1 jj jjvr jj
for all vi 2 Vi , i D 1; : : : ; r. This implies that ˛ is continuous by Example 12.7.

Example 12.9 (Extended real line). Let R WD R [ f˙1g be the extended real line.
Define a map 8
ˆ
<1;
ˆ x D 1I
'W R ! Œ1; 1; '.x/ WD 1Cjxj ; x 2 RI
x
ˆ
:̂
1; x D 1:
It is easy to check that ' is bijective. We endow R with a metric via transport of structure,
i.e., we define
d W R R ! R; d.x; y/ WD j'.x/ '.y/j:
Then d is a metric on R and we obtain an induced topology on R. As ' preserves the
metrics (by definition), it is a homeomorphism of topological spaces.
It is easy to check, that R R is an open subset with respect to this topology and
that the topology induced on this subspace is the usual topology on R (but note that the
restriction of d is not the standard metric on R).
Definition 12.10 (Open and closed maps). Let X and Y be topological spaces and let
f W X ! Y be a map. Then f is called open (respectively closed) if for every open
(respectively closed) subset Z of X its image f .Z/ is open (respectively closed) in Y .
The composition of open (respectively closed) maps is again open (respectively

closed). Every homeomorphism is open and closed. Conversely, a bijective continu-
ous map is a homeomorphism if and only if it is open (or, equivalently, closed).
Let B be a basis of the topology on X. As the image of a union of subsets is the union
of the images of these subsets, a map f W X ! Y is open if and only if f .U / is open in Y
for all U 2 B.
Proposition 12.11. A continuous map f W X ! Y is closed if and only if for all y 2 Y

and all open neighborhoods U of f 1 .y/ in X there exists an open neighborhood V of y
in Y such that f 1 .V / U .
Proof. Let f be closed. Let y 2 Y , U an open neighborhood of f 1 .y/. Then B WD

f .X nU / is closed in Y and y … B. Take V WD Y nB. Then f 1 .V / D X nf 1 .B/ U .
Conversely, the condition is also sufficient. Let A X be closed. We have to show
that W WD Y n f .A/ is open in Y . Choose y 2 W . Then U WD X n A is an open
neighborhood of f 1 .y/. Hence there exists y 2 V Y open such that f 1 .V / U
and hence V W .
12.3 Closure and Interior 249
12.3 Closure and Interior
Definition 12.12 (Closure, interior, and boundary). Let X be a topological space, Y

X a subset. The set \
Y WD A
A closed
Y A
is called the closure of Y (in X). It is the smallest closed subset of X that contains Y . The
set [
Y ı WD U
U open
U Y
is called the interior of Y (in X). It is the greatest open subset of X that is contained in
Y . The set
@Y WD Y n Y ı
is called the boundary of Y (in X).
Remark 12.13. Let X be a topological space and let Y be a subset of X. Then
1. .X n Y /ı D X n Y .
2. Y D f x 2 X I for all neighborhoods U of x one has U \ Y ¤ ;g. In the definition
of the right-hand set it suffices if U runs through a fundamental system of neighbor-
hoods of x.
Example 12.14. Let .V; jj jj/ be a normed real or complex vector space. Then for all
r 2 R>0 and x 2 X one has
Br .x/ D Br .x/; @Br .x0 / D f x 2 V I jjx x0 jj D r g :
Analogous equalities do not hold in a general metric space (Exercise 12.5).
Proposition 12.15. Let X and Y be topological spaces and let f W X ! Y be a map. If

f is continuous, then for every subset A of X one has f .A/ f .A/.
The converse also holds (Problem 12.9).
Proof. If B is a closed subset of Y containing f .A/, then f 1 .B/ is closed in X and

contains A. Hence f 1 .B/ contains A, in other words f .A/ B. Therefore f .A/ is
contained in the intersection of all such B.
Definition and Remark 12.16 (Dense subsets). Let X be a topological space. A sub-
set Y of X is called dense in X, if the following conditions (which are equivalent by
Remark 12.13 2) are satisfied:
(i) The closure of Y in X is equal to X.

(ii) For every non-empty open subset U of X one has U \ Y ¤ ;.
12.4 Construction of Topological Spaces
Definition 12.17 (Inverse and direct image topology). Let .Xi /i 2I be a family of topo-
logical spaces.
1. Let Y be a set and let fi W Y ! Xi be maps

for i 2 I . Then the topology on Y gen-
S ˚
erated by i 2I fi1 .Ui / I Ui Xi open is called the inverse image of the topology
of the .Xi /i under .fi /i .
It is the coarsest topology on Y such that all maps fi are continuous. In other words,
it satisfies the following universal property: If hW W ! Y is a map from a topological
space W , then h is continuous if and only if fi ı h is continuous for all i 2 I .
T
2. Let Z be a set and gi W Xi ! Z be maps for i 2 I . Then i 2I f W Z I gi1 .W /
is open in Xi g is a topology on Z, called the direct image of the topology of the .Xi /i
under .gi /i .
It is the finest topology on Z such that all maps gi are continuous. In other words, it
satisfies the following universal property: If hW Z ! W is a map to a topological space
W , then h is continuous if and only if h ı gi is continuous for all i 2 I .
A particularly important example is the subspace topology.
Example 12.18 (Subspaces). If Y is a subset of topological space X and f W Y ! X

is the inclusion, then the inverse topology of X under f is simply called the induced
topology on Y and the topological space Y is called a subspace of X.
In this case, the open (respectively closed) subsets of Y are those of the form A \ Y
for A X open (respectively closed).
Let Y be a subset of a finite-dimensional vector space V over R or over C. Then we

always endow Y with the topology induced by the unique topology on V that is given by
some norm on V .
Definition 12.19 (Topological embedding). A map j W Y ! X between topological

spaces is called a topological embedding, if j yields a homeomorphism Y ! j.Y /,
where j.Y / is endowed with the topology induced by the topology on X. An embedding
j W Y ! X is called open (respectively closed) if j is an open (respectively a closed) map.
12.4 Construction of Topological Spaces 251
Remark 12.20.
1. A topological embedding is injective and continuous. The converse does not hold in
general (Problem 12.10).
2. An injective continuous map j W Y ! X is a topological embedding if and only if for
every open subset V of Y there exists an open subset U of X such that j 1 .U / D V .
3. The composition of two topological embeddings is again a topological embedding (by
2).
4. Let j W Y ! X be a continuous map such that there exists a continuous map pW Y ! X
with p ı j D idY (i.e., j is a section of p), then for every open V Y one has
j 1 .p 1 .V // D V . Hence j is a topological embedding by 2.
5. A topological embedding j W Y ! X is open (respectively closed) if and only if j.Y /
is an open (respectively a closed) subset of X.
A very important example of the direct image topology are quotient spaces.
Example 12.21 (Quotient spaces). Let X be a topological space and let gW X ! Z be

a surjective map. Then the direct image topology of g is also called the quotient topology
and the topological space Z is called a quotient space of X.
A subset A of Z is open (respectively closed) if and only if g 1 .A/ is open (respectively
closed) in X. The universal property of the quotient topology is the following. Let Y
be a topological space. Then a map f W Z ! Y is continuous if and only if f ı g is
continuous.
An other important special case of the inverse image topology is the product topology.
Definition and Remark 12.22 (Product space). Let I be a set and let .Xi /i 2I be a family
Q
of topological spaces. Let X WD i 2I Xi be the cartesian product and let pi W X ! Xi be
the i-th projection. The inverse image on X of the topology of the .Xi /i under .pi /i is
called the product topology and the set X endowed with this topology is called the product
of the topological spaces .Xi /i .
Hence the product topology is the coarsest topology such that the projections pi are
continuous for all i. A basis of the product topology is given by open subsets of the form
Q
i 2I Ui , where Ui Xi is open and Ui D Xi for all but finitely many i 2 I .
As openness of maps can be checked on a basis of the topology, one sees that the
projections pi are open for all i.
Proposition 12.23 (Product and closure). Let .Xi /i 2I be a family of topological spaces
Q
and let Ai Xi be subsets for all i. Then the closure of i 2I Ai in the product space
Q Q
i 2I Xi is i 2I Ai .
Q
Proof. Let x D .xi /i be in the closure of i 2I Ai . Then for all i 2 I its projection
Q
pi .x/ D xi lies in Ai (Proposition 12.15) and hence x 2 i Ai .
Q Q
Conversely, let y D .yi /i 2 i Ai . The sets of the form i Ui containing y and with
Ui Xi open for all i 2 i and Ui D Xi for all but finitely many i 2 I form a fundamental
system of neighborhoods of y. For every i 2 I the open neighborhood Ui of yi contains
Q Q
a point ai 2 Ai . Hence i Ui contains .ai /i 2 i 2I Ai . This shows that y is in the
Q
closure of i 2I Ai .
Corollary 12.24. Let .Xi /i 2I be a family of topological spaces and let Ai Xi be non-
Q Q
empty subsets for all i. Then i 2I Ai is closed (respectively dense) in i 2I Xi if and only
if Ai is closed (respectively dense) in Xi for all i 2 I .
Remark 12.25. Let .X1 ; d1 / and .X2 ; d2 / be metric spaces. Then
d..x1 ; x2 /; .y1 ; y2 // WD maxfd1 .x1 ; y1 /; d2 .x2 ; y2 /g
defines a metric on X1 X2 . It is easy to check that d induces the product topology. In

particular, the product of two metrizable spaces is again metrizable.
More generally, one can show that the product of countably many metrizable spaces is
again metrizable (Problem 12.11).
Definition and Remark 12.26 (Sum of topological spaces). Let .Xi /i 2I be a family of
topological spaces. Let X be the disjoint union of all the Xi and let i W Xi ! X be the
inclusion. Endow the set X with the direct image topology of the Xi under i . We obtain
a topological space X such that Xi X is open and closed for all i 2 I . The space X is
`
called the sum or the coproduct of .Xi /i 2I and it is denoted by i 2I Xi .
The sum of topological spaces is a special case of the following more general construc-
tion of gluing topological spaces. Let .Ui /i 2I be a family of topological spaces, for all
i; j 2 I fix a subset Uij Ui , and for all i; j 2 I a continuous map 'j i W Uij ! Uj i such
that:
(a) Ui i D Ui and 'i i D idUi for all i 2 I ,

(b) the cocycle condition holds: 'kj ı 'j i D 'ki on Uij \ Ui k for all triples .i; j; k/ of I .
In the cocycle condition we implicitly assume that 'j i .Uij \ Ui k / Uj k , such that the
composition is meaningful.
Proposition 12.27 (Gluing of topological spaces). Suppose that Uij is open (respectively
closed) in Ui for all i; j 2 I . Then there exists a topological space X together with
morphisms i W Ui ! X, such that:
1. The map i is an open (respectively a closed) embedding for all i 2 I .

2. j ı 'j i D i on Uij for all i; j .
12.4 Construction of Topological Spaces 253
S
3. X D i i .Ui /.
4. i .Ui / \ j .Uj / D i .Uij / D j .Uj i / for all i; j 2 I .
Furthermore, .X; . i /i / has the following universal property: If Z is a topological space,

and for all i 2 I , i W Ui ! Z is a continuous map such that j ı 'j i D i on Uij for all
i; j 2 I , then there exists a unique continuous map W X ! Z with ı i D i for all
i 2 I.
In particular, .X; . i /i / is uniquely determined up to unique isomorphism.
Proof. First note that the cocycle conditions for the triples .i; j; i/ and .j; i; j / show that
'ij and 'j i are mutually inverse homeomorphisms. To define the underlying topological
`
space of X, we start with the sum i 2I Ui of the Ui and define an equivalence relation
on it as follows. Points xi 2 Ui , xj 2 Uj , i; j 2 I , are equivalent, if and only if
xi 2 Uij , xj 2 Uj i and xj D 'j i .xi /. The conditions (a) and (b) imply that is in fact an
equivalence relation. As a set, define X to be the set of equivalence classes,
a
X WD Ui = :
i 2I
The natural maps i W Ui ! X are injective, we have i .Uij / D i .Ui / \ j .Uj / for all
i; j 2 I , and properties (2) and (3) hold. We endow X with the direct image topology
of the Ui under the i . Hence a subset V of X is open (respectively closed) if and only
if i1 .V / is open (respectively closed) in Ui for all i 2 I . The universal property then
follows from the universal property of the direct image topology.
To show that i is an open (respectively a closed) embedding we have to show that
for all i 2 I a subset Ai of Ui is open (respectively closed) in Ui if and only if for all
j 2 J the subset j1 . i .Ai // is open (respectively closed) in Uj . By choosing j D i one
sees that the condition is clearly necessary. Conversely, if Ai is open (respectively closed)
in Ui , then we have j1 . i .Ai // D 'j i .Ai \ Uij / by construction. This set is open
(respectively closed) in Uj because 'j i is a homeomorphism and Uj i is open (respectively
closed) in Uj .
The following construction of fiber products of topological spaces plays an important

role in several places in the book.
Definition 12.28 (Fiber products). Let f W X ! S and gW Y ! S be continuous maps

of topological spaces. Define
X S Y WD f .x; y/ 2 X Y I f .x/ D g.y/ g
endowed with the inverse topology of X and Y under the projections pW X S Y ! X

and qW X S Y ! Y . This is also the topology induced by the product topology on X Y .
The topological space X S Y is called the fiber product of X and Y with respect to f
and g.
Example 12.29. Let f W X ! S and gW Y ! S be continuous maps of topological spaces.

Let pW X S Y ! X and qW X S Y ! Y be the projections.
1. Suppose that S consists of a single point. Then X S Y D X Y .

2. Suppose that Y consists of a single point y and set s WD g.y/. Then p induces
a homeomorphism X S Y ! f 1 .s/, where the fiber f 1 .s/ is endowed with the
subspace topology of X.
3. More generally, for arbitrary f and g the projection p induces for all y 2 Y a home-
omorphism q 1 .y/ ! f 1 .g.y//.
4. Let hW X ! Y be a continuous map such that g ı h D f . Then its graph h W X !
X S Y , x 7! .x; h.x// is a topological embedding by Remark 12.20 4 because it is
a section of the projection X S Y ! X.
5. Applying 4 to the special case X D Y , h D idX , the graph of idX
f W X ! X S X D f .x; x 0 / I f .x/ D f .x 0 / g ; x 7! .x; x/
is called the diagonal of f . The map f is injective if and only if f is a homeomor-

phism.
12.5 Local Properties
Definition 12.30 (Coverings). Let X be a topological space and let Y be a subset of X.

S
A family .Ai /i 2I of subsets of X is called a covering of Y in X if Y i 2I Ai .
A covering .Ai /i 2I of Y in X is called an open covering (respectively a closed cover-
ing) if all Ai are open (respectively closed) in X.
In the sequel we will see many examples of properties that are local in the sense that
they can be checked after “restriction to an open covering”. Of course one has to explain
what this means precisely. We make this precise in some cases. Let P be a property of
maps of topological spaces.
We say that P is local on the target (respectively local on the source), if for every map
f W X ! Y of topological spaces and for every open covering .Vi /i 2I of Y (respectively
for every open covering .Ui /i 2I of X) the map f has property P if and only if its restriction
f 1 .Vi / ! Vi (respectively its restriction Ui ! Y ) has property P for all i. Let us give
some easy examples.
1. The following properties are local on the target and local on the source: “continuous”,
“open”.
2. The following properties are local on the target: “injective”, “surjective”, “bijective”,
“homeomorphism”, “closed”, “open topological embedding”, “closed topological em-
bedding”.
12.5 Local Properties 255
As the properties “open topological embedding” and “closed topological embedding”

are both local on the target, one sees by considering inclusions of subspaces that for every
topological space X and every open covering .Ui /i of X a subset A of X is open (re-
spectively closed) in X if and only if A \ Ui is open (respectively closed) in Ui for all
i 2 I.
In fact there is a stronger result. To formulate it we need another example of a local
notion, which plays an important role when the notion of paracompact spaces is defined.
Definition 12.31 (Locally finite covering). Let X be a topological space. A family of

subsets .A / 2 is called locally finite if there exists an open covering .Ui /i 2I of X such
that each Ui meets only finitely many of the A .
Proposition 12.32. Let .Ai /i 2I be a family of subsets of a topological space X satisfying

one of the following properties:
(i) The interiors Aıi of Ai cover X.

(ii) .Ai /i 2I is a locally finite closed covering of X.
Then a subset B of X is open (respectively closed) in X if and only if B \ Ai is open

(respectively closed) in Ai for all i.
Proof. We have to show that the condition is sufficient. As .X n B/ \ Ai D Ai n .B \ Ai /,

it suffices by passing to the complement to consider either the case that for all i the subset
B \ Ai of Ai is open or that for all i it is closed.
Suppose that (i) is satisfied and suppose that B \Ai is open in Ai for all i. Then B \Aıi
S
is open in Aıi and hence open in X. By hypothesis, B D i .B \ Aıi / and hence B is
open.
Suppose that (ii) is satisfied and suppose that B \ Ai is closed in Ai for all i. As Ai is
closed in X, B \ Ai is closed in X. To see that B is closed we may work locally on X.
The hypothesis implies that .B \ Ai /i is locally finite and working locally on X we may
S
assume that all but finitely many of the B \ Ai are empty. But then B D i .B \ Ai / is
closed.
Corollary 12.33. Let .Ai /i 2I be a locally finite family of closed subsets of a topological
S
space. Then i 2I Ai is closed in X.
Corollary 12.34. Let .Ai /i be a covering of a topological space X such that the interiors
Aıi of Ai cover X or which is a locally finite closed covering. Let Y be a topological space
and let f W X ! Y be a map such that f jAi W Ai ! Y is continuous for all i. Then f is
continuous.
We introduce two local notions of properties already defined.

Definition and Remark 12.35 (Locally closed subsets). Let X be a topological space
and let Z X be a subset. Then Z is called locally closed if the following equivalent
conditions are satisfied:
S
(i) There exists a family .Ui /i 2I of open subsets Ui X such that Z i 2I Ui and
such that Z \ Ui is closed in Ui .
(ii) Z is an open subset of its closure ZN in X.
(iii) Z is the intersection of an open and a closed subset of X.
Proof. “(ii) ) (iii) ) (i)” is clear.

“(i) ) (ii)”. Let .Ui /i as in (i). As Z \ Ui is closed in Ui , we have Z \ Ui D ZN \ Ui ,
S
N Hence Z D N
which is open in Z. i .Z \ Ui / is open in Z.
Definition 12.36 (Local homeomorphisms). A continuous map f W X ! Y is called a

local homeomorphism if there exists an open covering .Ui /i of X such that f jUi W Ui ! Y
is an open topological embedding.
In other words, f W X ! Y is a local homeomorphism if and only if there exist open

coverings .Ui /i of X and .Vi /i of f .X/ in Y such that f induces for all i a homeomor-
phism Ui ! Vi .
Remark 12.37. Let f W X ! Y be a continuous map.
1. Every local homeomorphism is an open map. In particular, every injective local home-
omorphism is an open embedding.
2. If f is a local homeomorphism, then the diagonal f W X ! X Y X is an open
embedding. Indeed, let .Ui /i be an open covering of X such that f jUi is an open
embedding for all i. Then f jUi is an open embedding (with image Ui Y Ui ) for all
i. As f is injective, this shows that f is an open embedding.
3. Let gW Y ! Z be a local homeomorphism and suppose that g ı f is an open embed-
ding. Then f is injective and a local homeomorphism. Hence f is an open embedding.
12.6 Hausdorff Spaces
Definition and Proposition 12.38 (Hausdorff spaces). A topological space X is called

Hausdorff or a T2 -space if it satisfies the following equivalent properties:
(i) For all x; y 2 X with x ¤ y there exist neighborhoods U of x and V of y such that
U \ V D ;.
(ii) The diagonal X WD f .x; x/ I x 2 X g is closed in X X.
(iii) For every topological space Y and for all continuous maps f1 ; f2 W Y ! X the set
f y 2 Y I f1 .y/ D f2 .y/ g is closed in Y .
12.7 Connected Spaces and Locally Constant Maps 257
Proof. “(i) ) (ii)”. For .x; y/ 2 .X X/ n X clearly one has x ¤ y so that there
exist neighborhoods U of x and V of y in X such that U \ V D ;. Therefore U V is
a neighborhood of .x; y/ in X X that is contained in .X X/nX . Hence .X X/nX
is open in X X.
“(ii) ) (i)”. For x; y 2 X with x ¤ y we have .x; y/ 2 .X X/nX . As .X X/nX
is open, there exists a neighborhood of .x; y/ in .X X/ n X and by definition of the
product topology we can chose it to be of the form U V , where U and V are open
neighborhoods of x and y respectively. Then U \ V D ;.
“(ii) ) (iii)”. Let f W Y ! X X be the continuous map y 7! .f1 .x/; f2 .x//. Since
X X X is closed and f is continuous, f 1 .X / D f y 2 Y I f1 .y/ D f2 .y/ g is
closed in Y .
“(iii) ) (ii)”. Applying (iii) to the projections p1 ; p2 W X X ! X yields that
f .x; y/ 2 X X I p1 ..x; y// D p2 ..x; y// g D X is closed in X X.
Every metrizable topological space is Hausdorff. Every subspace of a Hausdorff topo-

logical space is Hausdorff. If .Xi /i is a family of Hausdorff topological spaces, then
Q
i 2I Xi is Hausdorff.
Property (iii) in Proposition 12.38 for Hausdorff spaces implies the following result.
Proposition 12.39. Let X and Y be topological spaces, let f; gW Y ! X. Suppose that

X is Hausdorff and that there exists a dense subset D of Y such that f jD D g jD . Then
f D g.
12.7 Connected Spaces and Locally Constant Maps
Definition 12.40. A non-empty topological space X is called connected if X and ; are

the only subsets of X that are both closed and open.
Note that we explicitly excluded the empty space to be connected.
Proposition 12.41 (Images and products of connected spaces).
1. Let f W X ! Y be a continuous surjective map. If X is connected, then Y is connected.

Q
2. Let .Xi /i 2I be a family of topological spaces. Then the product space i 2I Xi is
connected if and only if Xi is connected for all i 2 I .
Proof. 1. Let A Y be open and closed. Then f 1 .A/ X is also open and closed
because f is continuous and we have f 1 .A/ D ; or f 1 .A/ D X because X is con-
nected. Since f is surjective, we have A D f .f 1 .A// and therefore A D ; or A D Y .
Thus Y is connected.
Q
2. Assume that i 2I Xi is connected. For all j 2 I the j -th projection pj W
Q
i 2I Xi ! Xj is continuous and surjective and hence Xj is connected by 1. Q
Conversely, assume that Xi is connected for every i 2 I . Let ; ¤ U i 2I Xi
be open and closed. Let x 2 X and y 2 U such that x and y differ in at most a single
component, i.e., there exists i0 2 I with xi D yi for every i 2 I n fi0 g. Then
8
Y <x ; i 2 I n fi gI
i 0
j W X i0 ! Xi ; z 7! .zi /i zi WD
:z; i D i0 I
i 2I
is a section of pi0 and therefore a topological embedding. By assumption Xi0 is connected

so that j.Xi0 / is connected as well. Hence, y 2 j.Xi0 /\U j.Xi0 / is open and closed so
that j.Xi0 / \ U D j.Xi0 /. Since x 2 j.Xi0 / we have x 2 U . Using induction, this result
can be extend to the case where x and y differ in finitely many components. Now there
Q
exists a non-empty open subset W U of the form W D i 2I Wi such that Wi D Xi
for all but finitely many i 2 I . This implies that for every point x 2 X there exists a point
y 2 W such that x and y differ only in finitely many components. Thus W D X by the
above argument and hence U D X. This shows that X is connected.
Remark 12.42. Let A be a connected subspace of a topological space X. Then every

subspace B of X with A B AN is connected. Indeed, A is dense in B and hence
every non-empty open and closed subset Z of B meets A. As A is connected we find
A \ Z D A and hence Z D B, again because A is dense in B.
Recall that an interval in R is a subset I of R such that for all x; y 2 I and z 2 R with
x < z < y one has z 2 I .
Proposition 12.43 (Connected subspaces of R). A subspace of R is connected if and

only if it is an interval.
Proof. Let A R be a connected subset and assume that A is not an interval. Then there
exist a; b 2 A and c 2 R n A with a < c < b. Then A \ .1; c/ and A \ .c; 1/ are
disjoint non-empty open subsets of A such that their union is A. Thus they are both open
and closed in A. Contradiction.
Let A R be an interval and assume that there exists U A open and closed with
; ¤ U ¤ A. Then A n U has the same properties and after possibly replacing U by A n U
we find a 2 U , b 2 A n U with a < b. Because A is an interval we have Œa; b A so that
V WD U \ Œa; b ¨ Œa; b is open, closed and non-empty. Then V is bounded and closed
in R so that ˇ WD sup.V / has to be an element of V . In particular this implies ˇ ¤ b
and it is clear that we must have ˇ ¤ a. Since V Œa; b is open and a < ˇ < b, there
exists > 0 with .ˇ ; ˇ C / V . This contradicts ˇ D sup.V /, hence A has to be
connected.
12.7 Connected Spaces and Locally Constant Maps 259
Proposition 12.44. Let X be a topological space, let .Ai /i 2I be a family of connected

T S
subspaces such that i Ai ¤ ;. Then A WD i 2I Ai is a connected subspace of X.
T
Proof. Let x 2 i Ai . Let U A be open and closed. After possibly passing to A n U
we may assume that x 2 U . Then U \ Ai is open and closed in Ai and non-empty. As Ai
is connected, U \ Ai D Ai for all i and hence U D A.
Definition 12.45 (Connected components). Let X be a topological space and let x 2 X.

Then the union of all connected subspaces containing x is again connected by Proposi-
tion 12.44. It is therefore the largest connected subspace containing x. It is called the
connected component of x.
Remark 12.46. Let X be a topological space.
1. For x; y 2 X write x y if y belongs to the component of x. This relation is clearly

reflexive and symmetric. Moreover, it is transitive by Proposition 12.44 and hence an
equivalence relation. The equivalence classes are the connected components of X.
2. As the closure of a connected set is again connected (Remark 12.42), the connected
components of X are closed.
In general, connected components are not open (Problem 12.20), see however Propo-
sition 2.13 (or Problem 12.21 and Problem 12.22) for a positive result.
Definition and Remark 12.47 (Locally constant maps). Let X be a topological space,
M a set, and tW X ! M a map.
1. The map t is called constant if t.x/ D t.y/ for all x; y 2 X.

2. The map t is called locally constant if the following two equivalent conditions are
satisfied:
(i) There exists an open covering .Ui /i of X such that t jUi is constant for all i 2 I .
(ii) The map t is continuous if we endow M with the discrete topology.
Proposition 12.48. A topological space X is connected if and only if every locally con-
stant map f on X is constant.
Proof. Let X be connected and let f W X ! S be locally constant. Then f .X/ is con-
nected in the discrete space S (Proposition 12.41 1) and hence consists of at most a single
point.
Conversely, assume that X is not connected. Hence there exist non-empty disjoint open
subsets A; B X whose union is X. Then the map f from X to a discrete space with
two elements a and b with f .A/ D fag and f .B/ D fbg is continuous.
12.8 Compact Spaces
Definition 12.49 ((Locally) compact spaces). Let X be a topological space.
1. X is called compact if every open covering of X has a finite subcovering.

2. X is called locally compact if every point has a fundamental system of compact neigh-
borhoods.
The definitions of “compact” and “locally compact” differ within the literature:
Often compact spaces as defined above are called quasi-compact. Then compact spaces
are defined as quasi-compact spaces that are also Hausdorff.
Locally compact spaces are very often defined as spaces such that each point has a com-
pact neighborhood – a condition that is strictly weaker than our definition of “locally
compact” (Problem 12.32). For Hausdorff spaces both of these definitions of “locally
compact” are equivalent (Proposition 12.54).
Example 12.50. Let V be a normed vector space over R or over C. Then a basic result
of analysis shows that V is locally compact if and only if V is finite-dimensional (see also
Problem 12.28). In this case the Heine–Borel theorem shows that a subset of V is compact
if and only if it is closed in V and bounded (see also Problem 12.27).
Proposition 12.51. Let X be a compact topological space.
1. Every closed subspace of X is compact.

2. Let f W X ! Y be a continuous surjective map of topological spaces. Then Y is
compact.
Proof. 1. Let X be a compact topological space, Y X a closed subspace and .Ui /i

an open covering of Y . By definition of the induced topology there exist Vi X open
with Vi \ Y D Ui so that we obtain an open covering of X by adding the open subset
X n Y to .Vi /i . Since X is compact, we find finitely many i1 ; : : : ; in such that X D
.X n Y / [ Vi1 [ [ Vin and therefore Y D X \ Y D Ui1 [ [ Uin is a finite subcovering
of .Ui /i .
2. Let .Vj /j 2J be an open covering of Y . Then .f 1 .Vj //j 2J is an open covering of
S
X. As X is compact, there exist j1 ; : : : ; jn 2 J such that X D niD1 f 1 .Vji /. Hence
Sn
Y i D1 Vji . Therefore Y is compact.
Proposition 12.52. Let f W X ! Y be a continuous map. Suppose that X is compact and

that Y is Hausdorff. Then the subspace f .X/ of Y is closed in Y and compact.
12.8 Compact Spaces 261
Proof. As Y is Hausdorff, the subspace f .X/ is Hausdorff. By Proposition 12.51 2,

f .X/ is also compact.
It remains to show that Y n f .X/ is open in Y . Let y 2 Y n f .X/. For all z 2
f .X/ there exist open neighborhoods Uz of z and Vz of y with Uz \ Vz D ; (because
Y is Hausdorff). Then .Uz /z is an open covering of f .X/. As f .X/ is compact we
S T
find z1 ; : : : ; zn 2 f .X/ with f .X/ 1i n Uzi . Then 1i n Vzi Y n f .X/ is
a neighborhood of y. Hence Y n f .X/ is open in Y .
Corollary 12.53. Let X be compact and let Y be Hausdorff. Then every injective con-
tinuous map f W X ! Y is a closed embedding. In particular, every bijective continuous
map X ! Y is a homeomorphism.
Proof. It suffices to show that f is a closed map. Let A X be a closed subspace.

As X is compact, A is compact (Proposition 12.51). Hence f .A/ is closed in Y by
Proposition 12.52.
Proposition 12.54. A Hausdorff topological space X is locally compact if and only if

every point of X has a compact neighborhood.
Proof. We have to show that the condition is sufficient. Let x 2 X and let C be a compact
neighborhood of x. As X is Hausdorff, C is closed in X by Corollary 12.53. Let V be
any neighborhood of x. We have to show that V contains a compact neighborhood of x.
Replacing V by the interior of V \ C we may assume that V C and that V is open
in X. Then C n V is closed in C and hence compact. As X and hence C is Hausdorff,
for each z 2 C n V there exist in C open neighborhoods Wz of z and Uz of x such that
Wz \ Uz D ;. As C n V is compact, there exist finitely many z1 ; : : : ; zn 2 C n V such
S T
that C n V W WD 1i n Wzi . Then U WD 1i n Uzi is an open neighborhood of
x in C not meeting W . Therefore its closure UN is contained in C n W and hence in V .
Moreover, UN is compact (being closed in C ) and a neighborhood because U is open in V
and V is open in X.
Proposition 12.55. Let X be a topological space and let S be a set of open subsets of X
that generates the topology on X. Then X is compact if and only if every covering of X
by elements in S has a finite subcovering.
Proof. The condition is clearly necessary. To show that it is sufficient assume that X is
not compact. Denote by T the topology on X. As X is not compact, the set
8 9
< [ [ =
U WD AT IX D U and U ¤ X for every finite subset A0 of A
: ;
U 2A U 2A0
is non-empty. The set is partially ordered by inclusion and if U0 is a totally ordered subset
S
of U, then A2U0 A 2 U. Hence we can apply Zorn’s lemma (see Proposition 13.28)
and see that there exists a maximal element A in U. Set B WD A \ S.
To obtain a contradiction it remains to show that B is a covering of X (then B is a cov-
ering of X by elements in S that has no finite subcovering, otherwise A also would have
S
a finite subcovering). Assume that there exists x 2 X n U 2B U . Choose U 2 A with x 2
T
U . As S generates the topology, we find V1 ; : : : ; Vr 2 S such that x 2 1i r Vi U . As
Vj contains x we have Vj … A. Now A [ fVj g has a finite subcovering by the maximality
of A, i.e., for all j we find a finite union Wj of elements in A such that Wj [ Vj D X.
T S S
Hence . 1j r Vj / [ . 1j r Wj / D X and hence X D U [ . 1j r Wj /. Hence A
has a finite subcovering. This contradiction proves that B is a covering.
Theorem 12.56 (Theorem of Tychonoff). Let .Xi /i 2I be a family of compact spaces.

Q
Then the product space i 2I Xi is compact.
Q Q
Proof. Let pi W X WD i 2I Xi ! Xi be the projection. The topology on i 2I Xi is
generated by open subsets of the form pi1 .Ui / with Ui Xi open. Let A be a cover-
ing consisting of such sets. By Proposition 12.55 it suffices to show that A has a finite
subcovering. ˚
For all i 2 I set Ai WD U Xi open I pi1 .U / 2 A . As A is a covering con-
sisting of sets of the form pi1 .Ui /, there exists j 2 I such that Aj is a covering of
S
Xj . As Xj is compact, there exist U1 ; : : : ; Ur 2 Aj such that Xj D 1kr Uk . Then
n o
pj1 .Uk / I k D 1; : : : ; r is an open covering of X, which is a subcovering of A by
definition of Aj .
12.9 Topological Groups
Definition 12.57 (Topological group). A topological group is a set G endowed with the
structure of a topological space and of a group such that the maps
G G ! G; .g; g 0 / 7! gg 0 ; G ! G; g 7! g 1
are continuous (where we endow G G with the product topology).
A group G endowed with a topology is a topological group if and only if the map
G G ! G, .g; g 0 / 7! gg 01 , is continuous.
Let G be a topological group, a 2 G. Then left translation g 7! ag and right translation
g 7! ga are homeomorphisms G ! G. In particular the topology of G is uniquely
determined by a fundamental system of neighborhoods of one element of G.
Let G be a topological group and let H G be a subgroup. We endow the set of
cosets G=H with the quotient topology (Example 12.21).
12.9 Topological Groups 263
Proposition 12.58. Let G be a topological group, H G a subgroup.
1. The canonical homomorphism G ! G=H is open.

2. The closure H is a subgroup of G.
3. If H is locally closed in G, then H is closed in G.
4. If H contains a non-empty open subset U of G, then H is open and closed in G.
5. H is open in G (respectively closed in G) if and only if G=H is discrete (respectively
Hausdorff).
6. Let G1 and G2 be topological groups and let Hi Gi be subgroups. Then the canoni-
cal bijective map ˛W .G1 G2 /=.H1 H2 / ! G1 =H1 G2 =H2 is a homeomorphism.
7. If H is a normal subgroup, G=H is a topological group.
Applying 5 to H D feg one sees that G is discrete (respectively Hausdorff) if and only
if feg is open (respectively closed) in G.
S
Proof. 1. Let U G be open. Then p 1 .p.U // D h2H hU is open in G. Hence
p.U / is open by definition of the quotient topology.
2. Let aW G G ! G be the continuous map .g; h/ 7! gh1 . Then
a.H H / D a.H H / a.H H / D H :
This shows 2.
3. We may assume that H is open and dense in G (by replacing G by the subgroup
H ). Then for all g 2 G the two cosets gH and H have non-empty intersection hence
they are equal, i.e., g 2 H .
S
4. We have H D h2H hU , hence H is open G. Therefore it is also closed by 3.
5. The quotient G=H is discrete if and only if gH is open in G for all g 2 G if and
only if H is open in G because left translation is a homeomorphism.
If G=H is Hausdorff, then eH 2 G=H is a closed point and its inverse image H in G is
closed. Conversely, if H is closed, then H D H eH is a closed point in the quotient space
H nG=H . Hence its inverse image under the continuous map G=H G=H ! H nG=H ,
.g1 H; g2 H / 7! Hg21 g1 H is closed. But this is the diagonal of G=H G=H .
6. The map ˛ is continuous by the universal property of the quotient topology (Ex-
ample 12.21). Moreover, G1 G2 ! G1 =H1 G2 =H2 is open by 1. Hence ˛ is also
open.
7. The composition of G G ! G, .g; g 0 / 7! gg 01 and of G ! G=H induces
a continuous map G=H G=H Š .G G/=.H H / ! G=H by the universal property
of the quotient topology (Example 12.21).
Corollary 12.59. Let G be a connected topological group and let U be a non-empty open
subset of G. Then U generates the group G.
Proof. By Proposition 12.58 4, the subgroup H generated by U is open and closed in G.

Hence H D G because G is connected.
Proposition 12.60. Let G be a topological group with neutral element e. Let G 0 be the
connected component of e. Then G 0 is a closed normal subgroup of G.
The subgroup G 0 is called the identity component of G.
Proof. For all g 2 G 0 the subset g 1 G 0 is connected and contains e. Therefore g 1 h 2

G 0 for all g; h 2 G 0 , which shows that G 0 is a subgroup. This subgroup is stable under
all homeomorphisms G ! G sending e to e, in particular G 0 is stable under inner auto-
morphisms. Thus G 0 is normal. Finally G 0 is closed because all connected components
are closed (Remark 12.46 2).
12.10 Problems
Problem 12.1. Classify (up to homeomorphism) all topological spaces with 2 points.
Problem 12.2. Let X be a set and let B be a set of subsets of X. Show that the following
(i) B is a basis of a topology on X.

S
(ii) One has X D U 2B U and for all U; V 2 B and x 2 U \ V there exists W 2 B with
x 2 W U \V.
Problem 12.3. Let X be a set. Show that d.x; y/ WD 0 for x D y and d.x; y/ WD 1
for x ¤ y defines a metric on X whose induced topology is the discrete topology. This
metric is also called the trivial metric.
Problem 12.4. Let p be a prime number. Fix a real number 0 < < 1. Define the p-adic
absolute value on Q by j0jp WD 0 and jp n ab jp WD n for n 2 Z, a; b 2 Z n f0g such that p
does not divide ab.
1. Show that the topology on Q (the p-adic topology) induced by the p-adic metric
dp .x; y/ WD jx yjp does not depend on the choice of .
2. Show that for different prime numbers p and ` the p-adic and the `-adic topology on
Q are different. Show that they are also different from the topology on Q induced by
usual absolute value.
Problem 12.5. Show that the trivial metric (Problem 12.3) and the p-adic metric on Q
(Problem 12.4) both give examples of metric spaces, where the closure of Br .x/ is not
Br .x/ in general.
12.10 Problems 265
Problem 12.6. Let .X; d / be a metric space. Show that d 0 W X X ! R, d 0 .x; y/ WD

minfd.x; y/; 1g is a metric on X and that d and d 0 induce the same topology on X.
Problem 12.7. A topological space is called a T0 -space or a Kolmogorov space if for

every pair of distinct points one of them has a neighborhood not containing the other. Let
X be a set, p 2 X. Define a subset U of X to be open if U D ; or if p 2 U . Show that
this defines a topology on X (called the particular point topology on X). Show that X is
a T0 -space and that the closure of fpg is X (in particular fpg is not closed if X has more
than a single element).
Problem 12.8. Let X be a topological space. Write x y if an open subset of X

contains x if and only if it contains y. Show that is an equivalence relation and that
X= , endowed with the quotient topology, is a T0 -space.
Problem 12.9. Let X and Y be topological spaces. Let f W X ! Y be a map. Show that
the following assertions are equivalent:
(i) f is continuous.
(ii) For every subset A of X one has f .A/ f .A/.
Problem 12.10. Show that the following maps are bijective and continuous but not
a homeomorphism:
1. The map idR W Rd ! R, where Rd is the set of real numbers endowed with the discrete
topology.
2. The map Œ0; 1/ ! S 1 WD f z 2 C I jzj D 1 g, t 7! e 2i t .
Problem 12.11. Let .Xn ; dn /n be a countable family of metric spaces each also consid-
Q
ered as a topological space. Let X D n Xn be the product space. Show that
X 1 dn .xn ; yn /
d W X X ! R; .x; y/ 7!
n
2n 1 C dn .xn ; yn /
defines a metric on X that induces the product topology.
1. Let Z Y X be subspaces. Show that Z is dense in X if and only if Z is dense in

Y and Y is dense in X.
2. Let Y be a subspace and let .Ui /i be an open covering of X. Show that Y is dense in
X if and only if Y \ Ui is dense in Ui for all i.
Problem 12.13. Let f W X ! Y be a continuous map. Let .Bi /i be a family of sub-

sets of Y and suppose (a) that the interiors cover Y or (b) that .Bi /i is a locally finite
closed covering. Show that f is open (respectively closed) if and only if its restriction
f 1 .Bi / ! Bi is open (respectively closed) for all i.
Problem 12.14. Let .Xi /i 2I be a family of non-empty topological spaces. Show that the
Q
product space i 2I Xi is Hausdorff if and only if Xi is Hausdorff for all i 2 I .
Problem 12.15. Let X be a topological space and let .Ui /i 2I be an open covering.
1. Give an example where Ui is Hausdorff for all i 2 I but X is ˚not Hausdorff.

2. Show that X is Hausdorff if and only if for all i; j 2 I the set .x; x/ I x 2 Ui \ Uj
is closed in Ui Uj .
Problem 12.16. Let X be a set. Define a subset U of X to be open if U D ; or if X n U

is a finite set.
1. Show that this defines a topology on X. It is called the cofinite topology on X.

2. Let X be endowed with the cofinite topology. Show that the following assertions are
equivalent:
(i) X is discrete.
(ii) X is finite
(iii) X is Hausdorff.
3. Show that X is compact.
4. Let X be infinite. Show that every non-empty open subspace of X is connected.
Problem 12.17. Let f W X ! Y be a surjective map of topological spaces. Suppose that

f is open or closed. Show that Y has the quotient topology induced by X.
Problem 12.18. Let f W X ! Y be a surjective map of topological spaces and suppose

that Y has the quotient topology induced by X. Show that if Y is connected and all fibers
of f are connected, then X is connected.
Problem 12.19. A topological space X is called totally disconnected if the connected

component of each point consists of the point alone.
1. Show that the subspace Q of R is totally disconnected.

2. Show that Q endowed with the p-adic topology (Problem 12.4) is totally disconnected.
3. Let X be a topological space, let XN be the set of connected components endowed with
the quotient topology with respect to the natural map X ! X. N Show that XN is totally
disconnected.
12.10 Problems 267
Problem 12.20.
1. Let X be a topological space. Show that the connected component of x 2 X is con-

tained in the intersection of all open and closed subsets of X containing x.
2. Consider the following subspace of R2 :
X WD f .1=n; y/ I n 2 N; y 2 Œ1; 1 g [ f .0; y/ I y 2 Œ1; 1 n f0g g :
Show that there exists x 2 X whose connected component is different from the inter-
section of the open and closed subsets of X containing x.
Problem 12.21. Let X be a topological space. Show that the following three conditions
are equivalent:
(i) The connected components of X are open.

(ii) For every x 2 X there exists a neighborhood of x that is contained in every subset
that is both open and closed and that contains x.
(iii) For every x 2 X the intersection of all open and closed subsets containing x is an
open subset of X.
Problem 12.22. A topological space X is called locally connected if every point of X

has a fundamental system of connected neighborhoods.
1. Show that a topological space X is locally connected if and only if every connected
component of every open subspace is open.
2. Show that every quotient space of a locally connected space is again locally connected.
Problem 12.23. Let .Xi /i 2I be a family of non-empty topological spaces and let X D
Q
i Xi be the product space. Show that X is locally connected (Problem 12.22) if and only
if each Xi is locally connected and Xi is connected for all but finitely many i.
Problem 12.24. Let .X; / be a totally ordered set (Definition 13.20) and endow X
with the topology generated by sets of the form X<a WD f x 2 X I x < a g and X>a WD
f x 2 X I x > a g with a 2 X. This topology is called the order topology on X.
1. Show that a basis of the topology is given by the open intervals, i.e., sets of the form
X<a , X>a and .a; b/ WD f x 2 X I a < x < b g for a; b 2 X.
2. Show that X is a Hausdorff space.
3. Show that X is totally disconnected (Problem 12.19) if X is well ordered (Defini-
tion 13.24).
Problem 12.25. Let .X; d / be a metric space. Show that if C X is compact, then C is
closed and bounded in X. Show that the converse does not hold in general.
Problem 12.26. Let a; b 2 R with a < b. Show that Œa; b is compact (without using the
Heine–Borel Theorem).
Problem 12.27. Let V be a finite-dimensional K-vector space and let jj jj be a norm on V .

Show the Heine–Borel Theorem: A subspace of V is compact if and only if it is closed in
V and bounded.
Problem 12.28. Let .V; jj jj/ be a normed K-vector space. Show that the following asser-
tions are equivalent:
(i) V is locally compact.

(ii) B1 .0/ is compact.
(iii) V is finite-dimensional.
Sn
Hint: Show that if B1 .0/ i D1 B 1 .vi /, then fv1 ; : : : ; vn g is a generating system of V
2
to see that (ii) implies (iii).
Problem 12.29 (One-point compactification or Alexandroff compactification). Let X

be a topological space and set X WD X [ f1g, with 1 … X. Define a subset U of X
to be open if U is an open subset of X or if U contains 1 and X n U is a closed and
compact subspace of X.
1. Show that this defines a topology on X , that the inclusion X ! X is an open embed-
ding, and that X is compact. The space X is called the one-point compactification
or Alexandroff compactification of X.
2. Show that X is Hausdorff if and only if X is Hausdorff and locally compact.
Problem 12.30. Show that no point of Q (considered as a subspace of R) has a compact

neighborhood. In particular Q is not locally compact.
Problem 12.31. Let X be a topological space and let A be a subset of X. Then A is

called relatively compact in X if A is contained in a compact subspace of X.
Now suppose that X is Hausdorff. Show that a A is relatively compact in X if and only
if AN is compact.
12.10 Problems 269
Problem 12.32. Consider the following conditions on a topological space X:
(i) Every point of X has a compact neighborhood.

(ii) Every point of X has a closed compact neighborhood.
(iii) X is locally compact.
1. Show that (ii) and (iii) each imply (i).

2. Show that the one-point compactification of Q (Problem 12.29) satisfies (ii) but not
(iii).
3. Show that the particular point topology on an infinite set (Problem 12.7) satisfies (iii)
but not (ii).
4. Suppose that X is Hausdorff. Show that (i), (ii), and (iii) are equivalent.
Problem 12.33. Let X be a Hausdorff space.
1. Show that every locally compact subspace of X is locally closed.

2. Suppose that X is locally compact. Show that every locally closed subspace of X is
locally compact.
Problem 12.34. Let .Xi /i be a family of non-empty topological spaces. Show that the
Q
product space i Xi is locally compact and Hausdorff if and only if each space Xi is
locally compact and Hausdorff and Xi is compact for all but finitely many i.
Problem 12.35. Let X be a topological space. A subset A of X is called nowhere dense

N ı D ;. Show that the union of a locally finite family of nowhere dense subsets is
if .A/
again nowhere dense. Show that the boundary of a closed or of an open subset is nowhere
dense.
Problem 12.36. Let G be a topological group and let H be a normal subgroup. Show
that its closure HN is a normal subgroup of G.
Problem 12.37. Let G be a topological group and let G 0 be its identity component. Show
that G=G 0 is totally disconnected (Problem 12.19).
Problem 12.38. Let G be a Hausdorff locally compact topological group and let H be
a closed subgroup.
1. Show that G=H is Hausdorff and locally compact.

2. Show that the identity component G 0 of G is the intersection of all open subgroups of
G.
3. Show that the connected components of G=H are the closures of the images of the
connected components of G.
Appendix B: The Language of Categories
13
The goal of this appendix is to make the reader familiar with the language of categories.
It is by no means a rigorous introduction to category theory. We will explain the most
important notions and give many examples. There will be almost no proofs in this chapter.
Many of the claims are easy to check and are left to the reader as “exercise”. A complete
exposition can for instance be found in the book [KS2] by M. Kashiwara and P. Schapira.
In particular it contains proofs of the statements that we leave as “exercises”. For some of
the more difficult results explicit references are given.
We start by introducing categories and functors. Important concepts are the notion of an
equivalence of categories and, more generally, the notion of adjoint functors. We then give
a short reminder on several notions of ordered sets and prove the principles of transfinite
induction and transfinite recursion. In the last section we introduce limits and colimits in
categories. This allows us to unify several important constructions such as kernels, prod-
ucts, fiber products, and final objects (which are all special cases of limits) or cokernels,
coproducts, pushouts, and initial objects (which are all special cases of colimits).
13.1 Categories
Definition 13.1. A category C consists of:1
1
Here we ignore all set-theoretical issues. To avoid set-theoretical difficulties one should work
with a fixed universe in the sense of [KS2] Definition 1.1.1 and assume that the sets of morphisms
between two objects always lies in the given universe. Then a category C is called small if Ob.C /
is in that universe. Moreover it then will be sometimes necessary to pass to a bigger universe, for
instance when considering the category of functors between two categories (see Definition 13.13).
Finally one adds also the axiom that every set is an element of some universe to the axioms of
Zermelo–Fraenkel set theory. We refer to [SGA4] Exp. I for details. Alternatively one can also
work with classes as explained in [Sch].

DOI 10.1007/978-3-658-10633-1_13
272 13 Appendix B: The Language of Categories
(a) a set Ob.C / of objects,

(b) for any two objects X and Y a set HomC .X; Y / D Hom.X; Y / of morphisms from
X to Y ,
(c) for any three objects X, Y , Z a composition map
Hom.X; Y / Hom.Y; Z/ ! Hom.X; Z/; .f; g/ 7! g ı f
such that all morphisms sets are disjoint (hence every morphism f 2 Hom.X; Y / has
unique source X and a unique target Y ) and such that:
1. the composition of morphisms is associative (i.e., for all objects X, Y , Z, W and for all
f 2 Hom.X; Y /, g 2 Hom.Y; Z/, h 2 Hom.Z; W / one has h ı .g ı f / D .h ı g/ ı f ,
2. for all objects X there exists a morphism idX 2 Hom.X; X/, called the identity of X
such that for all objects Y and for all morphisms f 2 Hom.X; Y / and g 2 Hom.Y; X/
one has f ı idX D f and idX ıg D g.
A category is called finite if it has only finitely many objects and morphisms.
Remark 13.2. For every object X of a category C the morphism idX is uniquely deter-
Q X are identities of X, then id
mined by Condition 2 for if idX and id Q X D id
Q X ı idX D idX .
Definition and Remark 13.3. A morphism f W X ! Y in a category is called an isomor-

phism if there exists a morphism gW Y ! X such that f ı g D idY and g ı f D idX . We
often write f W X ! Y to indicate that f is an isomorphism. We also write X Š Y and
say that X and Y are isomorphic if there exists an isomorphism X ! Y .
An morphism (respectively an isomorphism) with the same source and target X is
called an endomorphism (respectively an automorphism) of X. Composition yields the
structure of a monoid on the set EndC .X/ of endomorphisms of X and the structure of
a group on the set AutC .X/ of automorphisms.
Definition 13.4. A subcategory of a category C is a category C 0 such that every object of

C 0 is an object of C and such that HomC 0 .X 0 ; Y 0 / HomC .X 0 ; Y 0 / for any pair .X 0 ; Y 0 /
of objects of C 0 , compatibly with composition of morphisms and identity elements. The
subcategory C 0 is called full if HomC 0 .X 0 ; Y 0 / D HomC .X 0 ; Y 0 / for all objects X 0 and Y 0
of C 0 .
Example 13.5.
1. (Sets) the category of sets: Objects are sets, for two sets X and Y a morphism is a map
X ! Y , composition in the category is the usual composition of maps, and the identity
of a set X is the usual identity map idX . An isomorphism in (Sets) is simply a bijective
map.
13.1 Categories 273
2. (Grp) the category of groups: Objects are groups, morphisms are group homomor-
phisms, composition is the usual composition of group homomorphisms, and the iden-
tity is the usual identity. An isomorphism in (Grp) is a group isomorphism.
3. (Ab) is the full subcategory of (Grp) of abelian groups: Objects are abelian groups,
morphisms are group homomorphisms, composition is the usual composition of group
homomorphisms, and the identity is the usual identity.
4. Fix a group G. Then (G-Sets) denotes the category of G-sets: Objects are sets with
a left action by the group G and morphisms are G-equivariant maps. By (Sets-G) we
denote the category of sets with a right action by G.
5. (Ring) the category of rings: Objects are rings (always assumed to have a unit) and
morphisms are ring homomorphisms (preserving the unit).
6. (Top) the category of topological spaces: Objects are topological spaces and mor-
phisms are continuous maps. An isomorphism is a homeomorphism.
7. Let R be a ring. Then (R-Mod) denotes the category of left R-modules: Objects are
left R-modules, morphisms are R-linear maps. An isomorphism is a bijective R-linear
map because its inverse is automatically R-linear again.
If R D k is a field, we obtain the category of k-vector spaces, usually denoted by
(k-Vec) instead of (k-Mod).
Definition 13.6. Let C be a category. A morphism f W X ! Y in C is called a monomor-

phism (respectively epimorphism) if for all objects Z in C and all morphisms g; hW Z ! X
(respectively g; hW Y ! Z) one has
f ıg Df ıh)g Dh (respectively g ı f D h ı f ) g D h).
Example 13.7. In the category of sets (respectively of groups), monomorphisms are in-
jective maps (respectively group homomorphisms) and epimorphisms are surjective maps
(respectively group homomorphisms).
In the category of rings the inclusion Z ! Q is a monomorphism and an epimor-
phism. This gives in particular an example of a morphism that is a monomorphism and an
epimorphism but not an isomorphism.
Definition 13.8. For every category C the opposite category, denoted by C opp , is the
category with the same objects as C and where for two objects X and Y of C opp we set
HomC opp .X; Y / WD HomC .Y; X/ with the obvious composition law.
Definition 13.9. Let C and D be categories. We define the product category C D by

Ob.C D/ WD Ob.C / Ob.D/ and HomC D ..X1 ; Y1 /; .X2 ; Y2 // WD HomC .X1 ; X2 /
HomC .Y1 ; Y2 /. Composition of morphisms is defined componentwise.
13.2 Functors
Definition 13.10. Given categories C and D, a (covariant) functor F W C ! D is given by

attaching to each object C of C an object F .C / of D, and to each morphism f W C ! C 0
in C a morphism F .f /W F .C / ! F .C 0 /, compatible with composition of morphisms, i.e.,
F .g ı f / D F .g/ ı F .f / whenever the composition is defined, and preserving identity
elements, i.e., F .idC / D idF .C / .
If F W C ! D and GW D ! E are functors, we write G ıF W C ! E for the composition.
A covariant functor C opp ! D is also called a contravariant functor from C to D.
By a “functor” we will always mean a covariant functor. If we write F W C ! D, then

we always mean that F is covariant. A contravariant functor F from C to D will always
be denoted by F W C opp ! D. It attaches to each object C in C an object F .C / in D
and to each morphism f W C ! C 0 in C a morphism F .f /W F .C 0 / ! F .C / such that
F .g ı f / D F .f / ı F .g/ and F .idC / D idF .C / .
Remark 13.11. Let F W C ! D be a functor. Let C and C 0 be objects in C and let

f W C ! C 0 be an isomorphism in C . Then F .f / is an isomorphism.
Indeed, as f is an isomorphism, there exists a morphism gW C 0 ! C such that g ı f D
idC and f ı g D idC 0 . Hence F .g/ ı F .f / D F .g ı f / D F .idC / D idF .C / and similarly
F .f / ı F .g/ D idF .C 0 / . Therefore F .f / is an isomorphism.
Example 13.12. A simple example is the functor that “forgets” some structure. An ex-
ample is the functor F W (Grp) ! (Sets) that attaches to every group the underlying set and
that sends every group homomorphism to itself (but now considered as a map of sets).
Definition and Remark 13.13. For two functors F; GW C ! D we call a family of mor-
phisms ˛.S/W F .S/ ! G.S/ for every object S of C functorial in S if for every morphism
f W T ! S in C the diagram
˛.T /
F .T / G.T /
F .f / G.f /
˛.S /
F .S/ G.S/
commutes. We also say that ˛ is a morphism of functors F ! G.

If ˇW G ! H is a second morphism of functors, we define the composition ˇ ı ˛ by
.ˇ ı ˛/.S/ D ˇ.S/ ı ˛.S/. With this notion of morphism we obtain the category of
all functors from C to D, denoted by Func.C ; D/ or DC . The identity idF is given by
idF .S / , where S runs through all objects of C . In particular we obtain the notion of an
isomorphism F ! G of functors.
13.2 Functors 275
Example 13.14. Let k be a field.
1. For every k-vector space V let V _ WD Homk .V; k/ be its dual space and for every
k-linear map uW V ! W let u_ W W _ ! V _ , 7! ı u be its dual homomorphism.
We obtain a functor . /_ W (k-Vec)opp ! (k-Vec).
2. For every k-vector space V one has the biduality homomorphism of k-vector space
V W V ! .V _ /_ ; v 7! . 7! .v//; v 2 V; 2 V _:
Moreover, if uW V ! W is k-linear, then the diagram
V
V .V _ /_
_
u .u /_
W
W .W _ /_
is commutative. In other words, the V for V a k-vector space form a morphism of

functors id(k-Vec) ! ._ ı _ /.
3. If V is a finite-dimensional vector space, then V _ if finite-dimensional and V is an iso-
morphisms. Hence . /_ restricts to a functor (k-vec) ! (k-vec), where (k-vec) denotes
the full subcategory of (k-Vec) of finite-dimensional k-vector spaces, and V 7! V is
an isomorphism of functors id(k-vec) ! ._ ı _ /.
Definition 13.15. Let F W C ! D be a functor.
1. F is called faithful (respectively fully faithful) if for all objects X and Y of C the
map HomC .X; Y / ! HomD .F .X/; F .Y //, f 7! F .f / is injective (respectively
bijective).
2. F is called essentially surjective if for every object Y of D there exists an object X of
C and an isomorphism F .X/ Š Y .
3. F is called an equivalence of categories if it is fully faithful and essentially surjective.
There are analogous notions for contravariant functors. A contravariant functor that is an
equivalence of categories is sometimes also called an anti-equivalence of categories.
Theorem and Definition 13.16. A functor F W C ! D is an equivalence of categories

if and only if there exists a quasi-inverse functor G, i.e., a functor GW D ! C such that
G ı F Š idC and F ı G Š idD .
Proof. [KS2] Proposition 1.3.13.

Definition 13.17. Let C and D be categories and let F W C ! D and GW D ! C be

functors. Then G is said to be right adjoint to F and F is said to be left adjoint to G if
for all objects X in C and Y in D there exists a bijection
HomC .X; G.Y // Š HomD .F .X/; Y /;
which is functorial in X and in Y .
Example 13.18. If F W C ! D is an equivalence of categories, then a quasi-inverse func-

tor GW D ! C is right adjoint and left adjoint to F . Indeed for all objects X in C and Y
in D we have bijections, functorial in X and in Y
HomC .X; G.Y // ! HomC .G.F .X//; G.Y // ! HomD .F .X/; Y /:
This shows that F is left adjoint to G. A similar argument shows that F is also right
adjoint to G.
Example 13.19. Let k be a field and let F W (k-Vec) ! (Sets) be the forgetful functor.
Then F is right adjoint to the functor G that sends a set I to the k-vector space
k .I / D f .xi /i 2I I xi 2 k, xi D 0 for all but finitely many ig
and which sends a map aW I ! J to the unique k-linear map G.a/W k .I / ! k .J / such
that G.a/.ei / D ea.i / where .ei /i 2I (respectively .ej /j 2J ) is the standard basis of k .I /
(respectively k .J / ).
Indeed for every set I and every k-vector space V we have a bijection, functorial in I
and in V ,
Hom(k-Vec) .k .I / ; V / ! Hom(Sets) .I; V /; f 7! .i 7! f .ei //:
13.3 Digression: Ordered Sets, Transfinite Induction,

and Zorn’s Lemma
Definition 13.20 (Ordered sets). Let I be a set.
1. A relation on I is called a partial preorder or simply a preorder, if i i for all

i 2 I and i j , j k imply i k for all i; j; k 2 I .
2. A partial preorder on I is called a partial order or simply an order if i j and
j i imply i D j for all i; j 2 I .
3. A partial preorder is called filtered if I ¤ ; and if for all i; j 2 I there exists a
k 2 I with i k and j k.
13.3 Digression: Ordered Sets, Transfinite Induction, and Zorn’s Lemma 277
4. A partial order is called a total order if for all i; j 2 I one has i j or j i.
If is a partial preorder on a set I and i; j 2 I , we write i < j if i j and i ¤ j ; we

write i j if j i, and i > j if j < i. We also set Ii WD f j 2 I I j i g. Similarly
we define I<i , Ii , and I>i .
If .I; / is a preordered set, then every subset J endowed with the induced relation
(i.e., with the restriction of to J J ) is again preordered. If I is partially (respectively
totally) ordered, so is J . Every total order is filtered.
Example 13.21.
1. Let I be any set. The discrete order in I is defined by i j , i D j . This is

a partial order. It is neither filtered nor a total order if I consists of more than one
element.
2. The chaotic order on a set I is the preorder such that i j for all i; j 2 I .
3. The real numbers together with the usual ordering form a totally ordered set.
4. For n; m 2 N we write njm if n divides m (i.e., there exists k 2 N such that m D k n).
Then j is a filtered partial order on N, which is not a total order.
5. Let X be a topological space, let x 2 X. Let U.x/ be the set of open neighborhoods
of x in X. Endow U.x/ with the opposite of the inclusion relation, i.e., U V if
V U . Then is a filtered partial order on U.x/.
Definition 13.22. Let .I; / be a partially ordered set.
1. An element i0 2 I is called minimal (respectively maximal) if there exists no element

i 2 I with i < i0 (respectively i > i0 ).
2. An element i0 2 I is called the smallest element (respectively the greatest element) if
i i0 (respectively i i0 ) for all i 2 I .
3. Let S I be a subset. An element i0 2 I is called upper bound of S (respectively
lower bound of S) if i i0 (respectively i i0 ) for all i 2 S.
A smallest and a greatest element of a partially ordered set is unique (if it exists). If
I is totally ordered, then every minimal (respectively maximal) element is the smallest
(respectively greatest) element.
Example 13.23. Consider the sets N and N2 with the dividing order (Example 13.21 4).
In N2 the prime numbers are the minimal elements but N2 has no smallest element. In
N, 1 is a smallest element with respect to j.
Definition 13.24 (Well orders). A partial order on a set I is called a well order if every
non-empty subset of I has a smallest element.
A well order is always a total order: For i; j 2 I , the set fi; j g has a smallest element.
Every subset of a well-ordered set is again well-ordered with the induced order. Every
element i of a well-ordered set is either the greatest element or it has a successor iC,
which is defined to be the smallest element of f j 2 I I j > i g. We then have I<i C D
Ii .
The natural numbers with the usual order are a well-ordered set. On the real numbers
the usual order is not a well order (the open interval .0; 1/ has no smallest element). We
have the following principle of transfinite induction2 .
Proposition 13.25 (Transfinite induction). Let .I; / be a partially ordered set such
that every non-empty subset of I has a minimal element (e.g., if .I; / is well ordered).
Let J I be a subset such that for all i 2 I one has
I<i J ) i 2 J:
Then I D J .
Very often transfinite induction is applied to prove that a property P.i/, depending on
i 2 I , holds for all i 2 I by setting J WD f i 2 I I P.i/ holdsg. The special case I D N
with the usual order corresponds to the standard mathematical induction.
Proof. Assume I n J ¤ ;. By hypothesis there exists a minimal element i0 in I n J . But

then I<i0 J and hence i0 2 J by our hypothesis. This is a contradiction.
A method related to transfinite induction is the method of transfinite recursion, which

we state as follows.
Proposition 13.26 (Transfinite recursion). Let .I; / be a well-ordered set, let X be

a set, and let W be the set of pairs .i; f i /, where i 2 I and f i W I<i ! X is a map. Let
S W be a subset and let GW S ! X be a map satisfying the following two conditions:
(a) If .i; f i / 2 S, then .j; f i jI<j / 2 S for all j 2 I<i .

(b) If .i; f i / 2 S and i is not the greatest element of I , then .iC; .f i /C / 2 S, where
.f i /C W I<i C ! X with .f i /CjI<i WD f i and .f i /C .i/ WD G.i; f i /.
Then there exists a unique map f W I ! X such that .i; f jI<i / 2 S and f .i/ D G.i; f jI<i /
for all i 2 I .
2
Partial orders satisfying the hypothesis of Proposition 13.25 are also called noetherian. Hence
Proposition 13.25 is also called the principle of noetherian induction.
13.3 Digression: Ordered Sets, Transfinite Induction, and Zorn’s Lemma 279
For instance, to construct a family of subsets .Ai /i 2I of some set Z indexed by a well-
ordered set, it suffices to construct Ai from .Aj /j <i : Let X be the power set of Z, consider
.Aj /j <i as a map Ai W I<i ! X, and let G be the map that attaches to .i; Ai / the set Ai .
Often the construction of Ai from .Aj /j <i will not be possible for an arbitrary family
.Aj /j <i of subsets and one has to restrict to a certain subset S of such families.
Proof. Uniqueness. Let f ¤ fQ be two such maps and let i 2 I be the smallest ele-
ment such that f .i/ ¤ fQ.i/. Then f .i/ D G.i; f jI<i / D G.i; fQjI<i / D fQ.i/. This is
a contradiction.
Existence. Let us call a subset J of I a segment if j 2 J , i 2 I , and i j imply
i 2 J . Clearly the union of segments is again a segment. If J ¤ I let i 2 I be the
smallest element in I n J . Then J D I<i . In other words, the segments ¤ I are the
subsets J D I<i for some i 2 I . The set of segments of I is well ordered by inclusion.
Let S be the set of segments J of I such that there exists a map fJ W J ! X such that
.j; f jI<j / 2 S and f .j / D G.j; f jI<j / for all j 2 J . As J is well ordered, the above
uniqueness argument shows that fJ is unique if it exists. If J K are segments in S,
then fK jJ D fJ . Hence the union of segments in S is again a segment in S. Moreover
suppose I<i 2 S for some i 2 I . Then I<i C 2 S with fI<i C WD .fI<i /C .
We conclude by showing that S contains every segment of I . Assume that there exists
a segment J of I with J … S. As the set of segments is well ordered we may assume that
J is the smallest segment with J … S. If J has no greatest element, then J is the union
of the smaller segments J<j for j 2 J . This is not possible because S is stable under
union. But if J has greatest element j , then J D J<j C 2 S because J<j 2 S. This is
a contradiction.
Transfinite induction and recursion is particularly useful because of the following re-
sult.
Proposition 13.27 (Well-ordering theorem). Every set can be endowed with a well or-
der.
The well-ordering theorem is equivalent to the axiom of choice and also to Zorn’s
Lemma:
Proposition 13.28 (Zorn’s Lemma). Let .I; / be a ordered set such that every well-
ordered subset S has an upper bound in I . Then there exist maximal elements in I .
Very often Zorn’s Lemma is formulated with the stronger hypothesis that every totally
ordered subset has an upper bound.
We do not prove here that the axiom of choice implies the well-ordering theorem and
Zorn’s Lemma but refer to [BouTS] III §2.3, Theorem 1 and [BouTS] III §2.4, Proposi-
tion 4.
13.4 Limits and Colimits
In this section I always denotes a small category.
Definition 13.29. Let C be a category.
1. An I -diagram in C is a functor XW I ! C . A morphism of I -diagrams is a morphism

of functors. We obtain the category of I -diagrams in C , denoted by C I . We often write
Xi instead of X.i/ for an object i in I .
2. A colimit or inductive limit of an I -diagram XW I ! C is an object
colim X D colim Xi D lim Xi

I i 2I !
i 2I
in C together with morphisms si W Xi ! colimI X in C for all objects i in I such

that:
(a) for every morphism 'W i ! j in I one has si D sj ı X.'/,
(b) for every object Z in C and for all morphisms ti W Xi ! Z such that for all mor-
phisms 'W i ! j in I one has ti D tj ı X.'/ there exists a unique morphism
qW colimI X ! Z such that ti D t ı si .
3. A limit or projective limit of a diagram XW I ! C is an object
lim X D lim Xi D lim Xi

I i 2I
i 2I
in C together with morphisms pi W limI X ! Xi in C for all objects i 2 I such that:

(a) for every morphism 'W j ! i in I , one has pi D X.'/ ı pj ,
(b) for every object Z in C and for all morphism qi W Z ! Xi such that qi D X.'/ıqj
for all morphism 'W j ! i in I , there exists a unique morphism qW Z ! limI X
such that qi D pi ı q for all i 2 Ob.I /.
Remark 13.30. Limits and colimits are (if they exist) unique up to unique isomorphism
by the uniqueness requirement in the definition.
Remark 13.31. Suppose that in a category C all colimits of I -diagrams exist. Every
morphism of functors XW I ! C to Y W I ! C induces a morphism colimI X ! colimI Y
and we obtain a functor
colimW C I ! C :
I
I
Similarly we have a functor limI W C ! C if in C all limits of I -diagrams exist.
13.4 Limits and Colimits 281
Remark 13.32. Every preordered set I can be made into a category, again denoted by
I . The objects are the elements of I and for two elements i; j 2 I the set of morphisms
HomI .i; j / consists of one element if i j and is empty otherwise. There is a unique
way to define a composition law making I into a category.
An I -diagram in a category C is then also called an inductive system in C indexed
by I . It is a family ..Xi /i 2I ; .'j i /i j /, where Xi is an object in C for all i 2 I and
'j i W Xi ! Xj is a morphism in C for all i; j 2 I with i j such that 'i i D idXi for all i
and 'kj ı 'j i D 'ki for all i j k.
An I opp -diagram in C is also called a projective system in C indexed by I . Of course,
one can switch between the notions of inductive and projective systems by taking the op-
posite order. Usually one considers colimits of inductive systems and limits of projective
systems.
In a given category C only some limits or colimits may exist. A category in which
arbitrary limits (respectively colimits) exist is called complete (respectively cocomplete).
A category in which limits (respectively colimits) of all I -diagrams exist for arbitrary finite
categories I is called finitely complete (respectively finitely cocomplete). The category of
sets is complete and cocomplete as the following example shows.
Example 13.33. Let I be a small category, I WD Ob.I /, and let XW I ! (Sets) be an

I -diagram in the category of sets.
1. The limit limI X exists in (Sets) and can be described by

( )
Y
lim X D .xi /i 2I 2 Xi I 8 'W i ! j in I : X.'/.xi / D xj : (13.1)
I
i 2I
For j 2 I the map pj W limI X ! Xj is given by the projection .xi /i 2I 7! xj .

2. The colimit colimI X exists in (Sets) and can be described by
!
a
colim X D Xi = ; (13.2)
I
i 2I
`
where i 2I Xi is the disjoint union of the sets Xi and where is the equivalence
relation generated by the relation xi xj if xi 2 Xi , xj 2 Xj and X.'/.xi / D xj
for some 'W i ! j . For j 2 I the map sj W Xj ! colimI X is given by attaching to
`
xj 2 Xj the equivalence class of xj 2 i 2I Xi .
The definition of the equivalence relation in the description of the colimit as an

equivalence relation generated by a relation makes it somewhat cumbersome to work with
. We will use the colimit usually in the case that the index category I is filtered in the
sense of Definition 13.37. In that case the description of simplifies (see Example 13.38).
Remark 13.34. Let XW I ! C , i 7! Xi be a diagram in a category C . Then the universal

property of lim Xi and colim Xi can also by definition be described as follows.
An object limI X in C together with morphisms pi W limI X ! Xi for all objects i of I
is a limit of X in C if and only if for all objects Y in C the map
u7!.pi ıu/i
HomC .Y; lim Xi / ! lim HomC .Y; Xi /
I I
is bijective, where the right-hand side denotes the limit in the category of sets.
Similarly, an object colimI X in C together with morphisms si W Xi ! colimI X for all
objects i of I is a colimit of X in C if and only if for all objects Y in C the map
u7!.uısi /i
HomC .colim Xi ; Y / ! lim HomC .Xi ; Y /
I I
is bijective.
The category of topological spaces is also complete and cocomplete.
Example 13.35. If XW I ! (Top) is a diagram in the category of topological spaces, then

limI X exists in (Top). The underlying set of limI X is the limit of sets described in (13.1)
and its topology is the subspace topology of the product topology.
Similarly colimI X exists in (Top). Its underlying set is given by (13.2) and its topology
`
is the quotient topology of the natural topology on i Xi .
One can also show that the category of groups is complete and cocomplete (Prob-
lem 13.11 and Problem 13.21). In Sect. 14.1 we will see that the category of left R-
modules (R a fixed ring) is complete and cocomplete.
Remark 13.36. Let I and J be (small) categories and let C be a category such that limits
(respectively colimits) of all I -diagrams and all J -diagrams in C exist. Let XW I J ! C ,
.i; j / 7! Xij be a diagram in C .
Then because of the definition of limits (respectively colimits) via a universal property
one obtains that limI J X (respectively colimI J X) exists and one has isomorphisms
lim Xij Š lim lim Xij Š lim lim Xij

i;j i j j i

resp. colim Xij Š colim colim Xij Š colim colim Xij :
i;j i j j i
Definition 13.37. A category I is called filtered if Ob.I / is non-empty and if the follow-
ing two conditions are satisfied:
(a) For all objects i and j in I there exists an object k and morphisms i ! k and j ! k.
(b) For all objects i and j and all morphisms f; gW i ! j there exists a morphism hW j !
k such that h ı f D h ı g.
For instance, a partially ordered set I is filtered if and only if the attached category is
a filtered category.
Example 13.38. Let I be a filtered category and let XW I ! (Sets) be an I -diagram in the
`
category of sets. In this case one has colimI X D . i 2I Xi /= where for xi 2 Xi and
xj 2 Xj one defines xi xj if there exist morphisms 'W i ! k and W j ! k such that
X.'/.xi / D X. /.xj / (the properties of a filtered category imply that is an equivalence
relation).
Filtered colimits in the category of sets commute with finite limits ([KS2] Theo-
rem 3.1.6).
Proposition 13.39. Let I be a filtered category, let J be a finite category and let XW I
J ! (Sets) be a diagram in the category of sets. Then the universal properties of limit
and colimit yield an isomorphism in the category of sets
colim lim X ! lim colim X:

I J J I
There is also a converse to this proposition (Problem 13.18).

There are several important special cases of limits and colimits.
Definition 13.40 (Products and coproducts). Let I be a set with the discrete order (i.e.,
i j , i D j ). Consider I as a category (equivalent to I opp ) and let XW I ! C be
a diagram in a category C .
Q
1. Then i 2I Xi WD limI X is called the product of X (if it exists). It is equipped with
Q
morphisms pj W i 2I Xi ! Xj for all j 2 I , called projections. It is characterized by
the following universal property: For every object Z in C and every family .fi W Z !
Q
Xi /i 2I of morphisms there exists a unique morphism f W Z ! i Xi such that fi D
pi ı f .
`
2. Dually, i 2I Xi WD colimI X is called the coproduct of X or the direct sum of X (if it
exists).
For I D f1; : : : ; ng we also write X1 Xn for the product and X1 q q Xn for

the coproduct.
Example 13.41 (Final and initial objects). Take I D ; in Definition 13.40. Then there
is a unique I -diagram in every category C . Its limit (respectively its colimit) is an object
P in C such that for every object X in C there exists a unique morphism X ! P (respec-
tively P ! X). The object P is called a final object (respectively an initial object). It is
unique up to unique isomorphism (if it exists).
Example 13.42. In the category of sets (respectively the category of topological spaces)
products and coproducts have the usual description:
1. The product of a family .Xi /i 2I of sets (respectively topological spaces) is the usual
cartesian product X of sets (respectively of topological spaces) together with its pro-
jections pi W X ! Xi . For any set (respectively topological space) Z and any family of
maps (respectively continuous maps) .fi W Z ! Xi /i 2I there is a unique map (respec-
tively a unique continuous map) f W Z ! X such that pi ı f D fi , namely the map
(respectively the continuous map) z 7! .fi .z//i 2I . The final object is the singleton,
i.e., the set (respectively topological space) consisting of a single element.
2. The coproduct of the family .Xi /i 2I is the disjoint union of the sets Xi (respectively
the sum of the topological spaces Xi in the sense of Definition 12.26).
The initial object is ;.
Remark and Definition 13.43 (Fiber products and pushouts). Let I be the category
with three objects j , i1 , and i2 and whose only morphisms except the identities are two
morphisms i1 ! j and i2 ! j . We represent I schematically by
i1 j i2
Then an I -diagram X in a category C is a diagram of morphisms in C of the form
f1 f2
X1 Y X2 :
The limit of X, if it exists, is called the fiber product of X1 and X2 over Y . It consists of
an object denoted by X1 Y X2 and morphisms pi W X1 Y X2 ! Xi for i D 1; 2, called
projections. It is characterized by the following universal property: For every object Z in
C and all morphisms g1 W Z ! X1 and g2 W Z ! X2 such that f1 ı g1 D f2 ı g2 there exists
a unique morphism gW Z ! X1 Y X2 such that pi ı g D gi for i D 1; 2. We indicate
this universal property by the following diagram:
Z
8g2
9Šg
p1
X1 Y X2 X1 (13.3)
8g1
p2 f1
X2 Y:
f2
The morphism g is also denoted by .g1 ; g2 /Y .

Dually, there is the notion of a pushout in a category C , which is the colimit of an
inductive system J ! C , where J is represented schematically by
The pushout of a J -diagram X1 Y ! X2 in C is denoted by X1 qY X2 .
Example 13.44. Let f W X ! Y be a morphism in a category and assume that the fiber
product of
f f
X Y X:
exists. Then f WD .idX ; idX /Y W X ! X Y X is called the diagonal of f .
Remark and Definition 13.45. Let C be a category and f W X ! Y a morphism in C .

Then f is a monomorphism if and only if the fiber product X Y X exists and the diagonal
f W X ! X Y X is an isomorphism.
One has a similar characterization of epimorphisms via pushouts.
Example 13.46. Consider the category of topological spaces. Then fiber products always
exist in this category and are given by fiber products in the sense of Definition 12.28. The
diagonal of a continuous map f W X ! Y is the diagonal f W X ! X Y X, x 7! .x; x/.
Let F W C ! D be a functor between categories. Let XW I ! C be a diagram in C such

that the limits limI X and limI .F ı X/ exist in C and D, respectively. For every object i
in I the morphism limI X ! Xi induces by application of F a morphism F .limI X/ !
F .Xi /. The family of these morphisms corresponds by the universal property of limI .F ı
X/ to a morphism
F .lim X/ ! lim.F ı X/: (13.4)
I I
We say that F commutes with limits if for every diagram XW I ! C such that its limit
limI X exists in C , the limit of F ı X exists in D and the morphism (13.4) is an isomor-
phism.
Dually, there is the notion a functor that commutes with colimits.
Proposition 13.47. Let F W C ! D be a functor.
1. Suppose that F is right adjoint to some functor GW D ! C . Then F commutes with

limits.
2. Dually, suppose that F is left adjoint to some functor. Then F commutes with colimits.
Proof. [KS2] Proposition 2.1.10.

Definition 13.48. Let C be a category that is finitely cocomplete, i.e., for all diagrams
XW I ! C , where I is a finite category, the colimit colimI X exists in C . Then a functor
F W C ! C 0 is called right exact if it commutes with finite colimits.
Similarly we define for a finitely complete category C a functor F W C ! C 0 to be left
exact if it commutes with finite limits.
Remark 13.49. Let C be a finitely complete category and let F W C ! D be a left ex-
act functor. As finite products, final objects, and fiber products are all special cases of
finite limits, F commutes with all these constructions. Moreover, the characterization
of monomorphisms in Remark 13.45 implies that F also maps monomorphisms in C to
monomorphisms in D.
Example 13.50. Let I be a filtered category. Then Proposition 13.39 shows that the
category of (Sets)I of I -diagrams of sets is finitely complete and that colimI W (Sets)I !
(Sets) (Remark 13.31) is left exact.
In particular colimI sends monomorphisms in (Sets)I to monomorphism in (Sets) (i.e.,
to an injective map).
13.5 Problems
Problem 13.1. Let .I; / and .J; / be ordered sets. A map f W I ! J is called increas-
ing if i i 0 implies f .i/ f .i 0 / for all i; i 0 2 I . Show that ordered sets (as objects) and
increasing functions (as morphisms) define a category.
Problem 13.2. Let C be a category and idC W C ! C the identity functor. Show that the
monoid EndFunc.C ;C / .idC / is commutative.
Problem 13.3. Let .I; / be an ordered set. Let Tr be the set of subsets U I satisfying
the following condition: For all i 2 U and j 2 I with i j one has j 2 U .
1. Show that Tr is a topology on I , the so-called right order topology on I .

2. Show that a map f W I ! J between partially ordered sets .I; / and .J; / is in-
creasing if and only if it is continuous for the right order topologies on I and J . In
particular we obtain a fully faithful functor T from the category of ordered sets (where
morphisms are increasing maps) to the category of topological spaces.
3. Show that T yields an equivalence of the category of ordered sets with the category of
topological spaces X that are T0 -spaces (Problem 12.7) such that the intersection of an
arbitrary family of open subsets is again open (considered as a full subcategory of the
category of topological spaces).
Similarly, one can also define the left order topology on I in which the open subsets U
are those such that for all i 2 U and j 2 I with i j one has j 2 U .
13.5 Problems 287
Problem 13.4. Show that the construction in Problem 12.8 yields a functor from the
category of topological spaces to the full subcategory of T0 -spaces that is left adjoint to
the inclusion functor. Deduce that any limit of T0 -spaces in the category of topological
spaces is again a T0 -space.
Problem 13.5. Let C be a (small) category and let idW C ! C be the identity functor.
Show that if colimC id exists in C , then it is a final object of C .
Problem 13.6. Let G be a group. Convince yourself that one can consider G as a category
with a single object whose set of endomorphisms is G and whose composition law is given
by multiplication in G.
1. Show that a G-diagram in the category of sets is the same as a G-set X (i.e., X is a set
endowed with a left G-action). Show that a morphisms of G-diagrams is the same as
a G-equivariant map between G-sets.
2. Let X be a G-set considered as a G-diagram. Show that limG X D X G WD
f x 2 X I gx D x 8 g 2 G g and that colimG X D GnX (the set of G-orbits).
Problem 13.7. A (small) category in which every morphism is an isomorphism is called

a groupoid. A category C is called connected if it is non-empty and if for all pairs of
objects X and Y in C there exists a finite sequence X D X0 ; X1 ; : : : ; Xn D Y of objects
Xi in C such that for all i D 1; : : : ; n one of the sets HomC .Xi 1 ; Xi / or HomC .Xi ; Xi 1 /
is non-empty.
Now let C be a non-empty category. Show that C is a connected groupoid if and only
if C is equivalent to the category defined by a group (Problem 13.6).
Problem 13.8. Let I and C be categories. Let X be an object in C and let "X W I ! C be
the constant functor with value X (i.e., "X .i/ D X for every object i in I and "X .'/ D idX
for every morphism ' in I ). Suppose that I is connected (Problem 13.7). Show that
X ! limI "X and colimI "X ! X.
Problem 13.9. Let I be the category with two objects 0 and 1 whose only morphisms
other than the identities are two morphisms 0 ! 1. Hence an I -diagram X in a category
C is simply a pair of two morphisms f; gW X0 ! X1 in C . We call Ker.f; g/ WD limI X
(respectively Coker.f; g/ WD colimI X) the equalizer (respectively coequalizer) of f and
g (if it exists).
1. Let g; hW X0 ! X1 be a pair of morphisms in C such that Ker.f; g/ (respectively

Coker.f; g/) exists. Let iW Ker.f; g/ ! X0 (respectively pW X1 ! Coker.f; g/) be
the natural morphism. Show that i is a monomorphism (respectively that p is an
epimorphism).
2. Let C be the category of sets and let f; gW X0 ! X1 be maps of sets. Show that
Ker.f; g/ D f x 2 X0 I f .x/ D g.x/ g. Describe Coker.f; g/.
Problem 13.10. Let C be a category.
1. Show that C is complete if and only if all products and all equalizers (Problem 13.9)
exist in C .
(i) C is finitely complete.
(ii) C admits all finite products and all equalizers.
(iii) C admits all fiber products and has a final object.
Dualize the above statements to obtain criteria for C to be (finitely) cocomplete.
Problem 13.11. Let C be either the category of all groups, of all rings, or of all left R-
modules (R a fixed ring). Show that all these categories are complete and that the forgetful
functor from C to the category of sets commutes with limits.
Problem 13.12.
1. Show that in the category (Top) of topological spaces a continuous map f W X !

Y is a monomorphism (respectively an epimorphism) if and only if f is injective
(respectively surjective).
2. Let C be the category of all Hausdorff spaces (considered as a full subcategory of
(Top)). Show that again the monomorphisms in C are the injective continuous maps
but that every continuous map with dense image is an epimorphism. Are there other
epimorphisms?
Problem 13.13. Let I be a category and let XW I ! (Top) be an I -diagram in the category
of topological spaces. Show that if Xi is Hausdorff for all i in I , then limI X is Hausdorff.
Problem 13.14. Let I be a small category and let i 7! Xi be an I -diagram of topological

spaces.
1. Suppose that Xi is Hausdorff and compact for all objects i of I . Show that limi Xi
(limit in the category of topological spaces) is Hausdorff and compact.
2. Let I D N opp be the category attached to the set N endowed with the opposite of the
usual order. For i 2 N let Xi D N be endowed with the unique topology such that the
sets f1g; : : : ; fig are open and closed in Xi and such that the only open subsets of N>i
are ; and N>i itself. Let the transition maps be the identity of N.
Show that Xi is compact but that limi Xi D N with the discrete topology. In particular
limi Xi is not compact.
13.5 Problems 289
Problem 13.15. Let G be an abelian group, let F .G/ be the set of finitely generated
subgroups of G and let C .G/ be the set of cyclic subgroups of G. Endow both sets with
the partial order given by and consider them as categories. Let W F .G/ ! (Grp) and
W C .G/ ! (Grp) be the inclusion functors.
1. Show that the universal property of the colimit yields maps
f W colim ! G; gW colim ! G:
F .G/ C .G/
2. Show that F .G/ is filtered and that f is an isomorphism.

3. Show that g is always surjective and that it is an isomorphism if the inclusion order on
C .G/ is filtered.
4. Show that g is not an isomorphism for G D Z=2Z Z=2Z.
Problem 13.16. Let I be a category. Show that I is filtered if and only if for every
finite category J and any functor F W J ! I there exists an object i of I such that
limj 2J HomI .F .j /; i/ ¤ ;.
Problem 13.17. Show that the categories (Sets) and (Sets)opp are not equivalent.
Problem 13.18. Let I be a category such that for every finite category J and every dia-
gram XW I J ! (Sets) in the category of sets the map
colim lim X ! lim colim X:

I J J I
is bijective. Show that I is filtered.
Problem 13.19. Describe the gluing of topological spaces as a colimit.
Problem 13.20. Let R be a commutative ring. Show that each of the following forgetful
functors has a left adjoint functor:
1. The functor (Top) ! (Sets) sending a topological space to its underlying set.
2. The functor from the category (R-Alg) of commutative R-algebras to (Sets) sending
an R-algebra to its underlying set.
3. The functor from (R-Alg) to the category of commutative monoids sending an R-
algebra to .R; /.
4. The functor (R-Alg) ! (Ab) sending an R-algebra to its underlying additive group
.R; C/.
Problem 13.21. The category of groups is cocomplete:
1. Show that the forgetful functor from the category of groups to the category of sets has
a left adjoint F W (Sets) ! (Grp). Show that the canonical map of sets S ! F .S/ is
injective and its image generates the group F .S/. For a set S the group F .S/ is called
the free group generated by S.
2. Show that every group G is isomorphic to a quotient of the form F .X/=N for some
set X and some normal subgroup N of F .X/. If R F .X/ is a subset such that N
is the smallest normal subgroup containing R, then one says that G is generated by X
with relators R and we write G D hXI Ri.
3. Let .Gi /i 2I be a family of groups. Show that their coproduct exists in the category of
groups. It is often called the free product of the Gi .
` `
Hint: If Gi D hXi I Ri i, show that h i Xi I i Ri i is a coproduct of the Gi .
4. Show that the coequalizer of every pair of group homomorphisms f; gW G0 ! G1
exists in the category of groups. Deduce that the category of groups is cocomplete
(use Problem 13.10).
Problem 13.22. Let .Mon/ be the category of monoids and monoid homomorphisms.
1. For M a monoid let M WD f m 2 M I 9 n 2 M W mn D nm D e g be the set of

invertible elements. Show that M is a group and that every homomorphism of
monoids M ! N induces a group homomorphism M ! N . We obtain a functor
. / W .Mon/ ! (Grp). Show that this functor is right adjoint to the inclusion functor
(Grp) ! .Mon/.
2. For M a commutative monoid (written additively) let M gp be its Grothendieck group,
i.e., M gp D .M M /= with .m; m0 / .n; n0 / W, 9 s 2 M W sCmCn0 D sCm0 Cn.
Show that is indeed an equivalence relation, that the componentwise addition on
M M induces the structure of an abelian group on M gp , and that M ! M gp ,
m 7! Œm; 0 is a homomorphism of monoids, called canonical. Show that every ho-
momorphism of commutative monoids M ! N induces a homomorphism of abelian
groups M gp ! N gp . We obtain a functor . /gp from the category of commutative
monoids to the category of abelian groups. Show that this functor is left adjoint to the
inclusion functor.
Appendix C: Basic Algebra
14
14.1 The Category of Modules over a Ring
Modules
Let R be a ring (with unit, but not necessarily commutative). We recall very briefly some
notions about R-modules: A left R-module is a “vector space over R”, i.e., it is an abelian
group .M; C/ together with a scalar multiplication
R M ! M; .a; m/ 7! am
such that for all a; b 2 R, m; m0 2 M one has
.aCb/m D amCbm; a.mCm0/ D amCam0; .ab/m D a.bm/; 1m D m:
In the sequel we will usually speak simply of R-modules. For R-modules M and N , a
homomorphism of R-modules or an R-linear map from M to N is a map uW M ! N such
that for all a 2 R, m; m0 2 M one has
u.am C m0 / D au.m/ C u.m0 /:
We obtain the categories of R-modules denoted by (R-Mod), where composition is de-

fined as the usual composition of maps. There is a unique way to define a scalar multipli-
cation on the trivial group f0g. This R-module is called the zero module and it is denoted
by 0. Every bijective R-linear map is an isomorphism because the inverse of a bijective
R-linear map is automatically R-linear again.
The set of homomorphisms of R-modules M ! N is denoted by HomR .M; N /. For
u; w 2 HomR .M; N / we define uCw 2 HomR .M; N / by .uCw/.m/ WD u.m/Cw.m/.
This defines the structure of an abelian group on HomR .M; N /. Let
C WD Cent.R/ WD f a 2 R I ab D ba for all b 2 Rg :

DOI 10.1007/978-3-658-10633-1_14
292 14 Appendix C: Basic Algebra
be the center of the ring R. This is a commutative subring of R. For a 2 C the map
auW M ! N , au.m/ WD a.u.m// is again R-linear and we obtain the structure of a
C -module on the abelian group HomR .M; N /.
Example 14.1.
1. If R D k is a field, an R-module is simply a k-vector space.

2. If A is an abelian group, it is a Z-module by
na WD „
a C ƒ‚
C…
a for n 0; na WD .n/a for n < 0:
n times
Conversely, forgetting the scalar multiplication makes every Z-module into an abelian
group. Therefore we see that it is the same to give an abelian group or to give a Z-
module.
Let M be an R-module. A subset N of M is called an R-submodule if N is a subgroup

of .M; C/ and if an 2 N for all a 2 R, n 2 N . Then addition and scalar multiplication
of M induce on N the structure of an R-module.
T
If .Ni /i 2I is a family of R-submodules of M , then i Ni is an R-submodule. In
particular there exists for every subset S a smallest R-submodule of M , containing S. It
is called the R-submodule generated by S and it is denoted by hSi or hSiR .
A subset S M such that hSi D M is called a generating set of M . The R-module
M is called a finitely generated if there exists a finite generating set of M .
Let N be an R-submodule of M . Then the scalar multiplication of M induces a scalar
multiplication of the quotient group M=N making M=N into an R-module, called the
quotient module of M by N .
Categorical Constructions of Modules

Let uW M ! N be a homomorphism of R-modules. Let N 0 N be an R-submodule.
Then u1 .N 0 / is an R-submodule of M . In particular
Ker.u/ WD f m 2 M I u.m/ D 0 g
is an R-submodule of M , called the kernel of u. As for groups one sees that u is injective
if and only if Ker.u/ D 0.
Let M 0 M be an R-submodule. Then u.M 0 / is an R-submodule of N . In particular
Im.u/ WD u.M / is an R-submodule of N , called the image of u.
We call Coker.u/ WD N= Im.u/ and Coim.u/ WD M= Ker.u/ the cokernel and the
coimage of u, respectively. Then u is surjective if and only Coker.u/ D 0.
As for groups one sees that every R-linear map uW M ! N induces an isomorphism
of R-modules
N Coim.u/ ! Im.u/:
uW
14.1 The Category of Modules over a Ring 293
We now construct limits and colimits in the category of R-modules. In particular we

will see that the category of R-modules is complete and cocomplete.
Q
Let .Mi /i 2I be a family of modules. The cartesian product i Mi with componen-
twise addition and scalar multiplication is an R-module. Together with the projections
Q
prj W i Mi ! Mj it is a product in the category of R-modules in the sense of Defini-
tion 13.40. In other words, for every R-module N the map
!
Y u7!.pri ıu/i Y
HomR N; Mi ! Hom(R-Mod) .N; Mi / (14.1)
i 2I i 2I
is bijective. Note that (14.1) is C -linear, where C is the center of R.

More generally, let I be a small category and let M W I ! (R-Mod), i 7! Mi , be an
I -diagram of R-modules. Then
( )
Y
lim M WD .mi /i 2I 2 Mi I 8 'W i ! j in I : M.'/.mi / D mj
I
i 2I
Q
is an R-submodule of i 2I Mi and the projections limI M ! Mi yield a bijection

HomR N; lim M ! lim HomR .N; Mi /; (14.2)
I I
which shows that limI M is indeed the limit of the I -diagram M (Remark 13.34). Again,
(14.2) is a C -linear map.
To construct colimits in the category of R-modules, we proceed as in the category of
sets (Remark 13.33): We define them as suitable quotients of coproducts. Hence let us
construct coproducts (called direct sums for modules) for a family of modules .Mi /i 2I
L Q
first. The submodule i 2I Mi of i Mi consisting of tuples .mi /i with mi D 0 for
all but finitely many i 2 I is called the direct sum of .Mi /i . Together with the maps
L
sj W Mj ! i 2I Mi , mj 7! .mi /i , where mi D 0 for all i ¤ j , it is a coproduct in
the category of R-modules: For every R-module N and every family .ui W Mi ! N /i of
L
R-linear maps there exists a unique R-linear map uW i 2I Mi ! N such that u ı si D ui
for all i 2 I . In other words, the map
!
M Y
Hom(R-Mod) Mi ; N ! Hom(R-Mod) .Mi ; N /;
i 2I i 2I (14.3)
u 7! .u ı si /i
P
is bijective. An inverse map is given by .ui /i 7! ..mi /i 7! i 2I ui .mi / (a finite sum
because all but finitely many mi are zero).
Now let M W I ! (R-Mod), i 7! Mi , be an I -diagram of R-modules. For all objects j
L L
of I we consider Mj as a submodule of i 2Ob.I / Mi via sj W Mj ! i 2Ob.I / Mi . Let N
L
be the submodule of i 2Ob.I / Mi generated by all mi mj with mi 2 Mi , mj 2 Mj and
M.'/.mi / D mj for some morphism 'W i ! j in I . Let
0 1
M
colim M WD @ Mi A =N
I
i 2Ob.I /
and let ti W Mi ! colimI M be the composition of si with the projection. Then composition
with all ti yields for all R-modules a bijective map

HomR colim M; N ! lim HomR .Mi ; N / (14.4)
I I
and hence colimI M is indeed the colimit in the category of R-modules. Again, (14.4) is
C -linear, where C is the center of R.
L Q
If I is a finite set, then i 2I Mi D i 2I Mi . For I D ; we obtain the zero module,
which is an initial and a final object in the category of R-modules.
Q
If Mi D M for all i, then we write M I (respectively M .I / ) instead of i 2I Mi (re-
L
spectively i 2I Mi ). For every integer n 0 we set M n WD M f1;:::;ng D M .f1;:::;ng/ . In
particular we have for M D R the R-modules RI , R.I / and Rn . We also usually write
ej WD sj .1/ 2 R.I / .
Example 14.2. Consider the following diagram of R-modules:
u
M N:
0
Then the limit of this diagram consists of pairs .m; n/ 2 M N such that u.m/ D n and
n D 0. Hence it is identified with Ker.u/.
The colimit of this diagram is the quotient of M ˚ N by the submodule E generated
of all elements of the form .m; 0/ .0; n/ with u.m/ D n and of all elements of the form
.m; 0/. Then E is the kernel of the projection M ˚ N ! N= Im.u/. Hence the colimit is
identified with Coker.u/.
Note that the forgetful functor (R-Mod) ! (Sets) commutes with limits. This we can
also deduce from Proposition 13.47 because the forgetful functor has a right adjoint as
we see in the next subsection. The forgetful functor does not commute with arbitrary
colimits as the example of coproducts shows. It does commute with filtered colimits
(Problem 14.3).
Free Modules and Matrices

As with vector spaces (Example 13.19) one sees that the functor I 7! R.I / is left adjoint
to the forgetful functor (R-Mod) ! (Sets). In other words, we have for every R-module
14.1 The Category of Modules over a Ring 295
a bijection, functorial in I and in M ,
Hom(R-Mod) .R.I / ; M / ! Hom(Sets) .I; M / D M I ;

(14.5)
u 7! .u.ei //i 2I ;
where ei 2 R.I / is the tuple whose j -th entry is ıij for all j 2 I . Hence to give a family of
elements m D .mi /i with mi 2 M is equivalent to giving an R-linear map um W R.I / ! M .
One can also view (14.5) as a special case of (14.3).
We call m linearly independent (respectively a generating system, respectively a basis)
if um is injective (respectively surjective, respectively an isomorphism).
An R-module is called free if it admits a basis, i.e., if it is isomorphic to an R-module
of the form R.I / for some set I .
In contrast to vector spaces, R-modules are usually not free (Problem 14.6).
Remark 14.3. Let .mi /i 2I be a generating system (respectively a linearly independent

system) of an R-module M and let uW M ! N be a surjective (respectively injective)
homomorphism of A-modules. Then .u.mi //i 2I is a generating system (respectively a lin-
early independent system) of N .
As for linear maps between vector spaces one can express linear maps between free
modules by matrices after choosing a basis. For integers m; n 0 we denote by Mnm .R/
the set of matrices with n rows and m columns and with coefficients in R. We also set
Mn .R/ WD Mnn .R/. The usual addition and scalar multiplication of matrices makes
Mnm .R/ into an R-module. It is a free R-module: A basis is given by the matrices Eij
with all entries equal to 0 except the .i; j /-th entry which is equal to 1.
One has the usual matrix multiplication (p 0 some integer)
Mpn .R/ Mnm .R/ ! Mpm .R/;

0 1
X
n
..aij /i;j ; .bj k /j;k / 7! @ aij bj k A :
j D1
i;k
Note that here the order of multiplication of summands usually matters if R is not com-
mutative.
Let M and N be R-modules such that there exists a basis .e1 ; : : : ; em / of M and a basis
.f1 ; : : : ; fn / of N . Let uW M ! N be an R-linear map. Then for every j D 1; : : : ; m
there exist unique aij , i D 1; : : : ; m such that
X
n
u.ej / D aij fi
i D1
and .aij /i;j 2 Mnm .R/ is the matrix of u with respect to the bases .ei / and .fj /. As
for vector spaces, composition of linear maps is expressed by multiplication of the corre-
sponding matrices.
Exact Sequences
A sequence of homomorphisms of R-modules
ui 1 ui
: : : ! Mi 1 ! Mi ! Mi C1 ! : : :
is called a complex (respectively is called an exact) if ui ı ui 1 D 0 (respectively if

Im.ui 1 / D Ker.ui /) for all i. A short exact sequence is an exact sequence of the form
0 ! M 0 ! M ! M 00 ! 0:
Remark 14.4. Let R be a ring and let Mi , i D 1; 2; : : : be R-modules.

f
1. The sequence 0 ! M1 ! M2 is exact if and only if f is injective. The sequence
f
M1 ! M2 ! 0 is exact if and only if f is surjective.
f1 f2 f3
2. Let M1 ! M2 ! M3 ! M4 be an exact sequence of R-modules. Then f2
induces an isomorphism of R-modules Coker.f1 / ! Ker.f3 /. In particular, f1 is
surjective if and only if f3 is injective.
i p
Definition and Remark 14.5. A short exact sequence 0 ! M 0 ! M ! M 00 ! 0 of
R-modules is called split if the following equivalent conditions are satisfied:
(i) There exists a section sW M 00 ! M of p (i.e., a homomorphism of R-modules

sW M 00 ! M such that p ı s D idM 00 ).
(ii) There exists an R-submodule MQ 00 of M such that p j MQ 00 W MQ 00 ! M 00 is an isomor-
phism.
(iii) There exists a retraction rW M ! M 0 of i (i.e., a homomorphism of R-modules
rW M ! M 0 such that r ı i D idM 0 ).
In this case one has isomorphisms of R-modules
M 0 ˚ M 00 Š M;
.m0 ; m00 / 7! i.m0 / C s.m00 /;
.r.m/; p.m// 7 m:
i p
Example 14.6. Every short exact sequence 0 ! M 0 ! M ! M 00 ! 0 of R-modules
with M 00 a free R-module is split (more generally, this holds whenever M 00 is a projective
R-module, see Problem 14.4). Let .m00i /i 2I be a basis of M 00 . As p is surjective, there
exist mi 2 M with p.mi / D m00i for all i. Let sW M 00 ! M be the unique R-linear map
with s.m00i / D mi for all i. Then s is a section of p.
14.2 Multilinear Algebra and Tensor Products 297
14.2 Multilinear Algebra and Tensor Products
In this section we always denote by A a commutative ring and by M an A-module.
Modules of Multilinear Maps

Let r 2 N0 and let .Ni /1i r be a finite family of A-modules. A map
˛W N1 Nr ! M
is called r-multilinear if it is linear in every component. The sum of two r-multilinear

maps and the product of an r-multilinear map with a scalar in A is again an r-multilinear
map (for the second assertion we need that A is commutative). Hence the set of all such
r-multilinear maps obtains a structure of an A-module, denoted by LA .N1 ; : : : ; Nr I M /.
For r D 1, we obtain the A-module HomA .N; M / WD LA .N I M / consisting of all
A-linear maps N ! M .
A 0-multilinear map is an arbitrary map "W f0g ! M . Identifying " with ".0/ gives an
identification of 0-linear M -valued maps and elements in M .
Finally, 2-multilinear maps are also called bilinear.
Example 14.7. For every A-module the map HomA .A; M / ! M , u 7! u.1/ is an
isomorphism of A-modules, which we will often use to identify HomA .A; M / and M .
The formation of HomA .M; N / is functorial in M and in N . Let uW M ! M 0 be an

A-linear map. Then
HomA .u; N /W HomA .M 0 ; N / ! HomA .M; N /; v 7! v ı u
is A-linear. We obtain a contravariant functor HomA .; N / from the category of A-modules
to itself.
Similarly, one has a (covariant) functor HomA .M; /W (A-Mod) ! (A-Mod).
Remark 14.8. Let .Mi /i 2I be a family of A-modules and let N be an A-module. The
bijection (14.3) is an isomorphism of A-modules. As multilinear maps are defined as
componentwise linear, it follows that for all integers 1 s r and for all A-modules
P1 ; : : : ; Ps1 ; PsC1 ; : : : ; Pr there is an isomorphism of A-modules, functorial if Mi , Pj
and N , !
M
LA P1 ; : : : ; Ps1 ; Mi ; PsC1 ; : : : ; Pr I N
i 2I (14.6)
Y
! LA .P1 ; : : : ; Ps1 ; Mi ; PsC1 ; : : : ; Pr I N /:
i 2I
Definition and Remark 14.9 (Dual module). For every A-module M the A-module
M _ WD HomA .M; A/ is called the dual of M . As usual there is a functorial map
M W M ! .M _ /_ ; m 7! . 7! .m//: (14.7)
We have A_ D A by Example 14.7. Hence for every set I we have
.A.I / /_ D AI
by (14.6).
In particular we see for that every A-module M with a basis of n 2 N0 elements its dual
_
M has also a basis with n elements. Moreover the biduality map M is an isomorphism
of A-modules.
Algebras
An A-algebra (with unit) is an A-module C together with a bilinear map C C ! C ,
.c; d / 7! c d , such that .C; C; / is a ring. Hence we always assume that the mul-
tiplication in an algebra is associative and has a unit. If the ring multiplication is also
commutative, C is called a commutative algebra. As usual we will often write cd instead
of c d .
Remark 14.10. An A-algebra C is the same as a ring C together with a homomor-

phism of rings W A ! C such that the image of is contained in the center Cent.C / WD
f c 2 C I cd D dc for all d 2 C g of C . Such a ring C becomes an A-module via the
scalar multiplication .a; c/ 7! .a/c for a 2 A, c 2 C . Conversely, given an A-algebra
structure on C , the ring homomorphism is defined by .a/ WD a1C for a 2 A.
As Z-modules are nothing but abelian groups, Z-algebras are nothing but rings.
A map 'W C ! C 0 between A-algebras C and C 0 is called a homomorphism of A-
algebras if
(a) ' is A-linear,

(b) '.cd / D '.c/'.d / for all c; d 2 C and '.1/ D 1.
We obtain the category of A-algebras, denoted by (A-Alg).

Our primary source of algebras will be algebras of R-valued or C-valued functions, for
instance:
Example 14.11. Let X be a topological space and denote by CX IK .X/ the set of con-
tinuous maps X ! K (we refer to Example 3.5 for an explanation for this cumbersome
notation). Then CX IK .X/ is a K-algebra with the usual addition of functions, scalar mul-
tiplication of a function with a real number, and multiplication of functions.
If F W X ! Y is a continuous map of topological spaces,
F W CY IK .Y / ! CX IK .X/; g 7! g ı F
is a homomorphism of K-algebras.
Example 14.12. Let M be an A-module. Then the composition of linear maps defines
on the A-module EndA .M / D HomA .M; M / the structure of an A-algebra. If M admits
a finite basis .e1 ; : : : ; en /, then attaching to a linear map its matrix with respect to .ei /i
yields an isomorphism of A-algebras EndA .M / ! Mn .A/, where the multiplication on
the A-module Mn .A/ is given by matrix multiplication.
Tensor Products
Let M and N be A-modules. Then there exists an A-module M ˝A N and an A-bilinear
map W M N ! M ˝A N , denoted by .m; n/ 7! m ˝ n, which is universal, i.e., for all
A-modules P and for all A-bilinear maps bW M N ! P there exists a unique A-linear
map bWN M ˝A N ! P such that b.m; n/ D b.m N ˝ n/ for all .m; n/ 2 M N . The pair
.M ˝A N; / is determined uniquely up to unique isomorphism by this definition. The
A-module M ˝A N is called the tensor product of M and N over A. We will also often
write M ˝ N if it is clear over which ring we form the tensor product.
The construction of M ˝A N is rather straightforward: For every A-module P , the set
HomA .A.M N / ; P / is the set of all maps ˛W M N ! P (14.5). Such a map ˛ is bilinear
if and only if
˛.am C m0 ; n/ .a˛.m; n/ C ˛.m0 ; n// D 0

and ˛.m; an C n0 / .a˛.m; n/ C ˛.m; n0 // D 0
for all a 2 A, m; m0 2 M , n; n0 2 N . Hence, if we denote by Q the A-submodule of

A.M N / generated by all elements of the form
e.amCm0 ;n/ .ae.m;n/ C e.m0 ;n/ / and e.m;anCn0 / .ae.m;n/ C e.m;n0 / /;
then an A-linear map uW A.M N / ! P corresponds to a bilinear map ˛ if and only if

u.Q/ D 0. Hence we can simply set M ˝A N WD A.M N / =Q and define m ˝ n as the
image of e.m;n/ in M ˝A N .
Remark 14.13. As the e.m;n/ for m 2 M and n 2 N generate A.M N / , their images
m ˝ n in M ˝A N form a generating system of M ˝A N (Remark 14.3) (but usually not
every element of M ˝A N is of this form).
In particular, every A-linear map uW M ˝A N ! P is already uniquely determined by
the elements u.m ˝ n/ for m 2 M and n 2 N . Conversely, a map given by specifying
elements u.m ˝ n/ for all m 2 M and n 2 N in an A-module P extends to a well-defined

A-linear map M ˝A N ! P if and only if
u.am ˝ n/ D u.m ˝ an/ D au.m ˝ n/;

u..m C m0 / ˝ n/ D u.m ˝ n/ C u.m0 ˝ n/; (14.8)
0 0
u.m ˝ .n C n // D u.m ˝ n/ C u.m ˝ n /
for all a 2 A, m; m0 2 M , and n; n0 2 N .
The maps m ˝ n 7! n ˝ m and n ˝ m 7! m ˝ n define mutually inverse isomorphisms,

functorial in M and N ,
M ˝A N Š N ˝A M; (14.9)
which we often use to identify these modules. Similarly, the map
M ˝ .N ˝ P / ! .M ˝ N / ˝ P; m ˝ .n ˝ p/ 7! .m ˝ n/ ˝ p (14.10)
defines an A-linear isomorphism, which we use to identify these modules. We simply

write M ˝A N ˝A P or M ˝ N ˝ P and m ˝ n ˝ p for m 2 M , n 2 N , p 2 P . We
use similar notations for the tensor product of finitely many modules.
Remark 14.14. For A-modules M1 ; : : : ; Mr the r-multilinear map
W M1 Mr ! M1 ˝ ˝ Mr ; .m1 ; : : : ; mr / 7! m1 ˝ ˝ mr
is universal, i.e., for every A-module N the A-linear map
HomA .M1 ˝ ˝ Mr ; N / ! LA .M1 ; : : : ; Mr I N /; u 7! u ı (14.11)
is bijective. In other words, for every r-multilinear map ˛W M1 Mr ! N there

exists a unique linear map uW M1 ˝ ˝ Mr ! N such that u ı D ˛. One simply
checks that
uW m1 ˝ ˝ mr 7! ˛.m1 ; : : : ; mr /
is well defined.
The tensor product is functorial. Let uW M ! M 0 and vW N ! N 0 be A-linear maps.

Then m ˝ n 7! u.m/ ˝ v.n/ defines a well-defined A-linear map
u ˝ vW M ˝A N ! M 0 ˝A N 0 :
We obtain a functor
./ ˝ N W (A-Mod) ! (A-Mod); M 7! M ˝A N; u 7! u ˝ idN : (14.12)
It is isomorphic to the analogously defined functor N ˝ ./ by (14.9).

Remark 14.15. Suppose that uW M ! M 0 and vW N ! N 0 are surjective. Then u ˝ v

is surjective because its image contains the generating set of all m ˝ n for m 2 M and
n 2 N.
Even if v D idN , it is in general not true that u ı idN is injective if u is injective

(Problem 14.8; see however Problem 14.9).
Remark 14.16. The maps A ˝A M ! M , a ˝ m 7! am, and M ! A ˝A M ,

m 7! 1 ˝ m yield mutually inverse isomorphisms
A ˝A M Š M; (14.13)
which we will usually use to identify these two modules.
Remark 14.17. As we carefully recorded the universal property of all our constructions
with modules, we can use this to show several useful compatibilities. The key idea is the
fact that two A-modules M and N are isomorphic if and only if for every A-module P
there is a bijection of sets
HomA .M; P / Š HomA .N; P /; functorial in P . (14.14)
Indeed the condition is clearly necessary. Conversely, choosing P D N in (14.14), idN

corresponds to a map uW M ! N and choosing P D M in (14.14), idM corresponds to
a map vW N ! M . The functoriality in P implies that u ı v D idN and v ı u D idM .
As an example of this principle we show that tensor products commute with direct
sums. Let .Mi /i be a family of A-modules and let N be an A-module. Then there is
a functorial isomorphism
!
M M
Mi ˝A N Š .Mi ˝A N /: (14.15)
i 2I i 2I
Indeed, let P be an A-module. Then there are isomorphisms of A-modules (in particular
bijective maps), functorial in P ,
! ! !
M M
HomA Mi ˝A N; P Š LA Mi ; N I P
(14.11)
i 2I i 2I
Y
Š LA .Mi ; N I P /
(14.6)
i 2I
Y (14.16)
Š HomA .Mi ˝A N; P /
(14.11)
i 2I
!
M
Š HomA .Mi ˝A N /; P :
(14.6)
i 2I
As all the isomorphisms in (14.16) are also functorial in the Mi and in N , the isomor-
phism (14.15) is functorial in the Mi and in N .
Remark 14.18. Let M and N be free A-modules with basis m D .mi /i 2I and n D
P
.nj /j 2J , respectively. Hence the maps um W A.I / ! M , .ai /i 2I 7! i 2I ai mi and un are
isomorphisms of A-modules. Then we can combine several of the above isomorphisms to
obtain an isomorphism
M ˝A N Š A.I / ˝A A.J / Š A ˝A A/.I J Š A.I J /

(14.15) (14.13)
sending mi ˝ nj to e.i;j / . In particular, M ˝A N is again a free module with basis

.mi ˝ nj /.i;j /2I J .
Another useful fact is that the tensor product and the Hom-functor are adjoint to each
other. This is based on the easy observation that for all A-modules M , N , and P the map
LA .M; N I P / ! HomA .M; HomA .N; P //; ˛ 7! .m 7! .n 7! ˛.m; n//
is an isomorphism of A-modules, functorial in M , N , and P : An inverse is given by

u 7! ..m; n/ 7! u.m/.n//. As LA .M; N I P / D HomA .M ˝ N; P / we obtain a functorial
isomorphism
HomA .M ˝ N; P / Š HomA .M; HomA .N; P //: (14.17)
In particular, the covariant functor ./ ˝ N from (A-Mod) to itself is left adjoint to the
functor HomA .N; /. Hence Proposition 13.47 shows that ./ ˝ N commutes with col-
imits and is in particular right exact. This observation also gives a new argument, why
tensor products commute with direct sums because the direct sum is, as a coproduct in the
category of modules, a special case of a colimit.
There is also a functorial homomorphism of A-modules
HomA .M; P / ˝A N ! HomA .M; P ˝A N /;

(14.18)
u ˝ n 7! .m 7! u.m/ ˝ n/:
This is an isomorphism if M or N is free with a finite basis. Indeed, assume that M Š

An for some integer n 0. As (14.18) is functorial, we may assume that M D An .
As the functors HomA .; P / and HomA .; P ˝A N / commute with finite direct sums
(Remark 14.8), we may assume M D A. Now the claim follows from Remark 14.7. The
proof in the case that N Š An is similar using that tensor products commute with direct
sums (14.15) and that Q ˝A A D Q for every A-module Q (14.13).
Taking P D A in (14.18) we obtain a functorial homomorphism of A-modules
ıW M _ ˝A N ! HomA .M; N /; (14.19)
which is an isomorphism if M or N are free with finite basis.

14.3 Tensor Algebra, Exterior Algebra, Symmetric Algebra 303
For every A-module Q we can compose the A-linear map HomA .Q; ı/ with (14.17)
and obtain a functorial homomorphism of A-modules
HomA .Q; M _ ˝A N / ! HomA .Q ˝A M; N /; (14.20)
which is an isomorphism if M or N are free with a finite basis.
Example 14.19. Let M be a free module with a finite basis. Then (14.19) yields an
isomorphism
EndA .M / WD HomA .M; M / ! M _ ˝A M: (14.21)
Composition with the linear map M _ ˝ M ! A, . ; m/ 7! .m/, yields an A-linear map
trW EndA .M / ! A; (14.22)
called the trace map. Via the choice of a basis .e1 ; : : : ; en / of M we identify EndA .M /
and the matrix algebra Mn .A/. If .x 1 ; : : : ; x n / is the dual basis of M _ , then x i ˝ej 2 M _
corresponds via (14.21) to the matrix Eij , which has everywhere 0 coefficients except in
.i; j /-th place, where its coefficient is 1. Hence the above trace map coincides with the
usual trace map Mn .A/ ! A because both maps are linear and agree on the basis .Eij /i;j
of Mn .A/.
14.3 Tensor Algebra, Exterior Algebra, Symmetric Algebra
Tensor Algebra
For r 0 let
T r .M / WD TAr .M / WD M ˝r WD M ˝A ˝A M
„ ƒ‚ …
r times
be the r-th tensor power of M . Then T 0 .M / D A and T 1 .M / D M . Define the tensor

algebra of M M
T .M / WD TA .M / WD T r .M /
r0
with A-algebra structure defined as follows. For m WD m1 ˝ ˝ mr 2 T r .M / and

n WD n1 ˝ ˝ ns 2 T s .M / we define the product by
m ˝ n WD m1 ˝ ˝ mr ˝ n1 ˝ ˝ ns 2 T rCs .M /;
then extend by linearity. This is well defined and makes T .M / into an associative A-
algebra with 1 (but not necessarily commutative).
The A-algebra T .M / is a graded A-algebra, i.e., for w 2 T i .M /, v 2 T j .M / one has
w ˝ v 2 T i Cj .M / for all i; j 0.
If uW M ! N is a homomorphism of A-modules, then m1 ˝ ˝ mr 7! u.m1 / ˝

˝ u.mr / defines an A-linear map T r .u/W T r .M / ! T r .N /. The map
M
T .u/ WD T r .u/W T .M / ! T .N /
is a homomorphism of A-algebras.
The formation of T r .u/ and of T .u/ is compatible with composition and sends identi-
ties to identities. Hence we obtain functors
T r W (A-Mod) ! (A-Mod); T W (A-Mod) ! (A-Alg):
Remark 14.20. Denote by W M ! T 1 .M / T .M / the inclusion. Then .T .M /; / has

the following universal property: For every associative A-algebra B with 1 the map
Hom(A-Alg) .T .M /; B/ ! Hom(A-Mod) .M; B/; ' 7! ' ı
is bijective. In other words, the functor M 7! T .M / is left adjoint to the forgetful functor
(A-Alg) ! (A-Mod).
Exterior Powers
Let M and N be A-modules. An r-multilinear map ˛W M r ! N is called alternating if
˛.m1 ; : : : ; mr / D 0 whenever there exists i ¤ j with mi D mj .
Let Qr T r .M / be the A-submodule generated by the set
˚
Er WD m1 ˝ ˝ mr 2 T r .M / I there exist 1 i ¤ j r with mi D mj :
Then an r-multilinear map ˛W M r ! N is alternating if and only if the corresponding

A-linear map uW T r .M / ! N sends Qr to 0. We set
r .M / WD Ar .M / WD T r .M /=Qr
and call it the r-th exterior power of M . The image of m1 ˝ ˝ mr in r .M / is denoted

by m1 ^ ^ mr . We have Q0 D Q1 D 0 and hence 0 .M / D A and 1 .M / D M .
By construction, r .M / has the following universal property. Let
W M r ! r .M /; .m1 ; : : : ; mr / 7! m1 ^ ^ mr :
Then is alternating and for every A-module N the map
Hom(A-Mod) .r .M /; N / ! f ˛W M r ! N I ˛ alternatingg ; u 7! u ı
is bijective.
The direct sum

M M
.M / WD A .M / WD r .M / D T .M /= Qr
r0 r0
is called the exterior algebra of M . The multiplication on the A-algebra T .M / induces

a multiplication on .M / making the A-module .M / into an A-algebra. We denote the
multiplication in .M / by .!; / 7! ! ^ .
As elements of the form m1 ˝ ˝ mr , mi 2 M , generate T r .M /, their images
m1 ^ ^ mr generate r .M /. A map u given on these elements with values in an A-
module N extends (necessarily uniquely) to an A-linear map uW r .M / ! N if and only
if u satisfies linearity conditions in each component (as in (14.8)) and if u sends every
element in Er to 0.
The A-algebra .M / is a graded A-algebra that is graded commutative, i.e., for ! 2
i .M /, 2 j .M / one has
! ^ D .1/ij ^ !:
Let uW M ! N be a homomorphism of A-modules. Then the homomorphism T .u/

of A-algebras and the A-linear maps T r .u/ induce a homomorphism of A-algebras
.u/W .M / ! .N / and A-linear maps r .u/W r .M / ! r .N / given by m1 ^ ^
mr 7! u.m1 / ^ ^ u.mr /.
We obtain a functor M 7! .M / from the category (A-Mod) to the category of asso-
ciative graded commutative A-algebras and functors r W (A-Mod) ! (A-Mod).
Remark 14.21. Let M be an A-module. As .M / is graded commutative one has for
m1 ; m2 2 M that m1 ^ m2 D m2 ^ m1 in 2 .M /. As the group of permutations Sr is
generated by transpositions of i and i C 1 and as sgnW Sr ! f˙1g is a homomorphism of
groups, we deduce that for all m1 ; : : : ; mr 2 M and for every 2 Sr we have
m .1/ ^ ^ m .r/ D sgn. /m1 ^ ^ wr : (14.23)
For finitely generated free modules, the exterior power can be described as follows
([BouA1] III, §7.8 and §7.9). Let r; n 0 be integers. Define
Lr .n/ WD f .i1 ; : : : ; ir / I 1 i1 < < ir n g :
Let E be an A-module, e1 ; : : : ; en 2 E. For J D .i1 ; : : : ; ir / 2 Lr .n/ define eJ WD

ei1 ^ ^ eir 2 r .E/.
Proposition 14.22. Let E be a free A-module of rank n 2 N.
1. Let r 2 N0 . The exterior power r .E/ is a free A-module of rank nr (with nr D 0

if r > n). For 1 r n a family .e1 ; : : : ; en / of elements in E is a basis of E if and
only if .eJ /J 2Lr .n/ is a basis of r .E/.
2. Let F be a free A-module with basis .f1 ; : : : ; fm / with m 2 N0 and let uW E ! F be
P
an A-linear map given by u.ej / D m i D1 aij fi for a matrix .aij / 2 Mmn .A/. Then
the matrix of r .u/ with respect to the bases .eJ /J 2Lr .n/ of r .E/ and .fI /I 2Lr .m/ of
Vr
.F / is the matrix
.det.AI;J //I 2Lr .m/;J 2Lr .n/ ;
where I D fi1 < < ir g, J D fj1 < < jr g and
AI;J D .ai ;j / 1 r

1r
the submatrix of A consisting only of rows (respectively columns) numbered by ele-

ments in I (respectively in J ).
3. In particular, if r D n D m, then det.E/ WD n .E/ and det.F / WD n .F / are free
A-modules of rank 1 and det.u/ WD n .u/ is given by det.A/ with respect to the bases
e.1;:::;n/ and f.1;:::;n/ .
Let E be a free A-module of rank n 2 N and let uW E ! E be an A-linear endomor-

phism. Then det.u/ is an endomorphism of the free A-module det.E/ of rank 1 and hence
is given by multiplication with an element in A that we call again det.u/.
Proposition 14.23. The endomorphism u is bijective if and only if det.u/ is a unit in A.
Proof. Let .e1 ; : : : ; en / be a basis of E and set xi WD u.ei /. Then
x1 ^ ^ xn D det.u/.e1 ^ ^ en /: (*)
The linear map u is bijective if and only if .x1 ; : : : ; xn / is a basis of E. By Proposi-

tion 14.22 1 this is the case if and only if .x1 ^ ^ xn / is a basis of the free A-module
n .E/ of rank 1. As .e1 ^ ^ en / is a basis, (*) shows that .x1 ^ ^ xn / is a basis if
and only if det.u/ is a unit in A.
Remark 14.24. Let M be an A-module and let ˛1 ; : : : ; ˛r W M ! A be A-linear maps.

Then
M r ! A; .m1 ; : : : ; mr / 7! det..˛i .mj //1i;j r /
is an alternating r-multilinear map and hence corresponds to an A-linear map r .M / !
A, which we denote by ˛1 ^ ^ ˛r . The map
.M _ /r ! r .M /_ ; .˛1 ; : : : ; ˛r / 7! ˛1 ^ ^ ˛r
is r-multilinear and alternating. Hence it corresponds to an A-linear map
r .M _ / ! r .M /_ : (14.24)
Now suppose that M is finitely generated and free. We claim that (14.24) is then an
isomorphism. Indeed, let .e1 ; : : : ; en / be a basis of M , and let ."1 ; : : : ; "n / be the dual
basis of M _ . Then (14.24) sends for J 2 Lr .n/ the J -th basis vector "J of r .M _ / to
the J -th basis vector in the dual basis of the basis .eJ /J 2Lr .n/ of r .M /. In particular our
claim follows.
Symmetric Products
A similar construction as for alternating multilinear maps also works for symmetric mul-
tilinear maps. Again let M and N be A-modules. An r-multilinear map ˛W M r ! N
is called symmetric if ˛.m1 ; : : : ; mr / D ˛.m .1/ ; : : : ; m .r/ / for all permutations in the
symmetric group Sr of f1; : : : ; rg.
Let Pr T r .M / be the A-submodule generated by all elements of the form
.m1 ˝ ˝ mr / .m .1/ ˝ ˝ m .r/ /
with mi 2 M and 2 Sr . Again an r-multilinear map ˛W M r ! N is symmetric if and

only if the corresponding A-linear map uW T r .M / ! N sends Pr to 0. We set
Symr .M / WD SymAr .M / WD T r .M /=Pr
and call it the r-th symmetric power of M . The image of m1 ˝ ˝ mr in Symr .M /

is denoted by m1 m2 mr . We have P0 D P1 D 0 and hence Sym0 .M / D A and
Sym1 .M / D M . The A-module Symr .M / has again a universal property, now with
respect to symmetric r-multilinear maps. As above, the direct sum
M
Sym.M / WD SymA .M / WD Symr .M /
r0
is a graded A-algebra, called the symmetric algebra of M It is commutative in the usual

sense. Again all these constructions are functorial yielding functors Sym from (A-Mod)
to the category of commutative graded algebras and Symr from (A-Mod) to itself. Finally,
by [BouA1] III, §6.6, Theorem 1 we have the following result about the symmetric powers
of free modules.
.I /
Proposition 14.25. Let M be a free module with basis .ei /i 2I . Let N0 be the set of
maps ˇW I ! N0 such that ˇ.i/ D 0 for all but finitely many i. Write
Y ˇ.i /
e ˇ WD ei ;
i 2I
product in the commutative A-algebra Sym.M /. Then .e ˇ /ˇ2N .I / is a basis of the A-

P 0
module Sym.M /. For all r 0 the elements e ˇ with i 2I ˇ.i/ D r form a basis of
Symr .M /.
14.4 Modules over Commutative Rings
Now let A be a commutative ring. We denote by A WD f a 2 A I 9 b 2 A W ab D 1 g its

group of units.
Recall that an ideal of A is an A-submodule of A. In other words, an ideal a of A is
a subgroup a of .A; C/ such that ba 2 a for all b 2 A and a 2 a.
Trivial examples of ideals of A are f0g and A itself. For an ideal a of A one has a D A
if and only if 1 2 a. The commutative ring A is a field if and only if A ¤ f0g and f0g and
A are the only ideals of A.
Remark 14.26. If a is an ideal of A then the multiplication A A ! A induces a map

A=a A=a ! A=a on the quotient group A=a making A=a again into a commutative
ring. The projection W A ! A=a is then a ring homomorphism.
N yields a bijection between the set of ideals in A=a
Sending an ideal bN of A=a to 1 .b/
and the set of ideals in A containing a.
Definition 14.27. Let a A be an ideal and M an A-module. The product of a and M

is the A-submodule of M defined by
n Xn o
aM WD ai mi I n 2 N0 ; ai 2 a; mi 2 M :
i D1
Proposition 14.28. Let a A be an ideal and M an A-module. Then the map uW M !

M ˝A A=a, m 7! m˝1 induces an isomorphism of A-modules M=aM ! M ˝A .A=a/.
Proof. For a 2 A or m 2 M let aN (respectively m)

N be its image in A=a (respectively in
M=aM ). For a 2 a and m 2 M we have .a m/ ˝ 1 D m ˝ .a 1/ D m ˝ aN D 0
so that aM is contained in the kernel of u. Denote by iW M=aM ! M ˝A .A=a/ the
induced homomorphism. Conversely, define a map M .A=a/ ! M=aM by .m; a/ N 7!
am. This assignment is well defined and A-bilinear so that it induces an A-linear map
j W M ˝A .A=a/ ! M=aM .
Now j ı i D idM=aM hence i must be injective. Moreover i is also surjective because
M ˝A .A=a/ is generated by elements of the form m ˝ aN D .a m/ ˝ 1N for m 2 M and
a 2 A. Hence i and j are mutually inverse isomorphisms.
Definition 14.29. An ideal m ¨ A is called maximal if there exists no proper ideal a ¨ A

with m ¨ a.
14.4 Modules over Commutative Rings 309
Remark 14.30. Fix an ideal a ¨ A. If .ai /i 2I is a family of proper ideals of A, totally

S
ordered by inclusion such that a ai for all i 2 I , then i ai is a proper ideal of A
containing a. Hence we can employ Zorn’s lemma 13.28 to see that every proper ideal a
of a commutative ring is contained in a maximal ideal. Applying this to a D f0g, we see
in particular there always exist maximal ideals of A if A ¤ f0g.
Remark 14.31. An ideal a of A is maximal if and only if A=a is a field (Remark 14.26).
Definition 14.32. Let 'W A ! B be a homomorphism of commutative rings. For every

A-module M the base change of M along ' is defined as MB WD M ˝A B. There is
a unique A-linear map MB B ! MB with .m ˝ b; b/ Q 7! m ˝ .b b/,
Q making MB into
a B-module.
If f W M ! N is an A-linear map, then fB WD f ˝ idB W MB ! NB is B-linear.
These constructions define a functor . /B from the category of A-modules to the cate-
gory of B-modules.
For commutative rings A ¤ f0g free A-modules have a well-defined rank:
Definition and Remark 14.33 (Rank of a free module). Let A ¤ 0 and let M be a free
A-module, hence M Š A.I / for some set I . Then the cardinality of I is called the rank
of M . It is denoted by rkA .M /.
To see that the rank is well defined it suffices to prove that the existence of an A-linear
isomorphism f W A.I / ! A.J / implies that I and J have the same cardinality. As A ¤ 0,
we can choose a maximal ideal m A and set k WD A=m. One has k .I / Š A.I / ˝A k
because tensor products commute with direct sums. Hence
fk W k .I / Š A.I / ˝A k ! A.J / ˝A k Š k .J /
is also an isomorphism (of k-vector spaces). Since k is a field (Remark 14.31), it is

a standard result of linear algebra that I and J have the same cardinality.
For free modules over non-commutative rings the rank is in general not well defined:
There exist non-commutative rings B and an isomorphism of B-modules B ! B 2 (Prob-
lem 14.28).
Example 14.34. Let M and N be free A-modules of finite rank m and n respectively and
let r 0 be an integer. Above we have expressed bases of M ˝ N , r .M /, and Symr .A/
in terms of bases of M and N . In particular we see that
rkA .M ˚ N / D m C n; rkA .M ˝A N / D mn;

rkA .HomA .M; N // D mn; rkA .M _ / D m;
! !
m mCr 1
rkA . .M // D
r
; rkA .Sym .M // D
r
:
r r
Remark 14.35. Let M be a free A-module and let 'W A ! B be a homomorphism of

commutative rings. Choose an isomorphism f W M ! A.I / for some set I . Its base
change via ' is an isomorphism fB W MB ! B .I / because tensor products commute with
direct sums. Hence MB is a free B-module of the same rank as the free A-module M .
Definition 14.36 (Local ring). A commutative ring A is called local if it has a unique
maximal ideal m A. The field A=m is called the residue field of A.
In particular a local ring is always non-zero.
Remark 14.37. A commutative ring A is local if and only if A n A is an ideal. In this

case m WD A n A is the unique maximal ideal of A. In particular for a 2 m we have
1 C a 2 A .
Proposition 14.38 (Nakayama’s lemma). Let A be a local ring with residue field k, let
M 0 and M be A-modules, and let f W M 0 ! M be an A-linear map. Suppose that M
is a finitely generated A-module. Then f is surjective if and only if the induced map
fNW M 0 =mM 0 ! M=mM is surjective.
Proof. We have a commutative diagram
f
M0 M
fN
M 0 mM 0 M=mM
where vertical maps are the projections. In particular the condition is necessary. Now let
fN be surjective. Let N WD Coker.f /. Our aim is to show N D 0. As fN is surjective, we
have N=mN D 0 and hence N D mN . As a quotient of M , N is again finitely generated.
Hence it remains to show that for a finitely generated A-module N the equality N D mN
implies N D 0.
Assume N ¤ 0 and N D mN . Since N is finitely generated, there exists a minimal
set of generators n1 ; : : : ; nr 2 N for r 2 N. As nr 2 N D mN , we find a1 ; : : : ; an 2 m
with nr D a1 n1 C : : : C ar nr . Since A is local, we find 1 ar 2 A so that nr D
.1 ar /1 .a1 n1 C : : : C nr1 nr1 / and therefore n1 ; : : : ; nr1 already generate N . This
contradicts the minimality of r 2 N.
Proposition 14.39. Let A be a local ring with maximal ideal m, k WD A=m. Let E be
a free A-module of finite rank and let uW E ! E be an A-linear endomorphism. Then u
N E=mE ! E=mE induced by u is
is an automorphism if and only if the k-linear map uW
an automorphism.
14.5 Problems 311
Proof. The determinant det.u/ N 2 k is the image of det.u/ 2 A. Hence det.u/ N is a unit in
N ¤ 0) if and only if det.u/ is a unit in A (i.e., det.u/ … m). We conclude by
k (i.e., det.u/
Proposition 14.23.
The conclusion of Proposition 14.39 also holds for arbitrary finitely generated A-
modules E: If uN is an automorphism, then u is surjective by Nakayama’s lemma. Then
conclude by Problem 14.30.
14.5 Problems
In the problem section let R be always a ring and A a commutative ring.
Problem 14.1. Let R be a ring. Show that a morphism in the category of R-modules is
a monomorphism (respectively an epimorphism) if and only if it is injective (respectively
surjective).
Problem 14.2. Let u; vW M ! N be a homomorphisms of R-modules. Show that

Ker.u; v/ WD Ker.u v/ (respectively that Coker.u; v/ WD Coker.u v/) is an equal-
izer (respectively a coequalizer) in the category of R-modules (Problem 13.9). Deduce
that the category of R-modules is complete and cocomplete (use Problem 13.10).
Problem 14.3. Let I be a small filtered category and let M W I ! (R-Mod) be an I -

diagram of R-modules. By composition with the forgetful functor (R-Mod) ! (Sets)
`
we also obtain an I -diagram MQ of sets. Define on colimI MQ D . i 2Ob.I / MQ i /= an
addition and a scalar multiplication as follows. For a 2 R and m; m0 2 colimI MQ choose
representatives m Q i 2 MQ i D Mi and m Q j0 2 MQ j D Mj and choose morphisms 'W i ! k
and W j ! k in I for some object k in I . Define m C m0 as the equivalence class of
MQ .'/.mQ i / C MQ . /.m Q j0 /. Define am as the equivalence class of am Q i . Show that this
defines the structure of a left R-module on colimI MQ , that the canonical maps si W M D
MQ i ! colimI MQ are R-linear, and that .colimI MQ ; .si /i / is a colimit in the category of
R-modules.
Problem 14.4. Let P be an R-module. Show that the following conditions on P are
equivalent:
(i) Every short exact sequence of R-modules of the form 0 ! M ! N ! P ! 0 is

split.
(ii) For every surjective homomorphism pW MQ ! M of R-modules and for every R-
linear map uW P ! M there exists an R-linear map uW Q P ! MQ such that p ı uQ D u.
(iii) P is a direct summand of a free module (i.e., there exists an R-module P 0 such that
P ˚ P 0 is a free R-module).
If P satisfies these conditions, P is called projective.

Show that an R-module is projective and finitely generated if and only if it is a direct
summand of a finitely generated free R-module.
Problem 14.5. Let P be an A-module. Show that P is projective and finitely generated
(Problem 14.4) if and only if the homomorphism P _ ˝A P ! EndA .P / (14.19) is an
isomorphism.
Problem 14.6. A non-zero commutative ring R is called an integral domain if for a; b 2

R the equality ab D 0 implies a D 0 or b D 0. Suppose now that R is an integral domain.
An R-module M is called torsion free if for a 2 R and m 2 M the equality am D 0
implies a D 0 or m D 0.
1. Show that Z and kŒT (k a field) are integral domains. Let n 1 be an integer. Show
that Z=nZ is an integral domain if and only if n is a prime number.
2. Show that every free R-module is torsion free.
3. Let a 2 R with a ¤ 0. Show that the R-module R=.a/ is not torsion free (and hence
not free).
4. The field of fractions of R, denoted by Frac.R/, is the set of pairs .a; s/ with a 2 A
and s 2 A n f0g modulo the equivalence relation .a; s/ .b; t/ , at D bs. The
equivalence class of .a; s/ is denoted by as . Define addition and multiplication as usual
for fractions and show that this is well defined and endows Frac.R/ with the structure
of a field. Observe that Frac.Z/ D Q.
5. Show that the map W R ! Frac.R/, a 7! a1 is an injective ring homomorphism satisfy-
ing the following universal property: If K is a field and 'W R ! K is an injective ring
homomorphism, then there exists a unique homomorphism ' 0 W Frac.R/ ! K such
that ' 0 ı D '.
6. Show that Frac.R/ is a torsion free R-module. Show that Frac.R/ is not a free R-
module if R is not a field.
Problem 14.7. Let n; m 2 Z be non-zero and let d be a greatest common divisor of m

and n. Show that the rings (considered as Z-algebras) Z=mZ ˝Z Z=nZ and Z=d Z are
isomorphic (in particular Z=mZ ˝Z Z=nZ D 0 if m and n are coprime).
Problem 14.8. For integers r; n 2 consider the inclusion uW rZ=nrZ ,! Z=nrZ of

Z=nrZ-modules. Show that u ˝ idZ=rZ is not injective.
Problem 14.9. Let E be an A-module.

(i) For every injective A-linear map uW M ! M 0 the map u˝idE W M ˝E ! M 0 ˝E
is injective.
14.5 Problems 313
i p
(ii) For every short exact sequence 0 ! M 0 ! M ! M 00 ! 0 of A-modules the
i ˝idE p˝idE
sequence 0 ! M 0 ˝ E ! M ˝ E ! M 00 ˝ E ! 0 is exact.
In this case E is called flat.
2. Show that every free A-module is flat. Deduce that every projective A-module (Prob-
lem 14.4) is flat.
3. Suppose that A is an integral domain (Problem 14.6). Show that a flat A-module is
torsion free.
Problem 14.10. Let M and N be A-modules. An r-multilinear map ˛W M r ! N is

called skew-symmetric if ˛.m .1/ ; : : : ; m .r/ / D sgn. /˛.m1 ; : : : ; mr / for every permuta-
tion 2 Sr and for all m1 ; : : : ; mr 2 M .
1. Show that every alternating r-multilinear map M r ! N is skew-symmetric.

2. Show that an r-multilinear map ˛W M r ! N is alternating if and only if
˛.m1 ; : : : ; mr / D 0 whenever there exists 1 i r 1 such that mi D mi C1 .
3. Show that every skew-symmetric r-multilinear map M r ! N is alternating if the map
N ! N , n 7! 2n is injective.
The injectivity of N ! N , n 7! 2n, is satisfied if 2 is a unit in A (for instance if A is

a commutative algebra over a field of characteristic ¤ 2).
Problem 14.11. Let M be an A-module and let r 0 be an integer. Show that

X
m1 ^ ^ mr 7! sgn. /.m .1/ ˝ ˝ m .r/ /
2Sr
defines an A-linear map ar W r .M / ! T r .M /. If pr W T r .M / ! r .M / is the canonical

projection, then pr ı ar D .rŠ/ idr .M / . Show that ar is injective if M is a free A-module
or if rŠ is invertible in A. Give an example of a ring A and an A-module M such that a2
is not injective.
Problem 14.12. Let M be an A-module, r 0 an integer, ! 2 r .M /. Show that if r is

odd, then ! ^ ! D 0. Give an example of a module M and an element ! 2 2 .M / such
that ! ^ ! ¤ 0.
Problem 14.13. Let M be an A-module generated by a single element. Show that

Sym.M / D T .M / and that r .M / D 0 for all r 2.
Problem 14.14. Let R be a commutative ring, let n 2 be an integer, and let

A D RŒT1 ; : : : ; Tn be the ring of polynomials over R in n indeterminates. Let a WD
.T1 ; : : : ; Tn / be the ideal in A generated by T1 ; : : : ; Tn . Show that An .a/ D A=a D R.
Problem 14.15. Let uW M ! N be a homomorphism of A-modules.
1. Suppose that u is surjective. Show that T r .u/, r .u/, and Symr .u/ are surjective for
all r 0.
2. Show that an analogous assertion for “injective” does not hold.
Hint: For T r take the inclusion Z-modules uW 2Z=4Z ! Z=4Z and use Problem 14.13
for Symr . For r use Problem 14.14.
3. Suppose that u is injective and u.M / is a direct summand of N (i.e., there exists
a submodule N 0 of N such that N D u.M / ˚ N 0 ). Show that T r .u/, r .u/, and
Symr .u/ are injective for all r 0 and their images are direct summands (see also
Problem 14.17).
Problem 14.16. Let M be a projective A-module (Problem 14.4), m1 ; : : : ; mr 2 M .

Show that .m1 ; : : : ; mr / is linearly independent if and only if there exists no 0 ¤ a 2 A
such that am1 ^ ^ mr D 0.
Hint: Consider first the case that M is a free A-module.
Problem 14.17. Let uW M ! N be an injective homomorphism of projective A-modules

(Problem 14.4). Show that r .u/ is injective for all r 0.
Problem 14.18. Let m; n 2 N0 and uW Am ! An be an A-linear map. Show that if u is

injective (respectively surjective), then m n (respectively m n).
Hint: One could use Problem 14.17 and Problem 14.15.
Problem 14.19. Let M be a free A-module or rank d 2 N0 and let uW M ! M be an

A-linear endomorphism. Show that u is injective if and only if d .u/ is injective.
Problem 14.20. Let d 2 N0 and let M be an A-module generated by elements

m1 ; : : : ; md . Suppose that there exists an A-linear injection Ad ,! M . Show that
.m1 ; : : : ; md / is a basis of M .
Hint: Use Problem 14.19.
Problem 14.21. For every A-module M let M W M ! T 1 .M / T .M / be the inclu-

sion. Show that for every homomorphism uW M ! N of A-modules T .u/ is the unique
homomorphism of A-algebras such that
u
M N
M N
T .u/
T .M / T .N /
commutes. Formulate and prove an analogous statement for the exterior algebra and the
symmetric algebra.
14.5 Problems 315
0
Problem 14.22. Let M be an A-module and let M W M ! 1 .M / ,! .M / be the
inclusion. Show that for every A-algebra C and for every A-linear map uW M ! C
with u.m/2 D 0 for all m 2 M there exists a unique homomorphism of A-algebras
0
Q .M / ! C such that u D uQ ı M
uW .
Problem 14.23. Let A be an integral domain and let K be its field of fractions (Prob-
lem 14.6). Show that A2 .K/ D 0.
Problem 14.24. Let M be an A-module and r 2 N0 .
1. Suppose that r .M / D 0 (respectively that r .M / is a finitely generated A-module).

Show that s .M / has the same property for all s r.
2. Let uW M ! N be a homomorphism of A-modules and suppose that r .u/ D 0.
Show that s .u/ D 0 for all s r.
Problem 14.25. Let M be an A-module, let r 0 be an integer and assume that r .M /
is generated by d 2 N0 elements. Show that s .M / D 0 for all s > r C d .
Problem 14.26. Let M be an A-module, let N be a free A-submodule of M of rank

d 2 N, and let .e/ be a basis of the free A-module d .N / of rank 1. Suppose that N is
a direct summand of M . Show that an element m 2 M is contained in N if and only if
e ^ m D 0.
Problem 14.27. Show that the functor M 7! Sym.M / is left adjoint to the forgetful
functor from the category of commutative A-algebras to (A-Mod).
Problem 14.28. Let k be a field, let V be a k-vector space with basis .ei /i 2N , and let
R D Endk .V / be the ring of endomorphisms. Define f1 ; f2 2 R by f1 .ei / D e2i 1
and f2 .ei / D e2i . Show that .f1 ; f2 / is an R-basis of the R-left module R. Deduce that
Rn Š Rm for all n; m 2 N.
Problem 14.29. Show the following version of Nakayama’s lemma. Let M be a finitely
generated module over a commutative ring A, let a A be an ideal. If M D aM , then
there exists a 2 A such that aM D 0 and a 1 mod a.
Problem 14.30. Let M be a finitely generated module over a commutative ring A and let
uW M ! M be a surjective A-linear endomorphism. Show that u is bijective.
Hint: View M as an AŒT -module by setting T m WD u.m/ and use Problem 14.29.
Note that injective A-linear endomorphisms u are not necessarily bijective (e.g., M D
A D Z and uW n 7! 2n).
Problem 14.31. Show that every finitely generated direct summand of a finitely generated
free module over a local ring A is again free. In other words (Problem 14.4), every finitely
generated projective A-module is free.
Appendix D: Homological Algebra
15
In this appendix we recall some basic notions from homological algebra. We start by
defining complexes of modules and the notion of homotopy between morphisms of com-
plexes. This is in fact the cohomological shadow of the topological notion of homotopy
(see the remark after Proposition 11.12). One obtains the homotopy category of modules,
which is a fundamental notion in homological algebra. After a short interlude on diagram
chases we introduce a central notion for the definition of cohomology: injective modules
and K-injective complexes. Until then all notions were explained for modules over a ring,
but in fact they make sense much more generally in arbitrary abelian categories. This is
explained in the last section of this appendix.
Notation: If not otherwise specified, R always denotes a ring, not necessarily commu-
tative. An R-module is always a left R-module.
15.1 Homotopy Category of Modules
Definition 15.1 (Complexes). A sequence

d i 1 di
M D . ! M i 1 ! M i ! M i C1 ! : : : /
of homomorphisms of R-modules is called a complex if d i ı d i 1 D 0 for all i 2 Z

(equivalently Im.d i 1 / Ker.d i /). Sometimes we write dM
i
instead of d i .
A morphism of complexes of R-modules uW M ! N is a family ui W M i ! N i of
homomorphisms of R-modules such that for all i 2 Z the diagram
i
dM
Mi M i C1
(15.1)
ui ui C1
dNi
Ni N i C1
DOI 10.1007/978-3-658-10633-1_15
318 15 Appendix D: Homological Algebra
commutes. We obtain the category (Com(R)) of complexes of R-modules.

A complex M is called bounded below if there exists a 2 Z such that M i D 0 for all
i < a.
We consider every R-module M as a complex M with M 0 D M and M i D 0 for all

i ¤ 0.
Definition 15.2 (Cohomology and quasi-isomorphisms). Let M be a complex of R-

modules. Then
H i .M / WD Ker.d i /= Im.d i 1 /
is called the cohomology of the complex M . For every morphism uW M ! N of
complexes of R-modules the commutativity of (15.1) shows ui .Ker.dM
i
// Ker.dNi /
i 1 i 1
and u .Im.dM // Im.dN /. Hence u induces a homomorphism of R-modules
i i
H i .u/W H i .M / ! H i .N /
for all i 2 Z.
The morphism u is called a quasi-isomorphism or shorter qis if H i .u/ is an isomor-
phism for all i 2 Z.
d i 1 di
Example 15.3. Let ! M i 1 ! M i ! M i C1 ! : : : be an exact sequence. Fix
j 2 Z. Then the vertical arrows of
d j 1
::: M j 1 Mj 0 :::
uj 1 WD0 uj WDd j uj C1 WD0
d j C1
::: 0 M j C1 M j C2 :::
yield a quasi-isomorphism of complexes: This is clear in degree ¤ j . In degree j , d j

induces an isomorphism M j = Im.d j 1 / D M j = Ker.d j / ! Im.d j / D Ker.d j C1 /.
Definition 15.4 (Homotopy of morphisms of complexes). Let M and N be complexes

of R-modules and let u; vW M ! N be two morphisms of complexes.
1. A homotopy h between u and v is a family of homomorphisms hi W M i ! N i 1 , i 2 Z,

such that
ui v i D dNi 1 ı hi C hi C1 ı dM
i
:
2. The morphisms u and v are called homotopic if there exists a homotopy between u
and v. We then write u ' v.
15.1 Homotopy Category of Modules 319
Remark 15.5. Let M and N be complexes of R-modules.
1. The relation ' is an equivalence relation on Hom(Com(R)) .M ; N /: h D 0 defines

a homotopy between u and u. If h is a homotopy between u and v, then h is a ho-
motopy between v and u. If h is a homotopy between u and v, and h0 is a homotopy
between v and w, then h C h0 is a homotopy between u and w.
2. If u ' v then u C w ' v C w for every w 2 Hom(Com(R)) .M ; N /.
3. Let tW L ! M , u; vW M ! N , wW N ! P be morphisms of complexes of R-
modules. Let h be a homotopy between u and v. Then .w i 1 ıhi ıt i /i 2Z is a homotopy
between w ı u ı t and w ı v ı t.
Remark 15.5 allows us to define:
Definition 15.6 (Homotopy category of modules). The homotopy category of complexes

of R-modules is the following category K.R/:
(a) Objects are complexes of R-modules.

(b) For two complexes M and N we set
HomK.R/.M ; N / WD Hom(Com(R)) .M ; N /= ' :
Addition of morphism of complexes endows HomK.R/ .M ; N / with the structure of

an abelian group.
(c) The identity is the identity morphism of complexes. Composition is given by compo-
sition of morphisms of complexes (this is well defined by Remark 15.5).
Proposition 15.7. Let u; vW M ! N be morphisms of complexes of R-modules that are

homotopic. Then
H p .u/ D H p .v/W H p .M / ! H p .N /
for all p 2 Z.
In particular we see that it makes sense to say that a morphism in K.R/ is a quasi-
isomorphism.
Proof. By considering u v it suffices to show: If there exists a homotopy h of u and 0,

p p1
then H p .u/ D 0 for all p. Let iW Ker.dM / ! M p be the inclusion. As up D dN ı hp C
p p1 p1
hpC1 ı dM we find that up ı i D dN ı hp ı i and hence factors through Im.dN /. Hence
H p .u/ D 0.
Remark 15.8. One also has the notion of a complex with decreasing numbering, i.e., of
di C1 di
complexes M of the form ! Mi C1 ! Mi ! Mi 1 ! : : : with di ı di C1 D 0. If
we want to distinguish between these notions, we call complexes with increasing number-
ing (i.e., complexes as defined in Definition 15.1) cochain complexes and complexes with
decreasing numbering chain complexes. All of the above definitions and results can also
be made for chain complexes.
For the proof of the next two less trivial results we refer to [GeMa] Chap. III, §4,
Theorem 4.
Lemma 15.9. Every diagram in K.R/ of the form
s
M N
qi s
u
P ;
where s is a quasi-isomorphism, can be completed to a diagram
s
M N
qi s
u v (15.2)
t
P Q ;
qi s
that commutes in K.R/ and such that t is a quasi-isomorphism.
It is in general not possible to find the commutative diagram (15.2) in (Com(R)).
Lemma 15.10. Let wW I ! J be a morphism of complexes. Then the following asser-

tions are equivalent:
qi s
(i) There exists a quasi-isomorphism sW M ! I such that w ı s ' 0.
qi s
(ii) There exists a quasi-isomorphism tW J ! N such that t ı w ' 0.
Finally, we will use the following result for which we refer to [Stacks] Tag 05T6.
Lemma 15.11. Let I be a set of objects in (R-Mod) containing 0 such that for every R-
module M there exists an injective homomorphism M ! I of R-modules with I 2 I . Let
a 2 Z. Then for every bounded below complex M of R-modules with M p D 0 for p < a
qi s
there exists a quasi-isomorphism M ! I with I p 2 I , with M p ! I p injective for
all p, and with I p D 0 for all p < a.
15.2 Diagram Chases 321
15.2 Diagram Chases
Lemma 15.12. (Five lemma) Consider a commutative diagram of R-modules with exact
rows
M1 M2 M3 M4 M5
u1 u2 u3 u4 u5
N1 N2 N3 N4 N5 :
Assume that u2 and u4 are isomorphisms, u1 is surjective and u5 is injective. Then u3 is
an isomorphism.
Proof. This is a simple exercise of diagram chasing, which we leave to the reader. Al-
ternatively, one can also work simply with the universal properties of kernel and coker-
nel. Such a proof then will also generalize to arbitrary abelian categories (see Defini-
tion 15.25). Let us illustrate this by proving that Coker.u3 / D 0.
Denote by vi W Mi ! Mi C1 and wi W Ni ! Ni C1 the homomorphisms in the diagram.
Let n3 W N3 ! P be a homomorphism such that n3 ıu3 D 0. We have to show that n3 D 0.
As n3 ı w2 ı u2 D n3 ı u3 ı v2 D 0 and as Coker.u2 / D 0 we find n3 ı w2 D 0. Hence
Ker.w3 / D Im.w2 / Ker.n3 / and there exists n4 W Im.w3 / ! P with n4 ı w3 D n3 . As
n4 ı u4 ı v3 D 0 we find
Ker.w4 ı u4 / D Ker.u5 ı v4 / D Ker.v4 / D Im.v3 / Ker.n4 ı u4 /
and hence Im.w3 / D Ker.w4 / Ker.n4 / because u4 is an isomorphism. Therefore

n4 D 0 and hence n3 D n4 ı w3 D 0.
Lemma 15.13. (Snake lemma) A commutative diagram of R-modules with exact rows
p
M1 M2 M3 0
u1 u2 u3
i
0 N1 N2 N3 :
induces an exact sequence

@
Ker.u1 / ! Ker.u2 / ! Ker.u3 / ! Coker.u1 / ! Coker.u2 / ! Coker.u3 /;
where “@ D i 1 ı u2 ı p 1 ”.
The informal definition of @ means that for x 2 Ker.u3 / one first chooses z 2 M2 with
p.z/ D x and then defines @.x/ as the image of an element y 2 N1 with i.y/ D u2 .z/.
Note that if M1 ! M2 is injective, then Ker.u1 / ! Ker.u2 / is injective. If N2 ! N3
is surjective, then Coker.u2 / ! Coker.u3 / is surjective.
Proof. Diagram chase.

u v
Lemma 15.14. Let 0 ! M ! N ! P ! 0 be a short exact sequence of
complexes of R-modules (i.e., for all i 2 Z the sequence of R-modules 0 ! M i !
N i ! P i ! 0 is exact). Then there exist for all i 2 Z homomorphisms ıW H i .P / !
H i C1 .M / such that the sequence
ı H i .u/ H i .v/
: : : ! H i .M / ! H i .N / ! H i .P /
ı H i C1 .u/ H i C1 .v/
! H i C1 .M / ! H i C1 .N / ! H i C1 .P / (15.3)
ı
! : : :
is exact and such that this long exact sequence is functorial for morphisms of exact se-
quences.
Proof. Applying the the snake lemma to
0 M i 1 N i 1 P i 1 0
d i 1 d i 1 d i 1
0 Mi Ni Pi 0
shows that the rows of the following commutative diagram are exact:
i 1
Coker.dM / Coker.dNi 1 / Coker.dPi 1 / 0
i
dM dNi dPi (*)
i C1
0 Ker.dM / Ker.dNi C1 / Ker.dPi C1 /:
Note that
i
dM
i 1 i C1
Ker.Coker.dM / ! Ker.dM // D H i 1 .M /;
i
dM
i 1 i C1
Coker.Coker.dM / ! Ker.dM // D H i .M /:
Hence applying the snake lemma to (*) yields an exact sequence

ı
H i .M / ! H i .N / ! H i .P / ! H i C1 .M / ! H i C1 .N / ! H i C1 .P /
and the long exact sequence (15.3) is obtained by pasting these sequences together. We
omit the proof of the functoriality of ı in morphisms of exact sequences.
15.3 Injective Modules and K-injective Complexes 323
15.3 Injective Modules and K-injective Complexes
We will define the cohomology of sheaves – or more general derived functors – using
injective objects and K-injective complexes. These are introduced here.
Definition 15.15 (Injective module). An R-module I is called injective if for every dia-
gram of R-modules
i
M0 M
u
I
Q M ! I such that uQ ı i D
with i injective, there exists a homomorphism of R-modules uW
u.
The notion of an injective module is dual to the notion of a projective module (Prob-
lem 14.4).
Proposition 15.16. Let I be an R-module. Then the following assertions are equivalent:
(i) I is an injective R-module.

(ii) Every short exact sequence of R-modules of the form 0 ! I ! M ! M 00 ! 0
splits.
(iii) For every complex M of R-modules every quasi-isomorphism sW I ! M has a left
inverse in K.R/ (i.e., there exists a morphism of complexes rW M ! I , where I is
considered as complex concentrated in degree 0, such that r ı s ' idM ).
i
Proof. “(i) ) (ii)”. Let 0 ! I ! M ! M 00 ! 0 be an exact sequence. As I is
Q M ! I with uQ ı i D idI . Hence the sequence
injective, there exists a homomorphism uW
splits.
“(ii) ) (iii)”. Let sW I ! M be a quasi-isomorphism. In particular H a .M / D 0
for a ¤ 0. Consider the complex 0 M given by
: : : ! 0 ! Coker.M 1 ! M 0 / ! M 1 ! M 2 ! : : :
where the cokernel term is in degree 0. Then M ! 0 M is a quasi-isomorphism

because H a .M / D 0 for a < 0. Hence we may assume that M a D 0 for a < 0. Then
s0 W I ! M 0
is injective and hence has a left inverse by applying (ii) to the exact sequence
s0
0 ! I ! M 0 ! Coker.s 0 / ! 0. This yields a left inverse of s.
“(iii) ) (i)”. Let t 0 W M ! N 0 be injective and uW M ! I be arbitrary. Let N be
the complex that is the canonical homomorphism N 0 ! Coker.i/ in degree 0 and 1 and
that vanishes in all other degrees. Then t 0 yields a quasi-isomorphism tW M ! N . By

Lemma 15.9 there exists in K.R/ a commutative diagram
t
M N
u u0
s
I P ;
Q N0 ! I
where s is a quasi-isomorphism. By (iii) there exists a left inverse r for s. Let uW
0 0
be the map given by r ı u in degree 0. We have r ı u ı t ' r ı s ı u ' u and hence
.r ı u0 / ı t D u because homotopic homomorphisms between complexes concentrated in
degree 0 are equal. In particular we find uQ ı t 0 D u.
Example 15.17.
1. If R D k is a field, then every k-vector space is injective because every short exact
sequence of k-vector spaces splits (by Example 14.6).
2. Let R be a principal ideal domain (e.g., R D Z). Then an R-module M is injective if
and only if a 2 R, a ¤ 0, the scalar multiplication M ! M , m 7! am, is surjective
([BouA3] §1.7, see also Problem 15.14).
For instance, let K D Frac R be the field of fractions of R. Then K and K=R are
injective R-modules.
Definition 15.18. A complex I of R-modules is called K-injective if for every complex

M and for every quasi-isomorphism uW I ! M there exists a left inverse in the cate-
gory K.R/ (in other words, there exists a morphism of complexes rW M ! I such that
r ı u ' idM ).
Proposition 15.19. Let I be a complex that is bounded below and such that I n is an
injective R-module for all n 2 Z. Then the complex I is K-injective.
Proof. If I is concentrated in degree 0, this is the implication “(i) ) (iii)” of Proposi-

tion 15.16. For the general case we refer to [Stacks] Tag 070J.
Lemma 15.20. Consider a diagram in K.R/

s
M I
qi s
u
qi s
N J ;
where the horizontal morphisms are quasi-isomorphisms and assume that J is K-

injective. Then there exists a unique morphism wW I ! J in K.R/ making the diagram
commutative.
15.4 Abelian Categories 325
Proof. Existence. By Lemma 15.9 we find a diagram
qi s
M I
u
N v
qi s
qi s
J t
K :
As J is K-injective, there exists a left inverse rW K ! J of t in K.R/ and we may set

w WD r ı v.
Uniqueness. For the uniqueness it suffices to show that if u ' 0 then one has neces-
sarily w ' 0. But u ' 0 implies w ı s ' 0 and hence there exists a quasi-isomorphism
tW J ! P such that t ı w ' 0 by Lemma 15.10. As J is K-injective, there exists
a left-inverse r of t. Hence w ' .r ı t/ ı w ' 0.
15.4 Abelian Categories
All of the above notions for R-modules can be generalized to so-called abelian categories.
These are categories that capture those properties of the category of R-modules that are
essential to define most notions in homological algebra: One has an abelian group struc-
ture on the set of all morphisms between two objects, there exist finite direct sums and
finite direct products, and for every morphism there exist kernels, cokernels (and hence
images and coimages) and a morphism always induces an isomorphism from its coimage
to its image (“fundamental homomorphism theorem”).
Definition 15.21 (Preadditive category). A preadditive category is a category A to-

gether with the structure of an abelian group on HomA .X; Y / for all objects X and Y
such that the composition HomA .X; Y / HomA .Y; Z/ ! HomA .X; Z/ is Z-bilinear
for all objects X, Y , and Z of A.
A functor F W A ! B between preadditive categories is called additive if the map
HomA .X; Y / ! HomB .F .X/; F .Y // given by F is a homomorphism of abelian groups
for all objects X and Y of A.
The category of R-modules (R a fixed ring) together with the structure of abelian
groups on the sets of homomorphisms defined by addition of R-linear maps is a preaddi-
tive category.
Definition and Remark 15.22 (Zero object). Let A be a preadditive category. An object
Z of A is called a zero object if it satisfies the following equivalent properties:
(i) Z is an initial object in A.

(ii) Z is a final object in A.
(iii) idZ D 0 2 HomA .Z; Z/.
Indeed, clearly (i) and (ii) both imply that HomA .Z; Z/ consists of a single element,
hence idZ D 0. Conversely, if (iii) holds, then for every morphism uW X ! Z (respec-
tively vW Z ! Y ) one has by bilinearity of composition that u D idZ ıu D 0 (respectively
v D v ı idZ D 0). Hence (ii) and (i) hold.
A zero object is unique up to unique isomorphism (if it exists) and is denoted by 0.
Condition (iii) shows that if F W A ! B is an additive functor between preadditive

categories, then F sends a zero object of A to a zero object of B.
Next we define kernels, cokernels, images, and coimages in preadditive categories.
Remark 15.23 (Kernel, cokernel, image, and coimage). Let A be a preadditive cate-
gory in which all finite limits and finite colimits exist, and let uW X ! Y be a morphism
in A. Consider the following diagram in A:
u
X Y:
0
Then its limit (respectively its colimit) is called the kernel of u (respectively cokernel of
u), denoted by Ker.u/ ! X (respectively Y ! Coker.u/).
Kernels have the following universal property: A morphism iW K ! X is a kernel of u
if and only if u ı i D 0 and if for any morphism i 0 W K 0 ! X with u ı i 0 D 0 there exists
a unique morphism vW K 0 ! K such that i 0 D i ı v.
Dually, a morphism pW Y ! X is a cokernel of u if and only if p ı u D 0 and if for
every morphism p 0 W Y ! C 0 with p 0 ı u D 0 there exists a unique morphism vW C ! C 0
with v ı p D p 0 .
The image of u is defined as Im.u/ WD Ker.Y ! Coker.u// and the coimage of u as
Coim.u/ WD Coker.Ker.u/ ! X/.
If A is the category of R-modules for some ring R, then these notions of kernel and
cokernel coincide with the usual notions by Example 14.2. Therefore the same holds for
images and coimages.
Remark 15.24. Let uW X ! Y be a morphism in a preadditive category A in which all

finite limits and finite colimits exist. Then there is a unique factorization of u
uN
X ! Coim.u/ ! Im.u/ ! Y: (15.4)
Indeed, as Ker.u/ ! X ! Y is zero, there exists a unique factorization X !

Coim.u/ ! Y of u. The composition Coim.u/ ! Y ! Coker.u/ is the unique mor-
15.4 Abelian Categories 327
phism giving rise to X ! X ! Coker.u/, which is zero. Hence Coim.u/ ! Coker.u/

is zero and there exists a unique factorization Coim.u/ ! Im.u/ ! Y of Coim.u/ ! Y .
Definition 15.25 (Abelian category). A preadditive category is called abelian if all fi-
nite limits and finite colimits exist and if for every morphism u the induced morphism
Coim.u/ ! Im.u/ is an isomorphism.
The hypothesis that all finite limits and colimits exist is highly redundant (see Prob-
lem 15.10).
We call a morphism uW X ! Y in an abelian category injective or X a subobject of Y
if Ker.u/ D 0 and write X Y . The morphism u is called surjective or Y a quotient of
X if Coker.u/ D 0. If X Y is a subobject, then we write Y =X WD Coker.X ! Y /.
Remark 15.26. Many notions like “(split) exact sequences” and results for R-modules
work verbatim in the same way in an arbitrary abelian category. This holds in particular
for all of the notions and results in Sects. 15.1–15.3. In particular we have the category
Com.A/ and the homotopy category K.A/ of complexes in A and the notions of injective
objects in A and of K-injective complexes in A. Note that all references given above for
results that we did not prove formulate their results in arbitrary abelian categories.
Recall that we defined in Definition 13.48 a functor to be left exact (respectively right
exact) if it commutes with finite limits (respectively finite colimits).
Remark 15.27 (Left exact functors). Let F W A ! B be an additive functor between

abelian categories. Then F is left exact (respectively right exact) if and only if for every
exact sequence 0 ! X 0 ! X ! X 00 (respectively X 0 ! X ! X 00 ! 0) in A the
sequence 0 ! F .X 0 / ! F .X/ ! F .X 00 / (respectively F .X 0 / ! F .X/ ! F .X 00 / !
0) is exact.
The easy proof is left to the reader (see also [KS2] Proposition 8.3.18).
One can show that every left or right exact functor is automatically additive (Prob-
lem 15.9).
Definition 15.28 (Existence of injective and K-injective resolutions). Let A be an

abelian category. We say that A has injective and K-injective resolutions if for every
complex M in A one has the following quasi-isomorphisms:
qi s
(a) There exists a quasi-isomorphism sW M ! I , where I is a K-injective complex.
This is called a K-injective resolution of M .
(b) If there exists a 2 Z such that M n D 0 for all n < a, then one can in addition assume
that I n is an injective object in A for all n 2 Z, that I n D 0 for all n < a, and that
s n W M n ! I n is injective for all n. This is called an injective resolution of M .
Every complex I as in (b) is K-injective by Proposition 15.19.

If A has injective and K-injective resolutions, then there exists in particular for every
object X of A an exact sequence
0 ! X ! I 0 ! I 1 ! : : : ;
where I p is an injective object, because such an exact sequence is the same as a quasi-
qi s
isomorphism X ! I with I p D 0 for p < 0.
15.5 Problems
Problem 15.1. Let A be an abelian category and let M be a complex of objects in

p
A with boundary maps dM W M p ! M pC1 . For k 2 Z define the k-shifted complex
p kCp
M Œk by M Œkp WD M kCp and dM Œk WD .1/k dM . If uW M ! N is a morphism
of complexes, uŒkW M Œk ! N Œk denotes the morphism of complexes with uŒkp D
ukCp .
1. Show that Œk defines a functor Com.A/ ! Com.A/ that induces a functor
ŒkW K.A/ ! K.A/.
2. Let M and N be complexes of objects in A and let uW M ! N be a morphism
of complexes. Show that a homotopy u ' u is the same as a morphism of complexes
M ! N Œ1 .
Problem 15.2. Prove Lemma 15.11.
Problem 15.3. Let X1 and X2 be objects in a preadditive category A. Show that the
product .pi W X1 X2 ! Xi /i exists if and only if the coproduct .sj W Xj ! X1 q X2 /j
exists. Show that in this case there is a unique morphism rW X1 q X2 ! X1 X2 with
pi ı r ı sj D 0 for i ¤ j , and pi ı r ı sj D idXi for i D j . Show that r is an isomorphism.
Write X1 ˚ X2 for X1 q X2 D X1 X2 . A preadditive category in which all finite
products exist is called an additive category.
Problem 15.4. Let A be a preadditive category, X and Y objects in A, u1 ; u2 2

HomA .X; Y / and assume that X ˚ X and Y ˚ Y exist (Problem 15.3). Show that u1 C u2
coincides with the composition
X u1 ˚u2 ˙Y
X ! X ˚ X ! Y ˚ Y ! Y;
where X is the diagonal and where ˙Y is the unique morphism whose composition with
the two component maps Y ! Y ˚ Y is the identity.
15.5 Problems 329
Problem 15.5. Consider a complex M of R-modules as a Z-diagram n 7! M n in the

category of R-modules. What is limn M n and colimn M n ?
Problem 15.6. Prove Lemma 15.12 and Lemma 15.13 for diagrams in arbitrary abelian
categories.
Problem 15.7. Let A be an additive category (Problem 15.3) and let F W A ! (Sets) be
a functor commuting with finite products. Show that there is a functor FQ W A ! (Ab) such
that F is isomorphic to the composition of FQ with the forgetful functor (Ab) ! (Sets).
Show that FQ is unique up to unique isomorphism.
Hint: Problem 15.4.
Problem 15.8. Prove Remark 15.27.
Problem 15.9. Let A and B be additive categories (Problem 15.3) and let F W A ! B be
a functor. Show that F is additive if and only if F commutes with finite products.
Hint: To see that the condition is sufficient apply Problem 15.7 for all objects X to the
functors Y 7! HomA .X; Y / and Y 7! HomB .F .X/; F .Y //.
Problem 15.10. Let A be an additive category (Problem 15.3) such that every morphism
u has a kernel and a cokernel and such that the induced morphism Coim.u/ ! Im.u/ is
an isomorphism. Show that A is an abelian category.
Problem 15.11. Let A be an abelian category. Show that every bounded below complex
of objects in A has an injective resolution if and only if for every object X of A there
exists an injective morphism X ! I , where I is an injective object of A.
Problem 15.12. Let A be an abelian category and let .I˛ /˛ be a family of objects of
Q Q
A such that the product ˛ I˛ exists. Show that ˛ I˛ is injective if and only if I˛ is
injective for all ˛.
u v
Problem 15.13. Let 0 ! M 0 ! M ! M 00 be an exact sequence in an abelian
category A, and let a0 W M 0 ! C 0 and a00 W M 00 ! C 00 be injective morphisms in A.
Suppose that C 0 is an injective object in A. Show that there exists an injective morphism
aW M ! C 0 ˚ C 00 such that the diagram
u v
M0 M M0
a0 a a00
i p
C0 C 0 ˚ C 00 C 00
commutes, where i.c 0 / D .c 0 ; 0/ and p.c 0 ; c 00 / D c 00 .

Problem 15.14. Let R be a ring.
1. Show that an R-left module I is injective if and only if for every left ideal a of R and
for every R-linear map uW a ! I there exists x 2 I such that u.a/ D ax for all a 2 a.
Deduce that an abelian group A is an injective Z-module if and only if A is divisible
(i.e., A ! A, a 7! na is surjective for all 0 ¤ n 2 Z).
2. Show that the R-left module IR WD HomZ .R; Q=Z/ is injective and a cogenerator of
the category of R-left modules (i.e., for every R-left module M and for every 0 ¤ m 2
M there exists an R-linear map uW M ! IR with u.m/ ¤ 0).
Hom .M;IR /
3. For every R-module M set I.M / WD IR R and let eW M ! I.M /, m 7!
.u.m//u2HomR .M;IR / . Show that I.M / is an injective R-left module and that e is an
injective R-linear map.
4. Deduce from Problem 15.11 that in the category of R-left modules every bounded
below complex of objects has an injective resolution.
Appendix E: Local Analysis
16
In this appendix we recall some notions on differentiable and analytic functions of open
subsets of finite-dimensional K-vector spaces, where K either denotes the field of real
numbers R or the field of complex numbers C. These will be the local building blocks in
the theory of manifolds. Most of the results are topics in any standard calculus class and
we will give no proofs.
16.1 Differentiable Functions
Recall that all norms on a finite-dimensional K-vector space V are equivalent and hence
that there is a unique topology on V that is induced by some norm. If not stated otherwise,
we will always endow all subsets of V with the induced topology. For K-vector spaces V
and W we denote by HomK .V; W / the K-vector space of K-linear maps V ! W .
We recall some basic notions and definitions about differentiable functions. Our main
reference is [La]. In the sequel let V and W be finite-dimensional K-vector spaces and let
U V be an open subset.
Definition 16.1 (Differentialbility). Let f W U ! W be a map.
1. The map f is called differentiable or K-differentiable if for all u 2 U there exists a

(necessarily unique) K-linear map Df .u/W V ! W such that
jjf .u C h/ f .u/ Df .u/.h/ jj

lim D 0;
V 3h!0 jjhjj
where we choose some norms on V and W (the notion of differentiability is indepen-

dent of the choice because all norms are equivalent). The map
Df W U ! HomK .V; W /
DOI 10.1007/978-3-658-10633-1_16
332 16 Appendix E: Local Analysis
is also independent of the choice of norms and is called the derivative of f . If we

want to stress whether we are in the case K D R or K D C, we use the notions “real
differentiable” and “complex differentiable”, respectively.
2. Let f W U ! W be differentiable. For v 2 V we call
f .u C tv/ f .u/
Dv f .u/ WD .Df .u//.v/ D lim
K3t !0 t
the partial derivative at u in direction v. The map
Dv f W U ! W
is then called the partial derivative in direction v.

3. For ˛ 2 N0 the notion of a C ˛ -map and its ˛-fold derivative D ˛ f is defined in-
ductively. The map f is called a C 0 -map or K-C 0 -map if f is continuous. We
set D 0 f WD f . For all ˛ 2 N the map f is called a C ˛ -map or K-C ˛ -map
if f is K-differentiable and Df W U ! HomK .V; W / is a K-C ˛1 -map. We set
D ˛ f WD D.D ˛1 f /.
If f is a C ˛ -map for all ˛ 2 N0 , then f is called a C 1 -map or a smooth map.
If we want to stress whether we are in the case K D R or K D C, we speak of real
C ˛ -maps and complex C ˛ -maps, respectively.
If .e1 ; : : : ; en / 2 V is a basis and .x 1 ; : : : ; x n / 2 V _ is the dual basis, we also set
@f
@i f WD @x i f WD WD Dei f W U ! W: (16.1)
@x i
@f
Note that this is a misuse of notation because @i f D @x i depends for a fixed i only on the
choice of ei but not only on x i (as e i is not determined by x i alone but by the whole basis
.x 1 ; : : : ; x n /). We usually use this notion only if V D Kn , where we always choose the
standard basis when we write @i f .
Remark 16.2. Let ˛ 2 N0 [ f1g.
1. If f W U ! W is differentiable, then f is continuous. If f is C ˛ , then f is also C ˇ

for all ˇ ˛.
2. The set of all C ˛ -functions U ! W is denoted by C ˛ .U; W /. It is a K-subspace of
the vector space of all maps U ! W and for f; g 2 C ˛ .U; W / and a 2 K one has
D.af C g/ D aDf C Dg:

16.1 Differentiable Functions 333
Remark 16.3. Formally the real case K D R and the complex case K D C are very simi-
lar. Let V and W be finite-dimensional C-vector spaces, U V open and let f W U ! W
be a map. Of course, we may consider V and W also as finite-dimensional R-vector
spaces. Then by definition f is complex differentiable if and only if f is real differen-
tiable and Df .u/ is C-linear for all u 2 U .
On the other hand, the behavior of real and complex differentiable functions is very
different – a topic that we will touch upon again and again. For instance, in the real case
there exist for all ˛ 2 N0 functions f˛ W R ! R that are C ˛ but not C ˛C1 , for instance
8
<x ˛C1 ; x < 0I
f˛ .x/ WD
:x ˛C1 ; x 0:
But from elementary complex analysis we know that for K D C every differentiable
map f W U ! C, U C open, is automatically a C 1 -map. In fact a much stronger
assertion holds, namely f is automatically analytic (i.e., locally given by a convergent
power series). We will make this more precise below.
Calculus (for instance by induction on [La] XIII, Theorem 7.1) tells us that real C ˛ can
also be defined via partial derivations:
Proposition 16.4. Let K D R, let ˛ 2 N, and let X be a generating system of the

R-vector space V (e.g., a basis or X D V ). Then a map f W U ! W is a (real) C ˛ -
map if and only if for every tuple .v1 ; : : : ; v˛ / 2 X ˛ the iterated partial derivatives
Dv1 Dv2 Dv˛ f W U ! W exist and are continuous.
In the complex case an analogous result is made superfluous by the fact that every
complex differentiable map is automatically analytic (see Theorem 16.15).
The next two propositions are well known from elementary calculus – at least for
K D R. We deduce the complex case from the real case by considering complex dif-
ferentiable maps as real differentiable maps whose derivation has complex linear values
(Remark 16.3).
Proposition 16.5 (Chain rule). Let V1 , V2 , and V3 be finite-dimensional K-vector spaces,

let U1 V1 and U2 V2 be open subsets and let f W U1 ! V2 and gW U2 ! V3 be maps
with f1 .U1 / U2 . If f and g are differentiable, then g ı f is differentiable and for
u1 2 U1 and u2 WD f .u1 / one has
D.g ı f /.u1 / D Dg.u2 / ı Df .u1 /:
If f and g are C ˛ -maps, then g ı f is a C ˛ -map.

Proposition 16.6. Let F W U ! W be a map. Then F is differentiable with DF .u/ D 0

for all u 2 U if and only if F is locally constant.
Example 16.7. Let V , V1 ; : : : ; Vr , W be finite-dimensional K-vector spaces.
1. It follows immediately from the definition of the derivative that every K-linear map
f W V ! W is differentiable and we have Df .u/ D f for all u 2 V , i.e., Df W V !
HomK .V; W / is the constant map with value f . In particular f is a C 1 -map with
D r f D 0 for all r 2 by Proposition 16.6.
2. More generally, every r-multilinear map ˛W V WD V1 Vr ! W is a C 1 -map.
For u D .u1 ; : : : ; ur / 2 V we have
X
r
.D˛.u//.v/ D ˛.u1 ; : : : ; ui 1 ; vi ; ui C1 ; : : : ; ur /
i D1
for all v D .v1 ; : : : ; vr / 2 V .

3. A homogeneous polynomial map pW V ! W of degree r 2 N0 is by definition the
composition of the linear map V ! V V ! V , v 7! .v; : : : ; v/ followed by an
r-multilinear map V V ! W . Hence any polynomial map V ! W (defined
to be a sum of homogeneous polynomial maps, not necessarily of the same degree) is
a C 1 -map.
Example 16.8. Let F W U ! W be a C ˛ -map and let LW W ! W 0 be a linear map

between finite-dimensional K-vector spaces. Then an easy induction and the chain rule
shows that D r .L ı F /.u/ D L ı D r .u/ for all u 2 U and r ˛.
16.2 Analytic Functions
We recall some facts on analytic functions, i.e., of maps that are locally given by conver-
gent power series. For proofs we refer to [Ser] Part II, Chap. II.
Definition 16.9 (Analytic function). Let U Kn be open, u D .u1 ; : : : ; un / 2 U .

A function f W U ! K is called analytic at u if there exists an open neighborhood UQ of u
in U such that f j UQ is given by an absolutely convergent power series
X
f .x/ D ai1 ;:::;in .x1 u1 /i1 : : : .xn un /in ; x 2 UQ ; (16.2)
.i1 ;:::;in /2N0n
with ai1 ;:::;in 2 K for all x 2 UQ . Such a power series is called power series expansion of
f at u.
The function f W U ! K is called analytic if it is analytic at every point u of U . A map
F W U ! Km is called analytic if each component is analytic.
16.2 Analytic Functions 335
Example 16.10. Every polynomial map is analytic, in particular every linear form Kn !
K is analytic. Therefore every linear map Kn ! Km is analytic.
For r 2 .R>0 /n and u 2 Kn we call
Pr .u/ WD f x 2 Kn I jxi ui j < ri for all i D 1; : : : ; ng
the polydisc of radius r around u. Its closure Pr .u/ consists of those x 2 Kn with
jxi ui j ri for all i.
Proposition 16.11. Let U Kn be open, u 2 U . Let f be a power series as in (16.2).

P
Let r 2 .R>0 /n such that .i1 ;:::;in /2N n jai1 ;:::;in jr1i1 : : : rnin converges.
0
1. The power series f converges uniformly on Pr 0 .u/ for all r 0 2 .R>0 /n with ri0 < ri
for all i and it converges absolutely in all points x 2 Pr .u/. In particular it defines
a continuous map fQW Pr .u/ ! K.
2. The map fQ is analytic and there is a unique power series expansion of fQ at u, namely
the one given by f .
3. The map fQ is differentiable and for all j D 1; : : : ; n the partial derivative @j fQ is
given by the absolutely convergent power series
X
@j fQ.x/ D ai1 ;:::;in ij .x1 u1 /i1 : : : .xij uij /ij 1 : : : .xn un /in : (16.3)
.i1 ;:::;in /2N0n
In particular @j fQ is again analytic.
Corollary 16.12. Let U Kn be open. Every analytic map F W U ! Km is a C 1 -

function.
Proposition 16.13. Let V Km be open and let f W U ! Km and gW V ! Kp be

analytic maps such that f .U / V . Then g ı f W U ! Kp is analytic.
As linear maps are analytic, we can define the notion of an analytic function also in
a coordinate-free setting.
Remark and Definition 16.14. Let V and W be finite-dimensional K-vector spaces,

U V open. Then a map f W U ! W is called analytic if for one choice (equivalently,
for all choices) of K-linear isomorphisms xW V ! Kn and yW W ! Km the composition
y ı f ı x 1 jx.U / W x.U / ! Km is analytic.
To see the mentioned equivalence suppose that y ı f ı x 1 jx.U / is analytic and let
Q V ! Kn be another K-linear isomorphism. As we have just remarked that composi-
xW
tions of analytic with linear maps are again analytic, we see that
y ı f ı xQ 1 j x.U 1
Q 1 j x.U
Q / D y ı f ı x jx.U / ı x jU ı x Q /
is also analytic. The same argument works if one replaces y by another linear isomor-
phism.
Every analytic map f W U ! W is a C 1 -map. The composition of analytic maps is
again analytic.
Sometimes we will also call analytic maps C ! -maps. Every C ! -map is also a C ˛ -map
for all ˛ 2 N0 [ f1g. To ease the notation we define
b 0 WD N0 [ f1; !g;
N b WD N [ f1; !g:
N
We extend the usual total order on N0 to N b 0 by requesting n < 1 < ! for all n 2 N0 .
We also define 1 ˙ ˛ WD 1 and ! ˙ ˛ WD ! for all ˛ 2 N0 .
Then every C ˛ -map is also a C ˇ -map for all ˛; ˇ 2 N b 0 with ˇ ˛. For K D R
ˇ ˛
there always exist functions that are C but not C : For ˛ < ! an example was given in
Remark 16.3. A standard example of a real C 1 -function that is not analytic is given in
Problem 16.1.
In the complex case, a map is C 1 if and only if it is C ! . More precisely:
Definition and Theorem 16.15. Let K D C. Then a map f W U ! W is called holomor-

phic if the following equivalent assertions hold:
(i) f is complex differentiable.

(ii) f is analytic.
(iii) There exists a basis .e1 ; : : : ; en / of V such that the partial complex derivatives Dei f
exist for all i D 1; : : : ; n.
(iv) For all ˛ 2 N and for all tuples .v1 ; : : : ; v˛ / of vi 2 V the iterated partial derivatives
Dv1 Dv2 Dv˛ f W U ! W exist.
For the proof we essentially refer to the literature.
Proof. The implications “(i) ) (iii)” and “(iv) ) (iii)” are trivial and the implications
“(ii) ) (i)” and “(ii) ) (iv)” hold (for arbitrary K) because analytic maps are C 1
(Corollary 16.12). For the proof of the (difficult) implication “(iii) ) (ii)” we refer to
[Hoe] Theorem 2.2.8.
Theorem 16.16 (Inverse function theorem). Let ˛ 2 N b and let f W U ! W be a

C -map. Let uQ 2 U and assume that Df .u/W
˛
Q V ! W is bijective. Then there exist
open neighborhoods U0 of uQ in U , and W0 of wQ WD f .u/ Q in W , such that the restric-
tion f jU0 W U0 ! W0 is bijective and f j U0 W W0 ! U0 is a C ˛ -map with D.f j 1
1
Q D
U0 /.w/
Q 1 .
Df .u/
16.3 Higher Derivatives 337
Proof. For K D R and ˛ < ! this is a standard result from calculus ([La] XIV Theo-
rem 1.2). For K D C we only have to prove the result for ˛ D 1 by Theorem 16.15. But
then it follows from the real case, that f j 1
U0 W W0 ! U0 exists and is a real C -map whose
1
1
differential is complex linear at every point. Hence it is a complex C -map (Remark 16.3).
Finally, for the case K D R and ˛ D ! we refer to [KrPa] Theorem 2.5.1.
16.3 Higher Derivatives
Let us take a closer look at higher derivatives. If f W U ! W is a C ˛ -map with ˛ 2,

then D.Df / is a map U ! HomK .V; HomK .V; W //. More generally its r-th derivative
(for 1 r ˛) is a map
U ! HomK .V; HomK .V; HomK .V; : : : ; HomK .V; W / : : :///
forming “HomK .V; /” r times. To ease the bookkeeping we first identify the right-hand
side with HomK .V ˝r ; W / as follows.
For K-vector spaces V1 , V2 , and W recall that by definition of the tensor product there
is an isomorphism of K-vector spaces, functorial in V1 , V2 , and W
HomK .V1 ; HomK .V2 ; W // ! f ˇW V1 V2 ! W I ˇ is K-bilinearg

(16.4)
D HomK .V1 ˝K V2 ; W /
given by attaching to the linear map 'W V1 ! HomK .V2 ; W / the linear map ˇ' W v1 ˝ v2 7!
.'.v1 //.v2 /. More generally, there is a functorial isomorphism
HomK .V1 ; HomK .V2 ; HomK .V3 ; : : : ; HomK .Vr ; W / : : :///

(16.5)
! HomK .V1 ˝K ˝K Vr ; W /
for K-vector spaces V1 ; : : : ; Vr and W by identifying an element ' of the left-hand side
with the K-linear map
v1 ˝ ˝ vr 7! : : : '.v1 / .v2 / : : : .vr / :
Finally recall that we write V ˝r WD V ˝K ˝K V , r factors of V .

Hence if f W U ! W is a C ˛ -map, we can (and will) consider for all r ˛ and u 2 U
the r-th derivative D r f .u/ 2 HomK .V; HomK .V; : : : ; HomK .V; W / : : :// as a K-linear
map
D r f .u/W V ˝r ! W: (16.6)
Q
Remark 16.17. Let W D m i D1 Wi for (finite-dimensional) K-vector spaces Wi . For
a map f W U ! W let fi W U ! Wi be the i-th component. Then f is a C ˛ -map if and
only if fi is a C ˛ -map for all i. In this case for r ˛ one has
Y
m
D r f .u/ D .D r f1 .u/; : : : ; D r fm .u//W V ˝r ! W D Wi :
i D1
In both the real and the complex case we can describe higher derivatives via partial
derivatives as follows.
Remark 16.18. Let f W U ! W be a C ˛ -map. Let .e1 ; : : : ; en / be a K-basis of V defining

the partial derivative @j f in direction ej for j D 1; : : : ; n (16.1). For r ˛ and u 2 U
we consider D r f .u/ as a map V ˝r ! W via (16.5).
Let us first consider the case V D Kn , W D Km and r D 1. Then Df .u/W Kn ! Km
is given by the Jacobian matrix
Jf .u/ WD .@j fi .u//1i m;1j n 2 Mmn .K/:
Now let 1 r ˛ be arbitrary. Then the ej1 ˝ ˝ ejr for 1 j1 ; : : : ; jr n form

a basis of V ˝r and one has
D r f .u/.ej1 ˝ ˝ ejr / D .@jr : : : @j1 f /.u/ 2 W: (16.7)
Indeed, for K D R this is a well-known result from calculus. It can be proved by induction
on r using the description of the first derivative by the Jacobian matrix. The same proof
works for K D C.
P
More generally, let v1 ; : : : ; vr 2 V arbitrary vectors and vd D riD0 ai;d ei with ai;d 2
K. By linearity we deduce from (16.7)
X
D r f .u/.v1 ˝ ˝ vr / D ai1 ;1 air ;r .@jr : : : @j1 f /.u/: (16.8)
.i1 ;:::;ir /
If W D Km and f1 ; : : : ; fm W U ! K are the components of f , then .@jr : : : @j1 fi /.u/,

i D 1; : : : ; m, are the components of D r f .u/.ej1 ˝ ˝ ejr /.
Proposition 16.19. Let V and W be finite-dimensional K-vector spaces, let U V be

open and let f W U ! W be a C ˛ -map. For r ˛ consider the r-th derivative as a map
D r f W U ! HomK .V ˝r ; W /. Then for every permutation 2 Sr one has
Df .u/.v .1/ ˝ ˝ v .r/ / D Df .u/.v1 ˝ ˝ vr /
for all u 2 U , v1 ; : : : ; vr 2 V . In other words D r f .u/ induces a map
D r f .u/W Symr .V / ! W:
16.3 Higher Derivatives 339
Proof. In the case K D R this is a standard fact from calculus (e.g., [La] XIII Theo-
rem 6.2). In the complex case this follows from the fact that F is automatically analytic
and from the concrete description of the derivative of a power series (16.3).
Every C ˛ -function can be approximated via its Taylor expansion by polynomials. For
analytic functions (and hence for all differentiable functions in the case K D C) this is
clear by definition. Hence we focus on K D R.
Proposition 16.20. Suppose that K D R. Let f W U ! W be a C ˛ -map, u 2 U and

2 R>0 such that B .u/ U for some choice of a norm jj jj on V . Let v 2 V such
that jjvjj < . Write v ˝r for v ˝ ˝ v 2 V ˝r . Then one has for all r ˛ the Taylor
expansion
X r
D m f .u/.v ˝m /
f .u C v/ D C Rr;u .v/;
mD0
mŠ
with
Rr;u .y/
lim r D0
y!0 jjyjj
locally uniformly in u.
Proof. This is again a standard fact from calculus (see [La] XIII Theorem 6.3, in particular
the “Estimate of the Remainder” there).
Remark 16.21. Let f W U ! W be a (real or complex) C ˛ -map, u 2 U . Choose a basis

.e1 ; : : : ; en / of V defining partial derivatives @i D @x@ i . The summands in the Taylor
expansion have the following coordinate version (valid both for K D R and K D C). We
P
write v D niD1 vi ei with vi 2 K. Then
D r f .u/.v ˝r / X 1 @r f
D v1i1 vnin i1 .u/; (16.9)
rŠ i Š in Š
i CCi Dr 1 @x1 @xnin
1 n
where the sum is taken over all .i1 ; : : : ; in / 2 N0n whose sum is r. This follows from
(16.8) and because
rŠ
i1 Š : : : in Š
is the number of decompositions of a set E with r elements into a tuple .E1 ; : : : ; En / of
disjoint subsets with #Ej D ij .
Now suppose that f W U ! W is a real C 1 -map, let 2 R>0 such that B .u/ U for
P
some norm jj jj on V and let v D niD1 vi ei 2 V with jjvjj < . Then
X1 X
D m f .u/.v ˝m / @i11 @inn f .u/ i1
D v1 vnin (16.10)
mŠ i1 Š in Š
mD0 .i1 ;:::;in /2N0n
is called the Taylor series of f at u.

Remark 16.22. For an analytic map f W U ! W its power series expansion in some point
u 2 U is its Taylor series in u ([Ser] Part II, Chap. II p.73).
16.4 Problems
Problem 16.1. Define f W R ! R by f .x/ WD 0 for x 0 and by f .x/ WD exp.1=x/

for x > 0. Show that all derivatives in 0 exist and are equal to zero. Deduce that f is C 1
but not analytic.
Problem 16.2. Let V and W be finite-dimensional K-vector spaces and let r 2 N. Show
that the maps
T r W HomK .V; W / ! HomK .T r .V /; T r .W //; u 7! T r .u/

r W HomK .V; W / ! HomK .r .V /; r .W //; u 7! r .u/
Symr W HomK .V; W / ! HomK .Symr .V /; Symr .W //; u 7! Symr .u/
are homogeneous polynomial maps of degree r and deduce that these maps are analytic.
Problem 16.3. Let U Kn be open and let f W U ! K be analytic. Show that

f x 2 U I D m f .x/ D 0 for all m 0g is open and closed in U . Deduce that the follow-
ing assertions are equivalent if U is connected:
(i) There exists a point x 2 U such that @i11 @inn f .x/ D 0 for all .i1 ; : : : ; in / 2 N0n .
(ii) There exists a non-empty open subset W of U such that f jW D 0.
(iii) f D 0.
Problem 16.4. Let V be a finite-dimensional C-vector space, let U V be open and

connected, and let f W U ! C be a non-constant holomorphic map. Show that f is open.
Hint: Deduce this from the classical result for V D C.
References
[AJS] Alonso Tarrío L., López A.J., Souto Salorio M.J., Localization in Categories of Complexes
and Unbounded Resolutions, Canad. J. of Math. 52 (2000), 225–247.
[AmEs3] Amann H., Escher J.: Analysis III. Birkhäuser (2009)
[BouA1] Bourbaki N.: Algebra, Chaps. 1–3. Springer (1989)
[BouA2] Bourbaki N., Algebra, Chaps. 4–7. Springer (2003)
[BouA3] Bourbaki N.: Algèbre, Chap. 10. Springer (2006)
[BouGT1] Bourbaki N.: General topology, Chaps. 1–4, 2nd printing. Springer (1989)
[BouGT2] Bourbaki N.: General topology, Chaps. 5–10, 2nd printing. Springer (1989)
[BouLie1] Bourbaki N.: Lie Groups and Lie Algebras, Chaps. 1–3. Springer (1989)
[BouTS] Bourbake N.: Theory of Sets. Springer (2008)
[Br1] Bredon G.E.: Topology and Geometry. Springer (1993)
[Br2] Bredon G.E.: Sheaf Theory, 2nd edition. Springer (1997)
[Car] Cartan H.: Variétés analytiques réelles et variétés analytiques complexes. Bull. de la SMF 85,
77–99 (1957)
[GeMa] Gelfand S.I., Manin Yu.I.: Methods of Homological Algebra. Springer (1996)
[God] Godement R.: Topologie algébrique et théorie des faisceaux. Hermann (1958)
[Gra] Grauert H.: On Levi’s problem and the imbedding of real analytic manifolds. Ann. of Math.
68, 460–472 (1958)
[GuRo] Gunning R., Rossi H.: Analytic functions of several complex variables. Prentice-Hall
(1965)
[HiNe] Hilgert J., Neeb K.-H., Structure and Geometry of Lie Groups. Springer (2012)
[Hoe] Hörmander L.: An Introduction to Complex Analysis in Several Variables. North-Holland
(1990)
[Ive] Iversen B.: Cohomology of Sheaves. Springer (1968)
[Ker] Kervaire M.A.: A manifold which does not admit any differentiable structure. Comm. Math.
Helvet. 34, 257–270 (1960)
[KoPu] Koch W., Puppe D.: Differenzierbare Strukturen auf Mannigfaltigkeiten ohne abzählbare
Basis. Arch. Math. (Basel) 19, 95–102 (1968)
[KrPa] Krantz S.G., Parks H.R.: A Primer of Real Analytic Functions, 2nd edition. Birkhäuser
(2002)

DOI 10.1007/978-3-658-10633-1
342 References
[KS1] Kashiwara M., Schapira P.: Sheaves on Manifolds. Springer (1990)

[KS2] Kashiwara M., Schapira P.: Categories and Sheaves. Springer (2006)
[La] Lang S.: Real and Functional Analysis, 3rd edition. Springer (1993)
[LeeJe] Jeffrey Lee M.: Manifolds and Differential Geometry. Graduate Stud. in Math. 107, AMS
[LeeJo] John Lee M.: Introduction to Smooth Manifolds. Graduate Texts in Math. 218, Springer
(2006)
[Mil] Milnor J.: On manifolds homeomorphic to the 7-sphere. Ann. of Math. 64, 399–405 (1956)
[Sch] Schubert H.: Kategorien I. Akademie-Verlag Berlin (1970)
[Ser] Serre J.-P.: Lie Algebras and Lie Groups. Lect. Notes in Math. 1500, Springer (1992)
[SGA4] Grothendieck A., Artin M., et. al.: Séminaire de Géometrie Algébrique du Bois-Marie,
Théorie des Topos et cohomologie étale des schémas (1963–64). Lect. Notes in Math. 269,
270, 305, Springer (1972/73)
[Spa] Spaltenstein N.: Resolutions of unbounded complexes, Comp. Math. 65, 121–154 (1988)
[Stacks] de Jong J., et al.: The Stacks Project. http://stacks.math.columbia.edu
[Ste] Sternberg S.: Lectures on Differential Geometry. Prentice-Hall Mathematics Series (1964)
[tDi] tom Dieck T.: Algebraic topology. EMS Textbook in Mathematics (2008)
[Wel] Wells R.O.: Differential Analysis on Complex manifolds, 3rd edition. Springer
[Whi1] Whitney H.: Geometric Integration Theory. Princeton University Press (1957)
[Whi2] Whitney H.: Differentiable manifolds. Ann. of Math. 37, 645–680 (1936)
Index
(Ab), category of abelian groups, 273 A , units in a ring, 308

(Com(R)), category of complexes, 318 AutB .Z/, sheaf of automorphisms of
(Cov(X )), category of open coverings, 143 B-manifold Z, 154
(CovSp(X )), category of covering spaces of X , Aut.p/, sheaf of automorphisms of a morphism,
29 154
(Ét/X ), category of étalé spaces, 53 abelian category, 327
.FIBF;G .B//, category of fiber bundles, 159 abelian sheaf, 44
.FLF.X //, category of finite locally free action
OX -modules, 171 of a group on a set, 126
(Grp), category of groups, 273 faithful, 126
(G-Sets), category of sets with left G-action, free, 126
273 simply transitive, 126
(h-Top), homotopy category of topological transitive, 126
spaces, 22 of a sheaf of groups, 139
(k-Vec), category of k-vector spaces, 273 free, 149
(Mfd), category of manifolds, 77 transitive, 149
.OpenX /, category of open subsets, 42 proper, 128
(OX -Mod), category of OX -modules, 168 acyclic, 207
(PMfd), category of premanifolds, 77 additive category, 328
(PSh(X )), category of presheaves, 42 additive functor, 325
(Ring), category of rings, 273 Alexandroff compactification, 268
(R-Mod), category of left R-Modules, 273 algebra, 298
(Sets), category of sets, 272 commutative, 298
(Sets-G), category of sets with right G-action, graded, 303
273 alternating multilinear map, 304
.Sh.X //, category of sheaves, 44 analytic map, 334, 335
(Top), category of topological spaces, 273 anti-equivalence of categories, 275
.Tors.A//, category of A-torsors, 140 associated sheaf, 51
.Vec.B//, category of vector bundles, 167 atlas, 79
Œf , homotopy class of a continuous map, 22 equivalent, 80
Œn D f0; ; : : : ; ng, 233 G-bundle, 159
ŒX; Y , set of homotopy classes of continuous twisting, 154
maps X ! Y , 22 automorphism, 272
A B
aM , a ideal, M module, 308 Br .x0 /, closed ball, xi, 246
343
344 Index
Br .x0 /, open ball, xi, 246 filtered, 282

Baire space, 18 finite, 272
base change of module, 309 finitely cocomplete, 281
basis finitely complete, 281
of topology, 246 of vector bundles, 167
basis of a module, 295 opposite, 273
basis of OX -module, 171 preadditive, 325
Betti cohomology, 235 product, 273
Betti number, 243 small, 271
biholomorphic map, 76 category of objects over a base object, 154
bijective morphism of sheaves, 49 Čech 1-cocycle, 142
bilinear form of OX -modules, 187 cohomologous, 142
bilinear map, 297 Čech cohomology, 143
boundary, 249 on an open covering, 143
bounded below complex, 318 Čech cohomology class, 143
bundle center of a ring, 292
Möbius, 202 chain complex, 320
G-bundle atlas, 159 chain rule, 94
G-bundle structure, 159 change of charts, 79
chaotic order, 277
chart, 76
C
chart of a morphism, 80
C.Y /, cone of topological space, 237
Chart Principle, 99
C=B , category of B-objects, 154
closed covering of subspace, 254
C ˛ -function on subspace, 105
closed form, 183
C ˛ -map, 332
closed map, 248
CX˛ IW , sheaf of C ˛ -maps, 43
closed path, 23
C ˛ -map, 81
closed subset, 245
C ˛ -vector bundle, 165
closure, 249
C, complex numbers, xi
coarser topology, 245
Cent.R/, center of ring, 292
cochain complex, 320
Coim.w/, coimage of homomorphism of
cocomplete category, 281
OX -modules, 170, 292
1-cocycle
Coker.u/, 292
Čech, 142
Coker.f; g/, equalizer in a category, 287
cocycle condition, 74, 252
Coker.w/, cokernel of homomorphism of
coequalizer, 287
OX -modules, 170
cofinite topology, 266
colimI X , 280
cogenerator, 330
C opp , 273
` cohomologous, 142
i2I Ui , sum of ringed spaces, 75
cohomology class
[, Yoneda product, 232
Čech, 143
C ! -map, 336
cohomology of a complex, 318
CX IY , sheaf of continuous maps, 43
cohomology of a complex of sheaves, 209
cartesian square, 225
cohomology of a sheaf, 209
category, 271
cohomology set, 140
abelian, 327
cohomology with compact support, 229
additive, 328
coimage, 326
cocomplete, 281
coimage of homomorphism of OX -modules,
complete, 281
170
connected, 287
Index 345
coimage of linear map, 292 covering space, 29

cokernel, 326 c-soft, 230
cokernel of homomorphism of OX -modules, [, cup product, 232
170 cup product, 232
cokernel of linear map, 292
colimit, 280
D
colimit of OX -modules, 170
@Y , boundary, 249
commutative algebra, 298
@i , partial derivation, 96
commutative monoid, xi
ı, connecting homomorphism of cohomology,
compact space, 260
210
compactification, 268
.˛/, morphism of standard simplices, 234
complete category, 281
ıij , Kronecker delta, xi
complex, 317
n D Œn, standard simplex, 233
bounded below, 318
X , diagonal of topological space, 256
K-injective, 324
DerK .A; M /, derivations, 95
complex case, 76
det.F/, determinant of OX -module, 180
complex differentiable, 332
Df , derivative, 331
complex Lie groups, 123
dF .p/, derivative, 93
complex manifold, 76
f , diagonal of a morphism, 285
complex of modules, 296, 317
f , diagonal of a continuous map, 254
complex premanifold, 76
dim.M /, dimension of a premanifold, 78
complex type, 76
dimp .M /, dimension of a premanifold in
complex vector bundle, 165
L a point p, 78
complexification, 124
composition, 272 Li2I Fi , direct sum of OX -modules, 170
i Mi , direct sum of modules, 293
concatenation of paths, 23
din , i -th face of standard n-simplex, 234
cone of topological space, 237
d!, exterior derivation of differential form, 182
connected category, 287
D r f , higher derivative, 337
connected component, 259
Der K .OX /, sheaf of derivations, 179
connected space, 257
Dv f , partial derivative, 332
connecting homomorphisms, 210
Darboux’ theorem, 188
connecting map, 148
De Rham cohomology, 219
constant functor, 287
De Rham complex, 183
constant map, 259
degree of covering map, 29
constant presheaf, 43
dense, 250
constant sheaf, 52
derivation, 95
continuous, 247
derivative, 93, 332
contracted product of G-sheaves, 150
partial, 332
contractible space, 22
determinant of OX -module, 180
contravariant functor, 274
ı-functor, 206
coproduct in a category, 283
universal, 207
coproduct of ringed spaces, 75
diagonal of a morphism, 285
coproduct of topological spaces, 252
diagonal of continuous map, 254
countable set, xi, 1
diagram, 280
covariant functor, 274
diffeomorphism, 76
covering map, 29
differentiable map, 331
covering of subspace, 254
differential form, 181
closed, 254
dimension of premanifold, 78
open, 254
direct image of presheaf, 57
346 Index
direct image of topology, 250 exterior power of OX -module, 180

direct sum in a category, 283
direct sum of modules, 293
F
direct sum of OX -modules, 170
F.x/, fiber of an OX -module, 168
direct sum of vector bundles, 175
f , direct image functor, 57
direct summand of OX -module, 175
f C , inverse image functor as presheaf, 58
discrete order, 277
f 1 , inverse image functor, 58, 222
discrete topology, 245
f 1 , pullback of sections, 62
divisible abelian group, 330
.f; f [ /, morphism of ringed spaces, 71
divisor, 239
f [ , underlying sheaf homomorphism of f , 71
divisor on C, 191
FibF;G .B/, isomorphism classes of fiber
dual of a module, 298
bundles, 159
dual of OX -module, 172
X S Y , fiber product of topological spaces,
dual of vector bundle, 175
253
X1 Y X2 , fiber product in a category, 284
E Flagn .E/, flag manifold, 132
.E _ ; _ /, dual vector bundle, 175 F ˝OX G, tensor product of OX -modules, 186
ej , standard basis vector, 294 f ] , underlying sheaf homomorphism of f , 71
.E ˚ E 0 ; ˚ 0 /, direct sum of vector bundle, fŠ F, proper direct image, 67
175 Q sheafification, 51
F,
EX , constant sheaf, 52 F _ , dual of an OX -module, 172
p
ExtA .M ; N /, Ext group, 231 Fx , stalk of a sheaf, 46
embedding, 104 F jY , restriction of a sheaf to a subspace, 58
open, 73 F.Y /, evaluation of a sheaf in a subspace, 58
topological, 250 face of simplex, 234
closed, 250 faithful functor, 275
open, 250 faithful group action, 126
end point of path, 23 fiber bundle, 159
endomorphism, 272 fiber of OX -module, 168
epimorphism, 273 fiber product in a category, 284
equalizer, 287 fiber product of premanifolds, 112
equivalence of categories, 275 fiber product of topological spaces, 253
equivalent atlases, 80 field, xi
equivalent twisting atlases, 155 field of fractions, 312
equivariant, 126 filtered category, 282
essentially surjective functor, 275 filtered preorder, 276
étalé space, 53 final object, 284
Euler characteristic, 244 fine sheaf, 203
exact form, 183 finer topology, 245
exact sequence finite category, 272
of pointed sets, 147 finite locally free, 171
of sheaves of groups, 49 finitely cocomplete category, 281
exact sequence of groups, 49 finitely complete category, 281
exact sequence of modules, 296 finitely generated module, 292
exact sequence of OX -modules, 170 first countable space, 1
exact sequence of sheaves of groups, 49 flabby, 215
Ext group, 231 flasque, 215
exterior algebra, 305 flat module, 313
exterior power, 304 form, 181
Index 347
free action of a sheaf of groups, 149 Gluing of topological spaces, 252

free group, 290 Godement resolution, 216
free group action, 126 graded algebra, 303
free module, 295 graded commutative, 305
free product of groups, 290 graph morphism, 104
full subcategory, 272 graph of continuous map, 254
fully faithful functor, 275 Grassmann manifold, 132
function on subspace, 105 greatest element, 277
functor, 274 Grothendieck group, 290
additive, 325 group
anti-equivalence of categories, 275 topological, 262
commuting with colimit, 285 groupoid, 287
commuting with limit, 285
constant, 287 H
contravariant, 274 H , inverse homotopy, 21
covariant, 274 H K, product homotopy, 21
equivalence of categories, 275 H 1 .X; /, 140
essentially surjective, 275 HL 1 .X; /, Čech cohomology, 143
faithful, 275 H 1 .U; /, 141
fully faithful, 275 HL 1 .U; /, Čech cohomology, 143
left adjoint, 276 H, upper half plane, 133
left exact, 286 H i .M /, cohomology of a complex, 318
quasi-inverse, 275 H n .X; F /, cohomology, 209
right adjoint, 276 Hcn .X; F /, cohomology with compact support,
right exact, 286 229
functorial, 274 HDRn
.M /, De Rham cohomology, 219
fundamental group, 26 sing
Hn .X /, singular homology, 235
fundamental groupoid, 37 n
Hsing .X; G/, singular cohomology, 235
fundamental system of neighborhoods, 246
HomC .X; Y /, morphisms in a category, 272
', homotopy of continuous maps, 21
G ', homotopy of morphism of complexes, 318
G=F, quotient of OX -modules, 170 HomOX .F; F 0 /, homomorphism of
G=X , set of G-orbits, 126 OX -modules, 169
-acyclic, 207 HomOX .F; G/, sheaf of homomorphisms of
f g, image of a path , 23 OX -modules, 171
, inverse path, 23 HomR .M; N /, R-linear homomorphisms, 291
p
ı, concatenation of paths, 23 HA .X; A /, local cohomology group with
hS iR , R-module generated by a set, 292 support in A, 231
GLn;B , sheaf of invertible matrices, 157 f ' g .rel A/, homotopy relative to subspace,
h , graph of a continuous map, 254 25
sing
Grassd .E/, Grassmann manifold, 132 Hn .X; R/, singular homology, 236
Gx , stabilizer, 126 hairy ball theorem, 244
Gx, G-orbit, 126 Hausdorff space, 256
generating set of submodule, 292 hereditarily Lindelöf space, 17
generating system, 295 hereditarily paracompact space, 6
generating system of OX -module, 171 higher direct image with compact support, 229
generators of a group, 290 holomorphic form, 181
germ, 46 holomorphic function on subspace, 105
gluing datum for ringed spaces, 74 holomorphic map, 76, 336
348 Index
holomorphic vector bundle, 165 inductive limit, 280

homeomorphism, 247 inductive system, 281
local, 256 initial object, 284
homogeneous polynomial map, 334 initial subpremanifold, 121
homomorphism injective module, 323
of algebras, 298 injective morphism in an abelian category, 327
of Lie algebras, 190 injective morphism of sheaves, 49
of Lie groups, 123 injective resolution, 327
of modules, 291 integral domain, 312
of modules over a sheaf of rings, 167 integral monoid, 203
of monoids, xi interior, 249
of premodules over a sheaf of rings, 167 interval, 258
homotopic inverse homotopy, 21
morphisms of complexes, 318 inverse image of presheaf, 58
homotopic continuous maps, 21 inverse image of topology, 250
homotopy inverse path, 23
relative to a subspace, 25 isomorphism, 272
homotopy category of complexes, 319 of fiber bundles, 159
homotopy category of topological spaces, 22 of twists, 155
homotopy class of a continuous map, 22
homotopy equivalence, 22 J
homotopy of continuous maps, 21 Jacobi identity, 190
homotopy of morphisms of complexes, 318 Jacobian matrix, 338
homotopy of paths, 25
hypercohomology of a complex of sheaves, 210 K
K.R/, 319
Ker.f; g/, coequalizer in a category, 287
I Ker.w/, kernel of homomorphism of
Ii , I<i , Ii , I>i , 277 OX -modules, 169
Im.u/, 292 K, real or complex numbers, xi
Im.w/, image of homomorphism of .x/, residue field at a point x, 70
OX -modules, 170 Ker.u/, 292
IS , ideal of functions vanishing on S , 105 kernel, 326
Š, isomorphic, 272 kernel of homomorphism of OX -modules, 169
!, isomorphism, 272 Kernel of homomorphism of vector bundles,
ideal, 308 191
maximal, 308 kernel of linear map, 292
ideal of OX , 168 K-injective complex, 324
idempotent, 88 K-injective resolution, 327
indecomposable, 89 Klein bottle, 131
identity, 272 Kolmogorov space, 265
identity component, 264 K-projective, 214
image, 326
image of homomorphism of OX -modules, 170 L
image of linear map, 292 LA .N1 ; : : : ; Nr I M /, multilinear maps, 297
immersed subpremanifold, 120 rOX F, exterior power of OX -module, 180
immersion, 100 Lie.G/, Lie algebra of a Lie group, 190
increasing map, 286 limI X , 280
indecomposable idempotent, 89 A .M /, exterior algebra, 305
induced topology on subset, 250 Ln F , left derived functor, 214
Index 349
Ar .M /, exterior power, 304 M

LX , Lie derivative, 179 M=R, quotient premanifold, 115
Lagrangian, 137 MapX IE , sheaf of maps, 43
left adjoint functor, 276 M , complex, 317
left exact functor, 286 M _ , dual of a module, 298
left module, 291 M I , M module, 294
left order topology, 286 M .I / , M module, 294
Lemma M n , M module, 294
Nakayama, 315 Mnm .R/, matrices, 295
Lie algebra, 190 Mn .R/, quadratic matrices, 295
Lie bracket, 190 M S N , fiber product of premanifolds, 112
Lie derivative, 179, 192 M ˝A N , tensor product, 299
Lie group, 123 mx , maximal ideal of germs vanishing at x, 47,
Lie subgroup, 125 70
lifting, 30 M N , product of premanifolds, 85
limit, 280 manifold, 76
limit of OX -modules, 169 over a base manifold, 154
Lindelöf space, 1 projective, 111
line bundle, 165 Stein, 111
linear map, 291 with Lie group action, 126
linearly independent tuple, 295 matrix of a linear map, 295
local cohomology, 231 maximal element, 277
local diffeomorphism, 76 maximal ideal, 308
local direct summand of OX -module, 175 Mayer–Vietoris sequence, 228
local homeomorphism, 256 meager subspace, 18
local isomorphism of ringed spaces, 73 metrizable topological space, 246
local on source, 254 minimal element, 277
local on target, 254 Mittag-Leffler theorem, 221
local ring, 310 Möbius band, 86
local ring homomorphism, 71 Möbius bundle, 202
local system, 67 Möbius transformation, 133
locally biholomorphic map, 76 module, 291
locally closed, 256 dual, 298
locally compact space, 260 finitely generated, 292
locally connected space, 267 flat, 313
locally constant map, 259 free, 295
locally constant rank, 101 injective, 323
locally constant sheaf, 56 projective, 312
locally contractible space, 22 quotient, 292
locally finite, 255 torsion free, 312
locally free module over sheaf of rings, 167
finite, 171 monoid, xi
locally path connected space, 24 commutative, xi
locally ringed space, 70 integral, 203
long ray, 89 monoid homomorphism, xi
loop, 23 monomorphism, 273
lower bound, 277 morphism, 272
of algebras, 298
of complex premanifolds, 76
350 Index
of complexes, 317 OS sheaf of functions on subspace, 105

of covering spaces, 29 OX D OXhol , sheaf of holomorphic functions, 43,
of diagrams, 280 70
of étalé spaces, 53 OX -module of finite presentation, 185
of fiber bundles, 159, 161 OX -module of finite type, 185
of functors, 274 objects of a category, 272
of G-sets, 126 of finite presentation, 185
of Lie algebras, 190 of finite type, 185
of Lie groups, 123 one-point compactification, 268
of locally constant rank, 101 open covering of subspace, 254
of locally ringed spaces, 71 open embedding, 73
of manifolds, 76 open map, 248
of modules, 291 open refinement, 4
of modules over a sheaf of rings, 167 open subset, 245
of pointed covering spaces, 29 of metric space, 246
of pointed sets, 140 open subspace of a ringed space, 73
of premanifolds, 76 opposite category, 273
of premanifolds with Lie group action, 126 orbit, 126
of premodules over a sheaf of rings, 167 order, 276
of presheaves, 42 chaotic, 277
of ringed spaces, 71 discrete, 277
of sheaves, 44 order topology, 267
of sheaves with action of a sheaf of group,
140
P
of twists, 155
p-adic absolute value, 264
of vector bundles, 167
p-adic metric, 264
multilinear map, 297
p-adic topology, 264
˘.f /, fundamental groupoid functor, 37
N 0 .X /, set of path components, 24
N, natural numbers, xi 1 .X; x/, fundamental group, 26
N0 , non-negative integers, xi Pic.B/, Picard group, 165
b, completed natural numbers, xi
N ˘.X /, fundamental groupoid, 37
b0 , completed non-negative integers, xi
N P n .K/, projective n-space, 83
Nakayama’s lemma, 315 P .V /, projective space, 130
neighborhood, 246 Pr .u/, polydisc, 335
fundamental system, 246 PrincG .B/, isomorphism classes of principal
noetherian induction, 278 bundles, 162
noetherian order, 278 C D, product of categories, 273
Q
normal space, 9
Qi Mi , product of modules, 293
nowhere dense, 269 i2I Xi , product of topological spaces, 251
null homotopic continuous map, 22 Pr .u/, closed polydisc, 335
X1 qY X2 , pushout in a category, 285
O pairing of OX -modules, 187
Ob.C /, objects of a category C , 272 paracompact space, 4
OBIM , sheaf of morphisms, 157 partial derivative, 332
OB , sheaf of nowhere vanishing functions, 157 partial order, 276
˝M , de Rham complex, 183 partial preorder, 276
i
˝M , sheaf of i -forms, 181 particular point topology, 265
O.p; q/, orthogonal group, 127 partition of a section, 197
Index 351
path, 23 projective system, 281

closed, 23 proper action, 128
path component, 24 proper direct image, 67
path connected space, 24 proper map, 15
perfect pairing, 187 pseudotorsor, 140
Picard group, 165 pullback
piece-wise affine path, 37 of OX -modules, 189
Plücker embedding, 136 pullback map for sections, 62
pointed covering space, 29 pullback of fiber bundle, 161
pointed set, 140 pullback of twists, 157
polydisc, 335 pushout in a category, 285
polynomial map, 334
power series expansion, 334 Q
preadditive category, 325 qis, quasi-isomorphism, 318
premanifold, 75, 76 qis, 318
complex, 76 Q, rational numbers, xi
over a base premanifold, 154 quasi-inverse functor, 275
real analytic, 76 quasi-isomorphism, 318
smooth, 76 quotient bundle of vector bundle, 175
with Lie group action, 126 quotient module, 292
premodule over sheaf of rings, 167 quotient object, 327
preorder, 276 quotient of OX -modules, 170
presheaf, 41 quotient of vector bundle, 191
constant, 43 quotient premanifold, 115
of algebras, 42 quotient space, 251
of functions, 43 quotient topology, 251
of groups, 42
R
of rings, 42
rkA .M /, rank of a module, 309
on a basis, 45
rkp .F /, rank of morphism, 100
with values in a category, 42
rkx .F/, rank of finite locally free module, 171
principal bundle, 162
Rn F , right derived functor, 214
principle of unique continuation, 66
R, real numbers, xi
product
rank of module, 309
of ideal and module, 308
rank of morphism of premanifolds, 100
product category, 273
rank of OX -module, 171
product homotopy, 21
rank of vector bundle, 165, 167
product in a category, 283
real analytic form, 181
product of modules, 293
real analytic map, 76
product of OX -modules, 169
real analytic premanifold, 76
product of paths, 23
real C ˛ -map, 82
product topological space, 251
real case, 76
product topology, 251
real differentiable, 332
projection of a product, 283
real Lie group, 123
projection of fiber product, 284
real manifold, 76
projective, 214
real type, 76
projective limit, 280
refinement of a covering, 4
projective manifold, 111
refinement of twisting atlas, 155
projective module, 312
regular equivalence relation, 115
projective space, 83
regular space, 17
352 Index
regular submanifold, 106 fine, 203

regular value, 109 flabby, 215
relations in a group, 290 flasque, 215
relatively compact subset, 268 locally constant, 56
relatively Hausdorff subspace, 12 of algebras, 44
residue field, 310 of C ˛ -functions, 76
restriction map, 42 of groups, 44
right adjoint functor, 276 of modules, 44
right derived functor, 207, 214 of rings, 44
right exact functor, 286 skyscraper, 65
right order topology, 286 soft, 195
ring, xi structure, 70
local, 310 with action of a sheaf of group, 139
ring homomorphism sheaf on a basis, 45
local, 71 sheafification, 51
ringed space, 70 shift of complex, 328
locally, 70 short exact sequence, 296
simplex, 234
simply connected space, 27
S
simply transitive group action, 126
-compact space, 2
singleton, 284
˙n .X /, set of n-simplices, 234
singular chain, 234
SLn .K/, special linear group, 132
singular cochain, 204, 234
S n , n-sphere, 110
singular cohomology, 235
Sn .X /, set of singular n-chains, 234
singular homology, 235, 236
S n .X; G/, set of singular cochains, 234
singular simplex, 234
Sp2m .K/, symplectic group, 127
skew-symmetric multilinear map, 313
StabG .x/, stabilizer, 126
` skyscraper sheaf, 65
i2I Xi , sum of topological spaces, 252
small category, 271
supp.F/, support of an abelian sheaf, 50
smallest element, 277
supp.s/, support of a section, 50
smooth form, 181
sx , germ of a section, 46
smooth map, 76, 332
SymA .M /, symmetric algebra, 307
smooth premanifold, 76
SymAr .M /, symmetric power, 307
soft sheaf, 195
scalar multiplication, 291
source, 272
second countable space, 1
space
section of étalé space, 53
topological, 245
section of presheaf, 42
sphere, 110
segment, 279
split exact sequence, 296
semilocally simply connected space, 38
split subbundle, 177
separable space, 2
stabilizer, 126
separated map, 13
stalk, 46
set
standard simplex, 233
pointed, 140
star center of a star-shaped space, 23
set with group action, 126
star-shaped space, 23
sheaf, 44
start point of path, 23
abelian, 44
Stein manifold, 111
associated to a presheaf, 51
Stiefel-Whitney class of a line bundle, 201
constant, 52
structure sheaf, 70
c-soft, 230
Index 353
subbundle of vector bundle, 175 tensor product of modules, 299

subcategory, 272 tensor product of OX -modules, 186
full, 272 theorem
submanifold, 105 hairy ball, 244
open, 77 Mittag-Leffler, 221
regular, 106 of Darboux, 188
submersion, 100 universal coefficient, 236
submodule, 292 topological embedding, 250
submodule of OX -module, 168 closed, 250
subobject, 327 open, 250
subpremanifold, 105 topological group, 262
open, 77 topological space, 245
subpresheaf, 65 Baire, 18
subsheaf, 65 compact, 260
subspace, 250 connected, 257
meager, 18 contractible, 22
relatively Hausdorff, 12 coproduct, 252
successor, 278 discrete, 245
sum of ringed spaces, 75 fiber product, 253
sum of topological spaces, 252 first countable, 1
support, 50 gluing, 252
surjective morphism in an abelian category, 327 Hausdorff, 256
surjective morphism of sheaves, 49 hereditarily Lindelöf, 17
symmetric algebra, 307 hereditarily paracompact, 6
symmetric multilinear map, 307 Kolmogorov, 265
symmetric power, 307 Lindelöf, 1
symplectic form on an OX -module, 188 locally compact, 260
symplectic group, 137 locally connected, 267
locally contractible, 22
locally path connected, 24
T
metrizable, 246
T0 -space, 265
normal, 9
T1 -space, 10, 82
paracompact, 4
T2 -space, 256
path connected, 24
TF , tangent map, 178
product, 251
TM , tangent bundle, 177
quotient, 251
Tp .F /, tangent map, 93
regular, 17
Tp .M /, tangent space, 92
-compact, 2
tr, trace, 303
second countable, 1
TAr .M /, tensor power, 303
semilocally simply connected, 38
TM , tangent module, 178
separable, 2
T n , n-dimensional real torus, 86
simply connected, 27
tangent bundle, 177
subspace, 250
tangent map, 93
sum, 252
target, 272
T0 , 265
tautological vector bundle, 191
T1 , 10
Taylor expansion, 339
T2 , 256
Taylor series, 339
totally disconnected, 266
tensor algebra, 303
topology, 245
tensor power, 303
354 Index
coarser, 245 twisting atlas, 154

cofinite, 266
direct image, 250 U
discrete, 245 U.x/, set of open neighborhoods of x, 46
finer, 245 underlying C ˛ -structure, 81
generated by a set of subsets, 245 underlying real analytic premanifold, 82
induced by metric, 246 universal coefficient theorem, 236
induced on subset, 250 universal covering map, 29
inverse image, 250 universal ı-functor, 207
left order, 286 upper bound, 277
order, 267
p-adic, 264 V
particular point, 265 vanishing order, 118
product, 251 vector bundle, 165, 167
quotient, 251 C ˛ -, 165
right order, 286 complex, 165
torsion free module, 312 holomorphic, 165
torsor, 140 vector field, 178
pseudo, 140
trivial, 140 W
torus well order, 278
real 1-dimensional, 82 winding number, 39
real n-dimensional, 86
total order, 277 Y
totally disconnected space, 266 Y , closure, 249
trace, 303 Y ı , interior, 249
transfinite induction, 278 Yoneda product, 232
transfinite recursion, 278
transitive action of a sheaf of groups, 149 Z
transitive group action, 126 ZL 1 .U; /, Čech 1-cocycles, 142
transversal morphisms of premanifolds, 112 Z, integers, xi
trivial metric, 264 zero module, 291
trivial torsor, 140 zero object, 325
twist, 155 zero set, 110

T. Wedhorn. Manifolds, Sheaves, and Cohomology

Uploaded by

Copyright:

Available Formats

T. Wedhorn. Manifolds, Sheaves, and Cohomology

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

T. Wedhorn. Manifolds, Sheaves, and Cohomology

Uploaded by

Copyright:

Available Formats

Springer Studium Mathematik – Master

More information about this series at http://www.springer.com/series/13893

Springer Studium Mathematik – Master

Mathematics Subject Classification (2010): 14-0, 18-01

Editor: Ulrike Schmickler-Hirzebruch

Printed on acid-free paper

This Springer Spektrum imprint is published by Springer Nature

For a set I and elements i; j 2 I we use the “Kronecker delta”

We denote by N the set f1; 2; : : : g of natural numbers (without 0), N0 WD N [ f0g. Z

Br .x0 / WD f x 2 X I d.x; x0 / < r g ;

Let V be a finite-dimensional K-vector space. If not otherwise stated, we endow a subset

2 Algebraic Topological Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . 21

4.5 Basic Constructions and Further Examples of Manifolds . . . . . . . . 85

6 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

7 Torsors and Non-abelian Čech Cohomology . . . . . . . . . . . . . . . . . . . 139

9 Soft Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

10 Cohomology of Complexes of Sheaves . . . . . . . . . . . . . . . . . . . . . . . 205

10.3 Acyclic Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

11 Cohomology of Constant Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . 233

12 Appendix A: Basic Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

13 Appendix B: The Language of Categories . . . . . . . . . . . . . . . . . . . . 271

14 Appendix C: Basic Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

15 Appendix D: Homological Algebra . . . . . . . . . . . . . . . . . . . . . . . . . 317

15.4 Abelian Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

16 Appendix E: Local Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331

1.1 Countability Properties for Topological Spaces

Definition 1.1 (Countability properties). Let X be a topological space.

1. X is called first countable if every point of x has a countable neighborhood basis.

© Springer Fachmedien Wiesbaden 2016 1

4. X is called separable if it contains a countable dense subset.

Example 1.2. Let X be a topological space.

both form countable neighborhood bases of x.

1. Every subspace Z of a second countable space X is second countable: If BX a count-

Proposition 1.4. Let X be a topological space.

1. If X is second countable, then X is first countable, separable, and a Lindelöf space.

Proof. 1. Let B D f Bn I n 2 N g be a countable basis for X. Then for every point x 2 X

Proposition 1.5. Let G be a topological group.

Proof. 1. As G is connected, G is generated by C (Appendix ˚ Corollary 12.59).

Combining 1. and 3. of Proposition 1.5 we obtain:

1.2 Paracompact Spaces

Definition 1.7 (Refinement). Let X be a topological space and let U D .Ui /i 2I be

Any subcovering is a refinement.

Definition 1.8 (Paracompact spaces). A topological space X is called paracompact if

B WD f V \ W I V 2 B0 and W 2 B0 with W  Cx for some x 2 X g :

Proof of 2. Let U D .Ui /i be an open covering of X. Let W D .Wj /j 2J be the

Remark 1.11. Let X be a paracompact space. Then every closed subspace Y of X is

Open subspaces of paracompact spaces are not paracompact in general ([BouGT1]

Remark and Definition 1.12 (Hereditarily paracompact). A topological space X is

(i) Every open subspace of X is paracompact.

Proposition 1.13. Every metrizable topological space is hereditarily paracompact.

(i) If n m C p, then B2mp .x/ \ Vi;n D ; for all i 2 I .

B2mp \ B2n .y/ D ; for all y 2 Xi;n because n p C 1 and m C p p C 1. This

1.3 Normal Spaces

Theorem 1.15 (Urysohn’s theorem/Tietze extension theorem). Let X be a topological

Figure 1.1 Separation of two

(i) ) (ii) ) (iii) ) (i) ) (v) ) (iv) ) (i).

Set U1 WD X. Then X n B is a neighborhood of A and by (iii) we find an open subset

X n UN q is an open neighborhood of x. Furthermore, for y 2 X n UN q we have y … Up for

1. jgn .x/j .1=3/.2=3/n for x 2 X.

B WD f V \ W I V 2 B0 and W 2 B0 with W Cx for some x 2 X g :

(i) If n m C p, then B2mp .x/ \ Vi;n D ; for all i 2 I .

B2mp \ B2n .y/ D ; for all y 2 Xi;n because n p C 1 and m C p p C 1. This

Proof. “(i) ) (ii)”. Let y 2 Y and x1 ; x2 2 f 1 .y/ with x1 ¤ x2 . Then .x1 ; x2 /

Lemma 1.28. Let X and Y be topological spaces, A X and B Y compact subsets

Z WD f J I I J finite and non-emptyg [ fI g:

Endow Z with the topology generated by the sets VK WD f J 2 Z I K J g for every

closed in Y 0 because f is closed by assumption. Let A Y 0 Y X be a closed subset and

.Y 0 Y X/. By hypothesis, fy 0 g f 1 .g.y 0 // is compact and we may apply Lemma 1.28

Proposition 12.51). Now let B Y be closed. As f is surjective, we have g.B/ D

.Rn n f0g/ Œ0; 1 ! Rn n f0g; .x; t/ 7! .1 t/x=jjxjj C tx:

1. Let W Œ0; 1 ! X be a path. Define

W Œ0; 1 ! X; .t/ D .1 t/