Graduate Texts in Mathematics 153

Editorial Board
S. Axler F.w. Gehring K.A. Ribet
Graduate Texts in Mathematics

TAKEUTI/ZARING. Introduction to 33 HIRSCH. Differential Topology.

Axiomatic Set Theory. 2nd ed. 34 SPITZER. Principles of Random Walk.
2 OXTOBY. Measure and Category. 2nd ed. 2nd ed.
3 SCHAEFER. Topological Vector Spaces. 35 ALEXANDER/WERMER. Several Complex
2nd ed. Variables and Banach Algebras. 3rd ed.
4 HILTON/STAMMBACH. A Cour~e in 36 KELLEy/NAMIOKA et al. Linear Topological
Homological Algebra. 2nd ed. Spaces.
5 MAC LANE. Categories for the Working 37 MONK. Mathematical Logic.
Mathematician. 2nd ed. 38 GRAUERT/FRITZSCHE. Several Complex
6 HUGHES/PIPER. Projective Planes. Variables.
7 SERRE. A Course in Arithmetic. 39 ARVESON. An Invitation to CO-Algebras.
8 TAKEUTI/ZARING. Axiomatic Set Theory. 40 KEMENy/SNELLiKNAPP. Denumerable
9 HUMPHREYS. Introduction to Lie Algebras Markov Chains. 2nd ed.
and Representation Theory. 41 ApOSTOL. Modular Functions and Dirichlet
10 COHEN. A Course in Simple Homotopy Series in Number Theory.
Theory. 2nd ed.
11 CONWAY. Functions of One Complex 42 SERRE. Linear Representations of Finite
Variable I. 2nd ed. Groups.
12 BEALS. Advanced Mathematical Analysis. 43 GILLMAN/JERISON. Rings of Continuous
13 ANDERSON/FuLLER. Rings and Categories Functions.
of Modules. 2nd ed. 44 KENDIG. Elementary Algebraic Geometry.
14 GOLUBITSKY/GUILLEMIN. Stable Mappings 45 LOEVE. Probability Theory I. 4th ed.
and Their Singularities. 46 LOEVE. Probability Theory II. 4th ed.
15 BERBERIAN. Lectures in Functional 47 MOISE. Geometric Topology in
Analysis and Operator Theory. Dimensions 2 and 3.
16 WINTER. The Structure of Fields. 48 SACHS/WU. General Relativity for
17 ROSENBLAn. Random Processes. 2nd ed. Mathematicians.
18 HALMOS. Measure Theory. 49 GRUENBERG/WEIR. Linear Geometry.
19 HALMOS. A Hilbert Space Problem Book. 2nd ed.
2nd ed. 50 EDWARDS. Fermat's Last Theorem.
20 HUSEMOLLER. Fibre Bundles. 3rd ed. 51 KLINGENBERG. A Course in Differential
21 HUMPHREYS. Linear Algebraic Groups. Geometry.
22 BARNES/MACK. An Algebraic Introduction 52 HARTSHORNE. Algebraic Geometry.
to Mathematical Logic. 53 MANIN. A Course in Mathematical Logic.
23 GREUB. Linear Algebra. 4th ed. 54 GRAVER/WATKINS. Combinatorics with
24 HOLMES. Geometric Functional Analysis Emphasis on the Theory of Graphs.
and Its Applications. 55 BROWN/PEARCY. Introduction to Operator
25 HEWITT/STROMBERG. Real and Abstract Theory I: Elements of Functional
Analysis. Analysis.
26 MANES. Algebraic Theories. 56 MASSEY. Algebraic Topology: An
27 KELLEY. General Topology. Introduction.
28 ZARISKI/SAMUEL. Commutative Algebra. 57 CROWELL/Fox. Introduction to Knot
Vol.1. Theory.
29 ZARISKI/SAMUEL. Commutative Algebra. 58 KOBLITZ. p-adic Numbers, p-adic Analysis,
VoU!. and Zeta-Functions. 2nd ed.
30 JACOBSON. Lectures in Abstract Algebra I. 59 LANG. Cyclotomic Fields.
Basic Concepts. 60 ARNOLD. Mathematical Methods in
31 JACOBSON. Lectures in Abstract Algebra 11. Classical Mechanics. 2nd ed.
Linear Algebra. 61 WHITEHEAD. Elements of Homotopy
32 JACOBSON. Lectures in Abstract Algebra
111. Theory of Fields and Galois Theory. (continued after index)
William Fulton

Algebraic Topology
A First Course

With 137 Illustrations

~ Springer
William Fulton
Mathematics Department
University of Chicago
Chicago, IL 60637
USA

Editorial Board
S. Axler F.w. Gehring K.A. Ribet
Mathematics Department Mathematics Department Mathematics Department
San Francisco State East Hall University of California
University University of Michigan at Berkeley
San Francisco, CA 94132 Ann Arbor, MI 48109 Berkeley, CA 94720-3840
USA USA USA

Mathematics Subject Classifications (1991): 55-0 I

Library of Congress Cataloging-in-Publication Data

Fulton, William, 1939-
Algebraic topologylWilliam Fulton.
p. cm. - (Graduate texts in mathematics)
Includes bibliographical references and index.
ISBN-13: 978-0-387-94327-5
I. Algebraic topology. I. Title. II. Series.
QA612.F85 1995
514'.2-dc20 94-21786

ISBN-13: 978-0-387-94327-5 e-ISI3N: 978-1-4612-4180-5

DOl: 10.1007/978-1-4612-4180-5

© 1995 Springer Science+Business Media, Inc.

All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, Inc., 233 Spring Street, New York,
NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in
connection with any form of information storage and retrieval, electronic adaptation, computer
software, or by similar or dissimilar methodology now know or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks and similar terms, even if the
are not identified as such, is not to be taken as an expression of opinion as to whether or not they are
subject to proprietary rights.

(EB)

9 8 7 6 5

springeronline.com
To the memory of my parents
Preface

To the Teacher. This book is designed to introduce a student to some

of the important ideas of algebraic topology by emphasizing the re-
lations of these ideas with other areas of mathematics. Rather than
choosing one point of view of modem topology (homotopy theory,
simplicial complexes, singular theory, axiomatic homology, differ-
ential topology, etc.), we concentrate our attention on concrete prob-
lems in low dimensions, introducing only as much algebraic machin-
ery as necessary for the problems we meet. This makes it possible to
see a wider variety of important features of the subject than is usual
in a beginning text. The book is designed for students of mathematics
or science who are not aiming to become practicing algebraic topol-
ogists-without, we hope, discouraging budding topologists. We also
feel that this approach is in better harmony with the historical devel-
opment of the subject.
What would we like a student to know after a first course in to-
pology (assuming we reject the answer: half of what one would like
the student to know after a second course in topology)? Our answers
to this have guided the choice of material, which includes: under-
standing the relation between homology and integration, first on plane
domains, later on Riemann surfaces and in higher dimensions; wind-
ing numbers and degrees of mappings, fixed-point theorems; appli-
cations such as the Jordan curve theorem, invariance of domain; in-
dices of vector fields and Euler characteristics; fundamental groups
and covering spaces; the topology of surfaces, including intersection
numbers; relations with complex analysis, especially on Riemann sur-

vii
viii Preface

faces; ideas of homology, De Rham cohomology, Cech cohomology,

and the relations between them and with fundamental groups; meth-
ods of calculation such as the Mayer-Vietoris and Van Kampen theo-
rems; and a taste of the way algebra and "functorial" ideas are used
in the subject.
To achieve this variety at an elementary level, we have looked at
the first nontrivial instances of most of these notions: the first ho-
mology group, the first De Rham group, the first Cech group, etc.
In the case of the fundamental group and covering spaces, however,
we have bowed to tradition and included the whole story; here the
novelty is on the emphasis on coverings arising from group actions,
since these are what one is most likely to meet elsewhere in mathe-
matics.
We have tried to do this without assuming a graduate-level knowl-
edge or sophistication. The notes grew from undergraduate courses
taught at Brown University and the University of Chicago, where about
half the material was covered in one-semester and one-quarter courses.
By choosing what parts of the book to cover-and how many of the
challenging problems to assign-it should be possible to fashion courses
lasting from a quarter to a year, for students with many backgrounds.
Although we stress relations with analysis, the analysis we require or
develop is certainly not "hard analysis."
We start by studying questions on open sets in the plane that are
probably familiar from calculus: When are path integrals independent
of path? When are I-forms exact? (When do vector fields have po-
tential functions?) This leads to the notion of winding number, which
we introduce first for differentiable paths, and then for continuous
paths. We give a wide variety of applications of winding numbers,
both for their own interest and as a sampling of what can be done
with a little topology. This can be regarded as a glimpse of the general
principle that algebra can be used to distinguish topological features,
although the algebra (an integer!) is fairly meager.
We introduce the first De Rham cohomology group of a plane do-
main, which measures the failure of closed forms to be exact. We
use these groups, with the ideas of earlier chapters, to prove the Jor-
dan curve theorem. We also use winding numbers to study the sin-
gularities of vector fields. Then I-chains are introduced as convenient
objects to integrate over, and these are used to construct the first ho-
mology group. We show that for plane open sets homology, winding
numbers, and integrals all measure the same thing; the proof follows
ideas of Brouwer, Artin, and Ahlfors, by approximating with grids.
As a first excursion outside the plane, we apply these ideas to sur-
Preface ix

faces, seeing how the global topology of a surface relates to local

behavior of vector fields. We also include applications to complex
analysis. The ideas used in the proof of the Jordan curve theorem are
developed more fully into the Mayer-Vietoris story, which becomes
our main tool for calculations of homology and cohomology groups.
Standard facts about covering spaces and fundamental groups are
discussed, with emphasis on group actions. We emphasize the con-
struction of coverings by patching together trivial coverings, since
these ideas are widely used elsewhere in mathematics (vector bundles,
sheaf theory, etc.), and Cech cocycles and cohomology, which are
widely used in geometry and algebra; they also allow, following
Grothendieck, a very short proof of the Van Kampen theorem. We
prove the relation among the fundamental group, the first homology
group, the first De Rham cohomology group, and the first Cech co-
homology group, and the relation between cohomology classes, dif-
ferential forms, and the coverings arising from multivalued functions.
We then turn to the study of surfaces, especially compact oriented
surfaces. We include the standard classification theorem, and work
out the homology and cohomology, including the intersection pairing
and duality theorems in this context. This is used to give a brief in-
troduction to Riemann surfaces, emphasizing features that are acces-
sible with little background and have a topological flavor. In partic-
ular, we use our knowledge of coverings to construct the Riemann
surface of an algebraic curve; this construction is simple enough to
be better known than it is. The Riemann-Roch theorem is included,
since it epitomizes the way topology can influence analysis. Finally.
the last part of the book contains a hint of the directions the subject
can go in higher dimensions. Here we do include the construction and
basic properties of general singular (cubical) homology theory, and
use it for some basic applications. For those familiar with differential
forms on manifolds, we include the generalization of De Rham theory
and the duality theorems.
The variety of topics treated allows a similar variety of ways to use
this book in a course, since many chapters or sections can be skipped
without making others inaccessible. The first few chapters could be
used to follow or complement a course in point set topology. A course
with more algebraic topology could include the chapters on funda-
mental groups and covering spaces, and some of the chapters on sur-
faces. It is hoped that, even if a course does not get near the last third
of the book, students will be tempted to look there for some idea of
where the subject can lead. There is some progression in the level of
difficulty, both in the text and the problems. The last few chapters
x Preface

may be best suited for a graduate course or a year-long undergraduate

course for mathematics majors.
We should also point out some of the many topics that are omitted
or slighted in this treatment: relative theory, homotopy theory, fibra-
tions, simplicial complexes or simplicial approximation, cell com-
plexes, homology or cohomology with coefficients, and serious ho-
mological algebra.

To the Student. Algebraic topology can be thought of as the study of

the shapes of geometric objects. It is sometimes referred to in popular
accounts as "rubber-sheet geometry." In practice this means we are
looking for properties of spaces that are unchanged when one space
is deformed into another. "Doughnuts and teacups are topologically
the same." One problem of this type goes back to Euler: What re-
lations are there among the numbers of vertices, edges, and faces in
a convex polytope, such as a regular solid, in space? Another early
manifestation of a topological idea came also from Euler, in the
Konigsberg bridge problem: When can one trace out a graph without
traveling over any edge twice? Both these problems have a feature
that characterizes one of the main attractions, as well as the power,
of modern algebraic topology-that a global question, depending on
the overall shape of a geometric object, can be answered by data that
are collected locally. Since these are so appealing-and perhaps to
capture your interest while we turn to other topics-they are included
as problems with hints at the end of this Preface.
In fact, modern topology grew primarily out of its relation with
other subjects, particularly analysis. From this point of view, we are
interested in how the shape of a geometric object relates to, or con-
trols, the answers to problems in analysis. Some typical and histor-
ically important problems here are:

(i) whether differential forms w on a region that are closed (dw =

0) must be exact (w = djL) depends on the topology of the region;
(ii) the behavior of vector fields on a surface depends on the topol-
ogy of the surface; and
(iii) the behavior of integrals f dx/YR(x) depends on the topology
of the surface l = R(x), here with x and y complex variables.

In this book we will begin with the first of these problems, working
primarily in open sets in the plane. There is one disadvantage that
must be admitted right away: this geometry is certainly flat, and lacks
some of the appeal of doughnuts and teacups. Later in the book we
Preface xi

will in fact discuss generalizations to curved spaces like these, but at

the start we will stick to the plane, where the analysis is simpler. The
topology of open sets in the plane is more interesting than one might
think. For example, even the question of the number of connected
components can be challenging. The famous Jordan curve theorem,
which is one of our goals here, says that the complement of a plane
set that is homeomorphic to a circle always has two components-a
fact that will probably not surprise you, but whose proof is not so
obvious. We will also spend some time on the second problem, which
includes the popular problem of whether one can "comb the hair on
a billiard ball." We will include some applications to complex anal-
ysis, later discussing some of the ideas related to the third problem.
To read this book you need a basic understanding of fundamental
notions of the other topology, known as point set topology or general
topology. This means that you should know what is meant by words
like connected, open, closed, compact, and continuous, and some of
the basic facts about them. The notions we need are recalled in Ap-
pendix A; if most of this is familiar to you, you should have enough
prerequisites. Because of our approach via analysis, you will also
need to know some basic facts about calculus, mainly for functions
of one or two variables. These calculus facts are set out in Appendix
B. In algebra you will need some basic linear algebra, and basic no-
tions about groups, especially abelian groups, which are recalled or
proved in Appendix C.
There will be many sorts of exercises. Some exercises will be rou-
tine applications of or variations on what is done in the text. Those
requiring (we estimate) a little more work or ingenuity will be called
problems. Many will have hints at the end of the book, for you to
avoid looking at. There will also be some projects, which are things
to experiment on, speculate about, and try to develop on your own.
For example, one general project can be stated right away: as we go
along, try to find analogues in 3-space or n-space for what we do in
the plane. (Some of this project is carried out in Part XI.)

Problem 0.1. Suppose X is a graph, which has a finite number of

vertices (points) and edges (homeomorphic to a closed interval), with
each edge having its endpoints at vertices, and otherwise not inter-
secting each other. Assume X is connected. When, and how, can you
trace out X, traveling along each edge just once? Can you prove your
answer?
xii Preface

Exercise 0.2. Let v, e, andfbe the number of vertices, edges, and

faces on a convex polyhedron. Compute these numbers for the five
regular solids, for prisms, and some others. Find a relation among
them. Experiment with other polyhedral shapes .

• =6, e= 12, f=8

(Note: This problem is "experimental." Proofs are not expected.)

Acknowledgments. I would like to thank the students who came up

with good ideas that contributed to notes for the courses, especially
K. Ryan, J. Silverman, J. Linhart, J. Trowbridge, and G. Gutman.
Thanks also to my colleagues, for answering many of my questions
and making useful suggestions, especially A. Collino, J. Harris, R.
MacPherson, J.P. May, R. Narasimhan, M. Rothenberg, and S.
Weinberger. I am grateful to Chandler Fulton for making the draw-
ings. I would most like to thank those who first inspired me with
some of these ideas in courses about three decades ago: H. Federer,
J. Milnor, and J. Moore.

William Fulton
Preface to Corrected Edition
I am grateful to J. McClure, J. Buhler, D. Goldberg, and R.B. Burckel
for pointing out errors and misprints in the first edition, and for useful
suggestions.
Contents

Preface vii

PART I
CALCULUS IN THE PLANE

CHAPTER I
Path Integrals 3
I a. Differential Forms and Path Integrals 3
lb. When Are Path Integrals Independent of Path? 7
Ic. A Criterion for Exactness IO

CHAPTER 2
Angles and Deformations 17
2a. Angle Functions and Winding Numbers 17
2b. Reparametrizing and Deforming Paths 23
2c. Vector Fields and Fluid Flow 27

PART II
WINDING NUMBERS

CHAPTER 3
The Winding Number 35
3a. Definition of the Winding Number 35
3b. Homotopy and Reparametrization 38
3c. Varying the Point 42
3d. Degrees and Local Degrees 43

Xlll
xiv Contents

CHAPTER 4
Applications of Winding Numbers 48
4a. The Fundamental Theorem of Algebra 48
4b. Fixed Points and Retractions 49
4c. Antipodes 53
4d. Sandwiches 56

PART III
COHOMOLOGY AND HOMOLOGY, I

CHAPTER 5
De Rham Cohomology and the Jordan Curve Theorem 63
5a. Definitions of the De Rham Groups 63
5b. The Coboundary Map 65
5c. The Jordan Curve Theorem 68
5d. Applications and Variations 72

CHAPTER 6
Homology 78
6a. Chains, Cycles, and HoU 78
6b. Boundaries, H,U, and Winding Numbers 82
6c. Chains on Grids 85
6d. Maps and Homology 89
6e. The First Homology Group for General Spaces 91

PART IV
VECTOR FIELDS

CHAPTER 7
Indices of Vector Fields 97
7a. Vector Fields in the Plane 97
7b. Changing Coordinates 101
7c. Vector Fields on a Sphere 102

CHAPTER 8
Vector Fields on Surfaces 106
8a. Vector Fields on a Torus and Other Surfaces 106
8b. The Euler Characteristic 113

PART V
COHOMOLOGY AND HOMOLOGY, II

CHAPTER 9
Holes and Integrals 123
9a. Multiply Connected Regions 123
Contents xv

9b. Integration over Continuous Paths and Chains 127

9c. Periods of Integrals 130
9d. Complex Integration 131

CHAPTER 10
Mayer-Vietoris 137
lOa. The Boundary Map 137
lOb. Mayer-Vietoris for Homology 140
lOco Variations and Applications 144
IOd. Mayer-Vietoris for Cohomology 147

PART VI
COVERING SPACES AND FUNDAMENTAL GROUPS, I

CHAPTER 11
Covering Spaces 153
11 a. Definitions 153
lIb. Lifting Paths and Homotopies 156
lIc. G-Coverings 158
lId. Covering Transformations 163

CHAPTER 12
The Fundamental Group 165
12a. Definitions and Basic Properties 165
12b. Homotopy 170
12c. Fundamental Group and Homology 173

PART VII
COVERING SPACES AND FUNDAMENTAL GROUPS, II

CHAPTER 13
The Fundamental Group and Covering Spaces 179
13a. Fundamental Group and Coverings 179
13b. Automorphisms of Coverings 182
13c. The Universal Covering 186
13d. Coverings and Subgroups of the Fundamental Group 189

CHAPTER 14
The Van Kampen Theorem 193
14a. G-Coverings from the Universal Covering 193
14b. Patching Coverings Together 196
14c. The Van Kampen Theorem 197
14d. Applications: Graphs and Free Groups 201
xvi Contents

PART VIII
COHOMOLOGY AND HOMOLOGY, III

CHAPTER 15
Cohomology 207
15a. Patching Coverings and Cech Cohomology 207
15b. Cech Cohomology and Homology 210
15c. De Rham Cohomology and Homology 213
15d. Proof of Mayer-Vietoris for De Rham Cohomology 217

CHAPTER 16
Variations 219
16a. The Orientation Covering 219
16b. Coverings from I-Forms 220
16c. Another Cohomology Group 222
16d. G-Sets and Coverings 225
16e. Coverings and Group Homomorphisms 227
16f. G-Coverings and Cocycles 228

PART IX
TOPOLOGY OF SURFACES

CHAPTER 17
The Topology of Surfaces 233
17a. Triangulation and Polygons with Sides Identified 233
17b. Classification of Compact Oriented Surfaces 236
17c. The Fundamental Group of a Surface 242

CHAPTER 18
Cohomology on Surfaces 247
18a. I-Forms and Homology 247
18b. Integrals of 2-Forms 251
18c. Wedges and the Intersection Pairing 252
18d. De Rham Theory on Surfaces 256

PART X
RIEMANN SURFACES

CHAPTER 19
Riemann Surfaces 263
19a. Riemann Surfaces and Analytic Mappings 263
19b. Branched Coverings 268
19c. The Riemann-Hurwitz Formula 272
Contents xvii

CHAPTER 20
Riemann Surfaces and Algebraic Curves 277
20a. The Riemann Surface of an Algebraic Curve 277
20b. Meromorphic Functions on a Riemann Surface 281
20c. Holomorphic and Meromorphic I-Forms 284
20d. Riemann's Bilinear Relations and the Jacobian 289
20e. Elliptic and Hyperelliptic Curves 291

CHAPTER 21
The Riemann-Roch Theorem 295
21a. Spaces of Functions and I-Forms 295
21b. Adeles 299
21c. Riemann-Roch 303
21d. The Abel-Jacobi Theorem 306

PART XI
HIGHER DIMENSIONS

CHAPTER 22
Toward Higher Dimensions 317
22a. Holes and Forms in 3-Space 317
22b. Knots 320
22c. Higher Homotopy Groups 324
22d. Higher De Rham Cohomology 325
22e. Cohomology with Compact Supports 328

CHAPTER 23
Higher Homology 332
23a. Homology Groups 332
23b. Mayer-Vietoris for Homology 334
23c. Spheres and Degree 339
23d. Generalized Jordan Curve Theorem 343

CHAPTER 24
Duality 346
24a. Two Lemmas from Homological Algebra 346
24b. Homology and De Rham Cohomology 350
24c. Cohomology and Cohomology with Compact Supports 355
24d. Simplicial Complexes 359

APPENDICES

ApPENDIX A
Point Set Topology 367
AI. Some Basic Notions in Topology 367
A2. Connected Components 369
xviii Contents

A3. Patching 370

A4. Lebesgue Lemma 371

ApPENDIX B
Analysis 373
BI. Results from Plane Calculus 373
B2. Partition of Unity 375

ApPENDIX C
Algebra 378
CI. Linear Algebra 378
C2. Groups; Free Abelian Groups 380
C3. Polynomials; Gauss's Lemma 385

ApPENDIX 0
On Surfaces 387
01. Vector Fields on Plane Domains 387
02. Charts and Vector Fields 389
03. Differential Forms on a Surface 391

ApPENDIX E
Proof of Borsuk's Theorem 393

Hints and Answers 397

References 419
Index of Symbols 421
Index 425
 PART I

CALCULUS IN THE PLANE

In this first part we will recall some basic facts about differentiable
functions, forms, and vector fields, and integration over paths. Much
of this should be familiar, although perhaps from a different point of
view. At any rate, several of these notions will be needed later, so
we take this opportunity to fix the ideas and notation. And of course,
we will be looking particularly at the role played by the shapes (to-
pology) of the underlying regions where these things are defined. For
the facts that we use, see Appendix B either for precise statements
or proofs. Most of this material is included mainly for motivation,
and will be developed from a purely topological point of view later;
one fact proved in the first chapter-that a closed I-form on an open
rectangle is the differential of a function-will be used later.
In the second chapter we will see that for any smooth path not
passing through the origin, it is possible to define a smooth function
that measures how the angle is changing as one moves along the path.
This gives us a notion of winding number-how many times a closed
path "goes around" the origin. Facts about changing variables in in-
tegrals are used to see what happens to integrals and winding numbers
when paths are reparametrized and deformed. The third section in-
cludes a reinterpretation of the facts from the first chapter in vector
field language, and gives a physical interpretation of these ideas to
fluid flow. Although we will not use these facts in the book, we will
study vector fields later, and it should be useful to have some feeling
for them, if you don't already.
CHAPTER 1

Path Integrals

la. Differential Forms and Path Integrals

In this chapter, U will denote an open set in the plane 1R2, for ex-
ample, the unshaded part of

A smooth or C(6'" Junction on U is a function f: U -IR such that all

partial derivatives of all orders' exist and are continuous. In partic-
ular, its partial derivatives aJlax and aJlay are C(6'" functions on U.
Since in this chapter we will only consider C(6'" functions, we will
sometimes just call them functions .
A functionJ on U is called locally constant if every point of U has
a neighborhood on which J is constant.

I We will never need more than continuous second derivatives, and often much less.

The few functions that we actually use , however, will be infinitely differentiable .
The extra hypotheses are included so we never have to worry about differentiating
any function we meet. The analytically inclined reader may enjoy supplying minimal
hypotheses for each assertion.

3
4 1. Path Integrals

Exercise 1.1. Prove that a function on an open set U in the plane is

locally constant if and only if it is constant on each connected com-
ponent of U. In other words, defining a locally constant function on
U is the same as specifying a constant for each of its connected com-
ponents.

If f is locally constant, then af/ax = 0 and al/ay = 0 (identically,

as functions on U), as follows immediately from the definitions of
partial derivatives. The converse is also true and only slightly harder:

Proposition 1.2. If f is a smooth function on U, then f is locally

constant if and only if af/ ax = 0 and af/ ay = O.
Proof. The point is that, in a rectangular neighborhood of a point of
U, tbe condition af/ax = 0 means thatfis independent of x, i.e., that
f is constant along horizontal lines. Likewise af/ ay = 0 means that f
is constant along vertical lines, and both conditions make f constant
in the rectangle. 0

It may not be much, but there is a grain of topology in this:

Corollary 1.3. The open set U is connected if and only if every smooth
function f in U with af/ ax = 0 and af/ ay = 0 is constant. 0

A differential Ijorm, or just a Ijorm, on U is given by a pair of

smooth functions p and q on U. We will usually denote a I-form by
W, and we will write W = pdx + qdy. This can be regarded as just a
formal notation, with the dx and dy there merely to indicate what we
will do with I-forms, namely integrate them over paths. The pair of
functions (p, q) can also be identified with a vector field on U. For
this interpretation, see §2c in Chapter 2.
By a smooth path (just called a path in this chapter) in U, we mean
a mapping -y: [a, b]~ U from a bounded interval into U that is con-
tinuous on [a, b] and differentiable in the open interval (a, b); in ad-
dition, to avoid any trouble at the endpoints, we assume the two com-
ponent functions of -y can be extended to ~oo functions in some
neighborhood of [a,b]. So -y(t) = (x(t),y(t», where x and y are re-
strictions of smooth functions on an intervaf (a - 10, b + E), for some

2 In fact, there are many extensions of these functions to such neighborhoods, but
we will never care about values outside the interval [a, b]. The assumption is useful
to assure that the derivatives of these functions are continuous on the whole closed
interval [a, b].
lao Differential Fonns and Path Integrals 5

positive number E. We call 'Y(a) the initial point of 'Y, and 'Y(b) the
final point; 'Y(a) and 'Y(b) are called the endpoints, and we say that 'Y
is a path from 'Y(a) to 'Y(b).
With w = p dx + q dy as above, and 'Y a path given by the pair of
functions 'Y(t) = (x(t),y(t», the integral I"/w ofw along 'Y is defined
by the formula

1"/ = Jar
w b
(P(X(t) , y(t» dx + q(x(t) , y(t» dy ) dt.
dt dt
Note that the integrand is continuous on [a, b], so the integral exists,
as a limit of Riemann sums.
The question we will be concerned with is this: given a I-form w
on U, when does the integral I"/w depend only on the endpoints 'Y(a)
and 'Y(b) of 'Y, and not on the actual path between them?

Language is usually abused here, saying the integral is "independent

of path." This happens whenever there is a "potential function":

Proposition 1.4. If w = af/ ax dx + af/ ay dy, for some «6"" function f

on an open set containing the path 'Y, then

L=w f('Y(b» - f('Y(a».

Proof. Since, by the chain rule,

d dx ~
- (/('Y(t))) = -~ (x(t),y(t» ~
- + - (x(t),y(t»-,
& ~ & ~ &
the integral is

1"/ = Jar !!..

w
b

dt
(/('Y(t»)dt = f('Y(b» - f('Y(a» ,

the last step by the fundamental theorem of calculus. o

6 1. Path Integrals

We write df= af/axdx + af/aydy for this I-form, and say that w
is the differential of f if w = df.

Exercise 1.5. Show that df= dg on U if and only iff - g is locally

constant on U.

For an example, take U to be the right half plane, i.e., the set of
points (x,y) with x> D. Consider the functionfthat measures the an-
gle in polar coordinates, measured counterclockwise from the x-axis.
Analytically, f(x,y) = tan-I(y/x), so

df = I (Y)
--2 dx +
1 + (Y/X)2
I (1)- dy
1+ (Y/X)2 X
x

For example, if -y is any path in U from (1, -1) to (2,2), then

J-ydf= 'Tr/2, since that is the change in angle between the two points.

Although the function fix, y) = tan -I (y / x), at least as it stands, is

not defined where x = D, the expression we found for df makes sense
everywhere except at the origin, and is a smooth I-form on the open
set ~2, {(D, D)}. Let us denote this I-form by w1'l:

-ydx + xdy 2
w1'l = ~ 2 on ~ ,{(D,D)}.
+y
In fact, although y /x cannot be extended across the y-axis, the func-
tion tan-I(y/x) can, at least away from the origin. This is clear if we
think of it geometrically as the angle in polar coordinates, which can
be extended, for example, to the complement of the negative x-axis:
lb. When Are Path Integrals Independent of Path? 7

("')~o< • . ,

~ L~=o

-1t<~<o

However, there is trouble in trying to extend this angle function to

be well defined everywhere on ~2 \ {(O, O)}. In fact, we can use our
last proposition to show that there is no smooth function g on ~2 \ {(O, O)}
with dg = oo,'}. For example, if 'Y(t) = (cos(t), sin(t», O:s t:s 21T, is the
counterclockwise path around the unit circle, we calculate using the
definition of the path integral:

{oo,'} = f1f (-sin(t) • (-sin(t» + cos(t) . cos(t» dt

= f1f 1 dt = 21T.
Since "1(0) = 'Y(21T), it follows from Proposition 1.4 that oo,'} cannot be
the differential of any function.

Exercise 1.6. On which of the following open sets U is there a smooth

function g with dg = 00" on U? Prove your answers. (i) The upper half
plane {(x,y): y > O}. (ii) The union of the upper half plane and the
right half plane. (iii) The left half plane. (iv) The lower half plane.
(v) The complement of the negative x-axis. (vi) The annulus
{(x,y): I <X2 + l < 2}. (vii) Challenge. The points of the form
(ret cos(t) , ret sin(t» , 0 < t < 41T, 112 < r < 2.

Exercise 1.7. Is 00 = (xlix + ydy)/(x2 + y2)2 the differential of a func-

tion on ~2 \ {(O, O)}?

lb. When Are Path Integrals Independent of Path?

It will be useful to generalize the notion of smooth path in order to
allow integration over a sequence of such paths. Let us define a seg-
mented path "I to be a sequence of paths "11, "12, ... ,"In, where each
"Ii is a smooth path, and the final point of each "Ii is the initial point
of the next 'Yi+lo for i = 1, 2, ... , n - 1.
8 1. Path Integrals

We sometimes write 'Y = 'YI + ... + 'Yo for this segmented path. The
initial point of 'Y is defined to be the initial point of 'YI, and the final
point of'Y is defined to be the final point of 'Yo. The segmented path
is closed if the final point of 'Yo is the initial point of 'YI' If W is a 1-
form on an open set containing (the images of) these paths, we define

f'I
W = f
)'11
W + f
'12
W + ... + f
'In
W.

If 'Y is a segmented path in U from P to Q, and w = df in U, then it

follows from Proposition 1.4 (the "interior endpoints" canceling) that

Lw = f(Q) - f(P) .

We'll show now that the converse of this is also true:

Proposition 1.S. Let w be a 110rm on U. The following are equiv-

alent: (i) f'Y w = f 6 W for all segmented paths 'Y and l5 in U with the
same initial and final points; (ii) f W = 0 for all segmented paths
T

T in U that are closed; and (iii) w = df for some smooth function f

on U.
Proof. The preceding remark shows that (iii) implies (i). To show
that (ii) implies (i), we use the notion of the inverse of a path
(1: [a,b]~U, which is the path (1-1: [a,b]~U defined by
(1-I(t) = (1(b + a - t); note that the integral of any w along (1-1 is the
negative of the integral of w along (1 (cf. Exercise 2.12). Given'Y and
8 as in (ii), form the closed segmented path T which is first the se-
quence of paths making up 'Y, and then the inverses of the paths that
make up 8, but taken in the reverse order. Then fTW = J'Iw - Jaw,
from which the fact that (ii) implies (i) follows. That (i) implies (ii)
is obvious, by comparing a closed path with a constant path. To show
that (i) implies (iii), it is enough to find such a function on each
connected component of U, so we can assume U is connected, and
hence path-connected (see Appendix A2). Choose and fix an arbitrary
lb. When Are Path Integrals Independent of Path? 9

point Po in U, and define a function f on U by the formula

f(P) = J"Y 00,

where'Y is any segmented path from Po to Pin U. (See Exercise 1.9

below.) By assumption, this is a well-defined function on U. We claim
that aflax = p and aflay = q, where 00 = pdx + qdy. For the first, we
must look at the limit of
f(x + ax,y) - f(x,y)
ax
as ax approaches zero, and P = (x, y) is any point in U. To estimate
this, let (J be the path from (x,y) to (x + ax,y) given by the formula
(J(t) = (x + t, y) 0::0:;; t::O:;; ax, assuming for the moment that ax is pos-
itive. Let'Y be any segmented path from Po to P.

P = (x,y) _-+-_ CI
(x + 8x,y)

Po = (Xo,Yo)
Since 'Y + (J is a segmented path from Po to (x + ax, y),

f(x + ax,~ - f(x,y) = ~(L+cr 00 - L(0) = ~ (L (0)

= ~ (Ax p(x + t,y)dt.
axJo
By the mean value theorem, this last expression is equal to p(x*, y)
for some x* between x and x + ax. Letting ax~ 0, we have, since
p is continuous,

lim- 1 iAx p(x+t,y)dt = p(x,y),

ax-->oax 0

as required. If ax is negative, use instead the path (J(t) = (x - t,y),

o: 0:; t::o:;; laxl, and the argument is the same, with the modification
(L
~ (0) = I~I £axlp(X - t, y)( -1) dt

1 (Iaxl
= laxlJo p(x-t,y)dt=p(x*,y),
10 1. Path Integrals

with x + at :5x* :5x. The proof that dfjay = q is similar, interchang-

ing the roles of x and y, and is left as an exercise. D

Exercise 1.9. Show that an open set U in the plane is connected if

and only if there is a segmented path between any two points of U.
Challenge. Can you show that any two points in a connected open
set can be connected by an are, i.e., a path that is one-to-one, and
whose tangent vector never vanishes?

1c. A Criterion for Exactness

We want a practical criterion to tell if a given I-form 00 is the dif-
ferential of some function without going to the work of constructing
such a function. We shall need another fact from calculus, the equal-
ity of mixed partial derivatives: ajax(af/ay) = a/ay(af/ax). This
translates to a simple

Criterion 1.10. 00 = p dx + q dy cannot be the differential of a func-

tion unless aq / ax = ap / ay.

This necessary condition, however, is not always sufficient. For

example, if 00" is the I-form on U = ~2 \ {(O, O)} that we looked at
earlier, you can verify easily that 00" satisfies this condition-either
by direct calculation, or by the fact that any point in U has a neigh-
borhood on which the restriction of 00" is the differential of a func-
tion-but we have seen that 00" cannot be the differential of any func-
tion on U. We also saw that the restrictions of 00" to some simpler
open sets, like the right half plane, are the differentials of functions.
We will see that it is the topology of U that is controlling this situ-
ation.
A 2-form on U is an expression hdxdy, where h is a '€'" function
on U. Logically, as before, a 2-form can be identified with the func-
tion h that defines it, with the "dxdy" playing only a formal role.
The notation indicates that we will use 2-forms for integrating over
two-dimensional regions. All we will need is double integrals IIR h dx dy
over rectangles R = [a, b] x [c, d], defined as limits of Riemann sums.
If 00 = pdx + qdy is a I-form on U, define doo to be the 2-form

doo aq - ap)
= (- - dxdy.
ax ay
Ie. A Criterion for Exactness 11

So our criterion can be stated: if w = df, then doo = 0; or simply that

d(df) = 0 for all functions f.
Let R = [a, b] X [c, d] be a closed (bounded) rectangle, and con-
sider the four boundary segments:

Y3
(a,d) (b,d)
'-.
/

Y4/'-. R / f' Y2

'-.
(a,c) / (b,c)
YI

In formulas,
'Yl(t) = (t, c), a =5 t=5 b; 'Y2(t) = (b, t), c=5t=5d;
'Y3(t) = (t, d), a =5 t=5 b; 'Y4(t) = (a, t), c=5t=5d.
We will need Green's theorem for a rectangle (see Appendix B).
This says that if 00 is a I-form on an open set containing the rectangle
R, then

where

roo = f oo+f oo-foo-f oo.

JaR "'II "'12 "'13 "'14

We will need only the following consequence:

Lemma 1.11. If dw = 0, then faRW = 0, i.e.,

1 +1 =1 +1
~
00
~
00
~
00
~
00 •

We can apply this to show that, on the plane, or a half plane, or

any rectangle, the necessary condition doo = 0 is actually sufficient
for integrals to be path-independent:

Proposition 1.12. Let U be a product of two open finite or infinite

intervals, i.e.,
U = {(x,y):a<x<b and c<y<d},
12 1. Path Integrals

with -00:5 a < b:5 00 and -00:5 c < d:5 00. If w is any Ilorm on U
such that dw = 0, then there is a function f on U with w = df.
Proof. Fix a point Po = (Xo,Yo) in U. For P = (x,y) in U, letf(P) = f'Yw,
where "( = "(I + "(2 is the path shown:

Y2
(xo,Y) (x,y)
"-
/

YI/ I'

(xo,yo) (x,yo)

Assume for the moment that x ~ Xo and y ~ Yo. The formulas are
"(I(t) = (xo, Yo + t), 0:5 t:5 Y - Yo, and "(2(t) = (xo + t, y), O:s t:5 x - Xo.
The last calculation in the proof of Proposition 1.8 shows that afI ax = p,
where w = pdx + qdy. If y <Yo, the same is true, replacing "(I(t) by
(xo, Yo - t), 0:5 t:s Yo - y; and similarly if x < Xo, replace "(2(t) by
(xo - t,y), 0:5 t:5xo - x.
Similarly, define a function g by g(P) = f 'Y. w, where "(* is the path
that first goes horizontally from (xo, Yo) to (x, Yo) and then goes ver-
tically from (x,Yo) to (x,y). The same argument, again left as an ex-
ercise, shows that aglay = q. Our assumptions on U imply that the
closed rectangle with opposite comers at Po and P is contained in U,
so that Green's theorem can be applied, and it follows from Lemma
1.11 thatf(P) is equal to g(P). It follows that aflax = p and aflay = q,
which means that df = w. 0

Exercise 1.13. Show that the proposition is also true when U is the
inside ofadisk, i.e., U= {(x,y): (x- a)2 + (y - b)2<r}. Can you prove
it when U is any convex region, or any starshaped region? (Convex
means that the straight line between any two points in the region is
contained in the region, and starshaped means that there is a point Po
in the region such that for any point P in the region, the straight line
from Po to P is contained in the region.)

A I-form w is called closed if dw = 0, and it is exact if w = dffor

some functionf. So all exact forms are closed, and the last proposition
says that, when U is a rectangle, all closed forms are exact. There
are many other regions U for which this is true, besides rectangles
and those in the preceding exercise. For example, if U is the union
Ic. A Criterion for Exactness 13

of any two rectangles, each as in the proposition, then any closed 1-

form w in U is exact.

To see this, let U I and U2 be the two open rectangles whose union
is U. By the proposition, there are functionsfl in UI andJi in U2 such
that dfl = w on UI and df2 = w on U2. Since UI n U2 is connected,
and since d(Ji - fl) = w - w = 0 on UI n U2, it follows that f2 - fl is
a constant function on UI n U2. If we replacef2 by f2 - c, where c is
this constant, we can assume}; andf2 agree on U I n U2 • This means
that there is a functionfon U= UI U U2 whose restriction to UI isfl
and whose restriction to U2 isf2. Moreover, df= won all of U, since
this condition is a local condition, to be verified at each point of U;
and every point is either in UI or U2, where we know dfl = w and
dJi = w.
In fact, this argument proves:

Lemma 1.14. Suppose UI and U2 are open sets, and UI n U2 is con-

nected. Let U = UI U U2, and let w be a I-form on U. If the restric-
tions of w to Uland U2 are both exact, then w is exact on U. D
The connectedness of U I n U2 is crucial for this. For example, sup-
pose U I and U2 are the regions indicated, with the shaded overlap,
and the origin outside in the middle.

Let w be the restriction of W{j to U, then W{j is the differential of a

function on each of the regions, but not on their union, as can be seen
by integrating w along a path around the origin in U.
Exercise 1.15. Show that if U is a union of open sets U I , •••• Un.
14 1. Path Integrals

and w is a I-form on U such that the restriction of w to each Ui is

exact, and (U I U U2 U ... U Ui) n Ui+1 is connected for I ::5 i::5 n - I,
then w is exact on U.

If w is a closed I-form in an open set U, although w may not be

exact on all of U, it follows from Proposition 1.12 that it is always
locally exact. That is, any point in U has a neighborhood (say rect-
angular), so that the restriction of w to this neighborhood is the dif-
ferential of a function. This can be used to calculate path integrals,
by cutting the path into pieces, on each of which Proposition 1.4 can
be applied:

Proposition 1.16. Ifw is a closed I-form on U, and y: is

[a,b]~U
a smooth path, then there is a subdivision a = to < tl < ... < tn = b
and a collection of open subsets UI> ... , Un of U so that y maps
[ti-I> til into Ui, and the restriction of w to U i is the differential of a
function J;. Let Pi = y(t;). Then, for any such choices,

Lw = (f1(P1) - ft(P o»+ (fiP 2) - (f2(Pd)

+ ... + (fn(P n ) - fn(P n - I ».

Proof For each point P in 'V([a, b]), choose a neighborhood Up of P
on which the restriction of w is exact. The open sets 'V-I(Up) form an
open covering of the compact interval [a, b], so a finite number of
them cover the interval. From this it is not hard to construct the sub-
division. One quick way to do it is to use the Lebesgue covering
lemma (see §A4 of Appendix A), which guarantees that, if the sub-
division is small enough, each subinterval will be mapped into one
of the neighborhoods Up. Having fixed such a subdivision, choose
one of these open sets containing the image of [ti- I, til, call it Ui' and
choose a functionJ; on Ui with dJ; = won Ui. Let 'Vi: [ti-I> t;]~ Ui be
the restriction of 'V to [ti - I> t;]. Then

f~w=fw+fw+···+fw
, 'YI 'Y2 'Yn
n n
= 2: (J;('Vi(t;)) - J;('Vi(ti- I») = 2: (J;(P;) - HP i- ».
i=1 i=1
I 0

The following two lemmas will be used in Part III to prove the
Jordan curve theorem. They are special cases of general theorems to
be proved in Chapter 9, but we can prove them directly with the
Ie. A Criterion for Exactness 15

methods of this section. For any positive number r, let 'YP,T be the
counterclockwise circle of radius r about P:
'Ypit) = P + r(cos(21Tt), sin(21Tt)), O:5t:51.

Lemma 1.17. Suppose U = ~2\{p}, and let r>O. Ifw is a closed

110rm on U such that f ¥P.r W = 0, then w is exact.
Proof. Let U\> U2 , U3 , and U4 be the half planes to the right of P,
above P, to the left of P, and below P. By Proposition 1.12, there
are functions j; on U j , unique up to the addition of constants, with
dj; = w on U j • By adjusting the constants, we may assume 12 =II on
UI n U2, and.f3 =f2 on U2n U3 , and!4 =13 on U3 n U4. Then 14 =II + c
on U4 nUl for some constant c. By Proposition 1.16, f,,/p.rw = c. The
hypothesis implies that c = 0, which means exactly that the four func-
tions j; agree on overlaps, so define a function I on U such that
df=w. 0

Lemma 1.IS. Suppose U = ~2 \ {P, Q}, and let 0 < r < Ip - QI. II w
is a closed 110rm on U such that f ¥P.r W = 0 and f ¥Q.r W = 0, then w
is exact.
Proof. The proof is similar. Suppose for definiteness that Q is located
to the southwest of P. Let UI be the half plane to the right of P, and
choose II on UI so that w = dll on UI • Let U2 be the half plane above
P, and choosef2 on U2 so that w = dfz on U2 , and 12 =lIon UI n U2.
Let U3 be the quarter plane to the left of P and above Q, and choose
h on U3 so that w = dl3 on U3, and.f3 =J2 on U2n U3 •

U2

Let U4 be the quarter plane to the right of Q and below P, and

choose!4 on U4 so that w = dJ4 on U4 , and J4 =h on U3 n U4, Then
J4 and JI differ by a constant on U4 nUl, and the hypothesis that
f,,/p,rw = 0 implies as in the preceding lemma that this constant is O.
Let Us be the half plane to the left of Q, and choose Js on Us such
that w = dis on Us and is = 13 on U3n Us. Let U6 be the half plane
16 1. Path Integrals

below Q, and choose 16 on V 6 such that w = dl6 on V 6 and 16 =Is on

Vs n V 6 • The hypothesis that LQ " W = 0 implies that/6 =14 on V 4 n V 6 •
The functions j; on Vi agree on overlaps, defining a function I on V
such that w = dl, which completes the proof in this case.
The proof when Q is located southeast of P, or the special cases
when P and Q are on a horizontal or vertical line, are similar and left
to the reader. D

Problem 1.19. Generalize the preceding lemmas from one or two to

n points.

For any point P = (xo,Yo), define the I-form wP,~ on ~2 \ {P} by the
formula
-(y - Yo) dx + (x - xo)dy
(x - XO)2 + (y - YO)2

Problem 1.20. For any two points P and Q, show that the I-form
w = WP.~ - wQ,~ is exact on ~2 \ L, where L is the line segment from
P to Q. Challenge. Find a function whose differential is w.
CHAPTER 2

Angles and Deformations

2a. Angle Functions and Winding Numbers

Any point in the plane can be expressed in polar coordinates, i.e., it
can be written in the form (r cos({}), r sin({}» for some r ;::: 0 and some
real number {}. The radius r is unique, being the distance from the
origin, or the square root of the sums of the squares of the Cartesian
coordinates. At the origin, r = 0, and {} can be any number. We often
denote the origin simply by 0 instead of (0,0). Except for the origin,
the angle {} is determined only up to adding integral multiples of 2'iT.
We call any of these numbers an angle for the point.
Suppose 'Y: [a, b] ~ ~2 \ {O} is a «6" path to the complement of the
origin, given in Cartesian coordinates by 'Y(t) = (x(t),y(t)). We want
to describe this in polar coordinates, that is, to find «6" functions r(t)
and {}(t) so that
(2.1) 'Y(t) = r(t)(cos({}(t», sin({}(t») = (r(t) cos({}(t» , r(t)sin({}(t)))
for all a:$ t:$ b. There is no problem with the function r: it is the
distance from the origin, defined by
r(t) = ii'Y(t)ii = V X(t)2 + y(t)2.
The angle function is not so simple. At any time t, there are many
possible angles to choose, all differing by multiples of 2'iT. If we choose
them say all lying in the interval (-'iT, 'iT] they would be unique, but
then would not vary continuously if the point crosses the negative
x-axis.

17
18 2. Angles and Defonnations

The initial angle can be chosen arbitrarily. In other words, choose

any number {}a so that

'Y(a) = (x(a) , y(a» = (r(a) cos({}a) , r(a) sin({}a».

We will require that our function {}(t) satisfy the initial condition
{}(a) = {}a. Motivated by the discussion in § Ib, we have a candidate
for the derivative {}'(t), which is the rate of change of angle: it should
be

-y(t)x'(t) + x(t)y'(t) -y(t)x'(t) + x(t)y'(t)

{}' (t) = = --"----::----'-'-'--~
X(t)2 + y(t)2 r(t)2

So we can define {}(t) to be the unique function with this initial con-
dition and this derivative. That is, we define {}(t) by

{}(t) -
- {}a +
11 -y(T)X'(T) + X(T)y'(T) dT.
2
a r(T)

Proposition 2.2. With these definitions, the functions r(t) and t)(t)
are ~oo functions, and equation (2.1) is satisfied for all t in [a, b] .

Proof. The function r(t) is ~oo since it is a composite of ~oo functions,

and {}-(t) is ~oo since it is the integral of a C{6oo function. Let

u(t) = (cos({}(t»,sin({}-(t») and v(t) = (-sin({}(t»,cos({}(t))).

These are perpendicular unit vectors for all t. We want to show that

I
- 'Y(t) = u(t)
r(t)

for all t. We are assuming that they are equal for t = a. It suffices to
show that both sides of this equation have the same dot product with
vectors u(t) and v(t), since the difference (l/r(t»'Y(t) - u(t) would then
be perpendicular to two independent vectors, and so would be zero.
For the right side of the equation u(t), these dot products are iden-
tically I and 0, respectively, so it suffices to prove that the same is
true for the left side. Since we know the equality of the vectors for
t = a, it suffices to prove that the derivatives of these dot products
2a. Angle Functions and Winding Numbers 19

vanish. We are therefore reduced to showing that

!!.- (_1 'Y(t). U(t») == 0 and !!.- (_1 'Y(t)· V(t») - O.

dt r(t) dt r(t)

These are simple verifications, using the usual rules of calculus.

Omitting the variable t, and writing f in place of r'(t), etc., the first
of these derivatives is

rx - xf rj - yf x . y .
-,;.- cos({}) + -,;.- sin({}) + f {}( - sin({}» + ; {} (cos({}»

1 . .
= ? (cos({})[';'x - xrf + y';'{}] + sin({})[';'y - yrf - x';'{}]).

Using the identities rf = xi + yy and r2,ft = -yx + xj, one sees that the
terms in brackets vanish. The proof that the other derivative vanishes
is similar and left as an exercise. 0

Exercise 2.3. (a) Show that this function {}(t) is unique, and that in
fact it is the only continuous function with {}(a) = {}a such that (2.1)
holds. (b) Show that if {}a is replaced by {}a + 2'TTn for some n, then
the corresponding angle function is {}(t) + 2'TTn.

Using this angle function, we can define the (total signed) change
in angle of the path 'Y to be {}(b) - {}(a). Note by the preceding ex-
ercise that this is independent of the choice of initial angle. Equiva-
lently, this change in angle is

ib-y(t)X'(t)+X(t)y'(t) dt =
a r(t)2
r -ydx+xdy
J-y ~ +l
_ r
- J/)a.
If a segmented path does not pass through the origin, i.e., it is a
path in 1R2 \ {O}, we want to define its winding number, which should
be the "net" number of times 'Y goes around the origin, counting the
counterclockwise motion as positive, and the clockwise motion as
negative. In other words, it is the total signed change in angle, di-
vided by 2'TT. We will denote it by W('Y,O), the "0" indicating that
we are going around the origin. What we have just done shows what
the definition should be. We define the winding number of'Y around
20 2. Angles and Deformations

the origin by

W('Y,O) = f
_1 w{\
21T "Y
= f
_1 -ydx + xdy
21T"Y x 2 + l .

Proposition 2.4. For any closed segmented path that does not pass
through the origin, the winding number of the path around the origin
is an integer.

Proof. The fonnula given before the proposition defines a number

W('Y, 0) for any segmented path 'Y that does not pass through the or-
igin. Suppose 'Y starts at P and goes to Q, and we choose an angle
{}p for P. We claim that {}p + 21TW('Y, 0) is an angle for Q. This will
prove the result, for when P = Q it says that {}p and {}p + 21TW('Y, 0)
define the same angle, so they must differ by an integer multiple of
21T.
It suffices to prove the claim for a smooth path, since if 'Y is a sum
of smooth paths, the assertion for each of them implies it for the sum.
But when 'Y: [a.b]--,)~2\{0} is smooth, {}p+21TW('Y,0) is just the
value of our angle function {}(t) at t = b, starting with {}(a) = {}p, so
the result follows from Proposition 2.2. 0

Exercise 2.5. Use Proposition 1.16 to give another proof of this

proposition.

Exercise 2.6. Let 'Y(t) = (kcos(nt), ksin(nt)), 0 ~ t ~ 21T, where k is a

positive number and n is an integer. Show that W('Y, 0) = n.

In earlier centuries, before modem rigor required functions to be

single-valued, the I-fonn w{\ would have been written as the differ-
ential d{} of the multivalued function {}. (In fact, this is still a common
and useful notation for this I-fonn, as long as one realizes that {} is
not a function, so that Proposition 1.4 is not contradicted!) The graph
of this multivalued function can be visualized in 3-space, as the locus
of points (x. y. z) of the fonn (rcos({}), rsin(b), {}) for some r> 0 and
real number {}. This is closely related to the polar coordinate mapping

p: {(r,{}): r>O} --,) ~2\{O}, (r, {}) ~ (rcos({}), rsin({}».

Picturing these together:

2a. Angle Functions and Winding Numbers 21

1 (x,y,z) t----> (x,y)

This illustrates the fact that, although the distance from the origin
is a continuous function on ~2 \ {O}, the counterclockwise angle from
the x-axis cannot be defined continuously. What we have proved,
however, is that one can define such an angle continuously along a
curve. Proposition 2.2 (with Exercise 2.3) can be expressed geomet-
rically as follows:

Corollary 2.7. Let 'Y: [a, b] ~ ~2 \{O} be a smooth path with starting
point Pa = (racos(tJa), rasin(tJa». Then there is a unique smooth path
1: [a, b]~ R, where R = {(r, tJ): r> O} is the right half plane, with
starting point (ra, tJa), and with po 1= 'Y. D

We say that the path 'Y is the lifting of the path 'Y with starting point
(ra,1}a)' On the picture,
22 2. Angles and Deformations

•
! •
a b "(

Problem 2.8. Show that the locus of points (r cos({}), r sin({}), {}) is
one connected component of the surface in the complement of the
z-axis in 1R3 defined by the equation y' cos(z) = X· sin(z).
Exercise 2.9. Show that every point in 1R2 \ {O} has a neighborhood
V such that p-l(V) is a disjoint union of open sets Vi' each of which
is mapped homeomorphic ally (in fact, diffeomorphic ally) by p
onto V.
This exercise verifies that p is a "covering map," a class of maps
we will study in Chapter 11. Corollary 2.7 is a special case of a
general theorem about covering maps.
Winding numbers can be described around any point P = (xo, Yo) in
place of the origin. One can do this either by translating everything,
or directly by using the form
-(y - Yo) dx + (x - xo)dy
(x - XO)2 + (y - YO)2
2b. Reparametrizing and Defonning Paths 23

Exercise 2.10. For any closed segmented path -y: [a, b]~ ~2 \ {P},
define the winding number W(-y,P) of -y around P by the formula

W(-y,P) = _1 f wp,,'t.
2-rr J..,
Generalize the assertions of this section to these winding numbers.

2b. Reparametrizing and Deforming Paths

Path integrals do not depend on the parametrization of the path, in
the following sense. Suppose w is a I-form on an open set U in the
plane, and -y: [a, b] ~ U is a smooth path, and suppose IP: [a', b'] ~ [a, b]
is a C(6'" function (as usual, extending to a neighborhood of [a', b'D
that maps a' to a and b' to b. The path -y ° IP is called a reparame-
trization of -y.

Lemma 2.11. With these assumptions,

f
J..,.'!'
w f
= . , W.
Proof. Write w = pdx + qdy, and -yet) = (x(t),y(t)), and calculate, us-
ing the chain rule and change of variables formulas:

J..,.", w = f' [p(x(IP(s)),y(IP(s)))' (xoIP)'(s)

+ q(x(IP(s)),y(IP(s)))' (yoIP)'(s)] ds
f' [p(x(IP(s)) , y(IP(s))) . x' (IP(s))

+ q(x(IP(s)) , y(IP(s))) . y'(IP(s))] . (IP)'(s) ds

f [p(x(t) , yet)) . x'(t) + q(x(t),y(t)) . y'(t)]dt = J.., w. D

Exercise 2.12.1f-y: [a,b]~U is a path in U, let -y-I: [a,b]~U be

the same path traveled backward: -y-I(t) = -y(a + b - t). Show that

Show more generally that if IP: [a', b'] ~ [a, b] is any C(6'" function that
maps a' to b and b' to a, then f..,o'!'w = - f..,w.
24 2. Angles and Deformations

The following problem shows how one could avoid dealing with
segmented paths: by turning around the comers very slowly!

Problem 2.13. (a) Construct a '(6"" mapping 'P from [0,1] to [a, b]
taking 0 to a and 1 to b, such that all derivatives of'P vanish at 0
and at 1. (b) Use this to show that any smooth path can be repara-
metrized to a path with the same endpoints but such that all deriva-
tives vanish at the endpoints. (c) If "YI: [a, b] ~ U and "Y2: [b, c] ~ U
are '(6"" paths with "YI(b) = "Y2(b), and all derivatives of"Y1 and "Y2 vanish
at b, verify that the path "Y: [a, c]~ U that agrees with "YI on [a, b]
and "Y2 on [b, c] is a '(6"" path. (d) If"y ="YI + ... +"Yn is a segmented
path, show that there is a '(6"" path "Y*: [0, n] ~ U, so that the restric-
tion of "Y* to [k - 1, k] is a reparametrization of "Yk for each k.

Problem 2.14. Given a I-form 00 on an open set U, show that the

following are equivalent: (i) there is a function f on U with df= 00;
(ii) I'Y 00 = I8 00 whenever "Y and 8 are '(6"" paths in U with the same
initial and final points; and (iii) I'Y 00 = 0 whenever "Y is a '(6"" closed
path in U (i.e., with the initial point equal to the final point).

Problem 2.15. Given a I-form 00 on an open set U, show that the

following are equivalent: (i) doo = 0; (ii) I.Roo = 0 for all closed rect-
angles R contained in U; and (iii) every point in U has a neighborhood
such that I.Roo = 0 for all closed rectangles R contained in the neigh-
borhood. Is the same true if rectangles are replaced by disks?

Next we tum to the question of what happens when the path is

moved, or deformed. We consider first the case of a deformation of
paths with fixed endpoints. This will be given by a family of paths
"Y.. for simplicity all defined on the same interval [a, b], with the pa-
rameter s varying in another interval which we take to be the unit
interval [0, 1]. We will assume this is a smooth family, in the sense
that the coordinates of the point "Yit) are smooth functions of both s
and t. This means that we are given a mapping H from [a, b] X [0, 1]
to U, which we assume is '(6"" (this, as usual, means that the two
coordinate functions can be extended to be infinitely differentiable
functions on some open neighborhood of the rectangle). We assume
that H(a, s) = P and H(b, s) = Q for all 0:'5 s:'5 1. Set "Yit) = H(t, s),
so each "Ys is a path in U from P to Q.
2b. Reparametrizing and Deforming Paths 25

-H

o
a b

We call H a smooth homotopy from the path 'Yo to the path 'YI' We
say that two paths from an interval [a, b] to U, with the same initial
and final points, are smoothly homotopic in U if there is such a ho-
motopy from one to the other.

Proposition 2.16. If 'Y and l) are smoothly homotopic paths from P

to Q in an open set U, and w is a closed I-form in U, then

Proof. First we sketch a proof using only ideas from calculus. Let V
be a neighborhood of the rectangle R = [a, b] X [0, 1] mapped into U
by an extension of H, and let x(t, s) and y(t, s) be the coordinate func-
tions of this mapping from V to U. Define a "pull-back" form w* on
V by the formula

w* = (P(X(t, s), y(t, s» ax + q(x(t, s), y(t, s» ay ) dt

at at

+ (P(X(t, s), y(t, s» ax + q(x(t, s), y(t, s» ay ) ds.

as as
A little calculation, left to you, shows that dw* = 0 on V. By Green's
theorem, we therefore know that the integral of w* around the bound-
ary of R must be zero. Simpler calculations show that the integral of
w* along the bottom of the rectangle is J'/ w, that along the top is
Jaw, and that the integrals along the two sides are zero. Putting this
all together gives the required equality.
Here is another proof, more topological in flavor. Each point in
the image H(R) of the rectangle has a neighborhood on which w is
exact. Applying the Lebesgue lemma, it follows that, if we subdivide
the rectangle small enough, by choosing
a= to < tl < ... < tn = b and 0 = So < Sl < ... < Sm = 1 ,
then each subrectangle RiJ = [ti-I> til x [Sj_1> sJ is mapped by H into
an open set Ui,j on which w is the differential of some function fiJ.
26 2. Angles and Defonnations

R3,4
f
I Q

- H

Since w is the differential of a function on Ui,j' the integral of w 0 H

around the boundary of R i,j is zero, i.e., the integral along the bottom
and right side of R i,j is the same as the integral along the left side and
top. Now J-yw is the integral of w along the bottom and right side of
the original rectangle, and one can successively replace integrals over
the bottom and right sides by integrals over the left and top sides, of
each of the small rectangles, until one has the integral over the left
and top sides of the whole rectangle, which is Jsw.

More succinctly , integrate w over all the boundaries of the small

rectangles. Each inside edge is integrated over twice, once in each
direction, which leaves the integrals over the outside, giving

0=2:( w=( w=fw-(w .

iJ JaRij JaR "Y Jr. o
There is another kind of deformation or homotopy that is also im-
portant. This is a deformation of closed paths, through a family of
closed paths, but allowing the endpoints to vary , It is given by a ~oo
mapping H from a rectangle [a, b] x [0,1] into U, with the property
that H(a , s) = H(b , s) for every s in [0, 1]. Each of the paths "is given
by "iit) = H(t, s) is a closed path, starting and ending at the point
'T(s) = H(a, s) = H(b , s). These endpoints are allowed to vary along
the path 'T .

(a, 1 ) (b,l)

~ -
H

(a , O) (b , O)
2c. Vector Fields and Fluid Flow 27

We call H a smooth homotopy from the path 'Y = 'Yo to the path
8 = 'Y .. and we say that two closed paths 'Y and 8 are smoothly ho-
motopic if there is such a smooth homotopy between them.

Proposition 2.17. If y and {) are smoothly homotopic closed paths in

an open set U, and w is a closed I-form in U, then

Proof. Either of the proofs of the last proposition works for this one.
The only point to notice is that the integrals of W over the two sides
of the rectangle may not be zero, but since the two paths given by H
on these two sides are the same (namely or), these integrals fTW can-
cel. The details are left as an exercise. 0

When U is the complement of a point P, and W = (l/2'll')Wp,lb this

specializes to the important:

Corollary 2.18. If y and {) are smoothly homotopic closed paths in

1R 2 \{P}, then W(y,P)=W({),P).

Of course, it is crucial for this that the homotopy stays in the com-
plement of the point P!

Problem 2.19. Prove that being smoothly homotopic is an equiva-

lence relation.

2c. Vector Fields and Fluid Flow

We defined a I-form on an open set U in the plane to be a pair of
smooth functions p and q on U. Such a pair of functions can also be
identified with a vector field on U, which assigns to each point (x, y)
in U the vector
V(x, y) = p(x, y)i + q(x, y)j .
28 2. Angles and Deformations

The vector field corresponding to the I-form Wa is perpendicular

to the position vector, pointing in a counterclockwise direction, with
length that is the inverse of the distance to the origin:

The path integral

!'Y
pdx+ qdy = Ib (P(X(t),y(t»
a
dx
dt
+ q(x(t),y(t» dy ) dt
dt
can be written, using the dot product, as

f V('Y(t»· 'Y'(t) dt = f (V('Y(t» 'II~:~:~II) 11'Y'(t)11 dt,

which is the integral of the projection of the vector field along the
tangent to the curve. (For this last formula, we must assume the tan-
gent vector is not zero.) If the vector field represents a force, it is the
work done by the force along the curve.
The translation of the equation W = dj, or p = aj/ ax and q = aj/ ay,
into vector field language says that the corresponding vector field is
the gradient of f
aj aj
grad(f) = - i + -j.
ax ay
The function j (or sometimes - f) is called a potential function.
We'll finish this chapter with a quick sketch of the interpretation
of these ideas for case where the vector field gives the velocity of a
fluid flowing in an open set in the plane. So V(x,y) is the velocity
vector at (x,y). We assume the flow is in a steady state, which means
that, as written, this velocity vector depends only on the point (and
not on time). Let p and q be the components of V, as above.
2c. Vector Fields and Fluid Flow 29

The integral f -y p dx + q dy represents the circulation of the fluid along

the path 'Y, per unit of time. This can be seen from the interpretation
as the integral of the projection of the velocity vector in the tangent
direction along the curve. There is another important integral,
f-yqdx - pdy, that represents the flUX of the fluid across the path 'Y
(per unit of time). To see this, let N(t) = (-dy jdt, dxjdt) be the nor-
mal vector, which is perpendicular to and 90° counterclockwise from
the tangent vector.

Then

= L q(x(t),y(t))-dxdt - p(x(t),y(t))-ddtY) dt
b

a
(

f V('Y(t))· N(t) dt = f (V('Y(t)) 'I;~;~II) "'Y'(t)" dt.

This is the integral of the projection of the velocity vector in the nor-
mal direction, which measures the flow across 'Y, from right to left
in the direction along 'Y, per unit of time.
The flow is called irrotational if the circulation around all small
closed loops is zero. By Problem 2.15, this is true precisely when
dw=O, i.e., aqjax-apjay=o. The function aqjax-apjay is called
the curl of the vector field.
The flow is called incompressible if the net flow across small closed
loops is zero. Applying Problem 2.15 to the I-form qdx - pdy, this
is equivalent to the condition that apjax + aqjay = O. The function
ap j ax + aq j ay is called the divergence of the vector field.

Exercise 2.20. If R is a rectangle (or disk) in U, show that the integral

of the curl (identified with the 2-form (aqjax - apjay)dxdy) over R
is the circulation around aR, and the integral of the divergence is the
flux across aR.

If the open set U is a simple one, such as an open rectangle or disk,

30 2. Angles and Deformations

and if the flow is irrotational, then we know that there is a potential

functionj, i.e., V= grad(f). If the fluid is also incompressible, the
equation we just found says that

01 01
- 2+ - = 0
ox ol '
which is the condition for j to be a harmonic function.
Although we have no topological excuses for them, here are a handful
of typical applications of and variations on these notions.

Exercise 2.21. (a) Show that the following functions are harmonic:
(i) a + bx + cy for any a, b, c; (ii) x2 -l; (iii) log(r), r = V xl + y2
on the complement of the origin. (b) Find all harmonic polynomial
functions of x and y of degree at most three. (c) Find all harmonic
functions that have the form h(r).

Problem 2.22. (a) State and prove an analogue of Green's theorem

when R is the region between two rectangles, one contained in the
other. (b) State and prove an analogue of Green's theorem when R
is a disk, or the region between two concentric disks.

Problem 2.23. If R is a region, with boundary dR, for which Green's

theorem is known, prove the following two formulas of Green, for
functions j and g on a region containing the closure of R:

(i) 1 ( ogoy
aR
ag )
j. --dx + -dy
ax
= fLv· (:~ + ::~) + grad(f)' grad(g») dxdy;

(ii) 1 ( ogoy
aR
Og) - g. (a
j- --dx + -dy
ox
j OJ)
--dx + -dy
oy ox
= II (
R
(02g 02g)
j. -2+- -g. - + -
ox ol
(01 (1)) dxdy.
ox2 ol
Problem 2.24. (a) If j is harmonic in R and vanishes on fiR, show
that j must be identically zero. (b) If two harmonic functions on R
have the same restriction to oR, show that they must agree everywhere
on R.
2c. Vector Fields and Fluid Flow 31

Problem 2.25. If f is harmonic on a disk, show that the value of f

at the center of the disk is the average value of f on the boundary of
the disk.

Exercise 2.26. If V(x,y) = -(1/27r)(x,y)/II(x,y)1I2 and 'Y is a closed

path not passing through the origin, show that the flux across 'Y is the
winding number W('Y,O), i.e., W('Y, 0) = f:V('Y(t»· N(t)dt.

These physical notions suggest a way to generalize the notion of a

winding number to higher dimensions. Iff is a C(6'" map from a rect-
angle or disk R to 1R3 \ (0), let Nf(P) be the cross product of the col-
umns of the Jacobian matrix of f at P, and let
V(x,y,z) = (l/47r)(x,y,z)/II(x,y,z)W.
It is a good project to develop a notion of an "engulfing number,"
setting W(f,O) = ffR V(ftP»·Nf(P).
 PART II

WINDING NUMBERS

The notion of winding numbers is generalized to arbitrary continuous

paths, and the facts we proved in the smooth case using calculus are
proved here by purely topological arguments. We also look at what
happens when the point being wound around is varied. Finally, we
use the idea of winding numbers to define the degree of a mapping
from one circle to another, and to define the local degree of a mapping
from one open plane set to another.
In Chapter 4 there are several applications of winding numbers,
some written out, and many left as exercises and problems. Many of
these results generalize to higher dimensions; the names attached to
some of them refer to these generalizations.
CHAPTER 3

The Winding Number

3a. Definition of the Winding Number

For any point P in the plane, and any sector with vertex at P, we can
define a continuous (even «6") angle function on the sector, although
the choice is unique only up to adding integral multiples of 21l'. Here
angles are measured with reference to P, counterclockwise from the
horizontal line to the east of P:

\
(x,y)
\
\
\
\
\

If p = pp is the corresponding polar coordinate map:

p(r,{}) = P+(rcos({}),rsin({}»,
a sector is the image of a strip {(r, (}): r> 0 and {}! < {} < {}z}, where
{}! and {}2 are any real numbers with 0 < {}2 - {}! :::; 21l'. The fact that
this {} is a continuous function on the sector can be seen directly,

35
36 3. The Winding Number

using formulas like the arc tangent, or topologically as follows: p is

one-to-one and continuous from the strip onto the sector, and it maps
small open rectangles {( r, {t): a < r < b, a < {t < ~} onto small open
sectors (bounded by two straight lines and arcs of two circles), so pis
open as well as continuous; it is therefore a homeomorphism (in fact
a diffeomorphism). This means that rand {t are continuous functions
of x and y.
For any continuous path 'Y: [a, b]~ []~f \ {P}, we define its winding
number W( 'Y, P) as follows:

Step 1. Subdivide the interval into a = to::5 t 1 ::5 ...

::5 tn = b, so that
each subinterval [ti - h til is mapped into some sector with vertex at P.
Such a subdivision exists by the Lebesgue lemma, since each point
in the image of'Y is contained in some such sector.

Step 2. Choose such a sector Vi containing 'Y([t;-h ta) and a corre-

sponding angle function {ti on V;, for 1 ::5 i::5 n. Let Pi = 'Y(t;), 0 ::5 i ::5 n.
Define

W('Y,P)

Each term represents the net change in angle along that part of the
path.

Proposition 3.1. (a) The definition of a winding number is indepen-

dent of the choices made in Steps I and 2. (b) If Y is a closed path,
i.e., yCa) = yCb), then W(y,P) is an integer.
Proof. (a) To see that it is independent of the choices of the sectors
V; and the angle functions {t;, suppose V;' and {t;' were other choices.
Then {t; and {}-/ would differ by a constant (in fact, a mUltiple of 2"IT)
on the component of the intersection of V; and U;' that contains
'Y([t;-h t;]). So the difference in the values of {t; at the two points P;-l
and P; is the same as the difference in values of {t;' at these two
points, and adding over i shows that the winding number doesn't
change.
So it is enough to show that the definition is independent of the
3a. Definition of the Winding Number 37

choice of subdivision. Suppose we add one point to a given subdi-

vision that satisfies the condition in Step 1, say by inserting a point
t* between some ti- 1 and ti. If Vi and tti are chosen for [ti-I> ti], these
same Vi and tti can also be chosen for the two subintervals [tH , t*]
and [t*, til. And if p* = 'Y(t*), then
(tti(P i) - ttlP*» + (ttlP*) - tti(Pi- 1» = tti(P) - tti(Pi-l)'
so again the sum is unchanged. It follows that if we insert any finite
number of points into a given subdivision, the definition of the wind-
ing number will not change. But then, if two subdivisions both satisfy
the condition in Step I, the common refinement of both of them, ob-
tained by including all division points for each, will define the same
winding number as each of them.
(b) In general, the claim is that, even when 'Y is not closed, if tta
is an angle for the initial point 'Y(a) , then tta + 2'IT· W('Y,P) is an angle
for the endpoint 'Y(b). When the path is closed, this implies that W('Y,P)
is an integer. This claim is evident for the restriction of 'Y to each
subinterval [tH' til on which there is a continuous angle function, and
the general case follows by adding up the results (or inducting on the
number of subintervals). D
Exercise 3.2. Show that if 'Y is smooth, this definition agrees with
that in Chapter 2.

Exercise 3.3. Show that if 'Y: [a, b] --') V is a closed path, and V is
an open set in ~2 \ {P} on which there is a continuous angle function
(for example, a sector with vertex at P), then W('Y, P) = O.

Exercise 3.4. Show that winding numbers are invariant by transla-

tion, in the following sense. Let 'Y: [a, b] --') ~2 \ {P}, and let v be any
vector in the plane. Let'Y + v be the path defined by ('Y + v)(t) = 'Y(t) + v.
Show that
W('Y+v,P+v) = W('Y,P).

Problem 3.5. Show that for any continuous path 'Y: [a,b]--')~2\{p},
there are continuous functions r: [a, b] --') ~+ (the positive real num-
bers) and tt: [a,b]--')~, so that
'Y(t) = P+(r(t)cos(tt(t»,r(t)sin(tt(t))), a5t5b.

Show that r is uniquely determined, and tt is uniquely determined up

to adding a constant integral multiple of 2'IT. Show in fact that
r(t) = 11'Y(t) - pll, and if 'Y' denotes the restriction of 'Y to the interval
38 3. The Winding Number

[a, t] (so "/(u) = 'Y(u) for a:5 u:5 t), and -3a is an angle for 'Y(a), then
one may take
-3(t) = -3 a + 2'TT' W('Y t , P).
Equivalently, for any choice of -3a , there is a unique continuous map-
ping ')i: [a,b]~{(r,-3): r>O} such that ppo')i ="1 and ')i(a) = (r(a),-3a ),
where pp is the polar coordinate mapping. Such a ')i is called a lifting
of "I with starting point (r(a) , -3a ).

3b. Homotopy and Reparametrization

Suppose R = [a, b] X [c, d] is a closed rectangle, and f: R~ ~2 is a
continuous mapping. The restrictions of f to the four sides of the
rectangle define four paths "II, "12, "13, and "14:
'YI(t) = f(t, C), 'Y3(t) = f(t, d), a:5t:5b;
'Y2(S) = f(b, S), 'Y4(S) = f(a, s), c:5s:5d.

r
--------
/
'YI

Theorem 3.6. For any point P not in f(R),

W('Yh P)+W('Y2,P) = W('Y3,P) + W('Y4,P),
Proof. In fact, the second proofs we gave for Proposition 2.16 and
2.17 work equally well in this case. As there, use the Lebesgue lemma
to subdivide the rectangle into subrectangles Ri .j , such that f maps
R;.j into a sector Ui,j (with vertex at P), on which there is a continuous
angle function -3 iJ • Then, by the canceling of inside edges as before,
W(flaR, P) = 2: W(flaR
iJ
iJ , P) ,

where W(fl aR , P) is defined by the equation

W(flaR, P) = W('Yh P) + W('Y2, P) - W('Y3, P) - W('Y4, P),
3b. Homotopy and Reparametrization 39

and W(fiaRij' P) is defined similarly as the signed sum of the winding

numbers around the small rectangles. But these winding numbers around
the small rectangles are all zero, since each R i•j is mapped into a re-
gion where there is an angle function {}iJ' So
W('Yh P ) + W('Y2,P) - W('Y3,P) - W('Y4,P) = O. 0
This theorem implies that the winding numbers have the same in-
variance under homotopies as in the smooth case considered in §2b.
Again, there are two kinds of homotopies we want to consider, first
for paths with fixed endpoints, and second for closed paths. If
'Y: [a,b]~U and 8: [a,b]~U are paths with the same initial and
final points, a homotopy from 'Y to 8 with fixed endpoints is a con-
tinuous mapping H: [a, b] x [0, l]~ U such that
H(t,O) = 'Y(t) and H(t, 1) = 8(t) forall a::5 t::5 b;
H(a, s) = 'Y(a) = 8(a) and H(b, s) = 'Y(b) = 8(b) for all 0::5 s::5 I.
The paths 'Ys defined by 'Ys(t) = H(t, s) give a continuous family of
smooth paths from 'Yo = 'Y to 'Y I = 8. The paths 'Y and 8 are called
homotopic with fIXed endpoints if there is such a homotopy H.
On the other hand, if'Y and 8 are closed paths in U, again defined
on the same interval [a, b], a homotopy from 'Y to 8 through closed
paths is a continuous H: [a, b] x [0, l]~ U, such that
H(t,O) = 'Y(t) and H(t, 1) = 8(t) for all a ::5 t ::5 b ;
H(a, s) = H(b, s) for all 0 ::5 s::5 I .
The paths 'Y and 8 are called homotopic closed paths if there is such
a homotopy H.

Exercise 3.7. Prove that the relation of being homotopic with fixed
endpoints, or as closed paths, is an equivalence relation.

Corollary 3.8. If two paths 'Y and 8 in 1R2 \ {P} are homotopic. either
as paths with the same endpoints, or as closed paths, then
W('Y,P) = W(8,P).

Proof. This is an immediate consequence of the theorem, applied to

the homotopy H = f. In the first case, the winding numbers of the
constant paths from the sides of the rectangle are both zero, and in
the second case they are the same, so their winding numbers cancel
in~re~. 0

We next consider what happens to the winding number by a change

40 3. The Winding Number

of parameter, generalizing what we saw in the last chapter in the smooth

case.

Corollary 3.9. Let y: [a, b] ~ lI,f \{P} be a continuous path, and

cp: [c, d] ~ [a, b] a continuous function.
(a) If cp(c) = a and cp(d) = b, then W( yo cp, P) = W( y, P).
(b) If cp(c) = band cp(d) = a, then W( yo cp, P) = - W( y, P). In partic-
ular, if y-l(t) = y(a + b - t), a ~ t~ b, then
W('Y-1,P) = -W('Y,P).

Proof. For (a), define f: [a,b] x [c,d]~U by the formula

f(t, s) = 'Y(min(t + 'P(s) - a, b».
y(b)

Then f is continuous, since the minimum or composite of two con-

tinuous functions is continuous. The paths from the sides of f are
'Yl = 'Y, 'Y4 = 'Yo'll, and 'Yz and 'Y3 are constant paths at the point 'Y(b).
So Theorem 3.6 applies and gives (a). Similarly for (b), use
f(s, t) = 'Y(max(t + 'P(s) - b, a».
In this case 'Yl = 'Y, 'Yz = 'Yo'll, and 'Y3 and 'Y4 are constant. 0

Exercise 3.10. Give a direct proof of Corollary 3.9 from the defi-
nition of the winding number, in case the change of coordinates'll is
a monotone increasing or monotone decreasing function. (Monotone
increasing means that 'P(t) < 'P(s) if t < s.)

As another application, we have the "dog-on-a-Ieash" theorem of

Poincare and Bohl. This says if the leash is kept shorter than the
distance from the walker to the fire hydrant, then the walker and the
dog both wind around the hydrant the same number of times:
3b. Homotopy and Reparametrization 41

Ii(t)

~
Dog ------;;!t) vI

/' Man
/'

P /'
.....
Hydrant

Theorem 3.11 (Dog-on-a-Leash). Suppose y: [a, b]~ R2 \{P} and

5: [a, b] ~ R2 \ {P} are closed paths, and the line segment between
y(t) and 8(t) never hits the point P. Then
W(,,(,P) = W(8,P).
Proof. Define H: [a, b] X [0, l]~ R2 by the formula
H(t,s) = (1- s)"(t) + s8(t), as tS b, 0 ss S l.
This is a continuous homotopy from "( to 8 through closed paths. The
hypotheses imply that r maps the rectangle into R2 \ {Pl. The result
therefore follows from Corollary 3.8. 0

Corollary 3.12. Ify: [a,b]~R2 and 8: [a,b]~R2 are closed paths

such that IIy(t) - 8(t)11 < IIy(t) - pil for all t in [a, b], then
W(y,P) = W(8,P).
Proof. The hypothesis implies that neither path hits P, and that the
line segment between "(t) and 8(t) doesn't hit P. 0

Exercise 3.13. Show that if U is an open rectangle (bounded or un-

bounded), then any two paths from [a, b] to U with the same end-
points are homotopic, and any two closed paths in U are homotopic.
Show that the same is true for any convex open set U. Can you prove
it when U is just starshaped?

Problem 3.14. Let"( and 8 be paths from an interval to R2 \ {P} with

the same endpoints. Show that the following are equivalent:
(i) "( and 8 are homotopic in R2 \ {P};
(ii) W(,,(, P) = W(8, P); and
(iii) if:Y and 8 are liftings of "( and 8 with the same initial point, as
in Problem 3.5, then :y and 8 have the same final point.

Problem 3.15. Let"( and 8 be closed paths from a closed interval to

42 3. The Winding Number

~2 \ {Pl. Show that 'Y and 3 are homotopic through closed paths in
~2 \ {P} if and only if W('Y,P) = W(3,P).

3c. Varying the Point

We want to study what happens to the winding number if a closed
path 'Y: [a, b] ~ ~2 is fixed, and the point P is allowed to vary, but
always so that the path does not pass through P. Let us denote the
image of the path 'Y by Supp('Y), and call it the support of 'Y, i.e.,
Supp('Y) = 'Y([a, b]).
Since the interval is compact, the support is a compact, and hence
closed and bounded, subset of the plane. The complement of the sup-
port is an open set, which may have many-even infinitely many-
connected components, each of which is open. Since the support is
bounded, however, there is one connected component of ~2 \ Supp('Y)
that contains all points outside some large disk; this component is
called the unbounded component.

Proposition 3.16. As afunction ofP, thefunction W('Y,P) is constant

on each connected component of ~2 \ Supp( 'Y). It vanishes on the un-
bounded component.

Proof. For the first statement, we must show that W( 'Y, P) is a locally
constant function of P in ~2 \ Supp('Y). Given P, choose a disk around
P contained in ~2\SUpp('Y). We must show that W('Y,P' ) = W('Y,P)
for all p' in the disk. Let v = p' - P be the vector from P to P'. By
Exercise 3.4,
W('Y,P) = W('Y+v,P+v) = W('Y+V,P').
The homotopy H(t, s) = 'Y(t) + sv, a:5 t:5 b, 0:5 S, :5 1, is a homotopy
through closed paths from 'Y to 'Y + v that never hits the point pI,
3d. Degrees and Local Degrees 43

so
W('Y + v, PI) = W('Y, PI).
To show that the winding number vanishes on the unbounded com-
ponent, it suffices to show it vanishes on one such point. For ex-
ample, we can take P far out on the negative x-axis, so that the sup-
port of 'Y is contained in a half plane to the right of P. But then there
is an angle function on this half plane, so the winding number is zero
by Exercise 3.3. 0

The following exercise gives alternative proofs of the proposition:

Exercise 3.17. (a) Prove directly from the definition that for any path
'Y, whether closed or not, the function P~ W('Y, P) is continuous on
the complement of Supp('Y), and approaches zero when Ilpll goes to
infinity. (b) Use (a) and the fact that, when'Y is closed, W('Y,P) takes
values in the integers to give another proof of the proposition. (c) For
a closed path 'Y, show directly that W('Y,P) is constant on path com-
ponents of ~2 \ Supp('Y) by applying Theorem 3.6 to the mapping
nt, s) = 'Y(t) - (1(s) and the point P = 0, where (1 is any path in
~2 \ Supp('Y).

3d. Degrees and Local Degrees

If I is any closed interval, and C is any circle, a continuous closed
path 'Y from I to an open set U is essentially the same thing as a

cO
continuous mapping from C into U:

I •
I·
1 •
/,
This can be realized explicitly as follows. Let us assume that 1= [0, 1],
44 3. The Winding Number

since this is the most common convention. Suppose C has center (Xo, Yo)
and radius r. Let <p: I~C be the function that wraps I around C:
<p(t) = (xo, Yo) + (rcos(2'TTt), rsin(2'TTt», 0:5 t:5 1.
So <p is a one-to-one mapping of I onto C, except that <p(0) = <p(1).
It follows that if F is a mapping from C into an open set U, then
'Y = F 0 <p is a mapping from I to U with 'Y(O) = 'Y( 1), and that any such
'Y can be realized in this way for a unique F.

Lemma 3.18. The mapping 'Y is continuous if and only if F is con-

tinuous.
Proof. In fact, <p realizes C as the quotient space of I with its end-
points 0 and 1 identified. Concretely, this means that a subset X of
C is open in C if and only if <p -I(X) is open in I. This is easy to verify
directly, or one can argue that the quotient space is a compact Haus-
dorff space, and the induced mapping from it to C, being continuous
and one-to-one, must be a homeomorphism. If follows that for an
open subset V of U, F-1(V) is open if and only if 'Y-1(V) = <p -I(F-I(V»
is open, which proves the lemma. 0
For any continuous F: C ~ ~2 \ {P}, we can define the winding number
of F around P, denoted W(F,P), to be the winding number W('Y,P)
of the path 'Y = F 0 <po

Exercise 3.19. Identify ~2 with the complex numbers C, and let

f C~ C be the mapping that takes a complex number z to its nth
power 1', where n is an integer. Let C be any circle centered at the
origin, and let F be the restriction of f to C. Show that W(F, 0) = n.
If flz) = -z, show that W(F,O) = 1.

Proposition 3.20. Suppose C is the boundary of the closed disk D,

and F: C ~ ~2 \ {P} extends to a continuous function from D to ~2 \ {P}.
Then W(F, P) = o.
Proof. If D is the disk of radius r about the point (xo, Yo) as
above, and 'Y: [0, 1]~ ~2 \ {P} is the path corresponding to F, and
F: D~ ~2 \ {P} is such an extension of F, then
H(t,s) = F(xo,Yo)+s(rcos(2'TTt),rsin(2'TTt»),
0:5t:51, 0:5s:51,
gives a homotopy from 'Y to the constant path at the point F«xo, Yo».
This homotopy stays inside ~2 \ {P}, and since the winding number
of a constant path is zero, the claim follows from Corollary 3.8. 0
3d. Degrees and Local Degrees 45

Problem 3.21. If Fo and FI are mappings from a circle C to U, cor-

responding to paths 'Yo and 'YI from [0, 1] to U, show that 'Yo and 'YI
are homotopic through closed paths if and only if Fo and FI are ho-
motopic mappings, i.e., there is a continuous mapping
H: C X [0, 1] ~ U
with H(P X 0) = Fo(P) and H(P X 1) = FI(P) for all Pin C.

Problem 3.22. Show that the converse of Proposition 3.20 is true: if

W('Y, P) = 0, then 'Y has a continuous extension to a map from D to
~2 \ {Pl.

Problem 3.23. Let C be a circle centered at the origin, and let F: C ~ ~2

be a continuous mapping such that the vector F(P) is never tangent
to the curve C at P, i.e., the dot product F(P) • P is not zero for all
Pin C. Show that W(F, 0) = 1.

The same idea lets us define the degree of any continuous mapping
F from one circle C to another circle C', which measures how many
times the first circle is wound around the second by F. Let P' be the
center of the circle C', and define the degree of F by the formula
deg(F) = W(F, P').

Exercise 3.24. Show that one could take any point inside C' in place
of P'.

Exercise 3.25. (a) Show that if F is not surjective, then its degree is
zero. (b) Give an example of a surjective mapping from C to C' that
has degree zero. (c) Show that if F and G are homotopic mappings
from C to C', i.e., if there is a continuous mapping H from C X [0, 1]
to C', with F(Q) = H(Q X 0) and G(Q) = H(Q Xl) for all Q in C.
then F and G have the same degree. (d) Show that if C is the boundary
of the disk D, then F extends to a continuous mapping from D to C'
if and only if F is homotopic to a constant mapping from C to C',
and then the degree of F is zero.

Problem 3.26. (a) Prove that two mappings from C to C' have the
same degree if and only if they are homotopic. (b) Deduce that, if C
is the boundary of the disk D, and the degree of a mapping is zero,
then the mapping extends to a continuous mapping from D to C'. (c)
Deduce also that if Sl is the unit circle centered at the origin, and
F: Sl ~ Sl is a continuous mapping with degree n, then F is homo-
46 3. The Winding Number

topic to the mapping that takes (cos(t}), sin(t}» to (cos(nt}), sin(nt})),

i.e., in the terminology of complex numbers, the restriction of the
nth power mapping z t-+ z' to the unit circle.

Problem 3.27. If F: C ~ C' and G: C' ~ C" are continuous mappings

of circles, what can you say about the relation among the degrees of
F, G, and the composite GoF? Can you prove your answer?

These ideas can also be used to define an important notion of a

local degree. Suppose U and V are open sets in the plane, and F: U~ V
is a continuous mapping, and let P be a point in U. Assume that P
has some small neighborhood such that F(Q) ¥- F(P) for all Q in that
neighborhood with Q ¥- P. Choose a positive number r so that no point
within a distance r of P has the same image as P, and let C,(P) be
the circle of radius r about P. Then F restricts to a continuous map-
ping from ClP) to Ilf \ {F(P)}.

Define the local degree ofF at P, denoted degp(F), to be the winding

number of this mapping of the circle around the point F(P). In other
words, degp(F) = W('Y" F(P» , where
'Ylt) = F(P + r(cos(21Tt), sin(21Tt))), 0 s t s 1.
To know that this is well defined, we need

Lemma 3.28. This winding number is independent of choice of r.

Proof. Ifr' isanother,H(t, s) = F(P + «(1- s)r + sr')(cos(21Tt), sin(21Tt»)

gives a homotopy from 'Yr to 'Yr" 0

Equivalently, the local degree of F at P is the winding number of

the mapping from the unit circle Sl to itself given by
F(P + rQ) - F(P)
Q t-+ IIF(P + rQ) - F(p)II'

Problem 3.29. Show that if F: 1R2~ 1R2 is a linear mapping, given

3d. Degrees and Local Degrees 47

by a (2 x 2)-matrix, and the determinant is not zero, then the local

degree of F at the origin is + 1 if this determinant is positive, and -1
if this determinant is negative.

Problem 3.30. Show that if F: U ~ V is a ~'" mapping, and the Ja-

cobian determinant of F is not zero at P, then the local degree is
defined, and is + 1 or -1, depending on the sign of this determinant.

Problem 3.31. Suppose F: C~ C is given by a complex polynomial.

Show that the local degree of F at a complex number z is the mul-
tiplicity of z as a root of F(T) - F(z). (This multiplicity is the highest
power of T - z that divides F(T) - F(z).)

Problem 3.32. If F: C ~ C' is a map between circles in the plane,

and P is a point in C such that F(Q) # F(P) for all Q in a neighbor-
hood of P in C, give a precise definition of a local degree of F at P;
this should be + 1 if F is increasing at P, -1 if F is decreasing, and
o if F has a local maximum or minimum at P (all expressed in terms
of counterclockwise angles). Show that if p' is any point of C' such
that p-l(p') is finite, then
deg(F) = 2:
PEF- 1(P')
degpF,

where degpF is the local degree of Fat P. This implies in particular

that the right side is independent of choice of P'.
CHAPTER 4

Applications of Winding Numbers

4a. The Fundamental Theorem of Algebra

The set C of complex numbers is identified as usual with the real
plane ~2, the number z = x + iy being identified with the point (x, y).
We will use the fact that for any complex polynomial
g(T)=aor+a1r- I + ... +an-1T+an,
with coefficients ai complex numbers, the mapping z ~ g(z) is a con-
tinuous mapping from C to C. This follows from the fact that addition
and multiplication of complex numbers are continuous. The goal of
this section is to show that, if n > 0 and ao # 0, then the polynomial
has a root, i.e., g(z) = 0 for some z. We may divide by ao, so we can
assume g(T) has leading coefficient ao = 1. If g(T) has no root, g is
a mapping of C into the complement of the origin.
Restrict g to a circle Cr of radius r centered at the origin. This gives
a mapping from Cr to C \ {O}, which we denote by gr. Since gr extends
to a continuous mapping of the disk Dr of radius r into C \ {O}, it
follows from Proposition 3.20 that the winding number W(g" 0) must
be zero. The idea is to compare gr with the mappingJ,. given similarly
by the polynomial.fiT) = r. The restriction of this to the circle is
J,.(z) = z\ and a simple calculation shows that W(J,., 0) = n (see Ex-
ercise 3.19) . We will apply the Dog-on-a-Leash Theorem 3.11 to show
that, for r sufficiently large, J,. and gr must have the same winding
number, which will be the desired contradiction. For this, it suffices

48
4b. Fixed Points and Retractions 49

to show that for r sufficiently large,

If,.(z) - g,(z)1 < If,.(z) - 01 for zEC,.
Now If,.(z) - 01 = Iznl = 1"', and
If,.(z) - g,(z)1 = lalzn- I + ... + an-Iz + ani
:s lallr"-I + ... + lan-dr + lanl,
which is less than I'" if r is large, e.g., if laA < ri/n for all i. This
completes the proof of

Proposition 4.1 (Fundamental Theorem of Algebra). Any complex

polynomial of degree greater than zero has a root.

If Zl is a root of g(D, then g(D = (T - zd' h(D, where h(D is a

polynomial of degree n - 1. By induction, we see that g(D factors
into linear factors: g(D = ao' I17=I(T - z;).

Exercise 4.2. (a) Suppose! C~ C is a continuous function such that

for some R > 0, Iftz)1 < Izln for Izl = R. Show that zn + f(z) = 0 has a
solution z with Izl < R . (b) Suppose g: C ~ C is a continuous function
such that g(z) / zn approaches a nonzero constant as Izl ~ 00. Show that
g is surjective.

There is another topological proof of the Fundamental Theorem of

Algebra that is in some ways even simpler, but requires a little more
about the local structure of a mapping given by a polynomial. One
shows that g extends to a continuous mapping of the Riemann sphere
to itself, taking the point at infinity to itself (which is essentially what
the above calculation showed), and which is an open mapping (see
Problem 3.31, and § 19a for details). The image is compact, so closed,
and since both open and closed, it is the whole sphere.

4b. Fixed Points and Retractions

One of the most important applications of topological ideas in other
areas of mathematics, and in science in general, is to give criteria to
guarantee that a continuous mapping from a space to itself must have
a point that is mapped to itself.
We start with the case of a closed interval [a, b]. We claim that
any continuous function! [a,b]~[a,b] must have a fixed point. This
can be "seen" by looking at the graph of the mapping:
50 4. Applications of Winding Numbers

b -+---+------,1"-

a -+------,1'-----1-

a b

To prove it rigorously, consider the function g(x) =f(x) - x. This is

a continuous function on the interval [a, b], with g(a) 2: 0 and g(b):::; O.
Since the image g([a, b]) of the interval by g must be connected, it
must contain the interval [g(b), g(a)], so it must contain O. This means
that g(x) = 0 for some x, which says that fix) = x.
This is closely related to another property of an interval: that there
is no continuous mapping from an interval [a, b] onto its endpoints
[a, b] that maps a to a and b to b. This fact is obvious since the image
of a connected set must be connected. In general if Y is a subspace
of a topological space X, a retraction from X to Y is a continuous
mapping r: X - Y such that reP) = P for all P in Y. In this case Y is
called a retract of X. So there is no continuous retraction of an in-
terval onto its boundary. This proves again that any continuous func-
tion f from the interval [-I, 1] to itself must have fixed points, for
otherwise the mapping Xf-7 (x - f(x»/ix - f(x)i would be a continuous
retraction of [-I, I] onto {-I, I}.
Next we tum to the case of a closed disk D. We first show there
is no retraction onto its boundary circle C = aD, and then use this as
above to show that any map from the disk to itself must have a fixed
point.

Proposition 4.3. There is no retraction from a closed disk onto its

boundary circle.
Proof. If C is the boundary of the disk D, the identity mapping from
C to itself has degree 1. A retraction would give an extension of the
identity mapping to a mapping from D to C, which would imply by
Proposition 3.20 that the degree is O. 0

Proposition 4.4 (Brouwer). Any continuous mapping from a closed

disk to itself must have a fixed point.
Proof. We may assume the disk D is the unit disk centered at the
origin, with boundary C the unit circle (see Exercise 4.7 below). Sup-
4b. Fixed Points and Retractions 51

pose f D ~ D is a continuous mapping with no fixed point. The idea

is to define a mapping h: D ~ C that takes a point P to the point in
C hit by the ray from f(P) to P:

r(p)

hlP)

This mapping h will be the identity on C, so will be a retraction

of D onto C. To finish the proof, we must verify that h is continuous.
To do this, we find an explicit formula for h. We know that
P - f(P)
h(P) = P + t· U with U=---
lip - f(p)11 '
and t is the positive number determined by the property that Ilh(P)11 = 1.
A little calculation shows that
t = - P • U + VI - P • P + (p. U)2
is such a positive number. With this t, the above formula for h is then
continuous, and is easily checked to be the identity on C. D

The following exercise gives a proof with less calculation:

Exercise 4.5. Iff D2~ D2 has no fixed point, define g: D2~ [R2 \ {O}
by setting g(P) = P - f(P). Show that g(P) . P > 0 for all P in Sl, so
the restriction of g to Sl is homotopic to the identity mapping of Sl
(by the homotopy H(P x s) = (l - s)P + s g(P». Apply Proposition
3.20.

Exercise 4.6. Deduce Proposition 4.3 from Proposition 4.4.

One says that a topological space X has the fixed point property if
every continuous mapping f X ~ X has some fixed point, i.e., there
is a point P in X withf(P) = P. So any closed interval and any closed
disk have the fixed point property.
Exercise 4.7. Show that a space that is homeomorphic to a space
52 4. Applications of Winding Numbers

with the fixed point property has the fixed point property. Show that
if Y is a retract of a space X that has the fixed point property, then
Y has the fixed point property.

Exercise 4.8. Which of the following spaces have the fixed point
property? (i) a closed rectangle; (ii) the plane; (iii) an open interval;
(iv) an open disk; (v) a circle; (vi) a sphere S2; (vii) a torus Sl x Sl;
and (viii) a solid torus Sl x D2.

Exercise 4.9. Let D be a disk with boundary circle C, and letf D~ ~2

be a continuous mapping. Suppose P is a point in ~2 that is not in
the image C, and the winding number of the restriction f Ie of f to C
around P is not zero. Show that there is some point Q in D such that
f(Q) =P.

Exercise 4.10. Suppose D and D' are disks with boundary circles C
and C'. Suppose f D ~ ~2 is a continuous mapping that maps C into
C', such that the degree of this map from C to C' is not zero. Show
that f(D) must contain D' .

Exercise 4.11. Show that if f D2~ ~2 \ {O} is a continuous mapping,

there is some P in Sl = aD 2 and some A> 0 such that f(P) = A' P, and
there is some Q in Sl and some fl. < 0 such that f(Q) = fl. . Q.

Exercise 4.12. Suppose F is a continuous mapping from the positive

octant {(x, y, z): x ~ 0, y ~ 0, z ~ O} to itself. Show that there is a unit
vector P in this octant, and a nonnegative number A, such that
F(P) = A'P.
Exercise 4.13. If all the entries of a (2 x 2)-matrix are nonnegative,
show by direct calculation that at least one of its eigenvalues must be
nonnegative. Prove the same for a (3 x 3)-matrix A.

Exercise 4.14. Iff D2 ~ ~2 is a continuous mapping, show that either:

(i) there is either some point Q E D2 such that f(Q) = Q; or (ii) there
is some PI in Sl and some AI> 1 such thatf(Pd = Al . Ph and there
is some point P 2 in Sl and some A2 < 1 such thatf(P2) = A2' P 2 .

Exercise 4.15. If f C~C is a continuous mapping with no fixed

point, show that degree off must be 1. In particular, iff has no fixed
point, show that f must be surjective.
4c. Antipodes 53

Exercise 4.16. Iff: C ~ Il\f is continuous and W(j, P) # 0, show that

every ray emanating from P meetsflC).
Exercise 4.17. If f: D2~ ~2 is continuous and flP)· P # 0 for all P
in Sl, show that there is some Q in D2 with flQ) = O.

Problem 4.18. Let DOD = {(ao, aI, a2, ... ): ~;=oa/:s I}, the unit ball
in the metric space of infinite sequences such that ~;=o a/ < 00. (a)
Find a continuous mappingj: D'" ~ D'" that has no fixed point. (b) Find
a continuous retraction of DOD onto S'" = {(ao, aI, ... ): ~;=oa/ = I}.

4c. Antipodes
The antipode of a point in a circle C or sphere S is the opposite point,
i.e., the point hit by the ray from the point through the center. We
denote the antipode of P by P*. With the center at the origin, p* = - P.
The antipodal map is the mapping that takes each point to its anti-
pode.
Exercise 4.19. Show that the degree of the antipodal map from a
circle to itself is 1.
Lemma 4.20 (Borsuk). If C and C' are circles, and j: C ~ C' is a
continuous map such that flP*) = flP)* for all P, then the degree of
fis odd.
Proof. There is no loss of generality in assuming C = C' = SI. Let
'Y(t)=flcos(t),sin(t»,O:st:s'IT. Since 'Y('IT) = -'Y(O), the change in
angle along 'Y must be 2'ITn + 'IT for some integer n. So W('Y, 0) = n + 1/2 .
Now let
aCt) = flcos(t + 'IT), sin(t + 'IT» = -'Y(t) ,
54 4. Applications of Winding Numbers

It follows easily from the definition of the winding number that

W(a, 0) = W('Y, 0), and so deg(f) = W('Y, 0) + W(a, 0) = 2n + 1. 0
Lemma 4.21. There is no continuous mapping f from a sphere S to
a circle C such that f(P*) = f(P)* for all P in S.
Proof. Again we can take S = S2. Consider the mapping g: D2~ C
given by the formula
g(x,y) = f«x,y, Yl-x2 -l».
This is continuous, since the projection from the upper hemisphere
of the sphere to the disk (being one-to-one and continuous) is a ho-
meomorphism, and g is the composite of f with the inverse. Now
g(P*) =f(P*) =f(P)* = g(P)* for P in Sl, so by Lemma 4.20 the de-
gree of the restriction of g to Sl must be odd. But since this map
extends over the disk, its degree must be zero, a contradiction. 0

Proposition 4.22 (Borsuk-Vlam). For any continuous mapping

f S ~ ~2 from a sphere S to the plane, there is a point P in S such
that f(P) = f(P*).

So there are always two antipodal points on the earth with the same
temperature and humidity-or any other two real values, provided
they vary continuously over the earth.

Proof. We may take S = S2. If there is no such P, consider the func-

tion g: S2~SI given by
f(P) - f( -P)
g(P)
IIf(p) - f( - p)ll·
Then g(-P) = -g(P), contradicting Lemma 4.21. o
A fact which was unquestionably obvious until people started look-
ing for a proof is the fact that open sets of different dimensions cannot
be homeomorphic. (The fact that they cannot be diffeomorphic can
be reduced, using Jacobian matrices, to the fact that vector spaces of
different dimensions cannot be isomorphic.) The fact that there are
continuous maps from intervals onto squares makes the assertion less
obvious than might have been thought. The first case is easy: an open
interval cannot be homeomorphic to an open set in the plane or any
~n, for the reason that removing a point disconnects an interval, but
does not disconnect an open set in ~n, n ~ 2. The next case is less
obvious:
4c. Antipodes 55

Corollary 4.23 (Invariance of Dimension). An open set in 1R2 cannot

be homeomorphic to an open set in IRn for n::=: 3.
Proof. In fact, no subset of the plane can be homeomorphic to a set
which contains a solid ball in IR n, n::=: 3, for a homeomorphism from
a ball D n (of some small radius) in the plane would embed a two-
sphere S2 C D3 C Dn in the plane, contradicting the proposition. 0

As you must expect, these results are also true in higher dimen-
sions, but more machinery is needed to extend the proofs. We'll come
back to this in the last part of the book.

Exercise 4.24. Show that iff C ---+ C' is a map between circles such
that fiP*) =fiP) for all P, then the degree off is even.

Exercise 4.25. If f C ---+ 1R2 \ {Q} is a continuous mapping such that

Q lies on the line segment between fiP) and fiP*) for all P in C, show
that the winding number of f around Q is odd.

Exercise 4.26. Suppose D is a disk with boundary C, andf: D---+ 1R2

is a continuous mapping such thatfiP*) = -fiP) for all Pin C. Show
that there is some point Q in D with fiQ) = O.

Exercise 4.27. If f and g are continuous real-valued functions on a

sphere S such that fiP*) = -fiP) and g(P*) = - g(P) for all P, show
thatf and g must have a common zero on the sphere.

Exercise 4.28. State and prove the analogue of the Borsuk-Ulam

theorem for mappings from a circle to IR.

Exercise 4.29. Iff C ---+ C' is a continuous mapping between circles

such that fiP*) #- fiP) for all P, show that deg(f) #- 0, so f must be
surjective.

Problem 4.30. Letf S---+S' be a continuous mapping between spheres.

Show that if fiP) #- fiP*) for all P, then f must be surjective.

Exercise 4.31. Letf C---+ C be continuous. If deg(f) #- 1, show that

there is a P in C with fiP) = P and there is a Q in C with fiQ) = Q* .
56 4. Applications of Winding Numbers

4d. Sandwiches

A pleasant application of the Borsuk-Ulam theorem is the so-called

ham sandwich problem: to cut both slices of bread and the ham be-
tween in two equal parts with one slice of a knife.

Suppose we have three bounded objects A, B, C in space. The prob-

lem is to show that there is one plane that cuts each of them in half
(by volume). Here is what we need to know about the volume of each
object X = A, B, or C:
(i) For any fixed line l, there is a unique point in l such that the
plane perpendicular to l through the point cuts X in half. Call this
point P I•X '
(ii) The point in (i) varies continuously as the line varies continu-
ously. More precisely, take a big sphere S with X inside, and
consider for each Q E S the line l(Q) going from Q to its antipodal
point. Then the map from Q to PI(Q),X is continuous.
These properties are intuitively evident, and follow from elemen-
tary properties of volume, For example, the first follows from the
fact that the volume on one side increases continuously as the point
moves along the line. We will take them as axioms, or consider only
objects for which we know them. (For a detailed discussion, see Chinn
and Steenrod (1966).)
Given a body X inside a sphere S as above, define a continuous
real-valued function Ix: S~ IR by defining fx(Q) to be the distance
from Q to the point PI(Q),x. Note that fx(Q*) = d - fxCQ) , where d is
the diameter of the sphere, and Q* is the antipodal point to Q,
4d. Sandwiches 57

Now for three bodies, take S containing all three, and consider the
mapping g: S ~ ~2 given by
g(Q) = (fA(Q) - fdQ) ,fB(Q) - fdQ».
Then g(Q*) = -g(Q) for all Q, so by Proposition 4.22 some Q must
be mapped to the origin, which means that P1(Q),A = P1(Q),B = P1(Q),C, as
required. This proves

Proposition 4.32 (Stone-Tukey). Given three bounded measurable

objects A, B, and C in space, there is a plane that divides each in
half by volume.

The same idea is used in the following proposition, which looks

quite different.

Proposition 4.33 (Lustemik-Schnirelman-Borsuk). It is impossible

to cover a sphere with three closed sets, none of which contains a
pair of antipodal points.
Proof. Suppose the sphere S is covered by three such closed sets K\ ,
K 2 , and K 3 • For each i, define a real-valued continuous function.fi on
S, whose value at P is the minimum distance from P to K; (see the
following exercise). Consider the mapping g: S~ ~2 given by
g(P) = (fl(P) -HP) ,f2(P) - HP».
By Proposition 4.22, there is a point P with g(P*) = g(P). For such
P, .fi(P) - jj(P) = .fi(P*) - jj(P*) for all i and j. But P must be in one
of the sets K;, and p* in another K j • Since P fI. Kj and p* fI. K;,
o> -jj(P) = .fi(P) - liP) = .fi(P*) - jj(P*) = .fi(P*) > 0,
a contradiction. D

Exercise 4.34. Show that for any compact set K in space, the function
p on ~3 that maps a point P to its distance from K is continuous. Note
by compactness that if p(P) = r, then there is a point Q in K of dis-
tance r from P.

Exercise 4.35. Show that the unit ball D3 is not the union of three
closed sets, each with diameter less than 2.

Exercise 4.36. Show that the "three" in Proposition 4.33 cannot be

replaced by "four."
58 4. Applications of Winding Numbers

Exercise 4.37. State and prove the analogue of Proposition 4.33 for
a circle.

Exercise 4.38. Show how Proposition 4.33 implies Lemma 4.21, and
hence Proposition 4.22.

Exercise 4.39. For a subset X of a sphere, let X* = {P*: P EX}. Sup-

pose A, B, and C are disjoint closed subsets in a sphere, none con-
taining a pair of antipodal points. Show that the six sets A, B, C, A * ,
B*, and C* cannot cover the sphere.

Problem 4.40. True/False. If a sphere is a union of two closed sets

A and B, then either A or B must contain a closed connected set X
such that X* = X.

It is an excellent project to try to generalize the results of this chap-

ter, including the exercises, to higher dimensions, assuming, for ex-
ample, that there is no continuous retraction from D n onto sn-l, and
that there is no continuous mapping/from the sphere Sn to the sphere
Sn-l such thatftP*) = ftP)* for all P. This will be carried out in Chap-
ter 23.
 PART III

COHOMOLOGY AND HOMOLOGY, I

We have seen that the topology of an open set U in the plane is related
to the question of whether all closed I-forms on U are exact. This is
formalized by introducing the vector space HI U of closed forms mod-
ulo exact forms. What we learned in the first few chapters amounts
to some calculations of these fIrst De Rham cohomology groups. There
is also a Oth De Rham group HOU, which measures how many con-
nected components U has.
A central theme will be how the topology of a union U U V of two
open sets compares with the topology of U and V and the intersection
un v. We construct a linear map from H°(U n V) to HI(U U V), and
describe the kernel and cokernel of this map. We use these groups
and this map to prove the famous Jordan curve theorem, which says
that any subset of the plane homeomorphic to a circle separates the
plane into exactly two connected pieces, an "inside" and an "out-
side. "
60 Part III. Cohomology and Homology, I

This is one of those facts that seem intuitively evident, although a

complicated enough maze may raise a few doubts. In fact, it went
unquestioned until mathematicians realized that continuous mappings
can be pretty horrible-for example, that there can be a continuous
mapping from an interval onto a square. When X is a closed polygon,
it is not too hard to give an elementary proof, and you might enjoy
seeing if you can.
A basic question, which motivates Chapter 6 and the continuation
in Chapter 9, stems from three different ways one can compare two
closed paths 'Y and & in an open set U in the plane, supposing for
simplicity that they are differentiable:
(1) Are the winding numbers of 'Y and & around all points not in U
the same?
(2) Are all integrals of closed i-forms w on U along 'Y and & the
same?
(3) Are the paths homotopic, or related by deformations of some kind?
In studying questions like this, we will find it useful to generalize
and formalize some of the ideas of previous chapters. In Chapter 6,
paths and segmented paths are generalized to the notion of I-chains,
which are arbitrary sums and differences of (nonconstant) paths. We
define what it means for two such chains to be homologous: the dif-
ference should be a linear combination of boundaries of maps of a
rectangle into the region. For an open set in the plane, we show that
this notion is equivalent to saying that the two chains have the same
winding number around all points outside the open set. The main
technique is to approximate general paths by rectangular paths along
Part III. Cohomology and Homology, I 61

sides of a grid. 3 We introduce the first homology group H1U, which

is the group of closed I-chains up to homology, and a Oth group HoU,
which is a free abelian group on the connected components of U. In
Part V we will develop general tools for computing these homology
and cohomology groups.

3The use of grids in these chapters follows ideas of L.E.J. Brouwer, E. Artin, and
L. Ahlfors, see Ahlfors (1979).
CHAPTER 5

De Rham Cohomology and the Jordan

Curve Theorem

5a. Definitions of the De Rham Groups

Define, for an open set U in the plane: the zeroth De Rham group,
HOU = {locally constant functions on U}.
This is a vector space, by the ordinary addition of functions, and
multiplication of functions by real scalars. As we have seen, to give
a locally constant function on U is the same as giving a constant on
each connected component of U. If U has n connected components,
then HOU is an n-dimensional vector space; if Ub . . . , Un are the
connected components of U, and ei is the function that is i on Ui and
o on Uj for all j~ i, then ei, ... , en is a basis for HOU. Similarly,
if U has infinitely many components, HOU is an infinite-dimensional
vector space.
The closed i-forms on U also form a vector space, since the sum
of closed forms, and a constant times a closed i-form, are also closed.
The exact i-forms on U are a subspace of this vector space, since
d(fl + f2) = dj; + df2 and d(cf) = cdf. Whenever we hav~ a subspace
W of a vector space V, we can form the quotient space V /W of equiv-
alence classes: two vectors in V are equivalent if their difference is
in W (see Appendix C). Define the first De Rham cohomology group
of U, denoted H1U, by
H1U = {closed i-forms on U}/{exact i-forms on U}.

63
64 5. De Rham Cohomology and the Jordan Curve Theorem

If W is a I-form on U, we may write [w] for the equivalence class in

H1U containing the I-form w.
Later we will have general methods for calculating H1U. In this
section we use the ideas of Part I to compute a few simple examples.
For any point P = (xo, Yo), let Wp denote the I-form
1 I - (y - Yo) dx + (x - xo) dy
Wp = 27T Wp,,~ = 27T (x - XO)2 + (y - Yo)2

We have seen that Wp is a closed I-form on any open set not con-
taining P.

Proposition 5.1. (a) If U is an open rectangle, then H1U = O. (b) If

U= ~2\ {P}, then [wp] is a basis for H1U. (c) IfU= ~2\{p,Q}, with
P -,.6 Q, then [wp] and [wQ] form a basis for H1U.
Proof. Assertion (a) is a translation of Proposition 1.12. To prove
(b), fix a positive number r, and let 'YP,r denote the counterclockwise
circle of radius r about P:
'Yp,r(t) = P + r(cos(27Tt), sin(21l't)),
Note first that [wp] is not 0 in H1U, for if Wp were exact, its integral
around 'YP,r would be zero by Proposition 1.4, and we know that this
integral is 1. Let W be any closed I-form on U, and let c = f"'/p"w. To
show that [wp] spans H1U, it suffices to show that W - c· Wp is exact,
for then [w] = c· [wp] in H1U. Now W - c· Wp is a closed I-form on
U whose integral along 'Y is 0, and Lemma 1.17 implies that such a
form is exact.
For (c), fix a positive number r less than the distance between P
and Q. To see that [wp] and [wQ] are linearly independent, we must
show that a' Wp + b· wQ is not exact unless a and b are both zero.
The integral of this form along 'YP,r is a, and the integral along 'YQ,r
is b, and by Proposition 1.4, these integrals must vanish if the form
is exact. To show that [wp] and [wQ] span H1U, let W be any closed
I-form on U. Let a = f"'/p" W and let b = f"'/Q,' W. To complete the proof
of (c), we must show that W - a' Wp - b· wQ is exact on U, for then
[w] = a' [wp] + b· [wQ] in H1U. Now W - a' Wp - b· wQ is a closed 1-
form on U whose integrals along 'YP,r and 'YQ,r both vanish, and an
appeal to Lemma 1.18 completes the proof. 0

Problem 5.2. Generalize the proposition from one and two to n points.

There is one other result we will need later in this chapter:

5b. The Coboundary Map 65

Proposition 5.3. If A is a connected closed subset of ~2, and P and

Q are points in A, then [wp] = [wQ] in Hl(~2 \A).
Proof. In order to show that W = Wp - wQ is exact on ~2 \ A it suffices
by Proposition 1. 8 to show that f'Y W = 0 whenever -y is a segmented
closed path in ~2 \A. But

LW LWp- LWQ = W(-y,P)-W(-y,Q).

Now Proposition 3.16 implies that the winding numbers W(-y,P) and
W(y, Q) are equal, since P and Q belong to the same connected com-
ponent of ~2 \ Supp(-y) . D

Exercise 5.4. With A as in the proposition, and P in A, show that

[wp] = 0 in Hl(~2 \A) if and only if A is unbounded.

5b. The Coboundary Map

Our basic tool for studying HOU and HIU, and hence the topology of
U, will be a method to study the relations among these groups for
open sets U, V and their union U U V and intersection un V.

u
u v

That there might be some relation can be seen already in Lemma

1.14. This showed, for example, that if un V is connected, i.e.,
HO(Un V) has dimension at most one, and if H1U= 0, and H1V= 0,
then HI(U U V) must vanish as well. We want to generalize this to
more complicated open sets.
For any two open sets U and V in the plane, we will define a linear
map

To do this, we will need the following basic fact:

Lemma 5.5. Given open sets U and V in ~2, any C(6'" function on
66 5. De Rham Cohomology and the Jordan Curve Theorem

un V can be written as the difference 01 two ~oo functions, one that

extends to a ~oo lunction on U, the other to a ~oo function on v.
Proof. This is a consequence of the existence of a partition of unity,
for the simple (but not so trivial) case of the covering of U U V by
the open sets U and V. This says that there are ~oo functions 'P and
IjI on U U V such that 'P + IjI == I on U U V, such that the closure (in
U U V) of the support of 'P is contained in U, and the closure (in
U U V) of the support of IjI is contained in V. (The support of a func-
tion is the set where it is not zero.) For the proof, see Appendix B2.
To use it to prove the lemma, given a ~oo function I on un V,
define lIon U by the rule
_ {1jI.1 on un V,
II - 0 on u\unv.
The assumption on the support of IjI insures that any point in the set
U \ un V has a neighborhood disjoint from the support of 1jI, from
which it follows that II is ~oo on all of U. Similarly, define a ~oo
function 12 on V by

{ -'P.I on un V,
o on V\ un V.

Thenll-fi=(IjI+'P)I=lon unv, as required. o

Construction of 8: HO(UnV)-7H I (UUV). Given a locally con-
stant function I on U n V, use the lemma to find ~oo functions II and
12 on U and V respectively so that 1= II -12 on un V. Since I is
locally constant,
dj;-dI2 = d(fI-12) = dl = 0 onUnV.
This means dll and dl2 agree on un V, so there is a unique I-form
won U U V that agrees with dj; on U and with dl2 on V. This I-form
w is closed, since it is even exact on each of U and V.
We define 8(f) to be the equivalence class in HI(U U V) deter-
mined by this closed form w, i.e., 8(f) = [w]. For this to be well
defined, we must see how w depends on the way we write I as the
difference j; - AWe claim that a different choice would lead to a
closed I-form that differs from w by an exact I-form. To see this,
suppose II I and fi I were another choice of functions on U and V with
II' - N = Ion un V. (These primes have nothing to do with deriv-
atives!) Let w' be the I-form on un V that is dN on U and is dN
5b. The Coboundary Map 67

on V. Now sincefl' - H =t. - f2 on un v,

/.'-fl = H-f2
on un V, so there is a 'fJ"" function g on U U V that is fl' - fl on U
and isH - f2 on V. Then dg = w' - w, as required.

Lemma 5.6. The mapping 13 is a linear mapping of vector spaces,

i.e., 5(/+ g) = 13(/) + 13(g) and 13(c·f) = c ·13(/),for f, g locally con-
stant functions, and c a constant.
Proof. This is just a matter of making the choices "linearly," and is
better (and probably easier) to check for yourself than to read. Write
f= fl - f2 and g = g, - g2 as in the construction of 0(/) and o(g).
Thenf+ g = (/1 + gl) - (/2 + g2). If wf is the I-form that is dfl on U
and df2 on V, and Wg is the I-form that is dg l on U and dg 2 on V,
then wf+ Wg is the I-form that is d(/I + gl) on U and d(f2 + g2) on
V. Therefore 8(/+ g) is represented by the I-form wf+ wg , which by
definition represents the sum 8(/) + 8(g). This proves that 8 preserves
sums. The proof that it preserves multiplication by a scalar is similar,
and left as an exercise. 0

This map 8 is called the coboundary map. In order to use it to

compare un V with U U V, we need a description of its kernel and
its image. The following two propositions do this.

Proposition 5.7. A locally constant function f on un V is in the ker-

nel of 8 if and only if there are locally constant functions fl on U and
f2 on V so that f= fl - f2 on un V. In particular, if U and V are
connected, then the kernel of 8 consists of the constant functions on
unv.
Proof. If f= fl - fi with fl and fi locally constant on U and V, these
can be chosen for the construction of 8(f), and the corresponding
form W is zero. Conversely, if 8(/) is zero, the form W of the con-
struction from an equation f= fl - f2 must be exact. Write W = dg.
Then dfl = dg on U, and df2 = dg on V. This means thatfl - g is lo-
cally constant on U, and f2 - g is locally constant on V. And
f = fl - f2 = (/1 - g) - (fi - g)
is the difference of two such functions.
If U and V are connected, locally constant functions on them must
be constant, and since f is the difference of two such functions, f is
also constant. 0
68 5. De Rham Cohomology and the Jordan Curve Theorem

Exercise 5.S. Show that if U and V are connected, and HI(U U V) = 0,

then un V is also connected.

Proposition 5.9. The class [w] of a closed I-form won U U V is in

the image of () if and only if the restrictions of w to U and to V are
exact. In particular, if HIU = 0 and HIV= 0, then () is surjective.
Proof. By construction, if w is in the image of &, there are functions
fl on U andJz on V with w = dfl on U and w = df2 on V. Conversely,
if there are such functions, the functionf= fl - f2 on un V is locally
constant, since df= dfl - df2 = W - W = 0 on un V, and [w] = &(f)
by construction. D

5c. The Jordan Curve Theorem

If X is a closed subset of the plane, there is a close relation between
the topology of X and the topology of its complement. We prove some
important cases of this fact here.

Theorem 5.10 (Jordan Curve Theorem). If xc 1R2 is homeomorphic

to a circle, then its complement 1R2 \ X has two connected components,
one bounded, the other unbounded. Any neighborhood of any point
on X meets both of these components.
Proof. Let P and Q be any two points in X. By considering a homeo-
morphism of X with the circle, we can write X as a union of two
subsets A and B, each homeomorphic to a closed interval, with
AnB={p,Q}.

We will apply the results of the preceding section to the open sets
U = 1R2 \A and V= 1R2 \B. Note that
U U V = 1R2 \ {P, Q} and U n V = 1R2 \ X .
To show that 1R2 \ X has two components, we want to show that the
dimension of If1(U n V) = If1(1R2 \ X) is 2.
5c. The Jordan Curve Theorem 69

We know from Proposition 5.1(c) that Hi(U U V) = Hi(l~2 \ {P, Q})

is a vector space of dimension 2, with a basis the classes of Wp and
wQ. Each of U and V is the complement of a subset homeomorphic
to an interval. We will need an analogue of the Jordan curve theorem,
but where the circle is replaced by an interval:

Theorem 5.11. If Y C 1R2 is homeomorphic to a closed interval, then

1R2 \ Y is connected.

We postpone a discussion of this theorem to the end of this section,

and show now how to use it to prove the Jordan curve theorem. Con-
sider the boundary map
8: if(unV)=if(1R2 \X) ~ Hi(UUV) =H i(1R 2 \{p,Q}).
Our goal is to show that the image and kernel of 8 are both one di-
mensional, which will imply by the rank-nullity theorem (see Ap-
pendix C) that HO(1R 2\ X) is two dimensional, which means that 1R2 \ X
has two connected components.
Since, by Theorem 5.11, U and V are connected, we can apply
Proposition 5.7, so the kernel of 8 consists of the constant functions
on 1R2 \ X, which is therefore one dimensional. We claim that the im-
age of 8 consists of those linear combinations a' [wp] + b· [wQ] of the
basis elements that have a + b = O. This means that [wp] - [wQ] forms
a basis for the image of 8, and completes the proof of the claim that
the image is one dimensional.
By Proposition 5.9, the image of 8 consists of those linear com-
binations a . [wp] + b . [wQl such that the restrictions of a . Wp + b . WQ
to U and to V are exact. Since P and Q are in the same connected
component of 1R2 \ U and of 1R2 \ V, it follows from Proposition 5.3
that Wp - wQ is exact on U and on V. If a + b = 0, it follows that
a' wp+ b· wQ = a' (wp - wQ) is exact on Uandon V, soa' [wp] + b· [wQ]
is in the image of 8 if a + b = O. Conversely, suppose the restriction
of W= a . Wp + b· wQ to U (and to V) is exact. Let 'Y = 'YO.r be a circle
about the origin, with r so large that X is contained inside this circle.
Since W is exact on U, Proposition 1.4 guarantees that f.y W = O. Since
P and Q are inside the circle, f.., Wp = I and f.., wQ = 1, so

o= L W = a' L
Wp + b . J.., wQ = a + b.
This completes the proof that 1R2 \ X has two connected compo-
nents. Note that since X is bounded, one of these components must
contain everything outside some large disk; this is the unbounded
70 5. De Rham Cohomology and the Jordan Curve Theorem

component, while the other must be bounded. To verify the last as-
sertion of the theorem, suppose N is a neighborhood of some point
in X. We may divide X into two pieces, as in the preceding discus-
sion, so that one of them, say A, lies entirely in N. Take two points,
say Po and Phone in each of the two components of [R2 \ X.

By Theorem 5.11, [R2 \ B is connected, so there is a path 'Y(t), 0::5 t::5 1,

from Po to PI in [R2 \ B. This path must hit A, for if it didn't, it would
connect the two points in [R2 \ X. By looking at the first and last time
the path hits A, i.e., the minimal and maximal t such that 'Y(t) is in
A (which is a closed set in [0, 1]), it follows that N must contain points
of both components: for t near to but less than the minimum, 'Y(t) will
be in N and the component of Po, and for t near to but greater than
the maximum, 'Y(t) will be in N and the other component. 0

Remark 5.12. (1) It follows in particular that X is nowhere dense:

no point has a neighborhood contained in X. Note that an arbitrary
continuous image of a circle or interval need not have this property.
(2) The fact that points on X are "accessible" from both sides is
strengthened in Problem 10.24.

Proof of Theorem 5.11. The proof of Theorem 5.11 uses the same
ideas, but with one new wrinkle. Choose a homeomorphism from the
interval [0, 1] to Y. Let A be the subset of Y corresponding to the left
half of the interval [0,1/2], and B the subset corresponding to the right
e
half 12, 1], and Q = A n B the point corresponding to 1f2.

• • • ~ A
5c. The Jordan Curve Theorem 71

We apply the basic construction to U = ~2 \ A and V = ~2 \ B, with

U n V = ~2 \ Y and U U V = ~2 \ {Q}. It might seem that little will be
gained by this, since each of U, V, and un V is of the same form:
the complement of a subset homeomorphic to an interval. But we will
see that some progress has been made.
Suppose ~2 \ Y is not connected, and let Po and PI be points in two
different connected components. The claim we will prove is: Po and
P I must be in different connected components of ~2 \ A or in different
connected components of ~2 \ B (or both). We use the map
8: If(U n V) = If(~2 \ Y) ~ HI(U U V) = HI(~2 \ {Q}).
Now HI(~2 \ {Q}) is generated by the class [wQ] of wQ. By Propo-
sition 5.9 the image of 8 consists of classes a· [wQ] such that a· wQ
is exact on U and V. If a· WQ is exact on U, the integral of a· wQ
around a large circle is zero, which implies that a = O. So the image
of 8 is zero. In other words, every locally constant function on ~2 \ Y
is in the kernel of 8.
By Proposition 5.7, a function f in the kernel of 8 must have the
form fl - fz, where fl and f2 are locally constant functions on U and
on V, respectively. Takefto be any locally constant function on un V
that takes on different values at the two points Po and PI> which is
possible since they are in different components. If the claim were
false, and Po and PI were in the same component of U and of V, then
each offl andf2 would have to take on the same values at Po and PI.
But then their differencefwould also take on the same values, which
is a contradiction.
The progress made by proving this claim comes from the fact that
the two sets A and B are smaller than Y. To take advantage of this,
we can argue as follows: Let YI be one of the halves of Y such that
Po and PI are in different components of ~2 \ Y I • Repeat the argu-
ment, cutting YI into two pieces (corresponding to cutting the half
interval [0,1/2] or P/2, 1] into equal pieces). For one of the two pieces,
say Y2 , by the same argument, the two points Po and PI are still in
different components of ~2 \ Y2 • Continuing in this way, we get a
nested sequence of subsets
Y ::) YI ::) Y2 ::) Y3 ::) • • • ::) Yn ::) • • •

with the property that Po and PI are in different components of ~2 \ Yn

for all n, with the intersection of all these subsets Yn being a single
point Pin Y.
Since the complement of a point is connected, there is a path from
Po to PI in ~2 that doesn't pass through P. Some neighborhood N of
72 5. De Rham Cohomology and the Jordan Curve Theorem

P is disjoint from this path, and, for large n, Y. is contained in N.

But this forces Po and PI to be in the same component of ~2 \ Y., a
contradiction. This finishes the proof of Theorem 5.11, and hence of
the full Jordan curve theorem. 0

5d. Applications and Variations

The same ideas can be used to calculate the number of connected
components of the complements of many other subsets in the plane.
For many of these variations, we need the following generalization of
Proposition 5.1, which will be proved in Chapter 9 (see §9c and
Lemma 9.1).

(*) Let K be a compact, nonempty subset of the plane. (a) If K is

connected, then Hl(~2 \ K) is one-dimensional, generated by [wp]
for any PEK. (b) If K is not connected, and P and Q are in differ-
ent components of K, then [wp] and [wQ] are linearly independent
in Hl(~2 \ K); if K has exactly two connected components, then
[wp] and [wQ] form a basis for Hl(~2 \ K).

Exercise 5.13. Show that, if A and B are compact connected subsets

in the plane such that An B is not connected (and not empty), then
~2 \ (A U B) is not connected.

Exercise 5.14. Show that if Y is a subset of the plane homeomorphic

to a closed rectangle or a closed disk, then the complement is con-
nected.

Exercise 5.15. Show that if X is a subset of the plane homeomorphic

to a closed annulus, then the complement has two connected com-
ponents.

Exercise 5.16. Show that if X is a subset of the plane homeomorphic

to a figure 8, or a "theta" 8, then the complement has three connected
components.

Here is a simple application of the Jordan curve theorem. Let D be

a closed disk, DO its interior, and C its boundary circle.
5d. Applications and Variations 73

Proposition 5.17. Let f D~ ~2 be a continuous, one-to-one map-

ping. Then ~2 \f(C) has two connected components, which are
f(DO) and ~2 \f(D) .

In particular, fiDe) is an open subset of the plane.

Proof. Recall that a continuous, one-to-one mapping on a compact
set such as D or C must be a homeomorphism onto its image. So the
Jordan curve theorem applies to the image of C, and its complement
has two connected components. The first displayed set is connected
since it is the continuous image of a connected space, and the second
is connected by Exercise 5.14. They are disjoint, and their union is
~2 \f(C). It follows immediately that they must be the two compo-
nents of ~2 \f(C). In particular, since the components of an open
subset of the plane are open, it follows thatf(DO) is open. 0

The following is another intuitively "obvious" result that is not so

easy to prove by hand (try it!):

Corollary 5.18 (lnvariance of Domain). If U is an open set in the

plane, and F: U ~ ~2 is a continuous, one-to-one mapping, then F( U)
is an open subset of ~2, and F is a homeomorphism of U onto F(U).
Proof. Take any P in U, and a closed disk D containing P and con-
tained in U. By the proposition, the image of the interior of the disk
must be open. This gives an open neighborhood of the image point
in F(U), which implies that F(U) is open, and that F is a homeo-
morphism of U with F(U). 0

In particular, if two subsets of the plane are homeomorphic, and

one is an open subset, the other must also be open. So if one is a
domain (a connected open subset), the other must be as well.
Another application is another proof of a result from Chapter 4.

Corollary 5.19. There is no subset of the plane that is homeomorphic

to a two-sphere.
Proof. Suppose f: S ~ ~2 maps the sphere homeomorphic ally onto a
subset X of ~2. Let A and B be the upper and lower closed hemi-
spheres, with intersection C, and let A ° = A \ C and B O = B \ C. Note
that A is homeomorphic to a disk, with C corresponding to the bound-
ary. The complement of the image f( C) of the circle in the plane has
two connected components, and by the proposition (applied to the
map from A to the plane),J(AO) must be one of them. Sincef(AO) is
74 5. De Rham Cohomology and the Jordan Curve Theorem

contained inf(A), it must be the bounded component. The same rea-

soning applies to the map from B, so f(BO) must also be the bounded
component of the complement off(C). But thenf(AO) =f(BO) , which
contradicts the fact that f is one-to-one. D
This section concludes with a few related results, mostly in the
form of exercises and problems, which can be sampled according to
your interest or perseverance. The next proposition verifies that a Jor-
dan curve winds once around each point inside. (This result will not
be used elsewhere.)

Proposition 5.20. Let C be a circle, F: C ~ ~2 a one-to-one, con-

tinuous mapping. Then W(F, P) = ± 1 for P in the bounded component
of lI,f \ F(C).
Proof. In the course of the proof, we will make several assertions
about winding numbers, leaving their proofs as exercises in using the
properties proved in Chapter 3. The mapping F is a homeomorphism
onto its image X = F(C). By the Jordan curve theorem, the comple-
ment has two components, and we know the winding number is 0 for
points in the unbounded component, and it is constant in the bounded
component. So it suffices to find one point P with W(F, P) = ± 1.
Take a horizontal line L that has points of X on both sides of it. Take
a point 0 on L so that X lies in the half plane to the right of O. Let
Sand T be the nearest and farthest points from 0 in X n L, and let M
and N be points on X below and above L such that no other points
of X lie on the line segments from 0 to M and from 0 to N.

o T

The points M and N correspond by F to two points m and n of the

circle C. Let A be the image by F of the counterclockwise arc from
m to n, and let B be the image of the other clockwise arc from m to
n. Let "fA be a closed path starting at 0, going along the segment to
M, then traveling along A to N, and back to 0 along the segment;
5d. Applications and Variations 75

define 'YB similarly using B in place of A. We choose these paths along

A and B by using the mapping F, so that
W('YA,P) - W('YB,P) = W(F,P)
for all P not in X or the line segments from 0 to M or N.
The answer will depend on whether the point T is in A or in B.
Suppose first that T is in A. Then W('YB, T) = 0, so W('YB, T') = for °
all T' in A n L. For a point Q on the line between 0 and S, W('YB, Q) = 1.
It follows that S is not in A, so it must be in B. Since W( 'YA, Q) = 1,
we must have W('YA,S) = 1, and therefore W('YA,S') = 1 for all S' in
B n L. Let S' be the point in B n L that is farthest from 0, and let T'
be the point in A n L that is closest to 0, and let P be any point on
L between S' and T'. Then
W(F, P) = W('YA, P) - W('YB, P) = W('YA, S') - W('YB, T') 1.
If T is in B, a similar argument shows that W(F, P) = -1. 0

The proof gives a criterion to tell whether the winding number is

+ 1 or -1: it is + 1 when four points M, N, S, and T chosen as in
the proof have the same relative position as the four corresponding
points of the circle. (This proof, from Ahlfors (1979), also shows
directly, without the Jordan curve theorem, that the complement of
X has at least two components.)
It is a fact that if X is a subset of ~2 homeomorphic to a circle,
then the bounded component of the complement is homeomorphic to
an open disk, and the unbounded component is homeomorphic to a
disk minus a point. In fact, the Riemann mapping theorem of complex
analysis implies that any connected plane domain with H1U = is
analytically isomorphic to (in particular, diffeomorphic to) an open
°
disk. Even more is true: any homeomorphism of SI with X can be
extended to a homeomorphism from ~2 onto ~2, mapping the inside
and outside of the circle to the two components of ~2 \ X. This last
is called the Schoenflies theorem. For elementary (but not so simple)
proofs, see Newman (1939). This is a case where the analogue in
higher dimensions is more complicated: a homeomorphism of S2 in
~3, or of sn- I in ~n for n;::: 3, will separate the space into two con-

nected components, as we will see in Chapter 23, but the inside need
not be homeomorphic to an open disk; for a picture of a counterex-
ample, "Alexander's homed sphere," see Hocking and Young (1988).
However, if the embedding extends to an embedding of a product of
Sn-I with an interval into ~n, then this wild behavior cannot occur
(see Bredon (1993), §19).
76 5. De Rham Cohomology and the Jordan Curve Theorem

Problem 5.21. Prove the following strong form of Euler's theorem.

°
Let X be a subset of the plane that is a union of v 2:: 1 points and e 2::

°
edges. The edges are assumed to be images of continuous maps from
[0,1] to [R2, each of which maps and 1 into the set of vertices, and
maps the open interval (0, 1) one-to-one into the complement of the
set of vertices. In addition, these open edges are assumed to be dis-
joint. Suppose X has k connected components. Show that [R2 \ X has
f= e - v + k + 1 connected components, i.e.,
v - e +f = 2 + (k - 1).
You may enjoy comparing your argument in the last problem with
that given in Rademacher and Toeplitz (1957), § 12, as well as the
applications given there to the problem of coloring maps. Can you
spot where they make implicit assumptions amounting to what we
proved in this chapter?

Problem 5.22. Show that the following graphs cannot be embedded

in the plane: (i) the graph with vertices PI , P 2 , P 3, Ql , Q2, Q3, with
an edge between each Pi and each Qj; and (ii) the graph with five
vertices, and an edge between each pair of distinct vertices. (It is a
theorem of Kuratowski that any finite graph not containing a subgraph
homeomorphic to one of these two examples can be embedded in the
plane.)

Problem 5.23. Let U be any connected open set in the plane. (a)
Show that if X C U is homeomorphic to a closed interval, then U \ X
is connected. (b) Show that if XC U is homeomorphic to a circle,
then U \ X has two connected components. (c) If X C U is a graph as
in Problem 5.1, show that U \ X has e - v + k + 1 connected com-
ponents.

Exercise 5.24. Show that Theorem 5.11, the Jordan curve theorem
(without mention of bounded or unbounded components), and the re-
sult of Problem 5.21 remain valid when [R2 is replaced by a sphere.

Problem 5.25. Find two graphs in a sphere that are homeomorphic,

but such that there is no homeomorphism of the sphere taking one
onto the other.

Problem 5.26. Show that there is no one-to-one continuous map from

a Moebius band into the plane.
5d. Applications and Variations 77

Problem 5.27. Suppose X is a subset of the plane homeomorphic to

a circle, and PI and P 2 are points in the complement that are joined
by a path that crosses X n times. Show that PI and P 2 are in the same
component of the complement if n is even, and the opposite com-
ponent if n is odd. (A complete answer should include a precise def-
inition of what it means for a path to cross X at a point!)

It is again an excellent project to speculate on the higher-dimen-

sional generalizations of the results of this chapter. For example, if
you assume the fact that if X is any subset of [Rn homeomorphic to a
sphere Sn-I (resp. a ball D n), then [Rn \X has two (resp. one) connected
components, can you state and prove the invariance of domain for
open sets in [Rn?

Problem S.2S. A topological surface with boundary is defined to be

a Hausdorff space such that every point P has a neighborhood homeo-
morphic either to the open disk DO = {(x, y): .i + l < I} or the half
disk {(x,y) E DO: y ~ O}, with P corresponding to the origin; P is an
interior or boundary point according to which case occurs. (a) Show
that this notion is well defined: a point cannot be both an interior
point and a boundary point. (b) Show that homeomorphic surfaces
have homeomorphic boundaries (so the Moebius band is not homeo-
morphic to a cylinder).
CHAPTER 6

Homology

6a. Chains, Cycles, and BoU

As we have seen, it frequently happens that one wants to compute
winding numbers or integrals along a succession of paths, counting
some positively and some negatively. For example, the integral around
the boundary of a rectangle is the sum of integrals over two of its
sides, minus the sum of the integrals over the other two sides. In this
section we formalize these ideas, by introducing the notion of a 1-
chain. A I-chain 'Y in U is an expression of the form
'Y = nl'Yl + n2'Y2 + ... + nr'Yr,
where each 'Yi is a continuous path in U, and each ni is an integer.
For simplicity, so we will not have to mention the interval each path
might be defined on, we will take all paths from now on to be defined
on the unit interval [0, 1]. The paths that are constant, that is, that
map [0,1] to one point of U, can be ignored in the present story. For
example, their winding numbers are zero, all integrals over them are
zero. We will agree that if we meet a sum "Lni'Yi where some of the
'Yi are constant paths, we simply throwaway any constant paths that
occur. Another way to say this is that we identify two expressions
"Lni'Yi and "Ln;'Y; if their difference "Lni'Yi - "Ln;'Y; is a linear combi-
nation of constant paths.
We will make this notion more precise in a moment, but for now
we note that, whatever it means, it is clear how we should define the

78
6a. Chains, Cycles, and RoU 79

winding number of a I-chain 'Y with respect to P:

W('Y,P) = n1W('Yt. P) + ... +nrW('Y"P);
this will be defined provided P is not in the support of any of the
paths 'Yi. In Chapter 9 we will also define the integral of a closed 1-
form along a path, and it will extend additively in the same way to
integrals over I-chains. Two I-chains should be regarded as the same
when each path occurs with the same multiplicity in both I-chains.
A I-chain will have a unique expression as shown, provided the paths
'Yi are all taken to be distinct and nonconstant, and all the coefficients
are taken to be nonzero. (There is also the zero I-chain, written'Y = 0,
which has no paths at all.)
To make this precise, define a I-chain in U to be a function that
assigns to every nonconstant path in U some integer, with the property
that the function is zero for all but a finite number of paths. If
'Yl, ... , 'Yr are the paths for which the value is not zero, and the
value of this function on 'Yi is ni, we write the I-chain as
nl'Yl + ... + nr'Yr. From this definition it is clear how to add and
subtract I-chains: one just adds or subtracts the corresponding values
on each path, or the coefficients in such expressions. In this way the
I-chains form an abelian group,4 with the operation in the group writ-
ten additively. Any path 'Y is identified with the I-chain 1· 'Y, the
corresponding function taking the value 1 on 'Y and 0 on all other
paths.
In practice, we will not use this "functional" terminology, but just
write I-chains as formal linear combinations of paths. Either way,
specifying a I-chain is the same as specifying a finite number of paths,
and assigning an integer to each. It should be emphasized that in this
definition two paths are the same only if defined by exactly the same
mapping. However, the following problem indicates some common
variations that are possible.

Problem 6.1. Call two paths equivalent if they differ by a monotone

increasing reparametrization. Show that this is an equivalence rela-
tion. Show how to define an abelian group of I-chains by using equiv-
alence classes of paths instead of paths. Do the same if paths are
called equivalent when they are homotopic with the same endpoints.

4 If you are familiar with the language from algebra, the chains are elements of the
free abelian group on the set of nonconstant paths. See Appendix C for more about
free abelian groups.
80 6. Homology

Some particular paths and I-chains will be important. For any two
points P and Q, the straight path from P to Q will be the path
'Y(t) = P + t (Q - P) = (1 - t) P + t Q, O::5t::5l.
If R is a bounded rectangle with sides parallel to the axes, the
boundary aR of R is the I-chain 'YI + 'Y2 - 'Y3 - 'Y4, where each 'Yi is
the straight path shown:

(a,d) ,13 (b,d)

/

14 / , R /r'- 12

,
/
(a,c) (b,c)
11

If R = [a, b] X [c, d], then 'YI is the straight path from (a, c) to (b, c),
'Y2 from (b, c) to (b, d), 'Y3 from (a, d) to (b, d), and 'Y4 from (a, c) to
(a, d).
If D is a disk of radius r around a point P, the boundary 'Y = aD
is the counterclockwise path around the circle
"(t) = P + (rcos(21Tt), rsin(21Tt»,
We want to define the notion of a closed I-chain, also called a 1-
cycle. This should mean that each point occurs as many times as an
initial point as it does as a final point, counting multiplicities cor-
rectly. For example, any closed path, such as the preceding aD, is a
closed I-chain, and the boundary aR of a rectangle is closed. The
precise definition is as follows. For 'Y = nl'Yl + ... + nr'Yr, if 'Yi is a
path from Pi to Qi' we define 'Y to be a closed I-chain if for each
point T occurring as a starting or ending point of any 'Yi'

Exercise 6.2. Verify that the sum or difference of closed I-chains is

closed.

Define a O-chain in U to be a formal finite linear combination

mlP I + ... + msPs of points in U, with integer coefficients. Pre-
cisely, it is a function from U to 71. that is zero outside a finite set (or
6a. Chains, Cycles, and HoU 81

an element of the free abelian group on the set of points of U). In

practice it means to pick out a finite set of points and assign a positive
or negative multiplicity to each. For any I-chain 'Y = nl'Yl + ... + nr'Y"
where 'Yi: [0, I]~ U is a path, define the boundary a'Y of'Y to be the
O-chain
a'Y = n.('Y.(1) - 'Yl(O)) + ... + nr('Yr(1) - 'Yr(O)).
The preceding definition of a closed I-chain can be said simply in
this language: a I-chain is closed exactly when its boundary is zero.
Let ZoU be the group of O-chains. A O-chain ~ is called a O-bound-
ary if there is a I-chain 'Y such that ~ = a'Y. These O-boundaries form
a subgroup of ZoU which is denoted by BoU. For example, if P and
Q are in the same component of U, then the O-chain Q - P is in BoU,
since it is the boundary of any path from P to Q. The quotient group
ZoU /BoU is called the Oth homoLogy group of U, and is denoted HoU:
HoU = ZoU /BoU.
We will see that although the groups ZoU and BoU are large (even
uncountable), the quotient group is small: it simply measures how
many connected components U has:

Proposition 6.3. The group HoU is canonically isomorphic to the free

abelian group on the set of path-connected components of u.
Proof. Let F be the free abelian group on the set of path-connected
components of U. The map that takes a point to the path component
containing it determines a surjective homomorphism from ZoU to F.
We claim that the kernel of this homomorphism is exactly the group
of boundaries BoU. This will conclude the proof, since such a ho-
momorphism determines a canonical isomorphism of ZoU /BoU with
F (see Appendix C). Any boundary is in the kernel, since the end-
points of a path must be in the same component. Conversely, if a
O-cycle is in the kernel, the total of the coefficients appearing in front
of points in any given component must be zero. Such a O-cycle can
be written (not necessarily uniquely) in the form L(Qi - Pi), where,
in each term, Pi and Qi are in the same component. As we saw before
the proof, such a O-cycle is a boundary. 0
We will use this proposition to determine the number of connected
components of U, by finding other ways of calculating HoU. If we
show that HoU has rank n, we will know that U has exactly n con-
nected components. This depends on the algebraic fact that a free
abelian group has a well-defined rank; this is proved in Appendix C.
82 6. Homology

(Alternatively, one could replace all the integer coefficients in all our
O-chains and I-chains by real numbers. Then we would find that HoU
is a real vector space of dimension n, where n is the number of com-
ponents, and appeal to the fact that a vector space has a well-defined
dimension.)
There is a homomorphism from HoU to the integers 71., defined by
the map that takes each connected component of U to I. In other
words, it takes the class of a O-cycle ~ = "i.n;P; to the sum "i.n; of the
coefficients. This is called the degree homomorphism. It is an iso-
morphism exactly when U is connected.

6b. Boundaries, H\U, and Winding Numbers

The group of I-chains on U is denoted CIU. The subgroup of closed

I-chains, or I-cycles, is denoted ZI U. There are some closed I-chains
in U, called I-boundaries, that playa particularly simple role. They
will tum out to be exactly those I-chains for which winding numbers
around points not in U vanish, and for which all integrals of closed
I-forms in U also vanish. These come from boundaries of continuous
mappings f from a square R = [0, 1] x [0, 1] into U. For such a map-
ping, define the I-chain ar by the formula

af = 'VI + 'V2 - 'V3 - 'V4'

where 'VI> 'V2' 'V3, and 'V4 are the paths obtained by restricting r to the
four sides of the square, as in §3b. (We should note here that one or
more of these four paths 'VI could be constant paths, in which case
we omit them from the formula for af.) We call a I-chain 'V a bound-
ary, or a boundary I-chain, or I-boundary, in U if it can be written
as a finite linear combination (with integer coefficients) of boundaries
of such maps on rectangles. Two closed I-chains in U are homologous
if the difference between them is a boundary in U.
We will need to know that some other I-chains are boundaries. The
following lemma considers what happens when one reparametrizes,
subdivides, or deforms a path.

Lemma 6.4. (a) Let 'Y: [0, 1] ~ U be a path. Let '1': [0, 1] ~ [0, 1] be
a continuous function. If '1'(0) = 0 and '1'( 1) = 1, then 'Y - 'YO 'I' is a
boundary in U; if '1'(0) = 1 and cp( 1) = 0, then 'Y + 'YO 'I' is a boundary.
(b) Let y. [0, l]~ U be a path, let 0:$ c:$ 1, and let (F and T be
6b. Boundaries, RIU, and Winding Numbers 83

the restriction of'Y to [0, c] and [c, 1], but scaled to be defined on
the unit interval, i. e . ,
cr(t) = -y(c· t), 0 ~ t ~ 1; T(t) = -y(c + (1 - c) . t), 0 ~ t ~ 1.
Then 'Y - (T - T is a boundary in U.
(c) If'Y and l) are paths in U that are homotopic, either as paths
with the same endpoints, or as closed paths, then 'Y - l) is a boundary
in U.
Proof. Part (a) was proved in the course of proving Corollary 3.9.
Part (b) is trivial if c = 0 or c = 1. Otherwise, we construct a mapping
f: [0,1] x [0, l]~ U as indicated in the following diagram:

y(l)

A little scratch-work in analytic plane geometry produces the formula

for f:
f( ) = {-y(t) if s ~ t,
t, s -y(c. s + (1 - c) . t) if s ~ t.

Note that the two expressions for f agree where s = t, so they define
a continuous mapping. Moreover, f(t,O) = -y(t), f(1, s) = -y(1),
nt, 1) = T(t), and f(O, s) = cr(s), so af = -y - T - cr, which proves (b).
Part (c) follows directly from the definition. 0

Exercise 6.S. Show that if U is starshaped, then every closed I-chain

on U is a I-boundary.

The I-boundaries form a subgroup, denoted B1U, of the group Z,U

of l-cycles. The quotient group ZIU /B,U is called the first homology
group of U, and is denoted HIU. One of our aims in this book will
be to study and use this group. We will see that, as the Oth group
84 6. Homology

BoU measures the simplest topological fact about U-how many con-
nected components it has-the 1st group HIU measures how many
"holes" there are in U. Two closed I-chains are homologous exactly
when they have the same image in HIU, in which case we say that
they define the same homology class.
The support Supp('Y) of a I-chain 'Y in the plane is the union of the
supports of the paths that appear in 'Y with nonzero coefficients. If P
is a point not in the support, we define the winding number of 'Y
around P to be

Proposition 6.6. If Y is a closed I-chain, then,Jor any P not in the

support of y, W(y,P) is an integer.
Proof. Let 'Y = nl'YI + ... + nr'Yr be any I-chain, with 'Yj a path. For
each point T that occurs as an endpoint of any of the paths 'Yj, choose
an angle {tr for T with respect to P. (Such an angle is measured coun-
terclockwise from a horizontal line to the right from P; it is deter-
mined only up to adding multiples of 2'IT.) If 'Yj is a path from Pj to
Qj, then W('YioP) = (l/2'IT)({tQi - {tp) + N j for some integer N j. There-
fore,

W('Y,P)

Suppose a'Y = L~=I nj(Qj - Pj) = mlTI + ... + msTs. Then we have
I
... + ms{tr) + L njNj .
r
W('Y,P) = -(ml{tr, +
2'IT j=1

In particular, if'Y is closed, i.e., a'Y = 0, then the first sum vanishes,
so W( 'Y, P) is an integer. 0

Lemma 6.7. If y is a I-boundary in ~2 , {P}, then W( y, P) = o. If two

I-chains differ by a I-boundary in ~2, {P}, then they have the same
winding number around P.
Proof. If 'Y = ar is the boundary of a map from [0, 1] x [0, 1] into
~2, {P}, Theorem 3.6 implies that W('Y, P) = O. For a general bound-
ary 'Y=LnlaC), W('Y,P) = LnjW(aC,P) =0. 0
6c. Chains on Grids 85

Proposition 6.8. If 'Y is a closed I-chain on ~2, then the function

P ~ W( 'Y, P) is constant on connected components of ~2 \ Supp( 'Y),
and vanishes on the unbounded component.
Proof. To show that the function is locally constant, it suffices to
show that it is constant on a disk D about a point P that does not
meet the support of 'Y. We want to show that W('Y, P) = W('Y, Q), with
Q a point of D. Let v be the vector from P to Q. Let 'Y = ~~=lni'Yi'
with each 'Yi a path. We know from Exercise 3.4 that
W('Y,P)=W('Y+v,P+v)=W('Y+v,Q), where 'Y+v is the I-chain
~~=lni('Yi + v), so it is enough to show that

W('Y, Q) = W('Y + v, Q).

By the lemma, this follows if we verify that the difference of the
I-chains 'Y and 'Y + v is a I-boundary in ~2 \ {Q}. Define mappings C
from [0, 1] X [0, 1] to ~2 \ {Q} by the formula
C(t, s) = + s· v.
'Yi(t)
The boundary of C has the paths 'Yi and 'Yi + v on the bottom and top,
and straight line paths from endpoints of 'Yi to their translations by v,
along the sides. The fact that 'Y is closed means that these straight
line paths from the sides cancel in the sum ~~=Ini(ar;). Therefore,
, "
2:i=1 ni(an = 2:i=1 ni'Yi - 2:i=1 ni('Yi + v) = 'Y - ('Y + v).

This shows that 'Y - ('Y + v) is a boundary in ~2 \ {Q}. and concludes

the proof that the function is locally constant. On the unbounded com-
ponent, we may take the point P far to the left of the support of 'Y, so
that there is one angle function {} defined on all the paths occurring
in 'Y. With 'Y = ~;= lni'Yi, and 'Yi a path from Pi to Q;, we have
W('Y, P) = (l/2'lT)~~= In;({}(QJ - {}(PJ), which is zero since the
boundary of y is zero. 0

6c. Chains on Grids

A grid G will be a finite union of lines in the plane, each parallel to
x-axis or the y-axis, with at least two horizontal and two vertical lines.
These lines cut the plane into a finite number of rectangular regions,
some bounded and some unbounded. By a rectangular I-chain for a
given grid we shall mean a I-chain fJ.. of the form fJ.. = nlO"I + ... + n,O""
86 6. Homology

where each (1j is a straight path along one of the sides of the bounded
rectangles, from left to right or from bottom to top.

-+--f---_ bounded

-unbounded
---+-++--+-+-+---+--~

Let R I, . . . ,Rr be the bounded rectangles for the grid G, num-

bered in any order, and choose a point Pi in the interior of R for each
i. The next lemma proves an elementary but important fact: a closed
rectangular I-chain is completely determined by its winding numbers
about these points.

Lemma 6.9. If /L is a closed rectangular I-chain for G, then

Jl. = n.iJR. + ... + n,iJR, ,
where ni = W(/L, Pi)'
Proof. Since the winding number of iJRi around Pj is 1 if i = j and 0
otherwise, the winding number of each side of the displayed equation
around each Pj is the same. Let or = Jl. - z,niiJR i be the difference, which
has winding number zero around each Pj' We must show that or is
zero as a I-chain. Suppose an edge (1 occurs in or with nonzero coef-
ficient m. Suppose that (1 is a vertical or horizontal line between two
rectangles R and R' of the grid, with R to the left of or above R'.
Assume first that R is bounded:

R R'
/
R'

The trick is to consider the closed I-chain or' = or - m' aR. Let P and
6c. Chains on Grids 87

pi be interior points in Rand R', respectively. Since W(aR,p) =1

and W(aR, Pi) = 0, we have
W(T',P) = W(T,P) - m· W(aR,p) = -m,
W(T', Pi) = W(T, Pi) - m· W(aR, Pi) = 0.
But the edge IT does not appear in T', so P and pi belong to the same
connected component of the complement of the support of T', which
implies by Proposition 6.8 that W(T ' , P) = W(T' , Pi), a contradiction.
If R is unbounded, and R' is bounded, the argument is similar, using
T' =T+m·aR' . 0

Next we need an approximation lemma that will assure that, for

the purposes of winding numbers and integration, all paths and I-chains
can be replaced by rectangular I-chains.

Lemma 6.10. Let 'Y be any I-chain in an open set U. Then there is
a grid G, and a rectangular I-chain JL for G, with the support of JL
contained in U, such that 'Y - JL is a I-boundary in U. If 'Y is closed,
then JL is also closed.
Proof. By Lemma 6.4(b) we can subdivide any of the paths that occur
in 'Y, and the difference between 'Y and the I-chain with subdivided
paths will be a boundary. Using the Lebesgue lemma as usual to sub-
divide, we may therefore assume that each path 'Yj occurring in 'Y maps
[0,1] into some open rectangle Uj contained in U. Let Pj and Qj be
the starting and ending points of 'Yj. Take any grid G that has a vertical
and horizontal line passing through each Pj and each Qj. Let j.Lj be a
rectangular I-chain for G that goes from P to Q;, involving only edges
j

on the closed rectangle with comers at Pj and Qj; in particular, the

boundary of j.Lj is Qj - Pj. (Note that j.Lj is in Uj, so j.Lj is a chain in
U.)

1 1
Qi Pi

_( Ie ~ I'
'\. '\.
"""'"' ----.-

)
I I 1'1

'-
-1

Qi

~
I 1'1
Pi
"- "-
I I I
1 1 1
88 6. Homology

It suffices to verify that "yj - jJ.j is a boundary in Vj, for if "Y = 'Lnj"yj,
then jJ. = 'LnjjJ.j will be the required rectangular I-chain, with
"Y - jJ. = 'Lnj("Yi - jJ.;) a boundary. But since "Yi - jJ.; is a closed I-chain
on a starshaped open set V;, this follows from Exercise 6.5. D

We are now ready to prove the main goal of this chapter: that the
geometric condition for I-cycles to be homologous in an open set is
equivalent to the numerical condition of having the same winding
number around all points outside the open set.

Theorem 6.11. Suppose yand 5 are closed I-chains on an open set

V in the plane. Then the following are equivalent:
(1 ) y and 5 are homologous, i. e ., y - 5 is a boundary in V; and
(2) W(y,P) = W(5,P)for all points P not in V.
Proof. We saw in Lemma 6.7 that (1) implies (2). For the converse,
by looking at T = "Y - 8, it suffices to show that if T is a closed I-chain
such that W(T,P) = 0 for all P rt. V, then T is a boundary. By Lemma
6.10, there is a closed rectangular I-chain jJ. for some grid G so that
T - jJ. is a boundary in V. By Lemma 6.9, jJ. = 'LniJRi> where
nj= W(jJ.,PJ.
We claim that nj = 0 unless the entire closed rectangle R j is con-
tained in V. For if Q is a point of R; that is not in V, then W(T, Q) = 0
by assumption, so W(jJ., Q) = 0 by Lemma 6.7. Since the straight line
from P j to Q lies in the complement of the support of jJ., it follows
from Proposition 6.8 that W(jJ., Pi) = W(jJ., Q), so ni = W(jJ., Pi) = O.

Pi

Rj
u

It follows that jJ. = 'LniaR;, with each R; contained in V, and such a

I-chain is visibly a boundary in V. So T = (T - jJ.) + jJ. is also a
boundary. D

For example, if V = 1R2 \ {P} is the complement of a point, two

I-cycles are homologous in V exactly when their winding numbers
6d. Maps and Homology 89

around P coincide. In other words, the winding number gives an iso-

morphism
b1 1-4 W('Y,P),
where b1 denotes the class of a I-cycle 'Y in the homology group.
Exercise 6.12. Suppose V = 1R2 \ {PI> ... , Pn } is the complement of
n points in the plane. Show that the mapping that takes a closed 1-
chain 'Y to (W( 'Y, PI), . . . , W( 'Y, P n» determines an isomorphism of
HIV with the free abelian group lLn.

Exercise 6.13. State and prove the analogue of Theorem 6.11 when
'Y and & are arbitrary I-chains in V with the same boundary.

6d. Maps and Homology

If 'Y = nl'Yl + ... + nr'Yr is a I-chain in an open set V, with 'Yj paths,
and F: V ~ V'is a continuous mapping from V to another open set
V', define F *'Y to be the I-chain in V' defined by
F *'Y = nl(F 0 'YI) + ... + nr(F 0 'Yr).

F also maps O-chains in V to O-chains in V': F*('i.m;Pj) = 'i.m;F(P j).

Exercise 6.14. Show that 'Y 1-4 F *'Y is a homomorphism from the group
of I-chains in V to the group of I-chains in V'. If 'Y is a closed 1-
chain, show that F*'Y is also closed. Show in fact that F*(rJ'Y) = rJ(F*'Y)
for any I-chain. Show that if'Y is a boundary in V, then F*'Y is a
boundary in V'.

From this exercise and Theorem 6.11, we deduce the following

fact, which is not so obvious from the definition of the winding num-
ber:

Proposition 6.15. If 1 and () are closed I-chains in V with the same

winding number around all points not in V, then F *1 and F *() are
closed I-chains in V' with the same winding number around all points
not in V'. 0

For example, take V to be the region inside one disk D and outside
a disjoint union of closed disks AI> ... ,An contained in D. Let 'Y
90 6. Homology

be a circular path in U containing the disks Ai, and let 'Yi be a circular
path around Ai'

Since'Y and L'Yi have the same winding numbers around each point
not in U, we conclude that for any F: U ~ U', and any Q ft. U',
W(F*'Y,Q) = W(F°'Yl>Q) + ... +W(F°'Yn,Q).

Here is an application, which is a substantial generalization of Ex-

ercise 4.9.

Corollary 6.16. Let F: U~ [R2 be a continuous mapping, and sup-

pose y is a closed I-chain in U such that W( y, P) = 0 for all points
P not in U. Let Q be a point in [R2, not in F(Supp(y», such that
W(F *y, Q) ~ O. Then there is a point P in U such that W( y, P) ~ 0
and F(P) = Q.
Proof. Let Z = p-l(Q), a closed (conceivably empty) subset of U dis-
joint from Supp('Y) . Apply the proposition to the open set if = U \ Z,
the restriction of F to if, and the I-chains 'Y and 8 = O. If the assertion
of the corollary is false, then W('Y,P) = 0 for all P in Z, so W('Y,P) = 0
for all P not in if, so 'Y is homologous to 0 in if. But since Q is
not in F(if), the fact that W(F *'Y, Q) ~ 0 shows that F *'Y is not ho-
mologous to 0 in F(if), contradicting Proposition 6.15. 0

For example, with U the complement of the gray area, the I-chain
'Y that is the sum of the paths shown has a nonzero winding number
only around points in the striped area, so any point with W(F *'Y, Q) ~ 0
would be the image of a point from one of the striped regions.
6e. The First Homology Group for General Spaces 91

Problem 6.17. Under the conditions of the corollary, assume that, at

each point P of rl(Q), the local degree of Fat P is defined. Show
that

W(F *'Y, Q) L
PEF- 1(Q)
degp(F) . W( 'Y, P) .

Problem 6.18. Let E be the closed set obtained from a closed disk
D by removing the interiors of k disjoint disks D I , •• • , Dk contained
in D, for some k~ O. Let C be the boundary of D, and Cj the bound-
ary of D j • Suppose F: E~ ~2 is a continuous mapping such that for
each point P in any of the boundary circles, the vector from P to
P + F(P) is not tangent to that circle. Show that, if k ¥- I, there must
be a point Q in E with F(Q) = O. What about the case k = I?

6e. The First Homology Group for General Spaces

The general definitions of this chapter make sense without change for
any topological space, although, of course, one does not have wind-
ing numbers in arbitrary spaces. Any topological space X has abelian
groups ZOX of O-chains, Be)( of O-boundaries, with Oth homology group
He)( = ZOX/Be)(, which is canonically isomorphic to the free abelian
group on the set of path-connected components of X. Similarly, one
has the abelian group C IX of I-chains on X, with the subgroups ZIX
of I-cycles and I-boundaries BIX, and the 1st homology group
HIX = ZIX/BIX. There are no changes in the definitions, other than
replacing a "U" by an "X."
92 6. Homology

Exercise 6.19. Show that a continuous mapping F: X ~ Y determines

a homomorphism from ZIX to ZIY taking BIX to BIY.

This determines a homomorphisms of the quotient groups, denoted

F*: HIX ~ H1Y.

Exercise 6.20. Show that if F: X ~ Y and G: Y ~ Z are continuous,

then (GoFh = G*oF* as homomorphisms from HIX to HIZ. If F is
the identity map on X, show that F * is the identity map on HIX.

The result of this exercise is expressed by saying that the diagram

F*
HIX .. HtY

(G-F~ /0.
HtZ
commutes: starting with an element in the upper left group H1X, map-
ping it to HIZ by either route gives the same answer.
For example, if Z = X, and F and G are homeomorphisms that are
inverses to each other, so that G °F is the identity map of X, then
(GoFh is the identity map of HIX, so the composite
F* G*
HIX ~ H1Y ~ HtX
is the identity map on HIX. Similarly FoG is the identity map on Y,
so G* 0 F * is the identity map on HIY. It follows that F * and G* are
inverse isomorphisms between HIX and HIY. In particular,

Proposition 6.21.lf X and Yare homeomorphic, then HIX and HIY

are isomorphic abelian groups.

Similarly, if Y is contained in X, and r: X ~ Y is a continuous re-

tract, the identity map on H1Y must factor into a composite of ho-
momorphismsHtY~HIX~HIY. For example, ifHtX= 0, andH1Y#0,
this shows that there can be no such retract.

Exercise 6.22. Compute HtX for X a circle and for X a disk, and
show again that a circle is not a retract of the disk it bounds.

Two continuous maps F and G from X to Y are homotopic if there

6e. The First Homology Group for General Spaces 93

is a continuous mapping H: X x [0, 1] ~ Y such that F(x) = H(x, 0)

and G(x) = H(x, 1) for all x in X.

Proposition 6.23. If F and G are homotopic maps from X to Y, then

F* = G*.
Proof. If 'Y = "Zn;'Y; is a I-cycle on X, set C(t, s) = H('Y;(t) , s). Then
F *'Y - G*'Y = "Zn;aC, the other terms canceling each other since 'Y is
a cycle. 0
A subspace Y of a space X is called a deformation retract if there
is a continuous retract r: X ~ Y such that the identity map from X to
X is homotopic to the map i 0 r, where i is the inclusion of Y in X. A
space X is called contractible if it contains a point that is a defor-
mation retract of X.

Exercise 6.24. (a) Show that the circle SI C ~2 \ {O} is a deformation

retract. (b) Give an example of a retract that is not a deformation
retract. (c) Show that any two maps from any space to a contractible
space are homotopic. In particular, every point in a contractible space
is a deformation retract of the space.

Exercise 6.25. Show that if Y is a deformation retract of X, then the

map from HIY to HIX determined by the inclusion of Y in X is an
isomorphism. Show that HI(X) = 0 if X is contractible.

Exercise 6.26. Show that F: X ~ Y determines a homomorphism from

ZoX to ZoY taking BoX to BoY, and so a homomorphism, also denoted
F*, from HoX to HoY. Verify the analogues of the assertions in Ex-
ercises 6.19 and 6.20, and Propositions 6.21 and 6.23.

If X is a subspace of the plane that is not open, one does have a

notion of the winding number around points not in X, but the situation
is more complicated, as the following problems indicate.

Problem 6.27 (For those who know the Tietze extension theorem).
Suppose X and X' are closed subsets of the plane, and F: X ~ X' is
a continuous mapping. Suppose 'Y and 8 are two closed I-chains on
X such that W('Y,P) = W(8,P) for all P not in X. Show that
W(F *'Y, P') = W(F *8, P') for all P' not in X'.

Problem 6.28. Let X be a closed set in the plane. Show that if'Y and
8 are homologous I-cycles on X, then W('Y,P) = W(8,P) for all P not
in X. Is the converse true?
 PART IV

VECTOR FIELDS

A mapping from an open set in the plane to the plane can be regarded
as a vector field, and winding numbers can be used to define the index
of a vector field at a singularity. The ideas of Chapter 6 can be used
to relate sums of indices to winding numbers around regions. This is
applied to show that vector fields on a sphere must have singularities:
one cannot comb the hair on a billiard ball. The same ideas are used
in the next chapter to study more interesting surfaces. These chapters
are inserted here to indicate some other interesting things one can do
with winding numbers; a reader in a hurry to move on can skip to
Part V or VI.
Chapter 8 sketches how some of the ideas we have studied in the
plane and on the sphere can be studied on more general surfaces. It
gives us a first chance to study some spaces other than plane regions
and spheres. In particular, we see how the "global" topology of the
surface puts restrictions on the "local" data of indices of a vector
field. We use this to discuss the Euler characteristic of a surface.
Some of the arguments in this section will depend on geometric con-
structions that will only be sketched, usually by drawing pictures
showing how to deform one surface into another. Later in the chapter
we discuss briefly what it would take to make these arguments rig-
orous, and later in the book we take up the study of surfaces more
systematically.
CHAPTER 7

Indices of Vector Fields

7a. Vector Fields in the Plane

We want to look at continuous vector fields on an open set U in the
plane, but allowing them to have a finite number of singularities. A
singularity will be a point at which either the vector field is not de-
fined, or a point where it is defined and is zero. A vector field on U
with singularities in Z will therefore be a continuous mapping
V: U \ Z ~ 1R2 \ {O},
where Z is a finite set in U. For vector fields arising from flow of a
fluid, the singularities may arise from "sources," where fluid is en-
tering the system, or "sinks," where it is leaving, or some other dis-
continuity.
Given such a vector field V, to each point P in U one can define
an integer called the index of Vat P, denoted IndexpV. To do this,
take a disk Dr of some radius r about P that does not meet Z at any
point except (perhaps) P. Let C be the boundary of this disk. The
restriction Vic, of V to C is a mapping from this circle to 1R2 \ {O}, so
it has a winding number (by §3d). Define the index to be this winding
number:
IndexpV = W(vlc" 0).
Lemma 7.1. (a) This definition is independent of choice of r.
(b) If P is not in Z, then IndexpV = O.

97
98 7. Indices of Vector Fields

Proof. The proof of (a) is the same as that of Lemma 3.28, and (b)
follows from Proposition 3.20. 0

From the definition, the index at P depends only on the restriction

of V to an arbitrarily small neighborhood of P. Here are some ex-
amples of vector fields with singularities at the origin, with the cor-
responding indices; the calculations are left as exercises. Instead of
drawing vectors at many points, it is more useful to draw some flow
lines, i.e., curves that are tangent to the vector field at each point.

Vector field V(x, y) Index at 0

• o
)

(2) (-y, x)

(3) (y, -x)

( -y X)
(4) x 2 + l' x 2 + l
7a. Vector Fields in the Plane 99

(5) (x,y) 1

(6) (-x, -y) 1

(7) (y,x) -1

Note that (2) and (3) are opposite as vector fields, as are (5) and (6),
but have the same index. Similarly, (4) is a positive multiple of (2).
These are special cases of Exercise 7.4, which shows that the mag-
nitude and sign of the vectors does not affect the index (which ex-
plains why the flow lines, without even their sense of direction, de-
termine the indices).

Exercise 7.2. Construct, for each integer n, a vector field with a sin-
gularity at the origin with index n.

Exercise 7.3. Let Vo and VI be continuous vector fields in a punctured

neighborhood U of P, and suppose there is a continuous mapping
H: U X [0, 1] ~ /R 2 such that
H(Q X 0) = Vo(Q), for all Q in U;
H(Q X t)~O for all Q and all 0 :s; t :s; 1.
Show that Vo and VI have the same index at P.
100 7. Indices of Vector Fields

Exercise 7.4. Show that if p is any continuous function defined in a

punctured neighborhood of P that is always positive or always neg-
ative in this neighborhood, then the index at P of p . V is the same as
the index at P of V.

The following proposition relates the behavior of a vector field along

the boundary of a region to the indices of the vector field at singular
points inside:

Proposition 7.5. Let V be a vector field with singularities in U. Sup-

°
pose 'Y is a closed I-chain in U whose support does not meet the
singular set Z of V, such that W( 'Y, P) = for all P not in U. Then

W(V*'Y,O) = LW('Y,P)·lndexpV.
PEZ

Proof. Let Z = {Ph' .. , Pr}, and let Dh ... , Dr be disjoint closed

disks centered at the points Ph ... , P" all contained in U. Let 'Yj
be the standard counterclockwise path around the boundary of D j , and
let nj = W('Y,P j). Then 'Y and nl'Yl + ... + nr'Yr have the same wind-
ing number around every point outside U \ Z. It follows from Prop-
osition 6.15 that
W(V*'Y,O) = n1W(V°'Yl,0) + ... + nrW(V°'Y" 0) ,
and this is the assertion to be proved. o
Corollary 7.6. If W(V *'Y, 0) # 0, then V must have at least one non-
vanishing index at a point P with W( 'Y, P) ~ 0. 0

The simplest case of the proposition is:

Corollary 7.7. If U contains a closed disk D, and V has no singu-

larities on the boundary circle C of D. Then

W(vlc, 0) = L IndexpV.
PED

If W(vlc, 0) ~ 0, V must have a singularity with nonvanishing index

inside D. 0

Problem 7.S. Generalize this discussion to allow the singularity set

Z to be infinite but discrete, i.e., Z is a closed subset of U such that
each P in Z has a neighborhood Up in U such that Up n Z = {P}. Show
that the sum in Proposition 7.5 is automatically finite.
7b. Changing Coordinates 101

Problem 7.9. Letfbe a Cf6'" function defined in V, defining a gradient

vector field V = grade!) on V. A point P is a critical point for f if
the gradient vanishes at P, and P is called nondegenerate if the "Hes-
sian" at P,

if
-(P)'-(P) -
a2f (a 2f
-(P)
)2
ax2 a/ axay
is not zero. (a) Show that, if P is a nondegenerate critical point, then

Indexp(V) = {_ 11 if f has a local maximum or minimum at P,

otherwise (when P is a saddle point).

(b) Suppose D is a disk in V with boundary C, andfhas only non-

degenerate critical points in D, with none on the boundary, and f is
constant on C. Show that the number of local maxima plus the num-
ber of local minima is one more than the number of saddle points.
"On a circular island, the number of peaks plus the number of valleys
is one more than the number of passes. "

7b. Changing Coordinates

Later we want to define the index of a vector field on a surface other
than an open set in the plane. To do this, the essential point is to
know that a change of coordinates does not change the index. This
result, which is intuitively obvious from pictures of vector fields, takes
some care to state and a little work to prove. To state it, suppose
'P: V ~ V'is a diffeomorphism from one open set in the plane onto
another; that is, 'P is Cf6"', one-to-one, and onto, and the inverse map
'P -I: V' _ V is also a Cf6x mapping. At any point P in V, we have the
Jacobian matrix

-(P)
au -(P)
au]
_ ax ay
J~,p - [ av av '
-(P) -(P)
ax ay

where 'P(x,y) = (u(x,y), v(x,y» in coordinates. This gives a linear

mapping from vectors in jR2 to vectors in jR2 (see Appendix C). If V
is a continuous vector field in V, define the vector field 'P* V in V'
102 7. Indices of Vector Fields

by the formula
('P*V)(P') = J",AV(P» ,
where P is the point in U mapped to P' by 'P, i.e., P = 'P-1(P'). If V
has singularities in the set Z, V' will have singularities in 'P(Z).

Lemma 7.10. With V and 'P*V as above, then,for any Pin U,

Index",(pl'P* V) = IndexpV .

The proof of this lemma is given in Appendix D. We will also want

to compare the indices of two different vector fields on a surface. The
following lemma will be used to reduce to the case where they agree
in a neighborhood of some point. We say that two vector fields V
and Wagree on a set A if V(P) = W(P) for all P in A. The proof is
also in Appendix D.
Lemma 7.11. Suppose V and Ware continuous vector fields with no
singularities on an open neighborhood U of a point P. Let D CUbe
a closed disk centered at P. Then..!here is a vector field if with no
sJ!tgularities on U such that: (i) V and V agree on U \ D; and (ii)
V and W agree on some neighborhood of P.

7c. Vector Fields on a Sphere

A vector field V on a sphere S assigns to each point P in S a vector
V(P) in the tangent space TpS to S at P, the mapping from P to V(P)
being continuous. If S = S2 is the standard sphere, and P = (x,y, z),
the tangent space is
TpS = {(a, b, c): (a, b, c)· (x,y, z) = ax + by + cz = O}.
As before, we may allow a finite set Z of singularities. In fact, one
of our goals is to show that any vector field on sphere must have
singularities.
7c. Vector Fields on a Sphere 103

We want to flatten out the sphere, say by stereographic projection

from a point on the sphere, so that the vector field determines a cor-
responding vector field on the plane, and we will use what we know
about vector fields on the plane.

(0,0,1)

.--------+----~-- y

We need the following, whose proof is left as an exercise:

Lemma 7.12. The inverse cp of this polar coordinate mapping takes

(x, y) in the plane to

( 2x 2y X2+l-1) S2,
'P(x, Y) = x 2 + l + l' ~ + l + l' x 2 + l + 1 in
The Jacobian matrix J""p of cp at P = (x,y) maps 1R2 one-to-one onto
the tangent space to S2 at the point cp(P), If V is a vector field on S2,
and cp*V is the vector field on 1R2 defined by the equation
J'P,p«'P*V)(P)) = V('P(P)) ,
then cp*V is continuous at P if V is continuous at cp(P).

Now suppose V is a continuous vector field with no singularities

on S2. Then 'P*V is a vector field on 1R2 with no singularities. Let C r
be a large circle centered at the origin in the plane. We know from
Corollary 7.7 that the winding number of 'P*V around Cr must be
zero. To get a contradiction, we must use the fact that V is continuous
and nonzero at the north pole. For r large, Cr can be thought of as a
small circle around the north pole. The winding number of 'P*V around
such a circle is not zero, even though it comes from a vector field
104 7. Indices of Vector Fields

that is not zero in the disk near the north pole. This can be seen by
unraveling what happens to a vector field on the sphere when the
sphere is flattened out. Think of the vectors as attached to a small
band, which is turned over:

The winding number of this vector field around this circle is 2, which
shows that no such vector field can exist.

Problem 7.13. Give an analytic proof of the fact that this winding
number is 2.

This shows more than the fact that a vector field on a sphere must
have singularities. If V is a vector field on S2 with singularities in a
finite set Z that does not include the north pole, we can define the
index IndexpVat a point P of Z to be the index of <p*V at the cor-
responding point <p -I(p). Then the proof shows that, for any such
vector field V on S2 ,

2: IndexpV = 2.
PEZ

In order to include the north pole in these considerations, we also

consider a stereographic projection from the south pole. To make the
orientations match in our two charts, we first reflect in the x-axis. So
we define $: 1R2 ~ S2 by
(_ 2x - 2y - x 2 -l + 1)
$(x,Y) = \,x2+l+ l' x +/+ l' x2 +/+ 1 .
2

Exercise 7.14. Show that the composite <p -1 0 $ takes z = x + iy to

l/z.
7c. Vector Fields on a Sphere 105

Now for any vector field V with a finite number of singularities on

S2, and any point P, we can define IndexpVas either the index of
'I'*V or of I/I*V at the corresponding point. It follows from Lemma
7.10 that these indices agree, if both are defined. In fact, we could
use a stereographic projection from any point; all the coordinate trans-
formations as above are C(6"".

Proposition 7.15. For any vector field with singularities V on S2,

2: IndexpV = 2.
PEZ

Proof. Instead of arguing as we did above, we can argue in two steps:

Step 1. There is a vector field Von S2 with LpEZ Indexp(V) = 2. For

example, if W is the vector field on ~2 given in (8), i.e.,
W(x, y) = (x2 -l, 2xy), then V = '1'* W is a vector field on the com-
plement of the north pole with one singularity of index 2 at the south
pole. A short calculation shows that V extends continuously to the
north pole, with value there the vector (-2,0,0).

Step 2. We show that the sum of the indices of any two such vector
fields V and Won S2 is the same. Let P be a point where neither has
a_singularity. By Lemma 7.11, replacing V by another vector field
V with the same indices as V, we can assume that V and Wagree in
some neighborhood of P. Then using stereographic projection from
P, one has two vector fields on the plane that agree outside some
large disk that contains all the singularities of either vector field. Tak-
ing a larger circle Cn their winding numbers around Cr will be the
same, and an application of Corollary 7.7 shows that the sum of their
indices is the same. 0

Exercise 7.16. Give an alternative proof of Step 1 by finding a vector

field on S2 with two singular points, each with index 1.

Problem 7.17. (a) If f S2~ ~3 is a continuous mapping, show that

there is some point Pin S2 and some real number A so thatfiP) = AP.
(b) If f S2 ~ S2 is a continuous mapping, show that there is some
point P in S2 such that fiP) = P or fiP) = - P.

Problem 7.18. Give a mathematical formulation and proof of the

statement: "On a spherical planet, the number of peaks plus the num-
ber of valleys is two more than the number of passes."
CHAPTER 8

Vector Fields on Surfaces

8a. Vector Fields on a Torus and Other Surfaces

Let us look next at a surface X which is a torus, i.e., the surface of
a doughnut. This can be realized concretely in several ways: as a
surface of revolution, by an explicit equation in 3-space, as a Carte-
sian product Sl x Sl, or by taking a square or rectangle and identifying
opposite edges:

It is clear by any of these descriptions that there are vector fields

on X that have no singularities at all. In order to state the analogue
of Proposition 7.15 for a torus, we need to define the index of a vector
field at a point on X. One way to do this is to realize X as a quotient

106
Sa. Vector Fields on a Torus and Other Surfaces 107

space of the plane ~2, identifying two points if their difference is in

the lattice 71? This amounts to identifying the opposite sides of the
unit square [0,1] x [0, 1]. We have a mapping

aEd giving a vector field V on X is the same as giving a vector field

V on ~2 that is unchanged by translation by any vector in 7l? . We
can define the index IndexpV to be the index of the corresponding
vector field V- at any point of ~ 2 that maps to P. As before, we allow
a finite set Z of singularities, and require V to be continuous outside
Z. Then we have the analogous proposition:

Proposition 8.1. For any vector field with singularities V on a

torus X,

2: IndexpV
PEZ
°.
Proof. Take a square R = [a, a + 1] x [b, b + 1] so that the image in
X of the boundary aR does ,.!lot hit the singularity set Z. Look at the
corresponding vector field V..9n a neighborhood of R. By Proposition
~5 the winding number of V around aR is the sum of the indices of
V inside R. This winding number is zero, since the vector field is the
same on opposite sides of the square. And the indices inside are the
indices of Von X. 0

For example, realizing X as a surface of revolution in space as in

the above picture, the projection of a vertical vector (0,0,1) onto the
tangent space at each point gives a vector field with singularities at
the four points with horizontal tangents. The indices at the points at
the top and bottom (where the height has local maximum and mini-
mum) are +1, while those at the two saddle points are -1.

Exercise 8.2. Show that the following formula defines a diffeomorph-

ism of ~2 /71..2 = Sl X Sl with a surface of revolution in ~3. Take 0< r < R
and define the map by

(x, y) ~ (R + rcos(21TY» . (0, COS(21TX), sin(21Tx»

+ rsin(21TY)· (1,0,0).
108 8. Vector Fields on Surfaces

Find the vector field on the plane corresponding to the above vector
field, and verify that the indices are 1, -1, -1, and 1.

We want to generalize what we have just seen from the sphere and
torus to other surfaces, in particular to the surface of a doughnut with
g holes, or a "sphere with g handles":

We will argue geometrically and rather loosely for now, and post-
pone until later a discussion of what needs to be done to make the
arguments precise and rigorous. From the second picture, taking V to
be the vector field with V(P) the projection on the tangent space TpX
of a vertical vector (0,0,1) as before, we see that there are two points
with index 1 (at the top and bottom), and 2g points with index -1
(at the saddles). This gives one vector field the sum of whose indices
is 2 - 2g. The claim is that this is always the case.

Theorem 8.3 (Poincare-Hopf). Let X be a sphere with g handles.

For any vector field V with singularities on X, the sum of the indices
of V at the singular points is 2 - 2g.
Proof. Having seen one such vector field, it is enough to show that
any two vector fields have the same sum of indices. By Lemma 7.11,
we can take a disk in X where both have no singularities, and modify
one so that they agree on such a disk. The idea of the proof is to
mimic the proof for a sphere: take a circle C in such a disk, and punch
out a smaller disk D (inside the circle) from the surface, and spread
the complement X \ D out on the plane. The two vector fields will
then have the same winding number around the curve C, so they will
have the same sum of indices by Proposition 7.5.
8a. Vector Fields on a Torus and Other Surfaces 109

For g> 0, however, this complement is not diffeomorphic (or ho-

meomorphic) to a plane domain . However, it can be realized with a
mapping <p: X \ D~ 1R2 that is a local diffeomorphism, i.e., every point
P in X \ D has a neighborhood that is mapped diffeomorphically (with
a diffeomorphic inverse) onto its image. To visualize this, look first
at the torus. The complement of a disk is formed of two bands joined
together. The mapping <p from X \ D can be visualized by picturing
the bands over the plane.

The image of the circle C goes around near the boundary. We can
define the winding number W(VlC. 0) by cutting C up into pieces , each
of which is mapped one-to-one by <p into the plane, and using the
usual definition on these pieces. We can also define the index of V
at a point P of X \ D by using the local diffeomorphism <p* to identify
V with a vector field <p*V near <pep), and defining IndexpV to be
the index of <p* V at <pcP). To finish the proof, it is enough to show
that

2:
PEXIJJ
IndexpV = W(vlc, 0) .

To see this, one can add some crosscuts, being careful not to go through
110 8. Vector Fields on Surfaces

any singularities, and apply Proposition 7.5 to the restriction of V to

each of the pieces.

As usual, the added pieces get counted twice with opposite signs, so
they cancel, and one is left with the displayed equation.
The same works for any genus g, although the visualization is a
little harder. We claim that, when a closed disk D is removed from
X, the complement can be realized over the plane as indicated (for
g= 3):

Once we have this, the same argument completes the proof of the
theorem. To realize X \ D this way, an essential point is that, as a
larger and larger disk is removed from X, the complements are all
diffeomorphic. To aid in visualization, the situation for g = 1 is re-
done this way on a separate page. The picture after that shows a disk
with "fingers," which can be sewn to the back of the above figure,
giving a sphere with g handles.
8a. Vector Fields on a Torus and Other Surfaces 111

o ~

y
112 8. Vector Fields on Surfaces

Exercise 8.4. Describe a vector field on X that has exactly 2g - 2

singular points, each with index -1.

It is time to discuss what it might take to make this sort of argument

rigorous. First, of course, one should give precise definitions of all
the objects involved: surfaces, a sphere with g handles, a vector field
on a surface, and the index of a vector field at a point. This not being
a course on manifolds, we will not go through all the details (but see
Appendix D for somE: of them). A key point, however, is that a sur-
face is covered by the images of coordinate charts 'P,,: U,,~X, that
are homeomorphisms from open sets U" in the plane to open sets
'P,,( UIX) in X, and the change of coordinate mappings 'P~ -I a 'P", defined
from part of U" to part of U~, should be '{6"". For the sphere, for
example, stereographic projections from the two poles gave coordi-
nate charts 'P and ljI, and for the torus, the mapping p: ~2~ X = Sl X Sl,
restricted to small open sets in the plane, gives charts on X.
A vector field V on X then determines a vector field V" on U", and
these are related by the condition that ('P~ -I a 'P"MV,,) = V ~ on the open
subsets where both are defined. In fact, a vector field on X can be
defined as a collection of such vector fields V"' related by these com-
patibilities under changes of coordinates. The index of V at a point
P in X can be defined as the index of V" at the point P", if 'P,,(PIX) = P.
The key Lemma 7.10 implies that this is independent of choice of
chart near P. Note that the surface X is assumed to have a differen-
tiable structure, but that the vector field is only assumed to be con-
tinuous in the complement of a finite set.
The description of surfaces via cutting and pasting, which we have
indicated in pictures, could be done explicitly in coordinates, as a
8b. The Euler Characteristic 113

large and not very pleasant exercise. For example, one can verify that
removing a slightly larger disk from a surface gives diffeomorphic
complements. The tools for doing this sort of argument properly are
developed systematically in the subject of differential topology, see
Milnor (1965), Wallace (1968), and Guillemin and Pollack (1974).

Exercise 8.5. What can you say about the number of peaks, valleys,
and passes on a planet shaped like a sphere with g handles?

Exercise 8.6. Does a sphere with g handles have the fixed point prop-
erty?

8b. The Euler Characteristic

Suppose X is a compact surface, and we have a triangulation of X.
This means that X is cut into pieces homeomorphic to triangles, fitting
together along the edges. We will have a certain number v of vertices
(points), a number e of edges (homeomorphic to closed intervals),
and a number / of faces (homeomorphic to closed triangles). These
homeomorphisms are assumed to take the ends of intervals to two
distinct vertices, and the three boundary pieces of a triangle onto three
distinct edges.

As you may have discovered in an exercise in the Preface, the

number v - e + / is independent of the triangulation:

Proposition 8.7. For any triangulation 0/ a sphere X with g handles,

v - e +/ = 2 - 2g .
Proof. The idea is to construct a vector field V on X with one sin-
gularity of index 1 for each vertex, one of index -1 for each edge,
and one of index 1 for each face. Then the proposition follows from
the Poincare-Hopf theorem. To do this, do a "barycentric subdivi-
114 8. Vector Fields on Surfaces

sion": put a new vertex in each edge, and one in each face, and con-
nect them as shown.

Then construct a vector field on X, so that in each triangle it looks

like:

Construct it first along the edges: on the old edges pointing from the
old vertices to the new ones added in the middle of the edges, and
on the new edges pointing toward the new vertices in the faces; then
fill in over the new triangles to make it continuous. If this is done,
one has a vector field V whose singularities are at the vertices, and
whose indices are: + 1 if P is an old vertex; -1 if P is a new vertex
along an edge; and + 1 if P is a new vertex in a face. There are v of
the first, e of the second, andf of the third, so by the Poincare-Hopf
theorem, v·(+1)+e·(-1)+f·(+1)=2-2g. D

The number v - e + f, for any triangulation, which is the same as

the sum of the indices of any vector field, is called the Euler char-
acteristic of the surface.

Exercise 8.8. Construct a triangulation on the sphere with g handles,

and verify the proposition for this triangulation.
8b. The Euler Characteristic 115

Problem 8.9. (a) Show that, for any triangulation of a sphere with
g handles:

(i) 2e = 3f;
(ii) e:5 1/2 v' (v - 1); and
(iii) v 2= 112(7 + V49 - 24(2 - 2g».
(b) Find lower bounds for v, e, and f for g = 0 and g = 1, and
construct triangulations achieving these lower bounds.

If N is the largest integer less than or equal to the number

112(7 + V49 - 24(2 - 2g», it is a fact that any map on X can be col-
ored with N colors, and N is the smallest number for which this is
true. See Rademacher and Toeplitz (1957) and Coxeter (1989). Sur-
prisingly, this is much easier for g> 0 than for g = o.

Exercise 8.10. Generalize Proposition 8.7 to allow arbitrary convex

polygons for faces in place of triangles.

Problem 8.11. Suppose a sphere with g handles is decomposed as in

the preceding exercise, but with each polygon having the same num-
ber p of edges, and assume that each vertex lies on the same number
q of edges. (a) Show that l/p + l/q = 1/2 + (1- g)/e. (b) When g = 0,
find all positive integers p, q, and e that satisfy this equation, and
show that all possibilities are realized by the boundaries of the five
platonic (regular) solids.

A sphere with g handles can be oriented, i.e., one can coherently

(continuously) define a notion of counterclockwise (or "which way is
up") in the neighborhood of any point. (See Appendix D for a precise
definition.) It is a fact that the only compact surfaces that can be
oriented are diffeomorphic to spheres with g handles. At least with
the assumption that the surface can be triangulated we will prove this
in Chapter 17. There are also compact surfaces that cannot be ori-
ented. One example is the projective plane ~p2, which can be real-
ized as a quotient space of the sphere S2, by identifying each point
with its antipodal point. The projection from S2 to ~p2 is two-to-one,
but a local diffeomorphism. One can also get ~p2 by taking the upper
hemisphere, and identifying opposite points on the boundary equator,
which is the same as identifying opposite points on the boundary of
a disk. Hence the projective plane can also be realized by identifying
the opposite sides of a rectangle as shown:
116 8. Vector Fields on Surfaces

Exercise 8.12. Construct a vector field on ~p2 that has one singular
point with index I. Show that the sum of the indices of any vector
field on ~p2 is 1. Triangulate ~p2 and compute its Euler character-
istic.

Another nonorientable surface is the Klein bottle, which can be

realized by identifying the opposite sides of a rectangle as shown:

Exercise 8.13. What is the Euler characteristic of the Klein bottle?

The Moebius band is a nonorientable surface whose boundary is a

circle:

tL---_+~
Or one may construct a Moebius band by taking an annulus, and iden-
tifying opposite points of one of the boundary circles:
8b. The Euler Characteristic 117

Exercise 8.14. (a) Show that these two descriptions agree. (b) What
do you get when you sew two Moebius bands together along their
boundary circles?

Exercise 8.15. Find a vector field on the Moebius band with the
boundary behavior shown and one singular point.

What is the index at the singular point?

Starting with any surface, one can cut out a disk, and paste back
in a Moebius band, by identifying points on the boundary circles. This
is called a crosscap.

Project 8.16. Investigate the surfaces that arise this way, especially
the sums of indices of vector fields and the notion of Euler charac-
teristic. What do you get when you do this to a sphere? What happens
to the Euler characteristic of a surface when this is done to it? What
is the Euler characteristic of the surface obtained by punching h dis-
joint disks from a sphere with g handles, and sewing h Moebius bands
onto their boundaries. Can you realize the Klein bottle this way? Can
two of these be homeomorphic, if you start with a different g and h?
When?

There is a beautiful proof of the Poincare-Hopf theorem for a com-

pact surface X that is oriented, using an inner product, varying con-
118 8. Vector Fields on Surfaces

tinuously, on each tangent space. (If the surface is embedded in 3-space,

one can identify each tangent space TpX as a subspace of 1R 3 , and use
the standard inner product on 1R3.) To show that two vector fields V
and Won X have the same sum of indices, triangulate X so all the
singularities of V or W are inside triangles, and so that each triangle
contains at most one singularity.

If P is a singular point in a triangle T, one can see that the number

IndexpV - IndexpW is the change in angle of the vector field V - W
around the boundary of T, divided by 2'TT; this change in angle is zero
if there is no singularity in T. Adding over all the triangles, noting
the cancellations along the edges, the theorem follows. (See Hopf
(1983).)
There is a related proof, closer to our first proof for a torus. As
we will see in Chapter 17, the surface X can be realized by identifying
sides as indicated on a plane polygon R with 4g sides:

~2 ~I

Exercise 8.17. Show that the resulting surface is a sphere with g

handles.

Given vector fields V and Won X, one can find such a realization
8b. The Euler Characteristic 119

of X so that none of the edges go thro~h sin~ularities of V or W.

Then V and W determine vector fields Vand Won the polygon, and

2: IndexpV - IndexpW = 2: Indexp V- IndexpW

PEX PER

= -1 -
(change in angle of V - W around aR) = 0,
2'lT
the last since the changes over identified edges of the boundary can-
cel.
We saw that the Euler characteristic is 2 - 2g by looking at the
vector field of a fluid flowing down the surface, or the gradient of
the height function for the surface sitting nicely in space. That picture
also shows how one can build up the topology of X by looking at the
portion of X whose height is at most h, and seeing how the topology
changes as h increases. One sees that changes occur only when the
height crosses the singularities, and that the change there is controlled
by the indices of these singularities. This is the beginning of the beau-
tiful subject of Morse theory. See Milnor (1963).
There is an important method for reducing some problems about
nonorientable surfaces to the case of orientable surfa.£es. For a non-
orientable surface_X, there is an orientable surface X and a two-to-
one mappirtg p: X ~ X, which is a local diffeomorphism. The two
points in X over a point P in X correspond to the two ways t2, orient
X near P. For example, if X is the projective plane, then X is the
sphere. We will discuss this in more detail when we come to covering
spaces, see § 16a.

Probl~ 8.18. Show that a vector field V on X d.!:,termines a vector

field V on X, and that the sum of the indices of V is twice the sum
of the indices of V. Deduce that if X is any compact surface, the sum
of indices of all vector fields with singularities on X is the same. If
X is the surface constructed by sewing h Moebius .!>ands to a sphere
with g handles with h disks removed, show that X is a sphere with
2g + h - 1 handles.
 PART V

COHOMOLOGY AND
HOMOLOGY, II

In Chapter 9 the first homology group HI U of a plane region U is

computed for a plane region "with n holes." The notion of the integral
of a closed ~'" i-form over any continuous path or any i-chain is
defined. We show that these integrals are the same over homologous
i-chains. This leads to useful methods for computing integrals and
winding numbers, and relating the two notions; these are described
in the third section.
The last section of Chapter 9, which is optional, takes a look at
how these ideas are used in complex analysis. This is written as a
(very) short course in complex analysis. It is self-contained, except
for some calculations left as exercises, but in practice it will probably
be most useful to those who have seen some of it before. For ex-
ample, if you have seen Cauchy's formula and the residue theorem
for regions such as disks and rectangles, this will show how the ideas
of topology lead to the appropriate generalizations involving winding
numbers. (This section includes all the analysis that will be needed
when we study Riemann surfaces in Part X.)
The basic theme of the Mayer-Vietoris story in Chapter 10 is to
find relations among the homology and cohomology groups of two
open sets and their union and intersection. This possibility of com-
paring different homology groups Hk , for different k as well as dif-
ferent spaces, is a salient feature of algebraic topology. For coho-
mology, the beginnings of this story were seen in Chapter 5. This
122 Part V. Cohomology and Homology, II

chapter proves analogous results for homology, and then completes

the cohomology story. This package of results, called the Mayer-
Vietoris theorem, gives a powerful tool for calculating homology and
cohomology groups.
It will be evident that the cohomology and homology groups behave
in a similar, or more precisely, dual fashion. This duality will be
made explicit in Part VIII. In fact, the proof of the full Mayer-Vie-
toris theorem for cohomology of plane domains will depend on this
duality.
CHAPTER 9

Holes and Integrals

9a. Multiply Connected Regions

Let U be an open set in the plane, and let A = /R 2 \ U. We know that
for a fixed closed chain 'Y in an open set U, the function P ~ W( 'Y, P)
is constant on connected components of /R 2 \ Supp('Y). It is therefore
constant on connected components of A, since each connected com-
ponent of A is contained in some connected component of /R 2 \ Supp('Y).
If A is a connected component of /R 2 \ U, we write W('Y,A) for the
value of W( 'Y, P) for all P in A, and call it the winding number of 'Y
around A. Note that the connected components of a closed set such
as /R 2 \ U are closed, but they need not be path-connected (see Ex-
ercise 9.5).
Our goal is to calculate the first homology group of U, at least if
this complement A is not too complicated. The idea is to find a closed
path or I-chain that "goes once" around each "piece" of A, and to
show that these give a free basis for H1U. The following lemma gives
a precise statement and rigorous proof that this is possible:

Lemma 9.1. Suppose A is a disjoint union of two closed sets Band

C, with B bounded. Then there is a closed I-chain I'in U such that
W(I', P) = I for all Pin Band W(I',P) = 0 for all Pin C.

Proof. By the compactness of B, there is a positive E so that every

point of B is at least distance E away from any point of C. Take a

123
124 9. Holes and Integrals

grid G so that its bounded (closed) rectangles cover B, but such that
none of these rectangles meets both Band C, and so that none of the
infinite rectangles meets B. This can be achieved by taking an infinite
grid so that the distances between parallel lines is less than £/Y2,
and letting G be the collection of lines that hit B.

The idea is to define 'Y to be the sum of the boundaries of the

(closed) bounded rectangles in the grid that meet B :

Clearly 'Y is a closed I-chain, since it is the sum of closed I-chains.

We claim next that 'Y is a I-chain in U. To see this, suppose an edge
CT occurs with a nonzero coefficient in 'Y, but that CT is not contained
in U . We know that CT cannot meet C, since it is in a rectangle
that meets B, and none of the rectangles meets both. So CT must meet
B. But then each of the two (bounded) rectangles that CT separates
meets B, so they occur in the displayed sum. Their contributions to
CT therefore cancel, which shows that CT cannot occur in 'Y.
Next we show that W( 'Y, P) = I for P in B . Any point P in B is in
one of the closed bounded rectangles, say R I , and we let Q be a point
in the interior of RI . Using the fact that P and Q are in the same
component of the complement of the support of 'Y, we have
9a. Multiply Connected Regions 125

W('Y, P) = W('Y, Q) = L W(ilRko Q)

k
= W(ilRIo Q) = 1.

Similarly, if PEC, then W('Y,P) = LkW(ilRkoP) =0. D

A I-chain 'Y having the property of the lemma is certainly not unique,
but any other I-chain with this property would have the same winding
numbers around all points not in U, and it follows from Theorem 6.11
that it would be homologous to 'Y.
We next describe what it means for U to have "n holes." First we
describe the "infinite" part of the complement A, denoted by A,,,, which
will be a closed set with the property that winding numbers of closed
I-chains in U around points in A,,, are always zero. In most examples
it is obvious what A"" should be, but the general definition is a little
complicated. Certainly A"" should contain any unbounded connected
component of A; to assure that we get a closed set, we define A"" by
the following:
A", = {P E A: for any 10 > 0 there is a connected subset C
of A containing a point within distance 10 of P
and a point farther than I /10 from the origin}.

Lemma 9.2. The set A", is a closed subset of A, and W(y,P) = ofor
all closed I-chains y in U and all points P in A"" .
Proof. If P is in the closure of A"" take a sequence Pn in A"" ap-
proaching P, and a connected set en as in the definition for P n for
some En, with En ~ 0 as n ~ 00. It follows immediately that P is in
A"". For any compact subset K of U, it follows from the definition
that any point of A"" is in the unbounded component of ~2 \ K. Note
that the support of any I-chain is compact. By Proposition 6.8 the
winding number of any closed I-chain around a point of A", must be
zero. D

Note that A"" can have more than one connected component, or it
can be empty. Now suppose that
~2 \ U = Al U A2 U . . . U An U A"" ,

where each Ai' 1::5 i::5 n, is a connected, closed, bounded, nonempty

set, and these n + 1 sets are disjoint. Such an open set U is sometimes
called multiply connected, or (n + I)-connected. Not every U has such
a complement, see Exercise 9.5.
126 9. Holes and Integrals

By Lemma 9.1, for each i between I and n there is a closed chain

'Yi in V so that W('Yi ,Ai) = 1, and W('Yi ,A) = 0 for j;:of i. Such I-chains
are not unique, but any other choices are homologous to these.

Proposition 9.3. Any closed I-chain y on V is homologous to the 1-

chainm,y,+ ... +mnYn, wheremi=W(y,A i).
Proof. With mi = W('Y,A i), '1 and~mi'Yi have the same winding num-
bers around all points not in V, and the result follows from Theorem
6.11. 0
The integers mj = W('Y,A j) are uniquely determined by the homol-
ogy class '1, again by Theorem 6.11. In other words:

CoroUary 9.4. The homology classes of the closed chains y" ... , Yn
form a free basis of H, V, giving an isomorphism H, V == zn.

Exercise 9.5. (a) Let V be the set of points (x, y) such that Iyl < 1 or
2n < x < 2n + 1 for some integer n. Show that Aoo is the union of an
infinite number of connected components. (b) Let A be the union of
the interval {(O,y): O:5y:5 I} and the set {(x,sin(l/x»:x>O}. Show
that A is closed and connected, but not path-connected. (c) Let A be
the union of the origin and the points (I/n,O) for n a positive integer.
Show that each point of A is a connected component of A, but the
origin has no neighborhood disjoint from the other components.

Exercise 9.6. Show that if V C V' and V is n-connected, and V'is

n' -connected, and n > n', then there is no retract from V' onto V.

Problem 9.7. (a) Show that if V is the complement of the set N of

nonnegative integers (identify n with (n,O», then H,V is isomorphic
9b. Integration over Continuous Paths and Chains 127

to the free abelian group on the points in N. In particular, HIV is not

finitely generated. (b) Compute HIV when V is the complement of
the set in Exercise 9.5(c).

Problem 9.S. (a) If V is bounded, show that A" is a connected com-

ponent of R2 \ V. (b) If the plane is identified with the complement
of the north pole in a sphere S (by stereographic projection, see §7c),
show that the union of Ace and the north pole is the connected com-
ponent of the complement of V in S that contains the north pole.

Problem 9.9. Generalize Corollary 9.4 as follows. Suppose V is any

open set in the plane, and K is a compact subset of V that has n con-
nected components K 1 , • • • ,Kn. There is a homomorphism
HI(V\K) ~ HIV(f)7Lffi . .. (f)7L == H IVffi7L n ,
defined as follows. Given a closed I-chain 'Y on V \ K, the first com-
ponent of this map takes the class of 'Y in HI (V \ K) to the class of 'Y
in HI V, and the other components take the class of 'Y to the winding
numbers of'Y around the components K), ... , Kn. Show that this
homomorphism is an isomorphism.

9b. Integration over Continuous Paths and Chains

We want to define the notion of the integral f-y w of a closed ~'" I-form
w on an open set V over an arbitrary closed path 'Y in V. We cannot
use calculus, but the idea we used for defining the winding number
works perfectly well. If 'Y is defined on an interval [a, b], subdivide
the interval into a = 10 S II S . . . S tn = b, so that each subinterval
[Ii-I, til is mapped by 'Y into some open rectangle Vi contained in V.
Such a subdivision exists by the Lebesgue lemma, since each point
in the image of'Y is contained in some such rectangle. The restriction
of w to Vi is exact (by Proposition 1.12), so we may find a ~oc func-
tion j; on Vi such that dj; = w on Vi. Let Pi = 'Y(li ), 0 SiS n. Define
the integral f"Y w by

L=w (HP I) - HPo»+ (.fz(P2) - fz(PI)

+ ... + (fn(P n) - fn(P n- ». 1

The proof that this definition is independent of choices is almost

the same as that of Proposition 3 .1. Since the intersection of two
128 9. Holes and Integrals

rectangles containing 'Y([ti- 1, ti)) is connected, and J; is unique up to

adding a constant on a connected set, the sum is independent of choices
of Ui and j;. The rest of the proof is identical, by observing that re-
fining a subdivision by adding a point doesn't change the answer.
If 'Y is a C(6'" path, it follows also from Proposition 1.16 that this
definition of J'/w agrees with that using calculus.
We extend the notion to all I-chains 'Y = nl'Yl + ... + nr'Yn with
'Yi paths, by linearity:

(w = nl ( w + ... + nr ( W.
),/ )'/1 ),/,
Note that the winding number is a special case of integral:

W('Y,P) = f.v Wp,

where, if P = (xo, Yo),
-(y - Yo) dx + (x - xo) dy
wp = -Wplt
21T . (x - XO)2 + (y - Yo)2

Exercise 9.10. If W = df is exact, and'Y is a I-chain with boundary

iJ'Y = ":imjPj, show that J'/w = ":imd(p). In particular, J'/df= 0 if'Y is
closed.

Proposition 9.11. If y and S are homologous I-chains in U, then

Proof. The proof is the same as the proof of Lemma 6.7, which refers
to Theorem 3.6. One simply changes winding numbers to integrals,
sectors U iJ to open rectangles, and angle functions 1}ij to arbitrary
functions J;J with dJ;J = w on U iJ . 0

It follows in particular that the integral of w is the same over two

paths that are homotopic closed paths, or homotopic paths with the
same endpoints. It also follows that the integrals are unchanged by
the reparametrization of paths.

Corollary 9.12. Given two closed I-chains y and S on U, the fol-

lowing are equivalent:
(1) W(y,P) = W(S,P) for all P not in U;
(2) J'/w = Jaw for all closed I-forms w on U; and
9b. Integration over Continuous Paths and Chains 129

(3) '}'- (j is a boundary I-chain in U.

Proof. The implications (3):::} (2):::} (1) have just been seen, and
(1)¢::>(3) is Theorem 6.11. D

Corollary 9.13. Let U be any open set in the plane. Then the fol-
lowing are equivalent:
(1) We'}', P) = 0 for all closed chains'}' in U and all P fI. U;
(2) I"Y W = 0 for all closed chains '}' in U and all closed 110rms w
on U;
(3) every closed I-chain is a boundary: H1U = 0; and
(4) every closed 110rm won U is exact: H1U = O.
These conditions hold whenever U is I-connected, i.e., ~2 \ U = A.,.
Proof. The equivalence of the first three conditions follows from the
preceding corollary. We have (4):::} (2) by Exercise 9.10. Con-
versely, (2) implies that I"Y W = fr, w whenever 'Y and 8 are segmented
paths with the same endpoints, and w is any closed I-form; we saw
in Chapter I that this makes w exact, which is (4). D
We know that homotopic closed paths are homologous.

Problem 9.14. (a) Show that homologous closed paths must be ho-
motopic when U is an open rectangle or any convex open set. (b) Do
the same when U is the complement of a point, or an annulus
U = {(x,y):rI 2«x-xo)2+(Y-Yo)2<r/}.
(c) Challenge. What if U is the complement of two points?

Problem 9.15. Suppose 'Y = 'ini'Yi is a closed I-chain in an open set

U such that each '}'i is a <{SO<> path. If 'Y is homologous to zero, show
that one can write 'Y = 'im/af), where each fj is a <{S'" map from the
unit square to U.

Problem 9.16. If X is any closed, connected subset of the plane, and

U is a bounded connected component of ~2 \ X, show that every closed
I-form on U is exact.

Problem 9.17. (a) If w, = p(x,y, t)dx + q(x,y, t)dy is a continuously

varying family of closed <{SO<> I-forms on U, i.e., the functions p and
q are continuous on U x [a, b], show that the function t~ I-yw, is a
continuous function of t. (b) Use this to give another proof of Prop-
osition 3.16.
130 9. Holes and Integrals

9c. Periods of Integrals

With U an (n + I)-connected region as above, and 00 any closed l-fonn
on U, define the period (or module of periodicity) of 00 around Ai to
be the integral of 00 along 'Yi' with 'Yi as defined just before Proposition
9.3. By Proposition 9.11, this is independent of choice of 'Yi' Denote
this period by Pi(W) or P(w,AJ

Pi(W) = P(w,Ai) = {w.

These numbers detennine integrals of 00 along any closed path, pro-

vided one knows the winding number of the path around each Ai:

Proposition 9.18. For any closed I-chain y and closed l{orm 00,

L 00 = W('Y,A.)P.(w) + W('Y,A2)P2(W) + ... +W('Y,An)Piw).

Proof. Since 'Y and Li W('Y,AJ . 'Yi are homologous, this follows from
Proposition 9.11. 0

Applying Corollary 9.12, we have:

Corollary 9.19. A closed l{orm 00 on U is exact if and only if all of

its periods P;(w) are zero.

For integrals along paths that are not closed, if 'Y and 8 are two
paths with the same endpoints, applying the proposition to 'Y - 8, we
have

L 00- Jaw = m.p.(w) + m2P2(w) + ... + mnPn(w) ,

with m., ... , mn integers. This means that the integral is detennined
up to adding integral combinations of the periods. For example, when
U = ~2 \ {P} and 00 = WP.i) (see Problem 2.10), there is only one pe-
riod, which is 2'7T, and we recover the fact that the integral is deter-
mined up to integral multiples of 2'7T.

Exercise 9.20. Compute the integral I'Yw, where 00 is the I-form

_ ~ -ydx+ (x- n)dy
W-L.J 2 2 '
n=. (x-n) +y
and 'Y(t) = (tcos(t),tsin(t)), 0:5t:56'7T.
9d. Complex Integration 131

Exercise 9.21. Show that, given U as above, for any real numbers
Ph ... , Pn there is a closed I-form 00 with these periods.
This means that the linear mapping from the vector space of closed
I-forms on U to IW, W~(PI(W), . • • , Pn(w), is surjective, and by
Corollary 9. 19 the kernel is the space of exact I-forms. This sets up
an isomorphism (see §Cl)
{closed I-forms on U}/{exact I-forms on U} == IRn,
i.e., the De Rham group HI U is an n-dimensional vector space. If Pi
is any point in A;. 1:5 i:5 n, the classes [wPJ form a basis for HI(U).
Problem 9.22. (a) Suppose n = 2, and the periods of 00 are PI and
lJ2' Let P and Q be two fixed points in U. Show that if the periods
are not zero, and the ratio ptlp2 is rational, there is a number r so
that if 'Y and 8 are any two paths from P to Q, then f"Y 00 - f I) 00 is an
integer times r. (b) Show that, if PI/P2 is not rational, and U is con-
nected, there is no such r.

9d. Complex Integration

The plane 1R2 can be identified with the complex numbers C, the pair
(x, y) being identified with z = x + iy. Functions on open sets in the
plane will be written as functions of z.
A complex 110rm 00 on an open set U in the plane is given by a
pair of ordinary (real) I-forms WI and 002, written in the form
00 = WI + iW2'
Define dw = dWI + i dW2, and for a function f = u + iv, with u and v
real-valued functions on U, set df= du + idv. The form 00 is closed
if dw = 0, and exact if 00 = df. For example, we have the I-form dz
defined by dz = dx + i dy. A complex I-form can by multiplied by a
complex-valued function: if 00 is as above, and f= u + iv, then f· 00
is the complex I-form
f' 00 = (u + iv)' (WI + i(2) = (UWI - V(2) + i(UW2 + VWI)'
Exercise 9.23. (a) Iff= u + iv, with u and v~., functions, show that
the I-formf{z)dz is closed if and only if u and v satisfy the Cauchy-
Riemann equations:
au av au av
and
ax ay ay ax
132 9. Holes and Integrals

(b) Show that if the conditions in (a) are satisfied, then the complex
derivative f'(a) = limz--->a(J(z) - f(a»/(z - a) exists at each point a
in U.

Let us calIf analytic if its real and imaginary parts satisfy the Cau-
chy-Riemann equations. For example, iff is locally expandable in a
power series, then f is analytic, so f(z) dz is closed. Indeed, if
f(z) = L;=oaiz - zor in a disk around zo, thenf= dg on that disk, with

~ an
g(z) = +1
LJ - - (z - zor .
n=O n + 1

For an example that is closed but not exact, consider dz/z on the
complement of the origin:
dz dx + idy (x - iy)(dx + i dy)
=
z x+iy x 2 +l
= xdx+ydy +i -ydx+xdy
= d(1og(r» + iw~
~+l x 2 +l
Similarly, dz/(z - a) = d(log(lz - aln + iWa.ih with wa.~ as defined in
Chapter 2.
If -y is a path or chain in U, the integral of 00 = WI + iW2 along -y is
defined by

L L + iL
00 = WI 002'

For example, if 'Y(t) = a + r' eit , O:s t:s 2'11', is a circle around the point
a, then

f ~ = i f Wa.~ =
,/z-a '/
2'11'i.

In general, if 'Y is any chain not containing a in its support, we see

similarly that

~
f,/z-a = if'/ Wa.~ = 2'11'i' W(-y, a).

The main fact about complex integrals is

Theorem 9.24 (Cauchy Integral Theorem). If l' is a closed chain in

U whose winding number around any point not in U is zero, and f
is an analytic function in U, then for any a in U that is not in the
9d. Complex Integration 133

support 0/ 'Y,

W(-y,a)'f(a) I
= -.
2m
f
-f(z)
dz.
-y z- a
Proof. Look at F(z) = (f(z) - f(a»/(z - a), which is analytic on U \ {a}.
Using the formula proved just before the statement of the theorem,
the displayed formula is equivalent to the formula f-yF(z)dz = O. Let
I>lt) = a + re2-rrit, 0:5 t:5 I, with r small enough so the disk of radius
r around a is contained in U. Let n = W(-y, a). Since
W(-y, P) = n' W(I>" P) for all P rt. U \ {a}, we know by Corollary 9.12
that LF(z)dz=n'fa,F(z)dz, so it suffices to prove that
lim,......ofa,F(z)dz = O. But by Exercise 9.23(b), F(z) has a limit as z
approaches a, and the fact that fa,F(z)dz approaches zero as the ra-
dius goes to zero follows easily (see Exercise B.8). 0
The simplest form of Cauchy's formula is when -y is a circle about
a. Since the right side of Cauchy's formula is expandable in a power
series in a neighborhood of a, this implies in particular the fact that
any analytic function is locally expandable in a power series.

Exercise 9.25. (a) Let U be an open set containing two concentric

circles and the region between them. For a in the region between the
circles, show that

f(a) = -l-f
27ri
f(z) dz -
-y,Z - a 27ri
-I-f f(z) dz
-y,Z - a '
where "Ir and "I. are counterclockwise paths around the larger and
smaller circle. (b) Deduce Riemann's theorem on removable singu-
larities: if / is analytic and bounded in a punctured neighborhood of
a point b, then/extends to an analytic function in a full neighborhood
of b.
A related application is to the general residue theorem. If / is an-
alytic in a punctured neighborhood of a point a (i.e., in some U \ {a},
for U a neighborhood of a), the residue of f at a, denoted Resa(f),
is defined by

Resa(f) = ~ rf(z) dz ,
2m Ja
where I> is a small counterclockwise circle around a. By the Cauchy
integral theorem, this is independent of the circle chosen.

Exercise 9.26. If/is given by a converging series L;=-m cn(z - ar in

134 9. Holes and Integrals

a punctured neighborhood of a (i. e. ,f has at most a pole at a), show

that Resa(f) = C-l.

Theorem 9.27 (Residue Theorem). Iff is analytic in U\{al, ... ,ar },

and y is a closed I-chain in U \ {aI, ... , a r} such that W( y, P) = 0
for all P not in U, then
1
L W('Y,ai)·Resa,(f).
r
-.!.f(z)dz =
2'7Tl "1 i=l

Proof. Take disjoint small circles 8i around ai. Then, with ni = W('Y, ai),
'Y and Ln,-8i have the same winding number around all points not in
U \ {aJ, ... , arlo It follows from Corollary 9.12 that the integral of
the closed I-form .f(z) dz around 'Y and around Lni8i gives the same
answer. 0

Problem 9.28. Extend the residue theorem to allow f to be analytic

outside any discrete set S in U (i.e., each point in U has a neigh-
borhood containing at most one point of S).

Exercise 9.29. Suppose U is an (n + I)-connected open set as in the

first section, with ~2 \ U = Al U ... U An U A"" and f is analytic on
U; let lJi be the period of f(z) dz around Ai. Show that. for a closed
chain 'Y on U,

Exercise 9.30. Deduce the Cauchy Integral Theorem directly from

the Residue Theorem.

Suppose now fis meromorphic in U, i.e., near every point a in U

one can write .f(z) = (z - at . h(z), with h analytic at a, h(a) oj:. 0, and
m an integer. The order off at a, denoted orda(f), is defined to be
this integer m.

Exercise 9.31. Show that, withf and h as above, if 8 is the boundary

of a disk about a such that h is nowhere zero in the disk, then
orda(f) = W(fo 8, 0) .

Theorem 9.32 (Argument Principle). Suppose f is meromorphic in

9d. Complex Integration 135

U, and 'Y is a closed I-chain in U not passing through any zero or

poLe off, such that W( 'Y, P) = 0 for aLL P not in U. Then

W(f°-Y,O) = 2: W(-y, a)' ordif) ,

where the sum is over the (finitely many) zeros or poLes a for which
W( 'Y, a) ,., O.
Proof. If {a" ... , ar} are the zeros and poles around which -y has
a nonzero winding number (see Problem 9.28), take a small circle 8;
around a; as in the proof of the Residue Theorem. Applying Propo-
sition 6.15 to the mappingf: U \ {a" ... , ar}~ U' = C \ {O}, we get
the formula W(f°-Y, 0) = ~~~, W(-y, a;)' W(f°8;, 0). D

The following problems give the analytic interpretation, and a typ-

ical application, of the Argument Principle:

Problem 9.33. (a) If f is meromorphic at a, show that

orda(f) = ResaC)·

(b) Under the conditions of the Argument Principle, use the Res-
idue Theorem to show that

-
21Ti
1 ff'(Z)
'I
--dz
f(z)
2: W(-y, a)' orda(f).
a

(c) Show directly that

I ff'(Z)
-. -dz = W(f0-Y,O)
2m 'I f(z)
by interpreting the integrand as d(log(f(z» = d(loglf(z)l) + i . d(arg(ftz»,
with loglftz)1 a well-defined function, and arg(f(z» the multivalued
angle function.

Problem 9.34. (a) Let U,f, and -y be as in the Argument Principle.

If g is analytic function on U, show that

~f g(z/'(z) dz =
2m 'I f(z)
2: W(-y, a)· g(a)· orda(f).
a

(b) Suppose the restriction of f to a closed disk D around Zo in U

is one-to-one, and let -y be the boundary of D. Show that the function
136 9. Holes and Integrals

that takes w to (l/21Ti) Lzj'(z)/(f(z) - w)dz defines an inverse func-

tion to f in a neighborhood of f(zo).

The following is Rouche' s theorem:

Problem 9.35. Suppose the chain 'Y is homologous to zero in U, and

suppose f and g are analytic functions in U such that
if(z) - g(z)i < if(z)i + ig(z)i
for all z in the support of 'Y. Show that

2: W('Y, a)' orda(f) = 2: W('Y, a) . orda(g)·

a a

In particular, if the winding number of 'Y is 1 for points a in some

region V, and the winding number is zero elsewhere, then f and g
have the same number of zeros in V, counting multiplicities.
CHAPTER 10

Mayer- Vietoris

lOa. The Boundary Map

For open sets U and V in the plane (or any topological space) we will
define a homomorphism
iJ: HI(UUV) ~ Ho(UnV),
called the boundary map.5 To do this we need a lemma.

Lemma 10.1. If'Y is a I-cycle on V U V, there are I-chains 'YI on V

and Y2 on V such that YI + Y2 is homologous to Y on U U V.
Proof. We know from Lemma 6.4(b) that if a path is subdivided, it
is homologous to the sum of the paths into which it is divided. By
the Lebesgue lemma, each path occurring in "I can be subdivided so
that the image of each piece is in U or in V. So "I is homologous to
a sum 2. njTi> where each Tj is a path in U or in V (or both). Then "II
can be taken to be the sum of those njTj for which Tj is a path in V,
and "12 can be the sum of the others. 0

Construction of the boundary map iJ: HI(U U V)~ Ho(U n V). Re-
call that HI(U U V) = ZI(U U V)/BI(U U V) is the group of I-cycles,

5 This symbol a is certainly overworked in this subject. So far we have used it for
the boundaries of I-chains and for maps of rectangles. Rather than introducing dif-
ferent notations for the different uses, we try for clarity by saying in each case to
what sort of object the "iJ" is being applied.

137
138 10. Mayer-Vietoris

modulo the subgroup of I-boundaries, on U U V. And

Ho(U n V) = Zo(U n V)/Bo(U U V) is the group of O-cycles modulo
O-boundaries on un V. We will write ['y] for the homology class in
HI (U U V) defined by a I-cycle 'V on U U V, and we write in m
Ho(U n V) for the class defined by a zero cycle' on un V.
Given a homology class IX in HI(U U V), by the lemma, we may
choose a I-cycle that represents IX and has the form 'VI + 'V2' where
'VI is a I-chain on U and 'V2 is a I-chain on V; in symbols, ['VI + 'V2] = IX.
Since the boundary of this I-cycle is zero, we have a('VI) = -a(-y2).
This O-cycle a('VI) = -a('V2) is a O-cycle on U and on V, so it is a 0-
cycle on unv. We define a:HI(UUV)-7Ho(UnV) by sending IX
to the class of this O-cycle:
a(IX) = a(['V I + 'V2]) = [a('VI)] = -[a('V2)].

Note that, although the zero cycle a('VI) = -a('V2) is a boundary on U

and on V, it need not be a boundary on un V:

The hardest part of our task is to show that this class [a'Vl] is well
defined.

Lemma 10.2. The class of a')'l in Ho(U n V) is independent of the

choice of ')'1 and ')'2·
Proof. Before giving the proof, we need a more canonical way to
subdivide our I-chains, by cutting each one exactly in half. For any
path 'V: [0,1]-7 U, define the I-chain S('V) by the formula S('V) = (J + T,
where (J and T are the restrictions of 'V to the two halves of the in-
terval, but rescaled as in Lemma 6.4(b). Extend this operator S lin-
early to all I-chains 'V = 'i, ni'Vi by setting S('V) = 'i, niS('Vi). It follows
from this definition that the boundary of the I-chain S('V) is the same
as the boundary of 'V.
lOa. The Boundary Map 139

If r: [0, 1] X [0, 1] ~ U is a map of the unit square, we can simi-

larly subdivide r into four pieces r O), r(2), r(3), and r(4) as indicated:

n l ) nZ)

Each restriction is rescaled to be a map from [0,1] x [0,1] to U. In

formulas,
r(I)(t, s) = r(lM, 1/2JS ), r(Z)(t, s) = r(llz + 1/2/, 11zs) ,
r(3)(t, s) = r(llzt, liz + 1/2JS ), r(4)(t, s) = [(1/2 + 1/2t, liz + 1/2JS).

It is clear from the picture, and easy to verify from the formulas, that
this subdivision is compatible with taking the boundary, i.e.,
scar) = ar(1) + ar(2) + ar(3) + a[<4) ,
the inside boundaries canceling as usual.
Now we prove the lemma. Suppose the class a is also represented
by the cycle 'YI' + 'Y2' for I-chains 'YI' on U and 'Y2' on V. Since both
sums represent 'Y, we know that ('YI + 'Y2) - ('YI' + 'Yz') is a boundary
on UU V, so we can write
('YI + 'Y2) - ('YI' + 'Y2') = ~ njac

for some maps r j from rectangles to U U V. We must show that

[0'YI] = [0'YI'] in Ho(U n V), i.e., that O"YI - l1"YI' is a boundary of some
I-chain on un v. We apply the subdivision operator S to each side
of the displayed equation. On the left, each of the four I-chains is
replaced by another with the same support and the same boundary.
On the right, each ac
is replaced by a sum of the boundaries of the
four subdivisions of C. SO we have an equation of the same form,
but with all the C's cut into quarters. The operator S can be applied
again, which subdivides each of these quarters into quarters, and so
on, dividing the smaller squares into quarters. When applied p times,
we have an equation
SP('YI)+SP('Y2)-SP('Y)')-SI'('Yz') = ~njSP(or).

By the Lebesgue lemma, the restrictions of r j to small enough por-

tions of the rectangles must be mapped into U or into V. It follows
that for some large p the right side of this equation can be written in
the form T) + Tz, where T) is a I-boundary on U and T2 is a I-boundary
140 10. Mayer-Vietoris

on V. This gives an equation of I-chains:

SP('Yl) - SP('Yl') - 71 = SP('Y2') - SP('Y2) + 72.
The left side is a I-chain on U, the right side a I-chain on V, so the
I-chain they define is a I-chain on un V. And the boundary of this
chain is
a(SP('Yl) - SP('Yl') - 71) = a'Yl - a'Yl' - a71 = a'Yl - a'Yl' ,
which shows that a'Yl - a'Yl' is a O-boundary on un V, and completes
~~~ 0
Lemma 10.3. The boundary operator a: H 1(UU V)~Ho(Un V) is a
homomorphism of abelian groups.
Proof. This follows readily from the definition. For if a is represented
by 'Yl + 'Y2, and a' is represented by 'Yl' + 'Y2', with 'Yl and 'Yl' I-chains
on U and 'Y2 and 'Y2' I-chains on V, then a ± a' is represented by
('Yl ± 'Yl') + ('Y2 ± 'Y/), so
a(a ± a') = [a('Y1 ± 'Yl')] = [a('Yl)] ± [a('Yl')] = aa ± aa'. 0

lOb. Mayer-Vietoris for Homology

If U 1 C u2 , we saw in Chapter 6 that there are homomorphisms from
H OU 1 to H OU2 and from H 1U 1 to H 1U2. Given two open sets U and V
we therefore have a diagram
HP
/~
- a Ho(UnV)

The Mayer-Vietoris story gives the relations among all these groups
and homomorphisms. This can be described in a series of assertions,
moving from right to left in the diagram.

MV(i). Any element in Ho(U U V) is the sum of the images of an ele-

ment in HoU and an element in HoV.
lOb. Mayer-Vietoris for Homology 141

Proof. Any O-chain that represents an element in Ho(U U V) can be

written as a sum of a O-chain in U and a O-chain in V. 0

MV(ii). An element in HoU and an element in HoV have the same

image in Ho(U U V) if and only if they come from some element in
Ho(Un V).
Proof. Let b and c be O-chains representing elements in HoU and H 0V.
If they have the same image in Ho(U U V), there is a I-chain 'Y on
U U V with a'Y = b - c. By subdividing the paths in 'Y to be suffi-
ciently small, so that the image of each subdivided path is contained
in U or in V, we may find I-chains 'YI on U and 'Y2 on V so that
a('Y1 + 'Y2) = a'Y. Then b - a'Yl = c + a'Y2, and the left side of this is a
O-chain on U, and the right side is a O-chain on V, so
a = b - a'Yl = c + a'Y2 is a O-chain on un V. Since a is homologous
to b on U and to c on V, the class in Ho(U n V) represented by a maps
to the classes represented by b in HoU and c in HoV. The converse
is immediate from the definitions. 0

MV(iii). An element in Ho(U n V) maps to zero in HoU and in HoV

if and only if it is the image by a of some element in HI(U U V).
Proof. We saw during the definition of the boundary a that a bound-
ary always has this form. Conversely, if ~ = a'Yl and ~ = a'Y2 for I-chains
'YI on U and 'Y2 on V, then m
= a(a), where a is the class represented
by the I-cycle 'YI - 'Y2· 0

MV(iv). An element in H 1CU U V) maps to zero in HoCU n V) if and

only if it is the sum of an element coming from HI U and an element
coming from HIV.
Proof. If a = hi + 'Y2] for some I-cycle 'YI on U and some I-cycle
'Y2 on V, we may take 'YI and 'Y2 for the I-chains in the definition of
the boundary, so a(a) = [a'Yd = O. Conversely, if a is in the kernel
of a, and we write a = hi + 'Y2], with 'Yi as in the definition of the
boundary, then a'Yl = -a'Y2 is a O-boundary on un V, so a'Yl = aT for
some I-chain T on un V. Therefore,
a = + ('Y2 + T)],
[('YI - T)
and 'YI - T is a I-cycle on U and 'Y2 + T is a I-cycle on V. 0

MV(v). An element in HIU and an element in HIV have the same

image in HI(U U V) if and only if they come from some element in
HI(Un V).
142 10. Mayer-Vietoris

Proof. Let 13 and "I be I-cycles on U and V representing the two

elements. If they have the same image in HI(U U V), there is an equa-
tion
13-"1 = LniaC
for some maps C from [0, 1] x [0, 1] to U U V. We apply the sub-
dividing operator S, which was introduced in the proof of Lemma
10.2, to this equation. As in that lemma, if S is applied sufficiently
many times, the images of the subdivided rectangles will be contained
in U or in V, and we will have an equation
SPI3 - SP'Y = L niSP(aC) = 81 + 82 ,
where 81 is a I-boundary on U and 82 is a I-boundary on V. Lemma
6.4(b) proves that S takes a I-cycle to a I-cycle that is homologous
to it. So snl3 - 8 1 = SP'Y + 82 is a I-cycle on un V that is homologous
to 13 on U and to "I on V, as required. Again, the converse is
obvious. 0
In the case of open sets in the plane, there is one more assertion
that completes the Mayer-Vietoris story.

MV(vi). If U and V are open subsets of the plane, an element in

HI (U n V) is zero if and only if its images in HI U and in HI V are
zero.
Proof. If "I is a I-cycle representing the element in HI (U n V), if its
images are zero in HIU and HIV, then the winding number W('Y,P)
is zero for all P not in U and all P not in V. This means that W( "I, P)
vanishes for all P not in un V, and by Theorem 6.11 this implies
that "I is homologous to zero on un V. 0

°
Exercise 10.4. If HI U = and HI V = 0, show that the kernel of a is
zero. If in addition un V is connected, show that HI(U U V) = 0.

Exercise 10.5. If U and V are connected, show that the image of a

consists of all classes in Ho( un V) of degree zero.

Exercise 10.6. Show that if U and V are connected, and HI(U U V) = 0,

then un V is also connected.

Exercise 10.7. If un V is connected, and HI(U n V) = 0, show that

HI (U U V) is isomorphic to the direct sum of HI U and HI V.

Exercise 10.S. If the homomorphism from HI(Un V) to HIV is in-

lOb. Mayer-Vietoris for Homology 143

jective, show that the homomorphism from HIU to HI(U U V) is also

injective.

Exercise 10.9. Suppose A and B are disjoint closed subsets in a space

X, and HIX = O. Show that if X \ A and X \ B are path-connected, then
X \ (A U B) is also path-connected.

Let us use Mayer-Vietoris to work out an example. We start with

a familiar situation, with U and V as simple as possible, which arises
when we take the union of two open sets each diffeomorphic to disks.

Suppose U and V are connected, with HI U = 0 = HI V. Suppose un V

has m;::: 1 connected components. We claim that HI(U U V) is a free
abelian group with m - 1 generators. By MV(iv), the boundary map
iJ is injective. and by MV(iii), the image of iJ consists of elements of
Ho(U n V) that restrict to zero in HoU = lL and HoV = lL. Now Ho(U n V)
is the free abelian group on the m connected components of U n v.
so the image of iJ is the kernel of the homomorphism from this free
abelian group to 7L that maps each component to 1. It is easy to see
that this kernel is the free abelian group on m - 1 generators, which
completes the proof. For example, if the components of un V are
numbered WI> ...• Wm • and ei is the class that puts coefficient 1 in
front of Wi' -1 in front of Wi+1> and 0 in front of the other com-
ponents, then the classes el> • • . , em-I form a basis for this kernel.
In particular, this recovers the fact that if U is the complement of
a point, then HIU = 7L.

Exercise 10.10. Use Mayer-Vietoris to recover the fact that, if U is

the complement of n points in the plane, then HIU is a free abelian
group with n generators. Show that small circles around each of the
points gives a basis for HIU.
144 10. Mayer-Vietoris

lOco Variations and Applications

The assertions of the Mayer-Vietoris story can be put in fancy lan-
guage (but with no change in content) as follows. Define, for k = 0
and k = 1, a homomorphism
+ : HkU(J)HN ~ Hk(UU V)
that takes a pair (13, -y) to the sum of the image of 13 and the image
of -y in HiUU V). In this language, MV(i) says that this homomor-
phism from HoUffiHoV~Ho(UUV) is surjective. Similarly, define
a homomorphism
- : Hiun V) ~ HkU(J)HkV
that sends a class IX to the pair (13, --y), where 13 is the image of IX
in HkU and -y is the image of IX in HN. Assertion MV(ii) says that
the image of this homomorphism from Ho( U n V) to HoU (J) HoV is the
kernel of the homomorphism from HoU(J)HoV to Ho(UU V).
Given a sequence . . . ~ An- 1~ An ~ An+ 1 ~ • • • of abelian groups
and homomorphisms between them, the sequence is called exact at
An if the kernel of the map from An to An+ 1 is equal to the image of
the map from A n- 1 to An. The sequence is called exact if it is exact
at every group in the sequence. For example, the exactness of a se-
quence A~B~O (at B) means precisely that the map from A to B
is surjective, and the exactness of O~ A ~ B (at A) means that the
map from A to B is injective. Assertions MV(i)-MV(vi) can be sum-
marized in the

Theorem 10.11 (Mayer-Vietoris Theorem for Homology). For any

open sets U and V of a topological space, the sequence
H1(Un V)~HIUffiHIV~HI(UU V)
~Ho(Un V)~HoU(J)HoV~Ho(UU V)~O
is exact. If U and V are open subsets of the plane, the sequence
O~Hl(Un V)~HIUffiHIV~HI(UU V)
~Ho(Un V)~HoUffiHoV~Ho(UU V)~O
is exact.

Exercise 10.12. Verify that the exactness of the Mayer-Vietoris se-

quence at each term is indeed equivalent to the assertions MV(i)-
MV(vi).
IOc. Variations and Applications 145

Exercise 10.13. Suppose A and B are disjoint closed subsets in the

plane. Use Mayer-Vietoris to construct an isomorphism
H I(lR 2 \ (A U B» == HI(lI~? \ A) Ee H I(lR 2 \ B).
Show that the number of connected components of 1R2 \ (A U B) is one
less than the sum of the numbers of connected components of 1R2 \ A
and of 1R2 \ B. Generalize to any finite number of disjoint closed sub-
sets.

Exercise 10.14. Let U be an open subset of the plane, and K a com-

pact subset of U. Show that H I(U\K)==H IUEeH I(1R 2 \K). If K has
n connected components, recover the isomorphism HI (U \ K) == HI U Ee 7Ln
of Problem 9.9.

Exercise 10.15. If X is a finite graph (see Problem 5.21) with v ver-

tices and e edges, and X has k connected components, show that HIX
is a free abelian group with e - v + k generators.

Exercise 10.16. If X is a finite graph in the plane with k connected

components, show that H I (1R 2 \ X) is a free abelian group with k gen-
erators.

Problem 10.17. For any nonempty space X, we considered in §9a

the degree homomorphism deg: Hr)( ~ 7L that takes the class of a zero
cycle 'in;?; to t!le sum 'in; of the coefficients. The reduced Oth ho-
mology group Hr)( of a space X is definectto be the kernel of this
J!!.ap. So X is path-connected if and only if Hr)( is zero. (a) Show that
Hr)( is a free abelian group with rank one less that the number of
path-connected components of X. (b) Show that, if un V is not empty,
there is an exact sequence
HI(Un V)~HIU(f)HIV~HI(UU V)
a- --
~Ho(Un V)~HoUEeHoV~Ho(UU V)~O.
-+-
This often simplifies computations, since the groups involved are a
little smaller.

Problem 10.1S. Let U be a connected open set in S2, and let X be a

subset of S2 homeomorphic to a closed interval such that all of X
except for the endpoints is contained in U. Show that U \ X is dis-
connected if and only if the endpoints of X lie in the same connected
component of S2 \ U.

Problem 10.19. If X is an open set in the plane, and Hr)( and HIX
146 10. Mayer-Vietoris

have finite ranks, the Euler characteristic of X, denoted X(X), is de-

fined by
x(X) = rank(HoX) - rank(H1X).
Suppose U and V are open sets in the plane, and three of the four
open sets U, V, un V, and U U V have homology of finite ranks.
Show that the fourth does also, and that
x(U U V) + X(U n V) = X(U) + X(V)·
Using the Mayer-Vietoris theorem, homology groups can be used
instead of cohomology to prove the Jordan Curve Theorem:

Problem 10.20. With U and Vas in the proof of Theorem 5.10, show
that the image of the boundary map
a:H 1(UUV)=H 1(1R 2 \{P,Q}) ~ H o(UnV)=Ho(lR 2 \X)
is free with one generator, and deduce that HO(1R2 \X) is a free abelian
group with two generators, so the complement of X has two connected
components. Similarly with U and Vas in the proof of Theorem 5.11,
show that the image of a is zero, and deduce that HO(1R2 \ Y) is a free
abelian group with one generator, so 1R2 \ Y is connected.

The following are some complements to the Jordan Curve Theorem

that can be proved by combining Mayer-Vietoris with Corollary 9.4
or Problem 9.9:

Problem 10.21. A compact set K of an open set U is said to separate

two points if they belong to different connected components of U \ K.
Prove the following version of "Alexander's Lemma": If Y and Z are
compact subsets of U, with Y n Z connected, and P and Q are two
points in U \ (Y U Z) that are not separated by Y or by Z, then they
are not separated by Y U Z.

Problem 10.22. Let X be a subset of the plane homeomorphic to a

circle, P a point on X, and D a disk centered at P. Let A be an open
arc of X containing P and contained in D, and let B be the comple-
mentary closed arc. Let E be a disk centered at P that doesn't meet
B. Show that if two points in E are not separated by X, then they can
be connected by a path in D \ X.

Problem 10.23. Let X be a subset of the plane homeomorphic to a

circle, U a connected component of the 1R2 \ X. Show that, for any
IOd. Mayer-Vietoris for Cohomology 147

E > 0, there is a 8> 0 such that any two points of V within distance
8 of each other can be connected by a path in the intersection of V
with a disk of radius E. Note that this is false when X is homeo-
morphic to an interval.

Problem 10.24. Let X be a subset of the plane homeomorphic to a

circle, P a point on X, and Q a point not on X. Show that there is a
path 'Y: [0, l]~ ~2 with 'Y(O) = Q, 'YO) = P, and 'Y(t) ft X for 0 ~ t < 1.

Problem 10.25. Suppose Y and Z are compact, connected subsets of

an open set V, and Y n Z has n connected components. (a) Show that
V \ (Y U Z) has at least n connected components. (b) If V \ Y and V \ Z
are connected, show that V \ (Y U Z) has exactly n connected com-
ponents.

IOd. Mayer-Vietoris for Cohomology

The Mayer-Vietoris story for cohomology is similar to-in fact, dual
to-that for homology. If V' is an open subset of a plane open set
V, the restriction of I-forms from V to V' takes closed forms on V
to closed forms on V', and exact forms on V to exact forms on V',
so we have a linear map of vector spaces
HI(U)~HI(V').

There is also a linear map lfJ(U)~ lfJ(U') that takes a locally constant
function on V to its restriction to U'. If U and V are open sets in the
plane, we therefore have a diagram
HOV HIV

HI(VUV) HI(VnV)

H1V
where the maps in each diagram are determined by restrictions, and
8 is the coboundary map defined in Chapter 5. The Mayer-Vietoris
story for De Rham cohomology, for open sets in the plane, is the
combination of six assertions:
148 10. Mayer-Vietoris

MV(i). An element in If(UU V) maps to zero in lfu and in lfv if

and only if it is zero.

MV(ii). An element in lfu and an element in lfv have the same

u
image in If( n V) if and only if they come from some element in
If(UU V).

MV(iii). An element in If(U n V) maps to zero in Hl(U U V) if and

only if it is the difference of an element coming from lfu and an
element coming from lfv.

MV(iv). An element in HI(UU V) maps to zero in HIU and HIV if

and only if it is the image by {) of an element in If(U n V).

MV(v). An element in HIU and an element in HIV have the same

image in HI(U n V) if and only if they come from some element in
HI(UU V).

MV(vi). Any element in HI(U n V) is the difference of an element

coming from HIU and an element coming from HIV.

Of these, MV(i) and MV(ii) are straightforward exercises, and

MV(iii) and MV(iv) are Propositions 5.7 and 5.9. We prove the non-
trivial assertion in MV(v). Given closed I-forms a on U and ~ on V,
such that there is a function f on U n V with a - ~ = df on U n V, we
must construct a closed I-form w on U U V that differs from a by an
exact form on U, and from ~ by an exact form on V. That is, we
want functions fl on U and f2 on V such that
a-dfl = ~-df2

on un V. In other words, we want dfl - df2 = df. So it will be enough

to find fl and f2 so that fl - f2 = f on U n V. The existence of such f
follows from Lemma 5.5.
The last case, MV(vi), however is not so obvious. We postpone
the proof until Chapter 15, where we make rigorous the "duality"
between homology and cohomology. With this duality, in fact, all six
assertions for cohomology will be seen to follow from the six asser-
tions for homology. D
The cohomology version of Mayer-Vietoris also has its concise
expression as an exact sequence. Define linear maps
+ : Jt(UUV) ~ HkUffiHkV
lOd. Mayer-Vietoris for Cohomology 149

by sending a class to the pair consisting of the restrictions of the class

to each open set. Define linear maps
- : HkU~HkV ~ H\Un V)
by taking a pair (WI, W2) to the difference wdunv - w21unv of its re-
strictions. As in the case for homology, the assertions MV(i)-MV(vi)
are equivalent to the exactness of a sequence:

Theorem 10.26 (Mayer-Vietoris Theorem for Cohomology). For any

open sets U and V in the plane, the sequence
o~I-f(UU V)~I-fU(fJI-fv~I-f(Un V)
.3,.HI(UU V)~HIU(fJHlv~HI(Un V)~O

is exact.

It is probably worth remarking that the choice of signs "+" and

"-" in both the homology and cohomology versions of Mayer-Vie-
toris is perfectly arbitrary; those we have chosen will be convenient
later.

The following two exercises use Mayer-Vietoris for cohomology,

including MV(vi), to strengthen some facts we saw in Chapter 9:

Exercise 10.27. Suppose A and B are disjoint closed subsets in the

plane. Use Mayer-Vietoris to construct an isomorphism
HI(~2\(AUB» == HI(~2\A)EBHI(~2\B).

Generalize to the complement of any finite number of disjoint closed

subsets.

Exercise 10.28. If K is a compact subset of an open set U, and K

has n connected components, construct an isomorphism
HI(U\K) == HIU(fJIR\
where the maps to the second factor are given by periods around the
components of K.

Exercise 10.29. If U is not empty, define ifu to be the quotient

space I-fu fIR of the locally constant functions on U by the subspace
of constanl functions. This is called the reduced Oth cohomology group.
So dim(I-fU) = dim(I-fU) - 1, and U is connected exactly when
ifu is zero. Show that, if un V is not empty, there is an exact
150 10. Mayer-Vietoris

sequence
0->.> jjJ(UU V)~ jjJUffijjJV~jjJ(U n V)
~Hl(UU V)~H1U(f)H1V~Hl(Un V)->.>O.
The result in the following problem was proved by Brouwer:

Problem 10.30. Let K be any compact connected subset of 1R2, U a

connected component of 1R2 \ K. Show that the boundary au of U is
connected. Is the same true if K is only closed and connected in 1R2?
 PART VI

COVERING SPACES AND

FUNDAMENTAL GROUPS, I

In Chapter 11 we introduce the notion of covering maps, which are

generalizations of the polar coordinate mapping, and study their basic
properties. Facts about lifting paths and homotopies will generalize
what we saw for this special case, which amounted to the basic prop-
erties of winding numbers. Many coverings, including the polar co-
ordinate mapping, are examples of G-coverings, arising from an ac-
tion of a group G on a space, and we emphasize those that arise this
way.
We have studied closed paths, and seen that homotopic paths have
similar properties. In Chapter 12 we formally introduce the funda-
mental group, which is the set of homotopy equivalence classes of
closed paths starting and ending at a fIxed point, the equivalence given
by homotopy. In the last section we see how it is related to the fIrst
homology group.
CHAPTER 11

Covering Spaces

lla. Definitions
If X and Y are topological spaces, a covering map is a continuous
mapping p: Y ~ X with the property that each point of X has an open
neighborhood N such that p-l(N) is a disjoint union of open sets, each
of which is mapped homeomorphic ally by ponto N. (If N is con-
nected, these must be the components of p -1(N).) One says that p is
evenly covered over such N. Such a covering map is called a covering
o/X.
An isomorphism between coverings p: Y ~ X and p': Y' ~ X is a
homeomorphism 'P: Y ~ Y' such that p' 'P = p. A covering is called
0

trivial if, in the definition, one may take N to be all of X. Equiva-

lently, a covering is trivial if it is isomorphic to the projection of a
product X x T onto X, where T is any set with the discrete topology
(all points are closed). So any covering is locally trivial.

153
154 11. Covering Spaces

The first nontrivial covering is the mapping p: IR--+Sl given by

p({}) = (cos(2'IT{}), sin(2'IT{}»:

( . ) ( .
or in a vertical picture:
) ( . . ) ( )
p

We saw in Chapter 2 that the polar coordinate mapping

p : {(r, {}) E 1R2: r> O} --+ 1R2 \ {(O, O)}
given by p(r, {}) = (r cos(tt), r sin({})), is a covering. Another is the
mapping Pn: Sl--+ Sl, for any integer n ~ 1, given by
(cos(2'IT{}), sin(2'IT{}» ~ (cos(2'ITn{}), sin(2'ITn{})),
or, using complex numbers, by z--+ zn; this can be visualized by join-
ing the loose ends in the last picture.

Exercise 11.1. Show that the following are covering maps, where
C* = C \ {O} is the group of nonzero complex numbers: (i) the nth
power mapping C* --+ C*, z--+ z"; and (ii) the exponential mapping
exp: C--+ C*.

Exercise 11.2. Let p: Y --+ X be a covering. If X' is any subspace of

X, verify that the restriction Y' = p-l(X')--+X' is a covering.

Exercise 11.3. If p: Y --+ X is a covering of an open subset X in the

plane, show that Y can be given the structure of a differentiable sur-
face in such a way that p is a local diffeomorphism. The components
V of p-l(N), for N a connected open set in X over which the covering
lla. Definitions 155

is trivial, together with the homeomorphisms of N with V given by

the inverse of p, give charts covering Y. Similarly, any covering of
any manifold has a natural manifold structure.

Exercise 11.4. Show that if X is connected, all fibers of a covering

Y ~ X have the same cardinality.

When each p-l(X) has cardinality a finite number n, the covering

is called an n-sheeted covering. It should be emphasized, however,
that one cannot distinguish n different "sheets," unless the covering
is trivial.
If p: Y~X is a covering, andf:Z~X is a continuous mapping,
a continuous mapping 1: Z ~ Y such that poI = f is called a lifting
off,

In the next section we will discuss the question of whether such lift-
ings exist. The following lemma discusses uniqueness: if Z is con-
nected, a lifting is determined by where it maps anyone point.

Lemma 11.5. Let p: Y~X be a covering, and let Z be a connected

topological space. Suppose 11 and 12 are continuous mappings from
Z to Y such that po 11 =P 012' If 11 (z) =12(z) for one point z in Z,
then 11 = 12'
Proof. It suffices to show that the set in Z where the mappings agree
is open, and its complement where they disagree is also open. If w
is in the set where they agree, take a neighborhood N of po j;(w) that
is evenly covered by p. Let p-l(N) be a disjoint union of open sets
N ex , with each N ex , mapped homeomorphically to N by p. By conti-
nuity, 11 and 12 must map a neighborhood V of w into the same Nex ,
and since po 11 = P 012, 11 and 12 must agree on V. Similarly, if w is
in the set where the mappings do not agree, 11 and 12 must map a
neighborhood V of w into two different (and hence disjoint) Nex's, so
they disagree on V. 0
156 11. Covering Spaces

11 b. Lifting Paths and Homotopies

The work we did in studying the winding number will be abstracted
and formalized in the following propositions. In §2a we saw that de-
fining an angle function amounted to lifting a closed path -y in 1R2 \ {O}
to a path 'Y in the right half plane so that the composite of 'Y with the
polar coordinate covering p is the given path -y; the difference in sec-
ond coordinates of 'Y from start to end is then the change in angle
along -y . We will now see that this is a general property of covering
spaces.

Proposition 11.6 (Path Lifting). Let p: Y~X be a covering, and let

y: [a,b]~X be a continuous path in X. Let y be a point of Y with
p(y) = y(a). Then there is a unique continuous path y: [a, b] ~ Y such
that yea) = y and po YCt) = yet) for all t in the interval [a, b].
Proof. The uniqueness comes from Lemma 11.5. When the covering
is trivial, the proposition is obvious: there is a unique component of
Y that contains y and maps homeomorphically to the component of X
that contains -y([a, b]), and one must simply lift the path to that com-
ponent using the inverse homeomorphism. For the general case, we
apply the Lebesgue lemma (Appendix A) to the open sets -y-I(N),
where N varies over open sets in X that are evenly covered by p. This
gives a subdivision a = to:5 ... :5 tn = b such that each -y([I;_1> til) is
contained in some open set that is evenly covered by p. By the trivial
case, there is a lifting of the restriction of -y to [to, t l ], giving a path
in Y from y to some point YI' Similarly, there is a lifting of the re-
striction of -y to [tl> t 2] that starts at YI; and one proceeds in n steps
until one has lifted the whole path. 0
It follows in particular that the final point 'Y(b) of the lifting is de-
termined by -y and by the initial point Y . We denote this point by Y * -y ,
so
y*-y = 'Y(b).
If this is applied to the polar coordinate covering, and y = (ro, 1}0) and
y * -y = (rl> 1} I) are the initial and final points of the lifting of a path
-y, then the difference 1}1 -1}0 is the total change of angle along -y.

Exercise 11.7. Verify this last statement.

One consequence of the path-lifting proposition is the fact that any

covering of an interval must be trivial.
lIb. Lifting Paths and Homotopies 157

The fact that homotopic paths in ~2 \ {O} with the same endpoints
have the same total change of angle around zero is equivalent to the
fact that their liftings are homotopic. This too is a general fact about
coverings.

Proposition 11.8 (Homotopy Lifting). Let p: Y---+X be a covering,

and let H be a homotopy of paths in X, i.e., H: [a,b] X [0, 1]---+X is
a continuous mapping. Let 'Yo(t) = H(t, 0), a:S t :s b, be the initiaipath.
Suppose 'Yo is a lifting of 'Yo.Jhen there is a unique lifting H of H
wh~e initialpa!..h is 'Yo, i.e., H: [a,b] X [0, 1]---+Yis continuous, with
poH =H and H(t, 0) = 'Yo(t), a:S t:s b.
Proof. The proof is very much the same as for Proposition 11.6. First
apply the Lebesgue lemma to know that there are subdivisions
a = to < tl < ... < tn = b and 0 = So < Sl < ... < Sm = 1 so that if RiJ
is the rectangle [ti- lo til X [Sj-lo sJ, then eac!!.H(Ri) is contained in an
evenly covered open set. Then the lifting H is constructed over each
piece RiJ' say first working across the bottom row, lifting the restric-
tion of H to RI,lo R 2 ,lo ' .. , Rn,lo then doing the same for the next
row R I ,2, R 2,2, . • , , Rn,2, and so on until the entire rectangle has been
covered. 0

If H is a homotopy of paths from x to x' in X, i.e" H(a, s) = x and

H(b, s) = x' for allO:S s:S 1, and if.:yo is a path from y to y', then the
lifted homotopy H is a homotopy of paths from y to y', (The fact that
H is constant on the sides of the rectangle is guaranteed by the unique-
ness of the lifting of the restriction of H to these sides.) In particular,
if H is a homotopy from 'Yo to 'Y lo then
y * 'Yo = Y * 'YI .
For the polar coordinate mapping, this is the preceding assertion about
homotopic paths having the same change of angle.

Exercise 11.9. Use the homotopy lifting proposition to show that

homotopic closed paths in [R2 \ {O} have the same winding number.
(The paths in the homotopy are assumed to be closed, but the end-
points can vary during the homotopy.)

Exercise 11.10. Use the homotopy lifting proposition to prove that

any covering of a rectangle (closed or open) must be trivial. Deduce
the same for any space homeomorphic to a rectangle, such as a disk.

Exercise 11.11. Suppose p: Y ---+ X is a covering map, with X a locally

158 11. Covering Spaces

connected space (any neighborhood of a point contains a connected

neighborhood of the point). (a) Show that X is a union of connected
open sets N such that each connected component of p-I(N) is mapped
homeomorphic ally by ponto N. (b) Let Y' be a connected component
of Y. Show that image X' = p(Y') is a connected component of X, and
the restriction Y' ~ X' is a covering map.

Exercise 11.12. Suppose p: Y ~ X and p': Y' ~ X are covering maps,

and 'P: Y ~ Y' is a continuous map such that p' 0'P = p. Suppose X,
Y, and Y' are connected, and X is locally connected. Show that 'P is
a covering map.

Exercise 11.13. If p: S' ~ S is an n-sheeted covering, and S is a com-

pact surface, show that S' is also a compact surface. Show, using
vector fields and/or triangulations, that the Euler characteristic of S'
is n times the Euler characteristic of S. If Sand S' are spheres with
g and g' handles, show that g' = ng - n + I.

Exercise 11.14. Let X be the space that consist of two circles A and
B joined at a point P. Let Y be a space that is two circles and four
half-circles, joined as shown, and let p: Y~ X be mapping that takes
each piece of Y to the correspondingly labeled piece of X as indicated:

Show that p is a three-sheeted covering. Let -y be the path in X, start-

ing at P, that goes first around the circle A counterclockwise, then
around B counterclockwise, then around A clockwise, then around B
clockwise. Find the three liftings of -y. Deduce that -y is not homotopic
to the constant path at P. Use this to solve Problem 9.14.

lIe. G-Coverings
Many important covering spaces arise from the action of a group G
on a space Y, with X the space of orbits. Recall that an action of a
Hc. G-Coverings 159

group G on Y (on the left) is a mapping G x Y~Y, (g,y)~g.y,

satisfying:
(1) g. (h· y) = (g. h) . y for all g and h in G and y in Y;
(2) e· y = y for all y in Y, where e is the identity in G; and
(3) the mapping y ~ g . y is a homeomorphism of Y for all g in G.
In other words, G defines a group of homeomorphisms of Y. Two
points y and y' are in the same orbit if there is an element g in G
that maps one to the other: y' = g. y. Since G is a group, this is an
equivalence relation. Let X = Y/G be the set of orbits, or equivalence
classes. There is a projection p: Y ~ X that maps a point to the orbit
containing it. The space X is equipped with the quotient topology
(i.e., a set U in X is defined to be open when p-l(U) is open in Y).

Exercise 11.15. (a) The group 71. acts on IR by translation: n· r = r + n.

Show that the quotient IR~ 1R/71. can be identified with the covering
map from IR to Sl described in the first section.
(b) Show that the polar coordinate covering map is the quotient of
the right half plane by a 7l.-action.
(c) The group G =1J-Ln of nth roots of unity (which is a cyclic group
of order n) in C acts by multiplication on Sl, regarded as the complex
numbers of absolute value 1. Compare the quotient map with the cov-
ering Pn: Sl ~ Sl described in § 11 a.

For another example, the group with two elements acts on the
n-sphere with the nontrivial element taking a point to its antipodal
point; i.e., G = {± 1} == 7l./2Z acts on Sn, with ±1 . P = ±P. The quo-
tient space is the real projective space IRlPn, and the quotient mapping
Sn~ IRlPn is a two-sheeted covering.
We say that G acts evenl/ if any point in Y has a neighborhood V
such that g . V and h . V are disjoint for any distinct elements g and h
in G.

Exercise 11.16. The group j.Ln of nth roots of unity acts on C by

6 The standard tenninology for the notion we are calling "even" is the mouthful
"properly discontinuous." The word "discontinuous" does not mean that anything
is not continuous or otherwise badly behaved; it means that the orbits are discrete
subsets of Y. The word "properly" refers to the fact that every compact set only
meets finitely many of its translates. If this were not bad enough, the use of "prop-
erly discontinuously" in the literature is inconsistent, in that often it is only required
that each point have a neighborhood V such that at most finitely many translates
g. V of V can intersect. In this case the word "freely" is added, so our "evenly" is
then "freely and properly discontinuously."
160 11. Covering Spaces

multiplication. Show that the action is not even, but that the action
on the open subset C* = C \ {O} is even.

Lemma 11.17. If a group G acts evenly on Y, then the projection

p: Y~Y/G is a covering map.
Proof. The map p is continuous, and open since for V open in Y, the
set p-l(p(V)) is the union of open sets g' V, g E G. If V is taken as
in the definition of an even action, then this union is a disjoint union.
It is enough to prove that for such V, the mapping from each g . V to
p(V) induced by p is a bijection, for then it follows that p is evenly
covered over p(V). This is a straightforward verification: since
p(g' y) = pry) for y in V, it is surjective; if p(g' YI) = p(g' Y2), there
is some h in G with h· g . YI = g . Y2, and the fact that the action is
even implies that h is the identity. 0
A covering p: Y ~ X is called a G-covering if it arises in this way
from an even action of G on Y. An isomorphism of G-coverings is
an isomorphism of coverings that commutes with the action of G; i.e.,
an isomorphism of the G-covering p: Y~ X with the G-covering
p': Y' ~ X is a homeomorphism <p: Y ~ Y' such that p' 0 <p = p and
<p(g' y) = g' <p(y) for g in G and Y in Y. The trivial G-covering of X
is the product X X G ~ X, where G acts by (left) multiplication on
the second factor.

Lemma 11.1S. Any G-covering is locally trivial as a G-covering,

i.e., if p: Y~X is a G-covering, then any point in X has a neigh-
borhood N such that the G-covering p-l(N)~N is isomorphic to the
trivial G-covering N x G~ N.
Proof. In fact, if N = p(V) as in the proof of Lemma 11.17, such a
local trivialization is given by
p-l(N) :3 g' v ~ p(v) X g E N x G. 0

Exercise 11.19. Verify that the actions described in Exercise 11.15

are all even. Show that the action of 7l.. n on IR n by translation is even.
Identify the quotient IRn /7l.. n with the n-dimensional torus
(sl)n = Sl X . . . X Sl.

Exercise 11.20. Show that any two-sheeted covering has a unique

structure of G-covering, where G = 7l../27l.. is the group of order two.

Exercise 11.21. Show that the three-sheeted covering of Exercise 11.14

is not a G-covering.
llc. G-Coverings 161

Exercise 11.22. A section of a covering p: Y ~ X is a continuous

mapping s: X ~ Y such that po s is the identity mapping of X. Show
that if a G-covering has a section, then the covering is a trivial
G-covering.

Exercise 11.23. If p: Y ~ X = Y/G is a G-covering that is trivial as

a covering, show that it is isomorphic to the trivial G-covering.

Exercise 11.24. Letp: Y~X= Y/G be a G-covering, and let 'PI and
'P2 be isomorphisms of G-coverings from Y to Y. If X is connected,
and 'PI and 'P2 agree at one point of Y, show that 'PI = 'P2.

Exercise 11.25. Let G be the subgroup of the group of homeomor-

phisms of the plane to itself generated by the translation (x, y) ~ (x + 1, y)
and by the mapping (x, y) ~ ( - x, y + 1). Show that this action of G
on ~2 is even, and identify the quotient ~2 /G with the Klein bottle.

Exercise 11.26. Let G be the subgroup of the group of homeomor-

phisms of the plane to itself generated by the translation
(x, y) ~ (x + 1, -y). Show that this action is even, and identify the
quotient with a Moebius band.

Exercise 11.27. If G acts evenly on a space Y, and H is a subgroup

of G, show that H also acts evenly. Show that the natural map from
Y/H to Y/G is a covering mapping. If n is the index of H in G, this
is an n-sheeted covering. Carry this out when G is the group of trans-
formations of the plane from Exercise 11.25, and H is the subgroup
generated by the two homeomorphisms (x, y) ~ (x + 1, y) and
(x, y) ~ (x, y + 2). Identify ~2 /H with a torus, and describe the re-
sulting two-sheeted covering of the Klein bottle.

Exercise 11.28. If a finite group G acts on a Hausdorff space Y, and

there are no fixed points (i.e., no y is fixed by any g in G except the
identity element), show that the action is even.

Exercise 11.29. If G acts evenly as a group of diffeomorphisms of

a differentiable manifold Y, show how to give Y/G the structure of
a differentiable manifold in such a way that the projection Y~ Y/G
is a local diffeomorphism.

Exercise 11.30. Let G =Jj.J.n be the group of nth roots of unity, as in

Exercise 11.15. The odd-dimensional sphere s2m-1 can be realized as
S2m-l = {(z!> ... , zm) E em: Izd 2 + ... + IZml2 = I}.
162 11. Covering Spaces

The group G acts on S'lm-I by ,. (Zh ... , zm) = ('Zh ... , 'zm). Show
that this action is even. When n is prime, the quotient space S'lm-I/Jl.n
is a manifold called a Lens space.

Exercise 11.31. Suppose CSJ is a topological group, i.e., a topological

space which is also a group such that the multiplication and inverse
maps are continuous. Suppose G is a discrete subgroup of CSJ, i.e.,
there is a neighborhood N of the identity e in CSJ such that N n G = {e}.
Show that left multiplication by G on CSJ is an even action. This makes
CSJ a G-covering of the space G\CSJ of right cosets of G in CSJ. Many
of our coverings have this form, e.g., IW~(Slr is the quotient by
the subgroup zn; exp: C~ C* is the quotient of C by 2-rriZ.

Exercise 11.32. Let y= IW\{O}, let r be any real number but 0, 1,

or -1. Let G = Z act on Y by m' v = ,mv for m E Z, v E Y. Show that
the action is even, and show that Y /G is homeomorphic to the product
SI X sn-I.

Problem 11.33 (For those who know about quaternions). The three-
sphere S3 can be identified with the set of unit quaternions:
S3 = {r + xi + yj + zk: ? + r + l + l = I},
which makes it a topological group. Identify 1R3 with the set of pure
quaternions {xi + yj + zk}. For q E S3, the mapping v ~ q . v . q - J de-
fines an orthogonal transformation of 1R3 with determinant 1, i.e., an
element of SO(3). Show that the resulting map S3~SO(3) is a sur-
jective homomorphism of groups, with kernel {± I}. Deduce that this
is two-sheeted covering. (This is the spin group Spin(3), which, with
its generalizations Spin(n)~ SO(n) for all n;:::: 3, appear frequently in
modem mathematics and physics.)

Problem 11.34. Suppose G is a subgroup of the group of distance-

preserving transformations of the plane 1R2, that satisfies a uniformity
condition: there is ad> 0 such that for all points P in the plane and
all g in G other than the identity, Ilg' P - pil ;:::: d.
(a) Show that the action of G on the plane 1R2 is even. Let
p: 1R2 ~ 1R2/G = X be the resulting G-covering. Defme a distance function
onXby
dist(Qh Q2) = Min{IIP 1 - P 211: p(P 1) = QI andp(P2) = Q2}'
(b) Show that this distance function defines a metric on X.
(c) Prove the analogous result for uniform actions on 1R3 (or on any
lid. Covering Transfonnations 163

locally compact metric space). This puts a geometric structure on the

quotient space X which is locally Euclidean. The reader is invited to
look at the resulting "locally Euclidean geometry" on X for some of
the groups we have seen. For a general discussion, with a list of
possibilities for ~2 and ~3, see Nikulin and Shafarevich (1987).

lId. Covering Transfonnations

For any covering p: Y --+ X there is a group Aut(YIX) of covering
transformations, or deck transformations:
Aut(YIX) = {<p: Y --+ Y : <p is a homeomorphism and p 0 <p = p} .
This is a group by composition of mappings, and it acts on Y in the
sense of the preceding section; it is called the automorphism group
of the covering.

Exercise 11.35. If Y --+ X is a trivial n-sheeted covering, and X is

connected, show that Aut(YIX) is isomorphic to the symmetric group
on n letters.

Exercise 11.36. If p: Y --+ X is the covering of Exercise 11.14, show

that Aut(YIX) contains only the identity element.

If the covering is a G-covering, there is a canonical homomorphism

from G to Aut(YIX) that takes g to the homeomorphism y 1--+ g . y. By
the definition of even action, this homomorphism is injective. It need
not be surjective; for example, it is not surjective if the covering is
a trivial G-covering, see Exercise 11.35.

Proposition 11.37. If p: Y --+ X is a G-covering, and Y is connected,

then the canonical homomorphism G--+ Aut(YIX) is an isomorphism.
Proof. Fix any point y in Y. Given <p in Aut(Y IX), since y and <p(y)
lie in the same orbit, there is a g in G such that g' y = <p(y). The
covering transformation determined by g and <p both take y to <p(y),
so by Lemma 11.5 they coincide. 0

For a G-covering, G acts transitively on each fiber p-l(X) of p; that

is, for y and y' in a fiber, there is a gin G with g' y = y'. In addition,
this action is faithful: the element g taking y to y' is unique. The
following proposition gives a converse to this:
164 11. Covering Spaces

Proposition 11.38. Let p: Y-+X be a covering, with Y connected

and X locally connected. Then Aut(YIX) acts evenly on Y. If Aut(YIX)
acts transitively on a fiber of p, then the covering is a G-covering,
with G = Aut(YIX).
Proof. We show first that the action is even. For yin Y, let N be a
neighborhood of p(y) evenly covered by p, and let V be the neigh-
borhood of y mapping homeomorphically to N by p. If 'P and 'P' are
distinct covering transformations, then 'P(V) and 'P'(V) must be dis-
joint, for if not, then 'P -1 0 'P' has a fixed point in V, and Lemma 11.5
implies that 'P -I 0 'P I is the identity.
Let G = Aut(YIX). The covering p factors into the composite of
the projection Y -+ Y IG followed by a mapping ft: Y /G-+ X. It fol-
lows easily from the definitions that this p is a covering mapping;
indeed, any open connected set of X that is evenly covered by p is
evenly covered by p. If x is a point in X such that G acts transitively
on p -l(X), then the fiber of p over x has only one point. By Exercise
11.4, all fibers of p have one point, so p is a homeomorphism. It fol-
lows that p is a G-covering. 0

Problem 11.39. Let p: Y -+ X be a covering, with Y connected and X

path-connected. Let xEX, and set S = p-I(X). Any automorphism of
the covering restricts to a permutation of S, and the automorphism is
determined by this restriction, so Aut(YIX) C Aut(S). Show that
Aut(YIX) == {'P E Aut(S): 'P(z * 0") = 'P(z) * 0" for all zE S
and all closed paths 0" at x}.
CHAPTER 12

The Fundamental Group

12a. Definitions and Basic Properties

The aim of this section is to make the homotopy equivalence classes
of paths that start and end at a fixed point in a space into a group.
We will see later how this group tells us about the covering spaces
of X, as well as determining the first homology and cohomology groups
(when X is an open set in the plane). In this chapter it will be con-
venient again to have all paths defined on the same interval. So a
path in a topological space X will be a continuous map "(: [0, 1]~X.
We say that "'I is a path from the point x = "'1(0) to the point x' = "'1(1).
In this chapter a homotopy of paths will always fix the endpoints x
and x', i.e., H(O, s) = x and H(1, s) = x' for all 0 ~ s ~ 1.
If U is a path from a point x to a point x', and T is a path from x'
to a point x', there is a product path, denoted U· T, which is a path
from x to x'. It is the path that first traverses u, then T, but it must
do so at double speed to complete the trip in the same unit time:
U(2t) , 0 ~ t ~ 112,
U·T ( t) = {
T(2t - 1), 112 ~ t~ I.

If U is a path from x to x', there is an inverse path U -[ from x' to x:

O~t~l.

For any point x, let ex be the constant path at x:

eit) = x, O~t~l.

165
166 12. The Fundamental Group

Exercise 12.1. Show that, for paths from a point x to a point X', the
relation of being homotopic is an equivalence relation.

Next we verify that, up to homotopy, these operations satisfy (where

defined!) the group axioms. For example, if 0" is a path from x to x',
there is a homotopy from the path Ex' 0" to 0". This is done by adjusting
the time of waiting at x, as indicated in the diagram:

cr

We write down the corresponding homotopy, which is constant on

the vertical and slanted lines drawn; on horizontal lines, it is the same
as indicated on the top and bottom, but adjusted proportionally:
X,
{ 1
H(t,s) = - -f2(I-S)),
0"(t- --- l/2CI-s):5t:5l.
1 - 112(1 - s)

Exercise 12.2. Construct a homotopy from 0" • Ex' to 0".

In these cases, and those that follow, it is not hard to write down
explicit formulas for the homotopy, using a little plane geometry to
map rectangles and triangles onto each other, to interpolate between
the values we want on boundary lines. It may be worth pointing out
that one could also appeal to general results about maps between con-
vex sets that guarantee the existence of such maps; the formulas them-
selves are not important.
If 0" is a path from x to x', '1" a path from x' to x", and /Jo is a path
from x' to x", there is a homotopy from (0"' '1") • /Jo to 0"' ('1"' /Jo) con-
structed similarly from the diagram
12a. Definitions and Basic Properties 167

Exercise 12.3. (a) Write down the homotopy indicated by this dia-
gram. (b) Define the path (1 • or . J.L by the formula
(1(3t) , 0:5 t:5 1/3,
(1' or' J.L(t) = { or(3t -1), 113:5 t:5 %,
J.L(3t - 2), 213:5 t:51.
Show that (1' or' J.L is homotopic to the paths «(1' or)' J.L and (1' (or' J.L).

There is also a homotopy from the path (1 . (1 -1 to the constant path

Ex' This family of paths does part of the trip specified by (1, rests at
the point it has reached, then returns. This is indicated on the fol-
lowing diagram, but note that this time the function is not constant
on the diagonal lines:

The homotopy is
(1(2t) , 0:5 t:5 1/2(1 - s),
H(t, s) = { (1(1 - s), 112(1 - s):5 t:5 1/2(1 + s),
(1(2 - 2t), 1/2(1 + s) :5 t:5 1 .
168 12. The Fundamental Group

We need one more homotopy: if two paths a and a' are homotopic
by a homotopy HI, and T and T' are homotopic by a homotopy H 2 ,
then a' T and a'· T' are homotopic, by the homotopy
HI(2t,S), 0:5t:51/2,
H(t, s) = {
H 2(2t-l,s), 112:5t:51.

Now, for a point x in X, a loop at x is a path that starts and ends

at x. Define the fundamental group of X with base point x, denoted
'TTI(X,X), to be the set of equivalence classes of loops at x, where the
equivalence is by homotopy (see Exercise 12.1). We write ['y] for the
class of the loop 'Y. The identity is the class e = [Ex] of the constant
path Ex' Define a product by [a]' [T] = [a' T]. The last displayed ho-
motopy implies that this product is well defined on the equivalence
classes. The others imply the equations
e' [a] = [a], = e,
[a]' e [a]' [a-I] = e,
= [a]' ([T] . [11-]),
([a] . [T])' [11-]
Since (a-I)-I = a, the equation [a-I]. [a] = e follows. So
in 'TTI(X,X).
this product makes 'TTI(X,X) into a group. From Exercise 12.3(b) we
deduce that the product ([a] . [T]) . [11-] is also equal to [a' T' 11-].
It should be emphasized that endpoints must be fixed during the
homotopies discussed here. Otherwise any loop 'Y would be homo-
topic to a constant loop, by the homotopy H(t, s) = "((1 - s)t).
If f X ~ Y is a continuous function, and fix) = y, then f determines
a homomorphism of groups
h: 'TT\(X,x) ~ 'TT\(Y,y) ,

that takes [a] to [faa].

Exercise 12.4. Verify that this is well defined and a group homo-
morphism.

The fundamental group is a "covariant functor, on the category of

pointed spaces." This means that, if we also have a map g: Y ~ Z
with g(y)=z, so that we also have g*: 'TT\(Y,Y)~'TT\(Z,z), then
(gofh = g*of*, i.e., the diagram
f*
1t\(X, x) ~ 1t\(y, y)

(goj)~ ~*
1t\(2, z)
12a. Definitions and Basic Properties 169

commutes. In addition, if/is the identity, then/*: '7TI(X,X)~'7TI(X,X)

is the identity mapping.
Some of the simple applications of fundamental groups use nothing
more than this functoriality. For example, if we know that '7T 1(X, x)
is a "complicated" group, and '7TI(Y,Y) is a "simple" group, so that
there are no homomorphisms of groups such that the composite
'7TI(X,X)~'7TI(Y,X)~'7TI(X,X) is the identity, then there can be no
continuous maps f X ~ Y and g: Y ~ X such that go/ is the identity
map on X. For example, if '7TI(X,X) is not the trivial group, and '7TI(Y,Y)
is trivial, then X cannot be embedded in Y as a retract.

Exercise 12.5. Show that '7TI(D 2 , x) = {e} for any x in the disk, and
that the winding number determines an isomorphism of '7TI(SI, (1,0))
with 7L. Deduce that Sl is not a retract of D2.

Exercise 12.6. Show that if X C [W is a subspace that is starshaped

about the point x, then '7TI(X,X) = {e}.

Although the definition of fundamental group depends on the choice

of base point, one gets the same group (up to isomorphism) if one
chooses another base point, at least if X is path-connected, or if the
two points can be connected by a path. Suppose T is a path from x
to x'. Define a map
T#: '7TI(X,X) ~ '7TI(X,X ')
by ['Y] ~ [T- 1. ('Y' T)] = [(T- 1. 'Y)' T]. As before, this is well defined,
and is a homomorphism of groups. It is an isomorphism, since (T- 1)#
gives the inverse homomorphism from '7TI(X,X' ) to '7TI(X,X).

Exercise 12.7. (a) Verify these assertions. (b) If T' is another path
from x to x', show that
(T')#[-y] = [pr l • (T#[-y])' [p],
where p is the loop T-I . T'. In particular, if T and T' are homotopic
paths from x to x', they determine the same isomorphism on funda-
mental groups. In general, the displayed equation means that the iso-
morphism from '7TI(X,X) to '7TI(X,X' ) depends on the choice of path
from x to x' only up to inner automorphism.

For this reason, if X is a path-connected space, one often speaks

of "the fundamental group" of X, without referring to a base point.
This will cause no confusion as long as we are only interested in the
group up to isomorphism.
170 12. The Fundamental Group

The fundamental group of a ball D n or of ~n is trivial, say by Ex-

ercise 12.6. We have seen that the circle has an infinite cyclic fun-
damental group. Let's look next at the n-sphere Sn, for n 2:: 2. This is
almost as simple as it seems. The complement of any point in Sn is
homeomorphic to ~n, so any loop that misses any point is homotopic
to the constant loop. Although it is possible for a continuous loop to
map onto the n-sphere, for any such n, this is not a serious obstruc-
tion. We will see a more general reason later, but for now it can be
seen directly:

Exercise 12.S. Show that any path in Sn, n 2:: 2, is homotopic to a

path whose image contains no neighborhood of any of its points. De-
duce that the fundamental group of Sn is trivial if n 2:: 2.

Exercise 12.9. Show that the fundamental group of a Cartesian prod-

uct is the product of fundamental groups of the spaces:
'TTI(X x Y,x x y) == 'TTI(X,X) x 'TTI(Y,y).
Problem 12.10: If @ is a topological group, show that 'TTI(@, e) is
commutative.

12b. Homotopy
Two continuous maps 10: X - Y and II: X - Y are homotopic if there
is a continuous mapping H: X X [0, 1] - Y such that H(z, 0) =Jo(z) and
H(z, 1) =!I(z) for all points z in X; H is a homotopy Irom 10 to II' We
want to say that homotopic maps determined the same homomor-
phism on fundamental groups, but to make this precise we must keep
track of base points. Let x be a base point in X, and let Yo =Jo(x) and
YI =!I(x). The mapping T(t) = H(x, t) is a path from Yo to YI'

Proposition 12.11. The diagram

~1tI(Y,yo)

1t.(X, x) 1 't#

~1t.(Y,y.)
commutes, i.e., T#o(Joh = (fl)*'
Proof. Let 'Y be a loop at x. We need to construct a homotopy from
12b. Homotopy 171

the path or -I . «fo ° 'Y) . or) to the path II ° 'Y. Consider the homotopy h
from [0, 1] X [0, 1] to Y given by h(t, s) = H('Y(t) , s):

h y
t

This provides a kind of homotopy between the path or-I. «/0 °'Y) . or)
around the two sides and the bottom of the square and the path II ° 'Y
that goes across the top. D

1\

1\

~ ~

Exercise 12.12. Turn this into a homotopy from or-I«foo'Y). or) to

II ° 'Y by constructing a continuous map from the square to itself that
is the identity on the top side, is constant on the two sides, and maps
the bottom side to the three sides, as indicated.

As a special case, we have the

Corollary 12.13. If /o(x) = II (x) = y, and H is a homotopy from /0 to

II such that H(x, s) = y lor all s, then
(/0)* = (fl)* : 'lT1(X,X) ~ 'lT1(Y,y).

Two spaces X and Y are said to have the same homotopy type if
there are continuous maps f X ~ Y and g: Y ~ X such that goI is hom-
172 12. The Fundamental Group

otopic to the identity map of X and fo g is homotopic to the identity

map of Y. The map f is called a homotopy equivalence if there is such
a g.

Exercise 12.14. (a) Show that having the same homotopy type is an
equivalence relation. (b) Show that a homotopy equivalence f X ~ Y
determines an isomorphismf*: 1T1(X,X)~1TI(Y,fix» of fundamental
groups. In particular, if i: X ~ Y embeds X as a deformation retract
of Y, then i*: 1T1(X,X)~1TI(Y,i(x» is an isomorphism.

Exercise 12.15. Classify the following spaces according to homotopy

type: (i) a point; (ii) a closed disk; (iii) a circle; (iv) ~2; (v) ~n; (vi) the
complement of a point in the plane; (vii) two circles joined at a point;
(viii) an annulus; and (ix) the complement of two points in the plane.

Exercise 12.16. Show that the mapping from S2 to itself that takes
(x,y,z) to (-x, -y,z) is homotopic to the identity mapping, and the
mapping that takes (x,y,z) to (x,y, -z) is homotopic to the antipodal
mapping.

Exercise 12.17. Iff sn ~ Sn is a continuous mapping such that fiP) #- P

for all P, show that f is homotopic to the antipodal mapping. If
f(P) #- - P for all P, show that f is homotopic to the identity map-
ping.

Problem 12.18. If n is odd, show that the identity mapping on sn is

homotopic to the antipodal mapping.

It is a general fact that for even n, the antipodal map on Sn is not

homotopic to the identity map. We will prove this in Chapter 23.

Problem 12.19. An orthogonal (n + 1) x (n + 1) matrix determines a

mapping from Sn to itself. Show that two such mappings are ho-
motopic if and only if they have the same determinant. If n is even,
show that such a mapping is homotopic to the identity if the deter-
minant is 1, and to the antipodal mapping if the determinant is - 1.

Problem 12.20. Compute the fundamental group of the space GLi([R)

of (2 x 2)-matrices with positive determinant, which gets its topology
as an open subspace of [R4.
12c. Fundamental Group and Homology 173

12c. Fundamental Group and Homology

For any topological space X, with base point x in X, there is a ho-
momorphism from the fundamental group 1TI(X,X) to the homology
group H1X, that takes the class ['Y] of a loop 'Y at x to the homology
class of 'Y, regarded as a closed path or I-chain. It takes the constant
path Ex at x to O. The fact that it is well defined amounts to the fact
that homotopic paths define the same homology class, as we saw in
Lemma 6.4. The same lemma showed that the homology class of a
product a . 'I" of loops is the sum of the homology classes of the loops,
which shows that the mapping is a homomorphism of groups.

Exercise 12.21. If f X ~ Y is a continuous mapping, and fix) = y,

show that the following diagram commutes:
f*
1t1(X, x) - - 1t1(Y, y)

t t
- f*
Since H1X is an abelian group, this homomorphism must vanish on
all commutators a· b . a -I b -I in 1T 1(X, x), so it must vanish on the
commutator subgroup [1TI(X,X),1TI(X,X)] that consists of all finite
products of commutators. This is a normal subgroup of 1TI(X,X), and
the quotient group is sometimes called the abelianized fundamental
group of X, and denoted 1Tl(X,X)abel' So we have a homomorphism
1TI(X,X)abel = 1TI(X,X)/[1Tl(X,X), 1TI(X,X)] ~ HIX,
Since the fundamental group 1TI(X,X) depends only on the path-
connected component of X that contains x, we cannot expect the fun-
damental group to determine the homology group for disconnected
spaces. But except for this, the fundamental group determines the
homology group:

Proposition 12.22. If X is a path-connected space, then the canonical

homomorphism from 7TI(X,X)abel to H1X is an isomorphism.
Proof. We must define a homomorphism from the abelian group ZIX
of I-cycles to 1TI(X,X)abel, and show that the I-boundaries B1X map
to zero. This will give a map back from H1X to 1TI(X,X)abel. To define
the map, let 'Y = Lini'Yi be a I-cycle, with paths 'Yi going from points
a(i) to b(i) in X. For each point c that occurs as an endpoint of any
174 12. The Fundamental Group

'Yi, choose a path Te from x to c. Let 'Y/ be the loop at x defined by

'Yi
I
= Ta(l)' 'Yi . Tb(i)
-I
,
where we are using the notation of Exercise 12.3(b). Define the map
from ZIX to 'TT\(X,X).bel by sending 'Y = Lini'Yi to the class of
TI i['Y/J\ note that the order in products is unimportant since the group
'TTl (X, X).bel is abelian.
We verify first that this is independent of the choice of paths Te'
Suppose Te is another path from x to c, for each c. Let "1/ = l'a(/) • 'Yi . 1'b(1) -I.
Let {}c be the loop 1'c' Te- I. Then ["1/] = [{}a(i)]' ['Y/J' [{}b(/)-I], so

n['Y/r = n['Y/r (n [{}a(i)r n

I
i

I
i •

,
i •

I
[{}b(i)r n ) = n
I
['Y/r' ,

the last equation using the fact that 'Y is a I-cycle, so each point c
occurs as many times as a starting point a(i) as ending point b(i).
It follows from the definition that this map is a homomorphism
from ZIX to 'TTl (X, X).bel' We next verify that this homomorphism maps
boundary cycles to zero. It suffices to show that 'Y maps to zero when
'Y = (If = 'YI + 'Y2 - 'Y3 - 'Y4,
where f: [0, I] X [0, I]~X is a continuous mapping, and the bound-
ary is as described in §6a. Let TI and T2 be paths from x to the starting
and ending points of 'YI, and let T3 and T4 be paths from x to the
starting and ending points of 'Y3' Then 'Y maps to the class of
[TI . 'YI . T2 -I] • [T2' 'Y2 . T4 -I] . [T4' 'Y3 -I. T3 -I] • [T3 . 'Y4 -I. TI- 1]
= [TI . 'YI . 'Y2' 'Y3 -I. 'Y4 -I. TI- I] = [(TI . 'YI . 'Y2)' (TI . 'Y4' 'Y3)-1].
To see that this last class is trivial in 'TTI(X,X), it suffices to show that
the paths TI . 'YI . 'Y2 and TI . 'Y4 . 'Y3 are homotopic with fixed endpoints.
For this it is enough to show that 'YI . 'Y2 and 'Y4' 'Y3 are homotopic with
fixed endpoints. This is evident from the picture:

14 13 13
'\. '\.
/ /

- 14 0:
0:
12
- r

"- "-
/ /
11 12 11

An explicit homotopy is given by the formula

12c. Fundamental Group and Homology 175

f(2t(1 - s), 2ts), Os t S 1/2,

H(t, s) = {
f«2t - 1)s + 1 - s, (2t - 1)(1 - s) + s), 1/2 StS1.

Finally, we must check that the two homomorphisms we have de-

fined are inverse to each other. It is immediate from the definitions
that the composite
'lTI(X,X)abel ~ HIX ~ 'lTI(X, X) abel
is the identity. For the other composite, for a 1-cycle 'Y = 2,ini'Yi, and
'Y/ defined as above, the element IIib/r; in 'IT,(X,X)abel maps to

using the fact that the boundary of'Y is zero. D

So far, all the fundamental groups we have calculated explicitly

have been abelian, in which case the proposition says that
'lTI(X,x)~H,(X). We will soon calculate other examples with a non-
abelian fundamental group. For now, however, we can appeal to Ex-
ercise 11.14 to find an example with a nonabelian fundamental group.
 PART VII

COVERING SPACES AND

FUNDAMENTAL GROUPS, II

In this part we will see that the two basic notions of covering spaces
and fundamental groups are intimately related. In Chapter 13 we see
how knowledge of the fundamental group controls the possible cov-
erings a space may have: coverings correspond to subgroups of the
fundamental group. There is a universal covering, from which all other
coverings can be constructed. In Chapter 14 we use this the other way
to prove the Van Kampen theorem, which relates the fundamental
group of a union of two spaces to the fundamental groups of the two
spaces and the fundamental group of their intersection. This can be
regarded as the analogue for the fundamental group of the Mayer-
Vietoris theorem for the homology group. The proof we give depends
on a correspondence between G-coverings and homomorphisms from
the fundamental group to G.

Caution to Beginners. The theorems relating coverings and the fun-

damental group are stated in their natural generality, with various
technical conditions about the spaces involved. This has two advan-
tages: you will have a reasonably complete story when you are fin-
ished, and the precise conditions sometimes help in shaping the proofs.
However, this also has a large disadvantage: the proliferation of these
technical conditions can get in the way of the main ideas. At least
178 Part VII. Covering Spaces and Fundamental Groups, II

for a first reading, it is probably a good idea to simply assume all

spaces arising are reasonably behaved-for example, that they are
open sets in the plane or a surface or manifold, or perhaps a finite
graph. These cases, in fact, will suffice for the applications consid-
ered in this text.
CHAPTER 13

The Fundamental Group

and Covering Spaces

13a. Fundamental Group and Coverings

We first have the basic:

Proposition 13.1. If p: Y -+ X is a covering, and p(y) = x, then the

induced homomorphism p*: '7T1(Y,Y)-+ '7T1(X,X) is an injection.

Proof. This is a consequence of the lifting properties of Chapter 11.

We must show that the kernel of p* is {e}. If a is a loop at y, and
p* [a] = e, there is a homotopy H from po a to th~ constant path Ex
at x. By homotopy lifting, H lifts to a homotopy H from a to some
path. Since JLmaps the sides and top of the unit square to the point
x, its lifting H (by uniqueness <2f path-lifting) maps the sides and top
of the square to the point y. So H is a homotopy from a to the constant
path Ey, and [a] = e, as required. 0

Exercise 13.2. (a) If a is a loop at x, and IT is the unique lifting of

a to a path starting at y, show that IT ends at y if and only if its class
is in the image of P*, i.e., [a] Ep*(1TI(Y,y». (b) If a and a' are two
paths in X from x to x', and IT and fj' are the lifts to paths in Y starting
at y, show that fj and fj' have the same endpoint if and only if [a'· a-I]
is in p*( 1T1(Y' y». In the notation of § 11 b,
y*a = y*a' ¢:::> [a' . a-I] EP*(1TJ(Y,y».

179
180 13. The Fundamental Group and Covering Spaces

Exercise 13.3. If y' is another point in Y with p(y') = x, and y and

y' can be connected by a path in Y, show that the image of 7T\(Y, y')
in 7I"\(X,x) is a subgroup conjugate to the image of 7I"\(Y,y). In fact,
if (J' is a path from y' to y, and 'Y = P 0 (J' is its image, show that

P*(7I"\(Y,y'» = ['Y]. P*(7I"\(Y,y»· br\ .

Given a covering p: Y ~ X, for any point y with p(y) = x, and any
loop (J' at x, we defined y * (J' to be the endpoint of the lift of (J' that
starts aty. This pointy*(J' is also inp-\(x). We saw in §l1b that if
(J" is a loop homotopic to (J', then y * (J" = Y * (J'. For any homotopy
class [(J'] in 71" \ (X , x), we can therefore define y * [(J'] to be y * (J'. This
defines a right action of the fundamental group 7I"\(X,x) on the fiber
p-\(x):

y x [(J'] f-+ y* [(J'],

taking y x [(J'] to the endpoint of the lift of (J' that starts at y.

Exercise 13.4. Show that this is a right group action; in particular,

y * ([ (J'] . [,.]) = (y * [(J']) * [,.]. If Y is path-connected, show that this ac-
tion is transitive: for any y and y' in p-\(x) there is some [(J'] with
y * [(J'] = y'. Show that the subgroup that acts trivially on a point y is
exactly P*(7T\(Y,y». For fixed y, show that this defines a one-to-one
correspondence between set of right cosets 1T\(X,X)/P*(1T\(Y,y» and
the fiber p-\(x). In particular, the index of the subgroup P*(1T\(Y,y»
in 7I"\(X,x) is the number of sheets of the covering.

We saw in the last chapter that fundamental groups provide an ob-

struction to the existence of mappings: if there is no map between the
groups, there cannot be a map between the spaces. The following
proposition shows that, in the case of covering maps, the converse
holds: if the fundamental groups say there may be a map, then there
will be. Recall that a space X is locally path-connected if any neigh-
borhood of any point contains a neighborhood that is path-connected.

Proposition 13.5. Suppose p: Y~X is a covering, andf. Z~X is a

continuous mapping with Z a connected and locally path-connected
space. Let x EX, y E Y, z EZ, be points with p(y) =f(z) = x. In order
If!! there to be a continuous mapping J Z~ Y with po J =I and
H(z) = y, it is necessary and sufficient thath(7T\(Z,z» be contained
in P*(7T\(Y,y»:
13a. Fundamental Group and Coverings 181

/-1]
/~/ P

Z-X
/
Such a lifting j, when it exists, is unique.
Proof. The necessity is clear from the functoriality of the fundamental
group, and the uniqueness is a special case of Lemma 11.5. For the
converse, to construct j, given a point w in Z, choose a path 'Y in Z
from z to w, and let a =/ 0 'Y, which is a path starting at x in X. Define
j(w) to be y *a; that is, j(w) is the endpoint of the path that lifts a
and starts at y. We must first show that this is independent of the
choice of path. If 'Y' is another path from z to w, then
/o('Y"'Y-I)=a"a- 1 is a loop atx. By the hypothesis, [a"a- I ] is
in the image of p*. By Exercise 13.2(b) it follows that (j' and (j end
at the same point.
We must verify that this mapping j is continuous at the point w.
Let N be any neighborhood of .f{w) that is evenly covered by p, let
V be the open set in p-I(N) that maps homeomorphic ally onto Nand
that contains j(w), and choose a path-connected neighborhood U of
w so that.f{U)CN. We need to show that! maps U into V. For all
points w' in U, we may find a path Ct from w to w' in U, and then
we can use 'Y' Ct as the path from z to w'. The lifting of
/ 0 ('Y . Ct) = (f° 'Y) • (f ° Ct) is obtained by first lifting / 0 'Y to (j, then

lifting /0 Ct. Since the latter lifting stays in V, this shows that
.f{U)C~ 0
Corollary 13.6. Let X be a connected and locally path-connected
space. Let p: Y --7 X and p': y' --7 X be two covering maps, with Y and
Y' connected and let p(y) = x and p'(y') = x. In order for there to be
an isomorphism between the coverings preserving the base points, it
is necessary and sufficient that
P*(7rI(Y,y)) = p' *(7r I(Y' ,y')).
Proof. The necessity is clear, and if these subgroups agree, the prop-
osition gives maps '1': Y --7 y' and 1\1: Y' --7 Y preserving the base points
and compatible with projections. (Note that X being locally path-con-
nected implies that covering spaces Y and Y' are also locally path-
connected.) Applying Lemma 11.5 to 1\10'1' and '1'01\1 shows that they
are isomorphisms. 0
182 13. The Fundamental Group and Covering Spaces

Exercise 13.7. Show that two connected coverings of such an X are

isomorphic, with an isomorphism that may not preserve base points,
if and only if the images of their fundamental groups are conjugate
subgroups of the fundamental group of X.

A path-connected space X is called simply connected if its funda-

mental group is the trivial group. Note that this is independent of
choice of the base point. In the preceding proposition, if Z is simply
J
connected, it follows that the liftings always exist.

Corollary 13.8. A simply connected and locally path-connected space

has only trivial coverings.
Proof. Suppose p: Y ~ X is a covering, with X simply connected. Fix
x in X, and for eachy Ep-l(X) apply the proposition to get a unique
continuous map Sy: X ~ Y with Sy(x) = y and pOSy the identity on X.
This gives an isomorphism from the trivial covering X x p-I(X) to Y,
by x X yf--+Sy(x). D

Exercise 13.9. Verify that this map is a homeomorphism from the

space X x p-I(X) onto Y.

Problem 13.10. The hypotheses of locally path-connected are needed

for the truth of Propositions 13.5 and the two corollaries. Challenge.
Find an example of a path-connected space X with trivial fundamental
group that has a nontrivial connected covering p: Y~X.

13b. Automorphisms of Coverings

Our next goal is to relate the fundamental group of X to the auto-
morphism group of a covering of X. Let p: Y ~ X be a covering, with
base point y in Y chosen so that p(y) = x, and assume Y is path-con-
nected. We want to make 'TTI(X,X) act on the left on Y. Given an
element [cr] in 'TTI(X,X) and a point z in Y, we therefore want to define
a point [cr] . z in Y. Let y' = y * [cr] be the endpoint of the lift of cr to
a path that starts at y. Choose a path 'Y from y to z in Y. Since
p(y') = p(y) = x, and po 'Y is a path starting at x, we have a point
y' * (p ° 'Y) that is the endpoint of the lift of the path po 'Y that starts
at y'. We want to define [cr] . z to be the point y' * (p ° 'Y). Denote this
pointy'*(p°'Y) temporarily by w(z,cr,'Y):
13b. Automorphisms of Coverings 183

y' [cr]·z = w(z.cr.y)

Equivalently, w(z, a, 'Y) = (y * a) * (p ° 'Y) = y * (a· (p ° 'Y». Sup-

pose, however, that we chose another path 'Y' from y to z. To have
w(z, a, 'Y') = w(z, a, 'Y), we want the lifts of a . (p ° 'Y') and a· (p ° 'Y)
that start at y to end at the same point. By Exercise 13.2(b), this is
the case precisely if the class [(a· (p 0'Y')) . (a' (p °'Y»-I] is in
P*(7TI(Y,y». Note that 'Y'. 'Y- I is a loop at y, and so
[(a· (p0'Y'». (a' (p0'Y»-I] = [(a' (p0'Y'). (po('Y- I». a-I)]
= [a]' [po('Y'. 'Y- I)]. [arl
= [a]' p*([-y" 'Y- I]). [arl ,
which is an element of [a] ·P*('ll'I(Y,y»· [arl. To know that this is
in P*('ll'I(Y,y», we need P*('ll'I(Y,y» to be a normal 7 subgroup of
'll'1(X,X). In this case we see that w(z, a, 'Y) is independent of choice
of 'Y, and depends only on z and the homotopy class [a] of a.
Assume then thatp*('ll'I(Y,y» is a normal subgroup of 'll'1(X, x). The
above construction determines a mapping
[a] X z ~ [a]' z,

where [a]' z = w(z, a, 'Y). Note that [a]· z is in the same fiber of pas
z. Next we want to show that this is a left action of 'll'1(X,X) on Y.
The fact that ([a] • [T]) . Z = [a] . ([T] . z) follows readily from the -def-
inition. For if 'Y is a path from y to z, then the lift of T • (p ° 'Y), starting
at y, is a path from y to [T]' z. It follows that [a]' ([T]' z) is the end-
point of the lift of the path a' (T . (p ° 'Y» that starts at y. The path
(a' T) . (p °'Y) is homotopic to a· (T' (po 'Y», so its lift at y has the
same endpoint, and this endpoint is ([a] . [T]) . z. The fact that [Ex] . Z = Z

7 Recall that a subgroup H of a group G is a nonnal subgroup if g . H . g -I C H for

all g in G.
184 13. The Fundamental Group and Covering Spaces

follows from the fact that [E..J . z = (y * E..) * 'Y = Y * 'Y, and y * 'Y = z by
definition.
To prove that, for fixed [u], the map z~ [u]· z is continuous, we
assume in addition that X is locally path-connected. To see the con-
tinuity near a point z, take, as in the preceding proposition, a path-
connected neighborhood N of p(z) that is evenly covered by p. Let V
and V' be the components of p-l(N) that contain z and z' = [u]· z.
We must show that [u] . V is contained in V'. For v in V, let a be a
path from z to v in V. If'Y is a path from y and z, then 'Y. a can be
used as the path from y to v, from which it follows that [u] . v is the
endpoint of the lift of po a that starts at z'. This lift is in V', which
concludes the proof of continuity.
Summarizing, we have constructed a homomorphism from 'lTl(X, x)
to the group Aut(Y/X) of covering transformations. We claim next
that this is surjective. Let <p: Y ~ Y be a covering transformation, and
suppose <p(y) = y'. It suffices to find an element [u] in 'lTl(X,X) with
[u] . y = y', since, Y being connected, two covering transformations
that agree at one point must be identical. Let 'Y be a path from y to
y'. Let u = p 0 'Y. Then [u] . y is the endpoint of the lift of the path u
that starts at y. Since this lift is 'Y, [u]· Y = y', as required.
Finally, we compute the kernel of this homomorphism from 'lTl(X,X)
to Aut(Y/X). As in the preceding step, it suffices to see which [u]
act trivially on y, and this happens when the lift of u at y ends at y,
i.e., when [u] is in P*('lTl(Y,y». Putting this all together, we have
the:

Theorem 13.11. Let p: Y~X be a covering, with Y connected and

X locally path-connected, and let p(y) = x. If p*( 7Tl(Y' y» is a normal
subgroup of 7Tl(X,X), then there is a canonical isomorphism
'lTl(X,X)/P*('lTl(Y,y» ~ Aut(Y/X).

The covering is a G-covering, with G being the quotient group

7Tl(X, x)/p*( 7Tl(Y' y».

The last statement follows from Proposition 11.38. A covering p: Y ~ X

is called regular if P*('lTl(Y,y» is a normal subgroup of'lTl(X,x).

Exercise 13.12. Still assuming Y connected and X locally path-con-

nected, but without assuming H = P*('lTl(Y,y» is a normal subgroup
of 'lTl(Y'Y)' let N be the normalizer of H in 'lTl(X,X), i.e., N is the
subgroup of elements g in 'lTl(X,X) such that g. H· g-l C H. Show that
l3b. Automorphisms of Coverings 185

N acts on the left on Y, and that this detennines an isomorphism

NIH == Aut(YIX) .

Exercise 13.13. Show that the following are equivalent (with Y con-
nected and X locally path-connected): (i) the covering is regular;
(ii) the action of Aut(YIX) on p -lex) is transitive; and (iii) for every
loop (J' at x, if one lifting of (J' is closed, then all liftings are closed.

Exercise 13.14. Show that any G-covering p: Y~X, with Y con-

nected and locally path-connected, is a regular covering.

Corollary 13.15. If p: Y ~ X is a covering, with Y simply connected

and X locally path-connected, then 7Tl(X,X) ==Aut(YIX).

Corollary 13.16. If a group G acts evenly on a simply connected and

locally path-connected space Y, and X = Y IG is the orbit space, then
the fundamental group of X is isomorphic to G.

In fact, we know by Proposition 11.37 that G is canonically iso-

morphic to the group of automorphisms of Y over X. Choosing a point
yin Y over a point x in X detennines an isomorphism of 1Tl(X,X) with
Aut(YIX) = G. 0

In general, the isomorphism of G with 1Tl(X,X) depends on the choice

of base point y, but only up to inner automorphism. In particular, in
case the group is abelian, the isomorphism is independent of choices.
For example, this corollary, applied to the mapping from ~ to Sl,
implies again that 1Tl(SI, x) = Z.
Applied to the two-sheeted covering p: Sn ~ ~pn from the sphere
to the projective space, it follows that
1Tl(IRPn, x) = Z/2Z for n?: 2,
where x is any point in IRpn.
This can be used to give a more conceptual explanation of some-
thing we saw in Chapter 4: if n ?: 2, there can be no continuous map-
ping from Sn to Sl with g( - P) = - g(P) for all P. Such a map would
define a continuous mapping h: IRpn~ IRpl on quotient spaces:

~Sl
t/5'/~ t
S"

p p'

~1Jl>"h~lJl>l,
186 13. The Fundamental Group and Covering Spaces

where p': Sl---+ 1RiP>1 is the corresponding mapping for n = 1. Since

1T1(1RiP>1, h(x» == Z, the mapping h*: 1T1(1RiP>\ x)---+ 1T1(1RiP>1, h(x» must
be trivial. Choosing a point yin Sn with p(y) = x, by Proposition 13.5
there is a continuous mapping h from lRiP>n to Sl so that p' 0 h = hand
h(x) = g(y). Now hop and g are two mappings from Sn to Sl that map
y to g(y), and both, when followed by p', are the map hop. By Lemma
11.5, hop = g. But hop always takes P and -P to the same point,
while g never does. So such g cannot exist.

Exercise 13.17. (a) Compute the fundamental group of the Lens spaces
of Exercise 11. 30. (b) If the Klein bottle is constructed by identifying
sides as shown:

-0-
b

show that the fundamental group has two generators a and b, with
one relation abab- I = e. In particular, the fundamental group is not
abelian. (c) The torus is a two-sheeted covering of the Klein bottle,
as in Exercise 11.27. Describe the image of the fundamental group
of the torus in the fundamental group of the Klein bottle, and verify
that it is a normal subgroup.

Exercise 13.18. If p: Y ---+ X is the three-sheeted covering of Exercise

11.14, show that P*(1TI(Y,y» is not a normal subgroup of 1T1(X,X).
In particular, 1T1(X,X) is not an abelian group.

13c. The Universal Covering

In this section we assume that X is a connected and locally path-
connected space. A covering p: Y ---+ X is called a universal covering
if Y is simply connected. It follows from Corollary 13.6 that such a
covering, if it exists, is unique, and unique up to canonical isomor-
phism if base points are specified. As we have seen, the fundamental
group of X will be isomorphic to the automorphism group of this
Bc. The Universal Covering 187

covering. The aim of this section is to show that, if one additional

property is satisfied, a universal covering always exists.
Suppose we have a universal covering p: Y~ X. Any point in X
has an evenly covered path-connected neighborhood N. Any loop (J
in N lifts to a loop cr in Y, and, since Y is simply connected, this loop
is homotopic in Y to a constant path. It follows that the loop (J = p 0 cr
is homotopic to a constant path in X. A space X is called semilocally
simply connected if every point has a neighborhood such that every
loop in the neighborhood is homotopic in X to a constant path. So
being semilocally simply connected is a necessary condition for the
existence of a universal covering. Note that if X is locally simply
connected, i.e., if every neighborhood of a point contains a neigh-
borhood that is simply connected, then X is semilocally simply con-
nected.
The spaces one generally meets, and those we have considered in
this book, are all locally simply connected. Any open set in the plane
or in [Rn, or any manifold, or any finite graph, is locally simply con-
nected. In fact, one has to work a little to produce spaces that are not
locally simply connected or semilocally simply connected.

Exercise 13.19. For a positive integer n, let Cn be the circle of radius

1/2n centered at the point (1/2", 0). Let C C ~2 be the union of all
the circles Cn; C is sometimes called a clamshell. (a) Show that C is
connected and locally path-connected, but not semilocally simply
connected. Let X be the cone over C, i. e., X C [R3 is the union of all
line segments from points in C to the point (0,0, 1). (b) Show that
X is semilocally simply connected but not locally simply connected.

Suppose we have a universal coveringp: Y~X, withp(y) =x. For

any point z in Y, there is a path 'Y from y to z, which is unique up to
188 13. The Fundamental Group and Covering Spaces

homotopy since Y is simply connected. The image ex = p "I is a path

0

from x to p(z), unique up to homotopy (with fixed endpoints). Con-

versely, given a path ex in X starting at x, it detennines a point z = y * ex
in Y. This identifies Y, at least as a set, with the set of homotopy
classes of paths in X that start at a given point x. The idea to the proof
of the following theorem is to use this observation in reverse, by using
these homotopy classes to construct the universal covering.

Theorem 13.20. A connected and locally path-connected space X has

a universal covering if and only if it is semilocally simply connected.
Proof. To construct a universal covering, fix x in X and define X to
be the set of homotopy classes [-y] of paths "I in X that start at x, the
homotopies as usual being required to fix endpoJEts. Assigning to
each such class its en~oint defines a function u: X-X. Our task is
t.2, put a topology on X so that this is a covering map, and show that
X is simply connected.
Call an open set N in X good if N is path-connected, and every
loop in N at a point z in N is homotopic to the constant path ez in X,.:..
If "I is a path from x to a point z in N, let N h ) be the subset of X
consisting of the homotopy classes ["I' ex], where ex is any path in N
that starts at z. Note that N['Y) depends only on the homotopy class [-y]
of "I. Here are some of the properties of these sets:
(I) if J3 is a path in a good N starting at the endpoint of "I, then
N h · 13 )=N['Y);
(2) if a good N is contained in a good N', then N h ) C N'h); and
(3) if "I and "I' are two paths from x to a point z in a good N, then
N)'Y) = N h ,) if "I and "I' are homotopic, and N)'Y) n N['Y') = 0
otherwise. 0
Exercise 13.21. Verify these properties.

Now define a subset 0 of X to be open if, for any [-y] in 0, there

is a good neighborhood N of the endpoint of "I with N['Y) cO. It fol-
low§..from properties (1) and (2) that these open sets fonn a topology
on X, and that each of these sets N['YJ is open, and the projection from
NI'Y) to N is continuous. In fact, this projection is a homeomorphism,
the inverse being given by the map that takes w in N to ["I' ex], where
ex is any path in N from the endpoint of "I to w. This is independent
of choice of ex, since if ex' is another, ex and ex' are homotopic in X,
so "I' ex is homotopic to "I' ex'. This projection is continuous since for
smaller good N~ N' maps to N'['Y)'
The map u: X - X is evenly covered over any good N, since the
13d. Coverings and Subgroups of the Fundamental Group 189

inverse image of N is a disjoint union of the open sets N[-Yl' where ['Y]
varies over the homotopy classes of paths from x to any given point
z of N.
For any path 'Y starting at x in X, and for s between 0 and 1, let
'Ys be the eath defined by 'Yit) = 'Y(st), 0:5 t:5 1. The mapping
.y: [0, 1] ~ X defined by .y(s) = ["(s] is the unique lift of 'Y to a path
iQ. X that starts at the base point t= [Ex]. In particular, it is a p,~th in
X from £ to ['Y], showing that X is connected. Any loop in X at £
has the form .y for a unique loop 'Y at x. For .y to be a loop, the
endpoint 'YI = 'Y must be homotopic to the constant path at x, which
implies by the lifting of homot~ies that .y is homotopic to the con-
stant path at £. This shows that X is simply connected, and completes
the proof of the theorem. 0

Problem 13.22. (a) Suppose X is locally simply connected. Show

that, if p: Y~X and q: Z~Y are covering maps, thenpoq: Z~X is
also a covering map. (b) Find a counterexample to (a) when X is a
clamshell.

13d. Coverings and Subgroups of the

Fundamental Group
The theorem of the preceding section will determine correspondence
between subgroups of the fundamental group and coverings. For the
following proposition, assume that X is connected, locally path-con-
nected, and semi locally simply connected, so that X has a universal
covering.

Proposition 13.23. (a) For every subgroup H of '1T1(X,X) there is a

connected covering PH: YH~ X, with a base point YH E PH -I(x) so that
the image of'1TI(YH, YH) in '1T1(X, x) is H. Any other such covering (with
choice of base point) is canonically isomorphic to this one.
(b) If K is another subgroup of '1T1(X,X) containing H, there is a
unique continuous mapping PH,K: YH~ YK that maps YH to YK and is
compatible with the projections to X. This mapping is a covering map-
ping, and if H is a normal subgroup of K, it is a G-covering with
G=K/H.
Proof. Let u: X ~ X bejl universal covering, with u(£) = x, and iden-
tify EI(X,x) with Aut(X/!D. ~y subgroup H of_'1TI(X,x) acts evenly
on X, and the quotient X~X/H= YH makes X the universal cov-
190 13. The Fundamental Group and Covering Spaces

ering of YH. The fundamental group of YH (with base point YH the

imas,e of x) is canonically isomorphic to H. The projection from YH
to X/'ITl(X,X) =X is a covering, and the image of its fundamental
group in 'ITl(X,X) is H. !! H is ~ntained in K, there is a canonical
map on the orbit spaces X /H ~ X / K. The remaining verifications are
left to the reader, using the results of the preceding section. 0

As seen in the pr2,.of, the covering YH~X corr~sponding to H can

be id£.l1tified with X/H~X, where H acts on X as a subgroup of
Aut(X/X) = 'ITl(X,X). The regular coverings correspond to normal
subgroups H. This means (see Exe~ise 13.14) that every connected
G-covering p: Y~X has the form X/H~X, with
'ITl(X,x)/H == Aut(Y/X) == G.
The correspondence of the proposition is similar to that seen in
Galois theory, where subgroups correspond to field extensions, smaller
subgroups corresponding to larger extensions. Here smaller subgroups
of the fundamental group correspond to larger coverings of the space:
X,x {e}
~ n
H
n
K
n
X,X G
If H is a normal subgroup of K, then the covering YH~ YK is a K/H-
covering.
For example, if X = Sl is a circle, the coverings correspond to
subgroups of 'IT 1(Sl , (1, 0)) = 7L. The trivial group corresponds to the
universal covering ~ ~ S\ and the subgroups 7Ln C 7L correspond to
the n-sheeted covering Pn: Sl~Sl considered in §l1a. Up to iso-
morphism, these are the only connected coverings of a circle.

Exercise 13.24. If 'ITl(X,X) is abelian, show that every connected cov-

ering of X is regular.

One simple space whose universal covering we have not yet con-
structed is the figure 8 space
13d. Coverings and Subgroups of the Fundamental Group 191

or the join of two circles at a point. This universal covering can be

constructed as an infinite tree. Here part of it is pictured with seg-
ments of decreasing size, so that they don't overlap in the plane, but
one should imagine that they are all of the same lengths.

a
a

b
a
a
b a

a
a
b

Each segment maps to one of the loops of the figure 8 space, the
horizontal segments to one and the vertical segments to the other.

Problem 13.25. (a) Show that this space is simply connected, and is
the universal covering of the figure 8 space X. (b) If you know what
a free group is, show that a free group with two generators acts evenly
on this universal covering, with orbit space X. Deduce that the fun-
damental group of X is this free group. (c) Generalize to the space
obtained by joining n circles at a point.

In Chapter 14 we will see a more general method of calculating

192 13. The Fundamental Group and Covering Spaces

fundamental groups, which will include another proof of the result of

the preceding problem.
The covering corresponding to the commutator subgro~ of the fun-
damental group of X is a covering we may denote by X.bel :
X.bel = i';[1Tl(X, x), 1Tl(X, x)] ~ X.
This is a G-covering, with G = 1Tl(X, X)/[1Tl(X, x), 1Tl(X,X)], which is
the first homology group HlX by Proposition 12.22. It is sometimes
called the universal abelian covering of X. In general a covering is
called abelian if it is regular with abelian automorphism group.

Exercise 9.26. Show that any connected abelian covering of X has

the form X.bel/H, for some subgroup H of HlX,

We saw in Chapter 6 that for open sets in the plane, the homology
class of a closed I-chain is determined by winding numbers around
points not in the set. The next exercise shows that this is true for
other nice subsets, but the following problem shows it is not true in
general.

Exercise 13.27. Suppose X is a subset of the plane that is contained

in some open set U for which there is a retract r: U ~ X. Show that
a closed I-chain 'Y on X is homologous to zero if and only if W( 'Y, P) = 0
for all P not in X.

Problem 13.28. Challenge. Let X be the clamshell of Exercise 13.19.

Give an example of a closed path 'Y on X such that W( 'Y, P) = 0 for
all P not in X, but such that 'Y is not a I-boundary on X.

Exercise 13.29. The universal covering of the complement of the

origin in the plane can be realized as the right half plane, via the
polar coordinate mapping (r,{})~(rcos({}),rsin({}», and as the en-
tire complex plane C via the mapping z~exp(z). Find an isomorph-
ism between these coverings.
CHAPTER 14

The Van Kampen Theorem

14ao G-Coverings from the Universal Covering

In this section X will denote a connected, locally path-connected, and
semilocaUy jimply connected space, so X has a universal covering,
denoted u: X ~ X. All spaces will have base points, and aU maps will
be assumed to take base points to b~se points. The base point of X
is denoted x, and the base point of X over x is denoted x.
We have seen in §13d that for every G-covering p: Y~X, with Y
connected, and with base point y, there is a surjective homomorphism
of '1T,(X,x) onto G. If H is the kel"!!,el of this homomorphism, so
G "= 'IT,(X,x)/H, Y is the quotient of X by the action of H, with y the
image of x. We want to extend this correspondence to G-coverings
that may not be connected. In this case there will only be a homo-
morphism from 7T,(X,X) to G, which need not be surjective. Here we
will set up this correspondence between G-coverings and homomor-
phisms directly and rather briefly, omitting some verifications. Other
ways to carry this out, with a more general context for these con-
structions, together with more details about the verifications, are de-
scribed §16d and §16e.
Suppose P: 7T,(X,X)~G is a homomorphism from the fundamental
group of X to any group G . We will construct from P a G-covering
pp: Yp ~ X, together with a base point YP ig Yp over x. Give G the
discretej:opology, so the Cartesian product X X G is a product of cop-
ies of X, one for each element in G. The group 7T, (X, x) acts on the
left on X X G by the rule
[u]o(zXg) = [u]ozXgop([ur') = [u]ozXg'p([u]r',

193
194 14. The Van Kampen Theorem

for [(J] E 'lTl(X, x), z EX, g E G. Here [(J] . z is the action of 'lTl(X, x)
on X that was described in §13b, and g' p([(Jrl) is the product in
the group G. Define Yp to be the quotient of Xx G by this action of
'lTl(X, x):
Yp = XX G/'lTl(X, x) ,
and let YP be the image of the point f x e in Yp. Let (z X g) denote
the image in Yp of the"point z X g in X X G. No~ that, by the above
action of 'lTl(X,X) on X X G, we have, for z in X, g in G, and [(J] in
'lTl(X, x),
([(J]'ZXg) = (zXg·p([(J]».
Define Pp: Yp ~ X by taking (z X g) to u(z).
The group G acts on Yp by the formula h· (z X g) = (z X h· g), for
hand g in G and z in X. (Note that using the right side of G for the
left action of 'lTl(X, x) frees up the left side of G for a left action of
G!) We claim that this is an even action, making Pp: Yp~X a G-
covering. To prove th~, let N be any open set in X over which the
universal covering u: X ~ X is trivial. By Lemma 1l.18 there is an
isomorphism of u-1(N) with the product covering N X 'lTl(X,X), on
which 'lTl(X,X) acts on the left on the second factor. This gives ho-
meomorphisms
pp-l(N) == (NX 'lTl(X, x» X G/'lTl(X, x) == NXG,
the latter homeomorphism by «u X [(J]) X g)~ U X g' p([(J]). (The map
back takes u X g to «u X e) X g).) These homeomorphisms are com-
patible with the projections to N, and it follows that, over N, the
action of G is even and the covering is a G-covering. Since X is
covered by such open sets N, the same is true for the map Pp from
Yp to X.
Conversely, suppose p: Y ~ X is a G-covering, with a base point Y
over x. From this we construct a homomorphism p from 'lTl(X,X) to
G. For each [(J] in 'lTl(X,X) the element p([(JD in G is determined by
the formula
p([(J]) . Y = Y * (J ,
where Y * (J is the endpoint of the lift of the path (J that starts at y.
We will need two facts about this operation:
(i) (z*(J) *,. = z* «(J''') for z Ep-l(X), (J a loop at x, and,. a
path starting at x;
(ii) g' (z*'Y) = (g. z) *'Y for g E G, z Ep-l(X), and'Y a path
starting at x.
14a. G-Coverings from the Universal Covering 195

The first of these facts is immediate from the definition. The second
follows from the fact that if 'Y is lifting of -y starting at z, then the
path tt--') g. 'Y(t), 0:5 t:5 1, is a lifting of -y that starts at g. z. The
endpoint of this path, which is (g. z) * -y by definition, is g. 'YO), and
since 'Y(l) = z * -y, (ii) follows.
We claim now that the p defined above is a homomorphism. This
is a calculation, using (ii) and (i):

(p([cr])· p([,.])). y = p([cr])· (p([,.]). y) = p([cr])· (y*,.)

= (p([cr])·y)*,.=(y*cr)*,.
= y*(cr·,.)=p([cr]·[,.])·y.

Proposition 14.1. The above constructions determine a one-to-one

correspondence between the set of homomorphisms from '7TI(X,X) to
the group G and the set of G-coverings with base point, up to iso-
morphism:

Hom(1TI(X,X),G) B {G-coverings}/isomorphism.

Proof. Given a G-covering p: Y ~ X with base points, from which we

constructed a homomorphism p, we must now show that the given
covering is isomorphic to the coveri!!.Spp: Yp~X constructed from p.
To map Yp to Y, we need to map X x G to Y, and show that orbits
by 1TI(X,X) have the same image. For this we identify the universal
covering X as the space of homotopy classes of paths in X starting
at x. Define a map

XXG~Y,

This is easily checked to be continuous. We must check that an equiv-

alent point ([cr] . [-y]) x (g. p([cr])-I) maps to the same point. By (i)
and (ii), this point maps to

(g·p([cr])-I)·(y*(cr·-y» = (g·p([cr])-I)·«y*cr)*-y)
= «g. p([cr])-I) . (y * cr» *-y
= (g. «y * cr) * cr -I» * [-y]
*
(g. (y* (cr· cr- I))) [-y] = (g. y) * [-y],
as required. Since the map takes the same values on equivalent points,
it gives a mapping from the quotient Yp to Y, which is a mapping of
covering spaces of X. This is easily checked to be a mapping of G-
coverings, from which it follows that it must be an isomorphism.
Conversely, starting with a homomorphism p, we constructed a G-
196 14. The Van Kampen Theorem

covering Yp ~ X, from which we constructed another homomorphism,

say p. We must verify that p = p. Now for [0'] in 1Tl(X,X),
p([O'])'(xXe) (.fXe)*O' = (x*O'Xe)
= ([O']'xXe) = (xXe'p([O']» = (xXp([O']»
= p([O'D' (xx e).
This shows that p([O']) = p([O']), which concludes the proof. 0

Exercise 14.2. If p: Y~ X is the G-covering corresponding to a ho-

momorphism p: 1Tl(X,X)~ G, and X' is a subspace of X that also has
a universal covering, with x in X', show that the restriction p -l(X')~ X'
of this covering to X' is the G-covering corresponding to the com-
posite homomorphism poi*, where i*: 1Tl(X',X)~1Tl(X,X) is induced
by the inclusion i of X' in X.

Exercise 14.3. Show that, if base points are ignored, two G-coverings
Yp and Yp' are isomorphic G-coverings if and only if the homomor-
phisms p and p' are conjugate, i.e., there is some gin G such that
p'([O']) = g' p([O']). g-l for all [0'] E 1Tl(X,X).

14b. Patching Coverings Together

Suppose X is a union of two open sets U and V. A covering of X
restricts to coverings of U and V, which are isomorphic over un V.
Conversely, suppose we have coverings P2: Y l ~ U and P2: Y2~ V,
and we have an isomorphism of coverings
it: Pl-l(U n V)~ P2 -l(U n V)
of un v. Then one may patch (or "glue," or "clutch") these together
to get a covering p: Y ~ X, together with isomorphisms of coverings
'PI: Yl~p-l(U), 'P2: Y2~p-l(V)
of U and of V, so that, over un V, it = 'P2"1 °'Pl'
One can construct Y as the quotient space of the disjoint union
Yl U Y2 , by the equivalence relation that identifies a point Yl in
Pl-l(U n V) with the point it(Yl) in P2 -l(U n V). (See Appendix A3.)
Since it is compatible with maps to X, one gets a mapping P from Y
to X. Since the map from Y l to Y is a homeomorphism onto its image
p-1U, which is open in Y, and similarly Y2 maps homeomorphically
onto P -1 V, one sees that the restriction of P to the inverse image of
14c. The Van Kampen Theorem 197

U is isomorphic to YI ~ U, and the restriction over V is isomorphic

to Y2~ V. From this it follows in particular that p is a covering map.
If each of Y I ~ U and Y2~ V is a G-covering, for a fixed group
G, and {} is an isomorphism of G-coverings, then Y ~ X gets a unique
structure of a G-covering in such a way that the maps from YI and
Y2 commute with the action of G.
Occasionally the following generalization is useful. Suppose we have
a collection Xa of open sets, a E 91, whose union is X, and a collec-
tion Pa: Ya~ Xa of covering maps. Suppose, for each a and [3, we
have an isomorphism
{}~a: Pa-I(XanX~) ~ p~-I(XanX~)

of coverings of Xa n X~. Assume these are compatible, i.e.,

(I) {}aa is the identity on Va; and
(2) {}"Ya={}"y~o{}~a onpa-I(XanX~nX"y) for all a, [3, "(Ed.
Then one can patch these coverings together to obtain a covering
p: Y~X. One has isomorphisms 'Pa: Ya~ p-I(Xa) of coverings of Xa,
such that {}~a = 'P~-I 0 'Pa on Pa -1(Xa n X~). In addition, the space Y is
the union of the open sets 'Pa(Ya).
One constructs Y as the quotient space UaE.Id YaiR of the disjoint
union of the Ya by the equivalence relation determined by the {}~a's.
The assertions about Y and the 'Pa are general facts about patching
spaces together, as proved in Appendix A3. The map p is determined
by the equations p°'Pa == Pa on Va. Since 'Pa is a homeomorphism of
Ya onto p-I(Xa), it follows that p is a covering map.
If each Pa: Ya~Xa is a G-covering, with fixed G, and each 6~a is
an isomorphism of G-coverings, then there is a unique action of G
on Y so that each 'Pa commutes with the action of G, i.e.,
'Pa(g' Ya) = g' 'Pa(Ya) for g in G and Ya in Va' This gives the patched
covering p: Y ~ X the structure of a G-covering, so that each 'Pa is
an isomorphism of G-coverings.

14c. The Van Kampen Theorem

The Van Kampen theorem describes the fundamental group of a union
of two spaces in terms of the fundamental group of each and of their
intersection, under suitable hypotheses. Let X be a space that is a
union of two open subspaces U and V. Assume that each of the spaces
U, V and their intersection U n V is path-connected, and let x be a
198 14. The Van Kampen Theorem

point in the intersection. Assume also that all these spaces X, U, V,

and un V have universal covering spaces; this is the case, for ex-
ample, if X is locally simply connected. We have a commutative dia-
gram of homomorphisms of fundamental groups:
1t1(U,x)

y~
ItI(Un V, x)

1t1(V, x)

The maps are induced by the inclusions of subspaces, and commu-

tativity means that jl ° i l = jz ° i2.
We will describe how '!T,(X,x) is determined by the other groups
(and the above maps between them). The description will not be di-
rect, but will be by a universal property. Note that any homomor-
phism h from '!TI(X,X) to a group G determines a pair of homomor-
phisms hI = hoj, from '!TI(U,X) to G and h2 = hojz from '!TI(V,X) to
G; the two homomorphisms h,oi, and h 2o i 2 from '!T,(Un V,x) to G
determined by these are the same. The Van Kampen theorem says
that '!TI(X,X) is the "universal" group with this property.
1t 1(U,x)

y~
G.

Theorem 14.4 (Seifert-van Kampen). For any homomorphisms

hI: '!TI(U,X)~G and h 2 : '!T,(V,x)~G,

such that h, ° i, = h2 ° i 2, there is a unique homomorphism

h: '!T,(X,x) ~ G,

such that hOj, = h, and h oj2 = h2 •

Exercise 14.5. Show that '!T,(X, x), together with the homomorphisms
14c. The Van Kampen Theorem 199

jl and h, is detennined up to canonical isomorphism by the universal

property.

Exercise 14.6. Use the universal property to show that 1T1(X,X) is

generated by the images of 1T1(U,X) and 1T1(V,X). Can you prove this
assertion directly?

A version of the Van Kampen theorem was found first by Seifert,

and the theorem is also known as the Seifert-Van Kampen theorem.
The version given here, via universal properties, was given by Fox,
see Crowell and Fox (1963). The usual proof of the Van Kampen
theorem (without the hypotheses that the spaces all have universal
coverings) is rather technical, and for it we refer to Crowell and Fox
(1963) or Massey (1991). Here we will give a quick proof, due to
Grothendieck (see Godbillon (1971», using the correspondence be-
tween homomorphisms from fundamental groups to a group G and
G-coverings. The assumptions assure that each of the spaces X, U,
V, and un V has a universal covering space, and that homomor-
phisms from their fundamental groups to a group G correspond to
G-coverings.
In particular, the homomorphisms hi and h2 detennine G-coverings
Y I ~ U and Y2~ V, together with base points YI and Y2 over x. The
fact that hi 0 i I is equal to h20 i2 means that the restrictions of these
coverings to un V are isomorphic G-coverings, and since un V is
connected, there is a unique isomorphism between these G-coverings
that maps the base point YI to the base point Y2. (The uniqueness is
a special case of Exercise 11.24.) By the construction of the preceding
section, these two coverings patch together, using this isomorphism
over the intersection. This gives a G-covering Y ~ X that restricts to
the two given G-coverings (and has the same base point). This
G-covering corresponds to a homomorphism h from 1T1(X,X) to G,
and the fact that the restricted coverings agree means precisely that
hojl = hi and h oj2 = h2· D

Corollary 14.7. If U and V are simply connected, then X is simply

connected.

Note the important hypothesis in all these theorems, that all spaces,
induding the intersection un V, are connected. It does not apply to
the annulus, written as a union of two sets homeomorphic to disks!

Exercise 14.8. If V is simply connected, show thatjl: 1T1(U, x)~ 1T1(X, x)

200 14. The Van Kampen Theorem

is sUljective, with kernel the smallest normal subgroup of 'TT\(X,x) that

contains the image of i\: 'TT\(V n V, x)~ 'TT\(V, x).

Corollary 14.9. If V n V is simply connected, then, for any G,

Hom('TT\(X,x),G) = Hom('TT\(V,x),G) x Hom('TT\(V,x),G).
This means that 'TT\(X,x) is the free product of 'TT\(V,x) and 'TT\(V,x).

Exercise 14.10. If V n V is simply connected, show that the inclusion

mappings j\ and j2 are one-to-one.

The following is a useful generalization of Van Kampen's theorem,

which can be used to compute the fundamental group of an increasing
union of spaces, each of whose fundamental groups in known. The
proof is identical to that of the preceding theorem, using the general
patching construction of the preceding section.
Suppose a space X is a union of a family of open subspaces Xa ,
a E.iJ., with the property that the intersection of any two of these
subspaces is in the family. Assume that X and each Xa is path-con-
nected and has a universal covering, and that the intersection of all
the Xa contains a point x. When XII is contained in Xa let iall be the
map from 'TT \(XII , x) to 'TT\ (Xa , x) determined by the inclusion, and let
ja be the map from 'TT\(Xa,x) to 'TT\(X,x) determined by inclusion.

Theorem 14.11. With these hypotheses, 7T\(X,X) is the direct limit of

the groups 7T\(XaoX). That is,for any group G, and any collection of
homomorphisms ha from 7T\ (Xa , x) to G such that h(3 = ha 0 ia(3 when-
ever X(3CXa, there is a unique homomorphism hfrom 7T\(X,X) to G
such that ha = hojafor all a. D

The preceding theorem is recovered by taking the family to consist

of V, V, and vn V.
Although this version of Van Kampen's theorem is stated with each
subspace Xa. open in X, it can often be applied to subspaces that are
not open. For example, if each Xa is contained in an open set Va, of
which it is a deformation retract, with VII e Va. whenever XII exa ,
and the hypotheses of Theorem 14.11 apply to these Va, then 'TT\(X, x)
is the direct limit of the groups 'TT\(Xa,x). This follows from the fact
that each 'TT\(Xa, x)~ 'TT\(Va,x) is an isomorphism. Without some such
hypotheses, however, the theorem is false. For example, if A and B
are copies of a cone over a clamshell (see Exercise 13.19), joined
together at the one point where all the circles are tangent, then the
14d. Applications: Graphs and Free Groups 201

spaces A and B are simply connected, and A n B is a point, but A U B

is not simply connected. (In fact, A U B is an example of a space that
is not simply connected but which has no nontrivial connected cov-
erings.)

Exercise 14.12. Show if a space X is a union of a family of open

subspaces Xa such that the intersection of any two sets in the family
is also in the family, then H1X is the direct limit of the groups H1Xa •

14d. Applications: Graphs and Free Groups

One simple application of the Van Kampen theorem is a result we
looked at earlier: the n-sphere Sn is simply connected if n ?: 2. To see
this now, write the sphere as a union of two hemispheres each homeo-
morphic to n-dimensional disks, with the intersection homeomorphic
to Sn-l. It follows from Corollary 14.7 that the fundamental group of
Sn is trivial. (The assumption n?: 2 is used to confirm that S·-l is
connected.)
Consider next a figure 8:

This is the union X of two circles U and V meeting at a point x. Let

'YI and 'Y2 be loops, one around each circle. The fundamental group
of each circle is infinite cyclic, generated by the classes of these loops.
It follows that to give a homomorphism from 'iT,(X,x) to a group G
is the same as specifying arbitrary elements gl and g2 in G: there is
a unique homomorphism from 'iTl(X,X) to G mapping l'Y,] to gl and
['Y2] to g2' This means that 'iT,(X,x) is thefree group on the generators
['Yd and ['Y2]'

Exercise 14.13. Let a = l'Yd and b = l'Y2]' Show that every element
in 'iT1(X,X) has a unique expression in the form

where the mj are integers, all nonzero except perhaps the first and
last. The identity element is e = aObo.

The· free group on two generators a and b can be constructed di-

202 14. The Van Kampen Theorem

rectly (and algebraically) as the set of all words of this form, with
products defined by juxtaposition of words, canceling to get a legit-
imate word. It is straightforward (if a little awkward) to show by hand
that this forms a group, and to see that it satisfies the above universal
property. With the use of the VanKampen theorem, we can avoid
this, by constructing the free group as the fundamental group of this
figure 8 space.
The free group Fn on n generators a., ... , an is defined similarly:
it is generated by these elements, and, for any group G and any ele-
ments gh ... , gn in G, there is a unique homomorphism from Fn
to G taking a; to g; for 1 :::s i:::s n. Again, it can be constructed purely
algebraically using words in these letters, or as a fundamental group:

Exercise 14.14. Verify that the fundamental group of the space ob-
tained by joining n circles at a point is the free group on n generators.
Use this to show that the fundamental group of the complement of n
points in the plane is free on n generators.

Exercise 14.15. Let X be a connected finite graph. (a) Show that, for
any edge between two distinct vertices, X is homotopy equivalent to
the graph obtained by removing the edge and identifying its two end-
points. (b) Show that X is homotopy equivalent to the graph obtained
by joining n circles at a point, where, if the graph has v vertices and
e edges, n = e - v + 1. (c) Show that n is the "connectivity" of the
graph, i.e., the largest number of edges one can remove from the
graph (leaving the vertices), so that what is left remains connected.

One can use this result to give a simple proof of a rather surprising
fact about free groups.

Proposition 14.16.lfG is afree group on n generators, and H is a

subgroup of G that has finite index d in G, then H is a free group,
with dn - d + 1 generators.
Proof. Take G to be the fundamental group of a connected graph X
that has v vertices and e edges, with n = e - v + 1. For simplicity we
14d. Applications: Graphs and Free Groups 203

assume each edge of X connects two distinct vertices. The subgroup

H corresponds to a connected covering p: Y ---+ X with d sheets, with
the fundamental group of Y isomorphic to H. It is not hard to verify
that Y is a connected graph. In fact, the d points over each of the
vertices of X can be taken as vertices of Y, and (since a covering is
trivial over an interval), the d components of the inverse image of
each edge are edges of Y. Since Y is a graph, its fundamental group
is free, with
de - dv + 1 = d(e - v + 1) - d +1 = dn - d +1
generators. o
If U is the plane domain that is the complement of two points, then
U has the figure 8 as a deformation retract. So U has the same fun-
damental group. In particular, this is not an abelian group. For ex-
ample, the path "(1' "(2' "(1-1. "(2- 1 is not homotopic to a constant path
(although all integrals of all closed I-forms over this path are trivial).

Problem 14.17. Use the Van Kampen theorem to compute the fun-
damental groups of the complement of n points (or small disks) in:
(1) a two-sphere; (2) a torus; and (3) a projective plane.

Problem 14.18. Describe the fundamental group of ~2 \ Z, where Z

is the set Z of all integers, or the set Z2 of lattice points, or any infinite
discrete set.

Problem 14.19. Show that a free group on two generators contains

a subgroup that is not finitely generated, in fact, a subgroup that is
a free group on an infinite number of variables.

Problem 14.20. Use the Van Kampen theorem to compute the fun-
damental groups of: (1) the sphere with g handles; (2) the complement
of n points in the sphere with g handles; and (3) the sphere with h
crosscaps.
 PART VIII

COHOMOLOGY AND
HOMOLOGY, III

We have seen that coverings of a space can be described by giving

coverings on open sets and patching together the coverings over the
intersections ofthese open sets. In particular, one can start with trivial
coverings over small open sets, and patch them together, and any
covering arises this way. This process is formalized in Chapter 15,
and the data that describe such coverings made into Cech cohomology
classes. For G-coverings, when G is an abelian group, these classes
form a group which we see is "dual" to the homology group, i.e., it
is isomorphic to the group of homomorphisms from the first homol-
ogy group into G.
This is applied to show that the first De Rham group H1X of an
open set in the plane is isomorphic to the dual Hom(H1X, ~) of the
homology group, which gives the culmination of our experience that
the homology group and the De Rham group are measuring the same
thing about X. This allows us to translate results about homology into
corresponding results about cohomology, and in particular to finish
the proof of the Mayer-Vietoris theorem for cohomology. (Another
proof of this fact will be given in the last chapter of this book.)
Chapter 16, which is optional, contains several miscellaneous vari-
ations, applications, and generalities on the same themes, with many
of the details left as exercises. The patching construction is used to
describe the orientation covering of a manifold. The construction of
a covering of a plane domain from a closed I-form, which follows
from the general results of Chapter 15, is carried out here directly
using the language of "germs" of functions; in particular, the cov-
206 Part VIII. Cohomology and Homology, III

erings are seen as graphs of multivalued functions, similar to the polar

coordinate covering of Part I. We also describe briefly another co-
homology theory.
The last few sections of Chapter 16 generalize the constructions
relating coverings with actions of groups and group homomorphisms
that were used in Chapters 14 and 15. The added generality, while
not needed in this book, may help to make the constructions more
understandable by putting them in their natural context. In addition,
the exercises (with their hints) carry out proofs of general facts about
these constructions, special cases of which were used in Part VII.
CHAPTER 15

Cohomology

15a. Patching Coverings and Cech Cohomology

Since a G-covering is locally trivial, it can be constructed by patching
together trivial coverings. In this section we specify exactly what data
are needed to carry this out. First, we need to know what this pasting
data looks like.

Lemma 15.1. IfY=X x G~X is the trivial G-covering, and h: X~G

is any locally constant function, then the mapping
x xg I-? X X g . h(x)

from Y to Y is an isomorphism of G-coverings, and every isomor-

phism ofG-coverings has thisformfor a unique locally constantfunc-
tion h.
Proof. It is evident that such a map is continuous and compatible with
left multiplication by G, so it is an isomorphism of G-coverings. Con-
versely, if 'P: Y ~ Y is compatible with left multiplication by G, 'P
must take each point x x g to x X g' h(x) for some h(x) in G. For 'P
to be continuous, the map XI-? h(x) must be locally constant. 0

If p: Y ~ X is any G-covering, one can find a collection

au == {Va: a E.sa} of open sets whose union is X such that the restric-
tion of the covering to each Ua is a trivial G-covering. Choose iso-
morphisms of G-coverings
'Pa: Va X G ~ p-'(U,J.

207
208 15. Cohomology

On the overlaps Ua n U[3 that are not empty we have "transition" iso-
morphisms

given by the restriction of <Pa followed by the restriction of <P[3 -I. These
transition isomorphisms have the form x x g ~ x X g . ga[3(x) for some
(unique) locally constant functions ga[3: Ua n U!,>~ G, by the preced-
ing lemma. The collection of these ga!'> satisfy three properties:

(i) gaa = e (the identity in G) for all n;

(ii) g!'>a = (g",!,»-I for all n, 13; and
(iii) g",!,> = ga!,> • gp-y on Ua n U[3 n U-y for all n, 13, 'Y.

The first two of these are obvious, and the last follows from the equa-
tion <p-y -I 0 <p", = <p-y -I 0 <p[3 0 <P!,> -I 0 <p",. A collection {g",!'>} of locally con-
stant functions 8 satisfying (i)-(iii) is called a Cech cocycle on au with
coefficients in G.
Conversely, given au and a Cech cocycle {ga!'>} on au with coeffi-
cients in G, one can use it as "gluing data" to construct a G-covering,
together with a trivialization over each Ua , so that the resulting co-
cycle is the given one. To do this, take the disjoint union of all the
products U", x G (where G has the discrete topology, i.e., its points
are open), and define an equivalence relation by defining x x g in
Ua x G to be equivalent to x x g' ga!'>(x) in U!'> x G for x in Ua n Up.
Properties (i)-(iii) guarantee precisely that this is an equivalence re-
lation. Define Y to be the set of equivalence classes, with the quotient
topology. Since the equivalence relation is compatible with left mul-
tiplication by G and with the projections to X, the space Y gets an
action of G and a projection p from Y to X so that the resulting maps
Ua x G~ Yare G-maps and compatible with the maps to X. In other
words, this is the patching described in § 14b, using the transition
functions {t!'>"" where {tp",(x x g) = x x g . g",!,>(x).
The cocycle is not uniquely determined by the G-covering, even
with a fixed choice of au, since it depends on the choice of triviali-
zations <p",. But if {<p",'} is another choice of trivializations, using the
lemma again, there is for each n a unique locally constant function

8 Note that if the collection U is chosen so that every Ua n Up is connected, then

these gap are constant, i.e., they are just elements of G. This case will suffice for
our applications, and the reader is invited to make this simplifying assumption from
the start.
15. Patching Coverings and tech Cohomology 209

ha: V a ~ G so that the diagram

q> UxG
a

n-'(U.) ~ 1xxg~xxg·h.(x)
Va xG
commutes. From this it follows that the cocycle {gal3'} for the trivial-
izations 'Pa' is related to that {gal3} for the 'Pa by the equations
gal3' = (ha)-" gal3' h13 •
Two Cech cocycles {gal3} and {gal3'} are said to be cohomologous if
there are locally constant functions ha: V a ~ G such that
gal3' = (ha)-' . gal3' hl3 on Va n VI3 for all n, 13 such that Va n VI3 is
nonempty. This is easily checked to be an equivalence relation, and
the equivalence classes are called tech cohomology classes on OU, with
coefficients in G. The set of these is denoted HI(ou';G).

Exercise 15.2. Verify that the G-coverings constructed from coho-

mologous cocycles are isomorphic.

Exercise 15.3. Let OU be a covering of an annulus by three sets ho-

meomorphic to disks, as shown:

For g in G let c(g) be the cocycle determined by setting gl2 = g23 = e

and g31 = g. (a) Show that every Cech cocycle on OU with coefficients
in G is cohomologous to such a cocycle, and show that c(g) is co-
homologous to c(g') if and only if g and g' are conjugate in G. This
gives a bijection between the set of conjugacy classes in G and H'(OU; G).
(b) If G is cyclic of prime order, show that the coverings correspond-
ing to any two distinct elements that are not the identity in G are
isomorphic as coverings, but not as G-coverings.
210 15. Cohomology

What we have done in this section amounts to setting up a one-to-

one correspondence between HI(au; G) and
{G-coverings of X that are trivial over each V,,}/=.
If the Va are chosen to be simple open sets like disks or rectangles,
or any simply connected and locally path-connected open sets, then
every covering is trivial over them (by Corollary 13.8), so we will
have all G-coverings of X classified by Cech cohomology classes.
Thus we have:

Proposition 15.4. If each Va is simply connected and locally path-

connected, then the set of G-coverings of X, up to isomorphism, is
in one-to-one correspondence with the Cech cohomology set HI(au; G).

Now suppose X is a connected space that has a universal covering.

From Exercise 14.3 we know that the set of G-coverings up to iso-
morphism is in one-to-one correspondence with the set of homomor-
phisms from 'lT1(X,X) to G, up to conjugacy. Putting this together,
and continuing to assume that the open sets in au are simply con-
nected, we have

Corollary 15.5. With these assumptions, there is a bijection

HI(6U; G) +-+ Hom('lT.(X, x), G)/conjugacy.

I5b. Cech Cohomology and Homology

If we want to have a well-defined homomorphism from 'IT.(X,x) to
G, not just up to conjugacy, we know that we must consider coverings
Y~X together with a base point yin Y over x. To do this with the
patching, for those open sets Va that contain x, fix the local trivial-
izations 'Pa: Va X G~p-I(Va) so that 'Pa(x x e) = y. The correspond-
ing cocycle then has gall(x) = e whenever x is in Va n VII; another
cocycle defines an isomorphic covering preserving base points when
it is of the form (ha)-I . gal3 . hl3' with ha: Va~ G locally constant with
the restriction that ha{x) = e whenever x is in Va' Such equivalence
classes form a set we can denote by HI(6U, x; G). This sets up bijec-
tions
HI(6U, x; G) +-+ {G-coverings with base point}/=
+-+ Hom('lTI(X,x), G).
I5h. Cech Cohomology and Homology 211

We will need a prescription to make the homomorphism p from

1TI(X,X) to G explicit, when the covering is given as above by a co-
cycle. As one should expect, the answer is the product of values of
the transition functions as one moves around a path, just as the wind-
ing number is the sum of changes in angle. If 'Y is a loop at x, sub-
divide the unit interval, 0 = to < tl < ... < tn = 1 such that 'Y maps
each subinterval [ti-I. t;] into one of the open sets Ua(i)' Let Xi = 'Y(tJ.

Lemma 15.6. The homomorphism p: 7TI(X,X)~G corresponding to

the G-covering p: Y ~ X takes the homotopy class of 'Y to

p(['Y)) = hi' h2 • •••• hn- I . hn'

where hi = ga(l)a(Hl)(Xi) for 1::5 i::5 n - 1, and hn = ga(n)a(l)(Xn)'
Proof. Let'Y be the lift of'Y starting at y. We must compute 'Y(ti) for
each i. At the start, )1(0) = Y = 'Pa(llx x e). For 0::5 t::5 t l , by conti-
nuity, 'Y(t) = 'Pa(l)('Y(t) x e). So
'Y(tl) = 'Pa(I)(XI x e) = 'Pa(2ixl X e' ga(l)a(2)(XI» = 'Pa(2)(XI X hi)'

Going along each piece of the path in the same way, we find
'Y(tJ = 'Pa(O(Xi X hi . . . . . hi-I) = 'Pa(Hl)(Xi x hi . . . . . hJ.
At the end, this gives
'Y(1) = 'Pa(l)(x x hi . . . . . hn) = (hi' . . . . hn)' 'Pa(I)(X x e)
= (hi' . . . . hn ) • y .
Since 'Y(1) = p([-y)) . y by definition, the lemma follows. D

Now we specialize to the case where G is an abelian group. In this

case any homomorphism from 1TI(X, x) to G must send any commu-
tator aba-Ib- I in 1TI(X,X) to the identity of G, so it must define a
212 15. Cohomology

homomorphism on the abelian quotient group 'IT I(X, X)abel' By Prop-

osition 12.22, we have an isomorphism 'lT1(X,X)abel=HIX. And con-
jugate homomorphisms to an abelian group must be equal. Putting all
this together, and assuming X and 0l1, are as in Proposition 15.4, we
have:

Corollary 15.7 (Hurewicz). If G is abelian, there are canonical bi-

jections
HI(Ol1,; G) ~ Hom('lTI(X, x), G) ~ {G-coverings ofX}/=
~ Hom('lTI(X,x)/['lTI(X, x), 'IT I(X, x)] , G)
~ Hom(HIX, G).
For a space X that may not be connected, we have:

Corollary 15.8. If X is a locally simply connected space, and 0l1, is

a collection of simply connected open sets whose union is X, then
there are canonical bijections
HI (0l1,; G) ~ {G-coverings of X} / = ~ Hom(H IX, G) .
Proof. To give a G-covering of X is the same as giving a G-covering
of each connected component of X, and to give a homomorphism
from HIX to G is the same as giving a homomorphism from the first
homology group of each connected component of X to G. So the re-
sult follows from Corollary 15.7. 0

If G is abelian, the Cech cocycles on 0l1, with coefficients in G form

an abelian group, by multiplication: {gal3}' {gal3'} = {gal3}' {gal3'}. Call
a cocycle {gal3} a coboundary if there are ha: Ua ~ G for each a such
that gal3 = ha -I. hfl.

Exercise 15.9. Assume that G is abelian. Show that the coboundaries

form a subgroup of the cocycles, and that HI(Ol1,; G) is the quotient
group of cocycles modulo coboundaries. (Caution: If G is not abe-
lian, HI(Ol1,; G) has no natural group structure, see Exercise 15.3.) If
G is abelian, give Hom(H, G) the structure of an abelian group, for
any group H. Show that the bijections of Corollary 15.7 are iso-
morphisms of abelian groups.

Exercise 15.10. Define the Oth Cech group Jf(Ol1,; G) for an open
covering 0l1, of a space X with coefficients in a group by defining a
class (or cocycle-there is no equivalence relation) to be a collection
of locally constant functions ga: Ua~ G such that go. = gl3 on Ua n U13 •
15c. De Rham Cohomology and Homology 213

Show that !f(at!; G) is the direct product of copies of G, one for each
connected component of X. In particular, for X open in the plane,
!f(at!;~) is isomorphic to the space !f(X) of locally constant func-
tions.

15c. De Rham Cohomology and Homology

We want to compare the De Rham group HIX with the first homology
group H1X, when X is an open set in the plane. If W is a closed I-form,
and "I a closed I-chain, we defined the integral I-yw in Chapter 9. For
fixed w, this map "I ~ !-v W is a homomorphism from the group ZIX
of closed I-chains to II\t Proposition 9. 11 says that this integral is the
same for homologous I-chains. In other words, the map vanishes on
I-boundaries B1X, and therefore defines a homomorphism from H1X
to ~. Let us denote this homomorphism, at least temporarily, by 'Pw,
so that 'Pw(['Y]) = I-yw.
The set Hom(HIX,~) of homomorphisms from H1X to ~ has a nat-
ural structure of vector space: the sum 'P + 1\1 of two homomorphisms
is defined by ('P + 1\1)(["1]) = 'P(b]) + l\I(b]), and multiplication of 'P
by a scalar r by (r· 'P)(b]) = r· 'P(b]). (Note that this works for any
group in place of HIX.) The above map that assigns 'Pw to the closed
1-form W is a linear mapping of vector spaces
{closed I-forms on X} ~ Hom(HIX,~).
This follows from the equation !-v(rlwl + rzwz) = I-yrlWl + I-yrzwz. If
the I-form w is exact, the homomorphism 'Pw is zero. In fact, if w = dj,
and "I is any I-chain, with boundary 0"1 = ~m;Pj, then

It follows that the above mapping vanishes on the subspace of exact

I-forms, so it defines a linear map on the quotient space HIX. That
is, we have a natural linear map of vector spaces

If ~ EHIX and a EHtX, we may write simply Ia~ in place of I-yw,

where w is a representative of ~ and "I a representative of a. The goal
of this section is to show that this mapping is always an isomorphism.
214 15. Cohomology

Theorem 15.11.lfX is an open set in the plane, then the canonical

homomorphism

is an isomorphism.

The fact that this map is one-to-one is not hard to see, for if a closed
I-form w has all integrals f'Yw vanishing for all closed paths 'Y on X,
we know from Proposition 1. 8 that w is exact. The fact that every
homomorphism comes from a I-form, however, will take some work.
As a warm-up you may consider the special case:

Exercise 15.12. Show that the homomorphism HIX ~ Hom(HIX, ~)

is an isomorphism when X is multiply connected.

Exercise 15.13. Show that, if V is an open subset of X, the diagram

HIX ~ BIV

t t
Hom(H,X, IR) ---+ Hom(H, V, IR)
commutes, where the map on the bottom is determined by the map
from HIV to HIX,

In order to prove the theorem, we specialize the results of § I5b to

the case where G = ~ is the additive group of real numbers, and X
is an open set in the plane. Let au = {Va: IX E~} be a collection of
open rectangles whose union is X. We may find such a collection so
that any point in X is contained in only finitely many Va (see Lemma
A.20 and Lemma 24.10). By Corollary 15.8 we have a bijection be-
tween Hom(HIX, ~) and HI(au; ~). To prove the theorem we will con-
struct a map from HI(au;~) to the De Rham group HIX, and then
show that all these maps are compatible and isomorphisms.
An element of HI(au;~) is determined by a tech cocycle {ga~},
where the ga~ are locally constant functions on Va n V~. We want to
produce from this a closed I-form w on X, well defined up to the
addition of an exact I-form. If we can find some «6" functions fa on
Va so thatfa - f~ = ga~ on Va n V~, then dfa = dh on the overlaps, so
there is a unique I-form won V that is dfa on each Va. This will be
the I-form we are after. The existence of such functions fa follows
from a general lemma:

Lemma 15.14. Let {fafJ} be a collection of «6" junctions,jafJ on Va n V fJ ,

15c. De Rham Cohomology and Homology 215

satisfying the cocycle conditions: (i) faa = 0; (ii) f{3a = -fa{3; and (iii)
fay = fa{3 +f{3y on Va n V{3 n Vr Then there are C(J,'" functions fa on Va,
for all a, such that
fa - fr. = far. on Va n Vr..
Proof. In order to solve these equations fa - fr. = far., we use the ex-
istence of a partition of unity subordinate to the covering OJL (Appen-
dix B2). This says there are C(J,'" functions 'Pa on X, with the closure
of the support of 'Pa (in X) contained in Va, with only finitely many
'Pa nonzero in a neighborhood of any point, and with La 'Pa == 1. For
each (l and ~ define a C(J,'" function hOor. on Va by the formula

h r. = {'Pr. "far. on Va n Vr.,

a 0 on Va \ Van V r. .

Now set fa = Lr. har.. It is an easy exercise to verify that har. and fa
are C(J,'" functions. To complete the proof, we calculate:

'Y 'Y

= 2:'Y 'Pifar.) = (2:'Y 'P'Y)rar. = 1 ·far. = far.. 0

We must verify that the construction made before the lemma gives
a well-defined map from Hl(OJL;~) to HIX. Suppose first that {fa'} is
another collection of functions, withfa' on Va such thatfa' - Jr.' = gar.
on Van V f3 , giving rise to the I-form w' that is dfa' on Va. The
functions fa' - fa on Va agree on the overlaps, so define a function f
on X; and w' - w = df, so w' and w define the same element of HIX.
We must also show that if {gar.'} is a Cech cocycle that is cohomolo-
gous to {gar.}, then they determine the same class in HIX. We are
given locally constant functions ha on Va such that

Iffa - fr. = gar. on Va n Vr., then (fa - ha) - (fr. - hr.) = gar.' on Va n Vr.'
and since each ha is locally constant, d(fa - ha) = dfa, so they define
the same I-form.
Summarizing what we have so far, we have maps
H1X ~ Hom(HIX, IR)

I
Hl(ll ; IR) ~
t
{IR -coverings} /= .
216 15. Cohomology

We want to show that all of these maps are bijections, and that
going once clockwise around the diagram, starting at any place, is
the identity map on that vector space. We have proved that the top
horizontal map in this diagram is one-to-one. It therefore suffices to
show that a trip around the diagram, starting with an element P in
Hom(H 1X,IIl), takes p to itself. Let {fal3} be a cocycle for the covering
Pp: Yp~X defined from this homomorphism. Use Lemma 15.14 to
write fal3 = fa - fl3' and let w be the I-form that is dfa on U a' It follows
from Proposition 12.22 that H1X is generated by the classes of closed
paths in X. To conclude the proof, it therefore suffices to show that

Lw = p(['y])

when "I is any closed path in X. Both sides depend only on the con-
nected component of X containing the image of "I, so we may assume
X is connected. Subdivide the path as in Lemma 15.6, and let "Ii de-
note the restriction of "I to the ith piece [ti-!' tJ By Lemma 15.6,
p(['YD = L7~lhio where hi =fa(l)a(i+I)(Xi) for 1 ::5 i::5 n, with cr(n + 1) = cr(1).
Therefore,
n n

p(['YD 2:fa(i)(Xi) - fa(i+I)(Xi) = 2:fa(i)(Xi) - fa(i)(Xi-l)

;=1 i~l

iLdfa(i) = i L w =
;=1 -V, 1=1 "V,
i -v
w.

This completes the proof of Theorem 15.11. o

Corollary 15.15. There are canonical bijections
HIX ~ Hom(H 1X,IIl) ~ {ill-coverings}/= ~ Hl(au; Ill).

Exercise 15.16. Verify directly that the maps of this corollary are
linear maps between vector spaces.

The theorem can be used to calculate the De Rham group of more

complicated sets in the plane than multiply connected sets, even when
it may be difficult to write down differential forms explicitly:

Problem 15.17. (a) Show that if X is the complement of the set N

of nonnegative integers, then HIX is isomorphic to the space of all
infinite sequences (aO,al, ... ) of real numbers. In particular, HIX
is infinite dimensional. (b) Compute HIX when X is the complement
of the set (0, I, 112, 1/3, 114, ... ). (c) What if X is the complement
of any infinite discrete set?
15d. Proof of Mayer-Vietoris for De Rham Cohomology 217

15d. Proof of Mayer-Vietoris for

De Rham Cohomology
Now that we have explicitly identified the De Rham cohomology group
with the dual of the homology group, we can finish the proof of the
Mayer-Vietoris theorem stated in §IOd for the De Rham groups. What
was missing is the assertion MV(vi) that the map denoted" -" from
H1UffiHIV to Hl(U n V) is surjective. We also have a linear map

H1UffiHIV ~ Hom(Hl(U)EBHl(V)'~)

that takes a pair (a, 13) to the homomorphism that takes a pair (a, b)
in H1(U)EBH1(V) to faa + fb 13. It is a simple exercise to verify, using
Theorem 15.11, that this map is also an isomorphism.
The homomorphism "-" from HI (U n V) to HI U EB HI V determines
a homomorphism
Hom(Hl(U)EBHl(V)'~) ~ Hom(H1(Un V),~),

and it is a general fact, proved in Appendix C (Lemma C.IO), that,

since the map that determines it is one-to-one, this map is surjective.
Consider the diagram
)0

Hom(H.(U)$H.(V), IR) - Hom(H.(Un V),IR)

where the maps are as just described. We know that each of the ver-
tical maps is an isomorphism, and that the bottom horizontal map is
surjective. It follows that the top horizontal map is also surjective,
which is the assertion to be proved, provided we check that the dia-
gram commutes. But this is entirely straightforward. Given a class
(a,l3) in H1UffiHIV, going either way around the diagram, it goes
to the homomorphism that takes an element 'Y in H1(U n V) to the
number f'Y a - f'Y 13· D
In fact, the whole Mayer-Vietoris sequence in cohomology is dual
to, and can be deduced from, the Mayer-Vietoris sequence in ho-
mology. The Oth De Rham group lfx is dual to the Oth homology
group HQX, as follows. For any function f on X and any O-cycle
~ = "'2:.m j P j on X, define fiX) = "'2:.mJ(PJ. Iff is locally constant, and ~
is a O-boundary, then fiX) = 0, as follows from the fact that f must
take the same values at the two endpoints of any path. This deter-
218 15. Cohomology

mines a linear map

lfx ~ Hom(HoX, ~),

Exercise 15.18. Use the fact that HoX is the free abelian group on
the connected components of X to prove that this map from lfx to
Hom(HoX, ID is an isol!!.orphism. Show that the reduced groups are
also dual: lfx == Hom(HoX, ~).

For simplicity now, for any abelian group A, write A * for Hom(A, ~).
Note that a homomorphism A ~ B determines a linear map B* ~ A *.
A direct sum decomposition A = B E9 C determines a direct sum de-
composition A * = B* E9 C*. The dual to the Mayer-Vietoris homol-
ogy sequence is the sequence of vector spaces:
0 - Ho(UU V)* ~ HoU*$ HoV* ....=..... Ho(Un V)*

L H1(UUV)* ~H1U*$H1V*""=""'Hl(UnV)*- 0
The fact that the homology sequence is exact (Theorem 10.5) implies
that this is an exact sequence of vector spaces (see Appendix C2).
The isomorphisms from the preceding section can be used to map
each term in the Mayer-Vietoris cohomology sequence to the cor-
responding term in the above sequence:
0 - JIO(UuV) ~ HOUmHoV ""=""'H°(UnV) L H1(UUV)

~Hlu*mH1V*""="'" Hl(UnV)*-O.
Problem 15.19. (a) Show that each square in the above diagram com-
mutes. (b) Use this to deduce the exactness of the cohomology se-
quence from the exactness of the homology sequence.
CHAPTER 16

Variations

16a. The Orientation Covering

Every manifold M has a canonical two-sheeted covering p: M~M,
called the orientation covering, whose fiber over a point P is the two
ways to orient M at P. Cech cocycles provide a convenient way to
construct this covering. Let G = {± I} be the group of order two, and
take an open covering au = {U(l} of M to be images of the coordinate
charts 'P(l: V(l~UaCM, with V(l open in IW. The Jacobian determi-
nant of the change of coordinates from V(l to Vi3 has a locally constant
sign, which gives a locally constant function from U a n U i3 to {± I}.
The_chain rule for lacobians implies that this is a cocycle. Define
p: M~M to be the resulting {±l}-covering. An orientation of M can
be define!! to be a section of this covering, i.e., a continuous mapping
IT: M ~ M such that po IT is the identity map on M.

Exercise 16.1. (a) Show that this orientation covering is independent

01 the choice of coordinate charts. (b) If M is connected, show that
M has one or two connected compone1}!.s: one if M is nonorientable,
two if M is orientable. (c) Show that M is always orientable.

Exercise 16.2. When M = !R1P2 show that the orientation covering is

isomorphic to the covering S2 ~ !R1P2. Do the same for !RlPn , n > 2.
Identify the orientation covering for the Klein bottle.

219
220 16. Variations

16b. Coverings from I-Forms

From Corollary 15.7 and Theorem 15.11 we have a canonical bijec-
tion
HIX ~ {~-coverings of X}/=-.
It follows that there is an ~-covering p.,: X., ~ X of an open set X in
the plane corresponding to a closed I-form w on X, with two I-forms
giving isomorphic coverings when their difference is exact. To con-
struct this covering, following the procedure given in § 15c, one takes
a collection au of open sets Ua on each of which w = dla for some
function 10. , and then uses the transition functions gal3 =10. - 113 to patch
together trivial ~-coverings of each Ua into a global ~-covering of
X. In this section we describe a more direct way to construct this
covering.
The idea is to put all graphs of functions I on open subsets of X
such that dl= w together into one big covering space. A good lan-
guage for describing this is that of germs. A germ of a C(6" function
at a point P in X is an equivalence class of C(6" functions defined in
neighborhoods of P. Two functions lIon a neighborhood NI and.li
on N2 are defined to be equivalent if there is a neighborhood N of P
contained in NI nN2 such that/l and/2 are equal on N.

These germs can be added, multiplied, and differentiated, just like

functions on definite open sets.

Exercise 16.3. Verify that this is an equivalence relation, and that

these operations preserve equivalence classes.

A germ at P has a value at P (the value I(P) of any representative

f), but it does not have a value at any other point. The idea is simply
that we only care about functions near the point P, and we allow
ourselves to shrink the neighborhoods arbitrarily. Given a function I
16b. Coverings from I-Forms 221

on a neighborhood of P, the equivalence class containing it is called

the germ defined by f at P.
Since derivatives of germs make sense, it makes sense to say that
the differential of a germ at a point is a given I-form; if the I-form
is w = P dx + q dy, the differential of a germ of f at P is w if
af af
- (P) = p(P) and - (P) = q(P) .
ax ay
If w is a closed I-form on X, define X", to be the set of all germs of
functions at points of X whose differential is w. We will make this
into a topological space, so that the map p", : X", ~ X, that takes a germ
to the point at which it is a germ, is a covering map. For each open
set N in X and functionf on N such that df= won N, define a basic
open set Nf in X",:
Nf = {germs at points of N defined by f} .
A set in X", is defined to be open if it is a union of basic sets Nf .
If N is an open set in X such that w is exact on N (for example,
any open disk or rectangle in X), then PO) -I(N) is a disjoint union of
the open sets Nf , where f runs through all functions on N whose dif-
ferential is w. The point is that if f is one such function, then all
others, on N or on any open subset of N, are of the form f + c for
some locally constant function c. This shows that the covering is triv-
ial over N, and, since any point in X has such neighborhoods, this
shows that we have a covering space:
p",: X",~ X.
In addition, this is an ~-covering. The action of a real number on X",
takes a germ to the sum of the germ and the real number. The pre-
ceding discussion shows that this is an even action of ~ on X"" and
that the covering is an ~-covering.

Exercise 16.4. Verify that when X = ~2, {(O, O)} and w = d{}, then
the polar coordinate covering of X is a connected component of X",.

Choosing any functionf on any connected open subset N of X with

df= won N determines a connected component of X",. Namely, this
is the connected component containing the open set Nf .

Exercise 16.5. Show that this component is the union of all open sets
of the form N'!, with N' a connected open set in X and df' = w on N'
such that there is a chain of connected open sets N = No,
222 16. Variations

N1 , • • • ,NT = N' in X, with functions /; on Ni such that d/; = w, and

/; =/;+1 on Ni n Ni+ l •

In general, there is a natural bijection between X", and the product

X x ~, that takes a germ at P to the pair P x r, where r is the value
of the germ at P. If ~ is has its usual topology, this bijection is con-
tinuous, but it is not a homeomorphism. In the polar coordinate ex-
ample, each connected component of X", maps to a closed submani-
fold of the product X x IR; these are the graphs of the multi valued
functions "-3 + c." In general, however, this need not be the case.
Problem 6.22 shows that, when X is the complement of two points
P and Q, and w is a linear combination r d-3 p + s d-3 p with r / s irra-
tional, the sheets of a component of X", can come arbitrarily close
together, if regarded as a subset of X x IR. With this understanding,
the connected components of X", can be regarded as the multivalued
functions on X whose differential is w-but their graphs need not be
closed in X x ~.

Problem 16.6. (a) Show that this covering POl: X", ~ X is isomorphic
to the covering constructed with cocycles. (b) Show directly that I-forms
that differ by an exact form determine isomorphic coverings.

Problem 16.7. Assume X is connected, and let Y be a connected

component of X",. Show that Y ~ X is a G-covering, where G is the
period module, i.e., G is the subgroup of IR generated by the periods
of w; these periods are the numbers f"'lw, as "I varies over loops at x
(or equivalently, closed I-chains on X).

16c. Another Cohomology Group

In addition to the De Rham and tech cohomology groups, there are
cohomology groups HO(X; G) and HI(X; G) that one can assign to any
topological space X, where G can be any abelian group. Lacking smooth
functions and I-forms to evaluate on points and paths, one makes
these groups directly out of objects that are defined by their values
on points and paths. Define a O-cochain to be an arbitrary function
from X to G. Define a I-cochain to be a function from the set of all
continuous paths on X to G, with the requirement that the function
takes all constant paths to the identity element 0 in G. Note that,
unlike the case of O-chains or I-chains, cochains can assign nonzero
16c. Another Cohomology Group 223

elements of G to arbitrarily many elements. Since the set of functions

from any set to an abelian group G is an abelian group (by adding
the values in G), the O-cochains form a group denoted CO(X;G), and
the l-cochains form a group CI(X;G).
If a: X ~ G is a O-cochain, define the coboundary Sa to be the
l-cochain whose value on a path 'Y is a('Y(1» - a('Y(O». Define the
Oth cohomology group HO(X; G) be the set of O-cochains whose boundary
is zero.

Exercise 16.8. Show that giving an element of YO(X; G) is equivalent

to giving a function from the set of path-connected components of X
to G. Deduce an isomorphism HO(X;G)=Hom(HoX,G).

If c is a l-cochain, it defines a homomorphism from the I-chains

°
CIX to G by the formula C(Lni'Yi) = Lnic('Y). A l-cochain c is called
a l-cocycle if c(af) = for all continuous f: [0,1] x [0, l]~X. The
l-cocycles form a subgroup ZI(X;G) of CI(X;G). Every coboundary
Sa is a l-cocycle, so the l-coboundaries form a subgroup B I(X; G) of
ZI(X; G). The quotient group is the first cohomology group
HI(X;G) = ZI(X;G)/BI(X;G).
There is a natural homomorphism from HI(X; G) to Hom(XIX, G),
that takes the class of a l-cocycle c to the homomorphism that takes
the homology class of a I-cycle 'Y to the element c('Y).

Exercise 16.9. Verify that this is a well-defined homomorphism.

Proposition 16.10. The homomorphism HI(X;G)~Hom(HIX, G) is

an isomorphism.
Proof. To give a homomorphism from HIX = ZIX/BIX to G is equiv-
alent to giving a homomorphism from ZIX to G that vanishes on BIX;
that is,
Hom(HIX, G) = Kemel(Hom(ZIX, G)~ Hom(BIX, G».
We have homomorphisms
BI(X; G) ~ ZI(X; G) ~ Hom(ZIX, G) ~ Hom(BIX, G),
where the middle map takes the cocycle c to the homomorphism
'Y~ c('Y). The assertion of the proposition is easily seen to be equiv-
alent to the assertion that this sequence is exact at the two middle
groups.
The fact that the image of each map is contained in the next is the
224 16. Variations

content of the preceding exercise. To prove the opposite inclusions,

we need to choose arbitrarily a point Xa in each path component Xa
of X, and choose for each point y in X an arbitrary path 'Ty starting at
the chosen point Xa of the component containing y and ending at y.
Suppose c in ZI(X; G) maps to the zero homomorphism from ZIX
to G. We want to construct a O-chain whose coboundary is c. Define
the O-chain a by the formula a(y) = c('Ty). We claim that (8a)('Y) = c('Y)
for all paths 'Y. In fact,
(8a)('Y) = a('Y(l» - a('Y(O» = c('T-y(I) - c('T-y(O»'
and c('T'Y(I) - c('T'Y(O» = c('Y) since 'T'Y(I) - 'T'Y(O) - 'Y is a I-cycle, and c is
assumed to vanish on I-cycles.
To finish the proof, we must show that if f: ZIX ~ G is a homo-
morphism that vanishes on B IX, then f comes from some I-cocycle
c. Define the cochain c by the formula
c('Y) = f('Y + 'T'Y(O) - 'T'Y(I»
for any path 'Y. Equivalently, for any I-chain 'Y, c('Y) = f('Y - "L.m;'Ty),
where "L.m;y; = iJ('Y). In particular, if'Y is a I-cycle, c('Y) = f('Y) , and
if'Y=iJr, c(iJf) =f(iJf) =0. So c is a I-cocycle that maps tof. 0

This proposition implies that HI(X; G) is isomorphic to the Cech

group HI (au; G), provided au is a suitable cover of a nice space, so
that Corollary 15.7 applies.

Exercise 16.11. When X is an open set in the plane, and G = IR,

construct homomorphisms HOX ~ HO(X; IR) and HIX ~ HI(X; IR) from
the De Rham groups to these cohomology groups, and show that they
are isomorphisms.

There is also a Mayer-Vietoris theorem for these cohomology groups.

If U' is an open subset of U there are natural restriction maps from
HO(U; G) to HO(U'; G) and from HI(U; G) to HI(U'; G). If X is a union
of two open sets U and V, there is a coboundary map
8: H°(UnV;G) ~ HI(UUV;G).
To define this, given a O-cocycle a on un
V, extend a to a O-cochain
ii on all of U U V by defining ii to be zero on all points not in V. un
Then define the I-cocycle 8(a) on U U V whose value on a path 'Y is
obtained by writing 'Y = 'Y I + 'Y2 + 'T, where 'Y I is a I-chain on U, 'Y2 a
I-chain on V, and'T is a I-boundary (see the proof of Lemma 10.2),
and setting 8(a)('Y) = ii(iJ'YI).
16d. G-Sets and Coverings 225

Problem 16.12. Show that this definition is independent of choices,

and verify that the resulting Mayer-Vietoris sequence
o ~ JIO(UuV;G) ~ HO(U;G)$HO(V;G) ~ HO(Un V;G)

-4 Hl(UU V;G) ~ Hl(U;G) $ Hl(V;G) ~ Hl(Un V;G)

is exact.

16d. G-Sets and Coverings

In this section and the next we describe two general constructions,
which may help to put the constructions of § 14a in context. The ex-
ercises verify the assertions made about these constructions.
Let p: Y ~ X = Y/G be a G-covering, without base point for the
moment. Suppose T is any set, and we have a left action of G on T;
we say that T is a G-set. Give T the discrete topology, so an action
of G on T is a mapping from G x T to T satisfying properties (1) and
(2) of § llc. The group G acts on the left on Y X T by the formula
g. (y x t) = g' y x g. t. Define YT to be the space of orbits:

YT = (Yx T)/G.
Write (y x t) in YT for the orbit containing y x t. Let PT: YT~ X be
the mapping that sends (y x t) to p(y).

Exercise 16.13. Show that the mappingpT: YT~X is a covering map.

If T and T' are sets with G-actions, a map of G-sets is a function

ql:T ~ T' such that ql(g' t) = g . ql(t) for all t in T and g in G. A map
of G-sets is an isomorphism if it is bijective, so there is an inverse
mapping of G-sets from T' to T. A map ql: T ~ T' of G-sets deter-
mines a continuous mapping from Y x T to Y X T', taking y x t to
y x ql(t). This is compatible with the actions of G, so it determines a
continuous mapping from YT to YT', taking (y x t) to (y x ql(t», which
commutes with the projections to X. If ql is an isomorphism, this is
an isomorphism of coverings. Conversely, we have:

Exercise 16.14. Let p: Y~X = Y/G be a G-covering, with Y con-

nected. Show that the two G-sets determine isomorphic coverings of
X if and only if the G-sets are isomorphic. Show in fact that any
continuous mapping f from YT to YT' commuting with the projections
to X comes from a map of G-sets from T to T'.
226 16. Variations

Exercise 16.15. If T ~ T' and T' ~ T" are maps of G-sets, show that
the composite of YT~ YT, and YT'~ YT" is the mapping detennined by
the composite T~T".

Suppose the action of G on a set T is transitive: for any tl and t2

in T, there is some g in G such that g' tl = t2. If a point t is chosen,
let He G be the subgroup of elements of G that fix t, i.e.,
H = {g E G: g' t = t}. Then the G-set G IH of left cosets is isomorphic
to the G-set T, by mapping the coset gH containing g to the point
g' t.

Exercise 16.16. Show that two transitive G-sets are isomorphic if and
only if the corresponding subgroups of G are conjugate.

In Exercise 11.27 we saw how a subgroup H of G detennines a

covering YIH~X.

Exercise 16.17. If T = G IH is a set of left cosets, show that the cov-

ering YT~X is isomorphic to the covering YIH~X.

Exercise 16.18. If T is a disjoint union of G-sets Ta , show that the

covering YT~ X is a disjoint union of the coverings YT. ~ X.

An arbitrary G-set T is a disjoint union of its orbits Ta. By the

preceding exercises the corresponding covering YT~ X is a disjoint
union of coverings of the fonn Y IHa, for Ha subgroups of G.
Now suppose X is conneSled and locally path-connec~d, and has
a universal covering space X ~ X. Choose a point £ of X lying over
x in X. We have seen that, with these choices, the universal covering
is a 'lT1(X,x)-covering. So everx action of the fundamental group on
a set T detennines a covering XT~X, Combining what we have just
proved with Proposition 13.23, we have:

Proposition 16.19. Every cOJ.,ering of X is isomorphic to one obtained

from the universal covering X ~ X by a left action of 1T1(X,X) on some
set T. This covering is connected if and only if the action on T is
transitive. Two such coverings are isomorphic if and only if the 1T1(X,X)-
sets are isomorphic.

A left action of a group G on a set T is the same as a homomor-

phism of G to the group Aut(T) of pennutations of T. Two such
homomorphisms give isomorphic G-sets exactly when the homo-
16e. Coverings and Group Homomorphisms 227

morphisms are conjugate. In particular, taking G = 1l'1(X,X) and

T= {l, ... , n}, so Aut(T) is the symmetric group @S. on n letters,
we have a canonical bijection
in-sheeted coverings of Xl/=. ~ Hom(1l'1(X,x),@Sn)/conjugacy.

16e. Coverings and Group Homomorphisms

Suppose p: Y~X is a G-covering, and $: G~G' is any homomor-
phism of groups. If we make G' into a left G-set, the construction of
the preceding section can be used to construct a covering of X that is
locally a product X x G' ~ X. We want to do this in such a way that
this covering can be made into a G' -covering. The simplest way to
make G' into a left G-set is by defining, for g in G and g' in G',
g . g' to be $(g) . g', the latter using the group product in G'. How-
ever, we will use another, which will allow us to define a compatible
left action of G' on the result. Define the left action of G on G' by
g' g' = g' '$(g-I) = g' '$(g)-I for gEG and g' E G' .
It is straightforward to check that this is a left action of G on G'.
Now define p($): Y($)~X to be the covering constructed from this
left G-set. That is, Y($) is the quotient of Y x G' by the left action
of G by g . (y x g') = (g. y x g' '$(g -I), with projection from Y($) to
X determined by p on the first factor. Make G' act on Y($) by
g' . (y x h') = (y x g' . h').

Exercise 16.20. Show that Y($)~X is a G'-covering of X. If Y has

a base point y, Y($) gets the base point (y x e). When p is the uni-
versal covering of X, this is the construction of Proposition 14.1.

Exercise 16.21. If Yp ~ X is the G-covering determined by a ho-

momorphism p: 11'1 (X ,x) ~ G, and $: G ~ G' is a homomorphism, show
that the G'-covering Yp($)~X constructed from Yp~X and $ is iso-
morphic to the G' -covering YIJJop ~ X.

Exercise 16.22. Let p: Y ~ X be a G-covering, with Y connected. Let

$I:G~G' and $2:G~G' be two homomorphisms. Show that, if
base points are taken into account, the coverings Y($I) and Y($2) are
isomorphic if and only if $1 and $2 are equal. Without base points,
they are isomorphic if and only if $1 and $2 are conjugate, i.e., there
is some g' in G' such that $2(g) = g' '$I(g)' (g,)-I for all g in G.
228 16. Variations

Exercise 16.23. Show that if G is abelian, anx G-covering of X is

obtained from the universal abelian covering X abel ~ X by a unique
homomorphism 1\1: H,X~G.

16f. G-Coverings and Cocyc1es

In § 16d we constructed from a G-covering p: Y ~ X and a left action
of G on a set T, a new covering PT: YT~X, where YT is the quotient
of (Y x T)/G, with G acting simultaneously on the left on Yand on
T. This can also be described by a cocycle. Take a cocycle {ga!3} of
transition functions from a trivialization of the covering Y ~ X over
OJL = {Ua}, and use them to glue together the disjoint union of the
Va X T, identifying x X t in Va X T with x X g!3a(x)· tin V!3 X T if x is
in van V!3.

Exercise 16.24. Verify that this covering defined by a cocycle is iso-

morphic to that defined in § 16d.

If 1\1: G~ G' is a homomorphism of groups, then 1\1 determines a

map
1\1*: H'(6U; G) ~ H'C6U; G')

that takes a class represented by a cocycle {ga!3} to the class of {1\I(ga!3)}.

(Exercise: Verify that this is well defined.) This means that any G-
covering of X that is trivial over each Va determines a G' -covering
of X, also trivial over each Va: if the G-covering is constructed by
gluing the products Vax G with the transition functions
x X g ~ x X g . ga!3(x) , then the corresponding G' -covering is con-
structed by gluing the products Va X G' with the transition functions
x X g' ~x X g'. l\I(ga!3(x».

Problem 16.25. Show that this construction of a G' -covering from a

G-covering agrees with the construction in § 16d as the quotient of
Y X G' by the action of G: g. (y X g') = g. y X g'. l\I(g-I): In partic-
ular, it is independent of choice of OJL.

Exercise 16.26. A homomorphism 1\1: G~ G' determines a map from

Hom(1TI(X,X), G) to Hom(1TI(X,X), G'), taking p to 1\10 p. Show that
this is compatible with the bijections of Proposition 14.1.
16f. G-Coverings and Cocyc1es 229

In our definition of Cech cohomology Hl(UU; G) there has been no

topology on the group G, or, more accurately, we have equipped G
with the discrete topology. Cech cohomology is also important when
G is a group with a more interesting topology. One can define a set
Hl(UU; G) the same way, requiring the functions gal3 and ha in the def-
inition to be continuous maps from V 0.13 and Va to G, respectively.
From a cocycle {fal3} one constructs in the same way a (principal) G-
bundle, i.e., a space Y, on which G acts continuously, with a mapping
p: Y ~ X so that p(g' Y) = p(y), locally isomorphic to the projection
X x G~ X (but in this product G is a topological space, not neces-
sarily discrete).
If the topological group G acts continuously on the left on a space
V, then a G-bundle determines a bundle with fiber V. The construction
is as in the discrete case: take trivial coverings Va X V, and glue them
together using the identifications from Va X V to VI3 X V over Va n VI3
by X X v~x X gl3a(X)' v; or one may construct the quotient (Y X V)/G.
The fundamental example is the case G = GLnIR (with its usual to-
pology as an open subset of IRn 2), with V = IRn, and the resulting bun-
dle is called a vector bundle of rank n.

Problem 16.27. For n = I, and UU a covering of a plane open set V

as above, show that the homomorphism from Hl(UU; {± I}) to
Hl(UU; GL 1 IR) determined by the inclusion of {± I} in GL 1 IR is an iso-
morphism. Equivalently, giving a line bundle is equivalent to giving
a two-sheeted covering. Prove this directly. Compute this group for
V an annulus. Over a circle in the annulus you should find two line
bundles, corresponding to a cylinder and a Moebius band.

Problem 16.28. Show how to make the set of all tangent vectors to
the sphere S2 C 1R3 into a vector bundle of rank 2. Generalize to con-
struct the tangent bundle of an arbitrary manifold.
 PART IX

TOPOLOGY OF SURFACES

The goal is to extend what we have done for open sets in the plane
to open sets in surfaces, especially compact orientable surfaces. Some
of this is straightforward generalization, valid for any surface, given
basic notions about coordinate charts, but there are some new features
that are special to the compact and orientable case.
We show that every compact orientable surface is homeomorphic
to a sphere with g handles, at least under the assumption that the
surface can be triangulated. Any such surface can be realized by tak-
ing a convex polygon with 4g sides, and making suitable identifica-
tions on the boundary. From this description it is routine to compute
the fundamental group and first homology group.
In Chapter 18 we show that for any differentiable surface X, there
is a canonical isomorphism between the De Rham group HIX of closed
mod exact I-forms, and the dual group Hom(H1X, IR) to the homology
group. This leads to a definition and calculation of an intersection
pairing on the homology group HIX.
Note: A surface is assumed to be connected unless otherwise stated.
See Appendix D for foundational facts about surfaces.
CHAPTER 17

The Topology of Surfaces

17a. Triangulation and Polygons

with Sides Identified

Our aim in this section is to show that a compact orientable surface

is homeomorphic to a sphere with handles, under the assumption that
the surface can be triangulated. The idea is to "flatten" it out, to
realize the surface as a polygon in the plane, with certain identifi-
cations on the boundary edges, and then, by cutting and pasting, to
simplify these realizations until we can recognize the result.
A triangulation of a compact surface X has a finite number f of
faces, each of which is a subset of X homeomorphic to an ordinary
(closed) triangle in the plane, with three edges homeomorphic to closed
intervals, and three vertices that are points. Two edges can meet only
at a vertex, and two faces meet only at one vertex or exactly along
a common edge; in the latter case the faces are called adjacent. There
are many choices for such homeomorphisms between plane triangles
and faces of the triangulation. In order to be able to compare the
homeomorphisms on adjacent faces in X, we first choose a homeo-
morphism of each edge on X with the interval [0,1]. We can then
find homeomorphisms of each face on X with a plane triangle so that,
on the edges, the maps are those determined by these homeomorphisms:

233
234 17. The Topology of Surfaces

/1-+ (l-t)P + tQ

o )

If two triangles correspond to adjacent faces, this means that the re-
sulting homeomorphism between corresponding sides PQ and P' Q'
is given by the "affine" map (1- t)P + tQ~(1- t)P' + tQ'. When
we identify sides of polygons, it will always be by such affine homeo-
morphisms, so there will be no ambiguity in the results.
An orientation of X determines a counterclockwise direction around
the boundary of each triangle, with adjacent triangles determining op-
posite directions along their common edge:

s·

Choose one face 'T! in the triangulation, and choose a homeomor-

phism of'T! with a plane triangle II! == T!, with counterclockwise ori-
entation. Choose a face 'T2 adjacent to 'T!, and extend the homeomor-
phism along their common edge to a homeomorphism of'T2 with a
triangle T2 adjacent to T!:

p ....-----..."

s
17a. Triangulation and Polygons with Sides Identified 235

Choose T2 so the union of TI and T2 is a convex quadrilateral II2.

This gives a homeomorphism from 'TI U'T2 to II 2. Now choose 'T3 ad-
jacent to 'TI or 'T2 (if 'T3 is adjacent to both, choose one of them ar-
bitrarily), and extend the map from the common edge of'T3 and the
(chosen) 'TI or'T2 to a homeomorphism from 'T3 to a triangle T3 adjacent
to the corresponding side of II 2. Let II3 be the resulting five-sided
polygon, which we can take to be convex:

The arrows indicate the identification made in the map from the
figure on the right onto that on the left. Continue in this way, until
all f of the faces have been used. We then have a convex polygon
III = II with f + 2 sides. Each side of II will correspond to a common
edge of two faces on X, and each edge occurring this way will cor-
respond to two such sides of II. Thus the sides of II will be paired
off, and we have a continuous map from II to X that realizes X as a
quotient space of II, obtained by identifying corresponding points on
corresponding sides. Note that, when traveling around II in a coun-
terclockwise direction, one travels along two corresponding sides in
opposite directions.
This realization of X as an identification space of a polygon is half
way to our goal. The data used in constructing X from the polygon
can be encoded by taking an alphabet with m letters, with m = 1f2(f + 2),
and writing down a sequence of 2m symbols, each of which is a letter
a in the alphabet, or its "inverse" a -I, with each of these symbols
occurring just once. For example, the space constructed by making
the identifications indicated in the following diagram
236 17. The Topology of Surfaces

can be described by the code ala2a 3a 2 -l al -I a3 -I.

Exercise 17.1. Show that this code represents a torus.

Exercise 17.2. If the number of sides is four, show that X is homeo-
morphic to a sphere or a torus.
There is a good deal of arbitrariness in the choice of polygon and
its code, even for a given triangulation and ordered choice of faces
of X. The polygon can certainly be replaced by any other convex
polygon with the same number of sides. (For simplicity, one may use
a regular polygon.) Of course, even when the polygon and the iden-
tifications are fixed, the choice of alphabet is arbitrary, as is the choice
of which of a pair of corresponding sides is written as a letter a of
the alphabet and which as a -\. In addition, the place where one starts
listing the sides is arbitrary, and the code can therefore by cyclically
permuted. Note that the identifications are made by identifying cor-
responding points on corresponding sides; some vertices of IT may
become identified by this, and others may not.

17b. Classification of Compact Oriented Surfaces

For the rest of the argument we do not need the triangulation, but
only this description as a polygon with sides identified according to
a code. We want to prescribe some rules for simplifying a polygonal
presentation of a surface. We proceed in several steps. By Exercise
17.2 we can assume the polygon has 2m sides, with m~3.
Step 1. If a letter a and its inverse a -I occur successively in the code,
they can both be omitted. That is, X is homeomorphic to the surface
obtained from a convex polygon with 2m - 2 sides, using the same
code in the same order but with a and a-I omitted. This can be seen
from the pictures
17b. Classification of Compact Oriented Surfaces 237

y
x
---~x

The point is that if one first carries out the identification of a with
a -I, one gets a space homeomorphic to a convex polygon, so that the
remaining identifications are prescribed by the code with a and a-I
omitted.
Step 2. It can be assumed that all the vertices of the polygon map to
the same point of X. To show this, it suffices, if not all the vertices
have the same image, to show how to increase by one the number of
vertices mapping to a given point x, without changing the total num-
ber of vertices. There is a side, say labeled a, that joins two vertices,
one of which is mapped to x and the other to a point y not equal to
x. Let ~ be the side of II adjacent to a at the point mapping to y.
Draw the diagonal -y as shown on the polygon, and cut off the triangle
formed by a, ~, and -y, and attach it to the polygon by identifying ~
with ~-I:

~-I
y

The surface X is the identification space of this new polygon, and it

has the same number of sides, but one more vertex mapping to x.
Step 3. By steps 1 and 2 we may assume that all vertices map to the
same point x in X, and that no edge is adjacent to its inverse. We
claim next that for any edge a there must be an edge ~ lying between
a and a -I so that ~ -I lies between a -I and a; that is, in a code, after
cyclically permuting if necessary, they occur in the order
Q. -I Q.-I
... a ... t-' ••• a ... t-' •••

If not, we could construct the identification space X by first identi-

 238 17. The Topology of Surfaces

fying (X with (X -I, and then separately doing all the identification pre-
scribed by edges lying (counterclockwise) between (X and (X - \ and
those prescribed by edges lying (counterclockwise) between (X -I and
(X. The two endpoints of (X never get identified in this process, con-

tradicting the assumption that all the vertices map to the same point
of x.
Now we simplify the code as follows: Choose an edge (x, and take
an edge f3 so that the edges (x, f3, (X - I, and f3 -I occur in this order,
but perhaps with other edges in between some of these (denoted A,
B, C, and D in the following diagram). Perform the following se-
quence of cutting and pasting moves:

~-I

A
D

I)-I
y
/ c

c
I)-I

A
,-I
17b. Classification of Compact Oriented Surfaces 239

This gives a new polygon with the same properties and the same num-
ber of vertices, which also represents X, but now it has a successive
sequence 8· -y' 8- 1 • -y-I. In addition, if other such sequences occurred
elsewhere in the polygon (i.e., in one of the parts labeled A, B, C,
or D), such sequences are not disturbed by this procedure. So we may
continue this process, until we have represented X as a polygon with
sides identified according to a code

(17.3)
f.I. -I f.I. -I f.I. -I f.I. -I f.I. -I f.I.- 1
<XI • I-'I • <XI • I-'I • <X2 • I-'2 • <X2 • I-'2 • • • • • <Xg ' I-'g' <Xg • I-'g •

~I

This is called a normal form. From this one can see directly that the
identification space is a sphere with g handles. First make the iden-
tification of each ~i with ~i-I:
240 17. The Topology of Surfaces

or
17b. Classification of Compact Oriented Surfaces 241

Then identify each ai with ai-I:

Theorem 17.4. If X is a (triangulable) compact orientable surface.

then X is homeomorphic to a sphere with g handles for some non-
negative integer g.

A similar procedure leads to a normal form for nonorientable com-

pact surfaces. The projective plane can be realized by the code a' a.
identifying the opposite sides of a "two-sided" polygon

a a

Problem 17.5. Show that a (triangulable) compact nonorientable sur-

242 17. The Topology of Surfaces

face has a normal form given by a code 0.1' 0.1 • 0.2 • 0.2' . • • • o.h· o.h,
for some positive integer h.

For a discussion of triangulation of arbitrary compact surfaces, to-

gether with a more modern proof of Theorem 17.4, see Armstrong
(1983).

17c. The Fundamental Group of a Surface

The normal form representation of a compact orientable surface X as
a 4g-sided polygon IT with sides identified according to the code (17.3)
can be used to compute the fundamental group of X. Let x be the
point in X that is the image of the vertices in IT, and let o.i and l3i be
the loops in X that are the images of the corresponding sides of IT.
Let F 2g be the free group on 2g generators ai' b l , • • • , a g , bg •
There is a homomorphism of groups from F2g to the fundamental group
'lT1(X,X) that takes ai to o.i and bi to l3i for l:5i:5g. We claim first
that the element
cg = al' b I' al -I • b I -I • a2' b2' a2 -I • b2 -I • • . • • a g ' bg' a g -I • b-
g
I

in F 2g maps to the identity element of 'IT I (X, x) by this homomorphism.

In fact, the product path
R -I R -I R -I R -I R -I R-I
0.1 • PI • 0.1 • PI • 0.2 • P2 • 0.2 • P2 • • . . • o.g' pg' o.g • Pg

is homotopic to the constant path at x, since the corresponding path

around the sides of IT is homotopic to a constant path in IT (since IT
is convex), and composition with the continuous map from IT to X
gives a homotopy in X.
Let Ng be the least normal subgroup of F2g containing cg , i.e., Ng
is the subgroup generated by all elements of the form U' cg ' u- I for
all u in F 2g • From what we have just seen, Ng is in the kernel of the
homomorphism from F 2g to 'lT1(X,X), so we have a homomorphism
F 2g /Ng - 'lTI(X,X).

Proposition 17.6. This homomorphism F 2g /Ng -'lTI(X,x) is an iso-

morphism.
Proof. We will apply the Van Kampen theorem. Let U be the image
in X of the complement of a small disk in the middle of IT, and let
V be the image in X of an open set that contains this disk as shown:
17c. The Fundamental Group of a Surface 243

Let K be the image of the boundary of II in X. This K is a graph

that consists of 2g loops at x, so 'IT,(K, x) is the free group on gen-
erators ai, b l , • • • , ag , bg , i.e., 'IT,(K, x) = F 2g • By radial projection
from the center of the disk, one sees that K is a deformation retract
of U. We therefore have an isomorphism
'ITI(U, x) =:; 'ITI(K, x),

so 'ITI(U,X) is also the free group F 2g on the same generators. Now V

is homeomorphic to a disk, so 'ITI(V,X) = kt} is trivial, and un V has
a circle for a deformation retract, so 'ITI(U n V,x) = 7L The inclusion
of un V in U takes a generator of 'IT I(U n V, x) to the element
c g =al' b l ·al- I • . . . ·bg - ' • By Van Kampen's theorem, for any group

G, to give a homomorphism from 'ITI(X, x) to G is the same as giving

a homomorphism from 'IT I(U, x) = F2g to G (and a homomorphism from
'ITI(V,X) = {e} to G) in such a way that the composite

7L = 'IT1(Un V,x) - 'ITI(U,X) - G

takes a generator of 7L to the identity element of G. This means pre-
cisely that one has a homomorphism from F 2g to G such that cg maps
to the identity, or equivalently, that N g maps to the identity. That is,
Hom('IT,(X,x), G) = Hom(F2g /Ng , G).

This implies that the map from F 2g /Ng to 'ITI(X,X) is an isomorphism,

cf. Exercise 14.5. (See also Exercise 14.8, which applies to this sit-
uation directly.) 0

Corollary 17.7. Thefirst homology group HIX is afree abelian group

of rank 2g, with basis the image of the loops ai, f31, •.• , a g , ~g.
Proof. Let A 2g be the free abelian group on 2g generators. The ca-
nonical map from F 2g to A 2g maps N g to 0, since cg is in the com-
mutator subgroup. This determines a homomorphism from
244 17. The Topology of Surfaces

'Tl'1(X,X) =F2g /Ng to A 2g , and therefore a homomorphism

H1(X) = 'Tl'1(X,X)/['Tl'}(X,X),'Tl'I(X,X)] ~ A 2g •
Since the images of a}, b 1, . . . , a g , bg generate H1(X), and their
images in A 2g are linearly independent, it follows that they are linearly
independent in H}(X) and that this map is an isomorphism. 0

In particular, since the rank of a free abelian group is an invariant

of the group, this shows that the number g of Theorem 17.4 is in-
dependent of choices, and depends only on the topology of X. It is
called the genus of the surface. Together with what we saw in Chapter
8, this shows that for any triangulation of X, the number v of vertices,
e of edges, and! of faces satisfies the Euler equation
v - e +! = 2 - 2g .

Problem 17.8. For an nonorientable compact surface with normal

form al . a} . a2 . a2' . . . . ah . ah as in Problem 17.5, show that the
fundamental group is the quotient of the free group Fh on h generators
a}, . . . , ah by the least normal subgroup that contains the element
a}2. a22. .... a/. Show that the first homology group is isomorphic
to 7lJJ(h-l)tf) (71./271.). In particular, the number h is independent of all
choices.

Exercise 17.9. Prove that for any triangulation of a compact non-

orientable surface, with h as above, the number v of vertices, e of
edges, and! of faces satisfies the Euler equation
v-e+! = 2-h.
For homology there is no need to use paths with the same base points.
It will be convenient to change the paths ai and ~i to paths that we
denote by ai and bi as shown:

Exercise 17.10. Show that the loops ai and a i define the same classes
l7c. The Fundamental Group of a Surface 245

in HI(X), and similarly for ~i and bi. In particular, the 2g classes

determined by a 1> b 1, • • • , ag , bg form a free basis for H I (X) .
The polygonal normal form of a surface can also be used to de-
scribe its universal covering space. When g == 0, of course, X is a
sphere, so simply connected. When g == 1, X is a torus, which we
have seen has the plane 1R2 as its universal covering. One can con-
struct this universal covering from its representation as a rectangle
with sides identified, by pasting the sides together but without iden-
tifying opposite sides. For g 2: 2, the universal covering can be re-
alized in a similar way by gluing together copies of the polygon IT,
but one cannot do this in a metrical way in the ordinary plane. How-
ever, in a hyperbolic plane this can be done, and one can see that the
universal covering can be realized as a hyperbolic plane. The hyper-
bolic plane is homeomorphic to an open disk (or to 1R2), so from this
one sees that for all g 2: 2 the universal covering is an open disk. For
more on the hyperbolic plane, see Hilbert and Cohn-Vossen (1952).
Exercise 17.11. Give another proof of the results of this section by
computing the fundamental group of the complement of 2g disjoint
disks in a two-sphere, and then showing what happens when one sews
g handles (each homeomorphic to Sl x [0, 1]) onto pairs of the bound-
aries of these disks.
Problem 17.12. Compute the fundamental group and first homology
group of the space obtained from a surface X of genus g by removing
n disjoint disks or points.
There is another important operation on the first homology group
HIX of an oriented compact surface, an intersection pairing, that as-
signs a number «1, T) to two classes (1 and T in HIX. If the classes
are represented by loops that meet each other transversally in a finite
number of points, this number is the sum of the numbers ± 1 assigned
to each point of intersection, with the number ± 1 assigned according
to the following picture:

't

+1 -I
246 17. The Topology of Surfaces

The number is + 1 if the direction from (J' to ,. is counterclockwise,

and -1 if it is clockwise. (Note that this notion depends on having
an orientation for X.) What takes some proof is showing that this
gives a well-defined number: that any two classes in H1X have rep-
resentatives that meet transversally (which is not difficult), and then
that the number one gets is independent of choices (which is more
difficult). Rather than doing this directly, we will construct this in-
tersection pairing by another procedure in the next chapter, by relating
it to the wedge product of I-forms.
CHAPTER 18

Cohomology on Surfaces

I8a. I-Forms and Homology

On any «6'" surface X, just as on an open set in the plane, we have a
notion of «6'" functions, I-forms, and 2-forms w, and we have linear
maps d, that take a functionfto a i-form df, and a I-form w to a 2-
form dw. (See Appendix D3 for definitions and basic properties.)
Therefore we have a notion of a i-form w being closed (dw = 0) or
exact (w = df for some f). All exact forms are closed, so we can
define the first De Rham cohomology group as before:
HIX = {closed I-forms onX}/{exact I-forms onX}.
Just as in the case of open sets in the plane, if -y: [a, b] ~ X is a
continuous path, and w is a closed I-form on X, we can define an
integral J-yw. As before, if -y is differentiable, this can be done by
calculus: J-yw = J:-y*w. In general, as in the plane, we can subdivide
the interval by a = to::;; t(::;; . . . ::;; tn = b, so that -y([ti - 1 , ta) is con-
tained in an open set Vi on which w = dj;, and then

f-y
w = i/;(-y(tJ) - /;(-y(ti -
i=l
1))·

The same argument as in §9b shows that this is independent of choices.

As before, this extends from integrals over paths to integrals over 1-
chains, and the same argument shows that the integral over a bound-

247
248 18. Cohomology on Surfaces

ary is zero, and that the integral of df over a cycle is zero. It follows
that we again have a canonical map

HIX ~ Hom(HIX, IR)

that takes the class of w to the homomorphism [-y] ~ f.v w.

We want to calculate HIX when X is a compact oriented surface.
In particular, we want to show that this canonical map is an iso-
morphism. This could be done by the Mayer-Vietoris argument (which
generalizes from open sets in the plane), but we will do it by explicitly
constructing a basis, which will be useful later.
We take a model of X as in § I7c , with a basis for HIX given by
the loops aI, b l , . . . , a g , bg as indicated in the picture there. For
each loop a; we will construct a closed I-form U; , and for each b; a
closed I-form 13;, and we will show that the classes of these 2g I-forms
form a basis of HIX. The forms U; and 13; will depend on several
choices , but we will see that their classes in HIX depend only on the
homology classes of the 2g loops aI , b l , . . . , a g , bg •
For each of these loops we can find an open set V containing it that
is diffeomorphic to an annulus U = {(x , y): a 2 < x 2 + / < e2 }, with the
loop corresponding to a counterclockwise circle around the middle of
the annulus , i.e. , to the path t~(bcos(21Tt),bsin(21Tt», 0:5 t:5 I , for
some a<b<e:

We take these neighborhoods V to be disjoint , except for those con-

taining the same a; and b;, which can be taken to intersect in a set
diffeomorphic to an open rectangle.
18a. I-Forms and Homology 249

Consider one such loop, either ai or b i , and fix such a diffeo-

morphism <p from a neighborhood V of the loop to such an annulus
U. From Exercise B.14 we can find a C(6'" function \jJ on the plane that
is identically I outside the big circle and identically 0 inside the small
circle. In fact, for any E with 0 < E < 1/2 (c - a), we can find \jJ so
that
I if x 2 + l ;::: (c - d,
{
\jJ(x,y) = 0 if x 2 + l:5 (a + d.

The function \jJ 0 <p is a C(6oc function on V that is constant on a neigh-

borhood of the boundary of V. Its differential d( \jJ 0 <p) is therefore a
I-form that is identically zero on a neighborhood of the boundary of
V. We can therefore define a C(6'" I-form on all of X by defining it to
be d(\jJ 0 <p) on V, and identically 0 outside V. This is certainly a closed
I-form. In fact, its restriction to V is exact, and its restriction to an
open set V' with V U V' = X is zero. (Note that the form is not exact
on X, however, as follows for example from the following lemma.)
This is the I-form we wanted to construct. We denote it by (Xi if the
given loop is ai, and l3i if the given loop is hi'

Lemma 18.1. These 110rms (Xi and f3i have integrals:

(i) (l3i = 0 Jor all i andj, and

Jbj
Ll3i = {b
j
~! ~~';
(ii) 1=
aj
(Xi 0 Joralliandj, and ((X.={-1
Jb
j I 0
if i=j,
if i ¥- j .
Proof. This is a direct calculation. All of the assertions that integrals
250 18. Cohomology on Surfaces

vanish follow from the fact that they are integrals of I-forms over a
loop, and these forms are exact on an open set containing the loop.
For the integral of 13; along a;, cut a; into two pieces, one inside the
annulus as shown, the other outside.

For the piece inside the annulus, 13; is the differential of a function
whose value at the initial point is 0 and at the end point is 1, so the
integral of 13; along this piece is 1. For the second, 13; is identically
zero. So the integral is I + 0 = I, as asserted. The argument is similar
for the integral of (X; along b;, but this time the function is I at the
initial point and 0 at the final point of the part of the path in the
annulus, so the integral is -1. 0

Proposition 18.2. (1) The canonical map HIX ~ Hom(H1X, JR.) is an

isomorphism.
(2) The classes of the l-forms Ul, {31, ... , u g, {3g form a basis
for HIX.
Proof. The same argument as in the plane shows that this canonical
map is one-to-one. For if W is a closed I-form and I'Yw = 0 for all
closed paths 'Y, then one can define a functionf on X by fixing a point
Po and defining f(P) to be the integral of w along any path from Po
to P. The assumptions make this a well-defined function, and the
proof that df= w is the same as for open sets in the plane (see Prop-
osition 1.8); indeed, the assertion df= w is a local assertion, so it can
be verified on coordinate neighborhoods.
To see that the canonical map is subjective, a homomorphism
h: H1X ~ JR. is determined by its values on a basis, so suppose h([aj]} = rj
and h([bj]} = Sj for some numbers rj and Sj' I $,j$, g. By the lemma,
the closed I-form w = Lf=l(r; 13; - s;(X;)has IaJw = rj and IbjW = Sj, which
means that the canonical map takes W to h. 0

Exercise 18.3. Show that for closed I-forms W on the surface X one
has a "module of periods" story as in Chapter 9: if the integrals of
18b. Integrals of 2-Forms 251

along the basic loops are known, all integrals are determined. Show
CI)

in fact that if Iaj CI) = rj and h CI) = Sj' then for any closed i-chain 'Y,
g

Joy CI) = ~ (miri + nisi),

for some integers mi and ni.

Note in particular that the classes [(Xi] and [13i] of (Xi and 13i in HIX
are determined by their integrals, so are independent of the choices
we made in defining them. They depend in fact only on the choice
of the basis for HIX determined by the ai and bi •

18b. Integrals of 2-Forms

Once an orientation of our surface is chosen, one can integrate '(600
2-forms. The integral of the 2-form v on X will be a real number,
denoted IIx v. More generally, if X is any oriented surface, and v is
a 2-form with compact support (i.e., which is identically zero outside
a compact subset of X), one can define IIxv by the following pro-
cedure. Choose an atlas of coordinate charts 'Pa: Ua--,;X, say with
each Ua an open rectangle in the plane, with all charts compatible
with the given orientation, and satisfying the condition that any point
of X has a neighborhood meeting only finitely many 'Pa(Ua) (see Ap-
pendix A4 and Lemma 24.10).
To give a 2-form v on X is the same as giving a 2-form vadxdy
on Ua (which the same as prescribing a function Va on Ua), such that
these 2-forms are compatible under changes of coordinates. First sup-
pose v is a 2-form that is zero except on a compact subset of one of
the open sets 'PiUa). Then one can define the integral of v by

II = II
x
v
Ua
vadxdy,

where the integral on the right is the ordinary Riemann integral of a

(or continuous) function Va on a rectangle. In general, choose a
'(600
partition of unity {I/Ia} so that the closure of the support of each I/Ia is
contained in 'Pa(Ua). Define the integral of v to be the sum of the
integrals of the I/Iav, i. e. ,
252 18. Cohomology on Surfaces

where the integral of each t\s",v is defined by the preceding case, since
it is zero except on 'P",(U",). Note by the compactness of the support
of v that only finitely many t\s",v can be nonzero, so this sum is finite.

Exercise 18.4. (a) Verify that this definition is independent of the

choice of partition of unity and the choice of coordinate charts.
(b) Verify that the integral is linear: ffx(rlvl + r2v2) = ffxrlvl + ffxr2v2
for 2-forms VI and V2 and real numbers rl and r2. (c) Show that, if
the other orientation is chosen, then the integral changes sign.

We need the following version of the Green-Stokes theorem for a

surface:

Proposition 18.5. If v = dw, where w is a Ilorm with compact sup-

port on an oriented surface X, then ff x v = o.
Proof. Choosing an atlas and partition of unity as in the definition,
since w = ~",t\s",w,

Now t\s",w is a I-form with compact support on 'P",(U",), which cor-

responds to a I-form /-1-", with compact support on the open rectangle
U",. By definition, ffxd(t\s",w) = ffu.d(/-I-a), so it suffices to prove that
this last integral is zero. But /-1-0< has compact support, so it vanishes
on the boundary of the rectangle. By Green's theorem for a rectangle
(Lemma 1. 11),

18c. Wedges and the Intersection Pairing

If X is a compact oriented surface, and w and /-I- are I-forms on X,
we can define a real number (w, /-1-) by the formula
(w, /-1-) = ffxw/\/-I-.
where w/\ /-I- is the wedge product (Appendix D3). From basic prop-
erties of the wedge product (see Exercise D.7) we have:
18c. Wedges and the Intersection Pairing 253

(i) (riwi + r2w2, /-1) = rl(wl, /-1) + r2(w2, /-1) for I-forms WI, W2, and
/-1, and real numbers rl and r2;
(ii) (w, /-1) = -(/-1, w) for any I-forms wand /-1; and
(iii) (dj, /-1) = 0 for a function j and a closed I-form /-1.
Condition (i) says that (w, /-1) is linear in the first factor. Similarly,
or using (ii), it is linear in the second, i.e., it is a bilinear pairing.
Condition (ii) says that this pairing is skew-symmetric. Condition (iii),
together with (ii), says that (w, /-1) = 0 if either w or /-1 is exact. It
follows that we can define a mapping
HIX X HIX ~ ~, [w] x [/-1] ~ (w, /-1) = IIxw/\ /-1.
The point is that, since the integral vanishes when either is exact, the
result is independent of choice of a closed I-form in an equivalence
class. This is also a bilinear and skew-symmetric pairing.

Lemma 18.6. Let {Xi and f3i be the 110rms constructed in §18a. Then
({Xi, (Xj) = 0 = (f3i' f3j) jor all i and j, and

(ai'
_{I
13) - 0
if i = j,
if i"# j .

Proof. As in Lemma 18. 1, this is a direct calculation. The pairings

that are asserted to be 0 are evident, since in these cases the wedge
products of the forms are zero. This uses the fact that a wedge product
w /\ /-1 vanishes where either w or /-1 vanishes, and the fact that w /\ w = 0
for any I-form w. It therefore suffices to verify that (ai' 13,) = 1, i.e.,
that IIx ai /\ l3i = 1, for all i. The form ai /\ l3i vanishes off the region
that is the intersection of the two annuli around the loops ai and bi ,
and this region is diffeomorphic to a rectangle. In a coordinate patch
we have the picture

,,'
b·

'V /
aj

On this rectangle, ai = dj; and l3i = dg i , where j; is a function that is

o on the right side of the rectangle and I on the left side, and gi is a
254 18. Cohomology on Surfaces

function that is 0 on the top of the rectangle and 1 on the bottom. If

R is this rectangle,

II
x
n;A 13; = JJ d};Adg;
R
= II
R
d(k dg;) = Lk dg;,

the last step by Green's theorem on the rectangle R. Now since dg;
vanishes on the bottom and top edges of the rectangle, /;. dg; is zero
on all sides of the rectangle except for the left vertical edge 'Y4, where
it is equal to dg;. So

rk
JaR
dg; = -]
'Y4
dg; = -(ghil» - g;('YiO») = -(0 - 1) = 1,

which finishes the proof. 0

Exercise 18.7. For any closed I-form w on X, show that

(nj'w) = Lw
J
and (l3j'w) = i w.J

Exercise 18.8. Show that, for any closed I-forms f.L and v,

Proposition 18.9. The pairing H1X X H1X ~ IR is a perfect pairing,

i.e.,for any linear map q;: H1X~IR, there is a unique win H1X such
that q;(p,) = (Cd, p,) for all p, in H1X.

Taking 'P = 0, this says in particular that if wE H1X and (w, f.L) =0
for all f.L in H1X, then w = O.
Proof. In this proof we identify closed I-forms with the classes they
define in H1X. If w = ~f=l(r;n; + s;I3;) in H1X, then by Lemma 18.6
we have

2: (r;(n;, 13) + s;(I3;, 13)

g

(w,l3) = = rj'
;=1

-2: (r;(nj, n;) + s;(nj' 13;»

g

(w, n) = -(nj' w) = = -Sj'

i=l

The homomorphism determined by w therefore takes nj to -Sj and I3j

to rj. If this homomorphism is zero, all Sj and rj must be zero, which
means that w is zero in H1X. Conversely, any homomorphism 'P is
18c. Wedges and the Intersection Pairing 255

determined by the values it takes on a basis of HIX, and if we define

rj to be cp(l~j) and Sj = -cp«(X), the same equations show that

00 = ~f=l(ri(Xi + sJ3 i) is a class in HIX with (00, /-1) = cp(/-1) for all /-1. 0

Corollary 18.10. For any 'Y in HIX, there is a unique class 001 in
HIX such that J1P- = JJxOO11\p-jor all p- in HIX.
Proof. Given 'Y, the map /-1~ J-y /-1 is a homomorphism from HIX to
IR, so the proposition gives a unique class oo-y with the required
property. 0

Exercise 18.11. (a) For 'Y = ai, the corresponding class ooa; is repre-
sented by the form (Xi; and for 'Y = b i , the corresponding class oob; is
represented by the form ~i. This shows again that the classes of the
forms (Xi and ~i depend only on the choice of ai, ... , bg • (b) Show
that the map from HIX to HIX taking 'Y to oo-y is a homomorphism.

Problem 18.12. Suppose 'Y is represented by a map from the circle

Sl to X that extends to a diffeomorphism from an annulus containing
Sl to an open subset of X. Use this annulus as in § 17 a to construct a
differential form 00 that vanishes outside the image of the annulus,
and, on the annulus, is the differential of a function that increases
from 0 to 1 as one moves from the inside to the outside of the annulus.
Show that 00 represents the class oo-y of the corollary.

We can use the corollary to define an intersection number (<T, 'T)

for any <T and 'T in HIX, as we indicated at the end of the last chapter.
Namely, let 00" and OOT be the cohomology classes given by the cor-
ollary, and define (<T, 'T) to be (00", ooT ) , i.e.,

Proposition 18.13. (1) This pairing is a bilinear skew-symmetric pairing

on HIX, i.e.,
(mlcrl + m2cr2, 'T) ml(cr I , 'T) + m2(cr2, 'T) ,
(cr, ml'Tl + m2'T2) = ml(cr, 'TI) + m2(cr, 'T2),
('T, cr) -(cr, 'T),
jor all <T, T, (7'1, (7'2, TI, T2 in HIX and all integers ml and m2.
256 18. Cohomology on Surfaces

(2) With the basis aj and bj for H1X as before, (aj, a) = (b j, b) = 0

for all i and j, and

_{I
(aj, b) - 0
if i = j,
if i ~ j .

(3) The number (u, T) is always an integer.

(4) For any u in HIX, the map T~ (u, T) is a homomorphism from
H1X to lL, and every homomorphism from H1X to lL arises in this way
from a unique u in HIX,
Proof. Part (1) follows from the corresponding assertions for the pair-
ing ( , ) on cohomology classes. Part (2) follows from Lemma 18.6
and Exercise 18.11: for example, (aj, bj ) = (ai> 13), which is 1 if i = j
and 0 otherwise. Since the aj and bj form a basis for HIX, (3) follows
from (1) and (2). The proof of (4) is the same as before: the element
u = "Lr.aj + sjbj has (u, a) = -Sj and (u, b) = rj, with this time rj and
Sj integers, and a homomorphism is determined by arbitrarily speci-
fying the integers to which each element of a basis maps. 0

Problem 18.14. Suppose u and'T are represented by maps from Sl

to X that extend to diffeomorphisms of annuli with open sets in X,
so that the images cross transversally at a finite number of points.
Show how to assign numbers + 1 or - 1 to each intersection point
(see the end of §17), in such a way that (u,'T) is the sum of these
numbers.

It should be pointed out that, as we have constructed it here, the

intersection pairing uses a differentiable structure on the surface.
However, as the preceding problem indicates; it really depends only
on the topology.

Problem 18.15. Prove this assertion.

18d. De Rham Theory on Surfaces

We have concentrated on the first De Rham group HIX. As in the
plane, on any surface one has the Oth group HOX, which is the space
of locally constant functions on X. If X is connected, such functions
are constant, and HOX is just the space ~ of constant functions.
18d. De Rham Theory on Surfaces 257

We can define the second De Rham group H2X of a differentiable

surface X by
H2X = {2-fonns on X}/{exact 2-fonns on X}.
If X is a compact oriented surface, it follows from Proposition 18.5
that integration defines a canonical map

[v] ~ II
x
v.

It is easy to see that this map is surjective, by constructing a fonn v

that vanishes outside one coordinate neighborhood 'Pa(Ua), and in Ua
has an expression Va dx dy, with Va a nonnegative function with given
integral. We claim that, in fact, this canonical map is an isomor-
phism. This is equivalent to the

Claim 18.16. If IIx v = 0, then v must be exact.

Equivalently, the dimension of H2X is (at most) one. With this, we

will know all the De Rham cohomology groups of a compact surface
of genus g:
HOX = IR, HIX = IRffi2 g, H2X = IR.
We will show how to extend the Mayer-Vietoris story to include
the second cohomology group, which will in particular prove this claim.
We will also see that if X is not compact, or if X is not orientable,
then H2X = 0: every closed 2-form is exact.

Exercise 18.17. Show directly from the definition that, if U is an

open rectangle, every 2-fonn is exact. Conclude that H 2U = 0 if U is
diffeomorphic to an open rectangle.

Exercise 18.18. Suppose a surface X is a disjoint union of a finite or

infinite number of open sets Ui • Show that specifying a k-fonn on X
is the same as specifying a k-fonn on each Ui , and specifying a class
in HkX is the same as specifying a class in HkUi for each i. In other
words, HkX is the direct product of the groups HkU i •

We want to compare the cohomology groups for different open sets.

Note first that if UI is a subset of U2 , any differential function or
fonn on U2 detennines by restriction a differential function or fonn
on U I • This restriction commutes with the boundary maps d (which
amounts to the obvious fact that partial derivatives of a function re-
258 18. Cohomology on Surfaces

stricted from a larger open set to a subset are the restrictions of the
partial derivatives). This means that restriction takes closed forms to
closed forms, and exact forms to exact forms, and hence determines
linear maps, also called restriction maps:
H k(U2) ~ Hk(U I ), k = 0,1,2.

Exercise 18.19. If U I C U 2C U 3 , show that the restriction map from

Hk(U3) to Hk(U I ) is the composite of the restriction map from Hk(U3)
to H k(U2) followed by the restriction map from H k(U2) to Hk(U I).

Exercise 18.20. If U and V are open sets in a surface X, and 00 is a

k-form on un V (with k = I or 2), show that there are k-forms WI on
U and 002 on V so that 00=001-002 on unv.
Given open sets U and V in a surface X, the contructions of Chapter
10 extend without change, giving a coboundary map
8: HO(Un V) ~ HI(UUV).
The Mayer-Vietoris properties MV(i)-MV(v) of §IOd continue to
hold for these spaces and maps. Similarly, we construct a coboundary
map
8:HI(UnV) ~ H2(UUV).
The construction of a map from {closed I-forms on un V} to H2(U U V)
proceeds exactly as before: Given a closed I-form 00 on un V, use
Exercise 18.20 to write 00 as WI - 002, with WI and 002 I-forms on U
and V, respectively. The 2-forms dWI and dW2 agree on un V, so
there is a unique 2-form fJ. on U U V that agrees with dWI on U and
with dW2 on V. Define 8(00) to be the class of this 2-form fJ.: 8(00) = [fJ.].
Exactly as before, one checks that this is independent of the choice
of WI and 002, and is a linear map. To see that it defines a homo-
morphism on the quotient space
HI(U n V) = {closed I-forms on un V}/{exact I-forms on un V},
we must show that the map just defined vanishes on the exact I-forms.
If 00 = dl, write I = II -/2, with II and 12 functions on U and V, re-
spectively. Then in our construction of the coboundary of 00 we may
take WI = dll and 002 = d12 , from which it follows that dWI and dW2 are
both zero, so fJ. = 0 and 8([00]) = 0, as required.
We claim next that properties (i)-(v) continue:

n
MV(vi). Given 00 in HI(U V), 8(00) = 0 if and only if 00 = a - ~ lor
some a in HIU and ~ in HIV.
18d. De Rham Theory on Surfaces 259

MV(vii). Given J.L in H2(U U V), J.L restricts to zero in H 2U and H2V
if and only if J.L = 8(w) for some w in HI(U n V).

MV(viii). Given a in H 2U and f3 in H 2V, a and f3 have the same

restriction to H2(U n V) if and only if a and f3 are the restrictions of
some element in H2(U U V).

The proofs are exactly the same as before, and are left as exercises.
In addition, we have, as an immediate consequence of Exercise 18.20:

MV(ix). Any class J.L in H2(U n V) can be written as the difference

of a class a in H 2U and a class f3 in H2V.

Putting this together, using the fancy language as before, we have

the full

Theorem 18.21 (Mayer-Vietoris Theorem). For any open sets U and

V in a surface, there is an exact sequence

o- ----=- HO(U n V)
H°(U uV) ~ HOU EEl HOV
~ HI(UU V) ~ HIU EEl HIV ----=- HI(U n Y)
~ H2(UU V) ~ H 2U EEl H2y --=- H2(U n Y) - o.
Problem 18.22. Use Mayer-Vietoris to show that H2X = 0 for any
open subset of the plane.

Proposition 18.23. If X is a surface diffeomorphic to a sphere with

g handles, then dim H2X = 1.
Proof. We can write X as the union of the same two open sets U and
Y that we used in the proof of Proposition 17.6. Since V is diffeo-
morphic to an open disk, H I Y=O=H2V, see Exercise 18.17. We
leave it as an exercise to show similarly that H 2U = o. We know that
HI(U n V) == lit The proof of Proposition 17.6 shows that the map
from HIU to HIX induced by inclusion is an isomorphism. It follows
from this and Proposition 18.2 that the restriction map from HIX to
HIU is an isomorphism. The relevant part of the Mayer-Vietoris se-
quence is therefore
HIX ~ HIUffiO ~ HI(Un V) ~ H2X ~ OEBO.

lt follows that the map from HI(U n V) to H2X is an isomorphism.

Since HI(U n V) == IR, this shows that H2X is one dimensional. 0
260 18. Cohomology on Surfaces

Exercise 18.24. Give another proof of the proposition by writing X

as a union of a sphere with 2g disjoint disks removed, and the union
of g handles, each diffeomorphic to a cylinder Sl X (0, o.

Exercise 18.25. If X is a nonorientable compact surface, use Mayer-

Vietoris to show that H2X = O. If X is written in normal form as in
Problem 17.5, show that the dimension of HIX is h - 1. Show that
for all the compact surfaces, if triangulated with v vertices, e edges,
andjfaces,
v - e +j = dim(JtX) - dim(HIX) + dim(H2X) .

Problem 18.26. (a) If f Y~X is a differentiable map of surfaces,

show how to define pull-backsj*w of forms from X to Y, determining
mapsj*: HkX ~Hky for k = 0, 1,2. (b) Iff is an n-sheeted covering
map, show how to define push-forwards f*w of forms from Y to X,
definingj*: Hky~ HkX. (c) Withfas in (b), show that the composite
h of*: HkX ~ HkX is multiplication by n. (d) If X is a compact non-
orientable surface, and f: X~ X is its orientation covering, show that
If xf*w = 0 for any 2-form w on X. Conclude again that H2X = O.

We will see in Chapter 24 that there are corresponding homology

groups HzX, with a corresponding Mayer-Vietoris sequence, and iso-
morphisms H2X == Hom(HzX, ~). We will also see in Chapter 24 that
H2X vanishes for any noncompact surface.
 PART X

RIEMANN SURFACES

Many of the ideas about covering spaces, homology, and cohomol-

ogy, can be used in the study of Riemann surfaces. A Riemann sur-
face is a differentiable surface with a complex analytic structure.
Compact Riemann surfaces arise by taking a finite-sheeted covering
of the complement of a finite set in the two-sphere S2, and then filling
in appropriately over the branch points. We prove the Riemann-Hur-
witz formula that computes the genus (number of handles) of a surface
arising this way.
If F(z, w) is an irreducible polynomial in two complex variables, it
defines w as an algebraic function of z, and one is interested in the
behavior of integrals of the form Idz/w, or more generally IR(z, w)dz,
where R is any rational function. Associated with F there is a complex
plane curve F(z, w) = 0; as a subset of ([} it is, except for a possible
finite number of singularities, a Riemann surface, and the above pro-
cess compactifies this surface. One of the key discoveries of the nine-
teenth century was how the topology of this surface, i.e., the genus
g, controls much of the analysis related to the algebraic function w
and its integrals. The integrals between two points are defined up to
"periods," which are integrals along classes in the first homology
group of the Riemann surface.
In Chapter 21 we use what we have learned about the topology and
differential forms on a surface to prove the celebrated Riemann-Roch
theorem, which shows how that topology (genus) of a surface controls
the kinds of meromorphic functions and forms one can find on the
surface. We also prove the Abel-Jacobi theorem, which similarly re-
262 Part X. Riemann Surfaces

lates the possible integrals on the surface to its topology. We prove

these theorems only for Riemann surfaces that are known to arise
from a plane curve, i.e., from an irreducible polynomial F(Z, W). (It
is true that all compact Riemann surfaces arise this way, but we do
not prove this here.) In this case there is a short proof of Weil that
uses the algebraic notion of adeles, together with the few facts from
analysis and topology that we have available.
CHAPTER 19

Riemann Surfaces

19a. Riemann Surfaces and Analytic Mappings

A Riemann surface X is a connected surface with a special collection
of coordinate charts 'Pa: V a ~ X. As before, Va is a subset of 1R2, but
now we identify 1R2 with the complex numbers C. The requirement
to be a Riemann surface is that the change of coordinate mappings
'P~a from Va~ C Va to V[3a C V[3 are not just C€"', but they must also be
analytic, or holomorphic. Recall (see §9d) that an analytic function
f on an open set in C is a complex-valued function that is locally
expandable in a power series, i.e., at each point Zo in the open set,
there is a power series L;~oan(z - zor that converges to f(z) for all z
in some neighborhood of Zo. As before, another atlas of charts is
compatible with a given one (and defines the same Riemann surface)
if the changes of coordinates from charts in one to charts in the other
are all analytic.
Any Riemann surface has a natural orientation. This follows from
the fact that, when the analytic changes of coordinates are written out
in terms of their real and imaginary parts, the Jacobian of the result
has positive determinant. We will consider only connected Riemann
surfaces.

Exercise 19.1. Show that in fact, if w =f(z) , with w = u + iv and

z = x + iy, the determinant of the Jacobian of the map from (x,y) to
(u, v) is the square of the absolute value of the complex derivative
f'(z).

263
264 19. Riemann Surfaces

We have seen several examples of Riemann surfaces. Of course,

( itself is one, with the identity coordinate chart (~C, and any open
subset of ( (or any Riemann surface) is also a Riemann surface. The
sphere S2 is a compact Riemann surface, with the charts given by
spherical projection from the north and south poles. Indeed, we saw
(Exercise 7. 14) that the change of coordinates mapping in this case
is given by the map z~ liz from (\ {O} to (\ {O}. (When we speak
of S2, it will always be regarded as a Riemann surface with these
coordinate charts.) The torus [R2/Z2 == (/(Z + Zi) is a compact Rie-
mann surface, using the projections from small open sets in ( to the
quotient for coordinate charts.
Any compact oriented surface can be given a structure of a Rie-
mann surface (as we shall see later), but, except for the sphere S2,
there are infinitely many nonequivalent Riemann surfaces with the
same underlying surface. For example, if A C ( is any lattice, i.e.,
a subgroup generated by two elements that give a basis for ( as a
real vector space, then (I A is likewise a compact Riemann surface.
These are all homeomorphic (and diffeomorphic) to each other, but
they are generally not isomorphic Riemann surfaces (see Exercise
19.17). This is part of the general subject of "moduli of Riemann
surfaces" -the study of the set of all Riemann surfaces of a given
genus-that we can only hint at in this book.
If X is a Riemann surface, it makes sense to say that a function
f: X ~ ( is analytic (or holomorphic): f is analytic if for each coor-
dinate chart <P"': U",~X, the compositefo<p", is an analytic function
on the open set U'" in C. More generally, if X and Y are Riemann
surfaces, a mappingf: X ~ Y is analytic at a point P in X if there are
charts <p: U ~ X and t/!: V ~ Y mapping to neighborhoods of P and
f(P), respectively, so thatf(<p(U) C t/!(V), and the composite t/!-I ofo 'fJ
is an analytic function from U to V:

x y

o o
19a. Riemann Surfaces and Analytic Mappings 265

This condition is independent of the choice of coordinates <p and tV.

In fact, we can choose both U and V to be disks centered at the origins,
with these origins mapped to P and f(P). In this case the composite
h=tV-1ofo<p has the form h(Z)=L~=la"zn, for some converging power
series. As we will soon see, the order of vanishing of h at the origin,
i.e., the smallest integer e such that ae ¥= 0, is independent of the
choice of coordinates. This integer is called the ramification index of
fat P, and we will denote it by ef(P)' or just e(P) when one function
f is being considered. (Iff is constant, of course, h is identically 0,
and e = co, but we will not be interested in constant maps.) The point
P is called a ramification point for f if ef(P) > 1.
We claim next that we can change the coordinate chart <p so that
the composite tV -I of0 <p is the function z ~ ze. To see this, write
h(z) = l· g(z), where g(z) is an analytic function at the origin and
g(O) ¥= O. (g(z) = L~=oan+ezn.) Such an analytic function g, possibly in
a smaller disk around the origin, can be written as the eth power of
an analytic function k(z); for example, g(z)/a e maps 0 to 1, so maps
a neighborhood of 0 to the right half-plane, so one can compose it
with a branch of the log function
1
-2:
oc

log(z) = -(1 - z)" for Iz - 11 < 1 ,

n=1 n
and then set

k(z) = a' expC 10g(g(Z)/ae») ,

where a is any eth root of ae and exp(z) = L~=o(1/n!)zn.

Exercise 19.2. Verify that k(z)e = g(z) in some disk containing the
origin. Show that there are exactly e choices for such k (up to shrink-
ing the disks they are defined on), obtained by the e choices for the
eth root a of ae •

Now we have written h(z) = (z· k(z»" where k is an analytic func-

tion with k(O) ¥= O. The mapping z ~ z· k(z) is an analytic isomor-
phism in some neighborhood of the origin, since its derivative does
not vanish at the origin. Therefore, we can define a new coordinate
chart (j5 (from some small disk V' to X) so that (j5 (z) = <p(z . k(z» for
all small z. It follows that, with this new coordinate chart (j5, the
composite tV of (j5 is just the map z ~ ze, as required.
0

The mapping z ~ ze is familiar: it maps 0 to 0, and outside the

266 19. Riemann Surfaces

origin it is an e-sheeted covering map. This shows that there are

neighborhoods U of P and V of f(P) so that f maps U to V, and the
mapping U \ {P}~ V \ {f(P)} is an e-sheeted covering. This shows in
particular that the number e is independent of choices, and in fact
depends only on the topology of the map f near P.
Note another important consequence of this local structure: a non-
constant analytic mapping is always an open mapping, i.e., the image
of any open set is open. In particular, if X is compact and Y is not
compact, the only analytic mappings from X to Yare constants. When
Y = C this openness is a strong form of the maximum principal: one
cannot have a point Po such that If(P o)I~ If(P) I for P in a neighbor-
hood of Po.

Proposition 19.3. Let f: X ~ Y be a nonconstant analytic map be-

tween compact Riemann surfaces.
(1) There are a finite number of ramification points. Let R ex be
the set of ramification points, and set S = fiR) e Y.
(2) The map from X V-I(S) to Y \ S determined by f is an n-sheeted
covering map for some finite number n. This integer n is called
the degree off.
(3) For any point Q in Y, ~PEf-'cQ)eJP) = n.

Proof. (1) follows from the fact that for any point P, there is a neigh-
borhood of P that contains no other ramification point, using the com-
pactness of X to cover it by a finite number of such neighborhoods.
It follows similarly thatrl(Q) is finite for all points Q in Y (consider
a possible limit point of the setrl(Q».
To prove (2), let Q fI. S, and rl(Q) = {PI, ... ,Pn }. There are
neighborhoods Ui of Pi and Vi of Q such that f maps Ui homeomor-
phically onto Vi. Shrinking the Ui if necessary, we may assume they
are disjoint, and that Vi contains no point of S. For any connected
neighborhood Vof Q contained in vln ... nvn , let U/ = Ui nf-I(V);
note that f maps each U/ homeomorphic ally onto V. We claim that
for V sufficiently small, rl(V) is the disjoint union of the sets U/,
from which it follows that V is evenly covered, so f is a covering in
a neighborhood of Q. To prove this claim, suppose on the contrary
that there is a sequence of neighborhoods Ni of Q whose intersection
is {Q}, such that there is a point P/ in rl(NJ with P/ not in
UI U ... U Un. By the compactness of X, a subsequence of these P;
must have a limit point P' in X. By the continuity off, f(P') = Q, so
P' = Pj for some j. But this contradicts the assumption that the points
19a. Riemann Surfaces and Analytic Mappings 267

P;' are not in Uj for all i andj. (In fact, having proved (2), it follows
that the whole neighborhood V= nVi is evenly covered.)
For (3), again letf-I(Q) = {PI' ... , Pm}, and find neighborhoods
Ui of Pi and Vi of Q such that f maps Ui onto Vi, so that in local
coordinates it is the map z >-+ Ze(Pf ) , so it is elPi ) to I except at the
point Pi' Again if V is a neighborhood of Q contained in the inter-
section of the Vi' but not containing any other point of S, then there
are ~ef(Pi) points over a point Q' in V except for the point Q. By
(2), this sum must be the number n of sheets of the covering. 0

A meromorphic function f on a Riemann surface X is the same as

an analytic function f: X --+ S2 from X to the Riemann sphere. Equiv-
alently, f is an analytic function from X \ S to C, with S a discrete
subset of X, and for each point P in X, there is a coordinate chart
'P: U--+X taking 0 to P, so that on U\ {P}, fo 'P(z) = l· g(z), for some
integer k and some analytic function g that is nonvanishing in U. The
integer k is called the order of fat P, and is denoted ordp(f). It is
independent of the choice of the local coordinate. If k is positive, one
says that f has a zero of order k, or vanishes to order k, and if k is
negative, we say that f has a pole of order - k.

Exercise 19.4. Show that f is meromorphic at P if and only if there

is a coordinate chart as above and an integer k so that the function
Z-k'(fo'P)(z) is bounded as z approaches O.

In terms of the mapping f X --+ S2, the order off at P is positive if

f(P) = 0, negative if f(P) = 00, and zero otherwise. If f(P) = 0, the
order off at P is just the ramification index ef(P) off at P. Similarly,
if f(P) = 00, the order of f at P is minus the ramification index at P,
since liz is a local parameter for the Riemann sphere at 00. If X is
compact, we know that the sum of the ramification indices over any
point in S2 is the degree n off. Assertion (3) of the proposition says
that f takes on all values the same number of times, counting mul-
tiplicity correctly. In particular, for the values 0 and 00, this gives:

Corollary 19.5. For any nonconstant meromorphic function f on a

compact Riemann surface X,
L ordp(f) = O.
PEX

Exercise 19.6. Let Pt(z) and P2(Z) be polynomials in z of degrees d t

and d 2 , with no common factors. (a) Show thatf(z) = PI(z)lp2(z) de-
268 19. Riemann Surfaces

tennines an analytic mapping from S2 = C n {oo} to itself. (b) Show

that the degree of this mapping is the maximum of the integers d l and
d2 • (c) Show that the ramification index of this mapping at 00 is idl - d 2i.
(d) Show that every analytic mapping from S2 to S2 has this form.

19b. Branched Coverings

If Y is a Riemann surface, and p: X ~ Y is a covering mapping, then
there is a unique structure of a Riemann surface on X so that p is an
analytic mapping. In fact, one can choose charts 'Pa: Ua ~ Y for Y so
that each 'Pa(Ua) is evenly covered by p; each component Va,; of
p-I('Pa(Ua» maps homeomorphically to 'Pa(Ua) by p, and the com-
posite
Ua ~ 'Pa(Ua) ~ Va.;
of 'Pa and the inverse of p, gives a coordinate chart on X. It is straight-
forward to verify that the changes of coordinates for these charts are
analytic, so define a Riemann surface structure on X.

Exercise 19.7. If Y is a Riemann surface, and G is a group acting

evenly on Y by analytic isomorphisms, show that Y/G can be given
the structure of a Riemann surface so that Y ~ Y /G is analytic.

An important case of this is the fact that the universal covering X of

a Riemann surface is a Riemann surface, and the fundamental group
'll'1(X, x) acts as a group of analytic isomorphisms of X. When X = S2,
X = X; when X = C/A for a lattice A, X = C. It is a fact of complex
analysis (the uniJormization theorem) that in all other cases, the uni-
versal covering is isomorphic to the upper half plane H (or an open
disk). The automorphisms of H all have the fonn z~(az+b)/(cz+d)
where (~ ~) is a real matrix of detenninant I. This means that X can
be realized as the quotient of H by a subgroup of SL2(~)/{±I} acting
evenly on H. We will not have more to say about this situation, which
is a fundamental area of mathematics in its own right.
In the last section we saw that any (nonconstant) analytic map X ~ Y
of compact Riemann surfaces determines a finite-sheeted covering
X \f-I(S)~ Y\ S, for some finite set S in Y. Our goal in this section
is to reverse this process. Given a Riemann surface Y, a finite subset
S of Y, and a finite-sheeted topological covering p: X" ~ Y \ S, by what
we just proved, )(' gets a structure of a Riemann surface so that this
19b. Branched Coverings 269

mapping is analytic. We want to "fill in" the missing points over the
points of S, embedding r in a Riemann surface X, so that we have
an analytic mapping/from X to Y compatible with the given covering:
Xo C X

p 1 1/
Y\S c Y.
If Y is compact, we want X to be compact. In general, we want the
mapping/to be proper: for any compact set K of Y, the inverse image
rI(K) should be a compact set in X. The problem is local on Y: we
need to fill in the covering over each point of S. We will look first
at the "local model" of a Riemann surface-an open disk.
Let D = {z: Izl < I}, DO = D \ {O}. We know all about the coverings
of DO. In fact, its fundamental group is lL (since it contains a circle
as a deformation retract), so connected finite-sheeted coverings cor-
respond to subgroups of finite index in lL, and the only such subgroups
of lL are the groups elL for e a positive integer. The e-sheeted covering
corresponding to this subgroup is
Z f--o? ze.
This means that if p: EO ~ DO is any e-sheeted connected covering,
there is a homeomorphism 1\1: DO ~ eo so that po 1\1 = Pe.

This homeomorphism 1\1 is not uniquely determined, depending, as

we know, on a choice of where a base point is mapped. In particular,
there are exactly e such homeomorphisms, corresponding to the e
points of eo over a given base point. The other choices of 1\1 have the
form z f--o? l\I(tk' z), where tk = exp(2'7Tkije) is one of the eth roots of
unity, 1 s k S e - 1.
It is now clear how to fill in the covering p: eo ~ DO. Define E to
be the union of eo with one other point, and put the structure of a
270 19. Riemann Surfaces

Riemann surface on E so that the extension of IjI from DO to D, map-

ping 0 to the added point, is an isomorphism of D with E. Note that
this does not depend on the choice of 1jI, since the map z ~ ~k • z is
an analytic isomorphism.
Now return to the given covering p: X"~ Y\ S. Given Q in S, we
can find a coordinate chart '1': D~ <p(D) = U C Y, with D the open
unit disk in the plane, and '1'(0) = Q, such that U does not contain
any other point of S but Q. Let if = U \ {Q}. The covering p restricts
to a covering of if, so p-I(if) is a disjoint union of connected open
sets Vlo, ... , Vmo, with each Vio~ if a connected covering, say with
ei sheets. By what we just saw, one can find homeomorphisms
ljIi: DO ~ vt such that the diagram

wO @v;o 'IIi
-

, H"; j j P

0·0 cp
~uo

commutes, i.e., P(ljIi(Z)) = <p(zei ), with ljIi unique up to first mUltiplying

z by an eith root of unity. We can therefore add one point to each
Vio, getting spaces Vi so that each ljIi extends to a homeomorphism
from D to Vi' Taking these extensions as charts, the disjoint union
of the Vi becomes a Riemann surface. The map from vt to if extends
to an analytic map from Vi to U that has ramification index ej at the
added point.
If this is done at each point of S, one gets a space X that is the
union of X" with a finite number of points. These local charts give X
the structure of a Riemann surface (noting that the added charts are
compatible with the given charts on X"), and the covering p is ex-
tended to an analytic mapping f: X ~ Y.

Exercise 19.8. Verify that this defines a Riemann surface X, and the
map f is analytic and proper.
19b. Branched Coverings 271

These results are summarized in the following proposition:

Proposition 19.9. Let Y be a Riemann surface, S a finite subset of

Y, andp: X"~ Y\ S afinite-sheeted covering map, with X" connected.
Then there is an embedding of X" as an open subset of a Riemann
surface X that is a union of X" and a finite set, so that p extends to
a proper analytic mapping from X to Y.

Exercise 19.10. Show that X, with its Riemann surface structure, is

unique up to canonical isomorphism: if X" C X' ~ Y is another, there
is a unique isomorphism of Riemann surfaces from X to X I compatible
with the inclusions of X" and mappings to Y.

Exercise 19.11. Extend the proposition to the case where S is an

infinite but discrete subset of Y.

One reason for the importance of this proposition is that, since we

know the fundamental group of Y \ S (see Problem 17.12 for the gen-
eral case), we know all possible finite-sheeted coverings. The main
case of concern to us will be Y = S2, with S a set of r points. By
choosing disjoint arcs from a base point Qo to these points, we can
number the points S = {Qt, ... , Qr} so that the arcs to them occur
in this order, going counterclockwise around Qo. Then one can con-
struct loops (1t, . . . , (1r that go from Qo along an arc to a point near
Qi' make a counterclockwise circle around Qi' and go back along the
arc to Qo.
272 19. Riemann Surfaces

Exercise 19.12. Show that 1T)(S2 \ {Q), ... , Qr}, Qo) is the free group
Fr on the generators <T), ••• , <Tn modulo the least normal subgroup
containing <T)' • • • • <Tr • This is isomorphic to a free group on r - 1
generators <T), ••• , <Tr -).

Remark 19.13. In §16d we described a correspondence between n-

sheeted coverings of a space and actions of its fundamental group on
a finite set T with n elements. To give an action of this fundamental
group on a set T is the same as giving r permutations s), . . . , Sr of
T, with the requirement that S) • • . . • Sr is the identity permutation.
So any such data determine a branched covering of the sphere. The
classical way to do this was to use the fact that the complement of
the r arcs drawn is simply connected, so a covering over this is a
disjoint union of r copies of it. Labeling these by elements of T, the
permutations are used to describe how to glue the sheets together across
the arcs, to get the Riemann surface X. (Note, however, that the arcs
are only a tool used to give nice generators for the fundamental group,
and are not necessary for describing the covering.)

Exercise 19.14. (a) Carry out this construction, and show that it agrees
with that described before. (b) Show that the covering corresponding
to this choice of permutations is connected if and only if they generate
a transitive subgroup of the automorphisms of T. (Recall that a group
of permutations is transitive if, for any elements t) and t2 of T, there
is a permutation in the group that takes t) to t2') (c) Let X be the
corresponding Riemann surface, withf: X~S2 the analytic mapping.
For each i, write T as a disjoint union of sets T i•j so that Si permutes
the elements of each Ti,j cyclically. Show that there is one point of
X over Qi for each set Ti,j' and the cardinality of Ti,j is the ramification
index of f at this point.

19c. The Riemann-Hurwitz Formula

Given an analytic mapping f: X ~ Y between compact Riemann sur-

faces, our next aim is to describe the topology of X in terms of the
topology of Y and the local behavior around the branch points.
19c. The Riemann-Hurwitz Formula 273

Theorem 19.15 (Riemann-Hurwitz). Letf: X~ Y be an analytic map

of degree n between compact Riemann surfaces. If Y can be trian-
gulated, so can X, and the genus gx of X and the genus gy of Yare
related by the formula

2gx - 2 = n(2gy - 2) + L (ef(P) -

PEX
1).

Proof. By refining the triangulation of Y if necessary, for example,

using the barycentric subdivision described in §8b, we can assume
the triangulation has the property that each point Q of S is a vertex
in the triangulation. We will construct from this an explicit triangu-
lation of X. It is simplest to describe the vertices and the "open"
edges (homeomorphic to (0,1)) and the open faces (homeomorphic
to the interior of a plane triangle) of this triangulation of X: these are
exactly the components of the inverse images by f of the corre-
sponding vertices, open edges, and open faces of the triangulation of
Y. The point is that each of the open edges and faces on Y is in the
locus where the map is a covering, and, since edges and faces are
simply connected, the restriction of the covering to each of them is
trivial, which means that there are n subsets of X that map homeo-
morphically onto each of them, and these are the specified open edges
and faces.
For each closed edge and face on Y the triangulation specifies a
homeomorphism with a closed interval or a closed triangle. Com-
posing with the projections, we have compatible homeomorphisms of
the open edges and faces on X with the interiors of these intervals or
triangles. To see that we have a triangulation, we must verify that
these homeomorphisms extend to the closed intervals and triangles.
This is clear when the image edges or triangles in Y do not have
vertices in S, since in that case the coverings are trivial over the entire
closed triangle. The same argument shows that the homeomorphisms
extend (uniquely, by continuity) to the closures of the intervals and
triangles, except possibly in an arbitrarily small neighborhood of a
vertex mapping to a point in rl(S). To see that it extends continu-
ously to such points is a local question, so we can assume we are in
the situation z ~ ze of the mapping from the disk to itself. The cor-
responding vertex is sent to the center of the disk, and the continuity
of the map at the vertex follows from the fact that Izl ~ 0 if and only
if Izel~o. The following picture shows the two types of behavior:
274 19. Riemann Surfaces

//q x
I /
I / ~ " , ", \. \ \

1/
/
4'-..J ,, \.
\.
\
\
I / / \. \
1/ / , ," \

.'<J
1// \. \

<J-q y
q

Now suppose the triangulation of Y has v vertices, e edges, andf

faces. By the construction of the triangulation of X, it has n"l faces
and n' e edges. But the number of vertices is not n' v, since over
each point Q of S there may be fewer than n points. In fact, by the
equation "2,ef(P) = n, the sum taken over the points P inr'(Q), the
number of points over Q is n - "2,(ef(P) - 1). It follows that the num-
ber of vertices in the triangulation of X is nov - "2,(ef(P) - 1), with
the sum over all P inr'(S), or in X, since ef(P) = 1 if P is not in
r 1(S). The Euler characteristic of X is therefore

nov - L (e/P) - 1) - n e + n of = n(v - e +f) -

0 L (ef(P) - 1).
PEX PEX

Replacing the left side by 2 - 2gx , and v - e +f by 2 - 2gy, the for-

mula of the proposition results. D

Exercise 19.16. Give an alternative proof of the theorem by trian-

gulating Y so that each point of S lies inside a face, and no face
contains two points of S.

For example, when Y = S2, we have Riemann's formula

L (ef(P) -
PEX
1) = 2gx + 2n - 2.

If X is also the sphere, this gives

L (ef(P)-1)
PEX
2n-2.
19c. The Riemann-Hurwitz Fonnula 275

The Riemann-Hurwitz formula can be useful to limit the possibil-

ities for mappings between Riemann surfaces. Consider for example
a nonconstant analytic mapping! CIA-CIA', where A and A' are
two lattices in C. Since both have genus 1, it follows from the Rie-
mann-Hurwitz formula that ef(P) = 1 for all P, i.e., f must be un-
ramified. Now f lifts to an analytic mapping 1: C-C so that the
diagram
1
t t
C~C

p p'
C/A~ Cli\.,
commutes. In fact, since C is simply connected, the composite fo p
lifts through the covering p' by Proposition 13.5, to produce a con-
1
tinuous map " and is automatically analytic since the projections
from C to the quotient spaces are local isomorphisms. Similarly, since
1
f is unramified, is also unramified. From the topology one can see
1
also that extends continuously to a map from S2 = C n {oo} to itself,
taking 00 to 00. From Exercise 19.4 this extension, also denoted 1, is
an analytic mapping from S2 to S2. If the degree of 1 is n, the sum
ofthe ef(p) - 1 can be at most n - 1 (since ramification can take place
only over (0), and this contradicts the Riemann-Hurwitz formula un-
less n = 1. By Exercise 19.6, we deduce that J(z) = Az + 1-.1. for some
complex numbers A and 1-.1., with A ~ O. In particular, A' A + 1-.1. cA'.
Conversely, any such A and 1-.1. determine an analytic mapping f. This
puts strong restrictions on the relations between the lattices, and on
the possible maps f.

Exercise 19.17. (a) Show that for any lattice A there is a nonzero
complex number a so that a . A is generated by 1 and T, where T is
a number in the upper half plane. So every C/A is isomorphic to one
of the form C/ClL + Z· T). (b) Show that for two numbers T and T' in
the upper half plane, C/(Z + Z· T) is isomorphic to C/(Z + 7L. T') if
and only if T' = (aT + b)/(cr + d) for some integers a, b, e, d with
ad - be = 1. (c) Show that the only analytic maps from the Riemann
surface C/(7L + 7L. T) to itself that take the image of 0 to itself are
given by multiplication by an integer n, unless T satisfies a quadratic
polynomial with integer coefficients.

It follows from the preceding exercise that the set of compact Rie-
mann surfaces that are isomorphic to some C / A are in one-to-one
correspondence with the space H/S~(Z) of orbits of the discrete group
276 19. Riemann Surfaces

SL2(Z) on the upper half plane H. In particular, not all such Riemann
surfaces are isomorphic as Riemann surfaces, although they are all
diffeomorphic to Sl x Sl. We will see later that every compact Rie-
mann surface of genus 1 has the form Cj A, so HjSL2 (Z) is the mod-
uli space of compact Riemann surfaces of genus 1.

Exercise 19.18. Let/: X ~ Y be a nonconstant analytic map between

compact Riemann surfaces. Show that gx ;:::: gy, and gx > gy unless gy = 0
or gy = 1 or I is an isomorphism.
CHAPTER 20

Riemann Surfaces and Algebraic Curves

20a. The Riemann Surface of an Algebraic Curve

If F(Z, W) is a polynomial in two variables, with complex coeffi-
cients, that is not simply a constant, its zero set
C = {(z, w) E c 2 : F(z, w) = o}
is called a "complex affine plane curve." Identifying C2 with R4 , C
is defined by two real equations: the vanishing of the real and imag-
inary parts of F(z, w). We may therefore expect C to be a surface,
and this expectation is generally true, except that, just as in the case
of real curves, C may have singularities. We will use the construction
of the preceding section to "remove" these singularities, and also add
some points over them and "at infinity," to get a compact Riemann
surface. In fact, if F is not irreducible, the surface we get will be the
disjoint union of the surfaces we get from the irreducible factors of
F, so we assume for now that F is an irreducible polynomial, i.e., it
has no nontrivial factors but constants. Write
F(Z, W) = ao(Z)W n + a\(Z)W n-\ + ... + an-\(Z)W + aiZ),
with a;(Z) a polynomial in Z alone, and ao(Z) "" o. We may also as-
sume that n is positive, for otherwise F = bZ + c, and C is isomorphic
to C, given by the projection to the second factor. We will need a
little piece of algebra. Let Fw = aF / aw, which is a polynomial of

277
278 20. Riemann Surfaces and Algebraic Curves

degree n - 1 in W:

Fw = -iJF = n· ao(Z)W'-
1
+ (n - 1)· al(Z)W'-
2
+ ... + an-I(Z).
iJW

Lemma 20.1. There are polynomials B(Z, W), C(Z, W), and d(Z),
with d(Z) ¥- 0, so that
B(Z, W) . F(Z, W) + C(Z, W) . F w(Z, W) = d(Z).
Proof. We use the lemma of Gauss (Appendix C3): if F is irreducible
in iC[Z, W], then F is also irreducible in iC(Z)[W], where C(Z) is the
field of rational functions in Z. The equation of Lemma 20.1 can be
found (and computed) from the Euclidean algorithm in the ring
iC(Z)[W], as follows. Divide F by F w to get, after clearing denomi-
nators, an equation of polynomials
bo·F = QI·Fw+R "
where bo E iC[Z] , and R, is a polynomial of degree less than n - 1 in
W. If the degree of RI in W is positive, divide Fw by R 1 , getting an
equation
b,·Fw = Q2·RI +R2;
continuing, find equations b;· R;_I = Q;+I . R; + Ri+l> until for some k,
RH , has degree 0 in W. Note that R H , 0;6 0, since otherwise Rk would
divide RH , then Rk- 2, ... , and finally Fw and F, contradicting the
fact that F is irreducible.
Set d = RH I. To find an equation of the form required amounts to
showing that d is in the ideal in iC[Z, W] generated by F and F w; this
ideal contains R, by the first equation, then R2 by the second, and so
on, until finally it contains RHI = d. 0

If one takes an equation as in the lemma so that B, C, and d are

not all divisible by any nonconstant polynomial in Z, then d will be
unique up to multiplication by a constant; this d is called the dis-
criminant of F with respect to W. For our purposes any d will do.

Exercise 20.2. For F = W2 + b(Z)W + c(Z), show that

d(Z) = b(Z)2 - 4c(Z). For F = W 3 + b(Z)W + c(Z), show that
d(Z) = 4b(Z)3 + 27c(Z)2.

Let P be the first projection from C to iC: p(z, w) = z. We will show

that, if a finite number of points are removed, p becomes a covering
map.
20a. The Riemann Surface of an Algebraic Curve 279

Lemma 20.3. There is a finite subset S of C such that the projection

from C \ P-I(S) to C \ S is a finite covering with n sheets.
Proof. From Lemma 20.1 we draw the conclusion that if z E C and
d(z)"# 0, then there is no w with F(z, w) = 0 and Fw(z, w) = O. This
means that the equation F(z, W) = 0 has no multiple roots. If in ad-
dition ao(z)"# 0, then this equation has n distinct roots. We take S to
be the set where ao(z) . d(z) "# O.
Suppose Zo ft. S, and let WI, . . . , Wn be the roots of the equation
F(zo, W) = O. We want to find analytic functions gl, ... , g. defined
in a neighborhood of Zo so that gbo) = Wi and F(z, gi(Z» == O. One way
to do this is to apply the implicit function theorem, either for a subset
of 1R4 defined by two real equations, or the complex analogue for a
subset of C2 defined by one equation. Another is to use the Argument
Principle. For this, take small disjoint closed disks around these n
points, and let 'Vi be a counterclockwise path around the boundary of
the disk around Wi. For z near Zo and W on a circle 'Vi, F(z, w)"# 0 by
continuity. For z near Zo,

-
I f Fw(z,w)
dw = 1
21Ti 'Ii F(z, w) ,

since the integral is an integer that is I when z = Zo, and it varies

continuously with z. This means that there is exactly one root of the
equation F(z, W) = 0 inside 'Vi. In fact, from Problem 9.34, this root
is given by the formula

glz) = -
1 i w
Fw(z,w)
dw.
21Ti 'Ii F(z, w)
In particular, take U to be a disk around Zo where all these functions
gi are defined, and set Vi = {(z, gb»: z E U}. Since these points give
all possible roots of F(z, W) = 0 over z in U, we see that p-I(U) is
the union of open sets Vi' and p maps each Vi homeomorphically onto
V, with an inverse given by Zt-+(Z,gi(Z». 0

Now regard C C S2 = C U {co} as usual, and enlarge S by including

the point at infinity in it. By the lemma, we have a covering map
r~s2 \ S, with r = {(z, w): F(z, w) = 0 and z ft. S}. This covering
map gives r the structure of a Riemann surface, except that we must
prove it is connected:

Lemma 20.4. If F(Z, W) is irreducible, then r is connected.

Proof. Let yo be a connected component of r. If r is not connected,
280 20. Riemann Surfaces and Algebraic Curves

then the covering X" ~ S2 \ S restricts to a covering yo ~ S2 \ S with

m < n sheets (see Exercise 11.11). For each z not in S, let el(z),
e2(z) , ... , em(z) be the elementary symmetric functions in the m val-
n
ues of won the points in p-l(Z) YO. That is, el(z) is the sum of these
m values, e2(z) is the sum of all products of pairs of these values, and
so on until em(z) is the product of all m values. These functions ei are
clearly analytic on S2 \ S. They are in fact meromorphic on S2, as
follows from the fact that, for a in S, after multiplying by some (z - a)\
the roots w;(z) approach 0 (see Exercise 19.4). Consider the poly-
nomial
G = W" - el(Z)W"-1 + e2(Z)W"-2 - ... + (-l)mem(Z)
in C(Z)[W]. For any z not in S,
G(z, W) Il
PEp-'(z)nY"
(W-w(P)),

F(z, W) ao(z) Il
PEp-lIz)
(W - w(P» .

It follows that G divides F in C(Z)[W], since the remainder obtained

by dividing F by G would be a polynomial whose coefficients are
rational functions of Z vanishing on an infinite set. This contradicts
the irreducibility of F, and completes the proof. 0

By Proposition 19.9 this covering can be extended to a compact

Riemann surface X, together with an analytic map f: X ~ S2. This Rie-
mann surface is called the Riemann surface of the algebraic curve C
or of the polynomial F. It is easily seen to be independent of choice
of S. In fact, the following exercise shows more.

Exercise 20.5. A point P = (zo, wo) is called a nonsingular point of

C if either: (i) F w(zo , wo) ¥- 0; or (ii) Fz(zo, wo) ¥- 0; or both. (a) Show
that the canonical map from X" to C extends to an isomorphism of a
neighborhood in X with C in a neighborhood of P. (b) Show that the
ramification index of z: X ~ S2 at the corresponding point is 1 in case
(i), and the order of vanishing at Wo of the function w ~ F(zo , w) in
case (ii).

It is a theorem in analysis that every compact Riemann surface is

the Riemann surface of an algebraic curve. The problem for proving
this is to produce the meromorphic functions "z" and "w" on X. It
is not at all obvious that there are any nonconstant merom orphic func-
tions. For those that arise as branched coverings of S2 (which is the
20b. Meromorphic Functions on a Riemann Surface 281

same as producing one nonconstant meromorphic function z), it is not

obvious how to produce another not in (z); once one produces a
meromorphic function w that takes n distinct values on the n points
in X over some given point of S2, however, it is not hard to see that
w satisfies some irreducible equation F(z, w) == 0, and that X is the
Riemann surface of this polynomial.
Let us work out the example with F(Z, W) = W 3 + Z 3W + Z. By
Exercise 20.2, d(Z) = 4Z 9 + 27Z 2 , so the possible branch points are

°
where z = 0, z = 00, and the solutions of Z7 = -27/4. The points of C
over the finite points are nonsingular. Over there is one point (0,0),
and the ramification index of z: X ~ S2 is 3. Over each of the seventh
roots of -27/4 there are two points, so one of each must have ram-
ification index 2 and the other must be unramified. The point at in-
finity can be analyzed by making the substitution Z' = I/Z, but one
can see from the Riemann-Hurwitz formula that the sum of the num-
bers ez(P) - 1 for the points P over 00 must be odd, so there must be
two points over 00, with ramification indices 2 and 1. The genus gx
of the Riemann surface is given by
2gx - 2 = -2n + ~(eiP) - 1) = -6 + 2 + 7 . 1 + 1,
so the genus is 3.

Exercise 20.6. For each of the following polynomials, compute a set

S as in Lemma 20.3, compute the ramification indices of the points
over S in the Riemann surface, and compute the genus of the Riemann
surface: (i) W2 - n~_.(Z - ai), a., ... , am distinct complex num-
bers; (ii) 4W3 - 3Z 2W + Z3 - 2Z; (iii) W 3 - Z6 + I; (iv) W 3 - 3W 2 + Z6;
and (v) W" + zm + l.
Exercise 20.7. (a) If a compact Riemann surface X has a mero-
morphic function with only one pole of order 1, show that X is iso-
morphic to S2. (b) If X has a meromorphic function with two poles
of order 1, or one pole of order 2, show that it gives a two-sheeted
covering X ~ S2. (c) Show that, for a given set of 2g + 2 points in
S2, there is, up to isomorphism, exactly one two-sheeted covering that
branches at these points.

20b. Meromorphic Functions on a

Riemann Surface
The meromorphic functions on a Riemann surface X form a field,
which we denote by M(X) or just M. If X is the Riemann surface of
282 20. Riemann Surfaces and Algebraic Curves

the polynomial F(Z, W), then the functions Z and w (which come from
the two projections of C C C2 to the axes) are seen as in Lemma 20.4
to be meromorphic functions on X, as are any rational functions of Z
and w. Such rational functions form a subfield of M, which we can
denote by C(z, w).

Proposition 20.8. Every meromorphic function on X is a rational

function of Z and w:
M = M(X) = C(z, w)
= C(z) + C(z)· w + C(z)· w2 + ... + C(z)· w·- I .
Proof. Note first that any element in C(z)[w] can be written as a poly-
nomial of degree at most n - 1 is w, as seen by dividing by F. So it
suffices to show that any meromorphic function h on X is in C(z)[w].
From Lemma 20.1 we have an equation
d(z)·h = B(z,w)·F(z,w)·h+C(z,w)·Fw(z,w)·h
= C(z,w)·Fw(z,w)·h,
so it suffices to show that Fw(z, w)· h is in C(z)[w]. For z not in the
branch set S, let PI, . . . , Pn be the points of X over z. Then
•
F(z, W) = ao(z)· TI (W -
j=1
w(Pj »,
•
Fw(z, W) ao(z)·2:
i=1
TI (W - w(P) ,
i"'i

so, for 1 ::;; k::;; n,

h(Pk)· F w(z, w(P k» = ao(z)· h(Pk) . TI (w(Pk) - w(Pj ».

i"'k
Now consider the expression

2: h(P TI (T - w(P ».
n

ao(z) . i) • j
i=1 j"'i

On the one hand, as in Lemma 20.4, this can be written in the form
"i,~-=~bm(z)Tm with each b m meromorphic on S2, so bmE C(z). On the
other hand, the preceding calculation shows that
h(Pk )· Fw(z, w(P k » = ao(z)· h(P TI (w(P k )· k) - w(Pj »
2: bm(z)w(Pk)m .
n-l

=
m=O
20b. Meromorphic Functions on a Riemann Surface 283

This means that h· Fw(z, w) and Lbm(z)wm agree on the complement

of a finite set, which implies that they are equal. 0
Exercise 20.9. If Fizo, wo) ¥= 0, show that the ramification index of
z: X ~ S2 at the point corresponding to (zo, wo) is one more than the
order of the meromorphic function F w(z, w) at (zo, wo).

With Y = S2 and f X ~ Y the mapping given by z, and with

p: r ~ f" = Y \ S the covering space obtained by throwing away the
branch points, consider the three groups:
Aut(r If") {continuous '1': r~ r: p a 'I' = 'P};
Aut(XIY) = {analytic h: X~X:foh = h};
Aut(M(X)/C(z» = {field homomorphisms {): M(X)~M(X): {) is
the identity on C(z)}.
The first is topological, the second analytic, the third algebraic. We
claim that they are the same, after reversing the order of multipli-
cation in the third group:
Aut(r If") == Aut(XIY) == Aut(M(X)/C(z)rpp •
Exercise 20.10. Prove this by showing that every deck transformation
'I' extends uniquely to an analytic isomorphism h, and showing that
every automorphism of M(X) that is the identity on C(z) has the form
ft-+foh for a unique h. If you know some Galois theory, show that
M(X) is a Galois extension of C(z) if and only if the covering r of
f" is a regular covering. Show that the Riemann surfaces of two al-
gebraic curves are isomorphic if and only if their fields of mero-
morphic functions are isomorphic C-algebras. Show that for any finite
group G there is a Galois extension L of C(z) whose Galois group
is G.

In fact, for a given compact Riemann surface Y, to give a compact

Riemann surface X with a nonconstant analytic map from X to Y is
equivalent to giving a finite-sheeted topological covering r ~ f" of
the complement of a finite set, or to specify a finite extension M(X)
of the field M(Y) of meromorphic functions. In this setting, the sim-
ilarity seen in Proposition 13.23 between coverings and field exten-
sions is more than just an analogy.
In most expositions the Riemann surface of a polynomial is con-
structed, following Weierstrass, by starting with a germ of an analytic
function w(z) satisfying the equation F(z, w(z» = 0, and analytically
284 20. Riemann Surfaces and Algebraic Curves

continuing it around the plane. This approach, however, does not take
advantage of the fact that the algebraic curve C is already, except for
the modification and addition of a finite number of points, the desired
Riemann surface.

Problem 20.11. Carry out this construction, using the ideas of § 16b,
and show that it gives the same Riemann surface.

20c. Holomorphic and Meromorphic I-Forms

For any differentiable surface, as in Chapter 9, we can consider not

just real differentiable 1-forms but complex ones as well, where a
complex 1-form is given by WI + iW2, with WI and W2 real 1-forms.
Again we can consider closed and exact forms, with WI + iW2 being
closed (resp. exact) when each of WI and W2 is closed (resp. exact).
The corresponding group of closed complex 1-forms modulo exact
complex I-forms is denoted HI(X;C). It is a complex vector space,
which can be identified with HI(X)$iHI(X). If X is compact of ge-
nus gx, H I (X; C) is a complex vector space of dimension 2gx .
When X is a compact Riemann surface, and not just a differentiable
surface, there are some special closed complex I-forms, called holo-
morphic 1-forms. They are the I-forms that in local coordinates
'Pa: Ua~X, with Ua C C, have the form

!a dz = (u a + iva)(dx + i dy) = (u a dx - Va dy) + i(va dx + Ua dy),

where/o,(z)=!a(x+iy)=ua(x,y)+iva(x,y) is an analytic function. To
define a global I-form, using the notation of § 19a, we must have

d'PJ3a
where 'PJ3a I =- -
dz
is the complex derivative. As before, the Cauchy-Riemann equations
aua/ax = ava/ay and aua/ay = -ava/ax say that such a I-form is closed.
The holomorphic 1-forms form a complex vector space, sometimes
denoted 0.1.0(X), or just 0. 1.0. A holomorphic I-form W is exact pre-
cisely when there is an analytic function g on X with dg = w. In par-
ticular, if X is compact, every analytic function is constant, and so if
W is exact, then W = O. This means that the natural map from 0. 1,0 to
HI(X; C) is injective. We regard 0.1.0 as a complex subspace of HI(X; C).
20e. Holomorphie and Meromorphie I-Forms 285

There is also a notion of an antiholomorphic I-form. This is a I-form

that locally has the fonn

fa dz = (U a - iva)(dx - i dy) = (U a dx - Va dy) + i( -Va dx - Ua dy)

with fa = U a + iVa an analytic function as above. Again these form a
complex vector space, denoted OO.I(X) or 0°,1, and again these are
closed forms, and the only antiholomorphic fonns which are exact
are differentials of complex conjugates g of complex analytic func-
tions. Again, for X compact, we regard 0°,1 as a subspace of HI(X; C).
It follows readily from the definitions that no nonzero I-form can
be both holomorphic and antiholomorphic: 0 1,0 n 0°. 1= O. For any
surface X the space HI(X; C) has a complex conjugation operator, that
takes W = WI + iW2 to W = WI - iW2' This is not a complex linear map,
but is conjugate linear: it is linear as a map of real spaces, and for
a complex number c, cw = cwo Complex conjugation takes 0 1,0 to 0°. 1
and 0°,1 to 0 1.°. The following is one of the major "existence theo-
rems" about compact Riemann surfaces:

Theorem. For a compact Riemann surface X, HI(X; C) = Ol.°EBOO. I .

Equivalently, dim(OI,o) = dim(Oo, I) = gx.

In light of the preceding paragraph, the theorem is equivalent to

showing that there are gx linearly independent holomorphic I-fonns
on X. We will say nothing about the proof of this theorem for a gen-
eral compact Riemann surface, except to say that producing such 1-
forms is closely related to producing meromorphic functions, which
we have already discussed. For example, if WI and W2 are two in-
dependent holomorphic I-forms, then W2 =f· WI, where f is a non-
constant meromorphic function. When the Riemann surface comes
from an algebraic curve, however, we will prove this theorem in the
next chapter. If the curve has a sufficiently nice fonn, however, it
can be proved directly, as in the following problems:

Problem 20.12. Suppose F(Z, W) = L7=oa;(Z)W n - i is the polynomial

as in §20a, and assume: (i) the curve C is nonsingular, i.e., there are
no points (zo, wo) at which F, F w, and Fz all vanish; and (ii) the degree
of ai(Z) as a polynomial in Z is at most i, and if Ai is the coefficient
of Zi in ai' the equation L7=0 Ait n - i = 0 has n distinct roots. (a) Show
that X has n points over 00 E S2. (b) Show that h = F w(z, w) has a pole
of order n - 1 at each of these n points. (c) Show that the genus of
286 20. Riemann Surfaces and Algebraic Curves

X is (n - I)(n - 2)/2. (d) Show that

g(z, w)
{ -=-----dz: g is a polynomial of degree at
Fw(z,w) }
most n - 3 in z and w ,

which is a complex vector space of dimension gx.

The following problem generalizes this to allow the curve C to have

the simplest possible singularities: simple nodes. A node is a point
where the two partial derivatives vanish, but the Hessian
2 2 2
( a F)2 _ a F. a F
azaw ai aw2
is not zero. Geometrically, this means the curve C has two nonsin-
gular branches that cross transversely; in particular, there are two points
of the Riemann surface over a node of C. The corresponding picture
of a real curve (W 2 = Z2 + Z3) is:

Problem 20.13. Generalize the preceding problem to allow C to have

a certain number 8 of nodes, but no other singularities, but continue
to assume (ii). Show that the genus of X is (n - I)(n - 2)/2 - 8, and
that the holomorphic I-forms on X are exactly those that have the
form (g(z, w)/Fw(z, w»dz where g is a polynomial of degree at most
n - 3 in z and w that vanishes at the nodes of c.

It is an important fact from the theory of algebraic curves that, after

suitable algebraic transformations, every algebraic curve can be put
in the form considered in the preceding problem. Note that it is nec-
essary to allow singularities, since not every genus has the form
(n - l)(n - 2)/2. For a nice discussion of this, see Griffiths (1989).
20c. Holomorphic and Meromorphic I-Forms 287

One can also consider meromorphic I-forms on X, which can be

defined as for holomorphic I-forms, locally given by fadz, but with
fa only required to be meromorphic. If W is a meromorphic I-form,
the order of W at a point P, denoted ordp(w), is defined to be the order
of vanishing of fa at the corresponding point in a coordinate disk. For
example, when X = S2, and W = dz = _(Z,)-2 dz', with z' = liz, then
W has a pole of order 2 at the point at infinity. In general, we have

Proposition 20.14. For any nonzero meromorphic ljorm w on a

compact Riemann surface X,
L ordp(w) = 2gx - 2.
PEX

Proof. It is enough to prove the formula for one such w, since any
other has the form h . W for some meromorphic function h, and
L ordp(h· w) = L ordp(h) + ordp(w) = 0 + L ordp(w)
PEX PEX PEX

by Corollary 19.5. Assume that X comes equipped with a nonconstant

meromorphic function! X~S2. We take W = df= f*(dz). Near a point
P of X, where the mapping in local coordinates is t~ te, a mero-
morphic form on S2 with local expression g(z) dz pulls back to one
on X with local expression g(te)et e - I dt. Therefore ordp(w) =
(e/P) - 1) + e/P) ordf(p)(dz), and since the sum of e/P) for P map-
ping to the point at infinity is n, and ord.,,(dz) = -2,
L ordp(w) = L (ef(P) -
PEX PEX
1) + n(-2);

this is 2gx - 2 by the Riemann-Hurwitz formula. o

Another proof can be given by appealing to the result of Chapter
8. Given a meromorphic I-form W with local expressionfadz, define
a vector field V whose expression in the same local coordinates is
given by Va = Ilfa, i.e., if fa = U a + iVa, then

Exercise 20.15. Verify that these Va define a vector field V on X,

and that IndexpV = -ordp(w).

Proposition 20.14 therefore also follows from Theorem 8.3.

288 20. Riemann Surfaces and Algebraic Curves

Exercise 20.16. Reverse the argument in the above proof to give an-
other proof of the Riemann-Hurwitz formula for f: X ~ Y, under the
assumption that Y has a nonzero meromorphic I-form.

Define the residue of a meromorphic I-form w at a point P on a

Riemann surface X, denoted Respw, to be (l/27Ti)fy w, where"y is a
small counterclockwise circle around P not surrounding any point ex-
cept P where w is not holomorphic.

Exercise 20.17. (a) Show that this is well defined. (b) If z is a local
coordinate at P, and w =f(z)dz, with f(z) = ~~=_manzn, show that
Resp(w) = a_I.
Proposition 20.18 (Residue Formula). If w is a meromorphic I-form
on a compact Riemann surface X, then
2: Resp(w)
PEX
= o.

Proof. We know that we can realize X as a polygon II with sides

identified. From the construction we see that the map from II to X
can be taken to be differentiable, and, moving the sides slightly, we
may assume the image of each side is disjoint from the set of poles
of w. The I-form w then determines a closed I-form Wi on II, and
we must show that the sum of the integrals of Wi around small circles
around the poles of Wi is zero. Corollary 9.12 implies that this sum
is the same as the integral of Wi around the boundary of II, and this
integral vanishes since the integrals over the sides that get identified
in X cancel in pairs. D

The following problem gives another proof for Riemann surfaces

coming from algebraic curves:

Problem 20.19. (a) Prove the Residue Formula directly when X = S2.
(b) With z: X ~ S2 as in §20b, and w = fdz, withfE M(X), show that,
for QES 2 ,

2:
PEz-'(Q)
Resp(fdz) = ResQ(gdz) ,

where g in C(z) is the trace of the C(z)-linear endomorphism of M(X)

that is left mUltiplication by f. (c) Deduce the Residue Formula for
w =fdz from (a) and (b).
20d. Riemann's Bilinear Relations and the Jacobian 289

20d. Riemann's Bilinear Relations and the Jacobian

We have seen that the space nl.O(X) of holomorphic I-forms on X is
a subspace of the De Rham group HI(X; C). The way nl,o(X) sits in
HI(X; C) is important both in studying functions on X and in studying
the moduli of Riemann surfaces of given genus. Here we indicate how
some of this is related to the facts about homology that we proved in
Chapter 18. For this we use the pairing (00, v) == ffxwA v defined on
closed I-forms and their cohomology classes, but extended linearly
to those with complex coefficients as usual. When applied to holo-
morphic I-forms, there are two simple consequences of the definition:
(i) (00, v) == 0 if 00 and v are holomorphic; and
(ii) i· (00, w) > 0 if 00 is nonzero and holomorphic.

The first follows from the fact that dz A dz == O. For the second, if, in
local coordinates, 00 == fez) dz == f(z)(dx + i dy), then
i'wAw == i·lf(z)1 2(dx+idy)A(dx-idy) == 2If(zWdxAdy,

which is strictly positive wherever f is not zero, so its integral is pos-

itive.
Taking a basis ai' ... , ag , b l , • • • , bg for homology as in Chapter
18, and applying Exercise 18.8, we deduce, for 00 and v holomorphic
as above:

Proposition 20.20 (Riemann's Bilinear Relations).

(1) ~IiwLiv == ~IivLiw;

(2) i· ~ (Ii 00 Li Ii Li (0)
W- W > O.

Corollary 20.21. There is a unique basis WI, • . • , Wg for the space

of holomorphic Ijorms so that

1ai
Wk == { I if j == k,
0 otherwise.

Proof. This follows immediately from the fact that HI(X; C) is the
direct sum of nl,o(X) and no,I(X), together with the fact that inte-
grating over cycles gives an isomorphism HI(X; C) == Hom(HIX, C),
D
290 20. Riemann Surfaces and Algebraic Curves

The integrals of this basis over the other basis elements b l , ••• , bg
carry important information. Let

TOk
J, = 1 b.
WoJ' l~j,k~g.

This gives a (g x g)-matrix of complex numbers Z = (Tj,k), called the

(normalized) period matrix corresponding to the homology basis. This
matrix Z is far from arbitrary. It follows from what we have just seen
that Z is nonsingular. Moreover,

Corollary 20.22. (z) The matrix Z is symmetric, i.e., Tk.j = Tj,k' (ii)
The matrix Im(Z) whose entries are the imaginary parts of the entries
of Z is a positive-definite symmetric matrix.
Proof. (i) is an immediate consequence of the first of Riemann's bi-
linear relations, applied to the forms Wj and Wk'
(ii) follows from the second, applied to a form W = 'Lj= I tjWj, with
tj arbitrary real numbers, not all zero:

°< i· (w, (0) = i' 2:

j,k
tA(Tk,j - Tj.k) = 22:
j,k
t/k(lm(Tj,k)' 0
Of course, the period matrix is not unique, depending as it does
on the choice of a homology basis, but if another basis is chosen,
with the same intersection numbers, they will differ by a nonsingular
matrix with integral entries that preserves this intersection pairing.
As in the plane, integrals between two points on X are determined
up to these periods. Periods also playa role in a fundamental theorem
of Abel, which concerns the question of when one can find a mero-
morphic function on a given Riemann surface with zeros and poles
of given orders at given points.

Exercise 20.23. Show that for X = 8 2 , for any finite set of points Pi
and any given integers mj, provided 'Lmi = 0, there is a meromorphic
function f on X with ordpi(f) = mi for all i.

For g> 0, taking any basis WI, . • • , Wg of the holomorphic 1-

forms, we have a mapping from the homology group HIX to Cg given
by

It follows from what we have seen above that this map embeds HIX
in Cg as a lattice, i.e., the image of the basis elements ai, ... , bg
20e. Elliptic and Hyperelliptic Curves 291

fonn a basis for cg as a real vector space. If this lattice is denoted

by A, then the quotient space (and group) Cg I A is called the Jacobian
of X, and denoted J(X); it is homeomorphic to ~2g 17L2g , so to the
Cartesian product of 2g circles.
If P and Q are any two points in X, they detennine a point denoted
[Q - P] in J(X), by the fonnula

[Q - P] = (f WI, ••• , f Wg) E Cg I A = J(X) ,

where the notation means to integrate along any path from P to Q;

the resulting vector is defined up to an element in A. Similarly, given
any O-cycle D = 'imiPi of degree zero on X, one can define a point
[D] in the Jacobian by writing D = 'i(Qj - Pj) and setting [D] equal
to 'i[Qj - Pj]. Equivalently, fix a point Po in X, and define the Abel-
Jacobi mapping A: X ~J(X) by the fonnula A(P) = [P - Po]. Then
['imRJ = 'imiA(Pi).
Exercise 20.24. §how that this gives a well-defined homomorphism
from the group ZJ( of O-cycles of degree zero on X to J(X).

The map from HIX to Cg can be defined intrinsically, without

choosing a basis of holomorphic I-fonns, as the map

where n1,o(X)* is the dual space of complex-valued functions on n1,o(X).

This gives J(X) = nl,o(X)* IHI(X) , without choices.
The divisor Div(f) of a meromorphic function f is the O-cycle
'i ordp(f)P. In the next chapter we will prove Abel's theorem that a
zero cycle D = 'imRi is the divisor of a meromorphic function if and
only if its degree 'imi is zero and [D] is zero in J(X). Equivalently,
D is the boundary of a I-chain 'Y such that f'Y W = 0 for all holomorphic
l-fonns w. We will also l'rove the Jacobi inversion theorem that the
Abel-Jacobi map from ZJ( to J(X) is surjective.

20e. Elliptic and Hyperelliptic Curves

Before turning to the general situation it may be helpful to work out
the simplest nontrivial case in detail. Consider a Riemann surface of
the fonn CIA, where A is a lattice. By Exercise 19.17 we may choose
292 20. Riemann Surfaces and Algebraic Curves

such a lattice to be generated by I and 1", where 1" is in the upper half
plane. Realizing X by identifying the sides of the parallelogram

the images of the indicated sides can be taken as the standard basis
al and b l for HIX. Now X has a natural holomorphic I-form w, whose
pull-back to C is just the I-form dz, with z the standard coordinate
on IC. The integral of w along al is clearly 1, and the integral of w
along b l is 1". (Note that the corresponding period matrix is Z= (1"),
in agreement with Corollary 20.22.) The mapping from HIX to C = cg,
'Y I--? f'Y w, has image A generated by 1 and 1". Fix the point Po in X to
be the image of the origin in C. The Abel-Jacobi mapping from X
to CIA then takes a point P to [P - Pol = f:o w. By looking at this
mapping on the parallelogram spanned by I and 1", we see that this
mapping is an isomorphism, i.e., A: X ~ CIA.
Now suppose X is any compact Riemann surface of genus 1, which
we assume to have a nonzero holomorphic I-form w. We have the
mapping HIX ~ C taking a homology class 'Y to L w. The image is
a lattice A in C. If we take a basis ai, b l for HIX as usual, we can
take w so that fa, W = I, in which case A is generated by 1 and
1" = fbI w. As a very special case of Corollary 20.22 it follows that 1"
is in the upper half-plane. Fix a point Po in X.

Proposition 20.25. The Abel-Jacobi mapping

X ~ CIA, P I--? [P-Pol = (w,

Jpo
is an isomorphism of X with CIA.
Proof. Since the Abel-Jacobi mapping is analytic and nonconstant,
it is unramified by the Riemann-Hurwitz formula, and therefore a
covering map. We know from Chapter 13 that all finite coverings of
CIA are given by subgroups of the fundamental groups, which means
that X has the form Cj A', where A' C A is a subgroup of finite index.
20e. Elliptic and Hyperelliptic Curves 293

But we proved directly that for any X of this fonn, the Abel-Jacobi
map is an isomorphism. D

If a Riemann surface X is a two-sheeted covering of SZ with four

branch points, then by the Riemann-Hurwitz fonnula it has genus 1.
By changing the map by an automorphism of SZ one can take these
four branch points to be 0, 1, 00, and another AE C. It follows easily
(see below for a more general situation) that X is the Riemann surface
of the curve WZ = Z(Z - l)(Z - A). Since dz/w is a holomorphic 1-
fonn on this Riemann surface, we know from what we have just seen
that if we fix Po on X, the map

Pdz - lP dz
P~ l
Po w PoVz(z - l)(z - A)

gives an isomorphism from X to C/A. We will see that every Rie-

mann surface of genus 1 is isomorphic to those arising this way, at
least under the assumption that X comes from an algebraic curve.
Most complex analysis texts have a chapter on such "elliptic inte-
grals." (The Weierstrass ,p-function is used to find a two-sheeted
branched covering from C / A to Sz.)
We end this brief excursion by looking at the special case of Rie-
mann surfaces that can be realized as two-sheeted branched coverings
of the sphere. By the Riemann-Hurwitz fonnula, such a covering
must have an even number of branch points, namely, 2gx + 2. By
Exercise 20.7, there is only one two-sheeted covering of SZ with a
given set of an even number of branch points. If these are the points
a[, ... , am in C, possibly together with the point 00 (if m is odd),
this can be realized as the Riemann surface of the algebraic curve

TI (Z - aj)
m
W2 - = o.
j=[

These curves, and the corresponding Riemann surfaces, are called

hyperelliptic. All Riemann surfaces of genus 2 are in fact hyperellip-
tic, but for genus greater than 2, not all Riemann surfaces arise this
way.
The topology of a hyperelliptic surface can be seen directly, by
cutting slits in the sphere along arcs from a[ to az, a3 to a4, . . . , a2g+ [
to azg+z.
294 20. Riemann Surfaces and Algebraic Curves

The two-sheeted covering over the complement of these slits is trivial,

so the Riemann surface can be constructed as in the picture:

Exercise 20.26. If m = 2gx + 2, verify that the I-forms i dz/w, for

0:5 i:5 g - I, are holomorphic, and therefore give a basis for the hol-
omorphic I-forms.
CHAPTER 21

The Riemann-Roch Theorem

2Ia. Spaces of Functions and I-Forms

Fix a compact Riemann surface x, and let g = gx be its genus, M its
field of meromorphic functions, and 0 the space of meromorphic 1-
forms on X. A divisor D = L.mpP on X is just another word for a 0-
chain. That is, it assigns an integer mp to each point P in X, with
only finitely many being nonzero. We say that the order of D at P
is mp, and write ordp(D) = mp. The divisors on X form an abelian
group. As for O-chains, the degree of a divisor is the sum of the
coefficients: deg(D) = ~mp. If E = ~npP is another divisor, we write
E ~ D to mean that np ~ mp for all P in X. A divisor D is called
effective if each coefficient mp is nonnegative, i.e., D ~ o.
Any nonzero meromorphic function f on X determines a divisor
Div(f) = L. ordp(f) P .
Similarly, any nonzero meromorphic I-form W on X determines a di-
visor
Div(w) = L.ordp(w)p.

Corollary 19.5 and Proposition 20.14 say that

deg(Div(f» =0 and deg(Div(w» = 2g - 2.

Our goal in this chapter is to fmd meromorphic functions and I-forms

with prescribed, or at least controlled behavior. For example, we want
to find functions with poles only at certain points, and with the orders

295
296 21. The Riemann-Roch Theorem

of poles at these points not exceeding some bounds. For a divisor

D = L mpP on X let
L(D) = {fEM: ordp(f) ~ -mp for all P EX}
= {fEM: Div(f) + D ~O}.
This set of functions L(D) is a complex subspace of M. Similarly, let
.o.(D) = {w E.o.: ordp(w) ~ mp for all P EX}
= {w E.o.: Div(w) ~D},
a complex subspace of .0.. For example, .0.(0) is the space .0.1.0 of
holomorphic I-forms on X. Note that L(D) allows poles at the points
P where mp> 0, while .o.(D) requires zeros at the same points.

Lemma 21.1. (a) L(D) = 0 if deg(D) < 0, and .o.(D) = 0 if

deg(D) > 2g - 2.
(b) For any D and any point Q in X, L(D) CL(D + Q), and
.o.(D) :J .o.(D + Q). In addition,
dim(L(D + Q)/L(D»:5 1 and dim(.o.(D)/.o.(D + Q» :5 I .
(c) L(D) and .o.(D) are finite-dimensional vector spaces.
Proof. (a) follows from the fact that deg(Div(f» = 0 and
deg(Div(w» = 2g - 2. To prove (b), fix a local coordinate function z
at Q, and let m = ordQ(D). Any f in L(D + Q) has a local expression
h(z)/zm+l, with h holomorphic at O. The map which assigns h(O) to
f determines a homomorphism of complex vector spaces from L(D + Q)
to C, whose kernel is exactly L(D). This shows that either
L(D) = L(D + Q) or the quotient L(D + Q)/L(D) is one dimensional.
Similarly, any w in .o.(D) has a local expression h(z)~dz for h,hol-
omorphic, and assigning h(O) to w determines a map from .o.(D) to
C whose kernel is .o.(D + Q).
It follows from (b) that L(D) is finite dimensional if and only if
L(D + Q) is finite dimensional. Since one can get from any D to a
divisor of negative degree by a finite number of subtractions of a
point, the fact that L(D) is finite dimensional follows from (a). Sim-
ilarly for .o.(D), one can add points until the degree gets larger than
2g-2. 0

Exercise 21.2. If D:5 E, show that

dim(L(E» - dim(L(D» :5 deg(E) - deg(D).
One sees from the preceding proof that dim(L(D»:S deg(D) + 1.
21a. Spaces of Functions and I-Forms 297

For example, L(O) = C has dimension 1. If D = Q is a point, however,

we see from Exercise 20.7 that L(Q) is also C unless X=- S2.

Exercise 21.3. If X = S2, show that dim(L(D» = deg(D) + 1 when-

ever deg(D) ~ -1.

Lemma 21.4. (a) For any nonzero meromorphic functionf and any
divisor D,
dim(L(D» = dim(L(D + Div(f)).
(b) For any nonzero meromorphic I-form wand any divisor D,
dim(!1(D» = dim(L(Div(w) - D».
Proof. We have isomorphisms
L(D) ~ L(D + Div(f», h ~ h'f,
and
L(Div(w) - D) ~ !1(D) , h ~ h·w,
from which the lemma follows. D
Although it has been fairly easy to get an upper bound for the size
of L(D) , it is not so easy to get lower bounds, i.e., to show that there
must be functions with given poles. When X comes from an algebraic
a
curve, however, Proposition 20.8 gives first step, for at least one
divisor. Take z: X ~ S2 as in that proposition, and let E be the divisor
of poles of z, that is,
E = 2:
z(P)=OO
eiP)P.

This is a divisor of degree n on X, where n is the degree of the map-

ping z.

Lemma 21.5. For this divisor E, there is a constant k such that for
all integers m,
dim(L(mE» ~ deg(mE) + 1 - k = mn + 1 - k.
Proof. We need the following fact: for any meromorphic function h
on X, there is a nonzero polynomial p(z) in C[z] and an integer t so
that p(z)' h is in L(tE). To prove this, we need only find p(z) so that
p(z)' h has no poles outside E, and one sees that II(z - z(P»-OrdP(h) ,
the product over all P such that z(P)"# 00 and ordp(h) < 0, is such a
polynomial.
298 21. The Riemann-Roch Theorem

We saw in Proposition 20.8 (see Lemma C.19) that M is a vector

space over lC(z) of dimension n. By the fact proved in the preceding
paragraph, we can find a basis hi, ... , hn for Mover lC(z) and an
integer t so that each hi is in L(tE). Now for m = t + s, s ~ 0, the
(s + 1) . n functions t·
hi' 0 $, j $, s, 1 $, i $, n, are all in L(mE). This
means that, for such m, the dimension of L(mE) is at least
(m - t + 1) . n = mn + 1 - k for some constant k. Increasing k if nec-
essary, one may also achieve this inequality for the finite number of
m with 0 $, m < t. The inequality is automatic for m < 0 and any k ~ 0,
so the lemma is proved. 0
Lemma 21.6. There are integers k and N so that
dim(L(D» ~ deg(D) + I - k
for all divisors D on X, with equality
dim(L(D» = deg(D) + 1 - k if deg(D) ~ N.

Proof. Choose E as above, and define k to be the smallest integer so

that Lemma 21.5 holds for k. Suppose that D is a divisor on X such
that D $, mE for some integer m. It follows from Exercise 21.2 that
dim(L(D)) ~ deg(D) + dim(L(mE» - deg(mE) ~ deg(D) + 1 - k,
which proves the required inequality for such a divisor D. Given any
divisor D on X, there is a nonzero meromorphic function h such that
D - Div(h) $, mE for some integer m. Indeed, as in the preceding lemma,
one can take h to be II(z - z(p)trdp(D), the product over all P with
z(P) ¥- 00. Then by Lemma 21.4(a) and the result just proved,
dim(L(D» = dim(L(D - Div(h)))
~ deg(D - Div(h» + 1 - k = deg(D) + 1 - k.
By the minimality of k, there is some divisor Do such that the di-
mension of L(Do) is deg(Do) + 1 - k. From Exercise 21.2 it follows
that for any divisor D such that D ~ Do,
dim(L(D» $, + dim(L(D o» - deg(Do) $, deg(D) + I - k,
deg(D)
so dim(L(D» = deg(D) + 1 - k for any such D.
Let N = deg(Do) + k. If the degree of D is at least N, then the degree
of D - Do is at least k, so the dimension of L(D - Do) is at least
k + 1 - k = 1. There is therefore a nonzero function fin L(D - Do),
which means that D + Div(f) ~ Do, and so
dim(L(D» = dim(L(D + Div(!))
= deg(D + Div(f» + 1 - k deg(D) + 1 - k,
which proves the lemma. o
2Ib. Adeles 299

There can be only one integer k with the second property in the
lemma, so k depends only on the Riemann surface X. We will see in
§21c that k is the genus of X, and that N can be taken to be 2g - 2.

21b. Adeles
Lemma 21.1 (b) says that each of the two subspaces L(D) C L(D + Q)
and n(D + Q) C n(D) are either equalities or subspaces of codimen-
sion one. These cannot both be subspaces of codimension one, for if
w is in n(D) andfis in L(D + Q), thenf· w is a meromorphic I-form
with at most one simple pole at Q; the Residue Formula then implies
that ResQ(f' w) = 0, which means that f· w does not have a pole at
Q, and hence either w is in n(D + Q) orfis in L(D). We will even-
tually see that one of these inclusions is an equality exactly when the
other is not, and this is the core of the proof of the Riemann-Roch
theorem. What we will do in this section is to prove a kind of local
version of this assertion.
For a point P in X, let us denote by Mp the germs of meromorphic
functions at P. These germs are defined as in §16b, by taking equiv-
alence classes of meromorphic functions in neighborhoods of P, two
being equivalent if they agree on some (punctured) neighborhood of
P. If z is a local coordinate at P, any such germ has a unique power
series expansion ~;=_manzn. Iffis in M p , and w is a meromorphic 1-
form on X, the residue Resp(f' w) can be defined to be 1/2'ITi times
the integral of f' w around a small counterclockwise circle around P.
In local coordinates, f' w can be written ~bnzn dz, and this residue is
b- t •
Define an adele on X to be the assignment of a germfp of a mero-
morphic function at P for every point P in X, with the property that
ordp(fp);::: 0 (i.e.,Jp is holomorphic at P) for all but finitely many P.
We write f= (fp) for the adele defined by such a collection of func-
tionsfp. These adeles form a complex vector space, which we denote
by R. Any meromorphic function f on X determines an adele, by as-
signing the germ off at P to each P, so the field M of meromorphic
functions is a subspace of R. An adele can be thought of as a kind
of "discontinuous function" on X. Since there is no relation between
the "values" fp at different points of X, it is remarkable that they can
be a useful tool.
If f = (fp) is an adele, then Resp(fp' w) = 0 for all but finitely many
P (those wherefp or w has a pole). We can therefore add the residues
300 21. The Riemann-Roch Theorem

Resp(fp . U) over all P in X, getting a complex number. In other words,

U) defines a homomorphism

<p",: R ~ C, f = (fp) ~ 2: Resp(fp'

PEX
U),

which is a linear map of complex vector spaces. If D = LmpP is a

divisor such that U) is in O(D), and ordp(fp) ~ -mp for all P, then
fp' U) is holomorphic at P, so the residue is zero. Define R(D) C R by
the formula
R(D) = {f = (fp) E R: ordp(fp) ~ -ordp(D) for all P EX}.
This means that the homomorphism <p", vanishes on R(D). In addition,
the Residue Formula says that if these fp all come from one mero-
morphic function f on X, then LpEX Resp(f' U) = 0, so <p", also van-
ishes on the subspace M of R. It follows that <p", determines a ho-
momorphism (still denoted <p.,)
<p",: R/(R(D) + M) ~ C.
Define SeD) to be this complex vector space R/(R(D) + M), and de-
fine O'(D) to be the dual space
O'(D) = S(D)* = HomdR/(R(D) + M), C).
Then 'P., is an element of this space O'(D). What we have done is
construct a natural homomorphism from OeD) to O'(D), taking U) to
<p.,. In the next section we will show that this homomorphism is an
isomorphism.

Exercise 21.7. Show that the homomorphism from OeD) to O'(D) is

injective, and that a meromorphic I-form U) is in OeD) if and only if
<p", vanishes on R(D).

In this section we prove that O'(D) has some properties we would

like O(D) to have. As in the preceding section, the main idea is to
compare O'(D) and O'(D + Q) for Q a point in X.
Since R(D) is contained in R(D + Q), there is a canonical surjection
from R/(R(D) + M) onto R/(R(D + Q) + M), i.e., from SeD) onto
SeD + Q). The kernel is (R(D + Q) + M)/(R(D) + M).

Lemma 21.8. For any divisor D and point Q,

dim«R(D + Q) + M)/(R(D) + M» ~ 1,
with equality if and only if L(D) = L(D + Q).
21b. Adeles 301

Proof. Choose a germ gQ at Q such that ordQ(gQ) = -ordQ(D) - 1,

and let gp = 0 for P ~ Q. The adele g = (gp) gives a generator of the
quotient space in the lemma. Indeed ifJ= (fp) is any adele in R(D + Q),
then there is some scalar A so that J - Ag is in R(D). This element
will be nonzero exactly when there is no h in M with g - h in R(D),
which says exactly that there is no h in L(D + Q) that is not in
L~. 0

Let k be the integer from Lemma 21.6.

Lemma 21.9.
(a) IfD is a divisor such that dim(L(D» = deg(D) + 1 - k, then
R(D) + M = R, so SeD) = 0 and fl'(D) = O.
(b) For any D the space SeD) has finite dimension, so its dual space
fl'(D) has the same finite dimension.
(c) For any divisor D and point Q, fl'(D + Q) is a subspace oj fl'(D) ,
and
dim(fl'(D)/fl'(D + Q» :::; 1,
with equality if and only if L(D) = L(D + Q).
(d) For any nonzero meromorphic Junction J on X,
dim(fl'(D + Div(f))) :;: dim(fl'(D».
Proof. If dim(L(D» = deg(D) + 1 - k, then for any point Q we know
that dim(L(D + Q» = deg(D + Q) + 1 - k (see Exercise 21.2), i.e.,
L(D + Q) ~ L(D). By the preceding lemma, this means that
R(D + Q) C R(D) + M. Continuing to add points to D + Q, we see
that R(E) C R(D) + M for all divisors E such that E ~ D. But any ele-
ment of R is in R(E) for some such E, so R = R(D) + M, which
proves (a).
For (b), take a sequence of surjections S(D)-S(D + QI)_
SeD + QI + Q2)- ... -SeE), until E is large enough so (a) implies
that See) is zero. Lemma 21.8 implies that the kernel of each of these
surjections is at most one dimensional, so by induction each SeD)
must be finite dimensional.
Dual to the exact sequence
o~ (L(D + Q) + M)/(L(D) + M) ~ SeD) ~ SeD + Q) ~ 0
is the exact sequence
o~ fl'(D + Q) ~ fl'(D) ~ «L(D + Q) + M)/(L(D) + M»* ~ O.
This shows that the inclusion fl'(D + Q) ~ fl'(D) is either an iso-
302 21. The Riemann-Roch Theorem

morphism or its cokemel has dimension one. By the preceding lemma,

we see that the latter occurs exactly when L(D) = L(D + Q), which
proves (c).
For (d), there is a natural isomorphism from R(D + Div(f)) to R(D)
that takes f to f· f. This determines an isomorphism from S(D + Div(f))
to S(D), and, taking duals, from O'(D) to O'(D + Div(f)). 0

Lemma 21.10. For any divisor D on X,

dim(L(D)) = deg(D) + 1- k + dim(O'(D)).
Proof. This equation is certainly true if deg(D) ~ N, with N as in
Lemma 21.6, for then dim(L(D)) = deg(D) + 1 - k and dim(O'(D)) = 0
by Lemma 21.6 and Lemma 21.9(a). Since we can get between any
two divisors by successively adding and subtracting points, it suffices
to show that the equation is true for a divisor D if and only if it is
true for D + Q, where Q is any point. Comparing the two equations,
what must be proved is that
dim(L(D + Q)) - dim(L(D)) + dim(O'(D)) - dim(O'(D + Q)) = 1,

and this is simply a translation of (c) in the preceding lemma. 0

Let 0' be the union of all O'(D), taken over all divisors D. An
element of 0' is a homomorphism from R to C which vanishes on M
and vanishes on some (unspecified) R(D). The space n of mero-
morphic differentials on X maps to n', and we want to see that this
is an isomorphism. We have seen that if w is any nonzero mero-
morphic differential, any other can be written in the form f· w for
some meromorphic functionJ. This means that n is a one-dimensional
vector space over the field M. The space 0' is also a vector space
over M, by the rule that if f is in M and '1': R~ C, then f· 'I' is the
homomorphism which takes f to 'P(f' f).

Lemma 21.11. The dimension of 0' over M is 1.

Proof. We know that 0' is not zero, for example, by applying the
preceding lemma for D of small degree to see that O'(D) ~ O. To
complete the proof we must show that two elements 'I' and I\! in 0'
cannot be independent over M. If this were the case, then for any
elements hi, ... , hn of M which are independent over C, it would
follow that hl'P, ... , hn'P, hll\!, ... , hnl\! are elements of 0' which
are independent over C. Take a divisor E which is large enough so
that 'I' and I\! are in O'(E). Suppose the functions hi are a basis for
L(D) for some D. Then the 2n products hi'P and hil\! are in O'(E - D),
2Ic. Riemann-Roch 303

so
dim(!l'(E - D» ~ 2dim(L(D» ~ 2(deg(D) + 1- k).
By Lemma 21.10,
dim(O'(E - D» = k - 1 - deg(E - D) + dim(L(E - D».
Now deg(E - D) = deg(E) - deg(D), and L(E - D) = 0 provided
deg(E - D) < O. So if we take any D with deg(D) > deg(E), the dis-
plays lead to the inequality
2(deg(D) + 1 - k) s; deg(D) - deg(E) + k - 1 ,
which says that deg(D) s; 3k - 3 - deg(E). But we may take D of
arbitrarily large degree, which is a contradiction. D

The canonical homomorphism from the space 0 of meromorphic

differentials to the space 0' is a homomorphism of vector spaces over
the field M. It is not identically zero by Exercise 21.7, since one can
certainly find meromorphic differentials wand adeles f such that there
is exactly one point P at which Ip . w has a simple pole. But since
both vector spaces have dimension one over M, it follows that the
map O~ 0' is an isomorphism. We saw at the beginning that the
subspace O(D) of 0 is mapped into O'(D) by this map, and it follows
from Exercise 21.7 that O(D) maps isomorphic ally onto O'(D). So
we have proved

Proposition 21.12. The canonical map O(D)~O'(D) is an iso-

morphism.

Exercise 21.13. Given germs 11, ... ,In of meromorphic functions

at distinct points PI> ... , Pn of X, and integers ml> ... , mn , show
that there is a meromorphic function I so that ordpi(f - /;) ~ mi for
1 s; i s; n.

2Ie. Riemann-Roeh
From Lemma 21.10 and Proposition 21.12 we have the formula
(*) dim(L(D» = deg(D) + 1- k + dim(O(D»
for all divisors D on X, valid for some integer k which we do not yet
know. We can specialize (*) to some cases where we know some-
304 21. The Riemann-Roch Theorem

thing. For example, if we take D = 0, L(O) = C, and the formula says

that I = 0 + I - k + dim(O(O», i.e., that the space of holomorphic
I-forms has dimension k. Fix any meromorphic divisor 00, and let K
be the divisor of 00, which we know is a divisor of degree 2g - 2.
Applying Lemma 21.4(b) with D = K, we get
dim(O(K» = dim(L(K - K» = dim(L(O» = I,
and applying the same lemma with D = 0, we get
dim(L(K» = dim(O(O» = k.
Now apply (*) with D = K, yielding
k = (2g - 2) + I - k + 1 ,
which means that k must be g. So we have proved:

Theorem 21.14 (Riemann-Roch Theorem). If X is the Riemann sur-

face of an algebraic curve, then for any divisor D on X,
dim(L(D» = deg(D) + I - g + dim(O(D»
= deg(D) + I - g + dim(L(K - D»,
where K is the divisor of any nonzero meromorphic Ijorm on X.

Corollary 21.15. The space of holomorphic differentials has dimen-

sion g.

This proves that H'(X;C) = OI,O(X)EBOO,I(X).

Corollary 21.16. dim(L(D» ~ deg(D) + 1 - g, with equality when-

ever deg(D) ~ 2g - I.

Corollary 21.17. For any two points P and Q on X, there is a mero-

morphic Ijorm cp with simple poles at P and Q and no other poles.
Proof. Riemann-Roch for D = -P - Q gives
o= -2+ l-g+dim(O(-P-Q»,
or dim(O( -P - Q» = g + 1. This means that there is a meromorphic
1-form 'P that is in O( - P - Q) but not in 0(0). Since the sum of the
residues is zero, 'P must have simple poles at both P and Q, with the
residue at Q being minus the residue at P. D

Corollary 21.18. If gx = 0, then X is isomorphic to S2.

Proof. Take any point P. Since dim(L(P» ~ deg(P) + 1 - g = 2, there
21e. Riemann-Roch 305

is a nonconstant meromorphic function f with at most one pole. This

is a mapping from X to S2 of degree 1, so is an isomorphism. 0
Exercise 21.19. If gx = 1, show that there is an analytic mapping
f: X ~ S2 of degree 2. Deduce that X is the Riemann surface of a
curve W2 = Z(Z - l)(Z - A), A~ 0, 1.

Exercise 21.20. Assume that g = gx ~ 1. (a) Show that there are g

distinct points PI, ... ,Pg on X so that a(P1 + ... + P g ) = O.
(b) Show that there are points PI,..., P g on X so that
a(P1 + ... + Pg ) ~ O. (c) If g ~ 2, show that there is an analytic
mapping f: X ~ S2 of degree at most g. In particular, if g = 2, X is
hyperelliptic.

Exercise 21.21. Show that the I-form 'P of Corollary 21.17 is unique
up to multiplying by a nonzero scalar and adding a holomorphic 1-
form. Show that there is a unique such 'P whose residue at P is 1,
whose residue at Q is -1, and so that I -y 'P is purely imaginative for
all closed paths 'Y in X \ {P, Q}.

Exercise 21.22. A real-valued function u on a Riemann surface is

harmonic if it is locally the real part of an analytic function. A func-
tion which is harmonic in a punctured neighborhood of a point P is
said to have a logarithmic pole at P if, with z a local coordinate at
P, there is a nonzero real scalar a so that u - a . 10g(lzl) extends to be
harmonic in a neighborhood of P. Show that for any two points P
and Q on X, there is a harmonic function u on X \ {p, Q} that has
logarithmic poles at P and Q.

Exercise 21.23. For any point P on X, show that there is a mero-

morphic differential 'P on X with a double pole at P. Deduce that there
is a harmonic function u on X \ {P} so that, if z = x + iy is a local
coordinate at P, then u - x/(x2 + l) is harmonic near P.

Historically, the arguments went in the reverse order: harmonic

functions were found with the properties in the preceding exercises,
and these were used to find meromorphic I-forms and to prove Rie-
mann-Roch. By regarding harmonic functions as integrals of fluid
flows or electric fields on X, one can give intuitive arguments for
their existence, say by putting sources and sinks at the points P and
Q. For a lively discussion along these lines, see Klein (1893).

Exercise 21.24. Given any sequence Ph P 2 , ••• ,Pn , ••• of points

306 21. The Riemann-Roch Theorem

in X, show that there are exactly g positive integers k, all in the in-
terval [1, 2g - 1], such that
L(P , + ... +Pk-I) = L(p, + ... +Pk ).
When all Pi are taken to be a fixed point P, these integers are called
the Weierstrass gaps at P.

Exercise 21.25. Suppose D and E are divisors on X such that D + E

is the divisor of a meromorphic I-form. Prove Brill-Noether rec-
iprocity: dim(L(D» - dim(L(E» = If2(deg(D) - deg(E».

Exercise 21.26. If z: X ~ S2 is an analytic mapping of degree n, and

Q in S2 is a point such that z-I(Q) = {P I, • • . , Pn} has n distinct
points, use Riemann-Roch to show that there is a meromorphic func-
tion w on X so that w(P I), ... , w(P n) are distinct complex numbers.
If F(z, w) = 0 is the irreducible equation for w over (:(z) (with de-
nominators cleared), show that X is isomorphic to the Riemann sur-
face of F.

Once one knows Riemann-Roch for a general compact Riemann

surface X, the preceding exercise shows that X comes from an alge-
braic curve.

Exercise 21.27. Show that the Riemann surface X of the polynomial

W4 + Z4 - 1 = 0 has genus 3, but X is not hyperelliptic. Show in fact
that dimO(2P) = 1 for all P in X.

2Id. The Abel-Jacobi Theorem

In this section we prove the assertions made in §20d. The first is
Abel's criterion for when a divisor D = '2:.miPi is a divisor of a mero-
morphic function: it must have degree zero and be in the kernel of
the Abel-Jacobi map. We use the notation of that section.

Theorem 21.28 (Abel's Theorem). There is a meromorphic function

f on X with Div(f) = D if and only if deg(D) = 0 and [D] = 0 in J(X).

We first sketch the proof of the necessity of these conditions. Sup-

pose f is a nonconstant meromorphic function on X, giving a mapping
of degree n from X to s2, with branch set S. Corollary 19.5 shows
2Id. The Abel-Jacobi Theorem 307

that deg(Div(f» = O. Fix a point Po in X. Consider the mapping from

S2 \ S to J(X) that takes a point Q to the point L7=1 [Pi - Po], where
PI ... , Pn are the points of X in rl(Q). It is not hard to see that
this extends continuously to the branch points, giving a continuous
mapping from S2 to J(X). Since S2 is simply connected, by Propo-
sition 13.5 this mapping factors: S2~ Cg ~ U / A. By looking locally,
one can verify that each of the g coordinate maps are analytic func-
tions on S2. But analytic functions on S2 are constant, so the given
map from S2 to J(X) must be constant. The fact that the value at 0
is equal to the value at 00 is precisely the condition that
[Div(f)] = [L ordp(f) . P] = 0 in J(X).
We tum now to the converse. Let D be a divisor of degree zero in
the kernel of the Abel-Jacobi map. We must show that D is the divi-
sor of a meromorphic function. The motivation for the proof comes
from the fact that if J is a meromorphic function on X, and we set
'P = dJ/J, then'P is a meromorphic differential on X with at most sim-
ple poles and, in fact, for any P, Resp('P) = ordp(f). We will look for
a meromorphic I-form 'P with at most simple poles among the points
appearing in D, such that Resp('P) = ordp(D) for all P. Then we will
define a function J on X by the formula J(P) = exp(f~o 'P), where Po
is a fixed point. Provided this is well defined, it will satisfy the equa-
tion 'P = dJlJ, and so we will have ordp(j) = ordp(D) for all P, so
Div(f) =D.
Since the degree of D is zero, we may write D = L~=I(Pi - Q;), for
some points PI,"" Qr (not necessarily unique). Let
S={P., ... ,P"Q., ... ,Qr}' We know by Corollary 21.17 that
there is a meromorphic I-form 'Pi with simple poles at Pi and Qi (only),
and with residues I at Pi and -1 at Qi' Let 'P = ~~=I 'Pi' We want to
defineJ(P) = exp(f~o'P), where the integral is along any path from Po
to P in X \ S. This will be well defined provided the integral of 'P
along any closed path T in X \ S is in 2-rrilL. The form 'P is only defined
up to the addition of a holomorphic I-form, so the proof of the Abel-
Jacobi theorem is reduced to the

Claim 21.29. There is a holomorphic I{orm w so that fA(fI- w) is

in 2 TrilL Jor all I-cycles T on X \ S.

We need the following refinement of Exercise 18.8. We take 2g

closed arcs aj and bj as in §17c. Cutting the surface open along these
arcs, we realize it as the identification space of a plane polygon II
with sides identified. These choices can be made so that the map from
II to X is a diffeomorphism on the interior of II, and has a ~"" exten-
308 21. The Riemann-Roch Theorem

sion to a neighborhood of II. By means of this map I-forms on X

correspond to I-forms on II. Let 00 be a closed I-form on X, and
define a function h on the closure of II by the formula h(P) = f~ow,
for some fixed point Po in II. Let 'I' be a closed I-form defined on a
neighborhood of the union of these 2g arcs in X, so 'I' determines a
I-form on a neighborhood of the boundary all of II.

Lemma 21.30.

Proof. Note first that if P and p' are corresponding points of the
boundary edges aj and aj -I of II, then

h(P) - h(P' ) = -L 00 ,
}

as one sees by integrating along a path from p' to P, noting that the
integrals over corresponding parts of aj and aj -I cancel.

Therefore

Lh'P+ L_,h'P L (h(P) - h(P I » 'P(P)

J J J

( (- Jb( (0)'1'
Jaj
= - (
Jbj
w· ( 'I'
Jaj
j

Similarly if Q and Q' are corresponding points of bj and bj - I , then

h(Q) - h(Q') = Lw,

J

so

Lh'P+L_,h'P L (h(Q) - h(Q'» 'P(Q)

J J J
2id. The Abel-Jacobi Theorem 309

Adding over all the edges of the boundary of IT gives the identity of
~~~. 0
To apply the lemma in our situation, the arcs aj' bj must be taken
so that none of them goes through a point of S. Let U j be a path from
Qj to P j which does not hit any of the arcs aj or bj .

Lemma 21.31. For any holomorphic Ijorm won X,

Proof. We apply the preceding lemma. We must evaluate Ianh<p, with

h(P) = I~o w. The Residue Formula in the polygon IT gives

in h<p = (21Ti) ~ Resp(h<p) = (21Ti)(~ h(P;) - ~ h(Q;)) ,

the last since <pj has simple poles, with residues 1 and -1 at P j and
Qj. And

by definition. So Lemma 21.30 gives the required conclusion. 0

The hypothesis that D is in the kernel of the Abel-Jacobi map means

that there is a I-cycle 'Y such that L~= I I!Ii W = I~ w for all holomorphic
i-forms w. So we have, for all holomorphic w,

We can now prove Claim 21.29. Since a basis of H,(X \ S) is given

by the cycles aj' bj , and small circles around the points in S (either
by Mayer-Vietoris or Problem 17.12), it suffices to find a holo-
morphic I-form w such that the integral of <p - w around all these
cycles is in 21TiZ. Note that for any such <p, the integral around a
small circle around a point in S is in 21TiZ, since all the residues of
<p are integers. To start, let 'A.j = I aj <p. Subtracting the holomorphic
1-form L'A.j Wj from <p, we can assume that I aj <p = 0 for all j. The pre-
ceding formula, with w = Wk, gives

(21Ti) f
"Y
Wk = ( <p.
Jb,
310 21. The Riemann-Roch Theorem

Write 'Y = "J.mA + njbj , with integer coefficients mj' nj. Then

L ook = ~ (mj Ii ook + nj ii ook)

ii ik
g g

= mk + ~ nj ook = mk + ~ nj ooj.

the last step by the symmetry of Corollary 20.22. But now if we set
00= 27Ti"J.njooj, this shows that fbk (IP - 00) = 27Timk. Therefore

1ale
(IP-OO)= -1 ale
00 = - 27Ti l2:
ak
njooj= - 27Tink,

and this completes the proof of the claim, and hence of Abel's theorem.

Theorem 21.32 (Jacobi Inversion). The Abel-Jacobi map from the

group of cycles of degree zero to the Jacobian J(X) is surjective.

This means that we have an exact sequence

°~ C* ~ M(X)* ~ Zo(X) ~ J(X) ~ 0,
which realizes the torus J(X) = Cg / A as the quotient of the group of
divisors of degree zero by the subgroup of divisors of meromorphic
functions.
The proof, which we only sketch, requires a few basic facts about
holomorphic mappings of several complex variables. Let Xg be the g-
fold Cartesian product of X with itself, which is a g-dimensional com-
plex manifold. Let
U g : xg ~ J(X)
be the map which takes (PI, ... , P g ) to A(LP j - gPo), where A is
the Abel-Jacobi map, and Po is any fixed point on X. It suffices to
prove that u g is surjective. The Jacobian J(X) = P / A gets the struc-
ture of a complex manifold so that the quotient mapping Cg ~ J(X)
is a local isomorphism.

Exercise 21.33. Verify that u g is a holomorphic (analytic) mapping

of complex manifolds.

We claim next that there are distinct points PI, . . . , P g in X such

that the Jacobian determinant of u g at the point (PI' ... , P g ) is not
zero. Once this is verified, it follows that the image of u g contains
an open set. Since the image of the Abel-Jacobi map is a subgroup,
21d. The Abel-Jacobi Theorem 311

and J(X) is compact, it follows that the image must be all of J(X).
To prove the claim, take any distinct points PI, . . . , Pg' Let Zj be
a local coordinate at Pj , and write Wi =/;,j dZj near Pj .

Exercise 21.34. Verify that, in suitable coordinates, the Jacobian ma-

trix of cxg at (PI, ... , Pg ) is (/;,j(Pj)),

Now take the points PI, ... , Pg as in Exercise 21.20(a). If the

Jacobian determinant det(/;.j(P)) of cxg vanishes at (PI, ... ,Pg ),
then there are g complex numbers AI, ... , Ag , not all zero, such
that
(AIWI + ... + AgWg)(P) = 0
for all j. But this means that n(P I + ... + Pg) is not zero, which
contradicts Exercise 21.20(a).
 PART XI

HIGHER DIMENSIONS

This last part is designed to introduce the reader to a few of the higher-
dimensional generalizations of the ideas we have studied in earlier
chapters, both to unify these ideas, and to indicate a few of the di-
rections one may go if one continues in algebraic topology. It is not
written as the culmination or goal of the rest of the course, but rather
as a brief introduction to the general theory. How accessible or useful
it may be depends on several factors, such as background in manifold
theory, and ability to generalize from the special cases we have seen
to higher dimensions (an ability, it seems to me, often underestimated
in our teaching). For systematic developments of the ideas of this
part, the books of Bredon (1993), Bott and Tu (1980), Greenberg and
Harper (1981), and Massey (1991) are recommended.
We have studied the first homology group H1X, the fundamental
group 'lTl(X,X), and the first De Rham cohomology group H1X, which
were sufficient to capture most of the topology of the spaces we have
been most concerned with: open sets in the plane and surfaces, and
an occasional graph. Each of these is the first in a sequence of groups
that are used to study similar questions for higher-dimensional spaces.
In contrast to many earlier chapters, the tone in this final part is de-
signed to be more formal, concise, and abstract; we are depending
on your experience with special cases and low-dimensional examples
for motivation.
We start by recalling some three-dimensional calculus, to indicate
the sort of "topology" these higher groups might measure. Then we
take a quick look at knots in 3-space, mainly because knot theory is
314 Part XI. Higher Dimensions

an interesting and important subject in its own right, and also because
it gives us a chance to use some of the tools developed in earlier
chapters. We also define the higher homotopy groups of a space and
the De Rham cohomology groups.
In the next chapter we define higher homology groups, and prove
their basic properties. We indicate some of the ways they can be used
to extend ideas we have looked at in the plane or on surfaces to higher
dimensions. In particular, they give a simple extension of the notion
of degree, and they lead to generalizations of the Jordan curve theo-
rem.
In the final chapter we include a couple of "diagram chasing" facts
from algebra, one of which can be used to compare different ho-
mology and cohomology groups, the other to construct long exact
sequences such as Mayer- Vietoris sequences. (Having proved this
once and for all, one does not have to keep doing the same sort of
manipulations we have done to define boundary and coboundary maps.)
Finally, we give proofs of the basic duality theorems between ho-
mology and De Rham cohomology on manifolds.
The sections involving higher De Rham cohomology on manifolds
are written for those with a working knowledge of differential forms
on differentiable manifolds. A reader without this background can
skip or skim this part (or stick to low dimensions and/or open sets
in ~n). The construction of higher homology groups and its appli-
cations to higher-dimensional analogues of the theorems we saw at
the beginning of the course do not depend on any of this, however,
and a reader who has mastered the earlier chapters should be able to
work through this without any gaps. We have included Borsuk's theo-
rem that maps of spheres that preserve antipodal pairs have odd de-
gree, since this allows generalizing all the results we proved about
winding numbers to higher dimensions.
A few remarks about other approaches may be in order. It is pos-
sible to define the degree of a mapping of a sphere to itself without
the notions of homology, and to prove many of its properties. Those
with a knowledge of differential topology can do this by approxi-
mating a continuous map by a differentiable one, and following the
pattern of Problem 3.32 for the winding number. (For a nice discus-
sion in the ~oo context, see Milnor (1965)). There are also elementary
approaches using simplicial approximations, although considerable care
is required to make the arguments rigorous. (In fact, the difficulties
in this approach seem to us much greater than those involved in de-
veloping general homology theory-not to mention the fact that, hav-
ing done the latter, one can apply it in many other situations.)
Part XI. Higher Dimensions 315

We might also mention that we have used cubes rather than the
simplices that are common in many other treatments. Simplices have
a slight advantage that one has no "degenerate" maps to ignore, but
cubes are simpler for homotopies and product spaces in general, and
they are more convenient for integrating differential forms. (In fact,
it is not hard-as we do in §24d-to use cubes to calculate the sim-
plicial homology of spaces that are triangulated.)
Finally, a word on terminology: here manifolds always are assumed
to have a countable basis for their topology.
CHAPTER 22

Toward Higher Dimensions

22a. Holes and Forms in 3-Space

On an open set U in [R3 we have: O-forms, which are just C(;oo functions
on U; Ijorms, which are expressions
pdx + qdy + rdz,
where p and q and rare C€oc functions on U; 2-forms, which are expres-
sions
udydz + vdxdz + wdxdy,
where u and v and w are C(;oo functions on U; and 3-forms, which are
expressions
hdxdydz,
where h is a C(;oo function on U.
These fonns are designed for integrating, just as in the plane. A 0-
fonn is evaluated at points. The integral of a I-fonn over a differ-
entiable path 'Y: [a, b] ~ U is defined exactly as for the plane:

f"I
pdx + qdy + rdz = ia
b
(P('Y(t» dx + q('Y(t» dy + r('Y(t» dz) dt,
dt dt dt
where 'Y(t) = (x(t) , y(t), z(t». A 2-fonn can be integrated over a dif-

317
318 22. Toward Higher Dimensions

ferentiable map f: R-,; U from a rectangle R = [a, b] x [c, d] to U:

II
r
udydz + vdxdz + wdxdy =

If( a(y,z) a(x,z) a(x,y))

u(f(s, t)) - - + v(f(s, t)) - - + w(f(s, t)) - - " ds dt ,
a(s, t) a(s, t) a(s, t)
R

where, for [(s,t) = (x(s, t),y(s,t), z(s,t)), and a(x,y)/a(s,t) denotes

ax/as ay/at- ax/at ay/as, and similarly for the other terms. A 3-
form is integrated over a differentiable map II: B -'; U where B is a
rectangular box [a, b] x [c, d] X [e,f]:

Iff hdxdydz = Iff h(II(s, t, u))

a(x,y,z)
a(s, t, u)
dsdtdu,
II B

where II(s, t, u) = (x(s, t, U),y(s, t, U),z(s, t, u)), and a(x,y,z)/a(s,t,u)

denotes the Jacobian determinant.
The differential df of a O-form f is a I-form defined by
af af af
df = -dx+-dy+-dz.
ax ay az
The fundamental theorem of calculus gives I'/df= f('Y(b)) - f('Y(a))
for'Y a path as above. The differential of a I-form is a 2-form:
d(pdx + qdy + rdz)

= (ar _ aq ) dydz + (ar _ ap ) dxdz + (a q _ ap ) dxdy,

~ ~ b ~ b ~

and Green's theorem (for rectangles in the plane) gives II rdw = Iarw,
where the integral around the boundary of the rectangle is defined as
in Part 1. Finally, the differential of a 2-form is a 3-form:

d(udydz + vdxdz + wdxdy) = (au _ av + aw) dxdydz,

ax ay az
and Stokes theorem says that III II dw = IIall w.

Exercise 22.1. State and prove Stokes' theorem for a box, and define
the boundary of II as a sum and difference of the restrictions of II to
the six sides of the box, assigning correct signs to each so the above
formula holds.

We define the differential of a 3-form to be O. A simple calculation

22a. Holes and Forms in 3-Space 319

using the equality of mixed derivatives shows that if f is a O-form,

then d(df) = 0, and if w is a I-form, then d(dw) = O. A k-form w is
closed if dw = 0, and exact if w = dlJ- for some (k - I)-form IJ-. So all
exact forms are closed, and we have the same question as in the plane:
when are closed forms exact? We can define the De Rham groups
ltv as before: for k = 0, 1, 2, or 3,
HkV = {closed k-forms on V}/{exact k-forms on V}.
The question becomes: How does the topology of V influence the size
of the vector spaces HkV?
For I-forms, the answer is very similar to the case of open sets in
the plane. A closed I-form w on V is exact if and only if integral of
w over paths in V depend only on the endpoints, or all integrals over
closed paths are zero. For example, if U is the complement of the z-
axis, the I-form w = (-y dx + x dy) / (~ + /) is closed but not exact;
as we know, an integral of w around a circle in the xy-plane is 21T.
Notice that taking a point or a closed ball out of ~3 does not count
as a "hole" as far as I-forms in 3-space is concerned. In fact:

Exercise 22.2. Show that if HI U = 0 then every closed I-form on V

is exact.

For 2-forms, however, if U is the complement of a point, there are

closed forms that are not exact. For example, let
xdydz - ydxdz + zdxdy
w =

Exercise 22.3. Show that w is closed. Fix a positive number p, and

let f: [0,21T] x [_1/21T, 1/21T]~ ~3 be the spherical coordinate map-
ping, i.e.,
f(tt, <p) = (p cos({) cos(<p), p sin({) cos(<p), p sin(<p».
Compute the integral of w over f, and deduce that w is not exact.

Exercise 22.4. Use this 2-form w to define the engulfing number around
o of a differentiable map from S2 to ~3 \ {O}. Can you prove any an-
alogues of the winding number?

We will see in the next chapter how to define second homology

groups H2 U that have the same relation to 2-forms and H2U as the
first homology H1U has to I-forms and HIV.
Some of these ideas may be more familiar in vector field language:
320 22. Toward Higher Dimensions

a I-form pdx + qdy + rdz can be identified with the vector field
pi + qj + rk, the 2-form udydz + vdxdz + wdxdy with the vector field
ui - vj + wk, and the 3-form hdxdydz with the function h. In this
language, the differential df of a function corresponds to the gradient
grad(f), the differential of a I-form to the curl of a vector field, and
the differential of a 2-form becomes the divergence of a vector field:
ox of of
grad(f) = -i+-j +-k;
ox oy oz

curl(pi + qj + rk) = (or _ Oq)i + (OP _ or)j + (Oq _ OP)k;

oy oz oz ox ox oy
ou ov ow
div(ui + vj + wk) = - + - + -;
ox oy OZ
and the equations dod = 0 say that curl 0 grad = 0 and div 0 curl = o.
The integral of the vector field corresponding to a 2-form over a sur-
face can be interpreted as the integral of a dot product with an out-
ward-pointing normal.

Problem 22.5. (a) Generalize the discussion of §2c to fluid flows on

open sets in 3-space. Interpret the engulfing number as the flux across
a surface of a flow with source at the origin (see Exercise 2.26).
(b) Define harmonic functions of three variables, and generalize Ex-
ercise 2.21 and Problems 2.22-2.25.

22b. Knots
A knot is a subset K of rR3 or the 3-sphere S3 that is homeomorphic
to a circle. Call two knots equivalent if there is a homeomorphism of
rR3 (or S3) with itself that takes one homeomorphic ally onto the other,
and is orientation-preserving. (For a precise definition of "orientation-
preserving," see §23c.) A weaker notion is similarity, which is the
same except ignoring orientation, so "mirror image" knots are always
similar. The generalized Jordan curve theorem, which we will prove
in the next chapter, implies that the complement of X is connected,
that its first homology group is infinite cyclic, and its other homology
groups vanish. In particular, these groups are the same for all knots.
However, the fundamental group 'iT I(S3 \ X) or 'iT I(rR 3 \ X) can some-
times be used to distinguish knots from each other. Note that these
fundamental groups are the same for similar knots, so they can be
used as a possible invariant.
22b. Knots 321

Exercise 22.6. If K C ~3 is a knot, and ~3 is identified with the com-

plement of a point in S3, by stereographic projection, show that the
fundamental group of ~3 \ K is isomorphic to the fundamental group
of S3 \K.

The standard embedding K = Sl C ~2 C ~3 C S3 is called the trivial

knot, and any knot equivalent to this is called trivial. We want to use
fundamental groups to give one example of a knot that is not trivial.
We will do this by identifying S3 C ~4 = (:2 with the set of pairs (z, w)
of complex numbers such that Izl2 + Iwl 2= 1, and constructing a knot
K that is the intersection of S3 with a complex plane curve that has
a singUlarity at the origin. This is more than a convenient way to find
an example: the knots that arise this way are important invariants of
singularities of plane curves! In this language, the trivial knot can be
realized as the intersection of S3, with the "curve" w = o.

Exercise 22.7. Show that the fundamental group of the complement

of the trivial knot is isomorphic to Z. Show, in fact, that the circle
{CO, w): Iwl = I} is a deformation retract of the complement of circle
{(z,O): Izl = I} in S3.

In nature coverings often arise by starting with a mapping that is

not a covering, but becomes one after throwing away a locus where
it fails to be a covering. We have seen this for an analytic mapping
between Riemann surfaces in Chapter 19, where only a finite set had
to be thrown away. With appropriate hypotheses, such a mapping is
called a "branched covering," with the bad set the "branch locus."
Here is another example. Consider the mapping /: (} ~ ([2 given by
the formula/(u, v) = (u, v3 + uv). The inverse image of a point (z, w)
has three points if the equation v 3 + zv = w has three distinct solutions
for the variable v, and one or two points otherwise.

Exercise 22.8. Show that rl(z, w) has three points if and only if
4i + 27w2 ¥- o. If 4z3 + 27w2 = 0, but (z, w) ¥- (0,0), the inverse im-
age has two points, and for (z, w) = (0,0), the inverse image has one
point.

Let V C (:2 be the plane curve 4z3 + 27w2 = 0, which is the branch
locus of the above mapping f. Let K be the intersection of V with S3:
V = {(z,w):4z 3 +27w2 =0}, K = vns 3 •
We claim first that K is homeomorphic to a circle.
322 22. Toward Higher Dimensions

Exercise 22.9. Show that the mapping e2'frit ~ ( _ae4'frit, beMrit ), where
a is the positive solution to the equation 4a 3 + 27a 2 = 27 and
b = ~, is a homeomorphism of S I onto K.

We will consider the mapping (:2 \rl(V)~ (:2 \ V determined by

j, and the restriction
p: Y=r l (S3\K) ~ S3\K=X, (u,v) ~ (u,v3 +uv).

Take x = (1, 0) as the base point in X, and y = (1, 0) as the base point
in Y. Note that p-I(X) = {(1, 0), (1, i), (1, -i)}.

Claim 22.10. (1) p is a three-sheeted covering map; (2) Y is con-

nected; and (3) p is not a regular covering.

It follows from this claim that 'TT1(S3 \ K, x) is not abelian, since

every connected covering of a manifold with an abelian fundamental
group is regular. In particular, K is not a trivial knot. We leave the
proofs of (1)-(3) as exercises, with the following comments. The
essential point of (1) is showing that the roots of a polynomial are,
locally where the roots are distinct, continuous functions of the coef-
ficients. In fact, they are analytic functions, by the same argument
as in §20a. For (2), it suffices to show that the three points of rl(x)
can be connected by paths. Consider the loop -y(t) = (e2'frit, 0), O:s t:s 1,
at x. This lifts to the loop :V1(t) = (e 2'frit,O) at y, and to the path
:;;2(t) = (e 2'frit, ie'frit) that goes from (1, i) to (1, -i) in Y.

Exercise 22.11. Find a path of the form <r(t) = (A(t), iJl.(t», O:s t:s 1,
with A(t) > 0 and Jl.(t);::: 0 for all t, that goes from (1,0) to (1, i) in Y.

This exercise shows that Y is connected, and since :;; I is closed,

and 12is not, the covering is not regular.
In fact, up to equivalence, K is an example of a torus knot. The
torus T = rFe II? sits in S3 by the mapping that takes (x, y) to
«(1/V2)i'frix, (1/V2)e2'friY). For any relatively prime pair of positive
integers p and q, the image in the torus of the line with equation
qy = px in ~2 is a knot that winds p times around the torus one way
while it winds q times around the other way. This is called a torus
knot of type (p, q).

Exercise 22.12. (a) Show that the knot K = V n S3 considered above

is equivalent to a torus knot of type (2,3). (b) Show that for relatively
prime positive integers p and q the intersection of zq = w P with S3 is
a torus knot of type (p, q).
22b. Knots 323

The VanKampen theorem can be used to calculate the fundamental

group of the complement of any torus knot. The 3-sphere is the union
of two solid tori

so T=A n B = {(z, w): Izl = Iwl = 1/Y2} is a torus.

Problem 22.13. If K is a torus knot of type (p, q) in T, show that
the fundamental groups of A \ K, B \ K, and T\ K are infinite cyclic,
and the generator of the fundamental group of T \ K maps to the pth
and qth powers of generators of the fundamental groups of A \ K and
B \ K. Apply the Van Kampen theorem to show that the fundamental
group of S3 \ K has two generators a and b, and one relation a P • bq = e,
i.e., the fundamental group is F2/N, where F2 is the free group on a
and b, and N is the least normal subgroup containing a P ' bq •

For a knot of type (2,3), for example, one can see again that this
group is not abelian by mapping it onto the symmetric group @:i 3 on
three letters, sending a to the transposition (1 2) and b to the per-
mutation (1 2 3).
There are many knots that are not torus knots. For example, one
can take a torus knot, and take a small tube around it, which is ho-
meomorphic to another torus, and put a torus knot on this. Repeating
this construction arbitrarily often gives a class of knots which, re-
markably, are exactly the knots one gets from singularities of plane
curves. There are many other knots, however. Moreover, there are
some "wild" knots, such as "Antoine's necklace":

The fundamental group of the complement of this knot is not even

finitely generated. A piece of this is an embedding of a closed interval
in ~3 such that the complement is not simply connected.
324 22. Toward Higher Dimensions

22c. Higher Homotopy Groups

The higher homotopy groups 'ITk(X , x) are easier to define than higher
homology or cohomology groups, although their calculation turns out
to be far more challenging. Fix a base point So in the sphere S\ say
the north pole: So = (0, ... ,0,1). Define 'ITiX,x) to be the set of
homotopy classes of maps from Sk to X that map So to x; here a ho-
motopy between two such maps must preserve basepoints throughout
the homotopy, i.e., H is a continuous map from Sk x [0, 1] to X, with
H(so x t) = x for all 0::5 t::5 1. One can also define 'ITk(X, x) as the set
of homotopy classes of maps from the standard k-cube [k to X that
map the boundary of the cube to x, with homotopies also mapping
the boundary to x throughout.

Exercise 22.14. (a) Show that these two definitions agree by showing
that Sk is homeomorphic to the space obtained by identifying all points
of the boundary of [k to a point. (b) Show that 'ITk(X, x) = 0 for all
k> 0 if X is contractible. (c) Show that a map f: X ~ Y determines
mapsf*: 'ITk(X,x)~'ITlY,f(x)), which are functorial, and that maps
that are homotopic through basepoint-preserving homotopies deter-
mine the same map on homotopy groups. (d) Show that 'ITk(Sn, so) = 0
for O<k<n.

The sets 'ITiX,x) can be made into groups, much as for the fun-
damental group. Using the definition by cubes, one can "multiply"
two maps f and A from [k to X, defining f· A by using the first
coordinate:
_ {f(2t l ' t2 , • • • ,tk), 0::5 II ::5 1/2,
f· A(t l , · · · , tk ) - 1
A(2t l -1,t2 , • • • ,tk ), /2::5t l ::51.

Exercise 22.15. (a) Show that this operation is well defined on ho-
motopy classes, and makes 'ITlX,x) into a group. (b) Show that the
maps f* of Exercise 22.14 are homomorphisms of groups.

Problem 22.16. Show that, for all k> 1, the group 'ITk(X,X) is abe-
lian.

It is a fact that 'IT.(S·, so) = Z, although this is quite a bit harder to

prove. Note that this gives a strong notion of degree for maps of sn
to sn: it defines the degree, and shows that maps are classified up to
homotopy by their degree. In the next chapter we will use chains to
22d. Higher De Rham Cohomology 325

define homology groups Hk(X), which are easier to calculate, and we

will show that Hisn) = 7L and Hlsn) = 0 for all k> 0, k =Ie- n. In stark
contrast with the homology groups, for k> n, the groups 'ITk(sn, so)
need not be trivial.

Exercise 22.17. Identifying S3 with {(z, w) E C2: Izl2 + Iwl 2= I}, show
that the map that takes (z, w) to w / z E C C S2 determines a continuous
mapping from S3 to S2.

It is a fact that 'IT3(S2, so) =:; 7L, with generator given by the mapping
of the preceding exercise, which is called a Hop! mapping (see Hilton
(1961». We will show in the next chapter that the fibers of the Hopf
mapping are circles that are "linked" together in S3, which is at least
an indication of the nontriviality of the Hopf map. It should be pointed
out, however, that in spite of enormous effort, which have produced
calculations of many special cases, the groups 'ITlS n, so) are far from
known in general.

22d. Higher De Rham Cohomology

All of the discussion of §22a generalizes to n variables, a k-form
being an expression
2./;li2' . .ikdxil dxi2 ... dxik = 2./Jdx[
the sum over all 1:5 i. < ... < ik :5 n, with the coefficients h1i2' . . ik
functions on an open set in IW. There is a differential d that takes
C{6'"
a k-form to a (k + I)-form, and again dod = O. For U open in ~n one
then gets De Rham groups HkU, the space of closed k-forms modulo
the space of exact k-forms. The main complication is in keeping track
of the signs. This is best done by introducing formally the "exterior
algebra" structure that is already apparent in the plane and 3-space:
one allows the differentials to be written in arbitrary order, but put
in a sign whenever they are interchanged: dydx = -dxdy, together
setting dxdx = O. (The usual notation for this exterior product is the
wedge "/\," so one writes dx /\ dy in place of our dx dy.) Working this
out properly belongs in an advanced calculus course.
As with surfaces, one can define k-forms on an arbitrary differ-
entiable manifold as collections of k-forms on the coordinate neigh-
borhoods of a chart that transform properly on overlaps. If you know
about differential forms on a manifold, it is not difficult to generalize
326 22. Toward Higher Dimensions

the idea of De Rham cohomology. Again there are differentials d from

k-forms to (k + I)-forms, with dod = O. In this section we will assume
familiarity with notions of manifolds and differential forms. If this
applies to you, fine; if not, you can either stick to low dimensions
where we have done it by hand, or assume this formalism in gen-
eral-or you can turn immediately to the next chapter, which does
not depend on any of this.
If X is a differentiable manifold, one can define the De Rham co-
homology group HkX as the vector space of closed k-forms modulo
the subspace of exact k-forms. If f: X ~ Y is a differentiable map,
there is a notion of pull-back of k-forms from Y to X, taking CJ) on Y
to f*CJ) on X. This commutes with the differential, so determines a
(functorial) homomorphism /*: Hky ~ HkX.

Exercise 22.18. If X is a disjoint union of a finite or infinite number

of manifolds Xi' show that HkX is the direct product of the HkXi , i.e.,
specifying a class on X is the same as specifying a class on each Xi .

The Mayer-Vietoris exact sequence is defined with almost no change

from the case with HO and HI. To define
&:Hk(UnV) ~ Hk+I(UUV),

as before, one uses a partition of unity subordinate to an open cov-

ering X = U U V by U and V to write a closed k-form CJ) on un V as
the difference III - 112 of a k-form ilion U and a k-form 112 on V;
there is a closed (k + I)-form on U U V that is dillon U and dll2 on
V, and this (k + I)-form represents the image in Hk+I(U U V) of the
class represented by CJ) in F(U n V). As before (and see §24a), one
proves:

Mayer-Vietoris Theorem 22.19. For any open sets U and V in a

manifold of dimension n, there is an exact sequence

_
0 - HO(UuV) -4HoUeHoV ~ HO(UnV)
-4 HI(UuV) -4 HIUeHIV ~ HI(UnV) -4
... ...
-4 H"(UUV) -4 H"UeH"V ~ Hn(UnV) - 0
To calculate these groups, one needs in addition the

Poincare Lemma 22.20. If p: X x ~~X is the projection, then

p*: HkX~Hk(X x~) is an isomorphism.
22d. Higher De Rham Cohomology 327

The inverse isomorphism to p* is S*: Hk(X x ~)~Hk(X), where S

is the embedding x ~ x x 0 of X in X x ~. The problem is to show
that p* 0 s* = (s 0 p)* is the identity. The idea of the proof is to con-
struct a linear map H from the space of k-forms on X x ~ to the space
of (k - I)-forms on X x ~, for each k, such that for any form w on
XX~,

(22.21) w - p* os*(w) = d(H(w» + H(d(w».

(Note that the two H's and the two d's in this equation are defined
on different spaces of forms!) It follows that if w is closed, then
w - p*os*(w) = d(H(w», so w and p*os*(w) defme the same De Rham
cohomology class.

Problem 22.22. Show that any k-form w on X x ~ has a unique

expression as a sum of a k-form not involving dt, where t is the co-
ordinate on ~, and one of the form dt /\ fJ., where fJ. is, in local co-
ordinates, a sum of expressions f· dx[, with the Xi coordinates on X,
andfis a function on the product of the coordinate neighborhood with
R Define H of such a form to be the form obtained by integrating
fJ. with respect to the variable in ~ (so forms not involving dt are
mapped to 0). For example, if XC~n, and fJ. is the formfdx[, then
H(dt/\fJ.) is the form gdx[, where

g(Xl, ... ,x.,t) = J:f(Xl' . .. ,xn,s)ds.

Show that this operator is well defined and satisfies (22.21).

For any real number t, if St: X ~ X x ~ maps x to x X t, then, since

pOSt is the identity, it follows that St*: Hk(X X ~)~Hk(X) is the in-
verse to p*: Hk(X)~Hk(X x ~). In particular, the maps St* are the
same for all t. This implies that if F: X x ~~ Y is differentiable, all
the maps Ft : X~ Y, Flx) = F(x x t), determine the same maps
Ft*: HkY~HkX. Indeed, Ft = Fost , so Ft* = st*oF* is independent
of t.

Problem 22.23. (a) Use the Poincare lemma and Mayer-Vietoris to

calculate the De Rham cohomology of ~n, S", and ~n \ {O}. (b) Show
that the (n - 1)-form Wn-l defined on ~n \ {O} by
n

L(_1)i-l Xi dx 1/\ . .. /\dxi-1/\dxi+ 1/\ . .. /\dx n

i=l

(xi + . . . + X~)"/2
328 22. Toward Higher Dimensions

is closed and gives a generator of Hn- 1(lR n \{O}). If f: sn-l---+ IRn \ {O}
is differentiable, this can be used to define a higher-dimensional winding
number, or "engulfing number": W(j,O) is the integral off*wn -l over
sn-l, divided by the integral of W n -l over Sn-l.

22e. Cohomology with Compact Supports

There is another way to use differential forms to construct coho-
mology groups, for open sets in IR n or any ~'" manifolds, which we
sketch briefly here. These cohomology groups are called De Rham
groups with compact supports, and denoted H~X, the subscript c
standing for "compact." These are defined exactly as for the ordinary
De Rham groups, but using differential forms with compact support,
i.e., forms for which there is some compact set K contained in X such
that the form vanishes identically outside K. Define H~X to be the
quotient space of the closed k-forms with compact support by the
subspace of forms that are differentials of (k - I)-forms with compact
support.
If X is compact, of course, all forms have compact support, so
H~X = HkX. In spite of this and the similarity of definition, however,
these groups are quite different on noncompact manifolds. For ex-
ample, locally constant functions on a noncompact connected space
can never have compact support unless they are identically zero:
H~X = 0 if X is connected and not compact.
In fact, we will see that the theories Hk and H~ behave in an opposite,
or dual, way. By using a partition of unity as in Chapter 18, if X is
an oriented n-manifold, one can integrate an n-form with compact
support over the whole manifold. Using Stokes' theorem, it follows
similarly that the integral of the differential of a closed (n - I)-form
with compact support is zero, so one has a map

It is easy to produce n-forms that are positive on a small piece of a

coordinate neighborhood, and zero elsewhere, to see that this map is
not zero. We will see in Chapter 24 that, if X is connected as well,
then this map is an isomorphism.

Exercise 22.24. If X is a disjoint union of a finite or infinite number

of open sets Xi, show that H~X is the direct sum of the H~Xi' i.e.,
22e. Cohomology with Compact Supports 329

specifying a class on X is the same as specifying a class on each Xi'

except that all but a finite number must be zero.

There is a Mayer-Vietoris exact sequence for cohomology with

compact supports, but it is different from that without supports. First
of all, there are no restriction maps, since if W has compact support
on an open set, its restriction to an open subset may no longer have
compact support. In fact, the maps go the other way: if VI is an open
subset of V z , any k-form W with compact support on VI extends by
zero outside VI to define a k-form with (the same) compact support
on V z . Since any point in Vz \ VI has a neighborhood not meeting the
support of w, this extension is <:(6"". This extension commutes with the
differential map d, so determines a linear map
H~(VI) ~ H~(Uz).
In particular, for two open sets V and V, we have diagrams
H~(V)

/~
H~(VUV) .

H~(V)
We want to define a coboundary map 8: H~(UU V)~H!+I(Vn V).
Given a class in H~(U n V), represent it by a closed k-form W with
compact supports. We can write W as a sum WI + Wz, with WI and W2
k-forms with compact supports in V and V, respectively. In fact, if
tVI + tVz = I is a partition of unity subordinate to V and V, then WI = tVI • W
and Wz = tVz· W are such forms. Let" be the (k + I)-form on V n V
that is (the restriction 00 dWI . From the equation 0 = dw = dWI + dW2,
we have dWI = -dwz on V n V, so the support of" is contained in
the intersection of the supports of WI and Wz. In particular, " has
compact support on V n V. Clearly" is closed. Set
8([wD = ["l.
It is not hard to verify that this is independent of the choice, and it
is the familiar (by now) argument (and see §24a) to prove the

Mayer-Vietoris Theorem 22.25 (Compact Supports). For open sets

V and V in an n-dimensional manifold, there is an exact sequence
330 22. Toward Higher Dimensions

o-

... _
m(Un V) ~ H~U E9 H~V --=--. H~(U UV)
~ H~(UnV) ~ H~UE9H~V --=--. H~(UUV) ~

~ H~(unv) ~ H~UE9H~V ~ H~(UUV)

...
- o.
As before, to complete the basic tools for calculating these groups,
we need to compare a manifold X and X x IR. This time the projection
p from X x IR to X determines homomorphisms

p*: H~(X x IR) ~ H~-I(X),

by "integrating along the fiber," as follows. As in Problem 22.22,

one can write a k-form with compact support of X x IR as a sum of
a form not involving dt, and a k-form dt 1\ fJ., where fJ. is, in local
coordinates, a sum of expressions f· dxl , with the Xi coordinates on
X, and f a function on X x IR. Define p* of such a form dt 1\f· dxl to
be g . dxl , where

g(xt. ... ,xn) = f/(X1, ... ,xn,t)dt

(so forms not involving dt are mapped to zero). Note that these in-
tegrals are really over finite intervals, by the assumption of compact
support. One checks that this is well defined, and that p*(dw) = d(p*w),
so p* determines a map on cohomology classes as indicated. If
s: X ~ X x IR is the inclusion X ~ x x 0, there is a map
s*: H~-I(X) ~ H~(X x IR)

determined by sending a form w to p(t) dt 1\ w, where p is any function

with compact support on IR such that f':"" p(t) dt = 1. One checks that
this commutes with d, so defines a map on cohomology, and that
p* 0 s* is the identity.

Poincare Lemma 22.26 (Compact Supports). For any manifold X,

p*: H~(X X IR) ~ H~-I(X) is an isomorphism.

Problem 22.27. Prove this by constructing an operator H from k-

forms with compact support to (k - I)-forms with compact support
on X x IR. This operator should vanish on forms without "dt," and
take dt 1\ fJ., where fJ. is, in local coordinates, a sum of expressions
22e. Cohomology with Compact Supports 331

f· dx/, to g' f.L, with

g(XI, ... ,xn,t) = L/(Xl,' .. ,xn,s)ds

-Lo p(s)ds (/(Xh ... ,xn,s)ds.

Show that for any k-form w with compact support, w - s*p*w =
d(H(w)) +H(d(w)) , and deduce that P* and s* determine inverse iso-
morphisms.

Exercise 22.28. Calculate H~X, when X is 1Rn, Sn, and IRn \ {O}.
CHAPTER 23

Higher Homology

23a. Homology Groups

The groups HeX and H,X are the beginning of a series of abelian
groups HkX, defined for any topological space X. Define a k-cube in
X to be a continuous map r:
[k ~ X, where [k is the k-dimensional
cube, i.e., [= [0, 1], so
[k = [0,1] x ... x [0, 1] C IRk.
For any such mapping r,and any integer i between 1 and k, and any
o:5 S :5 1, define a (k - 1)-cube a: r, which is obtained by restricting
r to the slice of the ith coordinate at s:
a:r:[k-'~X, a:r(t" .. . ,tk-,)=f(t" ... ,ti-"S,ti , ... ,tk-,).
Call r degenerate if, for some i, aIr is a constant function of s, and
nondegenerate otherwise. (When k = 1, r is a path, and degenerate
is the same as a constant path.) By convention, [0 = {O}, so as O-cube
is given by a point in X; no O-cube is regarded as degenerate.
Let eX be the free abelian group on the nondegenerate k-cubes in
X, so an element of CkX is a finite linear combination Ln,,.fa' with
e, a k-cube in X and na an integer. An element of CkX is called a
cubical k-chain on X. It is useful to regard any finite linear combi-
nation Lnara of arbitrary k-cubes as an element of CkX, by simply
discarding any degenerate k-cube ra that appears. (In other words,
CkX is identified with the quotient of the free abelian group on all k-
cubes in X, modulo the subgroup generated by degenerate k-cubes.)

332
23a. Homology Groups 333

If f: It~X is a k-cube in X, its boundary af in Ck-IX is defined

by the formula
k

af = L(-li(a?f-a:f),
;=1

which is the sum of the 2k faces of f, each with a coefficient of + 1

or -1. (Note that, even if f is nondegenerate, some of the a?f or
at f occurring in the formula can be degenerate, so they are dis-
carded.) This is extended linearly to a homomorphism
a: CkX ~ Ck-IX
by the formula a(Lnafa) = Lna(af a). A k-chain is called a k-cycle if
its boundary is zero; the k-cycles form a subgroup ZkX of CkX. The
boundaries of (k + I)-chains form a subgroup BkX of CkX.

Exercise 23.1. Show that for any (k + I)-cube f, a(af) = 0 in Ck-IX.

From this exercise it follows that BkX is a subgroup of ZkX, so we

can define the kth homology group of X to be quotient
HkX = ZkX/BkX .
Exercise 23.2. (a) Show that HkX = 0 if X is a point and k> O. (Note
that this would not be true if degenerate cubes had not been dis-
carded.) (b) Verify that for k = 0 and k = I, these are the groups we
studied in Chapter 6. (c) Show how any continuous mappingf: X ~ Y
determines homomorphisms h: HkX ~ HkY, and show that these are
functorial in the sense of Exercise 6.20. (d) Construct homomor-
phisms from '7TlX,x) to HkX that are compatible with the maps of
(c) and Exercise 22.14.

Proposition 23.3. If f and g are homotopic maps from X to Y, then

hand g* determine the same homomorphisms from HkX to HkY.
Proof. Suppose H: X x [0, I]~ Y is a homotopy from f to g, and
f: Ik ~ X is a k-cube; define a (k + I)-cube R(f) by the formula
R(f)(s, t l , ••• ,tk) = H(f(tt. ... ,tk) x s) .
If Lnafa is a k-cycle, a little calculation shows that the boundary of
LnaR(fa) is Lna(fo fa) - Lna(g 0 fa), which completes the proof. A
more elegant way to see this is to extend R by linearity to a map
R: CkX ~ CH IX, R(LnafJ = LnaR(fa). Then a formal calculation shows
334 23. Higher Homology

that
g*-!* = aoR+Roa

as homomorphisms from CkX to CkX. Then if z is a k-cycle,

g*(z) - !*(z) = a 0 R(z) + R 0 a(z) = a(R(z)) ,
which shows that g*(z) and !*(z) differ by a boundary. o
It follows from the proposition that if Y C X is a deformation re-
tract, then HkY --7 HkX is an isomorphism for all k. For example, if X
is contractible, then HkX = 0 for all k> o.

23b. Mayer-Vietoris for Homology

To calculate the higher homology groups of more interesting spaces,
we want to extend the Mayer-Vietoris sequence to these higher groups:
if U and V are open subsets of a space, there is an exact sequence
... --4 HiunV) ~ HkU $ HoV ~ Hk(UUV)
--4 Hk_l(un V) ~ Hk_1U$ Ht_1V ~ Hk_I(UUV) ~ ...
The idea is very much as in Chapter 10. The key is the definition of
the "boundary maps" a: H k ( U U V) --7 H k - 1(U n V). For this, we show
that any class in H k( U U V) can be represented by a k-cycle z that is
a sum of a chain cIon U and a chain C2 on V. Then aCI = -aC2 is a
cycle on un V, and this represents the image class in Hk-I(U n V).
To carry out the construction of the boundary map, we need a sys-
tematic way to subdivide k-cubes and chains into sums of smaller
chains, generalizing the constructions we used in Chapter 10. Given
a k-cube f: /k--7 X, we define S(n to be the sum of the 2k k-cubes
obtained by restricting f to each of the 2k subcubes obtained by sub-
dividing the cube:
23b. Mayer-Vietoris for Homology 335

Each of these restrictions must be renormalized, to be defined on the

cube Ik with sides of length 1. In symbols,
S(f) = ~f 810 ••• ,Ek'
the sum over all 2k choices of Ei = 0 or 1, and f EI ,... ,Ek is the k-cube
defined by the formula
f 81o •. . ,Ek(t l , ••• ,tk) = fef2(tl + El), ... ,1/2(tk + Ek»'
This is extended by linearity to give a homomorphism S: CkX ~ CkX,
i.e., by defining S(~nafa) = ~naS(fa). If the boundary of S(f) is cal-
culated, all the terms corresponding to inner faces cancel, and one
gets the result of subdividing the boundary of f. In symbols,
(23.4) aes = sea,
as homomorphisms from CkX to Ck-IX,

Exercise 23.5. Verify this formula.

We can iterate this subdivision operation, defining for any k-chain

c new k-chains S(c), S2(C) = S(S(c» , S3(C) = S(S(S(c))), and so on.

Lemma 23.6. If X is a union of two open sets U and V, and c is a

k-chain on X, then, for sufficiently large p, SP(c) can be written as a
sum Cl + C2 where Cl is a k-chain on U and C2 is a k-chain on V.
Proof. This is an immediate consequence of the Lebesgue lemma,
since, if f: Ik~ X is a k-cube, the image of each k-cube appear-
ing in SP(f) is the image of a subcube of Ik with sides of length
1/~. D

To use this construction and lemma, we want to know that, if z is

a k-cycle, then S(z) is a k-cycle defining the same homology class as
z. For this we proceed as follows. Let a: [0, l]~ [0,1] be defined
by the formula
2t, 0 ~ t ~ 1/2,
aCt) = { 1, 1/2~t~l.

If f is a k-cube in X, define a k-cube A(f) by the formula

A(f)(t l , ... ,tk) = f(a(tl), ... ,a(tk»'
Note that the "first comer" A(f>o,. .. ,0 of this k-cube is f, and all the
other A(f)EIo' .. ,Ek are degenerate. Extend this by linearity as usual to
336 23. Higher Homology

a homomorphism A: CkX ~ CkX. (If k = 0, define A to be the identity

map.) By the observation just made, we have
(23.7) SoA = I,
where I: CkX ~ CkX is the identity map. Now define, for a k-cube f,
a (k + 1)-cube H(f) by the formula
H(f)(s, t" ... , tk) = [«(1 - s)a(t,) + st" ... , (1 - s)a(tk) + stk),
and extend by linearity to a homomorphism H: CkX ~ Ck+,X. (If k = 0,
set H = 0.) Note that a?(H(f» = A(f) and a:(H(f» = f. From this one
sees that
(23.8) aoH+Hoa = I-A,
as homomorphisms from CkX to CkX.

Exercise 23.9. Verify this formula.

For each p "2. 1, define a homomorphism Rp: CkX ~ Ck+ ,X by the

formula
Rp =SoHo(/+S+S 2 + ... +SP-').
Then we have, for all p "2. 1, the identity
(23.10) aoRp+Rpoa = SP-I.
In fact, this is a formal calculation, following from (23.4), (23.7),
and (23.8), as follows. Whenp= 1, R, =SoH, and
aoSoH + SoHoa
SoaoH+SoHoa
So(aoH + Hoa) = So(/- A)
S-SoA=S-I.
For p> 1, Rp=R,oSp, where Sp=I+S+ ... +SP-'. We use the
case p = 1 in the form a0 R, = S - I - R, 0 a, and we use the fact that
a commutes with Sp by (23.4), together with the identity
9' - I = (S - I) 0 Sp. Calculating, we have
aoR, oSp + Rpoa
(S -1- R, oa)oSp + Rpoa
(S - /)oSp + (-Rl oa)oSp + Rpoa
SP-I- R, ospoa + Rpoa
SP -1- Rpoi) + Rpoa =SP -I,
as asserted.
23b. Mayer-Vietoris for Homology 337

Now suppose that X = U U V. The definition of the boundary ho-

momorphism from HiU U V) to Hk-I(U n V) depends on the follow-
ing lemma:

Lemma 23.11. (a) Any homology class in H,;K can be represented

by a cycle z on X of the form z = CI + C2, where CI is a k-chain on U
and C2 is a k-chain on V. (b) The (k - I)-chain aCI = -aC2 is a cycle
on un V, and its homology class in Hk-I(U n V) is independent of
choice of CI and C2'
Proof. For (a), take any cycle c that represents the homology class.
By Lemma 23.6, for some p ~ 1, the chain S Pc can be written as the
sum of a chain cion U and a chain C2 on V. By (23.10),
SPc - c = a(Rp(c» + Ria(c» = a(Rp(c» ,
from which it follows that z = SPc is a cycle representing the same
homology class as c.
For (b), suppose z' = CI' + C2' is another representative of the same
form for the same homology class. There is a (k + I)-chain w on X
with a(w) = z' - z. By Lemma 23.6, there is a p ~ 1 such that Spew)
can be written as a sum of a chain on U and a chain on V. Applying
(23.10) to the chain a(w), we have
z' - z = a(w) = SP(a(w» - Ria(a(w))) - a(Rp(z' - z»
= a(SP(w» - a(Rp(z' - z» = a(SP(w) - Rp(z' - z».
From the formula for Rp it follows that Rp takes a chain on U to a
chain on U and a chain on V to a chain on V. We know that z' - z
is a sum of a chain on U and a chain on V, and it follows that Rp(z' - z)
is also. It follows that there are (k + I)-chains YI and Y2 on U and V
such that Spew) - Rp(z' - z) = YI + Y2, so
Z' - Z = a(YI + Y2) .
This means that we have an equality of k-chains
CI' - CI - a(YI) = -(C2' - C2 - a(Y2» '
the left side of which is a chain on U and the right side is a chain on
V. This chain, denoted x, is a chain on un V, and it follows that
a(x) = a(cl') - a(CI) ,
so the cycles a(cl') and a(cl) differ by a boundary on un V, as
asserted. D
Define a: HkX~Hk-I(Un V) by taking the homology class [z] of
338 23. Higher Homology

a cycle z of the form CI + C2 with CI and C2 chains on V and V, to the

homology class [a(cI)] of the cycle a(cI) = -a(C2) on V n V. It fol-
lows from Lemma 23.11 that this definition makes sense. The proof
that it is a homomorphism, and that the resulting Mayer-Vietoris se-
quence is exact, is precisely the same as in Chapter 10, so will not
be repeated; the general algebra for this will be described in §24a.
In fact, the above argument shows something more. Let X be any
au
space, and = {Va: a E.sil} any open covering of X. Let Ck(X)i1U be
the subgroup of Ck(X) consisting of linear combinations of nonde-
generate k-cubes f: /k ~ X such that the image of f is contained in
one (or more) of the open sets Va. These cubes are said to be small
with respect to au. The boundary operator maps Ck(X)i1U to Ck_I(X)i1U,
so we can define the corresponding homology groups Hk(X)i1U. There
is a natural map from H k(X)i1U to Hk(X). The following proposition
says that we can always calculate homology by using chains that are
small with respect to any convenient covering.

Proposition 23.12. The natural map Hk(X)i1U~ Hk(X) is an isomor-

phism.

Exercise 23.13. Use (23.4), (23.7). (23.8), and (23.10) to prove this
proposition.

This proposition also gives a more concise way to construct the

Mayer-Vietoris exact sequence:

Exercise 23.14. For au = {V, V}, construct a homomorphism from

H k(X)i1U to Hk-I(V n V), and show that there is a long exact sequence
.. . ~Hk(Vn V)~HkV(f)HN~HiX)i1U~Hk_I(Vn V)~ . . . .
Combine with the proposition to get the full Mayer-Vietoris se-
quence.

The following is a useful general consequence of the Mayer-Vietoris

sequence.

Exercise 23.15. Suppose a space X is a union of some open sets

VI, ... , Vp such that all homology groups Hk(y) vanish for any
intersection Y = ViI n ... n Vi, of these open sets and all k> o. (a)
Show that HiX) = 0 for k ~ p. (b) If, in addition, each intersection
Y is connected, and p ~ 2, show that Hp-I(X) = O. (c) Finally, if each
intersection Y is connected and nonempty, show that Hk(X) = 0 for all
k>O.
23c. Spheres and Degree 339

23c. Spheres and Degree

We saw that Sn is simply connected if n ~ 2, so HIS n = 0. With Mayer-
Vietoris, one can calculate the homology groups of all spheres Sn.
One can cover Sn by two open sets U and Veach homeomorphic to
open disks in IRn, whose intersection is homeomorphic to Sn-I X f' for
an open interval f'. Mayer-Vietoris gives an isomorphism
i): Hk(Sn) -=+ Hk-\(sn-I xf') =:; H k-\(sn-I)
for all k > 1. From this one sees that
H(Sn) {7L k ?
k
=:;
° if = or n,
otherwise.

Exercise 23.16. Show that the complements of the south and north
poles (0, ... ,0, -1) and (0, ... ,0,1) satisfy the conditions for U
and V, and use Mayer-Vietoris to complete the proof of this calcu-
lation.

Equipped with homology groups, the definition of the degree of a

continuous map I: Sn ~ Sn is easy: Since HnSn =:; 71.., the induced map
1*: HnSn~ HnSn is multiplication by an integer, and this integer is
defined to be the degree of I, denoted deg(f). Equivalently, if [z] is
a generator for Hnsn, deg(f) is the integer such that/*([z]) = deg(f)· [z].
(Note that using the other generator -[z] leads to the same degree.)

Exercise 23.17. (a) Show that homotopic maps from sn to sn have

the same degree. (b) Show that if f: Sn ~ Sn extends to a continuous
map from D n + 1 to S", then deg(f) = 0. (c) Show that if f and g are
maps from sn to sn, then deg(gof) = deg(g)· deg(f). (d) Show that
for any integer d and any n ~ 1 there are maps f: Sn ~ sn of degree d.

In fact the converses of (a) and (b) of the preceding exercise are
true, but more difficult.
Having the notion of degree, we can define the generalization of
winding number: the engulfing number W(f, P) of a continuous map
I: sn-I ~ IRn \ {P} around P. This can be defined to be the degree of
the map that follows f by projection onto a sphere around P, i.e.,
define W(f, P) to be the degree of the map
I(x) -P
x t-+ .
11/(x) - pil
340 23. Higher Homology

Problem 23.18. (a) Show that, as a function of P, this number is

constant on connected components of IW \f(sn-I). (b) State and prove
an n-dimensional analogue of the dog-on-a-leash theorem. (c) Show
that for f Sn~ IRn \ {OJ differentiable, this definition agrees with that
in Problem 22.23.

Similarly, one can define the local degree of a continuous mapping

f U ~ V between open sets in IRn at a point P in U, provided there
is a neighborhood Up of P such that no other point of Up has the same
image as P. This is the degree of the mapping
f(P + rx) - f(P)
sn-I ~ sn-I, x ~ ~----'----=---'----

Ilf(P + rx) - f(P)11 '

for any positive r so that Up contains the ball of radius r around P.
This can be used for example to define the notion of a homeomor-
phism from IRn to IRn (or Sn to Sn, or X to X for any oriented manifold)
being orientation preserving or orientation reversing, according as
the local degree at any point is + 1 or -1.

Exercise 23.19. (a) Show that this local degree is a continuous func-
tion of the point, so the notion of orientation preserving or reversing
is well defined. (b) Show that a homeomorphism of sn is orientation
preserving or reversing according as its degree is + 1 or -1. Show
that a homeomorphism of IR n extends to a homeomorphism of
Sn = IRn U {oo}, whose degree therefore determines whether the origi-
nal map is orientation preserving. (c) For a diffeomorphism, show
that the local degree is given by the sign of the determinant of the
Jacobian.

With the concept of degree, the following assertions of Borsuk and

Brouwer are proved just as before, and the proofs are left as exercises.

Theorem 23.20. (1) There is no retraction from D n+ I onto Sn.

(2) Any continuous mapping from a closed disk D n to itself must
have a fIXed point.

Exercise 23.21. Generalize the results of Exercises 4.9-4.17.

Problem 23.22. Show that the degree of the antipodal map from sn
to Sn is 1 if n is odd and -1 if n is even. In particular, the antipodal
map is not homotopic to the identity if n is even.
23c. Spheres and Degree 341

Problem 23.23. (a) Show that no even-dimensional sphere can have

a nowhere vanishing vector field. (b) Construct on every odd-dimen-
sional sphere a nowhere vanishing vector field.

To extend the results about antipodal mappings, we need the fol-

lowing generalization of Borsuk's Lemma 4.20. As there, we denote
by p* = -P the antipode of a point Pin Sn.

Theorem 23.24. Letf: sn~sn be a continuous map.

(a) If f(P*) = f(P)* for all Pin Sn, then the degree off is odd.
(b) If f(P*) = f(P) for all P in Sn, then the degree off is even.
The proof of this requires some new ideas, and is postponed to
Appendix E. Assuming this theorem, we can draw the expected con-
sequences:

Corollary 23.25. (a) Ifm < n, there is no continuous mappingf: Sn~Sm

such that f(P*) = f(P)* for all P in Sn.
(b) Any continuous mapping f: Sn ~ IRn must map some pair of an-
tipodal points to the same point.
(c) An open set in IRn cannot be homeomorphic to an open set in
IRmifn~m.
(d) It is impossible to cover Sn with n + 1 closed sets, none of which
contains a pair of antipodal points.

Exercise 23.26. Prove this corollary.

Problem 23.27. Let f: Sn ~ Sn be continuous. (a) If f(P*) #- f(P) for

all P, show that deg(f) is odd. (b) If f(P*) #- f(P) * for all P, show
that deg(f) is even. (c) Show that the only nontrivial group that can
act freely on an even-dimensional sphere is the group with two ele-
ments.

Exercise 23.28. Show that if n > 1 any continuous mapping from Sn

to IRr must map some pair of antipodal points to the same point.

Exercise 23.29. Prove that if n + 1 bounded measurable objects are

given in IRn, then there is a hyperplane that cuts each in half.

Exercise 23.30. (a) State and prove n-dimensional analogues of Ex-

ercises 4.24-4.31 and 4.34-4.39. (b) Define the index of a vector
field on an open set in IRn at an isolated singular point, and state and
prove the n-dimensional analogues of Proposition 7.5 and its corol-
laries.
342 23. Higher Homology

Exercise 23.31. Use Mayer-Vietoris to compute the homology groups

of a torus Sl x Sl, or of any product Sm x Sn.

Homology can be used to define a notion of degree in many other

contexts. Here is an important illustration. A link in ~3 is a disjoint
union of knots. Equivalence is defined just as for knots. A link can
be nontrivial even if all the knots occurring in it are trivial. There is
a linking number that measures how many times two knots intertwine
with each other.

o ±I ±2

Suppose K and L are disjoint knots. Define a mapping

x-y
xXy ~ I~_YII'
which assigns to the pair (x, y) the direction from x to y. This mapping
F determines a homomorphism F *: H 2(K x L) --') H 2(S2). Both of these
homology groups are isomorphic with 7L. Choosing orientations of
each identifies them with 7L, and then F * is multiplication by some
integer, which is defined to be the linking number I(K,L). If a stan-
dard orientation is fixed for S2, the sign of I(K, L) depends on ori-
entations chosen for K and L; it changes sign if either of these ori-
entations are changed. The linking number also changes sign if the
roles of K and L are reversed. Note that if K and L are far apart, then
F will not be surjective, so this linking number is zero.

Exercise 23.32. (a) Show that the linking number of the two circles
in Exercise 22.7 is ± 1, and therefore the linking number of two fibers
of the Hopf mapping of Exercise 22.17 is ± 1. (b) Show that the
intersection of a small sphere S3 around a singUlarity of a plane curve
that is a node (see §20c) is two circles whose linking number is ± 1.

It is a fact (see §24c) that the top homology group HnX of an ori-
ented n-manifold X is Z (the orientation determining a choice of gen-
erator). It follows that any continuous map f: X --') Y between oriented
n-manifolds has a degree.
23d. Generalized Jordan Curve Theorem 343

Problem 23.33. (a) Show that if X is a compact oriented surface,

then HzX =- lL. (b) If X is a compact nonorientable surface, show that
HzX = O. (c) Iff X ~ Y is a nonconstant analytic map between com-
pact Riemann surfaces, show that the degree defined by homology is
the same as the number of sheets of the corresponding branched cov-
ering.

23d. Generalized Jordan Curve Theorem

There is a vast generalization of the Jordan curve theorem to higher
dimensions. This can be stated as follows:

Theorem 23.34. If xeS' is homeomorphic to a sphere Sm, then m ::5 n,

and m < n unless X = S'. If m < n, the homology groups of the com-
plement are
lL EB lL if m = n - 1 and k = 0,
Hk(Sn\X) =- { lL if m<n-l and k=O or k=n-l-m,
o otherwise.
In particular, the complement has two components ifm = n - 1, and one
ifm<n-l.

The essential point of this theorem is the assertion that the ho-
mology of the complement is the same for all embeddings of Sm in
Sn. The proof follows the pattern for the Jordan Curve Theorem in
the plane very closely, using the full Mayer-Vietoris theorem. We
discuss only the new features, leaving details to the reader. First we
have the analogous result for embeddings of cubes in Sn.

Proposition 23.35. If X C Sn is homeomorphic to r, then S" \ X is

connected and Hk(Sn \ X) = 0 for all k > o.
Proof. This is by induction on m, the case m = 0 being clear since
the complement of a point is homeomorphic to [R", so contractible.
For m > 0 write r as the union of two halves whose intersection is
homeomorphic to r- I ; this makes X a union of two subspaces A and
B. Applying Mayer-Vietoris to U = Sn \ A and V = Sn \ B, and know-
ing about U U V by induction, we have
O=Hk+I(UUV) ~ Hk(UnV) ~ HkUEBHN.
From this it follows that if z is a k-cycle on Sn \ X that is not a bound-
344 23. Higher Homology

ary, then it is not a boundary on Sn \ A or sn \ B. Continuing to cut

the cubes in half, passing to the limit as in Chapter 5, we find that
z is not a boundary on Sn \ {x} for x a point, from which the conclusion
follows easily. 0

To prove the theorem, also by induction on m, write X as the union

of two closed sets A and B homeomorphic to the upper and lower
hemispheres of the sphere Sm. Each of A and B is homeomorphic to
fR, and AnB is homeomorphic to Sm-I. Applying Mayer-Vietoris
and the proposition to U = sn \ A and V = Sn \ B, we get
o~ Hk+1(S" \A n B) ~ Hk(S" \X) ~ 0
if k> 0, and
O~HI(sn\A nB)~Ho(S" \X)~HoUtJJHoV~Ho(S" \A nB)~O.

We know about Sn \ An B by induction, so the first display computes

Hk(sn \X) for k> O. If m < n - 1, H1(Sn \A nB) = 0, and the second
gives
o~ Ho(S" \X) ~ ltJJl ~ l ~ 0,
from which it follows easily that Ho(S" \ X) == l. If m = n - 1, then
H1(sn \ An B) == l, and from
° l Ho(S" \X) ltJJl l.. °
~ ~ ~ ~ ~
we find similarly that Ho(Sn \X) == lEBl... If m = n, from
° Ho(Sn\X) lEBl ltJJl.. °
~ ~ ~ ~

we see that Ho(Sn \ X) = 0, so X = Sn. From this it follows that no

larger sphere can be embedded in Sn, since an n-dimensional sub-
sphere would already map to the whole Sn. 0

Exercise 23.36. State and prove analogous results for X C ~" homeo-
morphic to a sphere.

Exercise 23.37. If F: Dn~ [Rn is continuous and one-to-one, show

that IRn \ F(sn-I) has two connected components, one the image of the
interior of Dn , the other the complement of F(D"). Deduce the in-
variance of domain: if U is open in ~n, and F: U ~ IRn is a one-to-
one continuous mapping, then F(U) is open. Prove the invariance of
dimension: open sets in [Rn and IRm cannot be homeomorphic if n ¥- m.

It is also true that if F: S"~ IRn+1 is an embedding, then its en-

23d. Generalized Jordan Curve Theorem 345

gulfing number around points in the two components of the comple-

ments is ± 1 for the bounded component and 0 for the unbounded
component (see Proposition 5.20), but this requires more machinery.

Exercise 23.38. Suppose A and B are disjoint closed subsets of Sn,

n > 1, and two points are given in the complement of A U B. Show
that if A and B do not separate these points, neither does A U B.

The Mayer-Vietoris sequence can also be used to show that a com-

pact nonorientable surface cannot be topologically embedded in ~3.
We sketch a proof in the following problems:

Problem 23.39. (a) Show that if X e ~3 is homeomorphic to a Moe-

bius band, then Hk(~3 \ X) == 7L if k = 0 or 1, and Hk(~3 \ X) = 0 oth-
erwise. (b) With X as in (a), if Yex corresponds to the boundary
circle, show that the map
HI(~3 \ X) ~ HI(~3 \ y) == 7L
determined by the inclusion of ~3 \ X in ~3 \ Y takes a generator of
the first group to twice a generator of the second.

Suppose a subspace X of ~3 were homeomorphic to the projective

plane. Write X as a union of a space A homeomorphic to a Moebius
band and B homeomorphic to a disk, with A n B homeomorphic to
the boundary circle of each. Then with U = ~3 \ A, V = ~3 \ B, Mayer-
Vietoris and Exercise 23.36 give an exact sequence
Hl(~3 \A)E90 ~ Hl(~3 \A n B) ~ Ho(~3 \X).
By the preceding problem, the image of a generator of HI(~3 \A n B)
maps to an element IX in Ho(~3 \ X) that is nonzero, but 2· IX = O.
However, we know that the Oth homology group of any space is a
free abelian group, which has no such element.

Problem 23.40. Show similarly that none of the nonorientable com-

pact surfaces can be embedded in ~3.

The following problem generalizes two of the main results of Chap-

ters 6 and 9:

Problem 23.41. Let U be an open subset of ~n. (a) Show that two
classes in Hn-IU are equal if and only if they map to equal classes
in H._I(~n \ {P}) for all P not in U. (b) Show that if the complement
of U in S· = ~. U {co} is a disjoint union of m + 1 compact connected
sets, then H.-l U is a free abelian group on m generators.
CHAPTER 24

Duality

24a. Two Lemmas from Homological Algebra

We frequently want to compare different homology and cohomology
groups, when we have exact Mayer-Vietoris sequences for each, and
maps between them. Assuming that most of the maps are isomor-
phisms, we want to deduce that the others are as well. There is a
general algebraic fact that can be used for this:

Lemma 24.1 (Five-Lemma). Given a commutative diagram

A--B --C--D __ E

!
A ' - - B' - - C' - - D' _ _ E'
of abelian groups, such that the rows are exact sequences, and all
the vertical maps but the middle one are isomorphisms, then the mid-
dle map from C to C' must also be an isomorphism.

The proof is by a "diagram chase," which is much easier and en-

joyable to do for oneself than to follow when someone else does it.
Here is how to show that the map is one-to-one. If c in C maps to 0
in C', then its image in D maps to 0 in D' (by commutativity of the
diagram), so c maps to 0 in D (since D~D' is injective), so c comes
from some element b in B (by exactness of the top row). The element
b maps to an element b' in B' that maps to 0 in C' (why?), that

346
24a. Two Lemmas from Homological Algebra 347

therefore comes from an element a' in A'. This element a' comes
from some element a in A, and this element a must map to b since
they have the same images in B'. Since a maps to 0 in C, and b maps
to c, c must be o.
Exercise 24.2. (a) Prove similarly that C ~ C' is surjective. (b) Show
that the five-lemma is also valid under the following weaker as-
sumptions, still assuming the rows are exact: (i) each square either
commutes or commutes up to sign, i.e., the composite going around
one way is plus or minus the composite going around the other way;
and (ii) the maps B~B' and D~D' are isomorphisms, A~A' is
surjective, and E ~ E' is injective.
If you like this diagram chasing, there is a general process that
constructs long exact sequences, which can be used to construct all
the Mayer-Vietoris sequences we have seen. For this, one has a com-
mutative diagram of abelian groups

where the rows are exact, and the composite of any two successive
maps in the columns is zero. One says that the columns are chain
complexes. The diagram is abridged to saying one has a short exact
sequence of chain complexes
o~ C*' ~ C* ~ C*" ~ o.
For each chain complex (column) one can form homology groups.
For the center column,
HiC*) = Zk(C*)/Bk(C*) ,
348 24. Duality

where Zk(C*) = Kemel(Ck~ Ck- l ) are the k-cycles, and Bk(C*) =

Image(Ck+1 ~ Ck) are the k-boundaries. Similarly for the other two
columns. There are maps from Hk(C*') to Hk(C*) and from Hk(C*)
to Hk(C*'), determined by the horizontal maps in the diagram. For
example, the map from C/ maps Zk(C*') to Zk(C*) and BiC*') to
Bk(C*), so it determines a homomorphism on the quotient group. The
interesting maps are boundary homomorphisms
0: HlC*") ~ Hk-I(C/).

To define these, take a representative i' in ZlC*") of a class in HiC*').

Choose an element c in Ck that maps onto z". Let c be the image of
c in Ck-1- Then c maps to 0 in Ck-/', since it has the same image
there as z" does, and z" is a cycle. So c comes from an element c' in
Ck- l '.

Exercise 24.3. Show that this element c' is a (k - I)-cycle, and its
homology class in Hk-I(C*') is independent of choices of the repre-
sentative z" and the element c that maps onto z".

The homology class of c' is defined to be the boundary of the ho-

mology class of z": o([z"]) = [c']. One checks easily that 0 is a ho-
momorphism of abelian groups.

Proposition 24.4. The resulting sequence

• • • --'? Hk+ ICC /') --'? Hk(C *') --'? Hk(C *) --'? Hk(C/,) --'? Hk-I(C/') --'? ••

is exact.

The proof is some more diagram chasing, which again we leave as

an exercise. There is a similar result when the vertical maps in the
diagram go up rather than down. Usually then the indexing is by up-
per indices, so we have "cochain complexes"
C*: .. . ~Ck-I~Ck~Ck+I~ . ..
A short exact sequence o~ C*' ~ C* ~ C*" ~ 0 determines a long
exact sequence of their cohomology groups
.. . ~Hk-I(C*")~Hk(C*')~Hk(C*)
~Hk(C*")--'?Hk+I(C*')~ ..
Let us see how some of the Mayer-Vietoris sequences we have
seen earlier fall out of this formalism. For example, if X is a ~""
manifold, and CkX denotes the vector space of ~oo k-forms on X, and
24a. Two Lemmas from Homological Algebra 349

U and V are open sets in X, there is an exact sequence

o~ C*(U u V) ~ C*(U) EEl C*(V) ~ C*(U n V) ~ 0
of cochain complexes. The first map takes a form w on U U V to the
pair (wlu,wlv), and the second takes a pair (WI>W2) to the difference
wdunv - w2lunv. The exactness of this sequence is clear from the def-
initions except for the surjectivity of the second map, and that was
proved using a partition of unity in Chapter 10. The Mayer-Vietoris
sequence then results immediately from Proposition 24.4.
Similarly, for cohomology with compact support, one has a short
exact sequence of cochain complexes
o~ ct(U n V) ~ Ct(U) EEl ct(V) ~ Ct(U U V) ~ 0,
where the first map takes a form w on un V to the pair (w u, -w v),
where W U denotes the extension by 0 from un V to U, and similarly
for V; the second map takes (WI>W2) to Wl uUV +wtuv . Again, ex-
actness follows from a partition of unity argument, and the Mayer-
Vietoris exact sequence results.
For homology, if a space X is a union of open sets U and V, let
CtCX)"U denote the k-chains that are small with respect to the covering
au = {U, V} of X. Then there is an exact sequence
o~ c*(Un V) ~ C*(U)EElC*(V) ~ C*(X)'lL ~ 0

of chain complexes, the first taking a chain on un V to the pair con-

sisting of its images on U and on V, and the second taking a pair to
the difference of their images on X. The exactness is immediate, giv-
ing an exact sequence
~ HHl(C*(X)"U) ~ Hiun V) ~ HiU)EElHiV)
~ HiC*(X)'lL) ~ Hk-I(U n V) ~

To complete the proof, one appeals to Proposition 23.12, which says

that Hk(C*(X)'lL) =HiC*(X» = Hk(X),

Exercise 24.5. If 0 ~ C' ~ C~ C" ~ 0 is an exact sequence of free

abelian groups, and G is any abelian group, show that
o~ Hom(C", G) ~ Hom(C, G) ~ Hom(C', G) ~ 0

is also an exact sequence. IfO~C*'~C*~C*"~O is an exact se-

quence of complexes of free abelian groups, this gives an exact se-
quence O~Hom(C*",G)~Hom(C*,G)~Hom(C*',G)~O of co-
chain complexes, and hence an exact sequence of cohomology groups.
350 24. Duality

Exercise 24.6. Let C * and C *' be chain complexes with boundary

*'
maps denoted ih: Cc~ Ck- 1 and iJ/: Ck ' ~ Ck- 1', respectively. Define
a map of chain complexes f*: C *~ C to be a collection of homo-
morphisms.fk: Ck~C/ that commute with the boundary maps. Show
that such mapsh determines homomorphisms from HiC*) to Hk(C/).
Call two maps f* and g* chain homotopic if there is a collection of
maps, Hk: Ck~Ck+l' such that
gk-fk = iJk+loHk+Hk-loiJk
for all k. Show that f* and g* then determine the same maps from
Hk(C*) to Hk(C*') for all k.

If C*' ~ C* is a map of chain complexes such that each C/ ~ Ck

is one-to-one, one can define ct to be Ck/Ck', getting an exact se-
quence O~C*'~C*~C*/C*'~O, so a long exact homology se-
quence. For example, if Y is a subspace of a topological space X,
then C*Y~C*X is one-to-one, so one can define a quotient chain
complex C*X/C*Y. The homology groups of this complex are de-
noted Hk(X, y), and are called the relative homology groups. They
fit in a long exact sequence
.. . ~Hk+l(X, Y)~Hk(Y)~Hk(X)~Hk(X, Y)~Hk-l(Y)~' ..
For suitably nice spaces, these relative groups are isomorphic to the
homology groups of the space obtained by collapsing (identifying) Y
to a point. In many treatments of algebraic topology, these relative
groups, and the above sequence, are used for calculation in most sit-
uations where we have used the Mayer-Vietoris sequence.

24b. Homology and De Rham Cohomology

In this section we want to prove that the De Rham cohomology groups

HkX of a manifold are dual to the homology groups HkX, i.e., we
want to construct an isomorphism
HkX"':;' Hom(HkX, IR),
generalizing what we did for surfaces for k = 1. The idea is similar:
one wants to integrate k-forms over k-cubes. This makes sense for
differentiable k-cubes, but there is a problem of how to define this
for continuous k-cubes that are not differentiable-a problem that we
avoided for k = 1 by cutting up the path, locally writing the k-form
24b. Homology and De Rham Cohomology 351

as the differential of a (k - I)-form J.L, and evaluating J.L over the end-
points of the subdivided path. For k> 1, unless the restriction of the
k-cube to its boundary is differentiable, this will not work. A more
systematic procedure, that does work, is to show that the homology
HkX can be computed by using only differentiable cubes.
A cube f: /k ~ X is a «6" cube if it extends to a «6" mapping on
some neighborhood of the cube /k C lit. Define C;X to be the free
abelian group on the nondegenerate «6" k-cubes. The boundary afrom
the k-chains to the (k - i)-chains takes «600 cubes to «6" cubes, so we
can define the «6" homology groups H;X to be the quotient of the
closed «6" k-chains modulo the subgroup consisting of boundaries of
«6" (k + i)-chains. There is an obvious map

which, in the language of the preceding section, is given by the map

of chain complexes C;X ~ C*X. We will show that this is an iso-
morphism.
If w is a k-form on X, and f: /k~ X is a «6" cube, we can define
the integral of w over f by

Ir w = Ll*(W),
where f*(w) is the pull-back form; a form on the cube can be written
f(x i> • • . , Xk) dx 1 /\ dx2 /\. • . /\ dxb

and the integral of such a form is the usual Riemann integral of the
continuous function f on the cube.

Exercise 24.7. Prove "Stokes' theorem" in this context: if w is a

(k - i)-form, then

r dw
Jr
= r w.
Jar
From Stokes' theorem, it follows as in the case k = 1 that there is
a map

dx ~ Hom(H;X, IR),

We will see that this is also an isomorphism. Combining these two

isomorphisms will give the duality we were after.
352 24. Duality

To prove these isomorphisms, we need to know that the groups

H; X have many of the same properties as the purely topological groups
HkX, For example,

Exercise 24.8. (a) Show that a C(5'" mapping f: X ~ Y of manifolds

determines functorial homomorphismsh: H;X~H;Y. (b) If two maps
from X to Yare homotopic by a C(5'" mapping F: X X l' ~ Y (where I'
is an open interval containing [0, 1]), then they determine the same
homomorphisms. (c) Deduce that H;U = 0 for k> 0, and H~U = 7L
if U is a starshaped open set in ~n.

Similarly, one has Mayer-Vietoris exact sequences just as for the

groups HkX, and compatible with the maps from groups H; to the
Hk • In fact, the same construction works in the C(5" case, noting that
the subdivision operators used to cut cubes into small pieces preserve
C(5'" chains. One modification needs to be made in our proof, however,
since the operator A we used in Chapter 23 used a function that is
only piecewise differentiable.

Exercise 24.9. Change the function a used in §23b to a C(5'" function

from [0, 1] to [0, 1] such that a(O) = 0 and a(t) = 1 if t;::::: 1/2. With
any such a, show that, for any chain r, SoA(f) - r is a boundary.
Use this to complete the proof of Mayer-Vietoris for these groups.

To prove these isomorphisms, we need a way to build up arbitrary

manifolds out of simple pieces. The following general lemma will
suffice for our purposes. Let us call an open rectangle in ~n an open
rectangular solid with sides parallel to the axes, i.e., an open set of
the form (ai, b l ) x ... X (an, bn).

Lemma 24.10. If X is an open set in ~n, then X can be written as

the union of two open sets U and V such that each of U and V and
U n V is a disjoint union of open sets, each of which is a finite union
of open rectangles.
Proof. Take compact sets KI C K2 C ... as in the Lemma A.20.
Construct a sequence of open sets Up as follows. Let UI be a finite
union of rectangles covering KI , with the closure of each contained
in the interior of K2 • Let U2 be a finite union of rectangles covering
K2 \ Int(K I), with the closure of each contained in the interior of K3 •
Inductively, let Up be a finite union of rectangles that covers the com-
pact set Kp \ Int(Kp _ d, the closure of each contained in the interior of
24b. Homology and De Rham Cohomology 353

Kp+l and in the complement of K p - 2 , and not meeting Up - 2 • Now let

U be the union of the union of all Up with p even, and let V be the
union of the union of all Up with p odd. D

Lemma 24.11. If X is a differentiable n-manifold, then X can be

written as the union of two open sets U and V such that each of U
and V and U n V is a disjoint union of open sets, each of which is a
finite union of open sets diffeomorphic to open sets in IR".
Proof. The argument is the same. Remark A.21 shows that X is a
union of compact sets Kl C K2 C ... with the same properties. Then
the preceding proof, with "rectangle" replaced by "open set diffeo-
morphic to an open set in IRn" goes over without change. D

Theorem 24.12. For any manifold X the natural maps lI';X ~ HkX
are isomorphisms.
Proof. Let us write "T(X)" for the statement that the maps from
lI';X to HkX are isomorphisms for all k. There are three tools:
(1) T(U) is true when U is an open rectangle in IRn.
(2) If U and V are open in a manifold, and if T(U), T(V), and T(U n V)
are true, then T(U U V) is true.
(3) If X is a disjoint union of open manifolds X a , and each T(Xa) is
true, then T(X) is true.
With what we have seen, each of these is easy to prove. (1) follows
from the fact that H;U and HkU vanish for k>O, and both are nat-
urally isomorphic to Z when k = 0, cf. Exercise 24.8. (2) follows
from the fact that we have Mayer-Vietoris exact sequences for each,
with compatible maps between them:
H;unV-H;UeH;V- H;UUV-H':.-lunV-H;_IUeH;Y

~ ~ ~ ~ ~
Hk U n V - Hk U e Hk V - Hk Uu V - Hk_1un V - Hk_1U e Hk_1V

so the five-lemma shows that the middle map is an isomorphism if

the others are. (3) follows from the fact that to specify a class of either
kind on X is equivalent to specifying a class on each Xa , with all but
a finite number of these classes being zero (i.e., HkX is the direct
sum of the groups HkXa , and similarly for H;).
We can now use these tools to prove the theorem. We first show
that T(X) is true whenever X C IR" is a finite union of open rectangles.
This is by induction on the number of rectangles. (1) takes care of
354 24. Duality

one rectangle, and if X is a union of p rectangles, let U be the union

of p - 1 of them and let V be the other. Then T(U) and T(V) are true
by induction, and T(U n V) is true since un V is also a union of at
most p - 1 rectangles, since the intersection of two rectangles is either
empty or a rectangle. Then T(X) is true by (2).
Next we show that T(X) is true whenever X is an open set in ~n.
By Lemma 24.10 one can write X as a union of two open sets U and
V, such that each of U and V and un V is a disjoint union of open
sets, each of which is a finite union of open rectangles. Applying the
preceding step and (3) we know that T(U) and T(V) and T(U n V)
are true, and by (2) again we know that T(X) is true.
Note that since a diffeomorphism between manifolds determines an
isomorphism between the corresponding groups, it follows that T(X)
is true for any set diffeomorphic to an open set in ~n. The same in-
ductive argument as for rectangles shows that T(X) is true when X is
a finite union of open sets, each diffeomorphic to an open set in ~n.
For the general case, Lemma 24.11 shows that any manifold X is a
union of two open sets U and V such that each of U and V and U n V
is a disjoint union of open sets, each of which is diffeomorphic to a
finite union of open sets in ~n. By the last step and (3) again, T(U)
and T(V) and T(U n V) are true, and a final application of (2) shows
that T(X) is true. 0

Theorem 24.13. For any manifold X the natural maps from H"x to
Hom(H;X, ~) are isomorphisms.
Proof. The proof follows exactly the same format, with T(X) being
the statement that these maps are isomorphisms for all k. Once (1)-
(3) are proved, in fact, the proof is identical. The proof of (1) is the
same, and (3) follows from the fact that to specify a class of either
kind is equivalent to specifying a class on each Xa (i.e., HkX is the
direct product of the groups HkXa , and similarly for Hom(H;X, ~».
To prove (2), we need to compare the cohomology Mayer-Vietoris
sequence with the dual of the Mayer-Vietoris sequence in homology.
For brevity write H';X* in place of Hom(H;X, ~). We have a diagram

An application of the five-lemma, together with the following ex-

ercise, finishes the proof. 0
24c. Cohomology and Cohomology with Compact Supports 355

Exercise 24.14. Show that this diagram commutes.

These theorems justify the use of «6"" techniques in studying the

topology of a differentiable manifold. For example, they show that
the De Rham groups depend only on the underlying topology of the
manifold. Combining the isomorphisms of the two theorems, one has
justified writing Iz w for z a continuous k-cycle and w a closed «6"" k-
form on a manifold.

Problem 24.15. Show that for U open in ~n. two classes 'Tl and 'T2
in H n- 1 U are equal if and only if IT, w = IT2 w for all closed (n - 1)-
forms won U.

Exercise 24.16. Let X be an n-manifold that can be covered by a

finite number of open sets such that any intersection of them is dif-
feomorphic to a convex open set in ~n. (It is a fact, proved by using
a Riemannian metric and geodesics, that any compact manifold has
such an open cover.) Show that each HkX is a finitely generated abe-
lian group, and that each HkX and H~X is a finite-dimensional vector
space.

24c. Cohomology and Cohomology

with Compact Supports
In higher dimensions, except in simple cases in the Poincare lemmas,
we have not yet used the higher-dimensional versions of wedging forms
that we used on surfaces in Chapter 18. In general the wedge w t'l.L
of a k-form w and an I-form I.L is a (k + I)-form. This operation is
linear in each factor, and satisfies the identities:
(i) I.LAw=(-lt ' WAI.L; and
(ii) d(wA I.L) = dWAI.L + (-ltWAdl.L.
Again, we assume these properties from advanced calculus. If either
w or I.L has compact support, then w A I.L has compact support, since
the support of the wedge product is contained in the intersection of
the supports of the factors. It follows from (ii) that the wedge product
of two closed forms is closed, and that, if one is closed and the other
is exact, the wedge product is exact. From this it follows that the
wedge product determine products on the cohomology groups
A : dx X H'X ~ Hk+'X
356 24. Duality

and

each by the formula [w] x [fJ.]i-7 [w]A [fJ.] = [wAfJ.]'

Exercise 24.17. Verify that these are well-defined bilinear mappings.

Show that the first satisfies the formula [fJ.] A [w] = (- Itt [w] A [fJ.].
Prove that these products are associative where defined.

Now suppose X is oriented. As we saw in §22e, integrating over

the manifold gives a mapping H~X ~ II\t So we have homomorphisms
HkX x H~-kX ~ H~X ~ IR.
This determines linear maps qjJx: HkX ~ Hom(H~-k X, IR). Explicitly,
qjJx takes the class of a closed k-form w to the homomorphism that
takes the class of a closed (n - k)-form fJ. with compact support to
the integral Ix w A fJ..

Theorem 24.18. For any oriented manifold X the duality maps

qjJx: HkX ~ Hom(H~-kX, IR)
are isomorphisms.
Proof. The proof is almost identical to that for Theorems 24.12 and
24.13. This time, for (1), note that HOU = ~ and H~U =- ~ for U an
open rectangle, and all other groups vanish, by the Poincare lemmas;
and since 1 E ~U maps to the nonzero homomorphism that is inte-
gration over U, the map is an isomorphism. For (2), one again has
maps from the Mayer-Vietoris sequence for Hk to the dual of the
sequence for the H~-k. This time the signs involved in the definition
mean that the key square in the diagram only commutes up to sign,
but that is good enough to apply the five-lemma, cf. Exercise 24.2.
We leave the calculation of these signs as an exercise. 0

This duality theorem has several corollaries that were not obvious
before. For example, the simple fact that H~X = 0 whenever X is a
connected but not compact manifold implies the

Corollary 24.19. If X is a connected, oriented, but noncompact n-

manifold, then HnX = o.

Corollary 24.20. If X is a connected oriented n-manifold, then the

map H~~IR, wi-7fxw, is an isomorphism.
24c. Cohomology and Cohomology with Compact Supports 357

Proof. Since HOX = IR, it follows from the theorem that H~X is one
dimensional, and we have seen that the map H~ ~ IR is not zero.
D

Problem 24.21. Let X be a nonorientable connected n-manifold, and

let p: X~ X be the orientation covering of § 16a. (a) Construct maps
p*: H~X ~ H~X and p*: H~X~ H~X so that p* 0 p* .... = 2 ..... and
p* °P*w = W + T*W, where T: X~ X is the nontrivial deck transfor-
mation. (b) Deduce that p* embeds H~X as the subspace of H~X
consisting of classes of the form W + T*W. (c) Show that H~X = O. In
particular, if X is compact, then HnX = O.

Corollary 24.22 (Poincare Duality). If X is a compact oriented n-

manifold, then the pairing HkX X Hn-kX ~ IR is a perfect pairing, i.e.,
for any linear map lp: F-kX ~ IR, there is a unique w in HkX such
that lp(JL) = Ixw/\JLjor all JL in Hn-kX.

Problem 24.23. (a) Use this corollary to prove that dx is finite di-
mensional. (b) If n = 2m, with m odd, show that the dimension of
HmX is even, and deduce that the Euler characteristic
n

L(-ltdim(HkX)
k=O

must be even.

This puts strong restrictions on the homology and cohomology groups

of a compact oriented n-manifold. For example, the dimension of HkX
must equal the dimension of Hn-kX. The skew-commutative algebra
structure on the direct sum of the cohomology groups is also useful
in many applications.
As we saw for Riemann surfaces these duality theorems can be used
to define an intersection number (0:,13) for homology classes 0: in HpX
and 13 in H n-pX, when X is an oriented n-manifold. As in that case,
it is possible to do this directly and geometrically, by finding repre-
sentative cycles that meet transversally, and counting the points of
intersection with an appropriate sign. This takes quite a bit of work,
however, and one can use duality to define the intersection number
quickly: A class 0: in HpX determines a linear map from HPX to IR
by .... ~ I a .... , and by Poincare duality there is a unique class w" in
Hn-pX so that Ia .... = Ixw,,/\ .... for all .... in HPX. By the same construction,
358 24. Duality

~ in Hn-pX determines wj3 in HPX. So we can define

(a,~) = !xw a I\Wj3.

This is a bilinear pairing, satisfying (~, a) = (-IY"(n- p )(a, ~). The fact
that (a, ~) is always an integer, however, is not so obvious from this
definition, although it can often be verified directly by making con-
structions for representatives of Wa and wj3, as we did for surfaces in
Chapter 18.

Exercise 24.24. Suppose X is oriented but not necessarily compact,

and X has an open cover as in Exercise 24.16. Construct a homo-
morphism

characterized by the equality fa J..L = f X Wa 1\ J..L for all J..L in HPX.

Exercise 24.25. Suppose a topological space is a union of an increas-

ing family of open subsets Ui , U I C U2 C . . . . Show that any ele-
ment of HtX is the image of an element of some HtUi , and that ai
in HtUi and aj in HkUj determine the same element of HkX if and
only if there is some m ~ max(i,j) such that ai and aj have the same
image in HkUm • This is expressed by saying that HkX is the direct
limit of the HkU;, and written
HkX = lim HkU; .
~

Exercise 24.26. Suppose a manifold X is an increasing union of open

subsets Ui , U I C U2 C .... (a) Use duality to deduce that giving a
class 'T] in HkX is equivalent to giving a collection of classes 'T]i in
HkUi for all i such that 'T]i restricts to "li if i > j. This says that HkX
is the inverse limit of the HkUi , and is written
HkX= ~HkUi'

(Note that this is not obvious from the definition of De Rham groups,
even for open sets in ~n.) (b) Show that H~X is the direct limit of
the H~Ui:

In fact, one can construct cohomology groups H\X; lL) for any space
X, which are finitely generated abelian groups for manifolds as in
Exercise 24.16, and one can find an analogue of the wedge product
24d. Simplicial Complexes 359

for these groups; after proving appropriate duality theorems, one has
a construction of the intersection pairing whose values are integers.
This could be a next chapter, if this book didn't end here. At least
now we can give a quick definition of these groups, or of cohomology
groups Hk(X; G) with coefficients in any abelian group G, general-
izing directly the discussion in §16c. Define a k-cochain to be an
arbitrary function that assigns to every nondegenerate k-cube in X an
element of G; these form a group Ck(X; G). If c is a k-cochain, define
the coboundary 8(c) of c to be the (k + l)-cochain defined by the
formula 8(c)(f) = c(af), where a cochain is extended linearly to be
defined on all chains. Then 8 0 8 = 0, so one can define
Hk(X; G) = Zk(X; G)/Bk(X; G), where Zk(X; G) is the group of k-co-
cycles (whose boundary is zero), and Bk(X; G) is the group of k-co-
boundaries (of (k - l)-cochains).

Exercise 24.27. Prove that these groups satisfy the same properties
as homology groups, but "dual." For example, maps f: X ~ Y deter-
mine (functorial) homomorphisms f*: Hky ~ HkX, homotopic maps
determine the same maps on cohomology groups. State and prove the
Mayer-Vietoris theorem for these groups. Construct homomorphisms
from Hk(X; G) to Hom(H kX, G), and show that these are isomor-
phisms if G = lIt

Project 24.28. If G is an abelian group, and UU is an open covering

of a space X, define and study Cech groups Hk(UU; G) generalizing
the groups H'(UU; G) studied in Chapter 15.

24d. Simplicial Complexes

We have seen the usefulness of triangulating a surface. Many spaces
that arise in nature, including many which are not manifolds, admit
triangulations. When a space is triangulated, there is a much smaller
chain complex that can be used to compute its homology. The general
methods of §24a can be used to show that this complex computes the
same homology as that using cubical chains.
A (finite) abstract simplicial complex is a finite set V, called the
vertices, and a collection K of subsets of V, called the (abstract) sim-
plices, with the property that every subset of a simplex is a simplex.
One usually assumes also that every set {v} for v in V is a simpllx,
and one says that K is the simplicial complex. An n-simplex is a set
a in K with n + 1 elements. A subset T of a simplex a is called a
face of a.
360 24. Duality

A set of n + 1 points Po, . . ,Pn in a vector space is called af-

finely independent ifthere is no relation toPo + tlP I + ... + tnPn = 0
with to, ... , tn real numbers satisfying to + t I + ... + tn = 0 with
not all ti = O. Equivalently, the vectors PI - Po, P2 - Po, ... , Pn - Po
are linearly independent. In this case the set of points
{toP o + tlP I + ... + tnPn: li2:0, 10 + II + ... + tn = I}
is called the (geometric) simplex spanned by the points. It is homeo-
morphic to an n-dimensional disk.
The realization IKI of an abstract simplicial complex K can be con-
structed by taking the vertices V to be the basis vectors for a vector
space, and defining IKI to be the union of the geometric simplices
spanned by the abstract simplices in K. In practice one often takes
the vertices in a smaller vector space, provided those in any simplex
are affinely independent, and two geometric simplices are either dis-
joint or meet only along common faces.

We want to write down a chain complex for the simplicial complex

K. This is simplest if K is ordered. This means that a partial ordering
is given for the vertices, such that the vertices of each simplex are
totally ordered. Each simplex a then has a unique representation
a = (vo, ... ,vn) where the vertices of a are listed in order. The chain
complex C*K of the ordered simplicial complex K is defined as fol-
lows: CnK is the free abelian group on the n-simplices of K, and the
boundary iJ: CnK ~ Cn-IK is defined by
n

(24.29) iJ«vo,··. ,Vn» = 2:<-1)i(vo,'" ,Vi- \>Vi+ I , · · · ,Vn).

i= O

Exercise 24.30. Verify that the composite iJ 0 iJ: CnK ~ Cn- 2 K is zero.

The nth homology group Hn(C*K) of this complex is denoted HnK.

Exercise 24.31. Suppose K has a vertex Vo with the property that for
every simplex a in K, the subset consisting of a and va is also in K;
24d. Simplicial Complexes 361

denote this simplex by (vo, (J). (Geometrically, IKI is a cone with

vertex vo.) Assume that K is ordered so that Vo comes before all other
vertices. Define maps H: CnK ~ Cn+IK by the formula

( ) == {(vo, (J) if Vo is not a vertex of (J,

H (J 0 1' fVo'IS a vertex 0 f (J.

Show that a0 H + H 0 a is the identity map on Cn K for n > O. Deduce

that HnK == 0 for n> 0, and that HoK == 7L.

A subcomplex L of a simplicial complex K is subset of the simplices

in K such that whenever a simplex (J is in L, so are all its faces; then
L is a simplicial complex, with its vertices a subset of the vertices of
V. An ordering of K determines an ordering of L, and one has a ca-
nonical map C*L~C*K, determining homomorphisms HnL~HnK
on homology groups. If LI and L2 are subcomplexes of K, the inter-
section LI n L2 and union LI U L2 are also sUbcomplexes. These maps
determine an exact sequence of chain complexes
O~C*(LI nL2)~ C*L I ffiC*L2~C*(LI UL2)~0,

which determines a long exact Mayer-Vietoris sequence

.. . ~Hn+I(LI UL2)~Hn(LI nL2)~HnLI ffiH nL 2
~Hn(LI UL2)~' ..
We want to compare the homology of K with the homology of its
geometric realization IK I. For each ordered simplex (J == (vo, . . . ,vn)
we need to define a cubical n-chain [., == f(vo, .. .,v.) in IKI. If n == 0, fO'
is the constant O-chain at Vo. If n == 1, f is the path from Vo to
0'

VI: fO'(t) == tvl + (1 - t)vo· In general, define fO': r~ IKI inductively

by the formula
f O'(tl, ... ,tn) == tnvn + (1 - tn) f(yo, ... ,y._.>(t l , ... ,tn-I)'
Writing this out, we have
n

(24.32) fO'(tI,"" tn ) == 2: tk(1 - tk+I)' .... (1 - tn)v*,

k=O

where, when k == 0, to is set equal to 1.

Proposition 24.33. (a) The map (J'~ fO' determines a homomorphism

C*K ~C*IKI of chain complexes. (b) The induced homomorphisms
H$ ~ HnlKI are isomorphisms.
362 24. Duality

Proof. For (a), we must show that afa = L7=o(-IYf(vo ... ,~;, ... ,v ,where
n)

the " denotes an omitted vertex. From the definition of afa as

L~= I ( -1 y( a?f a - a: fa), this follows from the following three calcu-
lations, which are simple exercises, using (24.32):
(i) alfa = f(v" ... ,vn);
(ii) ai fa is a degenerate (n - 1)-cube if i > 1; and
I

(iii) a~fa = f(vo, .... ~;, ... ,v

n )·

The proof of (b) will be by the (by now) familiar induction using
Mayer-Vietoris, as follows. If L J and L2 are subcomplexes of K, we
have a commutative diagram

o- C*(IL\ II ~I) --=--+- C*ILJI E9 C*I~I ~ C*(ILJU ~I) - 0.

This gives a corresponding commutative diagram of long exact se-
quences, and the five-lemma shows that if (b) is true for L J and L2
and LJ n L2, then (b) is also true for L JU L2 .
We can now prove (b) by induction on the number of simplices in
the simplicial complex K. Take any vertex v of K. Let L J be the
subcomplex consisting of all simplices of K that are contained in a
simplex of K that contains v, and let L2 be the subcomplex consisting
of all simplices of K that do not contain v. Then (b) is known for L J
by Exercise 24.31, and (b) is known for L2 and L\ n L2 by induction
on the number of vertices. The preceding argument then shows that
(b) holds for K=LJ UL 2 • 0
Corollary 24.34. If K and L are simplicial complexes whose geo-
metric realizations are homeomorphic, then HnK=HnLfor all n.
Proof. This follows from the fact that a homeomorphism between
spaces induces an isomorphism between their homology groups. 0

In the early days of algebraic topology, the homology of a compact

space X was defined by triangulating the space, i.e., finding a homeo-
morphism between some IK I and X, and taking the homology H *K.
With this as the definition the assertion of the preceding corollary-
that homology is a topological invariant of the space-was a serious
problem.
The preceding discussion depended on a choice of ordering of the
simplicial complex, which is how one would usually use the result in
24d. Simplicial Complexes 363

calculations. The following exercise shows how this can be circum-

vented:

Exercise 24.35. For an abstract simplicial complex K, define C,.K to

be the quotient of the free abelian group on the set of symbols
(vo, ... ,vn), where Vo, ... , Vn is an (ordered) set of vertices span-
ning an n-simplex of K, modulo the subgroup generated by relations
(vo, ... ,vn) - sgn(T)(vT(O), ... ,vT(n»,
for all permutations T in the symmetric group n+1 , where sgn(T) = ± 1
is the sign of the permutation. Then CnK is a free abelian group of
rank equal to the number of n-simplices, but with basis elements only
specified up to multiplication by ± 1. (a) Show that formula (24.29)
determines a boundary map iJ: CnK --+ Cn-IK, with iJ 0 iJ = O. (b) Given
an ordered n-simplex (vo, ... ,vn), define f(vo ... .,vn) to be f(bo, .. .,bn),
where bk is the barycenter of the simplex spanned by the first k + 1
vertices, i.e., bk = I/(k + l)(vo + VI + ... + Vk). Define a map from
C*K to C*IKI by sending (vo, ... , vn) to the sum
~sgn(T)f(vT(O)' .. "VT(n»' the sum over all T in @Sn+I' Show that this de-
termines a homomorphism of chain complexes, and show that the
resulting map in homology is an isomorphism. (c) Show that an or-
dering of K determines an isomorphism of the complex defined earlier
with the complex defined in this exercise.

Problem 24.36. If C; is the number of i-simplices in K, show that the

Euler characteristic is the alternating sum of the numbers of simplices:
~(-lic; = ~(-l); dim(H;(IKI» ,

generalizing what we have seen for surfaces.

Problem 24.37. (a) If OU = {Un V E V} is a finite collection of open

sets whose union is a space X, define a simplicial complex, called
the nerve of OU and denoted N(OU) , by taking V to be the vertices, and
defining the simplices to be the subsets S such that the intersection
of the U v for V in S is nonempty. Verify that N(OU) is a simplicial
complex.
(b) If K is any simplicial complex, and v is a vertex in K, define
an open set St(v) in IKI, called the star of v, to be the union the
"interiors" of the simplices that contain v, i.e., St(v) is the comple-
ment in IKI of the union of those Icrl for which cr does not contain v.
Show that the open sets {St(v), VE V} form an open covering of IKI,
and that the nerve of this covering is the same as K.
364 24. Duality

(c) Suppose UU is an open covering of X as in (a), with the property

that for all vo, ... , Vr in V, Uvo n ... n UVr is connected and
HiUvo n ... n Uv,) = 0 for all k> O. Construct a homomorphism of
chain complexes from C*(N(UU» to C*X, and show that it determines
an isomorphism from Hk(N(UU» to HkX for all k.
 APPENDICES

These appendices collect some facts used in the text. The beginnings of
Appendices A, B, and C state definitions and basic results from point set
topology, calculus, and algebra that should be reasonably familiar, together
with proofs of a few basic results that may be slightly less so. Each of these
appendices ends with some more technical results that may be consulted as
the need arises. Appendix D contains two technical lemmas about vector
fields in the plane, as well as some basic definitions about coordinate charts
and differential forms on surfaces. Appendix E contains a proof of Borsuk's
general theorem on antipodal maps that was stated in Chapter 23.

Conventions and Notation

A closed rectangle in ~2 has sides parallel to the axes, so is a subset of the
form [a,b] x [c,d], with a<b and c<d. An open rectangle is a product
of two open intervals, usually finite, but we occasionally allow them to be
infinite.
The unit intervall is [0, I] = {x E~: 0 ~ x ~ I}.
The n-dimensional disk Dn is
D n = {(Xl, . . . ,Xn)E~n:XI2+ ... +xn2~1}.

The n-sphere S" is

S" = {(Xl,· .. ,xn+I)E~n+I:XI2+ ... +Xn+/= 1}.
The origin (0,0, ... ,0) in ~n is often denoted simply by O.
APPENDIX A

Point Set Topology

AI. Some Basic Notions in Topology

A topology on a set X is a collection of subsets, called the open sets, in-
cluding X itself and the empty set, such that any union of open sets is open,
and any finite intersection of open sets is open. A topological space is a set
X together with a topology. A collection of open sets is a basis for the open
sets if any open set is a union of sets in the basis. For example, if X is a
metric space, the open balls B.(x) = {y EX: distance (y,x) < e} form a basis
for a topology on X. In particular, Euclidean space IRn with its usual distance
function is a topological space. A neighborhood of a point in a topological
space is an open set containing the point-or, occasionally, any set con-
taining such an open set.
Any subset Y of a topological space is a topological space with the induced
topology: the open sets are those of the form un Y, for U open in X. Such
Y is called a topological subspace of X. In particular, any subset of IRn is a
topological space. A subset Y is closed if its complement is open. A map
f: X - Y from one topological space to another is continuous if 11(U) is
open in X for every open set U in Y. A bijectionf: X - Y is a homeomorph-
ism if f and 1 1 are continuous.
A topological space X is Hausdorff if, for any two distinct points in X,
there are disjoint open sets, one containing one of the points, the other con-
taining the other. Any metric space is Hausdorff. Although we seldom need
to assume spaces are Hausdorff, the reader will lose little by assuming that
all spaces occurring in the book are Hausdorff.
A subset K of a space X is called compact, if, for any collection of open

367
368 Appendix A. Point Set Topology

sets {Ua: a E stl} such that K is contained in the union of the Ua, there is a
finite subset {a(I), ... , a(m)} of stl so that K is contained in the union
Ua(l) U ... U Ua(m). The following are some basic facts about compact spaces:

(A. 1) Iff: X -+ Y is continuous, and K is a compact subset of X, then f(K)

is a compact subset of Y.
(A.2) A compact subset of a Hausdorff space is closed.
(A.3) If f: X -+ Y is continuous and bijective, and X is compact and Y is
Hausdorff, then f is a homeomorphism.
(A.4) A subset K of~' is compact if and only if it is closed and bounded.

Exercise A.S. If K and L are disjoint compact subsets in a Hausdorff space

X, show that there are disjoint open sets in X, one containing K, the other
containing L.

Exercise A.6. If K is compact, and, for each positive integer n, A. is a

nonempty subset of K, show that there is a limit point, i.e., a point P in K
such that every neighborhood of P meets A. for an infinite number of in-
tegers n.

Exercise A.7. (a) Show that a rectangle [a,b] x [c,d] is homeomorphic to

the closed unit disk {(x,y): ~ + l::; I}. (b) Show that ~2 is homeomorphic
to the open unit disk {(x,y): ~ + l < I}.

A subset X of ~. iE convex if, for any points P and Q in X, the line

segment {t· P + (1 - t)· Q: 0::; t::; I} from P to Q is contained in X.

Problem A.S. Show that any compact, convex subset of ~. that contains a
nonempty open set is homeomorphic to the closed unit disk
D' = {(XI, ... ,x.):x/+ ... +x.2 ::;1}.
If aK is the boundary of K, i.e., aK is the set of points of K such that every
neighborhood contains points inside and outside K, show that there is a ho-
meomorphism from K to D n that maps aK homeomorphically onto the boundary
Sn-I = aD n •

If X and Yare topological spaces, the Cartesian products U x V of open

sets U in X and V in Y form a basis for a topology in the Cartesian product
X X Y, called the product topology.
If X and Y are topological spaces, the disjoint union X i l Y is a topological
space. A set in the disjoint union is open when it is the union of an open
set in X and an open set in Y. More generally, if {Xa: a E.sIl} is any collection
of topological spaces, the disjoint union ilXa is a topological space, with
open sets disjoint unions of open sets in each Xa.
Any set X can be made into a topological space with the discrete topology,
in which every subset of X is open. Equivalently, all points are open.
A2. Connected Components 369

The interior of a subset A of a topological space, denoted Int(A), is the

set of points that have a neighborhood contained in A. The closure of a
subset A, denoted A, is the intersection of all closed sets containing A.

A2. Connected Components

A topological space X is connected if it cannot be written as a union of two
nonempty disjoint sets, each of which is open in X.

Exercise A.9. Show that each of the following is equivalent to X being

connected: (i) X has no nonempty proper subset that is both open and closed;
(ii) X cannot be written as the union of two nonempty disjoint closed subsets;
and (iii) there is no continuous mapping from X onto the discrete space
{O, I}.

The following are basic facts about connected spaces:

(A.lO) Iff: X- Y is a continuous, surjective mapping, and X is connected,
then Y is connected.
(A.I1) If X is a subspace of a space Y, and X is connected, then the closure
X of X in Y is also connected.
(A.12) If X is a union of a family of subspaces Xa , each of which is con-
nected, and each pair of which have nonempty intersection, then X
is connected.
(A.I3) The connected subsets of IR are the intervals.
A connected component of X is a connected subset that is not contained
in any larger connected subset. Each connected component is closed in X.
Any two connected components of X are disjoint. The union of all connected
subsets of X containing a point x is a connected component, called the con-
nected component of x in X. The space X is a disjoint union of its connected
components.

Exercise A.14. Let XC 1R2 be the union of the points (0,0), (0,1), and the
lines {lin} x [0, 1], n = 1, 2, .... Show that these are the connected com-
ponents of X, but whenever X is written as the union of two open and closed
subsets, the points (0,0) and (0, 1) belong to the same subset.

A space is called locally connected if for every neighborhood V of every

point x, there is a connected open neighborhood U of x that is contained
in V.

Exercise A.IS. If X is locally connected, show that all the connected com-
ponents of X are open in X.
370 Appendix A. Point Set Topology

A space X is path-connected if, for any two points x and y in X, there is

a continuous mapping 'Y from an interval [a, b] to X that maps a to x and b
to y. A space is locally path-connected if every neighborhood of every point
contains a path-connected neighborhood of the point.

Exercise A.16. If X is locally path-connected, show that all the connected

components of X are path-connected and open in X.

In particular, for U open in the plane or IRn, the connected components

of U are open and path-connected.

A3. Patching
If a topological space X is the union of two sets A and B, both open or both
closed, and f; A ~ Y and g; B ~ Y are continuous mappings from A and B
to a space Y, such that f and g agree on A n B, then there is a unique con-
tinuous mapping h from X to Y that agrees with f on A and with g on B.
If Y is a topological space, and R is any equivalence relation on Y, the
set Y /R of equivalence classes is given the quotient topology; a set U is
open in Y /R exactly when its inverse image in Y is open. If f; Y ~ Z is a
continuous mapping that maps all points in each equivalence class to the
same point, then f determines a continuous mapping 7: Y / R ~ Z so that the
composite Y~Y/R~Z isf.
Suppose Y I and Y2 are two spaces, with open subsets VI of Y I and V 2 of
Y2 , and a homeomorphism -It; UI ~ U2 is given between them. Then one can
patch (or glue, or clutch) the spaces Y I and Y2 together, to form a space Y.
There will be maps 'PI; Y I~ Y and 'P2; Y2~ Y; Y will be the union of the
open subsets 'PI(Y I) and 'Pz(Yz), each 'Pi will map Yi homeomorphic ally onto
'PI(Yi ), with 'PI(U I) = 'Pz(Uz), and -It will be the composite 'Pz -1 0 'PIon U I .

One can construct Yas the quotient space YlilYz!R, where R is the equiv-
alence relation consisting of pairs (UI ,-It(UI» for UI in U I , and of course the
symmetric pairs (-It(UI), UI), together with all pairs (YI ,YI) for YI in Y1 and
all pairs (Yz, yz) for Yz in Yz ·
A4. Lebesgue Lemma 371

More generally, suppose we have a collection fa of spaces, for a in an

index set 91., and, for each a and (3 in 91., we have an open subset Ua~ of
fa, and a homeomorphism
it~a: Ua~ -'» U~a.
These should satisfy the conditions:
(1) Uaa = fa, and itaa is the identity on fa; and
(2) for any a, (3, and 'Y in 91., it~a(Uap n Ua-y) C U~-y and

In particular, ita~ 0 it pa is the identity on Ua~' Set

where R is the equivalence relation: Ya in fa is equivalent to Y~ in fp if and

only if Ya E Ua~, Y~ E U~a, and itpa(Ya) = y~.
Let 'Pa be the map from fa to f that takes a point to its equivalence class.
Give f the quotient topology, which means that a set U in f is open if and
only if each 'Pa -I(U) is open in fa.

Lemma A.17. (1) Each ~a<fa) is open in f;

(2) ~a is a homeomorphism of fa onto ~a(fa);
(3) f is the union of the sets ~a(fa);
(4) ~a<UafJ) = ~fj(Ufja); and
(5) on Uafj' iJfja = ~fJ-I 0 ~a'
Proof. The fact that 'Pa is one-to-one onto its image, and the assertions (3)-
(5), are set-theoretic verifications, and left to the reader. The topology on
f is defined to make each 'Pa continuous. To prove (1) and (2), it suffices
to verify that if U is open in some fa, then 'f'a(U) is open in f, i.e., that,
for all (3, 'P~ -1('Pa(U» is open in f~. But 'P~ -1('Pa(U» = it~a(U n Ua~), which
is open since un Uap is open in Uail and itpa is a homeomorphism. D

Exercise A.IS. Make a similar construction if each Ua~ is a closed subset

of fa.

A4. Lebesgue Lemma

We make frequent use of the following lemma:

Lemma A.19. (Lebesgue Lemma). Given any covering of a compact metric

space K by open sets, there is an 13 > 0 such that any subset of K of diameter
less than 13 is contained in some open set in the covering.
372 Appendix A. Point Set Topology

Proof. If not, there is for every integer n a subset An of K with diameter

less than lin and not contained in any open set of the covering. From the
fact that K is compact it follows that there is a limit point P, see Exercise
A.6. Let U be an open set of the covering that contains P, and take r > 0
so all points within distance r of P are contained in U. There must be (in-
finitely many) n with 1/n < r /2 such that An meets the open ball of radius
r/2 around P. But such A. must be contained in U, a contradiction. 0

The following lemma will be used in Appendix B to construct a partition

of unity:

Lemma A.20. If U is an open set in ~n, there is a sequence of compact

subsets K I , Kz , ... , whose union is U, and so that
KI C Int(Kz) C K z C Int(K3 ) C ... C Kn C Int(Kn + l ) C ....
Proof. Start with any countable sequence of open sets Ui that cover U
such that the closure Ui is compact and contained in U; for example,
one can take the Ui to be balls at centers with rational coordinates and
rational radii. Take KI = UI . Then take K z = UI U ... U Up, where p is
minimal such that KI is contained in UI U ... U Up, and so on: if
Km = UI U ... U Us, take Km+ I = UI U .. U U, where t is minimal so that
Km is contained in VI U ... U V,. 0

Remark A.21. The lemma is true, with the same proof, when V is replaced
by any manifold whose. topology has a countable basis of open sets.
APPENDIX B

Analysis

B 1. Results from Plane Calculus

We list the basic results from calculus that were used in Chapters I and 2.
As in those chapters, for simplicity, differentiable functions on a closed
interval or rectangle are assumed to have differentiable extensions to some
open neighborhood. Integrals of continuous functions on a closed interval,
or a closed rectangle, are defined as limits of Riemann sums. The next five
basic facts from calculus are stated for easy reference, in the form we need.
Consult your favorite calculus book for proofs.

(B. I) Fundamental Theorem of Calculus. If a continuous function f is the

derivative of a function F on an interval [a, b], then

ff(X)dx = F(b)-F(a).

(B.2) Mean Value Theorem. Iff is continuous on an interval [a,b], there

is an x* with a < x* < b such that

- 1 Lb f(x) dx = f(x*).
b-a a

(B.3) Chain Rule. If y(t) = (x(t),y(t)), a:=; t:=; b is a differentiable path on

an interval [a, b], and f is a differentiable function on a neighborhood of

373
374 Appendix B. Analysis

,},([a, bD, then fo 'Y is differentiable on [a, b], and

d ~ ~ ~ ~
-(f(,,/(t))) = -(x(t),y(t»- + -(x(t),y(t»-.
& ~ & ~ &
(B.4) Equality of Mixed Partial Derivatives. Iff is a ~oo function on an
open set in the plane, then

(B.S) Fubini's Theorem. If f is a continuous function on a rectangle

R = [a, b] x [c, d], then

JJf(x,y)~dy = f[ff(x,Y)dY]~ = f[ff(x,Y)~]dY.

R

Proposition B.6 (Green's Theorem for a Rectangle). Ifp and q are contin-
uously differentiable functions on a rectangle R = [a, b] x [c, d], then

II (:: -:~) ~dy f = p(x, c)~ + f q(b,y)dy

f
R

- fp(X,d)~- q(a,y)dy.

Proof. By Fubini's theorem and the fundamental theorem of calculus,

JJ::~dY = f[J:::~]dY = f[q(b,y)-q(a,Y)]dY;

R

II:~ ~dy
R
= f [J::~ dY]~ = f [p(x,d) - p(x, c)]~.
Green's theorem results by subtracting these two equations. o
Writing W = p(x,y)~ + q(x,y)dy, this says that

fJ dW 1w+l w-i w-i W,

R
=
'VI "12 "13 "14

where 'VI> 'V2, 'V3, and "/4 are the four sides of the rectangles, as in
Chapter 1.

Corollary B.7. If dw = 0, then

( w+l W i 'Y3 w+l'Y4 w.

),YI 12
B2. Partition of Unity 375

Exercise B.S. Iffis continuous on [a,b], and If(t)I:SM on [a,b], show

that II:.t(t) dtl :s M· (b - a).

The following result will be used in Appendix D:

Lemma B.9. Iff is a C(6"" function in a neighborhood of P = (a, b) in 1R2,

with f(P) = 0, then there are C(6"" functions fl and fz so that
f(x,y) = (x - a)!I(x,y) + (y - b)fz(x,y)

in a neighborhood of P.
Proof. We may assume P = (0,0). By the fundamental theorem of calculus
and the chain rule,

f(x, y) = (~- (f(tx, ty)) dt

Jo at
= x [I af (tx, ty) dt + y [I af (tx, ty) dt .
Jo ax Jo ay
The functions

fl(x, y) = (I af (tx, ty) dt

Jo ax
and fz(x, y) l laf
-(tx,ty)dt
o ay

are the required C(6"" functions. D

B2. Partition of Unity

For construction of the Mayer-Vietoris sequence for open sets in the plane,
we need the following result:

Proposition B.10 (Partition of Unity). Suppose an open set V in IRn is the

union of a sequence VI, V 2 , ••• of open sets with the property that each
point is contained in Vi for only finitely many i. Then there is a sequence
of nonnegative C(6"" functions rpi on V such that the closure (in U) of the
support of rpi is contained in Vi' and };~~I rpi == 1 on V.

(We will construct these functions so that only finitely many <Pi are nonzero
in a neighborhood of any point of V, so the sum is a well-defined C(6"" func-
tion.)
Proof. There are several steps.
376 Appendix B. Analysis

Step 1. There is a '(6'" function f on IR that is zero on the negative half line
and positive on the positive half line. Such a function is

if x>O,
if x:s O.

Exercise B.11. Verify that this is infinitely differentiable by showing that

any derivative of exp(-l/x) has the fonu (P(x)/xm)exp(-l/x) for some
polynomial P and some exponent m.

Step 2. Given a bounded rectangle (a., b.) x ... x (an, bn), there is a '(6'"
function h in IRn that is positive on the rectangle, and zero outside the rect-
angle. In fact, the function g(x) = f(x) ·f(l - x) is positive on (0, 1) and zero
outside this interval, and this leads to the required function

h(x., . .. , xn ) = IIg x·-a·

n
1=1
b (
a,
l

I
, _ I,
) •

Step 3. If A is a compact subset of V, there are '(6'" functions (Ji so that the
closure of the support of (Ji is contained in Vi' and the sum 2~=. (Ji is every-
where positive on A. To construct them cover A by a finite number of rect-
angles Ra such that the closure of each Ra is contained in some Vi' and use
Step 2 to construct ha that are positive on Ra and zero outside. Take (J. to
be the sum of those ha such that Ra is contained in V., let (J2 be the sum
of those among the others such that the closure of the support is contained
in V2> and continue in this way until all ha are used; set the other (Ji equal
to zero.

Step 4. To complete the proof, write V as an increasing union of compact

sets K. C K2 C ... as in Lemma A.20. Let Aj = K j \ Int(Kj _.) (where we set
Kj = 0 for j:S 0). Let Wj = Int(Kj +.) \ Kj- 2 • Note that Aj is a compact subset
of the open set Wj . Apply Step 3 to each compact set Aj C Wj with its open
covering {Vi n W), obtaining functions (Jij so that the closure of the support
of (Jij is contained in Vi n Wj and with 2~=. (Jij everywhere positive on Aj .
Define ljIi to be the sum 2~~. (Jij' Only finitely many (at most three) tenus
in this sum are nonzero in some neighborhood of any point, so ljIi is a '(6""
function the closure of whose support is contained in Vi . The sum
IjI = 2~=. ljIi is similarly a '(6'" function, and I\J is positive on each Aj , so it is
positive on all of V. Now

'Pi

satisfies all the required conditions. o

Exercise B.12 (Partition of Unity for Arbitrary Coverings). Suppose
B2. Partition of Unity 377

au = {U",: a E .Ill} is an arbitrary open covering of an open set U in IRn. Show

that there is a sequence of nonnegative ~oo functions 'PI , 'P2, . . . on U such
that: (i) the closure (in U) of the support of 'Pi is contained in some open
set Ua(i); (ii) for each P in U there is a neighborhood of P such that only
finitely many 'Pi are nonzero on the neighborhood; and (iii) L~= I 'Pi == 1.

Exercise B.13. Extend these results on partitions of unity to the case where
U is any manifold that has a countable basis of open sets.

Exercise B.14. For 0 <'1 <'2 construct a ~oo function IjJ on the plane that
'1
vanishes on the disk of radius centered at the origin, and is identically 1
outside the disk of radius '2 centered at the origin, and takes values in the
interval (0,1) between the two circles.
APPENDIX C

Algebra

Cl. Linear Algebra

In this book, unless otherwise stated, vector spaces are real vector spaces.
The vector space IR n consists of n-tuples (XI, . . . ,xn) of real numbers, with
coordinatewise addition and multiplication by scalars. A set of elements {e.,}
is a basis for a vector space if every element in the space has a unique
expression in the form ~aea, for some real numbers Xa with only finitely
many Xa nonzero. A vector space isfinite dimensional if it has a finite basis.
The number of elements in a basis is independent of choice of basis, and is
the dimension of the space; the dimension of V is denoted dim(V). Choosing
a basis el, . . . , en for V sets up an isomorphism of V with IW, with the
vector Xlel + ... + Xnen in V corresponding to (XI' . . . ,xn) in IRn. The
standard basis of IRn is the basis {e;} where ei has a 1 for its ith coordinate,
and zeros for the other coordinates.
If L: V~ W is a linear mapping, the kernel Ker(L) is the subspace of V
consisting of vectors mapped to zero, and the image Im(L) is the subspace
of W consisting of vectors that can be written L(v) for some v in V.
The rank-nullity theorem asserts that if L: V ~ W is a linear mapping of
finite-dimensional vector spaces,
dim(Ker(L» + dim(lm(L» = dim(V).
If W is a subspace of a vector space V, the quotient space V /W is defined
to be the set of equivalence classes of elements of V, two vectors in V being
equivalent if their difference is in W. This set V/W has a natural structure
of a vector space, so that the mapping from V to V/W that takes a vector

378
C 1. Linear Algebra 379

to its equivalence class is a linear mapping of vector spaces. The kernel of

this mapping from V to V /W is W.
Conversely, if V ~ U is a surjective linear mapping of vector spaces, and
W is the kernel, this determines an isomorphism of V/W with U.
Suppose L: V ~ V'is a linear mapping of vector spaces, and W is a sub-
space of V, and W' a subspace of V'. If L(W) is contained in W', then L
determines a linear mapping.
V/W ~ V'/W'
of quotient spaces, which takes the class of v in V to the class of L(v) in
V'.
If V and Ware vector spaces, the direct sum VE9W can be defined as the
set of pairs (v, w), with v in V and w in W, with addition defined by
(v, w) + (v', w') = (v + v', w + w'), and multiplication by scalars by
r' (v, w) = (r' v, r' w). For example, IR' is the direct sum of n copies of R
More generally, given any collection Va of vector spaces, for IX in some
index set .sIl, an element of direct sum E9Va is determined by specifying a
vector Va in Va for each IX in .sIl, with the added condition that Va can be
nonzero for only finitely many IX. Addition and multiplication by scalars are
defined component by component, as for two factors. The same definition,
but without the restriction that only finitely many are nonzero, defines the
direct product, denoted IIVa •
For vectors u = (XI, ... ,x,) and v = (YI' ... ,y,) in IR', the dot product
is the number U' v = XIYI + ... + XnYn' The length of u is Ilull = VU:U. The
projection of u on v is the vector tv, where t = (u' v)/(v' v); the length of
this projection is lu· vl/llvll.
An m by n matrix A = (ai.j) determines a linear mapping L: IRn ~ IRmthat
takes ej to L(e) = 'J.7'=laj.jej. Every linear mapping from IR' to IRm arises from
a unique such matrix. If M: IRm~ IRI corresponds to an I by m matrix B = (aj,k),
the composite MoL: IR n~ IRI corresponds to the product matrix B . A, where
the (i, J) entry of B' A is 'J.~~lbi.kak.j' We need this mainly for (2 x 2) ma-
trices, where a matrix A = [~ ~] is a matrix, the linear mapping corre-
sponding to A takes a vector.v = (x,y) to the vector (ax + by, ex + dy). The
determinant of A, denoted det(A), is ad - bc. If the determinate is nonzero,
A is invertible, with inverse

A-I = _1_[ d -b].

ad-bc -c a
The invertible (2 x 2)-matrices form a group, denoted GL2 R This group
has a topology, determined by its embedding as an open subset of 1R4 : the
complement of the set of (a, b, c, d) with ad - bc = O.

Exercise C.I. Show that the product GLzIR x GLzIR~GLzIR, A XB~A 'B,
and the inverse map GL21R~ GL21R, A ~ A -I, are continuous mappings.
380 Appendix C. Algebra

Lemma C.2.1f det(A) > 0, there is a path y. [a, b] ~ GL21R such that ')'(a) = A
and y(b) = [b ~J. If det(A) < 0, there is a path y: [a, b] ~ GL21R such that
')'(a)=Aand')'(b)=[b -~J.
Proof. We will find a sequence of paths, each taking the matrix to a simpler
one. For example, by multiplying a column of a matrix by t, with t varying
in [a, 13], for a and 13 positive, we can find a path changing the lengths of
the column. In particular, we can assume the first column of A is a unit
vector, so it can be written in the form (cos({}), sin({})) for some {} in [0, 2'IT].
Then the path

(t) = [COS(t) Sin(t)] . A

"( -sin(t) cos(t) ,
takes A to a matrix whose first column is (1,0). Then one can gradually
project the second column on the line perpendicular to the first, via the path

() = [ 1 (1 - t)b]
"(t 0 d '
to get to a matrix where the second column is (0, d). Changing the length
of the second column as at the beginning, we can get it either to (0,1) or
to (0, -1), as asserted. 0

Problem C.3. Generalize to GLn IR, showing that GLn IR has two connected
components for all n;::' 1.

C2. Groups; Free Abelian Groups

A set of elements in a group G generates G if every element in G can be
written as a (finite) product of elements in the set and inverses of elements
in the set.
If H is a subgroup of a group G, a left coset is a subset of G of the form
g. H = {g. h: hE H}. The group G is a disjoint union of its left cosets; gl
and g2 are in the same left coset exactly when there is an element h in H
with gl . h = g2. The set of left cosets is denoted by G /H. There is a natural
map 'IT from G onto G /H that takes an element in G to the coset containing
it. A subgroup H is a normal subgroup if, for all g in G and h in H, g . h . g-I
is in H. In this case G/H gets the structure of a group, in such a way that
the natural map 'IT: G~G/H is a homomorphism of groups.
The identity element in a group G is usually denoted bye, or eG if there
is chance of confusion. If <p: G ~ G' is a homomorphism of groups, the
kernel N = {g E G: <p(g) = ed is a normal subgroup of G, denoted Ker(<p).
Then <p determines a one-to-one homomorphism ip: G /Ker( <p) ~ G' such that
C2. Groups; Free Abelian Groups 381

<p = (j) 0 1T. If <p is surjective, then (j) is an isomorphism. More generally, if
N is any normal subgroup of G, a homomorphism from G/N to a group G'
determines a homomorphism from G to G' such that N is contained in its
kernel. The image <peG) of any homomorphism <p: G~ G' is a subgroup of
G', denoted Im(<p), and <p determines an isomorphism of G/Ker(<p) with
Im(<p). If NeG and N' C G' are normal subgroups, and <p: G~ G' is a ho-
momorphism such that <peN) eN', then <p determines a homomorphism
(j): G/N~G'/N'.
The set of homomorphisms from G to G' is denoted by Hom(G, G'). So
if N is a normal subgroup of G,
Hom(G/N,G') ~ {<pEHom(G,G'):<p(N)=ed.
An important normal subgroup of a group G is the commutator subgroup,
denoted [G, G]. This consists of all finite products
g,h,g,-'h,-' . g2h2g2 -lh2-I •...• g.h,.g. -'h. - I ,

for elements g I, hi, gh h2' . . . , g., h. in G. The normality of this subgroup

comes from the identity g. (ab)· g-I = (g. a· g -I). (g. b· g-I). If A is an abelian
group, any homomorphism of G to A sends all commutators to the identity,
so
Hom(G/[G,G],A) ~ Hom(G,A).
We usually use an additive notation for the product in abelian groups,
writing g + h instead of g. h, with the identity element denoted o. The group
of integers under addition, which is the infinite cyclic group, is denoted l.
The abelian group with just one element is often denoted o. If A is an abelian
group, and X is any set, the set of functions from X to A has a natural
structure of abelian group, with (f + g)(x) = f(x) + g(x). In particular, if G
is any group, the set of homomorphisms Hom(G,A) has the structure of an
abelian group.

Exercise C.4. If <p: G~ G' is a homomorphism of groups, show that the

mapping from Hom(G',A) to Hom(G,A) that takes IjI to ljIo<p is a homo-
morphism of abelian groups.

If A and B are abelian groups, the direct sum A EB B is the set of pairs
(a, b), with a in A and b in B, with addition defined by
(a, b) + (a', b') = (a + a', b + b'). More generally, given any collection Aa
of abelian groups, the direct sum EBAa consists of collections faa}, with aa
in A a , with the condition that aa can be nonzero for only finitely many 0:.
Addition is defined component by component, as for two factors. For ex-
ample, l' is the direct sum of n copies of l. The same definition, but with-
out the restriction that only finitely many are nonzero, defines the direct
product, denoted IIAa.

Exercise C.S. If an abelian group C contains subgroups A and B such that

382 Appendix C. Algebra

every element of C can be written as a sum of an element in A and an

element in B, and A n B = {O}, show that A EEl B is isomorphic to C.

Exercise C.6. For any collection A" of abelian groups, and any abelian
group B, construct an isomorphism
Hom(EElA" ,B) == II Hom(A" ,B) .
An abelian group A is a free abelian group, with basis {e a }, if every ele-
ment in the group has a unique expression in the form Lnaea , for some in-
tegers na, with only finitely many na nonzero. If the number of elements in
a basis is a finite number n, we say A is a free abelian group of rank n. As
we will see below, this number is independent of choice of basis.

Exercise C.7. (a) If A and B are free abelian groups, show that AEElB is
free abelian, and if the ranks are finite, rank(AEElB) = rank(A) + rank(B).
(b) If F is free abelian, and A is abelian, and Ij): A - F is a surjective ho-
momorphism, show that A is isomorphic to the direct sum of F and Ker(Ij)).

More generally, a set {e,,} of elements in an abelian group A is called

linearly independent if no linear combination of them is zero, i.e., there is
no set of integers {n a }, with only finitely many nonzero, but not all zero,
such that Lnae" = O. The maximum number of elements in a linearly inde-
pendent set in A is called the rank of A . We will prove at the end of this
section that any two maximal linearly independent sets have the same num-
ber of elements, at least when this number is finite.
Unlike the case with vector spaces, however, a maximal set of linearly
independent elements in an abelian group need not generate the group. For
example, a finite abelian group has no independent elements. Even for groups
with no elements of finite order, however, it is not true:

Exercise C.S. Show that the rank of the abelian group II) of rational numbers
is 1.

For any set X, the free abelian group on X, denoted F(X), can be defined
as the set of finite formal linear combinations Lnxx, with nx integers, the
sum over a finite subset of X. The addition is defined coordinate-wise:
Lnxx + Lmxx = L(nx + mx)x. More precisely, define F(X) to be the set of
functions from X to 1L. that are zero except on a finite subset of X. This is
an abelian subgroup of the abelian group of functions from X to 1L.. To the
function f: X -1L. is associated the expression 'i/(x)x. The element x cor-
responds to the function that takes x to I and all other elements of X to zero.
These elements form a basis for F(X).
For any abelian group A, and any function Ij) from a set X to A, there is
a unique homomorphism from F(X) to A that takes LnxX to Lnxlj)(x). In par-
ticular, if Ij): X - Y is any function, it determines a homomorphism from
C2. Groups; Free Abelian Groups 383

F(X) to F(Y), taking ~nxX to ~nx<p(x), or taking n\x\ + ... + n,xr to

n\<p(x\) + ... + nr<p(xr).

Exercise C.9. If <p: X ~ Y is one-to-one, show that F(X) ~ F(Y) is one-to-

one, and if <p: X ~ Y is surjective, show that F(X)~ F(Y) is surjective.

If A is an abelian group, then the set of homomorphisms Hom(A, IR) from

A to IR forms a real vector space, with scalar multiplication by the rule
(rf)(x) = r ·f(x) forfin Hom(A, IR), r a real number, and x inA. If <p: A~A'
is a homomorphism of abelian groups, then there is a linear mapping
<p*: Hom(A', IR)~ Hom(A, IR) of vector spaces, defined by the formula
<p*(f) = fo <po

Lemma C.I0.lffP:A~A' is one-to-one, then fP* is surjective.

Proof. We need some preliminaries. There exists a set W3 = {x,,: a E stl} of

elements in A', such that:

(i) no fmite linear combination ~n..x" with integer coefficients is in the

image of <p unless all nIl are zero; and
(ii) W3 is maximal with this property.

This is a consequence of Zorn's lemma, exactly as in the proof that every

vector space has a basis. Of course, there may be many such sets W3, but
we fix one. It follows that for any element x in A', there is a nonzero integer
n and integers n", all zero except for finitely many, so that nx - ~n..x" is
in <p(A); otherwise one could enlarge W3 by adding x to it. Therefore for any
x in A' there is at least one equation of the form

(C.Il)

with y in A, n not O.
Given f in Hom(A, IR), we define g in Hom(A', IR) by setting

1
g(x) = - (f(y)) ,
n

for any integer n;60 and yEA so that (C.Il) holds. To see that g is well
defined, suppose also

mx = ~m..x" + <p(z) ,

with z EA and m;6 O. Then

(mnx - m<p( y)) - (nmx - n<p(z))

<p(nz) - <p(my) = <p(nz - my) .
384 Appendix C. Algebra

By (i), this element must be zero, and since q> is one-to-one, nz must be
equal to my. Therefore,
I 1 I 1 I
-(f(z» -(f(nz» -(f(my» -(mJ(y» -(f(y» ,
m mn mn mn n
as required.
The proof that g is a homomorphism is similar, for if x and x' are two
elements of A', write

Then mn(x ± x') = ~(mna ± nma)xa + q>(my ± nz), so

I I I
g(x ± x') = - (f(my ± nz» = - J(y) ± - J(z) = g(x) ± g(x') .
mn n m
And, by definition, if x = q>(y), then g(x) = J(y) , so q>*(g) = J. D

Corollary C.12. Suppose S is a set with n elements in an abelian group A,

and S is a maximal set oj linearly independent elements. Then the dimension
oj Hom(A, IR) is equal to n. In particular, any two maximal sets oj linearly
independent elements in A have the same number oj elements.
Proof Let F be the free abelian group on S, and q>: F~A the natural map.
The linear independence of S assures that q> is one-to-one. The maximality
of S assures that the set 00 considered in the proof of the lemma is empty.
The proof of the lemma shows that the map q>* from Hom(A, IR) to Hom(F, IR)
is an isomorphism. The functions that take value I on a given element in
S, and value 0 on the other elements, give a basis of Hom(F, IR) with n
elements, and this shows that the dimension of Hom(A, IR) is n. Note that
if S were a maximal set of linearly independent elements that were infinite,
the same argument shows that Hom(A, IR) == Hom(F, IR) is infinite
dimensional. D

Exercise C.13. Show conversely that if Hom(A, IR) has finite dimension n,
then A has rank n.

Exercise C.14. If B~C is surjective, with kernel A, show that if two of

the three abelian groups A, B, and C have finite ranks, so does the third,
and rank(B) = rank(A) + rank(C).

Given homomorphism q>: A~B and 1/1: B~C, one says that the sequence
A ~ B ~ C is exact, or exact at B if the image of q> is equal to the kernel
of 1/1. To say that the sequence O~ A ~ B is exact is the same as saying the
map from A to B is one-to-one, and to say that A~B~O is exact is the
same as saying the map from A to B is surjective.
C3. Polynomials; Gauss's Lemma 385

Problem C.1S. If A-B-C is exact at B, show that the dual sequence

Hom(C, IR)- Hom(B, IR)- Hom(A, IR) is exact at Hom(B, IR).

A sequence Ao-A I - • • • -An-An+ 1 of abelian groups and homo-

morphisms between them is called exact if it is exact at each of the groups
A;, for 1:s i:Sn.

Problem C.16. Show that if O-A I - • • • -An-O is exact, and each of

the abelian groups A; has finite rank, then

2: (-Ii rank(A;)
n
O.
;=1

C3. Polynomials; Gauss's Lemma

For any field K the ring of polynomials K[X] in a variable X over K is a
unique factorization domain. In fact, every nonzero Pin K[X] has a unique
factorization P = a' lIP;"i, with a in K and each P; an irreducible polynomial
that is monic. 9 This follows from the fact that one has a division algorithm
for polynomials, just as one has for integers. In particular, any finite col-
lection of nonzero polynomials has a greatest common divisor, which is
unique if, in addition, it is required to be monic. The quotient field of K[X],
consisting of all ratios P /Q, Q ~ 0, is denoted K(X).
The ring of polynomials K[X, Y] in two variables X and Y is a subring of
the ring K(X)[y], which, by what we have just seen, is a unique factori-
zation domain.

Lemma C.17 (Gauss). Let F be a polynomial in K[X, Y]. If F is irreducible

in K(X) [Y], then F is irreducible in K[X, Y].

Proof. Given F in K[X, Y], write F = ao(X) + al(X)Y + ... + an(X)y n, with
a;(X) in K[X]. The greatest common divisor of ao(X), ... , anCX) is called
the content of F, and denoted c(F). Call F primitive if c(F) = 1.
We show first that the product of two primitive polynomials is also
pnmltlve. To see this, suppose F = ao + al Y + ... + anyn and
G = bo + b l Y + ... + bmym are primitive. Suppose a nonconstant polyno-
mial p = p(X) divides all the coefficients of F· G. Take the minimal i and
j such that p does not divide a; and bj • Then the coefficient of yi+j in F· G
has the form

9 A monic polynomial is one of the form Xm + a1X m- 1 + ... + am'

386 Appendix C. Algebra

Since all the tenns but the first are divisible by p, and the first is not, this
is a contradiction.
It follows from this that for any two polynomials F and G in K[X, Y),
c(F' G) = c(F)' c(G). To see this, write F = c(F)· F l , G = c(G)' G l , with
Fl and G l primitive. Then F· G = c(F) . c(G)' Fl' G l , and Fl . G l is prim-
itive, from which it follows that the content of F· G is c(F)· c(G).
Given any nonzero G in K(X)[Y) , one can write G = g . G l , with g in
K(X) and G l a primitive polynomial in K[X, Y). Now suppose F is an ir-
reducible polynomial in K[X, Y), and that F factors in K(X)[Y) into G· H,
with both G and H of positive degree in Y. Write G = g' G l , H = h· HI,
with G l and HI primitive, and g = p/q, h = r/s, withp, q, r, s EK[X). Then
q·s·F=p·r·GI·H I in K[X,Y). It follows that q·s·c(F)=p·r. Hence
F = c(F)' Gl • HI, which contradicts the irreducibility of F in K[X, Y). D

Exercise C.tS. Show that, for any field K, K[X, Y) is a unique factorization
domain. Generalize to polynomials in n variables.

If P is a polynomial in K[X), the residue class ring K[X)/(P) is the set

of equivalence classes of polynomials in K[X), two being equivalent when
their difference is divisible by·P. The residue classes K[X)/(P) have the
structure of a ring so that the natural map K[X]-K[X]/(P) is a homo-
morphism of rings.

Lemma C.t9. If P has degree n, then the images of I, X, ... ,x n- l form

a basis for K[Xl/(P) over K.
Proof. These elements span, since, by dividing by P, any polynomial is
equivalent to a polynomial of degree less than n. They are linearly inde-
pendent, since no nonzero polynomial of degree less than n is divisible
by P. D

Exercise C.20. Show that K[X]/(P) is a field if and only if P is irreducible.

APPENDIX D

On Surfaces

D 1. Vector Fields on Plane Domains

The object of this section is to prove Lemmas 7.10 and 7.11; we refer to
Chapter 7 for notation. Suppose <p: U -') U ' is a diffeomorphism from one
open set in the plane onto another. If <p(x,y) = (u(x,y), v(x,y» in coordi-
nates, at any point Pin U, we have the Jacobian matrix

au (P) au (P)]
J~.P =
[ ~~ ~~ ,
-(P) -(P)
ax oy
which we regard as a linear mapping from vectors in ~2 to vectors
in ~2 (see Appendix C). If V is a continuous vector field in U, the
vector field 'P* V in U' is defined by the formula
(<p*V)(P' ) = J~.p(V(P»,

where P is the point in U mapped to p' by cp, i.e., P = cp-l(p l ) . If V has

singularities in the set Z, cp* V will have singularities in cp(Z).

Lemma D.l. Index~(p)(cp*V) = IndexpV.

Proof. There is no loss in generality by assuming that P and P' are the origin
0, that U is a disk containing the origin, and that V is not zero in U\{O}.

387
388 Appendix D. On Surfaces

Let J be the Jacobian of 'P at O. Our first goal is to show that

(D.2)

This will reduce the problem to the easier case of a linear mapping. We
want to construct a homotopy from 'P to J. Define

K: UX [0, 1] ~ 1R2, QXt ~ { ~'P(t.

t
Q), O<t::::; 1,
J(Q), t=O.

Claim D.3. This mapping K is C(6"'.

To prove this claim, we use Lemma B.9, replacing U by a smaller disk

if necessary, so we can write

with C(6'" functions Ul, U2, VI , and V2' Then

K«x, y) x t) = (XUI(tx, ty) + YU2(tx, ty), XVI(tX, ty) + yvitx, ty»
for all 0:5 t:5 I, and this expression is clearly C(6"'.

Now H(Q x t) = (K, hey) gives a homotopy from J *V to 'P* V in the sense
of Exercise 7.3, and (D.2) follows from that exercise.
We are therefore reduced to showing that Indexo(J* V) = Indexo(V) for any
invertible linear mapping J. Now we use Lemma C.2 to know that there is
a path in the space of invertible matrices from J either to the identity matrix
I, or to the matrix I' = [~ ~I J. If such a path is given by a formula t ~ J"
a:5 t:5 b, then the homotopy H(Q x t) = (J,)*(V) gives a homotopy from
J*V to I*Vor to I'*V, and the same exercise shows that the index doesn't
change. Of course I*V= V, so all that remains is to prove that
IndeXo(I' *V) = IndeXo(V).
If Vex, y) = (p(x, y), q(x, y», then by the definition of I' *,
(I'*V)(x,y) = (p(x, -y), -q(x, -y».

So one is reduced to the elementary problem of showing that if

F(x,y) = (p(x,y), q(x,y», and R(x,y) = (x, -y), the mappings RoFoR and
F, when restricted to a small circle, have the same winding number around
the origin. This is easy to do directly from the definition of winding number,
and we leave the details as an exercise. (This is also special case of the fact
that the degree of a composite of mappings of circles is the product of the
degrees of the mappings, see Problem 3.27. In this way one can argue di-
rectly with any linear mapping J, since J *V = J 0 VO J -[ .) D

Now we consider the other lemma from Chapter 7.

D2. Charts and Vector Fields 389

Lemma D.4. Suppose V and Ware continuous vector fields with no sin-
gularities on an open set U containing a point ~ Let D CUbe a closed
disk centered at,.!'. Then there is a vector fieJd V with no singularities on
U such that (i) V and V agree on U \ D; (ii) V and Wagree on some neigh-
borhood of P.

Proof. Suppose first that the dot product V(P) . W(P) is positive. Shrinking
the disk, we may assume that V(Q) . W(Q) > 0 for all Q in D. As in Step 2
of the proof of Proposition B.lO (see Exercise B.14), there is a ~~ function
p that is identically 1 in a neighborhood of P, and identically 0 outside D,
and taking values in [0,1]. Let

-
V (Q) = (1 - p(Q»V(Q) + p(Q)W(Q).

Then V(Q)' V(Q) > 0 for all Q in U, so V has no singularities, and con-
ditions (i) and (ii) are clear.
For the general case, it therefore suffices to find a vector field VI with
no singularities that agrees with V outside D, and such that VI(P)' W(P) is
positive. This can be done by rotating V inside D. With the same func-
tion p, and -fr the angle from the vector V(P) to the vector W(P), we can
take

[ COS(P(Q)-fr) -Sin(p(Q)-fr)] . V(Q)

D
sin (p(Q)-fr) cos(p(Q )-fr) .

D2. Charts and Vector Fields

We start with a brief definition of a (smooth) surface X, and define what

we mean by a vector field on X and the index of a vector field at a point
on X. A surface X with an atlas of charts is a Hausdorff topological space,
equipped with a collection of homeomorphisms

with Ua open in the plane [R2, and 'Pa(Ua) open in X; the a are in some index
set. The surface X should be the union of these open sets 'Pa(Ua). Let
Ua~ = 'Pa -1('Pa(Ua) n 'P~(U~». These charts determine change of coordi-
nate mappings 'P~a = 'P~ -I 0 'Po<' which are homeomorphisms from U0<13 to
U13a •
390 Appendix D. On Surfaces

qIOl

U~U~ _____________
qlP_Ol_________ ~~ ~.

To give X a differentiable, or smooth, structure, the requirement is that these

changes of coordinates CPJ3a should be '(6'" mappings for all n and ~. One
then has a notion of a differentiable function on an open subset U of X: it
is function f such that f 0 CPa is differentiable on CPa -1 (U) n Ua for all n.
Another collection of charts {Wa': Ua'~ X} is said to be equivalent to this
one if all the changes of coordinates from one to the other are all '(6"', i.e.,
all CPa -1 0Wa' are '(6'" where defined. We say that this collection defines the
same surface. More precisely, a smooth (or '(6"') surface is the topological
space X together with an equivalence class of families of charts. For the
sphere S2, the two mappings cP and Wwe obtained from stereographic pro-
jection in §7c form a family of charts. Stereographic projection from other
points gives charts that are compatible in this sense.

°
If f is a '(6'" function on some open set in 1R3 , and X is the locus where
f(x,y, z) = and grad(f) ¥ 0, then X is a smooth surface. If for example
(aflaz)(p) ¥ 0, the implicit function theorem says that projection from X to
the xy-plane is locally one-to-one near P, and the inverse to this projection
provides a chart near P.
If X is a surface given by charts as above, a vector field V on X can be
defined as a compatible collection of vector fields Va in each of the coor-
dinate neighborhoods Ua' The compatibility means that

(CPJ3a)*(Valuap) = vJ3lup.'

where (CPJ3a)* is defined as in the preceding section. The vector field has a
singularity at a point P in X if, with some n such that CPa(Pa) = P, the cor-
responding Va has a singularity at Pa in Ua . By Lemma D.I the index of
Va at Pais independent of choice of coordinate chart. This index is defined
to be the index of V at P.
The surface X is orientable if it has an atlas of charts such that all the
determinants of the lacobians of the change of coordinate mappings are pos-
itive. An orientation is a choice of such an atlas, with two atlases defining
D3. Differential Fonns on a Surface 391

the same orientation if all changes of coordinates from one to the other have
positive detenninants of Jacobians. An orientable surface has two orienta-
tions, with two orientations defining the opposite orientation if all changes
of coordinates from one to the other have negative Jacobian detenninants.

D3. Differential Forms on a Surface

If '1': u~ U' is a diffeomorphism from one open set in the plane to another,
and 00 = fdx + gdy is a f.(j,oo I-fonn on U', one can define a pull-back I-fonn
'1'*00 on U by the fonnula

'P*(fdx+gdy) = f('P(X,y)).(au(X,y)dx+ au (x, y)dy)

ax ay
+ g('P(x,y». (av (x,y)dx + av (x,y)dy)
ax ay
= ((f0'P). au + (go '1') . av) dx
ax ax
+ ((f ° '1') . au + (g ° '1') • av) dy ,
ay ay
where 'P(x,y) = (u(x,y), v(x,y». Similarly, if 00 = hdxdy is a 2-fonn on U',
the pull-back 2-fonn '1'*00 on U is defined by the fonnula

auav auav)
'P*(hdxdy) = (h°'P). ( - - - - - dxdy.
axay ayax

Exercise D.S. (a) If 1jI: U' ~ U' is a diffeomorphism, and 00 is a I-fonn or

2-fonn on U', show that 'P*(IjI*oo) = (ljIo '1')*(00). (b) Show that 'P*(dj) = d('P*f)
for a f.(j,oo functionf on U', and 'P*(doo) = d('P*oo) for 00 a I-fonn on U'.

Given a surface X with an atlas of charts 'Pa: Ua ~ X as in §D2, a function

(or O-fonn) is given by a collection of functions fa on Ua such that they
agree on the overlaps: fa = fll ° 'Plla on Uall· Define a onejorm 00 on X to be
a collection of I-fonns OOa on Ua that agree on the overlaps, i.e., such that
ooalua~ = ('Plla)*(oolllu",,)
for all pairs a and 13. A two-form is defined likewise, taking the OOa to be
2-fonns on Ua.
If f is a f.(j," function (or zerojorm) on X, its differential df is a I-fonn,
defined to be the I-fonn d(f°'Pa) on Ua • Similarly, if 00 is a I-fonn on X,
given by I-fonns OOa on Ua, the differential of 00 is the 2-fonn doo defined
to be the 2-fonn dOO a on Ua •
392 Appendix D. On Surfaces

Exercise D.6. (a) Verify that these formulas define I-forms and 2-forms on
X. (b) Verify that d is linear, i.e., d(riwi + r2w2) = rld(wl) + r2d(w2) for real
numbers rl and r2 and O-forms or I-forms WI and 002' (c) Verify that d(df) = O.
(d) Show that for k = 0, I, and 2, a k-form for one atlas determines a k-
form for any other atlas, and that this is compatible with the definition of
differential.

There is also a wedge product /\ that takes two I-forms 00 and IL and
produces a 2-form 00/\ IL. For an open set U in the plane, if 00 = fdx + gdy,
and IL = hdx + kdy, for f, g, h, and k C(&'" functions on U, then
w/\IL = (fdx+gdy)/\(hdx+kdy)=(f·k-g·h)dxdy.
If 00 and IL are I-forms on X given on Ua by Wa and lLa, respectively, then
00 /\ IL is the 2-form given on Ua by Wa /\ lLa, with Wa /\ lLa defined by the
displayed formula. Iffis a function and 00 is a I-form (or 2-form), then/' 00
(defined locally by fa' W a) is a I-form (or 2-form).

Exercise D.7. (a) Verify that W/\IL is a 2-form. (b) Verify the following
properties of the wedge product:
(i)
for 001,002, and IL I-forms, andfl andf2 functions;
(ii) IL/\W = -w/\IL
for IL and 00 I-forms; and
(iii)
for f a function and IL a I-form.

In fact, all the results of this appendix generalize from two to n dimen-
sions, leading to the notion of a smooth manifold of dimension n, a vector
field on a manifold, the index of a vector field, an orientation of a manifold,
k-forms on a manifold (0::::; k::::; n), with differential from k-forms to (k + 1)-
forms, wedge products of k-forms and I-forms being (k + I)-forms, with sim-
ilar properties.
APPENDIX E

Proof of Borsuk's Theorem

This appendix contains a proof of Borsuk's theorem as stated in §23c. It

assumes a knowledge of §23a and §23b.
So far in this book we have considered chains 2:n;fj with integer coeffi-
cients n j • In fact, one can use coefficients in any abelian group G, and one
gets chains Ck(X; G), cycles Zk(X; G), and boundaries Bk(X; G), so homol-
ogy groups Hk(X; G) = Zk(X; G)/Bk(X; G). All the formal properties proved
about ordinary homology groups extend without change to these groups, and
many of the calculations are similar. For example, the Mayer-Vietoris theo-
rem is true without change.
It is often useful to look at coefficients in 71./p71., the integers modulo a
prime p. In this case there is a natural homomorphism from each HiX) to
Hk(X; 71./p71.,) obtained by reducing all coefficients modulo p. This gives
homomorphisms

These homomorphisms are isomorphisms if X is a sphere, as one sees by

tracing through the Mayer-Vietoris argument computing the homology of a
sphere. The following exercise shows that it is not always an isomorphism,
however.

Exercise E.1. (a) Show that the above map is an isomorphism if X is a

compact oriented surface. (b) Show that, for X the real projective plane, or
any compact surface, H 2(X; 71./271.) = 71./271..

These homology groups are particularly useful when p = 2. In fact, every-

393
394 Appendix E. Proof of Borsuk's Theorem

thing becomes a little easier, since one can ignore all signs entirely. Since
these are the only ones we will consider here, we denote them by RkX:
RkX = Hk(X; 1./21.).
Similarly, we denote by C~ the chain complex C*(X;1./2l.).
It follows from the isomorphism Hn(Sn)/2Hn(sn) == Rn(sn) that for any
continuous map f: Sn~sn, the degree of f is odd if the homomorphism
f*: Rn(sn) ~ Rn(sn) is not zero, and even if f* is zero. We denote the an-
tipodal map by
T: Sn ~ So.

Let 11": sn~ !RlIln be the two-sheeted covering map that identifies antipodal
points. We will prove Borsuk's theorem by using the chain complex and
homology of these spaces and maps with coefficients in 1./21.. For this we
need a general lemma.
Let 11": X ~ Y be any two-sheeted covering map, and let T: X ~ X be the
map which interchanges the two points in 11" -I(p) for each P in Y. The cor-
responding map 11"*: C~ ~ C*Y of chain complexes is surjective. In fact,
if f: Ik~ Y is a k-cube, there are exactly two k-cubes Al and A2 = To Al in
X with 11" 0 Aj = f. This is proved exactly as we proved the lifting of hom-
otopies in § lIb. The mapping f 1-+ Al + A2 determines a homomorphism
t*: C*Y~ C*X of chain complexes, called the transfer (see Problem 18.26).
A chain is in the kernel of 11"* exactly when each cube To A occurs with the
same coefficient as A. We therefore have an exact sequence of chain com-
plexes

(E.2)

This gives a long exact sequence in homology:

Lemma E.3. (i) Lett X~X be a continuous map such thatfoT= Tof, and
let g: Y~ Y be the map determined by the condition go 7T = 7T Of. Then the
diagram
0 - c * y ~CX ~c*y - 0
*
g*! f.! g.!

0 - C . Y ~C*X ~cy - 0
commutes.
•
(ii) Let t X ~ X be a continuous map such that fo T = f, and let g: Y ~ Y
be the map determined by the condition g07T= 7T Of. Then the diagram
Appendix E. Proof of Borsuk's Theorem

0
395

~ C*X ~ c*y -0
01
-c*y

0- -0
f*l g*l

c*y ~c*x ~ C*Y

commutes.
Proof. Both of these are straightforward from the definitions. The right squares
commute by functoriality: g* 'IT * = (g 'IT)* = (fo 'IT)* = f* 'IT *. For the left
0 0 0

squares, let f be a cube in Y, AI and A2 its two liftings, so

f*t*[f] = [fo AI] + [fo A2]. In case (i), fo AI and fo A2 are the two liftings
of go f, so t*g*[f] = [fo Ad + [fo A2], as required. In case (ii),
foA 2 = foToA I = foAl' so f*t*[f] = 2[foAd = o. 0
It follows from this lemma that the corresponding maps between the long
exact homology sequences commute. We apply this now to X = sn, Y = IRpn,
'IT: X ~ Y the covering, and T the antipodal map. The long exact sequence
arising from (E.2) takes the form

o- HnY ~ fi.x ~ Hn Y ~ fin_IY - 0

-
0 - Hn_IY -
a -Hn_ Y - 2 0 - ...
- a t* - - 1t -
- 0- HIY - HoY ~ HoX ~ HoY - - O.
Since HnX= 7L127L, we must have HnY¥-O, so t*: HnY~HnX is an iso-
morphism. Hence 'IT*:HnX~HnY is zero, so a:HnY~Hn-IY is an iso-
morphism. Continuing, we see that a: H;Y~H;_IY is ari isomorphism for
all i = I, ... , n. In particular, H;Y = 7L127L for i = 0, ... , n.
Now supposef: S"~Sn is a map withfoT= Tof. Borsuk's theorem (Theorem
23.24(a» states that the degree of f is odd, which is equivalent to saying
thatf*: HnX~HnX is an isomorphism. Applying Lemma E.3(i), we have
commuting diagrams
fiY~HX
n ==

1 1
11

g* f*
- - t*
HnY~ Hn X ,

the right diagrams valid for i = I, ... , n. Since g*: HoY ~ Ho Y is an iso-
morphism, it follows from the right squares and induction on i that the ho-
momorphism g*:H;Y~H;Y is an isomorphism for every i=O, ... , n.
Then the left square implies thatf*: HnX ~ HnX is an isomorphism as well,
and this completes the proof.
Similarly, iff: sn~sn is a map withfoT= f, we have by (ii) of the lemma
a commutative diagram
396 Appendix E. Proof of Borsuk's Theorem

iif~HX

01 = 1·1
Hnf ~ .. H~.
This implies that f*: Iinx ~ Iinx is zero, which means that f has even de-
gree, and completes the proof of Theorem 23.24. 0

The constructions of this section are part of a general development of P. A.

Smith to study spaces equipped with periodic transformations like T. For a
proof of Borsuk's theorem using simplicial approximations, see Armstrong
(1983). For a proof using differential topology, see Guillemin and Pollack
(1974).
Hints and Answers

Hints and/or answers are given for some of the exercises and problems,
especially those used in the text, or those that are hard.

0.1. Hint: The answer depends only on the numbers of edges that emanate
from each vertex. What happens to these numbers when you travel, erasing
the edges as you travel over them? When do you get stuck at a vertex?
Answer: There is always an even number of vertices such that the number
of edges emanating from the vertex is odd (if an edge has both ends at a
vertex, it counts twice). If this number is greater than 2, the graph cannot
be traced. If the number is 2, it can be traced, but only by starting at one
of these, and (necessarily) ending at the other. If the number is 0, you can
start anywhere, and will end at where you start. To see that one can do it
under these conditions, one way is to make any trip, starting at an odd vertex
if there are two such, continuing until you get stuck. Then make another
trip, but adding a side trip along untraveled roads, until you (necessarily)
get back to the old route at the same point. Each trip becomes longer, until
the whole is traced.
0.2. See Chapter 8.

1.6. All but (vi).

1. 7 . Yes. Find such a function by integrating.
1.9. For the challenge, if P is in the closure of the points one can connect
to Po by such an arc, take a disk D around Pin U, take an are from Po to

397
398 Hints and Answers

a point inside D, and look at the first time the arc hits the boundary of D
with inward pointing tangent; splice on to this arc (see §B2 for similar con-
structions) to get to any point inside D.
I. 13. See Chapter 9 for more general results.
1.20. For the challenge, use the law of cosines.
2.3. Show that the derivative of {}(t) must be
-y(t)x'(t) + x(t)y'(t)
X(t)2 + y(t)2
Or show that two such functions differ by a multiple of 271', which must be
o by continuity.
2.9. Consider neighborhoods given in polar coordinates by {}, < {} < {}2 and
r, <r<r2, with {}2-{}' <271'.

2.13. See §B2 for the construction of such functions.

2.19. For formulas see Chapter 12, and use Problem 2.13.
2.22. Either argue directly, as in Appendix B, or use polar coordinates to
map a rectangle onto the disk, and integrate the pull-back of the I-form as
in the first proof of Proposition 2.16.
2.24. Apply Green's formula (i) with g = f.
2.25. Hint: Apply Green's formula (ii), where R is the region inside the
disk and outside a small disk around the point, with g of the form a + b log(r),
where r is the distance from the center of the disk. Pass to the limit as the
radius of the small disk approaches O.
3.4. Use the definition. Choices of subdivision and sector Vi and {}i for "Y
and P determine the same subdivision for "Y + v, choices Vi + v for sectors,
and translated angle functions, so that the changes in angle along each piece
are the same for each.
3.5. Use Exercise 2.9. See §llb for a generalization.
3.7. See Chapter 12 for formulas.
3.10. Apply the Lebesgue lemma to "y0Cf', to obtain a subdivision
S; tn' = b' , such that "Y ° Cf' maps each subinterval into a sector.
a' s; to'S; . . .

3.13. In the starshaped case, show that any closed path is homotopic to a
constant path, and use Exercise 3.7.
3.15. Use Problem 3.14 to construct a homotopy between the lifted paths.
3.22. Use Problem 3.21 and Problem 3.14.
3.23. Use a homotopy H(P x s) = (l - s)F(P) ± sP, which is a homotopy
from F to the mapping P~P or to P~ -P.
3.25. Part (d) uses Problem 3.22.
Hints and Answers 399

3.27. The degree of OoF is the sum of the degrees of F and O. For the
proof, use Problem 3.26.
3.29. Use the fact that the group GL2(1R) of two by two invertible matrices
has two connected components, cf. Appendix C.
3.30. Deform from the map to its linear approximation. See §D1.
3.31. See §19a for the local structure of general analytic mappings.
3.32. Compute the change in angle of F along the arcs between points in
r1(p').

4.6. If r were a retraction, F(P) = -r(P) would have no fixed point.

4.7. Given a mapping f of Y to itself, consider the composite i of0 r, where
i is the inclusion of Y in X and r is the retract.
4.8. (i). Note that (ii) and (iv) are homeomorphic.
4.11. Compare the restriction of f to Sl with the identity mapping and the
antipodal mapping, using the dog-on-a-Ieash theorem. Or look at fixed points
of x~ ±f(x)/lf(x)l.
4.12. The unit vectors in the octant form a space homeomorphic to a disk,
and, if no such vector is mapped to zero, then F(P)/IIF(P)1i must have a
fixed point.
4.13. Use the preceding exercise.
4.14. Look at the mapping P ~ f(P) - P, and use Exercise 4.11.
4.15. Show thatfis homotopic to the antipodal map.
4.17. See Problem 3.23.
4.18. MapD"" to S""by a formula (aO,al, .. .)~(t,aO,al' .. .).
4.24. See the proof of Lemma 4.20.
4.27. See Lemma 4.21.
4.30. If not, do a spherical projection from a point not in the image.
4.31. Iff(P)~P for all P,fis homotopic to the antipodal map, while if
f(P) ~ p* for all P, f is homotopic to the identity map.
4.38. Choose three arcs covering the circle without antipodal points, and
look at their inverse images in the sphere.
4.39. Look atAUB*, BUC*, and CUA*.
4.40. Tennis anyone?
5.4. If A is unbounded, the same proof shows that Wp is exact. If A is bounded,
the integral of Wp around a large circle is nonzero.
400 Hints and Answers

5.13. Take U = 1R2 \ A, V = 1R2 \ B, and show that the image of & and the
kernel of & each have dimension at least 1.
5.14. Divide the rectangle in half, with the intersection an interval. Argue
as in Theorem 5.1, and use the fact (*) to know about the first cohomology
groups of the complements.
5.16. Write X = A U B, with A and B homeomorphic to circles, and A n B
a point or homeomorphic to an interval.
5.21. Induct on e. Let A be the union of the vertices and e - I ofthe edges,
and let B be the other (closed) edge, and set U = 1R2 \ A and V = 1R2 \ B,
arguing separately the cases when B has one endpoint or two, and, when
two, whether they are in the same component of A or not.
5.22. Analyze the connected components of the complement as the edges
are added.
5.23. Take V= 1R2\X, so UU V= 1R2 and un V= U\X.

5.~.~ Q
5.26. Look at the image of a circle around the band, and the image of its
complement.
5.27. For example, the situation should look locally-via a homeomor-
phism-like the two axes crossing at the origin.
5.28. Both follow from Corollary 5.18.

6.5. If 'Y=~nl'fi' then 'Y=~ni(ari)' where C(t,s)=(1-s)·'Y;(t)+s·po,

where Po is the point with respect to which U is starshaped.
6.12. Use Theorem 6.11. See §9a for generalizations.
6.14. If r: [0,1] x [0, 1] - U, the boundary of Fo r is F *(ar).
6.17. See Proposition 7.5 and Problem 7.8.
6.18. See Problem 3.23.
6.24. (b) A point on a circle is a retract but not a deformation retract.
6.25. Use Exercise 6.20 and Proposition 6.23 to show that r* is the inverse
isomorphism to i*.
6.27. If P' is a point not in X', use Tietze to extend F to a continuous
mapping from an open neighborhood U of X to 1R2 \ {P'}. Apply Theorem
6.11.
Hints and Answers 401

6.28. The converse is false! For a counterexample, see Problem 13.28.

7.2. Consider (!R(x + iy)", ~(x + iyn and (!R(x + iy)", - ~(x + iy)"), where
!R and ~ denote the real and imaginary parts.
7.3. If 'Y is a path around P as usual, H°'Y gives a homotopy in Ilf\{O}
from Vo°'Y to VI°'Y.
7.4. In the first case, consider the homotopy
H(t, s) = s· V('Yr(t)) + (1 - s)· p('Ylt))· V('Yr(t)) ,

oS t s 1, 0 s s s 1, where 'Yr is a path around a small circle around P. For

the second, compare V and - V.
7.8. If an infinite number of points Pj in Z have W('Y,Pj ) =F 0, such points
lie in a bounded set, so they must have a limit point P. Since P cannot be
in U, W('Y,P) = 0, and this contradicts Proposition 6.8.
7.9. Recall thatfhas a local maximum (resp. minimum) at P if the Hessian
is positive and (azJ/a~)(p) < 0 (resp. (a1ja~)(p) > 0); and f has a saddle
point at P if the Hessian is negative. Use Problem 3.30.

/::x l [:: :~~: ~n '

7.13. Show that

(~*V)(x, y) = [Y~~ 2

with (a, b) a nonzero vector, and where p(x,y) and q(x,y) approach zero as
Ixl and Iyl approach infinity. Restrict to a large circle, and do a homotopy
(see Appendix D), to deform [:::~~:~n first to [:], and then to [b].
7.16. See the first picture in this section.
7.17. Look at Yep) = f(P) - U(P)· P)P.

8.4. Thinking of a horizontal doughnut with g holes, put a source on top

between each of the holes, and a sink directly under each source.
8.5. #peaks + #valleys - #passes = 2 - 2g.
8.6. No.
8.lD. Triangulate each of the polygons by putting a new vertex in its center.
8.12. Lift the vector field or triangulation to S2.
8.13. O.
8.14. (b) A Klein bottle.
8.15. -1.
402 Hints and Answers

9.6. There is no homomorphism from 7L n ' onto 7L n if n > n', see Exercise
C.I4.

9.7. For (a), take a small circle "Ij around i for each iE N, and show that
any closed I-chain is homologous to a finite sum 2,n j"lj. The answer is the
same for (b), since the spaces are homeomorphic, for example by the map
z~ I/z from C \ {O} to C \ {O}.

9.9. The fact that the map is one-to-one follows from Theorem 6.11. For
surjectivity, one can produce I-cycles with arbitrary winding numbers around
each K j by Lemma 9.1, so the essential point is to show that the map from
HI(V \ K) to HI V is surjective. Take a grid so that no rectangle meets a point
of K and a point not in U. If "I is a I-cycle on V, we know "I is homologous
to a sum of the form 2,niaRi' Let "I' = 2,R;nK=On i aR i . Then "I' is a I-cycle on
V \ K that is homologous to "I on V. (See also Exercise 10.14.)

9.14. Consider the path . See Exercise 11.14.

9.15. Use Corollary 9.12. If each "Ij is 't:"', so are all the constructions made
in the proof that "I is a boundary.

9.16. Show that 1R2 \ V is connected. Note that if V is any connected com-
ponent of [R2 \ X, then V C V U X and V meets X, so the union of any such
V with X is connected.

9.17. For (a), take a subdivision and rectangles V; as in the definition, but
with the additional properties that each side of each Vj is of length at most
1, and the closure of V; is contained in V. Let OJ be the point in the center
of Vj • Given a point to, and an E > 0, show that there is a 8 > 0 so that

Ip(x, y, t) - p(x, y, to)1 < E/2n and Iq(x, y, t) - q(x, y, to) I < E/2n

for It - tol < 8. For each such t, leth" be the function on Vj so that d(h,,) = w,
on V; and such that h"(O;) = O. Use the construction of Proposition 1.12,
with the fact that the integrals are taken over segments of length at most 112
(see Exercise B.8) to show that Ih,,(p) - h"o(p)1 < E/2n for all P in Vi' and
for all It - tol < 8. Deduce that for It - tol < 8,
n

2: ((hiPj) - h"o(P;)) -
i=l
(h"cP j-,) - h"oCPj-,))

:S 2n'(E/2n) = E.

For (b), note that integral-valued locally constant functions are constant, and
Hints and Answers 403
they vanish if they are small.
9.20. 53'IT.
9.21. Let Pi be any point in Ai' and set W = ~ ~iWPi' where wPi is the I-form
(1/2'IT)wp;.il that measures change in angle around Pi.

9.22. For (b), use the equation displayed after Corollary 9.19, with
(m\> m2) = (1,0) and (0, 1).

9.23. Approximate u(a + Ax + illy) - u(a) by (aulax)(a)Ax + (aulay)(a) lly,

and similarly for v.
9.25. For (b), take 'Y. a circle of radius E around the singUlarity, and let E
approach O.
9.28. See Problem 7.8.
9.31. See Exercise 7.4.
9.35. Use the dog-on-a-Ieash theorem (Theorem 3.11).

10.5. On a path-connected space, O-cycles of degree zero are boundaries.

10.14. Apply Mayer-Vietoris, with V = [R2 \ K. Use Corollary 9.4 to cal-
culate HIV.
10.18. Identify S2 with [R2U{oo}. Suppose X\Xnu={p,Q}, with P and
Q in [R2. Let V = S2 \ X. Use Mayer-Vietoris to see that un V is discon-
nected exactly when H 1U_H 1(S2\{p,Q}) is zero, or equivalently, when
W( 'Y, P) = W( 'Y, Q) for all I-cycles 'Y on U.

10.19. See Problem C.16.

10.20. Use Corollary 9.4 to compute H1U, H1V, and H1(UUV); and com-
pute the kernel of a. For the last part, argue as in the proof of Theorem 5.11,
using the inclusion of Ho(Un V) in Ho(U)(BHo(V) to show that if points Po
and P I are in different connected components of U and V, then they are in
different connected components of un V.
10.21. Use Exercise 10.14 with MV(iii) and MV(iv).
10.22. Apply Alexander's lemma, with the compact sets X and aD UB.
10.23. If not, find points p. and Q. within distance lin that cannot be so
joined, and apply the preceding problem to a limiting point P.
10.24. Let E. = lin, and take corresponding 8. > 0 from the preceding
problem. Take a sequence of points Q. in the same component as Q, with
the distance from Q. to P at most 8./2. Connect Qn to Qn+1 by a path in a
disk of radius En. Join these paths together, with the nth path defined on a
subinterval of length I 12n .
10.25. Use Problem 9.9 with MV(iii) and MV(iv).
404 Hints and Answers

10.28. Take V = [R2 \ K, and use the isomorphism described after Exercise
9.21.

10.30. If not, use grids to find a closed path 'Y in [R2 \ au that has different
winding numbers about two points of au. Note that au c K. Since 'Y is
connected and does not meet au = V \ U, 'Y must be contained in U or in
[R2 \ V. If 'Y is contained in U, then'Y has the same winding number about
all points in K, since K is connected and contained in [R2 \ U; this contradicts
the fact that au c K. If 'Y does not meet V, since V is connected, the winding
number of 'Y is constant around points in V, contracting the fact that
aucV.
If K is closed and connected, the same is true. To see it, let K be the
closure of Kin S2 = [R2 U {oo}, let P be a point in U, and apply the preceding
case to KCS2\{P}=R2.

11.1. For (ii) you can use the identity exp(x + iy) = eX(cos(y) + i sin(y».

11.4. Show that the set in X where the cardinality is n is open and closed.

11.10. Write an open rectangle as an increasing union of closed rectangles.

11.11. For (a), when X is locally connected, the evenly covered neighbor-
hoods N can be taken to be connected, and if p-I(N) is a disjoint union of
open sets Na , each mapping homeomorphically to N, then these Na are the
connected components of p-I(N). (b) follows, since Y' will be a union of
those Na that it meets.

11.13. For the triangulation, use the lifting propositions to lift any trian-
gulation of S to a triangulation of S'.

11.15. For (b), for n E l. and (r, -l1) in the right half plane, the action is
given by n' (r, -l1) = (r, -l1 + 2'ITn).

11.22. The map from X x G to Y by x x g~ g' sex) is a G-isomorphism.

11.23. Use the preceding exercise.

11.24. It's enough to look where the covering is trivial, as in the proof of
Lemma 11.5.

11.28. Given y in Y, take disjoint neighborhoods Ug of g' y, one for each

V be the intersection of the open sets g -I . Ug •
g in G, and let

11.39. Given such an automorphism IP of S, define an automorphism of the

covering by the formula y * 'Y ~ lP(y) * 'Y for any yES and any path 'Y starting
at x. Show that this is independent of choices.
Hints and Answers 405

12.1. Use the same fonnulas as in the preceding displays, but for the second
variable s.
12.3. For (a),

H(t, s) T(4t-s-l),

f.L (4t ; ~ ~ 2), 1/4(2 + s) ~ t ~ 1.

12.5. To compute 'Tr1(SI, (1,0», see Problem 3.14, which is an easy con-
sequence of the propositions in Chapter 11.
12.6. Use the homotopy H(v, s) = sx + (1 - s)v.
12.8. Cut the path into a finite number pieces that map into hemispheres,
say, and replace each piece by a homotopic arc with the same endpoints,
for example an arc along a great circle.
12.10. If a and T are loops at e, consider the homotopy H(t, s) = a(t) . T(S).
12.12. Map (t,s) to
(0,1 - 2t),
{ ( 4t+2S-2
3s + 1 '
s) , 112(1 - s) ~ t ~ 1/4(S + 3),

(1, 4t - 3),
Follow this by h to achieve the homotopy.
12.15. (i), (ii), (iv), and (v) are equivalent; (iii), (vi), and (viii) are equiv-
alent; (vii) and (ix) are equivalent.
12.18. For Sl, let H«xi ,X2) x s) = rotation by s· 'Tr acting on (XI, X2). For
larger n, use the same fonnulas for each successive pair of the n + 1 co-
ordinates on Sn C IRn+ I .
12.20. Show that it has a circle SO(2) as a defonnation retract.

13.10. Take xc 1R2 to be the union of the lines L(n) = {lin} x [0, I], for
all positive integers n, and three lines L= [-1, 1] x {I}, M={-I} x [0, 1],
and N = [-1,0] x {OJ. Let X be the point (0,0). Take two copies XI and X2
of X, and denote by subscripts the corresponding lines and points in XI and
X2 • Take Y to be the disjoint union of XI and X2 , topologized as usual except
near the points XI and X2. For a disk U of radius E < 1 about (0,0) in the
plane, define a neighborhood U(Xi) in Xi by
U(XI) = (N I n U) u U (L(n)1 n U) u U (L(nh n U),
nodd neven

U(X2) = (N2 n U) u U (L(n)2 n U)U U (L(n)1 n U) .

nodd neven
406 Hints and Answers

This space Y is connected, and the natural map from Y to X is a covering

map.

13.12. The proof is exactly the same as for the theorem.

13.17. By Corollary 13.16, the answer to (a) is 7L/n7L. For (b), the fun-
damental group is the group of translations of the plane described in Exercise
11.25. The subgroup in (c) is generated by a and b2 •

13.21. For (1), note that -y. ex is homotopic to (-y.~). (!~-l. ex). For (3), if
-y. ex is homotopic to -y' .~, with ex and ~ paths in N from z to w, then ex. ~-l
is homotopic to Ez in X, so -y is homotopic to -y. (ex· ~-l), so to (-y. ex). ~-l,
so to (-y' . ~). ~-l, so to -y' . (~. ~-l), and so to -y'.

13.22. Take Y to be the union of IR with a copy X. of a clamshell-but

without its outer circle-attached at each point n in 7L C IR, and map Y to
X by wrapping IR around the outer circle of X once between each integer.
Take z~ Y to be a covering that is nontrivial over a different circle of each
X•.

13.27. For ¢;, write -y as a boundary on U, and apply r to both sides.

13.28. Let -y be a path which, on [0,1/2] first goes around C 1 counterclock-

wise, then C2 counterclockwise, then C 1 clockwise, then C2 clockwise. On
[1/2,3/4] it does the same but using C3 and C4, and so on, on intervals of
length 1/2· using the circles C2n - 1 and C 2n • For all k there are homomor-
phisms from the fundamental group to the free group Fk with k generators,
obtained by using the first k circles (see Problem 13.25 or §14d). The image
of [-y] in F 2. is a commutator of n elements, but not of fewer than n elements.
So [-y] cannot be the commutator of any finite number of elements.

13.29. (r,{})~log(r)+H}.

14.5. If f and f' were two such groups, the universal property for each
would give maps from f to f' and from f' to f, and the two composites
f ~ f' ~ f and f' ~ f ~ f' would be the identity maps by the uniqueness
of such homomorphisms.

14.6. The subgroup of '1Tl(X,X) that is generated by these images has the
same universal property. It can be proved directly by subdividing the paths.

14.10. Take G = '1Tl(U, x) and then G = '1Tl(V, x).

14.13. See Problem 13.25 for the uniqueness.

14.15. If X' is X with an edge collapsed, map X' back to X as indicated:

Hints and Answers 407

The map is the identity except on edges adjacent to the vertex the edge is
collapsed to. Send this vertex to the midpoint of the collapsing edge, and
send half of each edge adjacent to this vertex to half of the edge that was
collapsed, and spread the other half over the full edge. Check that the two
composites are homotopic to the identity maps on X and X' .
14.17. For one point in a torus, the complement has a figure 8 as a defor-
mation retract.
14.18. It is a free group on a countably infinite number of generators. Use
Theorem 14. 11.
14.19. The complement of an infinite discrete set in the plane is a covering
space of the complement of a point in a torus.
14.20. The sphere with two handles can be obtained by joining the com-
plements of disks in two tori. For another approach, see Chapter 17.

15.17. See Problem 9.7. The answer is the same for (b), and (c), see Prob-
lem 14.18.
15.19. See Chapter 24 for more general results.

16.5. Show that the set defined this way is open and closed.
16.6. With 00 = dfa on Va' trivialize the covering over Va, by mapping
POl -1(Ua)~ Va X IR by taking a germfat P to P X (f(P).,.. fa(P». Check that
ga~ =fa - fr, are transition functions for this covering. If 00' = 00 + dg, let
fa' = fa + g, and one obtains the same transition functions.
16.11. HOX ~ HO(X; IR) comes from the fact that a locally constant function
is a function on X with coboundary zero; H1X ~ Hl(X; IR) comes from the
fact that a closed I-form defines by integration a function on paths that is
a l-cocycle. It follows from Proposition 16.10 and Theorem 15.11 (and
Exercise 15.18) that these maps are isomorphisms.
16.12. Use the ideas of Lemma 10.2. (Caution: Proposition 16.10 and Mayer-
Vietoris for homology can be used directly for example when G = IR, but
not in general.) See §24a for the general story.
16.13. Since the given covering is locally isomorphic to the trivial G-cov-
408 Hints and Answers

ering, it suffices to prove that PT is a trivial covering when the given cov-
ering is the trivial G-covering. If the given covering is the projection from
the product X x G -+ X, there is a canonical mapping
YT = «X x G) x T)/G -+ XxT
determined by the map «x x g) x t)-x x g-I. t. To see that this is well
defined on orbits, note that, for h in G,
h· «x x g) x t) = «x x h· g) x h· t)
- xX(hg)-lh·t = xXg-1.t.
Check that this is a homeomorphism, with its inverse determined by sending
x x t to «x x e) x t).
16.14. For any y in Y and t in T, there is a unique element <I>(y, t) in T' so
that
f«y x t» = (y x <I>(y, t» for all y in Y and t in T.
For fixed t, the mapping y - <I>(y, t) from Y to T' is locally constant, since
it is constant on each piece of p-I(N), for N an evenly covered set in X.
But since Y is connected, a locally constant function is constant; therefore
<I>(y, t) = Ip(t) for some function Ip: T -+ T'. To see that Ip is a map of G-sets,
calculate:
(y x Ip(g' t» = f«y x g' t» = f«g-I . y x t»
= (g-l·yXIp(t» = (yXg'lp(t»,
from which the equation Ip(g' t) = g . <p(t) follows. By the definition, the
mapping from YT to YT' determined by Ip is f.
16.17. There is a mapping from Y/H to (Y x G /H)/G that takes the H-orbit
of a point y to the G-orbit of the point y x H. The inverse mapping from
(Y x G /H)/G-+ Y/H is given by sending the G-orbit of y x gH to the H-
orbit of g-I . y.
16.22. By Exercise 16.14, an isomorphism f Y(I/II)-+ Y(1/I2) is given by a
map Ip: G' -+ G' of G-sets. So
(i) f«z x g'» = (z X Ip(g'» for all z E Y and g' E G'; and
(ii) Ip(g' '1/I1(g-I» = Ip(g') 'I/Iz(g-I) for all g' E G' and g E G.
Since f preserves base points,
f«y x e» = (y X Ip(e» = (y x e),
so we must have Ip(e) = e. Since f is a mapping of G'-coverings, we must
have
f«z x g'» = f(g' . (z x e» = g' -j«z x e» = g'. (z X e) = (z X g').
Therefore Ip(g') = g' for all g', and applying (ii) with g' = e we see that
I/II(g -I) = 1/12(g-l) for all g in G, so 1/11 = 1/12'
Hints and Answers 409

16.24. Over Va, where p-I(Va) == Va X G, identify (p-I(Va) X T)/G with

the product Va X T by (x X g X t)~x X g-I. t. The transition from Va X T
to V~ X T is
x X t ~ (x X eXt) ~ (x X gaf\(x)· eXt) ~ x X gafl(X)-1 . t = x X gfla(x)· t.

16.25. Consider first the case of a trivial covering Y = X X G. Then

(Y X G')/G = «X X G) X G')/G ~ X X G', x X g X g' ~ x X g' .1\J(g).

Choose trivializations p - \ Va) == VaX G ofthe covering p, so thatthe resulting

transitions are given by the cocycle {gafl}. Identify p,-I(Va) with Va X G' by
the displayed isomorphism. The transition from Va X G' to VfI X G' (over
van VfI) is
x X g' ~ (x X e X g') ~ (x X gaj3(x) . e X g') ~ x X g' ·1\J(gaj3(x» .

16.28. The transition functions are given by the Jacobian determinants of

the change of coordinate mappings.

17.10.

17.12. The fundamental group is the free group with 2g + n generators

ai' b l , • • • , ag , bg , d l , • • • , dn , divided by the least normal subgroup
containing

cg = al· bI· al -I . b I -I •.... ag • bg. ag -I • bg-I • d I · · · · · dn·

The result is a free group on 2g + n - 1 generators, if n 2: 1 (since one can

write dn in terms of the other generators).

18.3. These integers are determined by writing the class of 'Y in terms of
the basis: b] = ~f=l(m;[a;] + n;[b;]).
410 Hints and Answers

18.4. If {I\J~'} is another partition of unity subordinate to another atlas of

charts,

The linearity is immediate from the definitions. For the opposite orientation,
one can use the same charts but with the x and y axes interchanged, so in
local coordinates the form v is expressed as -v"dydx, and all the integrals
get replaced by their negatives.
18.7. Since the I-forms a j and I3j form a basis for H'X, it suffices to prove
these formulas when w is one of the I-forms aj or I3j. Lemmas 18.1 and
18.6 imply these formulas. For example, (aj' a j) and faj aj are both 0, as
are (l3j , I3j) and h °
I3j, and (aj , 13;) and faj I3j are both 1 if i = j, and oth-

°
erwise. Finally, (l3j ,aj) = -(aj,l3j), and -(aj,l3j) and fbja j are both -1 if
i = j and otherwise.
18.8. By linearity, it is enough to do it for f..l = a j and f..l = 13;, and it then
follows from Exercise 18.7 and Lemma 18.1.
18.11. See Exercise 18.7 for (a).
18.12. Changing the aj and hj if necessary, one can assume they cross the
annulus transversally. Calculate for f..l = a j and I3j as above.
18.14. Use Problem 18.12.
18.15. It suffices to show that another choice of differentiable structure gives
the same intersection numbers for pairs taken from basis elements aj and hj •
18.20. Use a partition of unity, as in Lemma 5.5.
°
18.22. If H2U = and H2V = 0, it follows from this and the Mayer-Vietoris
sequence in § 16 that H2( U U V) = 0. Use Exercise 18. 18 and the fact that X
can be built from rectangles, see Lemma 24.10.
18.25. One can realize the surface by removing h disks from a sphere, and
gluing in h Moebius bands. Apply Mayer-Vietoris.
18.26. IfJis a finite covering, and g is a function on Y, defineJ*(g) to be
the function whose value at x is the sum of the values of g at the points of
J-'(x). A similar definition works for forms.

19.1. f'(z) = Ux + ivx = Vy - iuy, and the Jacobian determinant is uxvy - vxuy-
19.4. Apply Riemann's theorem on removable singularities (Exercise 9.25).
19.6. For (d), multiply the function by a suitable P2(Z)/P,(z) so that the
result has no poles, so is constant.
Hints and Answers 411

19.8. This follows from the fact that the map Z ~ z· from D to D is proper,
and the fact that only finitely many points are added.
19.17. (b) If A·(Z+b')CZ+b, A'T'==aT+b, A·I ==CT+d; the de-
terminant is ± 1 exactly when A' (Z + b') == Z + b, and it is positive if T
and T' are both in the upper half plane.

20.5. To prove that C is a Riemann surface near such a point, show that
one or the other projection to an axis is a local isomorphism.
20.6. The genus in each case is: (i) (m - 2)/2 if m is even, (m - 1)/2 if m
is odd; (ii) 0; (iii) 4; (iv) 1; and (v) (m - I)(m - 2)/2.
20.7. (c) With the notation of Exercise 19.12, the covering is given by
assigning each (Ij to the unique transposition (l 2).
20.12. The assumptions in (ii) guarantee that X has n distinct points over
00,all therefore with ramification index 1. To see this, set z' == I/z and w' = w/z;
they satisfy an equation G(z', w') = };7=ob;(z')(W,)"-1 = 0, where
bj(z') = (z'iall/z') is a polynomial in z' with constant term Aj; the n roots
r
to the equation };7=o Aj i = 0 give n points on X over 00.
Consider the zeros and poles of the meromorphic function h. To see what
happens at the points over 00, make the change of coordinates as above. A
calculation shows that h = (z,)I-nGw'(z', w'), so h must have pole of order
n - 1 at each of these n points. At the other points of X, i.e., the points of
C, h has no poles, so by Corollary 19.5 the sum of the orders of zeros of
h at the points of C must be n(n - 1). By Exercise 20.9, this sum is the
sum of e(P) - lover all ramification points of the mapping z: X ~ S2. Apply
Riemann-Hurwitz to prove (c).
The space of polynomials of degree at most n - 3 in Z and W has a basis
the monomials Zi Wi with i + j ~ n - 3, and the number of these is
(n-2)+(n-3)+ ... +2+ 1 = (n-I)(n-2)/2.
If we verify that g' dz/h is a holomorphic I-form for each g = ZiWi , it fol-
lows that we have produced gx holomorphic I-forms, showing at once that
these are all of the holomorphic I-forms, and that their dimension is gx. At
the points in C, the form dz/h is holomorphic, as follows from the equation
o= d(F(z,w» = F.(z,w)dz+Fw(z,w)dw,
so dz/h = -dw/F.(z, w), and one of the denominators is nonzero at each
point of C. At infinity, since dz = -(z,)-2dz',

.. dz
z'w1 - - -
Fw(z, w)
412 Hints and Answers

from which we see that these I-forms are also holomorphic at points over
00.

20.13. The proof is essentially the same, except that Fw(z, w) vanishes at
the 28 points of X lying over the nodes, as well as at the branch points.
Note that the space of polynomials of degree at most n - 3 vanishing at 8
points always has dimension at least (n - l)(n - 2)/2 - 8, so the construc-
tion produces at least gx independent holomorphic I-forms. But since
dim(OI,o) $ g, the inequalities are equalities-which shows, in fact, that
conditions to vanish at the nodes are all independent.
20.16. Compare the sum of the orders of f*oo with those of 00, for 00 a
meromorphic I-form on y,
20.17. See Exercise 9.26.
20.19. For (a), since the residue is linear, it is enough to prove it when
00= (z - a)m dz, a E C, mE 7L. (b) can be reduced to a local calculation, over
a disk in S2, where one has explicit formulas for the map z.
20.24. The point is that, up to periods, integrating from P to Q and then
from Q to R is the same as integrating from P to R,

21.2. Write E = D + QI + ... + Q, and apply Lemma 21.1 r times.

21.13. Define the adele f to be /; at Pi and 0 elsewhere. Take D of large
degree so that M + R(D) = R, and with ordp,(D) 2: mi for all i. There is there-
fore anfin M so thatf-f is in R(D).
21.20, For (a), if points PI"'" P k have been found so that
»
dim(O(P I + ... + Pk = g - k, and k<g, take any nonzero 00 in
O(P I + .. , + P k ), and let PHI be any point which is not a zero of w. For
(b), change the last Pg to be a zero of 00'= 0 in O(P I + ... + Pg-I)' For
points as in (b), Riemann-Roch implies that dim(L(P I + ... + Pg» 2: 2, so
there is a nonconstant function with at most g poles.
21.21. Multiplying <p by a scalar, we may arrange so the residues of <p at
P and Q are as stated. This <p is unique up to adding a holomorphic oo. Take
aj' bj , ooj as in §20d, with the aj and bj not passing through P or Q. Use

Corollary 20.22 to show that there are unique complex numbers Aj so that
the integral of <p - ~Ajooj over the cycles ak and bk are all purely imaginary.
21.22, Take <p as in the preceding exercise, and take u to be an integral of
the real part of <po
21.23, An element in O( -2P) that is not in 0(0) must have a double pole,
since it cannot have a residue at P. Multiply by a scalar to get <p + dz/z 2
holomorphic near P, change <p by a holomorphic I-form to get all its periods
purely imaginary, and integrate the real part of <p to get u.
21.24. Consider the sequence C=L(O)CL(PI)CL(P I +P2 )C . .. ,with
Hints and Answers 413

the dimensions going up by

k=2g-1.
° or 1 at each step, reaching dimension g when

21.25. Apply Riemann-Roch, with E = K - D.

21.27. The holomorphic differentials have the form [(a + bz + cw)/w] dz,
with a, b, c E C. Check that no such form vanishes to order 2 at any point
ofX.

22.1. See the proof of Green's theorem in Appendix B. For the correct
signs, see §23a.
22.2. Extend the definition of integral over continuous paths as in the plane,
and use the same arguments as in the planar case.
22.3. 4'Tr.
22.4. If F: S2~1R3\{O}, define W(F,O) to be (l/4'Tr)IIForW.
22.6. Use Van Kampen, with one open set a ball around the missing point.

22.7. A homotopy is H«z, w) x s) = (sz, Yl- ilzI21:1).

22.9. To show the map is surjective, if (z, w) is in K, let z' = -z/a and
w' = w/b, and verify that (Z,)3 = (W,)2 and Iz'l = Iw'l = 1.
22.11. For example, with p(t) = 1 - t, one may take
A = If2(_p2 + Y4 + 4p3 + p4) and IL = VA=P.
22.13. The generators for the fundamental groups of A \ K and B \ K are the
circles around the middles of the solid tori. The generator for the funda-
mental group of T \ K is a path as indicated:

To appeal to Van Kampen, A and B must be replaced by open neighborhoods

of which they are retracts.
414 Hints and Answers

22.14. For (a), construct a mapping from [0, It to Sk that maps the bound-
ary to So, and is a homeomorphism from the interior of the cube to the
complement of so. For (d), show that a map from Sk to sn, with k < n, is
homotopic to one that misses the south pole, say by triangulating Sk into
small simplices, and approximating the map by a "simplicial" map. See,
e.g., Hilton (1961).
22.16. Find homotopies that stretch and slide between the maps indicated
in the diagram:

r x x

r A r x A x x x

x x A

x x r

x x x A x r A r

A x x

For example, the second homotopy maps (t l , ••• , t k , s) to

f<3tl' (3t2 - 2s)/(3 - 2s), t3 , • • • ,tk), O:s tl :s 1/3 , 2s/3:S t2 :S 1,

{ A(3t l - 1, 3t2 /(3 - 2s), t3 , • • • ,tk ), 213:S tl :s 1,0:s tl :s 1 - 2s/3,
x, otherwise.

The third slides the squares around clockwise.

22.22. Details can be found in many texts, e.g., Bott and Tu (1980).
22.23. The form is closed by calculation, and it is not exact by Stokes theo-
rem, since its integral over Sn-I is not zero.
22.27. See Bott and Tu (1980).

23.1. This is formal, using the identities

for i<j,
and s and s' each taking values 0 and 1. The signs cancel because of the
shift in subscript from j to j - 1.
23.5. Check that (a? - a!) oS = So (a? - a!).
23.15. Induct on p, and use Mayer-Vietoris for U = VI U ... U Up-I and
Hints and Answers 415

V= Vp- Note that vn V= (VI n V) u ... U (Vp _ 1 n V), so the inductive

hypotheses imply to V, V, and V n V.
23.17. For (d), argue by induction, using Mayer-Vietoris. If g: S"-I_S"-I
has degree d, show thatf: S"_Sn given by the formula
f(XI, ... ,Xn+l) = (g(XI' ... ,xn),X"+I)
also has degree d.
23.19. For (c), use Problem C.3 and the n-dimensional version of Lemma
B.9; see the proof of Claim D.3.
23.22. Induct on n, comparing the action of the antipodal map with the
Mayer-Vietoris isomorphism.
23.23. For (a), such a vector field gives a mapf: s"_sn that has no point
P mapped to P or to - P. Such a map is homotopic to the identity and to
the antipodal map. Use the preceding problem. For (b), consider the map-
ping (XI,X2," . ,Xn+I)-(X2,-XI,'" ,xn+l>-xn),
23.26. For (a), regard SmCS", and note that a mapf: Sn_Sn that is not
surjective has degree zero. The other proofs are essentially the same as in
Chapter 4.
23.27. In the situation of (a), there is a homotopy from f to a map g to
which Theorem 23.24 applies, given by homotopic to the map g given by
f(P) - sf(P*)
H(P x s) = Ilf(P) - sf(p*)II'

For (b) show similarly thatfis homotopic to a map g with g(P*) = g(P) for
all P. (c) Any automorphism g without fixed points has degree - 1, so if g
and h are nontrivial automorphisms, since g . h has degree 1, g . h must be
the identity.
23.28. Lift to a map from sn to sn and apply Problem 23.27.
23.31. If m=n, HMmxs") is Z for k=O and k=n+m, and ZEeZ if
k = n, and 0 for other k. If m ¥- n, the answer is Z for k = 0, m, n, and
m + n, and 0 for other k. For the proof, induct on n, using sm x V and
Sm x V, with V and V as before.
23.36. Identify ~n with the complement of a point Pin sn, and use Mayer-
Vietoris for this open set and a small neighborhood of P.
23.37. See Proposition 5.17 and Corollary 5.18.

23.39. Cut the band in half: . Look at the

exact sequence
416 Hints and Answers

and the maps to the terms in the corresponding sequence with A and B re-
placed by Any and B n Y, and X replaced by Y.
23.41. Use n-dimensional grids, and generalize the arguments of Chapters
6 and 9.

2..1.5. The fact that C" is free abelian means that one can find a subgroup
C" of C that maps isomorphically onto C", and then C is the direct sum of
E" and the image of C'; and Hom( - ,G) preserves direct sums.
24.7. This amounts to an identity on [k, which is proved just as in the case
of Green's theorem for a rectangle by a calculation using Fubini's theorem,
as in Proposition B.6.
24.9. This is the higher-dimensional version ofreparametrization for paths.
Here, for any k-cube r, SoA(r) - r = aA, where
A(s, t l , • •• ,tk) = r(s '1X(2t l ) + (1 - s)· t l , • •• ,s '1X(2tk) + (1 - s)· tk),
noting that the other terms in the boundary are degenerate. Defining Sp and
Rp by the same formulas as in §23b, one calculates that
SP-l = aoRp+Rpoa+(SoA-J)oSp,
and the rest of the proof is the same as before.
24.14. This is clear except where the coboundary and dual of the boundary
are involved. Let W = WI - W2 be a closed (k - I)-form representing a class
[w] in H"-I(U n V), with WI and W2 (k - I)-forms on U and V, and let z = CI + C2
represent a class [z] in lI';(U U V), where CI and C2 are chains on U and V
respectively. Then

/)([w))([z)) = L + L = L. + L
dWI dw2 WI W2 = [w](a([z)).

24.15. See Problem 23.41.

24.16. Use Mayer-Vietoris and induct on the number of open sets in the
cover.
24.21. Set (p*w)(x) = W(YI) + W(Y2) where p -I(X) = {YI ,Y2}' For (c), note
that Ix p*w = 0 for any n-form W with compact support on X.
24.23. For (a), use HkX:::= (Hn-kX)* :::= «HkX)*)*, and the general fact that
if V is a vector space isomorphic to (V*)* , then V must be finite dimensional.
For (b), use the linear algebra fact that a finite-dimensional vector space
with a nondegenerate skew-symmetric form must be even dimensional.
24.24. Since the spaces are finite dimensional, H~-PX is isomorphic to (HPX) *.
24.25. This follows from the fact that all cubes r have compact image in
X. (Compare the special case used in the argument in §5c.)
Hints and Answers 417

24.26. (b) follows from the definitions, just as for homology.

24.27. Use Exercise 24.S.
24.36. Consider for each n the exact sequences O~Zn~Cn~Bn-1 ~ 0
and O~Bn~Zn~Hn~O, and use Problem C.16.
24.37. For (c), fix an ordering on N(au). For each vertex v, choose a point
Cv in U v , regarded as a O-chain on U v • If (vo, VI) is a I-simplex, then, since
Uvo U UV\ is connected, there is a path from Vo to VI in Uvo U UV\' which de-
termines a I-chain C(VO.VI). Construct, by induction on n, for each n-simplex
(vo, ... ,vn), a chain c(vo ..... v.) in Cn(UvoU ... U Uv), such that
n

a(c(vo .. . .. v) = 2: (-l)ic(vo .. . ..
;=0
Vi •. . .• v.)·

The existence of c(vo .. . .. v.) follows from the fact that the right side is an
(n - I)-cycle on Uvo U ... U Uv., and Hn-l(Uvo U ... U UvJ = 0 by Ex-
ercise 23.1S(b). This gives a map C*(N(au»~ C~. To see that the resulting
map on homology is an isomorphism, induct on the number of open sets;
take any V and construct subcomplexes LI and L2 as in the proof of Prop-
osition 24.33, and compare the corresponding Mayer-Vietoris sequence for
LI and L2 with that of the covering of X by Uvand Uv'>'v Uv'.

A.6. If not, each point has a neighborhood meeting only finitely many, and
K would be contained in a finite union of such neighborhoods.
A.8. Without loss of generality, one may assume K contains a neighbor-
hood of the origin. Map aK to sn-I by mapping P to P Illpll. This is contin-
uous and bijective, so a homeomorphism. Let f: Sn-I ~ aK be the inverse
map. Define F: D n ~ K by F(O) = 0, and F(P) = IIPII· j{P fllPll) for P "# O. Then
F is continuous and bijective, so a homeomorphism.
A.16. A connected and locally path-connected space is path-connected, as
seen by showing that the set of points that can be connected to a given point
by a path is open and closed.

B.12. Find a countable covering of U and so that each open set in the cov-
ering is contained in some Ua' and so that any point has a neighborhood
that only meets finitely many of the open sets.
B.14. With g as in Step (2), let hex) = f~g(t) dtl f6g(t) dt, and set
= her - rl)/r2 - rl), where r = II(x, y)ll.
$(x, y)

e e
C. 7. If {e a } is a basis for F, and \p( a ) = ea , these a generate a free abelian
subgroup F of A, and A is the direct sum of Ker(\p) and F by Exercise C.S.
C.IS. Let (l be the map from A to B, 13 the map from B to C, and let C'
be the image of 13, 13': B ~ C' the induced surjection . Iff in Hom(B, IR) maps
418 Hints and Answers

=
to 0 in Hom(A, IR), f vanishes on the image of a. Since B /Image(a) C' ,
there is a homomorphism g' from C' to IR such that g' 0 W= f By the lemma,
there is a homomorphism g from C to IR that restricts to g' on C'. Then
go f3 = f, which means that g in Hom( C, IR) maps to f.
C.16. Let B =A2/Al' and show that there are exact sequences
O-AI-A2-B-O and O-B-A3- ... -An-O.
References

Included in this list are books and articles that were referred to in the text.
They can be consulted for more on topics covered in this book.

L.V. Ahlfors, Complex Analysis, 3rd edn, McGraw-Hill, 1979.

M.A. Armstrong, Basic Topology, Springer-Verlag, 1983.
R. Bott and L.W. Tu, Differential Forms in Algebraic Topology, Springer-
Verlag, 1980.
G.E. Bredon, Topology and Geometry, Springer-Verlag, 1993.
S.S. Cairns, Introductory Topology, The Ronald Press, 1961.
C. Chevalley, Introduction to the Theory of Algebraic Functions of One
Variable, American Mathematical Society, 1951.
W.G. Chinn and N.E. Steenrod, First Concepts of Topology, Random House,
1966.
H.S.M. Coxeter, Introduction to Geometry, 2nd edn, Wiley, 1961, 1980;
Wiley Classics, 1989.
R.H. Crowell and R.H. Fox, Introduction to Knot Theory, Springer-Verlag,
1963.
1. Dieudonne, A History of Algebraic and Differential Topology, 1990-1960,
Birkhauser, 1989.
C. Godbillon, Elements de Topologie Algebrique, Hermann, 1971.
M.J. Greenberg and J.R. Harper, Algebraic Topology, A First Course, Ben-
jamin/Cummings, 1981.
P.A. Griffiths, Introduction to Algebraic Curves, American Mathematical
Society, 1989.
V. Guillemin and A. Pollack, Differential Topology, Prentice-Hall, 1974.

419
420 References

D. Hilbert and S. Cohn-Vossen, Geometry and the Imagination, Chelsea,

1952.
P.I. Hilton, An Introduction to Homotopy Theory, Cambridge University
Press, 1961.
J.G. Hocking and G.S. Young, Topology, Addison-Wesley, 1961, Dover,
1988.
H. Hopf, Differential Geometry in the Large, Springer Lecture Notes 1000,
1983.
F. Klein, On Riemann's Theory of Algebraic Functions and their Integrals,
MacMillan and Bowes, 1893; Dover, 1963.
S. Lang, Introduction to Algebraic and Abelian Functions, 2nd edn, Springer-
Verlag, 1982.
W.S. Massey, A Basic Course in Algebraic Topology, Springer-Verlag, 1991.
J. Milnor, Morse Theory, Annals of Mathematics Studies, Princeton Uni-
versity Press, 1963.
J. Milnor, Topology from the Differentiable Viewpoint, University Press of
Virginia, 1965.
R. Narasimhan, Compact Riemann Surfaces, Birkhauser, 1992.
M.H.A. Newman, Elements of the Topology of Plane Sets of Points, Cam-
bridge University Press, 1939.
V.V. Nikulin and I.R. Shafarevich, Geometries and Groups, Springer-Verlag
Universitext, 1987.
H. Rademacher and O. Toeplitz, The Enjoyment of Mathematics, Princeton
University Press, 1957.
E.G. Rees, Notes on Geometry, Springer-Verlag, 1983.
G. Springer, Introduction to Riemann Surfaces, Addison-Wesley, 1957.
A. Wallace, Differential Topology, First Steps, W.A. Benjamin, 1968.
Index of Symbols

For some general notation, see page degp(F), local degree, 46

365. D, disk, 50
C = aD, boundary circle of disk, 50,
CQ,OO,smooth, 3 80
I-form, differential, 4 P*, antipode of P, 53
J~w, integral of I-form along smooth HOU, Oth De Rham cohomology
path, 5 group, 63
along segmented path, 8 HI U, 1st De Rham cohomology
along a continuous path or I-chain, group, 63
127, 132 [w], cohomology class of form w, 64
d, df, differential of function, form, Wp = (]I27T)wp,{), 64
6,247,318,326,329,391 0, coboundarymap, 65-67, 224, 326
w{}, I-form for angle, 6 I-chain, 78-79
aR, boundary of rectangle, 11, 80 O-chain, 80-81
",p,p path around circle, 15 ZoU, group of O-chains on U, 81, 91
Wp,{}, I-form for angle around P, 16, BoU, group of O-boundaries on U, 81,
22 91
{}(t), angle function along path, 18 HoU = ZoU /BoU, Oth homology group
W(""O), winding number around 0, on U, 81, 91
19 C]U, group of I-chains on U, 82, 91
:y, lifting of path, 21, 156 ZIU, group of I-chains on U, 82, 91
W("" P), winding number of", around B 1U, group of I-boundaries on U, 83,
P, 23, 36, 84 91
",-t, inverse of path, 23, 165 H]U=ZIU/B1U, 1st homology group
grad (f), gradient, 28 on U, 83, 91
Supp(",), support of path or chain, F *, map on chains or homology in-
42 duced by F, 89, 92
W(F,P), winding number, 44, 328 IndexpV, index of vector field at
deg(F), degree of mapping of cir- point, 97, 104, 107
cles,45 vic, restriction of V to C, 97

421
422 Index of Symbols

TpS, tangent space, 102 HI(fJU;G), first Cech cohomology set,

g, genus of surface, 108, 112 209
IRp2, projective plane, 115 E[(fJU,x;G), with base point, 210
W('Y,A), winding number around set M~M, orientation covering, 219
A, 123 p",: X"'~X, covering from I-form,
A"" infinite part of complement, 125 221
(n + I)-connected plane domain, 125 HO(X;G), HI(X;G), cohomology,
lJj(w) = lJ(w,A j ) , period, 130 222-225
Resif), residue off at a, 133 YT~X, covering from G-set and G-
ordif), order off at a, 134 covering, 225
a, boundary map, 137-140, 334 Y(IjI)~X, covering from 1jI: G~G'
S, subdivision operator, 139, 334- and G-covering, 227
335 aj and bj, basic loops on a surface,
MV(i) to MV(vi), Mayer-Vietoris 244
properties, 140-142, 148, 258 (0", T), intersection number, 245-246,
+, -, maps on homology and co- 255-256, 357-358
_ homology, 144, 148-149 (J.j and I3j, basic I-forms on a surface,
HoX, reduced homology group, 145 248-251
flOx, reduced cohomology group, ffxv, integral of 2-form on surface,
149 251
n-sheeted covering, 155 1'1, wedge of forms, 252, 325, 355,
y * 'Y, endpoint of lift of 'Y starting at 392
y, 156-157 (w,I1), intersection number for 1-
IRpn, real projective space, 159 forms, 252, 289
Aut(YIX), group of deck transfor- H 2X, second De Rham group, 257
mations, 163 e(P) = e/P) , ramification index, 265
0"' T, product of paths, 165 ordp(f) , 267
En constant path at x, 165 CIA, Riemann surface of genus 1,
'!TI(X,X), fundamental group of X at 264, 275-276, 291-293
x, 168 gx, genus of Riemann surface X, 273
['Y], class of loop 'Y in '!TI(X,X), 168 Fw(Z, W), partial derivative, 277-278
e = [Ex], identity in '!TI(X, x), 168 M = M(X) , field of meromorphic
T#, map induced by path T, 169 functions, 281
'!TI(X, X)abeh 173 C(z, w), field of rational functions in
y * [0"], endpoint of lift of 0" starting z and w, 282
at y, 180 n = nl,o = nl,o(X), space of holo-
[0"] . z, left action of fundamental morphic I-forms, 284
group on covering, 182-184 nO,1 = nO.I(X), antiholomorphic 1-
PH: YH~X' covering from forms, 285
He '!TI(X, x), 189 ordp(w), order of meromorphic 1-
.Kabel, universal abelian covering, 192 form, 287
Pp: Yp ~ X, covering from Resp(w), residue of meromorphic 1-
p: '!TI(X,X)~G, 193-194 form, 288, 299
(z x g), element of Yp from z E Y, WI, • • • , w g , basis of holomorphic
gEG,194 I-forms, 289
Index of Symbols 423

Z = (Tj.k), period matrix, 290 A, A(r), operator on chains, 335

A, Abel-Jacobi mapping, 291 Rp, operator on chains, 335
Div(!), divisor off, 291, 295 Hk(X)CiJl, homology with small cubes,
ordp(D), order of divisor at point, 295 338
deg(D), degree of divisor, 295 a, boundary homomorphism, 348
E 2: D, E - D is effective, 295 H;X, homology using <(;x cubes, 351
Div(w), divisor of meromorphic 1- qjJx, duality map, 356
fonn,295 Hk(X; Z), Hk(X; G), cohomology
L(D), functions with poles allowed groups, 358-359
at D, 296 IKI, realization of simplicial com-
a(D), meromorphic I-fonns with plex, 360
zeros at D, 296 C*K, chain complex of simplicial
E, divisor of poles, 297 complex, 360, 363
f = (jp), adele, 299 HnK, homology group of simplicial
R, space of adeles, 299 complex, 360
R(D), adeles with poles allowed at fa, cubical chain of simplex, 361
Int(A), interior of A, 369
D,300
SeD) = R/(R(D) + M), 300
A, closure of A, 369
Ker, kernel, 378, 380
a'(D), dual space to SeD), 300
1m, image, 378, 381
ai, union of all a'(D), 302
V /W, quotient space, 378
HkU, De Rham cohomology group,
VEBW, EBV", direct sum, 379, 381
319,325 IIV", direct product, 379, 381
'lTiX, x), higher homotopy group, 324 v' w, dot product, 379
H;X, De Rham cohomology with Ilvll, length of v, 379
compact supports, 328 GLnlR, invertible matrices, 379-380
af, boundary slices, 332 G /H, quotient group, 380
CkX, k-chains on X, 332 Hom(G, G'), set of homomor-
a, af, boundary of cube or chain, 333 phisms, 381
HkX = ZkX/BkX, homology group, F(X), free abelian group on X, 382
333 H k (X), homology with Z/2Z coef-
R, R(f), operator on chains, 333 ficients, 393
Index

abelian covering, 192 bilinear pairing, 253

abelianized fundamental group, 173 Borsuk-Ulam theorem, 54, 185,341,
Abel-Jacobi mapping, 291-292 393-396
Abel-Jacobi theorem, 291, 306-311 boundary, 77, 368
Abel's theorem, 291, 306 circle, 50, 80
abstract simplicial complex, 359 map, 137-140, 337-338
action of group on space, 158-159 I-chain, 82
adele, 299 branched coverings, 268-272, 279,
adjacent faces, 233 321
affinely independent vectors, 360 branch locus, 321
Alexander's Lemma, 146 Brill-Noether reciprocity, 306
analytic function, 132-133, 263, 264 Brouwer theorems, 50, 150, 340
angle function, 6-7
along path, 17-19,36-37
antiholomorphic I-forms, 284 Cartesian product, 170, 368
antipode, antipodal map, 53-55, 172, Cauchy integral theorem, formula,
340-341 132-133
Antoine's necklace, 323 Cauchy-Riemann equations, 131
approximation lemma, 87 Cech
Argument Principle, 134-135 with base point, 210
atlas of charts, 389 coboundary, 212
automorphism group of covering, cocycle, 208
163, 182-186, 189 cohomology class, group, 209,
212, 228-229
chain, 78-82, 332
barycentric subdivision, 113, 273, chain complex
363 chain homotopy of maps of, 350
basis exact sequence of, 347
of free abelian group, 382 homology of, 347
of vector space, 378 map of, 350

425
426 Index

chain rule, 373-374 complex affine plane, 277

change in angle, 19, 36-37 of a polynomial, 277
change of coordinates, 389 Riemann surface of, 277-281
chart, 389 cycle, 82
circulation of fluid, 29
clamshell, 187, 192
closed form, 12,247,319 deck transformation, 163
closed I-chain, 80 deformation retract, 93
closure, 369 degenerate cube, 332
coboundmy map, 65-68, 223, 326 degree
coboundmy of cochain, 223 of divisor, 295
cochain, 222 homomorphism on O-cycles, 82
cochain complex, 348 of mapping of circles, 45, 53-55
code for triangulation of surface, 235- of mapping of oriented manifolds,
236, 241-242 342-343
cohomology, 222-225, 358-359; see of mapping of spheres, 324, 339
also De Rham De Rham cohomology groups, 63-
of cochain complex, 348 65, 256-260, 325-328
with coefficients, 349 with compact supports, 328-331
coloring, 76, 115 and homology, 213-216
commutator subgroup, 173 differentiable function on a surface,
commute (diagram), 92 390
compact, 367-368 differential form, 4, 247, 251, 317,
complex conjugation operator, 285 391
complex integration, 131-136 differential of function, form, 6, 247,
complex I-form, 131 318, 326, 329, 391
cone, 361 differential topology, 113
conjugate homomorphisms, 196 dimension of vector space, 378
connected component, 369 direct limit, 200-201, 358
connected space, 369 direct product, 379, 381
continuous map, 367 direct sum, 379, 381
contractible space, 93 discrete set, 100, 134, 203, 271
convex, 12, 368 discrete topology, 153,368
covering discriminant, 278
map, 153 disjoint union, 368
n-sheeted, 155 divergence, 29, 320
from I-form, 220-222 divisor, 295
of space, 153 of function, 291, 295
transformation, 163 of meromorphic I-form, 295
critical point, nondegenerate, 101 Dog-on-a-Leash theorem, 41
crosscap, 117, 203 duality, 148, 213, 223, 350-359
cube, k-cube, 332
«600, 351
curl, 29, 320 edge, 233
curve effective divisor, 295
Index 427

eigenvalues, 52 genus, 244, 273, 281, 285-286

elliptic curve, 291-293 geometry, locally Euclidean, 162-
endpoint of path, 5 163
engulfmg number, 31, 319, 328, 339 germ, 230-231, 283-284
equality of mixed partial derivatives, gradient, 28, 320
374 graph (finite), xi, 75-76, 145,202
Euler characteristic, xii, 113-119, Green's formulas, 30
146, 357, 363 Green's theorem for a rectangle, 11,
Euler's equation, theorem, 75, 113, 374
244, 260, 363 grid, 85
even action, 159, 161 groups acting on spheres, 341
evenly covered, 153 G-sets, maps and isomorphisms of,
exact form, 12, 247, 319 225
exactness of forms, 10-16
exact sequence, 144, 348, 384-385
exponential mapping, 154, 192,265
ham sandwich theorem, 57, 341
harmonic function, 30-31, 305
Hausdorff space, 367
face, 233, 359 Hessian, 101, 286
final point of path, 5 holes, 125
Five-Lemma, 346-347 holomorphic function, 264
fixed point, 49-53, 340 holomorphic I-form, 284-285, 294
fixed point property, 51 homeomorphism, 367
fluid flow, 28-31, 305, 320 homologous I-chains, 82, 88
flux of fluid, 29 homology class, 84
free abelian group, 382 homology group, 81, 83, 332-334
on a set, 382 of chain complex, 347
free group, 191,201-203 with coefficients, 393
free product, 200 homomorphism from G-covering,
Fubini's theorem, 374 194-195, 211
functor, functorial, 92, 168, 326, 359 homotopic
fundamental group, 168-172 maps, 92, 170, 324, 333
and coverings, 179-182, 189-192 paths, 39
and homology, 173-175 homotopy
fundamental theorem of algebra, 48- of closed paths, 27, 39, 165
49 equivalence, 172, 202
fundamental theorem of calculus, 373 groups, 324-325
of paths with fixed endpoints, 25,
39
Galois theory, 190, 283 type, 171-172
Gauss's Lemma, 385 Hopf mapping, 325
G-covering, 158-164 Hurewicz theorem, 212
isomorphism of, 160 hyperbolic plane, 245
trivial, 160 hyperelliptic curve, Riemann sur-
generators of group, 380 face, 293-294, 306
428 Index

incompressible flow, 29 lattice, 264, 290-291

independent of path, integral, 5, 8 Lebesgue Lemma, 371-372
index of vector field at point, 97, 102, left coset, 380
104, 107, 112,287,341,387, Lens space, 162, 186
390 lifting
initial point of path, 5 of homotopy, 157
integral of mapping, 155, 180-181
of k-form over a k-cube, 351 of path, 21, 156
of I-form along a smooth path, 5, linearly independent, 382, 384
317 link, linking number, 325, 342
along a continuous path or chain, local degree, 46, 340
127, 213, 247 locally connected space, 157-158,
of 2-form, 251-252, 318 369
integrating along a fiber, 330 locally constant, 3
interior, 77, 369 locally path-connected space, 180,
intersection pairing 370
on manifold, 357-359 locally simply connected, 187
on surface, 245-246, 252-256 logarithmic pole, 305
invariance of dimension, 55, 341, 344 log function, 265
invariance of domain, 73, 344 loop at a point, 168
inverse limit, 358 Lustemik-Schnirelman-Borsuk
inverse of path, 23, 165 theorem, 57, 341
irrotational flow, 29
isomorphism of coverings, 153
manifold, 315
matrix, 379
Jacobian matrix or determinant, 47, Mayer-Vietoris
54,263,311,387 for cohomology, 147-150, 217-
Jacobian of Reimann surface, 291 218, 258-259, 326, 348-349
Jacobi Inversion Theorem, 291, 310 for cohomology with compact
Jordan curve theorem, 59-60, 68-77, supports, 329-330, 349
146 for homology, 140-147, 334-339,
generalized, 343-345 349,352
for simplicial complexes, 362
mean value theorem, 373
meromorphic function, 134, 267,
Klein bottle, 116, 161, 186,219,401 281-284
knot, 320 meromorphic I-form, 287
equivalence of, 320 module of periodicity, 130
fundamental group of comple- moduli of Riemann surfaces, 264,
ment, 320-323 275-276, 289
similarity of, 320 Moebius band, 77, 116, 161, 229,
trivial, 321 345
Konigsberg bridges, xi-xii Morse theory, 119
Kuratowski theorem, 76 multiplicity of root, 47, 136
Index 429

multiply connected, 125 Poincan!-Bohl theorem, 40

Poincare duality, 357
Poincare-Hopf theorem, 108, 117-
neighborhood, 367 118
nerve of open covering, 363 Poincare Lemma, 326-327, 330-331
node, 286 polar coordinate mapping, 20-23,
nondegenerate cube, 332 154, 192, 221
nonorientable surface, 115, 241, 244, pole, of order k, 134, 267
345 potential function,S, 28
can't be embedded in 3-space, 345 product of paths, 165
nonsingular point of complex curve, projection from space to quotient
280 space, 159
normal form of a surface, 239, 241- projection of a vector on another, 379
242 projective plane, liS, 219
normal subgroup, 183, 380 properly discontinuous, 159
normalizer, 184 proper mapping, 269

quaternions, 162
open mapping, 266 quotient space, 378
open rectangle in [Rn, 352 quotient topology, 370
orbit, 159
order of meromorphic function at ramification index, 265-267
point, 134, 267 ramification point, 265-267
orientable, 390 rank-nullity theorem, 378
orientation, 390 rank of abelian group, 382, 384
covering, 219, 357 rectangular I-chain, 85
of manifold, 219 reduced homology group, 145
preserving or reversing, 340 regular covering, 184, 190
of Riemann surface, 263 relative homology groups, 350
oriented surface, 115, 390 reparametrization of a path, 23, 40,
orthogonal matrix, 172 82
residue, 133-134,288,299
Residue Formula, 288
partition of unity, 66, 375-377 Residue Theorem, 134
patching spaces, 370 restriction maps in cohomology, 68,
patching coverings, 196-197, 207- 147,258
210 retract, retraction, 50, 340
path, 4, 36, 164 Riemann
path-connected space, 370 mapping theorem, 75
path integrals, 3-16 theorem on removable singulari-
peaks, valleys, passes, 101, 105, 113 ties, 133
perfect pairing, 254 Riemann-Hutwitz formula, 272-276,
period, 130-131 288
period matrix, 290 Riemann-Roch theorem, 303-306
period module, 130, 222, 250-251 Riemann's bilinear relations, 289
430 Index

Riemann's formula, 274 fundamental group of, 242-245,

Riemann surface, 263 272
of algebraic curve, 277-281 homology of, 243
Rouche's theorem, 136 smooth,390

tangent bundle, 229

saddle point, 10 I tangent space, 102, 118
Schoenflies theorem, 75 topological group, 162
section of covering, 161 topological space, 367
segmented path, 7 torus, 106, 186
Seifert-van Kampen theorem, 198 torus knot, 322-323
semilocally simply connected, 187 transfer, 260, 394
separate, 146, 345 transitive action, 163, 226, 272
simplex, simplices, 359-360 triangulation, 113, 233-242, 273
simplicial complex, 359 trivial covering, 153
chain complex of, 360
ordered, 360
simply connected space, 182 uniformization theorem, 268
singularity unique factorization domain, 385
of curve, 277, 280, 286, 321 universal abelian covering, 192
of vector field, 97 universal covering, 186-192, 193-
skew-symmetric pairing, 253 196, 226
small cubes, 338 of surface, 245
smooth function, 3 universal property, 198
smooth homotopy, 25, 27
smooth path, 4
Van Kampen theorem, 197-201
spheres, 339
vector bundle, 229
sphere with g handles, 108, 112, 118,
vector field, 27, 97, ll2, 287, 319-
203, 241, 259
320
spherical coordinate mapping, 319 on spheres, 102-105, 341
spin group, 162 on surface, 390
standard basis, 378
vertex, 233, 361
star of vertex, 363
starshaped, 12, 83
Stokes' theorem, 318, 351 wedge product of forms, 252, 325,
Stone-Tukey theorem, 57, 341 355, 392
straight path, 80 Weierstrass gaps, 306
subcomplex of complex, 361 Weierstrass $J-function, 293
subspace, 367 winding number, 19,23,36, 73, 84,
surface, 106-ll9, 231, 233-260, 157,328
389-391 around a set, 123
with boundary, 76
classification of, 236-242
cohomology on, 247-260 zero of order k, 267
Graduate Texts in Mathematics
(continued from page il)

62 KARGAPOLOV/MERLZJAKOV. Fundamentals 93 DUBROVIN/FoMENKOINOVIl<OV. Modern

of the Theory of Groups. Geometry-Methods and Applications.
63 BOLLOBAS. Graph Theory. Part !. 2nd ed.
64 EDWARDS. Fourier Series. Vol. I 2nd ed. 94 WARNER. Foundations of Differentiable
65 WELLS. Differential Analysis on Complex Manifolds and Lie Groups.
Manifolds. 2nd ed. 95 SHIRYAEV. Probability. 2nd ed.
66 WATERHOUSE. Introduction to Affine 96 CONWAY. A Course in Functional
Group Schemes. Analysis. 2nd ed.
67 SERRE. Local Fields. 97 KOBLITZ. Introduction to Elliptic Curves
68 WEIDMANN. Linear Operators in Hilbert and Modular Forms. 2nd ed.
Spaces. 98 BROCKERIToM DIECK. Representations of
69 LANG. Cyclotomic Fields II. Compact Lie Groups.
70 MASSEY. Singular Homology Theory. 99 GRovE/BENSON. Finite Reflection Groups.
71 FARKAS/KRA. Riemann Surfaces. 2nd ed. 2nd ed.
72 STILLWELL. Classical Topology and 100 BERG/CHRISTENSEN/REssEL. Harmonic
Combinatorial Group Theory. 2nd ed. Analysis on Semigroups: Theory of
73 HUNGERFORD. Algebra. Positive Definite and Related Functions.
74 DAVENPORT. Multiplicative Number 101 EDWARDS. Galois Theory.
Theory. 2nd ed. 102 VARADARAJAN. Lie Groups, Lie Algebras
75 HOCHSCHILD. Basic Theory of Algebraic and Their Representations.
Groups and Lie Algebras. 103 LANG. Complex Analysis. 3rd ed.
76 IITAKA. Algebraic Geometry. 104 DUBROVIN/FoMENKOINOVIKOV. Modern
77 HECKE. Lectures on the Theory of Geometry-Methods and Applications.
Algebraic Numbers. Part II.
78 BURRIS/SANKAPPANAVAR. A Course in 105 LANG. SL,(R).
Universal Algebra. 106 SILVERMAN. The Arithmetic of Elliptic
79 WALTERS. An Introduction to Ergodic Curves.
Theory. 107 OLVER. Applications of Lie Groups to
80 ROBINSON. A Course in the Theory of Differential Equations. 2nd ed.
Groups. 2nd ed. 108 RANGE. Holomorphic Functions and
81 FORSTER. Lectures on Riemann Surfaces. Integral Representations in Several
82 BOTT/Tu. Differential Forms in Algebraic Complex Variables.
Topology. 109 LEHTO. Univalent Functions and
83 WASHINGTON. Introduction to Cyclotomic TeichmUller Spaces.
Fields. 2nd ed. 110 LANG. Algebraic Number Theory.
84 IRELAND/RoSEN. A Classical Introduction III HUSEMOLLER. Elliptic Curves.
to Modern Number Theory. 2nd ed. 112 LANG. Elliptic Functions.
85 EDWARDS. Fourier Series. Vol. II. 2nd ed. 113 KARATZAS/SHREVE. Brownian Motion and
86 VAN LINT. Introduction to Coding Theory. Stochastic Calculus. 2nd ed.
2nd ed. 114 KOBLITZ. A Course in Number Theory and
87 BROWN. Cohomology of Groups. Cryptography. 2nd ed.
88 PIERCE. Associati ve Algebras. 115 BERGERIGOSTIAUX. Differential Geometry:
89 LANG. Introduction to Algebraic and Manifolds, Curves, and Surfaces.
Abelian Functions. 2nd ed. 116 KELLEy/SRINIVASAN. Measure and Integral.
90 BR0NDSTED. An Introduction to Convex Vol.!.
Polytopes. 117 SERRE. Algebraic Groups and Class Fields.
91 BEARDON. On the Geometry of Discrete 118 PEDERSEN. Analysis Now.
Groups. 119 ROTMAN. An Introduction to Algebraic
92 DIESTEL. Sequences and Series in Banach Topology.
Spaces.
120 ZIEMER. Weakly Differentiable Functions: 149 RATCLIFFE. Foundations of
Sobolev Spaces and Functions of Bounded Hyperbolic Manifolds.
Variation. 150 EISENBUD. Commutative Algebra
121 LANG. Cyclotomic Fields I and II. with a View Toward Algebraic
Combined 2nd ed. Geometry.
122 REMMERT. Theory of Complex Functions. 151 SILVERMAN. Advanced Topics in
Readings in Mathematics the Arithmetic of Elliptic Curves.
123 EBBINGHAUS/HERMES et al. Numbers. 152 ZIEGLER. Lectures on Polytopes.
Readings in Mathematics 153 FULTON. Algebraic Topology: A
124 DUBROYIN/FoMENKOINOYIKOY. Modern First Course.
Geometry-Methods and Applications. 154 BROWN/PEARCY. An Introduction to
Part III. Analysis.
125 BERENSTEIN/GAY. Complex Variables: An 155 KASSEL. Quantum Groups.
Introduction. 156 KECHRIS. Classical Descriptive Set
126 BOREL. Linear Algebraic Groups. 2nd ed. Theory.
127 MASSEY. A Basic Course in Algebraic 157 MALLIAYIN. Integration and
Topology. Probability.
128 RAUCH. Partial Differential Equations. 158 ROMAN. Field Theory.
129 FULTON/HARRIS. Representation Theory: A 159 CONWAY. Functions of One
First Course. Complex Variable II.
Readings in Mathematics 160 LANG. Differential and Riemannian
130 DODSON/POSTON. Tensor Geometry. Manifolds.
131 LAM. A First Course in Noncommutative 161 BORWEIN/ERDELYI. Polynomials and
Rings. Polynomial Inequalities.
132 BEARDON. Iteration of Rational Functions. 162 ALPERIN/BELL. Groups and
133 HARRIS. Algebraic Geometry: A First Representations.
Course. 163 DIXON/MoRTIMER. Permutation
134 ROMAN. Coding and Information Theory. Groups.
135 ROMAN. Advanced Linear Algebra. 164 NATHANSON. Additive Number Theory:
136 ADKINS/WEINTRAUB. Algebra: An The Classical Bases.
Approach via Module Theory. 165 NATHANSON. Additive Number Theory:
137 AXLERIBoURDON/RAMEY. Harmonic Inverse Problems and the Geometry of
Function Theory. Sumsets.
138 COHEN. A Course in Computational 166 SHARPE. Differential Geometry: Cartan's
Algebraic Number Theory. Generalization of Klein's Erlangen
139 BREDON. Topology and Geometry. Program.
140 AUBIN. Optima and Equilibria. An 167 MORANDI. Field and Galois Theory.
Introduction to Nonlinear Analysis. 168 EWALD. Combinatorial Convexity and
141 BECKERIWEISPFENNING/KREDEL. Grobner Algebraic Geometry.
Bases. A Computational Approach to 169 BHATIA. Matrix Analysis.
Commutative Algebra. 170 BREDON. SheafTheory. 2nd ed.
142 LANG. Real and Functional Analysis. 171 PETERSEN. Riemannian Geometry.
3rd ed. 172 REMMERT. Classical Topics in Complex
143 Doo8. Measure Theory. Function Theory.
144 DENNIS/FARB. Noncommutative 173 DIESTEL. Graph Theory. 2nd ed.
Algebra. 174 BRIDGES. Foundations of Real and
145 VICK. Homology Theory. An Abstract Analysis.
Introduction to Algebraic Topology. 175 LICKORISH. An Introduction to Knot
2nd ed. Theory.
146 BRIDGES. Computability: A 176 LEE. Riemannian Manifolds.
Mathematical Sketchbook. 177 NEWMAN. Analytic Number Theory.
147 ROSENBERG. Algebraic K-Tbeory 178 CLARKEILEDY AEY/STERN/WOLENSKI.
and Its Applications. Nonsmooth Analysis and Control
148 ROTMAN. An Introduction to the Theory.
Theory of Groups. 4th ed. 179 DOUGLAS. Banach Algebra Techniques in
Operator Theory. 2nd ed.
180 SRIVASTA VA. A Course on Borel Sets. 192 HIRSCH/LACOMBE. Elements of Functional
181 KRESS. Numerical Analysis. Analysis.
182 WALTER. Ordinary Differential 193 COHEN. Advanced Topics in
Equations. Computational Number Theory.
183 MEGGINSON. An Introduction to Banach 194 ENGELINAGEL. One-Parameter Semigroups
Space Theory. for Linear Evolution Equations.
184 BOLLOBAS. Modern Graph Theory. 195 NATHANSON. Elementary Methods in
185 COx/LITTLE/O'SHEA. Using Algebraic Number Theory.
Geometry. 196 OSBORNE. Basic Homological Algebra.
186 RAMAKRISHNANIV ALENZA. Fourier 197 EISENBUD/ HARRIS. The Geometry of
Analysis on Number Fields. Schemes.
187 HARRiS/MORRISON. Moduli of Curves. 198 ROBERT. A Course in p-adic Analysis.
188 GOLDBLATT. Lectures on the Hyperreals: 199 HEDENMALM/KoRENBLUM/ZHU. Theory
An Introduction to Nonstandard Analysis. of Bergman Spaces.
189 LAM. Lectures on Modules and Rings. 200 BAO/CHERN/SHEN. An Introduction to
190 ESMONDE/MuRTY. Problems in Algebraic Riemann-Finsler Geometry.
Number Theory. 201 HINDRY/SILVERMAN. Diophantine
191 LANG. Fundamentals of Differential Geometry.
Geometry.