Algebraic Topology EMS Textbooks in Mathematics

Dieck_titelei 1.8.
2008 13:23 Uhr Seite 1

EMS Textbooks in Mathematics
EMS Textbooks in Mathematics is a book series aimed at students or professional mathematici-
ans seeking an introduction into a particular field. The individual volumes are intended to provide
not only relevant techniques, results and their applications, but afford insight into the motivations
and ideas behind the theory. Suitably designed exercises help to master the subject and prepare
the reader for the study of more advanced and specialized literature.
Jrn Justesen and Tom Hholdt, A Course In Error-Correcting Codes
Markus Stroppel, Locally Compact Groups
Peter Kunkel and Volker Mehrmann, Differential-Algebraic Equations
Dorothee D. Haroske and Hans Triebel, Distributions, Sobolev Spaces, Elliptic Equations
Thomas Timmermann, An Invitation to Quantum Groups and Duality
Marek Jarnicki and Peter Pflug, First Steps in Several Complex Variables: Reinhardt Domains
Oleg Bogopolski, Introduction to Group Theory
Dieck_titelei 1.8.2008 13:23 Uhr Seite 2
Tammo tom Dieck
Algebraic Topology
Author:
Tammo tom Dieck
Mathematisches Institut
Georg-August-Universitt Gttingen
Bunsenstrasse 35
37073 Gttingen
Germany
E-mail: [email protected]
2000 Mathematics Subject Classification: 55-01, 57-01
Key words: Covering spaces, fibrations, cofibrations, homotopy groups, cell complexes, fibre
bundles, vector bundles, classifying spaces, singular and axiomatic homology and cohomology,
smooth manifolds, duality, characteristic classes, bordism.
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
ISBN 978-3-03719-048-7
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation,
broadcasting, reproduction on microfilms or in other ways, and storage in data banks. For any kind of
use permission of the copyright owner must be obtained.
2008 European Mathematical Society
Contact address:
European Mathematical Society Publishing House
Seminar for Applied Mathematics
ETH-Zentrum FLI C4
CH-8092 Zrich
Switzerland
Phone: +41 (0)44 632 34 36
Email: [email protected]
Homepage: www.ems-ph.org
Typeset using the authors TEX files: I. Zimmermann, Freiburg
Printed on acid-free paper produced from chlorine-free pulp. TCF
Printed in Germany
9 8 7 6 5 4 3 2 1
Preface
Algebraic topology is the interplay between continuous and discrete mathe-
matics. Continuous mathematics is formulated in its general form in the language
of topological spaces and continuous maps. Discrete mathematics is used to express
the concepts of algebra and combinatorics. In mathematical language: we use the
real numbers to conceptualize continuous forms and we model these forms with the
use of the integers. For example, our intuitive idea of time supposes a continuous
process without gaps, an unceasing succession of moments. But in practice we
use discrete models, machines or natural processes which we dene to be periodic.
Likewise we conceive of a space as a continuum but we model that space as a set
of discrete forms. Thus the essence of time and space is of a topological nature but
algebraic topology allows their realizations to be of an algebraic nature.
Classical algebraic topology consists in the construction and use of functors
from some category of topological spaces into an algebraic category, say of groups.
But one can also postulate that global qualitative geometry is itself of an algebraic
nature. Consequently there are two important viewpoints fromwhich one can study
algebraic topology: homology and homotopy.
Homology, invented by Henri Poincar, is without doubt one of the most inge-
nious and inuential inventions in mathematics. The basic idea of homology is that
we start with a geometric object (a space) which is given by combinatorial data (a
simplicial complex). Then the linear algebra and boundary relations determined by
these data are used to produce homology groups.
In this book, the chapters on singular homology, homology, homological algebra
and cellular homology constitute an introduction to homology theory (construction,
axiomatic analysis, classical applications). The chapters require a parallel reading
this indicates the complexity of the material which does not have a simple intuitive
explanation. If one knows or accepts some results about manifolds, one should read
the construction of bordism homology. It appears in the nal chapter but offers a
simple explanation of the idea of homology.
The second aspect of algebraic topology, homotopy theory, begins again with the
construction of functors from topology to algebra. But this approach is important
from another view point. Homotopy theory shows that the category of topological
spaces has itself a kind of (hidden) algebraic structure. This becomes immediately
clear in the introductory chapters on the fundamental group and covering space
theory. The study of algebraic topology is often begun with these topics. The
notions of bration and cobration, which are at rst sight of a technical nature,
are used to indicate that an arbitrary continuous map has something like a kernel
and a cokernel the beginning of the internal algebraic structure of topology. (The
chapter on homotopy groups, which is essential to this book, should also be studied
vi Preface
for its applications beyond our present study.) In the ensuing chapter on duality
the analogy to algebra becomes clearer: For a suitable class of spaces there exists
a duality theory which resembles formally the duality between a vector space and
its dual space.
The rst main theorem of algebraic topology is the BrouwerHopf degree the-
orem. We prove this theorem by elementary methods from homotopy theory. It is
a fairly direct consequence of the BlakersMassey excision theorem for which we
present the elementary proof of Dieter Puppe. Later we indicate proofs of the de-
gree theorem based on homology and then on differential topology. It is absolutely
essential to understand this theorem from these three view points. The theorem
says that the set of self-maps of a positive dimensional sphere under the homotopy
relation has the structure of a (homotopically dened) ring and this ring is the
ring of integers.
The second part of the book develops further theoretical concepts (like coho-
mology) and presents more advanced applications to manifolds, bundles, homotopy
theory, characteristic classes and bordism theory. The reader is strongly urged to
read the introduction to each of the chapters in order to obtain more coherent infor-
mation about the contents of the book.
Words in boldface italic are dened at the place where they appear even if there
is no indication of a formal denition. In addition, there is a list of standard or global
symbols. The problem sections contain exercises, examples, counter-examples and
further results, and also sometimes ask the reader to extend concepts in further
detail. It is not assumed that all of the problems will be completely worked out, but
it is strongly recommended that they all be read. Also, the reader will nd some
familiarity with the full bibliography, not just the references cited in the text, to be
crucial for further studies. More background material about spaces and manifolds
may, at least for a while, be obtained from the authors home page.
I would like to thank Irene Zimmermann and Manfred Karbe for their help and
effort in preparing the manuscript for publication.
Gttingen, September 2008 Tammo tom Dieck
Contents
Preface v
1 Topological Spaces 1
1.1 Basic Notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Subspaces. Quotient Spaces . . . . . . . . . . . . . . . . . . . . 5
1.3 Products and Sums . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Proper Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Paracompact Spaces . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7 Topological Groups . . . . . . . . . . . . . . . . . . . . . . . . 15
1.8 Transformation Groups . . . . . . . . . . . . . . . . . . . . . . . 17
1.9 Projective Spaces. Grassmann Manifolds . . . . . . . . . . . . . 21
2 The Fundamental Group 24
2.1 The Notion of Homotopy . . . . . . . . . . . . . . . . . . . . . 25
2.2 Further Homotopy Notions . . . . . . . . . . . . . . . . . . . . . 30
2.3 Standard Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Mapping Spaces and Homotopy . . . . . . . . . . . . . . . . . . 37
2.5 The Fundamental Groupoid . . . . . . . . . . . . . . . . . . . . 41
2.6 The Theorem of Seifert and van Kampen . . . . . . . . . . . . . 45
2.7 The Fundamental Group of the Circle . . . . . . . . . . . . . . . 47
2.8 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.9 Homotopy Groupoids . . . . . . . . . . . . . . . . . . . . . . . 58
3 Covering Spaces 62
3.1 Locally Trivial Maps. Covering Spaces . . . . . . . . . . . . . . 62
3.2 Fibre Transport. Exact Sequence . . . . . . . . . . . . . . . . . 66
3.3 Classication of Coverings . . . . . . . . . . . . . . . . . . . . . 70
3.4 Connected Groupoids . . . . . . . . . . . . . . . . . . . . . . . 72
3.5 Existence of Liftings . . . . . . . . . . . . . . . . . . . . . . . . 76
3.6 The Universal Covering . . . . . . . . . . . . . . . . . . . . . . 78
4 Elementary Homotopy Theory 81
4.1 The Mapping Cylinder . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 The Double Mapping Cylinder . . . . . . . . . . . . . . . . . . . 84
4.3 Suspension. Homotopy Groups . . . . . . . . . . . . . . . . . . 86
4.4 Loop Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
viii Contents
4.5 Groups and Cogroups . . . . . . . . . . . . . . . . . . . . . . . 90
4.6 The Cobre Sequence . . . . . . . . . . . . . . . . . . . . . . . 92
4.7 The Fibre Sequence . . . . . . . . . . . . . . . . . . . . . . . . 97
5 Cobrations and Fibrations 101
5.1 The Homotopy Extension Property . . . . . . . . . . . . . . . . 101
5.2 Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.3 Replacing a Map by a Cobration . . . . . . . . . . . . . . . . . 110
5.4 Characterization of Cobrations . . . . . . . . . . . . . . . . . . 113
5.5 The Homotopy Lifting Property . . . . . . . . . . . . . . . . . . 115
5.6 Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.7 Replacing a Map by a Fibration . . . . . . . . . . . . . . . . . . 120
6 Homotopy Groups 121
6.1 The Exact Sequence of Homotopy Groups . . . . . . . . . . . . 122
6.2 The Role of the Base Point . . . . . . . . . . . . . . . . . . . . . 126
6.3 Serre Fibrations . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.4 The Excision Theorem . . . . . . . . . . . . . . . . . . . . . . . 133
6.5 The Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.6 The Brouwer Fixed Point Theorem . . . . . . . . . . . . . . . . 137
6.7 Higher Connectivity . . . . . . . . . . . . . . . . . . . . . . . . 141
6.8 Classical Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6.9 Proof of the Excision Theorem . . . . . . . . . . . . . . . . . . . 148
6.10 Further Applications of Excision . . . . . . . . . . . . . . . . . . 152
7 Stable Homotopy. Duality 159
7.1 A Stable Category . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.2 Mapping Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.3 Euclidean Complements . . . . . . . . . . . . . . . . . . . . . . 168
7.4 The Complement Duality Functor . . . . . . . . . . . . . . . . . 169
7.5 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.6 Homology and Cohomology for Pointed Spaces . . . . . . . . . 179
7.7 Spectral Homology and Cohomology . . . . . . . . . . . . . . . 181
7.8 Alexander Duality . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.9 Compactly Generated Spaces . . . . . . . . . . . . . . . . . . . 186
8 Cell Complexes 196
8.1 Simplicial Complexes . . . . . . . . . . . . . . . . . . . . . . . 197
8.2 Whitehead Complexes . . . . . . . . . . . . . . . . . . . . . . . 199
8.3 CW-Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.4 Weak Homotopy Equivalences . . . . . . . . . . . . . . . . . . . 207
8.5 Cellular Approximation . . . . . . . . . . . . . . . . . . . . . . 210
8.6 CW-Approximation . . . . . . . . . . . . . . . . . . . . . . . . 211
Contents ix
8.7 Homotopy Classication . . . . . . . . . . . . . . . . . . . . . . 216
8.8 EilenbergMac Lane Spaces . . . . . . . . . . . . . . . . . . . . 217
9 Singular Homology 223
9.1 Singular Homology Groups . . . . . . . . . . . . . . . . . . . . 224
9.2 The Fundamental Group . . . . . . . . . . . . . . . . . . . . . . 227
9.3 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
9.4 Barycentric Subdivision. Excision . . . . . . . . . . . . . . . . . 231
9.5 Weak Equivalences and Homology . . . . . . . . . . . . . . . . 235
9.6 Homology with Coefcients . . . . . . . . . . . . . . . . . . . . 237
9.7 The Theorem of Eilenberg and Zilber . . . . . . . . . . . . . . . 238
9.8 The Homology Product . . . . . . . . . . . . . . . . . . . . . . 241
10 Homology 244
10.1 The Axioms of Eilenberg and Steenrod . . . . . . . . . . . . . . 244
10.2 Elementary Consequences of the Axioms . . . . . . . . . . . . . 246
10.3 Jordan Curves. Invariance of Domain . . . . . . . . . . . . . . . 249
10.4 Reduced Homology Groups . . . . . . . . . . . . . . . . . . . . 252
10.5 The Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
10.6 The Theorem of Borsuk and Ulam . . . . . . . . . . . . . . . . . 261
10.7 MayerVietoris Sequences . . . . . . . . . . . . . . . . . . . . . 265
10.8 Colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.9 Suspension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
11 Homological Algebra 275
11.1 Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
11.2 Exact Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . 279
11.3 Chain Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . 283
11.4 Cochain complexes . . . . . . . . . . . . . . . . . . . . . . . . . 285
11.5 Natural Chain Maps and Homotopies . . . . . . . . . . . . . . . 286
11.6 Chain Equivalences . . . . . . . . . . . . . . . . . . . . . . . . 287
11.7 Linear Algebra of Chain Complexes . . . . . . . . . . . . . . . . 289
11.8 The Functors Tor and Ext . . . . . . . . . . . . . . . . . . . . . 292
11.9 Universal Coefcients . . . . . . . . . . . . . . . . . . . . . . . 295
11.10 The Knneth Formula . . . . . . . . . . . . . . . . . . . . . . . 298
12 Cellular Homology 300
12.1 Cellular Chain Complexes . . . . . . . . . . . . . . . . . . . . . 300
12.2 Cellular Homology equals Homology . . . . . . . . . . . . . . . 304
12.3 Simplicial Complexes . . . . . . . . . . . . . . . . . . . . . . . 306
12.4 The Euler Characteristic . . . . . . . . . . . . . . . . . . . . . . 308
12.5 Euler Characteristic of Surfaces . . . . . . . . . . . . . . . . . . 311
x Contents
13 Partitions of Unity in Homotopy Theory 318
13.1 Partitions of Unity . . . . . . . . . . . . . . . . . . . . . . . . . 318
13.2 The Homotopy Colimit of a Covering . . . . . . . . . . . . . . . 321
13.3 Homotopy Equivalences . . . . . . . . . . . . . . . . . . . . . . 324
13.4 Fibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
14 Bundles 328
14.1 Principal Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . 328
14.2 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
14.3 The Homotopy Theorem . . . . . . . . . . . . . . . . . . . . . . 342
14.4 Universal Bundles. Classifying Spaces . . . . . . . . . . . . . . 344
14.5 Algebra of Vector Bundles . . . . . . . . . . . . . . . . . . . . . 351
14.6 Grothendieck Rings of Vector Bundles . . . . . . . . . . . . . . 355
15 Manifolds 358
15.1 Differentiable Manifolds . . . . . . . . . . . . . . . . . . . . . . 358
15.2 Tangent Spaces and Differentials . . . . . . . . . . . . . . . . . 362
15.3 Smooth Transformation Groups . . . . . . . . . . . . . . . . . . 366
15.4 Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . 369
15.5 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
15.6 Tangent Bundle. Normal Bundle . . . . . . . . . . . . . . . . . . 374
15.7 Embeddings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
15.8 Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
15.9 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
15.10 Gluing along Boundaries . . . . . . . . . . . . . . . . . . . . . . 388
16 Homology of Manifolds 392
16.1 Local Homology Groups . . . . . . . . . . . . . . . . . . . . . . 392
16.2 Homological Orientations . . . . . . . . . . . . . . . . . . . . . 394
16.3 Homology in the Dimension of the Manifold . . . . . . . . . . . 396
16.4 Fundamental Class and Degree . . . . . . . . . . . . . . . . . . 399
16.6 Winding and Linking Numbers . . . . . . . . . . . . . . . . . . 403
17 Cohomology 405
17.1 Axiomatic Cohomology . . . . . . . . . . . . . . . . . . . . . . 405
17.2 Multiplicative Cohomology Theories . . . . . . . . . . . . . . . 409
17.3 External Products . . . . . . . . . . . . . . . . . . . . . . . . . . 413
17.4 Singular Cohomology . . . . . . . . . . . . . . . . . . . . . . . 416
17.5 EilenbergMac Lane Spaces and Cohomology . . . . . . . . . . 419
17.6 The Cup Product in Singular Cohomology . . . . . . . . . . . . 422
17.7 Fibration over Spheres . . . . . . . . . . . . . . . . . . . . . . . 425
17.8 The Theorem of Leray and Hirsch . . . . . . . . . . . . . . . . . 427
Contents xi
17.9 The Thom Isomorphism . . . . . . . . . . . . . . . . . . . . . . 431
18 Duality 438
18.1 The Cap Product . . . . . . . . . . . . . . . . . . . . . . . . . . 438
18.2 Duality Pairings . . . . . . . . . . . . . . . . . . . . . . . . . . 441
18.3 The Duality Theorem . . . . . . . . . . . . . . . . . . . . . . . . 444
18.4 Euclidean Neighbourhood Retracts . . . . . . . . . . . . . . . . 447
18.5 Proof of the Duality Theorem . . . . . . . . . . . . . . . . . . . 451
18.7 The Intersection Form. Signature . . . . . . . . . . . . . . . . . 457
18.8 The Euler Number . . . . . . . . . . . . . . . . . . . . . . . . . 461
18.9 Euler Class and Euler Characteristic . . . . . . . . . . . . . . . . 464
19 Characteristic Classes 467
19.1 Projective Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 468
19.2 Projective Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 471
19.3 Chern Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
19.4 StiefelWhitney Classes . . . . . . . . . . . . . . . . . . . . . . 478
19.5 Pontrjagin Classes . . . . . . . . . . . . . . . . . . . . . . . . . 479
19.6 Hopf Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
19.7 Hopf Algebras and Classifying Spaces . . . . . . . . . . . . . . . 486
19.8 Characteristic Numbers . . . . . . . . . . . . . . . . . . . . . . 491
20 Homology and Homotopy 495
20.1 The Theorem of Hurewicz . . . . . . . . . . . . . . . . . . . . . 495
20.2 Realization of Chain Complexes . . . . . . . . . . . . . . . . . . 501
20.3 Serre Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
20.4 Qualitative Homology of Fibrations . . . . . . . . . . . . . . . . 505
20.5 Consequences of the Fibration Theorem . . . . . . . . . . . . . . 508
20.6 Hurewicz and Whitehead Theorems modulo Serre classes . . . . . 510
20.7 Cohomology of EilenbergMac Lane Spaces . . . . . . . . . . . 513
20.8 Homotopy Groups of Spheres . . . . . . . . . . . . . . . . . . . 514
20.9 Rational Homology Theories . . . . . . . . . . . . . . . . . . . . 518
21 Bordism 521
21.1 Bordism Homology . . . . . . . . . . . . . . . . . . . . . . . . 521
21.2 The Theorem of Pontrjagin and Thom . . . . . . . . . . . . . . . 529
21.3 Bordism and Thom Spectra . . . . . . . . . . . . . . . . . . . . 535
21.4 Oriented Bordism . . . . . . . . . . . . . . . . . . . . . . . . . 537
Bibliography 541
Symbols 551
Index 557
Chapter 1
Topological Spaces
In this chapter we collect the basic terminology about topological spaces and some
elementary results (without proofs). I assume that the reader has some experience
with point-set topology including the notion of compactness. We introduce a num-
ber of examples and standard spaces that will be used throughout the book. Perhaps
the reader has not met quotient spaces. Quotient spaces give precision to the in-
tuitive concept of gluing and pasting. They comprise adjunction spaces, pushouts,
attaching of spaces (in particular cells), orbit spaces of group actions. In the main
text we deal with other topics: Mapping spaces and compact open topology, bun-
dles, cell complexes, manifolds, partitions of unity, compactly generated spaces.
Transformation groups are another topic of this chapter. Whenever you study a
mathematical object you should consider its symmetries. In topology one uses, of
course, continuous symmetries. They are called actions of a topological group on
a space or transformation groups. In this chapter we assemble notions and results
about the general topology of transformation groups. We use the material later for
several purposes:
Important spaces like spheres, projective spaces and Grassmann manifolds
have a high degree of symmetry which comes from linear algebra (matrix
multiplication).
The fundamental group of a space has a somewhat formal denition. In the
theory of covering spaces the fundamental group is exhibited as a symmetry
group. This hidden symmetry, which is associated to a space, will inuence
several other of its geometric investigations.
The theory of bre bundles and vector bundles makes essential use of the
concept of a transformation group. Important information about a manifold
is codied in its tangent bundle. We will apply the tools of algebraic topology
to bundles (characteristic classes; classifying spaces).
We should point out that large parts of algebraic topology can be generalized
to the setting of transformation groups (equivariant topology). At a few occasions
later we point out such generalizations.
1.1 Basic Notions
Atopology on a set X is a set Oof subsets of X, called open sets, with the properties:
(1) The union of an arbitrary family of open sets is open. (2) The intersection of a
nite family of open sets is open. (3) The empty set 0and X are open. Atopological
2 Chapter 1. Topological Spaces
space (X. O) consists of a set X and a topology O on X. The sets in O are the open
sets of the topological space (X. O). We usually denote a topological space just by
the underlying set X. A set X is closed in (X. O) if the complement X is
open in (X. O). Closed sets have properties dual to (1)(3): The intersection of an
arbitrary family of closed sets is closed; the union of a nite family of closed sets
is closed; the empty set 0 and X are closed. A subset B of a topology O is a basis
of O if each U O is a union of elements of B. (The empty set is the union of the
empty family.) A subset S of O is a subbasis of O if the set of nite intersections
of elements in S is a basis of O. (The space X is the intersection of the empty
family.)
A map : X Y between topological spaces is continuous if the pre-image
-1
(V ) of each open set V of Y is open in X. Dually: A map is continuous
if the pre-image of each closed set is closed. The identity id(X): X X is
always continuous, and the composition of continuous maps is continuous. Hence
topological spaces and continuous maps form a category. We denote it by TOP.
A homeomorphism : X Y is a continuous map with a continuous inverse
g: Y X. Spaces X and Y are homeomorphic if there exists a homeomorphism
between them. A map : X Y between topological spaces is open (closed) if
the image of each open (closed) set is again open (closed).
In the sequel we assume that a map between topological spaces is continuous
if nothing else is specied or obvious. A set map is a map which is not assumed to
be continuous at the outset.
We x a topological space X and a subset . The intersection of the closed
sets which contain is denoted

and called closure of in X. A set is dense
in X if

= X. The interior of is the union of the open sets contained in .
We denote the interior by
. A point in
is an interior point of . A subset is

nowhere dense if the interior of its closure is empty. The boundary of in X is
Bd() =

(X ).
An open subset U of X which contains is an open neighbourhood of in X.
Aset T is a neighbourhood of if it contains an open neighbourhood. Asystemof
neighbourhoods of the point . is a neighbourhood basis of . if each neighbourhood
of . contains one of the system.
Given two topological spaces X and Y , a map : X Y is said to be continu-
ous at . X if for eachneighbourhoodV of (.) there exists a neighbourhoodU of
. such that (U) V ; it sufces to consider a neighbourhood basis of . and (.).
Suppose O
1
and O
2
are topologies on X. If O
1
O
2
, then O
2
is ner than
O
1
and O
1
coarser than O
2
. The topology O
2
is ner than O
1
if and only if the
identity (X. O
2
) (X. O
1
) is continuous. The set of all subsets of X is the nest
topology; it is the discrete topology and the resulting space a discrete space. All
maps : X Y from a discrete space X are continuous. The coarsest topology
on X consists of 0 and X alone. If (O
}
[ J) is a family of topologies on X,
then their intersection is a topology.
1.1. Basic Notions 3
We list some properties which a space X may have.
(T
1
) One-point subspaces are closed.
(T
2
) Any two points have disjoint neighbourhoods.
(T
3
) Given a point . X and a closed subset X not containing ., there exist
disjoint neighbourhoods U of . and V of .
(T
4
) Any two disjoint closed subsets have disjoint neighbourhoods.
We say X satises the separation axiom T
}
(or X is a T
}
-space), if X has property
T
}
. The separation axioms are of a technical nature, but they serve the purpose of
clarifying the concepts.
A T
2
-space is called a Hausdorff space or separated. A space satisfying T
1
and T
3
is said to be regular. A space satisfying T
1
and T
4
is called normal. In
a regular space, each neighbourhood of a point contains a closed neighbourhood.
A space X is called completely regular if it is separated and for each . X and
0 ,= X closed, . , there exists a continuous function : X 0. 1| such
that (.) = 1 and () = {0].
A remarkable consequence of the separation property T
4
is the existence of
many real-valued continuous functions. The Urysohn existence theorem (1.1.1)
shows that normal spaces are completely regular.
(1.1.1) Theorem (Urysohn). Let X be a T
4
-space and suppose that and T are
disjoint closed subsets of X. Then there exists a continuous function : X 0. 1|
with () {0] and (T) {1].
(1.1.2) Theorem (Tietze). Let X be a T
4
-space and X closed. Then each
continuous map : 0. 1| has a continuous extension : X 0. 1|.
A continuous map : R
n
from a closed subset of a T
4
-space X has a
continuous extension to X.
Many examples of topological spaces arise from metric spaces. Metric spaces
are important in their own right. Ametric J on a set X is a map J : XX 0. o
with the properties:
(1) J(.. ,) = 0 if and only if . = ,.
(2) J(.. ,) = J(,. .) for all .. , X.
(3) J(.. :) _ J(.. ,) J(,. :) for all .. ,. : X (triangle inequality).
We call J(.. ,) the distance between . and , with respect to the metric J. Ametric
space (X. J) consists of a set X and a metric J on X.
Let (X. J) be a metric space. The set U
t
(.) = {, X [ J(.. ,) < c] is
the "-neighbourhood of .. We call U X open with respect to J if for each
. X there exists c > 0 such that U
t
(.) U. The systemO
d
of subsets U which
are open with respect to J is a topology on X, the underlying topology of the
metric space, and the c-neighbourhoods of all points are a basis for this topology.
Subsets of the form U
t
(.) are open with respect to J. For the proof, let , U
t
(.)
and 0 < j < c J(.. ,). Then, by the triangle inequality, U
)
(,) U
t
(.). A
space (X. O) is metrizable if there exists a metric J on X such that O = O
d
.
Metrizable spaces have countable neighbourhood bases of points: Take the U
t
(.)
with rational c. A set U is a neighbourhood of . if and only if there exists an
c > 0 such that U
t
(.) U. For metric spaces our denition of continuity is
equivalent to the familiar denition of calculus: A map : X Y between
metric spaces is continuous at a X if for each c > 0 there exists > 0 such that
J(a. .) < implies J((a). (.)) < c. Continuityonlydepends onthe underlying
topology. But a metric is a ner and more rigid structure; one can compare the size
of neighbourhoods of different points and one can dene uniform continuity. A
map : (X. J
1
) (Y. J
2
) between metric spaces is uniformly continuous if for
each c > 0 there exists > 0 such that J
1
(.. ,) < implies J
2
((.). (,)) < c.
A sequence
n
: X Y of maps into a metric space (Y. J) converges uniformly
to : X Y if for each c > 0 there exists N such that for n > N and . X
the inequality J((.).
n
(.)) < c holds. If the
n
are continuous functions from a
topological space X which converge uniformly to , then one shows as in calculus
that is continuous.
A set in a metric space (X. J) is bounded if {J(.. ,) [ .. , ] is bounded
in R. The supremumof the latter set is then the diameter of . We dene J(.. ) =
inf{J(.. a) [ a ] as the distance of . from ,= 0. The relation [J(.. )
J(,. )[ _ J(.. ) shows that the map X R, . J(.. ) is uniformly
continuous. The relation J(.. ) = 0 is equivalent to .

.
If and T are disjoint, non-empty, closed sets in X, then
: X 0. 1|. . J(.. )(J(.. ) J(.. T))
-1
is a continuous function with () = {0] and (T) = {1]. Let 0 < a < b < 1.
Then 0. a and |b. 1| are open in 0. 1|, and their pre-images under are disjoint
open neighbourhoods of and T. Hence a metric space is normal.
Adirected set (1. _) consists of a set 1 and a relation _on 1 such that: (1) i _ i
for all i 1. (2) i _ , _ k implies i _ k. (3) For each pair i. 1 there
exists k 1 such that i _ k, _ k. We also write _ i for i _ . The set N
with the usual order is directed. The set U(.) of neighbourhoods of . is directed
by U _ V =V U.
Anet with directed index set 1 in X is a map 1 X, i .
i
. We write (.
i
)
iJ
or just (.
i
) for such a net. A net (.
i
) in a topological space X converges to .,
notation . = lim.
i
, provided for each neighbourhood U of . there exists i 1
such that .
}
U for _ i . If one chooses from each U U(.) a point .
U
, then
the net (.
U
) with index set U(.) converges to .. A point . is an accumulation
value of the net (.
i
)
iJ
if for each neighbourhood U of . and each i 1 there
exists _ i such that .
}
U. Let 1 and J be directed sets. A map h: 1 J
is nal if for each J there exists i 1 such that h(i ) _ . A subnet of a net
(.
}
)
}J
is a net of the form i .
h(i)
with a nal map h: 1 J. If a subnet of
1.2. Subspaces. Quotient Spaces 5
(.
}
)
}J
converges to :, then : is an accumulation value. Each accumulation value
is a convergence point of a subnet.
Let (.
}
)
}J
be a net in X. For J let J( ) be the closure of {.
k
[ k _ ].
Then J =
_
}J
J( ) is the set of accumulation values of the net.
1.2 Subspaces. Quotient Spaces
It is a classical idea and method to dene geometric objects (spaces) as subsets of
Euclidean spaces, e.g., as solution sets of a system of equations. But it is important
to observe that such objects have absolute properties which are independent of
their position in the ambient space. In the topological context this absolute property
is the subspace topology.
Let (X. O) be a topological space and X a subset. Then
O[ = {U [ there exists V O with U = V ]
is a topology on . It is called the induced topology, the subspace topology, or the
relative topology. The space (. O[) is called a subspace of (X. O); we usually
say: is a subspace of X. Acontinuous map : (Y. S) (X. O) is an embedding
if it is injective and (Y. S) ((Y ). O[(Y )), , (,) a homeomorphism.
From the denition one veries:
Let be a subspace of X. Then the inclusion i : X, a a is continuous.
Let Y be a space and : Y X a set map with (Y ) . Then is continuous
if and only if : Y , , (,) is continuous.
(1.2.1) Proposition. Let i : Y X be an injective set map between spaces. The
following are equivalent:
(1) i is an embedding.
(2) A set map g: 7 Y from any topological space 7 is continuous if and only
if ig: 7 X is continuous.
Property (2) characterizes embeddings i in categorical terms. We call this
property the universal property of an embedding.
Suppose T X are subspaces. If is closed in T and T closed in X, then
is closed in X. Similarly for open subspaces. But in general, an open (closed)
subset of T must not be open (closed) in X. The next proposition will be used
many times without further reference.
(1.2.2) Proposition. Let : X Y be a set map between topological spaces and
let X be the union of the subsets (X
}
[ J). If the X
}
are open and the maps
}
= [X
}
continuous, then is continuous. A similar assertion holds if the X
}
are closed and J is nite.
A subset of a space X is a retract of X if there exists a retraction r : X ,
i.e., a continuous map r : X such that r[ = id(). A continuous map
s : T 1 is a section of the continuous map : 1 T if s = id(T). In that
case s is an embedding onto its image.
In geometric and algebraic topology many of the important spaces are con-
structed as quotient spaces. They are obtained froma given space by an equivalence
relation. Although the quotient topology is easily dened, formally, it takes some
time to work with it. In several branches of mathematics quotient objects are more
difcult to handle than subobjects. Even if one starts with a nice and well-known
space, its quotient spaces may have strange properties; usually one has to add a
number of hypotheses in order to exclude unwanted phenomena. Quotient spaces
do not, in general, inherit desirable properties from the original space.
Let X be a topological space and : X Y a surjective map onto a set Y .
Then S = {U Y [
-1
(U) open in X] is a topology on Y . This is the nest
topology on Y such that is continuous. We call S the quotient topology on Y
with respect to . A surjective map : X Y between topological spaces is
called an identication or quotient map if it has the following property: U Y
open =
-1
(U) X open. If : X Y is a quotient map, then Y is called a
quotient space of X.
We recall that a surjective map : X Y is essentially the same thing as an
equivalence relation on X. If 1 is an equivalence relation on X, then X,1 denotes
the set of equivalence classes. The canonical map : X X,1assigns to . X
its equivalence class. If : X Y is surjective, then . ~ , = (.) = (,) is
an equivalence relation 1
(
on X. There is a canonical bijection : X,1
(
Y
such that = . The quotient spaceX,1 is dened to be the set X,1 together
with the quotient topology of the canonical map : X X,1.
If X, we denote
1
by X, the space obtained from X by identifying to
a point. In the case that = 0, we understand by this symbol together with a
disjoint point (topological sum (1.3.4)).
(1.2.3) Proposition. Let : X Y be a surjective map between spaces. The
following are equivalent:
(1) is a quotient map.
(2) A set map g: Y 7 into any topological space 7 is continuous if and only
if g : X 7 is continuous.
Property (2) characterizes quotient maps in categorical terms. We call this
property the universal property of a quotient map.
A subset of X is saturated with respect to an equivalence relation if it is a union
of equivalence classes.
Let : X be an inclusion and : Y a continuous map. We identify in
the topological sumXY for each a the point a X with the point (a) Y ,
1
A similar notation is used for factor groups and orbit spaces.
1.2. Subspaces. Quotient Spaces 7
i.e., we consider the equivalence relation on the topological sum(1.3.4) XY with
equivalence classes {:] for : , () and
-1
(:) {:] for : (). The
quotient space 7 is sometimes denoted by Y L
(
X and called the adjunction space
obtained by attaching X via to Y . The canonical inclusions X X Y and
Y X Y induce maps J : X Y L
(
X and J : Y Y L
(
X. The diagram
Y
J
X
T
7 = Y L
(
X
is a pushout in TOP. If is an embedding, then J is an embedding.
(1.2.4) Proposition. Let be a closed embedding. Then the data of the pushout
have the following properties:
(1) J is a closed embedding.
(2) J restricted to X is an open embedding.
(3) If X, Y are T
1
-spaces (T
4
-spaces), then Y L
(
X is a T
1
-space (T
4
-space).
(4) If is a quotient map, then J is a quotient map.
Because of (1) and (2) we identify X with the open subspace J(X )
and Y with the closed subspace J(Y ). In this sense, Y L
(
X is the union of the
disjoint subsets X and Y .
(1.2.5) Proposition. The space Y L
(
X is a Hausdorff space, provided the fol-
lowing holds: Y is a Hausdorff space, X is regular, and is a retract of an open
neighbourhood in X.
Problems
1. Let : X Y and g: Y 7 be continuous maps. If and g are embeddings, then
g is an embedding. If g and g are embeddings, then g is an embedding. If g = id, then
is an embedding. An embedding is open (closed) if and only if its image is open (closed).
If : X Y is a homeomorphism and X, then the map (), induced by , is
a homeomorphism.
2. Let : X Y be a quotient map. Let T be open or closed in Y and set =
-1
(T).
Then the restriction g: T of is a quotient map.
3. Let : X Y be surjective, continuous and open (or closed). Then is a quotient map.
The restriction
!
:
-1
(T) T is open (or closed) for each T Y , hence a quotient
map.
4. The exponential map exp: C C
= C {0] is open. Similarly : R S

1
,
t exp(2i t ) is open. The kernel of is Z. Let q : R R,Z be the quotient map
onto the factor group. There is a bijective map : R,Z S
1
which satises q = .
Since and q are quotient maps, is a homeomorphism. The continuous periodic functions
: R R. (. 1) = (.) therefore correspond to continuous maps R,Z R and to
continuous maps S
1
R via composition with q or . In a similar manner one obtains a
homeomorphism C,Z C
.
5. Let : T and g: T C be continuous. If and g are quotient maps, then g is
a quotient map. If g is a quotient map, then g is a quotient map. If g = id, then g is a
quotient map.
1.3 Products and Sums
Let ((X
}
. O
}
) [ J) be a family of topological spaces. The product set X =
}J
X
}
is the set of all families (.
}
[ J) with .
}
X
}
. We have the
projection pr
i
: X X
i
, (.
}
) .
i
into the i -th factor. Let X
}
. Y
}
be topological
spaces and
}
: X
}
Y
}
maps. The product map
}
:

X
}

Y
}
is dened
as (.
}
[ J) (
}
(.
}
) [ J). Given maps
}
: Y X
}
we denote by
(
}
) = (
}
[ J): Y
}
X
}
the map with components pr
i
(
}
) =
i
.
The family of all pre-images pr
-1
}
(U
}
), U
}
X
}
open in X
}
(for varying ),
is the subbasis for the product topology O on X. We call (X. O) the topological
product of the spaces (X
}
. O
}
). The next proposition shows that X =

X
}
together with the projections pr
}
is a categorical product of the family (X
}
) in the
category TOP. Note that for innite J, open sets in the product are quite large; a
product
U
}
, U
}
X
}
open, is then in general not an open subset of
X
}
.
(1.3.1) Proposition. The product topology is the coarsest topology for which all
projections pr
}
are continuous. A set map : Y X from a space Y into X is
continuous if and only if all maps pr
}
are continuous. The product =
}

}
of continuous maps
}
: X
}
Y
}
is continuous.
The product of X
1
. X
2
is denoted X
1
X
2
, and we use
1
2
for the product of
maps. The identity id: X
1
(X
2
X
3
) (X
1
X
2
) X
3
is a homeomorphism.
In general, the topological product is associative, i.e., compatible with arbitrary
bracketing. The canonical identication R
k
R
I
= R
kI
is a homeomorphism.
1.3.2 Pullback. Let : X T and g: Y T be continuous maps. Let
7 = {(.. ,) X Y [ (.) = g(,)] with the subspace topology of X Y . We
have the projections ontothe factors J : 7 Y andG: 7 X. The commutative
diagram
7
T
X
(
T
is a pullback in TOP. The space 7 is sometimes written 7 = X
B
Y and called the
product of X and Y over T (the product in the category TOP
B
of spaces over T).
1.3. Products and Sums 9
Pullbacks allowone to convert liftings into sections. Let i : X and a:
Y such that ga = [is given. The assignment (o : X 7) (J o : X Y )
sets up a bijection between sections of G with Jo[ = a and maps : X Y
such that [ = a and g = .
(1.3.3) Proposition. Let : X Y be surjective, continuous, and open. Then Y
is separated if and only if 1 = {(.
1
. .
2
) [ (.
1
) = (.
2
)] is closed in XX.
Let (X
}
[ J) be a family of non-empty pairwise disjoint spaces. The set
O = {U lX
}
[ U X
}
X
}
open for all ] is a topology on the disjoint
union lX
}
. We call (lX
}
. O) the topological sum of the X
}
. A sum of two
spaces is denoted X
1
X
2
. The following assertions are easily veried from
the denitions. They show that the topological sum together with the canonical
inclusions X
}
lX
}
is a categorical sum in TOP. Given maps
}
: X
}
7 we
denote by (
}
): lX
}
7 the map with restriction
}
to X
}
.
(1.3.4) Proposition. A topological sumhas the following properties: The subspace
topology of X
}
in lX
}
is the original topology. Let the space X be the union of the
family (X
}
[ J) of pairwise disjoint subsets. Then X is the topological sum of
the subspaces X
}
if and only if the X
}
are open. : lX
}
Y is continuous if
each [X
}
: X
}
Y is continuous.
1.3.5 Pushout. Let : X and : T be continuous maps and form a
pushout diagram
T
J
X
T
Y
in the category SET of sets. Then Y is obtainable as a quotient of X T. We give
Y the quotient topology via (J. J ): X T Y . Then the resulting diagram is
a pushout in TOP. The space Y is sometimes written X
T and called the sum

of X and T under (the sum in the category TOP
of spaces under ).
1.3.6 Clutching. An important method for the construction of spaces is to paste
open subsets; see the example (1.3.8) for the simplest case. Let (U
}
[ J) be a
family of sets. Assume that for each pair (i. ) J J a subset U
}
i
U
i
is given
as well as a bijection g
}
i
: U
}
i
U
i
}
. We call the families (U
}
. U
k
}
. g
k
}
) a clutching
datum if:
(1) U
}
= U
}
}
and g
}
}
= id.
(2) For each triple (i. . k) J J J the map g
}
i
induces a bijection
g
}
i
: U
}
i
U
k
i
U
i
}
U
k
}
and g
k
}
g
}
i
= g
k
i
holds, considered as maps from U
}
i
U
k
i
to U
}
k
U
i
k
.
Given a clutching datum, we have an equivalence relation on the disjoint sum
l
}J
U
}
:
. U
i
~ , U
}
= . U
}
i
and g
}
i
(.) = ,.
Let X denote the set of equivalence classes and let h
i
: U
i
X be the map which
sends . U
i
to its class. Then h
i
is injective. Set U(i ) = image h
i
, then
U(i ) U( ) = h
i
(U
}
i
).
Conversely, assume that X is a quotient of l
}J
U
}
such that each h
i
: U
i
X
is injective with image U(i ). Let U
}
i
= h
-1
i
(U(i ) U( )) and g
}
i
= h
-1
}

h
i
: U
}
i
U
i
}
. Then the (U
i
. U
}
i
. g
}
i
) are a clutching datum. If we apply the
construction above to this datum, we get back X and the h
i
.
(1.3.7) Proposition. Let (U
i
. U
}
i
. g
}
i
) be a clutching datum. Assume that the U
i
are topological spaces, the U
}
i
U
i
open subsets, and the g
}
i
: U
}
i
U
i
}
homeo-
morphisms. Let X carry the quotient topology with respect to the quotient map
: l
}J
U
}
X. Then the following holds:
(1) The map h
i
is a homeomorphism onto an open subset of X and is open.
(2) Suppose the U
i
are Hausdorff spaces. Then X is a Hausdorff space if and
only if for each pair (i. ) the map ;
}
i
: U
}
i
U
i
U
}
, . (.. g
}
i
(.)) is
a closed embedding.
1.3.8 Euclidean space with two origins. The simplest case is obtained from
open subsets V
}
U
}
, = 1. 2, and a homeomorphism : V
1
V
2
. Then
X = U
1
L
U
2
is obtained from the topological sum U
1
U
2
by identifying
V
1
with () V
2
. Let U
1
= U
2
= R
n
and V
1
= V
2
= R
n
0. Let = id.
Then the graph of in R
n
R
n
is not closed. The resulting locally Euclidean space
is not Hausdorff. If we use (.) = . [.[
-2
, then the result is homeomorphic to
S
n
(see (2.3.2)).
Suppose a space X is the union of subspaces (X
}
[ J). We say X carries
the colimit topology with respect to this family if one of the equivalent statements
hold:
(1) The canonical map

}J
X
}
X (the inclusion on each summand) is a
quotient map.
(2) C is closed in X if and only if X
}
C is closed in X
}
for each .
(3) A set map : X 7 into a space 7 is continuous if and only if the restric-
tions [X
}
: X
}
7 are continuous.
(1.3.9) Example. Let X be a set which is covered by a family (X
}
[ J) of
subsets. Suppose each X
}
carries a topology such that the subspace topologies of
1.4. Compact Spaces 11
X
i
X
}
in X
i
and X
}
coincide and these subspaces are closed. Then there is a
unique topology on X which induces on X
}
the given topology. The space X has
the colimit topology with respect to the X
}
.
Problems
1. Let (X
[ J) be spaces and
non-empty subspaces. Then

J

. The product
is closed if and only if the
are closed.
2. The projections pr
k
:

X
k
are open maps, and in particular quotient maps.
(The X
are non-empty.)
3. A space X is separated if and only if the diagonal D = {(.. .) [ . X] is closed in
XX. Let . g: X Y be continuous maps into a Hausdorff space. Then the coincidence
set = {. [ (.) = g(.)] is closed in X. Hint: Use (1.3.3).
4. A discrete space is the topological sum of its points. There is always a canonical homeo-
morphism X l
(X Y
). For each , Y the map X X Y , . (.. ,)

is an embedding. If : X Y is continuous, then ; : X X Y , . (.. (.)) is an
embedding. If Y is a Hausdorff space, then ; is closed.
1.4 Compact Spaces
A family = (
}
[ J) of subsets of X is a covering of X if X is the union of
the
}
. A covering T = (T
k
[ k 1) of X is a renement of if for each k 1
there exists J such that T
k

}
. If X is a topological space, a covering
= (
}
[ J) is called open (closed) if each
}
is open (closed). A covering
T = (T
k
[ k 1) is a subcovering of if 1 J and T
k
=
k
for k 1.
We say T is nite or countable if 1 is nite or countable. A covering is locally
nite if each . U has a neighbourhood U such that U
}
= 0 only for a nite
number of J. It is called point-nite if each . X is contained only in a nite
number of
}
.
A space X is compact if each open covering has a nite subcovering. (In some
texts this property is called quasi-compact.) By passage to complements we see:
If X is compact, then any family of closed sets with empty intersection contains a
nite family with empty intersection. A set in a space X is relatively compact
if its closure is compact. We recall from calculus the fundamental HeineBorel
Theorem: The unit interval 1 = 0. 1| is compact.
A space X is compact if and only if each net in X has a convergent subnet
(an accumulation value). A discrete closed set in a compact space is nite. Let
X be compact, X closed and : X Y continuous; then and (X) are
compact.
(1.4.1) Proposition. Let T, C be compact subsets of spaces X, Y , respectively.
Let Ube a family of open subsets of X Y which cover T C. Then there exist
open neighbourhoods U of T and V of C such that U V is covered by a nite
subfamily of U. In particular the product of two compact spaces is compact.
One can showthat an arbitrary product of compact spaces is compact (Theorem
of Tychonoff ).
(1.4.2) Proposition. Let T and C be disjoint compact subsets of a Hausdorff
space X. Then T and C have disjoint open neighbourhoods. A compact Hausdorff
space is normal. A compact subset C of a Hausdorff space X is closed.
(1.4.3) Proposition. A continuous map : X Y from a compact space into
a Hausdorff space is closed. If, moreover, is injective (bijective), then is an
embedding (homeomorphism). If is surjective, then it is a quotient map.
(1.4.4) Proposition. Let X be a compact Hausdorff space and : X Y a
quotient map. The following assertions are equivalent:
(1) Y is a Hausdorff space.
(2) is closed.
(3) 1 = {(.
1
. .
2
) [ (.
1
) = (.
2
)] is closed in X X.
Let X be a union of subspaces X
1
X
2
. Recall that X carries the
colimit-topology with respect to the ltration (X
i
) if X is open (closed) if
and only if each intersection X
n
is open (closed) in X
n
. We then call X the
colimit of the ascending sequence (X
i
). (This is a colimit in the categorical sense.)
(1.4.5) Proposition. Suppose X is the colimit of the sequence X
1
X
2
.
Suppose points in X
i
are closed. Then each compact subset 1 of X is contained
in some X
k
.
Aspace is locally compact if eachneighbourhoodof a point . contains a compact
neighbourhood. An open subset of a locally compact space is again locally compact.
Let X be a Hausdorff space and assume that each point has a compact neigh-
bourhood. Let U be a neighbourhood of . and 1 a compact neighbourhood. Since
1 is normal, 1U contains a closed neighbourhood 1of . in 1. Then 1is com-
pact and a neighbourhood of . in X. Therefore X is locally compact. In particular,
a compact Hausdorff space is locally compact. If X and Y are locally compact,
then X Y is locally compact.
Let X be a topological space. An embedding : X Y is a compactication
of X if Y is compact and (X) dense in Y .
The following theoremyields a compactication by a single point. It is called the
Alexandroff compactication or the one-point compactication. The additional
point is the point at innity. In a general compactication : X Y , one calls
the points in Y (X) the points at innity.
1.4. Compact Spaces 13
(1.4.6) Theorem. Let X be a locally compact Hausdorff space. Up to homeomor-
phism, there exists a unique compactication : X Y by a compact Hausdorff
space such that Y (X) consists of a single point.
(1.4.7) Proposition. Let the locally compact space be a union of compact subsets
(1
i
[ i N). Then there exists a sequence (U
i
[ i N) of open subsets with union
X such that each

U
i
is compact and contained in U
i1
.
(1.4.8) Theorem. Let the locally compact Hausdorff space M = 0 be a union of
closed subsets M
n
, n N. Then at least one of the M
n
contains an interior point.
A subset H of a space G is called locally closed, if each . H has a neigh-

bourhood V
x
in G such that H V
x
is closed in G.
(1.4.9) Proposition. (1) Let be locally closed in X. Then = U C with U
open and C closed. Conversely, if X is regular, then an intersection U C, U
open, C closed, is locally closed.
(2) A locally compact set in a Hausdorff space X is locally closed.
(3) A locally closed set in a locally compact space is locally compact.
Problems
1. D
n
,S
n-1
is homeomorphic to S
n
. For the proof verify that
D
n
S
n
. .
_
2
_
1 [.[
2
.. 2[.[
2
1
_
induces a bijection D
n
,S
n-1
S
n
.
2. Let : X C R be continuous. Assume that C is compact and set g(.) =
sup{(.. c) [ c C]. Then g: X R is continuous.
3. Let X be the colimit of an ascending sequence of spaces X
1
X
2
. Then the X
i
are subspaces of X. If X
i
X
i1
is always closed, then the X
are closed in X.
4. Let R
c
be the vector space of all sequences (.
1
. .
2
. . . . ) of real numbers which are
eventually zero. Let R
n
be the subspace of sequences with .
= 0 for > n. Give R

c
the
colimit topology with respect to the subspaces R
n
. Then addition of vectors is a continuous
map R
c
R
c
R
c
. Scalar multiplication is a continuous map RR
c
R
c
. (Thus
R
c
is a topological vector space.) A neighbourhood basis of zero consists of the intersec-
tion of R
c
with products of the form

i_1
| c
i
. c
i
. The space R
c
with this topology
is not metrizable. The space has also the colimit topology with respect to the set of nite-
dimensional linear subspaces. One can also consider the metric topology with respect to the
metric J((.
i
). (,
i
)) =
_
i
(.
i
,
i
)
2
_
1/2
; denote it by R
c
o
. The identity R
c
R
c
o
is
continuous. The space R
c
is separated.
1.5 Proper Maps
A continuous map : X Y is called proper if it is closed and the pre-images
-1
(,). , Y are compact.
(1.5.1) Proposition. Let 1 be compact. Then pr : X 1 X is proper. If
: X Y is proper and 1 Y compact, then
-1
(1) is compact. Let and
g be proper; then g is proper.
As a generalization of the theorem of Tychonoff one can show that an arbitrary
product of proper maps is proper.
(1.5.2) Proposition. Let : X X
t
and g: X
t
X
tt
be continuous.
(1) If and g are proper, then g is proper.
(2) If g is proper and surjective, then g is proper.
(3) If g is proper and g injective, then is proper.
(1.5.3) Proposition. Let : X Y be injective. Then the following are equiva-
lent:
(1) is proper.
(2) is closed.
(3) is a homeomorphism onto a closed subspace.
(1.5.4) Proposition. Let : X Y be continuous.
(1) If is proper, then for each T Y the restriction
B
:
-1
(T) T of
is proper.
(2) Let (U
}
[ J) be a covering of Y such that the canonical map
:

}J
U
}
Y is a quotient map. If each restriction
}
:
-1
(U
}
) U
}
is proper, then is proper.
(1.5.5) Proposition. Let be a continuous map of a Hausdorff space X into a
locally compact Hausdorff space Y . Then is proper if and only if each compact
set 1 Y has a compact pre-image. If is proper, then X is locally compact.
(1.5.6) Proposition. Let : X X
t
and g: X
t
X
tt
be continuous and assume
that g is proper. If X
t
is a Hausdorff space, then is proper.
(1.5.7) Theorem. A continuous map : X Y is proper if and only if for each
space T the product id: X T Y T is closed.
Problems
1. A map : X Y is proper if and only if the following holds: For each net (.
) in X
and each accumulation value , of ((.
)) there exists an accumulation value . of (.
) such
1.6. Paracompact Spaces 15
that (.) = ,.
2. Let X and Y be locally compact Hausdorff spaces, let : X Y be continuous and
: X
the extension to the one-point compactication. Then
is continuous,
if is proper.
3. The restriction of a proper map to a closed subset is proper.
4. Let : X Y be proper and X a Hausdorff space. Then the subspace (X) of Y is a
Hausdorff space.
5. Let : X Y be continuous. Let 1be the equivalence relation on X induced by , and
denote by : X X,1 the quotient map, by h: X,1 (X) the canonical bijection,
and let i : (X) Y . Then = i h is the canonical decomposition of . The map
is proper if and only if is proper, h a homeomorphism, and (X) Y closed.
1.6 Paracompact Spaces
Let A = (U
}
[ J) be an open covering of the space X. An open covering
B = (T
}
[ J) is called a shrinking of A if for each J we have the
inclusion

T
}
U
}
.
A point-nite open covering of a normal space has a shrinking.
A space X is called paracompact if it is a Hausdorff space and if each open
covering has an open, locally nite renement. A closed subset of a paracompact
space is paracompact. A compact space is paracompact.
Aparacompact space is normal. Suppose the locallycompact Hausdorff space X
is a countable unionof compact sets. ThenX is paracompact. Let X be paracompact
and 1 be compact Hausdorff. Then X 1 is paracompact. A metric space is
paracompact.
1.7 Topological Groups
A topological group (G. m. O) consists of a group (G. m) with multiplication
m: G G G, (g. h) m(g. h) = gh and a topology O on G such that
the multiplication m and the inverse | : G G, g g
-1
are continuous. We de-
note a topological group (G. m. O) usually just by the letter G. The neutral element
will be denoted by e (also 1 is in use and 0 for abelian groups). The left translation
l
: G G, . g. by g G in a topological group is continuous, and the rules

l
l
h
= l
h
and l
e
= id show it to be a homeomorphism. For subsets and T of a
group G we use notations like aT = {ab [ b T], T = {ab [ a . b T],
2
= ,
-1
= {a
-1
[ a ], and similar ones.
A group G together with the discrete topology on the set G is a topological
group, called a discrete (topological) group.
The additive groups of the real numbers R, complex numbers C, and quaternions
Hwith their ordinary topology are topological groups, similarly the multiplicative
groups R
+
, C
+
and H
+
of the non-zero elements. The multiplicative group R
+
of the positive real numbers is an open subgroup of R

+
and a topological group.
The complex numbers of norm 1 are a compact topological group S
1
with respect
to multiplication. The exponential function exp: R R
+
is a continuous ho-
momorphism with the logarithm function as a continuous inverse. The complex
exponential function exp: C C
+
is a surjective homomorphism with kernel
{2i n [ n Z], a discrete subgroup of C.
The main examples of topological groups are matrix groups. In the vector
space M
n
(R) of real (n. n)-matrices let GL
n
(R) be the subspace of the invertible
matrices. Since the determinant is a continuous map, this is an open subspace.
Matrix multiplication and passage to the inverse are continuous, since they are
given by rational functions in the matrix entries. This makes the general linear
group GL
n
(R) into a topological group. Similarly for GL
n
(C). The determinant
is a continuous homomorphism det : GL
n
(R) R
+
with kernel the special linear
group SL
n
(R); similarly in the complex case.
Let O(n) = { M
n
(R) [
t
= 1] be the group of orthogonal (n. n)-
matrices (
t
transpose of ; 1 unit matrix). The set O(n) is a compact subset in
M
n
(R). Hence O(n) is a compact topological group (the orthogonal group). The
open and closed subspace SO(n) = { O(n) [ det() = 1] of O(n) is the special
orthogonal group. Similarly the subgroup U(n) = { M
n
(C) [
t

= 1]
of unitary (n. n)-matrices is a compact topological group (unitary group). The
topological groups SO(2), U(1), and S
1
are isomorphic. The special unitary group
SU(n) is the compact subgroup of U(n) of matrices with determinant 1. The
multiplicative group of quaternions of norm 1 provides S
3
with the structure of
a topological group. This group is isomorphic to SU(2). From linear algebra
one knows about a surjective homomorphism SU(2) SO(3) with kernel 1
(a twofold covering); for this and other related facts see the nice discussion in
[27, Kapitel IX]. For more information about matrix groups, also from the view-
point of manifolds and Lie groups, see [29]; there you can nd, among others,
the symplectic groups Sp(n) and the Spinor groups Spin(n). The isomorphisms
SU(2) Spin(3) Sp(1) hold, and these spaces are homeomorphic to S
3
.
If G and H are topological groups, then the direct product G H with the
product topology is a topological group. The n-fold product S
1
S
1
is called
an n-dimensional torus.
The trivial subgroup is often denoted by 1 (in a multiplicative notation) or by 0
(in an additive notation). The neutral element will also be denoted 1 or 0. The
symbol H C G is used for a normal subgroup H of G. The notation H ~ 1 or
H ~
G
1 means that H and 1 are conjugate subgroups of G.
A homomorphism : G H between topological groups is continuous if it
is continuous at the neutral element e.
If G is a topological group and H G a subgroup, then H, with the subspace
topology, is a topological group (called a topological subgroup). If H G is a
1.8. Transformation Groups 17
subgroup, then the closure of H is also. If H is a normal subgroup, then

H is also.
1.8 Transformation Groups
A left action of a topological group G on a topological space X is a continuous
map j: G X X, (g. .) g. such that g(h.) = (gh). and e. = . for
g. h G, e G the unit, and . X. A (left) G-space (X. j) consists of a space
X and a left action j of G on X. The homeomorphism l
: X X, . g. is
called left translation by g. We also use right actions X G X, (.. g) .g;
they satisfy (.h)g = .(hg) and .e = .. For X and 1 G we let
1 = {ka [ k 1. a ]. An action is effective if g. = . for all . X implies
g = e. The trivial action has g. = . for g G and . X.
The set 1 = {(.. g.) [ . X. g G] is an equivalence relation on X.
The set of equivalence classes X mod 1 is denoted by X,G. The quotient map
q : X X,G is used to provide X,G with the quotient topology. The resulting
space X,G is called the orbit space of the G-space X. A more systematic notation
for the orbit space of a left action would be G\X. The equivalence class of . X
is the orbit G. through .. An action is transitive if it consists of a single orbit.
The set G
x
= {g G [ g. = .] is a subgroup of G, the isotropy group or
the stabilizer of the G-space X at .. An action is free if all isotropy groups are
trivial. We have G
x
= gG
x
g
-1
. Therefore the set Iso(X) of isotropy groups of X
consists of complete conjugacy classes of subgroups. If it contains a nite number
of conjugacy classes, we say X has nite orbit type.
A subset of a G-space is called G-stable or G-invariant if g G and a
implies ga . A G-stable subset is also called a G-subspace. For each
subgroup H of G there is an H-xed point set of X,
X
1
= {. X [ h. = . for all h H].
Suppose X and Y are G-spaces. A map : X Y is called a G-map or
a G-equivariant map if for g G and . X the relation (g.) = g(.)
holds. In general, the term equivariant refers to something related to a group
action. Left G-spaces and G-equivariant maps form the category G- TOP. This
category has products: If (X
}
[ J) is a family of G-spaces, then the topological
product

}
X
}
together with the diagonal action (g. (.
}
)) (g.
}
) is a product
in this category. A G-map : X Y induces by passage to the orbit spaces a
map ,G: X,G Y,G. We have the notion of an equivariant homotopy or
G-homotopy H
t
: this is a homotopy such that each H
t
is a G-map.
(1.8.1) Proposition. Let X be a G-space, G and T X. If T is open then
T is open. The orbit map : X X,G is open.
Proof. l
o
(T) is open, since l
o
is a homeomorphism. Hence
_
o
l
o
(T) = T
is a union of open sets. Let U be open. Then
-1
(U) =
_
G
l
(U) is open,
hence (U) is open, by denition of the quotient topology.
(1.8.2) Proposition. (1) Let H be a subgroup of the topological group G. Let the
set G,H of cosets gH carry the quotient topology with respect to : G G,H,
g gH. Then l : G G,H G,H, (.. gH) .gH is a continuous action.
(2) G,H is separated if and only if H is closed in G. In particular, G is
separated if {e] is closed.
(3) Let H be normal in G. Then the factor group G,H with quotient topology
is a topological group.
A space G,H with the G-action by left multiplication is called a homogeneous
space. The space of left cosets Hg is H\G; it carries a right action.
(1.8.3) Example. Homogeneous spaces are important spaces in geometry.
The orthogonal group O(n 1) acts on the sphere S
n
by matrix multiplication
(. ) . The action is transitive. The isotropy group of e
1
= (1. 0. . . . . 0)
is O(n), here considered as the block matrices
_
1 0
0 B
_
with T O(n). We obtain
a homeomorphism of O(n 1)-spaces O(n 1),O(n) S
n
. In the complex
case we obtain a homeomorphism U(n 1),U(n) S
2n1
, in the quaternionic
case a homeomorphism Sp(n 1),Sp(n) S
4n3
. Other important homoge-
neous spaces are the projective spaces, the Grassmann manifolds, and the Stiefel
manifolds to be discussed later.
(1.8.4) Proposition. (1) If . X is closed, then G
x
is closed in G.
(2) If X is a Hausdorff space, then X
1
is closed.
(3) Let be a G-stable subset of the G-space X. Then ,G carries the
subspace topology of X,G. In particular X
G
X X,G is an embedding.
(4) Let T X be closed and X. Then {g G [ g T] is closed in G.
(5) Let T X be closed. Then {g G [ gT = T] is closed.
Proof. (1) The isotropy group G
x
is the pre-image of . under the continuous map
G X, g g.. (2) The set X
= {. X [ g. = .] is the pre-image of
the diagonal under X X X, . (.. g.), and X
1
=
_
1
X
. (3) Let
C ,G be open with respect to the quotient map ,G. Then
-1
(C)
is open, and we can write
-1
(C) = U with an open subset U X. We
have U = GU, since is G-stable. We conclude C = (
-1
C) =
,G(GU). Since GU is open, (GU) is open, hence C is open in the subspace
topology. By continuity of ,G X,G, an open subset in the subspace topology
is open in ,G. (4) r
o
: G X, g ga is continuous, hence r
-1
o
(T) = {g
G [ ga T] closed and therefore
_
o
r
-1
o
(T) = {g G [ g T] closed. (5)
The set {g [ gT = T] = {g [ gT T] {g [ g
-1
T T] is closed, by (4).
(1.8.5) Proposition. Let r : G X X be a G-action, G and T X.
1.8. Transformation Groups 19
(1) If and T are compact, then T is compact.
(2) If is compact, then the restriction X X of r is proper. If, moreover,
T is closed, then T is closed.
(3) If G is compact, then the orbit map is proper. Thus X is compact if and
only if X,G is compact.
(4) If G is compact and X separated, then X,G is separated.
(5) Let G be compact, a G-stable closed subset and U a neighbourhood of
in X. Then U contains a G-stable neighbourhood of .
Proof. (1) T G X is compact as a product of compact spaces. Hence
the continuous image T of T under r : G X X is compact. (2) The
homeomorphismX X, (s. .) (s. s.) transforms r into the projection
pr : X X. The projection is proper, since is compact (see (1.5.1)). Hence
the image T of the closed set T is closed. (3) Let X be closed.
Then
-1
() = G is closed, by (2). Hence () is closed, by denition of
the quotient topology. The pre-images of points are orbits; they are compact as
continuous images of G. (4) Since is proper, so is . Hence the image of
the diagonal under is closed. (5) Let U be open. Then (X U) is disjoint
to (). By (4), X
-1
(X U) is open and a G-stable neighbourhood of
contained in U.
The orbit category Or(G) is the category of homogeneous G-spaces G,H, H
closed in G, and G-maps. There exists a G-map G,H G,1 if and only if
H is conjugate to a subgroup of 1. If a
-1
Ha _ 1, then 1
o
: G,H G,1,
gH ga1 is a G-map and each G-map G,H G,1 has this form; moreover
1
o
= 1
b
if and only if a
-1
b 1.
An action GV V on a real (or complex) vector space V is called a real (or
complex) representation of G if the left translations are linear maps. After choice of
a basis, a representation amounts to a continuous homomorphismfromGto GL
n
(R)
or GL
n
(C). A homomorphism G O(n) or G U(n) is called an orthogonal
or unitary representation. Geometrically, an orthogonal representation is given by
an action G V V with an invariant scalar product (. ). The latter means
(g. gn) = (. n) for g G and . n V . In an orthogonal representation, the
unit sphere S(V ) = { V [ (. ) = 1] is G-stable.
Let 1 be a right G-space and J a left G-space. We denote by 1
G
J the
orbit space of the G-action G (1 J) 1 J, (g. (.. ,)) (.g
-1
. g,). A
G-map : J
1
J
2
induces a continuous map
id
G
: 1
G
J
1
1
G
J
2
. (.. ,) (.. (.)).
If 1 carries a left 1-action which commutes with the right G-action (i.e., k(.g) =
(k.)g), then 1
G
J carries an induced 1-action (k. (.. ,)) (k.. ,). This
construction can in particular be applied in the case that 1 = 1, G a subgroup
of 1 and the G- and 1-actions on 1 are given by right and left multiplication.
The assignments J 1
G
J and id
G
yield the induction functor
ind
1
G
: G- TOP 1- TOP. This functor is left adjoint to the restriction functor
res
1
G
: 1- TOP G- TOP which is obtained by regarding a 1-space as a G-space.
The natural adjunction
TOP
1
(ind
1
G
X. Y ) TOP
G
(X. res
1
G
Y )
sends a G-map : X Y to the 1-map (k. .) k(.); in the other direction
one restricts a map to X G
G
X 1
G
X. (Here TOP
1
denotes the set of
1-equivariant maps.)
(1.8.6) Theorem. Suppose the Hausdorff groupGis locally compact withcountable
basis. Let X be a locally compact Hausdorff space and G X X a transitive
action. Then for each . X the map b : G X, g g. is open and the induced
map

b : G,G
x
X a homeomorphism.
Proof. If b is open, then

b is a homeomorphism. Let W be a neighbourhood of e,
(T
i
[ i N) a countable basis, and g
-1
i
T
i
. For each g G there exists a
such that T
}
Wg
-1
, g g
}
W. Therefore the g
}
W cover the group.
Let V G be open and g V . There exists a compact neighbourhood W
of e such that W = W
-1
and gW
2
V . Since G is the union of the g
}
W and
the action is transitive, X =
_
g
}
W.. Since W is compact and b continuous,
g
}
W. is compact and hence closed in X. By (1.4.8), there exist such that g
}
W.
contains an interior point, and therefore W. contains an interior point n.. Then .
is an interior point of n
-1
W. W
2
. and hence g. = (g) an interior point of
gW
2
. V. = (V ). This shows that is open.
(1.8.7) Corollary. Let the locally compact Hausdorff group G with countable basis
act on a locally compact Hausdorff space X. An orbit is locally compact if and only
if it is locally closed. An orbit is a homogeneous space with respect to the isotropy
group of each of its points if and only if it is locally closed.
Problems
1. Let H be a normal subgroup of G and X a G-space. Restricting the group action to H,
we obtain an H-space X. The orbit space H\X carries then an induced G,H-action.
2. Let a pushout in TOP be given with G-spaces , T, X and G-maps , :
T
J
X
1
Y .
Let G be locally compact. Then there exists a unique G action on Y such that J. J become
G-maps. The diagram is then a pushout in G-TOP. Hint: (2.4.6)
1.9. Projective Spaces. Grassmann Manifolds 21
3. Let Y be a 1-space and G a subgroupp of 1. Then 1
G
Y 1,G Y , (k. ,)
(kG. k,) is a 1-homeomorphism. If X is a G-space, then
1
G
(X Y ) (1
G
X) Y. (g. (.. ,)) ((g. .). g,)
is a 1-homeomorphism.
4. Let H be a closed subgroup of G. Then G,H is a Hausdorff space and therefore J =
G,H
!
is closed. The relation gH G,H
!
is equivalent to g
-1
Hg H. Hence
{g G [ g
-1
Hg H] is closed in G. The normalizer NH = {g G [ g
-1
Hg = H] of
H in G is closed in G. The group H is a normal subgroup of NH and NH,H = W
G
H =
WH is the Weyl group of H in G. The group NH always acts on the xed set X
!
, by
restricting the given G-action to NH. The action
G,H WH G,H. (gH. nH) gnH
is a free right action by G-automorphisms of G,H.
1.9 Projective Spaces. Grassmann Manifolds
Let 1(R
n1
) = R1
n
be the set of one-dimensional subspaces of the vector space
R
n1
. A one-dimensional subspace of V is spanned by . V 0. The vectors .
and , span the same subspace if and only if . = z, for some z R
+
= R 0.
We therefore consider 1(R
n1
) as the orbit space of the action
R
+
(R
n1
0) R
n1
0. (z. .) z..
The quotient map : R
n1
0 1(R
n1
) provides 1(R
n1
) with the quotient
topology. The space R1
n
is the n-dimensional real projective space. We set
(.
0
. . . . . .
n
) = .
0
. . . . . .
n
| and call .
0
. . . . . .
n
the homogeneous coordinates
of the point .
0
. . . . . .
n
|.
In a similar manner we consider the set 1(C
n1
) = C1
n
of one-dimensional
subspaces of C
n1
as the orbit space of the action
C
+
(C
n1
0) C
n1
0. (z. :) z:.
We have again a quotient map : C
n1
0 1(C
n1
). The space C1
n
is called
the n-dimensional complex projective space. (It is 2n-dimensional as a manifold.)
We describe the projective spaces in a different manner as orbit spaces. The
subgroup G = {1] R
+
acts on S
n
R
n1
by (z. .) z., called the
antipodal involution. The inclusion i : S
n
R
n1
induces a continuous bijective
map | : S
n
,G (R
n1
0),R
+
. The map : R
n1
0 S
n
, . [.[
-1
.
induces an inverse. The quotient S
n
,Gis compact, since S
n
is compact. By (1.4.4),
the quotient is a Hausdorff space. In a similar manner one treats C1
n
, but now
with respect to the action S
1
S
2n1
S
2n1
, (z. :) z: of S
1
on the unit
sphere S
2n1
C
n1
0.
Projective spaces are homogeneous spaces. Consider the action of O(n1) on
R
n1
by matrix multiplication. If V 1(R
n1
) is a one-dimensional space and
O(n 1), then V 1(R
n1
). We obtain an induced action
O(n 1) 1(R
n1
) 1(R
n1
).
This action is transitive. The isotropy group of 1. 0. . . . . 0| consists of the matrices
_
z 0
0 T
_
. z O(1). T O(n).
We consider these matrices as the subgroupO(1)O(n) of O(n1). The assignment
e
1
induces an O(n 1)-equivariant homeomorphism
b : O(n 1),(O(1) O(n)) 1(R
n1
).
The action of O(n 1) on 1(R
n1
) is continuous; this follows easily from the
continuity of the action O(n 1) (R
n1
0) R
n1
0 and the denition
of the quotient topology. Therefore b is a bijective continuous map of a compact
space into a Hausdorff space. In a similar manner we obtain a U(n1)-equivariant
homeomorphism U(n 1),(U(1) U(n)) 1(C
n1
).
Finally, one can dene the quaternionic projective space H1
n
in a similar
manner as a quotient of H
n1
0 or as a quotient of S
4n3
.
We generalize projective spaces. Let W be an n-dimensional real vector space.
We denote by G
k
(W) the set of k-dimensional subspaces of W. We dene a
topology on G
k
(W). Suppose W carries an inner product. Let V
k
(W) denote
the set of orthonormal sequences (n
1
. . . . . n
k
) in W considered as a subspace
of W
k
. We call V
k
(W) the Stiefel manifold of orthonormal k-frames in W. We
have a projection : V
k
(W) G
k
(W) which sends (n
1
. . . . . n
k
) to the subspace
n
1
. . . . . n
k
| spanned by this sequence. We give G
k
(W) the quotient topology
determined by . The space G
k
(W) can be obtained as a homogeneous space. Let
W = R
n
with standard inner product and standard basis e
1
. . . . . e
n
. We have a
continuous action of O(n) on V
k
(R
n
) and G
k
(R
n
) dened by (. (
1
. . . . .
k
))
(
1
. . . . .
k
) and such that becomes O(n)-equivariant. The isotropy groups
of (e
1
. . . . . e
k
) and e
1
. . . . . e
k
| consist of the matrices
_
1
k
0
0 T
_
.
_
0
0 T
_
. O(k). T O(n k)
respectively. The map (e
1
. . . . . e
k
) induces equivariant homeomorphisms
in the diagram
O(n),O(n k)

V
k
(R
n
)
O(n),(O(k) O(n k))

G
k
(R
n
).
1.9. Projective Spaces. Grassmann Manifolds 23
This shows that G
k
(R
n
) is a compact Hausdorff space. It is called the Grassmann
manifold of k-dimensional subspaces of R
n
.
In a similar manner we can work with the k-dimensional complex subspaces of
C
n
and obtain an analogous diagram of U(n)-spaces (complex Stiefel and Grass-
mann manifolds):
U(n),U(n k)

V
k
(C
n
)
U(n),(U(k) U(n k))

G
k
(C
n
).
Chapter 2
The Fundamental Group
In this chapter we introduce the homotopy notion and the rst of a series of algebraic
invariants associated to a topological space: the fundamental group.
Almost every topic of algebraic topology uses the homotopy notion. Therefore it
is necessary to begin with this notion. Ahomotopy is a continuous family h
t
: X
Y of continuous maps which depends on a real parameter t 0. 1|. (One may
interpret this as a time-dependent process.) The maps
0
and
1
are then called
homotopic, and being homotopic is an equivalence relation on the set of continuous
maps X Y . This equivalence relation leads to a quotient category of the category
TOP of topological spaces and continuous maps, the homotopy category h-TOP.
The importance of this notion is seen from several facts.
(1) The classical tools of algebraic topology are functors from a category of
spaces to an algebraic category, say of abelian groups. These functors turn
out to be homotopy invariant, i.e., homotopic maps have the same value under
the functor.
(2) One can change maps by homotopies and spaces by homotopy equivalences.
This fact allows for a great exibility. But still global geometric information
is retained. Basic principles of topology are deformation and approximation.
One idea of deformation is made precise by the notion of homotopy. Conti-
nuity is an ungeometric notion. So often one has to deform a continuous map
into a map with better properties.
(3) The homotopy notion leads in an almost tautological way to algebraic struc-
tures and categorical structures. In this chapter we learn about the simplest
example, the fundamental group and the fundamental groupoid.
The passage to the homotopy category is not always a suitable view-point. In gen-
eral it is better to stay in the category TOP of topological spaces and continuous
maps (space level as opposed to homotopy level). We thus consider homotopy
as an additional structure. Then classical concepts can be generalized by using the
homotopy notion. For instance one considers diagrams which are only commu-
tative up to homotopy and the homotopies involved will be treated as additional
information. One can also dene generalized group objects where multiplication
is only associative up to homotopy. And so on.
The passage from TOP to h-TOP may be interpreted as a passage from contin-
uous mathematics to discrete mathematics.
The homotopy notion allows us to apply algebraic concepts to continuous maps.
It is not very sensible to talk about the kernel or cokernel of a continuous map.
2.1. The Notion of Homotopy 25
But we will see later that there exist notions of homotopy-kernels (then called
homotopy bres) and homotopy-cokernels (then called homotopy cobres). This
is the more modern view-point of a large variety of homotopy constructions. In
general terms: The idea is to replace the categorical notions limit and colimit by
appropriate homotopy notions.
The prototype of a functor from spaces to groups is the fundamental group
functor. Historically it is the rst of such functors. It was introduced by Poincar,
in different context and terminology. In general it is difcult to determine the
fundamental group of a space. Usually one builds up a space from simpler pieces
and then one studies the interrelation between the groups of the pieces. This uses the
functorial aspect and asks for formal properties of the functor. We prove the basic
theorem of Seifert and van Kampen which roughly says that the functor transforms
suitable pushouts of spaces into pushouts of groups. This may not be the type of
algebra the reader is used to, and it can in fact be quite complicated. We describe
some relatedalgebra (presentationof groups bygenerators andrelations) anddiscuss
a number of geometric results which seem plausible from our intuition but which
cannot be proved (in a systematic way) without algebraic topology. The results are
of the type that two given spaces are not homeomorphic and this follows, if their
fundamental groups are different. Finally we show that each group can be realized
as a fundamental group (this is the origin of the idea to apply topology to group
theory).
The study of the fundamental group can be continued with the covering space
theory where the fundamental group is exhibited as a symmetry group. This sym-
metry inuences almost every other tool of algebraic topology (although we do not
always carry out this inuence in this text).
The chapter contains two sections on point-set topology. We discuss standard
spaces like spheres, disks, cells, simplices; they will be used in many different
contexts. We present the compact-open topology on spaces of continuous maps;
they will be used for the dual denition of homotopy as a continuous family of
paths, and this duality will henceforth be applied to many homotopy constructions
and notions.
2.1 The Notion of Homotopy
A path in a topological space X from . to , is a continuous map u: a. b| X
such that u(a) = . and u(b) = ,. We say that the path connects the points u(a)
and u(b). We can reparametrize and use the unit interval as a source 0. 1| X,
t u((1 t )a t b). In the general theory we mostly use the unit interval. If
u: 0. 1| X is a path from . to ,, then the inverse path u
-
: t n(1 t ) is a
path from , to .. If : 0. 1| X is another path from , to :, then the product
path u + , dened by t u(2t ) for t _ 1,2 and (2t 1) for t _ 1,2, is a path
from . to :. We also have the constant path k
x
with value ..
26 Chapter 2. The Fundamental Group
From these remarks we see that being connectible by paths is an equivalence
relation on X. An equivalence class is called a path component of X. We denote
by
0
(X) the set of path components and by .| the path component of the point ..
Aspace X is said to be path connected or 0-connected if it has one of the following
equivalent properties:
(1)
0
(X) consists of a single element.
(2) Any two points can be joined by a path.
(3) Any continuous map : d1 = {0. 1] X has a continuous extension
J : 1 X.
(Later we study the higher dimensional analogous problemof extending maps from
the boundary d1
n
of the n-dimensional cube to 1
n
.)
A map : X Y induces
0
( ):
0
(X)
0
(Y ), .| (.)|. In this
way
0
becomes a functor from the category TOP of topological spaces to the
category SET of sets
1
. We will see that this functor is the beginning of algebraic
topology, although there is no algebra yet.
Thinking in terms of categories and functors is a basic method in (algebraic)
topology. The size of
0
(X) is a topological property of the space X. A functor
transports isomorphisms to isomorphisms. Thus a homeomorphism induces
a bijection
0
( ). Suppose : X Y is a homeomorphism; then induces a
homeomorphismX Y () for each subset X. Suppose : R R
n
is a homeomorphism; the space R. has two path components (intermediate value
theorem of calculus), and R
n
, is path connected for n > 1; we apply the functor
0
and conclude that R is not homeomorphic to R
n
for n > 1. This example seems
almost trivial, but the reasoning is typical. Here is another simple example of this
type: The subspace X = R 0 L 0 R of R
2
is not homeomorphic to R since
X contains a point . = (0. 0) such that
0
(X .) has four elements whereas
0
(R ,) has always two elements.
2.1.1 Pathcategories. Forming the product path is not an associative composition.
We can remedy this defect by using parameter intervals of different length. So let
us consider paths of the form u: 0. a| X, : 0. b| X with u(a) = (0)
and a. b _ 0. Their composition u = n is the path 0. a b| X with
n(t ) = u(t ) for 0 _ t _ a and n(t ) = (a t ) for a _ t _ a b. In this
manner we obtain a category W(X): Objects are the points of X; a morphism from
. to , is a path u: 0. a| X with u(0) = .. u(a) = , for some a _ 0; and
composition of morphisms is as dened before; the path 0. 0| X with value .
is the identity of the object .. A continuous map : X Y induces a functor
W( ): W(X) W(Y ), . (.), u u.
1
Our general conventions: space = topological space, map = continuous map. A set map between
spaces is a map which is not assumed to be continuous at the outset.
Aspace is connected if it is not the topological sumof two non-empty subspaces.
Thus X is disconnected if and only if X contains a subset X which is open, closed,
and different from 0 and X. A decomposition of X is a pair U. V of open, non-
empty, disjoint subsets with union X. A space X is disconnected if and only if
there exists a continuous surjective map : X {0. 1]; a decomposition is given
by U =
-1
(0), V =
-1
(1). The continuous image of a connected space
is connected. Recall from calculus: R is connected if and only if is an
interval. (An interval is a subset which contains with .. , also .. ,|.)
(2.1.2) Proposition. Let (
}
[ J) be a family of connected subsets of X such
that
i

}
,= 0 for all i. . Then
_
}

}
= Y is connected. Let be connected
and T

. Then T is connected.
The union of the connected sets in X which contain . is thus a closed connected
subset. We call it the component X(.) of . in X. If , X(.), then X(,) = X(.).
A component of X is a maximal connected subset. Any space is the disjoint union
of its components. A space is totally disconnected if its components consist of
single points. Since intervals are connected a path connected space is connected.
A product
}
X
}
is connected if each X
}
is connected. The component of
(.
}
)
}
X
}
is the product of the components of the .
}
.
(2.1.3) Example. The space
X = 1. 0| 0 L 0 1. 1| L {(.. sin(.
-1
) [ 0 < . _ 1]
is connected but not path connected. The union S of X with {1] 2. 0| L
2. 2| {2] is called the pseudo-circle. The complement R
2
S has two path
components.
A pseudo-circle S has a sequence 1
1
1
2
of compact neighbourhoods
with
_
i
1
i
= S and 1
i
homeomorphic to S
1
0. 1|.
Let X and Y be topological spaces and . g: X Y continuous maps. A
homotopy from to g is a continuous map
H: X 0. 1| Y. (.. t ) H(.. t ) = H
t
(.)
such that (.) = H(.. 0) and g(.) = H(.. 1) for . X, i.e., = H
0
and
g = H
1
. We denote this situation by H: . g. One can consider a homotopy as
a dynamical process, the parameter t is the time and H
t
is a time-dependent family
of maps. One also says that is deformed continuously into g. Another (dual)
view-point of a homotopy is: a parametrized family of paths. We use the letter 1 for
the unit interval 0. 1|. If we write a homotopy in the form H
t
, we understand that
H: X 1 Y , (.. t ) H
t
(.) is continuous in both variables simultaneously.
We call and g homotopic if there exists a homotopy from to g. (One can, of
course, dene homotopies with 0. 1| X. While this does not affect the theory, it
does make a difference when orientations play a role.)
The homotopy relation . is an equivalence relation on the set of continuous
maps X Y . Given H: . g, the inverse homotopy H
-
: (.. t ) H(.. 1t )
shows g . . Let 1: . g and 1: g . h be given. The product homotopy
1 + 1 is dened by
(1 + 1)(.. t ) =
_
1(.. 2t ). 0 _ t _
1
2
.
1(.. 2t 1).
1
2
_ t _ 1.
and shows . h. The constant homotopy H(.. t ) = (.) shows . .
The equivalence class of is denoted | and called the homotopy class of .
We denote by X. Y | the set of homotopy classes | of maps : X Y . A
homotopy H
t
: X Y is said to be relative to X if the restriction H
t
[ does
not depend on t (is constant on ). We use the notation H: . g (rel ) in this
case.
The homotopyrelationis compatible withthe compositionof maps: Let H: .
g: X Y and G: k . l : Y 7 be given; then
(.. t ) G(H(.. t ). t ) = G
t
H
t
(.)
is a homotopy fromk to lg. We see that topological spaces and homotopy classes
of maps form a quotient category of TOP, the homotopy category h-TOP, when
composition of homotopy classes is induced by composition of representing maps.
If : X Y represents an isomorphism in h-TOP, then is called a homotopy
equivalence or h-equivalence. In explicit terms this means: : X Y is a
homotopy equivalence if there exists g: Y X, a homotopy inverse of , such
that g and g are both homotopic to the identity. Spaces X and Y are homotopy
equivalent or of the same homotopy type if there exists a homotopy equivalence
X Y . A space is contractible if it is homotopy equivalent to a point. A map
: X Y is null homotopic if it is homotopic to a constant map; a null homotopy
of is a homotopy between and a constant map. Anull homotopy of the identity
id(X) is a contraction of the space X.
2.1.4 Categories of homotopies. We generalize (2.1.1) and dene a category
W(X. Y ). The objects are the continuous maps : X Y . A morphism from
to g is a homotopy H: X 0. a| Y with H
0
= and H
o
= g. Composition
is dened as in (2.1.1).
As in any category we also have the Hom-functors in h-TOP. Given : X Y ,
we use the notation
+
: 7. X| 7. Y |. g g.
+
: Y. 7| X. 7|. h h
for the induced maps
2
. The reader should recall a little reasoning with Hom-
functors, as follows. The map is an h-equivalence, i.e., an isomorphism in
h-TOP if and only if
+
is always bijective; similarly for
+
. If : X Y
has a right homotopy inverse h: Y X, i.e., h . id, and a left homotopy
inverse g: Y X, i.e., g . id, then is an h-equivalence. If two of the maps
: X Y , g: Y 7, and g are h-equivalences, then so is the third.
Homotopy is compatible with sums and products. Let
i
:

}J
X
}
X
i
be
the projection onto the i -th factor. Then
Y.
}J
X
}
|
}J
Y. X
}
|. | (
i
|)
is a well-dened bijection. Let i
k
: X
k

}J
X
}
be the canonical inclusion of
the k-th summand. Then
}J
X
}
. Y |
}J
X
}
. Y |. | ( i
k
|)
is a well-dened bijection. In other words: sum and product in TOP also repre-
sent sum and product in h-TOP. (Problems arise when it comes to pullbacks and
pushouts.)
Let 1 be a point. Amap 1 Y can be identied with its image and a homotopy
1 1 Y can be identied with a path. The Hom-functor 1. | is therefore
essentially the same thing as the functor
0
.
2.1.5 Linear homotopy. Given maps . g: X , R
n
. Suppose that
the line-segment from (.) to g(.) is always contained in . Then H(.. t ) =
(1 t )(.) tg(.) is a homotopy from to g (linear homotopy). It will turn out
that many homotopies are constructed from linear homotopies.
A set R
n
is star-shaped with respect to a
0
if for each a the
line-segment from a
0
to a is contained in . If is star-shaped, then H(a. t ) =
(1 t )a t a
0
is a null homotopy of the identity. Hence star-shaped sets are
contractible. A set C R
n
is convex if and only if it is star-shaped with respect to
each of its points.
Note: If = R
n
and a
0
= 0, then each H
t
, t < 1, is a homeomorphism, and
only in the very last moment is H
1
constant! This is less mysterious, if we look at
the paths t H(.. t ).
The reader should now recall the notion of a quotient map (identication), its
universal property, and the fact that the product of a quotient map by the identity of
a locally compact space is again a quotient map (see (2.4.6)).
(2.1.6) Proposition. Let : X Y be a quotient map. Suppose H
t
: Y 7 is
a family of set maps such that H
t
is a homotopy. Then H
t
is a homotopy.
2
As a general principle we use a lower indexfor covariant functors andanupper indexfor contravariant
functors. If we applya (covariant) functor toa morphism( we oftencall the result the inducedmorphism
and denote it simply by (
if the functor is clear from the context.

Proof. The product id: X1 Y 1 is an identication, since 1 is compact.
The composition H ( id) is continuous and therefore H is continuous.
(2.1.7) Proposition. Let H
t
: X X be a homotopy of the identity H
0
= id(X)
such that the subspace 0 ,= X is always mapped into itself, H
t
() .
Suppose H
1
is constant on . Then the projection : X X, ( identied to
a point) is an h-equivalence.
Proof. Since H
1
() is a point, there exists a map q : X, X such that
q = H
1
. By assumption, this composition is homotopic to the identity. The
map H
t
: X X, factorizes over and yields 1
t
: X, X, such that
1
t
= H
t
. By (2.1.6), 1
t
is a homotopy, 1
0
= id and 1
1
= q.
Problems
1. Suppose there exists a homeomorphism R X Y . Then X or Y is a point.
2. Let : X Y be surjective. If X is (path) connected, then Y is (path) connected.
3. Let C be a countable subset of R
n
, n _ 2. Show that R
n
C is path connected.
4. The unitary group U(n) and the general linear group GL
n
(C) are path connected. The
orthogonal group O(n) and the general linear group GL
n
(R) have two path components; one
of them consists of matrices with positive determinant.
5. Let U R
n
be open. The path components of U are open and coincide with the
components. The set of path components is nite or countably innite. An open subset of R
is a disjoint union of open intervals.
6. List theorems of point-set topology which showthat the product homotopy and the inverse
homotopy are continuous. Do the same for the linear homotopy in 2.1.5.
7. A space X is contractible if and only if the identity id(X) is null homotopic.
8. g is null homotopic, if or g is null homotopic.
9. Let be contractible. Then any two maps X are homotopic.
10. The inclusions O(n) GL
n
(R) and U(n) GL
n
(C) are homotopy equivalences.
Let 1(n) denote the space of positive denite real (n. n)-matrices. Then O(n) 1(n)
GL
n
(R), (X. 1) X1 is a homeomorphism; 1(n) is star-like with respect to the unit
matrix.
11. There exist contractible and non-contractible spaces consisting of two points.
2.2 Further Homotopy Notions
The homotopy notion can be adapted to a variety of other contexts and categories:
Consider homotopies which preserve some additional structure of a category. We
describe some examples from which the general idea emerges. This section only
contains terminology.
The construction of group structures on homotopy sets uses the category of
pointed spaces, as we will see shortly. We call a pair (X. .
0
) consisting of a space
2.2. Further Homotopy Notions 31
X and a base point .
0
X a pointed space. A pointed map : (X. .
0
) (Y. ,
0
)
is a continuous map : X Y which sends the base point to the base point. A
homotopy H: X 1 Y is pointed if H
t
is pointed for each t 1. We de-
note by X. Y |
0
the set of pointed homotopy classes (xed base points assumed) or
by (X. .
0
). (Y. ,
0
)|. We obtain related notions: pointed homotopy equivalence,
pointed contractible, pointed null homotopy. We denote the category of pointed
spaces and pointed maps by TOP
0
, and by h-TOP
0
the associated homotopy cat-
egory. Often a base point will just be denoted by +. Also a set with a single
element will be denoted by its element. The choice of a base point is an addi-
tional structure. There is a functor from TOP to TOP
0
which sends a space X to
X
= X {+], i.e., to X with an additional base point added (topological sum),

with the obvious extension to pointed maps. This functor is compatible with homo-
topies. We also have the forgetful functor from TOP
0
to TOP. They are adjoint
TOP
0
((X). Y ) TOP(X. Y ), and similarly for the homotopy categories.
The category TOP
0
has sums and products. Suppose (X
}
. .
}
) is a family of
pointed spaces. The family (.
}
) of base points is taken as base point in the product
}
X
}
; this yields the pointed product. Let
_
}J
X
}
be the quotient of
}J
X
}
where all base points are identied to a single new base point. We have canonical
pointed maps i
k
: X
k

_
}
X
}
which arise from the canonical inclusions X
k

}
X
}
. The wedge, also called the bouquet,
_
}
X
}
of the pointed spaces X
}
together with the i
k
is the pointed sum in TOP
0
.
The sum and the product in TOP
0
also represent the sum and the product in
h-TOP
0
(use (2.1.6)).
Let (. a) and (T. b) be pointed spaces. Their smash product is
. T = T, b L a T = T, T.
(This is not a categorical product. It is rather analogous to the tensor product.) The
smash product is a functor in two variables and also compatible with homotopies:
Given : C. g: T D we have the induced map
. g: . T C . D. (a. b) ((a). g(b)).
and homotopies
t
. g
t
induce a homotopy
t
.g
t
. Unfortunately, there are point-set
topological problems with the associativity of the smash product (see Problem 14).
A pair .X; A/ of topological spaces consists of a space X and a subspace .
A morphism : (X. ) (Y. T) between pairs is a map : X Y such that
() T. In this way we obtain the category of pairs TOP(2). A homotopy
H in this category is assumed to have each H
t
a morphism of pairs. We write
(X. ). (Y. T)| for the associated homotopy sets and h-TOP(2) for the homotopy
category.
If (X. ) is a pair, we usually consider the quotient space X, as a pointed
space ( identied to a point) with base point {]. If = 0, then X, = X
is
X with a separate base point.
(2.2.1) Note. A continuous map : (X. ) (Y. +) into a pointed space induces
a pointed map

: X, Y . The assignment

induces a bijection
(X. ). (Y. +)| X,. Y |
0
. A verication uses (2.1.6).
We use the notation
(X. ) (Y. T) = (X Y. X T L Y ).
althoughthis is not a categorical product. Withthis notation(1
n
. d1
n
)(1
n
. d1
n
)=
(1
nn
. d1
nn
). In a similar manner we treat other congurations, e.g., triples
(X. . T) of spaces T X and the category TOP(3) of triples.
Let 1 and T be xed spaces. The category TOP
1
of spaces under 1 has as
objects the maps i : 1 X. A morphism from i : 1 X to : 1 Y is a
map : X Y such that i = . The category TOP
B
of spaces over T has as
objects the maps : X T. A morphism from : X T to q : Y T is a
map : X Y such that q = . If T is a point, then TOP
B
can be identied
with TOP, since each space has a unique map to a point. If 1 = {+] is a point, then
TOP
1
is the same as TOP
0
. If : X T is given, then
-1
(b) is called the bre
of over b; in this context, T is the base space and X the total space of . A map
in TOP
B
will also be called brewise or bre preserving.
Categories like TOP
1
or TOP
B
have an associated notion of homotopy. A
homotopy H
t
is in TOP
1
if each H
t
is a morphism in this category. A similar
denition is used for TOP
B
. A homotopy in TOP
B
will also be called brewise
or bre preserving. Again, being homotopic is an equivalence relation in these
categories. We denote by X. Y |
1
the set of homotopy classes in TOP
1
, and
by X. Y |
B
the set of homotopy classes in TOP
B
. The homotopy categories are
h-TOP
1
and h-TOP
B
. Note that a homotopy equivalence in TOP
B
, i.e., a brewise
homotopy equivalence, from : X T to q : Y T induces for each b T
a homotopy equivalence
-1
(b) q
-1
(b) between the bres over b, so this is a
continuous family of ordinary homotopy equivalences, parametrized by T ([96],
[97], [38], [128]).
A morphism r from i : 1 X to id: 1 1 in TOP
1
is a map r : X 1
such that ri = id(X). It is called a retraction of i . If it exists, then i is an
embedding. If i : 1 X we then call 1 a retract of X. The retraction r of
i : 1 X is a homotopy equivalence inTOP
1
if and only if there exists a homotopy
h
t
: X X relative to 1 such that h
0
= id and h
1
= i r. In this case we call 1 a
deformation retract of X. The inclusion S
n
R
n1
0 is a deformation retract.
A morphism s from id: T T to : 1 T in TOP
B
is a map s : T 1
such that s = id(T). It is called a section of . If : 1 T is homotopy
equivalent in TOP
B
to id(T) we call shrinkable. All bres of a shrinkable map
are contractible.
2.2. Further Homotopy Notions 33
Problems
1. Let ((X
. .
) [ J) be a family of pointed spaces. Let

_
/
be the subset of
those points (a
where all but one a
are equal to the base point. There is a

canonical bijective continuous map
_

_
/
. If J is nite, then this map is a

homeomorphism. If J is innite and (X
. .
) = (1. 0), then it is not a homeomorphism.

2. The canonical maps i
k
: X
k

_
are embeddings.
3. The comb space X is dened as T 0. 1| L0. 1| {1] with T = {n
-1
[ n N] L{0].
Then X is contractible but not pointed contractible with respect to (0. 0). Let Y = X
be another comb space. Then X L Y X Y is not contractible. Since (X L Y ),Y is
homeomorphic to X, we see that it does not sufce in (2.1.7) to assume that is contractible.
These counterexamples indicate the need for base points with additional (local) proper-
ties.
4. There exists a contractible subspace X R
2
which is not pointed contractible to any of
its points.
5. Let the homotopy H
I
in (2.1.7) be pointed with respect to some base point a . Show
that : X X, is a pointed h-equivalence. Is (X. ) h-equivalent to (X. {a])?
6. Show that (2.1.7) yields a homotopy equivalence of pairs (X. ) (X,. +).
7. The inclusion (1. d1) (1. 1 {1,2]) is not an h-equivalence in TOP(2) although the
component maps 1 1 and d1 1 {1,2] are h-equivalences.
8. Let 1 R
2
consist of k points. Show, heuristically, that the complement R
2
1 is
h-equivalent to the k-fold sum
_
k
1
S
1
.
9. Remove a point fromthe torus S
1
S
1
and showthat the result is h-equivalent to S
1
S
1
.
Is there an analogous result when you remove a point from S
m
S
n
?
10. Construct an inclusion X which is a retract and a homotopy equivalence but not a
deformation retract.
11. Construct a map : 1 T such that all bres
-1
(b) are contractible but which does
not have a section. Construct an h-equivalence : 1 T which has a section but which is
not shrinkable.
12. What is the sum of two objects in TOP
!
? What is the product of two objects in TOP
!
?
13. A pullback of a shrinkable map is shrinkable. A pushout of a deformation retract is a
14. Let Y. 7 be compact or X. 7 be locally compact. Then the canonical bijection (the
identity) (X . Y ) . 7 X . (Y . 7) is a homeomorphism. (In the category of com-
pactly generated spaces (with its associated product and smash product!) the map is always
a homeomorphism. See also [155, Satz 18].)
15. Let
_
. T)
_ _
_
. T be the canonical map which is on each sum-
mand
k
. T induced by the inclusion
k

_
. Show that this map is a home-

omorphism if the index set is nite. Show that in this case both spaces are quotients of
(l
) T l
T).
16. Let be a compact subset of X and : X X, be the quotient map. Then for each
space Y the product id(Y ) is a quotient map. If X is a Hausdorff space, then is proper
and id closed.
17. The canonical map X 1 X 1,d1 X .1,d1 is a quotient map which induces
a homeomorphism X = X 1,(X d1 L {+] 1) X . 1,d1.
18. There is a canonical bijective continuous map (XY ),(XTLY ) X,.Y,T
(the identity on representatives). It is a homeomorphism if X Y X,.Y,T is a quo-
tient map, e.g., if X and Y,T are locally compact (or in the category of compactly generated
spaces).
2.3 Standard Spaces
Standard spaces are Euclidean spaces, disks, cells, spheres, cubes and simplices.
We collect notation and elementary results about such spaces. The material will be
used almost everywhere in this book. We begin with a list of spaces. The Euclidean
norm is [.[.
R
n
Euclidean space
D
n
= {. R
n
[ [.[ _ 1] n-dimensional disk
S
n-1
= {. D
n
[ [.[ = 1] = dD
n
(n 1)-dimensional sphere
1
n
= D
n
S
n-1
n-dimensional cell
1
n
= {. R
n
[ 0 _ .
i
_ 1] n-dimensional cube
d1
n
= {. 1
n
[ .
i
= 0. 1 for some i ] boundary of 1
n
^
n
= ^n| = {. R
n1
[ .
i
_ 0.
i
.
i
= 1] n-dimensional simplex
d^
n
= {(.
i
) ^
n
[ some .
i
= 0] boundary of ^
n
The spaces D
n
, 1
n
, 1
n
and ^
n
are convex and hence contractible. We think of
1
0
= {0]. The spaces D
0
, 1
0
, and ^
0
are singletons, and S
-1
= dD
0
, d^
0
are
empty. In the case of ^
n
we use the indexing t = (t
0
. . . . . t
n
) ^
n
; the subset
d
i
^
n
= {t ^
n
[ t
i
= 0] is the i -th face of ^
n
; hence d^
n
=
_
n
i=0
d
i
^
n
.
It is useful to observe that certain standard spaces are homeomorphic. Ageneral
result of this type is:
(2.3.1) Proposition. Let 1 R
n
be a compact convex subset with non-empty
interior 1
. Then there exists a homeomorphism of pairs (D

n
. S
n-1
) (1. d1)
which sends 0 D
n
to a pre-assigned . 1
.
Proof. Let 1 R
n
be closed and compact and 0 1
. Verify that a ray from0 in-

tersects the boundary d1 of 1 in R
n
in exactly one point. The map : d1 S
n-1
,
. .,[.[ is a homeomorphism. The continuous map : S
n-1
0. 1| 1,
(.. t ) t
-1
(.) factors over q : S
n-1
0. 1| D
n
, (.. t ) t. and yields a
bijective map k : D
n
1, hence a homeomorphism (use (1.4.3)).
This proposition can be used to deduce a homeomorphism (D
n
. S
n-1
)
(1
n
. d1
n
). The simplex ^
n
is a compact convex subset with interior points in the
hyperplane {. R
n1
[
i
.
i
= 1]. From this fact we deduce a homeomorphism
(D
n
. dD
n
) (^
n
. d^
n
).
The sphere S
n
, as a homeomorphismtype, will appear in many different shapes.
2.3. Standard Spaces 35
(2.3.2) Example. Let N = e
n1
= (0. . . . . 0. 1) R
n1
. We dene the stere-
ographic projection
1
: S
n
{e
n1
] R
n
; the point
1
(.) is the intersec-
tion of the line through e
n1
and . with the hyperplane R
n
0 = R
n
. One
computes
1
(.
1
. . . . . .
n1
) = (1 .
n1
)
-1
(.
1
. . . . . .
n
). The inverse map is
1
: . ((1 [.[)
2
)
-1
(2.. [.[
2
1). We also have the stereographic projec-
tion
S
: S
n
{e
n1
] R
n
and the transition map is
S

-1
1
(,) = [,[
-2
,.
From the stereographic projection we obtain S
n
as a specic model of the one-
point compactication R
n
L {o] by extending
1
(o) = e
n1
. We also write
S
V
= V L {o] for the one-point compactication of a nite-dimensional real
vector space V .
2.3.3 Spheres. Let , S
n
. From (2.3.2) we see that S
n
, is homeomorphic to
R
n
and hence contractible. Thus, if X S
n
is not surjective, it is null homotopic.
The inclusion i : S
n
R
n1
{0] is an h-equivalence with homotopy inverse
N : R
n1
{0] S
n
, . [.[
-1
.. A homotopy (rel S
n
) from i N to the
identity is the linear homotopy (.. t ) t. (1 t )iN(.). Moreover N i = id.
We see that i is a deformation retract.
Under suitable circumstances each map in a small neighbourhood of is al-
ready homotopic to . For a general theorem to this effect see (15.8.3). Here we
only give a simple, but typical, example. Let . g: X S
n
be maps such that
[(.) g(.)[ < 2. Then they are homotopic by a linear homotopy when viewed
as maps into R
n1
{0]. We compose with N and see that . g.
If : S
n
S
n
is a continuous map, then there exists (by the theorem of
StoneWeierstrass, say) a C
o
-map g: S
n
S
n
such that [(.) g(.)[ < 2.
This indicates another use of homotopies: Improve maps up to homotopy. If one
uses some analysis, namely (the easy part of) the theorem of Sard about the density
of regular values, one sees that for m < n a C
o
-map S
n
S
n
is not surjective
and hence null homotopic. (Later we prove this fact by other methods.) There exist
surjective continuous maps S
1
S
2
(Peano curves); this ungeometric behaviour
of continuous maps is the source for many of the technical difculties in topology.
(2.3.4) Proposition. The map : S
n-1
1 D
n
, (.. t ) (1t ). is a quotient
map. Given J : D
n
X, the composition J: S
n-1
1 X is a null homotopy
of = J[S
n-1
. Each null homotopy of a map : S
n-1
X arises from a
unique J.
Proof. Since a null homotopy H of : S
n-1
X sends S
n-1
1 to a point, it
factors through the quotient map q : S
n-1
1 S
n-1
1,S
n-1
1. Thus null
homotopies H correspond via

H

Hq to maps

H: S
n-1
1,S
n-1
1 X.
The map induces a homeomorphism : S
n-1
1,S
n-1
1 D
n
(use (1.4.3)).
Hence there exists a unique J such that J =

H.
Let us use the notation
S(n) = 1
n
,d1
n
. S
(n)
= R
n
L {o].
since these spaces are homeomorphic to S
n
. The canonical map 1
nn
,d1
nn
1
n
d1
n
.1
n
,d1
n
which is the identity on representatives is a pointed homeomor-
phism. If V and W are nite-dimensional real vector spaces, we have a canonical
pointed homeomorphism S
V
.S
W
S
V W
which is the identity away from the
base point. The homeomorphism |0. 1 R, s
2x-1
x(1-x)
induces a homeomor-
phism; : S(1) S
(1)
which transports t 1t into the antipodal map . .
on R. We obtain an induced homeomorphism
;
n
: S(n) = S(1) . . S(1) S
(1)
. . S
(1)
= S
(n)
of the n-fold smash products.
(2.3.5) Example. A retraction r : D
n
S
n-1
1 L D
n
0 is r(.. t ) =
(2(.. t )
-1
.. (.. t )2t ) with (.. t ) = max(2[.[. 2t ). (See Figure 2.1, a
central projection fromthe point (0. 2).) Given a map : 1
n
X and a homotopy
h: d1
n
1 X with h
0
= [d1
n
combine to a map g: 1
n
0 Ld1
n
1 X.
We compose with a retraction and obtain a homotopy H: 1
n
X which extends
h and begins at H
0
= . This homotopy extension property is later studied more
generally under the name of cobration.
........................................
,
f
f
f
f
f
f
f
f
f
f
f
1
D
n
,
,
.
r(.)
(0. 2)
Figure 2.1. A retraction.
(2.3.6) Example. The assignment H: (.. t ) ((.. t )
-1
(1 t ) .. 2 (.. t ))
with the function (.. t ) = max(2[.[. 2 t ) yields a homeomorphism of pairs
(D
n
. S
n-1
) (1. 0) D
n
(1. 0), see Figure 2.2.
Similarly for (1
n
. d1
n
) in place of (D
n
. S
n-1
), since these two pairs are homeo-
morphic.
2.4. Mapping Spaces and Homotopy 37
e
e
e
e
e
e
e
e
e
e
a a b
c b c
a
t
b
t
a
t
c
t
b
t
c
t
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
D
n
1
Figure 2.2. A relative homeomorphism.
Problems
1. Construct a homeomorphism (D
m
. S
m-1
) (D
n
. S
n-1
) (D
mn
. S
mn-1
).
2. R
n1
{.] S
n
, : (: .),[: .[ is an h-equivalence.
3. Let D
n
= {(.
0
. . . . . .
n
) S
n
[ .
n
_ 0]. Show that the quotient map S
n
S
n
,D
n
is an h-equivalence.
4. Let
1
. . . . .
k
: C
n
C be linearly independent linear forms (k _ n). Then the com-
plement C
n

-1
(0) is homotopy equivalent to the product of k factors S

1
.
5. S
n
{(.. ,) S
n
S
n
[ . ,= ,], . (.. .) is an h-equivalence.
6. Let . g: X S
n
be maps such that always (.) ,= g(.). Then . g.
7. Let 1
n
be star-shaped with respect to 0. Showthat S
n-1
R
n
is a deformation
retract.
8. The projection : TS
n
= {(.. ) S
n
R
n1
[ . J ] S
n
, (.. ) . is
called the tangent bundle of S
n
. Show that admits a brewise homeomorphism with
pr : S
n
S
n
D S
n
, (.. ,) . (with D the diagonal).
2.4 Mapping Spaces and Homotopy
It is customary to endowsets of continuous maps with a topology. In this section we
reviewfrompoint-set topology the compact-open topology. It enables us to consider
a homotopy H: X 1 Y as a family of paths in Y , parametrized by X. This
dual aspect of the homotopy notion will be used quite often. It can be formalized;
but we use it more like a heuristic principle to dualize various constructions and
notions in homotopy theory (EckmannHilton duality).
We denote by Y
A
or J(X. Y ) the set of continuous maps X Y . For 1 X
and U Y we set W(1. U) = { Y
A
[ (1) U]. The compact-open
topology (CO-topology) on Y
A
is the topology which has as a subbasis the sets
of the form W(1. U) for compact 1 X and open U Y . In the sequel the
set Y
A
always carries the CO-topology. A continuous map : X Y induces
continuous maps
7
: X
7
Y
7
, g g and 7
(
: 7
Y
7
A
, g g .
(2.4.1) Proposition. Let X be locally compact. Then the evaluation e
A,Y
=
e : Y
A
X Y , (. .) (.) is continuous.
Proof. Let U be an open neighbourhood of (.). Since is continuous and X
locally compact, there exists a compact neighbourhood 1 of . such that (1) U.
The neighbourhood W(1. U) 1 of (. .) is therefore mapped under e into U.
This shows the continuity of e at (. .).
(2.4.2) Proposition. Let : X Y 7 be continuous. Then the adjoint map
.
: X 7
Y
,
.
(.)(,) = (.. ,) is continuous.
Proof. Let 1Y be compact andU 7open. It sufces toshowthat W(1. U) has
an open pre-image under
.
. Let
.
(.) W(1. U) and hence ({.] 1) U.
Since 1 is compact, there exists by (1.4.1) a neighbourhood V of . in X such that
V 1
-1
(U) and hence
.
(V ) W(1. U).
From (2.4.2) we obtain a set map : 7
AY
(7
Y
)
A
,
.
. Let e
Y,7
be continuous. A continuous map : X 7
Y
induces a continuous map
=
e
Y,7
(id
Y
): XY 7
Y
Y 7. Hence we obtain a set map : (7
Y
)
A
7
AY
,
.
(2.4.3) Proposition. Let e
Y,7
be continuous. Then and are inverse bijections.
Thus : X Y 7 is continuous if
: X Y 7 is continuous, and
: X Y 7 is continuous if
.
: X 7
Y
is continuous.
(2.4.4) Corollary. If h: XY 1 7 is a homotopy, then h
.
: X1 7
Y
is a
homotopy (see (2.4.2)). Hence XY. 7| X. 7
Y
|, |
.
| is well-dened.
If, moreover, e
Y,7
is continuous, e.g., Y locally compact, then this map is bijective
(see (2.4.3)).
2.4.5 Dual notion of homotopy. We have the continuous evaluation e
t
: Y
J
Y ,
n n(t ). A homotopy from
0
: X Y to
1
: X Y is a continuous
map h: X Y
J
such that e
t
h =
t
for c = 0. 1. The equivalence with our
original denition follows from (2.4.3): Since 1 is locally compact, continuous
maps X 1 Y correspond bijectively to continuous maps X Y
J
.
(2.4.6) Theorem. Let 7 be locally compact. Suppose : X Y is a quotient
map. Then id(7): X 7 Y 7 is a quotient map.
Proof. We verify for id the universal property of a quotient map: If h: Y 7
C is a set map and h ( id) is continuous, then h is continuous. The adjoint of
h ( id) is h
.
. By (2.4.2), it is continuous. Since is a quotient map, h
.
is continuous. Since 7 is locally compact, h is continuous, by (2.4.3).
(2.4.7) Theorem (Exponential law). Let X and Y be locally compact. Then the
adjunction map : 7
AY
(7
Y
)
A
is a homeomorphism.
2.4. Mapping Spaces and Homotopy 39
Proof. By (2.4.3), is continuous if
1
= e
A,7
Y ( id) is continuous. And
this map is continuous if
2
= e
Y,A
(
1
id) is continuous. One veries that
2
= e
AY,7
. The evaluations which appear are continuous by (2.4.1).
The inverse
-1
is continuous if e
AY,7
(
-1
id) is continuous, and this
map equals e
Y,7
(e
A,7
Y id).
Let (X. .) and (Y. ,) be pointed spaces. We denote by J
0
(X. Y ) the space of
pointed maps with CO-topology as a subspace of J(X. Y ). In J
0
(X. Y ) we use the
constant map as a base point. The adjoint
.
: X J(Y. 7) of : X Y 7
is a pointed map into J
0
(X. Y ) if and only if sends X , L . Y to the base
point of 7. Let : X Y X . Y = X Y,(X , L . Y ) be the quotient
map.
If g: X.Y 7is given, we denote the adjoint of g: XY X.Y 7
by
0
(g) and consider it as an element of J
0
(X. J
0
(Y. 7)). In this manner we
obtain a set map
0
: J
0
(X . Y. 7) J
0
(X. J
0
(Y. 7)).
The evaluation J
0
(X. Y ) X Y , (. .) (.) factors over the quotient
space J
0
(X. Y ) . X and induces e
0
= e
0
A,Y
: J
0
(X. Y ) . X Y . From (2.4.1)
we conclude:
(2.4.8) Proposition. Let X be locally compact. Then e
0
A,Y
is continuous.
Let e
0
A,Y
be continuous. From a pointed map : X J
0
(Y. 7) we ob-
tain
=
0
() = e
0
A,Y
( . id): X . Y 7, and hence a set map
0
: J
0
(X. J
0
(Y. 7)) J
0
(X . Y. 7).
0
A,Y
be continuous. Then
0
and
0
are inverse bijections.
(2.4.10) Corollary. Let h: (X . Y ) 1 7 be a pointed homotopy. Then
0
(h
t
): X J
0
(Y. 7) is a pointed homotopy and therefore
X . Y. 7|
0
X. J
0
(Y. 7)|
0
. |
0
( )|
is well dened. If, moreover, e
0
A,Y
is continuous, then this map is bijective.
By a proof formally similar to the proof of (2.4.7), we obtain the pointed version
of the exponential law.
(2.4.11) Theorem (Exponential law). Let X and Y be locally compact. Then the
pointed adjunction map
0
: J
0
(X . Y. 7) J
0
(X. J
0
(Y. 7)) is a homeomor-
phism.
(2.4.12) Lemma. Let k
o
: 7 denote the constant map with value a. Then
[: X
7
(X )
7
, (. a) (. k
o
) is continuous.
Proof. Let [(. a) W(1. U). This means: For . 1 we have ((.). a) U.
There exist open neighbourhoods V
1
of (1) in X and V
2
of a in such that
V
1
V
2
U. The inclusion [(W(1. V
1
) V
2
) W(1. U) shows the continuity
of [ at (. a).
(2.4.13) Proposition. Ahomotopy H
t
: X Y induces homotopies H
7
t
and 7
1
I
.
Proof. In the rst case we obtain, with a map [ from (2.4.12), a continuous map
H
7
[: X
7
1 (X 1)
7
Y
7
.
In the second case we use the composition
e ( id) (7
1
id): 7
Y
1 7
AJ
1 (7
A
)
J
1 7
A
which is continuous.
(2.4.14) Corollary. Let be a homotopy equivalence. Then the induced maps
J(7. X) J(7. Y ) and J(Y. 7) J(X. 7) are h-equivalences. If is a
pointedh-equivalence, the inducedmaps J
0
(7. X) J
0
(7. Y ) andJ
0
(Y. 7)
J
0
(X. 7) are pointed h-equivalences.
Problems
1. Verify that
.
and 7
are continuous.
2. An inclusion i : 7 Y induces an embedding i
: 7
.
3. The canonical map J
_
. Y
_
J(X
. Y ) is always a homeomorphism.
4. The canonical map J(X.
J(X. Y
), (pr
) is always bijective and

continuous. If X is locally compact, it is a homeomorphism.
5. Let : X Y be a surjective continuous map. Suppose the pre-image of a compact set
is compact. Then 7
)
: 7
Y
7
is an embedding.
6. We have a canonical bijective map J
0
_ _
J
X
. Y
_

J
J
0
(X
. Y ), since
_
is the sum in TOP

0
. If J is nite, it is a homeomorphism.
7. Let X. Y. U, and V be spaces. The Cartesian product of maps gives a map
: U
V
Y
(U V )
Y
. (. g) g.
Let X and Y be Hausdorff spaces. Then the map is continuous.
8. By denition of a product, a map X Y 7 is essentially the same thing as a pair of
maps X Y , X 7. In this sense, we obtain a tautological bijection t : (Y 7)
. Let X be a Hausdorff space. Then the tautological map t is a homeomorphism.

9. Let X and Y be locally compact. Then composition of maps 7
Y
Y
, (g. )
g is continuous.
10. Let (Y. +) be a pointed space, (X. ) a pair of spaces and : X X, the quotient
map. The space X,is pointed with base point {]. Let J((X. ). (Y. +)) be the subspace of
J(X. Y ) of the maps which send to the base point. Composition with induces a bijective
continuous map ; : J
0
(X,. Y ) J((X. ). (Y. +)); and a bijection of homotopy sets
2.5. The Fundamental Groupoid 41
X,. Y |
0
(X. ). (Y. +)|. If has compact pre-images of compact sets, then ; is a
homeomorphism.
11. Consider diagrams where the right-hand one is obtained by multiplying the left-hand
one with X:
T X
C

D, C X

D X.
If the left-hand diagram is a pushout in TOP and X locally compact, then the right-hand
diagram is a pushout in TOP. In TOP
0
the smash product with a locally compact space yields
again a pushout.
12. The CO-topology on the set of linear maps R
n
R is the standard topology.
13. Let X be a compact space and Y a metric space. Then the CO-topology on Y
is
induced by the supremum-metric.
2.5 The Fundamental Groupoid
Apath in the plane can be quite ungeometric: nowhere differentiable, innite length,
surjective onto 1 1 (a so-called Peano curve). We introduce an equivalence relation
on paths, and the equivalence classes still capture qualitative geometric properties
of the path. In particular a reparametrization of a path (different velocity) does
not change basic topological properties.
We consider homotopy classes relative to d1 of paths. A homotopy of paths
is always assumed to be relative to d1. A homotopy of paths between paths u
and with the same end points .
0
= u(0) = (0), .
1
= u(1) = (1) is a map
H: 1 1 X such that
H(s. 0) = u(s).
H(s. 1) = (s).
H(0. t ) = u(0) = (0).
H(1. t ) = u(1) = (1).
Thus for each t 1 we have a path H
t
: s H(s. t ) and all these paths have the
same end points. We write H: u . for this homotopy.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
.
0
] t
.
1
.
1
H
0
= u
H
1
=
H
t
H constant along dotted lines
Being homotopic in this sense is an equivalence relation on the set of all paths
from . to ,. The product operation is compatible with this relation as the next
proposition shows.
(2.5.1) Proposition. The product of paths has the following properties:
(1) Let : 1 1 be continuous and (0) = 0. (1) = 1. Then u . u.
(2) u
1
+ (u
2
+ u
2
) . (u
1
+ u
2
) + u
3
(if the products are dened ).
(3) u
1
. u
t
1
and u
2
. u
t
2
implies u
1
+ u
2
. u
t
1
+ u
t
2
.
(4) u + u
-
is always dened and homotopic to the constant path.
(5) k
u(0)
+ u . u . u + k
u(1)
.
Proof. (1) H: (s. t ) u(s(1 t ) t (s)) is a homotopy from u to u.
(2) The relation (u
1
+ (u
2
+ u
3
)) = (u
1
+ u
2
) + u
3
holds for dened as
(t ) = 2t for t _
1
4
, (t ) = t
1
4
for
1
4
_ t _
1
2
, (t ) =
t
2

1
2
for
1
2
_ t _ 1.
(3) Given J
i
: u
i
. u
t
i
then G: u
1
+ u
2
. u
t
1
+ u
t
2
for G dened as G(s. t ) =
J
1
(2s. t ) for 0 _ t _
1
2
and G(s. t ) = J
2
(2s 1. t ) for
1
2
_ t _ 1.
(4) The map J : 1 1 X dened as J(s. t ) = u(2s(1 t )) for 0 _ s _
1
2
and J(s. t ) = u(2(1 s)(1 t ) for
1
2
_ t _ 1 is a homotopy from u + u
-
to the
constant path. (At time t we only use the path from 0 to (1 t ) and compose it
with its inverse.)
(5) is proved again with the parameter invariance (1).
From homotopy classes of paths in X we obtain again a category, denoted
(X). The objects are the points of X. A morphism from . to , is a homotopy
class relative to d1 of paths from. to ,. A constant path represents an identity. If u
is a path from a to b and n a path from b to c, then we have the product path u +
from a to c, and the composition of morphisms is dened by | u| = u + |. In
this category each morphism has an inverse, i.e. is an isomorphism, represented by
the inverse path. A category with this property is called groupoid. Note that in a
groupoid the endomorphism set of each object becomes a group under composition
of morphisms. The category (X) is called the fundamental groupoid of X. The
automorphism group of the object . in this category is the fundamental group of
X with respect to the base point .. The usual rules of categorical notation force
us to dene the multiplication in this group by u| | = + u|. As long as
we are just interested in this group (and not in the categorical aspect), we use the
opposite multiplication u| | = u+| and denote this group by
1
(X. .). This is
the traditional fundamental group of the pointed space (X. .) (Poincar 1895 [151,
12]). An element in
1
(X. .) is represented by a closed path n based at . (i.e.,
n(0) = n(1) = .), also called a loop based at ..
(2.5.2) Remark. We can obtain the fundamental groupoid (X) as a quotient
category of the path category W(X). In that case we call paths u: 0. a| : 1 X
and : 0. b| X from. to , homotopic, if there exist constant paths with image ,
2.5. The Fundamental Groupoid 43
such that the compositions with u and , respectively, have the same domain of
denition 0. c| and the resulting composed paths are homotopic rel {0. c].
(2.5.3) Remark. The set
1
(X. .) has different interpretations. A loop based at
. is a map n: (1. d1) (X. .). It induces a pointed map n: 1,d1 X. The
exponential function
0
: 1 S
1
, t exp(2i t ) induces a pointed homeomor-
phism q : 1,d1 S
1
which sends the base point {d1] to the base point 1. There
exists a unique u: S
1
X such that uq = n. Altogether we obtain bijections
1
(X. .) = (1. d1). (X. .)| 1,d1. X|
0
S
1
. X|
0
.
induced by n| - n| = uq| -u|.
It is a general fact for groupoids that the automorphism groups Aut(.) =
(.. .) and Aut(,) = (,. ,) of objects .. , in are isomorphic, provided there
exists a morphism from . to ,. If (.. ,), then
(.. .) (,. ,).
-1
is an isomorphism. It depends on the choice of ; there is, in general, no canonical
isomorphism between these groups. Thus fundamental groups associated to base
points in the same path component are isomorphic, but not canonically.
A space is simply connected or 1-connected if it is path connected and its
fundamental group is trivial (consists of the neutral element alone).
A continuous map : X Y induces a homomorphism
1
( ):
1
(X. .)
1
(Y. (.)). u| u|
and, more generally, a functor
( ): (X) (Y ). . (.). u| u|.
In this way,
1
becomes a functor from TOP
0
to the category of groups, and a
functor from TOP to the category of small categories (small category: its objects
form a set). Homotopies correspond to natural transformations:
(2.5.4) Proposition. Let H: X 1 Y be a homotopy from to g. Each . X
yields the path H
x
: t H(.. t ) and the morphism H
x
| in (Y ) from (.) to
g(.). The H
x
| constitute a natural transformation (H) from ( ) to (g).
Proof. The claim says that for each path u: 1 X the relation u + H
u(1)
.
H
u(0)
+gu holds. We use 1 1 Y , (s. t ) H(u(s). t ). We obtain u+H
u(1)
as composition with a + b and H
u(0)
+ gu as composition with c + J, where a, b,
c, and J are the sides of the square: a(t ) = (t. 0), b(t ) = (1. t ), c(t ) = (0. t ),
J(t ) = (t. 1). But a + b and c + J are homotopic by a linear homotopy.
We express the commutativity of (2.5.4) in a different way. It says that
t
1
g
+
=
+
: X(.. ,) Y( .. ,).
where t
1
: (Y )(g.. g,) (Y )( .. ,) is the bijection a H
,
|
-1
aH
x
|.
The rule (1 + 1) = (1)(1) is obvious. Hence if is an h-equivalence
with h-inverse g: Y X, then ( )(g) and (g)( ) are naturally isomor-
phic to the identity functor, i.e., ( ) is an equivalence of categories. The natural
transformation (H) only depends on the homotopy class relative to Xd1 of H.
It is a general categorical fact that a natural equivalence of categories induces
a bijection of morphism sets. We prove this in the notation of our special case at
hand.
(2.5.5) Proposition. Let : X Y be a homotopy equivalence. Then the functor
( ): (X) (Y ) is an equivalence of categories. The induced maps between
morphism sets
+
: X(.. ,) Y( .. ,) are bijections. In particular,
1
( ):
1
(X. .)
1
(Y. (.)). n| n|
is an isomorphism for each . X. A contractible space is simply connected.
Proof. Let g: Y X be h-inverse to . Consider
X(.. ,)
(
Y( .. ,)
X(g .. g,)
(
Y(g .. g,).
Choose H: g . id(X). Then g
+
+
= (g )
+
= t
1
(id)
+
= t
1
is a bijection,
hence g
+
is surjective. In a similar manner one proves that
+
g
+
is a bijection, hence
g
+
is also injective. Since g
+
+
and g
+
are bijective we see that
+
is bijective.
The fundamental group forces us to work with pointed spaces. Usually the base
points serve some technical purpose and one has to study what happens when the
base point is changed. For pointed h-equivalences it would be immediately clear
that
1
( ) is an isomorphism. For the more general case (2.5.5) one needs some
argument like the one above.
(2.5.6) Proposition. Let (X. .
0
) and (Y. ,
0
) be pointed spaces and i
A
: X
X Y , . (.. ,
0
) and i
Y
: Y X Y , , (.
0
. ,). Then
1
(X. .
0
)
1
(Y. ,
0
)
1
(X Y. (.
0
. ,
0
)). (u. ) i
A
+
u i
Y
+

is a well-dened isomorphism with inverse : (pr
A
+
:. pr
Y
+
:).
Proof. Since homotopy is compatible with products we know already that the sec-
ond map is an isomorphism. In order to show that the rst map is a homomorphism
we have to verify that i
A
+
u commutes with i
Y
+
. Let now u and be actual paths
2.6. The Theorem of Seifert and van Kampen 45
and write n = (u) with the diagonal . With a notation introduced in the proof
of (2.5.4) we have (u )(a + b) = i
A
u + i
Y
and (u )(c + J) = i
Y
+ i
A
u.
We now use that , a + b, and c + J are homotopic (linear homotopy). It should be
clear that the two maps of the proposition are inverse to each other.
2.6 The Theorem of Seifert and van Kampen
Let a space X be the union of subsets X
0
. X
1
. A general problem is to derive prop-
erties of X from those of X
0
, X
1
, and X
01
= X
0
X
1
. (Similar problem for more
general unions.) Usually the covering has to satisfy certain reasonable conditions.
In this section we consider the fundamental groupoid and the fundamental group
under this aspect. The basic result is the theorem (2.6.2) of Seifert [166] and van
Kampen [100].
We rst prove an analogous and slightly more general result for groupoids [34].
The result is more formal but the proof is (notationally) simpler because we need
not take care of base points. Note that the hypothesis of the next theorem implies
that X is the pushout in TOP of the inclusions X
0
X
01
X
1
. Thus (2.6.1) says
that the functor preserves pushouts.
(2.6.1) Theorem (R. Brown). Let X
0
and X
1
be subspaces of X such that the
interiors cover X. Let i
: X
01
X
and
: X
X be the inclusions. Then

(X
01
)
(i
0
)
(i
1
)
(X
0
)
(}
0
)
(X
1
)
(}
1
)
(X)
is a pushout in the category of groupoids.
Proof. Let h
: (X
) be functors into a groupoid such that h

1
(i
1
) =
h
0
(i
0
). We have to show: There exists a unique functor z: (X) such that
h
1
= z(
1
) and h
0
= z(
0
). We begin with a couple of remarks.
A path n: a. b| U represents a morphism n| in (U) from n(a) to n(b)
if we compose it with an increasing homeomorphism : 0. 1| a. b|.
If a = t
0
< t
1
< < t
n
= b, then n represents the composition of the
morphisms n[t
i
. t
i1
||.
Suppose that n: 1 X is a path. Then there exists a decomposition 0 =
t
0
< t
1
< < t
n1
= 1 such that n(t
i
. t
i1
|) is contained in a set X
. Choose
; : {0. . . . . m] {0. 1] such that n(t
i
. t
i1
|) X
;(i)
. Consider n[t
i
. t
i1
| as
path n
i
in X
;(i)
. Then
n| = (
;(n)
)n
n
| (
;(0)
)n
0
|.
If z exists, then
(i) zn| = h
;(n)
n
n
| h
;(0)
n
0
|.
i.e., z is uniquely determined.
In order to show the existence of z, we have to dene zn| by (i). We have
to verify that this is well-dened. The commutativity h
0
(i
0
) = h
1
(h
1
) shows
that a different choice of ; yields the same result.
Since h
0
and h
1
are functors, we obtain the same result if we rene the decom-
position of the interval.
It remains to be shown that (i) only depends on the homotopy class of the path.
Let H: 1 1 X be a homotopy of paths from . to ,. There exists n Nsuch
that H sends each sub-square i,n. (i 1),n| ,n. ( 1),n| into one of the
sets X
(see (2.6.4)). We consider edge-paths in the subdivided square 1 1 which

differ by a sub-square, as indicated in the following gure (n = 5).
(0. 0)
(1. 1)
,
,
We apply H and obtain two paths in X. They yield the same result (i), since they
differ by a homotopy on some subinterval which stays inside one of the sets X
.
Changes of this type allow us to pass inductively from the H on the lower to H on
the upper boundary path from (0. 0) to (1. 1). These paths differ from H
0
and H
1
by composition with a constant path.
Finally, from the construction we conclude that z is a functor.
(2.6.2) Theorem (Seifertvan Kampen). Let X
0
and X
1
be subspaces of X such
that the interiors cover X. Let i
: X
01
= X
0
X
1
X
and
: X
X be the
inclusions. Suppose that X
0
. X
1
. X
01
are path connected with base point + X
01
.
Then
1
(X
01
. +)
i
1

i
0
1
(X
1
. +)
}
1
1
(X
0
. +)
}
0

1
(X. +)
is a pushout in the category of groups.
Proof. The theorem is a formal consequence of (2.6.1). In general, if 7 is path
connected and : 7 we have a retraction functor r : (7)
1
(7. :) onto the
full subcategory with object :. For each : 7 we choose a morphismu
x
(7)
2.7. The Fundamental Group of the Circle 47
from. to: suchthat u
z
= id. Thenr assigns u
,
u
-1
x
toa morphism: . ,. We
apply this to 7 = X
01
. X
0
. X
1
. X and : = + and choose a morphism u
x
(7)
if . is contained in 7. We obtain a commutative diagram of functors
(X
0
)
i
1
(X
01
)
i
01
(X
1
)
i
0
1
(X
0
. +)
1
(X
01
. +)

1
(X
1
. +).
Given homomorphisms
:
1
(X
. +) G into a group G (= a groupoid with a

single object) which agree on
1
(X
01
. +), we compose with r
and apply (2.6.1) to

obtain a functor (X) G. Its restriction to
1
(X. +) is the unique solution of
the pushout problem in (2.6.2).
(2.6.3) Remark. From the proof of (2.6.1) we see that each morphism in (X) is
a composition of morphisms in (X
0
) and (X
1
). Similarly, in (2.6.2) the group
1
(X. +) is generated by the images of
0+
and
1+
. This algebraic fact is not
immediately clear from the denition of a pushout.
We have used above the next fundamental result. It is impossible to prove a
geometric results about continuous maps without subdivision and approximation
procedures. In most of these procedures (2.6.4) will be used.
(2.6.4) Proposition (Lebesgue). Let X be a compact metric space. Let A be an
open covering of X. Then there exists c > 0 such that for each . X the c-neigh-
bourhood U
t
(.) is contained in some member of A. (An c with this property is
called a Lebesgue number of the covering.)
2.7 The Fundamental Group of the Circle
The space R is simply connected; (R) has a single morphism between any two
objects. We consider (R) as a topological groupoid: The object space is R, the
morphism space is R R, the source is (a. b) a, the range (a. b) b, the
identity a (a. a), and (b. c) (a. b) = (a. c) the composition.
The continuous map : R S
1
induces a functor (): (R) (S
1
).
It turns out that this functor is surjective on morphisms and provides us with an
algebraic description of (S
1
). So let us dene a topological groupoid G. The
object space is S
1
, the morphism space is S
1
R, the source (a. t ) a, the range
(a. t ) a exp(2i t ), the identity a (a. 0), and the composition (b. t )(a. s) =
(a. s t ).
The assignments a exp(2i a) and (a. b) (exp(2i a). b a) yield a
continuous functor (R) G. We will show that G (forgetting the topology) is
(S
1
).
We have the open covering of S
1
Cby X
0
= S
1
{1] and X
1
= S
1
{1]
with inclusions i
k
: X
01
X
k
and
k
: X
k
S
1
. The sets X
k
are contractible,
hence simply connected. Therefore there exists a single morphism (a. b)
k
: a b
between two objects a. b of (X
k
).
We have bijective maps
0
: |0. 1 X
0
and
1
: | 1,2. 1,2 X
1
given
by t exp(2i t ). We dene functors ;
k
: (X
k
) G by the identity on
objects and by ;
k
(a. b)
k
= (a.
-1
k
(b)
-1
k
(a)). Moreover we have a functor
: G (S
1
) whichis the identityonobjects andwhichsends the morphism(a. t )
of G to the class of the path 1 S
1
, s a exp(2i t s) from a to a exp(2i t ).
(The idea behind the denition of G is the fact that each path in S
1
is homotopic
to one of this normal form, see (2.7.9).) The following diagram is commutative.
(X
0
)
;
0
E
E
E
E
E
E
E
E
E
}
0
(X
01
)
i
0
t
t
t
t
t
t
t
t
t
i
1
J
J
J
J
J
J
J
J
J
G
(
(S
1
)
(X
1
)
;
1
y
y
y
y
y
y
y
y
y
}
1
(2.7.1) Proposition. The functor is an isomorphism.

Proof. We apply (2.6.1) to the pair (;
0
. ;
1
) and obtain a functor ; : (S
1
) G.
The uniqueness property of a pushout solution shows ; = id. In order to show
; = id we note that the morphisms of G are generated by the images of ;
0
and ;
1
. Given (a. t ) G(a. b), choose a decomposition t = t
1
t
n
such
that [t
i
[ < 1,2 for each r. Set a
0
= a and a
i
= a exp(2i(t
1
t
i
)). Then
(a. t ) = (a
n-1
. t
n
) (a
1
. t
2
) (a
0
. t
1
) in the groupoid G. Since [t
i
[ < 1,2
there exists for each r a k(r) {0. 1] such that a
i-1
exp(2i t
i
s) X
k(i)
for s 1.
Then (a
i-1
. t
i
) = ;
k(i)
(a
i-1
. a
i
)
k(i)
. Thus G(a. b) is generated by morphisms in
the images of the ;
k
.
The unit circle S
1
in the complex plane is the prototype of a loop. Typical
elements in the fundamental group are obtained by running ntimes around the circle.
Up to homotopy, there are no other possibilities. With (2.7.1) we have determined
the fundamental group
1
(S
1
. 1), namely as the automorphism group in (S
1
) of
the object 1. The automorphisms of the object 1 in G are the (1. n). n Z and
(1. n) is the loop t exp(2i nt ).
(2.7.2) Theorem. Let s
n
: 1 S
1
be the loop t exp(2i nt ). The assignment
: Z
1
(S
1
. 1), n s
n
| is an isomorphism.
The circle S
1
is a group with respect to multiplication of complex numbers.
We show that the composition law in
1
(S
1
. 1) can also be dened using this
multiplication.
More generally, assume that X is a space with a continuous multiplication
m: X X X. (.. ,) m(.. ,) = .,
and neutral element up to homotopy e X (the base point), i.e., the maps .
m(e. .) and . m(.. e) are both pointed homotopic to the identity. We call such
an object a monoid in h-TOP. (We do not assume that mis associative or commuta-
tive.) We dene a composition law on the pointed homotopy set Y. X|
0
, called the
m-product, by |. g| |
n
g| = g|; here g: , m((,). g(,))
is the ordinary pointwise multiplication. The constant map represents a two-sided
unit for the m-product. In a similar manner we dene by pointwise multiplication
of loops the m-product on
1
(X. e). The set
1
(X. e) S
1
. X|
0
now carries two
composition laws: the m-product and the +-product of the fundamental group.
(2.7.3) Proposition. Let (X. m) be a monoid in h-TOP. Then the +-product and the
m-product on
1
(X. e) coincide and the product is commutative.
Proof. Let k be the constant loop. Then for any two loops u and the relations
u + . (u k) + (k ) = (u + k) (k + )) . u .
u + . u . (k + u) ( + k) = (k ) + (u k) . + u
hold. In order to see the equalities, write down the denition of the maps.
(2.7.4) Lemma. The map : S
1
. S
1
|
0
S
1
. S
1
| which forgets the base point is
a bijection.
Proof. Given : S
1
S
1
we choose a path n: 1 S
1
from 1 to (1)
-1
.
Then (.. t ) (.)n(t ) is a homotopy from to a pointed map, hence is
surjective. Let H: S
1
1 S
1
be a homotopy between pointed maps; then
(.. t ) H(.. t ) H(1. t )
-1
is a pointed homotopy between the same maps, i.e.,
is injective.
If . g: X S
1
are continuous maps, then g: . (.)g(.) is again
continuous. This product of functions is compatible with homotopies and induces
the structure of an abelian group on X. S
1
|.
(2.7.5) Theorem. From (2.7.2), (2.7.4) and (2.5.3) we obtain an isomorphism J,
J : S
1
. S
1
| S
1
. S
1
|
0

1
(S
1
. 1) Z.
We call the integer J( ) = J( |) the degree of : S
1
S
1
. A standard map of
degree n is o
n
: : :
n
. A null homotopic map has degree zero.
(2.7.6) Example. A polynomial function
g: C C. g(:) = :
n
a
1
:
n-1
a
n
has a root (n _ 1).
Proof. Suppose g(:) ,= 0 for [:[ = 1. Then : S
1
S
1
, : g(:),[g(:)[
is dened. Suppose g is non-zero for [:[ _ 1. Then h(:. t ) = (t :) is a null
homotopy of . For t > 0 we have
k(:. t ) = :
n
t (a
1
:
n-1
a
2
t :
n-2
a
n
t
n-1
) = t
n
g(:,t ).
If g is non-zero for [:[ _ 1, then H(:. t ) = k(:. t ),[k(:. t )[ is a homotopy from
to o
n
. Thus if g has no root, then o
n
is null homotopic; this contradicts (2.7.5).
The classical approach to

1
(S
1
) uses topological properties of the exponential
function : R S
1
, t exp(2i t ). A lifting of n: a. b| S
1
along is a
map W : a. b| R with W = n; the value W(a) is the initial condition of the
lifting. Liftings always exist and depend continuously on the path and the initial
condition (see (2.7.8)).
(2.7.7) Proposition. Let : S
1
S
1
be given. Let J : 1 R be a lifting of
0
along . Then J(1) J(0) is the degree of .
Proof. Let g = (1)
-1
. Then (J( )) = g
0
|, by the denition of J in (2.7.5).
There exists a R such that (1) = exp(2i a) and J(0) = a. Then = J a
is a lifting of g
0
with initial condition 0. Hence J(1) J(0) = (1) (0) =
(1) = n Z. The homotopy (.. t ) (1 t )(.) t.(1) is a homotopy of
paths. Hence the loop g
0
is homotopic to s
n
. This shows (n) = g
0
|.
The next proposition will be proved in the chapter on covering spaces. It ex-
presses the fact that : R S
1
is bration.
(2.7.8) Proposition. Given a homotopy h: X 1 S
1
and an initial condition
a: X R such that a(.) = h(.. 0). Then there exists a unique homotopy
H: X 1 R such that H(.. 0) = a(.) and H = h.
(2.7.9) Example. Let n: 0. 1| S
1
be a path with n(0) = : = exp(2i a). Let
W : 0. 1| R be a lifting of n with W(0) = a. Suppose W(1) = b. Then W is,
by a linear homotopy, homotopic to t a t (b a) and hence n homotopic to
the path in normal form t : exp(2i(b a)t ).
2.7.10 The winding number. Let . C = R
2
. The map
r
x
: C\ {.] S
1
. : (: .),[: .[
is an h-equivalence and therefore S
1
. C\ {.]| S
1
. S
1
|, | r
x
| a bijec-
tion. The degree of r
x
is the winding number of with respect to .. We denote
it by W(. .). Maps
0
.
1
: S
1
C\ {.] are homotopic if and only if they have
the same winding number. If : S
1
C is given and n: 1 C a path with
(S
1
) n(1) = 0, then (.. t ) r
u(t)
is a homotopy. Therefore the winding
numbers of with respect to n(0) and n(1) are equal. The complement C\(S
1
)
decomposes into open path components, and the winding number with respect to .
is constant as long as . stays within a component.
Let u: 1 C {.] be a loop. Then there exists a unique continuous map
: S
1
C {.] such that
0
= u. The winding number of is then also
called the winding number of u, and we denote it by W(u. .).
The notion of the degree can be extended to other situations. Let h: S S
1
be
a homeomorphism and : S S any map; the degree of h h
-1
is independent
of the choice of a homeomorphism h and also called the degree J( ) of .
Problems
1. Let be a polynomial function on Cwhich has no root on S
1
. Then the number of roots
: with [:[ < 1 (counted with multiplicities) is equal to the winding number W([S
1
. 0).
What is the winding number of the function 1, with respect to 0?
2. (Properties of the degree.) J( g) = J( )J(g). A homeomorphism S
1
S
1
has
degree 1. If : S
1
S
1
has degree J( ) = 1, then there exists . S
1
such that
(.) = .. The map : : has degree 1.
Let u = exp(2i,n) be an n-th root of unity. Suppose h: S
1
S
1
satises h(u:) =
h(:). Then J(h) 0 mod n.
Let k. Zand assume that k is coprime to n. Let : S
1
S
1
satisfy the functional
equation (u
k
:) = u
(:). Then k J( ) mod n. If, conversely, this congruence is

satised with some integer J( ), then there exists a map of degree J( ) which satises
the functional equation. In particular an odd map , i.e., (:) = (:), has odd degree.
Suppose (:) = (:) for all :; then the degree of is odd. Suppose (:) = g(:)
for all :; then J( ) = J(g). Suppose J(g) 0 mod n for some n > 0; then there exists
h: S
1
S
1
such that g = h
n
.
3. Let U : 1 C be a lifting of u: 1 C
= C 0 along the covering 1 : C C
,
: exp(2i:). Then W(u. 0) = U(1) U(0).
4. Let ; : 0. 1| C
be a continuously differentiable path with initial point 1. Then

I : 0. 1| C, t
1
2i
_
)|0,I|
o z
z
is a continuously differentiable lifting of ; along 1
with initial point 0.
5. If u: 1 C {.] is a continuously differentiable loop, then W(u. .) =
1
2i
_
u
oz
z-\
.
6. Let GL
2
(R). Then the winding number of l
.
: S
1
R
2
0, . . with respect
to the origin is the sign of the determinant det().
7. Let : S
1
. X|
0
S
1
. X| be the map which forgets about the base point (pointed
homotopies versus free homotopies). Conjugate elements in the group S
1
. X|
0
have the
same image under . Hence induces a well-dened map : S
1
. X|
0
,(~) S
1
. X| from
the set of conjugacy classes. This map is injective, and surjective if X is path connected.
Thus is bijective if X is path connected and the fundamental group abelian.
2.8 Examples
The formal nature of the theoremof Seifert and van Kampen is simple, but the corre-
sponding algebra can be complicated. The setup usually leads to groups presented
by generators and relations. It may be difcult to understand a group presented in
this manner. For an introduction to this type of group theory see [122]; also [171]
and [39] are informative in this context. We report about some relevant algebra and
describe a number of examples and different applications of the fundamental group.
2.8.1 Spheres. If a space X is covered by two open simply connected subsets with
path connected intersection, then X is simply connected, since the pushout of two
trivial groups is trivial. Coverings of this type exist for the spheres S
n
for n _ 2.
Hence these spheres are simply connected.
2.8.2 Removing a point. The inclusion D
n
0 D
n
induces for n _ 3 an
isomorphism of fundamental groups; actually both groups are zero, since D
n
is
contractible and D
n
0 . S
n-1
.
Let M be a manifold of dimension n _ 3 and U M homeomorphic to D
n
under a homeomorphism that sends . to 0. Then M is the pushout of M {.]
and U. Theorem (2.6.2) implies that M {.] M induces an isomorphism of
fundamental groups.
Often we view the space S
n
as the one-point compactication R
n
L{o] of the
Euclidean space, see (2.3.2). Let 1 be a compact subset of R
n
for n _ 3. Then the
inclusion R
n
1 S
n
1 induces an isomorphism of fundamental groups.
2.8.3 Complements of spheres. Let S
n
0
= S
nn1
(R
n1
0) and S
n
1
=
S
nn1
(0 R
n1
). Then X = S
nn1
S
n
1
n
1
n
.
A homeomorphism S
n
1
n
X is (.. ,)
_
_
1 [,[
2
.. ,
_
. The space
Y = S
nn1
(S
n
0
L S
n
1
) is homeomorphic to S
n
S
n
|0. 1 via (.. ,. t )
(
_
1 t ..
_
t ,). Therefore the complement X is h-equivalent to S
n
and the com-
plement Y is h-equivalent to S
n
S
n
.
The fundamental group of S
3
({0] S
1
) is isomorphic to Z. If we view
S
3
= R
3
L{o], thenwe are consideringthe complement of the axis 7 = {(0. 0. :) [
: R] L {o]. The generator of the fundamental group is a loop that runs once
about the axis 7, represented by the standard sphere W = S
1
{0].
It is impossible to span a 2-disk with boundary W in the complement of 7,
because W represents a non-zero element in the fundamental group of the comple-
ment, see Figure 2.3. This is expressed by saying that W and 7 are linked in S
3
.
Apply the stereographic projection (2.3.2) to S
1
0 L 0 S
1
S
3
. The image
yields W L 7. The complement of W L 7 in R
3
is therefore isomorphic to the
fundamental group ZZof the torus S
1
S
1
. The reader should draw generators
of
1
(R
3
W L 7).
2.8. Examples 53
7
W
o
Figure 2.3. A standard circle in S
3
.
2.8.4 Presentation of groups by generators and relations. Let S be a set. A free
group with basis S consists of a group J(S) and a set map i : S J(S) which has
the following universal property: For each set map : S G into a group G there
exists a unique homomorphism : J(S) G such that i = . It turns out
that i is injective. Let us consider i as an inclusion and set S
-1
= {s
-1
[ s S]. A
word in the alphabet X = S lS
-1
is a sequence (.
1
. . . . . .
n
) of elements .
i
X.
The elements in J(S) are the products .
1
. . . .
n
corresponding to the words; the
neutral element belongs to the empty word; a word (.. .
-1
) also represents the
neutral element.
Let 1 be a set of words and

1 the image in J(S). Let N(1) be the normal sub-
group generated by

1. The factor group G = J(S),N(1) is the group presented
by the generators S and the relations 1. We denote this group by (S[1). It has the
following universal property: Let : S H be a set map into a group H. Assume
that for each (.
1
. . . . . .
n
) 1 the relation (.
1
) . . . (.
n
) = 1 holds in H. Then
there exists a unique homomorphism : G H such that (.) = (.) for each
. S.
Each group can be presented in the form (S [ 1) in many different ways. In
practice one uses a less formal notation. Here are a few examples.
(i) The cyclic group of order n has the presentation (a [ a
n
).
(ii) Let S = {.. ,]. Consider the word (.. .. ,
-1
. ,
-1
. ,
-1
) and 1 consisting
of this word. Then we can write G = (S[1) also in the form (.. , [ .
2
,
-3
)
or (.. , [ .
2
= ,
3
). The universal property says in this case that homomor-
phisms G H correspond bijectively to set maps : {.. ,] H such that
(.)
2
= (,)
3
.
(iii) (a. b [ ab = ba) is a presentation for the free abelian group with basis a, b.
2.8.5 Free product and pushout of groups. The sum (= coproduct) in the cate-
gory of groups is also called a free product. Let (G
}
[ J) be a family of groups.
The free product of this family consists of a group
kJ
G
k
together with a family
of homomorphisms |
}
: G
}

kJ
G
k
which have the universal property of a
sum in the category of groups. (The notation G
1
+ G
2
is used for the free product
of two groups.) Each family has a sum. Let G
}
= (S
}
[1
}
) and assume that the S
}
are disjoint. Let S = l
}
S
}
. The 1
}
are then words in the alphabet S lS
-1
. Let
1 =
_
}
1
}
. We have homomorphisms |
}
: (S
}
[1
}
) (S[1) which are induced
by S
}
S. These homomorphisms are a sum in the category of groups.
Let G and H be groups and i
1
: G G+H,
1
: H G+H be the canonical
maps which belong to the sum. Let J : 1 G, 1 : 1 H be homomorphisms
from a further group 1. Let N be the normal subgroup of G + H generated by
the elements {i
1
J(.)
1
1(.
-1
) [ . 1]. Let Q = (G + H),N and denote by
i : G Q, : H Q the composition of i
1
.
1
with the quotient map. Then
(i. ) is a pushout of (J. 1) in the category of groups. In the case that 1 and J are
inclusions (but sometimes also in the general case) one writes Q = G +
1
H.
Let S be a set and Z = Z
x
a copy of the additive group Zfor each s S. Then
the groups J(S) with basis S is also the free product
xS
Z.
2.8.6 Free products of fundamental groups. The free product
1
(X
0
) +
1
(X
1
)
arises geometrically if X
01
is simply connected.
Let X = S
1
S
1
with X
0
= Y S
1
, X
1
= S
1
Y , where Y = S
1
{1]. + =
1. Then (Y. 1) is pointed contractible; hence the inclusion of the summands S
1
X
0
. X
1
are pointed h-equivalences. One can apply (2.6.2) to the covering of X
by X
0
. X
1
. Since X
0
X
1
is pointed contractible, we see that
1
(X) is the free
product
1
(X
0
) +
1
(X
1
). Hence the inclusions of the summands S
1
S
1
S
1
yield a presentation of
1
(S
1
S
1
) as a free product
1
(S
1
) +
1
(S
1
) Z+ Z.
By induction one shows that
1
_ _
k
1
S
1
_
is the free group of rank k.
2.8.7 Plane without two points. The space R
2
{1] has as a deformation retract
the union X of the circles about 1 with radius 1,2 and the segment from 1,2
to 1,2, see Figure 2.4. (The reader should try to get an intuitive understanding of a
_
, , ,
1 1 0

u
Figure 2.4. Generators u. of
1
(R
2
1).
retraction. In order to give a formal proof, without writing down explicit formulas,
it is advisable to wait for the method of cobrations.) The fundamental group
1
(R
2
{1]. 0) is the free group Z + Z and generators are represented by two
2.8. Examples 55
small circles about 1 (of radius 1,2, say) connected linearly to the base point.
One can apply (2.6.2) to the covering of R
2
{1] by the punctured half-spaces
{(.. ,) [ . < 1,3. . ,= 1] and {(.. ,) [ 1,3 < .. . ,= 1] .
In the example in 2.8.6 one cannot apply (2.6.2) directly to the covering of
S
1
S
1
bythe twosummands, since the interiors donot cover the space. The general
method in cases like this is to rst thicken the subspaces up to h-equivalence. In
the next theorem we add a hypothesis which allows for a thickening.
(2.8.8) Theorem. Let ((X
}
. .
}
) [ J) be a family of pointed spaces with
the property: The base point .
}
has an open neighbourhood U
}
X
}
which is
pointed contractible to the base point. The inclusions of the summands induce
homomorphisms i
}
:
1
(X
}
. .
}
)
1
_ _
kJ
X
k
. .
_
. This family is a free product
of the groups
1
(X
}
. .
}
).
Proof. Let J = {1. 2]. We apply (2.6.2) to the covering X
1
U
2
. U
1
X
2
of
X
1
X
2
. The argument is as for 2.8.4. For nite J we use induction on [J[. Note
that
_
U
}

_
X
}
is an open, pointed contractible neighbourhood of the base
point.
Let nowJ be arbitrary. A path n: 1
_
X
}
has, by compactness of 1, an im-
age in
_ _
eT
X
e
_
_ _
}J~T
U
}
_
for a nite subset 1 J. This fact and the result
for 1 show that the canonical map
J
:
}J

1
(X
}
)
1
_ _
}J
X
}
_
is surjec-
tive. Each element .
}J

1
(X
}
) is contained in some
eT

1
(X
e
) for a
nite 1. Suppose . is contained in the kernel of
J
. Then a loop n: 1
_
eT
X
e
representing
T
. is null homotopic in
_
}J
X
}
and, again by compactness, null
homotopic in some larger nite wedge. The result for a nite index set now yields
. = 0.
2.8.9 Quotient groups. Let i : 1 G be a homomorphism of groups and denote
by N C G the normal subgroup generated by the image of i . Then
1

i
G
]
G,N
is a pushout in the category of groups, with the quotient map.
This situation arises geometrically in (2.6.2) if one of the spaces X
is simply
connected.
2.8.10 Attaching of a 2-cell. We start with a pushout diagram of spaces
S
1

T
J
D
2

X.
Then X is said to be obtained fromT by attaching a 2-cell via the attaching map .
(This construction will be studied in detail in the chapter on cell complexes.) Then
a suitable thickening shows that we can apply (2.6.2). Since D
2
is contractible, we
are in the situation of 2.8.9. Thus J
+
induces an isomorphism
1
(T),()
1
(X)
where () denotes the normal subgroup generated by | S
1
. T|
0
=
1
(T).
2.8.11 Attaching of a cone. Given a map : T from a path connected
space . The cone on is the space 1, 0. Let : C, a (a. 1)
denote the inclusion of into the cone. The cone is contractible, a contracting
homotopy is induced by h
t
(.. s) = (.. s(1 t )). Form a pushout
T
J
X.
Since C is contractible, J
+
1
(T),N
1
(X), where
N is the normal subgroup generated by the image of
+
.
2.8.12 Realizationof groups. We demonstrate that arbitrarygroups canbe realized
as fundamental groups. Let
: =
_
k1
S
1
_
I1
S
1
= T
be a pointed map. The inclusions of the summands yield a basis a
k

1
() and
b
I

1
(T) for the free groups and
+
(a
k
) = r
k
is a word in the b
t
I
, t Z. Let
N
I1
Z = G be the normal subgroup generated by the r
k
. Then G,N is the
group presented by generators and relations (b
I
. l 1 [ r
k
. k 1). Let C be
the cone on and dene X by a pushout diagram
C

X.
Then 2.8.11 shows
1
(X) G,N. Each group can be presented in the formG,N.
Note that X is a 2-dimensional cell complex.
2.8.13 Surfaces. The classication theory of compact connected surfaces presents
a surface as a quotient space of a regular 2n-gon, see e.g., [44, p. 75], [167], [123].
The edges are identied in pairs by a homeomorphism. The surface J is obtained
as a pushout of the type
S
1

_
n
1
S
1
D
2

J.
2.8. Examples 57
The attaching map is given in terms of the standard generators of
1
_ _
n
1
S
1
_
by
the so-called surface-word.
In order to save space we refer to [44, p. 8387] for the discussion of the
fundamental group of surfaces in general. We mention at least some results. They
will not be used in this text.
(2.8.14) Theorem. (1) The fundamental group of a closed connected orientable
surface J
of genus g _ 1 has the presentation
1
(J
) = (a
1
. b
1
. . . . . a
. b
[ a
1
b
1
a
-1
1
b
-1
1
. . . a
a
-1
b
-1
).
(2) The fundamental group of a closed connected non-orientable surface N
of
genus g has the presentation
1
(N
) = (a
1
. . . . . a
[ a
2
1
a
2
2
. . . a
2
).
(3) The fundamental group of a compact connected surface with non-empty
boundary is a free group. The number of generators is the nite number 1 (J
)
where (J
) is the so-called Euler characteristic.

(4) A simply connected surface is homeomorphic to R
2
or S
2
.
There are many different denitions of the genus. We mention a geometric
property: The genus of a closed connected orientable surface is the maximal number
g of disjointlyembeddedcircles suchthat their complement is connected. The genus
of a closed connected non-orientable surface is the maximal number g of disjointly
embedded Mbius bands such that their complement is connected. The sphere has
genus zero by the Jordan separation theorem.
Problems
1. Let S
1
R
2
0 R
3
be the standard circle. Let D = {(0. 0. t ) [ 2 _ t _ 2] and
S
2
(2) = {. R
3
[ [.[ = 2]. Then S
2
(2) L D is a deformation retract of X = R
3
S
1
.
The space X is h-equivalent to S
2
S
1
.
2. Consider the loop based at (0. 0) in the plane as shown in Figure 2.5. Determine which
_

_

_
, , ,
-1 1
Figure 2.5.
element in
1
(R
2
1) this loop represents in terms of the generators u. in 2.8.7. Deter-
mine the winding number about points in each of the six complementary regions.
3. Let (X
[ J) be a family of subspaces of X such that the interiors X
cover X.
Then each morphism in (X) is a composition of morphisms in the X
. If the intersections
X
i
X
are path connected and + X

i
X
, then
1
(X. +) is generated by loops in the
X
.
4. Let i
0
in (2.6.2) be an isomorphism. Then
1
is an isomorphism. This statement is a
general formal property of pushouts. If i
0
is surjective, then
1
is surjective.
5. Projective plane. The real projective plane 1
2
is dened as the quotient of S
2
by the
relation . ~ .. Let .
0
. .
1
. .
2
| denote the equivalence class of . = (.
0
. .
1
. .
2
). We can
also obtain 1
2
from S
1
by attaching a 2-cell
S
1

1
1
J
D
2

1
2
.
Here 1
1
= {.
0
. .
1
. 0|] 1
2
and(.
0
. .
1
) = .
0
. .
1
. 0|. The space 1
1
is homeomorphic
to S
1
via .
0
. .
1
. 0| :
2
. : = .
0
i.
1
; and corresponds to the standard map of degree
2. The map is . = (.
0
. .
1
) .
0
. .
1
.
_
1 [.[
2
|. As an application of 2.8.10 we
obtain
1
(1
2
) Z,2.
Another interpretation of the pushout: 1
2
is obtained from D
2
by identifying opposite
points of the boundary S
1
. The subspace {(.
0
. .
1
) [ [.[ _ 1,2] becomes in 1
2
a Mbius
band M. Thus 1
2
is obtainable from a Mbius band M and a 2-disk D by identication of
the boundary circles by a homeomorphism. The projective plane cannot be embedded into
R
3
, as we will prove in (18.3.7). There exist models in R
3
with self-intersections (technically,
the image of a smooth immersion.) The projective plane is a non-orientable surface.
6. Klein bottle. The Klein bottle 1 can be obtained from two Mbius bands M by an
identication of their boundary curves with a homeomorphism, 1 = M L
u^
M.
Apply the theorem of Seifert and van Kampen and obtain the presentation
1
(1) =
(a. b [ a
2
= b
2
). The elements a
2
. ab generate a free abelian subgroup of rank 2 and of
index 2 in the fundamental group. The element a
2
generates the center of this group, it is
represented by the central loop dM. The quotient by the center is isomorphic to Z,2 +Z,2.
The space M,dM is homeomorphic to the projective plane 1
2
. If we identify the central
dM to a point, we obtain a map q : 1 = M L
u^
M 1
2
1
2
. The induced map on the
fundamental group is the homomorphism onto Z,2 + Z,2.
2.9 Homotopy Groupoids
The homotopy category does not have good categorical properties. Therefore we
consider homotopy as an additional structure on the category TOP of topological
spaces. The categoryTOPwill be enriched: The set of morphisms (X. Y ) between
two objects carries the additional structure of a groupoid. The fundamental groupoid
is the special case in which X is a point.
2.9. Homotopy Groupoids 59
Recall that a category has the data: objects, morphisms, identities, and compo-
sition of morphisms. The data satisfy the axioms: composition of morphisms is
associative; identities are right and left neutral with respect to composition.
Let X and Y be topological spaces. We dene a category (X. Y ) and begin
with the data. The objects are the continuous maps X Y . A morphism from
: X Y to g: X Y is represented by a homotopy 1: . g. Two such
homotopies 1 and 1 dene the same morphism if they are homotopic relative to
X d1 with d1 = {0. 1] the boundary of 1. Let us use a second symbol J = 0. 1|
for the unit interval. This means: A map : (X 1) J Y is a homotopy
relative to Xd1, if (.. 0. t ) is independent of t and (.. 1. t ) is also independent
of t . Therefore
t
: X 1 Y. (.. s) (.. s. t ) is for each t J a homotopy
from to g. For this sort of relative homotopy one has, as before, the notion of
a product and an inverse, now with respect to the J-variable. Hence we obtain
an equivalence relation on the set of homotopies from to g. We now dene: A
morphism o : g in (X. Y ) is an equivalence class of homotopies relative
to X d1 from to g. Composition of morphisms, denoted ~, is dened by the
product of homotopies
1: . g. 1: g . h. 1| ~1| = 1 + 1| : h.
This is easily seen to be well-dened (use
t
+
J

t
). The identity of in (X. Y )
is represented by the constant homotopy k
(
: . .
The verication of the category axioms is based on the fact that a reparametriza-
tion of a homotopy does not change its class.
(2.9.1) Lemma. Let : 1 1 be a continuous map with (0) = 0 and (1) = 1.
Then 1 and 1 (id ) are homotopic relative to X d1.
Proof. (.. s. t ) = 1(.. (1 t )s t (s)) is a suitable homotopy.
(2.9.2) Proposition. The data for (X. Y ) satisfy the axioms of a category. The
category is a groupoid.
Proof. The associativity of the composition follows, because
(1 + 1) + M = 1 + (1 + M) (id ).
with dened by (t ) = 2t for t _
1
4
, (t ) = t
1
4
for
1
4
_ t _
1
2
, (t ) =
t
2
1
2
for
1
2
_ t _ 1.
Similarly, for each 1: . g the homotopies k
(
+1, 1, and 1+k
differ by
a parameter change. Therefore the constant homotopies represent the identities in
the category.
The inverse homotopy 1
-
represents an inverse of the morphism dened by 1.
Hence each morphism is an isomorphism. Proof: The assignments (.. s. t )
1(.. 2s(1t )) for 0 _ s _
1
2
and (.. s. t ) 1(.. 2(1s)(1t )) for
1
2
_ s _ 1
yield a homotopy relative to X d1 from 1 + 1
-
to the constant homotopy.
The endomorphism set of an object in a groupoid is a group with respect to
composition as group law. We thus see, from this view point, that the notion of
homotopy directly leads to algebraic objects. This fact is a general and systematic
approach to algebraic topology.
The homotopy categories of Section 2.2 have a similar enriched structure. If
we work, e.g., with pointed spaces and pointed homotopies, then we obtain for
pointed spaces X and Y a category
0
(X. Y ). The objects are pointed maps.
Morphisms are represented by pointed homotopies, and the equivalence is dened
by homotopies rel X d1 such that each
t
is a pointed homotopy.
The remainder of this section can be skipped on a rst reading. We study the
dependence of the groupoids (X. Y ) on X and Y . The formal structure of this
dependence canbe codiedinthe notionof a 2-category. Suppose given: U X
and : Y V . Composition with and yield a functor
#
=
#
(): (X. Y ) (X. V ).
which sends to and 1| to 1| and a functor
#
=
#
(): (X. Y ) (U. Y ).
which sends to and 1| to 1( id)|. They satisfy (
1
2
)
#
=
1
#
2
#
and
(
1
2
)
#
=
#
2
#
1
. These functors are compatible in the following sense:
(2.9.3) Proposition. Suppose 1: . g: X Y and 1: u . : Y 7 are
given. Then 1 1| =
#
1| ~
#
1| = g
#
1| ~ u
#
1|. Here 1 1: u .
g: X 1 7. (.. t ) 1(1(.. t ). t ).
Proof. We use the bi-homotopy 1 (1 id): X 1 1 7. Restriction to the
diagonal of 1 1 denes 1 1. Along the boundary of the square we have the
following situation.
u ug u1
1( id) 1(g id)

g 1
E
s
T
t
(s. t ) 1 1
11
g
#
1| ~u
#
1| is represented by u1 + 1(g id). If we compose the bi-homotopy
with id(X) ;, where ;(t ) = (2t. 0) for t _
1
2
and ;(t ) = (1. 2t 1) for t _
1
2
,
2.9. Homotopy Groupoids 61
we obtain u1 + 1(g id). In the same manner we obtain 1( id) + 1 if we
compose the bi-homotopy with id(.) , where (t ) = (0. 2t ) for t _
1
2
and
(t ) = (2t 1. 1) for t _
1
2
. The maps ; and are homotopic relative to d1 by a
linear homotopy in the square. They are also homotopic to the diagonal t (t. t )
of the square.
(2.9.4) Corollary. The homotopy 1 induces a natural transformation
1
#
: u
#

#
: (X. Y ) (X. 7).
The value of 1
#
at is
#
1|. The homotopy 1 induces a natural transformation
1
#
:
#
g
#
: (Y. 7) (X. 7).
The value of 1
#
at u is u
#
1|.
(2.9.5) Corollary. If u: Y 7 is an h-equivalence, then u
#
is an equivalence of
categories. Similarly in the contravariant case.
The data and assertions that we have obtained so far dene on TOP the structure
of a 2-category. In this context, the ordinary morphisms : X Y are called
1-morphisms and the morphisms 1| : . g are called 2-morphisms. The
composition ~ of 2-morphisms is called vertical composition. We also have a
horizontal composition of 2-morphisms dened as 1| 1| = 1 1|. Because
of (2.9.3) we need not dene via the diagonal homotopy; we can use instead
(2.9.3) as a denition 1| 1| =
#
1| ~
#
1| = g
#
1| ~u
#
1|.
(2.9.6) Note. Fromthis denition one veries the commutation rule of a 2-category
( ~;) ( ~) = ( ) ~(; ).
The following gure organizes the data (horizontal vertical).

[
h
[
T
u

[ ;
u
[
C
Conversely, one can derive (2.9.3) from the commutation rule (2.9.6). With the
constant homotopy k
u
of u we have
k
u
= u
#
. ; k
(
=
#
;. k
=
#
. ; k
= g
#
;
and this yields
; = (; ~k
u
) (k
~) = (; k
) ~(k
u
) = g
#
~u
#
.
In a similar manner one obtains ; =
#
~
#
;.
Chapter 3
Covering Spaces
A covering space is a locally trivial map with discrete bres. Objects of this type
can be classied by algebraic data related to the fundamental group. The reduction
of geometric properties to algebraic data is one of the aims of algebraic topology.
The main result of this chapter has some formal similarity with Galois theory.
A concise formulation of the classication states the equivalence of two cate-
gories. We denote by COV
B
the category of covering spaces of T; it is the full
subcategory of TOP
B
of spaces over T with objects the coverings of T. Under
some restrictions on the topology of T this category is equivalent to the category
TRA
B
= (T). SET| of functors (T) SET and natural transformations
between them. We call it the transport category. It is a natural idea that, when
you move from one place to another, you carry something along with you. This
transport of information is codied in moving along the bres of a map (here: of
a covering). We will show that the transport category is equivalent to something
more familiar: group actions on sets.
The second important aspect of covering space theory is the existence of a
universal covering of a space. The automorphism group of the universal covering
is the fundamental group of the space and in this manner the fundamental group
appears as a symmetry group. Moreover, the whole category of covering spaces
is obtainable by a simple construction (associated covering of bundle theory) from
the universal covering.
In this chapter we study coverings fromthe view-point of the fundamental group.
Another aspect belongs to bundle theory. In the chapter devoted to bundles we show
for instance that isomorphismclasses of n-fold coverings over a paracompact space
T correspond to homotopy classes T BS(n) into a so-called classifying space
BS(n).
3.1 Locally Trivial Maps. Covering Spaces
Let : 1 T be continuous and U T open. We assume that is surjec-
tive to avoid empty bres. A trivialization of over U is a homeomorphism
:
-1
(U) U J over U, i.e., a homeomorphism which satises pr
1
= .
This condition determines the space J up to homeomorphism, since induces a
homeomorphism of
-1
(u) with {u] J. The map is locally trivial if there
exists an open covering Uof T such that has a trivialization over each U U. A
locally trivial map is also called a bundle or bre bundle, and a local trivialization
a bundle chart. We say, is trivial over U, if there exists a bundle chart over U. If
3.1. Locally Trivial Maps. Covering Spaces 63
is locally trivial, then the set of those b T for which
-1
(b) is homeomorphic
to a xed space J is open and closed in T. Therefore it sufces for most purposes
to x the homeomorphism type of the bres. If the bres are homeomorphic to J,
we call J the typical bre. A locally trivial map is open, hence a quotient map.
A covering space or a covering
1
of T is a locally trivial map : 1 T with
discrete bres. If J is discrete (= all subsets are open and closed), then U J is
homeomorphic to the topological suml
xT
U {.]. The summands U {.] are
canonically homeomorphic to U. If :
-1
(U) U J is a trivialization, then
yields via restriction a homeomorphism of
-1
(U {.]) with U. A covering
is therefore a local homeomorphism. The summands
-1
(U {.]) = U
x
are the
sheets of the covering over U; the pre-image
-1
(U) is therefore the topological
sumof the sheets U
x
; the sheets are open in 1 and mapped homeomorphically onto
U under . If [J[ = n N, we talk about an n-fold covering. The trivial covering
with typical bre J is the projection pr : T J T. We say, U is admissible or
evenly covered if there exists a trivialization over U.
(3.1.1) Example. The exponential function : R S
1
, t exp(2i t ) is a cov-
ering with typical bre Z. For each t Rand (t ) = : we have a homeomorphism
-1
(S
1
:) =
nZ
|t n. t n 1 |t. t 1 Z.
and maps each summand homeomorphically.
(3.1.2) Proposition. Let : 1 T be a covering. Then the diagonal D of 11
is open and closed in 7 = {(.. ,) 1 1 [ (.) = (,)].
Proof. Let U
x
be an open neighbourhood of . which is mapped homeomorphically
under . Then 7 (U
x
U
x
) = W
x
is contained in D, and W
x
is an open
neighbourhood of (.. .) in 7. This shows that D is open.
Let . ,= , and (.) = (,). Let . U
x
and , U
,
be the sheets of over
the open set U T. Since . ,= ,, the intersection U
x
U
,
is empty. Hence
7 (U
x
U
,
) is an open neighbourhood of (.. ,) in 7 and disjoint to D. This
shows that also the complement 7 D is open.
Let : 1 T and : X T be maps; then J : X 1 is a lifting of
along , if J = .
(3.1.3) Proposition (Uniqueness of liftings). Let : 1 T be a covering. Let
J
0
. J
1
: X 1 be liftings of : X T. Suppose J
0
and J
1
agree somewhere.
If X is connected, then J
0
= J
1
.
Proof. (J
0
. J
1
) yield a map J : X 7. By assumption, J
-1
(D) is not empty,
and hence, by (3.1.2), open and closed. If X is connected, then J
-1
(D) = X, i.e.,
J
0
= J
1
.
1
Observe that the term covering has two quite different meanings in topology.
64 Chapter 3. Covering Spaces
(3.1.4) Proposition. Let q : 1 T 0. 1| be locally trivial with typical bre J.
Then T has an open cover Usuch that q is trivial over each set U 0. 1|, U U.
Proof. If q is trivial over U a. b| and over U b. c|, then q is trivial over
U a. c|. Two trivializations over U {b] differ by an automorphism, and this
automorphism can be extended over U b. c|. Use this extended automorphism
to change the trivialization over U b. c|, and then glue the trivializations. By
compactness of 1 there exist 0 = t
0
< t
1
< < t
n
= 1 and an open set U such
that q is trivial over U t
i
. t
i1
|.
For the classication of covering spaces we need spaces with suitable local
properties. A space X is called locally connected (locally path connected) if
for each . X and each neighbourhood U of . there exists a connected (path
connected) neighbourhood V of . which is contained in U. Both properties are
inherited by open subspaces.
(3.1.5) Proposition. The components of a locally connected space are open. The
path components of a locally path connected space Y are open and coincide with
the components.
Proof. Let 1 be the component of .. Let V be a connected neighbourhood of ..
Then 1LV is connected and therefore contained in 1. This shows that 1 is open.
Let U be a component of Y and 1 a path component of U. Then U 1 is a
union of path components, hence open. In the case that U ,= 1 we would obtain a
decomposition of U.
We see that each point in a locally path connected space has a neighbourhood
basis of open path connected sets.
(3.1.6) Remark. Let T be path connected and locally path connected. Since a
covering is a local homeomorphism, the total space 1 of a covering of T is locally
path connected. Let 1
t
be a component of 1 and
t
: 1
t
T the restriction of
. Then
t
is also a covering: The sets U, over which is trivial, can be taken as
path connected, and then a sheet over U is either contained in 1
t
or disjoint to 1
t
.
Since T is path connected, we see by path lifting (3.2.9) that
t
is surjective. By
(3.1.5), 1 is the topological sum of its components.
A left action G 1 1, (g. .) g. of a discrete group G on 1 is called
properly discontinuous if each . 1 has an open neighbourhood U such that
U gU = 0 for g ,= e. A properly discontinuous action is free. For more details
about this notion see the chapter on bundle theory, in particular (14.1.12).
A left G-principal covering consists of a covering : 1 T and a properly
discontinuous action of G on 1 such that (g.) = (.) for (g. .) G 1 and
such that the induced action on each bre is transitive.
3.1. Locally Trivial Maps. Covering Spaces 65
(3.1.7) Example. A left G-principal covering : 1 T induces a homeomor-
phism of the orbit space 1,G with T. The orbit map 1 1,G of a properly
discontinuous action is a G-principal covering.
A covering : 1 T has an automorphism group Aut(). An automorphism
is a homeomorphism : 1 1 such that = . Maps of this type are also
called deck transformations of . If is a left G-principal covering, then each
left translation l
: 1 1, . g. is an automorphism of . We thus obtain a

homomorphism l : G Aut(). Let 1 be connected. Then an automorphism
is determined by its value at a single point . 1, and (.) is a point in the bre
-1
((.)). Since G acts transitively on each bre, the map l is an isomorphism.
Thus the connected principal coverings are the connected coverings with the largest
possible automorphism group. Conversely, we can try to nd principal coverings
by studying the action of the automorphism group.
(3.1.8) Proposition. Let : 1 T be a covering.
(1) If 1 is connected, then the action of Aut() (and of each subgroup of Aut())
on 1 is properly discontinuous.
(2) Let T be locally path connected and let H be a subgroup of Aut(). Then
the map q : 1,H T induced by is a covering.
Proof. (1) Let . 1 and g Aut(). Let U be a neighbourhood of (.) which is
evenly covered, and let U
x
be a sheet over U containing .. For , U
x
gU
x
we
have (,) = (g
-1
,), since g
-1
is an automorphism. Hence , = g
-1
,, since
both elements are contained in U
x
. This shows g
-1
= id, hence U
x
gU
x
= 0
for g ,= e, and we see that the action is properly discontinuous.
(2) Let U T be open, path connected, and evenly covered. Let
-1
(U) =
_
}J
U
}
be the decomposition into the sheets over U. An element h H permutes
the sheets, since they are the path components of
-1
(U). The equivalence classes
with respect to H are therefore open in the quotient topology of 1,H and are
mapped bijectively and continuously under q. Since is open, so is q. Hence q is
trivial over U. Since T is locally path connected, it has an open covering by such
sets U.
A right G-principal covering : 1 T gives rise to associated coverings.
Let J be a set with left G-action. Denote by 1
G
J the quotient space of 1 J
under the equivalence relation (.. ) ~ (.g
-1
. g ) for . 1. J. g G.
The continuous map
T
: 1
G
J T, (.. ) (.) is a covering with typical
bre J.
A G-map [: J
1
J
2
induces a morphism of coverings
id
G
[: 1
G
J
1
1
G
J
2
. (.. ) (.. [( )).
We thus have obtained a functor associated coverings
(): G-SET COV
B
from the category G-SET of left G-sets and G-equivariant maps. We call a G-prin-
cipal covering : 1 T over the path connected space T universal if the functor
() is an equivalence of categories.
3.2 Fibre Transport. Exact Sequence
The relation of a covering space to the fundamental groupoid is obtained via path
lifting. For this purpose we now introduce the notion of a bration which will be
studied later in detail. A map : 1 T has the homotopy lifting property (HLP)
for the space X if the following holds: For each homotopy h: X1 T and each
map a: X 1 such that a(.) = hi(.), i(.) = (.. 0) there exists a homotopy
H: X 1 1 with H = h and Hi = a. We call H a lifting of h with initial
condition a. The map is called a bration if it has the HLP for all spaces.
(3.2.1) Example. A projection : T J T is a bration. Let a(.) =
(a
1
(.). a
2
(.)). The condition a = hi says a
1
(.) = h(.. 0). If we set H(.. t ) =
(h(.. t ). a
2
(.)), then H is a lifting of h with initial condition a.
(3.2.2) Theorem. A covering : 1 T is a bration.
Proof. Let the homotopy h: X 1 T and the initial condition a be given.
Since 1 is connected, a lifting with given initial condition is uniquely determined
(see (3.1.3)). Therefore it sufces to nd for each . X an open neighbourhood
V
x
such that h[V
x
1 admits a lifting with initial condition a[V
x
. By uniqueness
(3.1.3), these partial liftings combine to a well-dened continuous map.
By (3.2.3) there exists for each . T an open neighbourhood V
x
and an
n Nsuch that h maps V
x
i,n. (i 1),n| into a set U over which is trivial.
Since :
-1
(U) U is, by (3.2.1), a bration, h[V
x
i,n. (i 1),n| has a
lifting for each initial condition. Therefore we nd by induction over i a lifting of
h[V
x
0. i,n| with initial condition a[V
x
.
(3.2.3) Lemma. Let U be an open covering of T 0. 1|. For each b T there
exists an open neighbourhood V(b) of b in T and n = n(b) N such that for
0 _ i < n the set V(b) i,n. i,(n 1)| is contained in some member of U.
The fact that the lifted homotopy is uniquely determined implies that for a
covering : 1 T the diagram
1
J
]
1
e
0
1
T
J
e
0
!
1
]
T
3.2. Fibre Transport. Exact Sequence 67
is a pullback of topological spaces. Here 1
J
is the space of paths in 1 with
compact-open topology and e
0
T
(n) = n(0).
The homeomorphism (2.3.6) k : (1
n
. d1
n
) (1. 0) 1
n
(1. 0) of pairs is
used to solve the next homotopy lifting problem with a modied initial condition.
It reduces the problem to the HLP for 1
n
.
(3.2.4) Proposition. Let : 1 T have the HLP for the cube 1
n
. For each
commutative diagram
1
n
0 L d1
n
1
o
1
]
1
n
1
h
p
p
p
p
p
p
T
there exists H: 1
n
1 1 with Hi = a and H = h.
Let : 1 T be a map which has the HLP for a point and for 1. Write J
b
=
-1
(b). We associate to each path : 1 T from b to c a map
#
:
0
(J
b
)
0
(J
c
) which only depends on | (T). Let . J
b
. Choose a lifting
V : 1 1 of with V(0) = .. We set
#
.| = V(1)|. We have to show that this
assignment is well-dened. For this purpose assume given:
(1) u: 1 J
b
;
(2) h: 1 1 T a homotopy of paths from b to c;
(3) V
0
. V
1
: 1 1 liftings of h
0
, h
1
with initial points u(0). u(1).
These data yield a map a: 1 d1 L 0 1 1, dened by a(s. c) = V
t
(s) and
a(0. t ) = u(t ). The lifting H of h with initial condition a, according to (3.2.4),
yields a path t H(1. t ) in J
c
fromV
0
(1) to V
1
(1). This shows that the map
#
is
well-dened and depends only on the morphism | in the fundamental groupoid.
The rule n
#
#
= ( + n)
#
is easily veried from the denitions. Thus we have
shown:
(3.2.5) Proposition. The assignments b
0
(J
b
) and |
#
yield a functor
T
]
: (T) SET.
We call T
]
= T() the transport functor associated to .
The functor T
]
provides us with
0
(J
b
)
1
(T. b)
0
(J
b
). (.. |)
#
(.) = . |.
a right action of the fundamental group on the set
0
(J
b
). We write
0
(J. .) if .|
is chosen as base point of the set
0
(J). We use the action to dene
d
x
:
1
(T. b)
0
(J
b
. .). | . |.
The map d
x
is
1
(T. b)-equivariant, i.e., d
x
+ n| = (d
x
|) n|.
Recall that a sequence

T
C of pointed maps is exact at T if the image

of equals the kernel
-1
(+) of . Similarly for longer sequences. In this context
a group is pointed by its neutral element.
(3.2.6) Theorem. Let (.) = b and i : J
b
1. The sequence
1
(J
b
. .)
i

1
(1. .)
]

1
(T. b)
J
\

0
(J
b
. .)
i

0
(1. .)
]

0
(T. b)
is exact.
Proof. It is easily veried from the denitions that the composition of two maps is
the constant map. We consider the remaining four cases: kernel image.
Let u|
1
(1. .) and h: 1 1 T a null homotopy of u. Consider
the lifting problem for h with initial condition a: 1 0 L d1 1 1 with
a(s. 0) = u(s) and a(c. t ) = .. The lifting H of h is then a homotopy of loops
from u to a loop in the image of i .
Let d
x
| = .|. This means: There exists a lifting V of from. to V(1) .|.
Choose a path U : 1 J
b
from V(1) to .. Then V + U is a loop in 1, and its
class maps under
1
() to |, since U is constant.
Let
0
(i ),| = .|. There exists a path n: 1 1 from. to ,. The projection
= n is a loop and d
x
| = ,|, by denition of d
x
.
Let
0
(),| = b|. Thus there exists a path : 1 T from (,) to .. Let
V : 1 1 be a lifting of with initial point ,. Then V(1) J
b
, and V shows
0
(i )(V(1)| = ,|.
There is more algebraic structure in the sequence.
(3.2.7) Proposition. The pre-images of elements under d
x
are the left cosets of
1
(T. b) with respect to
+
1
(1. .). The pre-images of
0
(i ) are the orbits of the
1
(T. b)-action on
0
(J
b
. .).
Proof. Let d
x
u| = d
x
|. Choose liftings U, V of u, which start in ., and let
W : 1 J
b
be a path from U(1) to V(1). Then U + W + V
-
is a loop in 1, and
(U + W + V
-
) + . u, i.e., the elements u| and | are contained in the same
left coset. Conversely, elements in the same coset have the same image under d
x
.
(A similar assertion holds for right cosets.)
Suppose
0
(i )a| =
0
(i )b|. Then there exists a path n: 1 1 from a to b.
Set = n. Then a| | = b|. Conversely, elements in the same orbit have
equal image under
0
(i ).
We can apply (3.2.6) to a covering : 1 T. The bres are discrete. There-
fore
1
(J
b
. .) is the trivial group 1. Hence
+
:
1
(1. .)
1
(T. b) is injective.
We state (3.2.6) for a covering:
3.2. Fibre Transport. Exact Sequence 69
(3.2.8) Proposition. Let : 1 T be a covering over a path connected space T.
Then the sequence
1
1
(1. .)
]
1
(T. b)
J
\
0
(J
b
. .)
i
0
(1. .)
is exact and i
+
is surjective. (The sets J
b
=
0
(J
b
) and
0
(1) have . as base point,
and i : J
b
1.) Thus 1 is path connected if and only if
1
(T. b) acts transitively
on J
b
. The isotropy group of . J
b
is the image of
+
:
1
(1. .)
1
(T. b).
(3.2.9) Proposition (Path lifting). Let : 1 T be a covering. Let n: 1 T
be a path which begins at (e) = n(0). Then there exists a unique lifting of n
which begins in e. Two paths in 1 which start in the same point are homotopic if
and only if their images in T are homotopic.
Proof. The existence of the lifting follows from (3.2.2), applied to a point X, and
the uniqueness holds by (3.1.3).
Let h: 1 1 T be a homotopy of paths and H: 1 1 1 a lifting
of h. Since t H(c. t ) are continuous maps into a discrete bre, they are constant
(c = 0. 1). Hence H is a homotopy of paths.
Let u
0
. u
1
: 1 1 be paths which start in ., and suppose that u
0
and u
1
are homotopic. If we lift a homotopy between them with constant initial condition,
then the result is a homotopy between u
0
and u
1
.
Let : 1 T be a right G-principal covering. Each bre J
b
carries a free
right transitive G-action. From the construction of the transport functor it is im-
mediate that the bre transport T
]
n| : J
b
J
c
is G-equivariant. The left action
(a. .) a . = a
#
(.) of
b
= (T)(b. b) on J
b
commutes with the right
G-action; we say in this case that J
b
is a (
b
. G)-set. Fix . J
b
. For each
a
b
there exists a unique ;
x
(a) G such that a . = . ;
x
(a), since
the action of G is free and transitive. The assignment a ;
x
(a) is a homo-
morphism ;
x
:
b
G. Since
1
(T. b) is the opposite group to
b
, we set
x
(a) = ;
x
(a)
-1
. Then
x
:
1
(T. b) G is a homomorphism. Recall the
map d
x
:
1
(T. (.)) J
b
. If we compose it with the bijection j
x
: G J
b
,
g .g, we obtain j
x
x
= d
x
. Then (3.2.8) yields:
(3.2.10) Proposition. Let : 1 T be a right G-principal covering with path
connected total space. Then the sequence of groups and homomorphisms
1
1
(1. .)
]
1
(T. (.))

\
G 1
is exact ( for each . 1). The image of
+
is a normal subgroup. The space
1 is simply connected if and only if
x
is an isomorphism. Thus, if 1 is simply
connected, then G is isomorphic to the fundamental group of T.
If we apply this to the exponential covering R S
1
, a Z-principal covering,
we again obtain
1
(S
1
) Z.
The transport functor T
]
has an additional property, it is locally trivial in the
following sense. Let be trivial over U, and let b. c U. Then T
]
n| : J
b
J
c
is independent of the path n: 1 U from b to c. This is due to the fact that lifts
of paths inside U stay within a sheet over U.
3.3 Classication of Coverings
Let TRA
B
= (T). SET| denote the category of functors (T) SET (objects)
and natural transformations between them (morphisms). We call this category the
transport category.
Let : 1 T be a covering. We have constructed the associated transport
functor T
]
= T(): (T) SET. For a morphism: q between coverings
the restrictions
b
:
-1
(b) q
-1
(b) of to the bres yield a natural transfor-
mation T(): T() T(q) between the corresponding transport functors. So we
have obtained a functor
T : COV
B
TRA
B
.
A path connected space T is called a transport space if T is an equivalence of
categories.
The main theorem of this section gives conditions under which the transport
functor T is an equivalence.
(3.3.1) Note. Let : 1 T be a covering with simply connected 1. Then T is
path connected. If is trivial over U, then two paths in U between the same points
are homotopic in T.
Proof. The space T is path connected, since is assumed to be surjective and 1 is
path connected. Let u
0
. u
1
be paths in U between the same points. By (3.2.9) they
have liftings
0
.
1
which connect the same points. Since 1 is simply connected,
0
.
1
are homotopic in 1 and hence u
0
. u
1
are homotopic in T.
A set U T is transport-simple if two paths in U between the same points are
homotopic within T. A space T is semi-locally simply connected if it has an open
covering by transport-simple sets. We have just seen that this condition is implied
by the existence of a simply connected covering. We call T transport-local, if T
is path connected, locally path connected and semi-locally simply connected.
(3.3.2) Theorem(Classication I). Let T be path connected, locally path connected
and semi-locally simply connected. Then T is a transport space, i.e., T is an
equivalence of categories.
3.3. Classication of Coverings 71
Proof. We begin by constructing a functor
X: TRA
B
COV
B
in the opposite direction. Let : (T) SET be a functor. We construct an
associated covering = (): X() T. As a set, X() =

bB
(b), and
() sends (b) to b. Let U be the set of open, path connected and transport-
simple subsets of T. We dene bundle charts over sets U U. For b U we
dene
U,b
: U (b)
-1
(U). (u. :) (n):
with some path n in U fromb to u. By our assumption on U, the map
U,b
is well-
dened, i.e., the choice of n does not matter. By construction,
U,b
is bijective.
We claim: There exists a unique topology on X() such that the
U,b
of this type
are homeomorphisms onto open subsets. By general principles of gluing, we have
to verify that the transition maps
-1
V,c

U,b
: (U V ) (b) (U V ) (c)
are homeomorphisms. Let . U V and let W U V be an open, path
connected neighbourhood of .. Let u
x
be a path from b to . inside U, and
x
a
path from c to . inside V . Then for all , W
-1
V,c

U,b
(,. :) =
-1
V,c

U,b
(.. :).
because in order to dene
U,b
(,. :) we can take the product of u
x
with a path n
,
in W from . to ,, and similarly for
V,c
, so that the contribution of the piece n
,
cancels. This shows that the second component of
-1
V,c

U,b
is on W (b)
independent of . W. The continuity of the transition map is a consequence.
If :
1

2
is a natural transformation, then the morphism
X(): X(
1
) X(
2
). .
1
(b) (.)
2
(b).
induced by , is continuous with respect to the topologies just constructed and
hence a morphism of coverings. The continuity of X() follows from the fact, that
bundle charts
U,b
for X(
1
) and X(
2
) transform X() into
id (b): U
1
(b) U
2
(b).
This nishes the construction of the functor X: TRA
B
COV
B
.
We nowshowthat the functors T and X are mutually inverse equivalences of cat-
egories, i.e., that XT and TX are naturally isomorphic to the identity functor. From
our constructions we see immediately a canonical homeomorphismj: X(T())
1() over T for each covering : 1() T, namely set-theoretically the iden-
tity. We have to show that j is continuous. Let J
b
X(T()). Let W be a
neighbourhood of in 1(). Then there exists an open neighbourhood V W of
in 1() such that V is path connected and U = (V ) U. Let b = (). Then
we have the bundle charts
U,b
, and W =
U,b
(U ) is a neighbourhood of in
X(T()). From the construction of
U,
we see that j(W) V . This shows the
continuity at . It is easily veried that the homeomorphisms j constitute a natural
transformation. Conversely, we have to verify that the transport functor of ()
is . Using the bundle charts
U,b
this is rst veried for paths in U. But each
morphism in (T) is a composition of morphisms represented by paths in such
sets U.
Fix b T. A canonical functor is the Hom-functor (b. ) of the category
(T). Let
b
: 1
b
T denote the associated covering. We still assume that T is
transport-local. The automorphismgroup
b
= (b. b) of b in (T), the opposite
fundamental group
1
(T. b), acts on 1
b
brewise fromthe right by composition of
morphisms. The action is free and transitive on each bre. Via our bundle charts it
is easily veried that the action on 1
b
is continuous. Thus
b
is a right
b
-principal
covering. From (3.2.8) we see that 1
b
is simply connected. Thus we have shown:
(3.3.3) Theorem. The canonical covering
b
: 1
b
T associated to the Hom-
functor (b. ) has a simply connected total space. The right action of (b. b) on
the bres by composition of morphisms is the structure of a right principal covering
on
b
.
Problems
1. Let S be the pseudo-circle. The space S is simply connected. But S has non-trivial
connected principal coverings. They can be obtained by a pullback along suitable maps
S S
1
from Z,n-principal or Z-principal coverings of S
1
. In this sense S behaves like
S
1
. We see that certain local properties of T are necessary in order that T is an equivalence.
2. Let : T C be a continuous map. The pullback : X T of a covering q : Y C
along is a covering. Pulling back morphisms yields a functor
= COV( ): COV
C

COV
!
. The map induces a functor ( ): (T) (C), and composition with
functors (C) SET yields a functor TRA( ) : TRA
C
TRA
!
. These functors are
compatible T
!
COV( ) = TRA( ) T
C
: COV
C
TRA
!
.
3.4 Connected Groupoids
In this section the space T is assumed to be path connected.
A functor (T) SET is an algebraic object. The category of these functors
has an equivalent description in terms of more familiar algebraic objects, namely
group actions. We explain this equivalence.
Let be a connected groupoid (i.e. there exists at least one morphism between
any two objects) with object set T, e.g., = (T) for a path connected space T.
3.4. Connected Groupoids 73
Let (.. ,) denote the set of morphisms . , and =
b
= (b. b), the
automorphism group of b with respect to composition of morphisms. A functor
J : SET has an associated set J(b) with left
b
-action
b
J(b) J(b). (. .) J()(.).
A natural transformation : J G yields a map (b): J(b) G(b) which is
b
-equivariant. In this manner we obtain a functor
c
b
: . SET|
b
-SET
from the functor category of functors SET into the category of left
b
-sets
and equivariant maps.
We construct a functor j
b
in the opposite direction. So let be a
b
-set. The
Hom-functor (b. ) is a functor into the right
b
-sets, namely
b
acts on (b. .)
by composition of morphisms. These data yield the functor () = (b. )
t
.
(Here again
t
T denotes the quotient of T by the equivalence relation
(ag. b) ~ (a. gb). (a. g. b) T for left -sets and right -sets T.) A
b
-map : T induces a natural transformation ( ): () (T). This
nishes the denition of j
b
.
(3.4.1) Proposition. The functors c
b
and j
b
are mutually inverse equivalences of
categories.
Proof. The composition c
b
j
b
associates to a
b
-set the
b
-set (b. b)
t
,
with
b
-action g (. :). The isomorphisms
|
: (b. b)
t
. (. :) :
form a natural equivalence | : c
b
j
b
. Id.
The composition j
b
c
b
associates to a functor J : SET the functor
(b. )
t
J(b). The maps
T
(.): (b. .) J(b) J(.). (. :) J( ):
form a natural transformation, i.e., a morphism
T
: j
b
c
b
(J) J in . SET|.
Since is a connected groupoid, the
T
(.) are bijective, and therefore constitute
an isomorphism in the functor category. The
T
are a natural equivalence
: j
b
c
b
. Id.
In our previous notation TRA
B
= (T). SET|. From (3.3.2) and (3.4.1)
we obtain for each transport-local space an equivalence of categories COV
B

b
-SET, the composition of the transport functor T with c
b
. It associates to a
covering : 1 T the
b
-set J
b
. The inverse equivalence associates to a
b
-set
the covering
X(j
b
) =
xB
(T)(b. .)
t
T.
It is the covering 1
b
t
T associated to the
b
-principal covering (3.3.3).
Let : 1 T be a right G-principal covering with path connected T. Then
we have the functors
G- SET
(])
COV
B
T
TRA
B
t
b
:

b
- SET.
The composition associates to a G-set J the
b
-set J
b
G
J, where the
b
-action
is induced from the left
b
-action on J
b
.
Now suppose in addition that 1 is simply connected. Then we have a bijection
T
x
: J J
b

G
J, : .. :| for a xed . J
b
as well as the isomorphism
;
x
:
b
G, see (3.2.10). The relation ;
x
(a) : = a .. :| holds. So if we view
G-sets via ;
x
as
b
-sets, then the above composition of functors is the identity.
Thus we have shown:
(3.4.2) Proposition. Let : 1 T be a simply connected G-principal covering.
Then () is an equivalence of categories if and only if T is an equivalence of
categories.
(3.4.3) Theorem. The following properties of T are equivalent:
(1) T is a transport space, i.e., T is an equivalence of categories.
(2) T has a universal right G-principal covering : 1 T with simply con-
nected total space 1.
Proof. (1) = (2). Since c
b
T is an equivalence of categories, each object of
b
- SET is isomorphic to an object in the image of c
b
T . Thus there exists a
covering : 1 T such that its
b
-set J
b
is isomorphic to the
b
-set
b
. By
another property of an equivalence of categories, the morphisms correspond
under c
b
T bijectively to the
b
-maps J
b
J
b
. The
b
-morphisms
b

b
are the right translations by elements of
b
. Thus
b
acts simply and transitively
on 1. From (3.2.8) we see that 1 is simply connected.
The left action of the automorphism group Aut() on 1 is properly discon-
tinuous and the induced action on each bre is transitive. If we rewrite this as
a right action of the opposite group G, we obtain a right G-principal covering.
Proposition (3.4.2) now says that is universal.
(2) =(1) is a consequence of (3.4.2).
From a geometric view point the interesting coverings are those with connected
total space.
Let : 1 T be a universal right G-principal covering. A left G-set is
the disjoint sum of its orbits. We have a corresponding sum decomposition of the
total space 1
G
into the sum of 1
G
C, where C runs through the orbits
of . An orbit is a transitive G-set and isomorphic to a homogeneous set G,H
for some subgroup H of G. The homeomorphism 1
G
G,H 1,H shows
3.4. Connected Groupoids 75
that the summands 1
G
C are path connected. The action of H on 1 is properly
discontinuous and therefore 1 1,H an H-principal covering. Also the induced
map
1
: 1,H T is a covering.
The category of homogeneous G-sets and G-maps is the orbit category Or(G)
of G. The sets G,1 and G,1are isomorphic if and only if the subgroups 1 and 1
are conjugate in G. The isotropy groups of G,H are conjugate to H. The inclusion
of the subcategory Or(G) into the category of transitive G-sets is an equivalence.
Let : 1 T be a universal right G-principal covering with simply con-
nected 1. Then the functor () induces an equivalence of Or(G) with the cate-
gory of connected coverings of T. Each covering is thus isomorphic to a covering
of the form
1
: 1,H T for a subgroup H of G.
We x :
-1
(b) 1 and obtain an isomorphism
z
: G
1
(T. b). It
sends g G to the loop n
| where n
: 1 1 is a path from : to :g
-1
.
Let q : X T be a connected covering. We know that the induced homomor-
phism
+
:
1
(X. .)
1
(T. b) is injective. The image is called the characteris-
tic subgroup C(. .) of with respect to .. Let u: 1 X be a path from . to
,
-1
(b). Then n = u is a loop and C(. ,) = n|C(. .)n|
-1
, thus dif-
ferent base points in
-1
(b) yield conjugate characteristic subgroups. Conversely,
each subgroup conjugate to C(. .) arises this way.
We apply this to the covering
1
with : = :H 1,H. Then
(
1
)
+
(
1
(1,H. :) =
z
(H) = C(
1
. :).
We collect the results in the next theorem.
(3.4.4) Theorem (Classication II). Let T be a transport space. The category of
connected coverings of T is equivalent to the orbit category Or(
1
(T. b)). The
isomorphism class of a connected covering q : X T corresponds under this
equivalence to the isomorphism class of
1
(T. b),C(q. .) for any .
-1
(b).
The isomorphism class of a connected covering is determined by the conjugacy
class of its characteristic subgroup.
Problems
1. The automorphism group of
!
: 1,H T is NH,H, where NH denotes the normal-
izer of H in
1
(T. b). The covering is a principal covering (also called regular covering),
if and only if H is a normal subgroup of
1
(T. b).
2. The connected coverings of S
1
are, up to isomorphism, the maps
n
: : :
n
for n N
and : R S
1
, t exp(2i t ). These coverings are principal coverings.
3. Let T be a contractible space. Is the identity id: T T a universal G-principal covering
for the trivial group G?
3.5 Existence of Liftings
The following theorem(3.5.2) is interesting and important, because it asserts the ex-
istence of liftings under only the necessary algebraic conditions on the fundamental
groups.
(3.5.1) Lemma. Let n
0
and n
1
be paths in 1 beginning in .. Let u
i
= n
i
.
Then n
0
(1) = n
1
(1) if and only if u
0
(1) = u
1
(1) and u
0
+ u
-
1
| is contained in
1
(1. .).
Proof. If n
0
(1) = n
1
(1), then
+
n
0
+ n
-
1
| = u
0
+ u
-
1
|. Conversely: We lift
u
0
+u
-
1
with initial point .. Since u
0
+u
-
1
|
+
1
(1. .) there exists a loop which
is homotopic to u
0
+ u
-
1
, and which has a lifting with initial point .. By (3.2.9),
u
0
+ u
-
1
itself has a lifting as a loop. Therefore u
0
and u
1
have liftings with initial
point . and the same end point. These liftings are then necessarily n
0
and n
1
.
(3.5.2) Theorem. Let : 1 T be a covering. Suppose 7 is path connected
and locally path connected. Let : 7 T be a map with (:) = (.). Then
there exists a lifting : 7 1 of with (:) = . if and only if
+
1
(7. :) is
contained in
+
1
(1. .).
Proof. If a lifting exists, then the inclusion of groups holds by functoriality of
1
.
Suppose
+
1
(7. :)
+
1
(1. .). We begin by constructing as a set map.
Then we show its continuity.
Let :
0
7. There exists a path n from : to :
0
. Let : 1 1 be a lifting of
n starting in .. We want to dene by (:) = (1). Let n
1
be another path
from : to :
0
and
1
a lifting of n
1
starting in .. Then
(1) = n(1) = (:
0
) = n
1
(1) =
1
(1):
moreover
+
-
1
| =
+
n + n
-
1
|
+
1
(1. .).
By (3.5.1) we have (1) =
1
(1); this shows that is well-dened if we set
(:) = (1).
Continuity of . Let U be an open neighbourhood of (:
0
), such that is trivial
over (U) = V , and let : U V have the inverse homeomorphismq : V U.
Let W be a path connected neighbourhood of :
0
such that (W) V . We claim
(W) U. Let :
1
W and let n
1
be a path in W from :
0
to :
1
. Then n + n
1
is a path from : to :
1
, and
1
= + q n
1
a path with
1
= (n + n
1
) and
1
(0) = .. Thus
1
(1) U.
(3.5.3) Theorem. Let X be a topological group with neutral element . and let
: 1 X be a covering with path connected and locally path connected 1. For
each e
-1
(.) there exists a unique group structure on 1 which makes 1 into a
topological group with neutral element e and such that is a homomorphism.
3.5. Existence of Liftings 77
Proof. Construction of a group structure on 1. Let m: X X X be the group
multiplication. We try to nd M: 1 1 1 as a lift m( ) along with
M(e. e) = e. This can be done, by (3.5.2), if m
+
( )
+
1
(1 1)
+
1
(1).
This inclusion holds, since (using (2.7.3))
m
+
()
+
(n
1
. n
2
)| =n
1
n
2
| =n
1
+n
2
| =(n
1
+n
2
)| =
+
n
1
+n
2
|.
Fromthe uniqueness of liftings one shows that M is associative. In a similar manner
we see that (passage to) the inverse in X has a lifting to 1, and uniqueness of liftings
shows that the result is an inverse for the structure M.
Awell-known result of HermannWeyl is that a compact, connected, semi-simple
Lie group has a nite simply connected covering. See [29, V.7] about fundamental
groups of compact Lie groups. The group O(n) has two different two-fold coverings
which are non-trivial over SO(n). They are distinguished by the property that the
elements over the reections at hyperplanes have order 2 or 4 (n _ 1). We will see
that
1
(SO(n)) Z,2 for n _ 3. The corresponding simply connected covering
groups are the spinor groups Spin(n); see e.g., [29, I.6].
We repeat an earlier result in a different context. We do not assume that T has
a universal covering.
(3.5.4) Proposition. Let T be path connected and locally path connected. Cover-
ings
i
: (X
i
. .
i
) (T. b) with path connected total space are isomorphic if and
only if their characteristic subgroups are conjugate in
1
(T. b).
Proof. Since C(
1
. .
1
) and C(
2
. .
2
) are conjugate we can change the base point
.
2
such that the groups are equal. By (3.5.2), there exist morphisms
1
: (X
1
. .
1
)
(X
2
. .
2
) and
2
: (X
2
. .
2
) (X
1
. .
1
), and since
2
1
(.
1
) = .
1
.
1
2
(.
2
) = .
2
both compositions are the identity.
By functoriality of
1
we see that isomorphic coverings have conjugate charac-
teristic subgroups.
Problems
1. We have given a direct proof of (3.5.2), although it can also be derived from our previous
classication results. The existence of a lift is equivalent to the existence of a section in
the covering which is obtained by pullback along . The
1
(7. :)-action on the bre 1
b
of the pullback is obtained from the
1
(T. b)-action via
:
1
(7. :)
1
(T. b). The
existence of a section is equivalent to 1
b
having a xed point under the
1
(7. :)-action.
If the inclusion of groups holds as in the statement of the theorem, then a xed point exists
because the image of
is the isotropy group of the

1
(T. b)-action.
2. Let : 1 T be a covering with path connected and locally path connected total space.
The following are equivalent: (1) Aut() acts transitively on each bre of . (2) Aut()
acts transitively on some bre of . (3) The characteristic subgroup is normal in
1
(T. b).
(4) is an Aut()-principal covering.
3.6 The Universal Covering
We collect some of our results for the standard situation that T is path connected,
locally path connected and semi-locally simply connected space. Let us now call a
covering : 1 T a universal covering if 1 is simply connected.
(3.6.1) Theorem (Universal covering). Let T be as above.
(1) There exists up to isomorphism a unique universal covering : 1 T.
(2) The action of the automorphism group Aut() on 1 furnishes with the
structure of a left Aut()-principal covering.
(3) The group Aut() is isomorphic to
1
(T. b). Given .
-1
(b), an isomor-
phism |
x
: Aut()
1
(T. b) is obtained, if we assign to Aut() the
class of the loop n for a path n from . to (.).
(4) The space 1
b
is simply connected.
Proof. (1) Existence is shown in (3.3.3). Since T is locally path connected, the
total space of each covering has the same property. Let
i
: 1
i
T be simply
connected coverings with base points .
i

-1
i
(b). By (3.5.2), there exist mor-
phisms :
1

2
and :
2

1
such that (.
1
) = .
2
and (.
2
) = .
1
. By
uniqueness of liftings, and are the identity, i.e., and are isomorphisms.
This shows uniqueness.
(2) By (3.1.8), the action of Aut() on 1 is properly discontinuous. As in (1) one
shows that Aut() acts transitively on each bre of . The map Aut()\1 T,
induced by , is therefore a homeomorphism. Since 1 Aut()\1 is a principal
covering, so is .
(3) Since 1 is simply connected, there exists a unique homotopy class of paths
n from . to (.). Since . and (.) are contained in the same bre, n is a loop.
Therefore |
x
is well-dened. If we lift a loop u based at b to a path n beginning
in ., then there exists Aut() such that (.) = n(1). Hence |
x
is surjective.
Two paths starting in . have the same end point if and only if their images in T are
homotopic. Hence |
x
is injective. If is a path from . to (.) and n a path from
. to (.), then + n is a path from . to (.). Hence |
x
is a homomorphism.
(4) is shown in (3.3.3).
(3.6.2) Theorem (Classication III). Suppose that T has a universal covering
: 1 T. Then is a
1
(T. b)-principal covering. Each connected cover-
ing of T is isomorphic to a covering of the form 1,H T, H
1
(T. b) a
subgroup. This covering has H as a characteristic subgroup. Two such cover-
ings are isomorphic if and only if the corresponding subgroups of
1
(T. b) are
conjugate.
3.6. The Universal Covering 79
Problems
1. The product
c
1
S
1
is not semi-locally simply connected.
2. Is the product of a countably innite number of the universal covering of S
1
a covering?
3. Identify in S
1
the open upper and the open lower hemi-sphere to a point. The resulting
space X has four points. Show
1
(X) Z. Does X have a universal covering?
4. The quotient map : R
n
R
n
,Z
n
is a universal covering. The map q : R
n
T
n
,
(.
) (exp 2i.
) is a universal covering of the n-dimensional torus T

n
= S
1
S
1
.
Let : T
n
T
n
be a continuous automorphism, and let J : R
n
R
n
be a lifting of q
along q with J(0) = 0. The assignments . J(.) J(,) and . J(. ,) are liftings
of the same map with the same value for . = 0. Hence J(. ,) = J(.) J(,). From
this relation one deduces that J is a linear map. Since J(Z
n
) Z
n
, the map J is given
by a matrix GL
n
(Z). Conversely, each matrix in GL
n
(Z) gives us an automorphism
of T
n
. The group of continuous automorphisms of T
n
is therefore isomorphic to GL
n
(Z).
5. Classify the 2-fold coverings of S
1
S
1
and of S
1
S
1
S
1
. (Note that a subgroup of
index 2 is normal.)
6. The k-fold (k N) coverings of S
1
S
1
correspond to isomorphism classes of =
1
(S
1
S
1
) = (u) +()-sets of cardinality k. An action of on {1. . . . . k] is determined
by the action of u and , and these actions can be arbitrary permutations of {1. . . . . k].
Figure 3.1. The 3-fold regular coverings of S
1
S
1
.
Hence these actions correspond bijectively to the elements of S
k
S
k
(S
k
the symmetric
group). A bijection of {1. . . . . k] is an isomorphism of the actions corresponding to (u. )
and (u
/
.
/
) if and only if
-1
u
/
= u.
-1
/
= . The isomorphism classes of k-fold
coverings correspond therefore to the orbit set of the action
S
k
(S
k
S
k
) S
k
S
k
. (. u. ) (u
-1
.
-1
).
Consider the case k = 3 and S
3
= (. T [
3
= 1. T
2
= 1. TT
-1
=
-1
]. The
three conjugacy classes are represented by 1. . T. We can normalize the rst component
of each orbit correspondingly. If we x u, then the centralizer 7(u) of u acts on the second
component. We have 7(1) = S
3
, 7(T) = {1. T], and 7() = {1. .
2
]. This yields the
following representing pairs for the orbits:
(1. 1). (1. )cn. (1. T).
(. 1)cn. (. )cn. (.
2
)cn. (. T)c.
(T. 1). (T. )c. (T. T). (T. T)c.
The transitive actions (which yield connected coverings) have the addition c, the normal
subgroups (which yield regular coverings) have the addition n.
Draw gures for the connected coverings. For this purpose study the restrictions of the
coverings to the two summands S
1
; note that under restriction a connected covering may
become disconnected. Over each summand one has a 3-fold covering of S
1
; there are three
of them.
7. Classify the regular 4-fold coverings of S
1
S
1
.
8. The Klein bottle has three 2-fold connected coverings. One of them is a torus, the other
two are Klein bottles.
9. Let X be path connected and set Y = X 1,X d1. Show
1
(Y ) Z. Show that Y
has a simply connected universal Z-principal covering. Is Y always locally path connected?
10. Construct a transport space which is not locally path connected.
11. The space R
n
with two origins is obtained from R
n
R
n
by identifying . ,= 0 in the
rst summand with the same element in the second summand. Let M be the line with two
origins. Construct a universal covering of M and determine
1
(M). What can you say
about
1
of R
n
with two origins for n > 1?
12. Make the fundamental groupoid (T) into a topological groupoid with object space T.
(Hypothesis (3.6.1). Use (14.1.17).)
13. Let X be a compact Hausdorff space and H(X) the group of homeomorphism. Then
H(X) together with the CO-topology is a topological group and H(X) X X, (. .)
(.) a continuous group action.
14. The space C(S
1
. S
1
) with CO-topology becomes a topological group under pointwise
multiplication of maps.
15. There are two continuous homomorphisms e : C(S
1
. S
1
) S
1
, (1) and
J : C(S
1
. S
1
) Z, degree( ). Let M
0
(S
1
) be the kernel of (e. J). Let fur-
ther
n
: S
1
S
1
, : :
n
. The homomorphism s : S
1
Z C(S
1
. S
1
), (. n)
n
is continuous. The map
M
0
(S
1
) (S
1
Z) C(S
1
. S
1
). (. (. n)) s(. n)
is an isomorphism of topological groups. The space M
0
(S
1
) is isomorphic to the space V
of continuous functions : R R with (0) = 0 and (. 2) = (.) or, equivalently,
to the space of continuous functions : S
1
R with (1) = 0. The space V carries the
sup-norm and the induced CO-topology.
16. Let M(S
1
) be the group of homeomorphisms S
1
S
1
of degree 1 with CO-topology.
Each z S
1
yields a homeomorphism
: : z.. In this way S

1
becomes a subgroup
of M(S
1
). Let M
1
(S
1
) be the subgroup of homeomorphisms with (1) = 1. Then
S
1
M
1
(S
1
) M(S
1
). (z. h)
h
is a homomorphism. The space M
1
(S
1
) is homeomorphic tothe space H of homeomorphism
: 0. 1| 0. 1| with (0) = 0. The space H is contractible; a contraction is
I
(.) =
(1 t )(.) t.. Therefore the inclusion S
1
M(S
1
) is an h-equivalence. The space
H(S
1
) of homeomorphisms of S
1
is h-equivalent to O(2).
Chapter 4
Elementary Homotopy Theory
Further analysis and applications of the homotopy notion require a certain amount
of formal consideration. We deal with several related topics.
(1) The construction of auxiliary spaces from the basic homotopy cylinder
X 1: mapping cylinders, mapping cones, suspensions; and dual construc-
tions based on the path space X
J
. These elementary constructions are
related to the general problem of dening homotopy limits and homotopy
colimits.
(2) Natural group structures on Hom-functors in TOP
0
. By category theory they
arise from group and cogroup objects in this category. But we mainly work
with the explicit constructions: suspension and loop space.
(3) Exact sequences involving homotopy functors based on exact sequences
among pointed spaces (space level). These so-called cobre and bre se-
quences are a fundamental contribution of D. Puppe to homotopy theory
[155]. The exact sequences have a three-periodic structure, and it has by now
become clear that data of this type are an important structure in categories
with (formal) homotopy (triangulated categories).
As an application, we use a theorem about homotopy equivalences of mapping
cylinders to prove a gluing theorem for homotopy equivalences. The reader may
have seen partitions of unity. In homotopy theory they are used to reduce homotopy
colimits to ordinary colimits. Here we treat the simplest case: pushouts.
4.1 The Mapping Cylinder
Let : X Y be a map. We construct the mapping cylinder 7 = 7( ) of
via the pushout
X X
id (
( i
0
,i
1
)
X Y
( },J )
7( ) = X 1 Y,(.) ~ (.. 1).

X 1
o
7( )
J(,) = ,. (.) = (.. 0).
Here i
t
(.) = (.. t ). Since (i
0
. i
1
) is a closed embedding, the maps (. J ), and J
are closed embeddings. We also have the projection q : 7( ) Y , (.. t ) (.),
, ,. The relations q = and qJ = id hold. We denote elements in 7( ) by
82 Chapter 4. Elementary Homotopy Theory
their representatives in X 1 Y .
X
0 1
Y
.
,
,
(.)
1
The map Jq is homotopic to the identity relative to Y . The homotopy is the
identity on Y and contracts 1 relative 1 to 1.
We thus have a decomposition of into a closed embedding J and a homotopy
equivalence q. From the pushout property we see:
Continuous maps : 7( ) T correspond bijectively to pairs h: X1 T
and : Y T such that h(.. 1) = (.).
In the following we consider 7( ) as a space under X Y via the embedding
(inclusion) (. J ). We now study homotopy commutative diagrams
X
(
X
t
(
/
Y
t
together with homotopies :
t
. . In the case that the diagram is commu-
tative, the pair (. ) is a morphism from to
t
in the category of arrows in TOP.
We consider the data (. . ) as a generalized morphism. These data induce a
map = 7(. . ): 7( ) 7(
t
) dened by
(,) = (,). , Y. (.. s) =
_
((.). 2s). . X. s _ 1,2.
2x-1
(.). . X. s _ 1,2.
The diagram
X Y

7( )
7(,,)
X
t
Y
t
7(
t
)
is commutative. The composition of two such morphisms between mapping
cylinders is homotopic to a morphism of the same type. Suppose we are given
tt
: X
tt
Y
tt
,
t
: X
t
X
tt
,
t
: Y
t
Y
tt
, and a homotopy
t
:
tt
t
.
t
t
.
These data yield a composed homotopy
t
:
tt
t
.
t
dened by
(
t
)
t
=
_
t
2t
. t _ 1,2.
t

2t-1
. t _ 1,2.
(This is the product of the homotopies
t
t
and
t
t
.)
4.1. The Mapping Cylinder 83
(4.1.1) Lemma. There exists a homotopy
7(
t
.
t
.
t
) 7(. . ) . 7(
t
.
t
.
t
)
which is constant on X Y .
Proof. Both maps coincide on Y and differ on X1 by a parameter transformation
of 1.
We also change and by a homotopy. Suppose given homotopies
t
: X
X
t
and T
t
: Y Y
t
and a homotopy I
t
:
t
t
. T
t
. We assume, of course, that
I
t
= (I
t
x
) is continuous on X1 1. We get a homotopy 7(
t
. T
t
. I
t
): 7( )
7(
t
) which equals
t
T
t
: X Y X
t
Y
t
on these subspaces.
We use the fact that
t
. T
t
. induce a homotopy I
t
.
(4.1.2) Lemma. Suppose
t
. T
t
with
0
= . T
0
= and :
t
0
. T
0
are
given.Then there exists I
t
with I
0
= .
Proof. One applies a retraction X 1 1 X (d1 1 L 1 0) to the map
; : X(d1 1 L1 0) Y
t
dened by;(.. s. 0) = (.. s), ;(.. 0. t ) =
t
t
(.)
and ;(.. 1. t ) = T
t
(.).
Suppose now that X
tt
= X, Y
tt
= Y ,
tt
= and
t
. id,
t
. id,

t
.
t
t
. We choose homotopies
t
:
t
. id. T
t
:
t
. id.
t
:
t
.
t
t
.
As before, we have the composition =
t
. We use (4.1.2) to nd a homotopy
I
t
with I
0
= and I
t
:
t
. T
t
. Let I
1
-
be the inverse homotopy of I
1
.
Let C = 7(1
A
. 1
Y
. I
1
-
) 7(
t
.
t
.
t
): 7(
t
) 7( ); this morphism restricts
to
t
t
: X
t
Y
t
X Y .
(4.1.3) Proposition. There exists a homotopy from C 7(. . ) to the identity
which equals ((k +
t
) + k (k + T
t
) + k) on X Y ; here k denotes a constant
homotopy.
Proof. By (4.1.1) there exists a homotopy relative to X Y of the composition in
question to 7(1
A
. 1
Y
. I
1
-
) 7(
t
.
t
. ). By (4.1.2) we have a further homo-
topy to 7(1
A
. 1
Y
. I
1
-
) 7(1
A
. 1
Y
. I
1
), which equals
t
T
t
on X Y , and then
by (4.1.1) a homotopy to 7(1
A
. 1
Y
. I
1
-
I
1
), which is constant on X Y . The
homotopy I
1
-
I
1
: . is homotopic relative to X d1 to the constant homo-
topy k
(
of . We thus have an induced homotopy relative X Y to 7(1
A
. 1
Y
. k
(
)
and nally a homotopy to the identity (Problems 1 and 2).
(4.1.4) Theorem. Suppose and are homotopy equivalences. Then the map
7(. . ) is a homotopy equivalence.
Proof. The morphism C in (4.1.3) has a right homotopy inverse. We can apply
(4.1.1) and (4.1.3) to C and see that C also has a left homotopy inverse. Hence
C is a homotopy equivalence. From C 7(. . ) . id we now conclude that
7(. . ) is a homotopy equivalence.
Problems
1. Suppose and are homotopic relative to X d1. Then 7(. . ) and 7(. . ) are
homotopic relative to X Y .
2. In the case that
/
= we have the map 7(. ) : 7( ) 7(
/
) induced by
id . Let k be the constant homotopy. Then 7(. . k) . 7(. ) relative to X Y .
3. Let | denote the morphism in (X. Y
/
) represented by . We think of (. . |)
as a morphism from to . The composition is dened by (
/
.
/
.
/
|) (. . |) =
(
/
.
/
.
/
|). Show that we obtain in this manner a well-dened category. (This
denition works in any 2-category.)
4.2 The Double Mapping Cylinder
Given a pair of maps : T and g: C. The double mapping cylinder
7(. g) = 7(T
(

C) is the quotient of T 1 C with respect to

the relations (a) ~ (a. 0) and (a. 1) ~ g(a) for each a . We can also dene
it via a pushout

( i
0
,i
1
)
T C
( }
0
,}
1
)
1

7(. g).
The map (
0
.
1
) is a closed embedding. In the case that = id(), we can
identify 7(id(). g) = 7(g). We can also glue 7( ) and 7(g) along the common
subspace and obtain essentially 7(. g) (up to 1 L
0
1 1). A commutative
diagram
T
C
;
T
t

t
(
/
C
t
induces 7(. . ;): 7(. g) 7(
t
. g
t
), the quotient of id ;. We can
also generalize to an h-commutative diagram as in the previous section.
(4.2.1) Theorem. Suppose , , ; are h-equivalences. Then 7(. . ;) is an
h-equivalence.
4.2. The Double Mapping Cylinder 85
In order to use the results about the mapping cylinder, we present 7(. g),
up to canonical homeomorphism, also as the pushout of
: 7( ) and
B
: 7(g). Here the subspace 7( ) corresponds to the image of T
0. 1,2| in 7(. g) and 7(g) to the image of 1,2. 1| C. We view 7(. g)
as a space under T C. If we are given homotopies
B
:
t
. ,
C
: g
t
. ;g, we obtain an induced map
= 7(. .
B
) L
7(. ;.
C
): 7(. g) 7(
t
. g
t
)
which extends ;.
(4.2.2) Theorem. Let be an h-equivalence with h-inverse
t
and suppose and ;
have left h-inverses
t
. ;
t
. Choose homotopies
t
:
t
. id, T
t
:
t
. id,
C
t
: ;
t
; . id. Then there exists C: 7(
t
. g
t
) 7(. g) and a homotopy from
C to the identity which extends ((k + T
t
) + k (k +
t
) + k (k + C
t
) + k).
Proof. The hypotheses imply
t
t
. and ;
t
g
t
. g. We can apply (4.1.3)
and nd left homotopy inverses C
B
of 7(. .
B
) and C
C
of 7(. ;.
C
). Then
C = C
B
L
C
C
has the desired properties.
Theorem(4.2.1) is nowa consequence of (4.2.2). The reasoning is as for (4.1.4).
In general, the ordinary pushout of a pair of maps . g does not have good homo-
topy properties. One cannot expect to have a pushout in the homotopy category. A
pushout is a colimit, in the terminology of category theory. In homotopy theory one
replaces colimits by so-called homotopy colimits. We discuss this in the simplest
case of pushouts.
Given a diagram
X
0
(

(
-
X
-
}
-

X
(1)
and a homotopy h:
-
-
.
. We obtain an induced map : 7(

-
.
) X
which is the quotient of (
-
. h.
): X
-
X
0
1 X
X. We dene:
The diagram (1) together with the homotopy h is called a homotopy pushout or
homotopy cocartesian if the map is a homotopy equivalence. This denition is
in particular important if the diagram is commutative and h the constant homotopy.
Suppose we have inclusions
: X
0
X
and
: X
X such that X =
X
-
LX
. In the case that the interiors X
cover X, the space X is a pushout in the

category TOP. In many cases it is also the homotopy pushout; the next proposition
is implied by (4.2.4) and (4.2.5).
(4.2.3) Proposition. Suppose the covering X
of X is numerable (dened below).

Then X is the homotopy pushout of
: X
0
X
.
For the proof we rst compare 7(
-
.
) with the subspace N(X

-
. X
) =
X
-
0 L X
0
1 L X
1 of X 1. We have a canonical bijective map

: 7(
-
.
) N(X
-
. X
). Both spaces have a canonical projection to X

(denoted
7
.
1
), and is a map over X with respect to these projections.
(4.2.4) Lemma. The map is an h-equivalence over X and under X
.
Proof. Let ; : 1 1 be dened by ;(t ) = 0 for t _ 1,3, ;(t ) = 1 for t _ 2,3
and ;(t ) = 3t 1 for 1,3 _ t _ 2,3. We dene : N(X
-
X
) 7(
-
.
)
as id(X
0
) ; on X
0
1 and the identity otherwise. Homotopies . id and
. id are induced by a linear homotopy in the 1-coordinate. The reader should
verify that and the homotopies are continuous.
The covering X
of X is numerable if the projection

1
has a section. A
section o is determined by its second component s : X 0. 1|, and a function of
this type denes a section if and only if X X
-
s
-1
(0). X X
s
-1
(1).
(4.2.5) Lemma. Suppose
1
has a section o. Then
1
is shrinkable.
Proof. A homotopy o
1
. id over X is given by a linear homotopy in the
1-coordinate.
(4.2.6) Corollary. Suppose the covering X
is numerable. Then
7
is shrinkable.
(4.2.7) Theorem. Let (X. X
) and (Y. Y
) be numerable coverings. Suppose that

J : X Y is a map with J(X
) Y
. Assume that the induced partial maps

J
: X
andJ
0
: X
0
Y
0
are h-equivalences. ThenJ is anh-equivalence.
Proof. This is a consequence of (4.2.1) and (4.2.6).
The double mapping cylinder of the projections X X Y Y is called the
join X = Y of X and Y . It is the quotient space of X 1 Y under the relations
(.. 0. ,) ~ (.. 0. ,
t
) and (.. 1. ,) ~ (.
t
. 1. ,). Intuitively it says that each point
of X is connected with each point of Y by a unit interval. The reader should verify
S
n
= S
n
S
nn1
. One can also think of the join as CX Y L
AY
X CY
where CX denotes the cone on X.
4.3 Suspension. Homotopy Groups
We work with pointed spaces. Each object in the homotopy groupoid
0
(X. Y )
for TOP
0
has an automorphism group. We describe the automorphism group of the
constant map in a different manner.
4.3. Suspension. Homotopy Groups 87
A map 1: X 1 Y is a pointed homotopy from the constant map to itself
if and only if it sends the subspace X d1 L{.] 1 to the base point , of Y . The
quotient space
X = X 1,(X d1 L {.] 1)
is called the suspension of the pointed space (X. .). The base point of the suspen-
sion is the set which we identied to a point.
d
d
d
d
d
d
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
X
0
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
,
,
0
1
2
1
X
q
+ +
A homotopy 1: X 1 Y from the constant map to itself thus corresponds
to a pointed map

1: X Y , and homotopies relative to X d1 correspond to
pointed homotopies X Y . This leads us to the homotopy set X. Y |
0
. This
set carries a group structure (written additively) which is dened for representing
maps by
g: (.. t )
_
(.. 2t ). t _
1
2
g(.. 2t 1).
1
2
_ t.
(Again we consider the group opposite to the categorically dened group.) The
inverse of | is represented by (.. t ) (.. 1 t ). For this denition we do not
need the categorical considerations, but we have veried the group axioms.
If : X Y is a pointed map, then id(1) is compatible with passing to
the suspensions and induces : X Y , (.. t ) ((.). t ). In this manner
the suspension becomes a functor : TOP
0
TOP
0
. This functor is compatible
with homotopies: a pointed homotopy H
t
induces a pointed homotopy (H
t
).
There exists a canonical homeomorphism 1
kI
,d1
kI
= 1
k
,d1
k
. 1
I
,d1
I
which is the identity on representing elements in 1
kI
= 1
k
1
I
. We have for
each pointed space X canonical homeomorphisms
(X . 1
k
,d1
k
) . 1
I
,d1
I
X . 1
kI
,d1
kI
.
I
(
k
X)
kI
X.
We dene the k-fold suspension by
k
X = X . (1
k
,d1
k
). Note that
n
X is
canonically homeomorphic to X 1
n
,X d1
n
L{.] d1
n
. In the homotopy set
k
X. Y |
0
we have k composition laws, depending on which of the 1-coordinates
we use:
(
i
g)(.. t ) =
_
(.. t
1
. . . . . t
i-1
. 2t
i
. t
i1
. . . . ). t
i
_
1
2
.
g(.. t
1
. . . . . t
i-1
. 2t
i
1. t
i1
. . . . ).
1
2
_ t
i
.
We show in a moment that all these group structures coincide and that they are
abelian (n _ 2). For the purpose of the proof one veries directly the commutation
rule (unravel the denitions)
(a
1
b)
2
(c
1
J) = (a
2
c)
1
(b
2
J).
a b
c J
t
2
]
t
1
(4.3.1) Proposition. Suppose the set M carries two composition laws
1
and
2
with neutral elements e
i
. Suppose further that the commutation rule holds.
Then
1
=
2
= , e
1
= e
2
= e, and the composition is associative and
commutative.
Proof. The chain of equalities
a = a
2
e
2
= (a
1
e
1
)
2
(e
1
1
e
2
) = (a
2
e
1
)
1
(e
1
2
e
2
)
= (a
2
e
1
)
1
e
1
= a
2
e
1
shows that e
1
is a right unit for
2
. In a similar manner one shows that e
1
is a left
unit and that e
2
is a left and right unit for
1
. Therefore e
1
= e
1
2
e
2
= e
2
. The
equalities a
2
b = (a
1
e)
2
(e
1
b) = (a
2
e)
1
(e
2
b) = a
1
b show
1
=
2
= . From b c = (e b) (c e) = (e c) (b e) = c b
we obtain the commutativity. Finally a (b c) = (a e) (b c) =
(a b) (e c) = (a b) c shows associativity.
The suspension induces a map
+
: . Y |
0
. Y |
0
, | |, also
called suspension. If = X, then
+
is a homomorphism, because the addition
in X. Y |
0
is transformed by
+
into
1
.
Suppose X = S
0
= {e
1
] with base point e
1
. We have a canonical homeo-
morphism
1
n
,d1
n

n
S
0
S
0
1
n
,S
0
d1
n
L e
1
1
n
which sends . 1
n
to (e
1
. .).
The classical homotopy groups of a pointed space are important algebraic in-
variants. The n-th homotopy group is
n
(X) =
n
(X. .) = 1
n
,d1
n
. X|
0
= (1
n
. d1
n
). (X. .)|. n _ 1.
These groups are abelian for n _ 2. We can use each of the n coordinates to dene
the group structure.
4.4. Loop Space 89
4.4 Loop Space
We now dualize the concepts of the previous section. Let (Y. ,) be a pointed space.
The loop space CY of Y is the subspace of the path space Y
J
(with compact-open
topology) consisting of the loops in Y with base point ,, i.e.,
CY = {n Y
J
[ n(0) = n(1) = ,].
The constant loop k is the base point. A pointed map : Y 7 induces a pointed
map C : CY C7, n n. This yields the functor C: TOP
0
TOP
0
. It
is compatible with homotopies: Apointed homotopy H
t
yields a pointed homotopy
CH
t
. We can also dene the loop space as the space of pointed maps J
0
(1,d1. Y ).
The quotient map : 1 1,d1 induces Y
]
: Y
J{JJ
Y
J
and a homeomorphism
J
0
(1,d1. Y ) CY of the corresponding subspaces.
(4.4.1) Proposition. The product of loops denes a multiplication
m: CY CY CY. (u. ) u + .
It has the following properties:
(1) m is continuous.
(2) The maps u k + u and u u + k are pointed homotopic to the identity.
(3) m(m id) and m(id m) are pointed homotopic.
(4) The maps u u+u
-
and u u
-
+u are pointed homotopic to the constant
map.
Proof. (1) By (2.4.3) it sufces to prove continuity of the adjoint map
CY CY 1 Y. (u. . t ) (u + )(t ).
This map equals on the closed subspace CY CY 0.
1
2
| the evaluation (u. . t )
u(2t ) and is therefore continuous.
(2) Let h
t
: 1 1, s (1 t ) min(2s. 1) t . Then CY 1 CY ,
(u. t ) uh
t
is a homotopy from u u + k to the identity. Continuity is again
proved by passing to the adjoint.
(3) and (4) are proved in a similar manner; universal formulae for associativity
and the inverse do the job.
The loop product induces the m-sum on X. CY |
0
|
n
g| = m ( g) J|
with the diagonal J = (id. id): X X X. The functors and C are adjoint,
see (2.4.10). We compose a map X Y with the quotient map X 1 X.
The adjoint X Y
J
has an image contained in CY . In this manner we obtain a
bijection between morphisms X Y and X CY in TOP
0
. Moreover, this
adjunction induces a bijection X. Y |
0
X. CY |
0
, see (2.4.11). It transforms
the j-sum (see the next section) into the m-sum, and it is natural in the variables
X and Y .
In X. CY |
0
we have two composition laws
and
n
. They coincide and
are commutative. We prove this in a formal context in the next section.
4.5 Groups and Cogroups
We work in the category TOP
0
. Amonoid in h-TOP
0
is a pointed space M together
with a pointed map (multiplication) m: M M M such that . (+. .) and
. (.. +) are pointed homotopic to the identity. Spaces with this structure
are called Hopf spaces or H-spaces, in honour of H. Hopf [91]. In X. M|
0
we have the composition law
n
, dened as above for M = CY ; the constant
map represents the neutral element; and . M|
0
is a contravariant functor into the
category of monoids. Amonoid is a set together with a composition lawwith neutral
element. An H-space is associative if m(m id) . m(id m) and commutative
if m . mt with the interchange t(.. ,) = (,. .). An inverse for an H-space
is a map | : M M such that m(| id)J and m(id |)J are homotopic to the
constant map (J the diagonal). An associative H-space with inverse is a group
object in h- TOP
0
. By a general principle we have spelled out the denition in
TOP
0
. The axioms of a group are satised up to homotopy. A homomorphism (up
to homotopy) between H-spaces (M. m) and (N. n) is a map z: M N such that
n(z z) . zm. A subtle point in this context is the problem of coherence, e.g.,
can a homotopy-associative H-space be h-equivalent to a strictly associative one
(by a homomorphism up to homotopy)?
The loop space (C(X). m) is a group object in h- TOP
0
.
One can try other algebraic notions up to homotopy. Let (M. m) be an as-
sociative H-space and X a space. A left action of M on X in h- TOP
0
is a map
r : M X X such that r(mid) . r(id r) and . r(+. .) is homotopic to
the identity.
A comonoid in h-TOP
0
is a pointed space C together with a pointed map
(comultiplication) j: C C C such that pr
1
jand pr
2
jare pointed homotopic
to the identity. In C. Y |
0
we have the composition law
dened as
|
g| = ( g) j|
with the codiagonal (also called the folding map) = (id. id): Y Y Y . The
functor C. |
0
is a covariant functor into the category of monoids. The comultipli-
cation is coassociative up to homotopy if (id(C) j) j and (jid(C)) j are
pointed homotopic; it is cocommutative up to homotopy if j and tj are pointed
homotopic, with t : C C C C the interchange map. Let (C. j) and (D. v) be
4.5. Groups and Cogroups 91
monoids in h-TOP
0
; a cohomomorphism up to homotopy : C D is a pointed
map such that ( )j and v are pointed homotopic. A coinverse | : C C for
the comonoid C is a map such that (id |)j and (| id)j are both pointed ho-
motopic to the constant map. A coassociative comonoid with coinverse in h-TOP
0
is called a cogroup in h-TOP
0
. Let (C. j) be a coassociative comonoid and Y a
space. A left coaction of C on Y (up to homotopy) is a map j: Y C Y such
that (id j)j . (j id)j and pr
Y
j . id.
The suspension X is such a cogroup. We dene the comultiplication
j: X X X. j = i
1
i
2
as the sum of the two injections i
1
. i
2
: X X X. Explicitly, j(.. t ) =
(.. 2t ) in the rst summand for t _
1
2
, and j(.. t ) = (.. 2t 1) in the second
summand for
1
2
_ t .
d
d
d
d
d
d
d
d
d
d
d
d
d
d
d
d
d
d
....................
1
2
0
1
2
1
1
2
The previously dened group structure on X. Y |
0
is the j-sum.
(4.5.1) Proposition. Let (C. j) be a comonoid and (M. m) a monoid in h-TOP
0
.
The composition laws
and
n
in C. M|
0
coincide and are associative and
commutative.
Proof. We work in h-TOP
0
, as we should; thus morphisms are pointed homotopy
classes. We have the projections
k
: M M M and the injections i
I
: C
CC. Given : CC MM we set
kI
=
k
i
I
. Thenm =
1

n
and j = i
1
i
2
. From these relations we derive the commutation rule
(m )j = (
11

12
)
n
(
21

22
).
m(j) = (
11
n

21
)
(
12
n

22
).
Now apply (4.3.1).
Problems
1. Let X be a pointed space and suppose that the Hom-functor . X|
0
takes values in the
category of monoids. Then X carries, up to homotopy, a unique H-space structure m which
induces the monoid structures on . X|
0
as
m
. There is a similar result for Hom-functors
C. |
0
and comonoid structures on C.
2. Let S(k) = 1
k
,d1
k
. We have canonical homeomorphisms
C
k
(Y ) = J
0
(S(k). Y ) J((1
k
. d1
k
). (Y. +)) and C
k
C
l
(Y ) C
kl
(Y ).
3. The space J((1. 0). (Y. +)) Y
1
is pointed contractible.
4. Let J
2
(1. X) = {(u. ) X
1
X
1
[ u(1) = (0)]. The map j
2
: J
2
(1. X)
J(1. X), (u. ) u + is continuous.
5. Let J
3
(1. X) = {(u. . n) [ u(1) = (0), (1) = n(0)]. The two maps
j
3
. j
/
3
: J
3
(1. X) J(1. X). (u. . n) (u + ) + n. u + ( + n)
are homotopic over X X where the projection onto X X is given by (u. . n)
(u(0). n(1)).
6. Verify the homeomorphism J
0
(1,d1. Y ) CY .
4.6 The Cobre Sequence
A pointed map : (X. +) (Y. +) induces a pointed set map
+
: Y. T|
0
X. T|
0
. | |.
The kernel of
+
consists of the classes | such that is pointed null homotopic.
We work in the category TOP
0
and often omit pointed in the sequel. A homotopy
set Y. T|
0
is pointed by the constant map. A base point is often denoted by +.
A sequence

T
C of pointed set maps is exact if () =

-1
(+). A
sequence U
(
V
W in TOP
0
is called h-coexact if for each T the sequence
U. T|
0
V. T|
0
(
W. T|
0
is exact. If we apply this to id(W), we see that g is null homotopic.

A pointed homotopy X 1 T sends + 1 to the base point. Therefore
we use the cylinder X1 = X 1, + 1 in TOP
0
together with the embeddings
i
t
: X X1, . (.. t ) and the projection : X1 X, (.. t ) ., and we
consider morphisms X1 Y in TOP
0
as homotopies in TOP
0
.
The (pointed) cone CX over X is now dened as CX = X 1,X 0 L+ 1
with base point the identied set. The inclusion i
A
1
= i
1
: X CX, . (.. 1)
is an embedding. The maps h: CX T correspond to the homotopies of the
constant map to hi
1
(by composition with the projection X1 CX).
4.6. The Cobre Sequence 93
The mapping cone of is dened as C( ) = CX Y,(.. 1) ~ (.), or,
more formally, via a pushout
X
(
i
1
Y
(
1
CX
}
C( ).
r
r
r
r
r
r
Y CX
,
,
.
.
.
.
.
.
.
.
.
.
.
(.)
0 1
We denote the points of C( ) by their representing elements in X 1 Y . The
inclusionY CXY induces an embedding
1
: Y C( ), and CX CXY
induces : CX C( ). The pushout property says: The pairs : Y T,
h: CX T with = hi
1
, i.e., the pairs of and null homotopies of ,
correspond to maps : C( ) T with = h. If | is contained in the kernel
of
+
, then there exists : C( ) T with
1
= , i.e., | is contained in the
image of
+
1
. Moreover,
1
: X C( ) is null homotopic with null homotopy .
This shows that the sequence X
(
Y
(
1
C( ) is h-coexact.
We iterate the passage from to
1
and obtain the h-coexact sequence
X
(
Y
(
1
C( )
(
2
C(
1
)
(
3
C(
2
)
(
4
. . .
.
The further investigations replace the iterated mapping cones with homotopy equiv-
alent spaces which are more appealing. This uses the suspension. It will be impor-
tant that the suspension of a space arises in several ways as a quotient space; certain
canonical bijections have to be proved to be homeomorphisms.
In the next diagram the left squares are pushout squares and . ( ). q( ) are
quotient maps. The right vertical maps are homeomorphisms, see Problem 1. Now
X CX,i
1
X, by the identity on representatives. Therefore we view , ( ),
and q( ) as morphisms to X.
X
i
1
CX
]
CX,i
1
X
= X
Y
(
1
i
1
C( )
](( )
(
2
C( ),
1
Y
= X
CY
}
1
C(
1
)
q(( )
C(
1
),
1
CY
= X
(4.6.1) Note. The quotient map q( ) is a homotopy equivalence.
Proof. We dene a homotopy h
t
of the identity of C(
1
) which contracts CY along
the cone lines to the cone point and drags CX correspondingly
h
t
(.. s) =
_
(.. (1 t )s). (1 t )s _ 1.
((.). 2 (1 t )s). (1 t )s _ 1.
h
t
(,. s) = (,. (1 t )s).
In order to verify continuity, one checks that the denition is compatible with the
equivalence relation needed to dene C(
1
). The end h
1
of the homotopy has the
form s( ) q( ) with s( ): X C(
1
), (.. s) h
1
(.. s). The composition
q( ) s( ): (.. s) (.. min(2s. 1)) is also homotopic to the identity, as we
know from the discussion of the suspension. This shows that s( ) is h-inverse to
q( ).
We treat the next step in the same manner:
C( )
(
2
](( )
H
H
H
H
H
H
H
H
H
H
C(
1
)
(
3
q(( )
]((
1
)
H
H
H
H
H
H
H
H
H
H
C(
2
)
q((
1
)
X
(( )t
_ _ _ _
Y
with an h-equivalence q(
1
). Let | : X X, (.. t ) (.. 1t ) be the inverse
of the cogroup X. The last diagram is not commutative if we add the morphism
to it. Rather the following holds:
(4.6.2) Note. ( ) | q( ) . (
1
).
Proof. By (4.6.1) it sufces to study the composition with s( ). We know already
that (
1
)s( ): (.. s) ((.). min(1. 2(1s)) is homotopic to | : (.. s)
((.). 1 s).
4.6. The Cobre Sequence 95
An h-coexact sequence remains h-coexact if we replace some of its spaces by
h-equivalent ones. Since the homeomorphism | does not destroy exactness of a
sequence, we obtain from the preceding discussion that the sequence
X
(
Y
(
1
C( )
](( )
X
(
Y
is h-coexact.
We can continue this coexact sequence if we apply the procedure above to
instead of . The next step is then ( )
1
: Y C( ). But it turns out that we
can also use the suspension of the original map (
1
): Y C( ) in order to
continue with an h-coexact sequence. This is due to the next lemma.
(4.6.3) Lemma. There exists a homeomorphism t
1
: C( ) C( ) which
satises t
1
( )
1
= (
1
).
Proof. CX and CX are both quotients of X 1 1. Interchange of 1-
coordinates induces a homeomorphism t : CX CX which satises t
i
A
1
= (i
A
1
). We insert this into the pushout diagrams for C( ) and C( )
and obtain an induced t
1
. We use that a pushout in TOP
0
becomes a pushout again
if we apply (use C--adjunction).
We now continue in this manner and obtain altogether an innite h-coexact
sequence.
(4.6.4) Theorem. The sequence
X
(
Y
(
1
C( )
](( )
X
(
Y
(
1
C( )
](( )
2
X
2
(
. . .
is h-coexact. We call it the Puppe-sequence or the cobre sequence of ([155]).
The functor . T|
0
applied to the Puppe-sequence yields an exact sequence of
pointed sets; it consists from the fourth place onwards of groups and homomor-
phisms and from the seventh place onwards of abelian groups. See [49] for an
introduction to some other aspects of the cobre sequence.
The derivation of the cobre sequence uses only formal properties of the ho-
motopy notion. There exist several generalizations in an axiomatic context; for an
introduction see [69], [101], [18].
Let : X Y be a pointed map. Dene j: C( ) X C( ) by
j(.. t ) = ((.. 2t ). +) for 2t _ 1, j(.. t ) = (+. (.. 2t 1)) for 2t _ 1, and
j(,) = ,. This map is called the h-coaction of the h-cogroup X on C( ). This
terminology is justied by the next proposition.
r
r
r
r
rr
C( )
0
1
2
1
.
.
.
.
.
.
.
.
.
.
.
r
r
r
r
rr
C( )
d
d
d
d
d
d
X
(4.6.5) Proposition. The map j induces a left action
X. T|
0
C( ). T|
0
X C( ). T|
0

C( ). T|
0
. (. ) .
This action satises
1
( )
+
2
= ( )
+
(
1
2
). Moreover,
+
1
(
1
) =
+
1
(
2
)
if and only if there exists such that
1
=
2
. Thus
+
1
induces an injective
map of the orbits of the action.
Proof. That j
+
satises the axioms of a group action is proved as for the group
axioms involving X. Also the property involving ( ) is proved in the same
manner. It remains to verify the last statement.
Assume that . g: C( ) T are maps which become homotopic when re-
stricted to Y . Consider the subspaces C
0
= {(.. t ) [ 2t _ 1] C( ) and
C
1
= {(.. t ) [ 2t _ 1] L Y C( ). These inclusions are cobrations (see the
next chapter). Therefore we can change and g within their homotopy classes
such that [C
0
is constant and g[C
1
= [C
1
. Then g is constant on C
0
C
1
.
Therefore there exists h: X C( ) Y such that hj = g and h[C( ) = .
Let k = h[X. Then k| | = g|. Conversely, j (a. b) and b have the same
restriction to Y .
Problems
1. Let the left square in the next diagram be a pushout with an embedding and hence an
embedding J. Then J induces a homeomorphism

J of the quotient spaces.
X
)
X,
T
J
Y
q
Y,T
2. The map ( )
: X. T|
0
C( ). T|
0
induces an injective map of the left (or right)
cosets of X. T|
0
modulo the subgroup Im( )
.
4.7. The Fibre Sequence 97
4.7 The Fibre Sequence
The investigations in this section are dual to those of the preceding section. For
the purpose of this section we describe homotopies T 1 Y in a dual (adjoint)
form T Y
J
as maps into function spaces. A pointed map : (X. +) (Y. +)
induces a pointed set map
+
: T. X|
0
T. Y |
0
. | |.
A sequence U
(
V
W in TOP
0
is called h-exact, if for each T the sequence
T. U|
0
(

T. V |
0

T. W|
0
is exact. If we apply this to id(U), we see that g is null homotopic.
We need the dual form of the cone. Let J(Y ) = {n Y
J
[ n(0) = +] be the
space of paths which start in the base point of Y , with the constant path k
+
(t ) = +
as base point, and evaluation e
1
: JY Y , n n(1). Via adjunction we have
J
0
(T. JY ) J
0
(CT. Y ). The pointed maps h: T JY correspond to pointed
homotopies from the constant map to e
1
h, if we pass from h to the adjoint map
T 1 Y . We dene J( ) via a pullback
J( )
q
(
1
JY
e
1
J( ) = {(.. n) X JY [ (.) = n(1)]

X
(
1
(.. n) = .. q(.. n) = n.
The maps : T J( ) correspond to pairs =
1
: T X together with
the null homotopies q: T JY of . This shows that J( )
(
1
X
(
Y is
h-exact.
We now iterate the passage from to
1
. . .
(
4
J(
2
)
(
3
J(
1
)
(
2
J( )
(
1
X
(
Y
and show that J(
1
) and J(
2
) can be replaced, up to h-equivalence, by CY and
CX. We begin with the remark that
(
1
)
-1
(+) = {(.. n) [ n(0) = +. n(1) = (.). . = +]
can be identied with CY , via n (+. n). Let i( ): CY J( ) be the
associated inclusion of this bre of
1
. The space (by its pullback denition)
J(
1
) = {(.. n. ) [ n(0) = +. n(1) = (.). . = (1). (0) = +]
can be replaced by the homeomorphic space
J(
1
) = {(n. ) JY JX [ n(1) = (1)].
Then
2
becomes
2
: J(
1
) J( ), (n. ) ((1). n). The map
( ): CY J(
1
). n (n. k
+
)
satises
2
( ) = i( ).
(4.7.1) Note. The injection ( ) is a homotopy equivalence.
Proof. We construct a homotopy h
t
of the identity of J(
1
) which shrinks the path
to its beginning point and drags behind the path n correspondingly. We write
h
t
(n. ) = (h
1
t
(n. ). h
2
t
(n. )) JY JX and dene
h
1
t
(s) =
_
n(s(1 t )). s(1 t ) _ 1.
(2 (1 t )s). s(1 t ) _ 1.
h
2
t
(s) = (s(1 t )).
The end h
1
of the homotopy has the form ( ) r( ) with
r( ): J(
1
) CY. (n. ) n + ( )
-
.
The relation (r( ) ( ))(n) = n+k
-
+
shows that also r( ) ( ) is homotopic
to the identity. The continuity of h
t
is proved by passing to the adjoint maps.
We treat the next step in a similar manner.
J(
2
)
(
3
J(
1
)
CX
}((
1
)
i((
1
)
t
t
t
t
t
t
t
t
t
D(
CY
}(( )
i(
1
)() = (k
+
. ).
The upper triangle is commutative, and (4.7.2) applies to the lower one. The map
i(
1
) is the embedding of the bre over the base point. Let | : CY CY ,
n n
-
be the inverse.
(4.7.2) Note. ( ) | C . i(
1
).
Proof. We compose both sides with the h-equivalence r( ) from the proof of
(4.7.1). Then r( )i(
1
) equals k
+
+( )
-
, and this is obviously homotopic
to | C : ( )
-
.
As a consequence of the preceding discussion we see that the sequence
CX
D(
CY
i(( )
J( )
(
1
X
(
Y
is h-exact.
4.7. The Fibre Sequence 99
(4.7.3) Lemma. There exists a homeomorphism t
1
: J(C ) CJ( ) such that
(C )
1
= C(
1
) t
1
.
Proof. From the denitions and standard properties of mapping spaces we have
CJ( ) CX CJY and JC( ) CX JCY . We use the exponential law
for mapping spaces and consider CJY and JCY as subspaces of Y
JJ
. In the
rst case we have to use all maps which send d1 1 L1 0 to the base point, in the
second case all maps which send 1 d1 L 0 1 to the base point. Interchanging
the 1-coordinates yields a homeomorphism and it induces t
1
.
We now continue as in the previous section.
. . .
D
2
(
C
2
Y
Di(( )
CJ( )
D(
1
CX
D(
CY
i(( )
J( )
(
1
X
(
Y
is h-exact. We call it the bre sequence of . When we apply the functor T. |
0
to the bre sequence we obtain an exact sequence of pointed sets which consists
from the fourth place onwards of groups and homomorphisms and from the seventh
place onwards of abelian groups ([147]).
Problems
1. Work out the dual of (4.6.5).
2. Describe what happens to the bre sequence under adjunction. A map
a: T J( ) = {(.. n) X JY [ (.) = n(1)]
has two components b : T X and : T JY . Under adjunction, corresponds to
a map T: CT Y from the cone over T . The condition (.) = n(1) is equivalent to
the commutativity b = Ti
1
. This transition is also compatible with pointed homotopies,
and therefore we obtain a bijection a| T. J( )|
0
i
1
. |
0
T. b|. This bijection
transforms
1
into the restriction T. b| i

1
. |
0
T. X|
0
b|. In the next step we
have
T. CY |
0
i( )

T. Y |
0
;|
_
T. J( )|
0

i
1
. |
0
,
; . c|.
The image of ; is obtained in the following manner: With the quotient map : CT T
we have T = ; , and b is the constant map c.
3. There exist several relations between bre and cobre sequences.
The adjunction (. C) yields in TOP
0
the maps j: X CX (unit of the adjunction)
and c: CX X (counit of the adjunction). These are natural in the variable X. For each
: X Y we also have natural maps
j: J( ) CC( ). c: J( ) C( )
dened by
j(.. n)(t ) =
_
.. 2t |. t _ 1,2.
n(2 2t ). t _ 1,2.
and c adjoint to j. Verify the following assertions from the denitions.
(1) The next diagram is homotopy commutative
CY
i( )
J( )

1
X
)
CY
:
1
CC( )
:)( )
CX.
(2) Let i : X 7( ) be the inclusion and r : 7( ) Y the retraction. A path in
7( ) that starts in + and ends in X 0 yields under the projection to C( ) a loop. This
gives a map : J(i ) CC( ). The commutativity j J(r) . | holds.
(3) The next diagram is homotopy commutative
J(
1
)
i(
1
)
:q(
1
))
Y
t)
CX
:
CY .
Chapter 5
Cobrations and Fibrations
This chapter is also devoted to mostly formal homotopy theory. In it we study the
homotopy extension and lifting property.
An extension of : Y along i : X is a map J : X Y such that
Ji = . If i : X is an inclusion, then this is an extension in the ordinary sense.
Many topological problems can be given the form of an extension problem. It is
important to nd conditions on i under which the extendibility of only depends
on the homotopy class of . If this is the case, then is called a cobration.
The dual of the extension problem is the lifting problem. Suppose given maps
: 1 T and : X T. A lifting of along is a map J : X 1 such
that J = . We ask for conditions on such that the existence of a lifting only
depends on the homotopy class of . If this is the case, then is called a bration.
X
T
x
x
x
x
Y
(
1
]
X
T
x
x
x
x
(
T
Each map is the composition of a cobration and a homotopy equivalence and
(dually) the composition of a homotopy equivalence and a bration. The notions are
then used to dene homotopy bres (homotopy kernels) and homotopy cobres
(homotopy cokernels). Axiomatizations of certain parts of homotopy theory
(model categories) are based on these notions. The notions also have many
practical applications, e.g., to showing that maps are homotopy equivalences with
additional properties like brewise homotopy equivalences.
Another simple typical example: A base point . X is only good for homo-
topy theory if the inclusion {.] X is a cobration (or the homotopy invariant
weakening, a so-called h-cobration). This is then used to study the interrelation
between pointed and unpointed homotopy constructions, like pointed and unpointed
suspensions.
5.1 The Homotopy Extension Property
A map i : X has the homotopy extension property (HEP) for the space Y if
for each homotopy h: 1 Y and each map : X Y with i(a) = h(a. 0)
there exists a homotopy H: X 1 Y with H(.. 0) = (.) and H(i(a). t ) =
h(a. t ). We call H an extension of h with initial condition . The map i : X
102 Chapter 5. Cobrations and Fibrations
is a cobration if it has the HEP for all spaces. The data of the HEP are displayed
in the next diagram. We set i
A
t
: X X 1, . (.. t ) and e
0
(n) = n(0).
Y
J
e
0
X
1
}
}
}
}
(
Y
X
i
0

J
J
J
J
J
J
J
J
J
J
(
x
x
x
x
x
x
x
x
x
i
.
0
F
F
F
F
F
F
F
F
F
X 1
1
Y
1
iid
t
t
t
t
t
t
t
t
t
h
For a cobration i : X, the extendibility of only depends on its homotopy

class.
From this denition one cannot prove directly that a map is a cobration, but
it sufces to test the HEP for a universal space Y , the mapping cylinder 7(i ) of i .
Recall that 7(i ) is dened by a pushout
i
.
0
X
b
1
k
7(i ).
Pairs of maps : X Y and h: 1 Y with hi
0
= i then correspond
to maps o : 7(i ) Y with ob = and ok = h. We apply this to the pair
i
A
0
: X X 1 and i id: 1 X 1 and obtain s : 7(i ) X 1 such
that sb = i
A
0
and sk = i id.
Now suppose that i is a cobration. We use the HEP for the space 7(i ), the
initial condition b and the homotopy k. The HEP then provides us with a map
r : X 1 7(i ) such that ri
A
0
= b and r(i id) = k. We conclude from
rsb = ri
A
0
= b, rsk = r(i id) = k and the pushout property that rs = id(7(i )),
i.e., s is an embedding and r a retraction. Let r be a retraction of s. Given and
h, nd o as above and set H = or. Then H extends h with initial condition .
Altogether we have shown:
(5.1.1) Proposition. The following statements about i : X are equivalent:
(1) i is a cobration.
(2) i has the HEP for the mapping cylinder 7(i ).
(3) s : 7(i ) X 1 has a retraction.
A cobration i : X is an embedding; and i() is closed in X, if X is a
Hausdorff space (Problem 1). Therefore we restrict attention to closed cobrations
whenever this simplies the exposition. A pointed space (X. .) is called well-
pointed and the base point nondegenerate if {.] X is a closed cobration.
5.1. The Homotopy Extension Property 103
(5.1.2) Proposition. If i : X is a cobration, then there exists a retraction
r : X 1 X 0 L 1. If is closed in X and if there exists a retraction r,
then i is a cobration.
Proof. Let Y = X 0 L 1, (.) = (.. 0), and h(a. t ) = (a. t ). Apply the
HEP to obtain a retraction r = H.
If is closed in X, then g: X0L1 Y , g(.. 0) = (.), (a. t ) = h(a. t )
is continuous. A suitable extension H is given by gr.
(5.1.3) Example. Let r : X 1 X 0 L 1 be a retraction. Set r(.. t ) =
(r
1
(.. t ). r
2
(.. t )). Then
H: X 1 1 X 1. (.. t. s) (r
1
(.. t (1 s)). st (1 s)r
2
(.. t ))
is a homotopy relative to X 0 L 1 of r to the identity, i.e., a deformation
retraction.
(5.1.4) Example. The inclusions S
n-1
D
n
and d1
n
1
n
are cobrations. A
retraction r : D
n
S
n-1
1 L D
n
0 was constructed in (2.3.5).
It is an interesting fact that one need not assume to be closed. Strm [180,
Theorem 2] proved that an inclusion X is a cobration if and only if the
subspace X 0 L 1 is a retract of X 1.
If we multiplya retractionbyid(Y ) we obtainagaina retraction. Hence Y
X Y is a (closed) cobration for each Y , if i : X is a (closed) cobration.
Since we have proved (5.1.2) only for closed cobrations, we mention another
special case, to be used in a moment. Let Y be locally compact and i : X a
cobration. Then i id: Y X Y is a cobration. For a proof use the fact
that via adjunction and the exponential law for mapping spaces the HEP of i id
for 7 corresponds to the HEP of i for 7
Y
.
(5.1.5) Proposition. Let X and assume that 1 X 1 has the HEP
for Y . Given maps : 1 1 Y. H: X 1 Y.
t
: X 1 Y such
that
(a. s. 0) = H(a. s).
t
(.. 0) = H(.. c).
t
(a. t ) = (a. c. t )
c {0. 1], a , . X, s. t 1. Then there exists : X 1 1 Y such that
(a. s. t ) = (a. s. t ). (.. s. 0) = H(.. s). (.. c. t ) =
t
(.. t ).
Proof. H and
t
together yield a map : X (1 0 L d1 1) Y dened
by (.. s. 0) = H(.. s) and (.. c. t ) =
t
(.. t ). By our assumptions, and
coincide on (1 0 Ld1 1). Let k : (1 1. 1 0 Ld1 1) (1 1. 1 0)
be a homeomorphism of pairs. Since 1 X 1 has the HEP for Y , there
exists : X 1 1 Y which extends (1 k
-1
) and (1 k
-1
). The
map = (1 k) solves the extension problem.
(5.1.6) Proposition. Let i : X be a cobration. Then Xd1 L1 X1
is a cobration.
Proof. Given h: (1 LXd1)1 Y and an initial condition H: X1 Y ,
we set = h[ 1 1 and
t
(.. t ) = h(.. c. t ). Then we apply (5.1.5).
For = 0 we obtain from (5.1.5) that X d1 X 1 is a cobration, in
particular d1 1 and {0] 1 are cobrations. Induction over n shows again that
d1
n
1
n
is a cobration.
We list some special cases of (5.1.5) for a cobration X.
(5.1.7) Corollary. (1) Let : X 1 Y be a homotopy. Suppose = [1
is homotopic rel d1 to [. Then is homotopic rel X d1 to : X 1 Y
such that [ 1 = [.
(2) Let solve the extension problem for (. ) and the extension problem
for ([. g). Suppose . g rel and . [ rel d1. Then
1
.
1
rel .
(3) Let . : X 1 Y solve the extension problem for (h. ). Then there
exists a homotopy I : . rel X 0 L 1.
(5.1.8) Proposition. Let a pushout diagram in TOP be given.
T
J
X
T
Y
If has the HEP for 7, then J has the HEP for 7. If is a cobration, then J is
a cobration.
Proof. Suppose h: T 1 7 and : Y 7 are given such that h(b. 0) =
J(b) for b T. We use the fact that the product with 1 of a pushout is again a
pushout. Since is a cobration, there exists 1
t
: X 7 such that 1
0
=
and 1
t
= h
t
. By the pushout property, there exists H
t
: Y 7 such that
H
t
J = 1
t
and H
t
J = h
t
. The uniqueness property shows H
0
= , since both
maps have the same composition with J and J .
We call J the cobration induced from via cobase change along .
Example. If X is a cobration, then {] X,is a cobration. S
n-1
D
n
is a cobration, hence {S
n-1
] D
n
,S
n-1
is a cobration. The space D
n
,S
n-1
n
; therefore (S
n
. +) is well-pointed.
Example. If (X
}
) is a family of well-pointed spaces, then the wedge
_
}
X
}
is
well-pointed.
5.1. The Homotopy Extension Property 105
Our next result, the homotopy theoremfor cobration says, among other things,
that homotopic maps induce h-equivalent cobrations froma givencobrationunder
a cobase change.
Let : X be a cobration and
t
: . g: T a homotopy. We
consider two pushout diagrams.
T
}
T
}
X
T
Y
( X
G
Since is a cobration, there exists a homotopy

t
: X Y
(
with initial condition
0
= J and
t
=
(
t
. The pushout property of the Y
-diagram provides us
with a unique map k = k
such that k
=
(
and kG =
1
. (We use the notation
k
although the map depends on

1
.) Thus k
is a morphism of cobrations
k :

(
between objects in TOP
B
. Moreover kG . J. We now verify that
the homotopy class of k is independent of some of the choices involved. Let [
t
be another homotopy from to g which is homotopic to
t
relative to d1.
Let
t
: X Y
(
be an extension of
(
[
t
with initial condition
0
= J. Let
; : 1 1 T be a homotopy rel d1 from to [. These data give us on
X 0 1 L X 1 d1 a map I into Y
(
such that
I(.. 0. t ) = J(.). I(.. s. 0) = (.. s). I(.. s. 1) = (.. s).
By (5.1.5) there exists an extension, still denoted I, to X 1 1 such that
(
; =
I( id id). We multiply the Y
diagram by 1 and obtain again a pushout. It

provides us with a unique homotopy 1: Y
1 Y
(
such that 1(Gid) = I
1
and 1 (
id) =
(
pr where I
1
: X 1 Y
(
, (.. t ) I(.. 1. t ). By
construction, 1 is a homotopy under T fromk
to a corresponding map k
)
obtained
from [
t
and
t
. We thus have shown that the homotopy class k|
B
under T of k
only depends on the morphism | from to g in the groupoid (. T). Let us
write k| = |.
We verify that is a functor ([| ~|) = | [|. Let [: g . h:
T. Choose a homotopy
t
: X Y
with
0
= G and
t
=
[
t
. Then
k
)
: Y
h
Y
is determined by k
)
h
=
and k
)
H =
1
. (Here (H.
h
) are
the pushout data for (. h).) Since k
0
= k
G =
1
, we can form the product
homotopy +k
. It has the initial condition J and satises (+k
)( id) =
(
+ k
[ =
(
( + [). Hence k
+)
, constructed with this homotopy, is
determined by k
+)
H = k
1
= k
k
)
H and k
+)
h
=
(
= k
= k
k
)
h
.
Therefore k
k
)
represents ([| ~|).
Let h-COF
B
denote the full subcategory of h-TOP
B
with objects the cobrations
under T. Then we have shown above:
(5.1.9) Theorem. Let : X be a cobration. We assign to the object
: T in (. T) the induced cobration
(
: T Y
(
and to the mor-
phism | : g in (. T) the morphism k
| :

(
. These assignments
yield a contravariant functor
}
: (. T) h-COF
B
.
Since (. T) is a groupoid, k
| is always an isomorphism in h-TOP

B
. We
refer to this fact as the homotopy theorem for cobrations.
(5.1.10) Proposition. In the pushout (5.1.8) let be a cobration and a homotopy
equivalence. Then J is a homotopy equivalence.
Proof. With an h-inverse g: T of we form a pushout
T
Y
G
7.
Since g . id, there exists, by (5.1.9), an h-equivalence k : 7 X under
such that kGJ . id. Hence J has a left h-inverse and G a right h-inverse. Now
interchange the roles of J and G.
Problems
1. A cobration is an embedding. For the proof use that i
1
: 7(i ), a (a. 1) is an
embedding. From i
1
= rsi
1
= ri
1
i then conclude that i is an embedding.
Consider a cobration as an inclusion i : X. The image of s : 7(i ) X 1 is the
subset X 0 L 1. Since s is an embedding, this subset equals the mapping cylinder,
i.e., one can dene a continuous map X 0 L 1 by specifying its restrictions to X 0
and 1. (This is always so if is closed in X, and is a special property of i : X if i
is a cobration.)
Let X be a Hausdorff space. Then a cobration i : X is a closed embedding. Let
r : X 1 X 0 L 1 be a retraction. Then . is equivalent to r(.. 1) = (.. 1).
Hence is the coincidence set of the maps X X 1, . (.. 1), . r(.. 1) into a
Hausdorff space and therefore closed.
2. If i : 1 1, : 1 M have the HEP for Y , then i has the HEP for Y . A home-
omorphism is a cobration. 0 X is a cobration. The sum li
: l
lX
of
cobrations i
is a cobration.
3. Let : 1 Q be an h-equivalence and i : T a cobration. Then : 1 has
an extension to T if and only if has an extension to T. Suppose
0
.
1
: T 1 agree
on . If
0
and
1
are homotopic rel so are
0
.
1
.
4. Compression. Let X be a cobration and : (X. ) (Y. T) a map which is
homotopic as a map of pairs to k : (X. ) (T. T). Then is homotopic relative to to
a map g such that g(X) T.
5. Let X be a cobration and contractible. Then the quotient map X X, is a
homotopy equivalence.
5.2. Transport 107
6. The space C
/
X = X 1,X 1 is called the unpointed cone on X. We have the closed
inclusions : X C
/
X, . (.. 0) and b : {+] C
/
X, + {X 1]. Both maps are
cobrations.
7. Let : X be an inclusion. We have a pushout diagram
C
/
X
J
X L C
/
.
Since is a cobration, so is J. If is a cobration, then J is a cobration. There exists
a canonical homeomorphism X L C
/
,C
/
X,; it is induced by J. Since C
/
is
contractible, we obtain a homotopy equivalence X L C
/
X L C
/
,C
/
X,.
8. The unpointed suspension
/
X of a space X is obtained from X 1 if we identify each
of the sets X 0 and X 1 to a point. If + is a basepoint of X, we have the embedding
: 1
/
X, t (+. t ). If {+] X is a closed cobration, then is a closed (induced)
cobration. The quotient map
/
X X is a homotopy equivalence.
5.2 Transport
Let i : 1 be a cobration and : 1 1 X a homotopy. We dene a map
#
: (. i ). (X.
0
)|
1
(. i ). (X.
1
)|
1
.
called transport along , as follows: Let : X with i =
0
be given.
Choose a homotopy
t
: X with
0
= and
t
i =
t
. We dene
#
| =
1
|. Then (5.1.5) shows that
#
is well dened and only depends on the
homotopy class of rel 1 d1, i.e., on the morphism | (1. X). From the
construction we see ( + [)
#
= [
#
#
. Altogether we obtain:
(5.2.1) Proposition. Let i : 1 be a cobration. The assignments
0

i.
0
|
1
and |
#
yield a transport functor from (1. X) to sets. Since
(1. X) is a groupoid,
#
is always bijective.
The transport functor measures the difference between homotopic in TOP
1
and in TOP. The following is a direct consequence of the denitions.
(5.2.2) Proposition. Let i : 1 be a cobration. Let : (. i ) (X. g) and
t
: (. i ) (X. g
t
) be morphisms in TOP
1
. Then | =
t
|, if and only if there
exists | (1. X) from (X. g) to (X. g
t
) with
t
|
1
=
#
|
1
.
(5.2.3) Proposition. Let i : 1 be a cobration, g: 1 X a map, and
[: X 1 Y a homotopy. Then ([ (g id))
#
[
0+
= [
1+
, if we set
[
i+
| = [
i
|.
Proof. Use that [
t
is an extension of [
t
g and apply the denition.
(5.2.4) Proposition. Let : X Y be an ordinary homotopy equivalence and
i : 1 a cobration. Then
+
: (. i ). (X. g)|
1
(. i ). (Y. g)|
1
is
bijective.
Proof. Let g be h-inverse to and choose : id . g . Consider
i. |
1
(

i. |
1

i. g |
1
(

i. g |
1
.
Since g
+
+
= (g )
+
= ( id)|
#
id
+
, we conclude from (5.2.1) and (5.2.3) that
g
+
+
is bijective, hence g
+
is surjective. The bijectivity of
+
g
+
shows that g
+
is
also injective. Therefore g
+
is bijective and hence
+
is bijective too.
(5.2.5) Proposition. Let i : 1 X and : 1 Y be cobrations and : X
Y an h-equivalence such that i = . Then is an h-equivalence under 1.
Proof. By (5.2.4), we have a bijective map
+
: (Y. ). (X. i )|
1
(Y. ). (Y. )|
1
.
Hence there exists g| with
+
g|
1
= g|
1
= id|
1
. Since is an h-equivalence,
so is g. Since also g
+
is bijective, g has a homotopy right inverse under 1. Hence
g and are h-equivalences under 1.
(5.2.6) Proposition. Let i : X be a cobration and an h-equivalence. Then i
is a deformation retract.
Proof. The map i is a morphism from id() to i . By (5.2.5), i is an h-equivalence
under . This means: There exists a homotopy X 1 X rel from the identity
to a map r : X such that ri = id(), and this is what was claimed.
(5.2.7) Proposition. Given a commutative diagram
t
u
/

Y
t
X
t
with a cobration i and h-equivalences and

t
. Given : X and : . u.
Then there exists
t
:
t
X
t
and
t
:
t
t
. u
t
such that
t
i = and
t
t
i = g
t
.
Proof. We have bijective maps (note
t
= g . gu = u
t
i )
(g)
#

t
+
: (
t
. i ). (X
t
. )|
(
t
. i ). (Y
t
.
t
)|
(
t
. i ). (Y
t
. u
t
i )|
.
Let
t
:
t
X
t
be chosen such that (g)
#
t
+
t
|
= u
t
|
. This means:
t
i =
; and
t
t
has a transport along g to a map which is homotopic under to u
t
.
5.2. Transport 109
This yields a homotopy
tt
:
t
t
. u
t
such that
tt
(i(a). t ) = g(a. min(2t. 1)).
The homotopy + k : (a. t ) (a. min(2t. 1)) is homotopic rel d1 to . We
nowuse (5.1.6) in order to change this
tt
into another homotopy
t
with the desired
properties.
If we apply (5.2.7) in the case that u and u
t
are the identity we obtain the next
result (in different notation). It generalizes (5.2.5).
(5.2.8) Proposition. Given a commutative diagram
T
}
X
T
Y
with cobrations i , and h-equivalences and J. Given g: T and : g .
id, there exists G: Y X and : GJ . id such that G = ig and
t
i = i
t
.
In particular: (J. ) is an h-equivalence of pairs, and there exists a homotopy
inverse of the form (G. g): i .
(5.2.9) Proposition. Suppose a commutative diagram
X
0
o
1
(
0
X
1
o
2
(
1
X
2

(
2

Y
0
b
1
Y
2
b
2
Y
2

is given with cobration a
}
, b
}
and h-equivalences
}
. Let X be the colimit of the
a
}
and Y the colimit of the b
}
. Then the map : X Y induced by the
}
is a
Proof. We choose inductively h-equivalences J
n
: Y
n
X
n
such that a
n
J
n-1
=
J
n
b
n
and homotopies
n
: X
n
1 X
n
fromJ
n
n
to id(X
n
) such that a
n
n-1
=
n
(a
n
id). This is possible by (5.2.7). The J
n
and
n
induce J : Y X and
: X 1 X. J . id. Hence J is a left homotopy inverse of .
Problems
1. Let i : 1 and : 1 T be cobrations. Let : (T. ) (. i ) be a morphism
under 1, : X Y a continuous map, and : 1 1 X a homotopy. Then
(. i ). (X.
0
)|
!

#
,|
!
(. i ). (X.
1
)|
!
,|
!
(T. ). (Y.
0
)|
!
()
#
(T. ). (Y.
1
)|
!
commutes; here . |
!
| = |.
2. Apply the transport functor to pointed homotopy sets. Assume that the inclusion {+]
is a cobration. For each path n: 1 X we have the transport
n
#
: . (X. n(0))|
0
. (X. n(1))|
0
.
As a special case we obtain a right action of the fundamental group (transport along loops)
. X|
0

1
(X. +) . X|
0
. (.. ) . =
#
(.).
Let : . X|
0
. X| denote the forgetful map which disregards the base point. The
map induces an injective map from the orbits of the
1
-action into . X|. This map is
bijective, if X is path connected.
A space is said to be A-simple if for each path n the transport n
#
only depends on the
endpoints of n; equivalently, if for each . X the fundamental group
1
(X. .) acts trivially
on . (X. .)|
0
. If = S
n
, then we say n-simple instead of -simple. We call X simple if
it is -simple for each well-pointed .
3. The action on 1,d1. X|
0
=
1
(X) is given by conjugation. Hence this action is trivial
if and only if the fundamental group is abelian.
4. Let . X|
0
carry a composition law induced by a comultiplication on . Then n
#
is a
homomorphism. In particular
1
(X) acts by homomorphisms. (Thus, if the composition
law on . X|
0
is an abelian group, then this action makes this group into a right module
over the integral group ring Z
1
(X).)
5. Write S(1) = 1,d1 and
1
(X) = S(1). X|
0
. Thenwe canidentify. X|
0
1
(X. +)
S(1). X|
0
. The action of the previous problem is induced by a map v : S(1)
which can be obtained as follows. Extend the homotopy 1 S(1), t t S(1) to a
homotopy : 1 S(1) with the initial condition S(1) and set v =
1
.
Express in terms of v and the comultiplication of S(1) the fact that the induced map is a
group action (v is a coaction up to homotopy).
6. Let (X. e) be a path connected monoid in h-TOP
0
. Then the
1
(X. e)-action on . X|
0
is trivial.
5.3 Replacing a Map by a Cobration
We recall fromSection4.1the constructionof the mappingcylinder. Let : X Y
be a map. We construct the mapping cylinder 7 = 7( ) of via the pushout
X X
( id
( i
0
,i
1
)
Y X
( x,} )
7( ) = X 1 Y,(.) ~ (.. 0).

X 1
o
7( )
s(,) = ,. (.) = (.. 1).
Since (i
0
. i
1
) is a closedcobration, the maps (s. ), s and are closedcobrations.
We also have the projection q : 7( ) Y , (.. t ) (.), , ,. In the case
that : X Y , let : Y Y,X be the quotient map. We also have the quotient
5.3. Replacing a Map by a Cobration 111
map 1 : 7( ) C( ) = 7( ),(X) onto the mapping cone C( ). (Now we
consider the unpointed situation. The direction of the unit interval is different
from the one in the previous chapter.) We display the data in the next diagram. The
map r is induced by q.
X
(
Y
]
c(( )
A
A
A
A
A
A
A
A
A
A
A
A
A
x
Y,X
q = . qs = id
X
}
7( )
1
C( )
i
1s = c( ). q = r1
(5.3.1) Proposition. The following assertions hold:
(1) and s are cobrations.
(2) sq is homotopic to the identity relative to Y . Hence s is a deformation
retraction with h-inverse q.
(3) If is a cobration, then q is a homotopy equivalence under X and r the
induced homotopy equivalence.
(4) c( ): Y C( ) is a cobration.
Proof. (1) was already shown.
(2) The homotopy contracts the cylinder X 1 to X 0 and leaves Y xed,
h
t
(.. c) = (.. t c 1 t ), h
t
(,) = ,.
(3) is a consequence of (5.2.5).
(4) c( ) is induced from the cobration i
0
: X X 1,X 1 via cobase
change along .
We have constructed a factorization = q into a (closed) cobration and a
homotopy equivalence q. Factorizations of this type are unique in the following
sense. Suppose = q
t
t
: X 7
t
Y is another such factorization. Then
i q
t
: 7
t
7 satises i q
t
t
. i . Since
t
is a cobration, we can change i q
t
. k
such that k
t
= . Since i q
t
is an h-equivalence, the map k is an h-equivalence un-
der X, by (5.2.5). Also qk . q
t
. This expresses a uniqueness of the factorization.
If = q : X 7 Y is a factorization into a cobration and a homo-
topy equivalence q, then 7,(X) is called the (homotopical) cobre of . The
uniqueness of the factorization implies uniqueness up to homotopy equivalence of
the cobre. If : X Y is already a cobration, then Y Y,X is the projection
onto the cobre; in this case q : 7 Y is an h-equivalence under X.
The factorization of a map into a cobration and a homotopy equivalence is a
useful technical tool. The proof of the next proposition is a good example.
(5.3.2) Proposition. Let a pushout diagram
T
J
X
T
Y
with a cobration be given. Then the diagram is a homotopy pushout.
Proof. Let qi : 7( ) X be the factorization of . Since q is a homotopy
equivalence under , it induces a homotopy equivalence
q L
id: 7( ) L
T = 7(. ) X L
T = Y
of the adjunction spaces.
(5.3.3) Proposition. Let a commutative diagram
t
k
/

I
/
C
t
1
/
A
A
A
A
A
A
A
A
k
C
1
|
|
|
|
|
|
|
|
T
1
,}
}
}
}
}
}
}
D
B
B
B
B
B
B
B
B
T
t
1
/

D
t
be given. Suppose the inner and the outer square are homotopy cocartesian. If ,
, ; are homotopy equivalences, then is a homotopy equivalence.
Proof. From the data of the diagram we obtain a commutative diagram
7(k. l)
7(,,;)
7(k
t
. l
t
)
D
t
where and
t
are the canonical maps. By hypothesis, and
t
are homotopy
equivalences. By (4.2.1) the map 7(. . ;) is a homotopy equivalence.
(5.3.4) Proposition. Given a commutative diagram as in the previous proposition.
Assume that the squares are pushout diagrams. Then is induced by , , ;.
Suppose that , , ; are homotopy equivalences and that one of the maps k, l and
one of the maps k
t
, l
t
is a cobration. Then is a homotopy equivalence.
Proof. From (5.3.2) we see that the squares are homotopy cocartesian. Thus we
can apply (5.3.3).
5.4. Characterization of Cobrations 113
Problems
1. A map : X Y has a left homotopy inverse if and only if : X 7( ) has a
retraction r : 7( ) X. The map is a homotopy equivalence if and only if is a
2. In the case of a pointed map : (X. +) (Y. +) one has analogous factorizations into a
cobration and a homotopy equivalence. One replaces the mapping cylinder 7( ) with the
pointed mapping cylinder 7
0
( ) dened by the pushout
X X
id
{ i
0
,i
1
)
Y X
{ :, )
X1

7
0
( )
with the pointed cylinder X1 = X 1,{+] 1. The maps (i
0
. i
1
), (s. ), s and are
pointed cobrations. We have a diagram as for (5.3.1) with pointed homotopy equivalences
s. q and C
0
( ) = 7
0
( ),(X) the pointed mapping cone, the pointed cobre of .
3. (i
0
. i
1
): X X X1 is an embedding.
4. Let : X Y and g: Y 7 be pointed maps. We have canonical maps : C( )
C(g ) and : C(g ) C(g); is the identity on the cone and maps Y by g, and is the
identity on 7 and maps the cone by id. Show that is the pointed homotopy cobre
of .
5.4 Characterization of Cobrations
We look for conditions on X which imply that this inclusion is a cobration.
We begin by reformulating the existence of a retraction (5.1.2).
(5.4.1) Proposition. There exists a retraction r : X 1 1 L X 0 if and
only if the following holds: There exists a map u: X 0. o and a homotopy
: X 1 X such that:
(1) u
-1
(0)
(2) (.. 0) = . for . X
(3) (a. t ) = a for (a. t ) 1
(4) (.. t ) for t > u(.).
Proof. Suppose we are given a retraction r. We set (.. t ) = pr
1
r(.. t ) and
u(.) = max{t pr
2
r(.. t ) [ t 1]. For (4) note the following implications:
t > u(.), pr
2
r(.. t ) > 0, r(.. t ) 1, (.. t ) . The other properties
are immediate from the denition. Conversely, given u and , then r(.. t ) =
((.. t ). max(t u(.). 0)) is a retraction.
(5.4.2) Note. Let t
n
> u(.) be a sequence which converges to u(.). Then (4)
implies (.. u(.))

. If u(.) = 0, then . = (.. 0) = (.. u(.))

. Thus
= u
-1
(0). Therefore in a closed cobration X the subspace has the
remarkable property of being the zero-set of a continuous real-valued function.
(5.4.3) Lemma. Let u: X 1 and = u
-1
(0). Let : . g: X 7 rel .
Then there exists

: . g rel such that

(.. t ) =

(.. u(.)) =

(.. 1) for
t _ u(.).
Proof. We dene

by

(.. t ) = (.. 1) for t _ u(.) and by (.. t u(.)
-1
) for
t < u(.). For the continuity of

on C = {(.. t ) [ t _ u(.)] see Problem 1.
We call (X. ) a neighbourhood deformation retract (NDR), if there exist a
homotopy [: X 1 X and a function : X 1 such that:
(1) =
-1
(0)
(2) [(.. 0) = . for . X
(3) [(a. t ) = a for (a. t ) 1
(4) [(.. 1) for 1 > (.).
The pair ([. ) is said to be an NDR-presentation of (X. ).
(5.4.4) Proposition. (X. ) is a closed cobration if and only if it is an NDR.
Proof. If X is a closed cobration, then an NDR-presentation is obtained
from (5.4.1) and (5.4.2). For the converse, we modify an NDR-presentation ([. u)
by (5.4.3) and apply (5.4.1) to the result (
[. u).
(5.4.5) Theorem (Union Theorem). Let X, T X, and T X be
closed cobrations. Then L T X is a cobration.
Proof ([112]). Let : (LT) 1 7 be a homotopy and : X 7 an initial
condition. There exist extensions
: X1 7 of [1 and
B
: T1 7
of [T 1 with initial condition . The homotopies
and
B
coincide on
( T) 1. Therefore there exists :
.
B
rel ( T) 1 L X 0.
Let : X 1 X 1, ~ be the quotient map which identies each interval
{c] 1, c T to a point. Let T : 1 1 1 1 switch the factors. Then
(id T factors over id and yields C: (X 1, ~) 1 7.
Let u: X 1 and : X 1 be functions such that = u
-1
(0) and T =
-1
(0). Dene : X X 1, ~ by (.) = (.. u(.),(u(.) (.))) for
. T and by (.) = (.. 0) = (.. t ) for . T. Using the compactness
of 1 one shows the continuity of .
An extension of and is now given by C ( id).
(5.4.6) Theorem (Product Theorem). Let X and T Y be closed cobra-
tions. Then the inclusion X T L Y X Y is a cobration.
Proof. X X Y , X T X Y , and T = ( Y ) (X T)
X T X Y are cobrations. Now apply (5.4.5).
5.5. The Homotopy Lifting Property 115
Problems
1. Let C = {(.. t ) [ t _ u(.)] and q : X 1 C, (.. t ) (.. t u(.)). Then

q = .
It sufces to show that q is a quotient map. The map I : X 1 X 1 1, (.. t )
(.. t. u(.)) is an embedding onto a closed subspace D. The map m: 1 1 1, (a. b) ab
is proper, hence M = id m is closed. The restriction of M to D is closed, hence MI = q
is closed and therefore a quotient map.
2. The inclusion 0 L {n
-1
[ n N] 0. 1| is not a cobration. The inclusions
=
{0.
-1
] 1 are cobrations. Hence (5.4.5) does not hold for an innite number of cobra-
tions.
3. Set X = {a. b] with open sets 0. {a]. X for its topology. Then = {a] X is a
non-closed cobration. The product X L X X X is not a cobration.
4. Let
X be closed cobrations (1 _ _ n). For all o {1. . . . . n] let

o
=
_
o

X be a cobration. Then
_
n
1

X is a cobration.
5. Let and T be well-pointed spaces. Then . T is well-pointed.
5.5 The Homotopy Lifting Property
A map : 1 T has the homotopy lifting property (HLP) for the space X if the
following holds: For each homotopy h: X 1 T and each map a: X 1
such that a(.) = h(.. 0) there exists a homotopy H: X 1 1 with H = h
and H(.. 0) = a(.). We call H a lifting of h with initial condition a. The map
is called a bration (sometimes Hurewicz bration) if it has the HLP for all
spaces. It is called a Serre bration if it has the HLP for all cubes 1
n
, n N
0
.
Serre brations sufce for the investigation of homotopy groups. In order to see
the duality we can use the dual denition of homotopy and specify the data in the
right diagram. It uses the evaluation e
0
T
: 1
J
1, n n(0).
X
o
1
]
X 1
h
x
x
x
x
x
T
1
]
,}
}
}
}
}
}
}
}
T
1
J
e
0
1
C
C
C
C
C
C
C
C
]
1
|
|
|
|
|
|
|
|
X
1
T
J
e
0
!
A
A
A
A
A
A
A
A
We begin by introducing the dual W() of the mapping cylinder. It is dened
by the pullback
1
]
W()
b
W() = {(.. n) 1 T
J
[ (.) = n(0)].
T
T
J
e
0
!
k(.. n) = n. b(.. n) = ..
If we apply the pullback property to e
0
T
,
J
, we obtain a unique map r : 1
J
W(), ((0). ) such that br = e

0
T
and kr =
J
. If we apply the HLP to
(W(). b. k), we obtain a map s : W() 1
J
such that e
0
T
s = b and
J
s = k.
The relations brs = e
0
T
s = b and krs =
J
s = k imply rs = id, by uniqueness.
Therefore s is a section of r. Conversely, given data (a. h) for a homotopy lifting
problem. They combine to a map j: X W(). The composition H = sj with
a section s is a solution of the lifting problem. Therefore we have shown:
(5.5.1) Proposition. The following statements about : 1 T are equivalent:
(1) is a bration.
(2) has the HLP for W().
(3) r : 1
J
W() has a section.
(5.5.2) Proposition. Let : 1 T have the HLP for X. Let i : X be a
closed cobration and an h-equivalence. Let : X T be given and a: 1
a lifting of over , i.e., a = i . Then there exists a lifting J of which
extends a.
Proof. By (5.2.6) and (5.4.2) we know: There exists u: X 1 and : X1 X
rel such that = u
-1
(0),
1
= id(X),
0
(X) . Set r : X , .
0
(.).
Dene a new homotopy : X 1 X by (.. t ) = (.. t u(.)
-1
) for t < u(.)
and (.. t ) = (.. 1) = . for t _ u(.). We have seen in (5.4.3) that is
continuous. Apply the HLP to h = with initial condition b = ar : X 1.
The verication
h(.. 0) = (.. 0) = (.. 0) = r(.) = ar(.) = b(.)
shows that b is indeed an initial condition. Let H: X 1 1 solve the lifting
problem for h. b. Then one veries that J : X 1, . H(.. u(.)) has the
desired properties.
(5.5.3) Corollary. Let : 1 T have the HLP for X 1 and let i : X be
a closed cobration. Then each homotopy h: X 1 T with initial condition
given on 1 L X 0 has a lifting H: X 1 1 with this initial condition.
(5.5.4) Proposition. Let i : T be a (closed) cobration of locally compact
spaces. The restriction from T to yields a bration : 7
B
7
.
Let : X T be a bration. Then
7
: X
7
T
7
is a bration for locally
compact 7.
Proof. Use adjunction and the fact that X XT is a cobration for each X.
5.5. The Homotopy Lifting Property 117

(5.5.5) Proposition. Let : 1 T be a bration. Then r : 1
J
W(),
((0). ) is a bration.
Proof. A homotopy lifting problem for X and r is transformed via adjunction into
a lifting problem for and X 1 with initial condition given on the subspace
X (1 0 L 0 1).
(5.5.6) Proposition. Let : 1 T be a bration, T
0
T and 1
0
=
-1
(T
0
).
If T
0
T is a closed cobration, then 1
0
1 is a closed cobration.
Proof. Let u: T 1 and h: T 1 T be an NDR-presentation of T
0
T.
Let H: X 1 X solve the homotopy lifting problem for h( id) with initial
condition id(X). Dene 1: X X by 1(.. t ) = H(.. min(t. u(.))). Then
(1. u) is an NDR-presentation for X
0
X.
The proof of the next formal proposition is again left to the reader.
(5.5.7) Proposition. Let a pullback in TOP be given.
Y
T
X
]
C
(
T
If q has the HLPfor 7, then so also has . If is abration, thenq is abration.
We call q the bration induced from the bration via base change along .
In the case that : C T the restriction :
-1
(C) C can be taken as the
induced bration.
(5.5.8) Example. X
J
X
JJ
X X: n (n(0). n(1)) is a bration (5.5.4).
The evaluation e
1
: JY Y , n n(1) is a bration (restriction to + Y ).
Hence we have the induced bration
1
: J( ) Y .
The homotopy theorem for brations says, among other things, that homotopic
maps induce h-equivalent brations.
Let : X T be a bration and : . g: C T a homotopy. We consider
two pullback diagrams.
Y
(
T
X
]
X
]
C
(
T C
T
There exists a homotopy
t
: Y
(
X such that
0
= J and
t
=
t
(
. The
pullback property of the right square yields a map k = k
: Y
(
Y
such that
Gk =
1
and
k =
(
. Let [
t
: . g be homotopic to
t
by a homotopy
; : C 1 1 T relative to C d1. We obtain in a similar manner a map k
)
from
a lifting
t
of [
t
(
. Claim: The maps k
and k
)
are homotopic over C. In order to
verify this, we lift the homotopy ;(
(
id id): Y
(
1 1 T to a homotopy I
with initial data I(,. s. 0) = (,. s), I(,. s. 1) = (,. s), and I(,. 0. t ) = J(t )
by an application of (5.1.5). The homotopy H: (,. t ) I(,. 1. t ) yields, by the
pullback property of the right square, a homotopy 1: Y
(
1 Y
such that
G1 = H and
1 = pr
(
. By construction, 1 is a homotopy over C from k
(
to k
. The reader should now verify the functoriality k

+)
| = k
)
|k
|.
Let h-FIB
C
be the full subcategoryof h-TOP
C
withobjects the brations over C.
(5.5.9) Proposition. Let : X T be a bration. We assign to : C T the
induced bration
(
: Y
(
C and to the morphism | : g in (C. T) the
morphism k
|. This yields a functor (C. T) h-FIB

C
.
Since (C. T) is a groupoid, k
| is always an isomorphism in h-TOP

B
. This
fact we call the homotopy theorem for brations.
As a special case of (5.5.9) we obtain the bre transport. It generalizes the
bre transport in coverings. Let : 1 T be a bration and n: 1 T a path
from b to c. We obtain a homotopy equivalence T
]
n| : J
b
J
c
which only
depends on the homotopy class n| of n, and T
]
u + | = T
]
|T
]
u|. This yields
a functor T
]
: (T) h-TOP. In particular the bres over points in the same path
component of T are h-equivalent.
(5.5.10) Proposition. In the pullback (5.5.7) let be a bration and a homotopy
equivalence. Then J is a homotopy equivalence.
Proof. The proof is based on (5.5.9) and follows the pattern of (5.1.10).
(5.5.11) Remark. The notion of bration and cobration are not homotopy invari-
ant. The projection 1 0 L 0 1 1, (.. t ) . is not a bration, but the
map is over 1 h-equivalent to id. One denition of an h-bration : 1 T is
that homotopies X 1 T which are constant on X 0. c|. c > 0 can be lifted
with a given initial condition; a similar denition for homotopy extensions gives
the notion on an h-cobration. In [46] you can nd details about these notions.
Problems
1. A composition of brations is a bration. A product of brations is a bration. 0 T
is a bration.
2. Suppose : 1 T has the HLP for Y 1
n
. Then each homotopy h: Y 1
n
1 T
has a lifting to 1 with initial condition given on Y (1
n
0 L d1
n
1).
3. Let : 1 T 1 be a bration and
0
: 1
0
T its restriction to T 0 = T. Then
there exists a brewise h-equivalence from
0
id(1) to which is over T 0 the inclusion
1
0
1.
5.6. Transport 119
4. Go through the proof of (5.5.9) and verify a relative version. Let (C. D) be a closed
cobration. Consider only maps C T with a xed restriction J : D T and homotopies
relative to D. Let
1
: Y
1
T be the pullback of along J. Then the maps
have the
form (
.
1
): (Y
. Y
1
) (C. D). By (5.5.6), (Y
. Y
1
) is a closed cobration, and by
(5.5.3) the homotopies
I
can be chosen constant on Y
1
. The maps k
: Y
are then
the identity on Y
1
. The homotopy class of k
is unique as a map over C and under Y

1
.
5. Let (T. C) be a closed deformation retract with retraction r : T C. Let : X T be
a bration and
C
: X
C
C its restriction to C. Then there exists a retraction 1: X X
C
such that
C
1 = r.
5.6 Transport
We construct a dual transport functor. Let : 1 T be a bration, : Y 1 T
a homotopy and : Y 1 1 a lifting along with initial condition . We
dene
#
: (Y.
0
). (1. )|
B
(Y.
1
). (1. )|
B
. |
1
|.
One shows that this map is well dened and depends only on the homotopy class
of relative to Y d1 (see the analogous situation for cobrations). Moreover,
( + [)
#
= [
#
#
.
(5.6.1) Proposition. The assignments . |
B
and |
#
are a functor,
called transport functor, from (Y. T) into the category of sets. Since (Y. T) is
a groupoid,
#
is always bijective.
(5.6.2) Note. Let : 1 T be a bration and [: X 1 Y be a homotopy.
Then[
+
1
= g[
t
|
#
[
+
0
; here [
+
0
: g. | g[
0
. | is the compositionwith[
0
.
(5.6.3) Theorem. Let : X Y be an h-equivalence and : 1 T be a
bration. Then
+
: . |
B
. |
B
is bijective for each : Y T.
Proof. The proof is based on (5.6.1) and (5.6.2) and formally similar to the proof
of (5.2.4).
(5.6.4) Theorem. Let : X T and q : Y T be brations. Let h: X Y
be an h-equivalence and a map over T. Then is an h-equivalence over T.
Proof. The proof is based on (5.6.3) and formally similar to the proof of (5.2.5).
(5.6.5) Corollary. Let q : Y C be a bration and a homotopy equivalence.

Then q is shrinkable.
Let : 1 T be a bration. Then the canonical map r : 1
J
W() is
shrinkable (see (5.5.5)).
5.7 Replacing a Map by a Fibration
Let : X Y be a map. Consider the pullback
W( )

(q,])
Y
J
(e
0
,e
1
)
W( ) = {(.. n) X Y
J
[ (.) = n(0)].
X Y
( id
Y Y
q(.. n) = .. (.. n) = n(1).
Since (e
0
. e
1
) is a bration (see (5.5.8)), the maps (q. ), q and are brations.
Let s : X W( ), . (.. k
((x)
), with k
,
the constant path with value ,. Then
qs = id and s = . (The direction of the unit interval is again different from
the one in the previous chapter.) We display the data and some other to be explained
below in a diagram.
J
}
X
(
Y
=
J( )
J
(
1
u
u
u
u
u
u
u
u
u
W( )
]
Y
The map s is a shrinking of q; a homotopy h
t
: sq . id is given by h
t
(.. n) =
(.. n
t
), n
t
(s) = n((1 t )s). We therefore have a factorization = s into a
homotopy equivalence s and a bration . If =
t
s
t
is another factorization of
this type, then there exists a brewise homotopy equivalence k : W( ) W
t
such
that
t
k = and ks . s
t
. This expresses the uniqueness of the factorization.
Now suppose is a pointed map with base points +. Then W( ) is given the
base point (+. k
+
). The maps . q. s become pointed maps, and the homotopy h
t
is
pointed too. One veries that q and are pointed brations. Let J( ) =
-1
(+)
and J =
-1
(+) be the bres over the base point, with and J the inclusions.
The map q induces r. We call J( ) the homotopy bre of . We use the same
notion for the bre of any replacement of by a bration as above. If is already
a bration, then q is a brewise homotopy equivalence (5.6.4) and r the induced
homotopy equivalence; hence the actual bre is also the homotopy bre.
A map : X Y has a right homotopy inverse if and only if : W( ) X
has a section. It is a homotopy equivalence if and only if is shrinkable.
Chapter 6
Homotopy Groups
The rst fundamental theorem of algebraic topology is the BrouwerHopf degree
theorem. It says that the homotopy set S
n
. S
n
| has for n _ 1 a homotopically
dened ring structure. The ring is isomorphic to Z, the identity map corresponds to
1 Zand the constant map to 0 Z. The integer associated to a map : S
n
S
n
is called the degree of . We have proved this already for n = 1. Also in the general
case degree n roughly means that winds S
n
n-times around itself. In order
to give precision to this statement, one has to count the number of pre-images of
a regular value with signs. This is related to a geometric interpretation of the
degree in terms of differential topology.
Our homotopical proof of the degree theorem is embedded into a more general
investigation of homotopy groups. It will be a simple formal consequence of the
so-called excision theoremof Blakers and Massey. The elegant elementary proof of
this theorem is due to Dieter Puppe. It uses only elementary concepts of homotopy
theory, it does not even use the group structure. (The excision isomorphism is the
basic property of the homology groups introduced later where it holds without any
restrictions on the dimensions.)
Another consequence of the excision theoremis the famous suspension theorem
of Freudenthal. There is a simple geometric construction (the suspension) which
leads fromS
n
. S
n
| to S
n1
. S
n1
|. Freudenthals theorem says that this process
after a while is stable, i.e., induces a bijection of homotopy sets. This is the
origin of the so-called stable homotopy theory a theory which has developed into
a highly technical mathematical eld of independent interest and where homotopy
theory has better formal and algebraic properties. (Homology theory belongs to
stable homotopy.)
The degree theorem contains the weaker statement that the identity of S
n
is not
null homotopic. It has the following interpretation: If you extend the inclusion
S
n-1
R
n
to a continuous map : D
n
R
n
, then there exists a point . with
(.) = 0. For n = 1 this is the intermediate value theorem of calculus; the higher
dimensional analogue has other interesting consequences which we discuss under
the heading of the Brouwer xed point theorem.
This chapter contains the fundamental non-formal results of homotopy theory.
Based on these results, one can develop algebraic topology from the view-point
of homotopy theory. The chapter is essentially independent of the three previous
chapters. But in the last section we refer to the denition of a cobration and a
suspension.
122 Chapter 6. Homotopy Groups
6.1 The Exact Sequence of Homotopy Groups
Let 1
n
be the Cartesian product of n copies of the unit interval 1 = 0. 1|, and
d1
n
= {(t
1
. . . . . t
n
) 1
n
[ t
i
{0. 1] for at least one i ] its combinatorial
boundary (n _ 1). We set 1
0
= {:], a singleton, and d1
0
= 0. In 1
n
,d1
n
we
use d1
n
as base point. (For n = 0 this yields 1
0
,d1
0
= {:] {+], an additional
disjoint base point +.) The n-th homotopy group of a pointed space (X. +) is
n
(X. +) = (1
n
. d1
n
). (X. {+])| 1
n
,d1
n
. X|
0
with the group structure dened below. For n = 1 it is the fundamental group.
The denition of the set
n
(X. +) also makes sense for n = 0, and it can be
identied with the set
0
(X) of path components of X with +| as a base point. The
composition law on
n
(X. +) for n _ 1 is dened as follows. Suppose | and g|
in
n
(X. +) are given. Then | g| is represented by
i
g:
(1) (
i
g)(t
1
. . . . . t
n
) =
_
(t
1
. . . . . t
i-1
. 2t
i
. . . . . t
n
) for t
i
_
1
2
.
g(t
1
. . . . . t
i-1
. 2t
i
1. . . . . t
n
) for
1
2
_ t
i
.
As in the case of the fundamental group one shows that this composition law is a
group structure. The next result is a consequence of (4.3.1); a direct verication
along the same lines is easy. See also (2.7.3) and the isomorphism (2) below.
(6.1.1) Proposition. For n _ 2 the group
n
(X. +) is abelian, and the equality
1
=
i
holds for i _ 2.
We now dene relative homotopy groups (sets)
k
(X. . +) for a pointed pair
(X. ). For n _ 1, let J
n
= d1
n
1 L 1
n
{0] d1
n1
1
n
1 and
set J
0
= {0] 1. We denote by
n1
(X. . +) the set of homotopy classes
of maps of triples : (1
n1
. d1
n1
. J
n
) (X. . +). (Recall that this means
(d1
n1
) . (J
n
) {+], and for homotopies H we require H
t
for each
t 1 to be a map of triples.) Thus, with notation introduced earlier,
n1
(X. . +) = (1
n1
. d1
n1
. J
n
). (X. . +)|.
A group structure
i
, 1 _ i _ n is dened again by the formula (1) above. There
is no group structure in the case n = 0.
We now consider
n
as a functor. Composition with : (X. . +) (Y. T. +)
induces
+
:
n
(X. . +)
n
(Y. T. +); this is a homomorphism for n _ 2. Sim-
ilarly, : (X. +) (Y. +) induces for n _ 1 a homomorphism
+
:
n
(X. +)
n
(Y. +). The functor properties (g )
+
= g
+
+
and id
+
= id are clear. The
morphism
+
:
n
(X. +)
n
(X. . +) is obtained by interpreting the rst group
as
n
(X. {+]. +) and then using the map induced by the inclusion (X. {+]. +)
(X. . +). Maps which are pointed homotopic induce the same homomorphisms.
The group
n
(X. . +) is commutative for n _ 3.
6.1. The Exact Sequence of Homotopy Groups 123
Let h: (1
n1
. d1
n1
. J
n
) (X. . +) be given. We restrict to 1
n
= 1
n
{1]
and obtain a map dh: (1
n
. d1
n
) (. +). Passage to homotopy classes then yields
the boundary operator d:
n1
(X. . +)
n
(. +). The boundary operator is
a homomorphism for n _ 1. For n = 0 we have dh| = h(1)|.
We rewrite the homotopy groups in terms of mapping spaces. This is not strictly
necessary for the following investigations but sometimes technically convenient.
Let C
k
(X. +) be the space of maps 1
k
X which send d1
k
to the base point;
the constant mapis the base point. The space C
1
(X) = C(X) is the loopspace of X.
Given a map (1
n
. d1
n
) (X. +) we have the induced map

: 1
n-k
C
k
(X. +)
which sends u 1
n-k
to 1
k
X, (t
1
. . . . . t
k
) (t
1
. . . . . t
k
. u
1
. . . . . u
n-k
).
This adjunction is compatible with homotopies and induces a bijection
(2)
n
(X. +)
n-k
(C
k
(X. +). +).
Adjunction as above also yields a bijection
(3)
n1
(X. . +)
n1-k
(C
k
(X). C()
k
. +).
These isomorphisms are natural in (X. . +), compatible with the boundary opera-
tors, and the group structures.
(6.1.2) Theorem (Exact homotopy sequence). The sequence

n
(. +)
i
n
(X. +)
}
n
(X. . +)
J

1
(X. . +)
J
0
(. +)
i
0
(X. +)
is exact. The maps i
+
and
+
are induced by the inclusions.
Proof. We prove the exactness for the portion involving
0
and
1
in an elementary
manner. Exactness at
0
(. +) and the relations d
+
= 0 and
+
i
+
= 0 are left to
the reader.
Let n: 1 X represent an element in
1
(X. . +) with dn| = 0. This means:
There exists a path u: 1 with u(0) = n(1) and u(1) = +. The product n+u
is then a loop in X. The homotopy H which is dened by
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
n
n + u
u
-
+
H
t
H
t
(s) =
_
n(2s,(1 t )). 2s _ 1 t.
u(t 2(1 s)). 2s _ 1 t.
shows
+
n + u| = n|. Thus we have shown exactness at
1
(X. . +).
Given a loop n: 1 X. Let H: (1. d1. 0) 1 (X. . +) be a homotopy
from n to a constant path. Then u: s H(1. s) is a loop in . We restrict
H to the boundary of the square and compose it with a linear homotopy to prove
k
+
+ n . k
+
+ u.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....................
n
u H +
+
k
+
+ n . k
+
+ u.
u(t ) = H(1. t ).
Hence i
+
u| = i
+
k
+
+ u| = k
+
+ n| = n|.
We now apply this part of the exact sequence to the pairs (C
n
(X). C
n
()) and
obtain the other pieces of the sequence via adjunction.
(6.1.3) Remark. We previously introduced the mapping space
J(|) = {(a. n) X
J
[ n(0) = +. n(1) = a].
with base point (+. k), k : 1 {+] the constant path. This space is homeomorphic
to
J(X. ) = {n X
J
[ n(0) = +. n(1) ].
i(|) becomes the inclusion C(X) J(X. ), and |
1
the evaluation J(X. ) ,
n n(1). For n _ 1 we assign to : (1
n1
. d1
n1
. J
n
) (X. . +) the
adjoint map
.
: 1
n
J(X. ), dened by
.
(t
1
. . . . . t
n
)(t ) = (t
1
. . . . . t
n
. t ).
It sends d1
n
to the base point and induces a pointed map

: 1
n
,d1
n
J(X. ).
By standard properties of adjunction we see that the assignment |

| is a
well-dened bijection
(4)
n1
(X. . +)
n
(J(X. ). +).
and in fact a homomorphismwith respect to the composition laws
i
for 1 _ i _ n.
These considerations also make sense for n = 0. In the case that = {+], the
space J(X. ) is the loop space C(X).
The exact sequence is also obtained from the bre sequence of |. Under the
identications (4) the boundary operator is transformed into
|
1
+
: 1
n
,d1
n
. J(X. )|
0
1
n
,d1
n
. |
0
.
and
n1
(X. +)
n1
(X. . +) is transformed into
i(|)
+
: 1
n
,d1
n
. C(X)|
0
1
n
,d1
n
. J(X. . +)|
0
.
Now apply T = 1
n
,d1
n
to the bre sequence (4.7.4) of | : X to see the
exactness of a typical portion of the homotopy sequence.
6.1. The Exact Sequence of Homotopy Groups 125
The sequence (6.1.2) is compatible with maps : (X. . +) (Y. T. +). In
particular
+
d = d
+
.
(6.1.4) Remark. In the sequel it will be useful to have different interpretations for
elements in homotopy groups. (See also the discussion in Section 2.3.) We set
S(n) = 1
n
,d1
n
and D(n 1) = CS(n), the pointed cone on S(n). We have
homeomorphisms
S(n) d1
n1
,J
n
. D(n 1) 1
n1
,J
n
.
the rst one . (.. 1), the second one the identity on representatives in 1
n
1;
moreover we have the embedding S(n) D(n1), . (.. 1) which we consider
as an inclusion. These homeomorphisms allow us to write
n
(X. +) 1
n
,d1
n
. X|
0
= S(n). X|
0
.
n1
(X. . +) (1
n1
,J
n
. d1
n1
,J
n
). (X. )|
0
(D(n1). S(n)). (X. )|
0
.
and d:
n1
(X. . +)
n
(. +) is induced by the restriction from D(n 1)
to S(n).
The pointed cone on S
n
is D
n1
: We have a homeomorphism
S
n
1,(S
n
0 L e
n1
1) D
n1
. (.. t ) (1 t )e
n1
t..
Therefore we can also represent elements in
n1
(X. . +) by pointed maps
(D
n1
. S
n
) (X. ) and elements in
n
(X. +) by pointed maps S
n
X.
In comparing these different models for the homotopy groups it is important to
remember the homeomorphism between the standard objects (disks and spheres),
since there are two homotopy classes of homeomorphisms.
Problems
1.
n
(. . a) = 0. Given : (1
n
. d1
n
. J
n-1
) (. . a). Then a null homotopy is
I
(.
1
. . . . . .
n
) = (.
1
. .
2
. . . . . (1 t ).
n
).
2. Let + X
0
X
1
X
2
be a sequence of T
1
-spaces (i.e., points are closed).
Give X =
_
n_1
X
n
the colimit topology. Then a compact subset 1 X is contained
in some X
n
. Use this to show that the canonical maps
n
(X
i
. +)
n
(X. +) induce an
isomorphism colim
i

n
(X
i
. +)
n
(X. +).
3. Let (X. . T. b) be a pointed triple. Dene the boundary operator d:
n
(X. . b)
n-1
(. b)
n-1
(. T. b) as the composition of the previously dened operator with
the map induced by the inclusion. Show that the sequence

n
(. T. b)
n
(X. T. b)
n
(X. . b)
u
n-1
(. T. b)
is exact. The sequence ends with
1
(X. . b).
4. The group structure in
n1
(X. . +) is induced by an h-cogroup structure on
(D(n 1). S(n)) in the category of pointed pairs.
5. Let : (X. .) (Y. ,) be a pointed map. One can embed the induced morphism
:
n
(X. .)
n
(Y. ,) into an exact sequence which generalizes the case of an inclu-
sion . Let 7( ) be the pointed mapping cylinder of and = i : X 7( ) Y the
standard factorization into an inclusion and a homotopy equivalence, as explained in (5.3.1).
We can now insert the isomorphism
:
n
(7( ). +)
n
(Y. +) into the exact sequence
of the pair and obtain an exact sequence

n
(X. +)

n
(Y. +)
n
(7( ). X). +) .
One can dene the groups
n
(7( ). X. +) without using the mapping cylinder. Consider
commutative diagrams with pointed maps and .
d1
n
,J
n-1

1
n
,J
n-1

Y
We consider (. ): as a morphism in the category of pointed arrows. Let
n
( ) denote the set of homotopy classes of such morphisms. For : X Y we ob-
tain the previously dened
n
(Y. X. +). The projection : i induces an isomorphism
n
(7( ). X. +) =
n
(i )
n
( ). One can also use the bre sequence of .
6.2 The Role of the Base Point
We have to discuss the role of the base point. This uses the transport along paths.
Let a path : 1 X and : (1
n
. d1
n
) (X. (0)) be given. We consider
as a homotopy of the constant map d1
n
{(0)]. We extend the homotopy
t
to a homotopy V
t
: 1
n
X with initial condition = V
0
. An extension exists
because d1
n
1
n
is a cobration. The next proposition is a special case of (5.2.1)
and problems in that section. In order to be independent of that section, we also
repeat a proof in the present context.
(6.2.1) Proposition. The assignment V
0
| V
1
| is a well-dened map
#
:
n
(X. (0))
n
(X. (1))
which only depends on the morphism | in the fundamental groupoid (X). The
relation (+n)
#
= n
#
#
holds, and thus we obtain a transport functor from(X)
which assigns to .
0
X the group
n
(X. .
0
) and to a path the morphism
#
.
The map
#
is a homomorphism.
Proof. Let : . g be a homotopy of maps (1
n
. d1
n
) (X. .
0
) and [: . n
a homotopy of paths from.
0
to .
1
. Let V
t
: 1
n
X be a homotopy which extends
(. ) and W
t
a homotopy which extends (g. n). These data combine to a map on
T = 1
n
0 1 L d1
n
1 1 L 1
n
1d1 1
n2
as follows: On 1
n
0 1
6.2. The Role of the Base Point 127
we use , on d1
n
1 1 we use

[, on 1
n
1 0 we use V , and on 1
n
1 1
we use W. If we interchange the last two coordinates then T is transformed into
J
n1
. Therefore our map has an extension to 1
n2
, and its restriction to 1
n
11
is a homotopy from V
1
to W
1
. This shows the independence of the representatives
and . The other properties are clear from the construction.
There is a similar transport functor in the relative case. We start with a function
: (1
n
. d1
n
. J
n-1
) (X. . a
0
) and a path : 1 froma
0
to a
1
. We consider
the path as a homotopy of the constant map J
n-1
{a
0
]. Then we extend this
homotopy to a homotopy V
t
: (1
n
. d1
n
) (X. ). An extension exists because
J
n-1
d1
n
and d1
n
1
n
are cobrations.
(6.2.2) Proposition. The assignment V
0
| V
1
| is a well-dened map
#
:
n
(X. . a
0
)
n
(X. . a
1
)
which only depends on the morphism | in the fundamental groupoid (). For
n _ 2 the map
#
is a homomorphism. As above we have a transport functor
from ().
Since
#
is always bijective, homotopy groups associated to base points in the
same path component are isomorphic.
We list some naturality properties of the transport functors. As a special case of
the functor property we obtain right actions of the fundamental groups:
n
(X. .)
1
(X. .)
n
(X. .). (. ) =
#
().
n
(X. . a)
1
(. a)
n
(X. . a). (. ) =
#
().
We also have an action of
1
(. a) on
n
(X. a) via the natural homomorphism;
more generally, we can make the
n
(X. a) into a functor on () by viewing a
path in as a path in X. From the constructions we see:
(6.2.3) Proposition. The morphisms in the exact homotopy sequence of the pair
(X. ) are natural transformations of transport functors on (). In particular,
they are
1
(. +)-equivariant with respect to the actions above.
Continuous maps : (X. ) (Y. T) are compatible with the transport func-
tors
+
(n
#
()) = ( n)
#
(
+
()).
Let
t
: (X. ) (Y. T) be a homotopy and set n: t (a. t ). Then the diagram
n
(Y. T.
0
a)
u
#
n
(X. . a)
(
0
g
g
g
g
g
g
(
1
W
W
W
W
W
W
n
(Y. T.
1
a)
is commutative. As in the proof of (2.5.5) one uses this fact to show:
(6.2.4) Proposition. Let : (X. ) (Y. T) be an h-equivalence. Then the
induced map
+
:
n
(X. . a)
n
(Y. T. a) is bijective.
Suppose that that induces isomorphisms
}
()
}
(T) and
}
(X)
}
(Y ) for {n. n 1], n _ 1. Then the Five Lemma (11.1.4) implies that
+
:
n1
(X. . +)
n1
(Y. T. +) is an isomorphism. With some care, this also
holds for n = 0, see Problem 3.
Let : (X. ) (Y. T) be a map of pairs such that the individual maps
X Y and T induce for each base point in isomorphism for all
n
, then
+
:
n
(X. . a)
n
(Y. T. (a)) is bijective for each n _ 1 and each a .
For the case n = 1 see Problem 3.
The transport functors have special properties in low dimensions.
(6.2.5) Proposition. Let : 1 X be given. Then
#
:
1
(X. (0))
1
(X. (1))
is the map n|
-
|n||. In particular, the right action of
1
(X. .) on itself is
given by conjugation =
-1
.
(6.2.6) Proposition. Let .
1
. .
2

2
(X. . +) be given. Let : = d.
2

1
(. +).
Then .
1
: = (.
2
)
-1
.
1
.
2
(multiplicative notation for
2
).
Proof. We rst prove the claim in a universal situation and then transport it by
naturality to the general case. Set D = D(2). S = S(1).
Let |
1
. |
2

2
(D D. S S) be the elements represented by the inclusions
of the summands (D. S) (D D. S S). Set = d(|
2
)
1
(S S). From
(6.2.3) and (6.2.5) we compute
d(|
1
) = (d|
1
) =
-1
(d|
1
) = (d|
2
)
-1
(d|
1
)(d|
2
) = d(|
-1
2
|
1
|
2
).
Since D D is contractible, d is an isomorphism, hence |
1
|
2
= |
-1
2
|
1
|
2
.
Let now h: (D D. S S) (X. ) be a map such that hi
k
represents .
k
,
i.e., h
+
(|
k
) = .
k
. The computation
.
1
: = (h
+
|
1
) (dh
+
|
2
) = h
+
(|
1
) = h
+
(|
-1
2
|
1
|
2
) = .
-1
2
.
1
.
2
proves the assertion in the general case.
(6.2.7) Corollary. The image of the natural map
2
(X. +)
2
(X. . +) is con-
tained in the center.
The actions of the fundamental group also explain the difference between
pointed and free (=unpointed) homotopy classes.
(6.2.8) Proposition. Let S(n). X|
0
,(~) denote the orbit set of the
1
(X. +)-action
on S(n). X|
0
. The map S(n). X|
0
S(n). X| which forgets the base point
induces an injective map : S(n). X|
0
,(~) S(n). X|. For path connected X
the map is bijective. The forgetful map
n
(X. . +) = (D(n). S(n 1)). (X. )|
0
(D(n). S(n 1)). (X. )|
6.3. Serre Fibrations 129
induces an injective map of the orbits of the
1
(. +)-action; this map is bijective
if is path connected (n _ 2).
Problems
1. Let be path connected. Each element of
1
(X. . a) is represented by a loop in (X. a).
The map
:
1
(X. a)
1
(X. . a) induces a bijection of
1
(X. . a) with the right (or
left) cosets of
1
(X. a) modulo the image of i
:
1
(. a)
1
(X. a).
2. Let .
1
(X. . a) be represented by : 1 X with (1) and (0) = a. Let
n: 1 X be a loop in (X. a). The assignment (n|. |) n + | = n| | denes
a left action of the group
1
(X. a) on the set
1
(X. . a). The orbits of this action are the
pre-images of elements under d:
1
(X. . a)
0
(. a). Let (J. ): (X. ) (Y. T)
be a map of pairs. Then J
:
1
(X. . a)
1
(Y. T. (a)) is equivariant with respect
to the homomorphism J
:
1
(X. a)
1
(Y. (a)). Let |
1
(X. . a) with (1) =
u . The isotropy group of | is the image of
1
(. u) in
1
(X. a) with respect to
n| + n +
-
|. Find an example
0
.
1

1
(X. . a) such that
0
has trivial and
1
non-trivial isotropy group. It is in general impossible to dene a group structure on
1
(X. . a) such that
1
(X. a)
1
(X. . a) becomes a homomorphism.
3. Althoughthere is onlya restrictedalgebraic structure at the beginningof the exact sequence
we still have a Five Lemma type result. Let : (X. ) (Y. T) be a map of pairs. If
:
0
()
0
(T) and
:
1
(X. a)
1
(Y. (a)) are surjective and
:
0
(X)
0
(Y ) is injective, then
:
1
(X. . a)
1
(Y. T. (a)) is surjective. Suppose that for
each c the maps
:
1
(X. c)
1
(Y. (c)) and
:
0
()
0
(T) are injective
and
:
1
(. c)
1
(T. (c)) is surjective, then
:
1
(X. . a)
1
(Y. T. (a)) is
injective for each a .
4. Let (X. ) be a pair such that X is contractible. Then d:
q1
(X. . a)
q
(. a) is
for each q _ 0 and each a a bijection.
5. Let X be an h-equivalence. Then
n
(X. . a) = 0 for n _ 1 and a .
6. Let X carry the structure of an h-monoid. Then
1
(X) is abelian and the action of the
fundamental group on
n
(X. +) is trivial.
7. Give a proof of (6.2.8).
8. The
1
(X. +)-action on
n
(X. +) is induced by a map j
n
: S(n) S(n) S(1) by an
application of the functor . X|
0
. If we use the model D
n
,S
n-1
for the n-sphere, then an
explicit map j
n
is . (2.. +) for 2[.[ _ 1 and . (+. 2[.[ 1) for 2[.[ _ 1.
6.3 Serre Fibrations
The notion of a Serre bration is adapted to the investigation of homotopy groups,
only the homotopy lifting property for cubes is used.
(6.3.1) Theorem. Let : 1 T be a Serre bration. For T
0
T set 1
0
=
-1
T
0
. Choose base points + T
0
and + 1
0
with (+) = +. Then induces
for n _ 1 a bijection
+
:
n
(1. 1
0
. +)
n
(T. T
0
. +).
Proof.
+
surjective. Let .
n
(T. T
0
. +) be represented by
h: (1
n
. d1
n
. J
n-1
) (T. T
0
. +).
By (3.2.4), there exists a lifting H: 1
n
1 with H(J
n-1
) = {+] and H = h.
We then have H(d1
n
) 1
0
, and therefore H represents a pre-image of . under
+
.
+
injective. Let .
0
. .
1

n
(1. 1
0
. +) be represented by
0
.
1
and have
the same image under
+
. Then there exists a homotopy
t
: (1
n
. d1
n
. J
n-1
)
(T. T
0
. +) such that
0
(u) =
0
(u),
1
(u) =
1
(u) for u 1
n
. Consider the
subspace T = 1
n
d1 L J
n-1
1 and dene G: T 1 by
G(u. t ) =
_
t
(u). u 1
n
. t {0. 1].
+. u J
n-1
. t 1.
The set T d(1
n
1) is transformed into J
n
, if one interchanges the last two
coordinates. By (3.2.4) again, there exists a map H: 1
n
1 1 such that
H[T = G and H = . We can view H as a homotopy from
0
to
1
.
We use the isomorphism
n
(1. J. +)
n
(T. +), J =
-1
(+) in the exact
sequence of the pair (1. J. +) and obtain as a corollary to (6.3.1) the exact sequence
of a Serre bration:
(6.3.2) Theorem. For aSerre bration: 1 T withinclusioni : J =
-1
(b)
1 and . J the sequence

n
(J. .)
i

n
(1. .)
]

n
(T. b)
J

n-1
(J. .)
0
(1. .)
0
(T. b).
The new map d has the following description: Let : (1
n
. d1
n
) (T. b) be
given. View as 1
n-1
1 T. Lift to : 1
n
1, constant on J
n-1
. Then
d | is represented by [1
n-1
1. The very end of the sequence requires a little
extra argument. For additional algebraic structure at the beginning of the sequence
see the discussion of the special case in (3.2.7).
(6.3.3) Theorem. Let : 1 T be a continuous map and Ua set of subsets such
that the interiors cover T. Assume that for U U the map
U
:
-1
(U) U
induced by is a Serre bration. Then is a Serre bration.
Proof. A subdivision of width = 1,N, N N of 1
n
consists of the cubes
1(a
1
. . . . . a
n
) =

n
}=1
1(a
}
) where 1(k) = k,N. (k 1),N| for 0 _ k < N,
k Z. Ak-dimensional face of 1(a
1
. . . . . a
n
) is obtained by replacing nk of the
intervals 1(a
}
) by one of its boundary points. (The a
}
are integers, 0 _ a
}
< n.)
It sufces to work with an open covering U. Choose N such that each cube
1(a
1
. . . . . a
n
) 1(b) is mapped under h into some U U. This is possible by
6.3. Serre Fibrations 131
the Lebesgue lemma (2.6.4). Let V
k
1
n
denote the union of the k-dimensional
faces of the subdivision of 1
n
.
We have to solve a lifting problem for the space 1
n
with initial condition a. We
begin by extending a over 1
n
0. | to a lifting of h. We solve the lifting problems
1
n
0 L V
k-1
0. |
1(k-1)
1
]
1
n
0 L V
k
0. |
h
1(k)
l
l
l
l
l
l
l
l
l
T
for k = 0. . . . . n with V
-1
= 0 and H(1) = a by induction over k. Let W be
a k-dimensional cube and dW the union of its (k 1)-dimensional faces. We can
solve the lifting problems
W 0 L dW 0. |
1(k-1)
-1
U
]
I
W 0. |
h
1
V
k
k
k
k
k
k
k
k
U
by a map H
W
, since
U
is a Serre bration; here U U was chosen such that
h(W 0. |) U.
The H
W
combine to a continuous map H(k): V
k
0. | 1 which lifts h
and extends H(k 1). We dene H on the rst layer 1
n
0. | as H(n). We
now treat 1
n
. 2| similarly with initial condition given by H(n)[1
n
{] and
continue in this manner inductively.
(6.3.4) Example. Since a product projection is a bration we obtain from (6.3.3):
A locally trivial map is a Serre bration.
(6.3.5) Example. Let : 1 T be a covering with typical bre J. Since each
map 1
n
J is constant,
n
(J. +) is for n _ 1 the trivial group. The exact
sequence of then shows
+
:
n
(1)
n
(T) for n _ 2. The covering R S
1
then yields
n
(S
1
) 0 for n _ 2. Moreover we have the exact sequence
1
1
(1. +)
]
1
(T. +)
J
0
(J. +)
i
0
(1. +)
]
0
(T. +) 1
with the inclusion i : J =
-1
(+) 1 and
0
(J. +) = J. It yields for
: R S
1
the bijection d:
1
(S
1
) Z. A lifting of the loop s
n
: 1 S
1
,
t exp(2i nt ) with initial condition 0 is t nt . Hence ds
n
| = n. Thus we
have another method for the computation of
1
(S
1
).
(6.3.6) Example. Recall the Hopf bration : S
2n1
C1
n
(14.1.9). The exact
sequence (6.3.2) and
i
(S
1
) = 0 for i > 1 yield the isomorphisms
+
:
i
(S
2n1
)
i
(C1
n
). for i _ 3:
and in particular
i
(S
3
)
i
(S
2
) for i _ 3, since C1
1
2
(the Riemann sphere).
(6.3.7) Example. From linear algebra one knows a surjective homomorphism
SU(2) SO(3) with kernel {1] Z,2. The space SU(2) is homeomorphic to
S
3
. Hence SO(3) is homeomorphic to R1
3
and
1
(SO(3)) Z,2.
(6.3.8) Proposition. Let : (1
1
. 1
0
) T be a relative Serre bration, i.e.,
: 1
1
T is a Serre bration and the restriction of to 1
0
is also a Serre
bration. Let (J
b
1
. J
b
0
) be the pair of bres over (e) = b T. Then:
(1) The inclusion induces bijections
n
(J
b
1
. J
b
0
. e)
n
(1
1
. 1
0
. e).
(2)
0
(1
0
)
0
(1
1
) is surjective if and only if
0
(J
b
0
)
0
(J
b
1
) is surjective
for each b T.
Proof. (1) We rst prove the claim for n = 1 and begin with the surjectivity.
Let : (1. d1. 0) (1
1
. 1
0
. e) be given. The path ( )
-
: 1 T is lifted to
g: 1 1
0
with initial point (1). Then g(1) J
0
, and and +g represent the
same element in
1
(1
1
. 1
0
. e). The projection ( + g) is a null homotopic loop
with base point b. We lift a null homotopy to 1
1
with initial condition +g on 1 0
and constant on d1 1. The lifting is a homotopy (1. d1. 0) 1 (1
1
. 1
0
. e)
from + g to a map into (J
1
. J
0
. e). This proves the surjectivity.
Suppose
0
.
1
: (1. d1. 0) (J
1
. J
0
. e) are given, and let 1: (1. d1. 0)1
(1
1
. 1
0
. e) be a homotopyfrom
0
to
1
. We lift 1
-
to1: 1 1 1
0
withinitial
condition 1(s. 0) = 1(s. 1) and 1(0. t ) = 1(1. t ) = e. The homotopy (1+
2
1)
is a homotopy of loops which is relative to d1
2
homotopic to the constant map.
We lift a homotopy to 1
1
with initial condition 1 +
2
1 on 1
2
0 and constant on
d1
2
1. The end is a homotopy from
0
+k
e
to
1
+k
e
. This proves the injectivity.
The higher dimensional case is obtained by an application to the relative Serre
bration (C
n
J
1
. C
n
J
0
) (C
n
1
1
. C
n
1
0
) T.
(2) Suppose
0
(1
0
)
0
(1
1
) is surjective. The argument above for the
surjectivity is used to show the surjectivity of
0
(J
b
0
)
0
(J
b
1
). The other
implication is easy.
Problems
1. The 2-fold covering S
n
R1
n
yields for n _ 2 the isomorphism
1
(R1
n
) Z,2.
2. Prove directly the exactness of the sequence (6.3.2) without using (6.1.2).
3. The map C C, : :
2
has the HLP for 1
0
but not for 1
1
.
4. Let : (1. e) (T. b) be a Serre bration with bre J =
-1
(b). Then
C
n
(): C
n
(1. e) C
n
(T. b)
6.4. The Excision Theorem 133
is a Serre bration with bre C
n
(J. e).
6.4 The Excision Theorem
A basic result about homotopy groups is the excision theorem of Blakers and
Massey [22].
(6.4.1) Theorem (BlakersMassey). Let Y be the union of open subspaces Y
1
and
Y
2
with non-empty intersection Y
0
= Y
1
Y
2
. Suppose that
i
(Y
1
. Y
0
. +) = 0 for 0 < i < . _ 1
i
(Y
2
. Y
0
. +) = 0 for 0 < i < q. q _ 1
for each base point + Y
0
. Then the excision map, induced by the inclusion,
| :
n
(Y
2
. Y
0
. +)
n
(Y. Y
1
. +)
is surjective for 1 _ n _ q 2 and bijective for 1 _ n < q 2 ( for each
choice of the base point + Y
0
). In the case that = 1, there is no condition on
i
(Y
1
. Y
0
. +).
We defer the proof of this theorem for a while and begin with some applications
and examples. We state a special case which has a somewhat simpler proof and
already interesting applications. It is also a special case of (6.7.9).
(6.4.2) Proposition. Let Y be the union of open subspaces Y
1
and Y
2
with non-
empty intersection Y
0
. Suppose (Y
2
. Y
0
) = 0 is q-connected. Then (Y. Y
1
) is
q-connected.
We apply the excision theorem (6.4.1) to the homotopy group of spheres. We
use the following subspaces of S
n
, n _ 0,
D
n
= {(.
1
. . . . . .
n1
) S
n
[ .
n1
_ 0] H
n
= {. S
n
[ . ,= e
n1
].
We use R
n
R
n1
, (:
1
. . . . . :
n
) (:
1
. . . . . :
n
. 0) and similar inclusions for
subsets of R
n
. We choose + = e
1
as a base point; e
i
is the standard unit vector.
(6.4.3) Lemma. We have isomorphisms d:
i1
(D
n1
-
. S
n
. +)
i
(S
n
. +) for
i _ 0. n _ 0 and
i
(S
n
. +)
i
(S
n
. D
n
. +) for i _ 0. n _ 1.
Proof. In the rst case we use the exact sequence of the pair (D
n1
-
. S
n
). The
space D
n1
-
is contractible and hence
i
(D
n1
-
. +) = 0 for i _ 0 and n _ 0.
In the second case we consider similarly the exact sequence of (S
n
. D
n
). Note
that + = e
1
D
n
for n _ 1.
For n _ 0 we have a diagram with the isomorphisms (6.4.3)
i
(S
n
. +)
T

i1
(S
n1
. +)
i1
(D
n1
-
. S
n
. +)
J

i1
(S
n1
. D
n1
. +).
The morphism | is induced by the inclusion and 1 is dened so as to make the
diagram commutative. Note that the inductive proof of (1) in the next theorem only
uses (6.4.2).
(6.4.4) Theorem. (1)
i
(S
n
) = 0 for i < n.
(2) The homomorphism| is an isomorphismfor i _ 2n2 and an epimorphism
for i = 2n 1. A similar statement holds for 1.
Proof. Let N(n) be the statement (1) and 1(n) the statement (2). Obviously N(1)
holds. Assume N(n) holds. We then deduce 1(n). We apply the excision theorem
to (Y. Y
1
. Y
2
. Y
0
) = (S
n1
. D
n1
. D
n1
-
. S
n
). By N(n) and (6.4.3) we have
i
(S
n
)
i1
(D
n1
. S
n
) = 0 for 0 _ i < n. We use the excision theorem
for = q = n 1 and see that | is surjective for i 1 _ 2n and bijective for
i 1 _ 2n 1. Finally, 1(n) and N(n) imply N(n 1).
In order to have the correct hypotheses for the excision theorem, we thicken
the spaces, replace D
n
by H
n
and note that the inclusions D

n
H
n
and
S
n-1
H
n
H
n
-
are h-equivalences.
(6.4.5) Proposition. The homomorphism
i
(D
n1
-
. S
n
. +)
i
(D
n1
-
,S
n
. +)
induced by the quotient map is an isomorphism for i _ 2n1 and an epimorphism
for i = 2n.
Proof. Consider the commutative diagram
i
(D
n1
-
. S
n
. +)

t
i
(D
n1
-
,S
n
. +)
(1)
i
(S
n1
. D
n1
. +)
(2)

i
(S
n1
,D
n1
. +).
The map (1) is induced by a homeomorphism and the map (2) by a homotopy
equivalence, hence both are isomorphisms. Now apply (6.4.4).
The homomorphism 1 is essentially the suspension homomorphism. In order
to see this, let us work with (6.1.4). The suspension homomorphism
+
is the
composition
+
:
n
(X. +)
n1
(CX. X. +)
J

n1
(CX,X. +) =
n1
(X. +)
6.5. The Degree 135
with the quotient map q : D(n 1) D(n 1),S(n) = S(n 1).
The next result is the famous suspension theorem of Freudenthal ([66]).
(6.4.6) Theorem. The suspension
+
:
i
(S(n))
i1
(S(n1)) is an isomor-
phism for i _ 2n 2 and an epimorphism for i = 2n 1.
Proof. We have to show that q
+
:
i1
(CX. X)
i1
(CX,X) is for X = S(n)
an isomorphism(epimorphism) in the appropriate range. This follows from(6.4.5);
one has to use that S
n
is homeomorphic to S(n) and that D
n1
-
is the (pointed)
cone on S
n
.
(6.4.7) Theorem.
n
(S(n)) Z and
+
:
n
(S(n))
n1
(S(n 1)) is an
isomorphism (n _ 1). The group
n
(S(n)) is generated by the identity of S(n).
Proof. From the exact sequence
2
(S
3
)
2
(S
2
)
1
(S
1
) (S
3
) of the
Hopf bration S
1
S
3
S
2
and
}
(S
3
) = 0 for = 1. 2 we obtain an
isomorphism d:
2
(S
2
)
1
(S
1
) Z. From (6.4.6) we obtain a surjection
+
:
1
(S(1))
2
(S(2)); this is an isomorphism, since both groups are isomor-
phic to Z. For n _ 2, (6.4.6) gives directly an isomorphism
+
. We know that
1
(S(1)) Z is generated by the identity, and
+
respects the identity.
(6.4.8) Example. We continue the discussion of the Hopf brations (6.3.6). The
Hopf bration S
1
S
2n1
C1
n
and
i
(S
2n1
) = 0 for i _ 2n yield
2
(C1
n
)
1
(S
1
) Z and
i
(C1
n
) = 0 for 0 _ i _ 2n, i ,= 2. The inclu-
sion S
2n1
S
2n3
, : (:. 0) induces an embedding C1
n
C1
n1
. We
compare the corresponding Hopf brations and their exact sequences and conclude
2
(C1
n
)
2
(C1
n1
). Let C1
o
=
_
n_1
C1
n
be the colimit. The canonical
inclusion C1
n
C1
o
induces
i
(C1
n
)
i
(C1
o
) for i _ 2n. A proof uses
the fact that a compact subset of C1
o
is contained in some nite C1
1
. Therefore
C1
o
is a space with a single non-trivial homotopy group
2
(C1
o
) Z.
Note also the special case
3
(S
2
)
3
(S
3
) Z.
We have similar results for real projective spaces. The twofold coverings
Z,2 S
n
R1
n
are use to show that
1
(R1
2
)
1
(R1
3
)
1
(R1
o
) Z,2, induced by the inclusions,
i
(R1
n
)
i
(R1
n1
) for i < n
and
i
(R1
n
) = 0 for 0 _ i < n, i ,= 1. The space R1
o
has a single non-trivial
homotopy group
1
(R1
o
) Z,2.
6.5 The Degree
Let J :
n
(S(n)) Z be the isomorphism which sends id| to 1. If : S(n)
S(n) is a pointed map, then
+
:
n
(S(n))
n
(S(n)) is the multiplication by the
integer J( ) = J(
+
id|) = J( |). Since the map S(n). S(n)|
0
S(n). S(n)|
which forgets about the base point is bijective (see (6.2.8)), we can transport J
to a bijection J : S(n). S(n)| Z. The functoriality
+
g
+
= (g)
+
shows
J(g) = J( )J(g); therefore J(h) = 1 if h is a homeomorphism. The suspen-
sion sends | to . id|; hence J( ) = J( . id).
(6.5.1) Proposition. Given pointed maps : S(m) S(m) and g: S(n) S(n).
Then J( . g) = J( )J(g).
Proof. We use the factorization . g = ( . id)(id .g). The map . id is a
suspensionof , and suspension does not change the degree. Let t : S(m).S(n)
S(n) . S(m) interchange the factors. From t(g . id)t = id .g we conclude
J(id .g) = J(g . id) = J(g).
Let k
n
: S(n) S
n
be a homeomorphism. The bijection
S
n
. S
n
| S(n). S(n)|. | k
n
k
-1
n
|
is independent of the choice of k
n
. We use this bijection to transport J to a bijection
J : S
n
. S
n
| Z. If J( |) = k we call k the degree J( ) of . We still have
the properties J( )J(g) = J(g), J(id) = 1, J(h) = 1 for a homeomorphism
h. By a similar procedure we dene the degree J( ) for any self-map of a space
S which is homeomorphic to S(n).
Matrix multiplication l
: R
n
R
n
, . . induces for each GL
n
(R) a
pointed map 1
: S
(n)
S
(n)
. For the notation see (6.1.4).
(6.5.2) Proposition. The degree of 1
is the sign of the determinant det().

Proof. Let n: 1 GL
n
(R), t (t ) be a path. Then (.. t ) 1
(t)
. is a
homotopy. Hence J(1
) only depends on the path component of in GL

n
(R).
The group GL
n
(R) has two path components, distinguished by the sign of the
determinant. Thus it sufces to show that for some with det() = 1 we have
J(1
) = 1. By the preceding discussion and (6.1.4) we see that (.

1
. . . . . .
n
)
(.
1
. .
2
. . . . . .
n
) has degree 1.
The stereographic projection (6.1.4) now shows that the map S
n
S
n
which
changes the sign of the rst coordinate has degree 1.
(6.5.3) Proposition. Let O(n 1). Then z
: S
n
S
n
, . . has degree
det().
Proof. Again it sufces to verify this for appropriate elements in the two path
components of O(n 1), and this we have already achieved.
(6.5.4) Corollary. The map S
n
S
n
, . . has degree (1)
n1
.
A vector eld on S
n
is a continuous map J : S
n
R
n1
such that for each
. S
n
the vector J(.) is orthogonal to .. For the maximal number of linearly
independent vector elds see [3].
6.6. The Brouwer Fixed Point Theorem 137
(6.5.5) Theorem. There exists a vector eld J on S
n
such that J(.) ,= 0 for each
. S
n
if and only if n is odd.
Proof. Let n = 2k 1. Then
(.
1
. .
2
. . . . . .
2k-1
. .
2k
) (.
2
. .
1
. . . . . .
2k
. .
2k-1
)
is a vector eld with the desired property.
Let J be a vector eld such that J(.) ,= 0. Set V(.) = J(.),[J(.)[. Then
(.. t ) cos t . sin t V(.) is a homotopy from the identity to the antipodal
map. Hence the antipodal map has degree 1. By (6.5.4), n is odd.
(6.5.6) Proposition. Let t : S(m) .S(n) S(n) .S(m) interchange the factors.
Then J(t) = (1)
nn
.
Proof. By (6.5.2) we know the analogous assertion for the models S
(n)
.
6.6 The Brouwer Fixed Point Theorem
We prove the xed point theorem of Brouwer and a number of equivalent results.
As an application we discuss the problem of topological dimension.
Let us rst introduce some notation. Consider the cube
W = W
n
= {(.
i
) R
n
[ 1 _ .
i
_ 1]
with the faces C
i
() = {. W
n
[ .
i
= 1]. We say, T
i
W
n
separates C
i
()
and C
i
(), if T
i
is closed in W, and if
W T
i
= T
i
() L T
i
(). 0 = T
i
() T
i
(). C
i
() T
i
().
with open subsets T
i
() and T
i
() of W T
i
. The n-dimensional standard
simplex is ^
n
. Its boundary d^
n
is the union of the faces d
i
^
n
= {(t
0
. . . . . t
n
)
^
n
[ t
i
= 0].
(6.6.1) Theorem. The following statements are equivalent:
(1) A continuous map b : D
n
D
n
has a xed point (Brouwer Fixed Point
Theorem).
(2) There does not exist a continuous map r : D
n
S
n-1
which is the identity
on S
n-1
(Retraction Theorem).
(3) The identity of S
n-1
is not null homotopic (Homotopy Theorem).
(4) Let : D
n
R
n
be a continuous map such that (:) = : for : S
n-1
.
Then D
n
is contained in the image of .
(5) Let g: D
n
R
n
be continuous. Then there exists a xed point or there exists
: S
n-1
such that g(:) = z: with z > 1.
(6) Let
i
: W
n
R, 1 _ i _ n be functions such that
i
(.) < 0 for . C
i
()
and
i
(.) > 0 for . C
i
(). Then there exists . W
n
such that
i
(.) = 0
for each i (Intermediate Value Theorem).
(7) Suppose T
i
separates C
i
() and C
i
() for 1 _ i _ n. Then the intersection
T
1
T
2
T
n
is non-empty.
(8) Let T
0
. . . . . T
n
be a closed covering of ^
n
such that e
i
T
i
and d
i
^
n
T
i
.
Then
_
n
i=0
T
i
,= 0. The same conclusion holds if we assume that the T
i
are
open.
(9) Let T
0
. . . . . T
n
be a closed covering of ^
n
such that e
i
T
i
and d
i
^
n
T
i
=
0. Then
_
n
i=0
T
i
,= 0.
The xed point theoremexpresses a topological property of D
n
. If h: X D
n
is a homeomorphism and : X X a self-map, then h h
-1
has a xed point :
and therefore has the xed point h(:). We can apply (2) to the pairs (W
n
. dW
n
)
and (^
n
. d^
n
), since they are homeomorphic to (D
n
. S
n-1
). Statement (3) is also
equivalent to the inclusion S
n-1
R
n
{0] not being null homotopic (similarly
for dW
n
in place of S
n-1
).
Proof. (1) = (2). Suppose r is a retraction. Then . r(.) is a map without
xed point.
(2) = (3). The map r : D
n
S
n-1
which corresponds by (2.3.4) to a null
homotopy of the identity is a retraction.
(3) =(1). Suppose b has no xed point. Then
S
n-1
1 S
n-1
. (.. t )
. t b(.)
[. t b(.)[
= N(. t b(.))
is a homotopy from the identity to the map : . N(. b(.)). Since b
has no xed point, the formula for denes a map on the whole of D
n
, and
then (.. t ) (t.) is a homotopy from the constant map to . Thus is null
homotopic, and therefore also id(S
n-1
).
(2) =(4). If . is contained in the interior of D
n
, then there exists a retraction
r : R
n
. S
n-1
of S
n-1
R
n
.. If . is not contained in the image of ,
then r : D
n
S
n-1
contradicts the retraction theorem.
(4) =(5). Dene a map : D
n
R
n
by
(i) (.) = 2. g(2.). [.[ _ 1,2.
(ii) (.) = [.[
-1
. 2
_
1 [.[
_
g
_
[.[
-1
.
_
. [.[ _ 1,2.
For [.[ =
1
2
we obtain in both cases 2. g(2.). Thus is a well-dened
continuous map.
For [.[ = 1 we have (.) = .. By (4), there exists , with (,) = 0. If
[,[ _
1
2
, then (i) shows that 2, is a xed point. If [,[ >
1
2
, then [,[ ,= 1, and
(ii) shows the second case with z = (2 2[,[)
-1
> 1.
6.6. The Brouwer Fixed Point Theorem 139
(5) =(1). A special case.
(3) =(6). Set : W
n
R
n
, . (
1
(.). . . . .
n
(.)). Suppose (.) ,= 0 for
each . W
n
. Then : W
n
R
n
0. Consider h: (t. .) (1 t ). t (.).
If . C
i
(), i.e., .
i
< 0, then (1 t ).
i
t
i
(.) < 0 for each t 1. Hence
h
t
: dW
n
R
n
0 is a homotopy fromthe inclusion to . Since has an extension
to W
n
, it is null homotopic, but the inclusion is not null homotopic. Acontradiction.
(6) =(7). Let J denote the Euclidean distance. Dene
i
: W R by
i
(.) =
_
J(.. T
i
). . T
i
().
J(.. T
i
). . T
i
() L T
i
.
and apply (6).
(7) = (2). Let r : W
n
dW
n
be a retraction. We dene T
i
() =
r
-1
(.
i
> 0) and T
i
= r
-1
(.
i
= 0). We apply (7) and obtain a contradic-
tion.
(3) = (8). We use the functions
i
(.) = J(.. T
i
). Our assumptions imply
i
(e
i
) > 0, and
i
(.) = 0 provided . d
i
^
n
. If the T
i
have empty intersection,
then (.) = (
0
(.). . . . .
n
(.)) ,= 0 for every . ^
n
. This gives us a map
: ^
n
d^
n
. . (
i
(.))
-1
(.).
because, since the T
i
cover ^
n
, for each . at least one coordinate
i
(.) is zero. If
. d
i
^
n
, then (.) d
i
^
n
, hence (1 t ). t (.) d
i
^
n
for each t 0. 1|.
The identity of d^
n
is therefore homotopic to = [d^
n
. Since has the
extension it is null homotopic, and therefore also id(d^
n
) is null homotopic.
This contradicts (3).
Now suppose the T
i
are open. By a general result of point-set topology there
exist closed sets C
i
T
i
and the C
i
still form a covering. In order to make sure
that the C
i
satisfy the hypotheses of (8) we can replace the C
i
by C
i
L d
i
^
n
. The
rst part of the proof now shows that the C
i
have non-empty intersection.
(8) =(9). Set U
i
= ^
n
T
i
. Suppose the T
i
have empty intersection. Then
the U
i
cover ^
n
. Since the T
i
are a covering, the U
i
have empty intersection. By
construction, e
i
U
i
and d
i
^
n
U
i
. We therefore can apply (8) in the case
of the open covering by the U
i
and see that the U
i
have non-empty intersection.
Contradiction.
(9) = (2). Let
}
= {(t
0
. . . . . t
n
) d^
n
[ t
}
_ 1,n]. Let r : ^
n
d^
n
be
a retraction and set T
}
= r
-1
(
}
). Then (9) tells us that the T
}
have non-empty
intersection, and this is impossible.
Theorem (6.6.1) has many different proofs. For a proof which uses only basic
results in differential topology see [79]. Another interesting proof is based on a
combinatorial result, called Sperners Lemma [173].
The retraction theorem does not hold for innite-dimensional spaces. In [70,
Chapter 19] you can nd a proof that the unit disk of an innite-dimensional Banach
space admits a retraction onto its unit sphere.
Does there exist a sensible topological notion of dimension for suitable classes
of spaces? Greatest generality is not necessary at this point. As an example we
introduce the covering dimension of compact metric spaces X. (For dimension
theory in general see [94].) Let C be a nite covering of X and c > 0 a real
number. We call C an "-covering, if each member of C has diameter less than c.
We say C has order m, if at least one point is contained in m members but no point
in m1. The compact metric space X has covering dimension dimX = k, if there
exists for each c > 0 a nite closed c-covering of X of order k 1 and k N
0
is minimal with this property. Thus X is zero-dimensional in this sense, if there
exists for each c > 0 a nite partition of X into closed sets of diameter at most c.
We verify that this notion of dimension is a topological property.
(6.6.2) Proposition. Let X and Y be homeomorphic compact metric spaces. If X
is k-dimensional then also is Y .
Proof. Let h: X Y be a homeomorphism. Fix c > 0 and let Ube the covering of
Y by the open c-balls U
t
(,) = {. [ J(.. ,) < c]. (We use J for the metrics.) Let
be a Lebesgue number of the covering (h
-1
(U) [ U U). Since dimX = k,
there exists a nite closed -covering C of X of order k 1. The nite closed
covering D = (h(C) [ C C) of Y has then the order k 1, and since each
member of C is contained in a set h
-1
(U), the covering D is an c-covering. Thus
we have shown dimY _ k.
We nowshowthat dimY _ k, i.e., there exists > 0 such that each nite closed
-covering has order at least k 1. Let c > 0 be a corresponding number for X.
A homeomorphism g: Y X is uniformly continuous: There exists a > 0 such
that J(,
1
. ,
2
) < implies J(g(,
1
). g(,
2
)) < c. So if C is a -covering of Y , then
D = (g(C) [ C C) is an c-covering of X. Since D has order at least k 1, so
has C.
(6.6.3) Proposition. There exists c > 0 such that each nite closed c-covering
(T
}
[ J) of ^
n
has order at least n 1.
Proof. Let c be a Lebesgue number of the covering U
i
= ^
n
d
i
^
n
, i = 0. . . . . n.
Hence for each J there exist i such that T
}
U
i
, and the latter is equivalent
to T
}
d
i
^
n
= 0. Suppose e
k
T
}
. Since e
k
d
i
^
n
for i ,= k, we cannot
have T
}
U
i
; thus e
k
T
}
implies T
}
U
k
. Since each e
k
is contained in at
least one of the sets T
}
we conclude [J[ _ n 1. For each J we now choose
g( ) {0. . . . . n] such that T
}
d
(})
^
n
= 0 and set
k
= L{T
}
[ g( ) = k];
this is a closed set because J is nite. Each T
}
is contained in some
k
, hence
the
k
cover ^
n
. Moreover, by construction,
k
d
k
^
n
= 0. We can therefore
apply part (9) of (6.6.1) and nd an . in the intersection of the
k
. Hence for each
6.7. Higher Connectivity 141
k there exists i
k
such that . T
i
k
. Since each T
}
is contained in exactly one of
the sets
k
, the element . is contained in the n 1 members T
i
k
, k = 0. . . . . n of
the covering.
We can now compare the covering dimension and the algebraic dimension.
(6.6.4) Theorem. ^
n
has covering dimension n. A compact subset of R
n
has
covering dimension at most n.
Proof. By (6.6.3), ^
n
has covering dimension at least n. It remains to construct
nite closed c-coverings of order n 1 for each c. See Problem 4.
Problems
1. Let U. V be an open covering of 1
2
. Then there exists either a path u: 1 U such that
u(0) 1 0. u(1) 1 1 or a path : 1 V such that (0) 0 1. (1) 1 1.
2. Let U. V be an open covering of ^
2
. Then there exists a path component

U of U such
that

U d
i
^
2
,= 0 for each i or a path component of V with a similar property.
3. Generalize the preceding two exercises to n dimensions.
4. The following gure indicates the construction of closed c-coverings of order 3 for the
square.
Generalize this construction to the cube 1
n
by a suitable induction.
5. Suppose 1
n
is the union of a nite number of closed sets, none of which contains points
of two opposite faces. Then at least n 1 of these closed sets have a common point.
6.7 Higher Connectivity
For many applications it is important to know that the homotopy groups of a space
vanish in a certain range. We discuss several reformulations of this fact. In the
following
0
(X. .) =
0
(X) with base point .|. The space D
0
is a singleton and
S
-1
= 0.
(6.7.1) Proposition. Let n _ 0. The following are equivalent:
(1)
n
(X. .) = 0 for each . X.
(2) Each map S
n
X has an extension to D
n1
.
(3) Each map d1
n1
X has an extension to 1
n1
.
Proof. The case n = 0 is trivial. The equivalence of (2) and (3) is a consequence
of the homeomorphism (D
n1
. S
n
) (1
n1
. d1
n1
). Suppose : S
n
X is
given. Use e
1
= (1. 0. . . . ) S
n
as a base point and think of representing
an element in
n
(X. .). If (1) holds, then is pointed null homotopic. A null
homotopy S
n
1 X factors over the quotient map S
n
1 D
n1
, (.. t )
(1 t )e
1
t. and yields an extension of . Conversely, let an element of
n
(X. .) be represented by a pointed map : (S
n
. e
1
) (X. .). If this map has
an extension J to D
n1
, then (J. ) represents
n1
(X. X. .) = 0 with
d = .
(6.7.2) Proposition. Let n _ 0. Let : (D
n
. S
n-1
) (X. ) be homotopic as
a map of pairs to a map k : (D
n
. S
n-1
) (. ). Then is relative to S
n-1
homotopic to a map g such that g(D
n
) .
Proof. The case n = 0 is trivial. Let G
t
: (D
n
. S
n-1
) (X. ) be a homotopy
from to k according to the assumption. Dene z: D
n
1 D
n
1 by
z(.. t ) = (2(.. t )
-1
.. 2(.. t )) with the function (.. t ) = max(2[.[. 2t ).
Then H = Gz is a homotopy with the desired property from to g = H
1
.
e
e
e
e
e
e
..............................
a J
b c
D
n
0
1
2
........................................
a J
b c
(6.7.3) Proposition. Let n _ 1. The following assertions about (X. ) are equiv-
alent:
(1)
n
(X. . +) = 0 for each choice of + .
(2) Each map : (1
n
. d1
n
) (X. ) is as a map of pairs homotopic to a
constant map.
(3) Each map : (1
n
. d1
n
) (X. ) is homotopic rel d1
n
to a map into .
Proof. (1) =(2). Let : (1
n
. d1
n
) (X. ) be given. Since J
n-1
is contractible,
there exists a homotopy of the restriction : J
n-1
to a constant map. Since
J
n-1
d1
n
and d1
n
1
n
are cobrations, is as a map of pairs homotopic to
g: (1
n
. d1
n
) (X. ) such that g(J
n-1
) = {a
0
]. Since
n
(X. . a
0
) = 0, the
map g: (1
n
. d1
n
. J
n-1
) (X. . a
0
) is null homotopic as a map of triples.
(2) =(3). (6.7.2).
(3) = (1). Let : (1
n
. d1
n
. J
n-1
) (X. . +) be given. By assump-
tion (3) | is contained in the image of
n
(. . +) (X. . +). Now use
n
(. . +) = 0.
We call (X. ) n-compressible if one of the assertions in (6.7.3) holds. More
generally, we call a map : X Y n-compressible if the following holds: For
each commutative diagram
d1
n

X
(
1
n

Y
there exists : 1
n
X such that [d1
n
= and . relative to d1
n
. (This
amounts to part (3) in (6.7.3).) This notion is homotopy invariant in the following
sense:
(6.7.4) Proposition. Given : X Y and a homotopy equivalence : Y 7.
Then is n-compressible if and only is n-compressible.
(6.7.5) Proposition. Let n _ 0. The following assertions about (X. ) are equiv-
alent:
(1) Each map : (1
q
. d1
q
) (X. ), q {0. . . . . n] is relative to d1
q
homo-
topic to a map into .
(2) The inclusion : X induces for each base point a a bijection
+
:
q
(. a)
q
(X. a) for q < n and a surjection for q = n.
(3)
0
()
0
(X) is surjective, and
q
(X. . a) = 0 for q {1. . . . . n] and
each a .
Proof. (1) = (3). The surjectivity of
0
()
0
(X) is equivalent to (1) for
q = 0. The other cases follow from (6.7.3).
(2) =(3). This follows from the exact sequence (6.1.2).
A pair (X. ) is called n-connected if (1)(3) in (6.7.5) hold. We call (X. )
1-connected if the pair is n-connected for each n. A pair is o-connected if and
only if
+
:
n
(. a)
n
(X. a) is always bijective. If X ,= 0 but = 0 we say
that (X. ) is (1)-connected, and (0. 0) is o-connected.
(6.7.6) Proposition. Let n _ 0. The following assertions about X are equivalent:
(1)
q
(X. .) = 0 for 0 _ q _ n and . X.
(2) The pair (CX. X) is (n 1)-connected.
(3) Each map : d1
q
X, 0 _ q _ n 1 has an extension to 1
q
.
Proof. The cone CX is contractible. Therefore d:
q1
(CX. X. +)
q
(X. +).
This and (6.7.5) shows the equivalence of (1) and (2). The equivalence of (1) and
(3) uses (6.7.1).
A space X is n-connected if (1)(3) in (6.7.6) hold for X. Note that this is
compatible with our previous denitions for n = 0. 1.
Let : X Y be a map and X 7( ) the inclusion into the mapping
cylinder. Then is said to be n-connected if (7( ). X) is n-connected. We
then also say that is an n-equivalence. Thus is n-connected if and only if
+
:
q
(X. .)
q
(Y. (.)) is for each . X bijective (surjective) for q < n
(q = n). If is an o-equivalence we also say that is a weak (homotopy)
equivalence. Thus is a weak equivalence if and only if
+
:
n
(X. .)
n
(Y. (.)) is bijective for each n _ 0 and each . X.
(6.7.7) Proposition. Let (
1
.
0
): (1
1
. 1
0
) T be a relative Serre bration. Let
J
b
}
denote the bre of
}
over b. Then the following are equivalent:
(1) (1
1
. 1
0
) is n-connected.
(2) (J
b
1
. J
b
0
) is n-connected for each b T.
Proof. This is a direct consequence of (6.3.8).
The compression properties of an n-connected map can be generalized to pairs
of spaces which are regular unions of cubes of dimension at most n. We use this
generalization in the proof of theorem (6.7.9). Consider a subdivision of a cube 1
n
.
Let us call T a cube-complex if T is the union of cubes of this subdivision. A
subcomplex of T is then the union of a subset of the cubes in T. We understand
that T and contain with each cube all of its faces. The k-skeleton T(k) of T
consists of the cubes in T of dimension _ k; thus (k) = T(k) T.
(6.7.8) Proposition. Let : X Y be n-connected. Suppose (C. ) is a pair of
cube-complexes of dimension at most n. Then to each commutative diagram
X
(
Y
there exists : C X such that [ = and . relative to .
Proof. Induction over the number of cubes. Let T C such that C is
obtained from T by adding a cube W of highest dimension. Then dW T.
By induction there exists
t
: T X such that
t
[ = and a homotopy
H:
t
. [T relative to . Extend H to a homotopy of . The end
1
of this
homotopy satises
1
[T =
t
. We now use that is n-connected and extend
t
over W to : C X such that .
1
relative to T. Altogether we have
.
B

1
.
and [ =
t
[ = .
(6.7.9) Theorem. Let : (X. X
0
. X
1
) (Y. Y
0
. Y
1
) be a map such that the re-
strictions
i
: X
i
Y
i
are n-connected and
01
: X
0
X
1
Y
0
Y
1
is (n 1)-
connected. Suppose X = X
0
LX
1
and Y = Y
0
LY
1
. Then is an n-equivalence.
Proof. We use mapping cylinders to reduce to the case of inclusions : X
Y.
i
: X
i
Y
i
. Let (J. ): (1
n
. d1
n
) (Y. X) be given. We have to show that
this map is homotopic relative to d1
n
to a map into X. Let
i
= J
-1
(Y Y
i
) L
-1
(X X
i
).
These sets are closed and disjoint. By the Lebesgue lemma we choose a cubical
subdivision of 1
n
such that no cube W of the subdivision intersects both
0
and
1
. Let 1
}
be the union of the cubes W which satisfy
J(W) Y
i
. (W d1
n
) X
i
.
Then 1
i
is a cubical subcomplex and
1
n
= 1
0
L 1
1
. J(1
i
) Y
i
. (1
i
d1
n
) X
i
.
We denote by 1
-
the (n 1)-skeleton of a cubical complex; then 1 d1
n
=
1
-
d1
n
and 1
0
1
-
1
= 1
-
0
1
-
1
. We have a commutative square
X
01

Y
01
d1
n
1
01
(
01
1
-
01
.
T
01
01
J
J
J
J
J
Since (Y
01
. X
01
) is (n1)-connected there exists a homotopy relative to d1
n
1
01
from J
01
to a map g
01
: 1
-
01
X
01
. Dene g
0
: 1
0
(d1
n
L 1
-
) X
0
by
g
0
[1
0
d1
n
=
0
. g
0
[1
0
1
-
1
= g
01
.
(Both maps agree on the intersection.) The homotopy J
01
. g
01
and the constant
homotopy of
0
combine to a homotopy of J
0
[1
0
(d1
n
L 1
-
1
) to g
0
which is
constant on 1
0
d1
n
. Since the inclusion of a cube complex into another one is a
cobration, this homotopy can be extended to a homotopy [: 1
0
1 Y
0
from
J
0
to H
0
. We obtain a diagram
X
0

Y
0
1
0
(d1
n
L 1
-
1
)
1
0
1
0
h
0
N
N
N
N
N
N
N
where H
0
is homotopic to h
0
: 1
0
X
1
relative to 1
0
(d1
n
L 1
-
1
), since
(Y
0
. X
0
) is n-connected.
We prove the second part similarly. We obtain a map g
1
: 1
1
(d1
n
L1
-
0
) Y
1
with g
1
[1
1
d1
n
=
1
and g
1
[1
1
1
-
0
= g
01
and then
X
1

Y
1
1
1
(d1
n
L 1
-
0
)
1
1
.
1
1
h
1
N
N
N
N
N
N
N
The maps h
0
and h
1
coincide on 1
-
01
and yield a map h: 1
-
0
L1
-
1
X which is
homotopic relative d1
n
to J[1
-
0
L 1
-
1
; moreover h[d1
n
= . Let now W be an
n-dimensional cube, say with W 1
0
. Then dW 1
-
0
and h(dW) = h
0
(dW)
X
0
. Since (Y
0
. X
0
) is n-connected, we can deform the map relative to dW to a map
into X
0
.
(6.7.10) Corollary. Let : X Y be an n-connected map between well-pointed
spaces. Then : X Y is (n 1)-connected. If X is n-connected, then
X is (n 1)-connected. The sphere S
k1
is k-connected.
Proof. Let
t
X denote the unpointed suspension of X. This is a quotient of X 1
and covered by the open cones C
0
= X 0. 1,X 0 and C
1
= X|0. 1|,X 1
with intersection X|0. 1. We can apply (6.7.7) directly; the cones are contractible
and therefore the induced maps C
}
(X) C
}
(Y ) o-connected. In the case of a
well-pointed space X the quotient map
t
X X is an h-equivalence.
(6.7.11) Theorem. Let : X Y be a continuous map. Let (U
}
[ J) and
(V
}
[ J) open coverings of X and Y such that (U
}
) V
}
. Suppose that
for each nite 1 J the induced map
T
:
_
}T
U
}

_
}T
V
}
is a weak
equivalence. Then is a weak equivalence
Proof. By passage to the mapping cylinder we can assume that is an inclusion.
Let h: (1
n
. d1
n
) (Y. X) be given. We have to deform h relative to d1
n
into X.
By compactness of 1
n
it sufces to work with nite J. A simple induction reduces
the problem to J = {0. 1]. Then we apply (6.7.9).
Problems
1. Let Y = {0] L {n
-1
[ n N] and X the same set with the discrete topology. Then the
identity X Y is a weak equivalence but there does not exist a weak equivalence Y X.
2. Identify in S
1
the open sets {(.. ,) [ , > 0] and {(.. ,) [ , < 0] to a point. The quotient
map S
1
S onto the quotient space S, consisting of four points, is a weak equivalence
(but not a homotopy equivalence). In particular
1
(S) Z. Show that S has a universal
covering.
6.8 Classical Groups
We use exact sequences of Serre brations and deduce from our knowledge of
i
(S
n
) other results about homotopy groups of classical groups and Stiefel man-
ifolds. We use a uniform notation for the (skew) elds F = R. C. H and the
6.8. Classical Groups 147
corresponding groups (orthogonal, unitary, symplectic)
O(n) = O(n. R). SO(n) = SO(n. R).
U(n) = O(n. C). SU(n) = SO(n. C).
Sp(n) = O(n. H).
Let J = dim
R
F. The starting point are the (Serre) brations which arise from the
action of the orthogonal groups on the unit spheres by matrix multiplication
O(n. F)
}
O(n 1. F) S
d(n1)-1
.
SO(n. F)
}
SO(n 1. F) S
d(n1)-1
.
The inclusions of the groups arise from
_
0
0 1
_
. We also pass to the colimit
and obtain O(o. F) = colim
n
O(n. F) and SO(o. F) = colim
n
SO(n. F). From
i
(S
n
) = 0, i < n and the exact homotopy sequences of the brations we deduce
that the inclusions : O(n. F) O(n 1. F) and : SO(n. F) SO(n 1. F)
are J(n 1) 2 connected. By induction and passage to the colimit we obtain
(6.8.1) Proposition. For n < m _ o, the inclusions O(n. F) O(m. F) and
SO(n. F) SO(m. F) are J(n 1) 2 connected; in particular, the homomor-
phisms
i
(O(n. F))
i
(O(m. F)) are isomorphisms in the range i _ n2 (R),
i _ 2n 1 (C), and i _ 4n 1 (H).
We turn our attention to Stiefel manifolds of orthonormal k-frames in F
n
:
V
k
(R
n
) O(n),O(n k) SO(n),SO(n k).
V
k
(C
n
) U(n),U(n k) SU(n),SU(n k).
V
k
(H
n
) Sp(n),Sp(n k).
We have the corresponding (Serre) brations of the type H G G,H for
these homogeneous spaces. We use (6.8.1) in the exact homotopy sequences of
these brations and obtain:
(6.8.2) Proposition.
i
(V
k
(F
n
)) = 0 for i _ J(n k 1) 2.
We have the bration
: V
k1
(F
n1
) V
1
(F
n1
). (
1
. . . . .
k1
)
k1
.
The bre over e
k1
is homeomorphic to V
k
(F
n
); with | : (. 0) we obtain a
homeomorphism : (
1
. . . . .
k
) (|
1
. . . . . |
k
. e
k1
) onto this bre. From the
homotopy sequence of this bration we obtain
+
:
i
(V
k
(F
n
))
i
(V
k1
(F
n1
)) is an isomorphism for
i _ J(n 1) 3.
We use V
1
(F
n
) = S
dn-1
and
t
(S
t
) Zand obtain from(6.8.3) by induction
2(n-k)1
(V
k
(C
n
)) Z,
4(n-k)3
(V
k
(H
n
)) Z.
The real case is more complicated. The result is
n-k
(V
k
(R
n
))
_
Z. k = 1. or n k even .
Z,2. k _ 2. n k odd.
Proof. By (6.8.3) and induction it sufces to consider the case k = 2. Later we
compute the homology groups of V
2
(R
n
), and the theorem of Hurewicz will then
give us the desired result.
Problems
1. The group O(n) has two path components. The groups SO(n), U(n), SU(n), and Sp(n)
are path connected.
2. In low dimensions we have some special situations, namely
U(1) SO(2) S
1
.
Spin(3) SU(2) Sp(1) S
3
.
Z,2 SU(2) SO(3). a 2-fold covering,
SU(n) U(n) S
1
. a bration.
Use these data in order to verify
1
(SO(2))
1
(O(2)) Z.
1
(SO(3))
1
(SO(n)) Z,2. m _ 3.
1
(U(1))
1
(U(n)) Z. n _ 1.
1
(SU(n))
1
(Sp(n)) 0. n _ 1.
2
(SU(n))
2
(U(n))
2
(Sp(n)) 0. n _ 1.
2
(SO(n)) 0. n _ 3.
3
(U(2))
3
(U(k)) Z. k _ 2.
3
(SU(2))
3
(SU(k)) Z. k _ 2.
3
(Sp(1))
3
(Sp(k)) Z. k _ 1.
3
(SO(3)) Z.
6.9 Proof of the Excision Theorem
In this section we present an elementary proof of the excision theorem (6.4.1).
The proof is due to D. Puppe [46]. We derive the excision theorem from a more
6.9. Proof of the Excision Theorem 149
conceptual reformulation (6.9.3). The reformulation is more satisfactory, because
it is symmetric in Y
1
. Y
2
. In (6.4.1) we have a second conclusion with the roles
of Y
1
and Y
2
interchanged.
We begin with a technical lemma used in the proof.
A cube in R
n
, n _ 1 will be a subset of the form
W = W(a. . 1) = {. R
n
[ a
i
_ .
i
_ a
i
for i 1. a
i
= .
i
for i 1]
for a = (a
1
. . . . . a
n
) R
n
, > 0, 1 {1. . . . . n]. (1 can be empty.) We set
dimW = [1[. A face of W is a subset of the form
W
t
= {. W [ .
i
= a
i
for i 1
0
. .
}
= a
}
for 1
1
]
for some 1
0
1. 1
1
1. (W
t
can be empty.) Let dW denote the union
of all faces of W which are different from W. We use the following subsets of
W = W(a. . 1):
1
]
(W) =
. W [ .
i
< a
i

2
for at least values i 1
_
.
G
]
(W) =
. W [ .
i
> a
i

2
for at least values i 1
_
.
Here 1 _ _ n. For > dimW we let 1
]
(W) and G
]
(W) be the empty set.
(6.9.1) Lemma. Let : W Y and Y be given. Suppose that for _
dimW the inclusions
-1
() W
t
1
]
(W
t
) for all W
t
dW
hold. Then there exists a map g which is homotopic to relative to dW such that
g
-1
() 1
]
(W). (Similarly for G
]
in place of 1
]
.)
Proof. We can assume that W = 1
n
, n _ 1. We dene h: 1
n
1
n
in the
following manner: Let . = (
1
4
. . . . .
1
4
). For a ray , which begins in . we consider
its intersection 1(,) with d
_
0.
1
2
_
n
and its intersection Q(,) with d1
n
. Let h map
the segment from 1(,) to Q(,) onto the single point Q(,) and the segment from
. to 1(,) afnely to the segment from . to Q(,). Then h is homotopic relative to
d1
n
to the identity. We set g = h. Let : 1
n
and g(:) . If :
i
<
1
2
for all i ,
then : 1
n
(1
n
) 1
]
(1
n
). Suppose now that for at least one i we have :
i
_
1
2
,
then h(:) d1
n
and hence h(:) W
t
for some face W
t
with dimW
t
= n 1.
Since also h(:)
-1
(), by assumption h(:) 1
]
(W
t
). Hence we have for at
least coordinates
1
2
> h(:)
i
. By denition of h, we have h(:)
i
=
1
4
t (:
i

1
4
)
with t _ 1. We conclude that for at least coordinates
1
2
> :
i
.
The next theorem is the basic technical result. In it we deform a map 1
n
Y
into a kind of normal form. We call it the preparation theorem. Let Y be the union
of open subspaces Y
1
. Y
2
with non-empty intersection Y
0
. Let : 1
n
Y be
given. By the Lebesgue lemma (2.6.4) there exists a subdivision of 1
n
into cubes
W such that either (W) Y
1
or (W) Y
2
for each cube. In this situation we
claim:
(6.9.2) Theorem. Suppose (Y
1
. Y
0
) is -connected and (Y
2
. Y
0
) is q-connected
(. q _ 0). Then there exists a homotopy
t
of with the following properties:
(1) If (W) Y
}
, then
t
(W) Y
}
.
(2) If (W) Y
0
, then
t
is constant on W.
(3) If (W) Y
1
, then
-1
1
(Y
1
Y
0
) W 1
]1
(W).
(4) If (W) Y
2
, then
-1
1
(Y
2
Y
0
) W G
q1
(W).
Here W is any cube of the subdivision.
Proof. Let C
k
be the union of the cubes W with dimW _ k. We construct the
homotopy inductively over C
k
1.
Let dimW = 0. If (W) Y
0
we use condition (2). If (W) Y
1
. (W) ,
Y
2
, there exists a path in Y
1
from (W) to a point in Y
0
, since (Y
1
. Y
0
) is 0-
connected. We use this path as our homotopy on W. Then (1) and (3) hold.
Similarly if (W) Y
2
. (W) , Y
1
. Thus we have found a suitable homotopy
on C
0
. We extend this homotopy to the higher dimensional cubes by induction
over the dimension; we use that dW W is a cobration, and we take care of (1)
and (2).
Suppose we have changed by a homotopy such that (1) and (2) hold and (3),
(4) for cubes of dimension less than k. Call this map again . Let dimW = k. If
(W) Y
0
, we can use (2) for our homotopy. Let (W) Y
1
. (W) , Y
2
. If
dimW _ , there exists a homotopy
W
t
: W Y
1
relative to dW of [W with
W
1
(W) Y
0
, since (Y
1
. Y
0
) is -connected. If dimW > we use (6.9.1) in order
to nd a suitable homotopy of [W. We treat the case (W) Y
2
. (W) , Y
1
in
a similar manner. Again we extend the homotopy to the higher dimensional cubes.
This nishes the induction step.
Let us denote byJ(Y
1
. Y. Y
2
) the pathspace {n Y
J
[ n(0) Y
1
. n(1) Y
2
].
We have the subspace J(Y
1
. Y
1
. Y
0
).
(6.9.3) Theorem. Under the hypothesis of the previous theorem the inclusion
J(Y
1
. Y
1
. Y
0
) J(Y
1
. Y. Y
2
) is ( q 1)-connected.
Proof. Let a map : (1
n
. d1
n
) (J(Y
1
. Y. Y
2
). J(Y
1
. Y
1
. Y
0
)) be given where
n _ q 1. We have to deform this map of pairs into the subspace. By
adjunction, a map of this type corresponds to a map : 1
n
1 Y with the
following properties:
(1) (.. 0) Y
1
for . 1
n
,
(2) (X. 1) Y
2
for . 1
n
,
6.9. Proof of the Excision Theorem 151
(3) (,. t ) Y
1
for , d1
n
and t 1.
Let us call maps of this type admissible. The claim of the theorem is equivalent
to the statement, that can be deformed as an admissible map into a map with
image in Y
1
. We apply the preparation theorem to and obtain a certain map .
The deformation in (6.9.2) stays inside admissible maps. Consider the projection
: 1
n
1 1
n
. We claim that the images of
-1
(Y Y
1
) and
-1
(Y Y
2
)
under are disjoint. Let ,
-1
(Y Y
2
), , = (:) and :
-1
(Y Y
2
)W
for a cube W. Then : 1
]1
(W) and hence , has at least small coordinates. In
a similar manner we conclude from ,
-1
(Y Y
1
) that , has at least q large
coordinates. In the case that n < q the point , cannot have small and q large
coordinates.
The set
-1
(Y Y
1
) is disjoint to d1
n
, since (d1
n
) 1) . There exists
a continuous function t : 1
n
1 which assumes the value 0 on
-1
(Y Y
1
)
and the value 1 on d1
n
L (Y Y
2
). The homotopy
((.. t ). s) (.. (1 s)t st t(.))
is a homotopy of admissible maps from to a map with image in Y
1
.
(6.9.4) Theorem. Under the hypothesis of (6.9.2) the inclusion induces an isomor-
phism
}
(Y
1
. Y
0
)
}
(Y. Y
2
) for < q and an epimorphism for = q.
Proof. We have the path bration J(Y. Y. Y
2
) Y , n n(0). The pullback
along Y
1
Y yields the bration J(Y
1
. Y. Y
2
) Y
1
, n n(0). The bre over
+ is J(+. Y. Y
2
). We obtain a commutative diagram of brations:
J(+. Y
1
. Y
0
)
J(+. Y. Y
2
)
J(Y
1
. Y
1
. Y
0
)

J(Y
1
. Y. Y
2
)
=

.
The inclusion is ( q 1)-connected (see (6.9.3)). Hence has the same
connectivity (see (6.7.8)), i.e., the inclusion (Y
1
. Y
0
) (Y. Y
2
),
n
(J(+. Y
1
. Y
0
))

n
(J(+. Y. Y
2
))
n1
(Y
1
. Y
0
. +)

n1
(Y. Y
2
. +)
induces an isomorphism for n < q 1 and an epimorphism for n = q 1.

Problems
1. The hypothesis of (6.4.1) is a little different from the hypothesis of (6.9.4), since we did
not assume in (6.4.1) that (Y
1
. Y
0
) and (Y
2
. Y
0
) are 0-connected. Let Y
/
be the subset of
points that can be connected by a path to Y
0
. Show that Y
/
has the open cover Y
/
1
. Y
/
2
and the
inclusion induces isomorphisms
(Y
/
1
. Y
0
)
(Y
1
. Y
0
) and
(Y. Y
2
)
(Y
/
. Y
/
2
).
This reduces (6.4.1) to (6.9.4).
2. The map Y
0
J(Y
1
. Y
1
. Y
0
) which sends , Y
0
to the constant path with value , is
an h-equivalence.
3. The map a
1
: J(Y. Y. Y
1
) Y , n n(0) replaces the inclusion Y
1
Y by a bration.
There is a pullback diagram
J(Y
1
. Y. Y
2
)

J(Y. Y. Y
1
)
u
1
J(Y. Y. Y
2
)
u
2
Y .
Thus (6.9.3) compares the pushout Y of Y
1
Y
0
Y
2
and the pullback of a
1
. a
2
with
Y
0
. For generalizations see [73].
4. Show that the proof of (6.4.2) along the lines of this section does not need (6.9.1).
6.10 Further Applications of Excision
The excision theorem is a fundamental result in homotopy theory. For its appli-
cations it is useful to verify that it holds under different hypotheses. In the next
proposition we show and use that Y is the homotopy pushout.
(6.10.1) Proposition. Let a pushout diagram be given with a cobration ,
T
J
X
T
Y .
Suppose
i
(X. . a) = 0 for 0 < i < and each a , and
i
(. a) = 0 for
0 < i < q and each a . Then the map (J. )
+
:
n
(X. . a)
n
(Y. T. (a))
is surjective for 1 _ n _ q 2 and bijective for 1 _ n < q 2.
Proof. We modify the spaces up to h-equivalence such that (6.4.1) can be applied.
Let 7( ) = TL
(
0. 1| = T0. 1|,(a) ~ (a. 0) be the mappingcylinder
of with inclusion k : 7( ), a (a. 1) and projection : 7( ) T a
homotopy equivalence. We form the pushout diagrams
7( )
]
T
J
X
1
7
1
Y
6.10. Further Applications of Excision 153
with k = and 11 = J. Then 1 is a homotopy equivalence by (5.1.10)
and (1. ) induces an isomorphism of homotopy groups. Therefore it sufces to
analyze (1. k)
+
. The space 7 can be constructed as
7 = T L
(
0. 1| L X = 7( ) X,(a. 1) ~ a.
The map (1. k) is the composition of
(X. . a) (|0. 1| L X. |0. 1|. (a. 1)). . (.. 1)
with the inclusion | into (7. 7( ). (a. 1)). The rst map induces an isomorphism
of homotopy groups, by homotopy equivalence. In order to exhibit
n
(|) as an
isomorphism, we can pass to the base point (a. 1,2), by naturality of transport.
With this base point we have a commutative diagram
n
(|0. 1| L X. |0. 1|)

n
(7. 7( ))
n
(|0. 1| L X. |0. 1)

n
(7. T L 0. 1).
The vertical maps are isomorphism, by homotopy invariance. We apply (6.4.1) to

the bottom map. Note that
i
(|0. 1| L X. |0. 1|)
i
(X. ) and
i
(T L 0. 1. |0. 1)
i
(7( ). ).
again by homotopy invariance.
(6.10.2) Theorem (Quotient Theorem). Let X be a cobration. Let further
: (X. ) (X,. +) be the map which collapses to a point. Suppose that for
each base point a ,
i
(C. . a) = 0 for 0 < i < m.
i
(X. . a) = 0 for 0 < i < n.
Then
+
:
i
(X. . a)
i
(X,. +) is bijective for 0 < i < m n 2 and
surjective for i = n m 2.
Proof. By pushout excision,
i
(X. )
i
(X LC. C) is bijective (surjective)
in the indicated range. Note that d:
i
(C. . a)
i-1
(. a), so that the rst
hypothesis is a propertyof . The inclusionC XLCis aninducedcobration.
Since C is contractible, the projection : X LC X LC,C X, is a
(6.10.3) Corollary. Let X be a cobration. Assume that
i
() = 0 for
0 _ i _ m 1 and
i
(X) = 0 for 0 _ i _ m _ 2. Then
i
(X. )
i
(X,)
is an isomorphism for 0 < i _ 2m 1. We use this isomorphism in the exact
sequence of the pair (X. ) and obtain an exact sequence
2n-1
()
2n-1
(X)
2n-1
(X,)
2n-2
()

n1
(X)
n1
(X,)
n
() 0.
Asimilar exact sequence exists for an arbitrary pointed map : X where a typical
portion comes from the cobre sequence
i
()
(
i
(X)
(
1
i
(C( )).
We now generalize the suspension theorem. Let (X. +) be a pointed space.
Recall the suspension X and the homomorphism
+
:
n
(X)
n1
(X).
(6.10.4) Theorem. Let X be a well-pointed space. Suppose
i
(X) = 0 for 0 _
i _ n. Then
+
:
}
(X)
}1
(X) is bijective for 0 _ _ 2n and surjective
for = 2n 1.
Proof. Let CX = X1,(X1L{+]1) be the cone on X. We have an embedding
i : X CX, . .. 0| which we consider as an inclusion. The quotient CX,X
can be identied with X. From the assumption that {+] X is a cobration one
concludes that i is a cobration (Problem 1). Since CX is contractible, the exact
sequence of the pair (CX. X) yields an isomorphism d:
}1
(CX. X) .
}
(X).
The inverse isomorphism sends an element represented by : 1
n
X to the
element represented by id(1). From this fact we see
+
=
+
d
-1
:
}
(X)
}1
(CX. X)
J

}1
(X).
with the quotient map : CX CX,X = X. We can therefore prove the
theorem by showing that
+
is bijective or surjective in the same range. This
follows from the quotient theorem (6.10.2).
(6.10.5) Theorem. Let X and Y be well-pointed spaces. Assume
i
(X) = 0 for
i < (_ 2) and
i
(Y ) = 0 for i < q (_ 2). Then the inclusion X Y
X Y induces an isomorphism of the
i
-groups for i _ q 2. The groups
i
(X Y. X Y ) and
i
(X . Y ) are zero for i _ q 1.
Proof. We rst observe that :
i
(X Y )
i
(X Y ), induced by the inclu-
sion, is always surjective. The projections onto the factors induce isomorphisms
k :
i
(X Y )
i
(X)
i
(Y ). Let
A
: X X Y and
Y
: Y X Y
denote the inclusions. Let
s :
i
(X)
i
(Y )
i
(X Y ). (.. ,)
A
+
(.)
Y
+
(,).
Then sk is right inverse to . Hence the exact sequence of the pair (X Y. X Y )
yields an exact sequence
(+) 0
i1
(X Y. X Y )
i
(X Y )
i
(X Y ) 0.
In the case that i _ 2, the sequence splits, since we are then working with abelian
groups; hence
i
(X Y )
i
(X)
i
(Y )
i1
(X Y. X Y ). i _ 2.
Since the spaces are well-pointed, we can apply the theoremof Seifertvan Kampen
to (X Y. X. Y ) and see that
1
(X Y ) = 0. We now consider the diagram
i
(X Y. Y )
i
(X)
(1)
q
q
q
q
q
q
q
q
q
q
q

i
(X Y )

i
(X Y. X)
i
(Y )
(2)
m
m
m
m
m
m
m
m
m
m
m
m
with exact row and column. The diagonal arrows are always injective and split; this
is seen by composing with the projections.
Since the spaces are well-pointed, we can apply the pushout excision to the triad
(X Y. X. Y. +). It says that (1) and (2) are surjective for i _ q 2, and hence
bijective (since we already know the injectivity).
We now apply the Sum Lemma (11.1.2) to the diagram and conclude that
(
A
+
.
Y
+
) is an isomorphism, and therefore also the map of the theorem is an iso-
morphism. The exact sequence nowyields
i
(XY. XY ) = 0 for i _ q1.
We apply (6.10.2) to
i
(X Y. X Y )
i
(X . Y ). By what we have
already proved, we can apply this theorem with the data n = q 1 and
m = min( 1. q 1). We also need that X Y X Y is a cobration. This
is a consequence of the product theorem for cobrations.
(6.10.6) Proposition. Let (Y
}
[ J) be the family of path components of Y and
c
}
: Y
}
Y the inclusion. Then
(c
}
+
):

}J

k
(Y
}
. S
n
)
k
(Y
. S
n
)
is an isomorphism for k _ n.
Proof. Suppose Y is the topological sum of its path components. Then we have
a homeomorphism Y
. S
n
_
}J
Y
}
. S
n
, and the assertion follows for
nite J by induction on the cardinality of J from (6.10.5) and for general J then
by a compactness argument. For general Y it sufces to nd a 1-connected map
X Y such that X is the topological sum of its path components, because then
X
.S
n
Y
.S
n
is (n 1)-connected by (6.7.10) (and similarly for the path
components).
(6.10.7) Proposition. Let Y be k-connected (k _ 0) and 7 be l-connected
(l _ 1) and well-pointed. Then the natural maps
}
(7)
}
(Y 7. Y +)
}
(Y 7,Y +)
}
(7)
are isomorphisms for 0 < _ k l 1.
Proof. The rst map is always bijective for _ 1; this is a consequence of the
exact sequence of the pair (Y 7. Y +) and the isomorphism
}
(Y )
}
(7)
}
(Y 7). Since the composition of the maps is the identity, we see that the second
mapis always injective andthe thirdone surjective. Thus if
+
:
}
(Y 7. Y +)
}
(Y 7,Y + is surjective, then all maps are bijective. From our assumption
about 7 we conclude that
}
(Y 7. Y +) = 0 for 0 < _ l (thus there is no
condition for l = 0. 1). The quotient theorem now tells us that
+
is surjective
for 0 < _ k l 1.
(6.10.8) Corollary. Let Y be path connected. The natural maps
k
(S
n
)
k
(Y S
n
. + S
n
)
k
(Y S
n
, + S
n
)
k
(S
n
)
are isomorphisms for 1 _ k _ n.
(6.10.9) Proposition. Suppose
i
(X) = 0 for i < (_ 0) and
i
(Y ) = 0 for
i < q (_ 0). Then
i
(X = Y ) = 0 for i < q 1.
Proof. In the case that = 0 there is no condition on X. From the denition of
the join we see that X = Y is always path connected. For = 0 we claim that
i
(X = Y ) = 0 for i < q 1. Consider the diagram
X
X Y
pr
pr
pr
X X

{+]
and apply (6.7.9). In the general case the excision theorem says that the map
i
(CX Y. X Y )
i
(X =Y. X CY ) is an epimorphism for i < q 1.
Now use diagram chasing in the diagram
i
(X CY )

i
(X = Y )

i
(X = Y. X CY )

i-1
(X CY )
i
(X Y )

i
(CX Y )

i
(CX Y. X Y )

i-1
(X Y )
(a morphism between exact homotopy sequences).

The excision theorem in the formulation of (6.9.4) has a dual. Suppose given a
pullback diagram
1
T
X
(
T
with brations and g. The double mapping cylinder 7(J. G) can be considered
as the brewise join of and g. It has a canonical map j: 7(J. G) T.
(6.10.10) Proposition. Suppose is -connected and g is q-connected. Then j is
q 1-connected.
Proof. Use bre sequences and (6.10.9).
Problems
1. Let
0
(X) = 0 and
i
(Y ) = 0 for i < q(_ 2). Then
i
(X)
i
(X Y ) is an
isomorphism for i < q. Show also
2
(X Y. X Y ) = 0.
2. Let X and Y be 0-connected and well-pointed. Show
1
(X . Y ) = 0.
3. Show that
3
(D
2
. S
1
)
3
(D
2
,S
1
) is not surjective.
4. Show
1
(S
2
S
1
. S
1
) = 0. Showthat
2
(S
2
S
1
. S
1
)
2
(S
2
S
1
,S
1
)
2
(S
2
)
is surjective but not injective.
5. For X = Y = S
1
and i = 1 the sequence (+) does not split. The fundamental group
1
(S
1
S
1
) = Z+ Z has no subgroup isomorphic to ZZ.
6. Show that the diagram
X d1 L {+] 1
)
X
i
X 1

CX
with (.. 0) = .. (.. 1) = +. (+. t ) = + is a pushout.
7. If X is well-pointed, then X is well-pointed.
8. Some hypothesis like e.g. well-pointed is necessary in both (6.10.1) and (6.10.4). Let
= {0] L {n
-1
[ n Z] and = 0 X = 1, 1 with base point (0. 0).
Then
1
(X) and
1
() are uncountable;
:
0
(X)
1
(X) is not surjective. Note:
X is a cobration and X, is well-pointed.
9. Let e
1
. . . . . e
n1
be the standard basis of unit vectors in R
n1
, and let e
1
be the base
point of S
n
. A pointed homeomorphism h
n
: S
n
S
n1
is
h
n
: S
n
S
n1
. (.. t )
1
2
(e
1
.)
1
2
cos 2t (e
1
.)
1
2
[e
1
.[ sin 2t e
n2
where R
n1
= R
n1
0 R
n2
.
10. Let 1 R
n1
be compact. Show that each map : 1 S
n
has an extension to the
complement R
n1
1 of a nite set 1. One can choose 1 such that each component of
R
n1
1 contains at most one point of 1.
11. Determine
2n-1
(S
n
S
n
) for n _ 2.
12. Let
be a self-map of S
n()
. Show J(
1
=
2
) = J(
1
)J(
2
).
13. Let H:
3
(S
2
) Z be the isomorphism which sends (the class of the) Hopf map
j: S
3
S
2
to 1 (the Hopf invariant). Show that for : S
3
S
3
and g: S
2
S
2
the
relations H( ) = J( )H() and H(g ) = J(g)
2
H() hold.
Chapter 7
Stable Homotopy. Duality
The suspension theorem of Freudenthal indicates that homotopy theory simplies
by use of iterated suspensions. We use this idea to construct the simplest stable ho-
motopy category. Its construction does not need extensive technical considerations,
yet it has interesting applications. The term stable refers to the fact that iteration
of suspension induces after a while a bijection of homotopy classes.
We use the stable category to give an introduction to homotopical duality theory.
In this theory the stable homotopy type of a closed subspace X R
n
and its
complement R
n
X are compared. This elementary treatment of duality theory is
based on ideas of Albrecht Dold and Dieter Puppe; see in particular [54]. It is related
to the classical Alexander duality of homology theory and to SpanierWhitehead
duality.
We introduced a naive form of spectra and us them to dene spectral homology
and cohomology theories. The homotopical Euclidean complement duality is then
used to give a simple proof for theAlexander duality isomorphism. In a later chapter
we reconsider duality theory in the context of product structures.
7.1 A Stable Category
Pointedspaces X andY are calledstably homotopy equivalent, insymbols X .
x
Y ,
if there exists an integer k _ 0 such that the suspensions
k
X and
k
Y are
homotopy equivalent. Pointed maps . g: X Y are called stably homotopic,
in symbols .
x
g, if for some integer k the suspensions
k
and
k
g are
homotopic. We state some of the results to be proved in this chapter which use
these notions.
(7.1.1) Theorem (Stable Complement Theorem). Let X and Y be homeomorphic
closed subsets of the Euclidean space R
n
. Then the complements R
n
X and R
n
Y
are either both empty or they have the same stable homotopy type with respect to
arbitrary base points.
In general the complements themselves can have quite different homotopy type.
A typical example occurs in knot theory, the case that X Y S
1
are subsets
of R
3
. On the other hand the stable homotopy type still carries some interesting
geometric information: see (7.1.10).
(7.1.2) Theorem (Component Theorem). Let X and Y be closed homeomorphic
subsets of R
n
. Then
0
(R
n
X) and
0
(R
n
Y ) have the same cardinality.
160 Chapter 7. Stable Homotopy. Duality
Later we give another proof of Theorem (7.1.2) based on homology theory, see
(10.3.3). Fromthe component theoremone can deduce classical results: The Jordan
separation theorem (10.3.4) and the invariance of domain (10.3.7).
Theorem (7.1.1) is a direct consequence of (7.1.3). One can also compare
complements in different Euclidean spaces. The next result gives some information
about how many suspensions sufce.
(7.1.3) Theorem. Let X R
n
and Y R
n
be closed subsets and h: X Y a
homeomorphism. Suppose n _ m. Then the following holds:
(1) If R
n
,= X, then R
n
,= Y , and h induces a canonical homotopy equivalence
n1
(R
n
X) .
n1
(R
n
Y ) with respect to arbitrary base points.
(2) If R
n
= X and R
n
,= Y , then n < mand
n1
(R
n
Y ) . S
n
, i.e., R
n
Y
has the stable homotopy type of S
n-n-1
.
(3) If R
n
= X and R
n
= Y , then n = m.
In many cases the number of suspensions is not important. Since it also depends
on the situation, it is convenient to pass from homotopy classes to stable homotopy
classes. This idea leads to the simplest stable category.
The objects of our new category ST are pairs (X. n) of pointed spaces X and
integers n Z. The consideration of pairs is a technical device which allows for a
better formulation of some results. Thus we should comment on it right now.
The pair (X. 0) will be identied with X. The subcategory of the objects
(X. 0) = X with morphisms the so-called stable homotopy classes is the geo-
metric input. For positive n the pair (X. n) replaces the n-fold suspension
n
X.
But it will be convenient to have the object (X. n) also for negative n (desuspen-
sion). Here is an interesting example. In the situation of (7.1.3) the homotopy
equivalence
n1
(R
n
X)
n1
(R
n
Y ) induced by h represents in the cat-
egory ST an isomorphismh
-
: (R
n
X. n) (R
n
Y. m). In this formulation
it then makes sense to say that the assignment h h
-
is functor. (Otherwise we
would have to use a mess of different suspensions.) Thus if X is a space which
admits an embedding i : X R
n
as a proper closed subset for some n, then the
isomorphism type of (R
n
i(X). n) in ST is independent of the choice of the
embedding. Hence we have associated to X a dual object in ST (up to canonical
isomorphism).
Let
t
X = X . S
t
be the t -fold suspension of X. As a model for the sphere
S
t
we use either the one-point compactication R
t
L {o] or the quotient space
S(t ) = 1
t
,d1
t
. In these cases we have a canonical associative homeomorphism
S
o
. S
b
S
ob
which we usually treat as identity. Suppose n. m. k Z are
integers such that n k _ 0. mk _ 0. Then we have the suspension morphism
: X.S
nk
. Y .S
nk
|
0
X.S
nk1
. Y .S
nk1
|
0
. .id(S
1
).
We form the colimit over these morphisms, colim
k
X . S
nk
. Y . S
nk
|
0
. For
nk _ 2 the set X .S
nk
. Y .S
nk
|
0
carries the structure of an abelian group
7.1. A Stable Category 161
and is a homomorphism. The colimit inherits the structure of an abelian group.
We dene as morphism group in our category ST
ST((X. n). (Y. m)) = colim
k
X . S
nk
. Y . S
nk
|
0
.
Formation of the colimit means the following: An element of ST((X. n). (Y. m))
is represented by pointed maps
k
: X . S
nk
Y . S
nk
, and
k
,
I
. l _
k represent the same element of the colimit if
I-k
k
.
I
. Composition of
morphisms is dened by composition of representatives. Let
k
: X . S
nk
Y . S
nk
and g
I
: Y . S
nI
7 . S
]I
be representatives of morphisms and
let r _ k. l. Then the following composition of maps represents the composition
of the morphisms (dotted arrow):
X . S
ni
= X . S
nk
. S
i-k
-k
(
k
Y . S
nk
. S
i-k
=
7 . S
]i
= 7 . S
]I
. S
i-I
Y . S
nI
. S
i-I
.
-l
One veries that this denition does not depend on the choice of representatives.
The group structure is compatible with the composition
(
1
2
) =
1

2
. (
1
2
) =
1

2
.
The category ST has formal suspension automorphisms
]
: ST ST, Z
(X. n) (X. n ).
]
.
If : (X. n) (Y. m) is represented by
k
:
nk
X
nk
Y (with nk _ 0,
mk _ 0, k _ [[), then
]
is represented by
(
]
)
k
=
]
(
k
):
nk]
X
nk]
Y. _ 0.
(
]
)
k[][
=
k
:
nk][][
X
nk][][
Y. _ 0.
The rules
0
= id(ST) and
]

q
=
]q
show that
]
is an automorphism.
For > 0 we call
]
the -fold suspension and for < 0 the -fold desuspension.
We have a canonical isomorphism k
]
: (X. n) (
]
X. n ); it is represented
by the identity X . S
nk
(X . S
]
) . S
nk-]
for n k _ _ 0. We write
X for the object (X. 0). Thus for positive n the object (X. n) can be replaced by
n
X.
(7.1.4) Example. Pointed spaces X. Y are stably homotopy equivalent if and only
if they are isomorphic in ST. The image ST( ) of : X Y in ST(X. Y ) is called
the stable homotopy class of . Maps . g: X Y are stably homotopic if and
only if they represent the same element in ST(X. Y ). The groups ST(S
k
. S
0
) =
colim
n
nk
(S
n
) are the stable homotopy groups of the spheres.
(7.1.5) Example. It is in general difcult to determine morphismgroups in ST. But
we know that the category in non-trivial. The suspension theorem and the degree
theorem yield
ST(S
n
. S
n
) = colim
k
S
nk
. S
nk
|
0
Z.
The composition of morphisms corresponds to multiplication of integers.
(7.1.6) Proposition. Let Y be pathwise connected. We have the embeddingi : S
n
. S
n
, . (+. .) and the projection : Y
. S
n
S
n
, (,. .) . with
i = id. They induce isomorphisms of the
k
-groups for k _ n _ 1.
Proof. Let n = 1. Then Y
.S
1
Y S
1
,Y {+] is path connected. The base
point of Y
is non-degenerate. Hence the quotient

t
(Y
) (Y
) from the
unreduced suspension to the reduced suspension is an h-equivalence. The projection
Y 1 onto a point induces a 2-connected map between double mapping cylinders
t
(Y
) = 7(+ Y {+] +) 7
t
(+ 1 {+] +) =
t
(1
) S
1
.
From this fact one deduces the assertion for n = 1.
We now consider suspensions
k
(S
n
)
i
k
(Y
. S
n
)

k
(S
n
)
k1
(S
n1
)
i

k1
(Y
. S
n1
)
]

k1
(S
n1
).
The vertical morphisms are bijective (surjective) for k _ 2n2 (k = 2n1). For
n = 1
1
(Y
. S
1
) Z. Since
2
(Y
. S
2
) contains
2
(S
2
) Z as a direct
summand, we conclude that
Y
is an isomorphism. For n _ 2 we can use directly
the suspension theorem (6.10.4).
(7.1.7) Proposition. Let (Y
}
[ J) be the family of path components of Y and
c
}
: Y
}
Y the inclusion. Let n _ 2. Then
(c
}
+
):

}J
k
(Y
}
. S
n
)
k
(Y
. S
n
)
is an isomorphism for 0 _ k _ n. In particular
n
(Y
. S
n
) is a free abelian
group of rank [
0
(Y )[.
Proof. (7.1.6) and (6.10.6).
(7.1.8) Proposition. Let Y be well-pointed and n _ 2. Then
n
(Y . S
n
) is
a free abelian group of rank [
0
(Y )[ 1 and the suspension
n
(Y . S
n
)
n1
(Y . S
n1
) is an isomorphism.
7.1. A Stable Category 163
From the exact homotopy sequence of the pair (Y
.S
n
. S
n
) we conclude that
k
(Y
. S
n
. S
n
) = 0 for 0 < k < n. The quotient theorem (6.10.2) shows that
k
(Y
. S
n
. S
n
)
k
(Y
. S
n
,S
n
)
k
(Y . S
n
) is bijective (surjective)
for 0 < k _ 2n 2 (k = 2n 1). From the exact sequence 0
n
(S
n
)
n
(Y
.S
n
)
n
(Y
.S
n
. S
n
) 0 we deduce a similar exact sequence where
the relative group is replaced by
n
(Y .S
n
). The inclusion of
n
(S
n
) splits. Now
we can use (7.1.7).
(7.1.9) Corollary. Let X be a well-pointed space. Then ST(S
0
. X) is a free abelian
group of rank [
0
(X)[ 1.
The groupST(S
0
. X) onlydepends onthe stable homotopytype of X. Therefore
we can state:
(7.1.10) Corollary. Let X and Y be well-pointed spaces of the same stable homo-
topy type. Then [
0
(X)[ = [
0
(Y )[. Therefore (7.1.2) is a consequence of (7.1.3).
The category ST has a product structure induced by the smash product. The
category ST together with this additional structure is called in category theory
a symmetric tensor category (also called a symmetric monoidal category). The
tensor product of objects is dened by
(X. m) (Y. n) = (X . Y. mn).
Let
k
: X . S
nk
X
t
. S
n
/
k
. g
I
: Y . S
nI
Y
t
. S
n
/
I
be representing maps for morphisms : (X. m) (X
t
. m
t
) and g: (Y. n)
(Y
t
. n
t
). A representing morphism ( g)
kI
is dened to be (1)
k(nn
/
)
times
the composition t
t
(
k
. g
I
) t (dotted arrow)
X . Y . S
nknI
r
X . S
nk
. Y . S
nI
(
k
.
l
X
t
. Y
t
. S
n
/
kn
/
I
X
t
. S
n
/
k
. Y
t
. S
n
/
I
r
/
where t and t
t
interchange two factors in the middle. Now one has to verify:
(1) The denition does not depend on the representatives; (2) the functor property
(
t
g
t
)( g) =
t
g
t
g holds; (3) the tensor product is associative. These
requirements make it necessary to introduce signs in the denition. The neutral
object is (S
0
. 0). The symmetry c : (X. m) (Y. n) (Y. n) (X. m) is (1)
nn
times the morphism represented by the interchange map X . Y Y . X.
Problems
1. The spaces S
1
S
1
and S
1
S
1
S
2
are not homotopy equivalent. They have different
fundamental group. Their suspensions are homotopy equivalent.
2. The inclusion X Y X
Y induces for each pointed space Y a homeomorphism

(X Y ),(X {+]).
3. Let X and Y be well-pointed spaces. Then Y (X Y ),(X {+]), , (+. ,) is a
cobration.
4. Let 1 be a point. We have an embedding 1
. Y X
. Y and a canonical homeo-

morphism X . Y X
. Y,1
. Y .
7.2 Mapping Cones
We need a few technical results about mapping cones. Let : X Y be a pointed
map. We use as a model for the (unpointed) mapping cone C( ) the double mapping
cylinder 7(Y X +); it is the quotient of Y X 1 {+] under the relations
(.) ~ (.. 0). (.. 1) ~ +. The image of + is the basepoint. For an inclusion
| : X we write C(X. ) = C(|). For empty we have C(X. 0) = X
.
Since we will meet situations where products of quotient maps occur, we work in
the category of compactly generated spaces where such products are again quotient
maps. The mapping cone is a functor C : TOP(2) TOP
0
; a map of pairs
(J. ): (X. ) (Y. T) induces a pointed map C(J. ): C(X. ) C(Y. T),
and a homotopy (J
t
.
t
) induces a pointed homotopy C(J
t
.
t
). We note for further
use a consequence of (4.2.1):
(7.2.1) Proposition. If J and are h-equivalences, then C(J. ) is a pointed
h-equivalence.
(7.2.2) Example. We write C
n
= C(R
n
. R
n
0). This space will be our model for
the homotopy type of S
n
. In order to get a homotopy equivalence C
n
S
n
, we ob-
serve that S
n
is homeomorphic to the double mapping cylinder 7(+ S
n-1
+).
We have the canonical projection from C
n
= 7(R
n
R
n
0 +). An explicit
homotopy equivalence is (.. t ) (sin t
x
|x|
. cos t ), . (0. . . . . 0. 1).
(7.2.3) Example. Let X R
n
be a closed subspace. Then
C(R
n
. R
n
X) = 7(R
n
R
n
X +) . 7(+ R
n
X +) =
t
(R
n
X).
the unpointed suspension. If X = R
n
, then this space is h-equivalent to S
0
. If
X ,= R
n
, then R
n
X is well-pointed with respect to any point and
t
(R
n
X) is
h-equivalent to the pointed suspension (R
n
X).
We are mainly interested in the homotopy type of C(X. ) (under {+] X).
It is sometimes convenient to provide the set C(X. ) with a possibly different
7.2. Mapping Cones 165
topology which does not change the homotopy type. Set theoretically we can view
C(X. ) as the quotients C
1
(X. ) = (X 0 L 1), 1 or C
2
(X. ) =
(X 0 L 1 L X 1),X 1. We can provide C
1
and C
2
with the quotient
topology. Then we have canonical continuous maps : C(X. ) C
1
(X. ) and
q : C
1
(X. ) C
2
(X. ) which are the identity on representative elements.
(7.2.4) Lemma. The maps and q are homotopy equivalences under {+] X.
Proof. Dene : C
1
(X. ) C(X. ) by
(.. t ) = .. t _ 1,2. (a. t ) = (a. max(2t 1. 0)). (a. 1) = +.
One veries that this assignment is well-dened and continuous. A homotopy
. id is given by ((.. t ). s) (.. st (1 s) max(2t 1. 0)). A similar
formula works for . id. Dene q : C
2
(X. ) C
1
(X. ) by
q(.. t ) = (.. min(2t. 1)). t < 1. q(.. t ) = + = { 1]. t _ 1,2.
Again linear homotopies in the t -coordinate yield homotopies from q q and qq to
the identity.
(7.2.5) Proposition(Excision). Let U X andsuppose there exists afunction
t : X 1 such that U t
-1
(0) and t
-1
0. 1 . Then the inclusion of pairs
induces a pointed h-equivalence g: C(X U. U) C(X. ).
Proof. Set o(.) = max(2t(.) 1. 0). A homotopy inverse of g is the map
: (.. t ) (.. o(.)t ). The denition of uses the notation C
2
for the map-
ping cone. The homotopies fromg and g to the identity are obtained by a linear
homotopy in the t -coordinate.
(7.2.6) Remark. Mapping cones of inclusions are used at various occasions to
relate the category TOP(2) of pairs with the category TOP
0
of pointed spaces. We
make some general remarks which concern the relations. They will be relevant for
the investigation of homology and cohomology theories.
Let

h: TOP
0
C be a homotopy invariant functor. We dene an associated
functor h = 1

h: TOP(2) C by composition with the mapping cone functor
(X. ) C(X. ). The functor 1

h is homotopy invariant in a stronger sense: If
: (X. ) (Y. T) is a map of pairs such that the components : X Y and
: T are h-equivalences, then the induced map h(X. ) h(Y. T) is an
isomorphism (see (7.2.1)). Moreover h satises excision: Under the hypothesis of
(7.2.5) the inclusion induces an isomorphism h(X U. U) h(X. ).
Conversely, let h: TOP(2) C be a functor. We dene an associated functor
1h =

h on objects by 1h(X) = h(X. +) and with the obvious induced morphisms.
If h is homotopy invariant, then also 1h.
The composition 11is given by 11h(X. ) = h(C(X. ). +). We have natural
morphisms
h(C(X. ). +) h(C(X. ). C) h(C(X. ) U. C U) h(X. ).
Here Cis the cone onandU Cis the subspace witht -coordinates in1,2. 1|.
If h is strongly homotopy invariant and satises excision, then these morphisms are
isomorphisms, i.e., 11 is naturally isomorphic to the identity.
The composition 11 is given by 11

h(X) =

h(C(X. +)). There is a canonical
projection C(X. +) X. It is a pointed h-equivalence, if the inclusion {+] X
is a cobration. Thus if

h is homotopy invariant, the composition 11 is naturally
isomorphic to the identity on the subcategory of well-pointed spaces.
Let (X. ) and (Y. T) be two pairs. We call Y. X T excisive in X Y
if the canonical map : 7( Y T X T) Y L X T is a
(7.2.7) Proposition (Products). Let ( Y. X T) be excisive. Then there exists
a natural pointed homotopy equivalence
: C(X. ) . C(Y. T) C((X. ) (Y. T)).
It is dened by the assignments
(.. ,) (.. ,).
(a. s. ,) (a. ,. s).
(.. b. s) (.. b. s).
(a. s. b. t ) (a. b. max(s. t )).
(See the proof for an explanation of notation).
Proof. In the category of compactly generated spaces C(X. ) . C(Y. T) is a
quotient of
X Y 1 Y X T 1 1 T 1
under the following relations: (a. 0. ,) ~ (a. ,), (.. b. 0) ~ (.. b), (a. 0. b. t ) ~
(a. b. t ), (a. s. b. 0) ~ (a. s. b), and 1 T 1 L 1 T 1 is identied
to a base point +.
In a rst step we show that the smash product is homeomorphic to the double
mapping cylinder 7(X Y
t]
7 {+]) where
7 = 7( Y T X T).
This space is the quotient of
X Y ( Y T 1 X T) 1 T (1 1,1 0)
7.2. Mapping Cones 167
under the following relations: (.. b. 0) ~ (.. b), (a. ,. 0) ~ (a. ,), (a. b. t. 0) ~
(a. b), (a. b. 1. s) ~ (a. b. s), and (Y T 1 X T) 1 is identied
to a base point +.
The assignment
1 1 1 1. (u. )
_
(2u. ). u _ 1,2.
(. 2(1 u)). u _ 1,2.
induces a homeomorphism;
0
: 1 1,(1 0) 1 1. Its inverse
0
has the form
1 1 {(0. 0)] 1 (1 {0]). (s. t )
_
(1 t ,2s. s). s _ t.
(s,2t. t ). s _ t.
A homeomorphism : C(X. ) .C(Y. T) C(|) is now dened by (.. ,) =
(.. ,), (a. s. ,) = (a. ,. s), (.. b. t ) = (.. b. t ), (a. s. b. t ) = (a. b.
0
(s. t )).
The diagram
X Y
=
7
t]

]
{+]
=
X Y Y L X T
t

{+]
induces : C(|) C(|). It is a pointed h-equivalence if is an h-equivalence.
One veries that = .
(7.2.8) Remark. The maps are associative: For three pairs (X. ). (Y. T). (7. C)
the relation ( . id) = (id .) holds. They are also compatible with the
interchange map. Finally, they yield a natural transformation.
Problems
1. Verify that the map in the proof of 7.2.5 is continuous. Similar problem for the
homotopies.
2. Let (J. ): (X. ) (Y. T) be a map of pairs. If J is n-connected and (n 1)-
connected, then C(J. ) is n-connected.
3. Let X = L T and suppose that the interiors
. T
still cover X. Then the inclusion

induces a weak homotopy equivalence C(T. T) C(X. ).
4. Construct explicit h-equivalences C
n
R
n
L {o] = S
(n)
such that
C
m
. C
n

C
mn
S
(m)
. S
(n)

S
(mn)
is homotopy commutative.
7.3 Euclidean Complements
This section is devoted to the proof of (7.1.3). We need an interesting result from
general topology.
(7.3.1) Proposition. Let R
n
and T R
n
be closed subsets and let : T
be a homeomorphism. Then there exists a homeomorphism of pairs
J : (R
n
R
n
. 0) (R
n
R
n
. T 0)
such that J(a. 0) = ((a). 0) for a .
Proof. By the extension theorem of Tietze (1.1.2) there exists a continuous exten-
sion : R
n
R
n
of : T R
n
. The maps
: R
n
R
n
R
n
R
n
. (.. ,) (.. , (.))
are inverse homeomorphisms. Let G( ) = {(a. (a)) [ a ] denote the graph
of . Then
sends 0 homeomorphically to G( ) by (a. 0) (a. (a)).

Let [: R
n
R
n
be a Tietze extension of the inverse g of . Then we have
similar homeomorphisms
: R
n
R
n
R
n
R
n
. (,. .) (,. . [(,)).
The desiredhomeomorphismJ is the composition
-
t
where t interchanges
R
n
and R
n
(and sends G( ) to G(g)).
Let X R
n
and Y R
n
be closed subsets and : X Y a homeomorphism.
The induced homeomorphism J from (7.3.1) can be written as a homeomorphism
J : (R
n
. R
n
X) (R
n
. R
n
0) (R
n
. R
n
Y ) (R
n
. R
n
0).
We apply the mapping cone functor to J and use (7.2.2) and (7.2.7). The result is
a homotopy equivalence
C(R
n
. R
n
X) . S
n
. C(R
n
. R
n
Y ) . S
n
.
If X ,= R
n
and Y ,= R
n
we obtain together with (7.2.3)
n1
(R
n
X) .
n1
(R
n
Y ).
If X ,= R
n
then we have C(R
n
. R
n
X) . (R
n
X), and if X = R
n
then
we have C(R
n
. R
n
X) . S
0
.
Suppose X ,= R
n
but Y = R
n
. Then
n1
(R
n
X) . S
n
. Since n _ m the
homotopy group
n
(
n1
(R
n
X)) = 0 and
n
(S
n
) Z. This contradiction
shows that Y ,= R
n
.
7.4. The Complement Duality Functor 169
Suppose X = R
n
and Y ,= R
n
. Then n = mis excluded by the previous proof.
Thus
S
n
. C(R
n
. R
n
X) . S
n
. C(R
n
. R
n
Y ) . S
n
.
n1
(R
n
Y ).
If X = R
n
and Y = R
n
, then
S
n
.
n
C(R
n
. R
n
X) .
n
C(R
n
. R
n
Y ) . S
n
and therefore m = n.
This nishes the proof of (7.1.3).
7.4 The Complement Duality Functor
The complement duality functor is concerned with the stable homotopy type of
Euclidean complements R
n
X for closed subsets X R
n
. We consider an
associated category E. The objects are pairs (R
n
. X) where X is closed in R
n
.
A morphism (R
n
. X) (R
n
. Y ) is a proper map : X Y . The duality
functor is a contravariant functor D: E ST which assigns to (R
n
. X) the
object
-n
C(R
n
. R
n
X) = (C(R
n
. R
n
X). n). The associated morphism
D( ):
-n
C(R
n
. R
n
Y )
-n
C(R
n
. R
n
X) will be constructed via a
representing morphism D( )
nn
. Its construction needs some preparation.
Given the data X R
n
. Y R
n
and a proper map : X Y . Henceforth
we use the notation [T = (. T) for pairs T . Note that in this notation
[T C[D = T[C D. The basic step in the construction of the functor will
be an associated homotopy class
D
#
: R
n
[D
n
R
n
[Y R
n
[X R
n
[0.
Here D
n
again denotes the n-dimensional standard disk. A scaling function for a
proper map : X Y is a continuous function : Y |0. o with the property
((.)) _ [.[. . X.
The next lemma shows the existence of scalingfunctions withanadditional property.
(7.4.1) Lemma. There exists a positive continuous function [: 0. o |0. o
such that the inequality [([ .[) _ [.[ holds for . X. A scaling function in
the sense of the denition is then , [([,[).
Proof. The set
-1
D(t ) = max{. X [ [ .[ _ t ] is compact, since is proper.
Let

[(t ) be its norm maximum max{[.[ [ . X. [ .[ _ t ]. Then

[([ .[) =
max{[a[ [ a X. [a[ _ [ .[] _ [.[. The function

[: 0. o 0. o is
increasing. There exists a continuous increasing function [: 0. o |0. o such
that [(t ) _

[(t ) for each t _ 0.
The set of scaling functions is a positive convex cone. Let
1
.
2
be scaling
functions and 0 _ z _ 1; then z
1
(1 z)
2
is a scaling function. Let _ ;
if is a scaling function then also .
Let be a scaling function and set M() = {(.. ,) [ (,) _ [.[]. Then we
have a homeomorphism
R
nn
[D
n
Y R
nn
[M(). (.. ,) ((,) .. ,).
The graph G( ) = {(.. .) [ . X] of is contained in M(). We thus can
continue with the inclusion and obtain a map D
1
(. ) of pairs
R
nn
[D
n
Y R
nn
[G( ). (.. ,) ((,) .. ,).
The homotopy class of D
1
(. ) does not depend on the choice of the scaling
function: If
1
.
2
are scaling functions, then
(.. ,. t ) ((t
1
(,) (1 t )
2
(,)) .. ,)
is a homotopy from D
1
(.
2
) to D
1
(.
1
). A continuous map : X Y has a
Tietze extension

: R
n
R
n
. The homeomorphism (.. ,) (.. ,

(.)) of
R
nn
sends (.. (.)) to (.. 0). We obtain a homeomorphism of pairs
D
2
(.

): R
nn
[G( ) R
nn
[X 0.
The homotopy class is independent of the choice of the Tietze extension: The
homotopy (.. ,. t ) (.. , (1 t )

1
(.) t

2
(.)) proves this assertion. The
duality functor will be based on the composition
D
#
( ) = D
2
(.

) D
1
(. ): R
n
[D
n
R
n
[Y R
n
[X R
n
[0.
We have written D
#
( ), since the homotopy class is independent of the choice of the
scaling function and the Tietze extension. The morphism D :
-n
C(R
n
[Y )
-n
C(R
n
[X) is dened by a representative of the colimit:
(D )
nn
:
n
C(R
n
[Y )
n
C(R
n
[X).
Consider the composition
C(R
n
[Y ) . C
n
r
(-1)
nm
(T( )
nm
C
n
. C(R
n
[Y ) C(R
n
[D
n
) . C(R
n
[Y )
:
C(R
n
[X) . C
n
C(R
n
[X R
n
[0)
-1
C(R
n
[D
n
R
n
[Y ).
CT
#
(
Explanation. t interchanges the factors; the inclusion R[D

n
R
n
[0 induces a
homotopy equivalence C(R
n
[D
n
) C
n
; the morphisms comes from (7.2.7);
and CD
#
is obtained by applying the mapping cone to D
#
; nally, we multiply
the homotopy class of the composition by (1)
nn
. We take the freedom to use
(D )
nn
: C(R
n
[Y ) . C
n
C(R
n
[X) . C
n
as our model for D , i.e., we do
not compose with the h-equivalences of the type C
n
S
n
obtained in (7.2.2).
(7.4.2) Lemma. Let be an inclusion, : X Y R
n
. Then D has as
a representative the map C(R
n
[Y ) C(R
n
[X) induced by the inclusion. In
particular the identity of X is send to the identity.
Proof. We take the scaling function , [,[ 1 and extend by the identity.
Then D
#
( ) is the map (.. ,) (([,[ 1) .. , ([,[ 1) .). The map D
1
is (.. ,) (([,[ 1) .. ,) and the homotopy
(.. ,. t ) ((1 t )([,[ 1) . t (. ,). ,)
is a homotopy of pairs fromD
1
to (.. ,) (.,. ,). Hence D
2
D
1
is homotopic
to (.. ,) (. ,. .) and the homotopy (.. ,. t ) ((1 t ). ,. .) shows
it to be homotopic to (.. ,) (,. .). Now interchange the factors and observe
that . . has the degree (1)
n
= (1)
nn
.
Next we consider the case of a homeomorphism : X Y with inverse g.
Let g: R
n
R
n
be a Tietze extension of g. Then:
(7.4.3) Lemma. The maps (.. ,) (([,[ .. ,) and (.. ,) (. g(,). ,) are
as maps of pairs R
n
[D
n
R
n
[Y R
nn
[G( ) homotopic. Here [: 0. o
|0. o is a function such that [([ .[) _ [.[ and (r) = 1 [(r).
Proof. We use the linear homotopy ((1 t )(. g(,)) t ([,[) .. ,). Suppose
this element is contained in G( ). Then , Y and hence g(,) = g(,), and the
rst component equals g(,). We solve for . and obtain
. =
t
1 t [([,[)
g(,).
Then we take the norm
[.[ =
t
1 t [([,[)
[g,[ _
t [([,[)
1 t [([,[)
< 1.
Hence (.. ,) D
n
Y .
In the situation of the previous lemma the map D
#
( ) is homotopic to the
restriction of the homeomorphismR
n
[0 R
n
[Y R
n
[X R
n
[0 obtainable from
(7.3.1). Another special case is obtained from a homeomorphism h of R
n
and
X R
n
. Y = h(X) R
n
. In this case h and h
-1
are Tietze extensions.
For the verication of the functor property we start with the following data:
(R
n
. X), (R
n
. Y ), (R
]
. 7) and proper maps : X Y , g: Y 7. We have the
inclusion G(g) R
n
R
]
and the proper map h: X G(g), . ( .. g .).
(7.4.4) Proposition. The diagram
R
n
[D
n
R
n
[D
n
R
]
[7
r
nm
1
1T
#
R
n
[D
n
R
n
[D
n
R
]
[7
1T
#
(( )
R
n
[D
n
R
n
[Y R
]
[0
T
#
( 1
R
n
[D
n
R
n
[X R
]
[0
R
n
[X R
n
[0 R
]
[0
r
nm
1
R
n
[0 R
n
[X R
]
[0
is homotopy commutative. Here t
nn
are the appropriate interchange maps.
Proof. For the proof we use the intermediate morphism D
#
(h). In the sequel we
skip the notation for the scaling function and the Tietze extension. If
(
is a scaling
function for and
(
a scaling function for g , then
[
1
: Y 7 |0. o . (,. :)
(
(,). [
2
: Y 7 |0. o . (,. :)
(
(:)
are scaling functions for h. We have a factorization D
2
(h) = D
2
2
(h)D
1
2
(h) where
D
1
2
(h)(.. ,. :) = (.. ,

(.). :) and D
2
2
(h)(.. ,. :) = (.. ,. :

g (.)) and
we use

h = (

.

g ) with

g = g

. The diagram
R
nn]
[{.. 0. g .]
r
nm
1
T
2
2
(h)
R
nn]
[{0. .. g .]
1T
2
(( )
R
nn]
[{.. 0. 0]
r
nm
1
R
nn]
[{0. .. 0]
commutes. The notation {(.. 0. g .)] means that we take the set of all element of
the given form where . X. We verify that the diagram
R
n
[D
n
R
n
[D
n
R
]
[7
r
nm
1
R
n
[D
n
R
n
[D
n
R
]
[7
tT
1
(( )
R
nn]
[{.. 0. g .]
r
nm
1
R
nn]
[{0. .. g .]
with = D
1
2
h D
1
h (1 D
1
g) and | : R
n
[D
n
R
n
[0 commutes up to homo-
topy. The map is, with the choice
h
=
(
, the assignment
(.. ,. :) (
(
(:) ..
(:) ,

(
(
(:) .). :).
We use the linear homotopy (
(
(:) .. s(
(:) ,

(
(
(:) .) (1s),. :).
We verify that this is a homotopy of pairs, i.e., an element { .. 0. g( .)] only occurs
as the image of an element (.. ,. :) D
n
D
n
7. Thus assume
(i) . =
(
(:) . X;
(ii) s
(:) , s

(
(
(:) .) (1 s), = 0;
(iii) : = g( .).
Since
(
(:) . X, we can replace in (ii)

by . We apply
(
to (iii) and
obtain
(
(:) =
(
(g(
(
(:) .)) _
(
(:) [.[.
hence [.[ _ 1. The equation (ii) for s = 0 says , = 0, hence , D
n
. Thus
assume s ,= 0. Then (
(
(:) .) = (
(:) s
-1
1) ,. We apply g to this
equation and use (ii):
: = g(
(
(:) .) = g((
(:) s
-1
1) ,).
Finally we apply
to this equation and obtain
(:) =
g((
(:) s
-1
1) ,) _ (
(:) s
-1
1)[,[ _
(:)[,[.
and therefore [,[ _ 1.
Finally we show that the diagram
R
n
[D
n
R
n]
[G(g)
T
#
(h)
1T
2
R
nn]
[{.. 0. 0]
R
n
[D
n
R
n
[Y R
]
[0
T
#
(( )1
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
commutes up to homotopy. In this case we use for h the scaling function [
1
. Then
D
1
h: (.. ,. :) (
(
(,) .. ,. :) and
(D
#
1) (1 D
2
g)(.. ,. :) = (
(
(,) .. ,

(
(
(,) .). : g(,)).
D
#
(h)(.. ,. :) = (
(
(,) .. ,

(
(
(,) .). : g

(
(
(,) .)).
Again we use a linear homotopy with : g((1 t ), t

(
(
(,) .)) as the third
component and have to verify that it is a homotopy of pairs. Suppose the image is
contained in {.. 0. 0]. Then
(i)
(
(,) . X;
(ii) , =

(
(
(,) .)
(i)
= (
(
(,) .) Y ;
(iii) : = g((1 t ), t(
(
(,) .))
(ii)
= g(,).
(iii) shows that (,. :) G(g). We apply
(
to (ii) and see that [.[ _ 1. The three
diagrams in this proof combine to the h-commutativity of the diagram in (7.4.4).
(7.4.5) Proposition. Suppose that h: X 1 Y is a proper homotopy. Then

D(h
0
) = D(h
1
).
Proof. Let
0
: X X 1, . (.. 0). The map is the composition of the
homeomorphisma: X X0 and the inclusion b : X0 X1. Thus Da is an
isomorphismand Db is induced by the inclusion R
n1
[X1 R
n1
[X0; hence
Db is an isomorphism, since induced by a homotopy equivalence (use (7.4.2)).
Thus D
0
is an isomorphism. The composition pr
0
is the identity; hence D(pr)
is inverse to D(
0
). Asimilar argument for
1
shows that D
0
= D
1
. We conclude
that the maps h
t
= h
t
have the same image under D.
(7.4.6) Remark. The construction of the dual morphismis a little simpler for a map
between compact subsets of Euclidean spaces. Let X R
n
be compact. Choose a
disk D such that X D. Then the dual morphism is obtained from
1
n
[0 1
n
[Y R
n
[D R
n
[Y R
nn
[G( ) R
n
[X R
n
[0
where the last morphism is as before (.. ,) (.. ,

(.)). Also the proof of
the functoriality (7.4.4) is simpler in this case.
The composition (D
#
1)(1 D
#
g) is (.. ,. :) (.. ,

(.). : g(,)).
The other composition is (.. ,. :) (.. ,. : g

(.)). Then we use the homo-
topies of pairs (.. ,

(.). : g((1t ),t

(.))) and (.. ,t

(.). : g

(.)).
Problems
1. Verify in detail that the commutativity of the diagram in (7.4.4) implies that D is a
functor.
2. Use the homotopy invariance of the duality functor and generalize (7.1.3) as follows.
Suppose X R
n
and Y R
m
are closed subsets which are properly homotopy equivalent.
Let n _ m.
(1) If R
n
X ,= 0, then R
m
Y ,= 0.
(2) Let R
n
,= X. For each choice of a base point R
m
Y has the same stable homotopy
type as
n-m
(R
n
X).
(3) If R
n
X is empty and R
m
Y is non-empty, then R
m
Y has the stable homotopy
type of S
m-n-1
.
(4) If n = mthen the complements of X and Y have the same number of path components.
3. Let X R
n
and Y R
m
be closed subsets and : X Y a proper map. Consider the
closed subspace W R
m
R R
n
of points
(,. t. .) =
_
, Y. t = 0. . = 0
((1 t )(.). t. t.). . X. t 1.
Then W is homeomorphic to the mapping cylinder 7( ) of .
4. Let X R
n
and : X R
n
R
m
the standard embedding . (.. 0). Then D is
represented by the homotopy equivalence
C(R
n
[X R
m
[0)

C(R
n
[X) . C
m
. C(R
n
[X) . S
m
.
7.5. Duality 175
(Direct proof or an application of (7.4.3).)
5. Let X R
n
be compact. Suppose [.[ _ r > 0 for . X. Then the constant function
(t ) = r is a scaling function for each : X Y . Show that the map
C(R
m
[Y ) . C
n
:
C
n
. C(R
m
[Y ) C(R
n
[X) . C
m
which is obtained from the denition in (7.4.6) is homotopic to the map C(D
#
) of the
general denition.
7.5 Duality
We have associated to a proper map between closed subsets of Euclidean spaces
a dual morphism in the stable category ST. If X R
n
then the stable homotopy
type of R
n
X or C(R
n
[X) is to be considered as a dual object of X. There is a
categorical notion of duality in tensor categories.
Let and T be pointed spaces. An n-duality between (. T) consists of an
evaluation
c: T . S
n
and a coevaluation
j: S
n
. T
such that the following holds:
(1) The composition
(1 . c)(j . 1): S
n
. . T . . S
n
is homotopic to the interchange map t.
(2) The composition
(c . 1)(1 . j): T . S
n
T . . T S
n
. T
is homotopic to (1)
n
t.
We now construct an n-duality for (T. ) = (C(R
n
[1). 1
) where 1 R
n
is a suitable space. In the general denition of an n-duality above we now replace
S
n
by C
n
. The evaluation is dened to be
c: C(R
n
[1) . C(1. 0) C(R
n
[1 1[1)
C(d)
C(R
n
[0)
where J is the difference map
J : (R
n
1. (R
n
1) 1) (R
n
. R
n
0). (.. k) . k
as a map of pairs. This denition works for arbitrary 1 R
n
.
Let 1 R
n
be compact and D R
n
a large disk which contains 1. Let V be
an open neighbourhood of 1. Consider the following diagram
R
n
[0
)
1
R
n
[D

R
n
[1
V [V R
n
[1 V [V R
n
[1
i1
V [1
} L
with the diagonal ^: . (.. .). We apply the mapping cone functor and (7.2.1),
(7.2.7). The maps i and induce h-equivalences. We obtain
j
V
: C
n
V
. C(R
n
[1).
We want to replace V by 1 in order to obtain the desired map. This can be done
if we assume that there exists a retraction r : V 1 of 1 V . Then we can
compose with r
: V
and obtain a coevaluation

j: C
n
C(1. 0) . C(R
n
[1).
We call a closed subspace 1 R
n
a Euclidean neighbourhood retract (=ENR)
if there exists a retraction r : V 1 from a suitable neighbourhood V of 1 in R
n
.
We mention here that this is a property of 1 that does not depend on the particular
embedding into a Euclidean space; see (18.4.1).
The basic duality properties of c and j are:
(7.5.1) Proposition. The maps c andjare ann-duality for the pair (1
. C(R
n
[1)).
Proof. For the proof of the rst assertion we consider the diagram
R
n
[D 1[1

P
P
P
P
P
P
P
P
P
P
P
P
R
n
[1 1[1
V [1 1[1
Z1
}1
V [V R
n
[0 V [V R
n
[1 1[1
1d
with (.. ,) = (,. .), (.. ,) = (,. . ,), and ; = (1 J)(^ 1): (.. ,)
(.. . ,). The homotopy (.. ,. t ) (t. (1 t ),. . ,) shows that the right
square is h-commutative and the homotopy (.. ,. t ) (,. . t,) shows that the
triangle is h-commutative. The axiom (1) of an n-duality now follows if we write
out the morphisms according to their denition and use the result just proved.
For the proof of the axiom (2) we start with the diagram
R
n
[1 R
n
[D
1}
/
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
R
n
[1 R
n
[1
R
n
[1 V [1
1}
;
k
k
k
k
k
k
k
k
k
k
k
k
k
k
(1i1)(1Z)
R
n
[D R
n
[1 R
n
[1 1[1 R
n
[1
d1
7.5. Duality 177

with (.. ,) = (,. .), (.. ,) = (. ,. .), and ;(.. ,) = (. ,. ,). The
homotopy (.. ,. t ) (. t, (1t )r(,). ,) shows that the bottom triangle is h-
commutative; the homotopy (.. ,. t ) (.,. (1t ).t,) shows ; . (1 );
the homotopy (.. ,. t ) (t. ,. .) shows . (1
t
). Again we write out
the morphisms according to their denition and use this result.
Given a natural duality for objects via evaluations and coevaluations one can
dene the dual of an induced map. We verify that we recover in the case of compact
ENRthe morphisms constructed in the previous section. The following three propo-
sition verify that the duality maps have the properties predicted by the categorical
duality theory.
(7.5.2) Proposition. Let X R
n
be compact and a retract of a neighbourhood V .
The following diagram is homotopy commutative
C(R
n
[Y ) . C
n
1.)
r(T(
mn
)r
C(R
n
[Y ) . C(X. 0) . C(R
n
[X)
1.C(( ).1
C
n
. C(R
n
[X) C(R
n
[Y ) . C(Y. 0) . C(R
n
[X).
t
Y
.1
Proof. We reduce the problem to maps of pairs. We use the simplied denition
(7.4.6) of the duality map. First we have the basic reduction
R
n
[Y R
n
[0 R
n
[Y R
n
[D R
n
[Y R
n
[X R
n
[Y V [X.
Then the remaining composition R
n
[Y V [X R
n
[0 R
n
[X which involves j,
C( ), c is the assignment (,. .) (, r(.). .). The other map is (,. .)
(,

(.). .). Now we observe that we can arrange that

[V = r (by possibly
passing to a smaller neighbourhood, see Problem 1).
Dual maps are adjoint with respect to evaluation and coevaluation. This is the
content of (7.5.3) and (7.5.4).
(7.5.3) Proposition. The following diagram is homotopy commutative
C(R
n
[D) . C(R
n
[Y ) . C(X. 0)
T
#
( .1
1.1.C(( )
C(R
n
[X) . C(R
n
[0) . C(X. 0)
(t.1)r
C(R
n
[D) . C(R
n
[Y ) . C(Y. 0)
i.t
C
n
. C
n
.
Proof. Consider the diagram
R
n
[D
n
R
n
[Y X[X
T
#
( 1
11(
R
n
[X R
n
[0 X[X
(d1)r
R
n
[D
n
R
n
[Y Y [Y
id
R
n
[0 R
n
[0.
The composition down-right sends the element (a. b. .) to (a. b (.)) and the
composition right-down to ((b) a .. b

((b) a)). We use the homotopies
((b)a.. b

(t (b)a(1t ).)) and then ((1t )at ((b)a.). b(.)).
(7.5.4) Proposition. Let X and Y be compact and retracts of open neighbourhoods.

Then the following diagram is homotopy commutative
C
n
. C
n
).1
r(1.))
C(X. 0) . C(R
n
[X) . C
n
C(( ).1.1
C(Y. 0) . C(R
n
[D) . C(R
n
[Y )
1.T
#
(
C(Y. 0) . C(R
n
[X) . C
n
.
Proof. We unravel the denitions and deform suitable maps between pairs. The
composition (1 . D
#
)(t . 1)(1 . j) is induced by maps
R
n
[0 R
n
[0 R
n
[D R
n
[D

R
n
[D R
n
[Y
Y [Y R
n
[X R
n
[0 R
n
[D W[Y
with (.. ,) = (r
Y
(,). .. ,

(.)). Further investigations concern . We use
the next diagram
R
n
[D W[Y

S
S
S
S
S
S
S
S
S
S
S
S
S
S
R
n
[X W[Y
V [X W[Y
k
k
k
k
k
k
k
k
k
k
k
k
k
k
Y [Y R
n
[X R
n
[0 (V W)[G( )
U[G( ).
Let
z: V W 1 R
n
. (.. ,. t ) t, (1 t ) r
A
(.).
This homotopy is constant on G( ). Hence there exists an open neighbourhood U
of G( ) such that z(U 1) W. On U we consider the homotopy of given
by (r
Y
(t, (1 t ) r
A
(.)). .. ,

(.)). For t = 0 we obtain the morphism
( r
A
(.). .. ,

(.)) which is dened on V [XR
n
[Y . Consider the composition
(C( ) . 1 . 1)(j . 1). It is induced by
V [X R
n
[0 Y [Y R
n
[X R
n
[0. (.. ,) ( r
A
(.). .. ,).
Now we use the homotopy ( r
A
(.). .. , t

(.)). For t = 1 this homotopy is
dened on V [X R
n
[Y .
7.6. Homology and Cohomology for Pointed Spaces 179
(7.5.5) Remark. Let X R
n
be a compact ENR and : X X a continuous
map. From the associated n-duality we obtain a homotopy class
z
(
: S
n
)
X
. C(R
n
[X)
r
C(R
n
[X) . X
1.(
C(R
n
[X) . X
t
S
n
.
The degree J(z
(
) = 1( ) Z is an interesting invariant of the map , the
Lefschetz xed point index. If is the identity, then 1(id) is the Euler characteristic
of X. [51] [54]
Problems
1. Let be a closed subset of a normal space X. Let r : W be a retraction of an open
neighbourhood. Choose open sets U. V such that
U

U V

V W.
Choose a continuous function : X 0. 1| such that (U) = {1] and (X V ) = {0].
Let : 0. 1| be continuous. Dene J : X 0. 1| by J(.) = (.) r(.) for
.

V and J(.) = 0 for . X V . Then J is a Tietze extension of and J[U = r[U.
2. Verify directly that the homotopy class of the coevaluation j does not depend on the choice
of the retraction r : V X.
3. The n-dualities which we have constructed can be interpreted as representative elements
for morphisms in the category ST. We obtain
c: (C(R
n
[X). n) (X
. 0) (S
0
. 0)
j: (S
0
. 0) (X
. 0) (C(R
n
[X). n).
They satisfy the relations
(1 . c)(j . 1) = id. (c . 1)(1 . j) = id
which dene dualities in tensor categories.
7.6 Homology and Cohomology for Pointed Spaces
A homology theory for pointed spaces with values in the category 1-MOD of
left modules over the commutative ring 1 consists of a family (
h
n
[ n Z) of
functors

h
n
: TOP
0
1- MOD and a family (o
(n)
[ n Z) of natural suspension
isomorphisms o = o
(n)
:

h
n

h
n1
. These data are required to satisfy the
following axioms.
(1) Homotopy invariance. For each pointed homotopy
t
the equality

h
n
(
0
) =
h
n
(
1
) holds.
(2) Exactness. For each pointed map : X Y the induced sequence
h
n
(X)
(

h
n
(Y )
(
1
+

h
n
(C( ))
is exact.
Let (X
}
[ J) be a family of well-pointed spaces with inclusions i
: X

_
}J
X
}
of the summands. The theory is called additive, if
}J
h
n
(X
}
)

h
n
_ _
}J
X
}
_
. (.
}
)
}J
(i
}
)
+
(.
}
)
is always an isomorphism.
As a variant of the axioms we require the suspension isomorphisms and the
exact sequences only for well-pointed spaces.
If we apply the exactness axiom to the identity of a point 1 we see that
h
n
(1) = 0. If X and Y are well-pointed, then the inclusion and projection give
a cobre sequence X X Y Y . This is used to verify that the additivity
isomorphism holds for a nite number of well-pointed spaces. The groups

h
n
(S
0
)
are the coefcient groups of the theory.
A natural transformation of homology theories for pointed spaces consists of
a family of natural transformations

h
n
()

k
n
() which commute with the
suspension isomorphisms.
A cohomology theory for pointed spaces consists of a family of contravari-
ant functors

h
n
: TOP
0
1- MOD and natural suspension isomorphisms o =
o
(n)
:

h
n

h
n1
such that the analogous axioms (1) and (2) hold. The theory
is called additive, if
h
n
_ _
}J
X
}
_
}J
h
n
(X
}
). . ((i
}
)
+
(.))
is always an isomorphism for well-pointed spaces X
}
.
In Chapter 10 we dene homology theories by the axioms of Eilenberg and
Steenrod. They involve functors on TOP(2). We show in Section 10.4 that they
induce a homology theory for pointed spaces as dened above.
Given a homology theory

h
+
for pointed spaces we construct fromit a homology
theory for pairs of spaces as follows. We set h
n
(X. ) =

h
n
(C(X. )). It should
be clear that the h
n
are part of a homotopy invariant functor TOP(2) 1- MOD.
We dene the boundary operator as the composition
d: h
n
(X. ) =

h
n
(C(i
))
](i)

h
n
((
))

h
n-1
(
) = h
n-1
().
The isomorphism is the given suspension isomorphism of the theory

h
+
. The
EilenbergSteenrod exactness axioms holds; it is a consequence of the assumption
that

h
+
transforms a cobre sequence into an exact sequence and of the naturality
of the suspension isomorphism. The excision isomorphism follows from (7.2.5).
We need the additional hypothesis that the covering is numerable. Remark (7.2.6)
is relevant for the passage from one set of axioms to the other.
7.7. Spectral Homology and Cohomology 181
7.7 Spectral Homology and Cohomology
In this section we report about the homotopical construction of homology and
cohomology theories. We work in the category of compactly generated spaces. A
pre-spectrum consists of a family (7(n) [ n Z) of pointed spaces and a family
(e
n
: 7(n) 7(n 1) [ n Z) of pointed maps. Since we only work with
pre-spectra in this text, we henceforth just call them spectra. A spectrum is called
an -spectrum, if the maps c
n
: 7(n) C7(n 1) which are adjoint to e
n
are
pointed homotopy equivalences.
Let 7 = (7(n). c
n
) be an C-spectrum. We dene 7
n
(X) = X. 7(n)|
0
for a
pointed space X. Since 7(n) is up to h-equivalence a double loop space, namely
7(n) . C
2
7(n2), we see that 7
n
(X) is an abelian group, and we can view 7
n
as a contravariant and homotopy invariant functor TOP
0
Z- MOD. We dene
o : 7
n
(X) 7
n1
(X) via the structure maps and adjointness as
X. 7(n)|
0
(t
n
)

X. C7(n 1)|
0
X. 7(n 1)|
0
.
We thus have the data for a cohomology theory on TOP
0
. The axioms are satised
(Puppe sequence). The theory is additive.
We nowassociate a cohomologytheorytoanarbitraryspectrum7 = (7(n). e
n
).
For k _ 0 we have morphisms
b
k
n
:
k
X. 7(n)|
0

(
k
X). 7(n)|
0
(e
n
)
k1
X. 7(n 1)|
0
.
Let 7
n-k
(X) be the colimit over this system of morphisms. The b
k
n
are compatible
with pointed maps : X Y and induce homomorphisms of the colimit groups.
In this manner we consider 7
n
as a homotopy invariant, contravariant functor
TOP
0
Z- MOD. (The b
k
n
are for k _ 2 homomorphisms between abelian
groups.) The exactness axiom again follows directly from the cobre sequence.
The suspension isomorphism is obtained via the identity
k1
X. 7(n k 1)|
0

k
(X). 7(n k 1)|
0
which gives in the colimit 7
n
(X) 7
n1
(X). If the spectrumis an C-spectrum,
we get the same theory as before, since the canonical morphisms X. 7(n)|
0
7
n
(X) are natural isomorphisms of cohomology theories. Because of the colimit
process we need the spaces 7(k) only for k _ k
0
. We use this remark in the
following examples.
7.7.1 Sphere spectrum. We dene 7(n) = S(n) and e
n
: S(n) S(n 1) the
identity. We set o
k
(X) = colim
n
n
X. S
nk
|
0
and call this group the k-th stable
cohomotopy group of X.
7.7.2 Suspension spectrum. Let Y be a pointed space. We dene a spectrum with
spaces
n
Y and e
n
: (
n
Y )
n1
Y .
7.7.3 Smash product. Let 7 = (7(n). e
n
) be a spectrum and Y a pointed space.
The spectrum Y . 7 consists of the spaces Y . 7(n) and the maps
id .e
n
: (Y . 7(n)) Y . 7(n) Y . 7(n 1).
(Note that = .1,d1. Here and in other places we have to use the associativity
of the .-product. For this purpose it is convenient to work in the category of
k-spaces.) We write in this case
7
k
(X: Y ) = colim
n
n
X. Y . 7(n k)|
0
.
The functors 7
k
(: Y ) depend covariantly on Y : A pointed map : Y
1
Y
2
induces a natural transformation of cohomology theories 7
k
(: Y
1
) 7
k
(: Y
2
).
In general, the denition of the cohomology theory 7

+
() has to be improved,
since this theory may not be additive.
We now construct homology theories. Let
1 = (1(n). e
n
: 1(n) . S
1
1(n 1) [ n Z)
be a spectrum. We use spheres as pointed spaces and take as standard model the
one-point compactication S
n
= R
n
L{o]. If V is a vector space one often writes
S
V
= V L {o] with base point o. Then we have a canonical homeomorphism
S
V
.S
W
S
V W
, the identity away from the base point. A linear isomorphism
: V W induces a pointed map S
(
: S
V
S
W
.
The homology group 1
k
(X) of a pointed space X is dened as colimit over the
maps
b : S
nk
. X.1(n)|
0
S
nk
.S
1
. X.1(n).S
1
|
0
S
nk1
. X.1(n1)|
0
.
The rst map is . S
1
and the second map is induced by id .e
n
. For n k _ 2
the morphism b is a homomorphism between abelian groups.
It should be clear from the denition that 1
k
() is a functor on TOP
0
. We need
the suspension morphisms. We rst dene suspension morphisms
o
I
: 1
k
(7) 1
k1
(S
1
. 7).
They arise from the suspensions S
1
.
S
nk
. 7 . 1(n)|
0
S
nk1
. S
1
. 7 . 1(n)|
0
.
which are compatible with the maps b above. Then we set o = (1)
k
t
+
o
I
where
the map t : S
1
. 7 7 . S
1
interchanges the factors.
7.7. Spectral Homology and Cohomology 183
(7.7.4) Lemma. o
I
is an isomorphism.
Proof. Let . 1
k
(7) be contained in the kernel of o
I
. Then there exists |
S
nk
. 7.1(n)|
0
representing . such that 1 . is null homotopic. Consider the
diagram
S
1
. S
nk
1.(
r
1
S
1
. 7 . 1(n)
r
2
S
nk
. S
1
( .1
7 . 1(n) . S
1
1.e
7 . 1(n 1)
with permutation of factors t
1
and t
2
. Since 1 . is null homotopic, the rep-
resentative (1 . e) ( . 1) of . is also null homotopic. This shows that o
I
is
injective.
In order to prove surjectivity we consider the two-fold suspension. Let .
1
k2
(S
2
. 7) have the representative g: S
nk2
S
2
. 7 . 1(n). Then
: S
nk2

S
2
. 7 . 1(n)
r
7 . 1(n) . S
2
e
2
7 . 1(n 2)
represents an element , 1
k
(7). Here e
2
is the composition of the spectral
structure maps
1(n) . S
2
= (1(n) . S
1
) . S
1
1(n 1) . S
1
1(n 2).
We show o
2
I
(,) = .. Once we have proved this we see that the second o
I
is
surjective and injective and hence the same holds for the rst o
I
.
The proof of the claim is based on the next diagram with interchange maps
t. t
t
. t
tt
.
S
2
. S
nk2
1.
r
//

S
nk2
. S
2
.1
S
2
. (S
2
. 7 . 1(n))
r
//
1
1.r
S
2
. 7 . 1(n) . S
2
1.e
2
r
/
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
S
2
. (7 . 1(n) . S
2
)
1.e
2
S
2
. 7 . 1(n 2)
=

S
2
. 7 . 1(n 2)
The maps t
t
and t
tt
are homotopic to the identity, since we are interchanging a
sphere with an even-dimensional sphere. The composition of the left verticals
represents o
2
I
(,), and the composition of the right verticals represents ..
(7.7.5) Proposition. For each pointed map : Y 7 the sequence
1
k
(Y )
(

1
k
(7)
(
1

1
k
(C( ))
is exact.
Proof. The exactness is again a simple consequence of the cobre sequence. But
since the cobre sequence is inserted into the wrong covariant part, passage
to the colimit is now essential. Suppose : 1
k
(7) is contained in the kernel
of
1
+
. Then there exists a representing map h: S
nk
7 . 1(n) such that
(
1
. 1) h is null homotopic. The next diagram compares the cobre sequences
of id: S
nk
S
nk
and . 1: Y . 1(n) 7 . 1(n).
S
nk
id
1
C(id)
](id)
S
nk
. S
1
id
S
nk
. S
1
h.1
7 . 1(n)
(( .1)
1
(
1
.1
O
O
O
O
O
O
O
O
O
O
O
C( . 1)
](( .1)
Y . 1(n) . S
1
1.e
( .1.1
7 . 1(n) . S
1
1.e
C( ) . 1(n) Y . 1(n 1)
( .1
7 . 1(n 1)
The map is the canonical homeomorphism (in the category of k-spaces) which
makes the triangle commutative. Since ( .1)
1
h is null homotopic, there exists
H such that the rst square commutes. The map is induced from (h. H) by
passing to the quotients, therefore the second square commutes. It is a simple
consequence of the earlier discussion of the cobre sequence that the third square is
h-commutative (Problem1). The composition(1.e)(h.1) is another representative
of :, and the diagram shows that (1 . e) represents an element , 1
k
(Y ) such
that
+
, = :.
A similar proof shows that the 7
k
(X: Y ) form a homology theory in the vari-
able Y .
Problems
1. Consider the cobre sequences of two maps : T and
/
:
/
T
/
. In the
diagram
C( )
)( )
T
h
/

/
T
/

/
1
C(
/
)
)(
/
)
/

/
T
/

7.8. Alexander Duality 185
assume given h and H such that the rst square commutes. The map is induced from
(h. H) by passing to the quotients. Show that the third square commutes up to homotopy
(use (4.6.2)).
2. Show that the homology theory dened by a spectrum is additive (for families of well-
pointed spaces).
3. Show that a weak pointed h-equivalence between well-pointed spaces induces an isomor-
phism in spectral homology. The use of k-spaces is therefore not essential.
4. Let 7 be the sphere spectrum (7.7.1). Then, in the notation of (7.7.3),
ST((X. n). (Y. m)) = 7
m-n
(X: Y ).
the morphism set of the category ST of Section 7.1.
7.8 Alexander Duality
Let 1 = (1
n
. e(n): 1
n
. S
1
1
n1
) be a spectrum. Let j: S
n
T . ,
c: . T S
n
be an n-duality. The compositions
. S
t
. 1
kt
|
0
B.-

T . . S
t
. T . 1
kt
|
0
)
S
n
. S
t
. T . 1
kt
|
0
are compatible with the passage to the colimit and induce a homomorphism
D
-
: 1
k
() 1
n-k
(T).
The compositions
S
tn-k
. T.1
t
|
0
.
.S
tn-k
. .T.1
t
|
0
t

.S
nk-t
. S
n
.1
t
|
0
r

.S
nk-t
. 1
t
.S
n
|
0
e

.S
tn-k
. 1
tn
|
0
with the interchange map t are compatible with the passage to the colimit if we
multiply them by (1)
nt
. They induce a homomorphism
D
-
: 1
n-k
(T) 1
k
().
(7.8.1) Theorem (Alexander duality). The morphisms D
-
and D
-
are isomor-
phisms. They satisfy D
-
D
-
= (1)
nk
id and D
-
D
-
= (1)
nk
id.
Proof. The relations of the theorem are a direct consequence of the dening prop-
erties of an n-duality. The composition c
+
( . ) j
+
(T . ) equals
t
+
n
: .S
t
. 1
kt
|
0
S
n
..S
t
. S
n
.1
kt
|
0
.S
n
.S
t
. S
n
.1
kt
|
0
.
The denition of D
-
then involves the sign (1)
n(kt)
. This morphismdiffers from
a map in the direct systemfor 1
k
() by the interchange S
n
.S
t
S
t
.S
n
, hence
by a sign (1)
nt
. Hence the sign (1)
nk
remains. The second relation is veried
similarly.
Let 11
+
() and 1h
+
() be the homology and cohomology theories on TOP(2)
constructed from the theories 1
+
() and 1
+
(). If we use the n-duality between
X
and C(R
n
[X) for a compact ENR in R
n
we obtain isomorphisms
11
n-k
(R
n
. R
n
X) 11
k
(X). 11
n-k
(X) 11
+
(R
n
. R
n
X).
This is the usual appearance of Alexander duality.
In this setting one can also work with the bi-variant theory 7
k
(X: Y ) =
7
-k
(X: Y ). Then one obtains from an n-duality an adjointness isomorphism
7
k
( . X: Y ) 7
n-k
(X: T . Y ).
A homology theory h
+
() is dened on the category ST. Here one denes
h
I
((X. n)) = h
I-n
(X). Let a morphism ST((X. n). (Y. m)) be represented by
k
: X .S
nk
Y .S
nk
. The induced morphism is dened by commutativity
of the next diagram
h
I
((X. n))
=

h
l
(( )
h
I-n
(X)

nk
h
Ik
(X . S
nk
)
((
k
)
h
I
((Y. m))
=

h
I-n
(Y )

mk
h
Ik
(Y . S
nk
).
Given a homology theory h
+
() one can dene via the complement duality
functor a sort of cohomology for spaces which admit an embedding as a closed
subset of a Euclidean space. Let X be such a space. Choose an embedding i : X
R
n
and dene h
k
(iX) = h
n-k
(C(R
n
[iX)) = h
-k
(C(R
n
[iX). n). If : X
R
n
is another embedding, we have the homeomorphism i
-1
: iX Y and we
have the duality map D(i
-1
). The set of embeddings together with the morphisms
D(i
-1
) from i to form a contractible groupoid; it is a complicated replacement
for the space X. We obtain the induced contractible groupoid of the h
k
(iX). It
is equivalent to a group which we denote h
k
(X). From the complement duality
functor we obtain a well-dened homomorphism h
k
( ): h
k
(Y ) h
k
(X) for a
proper map : X Y ; in this way h
k
() becomes a contravariant functor. We
do not discuss in what sense the h
k
(X) can be made into a cohomology theory.
This cohomology theory is the correct one for duality theory in the sense that the
Alexander duality isomorphism h
k
(X) h
-k
(C(R
n
[X). n) holds for all spaces
in question (and not only for compact ENR). A similar devise can be applied to
a given cohomology theory. One obtains a homology theory which is again the
correct one for duality theory.
7.9 Compactly Generated Spaces
Several constructions in homotopy theory lead to problems in general topology. A
typical problem arises from the fact that a product of quotient maps is in general
7.9. Compactly Generated Spaces 187
no longer a quotient map. We met this problem already in the discussion of CW-
complexes. In this auxiliary section we report about some devices to deal with such
problems. The idea is to construct a category with better formal properties. One has
to pay a price and change some of the standard notions, e.g., redene topological
products.
A compact Hausdorff space will be called a ch-space. For the purpose of the
following investigations we also call a ch-space a test space and a continuous map
: C X of a test space C a test map. A space X is called weakly hausdorff or
wh-space, if the image of each test map is closed.
(7.9.1) Proposition. A Hausdorff space is a wh-space. A wh-space is a T
1
-space.
A space X is a wh-space if and only if each test map : 1 X is proper. If X is
a wh-space, then the image of each test map is a Hausdorff space. A subspace of a
wh-space is a wh-space. Products of wh-spaces are wh-spaces.
A subset of a topological space (X. T ) is said to be k-closed (k-open), if
for each test map : 1 X the pre-image
-1
() is closed (open) in 1. The
k-open sets in (X. T ) form a topology kT on X. A closed (open) subset is also
k-closed (k-open). Therefore kT is ner than T and the identity | = |
A
: kX X
is continuous. We set kX = k(X) = (X. kT ). Let : 1 X be a test map.
The same set map : 1 kX is then also continuous. For if U kX is open,
then U X is k-open, hence
-1
(U) 1 is open. Therefore |
A
induces for each
ch-space 1 a bijection.
TOP(1. kX)

TOP(1. X). |
A
.
Hence X and kX have the same k-open sets, i.e., k(kX) = kX. Atopological space
X is called k-space, if the k-closed sets are closed, i.e., if X = kX. Because of
k(kX) = kX the space kX is always a k-space. A k-space is also called compactly
generated. We let k-TOP be the full subcategory of TOP with objects the k-spaces.
A whk-space is a space which is a wh-space and a k-space.
The next proposition explains the denition of a k-space. We call a topology S
on X ch-denable, if there exists a family (
}
: 1
}
X [ J) of test maps
such that: X is S-closed =for each J the pre-image
-1
}
() is closed in
1
}
. We can rephrase this condition: The canonical map (
}
):

}
1
}
(X. S)
is a quotient map. A ch-denable topology is ner than T . We dene a partial
ordering on the set of ch-denable topologies by S
1
_ S
2
=S
1
S
2
.
(7.9.2) Proposition. The topology kT is the maximal ch-denable topology with
respect to the partial ordering.
Proof. By Zorns Lemma there exists a maximal ch-denable topology S. If this
topology is different from kT , then there exists an S-open set U, which is not
k-open. Hence there exists a test map t : 1 X such that t
-1
(U) is not open. If
we adjoin this test map to the dening family of S, we see that S is not maximal.
(7.9.3) Corollary. The k-spaces are the spaces which are quotients of a topological
sum of ch-spaces.
(7.9.4) Proposition. The following are equivalent:
(1) X is a k-space.
(2) Aset map : X Y is continuous if and only if for each test map t : 1 X
the composition t is continuous.
Proof. (1) = (2). Let U Y be open. In order to see that
-1
(U) is open it
sufces to show that this set is k-open, since X is a k-space. Let t : 1 X be a
test map and t continuous. Then k
-1
(
-1
(U)) is open, and this shows what we
want.
(2) =(1). We show that the identity X kX is continuous. This holds by (2)
and because X and kX have the same test maps.
(7.9.5) Proposition. Let : X Y be continuous. Then the same set map
k : kX kY is continuous.
The assignments X kX, k yield a functor k; moreover, we have the
inclusion functor i ,
k : TOP k- TOP. i : k- TOP TOP .
(7.9.6) Proposition. The functor k is right adjoint to the functor i .
Proof. A natural bijection is k- TOP(Y. kX) TOP(iY. X), | . This
map is certainly injective. If Y is a k-space and : Y X continuous, then
k : Y = kY kX is continuous; this is used to show surjectivity.
(7.9.7) Proposition. Let X be a wh-space. Then X is k-closed if and only if
for each ch-space 1 X the set 1 is closed in 1. In particular a wh-space
X is a k-space if and only if: X is closed =for each ch-space 1 X the
intersection 1 is closed in 1.
Proof. Let be k-closed. The inclusion 1 X of a ch-space is a test map. Hence
1 is closed in 1.
Conversely, suppose that satises the stated condition and let : 1 X be
a test map. Since X is a wh-space, (1) is a ch-space and therefore (1) is
closed in (1). Then
-1
() =
-1
((1) ) is closed in 1 =
-1
(1).
This shows: is k-closed.
Thus we see that wh-spaces have an internal characterization of their k-closed
sets. For wh-spaces therefore k(X) can be dened from internal properties of X.
If X is a wh-space, so is kX.
(7.9.8) Theorem. X is a k-space under one of the following conditions:
(1) X is metrizable.
(2) Each point of X has a countable neighbourhood basis.
(3) Each point of X has a neighbourhood which is a ch-space.
(4) For Q X and .

Qthere exists a ch-subspace 1 X such . is contained
in the closure of Q 1 in 1.
(5) For each Q X the following holds: Q 1 open (closed) in 1 for each
test space 1 X implies Q open (closed) in X.
Proof. (1) is a special case of (2).
(2) Let Q X and suppose that
-1
(Q) is closed for each test map : C
X. We have to show that Q is closed. Thus let a

Q and let (U
n
[ n N) be
a neighbourhood basis of a. For each n choose a
n
Q U
1
U
n
. Then
the sequence (a
n
) converges to a. The subspace 1 = {0. 1. 2
-1
. 3
-1
. . . . ] of R
is compact. The map : 1 X, (0) = a, (n
-1
) = a
n
is continuous, and
n
-1

-1
(Q). By assumption,
-1
(Q) is closed in 1, hence 0
-1
(Q), and
therefore a = (0) Q.
(3) = (4). Let Q X and suppose a

Q. We choose a ch-neighbourhood
1 of a and show that a is contained in the closure of Q 1 in 1. Thus let U be
a neighbourhood of a in 1. Then there exists a neighbourhood U
t
of a in X such
that U
t
1 U. Since U
t
1 is a neighbourhood of a in X and a

Q, we
conclude
U (Q 1) (U
t
1) (Q 1) = (U
t
1) Q = 0.
Hence a is contained in the closure of Q 1 in 1.
(4) =(5). Suppose Q1 is closed in 1 for every test subspace 1 X. Let
a

Q. By (4), there exists a test subspace 1
0
of X, such that a is contained in
the closure of Q1
0
in 1
0
. By the assumption (5), Q1
0
is closed in 1
0
; and
hence a Q 1
0
Q.
(5) Let
-1
(Q) be closed in 1 for each test map : 1 X. Then, in
particular, for each test subspace 1 X the set Q 1 is closed in 1. The
assumption (5) then says that Q is closed in X. This shows that X is a k-space.
(7.9.9) Theorem. Let : Y X be a quotient map and Y a k-space. Then X is

a k-space.
Proof. Let T X be k-closed. We have to show that T is closed, hence, since
is a quotient map, that
-1
(T) is closed in Y . Let g: D Y be a test map. Then
g
-1
(
-1
(T)) = (g)
-1
(T) is closed in D, because T is k-closed. Since Y is a
k-space,
-1
(T) is closed in Y .
(7.9.10) Proposition. Aclosed (open) subspace of a k-space is a k-space. The same
holds for whk-spaces.
Proof. Let be closed and T a subset such that
-1
(T) is closed in C for
test maps : C . We have to show: T is closed in or, equivalently, in X.
If g: D X is a test map, then g
-1
() is closed in D and hence compact,
since D is compact. The restriction of g yields a continuous map h: g
-1
() .
The set h
-1
(T) = g
-1
(T) is closed in g
-1
() and therefore in D, and this shows
that T is closed in X.
Let U be open in the k-space X. We write X as quotient q : 7 X of a sum
7 of ch-spaces (see 7.9.3). Then q : q
-1
(U) U is a quotient map and q
-1
(U)
as the topological sum of locally compact Hausdorff spaces is a k-space. Therefore
the quotient U is a k-space.
The second assertion follows, if we take (7.9.1) into account.
In general, a subspace of a k-space is not a k-space (see (7.9.23)). Let X be a
k-space and i : X the inclusion. Then the map k(i ): k() X = k(X) is
continuous. The next proposition shows that k(i ) has in the category k- TOP the
formal property of a subspace.
(7.9.11) Proposition. A map h: 7 k() from a k-space 7 into k() is contin-
uous if and only if k(i ) h is continuous.
Proof. If h is continuous then also is k(i )h. Conversely, let k(i )h be continuous.
We have k(i ) = i |
. Since i is the inclusion of a subspace, |
h is continuous;
(7.9.6) now shows that h is continuous.
(7.9.12) Theorem. The product in TOP of a k-space X with a locally compact
Hausdorff space Y is a k-space.
Proof. By (7.9.8), a locally compact Hausdorff space is a k-space. We write X as
quotient of 7, where 7 is a sum of ch-spaces, see (7.9.3). Since the product of
a quotient map with a locally compact space is again a quotient map, we see that
X Y is a quotient of the locally compact Hausdorff space, hence k-space, 7Y ,
and therefore a k-space by (7.9.9).
A product of k-spaces is not always a k-space (see (7.9.23)). Therefore one is
looking for a categorical product in the category k- TOP. Let (X
}
[ J) be a
family of k-spaces and

}
X
}
its product in the category TOP, i.e., the ordinary
topological product. We have a continuous map
}
= k(pr
}
): k
_
}
X
}
_
k(X
}
) = X
}
.
The following theoremis a special case of the fact that a right adjoint functor respects
limits.
(7.9.13) Theorem.
_
}
: k
_
}
X
}
_
X
}
[ J
_
is a product of (X
}
[ J)
in the category k- TOP.
Proof. We use (7.9.6) and the universal property of the topological product and
obtain, in short-hand notation, for a k-space T the canonical bijection
k- TOP
_
T. k
_
X
}
__
= TOP
_
T.
X
}
_
TOP(T. X
}
) =
k- TOP(T. X
}
).
and this is the claim.
In the case of two factors, we use the notation X
k
Y for the product in k- TOP
just dened. The next result shows that the wh-spaces are the formally hausdorff
spaces in the category k- TOP.
(7.9.14) Proposition. A k-space X is a wh-space if and only if the diagonal D
A
of
the product X
k
X is closed.
Proof. Let X be a wh-space. In order to verify that D
A
is closed, we have to
show that for each test map : 1 X
k
X the pre-image
-1
(D
A
) is closed.
Let
}
: 1 X be the -th component of . Then 1
}
=
}
(1) is a ch-space,
since X is a wh-space. Hence 1 = 1
1
L 1
2
X is a ch-space. The relation
-1
D
A
=
-1
((1 1) D
A
) shows that this set is closed.
Let D
A
be closed in X
k
X and : 1 X a test map. We have to show that
(1) X is closed. Let g: 1 X be another test map. Since X is a k-space,
we have to show that g
-1
(1) 1 is closed. We use the relation
g
-1
(1) = pr
2
(( g)
-1
D
A
).
Since D
A
is closed, the pre-image under g is closed and therefore also its image
under pr
2
as a compact set in a Hausdorff space.
Recall the mapping space J(X. Y ) with compact-open topology.
(7.9.15) Theorem. Let X and Y be k-spaces, and let : X
k
Y 7 be con-
tinuous. The adjoint map
.
: X kJ(Y. 7), which exists as a set map, is
continuous.
Proof. The map
.
: X kJ(Y. 7) is continuous, if for each test map t : C X
the composition
.
t is continuous. We use
.
t = ( (t id
Y
))
.
. Therefore it
sufces to assume that X is a ch-space. But then, by (7.9.12), X
k
Y = XY and
therefore
.
: X J(Y. 7) is continuous and hence also
.
: X kJ(Y. 7),
by (7.9.4).
(7.9.16) Theorem. Let Y be a k-space. Then the evaluation
e
Y,7
: kJ(Y. 7)
k
Y 7. (. ,) (,)
is continuous.
Proof. Let t : C kJ(Y. 7)
k
Y be a test map. We have to show the continuity
of e
Y,7
t . Let t
.
1
: C J(Y. 7) and t
2
: C Y be the continuous components
of t . We show rst: The adjoint t
1
: C Y 7 of t
.
1
is continuous. By (2.4.3),
this continuity is equivalent to the continuity of the second adjoint map t
1
: Y
J(C. 7). In order to show its continuity, we compose with a test map s : D Y .
But t
1
s = J(s. 7) t
.
1
is continuous. Moreover we have e
Y,7
t = t
1
(id. t
2
),
and the right-hand side is continuous.
A combination of (7.9.15) and (7.9.16) now yields the universal property of the
evaluation e
Y,7
for k-spaces:
(7.9.17) Proposition. Let X and Y be k-spaces. The assignments
.
and
g e
Y,7
(g
k
id
Y
) = g
~
are inverse bijections
TOP(X
k
Y. 7) TOP(X. kJ(Y. 7))
between these sets.
(7.9.18) Theorem. Let X, Y and 7 be k-spaces. Since e
Y,7
is continuous, we have
an induced set map
z: kJ(X. kJ(Y. 7)) kJ(X
k
Y. 7). e
Y,7
(
k
id
Y
) =
~
.
The map z is a homeomorphism.
Proof. We use the commutative diagram
kJ(X. kJ(Y. 7))
k
X
k
Y
e
1
id
2id id
kJ(Y. 7)
k
Y
e
2
kJ(X
k
Y. 7)
k
X
k
Y
e
3
7
with e
1
= e
A,kT(Y,7)
, e
2
= e
Y,7
, and e
3
= e
A
k
Y,7
. Since e
1
id and e
2
are
continuous, the universal property of e
3
shows that z is continuous; namely, using
the notation from (7.9.17), we have e
2
(e
1
id) = z
~
. The universal property of
e
1
provides us with a unique continuous map
j: kJ(X
k
Y. 7) kJ(X. kJ(Y. 7)).
.
.
such that e
1
(jid(X)) = e
.
3
, where e
.
3
: kJ(X
k
Y. 7)
k
X kJ(Y. 7) is
the adjoint of e
3
with respect to the variable Y . One checks that z and jare inverse
to each other, hence homeomorphisms.
(7.9.19) Theorem. Let X and Y be k-spaces, and : X X
t
and g: Y Y
t
be quotient maps. Then g: X
k
Y X
t
k
Y
t
is a quotient map.
Proof. It sufces to treat the case g = id, since a composition of quotient maps is
a quotient map. Using (7.9.18), the proof is now analogous to (2.4.6).
(7.9.20) Proposition. Let : X Y be a quotient map and X a whk-space. Then
Y is a whk-space if and only if 1 = {(.
1
. .
)
[ (.
1
) = (.
2
)] is closed in X
k
X.
Proof. The set 1 is the pre-image of D
Y
under . Since
k
is a quotient
map (7.9.19), D
Y
is closed if and only if 1 is closed. Now apply (7.9.9) and
(7.9.14).
(7.9.21) Proposition. Let Y and 7 be k-spaces and assume that 7 is a wh-space.
Then the mapping space kJ(Y. 7) is a wh-space. In particular, if Y and 7 are
whk-spaces, then kJ(Y. 7) is a whk-space.
Proof. Let
.
: 1 kJ(Y. 7) be a test map. We have to show that it has
a closed image hence is k-closed. For this purpose let g
.
: 1 kJ(Y. 7) be
another test map. It remains to show that the pre-image M of
.
(1) under g
.
is closed. We use the adjoint maps : 1 Y 7 and g: 1 Y 7. For
, Y let i
,
: 1 1 (1 Y )
k
(1 Y ), (k. l) (k. ,. l. ,). Then
M = pr
2
_ _
,Y
(( g)i
,
)
-1
D
7
_
. Since 7 is a wh-space and therefore the
diagonal D
7
is closed by (7.9.14), we see that M is closed.
We now consider pointed spaces. Let (X
}
[ J) be a family of pointed
k-spaces. Let

k
}
X
}
be its product in k- TOP. Let W
J
X
}
be the subset of the
product of those points for which at least one component equals the base point.
The smash product
_
k
}
X
}
is the quotient space
_
k
}
X
}
_
,W
J
X
}
. In the case that
J = {1. . . . . n] we denote this space by X
1
.
k
.
k
X
n
. Afamily of pointed maps
}
: X
}
Y
}
induces a pointed map
_
k
}
:
_
k
}
X
}

_
k
}
Y
}
.
Let X and Y be pointed k-spaces. Let J
0
(X. Y ) J(X. Y ) be the subspace of
pointed maps. We compose a pointed map : X .
k
Y 7 with the projections
: X
k
Y X .
k
Y . The adjoint ()
.
: X kJ(Y. 7) is continuous and has
an image contained in kJ
0
(Y. 7). We obtain a continuous map X kJ
0
(Y. 7)
which will be denoted by
.
.
The evaluation e
Y,7
induces e
0
Y,7
which makes the following diagram commu-
tative:
kJ
0
(Y. 7)
k
X
k(i)id
kJ(Y. 7)
k
X
e
Y,.
kJ
0
(Y. 7) .
k
X
e
0
Y,.
Y .
i is the inclusion and the quotient map. The continuity of k(i ) and e
Y,7
implies
the continuity of the pointed evaluation e
0
Y,7
. In analogy to (7.9.18) one proves:
(7.9.22) Theorem. Let X, Y and 7 be pointed k-spaces. The assignment
j
0
: kJ
0
(X .
k
Y. 7) kJ
0
(X. kJ
0
(Y. 7).
.
is a homeomorphism.
(7.9.23) Example. Let R,Z be obtained from R by identifying the subset Z to a
point (so this is not the factor group!). We denote by : R R,Z the quotient
map.
(1) The product id: R Q R,Z Qof quotient maps is not a quotient
map.
(2) The product R,Z Q is not a k-space, but the factors are k-spaces (see
(7.9.4)).
(3) The product R,ZR is a k-space (see (7.9.9) and (7.9.12)), but the subspace
R,Z Qis not a k-space by (2).
If 1 R,Z is compact, then there exists l N such that 1 l. l|.
Let (r
n
[ n N) be a strictly decreasing sequence of rational numbers with
limit
_
2. The set J =
_
m
1
2n
.
i
n
n
_
[ n. m N
_
R Q is saturated with
respect to id and closed in R Q.
The set G = (id)(J) is not closed in R,ZQ. Note that : = ((0). 0) , G;
but we show that :

G. Let U be a neighbourhood of :. Then there exists a
neighbourhood V of (0) in R,Z and c > 0 such that V ( | c. c Q) U.
Choose m Nsuch that m
-1
_
2 < 2
-1
c. The set
-1
(V ) is then a neighbourhood
of m in R, since m
-1
(0)
-1
(V ). Hence there exists > 0 such
that |m . m
-1
(V ). Now choose n N such that
1
2n
< and
r
n

_
2 < m
t
2
. Then ( id)(m
1
2n
.
i
n
n
) V ( | c. c Q) U holds,
because m
1
2n
|m. m
-1
(V ) and0 <
i
n
n
=
_
2
n

i
n
-
_
2
n
<
t
2
t
2
= c.
We see that U G ,= 0. This nishes the proof that :

G.
We nowsee that id is not a quotient map, since there exists a saturated closed
set J with non-closed image G.
The space R,Z Q is not a k-space. Let s : 1 R,Z Q be an arbitrary
test map. We show that s
-1
(G) is closed in 1 although G is not closed (this could
not occur in a k-space). The two projections pr
i
s(1) are compact and Hausdorff.
Hence there exists l N such that pr
1
s(1) l. l|. The inclusion
s(1) pr
1
s(1) pr
2
s(1) l l| pr
2
s(1)
then shows that we have s
-1
(G) = s
-1
(G l. l| pr
2
s(1)). But the set
G l. l| pr
2
s(1) is nite: By construction, J is a closed discrete subspace
of R Q; moreover, J l. l| pr
2
s(1) is nite as a closed discrete subspace
of the compact space l. l| pr
2
s(1); therefore also
( id)(J l. l| pr
2
s(1)) = G l. l| pr
2
s(1)
is nite. A nite set in a Hausdorff space is closed, and therefore s
-1
(G) as pre-
image of a closed set is closed itself.
(7.9.24) Example. It is stated already in [155, p. 336] that (Q . Q) . N
0
and
Q.(Q.N
0
) are not homeomorphic. In [128, p. 26 ] it is proved that the canonical
continuous bijection from the rst to the second is not a homeomorphism.
Problems
1. A space is a k-space if and only if it is a quotient of a locally compact Hausdorff space.
2. Let X
1
X
2
, let X
be a whk-space and let X
X
1
be closed. Then
X =
_
, with colimit topology, is a whk-space. If the X

i
are k-spaces, then X is a
k-space, being a quotient of the k-space
i
X
i
. If the X
i
are wh-spaces, hence T
1
-spaces,
then each test map : 1 X has an image which is contained in some X
i
and therefore
closed. If each inclusion is X
i
X
i1
closed, the image is also closed in X and therefore
X is a wh-space.
3. Let X and Y be k-spaces. Passage to adjoint maps induces bijections of homotopy sets
X
k
Y. 7| X. kJ(Y. 7)| and X . Y. 7|
0
X. kJ
0
(Y. 7)|
0
.
4. Let (X
[ J) be a family of k-spaces. Then the topological sum

J
X
is a
k-space. The product in k-TOP is compatible with sums.
5. Let a pushout of topological spaces with closed : X be given.
T
J
X
1
Y
Let X and T be whk-spaces. Then Y is a whk-space.
Chapter 8
Cell Complexes
The success of algebraic topology is largely due to the fact that one can describe
spaces of interest by discrete (or even nite) combinatorial data. Purely combina-
torial objects are simplicial complexes. Given such a complex, one denes from
its data by simple linear algebra the homology groups. It is then a remarkable
fact that these groups are independent of the combinatorial description and even
homotopy invariant. Simplicial complexes are a very rigid structure. A weaken-
ing of this structure is given by the cell complexes (CW-complexes in the sense of
J. H. C. Whitehead). They are more exible and better adapted to homotopy theory.
An n-cell in a space is a subset which is homeomorphic to the standard n-cell
1
n
= {. R
n
[ [.[ < 1]. A cell complex is a decomposition of a space into
cells. In order that one obtains something interesting, one has to add conditions
about the closure of the cells, and one has to relate the topology of the space to the
topology of the closed cells.
A nite cell complex is easily dened: a Hausdorff space X which is the union
of a nite number of cells. It e is an n-cell of this decomposition, then it is required
that there exists a continuous map : D
n
X which induces a homeomorphism
1
n
e and sends S
n-1
into the union of the cells up to dimension n 1.
From these data one obtains already an interesting invariant of X, the so-called
Euler characteristic. Let n(i ) denote the number of i -cells. Dene the combinatorial
Euler characteristic to be the alternating sum (X) =
i_0
(1)
i
n(i ). It is a non-
trivial fact that h-equivalent nite complexes have the same Euler characteristic.
The origin of the notion is the famous result of Leonhard Euler (~ 1752) that for a
sphere S
2
each polyhedral decomposition yields the value n(0) n(1) n(2) = 2.
In this chapter we present some point-set topology and elementary homotopy
theory of cell complexes. Then we demonstrate the use of cell complexes in the
construction of spaces with specic properties. In particular we construct so-called
EilenbergMac Lane spaces 1(. n). They have a single non-vanishing homotopy
group
n
(1(. n)) (here can be an arbitrary abelian group). EilenbergMac
Lane spaces can be used as building blocks for general homotopy types (Postnikov
systems). They also yield a homotopical denition of cohomology (and homology)
groups: The homotopy set X. 1(. n)| carries a natural structure of an abelian
group and is known to be a version of a cohomology group denoted H
n
(X: ).
8.1. Simplicial Complexes 197
8.1 Simplicial Complexes
Simplicial complexes are the objects of combinatorial topology.
A simplicial complex 1 = (1. S) consists of a set 1 of vertices and a set S
of nite non-empty subsets of 1. A set s S with q 1 elements is called a
q-simplex of 1. We require the following axioms:
(1) {e] S for each e 1.
(2) If t S and s t is non-empty, then s S.
If s S is a q-simplex, then q is called the dimension of s. If t s, then t is
a face of s. A 1-simplex of 1 is also called an edge of 1. The 0-simplices of
1 correspond to the elements of 1; a 0-simplex is called a vertex. A simplex is
determined by its 0-faces.
A simplicial complex is n-dimensional, if it contains at least one n-simplex but
no (n 1)-simplices. A subcomplex 1 of 1 consists of a set of simplices of 1
which contains with s also the faces of s. A 1-dimensional complex is a graph.
A complex 1 = (1. S) is nite if 1 is nite and locally nite if each vertex
is contained in a nite number of simplices. The n-skeleton 1
n
= (1. S
n
) of
1 = (1. S) is the subcomplex with S
n
= {s S [ dims _ n].
(8.1.1) Example. Let U = (U
}
[ J) be a covering of a set X by non-
empty sets U
}
. For a nite non-empty set 1 J let U
T
=
_
}T
U
}
and let
1(J) = {1 J [ U
T
,= 0]. Then (J. 1(J)) is a simplicial complex, called the
nerve N(U) of the covering U.
(8.1.2) Example. Let 1 be a set with a partial ordering _. The simplicial complex
(1. S
1
) associated to a partially ordered set has as simplices the totally ordered
nite subsets of 1.
(8.1.3) Example. Let 1 = (1. S) be a simplicial complex. Dene a partial order
on S by s _ t = s t . The simplicial complex 1
t
associated to this ordered set
is called the barycentric subdivision of 1.
Let 1 = (1. S) be a simplicial complex. We denote by [1[ the set of functions
: 1 1 such that
(1) {e 1 [ (e) > 0] is a simplex of 1.
(2)

eT
(e) = 1.
We regard [1[ as a subset of the product 1
T
. Let [1[
]
denote this set with the
subspace topology of the product topology. We have a metric J on [1[ dened by
J(. ) =
_
eT
((e) (e))
2
_1
2
.
We denote [1[ with this metric topology by [1[
n
. Each vertex e 1 gives us a con-
tinuous map e : [1[
n
1, (e). Therefore the identity [1[
n
[1[
]
is
198 Chapter 8. Cell Complexes
continuous. We leave it as an exercise to show that this map is actually a homeo-
morphism. The numbers ((e) [ e 1) are the barycentric coordinates of .
We dene a further topology on [1[. For s S let ^(s) be the standard
simplex {(t
e
) [1[ [ t
e
= 0 for e s]. Then [1[ is the union of the ^(s), and
we write [1[
c
for [1[ with the quotient topology dened by the canonical map
xS
^(s) [1[. The identity [1[
c
[1[
]
is continuous but not, in general, a
homeomorphism. The next proposition will be proved in the more general context
of simplicial diagrams.
(8.1.4) Proposition. [1[
c
[1[
]
is a homotopy equivalence.
In the sequel we write [1[ = [1[
c
and call this space the geometric realization
of 1. We dene [s[ [1[ as [s[ = { [1[ [ (e) ,= 0 = e s] and
call this set a closed simplex of [1[. For each simplex s of 1 the open simplex
(s) [1[ is the subspace (s) = { [1[ [ (e) ,= 0 =e s]. The complement
[s[ \ (s) = d[s[ is the combinatorial boundary of [s[; it is the geometric realization
of the subcomplex which consists of the proper faces of s. The set [1[ is the disjoint
union of the (s). s S.
Let [1[
n
be the union of the ^(s) with dims _ n.
(8.1.5) Proposition. The space [1[ is the colimit of the [1[
n
. The equality [1
n
[ =
[1[
n
holds. The canonical diagram
x,dimx=n
d^(s)

[1[
n-1
x,dimx=n
^(s)

[1[
n
is a pushout.
A homeomorphism t : [1[ X is called a triangulation of X. The triangu-
lation of surfaces was proved by Rad [161], the triangulation of 3-dimensional
manifolds by Moise (see [141] for references and proofs; [197, 7.5.1]). Differen-
tiable manifolds can be triangulated, and the triangulation can be chosen in such a
way that it is on each simplex a smooth embedding ([193]; [143]).
Since [1[
n
is separated and id: [1[ [1[
n
continuous, [1[ is separated. For
nite 1 the identity [1[ [1[
n
is a homeomorphism.
For each vertex e 1 the set St(e) = { [1[ [ (e) ,= 0] is called the star
of e. Since (e) is continuous, the set St(e) is open in [1[
d
and therefore
also in [1[. If we identify e with the function (e) = 1. (e
t
) = 0 for e ,= e
t
, then
St(e) is an open neighbourhood of e.
Points e
0
. . . . . e
k
of R
n
are afnely independent, if the relations z
i
e
i
= 0
and z
i
= 0 imply that each z
i
= 0. If e
0
. . . . . e
k
are afnely independent, then
the simplex

k
i=0
z
i
e
i
[ z
i
_ 0. z
i
= 1
_
spanned by e
0
. . . . . e
k
is the convex
hull of this set and homeomorphic to the k-dimensional standard simplex.
8.2. Whitehead Complexes 199
Let 1 = (1. S) be a simplicial complex and (.
e
[ e 1) a family of points
in R
n
. Consider the continuous map
: [1[ R
n
.
eT
(e).
e
.
If is an embedding, we call the image of a simplicial polyhedron in R
n
of type
1, and ([1[) is a realization of 1 as a polyhedron in R
n
.
Standard tools for the application of simplicial complexes in algebraic topology
are subdivision and simplicial approximation [67, p. 124].
Problems
1. id: [1[
m
[1[
)
is a homeomorphism.
2. Let 1 = (N
0
. S) be the simplicial complex where S consists of all nite subsets of N
0
.
The canonical map [1[
c
[1[
)
is not a homeomorphism.
3. Let 1be a subcomplex of 1. We can identify [1[ with a subset of [1[, and [1[ carries then
the subspace topology of [1[. If (1
[ J) is a family of subcomplex of 1, then

_
1
and
_
1
are subcomplexes and the relations

_
[1
[ = [
_
1
[ and
_
[1
[ = [
_
1
[
hold.
4. Let 1 be a simplicial complex. Then the following assertions are equivalent: (1) 1 is
locally nite. (2) [1[ is locally compact. (3) The identity [1[ [1[
o
is a homeomorphism.
(4) [1[ is metrizable. (5) Each point of [1[ has a countable neighbourhood basis. (See [44,
p. 65].)
5. Let 1 be a countable, locally nite simplicial complex of dimension at most n. Then 1
has a realization as a polyhedron in R
2n1
. (See [44, p. 66].)
8.2 Whitehead Complexes
We use the standard subsets of Euclidean spaces S
n-1
. D
n
. 1
n
= D
n
\ S
n-1
,
(n _ 1). We set S
-1
= 0 and let D
0
be a point, hence 1
0
= D
0
. A k-dimen-
sional cell (a k-cell) in a space X is a subset e which is, in its subspace topology,
homeomorphic to 1
k
. A point is always a 0-cell.
A Whitehead complex is a space X together with a decomposition into cells
(e
2
[ z ) such that:
(W1) X is a Hausdorff space.
(W2) For each n-cell e
2
there exists a characteristic map
2
: D
n
= D
n
2
X
which induces a homeomorphism 1
n
e
2
and sends S
n-1
into the union
X
n-1
of the cells up to dimension n 1.
(W3) The closure e
2
of each cell e
2
intersects only a nite number of cells.
(W4) X carries the colimit topology with respect to the family ( e
2
[ z ).
A subset of a Whitehead complex is a subcomplex if it is a union of cells and
the closure of eachcell inis containedin. We will see that a subcomplextogether
with its cells is itself a Whitehead complex. From the denition of a subcomplex
we see that intersections and unions of subcomplexes are again subcomplexes.
Therefore there exists a smallest subcomplex X(1) which contains a given set 1.
The decomposition of a Hausdorff space into its points always satises (W1)
(W3). We see that (W4) is an important condition. Condition (W3) is also called
(C), for closure nite. Condition (W4) is called (W), for weak topology. This is the
origin for the name CW-complex. In the next section we consider these complexes
from a different view-point and introduce the notion of a CW-complex.
(8.2.1) Lemma. Let : D
n
X be a continuous map into a Hausdorff space.
Let e = (1
n
). Then (D
n
) = e. In particular e is compact.
Proof. (D
n
) is a compact subset of a Hausdorff space and therefore closed. This
yields e = (1
n
) (D
n
) = (D
n
) = (
1
n
) (1
n
) = e.
(8.2.2) Example. Suppose X has a cell decomposition into a nite number of cells
such that properties (W1) and (W2) hold. Then X is a nite union of closures e of
cells and therefore compact by (8.2.1). Properties (W3) and (W4) are satised and
X is a Whitehead complex.
(8.2.3) Examples. The sphere S
n
has the structure of a Whitehead complex with
a single 0-cell and a single n-cell. The map
: D
n
S
n
. . (2
_
1 [.[
2
.. 2[.[
2
1)
sends S
n-1
to the 0-cell e
n1
= (0. . . . . 0. 1) and induces a homeomorphism of
1
n
with S
n
{e
n1
], hence is a characteristic map for the n-cell.
Fromthis cell decomposition we obtain a cell decomposition of D
n1
by adding
another (n 1)-cell 1
n1
with characteristic map the identity.
Another cell-composition of S
n
has two -cells for each {0. . . . . n] and
is obtained inductively from D
n
= {(.
i
) S
n
[ .
n1
_ 0] with intersection
S
n-1
= S
n-1
0. A characteristic map is D
n
D
n
, . (..
_
1 [.[
2
).
(8.2.4) Proposition. Let X be a Whitehead complex.
(1) A compact set 1 in X meets only a nite number of cells.
(2) A subcomplex which consists of a nite number of cells is compact and closed
in X.
(3) X(e) = X( e) is for each cell e a nite subcomplex.
(4) A compact subset of a Whitehead complex is contained in a nite subcomplex.
(5) X carries the colimit topology with respect to the nite subcomplexes.
(6) A subcomplex is closed in X.
Proof. (1) Let 1 be the set of cells which meet 1. For each e 1 we choose a
point .
e
1 e and set 7 = {.
e
[ e 1]. Let Y 7 be any subset. For each
cell of X the closure

is contained in the union of a nite number of cells. Thus
8.2. Whitehead Complexes 201
Y

is a nite set, hence closed in

since

is a Hausdorff space. The condition
(W4) now says that Y is closed in X and hence in 7. This tells us that 7 carries
the discrete topology and is closed in X. A discrete closed set in a compact space
is nite.
(2) Let = e
1
L Le
i
be a nite union of cells e
}
. Then

= e
1
L L e
i
,
by denition of a subcomplex. By (8.2.1), =

is compact and closed.
(3) Induction over dim(e). If e is a 0-cell, then e is a point and closed, hence a
subcomplex and X(e) = X( e) = e. Suppose X( ) is nite for each cell with
dim( ) < n. Let e be an n-cell with characteristic map .
The set (S
n-1
) is contained in the union of cells of dimension at most n 1,
hence is contained in e e.
Then e e = (S
n-1
) is compact, hence contained in a nite number of
cells e
1
. . . . . e
k
, by (1), which are contained in X
n-1
, by (W2). By induction
hypothesis, the set C = e L X(e
1
) L L X(e
k
) is a nite subcomplex which
contains e and hence X(e). Therefore X(e) is nite. Since X(e) is closed, by (2),
we have e X(e) and X( e) X(e).
(4) This is a consequence of (1) and (3).
(5) We show: X is closed if and only if for each nite subcomplex Y the
intersection Y is closed in Y .
Suppose the condition is satised, and let be an arbitrary cell. Then X( )
is closed in X( ), hence, by (2) and (3), closed in X; therefore

=
X( )

is closed in X and in

, hence closed in X by condition (W4).
(6) If Y is a nite subcomplex, then Y is a nite subcomplex, hence closed.
By (5), is closed.
(8.2.5) Proposition. A subcomplex Y of a Whitehead complex X is a Whitehead
complex.
Proof. Let e be a cell in Y and : D
n
X a characteristic map. Then (D
n
) =
e Y , since Y is closed. Hence can be taken as a characteristic map for Y .
It remains to verify condition (W4). Let 1 Y and suppose 1 e is closed
in e for each cell e in Y . We have to show that 1 is closed in Y . We show that 1
is closed in X. Let be a cell of X. By (W3),

is contained in a nite union
e
1
L L e
k
of cells. Let e
1
. . . . . e
}
be those which are contained in Y . Then
Y e
1
L L e
}
e
1
L L e
}
Y
since Y is a subcomplex. Hence
Y = (

e
1
) L L(

e
}
).

1 =

1Y =
_
}
k=1
(

e
k
1).
By assumption, e
k
1 is closed in e
k
; hence

e
k
1 is closed in

; therefore
1 is a nite union of sets which are closed in X.

(8.2.6) Proposition. Let X be a Whitehead complex. Then:
(1) X carries the colimit topology with respect to the family (X
n
[ n N
0
).
(2) Let (e
2
[ z (n)) be the family of n-cells of X with characteristic maps
2
: D
n
2
X
n
and restrictions
2
: S
n-1
2
X
n-1
. Then
2
S
n-1
2
i
=(
)

X
n-1
2
D
n
2
=(
)

X
n
is a pushout in TOP. (X
-1
= 0.)
Proof. (1) Suppose X
n
is closed in X
n
for each n. Then for each n-cell e of
X the set e = e X
n
is closed in e. By (W4), is closed in X.
(2) The diagram is a pushout of sets. Give X
n
the pushout topology and denote
this space by 7. By construction, the identity k : 7 X
n
is continuous. We show
that k is also closed. Let V 7 be closed. By denition of the pushout topology
this means:
(i) V X
n-1
is closed in X
n-1
.
(ii)
-1
(V ) D
n
2
is closed in D
n
2
, hence also compact.
We conclude that (
-1
(V ) D
n
2
) = V (D
n
2
) = V e
2
is closed in e
2
, being
a continuous image of a compact space in a Hausdorff space. From (i) and (ii) we
therefore conclude that for each cell e of X
n
the set V e is closed in e. Since X
n
is a Whitehead complex, V is closed in X
n
.
(8.2.7) Proposition. Let X be a Whitehead complex, pointed by a 0-cell. The
inclusions of the nite pointed subcomplexes J X induce a canonical map
colim
T

k
(J. +)
k
(X. +). This map is an isomorphism.
Recall from Section 7.9 the notion of a k-space and the k-space k(X) obtained
from a space X.
(8.2.8) Proposition. Let X have a cell decomposition such that (W1)(W3) hold
and such that each compact set is contained in a nite number of cells. Then k(X)
is a Whitehead complex with respect to the given cell decomposition and the same
characteristic maps. Moreover, X is aWhitehead complex if and only if k(X) = X.
Proof. Let : D
n
X be a characteristic map for the cell e. Since e is compact
it has the same topology in k(X). Hence : D
n
k(X) is continuous. Since
is a quotient map and
-1
(e) = 1
n
, we see that e has the same topology in k(X)
and X. Thus e is a cell in k(X) with characteristic map .
Let e be closed in e for each cell e. Let 1 k(X) be compact. By
hypothesis, 1 is contained in a nite number of cells, say 1 e
1
L Le
k
. Then
1 = (( e
1
) L L( e
k
)) 1 is closed in 1. Hence is k-closed.
8.3. CW-Complexes 203
Let X and Y be Whitehead complexes and e X, Y cells. Then
e X Y is a cell. From characteristic maps : D
n
X, : D
n
Y
for e. we obtain : D
n
D
n
X Y , and this can be considered as a
characteristic map for e . For this purpose use a homeomorphism
(D
nn
. S
nn-1
) (D
n
D
n
. D
n
S
n-1
L S
n-1
D
n
).
With this cell structure, X Y satises conditions (W1)(W3) in the denition
of a Whitehead complex. In general, property (W4) may not hold. In this case
one re-topologizes X Y such that the compact subsets do not change. The space
X
k
Y = k(X Y ) is then a Whitehead complex (see (8.2.8)).
Problems
1. R carries the structure of a Whitehead complex with 0-cells {n], n Z and 1-cells
|n. n 1, n Z. There is an analogous Whitehead complex structure W() on R
n
with
0-cells the set of points (k
1
. . . . . k
n
), k
Z, > 0 xed and the associated -cubes.

Thus, given a compact set 1 R
n
and a neighbourhood U of 1, there exists another
neighbourhood 1 of 1 contained in U such that 1 is a subcomplex of the complex W().
In this sense, compact subsets can be approximated by nite complexes.
2. The geometric realization of a simplicial complex is a Whitehead complex.
8.3 CW-Complexes
We now use (8.2.6) as a starting point for another denition of a cell complex. Let
(X. ) be a pair of spaces. We say, X is obtained from by attaching an n-cell,
if there exists a pushout
S
n-1

D
n

X.
Then is closed in X and X \ is homeomorphic to 1
n
via . We call X \ an
n-cell in X, its attaching map and its characteristic map.
(8.3.1) Proposition. Let a commutative diagram with closed embeddings . J be
given:
Y
J
X
T
7.
Suppose J induces a bijection X 7 Y . Then the diagram is a pushout,
provided that (1) J(X) 7 is closed; (2) J : X J(X) is a quotient map.
Condition (2) holds if X is compact and 7 Hausdorff.
Proof. Let g: X U and h: Y U be given such that g = h . The diagram
is a set-theoretical pushout. Therefore there exists a unique set map : 7 U
with J = g, J = h. Since J is a closed embedding, [J(Y ) is continuous.
Since J is a quotient map, [J(X) is continuous. Thus is continuous, since
J(X) and J(Y ) are closed sets which cover 7.
(8.3.2) Note. Let X be a Hausdorff space and a closed subset. Suppose there
exists a continuous map : D
n
X which induces a homeomorphism : 1
n
X \ . Then X is obtained from by attaching an n-cell.

Proof. We show(S
n-1
) . Suppose there exists s S
n-1
with (s) X\.
Then there exists a unique t 1
n
with (s) = (t ). Let V 1
n
, W D
n
be disjoint open neighbourhoods of t. s. Then (V ) X \ is open in X, since
: 1
n
X\is a homeomorphismand is closed in X. Since is continuous,
there exists an open neighbourhood W
1
W of s with (W
1
) (V ). This
contradicts the injectivity of [1
n
.
Thus provides us with a map : S
n-1
. We now use (8.3.1).
(8.3.3) Example. The projective space R1
n
is obtained fromR1
n-1
by attaching
an n-cell. The projective space C1
n
is obtained from C1
n-1
by attaching a 2n-
cell.
We recall that C1
n-1
is obtained from S
2n-1
by the equivalence relation
(:
1
. . . . . :
n
) ~ (z:
1
. . . . . z:
n
), z S
1
, or from C
n
\ 0 by : ~ z:, z C
+
.
The class of : is denoted :
1
. . . . . :
n
|. A characteristic map : D
2n
C1
n
is
. ..
_
1 [.[
2
|.
The space R1
n-1
is obtained fromS
n-1
by the relation : ~ :, or fromR
n
\0
by : ~ z:, z R
+
. A characteristic map : D
n
R1
n
is given by the same
formula as in the complex case.
We can also attach several n-cells simultaneously. We say X is obtained from
by attaching n-cells if there exists a pushout
}J
S
n-1
}

}J
D
n
}
X.
The index just enumerates different copies of the same space. Again, is then
closed in X and induces a homeomorphism of

}
1
n
}
with X \ . Therefore
X \ is a union of components and each component is an n-cell. (By invariance
of dimension, the integer n is determined by X \ .) We allow J = 0; in that case
= X. We write
}
= [D
n
}
and
}
= [S
n-1
and call
}
the characteristic
map of the n-cell (1
n
}
) and
}
its attaching map.
8.3. CW-Complexes 205
Let us give another interpretation: X = X() is the double mapping cylinder
of J
pr
S
n
J
where J is a discrete set. From this setting we see: If

is replaced by a homotopic map [, then X() and X([) are h-equivalent under .
Let : Y be a given map. Assume that X is obtained fromby attaching
n-cells via attaching maps (
}
):

}
S
n-1
}
. From the pushout denition of
the attaching process we obtain:
(8.3.4) Note. There exists an extension J : X Y of if and only if the maps

}
are null homotopic. We view a null homotopy of
}
as an extension to D
n
}
.
Then the extensions J correspond to the set of null homotopies of the
}
.
In view of this note we call the homotopy classes
}
| the obstructions to
extending .
Let be a subspace of X. ACW-decompositionof (X. ) consists of a sequence
of subspaces = X
-1
X
0
X
1
X such that:
(1) X = L
n_0
X
n
.
(2) For each n _ 0, the space X
n
is obtained from X
n-1
by attaching n-cells.
(3) X carries the colimit topology with respect to the family (X
n
).
X
k
is a subspace of the colimit X of a sequence X
}
X
}1
. If the
inclusions are closed, then X
k
is closed in X. This is an immediate consequence
of the denition of the colimit topology.
A pair (X. ) together with a CW-decomposition (X
n
[ n _ 1) is called a
relative CW-complex. In the case = 0 we call X a CW-complex. The space X
n
is the n-skeleton of (X. ) and (X
n
[ n _ 1) is the skeleton ltration. The cells
of X
n
\ X
n-1
are the n-cells of (X. ). We say, (X. ) is nite (countable etc.) if
X \ consists of a nite (countable etc.) number of cells. If X = X
n
, X = X
n-1
we denote by n = dim(X. ) the cellular dimension of (X. ). If = 0, then
is suppressed in the notation. We call X a CW-space if there exists some cellular
decomposition X
0
X
1
of X.
Let X be a Whitehead complex. From (8.2.6) we obtain a CW-decomposition
of X. The converse also holds: From a CW-decomposition we obtain a decompo-
sition into cells and characteristic maps; it remains to verify that X is a Hausdorff
space and carries the colimit topology with respect to the closures of cells (see
(8.3.8)).
In the context of CW-complexes (X. ), the symbol X
n
usually denotes the
n-skeleton and not the n-fold Cartesian product.
(8.3.5) Note. If (X. ) is a relative CW-complex, then also (X. X
n
) and (X
n
. )
are relative CW-complexes, with the obvious skeleton-ltration inherited from
(X
n
[ n _ 1).
(8.3.6) Example. From (8.3.3) we obtain cellular decompositions of C1
n
and
R1
n
. The union of the sequence R1
n
R
n1
denes the innite projective
space R1
o
as a CW-complex. It has a single n-cell for each n _ 0. Similarly, we
obtain C1
o
with a single cell in each even dimension.
(8.3.7) Example. The sphere S
n
has a CW-decomposition with a single 0-cell
and a single n-cell, and another CW-composition with two -cells for each
{0. . . . . n], see (8.2.3). The quotient map S
n
R1
n
sends each cell of the latter
homeomorphically onto a cell of R1
n
in the decomposition (8.3.6). We can also
form the colimit S
o
of S
n
S
n1
, a CW-complex with two cells in each
dimension.
The general topology of adjunction spaces and colimit topologies gives us the
next results.
(8.3.8) Proposition. Let (X. ) be a relative CW-complex. If is a T
1
-space, then
X is a T
1
-space and a compact subset of X meets only a nite number of cells. If
is a Hausdorff space, then X is a Hausdorff space. If is normal, then X is
normal. If is a Hausdorff space, then X carries the colimit topology with respect
to the family which consists of and the closures of cells.
Proof. We only verify the last statement. Let C be a subset of X and suppose C
in closed in and e closed in e for each cell e. We show inductively, that
C X
n
is closed in X
n
. This holds for n = 1 by assumption. The space X
n
is a quotient of 7
n
= X
n-1
D
n
}
. Each characteristic map
}
: D
n
}
e
}
is
a quotient map, since X is Hausdorff. From the assumptions we see that X
n
C
has a closed pre-image in 7
n
.
The considerations so far show that a CW-complex is a Whitehead complex.
(8.3.9) Proposition. Let (X. ) be a relative CW-complex. Then X is a
cobration.
Proof. We know that

S
n-1
}

D
n
}
is a cobration. Hence X
n-1
X
n
is
an induced cobration. Therefore the compositions X
n
X
nk
are cobrations.
Given : X 7 and a homotopy h
-1
: X
-1
1 7 of [X
-1
, we can extend
this inductivelytohomotopies h
n
: X
n
1 7suchthat h
n1
[X
n
1 = h
n
. Since
X 1 is the colimit of the X
n
1, the h
n
combine to a homotopy h: X 1 7.
Problems
1. The attaching map for the n-cells yields a homeomorphism
_
(D
n
,S
n-1
)
X,.
2. Let (X. ) and (Y. T) be relative CW-complexes. Consider X Y with the closed
subspaces
(X Y )
n
=
_
n1
i=-1
X
i
Y
n-i
. n _ 1.
8.4. Weak Homotopy Equivalences 207
In favorable cases, the ltration ((X Y )
n
[ n _ 1) is a CW-decomposition of the pair
(X Y. T).
Let Y be locally compact. Then (X Y )
n
is obtained from (X Y )
n-1
by attaching
n-cells.
3. Let (X. ) be a relative CW-complex and let C . Then (X,C. ,C) is a relative
CW-complex with CW-decomposition (X
n
,C). Moreover, X, is a CW-complex.
4. Let X be a subcomplex. Then X, is a CW-complex.
5. Let and T be subcomplexes of X. Then ,( T) is a subcomplex of X,T.
6. Let be a subcomplex of T and Y another CW-complex. Then .
k
Y is a subcomplex
of T .
k
Y .
7. Let be a CW-complex. Suppose X is obtained fromby attaching n-cells via attaching
maps :

S
n-1

n-1
. Then X is a CW-complex with CW-decomposition X
for < n and X
L (X ) for _ n, and is a subcomplex of X.

8. Let
0
.
1
:

,
S
n-1
be homotopic attaching maps. The spaces X(0). X(1) which

are obtained by attaching n-cells with
0
.
1
are h-equivalent under . (Homotopy theorem
for cobrations.)
9. Let X be a pointed CW-complex with base point + a 0-cell. Then the cone CX and the
suspension X are CW-complexes. (In statements of this type the reader is asked to nd a
canonical cell decomposition induced from the initial data.)
10. Let (X
[ J) be a family of pointed CW-complexes with base point a 0-cell. Then

_
J
X
has the structure of a CW-complex such that the summands are subcomplexes.
11. Let : 1 T be a Serre bration and (X. ) a CW-pair. Then each homotopy
h: X 1 T has a lifting along with given initial condition on X 0 L 1.
12. Suppose X is obtained from by attaching n-cells. Let : 1 X be a covering and
1
/
=
-1
(). Then 1 is obtained from 1
/
by attaching n-cells.
13. Let X be a CW-complex with n-skeleton X
n
and : 1 X a covering. Then 1 is a
CW-complex with n-skeleton 1
n
=
-1
(T
n
) such that maps the cells of 1 homeomor-
phically to cells of X. An automorphism of maps cells of 1 homeomorphically to cells.
14. Each neighbourhood U of a point . of a CW-complex contains a neighbourhood V
which is pointed contractible to .. A connected CW-complex has a universal covering. The
universal covering has a cell decomposition such that its automorphism group permutes the
cells freely.
15. Let X and Y be countable CW-complexes. Then X Y is a CW-complex in the product
topology.
8.4 Weak Homotopy Equivalences
We now study the notion of an n-connected map and of a weak homotopy equiva-
lence in the context of CW-complexes.
(8.4.1) Proposition. Let (Y. T) be n-connected. Then a map : (X. ) (Y. T)
from a relative CW-complex (X. ) of dimension dim(X. ) _ n is homotopic
relative to to a map into T. In the case that dim(X. ) < n the homotopy class
of X T is unique relative to .
Proof. Induction over the skeleton ltration. Suppose X is obtained from by
attaching q-cells via :

k
S
q-1
k
, q _ n. Consider a commutative diagram
S
q-1
k
T
}
D
q
k
X
T
Y .
Since (Y. T) is n-connected, J is homotopic relative to

S
q-1
k
to a map into
T. Since the left square is a pushout, we obtain a homotopy of J from a pair of
homotopies of J and Ji which coincide on
S
q-1
k
. Since we have homotopies
of J relative to
S
q-1
k
, we can use on the constant homotopy. Altogether we
obtain a homotopy of J relative to to a map into T.
For an arbitrary (X. ) with dim(X. ) _ n we apply this argument inductively.
Suppose we have a homotopy of relative to to a map g which sends X
k
into
T. By the argument just given we obtain a homotopy of g[X
k1
relative to X
k
which sends X
k1
into T. Since X
k1
X is a cobration, we extend this
homotopy to X. In the case that n = o, we have to concatenate an innite number
of homotopies. We use the rst homotopy on 0. 1,2| the second on 1,2. 3,4| and
so on. (Compare the proof of (8.5.4).) Suppose dim(X. ) < n. Let J
0
. J
1
: X
T be homotopic relative to to . We obtain from such homotopies a map
(X 1. X d1 L 1) (Y. T) which is the constant homotopy on . We
apply the previous argument to the pair (X 1. X d1 L1) of dimension _ n
and see that the homotopy class of the deformation X T of is unique relative
to .
(8.4.2) Theorem. Let h: T Y be n-connected, n _ 0. Then h
+
: X. T|
X. Y | is bijective (surjective) if X is a CW-complex with dimX < n (dimX _ n).
If h: T Y is pointed, then h
+
: X. T|
0
X. Y |
0
is injective (surjective) in the
same range.
Proof. By use of mapping cylinders we can assume that h is an inclusion. The
surjectivity follows if we apply (8.4.1) to the pair (X. 0). The injectivity follows,
if we apply it to the pair (X 1. X d1). In the pointed case we deform(X. +)
(Y. T) rel {+] to obtain surjectivity, and for the proof of injectivity we apply (8.4.1)
to the pair (X 1. X d1 L + 1).
(8.4.3) Theorem. Let : Y 7 be a map between CW-complexes.
(1) is a homotopy equivalence, if and only if for each b Y and each q _ 0
the induced map
+
:
q
(Y. b)
q
(7. (b)) is bijective.
8.4. Weak Homotopy Equivalences 209
(2) Suppose dimY _ k, dim7 _ k. Then is a homotopy equivalence if
+
is
bijective for q _ k.
Proof. (1) If
+
is always bijective, then
+
is a weak equivalence, hence the induced
map
+
: X. Y | X. 7| is bijective for all CW-complexes X (see (8.4.2)). By
category theory, represents an isomorphism in h-TOP: Take X = 7; then there
exists g: 7 Y such that g . id(7). Then g
+
is always bijective. Hence g
also has a right homotopy inverse.
(2)
+
: 7. Y | 7. 7| is surjective, since is k-connected (see (8.4.2)).
Hence there exists g: 7 Y such that g . id(7). Then g
+
:
q
(7)
q
(Y )
is bijective for q _ k, since
+
g
+
= id and
+
is bijective. Hence there exists
h: Y 7 with gh . id(Y ). Thus g has a left and a right h-inverse and is therefore
an h-equivalence. Fromg . id we then conclude that is an h-equivalence.
The importance of the last theorem lies in the fact that homotopy equivalence
can be tested algebraically. Note that the theorem does not say: If
q
(Y )
q
(7)
for each q, then Y and 7 are homotopy equivalent; it is important to have a map
which induces an isomorphism of homotopy groups. Mapping a space to a point
gives:
(8.4.4) Corollary. A CW-complex X is contractible if and only if
q
(X) = 0 for
q _ 0.
(8.4.5) Example. From
}
(S
n
) = 0 for < n and
}
(S
o
) = colim
n
}
(S
n
) we
conclude that the homotopy groups of S
o
are trivial. Hence S
o
is contractible.
(8.4.6) Example. A simply connected 1-dimensional complex is contractible. A
contractible 1-dimensional CW-complex is called a tree.
(8.4.7) Theorem. A connected CW-complex X contains a maximal (with respect
to inclusion) tree as subcomplex. A tree in X is maximal if and only if it contains
each 0-cell.
Proof. Let B denote the set of all trees in X, partially ordered by inclusion. Let
T T be a totallyorderedsubset. ThenC =
_
TT
T is contractible:
1
(C) = 0,
since a compact subset of C is contained in a nite subcomplex and therefore in
some T T . Thus, by Zorns lemma, there exist maximal trees.
Let T be a maximal tree. Consider the 1-cells which have at least one end point
in T. If the second end point is not contained in T, then T is obviously not maximal.
Therefore the union V of these 1-cells together with T form a subcomplex of X
1
,
and the remaining 1-cells together with their end points forma subcomplex X
1
V .
Since X is connected so is X
1
, hence V = X
1
, and T
0
= V
0
= X
0
.
Let T be a tree which contains X
0
. Let T
t
T be a strictly larger tree. Since
T is contractible, T
t
and T
t
,T are h-equivalent. Hence T
t
,T is contractible.
Since X
0
T, the space T
t
,T has the form
_
S
1
and is not simply connected.
Contradiction.
We now generalize the suspension theorem (6.10.4). Let X and Y be pointed
spaces. We have the suspension map
+
: X. Y |
0
X. Y |
0
. We use the
adjunction X. Y |
0
X. CY |
0
. The resulting map X. Y |
0
X. CY |
0
is then induced by the pointed map o : Y CY which assigns to , Y the
loop t ,. t | in Y .
(8.4.8) Theorem. Suppose
i
(Y ) = 0 for 0 _ i _ n. Then the suspension
+
: X. Y |
0
X. Y |
0
is bijective (surjective) if X is a CW-complex of di-
mension dimX _ 2n (dimX _ 2n 1).
Proof. By the suspension theorem (6.10.4), the map o is (2n1)-connected. Now
use the pointed version of (8.4.2).
(8.4.9) Theorem. Let X be a nite pointed CW-complex. Then
+
:
k
X.
k
Y |
0
k1
X.
k1
Y |
0
is bijective for dim(X) _ k 1.
Proof. We have dim
k
X = k dimX. The space Y is path connected. By
the theorem of Seifert and van Kampen,
2
Y is simply connected. From the
suspension theorem we conclude that
}
(
k
Y ) = 0 for 0 _ _ k 1. By the
previous theorem,
+
is a bijection for k dimX _ 2(k 1).
8.5 Cellular Approximation
(8.5.1) Proposition. Suppose X is obtained from by attaching (n 1)-cells.
Then (X. ) is n-connected.
Proof. We know that (D
n1
. S
n
) is n-connected. Now apply (6.4.2).
(8.5.2) Proposition. Let X be obtained from by attaching n-cells (n _ 1).
Suppose is simply connected. Then the quotient map induces an isomorphism
n
(X. )
n
(X,).
Proof. (8.5.1) and (6.10.2).
(8.5.3) Proposition. For each relative CW-complex (X. ) the pair (X. X
n
) is
n-connected.
Proof. From (8.5.1) we obtain by induction on k that (X
nk
. X
n
) is n-connected.
The compactness argument (8.3.8) nally shows (X. X
n
) to be n-connected.
Let X and Y be CW-complexes. Amap : X Y is cellular, if (X
n
) Y
n
for each n N
0
. The cellular approximation theorem (8.5.4) is an application of
(8.4.1).
8.6. CW-Approximation 211
(8.5.4) Theorem. A map : X Y is homotopic to a cellular map g: X Y .
If T X is a subcomplex and [T cellular, then the homotopy . g can be
chosen relative to T.
Proof. We showinductively that there exist homotopies H
n
: X1 Y such that
(1) H
0
0
= , H
n-1
1
= H
n
0
for n _ 1;
(2) H
n
1
(X
i
) Y
i
for i _ n;
(3) H
n
is constant on X
n-1
L T.
For the induction step we assume (X
i
) Y
i
for i < n. Let : (D
n
. S
n-1
)
(X
n
. X
n-1
) be a characteristic map of an n-cell not contained in T. The map
is homotopic relative to S
n-1
to a map into Y
n
, since (Y. Y
n
) is n-connected. A
corresponding homotopy is used to dene a homotopy of on the associated closed
n-cells. This process denes the homotopy on T L X
n
; and we extend it to X,
using the fact that T LX
n
X is a subcomplex and hence a cobration. We now
concatenate the homotopies H
n
:
H(.. t ) =
_
H
i
(.. 2
i1
(t 1 2
-i
)). 1 2
-i
_ t _ 1 2
-i-1
.
H
i
(.. 1). . X
i
. t = 1.
This map is continuous on X
i
1 and hence on X1, since this space is the colimit
of the X
i
1.
(8.5.5) Corollary. Let
0
.
1
: X Y be cellular maps which are homotopic.
Then there exists a homotopy between them such that (X
n
1) Y
n1
. If
0
.
1
are homotopic rel T, then can be chosen rel T.
Proof. Choose a homotopy :
0
.
1
rel T. Then maps X d1 LT 1 into
Y
n
. Now apply (8.5.4) to X d1 L T 1 X 1.
Problems
1. Let X be a subcomplex and : Y a cellular map. Then Y = X L
Y is a
CW-complex.
2. A CW-complex is path connected if and only if the 1-skeleton is path connected. The
components are equal to the path components, and the path components are open.
8.6 CW-Approximation
We show in this section, among other things, that each space is weakly homotopy
equivalent to a CW-complex. Our rst aim is to raise the connectivity of a map.
(8.6.1) Theorem. Let : Y be a k-connected map, k _ 1. Then there
exists for each n > k a relative CW-complex (X. ) with cells only in dimensions
{k 1. . . . . n], n _ o, and an n-connected extension J : X Y of . If
is CW-complex, then can be chosen as a subcomplex of X.
Proof. (Induction over n.) Recall that the map is k-connected if the induced
map
+
:
}
(. +)
}
(Y. (+)) is bijective for < k and surjective for = k
(no condition for k = 1). If we attach cells of dimension greater than k and
extend, then the extension remains k-connected. This fact allows for an inductive
construction.
Let n = 0, k = 1. Suppose
+
:
0
()
0
(Y ) is not surjective. Let
C = {c
}
[ J] be a family of points in Y which contains one element fromeach
path component
0
(Y ) \
+
0
(). Set X =
D
0
}
and dene J : X Y
by J[ = and J(D
0
}
) = {c
}
]. Then X is obtained from by attaching 0-cells
and J is a 0-connected extension of .
n = 1. Suppose : Y is 0-connected. Then
+
:
0
()
0
(Y ) is
surjective. Let c
-1
. c
1
be points in different path components of which have the
same image under
+
. Then : S
0
, (1) = c
1
, is an attaching map for a
1-cell. We can extend over L
D
1
by a path from (c
-
) to (c
). Treating
other pairs of path components similarly, we obtain an extension J
t
: X
t
Y of
over a relative 1-complex (X
t
. ) such that J
t
+
:
0
(X
t
)
0
(Y ) is bijective. The
bijectivity of J
t
+
follows from these facts: We have J
t
+
+
=
+
with the inclusion
: X
t
; the map is 0-connected; path components with the same image
under
+
have, by construction, the same image under
+
.
We still have to extend J
t
: X
t
Y to a relative 1-complex X X
t
such that
J
+
:
1
(X. .)
1
(Y. (.)) is surjective for each . X. Let J
}
: (D
1
. S
0
)
(Y. ,) be a familyof maps suchthat the J
}
|
1
(Y. ,) together withJ
t
+
(
1
(X
t
. .))
generate
1
(Y. ,), , = J
t
(.). Let X X
t
be obtained from X
t
by attaching 1-
cells with characteristic maps (
}
.
}
): (D
1
. S
0
) (X
t
. .). We extend J
t
to J
such that J
}
= J
}
. Then J
+
:
1
(X. .)
1
(Y. ,) is surjective.
n _ 2. Suppose : Y is (n 1)-connected. By the use of mapping
cylinders, we can assume that is an inclusion. Let (
}
.
}
): (D
n
. S
n-1
. e
0
)
(Y. . a) be a set of maps such that the ,
}
=
}
.
}
|
n
(Y. . a) generated the
1
(. a)-module
n
(Y. . a). We attach n-cells to by attaching maps
}
to obtain
X and extend to J by the null homotopies
}
of
}
. The characteristic map
of the n-cell with attaching map
}
represents .
}

n
(X. . a) and J
+
.
}
= ,
}
.
The map J induces a morphism of the exact homotopy sequence of (X. . a) into
the sequence of (Y. . a), and J
+
:
n
(X. . a)
n
(Y. . a) is surjective by
construction. Consider the diagram
n
()
=

n
(X)

T
(1)
n
(X. )

T
(2)
n-1
()

=
n-1
(X)
T
(3)
n
()

n
(Y )

n
(Y. )

n-1
()

n-1
(Y )

0.
The sequences end with 0, since
n-1
(X. ) = 0 and
n-1
()
n-1
(Y ) is
surjective by assumption. (2) is surjective. The Five Lemma shows us that (1)
is surjective and (3) injective. By induction hypothesis, (3) is already surjective.
Hence J is n-connected.
In order to obtain as a subcomplex of X, one works with cellular attaching
maps.
(8.6.2) Theorem. Let Y be a CW-complex such that
i
(Y ) = 0 for 0 _ i _ k.
Then Y is homotopy equivalent to a CW-complex X with X
k
= {+].
Proof. Start with the k-connected map : = {+] Y and extend it to a weak
equivalence J : X Y by attaching cells of dimension greater than k.
(8.6.3) Proposition. Let and T be pointed CW-complexes. Assume that is
(m 1)-connected and T is (n 1)-connected. Then .
k
T is (m n 1)-
connected.
Proof. We can assume that has no cells in dimensions less than m and n no cells
in dimensions less than n (except the base point). Then .
k
T has no cells in
dimensions less than mn.
(8.6.4) Theorem. Let (X
}
[ J) be a family of (n1)-connected CW-complexes.
Let |
k
: X
k

_
}J
X
}
be the inclusion of the k-th summand. Then
J
= (|
}+
: )
}J

n
(X
}
)
n
_ _
}J
X
}
_
is an isomorphism (n _ 2).
Proof. Let J be nite. Up to h-equivalence we can assume that X
}
has no cells
in dimensions less than n, except the base point. Then

}
X
}
is obtained from
_
}
X
}
by attaching cells of dimension _ 2n. Hence
n
_ _
X
}
_

n
_
X
}
_
is
an isomorphism for m _ 2n 2. From the diagram
n
_ _
X
}
_

n
_
X
}
_
(]
n
(X
}
)
(1)
n
(X
}
)
we conclude that
J
is an isomorphism.
Let now J be arbitrary. For each .
n
_ _
}J
X
}
_
there exists a nite 1 J
such that . is contained in the image of
n
_ _
T
X
}
_

n
_ _
J
X
}
_
, since a
compact subset is contained in a nite wedge. The result for 1 now shows that
J
is surjective. If .
1
and .
2
have the same image under
J
, then these elements
are contained in some nite sum
T
and, again by a compactness argument, they
have the same image under some
T
, if 1 is chosen large enough. This shows the
injectivity of
J
.
(8.6.5) Proposition. Suppose
}
(Y ) = 0 for > n. Let X be obtained from
by attaching cells of dimension _ n 2. Then X induces a bijection
X. Y | . Y |.
Proof. Surjective. Let : Y be given. Attach (n 2)-cells via maps
: S
n1
. Since : S
n1
Y is null homotopic, we can extend
over the (n 2)-cells. Continue in this manner.
Injective. Use the same argument for (X 1. X d1 L 1). The cells of
this relative complex have a dimension > n 2.
(8.6.6) Theorem. Let be an arbitrary space and k N
0
. There exists a relative
CW-complex (X. ) withcells only indimensions _ k2, suchthat
n
(X. .) = 0
for n > k and. X, andthe inducedmap
n
(. a)
n
(X. a) is anisomorphism
for n _ k and a .
Proof. We construct inductively for t _ 2 a sequence = X
k1
X
k2
X
kt
such that
n
(. a)
n
(X
kt
. a) for n _ k,
n
(X
kt
. a) = 0 for
k < n _ k t 1, and X
n1
is obtained from X
n
by attaching (m1)-cells.
The induction step: If we attach (m 1)-cells to X
n
by the attaching maps
}
: (S
n
}
. e
0
) (X
n
. a) to obtain X
n1
, then
n
(X
n
. a)
n
(X
n1
. a) for
n _ m 1. The exact sequence
n1
(X
n1
. X
n
. a)
J
n
(X
n
. a)
n
(X
n1
. a) 0
shows that the
}
| are in the image of d. Thus, if the
}
| generate
n
(X
n
. a),
then
n
(X
n1
. a) = 0.
(8.6.7) Example. We can attach cells of dimension _ n2 to S
n
to obtain a space
1(Z. n) which has a single non-trivial homotopy group
n
(1(Z. n)) Z. See
the section on EilenbergMac Lane spaces for a generalization.
Let i
A
n
: X Xn| be an inclusion of the type constructed in (8.6.6), namely
Xn| is obtained by attaching cells of dimension greater than n 1 such that
k
(Xn|) = 0 for k > n and i
A
n
k
(i
A
n
) for k _ n. Given
a map : X Y and i
Y
n
: Y Y m| for m _ n, there exists a unique homotopy
class
n,n
: Xn| Y m| such that i
Y
n
=
n,n
i
A
n
; this is a consequence
of (8.6.5). We let
A
n
: X(n) X be the homotopy bre of i
A
n
. We call
A
n
the
n-connective covering of X. The induced map
i
(
A
n
):
i
(X(n))
i
(X) is an
isomorphism for i > n and
i
(X(n)) = 0 for i _ n. The universal covering has
such properties in the case that n = 1. So we have a generalization, in the realm of
brations. Objects of this type occur in the theory of Postnikov decompositions of
a space, see e.g., [192].
As a consequence of (8.6.1) for = 0 we see that for each space Y there
exists a CW-complex X and a weak equivalence : X Y . We call such a weak
equivalence a CW-approximationof Y . Note that a weak equivalence between CW-
complexes is a homotopy equivalence (8.4.3). We show that CW-approximations
are unique up to homotopy and functorial in the homotopy category.
(8.6.8) Theorem. Let : Y
1
Y
2
be a continuous map and let
}
: X
}
Y
}
be
CW-approximations. Then there exists a map : X
1
X
2
such that
1
.
2
,
and the homotopy class of is uniquely determined by this property.
Proof. Since
2
is a weakequivalence,
2
: X
1
. X
2
| X
1
. Y
2
| is bijective. Hence
there exists a unique homotopy class such that
1
.
2
.
A domination of X by 1 consists of maps i : X 1. : 1 X and a
homotopy i . id(X).
(8.6.9) Proposition. Suppose M is dominated by a CW-complex X. Then M has
the homotopy type of a CW-complex.
Proof. Suppose i : M X and r : X M are given such that ri is homotopic to
the identity. There exists a CW-complex | : X Y and an extension 1: Y M
of r such that 1 induces an isomorphism of homotopy groups. Let = |i : M
X Y . Since 1 = ri . id, the composition 1 induces isomorphisms of
homotopy groups, hence so does . From 1 . we conclude that 1 induces
the identity on homotopy groups and is therefore a homotopy equivalence. Let k
be h-inverse to 1, then (1k) . id. Hence has the left inverse 1 and the right
inverse 1k and is therefore a homotopy equivalence.
A (half-exact) homotopy functor on the category C
0
of pointed connected CW-
spaces is a contravariant functor h: C
0
SET
0
into the category of pointed sets
(1) (Homotopy invariance) Pointed homotopic maps induce the same morphism.
(2) (MayerVietoris property) Suppose X is the union of subcomplexes and T.
If a h() and b h(T) are elements with the same restriction in h(T),
then there exists an element . h(X) with restrictions a and b.
(3) (Additivity) Let X =
_
}
X
}
with inclusions i
}
: X
}
X. Then
h(X)
}
h(X
}
). . (h(i
}
).)
is bijective.
(8.6.10) Theorem(E. H. Brown). For each homotopy functor h: C
0
SET
0
there
exist 1 C
0
and u h(1) such that
X. 1|
0
h(X). |
+
(u)
is bijective for each X C
0
.
In category theory one says that 1 is a representing object for the functor h.
The theorem is called the representability theorem of E. H. Brown. For a proof see
[31], [4].
(8.6.11) Example. Let h(X) = X. 7|
0
for a connected pointed space 7. Then h
is a homotopy functor. From (8.6.10) we obtain 1 C
0
and : 1 7 such that
+
: X. 1|
0
X. 7|
0
is always bijective, i.e., is a weak h-equivalence. Thus
we have obtained a CW-approximation X of 7.
Problems
1. As a consequence of (8.6.8) one can extend homotopy functors from CW-complexes
to arbitrary spaces. Let J be a functor from the category of CW-complexes such that
homotopic maps . g induce the same morphism J( ) = J(g). Then there is, up to
natural isomorphism, a unique extension of J to a homotopy invariant functor on TOP which
maps weak equivalences to isomorphisms.
2. A point is a CW-approximation of the pseudo-circle.
3. Determine the CW-approximation of {0] L {n
-1
[ n N].
4. Let X and Y be CW-complexes. Show that the identity X
k
Y X Y is a CW-
approximation.
5. Let (Y
[ J) be a family of well-pointed spaces and
: X
a family of
pointed CW-approximations. Then
_
is a CW-approximation. Give a counterexample

(with two spaces) in the case that the spaces are not well-pointed.
6. Let : T and g: C D be pointed weak homotopy equivalences between well-
pointed spaces. Then . g is a weak homotopy equivalence.
7. Verify from the axioms of a homotopy functor that h(1) for a point 1 contains a single
element.
8. Verify from the axioms of a homotopy functor that for each inclusion X in C
0
the
canonical sequence h(X,) h(X) h() is an exact sequence of pointed sets.
8.7 Homotopy Classication
In favorable cases the homotopy class of a map is determined by its effect on
homotopy groups.
(8.7.1) Theorem. Let X be an (n 1)-connected pointed CW-complex. Let Y be
a pointed space such that
i
(Y ) = 0 for i > n _ 2. Then
h
A
: X. Y |
0
Hom(
n
(X).
n
(Y )). |
+
is bijective.
Proof. The assertion only depends on the pointed homotopy type of X. We use
(8.6.2) and assume X
n-1
= {+]. The h
A
constitute a natural transformation in the
variable X. Since (X. X
n1
) is (n1)-connected, the inclusionX
n1
X induces
8.8. EilenbergMac Lane Spaces 217
an isomorphism on
n
. By (8.6.5), the restriction r : X. Y |
0
X
n1
. Y |
0
is a
bijection. Therefore it sufces to consider the case, that X has, apart from the base
point, only cells of dimension n and n1. Moreover, by the homotopy theoremfor
cobrations, we can assume that the attaching maps for the (n1)-cells are pointed.
In this case X is the mapping cone of a pointed map : =
_
S
n
k

_
S
n
}
= T.
We therefore have the exact cobre sequence
. Y |
0
(
T. Y |
0
X. Y |
0
. Y |
0
.
Our assumption about Y yields . Y |
0
=
_ _
S
n
k
. Y
_
0

k

n1
(Y ) = 0.
We apply the natural transformation h and obtain a commutative diagram
. Y |
0
h
.
T. Y |
0
h
!
X. Y |
0
h
Hom(
n
.
n
Y ) Hom(
n
T.
n
Y )
Hom(
n
X.
n
Y )
0.
As one of the consequences of the excision theorem we showed that the sequence
n
()
n
(T)
n
(X) 0 is exact, and therefore the bottom sequence of
the diagram is exact. We show that h
and h
B
are isomorphisms. If = S
n
, then
h
:
n
(Y ) = S
n
. Y |
0
Hom(
n
(S
n
).
n
(Y ))
is an isomorphism, since
n
(S
n
) is generated by the identity. In the case that
=
_
S
n
k
, we have a commutative diagram
_ _
S
n
k
. Y
_
0

h
_
c
n
k
S
n
k
. Y |
h
c
n
k
Hom
_
n
_ _
S
n
k
_
.
n
(Y )
_
(1)
Hom(
n
(S
n
k
).
n
(Y )).
The map (1) is induced by the isomorphism

n
(S
n
k
)
n
(
_
S
n
k
) and there-
fore an isomorphism. We now want to conclude from the diagram by a Five
Lemma type argument that h
A
is bijective. The proof of surjectivity does not
use the group structure. Injectivity follows, if
+
1
is injective. In order to see
this, one can use the general fact that . Y |
0
acts on X. Y |
0
and the orbits are
mapped injectively, or one uses that is, up to homotopy, a suspension, because
+
:
_ _
S
n-1
k
.
_
S
n-1
}
_
0
_ _
S
n
k
.
_
S
n
}
_
0
is surjective.
8.8 EilenbergMac Lane Spaces
Let be an abelian group. An EilenbergMac Lane space of type K.; n/ is a
CW-complex 1(. n) such that
n
(1(. n)) and
}
(1(. n)) 0 for ,= n.
In the cases n = 0. 1, the group can be non-abelian. In the case n = 0, we think
of 1(. 0) = with the discrete topology.
(8.8.1) Theorem. EilenbergMac Lane spaces 1(. n) exist.
Proof. Let n _ 2. There exists an exact sequence
0

J
1
J
0

0
with free abelian groups J
0
and J
1
. We x a basis (a
k
[ k 1) of J
1
and
(b
}
[ J) of J
0
. Then is determined by the matrix (a
k
) =

}
n(. k)b
}
.
We now construct a geometric realization of this algebraic situation. The group
n
_ _
k
S
n
k
_
k

n
(S
n
k
) is free abelian (8.6.4). Abasis is given by the canonical
inclusions S
n
I

_
S
n
k
. There exists a unique homotopy class : =
_
k
S
n
k

_
S
n
}
= T which realizes the matrix (n(. k)) with respect to these bases. Let
X = C( ) be the mapping cone of . Then the sequence
n
()
(

n
(T)
(
1

n
(X)

0
is exact. Hence
n
(X) . Also
i
(X) = 0 for i < n. We can now attach cells
of dimensions _ n 2 to X in order to obtain a 1(. n), see (8.6.6).
(8.8.2) Examples. The space S
1
is a 1(Z. 1). We know
1
(S
1
) Z, and from
the exact sequence of the universal covering R S
1
we knowthat
n
(S
1
) = 0 for
n _ 2. The space C1
o
is a model for 1(Z. 2). The space R1
o
is a 1(Z,2. 1).
The adjunction X. Y |
0
X. CY |
0
shows that C1(. n 1) has the ho-
motopy groups of a 1(. n). By a theorem of Milnor [132], [67], CY has the
homotopy type of a CW-complex if Y is a CW-complex. If one does not want to
use this result one has the weaker result that there exists a weak homotopy equiva-
lence 1(. n) C1(. n 1).
We now establish further properties of EilenbergMac Lane spaces. We begin
by showing that EilenbergMac Lane spaces are H-spaces. Then we construct
product pairings 1(. m) . 1(j. n) 1( j. mn). In this context j
denotes the tensor product of the abelian groups and j (alias Z-modules) over Z.
We call the space 1(. n) polarized, if we have chosen a xed isomorphism
:
n
(1(. n)) . If (1(. n). ) and (1(j. n). ) are polarized complexes,
the product 1(. n) 1(j. n) will be polarized by
n
(1(. n) 1(j. n))
n
(1(. n))
n
(1(j. n))

1

2
.
(8.8.3) Proposition. Having chosen polarizations, we obtain from (8.7.1) an iso-
morphism 1(. n). 1(j. n)|
0
Hom(. j).
(8.8.4) Theorem. Let be an abelian group. Then an EilenbergMac Lane com-
plex 1(. n) is a commutative group object in h-TOP.
Proof. Let 1 = (1(. n). ) be a polarized complex with base point a 0-cell
e. For an abelian group , the multiplication j: , (g. h) gh is
a homomorphism. Therefore there exists a map m: 1 1 1, unique up to
homotopy, which corresponds under (8.8.3) to j. Similarly, | : , g g
-1
is
a homomorphism and yields a map i : 1 1. Claim: (1. m. i ) is an associative
and commutative H-space. The maps m (m id) and m (id m) induce the
same homomorphism when
n
is applied; hence these maps are homotopic. In a
similar manner one shows that . m(.. e) is homotopic to the identity. Since
1 1 1 1 is a cobration, we can change m by a homotopy such that
m(.. e) = m(e. .) = .. We write . m(.. i(.)) as composition
1
d
1 1
id i
1 1
n
1
and apply
n
; the result is the constant homomorphism. Hence this map is null
homotopic. Commutativity is veried in a similar manner by applying
n
. See also
Problem 1.
For the construction of the product pairing we need a general result about
products for homotopy groups. We take the smash product of representatives
: 1
n
,d1
n
X, g: 1
n
,d1
n
Y and obtain a well-dened map
n
(X)
n
(Y )
nn
(X .
k
Y ). ( |. g|) . g| = | . g|.
We call this map the .-product for homotopy groups. It is natural in the variables
X and Y .
(8.8.5) Proposition. The .-product is bi-additive.
Proof. The additivity in the rst variable follows directly from the denition of the
addition, if we use
1
. We see the additivity in the second variable, if we use the
composition laws
1
and
n1
in the homotopy groups.
(8.8.6) Proposition. Let be an (m 1)-connected and T an (n 1)-connected
CW-complex. Then . T is (mn 1)-connected and the .-product
n
()
n
(T)
nn
( .
k
T)
is an isomorphism(m. n _ 2). If mor nequals 1, then one has to use the abelianized
groups.
Proof. The assertion about the connectivity was shown in (8.6.3). For = S
n
the
assertion holds by the suspension theorem. For =
_
}
S
n
}
we use the commuta-
tive diagram
n
_ _
S
n
}
_
n
(T)
.

nn
__ _
S
n
}
_
.
k
T
_
_
n
(S
n
}
)
_
n
(T)
(1)
nn
_ _
(S
n
}
. T)
_
(2)
(
n
(S
n
}
)
n
(T))
(3)
(5)
nn
(S
n
}
. T).
(4)
(1) is an isomorphism by (8.6.4). (2) is induced by a homeomorphism. (3) is an

isomorphism by algebra. (4) is an isomorphism by (8.6.4). (5) is an isomorphism
by the suspension theorem. This settles the case of a wedge of m-spheres. Next
we let be the mapping cone of a map : C D where C and D are wedges of
m-spheres. Then we have a commutative diagram with exact rows:
n
(C)
n
(Y )

n
(D)
n
(T)

n
()
n
(T)

nn
(C .
k
T)

nn
(D .
k
T)

nn
( .
k
T)

0.
The general case now follows from the observation that the inclusion
n1

induces an isomorphismon
n
and
n1
.
k
T .
k
T induces an isomorphism
on
nn
.
Let (1(G. m). ), (1(H. n). ) and (1(GH. mn). ;) be polarized Eilen-
bergMac Lane complexes for abelian groups G and H. A product is a map
;
n,n
: 1(G. m) .
k
1(H. n) 1(G H. mn)
such that the diagram
n
(1(G. m))
n
(1(H. n))
.

nn
(1(G. m) .
k
1(H. n))
(;
m,n
)
G H
;
nn
(1(G H. mn))
is commutative. Here we have use the .-product (8.8.5).
(8.8.7) Theorem. There exists a product. It is unique up to homotopy.
Proof. Let G and H be abelian groups. The rst non-trivial homotopy group of
1(G. m) .
k
1(H. n) is
nn
and it is isomorphic to G H, see (8.8.6). By
(8.6.6) there exists an inclusion
;
n,n
: 1(G. m) .
k
1(H. n) 1(G H. mn).
We can choose the polarization ; so that the diagram above becomes commutative.
Uniqueness follows from (8.7.1).
The products (8.8.7) are associative, i.e.,
;
nn,]
(;
n,n
id) . ;
n,n]
(id ;
n,]
).
The products are graded commutative in the following sense:
1(t. mn) ;
n,n
. (1)
nn
t
t
;
n,n
with the interchange maps t
t
: 1(m. G) . 1(n. H) 1(n. H) . 1(m. G) and
t : G H H G.
Let 1 be a commutative ring with 1. We think of the multiplication as being a
homomorphismj: 11 1betweenabeliangroups. Fromthis homomorphism
we obtain a unique homotopy class 1(j): 1(1 1. m) 1(1. m).
We compose 1(j) with ;
k,I
and obtain a product map
m
k,I
: 1(1. k) .
k
1(1. l) 1(1. k l).
Also these products are associative and graded commutative.
In the associated homotopy groups H
k
(X: 1) = X
. 1(1. k)|
0
we obtain
via (. g) m
k,I
( . g) products
H
k
(X: 1) H
I
(Y : 1) H
kI
(X Y : 1).
which are also associative and graded commutative. (See also Problem 3.) In a
similar manner we can start from an 1-module structure 1 M M on M.
Later, when we study singular cohomology, we show that for a CW-complex X the
group X
. 1(1. k)|
0
is naturally isomorphic to the singular cohomology group
H
k
(X: 1) with coefcients in the ring 1. This opens the way to a homotopical
study of cohomology. The product (8.8.7) can then be used to construct the so-called
cup product in cohomology.
Once singular cohomology theory is constructed one obtains from the repre-
sentability theorem of Brown EilenbergMac Lane spaces as representing objects.
8.8.8 EilenbergMac Lane spectra. Let be an abelian group. The Eilenberg
Mac Lane spectrum H consists of the family (1(. n) [ n N
0
) of Eilenberg
Mac Lane CW-spaces and maps e
n
: 1(. n) 1(. n 1) (an inclusion of
subcomplexes; attach cells to 1(. n) to obtain a 1(. n1)). This spectrum is
an C-spectrum. We have proved in any case that c
n
: 1(. n) C1(. n 1) is
a weak homotopy equivalence. This sufces if one wants to dene the cohomology
theory only for pointed CW-spaces.
Problems
1. From the natural isomorphism X. 1(. n)|
0
X. C
2
1(. n 2)|
0
we see that X
X. 1(. n)|
0
is a contravariant functor into the category of abelian groups. Therefore, by
category theory, there exists a unique (up to homotopy) structure of a commutative h-group
on 1(. n) inducing the group structures of this functor.
2. Let
m
(X),
n
(Y ), and t : X .
k
Y Y .
k
X the interchange map. Then
. = (1)
mn
. .
3. Let M be an 1-module. A left translation l
: M M, . r. is a homomorphism
of the abelian group M and induces therefore a map 1
: 1(M. k) 1(M. k). Use these

maps to dene a natural structure of an 1-module on X. 1(M. k)|
0
.
4. The simply connected surfaces are S
2
and R
2
[44, p. 87]. If a surface is different from
S
2
and R1
2
, then it is a 1(. 1).
5. Let 1() T() be a -principal covering with contractible 1(). Then T() is a
1(. 1). Spaces of the type T() will occur later as classifying spaces. There is a bijection
1(. 1). 1(j. 1)| Hom(. j), ~between homotopy classes and group homomorphisms
up to inner automorphisms.
6. Let S
c
be the colimit of the unit spheres S(C
n
) S(C
n1
) . This space carries
a free action of the cyclic group Z,m S
1
by scalar multiplication. Show that S
c
with
this action is a Z,m-principal covering. The quotient space is a CW-space T(Z,m) and
hence a 1(Z,m. 1).
7. Aconnected 1-dimensional CW-complex X is a 1(. 1). Determine fromthe topology
of X.
8. A connected non-closed surface (with or without boundary) is a 1(. 1).
Chapter 9
Singular Homology
Homology is the most ingenious invention in algebraic topology. Classically, the
denition of homology groups was based on the combinatorial data of simplicial
complexes. This denition did not yield directly a topological invariant. The
denition of homology groups and (dually) cohomology groups has gone through
various stages and generalizations.
The construction of the so-called singular homology groups by Eilenberg [56]
was one of the denitive settings. This theory is very elegant and almost entirely
algebraic. Very little topology is used as an input. And yet the homology groups are
dened for arbitrary spaces in an invariant manner. But one has to pay a price: The
denition is in no way intuitively plausible. If one does not mind jumping into cold
water, then one may well start algebraic topology with singular homology. Also
interesting geometric applications are easily obtainable.
In learning about homology, one has to followthree lines of thinking at the same
time: (1) The construction. (2) Homological algebra. (3) Axiomatic treatment.
(1) The construction of singular homology groups and the verication of its main
properties, now called the axioms of Eilenberg and Steenrod.
(2) Acertain amount of algebra, designed for use in homology theory (but also of
independent algebraic interest). It deals with diagrams, exact sequences, and
chain complexes. Later more advanced topics are needed: Tensor products,
linear algebra of chain complexes, derived functors and all that.
(3) The object that one constructs with singular homology is now called a ho-
mology theory, dened by the axioms of Eilenberg and Steenrod. Almost all
applications of homology are derived fromthese axioms. The axiomatic treat-
ment has other advantages. Various other homology and cohomology theories
are known, either constructed by special input (bordism theories, K-theories,
de Rham cohomology) or in a systematic manner via stable homotopy and
spectra.
The axioms of a homology or cohomology theory are easily motivated from
the view-point of homotopy theory. But we should point out that many results of
algebraic topology need the idea of homology: The reduction to combinatorial data
via cell complexes, chain complexes, spectral sequences, homological algebra, etc.
Reading this chapter requires a parallel reading of the chapter on homological
algebra. Already in the rst section we use the terminology of chain complexes and
their homology groups and results about exact sequences of homology groups.
224 Chapter 9. Singular Homology
9.1 Singular Homology Groups
The n-dimensional standard simplex is
^
n
= ^n| =
(t
0
. . . . . t
n
) R
n1

n
i=0
t
i
= 1. t
i
_ 0
_
R
n1
.
We set n| = {0. . . . . n]. A weakly increasing map : m| n| induces an afne
map
^(): ^m| ^n|.
n
i=0
t
i
e
i

n
i=0
t
i
e
(i)
.
Here e
i
is the standard unit vector, thus

n
i=0
t
i
e
i
= (t
0
. . . . . t
n
). These maps
satisfy the rules of a functor ^( ) = ^() ^() and ^(id) = id. Let
n
i
: n 1| n| be the injective map which misses the value i .
(9.1.1) Note.
n1
}

n
i
=
n1
i

n
}-1
, i < . (The composition misses i and .) We
write J
n
i
= ^(
n
i
). By functoriality, the J
n
i
satisfy the analogous commutation
rules.
A continuous map o : ^
n
X is called a singular n-simplex in X. The i -th
face of o is o J
n
i
. We denote by S
n
(X) the free abelian group with basis the
set of singular n-simplices in X. (We also set, for formal reasons, S
n
(X) = 0
in the case that n < 0 but disregard mostly this trivial case. If X = 0, we let
S
n
(X) = 0.) An element . S
n
(X) is called a singular n-chain. We think of .
as a formal nite linear combination . =

c
n
c
o, n
c
Z. In practice, we skip
a summand with n
c
= 0; also we write 1 o = o. We use without further notice
the algebraic fact that a homomorphism from S
n
(X) is determined by its values on
the basis elements o : ^
n
X, and these values can be prescribed arbitrarily. The
boundary operator d
q
is dened for q _ 1 by
d
q
: S
q
(X) S
q-1
(X). o
q
i=0
(1)
i
oJ
q
i
.
and for q _ 0 as the zero map. Basic for everything that follows is the
9.1.2 Boundary relation. d
q-1
d
q
= 0.
Proof. We decompose the sum ddo =

q
}=0
q-1
i=0
(1)
i}
oJ
q
}
J
q-1
i
into the
parts

i~}
and

i_}
. When we rewrite the rst sum using (9.1.1), the result is
the negative of the second sum.
The singular chain groups S
q
(X) and the boundary operators d
q
form a chain
complex, called the singular chain complex S
-
(X) of X. Its n-th homology group
is denoted H
n
(X) = H
n
(X: Z) and called singular homology group of X (with
coefcients in Z). A continuous map : X Y induces a homomorphism
#
= S
q
( ): S
q
(X) S
q
(Y ). o o.
9.1. Singular Homology Groups 225
The family of the S
q
( ) is a chain map S
-
( ): S
-
(X) S
-
(Y ). Thus we have
induced homomorphisms
+
= H
q
( ): H
q
(X) H
q
(Y ). In this manner, the
H
q
become functors from TOP into the category ABEL of abelian groups.
Let now i : X be an inclusion. We dene S
n
(X. ) as the cokernel of
S
n
(i ): S
n
() S
n
(X). Less formally: The group S
n
(X. ) is free abelian and
has as a basis the singular simplices o : ^
n
X with image not contained in
. Since S
n
(0) = 0, we identify canonically S
n
(X) = S
n
(X. 0). The boundary
operator of S
-
(X) induces a boundary operator d
n
: S
n
(X. ) S
n-1
(X. ) such
that the family of quotient homomorphisms S
n
(X) S
n
(X. ) is a chain map.
The homology groups H
n
(X. ) = H
n
(X. : Z) of S
-
(X. ) are the relative
singular homology groups of the pair (X. ) (with coefcients in Z). Acontinuous
map : (X. ) (Y. T) induces a chain map
-
: S
-
(X. ) S
-
(Y. T) and
homomorphisms
+
= H
q
( ): H
q
(X. ) H
q
(Y. T). In this way, H
q
becomes
a functor from TOP(2) to ABEL.
We apply (11.3.2) to the exact sequence of singular chain complexes
0 S
-
() S
-
(X) S
-
(X. ) 0
and obtain the associated exact homology sequence:
(9.1.3) Theorem. For each pair (X. ) the sequence

J
H
n
() H
n
(X) H
n
(X. )
J
H
n-1
()
is exact. The sequence terminates with H
0
(X) H
0
(X. ) 0. The undeco-
rated arrows are induced by the inclusions (. 0) (X. 0) and (X. 0) (X. ).
Let (X. . T) be a triple, i.e., T X. The inclusion S

-
() S
-
(X)
induces by passage to factor groups an inclusion S
-
(. T) S
-
(X. T), and its
cokernel can be identied with S
-
(X. ). We apply (11.3.2) to the exact sequence
of chain complexes
0 S
-
(. T) S
-
(X. T) S
-
(X. ) 0
and obtain the exact sequence of a triple

J
H
n
(. T) H
n
(X. T) H
n
(X. )
J
H
n-1
(. T) .
The boundary operator d: H
n
(X. T) H
n-1
(. T) in the exact sequence of a
triple is the composition of the boundary operator for (X. ) followed by the map
H
n-1
() H
n-1
(. T) induced by the inclusion.
It remains to verify that the connecting morphisms d constitute a natural trans-
formation, i.e., that for each map between triples : (X. . T) (X
t
.
t
. T
t
) the
diagram
H
k
(X. )
J
H
k-1
(. T)
(
H
k
(X
t
.
t
)
J
H
k-1
(
t
. T
t
)
is commutative. This is a special case of an analogous fact for morphisms be-
tween short exact sequences of chain complexes and their associated connecting
morphisms. We leave this as an exercise.
One cannot determine the groups H
q
(X. ) just from its denition (except in a
fewtrivial cases). Note that for open sets in Euclidean spaces the chain groups have
an uncountable basis. So it is clear that the setup only serves theoretical purposes.
Before we prove the basic properties of the homology functors (the axioms of
Eilenberg and Steenrod) we collect a few results which follow directly from the
denitions.
9.1.4 Point. Let X = 1 be a point. There is a unique singular n-simplex, hence
S
n
(1) Z, n _ 0. The boundary operators d
0
and d
2i1
are zero and d
2
. d
4
. . . .
are isomorphisms. Hence H
i
(1) = 0 for i ,= 0; and H
0
(1) Z, via the
homomorphism which sends the unique 0-simplex to 1 Z.
9.1.5 Additivity. Let (X
}
[ J) be the path components of X, and let
|
}
: (X
}
. X
}
) (X. ) be the inclusion. Then
}J
S
n
(X
}
. X
}
) S
n
(X. ). (.
}
)
}
|
}
#
(.
}
)
is an isomorphism. Similarly for H
n
instead of S
n
. The reason is that ^
n
is path
connected, and therefore o : ^
n
X has an image in one of the X
}
, so we can
sort the basis elements of S
n
(X) according to the components X
}
.
9.1.6 The groups H
0
. The group H
0
(X) is canonically isomorphic to the free
abelian group Z
0
(X) over the set
0
(X) of path components. We identify a
singular 0-simplex o : ^
0
X with the point o(^
0
). Then S
0
(X) is the free
abelian group on the points of X. A singular 1-simplex o : ^
1
X is essentially
the same thing as a path, only the domain of denition has been changed from 1
to ^
1
. We associate to o the path n
c
: 1 X, t o(1t. t ). Then d
0
o = n
c
(1)
and d
1
o = n
c
(0), hence do = d
0
od
1
o corresponds to the orientation convention
dn = n(1) n(0). If two points a. b X are in the same path component, then
the zero-simplices a and b are homologous. Hence we obtain a homomorphism
from Z
0
(X) into H
0
(X), if we assign to the path component of a its homology
class. We also have a homomorphismS
0
(X) Z
0
(X) which sends the singular
simplex of a X to the path component of a. This homomorphism sends the
image of d: S
1
(X) S
0
(X) to zero. Hence we obtain an inverse homomorphism
H
0
(X) Z
0
(X).
9.2. The Fundamental Group 227
9.2 The Fundamental Group
The signs which appear in the denition of the boundary operator have an inter-
pretation in low dimensions. They are a consequence of orientation conventions.
A singular 1-simplex o : ^
1
X is essentially the same thing as a path, only the
domain of denition has changed from 0. 1| to ^
1
. We associate to o the path
1 X, t o(1 t. t ). The inverse path is then o
-
(t
0
. t
1
) = o(t
1
. t
0
). The
product of paths has now the form
(o + t)(t
0
. t
1
) =
_
o(2t
0
1. 2t
1
). t
1
_ 1,2.
t(2t
0
. 2t
1
1). t
1
_ 1,2.
If we dene o: ^
2
X, (t
0
. t
1
. t
2
) (o + t)(t
0
t
1
,2. t
1
,2 t
2
), then one
veries do = o o + t t. A loop o : ^
1
X is a 1-cycle; let o| be its
homology class. Thus for loops o. t we have
(1) o + t| = o| t|.
(Here :| denotes the homology class of the cycle :.) Let k : ^
1
1 X
be a homotopy of paths ^
1
X. The map k factors over the quotient map
q : ^
1
1 ^
2
, (t
0
. t
1
. t ) (t
0
. t
1
(1 t ). t
1
t ) and yields o : ^
2
X. We
compute do = c k
1
k
0
, with a constant c. A constant 1-simplex is a boundary.
Hence k
0
k
1
is a boundary
k
0
| = k
1
| C
1
(X),T
1
(X).
In particular, homotopic loops yield the same element in H
1
(X). Thus we obtain
a well-dened map h
t
:
1
(X. .
0
) H
1
(X); by (1), it is a homomorphism. The
fundamental group is in general non-abelian. Therefore we modify h
t
algebraically
to take this fact into account. Each group G has the associated abelianized factor
group G
ob
= G,G. G|; the commutator group G. G| is the normal subgroup
generated by all commutators .,.
-1
,
-1
. A homomorphismG to an abelian
group factorizes uniquely over G
ob
. We apply this denition to h
t
and obtain a
homomorphism
h:
1
(X. .
0
)
ob
H
1
(X).
(9.2.1) Theorem. Let X be path connected. Then h is an isomorphism.
Proof. We construct a homomorphisminthe other direction. For . X we choose a
path u(.) from.
0
to .. We assign to a 1-simplex o : ^
1
X fromo
0
= o(1. 0) to
o
1
= o(0. 1) the class of the loop (u(o
0
) +o) +u(o
1
)
-
. We extend this assignment
linearly to a homomorphism l
t
: C
1
(X)
1
(X. .
0
)
ob
. Let t : ^
2
X be a
2-simplex with faces t
}
= tJ
}
. Since ^
2
is contractible, t
2
+ t
0
. t
1
. This
implies
l
t
(t
2
|) l
t
(t
0
|) = l
t
(t
2
| t
0
|) = l
t
(t
2
+ t
0
|) = l
t
(t
1
|).
Hence l
t
factors over C
1
(X),T
1
(X) and induces l : H
1
(X)
1
(X. .
0
)
ob
. By
construction, lh = id. We show that h is surjective. Let

a
c
o C
1
(X) be a
cycle. Then
a
c
o| =
a
c
(u(o
0
)| o| u(o
1
)|) =
a
c
(u(o
0
) + o) + u(o
1
)
-
|.
and the last element is contained in the image of h.
One of the rst applications of the homology axioms is the computation
H
1
(S
1
) Z. Granted the formal result that
1
(S
1
) is abelian, we obtain yet
another proof for
1
(S
1
) Z.
9.3 Homotopy
We prove in this section the homotopy invariance of the singular homology groups.
We begin with a special case.
9.3.1 Cone construction. Let X be a contractible space. Dene a chain map
c = (c
n
): S
-
(X) S
-
(X) by c
n
= 0 for n ,= 0 and by c
0
_
n
c
o
_
=
_
n
c
_
o
0
where o
0
: ^
0
{.
0
]. We associate to each homotopy h: X 1 X from the
identity to the constant map with value .
0
a chain homotopy s = (s
n
) from c to
the identity. The homomorphisms s : S
n-1
(X) S
n
(X) are obtained from a cone
construction. Let
q : ^
n-1
1 ^
n
. ((j
0
. . . . . j
n-1
). t ) (t. (1 t )j
0
. . . . . (1 t )j
n-1
).
Given o : ^
n-1
X, there exists a unique simplex s(o) = so : ^
n
X such that
h (o id) = s(o) q, since q is a quotient map. For the faces of so we verify
(so)J
i
= s(oJ
i-1
), for i > 0, and (so)J
0
= o. From these data we compute for
n > 1,
d(so) = (so)J
0
n
1
(1)
i-1
(so)J
i
= o
n-1
0
(1)
}
s(oJ
}-1
)
= o s(do)
and d(so) = o o
0
for a 0-simplex o. These relations imply ds sd = id c.
Note that c induces the zero map in dimensions n ,= 0.
(9.3.2) Proposition. Let X be contractible. Then H
n
(X) = 0 for n ,= 0.
The inclusions j
t
(X) = j
t
: X X 1, . (.. t ) induce chain maps
(j
t
n
) = j
t
-
: S
-
(X) S
-
(X 1). We consider these chain maps as natural
transformations between functors; the naturality says that for each continuous map
: X Y the commutation relation ( id)
-
j
t
(X)
-
= j
t
(Y )
-
-
holds.
(9.3.3) Theorem. There exists a natural chain homotopy s
-
from j
0
-
to j
1
-
.
9.3. Homotopy 229
Proof. We apply (11.5.1) to C = TOP, J
+
(X) = S
-
(X), G
+
(X) = S
-
(X 1)
and the natural transformation j
t
-
. The model set for J
n
consists of ^
n
, and the
corresponding b-element is the identity of ^
n
considered as a singular simplex.
From (9.3.2) we see that G
+
is acyclic. It should be clear that j
0
-
and j
1
-
induce the
same transformations in H
0
.
For the convenience of the reader we also rewrite the foregoing abstract proof in
explicit terms. See also Problem2 for an explicit chain homotopy and its geometric
meaning.
Proof. We have to show: There exist morphisms s
A
n
: S
n
(X) S
n1
(X1) such
that
(1
n
) ds
A
n
s
A
n-1
d = j
1
n
(X) j
0
n
(X)
(chain homotopy), and such that for continuous X Y the relations
(N
n
) ( id)
#
s
A
n
= s
Y
n

#
hold (naturality). We construct the s
n
inductively.
n = 0. In this case, s
0
sends the 0-simplex o : ^
0
{.] X to the 1-simplex
s
0
o : ^
1
X 1, (t
0
. t
1
) (.. t
1
). Then the computation
d(s
0
o) = (s
0
o)J
0
(s
0
o)J
1
= j
1
0
o j
0
0
o
shows that (1
0
) holds, and also (N
0
) is a direct consequence of the denitions.
Nowsuppose that the s
k
for k < nare given, and that they satisfy (1
k
) and (N
k
).
The identity of ^
n
is a singular n-simplex; let |
n
S
n
(^
n
) be the corresponding
element. The chain to be constructed s
n
|
n
should satisfy
d(s
n
|
n
) = j
1
n
(|
n
) j
0
n
(|
n
) s
n-1
d(|
n
).
The right-hand side is a cycle in S
n
(^
n
1), as the next computation shows.
d(j
1
n
(|
n
) j
0
n
(|
n
) s
n-1
d(|
n
))
= j
1
n-1
(d|
n
) j
0
n-1
(d|
n
) ds
n-1
(d|
n
)
= j
1
n-1
(d|
n
) j
0
n-1
(d|
n
) (j
1
n-1
(d|
n
) j
0
n-1
(d|
n
) s
n-2
dd|
n
) = 0.
We have used the relation (1
n-1
) for ds
n-1
d(|
n
) and that the j
t
-
are chain maps.
Since ^
n
1 is contractible, there exists, by (9.3.2), an a S
n1
(^
n
1) with the
property da = j
1
n
(|
n
) j
0
n
(|
n
) s
n-1
d(|
n
). We choose an a with this property and
dene s
n
(|
n
) = a and in general s
n
(o) = (o id)
#
a for o : ^
n
X; the required
naturality (N
n
) forces us to do so. We now verify (1
n
) and (N
n
). We compute
ds
n
(o) = d(o id)
#
a = (o id)
#
da
= (o id)
#
(j
1
n
|
n
j
0
n
|
n
s
n-1
d|
n
)
= j
1
n
o
#
|
n
j
0
n
o
#
|
n
s
n-1
o
#
d|
n
= j
1
n
o j
0
n
o s
n-1
do.
We have used: (o id)
#
is a chain map; choice of a; naturality of j
1
-
, j
0
-
, and
(N
n-1
); o
#
|
n
= o; o
#
is a chain map. Thus we have shown (1
n
). The equalities
( id)
#
s
n
(o) = ( id)
#
(o id)
#
a = (o id)
#
a = s
n
(o) = s
n
#
o
nally show the naturality (N
n
).
With (9.3.3) we control the universal situation. Let : (X. ) 1 (Y. T)
be a homotopy in TOP(2) from
0
to
1
. The s
n
in (9.3.3) induce by naturality
also a chain homotopy s
n
: S
n
(X. ) S
n1
(X 1. 1). The computation
d(
#
s
n
) (
#
s
n-1
)d =
#
ds
n
#
s
n-1
d =
#
(j
1
j
0
) =
1
#

0
#
proves the
#
s
n
to be a chain homotopy from
0
#
to
1
#
. Altogether we see:
(9.3.4) Theorem. Homotopic maps induce homotopic chain maps and hence the
same homomorphisms between the homology groups.
(9.3.5) Example. Let a
0
(X). a
1
(X): S
-
(X) S
-
(X) be chain maps, natural in
X, which coincide on S
0
(X). Then there exists a natural chain homotopy from a
0
to a
1
. This is a consequence of (11.5.1) for J
+
= G
+
and the models ^
n
as in the
proof of (9.3.3).
Problems
1. Let t
n
: ^
n
^
n
, (z
0
. . . . . z
n
) (z
n
. . . . . z
0
). Verify that S
n
(X) S
n
(X),
o (1)
(n1)n/2
ot
n
is a natural chain map. By (9.3.5), it is naturally homotopic to the
identity.
2. One can prove the homotopy invariance by constructing an explicit chain homotopy.
A natural construction would associate to a singular n-simplex o : ^
n
X the singular
prism o id: ^
n
1 X 1. The combinatorial (set-theoretic) boundary of ^
n
1
is ^
n
1 L ^
n
0 L (d^
n
) 1, and this corresponds exactly to the denition of a chain
homotopy, if one takes orientations into account. This idea works; one has to decompose
^
n
1 into simplices, and it sufces to do this algebraically.
In the prism^
n
1 let 0. 1. . . . . n denote the vertices of the base and 0
/
. 1
/
. . . . . n
/
those
of the top. In the notation for afne singular simplices introduced later, show that an explicit
formula for a = s
n
|
n
is
s
n
|
n
=
n
i=0
(1)
i
0. 1. . . . . i. i
/
. (i 1)
/
. . . . . n
/
|.
(This is a special case of the EilenbergMac Lane shufe morphism to be discussed in the
section on homology products.)
9.4. Barycentric Subdivision. Excision 231
9.4 Barycentric Subdivision. Excision
The basic property of homology is the excision theorem (9.4.7). It is this theo-
rem which allows for effective computations. Its proof is based on subdivision of
standard simplices. We have to work out the algebraic formof this subdivision rst.
Let D R
n
be convex, and let
0
. . . . .
]
be elements in D. The afne singular
simplex o : ^
]
D,
i
z
i
e
i

i
z
i
i
will be denoted o =
0
. . . . .
]
|. With
this notation
d
0
. . . . .
]
| =
]
i=0
(1)
i
0
. . . . .
i
. . . . .
]
|.
where
i
means that
i
has to be omitted from the string of vertices. For each
D we have the contracting homotopy D1 D, (.. t ) (1 t ). t . If
we apply the cone construction 9.3.1 to
0
. . . . .
]
| we obtain .
0
. . . . .
]
|. We
denote the chain homotopy associated to the contraction by S
]
(D) S
]1
(D),
c c. We have for c S
]
(D):
(1) d( c) =
_
c dc. > 0.
c c(c). = 0.
with c: S
0
(D) Z,
n
c
o
n
c
.
The barycenter of o =
0
. . . . .
]
| is o
=
1
]1
]
i=0
i
. We dene induc-
tively
B
]
(X) = B
]
: S
]
(X) S
]
(X)
to be the homomorphism which sends o : ^
]
X to B
]
(o) = o
#
B
]
(|
]
), where
B
]
(|
]
) is dened inductively as
(2) B
]
(|
]
) =
_
|
0
. = 0.
|
]
B
]-1
(d|
]
). > 0.
(9.4.1) Proposition. The B
]
constitute a natural chain map which is naturally
homotopic to the identity.
Proof. The equalities
#
Bo =
#
o
#
B(|
]
) = (o)
#
B(|
]
) = B(o) = B
#
o
prove the naturality. We verify by induction over that we have a chain map. Let
= 1. Then dB(|
1
) = d(|
1
B(d|
1
)) = d|
1
= Bd(|
1
). For > 1 we compute
dB|
]
= d(|
]
B(d|
]
)) = Bd|
]
|
]
dBd|
]
= Bd|
]
|
]
Bdd|
]
= Bd|
]
.
We have used: Denition; (1); inductive assumption; dd = 0. We now use this
special case and the naturality
Bdo = Bdo
#
|
]
= Bo
#
d|
]
= o
#
Bd|
]
= o
#
dB|
]
= do
#
B|
]
= dBo.
and this computation covers the general case.
The chain map B is naturally homotopic to the identity (see (9.3.5)).
Let U be a family of subsets of X such that their interiors cover X. We call
a singular simplex U-small, if its image is contained in some member of U. The
subgroup spanned by the U-small simplices is a subcomplex S
U
-
(X) of S
-
(X) with
homology groups denoted by H
U
n
(X).
(9.4.2) Lemma. The diameter J(
0
. . . . .
]
) of the afne simplex
0
. . . . .
]
| with
respect to the Euclidean norm is the maximum of the [
i

}
[.
Proof. Let .. ,
0
. . . . .
]
| and . =
z
}
}
. Then, because of
z
}
= 1,
[. ,[ = [
z
}
(
}
,)[ _
z
}
[
}
,[ _ max
}
[
}
,[.
This shows in particular [,
i
[ _ max
}
[
}

i
[; we insert this in the above
and obtain [. ,[ _ max
i,}
[
i

}
[; hence the diameter is at most as stated.
On the other hand, this value is clearly attained as the distance between two points.
(9.4.3) Lemma. Let

0
. . . . .
]
R
n
. Then B
]
0
. . . . .
]
| is a linear combination
of afne simplices with diameter at most
]
]1
J(
0
. . . . .
]
).
Proof. From the inductive denition (2) and the naturality of B we conclude
(3) B
0
. . . . .
]
| =
]
}=0
(1)
}
o
B
0
. . . . .
}
. . . . .
]
|
where o =
0
. . . . .
]
|.
We prove the claim by induction over . The assertion is obvious for = 0,
a point has diameter zero. By induction hypothesis, the simplices in the chain
B
0
. . . . .
}
. . . . .
]
| are afne of diameter at most
]-1
]
J(
0
. . . . .
}
. . . . .
]
) _
]-1
]
J(
0
. . . . .
]
). The simplices in B
0
. . . . .
]
| have vertices o
and vertices
from simplices in B
0
. . . . .
}
. . . . .
]
|. It sufces to evaluate the distance of o
from such vertices. It is less than or equal to sup([o
.[ [ .
0
. . . . .
]
|).
Let . =

z
}
}
. Then [o
.[ _ max [o

}
[, as in the proof of (9.4.2).
Moreover we have
[o

}
[ =
_
_
1
]1
_
i

i
_

}
_
_
_
1
]1
i
[
i

}
[
_
]
]1
max
i,}
[
i

}
[ =
]
]1
J(
0
. . . . .
]
).
Since ( 1), < ,( 1) we have veried, altogether, the claim.
(9.4.4) Lemma. Let o : ^
]
X be a singular simplex. Then there exists a k N
such that each simplex in the chain B
k
o has an image contained in a member of U.
(Here B
k
is the k-fold iteration of B.)
9.4. Barycentric Subdivision. Excision 233
Proof. We consider the open covering (o
-1
(U
)). U U of ^
]
. Let c > 0 be a
Lebesgue number of this covering. The simplices of B
k
o arise by an application of
o to the simplices in B
k
|
]
. From (9.4.3) we see that the diameter of these simplices
is at most (
]
]1
)
k
J(e
0
. . . . . e
]
). If k is large enough, this number is smaller than c.
(9.4.5) Theorem. The inclusion of chain complexes S

U
-
(X) S
-
(X) induces an
isomorphism H
U
+
(X) H
+
(X).
Proof. Let a S
U
n
(X) be a cycle which represents a homology class in the kernel.
Thus a = db with some b S
n1
(X). By (9.4.4), there exists k such that
B
k
(b) S
U
n1
(X) (apply (9.4.4) to the nite number of simplices in the linear
combination of b). By (9.4.1), there exists a natural chain homotopy T
k
between
B
k
and the identity. Therefore
B
k
(b) b = T
k
(db) dT
k
(b) = T
k
(a) dT
k
(b).
and we conclude
dB
k
(b) db = dT
k
(a). a = db = d(B
k
(b) T
k
(a)).
From the naturality of T
k
and the inclusion a S
U
n
(X) we see T
k
(a) S
U
n1
(X).
Therefore a is a boundary in S
U
-
(X). This shows the injectivity of the map in
question.
Let a S
n
(X) be a cycle. By (9.4.4), there exists k such that B
k
a S
U
n
(X).
We know that
B
k
a a = T
k
(da) dT
k
(a) = dT
k
(a).
Since B
k
is a chain map, B
k
a is a cycle. From the last equality we see that a
is homologous to a cycle in S
U
n
(X). This shows the surjectivity of the map in
question.
Let now (X. ) be a pair of spaces. We write U = (U [ U U) and
dene the chain complex S
U
-
(X. ) = S
U
-
(X),S
U
-
() with homology groups
H
U
+
(X. ). We obtain a commutative diagramof chain complexes with exact rows:
0

S
U
-
()
S
U
-

S
U
-
(X. )

0
0

S
-
()

S
-
(X)

S
-
(X. )

0.
Each row has its long exact homology sequence. We apply (9.4.5) to (X. U) and
(. U ), use the Five Lemma (11.2.7), and obtain:
(9.4.6) Theorem. The inclusion of chain complexes | : S
U
-
(X. ) S(X. ) in-
duces an isomorphism H
U
+
(X. ) H
+
(X. ). By an application of (11.6.3) we
see that the inclusion | is actually a chain equivalence.
(9.4.7) Theorem (Excision Theorem). Let Y = Y
1
L Y
2
. Then the inclusion
induces an isomorphism H
+
(Y
2
. Y
1
Y
2
) H
+
(Y. Y
1
). Let T X and
suppose that

T
. Then the inclusion (X T. T) (X. ) induces an

isomorphism H
+
(X T. T) H
+
(X. ). Again we can invoke (11.6.3) and
conclude that the inclusion actually induces chain equivalences between the chain
complexes under consideration.
Proof. The covering U = (Y
1
. Y
2
) satises the hypothesis of (9.4.5). By denition,
we have S
U
n
(X) = S
n
(Y
1
) S
n
(Y
2
) and also S
n
(Y
1
Y
2
) = S
n
(Y
1
) S
n
(Y
2
).
The inclusion S
-
(Y
2
) S
-
(Y ) induces therefore, by an isomorphism theorem of
elementary algebra,
S
n
(Y
2
)
S
n
(Y
1
Y
2
)
=
S
n
(Y
2
)
S
n
(Y
1
) S
n
(Y
2
)
S
n
(Y
1
) S
n
(Y
2
)
S
n
(Y
1
)
=
S
U
n
(Y )
S
n
(Y
1
)
.
By (9.4.5) and (11.2.7) we see, rstly, that S
U
-
(Y ),S
-
(Y
1
) S
-
(Y ),S
-
(Y
1
) and,
altogether, that S
-
(Y
2
),S
-
(Y
1
Y
2
) S
-
(Y ),S
-
(Y
1
) induces an isomorphism in
homology. The second statement is equivalent to the rst; we use X = Y , = Y
1
,
X T = Y
2
.
Problems
1. Let D R
m
and 1 R
n
be convex and let : D 1 be the restriction of a linear
map. Then
#
( c) = ()
#
(c).
2. Although not necessary for further investigations, it might be interesting to describe the
chain B
0
. . . . .
)
| in detail. We use (3) in the proof of (9.4.3). By (2), formula (3) also
holds for
0
. . . . .
)
|. This yields B
0
.
1
| =
01
.
1
|
01
.
0
| with barycenter
01
,
and for B
0
.
1
.
2
| we obtain in short-hand notation what is illustrated by the next gure.
012. 12. 2| 012. 12. 1| 012. 02. 2| 012. 02. 0| 012. 01. 1| 012. 01. 0|.

d
d
d
d
d
d
d
d

0

1
2
1
2
1
2
2
1
3
1
3
1
3
012. 12. 2|
One continues inductively in this manner. Let S( 1) denote the permutation group
of {0. . . . . ]. We associate to o =
0
. . . . .
)
| and S( 1) the simplex o
0
. . . . .
)
|, where
=
()
. . . . .
())
|
. With this notation the following holds:

Bo =
c()1)
sign()o
.
9.5. Weak Equivalences and Homology 235
9.5 Weak Equivalences and Homology
Although singular homology groups are dened for arbitrary topological spaces,
they only capture combinatorial information. The theory is determined by its values
on cell complexes. Technically this uses two facts: (1) a weak homotopy equiv-
alence induces isomorphisms of homology groups; (2) every topological space is
weakly equivalent to a CW-complex. One can use cell complexes to give proofs
by induction over the skeleta. Usually the situation for a single cell is quite trans-
parent, and this fact makes the inductive proofs easy to follow and to remember.
Once a theorem is known for cell complexes, it can formally be extended to general
topological spaces. We now prove this invariance property of singular homology
[56], [21].
Let (X. . +) be a pointed pair. Let ^k|
n
be the n-skeleton of the standard
simplicial complex^k| (this is the reasonfor switchingthe notationfor the standard
k-simplex). Let S
(n,)
k
(X) for n _ 0 denote the subgroup of S
k
(X) spanned by
the singular simplices o : ^k| X with the property
(#) o(^k|
n
) .
The groups (S
(n,)
k
(X) [ k _ 0) form the Eilenberg subcomplex S
(n,)
-
(X) of
S
-
(X).
(9.5.1) Theorem. Let (X. ) be n-connected. Then the inclusion of the Eilenberg
subcomplex : S
(n,)
-
(X) S
-
(X) is a chain equivalence.
Proof. We assign to a simplex o : ^k| X a homotopy 1(o): ^k| 1 X
such that
(1) 1(o)
0
= o,
(2) 1(o)
1
satises (#),
(3) 1(o)
t
= o, provided o satises already (#),
(4) 1(o) (J
k
i
id) = 1(o J
k
i
).
According to (3), the assignment 1 is dened for simplices which satisfy (#). For
the remaining simplices we use an inductive construction.
Suppose k = 0. Then o(^0|) X is a point. Since (X. ) is 0-connected,
there exists a path from this point to a point in . We choose a path of this type
as 1(o).
Suppose 1 is given for -simplices, < k. Then for each k-simplex o the
homotopy 1(o J
k
i
) is already dened, and the 1(o J
k
i
) combine to a homotopy
d^k| 1 X. Moreover 1(o)
0
is given. Altogether we obtain
1(o): (^k| 0 L d^k| 1. d^k| 1) (X. ).

Let k _ n. Then ^k|
n
= ^k|, and similarly for the faces. By the inductive
assumption,

1(o) sends d^k| 1 into .
There exists a homeomorphismk : ^k| 1 ^k| 1 which induces home-
omorphisms (see (2.3.6))
^k| 0 ^k| 0 L d^k| 1.
d^k| 0 d^k| 1.
d^k| 1 L ^k| 1 ^k| 1.
Since (X. ) is k-connected, the map
1(o) k : (^k| 0. d^k| 0) (X. )

can be extended to a homotopy Q: ^k| 1 X which is constant on d^k| 1
and sends ^k| 1 into . We now set 1(o) = Q k
-1
. Then 1(o) extends
1(o), hence (1) and (4) are satised, and (2) also holds by construction.
Let k > n. We use the cobration (^k|. d^k|) in order to extend

1(o) to
1(o). Since ^k|
n
d^k|, we see that 1(o)
1
satises (#).
We now dene j: S
k
(X) S
(n,)
k
(X) by o 1(o)
1
. Property (4) shows
that j is a chain map, and j = id holds by construction. We dene s : S
k
(X)
S
k1
(X) by s(o) = 1(o)
#
h(|
k
)
|
k
S
k
(^k|)
h
S
k1
(^k| 1)
1(c)
#
S
k1
(X) s(o)
where h is the natural chain homotopy between i
0
#
and i
1
#
, see (9.3.3). The compu-
tations
ds(o) = d(1(o)
#
h(|
k
)) = 1(o)
#
dh(|
k
)
= 1(o)
1#
(|
k
) 1(o)
0#
(|
k
) 1(o)
#
h(d|
k
)
= j(o) o 1(o)
#
h(d|
k
).
sd(o) = s
_
(1)
i
o J
k
i
_
=
(1)
i
1(o J
k
i
)
#
h(|
k-1
)
= 1(o)
#
_
(1)
i
J
k
i#
h(|
k-1
)
_
= 1(o)
#
h(d|
k
)
show that s is a chain homotopy between j and id.
For k _ n we have ^k|
n
= ^k| and therefore S
(n,)
k
(X) = S
k
(). The
chain equivalence (9.5.1) and the exact homology sequence of (X. ) now yield:
(9.5.2) Theorem. Let (X. ) be n-connected. Then H
k
()

H
k
(X) and
H
k
(X. ) = 0 for k _ n.
Let : X Y be a weak homotopy equivalence. We can assume that is an
inclusion (mapping cylinder and homotopy invariance).
9.6. Homology with Coefcients 237
(9.5.3) Theorem. A weak homotopy equivalence induces isomorphisms of the sin-
gular homology groups.
(9.5.4) Remark. Suppose that (X. . +) is a pointed pair and is pathwise con-
nected. Then we can dene a subcomplex S
(A,,+)
-
(X) of S
-
(X) where we require
in addition to (#) that o(^k|
0
) = {+]. Again the inclusion is a chain equivalence.
9.6 Homology with Coefcients

Let C
-
= (C
n
. c
n
) be a chain complex of abelian groups and let G be a further
abelian group. Then the groups C
n
G and the boundary operators c
n
id form
again a chain complex (the tensor product is taken over Z). We denote it by C
-
G.
We apply this process to the singular complex S
-
(X. ) and obtain the complex
S
-
(X. ) G of singular chains with coefcients in G. Its homology group
in dimension n is denoted H
n
(X. : G). The cases G = Z. Q. Z, are often
referred to as integral, rational, mod.p/ homology. Chains in S
n
(X. ) G can
be written as nite formal linear combinations

c
a
c
o, a
c
G of singular n-
simplices o; this accounts for the name chain with coefcients. The sequence
0 S
-
() S
-
(X) S
-
(X. ) 0 remains exact when tensored with G,
i.e., S
n
(X. ) G S
n
(X) G,S
n
() G. Therefore we still have the exact
homology sequence (11.3.2)
H
n
(: G) H
n
(X: G) H
n
(X. : G)
J
H
n-1
(: G)
and the analogous sequence for triples. The boundary operators d are again natural
transformations. If 0 G
t
G G
tt
G is an exact sequence of abelian
groups, then the tensor product with S
-
(X. ) yields again an exact sequence of
chain complexes and we obtain from (11.3.2) an exact sequence of the form
H
n
(X. : G
t
) H
n
(X. : G) H
n
(X. : G
tt
) H
n-1
(X. : G
t
) .
The passage fromC
-
toC
-
Gis compatible withchainmaps andchainhomotopies.
A chain equivalence induces a chain equivalence. This fact yields the homotopy
invariance of the homology groups H
n
(X. : G). The excision theorem still holds.
This is a consequence of (9.4.7): Under the hypothesis of the excision theorem, the
chain equivalence S
-
(Y
1
. Y
1
Y
2
) S
-
(Y. Y
2
) induces a chain equivalence when
tensored with G. Hence the functors H
n
(X. : G) satisfy the axioms of Eilenberg
and Steenrod for a homology theory. The dimension axiom holds: We have a
canonical isomorphism c
1
: H
0
(1) G for a point 1, which maps the homology
class of the chain ao to a, where o is the unique 0-simplex.
The application of (11.9.1) to topology uses the fact that the singular chain
complex consists of free abelian groups. Therefore we obtain:
(9.6.1) Theorem (Universal Coefcients). Let 1 be a principal ideal domain and
G an 1-module. There exists an exact sequence
0 H
n
(X. : 1)
T
G

H
n
(X. : G) Tor(H
n-1
(X. : 1). G) 0.
The sequence is natural in (X. ) and G. The sequence splits, the splitting is natural
in G, but not in (X. ).
The splitting statement means that H
n
(X. : G) can be determined as an abelian
group from homology with coefcients in Z, but the functor H
n
(: G) is not
the direct sum of the functors H
n
(: Z) G and Tor(H
n-1
(: Z). G). Here
is a consequence of (9.6.1): If : (X. ) (Y. T) induces an isomorphism
+
: H
+
(X. ) H
+
(Y. T) for + = n 1. n, then it induces also an isomorphism
H
n
(X. : G) H
n
(Y. T: G).
9.7 The Theorem of Eilenberg and Zilber
We study the homology of products. For this purpose we compare the chain com-
plexes S
-
(X) S
-
(Y ) and S
-
(X Y ). Both are values at (X. Y ) of a functor
TOP TOP CH
into the category of chain complexes which are zero in nega-

tive degrees. In dimension zero they essentially coincide. For . X let . S
0
(X)
also denote the basis element given by the singular simplex ^
0
{.] X.
Then S
-
(X Y ) has the basis (.. ,) and S
-
(X) S
-
(Y ) the basis . , for
(.. ,) X Y . Natural transformations 1 : S
-
() S
-
() S
-
( ) and
Q: S
-
() S
-
() S
-
() are called an EilenbergZilber morphisms if in
dimension zero always 1(. ,) = (.. ,) and Q(.. ,) = . ,. Both functors
are free and acyclic in the sense of (11.5.1). For (S
-
() S
-
())
n
we use the
models (^
k
. ^
n-k
) and the elements id id; for S
n
( ) we use the models
(^
n
. ^
n
) and the diagonal maps ^
n
^
n
^
n
. They account for the freeness.
The homology of the chain complexes S
-
(^
]
^
q
) is zero in positive dimensions,
since ^
]
^
q
is contractible; the homology of S
-
(^
]
)S
-
(^
q
) is zero in positive
dimensions, since the tensor product of chain complexes is compatible with chain
homotopies, and the chain complexes S
-
(^
]
) are homotopy equivalent to the triv-
ial complex. Similar statements hold for the analogous functors in three (or more)
variables like S
-
(X Y 7) or the corresponding three-fold tensor products. As
an application of (11.5.1) we obtain:
(9.7.1) Theorem. (1) EilenbergZilber morphisms 1 and Q exist. For each pair
(1. Q) of EilenbergZilber morphisms the compositions 1 Q and Q1 are nat-
urally homotopic to the identity. Hence the 1
A,Y
and Q
A,Y
are chain equivalences
and any two EilenbergZilber morphisms 1. 1
t
are naturally homotopic (similarly
for Q. Q
t
).
(2) An EilenbergZilber morphism 1 is associative and commutative up to
natural homotopy, i.e., the natural transformations 1
AY,7
(1
A,Y
1) and
9.7. The Theorem of Eilenberg and Zilber 239
1
A,Y 7
(1 1
Y,7
) from S
-
(X) S
-
(Y ) S
-
(7) to S
-
(X Y X) are
naturally homotopic and the transformations (t
A,Y
)
#
1
A,Y
and 1
Y,A
t
A,Y
are
naturally homotopic. Here t
A,Y
: X Y Y X interchanges the factors and
t
A,Y
(. ,) = (1)
[x[[,[
, ..
(3) An EilenbergZilber morphism Q is coassociative and cocommutative up
to natural homotopy, i.e., the natural transformations (Q
A,Y
1) Q
AY,7
and
(1Q
Y,7
)Q
A,Y 7
are naturally homotopic, and the transformations t
A,Y
Q
A,Y
and Q
Y,A
(t
A,Y
)
#
are naturally homotopic.
As a consequence one can determine the homology of X Y from the chain
complex S
-
(X) S
-
(Y ). We now turn to relative chain complexes and abbreviate
S = S
-
.
(9.7.2) Proposition. For EilenbergZilber transformations 1. Q and pairs of
spaces (X. ). (Y. T) we have a commutative diagram with short exact rows
S() S(Y ) S(X) S(T)

1
/
S(X) S(Y )
1
S(X. ) S(Y. T)
1
//
S( Y ) S(X T)

Q
/
S(X Y )

Q
S(X Y )
S( Y ) S(X T)
.
Q
//
The vertical maps are induced by 1 and Q. The compositions 1

t
Q
t
, Q
t
1
t
, 1
tt
Q
tt
,
Q
tt
1
tt
are naturally homotopic to the identity.
Proof. The naturality of 1 shows 1(S() S(Y )) S( Y ) and similarly
for Q. This shows that 1. Q induce by restriction 1
t
. Q
t
, and 1
tt
. Q
tt
are the
homomorphisms induced on the quotients. Since the homotopy 1Q . id is natural,
it maps S( Y ) S(X T) into itself and shows 1
t
Q
t
. id.
9.7.3 We can compose
1 : S(X. ) S(Y. T) S(X Y ),(S( Y ) S(X T))
with the map induced by the inclusion S(Y ) S(XT) S(Y LXT)
and obtain altogether natural chain maps
1 : S(X. ) S(Y. T) S((X. ) (Y. T)).
We call the pair ( Y. X T) excisive, if this chain map is a chain equivalence.

9.7.4 The natural chain map 1 induces natural chain maps for singular chain
complexes with coefcients. Let 1 be a commutative ring and M. N 1-modules.
S(X. : M) S(Y. T: N) = (S(X. ) M)
T
(S(Y. T) N)
(S(X. ) S(Y. T)) (M
T
N)
(S((X. ) (Y. T))) (M
T
N).
In many cases this chain map is followed by a homomorphism induced by a linear
map M
T
N 1. Examples are 1
T
1 1, . , ., in the case of a
ring 1 and 1
T
N N, . n .n in the case of an 1-module N.
Problems
1. There exist explicit EilenbergZilber morphisms whichhave further properties. Let . q
N. We use the notation n| = {0. 1. . . . . n]. A.p; q/-shufe is a map z: q| | q|
with z(0) = (0. 0) and z( q) = (. q) such that both components of z = (z
1
. z
2
) are
(weakly) increasing. Given z, there exists a permutation (j. v) = (j
1
. . . . . j
)
. v
1
. . . . . v
q
)
of 1. 2. . . . . q such that
1 _ j
1
< < j
)
_ q. 1 _ v
1
. . . . . v
q
_ q
and z
1
(j
) > z
1
(j
1) and z
2
(v
k
) > z
2
(v
k
1). We denote the signum of the
permutation (j. v) by c(z). If we interpret the points z(0). . . . . z( q) in the integral
lattice | q| as the vertices of an edge-path from (0. 0) to (. q), then the step z( )
z( 1) is horizontal or vertical of length 1. In the convex set ^
)
^
q
we have the afne
( q)-simplex z(0). . . . . z( q)|, also denoted z, and the set of these simplices form
a triangulation of the product when z runs through the (. q)-shufes (. q). (We do not
need this geometric fact, but it explains the idea of the construction.) Dene
1
:
),q
: S
)
(X) S
q
(Y ) S
)q
(X. Y ). o t
(),q)
c(z)((o t) z)
on a pair o. t of singular simplices.
The 1
:
),q
are a strictly associative EilenbergZilber morphism. We call it the shufe
morphism or the EilenbergMac Lane morphism.
2. An approximation of the diagonal is a natural chain map D: S
+
(X) S
+
(X) S
+
(X)
which coincides in dimension zero with . . .. (The name refers to the fact that the
diagonal of a cellular complex is not a cellular map, and so one looks for a homotopic cellular
approximation.) By an application of (11.5.1) one shows that any two approximations of the
diagonal are naturally chain homotopic.
3. The classical approximation of the diagonal is the AlexanderWhitney map.
Let o : ^
n
X be an n-simplex, n = q, 0 _ . q _ n. We have the afne maps
a
)
: ^
)
^
n
, e
i
e
i
and b
q
: ^
q
^
n
, e
i
e
n-qi
. They are used to dene
o
1
)
= o a
)
and o
2
q
= o b
q
.
The AlexanderWhitney approximation of the diagonal is dened by
Do
n
=
)q=n
o
1
)
o
2
q
. o
n
: ^
n
X
9.8. The Homology Product 241
and linear extension.
4. Given an approximation D of the diagonal one constructs from it an EilenbergZilber
morphism Q as the composition
S
+
(X Y )
1
Y
S
+
(X Y ) S
+
(X Y )
pr
#
pr
Y
#
S
+
(X) S
+
(Y ).
Let Q
.V
be the EilenbergZilber morphism obtained from the AlexanderWhitney ap-
proximation of the diagonal and call it the AlexanderWhitney morphism. The Alexander
Whitney morphism is strictly coassociative.
5. The EilenbergMac Lane morphism1M and the AlexanderWhitney morphismW are
also compatible in a certain sense:
S
+
(W X) S
+
(Y 7)
1^
.V.V
S
+
(W X Y 7)
(1I
,Y
1)
#
S
+
(W) S
+
(X) S
+
(Y ) S
+
(7)
1:
,Y
1
S
+
(W Y X 7)
.V
S
+
(W) S
+
(Y ) S
+
(X) S
+
(7)
1^1^
S
+
(W Y ) S
+
(X 7)
commutes.
9.8 The Homology Product
We pass to homology from the chain map 1 in (9.7.4)
H
+
(X. : M) H
+
(Y. T: N) = H
+
(S(X. : M)) H
+
(S(Y. T: N))
H
+
(S(X. : M) S(Y. T: N))
H
+
(S((X. ) (Y. T)): M N)
These maps are natural transformations, and we call them the homology product.
We use the notation . , . , for the homology product. In the case of
M = N = 1 we combine with the map induced by the canonical isomorphism
1
T
1 1 and obtain a homology product
H
i
(X. : 1) H
}
(Y. T: 1) H
i}
((X. ) (Y. T): 1).
Ingeneral we cancompose witha bilinear mapMN 1; for instance we canuse
an 1-module structure 1M M on M. We list some formal properties of the
homology product, for simplicity of notation only for homology with coefcients
in 1. We use the following notation: : (X. ) (X
t
.
t
) and g: (Y. T)
(Y
t
. T
t
) are continuous maps. Let (X T. Y ) be excisive in X Y . Then we
have two boundary operators
d
t
: H
n
((X. )(Y. T)) H
n-1
(XTLY. XT) H
n-1
((Y. T)).
d
tt
: H
n
((X. )(Y. T)) H
n-1
(XTLY. Y ) H
n-1
((X. )T).
t is the interchange mapt (u. ) = (. u). Let C = {c] be a point and1 H
0
(C: 1)
be represented by c 1.
9.8.1 Properties of the homology product.
( g)
+
(. ,) =
+
. g
+
(,).
d. , = d
t
(. ,).
. d, = (1)
[x[
d
tt
(. ,).
(. ,) : = . (, :).
. , = (1)
[x[[,[
t
+
(, .).
1 , = ,.
The algebraic Knneth formula (11.10.1) yields
(9.8.2) Theorem (Knneth Formula). Let 1 be an integral domain. Further, let
(Y. XT) be excisive in XY for singular homology. Then we have a natural
exact sequence
0

i}=n
H
i
(X. : 1) H
}
(Y. T: 1) H
n
((X. ) (Y. T): 1)

i}=n-1
H
i
(X. : 1) + H
}
(Y. T: 1) 0.
The sequence splits, but the splitting is not natural in the variable (X. ). For
homology with coefcients in a eld k we obtain an isomorphism
H
+
(X. : k)
k
H
+
(Y. T: k) H
+
((X. ) (Y. T): k)
as a special case. This isomorphism holds for an arbitrary commutative ring 1 if
the homology groups H
+
(X. : 1) are free 1-modules.
(9.8.3) Example. The homology class e H
1
(R. R 0: 1) 1 represented by
o 1 with the singular simplex o : ^
1
R, (t
0
. t
1
) 1 2t
0
is a generator. The
product , e , with e is an isomorphism
H
n
(Y. T: N) H
n1
((R. R 0) (Y. T): N)
for each 1-module N. This isomorphism can also be deduced from the axiomatic
properties 9.8.1 (see a similar deduction (17.3.1) in the case of cohomology the-
ories). The n-fold product e
n
= e e H
n
(R
n
. R
n
0: 1) serves as a
canonical generator (a homological 1-orientation of R
n
).
9.8. The Homology Product 243
Problems
1. Let
0
. . . .
n
be afnely independent points in R
n
and suppose that the origin 0 R
n
is
contained in the interior of the afne simplex =
0
. . . . .
n
|. We then have the singular
simplex o : ^
n
R
n
determined by o(e
i
) =
i
. Show that o represents a generator .
:
of
H
n
(R
n
. R
n
0: Z). We therefore have a relation .
:
= e
n
. Determine the sign, depending
on . You might rst consider the case n = 2 and make an intuitive guess. (This problem
indicates that keeping track of signs can be a nuisance.)
2. Verify the properties 9.8.1 of the homology product.
3. Study the axioms for a multiplicative cohomology theory and use 9.8.1 to dene multi-
plicative homology theories axiomatically.
Chapter 10
Homology
In this chapter we dene homology theories via the axioms of Eilenberg and Steen-
rod. From these axioms we derive some classical results: the Jordan separation
theorem; invariance of domain and dimension; degree and its determination by lo-
cal data; the theorem of BorsukUlam. The theorem of BorsukUlam is used for
a problem in combinatorics: the determination of the chromatic number of Kneser
graphs.
A second topic of the chapter is the derivation of some results of a general
nature: reduced homology; additivity; suspension isomorphisms; MayerVietoris
sequences; compatibility of homology with colimits.
10.1 The Axioms of Eilenberg and Steenrod
Recall the category TOP(2) of pairs of topological spaces. We use the functor
k : TOP(2) TOP(2) which sends (X. ) to (. 0) and : (X. ) (Y. T) to
the restriction : (. 0) (T. 0).
A homology theory for pairs of spaces consists of a family (h
n
[ n Z) of
covariant functors h
n
: TOP(2) 1- MOD and a family (d
n
[ n Z) of natural
transformations d
n
: h
n
h
n-1
k. These data are required to satisfy the following
axioms of Eilenberg and Steenrod [58], [59]:
(1) Homotopy invariance. For each homotopy
t
in TOP(2) the equality
h
n
(
0
) = h
n
(
1
) holds.
(2) Exact sequence. For each pair (X. ) the sequence
h
n1
(X. )
J
h
n
(. 0) h
n
(X. 0) h
n
(X. )
J

is exact. The undecorated homomorphisms are induced by the inclusions.
(3) Excision. Let (X. ) be a pair and U such that

U
. Then
the inclusion (X U. U) (X. ) induces an excision isomorphism
h
n
(X U. U) h
n
(X. ).
The excision axiom can be expressed in a different form. Suppose Y
1
. Y
2
are
subspaces of Y such that Y = Y
1
LY
2
. Then the inclusion induces an isomorphism
h
n
(Y
1
. Y
1
Y
2
) h
n
(Y. Y
2
).
The module h
n
(X. ) is the n-th homology group (module) of (X. ) in the
homology theory (we also say in degree or in dimension n). We set h
n
(X. 0) =
h
n
(X). The groups h
n
(X) are the absolute groups and the groups h
n
(X. ) are
10.1. The Axioms of Eilenberg and Steenrod 245
the relative groups. The homomorphisms d = d
n
are the boundary operators. We
often write h
n
( ) =
+
and call (as already above)
+
the induced morphism. The
homology groups h
n
(1) for a point 1 are the coefcient groups of the theory.
(More precisely, a group h
n
together with a compatible family of isomorphisms
c
1
: h
n
h
n
(1) for each point 1 is given.) In the case that h
n
= 0 for n ,= 0, we
say that the homology theory satises the dimension axiom and call the homology
theory an ordinary or classical one.
The exact sequence of a pair of spaces can be extended slightly to an exact
sequence for triples T X (see [59, p. 24], [189]). The boundary operator
for a triple is dened by
d: h
n
(X. ) h
n-1
() h
n-1
(. T):
the rst map is the previous boundary operator and the second map is induced by
the inclusion.
(10.1.1) Proposition. For each triple (X. . T) the sequence
h
n1
(X. )
J
h
n
(. T) h
n
(X. T) h
n
(X. )
J

is exact. The undecorated homomorphisms are induced by the inclusions.
We do not derive this proposition from the axioms right now (see 10.4.2 for a
proof which uses the homotopy invariance). In most constructions of homology
theories one veries this more general exact sequence directly from the denitions;
so we can treat it as an extended axiom.
A homology theory is called additive, if the homology groups are compatible
with topological sums. We formulate this as another axiom.
(4) Additivity. Let (X
}
.
}
). J be a family of pairs of spaces. Denote by
i
}
: (X
}
.
}
) (l
}
X
}
. l
}
}
) the canonical inclusions into the topological
sum. Then the additivity axiom says that
}J
h
n
(X
}
.
}
) h
n
(l
}
X
}
. l
}
}
). (.
}
)
}J
h
n
(i
}
)(.
}
)
is always an isomorphism. For nite families the additivity follows from the
axioms (see (10.2.1)).
Singular homology theory has further properties which may also be required in
an axiomatic treatment.
(5) Weak equivalence. A weak equivalence : X Y induces isomorphisms
+
: h
+
(X) h
+
(Y ) of the homology groups.
(6) Compact support. For each . h
n
(X. ) there exists a map : (1. 1)
(X. ) from a pair (1. 1) of compact Hausdorff spaces and : h
n
(1. 1)
with
+
: = ..
246 Chapter 10. Homology
If Axiom(5) holds, then the theory is determined by its restriction to CW-complexes
(actually to simplicial complexes).
Sometimes we have to compare different homology theories. Let h
+
= (h
n
. d
n
)
and k
+
= (k
n
. d
t
n
) be homology theories. A natural transformation
+
: h
+
k
+
of homology theories consists of a family
n
: h
n
k
n
of natural transformations
which are compatible with the boundary operators d
t
n
=
n-1
d
n
.
Problems
1. Let (h
n
. d
n
) be a homology theory with values in 1- MOD. Let (c
n
1) be a family of
units of the ring 1. Then (h
n
. c
n
d
n
) is again a homology theory. It is naturally isomorphic
to the original theory.
2. Given a homology theory (h
n
. d
n
) we can dene a new theory by shifting the indices
k
n
= h
nI
.
3. Let h
be a homology theory. For a xed space Y we dene a new homology theory

whose ingredients are the groups k
n
(X. ) = h
n
(X Y. Y ). The boundary operators
for the new theory are the boundary operators of the pair (X Y. Y ). The projections
pr : X Y X induce a natural transformation k
of homology theories. If h
is
additive then k
is additive.
4. Let h
be a homology theory with values in 1- MOD. Let M be a at 1-module, i.e., the

tensor product
!
M preserves exact sequences. Then the h
n
()
!
M and d
!
M are
the data of a new homology theory. Since the tensor product preserves direct sums, the new
theory is additive if h
was additive. In the case that 1 = Z one can take for M a subring
T of the rational numbers Q, in particular Qitself. It turns out that the rationalized theories
h
() Q are in many respects simpler than the original ones but still contain interesting
information.
5. If
is a family of homology theories ( J), then their direct sum
is again
a homology theory. One can combine this device with the shift of indices.
10.2 Elementary Consequences of the Axioms
We assume given a homology theory h
+
and derive some consequences of the
axioms of Eilenberg and Steenrod.
Suppose the inclusion X induces for = n. n 1 an isomorphism
h
}
() h
}
(X). Then h
n
(X. ) = 0. In particular h
n
(X. X) = 0 always, and
h
n
(0) = h
n
(0. 0) = 0. This is an immediate consequence of the exact sequence.
Let : X Y be an h-equivalence. Then
+
: h
n
(X) h
n
(Y ) is an isomor-
phism, by functoriality and homotopy invariance. If is, in addition, an inclusion,
then h
n
(X. Y ) = 0.
Let : (X. ) (Y. T) be a map such that the components : X Y and
: T induce isomorphisms of homology groups, e.g., the components are
10.2. Elementary Consequences of the Axioms 247
h-equivalences. Then the induced maps
+
: h
n
(X. ) h
n
(Y. T) are isomor-
phisms. The map induces a morphism from the homology sequence of (X. )
into the homology sequence of (Y. T), by functoriality and naturality of d. The
claim is then a consequence of the Five Lemma (11.1.4).
If X and are contractible, then h
n
(X. ) = 0. If (X. . T) is a triple and
X. T are contractible, then the exact sequence of a triple shows that d: h
n
(X. )
h
n-1
(. T) is an isomorphism.
Let the homology theory satisfy the dimension axiom. If X is contractible, then
h
k
(X) = 0 for k ,= 0. A null homotopic map Y Y therefore induces the zero
morphism on h
k
(Y ) for k ,= 0.
Let i : X and suppose there exists r : X such that ri . id. From
id = id
+
= (ri )
+
= r
+
i
+
we see that r
+
is a retraction of i
+
, hence i
+
is always
injective. Therefore the exact homology sequence decomposes into split short exact
sequences 0 h
n
() h
n
(X) h
n
(X. ) 0.
(10.2.1) Proposition (Finite Additivity). Let (X
}
.
}
). J be a nite family of
pairs of spaces. Denote by i
}
: (X
}
.
}
) (l
}
X
}
. l
}
}
) the canonical inclusions
into the topological sum. Then
}J
h
n
(X
}
.
}
) h
n
(l
}
X
}
. l
}
}
). (.
}
)
}J
h
n
(i
}
)(.
}
)
is an isomorphism.
Proof. As a consequence of the excision axiom we see that the inclusion always
induces an isomorphism h
n
(. T) h
n
( C. T C). It sufces to consider
the case J = {1. 2] and, by the Five Lemma, to deal with the absolute groups.
One applies the Sum Lemma (11.1.2) with
k
= h
n
(X
k
). T
k
= h
n
(X. X
k
). C =
h
n
(X
1
X
2
).
(10.2.2) Proposition. The identity |
n
of ^
n
is a cycle modulo d^
n
in singular ho-
mology theory. The group H
n
(^
n
. d^
n
) is isomorphic to Zand |
n
| is a generator.
Proof. The proof is by induction on n. Denote by s(i ) = J
n
i
^
n-1
the i -th face of
^
n
. Consider
h
k-1
(^
n-1
. d^
n-1
)
(d
n
i
)
h
k-1
(d^
n
. d^
n
s(i )
) h
k
(^
n
. d^
n
).
J
The space d^
n
s(i )
is contractible, a linear homotopy contracts it to the

point e
i
. Since ^
n
is also contractible, d is an isomorphism. The inclusion J
n
i
maps (^
n-1
. d^
n-1
) into the complement of e
i
, and as such it is a deformation
retraction of pairs. The excision of e
i
induces an isomorphism. Therefore (J
n
i
)
+
is the composition of two isomorphisms. If we always work with the rst face
map J
n
0
, we obtain by iteration an isomorphism h
k-n
(^
0
) h
k
(^
n
. d^
n
). (So
far we can work with any homology theory.)
Now consider the special case h
n
= H
n
of singular homology with coefcients
in Z. By denition of the boundary operator, d|
n
| = (1)
i
J
n
i
|, since the J
n
}
for
,= i are zero in the relative group. If we again work with J
n
0
, we see that the
isomorphism above sends the generator |
n-1
| to |
n
|.
If we work with H
k
for k ,= n, then the isomorphism above and the dimension
axiom tell us that H
k
(^
n
. d^
n
) = 0. The pair (^
n
. d^
n
) is homeomorphic to
(D
n
. S
n-1
). So we also see that H
k
(D
n
. S
n-1
) is zero for k ,= n and isomorphic
to Z for k = n.
In an additive homology theory we also have an additivity isomorphism for
pointed spaces if the base points are well-behaved. For a nite number of summands
we do not need the additivity axiom in (10.2.3).
(10.2.3) Proposition. Let (X
}
. 1
}
) be a family of pointed spaces and
_
}
X
}
=
(X. 1) the pointed sum with embedding i
}
: X
}
X of a summand. Assume that
the closure of 1 in X has a neighbourhood U such that h
+
(U. 1) h
+
(X. 1) is
the zero map. Then (i
}
+
):

}
h
+
(X
}
. 1
}
) h
+
(X. 1) is an isomorphism.
Proof. Set U
}
= U X
}
. Consider the diagram
0

h
+
(X. 1)
(1)
h
+
(X. U)
J
h
+
(U. 1)
h
+
(lX
}
. l1
}
)
(2)
h
+
(lX
}
. lU
}
)
(3)
h
+
(lU
}
. l1
}
).
(4)
The vertical morphisms are induced by the quotient maps. The horizontal mor-
phisms are part of the exact sequence of triples. The hypothesis implies that (1) is
injective. We use the additivity in order to conclude that (2) and (4) are injective.
This is due to the fact that we have the projections
}
: X X
}
, and
k
+
i
}
+
=
k}
.
We show that (3) is an isomorphism. Consider the diagram
h
+
(X 1. U 1)
h
+
(l(X
}
1
}
). l(U
}
1
}
))
=
h
+
(X. U) h
+
(lX
}
. lU
}
).
(3)
The isomorphisms hold by excision; here we use the assumption

1 U
. Diagram
chasing (Five Lemma) now shows that (2) is also surjective.
(10.2.4) Remark. The hypothesis of (10.2.3) is satised if 1
}
is closed in X
}
and
has a neighbourhood U
}
such that U
}
X
}
is pointed homotopic to the constant
map. These conditions hold if the spaces (X
}
. 1
}
) are well-pointed; see (5.4.4).
10.3. Jordan Curves. Invariance of Domain 249
10.2.5 Suspension. We dene the homological suspension isomorphism o =
o
(A,)
by the commutative diagram
h
n
(X. )
c
h
n
(1 X. 1 T)
(1)
h
n1
((1. d1) (X. ))
J
h
n
(d1 X L 1 . 0 X L 1 ).
(1) is an isomorphism: excise 0 X and then use an h-equivalence. The boundary
operator d for the triple sequence of (1 X. d1 X L1 . 0 X L1 ) is an
isomorphism, since 0XL1 1 X is an h-equivalence. We can interchange
the roles of 0 and 1; let o
-
denote this suspension isomorphism. By applying the
Hexagon Lemma (11.1.3) with center group h
n
(1 X. 1 Ld1 X) we obtain
o = o
-
(draw the appropriate diagram). For some purposes of homotopy theory
one has to use a similar suspension isomorphism dened with X 1.
The n-fold iteration of o provides us in particular with an isomorphism
o
n
: h
k
h
k
(1
0
) h
k1
(1. d1) h
kn
(1
n
. d1
n
).
10.3 Jordan Curves. Invariance of Domain
As a rst application of homology theory we prove a general duality theorem.
We then apply this general result to prove classical results: A generalized Jordan
separation theorem and the invariance of domain and dimension. For this section
see [53].
The propositions (10.3.1) and (10.3.2) can be proved in a similar manner for an
arbitrary homology (or, mutatis mutandis, cohomology) theory. For the applications
one needs homology groups which determine the cardinality of
0
(X) for open
subsets X of Euclidean spaces, and this holds, e.g., for singular homology H
+
().
(If one knows a little analysis, one can use de Rham cohomology for open subsets
of Euclidean spaces.)
(10.3.1) Proposition. Let R
n
be a closed subset. Then H
k
(R
n
. R
n
) and
H
k1
(R
n
R. R
n
R 0) are isomorphic.
Proof. We use the open subsets of R
n1
H
= (R
n
)| 1. o L|0. o
H
-
= (R
n
)| o. 1 L| o. 0
H
L H
-
= R
n1
0
H
H
-
= (R
n
)| 1. 1.
These data occur in the diagram
H
k
(R
n
. R
n
)
(3)
_ _ _ _ _ _ _ _ _ _ _ _ _ _
H
k1
(R
n1
. R
n1
0)
J
H
k
(R
n1
. H
H
-
) H
k
(H
. H
H
-
)
(2)
(1)
H
k
(H
L H
-
. H
-
).
The map d is the boundary operator of the triple (R
n1
. H
LH
-
. H
-
). The maps
(1), (2), and (3) are induced by the inclusions. We show that the morphisms in the
diagramare isomorphisms. The proof uses the fact that H
and H
-
are contractible;
the homotopy h
t
: H
, (.. s) (.. s t ) starts at the identity and has an

image in the contractible space R
n
|0. o. The map d is an isomorphism, because
H
k
(R
n1
. H
-
) = 0, by contractibility of H
-
. The map (1) is an excision. The
maps (2) and (3) are isomorphisms by homotopy invariance (R
n
= R
n
0).
(10.3.2) Theorem (Duality Theorem). Let and T be closed homeomorphic sub-
sets of R
n
. Then the groups H
k
(R
n
. R
n
) and H
k
(R
n
. R
n
T) are isomorphic
(k Z).
Proof. A homeomorphism : T yields a homeomorphism : R
n
R
n
R
n
R
n
, which sends 0 via 0 to T 0 (see (7.3.1)). We obtain an induced
isomorphism
H
kn
(R
n
R
n
. R
n
R
n
0) H
kn
(R
n
R
n
. R
n
R
n
T 0).
Now apply (10.3.1) n times.
(10.3.3) Theorem (Component Theorem). Let and T be closed homeomorphic
subsets of R
n
. Then
0
(R
n
) and
0
(R
n
T) have the same cardinality.
Proof. We use the fact that H
0
(R
n
) is the free abelian group on
0
(R
n
)
and the algebraic fact a free abelian group determines the cardinality of a basis (=
the rank). Suppose that ,= R
n
. Then we have an exact sequence
0 H
1
(R
n
. R
n
) H
0
(R
n
) H
0
(R
n
) 0.
This shows the rank of H
1
(R
n
. R
n
) is one less than the rank of H
0
(R
n
).
Hence if and T are different from R
n
, then the result follows from (10.3.2). We
see in the next section that = R
n
implies that T is open in R
n
and therefore,
since R
n
is connected, also equal to R
n
.
An injective continuous map : S
1
R
2
is an embedding, and its image is
called a Jordan curve.
10.3. Jordan Curves. Invariance of Domain 251
(10.3.4) Theorem (Jordan Separation Theorem). Let S R
n
be homeomorphic to
S
n-1
(n _ 2). Then R
n
S has two path components, the bounded interior J and
the unbounded exterior . Moreover, S is the set of boundary points of J and of .
Proof. The assertion holds in the elementary case S = S
n-1
. Hence, by (10.3.3),
the complement of S has two components. It remains to study the boundary points.
Let . S andlet V be anopenneighbourhoodof . inR
n
. ThenC = S(SV )
is closed in S and homeomorphic to a proper closed subset D of S
n-1
. Therefore
R
n
D is path connected and hence, by (10.3.3), also R
n
C. Let J and
q and n: 0. 1| R
n
C a path from to q. Then n
-1
(S) ,= 0. Let t
1
be
the minimum and t
2
the maximum of n
-1
(S). Then n(t
1
) and n(t
2
) are contained
in S V . Therefore n(t
1
) is limit point of n(0. t
1
) J and n(t
2
) limit point of
n( |t
2
. 1|) . Hence there exist t
3
0. t
1
with n(t
3
) J V and t
4
|t
2
. 1|
with n(t
4
) V . This shows that . is contained in the boundary of J and of .
(10.3.5) Remarks. For n = 2 one can improve the separation theorem. There
holds the theorem of Schoenies (for a topological proof see [141]):
Let J R
2
be a Jordan curve. Then there exists a homeomorphism : R
2
R
2
which maps J onto the standard circle.
There exists a stronger result. By the Riemann mapping theorem there exists
a holomorphic isomorphism from the interior of J onto the interior of S
1
; and
one can show that this isomorphism can be extended to a homeomorphism of the
closures. See e.g., [149].
For n _ 3 it is in general not true that the interior of an embedding S
n-1
R
n
is homeomorphic to an n-cell. One can construct an embedding D
3
R
3
such
that the complement is not simply connected. Under some regularity conditions on
the embedding the situation still resembles the standard embedding. There holds
the theorem of M. Brown [32] (see also [25, p. 236]):
Let : S
n-1
1. 1| S
n
be an embedding (n _ 1). Then the closure of
each component of S
n
(S
n-1
0) is homeomorphic to D
n
.
From the duality theorem (18.3.3) one can conclude that both components of
S
n
S have for an arbitrary embedding the integral singular homology groups of
a point.
(10.3.6) Theorem. Let R
n
be homeomorphic to D
k
, k _ n. Then R
n
is
path connected (n > 1).
Proof. D
k
is compact, hence is compact too. Therefore is closed in R
n
and the assertion follows from (10.3.3), since D
k
obviously has a path connected
complement.
(10.3.7) Theorem (Invariance of Domain). Let U R
n
be open and : U R
n
an injective continuous map. Then (U) is open in R
n
, and maps U homeomor-
phically onto (U).
Let V R
n
be homeomorphic to an open subset of U R
n
. Then V is open
in R
n
.
Proof. It sufces to show that (U) is open. It then follows that is open.
Let D = {. R
n
[ [. a[ _ ] U, and let S be the boundary of D.
It sufces to show that (D
) is open. We consider the case n _ 2 and leave

the case n = 1 as exercise. The set S as well as T = (S) are homeomorphic
to S
n-1
. Suppose U
1
, U
2
are the (open) components of R
n
T . Let U
1
be
unbounded. By (10.3.6), R
n
(D) is path connected, and therefore contained in
U
1
or U
2
. Since (D) is compact, the complement is unbounded. This implies
T L U
1
= R
n
U
2
(D) and then U
1
(D
). Since D
is connected, so
is (D
). The inclusion (D
) U
1
L U
2
shows that (D
) U
1
. Therefore
(D
) = U
1
, and this set is open.
The second statement follows from the rst, since by hypothesis there exists an
injective continuous map : U R
n
with image V .
(10.3.8) Theorem (Invariance of Dimension). Let U R
n
and V R
n
be non-
empty homeomorphic open subsets. Then m = n.
Proof. Let m < n. Then, by (10.3.7), U R
n
R
n
is open in R
n
, which is
impossible.
10.4 Reduced Homology Groups
The coefcient groups of a homology theory are important data of the theory but
they contain no information about spaces. We therefore split off these groups from
the homology groups h
n
(X).
Let X be a non-empty space and r : X 1 the unique map to a point. We set
h
n
(X) = kernel (r
+
: h
n
(X) h
n
(1))
and call this group a reduced homology group. The homomorphism h
n
( ) for a
continuous map : X Y restricts to

h
n
( ):

h
n
(X)

h
n
(Y ). In this way

h
n
becomes a homotopy invariant functor TOP 1- MOD. Let (X. +) be a pointed
space and i : 1 = {+] X be the inclusion of the base point. Then we have a
short exact sequence
0 h
n
(1)
i
h
n
(X)
}
h
n
(X. 1) 0
with a splitting r
+
of i
+
. Thus
+
induces isomorphisms

h
n
(X) h
n
(X. 1)
Coker(i
+
). By(11.1.1) we also have an isomorphism
h
n
(X) =

h
n
(X) i
+
(h
n
(1))

h
n
(X) h
n
10.4. Reduced Homology Groups 253
which is natural for pointed maps (but not, in general, for arbitrary maps), and the
canonical diagram
h
n
(X)

K
K
K
K
K
K
K
K
K
h
n
(X)

}
Coker i
+
r
r
r
r
r
r
r
r
r
r
h
n
(X. 1)
is commutative.
(10.4.1) Proposition. Let be non-empty. The image of the boundary operator
d: h
n
(X. ) h
n-1
() is contained in

h
n-1
(). The images of h
n
(X) and

h
n
(X)
in h
n
(X. ) coincide. The homology sequence for the reduced groups

h
n
()

h
n
(X) h
n
(X. )

h
n-1
()
is exact.
Proof. Map the exact sequence of (X. ) into the exact sequence of (1. 1) and
perform diagram chasing. The exactness is also a special case of (11.3.2).
From (10.4.1) we see: If is contractible, then

h
n
() = 0 and

h
n
(X)
h
n
(X. ) is an isomorphism. If X is contractible, then d: h
n
(X. )

h
n-1
() is
an isomorphism.
10.4.2 Triple sequence. Let (X. . +) be a pointed pair. The reduced homology
sequence of (X. ) is canonically isomorphic to the homology sequence of the
triple (X. . +). Hence the latter is exact.
Let CT denote the cone over T. The homology sequence of a triple (X. . T)
is, via excision, isomorphic to the sequence of (X L CT. L CT. CT), and the
latter, via h-equivalence isomorphic to the sequence of (X LCT. LCT. +). This
proves the exactness of the triple sequence.
Under the hypothesis of (10.2.3) or (10.2.4) we have for an additive homology
theory an isomorphism

}
h
+
(X
}
)

h
+
_ _
}
X
}
_
. For a nite number of sum-
mands we do not need the additivity axiom. We call
_
}
X
}
decomposable with
respect to

h
n
if the canonical map
h
n
(X
}
)

h
n
_ _
}
X
}
_
is an isomorphism.
(10.4.3) Proposition. Suppose X Y is decomposable with respect to h
n
. h
n1
.
Then the homology sequence of (X Y. X Y ) yields a split short exact sequence
0

h
n
(X Y )

h
n
(X Y ) h
n
(X Y. X Y ) 0.
Proof. The projections onto the factors induce

h
n
(X Y )

h
n
(X)

h
n
(Y ), and
together with the assumed decomposition isomorphism we obtain a left inverse to
the morphism

h
n
(X Y )

h
n
(X Y ).
Let (C. j) be a monoid in h-TOP, i.e., j: C C C is a pointed map such
that the composition with the inclusions of the summands is pointed homotopic to
the identity. Then we have the j-sum in each homotopy set C. Y |
0
, dened by
| g| = ( g)j| with the folding map = (id. id): Y Y Y . Let us
write h =

h
n
.
(10.4.4) Proposition. Assume that C C is decomposable with respect to h
n
. Then
the morphism
o: C. Y |
0
Hom(h(C). h(Y )). |
+
is a homomorphism.
h(C)

P
P
P
P
P
P
P
P
P
P
P
P
h(C C)
(( )

(1)
h(Y Y )

h(Y )
h(C) h(C)
(

h(Y ) h(Y )
(2)
o
o
o
o
o
o
o
o
o
o
o
o
with the diagonal J and the addition a(,. :) = , :. By our assumption about C,
the isomorphism (1) is induced by the projections onto the summands, and by our
assumption about j, the left triangle commutes. The morphism (2) and the inverse
of (1) are induced by the injection of the summands; this shows that the rectangle
and the right triangle commute. Now observe that a(
+
g
+
)J =
+
g
+
.
The hypothesis of (10.4.4) holds for the suspension C = X of a well-pointed
space X. We thus obtain in particular a homomorphism
o:
n
(X) Hom(
h
n
(S
n
).

h
n
(X))
for each pointed space X.
(10.4.5) Proposition. Let i : X be acobrationandlet : (X. ) (X,. +)
be the map which identies to a base point +. Then
+
: h
n
(X. ) h
n
(X,. +)
is an isomorphism. We can write this isomorphism also in the form q : h
n
(X. )
h
n
(X,).
Proof. Let X L C = (C X),(a. 1) ~ i(a) be the mapping cone of i . The
inclusion : (X. ) (X L C. C) induces an isomorphism in homology:
Excise the cone point 0 and apply a homotopy equivalence. For a cobration
we have an h-equivalence : XLC XLC,C X,. Hence q
+
=
+
+
is the composition of two isomorphisms.
10.4. Reduced Homology Groups 255
The isomorphism q also holds for = 0. In this case X, = X {+] and
is the inclusion (X. 0) (X. {+]). We use q to modify the exact sequence (10.4.1)
in the case of a cobration X

h
n
()

h
n
(X)

h
n
(X,)

h
n-1
() .
(10.4.6) Proposition. Let : X be a cobration and attach X to T via
: T. Then the map h
n
(X. ) h
n
(X L
T. T) induced by the inclusion

is an isomorphism.
Proof. We apply (10.4.5) to the homeomorphism X, X L
T,T.
(10.4.7) Proposition. Let : X Y be a pointed map between well-pointed
spaces and X
(
Y
(
1
C( ) the beginning of the cobre sequence. Then the
sequence

h
n
(X)
(

h
n
(Y )
(
1

h
n
(C( ))
is exact.
Proof. Let 7( ) = (X 1 Y ),(.. 1) ~ (.) be the (unpointed) mapping
cylinder of and X 7( ), . (.. 0) the canonical inclusion, a cobration.
Consider the commuting diagram
h
n
(X)

=
h
n
(7( ))

h
n
(7( ). X)
h
n
(X)
(

h
n
(Y )
c(( )

h
n
(7( ),X)
with the canonical inclusion c( ): Y 7( ),X. The top row is the exact se-
quence of the pair. The isomorphisms hold by homotopy invariance and (10.4.5).
Now we use that for a well-pointed pair the quotient map 7( ),X C( ) is a
homotopy equivalence, since a unit interval which is embedded as a cobration is
identied to a point.
For a well-pointed space X the inclusion X d1 L {+] 1 X 1 is
a cobration. The quotient is the suspension X. From (10.4.5) we obtain an
isomorphism
q : h
n
((X. +) (1. d1)) h
n
(X. +)

h
n
(X)
and a suspension isomorphism o :

h
n
(X)

h
n1
(X) which makes the diagram
h
n
(X. +)
c
h
n1
((X. +) (1. d1))
q
h
n
(X)
c

h
n1
(X)
commutative. In particular, we obtain h
n

h
n
(1
0
,d1
0
) h
nn
(1
n
,d1
n
).
Problems
1. Let
q
: S
n
S
n
be a map of degree q, and denote by M(q. n) its mapping cone.
Determine the groups and homomorphisms in the sequence (9.6.1) for the space M(q. n).
2. Let M(q. 1) = M(q). Use the cobre sequence of id
.
q
in order to derive an exact
sequence
0

h
n
(X) Z,q

h
n1
(X . M(q)) Tor(
h
n-1
(X). Z,q) 0.
This suggests dening for any homology theory

h
a theory with coefcients Z,q by
( . M(q)). Unfortunately the homotopy situation is more complicated than one would
expect (or wish), see [8]. Spaces of the type M(q. n) are sometimes called Moore spaces.
10.5 The Degree
In 10.2.5 we determined the homology groups of spheres from the axioms of a
homology theory h
+
. We describe yet another variant.
We use the standard subspaces D
n
. S
n-1
. 1
n
= D
n
S
n-1
of R
n
and D
n
=
{(.
1
. . . . . .
n1
) S
n
[ .
n1
_ 0]. The space D
0
= {D] is a singleton and
S
-1
= 0.
We dene a suspension isomorphism o
as the composition
o
:

h
k-1
(S
n-1
)
h
k
(D
n
-
. S
n-1
)
J
h
k
(S
n
. D
n
)

h
k
(S
n
).
The maps and d are isomorphisms, since D

n
is contractible, and s
is an iso-
morphism (compare (6.4.4)). Iteration of o
yields (suspension) isomorphisms

o
(n)
: h
k-n

h
k-n
(S
0
)

h
k
(S
n
). h
k
(S
n
) h
k-n
h
n
.
The rst isomorphism is determined by
h
n

h
n
(S
0
) h
n
(S
0
) h
n
(1) h
n
(1) h
n
h
n
. . (.. .).
We also have an isomorphismd: h
k
(D
n
. S
n-1
) h
k-1
(S
n-1
. e)

h
k-1
(S
n-1
).
In the case of ordinary homology H
+
(: G) with coefcients in G we obtain:
H
k
(S
n
: G)
G. k = n > 0. n > k = 0.
G G. k = n = 0.
0. otherwise.
Moreover H
n
(D
n
. S
n-1
: G) H
n
(R
n
. R
n
0: G) G.
A generator of H
n
(D
n
. S
n-1
: Z) Z is called a homological orientation of
D
n
; and a generator of H
n
(S
n
), n _ 1 a homological orientation of S
n
. Given
an orientation : H
n
(S
n
) and a continuous map : S
n
S
n
, we dene the
10.5. The Degree 257
(homological ) degree J( ) Z of by
+
(:) = J( ):. A different choice
of a generator does not effect the degree. We also dene the degree for n = 0:
The identity has degree 1, the antipodal map the degree 1 and a constant map
the degree 0. From the denition we see directly some properties of the degree:
J( g) = J( )J(g), J(id) = 1; a homotopy equivalence has degree 1; a null
homotopic map has degree zero.
(10.5.1) Proposition. Let h
+
be a homology theory such that h
0
(Point) = Z. Then
o: S
n
. S
n
| Hom(
h
n
(S
n
).

h
n
(S
n
)),
+
is an isomorphism (n _ 1).
Proof. The suspension isomorphism and the hypothesis yield

h
n
(S
n
) Z. Thus
the Hom-group is canonically isomorphic to Z. We now use that
n
(S
n
)
S
n
. S
n
| Z. The identity of S
n
is mapped to 1. By (10.4.4), o is a homo-
morphism, hence necessarily an isomorphism.
We have already dened a (homotopical) degree. From (10.5.1) we see that
the homotopical and the homological degree coincide. If one starts algebraic
topology with (singular) homology, then one has in any case the important homo-
topy invariant degree. Proposition (10.5.4) is not immediate fromthe homological
denition.
(10.5.2) Proposition. Dene an isomorphism o
-
:

h
k-1
(S
n-1
)

h
k
(S
n
) as in
the case of o
, but with the roles of D

n
interchanged. Then o
= o
-
.
h
k
(S
n
)
I
j
j
j
j
j
j
j
j
j
j
I
-
U
U
U
U
U
U
U
U
U
U
)
h
k
(S
n
. D
n
) h
k
(S
n
. D
n
-
)
h
k
(S
n
. S
n-1
)
}
U
U
U
U
U
U
U
U
}
-
i
i
i
i
i
i
i
i
i
J
h
k
(D
n
-
. S
n-1
)
i
-
j
j
j
j
j
j
j
j
J

T
T
T
T
T
T
T
T
h
k
(D
n
. S
n-1
)
i
U
U
U
U
U
U
U
U
J
-
j
j
j
j
j
j
j
j
x
-
h
k-1
(S
n-1
)
and apply the Hexagon Lemma (11.1.3).
(10.5.3) Proposition. Let O(n 1) and l
: S
n
S
n
, . .. Then l
+
is on

h
k
(S
n
) the multiplication by det(). The antipodal map . . on S
n
has
the degree (1)
n1
.
Proof. Let t : S
n
S
n
change the sign of the rst coordinate. Then t
+
o
= o
-
.
Hence t
+
= id, by (10.5.2). The group O(n 1) has two path components.
If and T are contained in the same component, then l
and l
B
are homotopic.
Representatives of the components are the unit matrix 1 and the diagonal matrix
T = Diag(1. 1. . . . . 1). The relations l
T
= t and l
T
= id now nish the proof.

n
S
n
be a map of degree J. Then induces on
h
k
(S
n
) the multiplication by J.
Proof. The cases J = 1. 0 are clear and J = 1 follows from (10.5.3). The
general case is then a consequence of (10.4.4) and our knowledge of
n
(S
n
).
(10.5.5) Proposition. Let GL
n
(R) and l
: R
n
R
n
, . .. Then l
+
is
on h
k
(R
n
. R
n
0) the multiplication by the sign c() = det(),[ det()[ of the
determinant.
Proof. We have isomorphisms h
k
(R
n
. R
n
0)
J

h
k-1
(R
n
0)

h
k-1
(S
n-1
)
which are compatible with the action of l
+
if O(n). In this case the claim
follows from (10.5.3). In the general case we use that GL
n
(R) has two path com-
ponents which are characterized by the sign of the determinant.
Let S
n
i
(a) = {. R
n1
[ [.a[ = r]. We have a canonical homeomorphism
h
i,o
: S
n
S
n
i
(a), . r. a and r
b
: R
n1
b S
n
, . N(. b). The
winding number of : S
n
i
(a) R
n1
b about b is dened as the degree of
r
b
h
i,o
.
10.5.6 Local degree. Let 1 S
n
be compact and different from S
n
and let
U be an open neighbourhood of 1. Then we have an excision isomorphism
H
n
(U. U 1) H
n
(S
n
. S
n
1). For a continuous map : S
n
S
n
we
let 1 =
-1
(). Consider the diagram
H
n
(S
n
)
(
H
n
(S
n
)
(1)
H
n
(S
n
. S
n
1)
(

H
n
(S
n
. S
n
)
H
n
(U. U 1)
(
I
H
n
(S
n
. S
n
)
=
with the restriction

U
of . The exact sequence of the pair (S
n
. S
n
) shows
that (1) is an isomorphism (n _ 1). Let : H
n
(S
n
) be a generator and :(U. 1)
its image in H
n
(U. U 1). Commutativity of the diagram shows
U
+
:
U,1
=
J( ):
S
n
,]
. Hence the degree only depends on the restriction
U
. For any compact
set 1 with (1) = {] and open neighbourhood W of 1 we dene the (partial)
10.5. The Degree 259
degree J(. 1) by
W
+
:(W. 1) = J(. 1):(S
n
. ); it is independent of the choice
of W.
(10.5.7) Lemma. Suppose U = U
1
LU
2
is the disjoint union of open sets U
}
. Set
1
}
= U
}
1. The inclusions induce the additivity isomorphism
(i
1
. i
2
): H
n
(U
1
. U
1
1
1
) H
n
(U
2
. U
2
1
2
) H
n
(U. U 1).
Then the relation :(U. 1) = i
1
:(U
1
. 1
1
) i
2
:(U
2
. 1
2
) holds.
Proof. Consider the diagram with M = S
n
:
H
n
(M)
n
n
n
n
n
n
n
n
n
n
n
n
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
H
n
(M. M 1)

H
n
(M. M 1
2
)
H
n
(U. U 1)
}
1
P
P
P
P
P
P
P
P
P
P
P
P
H
n
(U
2
. U
2
1
2
)
i
2
o
1
m
m
m
m
m
m
m
m
m
m
m
m
H
n
(U. U 1
2
).
There exist .
1
. .
2
such that :(U. 1) = i
1
.
1
i
2
.
2
. We compute
1
:(U. 1) = :(U. 1
2
) = a
1
:(U
2
. 1
2
) =
1
i
2
.
2
= a
1
.
2
.
This proves .
2
= :(U
2
. 1
2
).
As a consequence of this lemma we obtain the additivity of the degree J( ) =
J(. 1
1
) J(. 1
2
). Suppose 1 is a nite set, then we can choose U as a
disjoint union of open sets U
x
. . 1 such that U
x
1 = {.]. In that case
x1
H
n
(U
x
. U
x
.) H
n
(U. U 1); and H
n
(U
x
. U
x
.) Z, by exci-
sion and H
n
(D
n
. S
n-1
) Z. We call the integer J(. .) dened by
+
:
U
\
,x
=
J(. .):
S
n
,]
the local degree of at . With this notation we therefore have
J( ) =
x1
J(. .).
Let U R
n
be an open neighbourhood of the origin. Then we have an excision
isomorphismh
k
(U. U 0) h
k
(R
n
. R
n
0). Suppose g: U R
n
is a map with
the properties: (1) Continuously differentiable (a C
1
-map); (2) g
-1
(0) = {0]; (3)
the differential Dg(0) is invertible. Under these conditions we show:
(10.5.8) Proposition. The induced map
h
k
(R
n
. R
n
0) h
k
(U. U 0)
h
k
(R
n
. R
n
0)
is multiplication by the sign of the determinant of the differential Dg(0).
Proof. There exist continuous maps h
i
: U R
n
with g(.) =
n
i=1
.
i
h
i
(.) and
Dg(0)(.
1
. . . . . .
n
) =

n
i=1
.
i
h
i
(0). We dene a homotopy h
t
: (U. U 0)
(R
n
. R
n
0) from g to Dg(0)
h
t
(.) =
n
i=1
.
i
h
i
(t.) =
_
t
-1
g(t.). t > 0.
Dg(0). t = 0.
Now we use (10.5.5).
(10.5.9) Example. Let : S
n
S
n
be continuously differentiable and a regular
value of , i.e., the differential of in each point .
-1
() is bijective. Then
J(. .) = 1, and the plus-sign holds, if the differential respects the orientation
at .. For the proof express in terms of orientation preserving local charts and
apply (10.5.8).
If S is homeomorphic to S
n
and a self map of S we choose a homeomorphism
h: S
n
S and dene the degree of as the degree of h
-1
h. Let g: R
n
R
n
be a proper map. We dene its degree as the degree of the extension of g to the
one-point compactication.
A map : S
n
S
n
S
n
, n _ 1, has a bi-degree (a. b) Z Z where a is
the degree of . (.. ,) for a xed , and b the degree of , (.. ,) for a
xed ..
Problems
1. If : S
n
S
n
is not surjective, then is null homotopic and hence J( ) = 0.
2. Suppose (.) ,= . for each . S
n
, then h(.. t ) = t(.)(1t ). ,= 0 for t 0. 1|.
We compose with N : . .,[.[ and obtain in J(.. t ) = Nh(.. t ) a homotopy from the
identity to . If always (.) ,= ., then G(.. t ) = N(t. (1 t )(.)) is a homotopy
from to the antipodal map. Thus if J( ) ,= 1, there exists . such that (.) = . or
(.) = ..
3. A permutation z of (t
0
. . . . . t
n
) induces an afne homeomorphism of (^
n
. d^
n
). The
induced homomorphism in h
k
(^
n
. d^
n
) is the multiplication with the sign of the permuta-
tion z. The same holds for the linear permutation map l
induced by z on the vector space

N = {(t
0
. . . . . t
n
) [
i
t
i
= 0] and h
k
(N. N 0). One can compute the determinant of l
by using the decomposition R

n1
= N D with the diagonal D = {(t. . . . . t ) [ t R].
4. The map S
i
S
i
, (,. t ) (2t,. 2t
2
1), , R
i
, t R has degree 1 (1)
i1
.
The point (0. . . . . 0. 1) is a regular value.
5. Consider a complex polynomial as self-map of the Riemann sphere C1
1
S
2
. Then
the homological degree is the algebraic degree of the polynomial. A quotient = ,q of
two complex polynomials (without common divisor) of degree mand q of degree n can be
considered as a self-map of C1
1
. Show that the homological degree is a = max(m. n). In
homogeneous coordinates can be written as :. n| n
u
(:,n). n
u
q(:,n)|. Suppose
10.6. The Theorem of Borsuk and Ulam 261
c C is such that : (:) cq(:) has a pairwise different zeros. Then c. 1| is a regular
value of .
6. Consider S
3
as the topological group of quaternions of norm 1. Determine the degree of
: :
k
, k Z.
7. The map S
n
S
n
S
n
, (.. ) 2(.. ). has bi-degree (1 (1)
n-1
. 1). If
= (1. 0. . . . . 0) = e
0
, then e
0
is a regular value of . e
0
2(e
0
. .).. (Here (. )
is the standard inner product.)
8. Let a. b. . q N with a bq = 1. Then : C
2
C
2
, (.. ,) (.
)
,
q
. .
b
,
u
)
is proper and has degree 1. The point (1. 0) is a regular value.
10.6 The Theorem of Borsuk and Ulam
We describe another classical result which uses the homotopy notion in the presence
of a symmetry. As a rather striking application to a problem in combinatorics we
present the proof of Greene [74] for the determination of the chromatic number of
the Kneser graphs.
We have the antipodal symmetry . . on the Euclidean spaces. A map
: T which is equivariant with respect to this symmetry, i.e., which satises
(.) = (.), is called antipodal or odd; here and T are subsets of Euclidean
spaces that are invariant with respect to the antipodal symmetry. The additional
presence of the symmetry has remarkable consequences: Classical theorems known
under the name of BorsukUlam theorems and LusternikSchnirelmann theorems.
The basic result has a number of equivalent formulations.
(10.6.1) Theorem. The following assertions are equivalent:
(1) Let : S
n
R
n
be continuous. Then there exists . S
n
such that (.) =
(.).
(2) Let : S
n
R
n
be antipodal. Then there exists . S
n
such that (.) = 0.
(3) There does not exist an antipodal map : S
n
S
n-1
.
(4) There does not exist a continuous map : D
n
S
n-1
which restricts to an
antipodal map on the boundary.
(5) An antipodal map S
n-1
S
n-1
is not null homotopic.
(6) Suppose S
n
= J
1
L J
2
L L J
n1
with non-empty closed sets J
}
. Then
at least one of the sets J
}
contains an antipodal pair of points.
(7) Let S
n
=
1
L
2
L L
n1
and assume that each
}
is either open or
closed. Then at least one of the
}
contains an antipodal pair.
Proof. (1) =(2). By (1) there exists . with (.) = (.). Since is antipodal,
(.) = (.) and hence (.) = 0.
(2) =(3). The existence of an antipodal map contradicts (2).
(3) = (4). Let D
n
= {(.
0
. . . . . .
n
) S
n
[ .
n
_ 0]. The projection
h: D
n
D
n-1
onto the rst n 1 coordinates is a homeomorphism which is the
identity on the boundary. Suppose : D
n
S
n-1
is antipodal on the boundary.
Dene g: S
n
S
n-1
by
g(.) =
_
h(.). . D
n
.
h(.). . D
n
-
.
If . S
n-1
= D
n
D
n
-
, then h(.) = ., h(.) = . and h(.) = (.) =
(.) = h(.). Hence g is well-dened and continuous. One veries that
g is antipodal.
(4) = (5). If an antipodal map S
n-1
S
n-1
were null homotopic, then we
could extend this map to D
n
, contradicting (4). Conversely, if a map of type (4)
would exist, then the restriction to S
n-1
would contradict (5).
(1) =(6). We consider the function
: S
n
R
n
. . (J(.. J
1
). . . . . J(.. J
n
))
dened with the Euclidean distance J. By (1), there exists . such that (.) =
(.) = ,. If the i -th component of , is zero, then J(.. J
i
) = J(.. J
i
) = 0
and therefore .. . J
i
since J
i
is closed. If all coordinates of , are non-zero,
then . and . are not contained in
_
n
i=1
J
i
, so they are contained in J
n1
.
(6) = (3). There exists a closed covering J
1
. . . . . J
n1
of S
n-1
such that
no J
i
contains an antipodal pair, e.g., project the standard simplex onto the sphere
and take the images of the faces. Suppose : S
n
S
n-1
is antipodal. Then the
covering by the
-1
(J
i
) contradicts (6).
(3) = (2). If (.) ,= 0 for all ., then . [(.)[
-1
(.) is an antipodal
map S
n
S
n-1
.
(2) = (1). Given : S
n
R
n
. Then g(.) = (.) (.) is antipodal;
g(.) = 0 implies (.) = (.).
(6) = (7). Suppose for the moment that the
}
are open. Then we can nd
a closed shrinking and apply (6). In the general case let
1
. . . . .
}
be closed and
U
}1
. . . . . U
n1
open. Suppose there are no antipodal pairs in the U
}
. Thicken
the
i
to open c-neighbourhoods U
t
(
i
). Let c = n
-1
. By the case of an open
covering we nd an antipodal pair (.
n
. .
n
) in some U
t
(
i
). By passing to a
subsequence we nd an i _ and
lim
n-o
J(.
n
.
i
) = lim
n-o
J(.
n
.
i
) = 0.
A convergent subsequence yields an antipodal pair in
i
.
(7) =(3). As in the case (6) =(3).
Since the identity is antipodal, we see that (10.6.1) implies the retraction theo-
rem, see (6.6.1). Part (1) shows that R
n
does not contain a subset homeomorphic
to S
n
.
10.6. The Theorem of Borsuk and Ulam 263
(10.6.2) Lemma. Let J : S
n
S
n
be an odd map. Then J is homotopic as odd
map to a map g such that g(S
i
) S
i
for 0 _ i _ n.
Proof. Let : S
n
R1
n
be the quotient map and : R1
n
R1
n
be induced
by J on the orbit space. Choose a homotopy
t
from =
0
to a cellular map,
i.e.,
1
(R1
i
) R1
i
for 0 _ i _ n. Lift the homotopy
t
= h
t
to a homotopy
H
t
: S
n
S
n
with initial condition . Then H
t
is an odd map and H
1
= g has
the desired property.
We obtain a proof of (10.6.1) from
(10.6.3) Theorem. An odd map has odd degree.
Proof. The proof is by induction on the dimension of the sphere. Let : S
n
S
n
be an odd map. By (10.6.2) we can deform as odd map into a map g such that
g(S
n-1
) S
n-1
. The induction is now a consequence of (10.6.4).
n
S
n
be an odd map such that (S
n-1
) S
n-1
.
Then we have the degree D( ) of and the degree J( ) of its restriction to S
n-1
.
These degrees have the same parity.
Proof. We study the diagram in the proof of (10.5.2) more closely for singular
homology with coefcients in Z. We x a generator :

H
n-1
(S
n-1
) and dene
other generators by
d
= :. s
= n
. l
= n
.
We set i
= u
. The Sum Lemma tells us that u
is a Z-basis of the group

H
n
(S
n
. S
n-1
). Let T : S
n
S
n
be the antipodal map. Suppose given : S
n
S
n
such that (S
n-1
) S
n-1
and T = T . Then we have two degrees D( )
and J( ) dened by
+
(
) = D( )
.
+
(:) = J( ):.
Since u
is a Z-basis we can write

+
(u
) = au
bu
-
. Using this notation,
we show
J( ) = a b. D( ) = a b.
Hence J( ) and D( ) are congruent modulo 2. From
+
: =
+
du
= d
+
u
=
d(au
bu
-
) = (ab): we obtain the rst assertion J( ) = ab. Naturality of
the boundary operator d
-
T
+
= T
+
d
and T
+
: = (1)
n
: imply T
+
:
= (1)
n
:
-
and T
+
u
= (1)
n
u
-
. We conclude
+
u
-
= (1)
n
+
T
+
u
= (1)
n
T
+
+
u
= (1)
n
T
+
(au
bu
-
) = au
-
bu
.
The exactness of (j. d) shows that the image of j is generated by u
u
-
, hence
j(
) = c(u
u
-
) with c = 1. The computation
= s
i
-
:
= l
= c
(u
u
-
) = c
shows c = 1. The computation

D( )(u
) = D( )j
= j
+
=
+
j
=
+
(u
u
-
) = (a b)(u
u
-
)
nally yields the second assertion D( ) = a b.
(10.6.5) Example. The map : S
1
S
1
, : :
2k1
satises the hypothesis of
(10.6.4). We know already that D( ) = 2k 1 and J( ) = 1, hence a = k 1
and b = k. Let J
denote the singular 1-simplex represented by the path from 1

to 1 and J
-
the 1-simplex running from 1 to 1 (both counter-clockwise). Then
J
J
-
is a cycle and
= J
J
-
| is a natural choice of a generator. Then
J
-
represents u
. n
, and :
; and : = 1| 1|. By considering the simplex

J
-
, the relation
+
u
= J
-
| = (k 1)J
-
| kJ
| becomes apparent. For

k = 1 say, J
-
runs counter-clockwise from 1 to 1 to 1 to 1.
Let : S
1
S
1
be any self-map which commutes with the antipodal map.
We can multiply by a constant such that the new map g satises g(1) = 1. From
(10.6.4) we see that g and hence has odd degree, since g has degree 1 on S
0
.
We now apply the BorsukUlam theorem to a problem in combinatorics: The
determination of the chromatic number of the so-called Kneser graphs.
We begin by explaining the problem. Let n| = {1. . . . . n] and denote by N
k
the set of subsets of n| with k elements.
A graph consists of a set 1 of vertices and a set 1 of edges. Each edge has
two boundary points, and they are identied with one or two points in 1. (In other
terms: A graph is a 1-dimensional CW-complex.) The Kneser graph 1G
n,k
has
1 = N
k
. Vertices J
1
. J
2
are connected by an edge if they represent disjoint subsets
of n|.
Let C = k|, and call the elements of C colours. A k-colouring of a graph
(1. 1) is a map : 1 C such that (e
1
) ,= (e
2
) whenever e
1
and e
2
are
connected by an edge. The chromatic number of a graph (1. 1) is the smallest
k such that there exists a k-colouring. The following result was conjectured by
Martin Kneser (1955) [105]. This conjecture was proved by Lovsz [114] with
topological methods. The following ingenious proof was given by Greene [74].
(10.6.6) Theorem. Let k > 0 and n _ 2k 1. Then 1G
n,k
has the chromatic
number n 2k 2.
Proof. An explicit construction shows that the chromatic number is at most
n2k2. We associate to a set J with k elements the colour (J) = min(min(J).
10.7. MayerVietoris Sequences 265
n 2k 2). Suppose (J) = (J
t
) = i < n 2k 2. Then these sets are not
disjoint, since they contain the element i . If their colour is n 2k 2, then they
are contained in {n 2k 2. . . . . n], and then they cannot be disjoint.
Now we come to the topological part. Let J = 2n 2k 1. Choose a set
X S
d
with n elements and such that no hyperplane (through the origin) contains
more than J points of X. We consider the subsets of X with cardinality k as
the vertices of the Kneser graph 1G
n,k
. Suppose there exists a colouring with
J = n 2k 1 elements and we choose one. Let
i
= {. S
d
[ H(.) = {,[(.. ,) > 0] contains a k-tuple in X with colour i ]
and
d1
= S
d
(
1
L L
d
). The sets
1
. . . . .
d
are open and
d1
is
closed. By (10.6.1), one of the sets
i
contains an antipodal pair .. ..
If i _ J, then we have two disjoint k-tuples with colour i , one in H(.) the
other one in H(.). This contradicts the denition of a colouring.
Let i = J 1. Then, by denition of
1
. . . . .
d
, the half-space H(.) contains
at most k 1 points of X, and similarly for H(.). The set S
d
(H(.) LH(.))
is contained in a hyperplane and contains at least n2(k 1) = J 1 points, and
this contradicts the choice of X.
Problems
1. Let n be odd. Then id: S
n
S
n
are homotopic as odd maps.
2. Let . g: S
n
S
n
be odd maps with the same degree. Then they are homotopic as odd
maps.
3. Let J
0
. J
1
. . . . . J
n
be a family of odd integers with J
0
= 1. There exists an odd map
: S
n
S
n
such that (S
i
) S
i
and the map S
i
S
i
induced by has degree J
i
.
10.7 MayerVietoris Sequences
We derive further exact sequences for homology from the axioms, the so-called
MayerVietoris sequences.
Let h
+
be a homology theory. Let (X: . T) be a triad, i.e., . T X = LT.
The triad (X: . T) is said to be excisive for the homology theory if the inclusion
induces an (excision) isomorphism h
+
(. T) h
+
(X. T). The condition is
actually symmetric in and T. We write T = T.
(1) ( L T: . T) is excisive.
(2) ( L T: T. ) is excisive.
(3) | : h
+
(. T) h
+
(T. T) h
+
( L T. T) is an isomorphism.
(4) : h
+
( L T. T) h
+
( L T. ) h
+
( L T. T) is an isomorphism.
Proof. Apply the Sum Lemma (11.1.2) to the diagram
h
+
(. T)
U
U
U
U
U
U
U
U
h
+
( L T. T)
h
+
( L T. T)
i
i
i
i
i
i
i
i
U
U
U
U
U
U
U
U
h
+
(T. T)
i
i
i
i
i
i
i
i
h
+
( L T. ).
The morphisms are induced by the inclusions.
The boundary operator ^ of an excisive triad is dened by
^: h
n
(X) h
n
(X. T)

h
n
(. T)
J
h
n-1
(T).
This operator is part of the MayerVietoris sequence (=MVS) of the triad.
(10.7.2) Theorem. Let (LT: . T) be an excisive triad and C T. Then the
sequence
. . .
Z
h
n
(T. C)
(1)
h
n
(. C) h
n
(T. C)
(2)
h
n
( L T. C)
Z
h
n-1
(T)
(1)

is exact. The inclusions i
: T and i
B
: T T yield the rst map
. (i
+
.. i
B
+
.); and the inclusions
: LT and
B
: T LT yield
the second map (a. b)
+
a
B
+
b. If C = {+] is a point, we obtain the MVS
for reduced homology groups.
There exists another relative MVS. Let (LT: . T) be excisive andLT X.
We dene a boundary operator ^ by
^: h
n
(X. L T)
J
h
n-1
( L T. ) h
n-1
(T. T) h
n-1
(X. T).
. . .
Z
h
n
(X. T)
(1)
h
n
(X. ) h
n
(X. T)
(2)
h
n
(X. L T)
Z
h
n-1
(X. T)
(1)

is exact. The maps (1) and (2) are dened as in (10.7.2).
Proof. The homology sequences of the triples of (T. T. C) and (X. . C) yield
a commutative diagram (a ladder)

h
n
(T. C)

h
n
(. C)

h
n
(. T)

h
n-1
(T. C)

h
n
(T. C)

h
n
(X. C)

h
n
(X. T)

h
n-1
(T. C)

.
We apply (10.7.4) to this diagramand obtain (10.7.2). There exists a similar diagram
which compares the sequences of the triples (X. . T) and (X. L T. T), and
(10.7.4) yields now (10.7.3).
(10.7.4) Lemma. Suppose the following diagram of abelian groups and homomor-
phisms is commutative and has exact rows.

i
(
i
o
i
T
i
b
i
C
i
h
i
c
i
i-1
o
i-1

t
i
(
/
i
T
t
i
/
i
C
t
i
h
/
i

t
i-1
.
Assume moreover that the c
i
are isomorphisms. Then the sequence

i
(-(
i
,o
i
)
T
i

t
i
( b
i
,(
/
i
)

T
t
i
h
i
c
-1
i

/
i

i-1
.
is exact ([17, p. 433]).
We use abbreviations of the type 1 X = 1X, d1 X = d1X, 0 X = 0X.
We associate to a triad (X: . T) the subspace N = N(. T) = 0L1T L1T
of 1 X. Let : N(. T) X be the projection onto the second factor.
(1) The triad (X: . T) is excisive.
(2)
+
: h
+
(N(. T)) h
+
(X) is an isomorphism.
Proof. We have isomorphisms
h
+
(. T) h
+
(T. T) h
+
(0 1T. 0T 1T) h
+
(N. 1T).
by additivity, excision and h-equivalence. It transforms
+
into the map | of item
(3) in (10.7.1). Hence (1) and (2) are equivalent.
The excision axiomsays that the triad is excisive if X =
LT
. The auxiliary
space N(. T) allows us to transfer the problem into homotopy theory: A triad is
excisive for each homology theory, if : N X is an h-equivalence. Recall from
Section 3.3:
(10.7.6) Proposition. Suppose the covering , T of X is numerable. Then is an
h-equivalence.
We now give a second proof of the MVS; it uses the homotopy axiom but not
lemma (10.7.4). Let (X: . T) be a triad. Via excision and h-invariance we see that
the inclusion induces an isomorphism
h
n
((1. d1) T) h
n
(N. 0 L 1 T).
We rewrite the exact sequence of the pair (N. 0L1T). Using the suspension
isomorphism and the additivity, we obtain an exact sequence
h
n
() h
n
(T) h
n
(N) h
n-1
(T) .
If the triad is excisive, we can use (10.7.5) and replace h
+
(N) by h
+
(X). It is an
exercise to compare the boundary operators of the two constructions of the MVS.
There exists a more general MVSfor pairs of spaces. It comprises the previously
discussed cases.
(10.7.7) Theorem. Let (:
0
.
1
) (X: X
0
. X
1
) be two excisive triads. Set
X
01
= X
0
X
1
,
01
=
0

1
. Then there exists an exact MayerVietoris
sequence of the form
h
n
(X
01
.
01
) h
n
(X
0
.
0
) h
n
(X
1
.
1
) h
n
(X. ) .
Proof. Let N(X. ) = 0X
0
L1
01
L1X
1
. The sequence in question arises from a
rewriting of the exact sequence of the triple (N(X). N(X. ). N()). We consider
three typical terms.
(1)
+
: h
+
(N(X). N()) h
+
(X. ), by (10.7.5) and the hypotheses.
(2) The inclusions (X
}
.
}
) { ] (X
}
.
}
) (N(X. ). N()) induce an
isomorphism
h
+
(X
0
.
0
) h
+
(X
1
.
1
) h
+
(N(X. ). N()).
For the proof one excises 1,3. 2,3|
01
and then uses an h-equivalence and
additivity.
(3) The group h
+
(N(X). N(X. )) is isomorphic to h
+
(1X
01
. d1X
01
L
01
)
via inclusion, and the latter via suspension isomorphic to h
+-1
(X
01
.
01
). For the
proof one replaces N(X. ) by the thickened space
0X
0
L 0. 1,4|X
01
L 1
01
L 3,4|X
01
L 1X
1
.
Then one can excise 0X
0
and 1X
1
and use suitable h-equivalences.
It remains to identify the morphisms in the resulting sequence. The map
h
n
(N(X. ). N()) h
n
(N(X). N()) becomes
(
0
+
.
1
+
): h
n
(X
0
.
0
) h
n
(X
1
.
1
) h
n
(X. ).
The map d: h
n1
(N(X). N(X. )) h
n
(N(X. ). N() becomes, with our def-
inition of the suspension isomorphism,
(i
0
+
. i
1
+
): h
n
(X
01
.
01
) h
n
(X
0
.
0
) h
n
(X
1
.
1
).
The boundary operator ^ of the generalized MV-sequence becomes, in the special
cases X = X
0
= X
1
= X
01
and =
0
=
1
=
01
X
01
, the same as in the
previously discussed algebraic derivation of the MV-sequences.
(10.7.8) Example. Let i
1
. i
2
: S
n
S
n
S
n
, . (.. ,
0
), resp. . (.
0
. ,).
Then (i
1
+
. i
2
+
): H
n
(S
n
) H
n
(S
n
) H
n
(S
n
S
n
) is an isomorphism (n _ 1)
with inverse (pr
1
+
. pr
2
+
). We x a generator : H
n
(S
n
) and write :
}
= i
}
+
:. Then
(:
1
. :
2
) is a Z-basis of H
n
(S
n
S
n
). Let = (
1
.
2
): S
n
S
n
S
n
S
n
be a map with bi-degree (a. b) of
1
and bi-degree (c. J) of
2
. Then
+
(:
1
) =
a:
1
c:
2
and
+
(:
2
) = b:
1
J:
2
.
Construct a space, a (2n 1)-manifold, X by identifying in D
n1
S
n
D
n1
S
n
the point (.. ,) S
n
S
n
in the rst summand with (.. ,) in the
second summand via a homeomorphism = (
1
.
2
) of S
n
S
n
as above. The
two summands are embedded as X
1
and X
2
into X. We use the MV-sequence of
(X: X
1
. X
2
) to determine the integral homology of X. Let us consider a portion of
this sequence
H
n
(S
n
S
n
)
}
H
n
(D
n1
S
n
) H
n
(D
n1
S
n
) H
n
(X).
We use the Z-basis (:
1
. :
2
) as above. The inclusion S
n
D
n1
S
n
, . (0. .)
give us as image of : the generators u
1
. u
2
in the summands H
n
(D
n1
S
n
). The
image of the basis elements under is seen to be (:
1
) = cu
2
, (:
2
) = u
1
Ju
2
(we do not use the minus sign for the second summand). We conclude for c ,= 0
that H
n
(X) is the cyclic group of order [c[; the other homology groups of X are in
this case H
0
(X) Z H
2n1
(X) and H
}
(X) = 0 for ,= 0. n. 2n 1. We
leave the case c = 0 to the reader.
Problems
1. Let R
n
be the union of two open sets U and V .
(i) If U and V are path connected, then U V is path connected.
(ii) Suppose two of the sets
0
(U).
0
(V ).
0
(U V ) are nite, then the third set is
also nite and the relation
[
0
(U V )[ ([
0
(U)[ [
0
(V )[) [
0
(U L V )[ = 0
holds.
(iii) Suppose .. , U V can be connected by a path in U and in V . Then they can
be connected by a path in U V .
Can you prove these assertions without the use of homology directly from the denition
of path components?
2. Let the real projective plane 1 be presented as the union of a Mbius band M and a disk D,
glued together along the common boundary S
1
. Determine the groups and homomorphisms
in the MV-sequence of (1: M. D). Do the same for the Klein bottle (1: M. M). (Singular
homology with arbitrary coefcients.)
3. Let (X
1
. . . . . X
n
) be an open covering of X and (Y
1
. . . . . Y
n
) be an open covering of Y .
Let : X Y be a map such that (X
i
) Y
i
. Suppose that the restriction
_
u.
X
u

_
u.
Y
u
of induces a homology isomorphism for each 0 ,= {1. . . . . n]. Then
induces a homology isomorphism.
4. Suppose T and T T are closed cobrations. Then : N(. T) L T is
an h-equivalence.
5. The boundary operators in (10.7.2) and (10.7.3) which result when we interchange the
roles of and T differ from the original ones by 1. (Apply the Hexagon Lemma to the
two boundary operators.)
6. The triad (X d1 L1: X 0 L1. X 1 L1) is always excisive. (Excision
of X 0 L 0. 1,2 and h-equivalence.)
7. Verify the assertions about the morphisms in the sequence (10.7.7).
10.8 Colimits
The additivity axiom for a homology theory expresses a certain compatibility of
homology and colimits (namely sums). We show that this axiom has consequences
for other colimits.
Let (X
-
.
-
) be a sequence X
1
(
1
X
2
(
2
X
3
(
3
. . .
of continuous
maps
}
. Recall that a colimit (a direct limit) of this sequence consists of a space
X and continuous maps
k
: X
k
X with the following universal property:
(1)
k1
k
=
k
.
(2) If a
k
: X
k
Y is a family of maps such that a
k1
k
= a
k
, then there exists
a unique map a: X Y such that a
k
= a
k
.
(This denition can be used in any category.) Let us write
colim(X
-
.
-
) = colim(X
k
)
for the colimit. In the case that the
k
: X
k
X
k1
are inclusions, we can take as
colimit the union X =
_
i
X
i
together with the colimit topology: U X open if
and only if U X
n
open in X
n
for each n.
Colimits are in general not suitable for the purpose of homotopy theory, one has
to weaken the universal property up to homotopy. We will construct a so-called
homotopy colimit. Colimits of sequences allowa special and simpler treatment than
general colimits. A model of a homotopy colimit in the case of sequences is the
telescope. We identify in
i
X
i
i. i 1| the point (.
i
. i 1) with (
i
(.
i
). i 1)
for .
i
X
i
. Denote the result by
T = T(X
-
.
-
) = hocolim(X
-
.
-
).
We have injections
k
: X
k
T , . (.. k) and a homotopy k
k
:
k1
k
.
k
,
a linear homotopy in X
k
k. k 1|. Thus the telescope T consists of the mapping
cylinders of the maps
i
glued together. The data dene the homotopy colimit of
the sequence.
10.8. Colimits 271
Given maps a
k
: X
k
Y and homotopies h
k
: X
k
k. k 1| Y from
a
k1
k
to a
k
. Then there exists a map a: T Y such that
k
a = a
k
, and the
composition of the canonical map X
k
k. k 1| T with a is h
k
.
We have subspaces T
k
T , the image of

k-1
i=1
X
i
i. i 1| X
k
{k].
The canonical inclusion |
k
: X
k
T
k
is a homotopy equivalence (compare the
analogous situation for a mapping cylinder).
In homology we have the equality
k
+
=
k1
+

k
+
: h
n
(X
k
) h
n
(X
k1
)
h
n
(T ). By the universal property of the colimit of groups we therefore obtain a
homomorphism
| : colimh
n
(X
k
) h
n
(T(X
-
.
-
)).
(10.8.1) Theorem. In an additive homology theory | is an isomorphism.
Proof. We recall an algebraic construction of the colimit
1
o
1

2
o
2
. . .
of
abelian groups. Consider
k_1
k_1
k
. (.
k
) (.
k1
a
k
(.
k
)).
The cokernel is the colimit, together with the canonical maps (inclusion of the -th
summand composed with the projection)
}

k

k
colim
k
. We therefore
need a computation of h
n
(T ) which has this form. We cover T by the subspaces
= T
_
i_1
X
2i
{2i
1
2
]. T = T
_
i_1
X
2i-1
{2i
1
2
].
(10.8.2) Lemma. The inclusions
i_1
X
2i
{2i ] .
i_1
X
2i-1
{2i 1] T.
i_1
X
i
{i ] T
are h-equivalences, and (. T) is a numerable covering of T .
Because of this lemma we have a MayerVietoris sequence
h
n
( T)

h
n
() h
n
(T)

h
n
(T )
h
n
(X
}
)

}=0(2)
h
n
(X
}
)
}=1(2)
h
n
(X
}
).
The map has the form

(.
2i
) = (.
2i
.
2i
+
(.
2i
)) and (.
2i1
) = (
2i1
+
(.
2i1
). .
2i1
)
for .
}
h
n
(X
}
). We see that is injective; therefore we can obtain h
n
(T ) as
cokernel of . The automorphism (.
i
) ((1)
i
.
i
) transforms into the map
which was used in the algebraic denition of the colimit.
For applications we have to nd conditions under which the homotopy colimit
is h-equivalent to the colimit. We consider the case that the
k
: X
k
X
k1
are
inclusions, and denote the colimit by X =
_
k
X
k
. We change the denition of the
telescope slightly and consider it now as the subspace
T =
_
k
X
k
k. k 1| X 0. o .
The topology of T may be different, but the proof of (10.8.1) works equally well
in this case. The projection onto the rst factor yields : T X.
(10.8.3) Example. Let z: X 1. o be a function such that
s = (id. z): X X 1. o
has an image in T . Then s : X T is a section of . The composition s is
homotopic to the identity by the homotopy ((.. u). t ) (.. (1 t )u t z(.)).
This is a homotopy over X, hence is shrinkable. The property required by z
amounts to z(.) < i =. X
i-1
for each i .
Let (U
i
[ i N) be a numerable covering of X with locally nite numeration
(t
i
) (see the chapter on partitions of unity). Set X
k
=
_
k
i=1
U
i
and z =
o
i=1
i t
i
.
Then z(.) < i implies . X
i-1
.
(10.8.4) Proposition. Suppose the inclusions X
k
X
k1
are cobrations. Then
T X 1. o is a deformation retract.
Proof. Since X
k
X is a cobration, there exists by (5.1.3) a homotopy
h
k
t
: X 1. oX 1. o rel Y
k
= X
k
1. o LX k 1. o
from the identity to a retraction X 0. o Y
k
. The retraction 1
I
acts as
the identity on X
k
1. o for l > k, and therefore the innite composition
1
}
= 1
}2
1
}1
1
}
is a well-dened continuous map. Fromh
}
we obtain
a homotopy 1
}
. 1
}1
relative to Y
}
. We can concatenate these homotopies and
obtain a homotopy from the retraction 1
1
to the identity relative to T .
Problems
1. Let T be a subring. Find a system of homomorphisms Z Z such that the
colimit is T .
2. Let S
n
S
n
S
n
be a sequence of maps where each map has degree two. Let
X be the homotopy colimit. Show that
n
(X) Z1,2|, the ring of rational numbers with
denominators a power of two. What system of maps between S
n
would yield a homotopy
colimit Y such that
n
(Y ) Q?
3. Let X be a CW-complex and T the telescope of the skeleton ltration. Then the inclusion
T X 1. o induces isomorphisms of homotopy groups and is therefore a homotopy
equivalence. One can also apply (10.8.4) in this case.
10.9. Suspension 273
10.9 Suspension
Recall the homological suspension isomorphism 10.2.5. We use abbreviations of
the type 1 = 1, d1 = d1, 0 X = 0. We set k
n
(. T) =
h
n
((1. d1) (. T)). The k
+
() are the data for a new homology theory. The
boundary operator of this homology theory is dened for a triple (. T. C) by
d: h
n1
(1. d1 L 1T)
J
h
n
(d1 L T. d1 L 1C)

h
n
(1T. d1T L 1C).
In order to work with this denition we use
(10.9.1) Lemma. For each triple (. T. C) the triad (d1 L 1T: 1T. d1 L 1C)
is excisive.
Proof. The inclusion induces an isomorphism
h
n
(d1. d1T) h
n
(d1 L 1C. d1T L 1C).
excise
1
2
C and use a homotopy equivalence. If we use this also for T = C we
conclude that h
n
(d1L1C. d1T L1C) h
n
(d1L1T. 1T) is an isomorphism.
The exact sequence of the triple (1. d1 L 1T. d1) is transformed with
the isomorphism k
n
(T) = h
n
(1T. d1T) h
n
(1T L d1. d1) into the exact
sequence of (. T) for the k
+
-groups. Let U T and

U T
. The excision
isomorphism for the k
+
-theory claims that
h
n
(1 U). 1(T U) L d1( U)) h
n
(1. 1T L d1)
is an isomorphism. This is a consequence of 1U = 1

U (1T L d1)
and the
usual excision isomorphism. The isomorphisms o
(,B)
: h
n
(. T) k
n1
(. T)
dened in (10.2.5) form a natural transformation. The next proposition says that
they are natural transformations of homology theories of degree 1.
(10.9.2) Proposition. For each triple (. T. C) the diagram
h
n1
(. T)
J
c
(.,!)
h
n
(T. C)
-c
(!,C)
k
n2
(. T)
k
n1
(T. C)
is commutative.
Proof. By naturality it sufces to consider the case C = 0. The Hexagon Lemma
shows = for the maps
: h
n1
(1. 1T L d1) h
n
(1T L d1. d1)
h
n
(1T L 0. 0 L 1T) h
n-1
(0 L 1T. 0).
: h
n1
(1. 1T L d1) h
n
(1T L d1. 1T L 0)
h
n
(d1. 0 L 1T) h
n-1
(0 L 1T. 0).
and the center group h
n
(1T L d1. 1T L 0). Let be the isomorphism
: h
n-1
(1T L 0. 0)

h
n-1
(1T) h
n-1
(T).
By diagram chasing one veries o
(B)
=

d and o
(,B)
= d.
(10.9.3) Lemma. Let (. T. C) be a triple. Then we have an isomorphism
(|
1
. |
0
. | ): h
n
(. T)h
n
(. T)h
n
(1T. d1TL1C) h
n
(d1L1T. d1TL1C).
Here | is induced by the inclusion, and |
by a (v. a).
(10.9.4) Proposition. For each triple (. T. C) the diagram
h
n
(. T)
c
(.,!)
h
n1
(1. d1 L 1T)
J
h
n
(. T) h
n
((. T) h
n
(1T. d1T L 1C)

h
n
(d1 L 1T. d1T L 1C)
is commutative; here (.) = (.. .. o
(B,C)
d.), and the isomorphism is taken
from (10.9.3).
Proof. The assertion about the third component of follows from (10.9.2). The
other components require a little diagram chasing. For the verication it is helpful
to use the inverse isomorphism of given by the procedure of (10.7.1). The minus
sign in the second component is due to the fact that the suspension isomorphism
changes the sign if we interchange the roles of 0. 1, see 10.2.5.
Chapter 11
Homological Algebra
In this chapter we collect a number of algebraic denitions and results which are
used in homology theory. Reading of this chapter is absolutely essential, but it
only serves practical purposes and is not really designed for independent study.
Homological Algebra is also the name of a mathematical eld and the reader
may wish to look into the appropriate textbooks.
The main topics are diagrams and exact sequences, chain complexes, derived
functors, universal coefcients and Knneth theorems. We point out that one can
imitate a lot of homotopy theory in the realm of chain complexes. It may be helpful
to compare this somewhat simpler theory with the geometric homotopy theory.
11.1 Diagrams
Let 1 be a commutative ring and denote by 1- MOD the category of left 1-
modules and 1-linear maps. (The category ABEL of abelian groups can be iden-
tied with Z- MOD.) Recall that a sequence of 1-modules and 1-linear maps

i1
o
i1

i
o
i

i-1
is exact at
i
if Im(a
i1
) = Ker(a
i
) and
exact if it is exact at each
i
. The language of exact sequences is a convenient way
to talk about a variety of algebraic situations.
(1) 0
o
T exact = a injective. (We also use T for an injective
homomorphism.)
(2) T
b
C 0 exact =b surjective. (We also use T C for a surjective
homomorphism.)
(3) 0
o
T
b
C 0 exact =a is injective, b is surjective, b induces
an isomorphism of the cokernel of a with C.
(4) Let
(1)
T
(2)
C
(3)
D be exact. Then the following are equivalent:
(1) surjective = (2) zero = (3) injective.
An exact sequence of the form (3) is called a short exact sequence. It sometimes
happens that in a longer exact sequence every third morphism has the property (1)
(or (2), (3)). Then the sequence can be decomposed into short exact sequences.
Note that the exact homotopy sequences and the exact homology sequences have
period 3.
A family (M
}
[ J) of modules has a direct sum

}J
M
}
, the sum
in the category 1- MOD, and a product

}J
M
}
, the product in the category
276 Chapter 11. Homological Algebra
1- MOD. The underlying set of the product is the Cartesian product of the under-
lying sets, consisting of all families (.
}
M
}
[ J); the 1-module structure
is given by pointwise addition and scalar multiplication. The canonical projection
k
:

}
M
}
M
k
onto the k-th factor is part of the categorical product struc-
ture. The sum

}
M
}
is a submodule of the product and consists of all families
(.
}
) where all but a nite number of .
}
are zero. We have the canonical injection
i
k
: M
k

}
M
}
, dened by
k
i
k
= id and
k
i
I
= 0 for k ,= l. If
}
: M
}
N
is a family of 1-linear maps, then
(
}
):

M
}
N
denotes the morphism determined by (
}
) i
k
=
k
. If g
}
: 1 M
}
is a family
of 1-linear maps, then
(g
}
): 1
}
M
}
denotes the morphism determined by
k
(g
}
) = g
k
.
Let (
}
[ J) be a family of submodules of M. Then

}

}
denotes the
submodule generated by
_
}

}
. We say that M is the internal direct sum of the
}
if the map
}

}
M. (a
}
)
}
a
}
is an isomorphism. In that case we also write M =
}

}
. A submodule M
is called a direct summand of M if there exists a complement of , i.e., a submodule
T such that M is the internal direct sum of and T.
We assume known the structure theory of nitely generated abelian groups. An
element of nite order in an abelian group is called a torsion element. The torsion
elements form a subgroup, the torsion subgroup T(). The torsion subgroup of
,T() is trivial. If the group is nitely generated, then the torsion subgroup has
a complement J, and J is a free abelian group. The cardinality of a basis of J is
called the rank of . A nitely generated torsion group is the direct sum of cyclic
groups of prime power order Z,(
k
), and the number of factors isomorphic to
Z,(
k
) is uniquely determined by the group. A similar structure theory exists for
nitely generated modules over principal ideal domains.
A linear map : M M with the property = is called a projection
operator on M.
(11.1.1) Splitting Lemma. Let 0 1
(
J
G 0 be a short exact
sequence of modules. Then the following assertions are equivalent:
(1) The image of is a direct summand of J.
(2) There exists a homomorphism r : J 1 such that r = id.
(3) There exists a homomorphism s : G J such that gs = id.
If (1)(3) holds, we say the sequence splits. We then call r and s splittings, r a
retraction of , s a section of g. In case (2) r is a projection operator, hence we
11.1. Diagrams 277
have J = Im( r) Ker( r) = Im( ) Ker(r). In case (3) sg is a projection
operator, hence we have J = Im(sg) Ker(sg) = Im(s) Ker(g).
(11.1.2) Sum Lemma. Suppose given a commutative diagram in 1- MOD:
T
1
T
2
C
}
2
K
K
K
K
K
K
}
1
s
s
s
s
s
s
1
i
1
s
s
s
s
s
s
o
1
2
.
i
2
K
K
K
K
K
K
o
2
Assume
k
i
k
= 0 for k = 1. 2.
(1) If the a
k
are isomorphisms and (i
2
.
2
) is exact, then
(i
1
. i
2
):
1
2
C and (
2
.
1
): C T
2
T
1
are isomorphisms and (i
1
.
1
) is exact.
(2) If (i
1
. i
2
) is an isomorphismand (i
2
.
2
) is short exact, then a
1
is an isomor-
phism (
1
. a
2
are not needed). If (
2
.
1
) is an isomorphism and (i
1
.
1
) is short
exact, then a
1
is an isomorphism (i
2
. a
2
are not needed).
Proof. (1) The hypothesis implies (
2
.
1
) (i
1
. i
2
) = a
1
a
2
. We show that
(i
1
. i
2
) is surjective. Given c C we have
2
(c i
1
a
-1
1

2
(c)) = 0, by commuta-
tivity. Hence there exists by exactness .
2

2
such that c i
1
a
-1
1

2
(c) = i
2
(.
2
),
i.e., c is contained in the image of (i
1
. i
2
).
Let
1
(c) = 0andwrite c = i
1
.
1
i
2
.
2
. Then0 =
1
(c) =
1
i
1
.
1
1
i
2
.
2
=
1
i
2
.
2
= a
2
(.
2
), hence .
2
= 0 and c Im(i
1
).
(2) Exercise.
(11.1.3) Hexagon Lemma. Given a commutative diagram of abelian groups.
G
0
I
1
_r
r
r
r
r
r I
2
L
L
L
L
L
L
i
0
G
t
1
G
t
2
G
}
1
M
M
M
M
M
M
}
2

q
q
q
q
q
q
}
0
G
1
i
1
q
q
q
q
q
q
k
1
h
1

L
L
L
L
L
L
G
2
i
2
M
M
M
M
M
M
k
2
h
2
_r
r
r
r
r
r
G
t
0
Suppose that k
1
. k
2
are isomorphisms, (i
1
.
2
) exact, (i
2
.
1
) exact, and
0
i
0
= 0.
Then h
1
k
-1
1
l
1
= h
2
k
-1
2
l
2
.
Proof. The part i. . k satises the hypothesis of the Sum Lemma (11.1.2). Given
. G
0
there exist .
}
G
}
such that i
0
. = i
1
.
1
i
2
.
2
. We compute
0 =
0
i
0
. =
0
i
1
.
1
0
i
2
.
2
= h
1
.
1
h
2
.
2
.
l
1
. =
1
i
0
. =
1
i
1
.
1
i
2
. =
1
i
1
. = k
1
.
1
.
hence .
1
= k
-1
1
l
1
. and similarly .
2
= k
-1
2
l
2
..
(11.1.4) Five Lemma. Given a commutative diagram of groups and homomor-
phisms with exact rows:
C
;
1
e
t

/

T
t

/
C
t
;
/
D
t

/

1
t
.
(1) a surjective, b. J injective =c injective. (Here the 1-part of the diagram is
not needed.)
(2) b. J surjective, e injective =c surjective. (Here the -part of the diagram
is not needed.)
(3) a surjective, b. J bijective, e injective =c bijective.
Proof. For another proof see (11.2.7). We give here a direct proof by the method
called diagramchasing. One refers to diagramchasing whenever the proof (chasing
elements through the diagram) does not really require a mathematical idea, only
careful patience.
(1) Let c(n) = 0. Then ;
t
c(n) = J;(n) = 0, and injectivity of J shows
;(n) = 0. By exactness, () = n for some . Since
t
b() = c() = 0, we
have
t
(u
t
) = b() for some u
t
, by exactness, and a(u) = u
t
by surjectivity of
a. By injectivity of b and commutativity we see (u) = and hence by exactness
n = () = 0.
(2) Given n
t
C
t
. Choose . such that J(.) = ;
t
(c
t
). By exactness,
commutativity, and injectivity of e, we see (.) = 0 and hence ;(n) = . for
some n. By commutativity, c(n) and n
t
have the same image under ;
t
. Hence
n
t
= c(n)
t
(
t
) for some
t
. Then c(n ()) = c(n) c() = c(n)
t
b() =
c(n)
t
(
t
) = n
t
, i.e., n
t
is contained in the image of c.
(3) A consequence of (1) and (2).
11.2. Exact Sequences 279
Problems
1. Let be a projection operator on M. Then 1 is a projection operator. The equalities
Im(1 ) = Ker() and Ker(1 ) = Im() hold. Moreover M = Im() Im(1 ).
The submodule of M is a direct summand if and only if there exists a projection operator
with image .
2. Let (
[ J) be a nite family of modules. Suppose given linear maps i

k
:
k

and
l
:
l
such that
k
i
k
= id and
k
i
l
= 0 for k ,= l (we write
k
i
l
=
kl
in this
case). Then (
k
) (i
k
) = id and (i
k
) (
k
) =
is a projection operator. Hence

the following are equivalent: (1) (i
k
) is an isomorphism. (2) (
k
) is an isomorphism. (3)
= id.
3. Let be a prime number. Determine the number of subgroups of Z,(
k
) Z,(
l
).
4. Consider the group Z,(6)Z. Determine the subgroups of index 2. 3. 4. 5. 6. Determine
the number of complements of the torsion subgroup.
5. Let be a nitely generated abelian group. Then
Z
Q is a Q-vector space. Show
that its dimension is the rank of .
6. Let M
, J be submodules of M. The following assertions are equivalent:

(1)

is the direct sum of the M
.
(2) For each i J, M
i

,=i
M
= {0].
(3) Suppose
= 0, .
, almost all .
= 0, then .
= 0 for each J.
11.2 Exact Sequences
We start with a commutative diagram of modules.
T
b
C
;
t
o
/
T
t
b
/
C
t
It yields two derived diagrams (Ke = kernel, Ko = cokernel, Im = Image).
Ke()
o
(1)
Ke()
b
Ke(;)
Ke(a
t
)
o
Ke()
b
Ke(;) Im(b)
(2)
Ko()
o
/

(3)
Ko()
b
/

Ko(;)
t
Im() Ke(a
t
)
o
/

T
t
Im()
b
/

=
C
t
Im(;b)
(4)

The morphisms named a. b. a
t
. b
t
are induced by the original morphisms with the
same name by applying them to representatives. (1) and (2) are inclusions, (3) and
(4) are quotients.
(11.2.1) Proposition. Let (a. b) and (a
t
. b
t
) be exact. Then a connecting morphism
: Ke(;) Im(b)

t
Im() Ke(a
t
)
is dened by the correspondence (a
t
)
-1
b
-1
.
Proof. For : Ke(;) Im(b) there exists , T such that b(,) = :; since :
Ke(;) and ;b = b
t
we have (,) Ke(b
t
); since Ke(b
t
) Im(a
t
), there exists
.
t

t
such that a
t
(.
t
) = (,). We set (:) = .
t
and show that this assignment
is well-dened. If , T, b( ,) = :, then b(, ,) = 0; since Ke(b) Im(a),
there exists . such that a(.) = , ,. We have (,) ( ,) = a
t
(.),
because of a
t
= a, and with a
t
( .
t
) = ( ,) we obtain a
t
(.
t
.
t
(.)) = 0,
i.e., .
t
.
t
mod Im() Ke(a
t
) .
We add further hypotheses to the original diagram and list the consequences
for the derived diagrams. We leave the verication of (11.2.2), (11.2.3), (11.2.4),
(11.2.5) to the reader.
11.2.2 If a
t
is injective, then (1) and (3) are bijective. If b is surjective, then (2)
and (4) are bijective.
11.2.3 Let (a
t
. b
t
) be exact. Then
t
Im() 1e(a
t
)
o
/
T
t
Im()
b
/
C
t
Im(;b)
is exact. If, moreover, b is surjective, then (4) is bijective and therefore
Ko()
o
/
Ko()
b
/
Ko(;)
is exact. If b
t
is surjective, then the derived b
t
is surjective too.
11.2.4 Let (a. b) be exact. Then
Ke(a
t
)
o
Ke()
b
Ke(;) Im(b)
is exact. If, moreover, a
t
is injective, then (1) is bijective and therefore
Ke()
o
Ke()
b
Ke(;)
11.2. Exact Sequences 281
is exact. If a is injective, then the derived a is injective too.
11.2.5 Let (a. b) and (a
t
. b
t
) be exact. Then, as we have seen, is dened. Under
these assumptions the bottom lines of the derived diagrams together with yield
an exact sequence. (See the special case (11.2.6).)
(11.2.6) KernelCokernel Lemma. If in the original diagrama
t
is injective and b
surjective, then (1). (2). (3), and (4) are bijective and the kernel-cokernel-sequence
Ke()
o
Ke()
b
Ke(;)

Ko()
o
/
Ko()
b
/
Ko(;)
is exact.
Proof. We show the exactness at places involving ; the other cases have already
been dealt with. The relations b = 0 and a
t
= 0 hold by construction.
If the class of .
t
is contained in the kernel of a
t
, then there exists , such that
a
t
(.
t
) = (,). Hence : = b(,) Ke(;) by commutativity, and (:) = .
t
.
Suppose : Ke(;) is contained in the kernel of . Then there exists , such
that : = b(,). (,) = a
t
(.
t
) and (:) = (.) Im(). Then b(, a(.)) = :
and (, a(.)) = (,) a(.) = (,) a
t
(.) = (,) a
t
(.
t
) = 0. Hence
, a(.) is a pre-image of :.
We nowrelate the KernelCokernel Lemma to the Five Lemma (11.2.7); see also
(11.1.4). Given a commutative ve-term diagram of modules and homomorphisms
with exact rows.
C
;
1
e
t

/

T
t

/
C
t
;
/
D
t

/

1
t
We have three derived diagrams.
0

Ke()
D
d
1
e
T

b
Ko()

0
0

Ke(
t
)

D
t

/

1
t
,

t

/

T
t
Ko(
t
)

0,
0

Ko()

C
;
Ke()

0
0

Ko(
t
)

/
C
t
;
/
Ke(
t
)

0.
The rows of the rst two diagrams are exact for trivial reasons. The exactness of the
rows of the original diagram implies that the third diagram has exact rows. From
the considerations so far we obtain the exact sequences
Ke(e) Im() Ko(
J) Ko(J).
Ke(b) Ke(
b)
t
,(Im(a) Ke(
t
)).
0 Ke(
b) Ke(c) Ke(
J) Ko(
b) Ko(c) Ko(
J) 0.
This yields:
(11.2.7) Five Lemma. Given a ve-term diagram as above. Then the following
holds:
(1) Ke(
b) = 0. Ke(
J) = 0 = Ke(c) = 0.
(2) Ko(
b) = 0. Ko(
J) = 0 = Ko(c) = 0.
(3) Ke(b) = 0.
t
,(Im(a) Ke(
t
)) = 0 = Ke(
b) = 0.
(4) Ke(e) Im() = 0. Ko(J) = 0 = Ko(
J) = 0.
(5) a surjective, b. J injective = c injective. (Here the 1-part of the diagram
is not needed.)
(6) b. J surjective, e injective = c surjective. (Here the -part of the diagram
is not needed.)
(7) a surjective, b. J bijective, e injective = c bijective.
Problems
1. Given homomorphisms : T and g: T C between 1-modules. Then there is
a natural exact sequence
0 Ke( ) Ke(g ) Ke(g) Ko( ) Ko(g ) Ko(g) 0.
The connection to the previous considerations: The commutative diagram with exact rows
0

(1, )
T
{ -,1)

1
0
0

T
(,1)
C T
{ -1,)

C

0
can be viewed as an exact sequence of chain complexes. Its homology sequence (11.3.2) is
the desired sequence, if we identify the kernel and cokernel of g 1 with the corresponding
modules for g .
Describe the morphisms of the sequence and give also a direct proof.
11.3. Chain Complexes 283
11.3 Chain Complexes
The algebraic terminology of chain complexes arose from the denition of ho-
mology groups. Since then it also has become of independent interest in algebra
(homological algebra). The construction of (singular) homology proceeds in two
stages: First one associates to a space a so-called chain complex. Then the chain
complex yields, by algebra, the homology groups. The category of chain complexes
and chain maps has an associated homotopy structure.
We work in this section with the category 1- MOD of left modules over some
xed ring 1. A family
+
= (
n
[ n Z) of modules
n
is called a Z-graded
module. We call
n
the component of degree or dimension n. One sometimes
considers the direct sum

nZ
n
; then the elements in
n
are said to be ho-
mogeneous of degree n. Typical examples are polynomial rings; if k.. ,| is the
polynomial ring in two indeterminates .. , of degree 1 say, then the homogeneous
polynomials of degree n are spanned by .
i
,
n-i
for 0 _ i _ n, and in this manner
we consider k.. ,| as a graded k-module (actually a graded algebra, as dened
later). One can also consider formal power series; this would correspond to taking
the product
nZ
n
instead of the sum.
Let
+
and T
+
be Z-graded modules. A family
n
:
n
T
nk
of homomor-
phisms is called a morphism of degree k between the graded modules.
A sequence C
-
= (C
n
. d
n
[ n Z) of modules C
n
and homomorphisms
d
n
: C
n
C
n-1
, called boundary operators or differentials, is said to be a chain
complex, if for each n Zthe boundary relation d
n-1
d
n
= 0 holds. We associate
to a chain complex C
-
the modules
7
n
= 7
n
(C
-
) = Ker(d
n
: C
n
C
n-1
).
T
n
= T
n
(C
-
) = Im(d
n1
: C
n1
C
n
).
H
n
= H
n
(C
-
) = 7
n
,T
n
.
We call C
n
(7
n
, T
n
) the module of n-chains (n-cycles, n-boundaries) and H
n
the n-th homology module of the chain complex. (The boundary relation dd = 0
implies T
n
7
n
, and therefore H
n
is dened.) Two n-chains whose difference is
a boundary are said to be homologous. Often, in particular in the case 1 = Z, we
talk about homology groups.
Let C
-
= (C
n
. c
n
) and D
-
= (D
n
. J
n
) be chain complexes. A chain map
-
: C
-
D
-
is a sequence of homomorphisms
n
: C
n
D
n
which satisfy the
commutation rules J
n

n
=
n-1
c
n
. A chain map induces (by restriction
and passage to the factor groups) homomorphisms of the cycles, boundaries, and
homology groups
7
n
(
-
): 7
n
(C
-
) 7
n
(D
-
).
T
n
(
-
): T
n
(C
-
) T
n
(D
-
).
+
= H
n
(
-
): H
n
(C
-
) H
n
(D
-
).
A (short) exact sequence of chain complexes
0 C
t
(
C
C
tt
0
consists of chain maps and g such that 0 C
t
n
(
n
C
n
n
C
tt
n
0 is exact
for each n.
We certainly have the induced morphisms H
n
( ) and H
n
(g). Moreover, there
exists a connecting morphism d
n
: H
n
(C
tt
) H
n-1
(C
t
), also called boundary
operator, which is induced by the correspondence
-1
n-1
J
n
g
-1
n
.
C
n
d
n
C
tt
n
:
tt
:
t
C
t
n-1
(
n-1
C
n-1
(11.3.1) Lemma. For a cycle :
tt
C
tt
n
with pre-image : under g
n
the relation
g
n-1
J
n
: = J
tt
n
g
n
: = J
tt
n
:
tt
= 0 and exactness shows that there exists :
t
with
J
n
(:) =
n-1
(:
t
). The assignment :
tt
:
t
induces a well-dened homomorphism
d
n
: H
n
(C
tt
) H
n-1
(C
t
).
Proof. The relation
n-2
J
t
n-1
:
t
= J
n-1
n-1
:
t
= J
n-1
J
n
: = 0andthe injectivity
of
n-2
show that :
t
is a cycle. If we choose another pre-image :
n
n
t
of :
tt
,
then we have to replace :
t
by :
t
J
t
n
n
t
, so that the homology class of :
t
is well-
dened. Finally, if we change :
tt
by a boundary, we can replace : by the addition
of a boundary and hence J
n
: does not change.
(11.3.2) Proposition. The sequence
H
n
(C
t
)
(
H
n
(C)
H
n
(C
tt
)
J
n
H
n-1
(C
t
)
is exact.
Proof. The boundary operator J
n
: C
n
C
n-1
induces a homomorphism
J
n
: 1
n
= C
n
,T
n
7
n-1
.
and its kernel and cokernel are H
n
and H
n-1
. By (11.2.3) and (11.2.4) the rows of
the next diagram are exact.
1
t
n
(
n
d
/
n
1
n
d
n
1
tt
n

d
//
n
0
0

7
t
n-1
(
n-1
7
n-1
n-1
7
tt
n-1
The associated sequence (11.2.6) H
t
n
H
n
H
tt
n
J
H
t
n-1
H
n-1
H
tt
n-1
is the exact homology sequence.
11.4. Cochain complexes 285
Let . g: C = (C
n
. c
n
) D = (D
n
. J
n
) be chain maps. A chain homotopy
s from to g is a sequence s
n
: C
n
D
n1
of homomorphisms which satisfy
J
n1
s
n
s
n-1
c
n
= g
n

n
.
(This denition has two explanations; rstly, one can dene chain homotopy in
analogy to the topological denition by using the chain complex analogue of the
unit interval; secondly, it codies the boundary relation of a geometric homotopy.)
We call and g homotopic or chain homotopic, if there exists a chain homotopy
s from to g, in symbols s : . g. Chain homotopic is an equivalence
relation on the set of chain maps C D; the data s : . g and t : g . h
imply (s
n
t
n
): . h. This relation is also compatible with composition; if
s : .
t
: C D and t : g . g
t
: D 1, then (g
n1
s
n
): g . g
t
and
(t
n
n
): g . g
t
. We call : C D a chain equivalence, if there exists a
chain map g: D C and chain homotopies g . id and g . id.
(11.3.3) Proposition. Chain homotopic maps induce the same morphisms between
the homology groups.
Proof. Let . C
n
be a cycle. The homotopy relation g
n
(.)
n
(.) = J
n1
s
n
(.)
shows that
n
(.) and g
n
(.) are homologous.
11.4 Cochain complexes
Let C
-
= (C
n
. d
n
) be a chain complex of 1-modules. Let G be another 1-module.
We apply the functor Hom
T
(. 1) to C
-
and obtain a chain complex C
-
= (C
n
.
n
)
of 1-modules with C
n
= Hom
T
(C
n
. 1) and the 1-linear map
n
: C
n
= Hom
T
(C
n
. 1) Hom
T
(C
n1
. 1) = C
n1
dened by
n
() = (1)
n1
d
n1
for Hom(C
n
. 1).
For the choice of this sign see 11.7.4. The reader will nd different choices of
signs in the literature. Other choices will not effect the cohomology functors. But
there seems to be an agreement that our choice is the best one when it comes to
products.
Now some co terminology. A cochain complex C
-
= (C
n
.
n
) is a Z-
graded module (C
n
[ n Z) together with homomorphisms
n
: C
n
C
n1
,
called coboundary operators or differentials
1
, such that
n1
n
= 0. We set
7
n
= Ker
n
. T
n
= Im
n-1
. H
n
= 7
n
,T
n
and call C
n
. 7
n
. T
n
the module of n-cochains, n-cocycles, n-coboundaries, and
H
n
the n-th cohomology module of the cochain complex.
1
An important cochain complex arises from the exterior differentiation of differential forms. So one
should not use a co word here.
11.5 Natural Chain Maps and Homotopies
Let C be an arbitrary category and CH
the category of chain complexes (C

n
. c
n
) of
abelian groups with C
n
= 0 for n < 0 and chain maps. A functor J
+
: C CH
consists of a family of functors J

n
: C Z- MOD and natural transformations
J
T
n
: J
n
J
n-1
such that J
T
n-1
J
T
n
= 0. Anatural transformation
+
: J
+
G
+
between such functors is a family of natural transformations
n
: J
n
G
n
such
that J
G
n

n
=
n-1
J
T
n
. A natural chain homotopy s
+
:
+
. [
+
from
+
to [
+
is
a family s
n
: J
n
G
n1
of natural transformations such that
J
G
n1
s
n
s
n-1
J
T
n
= [
n

n
.
A functor J
n
: C Z- MOD is called free if there exists a family ((T
n,}
. b
n,}
) [
J(n)) of objects T
n,}
of C (called models) and elements b
n,}
J
n
(T
n,}
) such
that
J
n
( )(b
n,}
). J(n). Hom
C
(T
n,}
. X)
is for each object X of C a Z-basis of J
n
(X). A natural transformation
n
: J
n

G
n
from a free functor J
n
into another functor G
n
is then determined by the values
n
(b
n,}
) and the family of these values can be xed arbitrarily in order to obtain a
natural transformation. We call J
+
free if each J
n
is free. We call G
+
acyclic (with
respect to the families of models for J
+
) if the homology groups H
n
(G
+
(T
n,}
)) = 0
for n > 0 and each model T
n,}
.
(11.5.1) Theorem. Let J
+
be a free and G
+
be an acyclic functor. For each natural
transformation : H
0
J
0
H
0
G
0
there exists a natural transformation
+
: J
+
G
+
which induces . Any two natural transformations and [ with
this property are naturally chain homotopic ([57]).
Proof. We specify a natural transformation
0
by the condition that (b
0,}
) rep-
resents the homology class b
0,}
|. Let now
i
: J
i
G
i
be natural transfor-
mations (0 _ i < n) such that J
G
i

i
=
i-1
J
T
i
for 0 < i < n. Consider the
elements
n-1
J
T
n
b
n,}
G
n-1
(T
n,}
). For n = 1 this element represents 0 in
H
0
, by the construction of
0
. For n > 1 we see from the induction hypothesis
that J
G
n-1
n-1
J
T
n
b
n,}
=
n-2
J
T
n-1
J
T
n
b
n,}
= 0. Since G
+
is acyclic we nd
g
n,}
G
n
(T
n,}
) such that J
G
n
g
n,}
=
n-1
J
T
n
b
n,}
. We specify a natural trans-
formation by the conditions (b
n,}
) = g
n,}
. This transformation then satises
J
G
n

n
=
n-1
J
T
n
. This nishes the induction step.
Let now
+
and [
+
be given. Then [
0
(b
0,}
)
0
(b
0,}
) = J
G
1
c
0,}
for some
c
0,}
, since [
0
(b
0,}
) and
0
(b
0,}
) represent the same homology class. We dene
the transformation s
0
: J
0
G
1
by the condition s
0
(b
0,}
) = c
0,}
. Suppose now
that s
n
: J
n
G
n1
are given such that J
G
i1
s
i
s
i-1
J
T
i
= [
i
i
for 0 _ i < n
11.6. Chain Equivalences 287
(and s
-1
= 0). We compute with the induction hypothesis
J
G
n
([
n

n
s
n-1
J
T
n
)
= [
n-1
J
T
n

n-1
J
T
n
([
n-1
J
T
n

n-1
J
T
n
s
n-2
J
T
n-1
)J
T
n
= 0.
Thus, by acyclicity, we can choose c
n,}
G
n1
(T
n,}
) such that
J
G
n1
c
n,}
= ([
n

n
s
n-1
J
T
n
)(b
n,}
).
We now specify a natural transformation s
n
: J
n
G
n1
by s
n
(b
n,}
) = c
n,}
. It
then has the required property J
G
n1
s
n
= [
n

n
s
n-1
J
T
n
.
Problems
1. Let J
0
J
1
be a chain complex of free 1-modules J
i
and D
0
D
1
an
exact sequence of 1-modules. A chain map ( : J
i
D
i
[ i N
0
) induces a homomor-
phism H
0
(
): H
0
(J
) H
0
(D
). Given a homomorphism : H
0
(J
) H
0
(D
0
)
there exists up to chain homotopy a unique chain map (
i
) such that H
0
(
) = . This can
be obtained as a special case of (11.5.1).
The reader should now study the notion of a projective module (one denition is: direct
summand of a free module) and then show that a similar result holds if the J
i
are only
assumed to be projective.
An exact sequence of the form 0 1
0
1
1
with projective modules
1
i
is called a projective resolution of the module . The result stated at the beginning says
that projective resolutions are unique up to chain equivalence. (Fundamental Lemma of
homological algebra). Each module has a free resolution.
11.6 Chain Equivalences
A chain map which induces an isomorphism of homology groups is under certain
circumstances a chain equivalence. This is one of the results of this section.
The notion of a chain homotopy can be used to develop a homotopy theory of
chain complexes in analogy to the topological homotopy theory.
We have the null complex; the chain groups are zero in each dimension. A
chain complex is called contractible if it is chain equivalent to the null complex, or
equivalently, if the identity is chain homotopic to the null map. A chain complex is
said to be acyclic if its homology groups are zero.
Let : (1. J
1
) (1. J
1
) be a chain map. We construct a newchain complex
C , the mapping cone of , by
(C )
n
= 1
n
1
n-1
. J
C(
(,. .) = (J
1
, .. J
1
.).
This can also be written in matrix form
_
,
.
_
_
J
1
0 J
1
__
,
.
_
.
The suspension 1 of 1 is dened by (1)
n
= 1
n-1
and J
1
= J
1
.
The canonical injection and projection yield an exact sequence of chain complexes
0 1 C 1 0. Associated is an exact sequence (11.3.2), and
the boundary morphism d: H
n1
(1) H
n
(1) equals H
n
( ), if we use the
canonical identications H
n1
(1) H
n
(1). The next result shows a typical
difference between the topological and the algebraic homotopy theory.
(11.6.1) Theorem. Let C be contractible. Then is a chain equivalence.
Proof. The inclusion | : 1 C , , (,. 0) is null homotopic, since C is
contractible. Let s : | . 0 be a null homotopy. We write s(,) = (;(,). g(,))
11 (without notation for the dimensions). The condition ds sd = | then reads
(d;, g, ;d,. dg, gd,) = (,. 0).
i.e., dg = gd and d; ;d = id g. Hence g is a chain map, and because of the
;-relation, a right homotopy inverse of .
The projection k : C 1 is likewise null homotopic. Let t : k . 0 be a
null homotopy. We write t (,. .) = h(,) j(.). The equality dt t d = k then
means
dh, hd, dj. jd. h . = ..
hence dh = hd and dj jd = h id. Therefore h is a chain map and a left
homotopy inverse of .
(11.6.2) Proposition. Let 1 be acyclic and suppose that 7
n
1
n
is always a
direct summand. Then 1 is contractible.
Proof. We have the exact sequence 0 7
n
1
n
J
T
n-1
0, and since
1 is acyclic we conclude 7
n
= T
n
. Moreover there exists t
n-1
: T
n-1
1
n
with dt
n-1
= id, since 7
n
is a direct summand of 1
n
. We therefore have a direct
decomposition 1
n
= T
n
t
n-1
T
n-1
. We dene s : 1
n
1
n1
by s[T
n
= t
n
and s
n
[t
n-1
T
n-1
= 0. With these denitions one veries separately on T
n
as well
as on t
n-1
T
n-1
that ds sd is the identity, i.e., s is a null homotopy of the identity.
(11.6.3) Theorem. Let : 1 1be a chain map between chain complexes which
consist of free modules over a principal ideal domain 1. If induces isomorphisms
+
: H
+
(1) H
+
(1), then is a chain equivalence.
Proof. The exact homology sequence and the hypothesis imply that C is acyclic.
Asubmodule of a free 1-module is free. Hence the boundary groups of the complex
C are free, and therefore the exact sequence 0 7
n
C
n
T
n-1
0 splits.
Now we apply (11.6.1) and (11.6.2), in order to see that is a chain equivalence.
11.7. Linear Algebra of Chain Complexes 289

Inthe topological applications we oftenhave toworkwithlarge chaincomplexes.
In some situations it is useful to replace themby smaller chain equivalent complexes.
A graded 1-module = (
n
) is said to be of nite type if the modules
n
are
nitely generated 1-modules.
(11.6.4) Proposition. Let 1 be a principal ideal domain. Let C = (C
n
) be a chain
complex of free 1-modules such that its homology groups are nitely generated.
Then there exists a free chain complex D of nite type which is chain equivalent
to C.
Proof. Let J
n
be a nitely generated submodule of 7
n
(C) which is mapped onto
H
n
(C) under the quotient map 7
n
(C) H
n
(C), and denote by G
n
the kernel
of the epimorphism J
n
H
n
(C). Dene a chain complex D = (D
n
. J
n
) by
D
n
= J
n
G
n-1
and J
n
(.. ,) = (,. 0). Then D is a free chain complex of nite
type and H
n
(D) = J
n
,G
n
H
n
(C). Since G
n
is a free submodule of T
n
(C) we
can choose for each n a homomorphism
n
: G
n
C
n1
such that c
n1
n
(,) = ,
for each , G
n
. Dene [
n
: D
n
= J
n
G
n-1
C
n
, (.. ,) .
n-1
(,).
One veries that [ = ([
n
) is a chain map which induces an isomorphism of
homology groups. By (11.6.3), [ is a chain equivalence.
11.7 Linear Algebra of Chain Complexes
We work in the category 1- MOD for a commutative ring 1.
11.7.1 Graded modules. Let
-
= (
n
) and T
-
= (T
n
) be Z-graded left 1-
modules over a commutative ring 1. The tensor product
-
T
-
is the module
with
]q=n
]

T
T
q
as entry in degree n. If :
-

t
-
and g: T
-
T
t
-
are morphisms of some degree, then their tensor product g is dened by
( g)(a b) = (1)
[[[o[
(a) g(b).
Here [a[ denotes the degree of a. The formula for the tensor product obeys the
(heuristic) gradedsignrule: Whenever entities of degree . and, are interchanged,
then the sign (1)
x,
appears. The tensor product of objects and of morphisms is
associative and compatible with composition (in the graded sense)
( g) (
t
g
t
) = (1)
[[[(
/
[
t
gg
t
(sign rule). This composition is associative, as it should be. When we use the degree
as upper index (e.g., in cohomology), then the agreement
k
=
-k
is sometimes
suitable.
11.7.2 Graded algebras. A Z-graded 1-algebra
-
is a Z-graded 1-module
(
n
[ n Z) together with a family of 1-linear maps
T

}
i}
. . , . ,.
The algebra is associative, if always . (, :) = (. ,) : holds, and commutative,
if always . , = (1)
[x[[,[
, . holds (sign rule). A unit element 1
0
of the
algebra satises 1 . = . = . 1. Let M
-
= (M
n
) be a Z-graded 1-module. A
family
i
M
}
M
i}
. a . a .
of 1-linear maps is the structure of an
-
-module on M
-
, provided the associativity
a (b .) = (a b) . holds for a. b and . M. If has a unit element,
then the module is unital, provided 1 . = . always holds. Let
-
and T
-
be Z-
graded algebras. Their tensor product T is the tensor product of the underlying
graded modules ( T)
n
=

i}=n
i
T
}
together with the multiplication
(a b) (a
t
b
t
) = (1)
[b[[o
/
[
aa
t
bb
t
(sign rule). If and T are associative,
then T is associative. If both have a unit element 1, then 11 is a unit element
for the tensor product. If both algebras are commutative, then their tensor product
is commutative. The tensor product of graded algebras is an associative functor.
11.7.3 Tensor product of chain complexes. Let (
-
. J
) and (T
-
. J
B
) be chain
complexes. Then the graded module
-
T
-
is a chain complex with boundary
operator J = J
1 1 J
B
. Here we have to take the sign rule into account,
i.e.,
J(a b) = J
a b (1)
[o[
a J
B
b.
One veries JJ = 0, using this sign rule. Passage to homology induces
H
]
(
-
) H
q
(T
-
) H
]q
(
-
T
-
). a| b| a b|.
The tensor product of chain complexes is associative.
11.7.4 Dual chaincomplex. We regard the ground ring 1as a trivial chain complex
with 1 in degree 0 and zero modules otherwise. Let (
n
. d) be a chain complex.
We dene the dual graded 1-module by
+
-n
= Hom
T
(
n
. 1). We require a
boundary operator :
+
-n

+
-n-1
on the dual module such that the evaluation
c:
+
-

-
1
c:
+
-n
n
1. a (a).
c = 0 otherwise, becomes a chain map. This condition, c( a) = 0 and 11.7.3
yield for a
+
-n-1
n
0 = Jc( a) = c(J( a)
= c( a (1)
[[
da
= ()(a) (1)
[[
(da).
i.e., we have to dene = (1)
[[1
d.
11.7. Linear Algebra of Chain Complexes 291
11.7.5 Hom-Complex. For graded modules
-
and T
-
we let Hom(
-
. T
-
) be
the module with
oZ
Hom(
o
. T
on
) as component in degree n. On this Hom-
module we use the boundary operator
J(
i
) = (d
i
) ((1)
n
i
d)
for (
i
:
i
T
in
) , i.e., the a-component pr
o
(J ) Hom(
o
. T
on-1
) for
= (
o
) Hom(
-
. T
-
)
n
is dened to be
pr
o
(J ) = d
o
(1)
n
g
o-1
d.
One veries JJ = 0. This denition generalizes our convention about the dual
module.
11.7.6 Canonical maps. The following canonical maps from linear algebra are
chain maps.
(1) The composition
Hom(T. C) Hom(. T) Hom(. T). (
i
) (g
}
) (
I[[
g
I
).
(2) The adjunction
: Hom(T. C) Hom(. Hom(T. C)). (
i
)(.)(,) =
[x[[,[
(.,).
(3) The tautological map
; : Hom(C. C
t
) Hom(D. D
t
) Hom(C D. C
t
D
t
)
with ;( g)(. ,) = (1)
[[[x[
(.) g(,) (sign rule).
(4) The trace map z:
+
T Hom(. T), z( b)(a) = (1)
[o[[b[
(a)b.
Problems
1. Tensor product is compatible with chain homotopy. Let s : . g: C C
/
be a chain
homotopy. Then s id: id . g id: C D C
/
D is a chain homotopy.
2. A chain complex model of the unit interval is the chain complex 1
+
with two non-zero
groups 1
1
1 with basis e, 1
0
1 1 with basis e
0
. e
1
and boundary operator J(e) =
e
1
e
0
(in the topological context: the cellular chain complex of the unit interval). We use
this model to dene chain homotopies with the cylinder 1
+
C. Note
C
n
C
n
C
n-1
(1
+
C)
n
. (.
1
. .
0
. ,) e
1
.
1
e
0
.
0
e ,.
A chain map h: 1
+
C D consists, via these isomorphisms, of homomorphisms
h
I
n
: C
n
D
n
and s
n
: C
n
D
n1
. The h
I
+
are chain maps (t = 0. 1) and Js
n
(,) =
h
1
n
(,) h
0
n
(,) s
n-1
c,, i.e., s
+
: h
1
+
. h
0
+
is a chain homotopy in our previous denition.
3. Imitate the topological denition of the mapping cone and dene the mapping cone of a
chain map : C D as a quotient of 1
+
C D. The n-th chain group is then canonically
isomorphic to C
n-1
D
n
and the resulting boundary operator is the one we dened in the
section on chain equivalences. Consider also the mapping cylinder from this view-point.
11.8 The Functors Tor and Ext
Let 1 be a principal ideal ring. We work in the category 1- MOD; this comprises
the category of abelian groups (Z-modules). An exact sequence 0 J
1
J
0

0 with free modules J
1
. J
0
is a free resolution of . Since submodules of
free modules are free, it sufces to require that J
0
is free. Let J() denote the
free 1-module generated by the set . Denote the basis element of J() which
belongs to a by a|. We have a surjective homomorphism : J() ,
n
o
a|
n
o
a. Let 1() denote its kernel. The exact sequence
0 1()
i
J()
]
0
will be called the standard resolution of . We take the tensor product (over 1)
of this sequence with a module G, denote the kernel of i 1 by Tor
T
(. G) =
Tor(. G) and call it the torsion product of . G.
We now derive some elementary properties of torsion products. We show that
Tor(. G) can be determined fromany free resolution, and we make Tor(. ) into
a functor in two variables. In the next lemma we compare free resolutions.
(11.8.1) Lemma. Given a homomorphism :
t
and free resolution F and
F
t
of and
t
, there exists a commutative diagram
0

J
1
i
(
1
J
0
]
(
0
x
,
0 F
0

J
t
1
i
/
J
t
0
]
/
t
0 F
t
(without s). If (

1
.

0
) is another choice of homomorphisms making the diagram
commutative, then there exists a homomorphism s : J
0
J
t
1
with
0

t
0
= i
t
s
and
1

t
1
= si .
Proof. Let (.
k
) be a basis of J
0
. Choose .
t
k
J
t
0
such that
t
(.
t
k
) = (.
k
).
Dene
0
by
0
(.
k
) = .
t
k
. Then =
t
0
. Since
t
0
i = 0, there exists by
exactness of F
t
a unique
1
such that
0
i = i
t
1
. Since
t
(
0
t
0
) = =
0, the elements (
0

t
0
)(.
k
) are contained in the kernel of
t
. Hence we have
(
0

t
o
)(.
k
) = i
t
(,
k
) for suitable ,
k
. We dene s by s(.
k
) = ,
k
. From
i
t
(
1
t
1
) =
0
i
t
0
i = i
t
si and the injectivity of i
t
we conclude
1
t
1
= si .
We take the tensor product G of the diagram in (11.8.1). The homomorphism
1
1 induces a homomorphism Ker(i 1) Ker(i
t
1) and
1

t
1
= si
shows that this homomorphism does not depend on the choice of (
1
.
0
). Let us
denote this homomorphism by T( : F . F
t
). If g:
t

tt
is given and F
tt
a
11.8. The Functors Tor and Ext 293
free resolution of
tt
, then T(g: F
t
. F
tt
) T( : F . F
t
) = T(g : F . F
tt
). This
implies that an isomorphism induces an isomorphismT( : F . F
t
). In particular
each free resolution yields a unique isomorphism Ker(i 1) Tor(: G), if we
compare F with the standard resolution. The standard resolution is functorial in .
This fact is used to make Tor(. G) into a functor. It is clear that a homomorphism
G G
t
induces a homomorphisms Tor(. G) Tor(. G
t
). Hence Tor is also
a functor in the variable G (and the two functor structures commute).
If we view 0 J
1
J
0
0 in (11.8.1) as a chain complex, then (
1
.
0
) is
a chain map and s yields a chain homotopy between (
1
.
0
) and (
t
1
.
t
0
).
(11.8.2) Proposition. Elementary properties of torsion groups in the category of
abelian groups are:
(1) Let be a free abelian group. Then Tor(. G) = 0.
(2) Tor(Z,n. G) {g G [ ng = 0] G.
(3) If G is torsion free, then Tor(Z,n. G) = 0.
(4) Tor(Z,m. Z,n) Z,J with J the greatest common divisor of m. n.
(5) Adirect sumdecomposition
1
2
induces a direct sumdecomposition
Tor(. G) Tor(
1
. G) Tor(
2
. G).
Proof. (1) 0 0 0 is a free resolution. (2) Use the free resolution
0 Z
n
Z Z,n 0. (3) and (4) are consequences of (2). In order to verify
(5), use the direct sum of free resolutions.
We can also work with a resolution of the other variable. Let Q
1
Q
0
T
be a free resolution and dene Tor
t
(. T) = Ker( Q
1
Q
0
).
(11.8.3) Proposition. There exists a canonical isomorphism
Tor(. T) Tor
t
(. T).
Proof. Let 1
1
1
0
be a free resolution. From the resolutions of and T
we obtain a commutative diagram:
Tor(. T)
1
1
Q
1

1
1
Q
0

1
1
T
;
1
0
Q
1

1
0
Q
0

1
0
T
Tor
t
(. T)

Q
1

Q
0

T.
The KernelCokernel Lemma (11.2.6) yields an isomorphism of Tor(. T) =
Ker(;) with the submodule Tor
t
(. T) of Coker .
Interchangingthe tensor factors yields anisomorphismTor(T. ) Tor
t
(. T).
We combine this with (11.8.3) and see that the isomorphisms (11.8.2) also hold if
we interchange the variables. It is now no longer necessary to use the notation Tor
t
.
The functor Ext is dened in analogy to the functor Tor, the tensor product is
replaced by the Hom-functor.
Let 1 be a principal ideal domain and 0 1()
i
J()
]
0 the
standard free resolution of as above. We apply the functor Hom
T
(. T) to this
sequence. The cokernel of i
+
: Hom(J(). T) Hom(1(). T) is dened to be
Ext
T
(. T) = Ext(. T). We show that Ext(. T) can be determined from any
free resolution. We start with a diagram as in (11.8.1) and obtain a well-dened
homomorphismCoker(Hom(i. T)) Coker(Hom(i
t
. T)); in particular we obtain
an isomorphism Ext(. T) Coker(Hom(i. T)).
(11.8.4) Proposition. Elementary properties of Ext in the category of abelian
groups are:
(1) Ext(. T) = 0 for a free abelian group .
(2) Ext(Z,n. T) T,nT.
(3) Ext(Z,n. T) = 0 for T = Q. Q,Z. R.
(4) Ext(Z,m. Z,n) Z,(m. n).
(5) Ext(
1
2
. T) Ext(
1
. T) Ext(
2
. T).
The foregoing develops what we need in this text. We should at least mention
the general case. Let 0 C 1
0
1
1
be a projective resolution of the
1-module C and let be another 1-module. We apply Hom(. ) to the chain
complex 1
+
and obtain a cochain complex Hom(1
+
. ); its i -th cohomology group
(i _ 1) is denoted Ext
i
T
(C. ). Since projective resolutions are unique up to chain
equivalence, the Ext
i
T
-groups are unique up to isomorphism. For principal ideal
domains only Ext
1
occurs, since we have resolution of length 1. The notation Ext
has its origin in the notion of extensions of modules. An exact sequence
0 T
n-1
T
1
T
0
C 0
is called an n-fold extension of A by C. One can obtain Ext
n
T
(C. ) as certain
congruence classes of n-fold extension of by C, see [120, Chapter III]. Write
1 1
t
if there exists a commutative diagram
1: 0

C

=
T
n-1

T
0

0
1
t
: 0

C
t
T
t
n-1

T
t
0

0.
The congruence relation is generated by .
11.9. Universal Coefcients 295
Problems
1. Suppose Tor(. Z,) = 0 for each prime . Then the abelian group is torsion free.
2. The kernel of
Z
Q, a a 1 is the torsion subgroup of .
3. Does there exist a non-trivial abelian group such that F = 0 for each eld F?
11.9 Universal Coefcients
We still work in 1- MOD for a principal ideal domain 1. Let C = (C
n
. c
n
) be
a chain complex of modules. Then C G = (C
n
G. c
n
1) is again a chain
complex.
(11.9.1) Proposition (Universal Coefcients). Let C be a chain complex of free
modules. Then there exists an exact sequence
0 H
q
(C) G

H
q
(C G)
Tor(H
q-1
(C). G) 0.
The sequence is natural in C and G and splits. The homomorphism sends :| g
for a cycle : to the homology class : g|.
Proof. The sequence 0 7
n
C
n
c
n
T
n-1
0 is exact; T
n-1
is a submodule
of C
n-1
and hence free. Therefore the sequence splits and the induced sequence
0 7
n
G C
n
G T
n-1
G 0
is again a split exact sequence. We consider the totality of these sequences as an
exact sequence of chain complexes, the 7- and the T-complex have trivial boundary
operator. Associated to this short exact sequence of chain complexes is a long exact
homology sequence of the form
T
n
G
i1
7
n
G

H
n
(C G)

T
n-1
G
i1
7
n-1
G.
One veries that the boundary operator (11.3.1) of the homology sequence is i 1,
where i : T
n
7
n
. The sequence T
n
G 7
n
G H
n
G 0 is exact,
hence the cokernel of i 1 is H
n
(C) G, and the resulting map H
n
(C) G
H
n
(C G) is . The kernel of T
n-1
G 7
n-1
G is Tor(H
n-1
(C). G),
because 0 T
n-1
7
n-1
H
n-1
(C) 0 is a free resolution. Let r : C
n

7
n
be a splitting of 7
n
C
n
. Then
7
n
(C G) C
n
G
i1
7
n
G H
n
(C) G
maps T
n
(C G) to zero and induces j: H
n
(C G) H
n
(C)G with j = id,
i.e., a splitting of the universal coefcient sequence.
Let again C = (C
n
. c
n
) be a chain complex with free 1-modules C
n
. We
obtain the cochain complex with cochain groups Hom(C
n
. G) and cohomology
groups H
n
(C: G).
(11.9.2) Proposition (Universal Coefcients). There exists an exact sequence
0 Ext(H
n-1
(C). G) H
n
(C: G)

Hom(H
n
(C). G) 0.
The map sends the cohomology class of the cocycle : C
n
G to the homo-
morphismH
n
(C) G, c| (c). The sequence is natural with respect to chain
maps (variable C) and module homomorphisms (variable G). The sequence splits,
and the splitting is natural in G but not in C.
Proof. Again we start with the split exact sequence 0 7
n
C
n
T
n-1
0
and the induced exact sequence
0 Hom(7
n
. G) Hom(C
n
. G) Hom(T
n-1
. G) 0.
We consider the totality of these sequences as an exact sequence of cochain com-
plexes, the 7- and the T-complex have trivial coboundary operator. Associated
to this short exact sequence of cochain complexes is a long exact cohomology
sequence of the form
Hom(T
n
. G)
d
n
Hom(7
n
. G) H
n
(C: G) Hom(T
n-1
. G)
which induces a short exact sequence
(4) 0 Ker J
n

H
n
(C: G) Coker J
n-1
0.
We need:
(11.9.3) Lemma. The formal coboundary operator J
n
(without the additional sign
introduced earlier!) is the homomorphism induced by i : T
n
7
n
.
Proof. Let : 7
n
G be given. Then J
n
() is obtained as follows: Extend to
: C
n
G. Apply and nd a pre-image of ( ) = c
n1
in Hom(T
n
. G). One
veries that i is a pre-image.
From the exact sequence 0 T
n
7
n
H
n
(C) 0 we obtain the exact
sequence
Hom(T
n
. G)
i
Hom(7
n
. G) Hom(H
n
(C). G) 0.
We use it to identify the Ker i
+
with Hom(H
n
(C). G). One veries that is as
claimed in the statement (11.9.2). From the free presentation and the denition
11.9. Universal Coefcients 297
of Ext we thus obtain the exact sequence of the theorem. The naturality of this
sequence is a consequence of the construction. It remains to verify the splitting.
We choose a splitting r : C
n
7
n
of the inclusion 7
n
C
n
. Now consider the
diagram
0

7
n
(Hom(C. G))

Hom(C
n
. G)

Hom(C
n1
. G)
0

Hom(H
n
(C). G)

Hom(7
n
. G)
i

Hom(T
n
. G).
If Ker i
+
, then r
+
() = r Ker . The splitting is induced by Ker i
+

7
n
(Hom(C. G)), t .
Without going into the denition of Ext we see from the discussion:
(11.9.4) Proposition. Suppose H
n-1
(C) is a free 1-module. Then the homomor-
phism : H
n
(C: G) Hom(H
n
(C). G) in (11.9.2) is an isomorphism.
Proof. The sequence 0 T
n-1
7
n-1
H
n-1
0 splits and therefore the
cokernel of J
n-1
is zero.
Given a cochain complex C
-
= (C
q
.
q
) we can view it as a chain complex
C
-
= (C
q
. d
q
) by a shift of indices: We set C
q
= C
-q
and we dene d
q
: C
q

C
q-1
as
-q
: C
-q
C
-q1
. We can now rewrite (11.9.1):
(11.9.5) Proposition. Let C
-
be a cochain complex of free 1-modules. Then we
have a split exact sequence
0 H
q
(C
-
) G H
q
(C
-
G) Tor(H
q1
(C
-
). G) 0.
Let now C
-
be a chain complex of free modules. We apply (11.9.5) to the dual
cochain complex with C
q
= Hom(C
q
. 1) and cohomology groups H
q
(C: 1).
(11.9.6) Proposition. Let C be a free chain complex and G be a module such that
either H
+
(C) is of nite type or G is nitely generated. Then there exists a natural
exact sequence
0 H
]
(C) G H
q
(C: G) Tor(H
q1
(C). G) 0
and this sequence splits.
Proof. If G is nitely generated we have a canonical isomorphism of the form
Hom(C. 1) G Hom(C. G); we use this isomorphism in (11.9.5). If H
+
(C)
is of nite type we replace C by a chain equivalent complex C
t
of nite type (see
(11.6.4)). In that case we have again a canonical isomorphism Hom(C
t
. 1) G
Hom(C
t
. G). We apply now (11.9.5) to C
t
.
(11.9.7) Proposition. Let : C D be a chain map between complexes of free
abelian groups. Suppose that for each eld F the map F induces isomorphisms
of homology groups. Then is a chain equivalence.
Proof. Let C( ) denote the mapping cone of . The hypothesis implies that
H
+
(C( ) F) = 0. We use the universal coefcient sequence. It implies that
Tor(H
+
(C( )). Z,) = 0 for each prime . Hence H
+
(C( )) is torsion-free.
From H
+
(C( )) Q we conclude that H
+
(C( )) is a torsion group. Hence
H
+
(C( )) = 0. Now we use (11.6.3).
11.10 The Knneth Formula
Let C and Dbe chain complexes of 1-modules over a principal ideal domain 1. We
have the tensor product chain complex C
T
D and the associated homomorphism
: H
i
(C)
T
H
}
(D) H
i}
(C
T
D). .| ,| . ,|.
We use the notation + for Tor
T
. The next theorem and its proof generalizes the
universal coefcient formula (11.9.1).
(11.10.1) Theorem (Knneth Formula). Suppose C consists of free 1-modules.
Then there exists an exact sequence
0

i}=n
H
i
(C)
T
H
}
(D) H
n
(C
T
D)

i}=n-1
H
i
(C)+H
}
(D) 0.
If also D is a free complex, then the sequence splits.
Proof. We consider the graded modules 7(C) and T(C) of cycles and boundaries
as chain complexes with trivial boundary. Since 7(C) is free, we have the equalities
(canonical isomorphisms)
(7(C) 7(D))
n
= Ker(1 d: (7(C) D)
n
(7(C) D)
n-1
)
and
(7(C) T(D))
n
= Im(1 d: (7(C) D)
n1
(7(C) D)
n
).
and they imply H(7(C) D) 7(C) H(D) (homology commutes with the
tensor product by a free module). In a similar manner we obtain an isomorphism
H(T(C)D) T(C)H(D). We formthe tensor product of the free resolution
of chain complexes
0 T(C)
i
7(C) H(C) 0
11.10. The Knneth Formula 299
with H(D). We obtain the following exact sequence, referred to as (f), with
injective morphism (1) and surjective morphism (2)
H(C) + H(D)
(1)
T(C) H(D)
i1
7(C) H(D)
(2)
H(C) H(D)
H(T(C) D)
(i1)
H(7(C) D).
Let us use the notation (1|)
n
=
n-1
for a graded object . We tensor the
exact sequence of chain complexes 0 7(C) C T(C)1| 0 with D
and obtain an exact sequence
0 7(C) D C D (T(C) D)1| 0.
Its exact homology sequence has the form
. . . H(T(C) D)
(1)
H(7(C) D) H(C D)
H(T(C) D)1|
(1)
H(7(C) D)1| .
One veries that (1) is the map (i 1)
+
. Hence we obtain the exact sequence
0 Coker(i
+
) H(C D) Ker(i
+
)1| 0
which yields, together with the sequence (f), the exact sequence of the theorem.
Choose retractions r : C
n
7
n
(C) and s : D
n
7
n
(D). Then (C D)
n

H(C) H(D), c J r(c)| s(J)| sends the boundaries of (C D)
n
to
zero and induces a retraction j: H
n
(C D) (H(C) H(D))
n
of .
As in the case of the universal coefcient theorem we can rewrite (11.10.1) in
terms of cochain complexes. Under suitable niteness conditions we can then apply
the result to the dual complex of a chain complex and obtain:
(11.10.2) Theorem(Knneth Formula). Let C and D be free chain complexes such
that H
+
(C) or H
+
(D) is of nite type. Then there exists a functorial exact sequence
0

i}=n
H
i
(C) H
}
(D) H
n
(C D)

i}=n1
H
i
(C) +H
}
(D) 0
Chapter 12
Cellular Homology
In this chapter we nally show that ordinary homology theory is determined on
the category of cell complexes by the axioms of Eilenberg and Steenrod. From the
axioms one constructs the cellular chain complex of a CW-complex. This chain
complex depends on the skeletal ltration, and the boundary operators of the chain
complex are determined by the so-called incidence numbers; these are mapping
degrees derived from the attaching maps. The main theorem then says that the
algebraic homology groups of the cellular chain complex are isomorphic to the ho-
mology groups of the homology theory (if it satises the dimension axiom). From
this fact one obtains immediately qualitative results and explicit computations of
homology groups. Thus if X has k(n) n-cells, then H
n
(X: Z) is a subquotient
of the free abelian group of rank k(n). A nite cell complex has nitely gener-
ated homology groups. We deduce that the combinatorial Euler characteristic is a
homotopy invariant that can be computed from the homology groups.
In the case of a simplicial complex we show that singular homology is isomor-
phic to the classical combinatorial simplicial homology. In this context, simplicial
homology is a special case of cellular homology.
12.1 Cellular Chain Complexes
Let h
+
be an additive homology theory. Let X be obtained from by attaching
n-cells via (. ):

eT
(D
n
e
. S
n-1
e
) (X. ). The characteristic map of the
cell e is denoted by (
e
.
e
). The index e distinguishes different copies.
(12.1.1) Proposition. The induced map
n
= (
e
+
):

e
h
+
(D
n
e
. S
n-1
e
) h
+
(X. )
is an isomorphism.
Proof. By (10.4.6),
+
: h
+
_
e
(D
n
e
. S
n-1
e
)
_
h
+
(X. ) is an isomorphism.
Now apply the additivity isomorphism

e
h
+
(D
n
e
. S
n-1
e
) h
+
_
e
(D
n
e
. S
n-1
e
_
and compose it with
+
.
The isomorphism inverse to
n
is obtained as follows. Given : h
k
(X. ).
We use the inclusion
e
: (X. ) (X. X e) and the relative homeomorphism
e
: (D
n
. S
n-1
) (X. X e). Let :
e
h
k
(D
n
e
. S
n-1
e
) denote the image of :
under
: h
k
(X. )
]
c
h
k
(X. X e)
h
k
(D
n
e
. S
n-1
e
) :
e
.
12.1. Cellular Chain Complexes 301
Then : (:
e
[ e 1) is inverse to
n
.
Let X be a CW-complex. The boundary operator d: h
k1
(X
n1
. X
n
)
h
k
(X
n
. X
n-1
) of the triple (X
n1
. X
n
. X
n-1
) is transformed via the isomorphisms
(12.1.1) into a matrix of linear maps
m(e. ): h
k1
(D
n1
(
. S
n
(
) h
k
(D
n
e
. S
n-1
e
)
for each pair (. e) of an (n1)-cell and an n-cell e (as always in linear algebra).
Let |
e,(
be the composition
S
n
(
X
n
q
c
X
n
,(X
n
e)

c
D
n
,S
n-1
.
If we compose |
e,(
with an h-equivalence k
n
: D
n
,S
n-1
S
n
, then k
n
|
e,(
has
as a self-map of S
n
a degree J(e. ). We call J(e. ) the incidence number of the
pair (. e) of cells. The case n = 0 is special, so let us consider it separately. Note
that D
0
,S
-1
is the point D
0
= {0] together with a disjoint base point {+]. Let k
0
be given by k
0
(0) = 1 and k
0
(+) = 1. We have two 0-cells
(
(1) = e
(they could coincide). With these conventions J(. e
) = 1.
In the following considerations we use different notation d, d
t
, d
tt
for the bound-
ary operators.
h
k1
(D
n1
(
. S
n
(
)
n(e,( )
J
//
h
k
(D
n
e
. S
n-1
e
)
]
h
k
(S
n
(
)
t
c,

h
k
(D
n
e
,S
n-1
e
)
is commutative.
h
k1
(X
n1
. X
n
)
J
/

h
k
(X
n
)
}
h
k
(X
n
. X
n-1
)
]
c

h
k
(X
n
,X
n-1
)
h
k1
(D
n1
. S
n
)
J
//

h
k
(S
n
)
h
k
(X
n
. X
n
e)
]

h
k
(X
n
,X
n
e)
h
k
(D
n
. S
n-1
)

h
k
(D
n
,S
n-1
).
Given . h
k1
(D
n1
. S
n
). Then
+
m(e. ). is, by denition of m(e. ), the
image of . in

h
k
(D
n
,S
n-1
). Now use the commutativity of the diagram.
302 Chapter 12. Cellular Homology
(12.1.3) Corollary. Let o : h
k
(D
n
. S
n-1
) h
k1
(D
n1
. S
n
) be a suspension
isomorphism. Then m(e. ) o is the multiplication by J(e. ), provided the
relation d
tt
o = k
n
+

+
holds.
We now write the isomorphism (12.1.1) in a different form. We use an it-
erated suspension isomorphism o
n
: h
k-n
h
k
(D
n
. S
n-1
) in each summand.
Let C
n
(X) denote the free abelian group on the n-cells of (X. ). Elements in
C
n
(X)
Z
h
k-n
will be written as nite formal sums
e
e u
e
where u
e
h
k-n
;
the elements in C
n
(X) h
k-n
are called cellular n-chains with coefcients in
h
kn
. We thus have constructed an isomorphism
n
: C
n
(X)
Z
h
k-n
h
k
(X. ).
e
e u
e

e

e
+
o
n
(u
e
).
The matrix of incidence numbers provides us with the Z-linear map
M(n): C
n1
(X) C
n
(X).
e
J(e. )e.
The sum is nite: J(e. ) can only be non-zero if the image of
(
intersects e
(property (W3) of aWhitehead complex). Fromthe preceding discussion we obtain:
(12.1.4) Proposition. Suppose o and k are chosen such that the relation (12.1.3)
holds. Then the diagram
h
k1
(X
n1
. X
n
)
J
h
k
(X
n
. X
n-1
)
C
n1
(X) h
k-n
(
n1
(n)id
C
n
(X) h
k-n
(
n
is commutative.
The composition of the boundary operators (belonging to the appropriate triples)
h
n1
(X
n1
. X
n
)
J
h
n
(X
n
. X
n-1
)
J
h
n-1
(X
n-1
. X
n-2
)
is zero, because the part h
n
(X
n
) h
n
(X
n
. X
n-1
) h
n-1
(X
n-1
) of the ex-
act sequence of the pair (X
n
. X
n-1
) is contained in this composition. We set
h
n,k
(X) = h
nk
(X
n
. X
n-1
). Thus the groups (h
n,k
(X) [ n Z) together with
the boundary operators just considered form a chain complex h
-,k
(X).
(12.1.5) Proposition. The product M(n 1)M(n) of two adjacent incidence ma-
trices is zero. The cellular chain groups C
n
(X) together with the homomorphisms
M(n): C(n) C(n 1) form a chain complex C
-
(X). This chain complex is
called the cellular chain complex of X.
Proof. The relation M(n1)M(n) = 0 follows from (12.1.4) applied to the chain
complex H
-,0
(X) obtained from singular homology with coefcients in Z.
12.1. Cellular Chain Complexes 303
The cellular chaincomplexhas its algebraicallydenedhomologygroups. Inthe
next section we prove that in the case of an ordinary homology theory the algebraic
homology groups of the cellular chain complex are naturally isomorphic to the
homology groups of the theory. We should point out that the algebraic homology
groups of the chain complexes h
-,k
(X) only depend on the space and the coefcients
of the homology theory, so are essentially independent of the theory. Nevertheless,
they can be used to obtain further information about general homology theories
this is the topic of the so-called spectral sequences [130].
The denition of incidence numbers uses characteristic maps and a homotopy
equivalence k. These data are not part of the structure of a CW-complex so that
the incidence numbers are not completely determined by the CW-complex. The
choice of a characteristic map determines, as one says, an orientation of the cell. If
. : (D
n
. S
n-1
) (X
n
. X
n-1
) are two characteristic maps of a cell e, then
-1
: D
n
,S
n-1
X
n
,X
n
e D
n
,S
n-1
is a homeomorphism and hence has degree 1. One concludes that the incidence
numbers are dened up to sign by the CW-complex.
(12.1.6) Proposition. A cellular map : X Y induces a chain map with com-
ponents
+
: h
n
(X
n
. X
n-1
) h
n
(Y
n
. Y
n-1
). Homotopic cellular maps induce
chain homotopic maps.
Proof. The rst assertion is clear. Let . g: X Y be cellular maps and let
: X1 Y. . g be a homotopy between them. By the cellular approximation
theorem we can assume that is cellular, i.e., ((X 1)
n
) Y
n
. Note that
(X 1)
n
= X
n
d1 LX
n-1
1. We dene a chain homotopy as the composition
s
n
: h
n
(X
n
. X
n-1
)
c
h
n1
((X
n
. X
n-1
) (1. d1))
h
n1
(Y
n1
. Y
n
).
In order to verify the relation ds
n
= g
+

+
s
n-1
d we apply (10.9.4) to
(. T. C) = (X
n
. X
n-1
. X
n-2
) and compose with
+
.
(12.1.7) Proposition. Let j: k
+
() l
+
() be a natural transformation be-
tween additive homology theories such that j induces isomorphisms of the coef-
cient groups j: k
n
(1) l
n
(1), n Z, 1 a point. Then j is an isomorphism
k
+
(X) l
+
(X) for each CW-complex X.
Proof. Since j is compatible with the suspension isomorphismwe see from(12.1.1)
that j: k
+
(X
n
. X
n-1
) l
+
(X
n
. X
n-1
). Now one uses the exact homology se-
quences and the Five Lemma to prove by induction on n that j is an isomorphism
for n-dimensional complexes. For the general case one uses (10.8.1).
Problems
1. The map . (2
_
1 [.[
2
.. 2[.[
2
1) induces a homeomorphism k
n
. Let o be the
suspension isomorphism (10.2.5). Then commutativity holds in (12.1.3). For the proof show
that S
n
S
n
,D
n
D
n
-
,S
n-1

D
n
,S
n-1
k
n
S
n
, with the projection r which
deletes the last coordinate, has degree 1.
2. Prove M(n 1)M(n) = 0 without using homology by homotopy theoretic methods.
12.2 Cellular Homology equals Homology
Let H
+
() = H
+
(: G) be an ordinary additive homology theory with coefcients
in G (not necessarily singular homology). The cellular chain complex C
-
(X) =
C
-
(X: G) of a CW-complex X with respect to this theory has its algebraically
dened homology groups. It is a remarkable and important fact that these algebraic
homology groups are naturally isomorphic to the homology groups of the space X.
This result says that the homology groups are computable from the combinatorial
data (the incidence matrices) of the cellular complex.
(12.2.1) Theorem. The n-th homology group of the cellular chain complex C
-
(X)
is naturally isomorphic to H
n
(X).
Proof. We show that the isomorphism is induced by the correspondence
H
n
(X
n
. X
n-1
) H
n
(X
n
) H
n
(X).
We divide the proof into several steps.
(1) A basic input is H
k
(X
n
. X
n-1
) = 0 for k ,= 0; this follows from our
determination of the cellular chain groups in (12.1.1) and the dimension axiom.
(2) H
k
(X
n
) = 0 for k > n. Proof by induction on n. The result is clear
for X
0
by the dimension axiom. Let k > n 1. We have the exact sequence
H
k
(X
n
) H
k
(X
n1
) H
k
(X
n1
. X
n
). The rst group is zero by induction,
the third by (1).
(3) Since H
n-1
(X
n-2
) = 0, the map H
n-1
(X
n-1
) H
n-1
(X
n-1
. X
n-2
) is
injective. Hence the cycle group 7
n
of the cellular chain complex is the kernel of
d: H
n
(X
n
. X
n-1
) H
n-1
(X
n-1
).
(4) The exact sequence 0 H
n
(X
n
) H
n
(X
n
. X
n-1
) H
n-1
(X
n-1
)
induces an isomorphism (b): H
n
(X
n
) 7
n
.
(5) H
k
(X. X
n
) = 0 for k _ n. One shows by induction on t that the groups
H
k
(X
nt
. X
n
) are zero for t _ 0 and k _ n. We know that for an additive theory
the canonical map colim
t
H
k
(X
nt
. X
n
) H
k
(X. X
n
) is an isomorphism (see
(10.8.1) and (10.8.4)). For singular homology one can also use that a singular
chain has compact support and that a compact subset of X is contained in some
skeleton X
n
.
12.2. Cellular Homology equals Homology 305
(6) The map H
n
(X
n1
) H
n
(X) is an isomorphism. This follows from the
exact sequence of the pair (X. X
n1
) and (5).
(7) The diagram
H
n1
(X
n1
. X
n
)
J
7
n

7
n
,T
n

0
H
n1
(X
n1
. X
n
)
J
H
n
(X
n
)
(b)
H
n
(X
n1
)

v
v
v
v
v
v
v
v
v
v
v
v
(o)
0
H
n
(X)
shows us that we have an induced isomorphism (a) (Five Lemma).
(12.2.2) Corollary. Suppose X has a nite number of n-cells; then H
n
(X: Z) is a
nitely generated abelian group. Let X be n-dimensional; then H
k
(X: G) = 0 for
k > n.
(12.2.3) Example (Real projective space). The diagram
S
i-1
R1
i-1
D
i

R1
i
with attaching map : . ..
_
1 [.[
2
| is a pushout. The incidence map with
the homomorphism k
i-1
of Problem 1 in the previous section is computed to be
S
i-1
S
i-1
, (,. t ) (2t,. 2t
2
1) of degree 1(1)
i
. This yields the cellular
chain complex
C
0
0
C
1
2
C
2
0

with C
i
= Zfor 0 _ i _ nand boundary operator, alternatively, the zero morphism
and the multiplication by 2. The cellular chain complex with coefcients in the
abelian group G is of the same type (using the canonical identication ZG G).
Let
2
G = {g G [ 2g = 0]. We obtain the cellular homology
H
}
(R1
n
: G)
G. = 0.
G,2G. 0 < = 2k 1 < n.
2
G. 0 < = 2k _ n.
G n = 2k 1.
Similarly for R1
o
.
(12.2.4) Example. The sphere S
n
has a CW-decomposition with two i -cells in
each dimension i , 0 _ i _ n. The attaching diagram is
S
n-1
S
n-1
S
n-1
D
n
D
n
(
-
,
)

S
n
with
(.) = (..
_
1 [.[
2
) and
-
(.) =
(.). This attaching map

is G-equivariant, if the cyclic group G = {1. t [ t
2
= 1] acts on the spheres
by the antipodal map and on the left column by permutation of the summands.
The equivariant chain groups are therefore isomorphic to the group ring ZG =
Z 1 Z t . In order to determine the equivariant boundary operator we use
the fact that we know already the homology of this chain complex. If we add the
homology groups in dimension 0 and n we obtain an exact sequence
0 Z(c
n
)
)
ZG ZG ZG
t
Z 0.
The map c sends 1. t to 1. The kernel is generated by 1 t . From the geometry
we see that the rst boundary operator sends the generator 1 C
1
represented by
a suitably oriented 1-cell to (1 t ). We orient the cell such that the plus-sign
holds. Then J
1
is the multiplication by 1 t . The kernel of J
1
is thus generated by
1 t . We can again orient the 2-cells such that J
2
is multiplication by 1 t . If we
continue in this manner, we see that J
k
= 1t for k odd, and J
k
= 1t for k even.
The homology module H
n
(S
n
) = Z(c
n
) carries the t -action c
n
= (1)
n1
, the
degree of the antipodal map. One can, of course, determine the boundary operator
by a computation of degrees. We leave this as an exercise. Similar results hold
for S
o
.
Let X and Y be CW-complexes. The product inherits a cell decomposition.
The cross product induces an isomorphism
kI=n
H
k
(X
k
. X
k-1
) H
I
(Y
I
. Y
I-1
) H
n
((X Y )
n
. (X Y )
n-1
).
With a careful choice of cell orientations these isomorphisms combine to an iso-
morphism C
+
(X) C
+
(Y ) C
+
(X Y ) of cellular chain complexes.
12.3 Simplicial Complexes
We describe the classical combinatorial denition of homology groups of polyhe-
dra. These groups are isomorphic to the singular groups for this class of spaces.
The combinatorial homology groups of a nite polyhedron are nitely generated
abelian groups and they are zero above the dimension of the polyhedron. This nite
generation is not at all clear from the denition of the singular groups.
12.3. Simplicial Complexes 307
Recall that a simplicial complex 1 = (1. S) consists of a set 1 of vertices and
a set S of nite subsets of 1. A set s S with q 1 elements is called a q-simplex
of 1. We require the following axioms:
(1) A one-point subset of 1 is a simplex in S.
(2) s S and 0 ,= t s imply t S.
An ordering of a -simplex is a bijection {0. 1. . . . . ] s. An ordering of 1
is a partial order on 1 which induces a total ordering on each simplex. We write
s = (
0
. . . . .
]
), if the vertices of s satisfy
0
<
1
< <
]
in the given partial
ordering. Let C
]
(1) denote the free abelian group with basis the set of -simplices.
Its elements are called the simplicial -chains of 1. Now x an ordering of 1 and
dene a boundary operator
d: C
]
(1) C
]-1
(1). (
0
. . . . .
]
)
]
i=0
(1)
i
(
0
. . . . .
i
. . . . .
]
).
The symbol
i
means that this
i
is to be omitted from the string of vertices.
The boundary relation dd holds (we set C
]
(1) = 0 for _ 1). We denote
the -th homology group of this chain complex by H
]
(1). This is the classical
combinatorial homology group.
A simplicial complex 1 has a geometric realization [1[. An ordered simplex
s = (
0
. . . . .
]
) has an associated singular simplex
x
: ^
]
[1[. (t
0
. . . . . t
]
)
t
}
}
.
We extend s
x
by linearity to a homomorphism j
]
: C
]
(1) S
]
([1[). The
boundary operators are arranged so that j = (j
]
) is a chain map.
(12.3.1) Theorem. j induces isomorphisms 1
]
: H
]
(1) H
]
([1[).
Proof. We write X = [1[. Let S() be the set of -simplices. The characteristic
maps
x
: (^
]
. d^
]
) (X
]
. X
]-1
) yield an isomorphism (12.1.1),
]
:

xS(])
H
]
(^
]
x
. d^
]
x
) H
]
(X
]
. X
]-1
).
The identity of ^
]
represents a generator |
]
of H
]
(^
]
. d^
]
). Let .
x
be its image
under
x
+
. Then (.
x
[ s S()) is a Z-basis of H
x
(X
]
. X
]-1
). If we express
. H
]
(X
]
. X
]-1
) in terms of this basis, . =
x
n
x
.
x
, then n
x
is determined by
the image n
x
|
]
of . under
. H
]
(X
]
. X
]-1
) H
]
(X
]
. X
]
e
x
)
H
]
(^
]
. d^
]
) n
x
|
]
.
Here e
x
is the open simplex which belongs to s. Let s(i ) denote the i -th face of
^
]
and .
x(i)
H
]-1
(X
]-1
. X
]-2
) the corresponding basis element. We claim
d.
x
=
]
i=0
(1)
i
.
x(i)
. It is clear for geometric reasons that the expression of d.
x
in terms of the basis (.
t
[ t S( 1)) can have a non-zero coefcient only for
the .
x(i)
. The coefcient of .
x(i)
is seen from the commutative diagram
H
]
(^
]
. d^
]
)

:
H
]
(X
]
. X
]-1
)
J
H
]-1
(d^
]
. d^
]
s(i )
)

:

H
]-1
(X
]-1
. X
]-2
)
H
]-1
(^
]-1
. d^
]-1
)
(d
)
i
)
:(i)
H
]-1
(X
]-1
. X
]-1
e
x(i)
).
Note
x
J
]
i
=
x(i)
. The left column sends |
]
to (1)
i
|
]-1
. We have con-
structed so far an isomorphism of C
-
(1) with the cellular chain complex C
-
([1[)
of [1[. Let 1
]
: H
]
(C
-
(1)) H
]
(C
-
([1[)) be the induced isomorphism. Let
Q
]
: H
]
(C
-
([1[)) H
]
([1[) be the isomorphism in the proof of (12.2.1). Trac-
ing through the denitions one veries 1
]
= Q
]
1
]
. Hence 1
]
is the composition
of two isomorphisms.
An interesting consequence of (12.3.1) is that j: C
-
(1) S
-
([1[) is a chain
equivalence. Thus, for a nite complex 1, the singular complex of [1[ is chain
equivalent to a chain complex of nitely generated free abelian groups, zero above
the dimension of 1.
(12.3.2) Example. A circle S
1
can be triangulated by a regular n-gon with ver-
tices {e
0
. . . . . e
n-1
] and ordered simplices s
i
= (e
i
. e
i1
), 0 _ i _ n 1 and
mod n notation e
n
= e
0
. The cellular chain complex is given by d(e
i
. e
i1
) =
(e
i
) (e
i1
). The sum : =

n-1
i=0
s
i
is a 1-cycle. Let | : Z C
1
(1), 1 :
and c: C
0
(1) Z, e
}
1. Then the sequence
0 Z
t
C
1
(1)
J
C
0
(1)
t
Z 0
is exact. Hence | induces an isomorphism H
1
(C) Z.
Problems
1. Let 1 be the tetrahedral simplicial complex; it consists of 1 = {0. 1. 2. 3], and all subsets
are simplices. Verify H
i
(1) = 0 for n > 0. Generalize to an n-simplex.
12.4 The Euler Characteristic
Let X be a nite CW-complex and
i
(X) the number of its i -cells. The combina-
torial Euler characteristic of X is the alternating sum
(X) =
i_0
(1)
i
i
(X).
12.4. The Euler Characteristic 309
The fundamental and surprising property of this number is its topological invariance,
in fact its homotopy invariance it does not depend on the cellular decomposition
of the space. The origin is the famous result of Euler which says that in the case
X = S
2
the value (X) always equals 2 ([61], [62], [64]).
We prepare the investigation of the Euler characteristic by an algebraic result
about chain complexes. Let Mbe a category of 1-modules. An additive invariant
for Mwith values in the abelian group assigns to each module M in Man element
z(M) such that for each exact sequence 0 M
0
M
1
M
2
0 in M
the additivity
(1) z(M
0
) z(M
1
) z(M
2
) = 0
holds. For the zero-module M we have z(M) = 0 since there exists an exact
sequence 0 M M M 0. We consider only categories which contain
with a module also its submodules and its quotient modules as well as all exact
sequences between its objects. Let
C
+
: 0 C
k
J
k
C
k-1
C
1
J
1
C
0
J
0
0
be a chain complex in this category. Then its homology groups H
i
(C
+
) are also
contained in this category.
(12.4.1) Proposition. Let z be an additive invariant for M. Then for each chain
complex C
+
in M as above the following equality holds:
k
i=0
(1)
i
z(C
i
) =
k
i=0
(1)
i
z(H
i
(C
+
)).
Proof. Induction on the length k of C
+
. We set H
i
= H
i
(C
+
), T
i
= Imd
i1
,
7
i
= Ker d
i
. For k = 1 there exist, by denition of homology groups, exact
sequences
0 T
0
C
0
H
0
0. 0 H
1
C
1
T
0
0.
We apply the additivity (1) to both sequences and thereby obtain z(H
0
) z(H
1
) =
z(C
0
) z(C
1
). For the induction step we consider the sequences
C
t
+
: 0 C
k-1
C
0
0.
0 H
k
C
k
T
k-1
0. 0 T
k-1
7
k-1
H
k-1
0:
the last two are exact and the rst one is a chain complex. The homology groups
of the chain complex are, for k _ 2,
H
i
(C
t
+
) = H
i
(C
+
). 0 _ i _ k 2. H
k-1
(C
t
+
) = 7
k-1
.
We apply the induction hypothesis to C
t
+
and (1) to the other sequences. We obtain
the desired result by eliminating z(T
k-1
) and z(7
k-1
).
The relation of the combinatorial Euler characteristic to homology groups
goes back to Henri Poincar [150], [152]. The i -th Betti number, named after
Enrico Betti [20], b
i
(X) of X is the rank of H
i
(X: Z), i.e., the cardinality of a
basis of its free abelian part, or equivalently, the dimension of the Q-vector space
H
i
(X: Z) Q H
i
(X: Q). The result of Poincar says:
(12.4.2) Theorem. For each nite CW-complex X the combinatorial Euler char-
acteristic equals the homological Euler characteristic
i_0
(1)
i
b
i
(X).
Proof. For nitely generated abelian groups rank is an additive invari-
ant. We apply (12.4.1) to the cellular chain complex C(X) of X and observe that
rank C
i
(X) =
i
(X).
If z is an additive invariant for M and C
+
a chain complex of nite length in
M, then we call (C
+
) =

i_0
(1)
i
z(C
i
) =

i_0
(1)
i
z(H
i
(C
+
)) the Euler
characteristic of C
+
with respect to z.
(12.4.3) Proposition. Let
0 H
t
0
H
0
H
tt
0
H
t
1
H
1
H
tt
1
H
t
2
H
2

be an exact sequence of modules in M which consists eventually of zero-modules.
Let (H
+
) =
i_0
(1)
i
z(H
i
) and similarly for H
t
and H
tt
. Then
(H
t
+
) (H
+
) (H
tt
+
) = 0.
Proof. Apply (12.4.1) to the given exact sequence, considered as chain complex,
and order the terms according to H, H
t
, and H
tt
.
One candene the Euler characteristic byhomological methods for spaces which
are not necessarily nite CW-complexes. There are several possibilities depending
on the homology theory being used.
Let 1 be a principal ideal domain. We call (X. ) of nite 1-type if the groups
H
i
(X. : 1) are nitely generated 1-modules and only nitely many of them are
non-zero. In that case we have the associated homological Euler characteristic
(X. : 1) =
i_0
(1)
i
rank
T
H
i
(X. : 1).
(12.4.4) Proposition. If (X. ) is of nite Z-type, then it is of nite 1-type and the
equality (X. : Z) = (X. : 1) holds.
Proof. If (X. ) is of nite Z-type, then the singular complex S
-
(X. ) is chain
equivalent to a chain complex D
-
of nitely generated free abelian groups with
only nitely many of the D
n
non-zero (see (11.6.4)). Therefore
(X. : 1) =
i
(1)
i
rank
T
H
i
(D
-
1) =
i
(1)
i
rank
T
(D
i
1)
=
i
(1)
i
rank
Z
(D
i
) = (X. : Z).
by (12.4.1), and some elementary algebra.
12.5. Euler Characteristic of Surfaces 311
Proposition (12.4.3) has the following consequence. Suppose two of the spaces
, X, (X. ) are of nite 1-type. Then the third is of nite 1-type and the additivity
relation
(: 1) (X. : 1) = (X: 1)
holds. Let
0
.
1
be subspaces of X with MV-sequence, then
(
0
: 1) (
1
: 1) = (
0
L
1
: 1) (
0

1
: 1)
provided the spaces involved are of nite 1-type. Similarly in the relative case. Let
(X. ) and (Y. T) be of nite 1-type. Then the Knneth formula is used to show
that the product is of nite 1-type and the product formula
(X. : 1) (Y. T: 1) = ((X. ) (Y. T): 1)
holds. These relations should be clear for nite CW-complex and the combinatorial
Euler characteristic by counting cells.
For the more general case of Lefschetz invariants and xed point indices see
[51], [52], [109], [116].
12.5 Euler Characteristic of Surfaces
We report about the classical classication of surfaces and relate this to the Euler
characteristic. For details of the combinatorial or differentiable classication see
e.g., [167], [80], [123]. See also the chapter about manifolds.
Let J
1
and J
2
be connected surfaces. The connected sum J
1
#J
2
of these
surfaces is obtained as follows. Let D
}
J
}
be homeomorphic to the disk D
2
with boundary S
}
. In the topological sum J
1
D
1
J
2
D
2
we identify . S
1
with (.) S
2
via a homeomorphism : S
1
S
2
. The additivity of the Euler
characteristic is used to show
(J
1
) 1 (J
2
) 1 = (J
1
#J
2
).
i.e., the assignment J (J) 2 is additive with respect to the connected sum.
Let mJ denote the m-fold connected sum of J with itself. We have the standard
surfaces sphere S
2
, torus T , and projective plane 1. The Euler characteristics are
(S
2
) = 2. (mT ) = 2 2m. (n1) = 2 n.
If J is a compact surface with k boundary components, then we can attach k disks
D
2
along the components in order to obtain a closed surface J
+
. By additivity
(J
+
) = (J) k. Connected surfaces J
}
with the same number of bound-
ary components are homeomorphic if and only if the associated surfaces J
+
}
are
homeomorphic.
(12.5.1) Theorem. A closed connected surface is homeomorphic to exactly one of
the surfaces S
2
, mT with m _ 1, n1 with n _ 1. The n1 are the non-orientable
surfaces.
The homeomorphism type of a closed orientable surface is determined by the
orientation behaviour and the Euler characteristic. The homeomorphism type of
a compact connected surface with boundary is determined by the orientation be-
haviour, the Euler characteristic and the number of boundary components.
The sphere has genus 0, mT has genus m and n1 has genus n.
12.5.2 Platonic solids. A convex polyhedron is called regular if each vertex is the
end point of the same number of edges, say m, and each 2-dimensional face has
the same number of boundary edges, say n. If 1 is the number of vertices, 1 the
number of edges and J the number of 2-faces, then m1 = 21 and nJ = 21. We
insert this into the Euler relation 1 J = 1 2, divide by 21, and obtain
1
m
1
n
=
1
1
1
2
.
We have m _ 3, n _ 3. The equation has only the solutions which are displayed
in the next table.
m n 1 solid 1 J
3 3 6 tetrahedron 4 4
4 3 12 octahedron 6 8
3 4 12 cube 8 6
3 5 30 dodecahedron 20 12
5 3 30 icosahedron 12 20
12.5.3 Lines in the projective plane. Let G
1
. . . . . G
n
be lines in the projective
plane 1. We consider the resulting cells decomposition of 1. Let t
i
be the number
of points which are incident with r lines. We have the Euler characteristic relation
0

1

2
= 1 where
i
is the number of i -cells. Thus
0
= t
2
t
3
.
From an r-fold intersection point there start 2r edges. The sum over the vertices
yields
1
= 2t
2
3t
3
4t
4
. Let
n
denote the numbers of n-gons, then
2
=
x
. 2
1
=
s
x
. We insert these relations into the Euler characteristic
relation and obtain
i_2
(3 r)t
i

x_2
(3 s)
x
= 3
0

1
3
2
2
1
= 3.
We now assume that not all lines are incident with a single point; then we do not
have 2-gons. From 2
1
_ 3
2
and then
1
_ 3(
0
1) we conclude
t
2
_ 3
i_4
(r 3)t
i
.
Thus there always exist at least three double points.
(12.5.4) Proposition. Let X be a Hausdorff space and : X Y a local homeo-
morphism onto a connected space. Then is a covering with nitely many leaves
if and only if is proper.
Proof. (1) Suppose is proper. For n N let Y
n
= {, Y [ n _ [
-1
(,)[]. We
show that Y
n
is open and closed. Since Y is connected, either Y
n
= 0 or Y
n
= Y .
The inclusion Y
n
Y
n1
shows that there is a largest n such that Y
n
= Y and
Y
n1
= 0. Hence the bres have the cardinality n.
Let
-1
(,) = {.
1
. . . . . .
n
]. Since X is a Hausdorff space and a local
homeomorphism, there exist open pairwise disjoint sets U
i
.
i
which are mapped
homeomorphically under onto the same open set V ,. Hence each bre
-1
(:), : V has at least cardinality n. If , Y
n
, then m _ n and hence : Y
n
,
i.e., V Y
n
. This shows that Y
n
is open.
Let
-1
(,) = {.
1
. . . . .
t
], t < n, i.e., , , Y
n
. Let again the U
i
.
i
be open
with homeomorphic image V Y . Since is closed, being a proper map, the set
C = (X (U
1
L L U
t
)) is closed in Y . This set does not contain ,. Hence
Y C = W is an open neighbourhood of , and
-1
(W) =
-1
(Y )
-1
(X (U
1
L L U
t
)) U
1
L L U
t
.
This shows [
-1
(:)[ _ t for each : W, and the complement of Y
n
is seen to be
open.
We now know that all bres of have the same cardinality, and since is a
local homeomorphism it must be a covering.
(2) Suppose conversely that is an n-fold covering. We have to show that is
closed. A projection pr : T J T with a nite discrete set J as bre is closed.
Now we use (1.5.4).
A continuous map : X Y between surfaces is called a ramied covering if
for each . X there exist centered charts (U. . U
t
) about . and (V. [. V
t
) about
, = (.) with (U) V such that
[
-1
: (U) = U
t
C. : :
n
with n N. We call n 1 the ramication index of at .. In the case that n = 1
we say that is unramied at . and for n > 1 we call . a ramication point.
(12.5.5) Proposition. Let : X Y be a ramied covering between compact
connected surfaces. Let V be the image under of the ramication points. Then
: X
-1
() Y V is a covering with nitely many leaves.
Proof. For each , Y the set
-1
(,) X is closed and hence compact. The
pre-images
-1
(,) in a ramied covering are always discrete, hence nite. The
set V is also discrete and hence nite. The map is, as a continuous map between
compact Hausdorff spaces, closed. Thus we have shown that the map in question
is proper. Now we use (12.5.4).
(12.5.6) Proposition (RiemannHurwitz). Let : X Y be a ramied covering
between compact connected surfaces. Let 1
1
. . . . . 1
i
X be the ramication
points with ramication index (1
}
). Let n be the cardinality of the general bre.
Then for the Euler characteristics the relation
(X) = n(Y )
i
}=1
(1
}
)
holds.
Proof. Let Q
1
. . . . . Q
x
be the images of the ramication points. Choose pairwise
disjoint neighbourhoods D
1
. . . . . D
x
Y where D
}
is homeomorphic to a disk.
Then
: X
0
= X
_
x
}=1
-1
(D
}
) Y
_
}=1
D
}
= Y
0
is an n-fold covering (see (12.5.4)). We use the relation (X
0
) = n(Y
0
) for n-fold
coverings. If C is a nite set in a surface X, then (X C) = (X) [C[. Thus
we see
(X)
x
}=1
[
-1
(Q
}
)[ = n((Y ) s).
Moreover
ns
x
}=1
[
-1
(Q
}
)[ =
i
i=1
(1
i
).
since for
-1
(Q
}
) = {1
}
1
. . . . . 1
}
n(})
] the relation
n(})
t=1
((1
}
t
) 1) = n holds.
An interesting application of the RiemannHurwitz formula concerns actions

of nite groups on surfaces. Let J be a compact connected orientable surface and
G J J an effective orientation preserving action. We assume that this action
has the following properties:
(1) The isotropy group G
x
of each point . J is cyclic.
(2) There exist about each point . a centered chart : U R
2
such that U
is G
x
-invariant and transforms the G
x
-action on U into a representation
on R
2
, i.e., a suitable generator of G
x
acts on R
2
as rotation about an angle
2,[G
x
[.
In this case the orbit map : J J,G is a ramied covering, J,G is orientable,
and the ramication points are the points with non-trivial isotropy group. One can
showthat eachorientationpreservingactionhas the properties (1) and(2). Examples
are actions of a nite group G SO(3) on S
2
by matrix multiplication and of a
nite group G GL
2
(Z) on the torus T = R
2
,Z
2
by matrix multiplication.
The ramied coverings which arise as orbit maps from an action are of a more
special type. If . J is a ramication point, then so is each point in
-1
., and
these points have the same ramication index, since points in the same orbit have
conjugate isotropy groups. Let C
1
. . . . . C
i
be the orbits with non-trivial isotropy
group and let n
}
denote the order of the isotropy group of . C
i
; hence [C
i
[n
i
=
[G[. The RiemannHurwitz formula yields in this case:
12.5.7 RiemannHurwitz formula for group actions.
(J) = [G[
_
(J,G)
i
}=1
(1 1,n
}
)
_
.
In the case of a free action r = 0 and there is no sum.
12.5.8 Actions on spheres. Let J = S
2
and [G[ _ 2. Since (S
2
) = 2 we see
that (J,G) _ 0 is not compatible with 12.5.7, hence (J,G) = 2 and the orbit
space is again a sphere. We also see that r _ 3 and r = 0. 1 are not possible. For
r = 2 we have 2,[G[ = 1,n
1
1,n
2
, 2 = [C
1
[C
2
. Hence there are two xed
points (example: rotation about an axis). For r = 3 one veries that
1
2
[G[
=
1
n
1
1
n
2
1
n
3
has the solutions (for n
1
_ n
2
_ n
3
) displayed in the next table.
n
1
n
2
n
3
[G[
[G[,2 2 2 [G[
3 3 2 12
4 3 2 24
5 3 2 60
Examples are the standard actions of subgroups of SO(3), namely D
2n
(dihedral),
4
(tetrahedral), S
4
(octahedral),
5
(icosahedral). Up to homeomorphism there
are no other actions.
12.5.9 Action on the torus. Let J = T = S
1
S
1
be the torus, (J) = 0. The
RiemannHurwitz formula shows that for r _ 1 we must have (J,G) = 2. The
cases r _ 5 and r = 1. 2 are impossible. For r = 4 we must have n
1
= n
2
=
n
3
= n
4
= 2 and G = Z,2. For r = 3 the solutions of 12.5.7 are displayed in a
table.
n
1
n
2
n
3
3 3 3
2 3 6
2 4 4
Consider the matrices in SL
2
(Z)
=
_
0 1
1 1
_
.
_
1 1
1 0
_
.
The cyclic groups generated by .
2
.
3
realize cases 2 and 1 of the table and the
case r = 4 above. The matrix T realizes case 3 of the table.
Problems
1. Let Gact effectively on a closed orientable surface J of genus 2 preserving the orientation.
Then [G[ divides 48 or 10. There exist groups of orders 48 and 10 which act on a surface
of genus 2. The group of order 48 has a central subgroup C of order 2 and G,C is the
octahedral group S
4
acting on the sphere J,C. Study the solutions of 12.5.7 and determine
the groups which can act on J. Use covering space theory and work towards a topological
classication of the actions.
2. The nicest models of surfaces are of course Riemann surfaces. Here we assume known
the construction of a compact Riemann surface from a polynomial equation in two variables.
The equation ,
2
= (.) with 2g2 branch points denes a surface of genus g. Such curves
are called hyper-elliptic (g _ 2). It is known that all surfaces of genus 2 are hyper-elliptic.
A hyper-elliptic surface always has the hyper-elliptic involution 1(.. ,) = (.. ,). Here
are some examples. Let us write e(a) = exp(2i a).
(1) ,
2
= .(.
2
1)(.
2
4) has a Z,4-action generated by (.. ,) = (.. e(1,4),).
Note
2
= 1.
(2) ,
2
= (.
3
1),(.
3
8) has a Z,3-action generated by T(.. ,) = (e(1,3).. ,).
Since T commutes with 1, we obtain an action of Z,6.
(3) ,
2
= (.
3
1),(.
3
1) has a Z,6-action generated by C(.. ,) = (e(1,6).. 1,,).
Since C commutes with 1, we obtain an action of Z,6 Z,2. It has an action of Z,4
generated by D(.. ,) = (1,.. e(1,4),). Note D
2
= 1. The actions C and D do not
commute, in fact CD = DC
5
. Thus we obtain an action of a group J which is an extension
1 Z,2 J D
12
1
where Z,2 is generated by 1 and D
12
denotes the dihedral group of order 12.
(4) ,
2
= .(.
4
1) has the following automorphisms (see also [121, p. 94])
G(.. ,) = (e(1,4).. e(1,8),). G
8
= id. G
4
= 1.
H(.. ,) = (1,.. e(1,4),,.
3
). H
4
= id. H
2
= 1.
1(.. ,) = ((. i ),(. i ). 2
_
2e(1,8),,(. i )). 1
3
= 1.
The elements G. H. 1 generate a group of order 48. If we quotient out the central hyper-
elliptic involution we obtain the octahedral group of order 24 acting on the sphere. Thus
there also exists an action of a group of order 24 such that the quotient by the hyper-elliptic
involution is the tetrahedral group (and not the dihedral group D
12
, as in (3)).
(5) ,
2
= .
5
1 has an action of Z,5 generated by J(.. ,) = (e(1,5).. ,). It commutes
with 1 and gives an action of Z,10.
3. By an analysis of 12.5.7 one can show that, for an effective action of G on a closed
orientable surface of genus g _ 2, the inequality [G[ _ 84(g 1) holds. There exists a
group of order 168 which acts on a surface of genus 3 [63, p. 242].
Chapter 13
Partitions of Unity in Homotopy Theory
Partitions of unity and numerable coverings of a space are useful tools in order to
obtain global results from local data. (A related concept is that of a paracompact
space.) We present some notions about partitions of unity in the context of point-set
topology. Then we use them to show that, roughly, local homotopy equivalences
are global ones and a map is a bration if it is locally a bration (see the precise
statements in (13.3.1) and (13.4.1)). We apply the results to prove a theoremof Dold
about brewise homotopy equivalences (see (13.3.4)). Conceptually, partitions of
unity are used to relate the homotopy colimit of a covering to the colimit of the
covering; see (13.2.4) for a result of this type. There are many other results of this
type in the literature. This chapter only can serve as an introduction to this topic.
13.1 Partitions of Unity
Let t : X R be continuous. The closure of t
-1
(R 0) is the support supp(t )
of t . A family T = (t
}
: X R [ J) of continuous functions is said to be
locally nite if the family of supports (supp(t
}
) [ J) is locally nite. We call
T point nite if { J [ t
}
(.) = 0] is a nite set for each . X. We call a
locally nite T a partition of unity if the t
}
assume only non-negative values and
if for each . X we have

}J
t
}
(.) = 1. A covering U = (U
}
[ J) is
numerable if there exists a partition of unity T such that supp(t
}
) U
}
holds for
each J; the family T is then called a numeration of Uor a partition of unity
subordinate to U.
(13.1.1) Theorem. A locally nite open covering of a normal space is numerable.
Proof. Let U = (U
}
[ J) be a locally nite open covering of the normal space
X and V = (V
}
[ J) a shrinking of U and W = (W
}
[ J) a shrinking
of V . By the theorem of Urysohn, there exist continuous functions t
}
: X 0. 1|
which assume the value 1 on W
}
and the value 0 on the complement of V
}
. The
function t =

}J
t
}
: X 0. 1| is well-dened and continuous, since by local
niteness of V , in a suitable neighbourhood of a point only a nite number of t
}
are non-zero. We set
}
(.) = t
}
(.) t(.)
-1
. The functions (
}
[ J) are a
numeration of U.
(13.1.2) Lemma. Let the covering V = (V
k
[ k 1) be a renement of the
covering U = (U
}
[ J). If V is numerable, then also U is numerable.
13.1. Partitions of Unity 319
Proof. Let (
k
[ k 1) be a numeration of V . For each k 1 choose
a(k) J with V
k
U
o(k)
. This denes a map a: 1 J. We set g
}
(.) =
k,o(k)=}

k
(.); this is the zero function if the sum is empty. Then g
}
is con-
tinuous; the support of g
}
is contained in the union of the supports of the
k
with
a(k) = and is therefore contained in U
}
. Moreover, the sum of the g
}
is 1. The
family (g
}
[ J) is locally nite: If W is an open neighbourhood of . which
meets only a nite number of supports supp(
k
), k 1 1, 1 nite, then W
meets only the supports of the g
}
with a(1).
(13.1.3) Theorem. Each open covering of a paracompact space is numerable.
Proof. Let U = (U
}
[ J) be an open covering of the paracompact space X
and let V = (V
k
[ k 1) be a locally nite renement. Since X is normal, there
exists a numeration (
k
[ k 1) of V . Now apply the previous lemma.
(13.1.4) Lemma. Let (
}
: X 0. o [ J) be a family of continuous func-
tions such that U = (
-1
|0. o [ J) is a locally nite covering of X. Then U
is numerable and has, in particular, a shrinking.
Proof. Since U is locally nite, : . max(
}
(.) [ J) is continuous and
nowhere zero. We set g
}
(.) =
}
(.)(.)
-1
. Then
t
}
: X 0. 1|. . max(2g
}
(.) 1. 0)
is continuous. Since t
}
(.) > 0 if and only if g
}
(.) > 1,2, we have the inclusions
supp(t
}
) g
-1
}
1,4. o
-1
|0. o. For . X and i J with
i
(.) =
max(
}
(.)) we have t
i
(.) = 1. Hence the supports of the t
}
form a locally nite
covering of X, and the functions . t
i
(.),t (.), t (.) =

}J
t
}
(.) are a
numeration of U.
(13.1.5) Theorem. Let U = (U
}
[ J) be a covering of the space X. The
following assertions are equivalent:
(1) Uis numerable.
(2) There exists a family (s
o,n
: X 0. o [ a . n N) = S of continuous
functions s
o,n
(a) S, i.e., (s
-1
o,n
| 0. o), renes U.
(b) For each n the family (s
-1
o,n
|0. o [ a ) is locally nite.
(c) For each . X there exists (a. n) such that s
o,n
(.) > 0.
Proof. (1) =(2) is clear.
(2) = (1). (s
o,n
) is, by assumption, a countable union of locally nite fami-
lies. From these data we construct a locally nite family. By replacing s
o,n
with
s
o,n
,(1 s
o,n
) we can assume that s
o,n
has an image contained in 0. 1|. Let
q
i
(.) =
o, i~i
s
o,i
(.). r _ 1
320 Chapter 13. Partitions of Unity in Homotopy Theory
and q
i
(.) = 0 for r = 0. (The sum is nite for each . X.) Then q
i
and
o,i
(.) = max(0. s
o,i
(.) rq
i
(.))
are continuous. Let . X; then there exists s
o,k
with s
o,k
(.) = 0; we choose such
a function with minimal k; then q
k
(.) = 0,
o,k
(.) = s
o,k
(.). Therefore the sets
-1
o,k
|0. 1| also cover X. Choose N N such that N > k and s
o,k
(.) >
1
1
. Then
q
1
(.) >
1
1
, and therefore Nq
1
(,) > 1 for all , in a suitable neighbourhood of ..
In this neighbourhood, all
o,i
with r _ N vanish. Hence
(
-1
o,n
|0. 1| [ a . n N)
is a locally nite covering of X which renes U. We nish the proof by an appli-
cation of the previous lemma.
(13.1.6) Theorem. Let (U
}
[ J) be a numerable covering of T 0. 1|. Then
there exists a numerable covering (V
k
[ k 1) of T and a family (c(k) [ k 1)
of positive real numbers such that, for t
1
. t
2
0. 1|, t
1
< t
2
and [t
1
t
2
[ < c(k),
there exist a J with V
k
t
1
. t
2
| U
}
.
Proof. Let (t
}
[ J) be a numerationof (U
}
). For eachr-tuple k =(
1
. . . . .
i
)
J
i
, dene a continuous map
k
: T 1. .
i
i=1
min
_
t
}
i
(.. s) [ s
_
i-1
i1
.
i1
i1
__
.
Let 1 =
_
o
i=1
J
i
. We show that the V
k
=
-1
k
|0. 1| and c(k) =
1
2i
for k =
(
1
. . . . .
i
) satisfy the requirements of the theorem. Namely if [t
1
t
2
[ <
1
2i
,
there exists i with t
1
. t
2
|
i-1
i1
.
i1
i1
| and hence V
k
t
1
. t
2
| U
}
i
.
We show that (V
k
) is a covering. Let . T be given. Each point (.. t )
has an open neighbourhood of the form U(.. t ) V(.. t ) which is contained in a
suitable set W(i ) = t
-1
i
|0. 1| and meets only a nite number of the W( ). Suppose
V(.. t
1
). . . . . V(.. t
n
) cover the interval 1 = 0. 1|; let
2
i1
be a Lebesgue number
of this covering. We set U = U(.. t
1
) U(.. t
n
). Each set U
i-1
i1
.
i1
i1
| is
then contained in a suitable W(
i
). Hence . is contained in V
k
, k = (
1
. . . . .
i
).
There are only a nite number of J for which W( ) (U 1) = 0. Since
k
(.) = 0 implies the relation W(
i
) {.] 1 = 0, the family (V
k
[ k J
i
) is
locally nite for r xed. The existence of a numeration for (V
k
[ k 1) follows
now from theorem (13.1.5).
A family of continuous maps (t
}
: X 0. 1| [ J) is called a generalized
partition of unity if for each . X the family (t
}
(.) [ J) is summable with
sum 1.
13.2. The Homotopy Colimit of a Covering 321
(13.1.7) Lemma. Let (t
}
[ J) be a generalized partition of unity. Then
(t
-1
}
|0. 1| [ J) is a numerable covering.
Proof. Summability of (t
}
(a)) means: For each c > 0 there exists a nite set
1 J such that for all nite sets J 1 the inequality [1
}T
t
}
(a)[ > 1 c
holds. In that case V = {. [

}T
t
}
(.) > 1 c] is an open neighbourhood
of a. If k 1, . V and t
k
(.) > c, then t
k
(.)
}T
t
}
(.) > 1. This is
impossible. Hence for each a X there exists an open neighbourhood V(a) such
that only a nite number of functions t
}
have a value greater than c on V(a). Let
s
},n
(.) = max(t
}
(.) n
-1
. 0) for J and n N. By what we have just shown,
the s
},n
are locally nite for xed n. The claim now follows from (13.1.5).
It is a useful fact that arbitrary partitions of unity can be reduced to countable
ones. The method of proof is inspired by the barycentric subdivision of a simplicial
complex. Let U = (U
}
[ J) be a covering of the space 7 with subordinate
partition of unity T = (t
}
[ J). For each nite set 1 J we set q
T
(:) =
max(0. min
iT
t
i
(:) max
}T~S
t
}
(:)). The function q
e
is continuous, since T is
locally nite. From this denition one veries:
(13.1.8) Lemma. If q
T
(.) ,= 0 ,= q
T
(.), then either 1 J or J 1. The
family ([1[q
T
[ 1 J nite) is a locally nite partition of unity.
(13.1.9) Corollary. Let U(1) = q
-1
T
|0. 1|. Then U(1)U(J) ,= 0and [1[ = [J[
implies 1 = J. We set U
n
=

[T[=n
U(1) and dene t
n
: U
n
0. 1| by
t
n
[U(1) = [1[q
T
. Then (t
n
[ n N) is a numeration of (U
n
[ n N).
Suppose the functions t
}
are non-zero. Let N be the nerve of the covering
(t
-1
}
|0. 1| [ J). Then the nerve of the covering (q
-1
T
|0. 1| [ 1 J nite) is
the barycentric subdivision of N.
13.2 The Homotopy Colimit of a Covering
Let 1 = (V. S) be a simplicial complex. We consider it as a category: The objects
are the simplices, the morphisms are the inclusions of simplices. A contravariant
functor X: 1 TOP is called a simplicial K-diagram (in TOP). It associates to
each simplex s a space X
x
and to each inclusion t s a continuous map r
x
t
: X
x

X
t
. We also have a covariant functor ^: 1 TOP which ^(s) =

x
t
[
t
1.
= 1
_
, and for an inclusion t s we have the canonical inclusion
i
x
t
: ^(t ) ^(s). The geometric realization [X[ of a 1-diagramX is the quotient
of
x
X
x
^(s) with respect to the relation
X
x
^(s) (.. i
x
t
(a) ~ (r
x
t
(.). a)) X
t
^(t ).
Restriction to the n-skeleton 1
n
yields a functor X
n
: 1
n
TOP. In [X[ we have
the subspace [X[
n
which is the image of the X
x
^(s) with dims _ n. Since [X
n
[
is a quotient of the sumof these products we obtain a continuous map [X
n
[ [X[
n
.
(13.2.1) Proposition. The space [X[ is the colimit of the subspaces [X[
n
. The
canonical map [X
n
[ [X[
n
is a homeomorphism. There exists a canonical
pushout diagram
x,dimx=n
X
x
d^(s)

n
[X
n-1
[
x,dimx=n
X
x
^(s)

n
[X
n
[.
The attaching map
n
is dened as follows: d^(s) is the union of the i
x
t
^(t ), where
the t s have one element less than s. The map
n
is dened on X
x
i
x
t
^(s) by
r
x
t
(i
x
t
)
-1
composed with the canonical map into [X
n-1
[.
Let U = {U
}
[ J] be a covering of a space X. For each nite non-empty
1 J we write U
T
=
_
}
U
}
. We dene a subspace C(U) of X
}J
1
}
,
1
}
= 1, as the set of families , = (.: t
}
) such that:
(1) Only a nite number of the t
}
are non-zero.
(2)

}
t
}
= 1.
(3) If J(,) = { J [ t
}
,= 0] then . U
J(,)
.
We have coordinate maps pr = pr
C
: C(U) X, (.: t
}
) . and t
i
: C(U) 1,
(.: t
}
) t
i
. They are restrictions of the product projections and therefore continu-
ous. The t
}
form a point-nite partition of unity on C(U). We view C(U) via pr
C
as a space over X.
We dene a second space T(U) with the same underlying set but with a new
topology. Recall the nerve N(U) of the covering U. We have the simplicial N(U)-
diagram which associated to a simplex 1 of the nerve the space U
T
and to an
inclusion J 1 of simplices the inclusion r
T
T
: U
T
U
T
. The space T(U) is
the geometric realization of this N(U)-diagram. Thus T(U) is the quotient space
of
T
U
T
^(1) by the relation
U
T
^(1) (.. J
T
T
(a)) ~ (i
T
T
(.). a) U
T
^(J).
The sum is taken over the nite non-empty subsets 1 of J. Let ^(1)
be the
interior of ^(1) and d^(1) its boundary. Then each element of T(U) has a
unique representative of the form (.: t ) U
T
^(1)
for a unique 1. We can

interpret this element as an element of C(U), and in this manner we obtain a
bijection of sets j: T(U) C(U). This map is continuous, since the canonical
maps U
T
^(1) C(U) are continuous. The space T(U) has a projection
pr = pr
B
onto X and j is a map over X.
(13.2.2) Proposition. The map j is a homotopy equivalence over X.
Proof. We construct a map : C(U) T(U). For this purpose we choose a lo-
cally nite partition of unity(t
}
) subordinate to the open coveringt
-1
}
|0. 1| of C(U).
13.2. The Homotopy Colimit of a Covering 323
Then we dene
: , = (.: t
}
) (.: t
}
(,)) = :.
The map is well-dened and continuous: Let J(:), i.e., 0 ,= t
}
(:) = t
}
(,)
hence J(:) J(,) and . J(:); this shows that : T(U). Let W C(U) be an
open set such that J(W) = { [ t
}
[W ,= 0] is nite. Let , = (.: t
}
) W. Then
J(W) J(,), therefore factors on W through a map W U
J(W)
^(J(W)),
and this shows the continuity.
A homotopy j . id
C(U)
is dened by
(,. t ) = ((.: t
}
). t ) (.: t t
}
(1 t )t
}
).
This assignment is clearly well-dened and continuous. A homotopy j . id is
dened by the same formula (,: t ) (.: t t
}
(,) (1 t )t
}
j(,)). In order to
verify the continuity, we let again W be as above, but now considered as a subset
of T(U). We consider the composition with X
T
^(1) T(U). The formula
for the homotopy on the pre-image of W has an image in X
T
^(1).
Let T(U)
n
be the subspace of T(U) which is the image of the U
T
^(1) with
[1[ _ n 1. We state (13.2.1) for the special case at hand.
(13.2.3) Proposition. T(U) is the colimit of the sequence of subspaces T(U)
n
. The
canonical map
dimT_n
U
T
^(1) T(U)
n
is a quotient map. The inclusion
T
n-1
T
n
is obtained via a pushout diagram
dimT=n
U
T
d^(1)
k
n
T(U)
n-1
dimT=n
U
T
^(1)
1
n
T(U)
n
.
The map k
n
is dened on X
T
d^(1) as follows: Let J 1 be a proper subset.
Then k
n
is dened on X
T
^(J) by X
T
^(J) X
T
^(J) T
n-1
.
(13.2.4) Proposition. Let Ube numerable. Then the projections C(U) X and
T(U) X are shrinkable.
Proof. Let (t
}
[ J) a numeration of U. Then . (.: t
}
(.)) is a section
s of pr
C
and ((.: t
}
). t ) (.: t t
}
(1 t )t
}
(.)) is a homotopy from s pr
C
to
the identity over X. Thus pr
C
is shrinkable, and (13.2.2) shows that also pr
B
is
shrinkable.
For some applications we need a barycentric subdivision of T(U). Recall that
we have the barycentric subdivision N
t
(U) of the nerve of U. An n-simplex of
N
t
(U) is an ordered set t = (1
0
1
1
1
n
) such that U
T
n
,= 0. We
write q(t) = 1
n
. We have the N
t
(U)-diagram which associates to t the space
X
q(r)
and to o t the inclusion X
q(r)
X
q(c)
. Let T
t
(U) denote the geometric
realization of this N
t
(U)-diagram. Since the simplices are ordered, we can replace
^(t) by the standard simplex ^n| spanned by n| = {0. 1. . . . . n].
(13.2.5) Remark. In the case of the barycentric subdivision the pushout diagram
in (13.2.3) reads as follows:
r
n
X
q(r)
d^
n
k
n
}
n
T
t
(U)
n-1
J
n
r
n
X
q(r)
^
n
1
n
T
t
(U)
n
.
The sum is over the set
n
= {(o
0
. . . . . o
n
) [ o
0
o
n
. o
n
J nite], and
q(o
0
. . . . . o
n
) = o
n
.
13.3 Homotopy Equivalences
The main result (13.3.1) of this section asserts that being a homotopy equivalence
is in a certain sense a local property.
(13.3.1) Theorem. Let : X T and q : Y T be spaces over T and : X Y
a map over T. Let X = (X
}
[ J) be a covering of X and Y = (Y
}
[ J)
a covering of Y . Let (X
}
) Y
}
and assume that for each nite 1 J the
map
T
: X
T
Y
T
induced by is a homotopy equivalence over T. Then the
induced map T( ): T(X) T(Y) is a homotopy equivalence over T. Thus if
the coverings X and Y are numerable, then is a homotopy equivalence over T.
Proof. From (5.3.4) and (13.2.3) we prove inductively that the induced maps
T(X)
n
T(Y)
n
are h-equivalences. Now we use (5.2.9), in order to show that
T is an h-equivalence. In the case of numerable coverings we also use (13.2.4).
(13.3.2) Remark. In the situation of (13.3.1) we can conclude that is a homotopy

equivalence, if the projections
A
: T(X) X and
Y
are homotopy equivalences.
(13.3.3) Theorem. Let : X T and q : Y T be spaces over T and

: X Y a map over T. Let (U
}
[ J) be a numerable covering of T.
Let
}
:
-1
(U
}
) q
-1
(U
}
) be the map induced by over U
}
. Suppose each
}
is a brewise homotopy equivalence. Then is a brewise homotopy equivalence.
Proof. The hypothesis implies that is a brewise homotopy equivalence over
each set V V
}
. We can therefore apply (13.3.1).
13.4. Fibrations 325
We say that a covering (U
}
[ J) of T is null homotopic if every inclusion
U
}
T is null homotopic.
(13.3.4) Theorem. Let : X Y be a map over T from: X T to q : Y T.
Assume that and q are brations. Suppose T has a numerable null homotopic
covering (V
}
[ J) and that each path component of T contains a point b
such that is a homotopy equivalence over b. The is a brewise homotopy
equivalence.
Proof. Let V
}
T be homotopic to the constant map V
}
{b( )]. We can
assume that
b(})
: X
b(})
Y
b(})
is an h-equivalence. By the homotopy theorem
for brations we obtain a homotopy commutative diagram of maps over V
}
-1
(V
}
)
(
(1)
q
-1
(V
}
)
(2)
V
}
X
b(})
id (
b()
V
}
Y
b(})
with brewise homotopy equivalences (1) and (2). Hence
}
is a brewise equiva-
lence and we can apply (13.3.3)
Problems
1. For each J we let C(U)
= pr
-1
(U
) and similarly for T. Then the partial

projection maps pr
C
: C(U)
and pr
!
: T(U)
are shrinkable.
2. If the coverings in (13.3.1) are open, then
: 7. X| 7. Y | is for each paracompact

7 a bijection. The canonical projection : T(X) X induces for a paracompact space 7
a bijection
: 7. T(X)| 7. X|.
13.4 Fibrations
(13.4.1) Theorem. Let V = (V
}
[ J) be a covering of T and : 1 T
a continuous map. Assume that the map
}
:
-1
(T
}
) T
}
induced by is for
each J a bration. If the covering V is numerable, then is a bration. If the
covering V is open, then has the HLP for paracompact spaces.
Proof. We have to solve a homotopy lifting problem (left diagram)
X
o
1
]
Y

q
1
]
X 1
h
T X 1
h
T.
We form the pullback of along h (right diagram). The initial condition a yields
a section s
0
of q over X 0. A lifting of h with this initial condition amounts to
a section s of q which extends s
0
. We pull back the numerable covering of T to a
numerable coveringof X1. There exists a numerable coveringU = (U
k
[ k 1)
of X such that q is a bration over the sets U
k
1 (Problem 1). We begin by
constructing a lifting t : T(U) 1 1 of pr
B
which extends the lifting t
0
over
T(U) 0 determined by s
0
. The lifting is constructed inductively from partial
liftings t
n
over T(U)
n
1. The induction step is again based on the pushout
diagram (13.2.3), now multiplied by 1. The extension of the lifting t
n
amounts to
solving a lifting problem of the type
U
T
(d^(1) 1 L ^(1) 0)

Y
q
U
T
^(1) 1

j
j
j
j
j
j
j
j
j
j
X 1
and this is possible (by (5.5.3)), because U
T
^(1) 1 is mapped into a subset
over which q is a bration. If the covering U is numerable we compose it with a
section of pr
B
and obtain the desired extension of s
0
.
(13.4.2) Theorem. Let : Y X be a continuous map. Let Y = (Y
}
[ J)
be a family of subsets of Y and X = (X
}
[ J) a numerable covering of X.
Assume that (Y
}
) X
}
and that for nite 1 J the map
T
: Y
T
X
T
induced by is shrinkable. Then has a section. (Note that Y is not assumed to
be a covering of Y .)
Proof. We workwiththe barycentric subdivisionT
t
(X). We showthe existence of a
maps : T
t
(X) Y suchthat s = pr
B
. The proof does not use the numerabilityof
the covering. We construct inductively maps s
n
: T
t
(X)
n
Y with the appropriate
properties and an additional property which makes the induction work.
The map T
0
=

T
X
T
Y is given as follows: We choose sections X
T

Y
T
of
T
and compose them with the inclusion Y
T
Y .
Suppose s
n-1
is given. We want to extend
s
n-1
k
n
:

(X
q(r)
d^n|) 1
over
(X
q(r)
^n|). If t = (1
0
. . . . . 1
n
) we impose the additional hypothesis
that the image of X
q(r)
d^n| under s
n-1
k
n
is contained in Y
T
0
. The construction
of s
0
agrees with this requirement. Under this additional hypothesis we have a
commutative diagram
X
q(r)
d^n|
x
n-1
k
n
Y
T
0
X
q(r)

X
T
0
.
13.4. Fibrations 327
From (5.5.3) we see that s
n-1
k
n
can be extended over

X
q(r)
^n|. With an
extension we construct s
n
via the pushout (13.2.5). We show that s
n
satises the
additional hypothesis. Given t = (1
0
. . . . . 1
n1
) we describe
k
n1
: X
q(r)
d^n 1| T(U)
n
.
Let J
i
: ^n| ^n|
i
be the standard map onto the i -th face of ^n 1| with
inverse homeomorphism e
i
. Let d
i
: X
q(r)
X
q(t
i
r)
be the inclusion where
c
i
t = (1
0
. . . . . 1
i-1
. 1
i1
. . . . . 1
n1
). The restriction of k
n1
to the subset
X
q(r)
^n1|
i
is 1
n
(d
i
e
i
). By construction of s
n
the image of s
n
1
n
(d
i
e
i
)
is contained in X
T
0
(for i > 0) or X
T
1
(for i = 0). But X
T
1
X
T
0
, hence s
n
has
the desired property.
If X is numerable, then pr has a section t and st is a section of .
(13.4.3) Theorem. Let : X T be a continuous map and X = (X
}
[ J)
a numerable covering of X. Assume that for each nite 1 J the restriction
T
: X
T
T is a bration (an h-bration, shrinkable). Then is a bration (an
h-bration, shrinkable).
Proof. Recall that for a bration : X T the canonical map r : X
J
W()
is shrinkable (see (5.6.5)), and that is a bration if this map has a section (see
(5.5.1)). If the
T
are brations, then the r
T
: X
J
T
W(
T
) are shrinkable. The
W(
}
) form a numerable covering of W(). Theorem (13.4.2) shows that r has a
section, hence is a bration.
Assume that the
T
are shrinkable, i.e., homotopy equivalences over T. We
apply (13.3.1) and see that is shrinkable.
Assume that the
T
are h-brations. The map is an h-bration if and only if
the canonical map b : W() X is a homotopy equivalence over T (this can be
taken as a denition of an h-bration). The W(
}
) form a numerable covering of
W() and the b
T
: W(
T
) X
T
are homotopy equivalences over T, since
T
are h-brations. Thus we are in a position where (13.3.1) can be applied.
The hypothesis of (13.4.3) is satised if the X
T
are either empty or contractible.
Problems
1. If q : M N a. b| is a bration over N a. c| and N c. b|, then q is a bration.
2. Let X = (X
[ J) be a numerable covering of X. If the spaces X

1
have the
homotopy type of a CW-complex, then X has the homotopy type of a CW-complex.
3. Let : 1 T be an h-bration. Suppose T and each bre
-1
(b) have the homotopy
type of a CW-complex. Then 1 has the homotopy type of a CW-complex.
Chapter 14
Bundles
Bundles (also called bre bundles) are one of the main objects and tools in topology
and geometry. They are locally trivial maps with some additional structure. A
basic example in geometry is the tangent bundle of a smooth manifold and its
associated principal bundle. They codify the global information that is contained
in the transitions functions (coordinate changes).
The classication of bundles is reduced to a homotopy problem. This is achieved
via universal bundles and classifying spaces. We construct for each topological
group G the universal G-principal bundle EG BG over the so-called classifying
space BG. The isomorphism classes of numerable bundles over X are then in
bijection with the homotopy set X. BG|.
The classication of vector bundles is equivalent to the classication of their
associated principal bundles. A similar equivalence holds between n-fold covering
spaces and principal bundles for the symmetric group S
n
. This leads to a different
setting for the classication of coverings.
Fromthe set of (complex) vector bundles over a space X and their linear algebra
one constructs the Grothendieck ring 1(X). The famous Bott periodicity theorem
in one of its formulations is used to make the functor 1(X) part of a cohomology
theory, the so-called topological 1-theory. Unfortunately lack of space prevents us
from developing this very important aspect.
Classifying spaces and universal bundles have other uses, and the reader may
search in the literature for information.
The cohomology ring H
+
(BG) of the classifying space BG of a discrete group
G is also called the cohomology of the group. There is a purely algebraic theory
which deals with such objects.
If X is a G-space, one can form the associated bundle EG
G
X BG. This
bundle over BG can be interpreted as an invariant of the transformation group X.
The cohomology of EG
G
X is a module over the cohomology ring H
+
(BG)
(Borel-cohomology). The module structure contains some information about the
transformation group X, e.g., about its xed point set (see [7], [43]).
14.1 Principal Bundles
Let G be a topological group. In the general theory we use multiplicative notation
and denote the unit element of G by e. Let r : 1 G 1, (.. g) .g be a
continuous right action of G on 1, and : 1 T a continuous map. The pair
(. r) is called a (right) G-principal bundle if the following axioms hold:
14.1. Principal Bundles 329
(1) For . 1 and g G we have (.g) = (.).
(2) For each b T there exists an open neighbourhood U of b in T and a G-
homeomorphism :
-1
(U) U G which is a trivialization of over U
with typical bre G. Here G acts on U G by ((u. h). g) (u. hg).
If we talk about a G-principal bundle : 1 T, we understand a given action of G
on 1. From the axioms we see that G acts freely on 1. The map factors through
the orbit map q : 1 1,G and induces a continuous bijection h: 1,G T.
Since q and are open maps, hence quotient maps, h is a homeomorphism. Thus
G-principal bundles can be identied with suitable free right G-spaces. In contrast
to an arbitrary locally trivial map with typical bre G, the local trivializations in a
principal bundle have to be compatible with the group action. In a similar manner
we dene left principal bundles.
AG-principal bundle with a discrete group G is called a G-principal covering.
The continuity of the action r is in this case equivalent to the continuity of all
right translations r
: 1 1, . .g. This is due to the fact that 1 G is

homeomorphic to the topological sum l
G
1 {g], if G is discrete.
Let 1 G 1 be a free action and set C(1) = {(.. .g) [ . 1. g G].
We call t = t
T
: C(1) G, (.. .g) g the translation map of the action.
(14.1.1) Lemma. Let : 1 1,G be locally trivial. Then the translation map
is continuous.
Proof. Let W =
-1
(U) 1 be a G-stable open set which admits a trivialization
[: U G W. The pre-image of (W W)C(1) under [[ is {(u. g. u. h) [
u U. g. h G], and t ([ [) is the continuous map (u. g. u. h) g
-1
h.
A free G-action on 1 is called weakly proper if the translation map is continu-

ous. It is called proper if, in addition, C(1) is closed in 1 1.
(14.1.2) Proposition. A free action of G on 1 is weakly proper if and only if
0
t
: 1 G C(1), (.. g) (.. .g) is a homeomorphism.
Proof. The map [: C(1) 1 G, (.. ,) (.. t
T
(.. ,)) is a set-theoretical
inverse of 0
t
. It is continuous if and only if t
T
is continuous.
Let 1 carry a free right G-action and J a left G-action. We have a commutative
diagram
1 J
pr
1
1
]
1
G
J
q
1,G
with orbit maps 1 and and q = pr
1
,G.
330 Chapter 14. Bundles
(14.1.3) Proposition. A free right G action on 1 is weakly proper if and only if for
each left G-space J the diagram is a topological pullback.
Proof. We compare the diagram with the canonical pullback
X
b
1
]
1
G
J
q
1,G
with X = {((.. ). ,) (1
G
J) 1 [ (.) = (,)] and a = pr
1
, b =
pr
2
. There exists a unique map z: 1 J X such that bz = pr
1
, az = 1,
i.e., z(.. ) = ((.. ). .). The diagram in question is a pullback if and only
if z is a homeomorphism. Suppose this is the case for the left G-space G. The
homeomorphism 1
G
G 1, (.. g) .g transforms q into , X into C(1)
and z into (.. g) (.g. .). The latter is, in different notation, 0
t
. Hence 0
t
is a
homeomorphism if the diagram is a pullback for J = G.
Conversely, let the action be weakly proper. The map
j:

X = {((.. ). ,) [ (.) = (,)] 1J. ((.. ). ,) (,. t (.. ,)
-1
)
is continuous. One veries that j induces a map j: X 1 J. The equalities
jz(.. ) = j((.. ). .) = (.. t
-1
(.. .) ) = (.. )
show that j is an inverse of z.
(14.1.4) Proposition. Let G act freely and weakly properly on 1. The sections
of q : 1
G
J 1,G correspond bijectively to the maps : 1 J with the
property (.g) = g
-1
(.); here we assign to the section s
(
: . (.. (.)).
Proof. It should be clear that s
(
is a continuous section. Conversely, let s : T
1
G
J be a continuous section. We use the pullback diagram displayed before
(14.1.3). It yields an induced section o of pr
1
which is determined by the conditions
pr
1
o = id and 1 o = s . Let = pr
2
o : 1 J. Then pr
1
o(.g) =
.g = pr
1
(o(.)g), since o is a section and pr
1
a G-map. (The right action on
1 J is (e. . g) (eg. g
-1
).) The equalities 1o(.g) = s(.g) = s(.) =
1o(.) = 1(o(.)g) hold, since o is an induced section and 1 the orbit map. Since
o(.g) and o(.)g have the same image under 1 and pr
1
, these elements are equal;
we now apply the G-map pr
2
and obtain nally (.g) = g
-1
(.).
(14.1.5) Proposition. Let the free G-action on 1 be weakly proper. Then the orbit
map : 1 1,G = T is isomorphic to pr : T G T, if and only if has a
section.
Proof. Let s be a section of . Then T G 1, (b. g) s(b)g and 1
T G, . ((.). t (s.. .)) are inverse G-homeomorphisms, compatible with
the projections to T. Conversely, pr has a section and hence also the isomorphic
map .
(14.1.6) Proposition. Let X and Y be free G-spaces and : X Y a G-map. If
= ,G is a homeomorphismand Y weakly proper, then is a homeomorphism.
Proof. X is weakly proper, since the translation map of X is obtained from the
translation map of Y by composition with . We have to nd an inverse
: Y X. By (14.1.4), it corresponds to a section of
Y
: (Y X),G Y,G.
We have the section s : . (.. (.)) of
A
: (X Y ),G X,G. Let [ be the
inverse of . With the interchange map t : (X Y ),G (Y X),G we form
o = t s [. One veries that o is a section of
Y
.
Let a commutative diagram below with principal G-bundles and q be given,
and let J be a G-map. Then J or (J. ) is called a bundle map.
Y
T
X
]
C
(
T
If is a homeomorphism, then J is a homeomorphism (see (14.1.6)). If is the
identity, then J is called a bundle isomorphism.
Given a principal bundle : X T and a map : C T, we have a pullback
diagram as above with Y = {(c. .) [ (c) = (.)] C X. The maps q and
J are the restrictions of the projections onto the factors. The G-action on Y is
(c. .)g = (c. .g). If is trivial over V , then q is trivial over
-1
(V ). Therefore
q is a principal bundle, called the bundle induced from by . Also J is a bundle
map. Fromthe universal property of a pullback we see, that the bundle map diagram
above is a pullback.
(14.1.7) Proposition. Let U be a right G-space. The following are equivalent:
(1) There exists a G-map : U G.
(2) There exists a subset U such that m: G U, (a. g) ag is a
homeomorphism.
(3) The orbit map : U U,Gis G-homeomorphic over U,Gto the projection
pr : U,G G U,G.
(4) U is a free G-space, : U U,G has a section, and t
U
is continuous.
Proof. (1) = (2). Let =
-1
(e) and : U ,x . (.)
-1
(.). Then
(. ): U G is an inverse of m.
(2) =(3). The G-homeomorphism m induces a homeomorphism m,G of the
orbit spaces. We have the homeomorphismc: (G),G, a (a. e). With
these data m (c id) m,G = pr.
(3) = (4). If : U,G G U is a G-homeomorphism over U,G, then
the G-action is free and . (.. e) is a section of . The translation map of
U,G G is continuous and hence, via , also t
U
.
(4) = (1). Let s : U,G G be a section. Then U G, u t
U
(s(u). u)
is a G-map.
A right G-space U is called trivial if there exists a continuous G-map
: U G into the G-space G with right translation action. A right G-space
is called locally trivial if it has an open covering by trivial G-subspaces.
(14.1.8) Proposition. The total space 1 of a G-principal bundle is locally trivial.
If 1 is locally trivial, then 1 1,G is a G-principal bundle.
14.1.9 Hopf brations. Consider S
2n-1
C
n
as a free S
1
-space with action
induced from scalar multiplication. Let U
}
be the subset of points : = (:
k
) with
:
}
,= 0. The map : :
}
[:
}
[
-1
shows that U
}
is a trivial S
1
-space. The orbit
space of this action is C1
n-1
. The S
1
-principal bundle : S
2n-1
C1
n-1
,
i.e., the orbit map, is called a Hopf bration. There is a similar Z,2-principal
bundle S
n-1
R1
n-1
onto the real projective space and an S
3
-principal bundle
S
4n-1
H1
n-1
onto the quaternionic projective space.
(14.1.10) Proposition. Let : X Y be a G-map and
Y
: Y Y,G a G-
principal bundle. Then the diagram
X
(
Y
]
Y
X,G
({G
Y,G
is a pullback.
Proof. Let U Y be a G-set with a G-map h: U G. Then h :
-1
(U)
U G is a G-map. Hence
A
is a G-principal bundle. The diagram is therefore
a bundle map and hence a pullback.
We say, a map : X Y has local sections if each , Y has an open
neighbourhood V and a section s : V X of over V ; the latter means s() =
for all V .
(14.1.11) Proposition. Let G be a topological group and H a subgroup. The
quotient map q : G G,H is an H-principal bundle if and only if q has a section
over some neighbourhood of the unit coset.
Proof. A locally trivial map has local sections. Conversely, let s : U G be a
section of q over U. The map t
G
is continuous. For each k G we have the H-
equivariant map kq
-1
(U) H, kg s(q(g))
-1
g. We thus can apply (14.1.8)
and (14.1.7).
Actions with the properties of the next proposition were called earlier properly
discontinuous.
(14.1.12) Proposition. Let 1 G 1, (.. g) .g be a free right action of the
discrete group G. The following assertions are equivalent:
(1) The orbit map : 1 1,G is a G-principal covering.
(2) Each . 1 has a neighbourhood U, such that U Ug = 0 for each g ,= e.
(3) The set t
-1
(e) is open in C.
(4) The map t is continuous.
Proof. (1) =(4) holds by (14.1.1).
(4) =(3). The set {e] G is open, since G is discrete.
(3) = (2). We have t (.. .) = e. Since t
-1
(e) is open, there exists an open
neighbourhood U of . in 1 such that (U U) C t
-1
(e). Let U Ug = 0,
say = ug for u. U. Then (u. ) = (u. ug) (U U) C, hence
t (u. ug) = g = e.
(2) =(1). Let U be open. Then U G UG, (u. g) ug is a G-homeo-
morphism, hence UG an open trivial G-subspace.
(14.1.13) Example. Let G be a closed discrete subgroup of the topological group
1. Then the action G 1 1, (g. .) g. is free and has property (4) of the
previous proposition. Examples are Z R or Z C.
(14.1.14) Example. The map g: R S
1
, t exp(2i t ) has kernel Z. There-
fore there exists a bijective map : R,Z S
1
such that = g. Since g is an
open map, g is a quotient map and therefore a homeomorphism. By the previ-
ous example, g is therefore a covering. Similarly exp: C C
+
is seen to be a
covering.
(14.1.15) Example. Let G be a Lie group and H a closed subgroup. Then the quo-
tient map : G G,H is an H-principal bundle. In the chapter on differentiable
manifolds we show that G,H carries the structure of a smooth manifold such that
is a submersion. A submersion has always smooth local sections.
We construct locally trivial bundles from principal bundles. Let : 1 T
be a right G-principal bundle and J a left G-space. The projection 1 J 1
induces, via passage to orbit spaces, q : 1
G
J T. The map q is locally trivial
with typical bre J. A bundle chart :
-1
(U) U G of yields a bundle
chart
q
-1
(U) =
-1
(U)
G
J
G
id
(U G)
G
J U J
for q. We call q the associated bre bundle with typical bre J, and G is said to
be the structure group of this bre bundle. The structure group contains additional
information: The local trivializations have the property that the transition functions
are given by homeomorphisms of the bre which arise froman action of an element
of the group G.
Let : Y T be a right G-principal bundle. It may happen that there exists
a right H-principal bundle q : X Y for a subgroup H G and a G-homeo-
morphism j: X
1
G Y over T. In that case (q. j) is called a reduction
of the structure of . One can consider more generally a similar problem for
homomorphisms : H G.
(14.1.16) Example. Let 1 T be a G-principal bundle and H G a subgroup.
Then 1
G
G,H 1,H, (.. gH) .gH is a homeomorphism. Therefore
1,H 1,G, .H .G is isomorphic to the associated bundle 1
G
G,H
1,G. From a subgroup 1 H G we obtain in this manner G,1 G,H as
a bundle with structure group H and bre H,1, if G G,H has local sections.
If X is a G-space and H C G, then G,H acts on X,H by (.H. gH) .gH.
The quotient map X,H X,G induces a homeomorphism (X,H),(G,H)
X,G. In particular, 1,H 1,G becomes in this manner a G,H-principal
bundle.
One can topologise various algebraic notions in analogy to the passage from
groups to topological groups. An important notion in this respect is that of a small
category.
A (small) topological category C consists of an object space Ob(C) and a
morphism space Mor(C) such that the structure data
s : Mor(C) Ob(C) (source),
r : Mor(C) Ob(C) (range),
i : Ob(C) Mor(C) (identity),
c : Mor
2
(C) Mor(C) (composition)
are continuous. Here
Mor
2
(C) = {(. ) Mor(C) Mor(C) [ s() = r()]
carries the subspace topology of Mor(C)Mor(C). For these data the usual axioms
of a category hold. In a groupoid each morphism has an inverse. In a topological
14.2. Vector Bundles 335
groupoid we require in addition that passage to the inverse is a continuous map
Mor(C) Mor(C).
A topological group can be considered as a topological groupoid with object
space a point. Principal bundles yield a parameterized version.
(14.1.17) Example. Let 1 be a free right G-space with a weakly proper action.
Let : 1 T be a map which factors over the orbit map q : 1 1,G and
induces a homeomorphism 1,G T. We construct a topological groupoid with
object space T. The product : 1 1 T T factors over the orbit map
Q: 1 1 (1 1),G, (a. b) a. b| of the diagonal action and induces
(s. r): (1 1),G T T.
We dene (1 1),G as the morphism space of our category and s. r as source,
range. The diagonal of 1 induces i : T 1,G (11),G, and this is dened
to be the identity. We dene the composition by
a. b| .. ,| = .. b t (,. a)
-1
|
with the translation map t = t
T
of 1. One veries that composition is associative
and continuous. (The space Mor
2
is a quotient space of 1 C(1) 1.) The
morphism b. a| is inverse to a. b|, hence the inverse is continuous and we have
obtained a topological groupoid.
Problems
1. A free action of a nite group G on a Hausdorff space 1 is proper.
2. The action R
2
0R
R
2
0, ((.. ,). t ) (t.. t
-1
,) is a non-trivial R
-principal
bundle. Determine the orbits and the orbit space.
3. Let 1 be a space with a free right G-action. Then the translation map t
1
is continuous if
and only if the pullback of along is a trivial G-space.
4. The continuous maps 1
G
J
1
1
G
J
2
over 1,G correspond via (.. u)
(.. (.. u)) to the continuous maps : 1 J
1
J
2
with the equivariance condition
g(.. u) = (.g
-1
gu).
If 1 is connected and J
1
. J
2
discrete, then does not depend on . 1 and has the
form pr with a uniquely determined equivariant map : J
1
J
2
.
5. The groupoid (14.1.17) has further properties. There exists at least one morphism be-
tween any two objects. The map (s. r): (1 1),G T T is open.
6. Reconsider in the light of (14.1.17) the topological groupoid that was used in the deter-
mination of the fundamental groupoid of S
1
.
14.2 Vector Bundles
Vector bundles are, roughly speaking, continuous families of vector spaces. Sup-
pose given a continuous map : X T and the structure of an n-dimensional
R-vector space on each bre X
b
=
-1
(b). A bundle chart or a trivialization over
the open basic set U T for these data is a homeomorphism :
-1
(U) UR
n
over U which is brewise linear. A set of bundle charts is a bundle atlas, if their
basic domains cover T. The data : X T together with the vector space struc-
tures on the bres are an n-dimensional real vector bundle over the space T if a
bundle atlas exists. Thus a vector bundle is in particular a locally trivial map. In a
similar manner one denes complex vector bundles or quaternionic vector bundles.
A vector bundle : 1() T is called numerable, if there exists a numerable
covering U of T such that is trivial over the members U U. (This notion is
also used for other types of locally trivial bundles.) A bundle has nite type, if it
has a nite bundle atlas.
Let (U. ) and (V. [) be bundle charts for . Then the transition map is
[
-1
: (U V ) R
n
(U V ) R
n
. (.. ) (.. g
x
())
with g
x
GL
n
(R). The assignment g: U V GL
n
(R), . g
x
is continuous.
A bundle atlas is said to be orienting, if the g
x
have positive determinant. If
an orienting atlas exists, then the bundle is orientable. An orientation of a vector
bundle : 1 T consists of a vector space orientation of each bre
-1
(b) with
the property: For each . T there exists a bundle chart (U. ) about . such that
transports for each b U the given orientation on
-1
(b) into the standard
orientation of R
n
. A chart with this property is called positive with respect to the
given orientation. The positive charts forman orienting atlas, and for each orienting
atlas there exists a unique orientation such that its charts are positive with respect
to the orientation. A complex vector space has a canonical orientation. If one uses
this orientation in each bre, then the bundle, considered as a real bundle, is an
oriented bundle.
Let : 1() T and j: 1(j) C be real vector bundles. A bundle mor-
phism j over : T C is a commutative diagram
1()

1(j)
)
C
with a map which is brewise linear. If is bijective on bres, then we call the
bundle morphism a bundle map. Thus we have categories of vector bundles with
bundle morphisms or bundle maps as morphisms. The trivial n-dimensional bundle
is the product bundle pr : T R
n
T. More generally, we call a bundle trivial if
it is isomorphic to the product bundle.
(14.2.1) Proposition. A bundle map over the identity is a bundle isomorphism.
Proof. We have to show that the inverse of is continuous. Via bundle charts this
can be reduced to a bundle map between trivial bundles
U R
n
U R
n
. (u. ) (u. g
u
()).
In that case (u. ) (u. g
-1
u
()) is a continuous inverse, since u g
-1
u
is
continuous.
(14.2.2) Proposition. If the previous diagram is a pullback in TOP and j a vector
bundle, then there exists a unique structure of a vector bundle on such that the
diagram is a bundle map.
Proof. Since is bijective on bres, we dene the vector space structures in such
a way that becomes brewise linear. It remains to show the existence of bundle
charts.
If j is the product bundle, then can be taken as product bundle. If j has
a bundle chart over V , then has a bundle chart over
-1
(V ), by transitivity of
pullbacks.
We call in (14.2.2) the bundle induced from j along and write occasionally
=
+
j in this situation. The previous considerations show that a bundle map is
a pullback. (Compare the analogous situation for principal bundles.)
A bundle morphism over id(T) has for each b T a rank, the rank of the linear
map between the bres over b. It is then clear what we mean by a bundle morphism
of constant rank.
Asubset 1
t
1 of an n-dimensional real bundle : 1 T is a k-dimensional
subbundle of , if there exists an atlas of bundle charts (called adapted charts)
:
-1
(U) U R
n
such that (1
t
-1
(U)) = U (R
k
0). The restriction
: 1
t
T is then a vector bundle.
A vector bundle is to be considered as a continuous family of vector spaces. We
can apply constructions and notions from linear algebra to these vector spaces. We
begin with kernels, cokernels, and images.
(14.2.3) Proposition. Let :
1

2
be a bundle morphism over T of constant
rank and
b
the induced linear map on the bres over b. Then the following hold:
(1) Ker =
_
bB
Ker(
b
) 1(
1
) is a subbundle of
1
.
(2) Im() =
_
bB
Im(
b
) 1(
2
) is a subbundle of
2
.
(3) Suppose Coker() =
_
bB
1(
2
)
b
, Im(
b
) carries the quotient topology
from1(
2
). Then, with the canonical projection onto T, Coker() is a vector
bundle.
Proof. The problem in all three cases is the existence of bundle charts. This is a
local problem. Therefore it sufces to consider morphisms
: T R
n
T R
n
. (b. ) (b.
b
())
between trivial bundles.
We write 1
x
= Ker(
x
) and 1
x
= Im(
x
). We x b T and choose
complements R
n
= 1 1
t
and R
n
= 1 1
t
for 1 = 1
b
and 1 = 1
b
.
Let q : R
n
1 and : R
n
1 be the projections with Ker(q) = 1
t
and
Ker() = 1
t
. Then
;
x
: R
n
1
t
R
n
1. (. n) (
x
() n. q())
is anisomorphismfor . = b, hence alsoanisomorphismfor . ina neighbourhoodof
b. Thus let us assume without essential restriction that ;
x
is always an isomorphism.
Since
x
has constant rank k, we conclude that 1
x
1
t
= 0 and 1
x
1
t
= 0.
This fact is used to verify that
T R
n
T (1 1). (.. ) (..
x
(). q())
T (1
t
1
t
) T R
n
. (.. . n) (..
x
() n)
are brewise linear homeomorphisms. The rst one maps
_
xB
{.] 1
x
onto
T (0 1) and the second one T (1
t
0) onto
_
xB
{.] 1
x
. Moreover,
the second induces a bijection of T (0 1
t
) with
_
xB
{.] R
n
,1
x
. Thus we
have veried (1)(3) in the local situation.
14.2.4 Tangent bundle of the sphere. Let
TS
n
= {(.. ) [ (.. ) = 0] S
n
R
n1
with the projection : TS
n
S
n
onto the rst factor. The bre
-1
(.) is the
orthogonal complement of . in R
n1
. These data dene the tangent bundle of the
sphere. One can apply (14.2.3) to the family
x
: R
n1
R
n1
, (.. ). of
orthogonal projections.
Recall the stereographic projections
: S
n
{e
n1
] R
n
. The differential
of
-

-1
: R
n
0 R
n
0 at . is the linear map
R
n
R
n
.
[.[
2
2(.. ).
[.[
4
.
For [.[ = 1 we obtain the reection 2(.. ). at the hyperplane
orthogonal to .. Let U
=
-1
(S
n
{e
n1
]). The differential of
yields a
homeomorphism which is brewise linear and the diagram
U
R
n
R
n
pr
1
S
n
{e
n1
]

R
n
.
If we identify in R
n
R
n
R
n
R
n
the point (.. ) (R
n
0) R
n
in the
rst summand with
_
x
|x|
2
.
|x|
2
-2( x, )x
|x|
4
_
then we obtain a vector bundle over S =
(R
n
0 R
n
0),. ~ .[.[
-2
which is isomorphic to : TS
n
S
n
. We can
simplify the situation by identifying in D
n
R
n
D
n
R
n
the point (.. )
S
n-1
R
n
in the rst summand with (.. 2(.. ).) in the second summand.
In the tangent bundle we have the subspace of tangent vectors of length 1. In
our case this is the space {(u. ) S
n
S
n
[ (u. ) = 1], the Stiefel manifold
V
2
(R
n1
). We can obtain it from D
n
S
n-1
D
n
S
n-1
by the identica-
tion (.. ) ~ (.. 2(.. )). For even n we obtain from (10.7.8) the integral
homology of this Stiefel manifold.
For smallest structure groups of the tangent bundle of S
n
see [36]. The vector
eld problem [3] is a special case of this problem.
Important vector bundles in geometry are the tangent bundles of differentiable
manifolds and the normal bundles of immersed manifolds.
14.2.5 Tautological bundles. Let V be an n-dimensional real vector space and
G
k
(V ) the Grassmann manifold of the k-dimensional subspaces of V . We set
1
k
(V ) = {(.. ) [ . G
k
(V ). .] G
k
(V ) V.
We have the projections ;
k
(V ) = ;
k
: 1
k
(V ) G
k
(V ), (.. ) ., and the
bre over the element . G
k
(V ) is the subspace .. For this reason we call this
bundle the tautological bundle. It remains to verify that ;
k
is locally trivial. For
this purpose we recall the O(k)-principal bundle : S
k
(R
n
) G
k
(R
n
) from the
Stiefel manifold to the Grassmann manifold. The map
S
k
(R
n
)
O(k)
R
k
1
k
(R
n
).
((
1
. . . . .
k
). (z
1
. . . . . z
k
)) ((
1
. . . . .
k
).
z
}
}
)
is a brewise linear homeomorphism; it describes the tautological bundle as an
associated bre bundle.
Here is a different argument. Suppose . is spanned by (
1
. . . . .
k
) S
k
(R
n
);
then
x
: R
n
R
n
,

k
}=1
(.
}
)
}
is the orthogonal projection onto .. It
depends continuously on (
1
. . . . .
k
) and induces a continuous map G
k
(R
n
)
Hom(R
n
. R
n
), .
x
, and
G
k
(R
n
) R
n
G
k
(R
n
) R
n
. (.. ) (..
x
())
is a bundle morphism of constant rank with image 1
k
(R
n
). Now one can use
(14.2.3).
There exist analogous complex tautological bundles over the complex Grass-
mannians.
14.2.6 Line bundles over CP
n
. Let H: C
n1
0 C1
n
be the dening
C
+
-principal bundle. Let C(k) be the one-dimensional complex C
+
-representation
C
+
C C, (z. :) z
k
:. We obtain the associated complex line bundle
H(k) = (C
n1
0)
C
C(k) C1
n
.
Thus H(k) is the quotient of (C
n1
0)Cunder the equivalence relation (:. u) ~
(z:. z
k
u) for z C
+
. We also have the S
1
-principal bundle S
2n1
C
n
(the Hopf bundle). The inclusion S
2n1
C (C
n1
0) C induces a
homeomorphism
S
2n1
S
1 C(k) (C
n1
0)
C
C(k).
The inverse homeomorphism is induced by (:. u) ([:[
-1
:. [:[
-k
u). The
unit sphere bundle is S
2n1
S
1 S
1
(k). The assignment : (:. 1) induces a
homeomorphism
S
2n1
,C
[k[
S
2n1
S
1 S
1
(k).
Here C
n
S
1
is the subgroup of order m (roots of unity).
The bundles H(k) over C1
n
exist for 1 _ n _ o. We call the bundle H(1)
the canonical complex line bundle. For n = o it will serve as a universal one-
dimensional vector bundle.
The tautological bundle over C1
n
= G
1
(C
n1
) is H(1), since
(C
n1
0)
C
C(1) 1
1
(C
n1
). (.. u) (.|. u.)
is an isomorphism.
The sections of H(k) correspond to the functions : C
n1
0 C with the
property (z:) = z
k
(:), they are homogenous functions of degree k. This is the
reason to dene H(k) with C(k).
Let q : 1 T be a right G-principal bundle and V an n-dimensional repre-
sentation of G. Then the associated bundle : 1
G
V T is an n-dimensional
vector bundle. A bundle chart : q
-1
(U) U G for q induces a bundle chart
q
-1
(U)
G
V U V for , and the vector space structure on the bres of is
uniquely determined by the requirement that the bundle charts are brewise linear.
We nowshowthat vector bundles are always associated to principal bundles. Let
: X T be an n-dimensional real vector bundle. Let 1
b
= Iso(R
n
. X
b
) be the
space of linear isomorphisms. The group G = GL
n
(R) = Iso(R
n
. R
n
) = Aut(R
n
)
acts freely and transitively on 1
b
from the right by composition of linear maps. We
have the set map
q : 1() = 1 =
bB
1
b
T. 1
b
{b]
with brewise GL
n
(R)-action just explained. If :
-1
(U) U R
n
,
b
: X
b

R
n
is a bundle chart of , we dene
: q
-1
(U) =
bU
1
b
U Iso(R
n
. R
n
). 1
b
(b.
b
)
to be a bundle chart for q. The transition function for two such charts has the form
(U V ) Aut(R
n
) (U V ) Aut(R
n
). (b. ;) (b. [
b
-1
b
;).
This map is continuous, because b [
b
-1
b
is continuous. Therefore there
exists a unique topology on 1 in which the sets q
-1
(U) are open and the charts
homeomorphisms. The brewise GL
n
(R)-action on 1 now becomes continuous
and is equivariant. This shows q : 1 T to be a GL
n
(R)-principal bundle.
The evaluation Iso(R
n
. X
b
) R
n
X
b
, (. u) (u) induces an isomorphism
1()
GL
n
(R)
R
n
X() of vector bundles.
(14.2.7) Theorem. The assignment which associated to a GL
n
(R)-principal bundle
1 T the vector bundle 1
GL
n
(R)
R
n
T is an equivalence of the category of
GL
n
(R)-principal bundles with the category of n-dimensional real vector bundles;
the morphisms are in both cases the bundle maps.
Proof. The construction above shows that each vector bundle is, up to isomorphism,
in the image of this functor. The construction also associates to each bundle map
(J. ): j between vector bundles a bundle map between principal bundles
which is given on bres by
1()
b
= Iso(R
n
. X()
b
) Iso(R
n
. X(j)
((b)
) = 1(j)
((b)
.
and these isomorphisms are compatible with the original bundle maps, i.e., they
constitute a natural isomorphism. Therefore the functor is surjective on morphism
sets between two given objects. The injectivity is a consequence of the fact that a G-
map 1()
b
1(j)
((b)
is determined by the associated linear map 1()
b
G
R
n
1(j)
((b)
G
R
n
(where G = GL
n
(R)).
Problems
1. Determine an O(n)-principal bundle such that the associated vector bundle is the tangent
bundle of S
n
.
2. Consider in C
n1
the set of points (:
0
. . . . . :
n
) which satisfy the equations
:
2
0
:
2
1
:
2
n
= 0. [:
0
[
2
[:
1
[
2
[:
n
[
2
= 2.
Show that this space is homeomorphic to the Stiefel manifold V
2
(R
n1
). Show that V
2
(R
3
)
is the projective space R1
3
. Thus we have embedded this space into S
5
.
3. An n-dimensional real vector bundle is trivial if and only if it has n continuous sections
which are everywhere linearly independent.
4. The bundle H(k) over C1
1
is obtained from two trivial bundles pr
1
: C C C by
gluing over C
C with the transition maps (:. o) (:

-1
. :
-k
o).
5. The complex tangent bundle of C1
1
is H(2).
6. A bundle morphism J : of a real vector bundle which satises J
2
(.) = . for
each . 1() is called a complex structure on . If we dene in each bre the multiplica-
tion by i C as the map J, then becomes a complex vector bundle. One has to verify the
local triviality as a complex bundle.
7. An oriented real vector bundle is associated to a GL
n
(R)
-principal bundle. It is con-

structed as above using the orientation preserving isomorphisms Iso
(R
n
.
-1
(b)). There
exists an equivalence of categories analogous to (14.2.7).
Areductionof the structure groupfromGL
n
(R) toGL
n
(R)
corresponds tothe choice of

an orientation of the vector bundle. The reductions of a GL
n
(R)-bundle 1 T correspond
to the sections of 1,GL
n
(R)
T; the latter is a twofold covering, the orientation

covering.
8. The simplest non-trivial vector bundle is a line bundle over S
1
. Its total space X can
be viewed as the (open) Mbius band [140, Werke II, p. 484]. A formal denition is
X = S
1
G
R, where G = {1] acts on S
1
and R by (z. :) z:. It is also associated
to the G-principal bundle q : S
1
S
1
, : :
2
(where S
1
C). Suppose the bundle
were trivial. Then there would exist a nowhere vanishing section s : S
1
1, hence a map
o : S
1
R \ {0] satisfying o(:) = o(:). The latter contradicts the intermediate value
theorem of calculus.
In the same manner one constructs for each n _ 1 a non-trivial line bundle S
n
G
R
1
R1
n
= S
n
,G.
9. Let : X G,H be a G-map and =
-1
(eH). Then is a left H-space and we
have a bijective G-map J : G
!
X, (g. .) g.. The map J is a homeomorphismif
G G,H has local sections. If is a vector bundle and G acts by bundle automorphisms,
then is an H-representation.
14.3 The Homotopy Theorem
A locally trivial bundle is called numerable if it is trivial over the members of
a numerable covering of the base space. We show that homotopic maps induce
isomorphic bundles. We begin with the universal situation of a homotopy.
(14.3.1) Theorem. Let : 1 T 1 be a numerable, locally trivial bundle with
typical bre J. Then there exists a bundle map 1: 1 1 over r : T1 T1,
(b. t ) (b. 1) which is the identity on 1[T 1 and the morphism (1. r) is a
pullback.
Proof. (1) By (13.1.6), (3.1.4), and (13.1.8) we choose a numerable countable
covering (U
}
[ N) of T such that is trivial over U
}
1. We then choose
a numeration (t
}
) of (U
}
). Let t (.) = max(t
}
(.)) and set u
}
(.) = t
}
(.),t (.).
Then the support of u
}
is contained in U
}
and max{u
}
(.) [ N] = 1 holds. Let
r
}
: T 1 T 1. (.. t ) (.. max(u
}
(.). t )).
We dene over r
}
a bundle map 1
}
: 1 1: It is the identity in the complement
of
-1
(U
}
1), and over U
}
1 a trivialization U
}
1 G 1[U
}
1 transforms
14.3. The Homotopy Theorem 343
it into
(.. t. g) (.. max(u
}
(.). t ). g).
Then 1
}
is the identity on 1[T 1. From the construction we see that (1
}
. r
}
) is
a pullback. The desired bundle map 1 is the composition of the 1
}
according to
the ordering of N. This is sensible, since for each . 1 only a nite number of
1
}
(.) are different from .. The condition max{u
}
(.) [ J] = 1 shows that 1
is a map over r.
If we apply the previous proof to principal bundles (to vector bundles), then
(1. r) is a bundle map in the corresponding category of bundles.
Let : 1 T 1 be as in (14.3.1) and denote by
t
: 1
t
T t T
its restriction to T t . We obtain from (14.3.1) a pullback (1. r):
1
. The
map r induces from the product bundle
1
id: 1
1
1 T 1. We conclude
that there exists an isomorphism 1 1
1
1 of bundles which is the identity
1
1
= 1
1
1 over T 1.
(14.3.2) Theorem. Under the assumptions of (14.3.1) the bundles 1
0
and 1
1
are
isomorphic.
Proof. We have bundle maps 1
0
= 1[T 0 1 1
1
1
pr
1
1
.
14.3.3 Homotopy Theorem. Let q : 1 C be a numerable G-principal bundle
and h: T 1 C a homotopy. Then the bundles induced from along h
0
and
h
1
are isomorphic. A similar statement holds for vector bundles.
Proof. This follows from the previous theorem, since h
+
}
q = (h
+
q)
}
.
14.3.4 Homotopy lifting. Let
X

Y
q
C
be a bundle map between numerable G-principal bundles. Let h: T 1 C
be a homotopy with h
0
= . Then there exists a homotopy of bundle maps
H: X 1 Y with H
0
= and q H = h ( id).
Proof. There exists a diagram
X
t
Y
q
T
i
0
T 1
h
C
with two pullback squares and hi
0
= and

h| = . There exists an isomorphism
: X 1 7 of id with Q such that i
0
= |. The desired homotopy is
H =

h .
(14.3.5) Theorem. Let q : X C be a numerable locally trivial map.Then q is a
bration.
Proof. Given a homotopy h: T 1 C and an initial condition a: T X. We
pull back the bundle along h. The initial condition gives a section of the pullback
bundle over T0. We have seen that the bundle over a product T1 is isomorphic
to a product bundle, and in a product the section has an obvious extension to T 1.
We have remarked earlier that the extendibility of the section is equivalent to nding
a lifting of the homotopy with given initial condition.
14.4 Universal Bundles. Classifying Spaces
We denote by B(T. G) the set of isomorphism classes of numerable G-principal
bundles over T. (This is a set!) A continuous map : T C induces via
pullback a well-dened map B( ) =
+
: B(C. G) B(T. G). We thus obtain
a homotopy invariant functor B(. G).
Let
G
: EG BG be a numerable G-principal bundle and T. BG| the set of
homotopy classes T BG. Since homotopic maps induce isomorphic bundles,
we obtain a well-dened map
|
B
: T. BG| B(T. G). |
+
G
|.
The |
B
constitute a natural transformation.
We call the total space EG universal if each numerable free G-space 1 has up
to G-homotopy a unique G-map 1 EG. (Thus EG is a terminal object in the
appropriate homotopy category.) The corresponding bundle
G
: EG BG is also
called universal.
Let : 1() T be a numerable G-principal bundle. Then there exists a
G-map : 1() EG and an induced map

: T BG; and G-homotopic
maps induce homotopic maps between the base spaces. We assign to the class
| T. BG|. Isomorphic bundles yield the same homotopy class. Thus we

obtain a well-dened map k
B
: B(T. G) T. BG|, and the k
B
constitute a natural
transformation. The compositions |
B
k
B
and k
B
|
B
are the identity.
If
t
: E
t
G B
t
G is another universal bundle, then there exist bundle maps
: EG E
t
G, ; : E
t
G EG. The compositions ; and ; are homotopic to
the identity as bundle maps. In particular, the spaces BG and B
t
G are homotopy
equivalent. The space BG is called a classifying space of the group G. A map
k : T BG which induces from EG BG a given bundle q : 1 T is called a
classifying map of the bundle q. Hence:
14.4. Universal Bundles. Classifying Spaces 345
(14.4.1) Theorem (Classication Theorem). We assign to each isomorphism class
of numerable G-principal bundles the homotopy class of a classifying map and ob-
tain a well-dened bijection B(G. T) T. BG|. The inverse assigns to k : T
BG the bundle induced by k from the universal bundle.
(14.4.2) Theorem. There exist universal G-principal bundles.
The proof of the theorem will be given in three steps.
(1) Construction of the space EG (14.4.3).
(2) Proof that any two G-maps 1 EG are G-homotopic (see (14.4.4)).
(3) Proof that each numerable G-space 1 admits a G-map (see (14.4.5)).
14.4.3 The Milnor space. We present a construction of the universal bundle which
is due to Milnor [131]. It uses the notion of a join of a family of spaces. Let
(X
}
[ J) be a family of spaces X
}
. The join
X =
}J
X
}
is dened as follows. The elements of X are represented by families
(t
}
.
}
[ J). t
}
0. 1|. .
}
X
}
.
}J
t
}
= 1
in which only a nite number of t
}
are different from zero. The families (t
}
.
}
) and
(u
}
,
}
) represent the same element of X if and only if
(1) t
}
= u
}
for each J,
(2) .
}
= ,
}
whenever t
}
= 0.
The notation t
}
.
}
is short-hand for the pair (t
}
. .
}
). This is suggestive, since we
can replace 0.
}
by 0,
}
for arbitrary .
}
and ,
}
in X
}
. We therefore have coordinate
maps
t
}
: X 0. 1|. (t
i
.
i
) t
}
.
}
: t
-1
}
|0. 1| X
}
. (t
i
.
i
) .
}
.
The Milnor topology on X shall be the coarsest topology for which all t
}
and
}
are
continuous. This topology is characterized by the following universal property: A
map : Y X from any space Y is continuous if and only if the maps t
}
: Y
0. 1| and
}
:
-1
t
-1
}
|0. 1| X
}
are continuous. For a nite number of spaces
we use the notation X
1
= = X
n
for their join.
If the spaces X
}
are right G-spaces, then ((t
}
.
}
). g) (t
}
.
}
g) denes a con-
tinuous action of G on X. Continuity is veried with the universal property of the
join topology. The Milnor space is
EG = G = G = G = .
a join of a countably innite number of copies of G. We write BG = EG,G for
the orbit space and : EG BG for the orbit map.
It remains to show that EG BG is numerable. The coordinate functions t
}
are G-invariant and induce therefore functions t
}
on BG. The t
}
are a point-nite
partition of unity subordinate to the open covering by the V
}
,G, V
}
= t
-1
}
|0. 1|. The
bundle is trivial over V
}
,G, since we have, by construction, G-maps
}
: V
}
G.
(14.4.4) Proposition. Let 1 be a G-space. Any two G-maps . g: 1 EG are

G-homotopic.
Proof. We consider the coordinate form of (.) and g(.),
(t
1
(.)
1
(.). t
2
(.)
2
(.). . . . ) and (u
1
(.)g
1
(.). u
2
(.)g
2
(.). . . . ).
and show that and g are G-homotopic to maps with coordinate form
(t
1
(.)
1
(.). 0. t
2
(.)
2
(.). 0. . . . ) and (0. u
1
(.)g
1
(.). 0. u
2
(.)g
2
(.). . . . )
where 0 denotes an element of the form 0 ,. In order to achieve this, for say,
we construct a homotopy in an innite number of steps. The rst step has in the
homotopy parameter t the form
(t
1
1
. t t
2
2
. (1 t )t
2
2
. t t
3
3
. (1 t )t
3
3
. . . . ).
It removes the rst zero in the nal result just stated. We now iterate this process
appropriately. We obtain the desired homotopy by using the rst step on the interval
0.
1
2
|, the second step on the interval
1
2
.
3
4
|, and so on. The total homotopy is
continuous, since in each coordinate place only a nite number of homotopies are
relevant.
Having arrived at the two forms above, they are nowconnected by the homotopy
((1 t )t
1
1
. t u
1
g
1
. (1 t )t
2
2
. t u
2
g
2
. . . . ) in the parameter t .
(14.4.5) Proposition. Let 1 be a G-space. Let (U
n
[ n N) be an open covering
by G-trivial sets. Suppose there exists a point-nite partition of unity (
n
[ n N)
by G-invariant functions subordinate to the covering (U
n
). Then there exists a
G-map : 1 EG.
A numerable free G-space 1 admits a G-map 1 EG.
Proof. By denition of a G-trivial space, there exist G-maps
}
: U
}
G. The
desired map is now given by (:) = (
1
(:)
1
(:).
2
(:)
2
(:). . . . ). It is contin-
uous, by the universal property of the Milnor topology.
In order to apply the last result to the general case, we reduce arbitrary partitions
of unity to countable ones (see (13.1.8)).
(14.4.6) Proposition. The space EG is contractible.
Proof. We have already seen that there exists a homotopy of the identity to the
map (t
}
g
}
) (t
1
g
1
. 0. t
2
g
2
. t
3
g
3
. . . . ). The latter map has the null homotopy
((1 t )t
1
g
1
. t e. (1 t )t
2
g
2
. . . . ).
(14.4.7) Example. The locally trivial map : EG BG is a bration with con-
tractible total space by (14.3.5). We also have the path bration 1 BG with
contractible total space and bre CBG. We can turn a homotopy equivalence
EG 1 into a brewise map, and this map is then a brewise homotopy equiva-
lence. Hence we have a homotopy equivalence CBG . G. The exact homotopy
sequence then yields an isomorphism d:
n
(BG)
n-1
(G).
(14.4.8) Example. For a discrete group, BG is an EilenbergMac Lane space of
type 1(G. 1). The space BS
1
is an EilenbergMac Lane space 1(Z. 2). Models
for TZ,2 and BS
1
are R1
o
and C1
o
, respectively.
A continuous homomorphism : 1 1 induces the map
1(): EK EL. (t
i
k
i
) (t
i
(k
i
))
which is compatible with the projections to the classifying spaces. We obtain an
induced map T(): BK BL. In this manner T becomes a functor from the
category of topological groups into TOP.
(14.4.9) Proposition. An inner automorphism : 1 1, k uku
-1
induces a
map T() which is homotopic to the identity.
Proof. The map (t
i
k
i
) (t
i
uk
i
) is a 1-map and therefore 1-homotopic to the
identity. The assignment (t
i
k
i
) (t
i
uk
i
u
-1
) induces the same map between the
orbit spaces.
(14.4.10) Proposition. Let X be a free numerable G-space. Then the join 1 =
X = X = is a universal G-space.
Proof. As in the proof of 14.4.3 we see that any two G-maps into 1 are G-
homotopic. Since X is numerable, so is 1.
(14.4.11) Corollary. Let H be a subgroup of G. Assume that G is numerable as
H-space. Then EG is, considered as H-space, universal.
The next theorem characterizes universal bundles so that we need not rely on a
special construction.
(14.4.12) Theorem. A numerable G-principal bundle q : 1 T is universal if
and only if 1 is contractible (as a space without group action).
Proof. We know already that Milnors space EG is contractible. If is universal,
the G-space 1 is G-homotopy equivalent to EG and hence contractible.
Conversely, assume that 1 is contractible. Then the associated bre bundle
1
G
EG T has a contractible bre EGand is therefore shrinkable (use (13.3.3)).
Hence it has a section and any two sections are homotopic as sections. A section
corresponds to a bundle map : 1 EG (see (14.1.4). For the same reason there
exists a bundle map : EG 1. By 14.4.3, is homotopic to the identity as a
bundle map. In order to see that is homotopic id(1), we use that sections are
homotopic.
We compare classifying spaces of different groups and discuss the functorial
properties of classifying spaces. Let : 1 1 be a continuous homomorphism
between topological groups. We denote by

1 the 1-space
1 1 1. (k. l) (k) l.
The associated bundle 1(1)
1
1 T(1) inherits a right 1-action and is an
1-principal bundle. It has a classifying map T(): T(1) T(1). For the Milnor
bundle the homotopy class is the same as the one already dened. If : 1 M is
a further homomorphism, then the relation T()T() . T() is easily veried.
Let i : H G be the inclusion of a subgroup. We restrict the G-action to H
and obtain a free and contractible H-space res
1
EG. If G G,H is a numerable
H-principal bundle, then res
1
EG is numerable as H-space; hence we have in
this case in res
1
EG (res
1
EG),H as model for 1H BH. We then obtain,
because of EG
G
H EG,H, a map
Ti : BH = (EG),H (EG),G = BG.
which is a bre bundle with bre G,H. If G,H is contractible, then Ti is a
numerable bration with contractible bre, hence a homotopy equivalence. This
situation occurs for the inclusions O(n) GL
n
(R) and U(n) GL
n
(C), and in
general for the inclusion 1 G of a maximal compact subgroup 1 of a connected
Lie group G [84, p. 180].
(14.4.13) Proposition. The inclusions of subgroups induce homotopy equivalences
TO(n) TGL
n
(R) and TU(n) TGL
n
(C).
Let H be a normal subgroup of G. Then 1(G,H) 1(G) is a numerable
free G-space; hence (1(G,H) EG),G is a model for BG. (In general, for each
G-space X which is contractible, the product X EG is another model for EG.)
With this model and the orbit map of the projection 1(G,H) EG 1(G,H)
we obtain a map : BG T(G,H) which is a bre bundle with structure group
G,H and bre BH. In this case BH = EG,H with induced G,H-action. The map
and the inclusion i : BH BG of a bre are induced maps of the type Tj for
the cases i : H G and j: G G,H. Therefore we have a bre bundle
BH
Bi
BG
B]
T(G,H).
Principal bundles for a discrete group G are covering spaces; this holds in
particular for the universal bundle EG BG. Since EG is simply connected, it is
also the universal covering of BG. Thus two notions of universal meet. What is
the relation between these concepts?
Let T be a pathwise connected space with universal covering : 1 T, a
right -principal covering for =
1
(T). Let : G be a homomorphism.
We have as before the right G-principal covering 1
t
G. If and [ are conjugate
homomorphisms, i.e., if g(a)g
-1
= [(a) for a g G, then
1
t
G 1
t )
G. (.. h) (.. gh)
is anisomorphismof G-principal coverings. The assignment 1
t
Gis a map
: Hom(. G)
c
B(G. T) fromthe set of conjugacy classes of homomorphisms
(index c) to the set of isomorphism classes of G-principal bundles over T.
(14.4.14) Proposition. The map is a bijection.
(14.4.15) Example. Let G be discrete and abelian. Then conjugate homomor-
phisms are equal. If all coverings of T are numerable (say T paracompact), then
B(T. G) = T. BG|. A bijection
: Hom(. G) T. BG| = H
1
(T: G)
is obtained from (14.4.14). For the last equality note that BG is an EilenbergMac
Lane space and represents the rst cohomology.
(14.4.16) Example. The bration U(n),U(n 1) TU(n 1) TU(n) and
U(n),U(n 1) S
2n-1
show that the map i(n): TU(n 1) TU(n) induced
by the inclusion U(n 1) U(n) is (2n 1)-connected. The induced map
i(n)
+
: X. T(n 1)| X. TU(n)| is therefore bijective (surjective) for a CW-
complex X of dimension dimX < 2n1 (dimX _ 2n1). So if dimX _ 2n 2,
then a k-dimensional complex vector bundle over X is isomorphic to j(k n)c
for a unique isomorphism class j of an n-dimensional bundle j.
Problems
1. Work out a proof of (14.4.14).
2. The canonical diagram
1

CBK
:!()
1

CBL
is homotopy commutative. See (14.4.7).
3. The abelianized group
1
(T) is isomorphic to the homology group H
1
(T: Z). Therefore
we can write (14.4.15) in the form : Hom(H
1
(T). G) H
1
(T: G). The classifying map
: T BG of a G-principal bundle q : X T induces homomorphisms
:
1
(X)
1
(T) and
: H
1
(T) H
1
(BG). If G is abelian and discrete, then G
1
(BG)
H
1
(BG). We thus obtain a map
: T. BG| Hom(H
1
(T). G).
Under the hypotheses of (14.4.15), is inverse to .
4. We give an example of a non-numerable bundle. The equation .: ,
2
= 1 yields a
hyperboloid Q in R
3
. The action of the additive group R on R
3
c (.. ,. :) = (.. , c.. : 2c, c
2
.).
is free on Q, and Qbecomes an R-principal bundle. Numerable R-principal bundles are triv-
ial, since Ris contractible. The bundle Qis non-trivial. The orbit space is the non-Hausdorff
line with two origins. If the bundle were trivial it would have a section, and this would imply
that the orbit space is separated.
5. The join S
m
= S
n
mn1
, (t
1
:
1
. t
2
:
2
) (
_
t
1
:
1
.
_
t
2
:
2
) is
a homeomorphism. The join of k copies S
1
2k-1
. A suitable
homeomorphism respects the S
1
-action, if we let S
1
act on S
2k-1
by scalar multipli-
cation (z. ) z. The Milnor construction thus yields in this case the Hopf bundle
S
2k-1
C1
k-1
.
6. A suitable isomorphism
n
(BG)
n-1
(G) has the following interpretation. Let
: 1 S
n
be a G-principal bundle. Write S
n
= D
LD
-
, S
n-1
= D
D
-
as usual.
Then [D
and [D
-
are trivial. Choose trivializations t
:
-1
(D
) D
G. They
differ over S
n-1
by an automorphism
S
n-1
G S
n-1
G. (.. g) (..
\
(g))
of principal bundles. Hence
\
(g) =
\
(e)g, and .
\
(e) represents an element in
n-1
(G) which corresponds under the isomorphism in question to the classifying map of
.
7. The canonical map S
c
C1
c
is an S
1
-principal bundle with contractible total space.
Hence C1
c
is a model for BS
1
. This space is also an EilenbergMac Lane space of type
1(Z. 2). In a similar manner one has T(Z,2) = R1
c
= 1(Z,2. 1).
8. Suppose the X
are Hausdorff spaces. Then their join is a Hausdorff space.

9. The map (X
1
= X
2
) = X
3
X
1
= X
2
= X
3
which sends (u
1
(t
1
.
1
. t
2
.
2
). u
2
.
3
) to
(u
1
t
1
.
1
. u
1
t
2
.
2
. u
2
.
3
) is a homeomorphism. Discuss in general the associativity of the
join.
10. The join of a family (X
[ J) is a subspace of the product

J
CX
of cones,
when the cone CX = 1 X,0 X is given the Milnor topology with coordinate functions
t : (.. t ) t and t
-1
|0. 1| X, (.. t ) . (and not the quotient topology as previously
used in homotopy theory).
11. As a set, X
0
= = X
n
is a quotient of X
0
X
n
^n|. If the X
are compact
14.5. Algebra of Vector Bundles 351
Hausdorff spaces, then the join carries the quotient topology. In general the two topologies
yield h-equivalent spaces.
12. A theorem of type (14.2.7) can be proved in certain other situations. Let : 1 T
be a right S(n)-principal bundle for the symmetric group S(n) and J(n) = {1. . . . . n] the
standard left S(n)-set. Then
n
: 1
c(n)
J(n) T is an n-fold covering. The assignment

n
is part of an equivalence between the category of S(n)-principal bundles and n-fold
coverings. Isomorphism classes of n-fold numerable coverings of T are therefore classied
by T. BS(n)|.
13. Use the previous problemin order to classify n-fold coverings of S
1
S
1
via the bijection
to S
1
S
1
. BS(n)|.
14.5 Algebra of Vector Bundles
Let : 1() T and j: 1(j) C be vector bundles over the same eld. The
product j: 1() 1(j) T C is a vector bundle. Let T = C; we pull
back the product bundle along the diagonal J : T T T, b (b. b) and obtain
J
+
( j) = j, the Whitney sum of and j. The bre of j over b is the
direct sum of the bres
b
j
b
. A bundle j is called an inverse of , if j is
isomorphic to a trivial bundle.
A Riemannian metric on a real vector bundle is a continuous map
s : 1( ) R which is on each bre an inner product s(.) on 1()
x
. If
has a Riemannian metric and if : j is a brewise injective bundle morphism
over T, then Coker() is isomorphic to the brewise orthogonal complement
of Im() (this is a subbundle). We have therefore an isomorphism j .
Similarly for complex bundles and hermitian metrics.
(14.5.1) Proposition. A numerable vector bundle has a Riemannian metric.
Proof. Let Ube a covering of T with numeration (t
U
[ U U). A trivial bundle
certainly has a Riemannian metric; so let s
U
be a metric on [U. Then
U
t
U
s
U
is a Riemannian metric on . The sum is short-hand notation for the inner product
U
t
U
(.)s
U
(.) on the bre over ., and we agree that zero times undened =
zero.
(14.5.2) Proposition. A bundle : 1() T has an inverse if and only if it is
numerable of nite type.
Proof. Let
}
:
-1
(U
}
) U
}
R
n
be bundle charts and (t
}
) a numeration of
(U
}
), ( J, J nite). Then
: 1() T
}J
R
n
. . ((.): t
}
((.)) pr
2
}
(.) [ J)
is a brewise injective bundle morphism into a trivial bundle. The orthogonal
complement of the image of is an inverse of .
Let be a k-dimensional bundle with inverse j. From an isomorphism of j
to a trivial bundle we obtain a map : 1() 1( j) T R
n
pr
R
n
which
is injective on each bre. Given a map with this property we get a bundle map
;
k
(R
n
) into the tautological bundle by mapping .
b
to ((
b
). (.))
1
k
(R
n
) G
k
(R
n
)R
n
. (Verify that b (
b
) is continuous.) This map is called
the Gauss map of . The bundle ;
k
is numerable of nite type, as a bundle over
a compact Hausdorff space, hence the induced bundle has the same properties.
Standard constructions of linear algebra can be applied brewise to vector bun-

dles. Examples are:
V
+
dual space of V
+
dual bundle of
V W direct sum j Whitney sum
V W tensor product j tensor product
i
V i -th exterior power
i
i -th exterior power
Hom(V. W) homomorphisms Hom(. j) homomorphism bundle
Canonical isomorphisms between algebraic constructions yield canonical isomor-
phisms for the corresponding vector bundles. Examples are:
( j) ( ) (j )
Hom(. j)
+
j
k
( j)
i}=k
(
i

}
j).
In the last isomorphism
0
is the trivial one-dimensional bundle and
1
.
In order to prove such statements, one has to use that the constructions of linear
algebra are in an appropriate sense continuous. It sufces to consider an example,
say the tensor product. Let : 1() T and j: 1(j) T be real vector bundles.
The total space of j has the underlying set
_
bB
(
b
j
b
) = 1( j).
the disjoint union of the tensor products of the bres. Let :
-1
(U) U R
n
be a bundle chart of and [: j
-1
(U) U R
n
a chart of j. Then a bundle chart
for j over U should be
; :
_
bU
(
b
j
b
) U
}
(R
n
R
n
).
the bre
b
j
b
is mapped by the tensor product of the linear maps over b and
[ over b. At this point it is now important to observe that the transition maps
of such charts are homeomorphisms. Therefore there exists a unique topology on
1( j) such that the sources of the ; are open and the ; are homeomorphisms. In
14.5. Algebra of Vector Bundles 353
this manner we have obtained the data of the bundle j. In dealing with tensor
products one has to distinguish
R
for real bundles and
C
for complex bundles.
If we start with bundles : 1() T and j: 1(j) C we obtain in a similar
manner a bundle

j over TC with bres
b
j
c
. It is called the exterior tensor
product. Let T = C and let J : TT be the diagonal; then J
+
(

j) = j. Let
: T C T and q : T C C be the projections; then

j =
+
q
+
j.
(14.5.3) Example. Let : 1 T be an S
1
-principal bundle and : 1() =
1
S
1 C T the associated complex line bundle. Then is the unit-sphere
bundle of . Let C
n
S
1
be the cyclic subgroup of order m. The m-fold tensor
product
n
= is 1
S
1 C(m), where S
1
acts on Cby (z. :) z
n
:.
The unit-sphere bundle of j
n
is 1,C
n
T. If we use the model S
o
BS
1
for the universal S
1
-principal bundle, we obtain the canonical map BC
n
BS
1
as the sphere bundle of the m-fold tensor product of the universal line bundle.
(14.5.4) Example. Let : 1 T be a right G-principal bundle. Let V. W be
complex G-representations. Let
V
: 1
G
V T be the associated complex
vector bundle. Then there are canonical isomorphisms
V

W

V W
and
W

V W
. For the bundles H(k) over C1
n
the relations H(k)
C
H(l)
H(k l) hold.
14.5.5 Complex vector bundles over S
2
D CP
1
. We have the line bundles H(k)
over C1
1
for k Z. The total space is H(k) = (C
2
0)
C
Cwith equivalence
relation ((:
0
. :
1
). u) ~ ((z:
0
. z:
1
). z
k
u). Set j = H(1); then j
n
H(n). Let
be an arbitrary line bundle over C1
1
. We have the charts
0
: C U
0
=
{:
0
. :
1
| [ :
0
,= 0] and
1
: C U
1
{:
0
. :
1
| [ :
1
,= 0]. We pull back along
}
and obtain a trivial bundle. Let
}
: C C [U
}
be a trivialization. Then
-1
1

0
: C
+
C C
+
C has the form (:. u) (:
-1
. a
z
u) for some map
a: C
+
C
+
. The map a is homotopic to a map : :
-k
. We use a homotopy in
order to construct a bundle over C1
1
0. 1| which is over C1
1
0 given by the
gluing (:. u) (:
-1
. a
z
u) and over C1
1
1 by the gluing (:. u) (:
-1
. :
-k
u).
The latter gives H(k). By the homotopy theorem we see that is isomorphic
to H(k). Let B denote the set of isomorphism classes of complex line bundles
over C1
1
with tensor product as composition law (see Problem 4). We have just
seen that k : Z B, k H(k) is a surjective homomorphism. We know that
B C1
1
. C1
o
|
2
(C1
o
) Z. Therefore k has a trivial kernel, because
otherwise B would be a nite cyclic group. Altogether we have seen that the H(k)
represent the isomorphism classes of complex line bundles.
Nowwe use (14.4.16) and see that a k-dimensional bundle (k _ 1) is isomorphic
to H(n) (k 1)c for a unique n Z. Bundles over C1
1
have a cancellation
property: An isomorphism j implies j; this is again a consequence
of (14.4.16).
For the bundles H(k) over C1
1
the relations H(k) H(l) H(k l) c
hold. In order to prove this relation, we construct an isomorphism of j j
-1
to
the trivial bundle. We write the bundle in the form S
3
S
1 (C(1) C(1)). A
brewise map to C
2
is given by
((:
0
. :
1
). (u
0
. u
1
)) (:
0
u
0
:
1
u
1
. :
1
u
0
:
0
u
1
).
Observe that the matrix with rows (:
0
. :
1
). (:
1
. :
0
) is unitary. For u
0
the image is
the tautological bundle H(1), for u
1
the orthogonal complement H(1). We show
by induction j
k
(k 1)c = kj. This relation is clear for k = 1 and follows for
k = 2 from j j
-1
= 2c. Multiply j
k
(k 1)c = kj by j, add kc and cancel
(k 1)j; the desired relation for k 1 drops out. Suppose k. l N. Then
j
k
j
I
(k 1 l 1)c = (k l)j = j
kI
(k l 1)c.
and cancellation of (k 1 l 1)c gives j
k
j
I
= j
kI
c. We multiply this
relation by j
-k
. j
-I
, or j
-(kI)
in order to verify the remaining cases.
Problems
1. Let and j be vector bundles over T. An orientation of and j induces an orientation
of j, brewise the sum orientation of the vector spaces. If two of the bundles , j, and
j are orientable, then the third is orientable.
2. In a bundle with Riemannian metric the brewise orthogonal complement of a subbundle
is a subbundle.
3. Let : 1 T be an n-dimensional bundle with Riemannian metric. Then there exists a
bundle atlas such that the transition maps have an image in the orthogonal group O(n). The
structure group is therefore reducible to O(n). If the bundle is orientable, then the structure
group is reducible to SO(n).
4. Let and j be complex line bundles over T. Then
C
j is again a line bundle. The
bundle
C
is trivial; the assignments

b
b
C, (.. z) z(.) are an isomorphism
to the trivial bundle. The isomorphism classes of complex line bundles are an abelian group
with composition law the tensor product.
5. Let X be a normal space and Y X a closed subset. A section s : Y 1[Y over Y of
a numerable vector bundle : 1 X has an extension to a section over X.
6. Let : 1 X and q : J X be vector bundles. The bundle morphisms 1 J
correspond to the sections of Hom(1. J) X.
7. Let : 1 X and q : J X be numerable bundles over the normal space X. If
: 1[Y J[Y is an isomorphism over the closed set Y , then there exists an open neigh-
bourhood U of Y and an isomorphism : 1[U J[U which extends over Y .
8. The map C1
u
C1
b
C1
ub
, (.
i
|. ,
|) :
k
| with .
I
=

i=I
.
i
,
in-
duces from H(1) the exterior tensor product H(1)
H(1). In the case a = b = othe map

is associative and denes the structure of an H-space. It induces on T. C1
c
| the group
14.6. Grothendieck Rings of Vector Bundles 355
structure on the set of line bundles given by the tensor product.
9. Determine two-dimensional real bundles over R1
2
[44, p. 434].
14.6 Grothendieck Rings of Vector Bundles
Denote by V(X) the set of isomorphism classes of complex vector bundles over X.
The Whitney sum and the tensor product induce on V(X) two associative and
commutative composition laws (addition , multiplication ), and the distributive
law holds. Addition has a zero element, the 0-dimensional bundle; multiplication
has a unit element, the 1-dimensional trivial bundle.
A commutative monoid M is a set together with an associative and commuta-
tive composition law with zero element. A universal group for M is a homo-
morphism k : M 1(M) into an abelian group 1(M) such that each monoid-
homomorphism : M into an abelian group has a unique factorization
k = with a homomorphism : 1(M) . A monoid-homomorphism
: M N induces a homomorphism 1( ): 1(M) 1(N). Let N M be
a submonoid. We dene an equivalence relation on M by
. ~ , = there exist a. b N such that . a = , b.
Let : M M,N, . .| denote the quotient map onto the set of equivalence
classes. We obtain by .| ,| = . ,| a well-dened composition lawon M,N
which is a monoid structure.
In the product monoid M M we have the diagonal submonoid D(M) =
{(m. m)]. We set 1(M) = (MM),D(M) and k(.) = .. 0|. Then 1(M) is an
abelian group and k a universal homomorphism. Since .. 0| 0. .| = .. .| = 0,
we see 0. .| = k(.). The elements of 1(M) are formal differences .,, .. ,
M, k(.)k(,) = .. ,|, and ., = .
t
,
t
if and only if .,
t
: = .
t
,:
holds as equality in M for some : M.
We apply these concepts to M = V(X) and write 1(X) = 1(V(X)). The
tensor product induces a bi-additive map V(X) V(X) V(X). It induces a
bi-additive map in the 1-groups:
(14.6.1) Proposition. Let , T, C be abelian monoids and m: T C be
a bi-additive map. The there exists a unique bi-additive map 1(m) such that the
diagram
T
n
k
.
k
!
C
k
C
1() 1(T)
1(n)
1(C)
is commutative.
The bi-additive map induced by the tensor product is written as multiplication,
and 1(X) becomes in this way a commutative ring. This ring is often called the
Grothendieck ring of complex vector bundles. In general, the universal groups
1(M) are called Grothendieck groups.
We can apply the same construction to real vector bundles and obtain the
Grothendieck ring 1O(X). Other notations for these objects are 1O(X) = 1
R
(X)
and 1(X) = 1U(X) = 1
C
(X).
Pullback of bundles along : X Y induces a ring homomorphism 1( ) =
+
: 1(Y ) 1(X) and similarly for 1O. Homotopic maps induce the same
homomorphism.
(14.6.2) Example. 1(S
2
) is free abelian as an additive group with basis 1 and j.
The multiplicative structure is determined by j
2
= 2j 1. This is a consequence
of 14.5.5.
The inclusions U(n) U(n 1),
_
0
0 1
_
are used to dene U =
colimU(n), a topological group with the colimit topology. The inclusion of groups
U(n) U induces TU(n) TU. If we compose a classifying map X TU(n)
with this map, we call the result X TU the stable classifying map. Bundles
and ac are called stably equivalent, and they have the same stable classifying
map (up to homotopy).
(14.6.3) Proposition. Let X be a path connected compact Hausdorff space. Then
there exists a natural bijection 1(X) X. ZTU|. Here Z carries the discrete
topology.
Proof. Let | j| 1(X). A bundle j over a compact Hausdorff space has
an inverse bundle j
-
(see (14.5.2)). Hence j
-
| j j
-
| 1(X) is the
same element. Therefore each element in 1(X) can be written in the form | n.
Suppose | n = j| m. Then mc j nc for some . We
add an inverse of and arrive at a relation of the form ac j bc, i.e.,
and j are stably equivalent. The homotopy class of a stable classifying map
k
: X TU is therefore uniquely determined by the element | n. We dene

k : 1(X) X. Z TU| by sending | n to k
: X (dim n) TU.
Conversely, let : X Z TU be given. Since X is path connected, the
image is contained in some k TU. The compactness of X is used to verify that
admits a factorization X TU(n) TU. We obtain a well-dened inverse
map X. Z TU| 1(X), if we assign to : X TU(n) TU the element
+
n
;
n
| n k|.
For more general spaces the Grothendieck ring 1(X) can differ substantially
from the homotopy group X. Z TU|, e.g., for X = C1
o
. The latter is a kind
of completion of the Grothendieck ring.
14.6. Grothendieck Rings of Vector Bundles 357
The (exterior) tensor product of bundles yields a ring homomorphism
1(X)
Z
1(Y ) 1(X Y ).
A fundamental result is the periodicity theorem of Bott. One of its formulations is:
For compact spaces X the tensor product yields isomorphisms
1(X) 1(S
2
) 1(X S
2
). 1O(X) 1O(S
S
) 1O(X S
S
).
Starting from this isomorphism one constructs the cohomology theories which are
called 1-theories. For an introduction see [15], [9], [10], [12], [11], [13], [14],
[102]. For the Bott periodicity see also [6], [106], [19].
Chapter 15
Manifolds
This chapter contains an introduction to some concepts and results of differential
topology. For more details see [30], [44], [107]. We restrict attention to those
parts which are used in the proof of the so-called PontrjaginThom theorem in the
chapter on bordism theory. We do not summarize the results here, since the table
of contents should give enough information.
15.1 Differentiable Manifolds
A topological space X is n-dimensional locally Euclidean if each . X has an
open neighbourhood U which is homeomorphic to an open subset V of R
n
. A
homeomorphism h: U V is a chart or local coordinate system of X about .
with chart domain U. The inverse h
-1
: V U is a local parametrization of X
about .. If h(.) = 0, we say that h and h
-1
are centered at .. A set of charts is an
atlas for X if their domains cover X. If X is n-dimensional locally Euclidean, we
call n the dimension of X and write dimX = n. The dimension is well-dened,
by invariance of dimension.
An n-dimensional manifold or just n-manifold is an n-dimensional locally
Euclidean Hausdorff space with countable basis for its topology. Hence manifolds
are locally compact. A surface is a 2-manifold. A 0-manifold is a discrete space
with at most a countably innite number of points. The notation M
n
is used to
indicate that n = dimM.
Suppose (U
1
. h
1
. V
1
) and (U
2
. h
2
. V
2
) are charts of an n-manifold. Then we
have the associated coordinate change or transition function
h
2
h
-1
1
: h
1
(U
1
U
2
) h
2
(U
1
U
2
).
a homeomorphism between open subsets of Euclidean spaces.
Recall: A map : U V between open subsets of Euclidean spaces (U
R
n
. V R
n
) is a C
k
-map if it is k-times continuously differentiable in the ordinary
sense of analysis (1 _ k _ o). A continuous map is also called a C
0
-map. A
C
k
-map : U V has a differential D(.): R
n
R
n
at . U.
If the coordinate changes h
2
h
-1
1
and h
1
h
-1
2
are C
k
-maps, we call the charts
(U
1
. h
1
. V
1
) and(U
2
. h
2
. V
2
) C
k
-related (1 _ k _ o). Anatlas is a C
k
-atlas if any
two of its charts are C
k
-related. We call C
o
-maps smooth or just differentiable;
similarly, we talk about a smooth or differentiable atlas.
15.1. Differentiable Manifolds 359
(15.1.1) Proposition. Let Abe a smooth atlas for M. The totality of charts which
are smoothly related to all charts of A is a smooth atlas D(A). If A and B are
smooth atlases, then AL B is a smooth atlas if and only if D(A) = D(B). The
atlas D(A) is the uniquely determined maximal smooth atlas which contains A.
A differential structure on the n-manifold M is a maximal smooth atlas D
for M. The pair (M. D) is called a smooth manifold. A maximal atlas serves just
the purpose of this denition. Usually we work with a smaller atlas which then
generates a unique differential structure. Usually we omit the differential structure
from the notation; the charts of D are then called the charts of the differentiable
manifold M.
Let M and N be smooth manifolds. A map : M N is smooth at . M if
is continuous at . and if for charts (U. h. U
t
) about . and (V. k. V
t
) about (.)
the composition k h
-1
is differentiable at h(.). We call k h
-1
the expression
of f in local coordinates. The map is smooth if it is differentiable at each
point. The composition of smooth maps is smooth. Thus we have the category of
smooth manifolds and smooth maps. A diffeomorphism is a smooth map which
has a smooth inverse. Manifolds M and N are diffeomorphic if there exists a
diffeomorphism : M N.
Smoothmanifolds M andN have a product inthe categoryof smoothmanifolds.
The charts of the form (U V. g. U
t
V
t
) for charts (U. . U
t
) of M and
(V. g. V
t
) of N dene a smooth structure on MN. The projections onto the factors
are smooth. The canonical isomorphisms R
n
R
n
= R
nn
are diffeomorphisms.
A subset N of an n-manifold M is a k-dimensional submanifold of M if the
following holds: For each . N there exists a chart h: U U
t
of M about .
such that h(U N) = U
t
(R
k
0). A chart with this property is called adapted
to N. The difference n k is the codimension of N in M. (The subspace R
k
0
of R
n
may be replaced by any k-dimensional linear or afne subspace if this is
convenient.) If we identify R
k
0 = R
k
, then (U N. h. U
t
R
k
) is a chart of N.
If M is smooth, we call N a smooth submanifold of M if there exists about each
point an adapted chart from the differential structure of M. The totality of charts
(U N. h. U
t
R
k
) which arise fromadapted smooth charts of M is then a smooth
atlas for N. In this way, a differentiable submanifold becomes a smooth manifold,
and the inclusion N M is a smooth map. A smooth map : N M is a
smooth embedding if (N) M is a smooth submanifold and : N (N) a
diffeomorphism.
The spheres are manifolds which need an atlas with at least two charts. We have
the atlas with two charts (U
1
.
1
) and (U
S
.
S
) coming from the stereographic
projection (see (2.3.2)). The coordinate transformation is
S

-1
1
(,) = [,[
-2
,.
The differential of the coordinate transformation at . is ([.[
2
2(.. ).)
[.[
-4
. For [.[ = 1 we obtain the reection 2(.. ). in a hyperplane.
We now want to construct charts for the projective space R1
n
. The subset
360 Chapter 15. Manifolds
U
i
= {.
0
. . . . . .
n
| [ .
i
= 0] is open. The assignment
i
: U
i
R
n
. .
0
. . . . . .
n
| .
-1
i
(.
0
. . . . . .
i-1
. .
i1
. . . . . .
n
)
is a homeomorphism. These charts are smoothly related.
Charts for C1
n
can be dened by the same formulas. Note that C1
n
has
dimension 2n as a smooth manifold. (It is n-dimensional when viewed as a so-
called complex manifold.)
(15.1.2) Proposition. Let M be an n-manifold and U = (U
}
[ J) an open
covering of M. Then there exist charts (V
k
. h
k
. T
k
[ k N) of M with the
(1) Each V
k
is contained in some member of U.
(2) T
k
= U
3
(0) = {. R
n
[ [.[ < 3].
(3) The family (V
k
[ k N) is a locally nite covering of M.
In particular, each open cover has a locally nite renement, i.e., manifolds are
paracompact. If M is smooth, there exists a smooth partition of unity (o
k
[ k N)
subordinate to (V
k
). There also exists a smooth partition of unity (
}
[ J)
such that the support of
}
is contained in U
}
and at most a countable number of
the
}
are non-zero.
Proof. The space M is a locally compact Hausdorff space with a countable basis.
Therefore there exists an exhaustion
M
0
M
1
M
2
M =
_
o
i=1
M
i
by open sets M
i
such that

M
i
is compact and contained in M
i1
. Hence 1
i
=
M
i1
M
i
is compact. For each i we can nd a nite number of charts (V
. h
. T
),
T
= U
3
(0), such that V
U
}
for some and such that the h
-1
U
1
(0) cover 1
i
and such that V
M
i2

M
i-1
(M
-1
= 0). Then the V
form a locally nite,

countable covering of M, now denoted (V
k
. h
k
. T
k
[ k N).
The function z: R R, z(t ) = 0 for t _ 0, z(t ) = exp(1,t ) for t > 0, is
a C
o
-function. For c > 0, the function
t
(t ) = z(t )(z(t ) z(c t ))
-1
is C
o
and satises 0 _
t
_ 1,
t
(t ) = 0 = t _ 0,
t
(t ) = 1 = t _ c. Finally,
[: R
n
R, .
t
([.[ r) is a C
o
-map which satises 0 _ [(.) _ 1,
[(.) = 1 =. U
i
(0), [(.) = 0 =[.[ _ r c.
We use these functions [ for r = 1 and c = 1 and dene [
i
by [ h
i
on V
i
and as zero on the complement. Then the o
k
= s
-1
[
k
with s =
o
}=1
[
}
yield a
smooth, locally nite partition of unity subordinate to (V
k
[ k N).
The last statement follows from (13.1.2).
Let C
0
and C
1
be closed disjoint subsets of the smooth manifold M. Then there
exists a smooth function : M 0. 1| such that (C
}
) { ]; apply the previous
proposition to the covering by the U
}
= M C
}
.
15.1. Differentiable Manifolds 361
Let be a closed subset of the smooth manifold M and U an open neigh-
bourhood of in M. Let : U 0. 1| be smooth. Then there exists a smooth
function J : M 0. 1| such that J[ = [. For the proof choose a partition of
unity (
0
.
1
) subordinate to (U. M ). Then set J(.) =
0
(.)(.) for . U
and J(.) = 0 otherwise.
(15.1.3) Proposition. Let M be asubmanifoldof N. Asmoothfunction : M R
has a smooth extension J : N R.
Proof. From the denition of a submanifold we obtain for each M an open
neighbourhood U of in N and a smooth retraction r : U U M. Hence
we can nd an open covering (U
}
[ J) of M in N and smooth extensions
}
: U
}
R of [U
}
M. Let (
}
[ J) be a subordinate smooth partition of
unity and set J(.) =

}J

}
(.)
}
(.), where a summand is dened to be zero
if
}
(.) is not dened.
(15.1.4) Proposition. Let M be a smooth manifold. There exists a smooth proper
function : M R.
Proof. Afunction between Hausdorff spaces is proper if the pre-image of a compact
set is compact. We choose a countable partition of unity (t
k
[ k N) such that the
functions t
k
have compact support. Then we set =

o
k=1
k t
k
: M R. If
.
_
n
}=1
supp(t
}
), then 1 =

}_1
t
}
(.) =

}>n
t
}
(.) and therefore (.) =
}>n
t
}
(.) > n. Hence
-1
n. n| is contained in
_
n
}=1
supp(t
}
) and therefore
compact.
In working with submanifolds we often use, without further notice, the following
facts. Let M be a smooth manifold and M. Then is a submanifold if and only
if each a has an open neighbourhood U such that U is a submanifold of U.
(Being a submanifold is a local property.) Let : N
1
N
2
be a diffeomorphism.
ThenM
1
N
1
is a submanifoldif andonlyif (M
1
) = M
2
N
2
is a submanifold.
(Being a submanifold is invariant under diffeomorphisms.)
Important objects in mathematics are the group objects in the smooth category.
A Lie group consists of a smooth manifold G and a group structure on G such
that the group multiplication and the passage to the inverse are smooth maps. The
fundamental examples are the classical matrix groups. A basic result in this context
says that a closed subgroup of a Lie group is a submanifold and with the induced
structure a Lie group [84], [29].
Problems
1. The gluing procedure (1.3.7) can be adapted to the smooth category. The maps g
i
are
assumed to be diffeomorphisms, and the result will be a locally Euclidean space. Again one
has to take care that the result will become a Hausdorff space.
2. Let 1 be an n-dimensional real vector space 0 < r < n. We dene charts for the
Grassmann manifold G
(1) of r-dimensional subspaces of 1. Let 1 be a subspace of

codimension r in 1. Consider the set of complements in 1
U(1) = {J G
(1) [ J 1 = 1].
The sets are the chart domains. Let 1(1) = { Hom(1. 1) [
2
= . (1) = 1]
be the set of projections with image 1. Then 1(1) U(1), Ker() is a bijection.
The set 1(1) is an afne space for the vector space Hom(1,1. 1). Let : 1 1 and let
q : 1 1,1) be the quotient map. Then
Hom(1,1. 1) 1(1) 1(1). (. ) q
is a transitive free action. We choose a base point
0
1(1) in this afne space and obtain
a bijection
U(1) 1(1) Hom(1,1. 1). Ker()
0
.
The bijections are the charts for a smooth structure.
3. {(.. ,. :) R
3
[ :
2
.
3
3:.
2
3. :,
2
2, = 1] is a smooth submanifold of R
3
diffeomorphic to R
2
. If one considers the set of solutions (.. ,. :) C
2
, then one obtains
a smooth complex submanifold of C
3
which is contractible but not homeomorphic to C
2
(see [47]).
15.2 Tangent Spaces and Differentials
We associate to each point of a smooth m-manifold M an m-dimensional real
vector space T
]
(M), the tangent space of M at the point , and to each smooth
map : M N a linear map T
]
: T
]
(M) T
((])
(N), the differential of at
, such that the functor properties hold (chain rule)
T
]
(g ) = T
((])
g T
]
. T
]
(id) = id .
The elements of T
]
(M) are the tangent vectors of M at .
Since there exist many different constructions of tangent spaces, we dene them
by a universal property.
A tangent space of the m-dimensional smooth manifold M at consists of an
m-dimensional vector space T
]
(M) together with an isomorphism i
k
: T
]
M
R
n
for each chart k = (U. . U
t
) about such that for any two such charts k
and l = (V. [. V
t
) the isomorphism i
-1
I
i
k
is the differential of the coordinate
change [
-1
at (). If (T
t
]
M. i
t
k
) is another tangent space, then |
]
= i
-1
k

i
t
k
: T
t
]
M T
]
M is independent of the choice of k. Thus a tangent space is
determined, up to unique isomorphism, by the universal property. If we x a chart
k, an arbitrary m-dimensional vector space T
]
M, and an isomorphismi
k
: T
]
M
R
n
, then there exists a unique tangent space with underlying vector space T
]
M and
15.2. Tangent Spaces and Differentials 363
isomorphism i
k
; this follows from the chain rule of calculus. Often we talk about
the tangent space T
]
M and understand a suitable isomorphismi
k
: T
]
M R
n
as
structure datum.
Let : M
n
N
n
be a smooth map. Choose charts k = (U. . U
t
) about
M and l = (V. [. V
t
) about () N. There exists a unique linear map
T
]
which makes the diagram
T
]
M
T
)
(
T
((])
N
R
n
T()(
-1
)
i
k
R
n
i
l
commutative; the morphism at the bottom is the differential of [

-1
at ().
Again by the chain rule, T
]
is independent of the choice of k and l. Differentials,
dened in this manner, satisfy the chain rule. This denition is also compatible with
the universal maps |
]
for different choices of tangent spaces T
]
|
]
= |
((])
T
t
]
.
In abstract terms: Make a choice of T
]
(M) for each pair M. Then the T
]
M
and the T
]
constitute a functor from the category of pointed smooth manifolds
and pointed smooth maps to the category of real vector spaces. Different choices
of tangent spaces yield isomorphic functors.
The purpose of tangent spaces is to allow the denition of differentials. The ac-
tual vector spaces are adapted to the situation at hand and can serve other geometric
purposes (e.g., they can consist of geometric tangent vectors).
We call a smooth map an immersion if each differential T
x
is injective and
a submersion if each differential T
x
is surjective. The point . M is a regular
point of if T
x
is surjective. A point , N is a regular value of if each
.
-1
(,) is a regular point, and otherwise a singular value. If
-1
(,) = 0,
then , is also called a regular value.
(15.2.1) Rank Theorem. Let : M N be a smooth map from an m-manifold
into an n-manifold.
(1) If T
o
is bijective, then there exist open neighbourhoods U of a and V of
(a), such that induces a diffeomorphism : U V .
(2) If T
o
is injective, then there exist open neighbourhoods U of a, V of (a),
W of 0 R
n-n
and a diffeomorphismJ : U W V such that J(.. 0) = (.)
for . U.
(3) If T
o
is surjective, then there exist open neighbourhoods U of a, V of (a),
W of 0 R
n-n
andadiffeomorphismJ : U V W suchthat pr
V
J(.) = (.)
for . U with the projection pr
V
: V W V .
(4) Suppose T
x
has rank r for all . M. Then for each a M there
exist open neighbourhoods U of a, V of (a) and diffeomorphisms : U U
t
,
[: V V
t
onto open sets U
t
R
n
, V
t
R
n
such that (U) V and
[
-1
(.
1
. . . . . .
n
) = (.
1
. . . . . .
i
. 0. . . . . 0) for all (.
1
. . . . . .
n
) U
t
.
Proof. The assertions are of a local nature. Therefore we can, via local charts,
reduce to the case that M and N are open subsets of Euclidean spaces. Then these
assertions are known from calculus.
(15.2.2) Proposition. Let , be a regular value of the smooth map : M N.
Then 1 =
-1
(,) is a smooth submanifold of M. For each . 1, we can identify
T
x
1 with the kernel of T
x
.
Proof. Let . 1. The rank theorem (15.2.1) says that is in suitable local coor-
dinates about . and (.) a surjective linear map; hence 1 is locally a submanifold.
The differential of a constant map is zero. Hence T
x
1 is contained in the kernel
of T
x
. For reasons of dimension, the spaces coincide.
(15.2.3) Example. The differentials of the projections onto the factors yield an
isomorphismT
(x,,)
(MN) T
x
(M) T
,
(N) which we use as an identication.
With these identications, T
(x,,)
( g) = T
x
T
,
g for smooth maps and
g. Let h: M N 1 be a smooth map. Then T
(x,,)
h, being a linear map, is
determined by the restrictions to T
x
M and to T
,
N, hence can be computed from
the differentials of the partial maps h
1
: a h(a. ,) and h
2
: b h(.. b) via
T
(x,,)
h(u. ) = T
x
h
1
(u) T
,
h
2
().
(15.2.4) Proposition. Suppose : M N is an immersion which induces a
homeomorphism M (M). Then is a smooth embedding.
Proof. We rst show that (M) is a smooth submanifold of N of the same dimen-
sion as M. It sufces to verify this locally.
Choose U. V. W and J according to (15.2.1). Since U is open and M (M)
a homeomorphism, the set (U) is open in (M). Therefore (U) = (M) 1,
with some open set 1 N. The set 1 = V 1 is an open neighbourhood of b
in N, and 1(M) = (U) holds by construction. It sufces to show that (U)
is a submanifold of 1. We set Q = J
-1
1, and have a diffeomorphismJ : Q 1
which maps U 0 bijectively onto (U). Since U 0 is a submanifold of U W,
we see that (U) is a submanifold.
By assumption, : M (M) has a continuous inverse. This inverse is
smooth, since : M (M) has an injective differential, hence bijective for
dimensional reasons, and is therefore a local diffeomorphism.
(15.2.5) Proposition. Let : M N be a surjective submersion and g: N 1
a set map between smooth manifolds. If g is smooth, then g is smooth.
Proof. Let (.) = ,. By the rank theorem, there exist chart domains U about
. and V about , such that (U) = V and : U V has, in suitable local
coordinates, the form (.
1
. . . . . .
n
) (.
1
. . . . . .
n
). Hence there exists a smooth
map s : V U such that s(:) = : for all : V . Then g(:) = gs(:), and gs
is smooth. (The map s is called a local section of about ,.)
15.2. Tangent Spaces and Differentials 365
It is an important fact of analysis that most values are regular. Aset N in the
n-manifold N is said to have (Lebesgue) measure zero if for each chart (U. h. V )
of N the subset h(U ) has measure zero in R
n
. A subset of R
n
has measure
zero if it can be covered by a countable number of cubes with arbitrarily small total
volume. We use the fact that a diffeomorphism (in fact a C
1
-map) sends sets of
measure zero to sets of measure zero. An open (non-empty) subset of R
n
does not
have measure zero. The next theorem is a basic result for differential topology; in
order to save space we refer for its proof to the literature [136], [30], [177].
(15.2.6) Theorem (Sard). The set of singular values of a smooth map has measure
zero, and the set of regular values is dense.
Problems
1. An injective immersion of a compact manifold is a smooth embedding.
2. Let : M N be a smooth map which induces a homeomorphismM (M). If the
differential of has constant rank, then is a smooth embedding. By the rank theorem,
has to be an immersion, since is injective.
3. Let M be a smooth m-manifold and N M. The following assertions are equivalent:
(1) N is a k-dimensional smooth submanifold of M. (2) For each a N there exist an open
neighbourhood U of a in M and a smooth map : U R
m-k
such that the differential
D(u) has rank m k for all u U and such that N U =
-1
(0). (Submanifolds are
locally solution sets of regular equations.)
4. : R
n1
R, (.
0
. . . . . .
n
)

.
2
i
= [.[
2
has, away from the origin, a non-
zero differential. The sphere S
n
(c) =
-1
(c
2
) = {. R
n1
[ c = [.[] of radius
c > 0 is therefore a smooth submanifold of R
n1
. From Proposition (15.2.2) we obtain
T
\
S
n
(c) = { R
n1
[ . J ].
5. Let M(m. n) be the vector space of real (m. n)-matrices and M(m. n: k) for 0 _ k _
min(m. n) the subset of matrices of rank k. Then M(m. n: k) is a smooth submanifold of
M(m. n) of dimension k(mn k).
6. The subset S
k
(R
n
) = {(.
1
. . . . . .
k
) [ .
i
R
n
: .
1
. . . . . .
k
linearly independent ] of the
k-fold product of R
n
is called the Stiefel manifold of k-frames in R
n
. It can be identied
with M(k. n: k) and carries this structure of a smooth manifold.
7. The group O(n) of orthogonal (n. n)-matrices is a smooth submanifold of the vector space
M
n
(R) of real (n. n)-matrices. Let S
n
(R) be the subspace of symmetric matrices. The map
: M
n
(R) S
n
(R), T T
I
T is smooth, O(n) =
-1
(1), and has surjective
differential at each point O(n). The derivative at s = 0 of s (
I
sX
I
)( sX)
is
I
X X
I
; the differential of at is the linear map M
n
(R) S
n
(R), X
I
X X
I
. It is surjective, since the symmetric matrix S is the image of X =
1
2
S.
From (15.2.2) we obtain
T
.
O(n) = {X M
n
(R) [
I
X X
I
= 0].
and in particular for the unit matrix 1, T
1
O(n) = {X M
n
(R) [
I
= 0], the space
of skew-symmetric matrices. A local parametrization of O(n) about 1 can be obtained from
the exponential map T
1
O(n) O(n), X exp X =

c
0
X
k
,k. Group multiplication
and passage to the inverse are smooth maps.
8. Make a similar analysis of the unitary group U(n).
9. The Stiefel manifolds have an orthogonal version which generalizes the orthogonal group,
the Stiefel manifold of orthonormal k-frames. Let V
k
(R
n
) be the set of orthonormal k-
tuples (
1
. . . . .
k
),
R
n
. If we write
as row vector, then V

k
(R
n
) is a subset of the
vector space M = M(k. n: R) of real (k. n)-matrices. Let S = S
k
(R) again be the vector
space of symmetric (k. k)-matrices. Then : M S,
I
has the pre-image
-1
(1) = V
k
(R
n
). The differential of at is the linear map X X
I
X
I
and it is
surjective. Hence 1 is a regular value. The dimension of V
k
(R
n
) is (n k)k
1
2
k(k 1).
10. The deningmapR
n1
0 R1
n
is a submersion. Its restrictiontoS
n
is a submersion
and an immersion (a 2-fold regular covering).
11. The graph of a smooth function : R
n
R is a smooth submanifold of R
n1
.
12. Let Y be a smooth submanifold of 7 and X Y . Then X is a smooth submanifold
of Y if and only if it is a smooth submanifold of 7. If X is a smooth submanifold, then
there exists about each point . X a chart (U. . V ) of 7 such that (U X) as well as
(U Y ) are intersections of V with linear subspaces. (Charts adapted to X Y 7.
Similarly for inclusions of submanifolds X
1
X
2
X
.)
13. Let
k
(R
n
) be the k-th exterior power of R
n
. The action of O(n) on R
n
induces an
action on
k
(R
n
), a smooth representation. If we assign to a basis .(1). . . . . .(k) of a
k-dimensional subspace the element .(1) . ..(k)
k
(R
n
), we obtain a well-dened,
injective, O(n)-equivariant map : G
k
(R
n
) 1(
k
R
n
) (Plcker coordinates). The
image of is a smooth submanifold of 1(
k
R
n
), i.e., is an embedding of the Grassmann
manifold G
k
(R
n
).
14. The Segre embedding is the smooth embedding
R1
m
R1
n
R1
(m1)(n1)-1
. (.
i
|. ,
|) .
i
,
|.
For m = n = 1 the image is the quadric {s
0
. s
1
. s
2
. s
3
| [ s
0
s
3
s
1
s
2
= 0].
15. Let h: R
n1
R
n1
R
nk1
be a symmetric bilinear formsuch that . = 0. , = 0
implies h(.. ,) = 0. Let g: S
n
S
nk
, . h(.. .),[h(.. .)[. If g(.) = g(,), hence
h(.. .) = t
2
h(,. ,) with some t R, then h(. t,. . t,) = 0 and therefore . t, = 0
or . t, = 0. Hence g induces a smooth embedding R1
n
S
nk
. The bilinear form
h(.
0
. . . . . .
n
. ,
0
. . . . . ,
n
) = (:
0
. . . . . :
2n
) with :
k
=
i=k
.
i
,
yields an embedding
R1
n
S
2n
[89], [95].
16. Remove a point from S
1
S
1
and show (heuristically) that the result has an immersion
into R
2
. (Removing a point is the same as removing a big 2-cell!).
15.3 Smooth Transformation Groups
Let G be a Lie group and M a smooth manifold. We consider smooth action
G M M of G on M. The left translations l
: M M, . g. are
then diffeomorphisms. The map : G M, g g. is a smooth G-map with
image the orbit T = G. through .. We have an induced bijective G-equivariant
set map ; : G,G
x
T. The map has constant rank; this follows from the
equivariance. If 1
: G G and l
: M M denote the left translations by g,

15.3. Smooth Transformation Groups 367
then l
= 1
, and since 1
and l
are diffeomorphisms, we see that T

e
and
T
have the same rank.

(15.3.1) Proposition. Suppose the orbit T = G. is a smooth submanifold of M.
Then:
(1) : G T is a submersion.
(2) G
x
=
-1
(.) is a closed Lie subgroup of G.
(3) There exists a unique smooth structure on G,G
x
such that the quotient map
G G,G
x
is a submersion. The induced map ; : G,G
x
T is a diffeo-
morphism.
Proof. If would have somewhere a rank less than the dimension of T, the rank
would always be less than the dimension, by equivariance. This contradicts the
theorem of Sard. We transport via ; the smooth structure from T to G,G
x
. The
smooth structure is unique, since G G,G
x
is a submersion. The pre-image G
x
of a regular value is a closed submanifold.
The previous proposition gives us G
x
as a closed Lie subgroup. We need not
use the general theorem about closed subgroups being Lie subgroups.
(15.3.2) Example. The actionof SO(n) onS
n-1
bymatrixmultiplicationis smooth.
We obtain a resulting equivariant diffeomorphism S
n-1
SO(n),SO(n 1). In a
similar manner we obtain equivariant diffeomorphisms S
2n-1
U(n),U(n1)
SU(n),SU(n 1).
(15.3.3) Theorem. Let M be a smooth n-manifold. Let C M M be the graph
of an equivalence relation 1 on M, i.e., C = {(.. ,) [ . ~ ,]. Then the following
are equivalent:
(1) The set of equivalence classes N = M,1 carries the structure of a smooth
manifold such that the quotient map : M N is a submersion.
(2) C is a closed submanifold of MM and pr
1
: C M is a submersion.
(15.3.4) Theorem. Let M be a smooth G-manifold with free, proper action of the
Lie group G. Then the orbit space M,G carries a smooth structure and the orbit
map : M M,G is a submersion.
Proof. We verify the hypothesis of the quotient theorem (15.3.3). We have to show
that C is a closed submanifold. The set C is homeomorphic to the image of the
map : G M M M, (g. .) (.. g.), since the action is proper. We
show that is a smooth embedding. It sufces to show that is an immersion
(see (15.2.4)). The differential
T
(,x)
: T
G T
x
M T
x
M T
x
M
will be decomposed according to the two factors
T
(,x)
(u. ) = T
(. .)u T
x
(g. ).
The rst component of T
(. .)u is zero, since the rst component of the partial

map is constant. Thus if T
(,x)
(u. ) = 0, the component of T
x
(g. ) in T
x
M
is zero; but this component is . It remains to show that T
: T
G T
x
M
is injective for : G M, g g.. Since the action is free, the map is
injective; and has constant rank, by equivariance. An injective map of constant
rank has injective differential, by the rank theorem. Thus we have veried the rst
hypothesis of (15.3.3). The second one holds, since pr
1
= pr
2
shows that pr
1
is a submersion.
(15.3.5) Example. The cyclic group G = Z,m S
1
acts on C
n
by
Z,m C
n
C
n
. (. (:
1
. . . . . :
n
)) (
i
1
:
1
. . . . .
i
n
:
n
)
where r
}
Z. This action is a smooth representation. Suppose the integers r
}
are coprime to m. The induced action on the unit sphere is then a free G-manifold
S(r
1
. . . . . r
n
); the orbit manifold 1(r
1
. . . . . r
n
) is called a (generalized) lens space.
(15.3.6) Example. Let H be a closed Lie subgroup of the Lie group G. The
H-action on G by left translation is smooth and proper. The orbit space H\G
carries a smooth structure such that the quotient map G H\G is a submersion.
The G-action on H\G is smooth. One can consider the projective spaces, Stiefel
manifolds and Grassmann manifolds as homogeneous spaces from this view-point.
(15.3.7) Theorem. Let M be a smooth G-manifold. Then:

(1) An orbit C M is a smooth submanifold if and only if it is a locally closed
subset.
(2) If the orbit C is locally closed and . C, then there exists a unique smooth
structure on G,G
x
such that the orbit map G G,G
x
is a submersion. The
map G,G
x
C, gG
x
g. is a diffeomorphism. The G-action on G,G
x
is smooth.
(3) If the action is proper, then (1) and (2) hold for each orbit.
Proof. (1) : G C, g g. has constant rank by equivariance. Hence there
exists an open neighbourhood of e in G such that (U) is a submanifold of M.
Since C is locally closed in the locally compact space M, the set C is locally
compact and therefore : G C is an open map (see (1.8.6)). Hence there exists
an open set W in M such that C W = (U). Therefore C is a submanifold in a
neighbourhood of . and, by equivariance, also globally a submanifold.
15.4. Manifolds with Boundary 369
(2) Since C is locally closed, the submanifold C has a smooth structure. The
map has constant rank and is therefore a submersion. We now transport the
smooth structure from C to G,G
x
.
(3) The orbits of a proper action are closed.
15.4 Manifolds with Boundary
We now extend the notion of a manifold to that of a manifold with boundary. A
typical example is the n-dimensional disk D
n
= {. R
n
[ [.[ _ 1]. Other
examples are half-spaces. Let z: R
n
R be a non-zero linear form. We use the
corresponding half-space H(z) = {. R
n
[ z(.) _ 0]. Its boundary dH(z) is
the kernel of z. Typical half-spaces are R
n
= {(.
1
. . . . . .
n
) R
n
[ .
1
_ 0]. If
R
n
is any subset, we call : R
n
differentiable if for each a there
exists an open neighbourhood U of a in R
n
and a differentiable map J : U R
n
such that J[U = [U . We only apply this denition to open subsets
of half-spaces. In that case, the differential of J at a is independent of the
choice of the extension J and will be denoted D(a).
Let n _ 1 be an integer. An n-dimensional manifold with boundary or
@-manifold is a Hausdorff space M with countable basis such that each point has
an open neighbourhood which is homeomorphic to an open subset in a half-space
of R
n
. A homeomorphism h: U V , U open in M, V open in H(z) is called a
chart about . U with chart domain U. With this notion of chart we can dene
the notions: C
k
-related, atlas, differentiable structure. An n-dimensional smooth
manifold with boundary is therefore an n-dimensional manifold M with boundary
together with a (maximal) smooth C
o
-atlas on M.
Let M be a manifold with boundary. Its boundary dM is the following subset:
The point . is contained in dM if and only if there exists a chart (U. h. V ) about
. such that V H(z) and h(.) dH(z). The complement M dM is called
the interior In(M) of M. The following lemma shows that specifying a boundary
point does not depend on the choice of the chart (invariance of the boundary).
(15.4.1) Lemma. Let : V W be a diffeomorphism between open subsets V
H(z) andW H(j) of half-spaces inR
n
. Then(V dH(z)) = WdH(j).
(15.4.2) Proposition. Let M be an n-dimensional smooth manifold with boundary.
Then either dM = 0 or dM is an (n 1)-dimensional smooth manifold. The set
M dM is a smooth n-dimensional manifold with empty boundary.
The boundary of a manifold can be empty. Sometimes it is convenient to view
the empty set as an n-dimensional manifold. If dM = 0, we call M a manifold
without boundary. This coincides then with the notion introduced in the rst section.
In order to stress the absence of a boundary, we call a compact manifold without
boundary a closed manifold.
Amap : M N between smooth manifolds with boundary is called smooth
if it is continuous and C
o
-differentiable in local coordinates. Tangent spaces and
the differential are dened as for manifolds without boundary.
Let . dM and k = (U. h. V ) be a chart about . with V open in R
n
-
. Then the
pair (k. ), R
n
represents a vector n in the tangent space T
x
M. We say that
n is pointing outwards (pointing inwards, tangential) to dM if
1
> 0 (
1
< 0,
1
= 0, respectively). One veries that this disjunction is independent of the choice
of charts.
(15.4.3) Proposition. The inclusion : dM M is smooth and the differential
T
x
: T
x
(dM) T
x
M is injective. Its image consists of the vectors tangential to
dM. We consider T
x
as an inclusion.
The notion of a submanifold can have different meanings for manifolds with
boundary. We dene therefore submanifolds of type I and type II.
Let M be a smooth n-manifold with boundary. A subset N M is called a
k-dimensional smooth submanifold (of type I) if the following holds: For each
. N there exists a chart (U. h. V ), V R
n
-
open, of M about . such that
h(U N) = V (R
k
0). Such charts of M are adapted to N. The set
V (R
k
0) R
k
-
0 = R
k
-
is open in R
k
-
. Adiffeomorphismonto a submanifold
of type I is an embedding of type I. From this denition we draw the following
conclusions.
(15.4.4) Proposition. Let N M be a smooth submanifold of type I. The restric-
tions h: U N h(U N) of the charts (U. h. V ) adapted to N form a smooth
atlas for N which makes N into a smooth manifold with boundary. The relation
N dM = dN holds, and dN is a submanifold of dM.
Let M be a smooth n-manifold without boundary. A subset N M is a k-di-
mensional smooth submanifold (of type II) if the following holds: For each . N
there exists a chart (U. h. V ) of M about . such that h(U N) = V (R
k
-
0).
Such charts are adapted to N.
The intersection of D
n
with R
k
0 is a submanifold of type I (k < n). The
subset D
n
is a submanifold of type II of R
n
. The next two propositions provide a
general means for the construction of such submanifolds.
(15.4.5) Proposition. Let M be asmoothn-manifoldwithboundary. Let : M R
be smooth with regular value 0. Then
-1
0. o is a smooth submanifold of type II
of M with boundary
-1
(0).
Proof. We have to show that for each .
-1
0. o there exists a chart which
is adapted to this set. If (.) > 0, then . is contained in the open submanifold
-1
|0. o; hence the required charts exist. Let therefore (.) = 0. By the rank
theorem (15.2.1), has in suitable local coordinates the form (.
1
. . . . . .
n
) .
1
.
From this fact one easily obtains the adapted charts.
(15.4.6) Proposition. Let : M N be smooth and , (M) (N dN) be
a regular value of and [dM. Then 1 =
-1
(,) is a smooth submanifold of
type I of M with d1 = ( [dM)
-1
(,) = dM 1.
Proof. Being a submanifold of type I is a local property and invariant under diffeo-
morphisms. Therefore it sufces to consider a local situation. Let therefore U be
open in R
n
-
and : U R
n
a smooth map which has 0 R
n
as regular value for
and [dU (n _ 1. m > n).
We know already that
-1
(0) In(U) is a smooth submanifold of In(U). It
remains to show that there exist adapted charts about points . dU. Since . is
a regular point of [dU, the Jacobi matrix (D
i
}
(.) [ 2 _ i _ m. 1 _ _ n)
has rank n. By interchange of the coordinates .
2
. . . . . .
n
we can assume that the
matrix
(D
i
}
(.) [ m n 1 _ i _ n. 1 _ _ n)
has rank n. (This interchange is a diffeomorphism and therefore harmless.) Under
this assumption, : U R
n
-
, u (u
1
. . . . . u
n-n
.
1
(u). . . . .
n
(u)) has bijec-
tive differential at . and therefore yields, by part (1) of the rank theorem applied to
an extension of to an open set in R
n
, an adapted chart about ..
If only one of the two manifolds M and N has a non-empty boundary, say
M, then we dene M N as the manifold with boundary which has as charts the
products of charts for M and N. In that case d(M N) = dM N. If both M
and N have a boundary, then there appear corners along dM dN; later we shall
explain how to dene a differentiable structure on the product in this case.
Problems
1. The map
D
n
() = {(.. t ) [ t > 0. [.[
2
t
2
_ 1] | 1. 0| U
1
(0). (.. t )
_
I
_
1-]\]
2
1. .
_
is an adapted chart for S
n-1
= dD
n
D
n
.
2. Let T be a d-manifold. A smooth function : dT R has a smooth extension to T. A
smooth function g: R from a submanifold of type I or of type II of T has a smooth
extension to T.
3. Verify the invariance of the boundary for topological manifolds (use local homology
groups).
4. A d-manifold M is connected if and only if M dM is connected.
5. Let M be a d-manifold. There exists a smooth function : M 0. o such that
(dM) = {0] and T
\
,= 0 for each . dM.
6. Let : M R
k
be an injective immersion of a compact d-manifold. Then the image is
a submanifold of type II.
7. Verify that pointing inwards is well-dened, i.e., independent of the choice of charts.
8. Unfortunately is not quite trivial to classify smooth 1-dimensional manifolds by just
starting from the denitions. The reader may try to show that a connected 1-manifold
without boundary is diffeomorphic to R
1
or S
1
; and a d-manifold is diffeomorphic to 0. 1|
or 0. 1 .
15.5 Orientation
Let V be an n-dimensional real vector space. We call ordered bases b
1
. . . . . b
n
and c
1
. . . . . c
n
of V positively related if the determinant of the transition matrix
is positive. This relation is an equivalence relation on the set of bases with two
equivalence classes. An equivalence class is an orientation of V . We specify
orientations by their representatives. The standard orientation of R
n
is given by
the standard basis e
1
. . . . . e
n
, the rows of the unit matrix. Let W be a complex
vector space with complex basis n
1
. . . . . n
n
. Then n
1
. in
1
. . . . . n
n
. in
n
denes
an orientation of the underlying real vector space which is independent of the
choice of the basis. This is the orientation induced by the complex structure.
Let u
1
. . . . . u
n
be a basis of U and n
1
. . . . . n
n
a basis of W. In a direct sum
U W we dene the sum orientation by u
1
. . . . . u
n
. n
1
. . . . . n
n
. If o(V ) is
an orientation of V , we denote the opposite orientation (the occidentation) by
o(V ). A linear isomorphism : U V between oriented vector spaces is
called orientation preserving or positive if for the orientation u
1
. . . . . u
n
of U the
images (u
1
). . . . . (u
n
) yield the given orientation of V .
Let M be a smooth n-manifold with or without boundary. We call two charts
positively related if the Jacobi matrix of the coordinate change has always positive
determinant. Anatlas is calledorienting if anytwoof its charts are positivelyrelated.
We call M orientable, if M has an orienting atlas. An orientation of a manifold
is represented by an orienting atlas; and two such dene the same orientation if
their union contains only positively related charts. If M is oriented by an orienting
atlas, we call a chart positive with respect to the given orientation if it is positively
related to all charts of the orienting atlas. These denitions apply to manifolds of
positive dimension. An orientation of a zero-dimensional manifold M is a function
c : M {1].
Let M be an oriented n-manifold. There is an induced orientation on each of its
tangent spaces T
x
M. It is specied by the requirement that a positive chart (U. h. V )
induces a positive isomorphism T
x
h: T
x
M T
h(x)
V = R
n
with respect to the
standard orientation of R
n
. We can specify an orientation of M by the corresponding
orientations of the tangent spaces.
If M and N are oriented manifolds, the product orientation on M N is
specied by declaring the products (U V. k l. U
t
V
t
) of positive charts
(U. k. U
t
) of M and (V. l. V
t
) of N as positive. The canonical isomorphism
T
(x,,)
(M N) T
x
M T
,
N is then compatible with the sum orientation of
vector spaces. If N is a point, then the canonical identication M N M is
orientation preserving if and only if c(N) = 1. If M is oriented, then we denote
15.5. Orientation 373
the manifold with the opposite orientation by M.
Let M be an oriented manifold with boundary. For . dM we have a direct
decomposition T
x
(M) = N
x
T
x
(dM). Let n
x
N
x
be pointing outwards.
The boundary orientation of T
x
(dM) is dened by that orientation
1
. . . . .
n-1
for which n
x
.
1
. . . . .
n-1
is the given orientation of T
x
(M). These orientations
correspond to the boundary orientation of dM; one veries that the restriction of
positive charts for M yields an orienting atlas for dM.
In R
n
-
, the boundary dR
n
-
= 0 R
n-1
inherits the orientation dened by
e
2
. . . . . e
n
. Thus positive charts have to use R
n
-
.
Let D
2
R
2
carry the standard orientation of R
2
. Consider S
1
as boundary of
D
2
and give it the boundary orientation. An orienting vector in T
x
S
1
is then the
velocity vector of a counter-clockwise rotation. This orientation of S
1
is commonly
known as the positive orientation. In general if M R
2
is a two-dimensional
submanifold with boundary with orientation induced from the standard orientation
of R
2
, then the boundary orientation of the curve dM is the velocity vector of a
movement such that M lies to the left.
Let M be an oriented manifold with boundary and N an oriented manifold
without boundary. Then product and boundary orientation are related as follows
o(d(M N)) = o(dM N). o(d(N M)) = (1)
dim1
o(N dM).
The unit interval 1 = 0. 1| is furnished with the standard orientation of R.
Since the outward pointing vector in 0 yields the negative orientation, we specify
the orientation of d1 by c(0) = 1. c(1) = 1. We have d(1 M) = 0ML1M.
The boundary orientation of 0 M M is opposite to the original one and the
boundary orientation of 1 M M is the original one, if 1 M carries the
product orientation. We express these facts by the suggestive formula d(1 M) =
1 M 0 M. (These conventions suggest that homotopies should be dened
with the cylinder 1 X.)
A diffeomorphism : M N between oriented manifolds respects the ori-
entation if T
x
is for each . M orientation preserving. If M is connected, then
respects or reverses the orientation.
Problems
1. Show that a 1-manifold is orientable.
2. Let : M N be a smooth map and let be the pre-image of a regular value , N.
Suppose M is orientable, then is orientable.
We specify an orientation as follows. Let M and N be oriented. We have an exact
sequence 0 T
u
(1)
T
u
M
(2)
T
,
N 0, with inclusion (1) and differential T
u
at
(2). This orients T
u
as follows: Let
1
. . . . .
k
be a basis of T
u
, n
1
. . . . . n
l
a basis of
T
,
N, and u
1
. . . . . u
l
be pre-images in T
u
M; then
1
. . . . .
k
. u
1
. . . . . u
l
is required to be
the given orientation of T
u
M. These orientations induce an orientation of . This orientation
of is called the pre-image orientation.
3. Let : R
n
R, (.
i
)

.
2
i
and S
n-1
=
-1
(1). Then the pre-image orientation
coincides with the boundary orientation with respect to S
n-1
D
n
.
15.6 Tangent Bundle. Normal Bundle
The notions and concepts of bundle theory can now be adapted to the smooth
category. A smooth bundle : 1 T has a smooth bundle projection and the
bundle charts are assumed to be smooth. A smooth subbundle of a smooth vector
bundle has to be dened by smooth bundle charts. Let :
1

2
be a smooth
bundle morphism of constant rank; then Ker and Im are smooth subbundles.
The proof of (14.2.3) can also be used in this situation. A smooth vector bundle has
a smooth Riemannian metric; for the existence proof one uses a smooth partition
of unity and proceeds as in (14.5.1). Let be a smooth subbundle of the smooth
vector bundle j with Riemannian metric; then the orthogonal complement of in
j is a smooth subbundle.
Let M be a smooth n-manifold. Denote by TM the disjoint union of the tangent
spaces T
]
(M), M. We write a point of T
]
(M) TM in the form (. ) with
T
]
(M), for emphasis. We have the projection
: TM M, (. ) .
Each chart k = (U. h. V ) of M yields a bijection
k
: TU =
_
]U
T
]
(M) U R
n
. (. ) (. i
k
()).
Here i
k
is the morphism which is part of the denition of a tangent space. The map
k
is a map over U and linear on bres. The next theorem is a consequence of the
general gluing procedure.
(15.6.1) Theorem. There exists a unique structure of a smooth manifold on TM such
that the (TU.
k
. U R
n
) are charts of the differential structure. The projection
: TM M is then a smooth map, in fact a submersion. The vector space

structure on the bres of
give
the structure of an n-dimensional smooth

real vector bundle with the
k
as bundles charts.
The vector bundle
: TM M is called the tangent bundle of M. Asmooth

map : M N induces a smooth brewise map T : TM TN, (. )
((). T
]
()).
(15.6.2) Proposition. Let M R
q
be a smooth n-dimensional submanifold. Then
TM = {(.. ) [ . M. T
x
M] R
q
R
q
is a 2n-dimensional smooth submanifold.
15.6. Tangent Bundle. Normal Bundle 375
Proof. Write M locally as h
-1
(0) with a smooth map h: U R
q-n
of constant
rank q n. Then TM is locally the pre-image of zero under
U R
q
R
q-n
R
q-n
. (u. ) (h(u). Dh(u)()).
and this map has constant rank 2(qn); this can be seen by looking at the restrictions
to U 0 and u R
q
.
We can apply (15.6.2) to S
n
R
n1
and obtain the model of the tangent bundle
of S
n
, already used at other occasions.
Let the Lie group G act smoothly on M. We have an induced action
G TM TM. (b. ) (T l
).
This action is again smooth and the bundle projection is equivariant, i.e., TM M
is a smooth G-vector bundle.
(15.6.3) Proposition. Let : 1 M be a smooth G-vector bundle. Suppose the
action on M is free and proper. Then the orbit map 1,G M,G is a smooth
vector bundle. We have an induced bundle map ,G.
The differential T: TM T(M,G) of the orbit map is a bundle morphism
which factors over the orbit map TM (TM),G and induces a bundle morphism
q : (TM),G T(M,G) over M,G. The map is brewise surjective. If G is
discrete, then M and M,G have the same dimension, hence q is an isomorphism.
(15.6.4) Proposition. For a free, proper, smooth action of the discrete group G on
M we have a bundle isomorphism (TM),G T(M,G) induced by the orbit map
M M,G.
(15.6.5) Example. We have a bundle isomorphismTS
n
c (n1)c. If G = Z,2
acts on TS
n
via the differential of the antipodal map and trivially on c, then the said
isomorphism transforms the action into S
n
R
n1
S
n
R
n1
, (.. )
(.. ). We pass to the orbit spaces and obtain an isomorphism T(R1
n
) c
(n 1)j with the tautological line bundle j over R1
n
.
In the general case the map q : (TM),G T(M,G) has a kernel, a bundle
1 M,G with bre dimension dimG. See [44, IX.6] for details.
(15.6.6) Example. The dening map C
n1
0 (C
n1
0),C
+
= C1
n
yields
a surjective bundle map q : T(C
n1
0),C
+
T(C1
n
). The source of q is the
(n 1)-fold Whitney sum (n 1)j where 1(j) is the quotient of (C
n1
0) C
with respect to (:. .) ~ (z:. z.) for z C
+
and (:. .) (C
n1
0) C. The
kernel bundle of q is trivial: We have a canonical section of (n 1)j
C1
n
((C
n1
0) C
n1
),C
+
. :| (:. :), ~.
and the subbundle generated by this section is contained in the kernel of q. Hence
the complex tangent bundle of C1
n
satises T(C1
n
) c (n 1)j. For j see
H(1) in (14.2.6).
Let : 1 M be a smooth vector bundle. Then 1 is a smooth manifold and
we can ask for its tangent bundle.
(15.6.7) Proposition. There exists a canonical exact sequence
0
+
1

T1
+
TM 0
of vector bundles over 1, written in terms of total spaces.
Proof. The differential of is a bundle morphism T: T1 TM, and it induces
a bundle morphism : T1
+
TM which is brewise surjective, since is a
submersion. We consider the total space of
+
1 1 as 11 and the projection
onto the rst summand is the bundle projection. Let (. n) 1
x
1
x
. We dene
(. n) as the derivative of the curve t t n at t = 0. The bundle morphism
has an image contained in the kernel of and is brewise injective. Thus, for
reasons of dimension, the sequence is exact.
(15.6.8) Remark. We restrict the exact sequence givenin(15.6.7) tothe zero section
i : M 1. Since i = id we obtain an exact sequence
0 1

T1[M
TM 0
of vector bundles over M. For n 1
x
, . M the vector (n) T
x
1 is
the derivative at t = 0 of the curve t t n 1
x
. The bundle map has the
right inverse t i : TM T1[M. We therefore obtain a canonical isomorphism
(. T i ): 1 TM T1[M. We have written this in a formal manner. The
geometric meaning is that T
x
1 splits into the tangent vectors in direction of the
bre and the tangent vectors to the submanifold M. Since the tangent space of a
vector space is canonically identied with the vector space, we can consider as
an inclusion.
Let : M N be an immersion. Then T is brewise injective. We pull back
TN along and obtain a brewise injective bundle morphismi : TM
+
TN[M.
The quotient bundle is called the normal bundle of the immersion. In the case of a
submanifold M N the normal bundle v(M. N) of M in N is the quotient bundle
of TN[M by TM. If we give TN a smooth Riemannian metric, then we can take
the orthogonal complement of TM as a model for the normal bundle. The normal
bundle of S
n
R
n1
is the trivial bundle.
We will showthat the total space of the normal bundle of an embedding M N
describes a neighbourhood of M in N. We introduce some related terminology.
Let v : 1(v) M denote the smooth normal bundle. A tubular map is a smooth
map t : 1(v) N with the following properties:
15.6. Tangent Bundle. Normal Bundle 377
(1) It is the inclusion M N when restricted to the zero section.
(2) It embeds an open neighbourhood of the zero section onto an open neigh-
bourhood U of M in N.
(3) The differential of t , restricted to T1(v)[M, is a bundle morphism
t : T1(v)[M TN[M.
We compose with the inclusion (15.6.8)
: 1(v) 1(v) TM T1(v)[M
and the projection
: TN[M 1(v) = (TN[M),TM.
We require that t is the identity. If we use another model of the normal
bundle given by an isomorphism| : 1(v) (TN[M),TM, then we require
t = |.
The purpose of (3) is to exclude bundle automorphisms.
(15.6.9) Remark (Shrinking). Let t : 1(v) N be a tubular map for a subman-
ifold M. Then one can nd by the process of shrinking another tubular map that
embeds 1(v). There exists a smooth function c: M R such that
1
t
(v) = {, 1(v)
x
[ [,[ < c(.)] U.
Let z
)
(t ) = jt (j
2
t
2
)
-1{2
. Then z
)
: 0. o 0. j is a diffeomorphism
with derivative 1 at t = 0. We obtain an embedding
h: 1 1. , z
t(x)
([,[) [,[
-1
,. , 1(v)
x
.
Then g = h is a tubular map that embeds 1(v).
The image U of a tubular map t : 1(v) N which embeds 1(v) is called a
tubular neighbourhood of M in N.
Let M be an m-dimensional smooth submanifold M R
n
of codimension k.
We take N(M) = {(.. ) [ . M. J T
x
M] M R
n
as the normal bundle of
M R
n
.
(15.6.10) Proposition. N(M) is a smooth submanifold of M R
n
, and the pro-
jection N(M) M is a smooth vector bundle.
Proof. Let : R
n
R
k
be a linear map. Its transpose
t
with respect to the
standard inner product is dened by (. n) = (.
t
n). If is surjective,
then
t
is injective, and the relation image (
t
) = (kernel )
J
holds; moreover

t
GL
k
(R).
We dene M locally as the solution set: Suppose U R
n
is open, : U R
k
a submersion, and
-1
(0) = U M = W. We set N(M) (W R
n
) = N(W).
The smooth maps
: W R
n
W R
k
. (.. ) (.. T
x
()).
: W R
k
W R
n
. (.. ) (.. (T
x
)
t
())
satisfy
N(W) = Im. T(W) = Ker .
The composition is a diffeomorphism: it has the form (n. ) (n. g
u
())
with a smooth map W GL
k
(R), n g
u
and therefore (n. ) (n. g
-1
u
())
is a smooth inverse. Hence is a smooth embedding with image N(W) and
-1
[N(W) is a smooth bundle chart.
(15.6.11) Proposition. The map a: N(M) R
n
, (.. ) . is a tubular
map for M R
n
.
Proof. We show that a has a bijective differential at each point (.. 0) N(M).
Let N
x
M = T
x
M
J
. Since M R
n
we consider T
x
M as a subspace of R
n
. Then
T
(x,0)
N(M) is the subspace T
x
M N
x
M T
(x,0)
(M R
n
) = T
x
M R
n
. The
differential T
(x,0)
a is the identity on each of the subspaces T
x
M and N
x
M. There-
fore we can consider this differential as the map (u. ) u , i.e., essentially
as the identity.
It is now a general topological fact (15.6.13) that a embeds an open neighbour-
hood of the zero section. Finally it is not difcult to verify property (3) of a tubular
map.
(15.6.12) Corollary. If we transport the bundle projection N(M) M via the
embedding a we obtain a smooth retraction r : U M of an open neighbourhood
U of M R
n
.
(15.6.13) Theorem. Let : X Y be a local homeomorphism. Let X and
: () = T be a homeomorphism. Suppose that each neighbourhood
of T in Y contains a paracompact neighbourhood. Then there exists an open
neighbourhood U of in X which is mapped homeomorphically under onto an
open neighbourhood V of T in Y (see [30, p. 125]).
For embeddings of compact manifolds and their tubular maps one can apply
another argument as in the following proposition.
(15.6.14) Proposition. Let : X Y be a continuous map of a locally compact
space intoaHausdorff space. Let be injective onthe compact set X. Suppose
that each a has a neighbourhood U
o
in X on which is injective. Then there
exists a compact neighbourhood V of in X on which is an embedding.
15.7. Embeddings 379
Proof. The coincidence set 1 = {(.. ,) X X [ (.) = (,)] is closed in
X X, since Y is a Hausdorff space. Let D(T) be the diagonal of T X. If
is injective on U
o
, then (U
o
U
o
) 1 = D(U
o
). Thus our assumptions imply
that D(X) is open in 1 and hence W = X X \ (1 \ D(X)) open in X X.
By assumption, is contained in W. Since is compact and X locally
compact, there exists a compact neighbourhood V of such that V V W.
Then [V is injective and, being a map from a compact space into a Hausdorff
space, an embedding.
(15.6.15) Proposition. A submanifold M N has a tubular map.
Proof. We x an embedding of N R
n
. By (15.6.12) there exists an open neigh-
bourhood W of V in R
n
and a smooth retraction r : W V . The standard inner
product on R
n
induces a Riemannian metric on TN. We use as normal bundle for
M N the model
1 = {(.. ) M R
n
[ (T
x
M)
J
T
x
N].
Again we use the map : 1 R
n
, (.. ) . and set U =
-1
(W). Then
U is an open neighbourhood of the zero section of 1. The map g = r : U N is
the inclusion when restricted to the zero section. We claim that the differential of g
at points of the zero section is the identity, if we use the identication T
(x,0)
1 =
T
x
M 1
x
= T
x
N. On the summand T
x
M the differential T
(x,0)
g is obviously
the inclusion T
x
M T
x
V . For (.. ) 1
x
the curve t (.. t ) in 1 has
(.. ) as derivative at t = 0. Therefore we have to determine the derivative of
t r(. t ) at t = 0. The differential of r at (.. 0) is the orthogonal projection
R
n
T
x
N, if we use the retraction r in (15.6.12). The chain rule tells us that the
derivative of t r(. t ) at t = 0 is . We now apply again (15.6.13). One
veries property (3) of a tubular map.
15.7 Embeddings
This section is devoted to the embedding theorem of Whitney:
(15.7.1) Theorem. Asmooth n-manifold has an embedding as a closed submanifold
of R
2n1
.
We begin by showing that a compact n-manifold has an embedding into some
Euclidean space. Let : M R
t
be a smooth map from an n-manifold M. Let
(U
}
.
}
. U
3
(0)), {1. . . . . k] be a nite number of charts of M (see (15.1.2)
for the denition of U
3
(0)). Choose a smooth function t : R
n
0. 1| such that
t(.) = 0 for [.[ _ 2 and t(.) = 1 for [.[ _ 1. Dene o
}
: M R by
o
}
(.) = 0 for . U
}
and by o
}
(.) = t
}
(.) for . U
}
; then o
}
is a smooth
function on M. With the help of these functions we dene
: M R
t
(R R
n
) (R R
n
) = R
t
R
1
(.) = ((.): o
1
(.). o
1
(.)
1
(.): . . . : o
k
(.). o
k
(.)
k
(.)).
(k factors R R
n
), where o
}
(.)
}
(.) should be zero if
}
(.) is not dened. The
differential of this map has the rank n on W
}
=
-1
}
(U
1
(0)), as (W
}
) V
}
=
{(:: a
1
. .
1
: . . . : a
k
. .
k
) [ a
}
,= 0], and the composition of [W
}
with V
}
R
n
,
(:: a
1
. .
1
: . . . ) a
-1
}
.
}
is
}
. By construction, is injective on W =
_
k
}=1
W
}
,
since (a) = (b) implies o
}
(a) = o
}
(b) for each , and then
i
(a) =
i
(b)
holds for some i . Moreover, is equal to composed with R
t
R
t
R
1
on
the complement of the
-1
}
U
2
(0). Hence if is an (injective) immersion on the
open set U, then is an (injective) immersion on U L W. In particular, if M is
compact, we can apply this argument to an arbitrary map and M = W. Thus we
have shown:
(15.7.2) Note. A compact smooth manifold has a smooth embedding into some
Euclidean space.
We now try to lower the embedding dimension by applying a suitable parallel
projection.
Let R
q-1
= R
q-1
0 R
q
. For R
q
\ R
q-1
we consider the projection
: R
q
R
q-1
with direction , i.e., for . = .
0
z with .
0
R
q-1
and
z R we set
(.) = .
0
. In the sequel we only use vectors S
q-1
. Let
M R
q
. We remove the diagonal D and consider o : M M \ D S
q-1
,
(.. ,) N(. ,) = (. ,),[. ,[.
(15.7.3) Note.
[M is injective if and only if is not contained in the image

of o.
Proof. The equality
(.) =
(,). . = , and . = .
0
z. , = ,
0
j imply
. , = (z j) = 0, hence = N(. ,). Note o(.. ,) = o(,. .).
Let now M be a smooth n-manifold in R
q
. We use the bundle of unit vectors
STM = {(.. ) [ T
x
M. [[ = 1] M S
q-1
and its projection to the second factor t = pr
2
[STM: STM S
q-1
. The function
(.. ) [[
2
on TM R
q
R
q
has 1 as regular value with pre-image STM,
hence STM is a smooth submanifold of the tangent bundle TM.
(15.7.4) Note.
is an immersion if and only if is not contained in the image

of t.
Proof. The map
is an immersion if for each . M the kernel of T

x
has
trivial intersection with T
x
M. The differential of
is again
. Hence 0 ,= : =
(:) z T
x
M is contained in the kernel of T
x
if and only if : = z and

hence is a unit vector in T
x
M.
15.7. Embeddings 381
(15.7.5) Theorem. Let M be a smooth compact n-manifold. Let : M R
2n1
be a smooth map which is an embedding on a neighbourhood of a compact subset
M. Then there exists for each c > 0 an embedding g: M R
2n1
which
coincides on with and satises [(.) g(.)[ < c for . M.
Proof. Suppose embeds the open neighbourhood U of and let V U be a
compact neighbourhood of . We apply the construction in the beginning of this
section with chart domains U
}
which are contained in M \ V and such that the sets
W
}
cover M U. Then is an embedding on some neighbourhood of M \ U and
: M R
2n1
R
1
= R
q
. . ((.). (.))
is an embedding which coincides on V with (up to composition with the inclusion
R
2n1
R
q
). For 2n < q 1 the image of o is nowhere dense and for 2n 1 <
q 1 the image of t is nowhere dense (theorem of Sard). Therefore in each
neighbourhood of n S
q-1
there exist vectors such that
is an
injective immersion, hence an embedding since M is compact. By construction,
coincides on V with . If necessary, we replace with s (with small s)

such that [(.) (.)[ _ c,2 holds. We can write as composition of
with projections R
q
R
q-1
R
2n1
along the unit vectors (0. . . . . 1).
Sufciently small perturbations of these projections applied to yield a map g
such that [(.) g(.)[ < c, and, by the theorem of Sard, we nd among these
projections those for which g is an embedding.
The preceding considerations show that we need one dimension less for immer-
sions.
(15.7.6) Theorem. Let : M R
2n
be a smooth map from a compact n-mani-
fold. Then there exists for each c > 0 an immersion h: M R
2n
such that
[h(.) (.)[ < c for . M. If : M R
2n1
is a smooth embedding, then
the vectors S
2n
for which the projection
: M R
2n
is an immersion
are dense in S
2n
.
Let : M R be a smooth proper function from an n-manifold without
boundary. Let t R be a regular value and set =
-1
(t ). The manifold is
compact. There exists an open neighbourhood U of in M and a smooth retraction
r : U .
(15.7.7) Proposition. There exists an c > 0 and open neighbourhood V U of
such that (r. ): V |t c. t c is a diffeomorphism.
Proof. The map (r. ): U Rhas bijective differential at points of . Hence
there exists an open neighbourhood W U of such that (r. ) embeds W onto
an open neighbourhood of {t ] in R. Since is proper, each neighbourhood
W of contains a set of the formV =
-1
|t c. t c . The restriction of (r. )
to V has the required properties.
Ina similar manner one shows that a proper submersionis locallytrivial (theorem
of Ehresmann).
We nowshowthat a non-compact n-manifoldM has anembeddingintoR
2n1
as
a closed subset. For this purpose we choose a proper smooth function : M R
.
We then choose a sequence (t
k
[ k N) of regular values of such that t
k
< t
k1
and lim
k
t
k
= o. Let
k
=
-1
(t
k
) and M
k
=
-1
t
k
. t
k1
|. Choose c
k
> 0
small enough such that the intervals J
k
=|t
k
c
k
. t
k
c
k
are disjoint and such
that we have diffeomorphisms
-1
(J
k
)
k
J
k
of the type (15.7.7). We
then use (15.7.7) in order to nd embeddings
k
:
-1
(J
k
) R
2n
J
k
which
have as their second component. We then use the method of (15.7.5) to nd an
embedding M
k
R
2n
t
k
. t
k1
| which extends the embeddings
k
and
k1
in a neighbourhood of M
k
M
k1
. All these embeddings t together and yield an
embedding of M as a closed subset of R
2n1
.
A collar of a smooth d-manifold M is a diffeomorphism k : dM 0. 1 M
onto an open neighbourhood U of dM in M such that k(.. 0) = .. Instead of 0. 1
one can also use R
.
(15.7.8) Proposition. A smooth d-manifold M has a collar.
Proof. There exists an open neighbourhood U of dM in M and a smooth retraction
r : U dM. Choose a smooth function : M R
such that (dM) = {0]

and the derivative of at each point . dM is non-zero. Then (r. ): U
dM R
has bijective differential along dM. Therefore this map embeds an open
neighbourhood V of dM onto an open neighbourhood W of dM 0. Now choose
a smooth function c: dM R
such that {.] 0. c(.) W for each . dM.

Then compose dM 0. 1 dM R
, (.. s) (.. c(.)s) with the inverse of

the diffeomorphism V W.
(15.7.9) Theorem. Acompact smoothn-manifoldT with boundary M has a smooth
embedding of type I into D
2n1
.
Proof. Let : M S
2n
be an embedding. Choose a collar k : M 0. 1 U
onto the open neighbourhood U of M in T, and let l = (l
1
. l
2
) be its inverse. We
use the collar to extend to : T D
2n1
(.) =
_
max(0. 1 2l
2
(.))(l
1
(.)). . U.
0. . U.
Then is a smooth embedding on k(M 0.
1
2
). As in the proof of (15.7.4) we
approximate by a smooth embedding g: T D
2n1
which coincides with
on k(M 0.
1
4
) and which maps T M into the interior of D
2n1
. The image
of g is then a submanifold of type I of D
2n1
.
15.8. Approximation 383
15.8 Approximation
Let M and N be smooth manifolds and M a closed subset. We assume that
N R
]
is a submanifold and we give N the metric induced by this embedding.
(15.8.1) Theorem. Let : M N be continuous and [ smooth. Let : M
|0. o be continuous. Then there exists a smooth map g: M N which coincides
on with and satises [g(.) (.)[ < (.) for . M.
Proof. We start with the special case N = R. The fact that is smooth at .
means, by denition, that there exists an open neighbourhood U
x
of . and a smooth
function
x
: U
x
R which coincides on U
x
with . Having chosen
x
, we
shrink U
x
, such that for , U
x
the inequality [
x
(,) (,)[ < (,) holds.
Fix now . M . We choose an open neighbourhood U
x
of . in M
such that for , U
x
the inequality [(,) (.)[ < (,) holds. We dene
x
: U
x
R in this case by
x
(,) = (.).
Let (t
x
[ . M) be a smooth partition of unity subordinate to (U
x
[ . M).
The function g(,) =
x
t
x
(,)
x
(,) now has the required property.
From the case N = R one immediately obtains a similar result for N = R
]
.
The general case will now be reduced to the special case N = R
]
. For this purpose
we choose an open neighbourhood U of N in R
]
together with a smooth retraction
r : U N. We show in a moment:
(15.8.2) Lemma. There exists a continuous function c: M |0. o with the
properties:
(1) U
x
= U
t(x)
((.)) U for each . M.
(2) For each . M the diameter of r(U
x
) is smaller than (.).
Assuming this lemma, we apply (15.8.1) to N = R
]
and c instead of . This
provides us with a map g
1
: M R
]
which has an image contained in U. Then
g = r g
1
has the required properties.
Proof. We rst consider the situationlocally. Let . M be xed. Choose ;(.) > 0
and a neighbourhood W
x
of . such that (.) _ 2;(.) for , W
x
. Let
V
x
= r
-1
(U
;(x){2
((.)) N).
The distance j(.) = J((.). R
]
V
x
) is greater than zero. We shrink W
x
to a
neighbourhood 7
x
such that [(.) (,)[ <
1
4
j(.) for , 7
x
.
The function [7
x
satises the lemma with the constant function c = c
x
: ,
1
4
j(.). In order to see this, let , 7
x
and [: (,)[ <
1
4
j(.), i.e., : U
,
.
Then, by the triangle inequality, [: (.)[ <
1
2
j(.), and hence, by our choice of
j(.),
: V
x
U. r(:) U
;(x){2
((.)).
If :
1
. :
2
U
,
, then the triangle inequality yields [r(:
1
)r(:
2
)[ < ;(.) _
1
2
(.).
Therefore the diameter of r(U
,
) is smaller than (,).
After this local consideration we choose a partition of unity (t
x
[ . M)
subordinate to (7
x
[ . M). Then we dene c: M |0. o as c(.) =
o
1
4
t
o
(.)j(a). This function has the required properties.
(15.8.3) Proposition. Let : M N be continuous. For each continuous map
: M |0. o there exists a continuous map c: M |0. o with the following
property: Each continuous map g: M N with [g(.) (.)[ < c(.) and
[ = g[ is homotopic to by a homotopy J : M 0. 1| N such that
J(a. t ) = (a) for (a. t ) 0. 1| and [J(.. t ) (.)[ < (.) for (.. t )
M 0. 1|.
Proof. We choose r : U N and c: M |0. o as in (15.8.1) and (15.8.2). For
(.. t ) M 0. 1| we set H(.. t ) = t g(.) (1 t ) (.) U
t(x)
((.)). The
composition J(.. t ) = rH(.. t ) is then a homotopy with the required properties.
(15.8.4) Theorem. (1) Let : M N be continuous and [ smooth. Then

is homotopic relative to to a smooth map. If is proper and N closed in R
]
,
then is properly homotopic relative to to a smooth map.
(2) Let
0
.
1
: M N be smooth maps. Let
t
: M N be a homotopy
which is smooth on T = M0. c LM|1c. 1| L0. 1|. Then there exists a
smooth homotopy g
t
from
0
to
1
which coincides on 0. 1| with . If
t
is a
proper homotopy and N closed in R
]
, then g
t
can be chosen as a proper homotopy.
Proof. (1) We choose andc accordingto(15.8.3) andapply(15.8.1). Then (15.8.3)
yields a suitable homotopy. If is proper, bounded, and if [g(.)(.)[ < (.)
holds, then g is proper.
(2) We nowconsider M|0. 1 instead of M and its intersection with T instead
of and proceed as in (1).
15.9 Transversality
Let : M and g: T N be smooth maps. We form the pullback diagram
C
T
M
with C = {(a. b) [ (a) = g(b)] T. If g: T M, then we identify C
with
-1
(T). If also : M, then
-1
(T) = T. The space C can also
15.9. Transversality 385
be obtained as the pre-image of the diagonal of M M under g. The maps
and g are said to be transverse in (a. b) C if
T
o
(T
o
) T
b
g(T
b
T) = T
,
M.
, = (a) = g(b). They are called transverse if this condition is satised for all
points of C. If g: T M is the inclusion of a submanifold and (a) = b, then
we say that is transverse to T in a if
T
o
(T
o
M) T
b
T = T
b
M
holds. If this holds for each a
-1
(T), then is called transverse to T. We also
use this terminology if C is empty, i.e., we also call and g transverse in this case.
In the case that dimdimT < dimM, the transversality condition cannot hold.
Therefore and g are then transverse if and only if C is empty. A submersion
is transverse to every g.
In the special case T = {b] the map is transverse to T if and only if b is a
regular value of . We reduce the general situation to this case.
We use a little linear algebra: Let a: U V be a linear map and W V a
linear subspace; then a(U) W = V if and only if the composition of a with the
canonical projection : V V,W is surjective.
Let T M be a smooth submanifold. Let b T and suppose : Y R
k
is
a smooth map with regular value 0, dened on an open neighbourhood Y of b in
M such that T Y =
-1
(0). Then:
(15.9.1) Note. : M is transverse to T in a if and only if a is a regular
value of :
-1
(Y ) Y R
k
.
Proof. The space T
b
T is the kernel of T
b
. The composition of T
o
: T
o

T
b
M,T
b
T with the isomorphism T
b
M,T
b
T T
0
R
k
induced by T
b
: T
b
M
T
0
R
k
is T
o
( ). Now we apply the above remark from linear algebra.
(15.9.2) Proposition. Let : M and [d be smooth and transverse to the
submanifold T of M of codimension k. Suppose T and M have empty boundary.
Then C =
-1
(T) is empty or a submanifold of type I of of codimension k. The
equality T
o
C = (T
o
)
-1
(T
((o)
T) holds.
Let, in the situation of the last proposition, v(C. ) and v(T. M) be the normal
bundles. Then T induces a smooth bundle map v(C. ) v(T. M); for, by
denition of transversality, T
o
: T
o
,T
o
C T
((o)
,T
((o)
T is surjective and
then bijective for reasons of dimension.
From (15.9.1) we see that transversality is an open condition: If : M
is transverse in a to T, then it is transverse in all points of a suitable neighbourhood
of a, since a similar statement holds for regular points.
(15.9.3) Proposition. Let : M and g: T M be smooth and let , =
(a) = g(b). Then and g are transverse in (a. b) if and only if g is transverse
in (a. b) to the diagonal of M M.
Proof. Let U = T
o
(T
o
), V = T
b
g(T
b
T), W = T
,
M. The statement amounts
to: U V = W and (U V ) D(W) = W W are equivalent relations (D(W)
diagonal). By a small argument from linear algebra one veries this equivalence.
(15.9.4) Corollary. Suppose and g are transverse. Then C is a smooth subman-

ifold of T. Let c = (a. b) C. We have a diagram
T
c
C
TT
TG
T
b
T
T
T
o
T(
T
,
M.
It is bi-cartesian, i.e., (T. Tg) is surjective and the kernel is T
c
C. Therefore the
diagram induces an isomorphism of the cokernels of TG and Tg (and similarly of
TJ and T ).
(15.9.5) Corollary. Let a commutative diagram of smooth maps be given,
C
T
7
h
M.
Let be transverse to g and C as above. Then h is transverse to G if and only if
h is transverse to g.
Proof. The uses the isomorphisms of cokernels in (15.9.4).
(15.9.6) Corollary. We apply (15.9.5) to the diagram
M

i
:
{s]
W
(
M S
pr
S
and obtain: is transverse to i
x
: . (.. s) if and only if s is a regular value of
pr .
Let J : M S N be smooth and 7 N a smooth submanifold. Suppose
S, 7, and N have no boundary. For s S we set J
x
: M N, . J(.. s). We
consider J as a parametrized family of maps J
x
. Then:
15.9. Transversality 387
(15.9.7) Theorem. Suppose J : MS N and dJ = J[(dMS) are transverse
to 7. Then for almost all s S the maps J
x
and dJ
x
are both transverse to 7.
Proof. By (15.9.2), W = J
-1
(7) is a submanifold of M S with boundary
dW = W d(M S). Let : M S S be the projection. The theorem of
Sard yields the claim if we can show: If s S is a regular value of : W S,
then J
x
is transverse to 7, and if s S is a regular value of d : dW S, then
dJ
x
is transverse to 7. But this follows from (15.9.6).
(15.9.8) Theorem. Let : M N be a smooth map and 7 N a submanifold.
Suppose 7 and N have no boundary. Let C M be closed. Suppose is
transverse to 7 in points of C and d transverse to 7 in points of dM C. Then
there exists a smooth map g: M N which is homotopic to , coincides on C
with and is on M and dM transverse to 7.
Proof. We begin with the case C = 0. We use the following facts: N is diffeo-
morphic to a submanifold of some R
k
; there exists an open neighbourhood U of
N in R
k
and a submersion r : U N with r[N = id. Let S = 1
k
R
k
be the
open unit disk and set
J : M S N. (.. s) r((.) c(.)s).
Here c: M |0. o is a smooth function for which this denition of J makes
sense. We have J(.. 0) = (.). We claim: J and dJ are submersions. For the
proof we consider for xed . the map
S U
t
((.)). s (.) c(.)s:
it is the restriction of an afne automorphism of R
k
and hence a submersion. The
composition with r is then a submersion too. Therefore J and dJ are submersions,
since already the restrictions to the {.] S are submersions.
By (15.9.7), for almost all s S the maps J
x
and dJ
x
are transverse to 7. A
homotopy from J
x
to is M 1 N, (.. t ) J(.. st ).
Let nowC be arbitrary. There exists an open neighbourhood W of C in M such
that is transverse to 7 on W and d transverse to 7 on W dM. We choose a set
V which satises C V

V W
and a smooth function t : M 0. 1| such

that M \ W t
-1
(1). V t
-1
(0). Moreover we set o = t
2
. Then T
x
o = 0,
whenever t(.) = 0. We now modify the map J from the rst part of the proof
G: M S N. (.. s) J(.. o(.)s)
and claim: G is transverse to 7. For the proof we choose (.. s) G
-1
(7).
Suppose, to begin with, that o(.) = 0. Then S N, t G(.. t ) is, as a
composition of a diffeomorphism t o(.)t with the submersion t J(.. t ),
also a submersion and therefore G is regular at (.. s) and hence transverse to 7.
Suppose now that o(.) = 0. We compute T
(x,x)
G at (. n) T
x
M T
x
S =
T
x
X R
n
. Let
m: M S M S. (.. s) (.. o(.)s).
Then
T
(x,x)
m(. n) = (. o(.)n T
x
o()s).
The chain rule, applied to G = J m, yields
T
(x,x)
G(. n) = T
n(x,x)
J T
(x,x)
m(. n) = T
(x,0)
J(. 0) = T
x
().
since o(.) = 0. T
x
o = 0 and J(.. 0) = (.). Since o(.) = 0, by choice of
W and t, is transverse to 7 in ., hence since T
(x,x)
G and T
x
have the same
image also G is transverse to 7 in (.. s). A similar argument is applied to dG.
Then one nishes the proof as in the case C = 0.
15.10 Gluing along Boundaries
We use collars in order to dene a smooth structure if we glue manifolds with
boundaries along pieces of the boundary. Another use of collars is the denition of
a smooth structure on the product of two manifolds with boundary (smoothing of
corners).
15.10.1 Gluing along boundaries. Let M
1
and M
2
be d-manifolds. Let N
i
dM
i
be a union of components of dM
i
and let : N
1
N
2
be a diffeomorphism. We
denote by M = M
1
L
M
2
the space which is obtained from M
1
M
2
by the
identication of . N
1
with (.) N
2
. The image of M
i
in M is again denoted
by M
i
. Then M
i
M is closed and M
i
N
i
M open. We dene a smooth
structure on M. For this purpose we choose collars k
i
: R
-
N
i
M
i
with open
image U
i
M
i
. The map
k : R N
1
M. (t. .)
_
k
1
(t. .). t _ 0.
k
2
(t. (.)). t _ 0.
is an embedding with image U = U
1
L
U
2
. We dene a smooth structure (depend-
ing on k) by the requirement that M
i
N
i
M and k are smooth embeddings.
This is possible since the structures agree on (M
i
N
i
) U.
15.10.2 Products. Let M
1
and M
2
be smooth d-manifolds. We impose a canonical
smooth structure on M
1
M
2
(dM
1
dM
2
) by using products of charts for M
i
as
15.10. Gluing along Boundaries 389
charts. We nowchoose collars k
i
: R
-
dM
i
M
i
and consider the composition z,
R
2
-
dM
1
dM
2
tid
R
-
R
-
dM
1
dM
2
(1)
M
1
M
2
(R
-
dM
1
) (R
-
dM
2
).
k
1
k
2
Here : R
2
-
R
1
-
R
1
-
, (r. ) (r.
1
2

3t
4
), written in polar coordinates
(r. ), and (1) interchanges the second and third factor. There exists a unique
smooth structure on M
1
M
2
such that M
1
M
2
(dM
1
dM
2
) M
1
M
2
and z are diffeomorphisms onto open parts of M
1
M
2
.
15.10.3 Boundary pieces. Let T and C be smooth n-manifolds with boundary.
Let M be a smooth (n 1)-manifold with boundary and suppose that
B
: M dT.
C
: M dC
are smooth embeddings. We identify in T C the points
B
(m) with
C
(m) for
each m M. The result D carries a smooth structure with the following properties:
(1) T
B
(M) D is a smooth submanifold.
(2) C
C
(M) D is a smooth submanifold.
(3) | : M D, m
B
(m) ~
C
(m) is a smooth embedding as a submanifold
of type I.
(4) The boundary of D is diffeomorphic to the gluing of dT
B
(M)
with
dC
C
(M)
via
B
(m) ~
C
(m). m dM.
The assertions (1) and (2) are understood with respect to the canonical embeddings
T D C. We have to dene charts about the points of |(M), since the
conditions (1) and (2) specify what happens about the remaining points. For points
of |(M dM) we use collars of T and C and proceed as in 15.10.1. For |(dM)
we use the following device.
Choose collars k
B
: R
-
dT T and k : R
-
dM M and an embedding
t
B
: R dM dT such that the next diagram commutes,
R dM
r
!
dT
R
-
dM
L
M.
Here t
B
can essentially be considered as a tubular map, the normal bundle of (dM)
in dT is trivial. And k is half of this normal bundle.
Then we form
B
= k
B
(id t
B
): R
-
R dM T. For C we choose
in a similar manner k
C
and t
C
, but we require
C
k
-
= t
C
where k
-
(m. t ) =
k(m. t ). Then we dene
C
from k
C
and t
C
. The smooth structure in a neigh-
bourhood of |(dM) is now dened by the requirement that : R
-
RdM D
is a smooth embedding where
(r. [. m) =
_
B
(r. 2[ ,2. m).
t
2
_ [ _ .
C
(r. 2[ 3,2. m). _ [ _
3t
2
.
with the usual polar coordinates (r. [) in R
-
R.
15.10.4 Connected sum. Let M
1
and M
2
be n-manifolds. We choose smooth
embeddings s
i
: D
n
M
i
into the interiors of the manifolds. In M
1
s
1
(1
n
)
M
2
s
2
(1
n
) we identify s
1
(.) with s
2
(.) for . S
n-1
. The result is a smooth
manifold (15.10.1). We call it the connected sumM
1
#M
2
of M
1
and M
2
. Suppose
M
1
. M
2
are oriented connected manifolds, assume that s
1
preserves the orientation
and s
2
reverses it. Then M
1
#M
2
carries an orientation such that the M
i
s
i
(1
n
)
are oriented submanifolds. One can show by isotopy theory that the oriented dif-
feomorphism type is in this case independent of the choice of the s
i
.
15.10.5 Attaching handles. Let M be an n-manifold with boundary. Furthermore,
let s : S
k-1
D
n-k
dM be an embedding and identify in M D
k
D
n-k
the
points s(.) and .. The result carries a smooth structure (15.10.3) and is said to be
obtained by attaching a k-handle to M.
Attaching a 0-handle is the disjoint sum with D
n
. Attaching an n-handle means
that a hole with boundary S
n-1
is closed by inserting a disk. A fundamental
result asserts that each (smooth) manifold can be obtained by successive attaching
of handles. Aproof uses the so-calledMorse theory(see e.g., [134], [137]). Ahandle
decomposition of a manifold replaces a cellular decomposition, the advantage is
that the handles are themselves n-dimensional manifolds.
15.10.6 Elementary surgery. If M
t
arises from M by attaching a k-handle,
then dM
t
is obtained from dM by a process called elementary surgery. Let
h: S
k-1
D
n-k
X be an embedding into an (n 1)-manifold with image
U. Then X U
has a piece of the boundary which is via h diffeomorphic to

S
k-1
S
n-k-1
. We glue the boundary of D
k
S
n-k-1
with h; in symbols
X
t
= (X U
) L
h
D
k
S
n-k-1
.
The transition from X to X
t
is called elementary surgery of index k at X via
h. The method of surgery is very useful for the construction of manifolds with
prescribed topological properties. See [191], [162], [108] to get an impression of
surgery theory.
15.10. Gluing along Boundaries 391
Problems
1. The subsets of S
mn1
R
m1
R
n1
D
1
= {(.. ,) [ [.[
2
_
1
2
. [,[
2
_
1
2
]. D
2
= {(.. ,) [ [.[
2
_
1
2
. [,[
2
_
1
2
]
are diffeomorphic to D
1
S
m
D
n1
, D
2
D
m1
S
n
. They are smooth submani-
folds with boundary of S
mn1
. Hence S
mn1
can be obtained from S
m
D
n1
and
D
m1
S
n
by identifying the common boundary S
m
S
n
with the identity. A diffeomor-
phism D
1
S
m
D
n1
is (:. n) ([:[
-1
:.
_
2n).
2. Let M be a manifold with non-empty boundary. Identify two copies along the boundary
with the identity. The result is the double D(M) of M. Show that D(M) for a compact M
is the boundary of some compact manifold. (Hint: Rotate M about dM about 180 degrees.)
3. Show M#S
n
M for each n-manifold M.
4. Study the classication of closed connected surfaces. The orientable surfaces are S
2
and connected sums of tori T = S
1
S
1
. The non-orientable ones are connected sums of
projective planes 1 = R1
2
. The relation T #1 = 1#1#1 holds. The connected sum with
T is classically also called attaching of a handle.
Chapter 16
Homology of Manifolds
The singular homology groups of a cell complex vanish above its dimension. It
is an obvious question whether the same holds for a manifold. It is certainly
technically complicated to produce a cell decomposition of a manifold and also
an articial structure. Locally the manifold looks like a Euclidean space, so there
arises no problemlocally. The MayerVietoris sequences can be used to paste local
information, and we use this technique to prove the vanishing theorem.
The homology groups of an n-manifold M in dimension n also have special
properties. They can be used to dene and construct homological orientations of a
manifold. A local orientation about . M is a generator of the local homology
group H
n
(M. M .: Z) Z. In the case of a surface, the two generators corre-
spond to clockwise and counter-clockwise. If you pick a local orientation, then
you can transport it along paths, and this denes a functor from the fundamental
groupoid and hence a twofold covering. If the covering is trivial, then the manifold
is called orientable, and otherwise (as in the case of a Mbius-band) non-orientable.
Our rst aim in this chapter will be to construct the orientation covering and use
it to dene orientations as compatible families of local orientations.
In the case of a closed compact connected manifold we can dene a global
homological orientation to be a generator of H
n
(M: Z); we show that this group
is either zero or Z. In the setting of a triangulation of a manifold, the generator
is the sum of the n-dimensional simplices, oriented in a coherent manner. In a
non-orientable manifold it is impossible to orient the simplices coherently; but in
that case their sum still gives a generator in H
n
(M: Z,2), since Z,2-coefcients
mean that we can ignore orientations.
Once we have global orientations, we can dene the degree of a map between
oriented manifolds. This is analogous to the case of spheres already studied.
16.1 Local Homology Groups
Let h
+
() be a homology theory and M an n-dimensional manifold. Groups of the
type h
k
(M. M .) are called local homology groups. Let : U R
n
be a chart
of M centered at .. We excise M U and obtain an isomorphism
h
k
(M. M .) h
k
(U. U .)
h
k
(R
n
. R
n
0).
For singular homology with coefcients in G we see that H
n
(M. M .: G) G,
and the other local homology groups are zero. Let 1 be a commutative ring. Then
16.1. Local Homology Groups 393
H
n
(M. M .: 1) 1 is a free 1-module of rank 1. A generator, corresponding
to a unit of 1, is called a local R-orientation of M about .. We assemble the
totality of local homology groups into a covering.
Let 1 1 M. The homomorphismr
1
1
: h
k
(M. M1) h
k
(M. M1),
induced by the inclusion, is called restriction. We write r
1
x
in the case that 1 = {.].
(16.1.1) Lemma. Each neighbourhood W of . contains an open neighbourhood
U of . such that the restriction r
U
,
is for each , U an isomorphism.
Proof. Choose a chart : V R
n
with V W centered at .. Set U =
-1
(1
n
),
1
n
= {. R
n
[ [.[ < 1]. We have a commutative diagram
h
k
(M. M U)
i
I
,
h
k
(V. V U)
(1)
(3)
h
k
(M. M ,) h
k
(V. V ,).
(2)
with morphisms induced by inclusion. The maps (1) and (2) are excisions, and (3)
is an isomorphism, because V U V , is for each , U an h-equivalence
(see Problem 1).
We construct a covering o: h
k
(M. M v) M. As a set
h
k
(M. M v) =
x
h
k
(M. M .).
and h
k
(M. M .) is the bre of o over . (with discrete topology). Let U be an
open neighbourhood of . such that r
U
,
is an isomorphism for each , U. We
dene bundle charts
x,U
: U h
k
(M. M .) o
-1
(U). (,. a) r
U
,
(r
U
x
)
-1
(a).
We give h
k
(M. M v) the topology which makes
x,U
a homeomorphism onto an
open subset. We have to show that the transition maps
-1
,,V
x,U
: (U V ) h
k
(M. M .) (U V ) h
k
(M. M ,)
are continuous. Given : U V , choose : W U V such that r
W
u
is an
isomorphism for each n W. Consider now the diagram
h
k
(M. M .) h
k
(M. M U)
i
I
\
i
I
V
i
I
u
h
k
(M. M n)
h
k
(M. M W)
i
V
u
l
l
l
l
l
l
l
l
l
l
l
l
l
h
k
(M. M V )
i
1
V
i
1
,
i
1
u
h
k
(M. M ,).
394 Chapter 16. Homology of Manifolds
It shows
-1
,,V
x,U
= r
V
,
(r
V
W
)
-1
r
U
W
(r
U
x
)
-1
. Hence the second component of
-1
,,V
x,U
is independent of n W, and this shows the continuity of the tran-
sition map.
We take advantage of the fact that the bres are groups. For M we
denote by I() the set of continuous (= locally constant) sections over of
o: h
k
(M. M v) M. If s and t are sections, we can dene (s t )(a) =
s(a) t (a). One uses the bundle charts to see that s t is again continuous. Hence
I() is an abelian group. We denote by I
c
() I() the subgroup of sections
with compact support, i.e., of sections which have values zero away froma compact
set.
(16.1.2) Proposition. Let : h
k
(M. M U). Then , r
U
,
: is a continuous
section of o.
Proof. The bundle chart
x,U
transforms the constant section , (,. r
U
x
:) into
the section , r
U
,
:.
Problems
1. S
n-1
R
n
1
n
and R
n
e S
n-1
, , (, e),[, e[ are h-equivalences
(e 1
n
). The map S
n-1
S
n-1
, , (, e),[, e[ is homotopic to the identity.
These facts imply that R
n
1 R
n
e is an h-equivalence.
2. Let M be an n-dimensional manifold with boundary dM. Showthat . dM if and only if
H
n
(M. M.) = 0. Fromthis homological characterization of boundary points one obtains:
Let : M M be a homeomorphism. Then (dM) = dM and (MdM) = MdM.
3. Let : (C. 0) (D. 0) be a homeomorphism between open neighbourhoods of 0 in R
n
.
Then
: H
n
(C. C 0: 1) H
n
(D. D 0: 1) is multiplication by 1.
16.2 Homological Orientations
Let M be an n-manifold and M. An R-orientation of M along Ais a section
s I(: 1) of o: H
n
(M. Mv: 1) M such that s(a) H
n
(M. Ma: 1)
1is for each a a generator of this group. Thus s combines the local orientations
in a continuous manner. In the case that = M, we talk about an R-orientation
of M, and for 1 = Z we just talk about orientations. If an orientation exists, we
call M (homologically) orientable. If M is orientable along and T , then
M is orientable along T.
(16.2.1) Note. Let Ori(M) H
n
(M. M v: Z) be the subset of all generators of
all bres. Then the restriction Ori(M) M of o is a 2-fold covering of M, called
the orientation covering of M.
16.2. Homological Orientations 395
(1) M is orientable.
(2) M is orientable along compact subsets.
(3) The orientation covering is trivial.
(4) The covering o: H
n
(M. M v: Z) M is trivial.
Proof. (1) =(2). Special case.
(2) = (3). The orientation covering is trivial if and only if the covering over
each component is trivial. Therefore let M be connected. Then a 2-fold covering
M M is trivial if and only if

M is not connected, hence the components of

M
are also coverings.
Suppose Ori(M) M is non-trivial. Since Ori(M) is then connected, there
exists a path in Ori(M) between the two points of a given bre. The image S of such
a path is compact and connected, and the covering is non-trivial over S, since we can
connect two points of a bre in it. By the assumption (2), the orientation covering
is trivial over the compact set S, hence it has a section over S. Contradiction.
(3) = (4). Let s be a section of the orientation covering. Then M Z
H
n
(M. M v: Z), (.. k) ks(.) is a trivialization of o: It is a map over M,
bijective on bres, continuous, and a morphism between coverings.
(4) =(1). If o is trivial, then it has a section with values in the set of generators.
(16.2.3) Note. Ori(M) M is a twofold principal covering with automorphism

group C = {1. t [ t
2
= 1]. Let t act on G as multiplication by 1. Then the
associated covering Ori(M)
C
G is isomorphic to H
n
(M. M v: G).
Proof. The map
Ori(M) G H
n
(M. M v: Z) G H
n
(M. M v: G). (u. g) u g
induces the isomorphism. (The isomorphism is the brewise isomorphism
H
n
(M. M .: Z) G H
n
(M. M .: G) from the universal coefcient for-
mula.)
(16.2.4) Remark. The sections I(: G) of o over correspond bijectively to the
continuous maps z: Ori(M)[ G with the property z t = z. This is a
general fact about sections of associated bundles.
Problems
1. Let M be a smooth n-manifold with an orienting atlas. Then there exists a unique
homological Z-orientation such that the local orientations in H
n
(M. M .: Z) are mapped
via positive charts to a standard generator of H
n
(R
n
. R
n
0: Z). Conversely, if M is Z-
oriented, then M has an orienting atlas which produces the given Z-orientation.
2. Every manifold has a unique Z,2-orientation.
16.3 Homology in the Dimension of the Manifold
Let M be an n-manifold and M a closed subset. We use in this section singular
homology with coefcients in the abelian group G and sometimes suppress G in
the notation.
(16.3.1) Proposition. For each H
n
(M. M : G) the section
J
(): H
n
(M. M v: G). . r
x
()
of o (over ) is continuous and has compact support.
Proof. Let the chain c S
n
(M: G) represent the homology class . There exists
a compact set 1 such that c is a chain in 1. Let . 1. Then the image of c
under
S
n
(1: G) S
n
(M: G) S
n
(M. 1: G) S
n
(M. M .: G)
is zero. Since this image represents r
x
(), the support of J
() is contained in
1.
The continuity is a general fact (16.1.2).
From (16.3.1) we obtain a homomorphism
J
: H
n
(M. M : G) I
c
(: G). (. r
x
()).
(16.3.2) Theorem. Let M be closed.
(1) Then H
i
(M. M ) = 0 for i > n.
(2) The homomorphism J
: H
n
(M. M ) I
c
() is an isomorphism.
Proof. Let D(. 1) and D(. 2) stand for the statement that (1) and (2) holds for the
subset , respectively. We use the fact that J
is a natural transformation between

contravariant functors on the category of closed subsets of M and their inclusions.
The proof is a kind of induction over the complexity of . It will be divided into
several steps.
(1) D(. ). D(T. ). D(T. ) imply D(LT. ). For the proof we use
the relative MayerVietoris sequence for (M T: M . M T) and an
analogous sequence for sections. This leads us to consider the diagram
H
n1
(M. M ( T))
H
n
(M. M ( L T))
J
.L!
I
c
( L T)
H
n
(M. M ) H
n
(M. M T)
J
.
J
!
I
c
() I
c
(T)
H
n
(M. M ( T))
J
.!
I
c
( T).
16.3. Homology in the Dimension of the Manifold 397
The Five Lemma and the hypotheses now show D(LT. 2). The MayerVietoris
sequence alone yields D( L T. 1).
(2) D(. ) holds for compact convex subsets in a chart domain U, i.e.,
() = T is compact convex for a suitable chart : U R
n
. For the proof we
show that for . the restriction r
x
is an isomorphism. By an appropriate choice
of we can assume 0 T 1
n
. Then we have a commutative diagram
H
i
(D
n
. S
n-1
)
(1)
H
i
(R
n
. R
n
T)

H
i
(U. U )
(3)
H
i
(M. M )
i
.
\
H
i
(D
n
. S
n-1
)
(2)
H
i
(R
n
. R
n
0)

H
i
(U. U .)
(4)
H
i
(M. M .).
The maps (3) and (4) are excisions. The maps (1) and (2) are isomorphisms,
because S
n-1
R
n
T is an h-equivalence. The isomorphism r
x
shows, rstly,
that D(. 1) holds; and, secondly, D(. 2), since a section of a covering over a
connected set is determined by a single value.
(3) Suppose U, : U R
n
a chart, = 1
1
L L1
i
, (1
i
) compact
convex. We show D(. ) by induction on r. Let T = 1
1
L L 1
i-1
and
C = 1
i
. Then T and T C are unions of r 1 sets of type (2), hence D(T. )
and D(T C. ) holds by induction. Now use (1).
(4) Let 1 U, : U R
n
a chart, 1 compact. Let 1 W U, W open.
Then there exists a neighbourhood V of 1 inside W of type (3). In this case J
V
is
an isomorphism. The restrictions r
V
1
induce canonical maps from the colimits
colim
V
H
i
(M. M V )
(+)
colim
1
J
1
H
i
(M. M 1)
J
!
colim
V
I
c
(V )
(++)
I
c
(1),
where the colimit is taken over the directed set of neighbourhoods V of 1 of type
(3). What does this isomorphism statement mean in explicit terms? Firstly, an
element .
1
in the image has the form r
V
1
.
V
for a suitable V ; and secondly, if .
V
and .
W
have the same image .
1
, then they become equal under a restriction to a
suitable smaller neighbourhood. From this description it is then easy to verify that
J
1
is indeed an isomorphism. Suppose (+) and (++) are isomorphisms. Then we
obtain D(1. 1), and the isomorphisms J
V
yield D(1. 2).
The isomorphism (+) holds already for the singular chain groups. It uses the
fact that a chain has compact support; if the support is contained in M 1, then
already in M V for a suitable neighbourhood V of 1.
(++) is an isomorphism: See Problem 1.
(5) D(1. ) holds for arbitrary compact subsets 1, for 1 is a union of a nite
number of sets of type (4). Then we can use induction as in case (3).
(6) Let 1 = 1
1
L1
2
L1
3
L with compact 1
i
. Suppose there are pairwise
disjoint open neighbourhoods U
i
of 1
i
. By additivity of homology groups and
section groups, J
1
is the direct sum of the J
1
i
.
(7) Let nally be an arbitrary closed subset. Since M is locally compact
with countable basis, there exists an exhaustion M = L1
i
, 1
1
1
2
by
compact sets 1
i
such that 1
i
1
i1
. Set
i
= (1
i
1
i-1
), 1
0
= 0,
T =
_
i=2n
i
, C =
_
i=2n1
i
. Then D(T. ). D(C. ), and D(T C. )
hold by (6); the hypothesis of (6) follows from the fact that a manifold is a normal
space. Now D(. ) holds by (1), since = T L C.
(16.3.3) Theorem. Suppose is a closed connected subset of M. Then:
(1) H
n
(M. M : G) = 0, if is not compact.
(2) H
n
(M. M : G) G, if M is 1-orientable along and is compact.
Moreover H
n
(M. M : G) H
n
(M. M .: G) is an isomorphism for
each . .
(3) H
n
(M. M : Z)
2
G = {g G [ 2g = 0], if M is not orientable along
and is compact.
Proof. (1) Since is connected, a section in I(: G) is determined by its value at
a single point. If this value is non-zero, then the section is non-zero everywhere.
Therefore there do not exist non-zero sections with compact support over a non-
compact , and (16.3.2) shows H
n
(M. M : G) = 0.
(2) Let be compact. Then H
n
(M. M : G) I(: G). Again a section is
determined by its value at a single point. We have a commutative diagram
H
n
(M. M : G)

i
.
\
I(: G)
b
H
n
(M. M .: G)

I({.]: G).
If M is orientable along , then there exists in I() an element such that its value
at . is a generator. Hence b is an isomorphism and therefore also r
x
.
(3) A section in I(: G) corresponds to a continuous map z: Ori(M)[ G
with zt = z. If M is not orientable along , then Ori(M)[ is connected and
therefore z constant. The relation zt = z shows that the value of z is contained
in
2
G. In this case r
x
: H
n
(M. M : G) H
n
(M. M .: G) G is injective
and has image
2
G.
Theorem (16.3.3) can be considered as a duality result, since it relates an asser-
tion about with an assertion about M .
(16.3.4) Proposition. Let M be an n-manifold and M a closed connected
subset. Then the torsion subgroup of H
n-1
(M. M : Z) is of order 2 if is
compact and M non-orientable along , and is zero otherwise.
16.4. Fundamental Class and Degree 399
Proof. Let q N and suppose M is orientable along the compact set . Then
Z,q H
n
(M. M : Z,q)
H
n
(M. M : Z) Z,q H
n-1
(M. M : Z) + Z,q
Z,q H
n-1
(M. M : Z) + Z,q.
We have used: (16.3.3); universal coefcient theorem; again (16.3.3). This implies
that H
n-1
(M. M : Z) + Z,q = 0. Similarly for non-compact or q odd
0 H
n
(M. M : Z,q) H
n-1
(M. M : Z) + Z,q.
Since Tor(G. Z,q) {g G [ qg = 0], this shows that H
n-1
(M. M: Z) has
no q-torsion in these cases. If is compact and M non-orientable along , then
Z,2 H
n
(M. M : Z,4) H
n-1
(M. M : Z) + Z,4
by (16.3.3) and the universal coefcient formula. Since we know already that the
group in question has no odd torsion, we conclude that there exists a single non-zero
element of nite order and the order is 2.
Problems
1. Let s be a section over the compact set 1. For each . 1 there exists an open neigh-
bourhood U(.) and an extension s
\
of s[U(.) 1. Cover 1 by U(.
1
). . . . . U(.
). Let
W = {, [ s
\
i
(,) = s
\
(.) if , U(.
i
) U(.
)] and dene s(,). , W as the common

value. Show that W is open. Let s. s
/
be sections over V which agree on 1. Show that
they agree in a smaller neighbourhood V
1
V of 1. (These assertions hold for sections of
coverings.)
16.4 Fundamental Class and Degree
The next theorem is a special case of (16.3.3).
(16.4.1) Theorem. Let M be a compact connected n-manifold. Then one of the
following assertions holds:
(1) M is orientable, H
n
(M) Z, andfor each. M the restrictionH
n
(M)
H
n
(M. M .) is an isomorphism.
(2) M is non-orientable and H
n
(M) = 0.
Under the hypothesis of (16.4.1), the orientations of M correspond to the gen-
erators of H
n
(M). A generator will be called fundamental class or homological
orientation of the orientable manifold.
We now use fundamental classes in order to dene the degree; we proceed as
in the special case of S
n
. Let M and N be compact oriented n-manifolds. Let
N be connected and suppose M has components M
1
. . . . . M
i
. Then we have
fundamental classes :(M
}
) for the M
}
and :(M) H
n
(M)

}
H
n
(M
}
) is
the sum of the :(M
}
). For a continuous map : M N we dene its degree
J( ) Z via the relation
+
:(M) = J( ):(N). From this denition we see
immediately the following properties of the degree:
(1) The degree is a homotopy invariant.
(2) J( g) = J( )J(g).
(3) A homotopy equivalence has degree 1.
(4) If M = M
1
M
2
, then J( ) = J( [M
1
) J( [M
2
).
(5) If we pass in M or N to the opposite orientation, then the degree changes the
sign.
We nowcome to the computation of the degree in terms of local data of the map.
Let M and N be connected and set 1 =
-1
(). Let U be an open neighbourhood
of 1 in M. In the commutative diagram
:(M)
_
H
n
(M)
(
H
n
(N)
:(N)
_
H
n
(M. M 1)
(

H
n
(N. N )
:(U. 1) H
n
(U. U 1)
(
I

H
n
(N. N )
=
:(N. )
we have
U
+
:(U. 1) = J( ):(N. ). Thus the degree only depends on the restric-
tion
U
of to U. One can now extend the earlier investigations of self maps of
S
n
to this more general case. The additivity of the degree is proved in exactly the
same manner. Let 1 be nite. Choose U =
_
x1
U
x
where the U
x
are pair-wise
disjoint open neighbourhoods of .. We then have
x1
H
n
(U
x
. U
x
.) H
n
(U. U 1). H
n
(U
x
. U
x
.) Z.
The image :(U
x
. .) of :(M) is a generator, the local orientation determined by
the fundamental class :(M). The local degree J(. .) of about . is dened by
+
:(U
x
. .) = J(. .):(N. ). The additivity yields J( ) =
x1
J(. .).
(16.4.2) Remark. Let be a C
1
-map in a neighbourhood of .; this shall mean
the following. There exist charts : U
x
R
n
centered at . and [: V R
n
centered at such that (U
x
) V and g = [
-1
is a C
1
-map. We can
suppose that the charts preserve the local orientations; this shall mean for that
+
: H
n
(U
x
. U
x
.) H
n
(R
n
. R
n
0) sends :(U
x
. .) to the standard generator.
Such charts are called positive with respect to the given orientations. Suppose now
in addition that the differential of g at . is regular. Then J(. .) is the sign of the
determinant of the Differential Dg(0).
16.4. Fundamental Class and Degree 401
(16.4.3) Proposition. Let M be a connected, oriented, closed n-manifold. Then
there exists for each k Z a map : M
n
S
n
of degree k.
Proof. If : M S
n
has degree a and g: S
n
S
n
degree b, then g has degree
ab. Thus it sufces to realize a degree 1. Let : D
n
M be an embedding.
Then we have a map : M D
n
,S
n-1
which is the inverse of on U = (1
n
)
and sends M U to the base point. This map has degree 1.
Let the manifolds M and N be oriented by the fundamental classes :

H
n
(M) and :
1
H
n
(N). Then the homology product :
:
1
is a fundamental
class for M N, called the product orientation.
Problems
1. Let : M N be a covering of n-manifolds. Then the pullback of Ori(N) N along
is Ori(M) M.
2. Let : M N be a G-principal coveringbetweenconnectednmanifolds withorientable
M. Then N is orientable if and only if G acts by orientation-preserving homeomorphisms.
3. The manifold Ori(M) is always orientable.
4. R1
n
is orientable if and only if n is odd.
5. Let M be a closed oriented connected n-manifold. Suppose that M carries a CW-
decomposition with k-skeleton M
k
. The inclusion induces an injective map H
n
(M)
H
n
(M
n
. M
n-1
). The fundamental class is therefore represented by a cellular chain in
H
n
(M
n
. M
n-1
). If we orient the n-cells in accordance with the local orientations of the
manifold, then the fundamental class chain is the sum of the n-cells. This is the classical in-
terpretation of the fundamental class of a triangulated manifold. Asimilar assertion holds for
unoriented manifolds and coefcients in Z,2 and manifolds with boundary (to be considered
in the next section).
In a sense, a similar assertion should hold for non-compact manifolds; but the cellular
chain would have to be an innite sum. Therefore a fundamental class has to be dened via
an inverse limit.
6. Let M be a closed connected n-manifold. Then H
n
(M: Z,2) Z,2 and the restrictions
r
^
\
: H
n
(M: Z,2) H
n
(M. M .: Z,2) are isomorphisms.
7. Let : M N be a map between closed connected n-manifold. Then one can dene
the degree modulo 2 J
2
( ) Z,2; it is zero (one) if
: H
n
(M: Z,2) H
n
(N: Z,2)
is the zero map (an isomorphism). If this degree is non-zero, then is surjective. If the
manifolds are oriented, then J( ) mod 2 = J
2
( ).
8. Let G be a compact connected Lie group and let T be a maximal torus of G. The map
q : G,T T G, (g. t ) gtg
-1
has degree [W[. Here W = NT,T is the Weyl group.
Since q has non-zero degree, this map is surjective (see [29, IV.1]).
9. Let G be a compact connected Lie group and T a maximal torus of G. The degree of
: G G, g g
k
has degree k
, r = dimT . Let c T be an element such that the

powers of c are dense in T , then [
-1
(c)[ = k
,
-1
(c) T , and c is a regular value of
. [90]
10. Let : M N be a proper map between oriented connected n-manifolds. Dene the
degree of .
Let M be an n-dimensional manifold with boundary. We call : H
n
(M. dM) a
fundamental class if for each . M dM the restriction of : is a generator in
H
n
(M. M .).
(16.5.1) Theorem. Let M be a compact connected n-manifold with non-empty
boundary. Then one of the following assertions hold:
(1) H
n
(M. dM) Z, and a generator of this group is a fundamental class.
The image of a fundamental class under d: H
n
(M. dM) H
n-1
(dM) is a
fundamental class. The interior M dM is orientable.
(2) H
n
(M. dM) = 0, and M dM is not orientable.
Proof. Let k : 0. o dM U be a collar of M, i.e., a homeomorphism onto an
open neighbourhood U of dM such that k(0. .) = . for . dM. For simplicity
of notation we identify U with 0. o dM via k; similarly for subsets of U. In
this sense dM = 0 dM. We have isomorphisms
H
n
(M. dM) H
n
(M. 0. 1 dM) H
n
(M dM. |0. 1dM) I().
The rst one by h-equivalence; the second one by excision; the third one uses the
closed set
= M (0. 1 dM) M dM
and (16.3.2). The set is connected, hence I() Zor I() 0. If I() Z,
then MdM is orientable along . Instead of we can argue with the complement
of 0. c dM. Since each compact subset of M dM is contained in some such
complement, we see that M dM is orientable along compact subsets, hence
orientable (see (16.2.2)). The isomorphism H
n
(M dM. |0. 1 dM) I()
says that there exists an element : H
n
(M dM. |0. 1 dM) which restricts to
a generator of H
n
(M dM. M dM .) for each . . For the corresponding
element : H
n
(M. dM) a similar assertion holds for each . M dM, i.e., : is
a fundamental class (move around . within the collar).
It remains to show that d: is a fundamental class. The lower part of the diagram
(for . |0. 1 dM)
H
n-1
(dM)

H
n-1
(dM L . ) H
n-1
(d1 dM. 1 dM)
H
n
(M. dM)

R
R
R
R
R
R
R
R
R
R
R
R
R
J
H
n
(M. dM L )
H
n
(1 dM. d1 dM)
H
n
(M. M .) H
n
(1 dM. 1 dM .)
16.6. Winding and Linking Numbers 403

shows that : yields a fundamental class in H
n
(1 dM. d1 dM). The upper part
shows that this fundamental class corresponds to a fundamental class in H
n-1
(dM),
since fundamental classes are characterized by the fact that they are generators (for
each component of dM).
(16.5.2) Example. Suppose the n-manifold M is the boundary of the compact ori-
entable (n1)-manifoldT. We have the fundamental classes :(T) H
n1
(T. dT)
and :(M) = d:
B
H
n
(M). Let : M N be a map which has an extension
J : T N, then the degree of (if dened) is zero, J( ) = 0, for we have
+
:(M) =
+
d:(T) = J
+
i
+
d:(T) = 0, since i
+
d = 0 as consecutive morphisms
inthe exact homologysequence of the pair (T. M). We call maps
: M
N ori-
entable bordant if there exists a compact oriented manifold T with oriented bound-
ary dT = M
1
M
2
and an extension J : T N of (
1
.
2
): M
1
M
2
N.
The minus sign in dT = M
1
M
2
means d:(T) = :(M
1
) :(M
2
). Under these
assumptions we have J(
1
) = J(
2
). This fact is called the bordism invariance
of the degree; it generalizes the homotopy invariance.
Problems
1. Let M be a compact connected n-manifold with boundary. Then H
n
(M. dM: Z,2)
is isomorphic to Z,2; the non-zero element is a Z,2-fundamental class :(M: Z,2). The
restriction to H
n
(M. M.: Z,2) is for each . MdM an isomorphismand d:(M: Z,2)
is a Z,2-fundamental class for dM.
2. Show that the degree J
2
( ) is a bordism invariant.
16.6 Winding and Linking Numbers
Let M be a closed connected oriented n-manifold. Let : M R
n1
and a
Im( ). The winding number W(. a) of with respect to a is the degree of the
map
(,o
=
o
: M S
n
. . N((.) a)
where N : R
n1
0 S
n
, . [.[
-1
.. If
t
is a homotopy with a Im(
t
)
for each t , then W(
0
. a) = W(
t
. a).
(16.6.1) Theorem. Let M be the oriented boundary of the compact smooth oriented
manifold T. Let J : T R
n1
be smooth with regular value 0 and assume
0 (M). Then
W(. 0) =
x1
c(J. .). 1 = J
-1
(0). = J[dT
where c(J. .) {1] is the orientation behaviour of the differential T
x
J : T
B

T
0
(R
n1
).
Proof. Let D(.) TdT, . 1 be small disjoint disks about .. ThenG = NJ
is dened on C = T
_
x1
D(.), and by the bordism invariance of the degree
J(G[dT) =
x1
J(G[dD(.)). By (16.4.2), J(G[dD(.)) = c(J. .).
Let M and N be oriented closed submanifolds of R
k1
of dimensions m and n
with k = mn. Let M N carry the product orientation. The degree of the map
,1
= : M N S
k
. (.. ,) N(. ,)
is the linking number 1(M. N) of the pair (M. N). More generally, if the maps
j: M
n
R
k1
and v : N
n
R
k1
have disjoint images, then the degree of
(.. ,) N(j(.) v(,)) is the linking number of (j. v).
Problems
1. Let the n-manifold M be the oriented boundary of the smooth connected compact mani-
fold T. Suppose : M S
n
has degree zero. Then can be extended to T.
2. Show 1(M. N) = (1)
mn1
1(N. M).
3. Let . g: R R
3
be smooth embeddings with disjoint closed images. Dene a linking
number for the pair (. g) and justify the denition.
Chapter 17
Cohomology
The axioms for a cohomology theory are analogous to the axioms of a homology
theory. Now we consider contravariant functors. The reader should compare the
two denitions, also with respect to notation. One advantage of cohomology is an
additional internal product structure (called cup product) which will be explained
in subsequent sections. The product structure suggests to view the family (h
n
(X) [
n Z) as a single object; the product then furnishes it with the structure of a ring
(graded algebra). Apart fromthe additional information in the product structure, the
ring structure is also notationally convenient (for instance, a polynomial ring has a
better description than its additive group without using the multiplicative structure).
Singular cohomology is obtained from the singular chain complex by an ap-
plication of the Hom-functor. We present an explicit denition of the cup product
in singular cohomology (AlexanderWhitney). In more abstract terms the product
can also be obtained from the EilenbergZilber chain equivalences as in the case of
the homology product.
We use the product structure to prove a powerful theorem(LerayHirsch) which
says roughly that the cohomology of the total space of a bration is a free module
over the cohomology ring of the base, provided the bre is a free module and a basis
of the bre-cohomology can be lifted to the total space. In the case of a topological
product, the result is a special case of the Knneth theoremif the cohomology of the
bre is free, since the Ext-groups vanish in that case. One interesting application
is to vector bundles; the resulting so-called Thom isomorphism can be considered
as a twisted suspension isomorphism in that the suspension is replaced by a sphere
bundle. As a specic example we determine the cohomology rings of the projective
spaces.
17.1 Axiomatic Cohomology
17.1.1 The axioms. A cohomology theory for pairs of spaces with values in the
category of 1-modules consists of a family (h
n
[ n Z) of contravariant functors
h
n
: TOP(2) 1- MOD and a family (
n
[ n Z) of natural transformations
n
: h
n-1
k h
n
. These data are required to satisfy the following axioms.
(1) Homotopy invariance. Homotopic maps
0
and
1
between pairs of spaces
induce the same homomorphism, h
n
(
0
) = h
n
(
1
).
406 Chapter 17. Cohomology
(2) Exact sequence. For each pair (X. ) the sequence
h
n-1
(. 0)

h
n
(X. ) h
n
(X. 0) h
n
(. 0)

is exact. The undecorated arrows are induced by the inclusions.
(3) Excision. Let (X. ) be a pair and U such that

U
. Then
the inclusion (X U. U) (X. ) induces an excision isomorphism
h
n
(X. ) h
n
(X U. U).
We call h
n
(X. ) the n-th cohomology group of (X. ). The
n
are called the
coboundary operators. We write h
n
(X. 0) = h
n
(X) and h
n
( ) =
+
. Occasion-
ally we refer to the homomorphisms i
+
: h
n
(X) h
n
() induced by an inclusion
i : X as restriction. The groups h
n
(1) h
n
for a point 1 are said to be
the coefcient groups of the theory (compatible family of isomorphisms to a given
module h
n
). In the case that h
n
(1) = 0 for n ,= 0, we talk about an ordinary or
classical cohomology theory and say that the theory satises the dimension axiom.
The notation h
n
(X. ) = H
n
(X. : G) stands for an ordinary cohomology theory
with a given isomorphism h
0
(1) G.
The cohomology theory is additive if
h
n
_
}
X
}
.
}

}
_
}
h
n
(X
}
.
}
). . (h
n
(i
}
)(.))
is always an isomorphism (i
}
the inclusion of the -th summand). For nite J the
additivity isomorphism follows from the other axioms.
Several formal consequences of the homology axioms have analogues in coho-
mology and the proofs are similar. We mention some of them.
We begin with the exact sequence of a triple (X. . T). The coboundary op-
erator is in this case dened by : h
n-1
(. T) h
n-1
() h
n
(X. ). The
rst map is induced by the inclusion and the second map is the given coboundary
operator. For each triple (X. . T) the sequence
h
n-1
(. T)

h
n
(X. ) h
n
(X. T) h
n
(. T)

is exact. The undecorated arrows are restrictions.
17.1.2 Suspension. The suspension isomorphismo is dened by the commutative
diagram
h
n
(Y. T)
c

h
n
(0 Y. 0 T)
h
n1
((1. d1) (Y. T)) h
n
(d1 Y L 1 T. 1 Y L 1 T).
For homology we used a denition with the roles of 0. 1 interchanged.

17.1. Axiomatic Cohomology 407
17.1.3 Reduced cohomology. The reduced cohomology groups of a non-empty
space X are dened as

h
n
(X) = coker(
+
: h
n
(1) h
n
(X)) where : X 1
denotes the unique map to a point. The functors

h
n
() are homotopy invariant. For
a pointed space (X. +) we have the canonical split exact sequence
0 h
n
(X. +)
}
h
n
(X)
i
h
n
(+) 0.
The restriction induces an isomorphismh
n
(X. +)

h
n
(X) and we have isomor-
phisms
(.
+
): h
n
(X. +) h
n
h
n
(X). (q. i ): h
n
(X)

h
n
(X) h
n
with the quotient map q. The coboundary operator
n
: h
n-1
() h(X. ) factors
over the quotient map q : h
n-1
()

h
n-1
(). Passing to quotients yields the
exact sequence

h
n-1
()

h(X. )

h
n
(X)

h
n
()

for the reduced groups.
17.1.4 MayerVietoris sequence. A triad (X: . T) is excisive for the cohomol-
ogy theory if the inclusion induces an isomorphismh
+
(LT. ) h
+
(T. T).
This property can be characterized in different ways as in the case of a homology
theory, see (10.7.1) and (10.7.5). In particular the property is symmetric in . T.
We have exact MayerVietoris sequences for excisive triads. As in the case of
a homology theory one can derive some MV-sequences by diagram chasing. For
the general case of two excisive triads we use a method which we developed in the
case of homology theory; the MV-sequence was obtained as the exact sequence of
a triad of auxiliary spaces by some rewriting (see (10.7.6)). This procedure also
works for cohomology.
Let (:
0
.
1
) (X: X
0
. X
1
) be excisive triads. Set X
01
= X
0
X
1
and
01
=
0

1
. Then there exists an exact MayerVietoris sequence of the
following form
h
n
(X
01
.
01
)
(1)
h
n
(X
0
.
0
) h
n
(X
1
.
1
)
(2)
h
n
(X. )
Z
h
n-1
(X
01
.
01
) .
The map (1) is (.
0
. .
1
) i
+
0
.
0
i
+
1
.
1
with the inclusions i
; the components of
(2) are the restrictions. The connecting morphism in the case =
0
=
1
is the
composition
^: h
n-1
(X
01
. )

h
n
(X
0
. X
01
) h
n
(X. X
1
) h
n
(X. )
and the connecting morphism in the case X = X
0
= X
1
is
^: h
n
(X.
01
) h
n
(
0
.
01
) h
n
(.
1
)

h
n1
(X. ).
17.1.5 Limits. We have seen that an additive homology theory is compatible with
colimits. The situation for cohomology and limits is more complicated.
Let G
-
: G
1
G
2
]
1
G
3
]
2
. . .
]
3
be a sequence of groups and ho-

momorphisms. The group
iN
G
i
acts on the set
iN
G
i
(g
1
. g
2
. . . . ) (h
1
. h
2
. . . . ) = (g
1
h
1
1
(g
2
)
-1
. g
2
h
2
2
(g
3
)
-1
. . . . ).
The orbit set is denoted lim
1
(G
-
) = lim
1
(G
i
.
i
) = lim
1
(G
i
). A direct conse-
quence of this denition is:
Suppose the groups G
i
are abelian. Then we have an exact sequence
0 lim(G
i
.
i
)
iN
G
i
d
iN
G
i
lim
1
(G
i
.
i
) 0.
Here J(g
1
. g
2
. . . . ) = (g
1

1
(g
2
). g
2

2
(g
3
). . . . ) and lim is the limit of the
sequence. As a consequence of the ker-coker-sequence we obtain:
A short exact sequence
0 (G
t
i
.
t
i
) (G
i
.
i
) (G
tt
i
.
tt
i
) 0
of inverse systems induces an exact sequence
0 lim
0
(G
t
i
) lim
0
(G
i
) lim
0
(G
tt
i
)
lim
1
(G
t
i
) lim
1
(G
i
) lim
1
(G
tt
i
) 0.
Let nowan additive cohomology theory by given. We apply the MayerVietoris
sequence to the telescope T of a sequence X
1
X
2
of spaces with colimit X.
The result is [133]:
(17.1.6) Proposition. There exists an exact sequence
0 lim
1
(h
n-1
(X
i
)) h
n
(T ) lim(h
n
(X
i
)) 0.
Proof. We use the MV-sequence of the triad (T : . T) as in (10.8.2). It has the
form
h
n
(T ) h
n
() h
n
(T)
n
h
n
( T)
and yields the short exact sequence
0 Coker(
n-1
) h
n
(T ) Ker(
n
) 0.
17.2. Multiplicative Cohomology Theories 409
Thus we have to determine the kernel and the cokernel. By additivity of the coho-
mology theory we obtain
h
n
()
i=0(2)
h
n
(X
i
). h
n
(T)
i=1(2)
h
n
(X
i
). h
n
( T)
iN
h
n
(X
i
).
These isomorphisms transform
n
into
(.
1
. .
2
. .
3
. . . . ) (.
1
+
1
(.
2
). .
2

+
2
(.
3
). . . . ).
The isomorphism (a
i
) ((1)
i
a
i
) transforms this map nally into the map J in
the denition of lim and lim
1
.
(17.1.7) Proposition. Suppose the homomorphisms
i
between the abelian groups
G
i
are surjective. Then lim
1
(G
i
.
i
) = 0.
Proof. Let g = (g
1
. g
2
. . . . )

iN
G
i
. We have to show that g is contained
in the image of J, i.e., we have to solve the equations g
i
= .
i

i
(.
i1
) with
suitable .
i
G
i
. This is done inductively.
Problems
1. The system (G
i
.
i
) satises the Mittag-Lefer condition (= ML) if for each i there
exists such that for k _ the equality Im(G
ik
G
i
) = Im(G
i
G
i
) holds.
If (G
i
.
i
) satises ML, then lim
1
(G
i
.
i
) = 0. Thus if the groups G
i
are nite, then
lim
1
(G
i
.
i
) = 0. If the G
i
are countable and if lim
1
(G
i
.
i
) = 0, then the systemsatises
ML ([44, p. 154]).
2. Imitate the earlier investigation of cellular homology and show that H
(X) can be de-

termined from a cellular cochain complex which arises from a cellular decomposition of X.
3. Let j: k
() l
() we a natural transformation between additive cohomology theo-

ries which induces isomorphisms of the coefcient groups. Show that j is an isomorphism
for each CW-complex.
17.2 Multiplicative Cohomology Theories
Let h
+
be a cohomology theory with values in 1- MOD. A multiplicative structure
on this theory consists of a family of 1-linear maps (m. n Z)
h
n
(X. )
T
h
n
(X. T) h
nn
(X. L T). . , . ,.
dened for suitable triads (X: . T), and in any case for excisive (. T) in X. We
call . , the cup product of .. ,. The products are always dened if or T is
empty or if = T. In this section, tensor products will be taken over 1. The cup
product maps are required to satisfy the following axioms.
(1) Naturality. For maps of triads : (X: . T) (X
t
:
t
. T
t
) the commutativity
+
(. ,) =
+
.
+
, holds.
(2) Stability. Let (. T) be excisive in X. We use the restriction morphism
|
: h
}
(X. T) h
}
(. T) and the coboundary operator
: h
i
(. T)
h
i
( L T. T)

h
i1
(X: L T). The diagram
h
i
() h
}
(X. T)
1
1t
.
h
i
() h
}
(. T)
.

h
i}
(. T)
h
i1
(X. ) h
}
(X. T)
.

h
i}1
(X. L T)
is commutative.
(3) Stability. Let (. T) be excisive in X. We use the restriction morphism
|
B
: h
i
(X. ) h
i
(T. T) and the coboundary operator
B
: h
i
(T. T)
h
i
( L T. )

h
i1
(X. L T). The diagram
h
i
(X. ) h
}
(T)
1
t
!
1
h
i
(T. T) h
}
(T)
.

h
i}
(T. T)
h
i
(X. ) h
}1
(X. T)
.

h
i}1
(X. L T)
is commutative up to the sign (1)
i
.
(4) Unit element. There is given a unit 1 h
0
(1), 1 a xed point, as additional
structure datum. It induces 1 = 1
A
=
+
(1) h
0
(X), : X 1. Then
1 . = . 1 = . for each . h
n
(X. ).
(5) Associativity. (. ,) : = . (, :).
(6) Commutativity. . , = (1)
[x[[,[
, ..
One can also consider situations where (6) or (5) do not hold. This is the reason
for requiring (2) and (3) separately. For (5) it is required that the products are
dened. For the convenience of the reader we also display the properties in a table
and refer to the detailed description above.
17.2. Multiplicative Cohomology Theories 411
+
(. ,) =
+
.
+
,
(a) . =
(a |
.)
. (b) = (1)
[x[
B
(|
B
. b)
1 . = . = . 1
(. ,) : = . (, :)
. , = (1)
[x[[,[
, .
The cup product denes on h
+
(X. ) the structure of a Z-graded associative and
commutative algebra. If = 0 this algebra has a unit element 1
A
h
0
(X).
Moreover h
+
(X. ) becomes a unital graded left h
+
(X)-module. These structures
have the obvious naturality properties which follow from the axioms, e.g., a map
: X Y induces a unital algebra homomorphism
+
: h
+
(Y ) h
+
(X).
In particular the graded module h
+
= (h
n
) of the coefcient groups becomes a
unital commutative graded algebra. We make h
+
(X. ) into a unital left h
+
-module
via a . =
+
(a) . with : X 1 the unique map to a point. Morphisms
induced by continuous maps are then h
+
-linear. This is a particular case of the cross
product introduced later.
We list some consequences of naturality and stability. Let , h
n
(X. T) be
xed. Right multiplication by , yields a morphism of degree n from the exact
sequence of the pair into the exact sequence of the triple (X. L T. T). This
means: We have a commutative diagram
h
n
(X)

i
,
h
n
()

i
,
h
n1
(X. )
i
,
h
nn
(X. T)

h
nn
( L T. T)

h
nn1
(X. L T).
The r
,
in the middle is dened by multiplication of . h
n
() with the restriction of
, along h
n
(X. T) h
n
(. T) and then using h
n
(. T) h
n
(LT. T).
Let , h
n
(X) be xed. For each pair (. T) in X we obtain a product
r
,
: h
k
(. T) h
kn
(. T) by right multiplication with the restriction of , along
h
n
(X) h
n
(). The following diagram commutes:
h
n
()

i
,
h
n
(T)

i
,
h
n1
(. T)
i
,
h
nn
()

h
nn
(T)

h
nn1
(. T).
If (. T) is excisive, then the coboundary operators of the MV-sequences
^: h
n-1
( T. C) h
n
(. T) h
n
( L T. T) h
n
( L T. C).
^: h
n
(X. T) h
n
(. T) h
n
( L T. T) h
n1
(X. T)
commute with r
,
. Also in the general case 17.1.4, the boundary operator commutes
with products:
(17.2.1) Proposition. Let (:
0
.
1
) (X: X
0
. X
1
) be excisive triads. Let X
0
X and assume that also (LX

0
:
0
LX
0
0
.
1
L
0
1
) is excisive for X
0
= X
X
0
.
Then the diagram
h
i
(X
01
.
01
)
i
,
h
i[,[
(X
01
.
01
L X
0
01
)
Z
h
i1
(X. )
i
,
h
i1[,[
(X. L X
0
)
commutes. Here r
,
is right multiplication by , h(X. X
0
) (bottom) and multipli-
cation with the restriction of , to h(X
01
. X
0
01
) (top).
Proof. The coboundary operators are dened via suspension and appropriate in-
duced morphisms. The commutativity of the diagramthen amounts to the naturality
of the cup product and a compatibility (17.2.2) with the suspension.
h(1Y. 1T L d1Y ) h(1Y. 1Y
0
)
.

h(1Y. 1(T L 1
0
) L d1Y )
h(Y. T) h(Y. Y
0
)
cpr
.

h(Y. T L Y
0
)
c
commutes. (Notation: 1Y = 1 Y. d1Y = d1 Y etc.)

Proof. We use the associated cross product and (17.3.1), (17.3.3) in the computation
(e
1
.) (1
J
,) = (e
1
1
J
) (. ,).
(17.2.3) Example. Let (X. +) be a pointed space. The ring homomorphism
i : h
+
(X) h
+
has as kernel the two-sided h
+
-ideal h
+
(X. +). We have the
isomorphism h
+
(X. +) h
+
h
+
(X), (a. b) a
+
b, see 17.1.3. This
isomorphism transforms the cup product on h
+
(X) into the product
(a
1
. b
1
) (a
2
. b
2
) = (a
1
a
2
b
1
a
2
a
1
b
2
. b
1
b
2
).
(17.2.4) Example. The commutativity . . = (1)
[x[[x[
. . has for [.[
1 mod 2 the consequence 2(. .) = 0. Hence . . = 0, if multiplication by 2
is injective.
17.3. External Products 413
(17.2.5) Example. Let X =
1
L L
k
. Suppose a
i
h
n(i)
(X) are elements
in the kernel of the restriction h
n(i)
(X) h
n(i)
(
i
). Let b
i
h
n(i)
(X.
i
) be
a pre-image of a
i
. Then a
1
a
k
= 0, because this element is the image
of b
1
b
k
h
n
(X.
_
i

i
) = 0. This argument requires that the relative
products are dened.
This simple consequence of the existence of products has interesting geometric
consequences. The projective space C1
n
has a covering by n 1 afne (hence
contractible) open sets U
}
= {:
i
| [ :
}
,= 0]. Therefore a
0
a
n
= 0
for elements a
}
h
n(})
(C1
n
. +). Later we show the existence of an element
c H
2
(C1
n
: Z) with c
n
,= 0. Hence C1
n
cannot be covered by n contractible
open sets (more generally, by open sets U such that U C1
n
is null homotopic).
An analogous result holds for R1
n
.
(17.2.6) Example. The argument of the preceding example shows that products in
h
+
(S
n
. +), n _ 1 are trivial, a
1
a
2
= 0.
More generally, let X be a well-pointed space. Then products in h
+
(X. +)
are trivial. It sufces to prove this fact for the unreduced suspension
t
X; but this
space has a covering by two contractible open sets (cones).
Additively, the cohomology groups only depend on the stable homotopy type.
The product structure contains more subtle information.
(17.2.7) Example. Let : 1 T be any map. Then we make h
+
(1) into a
graded right h
+
(T)- module by the denition , . = ,
+
. for . h
+
(T)
and , h
+
(1). If : X Y is a morphism from : X T to q : Y T
in TOP
B
, then
+
: h
+
(Y ) h
+
(X) is h
+
(T)-linear. The same device works
for pairs of spaces over T. The coboundary operator is then also h
+
(T)-linear,
(, .) = , ..
17.3 External Products
Amultiplicative structure on a cohomology theory h
+
of external products consists
of a family of 1-linear maps (m. n Z)
h
n
(X. )
T
h
n
(Y. T) h
nn
((X. ) (Y. T)). . , . ,.
dened for a suitable class of pairs (X. ) and (Y. T) and in any case if the pair
(X T. Y ) is excisive in X Y . The products are dened if or T is empty.
These maps are required to satisfy the following axioms.
(1) Naturality. For continuous maps : (X. ) (X
t
.
t
) and g: (Y. T)
(Y
t
. T
t
) we have ( g)
+
(. ,) =
+
. g
+
,.
(2) Stability. Let (X T. Y ) be excisive. For . h
n
() and , h
n
(Y. T)
the relation . , =
t
(. ,) holds. Here
t
is the composition of the excision
isomorphism with for k = mn
t
: h
k
( (Y. T)) h
k
( Y L X T. X T)

h
k1
((X. ) (Y. T)).
(3) Stability. Let (X T. Y ) be excisive. For . h
n
(X. ) and , h
n
(T)
the relation . , = (1)
[x[
tt
(. ,) holds. Here
tt
is a similar composition of
excision with for k = mn
tt
: h
k
((X. ) T) h
k
(X T L Y. Y )

H
k1
((X. ) (Y. T)).
(4) Unit element. There is given 1 h
0
(1) as a further structure datum. It satises
1 . = . 1 = . (with respect to 1 X = X 1 = X).
(5) Associativity. (. ,) : = . (, :).
(6) Commutativity. Let t : X Y Y X, (.. ,) (,. .). Then for .
h
n
(X. ) and , h
n
(Y. T) the relation t
+
(. ,) = (1)
[x[[,[
, . holds.
Also in this case we display the properties in a table and refer to the detailed
explanation above.
( g)
+
(. ,) =
+
. g
+
,
. , =
t
(. ,)
. , = (1)
[x[
tt
(. ,)
1 . = . = . 1
(. ,) : = . (, :)
t
+
(. ,) = (1)
[x[[,[
, .
The termstability usually refers to compatibility with suspension. We explain this
for the present setup. As a rst applicationwe showthat the suspensionisomorphism
is given by multiplication with a standard element. Let e
1
h
1
(1. d1) be the image
of 1 h
0
under
1 h
0
h
0
(0) h
0
(d1. 1)
h
1
(1. d1) e
1
.
(17.3.1) Proposition. e
1
, = o(,).
h
n
(Y. T)

1

h
n
(d1Y. d1T)

h
n
(0Y. 0T)
h
n1
((1. d1) (Y. T) h
n
(d1Y L 1T. 1T)
h
n
(d1Y L 1T. 1Y L 1T).
17.3. External Products 415

Let

1 h
0
(d1) be the image of 1 h
0
(0) under h
0
(0) h
0
(d1. 1) h
0
(d1).
Then e
1
= (
1). Stability (2) of the cross product shows e

1
, =
-1
(
1 ,).
The maps and are isomorphisms by excision and h-equivalence. The outer
diagram path is , o(,).
Suppose we have a cup product. We construct an associated external product.
Let : (X. ) Y (X. ) and q : X (Y. T) (Y. T) be the projections
onto the factors. We dene for . h
n
(X. ) and , h
n
(Y. T) the product
. , =
+
, L q
+
,.
Conversely, suppose an external product is given. We dene an associated
-product. Let J : (X. L T) (X. ) (X. T) be the diagonal. Then we set
. , = J
+
(. ,).
(17.3.2) Proposition. The -product associated to a -product satises the axioms
of an external product. The -product associated to a -product satises the axioms
of an internal product. The processes and are inverse to each
other.
(17.3.3) Proposition. Let .
i
h
+
(X.
i
) and ,
i
h
+
(Y. T
i
). Then
(.
1
,
1
) (.
2
,
2
) = (1)
[x
2
[[,
1
[
(.
1
.
2
) (,
1
,
2
).
In particular h
+
(X) h
+
(Y ) h
+
(X Y ), . , . , is a homomorphism
of unital graded algebras.
(17.3.4) Proposition. Let s h
n
(S
n
. +) h
n
(S
n
) be the element which corre-
sponds to 1 h
0
under a suspension isomorphism. Then for each space J the
map
h
k
(J) h
k-n
(J) h
k
(J S
n
). (a. b) a 1 b s
is an isomorphism. These isomorphisms show that h
+
(J S
n
) is a free graded left
h
+
(J)-module with homogeneous basis 1 1, 1 s.
Proof. We start with the isomorphism
h
k-n
(J) h
k
(1
n
J. d1
n
J). . e
n
.
where e
n
= e
1
e
1
, see (17.3.1). Let s h
n
(S
n
. +) correspond to e
n
under
the isomorphismh
n
(S
n
. +) h
n
(1
n
,d1
n
. +) h
n
(1
n
. d1
n
). Then, by naturality,
h
k-n
(J) h
k
(S
n
J. + J). b s b
is an isomorphism. We now use the split exact sequence
0 h
k
(S
n
J. + J) h
k
(S
n
J) h
k
(J) 0
with splitting h
k
(J) h
k
(S
n
J), a 1 a = pr
+
(a) in order to see that
(a. b) 1 a s b is an isomorphism.
Under this isomorphism we have the following expression of the product struc-
ture
(a sb)(a
t
sb
t
) = aa
t
s(ba
t
(1)
[o[n
ab
t
)
since s
2
= 0.
We use commutativity to show that (a. b) a 1 b s is an isomorphism
as claimed. If we use 1 = 1
T
1
S
n and s = 1
T
s as basis elements for the
left h
+
(J)-module h
+
(J S
n
), then the h
+
(J) algebra is seen to be the graded
exterior algebra h
+
(J)s|,(s
2
).
Let T be a CW-complex. The skeleton ltration (J
k
h
i
(T) [ k N) on h
i
(T)
is dened by J
k
h
i
(T) = Ker(h
i
(T) h
i
(T
k-1
)).
(17.3.5) Proposition. The skeleton ltration is multiplicative: If a J
k
h
i
(T) and
b J
I
h
}
(T), then a b J
kI
h
i}
(T).
Proof. Choose pre-images a
t
h
i
(T. T
k-1
), b
t
h
I
(T. T
I-1
). Then a
t
b
t

h
i}
(TT. T
k-1
TLTT
I-1
). The product ab is the image of ab under
the diagonal J
+
. Acellular approximation J
t
: T TT of the diagonal J sends
T
kI-1
into (T T)
kI-1
and the latter is contained in T
k-1
T L T T
I-1
.
Naturality of the -product now shows that a b is contained in the image of
h
i}
(T. T
kI-1
) h
i}
(T).
(17.3.6) Corollary. Let n h
0
(T) be contained in J
1
h
0
(T). Then n
k
J
k
h
0
(T).
Thus if T is nite-dimensional, n is nilpotent and 1 n a unit.
Problems
1. Supply the proofs for (17.3.2) and (17.3.3).
2. Determine the algebras h
(S
n(1)
S
n(k)
) as graded h
-algebras (graded exterior

algebra).
17.4 Singular Cohomology
The singular cohomology theory is constructed from the singular chain complexes
by a purely algebraic process. The algebraic dual of the singular chain complex is
again a chain complex, and its homology groups are called cohomology groups. It
is customary to use a co terminology in this context.
Let C
-
= (C
n
. d
n
) be a chain complex of 1-modules. Let G be another 1-
module. We apply the functor Hom
T
(. G) to C
-
and obtain a chain complex
C
-
= (C
n
.
n
) of 1-modules with C
n
= Hom
T
(C
n
. G) and the 1-linear map
n
: C
n
= Hom
T
(C
n
. G) Hom
T
(C
n1
. G) = C
n1
17.4. Singular Cohomology 417
dened by
n
() = (1)
n1
d
n1
for Hom(C
n
. G).
For the choice of this sign see 11.7.4. The reader will nd different choices of
signs in the literature. Other choices will not effect the cohomology functors. But
there seems to be an agreement that our choice is the best one when it comes to
products.
For a pair of spaces (X. ) we have the singular chain complex S
-
(X. ). With
an abelian group G we set
S
n
(X. : G) = Hom(S
n
(X. ). G)
and use the coboundary operator above. Here we are using Hom
Z
. Elements in
S
n
(X. : G) are functions which associate to each singular simplex o : ^
n
X an
element of Gand the value 0 to simplices with image in . If Gis an 1-module, then
the set of these functions becomes an 1-module by pointwise addition and scalar
multiplication. A continuous map : (X. ) (Y. T) induces homomorphisms
#
= S
n
( ): S
n
(Y. T: G) S
n
(X. : G). (
#
)(o) = (o)
which are compatible with the coboundary operators. In this manner we obtain
a contravariant functor from TOP(2) into the category of cochain complexes of
1-modules. The n-th cohomology group of S
-
(X. : G) is denoted H
n
(X. : G)
and called the n-th singular cohomology module of (X. ) with coefcients in G.
We often write H
n
(X. : Z) = H
n
(X. ) and talk about integral cohomology,
in this case (G = Z,() mod- cohomology). Dualization of the split exact
sequence 0 S
-
() S
-
(X) S
-
(X. ) 0 yields again an exact sequence
0 S
-
(X. : G) S
-
(X: G) S
-
(: G) 0. It induces a long exact
cohomology sequence
H
n-1
(: G)

H
n
(X. : G) H
n
(X: G) H
n
(: G) .
(17.4.1) Remark. We recall the denition of the coboundary operator for the
present situation. Let : S
n-1
() G be a cocycle, i.e., d = 0. Extend
in an arbitrary manner to a function : S
n-1
(X) G. The element ( ) =
(1)
n
d: S
n
(X) G vanishes on S
n
(), since its restriction to S
n
() is d.
Therefore ( ) yields an element [ S
n
(X. : G). The coboundary operator is
then dened by the assignment | [|.
So far we have dened the data of a cohomology theory. If we apply Hom(. G)
to a chain homotopy we obtain a cochain homotopy; this yields the homotopy
invariance. The excision axiomholds, as the chain equivalence S
-
(XU. U) .
S
-
(X. ) induces a chain equivalence S
-
(X. : G) . S
-
(X U. U: G).
There exist several algebraic relations among homology and cohomology. The
rst one comes from the evaluation of the Hom-complex. Let us use coefcients in
a commutative ring 1. The evaluations
Hom(S
n
(X. ). 1) S
n
(X. : 1) 1. (c r) (c)r
induces a pairing (sometimes called Kronecker pairing)
H
n
(X. : 1) H
n
(X. : 1) 1. . , (.. ,).
This is due to the fact that the evaluations combine to a chain map, see 11.7.4.
(17.4.2) Proposition. For : (X. ) (Y. T), , H
n
(Y. T: 1), and .
H
n
(X. : 1) we have
(
+
(,). .) = (,.
+
(.)).
For a H
n-1
(: 1), b H
n
(X. : 1) we have
(a. b) (1)
n-1
(a. db) = 0.
Proof. We verify the second relation. Let a = |, Hom(S
n-1
(). Z). Let
: S
n-1
(X) Z be an extension of . Then (a) is represented by the homo-
morphism (1)
n
d: S
n
(X) Z (see (17.4.1)). Let , = c|, c S
n
(X). Then
(.. ,) = (1)
n
d(c). From d(c) S
n-1
() we conclude (d(c)) = (dc) =
(.. d,).
The canonical generator e
1
H
1
(1. d1) is represented by the singular simplex
s : ^
1
1, (t
0
. t
1
) t
1
. Let .| H
0
(X) denote the element represented by the
cochain which assumes the value 1 on . X and 0 otherwise.
1
be the generator which is the image of 0| H
0
(0)
under H
0
(0) H
0
(d1. 1)

H
1
(1. d1). Then (e
1
. e
1
) = 1.
The singular cohomology groups H
n
(X. : G) can be computed from the ho-
mology groups of (X. ). This is done via the universal coefcient formula. We
have developed the relevant algebra in (11.9.2). The application to topology starts
with the chain complex C = S(X. : 1) = S(X. )
Z
1 of singular chains
with coefcients in a principal ideal domain 1. It is a complex of free 1-modules.
Note that Hom
T
(S
n
(X. )
Z
1. G) Hom
Z
(S
n
(X. ). G) where in the second
group G is considered as abelian group (= Z-module). Then (11.9.2) yields the
universal coefcient formula for singular cohomology.
(17.4.4) Theorem. For each pair of spaces (X. ) and each 1-module G there
exists an exact sequence
0 Ext(H
n-1
(X. : 1). G) H
n
(X. : G) Hom(H
n
(X. : 1). G) 0.
The sequence is natural in (X. ) and in G and splits. In particular, we have
isomorphisms H
0
(X: G) Hom(H
0
(X). G) Map(
0
(X). G).
The statement that the splitting is not natural means that, although the term
in the middle is the direct sum of the adjacent terms, this is not a direct sum of
functors. There is some additional information that cannot be obtained directly
from the homology functors. The topological version of (11.9.6) is:
17.5. EilenbergMac Lane Spaces and Cohomology 419
(17.4.5) Theorem. Let 1 be a principal ideal domain. Assume that either
H
+
(X. : 1) is of nite type or the 1-module G is nitely generated. Then there
is a functorial exact sequence
0 H
n
(X. : 1) G H
n
(X. : G) Tor(H
n1
(X. : 1). G) 0.
The sequence splits.
(17.4.6) Proposition. Let M be a closed, connected, non-orientable n-manifold.
Then H
n
(M: Z) Z,2.
Proof. Since M is non-orientable, H
n
(M: Z) = 0. Theorem (17.4.4) for 1 =
G = Z then shows H
n
(M: Z) Ext(H
n-1
(M: Z). Z). From (16.3.4) we know
that H
n-1
(M: Z) Z,2 J with a nitely generated free abelian group. This is
also a consequence of Poincar duality, see (18.3.4).
Problems
1. Let 0 G
1
G
2
G
3
0 be an exact sequence of abelian groups. It induces a
short exact sequence of cochain complexes
0 Hom(S(X). G
1
) Hom(S(X). G
2
) Hom(S(X). G
3
) 0
and an associated long exact sequence of cohomology groups. The coboundary operator
: H
n
(X: G
3
) H
n1
(X: G
1
)
is a natural transformation of functors H
n
(: G
3
) H
n1
(: G
1
) and called a Bockstein
operator. A typical and interesting case arises from the exact sequence 0 Z,
Z,
2
Z, 0.
2. If the functor Hom(. G) preserves exact sequences, then no Ext-term appears in the
universal coefcient formula. Examples are
H
n
(X. : Q) Hom
Z
(H
n
(X. ). Q) Hom
Q
(H
n
(X. : Q). Q)
and H
n
(X: Z,) Hom
Z/)
(H
n
(X: Z,). Z,) for the prime eld Z, as coefcient
ring.
17.5 EilenbergMac Lane Spaces and Cohomology
The representability theorem(8.6.10) of Brown can be used to nd a natural isomor-
phismz: . 1(. n)| H
n
(: ) on the homotopy category of CW-complexes.
In this section we construct this isomorphism and give some applications. A nat-
ural transformation z is determined by its value on the identity of 1(. n), and
this value can be prescribed arbitrarily. Thus we have to nd a suitable element
|
n
H
n
(1(. n): ).
Let us begin with the special case n = 0. As a model for 1(. 0) we take
the abelian group with discrete topology. Since is discrete, a map X
is continuous if and only if it is locally constant. Moreover, all homotopies are
constant. Therefore X. 1(. 0)| = X. | is the group of locally constant maps.
A locally constant map is constant on a path component. Therefore the surjection
q : X
0
(X) induces an injective homomorphism
z
0
(X): X. | Map(
0
(X). ) Hom(H
0
(X). ) H
0
(X: ).
The last isomorphism is the one that appears in the universal coefcient formula.
If X is locally path connected, then the homomorphism induced by q is an isomor-
phism. Hence we have obtained a natural isomorphism on the category of locally
path connected spaces, in particular on the category of CW-complexes. If X is
connected but not path connected, then z
0
(X) is not an isomorphism.
Let now n _ 1 and write 1 = 1
n
= 1(. n). Consider the composition
|
n
H
n
(1: )
(1)
Hom(H
n
(1: Z). )
(2)
Hom(
n
(). )
(3)
Hom(. ) id .
The isomorphism(1) is the universal coefcient isomorphism. The isomorphism(2)
is induced by the Hurewicz isomorphism h:
n
(1. +) H
n
(1: Z), see (20.1.1).
It sends | S(n). 1|
0
=
n
(1) to
+
(:
n
) where :
n
H
n
(S(n): Z) is a suitable
generator. The isomorphism (3) is induced by a xed polarization j:
n
(1).
We dene |
n
as the element which corresponds to the identity of . Let z
n
be the
natural transformation which is determined by the condition z
n
id| = |
n
. Note
that category theory does not tell us yet that the z
n
(X): X. 1| H
n
(X: ) are
homomorphisms of abelian groups.
Let us compare z
n-1
and z
n
. It sufces to consider connected CW-complexes
andpointedhomotopyclasses. The diagramwiththe structure mape(n): 1
n-1

1
n
of the spectrum
X. 1
n-1
|
0

2
n-1
X. 1
n-1
|
0
e(n)

X. 1
n
|
0
2
n
H
n-1
(X: )
c

H
n
(X: )
is commutative up to sign, provided e(n)
+
(|
n
) = z
n
(e(n)) = o(|
n-1
). The
morphismz
n
(X) is a homomorphism, since the group structures are also induced
by the cogroup structure of X. In order to check the commutativity one has to
arrange and prove several things: (1) The Hurewicz homomorphisms commute with
suspensions. (2) The structure map e(n) and the polarizations satisfy e(n)
+

j
n-1
= j
n
:
n
(1
n
). (3) The homomorphisms from the universal coefcient
formula commute up to sign with suspension. Since the sign is not important for
the moment, we do not go into details.
17.5. EilenbergMac Lane Spaces and Cohomology 421
(17.5.1) Theorem. The transformation z
n
is an isomorphism on the category of
pointed CW-complexes.
Proof. We work with connected pointed CW-complexes. Both functors have value
0 for S
n
. m ,= n. The reader is asked to trace through the denitions and verify
that z
n
(S
n
) is an isomorphism. By additivity, z
n
(X) is an isomorphism for X a
wedge of spheres. Now one uses the cobre sequence
_
S
k-1
X
k-1
X
k
X
k
,X
k-1
X
k-1
and applies the functors . 1
n
|
0
and H
n
(: ). We use induction on k. Then z
n
is an isomorphism for
_
S
k-1
, X
k-1
, and X
k
,X
k-1
. The Five Lemma implies
that z
n
(X
k
) and hence also z
n
(X
k-1
) are surjective. By another application
of the Five Lemma we see that z
n
(X
k
) is also injective. This settles the case of
nite-dimensional CW-complexex. The general case follows fromthe fact that both
functors yield an isomorphism when applied to X
n1
X.
ACW-complex1(Z. n) canbe obtainedbyattachingcells of dimension _n 2
to S
n
. The cellular approximation theorem (8.5.4) tells us that the inclusion
i
n
: S
n
1(Z. n) induces for each CW-complex X of dimension at most n a
bijection
i
n
+
: X. S
n
|

X. 1(Z. n)|.
We combine this with the isomorphism z and obtain a bijection
X. S
n
|

H
n
(X: Z).
It sends a class | X. S
n
| tothe image of 1 H
n
(S
n
: Z) under
+
: H
n
(S
n
)
H
n
(X). Here we use the isomorphism H
n
(S
n
: ) which is (for an arbitrary
abelian group ) dened as the composition
H
n
(S
n
: ) Hom(H
n
(S
n
). ) Hom(
n
(S
n
). ) Hom(Z. ) .
Again we have used universal coefcients and the Hurewicz isomorphism.
(17.5.2) Proposition (Hopf). Let M = X be a closed connected n-manifold which
has an n-dimensional CW-decomposition. Suppose M is oriented by a fundamental
class :
. Then we have an isomorphism

H
n
(M: Z) Hom(H
n
(M). Z) Z
where the second isomorphism sends to (:
). The isomorphism M. S
n
|
H
n
(M: Z) Z maps the class | to the degree J( ) of .
(17.5.3) Proposition (Hopf). Let M be a closed, connected and non-orientable
n-manifold with a CW-decomposition. Then H
n
(M: Z) Z,2; hence we have a
bijection M. S
n
| Z,2. It sends | to the degree J
2
( ) modulo 2 of .
Proof. The universal coefcient theorem and H
n
(M: Z) = 0 show
Ext(H
n-1
(M). Z,2) H
n
(M: Z,2).
We use again that H
n-1
(M: Z) J Z,2 with a nitely generated free abelian
group J. This proves H
n
(M: Z,2) Z,2. Naturality of the universal coefcient
sequences is used to show that the canonical map H
n
(M: Z) H
n
(M: Z,2) is
an isomorphism. The commutative diagram
H
n
(S
n
: Z,2)
(
H
n
(M: Z,2)
Hom(H
n
(S
n
: Z,2). Z,2)

Hom(H
n
(M: Z,2). Z,2)
is used to show the assertion about the degree.
Problems
1. z is not always an isomorphism: = Z, n = 1 and the pseudo-circle.
17.6 The Cup Product in Singular Cohomology
Let 1 be a commutative ring. The cup product in singular cohomology with coef-
cients in 1 arises from a cup product on the cochain level
S
k
(X: 1) S
I
(X: 1) S
kI
(X: 1). [ [.
It is dened by
(1) ( [)(o) = (1)
[[[)[
(o[e
0
. . . . . e
k
|)[(o[e
k
. . . . . e
kI
|).
Here o : ^
kI
= e
0
. . . . . e
kI
| X is a singular (kl)-simplex. Let
0
. . . . .
n
|
be an afne n-simplex and t :
0
. . . . .
n
| X a continuous map. We denote by
t[
0
. . . . .
n
| the singular simplex obtained from the composition of t with the
map ^
n
0
. . . . .
n
|, e
}

}
. This explains the notation in (1).
(17.6.1) Proposition. The cup product is a chain map
S
-
(X: 1) S
-
(X: 1) S
-
(X: 1).
i.e., the following relation holds:
( [) = [ (1)
[[
[.
17.6. The Cup Product in Singular Cohomology 423
Proof. From the denition we compute ( [)(o) to be
(1)
([[1)[)[
k
i=0
(1)
i[[1
(o[e
0
. . . . . e
i
. . . . . e
k1
|)[(o[e
k1
. . . . . e
kI1
|)
and (1)
[[
( [)(o) to be
(1)
[[([)[1)
kI1
i=k
(1)
i[)[1
(o[e
0
. . . . . e
k
|)[(o[e
k
. . . . . e
i
. . . . . e
kI1
|).
If we add the two sums, the last term of the rst sum and the rst term of the second
sum cancel, the remaining terms yield (1)
[[[)[1
( [)(do), and this equals
(( [))(o).
We extend the cup product to the relative case. Suppose S
k
(X. : 1)
and S
I
(X. T: 1) are given. This means: vanishes on singular simplices
^
k
and [ vanishes on simplices ^
I
T. The relation (17.6.1) then shows
that L [ vanishes on the submodule S
kI
( T: 1) S
kI
(X: 1) generated
by simplices in and T. The pair (. T) is excisive for singular homology if
S
-
(T) = S
-
() S
-
(T) S
-
(LT) is a chain equivalence. The dual maps
S
-
( L T: 1) S
-
( T: 1) and S
-
(X. L T: 1) S
-
(X: T: 1) are
then chain equivalences. The last module consists of the cochains which vanish on
S
-
( T); let H
+
(X: T: 1) denote the corresponding cohomology group.
Thus we obtain a cup product (again a chain map)
S
-
(X. : 1) S
-
(X. T: 1) S
-
(X: T: 1).
We pass to cohomology, obtain H
+
(X. : 1)H
+
(X. T: 1) H
+
(X: T: 1)
and in the case of an excisive pair a cup product
H
+
(X. : 1) H
+
(X. T: 1) H
+
(X. L T: 1).
We now prove that the cup product satises the axioms of Section 2. From
the denition (1) we see that naturality and associativity hold on the cochain level.
The unit element 1
A
H
0
(X: 1) is represented by the cochain which assumes the
value 1 on each 0-simplex. Hence it acts as unit element on the cochain level. In
the relative case the associativity holds for the representing cocycles in the group
H
+
(X: T C: 1). In order that the products are dened one needs that the
pairs (. T), (T. C), ( L T. C) and (. T L C) are excisive.
Commutativity does not hold on the cochain level. We use: The homomor-
phisms
j: S
n
(X) S
n
(X). o c
n
o
n
with c
n
= (1)
(n1)n{2
and o = o[e
n
. e
n-1
. . . . . e
0
| form a natural chain map
which is naturally chain homotopic to the identity (see (9.3.5) and Problem1 in that
section). The cochain map j
#
induced by j satises
j
#
j
#
[ = (1)
[[[)[
j
#
( [).
Since j
#
induces the identity on cohomology, the commutativity relation follows.
The stability relation (2) is a consequence of the commutativity of the next
diagram (which exists without excisiveness). Coefcients are in 1.
H
i
()H
}
(X. T)

1
H
i
()H
}
(. . T)
.

H
i}
(. T)
H
i1
(X. )H
}
(X. T)
.

H
i}1
(X: T)
where
t
is the composition of H
i}
(. T)

H
i}
( T: T), which is
induced by the algebraic isomorphism
S
-
(),S
-
( T) S
-
( T),S
-
(T).
and : H
i}
( T: T) H
i}1
(X: T). In order to verify the commu-
tativity, one has to recall the construction of , see (17.4.1). Suppose | H
i
()
and [| H
}
(X. ) are given. Let S
i
(X) be an extension of ; then
vanishes on S
i
(), and the cochain S
i1
(X. ) represents |. The image
of | [| along the down-right path is represented by L [, and one veries
that the image along the right-down path is represented by ( L [). Since [ is a
cocycle, the representing elements coincide. A similar verication can be carried
out for stability (3).
One can deal with products from the view-point of EilenbergZilber transfor-
mations. We have the tautological chain map
S
-
(X: 1) S
-
(Y : 1) Hom(S
-
S
-
. 1 1).
We compose it with the ring multiplication 1 1 1 and an EilenbergZilber
transformation S
-
(X Y ) S
-
(X) S
-
(Y ) and obtain a chain map
S
-
(X: 1) S
-
(Y : 1) S
-
(X Y : 1). g g.
a -product on the cochain level. If the pair (Y. X T) is excisive, we obtain
a -product
S
-
(X. : 1) S
-
(Y. T: 1) S
-
((X. ) (Y. T): 1).
Our previous explicit denition of the cup product arises in this manner from the
AlexanderWhitney equivalence and the related approximation of the diagonal.
We can now apply the algebraic Knneth theorem for cohomology to the singular
cochain complexes and obtain:
17.7. Fibration over Spheres 425
(17.6.2) Theorem. Let 1 be a principal ideal domain and H
+
(Y. T: 1) of nite
type. Assume that the pair ( Y. X T) is excisive. Then there exists a natural
short exact sequence
0
i}=n
H
i
(X. : 1) H
}
(Y. T: 1) H
n
((X. ) (Y. T): 1)
i}=n1
H
i
(X. : 1) + H
}
(Y. T: 1)
Problems
1. We have dened the cup product for simplicity by explicit formulas. One can use instead
EilenbergZilber morphisms. Let us consider the absolute case. Consider the composition
S
q
(X: 1) S
q
(Y : 1) = Hom(S
)
X. 1) Hom(S
q
(Y ). 1)
Hom(S
)
(X) S
q
(Y ). 1) Hom(S
)q
(X Y ). 1)
where the rst morphism is the tautological map and the second induced by an Eilenberg
Zilber morphism S
+
(X Y ) S
+
(X) S
+
(Y ). Then this composition induces the
-product in cohomology. The cup product in the case X = Y is obtained by composition
with S
+
(X) S
+
(X X) induced by the diagonal. Instead we can go directly from
Hom(S
)
(X) S
q
(X). 1) to Hom(S
)q
(X). 1) by an approximation of the diagonal. Our
previous denition used the AlexanderWhitney diagonal.
17.7 Fibration over Spheres
Let : X S
n
be a bration (n _ 2). We write as usual S
n
= D
n
L D
n
-
and
S
n-1
= D
n
D
n
-
. Let b
0
S
n-1
be a base point. We set X
=
-1
(D
n
),
X
0
=
-1
(S
n-1
), and J =
-1
(b
0
). From the homotopy theorem of brations
we obtain the following result.
(17.7.1) Proposition. There exist h-equivalences
: D
n
X X
over
D
n
such that
(b
0
. ,) = , for , J. The h-inverses [
also satisfy [
(,) =
(b
0
. ,) andthe brewise homotopies of [
and
tothe identity are constant

on J. Since X
0
X
X are closed cobrations, we have a MayerVietoris

sequence for (X
. X
-
).
We use the data of (17.7.1) in order to rewrite the MV-sequence. We work with a
multiplicative cohomology theory. The embedding : J D
n
J, , (b
0
. ,)
is an h-equivalence. Therefore we have isomorphisms
i
+
: h
k
(X
h
k
(D
n
J)
}
h
k
(J).
The restriction of
gives us another isomorphism
: h
k
(X
0
)

h
k
(S
n-1
J).
We insert these isomorphisms into the MV-sequence of (X
. X
-
)
h
k
(X)

=
h
k
(X
) h
k
(X
-
)
i
h
k
(X
0
)
h
k
(X)
(1)
h
k
(J) h
k
(J)
(2)
h
k
(S
n-1
J).
The two components of (1) equal i
+
where i : J X. The rst component of (2)
is induced by the projection pr : S
n-1
J J. We write [
-

in the form
(s. ,) (s. (s. ,)). Then the second component of (2) is
+
. Both maps yield
the identity when composed with : J S
n-1
J, , (b
0
. ,). The product
structure provides us with an isomorphism
h
k
(J) h
k-n1
(J) h
k
(S
n-1
J). (a. b) 1 a s b.
We also use this isomorphism to change the MV-sequence. We set
+
(.) = 1 . s (.). (.) h
k-n1
(J).
The relation
+
(1 a s b) = a shows that
+
(.) has the displayed form.
(17.7.2) Theorem (Wang Sequence). There exists an exact sequence
h
k
(X)
i
h
k
(J)

h
k-n1
(J) h
k-n1
(X) .
The map is a derivation, i.e., (. L ,) = (.) L , (1)
[x[(n-1)
. L ,.
Proof. We start with the modied MV-sequence
h
k
(X)
(1)
h
k
(J) h
k
(J)
(2)
h
k
(J) h
k-n1
(J) .
The morphism (1) is as before, and (2) has the form (a. b) (a b. (b)). Then
we form the quotients with respect to the left h
k
(J) summands in order to obtain
the stated exact sequence.
Since
+
is a homomorphism and s
2
= 0, we obtain
+
(.,) = 1 ., s (.,)
+
(.)
+
(,) = (1 . s (.))(1 , s (,))
= 1 ., s (.) , (1)
[x[[x[
s . (,).
This proves the derivation property of .
We can, of course, also consider the MV-sequence in homology. It assumes
after an analogous rewriting the form
H
q
(J)
i
H
q
(X) H
q-n
(J)

H
q-1
(J) .
17.8. The Theorem of Leray and Hirsch 427
Problems
1. As an example for the use of the Wang sequence compute the integral cohomology ring
H
(CS
n1
) of the loop space of S
n1
. Use the path bration CS
n1
i
1
)
S
n1
with contractible 1.
If n = 0, then CS
1
is h-equivalent to the discrete space Z. So let n _ 1. Since 1
is contractible, the Wang sequence yields H
q
(CS
n1
) H
q-n
(CS
n1
) and therefore
H
k
(CS
n1
) Z for k 0 mod (n) and 0 otherwise. Similarly for cohomology
H
k
(CS
n1
) Z for k 0 mod (n) and zero otherwise. Using the isomorphism we
dene inductively elements :
0
= 1 and :
k
= :
k-1
for k _ 1.
Let n be even. Then k:
k
= :
k
1
for k _ 1. For the proof use induction over k and the
derivation property of .
The relation above yields the multiplication rule
:
k
:
l
=
_
k l
k
_
:
kl
.
A multiplicative structure of this type is called a polynomial ring with divided powers. With
coefcient ring Qone obtains a polynomial ring H
(CS
n1
: Q) Q:
1
|.
Let n be odd. Then :
1
:
2k
= :
2k1
, :
1
:
2k1
= 0, and :
k
2
= k:
2k
.
Again use induction and the derivation property. Since :
2
1
= :
2
1
, one has :
2
1
= 0.
Then (:
1
:
2k
) = (:
1
):
2k
:
1
(:
2k
) = :
2k
:
1
:
2k-1
= :
2k
= (:
2k1
), hence
:
1
:
2k
= :
2k1
, since is an isomorphism. Next compute :
1
:
2k1
= :
1
(:
1
:
2k
) =
:
2
1
:
2k
= 0. For the last formula use that is a derivation of even degree which maps
:
2k
to :
2k-2
. The induction runs then as for even n. The elements :
2k
generate a polynomial
algebra with divided powers and :
1
generates an exterior algebra.
17.8 The Theorem of Leray and Hirsch
The theorem of Leray and Hirsch determines the additive structure of the coho-
mology of the total space of a bration as the tensor product of the cohomology of
the base and the bre. We work with singular cohomology with coefcients in the
ring 1. A relative bration
(J. J
t
) (1. 1
t
) T
consists of a bration : 1 T such that the restriction
t
: 1
t
T to the
subspace 1
t
of 1 is also a bration. The bres of and
t
over a base point + T
are J and J
t
. The case 1
t
= 0 and hence J
t
= 0 is allowed. We assume that T
is path connected.
(17.8.1) Theorem (LerayHirsch). Let (J. J
t
)
i
(1. 1
t
)
]
T be a relative
bration. Assume that H
n
(J. J
t
) is for each n a nitely generated free 1-module.
Let c
}
H
+
(1. 1
t
) be a family of elements such that the restrictions i
+
(c
}
) form
an 1-basis of H
+
(J. J
t
). Then
1: H
+
(T) H
+
(J. J
t
) H
+
(1. 1
t
). b i
+
(c
}
)
+
(b) c
}
is an isomorphism of 1-modules. Thus H
+
(1) is a free graded H
+
(T)-module
with basis {c
}
].
We explain the statement of the theorem. The source of 1is a direct sumof mod-
ules H
k
(T) H
I
(J. J
t
). The elements i
+
c
which are contained in H

I
(J. J
t
)
are a nite 1-basis of the 1-module H
I
(J. J
t
). A basic property of the tensor
product says that each element has a unique expression of the form
i
+
(c
). b
H
k
(T).
By the conventions about tensor products of graded modules, 1 is a map of degree
zero between graded modules.
Proof. Let T be a subspace. We have the restricted brations (J. J
t
)
(1[. 1
t
[) with 1[ =
-1
() and 1
t
[ = 1
t
1[. The elements
c
}
yield by restriction elements c
}
[ H
+
(1[. 1
t
[) which again restrict to a
basis of H
+
(J. J
t
).
We rst prove the theorem for CW-complexes T by induction over the skeleta
T
n
. If T
0
= {+], then 1 has the form H
0
(T
0
) H
k
(J. J
t
) H
k
(J. J
t
) and
it is an isomorphism by the unit element property of the cup product.
Suppose the theoremholds for the (n1)-skeletonT
n-1
. We write T
n
= ULV ,
where U is obtained from T
n
by deleting a point in each open n-cell and V is the
union of the open n-cells. We use the MV-sequence of U. V and 1[U. 1[V and
obtain a commutative diagram
H
+
(U L V ) M
+
1
IL1
H
+
(1[U L V . 1
t
[U L V )
H
+
(U) M
+
H
+
(V ) M
+
1
I
1
1
H
+
(1[U. 1
t
[U) H
+
(1[V. 1
t
[V )
H
+
(U V ) M
+
1
I1
H
+
(1[U V . 1
t
[U V ).
The left column is the tensor product of the MV-sequence for U. V with the graded
module M
+
= H
+
(J. J
t
). It is exact, since the tensor product with a free module
preserves exactness. We showthat 1
U
. 1
V
and 1
UV
are isomorphisms. The Five
Lemma then shows that 1
ULV
is an isomorphism. This nishes the induction step.
17.8. The Theorem of Leray and Hirsch 429
Case U. We have the commutative diagram
H
+
(U) H
+
(J. J
t
)
1
I
H
+
(1[U. 1
t
[U)
H
+
(T
n-1
) H
+
(J. J
t
)

H
+
(1[T
n-1
. 1
t
[T
n-1
).
Since T
n-1
U, 1[T
n-1
1[U, and 1
t
[T
n-1
1
t
[U are deformation retracts,
the vertical maps are isomorphisms. We use the induction hypothesis and see that
1
U
is an isomorphism.
Case V . The set V is the disjoint union of the open n-cells V =

}
e
n
}
. We
obtain a commutative diagram
H
+
(V ) H
+
(J. J
t
)
1
1
(1)
H
+
(1[V. 1
t
[V )
(2)
_
H
+
(e
n
}
)
_
H
+
(J. J
t
)
(3)
H
+
(1[e
n
}
. 1
t
[e
n
}
)
=
_
H
+
(e
n
}
) H
+
(J. J
t
)
_
1
c
n
H
+
(1[e
n
}
. 1
t
[e
n
}
).
(1) and (2) are isomorphisms by the additivity of cohomology. The map (3) is the
direct sum of homomorphisms of the type
_
M
}
_
N

(M
}
N) with a
nitely generated free module N and other modules M
}
. In a situation like this
the tensor product commutes with the product. Hence (3) is an isomorphism. The
homomorphisms 1(e
n
}
) are isomorphisms, since e
n
}
is pointed contractible.
Case U V . We combine the arguments of the two previous cases. By additivity
and nite generation we reduce to the case of Ue
n
}
, a cell with a point deleted. This
space has the (n 1)-sphere as a deformation retract. By induction, the theorem
holds for an (n 1)-sphere.
From the nite skeleta we now pass to arbitrary CW-complexes via the lim
lim
1
-sequence (17.1.6). For general base spaces T we pull back the bration along
a CW-approximation.
(17.8.2) Example. Consider the product bration = pr
B
: T (J. J
t
) T.
Let H
+
(J. J
t
) be a free 1-module with homogeneous basis (J
}
[ J), nite
in each dimension. Let c
}
= pr
+
T
J
}
. Then i
+
c
}
= J
}
. Therefore (17.8.1) says in
this case that
H
+
(T) H
+
(J. J
t
) H
+
(T (J. J
t
)). . , . ,
is an isomorphism (of graded algebras). This is a special case of the Knneth
formula.
(17.8.3) Remark. The methods of proof for (17.8.1) (induction, MayerVietoris
sequences) gives also the following result. Suppose H
k
(J. J
t
) = 0 for k < n.
Then H
k
(1. 1
t
) = 0 for k < n.
Let now h
+
be an arbitrary additive and multiplicative cohomology theory and
(J. J
t
) (1. 1
t
) T a relative bration over a CW-complex. We prove a
Leray- Hirsch theorem in this more general situation. We assume now that there
is given a nite number of elements t
}
h
n(})
(1. 1
t
) such that the restrictions
t
}
[b h
n(})
(1
b
. 1
t
b
) to each bre over b are a basis of the graded h
+
-module
h
+
(1
b
. 1
t
b
). Under these assumptions:
(17.8.4) Theorem (LerayHirsch). h
+
(1. 1
t
) is a free left h
+
(T)-module with
basis (t
}
).
Proof. Let us denote by h
+
(C)(t ) the free graded h
+
(C)-module with (formal)
basis t
}
in degree n( ). We have the h
+
(C)-linear map of degree zero
(C): h
+
(C)(t ) h
+
(1[C. 1
t
[C)
which sends t
}
to t
}
[C. These maps are natural in the variable C T. We view
h
+
()(t ) as a cohomology theory, a direct sum of the theories h
+
() with shifted
degrees. Thus we have MayerVietoris sequences for this theory. If U and V are
open in T, we have a commutative diagram of MV-sequences.
h
+
(U L V )(t )
(ULV )
h
+
(1[U L V. 1
t
[U L V )
h
+
(U)(t ) h
+
(V )(t )
(U)(V )
h
+
(1[U. 1
t
[U) h
+
(1[V. 1
t
[V )
h
+
(U V )(t )
(UV )
h
+
(1[U V. 1
t
[U V )
We use this diagram as in the proof of (17.8.1). We need for the inductive proof
that (e) is an isomorphism for an open cell e. This follows from two facts:
(1) (1) is an isomorphism for a point 1 = {b] T, by our assumption about
the t
}
.
(2) (1
1
. 1
t
1
) (1[e. 1
t
[e) induces an isomorphism in cohomology, since
1
1
1[e is a homotopy equivalence by the homotopy theorem for brations.
The niteness of the set {t
}
] is used for the compatibility of products and nite
sums. The passage from the skeleta of T to T uses again (17.1.6).
There is a similar application of (17.8.4) as we explained in (17.8.2).
17.9. The Thom Isomorphism 431
17.9 The Thom Isomorphism
We again work with a cohomology theory which is additive and multiplicative.
Under the obvious niteness conditions (e.g., nite CW-complexes) additivity is
not needed.
Let (.
t
): (1. 1
t
) T be a relative bration over a CW-complex. A Thom
class for is an element t = t () h
n
(1. 1
t
) such that the restriction to each bre
t
b
h
n
(J
b
. J
t
b
) is a basis of the h
+
-module h
+
(J
b
. J
t
b
). We apply the theorem of
LerayHirsch (17.8.4) and obtain:
(17.9.1) Theorem (Thom Isomorphism). The Thom homomorphism
: h
k
(T) h
kn
(1. 1
t
). b
+
(b) t
is an isomorphism.
Let us further assume that induces an isomorphism
+
: H
+
(T) H
+
(1).
We use the Thomisomorphismand the isomorphism
+
in order to rewrite the exact
sequence of the pair (1. 1
t
); we set ^ =
-1
.
h
k-1
(1
t
)

L
L
L
L
L
L
L
L
L
L
h
k
(1. 1
t
)
}
h
k
(1)

h
k
(1
t
)
h
k-n
(T)

h
k
(T)
]
(]
/
)
v
v
v
v
v
v
v
v
v
Let e = e() H
n
(T) be the image of t under h
n
(1. 1
t
)
}
h
n
(1)
]
h
n
(T).
We call e the Euler class of with respect to t . From the denitions we verify that
J is the cup product with e, i.e., J(.) = . e.
(17.9.2) Theorem (Gysin Sequence). Let (1. 1
t
) T be a relative bration as
above such that
+
: h
+
(T) h
+
(1) with Thom class t and associated Euler class
e h
n
(T). Then we have an exact Gysin sequence
h
k-1
(1
t
) h
k-n
(T)
(1)
h
k
(T)
(2)
h
k
(1
t
) h
k-n1
(T) .
(1) is the cup product . . e and (2) is induced by
t
.
We discuss the existence of Thom classes for singular cohomology H
+
(: 1).
In this case it is not necessary to assume that T is a CW-complex (see (17.8.1)). We
have for each path n: 1 T from b to c a bre transport n
#
dened as follows:
Let q : (X. X
t
) 1 be the pullback of : (1. 1
t
) T along n. Then we have
isomorphisms induced by the inclusions
n
#
: H
n
(J
c
. J
t
c
)

H
n
(X. X
t
)

H
n
(J
b
. J
t
b
).
The homomorphismn
#
only depends on the class of nin the fundamental groupoid.
In this manner we obtain a transport functor bre cohomology.
(17.9.3) Proposition. We assume that H
+
(J
b
. J
t
b
: 1) is a free 1-module with a
basis element in H
n
(J
b
. J
t
b
: 1). A Thom class exists if and only if the transport
functor is trivial.
Proof. Let t be a Thomclass and n: 1 T a path fromb to c. Then n
#
i
+
c
= i
+
b
,
where i
b
denotes the inclusion of the bre over b. Thus n
#
sends the restricted
Thom class to the restricted Thom class and is therefore independent of the path.
(In general, the transport is trivial on the image of the i
+
c
.)
Let nowthe transport functor be trivial. Then we x a basis element in a particu-
lar bre H
n
(1
b
. 1
t
b
) and transport it to any other bre uniquely (T path connected).
A Thom class t
C
H
n
(1[C. 1
t
[C) for C T is called distinguished, if the re-
striction to each bre is the specied basis element. This requirement determines t
C
.
By a MV-argument and (17.8.3) we prove by induction that H
k
(1[T
n
. 1
t
[T
n
) = 0
for k < n and that a distinguished Thom class exists. Then we pass to the limit and
to general base spaces as in the proof of (17.8.1).
The preceding considerations can be applied to vector bundles. Let : 1 T
be a real n-dimensional vector bundle and 1
0
the complement of the zero section.
Then for each bre H
n
(1
b
. 1
0
b
) 1 and is a homotopy equivalence. A Thom
class t () H
n
(1. 1
0
: 1) is called an 1-orientation of . If it exists, we have
a Thom isomorphism and a Gysin sequence. We discuss the existence of Thom
classes and its relation to the geometric orientations.
(17.9.4) Theorem. There exists a Thom class of with respect to singular coho-
mology H
+
(: Z) if and only if the bundle is orientable. The Thom classes with
respect to H
+
(: Z) correspond bijectively to orientations.
Proof. Let us consider bundles over CW-complexes. Let t be a Thom class. Con-
sider a bundle chart : U R
n

-1
(U) over a path connected open U. The
image of t [U in H
n
(U (R
n
. R
n
0)) H
0
(U) under
+
and a canonical suspen-
sion isomorphism is an element c(U) which restricts to c(u) = 1 for each point
u U, and u c(u) is constant, since U is path connected. We can therefore
change the bundle chart by an automorphism of R
n
such that c(u) = 1 for each u.
Bundle charts with this property yield an orienting bundle atlas.
Conversely, suppose has an orienting atlas. Let : U R
n

-1
(U) be a
positive chart. From a canonical Thom class for U R
n
we obtain via a local
Thom class t
U
for
-1
(U). Two such local Thom classes restrict to the same Thom
class over the intersection of the basic domains, since the atlas is orienting. We can
now paste these local classes by the MayerVietoris technique in order to obtain a
global Thom class.
As a canonical generator of H
n
(R
n
. R
n
0: Z) we take the element e
(n)
satisfy-
ing the Kronecker pairing relation (e
(n)
. e
n
) = 1 where e
n
H
n
(R
n
. R
n
0: Z) is
the n-fold product e
1
e
1
of the canonical generator e
1
H
1
(R
1
. R
1
0: Z).
If a bundle is oriented and
b
: R
n
b
a positive isomorphism, then we require
+
b
t () = e
(n)
for its associated Thom class t ().
Let t (j) h
n
(1. 1
0
) be a Thom class of j: 1 T and e(j) h
n
(T) the
associated Euler class, dened as the restriction of t (j) to the zero section
t (j) h
n
(1. 1
0
) h
n
(1)
x
h
n
(T) e(j).
Here is a geometric property of the Euler class:
(17.9.5) Proposition. Suppose j has a section which is nowhere zero. Then the
Euler class is zero.
Proof. Let s
t
: T 1
0
be a map such that js
t
= id. The section s
t
is homotopic
to the zero section by a linear homotopy in each bre. Therefore e(j) is the image
of t (j) under a map
h
n
(1. 1
0
) h
n
(1) h
n
(1
0
)
x
/
h
n
(T)
and therefore zero.
The Thom classes and the Euler classes have certain naturality properties. Let
+
: j be a bundle map. If t (j) is a Thom class, then
+
t (j) is a Thom class
for and
+
e(j) is the corresponding Euler class. If : X T and j: Y C
are bundles with Thom classes, then the -product t () t (j) is a Thom class for
j and e() e(j) is the corresponding Euler class.
In general, Thom classes are not unique. Let us consider the case of a trivial
bundle = pr
2
: R
n
T T. It has a canonical Thom class pr
+
2
e
n
. If t ()
is an arbitrary Thom class, then it corresponds under the suspension isomorphism
h
0
(T) h
n
(R
n
T. R
n
0
T) to an element () with the property that its restriction
to each point b T is the element 1 h
0
(b). Under reasonable conditions, an
element with this property (call it a point-wise unit) is a (global) unit in the ring
h
0
(T). We call a Thom class for a numerable bundle strict if the restrictions to
the sets of a numberable covering correspond under bundle charts and suspension
isomorphism to a unit in h
0
.
(17.9.6) Proposition. Let Ube a numerable covering of X. Let c h
0
(X) be an
element such that its restriction to each U Uis a unit. Then c is a unit.
Proof. Let X = U L V and assume that (U. V ) is excisive. Let c[U = c
U
and
c[V = c
V
be a unit. Let j
U
. j
V
be inverse to c
U
. c
V
. Then j
U
and j
V
have
the same restriction to U V . By the exactness of the MV-sequence there exists
j h
0
(X) with j[U = j
U
. j[V = j
V
. Then . = cj 1 has restriction 0 in U
and V . Let .
U
h
0
(X. U) be a pre-image of . and similarly .
V
h
0
(X. V ).
Then .
U
.
V
= 0 and hence .
2
= 0. The relation 1 = cj(2 cj) shows that c is
a unit. Restrictions of units are units. By additivity, if X is the disjoint union of U
and c[U is a unit for each U U, then c is a unit. We nish the proof as in the
proof of (17.9.7).
The Thom isomorphism is a generalized (twisted) suspension isomorphism. It
is given by the product with a Thom class. Let : 1 T be an n-dimensional
real vector bundle and t () h
n
(1. 1
0
) a Thom class with respect to a given
multiplicative cohomology theory. The Thom homomorphism is the map
(): h
k
(T. ) h
kn
(1(). 1
0
() L 1(
)). . . t () =
+
(.) t ()
where
+
: h
k
(T. ) H
k
(1(). 1
0
()) is the homomorphism induced by .
The Thom homomorphism denes on h
+
(1. 1
0
) the structure of a left graded
h
+
(T)-module. The Thom homomorphism is natural with respect to bundle maps.
Let
1()
T
1(j)
)
T
(
C
be a bundle map. Let t (j) be a Thom class. We use t () = J
+
t (j) as the Thom
class for . Then the diagram
h
k
(C. D)
()
h
kn
(1(j). 1
0
(j) L 1(j
T
))
T
h
k
(T. )
())
h
kn
(1(j). 1
0
(j) L 1(j
))
is commutative. We assume that : (T. ) (C. D) is a map of pairs.
The Thom homomorphism is also compatible with the boundary operators. Let
t (
) be the restriction of t (). Then the diagram

h
k
()

(
.
)
h
k1
(T. )
()
h
kn
(1(
). 1
0
(
))
h
kn
(1
0
() L 1(
). 1
0
(
))
h
kn1
(1(). 1
0
() L 1(
))
is commutative.
The Thomhomomorphisms are also compatible with the morphisms in the MV-
sequence. We nowconsider the Thomhomomorphismunder a different hypothesis.
(17.9.7) Theorem. The Thom homomorphism of a numerable bundle with strict
Thom class is an isomorphism.
Proof. Let be a numerable bundle of nite type. By hypothesis, T has a nite
numerable covering U
1
. . . . . U
t
such that the bundle is trivial over each U
}
and the
Thom class is strict over U
}
. We prove the assertion by induction over t . For t = 1
it holds by the denition of a strict Thom class. For the induction step consider
C = U
1
L LU
t-1
and D = U
t
. By induction, the Thom homomorphism is an
isomorphism for C, D, and C D. Now we use that the Thom homomorphism is
compatible with the MV-sequence associated to C, D. By the Five Lemma we see
that () is an isomorphism over C L D.
Suppose the bundle is numerable over a numerable covering U. Assume that for
each U Uthe Thom class t (
U
) is strict. In that case (
U
) is an isomorphism.
For each V U the Thom class t (
V
) is also strict. There exists a numerable
covering (U
n
[ n N) such that [U
n
is numerable of nite type with strict Thom
class. Let (o
n
[ n N) be a numeration of (u
n
). Set : T 0. o , (.) =
n
no
n
(.). If .
_
n
}=1
supp(o
}
), then 1 =

}_1
o
}
(.) =

}>n
o
}
(.) and
therefore
(.) =
}>n
o
}
(.) _ (n 1)
}>n
o
}
(.) = n 1.
Hence
-1
0. n| is contained in
_
n
}=1
supp(o
}
). Hence
-1
|r. s is always con-
tained in a nite number of V
}
and therefore the bundle over such a set is numerable
of nite type. The sets C
n
=
-1
|2n 1. 2n 1 are open and disjoint. Over C
n
the bundle is of nite type. By additivity, we have for C =
_
C
n
the isomorphism
h
k
(1[C. 1
0
[C)
h
k
(1[C
n
. 1
0
[C
n
). The Thomclasses over C
n
yield a unique
Thom class over C. Now we use the same argument for D
n
=
-1
|2n. 2n 2 ,
D =
_
D
n
and C D =
_
-1
|n. n 1 and then apply the MV-argument
to C, D.
(17.9.8) Example. Let ;
1
(n): H(1) C1
n
be the canonical line bundle in-
troduced in (14.2.6). A complex vector bundle has a canonical orientation and
an associated Thom class (17.9.4). Let c H
2
(C1
n
) be the Euler class of
;
1
(n). The associated sphere bundle is the Hopf bration S
2n1
C1
n
. Since
H
k
(S
2n1
) = 0 for 0 < k < 2n 1 the multiplications by the Euler class
c H
2
(C1
n
) are isomorphisms
Z H
0
(C1
n
) H
2
(C1
n
) H
2n
(C1
n
)
and similarly H
k
(C1
n
) = 0 for odd k. We obtain the structure of the cohomology
ring
H
+
(C1
n
) Zc|,(c
n1
).
In the innite case we obtain H
+
(C1
o
: 1) 1c| where 1 is an arbitrary com-
mutative ring.
A similar argument for the real projective space yields for the cohomology ring
H
+
(R1
n
: Z,2) Z,2n|,(n
n1
) with n H
1
(R1
n
: Z,2) and for the innite
projective space H
+
(R1
o
: Z,2) Z,2n|.
(17.9.9) Example. The structure of the cohomology ring of R1
n
can be used to
give another proof of the BorsukUlam theorem: There does not exist an odd map
J : S
n
S
n-1
.
For the proof let n _ 3. (We can assume this after suspension.) Suppose
there exists an odd map J. It induces a map : R1
n
R1
n-1
of the orbit
spaces. Let : 1 S
n
be a path from . to .. Composed with the orbit map
n
: S
n
R1
n
we obtain a loop
n
that generates
1
(R1
n
) Z,2. The path
u = J from J(.) to J(.) = J(.) yields a loop that generates
1
(R1
n-1
).
Hence
+
:
1
(R1
n
)
1
(R1
n-1
) is an isomorphism. This fact implies (uni-
versal coefcients) that
+
is an isomorphism in H
1
(: Z,2). Since n
n
= 0 in
H
+
(R1
n-1
: Z,2) but n
n
,= 0 in H
+
(R1
n
: Z,2), we have arrived at a contradic-
tion.
Problems
1. A point-wise unit is a unit under one of the following conditions: (1) For singular coho-
mology H
(: 1). (2) T has a numerable null homotopic covering. (3) T is a CW-complex.

2. Prove the Thom isomorphism for vector bundles over general spaces and for singular co-
homology.
3. Let : 1() T and j: 1(j) T be vector bundles with Thom classes t () and t (j).
Dene a relative Thom homomorphism as the composition of . . t (),
h
k
(1(j). 1
0
(j)L1(j
.
)) h
kn
(1(j)1(). (1
0
(j)L1(j
.
))1()L1(j)1
0
())
with the map induced by
(1(j ). 1
0
(j ) L 1(j
.
.
))
(1(j) 1(). (1
0
(j) L 1(j
.
) 1() L 1(j) 1
0
()).
a kind of diagonal, on each bre given by (b. . n) ((b. ). (b. n)). This is the previously
dened map in the case that dimj = 0. The product t (j) t () is a Thom class and also
its restriction t (j ) to the diagonal. Using this Thom class one has the transitivity of the
Thom homomorphism (j)() = (j ).
4. Given i : X Y , r : Y X such that ri = id (a retract). Let : 1 X be a bundle
over X and r
= j the induced bundle. Let t () be a Thom class and t (j) its pullback. If
(j) is an isomorphism, then () is an isomorphism.
5. Let C
m
S
1
be the cyclic subgroup of m-th roots of unity. A model for the canonical
map
m
: BC
m
BS
1
is the sphere bundle of the m-fold tensor product j
m
= j j
of the canonical (universal) complex line bundle over BS
1
.
6. Let 1 be a commutative ring. Then H
(BS
1
: 1) 1c| where c is the Euler class of j.
7. We use coefcients in the ring 1. The Gysin sequence of
m
: BC
m
BS
1
splits into
short exact sequences
0 H
2k-1
(BC
m
) H
2k-2
(BS
1
)
mc
H
2k
(BS
1
)
)
m
H
2k
(BC
m
) 0.
This implies H
2k
(BC
m
) 1,m1, H
2k-1
(BC
m
)
m
1 for k > 0, where
m
1 is the
m-torsion (. 1 [ m. = 0) of 1. In even dimensions we have the multiplicative isomor-
phism H
2
(BS
1
),(mc) H
2
(BC
m
) induced by
m
.
8. The sphere bundle of the canonical bundle 1C
m

C
m
C BC
m
has a contractible
total space. Therefore the Gysin sequence of this bundle shows that the cup product
t : H
(BC
m
: 1) H
2
(BC
m
: 1) is an isomorphism for > 0; here t =
m
c.
9. The cup product
H
1
(BC
m
: 1) H
1
(BC
m
: 1) H
2
(BC
m
: 1)
is the 1-bilinear form
m
1
m
1 1,m1. (u. ) m(m 1),2 u.
Here one has to take the product of u.
m
1 1 and reduce it modulo m. Thus if m is
odd, this product is zero; and if m is even it is (u. ) m,2 u.
Chapter 18
Duality
We have already given an introduction to duality theory from the view point of
homotopy theory. In this chapter we present the more classical duality theory based
on product structures in homology and cohomology. Since we did not introduce
products for spectral homology and cohomology we will not directly relate the two
approaches of duality in this book.
Duality theory has several aspects. There is, rstly, the classical Poincar duality
theorem. It states that for a closed orientable n-dimensional manifold the groups
H
k
(M) and H
n-k
(M) are isomorphic. A consequence is that the cup product
pairing H
+
(M) H
n-+
(M) H
n
(M) is a regular bilinear form (say with
eld coefcients). This quadratic structure of a manifold is a basic ingredient in the
classication theory (surgery theory). The cup product pairing for a manifold has in
the context of homology an interpretation as intersection. Therefore the bilinear cup
product form is called the intersection form. In the case of a triangulated manifold
there exists the so-called dual cell decomposition, and the simplicial chain complex
is isomorphic to the cellular cochain complex of dual cells; this is a very strong
form of a combinatorial duality theorem [167].
The second aspect relates the cohomology of a closed subset 1 R
n
with the
homology of the complement R
n
1 (Alexander duality). This type of duality is
in fact a phenomenon of stable homotopy theory as we have explained earlier.
Both types of duality are related. In this chapter we prove in the axiomatic con-
text of generalized cohomology theories a theoremwhich compares the cohomology
of pairs (1. 1) of compact subsets of an oriented manifold M with the homology
of the dual pair (M 1. M 1). The duality isomorphism is constructed with the
cap product by the fundamental class. We construct the cap product for singular
theory.
18.1 The Cap Product
The cap product relates singular homology and cohomology with coefcients in
the ring 1. Let M and N be left 1-modules. The cap product consists of a family
of 1-linear maps
H
k
(X. : M) H
n
(X. LT: N) H
n-k
(X. T: M
T
N). . , . :,
and is dened for excisive pairs (. T) in X. (Compare the denition of the cup
product for singular cohomology.) If a linear map z: M N 1 is given,
18.1. The Cap Product 439
we compose with the induced map; then . : , H
n-k
(X. T: 1). This device is
typically applied in the cases M = 1and z: 1N N is an 1-module structure,
or M = N = is an 1-algebra and z: is the multiplication.
We rst dene a cap product for chains and cochains
S
k
(X: M) S
n
(X: N) S
n-k
(X: M N). c :c.
Given S
]
(X: M) and o : ^
]q
= e
0
. . . . . e
]q
| X, we set
:(o b) = (1)
]q
((o[e
q
. . . . . e
]q
| b)o[e
0
. . . . . e
q
|
and extend linearly. (Compare in this context the denition of the cup product.)
From this denition one veries the following properties.
(1) Let : X Y be continuous. Then
#
(
#
:c) = :
#
c.
(2) d( :c) = :c (1)
[[
:dc.
(3) ( [) :c = :([ :c).
(4) 1 :c = c.
Case (3) needs conventions about the coefcients. It can be applied in the case that
S
]
(X: 1). [ S
q
(X: ) and c S
n
(X: ) for an 1-algebra . In case
(4) we assume that 1 S
0
(X: 1) is the cocycle which send a 0-simplex to 1 and
c S
n
(X: N) for an 1-module N.
We now extend the denition to relative groups. If S
]
(X. : M)
S
]
(X: M) and c S
]q
(: N) S
]q
(T: N), then : c S
q
(T: M N).
Thus we have an induced cap product
S
]
(X. : M)
S
]q
(X: N)
S
]q
(: N) S
]q
(T: N)
S
q
(X: M N)
S
q
(T: M N)
.
Let . T be excisive. We use the chain equivalence S
-
() S
-
(T) S
-
(LT).
After passing to cohomology we obtain the cap product as stated in the beginning.
We list the
18.1.1 Properties of the cap product.
(1) For : (X: . T) (X
t
:
t
. T
t
), .
t
H
]
(X
t
.
t
: M), and for u
H
]q
(X. L T: N) the relation
+
(
+
.
t
:u) = .
t
:
+
u holds.
(2) Let . T be excisive,
B
: (T. T) (X. L T) the inclusion and
d
B
: H
]q
(X. L T)
J
H
]q-1
( L T. )

H
]q-1
(T. T).
Then for . H
]
(X. : M), , H
]q
(X. L T: N),
+
B
. :d
B
, = (1)
]
d(. ,) H
q-1
(T: M N).
440 Chapter 18. Duality
(3) Let . T be excisive,
: (. T) (X. T) the inclusion and

d
: H
]q
(X. L T)
J
H
]q-1
( L T. T)

H
]q-1
(. T).
Then for . H
]
(: M), , H
]q
(X. L T: N),
+
(. :d
,) = (1)
]1
. :, H
q-1
(X. T: M N).
(4) 1 :. = ., 1 H
0
(X), . H
n
(X. T).
(5) (. ,) : : = . : (, : :) H
n-]-q
(X. C: ) for . H
+
(X. : 1),
, H
+
(X. T: ). : H
+
(X. L T L C: ).
(6) Let c: H
0
(X: M N) M N denote the augmentation. For .
H
]
(X. : M), , H
]
(X. : N),
c(. :,) = (.. ,)
where (. ) is the Kronecker pairing.
We display again the properties in a table and refer to the detailed description
above.
+
(
+
.
t
:u) = .
t
:
+
u
+
B
. :d
B
, = (1)
[x[
d(. :,)
(
)
+
(. :d
,) = (1)
[x[1
. :,
1 :. = .
(. ,) :: = . :(, ::)
c(. :,) = (.. ,)
We use the algebra of the cap product and deduce the homological Thom iso-
morphism from the cohomological one.
(18.1.2) Theorem. Let : 1 T be an oriented n-dimensional real vector bundle
with Thom class t H
n
(1. 1
0
: Z). Then
t :: H
nk
(1. 1
0
: N) H
k
(1: N)
is an isomorphism.
Proof. Let : S
n
(1. 1
0
) be a cocycle which represents t . Then the family
S
nk
(1. 1
0
: N) S
k
(1: N). . : :.
18.2. Duality Pairings 441
is a chain map of degree n. This chain map is obtained from the corresponding
one for N = Z by taking the tensor product with N. It sufces to show that
the integral chain map induces an isomorphism of homology groups, and for this
purpose it sufces to show that for coefcients in a eld N = F an isomorphism is
induced (see (11.9.7)). The diagram
H
k
(1: F)
.t

H
kn
(1. 1
0
: F)

Hom(H
k
(1: F). F)
(tt)
Hom(H
kn
(1. 1
0
: F). F)
is commutative (by property (6) in 18.1.1) where is the isomorphism of the
universal coefcient theorem. Since t is an isomorphism, we conclude that t :is
an isomorphism.
Problems
1. The cap product for an excisive pair (. T) in X is induced by the following chain map
(coefcient group Z):
S
+
(X. ) S
+
(X. L T)
1t
S
+
(X. ) S
+
(X),(S
+
() S
+
(T))
11
S
+
(X. ) S
+
(X. T) S
+
(X. )
1:
S
+
(X. ) S
+
(X. ) S
+
(X. T)
s
ZS
+
(X. T) S
+
(X. T).
D is an approximation of the diagonal, t the graded interchange map, and c the evaluation.
The explicit form above is obtained from the AlexanderWhitney map D.
2. (. ,) :(a b) = (1)
|,||u|
(. :a) (, :b).
3. From the cap product one obtains the slant product . u .\u which makes the
following diagram commutative:
H
q
(X. ) H
n
((X. ) (Y. T))
\

pr
H
n-q
(Y. T)
H
q
((X. ) Y ) H
n
((X. ) (Y. T))
r

H
n-q
(X (Y. T)).
pr
The properties (1)(5) of the cap product can be translated into properties of the slant product,
and the cap product can be deduced from the slant product. (This is analogous to the - and
-product.)
18.2 Duality Pairings
We use the properties of the cap product in an axiomatic context. Let h
+
be a
cohomology theory and k
+
. h
+
homology theories with values in 1- MOD. A
duality pairing (a cap product) between these theories consists of a family of linear
maps
h
]
(X. ) k
]q
(X. L T) h
q
(X. T). . , . :,
dened for pairs (. T) which are excisive for the theories involved. They have the
(1) Naturality. For : (X: . T) (X
t
:
t
. T
t
) the relation
+
(
+
.
t
: u) =
.
t
:
+
u holds.
(2) Stability. Let . T be excisive. Dene the mappings
B
and d
B
as in (18.1.1).
Then
+
B
. :d
B
, = (1)
[x[
d(. :,).
(3) Stability. Let . T be excisive. Dene the mappings
and d
as in (18.1.1).
Then (
)
+
(. :d
,) = (1)
[x[1
. :,.
(4) Unit element. There is given a unit element 1 k
0
(1). The homomorphism
h
k
(1) h
-k
(1), . . :1 is assumed to be an isomorphism (1 a point).
(In the following investigations we deal for simplicity of notation only with the
case h
+
= k
+
.) Note that we do not assume given a multiplicative structure for the
cohomology and homology theories.
As a rst consequence of the axioms we state the compatibility of the cap product
with the suspension isomorphisms.
(18.2.1) Proposition. The following diagrams are commutative:
h
]
(X. ) h
]q
(X. L T)
t

pr
h
q
(X. T)
(-1)
)
c
h
]
(1X. 1) h
]q1
(1X. d1X L 1 L 1T)
t

h
q1
(1X. d1X L 1T),
h
]
(X. ) h
]q
(X. L T)
t

cc
h
q
(X. T)
h
]1
(1X. d1X L 1) h
]q1
(1X. d1X L 1 L 1T)
t

h
q
(1X. 1T).
(-1)
)
pr
(For the second diagram one should recall our conventions about the suspension
isomorphisms, they were different for homology and cohomology. Again we use
notations like 1X = 1 X.)
Proof. We consider the rst diagram in the case that = 0. The proof is based on
18.2. Duality Pairings 443
the next diagram.
h
]
(1X) h
]q
(1X. 1T)
t

h
q
(1X. 1T)
h
]
(d1X L 1T) h
]q
(d1X L 1T. 1T)
t

}
}
-1
h
q
(d1X L 1T. 1T)
}
-1
h
]
(d1X L 1T) h
]q
(d1X L 1T. 1T)
t

1i
h
q
(d1X L 1T)
i
h
]
(1X) h
]q1
(1X. d1X L 1T)
t

k
h
q1
(1X. d1X L 1T)
J
The maps i , , k are inclusions. The rst and the second square commute by
naturality. The third square is (1)
]
-commutative by stability (2).
n
h
n
(R
n
. R
n
0) be obtained from 1 h
0
(1) under
an iterated suspension isomorphism. Then
h
k
(R
n
) h
n-k
(R
n
. R
n
0). . . :e
n
is for k Z and n _ 1 an isomorphism.
Proof. This follows by induction on n. One uses the rst diagram in (18.2.1) and
an analogous suspension isomorphism with (R. R 0) in place of (1. d1).
(18.2.3) Proposition. Let (U. V ) and (U
t
. V
t
) be pairs of open subsets in the space
X = U LU
t
. Let h
n
(U LU
t
. V LV
t
) be a xed element. From it we produce
elements and via
h
n
(X. V L V
t
) h
n
(X. V L U
t
)

h
n
(U. V L UU
t
) .
h
n
(X. V L V
t
) h
n
(X. U L V
t
)

h
n
(U
t
. V
t
L UU
t
) .
Then the diagram
h
k-1
(U. V )

t
h
k-1
(UU
t
. V U
t
)

h
k
(U
t
. UU
t
)
t
h
n-k1
(U. UU
t
)
J
h
n-k1
(UU
t
. UV
t
)

h
n-k
(U
t
. V
t
)
is commutative. (We again have used notations like UU
t
= U U
t
.)
Proof. We use naturality and stability (2) and show that the down-right path of the
diagram is (1)
k-1
times the map
h
k-1
(U. V ) h
k-1
(UU
t
. V U
t
)
t
1
h
n-k
(UU
t
. UV
t
) h
n-k
(U
t
. V
t
)
and the right-down path (1)
k
times the analogous map where
1
is replaced by
1
; the element
1
is obtained from via the morphism
h
n
(U. V L UU
t
)
J
h
n-1
(V L UU
t
. V )

h
n-1
(UU
t
. V U
t
)
h
n-1
(UU
t
. (UV
t
) L (V U
t
))
1
and
1
from via the analogous composition in which the primed and unprimed
spaces are interchanged. Thus it remains to show
1
=
1
. This is essentially
a consequence of the Hexagon Lemma. One of the outer paths in the hexagon is
given by the composition
h
n
(U L U
t
. V L V
t
) h
n
(U L U
t
. U L V
t
)
h
n
(U
t
L V. (V L U
t
)(U L V
t
))
J
h
n-1
((U L V
t
)(V L U
t
). V L V
t
)
and the other path is obtained by interchanging the primed and unprimed objects.
The center of the hexagon is h
n
(U L U
t
. (U L V
t
)(V L U
t
)). We then compose
the outer paths of the hexagon with the excision
h
n-1
(UU
t
. UV
t
L V U
t
) h
n-1
((U L V
t
)(V L U
t
). V L V
t
):
then is mapped along the paths to
1
and
1
, respectively; this follows from the
original denition of the elements by a little rewriting. The displayed morphism
yields
1
.
18.3 The Duality Theorem
For the statement of the duality theorem we need two ingredients: A homological
orientation of a manifold and a duality homomorphism. We begin with the former.
Let M be an n-dimensional manifold. For 1 1 M we write
r
1
1
: h
+
(M. M 1) h
+
(M. M 1)
for the homomorphism induced by the inclusion, and r
1
x
in the case that 1 = {.].
An element o
1
h
n
(M. M 1) is said to be a homological orientation along 1
if for each , 1 and each chart : U R
n
centered at , the image of o
1
under
h
n
(M. M 1)
i
1
,
h
n
(M. M ,) h
n
(U. U ,)

h
n
(R
n
. R
n
0)
18.3. The Duality Theorem 445
is e
n
where e
n
is the element which arises from 1 h
0
under suspension. A
family (o
1
[ 1 M compact) is called coherent if for each compact pair 1 1
the restriction relation r
1
1
o
1
= o
1
holds. A coherent family (o
1
) of orientations is
a (homological ) orientation of M. If M is compact, then 1 = M is allowed and
an orientation is determined by the element o
h
n
(M), called the fundamental
class of M.
In order to state the duality theorem we need the denition of a duality ho-
momorphism. We x a homological orientation (o
1
) of M. Given closed sets
1 1 M and open sets V U M such that 1 V. 1 U. We x an
element : h
n
(M. M 1). From o
1
= : we obtain :
UV
11
via
: h
n
(M. M 1)

h
n
(M. (M 1) L V )
:
UV
11
h
n
(U 1. (U 1) L (V 1)).
(#)
The morphism (#) is an excision, since M (U 1) = (M U) L1 (closed) is

contained in (M 1) L V (open).
From :
UV
11
we obtain the homomorphism D
UV
11
via the commutative diagram
h
k
(U. V )

T
I1
!1
h
k
(U 1. V 1)
tz
I1
!1
h
n-k
(M 1. M 1) h
n-k
(U 1. U 1).
We state some naturality properties of these data. They are easy consequences
of the naturality of the cap product.
(18.3.1) Lemma. Let (1. 1) (U
t
. V
t
) (U. V ) and
i : (U
t
1. U
t
1. V
t
1) (U 1. U 1. V 1).
Then i
+
:
U
/
V
/
11
= :
UV
11
and D
UV
11
= D
U
/
V
/
11
i
+
.
(18.3.2) Lemma. Let (1
t
. 1
t
) (1. 1) (U. V ) and
: (U 1. U 1. V 1) (U 1
t
. U 1
t
. V 1
t
).
Then
+
:
UV
11
= :
UV
1
/
1
/
and
+
D
UV
1
/
1
/
= D
UV
11
.
The naturality (18.3.1) allows us to pass to the colimit over the neighbourhoods
(U. V ) of (1. 1) in M. We obtain a duality homomorphism
D
11
:
`
h
k
(1. 1) = colim
UV
h
k
(U. V ) h
n-k
(M 1. M 1).
We explain this with some remarks about colimits. An element . h
k
(U. V ) rep-
resents an element of the colimit. Two elements . h
k
(U. V ) and .
t
h
k
(U
t
. V
t
)
represent the same element if and only if there exists a neighbourhood (U
tt
. V
tt
)
with U
tt
U U
t
, V
tt
V V
t
such that . and .
t
have the same restric-
tion in h
k
(U
tt
. V
tt
). Thus we have canonical homomorphisms l
UV
: h
k
(U. V )
`
h
k
(1. 1). Via these homomorphisms the colimit is characterized by a univer-
sal property: If z
UV
: h
k
(U. V ) h is a family of homomorphisms such that
z
U
/
V
/ i = z
UV
for the restrictions i : h
k
(U. V ) h
k
(U
t
. V
t
), then there ex-
ists a unique homomorphism z:
`
h
k
(1. 1) h such that zl
UV
= z
UV
. The
restrictions h
k
(U. V ) h
k
(1. 1) are compatible in this sense, and we obtain a
canonical homomorphism
`
h
k
(1. 1) h
k
(1. 1). In sufciently regular situations
this homomorphism is an isomorphism; we explain this later.
(18.3.3) Duality Theorem. Let M be an oriented manifold. Then the duality
homomorphism D
11
is, for each compact pair (1. 1) in M, an isomorphism.
We postpone the proof and discuss some of its applications. Let M be compact
and M| h
n
(M) a fundamental class. In the case (1. 1) = (M. 0) we have
`
h
k
(M. 0) = h
k
(M) and D
11
is the cap product with M|. Thus we obtain as a
special case:
(18.3.4) Poincar DualityTheorem. Suppose the compact n-manifoldis oriented
by the fundamental class M| h
n
(M). Then
h
k
(M) h
n-k
(M). . . :M|
is an isomorphism.
A duality pairing exists for the singular theory
H
]
(X. : G) H
]q
(X. L T: 1) H
q
(X. T: G)
for commutative rings 1 and 1-modules G. The Euclidean space R
n
is orientable
for H
+
(: Z). Thus we have:
(18.3.5) Alexander Duality Theorem. For a compact pair (1. 1) in R
n
`
H
k
(1. 1: G) H
n-k
(R
n
1. R
n
1: G).
A similar isomorphism exists for S
n
in place of R
n
.
(18.3.6) Example. We generalize the Jordan separation theorem. Let M be a
connected and orientable (with respect to H
+
(: Z)) n-manifold. Suppose that
H
1
(M: 1) = 0. Let M be compact, ,= M. Then
`
H
n-1
(: 1) is a free
1-module, and [
0
(M )[ = 1 rank
`
H
n-1
(: 1).
18.4. Euclidean Neighbourhood Retracts 447
By duality
`
H
n-1
(: 1) H
1
(M. M : 1). The hypothesis shows
d: H
1
(M. M : 1)

H
0
(M : 1).
and the latter is a free 1-module of rank [
0
(M )[ 1.
(18.3.7) Example. H
2
(R1
2
: Z) Z,2. This is not a free Z-module. Hence the
projective plane cannot be embedded into S
3
. (A similar proof shows that R1
2n
has no embedding into S
2n1
.)
(18.3.8) Remark. From Alexander duality and the Thom isomorphism one can
deduce Poincar duality. Let M R
nt
be a smooth closed submanifold of
dimension n. Suppose we have an Alexander duality isomorphism h
k
(M)
h
nt-k
(R
nt
. R
nt
M). Let t : 1(v) U be a tubular map. We use t and
excision and obtain h
nt-k
(R
nt
. R
nt
M) h
nt-k
(1(v). 1
0
(v)). Sup-
pose the normal bundle v is oriented by a Thom class. Then we have a Thom-
isomorphism h
nt-k
(1(v). 1
0
(v)) h
n-k
(M). Altogether we obtain an iso-
morphism h
k
(M) h
n-k
(M) of Poincar duality type. A similar device works
if we start from an isomorphism h
k
(M) h
nk-t
(R
nt
. R
nt
M) and use a
homological Thom isomorphism. This approach would be used if one starts with
homotopical duality as a foundation stone.
Problems
1. Let D R
2
be connected and open. The following are equivalent: (1) D is homeomor-
phic to R
2
. (2) D is simply connected. (3) H
1
(D: Z) = 0. (4) H
1
(D: Z) = 0. (5) R
2
D
is connected. (6) The boundary of D is connected. (7) If J D is a Jordan curve, then D
contains the interior of J. [44, p. 394 ]
2. Let i : S
n
1(Z. n) be an inclusion of a subcomplex which induces an isomorphism
of
n
. For each compact subset 1 R
n1
the induced map i
: 1. S
n
| 1. 1(Z. n)|
is bijective.
3. Use cohomology H
n
(X: Z) = X. 1(Z. n)| dened with an EilenbergMac Lane com-
plex 1(Z. n). Then for a compact subset X in a Euclidean space
`
H
n
(X: Z) H
n
(X: Z).
Similar isomorphisms hold for the stable cohomotopy groups.
4. R
n
is orientable for each homology theory. Let 1 D(r) = {. [ [.[ _ r] be compact.
Dene o
!
as the image of the canonical class under h
n
(R
n
. R
n
0) h
n
(R
n
. R
n
D(r))
h
n
(R
n
. R
n
1).
18.4 Euclidean Neighbourhood Retracts
For applications it is interesting to compare
`
H
k
with H
k
.
Aspace X is called a Euclideanneighbourhood retract (ENR) if there exists an
embedding : X R
n
, an open neighbourhood U of (X) in R
n
and a retraction
r : U X, i.e., r = id(X).
Let X R
n
be a retract of an open set U, then X is closed in U and hence
locally compact: Let r : U X be a retraction; then X is the coincidence set of
r : U U and id: U U. Recall that a locally compact set Y in a Hausdorff
space 7 is locally closed, i.e., has the form Y =

Y W for an open set W in 7.
n
be a retract of an open neighbourhood. Let 7
be a metric space and Y 7 homeomorphic to X. Then Y is a retract of an open
neighbourhood V of Y in 7.
Proof. Let : X Y be a homeomorphism and r : U X a retraction. Then Y
is locally compact and we can write Y =

Y W with an open W 7. Then Y is
closed in W. Since W is a normal space, the map Y
(
-1
X R
n
has a continuous
extension h: W R
n
by the Tietze extension theorem. Let V = h
-1
(U). Then
rh: V Y is a retraction of Y V .
n
be locally compact. Then there exists an em-
bedding of X into R
n1
as a closed subset.
Proof. Let U R
n
be open (U ,= R
n
). Then
: U R
n
R. . (.. J(.. R
n
U)
-1
)
is an embedding. The image of is closed , since
(U) = {(.. t ) [ t J(.. R
n
U) = 1].
We can assume that X U is closed; then (X) is closed in (U), hence closed
in R
n1
.
(18.4.3) Proposition. Let X be an ENR. Suppose
0
.
1
: Y X are maps which
coincide on a subset T Y . Then there exists a neighbourhood W of T in Y and
a homotopy h:
0
[W .
1
[W relative to T.
Proof. Let X
i
U
i
X be a presentation as a retract with U R
n
open. Let
W = {, [ (1 t )i
0
(,) t i
1
(,) U for all t 1].
Then certainly T W. Since
z: Y 1 R
n
. (,. t ) (1 t )i
0
(,) t i
1
(,)
is continuous, z
-1
(U) is open. If {,] 1 z
-1
(U), then there exists an open
neighbourhood U
,
of , such that U
,
1 z
-1
(U). Hence W is open in Y . A
suitable homotopy h is now obtained as the restriction of z to W 1.
18.4. Euclidean Neighbourhood Retracts 449
(18.4.4) Remark. Suppose T X are Euclidean neighbourhood retracts. Then
there exists a retraction r : V T from an open neighbourhood V of T in X.
There further exists a neighbourhood W V of T in X such that : W V and
i (r[W): W T V are homotopic relative to T.
An ENR is locally contractible: Each neighbourhood V of . contains a neigh-
bourhood W of . such that W V is homotopic to W {.] V relative
to {.].
(18.4.5) Lemma. Let the Hausdorff space X = X
1
L LX
i
be a union of locally
compact open subsets X
}
which are homeomorphic to a subset of a Euclidean space.
Then X is homeomorphic to a closed subset of a Euclidean space.
Proof. There exist embeddings h
i
: X
i
R
n(i)
as a closed subset. We extend h
i
to a continuous map k
i
: X S
n(i)
= R
n(i)
L {o] by k
i
(X X
i
) = {o] (if
X ,= X
i
). The product (k
i
): X

i
i=1
S
n(i)
is an embedding. The product of
the spheres can be embedded into

i
R
n(i)1
, and then we can apply (18.4.2) if
necessary.
(18.4.6) Theorem. Let the Hausdorff space X = X
1
L LX
i
be a union of open
subsets X
i
which are ENRs. Then X is an ENR.
Proof. Induction on r. It sufces to consider X = X
0
L X
1
. By (18.4.5) we can
assume that X is a closed subset of some R
n
. Let r
i
: U
i
X
i
be retractions (see
(18.4.1)). Set
U
01
= r
-1
0
(X
0
X
1
) r
-1
1
(X
0
X
1
).
Then r
0
. r
1
: U
01
X
0
X
1
are retractions of a neighbourhood. The open subset
X
0
X
1
of the ENR X
0
is an ENR. Hence there exists X
0
X
1
V
01
U
01
such that r
0
. r
1
are homotopic on V
01
relative to X
0
X
1
by a homotopy r
t
(see
(18.4.3)). Let V
0
U
0
. V
1
U
1
be open neighbourhoods of X X
1
. X X
0
such that

V
0

V
1
= 0. Choose a continuous function t : R
n
0. 1| such that
t(V
0
) = 0 and t(V
1
) = 1. Let V = V
0
L V
01
L V
1
. Then j: V X, dened as
j[V
0
= r
0
[V
0
, j[V
1
= r
1
[V
1
, j(.) = r
r(x)
(.) for . V
01
is a suitable retraction.
(18.4.7) Corollary. A compact manifold is an ENR.

(18.4.8) Remark. Since an ENR is dominated by a CW-complex it has the homo-
topy type of a CW-complex. Acompact ENRis dominated by a nite CW-complex;
therefore its singular homology groups with coefcients in Zare nitely generated
abelian groups. This holds in particular for compact manifolds.
(18.4.9) Proposition. Let 1 be a compact ENR in an n-manifold M. Then the
canonical map j:
`
H
k
(1) H
k
(1) is an isomorphism.
Proof. We use that M is a metrizable space or, at least, that open subsets are normal.
Then 1 is a retract r : U 1 of an open neighbourhood U of 1 in M. Suppose
. H
k
(1); then r
+
(.) represents an element ` .
`
H
k
(1) with j( ` .) = .. This
shows that j is surjective. Let .
U
H
k
(U) represent an element in the kernel
of j.
Suppose there exists a neighbourhood V U of 1 and a homotopy from
: V U to i r : V 1 U. Then we have the situation
.
U
_
H
k
(U)
i

P
P
P
P
P
P
}
H
k
(1)
i

H
k
(V ).
.
V H
k
(V )
n
n
n
n
n
n
Since i
+
(.
U
) = 0, by assumption, we see that .
V
= 0; but .
V
represents the same
element as .
U
.
The homotopy exists by (18.4.3) if also U is an ENR. If we choose U as the
union of a nite number of sets which are homeomorphic to open subsets of R
n
,
then we can apply (18.4.6).
Let X be an ENR and X R
n
. Since the canonical homomorphism
`
h
+
(X)
h
+
(X) is an isomorphism,
`
h
+
(X) does not depend on the embedding X R
n
.
Let M be a compact n-manifold. Then H
k
(M: Z) and H
k
(M: Z) are nitely
generated abelian groups. For each eld F the groups H
k
(M: F) are nite-dimen-
sional vector spaces. The Euler characteristic (M: F) is independent of F and
equal to the Euler characteristic (M).
Let 1 M be a compact ENR. Then
H
n-i
(1: F
2
)
`
H
n-i
(1: F
2
) H
i
(M. M 1: F
2
)
and these are nite-dimensional vector spaces over the prime eld F
2
.
(18.4.10) Proposition. H
+
(1: F
2
) is nite-dimensional if and only if the same holds
for H
+
(M 1: F
2
). If niteness holds, then for the Euler characteristic
2
the
relation
2
(M) =
2
(M 1) (1)
n
2
(1)
holds. (Note that
2
(1) = (1).)
Proof. The rst statement follows from the exact homology sequence of the pair
(M. M 1). It also yields
2
(M) =
2
(M 1)
2
(M. M 1). The
equality
2
(M. M 1) = (1)
n
2
(1) is obtained, if we insert the consequence
dimH
i
(M. M 1: F
2
) = dimH
n-i
(1: F
2
) of the duality into the homological
denition of the Euler characteristic.
18.5. Proof of the Duality Theorem 451
(18.4.11) Corollary. Let M be a closed manifold of odd dimension. Then
(M) = 0. If 1 M is a compact ENR, then (1) = (M 1).
Problems
1. Let J be a compact, connected, non-orientable surface. The universal coefcient formula
and H
1
(J: Z) Z
-1
Z,2 showH
2
(J: Z) Z,2. Therefore J cannot be embedded
into S
3
.
2. Let S R
2
be the pseudo-circle. Show, heuristically, that the pseudo-circle has a system
of neighbourhoods U
1
U
2
with U
i
S
1
0. 1| and U
n
= S.
Then
`
H
1
(S: Z) H
1
(R
2
. R
2
S: Z)

H
0
(R S: Z) Z.
the last isomorphism because R
2
S has two path components. By the universal coefcient
formula H
1
(S: Z) Hom(H
1
(S). Z). The singular homology group H
1
(S: Z) is zero,
a singular 1-chain is always contained in a contractible subset. In fact, S has the weak
homotopy type of a point. This shows that singular theory is the wrong one for spaces like
S.
3. Let J R
3
be a connected orientable compact surface. Then R
3
J has two path
components (interior and exterior).
4. Let M = R
n1
and S R
n1
be homeomorphic to S
n
. Then S is an ENR and
`
H
n
(S: Z) H
n
(S: Z) Z. From the duality theorem one obtains that R
n1
S has
two path components.
18.5 Proof of the Duality Theorem
We have to collect some formal properties of the groups
`
h and of the duality ho-
momorphisms D. We want the
`
h
k
(1. 1) to be part of functors from the category
K(M) of compact pairs in M and inclusions.
Let (1
t
. 1
t
) (1. 1). The induced map
`
h
k
(1. 1)
`
h
k
(1
t
. 1
t
) sends
an element represented by . h
k
(U. V ) to the element which is represented
by the same .. This makes the
`
h
k
into functors on K(M) and the canoni-
cal maps
`
h
k
h
k
into natural transformations. We dene a coboundary op-
erator
`
:
`
h
k
(1)
`
h
k1
(1. 1) as follows. Let V 1 be open. Choose
U V as an open neighbourhood of 1. Then we map representing elements
via : h
k
(V ) h
k1
(U. V ). This process yields a well-dened
`
and the diagram
`
h
k
(1)
`
h
k1
(1. 1)
h
k
(1)

h
k1
(1. 1)
is commutative. From (18.3.2) we obtain by passage to the colimit:
(18.5.1) Lemma. The D
11
yield a natural transformation, i.e., the diagram
`
h
k
(1. 1)
T
!1
h
n-k
(M 1. M 1)
`
h
k
(1
t
. 1
t
)
T
!
/
1
/
h
n-k
(M 1
t
. M 1
t
)
is commutative for each inclusion (1
t
. 1
t
) (1. 1).
(18.5.2) Lemma. The sequence

`
h
k
(1)

`
h
k
(1)
`
h
k1
(1. 1)
is exact. Similarly for triples of compact subsets.
Proof. This is a special case of the general fact that a colimit over a directed set of
exact sequences is again exact. A direct verication from the denitions and the
exact sequences for the representing elements is not difcult. An example should
sufce. Suppose
`
(.) = 0 and let . be represented by .
1
h
k
(V ). We use
the representing element (.
1
) h
k1
(U. V ) for
`
(.). Since
`
(.) is zero, .
1
is
contained in the kernel of some restriction h
k1
(U. V ) h
k1
(U
t
. V
t
). Another
representative of . is the restriction .
2
h
k
(V
t
) of .
1
. By exactness, .
2
has a
pre-image in h
k
(U
t
), and it represents a pre-image of . in
`
h
k
(1).
(18.5.3) Lemma. Each compact pair 1. 1 is excisive for
`
h
k
, i.e.,
`
h
k
(1 L 1. 1)
`
h
k
(1. 1 1)
is an isomorphism.
Proof. This is a consequence of the isomorphisms h
k
(U LV. U) h
k
(V. U V )
for open neighbourhoods U 1, V 1.
(18.5.4) Corollary. For each compact pair there exist an exact MV-sequence

`
h
k
(1 L 1)
`
h
k
(1)
`
h
k
(1)
`
h
k
(1 1)

.
Proof. The MV-sequence is constructed by algebra from suitable exact sequences
using (18.5.2) and (18.5.3).
(18.5.5) Lemma. For each pair (1. 1) the diagram
`
h
k
(1)
T
1
h
n-k
(M. M 1)
J
`
h
k1
(1. 1)
T
!1
h
n-k-1
(M 1. M 1)
is commutative.
18.5. Proof of the Duality Theorem 453
Proof. By passage to the colimit this is a consequence of the commutativity of the
diagrams
h
k-1
(V )

T
1
1
h
k
(U. V )
T
I1
!1
h
n-k1
(M. M 1)
J
h
n-k
(M 1. M 1).
In order to verify this commutativity we apply (18.2.3) to the sets
(U. V. U
t
. V
t
) = (V. 0. U 1. U 1).
The element arises from o
1
via the excision h
n
(U. U 1) h
n
(M. M 1).
One now veries from the denitions that the element is :
V 0
10
and becomes
:
UV
11
. These data yield the commutative diagram
h
k-1
(V )
=
h
k
(U. V )
h
k-1
(V )

tz
1
1
h
k-1
(V 1)

h
k
(U 1. V 1)
tz
I1
!1
h
n-k1
(V. V 1)
J
h
n-k
(V 1. V 1)

h
n-k
(U 1. U 1)
h
n-k1
(M. M 1)
J
h
n-k
(M 1. M 1).
The upper and lower rectangles commute by naturality of d and .
For the proof of the duality theoremwe note that it sufces to consider the special
case D
1
:
`
h
k
(1) h
n-k
(M. M1), by the Five Lemma and the previous results.
The proof is based on the following principle.
(18.5.6) Theorem. Let D(1) be an assertion about compact subsets in M. Sup-
pose:
(1) D(1) holds for sets 1 in a chart domain which are mapped onto a convex
subset of R
n
under the coordinate map.
(2) If D(1). D(1), and D(1 1) hold, then also D(1 L 1) holds.
(3) Let 1
1
1
2
, 1 =
_
1
i
. If D(1
i
) holds for each i , then D(1)
holds.
Under these assumptions, D(1) holds for all compact 1.
Proof. Since an intersection of convex sets is convex, (1) and (2) yield by induction
on t that D(1
1
L L 1
t
) holds for compact subsets 1
i
of type (1) which are
contained in the same chart domain.
If 1 is a compact set in a chart domain, then 1 is the intersection of a sequence
1
1
1
2
where each 1
i
is a nite union of compact convex sets.
Each compact set 1 is a nite union of compact sets in chart domains. Again
D(1) follows by induction from (2).
We now verify (1)(3) of (18.5.6) in the case that D(1) is the assertion: D
1
is
an isomorphism.
(2) The duality homomorphisms D yield a morphism of the MV-sequence for
1. 1 into the MV-sequence of the complements; this follows from the fact that the
(co-)boundary operators of the MV-sequences are dened frominduced morphisms
and ordinary (co-)boundary operators. Now use the Five Lemma.
(1) Let : U R
n
be a chart, 1 U and (1) convex. We begin with the
special case of a point 1 and (1) = {0]. We have a commutative diagram
h
k
(R
n
)
(z
I
!
)
h
k
(U)

tz
I
!
`
h
k
(1)
T
!
h
n-k
(R
n
. R
n
0) h
n-k
(U. U 1)
h
n-k
(M. M 1).
The right square commutes by denition of D
1
; here :
U
1
is the image of the
orientation under the restriction h
n
(M. M 1) h
n
(U. U 1). The left
square commutes by naturality of the cap product. By denition of the orien-
tation,
+
(:
U
1
) = e
n
. The fact that :e
n
is an isomorphism follows from the
compatibility with suspension and the unit element axiom of the pairing. Hence
D(1) holds for a point 1.
Let now 1 be arbitrary and 1 1 a point. From naturality we see that
D(1) holds, if
`
h
k
(1)
`
h
k
(1) and h
n-k
(M. M 1) h
n-k
(M. M 1) are
isomorphisms.
The set X = (1), being compact convex, is the intersection of a sequence
of open neighbourhoods U
1
U
2
which are contractible onto 1, and each
neighbourhood of X contains eventually all U
}
. Hence the restriction h
k
(U
}
)
h
k
(U
}1
) are isomorphisms, and we see h
k
(U
}
)
`
h
k
(X) h
k
(X). This shows
the rst isomorphism. The second isomorphism is veried by standard methods
(excision, h-equivalence).
(3) We show that the canonical maps
colim
i
`
h
k
(1
i
)
`
h
k
(1). colim
i
h
n-k
(M. M 1
i
) h
n-k
(M. M 1)
are isomorphisms.
The rst isomorphism is an immediate consequence of the colim-denition.
Given .
`
h
k
(1) represented by , h
k
(U). There exists i such that 1
i
U.
Hence , represents an element in
`
h
k
(1
i
). This shows surjectivity, and injectivity
is shown by a similar argument.
The second isomorphism is easily seen for singular homology, if one uses that
singular chains have compact carrier. In the general case one uses that additive
homology theories commute with colimits.
We now treat duality for manifolds with boundary.
(18.6.1) Theorem. Let M be a compact n-manifold with boundary dM, oriented
by a fundamental class M| H
n
(M. dM: Z). Then
H
]
(M: G) H
n-]
(M. dM: G). . . :M|.
H
]
(M. dM: G) H
n-]
(M: G). . . :M|.
are isomorphisms for each coefcient group G.
Proof. By naturality and stability of the cap product the following diagram com-
mutes up to sign (coefcients are G); dM| = dM| is a fundamental class:
H
]
(M. dM)
}
tj
H
]
(M)
i

tj
H
]
(dM)

tJj
H
]1
(M. dM)
tj
H
n-]
(M)
}

H
n-]
(M. dM)
J
H
n-]-1
(dM)
i

H
n-]-1
(M).
We know already that :dM| is an isomorphism. Therefore it sufces to show that
the left-most vertical map is an isomorphism. We reduce the problem to the duality
already proved. We use the non-compact auxiliary manifold
1 = M L (dM 0. 1).
which is obtained by the identication . ~ (.. 0) for . dM. We also use the
subspaces
M(t ) = M L (dM 0. t |). 0 _ t < 1.
1(t ) = M L (dM 0. t ). 0 < t _ 1.
The 1(t ) are a conal system of open neighbourhoods of M = M(0) in 1. The
M(t ) are a compact exhaustion of 1. The inclusions M 1(t ) M(t ) are
h-equivalences. Let i(t ): (M. dM) (1(t ). 1(t ) M
). The diagram
H
]
(M)
tj
H
]
(1(t ))
i(t)
ti(t)
H
n-]
(M. dM)
i(t)

H
n-]
(1(t ). 1(t ) M
)
is commutative (naturality of the cap product). The map i(t )
+
is an isomorphism,
since i(t ) is an h-equivalence; the map i(t )
+
is an excision. Thus it sufces to show
that :i(t )
+
M| is an isomorphism. We use the duality theorem for 1. We have
isomorphisms
H
n
(1. 1 M(t )) H
n
(1. 1 M) H
n
(1. 1 M
) H
n
(M. dM).
Let :(t ) H
n
(1. 1 M(t )) correspond to the fundamental class M|. One
veries that :(t ) is an orientation along M(t ). Since the M(t ) form a com-
pact exhaustion, the coherent family of the :(t ) yields an orientation of 1. Let
n(t ) H
n
(1(t ). 1(t ) M) and (t ) H
n
(1(t ). 1(t ) M
) correspond to the
fundamental class under
H
n
(1(t ). 1(t ) M) H
n
(1(t ). 1(t ) M
) H
n
(M. dM).
By denition of the duality homomorphism, D
:
`
H
]
(M) H
n-]
(1. 1 M)
is the colimit of the maps
H
]
(1(t ))
tu(t)
H
n-]
(1(t ). 1(t ) M)

H
n-]
(1. 1 M).
Since H
]
(1(t )) H
]
(M) is an isomorphism, the canonical maps H
]
(1(t ))
`
H
]
(M) H
]
(M) are isomorphisms. Since D
is an isomorphism, so is :n(t ).
The diagram
H
]
(1(t ))
tu(t)
H
]
(1(t ))
i
t(t)
H
n-]
(1(t ). 1(t ) M)
i

H
n-]
(1(t ). 1(t ) M
)
is commutative by naturality of the cap product and (t ) = i
+
n(t ). Since (t ) =
i(t )
+
M|, the map :i(t )
+
M| is an isomorphism.
(18.6.2) Proposition. Let T be a compact (n 1)-manifold with boundary dT =

M. Then (M) = (1 (1)
n
)(T). In particular (M) is always even.
18.7. The Intersection Form. Signature 457
Proof. Let M = T L (dT 0. 1). Then T is a compact deformation retract
of M and M T dT|0. 1. dT. Hence (T) = (M) = (M T)
(1)
n1
(T) = (dT) (1)
n
(T).
(18.6.3) Example. R1
2n
is not the boundaryof a compact manifold, since (R1
2n
)
is odd. The same holds for an arbitrary nite product of even-dimensional real pro-
jective spaces.
Problems
1. Suppose dM = T is a decomposition into two closed submanifolds. The diagram

H
)
(M. )
r^|
H
)
(M)
r^|
H
)
()
ru^|

H
n-)
(M. T)

H
n-)
(M. dM)

H
n-)-1
(dM. )

is commutative up to sign. Hence :M| : H
)
(M. ) H
n-)
(M. T) is an isomorphism.
18.7 The Intersection Form. Signature
Let M be a closed n-manifold oriented by a fundamental class M| H
n
(M: K),
coefcients in a eld K. The evaluation on the fundamental class is H
n
(M)
K. . .M| = (.. M| ). We can also write this as the composition H
n
(M)
Hom(H
n
(M). K) Kwhere we evaluate a homomorphismonM|. The canonical
map c: H
0
(M) K allows us to write (.. M| ) = c(. M|).
(18.7.1) Proposition. The bilinear form
H
k
(M) H
n-k
(M)
.

H
n
(M)
( -,j )

K
is regular. We write this form also as (.. ,) . ,.
Proof. We use the rule (.,):M| = .:(,:M|). It gives us the commutative
diagram
H
k
(M) H
n-k
.

id tj
H
n
(M)
tj
H
k
(M) H
k
(M)
t

id
H
0
(M)
t
Hom(H
k
(M). K) H
k
(M)
eval
K.
The bilinear form in question is isomorphic to the Hom-evaluation, and the latter is
for each nite-dimensional vector space a regular form.
We take now R = K as coefcients and assume n = 4t . In that case
H
2t
(M) H
2t
(M) R. (.. ,) . ,
is a regular symmetric bilinear form. Recall from linear algebra: Let (V. ) be
a real vector space together with a symmetric bilinear form . Then V has a
decomposition V = V
V
-
V
0
such that the form is positive denite on V
,
negative denite on V
-
and zero on V
0
. By Sylvesters theorem the dimensions
of V
and V
-
are determined by . The integer dimV
dimV
-
is called the
signature of . We apply this to the intersection form and call
o(M) = dimH
2t
(M)
dimH
2t
(M)
-
the signature of the closed oriented 4t -manifold M. We also set o(M) = 0, if
the dimension of M is not divisible by 4. If M denotes the manifold with the
opposite orientation, then one has o(M) = o(M). If M = M
1
M
2
then
H
2t
(M) = H
2t
(M
1
)H
2t
(M
2
), the forms on M
1
and M
2
are orthogonal, hence
o(M
1
M
2
) = o(M
1
) o(M
2
).
(18.7.2) Proposition. The signature of C1
2n
with its natural orientation induced
by the complex structure is 1.
Proof. Since H
2n
(C1
2n
: Z)) is the free abelian group generated by c
n
, the claim
follows from (c
n
. C1
n
| ) = 1. For n = 1 this holds by the denition of the
rst Chern class (see (19.1.2)). Consider the map : (C1
1
)
n
C1
n
that sends
(a
1
. b
1
|. . . . . a
n
b
n
|) to c
0
. . . . . c
n
| where
n
}=1
(a
}
. b
}
,) =
n
i=0
c
}
.
}
,
n-}
.
Note that H
2
((C1
1
)
n
: Z) is the free abelian group with basis t
1
. . . . . t
n
where t
}
is the rst Chern class of pr
+
}
(j). One veries that
+
(c) = t
1
t
n
. This
implies
+
c
n
= n t
1
t
2
. . . t
n
. The map has degree n. These facts imply
n = (
+
c
n
. C1
1
|
n
) = (c
n
.
+
C1
1
|
n
) = (c
n
. n C1
n
| ) = n (c
n
. C1
n
| ).
hence (c
n
. C1
n
| ) = 1.
(18.7.3) Proposition. Let M and N be closed oriented manifolds. Give M N
the product orientation. Then o(M N) = o(M)o(N).
Proof. Let m = dimM and n = dimM. If m n , mod4 then o(M N) =
0 = o(M)o(N) by denition. In the case that m n = 4 we use the Knneth
isomorphism and consider the decomposition
H
2]
(M N) = H
n{2
(M) H
n{2
(N)
2i~n
_
H
i
(M) H
2]-i
(N) H
n-i
(M) H
n-2]i
(N)
_
.
The rst summand on the right-hand side is zero if m or n is odd. The form
on H
2]
(M N) is transformed via the Knneth isomorphism by the formula
18.7. The Intersection Form. Signature 459
(. ,) (.
t
,
t
) = (1)
[,[[x
/
[
(. .
t
)(, ,
t
). Products of elements in
different summands never contribute to the top dimension m n. Therefore the
signature to be computed is the sumof the signatures of the forms on the summands.
Consider the rst summand. If m,2 and n,2 are odd, then the form is zero. In
the other case let be a basis of H
n{2
(M) such that the formhas a diagonal matrix
with respect to this basis and let T be a basis of H
n{2
(N) with a similar property.
Then (a b [ a . b T) is a basis of H
n{2
(M) H
n{2
(N) for which the
form has a diagonal matrix. Then
o(M)o(N) =
_
o
a a
__
bB
b b
_
=
(o,b)B
(a b) (a b) = o(M N).
Now consider the summand for 2i < m. Choose bases of H
i
(M) and T of
H
2]-i
(N) and let
+
, T
+
be the dual bases of H
n-i
(M), H
n-2]i
(N) respec-
tively. Then
(a b a
+
b
+
. a b a
+
b
+
[ a . b T)
is a basis of the summand under consideration. The product of different basis
elements is zero, and (a b a
+
b
+
)
2
= (a b a
+
b
+
)
2
shows that
the number of positive squares equals the number of negative squares. Hence these
summands do not contribute to the signature.
There exists a version of the intersection form for cohomology with integral
coefcients. We begin again with the bilinear form
s : H
k
(M) H
n-k
(M) Z. (.. ,) (. ,)M|.
We denote by
^
the quotient of the abelian group by the subgroup of elements
of nite order. We obtain an induced bilinear form
s
^
: H
k
(M)
^
H
n-k
(M)
^
Z.
(18.7.4) Proposition. The form s
^
is regular, i.e., the adjoint homomorphism
H
k
(M)
^
Hom(H
n-k
(M)
^
. Z) is an isomorphism (and not just injective).
Proof. We use the fact that the evaluation H
k
(M: Z)
^
H
k
(M: Z)
^
Z is
a regular bilinear form over Z. By the universal coefcient formula, the kernel
of H
k
(M: Z) Hom(H
k
(M: Z). Z) is a nite abelian group; hence we have
isomorphisms
H
k
(M: Z)
^
Hom(H
k
(M: Z). Z) Hom(H
k
(M: Z)
^
. Z).
Now the proof is nished as before.
We return to eld coefcients. Let M be the oriented boundary of the compact
oriented manifold T. We set
k
= Im(i
+
: H
k
(T) H
k
(M)) with the inclusion
i : M T.
(18.7.5) Proposition. The kernel of
H
k
(M)

Hom(H
n-k
(M). K) Hom(
n-k
. K)
is
k
. The isomorphismis . (, (,.. M| )); the second map is the restric-
tionto
n-k
. Inparticular dimH
k
(M) = dim
k
dim
n-k
, andinthe case n =
2t we have dimH
t
(M) = 2 dim
k
and dimH
t
(M) = 2 dimKer(i
+
: H
t
(M)
H
t
(T)).
H
k
(T)
i

H
k
(M)

tj
H
k1
(T. M)
tBj
H
n-k
(M)
i

H
n-k
(M).
By stability of the cap product, the square commutes up to the sign (1)
k
. By
commutativity and duality
.
k
= (.) = 0 = (.) :T| = 0 = i
+
(. :M|) = 0.
The regularity of the pairing H
}
(M) H
}
(M) K, (.. ,) (.. ,) says that
i
+
(. : M|) = 0 is equivalent to (H
n-k
(T). i
+
(. : M|)) = 0. Properties of
pairings yield
(H
n-k
(T). i
+
(. :M| ) = (i
+
H
n-k
(T). . :M| )
= (
n-k
. . :M| )
= (
n-k
.. M| )
and we see that .
k
is equivalent to (
n-k
.. M| ) = 0, and the latter
describes the kernel of the map in the proposition.
(18.7.6) Example. If n = 2t and dimH
t
(M) is odd, then M is not a boundary of
a K-orientable compact manifold. This can be applied to R1
2n
(for K = Z,2) and
to C1
2n
(for K = R).
(18.7.7) Proposition. Let M be the boundary of a compact oriented (4k 1)-
manifold T. Then the signature of M is zero.
18.8. The Euler Number 461
Proof. It follows from Proposition (18.7.5) that the orthogonal complement of
2k
with respect to the intersection form on H
2k
(M: R) is
2k
, and 2 dim
2k
=
dimH
2k
(M: R). Linear algebra tells us that a symmetric bilinear form with these
properties has signature 0.
We generalize the preceding by taking advantage of the general duality isomor-
phism (coefcients in K). Let M be an n-manifold and 1 1 a compact pair in
M. Assume that M is K-oriented along 1. We dene a bilinear form
(+)
`
H
i
(1. 1) H
}
(M 1. M 1)
.
H
i}
(M. M 1)
as follows: Let (V. W) be a neighbourhood of (1. 1). We x an element ,
H
}
(M 1. M 1) and restrict it to H
}
(V 1. V 1). Then we have
H
i
(V. W) H
i
(V 1. W 1)
.,
H
i}
(V 1. W 1 L V 1)
H
i}
(M. W L (M 1)) H
i}
(M. M 1).
The colimit over the neighbourhoods (V. W) yields , in (+).
(18.7.8) Proposition. Let M be an n-manifold and 1 1 compact ENR in M.
Then
H
i
(1. 1) H
n-i
(M 1. M 1)
.

H
n
(M. M 1)
( -,o
!
)

K
is a regular bilinear form.
(18.7.9) Example. Let M be a compact oriented n-manifold for n 2 mod (4).
Then the Euler characteristic (M) is even.
The intersection form H
n{2
(M) H
n{2
(M) Q with coefcients in Q is
skew-symmetric and regular, since n,2 is odd. By linear algebra, a form of this
type only exists on even-dimensional vector spaces.
(M) =
n
i=0
(1)
i
dimH
i
(M)
= dimH
n{2
(M) 2
2i~n
(1)
i
dimH
i
(M):
we have used dimH
i
(M) = dimH
n-i
(M), and this holds because of H
i
(M)
Hom(H
i
(M). Q) and hence dimH
i
(M) = dimH
i
(M) = dimH
n-i
(M).
18.8 The Euler Number
Let : 1() M be ann-dimensional real vector bundle over the closedconnected
orientable manifold M. The manifold is oriented by a fundamental class M|
H
n
(M: Z) and the bundle by a Thom class t () H
n
(1. 1
0
: Z). Let s : M
1() be a section of and assume that the zero set N(s) is contained in the disjoint
sum D = D
1
L L D
i
of disks D
}
. The aim of this section is to determine the
Euler number e() = (s
+
t (). M| ) by local data. We assume given positive charts
}
: R
n
U
}
with disjoint images of M such that
}
(D
n
) = D
}
. The bundle is
trivial over U
}
. Let
D
n
R
n

pr
1([D
}
)
D
n

D
}
be a trivialization. We assume that
}
is positive withrespect tothe givenorientation
of . These data yield a commutative diagram
H
n
(1(). 1
0
())
x
H
n
(M)
H
n
(1([D) L 1
0
([M D
. 1
0
([D))
x
H
n
(. M D
H
n
(1([D). 1
0
([D))
x
H
n
(D. S)
}
H
n
(1([D
}
). 1
0
([D
}
))
x
}
H
n
(D
}
. S
}
).
Here S
}
is the boundary of D
}
and S =
_
}
S
}
. The restriction of s to D
}
is s
}
. The
image of t () in H
n
(1([D
}
). 1
0
([D
}
)) is the Thom class t ([D
}
). The vertical
maps have their counterpart in homology
H
n
(M)

H
n
(M. M D
) H
n
(D. S)
}
H
n
(D
}
. S
}
).
and M| is mapped to (D
}
|). By commutativity and naturality
(s
+
t (). M| ) =
}
(s
+
}
t ([D
}
). D
}
| ).
The bundle isomorphism (
}
.
}
) transports s
}
into a section
t
}
: D
n
D
n
R
n
. . (.. u
}
(.))
18.8. The Euler Number 463
of pr. Note that u
}
(S
n-1
) R
n
0. We have another commutative diagram
H
n
(1([D
}
). 1
0
([D
}
))
x
H
n
(D
}
. S
}
)
H
n
(D
n
R
n
. D
n
R
n
0
)
t
H
n
(D
n
. S
n-1
)
H
n
(R
n
. R
n
0
).
pr
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
The evaluation (s
+
}
t ([D
}
). D
}
| ) is the degree of u
}
: S
n-1
R
n
0
if we choose
the correct orientations. We explain this now and use the following computation:
(s
+
}
t ([D
}
). D
}
| ) = (s
+
}
t ([D
}
).
}+
) = (
+
}
s
+
}
t ([D
}
). e
n
)
= (t
+
}

+
}
t ([D
}
). e
n
) = (u
+
}
e
n
. e
n
)
= J(u
}
)(e
n
. e
n
).
The cohomological degree of u
}
is u
+
}
e
n
= J(u
}
)e
n
. With these denitions we
obtain:
(18.8.1) Proposition. e() =
_
i
}=1
J(u
}
)
_
(e
n
. e
n
).
If s has in D
}
an isolated zero, then J(u
}
) is called the index of this zero.
Problems
1. There always exists a section with a single zero.
2. The index can be computed for transverse zeros of a smooth section s : M 1 of a
smooth bundle. Consider the differential
T
\
s : T
\
M T
\
1 = T
\
M 1
\
.
Transversality means that the composition with the projection pr T
\
s : T
\
M 1
\
is an
isomorphism between oriented vector spaces. This isomorphism has a sign c(.) {1],
1 if the orientation is preserved. Show that c(.) is the local index.
3. The section
s : S
n
TS
n
S
n
R
n1
. . = (.
0
. . . . . .
n
) (.. (.
2
0
1. .
0
.
1
. . . . . .
0
.
n
))
has the transverse zeros (1. 0. . . . . 0) with index 1 and (1. 0. . . . . 0) with index (1)
n
.
4. Find a vector eld on S
2n
with a single zero (of index 2).
5. There exists a section without zeros if and only if the Euler number is zero.
We know already that the Euler number is zero, if there exists a non-vanishing section.
For the converse one has to use two facts: (1) There always exist sections with isolated zeros.
(2) There exists a cell D which contains every zero. Hence one has to consider a single local
index. This index is zero, and the corresponding map u: S R
n
0
is null homotopic. Thus
there exists an extension u: D R
n
0
. We use this extension to extend the section s over the
interior of D without zeros.
6. Consider the bundle (k): H(k) C1
1
. Let 1(:
0
. :
1
) =

k
=0
k
:
0
:
k-
1
be a
homogeneous polynomial of degree k. Then
o : C1
1
H(k). :
0
. :
1
| (:
0
. :
1
: 1(:
0
. :
1
))
is a section of (k). If 1(:
0
. :
1
) =
(a
:
1
b
:
0
) is the factorization into linear factors,
then the a
. b
| C1
1
are the zeros of o, with multiplicities.
7. Consider the bundle : S
n
Z/2
R
n
R1
n
, (.. :) .|. Then o : .
0
. . . . . .
n
|
((.
0
. . . . . .
n
). (.
1
. . . . . .
n
)) is a section with a single zero. The sections correspond to maps
: S
n
R
n
such that (.) = (.). One form of the theorem of BorsukUlam says
that maps of this type always have a zero. We would reprove this result, if we show that the
Euler class mod (2) is non-zero. The tautological bundle j over R1
n
has as Euler class the
non-zero element n of H
1
(R1
n
: Z,2). The Euler classes are multiplicative and = nj.
Hence e() = n
n
,= 0.
18.9 Euler Class and Euler Characteristic
Let M be a closed orientable n-manifold. We dene in a new manner the Thom
class of the tangent bundle. It is an element t (M) H
n
(M M. M M D)
such that for each . M the restriction of t (M) along
H
n
(M M. M M D) H
n
(. M. . (M .))
is a generator (integral coefcients, D the diagonal). The image of t (M) under the
composition
H
n
(M M. M M D) H
n
(M M)
d
H
n
(M)
(where J is the diagonal map) is now called the associated Euler class e(M) of M.
Let us use coefcients in a eld K. We still denote the image of the fundamental
class M| H
n
(M: Z) in H
n
(M: K) by M|. We use the product orientation
M M| = M| M|. Let v : 1(v) M be the normal bundle of the diagonal
J : M M M with disk- and sphere bundle D(v) and S(v) and tubular map
: D(v) M M. The fundamental class of M M| H
2n
(M M) induces
a fundamental class D(v)| H
2n
(D(v). S(v)) via
H
2n
(M M) H
2n
(M M. M M D)

H
2n
(D(v). S(v)).
18.9. Euler Class and Euler Characteristic 465
Let : =
+
(D(v)|. The diagram
H
n
(M M)
tj
H
n
(M M. M M D)
tz

H
n
(D(v). S(v))
tT()j
H
n
(M M)
=

H
n
(M M) H
n
(D(v))
}
commutes (naturality of the cap product). Suppose M is connected. From the

isomorphisms
H
n
(M) H
n
(D(v))
i
H
n
(D(v). S(v))
tT()j
we obtain an element t (v) that satises i

+
M| = t (v) : D(v)|. It is a generator
and therefore a Thom class. We dene t (M) H
n
(M M. M M D) by
+
t (M) = t (v). The image t(M) H
n
(M M) of t (M) is characterized by
the relation t : M M| = J
+
M|. From the denitions we see that J
+
t is the
image of t (v) under H
n
(D(v). S(v)) H
n
(D(v))
i
H
n
(M), hence the Euler
class e(v) of v.
Let T = {] be a basis of H
+
(M) and {
0
] the dual basis in H
+
(M) with
respect to the intersection form (
0
. M| ) =
, [
0
[ = n [[.
(18.9.1) Proposition. The image t(M) H
n
(M M) of t (M) is given by
t =
B
(1)
[[
0
H
n
(M M).
A consequence is
e(M) = J
+
t =
B
(1)
[[
0
.
(e(M). M| ) =
(1)(
0
. M| ) =
(1)
[[
= (M).
Proof. The Knneth isomorphism
H
+
(M) H
+
(M) H
+
(M M). u u .
tells us that there exists a relation of the form t =

;,B
(;. );
0
. The
following computations determine the coefcient (;. ). Let and be basis
elements of degree . Then
((
0
) t. M M| )
= (
0
. t :M M| ) = (
0
. J
+
M| )
= (J
+
(
0
). M| ) = (
0
. M| )
= (1)
](n-])
(
0
. M| ) = (1)
](n-])
.
A second computation gives
((
0
) t. M M| )
= ((
0
)
(;. );
0
. M M| )
=
(;. )(1)
[
0
[[;
0
[
(( ;
0
) (
0
). M| M| )
=
(;. )(1)
[
0
[[;
0
[
(1)
n([
0
[[[)
( ;
0
. M| )(
0
. M| ).
Only summands with ; = and = are non-zero. Thus this evaluation has the
value (. ) = (1)
]n
(collect the signs and compute modulo 2). We compare
the two results and obtain (. ) = (1)
]
.
Chapter 19
Characteristic Classes
Characteristic classes are cohomological invariants of bundles which are compatible
with bundle maps. Let h
+
() be a cohomology theory. An h
k
-valued characteristic
class for numerable n-dimensional complex vector bundles, say, assigns to each
such bundle : 1() T an element c() h
k
(T) such that for a bundle map
j over : T C the naturality property
+
c(j) = c() holds.
An assignment which has these properties is determined by its value c(;
n
)
h
k
(TU(n)) on the universal bundle ;
n
, and this value can be prescribed in an
arbitrary manner (Yoneda lemma). In other words, the elements of h
k
(TU(n))
correspond to this type of characteristic classes.
It turns out that in important cases characteristic classes are generated by a
few of them with distinguished properties, essentially a set of generators of the
cohomology of classifying spaces.
We work with a multiplicative and additive cohomology theory h
+
and bundles
are assumedtobe numerable. AC-orientationof the theoryassigns toeachn-dimen-
sional complex vector bundle (numerable, over a CW-complex,) : 1() T
a Thom class t () h
2n
(1(). 1
0
()) such that for a bundle map : 1()
1(j) the naturality
+
t (j) = t () holds and the Thom classes are multiplicative
t () t (j) = t ( j). If an assignment of this type is given, then the theory is
called C-oriented. In a similar manner we call a theory R-oriented, if for each n-
dimensional real vector bundle : 1() T a Thomclass t () h
n
(1(). 1
0
())
is given which is natural and multiplicative. It is a remarkable fact that structures
of this type are determined by 1-dimensional bundles.
(19.0.1) Theorem. A C-orientation is determined by its value
t (;
1
) h
2
(1(;
1
). 1
0
(;
1
))
on the universal 1-dimensional bundle ;
1
over C1
o
. Each Thom class t of ;
1
determines a C-orientation. A similar bijection exists between Thom classes of the
universal 1-dimensional real vector bundle over R1
o
and R-orientations.
An example of a C-oriented theory is H
+
(: Z); a complex vector bundle has
a canonical Thom class and these Thom classes are natural and multiplicative. One
can use an arbitrary commutative ring as coefcient ring.
An example of an R-oriented theory is H
+
(: Z,2); a real vector bundle has a
unique Thom class in this theory. One can use any commutative ring of character-
istic 2 as coefcient ring.
468 Chapter 19. Characteristic Classes
Suppose the theory is C-oriented. Then an n-dimensional complex vector bun-
dle over T has an Euler class e() h
2n
(T) associated to t (). Euler classes are
natural,
+
e(j) = e(
+
j), and multiplicative, e( j) = e() e(j).
The Thom classes have associated Thom homomorphisms. They are dened as
before by cup product with the Thom class
(): h
k
(T. ) h
k2n
(1(). 1()
0
L 1([)). .
+
(.) t ().
These Thom homomorphisms are natural and multiplicative as we have explained
earlier.
For an R-oriented theory we have natural and multiplicative Euler classes for
real vector bundles.
A proof of (19.0.1) is based on a determination of characteristic classes. We
present a construction of characteristic classes based on the cohomology of projec-
tive bundles. For this purpose, classifying spaces are not used. But they will of
course appear and they are necessary for a more global view-point.
19.1 Projective Spaces
Let j
n
: 1
n
C1
n-1
be the canonical bundle with total space
1
n
= C
n
0
C
C. (:. u) ~ (z:. zu).
We have the embedding v
n
: 1
n
C1
n
, (:. u) :. u|. The image is the
complement of the point + = 0. . . . . 0. 1|. Let t (j
n
) h
2
(1
n
. 1
0
n
) be a Thom
class. The Thom class yields the element t
n
h
2
(C1
n
) as the image under
h
2
(1
n
. 1
0
n
)

h
2
(C1
n
. C1
n
C1
n-1
) h
2
(C1
n
. +) h
2
(C1
n
).
The rst isomorphism is induced by v
n
. Note that v
n
sends the zero section
to C1
n-1
, the image under the embedding | : .
1
. . . . . .
n
| .
1
. . . . . .
n
. 0|.
The total space 1
n
of j
n
was denoted H(1) in (14.2.6). The bundle j
n
is the
(complex) normal bundle of the embedding | : C1
n-1
C1
n
. The embedding
v
n
is a tubular map; it also shows that C1
n
is the one-point compactication of 1
n
(see the denition of a Thom space in the nal chapter). The complement
C1
n
C1
n-1
is the afne subset U
n1
= {.
1
. . . . . .
n1
| [ .
n1
,= 0]. We
obtain a homomorphism h
+
t
n
| h
+
(C1
n
) of graded h
+
-algebras; it sends t
n1
n
to zero (see (17.2.5)) and induces a homomorphism of the quotient by the principal
ideal (t
n1
n
).
(19.1.1) Lemma. Let t (j
n-1
) h
2
(1
n-1
. 1
0
n-1
) be the Thom class obtained from
t (j
n
) by restriction along |. Let t
n-1
h
2
(C1
n-1
) be obtained from t (j
n-1
) as
explained above. Then:
19.1. Projective Spaces 469
(1) |
+
t
n
= t
n-1
.
(2) t
n-1
is the Euler class associated to t (j
n
).
Proof. (1) The embedding | is homotopic to the embedding |
1
: .
1
. . . . . .
n
|
0. .
1
. . . . . .
n
|. Thus it sufces to show |
+
1
t
n
= t
n-1
. We have a bundle map
|
2
: 1
n-1
1
n
which is compatible with the embeddings, i.e., v
n
|
2
= |
1
v
n-1
.
We apply cohomology to this commutativity and obtain the desired result.
(2) With the zero section s the diagram
h
2
(C1
n
. C1
n
C1
n-1
)

h
2
(C1
n
)
t
h
2
(1
n
. 1
0
n
)
x

h
2
(C1
n-1
)
commutes. The denition of the Euler class and (1) now yield the result.
(19.1.2) Lemma. For singular homology and cohomology the Kronecker pairing
relation (t
1
. C1
1
| ) = 1 holds. The element t
1
is by (19.1.1) also the Euler class
of j
2
and this is, by denition, the rst Chern class.
Proof. j
1
is a bundle over a point. We have the isomorphism : C 1
1
, :
(1. :). By denition of the canonical Thom class of a complex vector bundle
the Thom class t (j
1
) H
2
(1
1
. 1
0
1
) is mapped to the generator e
(2)
under
+
,
where e
(2)
is dened by the relation (e
(2)
. e
2
) = 1 (Kronecker pairing). The
element t
1
is the image of e
(2)
under H
2
(C. C0) H
2
(C1
1
. C1
1
C1
0
)
H
2
(C1
1
) and the fundamental class C1
1
| is mapped to e
2
under H
2
(C1
1
)
H
2
(C1
1
. C1
1
C1
0
) H
2
(C. C 0). Naturality of the Kronecker pairing
now gives the desired result.
(19.1.3) Theorem. The homomorphism just constructed is an isomorphism
h
+
(C1
n
) h
+
t
n
|,(t
n1
n
) of graded h
+
-algebras. In particular h
+
(C1
n
) is a
free h
+
-module with basis 1. t
n
. t
2
n
. . . . . t
n
n
.
Proof. Induction on n. We have the Thom isomorphism
h
k
(C1
n-1
) h
k2
(1
n
. 1
0
n
). . j
+
n
. L t (j
n
) = . t (j
n
).
By induction, h
+
(C1
n-1
) is a free h
+
-module with basis 1. t
n-1
. . . . . t
n-1
n-1
and
therefore h
+
(1
n
. 1
0
n
) is a free module with basis
1 t (j
n
). t
n-1
t (j
n
). . . . . t
n-1
n-1
t (j
n
).
We applythe isomorphismh
+
(1
n
. 1
0
n
) h
+
(C1
n
. +) constructedabove andclaim
that it sends t
k
n-1
t (j
n
) to t
k1
n
. Unraveling the denitions one shows that this
claim is a consequence of the naturality of the cup product and the fact (19.1.1) that
t
n-1
is the Euler class of t (j
n
).
We denote by h
+
T || the ring of homogeneous formal power series in T over
the graded ring h
+
. If T has the degree 2, then the power series in h
+
T || of degree
k consist of the series
}
a
}
T
}
with a
}
h
k-2}
. If h
+
is concentrated in degree
zero, then this coincides with the polynomial ring h
0
T |.
Let t (j
o
) h
2
(1
o
. 1
0
o
) be a Thom class and t (j
n
) its restriction. Let t
o

h
2
(C1
o
) and t
n
h
2
(C1
n
) be the corresponding elements. Since t
n
is the
restriction of t
o
let us write just t for all these elements. We have a surjective
restriction homomorphismh
+
(C1
n1
) h
+
(C1
n
). Thus the restrictions induce
an isomorphism (see (17.1.6)and (17.1.7))
h
+
(C1
o
) lim
n
h
+
(C1
n
) lim
n
h
+
t |,(t
n1
).
The algebraic limit is h
+
t ||. This shows:
(19.1.4) Theorem. h
+
(C1
o
) h
+
t ||.
We extend the previous results by a formal trick to products X C1
n
. Let
: X C1
n
C1
n
be the projection. We set u = u
n
=
+
(t
n
).
(19.1.5) Proposition. Consider h
+
(X C1
n
) as a graded h
+
(X)-algebra. Then
h
+
(X C1
n
) h
+
(X)u|,(u
n1
) and h
+
(X C1
o
) h
+
(X)u||.
Proof. The cohomology theory k
+
() = h
+
(X ) is additive and multiplicative,
and the coefcient algebra is h
+
(X). The multiplicative structure in k
+
() is
induced by the -product of h
+
() and the diagonal of X. The element u
o
now
plays the role of t
o
.
Let
i
: (C1
o
)
n
C1
o
be the projection onto the i -th factor, and set T
i
=
+
i
(t
o
). Then (19.1.5) implies:
(19.1.6) Proposition. h
+
((C1
o
)
o
) h
+
T
1
. . . . . T
n
||.
This statement uses algebraic identities of the type h
+
.. ,|| (h
+
.||),||
for graded formal power series rings.
Problems
1. An element
n
i=0
a
n
h
0
(C1
n
) is a unit if and only if a
0
h
0
is a unit.
2. An element u =

n
i=1
b
i
t
i
n
h
2
(C1
n
) is a Thom class of j
n1
if and only if b
1
is a
unit, and this holds if and only if u = ct
n
for a unit c h
0
(C1
n
).
3. Let t
n-1
be a Thomclass for j
n-1
and u
n
a Thomclass for j
n
. Then there exists a Thom
class t
n
for j
n
such that its restriction is t
n-1
.
19.2. Projective Bundles 471
19.2 Projective Bundles
Let : 1() T be an n-dimensional complex vector bundle. (For the moment
we work with bundles over spaces of the homotopy type of a CW-complex.) The
group C
+
acts brewise on 1
0
() by scalar multiplication. Let 1() be the orbit
space. The projection induces a projection
: 1() T. The bre

-1
(b)
is the projective space 1(
b
) of the vector space
-1
(b) =
b
. We call
the
projective bundle associated to .
There exists a canonical line bundle Q() 1() over 1(). Its total space is
dened as 1
0
()
C
C with respect to the relation (.. u) ~ (.z. zu). Thus over
each bre 1(
b
) we have a bundle canonically isomorphic to j
n
.
The construction of the projective bundle is compatible with bundle maps. Let
j: 1(j) C be a further bundle and : j a bundle map over : T C.
These data yield an induced bundle map
Q()
Q()
Q(j)
1()
1()
1(j).
We nowassume that we are given a Thomclass t
o
h
2
(C1
o
) of the universal
line bundle j
o
over C1
o
. Aclassifying map k
: 1() C1
o
of the line bundle
Q() 1() provides us with the element
t
= k
+
(t
o
) h
2
(1()).
We consider h
+
(1()) in the standard manner as left h
+
(T)-module, . , =
(.) ,.
(19.2.1) Example. Let = j
n1
: 1
n1
C1
n
. Then Q() 1() is canon-
ically isomorphic to j
n1
and t
= t
n
.
(19.2.2) Theorem. The h
+
(T)-module h
+
(1()) is free with basis
1. t
. t
2
. . . . . t
n-1
.
In particular
+
: h
+
(T) h
+
(1()) is injective.
Proof. This is a consequence of the LerayHirsch theorem (17.8.4) and the com-
putation (19.1.3).
(19.2.3) Corollary. There exist uniquely determined elements c
}
() h
2}
(T) such
that

n
}=0
(1)
}
c
}
()t
n-}
= 0.
since t
n
is a linear combination of the basis (c

0
() = 1).
(19.2.4) Remark. Here is a justication for the choice of the signs. Let = j
n1
.
Then t
t
n
= 0 and hence c
1
(j
n1
) = t
n
.
(19.2.5) Proposition. Let : j be a bundle map over : T C. Then the
naturality relation
+
(c
}
(j)) = c
}
() holds.
Proof. The homotopy relation k
)
1() . k
implies 1()
+
t
)
= t
. This yields
0 = 1()
+
_
}
(1)
}
c
}
(j)t
n-}
)
_
=
}
(1)
}
+
(c
}
(j))t
n-}
.
Comparing coefcients gives the claim. We have used the rule 1()
+
(a .) =
+
(a) 1()
+
(.), a h
+
(C) , . h
+
(1(j)) for the module structure; it is a
consequence of the naturality of the cup product.
(19.2.6) Proposition. Let and j be bundles over T. Then the sum formula
c
i
( j) =
i}=i
c
i
()c
}
(j).
holds. We set c
i
() = 0, if i > dim.
Proof. Consider the subspaces 1() 1( j) 1(j) and their open comple-
ments U = 1( j)1(j) and V = 1( j)1(). The inclusions 1() U
and 1(j) V are deformation retracts. Let s : 1() 1( j) 1(j),
.| .. 0| and : 1( j) 1(j) 1(), .. ,| .|. Then s = id and
s . id by the homotopy (.. ,|. z) .. z,|.
Let t = t
)
. Consider the elements (k = dim. l = dimj)
. =
k
i=0
(1)
i
c
i
()t
k-i
. , =
I
}=0
(1)
}
c
}
(j)t
I-}
.
Under the restriction h
+
(1( j)) h
+
(U) h
+
(1()) the element . is sent to
zero; this is a consequence of the denition of the c
i
(), the deformation retraction
and the naturality t [1() = t
. Hence . comes from an .

t
h
+
(1( j). U).
Similarly , comes from an element ,
t
h
+
(1( j). V ). Since U. V is an open
covering of 1( j), we see that .
t
,
t
= 0 and therefore ., = 0. We use the
denition of the c
i
( j) in the relation
., =
kI
i=0
(1)
i
_
i}=i
c
i
()c
}
(j)
_
t
i
and arrive at the desired sum formula by comparing coefcients.
19.3 Chern Classes
Let h
+
() be a cohomology theory with universal element t = t
o
h
2
(C1
o
).
Our rst aim is the computation of h
+
(TU(n)). Recall that TU(1) = C1
o
. The
space TU(n) is the basis of the universal n-dimensional complex vector bundle ;
n
19.3. Chern Classes 473
and ;
1
= j
o
. We use that h
+
(TU(1)
n
) h
+
T
1
. . . . . T
n
||, see (19.1.6). Let us
recall the ring of formal graded power series h
+
c
1
. . . . . c
n
||. The indeterminate
c
}
has degree 2 . The degree of a monomial in the c
}
is the sum of the degrees of
the factors
degree(c
k(1)
1
c
k(2)
2
. . . ) = 2k(1) 4k(2) .
A homogeneous power series of degree k is the formal sum of terms of the form
z
}
M
}
where M
}
is a monomial of degree m and z
}
h
k-n
. Thus we assign the
degree k to the elements in the coefcient group h
k
.
(19.3.1) Lemma. A classifying map : TU(n 1) TU(1) TU(n) of the
product ;
n-1
;
1
is the projective bundle of ;
n
.
Proof. Let U(n 1) U(1) U(n) be the subgroup of block diagonal matrices.
We obtain a map
: T(U(n 1) U(1)) = 1U(n),(U(n 1) U(1))
= 1U(n)
U(n)
(U(n),U(n 1) U(1))
TU(n).
A model for the universal vector bundle is ;
n
: 1U(n)
U(n)
C
n
TU(n). The
U(n)-matrix multiplication on C
n
induces a U(n)-action on the corresponding
projective space 1(C
n
). The projective bundle associated to the universal bundle ;
n
is 1U(n)
U(n)
1(C
n
) TU(n). We now use the U(n)-isomorphism 1(C
n
)
U(n),U(n 1) U(1). Hence is the projective bundle of ;
n
. We compose with
a canonical h-equivalence : TU(n 1) TU(1) T(U(n 1) U(1)).
It remains to show that = is a classifying map for ;
n-1
;
1
.
Let 1U(n 1) 1U(1) 1U(n) be a U(n 1) U(1)-map. From it we
obtain a bundle map
1(;
n-1
) 1(;
1
) = (1U(n 1) 1U(1))
U(n-1)U(1)
(C
n-1
C
1
)
1U(n)
U(n-1)U(1)
C
n
1U(n)
U(n)
C
n
= 1(;
n
).
It is a bundle map over .
(19.3.2) Theorem. Let k : TU(1)
n
TU(n) be a classifying map of the n-fold
Cartesian product of the universal line bundle. Then the following holds: The
inducedmap k
+
: h
+
(TU(n)) h
+
(TU(1)
n
) is injective. The image consists of the
power series which are symmetric in the variables T
1
. . . . . T
n
. Let c
i
h
2i
(TU(n))
be the element such that k
+
(c
i
) is the i -th elementary symmetric polynomial in
T
1
. . . . . T
n
. Then
h
+
(TU(n)) h
+
c
1
. . . . . c
n
||.
The elements c
1
. . . . . c
n
are those which were obtained from the projective bundle
associated to ;
n
by the methods of the previous section.
Proof. Let o S
n
be a permutation and also the corresponding permutation of
the factors of TU(1)
n
. Then o is covered by a bundle automorphism of the n-fold
product ;
n
1
= ;
1
;
1
. Hence ko is another classifyingmapof ;
n
1
andtherefore
homotopic to k. The permutation o induces on h
+
(TU(1)
n
) = h
+
T
1
. . . . . T
n
||
the corresponding permutation of the T
}
. Hence the image of k
+
is contained in the
symmetric subring, since k o . k. Let pr
}
: TU(1)
n
TU(1) be the projection
onto the -th factor. We write ;( ) = pr
+
}
(;
1
) so that ;
n
1
= ;(1) ;(n) and
T
}
= c
1
(;( )). We have the relation (naturality)
k
+
c
i
(;
n
) = c
i
(k
+
;
n
) = c
i
(;
n
1
) = c
i
(;(1) ;(n)).
By the sum formula (19.2.6) this equals
c
1
(;(1))c
i-1
(;(2) ;(n)) c
i
(;(2) ;(n)).
This is used to showby induction that this element is the i -th elementary symmetric
polynomial o
i
in the variables T
}
. We nowuse the algebraic fact that the symmetric
part of h
+
T
1
. . . . . T
n
|| equals the ring of graded power series h
+
o
1
. . . . . o
n
|| in
the elementary symmetric polynomials o
i
. This shows that the image of k
+
is as
claimed.
It remains to show that k
+
is injective. From (19.3.1), (19.2.2) and (19.1.5) we
obtain an injective map
+
: h
+
(TU(n)) h
+
(TU(n 1) TU(1)) h
+
(TU(n 1));(n)||.
This fact yields, by induction on n, the claimed injectivity.
Elements of h
+
(TU(n)) are calleduniversal h
+
()-valuedcharacteristic classes
for n-dimensional complex vector bundles. Given c h
+
(TU(n)) and a classifying
map : T TU(n) of the bundle over T we set c() =
+
(c) and call c()
a characteristic class. With this denition, the naturality
+
c(j) = c() holds for
each bundle map : j. Theorem (19.3.2) shows that it sufces to work with
c
i
. The corresponding characteristic class c
i
() is called the i -th Chern class of
with respect to the chosen Thom class t
o
. It is sometimes useful to consider the
total Chern class c() = 1 c
1
() c
2
() h
+
(T) of a bundle over T; the
sum formula then reads c( j) = c() c(j).
The preceding results can in particular be applied to integral singular coho-
mology. Complex vector bundles have a canonical orientation and a canonical
Thom class. There are two choices for the element t
o
, they differ by a sign.
We use the element that satises (t
1
. C1
1
| ) = 1 (Kronecker pairing), where
C1
1
| H
2
(C1
1
: Z) denotes the canonical fundamental class determined by the
complex structure.
Chern classes are stable characteristic classes, i.e., c
}
() = c
}
( c) if c
denotes the trivial 1-dimensional bundle; this follows fromthe sumformula (19.2.6)
and c
i
(c) = 0 for i > 0. This fact suggests that we pass to the limit n o.
(19.3.3) Example. The complex tangent bundle T C1
n
of the complex manifold
C1
n
satises T C1
n
c (n 1)j
n1
, see (15.6.6). Therefore the total Chern
class of this bundle is (1 c
1
(j
n1
))
n1
.
Let o: TU(n) TU(n1) be a classifying map for ;
n
c. Then o
+
c
i
= c
i
for i _ n and o
+
c
n1
= 0. Let U = colim
n
U(n), with respect to the inclusions
U(n) U(n 1).
_
0
0 1
_
.
be the stable unitary group. The classifying space TU is called the classifying
space for stable complex vector bundles. We think of this space as a homotopy
colimit (telescope) over the maps TU(n) TU(n 1). By passage to the limit
we obtain (since the lim
1
-term vanishes by (17.1.7)):
(19.3.4) Theorem. h
+
(TU) limh
+
TU(n) h
+
c
1
. c
2
. . . . ||.
(19.3.5) Example. Let k
n,n
: TU(m) TU(n) TU(m n) be a classifying
map for ;
n
;
n
= pr
+
1
;
n
pr
+
2
;
n
. We use the elements c
t
}
= c
}
(pr
+
1
;
n
) and
c
tt
}
= c
}
(pr
+
2
;
n
) and obtain
(1) h
+
(TU(m) TU(n)) h
+
c
t
1
. . . . . c
t
n
. c
tt
1
. . . . . c
tt
n
||.
Moreover, by the sum formula,
(2) k
+
n,n
c
k
=
i}=k
c
t
i
c
tt
}
.
The map k
+
n,n
is continuous in the sense that the effect on a formal power series in
the variables c
1
. . . . . c
nn
is obtained by inserting for c
k
the value (2).
For the proof of (1) one can use the theory h
+
(TU(m) ) and proceed as
for (19.3.2). In the case of integral singular cohomology one has the Knneth
isomorphism and k
+
n,n
becomes the homomorphism of algebras
(3) Zc
1
. c
2
. . . . . c
nn
| Zc
1
. . . . . c
n
| Zc
1
. . . . . c
n
|
determined by c
k

i}=k
c
i
c
}
. Since formal power series are not compatible
with tensor products, one has to use a suitably completed tensor product for general
theories if one wants a similar statement.
The maps k
n,n
combine in the colimit to a map k : TU TU TU. It is
associative and commutative up to homotopy and there exists a unit element. A
classifying map for the inverse bundle yields a homotopy inverse for k. With
these structures TU becomes a group object in h-TOP. The precise denition and
the detailed verication of these topological results are not entirely trivial. We
are content with the analogous algebraic result that we have a homomorphism
h
+
(TU) h
+
(TU TU) determined by the sum formula and continuity. The
inverse |
+
: h
+
(TU) h
+
(TU) is determined by the formal relation
(1 c
1
c
2
)(1 |
+
(c
1
) |
+
(c
2
) ) = 1.
It allows for an inductive computation |
+
c
1
= c
1
. |
+
c
2
= c
2
1
c
2
etc. We also have
the homomorphism of algebras (3) for TU. It will be important for the discussion
of the Hopf algebra structure.
(19.3.6) Example. We know already 1(Z. 2) = C1
o
= BS
1
= TU(1) .
TGL
1
(C). A numerable complex line bundle over X is determined by its classi-
fying map in
H
2
(X: Z) = X. C1
o
| = X. TU(1)|.
The corresponding element in c
1
() H
2
(X: Z) is the rst Chern class of .
(19.3.7) Proposition. The relation c
1
( j) = c
1
()c
1
(j) holds for line bundles
and j.
Proof. We begin with the universal situation. We know that H
2
(C1
o
: Z) Zis
generated by the rst Chern class c of the universal bundle ; = ;
1
. Let k : C1
o
C1
o
C1
o
be the classifying map of ;

;. Let pr
}
: C1
o
C1
o
C1
o
be the projection onto the -th factor. Then H
2
(C1
o
C1
o
: Z) has the Z-basis
T
1
. T
2
with T
}
= pr
+
}
(c). There exists a relation k
+
c
1
(;) = a
1
e
1
a
2
e
2
with
certain a
i
Z. Let i
1
: C1
o
C1
o
C1
o
. . (.. .
0
) for xed .
0
. Then
i
+
1
e
1
= c
1
(;), since pr
1
i
1
= id, and i
+
1
e
2
= 0 holds, since pr
2
i
1
is constant. We
compute
a
1
c
1
(;) = i
+
1
k
+
c
1
(;) = c
1
(i
+
1
k
+
;) = c
1
(i
+
1
(pr
+
1
; pr
+
2
;)) = c
1
(;).
since i
+
1
pr
+
1
; = ; and i
+
1
pr
+
2
; is the trivial bundle. Hence a
1
= 1, and similarly
we see a
2
= 1.
We continue with the proof. Let k
. k
)
: T C1
o
be classifying maps of
and j. Then c
1
() = k
+
c
1
(;) and similarly for j. With the diagonal J the
equalities j = J
+
(

j) = J
+
(k
k
)
)
+
(;

;) hold. This yields
c
1
( j) = c
1
(J
+
(k
k
)
)
+
(;

;))
= J
+
(k
k
)
)
+
(e
1
e
2
)
= J
+
(k
k
)
)
+
pr
+
1
c
1
(;) J
+
(k
k
)
)
+
pr
+
2
c
1
(;)
= k
+
c
1
(;) k
+
)
c
1
(;)
= c
1
() c
2
(j).
since k
= pr
1
(k
k
)
)J holds.
(19.3.8) Proposition. Let : 1() T be an n-dimensional vector bundle and
: 1() T the associated projective bundle. The induced bundle splits
() =
1

t
into the canonical line bundle
1
over the projective bundle
and another (n 1)-dimensional bundle
t
.
Proof. Think of as associated bundle 1
U(n)
C
n
T. Let H be the subgroup
U(n 1) U(1) of U(n). We obtain the pullback
1
1
C
n-1
C 1
U(n)
(U(n)
1
C
n
)
1
U(n)
C
n
1,H 1
U(n)
(U(n),H)

T
and this implies the assertion.
We nowiterate this process: We consider over 1() the projective bundle 1(
t
),
et cetera. Finally we arrive at a map (): J() T with the properties:
(1) ()
+
splits into a sum of line bundles.
(2) The induced map ()
+
: h
+
(T) h
+
(J()) is injective.
Assertion (2) is a consequence of (19.2.2).
A model for () is the ag bundle. The ag space J(V ) of the n-dimensional
vector space V consists of the sequences (=ags)
{0] = V
0
V
1
V
n
= V
of subspaces V
i
of dimension i . Let V carry a Hermitian form. Each ag has
an orthonormal basis b
1
. . . . . b
n
such that V
i
is spanned by b
1
. . . . . b
i
. The basis
vectors b
i
are determined by the ag up to scalars of norm 1. The group U(n)
acts transitively on the set of ags. The isotropy group of the standard ag is the
maximal torus T(n) of diagonal matrices. Hence we can viewJ(V ) as U(n),T(n).
The ag bundle associated to 1
U(n)
C
n
is then
(): J() = 1
U(n)
U(n),T(n) 1,T(n) T.
We can apply this construction to a nite number of bundles.
(19.3.9) Theorem (Splitting Principle). Let
1
. . . . .
k
be complex vector bundles
over T. Then there exists a map : X T such that
+
: h
+
(T) h
+
(X) is
injective and
+
(
}
) is for each
}
a sum of line bundles.
We now prove (19.0.1). Consider the exact cohomology sequence of the pair
(1(;
n
). 1
0
(;
n
)). We can use 1
0
(;
n
) as a model for TU(n 1). The projection
1(;
n
) TU(n) is an h-equivalence. Our computation of h
+
(TU(n)) shows that
we have a short exact sequence
0

h
2n
(1(;
n
). 1
0
(;
n
))

h
2n
(1(;
n
))

h
2n
(1
0
(;
n
)

0
h
2n
(TU(n))

h
2n
(TU(n 1)).
The element c
n
lies in the kernel of i
+
. It therefore has a unique pre-image
t (;
n
) h
2n
(1(;
n
). 1
0
(;
n
)). For an n-dimensional numerable bundle : 1()
T we dene t () h
2n
(1()1
0
()) to be the element k
+
t (;
n
) with a classifying
map k : ;
n
. Then the elements t () are natural with respect to bundle maps.
Fromthe relation k
+
n,n
(c
nn
) = c
t
n
c
tt
n
we conclude (by naturality) t (;
n
)t (;
n
) =
t (;
n
;
n
) and then t () t (j) = t ( j) for arbitrary numerable bundles.
The element t (;
1
) h
2
(1(;
1
). 1
0
(;
1
)) corresponds to the chosen element t
o

h
2
(C1
o
). The restriction of t (;
n
) to ;
1
;
1
yields t (;
1
) t (;
1
). This
is a Thom class, since products of Thom classes are Thom classes. This shows that
t (;
n
) is a Thom class.
19.4 StiefelWhitney Classes
The theory of Chern classes has a parallel theory for real vector bundles. Suppose
given an element t
o
h
1
(R1
o
. +) h
1
(R1
o
) such that its restriction to t
1

h
1
(R1
1
. +) is a generator of this h
0
-module. Then there exists an isomorphism
h
+
T |,(T
n1
) h
+
(R1
n
)
which sends T to the restriction t
n
of t
o
. This is then used to derive isomorphisms
h
+
(X R1
n
) h
+
(X)u|,(u
n1
).
h
+
(X R1
o
) h
+
(X)u||.
h
+
((R1
o
)
n
) h
+
T
1
. . . . . T
n
||.
The projective bundle 1() of a real vector bundle over T yields a free h
+
(T)-
module h
+
(1()) with basis 1. t
. . . . . t
n-1
, and there exists a relation
o
}=0
(1)
}
n
}
()t
n-}
= 0
with elements n
}
() h
}
(T) which satisfy the sum formula
n
i
( j) =
i}=i
n
i
()n
}
(j)
where n
0
() = 1 and n
}
() = 0 for > dim. These elements are natural
with respect to bundle maps, hence characteristic classes. We obtain an injective
19.5. Pontrjagin Classes 479
map k
+
: h
+
(TO(n)) h
+
(TO(1)
n
). The classes n
1
. . . . . n
n
which belong to the
universal n-dimensional bundle over TO(n) yield
h
+
(TO(n)) h
+
n
1
. . . . . n
n
||.
The image of n
}
under k
+
is the -th elementary symmetric polynomial in the
T
1
. . . . . T
n
. We pass to the limit n o and obtain h
+
TO as a ring of graded
power series in n
1
. n
2
. . . . with n
}
of degree . The n
}
are the universal Stiefel
Whitney classes. The StiefelWhitney classes are natural, stable, and satisfy the
sum formula. There holds a splitting principle for real bundles. We write n() =
1 n
1
() n
2
() for the total StiefelWhitney class of ; then the sum
formula reads n( j) = n() n(j).
The existence of the universal element has the consequence that the unit element
1 h
0
is of order 2, so that the cohomology groups consist of Z,2-vector spaces
and signs can be ignored. This result is due to the fact that multiplication by 1 in
the bres of vector bundles is a bundle map. If we apply this to the universal one-
dimensional bundle, then we see that this bundle map preserves the Thom class t
o
.
On the other hand, if we restrict to t
1
h
1
(R1
1
. +), this bundle map is of degree 1
and changes the sign of t
1
. Since t
1
corresponds under suspension to a unit of h
0
,
we conclude that 1 = 1 h
0
.
One shows as in the complex case that the theory is R-oriented. One can apply
these results to singular cohomology with coefcients in Z,2. There is a unique
choice for the universal element t
o
H
1
(R1
o
: Z,2). The resultingcharacteristic
classes are the classical StiefelWhitney classes.
(19.4.1) Example. The tangent bundle t of R1
n
satises t c (n 1)j
with the canonical line bundle j. The total StiefelWhitney class of t is therefore
(1 n)
n1
H
+
(R1
n
) Z,2n|,(n)
n1
. Suppose R1
n
has an immersion
into R
nk
. Then t has an inverse bundle of dimension k, the normal bundle of this
immersion. Suppose n = 2
i
. Properties of binomial numbers modulo 2 show that
n(t) = 1 n n
n
. If v is inverse to t, then n(t)n(v) = 1, and this implies in
our case n(v) = 1 n n
2
n
n-1
. This shows that an inverse bundle
must have dimension at least n 1. Therefore R1
n
has for n of the form 2
k
no
immersion into R
2n-2
.
19.5 Pontrjagin Classes
We now discuss characteristic classes for oriented bundles. Suppose : 1() T
is an oriented n-dimensional real vector bundle. The orientation determines a Thom
class t () H
n
(1(). 1
0
()) by the requirement that a positive isomorphism
i
b
: R
n
1(
b
) sends t () to the generator e
(n)
H
n
(R
n
. R
n
0) which is dual
to the generator e
n
= e
1
e
1
H
n
(R
n
. R
n
0), i.e., (e
(n)
. e
n
) = 1. This
denitionimplies e
(n)
e
(n)
= (1)
nn
e
(nn)
andhas the followingconsequences:
(19.5.1) Lemma. Let and j be oriented bundles and give j the sumorientation.
Then t ( j) = (1)
[[[)[
t () t (j) and e( j) = (1)
[[[)[
e()e(j). Here
[[ = dim.
Let : 1 T be a real vector bundle. It has a complexication
C
=
R
C.
We take with complex structure J(.. ,) = (,. .) on each bres as a model
for
C
.
Let be a complex n-dimensional bundle and
R
the underlying 2n-dimensional
real bundle. If
1
. . . . .
n
is a basis of a bre, then
1
. i
1
. . . . .
n
. i
n
is a basis of
the bre of
R
, and it denes the canonical orientation.
(19.5.2) Lemma. Let be an oriented n-dimensional real bundle. Then (
C
)
R
in our model for
C
above is isomorphic to (1)
n(n-1){2
as an oriented
bundle. The factor indicates the change of orientation, and carries the sum
orientation.
(19.5.3) Proposition. Let : 1 T be a complex n-dimensional bundle. Con-
sider it as a real bundle with orientation and canonical Thom class induced by the
complex structure. Then c
n
() = e() H
2n
(T: Z).
Proof. This holds for 1-dimensional bundles by denition of c
1
. The general case
follows by an application of the splitting principle and the sum formula.
We set
i
() = (1)
i
c
2i
(
C
) H
4i
(T: Z)
and call this characteristic class the i -th Pontrjagin class of . The bundle
C
is
isomorphic to the conjugate bundle

C
. The relation c
i
() = (1)
i
c
i
(
) holds in
general for conjugate bundles. Hence the odd Chern classes of
C
are elements
of order 2. This is a reason why we ignore them for the moment. The Pontrjagin
classes are by denition compatible with bundle maps (naturality) and they do not
change by the addition of a trivial bundle (stability). The next proposition justies
the choice of signs in the denition of the
}
.
(19.5.4) Proposition. Let be an oriented 2k-dimensional real bundle. Then
k
() = e()
2
.
Proof. We compute
k
() = (1)
k
c
2k
(
C
) = (1)
k
e
2k
((
C
)
R
) = (1)
k2k(2k-1){2
e( )
= e( ) = (1)
2k2k
e()
2
= e()
2
.
We have used (19.5.1), (19.5.2), and (19.5.3).
One can remove elements of order 2 if one uses the coefcient ring 1 = Z
1
2
|
of rational numbers with 2-power denominator (or, more generally, assumes that
1
2
1). The next theorem shows the universal nature of the Pontrjagin classes.
19.5. Pontrjagin Classes 481
(19.5.5) Theorem. Let
}
denote the Pontrjagin classes of the universal bundle
and e its Euler class. Then
H
+
(TSO(2n1): 1) 1
1
. . . .
n
|. H
+
(TSO(2n): 1) 1
1
. . . . .
n-1
. e|.
Proof. Induction over n. Let
n
: 1SO(n)
SO(n)
R
n
TSO(n) be the universal
oriented n-bundle and : TSO(n1) TSO(n) the classifying map of
n-1
c.
As model for we take the sphere bundle of
n
. Then we have a Gysin sequence
at our disposal. Write T
n
= TSO(n) for short.
Suppose nis even. Then, by induction, H
+
(T
n-1
) is generated by the Pontrjagin
classes, and
+
is surjective since the classes are stable. Hence the Gysin sequence
decomposes into short exact sequences. Let H
+
n
denote the algebra which is claimed
to be isomorphic to H
+
(T
n
). And let j
n
: H
+
n
H
+
(T
n
) be the homomorphism
which sends the formal elements
}
, e onto the cohomology classes with the same
name. We obtain a commutative diagram
0

H
i
(T
n
)
e
H
in
(T
n
)
]
H
in
(T
n-1
)

0
0

H
i
n
e
H
in
n
H
in
n-1
n-1
0.
By induction, j
n-1
is an isomorphism. By a second induction over i the left arrow
is an isomorphism. Now we apply the Five Lemma. In order to start the induction,
we note that by the Gysin sequence j
n
: H
i
n
H
i
(T
n
) is an isomorphism for
i < n.
Suppose n = 2m 1. The Euler class is zero, since we use the coefcient
ring 1. Hence the Gysin sequence yields a short exact sequence
0 H
}
(T
n
)
]
H
}
(T
n-1
) H
}-2n
(T
n
) 0.
Therefore H
+
(T
n
) is a subring of H
+
(T
n-1
) via
+
. The image of
+
contains the
subring 1
+
generated by
1
. . . . .
n
. We use
n
= e
2
. The induction hypothesis
implies
rank H
}
(T
n-1
) = rank 1
}
rank 1
}-2n
.
The Gysin sequence yields
rank H
}
(T
n-1
) = rank H
}
(T
n
) rank H
}-2n
(T
n
).
The equality rank 1
}
= rank H
}
(T
n
) is a consequence. If
+
H
}
(T
n
) ,= 1
}
then the image would contain elements of the form . e,, . 1
}
, , 1
}-2n
.
Such an element would be linearly independent of the basis elements of 1
+
. This
contradicts the equality of ranks.
(19.5.6) Example. Let be a complex bundle. Then (
R
)
C
is isomorphic to
.
An isomorphism from
R

R
with the complex structure (.. ,) (,. .) is
given by
(.. ,)
_
. i,
_
2
.
i. ,
_
2
_
.
Hence the
i
(
R
) = (1)
i
c
2i
(
) = (1)
i

ob=2i
(1)
b
c
o
()c
b
(). Since
SO(2) = U(1), an oriented plane bundle has a unique complex structure such
that
R
= . The total Pontrjagin class of is therefore 1 c
1
()
2
.
(19.5.7) Example. Let t = T C1
n
denote the complex tangent bundle of C1
n
.
Then t
R
is the real tangent bundle. In order to determine the Pontrjagin classes we
use (t
R
)
C
= t t. The total Chern class of this bundle is (1c)
n1
(1c)
n1
=
(1c
2
)
n1
if we write H
+
(C1
n
) = Zc|,(c
n1
) with c = c
1
(j
n1
), see (19.3.3).
Hence the total Pontrjagin class of C1
n
, i.e., of its tangent bundle with the canonical
orientation, is (1 c
2
)
n1
.
Problems
1. The Pontrjagin classes are stable. Under the hypothesis of (19.5.5) we obtain in the limit
H
(TSO: 1) 1
1
.
2
. . . . |. The sum formula
k
( j) =

i=k

i
()
(j)
holds (
0
= 1).
19.6 Hopf Algebras
We x a commutative ring 1 and work in the category 1-MOD of left 1-modules.
The tensor product of 1-modules M and N is denoted by M N. The natural
isomorphismt : MN N M, mn nmexpresses the commutativity
of the tensor product. We have canonical isomorphisms l : 1M M, zm
zm and r : M 1 M, mz zm. Co-homology will have coefcients in
1, if nothing else is specied.
An algebra (. m. e) in 1-MOD consists of an 1-module and linear maps
m: (multiplication), e : 1 (unit) such that m(e 1) = l,
m(1 e) = r. If m(m 1) = m(1 m) holds, then the algebra is associative,
and if mt = m holds, the algebra is commutative. Usually we write m(a b) =
a b = ab. We use similar denitions in the category of Z-graded 1-modules (with
its tensor product and interchange map).
(19.6.1) Example. Let X be a topological space. Then the graded 1-module
H
+
(X) becomes a (graded) associative and commutative algebra with multiplica-
tion
m: H
+
(X) H
+
(X) H
+
(X X) H
+
(X).
19.6. Hopf Algebras 483
where the rst map is the -product and the second map is induced by the diagonal
J : X X X. The unit is the map induced by the projection X 1 onto a
point 1.
A coalgebra (C. j. c) in 1-MOD consists of an 1-module C and linear maps
j: C C C (comultiplication), c: C 1 (counit) such that (c 1)j = l
-1
,
(1 c)j = r
-1
. If (j 1)j = (1 j)j holds, the coalgebra is coassociative,
and if tj = j holds, the coalgebra is cocommutative.
(19.6.2) Example. Let X be a topological space. Suppose H
+
(X) is a free
1-module. The graded 1-module H
+
(X) becomes a (graded) coassociative and
cocommutative coalgebra with comultiplication
j: H
+
(X) H
+
(X X) H
+
(X) H
+
(X)
where the rst map is induced by the diagonal J and the isomorphismis the Knneth
isomorphism. The counit is induced by X 1.
A homomorphism of algebras : (. m. e) (
t
. m
t
. e
t
) is a linear map
:
t
such that m = m( ) and e
t
= e. A homomorphism of
coalgebras [: (C. j. c) (C
t
. j
t
. c
t
) is a linear map [: C C
t
such that
([ [)j = j
t
[ and c
t
[ = c. A continuous map : X Y induces a homo-
morphism
+
: H
+
(Y ) H
+
(X) of the algebras (19.6.1) and a homomorphism
+
: H
+
(X) H
+
(Y ) of the coalgebras (19.6.2).
The tensor product of algebras (
i
. m
i
. e
i
) is the algebra (. m. e) with =
2
and m = (m
1
m
2
)(1t 1) and e = e
1
e
2
: 1 11
1
2
.
The multiplication m is determined by (a
1
a
2
)(b
1
b
2
) = a
1
b
1
a
2
b
2
(with
the appropriate signs in the case of graded algebras). The tensor product of co-
algebras (C
i
. j
i
. c
i
) is the coalgebra (C. j. c) with C = C
1
C
2
, comultiplication
j = (1 t 1)(j
1
j
2
) and counit c = c
1
c
2
: C
1
C
2
1 1 1.
Let (C. j. c) be a coalgebra. Let C
+
= Hom(C. 1) denote the dual module.
The data
m: C
+
C
+
(C C)
+

C
+
and e : 1 1
+
t
C
+
dene the dual algebra (C
+
. m. e) of the coalgebra. (The
rst map is the tautological homomorphism. It is an isomorphism if C is a nitely
generated, projective 1-module.)
Let (. m. e) be an algebra with a nitely generated, projective 1-module.
The data
j:
+
n
( )
+

+
+
and c:
+
e
1
+
1 dene the dual coalgebra (
+
. j. c) of the algebra.
In the case of graded modules we take the graded dual; if = (
n
[ n N
0
),
then the dual is (
n
= Hom(
n
. 1) [ n N
0
).
(19.6.3) Example. Let k : H
+
(X) Hom(H
+
(X). 1) be the map in the universal
coefcient sequence. Then k is an isomorphism of the algebra (19.6.1) onto the
dual algebra of the coalgebra (19.6.2).
(19.6.4) Proposition. Let C be a coalgebra and an algebra. Then Hom(C. )
carries the structure of an algebra with product + = m( )j, for .
Hom(C. ), and unit ec. The product + is called convolution.
Proof. The map (. ) is bilinear by construction. The (co-)associativity
of m and j is used to verify that + is associative. The unit and counit axioms yield
+ (ec) = m( ec)j = m(1 e)( 1)(1 c)j = .
Hence ec is a right unit.
A bialgebra (H. m. e. j. c) is an algebra (H. m. e) and a coalgebra (H. j. c)
such that j and c are homomorphisms of algebras. (Here H H carries the tensor
product structure of algebras.) The equality jm = (mm)(1 t 1)(j j)
expresses the fact that j is compatible with multiplication. The same equality says
that m is compatible with comultiplication. This and a similar interpretation of the
identities id = ce, je = (e e)j, m(cc) = cmis used to show that a bialgebra
can, equivalently, be dened by requiring that m and e are homomorphisms of
coalgebras. A homomorphism of bialgebras is an 1-linear map which is at the
same time a homomorphism of the underlying algebras and coalgebras.
An antipode for a bialgebra H is an s Hom(H. H) such that s is a two-
sided inverse of id(H) Hom(H. H) in the convolution algebra. A bialgebra with
antipode is called Hopf algebra.
(19.6.5) Example. Let X be an H-space with multiplication j: X X X and
neutral element .. Then
m: H
+
(X) H
+
(X) H
+
(X X)
p
H
+
(X)
is an algebra structure on H
+
(X) with unit induced by {.] X. Suppose H
+
(X) is
a free 1-module. Then the algebra structure mand the coalgebra structure (19.6.2)
dene on H
+
(X) the structure of a bialgebra. An inverse for the multiplication j
induces an antipode.
Suppose H
+
(X) is nitely generated and free in each dimension. Then
j: H
+
(X)
p
H
+
(X X) H
+
(X) H
+
(X)
is a coalgebra structure and together with the algebra structure (19.6.1) we obtain
a bialgebra. Again an inverse for j induces an antipode. The duality isomorphism
H
+
(X) Hom(H
+
(X). 1) is an isomorphism of the bialgebra onto the dual
bialgebra of H
+
(X).
19.6. Hopf Algebras 485
This situation was studied by Heinz Hopf [91]. The letter ^ for the comultipli-
cation (and even the term diagonal) has its origin in this topological context. For
background on Hopf algebras see [1], [182], [142], [138].
(19.6.6) Example. The space C1
o
is an H-space with multiplication the clas-
sifying map of the tensor product of the universal line bundle. The algebra struc-
ture is H
+
(C1
o
: Z) = Zc| with c the universal Chern class c
1
. Let C1
i
|
H
2i
(C1
o
: Z) denote the image of the fundamental class of C1
i
under the ho-
momorphism induced by the embedding C1
i
C1
o
. The coalgebra structure
is determined by j(c) = c 1 1 c, see (19.3.7). Since (c
n
. C1
n
| ) = 1
(see the proof of (18.7.2)), the dual Hopf algebra H
+
(C1
o
: Z) has an additive
basis .
i
= C1
i
|. i N
0
; by dualization of the cohomological coalgebra struc-
ture we obtain the multiplicative structure .
i
.
}
= (i. ).
i}
with (i. ) =
(i ),(i ). Geometrically this means that the map C1
i
C1
}
C1
i}
,
(.
0
. . . . . .
i
|. ,
0
. . . . . ,
}
|) :
0
. . . . . :
i}
| with :
k
=
ob=k
.
o
,
b
has degree
(i. ). The comultiplication in H
+
(C1
o
) is j(.
n
) =
i}=n
.
i
.
}
.
We generalize the Hom-duality of Hopf algebras and dene pairings. Let and
T be Hopf algebras. A pairing of Hopf algebras is a bilinear map T 1,
(a. b) (a. b) with the properties: For .. , and u. T
(.,. u) = (. ,. j(u)). (.. u) = (j(.). u ).
(1. u) = c(u). (.. 1) = c(.).
The bilinear form (. ) on T induces a bilinear form on T T
by (. ,. u ) = (1)
[,[[u[
(.. u)(,. ). This is used in the rst two axioms.
A pairing is called a duality between . T, if (.. u) = 0 for all u T implies
. = 0, and (.. u) = 0 for all . implies u = 0. An example of a pairing is the
Kronecker pairing H
+
(X) H
+
(X) 1 in the case of an H-space X.
An element . of a bialgebra H is called primitive, if j(.) = . 1 1 ..
Let 1(H) H be the 1-module of the primitive elements of H. The bracket
(.. ,) .. ,| = ., ,. denes the structure of a Lie algebra on 1(H). The
inclusion 1(H) H yields, by the universal property of the universal enveloping
algebra, a homomorphism | : U(1(H)) H. For cocommutative Hopf algebras
over a eld of characteristic zero with an additional technical condition, | is an
isomorphism [1, p. 110].
(19.6.7) Example. A coalgebra structure on the algebra of formal powers series
1.|| is, by denition, a (continuous) homomorphism j: 1.|| 1.
1
. .
2
||
with (j 1)j = (1 j)j and c(.) = 0. Here 1.
1
. .
2
|| is interpreted as a
completed tensor product 1.
1
||
1.
2
||. Then j is given by the power series
j(.) = J(.
1
. .
2
) with the properties
J(.. 0) = 0 = J(0. .). J(J(.. ,). :) = J(.. J(,. :)).
Such power series J are called formal group laws.
Problems
1. The Group algebra. Let G be a group and 1G the group algebra. The 1-module 1G
is the free 1-module on the set G, and the multiplication 1G 1G 1(G G) 1G
is the linear extension of the group multiplication. This algebra becomes a Hopf algebra, if
we dene the comultiplication by j(g) = g g for g G, the counit by c(g) = 1, and the
antipode by s(g) = g
-1
.
Let G be a nite group and O(G) the 1-algebra of all maps G 1 with pointwise
addition and multiplication. Identify O(G G) with O(G) O(G). Show that the group
multiplication m induces a comultiplication j = m
: O(G) O(G G). The data

c( ) = (1) and s( )(g) = (g
-1
) complete O(G) to a Hopf algebra. Evaluation
at g G denes an algebra homomorphism O(G) 1. Show that G is canonically
isomorphic to the group Hom(O(G). 1) of Problem 2.
An element g in a Hopf algebra H is called group-like if j(g) = g g and c(g) = 1.
The set of group-like elements in H is a group under multiplication. The inverse of g is s(g).
2. Let D be a Hopf algebra and a commutative algebra. The convolution product induces
on the set Hom(D. ) of algebra homomorphisms D the structure of a group.
3. Let H be a Hopf algebra with antipode s. Then s is an anti-homomorphismof algebras and
coalgebras, i.e., s(.,) = s(,)s(.), se = e, cs = s, t(s s)j = js. If H is commutative
or cocommutative, then s
2
= id.
4. Let H
1
and H
2
be Hopf algebras and : H
1
H
2
a homomorphism of bialgebras.
Then commutes with the antipodes.
5. Let 1 be a eld of characteristic > 0. Let = 1.|,(.
)
). The following data dene
a Hopf algebra structure on : j(.) = . 1 1 ., c(.) = 0, s(.) = ..
19.7 Hopf Algebras and Classifying Spaces
The homology and cohomology of classifying spaces TU, TO, TSOlead to a Hopf
algebra which we will study fromthe algebraic view-point in this section. The poly-
nomial algebra 1a| = 1a
1
. a
2
. . . . | becomes a Hopf algebra with coassociative
and cocommutative comultiplication determined by ^(a
n
) =

]q=n
a
]
a
q
and a
0
= 1. We consider the algebra as a graded algebra with a
i
of degree i .
(In the following we disregard the signs which appear in graded situations. An-
other device would be to assume that the a
i
have even degree, say degree 2i , or
that 1 has characteristic 2.) Let j = (j
1
. . . . . j
i
) N
i
0
be a multi-index with r
components. We use the notation a
p
= a
p
1
1
. . . a
p
i
. The monomials of type a
p
(for arbitrary r) form an 1-basis of 1a|. The homogeneous component 1a|
n
of
degree n is spanned by the monomials a
p
with [j[ = j
1
2j
2
rj
i
= n.
We have an embedding 1a
1
. . . . . a
n
|

1
1
. . . . .
n
| where a
}
is the -th
elementary symmetric polynomial in the
1
. . . . .
n
. The embedding respects the
grading if we give
}
the degree 1. The image is the subalgebra of symmetric
functions. The a
p
with j N
n
0
form an 1-basis of the symmetric polynomials.
Another, more obvious, 1-basis is obtained by starting with a monomial
19.7. Hopf Algebras and Classifying Spaces 487
J
=
i
1
1
. . .
i
n
n
and sum over the S
n
-orbit of 1 = (i
1
. . . . . i
n
). Let us write
1 ~ J if (
1
. . . . .
n
) is a permutation of (i
1
. . . . . i
n
). The polynomials
J
(
1
. . . . .
n
) =
J~J

J
form an 1-basis of the symmetric polynomials in 1
1
. . . . .
n
|. The family 1 =
(i
1
. . . . . i
n
) is called a partition of [1[ = i
1
i
2
i
n
; in the case that 1 ~ J,
we say that 1 and J yield the same unordered partition. We can write
J
as a
polynomial in the a
1
. . . . . a
n
and denote it by o
J
(a
1
. . . . . a
n
). The monomials
a
p
which are summands of o
J
have degree [j[ = [1[. Thus o
J
(a
1
. . . . . a
n
) =
o
J
(a
1
. . . . . a
n-1
. 0) = o
J
(a
1
. . . . . a
n-1
) for n > [1[.
(19.7.1) Lemma. If 1 is a partition of k and n _ k, then o
J
(a
1
. . . . . a
k
) is in-
dependent of n. We consider it as a polynomial in 1a|. In this way we obtain
another 1-basis of 1a| which consists of the polynomials o
J
. The homogeneous
component of degree n is spanned by the o
J
with 1 an (unordered) partition of n.
Consider the formal power series

U
n
=
n
}=1
(1 a
1
}
a
2
2
}
) 1a|
1
. . . . .
n
||.
The series has the form

p
a
p
T
(n)
p
(
1
. . . . .
n
) where the sum is taken over the
multi-indices j = (j
1
. . . . . j
i
) with [j[ =

}
j
}
_ n. The polynomial T
(n)
p
is
symmetric in the
1
. . . . .
n
. Hence we can write it as polynomial b
(n)
p
(b
1
. . . . . b
n
)
where b
k
is the k-th elementary symmetric polynomial in the variables
1
. . . . .
n
.
For j = (j
1
. . . . . j
i
) let 1(j) denote the multi-index (i
1
. . . . . i
n
) with i
=
for j
1
j
}-1
< v _ j
1
j
}
, i.e., we begin with j
1
entries 1, then j
2
entries 2 and so on; hence m = j
1
j
i
= [j[ and
n
k=1
i
k
= [1(j)[ = [j[,
i.e., 1(j) is a partition of [j[ with (weakly) increasing components. In the notation
introduced above
T
(n)
p
(
1
. . . . .
n
) =
J(p)
(
1
. . . . .
n
). b
(n)
p
(b
1
. . . . . b
n
) = o
J(p)
(b
1
. . . . . b
n
).
The polynomial b
(n)
p
only involves the variables b
1
. . . . . b
[J(p)[
and is independent
of n for n _ [1(j)[. We denote this stable version by b
p
. The b
p
form an 1-basis
of the symmetric polynomials in 1|. In this sense we can write formally
o
i=1
(1 a
1
}
a
2
2
}
) =
p
a
p
b
p
=
n_0
Un|
where Un| is the nite partial sum over the j with [j[ = n (although the innite
product itself is not dened). The reader may verify
U1| = a
1
b
1
U2| = a
2
1
b
2
a
2
b
2
1
2a
2
b
2
U3| = a
3
1
b
3
a
3
b
3
1
a
1
b
1
a
2
b
2
3a
3
b
3
3a
1
a
2
b
3
3a
3
b
1
b
2
.
The polynomials Un| are symmetric in the as and the bs
Un|(a
1
. . . . . a
n
: b
1
. . . . . b
n
) = Un|(b
1
. . . . . b
n
: a
1
. . . . . a
n
).
In order to see this note that U = lim
n,n
U
n,n
with
U
n,n
=
n
i=1
n
}=1
(1
i
}
)
=
n
}=1
(1 a
1
}
a
n
n
}
) =
n
i=1
(1 b
1
i
b
n
n
i
).
(19.7.2) Lemma. Let us write ^(a
p
) =
cr
a
p
cr
a
c
a
r
and b
c
b
r
=
p
b
p
cr
b
p
.
Then a
p
cr
= b
p
cr
.
Proof. The denition of the b
p
cr
implies the relation b
(n)
c
b
(n)
r
=
p
b
p
cr
b
(n)
p
. We
compute
p
_
c,r
a
p
cr
a
c
a
r
_
b
(n)
p
=
p
^(a
p
)b
(n)
p
=
(1 ^(a
1
)
i
^(a
2
)
2
i
)
=
(1 (a
1
1)
i
(a
2
1)
2
i
)
(1 (1 a
1
)
i
(1 a
2
)
2
i
)
=
_
c
(a
c
1)b
(n)
c
__
r
(1 a
r
)b
(n)
r
_
=
cr
(a
c
a
r
)b
(n)
c
b
(n)
r
=
cr
(a
c
b
r
)
p
b
p
cr
b
(n)
p
.
Now we compare coefcients and obtain
p
a
p
cr
b
(n)
p
=

p
b
p
cr
b
(n)
p
. The sum is
nite in each degree. We pass to the stable values b
p
and compare again coefcients.
Let Hom(1a|. 1) be the graded dual of 1a|. We can view this as the module
of 1-linear maps 1a| 1 which have non-zero value only at a nite number of
monomials. Let a
+
p
be the dual of a
p
, i.e., a
+
p
(a
c
) =
c
p
. The Hopf algebra structure
of 1a| induces a Hopf algebra structure on Hom(1a|. 1). The basic algebraic
result of this section is that the dual Hopf algebra is isomorphic to the original Hopf
algebra.
(19.7.3) Theorem. The homomorphism
: Hom(1a|. 1) 1b|.
p
(a
p
)b
p
is an isomorphism of Hopf algebras. The generator b
}
is dual to a
}
1
, that is,
((a
i
1
)+) = b
i
.
Proof. The dual basis element of a
p
is mapped to b
p
. Therefore is an 1-linear
isomorphism. It remains to show that is compatible with the multiplication and
the comultiplication.
19.7. Hopf Algebras and Classifying Spaces 489
We verify that is a homomorphism of algebras.
( )(g) =
_
c
(a
c
)b
c
__
r
g(a
r
)b
r
_
=
c,r
(a
c
)g(a
r
)
_
p
b
p
c,r
b
p
_
.
The coefcient of b
p
in ( g) is ( g)(a
p
) and
( g)(a
p
)=( g)(^a
p
)=( g)
_
c,r
a
p
cr
a
c
a
r
_
=
c,r
a
p
cr
(a
c
)g(a
r
).
Now we use the equality (19.7.2).
The denition of the comultiplication in Hom(1a|. 1) gives for the element
a
+
p
which is dual to a
p
the relation
^a
+
p
(a
c
a
r
) = a
+
p
(a
cr
) =
_
1. o t = j.
0. otherwise.
This means that ^(a
+
p
) =

cr=p
a
+
c
a
+
r
. Since ((a
i
1
)
+
) = b
i
, the gen-
erators of the algebras Hom(1a|. 1) and 1b| have the same coproduct. Since
we know already that is a homomorphism of algebras, we conclude that pre-
serves the comultiplication. In particular we also have for the b
p
the formula
^(b
p
) =
cr=p
b
c
b
r
.
The Hopf algebras which we have discussed have other interesting applications,
e.g., to the representation theory of symmetric groups, see [113].
(19.7.4) Remark. If we dene in (19.7.3) on the 1-module of all 1-linear maps,
then the image is the algebra 1b|| of formal power series. The homogeneous
components of degree n in 1b| and 1b|| coincide.
(19.7.5) Remark. The duality isomorphism (19.7.3) can be converted into a sym-
metric pairing : 1b| 1a| 1. The pairing is dened by (
+
,) = (,)
and satises (b
p
a
c
) =
c
p
.
Let : 1a| 1 be a homomorphism of 1-algebras. We restrict to the
component of degree n and obtain
n
: 1a|
n
1. We identify with the family
(
n
). The duality theorem sets up an isomorphism : Hom(1a|
n
. 1) 1b|
n
with the homogeneous part 1b|
n
of 1b|.
A graded group-like element 1 of 1b| is dened to be a sequence of polyno-
mials (1
n
(b
1
. . . . . b
n
) [ n N
0
) with 1
0
= 1 and 1
n
1b|
n
of degree n such
that
(1) ^1
n
=
i}=n
1
i
1
}
.
Since ^ is a homomorphism of algebras, the relation
^1
n
(b
1
. . . . . b
n
) = 1
n
(^b
1
. . . . . ^b
n
)
holds. The comultiplication has the form ^b
n
=

i}=n
b
i
b
}
(with b
0
= 1).
If we use two independent sets (b
t
i
) and (b
tt
i
) of formal variables, we can write the
condition (1) in the form
1
n
(b
t
1
b
tt
1
. b
t
2
b
t
1
b
tt
1
b
tt
2
. . . . .
i}=n
b
t
i
b
tt
}
)
=
i}=n
1
i
(b
t
1
. . . . b
t
i
)1
}
(b
tt
1
. . . . . b
tt
}
.)
The simplest example is 1
n
= b
n
.
(19.7.6) Proposition. The sequence (
n
) is an 1-algebra homomorphism if and
only if the sequence (1
n
) with 1
n
= (
n
) is a graded group-like element.
Proof. We use the duality pairing (19.7.5), now with the notation (. ,) =
(.. ,). Let (1
n
) be group-like and dene a linear map
n
: 1a|
n
1by
n
(,) =
(1
n
. ,). Then for . 1a|
i
and , 1a|
}
with i = n
n
(.,) = (1
n
. .,) = (^1
n
. . ,) = (1
i
. .)(1
}
. ,) =
i
(.)
}
(,).
Hence (
n
) is an algebra homomorphism.
Conversely, let : 1a| 1 be an algebra homomorphism with restriction
n
: 1a|
n
1 in degree n. We set 1
n
= (
n
). A similar computation as above
shows that (1
n
) is a group-like element.
(19.7.7) Remark. The algebra homomorphisms : 1a| 1 correspond to fam-
ilies of elements (z
i
1 [ i N) via ((a
i
) = z
i
). Given a family (z
i
) the
corresponding group-like element is obtained as follows. From
i
(1 z
1
t
i
z
2
t
2
i
) =
p
z
p
b
p
=
p
(a
p
)b
p
=
+
()
we see that 1
n
(b
1
. . . . . b
n
) is the component of degree n in
p
z
p
b
p
.
We now return to classifying spaces and apply the duality theorem (19.7.3). We
have the Kronecker pairing k : H
+
(TO: F
2
)H
+
(TO: F
2
) F
2
. ., (.. ,)
and the duality pairing (19.7.5) : F
2
n| F
2
u| F
2
, now with variables n. u
in place of a. b. We also have the isomorphism
+
: F
2
n| H
+
(TO: F
2
) from
the determination of the StiefelWhitney classes. We obtain an isomorphism of
Hopf algebras
+
: F
2
u| H
+
(TO: F
2
) determined via algebraic duality by the
compatibility relation (
+
..
+
,) = (. ,). The generators of a polynomial
algebra are not uniquely determined. Our algebraic considerations produce fromthe
universal StiefelWhitney classes as canonical generators of H
+
(TO: F
2
) canonical
generators of H
+
(TO: F
2
) via
+
.
In a similar manner we obtain isomorphisms H
+
(TU: Z) ZJ
1
. J
2
. . . .| (vari-
ables c. J) and H
+
(TSO: 1) 1q
1
. q
2
. . . .| (variables . q).
19.8. Characteristic Numbers 491
Problems
1. Verify the following polynomials b
,
for [j[ = 4 = [1(j)[:
b
(0,0,0,1)
= b
4
1
4b
2
1
b
2
2b
2
2
4b
1
b
3
4b
4
. 1(j) = (4):
b
(1,0,1)
= b
2
1
b
2
2b
2
2
b
1
b
3
4b
4
. 1(j) = (1. 3):
b
(0,2)
= b
2
2
2b
1
b
3
2b
4
. 1(j) = (2. 2):
b
(2,1)
= b
1
b
3
4b
4
. 1(j) = (1. 1. 2):
b
(4)
= b
4
. 1(j) = (1. 1. 1. 1).
These b
,
are the coefcients of a
,
in U4| =

,
a
,
b
,
. Check that U4| is symmetric in
the as and bs.
2. The assignment 1a| 1b| 1, a
o
b
,
(a
o
)
(b
,
) =
o
,
is a symmetric pairing.
(The formal element U =
,
a
,
b
,
could be called a symmetric copairing.)
19.8 Characteristic Numbers
Let k
: X TO(n) be a classifying map of an n-dimensional bundle. It induces

a ring homomorphism H
+
(TO(n): F
2
) H
+
(X: F
2
). We can also pass to the
stable classifying map X TO and obtain k
+
: H
+
(TO: F
2
) H
+
(X: F
2
).
This homomorphism codies the information which is obtainable from the Stiefel
Whitney classes. We use the isomorphism F
2
n| H
+
(TO: F
2
) and the duality
theorem (19.7.3). We use a slightly more general form. Let S
+
be a graded 1-
algebra; the grading should correspond to the grading of 1a|, there are no signs.
We obtain a graded algebra Hom
T
(1a|. S
+
) where the component of degree k
consists of the homomorphisms of degree k. The product in this algebra is dened
by convolution. Then we have:
(19.8.1) Theorem. There exists a canonical isomorphism
: Hom
T
(1a|. S
+
) S
+
b||
of graded 1-algebras. Here S
+
b|| = S
+
b
1
. b
2
. . . . || is the algebra of graded
formal power series in the b
i
of degree i . The isomorphism sends the 1-homo-
morphism : 1a| S
+
to the series
p
(a
p
)b
p
.
In our example we obtain fromk
+
: H
+
(TO: F
2
) H
+
(X: F
2
) a series ()
H
+
(X: F
2
)u|| of degree zero. The constant term is 1, the multiplicativity relation
( j) = ()(j) and the naturality (
+
j) =
+
(j) hold. For a line bundle
j we have (j) = 1 n
1
(j)u
1
n
1
(j)
2
u
2
. These properties characterize
the assignment ().
We can apply a similar process to oriented or complex bundles. In the case of
a complex oriented theory h
+
() we obtain series () h
+
(X)J|| which are
natural, multiplicative and assign to a complex line bundle j the series (j) =
1 c
1
(j)J
1
c
1
(j)
2
J
2
.
Interesting applications arise if we apply the process to the tangent bundle of
a manifold. Let us consider oriented closed n-manifolds M with classifying map
k
: M TSO of the stable oriented tangent bundle. We evaluate the homomor-

phism k
+
on the fundamental class M|

H
n
(TSO: 1) H
n
(M: 1) 1. . k
+
(.)M|.
By the Kronecker pairing duality H
n
(TSO: 1) Hom
T
(H
n
(TSO: 1). 1) this
homomorphism corresponds to an element in H
n
(TSO: 1), and this element is
k
+
M|, the image of the fundamental class M| H
n
(M: 1) under (k
)
+
, by the
naturality (k
+
(). M| ) = (. (k
)
+
M| ) of the pairing. Under the isomorphism
+
: 1q
1
. q
2
. . . .| H
+
(TSO: 1) the element k
+
M| corresponds to an element
that we denote
SO
(M) 1q
1
. q
2
. . . .|. From the denitions we obtain:
(19.8.2) Proposition. Let t
denote the oriented tangent bundle of M. Then
SO
(M) = ((t
). M| ).
the evaluation of the series (t
) on the fundamental class.

If H
n
(M) is a polynomial of degree n in the Pontrjagin classes, then
the element (number) M| is called the corresponding Pontrjagin number. In a
similar manner one denes a StiefelWhitney number by evaluating a polynomial
in the StiefelWhitney classes on the fundamental class. A closed n-manifold M
has an associated element
O
(M) F
2
u
1
. u
2
. . . .| H
n
(TO: F
2
), again the
image of the fundamental class under the map induced by the stable classifying
map k
: M TO of the tangent bundle.

(19.8.3) Example. Let us consider M = C1
2k
. The stable tangent bundle is
j
2k1
where is j is the canonical complex line bundle, now considered as oriented
bundle; see (15.6.6). By the multiplicativity of the -classes, we have for the tangent
bundle t
2k
of C1
2k
the relation
(t
2k
) = (j)
2k1
= (1
1
(j)q
1
1
(j)
2
q
2
)
2k1
= (1 c
2
q
1
c
4
q
2
)
2k1
where as usual H
+
(C1
2k
: 1) 1c|,(c
2k1
). Note that
1
(j) = c
2
, by (19.5.6).
The evaluation on the fundamental class yields the coefcient of c
2k
in this series,
since (c
2k
. C1
2k
| ) = 1 (see (18.7.2)). Modulo decomposable elements in the
indeterminates q
}
, i.e., modulo polynomials in the q
}
with < k, this value is
(2k 1)q
k
.
When we pass to rational coefcients 1 = Q we can divide by 2k 1 and
obtain:
19.8. Characteristic Numbers 493
(19.8.4) Proposition. The elements
SO
(C1
2k
). k Nare polynomial generators
of Qq|.
In bordismtheory it will be shown that the signature of an oriented 4k-manifold
only depends on its image in H
+
(TSO: Q) Qq|. And from the multiplica-
tivity of the signature it then follows that there exists an algebra homomorphism
s : H
+
(TSO: Q) Q such that the signature o(M) is obtained as the image of
this homomorphism o(M) = s(k
M
)
+
M|). We know that the generators C1
2k
have signature 1; see 18.7.2. The ring homomorphism s is determined by the val-
ues z
i
= s(q
i
) Q. Via the duality Q| Hom(Qq|. Q) the homomorphism
s corresponds to a group-like element (1
n
(
1
. . . . .
n
) [ n N) where 1
n
is a
polynomial in the Pontrjagin classes of degree 4n such that the evaluation on the
fundamental class is the signature, (1
n
. M
4k
| ) = o(M
4n
). If we expand the for-
mal product
i
(1 z
1
t
i
z
2
t
2
i
) and assume that
k
is the k-th elementary
symmetric polynomial in the t
i
(of degree 4), then 1
n
is the component of degree 4n.
The t
i
are obtained if we split the total Pontrjagin class (formally) into linear factors,
1
1
.
2
.
2
=
i
(1 t
i
.). Fortunately, nature has already split for us
the stable tangent bundle of C1
2n
, the total Pontrjagin class is (1 c
2
)
2n1
; i.e.,
we can take t
i
= c
2
in order to evaluate 1
n
on C1
2n
. This allows us to determine
the coefcients z
i
: The power series H(c) = 1 z
1
c
2
z
2
c
4
has the
property that the coefcient of c
2n
in H(c)
2n1
is 1. Hirzebruch [81, p. 14] has
found this power series
H(c) =
c
tanh c
=
2c
e
2c
1
c = 1
T
1
2
(2c)
2
T
2
4
(2c)
4
T
3
6
(2c)
6

where the T
}
are the so-called Bernoulli numbers. The rst four values are
T
1
=
1
6
. T
2
=
1
30
. T
3
=
1
42
. T
4
=
1
30
.
The corresponding coefcients in the power series are
z
1
=
1
3
. z
2
=
1
3
2
5
. z
3
=
2
3
3
5 7
. z
4
=
1
3
3
5
2
7
.
From these data we obtain the polynomials 1
n
if we insert in the universal polyno-
mials Un|(
1
. . . . .
n
: q
1
. . . . . q
n
) for q
}
the value z
}
. We have already listed the
polynomials U1|, U2|, U3|, U4|. The result is
1
1
=
1
3
1
.
1
2
=
1
45
(7
2

2
1
).
1
3
=
1
945
(62
3
13
2
1
2
3
1
).
1
4
=
1
14175
(381
4
71
3
1
19
2
2
22
2
2
1
3
4
1
).
The polynomials 1
n
are called the Hirzebruch L-polynomials.
Problems
1. Show that
SO
(M N) =
SO
(M)
SO
(N).
2. Let M be the oriented boundary of a compact manifold. Then
SO
(M) = 0. (See the
bordism invariance of the degree.)
3. Show that
O
(R1
2n
) = u
2n
modulo decomposable elements. Therefore these elements
can serve as polynomial generators of H
(TO: F
2
) F
2
u| in even dimensions.
4. The convolution product of the homomorphisms dened at the beginning of the section
satises k
+ k
)
= k
)
.
5. Determine
SO
(C1
2
) and
SO
(C1
4
).
Chapter 20
Homology and Homotopy
We begin this chapter with the theorem of Hurewicz which says in its simplest form
that for a simply connected space the rst non-zero homotopy group is isomorphic
to the rst non-zero integral homology group. In the case of the sphere S
n
this is
essentially the Hopf degree theorem. In our proof we use this theorem and other
consequences of the homotopy excision theorem. We indicate an independent proof
which only uses methods from homology theory and the Eilenberg subcomplexes
introduced earlier. The theorem of Hurewicz has the important consequence that
a map between simply connected CW-complexes is a homotopy equivalence if it
induces an isomorphism of the integral homology groups (theorem of Whitehead).
Another application is to the geometric realization of algebraic chain complexes as
cellular chain complexes. We will see that under suitable hypotheses we do not
need more cells in a homotopy type than the homology groups predict.
Since homotopy groups are difcult to compute it is desirable to have at least
some qualitative information about them. One of the striking results is the famous
theorem of Serre that the homotopy groups of spheres are nite groups, except in
the few cases already known to Hopf; in particular the stable homotopy groups of
spheres are nite (except
n
(S
n
)). Since for a nite abelian group the tensor
product Q = 0 and since homology theories are objects of stable homotopy,
this theorem has the remarkable consequence that rationalized homology theories
h
+
() Qcan be reduced to ordinary rational homology.
Along the way we obtain qualitative results in general. They concern, for in-
stance, statements about niteness or nite generation and are based on qualitative
generalizations of the theorem of Hurewicz. For the expert we point out that we do
not use the theory of spectral sequences for the proofs. Only elementary methods
like induction over skeleta enter. A basic technical theorem relates in a qualitative
manner the homology of the total space, bre and base of a bration. On the alge-
braic side we use so-called Serre classes of abelian groups: Properties like nite
generation are formalized. (In the long run this leads to localization of spaces and
categories.)
20.1 The Theorem of Hurewicz
The theoremof Hurewicz relates the homotopy and the homology groups of a space.
In this section H
+
denotes integral singular homology. Let (X. . +) be a pointed
pair of spaces.
496 Chapter 20. Homology and Homotopy
We dene natural homomorphisms, called Hurewicz homomorphisms,
h
(A,,+)
= h:
n
(X. . +) H
n
(X. ). n _ 2.
h
(A,+)
= h:
n
(X. +) H
n
(X). n _ 1.
such that the diagrams
n
(X. +)

h
n
(X. . +)
J
n-1
(. +)
h
H
n
(X)

H
n
(X. )
J
H
n-1
()
commute (compatibility with exact sequences). For this purpose we use the def-
inition
n
(X. +) = S(n). X|
0
and
n
(X. . +) = (D(n). S(n 1)). (X. )|
0
of the homotopy groups (see (6.1.4)). We choose generators :
n
H
n
(S(n))
and :
n
H
n
(D(n). S(n 1)) such that d :
n
= :
n-1
and q
+
( :
n
) = :
n
, where
q : D(n) D(n),S(n 1) = S(n) is the quotient map. If we x :
1
, then
the other generators are determined inductively by these conditions. We dene
h:
n
(X. . +) H
n
(X. ) by |
+
( :
n
) and h:
n
(X. +) H
n
(X) by
|
+
(:
n
). With our choice of generators the diagram above is then commuta-
tive. From(10.4.4) and the analogous result for the relative homotopy groups we see
that the maps h are homomorphisms. The singular simplex ^
1
1,d1 = S(1),
(t
0
. t
1
) t
1
represents a generator :
1
H
1
(S(1)). If we use this generator, then
h:
1
(X. +) H
1
(X) becomes the homomorphism which was shown in (9.2.1)
to induce an isomorphism
1
(X. +)
ob
H
1
(X) for 0-connected X.
Recall that we have a right action of the fundamental group
n
(X. . +)
1
(. +)
n
(X. . +). (.. ) .
via transport. We denote by
#
n
(X. . +) the quotient of
n
(X. . +) by the normal
subgroup generated by all elements of the form . . (additive notation in
n
).
Recall from (6.2.6) that
#
2
is abelian. Representative elements in
n
which differ
by transport are freely homotopic, i.e., homotopic disregarding the base point.
Therefore the Hurewicz homomorphism induces a homomorphism
h
#
:
#
n
(X. . +) H
n
(X. ).
The transport homomorphism
n
(X. . a
1
)
n
(X. . a
2
) along a path from
a
1
to a
2
induces an isomorphism of the
#
n
-groups, and this isomorphism is in-
dependent of the choice of the path. We can use this remark: An unpointed map
(D(n). S(n1)) (X. ) yields in each of the groups
#
n
(X. . a) a well-dened
element ( path connected). Thus, if
1
() is trivial, we can regard
#
n
(X. ) as
the homotopy set (D(n). S(n 1)). (X. )|. The group
#
1
(X. +) is dened to
be the abelianized group
1
(X. +)
ob
, i.e., the quotient by the commutator sub-
group. We set
#
n
(X. +) =
n
(X. +) for n _ 2 and again we have the Hurewicz
homomorphism h
#
:
#
n
(X. +) H
n
(X).
20.1. The Theorem of Hurewicz 497
(20.1.1) Theorem (Hurewicz). Let the space X be (n 1)-connected (n _ 1).
Then h
#
:
#
n
(X. +) H
n
Proof. We have already proved the theorem in the case that n = 1. So let n _ 2
and then
#
n
=
n
. Since weak homotopy equivalences induce isomorphisms in
homotopy and homology, we only need to prove the theoremfor CW-complexes X.
We can assume that X has a single 0-cell and no i -cells for 1 _ i _ n 1, see
(8.6.2). The inclusion X
n1
X induces isomorphisms
#
n
(X)
#
n
(X
n1
)
and H
n
(X) H
n
(X
n1
). Since the Hurewicz homomorphisms h form a natural
transformation of functors, it sufces to prove the theorem for (n1)-dimensional
complexes. In this case X is h-equivalent to the mapping cone of a map of the form
: =
_
S
n
}
T =
_
S
n
k
.
For X = S
n
the theorem holds by (10.5.1). By naturality and additivity it then
holds for pointed sums
_
S
n
}
. We have a commutative diagram
n
()

n
(T)

n
(X)
0
H
n
()

H
n
(T)

H
n
(X)

0
with exact rows. The exactness of the top row is a consequence of the homotopy
excision theorem.
(20.1.2) Corollary. Let X be simply connected and suppose that

H
i
(X) = 0 for
i < n. Then
i
(X. +) = 0 for i < n and h:
n
(X. +) H
n
(X).
Proof. (20.1.1) says, in different wording, that h:
}
(X) H
}
(X) for the smallest
such that
k
(X) = 0 for 1 _ k < .
(20.1.3) Theorem. Let (X. ) be a pair of simply connected CW-complexes. Sup-
pose H
i
(X. ) = 0 for i < n, n _ 2. Then
i
(X. ) = 0 for i < n and
h:
n
(X. ) H
n
Proof. Induction over n _ 2. We use a consequence of the homotopy excision
theorem: Let be simply connected and
i
(X. . +) = 0 for 0 < i < n. Then
n
(X. . +)
n
(X,. +) is an isomorphism. The theorem of Seifert and van
Kampen shows
1
(X,) = {e]. From H
i
(X. ) =

H
i
(X,) and (20.1.2) we
conclude
i
(X,) = 0 for i < n.
Let n = 2. Since X and are simply connected,
1
(X. . +) = 0 and the
diagram
2
(X. . +)

h
2
(X,. +)
H
2
(X. )

H
2
(X,)
shows that h is an isomorphism.
By induction we know that
}
(X. . +) = 0 for 0 < < n. Then a similar
diagram shows that h:
n
(X. . +) H
n
(20.1.4) Theorem. Let : X Y be a map between simply connected spaces.
Suppose
+
: H
i
(X) H
i
(Y ) is bijective for i < n and surjective for i = n
(n _ 2). Then
+
:
i
(X)
i
(Y ) is bijective for i < n and surjective for i = n.
Proof. We pass to the mapping cylinder and assume that is an inclusion. The hy-
pothesis is then equivalent to H
i
(Y. X) = 0 and the claimequivalent to
i
(Y. X) =
0 for i < n. Now we use (20.1.3).
Recall: A map : X Y between CW-complexes is an h-equivalence if
and only if
+
:
i
(X)
i
(Y ) for each i . Together with (20.1.4) we obtain a
homological version of this result:
(20.1.5) Theorem (Whitehead). Let : X Y be a map between simply con-
nected CW-complexes, which induces isomorphisms of homology groups. Then
In (20.1.5) one cannot dispense with the hypothesis that the spaces are simply
connected. There exist, e.g., so-called acyclic complexes X withreducedhomology
groups vanishing but with non-trivial fundamental group. Moreover it is important
that the isomorphism is induced by a map.
(20.1.6) Proposition. Let X be a simply connected CW-complex with integral ho-
mology of a sphere, H
+
(X) H
+
(S
n
), n _ 2. Then X is h-equivalent to S
n
.
Proof. By (20.1.2),
n
(X) H
n
(X), and this group is assumed to be isomorphic
to Z. Let : S
n
X represent a generator. Then
+
:
n
(S
n
)
n
(X) is an
isomorphism and also
+
: H
n
(S
n
) H
n
(X). Now we use (20.1.5).
The preceding proposition has interesting applications. It is known that a closed
connected n-manifold of the homotopy type of the n-sphere is actually homeomor-
phic to the n-sphere. Therefore these spheres are characterized by invariants of
algebraic topology.
(20.1.7) Example. The spaces S
n
S
n
S
2n
and S
n
S
n
are for n _ 2 simply
connected and have isomorphic homology groups. But they are not h-equivalent,
since their cohomology rings are different.
The homological theorem of Whitehead (20.1.5) no longer holds for spaces
which are not simply connected, even in the case when the map induces an isomor-
phism of the fundamental groups. But it sufces to consider the universal covering,
as the next theorem shows.
20.1. The Theorem of Hurewicz 499
(20.1.8) Theorem. Let : X Y be a map between connected CW-complexes
which induces an isomorphism
+
:
1
(X)
1
(Y ). Let :

X X and
q :

Y Y be the universal coverings. There exists a lifting J :

X

Y of ,
i.e., qJ = . Suppose J induces an isomorphism of the homology groups. Then
Proof. We choose isomorphisms
1
(X) G
1
(Y ) which transform
+
into
the identity of G. We then consider and q as G-principal bundles with left action
and J :

X

Y as a G-map. We obtain a morphismof the associated bre bundles.
X

T
EG
G

X

EG
G
T
BG
=
Y

EG
G

Y

BG
From the assumption and (20.1.5) we see that J is an h-equivalence. The exact
homotopy sequence and the Five Lemma showthat EG
G
J induces isomorphisms
of the homotopy groups.
We now consider the second associated bre bundles 1 : EG
G

X X and
Q: EG
G

Y Y . A section s of 1 arises from a map o :

X EG such
that o(g.) = o(.)g
-1
for . X and g G, see (14.1.4). A map of this type
is essentially the same thing as a classifying map of . Since the bre of 1 is
contractible, 1 induces isomorphisms of the homotopy groups, and the same holds
then for a section s of 1. We see that = Q(EG
G
J)s induces isomorphisms
of homotopy groups; hence is a homotopy equivalence.
(20.1.9) Corollary. In the situation of (20.1.8) J is a G-homotopy equivalence.
Proof. Let h: Y X be h-inverse to and H:

Y

X a lifting of H which is a
G-map. A homotopy of h can be lifted to a G-homotopy of HJ. The end of this
homotopy is a bundle automorphism.
Let G be a discrete group which acts on the pair (Y. T). The induced maps
of the left translations by group elements yield a left action of G on H
n
(Y. T) via
homomorphisms, i.e., H
n
(Y. T) becomes a module over the integral group ring
ZG of G.
Suppose X is obtained from by attaching n-cells (n _ 3). Let : Y X
be a universal covering and T =
-1
(). Then Y is obtained fromT by attaching
n-cells. The group =
1
(X) of deck transformations acts freely on the set
of n-cells in Y T. Hence H
n
(Y. T) is a free Z-module, the basis elements
correspond bijectively to the n-cells of X . Theorem (20.1.3) now yields:
(20.1.10) Theorem. Let X be a connected CW-complex and let n _ 3. Then
n
(X
n
. X
n-1
) is a free Z
1
(X
n-1
)-module. A basis of this module consists of the
characteristic maps of the n-cells. The map h
#
:
n
(X
n
. X
n-1
) H
n
(X
n
. X
n-1
)
is an isomorphism.
The exact sequence
n1
(X
n1
. X
n
)
n
(X
n
. X
n-1
)
n
(X
n1
. X
n-1
) 0
is for n _ 3 a sequence of Z
1
(X
n-1
) Z
1
(X
n
)-modules. Because of the
isomorphism
n
(X. X
n-1
)
n
(X
n1
. X
n-1
) the sequence is a presentation of
the Z
1
(X
n-1
)-module
n
(X. X
n-1
). The induced sequence of the
#
-groups is
also exact. This gives the following theorem for n _ 3.
(20.1.11) Theorem(Hurewicz). Let (X. ) be a CW-pair with connected X and .
Let (X. ) be (n 1)-connected (n _ 2). Then h
#
:
#
n
(X. . +) H
n
(X. ).
Proof. We now give a purely homological proof of the Hurewicz theorems which
also covers the relative case n = 2 for spaces which are not simply connected. The
proof is by induction. The induction starts with (9.2.1). We assume the absolute
theorem for 1 _ i _ n 1 and prove the relative theorem for n. We consider the
standard simplex ^k| = e
0
. . . . . e
k
| as the usual simplicial complex and denote
its l-skeleton by ^k|
I
. Let S
n-1
k
(X. . +) be the chain group spanned by simplices
o : ^k| X such that o(^k|
n-1
) and o(^k|
0
) = {+], modulo S
k
().
Let H
(n-1)
n
(X. . +) be the n-th homology group of the resulting chain complex.
The inclusion of chain complexes induces an isomorphism H
(n-1)
n
(X. . +)
H
n
(X. ) for an (n 1)-connected pair (X. ) with path connected , see (9.5.4).
We have to adapt the homotopy groups to the simplicial setup. We consider
elements of
n
(X. . +) as homotopy classes of maps : (^n|. d^n|. e
0
)
(X. . +). For this purpose we x a homeomorphism : (D(n). S(n 1). +)
(^n|. d^n|. e
0
) which sends the generator :
n
dened at the beginning to the
standard generator id(^n|)| H
n
(^n|. d^n|).
The Hurewicz homomorphism then sends the homotopy class of to
+
id|,
and this class is an element in H
(n-1)
n
(X. . +). We now construct an inverse
[: H
(n-1)
n
(X. . +)
#
n
(X. . +) of h. We assign to a singular simplex
o : (^n|. d^n|. ^n|
0
) (X. . +)
the element in
#
n
(X. . +) represented by o. If o(^n|) , then the cor-
responding homotopy class is zero. Since the
#
n
-group is abelian, we obtain a
well-dened homomorphism [: S
n-1
n
(X. . +)
#
n
(X. . +). The simplices o
are cycles (since S
n-1
n-1
(X. . +) = 0), and thus it remains to show that the com-
posite [ d: S
n-1
n1
(X. . +)
#
n
(X. . +) is trivial, in order to obtain [. We
reduce the problem to a universal situation. For this purpose we dene elements
b
n

n
(d^n 1|. ^n 1|
n-1
. e
0
),
b
2
= (J
3
0
| e
1
e
0
|)J
3
2
|J
3
1
|
-1
J
3
3
|
-1
.
b
n
= J
n1
0
| e
1
e
0
|
n1
i=1
(1)
i
J
n1
i
|. n _ 3.
20.2. Realization of Chain Complexes 501
where n| denotes the afne path class in ^n 1| from to n, and we use
the transport along this path. (Multiplicative notation in
2
, additive notation for
n _ 3. This denition corresponds to the homological boundary operator, but we
have to transport the face maps to the base point e
0
. For b
2
we have to pay attention
to the order of the factors, since the group is non-abelian.)
Let us write 1 = ^n 1| and let t : (1. 1
n-1
. 1
0
) (X. . +) be a basis
element of S
n-1
n1
(X. . +). Then
[dt| =
i
(1)
i
tJ
n1
i
| = t
#
+
b
n
|
#
= t
#
+
#
+
b
n
|
#
where denotes the inclusion d^n1| ^n1|. Thus it remains to show that
+
b
n
| = 0.
The skeleton 1
n-1
is (n 2)-connected (use e.g., the induction hypothe-
sis). Hence, by the inductive assumption,
#
n-1
(1
n-1
. e
0
) H
n-1
(1
n-1
) is
an isomorphism. The commutativity dh = hd now shows that db
n
| = 0, since
dhb
n
| = 0 by the fundamental boundary relation for singular homology. We have
the factorization
d:
n
(1
n
. 1
n-1
. e
0
)
}
n
(1. 1
n-1
. e
0
)
J
/
n-1
(1
n-1
. e
0
).
and d
t
is an isomorphism, since 1 is contractible. This nishes the inductive step
for the relative Hurewicz theorem. For n _ 2 the absolute theorem is a special case
of the relative theorem.
An interesting consequence of the homological proof of the Hurewicz theorem
is a new proof of the BrouwerHopf degree theorem
n
(S
n
) Z.
20.2 Realization of Chain Complexes
The computation of homology groups from the cellular chain complex shows that
one needs enough cells to realize the homology groups algebraically as the homol-
ogy groups of a chain complex. It is interesting to know that in certain cases a
converse holds. We work with integral homology.
(20.2.1) Theorem (Cell Theorem). Let Y be a 1-connected CW-complex. Suppose
H
}
(Y ) is nitely generated for _ n. Then Y is homotopy-equivalent to a CW-
complex 7 with nitely many -cells for _ n.
The proof of this theorem is based on a theorem which says that under suitable
hypotheses an algebraic chain complex can be realized as a cellular chain complex.
We describe the inductive construction of a realization. We start with the following:
20.2.2 Data and notation.
(1) Y is a CW-complex with i -skeleton Y
i
.
(2) 7
i
is an r-dimensional CW-complex.
(3) : 7
i
Y is a cellular map.
(4) C
i
(7) = H
i
(7
i
. 7
i-1
: Z) is the i -th cellular chain group.
(5) induces a chain map
-
: C
-
(7) C
-
(Y ).
(6) We attach (r 1)-cells to 7
i
such that can be extended to J:
lS
i
}

7
i
(
Y
i
lD
i1
}
7
i1
T
Y
i1
.
(7) From this diagram we obtain a resulting diagram of chain groups
i1
C
i
(7)
d
C
i-1
(7)
-1
C
i1
(Y )
d
/
1
C
i
(Y )
d
/
C
i-1
(Y )
with
i1
= H
i1
(7
i1
. 7
i
) a free abelian group with a basis given by the
(r 1)-cells, and [ induced by (J. ).
We now start from a diagram in which
i1
is a free abelian group with basis
(a
}
[ J). The horizontal parts should be chain complexes, i.e., J = 0. Can
this diagram be realized geometrically?
(20.2.3) Proposition. A realization exists, if the following holds:
(1)
+
: H
i
(7
i
) H
i
(Y ) is bijective for i _ r 1 and surjective for i = r;
(2) r _ 2;
(3) 7
i
and Y are 1-connected.
Proof. Suppose we are given for each J a diagram
S
i
b
7
i
(
D
i1
B
Y
i1
.
We attach (r 1)-cells to 7
i
with attaching maps b
}
to obtain 7
i1
and use the T
}
to extend to
i1
: 7
i1
Y
i1
. Then
i1
H
i1
(7
i1
. 7
i
) canonically
and basis preserving.
We consider as an inclusion. The assumption (1) is then equivalent to
H
i
(Y. 7
i
) = 0 for i _ r. Since r _ 2 and
1
(Y ) = 0 we also have
1
(Y
i1
) = 0.
20.2. Realization of Chain Complexes 503
Therefore we have the relative Hurewicz isomorphism h:
i1
(Y
i1
. 7
i
)
H
i1
(Y
i1
. 7
i
), since H
i
(Y
i1
. 7
i
) H
i
(Y. 7
i
) = 0 for i _ r. The diagram
represents anelement in
i1
(Y
i1
. 7
i
). Let .
}
= h(T
}
. b
}
|) H
i1
(Y
i1
. 7
i
)
be its image under the relative Hurewicz homomorphism. This element can be de-
termined by homological conditions. The correct maps and [ are obtained, if .
}
has the following properties:
(1) The image of .
}
under d: H
i1
(Y
i1
. 7
i
) H
i
(7
i
) H
i
(7
i
. 7
i-1
)
is (a
}
).
(2) The image of .
}
under ; : H
i1
(Y
i1
. 7
i
) H
i1
(Y
i1
. Y
i
) is [(a
}
).
We show that there exists a unique element .
}
with these properties. We know
H
i1
(Y. r. 7
i
) = 0 and H
}
(Y
i-1
7
i-1
) = 0, _ r for reasons of dimension.
The exact sequence of the triple (Y
i1
. Y
i
. 7
i
) shows that ; is injective. Hence
there exists at most one .
}
with the desired properties. The existence follows if we
showthat Im(d. ;) = Ker(
i
J
t
i1
). This follows by diagram chasing in the next
diagram with exact rows
H
i1
(Y
i1
. 7
i-1
)

H
i1
(Y
i1
. 7
i
)
J
H
i
(7
i
. 7
i-1
)

H
i
(Y
i1
. 7
i-1
)
H
i1
(Y
i1
. Y
i-1
)

H
i1
(Y
i1
. Y
i
)
d
/
1
H
i
(Y
i
. Y
i-1
)

H
i
(Y
i1
. Y
i-1
).
One uses that is surjective and injective.
Proof. Since Y is simply connected, we can assume that Y has a single 0-cell and
no 1-cells. We construct 7 inductively with 7
0
= {+] and 7
1
= {+]. We choose a
nite number of generators for
2
(Y ) H
2
(Y ) and representing maps S
2
Y
2
.
They yield a cellular map
2
: 7
2
=
_
S
2
Y , and the induced map is surjective
in H
2
and bijective in H
}
, < 2. This starts the inductive construction.
Suppose
i
: 7
i
Y is given such that
i+
: H
i
(7
i
) H
i
(Y ) is bijective
for i _ r 1 and surjective for i = r. We construct a diagram of type (7) in
20.2.2 as follows. We have H
i
(7
i
) = Ker(J) and H
i
(Y ) = Ker(J
t
i
), Im(J
t
i1
)
and the map (
i
)
+
: H
i
(7
i
) = Ker(J) H
i
(Y ) is surjective. Let
i1
be the
kernel of (
i
)
+
and : H
i
(7
i
) C
i
(7
i
). As a subgroup of C
i
(7
i
) it is free
abelian. Since 7
i
has, by induction, a nite number of r-cells, the group
i1
is nitely generated. By denition of
i1
, the image of
i
is contained in the
image of J
t
i1
. Hence there exists [ making the diagram commutative. We now
apply (20.2.3) in order to attach an (r 1)-cell for each basis element of
i1
and
to extend
i
to
t
i1
: 7
t
i1
Y . By construction, (
t
i1
)
+
is now bijective on
H
i
. If this map is not yet surjective on H
i1
we can achieve this by attaching more
(r 1)-cells with trivial attaching maps; if H
i1
(Y ) is nitely generated, we only
need a nite number of cells for this purpose. We continue in this manner as long as
H
+
(Y ) is nitely generated. After that point we do not care about nite generation.
The nal map : 7 Y is a homotopy equivalence by (20.1.5).
20.3 Serre Classes
Typical qualitative results in algebraic topology are statements of the type that the
homotopy or homology groups of a space are (in a certain range) nite or nitely
generated or that induced maps have nite or nitely generated kernel and cokernel.
A famous result of Serre [170] says that the homotopy groups of spheres are nite,
except in the cases already known to Hopf.
Here are three basic ideas of Serres approach:
(1) Properties like nite or nitely generated or rational isomorphism have
a formal structure. Only this structure matters and it is axiomatized in the
notion of a Serre class of abelian groups or modules.
(2) One has to relate homotopy groups and homology groups, since qualitative
results about homology groups are more accessible. The connection is based
on the Hurewicz homomorphism.
(3) For inductive proofs one has to relate the homology groups of the basis, bre,
and total space of a (Serre-)bration. This is the point where Serre uses the
method of spectral sequences.
Anon-empty class C of modules over a commutative ring 1is a Serre class if the
following holds: Let 0 T C 0 be an exact sequence of 1-modules.
Then T C if and only if . C C. We call C saturated if C implies that
arbitrary direct sums
}
of copies of are contained in C. The class consisting
of the trivial module alone is saturated. A morphism : M N between 1-
modules is a C-epimorphism(C-monomorphism) if the cokernel (kernel) of is in
C, and a C-isomorphism if it is a C-epi- and -monomorphism. We use certain facts
about these notions, especially the C-Five Lemma. The idea is to neglect modules
in C, or, as one says, to work modulo C; so, instead of C-isomorphism, we say
isomorphism modulo C. Here are some examples of Serre classes.
(1) The class containing only the trivial group.
(2) The class F of nite abelian groups.
(3) The class G of nitely generated abelian groups.
(4) Let 1 be a principal ideal domain. The class C consists of the (nitely
generated) 1-modules. If 1 is a eld, then we are considering the class of
(nite-dimensional) vector spaces.
(5) Let 1 be a principal ideal domain. The class C consists of the (nitely
generated) 1-torsion modules. A module M is a torsion module, if for each
. M there exists 0 = z 1 such that z. = 0.
(6) Let 1 N be a set of prime numbers. Let Z
1
Q denote the subring
of rational numbers with denominators not divisible by an element of 1. If
1 = 0, then Z = Q. If 1 = {], then Z
1
= Z
(])
is the localization
of Z at . If 1 contains the primes except , then Z
1
= Z
-1
| is the
20.4. Qualitative Homology of Fibrations 505
ring of rational numbers with denominators only -powers. The rings Z
1
are principal ideal domains. Let 1
t
be the complementary set of primes. An
abelian group is a 1
t
-torsion group if and only if
Z
Z
1
= 0.
Let C
1
be the class of 1
t
-torsion groups. Then a homomorphism : T
is a C
1
-isomorphism if and only if
Z
Z
1
is an ordinary isomorphism.
Similarly for epi- and monomorphism. This remark reduces the C
1
Five
Lemma to the ordinary Five Lemma after tensoring with Z
1
. This simplies
working with this class
1
.
(20.3.1) Proposition (Five Lemma mod C). In the next proposition we use the
same notation as in (11.2.7). The considerations of that section then yield directly
a proof of the following assertions.
(1) b C-epimorphism =

b C-epimorphism.
(2) J C-epimorphism, e C-monomorphism =

J C-epimorphism.
(3) The hypotheses of (1) and (2) imply: c C-epimorphism.
(4) J C-monomorphism =

J C-monomorphism.
(5) a C-epimorphism, b C-monomorphism =

b C-monomorphism.
(6) The hypotheses of (4) and (5) imply: c C-monomorphism.
The kernel-cokernel-sequence shows other properties of C-notions.
(20.3.2) Proposition. Given homomorphisms : T and g: T C between
1-modules.
(1) If and g are C-monomorphisms (-epimorphisms), then g is a C-mono-
morphism (-epimorphism).
(2) If two of the morphisms , g, and g are C-isomorphisms so is the third.
20.4 Qualitative Homology of Fibrations
In this section we work with singular homology with coefcients in the 1-mo-
dule M.
(20.4.1) Theorem(Fibration Theorem). Let : 1 T be a (Serre-) bration with
0-connected bres. Let (T. ) be a relative CW-complex with t -skeleton T
t
. We
assume that = T
-1
= T
0
= = T
x-1
for an s _ 0 and that there are only
a nite number of t -cells for t _ e. Finally we assume that H
i
(J: M) C for
0 < i < r and all bres J of . We write 1
t
=
-1
(T
t
). Then the following
holds:
(1) Let C be saturated. Then
+
: H
i
(1. 1
-1
: M) H
i
(T. T
-1
: M) is a C-
isomorphism for i _ r s 1 = and a C-epimorphism for i = 1.
1
In a more abstract setting one can construct localizations of categories so that C-isomorphisms
become isomorphisms in the localized category, etc.
(2) Let C be arbitrary. Then
+
is a C-isomorphism for i _ and a C-epi-
morphism for i = 1 where now = min(e 1. r s 1).
We remark that r _ 1. For r = 1 we make no further assumptions about J.
Since a weak homotopy equivalence induces isomorphisms in singular homo-
logy, we can assume without essential restriction that T is a CW-complex (pull
back the bration along a CW-approximation). We reduce the proof of the theorem
by a Five Lemma argument to the attaching of t -cells. We consider the following
situation. Let
(. ): l
o
(D
t
o
. S
t-1
o
) (T. T
t
)
be an attaching of t -cells. Let : 1 T be a bration and set 1
t
=
-1
(T
t
).
We assume that the bres are 0-connected and homotopy equivalent. We pull back
the bration along and obtain two pullback diagrams.
l1
o
I]
u
1
]
l1
t
o
I]
/
u
1
t
]
/
lD
t
o
T lS
t-1
o
T
t
We apply homology (always with coefcients in M) and obtain the diagram
H
i
(1. 1
t
)
]

H
i
(T. T
t
)
o
H
i
(1
o
. 1
t
o
)

u
]
u

o
H
i
(D
t
o
. S
t-1
o
).
We already know that

+
is an isomorphism. In order to show that
+
is an
isomorphism, we attach a single t -cell, to simplify the notation. Let T
0
be obtained
from T by deleting the center (0) of the cell.
(20.4.2) Lemma. Let : X T be a bration with restrictions
t
: X
t
T
t
and
0
: X
0
T
0
. Let q : Y D
t
be the pullback of , and similarly q
t
: Y
t
S
t-1
and q
0
: Y
0
D
t
0. Then
+
: h
i
(Y. Y
t
) h
i
(X. X
t
)
is an isomorphism for each homology theory.
Proof. We have a commutative diagram
h
i
(Y. Y
t
)

(1)
h
i
(X. X
t
)
(2)
h
i
(Y. Y
0
)

h
i
(X. X
0
)
h
i
(Y Y
t
. Y
0
Y
t
)
(3)
(5)
h
i
(X X
t
. X
0
X
t
).
(4)
20.4. Qualitative Homology of Fibrations 507

Since T
t
is a deformation retract and a bration, also X
t
is a deformation retract
of X
0
. Therefore (2) is an isomorphism by homotopy invariance. The map (4) is
an isomorphism by excision. For similar reasons (1) and (3) are isomorphisms.
Finally (5) is induced by a homeomorphism.
By the homotopy theorem for brations, a bration over D
t
is bre-homotopy
equivalent to a product projection D
t
J D
t
. We use such equivalences and a
suspension isomorphismH
i
(D
t
J. S
t-1
J) H
i-t
(J) and obtain altogether
a commutative diagram (1
o
a point and J
o
the bre over 1
o
):
H
i
(1. 1
t
)
]

H
i
(T. T
t
)
o
H
i-t
(J
o
)
o
H
i-t
(1
o
).
(20.4.3) Note. The considerations so far showthat the bottommap has the following
properties:
(1) Isomorphism for i _ t (since bres are 0-connected).
(2) Epimorphism always.
(3) Suppose H
i
(J
o
) C for 0 < i < r. Then each particular map H
i-t
(J
o
)
H
i-t
(1
o
) is a C-isomorphism for 0 < i t < r. Thus the total map is
a C-isomorphism if either C is saturated or if we attach a nite number of
cells.
We apply these considerations to a bration : 1 T over a relative CW-
complex (T. ) as in the statement of the theorem. In this situation the previous
considerations yield:
Let C be saturated. Then
+
: H
i
(1
t
. 1
t-1
) H
i
(T
t
. T
t-1
) is a C-iso-
morphism for each t _ 0, if i < r s. We only have to consider t _ s. By
(20.4.3), we have a C-isomorphism in the cases i _ t and i > t > i r. These
conditions hold for each t , if s > i r. If, in addition, there are only nitely many
t -cells for t _ e, then we have a C-isomorphism for arbitrary C and each t , if
i _ min(e 1. r s 1), by the same argument.
We nish the proof of theorem (20.4.1) with:
(20.4.4) Lemma. Let (T. ) be a relative CW-complex with t -skeleton T
t
. Let
: 1 T be a bration and 1
t
=
-1
(T
t
). Suppose
+
: H
i
(1
t
. 1
t-1
: M)
H
i
(T
t
. T
t-1
: M) is a C-isomorphism for each t _ 0 and each 0 _ i _ , then
+
: H
i
(1. 1
-1
: M) H
i
(T. T
-1
: M) is a C-isomorphism for 0 _ i _ and a
C-epimorphism for i = 1.
Proof. We show by induction on k _ 0 that
+
: H
i
(1
k
. 1
-1
) H
i
(T
k
. T
-1
)
is a C-isomorphism (k _ 0). For the induction step one uses the exact homology
sequence for the triple (1
k
. 1
k-1
. 1
-1
), and similarly for T, and applies the C-
Five Lemma to the resulting diagram induced by the various morphisms
+
. For
the epimorphism statement we also use part (2) in (20.4.3). The colimit k o
causes no problem for singular homology.
The statement of (20.4.1) is adapted to the method of proof. The hypotheses
can be weakened as follows.
(20.4.5) Remark. Let (X. ) be an (s 1)-connected pair of spaces. Then there
exists a weak relative homotopy equivalence (T. ) (X. ) from a relative CW-
complex (T. ) with = T
x-1
. We pull back a bration over q : 1 X along
this equivalence and use the fact that weak equivalences induce isomorphisms in
singular homology. Then part (1) of (20.4.1) yields that q
+
: H
}
(1. q
-1
(): M)
H
}
(X. : M) is a C- isomorphism for _ r s 1 and a C-epimorphism for
i = r s.
(20.4.6) Remark. Let X be a 1-connected space such that H
}
(X: Z) is nitely
generated for i _ e. Then there exists a weak equivalence T X such that T has
only a nite number of t -cells for t _ e. We use this result in the next section.
Problems
1. Suppose that H
(J: M) = 0 for 0 < < r. Let T be (s 1)-connected. Then
: H
(1. J: M) H
(T. +: M) is an isomorphism for _ r s 1. We can now

insert this isomorphism into the exact homology sequence of the pair (1. J) and obtain an
exact sequence
H
:-1
(J: M) H
1-1
(1: M) H
:-1
(T: M)
H
1
(J: M) H
1
(1: M) H
1
(T: M) 0
which is analogous to the exact sequence of homotopy groups. Compare these sequences
via the Hurewicz homomorphism (M = Z).
20.5 Consequences of the Fibration Theorem
We use exact sequences and the bration theorem to derive a number of results.
We consider a bration : 1 T and assume that T and J =
-1
(+) are
0-connected; then 1 is 0-connected too. We use the notation
7 C(r. M) = H
}
(7: M) C for 0 < < r _ o.
(M an 1-module. In the case that M = Z we write C(r). For r = 1 there is no
condition.) Let F , G denote the class of nite, nitely generated abelian groups,
respectively. We use homology with coefcients in the 1-module M if nothing
else is specied.
20.5. Consequences of the Fibration Theorem 509
(20.5.1) Remark. The homomorphism
+
: H
}
(1. J: M) H
}
(T. +: M) is al-
ways an isomorphism for _ 1 and an epimorphism for = 2 (see (20.4.1)). The
homomorphism
+
: H
}
(1: M) H
}
(T: M) is an isomorphism for = 0 and
an epimorphism for = 1.
(20.5.2) Theorem. Suppose J C(r. M). Let C be saturated. Suppose T is
s-connected. Then
+
: H
i
(1. J) H
i
(T. +) is a C-isomorphism for i _ r s
and a C-epimorphism for i = r s 1. Moreover
+
: H
i
(1) H
i
(T) is a
C-isomorphism for i < r and a C-epimorphism for i = r.
Proof. The rst statement is (20.4.1). For the second statement we use in addition
the exact homology sequence of the pair (1. J).
(20.5.3) Theorem. Let C be saturated.
(1) J. T C(r. M) =1 C(r. M).
(2) J C(r. M). 1 C(r 1. M) =T C(r 1. M).
(3) T C(r 1. M). 1 C(r. M), T 1-connected =J C(r. M).
Proof. (1) and (2) are consequences of the bration theoremand the exact homology
sequence of the pair (1. J). (3) is proved by induction on r. For r = 1 there is
nothing to prove. For the induction step consider
H
i
(T. +: M) H
i
(1. J: M)

(1)
H
i-1
(J: M)

H
i-1
(1: M)
(1) is a C isomorphism for r 1 s 1 = r, since s = 2 and J C(r 1. M)
by induction. From the hypotheses H
i
(T: +: M). H
i-1
(1: M) C we conclude
H
i-1
(J: M) C.
(20.5.4) Theorem. Let C be arbitrary and assume that T is 1-connected.
(1) J. T C(r. M). T G(r 1) =1 C(r. M).
(2) J C(r. M). 1 C(r 1. M). T G(r) =T C(r 1. M).
(3) T C(r 1. M). 1 C(r. M). T G(r) =J C(r. M).
Proof. As for (20.5.3). The 1-connectedness of T is needed in order to apply the
cell theorem (see (20.4.6)).
(20.5.5) Corollary. Suppose 1 is contractible (path bration over T). Then
J . CT, the loop space of T. Let T be simply connected. Let C be saturated.
Then T C(r 1. M) if and only if CT C(r. M). Moreover T G(r 1) if
and only if CT G(r). Similarly for F instead of G.
(20.5.6) Proposition. Let be a nitely generated abelian group. Then the
EilenbergMac Lane spaces 1(. n) are contained in G(o). If is nite, then
1(. n) F (o). Moreover 1(. 1) G(o). 1(. 1) C(o. M) implies
1(. n) C(o. M). If C is saturated, then 1(. 1) C(o. M) implies
1(. n) C(o. M).
Proof. We use the path bration 1(. n 1) 1 1(. n) with contractible
1 and induction with (20.5.3) and (20.5.4). In the case that n = 1, standard
constructions yield models for 1(. 1) with a nite number of cells in each di-
mension. One uses 1(
1
. 1) 1(
2
. 1) = 1(
1

2
. 1), 1(Z. 1) = S
1
, and
1(Z,m. 1) = S
o
,(Z,m).
Let X be a (k 1)-connected space (k _ 2). We attach cells of dimension
_ k 2 to X in order to kill the homotopy groups
}
(X) for _ k 1. The
resulting space in an EilenbergMac Lane space 1(. k), =
k
(X), and the
inclusion | : X 1(. k) induces an isomorphism
k
(|). We pull back the path
bration over 1(. k) and obtain a bration
1(. k 1) Y X
with k-connected Y , and q : Y X induces isomorphisms
}
(q), > k. This
follows from the exact homotopy sequence. If C and C is saturated, then
q
+
: H
}
(Y : M) H
}
(X: M) is a C-isomorphism for > 0, by the bration
theorem, since 1(. k 1) C(o). Similarly for arbitrary C when X is of nite
type.
20.6 Hurewicz andWhiteheadTheorems moduloSerre classes
Let C be a Serre class of abelian groups with the additional property: The groups
H
k
(1(. 1)) C whenever C and k > 0. In this section we work with
integral singular homology.
(20.6.1) Theorem (Hurewicz Theorem mod C). Suppose X is 1-connected and
n _ 2. Assume that either C is saturated or H
i
(X) is nitely generated for i < n
and C is arbitrary. Then the following assertions are equivalent:
(1) (n):
i
(X. +) C for 1 < i < n.
(2) H(n): H
i
(X) C for 1 < i < n.
If (n) or H(n) holds, the Hurewicz homomorphism h
n
:
n
(X. +) H
n
(X) is
a C-isomorphism.
Proof. The proof is by induction on n.
(1) Let CX 1X
(
X be the path bration with contractible 1X. It
provides us with a commutative diagram
n
(X. +)
h
n
(1X. CX. +)
t
n
(( )
n-1
(CX. +)
h
H
n
(X. +) H
n
(1X. CX)
1
n
(( )
H
n-1
(CX).
20.6. Hurewicz and Whitehead Theorems modulo Serre classes 511
The boundary maps d are isomorphisms, since 1X is contractible. By a basic
property of brations
n
( ) is an isomorphism.
By the ordinary Hurewicz theoremh
2
:
2
(X. +) H
2
(The method of the following proof can also be used to prove the classical Hurewicz
theorem.)
(2) Let n > 2. Assume that the theoremholds for n1 and that (n) holds. We
want to showthat H(n) holds and that h
n
: (X. +) H
n
(X) is a C-isomorphism.
We rst consider the special (3) case that
2
(X) = 0 and then reduce the general
case (4) to this special.
(3) Thus let
2
(X) = 0. Then
i
(CX)
i1
(X) C for 1 < i < n 1,
and CX is 1-connected. If C is saturated, then h
n-1
:
n-1
(CX) H
n-1
(CX)
is a C-isomorphism by induction. If H
i
(X) is nitely generated for i < n, then
by (20.5.3) H
i
(CX) is nitely generated for i < n 1 so that by induction h
n-1
is also an isomorphism in this case. The bration theorem shows that H
n
( )
in the diagram is a C-isomorphism. From the diagram we now see that h
n
is a
C-isomorphism. Also H
n-1
(X) C by induction.
(4) Let now
2
(X) =
2
be arbitrary. By assumption, this group is contained
in C. There exists a map ; : X 1(
2
. 2) which induces an isomorphism
2
.
We pull back the path bration along ;
X
2

11(
2
. 2)
(
X
;
1(
2
. 2).
Since
2
C, we have H
i
(1(
2
. 1)) C for i > 0, by the general assumption
in this section. Note that 1 = 1(
2
. 1) is the bre of and . The exact
homotopy sequence of ; is used to show that
+
:
i
(X
2
)
2
(X) for i > 2 and
that
1
(X
2
) 0
2
(X
2
). We can therefore apply the special case (3) to X
2
.
Consider the diagram
n
(X
2
)
h
n
(A
2
)

n
(X)
h
n
(A)
H
n
(X
2
)
C

H
n
(X).
In order to show that h
n
(X) is a C-isomorphism we show two things:
(i) h
n
(X
2
) is a C-isomorphism.
(ii)
+
: H
n
(X
2
) H
n
(X) is a C-isomorphism.
Part (i) follows from case (3) if we know that H
i
(X
2
) is nitely generated for
i < n. This follows from (20.5.3) applied to the bration 1 X
2
X, since X
is 1-connected and since
1
(1)
2
H
2
(X) is nitely generated.
For the proof of (ii) we rst observe that the canonical map : H
i
(X
2
)
H
i
(X
2
. 1) is a C-isomorphism for i > 0, since H
i
(1) C for i > 0 by the
general assumption of this section. The bration theorem and (20.4.6) show that
+
: H
i
(X
2
. 1) H
i
(X. +) is a C-isomorphism for 0 < i _ n. We now compose
with and obtain (ii).
(5) Now assume that H(k) holds for 2 _ k < n. Since H
i
(X) C for
2 _ i < k 1 the Hurewicz map
i
(X) H
i
(X) is a C-isomorphism for i _ k
by H(k), hence
i
(X) C for i _ k. By the rst part of the proof, h
k1
is a
C-isomorphism, hence (k 1) holds.
We list some consequences of the Hurewicz theorem. Note that the general
assumption of this section holds for the classes G and F .
(20.6.2) Theorem. Let X be a 1-connected space.
(1)
i
(X) is nitely generated for i < n if and only if H
i
(X: Z) is nitely
generated for i < n.
(2)
i
(X) is nite for i < n if and only if

H
i
(X: Z) is nite for i < n.
(3) If X is a nite CW-complex, then its homotopy groups are nitely generated.
(20.6.3) Theorem. Let C be a saturated Serre class. Let : X Y be a map

between1-connectedspaces with1-connectedhomotopy bre J. Thenthe following
are equivalent:
(1)
k
( ):
k
(X)
k
(Y ) is a C-isomorphism for k < n and a C-epimor-
phism for k = n.
(2) H
k
( ): H
k
(X) H
k
(Y ) is a C-isomorphism for k < n and a C-epimor-
phism for k = n.
Proof. The statement (1) is equivalent to
k
(J) C for k < n (exact homotopy
sequence). Suppose this holds. Then H
k
(X. J) H
k
(Y ) is a C isomorphism for
k _ n, by the bration theorem (20.4.1). From the exact homology sequence of
the pair (X. J) we now conclude that (2) holds, since H
}
(J) C by the Hurewicz
theorem. Here we use that
1
(J) = 0.
Suppose (2) holds. We show by induction that H
k
(J) C for k < n. Then we
apply again the Hurewicz theorem. The induction starts with n = 3. Since Y and
J are simply connected, the bration theorem shows that H
}
(X. J) H
}
(Y. +)
is a C-isomorphism for _ 3. The assumption (2) and the homology sequence
of the pair then show H
2
(J) C. The general induction step is of the same type.
Problems
1. Let T Q be a subring. Let X be a simply connected space such that H
n
(X: T )
H
n
(S
n
: T ). Then there exists a map S
n
X which induces an isomorphism in T -homo-
logy.
20.7. Cohomology of EilenbergMac Lane Spaces 513
2. The nite CW-complex X = S
1
S
2
has
1
(X) Z. The group
2
(X) is free abelian
with a countably innite number of generators. (Study the universal covering of X.)
3. Use the bration theoremand deduce for each 0-connected space a natural exact sequence
2
(X) H
2
(X) H
2
(1(
1
(X). 1): Z) 0.
20.7 Cohomology of EilenbergMac Lane Spaces
We compute the cohomology ring H
+
(1(Z. n): Q). Let : S
n
1(Z. n) =
1(n) induce an isomorphism
n
( ) of the n-th homotopy groups. Then also
+
: H
n
(1(Z. n): Q) H
n
(S
n
: Q) is an isomorphism and |
n
is dened such
that
+
(|
n
) H
n
(S
n
: Q) is a generator.
(20.7.1) Theorem. If n _ 2 is even, then H
+
(1(Z. n): Q) Q|
n
| (polynomial
ring). If n is odd, then
+
: H
+
(1(Z. n): Q) H
+
(S
n
: Q).
Proof. We work with rational cohomology. Since 1(Z. 1) . S
1
and 1(Z. 2) .
C1
o
we know already the cohomology ring for these spaces with coefcients in
Zand this implies (20.7.1) in these cases. We prove the theorem by induction on n,
and for this purpose we analyze the path-bration 1(Z. n 1) 1 1(Z. n)
with contractible 1. There are two cases for the induction step, depending on the
parity of n.
2k 1 = 2k. We have a relative bration : (1. 1
t
) 1(n) with 1 =
1(n)
J
. (n) = n(1). 1 = 1
t
=
-1
(+). The map : 1 1(n) is a homotopy
equivalence and 1
t
is contractible. Therefore
(1) H
n
(1. 1
t
) H
n
(1) H
n
(1(n)).
the latter induced by . The bres (J. J
t
) of have a contractible J and J
t
=
C1(n) = 1(n 1) is by induction a rational cohomology (n 1)-sphere (i.e.,
has the rational cohomology of S
n-1
). Hence H
k
(J. J
t
) H
k-1
(J
t
) Q for
k = n and 0 for k ,= n. Since 1(n) is simply connected, we have a Thom class
t
n
H
n
(1. 1
t
). We can assume that t
n
is mapped under (1) to |
n
, hence |
n
is the
Euler class e associated to t
n
. The Gysin sequence has the form (n = 2k)
H
}
(1(n))
e
H
}n
(1(n)) H
}n
(1) .
Since 1 is contractible and H
}
(1(n)) = 0 for 0 < < n, we see induc-
tively that the cup product with the Euler class e is an isomorphism H
}
(1(n))
H
}2k
(1(n)). Hence H
+
(1(n)) Q|
n
|.
2k = 2k 1. We reduce the problem to a Wang sequence. Let n = 2k 1.
We consider a pullback
Y

q
1
]
S
n
(
1(n),
where induces an isomorphism
n
( ). The Wang sequence for q has the form
H
}
(Y )
i
H
}
(1(n 1))
0
H
}1-n
(1(n 1)) .
We use H
+
(1(n 1)) Q|
n-1
| and the fact that 0 is a derivation. From the
denition of Y and the exact sequence of homotopy groups we see
}
(Y ) = 0. _ n.
}
(Y )
}
(S
n
). > n.
From the Hurewicz theorem and the universal coefcient theorem we conclude
that H
}
(Y ) = 0 for _ n. Hence 0 : H
2k
(1(n 1)) H
0
(1(n 1))
is an isomorphism. Using the derivation property of 0 we see inductively that
0 : H
2ki
(1(n 1)) H
2k(i-1)
(1(n 1)) is an isomorphism, and the Wang
sequence then shows us that

H
+
(Y ) = 0; and this implies

H
+
(Y ) = 0. Since Y
is the homotopy bre of , we conclude that
+
: H
+
(S
n
) H
+
(1(n)) is an
isomorphism (by (20.4.1) say), and similarly for cohomology. This nishes the
induction.
20.8 Homotopy Groups of Spheres
Let n > 1 be an odd integer. Let : S
n
1(Z. n) induce an isomorphism
n
( )
and denote by Y the homotopy bre of . In the previous section we have shown
that H
}
(Y : Q) = 0 for > 0. The Hurewicz theorem modulo the class of torsion
groups (=the rational Hurewicz theorem) shows us that the groups
}
(Y ) Qare
zero for N. From
}
(S
n
)
}
(Y ) for > n we see that also
}
(S
n
)Q = 0
for > n and odd n. Since we already know that the homotopy groups of spheres
are nitely generated we see:
(20.8.1) Theorem. Let n be an odd integer. Then the groups
}
(S
n
) are nite for
> n.
We now investigate the homotopy groups of S
2n
. Let V = V
2
(R
2n1
) denote
the Stiefel manifold of orthonormal pairs (.. ,) in R
2n1
. We have a bre bundle
S
2n-1
V S
2n
, and V is the unit-sphere bundle of the tangent bundle of S
2n
.
Recall from 14.2.4 the integral homology of V
H
q
(V )
Z. q = 0. 4n 1.
Z,2. q = 2n 1.
0. otherwise.
Let g: V S
4n-1
be a map of degree 1. Then g induces an isomorphism in
rational homology. Let J be the homotopy bre of g; it is simply connected.
From (20.6.3) we see that g
+
Q:
}
(V ) Q
}
(S
4n-1
) Q is always an
20.8. Homotopy Groups of Spheres 515
isomorphism. We use (20.8.1) and see that the homotopy groups of V are nite,
except
4n-1
(V ) Z 1, 1 nite. Now we go back to the bration V S
2n
and its homotopy sequence. It shows:
(20.8.2) Theorem.
}
(S
2n
) is nite for = 2n. 4n1 and
4n-1
(S
2n
) Z1,
1 nite.
The results about the homotopy groups of spheres enable us to prove a rened
rational Hurewicz theorem.
(20.8.3) Theorem. Let X be 1-connected. Suppose H
i
(X: Z) is nite for i < k
and nitely generated for i _ 2k 2 (k _ 2). Then the Hurewicz homomorphism
h:
i
(X) H
i
(X: Z) has nite kernel (cokernel) for r < 2k 1 (r _ 2k 1).
Proof. The case k = 2 causes no particular problem, since the Hurewicz homo-
morphism h:
n1
(X) H
n1
(X: Z) is surjective for each (m1)-connected
space (m _ 2). So let k _ 3. Since X is 1-connected, the Hurewicz theoremshows
that
i
(X) is nite for r < k and nitely generated for r _ 2k 2. We write
i
(X) = J
i
T
i
, J
i
free, and T
i
nite. We choose basis elements for J
i
and
representing maps. These representing elements provide us with a map of the form
: S = S
i(1)
S
i(t)
X
with k _ r( ) _ 2k 2. The canonical map
}

i
(S
i(})
)
i
_ _
}
S
n(})
_
is an isomorphism for r _ 2k 2 and an epimorphism for r = 2k 1. We can now
conclude that
+
:
i
(S)
i
(X) has nite kernel and cokernel for r _ 2k 2
and nite cokernel for r = 2k 1, since the homotopy groups of spheres are nite
in the relevant range. The homotopy bre J of has nite homotopy groups
}
(J) for _ 2k 2. If, moreover, J is 1-connected, then H
}
(J: Z) is nite in
the same range, by the Hurewicz theorem. The bration theorem then yields that
+
: H
i
(S) H
i
(X) is an F -isomorphism (F -epimorphism) for r _ 2k 2
(r = 2k 1). From our knowledge of the homotopy groups of spheres we see
directly that the theorem holds for S. The naturality of the Hurewicz theorem
applied to is now used to see that the desired result also holds for X.
We have used that J is 1-connected. This holds if
2
(X) = 0. Since
2
(X) is
nite by assumption (k _ 3), we can pass to the 2-connected cover q : X(2) X
of X. The map q induces F -isomorphisms in homology and homotopy. Therefore
it sufces to prove the theorem for X(2) to which the reasoning above applies.
If one is not interested in nite generation one obtains by a similar reasoning

(see also [103]):
(20.8.4) Theorem. Let X be a 1-connected space. Suppose H
i
(X: Q) = 0 for
r < k. Then the Hurewicz map
}
(X) Q H
}
(X: Q) is an isomorphism for
< 2k 2 and an epimorphism for = 2k 1.
This theorem indicates that homotopy theory becomes simpler over the ratio-
nals. In the so-called rational homotopy theory one constructs algebraic models
for the rationalized homotopy theory. For an exposition see [65].
We discuss an example. Consider the path bration
1(Z. 2) . C1(Z. 3) X 1(Z. 3) = 1
3
with contractible X. Let : S
3
1
3
induce an isomorphism in
3
and let
1
2
i
Y
]
S
3
be the induced bration. The homotopy groups
k
(Y ) are zero
for k _ 3 and
+
k
(Y )
k
(S
3
) for k _ 4.
3
(S
2
) Z and
n1
(S
n
) Z,2 for n _ 3.
Proof. We know already that
3
(S
2
) Z, generated by the Hopf map S
3
C1
1
S
2
. Fromthe Freudenthal suspensiontheoremwe knowthat the suspension
+
:
n1
(S
n
)
n2
(S
n1
) is surjective for n = 2 and bijective for n _ 3.
Therefore it sufces to determine
4
(S
3
). By the Hurewicz theorem,
4
(S
3
)
4
(Y ) H
4
(Y ). Thus it remains to compute H
4
(Y ). We rst determine the
cohomology.
(20.8.6) Proposition. The cohomology groups H
k
(Y ) of Y are: Z for k = 0,
0 for k 0 mod 2, and Z,n for k = 2n 1.
Proof. We use 1
2
= 1(Z. 2) = C1
o
and H
+
(1
2
) Zc| with c H
2
. The
Wang sequence of Y S
3
shows that
H
2n
(Y ) Ker . H
2n1
(Y ) Coker .
since H
+
(1
2
) = 0 for odd +. The group H
2n
(1
2
) Zis generated by c
n
. By the
universal coefcient formula H
}
(Y ) = 0 for = 1. 2. 3. Hence : H
2
(1
2
)
H
0
(1
2
) is an isomorphism. We can choose c such that (c) = 1. The derivation
property of yields then inductively (c
n
) = nc
n-1
.
The Wang sequence in homology yields in a similar manner
H
2n
(Y ) Coker
+
. H
2n1
(Y ) Ker
+
.
From the universal coefcient sequence
0 Ext (H
2n
(Y ). Z) H
2n1
(Y ) Hom(H
2n1
(Y ). Z) 0
and the fact that H
2n
(Y ) is a quotient of Z we obtain
H
2n
(Y ) Z,n. H
2n1
(Y ) = 0.
The special case H
4
(Y ) Z,2 now proves
4
(S
3
) Z,2.
20.8. Homotopy Groups of Spheres 517
We apply the Hurewicz theorem modulo the class of abelian -torsion groups
( prime) and see that the -primary component of
i
(S
3
) is zero for 3 < i < 2
and isomorphic to Z, for i = 2. In particular an innite number of homotopy
groups
n
(S
3
) is non-zero.
The determination of the homotopy groups of spheres is a difcult problem. You
can get an impression by looking into [163]. Individual computations are no longer
so interesting; general structural insight is still missing. Since the groups
nk
(S
n
)
do not change after suspension for k _ n2, by the Freudenthal theorem, they are
called the stable homotopy groups
S
k
.
We copy a table from [185]; a denotes a cyclic group of order a, and a b is
the product of cyclic groups of order a and b, and a
}
the -fold product of cyclic
groups of order a.
k 0 1 2 3 4 5 6 7 8 9
S
k
o 2 2 24 0 0 2 240 2
2
2
3
k 10 11 12 13 14 15 16 17 18 19
S
k
6 504 0 3 2
2
480 2 2
2
2
4
8 2 264 2
20.8.7 The Hopf invariant. The exceptional case
4n-1
(S
2n
) is interesting in
many respects. Already Hopf constructed a homomorphism h:
4n-1
(S
2n
) Z,
nowcalled the Hopf invariant, and gave a geometric interpretation in the simplicial
setting [88]. Let : S
4n-1
S
2n
be a smooth map; and let a, b be two regular
values. The pre-images M
o
=
-1
(a) and M
b
=
-1
(b) are closed orientable
(2n 1)-manifolds, they have a linking number, and this number is the Hopf
invariant of . It is easy to dene h( ), using cohomology. Let : S
2k-1
S
k
be given (k _ 2). Attach a 2k-cell to S
k
by and call the result X = X( ).
The inclusion i : S
k
X induces an isomorphism H
k
(X) H
k
(S
k
), and we
also have an isomorphism H
2k
(D
2k
. S
2k-1
) H
2k
(X. S
k
) H
2k
(X). The
integral cohomology groups H
}
(X) for ,= 0. k. 2k are zero. Choose generators
. H
k
(X) and , H
2k
(X). Then there holds a relation . . = h( ), in the
cohomology ring. The graded commutativity of the cup product is used to show
that h( ) = 0 for odd k. In the case k = 2n the integer h( ) is the Hopf invariant.
Since X( ) depends up to h-equivalence only on the homotopy class of , the
integer h( ) is a homotopy invariant. One shows the elementary properties of this
invariant:
(1) h is a homomorphism.
(2) If g: S
4n-1
S
4n-1
has degree J, then h(g) = Jh( ).
(3) If k : S
2n
S
2n
has degree J, then h(k ) = J
2
h( ).
Note that a map of degree J does not induce the multiplication by J, as opposed to
the general situation for cohomology theories.
It is an important problem to determine the image of h. Already Hopf showed
by an explicit construction that 2Z is always contained in the image. Here is the
Hopf construction. Start with a map u: S
k
S
k
S
k
. From it we obtain a map
: S
2k1
S
k1
via the diagram
S
k
S
k
1
uid
S
k
1
]
S
k
= S
k
(
S
k
,
where q is the projection onto the suspension and the projection onto the join. The
map u has a bi-degree (a. b). Hopf shows that (with suitable orientations chosen)
h( ) = ab. The map S
2n-1
S
2n-1
S
2n-1
, (.. ) (.. ). has bi-
degree (2. 1). If : R
n
R
n
R
n
is a bilinear map without zero divisors (i.e.,
(.. ,) = 0 implies that . or , is zero), then (R
n
. ) is called an n-dimensional
real division algebra. The induced map : S
n-1
S
n-1
S
n-1
, (.. ,)
(.. ,),[(.. ,)[ satises (c.. j,) = cj(.. ,) for c. j {1]. Hence has
a bi-degree (J
1
. J
2
) with odd J
}
. If there exist maps with odd Hopf invariant, then
there also exist maps with invariant 1, since 2Z is contained in the image of h. It
is a famous result of Adams [2] that maps : S
4n-1
S
2n
of Hopf invariant 1
only exist for n = 1. 2. 4. Hence there exist n-dimensional real division algebras
only for n = 1. 2. 4. 8. See [55] for this topic and the classical algebra related to
it. Once complex 1-theory is established as a cohomology theory, it is fairly easy
to solve the Hopf invariant 1 problem [5].
20.9 Rational Homology Theories
Recall the n-th stable homotopy group o
n
(X) = colim
k

nk
(X . S(k)) of the
pointed space X. The Hurewicz homomorphisms are compatible with suspension,
i.e., the diagram
n
(Y )

H
n
(Y )
n1
(Y )
n

H
n1
(Y )
is commutative for each well-pointed space Y . This follows from the inductive
denition of the Hurewicz homomorphisms; one has to use the same denition of
+
in homotopy and homology via the boundary operator of the pair (CY. Y ) and
the quotient map CY CY,Y = Y . We pass to the colimit and obtain the
20.9. Rational Homology Theories 519
stable Hurewicz homomorphism
h
n
: o
n
(X)

H
n
(X).
The h
n
are natural transformations of homology theories (on well-pointed spaces).
In order to see this one has to verify that the diagram
o
n
(X)
h
n
H
n
(X)
o
n1
(X . S(1))
h
n1

H
n1
(X . S(1))
is commutative. The commutativity of this diagram is a reason for introducing the
sign in the denition o = (1)
n
t
+
o
I
of the suspension isomorphism for spectral
homology.
The coefcients of the theory o
+
() are the stable homotopy groups of spheres
o
n
(S
0
) = colim
k

nk
(S
k
). These groups are nite for n > 0. Finite abelian
groups become zero when tensored with the rational numbers. We thus obtain a
natural transformation of homology theories
h
+
: o
+
(X)
Z
Q

H
+
(X)
Z
Q

H
+
(X: Q)
which are isomorphisms on the coefcients and therefore in general for pointed
CW-complexes.
This basic rational isomorphism is now used to show that any homology theory
h
+
with values in Q- MODcan be reduced to ordinary homology. We dene natural
maps
o
]
(X)
Z

h
q
(S
0
)

h
]q
(X).
Let . o
]
(X) be represented by : S( k) X . S(k). The image of
,

h
q
(S
0
) under
h
q
(S
0
)

h
q]k
(S( k))
(

h
q]k
(X . S(k))

h
]q
(X)
is independent of the chosen representative of .. We combine these homomor-
phisms
]q=n
o
]
(X)
Z

h
q
(S
0
)

h
n
(X)
and obtain a natural transformation of homology theories. Now assume that the
coefcients

h
q
(S
0
) are Q-vector spaces. Then for X = S
0
only the groups
o
0
(S
0
)
Z

h
n
(S
0
) are non-zero; and the induced map to

h
n
(S
0
) is an isomor-
phism, since o
0
(S
0
) Z by the degree and a map of degree k induces on

h
n
(S
0
)
the multiplication by k. We thus have shown:
(20.9.1) Theorem. Let

h
+
be an additive homology theory for pointed CW-com-
plexes with values in Q-vector spaces. Then we have an isomorphism
]q=n

H
]
(X:

h
q
(S
0
))
]q=n

H
]
(X)

h
q
(S
0
)
h
n
(X)
]q=n
o
]
(X)

h
q
(S
0
)
of homology theories.
If

k
+
() is an arbitrary additive homology theory we can apply the foregoing
to

h
+
() =

k
+
()
Z
Q.
Problems
1. The EilenbergMac Lane spectrum(1(Z. n) [ n N
0
) yields a homology theory which
is isomorphic to singular homology with integral coefcients.
2. One can dene the stable Hurewicz transformation from a morphism of the sphere spec-
trum into the EilenbergMac Lane spectrum which is obtained from maps S
n
1(Z. n)
that induce isomorphisms of
n
.
3. Dene a stable Hurewicz homomorphism h
n
: o
n
(X)

H
n
(X: Z) either from a map
of spectra or by an application of cohomology to representing maps of elements in o
n
(X).
Construct a natural commutative diagram
o
n
(X)
h
n
H
n
(X)
Hom(o
n
(X). Z)
(h
n
)

Hom(

H
n
(X). Z).
4. Give a proof of the (absolute) Hurewicz theorem by using the 1(Z. n)-denition of
homology. The proof uses: Let
(X) = 0 for < n _ 2; then
(X . 1(Z. k)) = 0 for

_ n k 1 and
nk
(X . 1(Z. k))
n
(X).
5. Derive an isomorphism

h
n
(X)

)q=n

H
)
(X: h
q
(S
0
)) for cohomology theories
with values in Q-vector spaces (X a nite pointed CW-complex).
6. Use a bration 1(Z. n) 1(Z. n) 1(Z,k. n) and derive a universal coefcient
sequence for homology with Z,k-coefcients.
7. An interesting example of a rational cohomology isomorphism is given by the Chern
character. It is a natural isomorphism of Q-algebras
ch: 1(X) Q
n
H
2n
(X: Q).
for nite complexes X, say, and which sends a complex line bundle j over X to the power
series
e
c
1
())
=
c
i=0
1
i
c
1
(j)
i
H
2
(X: Q).
Chapter 21
Bordism
We begin with the denition of bordism homology. The geometric idea of homol-
ogy is perhaps best understood from the view-point of bordism and manifolds. A
singular cycle is a map from a closed manifold to a space, and the boundary rela-
tion is induced by manifolds with boundary. Several of our earlier applications of
homology and homotopy can easily be obtained just from the existence of bordism
homology, e.g., the Brouwer xed point theorem, the generalized Jordan separation
theorem and the component theorem, and the theorem of BorsukUlam.
Bordism theory began with the fundamental work of Thom [184]. He deter-
mined the bordism ring of unoriented manifolds (the coefcient ring of the asso-
ciated bordism homology theory). This computation was based on a fundamental
relation between bordism and homotopy theory, the theorem of PontrjaginThom.
In the chapter on smooth manifolds we developed the material which we need for
the present proof of this theorem. One application of this theorem is the isomor-
phism between the geometric bordism theory and a spectral homology theory via
the Thom spectrum. From this reduction to homotopy we compute the rational
oriented bordism. Hirzebruch used this computation in the proof of his signature
theorem. This proof uses almost everything that we developed in this text.
21.1 Bordism Homology
We dene the bordism relation and construct the bordism homology theory. Mani-
folds are smooth.
Let X be a topological space. An n-dimensional singular manifold in X is a pair
(T. J) which consists of a compact n-dimensional manifold T and a continuous
map J : T X. The singular manifold d(T. J) = (dT. J[dT) is the boundary
of (T. J). If dT = 0, then (T. J) is closed.
Anull bordismof the closed singular manifold (M. ) in X is a triple (T. J. )
which consists of a singular manifold (T. J) in X and a diffeomorphism : M
dT such that (J[dT) = . If a null bordismexists, then (M. ) is null bordant.
Let (M
1
.
1
) and (M
2
.
2
) be singular manifolds in X of the same dimension.
We denote by (M
1
.
1
) (M
2
.
2
) the singular manifold (
1
.
2
): M
1
M
2

X. We say (M
1
.
1
) and (M
2
.
2
) are bordant, if (M
1
.
1
) (M
2
.
2
) is null
bordant. Anull bordismof (M
1
.
1
)(M
2
.
2
) is called bordismbetween(M
1
.
1
)
and (M
2
.
2
). The boundary dT of a bordism (T. J. ) between (M
1
.
1
) and
(M
2
.
2
) thus consists of a disjoint sum d
1
T d
2
T, and decomposes into two
diffeomorphisms
i
: M
i
d
i
T.
522 Chapter 21. Bordism
(21.1.1) Proposition. Bordant is an equivalence relation.
Proof. Let (M. ) be given. Set T = M1 and J = pr : M1 M X.
Then dT = M 0 M 1 is canonically diffeomorphic to M M and (T. J)
is a bordism between (M. ) and (M. ). The symmetry of the relation is a direct
consequence of the denition. Let (T. J.
i
: M
i
d
i
T) be a bordism between
(M
1
.
1
) and (M
2
.
2
) and (C. G. [
i
: M
i
d
i
C) a bordism between (M
2
.
2
)
and (M
3
.
3
). We identify in T C the subset d
2
T with d
2
C via . ~ [
2
-1
2
(.)
for . d
2
T. The result D carries a smooth structure, and the canonical maps T
D C are smooth embeddings (15.10.1). We can factor (J. G): T C X
over the quotient map T C D and get H: D X, and (D. H. (
1
. [
3
)) is
a bordism between (M
1
.
1
) and (M
3
.
3
).
We denote by M. | the bordism class of (M. ) and by N
n
(X) the set
of bordism classes of n-dimensional closed singular manifolds in X. The set
N
n
(X) carries an associative and commutative composition lawM. | N. g| =
M N. (. g)|. The reader may check that this is well-dened.
(21.1.2) Proposition. (N
n
(X). ) is an abelian group. Each element has order at
most 2.
Proof. The class of a null bordant manifold serves as neutral element, for ex-
ample the constant map S
n
X. (For the purpose at hand it is convenient to
think of the empty set as an n-dimensional manifold.) For each (M. ) the sum
(M M. (. )) is null bordant, hence M. | M. | = 0.
A continuous map : X Y induces a homomorphism
N
n
( ) =
+
: N
n
(X) N
n
(Y ). M. g| M. g|.
In this way N
n
() becomes a functor from TOP toABEL. Homotopic maps induce
the same homomorphism: If J : X 1 Y , . g is a homotopy, then (M 1.
J (hid)) is a bordismbetween (M. h) and (M. gh). If X is empty, we consider
N
n
(X) as the trivial group.
(21.1.3) Example. A 0-dimensional compact manifold M is a nite discrete set.
Hence (M. ) can be viewed as a family (.
1
. . . . . .
i
) of points in X. Points
.. , X are bordant if and only if they are contained in the same path component.
(Here you have to know 1-dimensional compact manifolds.) One concludes that
N
0
(X) is isomorphic to the Z,2-vector space over
0
(X).
(21.1.4) Proposition. Let h: 1 1be a diffeomorphism. Then 1. g| = 1. gh|.
Proof. Consider the bordism g pr : 1 1 X; on the boundary piece 1 1
we use the canonical diffeomorphism to 1, on the boundary piece we 1 0 we
compose the canonical diffeomorphism to 1 with h.
21.1. Bordism Homology 523
We now make the functors N
n
() part of a homology theory. But this time, for
variety, we do not begin with the denition of relative homology groups. The exact
homology sequence and the excision axiom are now replaced by a MayerVietoris
sequence.
Suppose X is the union of open sets X
0
and X
1
. We construct the boundary
operator
d: N
n
(X) N
n-1
(X
0
X
1
)
of the MayerVietoris sequence. Let M. | N
n
(X) be given. The sets M
i
=
-1
(X X
i
) are disjoint closed subsets of M.
(21.1.5) Lemma. There exists a smooth function : M 0. 1| such that:
(1) M
i

-1
(i ) for i {0. 1].
(2)
1
2
is a regular value of .
We call in (21.1.5) a separating function. If is a separating function, then
M
=
-1
(
1
2
) is a closed submanifold of M of dimension n 1 (or empty), and
induces by restriction
: M
X
0
X
1
.
If t = 0. 1 is another regular value of , then
-1
(t ) and
-1
(
1
2
) are bordant
via
-1
1
2
. t |. The choice of
1
2
is therefore immaterial. We think of M
| as
being given by any choice of a regular value t |0. 1 of with M
=
-1
(t ).
(21.1.6) Lemma. Let 1. | = 1. g| N
n
(X) and let . be separating func-
tions for (1. ). (1. g). Then 1
| = 1
. g
|.
Proof. We take advantage of (21.1.4). Let (T. J) be a bordism between (1. )
and (1. g) with dT = 1 1. There exists a smooth function ; : T 0. 1| such
that
;[1 = . ;[1 = . J
-1
(X X
}
) ;
-1
( ).
We choose a regular value t for ; and ;[dT and obtain a bordism ;
-1
(t ) between
some 1
and some 1
.
From (21.1.6) we obtain a well-dened boundary homomorphism
d: N
n
(X) N
n-1
(X
0
X
1
). M. | M
|.
(21.1.7) Proposition. Let X be the union of open subspaces X
0
and X
1
. Then the
sequence

J
N
n
(X
0
X
1
)
}
N
n
(X
0
) N
n
(X
1
)
k
N
n
(X)
J

is exact. Here (.) = (
0
+
(.).
1
+
(.)) and k(,. :) = k
0
+
,k
1
+
: with the inclusions
: X
0
X
1
X
and k
: X
X. The sequence ends with

k
N
0
(X) 0.
Proof. (1) Exactness at N
n-1
(X
0
X
1
). Suppose M. | N
n
(X) is given. Then
M is decomposed by M
into the parts T

0
=
-1
0.
1
2
| and T
1
=
-1
1
2
. 1| with
common boundary M
. Since (T
0
) X
1
, we see via T
0
that
1
+
dM. | is in
N
n-1
(X
1
) the zero element. This shows d = 0.
Suppose, conversely, that 1. | = 0. Then there exist singular manifolds
(T
i
. J
i
) in X
i
such that dT
0
= 1 = dT
1
and J
0
[1 = = J
1
[1. We identify T
0
and T
1
along 1and obtainM; the maps J
0
and J
1
canbe combined toJ : M X.
There exists a separating function on M such that M
= 1: With collars of 1
in T
0
and T
1
one obtains an embedding 1 0. 1| M which is the identity on
1 {
1
2
]; then one chooses such that (k. t ) = t for k 1,
1
4
_ t _
3
4
. By
construction, dM. J| = 1. |.
(2) Exactness at N
n
(X
0
) N
n
(X
1
). By denition, k = 0. Suppose
.
i
= M
i
.
i
| N
n
(X
i
) are given. If k(.
0
. .
1
) = 0 there exists a bordism
(T. J) in X between (M
0
. k
0
0
) and (M
1
. k
1
1
). Choose a smooth function
[: T 0. 1| such that:
(1) J
-1
(X X
1-i
) L M
i
[
-1
(i ) for i = 0. 1.
(2) [ has regular value
1
2
.
Let (N. ) = ([
-1
(
1
2
). J[[
-1
(
1
2
)). Then ([
-1
0.
1
2
|. J[[
-1
0.
1
2
|) is a bordism
between (N. ) and (M
0
.
0
) in X
0
; similarly for (M
1
.
1
). This shows N. | =
(.
0
. .
1
).
(3) Exactness at N
n
(X). The relation d k = 0 holds, since we can choose on
(M
0
. k
0
0
) (M
1
. k
1
1
) a separating function : M
0
M
1
0. 1| such that
-1
(
1
2
) is empty.
Conversely, let be a separating function for (M. ) in X and (T. J) a null
bordism of (M
) in X
0
X
1
. We decompose M along M
into the manifolds

T
1
=
-1
0.
1
2
| and T
0
=
-1
1
2
. 1| with dT
1
= M
= dT
0
. Then we identify T
and T
0
along M
= 1 and obtain a singular manifold (M

0
.
0
) = (T
0
L
1
T.
( [T
0
) L
1
J) in X
0
, and similarly (M
1
.
1
) in X
1
. Once we have shown that
in N
n
(X) the equality M
0
.
0
| M
1
.
1
| = M. | holds, we have veried the
exactness. We identify in M
0
1 M
1
1 the parts T1 in M
0
1 and M
1
1. The
resulting manifold 1 = (M
0
1)L
B1
(M
1
1) has the boundary (M
0
M
1
)M.
A suitable map J : 1 X is induced by (
0
.
1
) pr
1
: (M
0
M
1
) 1 X.
For the smooth structure on 1 see 15.10.3.
We now dene relative bordism groups N
n
(X. ) for pairs (X. ). Elements
of N
n
(X. ) are represented by maps : (M. dM) (X. ) from a compact n-
manifold M. Again we call (M. ) = (M. dM: ) a singular manifold in (X. ).
The bordism relation is a little more complicated. A bordism between (M
0
.
0
)
and (M
1
.
1
) is a pair (T. J) with the following properties:
(1) T is a compact (n 1)-manifold with boundary.
(2) dT is the union of three submanifolds with boundary M
0
, M
1
and M
t
, where
dM
t
= dM
0
dM
1
. M
i
M
t
= dM
i
.
(3) J[M
i
=
i
.
(4) J(M
t
) .
We call (M
0
.
0
) and (M
1
.
1
) bordant, if there exists a bordism between them.
Again bordant is an equivalence relation. For the proof one uses 15.10.3. The
sum in N
n
(X. ) is again induced by disjoint union. Each element in N
n
(X. )
has order at most 2. A continuous map : (X. ) (Y. T) induces a homo-
morphism N
n
( ) =
+
: N
n
(X. ) N
n
(Y. T) by composition with . If
0
and
1
are homotopic as maps between pairs, then N
n
(
0
) = N
n
(
1
). The as-
signment M. | dM. [dM| induces a homomorphism (boundary operator)
d: N
n
(X. ) N
n-1
(). For = 0 the equality N
n
(X. 0) = N
n
(X) holds.
(21.1.8) Lemma. Let M be a closed n-manifold and V M an n-dimensional
submanifold with boundary. If : M X is a map which sends M V into ,
then M. | = V. [V | in N
n
(X. ).
Proof. Consider J : M 1 X, (.. t ) (.). Then d(M 1) = M d1 and
V 1 LM 0 is a submanifold of d(M 1) whose complement is mapped under
J into . The denition of the bordism relation now yields the claim.
(21.1.9) Proposition. Let i : X and : X = (X. 0) (X. ). Then the
sequence

J
N
n
()
i
N
n
(X)
}
N
n
(X. )
J

}
N
0
(X. ) 0.
Proof. (1) Exactness at N
n
(). The relation i
+
d = 0 is a direct consequence
of the denitions. Let (T. J) be a null bordism of : M in X. Then
dT. J| = M. |.
(2) Exactness at N
n
(X). Let M. | N
n
() be given. Choose V = 0 in
(21.1.8). Then M. | = 0 in N
n
(X. ), and this shows
+
i
+
= 0.
Let
+
M. | = 0. Anull bordismof M. | in (X. ) is a bordismof (M. ) in
X to (1. g) such that g(1) . A bordism of this type shows i
+
1. g| = M. |.
(3) Exactness at N
n
(X. ). The relation di
+
= 0 is a direct consequence of the
denitions. Let dM. | = 0. Choose a null bordism T. J| of (dM. [dM). We
identify (M. ) and (T. ) along dM and obtain (C. g). Lemma (21.1.8) shows
+
C. g| = M. |.
A basic property of the relative groups is the excision property. It is possible to
give a proof with singular manifolds.
(21.1.10) Proposition. The inclusion i : (X U. U) (X. ) induces an
isomorphism i
+
: N
n
(X U. U) N
n
(X. ), provided

U
.
The bordismnotion can be adapted to manifolds with additional structure. Inter-
esting are oriented manifolds. Let M
0
and M
1
be closed oriented n-manifolds. An
oriented bordism between M
0
. M
1
is a smooth compact oriented (n1)-manifold
T with oriented boundary dT together with an orientation preserving diffeomor-
phism : M
1
M
0
dT. Here we have to use the convention about the boundary
orientation, and M
1
M
0
denotes the disjoint sum of the manifolds M
1
and M
0
where M
1
carries the given and M
0
the opposite orientation. Again this notion of
bordism is an equivalence relation. Singular manifolds are dened as before, and
we have bordismgroups C
n
(X) of oriented bordismclasses of singular n-manifolds
in X. But now elements in the bordism group no longer have order at most 2. For
a point 1 we have C
0
(1) = Z, C
i
(1) = 0 for 1 _ i _ 3. The assertion about
C
1
follows from the fact that S
1
is an oriented boundary; the known classication
of orientable surfaces as a sphere with handles shows that these surfaces are ori-
ented boundaries. It is a remarkable result that C
3
(1) = 0: Each oriented closed
3-manifold is an oriented boundary; for a proof of this theorem of Rohlin see [77].
The exact sequences (21.1.7) and (21.1.9) as well as (21.1.10) still hold for
the C-groups. The denition of the boundary operator d: C
n
(X. ) C
n-1
()
uses the boundary orientation. In order to dene the boundary operator of the
MV-sequence we have to orient M
. There exists an open neighbourhood U of

M
in M and a diffeomorphism : V =|1,2 c. 1,2 c M
U such that
()(t. .) = t . If M is oriented, we have the induced orientation of U, and we
orient V such that preserves the orientation. We orient M
such that V carries

the product orientation.
The idea of bordismcan be used to acquire an intuitive understanding of homol-
ogy. A compact n-manifold M has a fundamental class :
H
n
(M. dM: F
2
) and
d:
H
n-1
(dM: F
2
) is again a fundamental class. Let : (M. dM) (X. )
be a singular n-manifold. We set j( ) =
+
:
H
n
(X. : F
2
). In this manner
we obtain a well-dened homomorphism
j: N
n
(X. ) H
n
(X. : F
2
).
The morphisms j constitute a natural transformation of homology theories. One
of the basic results of bordism theory says that j is always surjective. This allows
us to view homology classes as being represented by singular manifolds. If, in
particular, is an embedding of manifolds, then we view the image of as a cycle
or a homology class. In bordism theory, the fundamental class of M is M itself,
i.e., the identity of M considered as a singular manifold.
The transformation j can be improved if we take tangent bundle information
into account. Let M be a compact n-manifold and denote by k
: M TO
the classifying map of the stable tangent bundle of M. For a singular manifold
: (M. dM) (X. ) we then have (. k
): (M. dM) (X. ) TO. Again

we take the image of the fundamental class
j | = (. k
)
+
:
H
n
((X. ) TO: F
2
).
We now obtain a natural transformation of homology theories
j = j(X. ): N
+
(X. ) H
+
((X. ) TO: F
2
).
and in particular for the coefcient ring, the Thom bordism ring N
+
of unoriented
manifolds,
j: N
+
H
+
(TO: F
2
).
A fundamental result says that j(X. ) is always injective [28, p. 185].
This transformation is also compatible with the multiplicative structures. The
algebra structure of H
+
(X TO: F
2
) is induced by the homology product and the
H-space structure m: TO TO TO which comes from the Whitney sum of
bundles. We obtain a natural homomorphism of graded algebras,
H
+
(X TO: F
2
) H
+
(Y TO: F
2
) H
+
(X TO Y TO: F
2
)
H
+
(X Y TO: F
2
):
the rst map is the homology product and the second is induced by the permutation
of factors and m.
Thom [184] determined the structure of the ring N
+
: It is a graded algebra
F
2
u
2
. u
4
. u
5
. . . . | with a generator u
k
in each dimension k which does not have the
form k = 2
t
1. One can take u
2n
= R1
2n
| as generators in even dimensions.
The ring H
+
(TO: F
2
) is isomorphic to F
2
a
1
. a
2
. a
3
. . . . | with a generator a
i
in
dimension i .
Another basic result says that there exists a natural isomorphism N
+
(X)
N
+
F
2
H
+
(X: F
2
) of multiplicative homology theories [160], [28, p. 185]. Thus
the homology theory N
+
() can be reduced to the determination of the coefcient
ring N
+
and singular homology with F
2
-coefcients.
For oriented manifolds the situation is more complicated. One can still dene
a multiplicative natural transformation of homology theories
j
D
: C
+
(X. ) H
+
((X. ) TSO: Z)
from the fundamental classes of oriented manifolds as above. But this time the
transformation is no longer injective and C
+
(X. ) H
+
(X. : Z) in general not
surjective. Also the theory C
+
() cannot be reduced to ordinary homology. But the
transformation still carries a lot of information. It induces a natural isomorphism
C
+
(X. ) Q H
+
((X. ) TSO: Q).
and, in particular,
(1) C
+
Q H
+
(TSO: Q)
by the stable classifying map of the tangent bundle (see (21.4.2)). The ring C
+
Q
is isomorphic to Q.
4
. .
S
. . . . | with a generator .
n
for each n 0(4). One can
take the .
4n
= C1
2n
| as polynomial generators (see (21.4.4)).
We have seen that the signature of oriented smooth manifolds denes a homo-
morphism o : C
+
Q Qof Q-algebras with oC1
2n
| = 1. The isomorphism
(1) tells us that this homomorphism can be determined from the stable tangent bun-
dle. Afamous formula of Hirzebruch [81], the so-called Hirzebruch 1-genus, gives
a polynomial 1
n
(
1
.
2
. . . . .
n
) in the Pontrjagin classes
1
. . . . .
n
such that the
evaluation on the fundamental class of an oriented 4n-manifold is the signature
(1
n
(
1
(M). . . . .
n
). M| ) = o(M).
It is a remarkable fact that the polynomials 1
n
have rational coefcients with large
denominators, but nevertheless the evaluation on the fundamental class is an integer.
Such integrality theorems have found a conceptual interpretation in the index theory
of Atiyah and Singer [16].
The homomorphisms N
n
H
n
(TO: F
2
) and C
n
H
n
(TSO: Z) have an in-
terpretation in terms of characteristic numbers and can be determined by evaluating
polynomials in the StiefelWhitney classes or Pontrjagin classes on the fundamental
class.
In order to prove some of the results above one starts from the fundamen-
tal results of Thom [184], the reduction of bordism to homotopy theory via the
PontrjaginThom construction.
Problems
1. C
0
(X) is naturally isomorphic to the free abelian group over
0
(X).
2. Give a proof of (21.1.10) with singular manifolds.
3. We had dened formally a boundary operator for the MV-sequence from the axioms of a
homology theory. Show that the boundary operator dened with separating functions coin-
cides with this formal boundary operator. Pay attention to signs in the oriented case.
4. The homology theories C
and N
have a product structure. Products of singular mani-

folds induce a bilinear map
C
m
(X. ) C
n
(Y. T) C
mn
(X Y. X T L Y ).
Verify the formal properties of a multiplicative structure, in particular the stability axiom.
5. Let M
1
and M
2
be oriented d-manifolds which are glued together along a component
N
i
dM
i
with a diffeomorphism : N
1
N
2
. Let M be the result. There exists an
orientation of M such that the canonical embeddings M
i
M are orientation preserving,
provided reverses the boundary orientations of the N
i
.
A collar k : R
-
dM M of an oriented manifold M is orientation preserving, if
R
-
R carries the standard orientation, dM the boundary orientation and R
-
dM the
product orientation.
In order to verify the transitivity of the oriented bordism relation one has to dene the
orientation of the bordism which is obtained by gluing the given bordisms in such a way that
the given bordisms are oriented submanifolds of the glued bordism.
6. The construction of the bordism MV-sequence suggests another set of axioms for a ho-
mology theory. Aone-space homology theory consists of a family h
n
: TOP 1- MODof
21.2. The Theorem of Pontrjagin and Thom 529
covariant homotopyinvariant functors anda familyof boundaryoperators d: h
n
(X
0
LX
1
)
h
n-1
(X
0
X
1
) for each triad (X: X
0
. X
1
) in which the X
are open in X. The bound-

ary operators are assumed to be natural with respect to maps of triads and the usual exact
MV-sequence should hold for each such triad.
Given these data, one denes relative groups for a pair (X. ) by
h
n
(X. ) = Coker(h
n
(C) h
n
(X L C))
where as usual C is the cone on . The relative groups are homotopy-invariant functors
on TOP(2). Dene a boundary operator d: h
n
(X. ) h
n-1
() and show that the usual
sequence of a pair is exact. In order to derive this sequence, consider the MV-sequence for the
triad (X LC: X LCX. XLC+). The excision isomorphismh
n
(X U. U)
h
n
(X. ) holds, provided there exists a function t : X 0. 1| with U t
-1
(0) and
t
-1
0. 1 , since under this assumptionthe canonical map(XU)LC(U) XLC
is a pointed h-equivalence.
21.2 The Theorem of Pontrjagin and Thom
The theorem of Pontrjagin and Thom relates homotopy theory and manifold theory.
It describes sets of bordism classes as homotopy sets. We begin by dening the
ingredients of the theorem.
Let Q be a smooth manifold without boundary. We denote by Q
c
= QL{o]
its one-point compactication. Let : 1() T be a smooth real vector bundle
over a closed manifold T. The one-point compactication M() = 1() L{o] is
called the Thom space of . The points at innity serve as base points. The pointed
homotopy set Q
c
. M()|
0
will be described as a bordism set.
A-submanifold of Qis a closed submanifold M together with a smooth bundle
map J : 1(v) 1() from its normal bundle
v = v
= v(M. Q): 1(v) M

into the given bundle . A bordism between two submanifolds (M
0
. J
0
) and
(M
1
. J
1
) is a compact submanifold W of Q 1 (of type I) such that
W (Q 0. 1,3) = M
0
0. 1,3 . W (Q|2,3. 1|) = M
1
|2,3. 1|
and dW = M
0
0 L M
1
1 together with a bundle map J : v(W. Q 1)
which extends J
0
and J
1
. Note that v(W. Q 1)[M 0 can be identied with
v(M
0
. Q). The relation of -bordism is an equivalence relation. We denote by
1(Q. ) the set of -bordism classes.
We dene a map 1 : 1(Q. ) Q
c
. M()|
0
, and the main theorem then as-
serts that 1 is a bijection. We choose an embedding of Qinto some Euclidean space.
Then Qinherits a Riemannian metric. The normal bundle v(M. Q): 1(v) M is
the orthogonal complement of TM TQ[M. From these data we had constructed
a tubular map t : 1(v) Q. The bundle v inherits a Riemannian metric, and
D
t
(v). 1
t
(). S
t
(v) are the subsets of vectors of norm[[ _ c. [[ < c. [[ =
c, respectively. There exists an c > 0 such that t embeds 1
2t
() onto an open neigh-
bourhood U
2t
of M in Q. We have a brewise diffeomorphismh
t
: 1
t
() 1()
dened on each bre by
h
t
(.) =
c.
_
c
2
[.[
2
.
From these data we construct from a -submanifold (M. J) a map
g
t
: Q
c
Q
c
,(Q
c
U
t
) D
t
(v),S
t
(v) M(v) M().
The rst map is the quotient map, the second a homeomorphism induced by the
inclusion D
t
(v) Q (note that D
t
,S
t
is a one-point compactication of 1
t
), the
third induced by h
t
, and the fourth induced by the bundle map J (a proper map).
The pointed homotopy class of g
t
is independent of the chosen (sufciently small)
c, by a linear homotopy in the bre. Let us set 1(M. J) = g
t
| Q
c
. M()|
0
.
We say that a map g
t
is obtained by a PontrjaginThom construction.
(21.2.1) Lemma. For -bordant manifolds (M
}
. J
}
) the classes 1(M
}
. J
}
) are
equal. Therefore we obtain a well-dened map 1 : 1(Q. ) Q
c
. M()|
0
.
Proof. We apply a PontrjaginThom construction to a -bordism. For this purpose
we use for Q 1 the product embedding. Let (W. J) be a -bordism. We obtain
a tubular map t : 1(v
W
) Q 1 which is over M
0
0. 1,3 the product of the
tubular map for M
0
with 0. 1,3, and similarly for the other end. For sufciently
small c, again t embeds 1
2t
(v
W
) onto a neighbourhood of W and we dene as
above a PontrjaginThom map
Q
c
1 Q
c
1,(Q
c
1 U
t
) D
t
(v
W
),S
t
(v
W
) M(v
W
) M().
and this map is a homotopy between the PontrjaginThom maps for (M
0
. J
0
) and
(M
1
. J
1
).
(21.2.2) Theorem (Pontrjagin, Thom). The PontrjaginThom map
1 : 1(Q. ) Q
c
. M()|
0
is a bijection.
Proof. We construct a map in the other direction. Let us observe that the maps
: Q
c
M() obtained by the PontrjaginThom construction are of a very
special type. They have the following properties:
(1) The map :
-1
(1()) 1() is proper, smooth and transverse to the
zero section T 1().
(2) There exists a tubular neighbourhood U
t
of M =
-1
(T) in Q such that
(.) = o=. U
t
.
(3) The map t h
-1
t
: 1(v
) 1
t
(v
) U
t
1() is a smooth bundle
map.
So we have to deform an arbitrary map into one having these three properties.
(i) Let g: Q
c
M() be given and set = g
-1
(1()). Then is open
in Q and g: 1() is a proper map. By the approximation theorem (15.8.4)
there exists a proper homotopy of g to a smooth map g
1
: 1(). Restrict
g
1
to a compact neighbourhood V of g
-1
1
(T) in such that V is a manifold and
g
1
(dV ) 1() T. By the transversality theorem (15.9.8) we nd a smooth
homotopy of g
1
[V to a map which is transverse to T and such that the homotopy
is constant in a neighbourhood of dV . We can therefore extend this homotopy
to a smooth proper homotopy of g
1
by a constant homotopy in the complement
of V . Since both homotopies are proper, they can be extended continuously to Q
c
by mapping the complement of to the base point. The result is a smooth map
g
2
: Q
c
M() which has property (1) above.
(ii) Let M = g
-1
2
(T). Let now c be small enough such that the tubular neigh-
bourhood U
2t
of M is contained in . Let : Q
c
0. 1| be a continuous function
which is smooth on Qand such that
-1
(0) = D
t{2
and
-1
0. 1 = U
t
. We dene
a homotopy of g
2
by
H
t
(.) =
_
(1 t(.))
-1
g
2
(.). . . t < 1: . U
t
. t = 1.
o. otherwise.
The map g
3
= H
1
has properties (1) and (2) above with U = U
t
.
(iii) Let = g
3
be a map obtained in step (ii). Consider the composition
th
-1
t
= h: 1(v
) 1
t
(v
) U
t
1().
This map is proper, smooth, and transverse to T 1(). The homotopy H
t
(.) =
t
-1
h(t.), dened for t > 0 can be extended to t = 0 by a bundle map : v

such that the resulting homotopy is smooth and proper. The map is the derivative
in the direction of the bres. In order to see what happens in the limit t 0, we
express h in local coordinates. Then h has the form
X R
n
Y R
n
. (.. ) (a(.. ). b(.. ))
with open sets X M, Y T and a(.. 0) = (.). b(.. 0) = 0. Then
H
t
(.. ) = (a(.. t ). t
-1
b(.. t )). The map lim
t-0
t
-1
b(.. t ) is the
differential of b
x
: T(.. ) at = 0. It is a bijective linear map, because is
transverse to the zero section.
(iv) We nowconstruct a map Qwhich is inverse to 1. Suppose g: Q
c
M()
is a map such that g: g
-1
1() 1() is proper, smooth, and transverse to the
zero section. Let M = g
-1
(T). Then the differential of g induces a smooth bundle
map J : v
. The map Q sends g| to the bordism class of the -submanifold

(M. J). By (i) we know that each homotopy class has a representative g with
the properties just used. If g
0
and g
1
are two such representatives, we choose a
homotopy between themwhich is constant on 0. 1,3 and on |2,3. 1| and apply the
method of (i) in order to obtain a homotopy h: Q
c
1 M(). g
0
. g
1
such that
h: h
-1
1() 1() is proper, smooth, transverse to the zero section, and constant
on 0. j and |1 j. 1|. The pre-image of the zero section and the differential of h
yield a -bordism. This argument shows that Q is well-dened. By construction,
Q1 is the identity. The arguments of (i)(iii) show that 1 is surjective.
(21.2.3) Example. Let Q = M be a closed connected n-manifold and the n-
dimensional bundle over a point. Then 1() = R
n
and M() = S
n
= R
n
L{o]. A
-submanifold of Qis a nite subset X together with an isomorphismJ
x
: T
x
M
R
n
for each . X. In the present situation Q
c
. M()
0
| = M. S
n
|. We are
therefore in the situation of the Hopf degree theorem.
Let M be oriented. If J
x
is orientation preserving, we set c(.) = 1, and
c(.) = 1 otherwise. Let c(X. J) =

xA
c(.) Z. The integer c(X. J)
characterizes the -bordism class. If we represent a homotopy class in M. S
n
| by
a smooth map : M S
n
with regular value 0 R
n
, then X =
-1
(0) and
J
x
= T
x
and c(X. J) is the degree of , as we have explained earlier. If we show
that c(X. J) characterizes the bordism class, then the PontrjaginThom theorem
gives a proof of the Hopf degree theorem for smooth manifolds.
If M is non-orientable, we have a similar situation, but this time we have to
consider c(X. J) modulo 2.
(21.2.4) Example. Let Q = R
nk
and 1() = R
n
. Then Q
c
. M()|
0
=
nk
(S
n
), but we disregard the group structure for the moment. A -submani-
fold is in this case a closed k-manifold M R
kn
together with a trivialization
of the normal bundle. A trivialization of a vector bundle is also called a framing
of the bundle. Since the normal bundle v is inverse to the tangent bundle t we see
that the tangent bundle is stably trivial and a framing v nc of the normal bundle
induces TM nc TM v (n k)c, a stable framing of M.
We denote by o
n
(k) the bordism set of a closed n-manifold with framing
TM kc (n k)c. The bordism relation is dened as follows. Let W be
a bordism between M
0
and M
1
and let : T W (k 1)c (n k)c be a
framing. Let |
0
: T W[M
o
TM
0
c where the positive part of c corresponds
to an inwards pointing vector. Similarly |
1
: T W[M
1
TM
1
c where now the
positive part corresponds to an outwards pointing vector. We obtain a framing
i
: TM
i
c (k 1)c T W[M
i
(k 1)c

(n k)c
of M
i
. We say in this case: (W. ) is a framed bordism between (M
0
.
0
) and
(M
1
.
1
).
The assignment M.
| M.
r
| is a well-dened map 1(R
nk
. kc)
o
n
(k). It is bijective for k > n 1.
In the PontrjaginThom theorem it is not necessary to assume that is a smooth
bundle over a closed manifold. In fact, can be an arbitrary bundle. In this case
we have to dene the Thom space in a different manner. Let have a Riemannian
metric; then we have the unit disk bundle D() and the unit sphere bundle S(). We
dene the Thom space now as the quotient space M() = D(),S(). A denition
that does not use the Riemannian metric runs as follows. The multiplicative group
R
+
of positive real numbers acts on the subset 1

0
() of non-zero vectors bre-
wise by scalar multiplication. Let S() be the orbit space with induced projection
s
: S() T. The mapping cylinder of s
is a space J
: D() T over T and

M() is dened to be the (unpointed) mapping cone of s
. From this denition we

see that a bundle map : j induces a pointed map M( ): M() M(j). In
the category of compactly generated spaces we have a canonical homeomorphism
M( j) M() .M(j). If j is the trivial one-dimensional bundle over a point,
this homeomorphism amounts to M( c) M() .S
(1)
with S
(1)
= RL{o].
We can now dene as before -submanifolds of Q and -bordisms. Also the
PontrjaginThom construction can be applied in this situation, and we obtain a
well-dened map
1 = 1
: 1(Q. ) Q
c
. M()|
0
.
These maps constitute a natural transformation between functors from the category
of n-dimensional bundles and bundle maps.
(21.2.5) Theorem (Pontrjagin, Thom). The PontrjaginThom map 1
is for each
bundle a bijection.
Proof. The proof is essentially a formal consequence of the special case (21.2.2),
based on the general techniques developed so far.
(i) In the proof of (21.2.2) we used smooth bundles . A bundle over a closed
manifold T is induced froma tautological bundle over some Grassmannian G
n
(R
1
)
by some map . The map is homotopic to a smooth map g and the bundle induced
by g is therefore smooth and isomorphic to . This fact allows us to work with
arbitrary bundles in (21.2.2). The PontrjaginThomconstruction itself does not use
a smooth bundle map.
(ii) Suppose i : X Y and r : Y X are maps with ri = id (a retraction). If
1 is bijective for bundles over Y and is a bundle over X, we pull back this bundle
to j = r
+
. From the naturality of 1 and the fact that 1
)
is bijective, we conclude
that 1
is bijective.
(iii) Let C be a compact smooth manifold with boundary. Let T denote the
double D(C) of C. Then C is a retract of T. Hence 1 is bijective for bundles
over C.
(iv) Let X be a nite CW-complex. A nite CW-complex is a retract of some
open set U R
n
. Choose a proper smooth function t : U R
such that
t (X) = {0]. Let > 0 be a regular value of t . Then X is a retract of the compact
smooth manifold t
-1
0. | with boundary. Therefore the theorem holds for bundles
over nite CW-complexes.
(v) Suppose is a bundle over a CW-complex. By compactness of Q
c
we see
that a map Q
c
M() has an image in the Thom space of the restriction of the
bundle to a nite subcomplex. This shows, using (iv), that 1
is surjective. An
analogous argument shows the injectivity.
(vi) If T is an arbitrary space we choose a CW-approximation : C T
and let j =
+
. One shows that the bundle map j induces a bijection
1(C. ) 1(T. j). This is due to the fact that a bundle map v is up to
homotopy the composition of a bundle map v j with the bundle map j
(manifolds have the homotopy type of a CW-complex), and homotopic bundle maps
yield, via the PontrjaginThom construction, homotopic maps. A similar argument
shows that a bijective map Q
c
. M(j)|
0
Q
c
. M()|
0
is induced.
Problems
1. Work out the classication of the -bordism classes in (21.2.3), thus completing the
sketched proof of the Hopf degree theorem.
2. Give a proof of (21.2.4). The source of the map in question uses embedded manifolds,
the range abstract manifolds. Thus one has to use the Whitney embedding theorem.
3. We use o
1
(k) to interpret the isomorphism
k1
(S
k
) Z,2 for k _ 3. Let (. )
represent an element of o
1
(k). The manifold is a disjoint sum of manifolds
diffeo-
morphic to S
1
. Let
be the framing of
induced by . We assign to (
. ) an element
J(
) Z,2. Let h: S
1
be a diffeomorphism. We think of S
1
as boundary of
D
2
and give D
2
R
2
the standard framing. This induces a standard framing o of TS
1
c
in which 1 c points outwards. This provides us with a framing
; : T kc
Thid
TS
1
c
oid
(k 1)c.
If is framed by , then : T kc (k 1)c orients the bundle T. We choose the
diffeomorphismh such that the composition with ; is orientation preserving. The homotopy
class of ; is independent of the chosen h. The framing differs from the standard framing
by a map
GL
k1
(R) . SO(k 1).
Composed with h
-1
we obtain a well-dened element in
S
1
. SO(k 1)|
1
SO(k 1) Z,2. k _ 2.
denoted by J(. ) Z,2. If consists of the components (
[ J) we set J(. ) =
J(
).
Showthat (. ) J(. ) induces for k _ 2 a well-dened isomorphismJ : o
1
(k)
Z,2.
21.3. Bordism and Thom Spectra 535
4. Give a similar interpretation of
3
(S
2
) Z. The difference to the case of the previous
problem is due to
1
SO(2) Z.
21.3 Bordism and Thom Spectra
The theorem of PontrjaginThom allows us to describe the bordism group N
n
(X)
as a homotopy group.
Let ;
n
: 1(;
n
) TO(n) be a universal n-dimensional real vector bundle and
MO(n) = M(;
n
) its Thom space. A classifying map ;
n
c ;
n1
induces a
pointed map e
n
: M(;
n
) M(;
n
c) M(;
n1
). The Thom spectrum MO
consists of the family (MO(n). e
n
). The associated homology and cohomology
groups of a pointed space Y are denoted MO
n
(Y ) and MO
n
(Y ).
(21.3.1) Theorem. There exists a natural isomorphism
T(X): N
n
(X) MO
n
(X
).
We will see that the isomorphism T(X) is obtained by a stable version of the
PontrjaginThom construction.
For each space X we denote by
n
(X) the product bundle id
A
;
n
. We dene
a map
k
(X): 1(R
nk
.
k
(X)) N
n
(X).
Let M. J| be an n-dimensional
k
(X)-submanifold of R
nk
. The rst component
J
1
of the
k
(X)-structure J = (J
1
. J
2
): 1(v
) X1(;) gives us the element
k
(X)M. J| = M. J
1
| N
n
(X). It is obvious that we obtain a well-dened
map
k
. There is a kind of suspension map
o : 1(R
nk
.
k
(X)) 1(R
nk1
.
k1
(X)).
For M. J| consider M
t
= M {0] R
nk
{0] R
nk1
. The normal
bundle of M
t
is 1(v
) c. We compose 1(v
) c 1(;
k
) c with the
classifying map 1(;
k
) c 1(;
k1
). From J we thus obtain a new struc-
ture J
t
= (J
1
. J
t
2
): 1(v
/ ) X 1(;
k1
). We set oM. J| = M
t
. J
t
|.
The commutativity
k1
o =
k
holds. Let 1
n
(X) denote the colimit over
the maps o : 1(R
nk
.
k
(X)) 1(R
nk1
.
k1
(X)). Altogether we obtain
(X): 1
n
(X) N
n
(X).
(21.3.2) Proposition. The map (X): 1
n
(X) N
n
(X) is bijective.
Proof. Surjective. Let M. | N
n
(X) be given. We can assume M R
nk
for
some k, by the Whitney embedding theorem. Let k
: v
;
k
be a classifying
map of the normal bundle. Then we have the
k
(X)-structure J = ( v
. k
),
and
k
(X)M. J| = M. | holds by construction.
Injective. Suppose M
0
. J
0
| and M
1
. J
1
| have the same image under (X).
We can assume that M
}
R
nk
for a suitable k. There exists a bordism T with
dT = M
0
M
1
and an extension : T X of (
0
.
1
) where
}
is the rst
component of J
}
. There exists an embedding T R
nkt
0. 1| such that
T (R
nk
R
t
0. 1,3 ) = M
0
0 0. 1,3 .
T (R
nk
R
t
|2,3. 1|) = M
1
0 |2,3. 1|.
By use of collars we can nd a bordism T such that
: C = M
0
0. 1,2 |1,2. 1| T
and dT = M
0
0M
1
1. We embed C R
nk
R
t
0. 1|, (m. s) (m. 0. s).
Then we choose a continuous function : T R
nkt
0. 1| such that
extends on D = M
0
0. j| M
1
1 j. 1| for some 1,3 < j < 1,2
and such that (T D) is contained in R
nkt
|1,3. 2,3 (Tietze extension
theorem). Suppose k t > n 1. We now approximate by an embedding
J : T R
nkt
0. 1| such that J(T D) R
nkt
|1,3. 2,3 and such
that J equals on M
0
0. 1,3 M
1
|2,3. 1|.
The bundle maps v
;
k
yield bundle maps v
t c ;
kt
. Since these
classifying maps are unique up to homotopy and since dT T is a cobration,
we can extend these maps to a bundle map v
B
;
kt
. We thus see that the
M
}
. J
}
| 1(R
nkt
.
k
(X)) have the same image in 1(R
nkt
.
kt
(X)).
The PontrjaginThom maps
1 : 1(R
nk
.
k
(X)) S
nk
. M(
k
(X))|
0

nk
(X
. MO(k))
are compatible with suspension 1 o = o 1. We obtain a bijection of the colimits
1 : 1
n
(X) MO
n
(X
) and hence a natural bijection

T = 1
-1
: N
n
(X) MO
n
(X
).
It remains to verify that T is a homomorphism. Let M
i
.
i
| N
n
(X) be given.
Choose embeddings
M
1

1
= {. [ .
nk
> 0] R
nk
. M
2

2
= {. [ .
nk
< 0] R
nk
and choose tubular neighbourhoods U
i

i
of M
i
. The PontrjaginThom con-
struction applied to M
1
M
2
yields a map which factors over the comultiplication
S
nk
S
nk
S
nk
. The restrictions to the summands are representatives of
1M
i
.
i
|.
For more details on bordismhomology and cohomology theories see [37], [181],
[160], [28].
21.4. Oriented Bordism 537
Problems
1. Given
0
.
1
: Q
c
M(). Let U M() be an open neighbourhood of T. Suppose
V =
-1
0
(U) =
-1
1
(U) and
0
[V =
1
[V . Then
0
and
1
are pointed homotopic.
21.4 Oriented Bordism
The oriented bordism homology theory is isomorphic to the spectral homology
theory of the Thom spectrum MSO = (MSO(n). e
n
) where MSO(n) is the Thom
space of the universal n-dimensional orientable vector bundle over TSO(n). The
isomorphism
C
n
(X) MSO
n
(X
)
is established as in the case of unoriented bordism. The PontrjaginThom con-
struction uses in this case an orientation of the normal bundle. An embedding
M
n
R
nk
induces a canonical isomorphism t(M) v(M) (n k)c in
which the normal bundle v(M) is the orthogonal complement of the tangent bun-
dle. An orientation of t(M) induces an orientation of v(M) such that brewise
t
x
(M) v
x
(M) R
nk
is orientation preserving.
(21.4.1) Lemma. MSO(k) is (k 1)-connected.
Proof. The canonical map s : TSO(k 1) TSO(k) can be taken as the sphere
bundle of the universal oriented k-dimensional bundle. From the homotopy se-
quence of this bration we see that s is (k 1)-connected. The homotopy se-
quence of s is isomorphic to the sequence of the pair (D
k
. S
k
) of the universal
(disk,sphere)-bundle over TSO(k). Hence (D
k
. S
k
) is (k 1)-connected. Since
TSO(k) is simply connected, we can apply (6.10.2) and see that
}
(D
k
. S
k
)
}
(D
k
,S
k
) =
}
(MSO(k)) is an isomorphism for _ k 1.
The suspension isomorphism
nk
(MSO(k))
nk1
(MSO(k)) is an
isomorphism for k _ n 2, since MSO(k) is (k 1)-connected (see (6.10.4)).
The spectral map
}
(MSO(k))
}
(MSO(k 1)) is an isomorphism for
_ 2k 1. In order to see this, we use the Whitehead theorem (20.1.4): The
spaces in question are simply connected. Thus it sufces to see that we have a
homology isomorphism in the same range.
H
}
(MSO(k))

Thom
H
}
(MSO(k 1))
Thom
H
}-k
(TSO(k))

H
}-k
(TSO(k 1))
The map TSO(k) TSO(k 1) is (k 1)-connected, hence the vertical maps
are isomorphisms for k _ k 1. These arguments show that we need
not pass to the colimit, colim
k

nk
MSO(k); we already have an isomorphism
C
n

nk
MSO(k) for k _ n 2. (The geometric reason for this stability result
is the strong form of the Whitney embedding theorem which we had not used in
(21.3.1).)
The assignment M| (k
)
+
M| H
n
(TSO: Q) of Section 19.8 induces a
ring homomorphism C
+
H
+
(TSO: Q) which we extend to a homomorphism
of Q-algebras t
+
: C
+
Q H
+
(TSO: Q). We can dene in a similar manner a
homomorphismv
+
: C
+
Q H
+
(TSO: Q) if we use classifying maps of stable
normal bundles.
(21.4.2) Theorem. The homomorphisms
v
+
: C
+
Q H
+
(TSO: Q) and t
+
: C
+
Q H
+
(TSO: Q)
are isomorphisms of graded algebras.
Proof. Lemma (21.4.1) and (20.8.3) imply that the Hurewicz homomorphism
i
(MSO(k)) H
k
(MSO(k))
has for r < 2k 1 both a nite kernel and a nite cokernel; hence it induces
an isomorphism
i
(MSO(k)) Q H
i
(MSO(k): Q) in this range. We also
have the homological ThomisomorphismH
i
(MSO(k): Q) H
i-k
(TSO(k): Q).
The previous considerations now show that we have an isomorphism C
n
Q
H
n
(TSO(k): Q). The computation of H
n
(TSO(k): Q) H
n
(TSO: Q) for k _
n 2 shows that the Q-vector space C
4n
Qhas dimension (n), the number of
partitions of n. From our computation of t
+
(C1
2o
) we see that t
+
: C
4n
Q
H
4n
(TSO: Q) is a surjective map between vector spaces of the same dimension,
hence an isomorphism.
The homomorphism v
+
is obtained from t
+
by composition with the antipode |
of the Hopf algebra H
+
(TSO: Q). It is determined by the formal relation
(1 q
1
q
2
)(1 |(q
1
) |(q
2
) ) = 1
which expresses the relation between the Pontrjagin classes of a bundle and its
inverse.
Together with our previous computations (19.8.4) we obtain:
(21.4.3) Theorem. The algebra C
+
Qis a polynomial Q-algebra in the gener-
ators C1
2n
|, n N.
We now collect various results and prove the Hirzebruch signature theorem.
(21.4.4) Theorem. The signature o(M) of an oriented closed 4n-manifold is ob-
tained by evaluating the Hirzebruch polynomial 1
n
in the Pontrjagin classes of M
on the fundamental class.
21.4. Oriented Bordism 539
Proof. From (18.7.3) and (18.7.7) we see that M o(M) induces a ring homo-
morphismC
+
Z. We extend it to a homomorphismof Q-algebras o : C
+
Q.
Via the isomorphism t
+
of Theorem (21.4.2) it corresponds to a homomorphism
s : H
+
(TSO: Q) Q such that o = s t
+
. The homomorphism s was used
at the end of Section 19.8 to determine polynomials 1
n
H
4n
(TSO: Q) in the
Pontrjagin classes such that (1
n
(). M| ) = o(M).
Problems
1. It is not necessary to use the computation of t
(C1
2u
) in the proof of (21.4.2). The
reader is asked to check that the diagram
C
n
T
nk
(MSO(k))
h
H
nk
(MSO(k))
Ir
H
n
(TSO) H
n
(TSO(k))
commutes (at least up to sign if one does not care about specic orientations). 1 is the
PontrjaginThom map, h is the Hurewicz homomorphism, t : is the homological Thom
isomorphism, the bottom map is induced by the stabilization.
Bibliography
[1] Abe, E., Hopf Algebras. Cambridge Tracts in Math. 74, Cambridge University Press,
1980. 485
[2] Adams, J. F., On the non-existence of elements of Hopf invariant one. Ann. of Math.
72 (1960), 20104. 518
[3] Adams, J. F., Vector elds on spheres. Ann. of Math. 75 (1962), 603632. 136, 339
[4] Adams, J. F., Avariant of E. H. Browns representability theorem. Topology 10 (1971),
185198. 216
[5] Adams, J. F., Atiyah, M. F., 1-theory and the Hopf invariant. Quart. J. Math. Oxford
17 (1966), 3138. 518
[6] Aguilar, M., Gitler, S., Prieto, C., Algebraic topology from a homotopical viewpoint.
Universitext, NewYork, Springer 2002. 357
[7] Allday, C., Puppe, V., Cohomological methods in transformation groups. Cambridge
Stud. Adv. Math. 32, Cambridge University Press, 1993. 328
[8] Araki, S., Toda, H., Multiplicative structure in mod q cohomology theories. I. Osaka
J. Math. 2 (1965), 71115; II - Osaka J. Math. 3 (1966), 81120. 256
[9] Atiyah, M. F., 1-theory and Reality. Quart. J. Math. Oxford (2) 17 (1966), 367386.
357
[10] Atiyah, M. F., Bott periodicity and the index of elliptic operators. Quart. J. Math.
Oxford (2) 72 (1968), 113140. 357
[11] Atiyah, M. F., 1-Theory. NewYork, W.A. Benjamin 1967. 357
[12] Atiyah, M. F., Collected Works, Vol IV. Oxford Science Publishers 1988. 357
[13] Atiyah, M. F., Bott, R., On the periodicity theorem for complex vector bundles. Acta
Math. 112 (1964), 229247. 357
[14] Atiyah, M. F., Bott, R., Shapiro A., Clifford modules. Topology 3 (1964), 338. 357
[15] Atiyah, M. F., Hirzebruch, F., Vector bundles and homogeneous spaces. Proc. Symp.
in Pure Math., Amer. Math. Soc. 3 (1961), 738. 357
[16] Atiyah, M. F., Singer, I. M., The index of elliptic operators I. Ann. of Math. 87 (1968),
484534. 528
[17] Barrat, M., Whitehead, J. H. C., The rst non-vanishing group of an (n1)-ad. Proc.
London Math. Soc. (3) 6 (1966), 417439. 267
[18] Baues, H. J., Algebraic homotopy. Cambridge Stud. Adv. Math. 15, Cambridge Uni-
versity Press, 1989. 95
[19] Behrens, M. J., A new proof of the Bott periodicity theorem. Topology Appl. 119
(2002), 167 -183. 357
[20] Betti, E., Sopra gli spazi di un numero qualunque dimensioni. Ann. Mat. Pura Appl.
2 (4) (1871), 140158. 310
542 Bibliography
[21] Blakers, A. L., Some relations between homology and homotopy groups. Ann. of Math.
49 (1948), 428461. 235
[22] Blakers, A. L., Massey, W.S., The homotopy groups of a triad II. Ann. of Math. 55
(1952), 192201. 133
[23] Borsuk, K., Drei Stze ber die n-dimensionale euklidische Sphre. Fund. Math. 20
(1933), 177190.
[24] Bredon, G. E., Introduction to compact transformation groups. Pure Appl. Math. 46,
NewYork, Academic Press 1972.
[25] Bredon, G. E., Topology and Geometry. Grad. Texts in Math. 139, NewYorkBerlin
Heidelberg, Springer 1997. 251
[26] Brinkmann, H.-B., Puppe D., Kategorien und Funktoren. Lecture Notes in Math. 18,
Berlin, Springer 1966.
[27] Brcker, Th., Lineare Algebra und Analytische Geometrie. Grundstud. Math., Basel,
Birkhuser, 2003. 16
[28] Brcker, Th., tomDieck, T., Kobordismentheorie. Lecture Notes in Math. 178, Berlin,
Springer 1970. 527, 536
[29] Brcker, Th., tom Dieck, T., Representations of Compact Lie Groups. Grad. Texts in
Math. 98, NewYork, Springer 1995. 16, 77, 361, 401
[30] Brcker, Th., Jnich, K., Einfhrung in die Differentialtopologie. Heidelberger
Taschenbcher 143, Berlin, Springer 1973. 358, 365, 378
[31] Brown, E. H., Cohomology theories. Ann. of Math. 75 (1962), 467484; Ann. of Math.
78 (1963), 201. 216
[32] Brown, M., A proof of the generalized Schoenies Theorem. Bull. Amer. Math. Soc.
66 (1960), 7476. 251
[33] Brown, M., Locally at imbeddings of topological manifolds. Ann. of Math. 75 (1962),
331341.
[34] Brown, R., Groupoids and van Kampens theorem. Proc. London Math. Soc. (3) 17
(1967), 385401. 45
[35] Burde, G., Zieschang, H., Knots. de Gruyter Stud. Math. 5, BerlinNewYork, W. de
Gruyter 2003.
[36]

Cadek, M., Crabb, M., G-structures on spheres. Proc. London Math. Soc. (3) 93
(2006), 791816. 339
[37] Conner, P. E., Floyd, E. E., Differentiable periodic maps. Ergeb. Math. Grenzgeb. 33,
BerlinGttingenHeidelberg, Springer 1964. 536
[38] Crabb, M., James, I., Fibrewise Homotopy Theory. Springer Monogr. Math., Berlin
HeidelbergNewYork, Springer 1998. 32
[39] Dicks, W., Dunwoody, M. J., Groups acting on graphs. Cambridge Stud. Adv. Math.
17, Cambridge University Press, 1989. 52
[40] tomDieck, T., Klassikation numerierbarer Bndel. Arch. Math. 17 (1966), 395399.
Bibliography 543
[41] tom Dieck, T., Partitions of unity in homotopy theory. Compositio Math. 23 (1971),
159167.
[42] tom Dieck, T., Transformation groups and representation theory. Lecture Notes in
Math. 766, Berlin, Springer 1979.
[43] tom Dieck, T., Transformation Groups. de Gruyter Stud. Math. 8, BerlinNewYork,
de Gruyter 1987. 328
[44] tom Dieck, T., Topologie. (2. Auflage). de Gruyter Lehrbuch, BerlinNew York, de
Gruyter 2000. 56, 57, 199, 222, 355, 358, 375, 409, 447
[45] tom Dieck, T., General topology (Preliminary version).
www.uni-math.gwdg.de/tammo
[46] tom Dieck, T., Kamps, K. H., Puppe. D., Homotopietheorie. Lecture Notes in Math.
157, Berlin, Springer 1970. 118, 148
[47] tomDieck, T., Petrie, T., Contractible afne surfaces of Kodaira dimension one. Japan.
J. Math. (1990), 147169. 362
[48] Dold, A., Partitions of unity in the theory of brations. Ann. of Math. 78 (1963),
223255.
[49] Dold, A., Halbexakte Homotopiefunktoren. Lecture Notes in Math. 12, Berlin,
Springer 1966. 95
[50] Dold, A., Die Homotopieerweiterungseigenschaft (= HEP) ist eine lokale Eigenschaft.
Invent. Math. 6 (1968), 185189.
[51] Dold, A., Lectures on Algebraic Topology. Grundlehren Math. Wiss. 200, Berlin
HeidelbergNewYork, Springer 1972. 179, 311
[52] Dold, A., The xed point index of bre-preserving maps. Invent. Math. 25 (1974),
281297. 311
[53] Dold, A., A simple proof of the Jordan-Alexander complement theorem. Amer. Math.
Monthly 100 (1993), 856857. 249
[54] Dold, A., Puppe, D., Duality, trace, and transfer. In Proceedings Int. Conf. Geometric
Topology, PWN - Polish Scientic Publishers, Warszawa 1980, 81- 102. 159, 179
[55] Ebbinghaus, H.-D., et al., Zahlen. Grundwissen Math. 1, Berlin, Springer 1983. 518
[56] Eilenberg, S., Singular homology theory. Ann. of Math. 45 (1944), 407447. 223, 235
[57] Eilenberg, S., Mac Lane, S., Acyclic models. Amer. J. Math. 79 (1953), 189199. 286
[58] Eilenberg, S., Steenrod, N., Axiomatic approach to homology theory. Proc. Nat. Acad.
Sci. 31 (1945), 117120. 244
[59] Eilenberg, S., Steenrod, N., Foundations of Algebraic Topology. Princeton University
Press, 1952. 244, 245
[60] Eilenberg, S., Zilber, J. A., On products of complexes. Amer. J. Math. 75 (1953), 200
-204.
[61] Euler, L., Demonstratio non nullarum insignium proprietatum quibus solida hedris
planis indusa sunt praedita. Novi commentarii academiae scientiarum Petropolitanae
4 (1752/3), 140160. Opera Mathematica Vd. 26, 94108 (1758) 309
544 Bibliography
[62] Euler, L., Elementa doctrinae solidorum. Novi commentarii academiae scientiarum
Petropolitanae 4 (1752/3), 109140. Opera Mathematica Vd. 26, 7193 (1758) 309
[63] Farkas, H. M., Kra, I., Riemann Surfaces. Grad. Texts in Math. 71, New York
HeidelbergBerlin, Springer 1980. 317
[64] Federico, P. J., Descartes on Polyhedra. Sources Hist. Math. Phys. Sci. 4, NewYork
HeidelbergBerlin, Springer 1982. 309
[65] Flix, Y., Halperin, S., Thomas, J.-C., Rational homotopy theory. Grad. Texts in Math.
205, NewYorkBerlinHeidelberg, Springer 2001. 516
[66] Freudenthal, H., ber die Klassen von Sphrenabbildungen I. Compositio Math.
(1937) 5, 299314. 135
[67] Fritsch, R., Piccinini, R. A., Cellular Structures in Topology. Cambridge University
Press, 1990. 199, 218
[68] Fulton, W., Algebraic Topology. A rst course. Grad. Texts in Math. 153, NewYork,
Springer 1995.
[69] Gabriel, P., M. Zisman, Z., Calculus of Fractions and Homotopy Theory. Ergeb. Math.
Grenzgeb. 35, BerlinHeidelbergNewYork, Springer 1967. 95
[70] Goebel, K., Kirk, W. A., Topics in metric xed point theory. Cambridge Stud. Adv.
Math. 28, Cambridge University Press, 1990. 140
[71] Goerss, R. G., Jardine, J. F., Simplicial homotopy theory. Progr. Math. 174, Basel,
Birkhuser 1999.
[72] Gray, B., Homotopy theory. Pure and Appl. Math. 64, New York, Academic Press
1975.
[73] Goodwillie, Th. G., Calculus II. Analytic functors. 1-Theory 5 (1991/1992), 295332.
152
[74] Greene, J. E., A new short proof of Knesers conjecture. Amer. Math. Monthly (2002),
918920. 261, 264
[75] Greub, W., Halperin, S., Vanstone R., Pure and Appl. Math. 47, Connections, Curva-
ture, and Cohomology. NewYork, Academic Press 1972.
[76] Grifths, H. B., The fundamental group of two spaces with a common point. Quart.
J. of Math. 5 (1954), 175190.
[77] Guillou, L., Marin, A. (Ed.), A la Recherche de la Topologie Perdue. Progr. Math. 62,
Boston, Birkhuser 1986. 526
[78] Hewitt, E., Ross, K. A., Abstract harmonic analysis I. Grundlehren Math. Wiss. 115,
BerlinGttingenHeidelberg, Springer 1963.
[79] Hirsch, M. W., A proof of the nonretractability of a cell onto its boundary. Proc. Amer.
Math. Soc. 14 (1963), 364365. 139
[80] Hirsch, M.W., Differential topology. New YorkHeidelbergBerlin, Springer 1976.
311
[81] Hirzebruch, F., Neue topologische Methoden in der algebraischen Geometrie. 2. Aufl.,
Ergeb. Math. Grenzgeb. 9, Berlin, Springer 1962. 493, 528
Bibliography 545
[82] Hirzebruch, F., Gesammelte Abhandlungen. Bde III. BerlinHeidelbergNewYork,
Springer 1987.
[83] Hirzebruch, F., Mayer K.H., O(n)-Mannigfaltigkeiten, exotische Sphren und Singu-
laritten. Lecture Notes in Math. 57, Berlin, Springer 1968.
[84] Hochschild, G., The structure of Lie groups. San Francisco, Holden-Day 1965. 348,
361
[85] Hocking, J. G., Young G. S., Topology. Reading Mass., Addison-Wesley 1961.
[86] Hopf, H., ber die Abbildungen der dreidimensionalen Sphre auf die Kugelche.
Math. Ann. 104 (1931), 637665.
[87] Hopf, H., Die Klassen der Abbildungen der n-dimensionalen Polyeder auf die n-
dimensionale Sphre. Comment. Math. Helv. 5 (1933), 3954.
[88] Hopf, H., ber die Abbildungen von Sphren auf Sphren niedrigerer Dimension.
Fund. Math. 25 (1935), 427440. 517
[89] Hopf, H., Systeme symmetrischer Bilinearformen und euklidische Modelle der pro-
jektiven Rume. Vierteljahresschrift der Naturforschenden Gesellschaft in Zrich 85
(1940), 165177. 366
[90] Hopf, H., ber den Rang geschlossener Liescher Gruppen. Comment. Math. Helv. 13
(1940/41), 119143. 401
[91] Hopf, H., ber die Topologie der Gruppen-Mannigfaltigkeiten und ihrer Verallge-
meinerungen. Ann. of Math. 42 (1941), 2252. 90, 485
[92] Hopf, H., Selecta Heinz Hopf. BerlinGttingenHeidelberg, Springer 1964.
[93] Huppert, B., Endliche Gruppen I. Grundlehren Math. Wiss. 134, Berlin, Springer
1979.
[94] Hurewicz, W., Wallman, H., Dimension theory. Princeton Math. Ser. 4, Princeton
University Press, 1948. 140
[95] James, I. M., Euclidean models of projective spaces. Bull. London Math. Soc. 3 (1971),
257276. 366
[96] James, I. M., General topology and homotopy theory. Berlin, Springer 1984. 32
[97] James, I. M., Fibrewise topology. CambridgeTracts inMath. 91, Cambridge University
Press, 1989. 32
[98] James, I. M. (Editor), Handbook of Algebraic Topology. Amsterdam, Elsevier 1995.
[99] James, I. M. (Editor), History of Topology. Amsterdam, Elsevier 1999.
[100] van Kampen, E. H., On the connection between the fundamental group of some related
spaces. Amer. J. Math. 55 (1933), 261267. 45
[101] Kamps, K. H., Porter, T., Abstract homotopy and simple homotopy theory. Singapore,
World Scientic 1997. 95
[102] Karoubi, M., 1-Theory. Grundlehren Math. Wiss. 226, Berlin, Springer 1978. 357
[103] Klaus, S., Kreck, M., A quick proof of the rational Hurewicz theorem and a compu-
tation of the rationnal homotopy groups of spheres. Math. Proc. Camb. Philos. Soc.
136 (2004), 617623. 515
546 Bibliography
[104] Klein, F., Gesammelte Mathematische Abhandlungen. Bd. I-III. Berlin, Springer
19211923.
[105] Kneser, M., Aufgabe 360. Jahresbericht der Deutschen Mathematiker-Vereinigung 58
(1955), 27. 264
[106] Kono, A., Tamaki, D., Generalized cohomology. Transl. Math. Monogr. 230, Provi-
dence, R.I., Amer. Math. Soc. 2006. 357
[107] Kosinski, A., Differential Manifolds. Pure andAppl. Math. 138, Academic Press 1992.
358
[108] Kreck, M., Surgery and duality. Ann. of Math. 149 (1999), 707754. 390
[109] Laitinen, E., Lck, W., Equivariant Lefschetz classes. Osaka J. Math. 26 (1989),
491525. 311
[110] Lamotke, K., Semisimpliziale algebraische Topologie. Grundlehren Math. Wiss. 147,
BerlinHeidelbergNewYork, Springer 1968.
[111] Lewis Jr., L. G., May, J. P., Steinberger, M., McClure, J. E., Equivariant stable homo-
topy theory. Lecture Notes in Math. 1213, Berlin, Springer 1986.
[112] Lillig, J., A union theorem for cobrations. Arch. Math. 24 (1973), 410415. 114
[113] Liulevicius, A., Arrows, symmetries and representation rings. J. Pure Appl. Algebra
19 (1980), 259273. 489
[114] Lovsz, L., Knesers conjecture, chromatic number and homotopy. J. Combin. Theory
Ser. A 25 (1978), 319324. 264
[115] Lck, W., Transformation groups and algebraic 1-theory. Lecture Notes in Math.
1408, Berlin, Springer 1989.
[116] Lck, W., The universal functorial Lefschetz invariant. Fund. Math. 161 (1999),
167215. 311
[117] Lck, W., Chern character for proper equivariant homology theories and applications
to 1- and 1-theory. J. Reine Angew. Math. 543 (2002), 193234.
[118] Lck, W., 1
2
-invariants: Theory and applications to geometry and 1-theory. Ergeb.
Math. Grenzgeb. 44, BerlinHeidelbergNewYork, Springer 2002.
[119] Lck, W., Algebraische Topologie. Wiesbaden, Vieweg 2005.
[120] MacLane, S., Homology. Grundlehren MathematischenWiss. 114, BerlinGttingen
Heidelberg, Springer 1963. 294
[121] Magnus, W., Noneuclidean tesselations and their groups. Pure and Appl. Math. 61,
NewYork, Academic Press 1974. 316
[122] Magnus, W., Karras, A., Solitar, D., Combinatorial group theory. New York, Inter-
science Publishers 1966. 52
[123] Massey, W. S., Algebraic Topology: An Introduction. New York, Harcourt, Brace &
World 1967. 56, 311
[124] Massey, W. S., Homology and cohomology theory. Monogr. Textbooks Pure Appl.
Math. 46, NewYork, Marcel Dekker 1978.
Bibliography 547
[125] Matouek, J., Using the Borsuk-Ulamtheorem. Universitext, BerlinHeidelbergNew
York, Springer 2003.
[126] May, J. P., Classifying spaces and brations. Mem. Amer. Math. Soc. 1 (1955), no.1,
155.
[127] May, J. P., A concise course in algebraic topology. Chicago Lecture Notes in Math.,
Chicago University Press, 1999.
[128] May, J. P., Sigurdsson, J., Parametrized homotopy theory. Math. Surveys Monogr.
132, Providence, R.I., Amer. Math. Soc. 2006. 32, 195
[129] Mayer, K. H., Algebraische Topologie. BaselBostonBerlin, Birkhuser 1989.
[130] McCleary, J., Ausers guide to spectral sequences. 2. ed., Cambridge Stud. Adv. Math.
58, Cambridge University Press, 2001. 303
[131] Milnor, J., Construction of universal bundles. II. Ann. of Math. 63 (1956), 430436.
345
[132] Milnor, J., On spaces having the homotopy type of a CW-complex. Trans. Amer.
Math. Soc. 90 (1958), 272280. 218
[133] Milnor, J., On axiomatic homology theory. Pacic J. Math. 12 (1962), 337342. 408
[134] Milnor, J., Morse Theory. Ann. of Math. Stud. 51, Princeton University Press, 1963.
390
[135] Milnor, J., Micro bundles. Topology 3, Suppl. 1 (1964), 5380.
[136] Milnor, J., Topology from the differentiable view point. Charlottesville University
Press, 1965. 365
[137] Milnor, J., Lectures on the h-cobordism theorem. Princeton University Press, 1965.
390
[138] Milnor, J., Moore, J. C., On the structure of Hopf algebras. Ann. of Math. 81 (1965),
211264 485
[139] Milnor, J., Stasheff, J. D., Characteristic classes. Ann. Math. Stud. 76, Princeton
University Press, 1974.
[140] Mbius, A. F., Gesammelte Werke. Bd. I -IV, Leipzig, Hirzel KG 1886. 342
[141] Moise, E. E., Geometric Topology in Dimensions 2 and 3. Grad. Texts in Math. 47,
BerlinHeidelbergNewYork, Springer 1977. 198, 251
[142] Montgomery, S., Hopf algebras and their actions on rings. Amer. Math. Soc. Regional
Conf. Series Math. 82, Washington, DC, 1993. 485
[143] Munkres, J. R., Elementary Differential Topology. Princeton University Press, 1966.
198
[144] Munkres, J. R., Topology. A rst Course. Englewood Cliffs, Prentice-Hall 1975.
[145] Munkres, J. R., Elements of Algebraic Topology. Reading, Mass., Addison-Wesley
1984.
[146] Nagata, J., Modern general topology. Groningen, Nordhoff 1968
[147] Nomura, Y., On mapping sequences. Nagoya Math. J. 17 (1960), 111145. 99
548 Bibliography
[148] Peano, G., Sur une courbe, qui remplit tout une aire plane. Math. Ann. 36 (1890),
157160.
[149] Pommerenke, Ch., Boundary behaviour of conformal maps. Grundlehren Math. Wiss.
299, Berlin, Springer 1992. 251
[150] Poincar, H., Sur la gnralisationdunthorme dEuler relatif aux polydres. Compt.
Rend. Acad. Sci. Paris 117 (1893), 144145. 310
[151] Poincar, H., Analysis situs. Journal de lcole Polytechnique 1 (1895), 1121. 42
[152] Poincar, H., Complment lAnalysis Situs. Rend. Circ. Mat. Palermo 13 (1899),
285343. 310
[153] Poincar, H., Cinquime complment lAnalysis Situs. Rend. Circ. Mat.Palermo 18
(1904), 45110.
[154] Poincar, H., uvres de Henri Poincar VI. Paris, Gauthier-Villars 1953.
[155] Puppe, D., Homotopiemengen und ihre induzierten Abbildungen. I. Math. Z. 69
(1958), 299344. 33, 81, 95, 195
[156] Puppe, D., Bemerkungen ber die Erweiterung von Homotopien. Arch. Math. 18
(1967), 8188.
[157] Puppe, D., Some well known weak homotopy equivalences are genuine homotopy
equivalences. Symposia Mathematica 5 (1971), 363374.
[158] Puppe, D., Homotopy cocomplete classes of spaces and the realization of the singular
complex. In: Topological Topics (I. M. James ed.), London Math. Soc. Lecture Note
Ser. 86, Cambridge University Press, 1983, 5569.
[159] von Querenburg, B., Mengentheoretische Topologie. BerlinHeidelbergNew York,
Springer 1979.
[160] Quillen, D., Elementary proofs of some results of cobordism using Steenrod opera-
tions. Adv. Math. 7 (1971), 2956. 527, 536
[161] Rad, T., ber den Begriff der Riemannschen Flche. Acta Sci. Math. (Szeged) 2
(1924), 101121. 198
[162] Ranicki, A., Algebraic and geometric surgery. Oxford Math. Monographs, Oxford
University Press, 2002. 390
[163] Ravenel, D. C., Complex cobordism and stable homotopy groups of spheres. New
York, Academic Press 1986. 517
[164] Schubert, H., Topologie. Stuttgart, Teubner, 4. Auage 1975.
[165] Segal, G., Classifying spaces and spectral sequences. Inst. Hautes tudes Sci. Publ.
Math. 34 (1968), 105112.
[166] Seifert, H., Konstruktion dreidimensionaler geschlossener Rume. Ber. Schs. Akad.
Wiss. 83 (1931), 2666. 45
[167] Seifert, H., Threllfall, W., Lehrbuch der Topologie. Leipzig, Teubner 1934. 56, 311,
438
[168] Selick, P., Introduction to homotopy theory. Fields Institute Monogr. 9, Amer. Math.
Soc., Providence, R.I. , 1997.
Bibliography 549
[169] Serre, J.-P., Homologie singulire des espaces brs. Ann. of Math. 54 (1951),
425505.
[170] Serre, J.-P., Groupes dhomotopie et classes de groupes abliens. Ann. of Math. 58
(1953), 258294. 504
[171] Serre, J.-P., Trees. BerlinHeidelbergNewYork, Springer 1980. 52
[172] Spanier, E. H., Algebraic Topology. NewYork, McGrawHill 1966.
[173] Sperner, E., Neuer Beweis fr die Invarianz der Dimensionszahl und des Gebietes.
Abh. Math. Sem. Univ. Hamburg 6 (1928), 265272. 139
[174] Steen, L. A., Seebach, J. A. jr., Counterexamples in topology. Holt, Rinehart ans
Winston 1970.
[175] Steenrod, N., The topology of bre bundles. Princeton Math. Ser. 14, Princeton Uni-
versity Press, 1951.
[176] Steenrod, N., A convenient category of topological spaces. Michigan Math. J. 14
(1967), 13152.
[177] Sternberg, S., Lectures on differential geometry. Englewood Cliffs, Prentice Hall 1964.
365
[178] Stcker, R., Zieschang, H., Algebraische Topologie. Mathematische Leitfden,
Stuttgart, Teubner 1988.
[179] Strm, A., Note on cobrations. Math. Scand. 19 (1966), 1114.
[180] Strm, A., Note on cobrations II. Math. Scand. 22 (1968), 130142. 103
[181] Stong, R. E., Notes on cobordism theory. Princeton University Press, 1968. 536
[182] Sweedler, M. E., Hopf algebras. Math. Lecture Note Ser., NewYork, Benjamin 1980.
485
[183] Switzer, R. M., Algebraic Topology Homotopy and Homology. Grundlehren Math.
Wiss. 212, BerlinHeidelbergNewYork, Springer 1975.
[184] Thom, R., Quelques proprits globales des varits diffrentiables. Comment. Math.
Helv. 28 (1954), 1786. 521, 527, 528
[185] Toda, H., Composition methods in homotopy groups of spheres. Ann. of Math. Stud.
49, Princeton University Press, 1962. 517
[186] Vogt, R., Convenient categories of topological spaces for homotopy theory. Arch.
Math. 22 (1971), 545555.
[187] Waldhausen F., Algebraic 1-theory of spaces. In: Proc. Conf., Rutgers NewBrunswick
1983, Lecture Notes in Math. 1126, Springer, 1985, 318419.
[188] Wall, C. T. C., Finiteness conditions for CW-complexes. Ann. of Math. 81 (1965),
5669; Finiteness conditions for CW-complexes II, Proc. London Math. Soc. 1966,
129139.
[189] Wall, C. T. C., On the exactness of interlocking sequences. Enseign. Math. 12 (1966),
95100. 245
[190] Wall, C. T. C., A geometric introduction to topology. Reading, Mass., Addison-Wesley
1972.
550 Bibliography
[191] Wall, C. T. C., Ranicki, A., Surgery on compact manifolds. Math, Surveys Monogr,
69, Providence, R.I., Amer. Math. Soc. 1999. 390
[192] Whitehead, G. W., Elements of homotopy theory. Grad. Texts in Math. 61, NewYork,
Springer 1978. 214
[193] Whitehead, J. H. C., On C
1
-complexes. Ann. of Math. 41 (1940), 809824. 198
[194] Whitehead, J. H. C., Combinatorial homotopy. Bull. Amer. Math. Soc. 55 (1949),
213245.
[195] Whitehead, J. H. C., Mathematical Works. Vol I IV. Oxford, Pergamon Press 1962.
[196] Whitney, H., On regular closed curves in the plane. Compositio Math. 4 (1937),
276286.
[197] Zieschang, H., Vogt, E., Coldewey, H.-D., Surfaces and planar discontinuous groups.
Lecture Notes in Math. 835, BerlinHeidelbergNewYork, Springer 1980. 198
Symbols
Numbers
N natural numbers {1. 2. 3. . . . ]
N
0
NL {0]
Z integers {0. 1. 2. . . . ]
Q rational numbers
R, R
, R
-
real numbers, non-negative, non-positive
C complex numbers
C
+
C 0, non-zero numbers
H quaternions
Z,m = Z,mZ integers modulo m
F
q
eld with q elements
Categories
TOP category of topological spaces and continuous maps
TOP
0
pointed spaces and pointed maps
TOP
1
spaces under the space 1
TOP
B
spaces over the space T
TOP(2) pairs (X. ) of a space X and a subspace
G- TOP spaces with an action of the topological group G
COV
B
covering spaces of the space T
h-TOP homotopy category of TOP
h-C homotopy category associated to a category C
with homotopy notion
1- MOD left modules over the ring 1
ABEL abelian groups
SET sets
G- SET sets with left G-action
(X) fundamental groupoid of X
Or(G) orbit category of the group G
552 Symbols
C. D| category of functors C D and natural transformations
TRA
B
transport category (T). SET|
Ker, Ke kernel
Coker, Ko cokernel
Im image
id. 1 identity
pr projection onto a factor of a product
(
}
) map into a product with components
}
(
}
) map from a sum with components
}
Spaces
R
n
Euclidean n-space
S
n
unit sphere in R
n1
D
n
unit disk in R
n
1
n
D
n
S
n-1
unit cell
1 = 0. 1| unit interval
1
n
n-fold Cartesian product 1 1 1
d1
n
combinatorial boundary of 1
n
S(n) 1
n
,d1
n
S
(n)
R
n
L {o]
^
n
= ^n| n-dimensional standard simplex
d^
n
combinatorial boundary of ^
n
R1
n
n-dimensional real projective space
C1
n
n-dimensional complex projective space
V
k
(F
n
) Stiefel manifold of orthonormal k-frames in F
n
G
k
(F
n
) Grassmann manifold of k-dimensional subspaces of F
n
EG universal free G-space
BG classifying space of the topological group G
1,G. G\1 orbit space of a G-action on 1
1
G
J balanced product
X
1
H-xed point set
1(. n) EilenbergMac Lane space of type (. n)
Symbols 553
X, X with X identied to a point
X
X with additional base point

Groups
GL
n
(F) group of (n. n)-matrices with entries in F = R. C. H
O(n) group of orthogonal matrices in GL
n
(R)
SO(n) matrices of determinant 1 in O(n)
U(n) group of unitary matrices in GL
n
(C)
SU(n) matrices of determinant 1 in U(n)
S
1
complex numbers of modulus 1
Spin(n) Spinor group, double covering of SO(n)
Sp(n) symplectic group
S
n
symmetric group
Relations
end of proof
end of numbered item
difference set
cup product
: cap product
~ equivalent (equivalence relation)
. homotopy-equivalent, homotopic
isomorphic, homeomorphic, diffeomorphic
direct sum
tensor product
pointed sum (bouquet)
= join
. smash product
+ induced morphism, general index, base point, free product
of groups
, l topological sum
[.[ Euclidean norm of .
554 Symbols
[:[ absolute value of complex number :
[X[ cardinality of X
Modules
lim limit (inverse limit)
colim colimit (direct limit)
lim
1
derived functor of lim
Tor torsion product, left derived of Hom
Ext. Ext
n
(. T) module of extensions, right derived of Hom

Bundles
1(). 1
0
() total space of vector bundle, without zero section
TM tangent bundle of the manifold M
1(X) = 1
C
(X) = 1U(X) Grothendieck ring of complex vector bundles
1O(X) = 1
R
(X) Grothendieck ring of real vector bundles
c. nc trivial bundle, of dimension n
M() Thom space of the bundle
MO(n). MSO(n) Thom space of the universal n-dimensional
(oriented) bundle
c
i
(). n
i
().
i
() i -th Chern, StiefelWhitney, Pontrjagin class
Homotopy
CX (pointed) cone on X
X suspension of X
7( ) (pointed) mapping cylinder of
7(. g) double mapping cylinder
C( ) mapping cone of
C(X. ) mapping cone of X
H
-
inverse of the homotopy H
1 + 1 product (concatenation) of homotopies 1. 1
X. Y | homotopy classes of maps X Y
X. Y |
0
pointed homotopy classes
X. Y |
1
. X. Y |
B
homotopy classes under 1, over T
Symbols 555
(X) fundamental groupoid of X
(X. Y ) homotopy groupoid
n
(X. . +) n-homotopy group of the pointed pair (X. . +)
X
Y
= J(X. Y ) space of maps X Y with compact-open topology
CX loop space of X
hocolim homotopy colimit
N(U) nerve of the covering U
T(U) geometric realization of the nerve
Co-Homology
S
q
(X) singular q-chains
S
-
(X. ) singular chain complex of (X. )
H
q
(X. : G) ordinary homology with coefcients in G
H
q
(X. : G) ordinary cohomology with coefcients in G
0
. . . . .
q
| afne simplex with vertices
}
h
+
. h
+
general homology, cohomology; or its
coefcient groups
h
+
.

h
+
reduced homology, cohomology
d boundary operator
coboundary operator
(X) Euler characteristic of X
N
+
(X) unoriented bordism
C
+
(X) oriented bordism
MO
k
(X). MSO
k
(X) unoriented, oriented bordism via Thom spectra
Index
accumulation value, 4
action
diagonal, 17
effective, 17
free, 17
left, 17
proper, 329
properly discontinuous, 64
right, 17
transitive, 17
trivial, 17
weakly proper, 329
acyclic, 287, 498
additive invariant, 309
additivity axiom, 245
adjoint map, 38
adjunction space, 7
afnely independent, 198
Alexander duality, 446
AlexanderWhitney map, 240
AlexanderWhitney morphism, 241
algebra, 482
dual, 483
antipodal involution, 21
antipode, 484
approximation of the diagonal, 240
AlexanderWhitney, 240
atlas, 358
orienting, 336, 372
attaching a space, 7
attaching map, 203
barycenter, 231
barycentric coordinates, 198
barycentric subdivision, 197
base change, 117
base point, 31
nondegenerate, 102
base space, 32
basis of a topology, 2
Bernoulli numbers, 493
Betti number, 310
bi-degree, 260
bialgebra, 484
Bockstein operator, 419
bordant
orientable, 403
bordism, 521, 524
oriented, 526
BorsukUlam theorem, 436
boundary, 2, 283
boundary operator, 123, 224, 245, 266,
283, 284
boundary relation, 224
bouquet, 31
Brouwer Fixed Point Theorem, 137
bundle, 62
associated, 334
framing, 532
induced, 331, 337
numerable, 342
principal, 328
projective, 471
universal, 344
bundle atlas, 336
bundle chart, 62, 336
bundle isomorphism, 331
bundle map, 331, 336
bundle morphism, 336
canonical complex line bundle, 340
cap product, 438
category
topological, 334
2-category, 60
1-morphism, 61
558 Index
2-morphism, 61
horizontal composition, 61
vertical composition, 61
cell, 203
attaching, 203, 204
k-dimensional, 199
cellular chains, 302
cellular approximation theorem, 210
cellular chain complex, 302
chain, 283
singular, 224
with coefcients, 237
chain complex, 283
acyclic, 287
cellular, 302
contractible, 287
exact sequence, 284
singular, 224
chain equivalence, 285
chain homotopic, 285
chain homotopy, 285
chain map, 283
characteristic class, 474
stable, 474
characteristic map, 199
characteristic subgroup, 75
chart, 358, 369
adapted, 337, 359, 366, 370
centered at a point, 358
C
k
-related, 358
domain, 358
positive, 336, 372, 400
positively related, 372
Chern class, 474
total, 474
chromatic number, 264
classication II, 75
classifying map, 344
stable, 356
classifying space, 344
closed simplex, 198
closure, 2
clutching datum, 9
coalgebra, 483
dual, 483
cobase change, 104
coboundary, 285
coboundary operator, 285, 406
cochain, 285
cochain complex, 285
cocycle, 285
codimension, 359
coefcient groups, 245, 406
cobration, 102
homotopy theorem, 106
induced, 104
cobre, 111
cobre sequence, 95
cogroup, 91
coherent, 445
cohomology
additive, 180
C-oriented, 467
for pointed spaces, 180
limits, 408
R-oriented, 467
singular, 417
cohomology group, 406
cohomology module, 285
cohomology theory, 405
additive, 406
multiplicative, 409
suspension, 406
cohomotopy group, 181
coincidence set, 11
colimit, 12
colimit topology, 10
collar, 382
commutation rule, 61, 88
commutator group, 227
comonoid, 90
compact, 11
locally, 12
compact-open topology, 37
Index 559
compactication, 12
Alexandroff, 12
one-point, 12
point at innity, 12
compactly generated, 187
complex
Whitehead, 199
complex structure, 342
component, 27
compressible, 143
comultiplication, 91
cone, 92, 107
cone construction, 228
connected sum, 390
connecting morphism, 280
connectivity
o-connected, 143
n-connected, 143
continuous, 2
contractible, 28, 287
contraction, 28
convergence
uniform, 4
convex, 29
convolution, 484
coordinate change, 358
copairing, 491
covering, 11
closed, 11
countable, 11
c-covering, 140
nite, 11
locally nite, 11
n-connective, 214
nerve, 197
null homotopic, 325
numerable, 318
numeration, 318
open, 11
order, 140
point-nite, 11
renement, 11
shrinking, 15
universal, 66
covering dimension, 140
covering space, 63
associated, 65
characteristic subgroup, 75
classication I, 70
classication II, 75
classication III, 78
deck transformation, 65
path lifting, 69
principal, 64, 329
regular, 75
transport functor, 67
trivial, 63
universal, 78
cup product, 409
CW-approximation, 215
CW-complex, 205
cellular dimension, 205
nite, 205
relative, 205
skeleton, 205
CW-decomposition, 205
CW-space, 205
cycle, 283
cylinder, 92
deck transformation, 65
decomposable, 253
decomposition of a space, 27
deformation retract, 32
degree, 49, 136
bordism invariance, 403
homological, 257
local, 259
modulo 2, 401
diagonal action, 17
diameter, 4
diffeomorphism, 359
differential, 283, 362
differential structure, 359
560 Index
dimension axiom, 245, 406
directed set, 4
distance, 3, 4
divided powers, 427
division algebra, 518
domination, 215
double mapping cylinder, 84
duality, 175
coevaluation, 175
evaluation, 175
duality pairing, 442
duality theorem, 446
edge, 197
Eilenberg subcomplex, 235
EilenbergMac Lane morphism, 240
EilenbergMac Lane space, 217
polarized, 218
product, 220
rational cohomology, 513
EilenbergMac Lane spectrum, 221
EilenbergZilber morphism, 238
elementary surgery, 390
embedding, 5
smooth, 359
universal property, 5
ENR, 448
equivariant, 17
Euclidean neighbourhood retract, 176,
448
Euclidean space
with two origins, 80
Euler characteristic, 179, 310
combinatorial, 308
relative homological, 310
Euler class, 431, 464
evaluation, 37
on the fundamental class, 457
evenly covered, 63
exact homology sequence, 225
exact sequence, 68
exact sequence of a triple, 225
excision map, 133
excision property, 525
excisive, 166, 239, 265, 407
extension, 101
initial condition, 101
extension of modules, 294
external product, 413
face, 197
bration, 66
homotopy theorem, 118
Hurewicz, 115
induced, 117
Serre, 115
bration theorem, 505
bre, 32
typical, 63
bre bundle, 62
bre sequence, 99
bre transport, 118
brewise, 32
nal, 4
ve lemma, 278, 282
mod C, 505
xed point set, 17
ag bundle, 477
ag space, 477
formal group law, 485
free group, 53
free product, 53
free resolution, 292
fundamental class, 399, 402, 445
fundamental group, 42
fundamental groupoid, 42
fundamental lemma, 287
Gauss map, 352
genus, 312
geometric realization, 198, 321
graded module, 283
graph, 197, 264
chromatic number, 264
Index 561
colouring, 264
Grassmann manifold, 23, 362
Grothendieck ring, 356
group
discrete, 15
free, 53
general linear, 16
normalizer, 21
orthogonal, 16
special linear, 16
special orthogonal, 16
special unitary, 16
topological, 15
topological subgroup, 16
torus, 16
unitary, 16
Weyl, 21
group algebra, 486
group object, 90
group-like, 486
groupoid, 42
topological, 335
Gysin sequence, 431
h-bration, 118
half-space, 369
Hausdorff space, 3
h-coexact, 92
h-cobration, 118
HEP, 101
h-equivalence, 28
h-exact, 97
hexagon lemma, 277
Hirzebruch 1-polynomials, 494
HLP, 66, 115
homeomorphism, 2
homogeneous space, 18
homological orientation, 444, 445
homologous, 283
homology
additive, 180
additivity axiom, 245
coefcient groups, 245
dimension axiom, 245
for pointed spaces, 179
group, 283
integral, rational, mod(), 237
module, 283
homology product, 242
homology group, 283
local, 392
reduced, 252
relative, 225
singular, 224
homology module, 283
homology theory, 244
one-space, 528
homotopic, 27
homotopy, 27, 38
constant, 28
equivariant, 17
G-homotopy, 17
inverse, 28
linear, 29
product, 28
relative, 28
homotopy category, 28
homotopy class, 28
homotopy cocartesian, 85
homotopy colimit, 270
homotopy equivalence, 28
weak, 144
homotopy equivalent, 28
homotopy extension property, 101
homotopy bre, 120
homotopy functor, 215
homotopy inverse, 28
homotopy lifting property = HLP, 66
homotopy pushout, 85
homotopy type, 28
Hopf algebra, 484
primitive element, 485
duality, 485
562 Index
group-like element, 486
pairing, 485
Hopf bration, 332
Hopf invariant, 517
Hopf space, 90
H-space, 90
Hurewicz bration, 115
Hurewicz homomorphism, 496
identication, 6
immersion, 363
incidence number, 301
index, 463
initial condition, 115
interior, 2
interior point, 2
Invariance of dimension, 252
Invariance of domain, 251
invariant scalar product, 19
involution
antipodal, 21
isotropy group, 17
join, 86, 345
Jordan curve, 250
Knneth formula, 298
homology, 242
KernelCokernel Lemma, 281
Klein bottle, 58
Kneser graph, 264
Kronecker pairing, 418
k-space, 187
Lebesgue lemma, 47
Lebesgue number, 47
Lefschetz xed point index, 179
left action, 17
left translation, 15, 17
lens space, 368
Lie group, 361
lifting, 50, 63, 101
initial condition, 50, 66
linking number, 404
local coordinate system, 358
local degree, 259
local parametrization, 358
local section, 332, 364
locally Euclidean, 358
locally nite, 318
locally trivial, 62, 332
loop, 42
loop space, 89
manifold, 358
boundary, 369
closed, 369
double, 391
interior, 369
orientable, 372
product, 359
smooth, 359
Stiefel, 365
manifold with boundary, 369
map
antipodal, 261
attaching, 203, 204
cellular, 210
characteristic, 199, 203, 204
closed, 2
continuous, 2
continuous at a point, 2
differentiable, 358, 369
n-equivalence, 144
odd, 261
open, 2
proper, 14
quotient, 6
regular point, 363
regular value, 363
singular value, 363
smooth, 358, 359, 370
transverse, 385
uniformly continuous, 4
weak equivalence, 144
Index 563
mapping cone, 93, 111, 287
mapping cylinder, 81, 110
MayerVietoris sequence, 266, 407
measure zero, 365
metric, 3
metric space, 3
Mittag-Lefer condition (=ML), 409
ML, 409
module
graded, 283
monoid, 90
in h-TOP, 49
Moore space, 256
morphism
connecting, 284
multiplicative structure, 409
MVS, 266
NDR, 114
NDR-presentation, 114
neighbourhood, 2
basis, 2
c, 3
open, 2
neighbourhood deformation retract, 114
nerve, 197
net, 4
convergent, 4
normal bundle, 376
normalizer, 21
null bordant, 521
null bordism, 521
null homotopic, 28
null homotopy, 28
numerable, 86, 318
numeration, 318
obstruction, 205
omega-spectrum, 181
open simplex, 198
orbit, 17
orbit category, 19, 75
orbit space, 17
orbit type
nite, 17
orientation, 372, 394
along a subset, 394
boundary, 373
complex structure, 372
C-oriented, 467
local, 393
opposite, 372
pre-image, 374
product, 372
R-oriented, 467
standard, 372
sum, 372
orientation covering, 342, 394
orthogonal group, 16
paracompact, 15
partition of unity, 318
generalized, 320
subordinate, 318
path, 25
constant, 25
homotopy, 41
inverse, 25
product, 25
path component, 26
path connected, 26
Plcker coordinates, 366
Poincar duality, 446
point nite, 318
pointed
homotopy, 31
homotopy equivalence, 31
map, 31
product, 31
space, 31
sum, 31
polyhedron, 199
Pontrjagin class, 480
Pontrjagin number, 492
564 Index
PontrjaginThom construction, 530
pre-spectrum, 181
preparation theorem, 149
primitive, 485
principal bundle, 328
principal covering, 64
product orientation, 401
products
cap product, 438, 442
cup product, 409
cup product in singular
cohomology, 422
external, 413
Kronecker pairing, 418
projective bundle, 471
projective plane, 58
projective resolution, 287
projective space, 21, 22
homogeneous coordinates, 21
proper, 14
properly discontinuous, 64
pseudo-circle, 27
pullback, 8
Puppe-sequence, 95
pushout, 9
quasi-compact, 11
quotient map, 6
universal property, 6
quotient space, 6
ramication index, 313
ramication point, 313
ramied covering, 313
rank theorem, 363
reduced cohomology groups, 407
reduced homology, 252
regular point, 363
regular value, 363
representation, 19
orthogonal, 19
unitary, 19
retract, 6, 32
retraction, 6, 32
RiemannHurwitz formula, 314
Riemannian metric, 351
right action, 17
saturated, 6
scaling function, 169
section, 6, 32
local, 332, 364
Segre embedding, 366
semi-locally simply connected, 70
separating function, 523
separation axiom, 3
Serre class, 504
saturated, 504
Serre bration, 115
exact sequence, 130
set
bounded, 4
closed, 2
dense, 2
nowhere dense, 2
open, 2
set map, 2
sheet, 63
short exact, 275
shrinkable, 32
shufe, 240
shufe morphism, 240
signature, 458
simplex, 197
combinatorial boundary, 198
dimension, 197
face, 197, 224
singular, 224
standard simplex, 224
simplicial complex
barycentric subdivision, 197
geometric realization, 198
simplicial complex, 197, 307
nite, 197
Index 565
locally nite, 197
q-simplex, 307
subcomplex, 197
vertex, 307
simplicial diagram, 321
simply connected, 43
1-connected, 43
singular value, 363
skeleton, 205
skeleton ltration, 205, 416
slant product, 441
smash product, 31, 193
smooth manifold, 359
space
weakly hausdorff, 187
adjunction, 7
-simple, 110
comb space, 33
compact, 11
compactly generated, 187
completely regular, 3
component, 27
connected, 27
covering, 63
decomposition, 27
discrete, 2
EilenbergMac Lane, 217
G-space, 17
Hausdorff, 3
homeomorphic, 2
homogeneous, 18
k-closed subset, 187
k-open subset, 187
locally compact, 12
locally connected, 64
locally path connected, 64
metric, 3
metrizable, 4
n-connected, 143
normal, 3
n-simple, 110
over, 32
paracompact, 15
pointed, 31
quasi-compact, 11
quotient, 6
regular, 3
separated, 3
simple, 110
topological, 2
totally disconnected, 27
transport-local, 70
under, 32
well-pointed, 102
spaces
triple of, 32
pair of, 31
spectrum, 181
EilenbergMac Lane, 221
spheres, 181
suspension, 182
sphere
stable homotopy groups, 517
sphere spectrum, 181
splitting, 276
splitting lemma, 276
splitting principle, 477
stabilizer, 17
stable unitary group, 475
stably homotopic, 159
stably homotopy equivalent, 159
standard resolution, 292
star, 198
star-shaped, 29
stereographic projection, 35
Stiefel manifold, 22, 365
StiefelWhitney class, 479
StiefelWhitney number, 492
structure group, 334
reduction, 334
subbasis of a topology, 2
subcomplex, 197, 199
566 Index
submanifold, 359
of type I, 370
of type II, 370
smooth, 359
submersion, 363
subnet, 4
subspace, 5
G-, 17
G-invariant, 17
G-stable, 17
locally closed, 13
relatively compact, 11
sum-lemma, 277
support of a function, 318
surface, 358
genus, 312
hyper-elliptic, 316
hyper-elliptic involution, 316
suspension, 87, 107, 288
homological, 249
suspension spectrum, 182
tangent bundle, 374
tangent space, 362
tangent vector, 362
pointing inwards, 370
pointing outwards, 370
telescope, 270
test map, 187
test space, 187
theorem
Alexander duality, 446
BlakersMassey, 133, 148
BorsukUlam, 464
Bott periodicity, 357
cellular approximation, 210
E. H. Brown representability, 216
Ehresmann, 382
Freudenthal suspension, 154
Hurewicz, 497
LerayHirsch, 427, 430
M. Brown, 251
Poincar duality, 446
Sard, 365
Schoenies, 251
Thom isomorphism, 431
Tietze, 3
Tychonoff, 12
Urysohn, 3
Whitehead, 498
Thom class, 431
strict, 433
tangent bundle, 464
Thom homomorphism, 434
Thom isomorphism, 431
topological group, 15
topological product, 8
topological subgroup, 16
topological sum, 9
topology, 1
basis, 2
coarser, 2
colimit, 12
compact open, 37
discrete, 2
ner, 2
product, 8
quotient, 6
relative, 5
subbasis, 2
subspace topology, 5
torsion product, 292
torus, 16
total space, 32
transition function, 358
transition map, 336
translation map, 329
transport, 107
transport category, 70
transport functor, 67, 107, 119
transport space, 70
transport-local, 70
Index 567
transport-simple, 70
transverse, 385
to a submanifold, 385
tree, 209
triangulation, 198
trivial G-space, 332
trivial over, 62
trivialization, 62
tubular map, 376
tubular neighbourhood, 377
typical bre, 63
unitary group, 16
universal coefcients, 295, 296
cohomology, 418
homology, 238
universal covering, 78
universal group, 355
vector bundle, 336
nite type, 336
inverse, 351
numerable, 336
orientable, 336
orientation, 336
subbundle, 337
tautological, 339
vector bundles
stably equivalent, 356
vector eld, 136
vertex, 197
Wang sequence, 426
weak equivalence, 144
wedge, 31
well-pointed, 102
Weyl group, 21
Whitehead complex, 199
subcomplex, 199
Whitney sum, 351
winding number, 50, 258, 403

Algebraic Topology EMS Textbooks in Mathematics

Uploaded by

Copyright:

Available Formats

Algebraic Topology EMS Textbooks in Mathematics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Algebraic Topology EMS Textbooks in Mathematics

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

Dieck_titelei 1.8.

2008 13:23 Uhr Seite 1

is an interior point of . A subset is

= C {0] is open. Similarly : R S

T and called the sum

non-empty subspaces. Then

is closed if and only if the

). For each , Y the map X X Y , . (.. ,)

A subset H of a space G is called locally closed, if each . H has a neigh-

= 0 for > n. Give R

)) there exists an accumulation value . of (.

the extension to the one-point compactication. Then

: G G, . g. by g G in a topological group is continuous, and the rules

of the positive real numbers is an open subgroup of R

O(n),(O(k) O(n k))

U(n),(U(k) U(n k))

if the functor is clear from the context.

= X {+], i.e., to X with an additional base point added (topological sum),

) [ J) be a family of pointed spaces. Let

where all but one a

are equal to the base point. There is a

. If J is nite, then this map is a

) = (1. 0), then it is not a homeomorphism.

. Show that this map is a home-

. Then there exists a homeomorphism of pairs (D

. Verify that a ray from0 in-

(0) is homotopy equivalent to the product of k factors S

(2.4.10) Corollary. Let h: (X . Y ) 1 7 be a pointed homotopy. Then

) is always bijective and

is the sum in TOP

. Let X be a Hausdorff space. Then the tautological map t is a homeomorphism.

X be the inclusions. Then

) be functors into a groupoid such that h

(see (2.6.4)). We consider edge-paths in the subdivided square 1 1 which

. +) G into a group G (= a groupoid with a

and apply (2.6.1) to

(2.7.1) Proposition. The functor is an isomorphism.

The classical approach to

(:). Then k J( ) mod n. If, conversely, this congruence is

= C 0 along the covering 1 : C C

be a continuously differentiable path with initial point 1. Then

of genus g _ 1 has the presentation

) is the so-called Euler characteristic.

[ J) be a family of subspaces of X such that the interiors X

are path connected and + X

1( id) 1(g id)

: 1 1, . g. is an automorphism of . We thus obtain a

C of pointed maps is exact at T if the image

is the isotropy group of the

) is a universal covering of the n-dimensional torus T

: : z.. In this way S

7( ) = X 1 Y,(.) ~ (.. 1).

C) is the quotient of T 1 C with respect to

. We obtain an induced map : 7(

. In the case that the interiors X

cover X, the space X is a pushout in the