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.
Jørn Justesen and Tom Høholdt, A Course In ErrorCorrecting Codes
Markus Stroppel, Locally Compact Groups
Peter Kunkel and Volker Mehrmann, DifferentialAlgebraic 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
Dieck_titelei 1.8.2008 13:23 Uhr Seite 3
Author:
Tammo tom Dieck
Mathematisches Institut
GeorgAugustUniversität Göttingen
Bunsenstrasse 3–5
37073 Göttingen
Germany
Email: tammo@unimath.gwdg.de
2000 Mathematics Subject Classification: 5501, 5701
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 9783037190487
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, reuse 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
ETHZentrum FLI C4
CH8092 Zürich
Switzerland
Phone: +41 (0)44 632 34 36
Email: info@emsph.org
Homepage: www.emsph.org
Typeset using the author’s TEX files: I. Zimmermann, Freiburg
Printed on acidfree paper produced from chlorinefree pulp. TCF
°°
Printed in Germany
9 8 7 6 5 4 3 2 1
Dieck_titelei 1.8.2008 13:23 Uhr Seite 4
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 deﬁne 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 Poincaré, is without doubt one of the most inge
nious and inﬂuential 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 ﬁnal 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 ﬁbration and coﬁbration, which are at ﬁrst 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 ﬁrst main theorem of algebraic topology is the Brouwer–Hopf degree the
orem. We prove this theorem by elementary methods from homotopy theory. It is
a fairly direct consequence of the Blakers–Massey 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 selfmaps of a positive dimensional sphere under the homotopy
relation has the structure of a (homotopically deﬁned) 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 deﬁned at the place where they appear even if there
is no indication of a formal deﬁnition. In addition, there is a list of standard or global
symbols. The problem sections contain exercises, examples, counterexamples 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 ﬁnd 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 author’s home page.
I would like to thank Irene Zimmermann and Manfred Karbe for their help and
effort in preparing the manuscript for publication.
Göttingen, 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 Classiﬁcation 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 Coﬁbre Sequence . . . . . . . . . . . . . . . . . . . . . . . 92
4.7 The Fibre Sequence . . . . . . . . . . . . . . . . . . . . . . . . 97
5 Coﬁbrations and Fibrations 101
5.1 The Homotopy Extension Property . . . . . . . . . . . . . . . . 101
5.2 Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.3 Replacing a Map by a Coﬁbration . . . . . . . . . . . . . . . . . 110
5.4 Characterization of Coﬁbrations . . . . . . . . . . . . . . . . . . 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 CWComplexes . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.4 Weak Homotopy Equivalences . . . . . . . . . . . . . . . . . . . 207
8.5 Cellular Approximation . . . . . . . . . . . . . . . . . . . . . . 210
8.6 CWApproximation . . . . . . . . . . . . . . . . . . . . . . . . 211
Contents ix
8.7 Homotopy Classiﬁcation . . . . . . . . . . . . . . . . . . . . . . 216
8.8 Eilenberg–Mac 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 Coefﬁcients . . . . . . . . . . . . . . . . . . . . 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 Mayer–Vietoris 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 Coefﬁcients . . . . . . . . . . . . . . . . . . . . . . . 295
11.10 The Künneth 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
16.5 Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . 402
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 Eilenberg–Mac 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
18.6 Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . 455
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 Stiefel–Whitney 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 Eilenberg–Mac 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 pointset 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 deﬁnition. 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 inﬂuence
several other of its geometric investigations.
• The theory of ﬁbre bundles and vector bundles makes essential use of the
concept of a transformation group. Important information about a manifold
is codiﬁed 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
ﬁnite 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 · Ç 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 ﬁnite 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 ﬁnite 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 preimage
¡
1
(V ) of each open set V of Y is open in X. Dually: A map is continuous
if the preimage 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 speciﬁed or obvious. A set map is a map which is not assumed to
be continuous at the outset.
We ﬁx 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 ·
ı
. A point in ·
ı
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) Ç V ; it sufﬁces to consider a neighbourhood basis of . and ¡(.).
Suppose O
1
and O
2
are topologies on X. If O
1
Ç O
2
, then O
2
is ﬁner than
O
1
and O
1
coarser than O
2
. The topology O
2
is ﬁner 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 ﬁnest
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
) Onepoint subspaces are closed.
(T
2
) Any two points have disjoint neighbourhoods.
(T
3
) Given a point . ÷ X and a closed subset · Ç 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 satisﬁes 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 ,= · Ç X closed, . … ·, there exists a continuous function ¡ : X ÷Œ0. 1 such
that ¡(.) = 1 and ¡(·) = {0].
A remarkable consequence of the separation property T
4
is the existence of
many realvalued 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 ÷Œ0. 1
with ¡(·) Ç {0] and ¡(T) Ç {1].
(1.1.2) Theorem (Tietze). Let X be a T
4
space and · Ç X closed. Then each
continuous map ¡ : · ÷Œ0. 1 has a continuous extension ¡ : X ÷Œ0. 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 : X×X ÷Œ0. oŒ
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 Ç X open with respect to J if for each
. ÷ X there exists c > 0 such that U
t
(.) Ç 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 cneighbourhoods 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
(.)
4 Chapter 1. Topological Spaces
and 0 < j < c ÷ J(.. ,). Then, by the triangle inequality, U
)
(,) Ç 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
(.) Ç U. For metric spaces our deﬁnition of continuity is
equivalent to the familiar deﬁnition of calculus: A map ¡ : X ÷ Y between
metric spaces is continuous at a ÷ X if for each c > 0 there exists ı > 0 such that
J(a. .) < ı implies J(¡(a). ¡(.)) < c. Continuityonlydepends onthe underlying
topology. But a metric is a ﬁner and more rigid structure; one can compare the size
of neighbourhoods of different points and one can deﬁne uniform continuity. A
map ¡ : (X. J
1
) ÷ (Y. J
2
) between metric spaces is uniformly continuous if for
each c > 0 there exists ı > 0 such that J
1
(.. ,) < ı 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 deﬁne 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, nonempty, closed sets in X, then
¡ : X ÷Œ0. 1. . ÷J(.. ·)(J(.. ·) ÷J(.. T))
1
is a continuous function with ¡(·) = {0] and ¡(T) = {1]. Let 0 < a < b < 1.
Then Œ0. aŒ and b. 1 are open in Œ0. 1, and their preimages 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 Ç U.
Anet with directed index set 1 in X is a map 1 ÷X, i ÷.
i
. We write (.
i
)
i÷J
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
)
i÷J
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 ﬁnal 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 ﬁnal 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 deﬁne 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 · Ç X a subset. Then
O[· = {U Ç · [ 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 deﬁnition one veriﬁes:
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 ) Ç ·. Then ¡ is continuous
if and only if ç : 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 · Ç T Ç 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 ﬁnite.
6 Chapter 1. Topological Spaces
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 deﬁned, formally, it takes some
time to work with it. In several branches of mathematics quotient objects are more
difﬁcult to handle than subobjects. Even if one starts with a nice and wellknown
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 Ç Y [ ¡
1
(U) open in X] is a topology on Y . This is the ﬁnest
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 identiﬁcation or quotient map if it has the following property: U Ç Y
open = ¡
1
(U) Ç 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 ç : X,1
(
÷ Y
such that ç¸ = ¡ . The quotient spaceX,1 is deﬁned to be the set X,1 together
with the quotient topology of the canonical map ¸: X ÷X,1.
If · Ç 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 ¸ : · Ç X be an inclusion and ¡ : · ÷Y a continuous map. We identify in
the topological sumX÷Y 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) X÷Y 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 · Ç 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 Ç Y , hence a quotient
map.
4. The exponential map exp: C ÷ C
·
= C · {0] is open. Similarly ¸: R ÷ S
1
,
t ÷ exp(2¬i 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 satisﬁes ˛ ı q = ¸.
8 Chapter 1. Topological Spaces
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 Š 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 deﬁned
as (.
}
[ ¸ ÷ J) ÷ (¡
}
(.
}
) [ ¸ ÷ J). Given maps ¡
}
: Y ÷ X
}
we denote by
(¡
}
) = (¡
}
[ ¸ ÷ J): Y ÷
}
X
}
the map with components pr
i
ı(¡
}
) = ¡
i
.
The family of all preimages pr
1
}
(U
}
), U
}
Ç 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 inﬁnite J, open sets in the product are quite large; a
product
U
}
, U
}
Ç 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
}
ı¡ 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 identiﬁcation R
k
× R
I
= R
k÷I
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
G
Y
¿
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 : · Ç X and a: · ÷
Y such that ga = ¡ [·is given. The assignment (o : X ÷7) ÷(J ıo : X ÷Y )
sets up a bijection between sections of G with Jo[· = a and maps ç : X ÷ Y
such that ç[· = a and gç = ¡ . Þ
(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 X×X.
Let (X
}
[ ¸ ÷ J) be a family of nonempty pairwise disjoint spaces. The set
O = {U Ç lX
}
[ U ¨ X
}
Ç 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 veriﬁed from
the deﬁnitions. 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 ÷
¹
T and called the sum
of X and T under · (the sum in the category TOP
¹
of spaces under ·). Þ
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
Ç 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
}
10 Chapter 1. Topological Spaces
and g
k
}
ı 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
}
ı
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
. Þ
(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
Ç 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
}
Ç U
}
, ¸ = 1. 2, and a homeomorphism ç : V
1
÷ V
2
. Then
X = U
1
L
ç
U
2
is obtained from the topological sum U
1
÷ U
2
by identifying
· ÷ V
1
with ç(·) ÷ V
2
. Let U
1
= U
2
= R
n
and V
1
= V
2
= R
n
· 0. Let ç = id.
Then the graph of ç in R
n
×R
n
is not closed. The resulting locally Euclidean space
is not Hausdorff. If we use ç(.) = . [.[
2
, then the result is homeomorphic to
S
n
(see (2.3.2)). Þ
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
}
. Þ
Problems
1. Let (X
¸
[ ¸ ÷ J) be spaces and ·
¸
Ç X
¸
nonempty subspaces. Then
¸÷J
¨
·
¸
=
¸÷J
·
¸
. The product
¸÷J
·
¸
is closed if and only if the ·
¸
are closed.
2. The projections pr
k
:
¸
X
¸
÷ X
k
are open maps, and in particular quotient maps.
(The X
¸
are nonempty.)
3. A space X is separated if and only if the diagonal D = {(.. .) [ . ÷ X] is closed in
X×X. 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
¸
Y
¸
Š 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 reﬁnement of ·if for each k ÷ 1
there exists ¸ ÷ J such that T
k
Ç ·
}
. 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 Ç J and T
k
= ·
k
for k ÷ 1.
We say T is ﬁnite or countable if 1 is ﬁnite or countable. A covering · is locally
ﬁnite if each . ÷ U has a neighbourhood U such that U ¨·
}
= 0 only for a ﬁnite
number of ¸ ÷ J. It is called pointﬁnite if each . ÷ X is contained only in a ﬁnite
number of ·
}
.
A space X is compact if each open covering has a ﬁnite subcovering. (In some
texts this property is called quasicompact.) By passage to complements we see:
If X is compact, then any family of closed sets with empty intersection contains a
ﬁnite family with empty intersection. A set · in a space X is relatively compact
if its closure is compact. We recall from calculus the fundamental Heine–Borel
Theorem: The unit interval 1 = Œ0. 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 ﬁnite. Let
X be compact, · Ç 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
12 Chapter 1. Topological Spaces
open neighbourhoods U of T and V of C such that U × V is covered by a ﬁnite
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
Ç X
2
Ç . Recall that X carries the
colimittopology with respect to the ﬁltration (X
i
) if · Ç 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
Ç X
2
Ç .
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, 1¨U 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 compactiﬁcation
of X if Y is compact and ¡(X) dense in Y .
The following theoremyields a compactiﬁcation by a single point. It is called the
Alexandroff compactiﬁcation or the onepoint compactiﬁcation. The additional
point is the point at inﬁnity. In a general compactiﬁcation ¡ : X ÷ Y , one calls
the points in Y · ¡(X) the points at inﬁnity.
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 compactiﬁcation ¡ : 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
i÷1
.
(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
n1
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
n1
÷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
Ç X
2
Ç . Then the X
i
are subspaces of X. If X
i
Ç X
i¬1
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 R×R
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
Œ. The space R
c
with this topology
is not metrizable. The space has also the colimit topology with respect to the set of ﬁnite
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.
14 Chapter 1. Topological Spaces
1.5 Proper Maps
A continuous map ¡ : X ÷ Y is called proper if it is closed and the preimages
¡
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 Ç 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 ı ¡ is proper.
(2) If g ı ¡ is proper and ¡ surjective, then g is proper.
(3) If g ı ¡ 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 Ç 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 Ç Y has a compact preimage. 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
¬
÷Y
¬
the extension to the onepoint compactiﬁcation. 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) Ç Y . Then ¡ = i ı h ı ¸ is the canonical decomposition of ¡ . The map
¡ is proper if and only if ¸ is proper, h a homeomorphism, and ¡(X) Ç 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
}
Ç U
}
.
A pointﬁnite 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 ﬁnite reﬁnement. 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
16 Chapter 1. Topological Spaces
groups R
+
, C
+
and H
+
of the nonzero 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
{2¬i 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 ﬁnd, among others,
the symplectic groups Sp(n) and the Spinor groups Spin(n). The isomorphisms
SU(2) Š Spin(3) Š 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 nfold product S
1
× ×S
1
is called
an ndimensional 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 Ç G a subgroup, then H, with the subspace
topology, is a topological group (called a topological subgroup). If H Ç 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) Gspace (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 · Ç X and 1 Ç 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 Gspace 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 Gspace 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 ﬁnite number
of conjugacy classes, we say X has ﬁnite orbit type.
A subset · of a Gspace is called Gstable or Ginvariant if g ÷ G and a ÷ ·
implies ga ÷ ·. A Gstable subset · is also called a Gsubspace. For each
subgroup H of G there is an Hﬁxed point set of X,
X
1
= {. ÷ X [ h. = . for all h ÷ H].
Suppose X and Y are Gspaces. A map ¡ : X ÷ Y is called a Gmap or
a Gequivariant 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 Gspaces and Gequivariant maps form the category G TOP. This
category has products: If (X
}
[ ¸ ÷ J) is a family of Gspaces, then the topological
product
}
X
}
together with the diagonal action (g. (.
}
)) ÷ (g.
}
) is a product
in this category. A Gmap ¡ : 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
Ghomotopy H
t
: this is a homotopy such that each H
t
is a Gmap.
(1.8.1) Proposition. Let X be a Gspace, · Ç G and T Ç X. If T is open then
·T is open. The orbit map ¸: X ÷X,G is open.
18 Chapter 1. Topological Spaces
Proof. l
o
(T) is open, since l
o
is a homeomorphism. Hence
_
o÷¹
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 deﬁnition 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 Gaction 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) Š S
n
. In the complex
case we obtain a homeomorphism U(n ÷ 1),U(n) Š S
2n÷1
, in the quaternionic
case a homeomorphism Sp(n ÷ 1),Sp(n) Š S
4n÷3
. Other important homoge
neous spaces are the projective spaces, the Grassmann manifolds, and the Stiefel
manifolds to be discussed later. Þ
(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 Gstable subset of the Gspace X. Then ·,G carries the
subspace topology of X,G. In particular X
G
÷X ÷X,G is an embedding.
(4) Let T Ç X be closed and · Ç X. Then {g ÷ G [ g· Ç T] is closed in G.
(5) Let T Ç X be closed. Then {g ÷ G [ gT = T] is closed.
Proof. (1) The isotropy group G
x
is the preimage of . under the continuous map
G ÷ X, g ÷ g.. (2) The set X
¿
= {. ÷ X [ g. = .] is the preimage of
the diagonal under X ÷ X × X, . ÷ (.. g.), and X
1
=
_
¿÷1
X
¿
. (3) Let
C Ç ·,G be open with respect to the quotient map · ÷·,G. Then ¸
1
(C) Ç ·
is open, and we can write ¸
1
(C) = · ¨ U with an open subset U Ç X. We
have · ¨ U = · ¨ GU, since · is Gstable. 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÷¹
r
1
o
(T) = {g ÷ G [ g· Ç T] closed. (5)
The set {g [ gT = T] = {g [ gT Ç T] ¨ {g [ g
1
T Ç T] is closed, by (4).
(1.8.5) Proposition. Let r : G × X ÷X be a Gaction, · Ç G and T Ç 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 Gstable closed subset and U a neighbourhood of ·
in X. Then U contains a Gstable neighbourhood of ·.
Proof. (1) · × T Ç 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
homeomorphism·×X ÷·×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 · Ç X be closed.
Then ¸
1
¸(·) = G· is closed, by (2). Hence ¸(·) is closed, by deﬁnition of
the quotient topology. The preimages 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 Gstable neighbourhood of ·
contained in U.
The orbit category Or(G) is the category of homogeneous Gspaces G,H, H
closed in G, and Gmaps. There exists a Gmap 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 Gmap and each Gmap G,H ÷ G,1 has this form; moreover
1
o
= 1
b
if and only if a
1
b ÷ 1.
An action G×V ÷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 Gstable.
Let 1 be a right Gspace and J a left Gspace. We denote by 1 ×
G
J the
orbit space of the Gaction G × (1 × J) ÷1 × J, (g. (.. ,)) ÷(.g
1
. g,). A
Gmap ¡ : J
1
÷J
2
induces a continuous map
id ×
G
¡ : 1 ×
G
J
1
÷1 ×
G
J
2
. (.. ,) ÷(.. ¡(.)).
If 1 carries a left 1action which commutes with the right Gaction (i.e., k(.g) =
(k.)g), then 1 ×
G
J carries an induced 1action (k. (.. ,)) ÷ (k.. ,). This
construction can in particular be applied in the case that 1 = 1, G a subgroup
20 Chapter 1. Topological Spaces
of 1 and the G and 1actions 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 1space as a Gspace.
The natural adjunction
TOP
1
(ind
1
G
X. Y ) Š TOP
G
(X. res
1
G
Y )
sends a Gmap ¡ : X ÷ Y to the 1map (k. .) ÷ k¡(.); in the other direction
one restricts a map to X Š G ×
G
X Ç 1 ×
G
X. (Here TOP
1
denotes the set of
1equivariant 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
}
Ç Wg
1
, g ÷ g
}
W. Therefore the g
}
W cover the group.
Let V Ç G be open and g ÷ V . There exists a compact neighbourhood W
of e such that W = W
1
and gW
2
Ç 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. Ç W
2
. and hence g. = ¸(g) an interior point of
gW
2
. Ç 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 Gspace. Restricting the group action to H,
we obtain an Hspace X. The orbit space H\X carries then an induced G,Haction.
2. Let a pushout in TOP be given with Gspaces ·, T, X and Gmaps ¸ , ¡ :
·
¸
¸
T
J
X
1
Y .
Let G be locally compact. Then there exists a unique G action on Y such that J. J become
Gmaps. The diagram is then a pushout in GTOP. Hint: (2.4.6)
1.9. Projective Spaces. Grassmann Manifolds 21
3. Let Y be a 1space and G a subgroupp of 1. Then 1 ×
G
Y ÷ 1,G × Y , (k. ,) ÷
(kG. k,) is a 1homeomorphism. If X is a Gspace, then
1 ×
G
(X × Y ) ÷(1 ×
G
X) × Y. (g. (.. ,)) ÷((g. .). g,)
is a 1homeomorphism.
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 Ç H. Hence
{g ÷ G [ g
1
Hg Ç 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 ﬁxed set X
!
, by
restricting the given Gaction to NH. The action
G,H × WH ÷G,H. (gH. nH) ÷gnH
is a free right action by Gautomorphisms of G,H.
1.9 Projective Spaces. Grassmann Manifolds
Let 1(R
n÷1
) = R1
n
be the set of onedimensional subspaces of the vector space
R
n÷1
. A onedimensional 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
n÷1
) as the orbit space of the action
R
+
× (R
n÷1
· 0) ÷R
n÷1
· 0. (z. .) ÷z..
The quotient map ¸: R
n÷1
· 0 ÷ 1(R
n÷1
) provides 1(R
n÷1
) with the quotient
topology. The space R1
n
is the ndimensional real projective space. We set
¸(.
0
. . . . . .
n
) = Œ.
0
. . . . . .
n
 and call .
0
. . . . . .
n
the homogeneous coordinates
of the point Œ.
0
. . . . . .
n
.
In a similar manner we consider the set 1(C
n÷1
) = C1
n
of onedimensional
subspaces of C
n÷1
as the orbit space of the action
C
+
× (C
n÷1
· 0) ÷C
n÷1
· 0. (z. :) ÷z:.
We have again a quotient map ¸: C
n÷1
·0 ÷1(C
n÷1
). The space C1
n
is called
the ndimensional complex projective space. (It is 2ndimensional as a manifold.)
We describe the projective spaces in a different manner as orbit spaces. The
subgroup G = {˙1] Ç R
+
acts on S
n
Ç R
n÷1
by (z. .) ÷ z., called the
antipodal involution. The inclusion i : S
n
÷R
n÷1
induces a continuous bijective
map  : S
n
,G ÷ (R
n÷1
· 0),R
+
. The map ¸ : R
n÷1
· 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
2n÷1
÷ S
2n÷1
, (z. :) ÷ z: of S
1
on the unit
sphere S
2n÷1
Ç C
n÷1
· 0.
22 Chapter 1. Topological Spaces
Projective spaces are homogeneous spaces. Consider the action of O(n÷1) on
R
n÷1
by matrix multiplication. If V ÷ 1(R
n÷1
) is a onedimensional space and
· ÷ O(n ÷1), then ·V ÷ 1(R
n÷1
). We obtain an induced action
O(n ÷1) × 1(R
n÷1
) ÷1(R
n÷1
).
This action is transitive. The isotropy group of Œ1. 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(n÷1). The assignment
· ÷·e
1
induces an O(n ÷1)equivariant homeomorphism
b : O(n ÷1),(O(1) × O(n)) Š 1(R
n÷1
).
The action of O(n ÷ 1) on 1(R
n÷1
) is continuous; this follows easily from the
continuity of the action O(n ÷ 1) × (R
n÷1
· 0) ÷ R
n÷1
· 0 and the deﬁnition
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(n÷1)equivariant
homeomorphism U(n ÷1),(U(1) × U(n)) Š 1(C
n÷1
).
Finally, one can deﬁne the quaternionic projective space H1
n
in a similar
manner as a quotient of H
n÷1
· 0 or as a quotient of S
4n÷3
.
We generalize projective spaces. Let W be an ndimensional real vector space.
We denote by G
k
(W) the set of kdimensional subspaces of W. We deﬁne 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 kframes in W. We
have a projection ¸: V
k
(W) ÷G
k
(W) which sends (n
1
. . . . . n
k
) to the subspace
Œn
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
) deﬁned by (·. (·
1
. . . . . ·
k
)) ÷
(··
1
. . . . . ··
k
) and such that ¸ becomes O(n)equivariant. The isotropy groups
of (e
1
. . . . . e
k
) and Œe
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)
Š
V
k
(R
n
)
O(n),(O(k) × O(n ÷ k))
Š
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 kdimensional subspaces of R
n
.
In a similar manner we can work with the kdimensional 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)
Š
V
k
(C
n
)
U(n),(U(k) × U(n ÷ k))
Š
G
k
(C
n
).
Chapter 2
The Fundamental Group
In this chapter we introduce the homotopy notion and the ﬁrst 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 ÷ Œ0. 1. (One may
interpret this as a “timedependent” 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 hTOP.
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 ﬂexibility. 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 viewpoint. 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 deﬁne generalized group objects where multiplication
is only associative up to homotopy. And so on.
The passage from TOP to hTOP 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 “homotopykernels” (then called
homotopy ﬁbres) and “homotopycokernels” (then called homotopy coﬁbres). This
is the more modern viewpoint 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 ﬁrst of such functors. It was introduced by Poincaré,
in different context and terminology. In general it is difﬁcult 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 inﬂuences almost every other tool of algebraic topology (although we do not
always carry out this inﬂuence in this text).
The chapter contains two sections on pointset topology. We discuss standard
spaces like spheres, disks, cells, simplices; they will be used in many different
contexts. We present the compactopen topology on spaces of continuous maps;
they will be used for the dual deﬁnition 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: Œa. 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 Œ0. 1 ÷ X,
t ÷ u((1 ÷ t )a ÷ t b). In the general theory we mostly use the unit interval. If
u: Œ0. 1 ÷ X is a path from . to ,, then the inverse path u

: t ÷ n(1 ÷ t ) is a
path from , to .. If ·: Œ0. 1 ÷ X is another path from , to :, then the product
path u + ·, deﬁned 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 Œ. the path component of the point ..
Aspace X is said to be path connected or 0connected 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 ndimensional cube to 1
n
.)
A map ¡ : X ÷ Y induces ¬
0
(¡ ): ¬
0
(X) ÷ ¬
0
(Y ), Œ. ÷ Œ¡(.). 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 · Ç 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: Œ0. a ÷ X, ·: Œ0. b ÷ X with u(a) = ·(0)
and a. b _ 0. Their composition · ı u = n is the path Œ0. 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: Œ0. a ÷ X with u(0) = .. u(a) = , for some a _ 0; and
composition of morphisms is as deﬁned before; the path Œ0. 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. Þ
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.
2.1. The Notion of Homotopy 27
Aspace is connected if it is not the topological sumof two nonempty 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: · Ç R is connected if and only if · is an
interval. (An interval is a subset which contains with .. , also Œ.. ,.)
(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 · Ç T Ç
¨
·. 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 = Œ÷1. 0 × 0 L 0 × Œ÷1. 1 L {(.. sin(¬.
1
) [ 0 < . _ 1]
is connected but not path connected. The union S of X with {˙1] × Œ÷2. 0 L
Œ÷2. 2 × {÷2] is called the pseudocircle. The complement R
2
· S has two path
components.
A pseudocircle S has a sequence 1
1
¸ 1
2
¸ of compact neighbourhoods
with
_
i
1
i
= S and 1
i
homeomorphic to S
1
× Œ0. 1. Þ
Let X and Y be topological spaces and ¡. g: X ÷ Y continuous maps. A
homotopy from ¡ to g is a continuous map
H: X × Œ0. 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 timedependent family
of maps. One also says that ¡ is deformed continuously into g. Another (dual)
viewpoint of a homotopy is: a parametrized family of paths. We use the letter 1 for
the unit interval Œ0. 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
28 Chapter 2. The Fundamental Group
course, deﬁne homotopies with Œ0. 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(.. 1÷t )
shows g . ¡ . Let 1: ¡ . g and 1: g . h be given. The product homotopy
1 + 1 is deﬁned 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 Œ¡  and called the homotopy class of ¡ .
We denote by ŒX. Y  the set of homotopy classes Œ¡  of maps ¡ : X ÷ Y . A
homotopy H
t
: X ÷Y is said to be relative to · Ç 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 hTOP, when
composition of homotopy classes is induced by composition of representing maps.
If ¡ : X ÷ Y represents an isomorphism in hTOP, then ¡ is called a homotopy
equivalence or hequivalence. 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 deﬁne a category
W(X. Y ). The objects are the continuous maps ¡ : X ÷ Y . A morphism from ¡
to g is a homotopy H: X × Œ0. a ÷Y with H
0
= ¡ and H
o
= g. Composition
is deﬁned as in (2.1.1). Þ
As in any category we also have the Homfunctors in hTOP. Given ¡ : X ÷Y ,
we use the notation
¡
+
: Œ7. X ÷Œ7. Y . g ÷¡g. ¡
+
: ŒY. 7 ÷ŒX. 7. h ÷h¡
2.1. The Notion of Homotopy 29
for the induced maps
2
. The reader should recall a little reasoning with Hom
functors, as follows. The map ¡ is an hequivalence, i.e., an isomorphism in
hTOP 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 hequivalence. If two of the maps
¡ : X ÷Y , g: Y ÷7, and g¡ are hequivalences, 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
ŒY.
}÷J
X
}
 ÷
}÷J
ŒY. X
}
. Œ¡  ÷(Œ¸
i
ı ¡ )
is a welldeﬁned bijection. Let i
k
: X
k
÷
}÷J
X
}
be the canonical inclusion of
the kth summand. Then
Œ
}÷J
X
}
. Y  ÷
}÷J
ŒX
}
. Y . Œ¡  ÷(Œ¡ ı i
k
)
is a welldeﬁned bijection. In other words: sum and product in TOP also repre
sent sum and product in hTOP. (Problems arise when it comes to pullbacks and
pushouts.)
Let 1 be a point. Amap 1 ÷Y can be identiﬁed with its image and a homotopy
1 × 1 ÷ Y can be identiﬁed with a path. The Homfunctor Œ1. ÷ is therefore
essentially the same thing as the functor ¬
0
.
2.1.5 Linear homotopy. Given maps ¡. g: X ÷ ·, · Ç R
n
. Suppose that
the linesegment 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 · Ç R
n
is starshaped with respect to a
0
÷ · if for each a ÷ · the
linesegment from a
0
to a is contained in ·. If · is starshaped, then H(a. t ) =
(1 ÷ t )a ÷ t a
0
is a null homotopy of the identity. Hence starshaped sets are
contractible. A set C Ç R
n
is convex if and only if it is starshaped 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 ). Þ
The reader should now recall the notion of a quotient map (identiﬁcation), 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
ı ¸ 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.
30 Chapter 2. The Fundamental Group
Proof. The product ¸×id: X×1 ÷Y ×1 is an identiﬁcation, since 1 is compact.
The composition H ı (¸ × 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 ,= · Ç X is always mapped into itself, H
t
(·) Ç ·.
Suppose H
1
is constant on ·. Then the projection ¸: X ÷ X,· (· identiﬁed to
a point) is an hequivalence.
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 ¸ ı 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 Ç R
n
be open. The path components of U are open and coincide with the
components. The set of path components is ﬁnite or countably inﬁnite. An open subset of R
is a disjoint union of open intervals.
6. List theorems of pointset 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) Ç GL
n
(R) and U(n) Ç GL
n
(C) are homotopy equivalences.
Let 1(n) denote the space of positive deﬁnite real (n. n)matrices. Then O(n) × 1(n) ÷
GL
n
(R), (X. 1) ÷ X1 is a homeomorphism; 1(n) is starlike with respect to the unit
matrix.
11. There exist contractible and noncontractible 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 ŒX. Y 
0
the set of pointed homotopy classes (ﬁxed base points assumed) or
by Œ(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 hTOP
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 ˇ from TOP
0
to TOP. They are adjoint
TOP
0
(˛(X). Y ) Š TOP(X. ˇY ), 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 identiﬁed 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
hTOP
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 pointset
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
¡(·) Ç 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
Œ(X. ·). (Y. T) for the associated homotopy sets and hTOP(2) for the homotopy
category.
If (X. ·) is a pair, we usually consider the quotient space X,· as a pointed
space (· identiﬁed to a point) with base point {·]. If · = 0, then X,· = X
÷
is
X with a separate base point.
32 Chapter 2. The Fundamental Group
(2.2.1) Note. A continuous map ¡ : (X. ·) ÷(Y. +) into a pointed space induces
a pointed map
¨
¡ : X,· ÷ Y . The assignment ¡ ÷
¨
¡ induces a bijection
Œ(X. ·). (Y. +) Š ŒX,·. Y 
0
. A veriﬁcation 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
n÷n
. d1
n÷n
). In a similar manner we treat other conﬁgurations, e.g., triples
(X. ·. T) of spaces · Ç T Ç X and the category TOP(3) of triples.
Let 1 and T be ﬁxed 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 identiﬁed
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 ﬁbre
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 ﬁbrewise or ﬁbre 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
deﬁnition is used for TOP
B
. A homotopy in TOP
B
will also be called ﬁbrewise
or ﬁbre preserving. Again, being homotopic is an equivalence relation in these
categories. We denote by ŒX. Y 
1
the set of homotopy classes in TOP
1
, and
by ŒX. Y 
B
the set of homotopy classes in TOP
B
. The homotopy categories are
hTOP
1
and hTOP
B
. Note that a homotopy equivalence in TOP
B
, i.e., a ﬁbrewise
homotopy equivalence, from ¸: X ÷ T to q : Y ÷ T induces for each b ÷ T
a homotopy equivalence ¸
1
(b) ÷ q
1
(b) between the ﬁbres 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 Ç X we then call 1 a retract of X. The retraction r of
i : 1 Ç 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
Ç R
n÷1
· 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 ﬁbres of a shrinkable map
are contractible.
2.2. Further Homotopy Notions 33
Problems
1. Let ((X
¸
. .
¸
) [ ¸ ÷ J) be a family of pointed spaces. Let
_
/
¸
X
¸
be the subset of
those points (a
¸
) ÷
¸
X
¸
where all but one a
¸
are equal to the base point. There is a
canonical bijective continuous map
_
¸
X
¸
÷
_
/
¸
X
¸
. If J is ﬁnite, then this map is a
homeomorphism. If J is inﬁnite and (X
¸
. .
¸
) = (1. 0), then it is not a homeomorphism.
2. The canonical maps i
k
: X
k
÷
_
¸
X
¸
are embeddings.
3. The comb space X is deﬁned as T ×Œ0. 1 LŒ0. 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 Š X Y is not contractible. Since (X L Y ),Y is
homeomorphic to X, we see that it does not sufﬁce 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 Ç 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 hequivalence. Is (X. ·) hequivalent 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 hequivalence in TOP(2) although the
component maps 1 ÷1 and d1 ÷1 · {1,2] are hequivalences.
8. Let 1 Ç R
2
consist of k points. Show, heuristically, that the complement R
2
· 1 is
hequivalent to the kfold sum
_
k
1
S
1
.
9. Remove a point fromthe torus S
1
×S
1
and showthat the result is hequivalent to S
1
S
1
.
Is there an analogous result when you remove a point from S
m
× S
n
?
10. Construct an inclusion · Ç X which is a retract and a homotopy equivalence but not a
deformation retract.
11. Construct a map ¸: 1 ÷T such that all ﬁbres ¸
1
(b) are contractible but which does
not have a section. Construct an hequivalence ¸: 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
deformation retract.
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 ﬁnite. Show that in this case both spaces are quotients of
(l
¸
·
¸
) × T Š 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) Š X . 1,d1.
18. There is a canonical bijective continuous map (X×Y ),(X×TL·×Y ) ÷X,·.Y,T
34 Chapter 2. The Fundamental Group
(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] ndimensional disk
S
n1
= {. ÷ D
n
[ [.[ = 1] = dD
n
(n ÷ 1)dimensional sphere
1
n
= D
n
· S
n1
ndimensional cell
1
n
= {. ÷ R
n
[ 0 _ .
i
_ 1] ndimensional cube
d1
n
= {. ÷ 1
n
[ .
i
= 0. 1 for some i ] boundary of 1
n
^
n
= ^Œn = {. ÷ R
n÷1
[ .
i
_ 0.
i
.
i
= 1] ndimensional 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 Ç R
n
be a compact convex subset with nonempty
interior 1
ı
. Then there exists a homeomorphism of pairs (D
n
. S
n1
) ÷(1. d1)
which sends 0 ÷ D
n
to a preassigned . ÷ 1
ı
.
Proof. Let 1 Ç R
n
be closed and compact and 0 ÷ 1
ı
. Verify that a ray from0 in
tersects the boundary d1 of 1 in R
n
in exactly one point. The map ¡ : d1 ÷S
n1
,
. ÷ .,[.[ is a homeomorphism. The continuous map ç : S
n1
× Œ0. 1 ÷1,
(.. t ) ÷ t¡
1
(.) factors over q : S
n1
× Œ0. 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
n1
) Š
(1
n
. d1
n
). The simplex ^
n
is a compact convex subset with interior points in the
hyperplane {. ÷ R
n÷1
[
i
.
i
= 1]. From this fact we deduce a homeomorphism
(D
n
. dD
n
) Š (^
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
n÷1
= (0. . . . . 0. 1) ÷ R
n÷1
. We deﬁne the stere
ographic projection ç
1
: S
n
· {e
n÷1
] ÷ R
n
; the point ç
1
(.) is the intersec
tion of the line through e
n÷1
and . with the hyperplane R
n
× 0 = R
n
. One
computes ç
1
(.
1
. . . . . .
n÷1
) = (1 ÷ .
n÷1
)
1
(.
1
. . . . . .
n
). The inverse map is
¬
1
: . ÷ ((1 ÷[.[)
2
)
1
(2.. [.[
2
÷ 1). We also have the stereographic projec
tion ç
S
: S
n
· {÷e
n÷1
] ÷R
n
and the transition map is ç
S
ı ç
1
1
(,) = [,[
2
,.
From the stereographic projection we obtain S
n
as a speciﬁc model of the one
point compactiﬁcation R
n
L {o] by extending ¬
1
(o) = e
n÷1
. We also write
S
V
= V L {o] for the onepoint compactiﬁcation of a ﬁnitedimensional real
vector space V . Þ
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
n÷1
·{0] is an hequivalence with homotopy inverse
N : R
n÷1
· {0] ÷ S
n
, . ÷ [.[
1
.. A homotopy (rel S
n
) from i ı N to the
identity is the linear homotopy (.. t ) ÷t. ÷(1 ÷t )iN(.). Moreover N ı 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
n÷1
· {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
Stone–Weierstrass, 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 difﬁculties in topology.Þ
(2.3.4) Proposition. The map ¸: S
n1
×1 ÷D
n
, (.. t ) ÷(1÷t ). is a quotient
map. Given J : D
n
÷X, the composition J¸: S
n1
×1 ÷X is a null homotopy
of ¡ = J[S
n1
. Each null homotopy of a map ¡ : S
n1
÷ X arises from a
unique J.
Proof. Since a null homotopy H of ¡ : S
n1
÷ X sends S
n1
× 1 to a point, it
factors through the quotient map q : S
n1
× 1 ÷ S
n1
× 1,S
n1
× 1. Thus null
homotopies H correspond via
¨
H ÷
¨
Hq to maps
¨
H: S
n1
× 1,S
n1
× 1 ÷ X.
The map ¸ induces a homeomorphism ¨ ¸: S
n1
×1,S
n1
×1 ÷D
n
(use (1.4.3)).
Hence there exists a unique J such that J ¨ ¸ =
¨
H.
36 Chapter 2. The Fundamental Group
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
n÷n
,d1
n÷n
÷
1
n
d1
n
.1
n
,d1
n
which is the identity on representatives is a pointed homeomor
phism. If V and W are ﬁnitedimensional real vector spaces, we have a canonical
pointed homeomorphism S
V
.S
W
Š S
V ˚W
which is the identity away from the
base point. The homeomorphism 0. 1Œ ÷ R, s ÷
2x1
x(1x)
induces a homeomor
phism; : S(1) ÷S
(1)
which transports t ÷1÷t 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 nfold smash products.
(2.3.5) Example. A retraction r : D
n
× ÷ S
n1
× 1 L D
n
× 0 is r(.. t ) =
(2˛(.. t )
1
.. ˛(.. t )÷2÷t ) with ˛(.. t ) = max(2[.[. 2÷t ). (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 coﬁbration. Þ
........................................
,
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
n1
) × (1. 0) Š D
n
× (1. 0), see Figure 2.2.
Similarly for (1
n
. d1
n
) in place of (D
n
. S
n1
), since these two pairs are homeo
morphic. Þ
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
m1
) × (D
n
. S
n1
) Š (D
m¬n
. S
m¬n1
).
2. R
n¬1
· {.] ÷S
n
, : ÷(: ÷ .),[: ÷ .[ is an hequivalence.
3. Let D
n
¬
= {(.
0
. . . . . .
n
) ÷ S
n
[ .
n
_ 0]. Show that the quotient map S
n
÷ S
n
,D
n
¬
is an hequivalence.
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 hequivalence.
6. Let ¡. g: X ÷S
n
be maps such that always ¡(.) ,= ÷g(.). Then ¡ . g.
7. Let · Ç 1
n
be starshaped with respect to 0. Showthat S
n1
Ç R
n
··is a deformation
retract.
8. The projection ¸: TS
n
= {(.. ·) ÷ S
n
× R
n¬1
[ . J ·] ÷ S
n
, (.. ·) ÷ . is
called the tangent bundle of S
n
. Show that ¸ admits a ﬁbrewise 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
reviewfrompointset topology the compactopen 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 (Eckmann–Hilton duality).
We denote by Y
A
or J(X. Y ) the set of continuous maps X ÷Y . For 1 Ç X
and U Ç Y we set W(1. U) = {¡ ÷ Y
A
[ ¡(1) Ç U]. The compactopen
topology (COtopology) on Y
A
is the topology which has as a subbasis the sets
of the form W(1. U) for compact 1 Ç X and open U Ç Y . In the sequel the
set Y
A
always carries the COtopology. 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.
38 Chapter 2. The Fundamental Group
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) Ç 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 1ÇY be compact andU Ç7open. It sufﬁces toshowthat W(1. U) has
an open preimage under ¡
.
. Let ¡
.
(.) ÷ W(1. U) and hence ¡({.] ×1) Ç U.
Since 1 is compact, there exists by (1.4.1) a neighbourhood V of . in X such that
V × 1 Ç ¡
1
(U) and hence ¡
.
(V ) Ç W(1. U).
From (2.4.2) we obtain a set map ˛: 7
A×Y
÷ (7
Y
)
A
, ¡ ÷ ¡
.
. Let e
Y,7
be continuous. A continuous map ç : X ÷ 7
Y
induces a continuous map ç
·
=
e
Y,7
ı(ç×id
Y
): X×Y ÷7
Y
×Y ÷7. Hence we obtain a set map ˇ: (7
Y
)
A
÷
7
A×Y
, ç ÷ç
·
.
(2.4.3) Proposition. Let e
Y,7
be continuous. Then ˛ and ˇ are inverse bijections.
Thus ç : X × Y ÷ 7 is continuous if ç
·
: X × Y ÷ 7 is continuous, and
¡ : X × Y ÷7 is continuous if ¡
.
: X ÷7
Y
is continuous.
(2.4.4) Corollary. If h: X×Y ×1 ÷7 is a homotopy, then h
.
: X×1 ÷7
Y
is a
homotopy (see (2.4.2)). Hence ŒX×Y. 7 ÷ŒX. 7
Y
, Œ¡  ÷Œ¡
.
 is welldeﬁned.
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
ı h = ¡
t
for c = 0. 1. The equivalence with our
original deﬁnition follows from (2.4.3): Since 1 is locally compact, continuous
maps X × 1 ÷Y correspond bijectively to continuous maps X ÷Y
J
. Þ
(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 ı (¸ × id) is continuous, then h is continuous. The adjoint of
h ı (¸ × id) is h
.
ı ¸. 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
A×Y
÷(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 ı (˛ × id) is continuous. And
this map is continuous if ˛
2
= e
Y,A
ı (˛
1
× id) is continuous. One veriﬁes that
˛
2
= e
A×Y,7
. The evaluations which appear are continuous by (2.4.1).
The inverse ˛
1
is continuous if e
A×Y,7
ı (˛
1
× id) is continuous, and this
map equals e
Y,7
ı (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 COtopology 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 gı¸: X×Y ÷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 ç : X ÷ J
0
(Y. 7) we ob
tain ç
·
= ˇ
0
(ç) = e
0
A,Y
ı (ç . id): X . Y ÷ 7, and hence a set map
ˇ
0
: J
0
(X. J
0
(Y. 7)) ÷J
0
(X . Y. 7).
(2.4.9) Proposition. Let e
0
A,Y
be continuous. Then ˛
0
and ˇ
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
ŒX . Y. 7
0
÷ŒX. J
0
(Y. 7)
0
. Œ¡  ÷Œ˛
0
(¡ )
is well deﬁned. 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
, (ç. a) ÷(ç. k
o
) is continuous.
40 Chapter 2. The Fundamental Group
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
Ç U. The inclusion [(W(1. V
1
) ×V
2
) Ç 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 ﬁrst case we obtain, with a map [ from (2.4.12), a continuous map
H
7
ı [: X
7
× 1 ÷(X × 1)
7
÷Y
7
.
In the second case we use the composition
e ı (˛ × id) ı (7
1
× id): 7
Y
× 1 ÷7
A×J
× 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 hequivalences. If ¡ is a
pointedhequivalence, the inducedmaps J
0
(7. X) ÷J
0
(7. Y ) andJ
0
(Y. 7) ÷
J
0
(X. 7) are pointed hequivalences.
Problems
1. Verify that ¡
.
and 7
¸
are continuous.
2. An inclusion i : 7 Ç Y induces an embedding i
³
: 7
³
÷Y
³
.
3. The canonical map J
_
¸
X
¸
. Y
_
÷
¸
J(X
¸
. Y ) is always a homeomorphism.
4. The canonical map J(X.
¸
Y
¸
) ÷
¸
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 preimage of a compact set
is compact. Then 7
)
: 7
Y
÷7
³
is an embedding.
6. We have a canonical bijective map J
0
_ _
¸÷J
X
¸
. Y
_
÷
¸÷J
J
0
(X
¸
. Y ), since
_
¸
X
¸
is the sum in TOP
0
. If J is ﬁnite, it is a homeomorphism.
7. Let X. Y. U, and V be spaces. The Cartesian product of maps gives a map
¬ : U
³
× V
Y
÷(U × V )
³×Y
. (¡. g) ÷¡ × g.
Let X and Y be Hausdorff spaces. Then the map ¬ is continuous.
8. By deﬁnition 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)
³
÷
Y
³
×7
³
. 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
³
÷7
³
, (g. ¡ ) ÷
g ı ¡ 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
ŒX,·. Y 
0
÷ Œ(X. ·). (Y. +). If ¸ has compact preimages of compact sets, then ; is a
homeomorphism.
11. Consider diagrams where the righthand one is obtained by multiplying the lefthand
one with X:
·
T
· × X
T × X
C
D, C × X
D × X.
If the lefthand diagram is a pushout in TOP and X locally compact, then the righthand
diagram is a pushout in TOP. In TOP
0
the smash product with a locally compact space yields
again a pushout.
12. The COtopology 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 COtopology on Y
³
is
induced by the supremummetric.
2.5 The Fundamental Groupoid
Apath in the plane can be quite ungeometric: nowhere differentiable, inﬁnite length,
surjective onto 1 ×1 (a socalled 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
42 Chapter 2. The Fundamental Group
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 deﬁned ).
(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 deﬁned 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 ˛ deﬁned 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 deﬁned 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 deﬁned 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 deﬁned by Œ· ı Œu = Œu + ·. 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 deﬁne the multiplication in this group by Œu ı Œ· = Œ· + u. As long as
we are just interested in this group (and not in the categorical aspect), we use the
opposite multiplication Œu Œ· = Œu+· and denote this group by ¬
1
(X. .). This is
the traditional fundamental group of the pointed space (X. .) (Poincaré 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: Œ0. a : 1 ÷ X
and ·: Œ0. 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
deﬁnition Œ0. c and the resulting composed paths are homotopic rel {0. c]. Þ
(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(2¬i 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. .) = Œ(1. d1). (X. .) Š Œ1,d1. X
0
Š ŒS
1
. X
0
.
induced by Œn Œ ¨ n = Œuq Œu. Þ
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
…(.. .) ÷…(,. ,). ˇ ÷˛ˇ˛
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 1connected 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. ¡(.)). Œu ÷Œ¡ u
and, more generally, a functor
…(¡ ): …(X) ÷…(Y ). . ÷¡(.). Œu ÷Œ¡ 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 ŒH
x
 in …(Y ) from ¡(.) to
g(.). The ŒH
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.
44 Chapter 2. The Fundamental Group
We express the commutativity of (2.5.4) in a different way. It says that
t
1
ı g
+
= ¡
+
: …X(.. ,) ÷…Y(¡ .. ¡,).
where t
1
: …(Y )(g.. g,) ÷…(Y )(¡ .. ¡,) is the bijection a ÷ŒH
,

1
aŒH
x
.
The rule …(1 + 1) = …(1)…(1) is obvious. Hence if ¡ is an hequivalence
with hinverse 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 X×d1 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. ¡(.)). Œn ÷Œ¡ n
is an isomorphism for each . ÷ X. A contractible space is simply connected.
Proof. Let g: Y ÷X be hinverse to ¡ . Consider
…X(.. ,)
(
·
÷÷…Y(¡ .. ¡,)
¿
·
÷÷…X(g¡ .. g¡,)
(
·
÷÷…Y(¡g¡ .. ¡g¡,).
Choose H: g¡ . id(X). Then g
+
¡
+
= (g¡ )
+
= t
1
ı (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 hequivalences ¡ 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 welldeﬁned 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 ﬁrst 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×·)ı with the diagonal ı. 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 ı, 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 ﬁrst 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
Ç 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
¡
) ÷ ƒ 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) ÷ƒsuch that
h
1
= z…(¸
1
) and h
0
= z…(¸
0
). We begin with a couple of remarks.
A path n: Œa. b ÷U represents a morphism Œn in …(U) from n(a) to n(b)
if we compose it with an increasing homeomorphism ˛: Œ0. 1 ÷Œa. b.
If a = t
0
< t
1
< < t
n
= b, then n represents the composition of the
morphisms Œn[Œt
i
. t
i÷1
.
Suppose that n: 1 ÷ X is a path. Then there exists a decomposition 0 =
t
0
< t
1
< < t
n÷1
= 1 such that n(Œt
i
. t
i÷1
) is contained in a set X
ı
¡
. Choose
; : {0. . . . . m] ÷ {0. 1] such that n(Œt
i
. t
i÷1
) Ç X
ı
;(i)
. Consider n[Œt
i
. t
i÷1
 as
path n
i
in X
;(i)
. Then
Œn = …(¸
;(n)
)Œn
n
 ı ı …(¸
;(0)
)Œn
0
.
46 Chapter 2. The Fundamental Group
If z exists, then
(i) zŒn = h
;(n)
Œn
n
 ı ı h
;(0)
Œn
0
.
i.e., z is uniquely determined.
In order to show the existence of z, we have to deﬁne zŒn by (i). We have
to verify that this is welldeﬁned. 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 reﬁne 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 subsquare Œi,n. (i ÷1),n × Œ¸,n. (¸ ÷1),n into one of the
sets X
ı
¡
(see (2.6.4)). We consider edgepaths in the subdivided square 1 ×1 which
differ by a subsquare, as indicated in the following ﬁgure (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
ı
¡
.
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 (Seifert–van 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 ç
¡
: ¬
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 deﬁnition of a pushout. Þ
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 cneigh
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) ı (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 deﬁne 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(2¬i t ), the identity a ÷(a. 0), and the composition (b. t )ı(a. s) =
(a. s ÷t ).
The assignments a ÷ exp(2¬i a) and (a. b) ÷ (exp(2¬i a). b ÷ a) yield a
continuous functor …(R) ÷ G. We will show that G (forgetting the topology) is
…(S
1
).
48 Chapter 2. The Fundamental Group
We have the open covering of S
1
Ç 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. 1Œ ÷ X
0
and ¡
1
:  ÷ 1,2. 1,2Œ ÷ X
1
given
by t ÷ exp(2¬i t ). We deﬁne 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(2¬i t s) from a to a exp(2¬i t ).
(The idea behind the deﬁnition 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(2¬i(t
1
÷ ÷t
i
)). Then
(a. t ) = (a
n1
. t
n
) ı ı (a
1
. t
2
) ı (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
i1
exp(2¬i t
i
s) ÷ X
k(i)
for s ÷ 1.
Then (a
i1
. t
i
) = ;
k(i)
(a
i1
. 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(2¬i nt ).
(2.7.2) Theorem. Let s
n
: 1 ÷ S
1
be the loop t ÷ exp(2¬i nt ). The assignment
ı : Z ÷¬
1
(S
1
. 1), n ÷Œs
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 deﬁned using this
multiplication.
2.7. The Fundamental Group of the Circle 49
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 hTOP. (We do not assume that mis associative or commuta
tive.) We deﬁne a composition law on the pointed homotopy set ŒY. X
0
, called the
mproduct, by Œ¡ . Œg ÷ Œ¡ 
n
Œg = Œ¡ g; here ¡ g: , ÷ m(¡(,). g(,))
is the ordinary pointwise multiplication. The constant map represents a twosided
unit for the mproduct. In a similar manner we deﬁne by pointwise multiplication
of loops the mproduct on ¬
1
(X. e). The set ¬
1
(X. e) Š ŒS
1
. X
0
now carries two
composition laws: the mproduct and the +product of the fundamental group.
(2.7.3) Proposition. Let (X. m) be a monoid in hTOP. Then the +product and the
mproduct 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 deﬁnition of the maps.
(2.7.4) Lemma. The map ·: ŒS
1
. S
1

0
÷ŒS
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 ŒX. S
1
.
(2.7.5) Theorem. From (2.7.2), (2.7.4) and (2.5.3) we obtain an isomorphism J,
J : ŒS
1
. S
1
 Š ŒS
1
. S
1

0
Š ¬
1
(S
1
. 1) Š Z.
We call the integer J(¡ ) = J(Œ¡ ) 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
:
n1
÷ ÷a
n
has a root (n _ 1).
50 Chapter 2. The Fundamental Group
Proof. Suppose g(:) ,= 0 for [:[ = 1. Then ¡ : S
1
÷ S
1
, : ÷ g(:),[g(:)[
is deﬁned. Suppose g is nonzero for [:[ _ 1. Then h(:. t ) = ¡(t :) is a null
homotopy of ¡ . For t > 0 we have
k(:. t ) = :
n
÷t (a
1
:
n1
÷a
2
t :
n2
÷ ÷a
n
t
n1
) = t
n
g(:,t ).
If g is nonzero 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(2¬i t ). A lifting of n: Œa. b ÷ S
1
along ¸ is a
map W : Œa. 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 ı(J(¡ )) = Œg¸
0
, by the deﬁnition of J in (2.7.5).
There exists a ÷ R such that ¡(1) = exp(2¬i a) and J(0) = a. Then ˆ = J ÷ a
is a lifting of g¸
0
with initial condition 0. Hence J(1) ÷ J(0) = ˆ(1) ÷ ˆ(0) =
ˆ(1) = n ÷ Z. The homotopy (.. t ) ÷ (1 ÷ t )ˆ(.) ÷t.ˆ(1) is a homotopy of
paths. Hence the loop g¸
0
is homotopic to s
n
. This shows ı(n) = Œg¸
0
.
The next proposition will be proved in the chapter on covering spaces. It ex
presses the fact that ¸: R ÷S
1
is ﬁbration.
(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: Œ0. 1 ÷S
1
be a path with n(0) = : = exp(2¬i a). Let
W : Œ0. 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(2¬i(b ÷ a)t ). Þ
2.7.10 The winding number. Let . ÷ C = R
2
. The map
r
x
: C\ {.] ÷S
1
. : ÷(: ÷ .),[: ÷ .[
is an hequivalence and therefore ŒS
1
. C\ {.] ÷ ŒS
1
. S
1
, Œ¡  ÷ Œr
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
2.7. The Fundamental Group of the Circle 51
¡(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 ¡ ı ¸
0
= u. The winding number of ¡ is then also
called the winding number of u, and we denote it by W(u. .). Þ
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(¡ ı 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(2¬i,n) be an nth root of unity. Suppose h: S
1
÷ S
1
satisﬁes 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
satisﬁed with some integer J(¡ ), then there exists a map ¡ of degree J(¡ ) which satisﬁes
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(2¬i:). Then W(u. 0) = U(1) ÷ U(0).
4. Let ; : Œ0. 1 ÷ C
·
be a continuously differentiable path with initial point 1. Then
I : Œ0. 1 ÷ C, t ÷
1
2¬i
_
)Œ0,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
2¬i
_
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 ·: ŒS
1
. X
0
÷ ŒS
1
. X be the map which forgets about the base point (pointed
homotopies versus free homotopies). Conjugate elements in the group ŒS
1
. X
0
have the
same image under ·. Hence · induces a welldeﬁned map ¨ ·: ŒS
1
. X
0
,(~) ÷ŒS
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.
52 Chapter 2. The Fundamental Group
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 difﬁcult 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. Þ
2.8.2 Removing a point. The inclusion D
n
· 0 Ç 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
n1
.
Let M be a manifold of dimension n _ 3 and U Ç 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 · {.] Ç M induces an isomorphism of
fundamental groups.
Often we view the space S
n
as the onepoint compactiﬁcation 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 Ç S
n
· 1 induces an isomorphism of fundamental groups. Þ
2.8.3 Complements of spheres. Let S
n
0
= S
n÷n÷1
¨ (R
n÷1
× 0) and S
n
1
=
S
n÷n÷1
¨ (0 × R
n÷1
). Then X = S
n÷n÷1
· S
n
1
is homeomorphic to S
n
× 1
n
.
A homeomorphism S
n
× 1
n
÷ X is (.. ,) ÷
_
_
1 ÷ [,[
2
.. ,
_
. The space
Y = S
n÷n÷1
· (S
n
0
L S
n
1
) is homeomorphic to S
n
× S
n
×0. 1Œ via (.. ,. t ) ÷
(
_
1 ÷ t ..
_
t ,). Therefore the complement X is hequivalent to S
n
and the com
plement Y is hequivalent 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 2disk with boundary W in the complement of 7,
because W represents a nonzero 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
Ç S
3
. The image
yields W L 7. The complement of W L 7 in R
3
is therefore isomorphic to the
fundamental group Z×Zof the torus S
1
×S
1
. The reader should draw generators
of ¬
1
(R
3
· W L 7). Þ
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 · ı 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.
Þ
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
∗k÷J
G
k
together with a family
54 Chapter 2. The Fundamental Group
of homomorphisms 
}
: G
}
÷
∗k÷J
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
}
Ç 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
∗x÷S
Z. Þ
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 hequivalences. 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
) Š Z+ Z.
By induction one shows that ¬
1
_ _
k
1
S
1
_
is the free group of rank k. Þ
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 coﬁbrations.) 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 halfspaces
{(.. ,) [ . < 1,3. . ,= ÷1] and {(.. ,) [ ÷1,3 < .. . ,= 1] . Þ
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 ﬁrst “thicken” the subspaces up to hequivalence. 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
}
Ç X
}
which is
pointed contractible to the base point. The inclusions of the summands induce
homomorphisms i
}
: ¬
1
(X
}
. .
}
) ÷¬
1
_ _
k÷J
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 ﬁnite J we use induction on [J[. Note
that
_
U
}
Ç
_
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
_ _
e÷T
X
e
_
_ _
}÷J~T
U
}
_
for a ﬁnite subset 1 Ç 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
∗e÷T
¬
1
(X
e
) for a
ﬁnite 1. Suppose . is contained in the kernel of ˛
J
. Then a loop n: 1 ÷
_
e÷T
X
e
representing ˛
T
. is null homotopic in
_
}÷J
X
}
and, again by compactness, null
homotopic in some larger ﬁnite wedge. The result for a ﬁnite 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
1
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. Þ
2.8.10 Attaching of a 2cell. We start with a pushout diagram of spaces
S
1
ç
}
T
J
D
2
ˆ
X.
56 Chapter 2. The Fundamental Group
Then X is said to be obtained fromT by attaching a 2cell via the attaching map ç.
(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),(ç) Š ¬
1
(X)
where (ç) denotes the normal subgroup generated by Œç ÷ ŒS
1
. T
0
= ¬
1
(T). Þ
2.8.11 Attaching of a cone. Given a map ç : · ÷ 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
·
ç
}
T
J
C·
ˆ
X.
Since C· is contractible, J
+
induces an isomorphism ¬
1
(T),N Š ¬
1
(X), where
N is the normal subgroup generated by the image of ç
+
. Þ
2.8.12 Realizationof groups. We demonstrate that arbitrarygroups canbe realized
as fundamental groups. Let
¡ : · =
_
k÷1
S
1
÷
_
I÷1
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 Ç
∗I÷1
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 deﬁne X by a pushout diagram
·
(
í
T
C·
X.
Then 2.8.11 shows ¬
1
(X) Š G,N. Each group can be presented in the formG,N.
Note that X is a 2dimensional cell complex. Þ
2.8.13 Surfaces. The classiﬁcation theory of compact connected surfaces presents
a surface as a quotient space of a regular 2ngon, see e.g., [44, p. 75], [167], [123].
The edges are identiﬁed in pairs by a homeomorphism. The surface J is obtained
as a pushout of the type
S
1
ç
_
n
1
S
1
D
2
ˆ
J.
2.8. Examples 57
The attaching map ç is given in terms of the standard generators of ¬
1
_ _
n
1
S
1
_
by
the socalled surfaceword.
In order to save space we refer to [44, p. 83–87] 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
¿
b
¿
a
1
¿
b
1
¿
).
(2) The fundamental group of a closed connected nonorientable 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 nonempty
boundary is a free group. The number of generators is the ﬁnite number 1 ÷¡(J
¿
)
where ¡(J
¿
) is the socalled Euler characteristic.
(4) A simply connected surface is homeomorphic to R
2
or S
2
.
There are many different deﬁnitions 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 nonorientable surface is the maximal number g of disjointly
embedded Möbius bands such that their complement is connected. The sphere has
genus zero by the Jordan separation theorem. Þ
Problems
1. Let S
1
Ç R
2
× 0 Ç 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 hequivalent 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.
58 Chapter 2. The Fundamental Group
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
ı
¸
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 deﬁned as the quotient of S
2
by the
relation . ~ ÷.. Let Œ.
0
. .
1
. .
2
 denote the equivalence class of . = (.
0
. .
1
. .
2
). We can
also obtain 1
2
from S
1
by attaching a 2cell
S
1
µ
¸
1
1
J
D
2
ˆ
1
2
.
Here 1
1
= {Œ.
0
. .
1
. 0] Ç 1
2
andç(.
0
. .
1
) = Œ.
0
. .
1
. 0. The space 1
1
is homeomorphic
to S
1
via Œ.
0
. .
1
. 0 ÷:
2
. : = .
0
÷i.
1
; and ç corresponds to the standard map of degree
2. The map ˆ is . = (.
0
. .
1
) ÷ Œ.
0
. .
1
.
_
1 ÷ [.[
2
. As an application of 2.8.10 we
obtain ¬
1
(1
2
) Š 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 Möbius
band M. Thus 1
2
is obtainable from a Möbius band M and a 2disk D by identiﬁcation 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 selfintersections (technically,
the image of a smooth immersion.) The projective plane is a nonorientable surface.
6. Klein bottle. The Klein bottle 1 can be obtained from two Möbius bands M by an
identiﬁcation 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 deﬁne 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 deﬁne 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 = Œ0. 1
for the unit interval. This means: A map ˆ: (X × 1) × J ÷ Y is a homotopy
relative to X×d1, if ˆ(.. 0. t ) is independent of t and ˆ(.. 1. t ) is also independent
of t . Therefore ˆ
t
: X ×1 ÷Y. (.. s) ÷ˆ(.. 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 Jvariable. Hence we obtain
an equivalence relation on the set of homotopies from ¡ to g. We now deﬁne: 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 deﬁned by the
product of homotopies
1: ¡ . g. 1: g . h. Œ1 ~Œ1 = Œ1 + 1 : ¡ ÷h.
This is easily seen to be welldeﬁned (use ˆ
t
+
J
‰
t
). The identity of ¡ in …(X. Y )
is represented by the constant homotopy k
(
: ¡ . ¡ .
The veriﬁcation 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 ı (id ×˛) are homotopic relative to X × d1.
Proof. ˆ(.. 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) ı (id × ˛).
with ˛ deﬁned 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 deﬁned by 1.
Hence each morphism is an isomorphism. Proof: The assignments (.. s. t ) ÷
1(.. 2s(1÷t )) for 0 _ s _
1
2
and (.. s. t ) ÷1(.. 2(1÷s)(1÷t )) for
1
2
_ s _ 1
yield a homotopy relative to X ×d1 from 1 + 1

to the constant homotopy.
60 Chapter 2. The Fundamental Group
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 deﬁned
by homotopies ˆ rel X × d1 such that each ˆ
t
is a pointed homotopy.
The remainder of this section can be skipped on a ﬁrst reading. We study the
dependence of the groupoids …(X. Y ) on X and Y . The formal structure of this
dependence canbe codiﬁedinthe notionof a 2category. Suppose given˛: U ÷X
and ˇ: Y ÷V . Composition with ˛ and ˇ yield a functor
ˇ
#
= …
#
(ˇ): …(X. Y ) ÷…(X. V ).
which sends ¡ to ˇ¡ and Œ1 to Œˇ1 and a functor
˛
#
= …
#
(˛): …(X. Y ) ÷…(U. Y ).
which sends ¡ to ¡ ˛ and Œ1 to Œ1(˛ × id). They satisfy (ˇ
1
ˇ
2
)
#
= ˇ
1
#
ˇ
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 Œ1 ˘ 1 = ·
#
Œ1 ~ ¡
#
Œ1 = g
#
Œ1 ~ u
#
Œ1. Here 1 ˘ 1: u¡ .
·g: X × 1 ÷7. (.. t ) ÷1(1(.. t ). t ).
Proof. We use the bihomotopy 1ı (1 ×id): X ×1 ×1 ÷7. Restriction to the
diagonal of 1 × 1 deﬁnes 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
1˘1
g
#
Œ1 ~u
#
Œ1 is represented by u1 + 1(g ×id). If we compose the bihomotopy
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 bihomotopy with id(.) × ı, where ı(t ) = (0. 2t ) for t _
1
2
and
ı(t ) = (2t ÷ 1. 1) for t _
1
2
. The maps ; and ı 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 ¡
#
Œ1. The homotopy 1 induces a natural transformation
1
#
: ¡
#
÷g
#
: …(Y. 7) ÷…(X. 7).
The value of 1
#
at u is u
#
Œ1.
(2.9.5) Corollary. If u: Y ÷7 is an hequivalence, then u
#
is an equivalence of
categories. Similarly in the contravariant case.
The data and assertions that we have obtained so far deﬁne on TOP the structure
of a 2category. In this context, the ordinary morphisms ¡ : X ÷ Y are called
1morphisms and the morphisms Œ1 : ¡ . g are called 2morphisms. The
composition ~ of 2morphisms is called vertical composition. We also have a
horizontal composition of 2morphisms deﬁned as Œ1 ˘ Œ1 = Œ1˘ 1. Because
of (2.9.3) we need not deﬁne ˘ via the diagonal homotopy; we can use instead
(2.9.3) as a deﬁnition Œ1 ˘ Œ1 = ·
#
Œ1 ~¡
#
Œ1 = g
#
Œ1 ~u
#
Œ1.
(2.9.6) Note. Fromthis deﬁnition one veriﬁes the commutation rule of a 2category
(ı ~;) ˘ (ˇ ~˛) = (ı ˘ ˇ) ~(; ˘ ˛).
The following ﬁgure organizes the data (horizontal – vertical).
·
(
¿
[ ˛
h
¸
[ ˇ
T
u
¡
[ ;
u
¸
[ ı
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 ﬁbres. Objects of this type
can be classiﬁed 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 classiﬁcation 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
= Œ…(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 codiﬁed in moving along the ﬁbres 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 viewpoint of the fundamental group.
Another aspect belongs to bundle theory. In the chapter devoted to bundles we show
for instance that isomorphismclasses of nfold coverings over a paracompact space
T correspond to homotopy classes T ÷ BS(n) into a socalled classifying space
BS(n).
3.1 Locally Trivial Maps. Covering Spaces
Let ¸: 1 ÷ T be continuous and U Ç T open. We assume that ¸ is surjec
tive to avoid empty ﬁbres. A trivialization of ¸ over U is a homeomorphism
ç : ¸
1
(U) ÷U ×J over U, i.e., a homeomorphism which satisﬁes pr
1
ıç = ¸.
This condition determines the space J up to homeomorphism, since ç 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 ﬁbre 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 ﬁxed space J is open and closed in T. Therefore it sufﬁces for most purposes
to ﬁx the homeomorphism type of the ﬁbres. If the ﬁbres are homeomorphic to J,
we call J the typical ﬁbre. 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 ﬁbres. If J is discrete (= all subsets are open and closed), then U × J is
homeomorphic to the topological suml
x÷T
U ×{.]. The summands U ×{.] are
canonically homeomorphic to U. If ç : ¸
1
(U) ÷U × J is a trivialization, then
¸ yields via restriction a homeomorphism of ç
1
(U × {.]) with U. A covering
is therefore a local homeomorphism. The summands ç
1
(U × {.]) = U
x
are the
sheets of the covering over U; the preimage ¸
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 nfold covering. The trivial covering
with typical ﬁbre 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(2¬i t ) is a cov
ering with typical ﬁbre Z. For each t ÷ Rand ¸(t ) = : we have a homeomorphism
¸
1
(S
1
· :) =
n÷Z
t ÷n. t ÷n ÷1Œ Š t. t ÷1Œ × Z.
and ¸ maps each summand homeomorphically. Þ
(3.1.2) Proposition. Let ¸: 1 ÷T be a covering. Then the diagonal D of 1×1
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 Ç 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 × Œ0. 1 be locally trivial with typical ﬁbre J.
Then T has an open cover Usuch that q is trivial over each set U ×Œ0. 1, U ÷ U.
Proof. If q is trivial over U × Œa. b and over U × Œb. c, then q is trivial over
U × Œa. c. Two trivializations over U × {b] differ by an automorphism, and this
automorphism can be extended over U × Œb. c. Use this extended automorphism
to change the trivialization over U × Œb. 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 × Œt
i
. t
i÷1
.
For the classiﬁcation 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. Þ
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 Gprincipal 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 ﬁbre is transitive.
3.1. Locally Trivial Maps. Covering Spaces 65
(3.1.7) Example. A left Gprincipal 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 Gprincipal covering. Þ
A covering ¸: 1 ÷T has an automorphism group Aut(¸). An automorphism
is a homeomorphism ˛: 1 ÷ 1 such that ¸ ı ˛ = ¸. Maps of this type are also
called deck transformations of ¸. If ¸ is a left Gprincipal 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 ﬁbre
¸
1
(¸(.)). Since G acts transitively on each ﬁbre, 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 ﬁnd 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 Ç 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 Gprincipal covering ¸: 1 ÷ T gives rise to associated coverings.
Let J be a set with left Gaction. 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
ﬁbre J.
A Gmap [: J
1
÷J
2
induces a morphism of coverings
id ×
G
[: 1 ×
G
J
1
÷1 ×
G
J
2
. (.. ¡ ) ÷(.. [(¡ )).
66 Chapter 3. Covering Spaces
We thus have obtained a functor “associated coverings”
·(¸): GSET ÷COV
B
from the category GSET of left Gsets and Gequivariant maps. We call a Gprin
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 ﬁbration 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: X×1 ÷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 ﬁbration if it has the HLP for all spaces.
(3.2.1) Example. A projection ¸: T × J ÷ T is a ﬁbration. 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. Þ
(3.2.2) Theorem. A covering ¸: 1 ÷T is a ﬁbration.
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 sufﬁces to ﬁnd 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 welldeﬁned 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
× Œi,n. (i ÷1),n into a set U over which ¸ is trivial.
Since ¸: ¸
1
(U) ÷ U is, by (3.2.1), a ﬁbration, h[V
x
× Œi,n. (i ÷ 1),n has a
lifting for each initial condition. Therefore we ﬁnd by induction over i a lifting of
h[V
x
× Œ0. i,n with initial condition a[V
x
.
(3.2.3) Lemma. Let U be an open covering of T × Œ0. 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) × Œi,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
compactopen 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 modiﬁed 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
i í
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 Œ· ÷ …(T). Let . ÷ J
b
. Choose a lifting
V : 1 ÷1 of · with V(0) = .. We set ·
#
Œ. = ŒV(1). We have to show that this
assignment is welldeﬁned. 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, deﬁned 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
welldeﬁned and depends only on the morphism Œ· in the fundamental groupoid.
The rule n
#
·
#
= (· + n)
#
is easily veriﬁed from the deﬁnitions. Thus we have
shown:
(3.2.5) Proposition. The assignments b ÷ ¬
0
(J
b
) and Œ· ÷ ·
#
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
). (.. Œ·) ÷·
#
(.) = . Œ·.
a right action of the fundamental group on the set ¬
0
(J
b
). We write ¬
0
(J. .) if Œ.
is chosen as base point of the set ¬
0
(J). We use the action to deﬁne
d
x
: ¬
1
(T. b) ÷¬
0
(J
b
. .). Œ· ÷. Œ·.
The map d
x
is ¬
1
(T. b)equivariant, i.e., d
x
Œ· + n = (d
x
Œ·) Œn.
68 Chapter 3. Covering Spaces
Recall that a sequence ·
˛
÷T
ˇ
÷C of pointed maps is exact at T if the image
of ˛ equals the kernel ˇ
1
(+) of ˇ. 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
Ç 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 veriﬁed from the deﬁnitions that the composition of two maps is
the constant map. We consider the remaining four cases: kernel Ç image.
Let Œu ÷ ¬
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
Œ· = Œ.. This means: There exists a lifting V of · from. to V(1) ÷ Œ..
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 Œ·, since ¸U is constant.
Let ¬
0
(i )Œ, = Œ.. There exists a path n: 1 ÷1 from. to ,. The projection
· = ¸n is a loop and d
x
Œ· = Œ,, by deﬁnition of d
x
.
Let ¬
0
(¸)Œ, = Œb. 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 )(ŒV(1) = Œ,.
There is more algebraic structure in the sequence.
(3.2.7) Proposition. The preimages of elements under d
x
are the left cosets of
¬
1
(T. b) with respect to ¸
+
¬
1
(1. .). The preimages of ¬
0
(i ) are the orbits of the
¬
1
(T. b)action on ¬
0
(J
b
. .).
Proof. Let d
x
Œu = d
x
Œ·. 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 Œu and Œ· 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 )Œa = ¬
0
(i )Œb. Then there exists a path n: 1 ÷1 from a to b.
Set · = ¸n. Then Œa Œ· = Œb. Conversely, elements in the same orbit have
equal image under ¬
0
(i ).
We can apply (3.2.6) to a covering ¸: 1 ÷T. The ﬁbres 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
Ç 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 ﬁbre, 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 Gprincipal covering. Each ﬁbre J
b
carries a free
right transitive Gaction. From the construction of the transport functor it is im
mediate that the ﬁbre transport T
]
Œn : J
b
÷ J
c
is Gequivariant. The left action
(a. .) ÷ a . = a
#
(.) of ¬
b
= …(T)(b. b) on J
b
commutes with the right
Gaction; 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
ı
x
(a) = ;
x
(a)
1
. Then ı
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
ı
x
= d
x
. Then (3.2.8) yields:
(3.2.10) Proposition. Let ¸: 1 ÷ T be a right Gprincipal covering with path
connected total space. Then the sequence of groups and homomorphisms
1 ÷¬
1
(1. .)
]
·
÷÷¬
1
(T. ¸(.))
ı
\
÷÷G ÷1
is exact ( for each . ÷ 1). The image of ¸
+
is a normal subgroup. The space
1 is simply connected if and only if ı
x
is an isomorphism. Thus, if 1 is simply
connected, then G is isomorphic to the fundamental group of T.
70 Chapter 3. Covering Spaces
If we apply this to the exponential covering R ÷ S
1
, a Zprincipal covering,
we again obtain ¬
1
(S
1
) Š 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
]
Œn : 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 Classiﬁcation of Coverings
Let TRA
B
= Œ…(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 ﬁbres 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 Ç T is transportsimple if two paths in U between the same points are
homotopic within T. A space T is semilocally simply connected if it has an open
covering by transportsimple sets. We have just seen that this condition is implied
by the existence of a simply connected covering. We call T transportlocal, if T
is path connected, locally path connected and semilocally simply connected.
(3.3.2) Theorem(Classiﬁcation I). Let T be path connected, locally path connected
and semilocally simply connected. Then T is a transport space, i.e., T is an
equivalence of categories.
3.3. Classiﬁcation of Coverings 71
Proof. We begin by constructing a functor
X: TRA
B
÷COV
B
in the opposite direction. Let ˆ: …(T) ÷ SET be a functor. We construct an
associated covering ¸ = ¸(ˆ): X(ˆ) ÷ T. As a set, X(ˆ) =
b÷B
ˆ(b), and
¸(ˆ) sends ˆ(b) to b. Let U be the set of open, path connected and transport
simple subsets of T. We deﬁne bundle charts over sets U ÷ U. For b ÷ U we
deﬁne
ç
U,b
: U × ˆ(b) ÷¸
1
(U). (u. :) ÷ˆ(n):
with some path n in U fromb to u. By our assumption on U, the map ç
U,b
is well
deﬁned, i.e., the choice of n does not matter. By construction, ç
U,b
is bijective.
We claim: There exists a unique topology on X(ˆ) such that the ç
U,b
of this type
are homeomorphisms onto open subsets. By general principles of gluing, we have
to verify that the transition maps
ç
1
V,c
ı ç
U,b
: (U ¨ V ) × ˆ(b) ÷(U ¨ V ) × ˆ(c)
are homeomorphisms. Let . ÷ U ¨ V and let W Ç 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
ç
1
V,c
ı ç
U,b
(,. :) = ç
1
V,c
ı ç
U,b
(.. :).
because in order to deﬁne ç
U,b
(,. :) we can take the product of u
x
with a path n
,
in W from . to ,, and similarly for ç
V,c
, so that the contribution of the piece n
,
cancels. This shows that the second component of ç
1
V,c
ı ç
U,b
is on W × ˆ(b)
independent of . ÷ W. The continuity of the transition map is a consequence.
If ˛: ˆ
1
÷ˆ
2
is a natural transformation, then the morphism
X(˛): X(ˆ
1
) ÷X(ˆ
2
). . ÷ ˆ
1
(b) ÷˛(.) ÷ ˆ
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 ˆ
U,b
for X(ˆ
1
) and X(ˆ
2
) transform X(˛) into
id ×˛(b): U × ˆ
1
(b) ÷U × ˆ
2
(b).
This ﬁnishes 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(¸)) Š
1(¸) over T for each covering ¸: 1(¸) ÷ T, namely settheoretically the iden
tity. We have to show that j is continuous. Let · ÷ J
b
Ç X(T(¸)). Let W be a
72 Chapter 3. Covering Spaces
neighbourhood of · in 1(¸). Then there exists an open neighbourhood V Ç W of
· in 1(¸) such that V is path connected and U = ¸(V ) ÷ U. Let b = ¸(·). Then
we have the bundle charts ç
U,b
, and W = ç
U,b
(U ×·) is a neighbourhood of · in
X(T(¸)). From the construction of ç
U,¡
we see that j(W) Ç V . This shows the
continuity at ·. It is easily veriﬁed that the homeomorphisms j constitute a natural
transformation. Conversely, we have to verify that the transport functor of ¸(ˆ)
is ˆ. Using the bundle charts ç
U,b
this is ﬁrst veriﬁed 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 Homfunctor …(b. ÷) of the category
…(T). Let ¸
b
: 1
b
÷T denote the associated covering. We still assume that T is
transportlocal. The automorphismgroup ¬
b
= …(b. b) of b in …(T), the opposite
fundamental group ¬
1
(T. b), acts on 1
b
ﬁbrewise fromthe right by composition of
morphisms. The action is free and transitive on each ﬁbre. Via our bundle charts it
is easily veriﬁed 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 ﬁbres by composition of morphisms is the structure of a right principal covering
on ¸
b
.
Problems
1. Let S be the pseudocircle. The space S is simply connected. But S has nontrivial
connected principal coverings. They can be obtained by a pullback along suitable maps
S ÷ S
1
from Z,nprincipal or Zprincipal 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
!
ı COV(¡ ) = TRA(¡ ) ı 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
: Œ…. 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
Homfunctor …(b. ‹) is a functor into the right ¬
b
sets, namely ¬
b
acts on …(b. .)
by composition of morphisms. These data yield the functor ˆ(·) = …(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 ˆ(¡ ): ˆ(·) ÷ˆ(T). This
ﬁnishes the deﬁnition 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

¹
: …(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
ˇ
T
(.): …(b. .) × J(b) ÷J(.). (¡. :) ÷J(¡ ):
form a natural transformation, i.e., a morphism ˇ
T
: j
b
c
b
(J) ÷ J in Œ…. SET.
Since … is a connected groupoid, the ˇ
T
(.) are bijective, and therefore constitute
an isomorphism in the functor category. The ˇ
T
are a natural equivalence
ˇ: j
b
c
b
. Id.
In our previous notation TRA
B
= Œ…(T). SET. From (3.3.2) and (3.4.1)
we obtain for each transportlocal 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
·) =
x÷B
…(T)(b. .) ×
t
· ÷T.
74 Chapter 3. Covering Spaces
It is the covering 1
b
×
t
· ÷T associated to the ¬
b
principal covering (3.3.3).
Let ¸: 1 ÷ T be a right Gprincipal covering with path connected T. Then
we have the functors
G SET
¹(])
COV
B
T
TRA
B
t
b
:
¬
b
 SET.
The composition associates to a Gset 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
ç
T
x
: J ÷ J
b
×
G
J, : ÷ Œ.. : for a ﬁxed . ÷ J
b
as well as the isomorphism
;
x
: ¬
b
÷G, see (3.2.10). The relation ;
x
(a) : = a Œ.. : holds. So if we view
Gsets 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 Gprincipal 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 Gprincipal covering ¸: 1 ÷ T with simply con
nected total space 1.
Proof. (1) = (2). Since c
b
ı T is an equivalence of categories, each object of
¬
b
 SET is isomorphic to an object in the image of c
b
ı 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
ı 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 ﬁbre is transitive. If we rewrite this as
a right action of the opposite group G, we obtain a right Gprincipal 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 Gprincipal covering. A left Gset · 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 Gset and isomorphic to a homogeneous set G,H
for some subgroup H of G. The homeomorphism 1 ×
G
G,H Š 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 Hprincipal covering. Also the induced
map ¸
1
: 1,H ÷T is a covering.
The category of homogeneous Gsets and Gmaps 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 Gsets is an equivalence.
Let ¸: 1 ÷ T be a universal right Gprincipal 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 ﬁx : ÷ ¸
1
(b) Ç 1 and obtain an isomorphism ı
z
: G ÷ ¬
1
(T. b). It
sends g ÷ G to the loop Œ¸ ı 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(¸. ,) = ŒnC(¸. .)Œn
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. ¨ :) = ı
z
(H) = C(¸
1
. ¨ :).
We collect the results in the next theorem.
(3.4.4) Theorem (Classiﬁcation 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(2¬i t ). These coverings are principal coverings.
3. Let T be a contractible space. Is the identity id: T ÷T a universal Gprincipal covering
for the trivial group G?
76 Chapter 3. Covering Spaces
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 Œu
0
+ u

1
 is contained in
¸
+
¬
1
(1. .).
Proof. If n
0
(1) = n
1
(1), then ¸
+
Œn
0
+ n

1
 = Œu
0
+ u

1
. Conversely: We lift
u
0
+u

1
with initial point .. Since Œu
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 ˆ: 7 ÷ 1 of ¡ with ˆ(:) = . 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. :) Ç ¸
+
¬
1
(1. .). We begin by constructing ˆ 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 deﬁne ˆ by ˆ(:) = ·(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
Œ¸· + ¸·

1
 = ¡
+
Œn + n

1
 ÷ ¸
+
¬
1
(1. .).
By (3.5.1) we have ·(1) = ·
1
(1); this shows that ˆ is welldeﬁned if we set
ˆ(:) = ·(1).
Continuity of ˆ. Let U be an open neighbourhood of ˆ(:
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) Ç V . We claim
ˆ(W) Ç 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
= ¡ ı (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 ﬁnd 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) Ç ¸
+
¬
1
(1).
This inclusion holds, since (using (2.7.3))
m
+
(¸×¸)
+
Œ(n
1
. n
2
) =Œ¸n
1
¸n
2
 =Œ¸n
1
+¸n
2
 =Œ¸(n
1
+n
2
) =¸
+
Œn
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.
Awellknown result of HermannWeyl is that a compact, connected, semisimple
Lie group has a ﬁnite simply connected covering. See [29, V.7] about fundamental
groups of compact Lie groups. The group O(n) has two different twofold coverings
which are nontrivial over SO(n). They are distinguished by the property that the
elements over the reﬂections at hyperplanes have order 2 or 4 (n _ 1). We will see
that ¬
1
(SO(n)) Š 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
classiﬁcation results. The existence of a lift ˆ is equivalent to the existence of a section in
the covering which is obtained by pullback along ¡ . The ¬
1
(7. :)action on the ﬁbre 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 ﬁxed point under the ¬
1
(7. :)action.
If the inclusion of groups holds as in the statement of the theorem, then a ﬁxed 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 ﬁbre of ¸. (2) Aut(¸)
acts transitively on some ﬁbre of ¸. (3) The characteristic subgroup is normal in ¬
1
(T. b).
(4) ¸ is an Aut(¸)principal covering.
78 Chapter 3. Covering Spaces
3.6 The Universal Covering
We collect some of our results for the standard situation that T is path connected,
locally path connected and semilocally 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 ˇ: ¸
2
÷ ¸
1
such that ˛(.
1
) = .
2
and ˇ(.
2
) = .
1
. By
uniqueness of liftings, ˛ˇ and ˇ˛ are the identity, i.e., ˛ and ˇ 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 ﬁbre 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 ﬁbre, ¸n is a loop.
Therefore 
x
is welldeﬁned. 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 ˇ(.), then · + ˇn is a path from . to ˛ˇ(.). Hence 
x
is a homomorphism.
(4) is shown in (3.3.3).
(3.6.2) Theorem (Classiﬁcation 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 Ç ¬
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 semilocally simply connected.
2. Is the product of a countably inﬁnite number of the universal covering of S
1
a covering?
3. Identify in S
1
the open upper and the open lower hemisphere to a point. The resulting
space X has four points. Show ¬
1
(X) Š 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 2¬i.
¸
) is a universal covering of the ndimensional 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
) Ç 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 2fold 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 kfold (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 3fold 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 kfold
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. T·T
1
= ·
1
]. The
three conjugacy classes are represented by 1. ·. T. We can normalize the ﬁrst component
of each orbit correspondingly. If we ﬁx 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.
80 Chapter 3. Covering Spaces
The transitive actions (which yield connected coverings) have the addition c, the normal
subgroups (which yield regular coverings) have the addition n.
Draw ﬁgures 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 3fold covering of S
1
; there are three
of them.
7. Classify the regular 4fold coverings of S
1
S
1
.
8. The Klein bottle has three 2fold 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 ) Š Z. Show that Y
has a simply connected universal Zprincipal 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
ﬁrst 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 COtopology is a topological group and H(X) ×X ÷X, (¡. .) ÷
¡(.) a continuous group action.
14. The space C(S
1
. S
1
) with COtopology 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 ç: R ÷R with ç(0) = 0 and ç(. ÷2¬) = ç(.) or, equivalently,
to the space of continuous functions ˛: S
1
÷ R with ˛(1) = 0. The space V carries the
supnorm and the induced COtopology.
16. Let M(S
1
) be the group of homeomorphisms S
1
÷S
1
of degree 1 with COtopology.
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) ÷¡
·
ı h
is a homomorphism. The space M
1
(S
1
) is homeomorphic tothe space H of homeomorphism
¡ : Œ0. 1 ÷ Œ0. 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 hequivalence. The space
H(S
1
) of homeomorphisms of S
1
is hequivalent 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 deﬁning homotopy limits and homotopy
colimits.
(2) Natural group structures on Homfunctors 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 socalled coﬁbre and ﬁbre se
quences are a fundamental contribution of D. Puppe to homotopy theory
[155]. The exact sequences have a threeperiodic 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 ˇ: 7(¡ ) ÷T correspond bijectively to pairs h: X×1 ÷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
(
˛
Y
ˇ
X
t
(
/
Y
t
together with homotopies ˆ: ¡
t
˛ . ˇ¡ . In the case that the diagram is commu
tative, the pair (˛. ˇ) is a morphism from¡ to ¡
t
in the category of arrows in TOP.
We consider the data (˛. ˇ. ˆ) as a generalized morphism. These data induce a
map ¡ = 7(˛. ˇ. ˆ): 7(¡ ) ÷7(¡
t
) deﬁned by
¡(,) = ˇ(,). , ÷ Y. ¡(.. s) =
_
(˛(.). 2s). . ÷ X. s _ 1,2.
ˆ
2x1
(.). . ÷ X. s _ 1,2.
The diagram
X ÷Y
˛÷ˇ
7(¡ )
7(˛,ˇ,ˆ)
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
, ˇ
t
: Y
t
÷Y
tt
, and a homotopy ˆ
t
: ¡
tt
˛
t
. ˇ
t
¡
t
.
These data yield a composed homotopy ˆ
t
˘ ˆ: ¡
tt
˛
t
˛ . ˇ
t
ˇ¡ deﬁned by
(ˆ
t
˘ ˆ)
t
=
_
ˆ
t
2t
ı ˛. t _ 1,2.
ˇ
t
ı ˆ
2t1
. t _ 1,2.
(This is the product of the homotopies ˆ
t
t
˛ and ˇ
t
ˆ
t
.)
4.1. The Mapping Cylinder 83
(4.1.1) Lemma. There exists a homotopy
7(˛
t
. ˇ
t
. ˆ
t
) ı 7(˛. ˇ. ˆ) . 7(˛
t
˛. ˇ
t
ˇ. ˆ
t
˘ ˆ)
which is constant on X ÷Y .
Proof. Both maps coincide on Y and differ on X×1 by a parameter transformation
of 1.
We also change ˛ and ˇ 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 X×1 ×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
. ˆ induce a homotopy I
t
.
(4.1.2) Lemma. Suppose ·
t
. T
t
with ·
0
= ˛. T
0
= ˇ and ˆ: ¡
t
·
0
. T
0
¡ are
given.Then there exists I
t
with I
0
= ˆ.
Proof. One applies a retraction X × 1 × 1 ÷ X × (d1 × 1 L 1 × 0) to the map
; : X×(d1 ×1 L1 ×0) ÷Y
t
deﬁned by;(.. s. 0) = ˆ(.. s), ;(.. 0. t ) = ¡
t
·
t
(.)
and ;(.. 1. t ) = T
t
¡(.).
Suppose now that X
tt
= X, Y
tt
= Y , ¡
tt
= ¡ and ˛
t
˛ . id, ˇ
t
ˇ . id,
¡ ˛
t
. ˇ
t
¡
t
. We choose homotopies
·
t
: ˛
t
˛ . id. T
t
: ˇ
t
ˇ . id. ˆ
t
: ¡ ˛
t
. ˇ
t
¡
t
.
As before, we have the composition ‰ = ˆ
t
˘ˆ. We use (4.1.2) to ﬁnd 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

) ı 7(˛
t
. ˇ
t
. ˆ
t
): 7(¡
t
) ÷7(¡ ); this morphism restricts
to ˛
t
÷ˇ
t
: X
t
÷Y
t
÷X ÷Y .
(4.1.3) Proposition. There exists a homotopy from Cı 7(˛. ˇ. ˆ) 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

) ı 7(˛
t
˛. ˇ
t
ˇ. ‰). By (4.1.2) we have a further homo
topy to 7(1
A
. 1
Y
. I
1

) ı7(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 ﬁnally a homotopy to the identity (Problems 1 and 2).
(4.1.4) Theorem. Suppose ˛ and ˇ are homotopy equivalences. Then the map
7(˛. ˇ. ˆ) is a homotopy equivalence.
84 Chapter 4. Elementary Homotopy Theory
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 ı 7(˛. ˇ. ˆ) . id we now conclude that
7(˛. ˇ. ˆ) is a homotopy equivalence.
Problems
1. Suppose ˆand ‰ are homotopic relative to X ×d1. Then 7(˛. ˇ. ˆ) and 7(˛. ˇ. ‰) are
homotopic relative to X ÷Y .
2. In the case that ¡
/
˛ = ˇ¡ we have the map 7(˛. ˇ) : 7(¡ ) ÷ 7(¡
/
) induced by
˛ ×id ÷ˇ. Let k be the constant homotopy. Then 7(˛. ˇ. k) . 7(˛. ˇ) relative to X ÷Y .
3. Let Œˆ denote the morphism in …(X. Y
/
) represented by ˆ. We think of (˛. ˇ. Œˆ)
as a morphism from ˛ to ˇ. The composition is deﬁned by (˛
/
. ˇ
/
. Œˆ
/
) ı (˛. ˇ. Œˆ) =
(˛
/
˛. ˇ
/
ˇ. Œˆ
/
˘ ˆ). Show that we obtain in this manner a welldeﬁned category. (This
deﬁnition works in any 2category.)
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 deﬁne
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 Š 1). A commutative
diagram
T
ˇ
·
(
¸
¿
˛
C
;
T
t
·
t
(
/
¸
¿
/
C
t
induces 7(ˇ. ˛. ;): 7(¡. g) ÷7(¡
t
. g
t
), the quotient of ˇ ÷˛ ×id ÷;. We can
also generalize to an hcommutative diagram as in the previous section.
(4.2.1) Theorem. Suppose ˇ, ˛, ; are hequivalences. Then 7(ˇ. ˛. ;) is an
hequivalence.
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 ¸
¹
: · ÷ 7(¡ ) and
¸
B
: · ÷ 7(g). Here the subspace 7(¡ ) corresponds to the image of T ÷· ×
Œ0. 1,2 in 7(¡. g) and 7(g) to the image of · × Œ1,2. 1 ÷C. We view 7(¡. g)
as a space under T ÷ · ÷ C. If we are given homotopies ˆ
B
: ¡
t
˛ . ˇ¡ ,
ˆ
C
: g
t
˛ . ;g, we obtain an induced map
‰ = 7(˛. ˇ. ˆ
B
) L
¹
7(˛. ;. ˆ
C
): 7(¡. g) ÷7(¡
t
. g
t
)
which extends ˇ ÷˛ ÷;.
(4.2.2) Theorem. Let ˛ be an hequivalence with hinverse ˛
t
and suppose ˇ and ;
have left hinverses ˇ
t
. ;
t
. Choose homotopies ·
t
: ˛
t
˛ . id, T
t
: ˇ
t
ˇ . id,
C
t
: ;
t
; . id. Then there exists C: 7(¡
t
. g
t
) ÷ 7(¡. g) and a homotopy from
Cı ‰ to the identity which extends ((k + T
t
) + k ÷(k + ·
t
) + k ÷(k + C
t
) + k).
Proof. The hypotheses imply ˇ
t
¡
t
. ¡ ˛ and ;
t
g
t
. g˛. We can apply (4.1.3)
and ﬁnd left homotopy inverses C
B
of 7(˛. ˇ. ˆ
B
) and C
C
of 7(˛. ;. ˆ
C
). Then
C = C
B
L
¹
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 socalled homotopy colimits. We discuss this in the simplest
case of pushouts.
Given a diagram
X
0
(
¬
(

X
÷
}
¬
X

}

X
(1)
and a homotopy h: ¸

¡

. ¸
÷
¡
÷
. We obtain an induced map ç : 7(¡

. ¡
÷
) ÷X
which is the quotient of (¸

. h. ¸
÷
): X

÷ X
0
× 1 ÷ X
÷
÷ X. We deﬁne:
The diagram (1) together with the homotopy h is called a homotopy pushout or
homotopy cocartesian if the map ç is a homotopy equivalence. This deﬁnition is
in particular important if the diagram is commutative and h the constant homotopy.
Suppose we have inclusions ¡
˙
: X
0
Ç X
˙
and ¸
˙
: X
˙
Ç X such that X =
X

LX
÷
. In the case that the interiors X
ı
˙
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 (deﬁned below).
Then X is the homotopy pushout of ¡
˙
: X
0
Ç X
˙
.
86 Chapter 4. Elementary Homotopy Theory
For the proof we ﬁrst 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 hequivalence over X and under X
˙
.
Proof. Let ; : 1 ÷ 1 be deﬁned by ;(t ) = 0 for t _ 1,3, ;(t ) = 1 for t _ 2,3
and ;(t ) = 3t ÷ 1 for 1,3 _ t _ 2,3. We deﬁne ˇ: N(X

X
÷
) ÷ 7(¡

. ¡
÷
)
as id(X
0
) × ; on X
0
× 1 and the identity otherwise. Homotopies ˛ˇ . id and
ˇ˛ . id are induced by a linear homotopy in the 1coordinate. The reader should
verify that ˇ 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 ÷Œ0. 1, and a function of
this type deﬁnes a section if and only if X · X

Ç s
1
(0). X · X
÷
Ç s
1
(1).
(4.2.5) Lemma. Suppose ¸
1
has a section o. Then ¸
1
is shrinkable.
Proof. A homotopy o ı ¸
1
. id over X is given by a linear homotopy in the
1coordinate.
(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
˙
) Ç Y
˙
. Assume that the induced partial maps
J
˙
: X
˙
÷Y
˙
andJ
0
: X
0
÷Y
0
are hequivalences. ThenJ is anhequivalence.
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
Š S
n÷n÷1
. One can also think of the join as CX × Y L
A×Y
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 identiﬁed 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 Œ†X. Y 
0
. This
set carries a group structure (written additively) which is deﬁned 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 deﬁned group.) The
inverse of Œ¡  is represented by (.. t ) ÷¡(.. 1 ÷t ). For this deﬁnition we do not
need the categorical considerations, but we have veriﬁed 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
k÷I
,d1
k÷I
= 1
k
,d1
k
. 1
I
,d1
I
which is the identity on representing elements in 1
k÷I
= 1
k
× 1
I
. We have for
each pointed space X canonical homeomorphisms
(X . 1
k
,d1
k
) . 1
I
,d1
I
Š X . 1
k÷I
,d1
k÷I
. †
I
(†
k
X) Š †
k÷I
X.
We deﬁne the kfold 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
Œ†
k
X. Y 
0
we have k composition laws, depending on which of the 1coordinates
we use:
(¡ ÷
i
g)(.. t ) =
_
¡(.. t
1
. . . . . t
i1
. 2t
i
. t
i÷1
. . . . ). t
i
_
1
2
.
g(.. t
1
. . . . . t
i1
. 2t
i
÷ 1. t
i÷1
. . . . ).
1
2
_ t
i
.
88 Chapter 4. Elementary Homotopy Theory
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 veriﬁes directly the commutation
rule (unravel the deﬁnitions)
(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 †
+
: Œ·. Y 
0
÷ Œ†·. †Y 
0
, Œ¡  ÷ Œ†¡ , also
called suspension. If · = †X, then †
+
is a homomorphism, because the addition
in Œ†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
Š †
n
S
0
Š 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 nth homotopy group is
¬
n
(X) = ¬
n
(X. .) = Œ1
n
,d1
n
. X
0
= Œ(1
n
. d1
n
). (X. .). n _ 1.
These groups are abelian for n _ 2. We can use each of the n coordinates to deﬁne
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 compactopen
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 ÷¡ ın. 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 deﬁne 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 deﬁnes 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 sufﬁces to prove continuity of the adjoint map
CY × CY × 1 ÷Y. (u. ·. t ) ÷(u + ·)(t ).
This map equals on the closed subspace CY ×CY ×Œ0.
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 msum on ŒX. CY 
0
Œ¡  ÷
n
Œg = Œmı (¡ × g) ı 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
90 Chapter 4. Elementary Homotopy Theory
bijection between morphisms †X ÷ Y and X ÷ CY in TOP
0
. Moreover, this
adjunction induces a bijection Œ†X. Y 
0
Š ŒX. CY 
0
, see (2.4.11). It transforms
the jsum (see the next section) into the msum, and it is natural in the variables
X and Y .
In Œ†X. CY 
0
we have two composition laws ÷
µ
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 hTOP
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 Hspaces, in honour of H. Hopf [91]. In ŒX. M
0
we have the composition law ÷
n
, deﬁned as above for M = CY ; the constant
map represents the neutral element; and Œ÷. M
0
is a contravariant functor into the
category of monoids. Amonoid is a set together with a composition lawwith neutral
element. An Hspace is associative if m(m × id) . m(id ×m) and commutative
if m . mt with the interchange t(.. ,) = (,. .). An inverse for an Hspace
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 Hspace with inverse is a group
object in h TOP
0
. By a general principle we have spelled out the deﬁnition in
TOP
0
. The axioms of a group are satisﬁed up to homotopy. A homomorphism (up
to homotopy) between Hspaces (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 homotopyassociative Hspace be hequivalent 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 Hspace 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(m×id) . r(id ×r) and . ÷r(+. .) is homotopic to
the identity.
A comonoid in hTOP
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 ŒC. Y 
0
we have the composition law ÷
µ
deﬁned as
Œ¡  ÷
µ
Œg = Œı ı (¡ g) ı j
with the codiagonal (also called the folding map) ı = (id. id): Y Y ÷Y . The
functor ŒC. ÷
0
is a covariant functor into the category of monoids. The comultipli
cation is coassociative up to homotopy if (id(C) j) ı j and (jid(C)) ı 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 hTOP
0
; a cohomomorphism up to homotopy ç : C ÷D is a pointed
map such that (ç ç)j and vç are pointed homotopic. A coinverse  : C ÷C for
the comonoid C is a map such that ı(id )j and ı( id)j are both pointed ho
motopic to the constant map. A coassociative comonoid with coinverse in hTOP
0
is called a cogroup in hTOP
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 deﬁne 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 ﬁrst 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
µ
÷÷
....................
1
2
0
1
2
1
1
2
The previously deﬁned group structure on Œ†X. Y 
0
is the jsum.
(4.5.1) Proposition. Let (C. j) be a comonoid and (M. m) a monoid in hTOP
0
.
The composition laws ÷
µ
and ÷
n
in ŒC. M
0
coincide and are associative and
commutative.
Proof. We work in hTOP
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 ÷M×M we set ¡
kI
= ¸
k
¡ i
I
. Thenm¡ = ¸
1
¡ ÷
n
¸
2
¡
and ¡j = ¡ i
1
÷
µ
¡ i
2
. From these relations we derive the commutation rule
(m¡ )j = (¡
11
÷
µ
¡
12
) ÷
n
(¡
21
÷
µ
¡
22
).
m(¡j) = (¡
11
÷
n
¡
21
) ÷
µ
(¡
12
÷
n
¡
22
).
Now apply (4.3.1).
Problems
1. Let X be a pointed space and suppose that the Homfunctor Œ÷. X
0
takes values in the
category of monoids. Then X carries, up to homotopy, a unique Hspace structure m which
92 Chapter 4. Elementary Homotopy Theory
induces the monoid structures on Œ·. X
0
as ÷
m
. There is a similar result for Homfunctors
ŒC. ÷
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 ) Š J((1
k
. d1
k
). (Y. +)) and C
k
C
l
(Y ) Š C
k¬l
(Y ).
3. The space J((1. 0). (Y. +)) Ç 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 ) Š CY .
4.6 The Coﬁbre Sequence
A pointed map ¡ : (X. +) ÷(Y. +) induces a pointed set map
¡
+
: ŒY. T
0
÷ŒX. T
0
. Œ˛ ÷Œ˛¡ .
The kernel of ¡
+
consists of the classes Œ˛ such that ˛¡ is pointed null homotopic.
We work in the category TOP
0
and often omit “pointed” in the sequel. A homotopy
set ŒY. T
0
is pointed by the constant map. A base point is often denoted by +.
A sequence ·
˛
÷÷ T
ˇ
÷÷ C of pointed set maps is exact if ˛(·) = ˇ
1
(+). A
sequence U
(
÷÷V
¿
÷÷W in TOP
0
is called hcoexact if for each T the sequence
ŒU. T
0
ŒV. T
0
(
·
¸
ŒW. 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 deﬁned as CX = X ×1,X ×0 L+ ×1
with base point the identiﬁed 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 Coﬁbre Sequence 93
The mapping cone of ¡ is deﬁned 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 Ç CX÷Y induces an embedding ¡
1
: Y ÷C(¡ ), and CX Ç CX÷Y
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 ˇ: C(¡ ) ÷ T with ˇ¸ = h. If Œ˛ is contained in the kernel
of ¡
+
, then there exists ˇ: C(¡ ) ÷ T with ˇ¡
1
= ˛, i.e., Œ˛ 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 hcoexact.
We iterate the passage from ¡ to ¡
1
and obtain the hcoexact 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 Š CX,i
1
X, by the identity on representatives. Therefore we view ¸, ¸(¡ ),
94 Chapter 4. Elementary Homotopy Theory
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 deﬁne 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 deﬁnition is compatible with the
equivalence relation needed to deﬁne C(¡
1
). The end h
1
of the homotopy has the
form s(¡ ) ı q(¡ ) with s(¡ ): †X ÷C(¡
1
), (.. s) ÷h
1
(.. s). The composition
q(¡ ) ı 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 hinverse 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
†(( )ıt
_ _ _ _
†Y
with an hequivalence q(¡
1
). Let  : †X ÷†X, (.. t ) ÷(.. 1÷t ) 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. †(¡ ) ı  ı q(¡ ) . ¸(¡
1
).
Proof. By (4.6.1) it sufﬁces to study the composition with s(¡ ). We know already
that ¸(¡
1
)s(¡ ): (.. s) ÷(¡(.). min(1. 2(1÷s)) is homotopic to †¡ ı : (.. s) ÷
(¡(.). 1 ÷ s).
4.6. The Coﬁbre Sequence 95
An hcoexact sequence remains hcoexact if we replace some of its spaces by
hequivalent 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 hcoexact.
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 hcoexact sequence. This is due to the next lemma.
(4.6.3) Lemma. There exists a homeomorphism t
1
: C(†¡ ) ÷ †C(¡ ) which
satisﬁes t
1
ı (†¡ )
1
= †(¡
1
).
Proof. C†X and †CX are both quotients of X × 1 × 1. Interchange of 1
coordinates induces a homeomorphism t : C†X ÷ †CX which satisﬁes t ı
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 inﬁnite hcoexact
sequence.
(4.6.4) Theorem. The sequence
X
(
Y
(
1
C(¡ )
](( )
†X
†(
†Y
†(
1
†C(¡ )
†](( )
†
2
X
†
2
(
. . .
is hcoexact. We call it the Puppesequence or the coﬁbre sequence of ¡ ([155]).
The functor Œ÷. T
0
applied to the Puppesequence 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 coﬁbre sequence.
The derivation of the coﬁbre 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. Deﬁne 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 hcoaction of the hcogroup †X on C(¡ ). This
terminology is justiﬁed by the next proposition.
96 Chapter 4. Elementary Homotopy Theory
¨
¨
¨
¨
¨¨
r
r
r
r
rr
C(¡ )
0
1
2
1
.
.
.
.
.
.
.
.
.
.
.
µ
÷÷
¨
¨
¨
¨
¨¨
r
r
r
r
rr
C(¡ )
d
d
d
d
d
d
†X
(4.6.5) Proposition. The map j induces a left action
Œ†X. T
0
×ŒC(¡ ). T
0
Š Œ†X C(¡ ). T
0
µ
·
÷÷ŒC(¡ ). T
0
. (˛. ˇ) ÷˛ ˇˇ.
This action satisﬁes ˛
1
ˇ¸(¡ )
+
˛
2
= ¸(¡ )
+
(˛
1
˛
2
). Moreover, ¡
+
1
(ˇ
1
) = ¡
+
1
(ˇ
2
)
if and only if there exists ˛ such that ˛ ˇˇ
1
= ˇ
2
. Thus ¡
+
1
induces an injective
map of the orbits of the action.
Proof. That j
+
satisﬁes 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] Ç C(¡ ) and
C
1
= {(.. t ) [ 2t _ 1] L Y Ç C(¡ ). These inclusions are coﬁbrations (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 Œk ˇŒ¡  = Œg. Conversely, jı (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
)
1
X,·
¯
1
T
J
Y
q
Y,T
2. The map ¸(¡ )
·
: Œ†X. T
0
÷ŒC(¡ ). T
0
induces an injective map of the left (or right)
cosets of Œ†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
¡
+
: ŒT. X
0
÷ŒT. Y 
0
. Œ˛ ÷Œ¡ ˛.
A sequence U
(
÷÷V
¿
÷÷W in TOP
0
is called hexact, if for each T the sequence
ŒT. U
0
(
·
ŒT. V 
0
¿
·
ŒT. 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 ) Š 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 deﬁne J(¡ ) via a pullback
J(¡ )
q
(
1
JY
e
1
J(¡ ) = {(.. n) ÷ X × JY [ ¡(.) = n(1)]
X
(
Y
¡
1
(.. n) = .. q(.. n) = n.
The maps ˇ: T ÷ J(¡ ) correspond to pairs ˛ = ¡
1
ˇ: T ÷ X together with
the null homotopies qˇ: T ÷JY of ¡ ˛. This shows that J(¡ )
(
1
÷÷X
(
÷÷Y is
hexact.
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 hequivalence, by CY and
CX. We begin with the remark that
(¡
1
)
1
(+) = {(.. n) [ n(0) = +. n(1) = ¡(.). . = +]
can be identiﬁed with CY , via n ÷ (+. n). Let i(¡ ): CY ÷ J(¡ ) be the
associated inclusion of this ﬁbre of ¡
1
. The space (by its pullback deﬁnition)
J(¡
1
) = {(.. n. ·) [ n(0) = +. n(1) = ¡(.). . = ·(1). ·(0) = +]
98 Chapter 4. Elementary Homotopy Theory
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
+
)
satisﬁes ¡
2
ı ¸(¡ ) = 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 deﬁne
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 ¸(¡ ) ı r(¡ ) with
r(¡ ): J(¡
1
) ÷CY. (n. ·) ÷n + (¡ ·)

.
The relation (r(¡ ) ı¸(¡ ))(n) = n+k

+
shows that also r(¡ ) ı¸(¡ ) 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 ﬁbre over the base point. Let  : CY ÷ CY ,
n ÷n

be the inverse.
(4.7.2) Note. ¸(¡ ) ı  ı C¡ . i(¡
1
).
Proof. We compose both sides with the hequivalence r(¡ ) from the proof of
(4.7.1). Then r(¡ )ıi(¡
1
) equals · ÷k
+
+(¡ ·)

, and this is obviously homotopic
to  ı C¡ : · ÷(¡ ·)

.
As a consequence of the preceding discussion we see that the sequence
CX
D(
CY
i(( )
J(¡ )
(
1
X
(
Y
is hexact.
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
) ı t
1
.
Proof. From the deﬁnitions and standard properties of mapping spaces we have
CJ(¡ ) Ç CX × CJY and JC(¡ ) Ç CX × JCY . We use the exponential law
for mapping spaces and consider CJY and JCY as subspaces of Y
J×J
. In the
ﬁrst 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 1coordinates yields a homeomorphism and it induces t
1
.
We now continue as in the previous section.
(4.7.4) Theorem. The sequence
. . .
D
2
(
C
2
Y
Di(( )
CJ(¡ )
D(
1
CX
D(
CY
i(( )
J(¡ )
(
1
X
(
Y
is hexact. We call it the ﬁbre sequence of ¡ . When we apply the functor ŒT. ÷
0
to the ﬁbre 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 ﬁbre sequence under adjunction. A map
a: T ÷J(¡ ) = {(.. n) ÷ X × JY [ ¡(.) = n(1)]
has two components b : T ÷ X and ˇ: T ÷ JY . Under adjunction, ˇ 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 Œa ÷ ŒT. J(¡ )
0
Š Œi
1
. ¡ 
0
÷ ŒT. b. This bijection
transforms ¡
1
·
into the restriction ŒT. b ÷ Œi
1
. ¡ 
0
÷ ŒT. X
0
÷ Œb. In the next step we
have
ŒT. CY 
0
i(¸ )
·
Š
Œ†T. Y 
0
Œ;
_
ŒT. J(¡ )
0
Š
Œi
1
. ¡ 
0
,
Œ; ı ¸. c.
The image of ; is obtained in the following manner: With the quotient map ¸: CT ÷†T
we have T = ; ı ¸, and b is the constant map c.
3. There exist several relations between ﬁbre and coﬁbre sequences.
The adjunction (†. C) yields in TOP
0
the maps j: X ÷C†X (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(¡ )
100 Chapter 4. Elementary Homotopy Theory
deﬁned by
j(.. n)(t ) =
_
Œ.. 2t . t _ 1,2.
n(2 ÷ 2t ). t _ 1,2.
and c adjoint to j. Verify the following assertions from the deﬁnitions.
(1) The next diagram is homotopy commutative
CY
i(¸ )
t
J(¡ )
¸
1
)
X
)
CY
:¸
1
CC(¡ )
:)(¸ )
C†X.
(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 ı J(r) .  ı ¬ holds.
(3) The next diagram is homotopy commutative
J(¡
1
)
i(¸
1
)
:q(¸
1
)ı)
Y
tı)
C†X
:†¸
C†Y .
Chapter 5
Coﬁbrations 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 : · Ç 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 ﬁnd conditions on i under which the extendibility of ¡ only depends
on the homotopy class of ¡ . If this is the case, then ¡ is called a coﬁbration.
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 ﬁbration.
X
T
¸x
x
x
x
Y ·
(
¸
i
¸
1
]
X
T
x
x
x
x
(
T
Each map is the composition of a coﬁbration and a homotopy equivalence and
(dually) the composition of a homotopy equivalence and a ﬁbration. The notions are
then used to deﬁne homotopy ﬁbres (“homotopy kernels”) and homotopy coﬁbres
(“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 ﬁbrewise homotopy equivalences.
Another simple typical example: A base point . ÷ X is only good for homo
topy theory if the inclusion {.] Ç X is a coﬁbration (or the homotopy invariant
weakening, a socalled hcoﬁbration). 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. Coﬁbrations and Fibrations
is a coﬁbration 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).
·
h
i
Y
J
e
0
X
1
}
}
}
}
(
Y
X
i
³
0
J
J
J
J
J
J
J
J
J
J
(
·
i
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
i×id
t
t
t
t
t
t
t
t
t
h
For a coﬁbration i : · ÷ X, the extendibility of ¡ only depends on its homotopy
class.
From this deﬁnition one cannot prove directly that a map is a coﬁbration, but
it sufﬁces to test the HEP for a universal space Y , the mapping cylinder 7(i ) of i .
Recall that 7(i ) is deﬁned by a pushout
·
i
i
.
0
X
b
· × 1
k
7(i ).
Pairs of maps ¡ : X ÷Y and h: ·×1 ÷Y with hi
¹
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 coﬁbration. 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, ﬁnd 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 coﬁbration.
(2) i has the HEP for the mapping cylinder 7(i ).
(3) s : 7(i ) ÷X × 1 has a retraction.
A coﬁbration 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 coﬁbrations
whenever this simpliﬁes the exposition. A pointed space (X. .) is called well
pointed and the base point nondegenerate if {.] Ç X is a closed coﬁbration.
5.1. The Homotopy Extension Property 103
(5.1.2) Proposition. If i : · Ç X is a coﬁbration, 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 coﬁbration.
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: X×0L·×1 ÷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
n1
Ç D
n
and d1
n
Ç 1
n
are coﬁbrations. A
retraction r : D
n
÷S
n1
× 1 L D
n
× 0 was constructed in (2.3.5). Þ
It is an interesting fact that one need not assume · to be closed. Strøm [180,
Theorem 2] proved that an inclusion · Ç X is a coﬁbration 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) coﬁbration for each Y , if i : · ÷ X is a (closed) coﬁbration.
Since we have proved (5.1.2) only for closed coﬁbrations, we mention another
special case, to be used in a moment. Let Y be locally compact and i : · ÷ X a
coﬁbration. Then i ×id: ·×Y ÷X ×Y is a coﬁbration. 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 · Ç X and assume that · × 1 Ç X × 1 has the HEP
for Y . Given maps ç : · × 1 × 1 ÷ Y. H: X × 1 ÷ Y. ¡
t
: X × 1 ÷ Y such
that
ç(a. s. 0) = H(a. s). ¡
t
(.. 0) = H(.. c). ¡
t
(a. t ) = ç(a. c. t )
c ÷ {0. 1], a ÷ ·, . ÷ X, s. t ÷ 1. Then there exists ˆ: X ×1 ×1 ÷Y such that
ˆ(a. s. t ) = ç(a. s. t ). ˆ(.. s. 0) = H(.. s). ˆ(.. c. t ) = ¡
t
(.. t ).
Proof. H and ¡
t
together yield a map ˛: X × (1 × 0 L d1 × 1) ÷ Y deﬁned
by ˛(.. s. 0) = H(.. s) and ˛(.. c. t ) = ¡
t
(.. t ). By our assumptions, ˛ and ç
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 ç ı (1 × k
1
) and ˛ ı (1 × k
1
). The
map ˆ = ‰ ı (1 × k) solves the extension problem.
104 Chapter 5. Coﬁbrations and Fibrations
(5.1.6) Proposition. Let i : · Ç X be a coﬁbration. Then X×d1 L·×1 Ç X×1
is a coﬁbration.
Proof. Given h: (·×1 LX×d1)×1 ÷Y and an initial condition H: X×1 ÷Y ,
we set ç = 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 Ç X × 1 is a coﬁbration, in
particular d1 Ç 1 and {0] Ç 1 are coﬁbrations. Induction over n shows again that
d1
n
Ç 1
n
is a coﬁbration.
We list some special cases of (5.1.5) for a coﬁbration · Ç X.
(5.1.7) Corollary. (1) Let ˆ: X ×1 ÷Y be a homotopy. Suppose ç = ˆ[·×1
is homotopic rel ·×d1 to [. Then ˆ is homotopic rel X ×d1 to ‰: X ×1 ÷Y
such that ‰[· × 1 = [.
(2) Let ˆ solve the extension problem for (ç. ¡ ) and ‰ the extension problem
for ([. g). Suppose ¡ . g rel · and ç . [ rel · × d1. Then ˆ
1
. ‰
1
rel ·.
(3) Let ˆ. ‰: X × 1 ÷ Y solve the extension problem for (h. ¡ ). Then there
exists a homotopy I : ˆ . ‰ 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 coﬁbration, then J is
a coﬁbration.
Proof. Suppose h: T × 1 ÷ 7 and ç : 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 coﬁbration, there exists 1
t
: X ÷ 7 such that 1
0
= ç¡
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
= ç, since both
maps have the same composition with J¸ and J¡ .
We call J the coﬁbration induced from ¸ via cobase change along ¡ .
Example. If · Ç X is a coﬁbration, then {·] Ç X,·is a coﬁbration. S
n1
Ç D
n
is a coﬁbration, hence {S
n1
] Ç D
n
,S
n1
is a coﬁbration. The space D
n
,S
n1
is homeomorphic to S
n
; therefore (S
n
. +) is wellpointed. Þ
Example. If (X
}
) is a family of wellpointed spaces, then the wedge
_
}
X
}
is
wellpointed. Þ
5.1. The Homotopy Extension Property 105
Our next result, the homotopy theoremfor coﬁbration says, among other things,
that homotopic maps induce hequivalent coﬁbrations froma givencoﬁbrationunder
a cobase change.
Let ¸ : · ÷ X be a coﬁbration and ç
t
: ¡ . g: · ÷ T a homotopy. We
consider two pushout diagrams.
·
(
}
T
}
¸
·
¿
}
T
}
¡
X
T
Y
( X
G
Y
¿
Since ¸ is a coﬁbration, there exists a homotopy ˆ
t
: X ÷Y
(
with initial condition
ˆ
0
= J and ˆ
t
¸ = ¸
(
ç
t
. The pushout property of the Y
¿
diagram provides us
with a unique map k = k
ç
such that k¸
¿
= ¸
(
and kG = ˆ
1
. (We use the notation
k
ç
although the map depends on ˆ
1
.) Thus k
ç
is a morphism of coﬁbrations
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 ç
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 ç 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) = ˆ(.. 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 1ı(G×id) = I
1
and 1 ı (¸
¿
× id) = ¸
(
ı pr where I
1
: X × 1 ÷ Y
(
, (.. t ) ÷ I(.. 1. t ). By
construction, 1 is a homotopy under T fromk
ç
to a corresponding map k
)
obtained
from [
t
and ‰
t
. We thus have shown that the homotopy class Œk
B
under T of k
only depends on the morphism Œç from ¡ to g in the groupoid …(·. T). Let us
write Œk = ˇŒç.
We verify that ˇ is a functor ˇ(Œ[ ~Œç) = ˇŒç ı ˇŒ[. 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
ç
‰
0
= k
ç
G = ˆ
1
, we can form the product
homotopy ˆ+k
ç
‰. It has the initial condition J and satisﬁes (ˆ+k
ç
‰)(¸ ×id) =
¸
(
ç + k
ç
¸
¿
[ = ¸
(
(ç + [). Hence k
ç+)
, constructed with this homotopy, is
determined by k
ç+)
H = k
ç
‰
1
= k
ç
k
)
H and k
ç+)
¸
h
= ¸
(
= k
ç
¸
¿
= k
ç
k
)
¸
h
.
Therefore k
ç
k
)
represents ˇ(Œ[ ~Œç).
Let hCOF
B
denote the full subcategory of hTOP
B
with objects the coﬁbrations
under T. Then we have shown above:
106 Chapter 5. Coﬁbrations and Fibrations
(5.1.9) Theorem. Let ¸ : · ÷ X be a coﬁbration. We assign to the object
¡ : · ÷ T in …(·. T) the induced coﬁbration ¸
(
: T ÷ Y
(
and to the mor
phism Œç : ¡ ÷ g in …(·. T) the morphism Œk
ç
 : ¸
¿
÷ ¸
(
. These assignments
yield a contravariant functor ˇ
}
: …(·. T) ÷hCOF
B
.
Since …(·. T) is a groupoid, Œk
ç
 is always an isomorphism in hTOP
B
. We
refer to this fact as the homotopy theorem for coﬁbrations.
(5.1.10) Proposition. In the pushout (5.1.8) let ¸ be a coﬁbration and ¡ a homotopy
equivalence. Then J is a homotopy equivalence.
Proof. With an hinverse g: T ÷· of ¡ we form a pushout
T
¿
J
·
i
Y
G
7.
Since g¡ . id, there exists, by (5.1.9), an hequivalence k : 7 ÷ X under ·
such that kGJ . id. Hence J has a left hinverse and G a right hinverse. Now
interchange the roles of J and G.
Problems
1. A coﬁbration is an embedding. For the proof use that i
1
: · ÷ 7(i ), a ÷ (a. 1) is an
embedding. From i
1
= rsi
1
= ri
³
1
i then conclude that i is an embedding.
Consider a coﬁbration as an inclusion i : · Ç 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 deﬁne 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 : · Ç X if i
is a coﬁbration.)
Let X be a Hausdorff space. Then a coﬁbration 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 coﬁbration. 0 Ç X is a coﬁbration. The sum li
¸
: l ·
¸
÷ lX
¸
of
coﬁbrations i
¸
: ·
¸
÷X
¸
is a coﬁbration.
3. Let ¸: 1 ÷ Q be an hequivalence and i : · Ç T a coﬁbration. 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 · Ç X be a coﬁbration 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) Ç T.
5. Let · Ç X be a coﬁbration 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
coﬁbrations.
7. Let ¡ : · Ç X be an inclusion. We have a pushout diagram
·
¸
¸
C
/
·
1
X
J
X L C
/
·.
Since ¸ is a coﬁbration, so is J. If ¡ is a coﬁbration, then J is a coﬁbration. There exists
a canonical homeomorphism X L C
/
·,C
/
· Š X,·; it is induced by J. Since C
/
· is
contractible, we obtain a homotopy equivalence X L C
/
· ÷X L C
/
·,C
/
· Š 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 {+] Ç X is a closed coﬁbration, then ¸ is a closed (induced)
coﬁbration. The quotient map †
/
X ÷†X is a homotopy equivalence.
5.2 Transport
Let i : 1 ÷· be a coﬁbration and ç : 1 × 1 ÷X a homotopy. We deﬁne a map
ç
#
: Œ(·. i ). (X. ç
0
)
1
÷Œ(·. i ). (X. ç
1
)
1
.
called transport along ç, as follows: Let ¡ : · ÷ X with ¡ i = ç
0
be given.
Choose a homotopy ˆ
t
: · ÷ X with ˆ
0
= ¡ and ˆ
t
i = ç
t
. We deﬁne
ç
#
Œ¡  = Œˆ
1
. Then (5.1.5) shows that ç
#
is well deﬁned and only depends on the
homotopy class of ç rel 1 × d1, i.e., on the morphism Œç ÷ …(1. X). From the
construction we see (ç + [)
#
= [
#
ç
#
. Altogether we obtain:
(5.2.1) Proposition. Let i : 1 ÷ · be a coﬁbration. The assignments ç
0
÷
Œi. ç
0

1
and Œç ÷ ç
#
yield a transport functor from …(1. X) to sets. Since
…(1. X) is a groupoid, ç
#
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 deﬁnitions.
(5.2.2) Proposition. Let i : 1 ÷· be a coﬁbration. Let ¡ : (·. i ) ÷(X. g) and
¡
t
: (·. i ) ÷(X. g
t
) be morphisms in TOP
1
. Then Œ¡  = Œ¡
t
, if and only if there
exists Œç ÷ …(1. X) from (X. g) to (X. g
t
) with Œ¡
t

1
= ç
#
Œ¡ 
1
.
(5.2.3) Proposition. Let i : 1 ÷ · be a coﬁbration, g: 1 ÷ X a map, and
[: X × 1 ÷ Y a homotopy. Then ([ ı (g × id))
#
ı [
0+
= [
1+
, if we set
[
i+
Œ¡  = Œ[
i
¡ .
Proof. Use that [
t
¡ is an extension of [
t
g and apply the deﬁnition.
108 Chapter 5. Coﬁbrations and Fibrations
(5.2.4) Proposition. Let ¡ : X ÷ Y be an ordinary homotopy equivalence and
i : 1 ÷ · a coﬁbration. Then ¡
+
: Œ(·. i ). (X. g)
1
÷ Œ(·. i ). (Y. ¡g)
1
is
bijective.
Proof. Let g be hinverse to ¡ and choose ç : id . g¡ . Consider
Œi. ·
1
(
·
Œi. ¡ ·
1
¿
·
Œi. g¡ ·
1
(
·
Œi. ¡g¡ ·
1
.
Since g
+
¡
+
= (g¡ )
+
= Œç(· ×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 coﬁbrations and ¡ : X ÷
Y an hequivalence such that ¡ i = ¸ . Then ¡ is an hequivalence under 1.
Proof. By (5.2.4), we have a bijective map
¡
+
: Œ(Y. ¸ ). (X. i )
1
÷Œ(Y. ¸ ). (Y. ¸ )
1
.
Hence there exists Œg with ¡
+
Œg
1
= Œ¡g
1
= Œid
1
. Since ¡ is an hequivalence,
so is g. Since also g
+
is bijective, g has a homotopy right inverse under 1. Hence
g and ¡ are hequivalences under 1.
(5.2.6) Proposition. Let i : · ÷X be a coﬁbration and an hequivalence. Then i
is a deformation retract.
Proof. The map i is a morphism from id(·) to i . By (5.2.5), i is an hequivalence
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
·
u
i
Y
¿
X
¸
¸
(
·
t
u
/
Y
t
X
t
¸
/
¸
with a coﬁbration i and hequivalences ¸ and ¸
t
. Given ·: · ÷X and ç : ¸· . u.
Then there exists ·
t
: ·
t
÷ X
t
and ç
t
: ¸
t
·
t
. u
t
such that ·
t
i = ¡ · and
ç
t
t
i = gç
t
.
Proof. We have bijective maps (note ¸
t
¡ · = g¸· . gu = u
t
i )
(gç)
#
ı ¸
t
+
: Œ(·
t
. i ). (X
t
. ¡ ·)
¹
÷Œ(·
t
. i ). (Y
t
. ¸
t
¡ ·)
¹
÷Œ(·
t
. i ). (Y
t
. u
t
i )
¹
.
Let ·
t
: ·
t
÷ X
t
be chosen such that (gç)
#
¸
t
+
Œ·
t

¹
= Œu
t

¹
. This means: ·
t
i =
¡ ·; and ¸
t
·
t
has a transport along gç to a map which is homotopic under · to u
t
.
5.2. Transport 109
This yields a homotopy ç
tt
: ¸
t
·
t
. u
t
such that ç
tt
(i(a). t ) = gç(a. min(2t. 1)).
The homotopy ç + k : (a. t ) ÷ç(a. min(2t. 1)) is homotopic rel ·×d1 to ç. We
nowuse (5.1.6) in order to change this ç
tt
into another homotopy ç
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
·
(
i
T
}
X
T
Y
with coﬁbrations i , ¸ and hequivalences ¡ and J. Given g: T ÷·and ç : g¡ .
id, there exists G: Y ÷ X and ˆ: GJ . id such that G¸ = ig and ˆ
t
i = iç
t
.
In particular: (J. ¡ ) is an hequivalence 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 coﬁbration a
}
, b
}
and hequivalences ¡
}
. 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
homotopy equivalence.
Proof. We choose inductively hequivalences J
n
: Y
n
÷ X
n
such that a
n
J
n1
=
J
n
b
n
and homotopies ç
n
: X
n
×1 ÷X
n
fromJ
n
¡
n
to id(X
n
) such that a
n
ç
n1
=
ç
n
(a
n
× id). This is possible by (5.2.7). The J
n
and ç
n
induce J : Y ÷ X and
ç : X × 1 ÷X. J¡ . id. Hence J is a left homotopy inverse of ¡ .
Problems
1. Let i : 1 ÷ · and ¸ : 1 ÷ T be coﬁbrations. Let ˛: (T. ¸ ) ÷ (·. i ) be a morphism
under 1, ¸ : X ÷Y a continuous map, and ç : 1 × 1 ÷X a homotopy. Then
Œ(·. i ). (X. ç
0
)
!
µ
#
Œ˛,¸
!
Œ(·. i ). (X. ç
1
)
!
Œ˛,¸
!
Œ(T. ¸ ). (Y. ¸ç
0
)
!
(¸µ)
#
Œ(T. ¸ ). (Y. ¸ç
1
)
!
110 Chapter 5. Coﬁbrations and Fibrations
commutes; here Œ˛. ¸
!
Œ¡  = Œ¸¡ ˛.
2. Apply the transport functor to pointed homotopy sets. Assume that the inclusion {+] Ç ·
is a coﬁbration. For each path n: 1 ÷X we have the transport
n
#
: Œ·. (X. n(0))
0
÷Œ·. (X. n(1))
0
.
As a special case we obtain a right action of the fundamental group (transport along loops)
Œ·. X
0
× ¬
1
(X. +) ÷Œ·. X
0
. (.. ˛) ÷. ˛ = ˛
#
(.).
Let ·: Œ·. X
0
÷ Œ·. X denote the forgetful map which disregards the base point. The
map · induces an injective map from the orbits of the ¬
1
action into Œ·. X. This map is
bijective, if X is path connected.
A space is said to be Asimple 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 Œ·. (X. .)
0
. If · = S
n
, then we say nsimple instead of ·simple. We call X simple if
it is ·simple for each wellpointed ·.
3. The action on Œ1,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 Œ·. 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 Œ·. 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) = ŒS(1). X
0
. Thenwe canidentifyŒ·. X
0
×¬
1
(X. +) Š
Œ·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 ç : · × 1 ÷ · S(1) with the initial condition · Ç · S(1) and set v = ç
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 hTOP
0
. Then the ¬
1
(X. e)action on Œ·. X
0
is trivial.
5.3 Replacing a Map by a Coﬁbration
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 closedcoﬁbration, the maps (s. ¸ ), s and¸ are closedcoﬁbrations.
We also have the projection q : 7(¡ ) ÷ Y , (.. t ) ÷ ¡(.), , ÷ ,. In the case
that ¡ : X Ç Y , let ¸: Y ÷Y,X be the quotient map. We also have the quotient
5.3. Replacing a Map by a Coﬁbration 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
q
¸
C(¡ )
i
¸
1s = c(¡ ). ¸q = r1
(5.3.1) Proposition. The following assertions hold:
(1) ¸ and s are coﬁbrations.
(2) sq is homotopic to the identity relative to Y . Hence s is a deformation
retraction with hinverse q.
(3) If ¡ is a coﬁbration, then q is a homotopy equivalence under X and r the
induced homotopy equivalence.
(4) c(¡ ): Y ÷C(¡ ) is a coﬁbration.
Proof. (1) was already shown.
(2) The homotopy contracts the cylinder X × 1 to X × 0 and leaves Y ﬁxed,
h
t
(.. c) = (.. t c ÷1 ÷ t ), h
t
(,) = ,.
(3) is a consequence of (5.2.5).
(4) c(¡ ) is induced from the coﬁbration i
0
: X ÷ X × 1,X × 1 via cobase
change along ¡ .
We have constructed a factorization ¡ = q¸ into a (closed) coﬁbration 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 satisﬁes i q
t
¸
t
. i . Since ¸
t
is a coﬁbration, we can change i q
t
. k
such that k¸
t
= ¸ . Since i q
t
is an hequivalence, the map k is an hequivalence 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 coﬁbration ¸ and a homo
topy equivalence q, then 7,¸(X) is called the (homotopical) coﬁbre of ¡ . The
uniqueness of the factorization implies uniqueness up to homotopy equivalence of
the coﬁbre. If ¡ : X Ç Y is already a coﬁbration, then Y ÷Y,X is the projection
onto the coﬁbre; in this case q : 7 ÷Y is an hequivalence under X.
The factorization of a map into a coﬁbration and a homotopy equivalence is a
useful technical tool. The proof of the next proposition is a good example.
112 Chapter 5. Coﬁbrations and Fibrations
(5.3.2) Proposition. Let a pushout diagram
·
(
}
T
J
X
T
Y
with a coﬁbration ¸ 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
¹
id: 7(¡ ) L
¹
T = 7(¡. ¸ ) ÷X L
¹
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
I
C
1
;








T
1
ˇ
,}
}
}
}
}
}
}
D
ı
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 ˛,
ˇ, ; are homotopy equivalences, then ı is a homotopy equivalence.
Proof. From the data of the diagram we obtain a commutative diagram
7(k. l)
7(ˇ,˛,;)
ç
7(k
t
. l
t
)
ç
/
D
ı
D
t
where ç and ç
t
are the canonical maps. By hypothesis, ç and ç
t
are homotopy
equivalences. By (4.2.1) the map 7(ˇ. ˛. ;) 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 ı is induced by ˛, ˇ, ;.
Suppose that ˛, ˇ, ; are homotopy equivalences and that one of the maps k, l and
one of the maps k
t
, l
t
is a coﬁbration. Then ı 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 Coﬁbrations 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
deformation retract.
2. In the case of a pointed map ¡ : (X. +) ÷(Y. +) one has analogous factorizations into a
coﬁbration and a homotopy equivalence. One replaces the mapping cylinder 7(¡ ) with the
pointed mapping cylinder 7
0
(¡ ) deﬁned 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 coﬁbrations. 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 coﬁbre 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 ˇ: C(g¡ ) ÷C(g); ˛ is the identity on the cone and maps Y by g, and ˇ is the
identity on 7 and maps the cone by ¡ × id. Show that ˇ is the pointed homotopy coﬁbre
of ˛.
5.4 Characterization of Coﬁbrations
We look for conditions on · Ç X which imply that this inclusion is a coﬁbration.
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 ÷ Œ0. oŒ and a homotopy
ç : X × 1 ÷X such that:
(1) · Ç u
1
(0)
(2) ç(.. 0) = . for . ÷ X
(3) ç(a. t ) = a for (a. t ) ÷ · × 1
(4) ç(.. t ) ÷ · for t > u(.).
Proof. Suppose we are given a retraction r. We set ç(.. t ) = pr
1
ı r(.. t ) and
u(.) = max{t ÷ pr
2
ı r(.. t ) [ t ÷ 1]. For (4) note the following implications:
t > u(.), pr
2
r(.. t ) > 0, r(.. t ) ÷ · × 1, ç(.. t ) ÷ ·. The other properties
are immediate from the deﬁnition. Conversely, given u and ç, then r(.. t ) =
(ç(.. 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 ç(.. u(.)) ÷
¨
·. If u(.) = 0, then . = ç(.. 0) = ç(.. u(.)) ÷
¨
·. Thus
114 Chapter 5. Coﬁbrations and Fibrations
¨
· = u
1
(0). Therefore in a closed coﬁbration · Ç X the subspace · has the
remarkable property of being the zeroset of a continuous realvalued function. Þ
(5.4.3) Lemma. Let u: X ÷1 and · = u
1
(0). Let ˆ: ¡ . g: X ÷7 rel ·.
Then there exists
¯
ˆ: ¡ . g rel · such that
¯
ˆ(.. t ) =
¯
ˆ(.. u(.)) =
¯
ˆ(.. 1) for
t _ u(.).
Proof. We deﬁne
¯
ˆ by
¯
ˆ(.. t ) = ˆ(.. 1) for t _ u(.) and by ˆ(.. t u(.)
1
) for
t < u(.). For the continuity of
¯
ˆ 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 NDRpresentation of (X. ·).
(5.4.4) Proposition. (X. ·) is a closed coﬁbration if and only if it is an NDR.
Proof. If · Ç X is a closed coﬁbration, then an NDRpresentation is obtained
from (5.4.1) and (5.4.2). For the converse, we modify an NDRpresentation ([. u)
by (5.4.3) and apply (5.4.1) to the result (
¯
[. u).
(5.4.5) Theorem (Union Theorem). Let · Ç X, T Ç X, and · ¨ T Ç X be
closed coﬁbrations. Then · L T Ç X is a coﬁbration.
Proof ([112]). Let ç : (·LT) ×1 ÷7 be a homotopy and ¡ : X ÷7 an initial
condition. There exist extensions ˆ
¹
: X×1 ÷7 of ç[·×1 and ˆ
B
: T×1 ÷7
of ç[T × 1 with initial condition ¡ . The homotopies ˆ
¹
and ˆ
B
coincide on
(· ¨ T) × 1. Therefore there exists ‰: ˆ
¹
. ˆ
B
rel (· ¨ T) × 1 L X × 0.
Let ¸: X × 1 ÷X × 1, ~ be the quotient map which identiﬁes each interval
{c] × 1, c ÷ · ¨ T to a point. Let T : 1 × 1 ÷ 1 × 1 switch the factors. Then
‰ ı (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). Deﬁne ¸ : 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 ç and ¡ is now given by Cı (¸ × id).
(5.4.6) Theorem (Product Theorem). Let · Ç X and T Ç Y be closed coﬁbra
tions. Then the inclusion X × T L · × Y Ç X × Y is a coﬁbration.
Proof. · × X Ç X × Y , X × T Ç X × Y , and · × T = (· × Y ) ¨ (X × T) Ç
X × T Ç X × Y are coﬁbrations. 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
¯
ˆq = ˆ.
It sufﬁces 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] Ç Œ0. 1 is not a coﬁbration. The inclusions ·
¸
=
{0. ¸
1
] Ç 1 are coﬁbrations. Hence (5.4.5) does not hold for an inﬁnite number of coﬁbra
tions.
3. Set X = {a. b] with open sets 0. {a]. X for its topology. Then · = {a] Ç X is a
nonclosed coﬁbration. The product X × · L · × X Ç X × X is not a coﬁbration.
4. Let ·
¸
Ç X be closed coﬁbrations (1 _ ¸ _ n). For all o Ç {1. . . . . n] let ·
o
=
_
¸÷o
·
¸
Ç X be a coﬁbration. Then
_
n
1
·
¸
Ç X is a coﬁbration.
5. Let · and T be wellpointed spaces. Then · . T is wellpointed.
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 ﬁbration (sometimes Hurewicz ﬁbration) if it has the HLP for all
spaces. It is called a Serre ﬁbration if it has the HLP for all cubes 1
n
, n ÷ N
0
.
Serre ﬁbrations sufﬁce for the investigation of homotopy groups. In order to see
the duality we can use the dual deﬁnition 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
i
³
0
1
]
X × 1
h
1
x
x
x
x
x
T
1
]
,}
}
}
}
}
}
}
}
T
1
J
e
0
1
¸C
C
C
C
C
C
C
C
]
1
¸







X
1
¸
o
¸
h
¸
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 deﬁned
by the pullback
1
]
W(¸)
b
¸
k
W(¸) = {(.. n) ÷ 1 × T
J
[ ¸(.) = n(0)].
T
T
J
e
0
!
¸
k(.. n) = n. b(.. n) = ..
116 Chapter 5. Coﬁbrations and Fibrations
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 ﬁbration.
(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 : · Ç X be a
closed coﬁbration and an hequivalence. 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 ç : X×1 ÷X
rel ·such that · = u
1
(0), ç
1
= id(X), ç
0
(X) Ç ·. Set r : X ÷·, . ÷ç
0
(.).
Deﬁne a new homotopy ˆ: X ×1 ÷X by ˆ(.. t ) = ç(.. t u(.)
1
) for t < u(.)
and ˆ(.. t ) = ç(.. 1) = . for t _ u(.). We have seen in (5.4.3) that ˆ is
continuous. Apply the HLP to h = ¡ ˆ with initial condition b = ar : X ÷ 1.
The veriﬁcation
h(.. 0) = ¡ ˆ(.. 0) = ¡ ç(.. 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 veriﬁes 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 : · Ç X be
a closed coﬁbration. 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.
Proof. This is a consequence of (5.1.3) and (5.5.2).
(5.5.4) Proposition. Let i : · Ç T be a (closed) coﬁbration of locally compact
spaces. The restriction from T to · yields a ﬁbration ¸: 7
B
÷7
¹
.
Let ¸: X ÷ T be a ﬁbration. Then ¸
7
: X
7
÷ T
7
is a ﬁbration for locally
compact 7.
Proof. Use adjunction and the fact that X×· ÷X×T is a coﬁbration for each X.
5.5. The Homotopy Lifting Property 117
(5.5.5) Proposition. Let ¸: 1 ÷ T be a ﬁbration. Then r : 1
J
÷ W(¸), · ÷
(·(0). ¸·) is a ﬁbration.
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 ﬁbration, T
0
Ç T and 1
0
= ¸
1
(T
0
).
If T
0
Ç T is a closed coﬁbration, then 1
0
Ç 1 is a closed coﬁbration.
Proof. Let u: T ÷ 1 and h: T × 1 ÷ T be an NDRpresentation of T
0
Ç T.
Let H: X × 1 ÷X solve the homotopy lifting problem for h(¸ × id) with initial
condition id(X). Deﬁne 1: X ÷ X by 1(.. t ) = H(.. min(t. u¸(.))). Then
(1. u¸) is an NDRpresentation for X
0
Ç 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
q
X
]
C
(
T
If q has the HLPfor 7, then so also has ¸. If ¸ is aﬁbration, thenq is aﬁbration.
We call q the ﬁbration induced from the ﬁbration ¸ via base change along ¡ .
In the case that ¡ : C Ç T the restriction ¸: ¸
1
(C) ÷ C can be taken as the
induced ﬁbration.
(5.5.8) Example. X
J
÷X
JJ
Š X ×X: n ÷(n(0). n(1)) is a ﬁbration (5.5.4).
The evaluation e
1
: JY ÷ Y , n ÷ n(1) is a ﬁbration (restriction to + × Y ).
Hence we have the induced ﬁbration ¡
1
: J(¡ ) ÷Y . Þ
The homotopy theorem for ﬁbrations says, among other things, that homotopic
maps induce hequivalent ﬁbrations.
Let ¸: X ÷T be a ﬁbration and ç : ¡ . g: C ÷T a homotopy. We consider
two pullback diagrams.
Y
(
T
]
¸
X
]
Y
¿
G
]
¡
X
]
C
(
T C
¿
T
There exists a homotopy ˆ
t
: Y
(
÷ X such that ˆ
0
= J and ¸ˆ
t
= ç
t
¸
(
. The
pullback property of the right square yields a map k = k
ç
: Y
(
÷ Y
¿
such that
Gk = ˆ
1
and ¸
¿
k = ¸
(
. Let [
t
: ¡ . g be homotopic to ç
t
by a homotopy
118 Chapter 5. Coﬁbrations and Fibrations
; : 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
ç
and k
)
are homotopic over C. In order to
verify this, we lift the homotopy ;ı(¸
(
×id ×id): Y
(
×1 ×1 ÷T to a homotopy I
with initial data I(,. s. 0) = ˆ(,. 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 ı¸
(
. By construction, 1 is a homotopy over C from k
(
to k
¿
. The reader should now verify the functoriality Œk
ç+)
 = Œk
)
Œk
ç
.
Let hFIB
C
be the full subcategoryof hTOP
C
withobjects the ﬁbrations over C.
(5.5.9) Proposition. Let ¸: X ÷ T be a ﬁbration. We assign to ¡ : C ÷ T the
induced ﬁbration ¸
(
: Y
(
÷ C and to the morphism Œç : ¡ ÷ g in …(C. T) the
morphism Œk
ç
. This yields a functor …(C. T) ÷hFIB
C
.
Since …(C. T) is a groupoid, Œk
ç
 is always an isomorphism in hTOP
B
. This
fact we call the homotopy theorem for ﬁbrations.
As a special case of (5.5.9) we obtain the ﬁbre transport. It generalizes the
ﬁbre transport in coverings. Let ¸: 1 ÷ T be a ﬁbration and n: 1 ÷ T a path
from b to c. We obtain a homotopy equivalence T
]
Œn : J
b
÷ J
c
which only
depends on the homotopy class Œn of n, and T
]
Œu + · = T
]
Œ·T
]
Œu. This yields
a functor T
]
: …(T) ÷hTOP. In particular the ﬁbres over points in the same path
component of T are hequivalent.
(5.5.10) Proposition. In the pullback (5.5.7) let ¸ be a ﬁbration 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 ﬁbration and coﬁbration are not homotopy invari
ant. The projection 1 × 0 L 0 × 1 ÷ 1, (.. t ) ÷ . is not a ﬁbration, but the
map is over 1 hequivalent to id. One deﬁnition of an hﬁbration ¸: 1 ÷ T is
that homotopies X × 1 ÷T which are constant on X × Œ0. c. c > 0 can be lifted
with a given initial condition; a similar deﬁnition for homotopy extensions gives
the notion on an hcoﬁbration. In [46] you can ﬁnd details about these notions.
Problems
1. A composition of ﬁbrations is a ﬁbration. A product of ﬁbrations is a ﬁbration. 0 ÷ T
is a ﬁbration.
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 ﬁbration and ¸
0
: 1
0
÷ T its restriction to T × 0 = T. Then
there exists a ﬁbrewise hequivalence 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
coﬁbration. Consider only maps C ÷T with a ﬁxed 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 coﬁbration, and by
(5.5.3) the homotopies ˆ
I
can be chosen constant on Y
1
. The maps k
µ
: Y
¸
÷Y
¡
are then
the identity on Y
1
. The homotopy class of k
µ
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 ﬁbration 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 ﬁbration, ç : Y ×1 ÷T
a homotopy and ˆ: Y × 1 ÷ 1 a lifting along ¸ with initial condition ¡ . We
deﬁne
ç
#
: Œ(Y. ç
0
). (1. ¸)
B
÷Œ(Y. ç
1
). (1. ¸)
B
. Œ¡  ÷Œˆ
1
.
One shows that this map is well deﬁned and depends only on the homotopy class
of ç relative to Y × d1 (see the analogous situation for coﬁbrations). Moreover,
(ç + [)
#
= [
#
ç
#
.
(5.6.1) Proposition. The assignments ¡ ÷ Œ¡. ¸
B
and Œç ÷ ç
#
are a functor,
called transport functor, from …(Y. T) into the category of sets. Since …(Y. T) is
a groupoid, ç
#
is always bijective.
(5.6.2) Note. Let ¸: 1 ÷ T be a ﬁbration and [: X × 1 ÷ Y be a homotopy.
Then[
+
1
= Œg[
t

#
[
+
0
; here [
+
0
: Œg. ¸ ÷Œg[
0
. ¸ is the compositionwith[
0
.
(5.6.3) Theorem. Let ¡ : X ÷ Y be an hequivalence and ¸: 1 ÷ T be a
ﬁbration. Then ¡
+
: Œ·. ¸
B
÷Œ·¡. ¸
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 ﬁbrations. Let h: X ÷ Y
be an hequivalence and a map over T. Then ¡ is an hequivalence 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 ﬁbration and a homotopy equivalence.
Then q is shrinkable.
Let ¸: 1 ÷ T be a ﬁbration. Then the canonical map r : 1
J
÷ W(¸) is
shrinkable (see (5.5.5)). Þ
120 Chapter 5. Coﬁbrations and Fibrations
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 ﬁbration (see (5.5.8)), the maps (q. ¸), q and ¸ are ﬁbrations.
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
(
x
Y
=
J(¡ )
J
i
¸
(
1
u
u
u
u
u
u
u
u
u
W(¡ )
]
q
¸
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 ﬁbration ¸. If ¡ = ¸
t
s
t
is another factorization of
this type, then there exists a ﬁbrewise 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 veriﬁes that q and ¸ are pointed ﬁbrations. Let J(¡ ) = ¸
1
(+)
and J = ¡
1
(+) be the ﬁbres over the base point, with ¸ and J the inclusions.
The map q induces r. We call J(¡ ) the homotopy ﬁbre of ¡ . We use the same
notion for the ﬁbre of any replacement of ¡ by a ﬁbration as above. If ¡ is already
a ﬁbration, then q is a ﬁbrewise homotopy equivalence (5.6.4) and r the induced
homotopy equivalence; hence the actual ﬁbre is also the homotopy ﬁbre.
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 ﬁrst fundamental theorem of algebraic topology is the Brouwer–Hopf degree
theorem. It says that the homotopy set ŒS
n
. S
n
 has for n _ 1 a homotopically
deﬁned 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
ntimes around itself. In order
to give precision to this statement, one has to count the number of preimages 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
socalled 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 fromŒS
n
. S
n
 to ŒS
n÷1
. S
n÷1
. Freudenthal’s theorem says that this process
after a while is “stable”, i.e., induces a bijection of homotopy sets. This is the
origin of the socalled stable homotopy theory – a theory which has developed into
a highly technical mathematical ﬁeld 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
n1
Ç 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 ﬁxed point theorem.
This chapter contains the fundamental nonformal results of homotopy theory.
Based on these results, one can develop algebraic topology from the viewpoint
of homotopy theory. The chapter is essentially independent of the three previous
chapters. But in the last section we refer to the deﬁnition of a coﬁbration 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 = Œ0. 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 nth homotopy group of a pointed space (X. +) is
¬
n
(X. +) = Œ(1
n
. d1
n
). (X. {+]) Š Œ1
n
,d1
n
. X
0
with the group structure deﬁned below. For n = 1 it is the fundamental group.
The deﬁnition of the set ¬
n
(X. +) also makes sense for n = 0, and it can be
identiﬁed with the set ¬
0
(X) of path components of X with Œ+ as a base point. The
composition law on ¬
n
(X. +) for n _ 1 is deﬁned as follows. Suppose Œ¡  and Œg
in ¬
n
(X. +) are given. Then Œ¡  ÷Œg is represented by ¡ ÷
i
g:
(1) (¡ ÷
i
g)(t
1
. . . . . t
n
) =
_
¡(t
1
. . . . . t
i1
. 2t
i
. . . . . t
n
) for t
i
_
1
2
.
g(t
1
. . . . . t
i1
. 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 veriﬁcation
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 deﬁne relative homotopy groups (sets) ¬
k
(X. ·. +) for a pointed pair
(X. ·). For n _ 1, let J
n
= d1
n
× 1 L 1
n
× {0] Ç d1
n÷1
Ç 1
n
× 1 and
set J
0
= {0] Ç 1. We denote by ¬
n÷1
(X. ·. +) the set of homotopy classes
of maps of triples ¡ : (1
n÷1
. d1
n÷1
. J
n
) ÷ (X. ·. +). (Recall that this means
¡(d1
n÷1
) Ç ·. ¡(J
n
) Ç {+], and for homotopies H we require H
t
for each
t ÷ 1 to be a map of triples.) Thus, with notation introduced earlier,
¬
n÷1
(X. ·. +) = Œ(1
n÷1
. d1
n÷1
. J
n
). (X. ·. +).
A group structure ÷
i
, 1 _ i _ n is deﬁned 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 ﬁrst group
as ¬
n
(X. {+]. +) and then using the map induced by the inclusion (X. {+]. +) Ç
(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
n÷1
. d1
n÷1
. 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: ¬
n÷1
(X. ·. +) ÷ ¬
n
(·. +). The boundary operator is
a homomorphism for n _ 1. For n = 0 we have dŒh = Œh(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
nk
÷C
k
(X. +)
which sends u ÷ 1
nk
to 1
k
÷ X, (t
1
. . . . . t
k
) ÷ ¡(t
1
. . . . . t
k
. u
1
. . . . . u
nk
).
This adjunction is compatible with homotopies and induces a bijection
(2) ¬
n
(X. +) Š ¬
nk
(C
k
(X. +). +).
Adjunction as above also yields a bijection
(3) ¬
n÷1
(X. ·. +) Š ¬
n÷1k
(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 dŒn = 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 deﬁned 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 ¸
+
Œn + u = Œn. Thus we have shown exactness at ¬
1
(X. ·. +).
124 Chapter 6. Homotopy Groups
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
+
Œu = i
+
Œk
+
+ u = Œk
+
+ n = Œn.
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) Ç J(X. ·), and 
1
the evaluation J(X. ·) ÷·,
n ÷ n(1). For n _ 1 we assign to ¡ : (1
n÷1
. d1
n÷1
. J
n
) ÷ (X. ·. +) the
adjoint map ¡
.
: 1
n
÷J(X. ·), deﬁned 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 Œ¡  ÷ Œ
¨
¡  is a
welldeﬁned bijection
(4) ¬
n÷1
(X. ·. +) Š ¬
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 ﬁbre sequence of . Under the
identiﬁcations (4) the boundary operator is transformed into

1
+
: Œ1
n
,d1
n
. J(X. ·)
0
÷Œ1
n
,d1
n
. ·
0
.
and ¬
n÷1
(X. +) ÷¬
n÷1
(X. ·. +) is transformed into
i()
+
: Œ1
n
,d1
n
. C(X)
0
÷Œ1
n
,d1
n
. J(X. ·. +)
0
.
Now apply T = 1
n
,d1
n
to the ﬁbre sequence (4.7.4) of  : · Ç X to see the
exactness of a typical portion of the homotopy sequence. Þ
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
n÷1
,J
n
. D(n ÷1) ÷1
n÷1
,J
n
.
the ﬁrst one . ÷ (.. 1), the second one the identity on representatives in 1
n
× 1;
moreover we have the embedding S(n) ÷D(n÷1), . ÷(.. 1) which we consider
as an inclusion. These homeomorphisms allow us to write
¬
n
(X. +) Š Œ1
n
,d1
n
. X
0
= ŒS(n). X
0
.
¬
n÷1
(X. ·. +) ŠŒ(1
n÷1
,J
n
. d1
n÷1
,J
n
). (X. ·)
0
ŠŒ(D(n÷1). S(n)). (X. ·)
0
.
and d: ¬
n÷1
(X. ·. +) ÷ ¬
n
(·. +) is induced by the restriction from D(n ÷ 1)
to S(n).
The pointed cone on S
n
is D
n÷1
: We have a homeomorphism
S
n
× 1,(S
n
× 0 L e
n÷1
× 1) ÷D
n÷1
. (.. t ) ÷(1 ÷ t )e
n÷1
÷t..
Therefore we can also represent elements in ¬
n÷1
(X. ·. +) by pointed maps
(D
n÷1
. 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. Þ
Problems
1. ¬
n
(·. ·. a) = 0. Given ¡ : (1
n
. d1
n
. J
n1
) ÷ (·. ·. a). Then a null homotopy is
¡
I
(.
1
. . . . . .
n
) = ¡(.
1
. .
2
. . . . . (1 ÷ t ).
n
).
2. Let + ÷ X
0
Ç X
1
Ç X
2
Ç 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 Ç 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
. +) Š ¬
n
(X. +).
3. Let (X. ·. T. b) be a pointed triple. Deﬁne the boundary operator d: ¬
n
(X. ·. b) ÷
¬
n1
(·. b) ÷ ¬
n1
(·. T. b) as the composition of the previously deﬁned operator with
the map induced by the inclusion. Show that the sequence
÷¬
n
(·. T. b) ÷¬
n
(X. T. b) ÷¬
n
(X. ·. b)
u
÷÷¬
n1
(·. T. b) ÷
is exact. The sequence ends with ¬
1
(X. ·. b).
4. The group structure in ¬
n¬1
(X. ·. +) is induced by an hcogroup structure on
(D(n ÷1). S(n)) in the category of pointed pairs.
126 Chapter 6. Homotopy Groups
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 deﬁne the groups ¬
n
(7(¡ ). X. +) without using the mapping cylinder. Consider
commutative diagrams with pointed maps ç and ˆ.
d1
n
,J
n1
µ
ï ¸
X
¸
1
n
,J
n1
ˆ
Y
We consider (ç. ˆ): ¸ ÷ ¡ as a morphism in the category of pointed arrows. Let
¬
n
(¡ ) denote the set of homotopy classes of such morphisms. For ¡ : X Ç Y we ob
tain the previously deﬁned ¬
n
(Y. X. +). The projection ¸: i ÷¡ induces an isomorphism
¬
n
(7(¡ ). X. +) = ¬
n
(i ) ÷¬
n
(¡ ). One can also use the ﬁbre 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
Ç 1
n
is a coﬁbration. 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 ŒV
0
 ÷ŒV
1
 is a welldeﬁned map
·
#
: ¬
n
(X. ·(0)) ÷¬
n
(X. ·(1))
which only depends on the morphism Œ· in the fundamental groupoid …(X). The
relation (·+n)
#
= n
#
ı·
#
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 ç : ¡ . 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 Ç 1
n÷2
as follows: On 1
n
× 0 × 1
6.2. The Role of the Base Point 127
we use ç, 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
n÷1
. Therefore our map has an extension to 1
n÷2
, and its restriction to 1
n
×1×1
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
n1
) ÷(X. ·. a
0
) and a path ·: 1 ÷·froma
0
to a
1
. We consider
the path as a homotopy of the constant map J
n1
÷ {a
0
]. Then we extend this
homotopy to a homotopy V
t
: (1
n
. d1
n
) ÷ (X. ·). An extension exists because
J
n1
Ç d1
n
and d1
n
Ç 1
n
are coﬁbrations.
(6.2.2) Proposition. The assignment ŒV
0
 ÷ŒV
1
 is a welldeﬁned map
·
#
: ¬
n
(X. ·. a
0
) ÷¬
n
(X. ·. a
1
)
which only depends on the morphism Œ· 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. .). (˛. ˇ) ÷˛ ˇ = ˇ
#
(˛).
¬
n
(X. ·. a) × ¬
1
(·. a) ÷¬
n
(X. ·. a). (˛. ˇ) ÷˛ ˇ = ˇ
#
(˛).
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:
128 Chapter 6. Homotopy Groups
(6.2.4) Proposition. Let ¡ : (X. ·) ÷ (Y. T) be an hequivalence. 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
¡
+
: ¬
n÷1
(X. ·. +) ÷¬
n÷1
(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 Œn ÷Œ·

ŒnŒ·. In particular, the right action of ¬
1
(X. .) on itself is
given by conjugation ˛ ˇ = ˇ
1
˛ˇ.
(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 ﬁrst 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 ŒS(n). X
0
,(~) denote the orbit set of the ¬
1
(X. +)action
on ŒS(n). X
0
. The map ŒS(n). X
0
÷ ŒS(n). X which forgets the base point
induces an injective map ·: ŒS(n). X
0
,(~) ÷ ŒS(n). X. For path connected X
the map · is bijective. The forgetful map
¬
n
(X. ·. +) = Œ(D(n). S(n ÷ 1)). (X. ·)
0
÷Œ(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 (Œn. Œ·) ÷ Œn + · = Œn Œ· deﬁnes
a left action of the group ¬
1
(X. a) on the set ¬
1
(X. ·. a). The orbits of this action are the
preimages 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 Œ· ÷ ¬
1
(X. ·. a) with ·(1) =
u ÷ ·. The isotropy group of Œ· is the image of ¬
1
(·. u) in ¬
1
(X. a) with respect to
Œn ÷ Œ· + n + ·

. Find an example ˛
0
. ˛
1
÷ ¬
1
(X. ·. a) such that ˛
0
has trivial and
˛
1
nontrivial isotropy group. It is in general impossible to deﬁne 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: ¬
q¬1
(X. ·. a) ÷¬
q
(·. a) is
for each q _ 0 and each a ÷ · a bijection.
5. Let · Ç X be an hequivalence. Then ¬
n
(X. ·. a) = 0 for n _ 1 and a ÷ ·.
6. Let X carry the structure of an hmonoid. 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 Œ÷. X
0
. If we use the model D
n
,S
n1
for the nsphere, 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 ﬁbration 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 ﬁbration. For T
0
Ç 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
. +).
130 Chapter 6. Homotopy Groups
Proof. ¸
+
surjective. Let . ÷ ¬
n
(T. T
0
. +) be represented by
h: (1
n
. d1
n
. J
n1
) ÷(T. T
0
. +).
By (3.2.4), there exists a lifting H: 1
n
÷ 1 with H(J
n1
) = {+] and ¸H = h.
We then have H(d1
n
) Ç 1
0
, and therefore H represents a preimage 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 ç
t
: (1
n
. d1
n
. J
n1
) ÷
(T. T
0
. +) such that ç
0
(u) = ¸¡
0
(u), ç
1
(u) = ¸¡
1
(u) for u ÷ 1
n
. Consider the
subspace T = 1
n
× d1 L J
n1
× 1 and deﬁne G: T ÷1 by
G(u. t ) =
_
¡
t
(u). u ÷ 1
n
. t ÷ {0. 1].
+. u ÷ J
n1
. t ÷ 1.
The set T Ç 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 = ç. We can view H as a homotopy from ¡
0
to ¡
1
.
We use the isomorphism ¬
n
(1. J. +) Š ¬
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 ﬁbration:
(6.3.2) Theorem. For aSerre ﬁbration¸: 1 ÷T withinclusioni : J = ¸
1
(b) Ç
1 and . ÷ J the sequence
÷¬
n
(J. .)
i
·
¬
n
(1. .)
]
·
¬
n
(T. b)
J
¬
n1
(J. .) ÷
is exact. The sequence ends with ¬
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
n1
× 1 ÷ T. Lift to ç: 1
n
÷ 1, constant on J
n1
. Then
dŒ¡  is represented by ç[1
n1
× 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 ﬁbration. Then ¸ is a Serre ﬁbration.
Proof. A subdivision of width ı = 1,N, N ÷ N of 1
n
consists of the cubes
1(a
1
. . . . . a
n
) =
n
}=1
1(a
}
) where 1(k) = Œk,N. (k ÷ 1),N for 0 _ k < N,
k ÷ Z. Akdimensional face of 1(a
1
. . . . . a
n
) is obtained by replacing n÷k of the
intervals 1(a
}
) by one of its boundary points. (The a
}
are integers, 0 _ a
}
< n.)
It sufﬁces 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
Ç 1
n
denote the union of the kdimensional
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
×Œ0. ı to a lifting of h. We solve the lifting problems
1
n
× 0 L V
k1
× Œ0. ı
1(k1)
í
1
]
1
n
× 0 L V
k
× Œ0. ı
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 kdimensional cube and dW the union of its (k ÷ 1)dimensional faces. We can
solve the lifting problems
W × 0 L dW × Œ0. ı
1(k1)
í
¸
1
U
]
I
W × Œ0. ı
h
1
V
k
k
k
k
k
k
k
k
U
by a map H
W
, since ¸
U
is a Serre ﬁbration; here U ÷ U was chosen such that
h(W × Œ0. ı) Ç U.
The H
W
combine to a continuous map H(k): V
k
× Œ0. ı ÷ 1 which lifts h
and extends H(k ÷ 1). We deﬁne H on the ﬁrst layer 1
n
× Œ0. ı as H(n). We
now treat 1
n
× Œı. 2ı similarly with initial condition given by H(n)[1
n
× {ı] and
continue in this manner inductively.
(6.3.4) Example. Since a product projection is a ﬁbration we obtain from (6.3.3):
A locally trivial map is a Serre ﬁbration. Þ
(6.3.5) Example. Let ¸: 1 ÷ T be a covering with typical ﬁbre 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) Š ¬
n
(T) for n _ 2. The covering R ÷S
1
then yields ¬
n
(S
1
) Š 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
(+) Ç 1 and ¬
0
(J. +) = J. It yields for
¸: R ÷ S
1
the bijection d: ¬
1
(S
1
) Š Z. A lifting of the loop s
n
: 1 ÷ S
1
,
t ÷ exp(2¬i nt ) with initial condition 0 is t ÷ nt . Hence dŒs
n
 = n. Thus we
have another method for the computation of ¬
1
(S
1
). Þ
132 Chapter 6. Homotopy Groups
(6.3.6) Example. Recall the Hopf ﬁbration ¸: S
2n÷1
÷C1
n
(14.1.9). The exact
sequence (6.3.2) and ¬
i
(S
1
) = 0 for i > 1 yield the isomorphisms
¸
+
: ¬
i
(S
2n÷1
) Š ¬
i
(C1
n
). for i _ 3:
and in particular ¬
i
(S
3
) Š ¬
i
(S
2
) for i _ 3, since C1
1
is homeomorphic to S
2
(the Riemann sphere). Þ
(6.3.7) Example. From linear algebra one knows a surjective homomorphism
SU(2) ÷SO(3) with kernel {˙1] Š Z,2. The space SU(2) is homeomorphic to
S
3
. Hence SO(3) is homeomorphic to R1
3
and ¬
1
(SO(3)) Š Z,2.
(6.3.8) Proposition. Let ¸: (1
1
. 1
0
) ÷ T be a relative Serre ﬁbration, i.e.,
¸: 1
1
÷ T is a Serre ﬁbration and the restriction of ¸ to 1
0
is also a Serre
ﬁbration. Let (J
b
1
. J
b
0
) be the pair of ﬁbres over ¸(e) = b ÷ T. Then:
(1) The inclusion induces bijections ¬
n
(J
b
1
. J
b
0
. e) Š ¬
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 ﬁrst 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
ﬁbration (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 2fold covering S
n
÷R1
n
yields for n _ 2 the isomorphism ¬
1
(R1
n
) Š 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 ﬁbration with ﬁbre J = ¸
1
(b). Then
C
n
(¸): C
n
(1. e) ÷C
n
(T. b)
6.4. The Excision Theorem 133
is a Serre ﬁbration with ﬁbre 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 (Blakers–Massey). Let Y be the union of open subspaces Y
1
and
Y
2
with nonempty 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 qconnected. Then (Y. Y
1
) is
qconnected.
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
. . . . . .
n÷1
) ÷ S
n
[ ˙.
n÷1
_ 0] Ç H
n
˙
= {. ÷ S
n
[ . ,= ÷e
n÷1
].
We use R
n
Ç R
n÷1
, (:
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: ¬
i÷1
(D
n÷1

. 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 ﬁrst case we use the exact sequence of the pair (D
n÷1

. S
n
). The
space D
n÷1

is contractible and hence ¬
i
(D
n÷1

. +) = 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.
134 Chapter 6. Homotopy Groups
For n _ 0 we have a diagram with the isomorphisms (6.4.3)
¬
i
(S
n
. +)
T
¬
i÷1
(S
n÷1
. +)
Š
¬
i÷1
(D
n÷1

. S
n
. +)
Š J
¸
t
¬
i÷1
(S
n÷1
. D
n÷1
÷
. +).
The morphism  is induced by the inclusion and 1 is deﬁned 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 _ 2n÷2 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
n÷1
. D
n÷1
÷
. D
n÷1

. S
n
). By N(n) and (6.4.3) we have
¬
i
(S
n
) Š ¬
i÷1
(D
n÷1
˙
. 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
˙ Ç H
n
˙
and
S
n1
Ç H
n
÷
¨ H
n

are hequivalences.
(6.4.5) Proposition. The homomorphism ¬
i
(D
n÷1

. S
n
. +) ÷ ¬
i
(D
n÷1

,S
n
. +)
induced by the quotient map is an isomorphism for i _ 2n÷1 and an epimorphism
for i = 2n.
Proof. Consider the commutative diagram
¬
i
(D
n÷1

. S
n
. +)
t
¬
i
(D
n÷1

,S
n
. +)
(1)
¬
i
(S
n÷1
. D
n÷1
÷
. +)
(2)
¬
i
(S
n÷1
,D
n÷1
÷
. +).
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. +) ¬
n÷1
(CX. X. +)
J
Š
¸
q
·
¬
n÷1
(CX,X. +) = ¬
n÷1
(†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)) ÷¬
i÷1
(S(n÷1)) is an isomor
phism for i _ 2n ÷ 2 and an epimorphism for i = 2n ÷ 1.
Proof. We have to show that q
+
: ¬
i÷1
(CX. X) ÷¬
i÷1
(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
n÷1

is the (pointed)
cone on S
n
.
(6.4.7) Theorem. ¬
n
(S(n)) Š Z and †
+
: ¬
n
(S(n)) ÷ ¬
n÷1
(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 ﬁbration S
1
÷ S
3
÷ S
2
and ¬
}
(S
3
) = 0 for ¸ = 1. 2 we obtain an
isomorphism d: ¬
2
(S
2
) Š ¬
1
(S
1
) Š 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)) Š Z is generated by the identity, and †
+
respects the identity.
(6.4.8) Example. We continue the discussion of the Hopf ﬁbrations (6.3.6). The
Hopf ﬁbration S
1
÷ S
2n÷1
÷ C1
n
and ¬
i
(S
2n÷1
) = 0 for i _ 2n yield
¬
2
(C1
n
) Š ¬
1
(S
1
) Š Z and ¬
i
(C1
n
) = 0 for 0 _ i _ 2n, i ,= 2. The inclu
sion S
2n÷1
÷ S
2n÷3
, : ÷ (:. 0) induces an embedding C1
n
Ç C1
n÷1
. We
compare the corresponding Hopf ﬁbrations and their exact sequences and conclude
¬
2
(C1
n
) Š ¬
2
(C1
n÷1
). Let C1
o
=
_
n_1
C1
n
be the colimit. The canonical
inclusion C1
n
Ç C1
o
induces ¬
i
(C1
n
) Š ¬
i
(C1
o
) for i _ 2n. A proof uses
the fact that a compact subset of C1
o
is contained in some ﬁnite C1
1
. Therefore
C1
o
is a space with a single nontrivial homotopy group ¬
2
(C1
o
) Š Z.
Note also the special case ¬
3
(S
2
) Š ¬
3
(S
3
) Š 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
) Š ¬
1
(R1
3
) Š Š
¬
1
(R1
o
) Š Z,2, induced by the inclusions, ¬
i
(R1
n
) Š ¬
i
(R1
n÷1
) for i < n
and ¬
i
(R1
n
) = 0 for 0 _ i < n, i ,= 1. The space R1
o
has a single nontrivial
homotopy group ¬
1
(R1
o
) Š Z,2. Þ
6.5 The Degree
Let J : ¬
n
(S(n)) ÷ Z be the isomorphism which sends Œid 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(¡
+
Œid) = J(Œ¡ ). Since the map ŒS(n). S(n)
0
÷ŒS(n). S(n)
which forgets about the base point is bijective (see (6.2.8)), we can transport J
to a bijection J : ŒS(n). S(n) ÷ Z. The functoriality ¡
+
g
+
= (¡g)
+
shows
136 Chapter 6. Homotopy Groups
J(¡g) = J(¡ )J(g); therefore J(h) = ˙1 if h is a homeomorphism. The suspen
sion sends Œ¡  to Œ¡ . 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
ŒS
n
. S
n
 ÷ŒS(n). S(n). Œ¡  ÷Œk
n
¡ k
1
n

is independent of the choice of k
n
. We use this bijection to transport J to a bijection
J : ŒS
n
. S
n
 ÷ Z. If J(Œ¡ ) = 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 deﬁne the degree J(¡ ) for any selfmap ¡ of a space
S which is homeomorphic to S(n).
Matrix multiplication l
¹
: R
n
÷R
n
, . ÷ ·. induces for each · ÷ GL
n
(R) a
pointed map 1
¹
: S
(n)
÷S
(n)
. For the notation see (6.1.4).
(6.5.2) Proposition. The degree of 1
¹
is the sign of the determinant det(·).
Proof. Let n: 1 ÷ GL
n
(R), t ÷ ·(t ) be a path. Then (.. t ) ÷ 1
¹(t)
. is a
homotopy. Hence J(1
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 sufﬁces to show that for some · with det(·) = ÷1 we have
J(1
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 ﬁrst coordinate has degree ÷1.
(6.5.3) Proposition. Let · ÷ O(n ÷1). Then z
¹
: S
n
÷S
n
, . ÷·. has degree
det(·).
Proof. Again it sufﬁces 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)
n÷1
.
A vector ﬁeld on S
n
is a continuous map J : S
n
÷ R
n÷1
such that for each
. ÷ S
n
the vector J(.) is orthogonal to .. For the maximal number of linearly
independent vector ﬁelds see [3].
6.6. The Brouwer Fixed Point Theorem 137
(6.5.5) Theorem. There exists a vector ﬁeld 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
. . . . . .
2k1
. .
2k
) ÷(.
2
. ÷.
1
. . . . . .
2k
. ÷.
2k1
)
is a vector ﬁeld with the desired property.
Let J be a vector ﬁeld 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 ﬁxed point theorem of Brouwer and a number of equivalent results.
As an application we discuss the problem of topological dimension.
Let us ﬁrst 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
Ç 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
(˙) Ç T
i
(˙).
with open subsets T
i
(÷) and T
i
(÷) of W · T
i
. The ndimensional 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 ﬁxed point (Brouwer Fixed Point
Theorem).
(2) There does not exist a continuous map r : D
n
÷ S
n1
which is the identity
on S
n1
(Retraction Theorem).
(3) The identity of S
n1
is not null homotopic (Homotopy Theorem).
(4) Let ¡ : D
n
÷ R
n
be a continuous map such that ¡(:) = : for : ÷ S
n1
.
Then D
n
is contained in the image of ¡ .
(5) Let g: D
n
÷R
n
be continuous. Then there exists a ﬁxed point or there exists
: ÷ S
n1
such that g(:) = z: with z > 1.
138 Chapter 6. Homotopy Groups
(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 nonempty.
(8) Let T
0
. . . . . T
n
be a closed covering of ^
n
such that e
i
… T
i
and d
i
^
n
Ç 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 ﬁxed point theoremexpresses a topological property of D
n
. If h: X ÷D
n
is a homeomorphism and ¡ : X ÷ X a selfmap, then h¡ h
1
has a ﬁxed point :
and therefore ¡ has the ﬁxed 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
n1
). Statement (3) is also
equivalent to the inclusion S
n1
Ç R
n
· {0] not being null homotopic (similarly
for dW
n
in place of S
n1
).
Proof. (1) = (2). Suppose r is a retraction. Then . ÷ ÷r(.) is a map without
ﬁxed point.
(2) = (3). The map r : D
n
÷ S
n1
which corresponds by (2.3.4) to a null
homotopy of the identity is a retraction.
(3) =(1). Suppose b has no ﬁxed point. Then
S
n1
× 1 ÷S
n1
. (.. t ) ÷
. ÷ t b(.)
[. ÷ t b(.)[
= N(. ÷ t b(.))
is a homotopy from the identity to the map ¡ : . ÷ N(. ÷ b(.)). Since b
has no ﬁxed point, the formula for ¡ deﬁnes 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
n1
).
(2) =(4). If . is contained in the interior of D
n
, then there exists a retraction
r : R
n
· . ÷ S
n1
of S
n1
Ç R
n
· .. If . is not contained in the image of ¡ ,
then r ı ¡ : D
n
÷S
n1
contradicts the retraction theorem.
(4) =(5). Deﬁne 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 welldeﬁned
continuous map.
For [.[ = 1 we have ¡(.) = .. By (4), there exists , with ¡(,) = 0. If
[,[ _
1
2
, then (i) shows that 2, is a ﬁxed 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. Deﬁne ·
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 deﬁne 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 ÷ Œ0. 1.
The identity of d^
n
is therefore homotopic to ˇ = ˛[d^
n
. Since ˇ 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 pointset topology there
exist closed sets C
i
Ç 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
ﬁrst part of the proof now shows that the C
i
have nonempty 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
Ç 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 nonempty 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 nonempty
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 Sperner’s Lemma [173].
140 Chapter 6. Homotopy Groups
The retraction theorem does not hold for inﬁnitedimensional spaces. In [70,
Chapter 19] you can ﬁnd a proof that the unit disk of an inﬁnitedimensional 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 ﬁnite 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 m÷1. The compact metric space X has covering dimension dimX = k, if there
exists for each c > 0 a ﬁnite closed ccovering of X of order k ÷ 1 and k ÷ N
0
is minimal with this property. Thus X is zerodimensional in this sense, if there
exists for each c > 0 a ﬁnite 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 kdimensional 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 cballs U
t
(,) = {. [ J(.. ,) < c]. (We use J for the metrics.) Let
ı be a Lebesgue number of the covering (h
1
(U) [ U ÷ U). Since dimX = k,
there exists a ﬁnite closed ıcovering C of X of order k ÷ 1. The ﬁnite 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 ccovering. Thus
we have shown dimY _ k.
We nowshowthat dimY _ k, i.e., there exists ı > 0 such that each ﬁnite closed
ıcovering 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 ı > 0 such
that J(,
1
. ,
2
) < ı implies J(g(,
1
). g(,
2
)) < c. So if C is a ıcovering of Y , then
D = (g(C) [ C ÷ C) is an ccovering of X. Since D has order at least k ÷1, so
has C.
(6.6.3) Proposition. There exists c > 0 such that each ﬁnite closed ccovering
(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
}
Ç 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
}
Ç U
i
; thus e
k
÷ T
}
implies T
}
Ç 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 ﬁnite. 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 ﬁnd 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
ﬁnite closed ccoverings 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 ﬁgure indicates the construction of closed ccoverings 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 ﬁnite 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 Œ.. 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
n÷1
.
(3) Each map d1
n÷1
÷X has an extension to 1
n÷1
.
142 Chapter 6. Homotopy Groups
Proof. The case n = 0 is trivial. The equivalence of (2) and (3) is a consequence
of the homeomorphism (D
n÷1
. S
n
) Š (1
n÷1
. d1
n÷1
). 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
n÷1
, (.. 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
n÷1
, then (J. ¡ ) represents ˇ ÷ ¬
n÷1
(X. X. .) = 0 with
dˇ = ˛.
(6.7.2) Proposition. Let n _ 0. Let ¡ : (D
n
. S
n1
) ÷ (X. ·) be homotopic as
a map of pairs to a map k : (D
n
. S
n1
) ÷ (·. ·). Then ¡ is relative to S
n1
homotopic to a map g such that g(D
n
) Ç ·.
Proof. The case n = 0 is trivial. Let G
t
: (D
n
. S
n1
) ÷ (X. ·) be a homotopy
from ¡ to k according to the assumption. Deﬁne z: D
n
× 1 ÷ D
n
× 1 by
z(.. t ) = (2˛(.. t )
1
.. 2÷˛(.. t )) with the function ˛(.. t ) = max(2[.[. 2÷t ).
Then H = Gız 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
n1
is contractible,
there exists a homotopy of the restriction ¡ : J
n1
÷· to a constant map. Since
J
n1
Ç d1
n
and d1
n
Ç 1
n
are coﬁbrations, ¡ is as a map of pairs homotopic to
g: (1
n
. d1
n
) ÷ (X. ·) such that g(J
n1
) = {a
0
]. Since ¬
n
(X. ·. a
0
) = 0, the
map g: (1
n
. d1
n
. J
n1
) ÷(X. ·. a
0
) is null homotopic as a map of triples.
(2) =(3). (6.7.2).
(3) = (1). Let ¡ : (1
n
. d1
n
. J
n1
) ÷ (X. ·. +) be given. By assump
tion (3) Œ¡  is contained in the image of ¬
n
(·. ·. +) ÷ ¬(X. ·. +). Now use
¬
n
(·. ·. +) = 0.
6.7. Higher Connectivity 143
We call (X. ·) ncompressible if one of the assertions in (6.7.3) holds. More
generally, we call a map ¡ : X ÷ Y ncompressible if the following holds: For
each commutative diagram
d1
n
ç
í
X
(
1
n
ˆ
Y
there exists ‰: 1
n
÷X such that ‰[d1
n
= ç and ¡ ‰ . ˆ 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 ncompressible if and only ¸¡ is ncompressible.
(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 nconnected if (1)–(3) in (6.7.5) hold. We call (X. ·)
1connected if the pair is nconnected for each n. A pair is oconnected 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 oconnected.
(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: ¬
q÷1
(CX. X. +) Š ¬
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 nconnected if (1)–(3) in (6.7.6) hold for X. Note that this is
compatible with our previous deﬁnitions for n = 0. 1.
144 Chapter 6. Homotopy Groups
Let ¡ : X ÷ Y be a map and X Ç 7(¡ ) the inclusion into the mapping
cylinder. Then ¡ is said to be nconnected if (7(¡ ). X) is nconnected. We
then also say that ¡ is an nequivalence. Thus ¡ is nconnected if and only if
¡
+
: ¬
q
(X. .) ÷ ¬
q
(Y. ¡(.)) is for each . ÷ X bijective (surjective) for q < n
(q = n). If ¡ is an oequivalence 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 ﬁbration. Let
J
b
}
denote the ﬁbre of ¸
}
over b. Then the following are equivalent:
(1) (1
1
. 1
0
) is nconnected.
(2) (J
b
1
. J
b
0
) is nconnected for each b ÷ T.
Proof. This is a direct consequence of (6.3.8).
The compression properties of an nconnected 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 cubecomplex 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 kskeleton 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 nconnected. Suppose (C. ·) is a pair of
cubecomplexes of dimension at most n. Then to each commutative diagram
·
ç
í
X
(
C
ˆ
Y
there exists ‰: C ÷X such that ‰[· = ç and ¡ ‰ . ˆ relative to ·.
Proof. Induction over the number of cubes. Let · Ç T Ç C such that C is
obtained from T by adding a cube W of highest dimension. Then dW Ç T.
By induction there exists ‰
t
: T ÷ X such that ‰
t
[· = ç and a homotopy
H: ¡ ‰
t
. ˆ[T relative to ·. Extend H to a homotopy of ˆ. The end ˆ
1
of this
homotopy satisﬁes ˆ
1
[T = ¡ ‰
t
. We now use that ¡ is nconnected and extend
‰
t
over W to ‰: C ÷ X such that ¡ ‰ . ˆ
1
relative to T. Altogether we have
¡ ‰ .
B
ˆ
1
.
¹
ˆ and ˆ[· = ‰
t
[· = ç.
(6.7.9) Theorem. Let ç : (X. X
0
. X
1
) ÷ (Y. Y
0
. Y
1
) be a map such that the re
strictions ç
i
: X
i
÷Y
i
are nconnected and ç
01
: X
0
¨X
1
÷Y
0
¨Y
1
is (n ÷1)
connected. Suppose X = X
ı
0
LX
ı
1
and Y = Y
ı
0
LY
ı
1
. Then ç is an nequivalence.
6.7. Higher Connectivity 145
Proof. We use mapping cylinders to reduce to the case of inclusions ç : X Ç
Y. ç
i
: X
i
Ç 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
ı
i
) L ¡
1
(X · X
ı
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) Ç Y
ı
i
. ¡(W ¨ d1
n
) Ç X
ı
i
.
Then 1
i
is a cubical subcomplex and
1
n
= 1
0
L 1
1
. J(1
i
) Ç Y
ı
i
. ¡(1
i
¨ d1
n
) Ç X
ı
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 (n÷1)connected there exists a homotopy relative to d1
n
¨1
01
from J
01
to a map g
01
: 1

01
÷X
01
. Deﬁne 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
coﬁbration, 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
)
¿
0
¸
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 nconnected.
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
1
¸
h
1
¸N
N
N
N
N
N
N
146 Chapter 6. Homotopy Groups
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
ndimensional cube, say with W Ç 1
0
. Then dW Ç 1

0
and h(dW) = h
0
(dW) Ç
X
0
. Since (Y
0
. X
0
) is nconnected, we can deform the map relative to dW to a map
into X
0
.
(6.7.10) Corollary. Let ¡ : X ÷Y be an nconnected map between wellpointed
spaces. Then †¡ : †X ÷ †Y is (n ÷ 1)connected. If X is nconnected, then
†X is (n ÷1)connected. The sphere S
k÷1
is kconnected.
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 ×Œ0. 1Œ,X ×0 and C
1
= X×0. 1,X ×1
with intersection X×0. 1Œ. We can apply (6.7.7) directly; the cones are contractible
and therefore the induced maps C
}
(X) ÷ C
}
(Y ) oconnected. In the case of a
wellpointed space X the quotient map †
t
X ÷†X is an hequivalence.
(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
}
) Ç V
}
. Suppose that
for each ﬁnite 1 Ç 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 sufﬁces to work with ﬁnite 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) Š Z. Show that S has a universal
covering.
6.8 Classical Groups
We use exact sequences of Serre ﬁbrations 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) ﬁelds 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) ﬁbrations 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(n÷1)1
.
SO(n. F)
}
÷SO(n ÷1. F) ÷S
d(n÷1)1
.
The inclusions ¸ of the groups arise from · ÷
_
¹ 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 ﬁbrations 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 _ n÷2 (R),
i _ 2n ÷ 1 (C), and i _ 4n ÷1 (H).
We turn our attention to Stiefel manifolds of orthonormal kframes in F
n
:
V
k
(R
n
) Š O(n),O(n ÷ k) Š SO(n),SO(n ÷ k).
V
k
(C
n
) Š U(n),U(n ÷ k) Š SU(n),SU(n ÷ k).
V
k
(H
n
) Š Sp(n),Sp(n ÷ k).
We have the corresponding (Serre) ﬁbrations of the type H ÷ G ÷ G,H for
these homogeneous spaces. We use (6.8.1) in the exact homotopy sequences of
these ﬁbrations and obtain:
(6.8.2) Proposition. ¬
i
(V
k
(F
n
)) = 0 for i _ J(n ÷ k ÷1) ÷ 2.
We have the ﬁbration
¸: V
k÷1
(F
n÷1
) ÷V
1
(F
n÷1
). (·
1
. . . . . ·
k÷1
) ÷·
k÷1
.
The ﬁbre over e
k÷1
is homeomorphic to V
k
(F
n
); with  : · ÷ (·. 0) we obtain a
homeomorphism ¸ : (·
1
. . . . . ·
k
) ÷(·
1
. . . . . ·
k
. e
k÷1
) onto this ﬁbre. From the
homotopy sequence of this ﬁbration we obtain
(6.8.3) Proposition. ¸
+
: ¬
i
(V
k
(F
n
)) ÷ ¬
i
(V
k÷1
(F
n÷1
)) is an isomorphism for
i _ J(n ÷1) ÷ 3.
148 Chapter 6. Homotopy Groups
We use V
1
(F
n
) = S
dn1
and ¬
t
(S
t
) Š Zand obtain from(6.8.3) by induction
(6.8.4) Proposition. ¬
2(nk)÷1
(V
k
(C
n
)) Š Z, ¬
4(nk)÷3
(V
k
(H
n
)) Š Z.
The real case is more complicated. The result is
(6.8.5) Proposition.
¬
nk
(V
k
(R
n
)) Š
_
Z. k = 1. or n ÷ k even .
Z,2. k _ 2. n ÷ k odd.
Proof. By (6.8.3) and induction it sufﬁces 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) Š SO(2) Š S
1
.
Spin(3) Š SU(2) Š Sp(1) Š S
3
.
Z,2 ÷SU(2) ÷SO(3). a 2fold covering,
SU(n) ÷U(n) ÷S
1
. a ﬁbration.
Use these data in order to verify
¬
1
(SO(2)) Š ¬
1
(O(2)) Š Z.
¬
1
(SO(3)) Š ¬
1
(SO(n)) Š Z,2. m _ 3.
¬
1
(U(1)) Š ¬
1
(U(n)) Š Z. n _ 1.
¬
1
(SU(n)) Š ¬
1
(Sp(n)) Š 0. n _ 1.
¬
2
(SU(n)) Š ¬
2
(U(n)) Š ¬
2
(Sp(n)) Š 0. n _ 1.
¬
2
(SO(n)) Š 0. n _ 3.
¬
3
(U(2)) Š ¬
3
(U(k)) Š Z. k _ 2.
¬
3
(SU(2)) Š ¬
3
(SU(k)) Š Z. k _ 2.
¬
3
(Sp(1)) Š ¬
3
(Sp(k)) Š Z. k _ 1.
¬
3
(SO(3)) Š 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. ı. 1) = {. ÷ R
n
[ a
i
_ .
i
_ a
i
÷ı for i ÷ 1. a
i
= .
i
for i … 1]
for a = (a
1
. . . . . a
n
) ÷ R
n
, ı > 0, 1 Ç {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
}
÷ı for ¸ ÷ 1
1
]
for some 1
0
Ç 1. 1
1
Ç 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. ı. 1):
1
]
(W) =
˚
. ÷ W [ .
i
< a
i
÷
ı
2
for at least ¸ values i ÷ 1
_
.
G
]
(W) =
˚
. ÷ W [ .
i
> a
i
÷
ı
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 · Ç Y be given. Suppose that for ¸ _
dimW the inclusions
¡
1
(·) ¨ W
t
Ç 1
]
(W
t
) for all W
t
Ç dW
hold. Then there exists a map g which is homotopic to ¡ relative to dW such that
g
1
(·) Ç 1
]
(W). (Similarly for G
]
in place of 1
]
.)
Proof. We can assume that W = 1
n
, n _ 1. We deﬁne 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(,) afﬁnely 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
) Ç 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 deﬁnition 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
150 Chapter 6. Homotopy Groups
of open subspaces Y
1
. Y
2
with nonempty 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) Ç Y
1
or ¡(W) Ç 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 qconnected
(¸. q _ 0). Then there exists a homotopy ¡
t
of ¡ with the following properties:
(1) If ¡(W) Ç Y
}
, then ¡
t
(W) Ç Y
}
.
(2) If ¡(W) Ç Y
0
, then ¡
t
is constant on W.
(3) If ¡(W) Ç Y
1
, then ¡
1
1
(Y
1
· Y
0
) ¨ W Ç 1
]÷1
(W).
(4) If ¡(W) Ç Y
2
, then ¡
1
1
(Y
2
· Y
0
) ¨ W Ç G
q÷1
(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) Ç Y
0
we use condition (2). If ¡(W) Ç Y
1
. ¡(W) ,Ç
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) Ç Y
2
. ¡(W) ,Ç 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 Ç W is a coﬁbration, 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) Ç Y
0
, we can use (2) for our homotopy. Let ¡(W) Ç Y
1
. ¡(W) ,Ç Y
2
. If
dimW _ ¸, there exists a homotopy ¡
W
t
: W ÷ Y
1
relative to dW of ¡ [W with
¡
W
1
(W) Ç Y
0
, since (Y
1
. Y
0
) is ¸connected. If dimW > ¸ we use (6.9.1) in order
to ﬁnd a suitable homotopy of ¡ [W. We treat the case ¡(W) Ç Y
2
. ¡(W) ,Ç Y
1
in
a similar manner. Again we extend the homotopy to the higher dimensional cubes.
This ﬁnishes 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
) Ç J(Y
1
. Y. Y
2
) is (¸ ÷q ÷ 1)connected.
Proof. Let a map ç : (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 ˆ: 1
n
× 1 ÷ Y with the
following properties:
(1) ˆ(.. 0) ÷ Y
1
for . ÷ 1
n
,
(2) ˆ(X. 1) ÷ Y
2
for . ÷ 1
n
,
6.9. Proof of the Excision Theorem 151
(3) ˆ(,. 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 ˆ can be deformed as an admissible map into a map with
image in Y
1
. We apply the preparation theorem to ˆ 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 ˆ
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) Ç ·. 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 ﬁbration J(Y. Y. Y
2
) ÷ Y , n ÷ n(0). The pullback
along Y
1
Ç Y yields the ﬁbration J(Y
1
. Y. Y
2
) ÷ Y
1
, n ÷ n(0). The ﬁbre over
+ is J(+. Y. Y
2
). We obtain a commutative diagram of ﬁbrations:
J(+. Y
1
. Y
0
)
ˇ
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 ˇ has the same
connectivity (see (6.7.8)), i.e., the inclusion (Y
1
. Y
0
) Ç (Y. Y
2
),
¬
n
(J(+. Y
1
. Y
0
))
Š
¬
n
(J(+. Y. Y
2
))
Š
¬
n÷1
(Y
1
. Y
0
. +)
¬
n÷1
(Y. Y
2
. +)
induces an isomorphism for n < ¸÷q ÷1 and an epimorphism for n = ¸÷q ÷1.
152 Chapter 6. Homotopy Groups
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 0connected. 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
) Š ¬
·
(Y
1
. Y
0
) and ¬
·
(Y. Y
2
) Š ¬
·
(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 hequivalence.
3. The map a
1
: J(Y. Y. Y
1
) ÷Y , n ÷n(0) replaces the inclusion Y
1
÷Y by a ﬁbration.
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 coﬁbration ¸ ,
·
(
}
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 hequivalence such that (6.4.1) can be applied.
Let 7(¡ ) = TL
(
·×Œ0. 1 = T÷·×Œ0. 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
·
k
}
7(¡ )
]
1
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 sufﬁces to
analyze (1. k)
+
. The space 7 can be constructed as
7 = T L
(
· × Œ0. 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 ﬁrst 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. 1Œ)
¸
¬
n
(7. T L · × Œ0. 1Œ).
¸
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) Š ¬
i
(X. ·) and
¬
i
(T L · × Œ0. 1Œ. ·×0. 1Œ) Š ¬
i
(7(¡ ). ·).
again by homotopy invariance.
(6.10.2) Theorem (Quotient Theorem). Let · Ç X be a coﬁbration. 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) Š ¬
i1
(·. a), so that the ﬁrst
hypothesis is a propertyof ·. The inclusionC· Ç XLC·is aninducedcoﬁbration.
Since C· is contractible, the projection ¸: X LC· ÷X LC·,C· Š X,· is a
homotopy equivalence.
(6.10.3) Corollary. Let · Ç X be a coﬁbration. Assume that ¬
i
(·) = 0 for
0 _ i _ m÷ 1 and ¬
i
(X) = 0 for 0 _ i _ m _ 2. Then ¬
i
(X. ·) ÷ ¬
i
(X,·)
154 Chapter 6. Homotopy Groups
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
¬
2n1
(·) ÷¬
2n1
(X) ÷¬
2n1
(X,·) ÷¬
2n2
(·)
÷ ÷¬
n÷1
(X) ÷¬
n÷1
(X,·) ÷¬
n
(·) ÷0.
Asimilar exact sequence exists for an arbitrary pointed map ¡ : ·X where a typical
portion comes from the coﬁbre 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) ÷¬
n÷1
(†X).
(6.10.4) Theorem. Let X be a wellpointed 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 = X×1,(X×1L{+]×1) be the cone on X. We have an embedding
i : X ÷CX, . ÷Œ.. 0 which we consider as an inclusion. The quotient CX,X
can be identiﬁed with †X. From the assumption that {+] Ç X is a coﬁbration one
concludes that i is a coﬁbration (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
†
+
= ¸
+
ı 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 wellpointed 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 ﬁrst 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 ) Š ¬
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 ÷¬
i÷1
(X × Y. X Y ) ÷¬
i
(X Y ) ÷¬
i
(X × Y ) ÷0.
6.10. Further Applications of Excision 155
In the case that i _ 2, the sequence splits, since we are then working with abelian
groups; hence
¬
i
(X Y ) Š ¬
i
(X) ˚¬
i
(Y ) ˚¬
i÷1
(X × Y. X Y ). i _ 2.
Since the spaces are wellpointed, we can apply the theoremof Seifert–van 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 wellpointed, 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
(X×Y. XY ) = 0 for i _ ¸÷q÷1.
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 coﬁbration. This
is a consequence of the product theorem for coﬁbrations.
(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
Š
_
}÷J
Y
÷
}
. S
n
, and the assertion follows for
ﬁnite 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 sufﬁces to ﬁnd a 1connected 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).
156 Chapter 6. Homotopy Groups
(6.10.7) Proposition. Let Y be kconnected (k _ 0) and 7 be lconnected
(l _ ÷1) and wellpointed. Then the natural maps
¬
}
(7) ÷¬
}
(Y × 7. Y × +) ÷¬
}
(Y × 7,Y × +) ÷¬
}
(7)
are isomorphisms for 0 < ¸ _ k ÷l ÷1.
Proof. The ﬁrst 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) Š
¬
}
(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 deﬁnition 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
Y
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 )
¬
i1
(X × CY )
¬
i
(X × Y )
¸
¬
i
(CX × Y )
¸
¬
i
(CX × Y. X × Y )
¸
¬
i1
(X × Y )
¸
(a morphism between exact homotopy sequences).
6.10. Further Applications of Excision 157
The excision theorem in the formulation of (6.9.4) has a dual. Suppose given a
pullback diagram
1
T
G
X
(
Y
¿
T
with ﬁbrations ¡ and g. The double mapping cylinder 7(J. G) can be considered
as the ﬁbrewise join of ¡ and g. It has a canonical map j: 7(J. G) ÷T.
(6.10.10) Proposition. Suppose ¡ is ¸connected and g is qconnected. Then j is
¸ ÷q ÷1connected.
Proof. Use ﬁbre 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 0connected and wellpointed. 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
) Š ¬
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 Z˚Z.
6. Show that the diagram
X × d1 L {+] × 1
)
ï
X
i
X × 1
CX
with ¸(.. 0) = .. ¸(.. 1) = +. ¸(+. t ) = + is a pushout.
7. If X is wellpointed, then †X is wellpointed.
8. Some hypothesis like e.g. wellpointed is necessary in both (6.10.1) and (6.10.4). Let
· = {0] L {n
1
[ n ÷ Z] and · = · × 0 Ç X = · × 1,· × 1 with base point (0. 0).
Then ¬
1
(†X) and ¬
1
(·) are uncountable; †
·
: ¬
0
(X) ÷¬
1
(†X) is not surjective. Note:
· Ç X is a coﬁbration and X,· is wellpointed.
9. Let e
1
. . . . . e
n¬1
be the standard basis of unit vectors in R
n¬1
, and let e
1
be the base
point of S
n
. A pointed homeomorphism h
n
: †S
n
Š S
n¬1
is
h
n
: †S
n
÷S
n¬1
. (.. t ) ÷
1
2
(e
1
÷.) ÷
1
2
cos 2¬t (e
1
÷.) ÷
1
2
[e
1
÷ .[ sin 2¬t e
n¬2
where R
n¬1
= R
n¬1
× 0 Ç R
n¬2
.
10. Let 1 Ç R
n¬1
be compact. Show that each map ¡ : 1 ÷ S
n
has an extension to the
complement R
n¬1
· 1 of a ﬁnite set 1. One can choose 1 such that each component of
R
n¬1
· 1 contains at most one point of 1.
158 Chapter 6. Homotopy Groups
11. Determine ¬
2n1
(S
n
S
n
) for n _ 2.
12. Let ¡
¸
be a selfmap 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(˛ ı ¡ ) = J(¡ )H(˛) and H(g ı ˛) = J(g)
2
H(˛) hold.
Chapter 7
Stable Homotopy. Duality
The suspension theorem of Freudenthal indicates that homotopy theory simpliﬁes
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 Ç 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 Spanier–Whitehead
duality.
We introduced a naive form of spectra and us them to deﬁne 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 Š Y Š 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 sufﬁce.
(7.1.3) Theorem. Let X Ç R
n
and Y Ç 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
†
n÷1
(R
n
· X) . †
n÷1
(R
n
· Y ) with respect to arbitrary base points.
(2) If R
n
= X and R
n
,= Y , then n < mand †
n÷1
(R
n
·Y ) . S
n
, i.e., R
n
·Y
has the stable homotopy type of S
nn1
.
(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 identiﬁed with X. The subcategory of the objects
(X. 0) = X with morphisms the socalled stable homotopy classes is the geo
metric input. For positive n the pair (X. n) replaces the nfold 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 †
n÷1
(R
n
·X) ÷†
n÷1
(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 onepoint compactiﬁcation 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
Š S
o÷b
which we usually treat as identity. Suppose n. m. k ÷ Z are
integers such that n ÷k _ 0. m÷k _ 0. Then we have the suspension morphism
†: ŒX.S
n÷k
. Y .S
n÷k

0
÷ŒX.S
n÷k÷1
. Y .S
n÷k÷1

0
. ¡ ÷¡ .id(S
1
).
We form the colimit over these morphisms, colim
k
ŒX . S
n÷k
. Y . S
n÷k

0
. For
n÷k _ 2 the set ŒX .S
n÷k
. Y .S
n÷k

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 deﬁne as morphism group in our category ST
ST((X. n). (Y. m)) = colim
k
ŒX . S
n÷k
. Y . S
n÷k

0
.
Formation of the colimit means the following: An element of ST((X. n). (Y. m))
is represented by pointed maps ¡
k
: X . S
n÷k
÷ Y . S
n÷k
, and ¡
k
, ¡
I
. l _
k represent the same element of the colimit if †
Ik
¡
k
. ¡
I
. Composition of
morphisms is deﬁned by composition of representatives. Let ¡
k
: X . S
n÷k
÷
Y . S
n÷k
and g
I
: Y . S
n÷I
÷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
n÷i
= X . S
n÷k
. S
ik
†
¡k
(
k
Y . S
n÷k
. S
ik
=
7 . S
]÷i
= 7 . S
]÷I
. S
iI
Y . S
n÷I
. S
iI
.
†
¡l
¿
l
¸
One veriﬁes that this deﬁnition does not depend on the choice of representatives.
The group structure is compatible with the composition
ˇ ı (˛
1
÷˛
2
) = ˇ ı ˛
1
÷ˇ ı ˛
2
. (ˇ
1
÷ˇ
2
) ı ˛ = ˇ
1
ı ˛ ÷ˇ
2
ı ˛.
The category ST has formal suspension automorphisms †
]
: ST ÷ST, ¸ ÷ Z
(X. n) ÷(X. n ÷¸). ¡ ÷†
]
¡.
If ¡ : (X. n) ÷(Y. m) is represented by ¡
k
: †
n÷k
X ÷†
n÷k
Y (with n÷k _ 0,
m÷k _ 0, k _ [¸[), then †
]
¡ is represented by
(†
]
¡ )
k
= †
]
(¡
k
): †
n÷k÷]
X ÷†
n÷k÷]
Y. ¸ _ 0.
(†
]
¡ )
k÷[][
= ¡
k
: †
n÷k÷]÷[][
X ÷†
n÷k÷]÷[][
Y. ¸ _ 0.
The rules †
0
= id(ST) and †
]
ı †
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
n÷k
÷ (X . S
]
) . S
n÷k]
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
¬
n÷k
(S
n
) are the stable homotopy groups of the spheres. Þ
162 Chapter 7. Stable Homotopy. Duality
(7.1.5) Example. It is in general difﬁcult to determine morphismgroups in ST. But
we know that the category in nontrivial. The suspension theorem and the degree
theorem yield
ST(S
n
. S
n
) = colim
k
ŒS
n÷k
. S
n÷k

0
Š Z.
The composition of morphisms corresponds to multiplication of integers. Þ
(7.1.6) Proposition. Let Y be pathwise connected. We have the embeddingi : S
n
÷
Y
÷
. 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
Š Y ×S
1
,Y ×{+] is path connected. The base
point of Y
÷
is nondegenerate. Hence the quotient †
t
(Y
÷
) ÷ †(Y
÷
) from the
unreduced suspension to the reduced suspension is an hequivalence. The projection
Y ÷1 onto a point induces a 2connected map between double mapping cylinders
†
t
(Y
÷
) = 7(+ ÷Y ÷{+] ÷+) ÷7
t
(+ ÷1 ÷{+] ÷+) = †
t
(1
÷
) Š S
1
.
From this fact one deduces the assertion for n = 1.
We now consider suspensions
¬
k
(S
n
)
i
·
†
¬
k
(Y
÷
. S
n
)
†
Y
]
·
¬
k
(S
n
)
†
¬
k÷1
(S
n÷1
)
i
·
¬
k÷1
(Y
÷
. S
n÷1
)
]
·
¬
k÷1
(S
n÷1
).
The vertical morphisms are bijective (surjective) for k _ 2n÷2 (k = 2n÷1). For
n = 1 ¬
1
(Y
÷
. S
1
) Š Z. Since ¬
2
(Y
÷
. S
2
) contains ¬
2
(S
2
) Š 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 wellpointed 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
) ÷
¬
n÷1
(Y . S
n÷1
) 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
) Š ¬
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 wellpointed space. Then ST(S
0
. X) is a free abelian
group of rank [¬
0
(X)[ ÷ 1. Þ
The groupST(S
0
. X) onlydepends onthe stable homotopytype of X. Therefore
we can state:
(7.1.10) Corollary. Let X and Y be wellpointed spaces of the same stable homo
topy type. Then [¬
0
(X)[ = [¬
0
(Y )[. Therefore (7.1.2) is a consequence of (7.1.3).
Þ
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 deﬁned by
(X. m) ˝(Y. n) = (X . Y. m÷n).
Let
¡
k
: X . S
n÷k
÷X
t
. S
n
/
÷k
. g
I
: Y . S
n÷I
÷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)
k÷I
is deﬁned to be (÷1)
k(n÷n
/
)
times
the composition t
t
ı (¡
k
. g
I
) ı t (dotted arrow)
X . Y . S
n÷k÷n÷I
r
X . S
n÷k
. Y . S
n÷I
(
k
.¿
l
X
t
. Y
t
. S
n
/
÷k÷n
/
÷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 deﬁnition 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 deﬁnition. 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.
164 Chapter 7. Stable Homotopy. Duality
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 wellpointed spaces. Then Y ÷ (X × Y ),(X × {+]), , ÷ (+. ,) is a
coﬁbration.
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
 : · Ç 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 hequivalences, then C(J. ¡ ) is a pointed
hequivalence.
(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
n1
÷+).
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). Þ
(7.2.3) Example. Let X Ç 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 hequivalent to S
0
. If
X ,= R
n
, then R
n
·X is wellpointed with respect to any point and †
t
(R
n
·X) is
hequivalent to the pointed suspension †(R
n
· X). Þ
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. Deﬁne ¨ ¸: C
1
(X. ·) ÷C(X. ·) by
¨ ¸(.. t ) = .. t _ 1,2. ¨ ¸(a. t ) = (a. max(2t ÷ 1. 0)). ¨ ¸(a. 1) = +.
One veriﬁes that this assignment is welldeﬁned and continuous. A homotopy
¸ ¨ ¸ . id is given by ((.. t ). s) ÷ (.. st ÷ (1 ÷ s) max(2t ÷ 1. 0)). A similar
formula works for ¨ ¸¸ . id. Deﬁne ¨ 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 Ç · Ç X andsuppose there exists afunction
t : X ÷ 1 such that U Ç t
1
(0) and t
1
Œ0. 1Œ Ç ·. Then the inclusion of pairs
induces a pointed hequivalence 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 deﬁnition of ¡ uses the notation C
2
for the map
ping cone. The homotopies from¡g 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 deﬁne 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 hequivalences, then the induced map h(X. ·) ÷ h(Y. T) is an
isomorphism (see (7.2.1)). Moreover h satisﬁes excision: Under the hypothesis of
(7.2.5) the inclusion induces an isomorphism h(X · U. · · U) Š h(X. ·).
Conversely, let h: TOP(2) ÷C be a functor. We deﬁne 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.
166 Chapter 7. Stable Homotopy. Duality
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 C·is the cone on·andU Ç C·is the subspace witht coordinates inŒ1,2. 1.
If h is strongly homotopy invariant and satisﬁes 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 hequivalence, if the inclusion {+] ÷ X
is a coﬁbration. Thus if
¯
h is homotopy invariant, the composition 11 is naturally
isomorphic to the identity on the subcategory of wellpointed spaces. Þ
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
homotopy equivalence.
(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 deﬁned 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 identiﬁed
to a base point +.
In a ﬁrst 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 identiﬁed
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 ˇ
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 ˇ: C(X. ·) .C(Y. T) ÷C(¸) is now deﬁned by ˇ(.. ,) =
(.. ,), ˇ(a. s. ,) = (a. ,. s), ˇ(.. b. t ) = (.. b. t ), ˇ(a. s. b. t ) = (a. b. ˇ
0
(s. t )).
The diagram
X × Y
=
7
t]
¸
]
{+]
=
X × Y · × Y L X × T
t
¸
{+]
induces ¬ : C(¸) ÷ C(). It is a pointed hequivalence if ¸ is an hequivalence.
One veriﬁes that ˛ = ¬ˇ.
(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. Þ
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 nconnected and ¡ (n ÷ 1)
connected, then C(J. ¡ ) is nconnected.
3. Let X = · L T and suppose that the interiors ·
ı
. T
ı
still cover X. Then the inclusion
induces a weak homotopy equivalence C(T. · ¨ T) ÷C(X. ·).
4. Construct explicit hequivalences C
n
÷R
n
L {o] = S
(n)
such that
C
m
. C
n
˛
C
m¬n
S
(m)
. S
(n)
Š
S
(m¬n)
is homotopy commutative.
168 Chapter 7. Stable Homotopy. Duality
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 · Ç R
n
and T Ç 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 ç : R
n
÷R
n
of ¡ : · ÷T Ç R
n
. The maps
ˆ
˙
: R
n
× R
n
÷R
n
× R
n
. (.. ,) ÷(.. , ˙ç(.))
are inverse homeomorphisms. Let G(¡ ) = {(a. ¡(a)) [ a ÷ ·] denote the graph
of ¡ . Then ˆ
÷
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‰

ıtıˆ
÷
where t interchanges
R
n
and R
n
(and sends G(¡ ) to G(g)).
Let X Ç R
n
and Y Ç 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)
†
n÷1
(R
n
· X) . †
n÷1
(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 †
n÷1
(R
n
· X) . S
n
. Since n _ m the
homotopy group ¬
n
(†
n÷1
(R
n
· X)) = 0 and ¬
n
(S
n
) Š 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
. †
n÷1
(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 ﬁnishes 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 Ç 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(¡ )
n÷n
. Its construction needs some preparation.
Given the data X Ç R
n
. Y Ç R
n
and a proper map ¡ : X ÷ Y . Henceforth
we use the notation ·[T = (·. ··T) for pairs T Ç ·. 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 ndimensional standard disk. A scaling function for a
proper map ¡ : X ÷Y is a continuous function ç : Y ÷0. oŒ with the property
ç(¡(.)) _ [.[. . ÷ X.
The next lemma shows the existence of scalingfunctions withanadditional property.
(7.4.1) Lemma. There exists a positive continuous function [: Œ0. oŒ ÷0. oŒ
such that the inequality [([¡ .[) _ [.[ holds for . ÷ X. A scaling function in
the sense of the deﬁnition 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
¯
[: Œ0. oŒ ÷ Œ0. oŒ is
increasing. There exists a continuous increasing function [: Œ0. oŒ ÷0. oŒ such
that [(t ) _
¯
[(t ) for each t _ 0.
170 Chapter 7. Stable Homotopy. Duality
The set of scaling functions is a positive convex cone. Let ç
1
. ç
2
be scaling
functions and 0 _ z _ 1; then zç
1
÷(1 ÷ z)ç
2
is a scaling function. Let ¯ ç _ ç;
if ç is a scaling function then also ¯ ç.
Let ç be a scaling function and set M(ç) = {(.. ,) [ ç(,) _ [.[]. Then we
have a homeomorphism
R
n÷n
[D
n
× Y ÷R
n÷n
[M(ç). (.. ,) ÷(ç(,) .. ,).
The graph G(¡ ) = {(.. ¡ .) [ . ÷ X] of ¡ is contained in M(ç). We thus can
continue with the inclusion and obtain a map D
1
(¡. ç) of pairs
R
n÷n
[D
n
× Y ÷R
n÷n
[G(¡ ). (.. ,) ÷(ç(,) .. ,).
The homotopy class of D
1
(¡. ç) does not depend on the choice of the scaling
function: If ç
1
. ç
2
are scaling functions, then
(.. ,. t ) ÷((t ç
1
(,) ÷(1 ÷ t )ç
2
(,)) .. ,)
is a homotopy from D
1
(¡. ç
2
) to D
1
(¡. ç
1
). A continuous map ¡ : X ÷ Y has a
Tietze extension
¯
¡ : R
n
÷R
n
. The homeomorphism (.. ,) ÷ (.. , ÷
¯
¡ (.)) of
R
n÷n
sends (.. ¡(.)) to (.. 0). We obtain a homeomorphism of pairs
D
2
(¡.
¯
¡ ): R
n÷n
[G(¡ ) ÷R
n÷n
[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
(¡.
¯
¡ ) ı D
1
(¡. ç): 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 deﬁned by a representative of the colimit:
(D¡ )
n÷n
: †
n
C(R
n
[Y ) ÷†
n
C(R
n
[X).
Consider the composition
C(R
n
[Y ) . C
n
r
(1)
nm
(T( )
n¬m
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
#
¡ ; ﬁnally, we multiply
the homotopy class of the composition by (÷1)
nn
. We take the freedom to use
(D¡ )
n÷n
: 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 hequivalences of the type C
n
÷S
n
obtained in (7.2.2).
7.4. The Complement Duality Functor 171
(7.4.2) Lemma. Let ¡ be an inclusion, ¡ : X Ç Y Ç 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
ıD
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)
n·n
.
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 (.. ,) ÷(ç([,[ .. ,) and (.. ,) ÷(.÷¯ g(,). ,) are
as maps of pairs R
n
[D
n
× R
n
[Y ÷ R
n÷n
[G(¡ ) homotopic. Here [: Œ0. oŒ ÷
0. oŒ is a function such that [([¡ .[) _ [.[ and ç(r) = 1 ÷[(r).
Proof. We use the linear homotopy ((1 ÷t )(. ÷ ¯ g(,)) ÷t ç([,[) .. ,). Suppose
this element is contained in G(¡ ). Then , ÷ Y and hence g(,) = ¯ g(,), and the
ﬁrst 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 Ç R
n
. Y = h(X) Ç R
n
. In this case h and h
1
are Tietze extensions.
For the veriﬁcation 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) Ç R
n
× R
]
and the proper map h: X ÷G(g), . ÷(¡ .. g¡ .).
172 Chapter 7. Stable Homotopy. Duality
(7.4.4) Proposition. The diagram
R
n
[D
n
× R
n
[D
n
× R
]
[7
r
nm
×1
1×T
#
¿
R
n
[D
n
× R
n
[D
n
× R
]
[7
1×T
#
(¿( )
R
n
[D
n
× R
n
[Y × R
]
[0
T
#
( ×1
R
n
[D
n
× R
n
[X × R
]
[0
í
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 ç
(
is a scaling
function for ¡ and ç
¿(
a scaling function for g¡ , then
[
1
: Y ×7 ÷0. oŒ . (,. :) ÷ç
(
(,). [
2
: Y ×7 ÷0. oŒ . (,. :) ÷ç
¿(
(:)
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
n÷n÷]
[{.. 0. g¡ .]
r
nm
×1
T
2
2
(h)
R
n÷n÷]
[{0. .. g¡ .]
1×T
2
(¿( )
R
n÷n÷]
[{.. 0. 0]
r
nm
×1
R
n÷n÷]
[{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
t×T
1
(¿( )
R
n÷n÷]
[{.. 0. g¡ .]
r
nm
×1
R
n÷n÷]
[{0. .. g¡ .]
with ¸ = D
1
2
h ı D
1
h ı (1 × D
1
g) and  : R
n
[D
n
Ç R
n
[0 commutes up to homo
topy. The map ¸ is, with the choice ç
h
= ç
¿(
, the assignment
(.. ,. :) ÷(ç
¿(
(:) .. ç
¿
(:) , ÷
¯
¡ (ç
¿(
(:) .). :).
We use the linear homotopy (ç
¿(
(:) .. s(ç
¿
(:) , ÷
¯
¡ (ç
¿(
(:) .) ÷(1÷s),. :).
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) ¯ . = ç
¿(
(:) . ÷ X;
7.4. The Complement Duality Functor 173
(ii) sç
¿
(:) , ÷ s
¯
¡ (ç
¿(
(:) .) ÷(1 ÷ s), = 0;
(iii) : = g¡( ¯ .).
Since ç
¿(
(:) . ÷ X, we can replace in (ii)
¯
¡ by ¡ . We apply ç
¿(
to (iii) and
obtain
ç
¿(
(:) = ç
¿(
(g¡(ç
¿(
(:) .)) _ ç
¿(
(:) [.[.
hence [.[ _ 1. The equation (ii) for s = 0 says , = 0, hence , ÷ D
n
. Thus
assume s ,= 0. Then ¡(ç
¿(
(:) .) = (ç
¿
(:) ÷s
1
÷ 1) ,. We apply g to this
equation and use (ii):
: = g¡(ç
¿(
(:) .) = g((ç
¿
(:) ÷s
1
÷ 1) ,).
Finally we apply ç
¿
to this equation and obtain
ç
¿
(:) = ç
¿
g((ç
¿
(:) ÷s
1
÷ 1) ,) _ (ç
¿
(:) ÷s
1
÷ 1)[,[ _ ç
¿
(:)[,[.
and therefore [,[ _ 1.
Finally we show that the diagram
R
n
[D
n
× R
n÷]
[G(g)
T
#
(h)
1×T
2
¿
R
n÷n÷]
[{.. 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: (.. ,. :) ÷(ç
(
(,) .. ,. :) and
(D
#
¡ × 1) ı (1 ı D
2
g)(.. ,. :) = (ç
(
(,) .. , ÷
¯
¡ (ç
(
(,) .). : ÷ ¯ g(,)).
D
#
(h)(.. ,. :) = (ç
(
(,) .. , ÷
¯
¡ (ç
(
(,) .). : ÷ ¯ g
¯
¡ (ç
(
(,) .)).
Again we use a linear homotopy with : ÷ ¯ g((1 ÷t ), ÷t
¯
¡ (ç
(
(,) .)) 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) ç
(
(,) . ÷ X;
(ii) , =
¯
¡ (ç
(
(,) .)
(i)
= ¡(ç
(
(,) .) ÷ Y ;
(iii) : = ¯ g((1 ÷ t ), ÷t¡(ç
(
(,) .))
(ii)
= g(,).
(iii) shows that (,. :) ÷ G(g). We apply ç
(
to (ii) and see that [.[ _ 1. The three
diagrams in this proof combine to the hcommutativity 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
).
174 Chapter 7. Stable Homotopy. Duality
Proof. Let ¸
0
: X ÷ X × 1, . ÷ (.. 0). The map is the composition of the
homeomorphisma: X ÷X×0 and the inclusion b : X×0 Ç X×1. Thus Da is an
isomorphismand Db is induced by the inclusion R
n÷1
[X×1 Ç R
n÷1
[X×0; hence
Db is an isomorphism, since induced by a homotopy equivalence (use (7.4.2)).
Thus D¸
0
is an isomorphism. The composition pr ı¸
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 ı ¸
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 Ç R
n
be compact. Choose a
disk D such that X Ç D. Then the dual morphism is obtained from
1
n
[0 × 1
n
[Y ¸ R
n
[D × R
n
[Y Ç R
n÷n
[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((1÷t ),÷t
¯
¡ (.))) and (.. ,÷t
¯
¡ (.). :÷¯ g
¯
¡ (.)).
Þ
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 Ç R
n
and Y Ç 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 †
nm
(R
n
· X).
(3) If R
n
·X is empty and R
m
·Y is nonempty, then R
m
·Y has the stable homotopy
type of S
mn1
.
(4) If n = mthen the complements of X and Y have the same number of path components.
3. Let X Ç R
n
and Y Ç R
m
be closed subsets and ¡ : X ÷Y a proper map. Consider the
closed subspace W Ç 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 Ç 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 Ç R
n
be compact. Suppose [.[ _ r > 0 for . ÷ X. Then the constant function
ç(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 deﬁnition in (7.4.6) is homotopic to the map C(D
#
¡ ) of the
general deﬁnition.
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 Ç 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 nduality 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 nduality for (T. ·) = (C(R
n
[1). 1
÷
) where 1 Ç R
n
is a suitable space. In the general deﬁnition of an nduality above we now replace
S
n
by C
n
. The evaluation is deﬁned 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 deﬁnition works for arbitrary 1 Ç R
n
.
176 Chapter 7. Stable Homotopy. Duality
Let 1 Ç R
n
be compact and D Ç 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
¸
i
¸
Ç
R
n
[1
V [V × R
n
[1 V [V × R
n
[1
i×1
¸
V [1
} L
¸
Z
¸
with the diagonal ^: . ÷(.. .). We apply the mapping cone functor and (7.2.1),
(7.2.7). The maps i and ¸ induce hequivalences. 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 Ç V . Then we can
compose with r
÷
: V
÷
÷1
÷
and obtain a coevaluation
j: C
n
÷C(1. 0) . C(R
n
[1).
We call a closed subspace 1 Ç 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 annduality for the pair (1
÷
. C(R
n
[1)).
Proof. For the proof of the ﬁrst 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
ˇ
V [1 × 1[1
Z×1
}×1
¸
V [V × R
n
[0 V [V × R
n
[1 × 1[1
1×d
¸
with ˛(.. ,) = (,. .), ˇ(.. ,) = (,. . ÷ ,), and ; = (1 × J)(^× 1): (.. ,) ÷
(.. . ÷,). The homotopy (.. ,. t ) ÷(t. ÷(1 ÷t ),. . ÷,) shows that the right
square is hcommutative and the homotopy (.. ,. t ) ÷(,. . ÷ t,) shows that the
triangle is hcommutative. The axiom (1) of an nduality now follows if we write
out the morphisms according to their deﬁnition 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
ˇ
R
n
[1 × V [1
1×}
¸
;
¸k
k
k
k
k
k
k
k
k
k
k
k
k
k
(1×i×1)(1×Z)
R
n
[D × R
n
[1 R
n
[1 × 1[1 × R
n
[1
d×1
¸
7.5. Duality 177
with ˛(.. ,) = (÷,. .), ˇ(.. ,) = (. ÷ ,. .), and ;(.. ,) = (. ÷ ,. ,). The
homotopy (.. ,. t ) ÷(. ÷t, ÷(1÷t )r(,). ,) shows that the bottom triangle is h
commutative; the homotopy (.. ,. t ) ÷(.÷,. (1÷t ).÷t,) shows ; . ˇ(1×¸ );
the homotopy (.. ,. t ) ÷ (t. ÷ ,. .) shows ˛ . ˇ(1 × ¸
t
). Again we write out
the morphisms according to their deﬁnition and use this result.
Given a natural duality for objects via evaluations and coevaluations one can
deﬁne 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 Ç R
n
be compact and a retract of a neighbourhood V .
The following diagram is homotopy commutative
C(R
n
[Y ) . C
n
1.)
³
r(T(
m¬n
)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 simpliﬁed deﬁnition
(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
1×1×(
R
n
[X × R
n
[0 × X[X
(d×1)r
R
n
[D
n
× R
n
[Y × Y [Y
i×d
R
n
[0 × R
n
[0.
178 Chapter 7. Stable Homotopy. Duality
The composition downright sends the element (a. b. .) to (a. b ÷ ¡(.)) and the
composition rightdown to (ç(b) a ÷.. b ÷
¯
¡ (ç(b) a)). We use the homotopies
(ç(b)a÷.. b÷
¯
¡ (t ç(b)a÷(1÷t ).)) and then ((1÷t )a÷t (ç(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 deﬁnitions 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) Ç 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 deﬁned on V [X×R
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
deﬁned on V [X × R
n
[Y .
7.6. Homology and Cohomology for Pointed Spaces 179
(7.5.5) Remark. Let X Ç R
n
be a compact ENR and ¡ : X ÷ X a continuous
map. From the associated nduality 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 ﬁxed point index. If ¡ is the identity, then 1(id) is the Euler characteristic
of X. [51] [54] Þ
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
· Ç U Ç
¨
U Ç V Ç
¨
V Ç W.
Choose a continuous function ç : X ÷ Œ0. 1 such that ç(U) = {1] and ç(X · V ) = {0].
Let ¡ : · ÷ Œ0. 1 be continuous. Deﬁne J : X ÷ Œ0. 1 by J(.) = ç(.) ¡ 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 ndualities 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 deﬁne dualities in tensor categories.
7.6 Homology and Cohomology for Pointed Spaces
A homology theory for pointed spaces with values in the category 1MOD 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
n÷1
ı †. 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.
180 Chapter 7. Stable Homotopy. Duality
Let (X
}
[ ¸ ÷ J) be a family of wellpointed 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 wellpointed 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 wellpointed, then the inclusion and projection give
a coﬁbre sequence X ÷ X Y ÷ Y . This is used to verify that the additivity
isomorphism holds for a ﬁnite number of wellpointed spaces. The groups
¯
h
n
(S
0
)
are the coefﬁcient 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
n÷1
ı † 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 wellpointed spaces X
}
.
In Chapter 10 we deﬁne 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 deﬁned 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 deﬁne the boundary operator as the composition
d: h
n
(X. ·) =
¯
h
n
(C(i
÷
))
](i)
·
¯
h
n
(†(·
÷
)) Š
¯
h
n1
(·
÷
) = h
n1
(·).
The isomorphism is the given suspension isomorphism of the theory
¯
h
+
. The
Eilenberg–Steenrod exactness axioms holds; it is a consequence of the assumption
that
¯
h
+
transforms a coﬁbre 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
prespectrum 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
prespectra 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 Cspectrum. We deﬁne 7
n
(X) = ŒX. 7(n)
0
for a
pointed space X. Since 7(n) is up to hequivalence a double loop space, namely
7(n) . C
2
7(n÷2), 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 deﬁne
o : 7
n
(X) Š 7
n÷1
(†X) via the structure maps and adjointness as
ŒX. 7(n)
0
(t
n
)
·
ŒX. C7(n ÷1)
0
Š Œ†X. 7(n ÷1)
0
.
We thus have the data for a cohomology theory on TOP
0
. The axioms are satisﬁed
(Puppe sequence). The theory is additive.
We nowassociate a cohomologytheorytoanarbitraryspectrum7 = (7(n). e
n
).
For k _ 0 we have morphisms
b
k
n
: Œ†
k
X. 7(n)
0
†
Œ†(†
k
X). †7(n)
0
(e
n
)
·
Œ†
k÷1
X. 7(n ÷1)
0
.
Let 7
nk
(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 coﬁbre sequence.
The suspension isomorphism is obtained via the identity
Œ†
k÷1
X. 7(n ÷k ÷1)
0
Š Œ†
k
(†X). 7(n ÷k ÷1)
0
which gives in the colimit 7
n
(X) Š 7
n÷1
(†X). If the spectrumis an Cspectrum,
we get the same theory as before, since the canonical morphisms ŒX. 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 deﬁne 7(n) = S(n) and e
n
: †S(n) Š S(n ÷1) the
identity. We set o
k
(X) = colim
n
Œ†
n
X. S
n÷k

0
and call this group the kth stable
cohomotopy group of X. Þ
182 Chapter 7. Stable Homotopy. Duality
7.7.2 Suspension spectrum. Let Y be a pointed space. We deﬁne a spectrum with
spaces †
n
Y and e
n
: †(†
n
Y ) Š †
n÷1
Y . Þ
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)) Š 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
kspaces.) We write in this case
7
k
(X: Y ) = colim
n
Œ†
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
).
Þ
In general, the deﬁnition 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
onepoint compactiﬁcation 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
Š 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 deﬁned as colimit over the
maps
b : ŒS
n÷k
. X.1(n)
0
÷ŒS
n÷k
.S
1
. X.1(n).S
1

0
÷ŒS
n÷k÷1
. X.1(n÷1)
0
.
The ﬁrst 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 deﬁnition that 1
k
(÷) is a functor on TOP
0
. We need
the suspension morphisms. We ﬁrst deﬁne suspension morphisms
o
I
: 1
k
(7) ÷1
k÷1
(S
1
. 7).
They arise from the suspensions S
1
. ÷
ŒS
n÷k
. 7 . 1(n)
0
÷ŒS
n÷k÷1
. 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 Œ¡  ÷
ŒS
n÷k
. 7.1(n)
0
representing . such that 1 .¡ is null homotopic. Consider the
diagram
S
1
. S
n÷k
1.(
r
1
S
1
. 7 . 1(n)
r
2
S
n÷k
. 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) ı (¡ . 1) of . is also null homotopic. This shows that o
I
is
injective.
In order to prove surjectivity we consider the twofold suspension. Let . ÷
1
k÷2
(S
2
. 7) have the representative g: S
n÷k÷2
÷S
2
. 7 . 1(n). Then
¡ : S
n÷k÷2
¿
÷÷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 ﬁrst o
I
.
The proof of the claim is based on the next diagram with interchange maps
t. t
t
. t
tt
.
S
2
. S
n÷k÷2
1.¿
r
//
S
n÷k÷2
. 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 evendimensional sphere. The composition of the left verticals
represents o
2
I
(,), and the composition of the right verticals represents ..
184 Chapter 7. Stable Homotopy. Duality
(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 coﬁbre sequence. But
since the coﬁbre 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
n÷k
÷ 7 . 1(n) such that
(¡
1
. 1) ı h is null homotopic. The next diagram compares the coﬁbre sequences
of id: S
n÷k
÷S
n÷k
and ¡ . 1: Y . 1(n) ÷7 . 1(n).
S
n÷k
id
1
h
C(id)
](id)
1
S
n÷k
. S
1
ˇ
id
S
n÷k
. S
1
h.1
7 . 1(n)
(( .1)
1
(
1
.1
O
O
O
O
O
O
O
O
O
O
O
C(¡ . 1)
ç
](( .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 ç is the canonical homeomorphism (in the category of kspaces) which
makes the triangle commutative. Since (¡ .1)
1
ı h is null homotopic, there exists
H such that the ﬁrst square commutes. The map ˇ 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 coﬁbre sequence that the third square is
hcommutative (Problem1). The composition(1.e)(h.1) is another representative
of :, and the diagram shows that (1 . e)ˇ 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 coﬁbre sequences of two maps ¡ : · ÷ T and ¡
/
: ·
/
÷ T
/
. In the
diagram
·
¸
T
¸
1
h
C(¡ )
)(¸ )
!
†·
†¸
ˇ
†T
†h
·
/
¸
/
T
/
¸
/
1
C(¡
/
)
)(¸
/
)
†·
/
†¸
/
†T
/
7.8. Alexander Duality 185
assume given h and H such that the ﬁrst square commutes. The map ˇ 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 deﬁned by a spectrum is additive (for families of well
pointed spaces).
3. Show that a weak pointed hequivalence between wellpointed spaces induces an isomor
phism in spectral homology. The use of kspaces 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
mn
(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
n÷1
) be a spectrum. Let j: S
n
÷ T . ·,
c: · . T ÷S
n
be an nduality. The compositions
Œ· . S
t
. 1
k÷t

0
B.
ŒT . · . S
t
. T . 1
k÷t

0
)
·
ŒS
n
. S
t
. T . 1
k÷t

0
are compatible with the passage to the colimit and induce a homomorphism
D

: 1
k
(·) ÷1
nk
(T).
The compositions
ŒS
t÷nk
. T.1
t

0
.¹
Œ·.S
t÷nk
. ·.T.1
t

0
t
·
Œ·.S
n÷kt
. S
n
.1
t

0
r
·
Œ·.S
n÷kt
. 1
t
.S
n

0
e
·
Œ·.S
t÷nk
. 1
t÷n

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
nk
(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 deﬁning prop
erties of an nduality. The composition c
+
ı (· . ÷) ı j
+
ı (T . ÷) equals
t
+
ı†
n
: Œ·.S
t
. 1
k÷t

0
÷ŒS
n
.·.S
t
. S
n
.1
k÷t

0
÷Œ·.S
n
.S
t
. S
n
.1
k÷t

0
.
The deﬁnition of D

then involves the sign (÷1)
n(k÷t)
. 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 veriﬁed
similarly.
186 Chapter 7. Stable Homotopy. Duality
Let 11
+
(÷) and 1h
+
(÷) be the homology and cohomology theories on TOP(2)
constructed from the theories 1
+
(÷) and 1
+
(÷). If we use the nduality between
X
÷
and C(R
n
[X) for a compact ENR in R
n
we obtain isomorphisms
11
nk
(R
n
. R
n
· X) Š 11
k
(X). 11
nk
(X) Š 11
+
(R
n
. R
n
· X).
This is the usual appearance of Alexander duality.
In this setting one can also work with the bivariant theory 7
k
(X: Y ) =
7
k
(X: Y ). Then one obtains from an nduality an adjointness isomorphism
7
k
(· . X: Y ) Š 7
nk
(X: T . Y ).
A homology theory h
+
(÷) is deﬁned on the category ST. Here one deﬁnes
h
I
((X. n)) = h
In
(X). Let a morphism¡ ÷ ST((X. n). (Y. m)) be represented by
¡
k
: X .S
n÷k
÷Y .S
n÷k
. The induced morphism is deﬁned by commutativity
of the next diagram
h
I
((X. n))
=
h
l
(( )
h
In
(X)
†
n¬k
h
I÷k
(X . S
n÷k
)
((
k
)
·
h
I
((Y. m))
=
h
In
(Y )
†
m¬k
h
I÷k
(Y . S
n÷k
).
Given a homology theory h
+
(÷) one can deﬁne 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 deﬁne h
k
(iX) = h
nk
(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 welldeﬁned 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) Š 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., redeﬁne topological
products.
A compact Hausdorff space will be called a chspace. For the purpose of the
following investigations we also call a chspace 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
whspace, if the image of each test map is closed.
(7.9.1) Proposition. A Hausdorff space is a whspace. A whspace is a T
1
space.
A space X is a whspace if and only if each test map ¡ : 1 ÷X is proper. If X is
a whspace, then the image of each test map is a Hausdorff space. A subspace of a
whspace is a whspace. Products of whspaces are whspaces.
A subset · of a topological space (X. T ) is said to be kclosed (kopen), if
for each test map ¡ : 1 ÷ X the preimage ¡
1
(·) is closed (open) in 1. The
kopen sets in (X. T ) form a topology kT on X. A closed (open) subset is also
kclosed (kopen). Therefore kT is ﬁner 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 Ç kX is open,
then U Ç X is kopen, hence ¡
1
(U) Ç 1 is open. Therefore 
A
induces for each
chspace 1 a bijection.
TOP(1. kX)
Š
÷÷TOP(1. X). ¡ ÷
A
ı ¡.
Hence X and kX have the same kopen sets, i.e., k(kX) = kX. Atopological space
X is called kspace, if the kclosed sets are closed, i.e., if X = kX. Because of
k(kX) = kX the space kX is always a kspace. A kspace is also called compactly
generated. We let kTOP be the full subcategory of TOP with objects the kspaces.
A whkspace is a space which is a whspace and a kspace.
The next proposition explains the deﬁnition of a kspace. We call a topology S
on X chdeﬁnable, if there exists a family (¡
}
: 1
}
÷ X [ ¸ ÷ J) of test maps
such that: · Ç X is Sclosed =for each ¸ ÷ J the preimage ¡
1
}
(·) is closed in
1
}
. We can rephrase this condition: The canonical map (¡
}
):
}
1
}
÷ (X. S)
is a quotient map. A chdeﬁnable topology is ﬁner than T . We deﬁne a partial
ordering on the set of chdeﬁnable topologies by S
1
_ S
2
=S
1
¸ S
2
.
(7.9.2) Proposition. The topology kT is the maximal chdeﬁnable topology with
respect to the partial ordering.
Proof. By Zorn’s Lemma there exists a maximal chdeﬁnable topology S. If this
topology is different from kT , then there exists an Sopen set U, which is not
kopen. Hence there exists a test map t : 1 ÷X such that t
1
(U) is not open. If
188 Chapter 7. Stable Homotopy. Duality
we adjoin this test map to the deﬁning family of S, we see that S is not maximal.
(7.9.3) Corollary. The kspaces are the spaces which are quotients of a topological
sum of chspaces.
(7.9.4) Proposition. The following are equivalent:
(1) X is a kspace.
(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 Ç Y be open. In order to see that ¡
1
(U) is open it
sufﬁces to show that this set is kopen, since X is a kspace. 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) Š TOP(iY. X), ¡ ÷  ı ¡ . This
map is certainly injective. If Y is a kspace 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 whspace. Then · Ç X is kclosed if and only if
for each chspace 1 Ç X the set · ¨ 1 is closed in 1. In particular a whspace
X is a kspace if and only if: · Ç X is closed =for each chspace 1 Ç X the
intersection · ¨ 1 is closed in 1.
Proof. Let ·be kclosed. The inclusion 1 Ç X of a chspace is a test map. Hence
· ¨ 1 is closed in 1.
Conversely, suppose that · satisﬁes the stated condition and let ¡ : 1 ÷X be
a test map. Since X is a whspace, ¡(1) is a chspace and therefore ¡(1) ¨ · is
closed in ¡(1). Then ¡
1
(·) = ¡
1
(¡(1) ¨ ·) is closed in 1 = ¡
1
¡(1).
This shows: · is kclosed.
Thus we see that whspaces have an internal characterization of their kclosed
sets. For whspaces therefore k(X) can be deﬁned from internal properties of X.
If X is a whspace, so is kX.
7.9. Compactly Generated Spaces 189
(7.9.8) Theorem. X is a kspace 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 chspace.
(4) For Q Ç X and . ÷
¨
Qthere exists a chsubspace 1 Ç X such . is contained
in the closure of Q¨ 1 in 1.
(5) For each Q Ç X the following holds: Q ¨ 1 open (closed) in 1 for each
test space 1 Ç X implies Q open (closed) in X.
Proof. (1) is a special case of (2).
(2) Let Q Ç 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 Ç X and suppose a ÷
¨
Q. We choose a chneighbourhood
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 Ç 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 Q¨1 is closed in 1 for every test subspace 1 Ç X. Let
a ÷
¨
Q. By (4), there exists a test subspace 1
0
of X, such that a is contained in
the closure of Q¨1
0
in 1
0
. By the assumption (5), Q¨1
0
is closed in 1
0
; and
hence a ÷ Q¨ 1
0
Ç Q.
(5) Let ¡
1
(Q) be closed in 1 for each test map ¡ : 1 ÷ X. Then, in
particular, for each test subspace 1 Ç 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 kspace.
(7.9.9) Theorem. Let ¸: Y ÷ X be a quotient map and Y a kspace. Then X is
a kspace.
Proof. Let T Ç X be kclosed. 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 kclosed. Since Y is a
kspace, ¸
1
(T) is closed in Y .
190 Chapter 7. Stable Homotopy. Duality
(7.9.10) Proposition. Aclosed (open) subspace of a kspace is a kspace. The same
holds for whkspaces.
Proof. Let · be closed and T Ç · 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 kspace X. We write X as quotient q : 7 ÷ X of a sum
7 of chspaces (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 kspace. Therefore
the quotient U is a kspace.
The second assertion follows, if we take (7.9.1) into account.
In general, a subspace of a kspace is not a kspace (see (7.9.23)). Let X be a
kspace and i : · Ç 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 kspace 7 into k(·) is contin
uous if and only if k(i ) ı h is continuous.
Proof. If h is continuous then also is k(i )ıh. Conversely, let k(i )ıh be continuous.
We have k(i ) = i ı 
¹
. Since i is the inclusion of a subspace, 
¹
ı h is continuous;
(7.9.6) now shows that h is continuous.
(7.9.12) Theorem. The product in TOP of a kspace X with a locally compact
Hausdorff space Y is a kspace.
Proof. By (7.9.8), a locally compact Hausdorff space is a kspace. We write X as
quotient of 7, where 7 is a sum of chspaces, 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 kspace, 7×Y ,
and therefore a kspace by (7.9.9).
A product of kspaces is not always a kspace (see (7.9.23)). Therefore one is
looking for a categorical product in the category k TOP. Let (X
}
[ ¸ ÷ J) be a
family of kspaces 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. Compactly Generated Spaces 191
(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 shorthand notation, for a kspace T the canonical bijection
k TOP
_
T. k
_
X
}
__
= TOP
_
T.
X
}
_
Š
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 deﬁned. The next result shows that the whspaces are the formally hausdorff
spaces in the category k TOP.
(7.9.14) Proposition. A kspace X is a whspace if and only if the diagonal D
A
of
the product X ×
k
X is closed.
Proof. Let X be a whspace. In order to verify that D
A
is closed, we have to
show that for each test map ¡ : 1 ÷ X ×
k
X the preimage ¡
1
(D
A
) is closed.
Let ¡
}
: 1 ÷ X be the ¸ th component of ¡ . Then 1
}
= ¡
}
(1) is a chspace,
since X is a whspace. Hence 1 = 1
1
L 1
2
Ç X is a chspace. 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) Ç X is closed. Let g: 1 ÷ X be another test map. Since X is a kspace,
we have to show that g
1
¡(1) Ç 1 is closed. We use the relation
g
1
¡(1) = pr
2
((¡ × g)
1
D
A
).
Since D
A
is closed, the preimage 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 compactopen topology.
(7.9.15) Theorem. Let X and Y be kspaces, 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 ¡
.
ıt is continuous. We use ¡
.
ıt = (¡ ı(t ×id
Y
))
.
. Therefore it
sufﬁces to assume that X is a chspace. But then, by (7.9.12), X×
k
Y = X×Y 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 kspace. Then the evaluation
e
Y,7
: kJ(Y. 7) ×
k
Y ÷7. (¡. ,) ÷¡(,)
is continuous.
192 Chapter 7. Stable Homotopy. Duality
Proof. Let t : C ÷kJ(Y. 7) ×
k
Y be a test map. We have to show the continuity
of e
Y,7
ı t . Let t
.
1
: C ÷J(Y. 7) and t
2
: C ÷Y be the continuous components
of t . We show ﬁrst: 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
ıs = J(s. 7) ıt
.
1
is continuous. Moreover we have e
Y,7
ıt = t
1
ı(id. t
2
),
and the righthand 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 kspaces:
(7.9.17) Proposition. Let X and Y be kspaces. The assignments ¡ ÷ ¡
.
and
g ÷e
Y,7
ı (g ×
k
id
Y
) = g
~
are inverse bijections
TOP(X ×
k
Y. 7) Š TOP(X. kJ(Y. 7))
between these sets.
(7.9.18) Theorem. Let X, Y and 7 be kspaces. 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
ı (¡ ×
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
2×id ×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
ı (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
ı (j×id(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 kspaces, 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.
7.9. Compactly Generated Spaces 193
Proof. It sufﬁces 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 whkspace. Then
Y is a whkspace if and only if 1 = {(.
1
. .
)
[ ¡(.
1
) = ¡(.
2
)] is closed in X×
k
X.
Proof. The set 1 is the preimage 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 kspaces and assume that 7 is a whspace.
Then the mapping space kJ(Y. 7) is a whspace. In particular, if Y and 7 are
whkspaces, then kJ(Y. 7) is a whkspace.
Proof. Let ¡
.
: 1 ÷ kJ(Y. 7) be a test map. We have to show that it has
a closed image hence is kclosed. For this purpose let g
.
: 1 ÷ kJ(Y. 7) be
another test map. It remains to show that the preimage 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 whspace 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
kspaces. 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 kspaces. Let J
0
(X. Y ) Ç 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:
194 Chapter 7. Stable Homotopy. Duality
(7.9.22) Theorem. Let X, Y and 7 be pointed kspaces. 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 kspace, but the factors are kspaces (see
(7.9.4)).
(3) The product R,Z×R is a kspace (see (7.9.9) and (7.9.12)), but the subspace
R,Z× Qis not a kspace by (2).
If 1 Ç R,Z is compact, then there exists l ÷ N such that 1 Ç ¸Œ÷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
_
Ç R × Q is saturated with
respect to ¸ × id and closed in R × Q.
The set G = (¸×id)(J) is not closed in R,Z×Q. 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. cŒ ¨Q) Ç 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) Ç ¸
1
(V ). Hence there exists ı > 0 such
that m ÷ ı. m ÷ ıŒ Ç ¸
1
(V ). Now choose n ÷ N such that
1
2n
< ı and
r
n
÷
_
2 < m
t
2
. Then (¸ × id)(m ÷
1
2n
.
i
n
n
) ÷ V × (  ÷ c. cŒ ¨Q) Ç U holds,
because m÷
1
2n
÷m÷ı. m÷ıŒÇ ¸
1
(V ) and0 <
i
n
n
=
_
2
n
÷
i
n

_
2
n
<
t
2
÷
t
2
= c.
We see that U ¨ G ,= 0. This ﬁnishes the proof that : ÷
¨
G.
We nowsee that ¸×id is not a quotient map, since there exists a saturated closed
set J with nonclosed image G.
The space R,Z × Q is not a kspace. 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 kspace). The two projections pr
i
s(1) are compact and Hausdorff.
Hence there exists l ÷ N such that pr
1
s(1) Ç ¸Œ÷l. l. The inclusion
s(1) Ç pr
1
s(1) × pr
2
s(1) Ç ¸Œ÷l ÷ l × pr
2
s(1)
then shows that we have s
1
(G) = s
1
(G ¨ ¸Œ÷l. l × pr
2
s(1)). But the set
G ¨¸Œ÷l. l ×pr
2
s(1) is ﬁnite: By construction, J is a closed discrete subspace
of R × Q; moreover, J ¨ Œ÷l. l × pr
2
s(1) is ﬁnite as a closed discrete subspace
of the compact space Œ÷l. l × pr
2
s(1); therefore also
(¸ × id)(J ¨ Œ÷l. l × pr
2
s(1)) = G ¨ ¸Œ÷l. l × pr
2
s(1)
7.9. Compactly Generated Spaces 195
is ﬁnite. A ﬁnite set in a Hausdorff space is closed, and therefore s
1
(G) as pre
image of a closed set is closed itself. Þ
(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 ﬁrst to the second is not a homeomorphism. Þ
Problems
1. A space is a kspace if and only if it is a quotient of a locally compact Hausdorff space.
2. Let X
1
Ç X
2
Ç , let X
¸
be a whkspace and let X
¸
Ç X
¸¬1
be closed. Then
X =
_
¸
X
¸
, with colimit topology, is a whkspace. If the X
i
are kspaces, then X is a
kspace, being a quotient of the kspace
i
X
i
. If the X
i
are whspaces, 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
Ç X
i¬1
closed, the image is also closed in X and therefore
X is a whspace.
3. Let X and Y be kspaces. Passage to adjoint maps induces bijections of homotopy sets
ŒX ×
k
Y. 7 Š ŒX. kJ(Y. 7) and ŒX . Y. 7
0
Š ŒX. kJ
0
(Y. 7)
0
.
4. Let (X
¸
[ ¸ ÷ J) be a family of kspaces. Then the topological sum
¸÷J
X
¸
is a
kspace. The product in kTOP is compatible with sums.
5. Let a pushout of topological spaces with closed ¸ : · Ç X be given.
·
¸
¸
T
J
X
1
Y
Let X and T be whkspaces. Then Y is a whkspace.
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 ﬁnite) combinatorial data. Purely combina
torial objects are simplicial complexes. Given such a complex, one deﬁnes 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 (CWcomplexes in the sense of
J. H. C. Whitehead). They are more ﬂexible and better adapted to homotopy theory.
An ncell in a space is a subset which is homeomorphic to the standard ncell
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 ﬁnite cell complex is easily deﬁned: a Hausdorff space X which is the union
of a ﬁnite number of cells. It e is an ncell of this decomposition, then it is required
that there exists a continuous map ç : D
n
÷ X which induces a homeomorphism
1
n
÷e and sends S
n1
into the union of the cells up to dimension n ÷ 1.
From these data one obtains already an interesting invariant of X, the socalled
Euler characteristic. Let n(i ) denote the number of i cells. Deﬁne the combinatorial
Euler characteristic to be the alternating sum ¡(X) =
i_0
(÷1)
i
n(i ). It is a non
trivial fact that hequivalent ﬁnite 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 pointset topology and elementary homotopy
theory of cell complexes. Then we demonstrate the use of cell complexes in the
construction of spaces with speciﬁc properties. In particular we construct socalled
Eilenberg–Mac Lane spaces 1(¬. n). They have a single nonvanishing homotopy
group ¬
n
(1(¬. n)) Š ¬ (here ¬ can be an arbitrary abelian group). Eilenberg–Mac
Lane spaces can be used as building blocks for general homotopy types (Postnikov
systems). They also yield a homotopical deﬁnition of cohomology (and homology)
groups: The homotopy set ŒX. 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 ﬁnite nonempty subsets of 1. A set s ÷ S with q ÷ 1 elements is called a
qsimplex of 1. We require the following axioms:
(1) {e] ÷ S for each e ÷ 1.
(2) If t ÷ S and s Ç t is nonempty, then s ÷ S.
If s ÷ S is a qsimplex, then q is called the dimension of s. If t Ç s, then t is
a face of s. A 1simplex of 1 is also called an edge of 1. The 0simplices of
1 correspond to the elements of 1; a 0simplex is called a vertex. A simplex is
determined by its 0faces.
A simplicial complex is ndimensional, if it contains at least one nsimplex 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 1dimensional complex is a graph.
A complex 1 = (1. S) is ﬁnite if 1 is ﬁnite and locally ﬁnite if each vertex
is contained in a ﬁnite number of simplices. The nskeleton 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 ﬁnite nonempty set 1 Ç J let U
T
=
_
}÷T
U
}
and let
1(J) = {1 Ç J [ U
T
,= 0]. Then (J. 1(J)) is a simplicial complex, called the
nerve N(U) of the covering U. Þ
(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
ﬁnite subsets of 1. Þ
(8.1.3) Example. Let 1 = (1. S) be a simplicial complex. Deﬁne a partial order
on S by s _ t = s Ç t . The simplicial complex 1
t
associated to this ordered set
is called the barycentric subdivision of 1. Þ
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)
e÷T
˛(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[ deﬁned by
J(˛. ˇ) =
_
e÷T
(˛(e) ÷ ˇ(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 deﬁne 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 deﬁned by the canonical map
x÷S
^(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 deﬁne [s[ Ç [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) Ç [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[
n1
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 Radó [161], the triangulation of 3dimensional
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
ﬁnite 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 afﬁnely 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 afﬁnely 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 kdimensional 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
. ˛ ÷
e÷T
˛(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 ﬁnite 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 ﬁnite. (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 ﬁnite simplicial complex of dimension at most n. Then 1
has a realization as a polyhedron in R
2n¬1
. (See [44, p. 66].)
8.2 Whitehead Complexes
We use the standard subsets of Euclidean spaces S
n1
. D
n
. 1
n
= D
n
\ S
n1
,
(n _ 1). We set S
1
= 0 and let D
0
be a point, hence 1
0
= D
0
. A kdimen
sional cell (a kcell) in a space X is a subset e which is, in its subspace topology,
homeomorphic to 1
k
. A point is always a 0cell.
A Whitehead complex is a space X together with a decomposition into cells
(e
2
[ z ÷ ƒ) such that:
(W1) X is a Hausdorff space.
(W2) For each ncell e
2
there exists a characteristic map ˆ
2
: D
n
= D
n
2
÷ X
which induces a homeomorphism 1
n
÷e
2
and sends S
n1
into the union
X
n1
of the cells up to dimension n ÷ 1.
(W3) The closure ¨ e
2
of each cell e
2
intersects only a ﬁnite number of cells.
(W4) X carries the colimit topology with respect to the family (¨ e
2
[ z ÷ ƒ).
A subset · of a Whitehead complex is a subcomplex if it is a union of cells and
the closure of eachcell in·is containedin·. We will see that a subcomplextogether
with its cells is itself a Whitehead complex. From the deﬁnition of a subcomplex
200 Chapter 8. Cell Complexes
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 satisﬁes (W1)–
(W3). We see that (W4) is an important condition. Condition (W3) is also called
(C), for closure ﬁnite. Condition (W4) is called (W), for weak topology. This is the
origin for the name CWcomplex. In the next section we consider these complexes
from a different viewpoint and introduce the notion of a CWcomplex.
(8.2.1) Lemma. Let ˆ: D
n
÷ X be a continuous map into a Hausdorff space.
Let e = ˆ(1
n
). Then ˆ(D
n
) = ¨ e. In particular ¨ e is compact.
Proof. ˆ(D
n
) is a compact subset of a Hausdorff space and therefore closed. This
yields ¨ e = ˆ(1
n
) Ç ˆ(D
n
) = ˆ(D
n
) = ˆ(
¨
1
n
) Ç ˆ(1
n
) = ¨ e.
(8.2.2) Example. Suppose X has a cell decomposition into a ﬁnite number of cells
such that properties (W1) and (W2) hold. Then X is a ﬁnite union of closures ¨ e of
cells and therefore compact by (8.2.1). Properties (W3) and (W4) are satisﬁed and
X is a Whitehead complex. Þ
(8.2.3) Examples. The sphere S
n
has the structure of a Whitehead complex with
a single 0cell and a single ncell. The map
ˆ: D
n
÷S
n
. . ÷(2
_
1 ÷ [.[
2
.. 2[.[
2
÷ 1)
sends S
n1
to the 0cell e
n÷1
= (0. . . . . 0. 1) and induces a homeomorphism of
1
n
with S
n
· {e
n÷1
], hence is a characteristic map for the ncell.
Fromthis cell decomposition we obtain a cell decomposition of D
n÷1
by adding
another (n ÷1)cell 1
n÷1
with characteristic map the identity.
Another cellcomposition of S
n
has two ¸ cells for each ¸ ÷ {0. . . . . n] and
is obtained inductively from D
n
˙
= {(.
i
) ÷ S
n
[ ˙.
n÷1
_ 0] with intersection
S
n1
= S
n1
×0. A characteristic map is D
n
÷D
n
˙
, . ÷(.. ˙
_
1 ÷ [.[
2
).Þ
(8.2.4) Proposition. Let X be a Whitehead complex.
(1) A compact set 1 in X meets only a ﬁnite number of cells.
(2) A subcomplex which consists of a ﬁnite number of cells is compact and closed
in X.
(3) X(e) = X(¨ e) is for each cell e a ﬁnite subcomplex.
(4) A compact subset of a Whitehead complex is contained in a ﬁnite subcomplex.
(5) X carries the colimit topology with respect to the ﬁnite 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 Ç 7 be any subset. For each
cell ¡ of X the closure
¨
¡ is contained in the union of a ﬁnite number of cells. Thus
8.2. Whitehead Complexes 201
Y ¨
¨
¡ is a ﬁnite 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 ﬁnite.
(2) Let · = e
1
L Le
i
be a ﬁnite union of cells e
}
. Then
¨
· = ¨ e
1
L L¨ e
i
Ç ·,
by deﬁnition of a subcomplex. By (8.2.1), · =
¨
· is compact and closed.
(3) Induction over dim(e). If e is a 0cell, then e is a point and closed, hence a
subcomplex and X(e) = X(¨ e) = e. Suppose X(¡ ) is ﬁnite for each cell ¡ with
dim(¡ ) < n. Let e be an ncell with characteristic map ˆ.
The set ˆ(S
n1
) is contained in the union of cells of dimension at most n ÷1,
hence is contained in ¨ e · e.
Then ¨ e · e = ˆ(S
n1
) is compact, hence contained in a ﬁnite number of
cells e
1
. . . . . e
k
, by (1), which are contained in X
n1
, by (W2). By induction
hypothesis, the set C = e L X(e
1
) L L X(e
k
) is a ﬁnite subcomplex which
contains e and hence X(e). Therefore X(e) is ﬁnite. Since X(e) is closed, by (2),
we have ¨ e Ç X(e) and X(¨ e) Ç X(e).
(4) This is a consequence of (1) and (3).
(5) We show: · Ç X is closed if and only if for each ﬁnite subcomplex Y the
intersection · ¨ Y is closed in Y .
Suppose the condition is satisﬁed, 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 ﬁnite subcomplex, then ·¨Y is a ﬁnite 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 ˆ: D
n
÷X a characteristic map. Then ˆ(D
n
) =
¨ e Ç Y , since Y is closed. Hence ˆ can be taken as a characteristic map for Y .
It remains to verify condition (W4). Let 1 Ç 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 ﬁnite union
e
1
L L e
k
of cells. Let e
1
. . . . . e
}
be those which are contained in Y . Then
¨
¡ ¨ Y Ç e
1
L L e
}
Ç ¨ e
1
L L ¨ e
}
Ç Y
since Y is a subcomplex. Hence
¨
¡ ¨Y = (
¨
¡ ¨¨ e
1
) L L(
¨
¡ ¨¨ e
}
).
¨
¡ ¨1 =
¨
¡ ¨1¨Y =
_
}
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 ﬁnite union of sets which are closed in X.
202 Chapter 8. Cell Complexes
(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 ÷ ƒ(n)) be the family of ncells of X with characteristic maps
ˆ
2
: D
n
2
÷X
n
and restrictions ç
2
: S
n1
2
÷X
n1
. Then
2
S
n1
2
i
ç=( ç
·
)
X
n1
í
2
D
n
2
ˆ=( ˆ
·
)
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 ncell 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 Ç 7 be closed. By deﬁnition of the pushout topology
this means:
(i) V ¨ X
n1
is closed in X
n1
.
(ii) ˆ
1
(V ) ¨ D
n
2
is closed in D
n
2
, hence also compact.
We conclude that ˆ(ˆ
1
(V ) ¨D
n
2
) = V ¨ˆ(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 0cell. The
inclusions of the ﬁnite pointed subcomplexes J Ç 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 kspace and the kspace 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 ﬁnite 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 ˆ: D
n
÷X be a characteristic map for the cell e. Since ¨ e is compact
it has the same topology in k(X). Hence ˆ: D
n
÷ k(X) is continuous. Since ˆ
is a quotient map and ˆ
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 ˆ.
Let · ¨ ¨ e be closed in ¨ e for each cell e. Let 1 Ç k(X) be compact. By
hypothesis, 1 is contained in a ﬁnite number of cells, say 1 Ç e
1
L Le
k
. Then
·¨1 = ((·¨¨ e
1
) L L(·¨¨ e
k
)) ¨1 is closed in 1. Hence ·is kclosed.
8.3. CWComplexes 203
Let X and Y be Whitehead complexes and e Ç X, ¡ Ç Y cells. Then
e × ¡ Ç X × Y is a cell. From characteristic maps ˆ: D
n
÷ X, ‰: D
n
÷ Y
for e. ¡ we obtain ˆ × ‰: D
n
× D
n
÷ X × Y , and this can be considered as a
characteristic map for e × ¡ . For this purpose use a homeomorphism
(D
n÷n
. S
n÷n1
) ÷(D
n
× D
n
. D
n
× S
n1
L S
n1
× D
n
).
With this cell structure, X × Y satisﬁes conditions (W1)–(W3) in the deﬁnition
of a Whitehead complex. In general, property (W4) may not hold. In this case
one retopologizes 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 0cells {n], n ÷ Z and 1cells
n. n ÷ 1Œ, n ÷ Z. There is an analogous Whitehead complex structure W(ı) on R
n
with
0cells the set of points ı(k
1
. . . . . k
n
), k
¸
÷ Z, ı > 0 ﬁxed and the associated ıcubes.
Thus, given a compact set 1 Ç 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(ı).
In this sense, compact subsets can be approximated by ﬁnite complexes.
2. The geometric realization of a simplicial complex is a Whitehead complex.
8.3 CWComplexes
We now use (8.2.6) as a starting point for another deﬁnition of a cell complex. Let
(X. ·) be a pair of spaces. We say, X is obtained from · by attaching an ncell,
if there exists a pushout
S
n1
ç
í
·
í
D
n
ˆ
X.
Then · is closed in X and X \ · is homeomorphic to 1
n
via ˆ. We call X \ · an
ncell in X, ç its attaching map and ˆ 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) Ç 7 is closed; (2) J : X ÷ J(X) is a quotient map.
Condition (2) holds if X is compact and 7 Hausdorff.
204 Chapter 8. Cell Complexes
Proof. Let g: X ÷U and h: Y ÷U be given such that g¸ = h¡ . The diagram
is a settheoretical pushout. Therefore there exists a unique set map ç : 7 ÷ U
with çJ = g, çJ = h. Since J is a closed embedding, ç[J(Y ) is continuous.
Since J is a quotient map, ç[J(X) is continuous. Thus ç 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 ˆ: D
n
÷X which induces a homeomorphism ˆ: 1
n
÷
X \ ·. Then X is obtained from · by attaching an ncell.
Proof. We showˆ(S
n1
) Ç ·. Suppose there exists s ÷ S
n1
with ˆ(s) ÷ X\·.
Then there exists a unique t ÷ 1
n
with ˆ(s) = ˆ(t ). Let V Ç 1
n
, W Ç D
n
be disjoint open neighbourhoods of t. s. Then ˆ(V ) Ç X \ · is open in X, since
ˆ: 1
n
÷X\·is a homeomorphismand ·is closed in X. Since ˆis continuous,
there exists an open neighbourhood W
1
Ç W of s with ˆ(W
1
) Ç ˆ(V ). This
contradicts the injectivity of ˆ[1
n
.
Thus ˆ provides us with a map ç : S
n1
÷·. We now use (8.3.1).
(8.3.3) Example. The projective space R1
n
is obtained fromR1
n1
by attaching
an ncell. The projective space C1
n
is obtained from C1
n1
by attaching a 2n
cell.
We recall that C1
n1
is obtained from S
2n1
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 Œ:
1
. . . . . :
n
. A characteristic map ˆ: D
2n
÷ C1
n
is
. ÷Œ..
_
1 ÷ [.[
2
.
The space R1
n1
is obtained fromS
n1
by the relation : ~ ÷:, or fromR
n
\0
by : ~ z:, z ÷ R
+
. A characteristic map ˆ: D
n
÷ R1
n
is given by the same
formula as in the complex case. Þ
We can also attach several ncells simultaneously. We say X is obtained from·
by attaching ncells if there exists a pushout
}÷J
S
n1
}
ç
í
·
í
}÷J
D
n
}
ˆ
X.
The index ¸ just enumerates different copies of the same space. Again, · is then
closed in X and ˆ induces a homeomorphism of
}
1
n
}
with X \ ·. Therefore
X \ · is a union of components and each component is an ncell. (By invariance
of dimension, the integer n is determined by X \ ·.) We allow J = 0; in that case
· = X. We write ˆ
}
= ˆ[D
n
}
and ç
}
= ç[S
n1
and call ˆ
}
the characteristic
map of the ncell ˆ(1
n
}
) and ç
}
its attaching map.
8.3. CWComplexes 205
Let us give another interpretation: X = X(ç) is the double mapping cylinder
of J
pr
÷÷ S
n
× J
ç
÷÷ · where J is a discrete set. From this setting we see: If ç
is replaced by a homotopic map [, then X(ç) and X([) are hequivalent under ·.
Let ¡ : · ÷Y be a given map. Assume that X is obtained from·by attaching
ncells via attaching maps (ç
}
):
}
S
n1
}
÷ ·. From the pushout deﬁnition of
the attaching process we obtain:
(8.3.4) Note. There exists an extension J : X ÷ Y of ¡ if and only if the maps
¡ ç
}
are null homotopic. We view a null homotopy of ¡ ç
}
as an extension to D
n
}
.
Then the extensions J correspond to the set of null homotopies of the ¡ ç
}
.
In view of this note we call the homotopy classes Œ¡ ç
}
 the obstructions to
extending ¡ .
Let ·be a subspace of X. ACWdecompositionof (X. ·) consists of a sequence
of subspaces · = X
1
Ç X
0
Ç X
1
Ç Ç X such that:
(1) X = L
n_0
X
n
.
(2) For each n _ 0, the space X
n
is obtained from X
n1
by attaching ncells.
(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
}
Ç X
}÷1
Ç . If the
inclusions are closed, then X
k
is closed in X. This is an immediate consequence
of the deﬁnition of the colimit topology.
A pair (X. ·) together with a CWdecomposition (X
n
[ n _ ÷1) is called a
relative CWcomplex. In the case · = 0 we call X a CWcomplex. The space X
n
is the nskeleton of (X. ·) and (X
n
[ n _ ÷1) is the skeleton ﬁltration. The cells
of X
n
\ X
n1
are the ncells of (X. ·). We say, (X. ·) is ﬁnite (countable etc.) if
X \ · consists of a ﬁnite (countable etc.) number of cells. If X = X
n
, X = X
n1
we denote by n = dim(X. ·) the cellular dimension of (X. ·). If · = 0, then ·
is suppressed in the notation. We call X a CWspace if there exists some cellular
decomposition X
0
Ç X
1
Ç of X.
Let X be a Whitehead complex. From (8.2.6) we obtain a CWdecomposition
of X. The converse also holds: From a CWdecomposition 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 CWcomplexes (X. ·), the symbol X
n
usually denotes the
nskeleton and not the nfold Cartesian product.
(8.3.5) Note. If (X. ·) is a relative CWcomplex, then also (X. X
n
) and (X
n
. ·)
are relative CWcomplexes, with the obvious skeletonﬁltration 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
Ç R
n÷1
Ç deﬁnes the inﬁnite projective
206 Chapter 8. Cell Complexes
space R1
o
as a CWcomplex. It has a single ncell for each n _ 0. Similarly, we
obtain C1
o
with a single cell in each even dimension. Þ
(8.3.7) Example. The sphere S
n
has a CWdecomposition with a single 0cell
and a single ncell, and another CWcomposition 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
Ç S
n÷1
Ç , a CWcomplex with two cells in each
dimension. Þ
The general topology of adjunction spaces and colimit topologies gives us the
next results.
(8.3.8) Proposition. Let (X. ·) be a relative CWcomplex. If ·is a T
1
space, then
X is a T
1
space and a compact subset of X meets only a ﬁnite 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
n1
÷
D
n
}
. Each characteristic map ˆ
}
: D
n
}
÷ ¨ e
}
is
a quotient map, since X is Hausdorff. From the assumptions we see that X
n
¨ C
has a closed preimage in 7
n
.
The considerations so far show that a CWcomplex is a Whitehead complex.
(8.3.9) Proposition. Let (X. ·) be a relative CWcomplex. Then · Ç X is a
coﬁbration.
Proof. We know that
S
n1
}
÷
D
n
}
is a coﬁbration. Hence X
n1
Ç X
n
is
an induced coﬁbration. Therefore the compositions X
n
Ç X
n÷k
are coﬁbrations.
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
n÷1
[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 ncells yields a homeomorphism
_
¸
(D
n
,S
n1
)
¸
Š X,·.
2. Let (X. ·) and (Y. T) be relative CWcomplexes. Consider X × Y with the closed
subspaces
(X × Y )
n
=
_
n¬1
i=1
X
i
× Y
ni
. n _ ÷1.
8.4. Weak Homotopy Equivalences 207
In favorable cases, the ﬁltration ((X × Y )
n
[ n _ ÷1) is a CWdecomposition of the pair
(X × Y. · × T).
Let Y be locally compact. Then (X × Y )
n
is obtained from (X × Y )
n1
by attaching
ncells.
3. Let (X. ·) be a relative CWcomplex and let C Ç ·. Then (X,C. ·,C) is a relative
CWcomplex with CWdecomposition (X
n
,C). Moreover, X,· is a CWcomplex.
4. Let · Ç X be a subcomplex. Then X,· is a CWcomplex.
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 CWcomplex. Then ·.
k
Y is a subcomplex
of T .
k
Y .
7. Let ·be a CWcomplex. Suppose X is obtained from·by attaching ncells via attaching
maps ç :
S
n1
¸
÷·
n1
. Then X is a CWcomplex with CWdecomposition X
¸
= ·
¸
for ¸ < n and X
¸
= ·
¸
L (X · ·) for ¸ _ n, and · is a subcomplex of X.
8. Let ç
0
. ç
1
:
,
S
n1
¸
÷·be homotopic attaching maps. The spaces X(0). X(1) which
are obtained by attaching ncells with ç
0
. ç
1
are hequivalent under ·. (Homotopy theorem
for coﬁbrations.)
9. Let X be a pointed CWcomplex with base point + a 0cell. Then the cone CX and the
suspension †X are CWcomplexes. (In statements of this type the reader is asked to ﬁnd a
canonical cell decomposition induced from the initial data.)
10. Let (X
¸
[ ¸ ÷ J) be a family of pointed CWcomplexes with base point a 0cell. Then
_
¸÷J
X
¸
has the structure of a CWcomplex such that the summands are subcomplexes.
11. Let ¸: 1 ÷ T be a Serre ﬁbration and (X. ·) a CWpair. 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 ncells. Let ¸: 1 ÷ X be a covering and
1
/
= ¸
1
(·). Then 1 is obtained from 1
/
by attaching ncells.
13. Let X be a CWcomplex with nskeleton X
n
and ¸: 1 ÷ X a covering. Then 1 is a
CWcomplex with nskeleton 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 CWcomplex contains a neighbourhood V
which is pointed contractible to .. A connected CWcomplex 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 CWcomplexes. Then X ×Y is a CWcomplex in the product
topology.
8.4 Weak Homotopy Equivalences
We now study the notion of an nconnected map and of a weak homotopy equiva
lence in the context of CWcomplexes.
(8.4.1) Proposition. Let (Y. T) be nconnected. Then a map ¡ : (X. ·) ÷(Y. T)
from a relative CWcomplex (X. ·) of dimension dim(X. ·) _ n is homotopic
208 Chapter 8. Cell Complexes
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 ﬁltration. Suppose X is obtained from · by
attaching qcells via ç :
k
S
q1
k
÷·, q _ n. Consider a commutative diagram
S
q1
k
ç
·
(
i í
T
} í
D
q
k
ˆ
X
T
Y .
Since (Y. T) is nconnected, Jˆ is homotopic relative to
S
q1
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 Jˆ and Ji which coincide on
S
q1
k
. Since we have homotopies
of Jˆ relative to
S
q1
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
k÷1
relative to X
k
which sends X
k÷1
into T. Since X
k÷1
Ç X is a coﬁbration, we extend this
homotopy to X. In the case that n = o, we have to concatenate an inﬁnite number
of homotopies. We use the ﬁrst homotopy on Œ0. 1,2 the second on Œ1,2. 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 L·×1) 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 nconnected, n _ 0. Then h
+
: ŒX. T ÷
ŒX. Y  is bijective (surjective) if X is a CWcomplex with dimX < n (dimX _ n).
If h: T ÷Y is pointed, then h
+
: ŒX. T
0
÷ŒX. 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 CWcomplexes.
(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 ¡
+
: ŒX. Y  ÷ ŒX. 7 is bijective for all CWcomplexes X (see (8.4.2)). By
category theory, ¡ represents an isomorphism in hTOP: 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) ¡
+
: Œ7. Y  ÷ Œ7. 7 is surjective, since ¡ is kconnected (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 hinverse and is therefore
an hequivalence. From¡g . id we then conclude that ¡ is an hequivalence.
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 ) Š ¬
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 CWcomplex 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.Þ
(8.4.6) Example. A simply connected 1dimensional complex is contractible. A
contractible 1dimensional CWcomplex is called a tree. Þ
(8.4.7) Theorem. A connected CWcomplex X contains a maximal (with respect
to inclusion) tree as subcomplex. A tree in X is maximal if and only if it contains
each 0cell.
Proof. Let B denote the set of all trees in X, partially ordered by inclusion. Let
T Ç T be a totallyorderedsubset. ThenC =
_
T÷T
T is contractible: ¬
1
(C) = 0,
since a compact subset of C is contained in a ﬁnite subcomplex and therefore in
some T ÷ T . Thus, by Zorn’s lemma, there exist maximal trees.
Let T be a maximal tree. Consider the 1cells 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 1cells together with T form a subcomplex of X
1
,
and the remaining 1cells 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 hequivalent. Hence T
t
,T is contractible.
Since X
0
Ç T, the space T
t
,T has the form
_
S
1
and is not simply connected.
Contradiction.
210 Chapter 8. Cell Complexes
We now generalize the suspension theorem (6.10.4). Let X and Y be pointed
spaces. We have the suspension map †
+
: ŒX. Y 
0
÷ Œ†X. †Y 
0
. We use the
adjunction Œ†X. †Y 
0
Š ŒX. C†Y 
0
. The resulting map ŒX. Y 
0
÷ ŒX. C†Y 
0
is then induced by the pointed map o : Y ÷ C†Y which assigns to , ÷ Y the
loop t ÷Œ,. t  in †Y .
(8.4.8) Theorem. Suppose ¬
i
(Y ) = 0 for 0 _ i _ n. Then the suspension
†
+
: ŒX. Y 
0
÷ Œ†X. †Y 
0
is bijective (surjective) if X is a CWcomplex of di
mension dimX _ 2n (dimX _ 2n ÷1).
Proof. By the suspension theorem (6.10.4), the map o is (2n÷1)connected. Now
use the pointed version of (8.4.2).
(8.4.9) Theorem. Let X be a ﬁnite pointed CWcomplex. Then
†
+
: Œ†
k
X. †
k
Y 
0
÷Œ†
k÷1
X. †
k÷1
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 nconnected.
Proof. We know that (D
n÷1
. S
n
) is nconnected. Now apply (6.4.2).
(8.5.2) Proposition. Let X be obtained from · by attaching ncells (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 CWcomplex (X. ·) the pair (X. X
n
) is
nconnected.
Proof. From (8.5.1) we obtain by induction on k that (X
n÷k
. X
n
) is nconnected.
The compactness argument (8.3.8) ﬁnally shows (X. X
n
) to be nconnected.
Let X and Y be CWcomplexes. Amap ¡ : X ÷Y is cellular, if ¡(X
n
) Ç Y
n
for each n ÷ N
0
. The cellular approximation theorem (8.5.4) is an application of
(8.4.1).
8.6. CWApproximation 211
(8.5.4) Theorem. A map ¡ : X ÷ Y is homotopic to a cellular map g: X ÷ Y .
If T Ç 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
: X×1 ÷Y such that
(1) H
0
0
= ¡ , H
n1
1
= H
n
0
for n _ 1;
(2) H
n
1
(X
i
) Ç Y
i
for i _ n;
(3) H
n
is constant on X
n1
L T.
For the induction step we assume ¡(X
i
) Ç Y
i
for i < n. Let ˆ: (D
n
. S
n1
) ÷
(X
n
. X
n1
) be a characteristic map of an ncell not contained in T. The map ¡ ıˆ
is homotopic relative to S
n1
to a map into Y
n
, since (Y. Y
n
) is nconnected. A
corresponding homotopy is used to deﬁne a homotopy of ¡ on the associated closed
ncells. This process deﬁnes the homotopy on T L X
n
; and we extend it to X,
using the fact that T LX
n
Ç X is a subcomplex and hence a coﬁbration. We now
concatenate the homotopies H
n
:
H(.. t ) =
_
H
i
(.. 2
i÷1
(t ÷ 1 ÷2
i
)). 1 ÷ 2
i
_ t _ 1 ÷ 2
i1
.
H
i
(.. 1). . ÷ X
i
. t = 1.
This map is continuous on X
i
×1 and hence on X×1, 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) Ç Y
n÷1
. 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 Ç X × 1.
Problems
1. Let · Ç X be a subcomplex and ¡ : · ÷ Y a cellular map. Then Y = X L
¸
Y is a
CWcomplex.
2. A CWcomplex is path connected if and only if the 1skeleton is path connected. The
components are equal to the path components, and the path components are open.
8.6 CWApproximation
We show in this section, among other things, that each space is weakly homotopy
equivalent to a CWcomplex. Our ﬁrst aim is to raise the connectivity of a map.
(8.6.1) Theorem. Let ¡ : · ÷ Y be a kconnected map, k _ ÷1. Then there
exists for each n > k a relative CWcomplex (X. ·) with cells only in dimensions
212 Chapter 8. Cell Complexes
¸ ÷ {k ÷1. . . . . n], n _ o, and an nconnected extension J : X ÷Y of ¡ . If ·
is CWcomplex, then · can be chosen as a subcomplex of X.
Proof. (Induction over n.) Recall that the map ¡ is kconnected 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 kconnected. 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 deﬁne J : X ÷ Y
by J[· = ¡ and J(D
0
}
) = {c
}
]. Then X is obtained from · by attaching 0cells
and J is a 0connected extension of ¡ .
n = 1. Suppose ¡ : · ÷ Y is 0connected. 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 ç : S
0
÷·, ç(˙1) = c
˙1
, is an attaching map for a
1cell. We can extend ¡ over · L
ç
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 1complex (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 0connected; 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 1complex 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 ŒJ
}
 ÷ ¬
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 (ˆ
}
. ç
}
): (D
1
. S
0
) ÷ (X
t
. .). We extend J
t
to J
such that J ı ˆ
}
= 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 (ˆ
}
. ç
}
): (D
n
. S
n1
. e
0
) ÷
(Y. ·. a) be a set of maps such that the ,
}
= Œˆ
}
. ç
}
 ÷ ¬
n
(Y. ·. a) generated the
¬
1
(·. a)module ¬
n
(Y. ·. a). We attach ncells to ·by attaching maps ç
}
to obtain
X and extend ¡ to J by the null homotopies ˆ
}
of ¡ ç
}
. The characteristic map
of the ncell with attaching map ç
}
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)
¬
n1
(·)
=
¬
n1
(X)
T
·
(3)
0
¬
n
(·)
¬
n
(Y )
¬
n
(Y. ·)
¬
n1
(·)
¬
n1
(Y )
0.
8.6. CWApproximation 213
The sequences end with 0, since ¬
n1
(X. ·) = 0 and ¬
n1
(·) ÷ ¬
n1
(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 nconnected.
In order to obtain · as a subcomplex of X, one works with cellular attaching
maps.
(8.6.2) Theorem. Let Y be a CWcomplex such that ¬
i
(Y ) = 0 for 0 _ i _ k.
Then Y is homotopy equivalent to a CWcomplex X with X
k
= {+].
Proof. Start with the kconnected 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 CWcomplexes. 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 m÷n.
(8.6.4) Theorem. Let (X
}
[ ¸ ÷ J) be a family of (n÷1)connected CWcomplexes.
Let 
k
: X
k
÷
_
}÷J
X
}
be the inclusion of the kth summand. Then
˛
J
= (
}+
: )
}÷J
¬
n
(X
}
) ÷¬
n
_ _
}÷J
X
}
_
is an isomorphism (n _ 2).
Proof. Let J be ﬁnite. Up to hequivalence 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
}
_
Š
¬
n
_
X
}
_
(]
¸·
) Š
¬
n
(X
}
)
˛
J
¸
(1)
Š
¬
n
(X
}
)
we conclude that ˛
J
is an isomorphism.
Let now J be arbitrary. For each . ÷ ¬
n
_ _
}÷J
X
}
_
there exists a ﬁnite 1 Ç J
such that . is contained in the image of ¬
n
_ _
T
X
}
_
÷ ¬
n
_ _
J
X
}
_
, since a
compact subset is contained in a ﬁnite 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 ﬁnite 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
.
214 Chapter 8. Cell Complexes
(8.6.5) Proposition. Suppose ¬
}
(Y ) = 0 for ¸ > n. Let X be obtained from
· by attaching cells of dimension _ n ÷ 2. Then · Ç X induces a bijection
ŒX. Y  ÷Œ·. Y .
Proof. Surjective. Let ¡ : · ÷ Y be given. Attach (n ÷ 2)cells via maps
ç : S
n÷1
÷ ·. Since ¡ ç : S
n÷1
÷ 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
CWcomplex (X. ·) withcells only indimensions ¸ _ k÷2, 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
k÷1
Ç X
k÷2
Ç
Ç X
k÷t
such that ¬
n
(·. a) Š ¬
n
(X
k÷t
. a) for n _ k, ¬
n
(X
k÷t
. a) = 0 for
k < n _ k ÷t ÷ 1, and X
n÷1
is obtained from X
n
by attaching (m÷1)cells.
The induction step: If we attach (m ÷ 1)cells to X
n
by the attaching maps
ç
}
: (S
n
}
. e
0
) ÷ (X
n
. a) to obtain X
n÷1
, then ¬
n
(X
n
. a) Š ¬
n
(X
n÷1
. a) for
n _ m÷ 1. The exact sequence
¬
n÷1
(X
n÷1
. X
n
. a)
J
÷÷¬
n
(X
n
. a) ÷¬
n
(X
n÷1
. a) ÷0
shows that the Œç
}
 are in the image of d. Thus, if the Œç
}
 generate ¬
n
(X
n
. a),
then ¬
n
(X
n÷1
. a) = 0.
(8.6.7) Example. We can attach cells of dimension _ n÷2 to S
n
to obtain a space
1(Z. n) which has a single nontrivial homotopy group ¬
n
(1(Z. n)) Š Z. See
the section on Eilenberg–Mac Lane spaces for a generalization. Þ
Let i
A
n
: X ÷ XŒn be an inclusion of the type constructed in (8.6.6), namely
XŒn is obtained by attaching cells of dimension greater than n ÷ 1 such that
¬
k
(XŒn) = 0 for k > n and i
A
n
induces an isomorphism¬
k
(i
A
n
) for k _ n. Given
a map ¡ : X ÷Y and i
Y
n
: Y ÷Y Œm for m _ n, there exists a unique homotopy
class ¡
n,n
: XŒn ÷ Y Œm such that i
Y
n
ı ¡ = ¡
n,n
ı i
A
n
; this is a consequence
of (8.6.5). We let ¸
A
n
: X(n) ÷ X be the homotopy ﬁbre of i
A
n
. We call ¸
A
n
the
nconnective 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
ﬁbrations. 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 CWcomplex X and a weak equivalence ¡ : X ÷Y . We call such a weak
8.6. CWApproximation 215
equivalence a CWapproximationof Y . Note that a weak equivalence between CW
complexes is a homotopy equivalence (8.4.3). We show that CWapproximations
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
CWapproximations. Then there exists a map ç : X
1
÷X
2
such that ¡ ˛
1
. ˛
2
ç,
and the homotopy class of ç is uniquely determined by this property.
Proof. Since ˛
2
is a weakequivalence, ˛
2
: ŒX
1
. X
2
 ÷ŒX
1
. Y
2
 is bijective. Hence
there exists a unique homotopy class ç such that ¡ ˛
1
. ˛
2
ç.
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 CWcomplex X. Then M has
the homotopy type of a CWcomplex.
Proof. Suppose i : M ÷X and r : X ÷M are given such that ri is homotopic to
the identity. There exists a CWcomplex  : X Ç 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 hinverse to ¸1, then ¸(1k) . id. Hence ¸ has the left inverse 1 and the right
inverse 1k and is therefore a homotopy equivalence.
A (halfexact) 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
with the properties:
(1) (Homotopy invariance) Pointed homotopic maps induce the same morphism.
(2) (Mayer–Vietoris 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
ŒX. 1
0
÷h(X). Œ¡  ÷¡
+
(u)
is bijective for each X ÷ C
0
.
216 Chapter 8. Cell Complexes
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) = ŒX. 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
¡
+
: ŒX. 1
0
÷ ŒX. 7
0
is always bijective, i.e., ¡ is a weak hequivalence. Thus
we have obtained a CWapproximation X of 7. Þ
Problems
1. As a consequence of (8.6.8) one can extend homotopy functors from CWcomplexes
to arbitrary spaces. Let J be a functor from the category of CWcomplexes 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 CWapproximation of the pseudocircle.
3. Determine the CWapproximation of {0] L {n
1
[ n ÷ N].
4. Let X and Y be CWcomplexes. Show that the identity X ×
k
Y ÷ X × Y is a CW
approximation.
5. Let (Y
¸
[ ¸ ÷ J) be a family of wellpointed spaces and ˛
¸
: X
¸
÷ Y
¸
a family of
pointed CWapproximations. Then
_
¸
˛
¸
is a CWapproximation. Give a counterexample
(with two spaces) in the case that the spaces are not wellpointed.
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 · Ç X in C
0
the
canonical sequence h(X,·) ÷h(X) ÷h(·) is an exact sequence of pointed sets.
8.7 Homotopy Classiﬁcation
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 CWcomplex. Let Y be
a pointed space such that ¬
i
(Y ) = 0 for i > n _ 2. Then
h
A
: ŒX. Y 
0
÷Hom(¬
n
(X). ¬
n
(Y )). Œ¡  ÷¡
+
is bijective.
Proof. The assertion only depends on the pointed homotopy type of X. We use
(8.6.2) and assume X
n1
= {+]. The h
A
constitute a natural transformation in the
variable X. Since (X. X
n÷1
) is (n÷1)connected, the inclusionX
n÷1
Ç X induces
8.8. Eilenberg–Mac Lane Spaces 217
an isomorphism on ¬
n
. By (8.6.5), the restriction r : ŒX. Y 
0
÷ ŒX
n÷1
. Y 
0
is a
bijection. Therefore it sufﬁces to consider the case, that X has, apart from the base
point, only cells of dimension n and n÷1. Moreover, by the homotopy theoremfor
coﬁbrations, we can assume that the attaching maps for the (n÷1)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 coﬁbre sequence
Œ·. Y 
0
(
·
÷÷ ŒT. Y 
0
÷ŒX. Y 
0
÷Œ†·. Y 
0
.
Our assumption about Y yields Œ†·. Y 
0
=
_ _
†S
n
k
. Y
_
0
Š
k
¬
n÷1
(Y ) = 0.
We apply the natural transformation h and obtain a commutative diagram
Œ·. Y 
0
h
.
ŒT. Y 
0
h
!
(
·
¸
ŒX. Y 
0
h
³
(
·
1
¸
0
¸
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
¹
and h
B
are isomorphisms. If · = S
n
, then
h
¹
: ¬
n
(Y ) = ŒS
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
Š
h
_
c
n
k
ŒS
n
k
. Y 
Š
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
) Š ¬
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 Œ†·. Y 
0
acts on ŒX. Y 
0
and the orbits are
mapped injectively, or one uses that ¡ is, up to homotopy, a suspension, because
†
+
:
_ _
S
n1
k
.
_
S
n1
}
_
0
÷
_ _
S
n
k
.
_
S
n
}
_
0
is surjective.
8.8 Eilenberg–Mac Lane Spaces
Let ¬ be an abelian group. An Eilenberg–Mac Lane space of type K.; n/ is a
CWcomplex 1(¬. n) such that ¬
n
(1(¬. n)) Š ¬ and ¬
}
(1(¬. n)) Š 0 for ¸ ,= n.
218 Chapter 8. Cell Complexes
In the cases n = 0. 1, the group ¬ can be nonabelian. In the case n = 0, we think
of 1(¬. 0) = ¬ with the discrete topology.
(8.8.1) Theorem. Eilenberg–Mac Lane spaces 1(¬. n) exist.
Proof. Let n _ 2. There exists an exact sequence
0
J
1
˛
J
0
ˇ
¬
0
with free abelian groups J
0
and J
1
. We ﬁx 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
_
Š
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) Š ¬. 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
) Š 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).
Þ
The adjunction Œ†X. Y 
0
Š ŒX. 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 CWcomplex if Y is a CWcomplex. 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 Eilenberg–Mac Lane spaces. We begin
by showing that Eilenberg–Mac Lane spaces are Hspaces. Then we construct
product pairings 1(¬. m) . 1(j. n) ÷ 1(¬ ˝j. m÷n). In this context ¬ ˝j
denotes the tensor product of the abelian groups ¬ and j (alias Zmodules) over Z.
We call the space 1(¬. n) polarized, if we have chosen a ﬁxed isomorphism
˛: ¬
n
(1(¬. n)) ÷ ¬. If (1(¬. n). ˛) and (1(j. n). ˇ) are polarized complexes,
the product 1(¬. n) × 1(j. n) will be polarized by
¬
n
(1(¬. n) × 1(j. n)) Š ¬
n
(1(¬. n)) × ¬
n
(1(j. n))
˛×ˇ
¬
1
× ¬
2
.
(8.8.3) Proposition. Having chosen polarizations, we obtain from (8.7.1) an iso
morphism Œ1(¬. n). 1(j. n)
0
Š Hom(¬. j).
8.8. Eilenberg–Mac Lane Spaces 219
(8.8.4) Theorem. Let ¬ be an abelian group. Then an Eilenberg–Mac Lane com
plex 1(¬. n) is a commutative group object in hTOP.
Proof. Let 1 = (1(¬. n). ˛) be a polarized complex with base point a 0cell
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 Hspace. The maps m ı (m × id) and m ı (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 Ç 1 × 1 is a coﬁbration, 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 veriﬁed 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 welldeﬁned map
¬
n
(X) × ¬
n
(Y ) ÷¬
n÷n
(X .
k
Y ). (Œ¡ . Œg) ÷Œ¡ . g = Œ¡  . Œg.
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 biadditive.
Proof. The additivity in the ﬁrst variable follows directly from the deﬁnition of the
addition, if we use ÷
1
. We see the additivity in the second variable, if we use the
composition laws ÷
1
and ÷
n÷1
in the homotopy groups.
(8.8.6) Proposition. Let · be an (m÷ 1)connected and T an (n ÷ 1)connected
CWcomplex. Then · . T is (m÷n ÷ 1)connected and the .product
¬
n
(·) ˝¬
n
(T) ÷¬
n÷n
(· .
k
T)
is an isomorphism(m. n _ 2). If mor nequals 1, then one has to use the abelianized
groups.
220 Chapter 8. Cell Complexes
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)
.
¬
n÷n
__ _
S
n
}
_
.
k
T
_
_
¬
n
(S
n
}
)
_
˝¬
n
(T)
(1)
¸
¬
n÷n
_ _
(S
n
}
. T)
_
(2)
¸
(¬
n
(S
n
}
) ˝¬
n
(T))
(3)
¸
(5)
¬
n÷n
(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 mspheres. Next
we let · be the mapping cone of a map ¡ : C ÷D where C and D are wedges of
mspheres. Then we have a commutative diagram with exact rows:
¬
n
(C) ˝¬
n
(Y )
Š
¬
n
(D) ˝¬
n
(T)
Š
¬
n
(·) ˝¬
n
(T)
0
¬
n÷n
(C .
k
T)
¬
n÷n
(D .
k
T)
¬
n÷n
(· .
k
T)
0.
The general case now follows from the observation that the inclusion ·
n÷1
÷ ·
induces an isomorphismon ¬
n
and ·
n÷1
.
k
T ÷·.
k
T induces an isomorphism
on ¬
n÷n
.
Let (1(G. m). ˛), (1(H. n). ˇ) and (1(G˝H. m÷n). ;) be polarized Eilen
berg–Mac 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. m÷n)
such that the diagram
¬
n
(1(G. m)) ˝¬
n
(1(H. n))
.
¬
n÷n
(1(G. m) .
k
1(H. n))
(;
m,n
)
·
G ˝H
;
˛˝ˇ
¸
¬
n÷n
(1(G ˝H. m÷n))
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.
8.8. Eilenberg–Mac Lane Spaces 221
Proof. Let G and H be abelian groups. The ﬁrst nontrivial homotopy group of
1(G. m) .
k
1(H. n) is ¬
n÷n
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. m÷n).
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.,
;
n÷n,]
ı (;
n,n
× id) . ;
n,n÷]
ı (id ×;
n,]
).
The products are graded commutative in the following sense:
1(t. m÷n) ı ;
n,n
. (÷1)
nn
t
t
ı ;
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: 1˝1 ÷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) = ŒX
÷
. 1(1. k)
0
we obtain
via (¡. g) ÷m
k,I
(¡ . g) products
H
k
(X: 1) ˝H
I
(Y : 1) ÷H
k÷I
(X × Y : 1).
which are also associative and graded commutative. (See also Problem 3.) In a
similar manner we can start from an 1module structure 1 ˝ M ÷ M on M.
Later, when we study singular cohomology, we show that for a CWcomplex X the
group ŒX
÷
. 1(1. k)
0
is naturally isomorphic to the singular cohomology group
H
k
(X: 1) with coefﬁcients 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 socalled
cup product in cohomology.
Once singular cohomology theory is constructed one obtains from the repre
sentability theorem of Brown Eilenberg–Mac Lane spaces as representing objects.
8.8.8 Eilenberg–Mac 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 CWspaces and maps e
n
: †1(·. n) ÷ 1(·. n ÷ 1) (an inclusion of
222 Chapter 8. Cell Complexes
subcomplexes; attach cells to †1(·. n) to obtain a 1(·. n÷1)). This spectrum is
an Cspectrum. We have proved in any case that c
n
: 1(·. n) ÷C1(·. n ÷1) is
a weak homotopy equivalence. This sufﬁces if one wants to deﬁne the cohomology
theory only for pointed CWspaces. Þ
Problems
1. From the natural isomorphism ŒX. 1(¬. n)
0
Š ŒX. C
2
1(¬. n ÷2)
0
we see that X ÷
ŒX. 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 hgroup
on 1(¬. n) inducing the group structures of this functor.
2. Let ˛ ÷ ¬
m
(X), ˇ ÷ ¬
n
(Y ), and t : X .
k
Y ÷ Y .
k
X the interchange map. Then
˛ . ˇ = (÷1)
mn
ˇ . ˛.
3. Let M be an 1module. 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 deﬁne a natural structure of an 1module on ŒX. 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
Œ1(¬. 1). 1(j. 1) Š 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
) Ç S(C
n¬1
) Ç . This space carries
a free action of the cyclic group Z,m Ç S
1
by scalar multiplication. Show that S
c
with
this action is a Z,mprincipal covering. The quotient space is a CWspace T(Z,m) and
hence a 1(Z,m. 1).
7. Aconnected 1dimensional CWcomplex X is a 1(¬. 1). Determine ¬ fromthe topology
of X.
8. A connected nonclosed surface (with or without boundary) is a 1(¬. 1).
Chapter 9
Singular Homology
Homology is the most ingenious invention in algebraic topology. Classically, the
deﬁnition of homology groups was based on the combinatorial data of simplicial
complexes. This deﬁnition did not yield directly a topological invariant. The
deﬁnition of homology groups and (dually) cohomology groups has gone through
various stages and generalizations.
The construction of the socalled singular homology groups by Eilenberg [56]
was one of the deﬁnitive settings. This theory is very elegant and almost entirely
algebraic. Very little topology is used as an input. And yet the homology groups are
deﬁned for arbitrary spaces in an invariant manner. But one has to pay a price: The
deﬁnition 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 veriﬁcation 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, deﬁned 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, Ktheories,
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 viewpoint 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 ﬁrst 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 ndimensional standard simplex is
^
n
= ^Œn =
˚
(t
0
. . . . . t
n
) ÷ R
n÷1
ˇ
ˇ
n
i=0
t
i
= 1. t
i
_ 0
_
Ç R
n÷1
.
We set Œn = {0. . . . . n]. A weakly increasing map ˛: Œm ÷Œn induces an afﬁne
map
^(˛): ^Œm ÷^Œn.
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 ^(˛ ı ˇ) = ^(˛) ı ^(ˇ) and ^(id) = id. Let
ı
n
i
: Œn ÷ 1 ÷Œn be the injective map which misses the value i .
(9.1.1) Note. ı
n÷1
}
ı
n
i
= ı
n÷1
i
ı
n
}1
, i < ¸ . (The composition misses i and ¸ .) We
write J
n
i
= ^(ı
n
i
). By functoriality, the J
n
i
satisfy the analogous commutation
rules.
A continuous map o : ^
n
÷ X is called a singular nsimplex in X. The i th
face of o is o ı J
n
i
. We denote by S
n
(X) the free abelian group with basis the
set of singular nsimplices 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 nchain. We think of .
as a formal ﬁnite 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 deﬁned for q _ 1 by
d
q
: S
q
(X) ÷S
q1
(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
q1
d
q
= 0.
Proof. We decompose the sum ddo =
q
}=0
q1
i=0
(÷1)
i÷}
oJ
q
}
J
q1
i
into the
parts
i~}
and
i_}
. When we rewrite the ﬁrst 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 nth homology group
is denoted H
n
(X) = H
n
(X: Z) and called singular homology group of X (with
coefﬁcients 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 : · Ç X be an inclusion. We deﬁne 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
n1
(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 coefﬁcients 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
n1
(·) ÷
is exact. The sequence terminates with H
0
(X) ÷ H
0
(X. ·) ÷ 0. The undeco
rated arrows are induced by the inclusions (·. 0) Ç (X. 0) and (X. 0) Ç (X. ·).
Let (X. ·. T) be a triple, i.e., T Ç · Ç X. The inclusion S

(·) ÷ S

(X)
induces by passage to factor groups an inclusion S

(·. T) ÷ S

(X. T), and its
cokernel can be identiﬁed 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
n1
(·. T) ÷ .
The boundary operator d: H
n
(X. T) ÷ H
n1
(·. T) in the exact sequence of a
triple is the composition of the boundary operator for (X. ·) followed by the map
H
n1
(·) ÷H
n1
(·. 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
226 Chapter 9. Singular Homology
diagram
H
k
(X. ·)
J
(
·
H
k1
(·. T)
(
·
H
k
(X
t
. ·
t
)
J
H
k1
(·
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 deﬁnition (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
deﬁnitions.
9.1.4 Point. Let X = 1 be a point. There is a unique singular nsimplex, hence
S
n
(1) Š Z, n _ 0. The boundary operators d
0
and d
2i÷1
are zero and d
2
. d
4
. . . .
are isomorphisms. Hence H
i
(1) = 0 for i ,= 0; and H
0
(1) Š Z, via the
homomorphism which sends the unique 0simplex to 1 ÷ Z. Þ
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
}
. Þ
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 0simplex o : ^
0
÷ X with the point o(^
0
). Then S
0
(X) is the free
abelian group on the points of X. A singular 1simplex o : ^
1
÷X is essentially
the same thing as a path, only the domain of deﬁnition has been changed from 1
to ^
1
. We associate to o the path n
c
: 1 ÷X, t ÷o(1÷t. t ). Then d
0
o = n
c
(1)
and d
1
o = n
c
(0), hence do = d
0
o÷d
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 zerosimplices 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). Þ
9.2. The Fundamental Group 227
9.2 The Fundamental Group
The signs which appear in the deﬁnition of the boundary operator have an inter
pretation in low dimensions. They are a consequence of orientation conventions.
A singular 1simplex o : ^
1
÷ X is essentially the same thing as a path, only the
domain of deﬁnition has changed from Œ0. 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 deﬁne o: ^
2
÷ X, (t
0
. t
1
. t
2
) ÷ (o + t)(t
0
÷ t
1
,2. t
1
,2 ÷ t
2
), then one
veriﬁes do = o ÷ o + t ÷ t. A loop o : ^
1
÷ X is a 1cycle; let Œo be its
homology class. Thus for loops o. t we have
(1) Œo + t = Œo ÷Œt.
(Here Œ: 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 1simplex is a boundary.
Hence k
0
÷ k
1
is a boundary
Œk
0
 = Œk
1
 ÷ C
1
(X),T
1
(X).
In particular, homotopic loops yield the same element in H
1
(X). Thus we obtain
a welldeﬁned map h
t
: ¬
1
(X. .
0
) ÷ H
1
(X); by (1), it is a homomorphism. The
fundamental group is in general nonabelian. Therefore we modify h
t
algebraically
to take this fact into account. Each group G has the associated abelianized factor
group G
ob
= G,ŒG. G; the commutator group ŒG. 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 deﬁnition 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 1simplex 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
2simplex with faces t
}
= tJ
}
. Since ^
2
is contractible, t
2
+ t
0
. t
1
. This
implies
l
t
(Œt
2
) ÷l
t
(Œt
0
) = l
t
(Œt
2
 ÷Œt
0
) = l
t
(Œt
2
+ t
0
) = l
t
(Œt
1
).
228 Chapter 9. Singular Homology
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
Œo =
a
c
(Œu(o
0
) ÷Œo ÷ Œu(o
1
)) =
a
c
Œ(u(o
0
) + o) + u(o
1
)

.
and the last element is contained in the image of h.
One of the ﬁrst applications of the homology axioms is the computation
H
1
(S
1
) Š Z. Granted the formal result that ¬
1
(S
1
) is abelian, we obtain yet
another proof for ¬
1
(S
1
) Š 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. Deﬁne 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
n1
(X) ÷S
n
(X) are obtained from a cone
construction. Let
q : ^
n1
× 1 ÷^
n
. ((j
0
. . . . . j
n1
). t ) ÷(t. (1 ÷ t )j
0
. . . . . (1 ÷ t )j
n1
).
Given o : ^
n1
÷X, there exists a unique simplex s(o) = so : ^
n
÷X such that
h ı (o × id) = s(o) ı q, since q is a quotient map. For the faces of so we verify
(so)J
i
= s(oJ
i1
), for i > 0, and (so)J
0
= o. From these data we compute for
n > 1,
d(so) = (so)J
0
÷
n
1
(÷1)
i1
(so)J
i
= o ÷
n1
0
(÷1)
}
s(oJ
}1
)
= o ÷ s(do)
and d(so) = o ÷ o
0
for a 0simplex o. These relations imply ds ÷ sd = id ÷c.
Note that c induces the zero map in dimensions n ,= 0. Þ
(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 belement 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
n÷1
(X×1) such
that
(1
n
) ds
A
n
÷s
A
n1
d = j
1
n
(X) ÷ j
0
n
(X)
(chain homotopy), and such that for continuous X ÷Y the relations
(N
n
) (¡ × id)
#
ı s
A
n
= s
Y
n
ı ¡
#
hold (naturality). We construct the s
n
inductively.
n = 0. In this case, s
0
sends the 0simplex o : ^
0
÷{.] Ç X to the 1simplex
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 deﬁnitions.
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 nsimplex; 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
n1
d(
n
).
The righthand side is a cycle in S
n
(^
n
× 1), as the next computation shows.
d(j
1
n
(
n
) ÷ j
0
n
(
n
) ÷ s
n1
d(
n
))
= j
1
n1
(d
n
) ÷ j
0
n1
(d
n
) ÷ ds
n1
(d
n
)
= j
1
n1
(d
n
) ÷ j
0
n1
(d
n
) ÷ (j
1
n1
(d
n
) ÷ j
0
n1
(d
n
) ÷ s
n2
dd
n
) = 0.
We have used the relation (1
n1
) for ds
n1
d(
n
) and that the j
t

are chain maps.
Since ^
n
×1 is contractible, there exists, by (9.3.2), an a ÷ S
n÷1
(^
n
×1) with the
property da = j
1
n
(
n
) ÷j
0
n
(
n
) ÷s
n1
d(
n
). We choose an a with this property and
deﬁne s
n
(
n
) = a and in general s
n
(o) = (o ×id)
#
a for o : ^
n
÷X; the required
230 Chapter 9. Singular Homology
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
n1
d
n
)
= j
1
n
o
#

n
÷ j
0
n
o
#

n
÷ s
n1
o
#
d
n
= j
1
n
o ÷ j
0
n
o ÷ s
n1
do.
We have used: (o × id)
#
is a chain map; choice of a; naturality of j
1

, j
0

, and
(N
n1
); 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
ﬁnally 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
n÷1
(X × 1. · × 1). The computation
d(¡
#
ı s
n
) ÷(¡
#
ı s
n1
)d = ¡
#
ds
n
÷¡
#
s
n1
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). Þ
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)
(n¬1)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 nsimplex o : ^
n
÷ X the singular
prism o × id: ^
n
× 1 ÷ X × 1. The combinatorial (settheoretic) boundary of ^
n
× 1
is ^
n
× 1 L ^
n
× 0 L (d^
n
) × 1, and this corresponds exactly to the deﬁnition of a chain
homotopy, if one takes orientations into account. This idea works; one has to decompose
^
n
× 1 into simplices, and it sufﬁces 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 afﬁne singular simplices introduced later, show that an explicit
formula for a = s
n

n
is
s
n

n
=
n
i=0
(÷1)
i
Œ0. 1. . . . . i. i
/
. (i ÷1)
/
. . . . . n
/
.
(This is a special case of the Eilenberg–Mac Lane shufﬂe 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 ﬁrst.
Let D Ç R
n
be convex, and let ·
0
. . . . . ·
]
be elements in D. The afﬁne singular
simplex o : ^
]
÷D,
i
z
i
e
i
÷
i
z
i
·
i
will be denoted o = Œ·
0
. . . . . ·
]
. With
this notation
dŒ·
0
. . . . . ·
]
 =
]
i=0
(÷1)
i
Œ·
0
. . . . . ¨ ·
i
. . . . . ·
]
.
where ¨ ·
i
means that ·
i
has to be omitted from the string of vertices. For each
· ÷ D we have the contracting homotopy D×1 ÷D, (.. t ) ÷(1 ÷t ). ÷t ·. If
we apply the cone construction 9.3.1 to Œ·
0
. . . . . ·
]
 we obtain Œ·. ·
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 = Œ·
0
. . . . . ·
]
 is o
ˇ
=
1
]÷1
]
i=0
·
i
. We deﬁne 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 deﬁned inductively as
(2) B
]
(
]
) =
_

0
. ¸ = 0.

ˇ
]
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(
ˇ
1
B(d
1
)) = d
1
= Bd(
1
). For ¸ > 1 we compute
dB
]
= d(
ˇ
]
B(d
]
)) = Bd
]
÷ 
ˇ
]
dBd
]
= Bd
]
÷ 
ˇ
]
Bdd
]
= Bd
]
.
We have used: Deﬁnition; (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.
232 Chapter 9. Singular Homology
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 Usmall, if its image is contained in some member of U. The
subgroup spanned by the Usmall 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 afﬁne simplex Œ·
0
. . . . . ·
]
 with
respect to the Euclidean norm is the maximum of the [·
i
÷ ·
}
[.
Proof. Let .. , ÷ Œ·
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
]
Œ·
0
. . . . . ·
]
 is a linear combination
of afﬁne simplices with diameter at most
]
]÷1
J(·
0
. . . . . ·
]
).
Proof. From the inductive deﬁnition (2) and the naturality of B we conclude
(3) BŒ·
0
. . . . . ·
]
 =
]
}=0
(÷1)
}
o
ˇ
BŒ·
0
. . . . . ¨ ·
}
. . . . . ·
]

where o = Œ·
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
BŒ·
0
. . . . . ¨ ·
}
. . . . . ·
]
 are afﬁne of diameter at most
]1
]
J(·
0
. . . . . ¨ ·
}
. . . . . ·
]
) _
]1
]
J(·
0
. . . . . ·
]
). The simplices in BŒ·
0
. . . . . ·
]
 have vertices o
ˇ
and vertices
from simplices in BŒ·
0
. . . . . ¨ ·
}
. . . . . ·
]
. It sufﬁces to evaluate the distance of o
ˇ
from such vertices. It is less than or equal to sup([o
ˇ
÷ .[ [ . ÷ Œ·
0
. . . . . ·
]
).
Let . =
z
}
·
}
. Then [o
ˇ
÷ .[ _ max [o
ˇ
÷ ·
}
[, as in the proof of (9.4.2).
Moreover we have
[o
ˇ
÷ ·
}
[ =
_
_
1
]÷1
_
i
·
i
_
÷ ·
}
_
_
_
1
]÷1
i
[·
i
÷ ·
}
[
_
]
]÷1
max
i,}
[·
i
÷ ·
}
[ =
]
]÷1
J(·
0
. . . . . ·
]
).
Since (¸ ÷ 1),¸ < ¸,(¸ ÷1) we have veriﬁed, 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 kfold iteration of B.)
9.4. Barycentric Subdivision. Excision 233
Proof. We consider the open covering (o
1
(U
ı
)). 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) Ç 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
n÷1
(X). By (9.4.4), there exists k such that
B
k
(b) ÷ S
U
n÷1
(X) (apply (9.4.4) to the ﬁnite 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
n÷1
(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
deﬁne the chain complex S
U

(X. ·) = S
U

(X),S
Uí¹

(·) with homology groups
H
U
+
(X. ·). We obtain a commutative diagramof chain complexes with exact rows:
0
S
Uí¹

(·)
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. ·) Š H
+
(X. ·). By an application of (11.6.3) we
see that the inclusion  is actually a chain equivalence.
234 Chapter 9. Singular Homology
(9.4.7) Theorem (Excision Theorem). Let Y = Y
ı
1
L Y
ı
2
. Then the inclusion
induces an isomorphism H
+
(Y
2
. Y
1
¨ Y
2
) Š H
+
(Y. Y
1
). Let T Ç · Ç X and
suppose that
¨
T Ç ·
ı
. Then the inclusion (X · T. · · T) ÷ (X. ·) induces an
isomorphism H
+
(X · T. · · T) Š 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
) satisﬁes the hypothesis of (9.4.5). By deﬁnition,
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
)
Š
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, ﬁrstly, 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 ﬁrst; we use X = Y , · = Y
1
,
X · T = Y
2
.
Problems
1. Let D Ç R
m
and 1 Ç 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 BŒ·
0
. . . . . ·
)
 in detail. We use (3) in the proof of (9.4.3). By (2), formula (3) also
holds for Œ·
0
. . . . . ·
)
. This yields BŒ·
0
. ·
1
 = Œ·
01
. ·
1
 ÷ Œ·
01
. ·
0
 with barycenter ·
01
,
and for BŒ·
0
. ·
1
. ·
2
 we obtain in shorthand notation what is illustrated by the next ﬁgure.
Œ012. 12. 2 ÷ Œ012. 12. 1 ÷ Œ012. 02. 2 ÷Œ012. 02. 0 ÷Œ012. 01. 1 ÷ Œ012. 01. 0.
÷ ÷
d
d
d
d
d
d
d
d
÷ ÷
÷ ÷
·
0
·
1
·
2
1
2
·
1
÷
1
2
·
2
1
3
·
0
÷
1
3
·
1
÷
1
3
·
2
Œ012. 12. 2
One continues inductively in this manner. Let S(¸ ÷ 1) denote the permutation group
of {0. . . . . ¸]. We associate to o = Œ·
0
. . . . . ·
)
 and ¬ ÷ S(¸ ÷ 1) the simplex o
¬
=
Œ·
¬
0
. . . . . ·
¬
)
, where ·
¬
¡
= Œ·
¬(¡)
. . . . . ·
¬())

ˇ
. 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 deﬁned 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 CWcomplex. 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 ^Œk
n
be the nskeleton of the standard
simplicial complex^Œk (this is the reasonfor switchingthe notationfor the standard
ksimplex). Let S
(n,¹)
k
(X) for n _ 0 denote the subgroup of S
k
(X) spanned by
the singular simplices o : ^Œk ÷X with the property
(#) o(^Œk
n
) Ç ·.
The groups (S
(n,¹)
k
(X) [ k _ 0) form the Eilenberg subcomplex S
(n,¹)

(X) of
S

(X).
(9.5.1) Theorem. Let (X. ·) be nconnected. Then the inclusion of the Eilenberg
subcomplex ˛: S
(n,¹)

(X) ÷S

(X) is a chain equivalence.
Proof. We assign to a simplex o : ^Œk ÷ X a homotopy 1(o): ^Œk × 1 ÷ X
such that
(1) 1(o)
0
= o,
(2) 1(o)
1
satisﬁes (#),
(3) 1(o)
t
= o, provided o satisﬁes already (#),
(4) 1(o) ı (J
k
i
× id) = 1(o ı J
k
i
).
According to (3), the assignment 1 is deﬁned for simplices which satisfy (#). For
the remaining simplices we use an inductive construction.
Suppose k = 0. Then o(^Œ0) ÷ X is a point. Since (X. ·) is 0connected,
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 ksimplex o the
homotopy 1(o ıJ
k
i
) is already deﬁned, and the 1(o ıJ
k
i
) combine to a homotopy
d^Œk × 1 ÷X. Moreover 1(o)
0
is given. Altogether we obtain
¯
1(o): (^Œk × 0 L d^Œk × 1. d^Œk × 1) ÷(X. ·).
Let k _ n. Then ^Œk
n
= ^Œk, and similarly for the faces. By the inductive
assumption,
¯
1(o) sends d^Œk × 1 into ·.
236 Chapter 9. Singular Homology
There exists a homeomorphismk : ^Œk ×1 ÷^Œk ×1 which induces home
omorphisms (see (2.3.6))
^Œk × 0 Š ^Œk × 0 L d^Œk × 1.
d^Œk × 0 Š d^Œk × 1.
d^Œk × 1 L ^Œk × 1 Š ^Œk × 1.
Since (X. ·) is kconnected, the map
¯
1(o) ı k : (^Œk × 0. d^Œk × 0) ÷(X. ·)
can be extended to a homotopy Q: ^Œk ×1 ÷X which is constant on d^Œk ×1
and sends ^Œk × 1 into ·. We now set 1(o) = Q ı k
1
. Then 1(o) extends
¯
1(o), hence (1) and (4) are satisﬁed, and (2) also holds by construction.
Let k > n. We use the coﬁbration (^Œk. d^Œk) in order to extend
¯
1(o) to
1(o). Since ^Œk
n
Ç d^Œk, we see that 1(o)
1
satisﬁes (#).
We now deﬁne j: S
k
(X) ÷ S
(n,¹)
k
(X) by o ÷ 1(o)
1
. Property (4) shows
that j is a chain map, and jı˛ = id holds by construction. We deﬁne s : S
k
(X) ÷
S
k÷1
(X) by s(o) = 1(o)
#
h(
k
)

k
÷ S
k
(^Œk)
h
S
k÷1
(^Œk × 1)
1(c)
#
S
k÷1
(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 ı J
k
i
_
=
(÷1)
i
1(o ı J
k
i
)
#
h(
k1
)
= 1(o)
#
_
(÷1)
i
J
k
i#
h(
k1
)
_
= 1(o)
#
h(d
k
)
show that s is a chain homotopy between ˛ ı j and id.
For k _ n we have ^Œk
n
= ^Œk and therefore S
(n,¹)
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 nconnected. Then H
k
(·)
Š
÷÷ 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 Coefﬁcients 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 deﬁne a subcomplex S
(A,¹,+)

(X) of S

(X) where we require
in addition to (#) that o(^Œk
0
) = {+]. Again the inclusion is a chain equivalence.
Þ
9.6 Homology with Coefﬁcients
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 coefﬁcients 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 ﬁnite formal linear combinations
c
a
c
o, a
c
÷ G of singular n
simplices o; this accounts for the name “chain with coefﬁcients”. The sequence
0 ÷ S

(·) ÷ S

(X) ÷ S

(X. ·) ÷ 0 remains exact when tensored with G,
i.e., S
n
(X. ·) ˝G Š 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
n1
(·: 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
n1
(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) Š G for a point 1, which maps the homology
class of the chain ao to a, where o is the unique 0simplex.
The application of (11.9.1) to topology uses the fact that the singular chain
complex consists of free abelian groups. Therefore we obtain:
238 Chapter 9. Singular Homology
(9.6.1) Theorem (Universal Coefﬁcients). Let 1 be a principal ideal domain and
G an 1module. There exists an exact sequence
0 ÷H
n
(X. ·: 1) ˝
T
G
˛
÷÷H
n
(X. ·: G) ÷Tor(H
n1
(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 coefﬁcients in Z, but the functor H
n
(÷: G) is not
the direct sum of the functors H
n
(÷: Z) ˝ G and Tor(H
n1
(÷: Z). G). Here
is a consequence of (9.6.1): If ¡ : (X. ·) ÷ (Y. T) induces an isomorphism
¡
+
: H
+
(X. ·) Š H
+
(Y. T) for + = n ÷ 1. n, then it induces also an isomorphism
H
n
(X. ·: G) Š 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
÷ {.] Ç 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 Eilenberg–Zilber 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
. ^
nk
) 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 threefold tensor products. As
an application of (11.5.1) we obtain:
(9.7.1) Theorem. (1) Eilenberg–Zilber morphisms 1 and Q exist. For each pair
(1. Q) of Eilenberg–Zilber morphisms the compositions 1 ıQ and Qı1 are nat
urally homotopic to the identity. Hence the 1
A,Y
and Q
A,Y
are chain equivalences
and any two Eilenberg–Zilber morphisms 1. 1
t
are naturally homotopic (similarly
for Q. Q
t
).
(2) An Eilenberg–Zilber morphism 1 is associative and commutative up to
natural homotopy, i.e., the natural transformations 1
A×Y,7
ı (1
A,Y
˝ 1) and
9.7. The Theorem of Eilenberg and Zilber 239
1
A,Y ×7
ı (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
)
#
ı 1
A,Y
and 1
Y,A
ı 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 Eilenberg–Zilber morphism Q is coassociative and cocommutative up
to natural homotopy, i.e., the natural transformations (Q
A,Y
˝1) ı Q
A×Y,7
and
(1˝Q
Y,7
)ıQ
A,Y ×7
are naturally homotopic, and the transformations t
A,Y
ıQ
A,Y
and Q
Y,A
ı (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 Eilenberg–Zilber 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 )) Ç 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(X×T) Ç S(·×Y LX×T)
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.
Þ
240 Chapter 9. Singular Homology
9.7.4 The natural chain map 1 induces natural chain maps for singular chain
complexes with coefﬁcients. Let 1 be a commutative ring and M. N 1modules.
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 1module N. Þ
Problems
1. There exist explicit Eilenberg–Zilber morphisms whichhave further properties. Let ¸. q ÷
N. We use the notation Œn = {0. 1. . . . . n]. A.p; q/shufﬂe is a map z: Œ¸÷q ÷Œ¸ ×Œq
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 Œ¸ × Œq as the vertices of an edgepath 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 afﬁne
(¸ ÷q)simplex Œz(0). . . . . z(¸ ÷q), also denoted z, and the set of these simplices form
a triangulation of the product when z runs through the (¸. q)shufﬂes †(¸. q). (We do not
need this geometric fact, but it explains the idea of the construction.) Deﬁne
1
:
),q
: S
)
(X) ˝S
q
(Y ) ÷S
)¬q
(X. Y ). o ˝t ÷
·÷†(),q)
c(z)((o × t) ı z)
on a pair o. t of singular simplices.
The 1
:
),q
are a strictly associative Eilenberg–Zilber morphism. We call it the shufﬂe
morphism or the Eilenberg–Mac 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 Alexander–Whitney map.
Let o : ^
n
÷ X be an nsimplex, n = ¸ ÷q, 0 _ ¸. q _ n. We have the afﬁne maps
a
)
: ^
)
÷ ^
n
, e
i
÷ e
i
and b
q
: ^
q
÷ ^
n
, e
i
÷ e
nq¬i
. They are used to deﬁne
o
1
)
= o ı a
)
and o
2
q
= o ı b
q
.
The Alexander–Whitney approximation of the diagonal is deﬁned 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 Eilenberg–Zilber
morphism Q as the composition
S
+
(X × Y )
1
³×Y
S
+
(X × Y ) ˝S
+
(X × Y )
pr
³
#
˝pr
Y
#
S
+
(X) ˝S
+
(Y ).
Let Q
.V
be the Eilenberg–Zilber morphism obtained from the Alexander–Whitney ap
proximation of the diagonal and call it the Alexander–Whitney morphism. The Alexander–
Whitney morphism is strictly coassociative.
5. The Eilenberg–Mac Lane morphism1M and the Alexander–Whitney morphism·W are
also compatible in a certain sense:
S
+
(W × X) ˝S
+
(Y × 7)
1^
.V˝.V
S
+
(W × X × Y × 7)
(1×I
³,Y
×1)
#
S
+
(W) ˝S
+
(X) ˝S
+
(Y ) ˝S
+
(7)
1˝:
³,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 mapM˝N ÷1; for instance we canuse
an 1module structure 1˝M ÷M on M. We list some formal properties of the
homology product, for simplicity of notation only for homology with coefﬁcients
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
n1
(X×TL·×Y. X×T) ÷H
n1
(·×(Y. T)).
242 Chapter 9. Singular Homology
d
tt
: H
n
((X. ·)×(Y. T)) ÷H
n1
(X×TL·×Y. ·×Y ) ÷H
n1
((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 Künneth formula (11.10.1) yields
(9.8.2) Theorem (Künneth Formula). Let 1 be an integral domain. Further, let
(·×Y. X×T) be excisive in X×Y 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÷}=n1
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 coefﬁcients in a ﬁeld k we obtain an isomorphism
H
+
(X. ·: k) ˝
k
H
+
(Y. T: k) Š 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 1modules.
(9.8.3) Example. The homology class e ÷ H
1
(R. R · 0: 1) Š 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) Š H
n÷1
((R. R · 0) × (Y. T): N)
for each 1module 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 nfold product e
n
= e × × e ÷ H
n
(R
n
. R
n
· 0: 1) serves as a
canonical generator (a homological 1orientation of R
n
). Þ
9.8. The Homology Product 243
Problems
1. Let ·
0
. . . . ·
n
be afﬁnely independent points in R
n
and suppose that the origin 0 ÷ R
n
is
contained in the interior of the afﬁne simplex · = Œ·
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 ﬁrst 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 deﬁne multi
plicative homology theories axiomatically.
Chapter 10
Homology
In this chapter we deﬁne 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 Borsuk–Ulam. The theorem of Borsuk–Ulam 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; Mayer–Vietoris
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
n1
ık. 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
n÷1
(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 Ç · such that
¨
U Ç ·
ı
. Then
the inclusion (X · U. · · U) ÷ (X. ·) induces an excision isomorphism
h
n
(X · U. · · U) Š 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
ı
1
LY
ı
2
. Then the inclusion induces an isomorphism
h
n
(Y
1
. Y
1
¨ Y
2
) Š h
n
(Y. Y
2
).
The module h
n
(X. ·) is the nth 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 coefﬁcient 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 satisﬁes 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 Ç · Ç X (see [59, p. 24], [189]). The boundary operator
for a triple is deﬁned by
d: h
n
(X. ·) ÷h
n1
(·) ÷h
n1
(·. T):
the ﬁrst 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
n÷1
(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 veriﬁes this more general exact sequence directly from the deﬁnitions;
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 ﬁnite 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) Š 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 CWcomplexes
(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 ç
+
: h
+
÷ k
+
of homology theories consists of a family ç
n
: h
n
÷k
n
of natural transformations
which are compatible with the boundary operators d
t
n
ı ç = ç
n1
ı 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 deﬁne a new theory by shifting the indices
k
n
= h
n¬I
.
3. Let h
·
be a homology theory. For a ﬁxed space Y we deﬁne 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
·
÷ h
·
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 ﬂat 1module, 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
¸
h
·
is a family of homology theories (¸ ÷ J), then their direct sum
¸
¸
h
·
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 · Ç X induces for ¸ = n. n ÷ 1 an isomorphism
h
}
(·) Š 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 hequivalence. 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
hequivalences. 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
n1
(·. 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 : · Ç 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 ﬁnite 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) Š h
n
(· ÷ C. T ÷ C). It sufﬁces 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 Œ
n
 is a generator.
Proof. The proof is by induction on n. Denote by s(i ) = J
n
i
^
n1
the i th face of
^
n
. Consider
h
k1
(^
n1
. d^
n1
)
(d
n
i
)
·
Š
h
k1
(d^
n
. d^
n
· s(i )
ı
) h
k
(^
n
. d^
n
).
J
Š
¸
The space d^
n
· s(i )
ı
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 (^
n1
. d^
n1
) 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 ﬁrst face
map J
n
0
, we obtain by iteration an isomorphism h
kn
(^
0
) Š h
k
(^
n
. d^
n
). (So
far we can work with any homology theory.)
248 Chapter 10. Homology
Now consider the special case h
n
= H
n
of singular homology with coefﬁcients
in Z. By deﬁnition of the boundary operator, dŒ
n
 = (÷1)
i
ŒJ
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 Œ
n1
 to Œ
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
n1
). So we also see that H
k
(D
n
. S
n1
) 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 wellbehaved. For a ﬁnite 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)
¸
J
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
}
+
= ı
k}
.
We show that (3) is an isomorphism. Consider the diagram
h
+
(X · 1. U · 1)
Š
h
+
(l(X
}
· 1
}
). l(U
}
· 1
}
))
=
¸
Š
h
+
(X. U) h
+
(lX
}
. lU
}
).
(3)
¸
The isomorphisms hold by excision; here we use the assumption
¨
1 Ç U
ı
. Diagram
chasing (Five Lemma) now shows that (2) is also surjective.
(10.2.4) Remark. The hypothesis of (10.2.3) is satisﬁed if 1
}
is closed in X
}
and
has a neighbourhood U
}
such that U
}
Ç X
}
is pointed homotopic to the constant
map. These conditions hold if the spaces (X
}
. 1
}
) are wellpointed; see (5.4.4). Þ
10.3. Jordan Curves. Invariance of Domain 249
10.2.5 Suspension. We deﬁne the homological suspension isomorphism o =
o
(A,¹)
by the commutative diagram
h
n
(X. ·)
c
h
n
(1 × X. 1 × T)
(1) Š
Š
¸
h
n÷1
((1. d1) × (X. ·))
J
Š
h
n
(d1 × X L 1 × ·. 0 × X L 1 × ·).
(1) is an isomorphism: excise 0 ×X and then use an hequivalence. The boundary
operator d for the triple sequence of (1 ×X. d1 ×X L1 ×·. 0 ×X L1 ×·) is an
isomorphism, since 0×XL1 ×· Ç 1 ×X is an hequivalence. 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 deﬁned with X × 1.
The nfold iteration of o provides us in particular with an isomorphism
o
n
: h
k
Š h
k
(1
0
) Š h
k÷1
(1. d1) Š Š h
k÷n
(1
n
. d1
n
). Þ
10.3 Jordan Curves. Invariance of Domain
As a ﬁrst 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 · Ç R
n
be a closed subset. Then H
k
(R
n
. R
n
· ·) and
H
k÷1
(R
n
× R. R
n
× R · · × 0) are isomorphic.
Proof. We use the open subsets of R
n÷1
H
÷
= (R
n
· ·)× ÷ 1. oŒ L·×0. oŒ
H

= (R
n
· ·)× ÷ o. 1Œ L·× ÷ o. 0Œ
H
÷
L H

= R
n÷1
· · × 0
H
÷
¨ H

= (R
n
· ·)× ÷ 1. 1Œ.
250 Chapter 10. Homology
These data occur in the diagram
H
k
(R
n
. R
n
· ·)
(3)
_ _ _ _ _ _ _ _ _ _ _ _ _ _
H
k÷1
(R
n÷1
. R
n÷1
· · × 0)
J
H
k
(R
n÷1
. 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
n÷1
. 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
÷
÷H
÷
, (.. s) ÷(.. s ÷t ) starts at the identity and has an
image in the contractible space R
n
×0. oŒ. The map d is an isomorphism, because
H
k
(R
n÷1
. 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
k÷n
(R
n
× R
n
. R
n
× R
n
· · × 0) Š H
k÷n
(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 Ç R
n
be homeomorphic to
S
n1
(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
n1
. 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·(S¨V )
is closed in S and homeomorphic to a proper closed subset D of S
n1
. Therefore
R
n
· D is path connected and hence, by (10.3.3), also R
n
· C. Let ¸ ÷ J and
q ÷ · and n: Œ0. 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(Œ0. t
1
Œ ) Ç J and n(t
2
) limit point of
n( t
2
. 1) Ç ·. Hence there exist t
3
÷ Œ0. t
1
Œ 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 Schoenﬂies (for a topological proof see [141]):
Let J Ç 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
n1
÷R
n
is homeomorphic to an ncell. 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
n1
× Œ÷1. 1 ÷ S
n
be an embedding (n _ 1). Then the closure of
each component of S
n
· ¡(S
n1
× 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. Þ
(10.3.6) Theorem. Let · Ç 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 Ç 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).
252 Chapter 10. Homology
Let V Ç R
n
be homeomorphic to an open subset of U Ç R
n
. Then V is open
in R
n
.
Proof. It sufﬁces to show that ¡(U) is open. It then follows that ¡ is open.
Let D = {. ÷ R
n
[ [. ÷ a[ _ ı] Ç U, and let S be the boundary of D.
It sufﬁces to show that ¡(D
ı
) 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
n1
. 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
Ç ¡(D) and then U
1
Ç ¡(D
ı
). Since D
ı
is connected, so
is ¡(D
ı
). The inclusion ¡(D
ı
) Ç U
1
L U
2
shows that ¡(D
ı
) Ç U
1
. Therefore
¡(D
ı
) = U
1
, and this set is open.
The second statement follows from the ﬁrst, since by hypothesis there exists an
injective continuous map ¡ : U ÷R
n
with image V .
(10.3.8) Theorem (Invariance of Dimension). Let U Ç R
n
and V Ç R
n
be non
empty homeomorphic open subsets. Then m = n.
Proof. Let m < n. Then, by (10.3.7), U Ç R
n
Ç R
n
is open in R
n
, which is
impossible.
10.4 Reduced Homology Groups
The coefﬁcient 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 nonempty 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 = {+] Ç 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) Š h
n
(X. 1) Š
Coker(i
+
). By(11.1.1) we also have an isomorphism
h
n
(X) =
¯
h
n
(X) ˚i
+
(h
n
(1)) Š
¯
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)
Š
K
K
K
K
K
K
K
K
K
h
n
(X)
}
·
Coker i
+
Š
¸r
r
r
r
r
r
r
r
r
r
h
n
(X. 1)
is commutative.
(10.4.1) Proposition. Let · be nonempty. The image of the boundary operator
d: h
n
(X. ·) ÷h
n1
(·) is contained in
¯
h
n1
(·). 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
n1
(·) ÷
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
n1
(·) 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 hequivalence isomorphic to the sequence of (X LCT. ·LCT. +). This
proves the exactness of the triple sequence. Þ
Under the hypothesis of (10.2.3) or (10.2.4) we have for an additive homology
theory an isomorphism
}
¯
h
+
(X
}
) Š
¯
h
+
_ _
}
X
}
_
. For a ﬁnite 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
n÷1
.
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 ).
254 Chapter 10. Homology
Let (C. j) be a monoid in hTOP, 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 jsum in each homotopy set ŒC. Y 
0
, deﬁned by
Œ¡  ÷Œg = Œı(¡ g)j with the folding map ı = (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: ŒC. Y 
0
÷Hom(h(C). h(Y )). Œ¡  ÷¡
+
is a homomorphism.
Proof. Consider the commutative diagram
h(C)
µ
·
d
P
P
P
P
P
P
P
P
P
P
P
P
h(C C)
(( ·¿)
·
Š (1)
h(Y Y )
ı
·
h(Y )
h(C) ˚h(C)
(
·
˚¿
·
h(Y ) ˚h(Y )
(2)
¸
o
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 wellpointed
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 : · Ç X be acoﬁbrationandlet ¸: (X. ·) ÷(X,·. +)
be the map which identiﬁes ·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. ·) Š
¯
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 coﬁbration
we have an hequivalence ¸: XLC· ÷XLC·,C· Š 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 coﬁbration · Ç X
÷
¯
h
n
(·) ÷
¯
h
n
(X) ÷
¯
h
n
(X,·) ÷
¯
h
n1
(·) ÷ .
(10.4.6) Proposition. Let ¸ : · Ç X be a coﬁbration and attach X to T via
¡ : · ÷T. Then the map h
n
(X. ·) ÷ h
n
(X L
¹
T. T) induced by the inclusion
is an isomorphism.
Proof. We apply (10.4.5) to the homeomorphism X,· ÷X L
¹
T,T.
(10.4.7) Proposition. Let ¡ : X ÷ Y be a pointed map between wellpointed
spaces and X
(
÷÷ Y
(
1
÷÷ C(¡ ) the beginning of the coﬁbre 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 Ç 7(¡ ), . ÷ (.. 0) the canonical inclusion, a coﬁbration.
Consider the commuting diagram
¯
h
n
(X)
=
¯
h
n
(7(¡ ))
Š
h
n
(7(¡ ). X)
Š
¯
h
n
(X)
(
·
¯
h
n
(Y )
c(( )
·
¯
h
n
(7(¡ ),X)
with the canonical inclusion c(¡ ): Y Ç 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 wellpointed pair the quotient map 7(¡ ),X ÷ C(¡ ) is a
homotopy equivalence, since a unit interval which is embedded as a coﬁbration is
identiﬁed to a point.
For a wellpointed space X the inclusion X × d1 L {+] × 1 Ç X × 1 is
a coﬁbration. The quotient is the suspension †X. From (10.4.5) we obtain an
isomorphism
q : h
n
((X. +) × (1. d1)) ÷h
n
(†X. +) Š
¯
h
n
(†X)
and a suspension isomorphism ¯ o :
¯
h
n
(X) Š
¯
h
n÷1
(†X) which makes the diagram
h
n
(X. +)
c
q
h
n÷1
((X. +) × (1. d1))
q
¯
h
n
(X)
¯ c
¯
h
n÷1
(†X)
commutative. In particular, we obtain h
n
Š
¯
h
n
(1
0
,d1
0
) Š h
n÷n
(1
n
,d1
n
).
256 Chapter 10. Homology
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 coﬁbre sequence of id
³
.¡
q
in order to derive an exact
sequence
0 ÷
¯
h
n
(X) ˝Z,q ÷
¯
h
n¬1
(X . M(q)) ÷Tor(
¯
h
n1
(X). Z,q) ÷0.
This suggests deﬁning for any homology theory
¯
h
·
a theory with coefﬁcients Z,q by
¯
h
·
(÷ . 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
n1
. 1
n
= D
n
· S
n1
of R
n
and D
n
˙
=
{(.
1
. . . . . .
n÷1
) ÷ S
n
[ ˙.
n÷1
_ 0]. The space D
0
= {D] is a singleton and
S
1
= 0.
We deﬁne a suspension isomorphism o
÷
as the composition
o
÷
:
¯
h
k1
(S
n1
)
h
k
(D
n

. S
n1
)
J
Š
¸
x
¬
Š
h
k
(S
n
. D
n
÷
)
¯
h
k
(S
n
).
Š
}
¸
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
kn
Š
¯
h
kn
(S
0
) Š
¯
h
k
(S
n
). h
k
(S
n
) Š h
kn
˚h
n
.
The ﬁrst isomorphism is determined by
h
n
÷
¯
h
n
(S
0
) Ç h
n
(S
0
) Š h
n
(÷1) ˚h
n
(÷1) Š h
n
˚h
n
. . ÷(.. ÷.).
We also have an isomorphismd: h
k
(D
n
. S
n1
) Š h
k1
(S
n1
. e) Š
¯
h
k1
(S
n1
).
In the case of ordinary homology H
+
(÷: G) with coefﬁcients in G we obtain:
H
k
(S
n
: G) Š
G. k = n > 0. n > k = 0.
G ˚G. k = n = 0.
0. otherwise.
Moreover H
n
(D
n
. S
n1
: G) Š H
n
(R
n
. R
n
· 0: G) Š G.
A generator of H
n
(D
n
. S
n1
: Z) Š 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 deﬁne 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 deﬁne 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 deﬁnition we see directly some properties of the degree:
J(¡ ı 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: ŒS
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
) Š Z. Thus
the Homgroup is canonically isomorphic to Z. We now use that ¬
n
(S
n
) Š
ŒS
n
. S
n
 Š 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 deﬁned 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
deﬁnition.
(10.5.2) Proposition. Deﬁne an isomorphism o

:
¯
h
k1
(S
n1
) ÷
¯
h
k
(S
n
) as in
the case of o
÷
, but with the roles of D
n
˙
interchanged. Then o
÷
= ÷o

.
Proof. Consider the commutative diagram
¯
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
n1
)
}
¬
¸U
U
U
U
U
U
U
U
}

i
i
i
i
i
i
i
i
i
J
h
k
(D
n

. S
n1
)
Š
x
¬
¸
i

j
j
j
j
j
j
j
j
J
¬
T
T
T
T
T
T
T
T
h
k
(D
n
÷
. S
n1
)
i
¬
¸U
U
U
U
U
U
U
U
J

¸·j
j
j
j
j
j
j
j
Š
x

¸
¯
h
k1
(S
n1
)
and apply the Hexagon Lemma (11.1.3).
(10.5.3) Proposition. Let · ÷ O(n ÷ 1) and l
¹
: S
n
÷ S
n
, . ÷ ·.. Then l
¹
+
is on
¯
h
k
(S
n
) the multiplication by det(·). The antipodal map . ÷÷. on S
n
has
the degree (÷1)
n÷1
.
258 Chapter 10. Homology
Proof. Let t : S
n
÷ S
n
change the sign of the ﬁrst 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
¹
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 ﬁnish the proof.
(10.5.4) Proposition. Let ¡ : S
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
¹
: R
n
÷ R
n
, . ÷ ·.. Then l
¹
+
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
k1
(R
n
· 0) Š
¯
h
k1
(S
n1
)
which are compatible with the action of l
¹
+
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
n÷1
[ [.÷a[ = r]. We have a canonical homeomorphism
h
i,o
: S
n
÷ S
n
i
(a), . ÷ r. ÷a and r
b
: R
n÷1
· b ÷ S
n
, . ÷ N(. ÷ b). The
winding number of ¡ : S
n
i
(a) ÷ R
n÷1
· b about b is deﬁned as the degree of
r
b
ı ¡ ı h
i,o
.
10.5.6 Local degree. Let 1 Ç 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) Š 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
·
Š
¸
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 deﬁne 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)
Š
¸
}
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
Š
¸
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 ﬁnite set, then we can choose U as a
disjoint union of open sets U
x
. . ÷ 1 such that U
x
¨ 1 = {.]. In that case
x÷1
H
n
(U
x
. U
x
· .) Š H
n
(U. U · 1); and H
n
(U
x
. U
x
· .) Š Z, by exci
sion and H
n
(D
n
. S
n1
) Š Z. We call the integer J(¡. .) deﬁned by ¡
+
:
U
\
,x
=
J(¡. .):
S
n
,]
the local degree of ¡ at ¸. With this notation we therefore have
J(¡ ) =
x÷1
J(¡. .). Þ
Let U Ç R
n
be an open neighbourhood of the origin. Then we have an excision
isomorphismh
k
(U. U ·0) Š 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) Š 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).
260 Chapter 10. Homology
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 deﬁne 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 plussign holds, if the differential respects the orientation
at .. For the proof express ¡ in terms of orientation preserving local charts and
apply (10.5.8). Þ
If S is homeomorphic to S
n
and ¡ a self map of S we choose a homeomorphism
h: S
n
÷S and deﬁne the degree of ¡ as the degree of h
1
¡ h. Let g: R
n
÷R
n
be a proper map. We deﬁne its degree as the degree of the extension of g to the
onepoint compactiﬁcation.
A map ¡ : S
n
× S
n
÷ S
n
, n _ 1, has a bidegree (a. b) ÷ Z × Z where a is
the degree of . ÷ ¡(.. ,) for a ﬁxed , and b the degree of , ÷ ¡(.. ,) for a
ﬁxed ..
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¡(.)÷(1÷t ). ,= 0 for t ÷ Œ0. 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 afﬁne 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
n¬1
= 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)
i¬1
.
The point (0. . . . . 0. 1) is a regular value.
5. Consider a complex polynomial as selfmap of the Riemann sphere C1
1
Š 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 selfmap of C1
1
. Show that the homological degree is a = max(m. n). In
homogeneous coordinates ¡ can be written as Œ:. n ÷Œn
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 Œc. 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 bidegree (1 ÷(÷1)
n1
. ÷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 satisﬁes
¡(÷.) = ÷¡(.), 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 Borsuk–Ulam theorems and Lusternik–Schnirelmann 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
n1
.
(4) There does not exist a continuous map ¡ : D
n
÷S
n1
which restricts to an
antipodal map on the boundary.
(5) An antipodal map S
n1
÷S
n1
is not null homotopic.
(6) Suppose S
n
= J
1
L J
2
L L J
n÷1
with nonempty 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 ·
n÷1
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
n1
onto the ﬁrst n ÷1 coordinates is a homeomorphism which is the
262 Chapter 10. Homology
identity on the boundary. Suppose ¡ : D
n
÷ S
n1
is antipodal on the boundary.
Deﬁne g: S
n
÷S
n1
by
g(.) =
_
¡ h(.). . ÷ D
n
÷
.
÷¡ h(÷.). . ÷ D
n

.
If . ÷ S
n1
= D
n
÷
¨ D
n

, then h(.) = ., h(÷.) = ÷. and ¡ h(.) = ¡(.) =
÷¡(÷.) = ÷¡ h(÷.). Hence g is welldeﬁned and continuous. One veriﬁes that
g is antipodal.
(4) = (5). If an antipodal map S
n1
÷ S
n1
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
n1
would contradict (5).
(1) =(6). We consider the function
¡ : S
n
÷R
n
. . ÷(J(.. J
1
). . . . . J(.. J
n
))
deﬁned 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 nonzero,
then . and ÷. are not contained in
_
n
i=1
J
i
, so they are contained in J
n÷1
.
(6) = (3). There exists a closed covering J
1
. . . . . J
n÷1
of S
n1
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
n1
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
n1
.
(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 ﬁnd
a closed shrinking and apply (6). In the general case let ·
1
. . . . . ·
}
be closed and
U
}÷1
. . . . . U
n÷1
open. Suppose there are no antipodal pairs in the U
}
. Thicken
the ·
i
to open cneighbourhoods U
t
(·
i
). Let c = n
1
. By the case of an open
covering we ﬁnd an antipodal pair (.
n
. ÷.
n
) in some U
t
(·
i
). By passing to a
subsequence we ﬁnd an i _ ¸ and
lim
no
J(.
n
. ·
i
) = lim
no
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
) Ç 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
) Ç 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
n1
) Ç S
n1
. The induction is now a consequence of (10.6.4).
(10.6.4) Proposition. Let ¡ : S
n
÷S
n
be an odd map such that ¡(S
n1
) Ç S
n1
.
Then we have the degree D(¡ ) of ¡ and the degree J(¡ ) of its restriction to S
n1
.
These degrees have the same parity.
Proof. We study the diagram in the proof of (10.5.2) more closely for singular
homology with coefﬁcients in Z. We ﬁx a generator : ÷
¯
H
n1
(S
n1
) and deﬁne
other generators by
d
˙
:
˙
= :. s
˙
:
˙
= n
˙
. l
˙
·
˙
= n
˙
.
We set i
÷
:
˙
= u
˙
. The Sum Lemma tells us that u
˙
is a Zbasis of the group
H
n
(S
n
. S
n1
). Let T : S
n
÷S
n
be the antipodal map. Suppose given ¡ : S
n
÷
S
n
such that ¡(S
n1
) Ç S
n1
and T¡ = ¡ T . Then we have two degrees D(¡ )
and J(¡ ) deﬁned by
¡
+
(·
÷
) = D(¡ )·
÷
. ¡
+
(:) = J(¡ ):.
Since u
˙
is a Zbasis 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

) = (a÷b): we obtain the ﬁrst assertion J(¡ ) = a÷b. 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
÷
.
264 Chapter 10. Homology
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
¸
÷
u
÷
= s
÷
i

:
÷
= l
÷
= ¸
÷
j·
÷
= c¸
÷
(u
÷
÷ u

) = c¸
÷
u
÷
shows c = 1. The computation
D(¡ )(u
÷
÷ u
÷
) = D(¡ )j·
÷
= j¡
+
·
÷
= ¡
+
j·
÷
= ¡
+
(u
÷
÷ u

) = (a ÷ b)(u
÷
÷ u

)
ﬁnally yields the second assertion D(¡ ) = a ÷ b.
(10.6.5) Example. The map ¡ : S
1
÷ S
1
, : ÷ :
2k÷1
satisﬁes 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 1simplex represented by the path from 1
to ÷1 and J

the 1simplex running from ÷1 to 1 (both counterclockwise). Then
J
÷
÷J

is a cycle and ·
÷
= ŒJ
÷
÷J

 is a natural choice of a generator. Then
J

represents u
÷
. n
÷
, and :
÷
; and : = Œ1 ÷ Œ÷1. By considering the simplex
¡J

, the relation ¡
+
u
÷
= Œ¡J

 = (k ÷1)ŒJ

 ÷kŒJ
÷
 becomes apparent. For
k = 1 say, ¡J

runs counterclockwise from ÷1 to 1 to ÷1 to 1.
Let ¡ : S
1
÷ S
1
be any selfmap which commutes with the antipodal map.
We can multiply ¡ by a constant such that the new map g satisﬁes g(1) = 1. From
(10.6.4) we see that g and hence ¡ has odd degree, since g has degree 1 on S
0
. Þ
We now apply the Borsuk–Ulam theorem to a problem in combinatorics: The
determination of the chromatic number of the socalled Kneser graphs.
We begin by explaining the problem. Let Œn = {1. . . . . n] and denote by N
k
the set of subsets of Œn 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 identiﬁed with one or two points in 1. (In other
terms: A graph is a 1dimensional CWcomplex.) 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 Œn.
Let C = Œk, and call the elements of C colours. A kcolouring 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 kcolouring. The following result was conjectured by
Martin Kneser (1955) [105]. This conjecture was proved by Lovász [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
n÷2k÷2. We associate to a set J with k elements the colour ç(J) = min(min(J).
10.7. Mayer–Vietoris Sequences 265
n ÷ 2k ÷2). Suppose ç(J) = ç(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 Ç 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 ktuple in X with colour i ]
and ·
d÷1
= S
d
· (·
1
L L ·
d
). The sets ·
1
. . . . . ·
d
are open and ·
d÷1
is
closed. By (10.6.1), one of the sets ·
i
contains an antipodal pair .. ÷..
If i _ J, then we have two disjoint ktuples with colour i , one in H(.) the
other one in H(÷.). This contradicts the deﬁnition of a colouring.
Let i = J ÷1. Then, by deﬁnition of ·
1
. . . . . ·
d
, the halfspace 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 n÷2(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
) Ç S
i
and the map S
i
÷S
i
induced by ¡ has degree J
i
.
10.7 Mayer–Vietoris Sequences
We derive further exact sequences for homology from the axioms, the socalled
Mayer–Vietoris sequences.
Let h
+
be a homology theory. Let (X: ·. T) be a triad, i.e., ·. T Ç X = ·LT.
The triad (X: ·. T) is said to be excisive for the homology theory if the inclusion
induces an (excision) isomorphism h
+
(·. · ¨ T) Š h
+
(X. T). The condition is
actually symmetric in · and T. We write ·T = · ¨ T.
(10.7.1) Proposition. The following are equivalent:
(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.
266 Chapter 10. Homology
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 deﬁned by
^: h
n
(X) ÷h
n
(X. T)
Š
÷÷ h
n
(·. ·T)
J
÷÷h
n1
(·T).
This operator is part of the Mayer–Vietoris sequence (=MVS) of the triad.
(10.7.2) Theorem. Let (·LT: ·. T) be an excisive triad and C Ç ·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
n1
(·T)
(1)
÷÷
is exact. The inclusions i
¹
: ·T Ç · and i
B
: ·T Ç T yield the ﬁrst map
. ÷(÷i
¹
+
.. i
B
+
.); and the inclusions ¸
¹
: · Ç ·LT and ¸
B
: T Ç ·LT yield
the second map (a. b) ÷ ¸
¹
+
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 and·LT Ç X.
We deﬁne a boundary operator ^ by
^: h
n
(X. · L T)
J
÷÷h
n1
(· L T. ·) Š h
n1
(T. ·T) ÷h
n1
(X. ·T).
(10.7.3) Theorem. The sequence
. . .
Z
h
n
(X. ·T)
(1)
h
n
(X. ·) ˚h
n
(X. T)
(2)
h
n
(X. · L T)
Z
h
n1
(X. ·T)
(1)
is exact. The maps (1) and (2) are deﬁned 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
n1
(·T. C)
h
n
(T. C)
h
n
(X. C)
h
n
(X. T)
h
n1
(T. C)
.
10.7. Mayer–Vietoris Sequences 267
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
¿
i
b
i
C
i
h
i
c
i
·
i1
o
i1
÷
÷
·
t
i
(
/
i
T
t
i
¿
/
i
C
t
i
h
/
i
·
t
i1
÷ .
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
·
i1
÷ .
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) = 0·L1·T L1T
of 1 × X. Let ¸: N(·. T) ÷X be the projection onto the second factor.
(10.7.5) Proposition. The following are equivalent:
(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) Š h
+
(0· ÷1T. 0·T ÷1·T) Š h
+
(N. 1·T).
by additivity, excision and hequivalence. 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 = ·
ı
LT
ı
. 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 hequivalence. Recall from
Section 3.3:
(10.7.6) Proposition. Suppose the covering ·, T of X is numerable. Then ¸ is an
hequivalence. Þ
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 hinvariance we see that
the inclusion induces an isomorphism
h
n
((1. d1) × ·T) Š h
n
(N. 0 × · L 1 × T).
268 Chapter 10. Homology
We rewrite the exact sequence of the pair (N. 0×·L1×T). Using the suspension
isomorphism and the additivity, we obtain an exact sequence
÷h
n
(·) ˚h
n
(T) ÷h
n
(N) ÷h
n1
(·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
) Ç (X: X
0
. X
1
) be two excisive triads. Set
X
01
= X
0
¨ X
1
, ·
01
= ·
0
¨ ·
1
. Then there exists an exact Mayer–Vietoris
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(·)) Š h
+
(X. ·), by (10.7.5) and the hypotheses.
(2) The inclusions (X
}
. ·
}
) Š {¸ ] × (X
}
. ·
}
) ÷ (N(X. ·). N(·)) induce an
isomorphism
h
+
(X
0
. ·
0
) ˚h
+
(X
1
. ·
1
) ÷h
+
(N(X. ·). N(·)).
For the proof one excises Œ1,3. 2,3 × ·
01
and then uses an hequivalence 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 Œ0. 1,4X
01
L 1·
01
L Œ3,4X
01
L 1X
1
.
Then one can excise 0X
0
and 1X
1
and use suitable hequivalences.
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
n÷1
(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 MVsequence becomes, in the special
cases X = X
0
= X
1
= X
01
and · = ·
0
= ·
1
= ·
01
Ç X
01
, the same as in the
previously discussed algebraic derivation of the MVsequences.
10.7. Mayer–Vietoris Sequences 269
(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 ﬁx a generator : ÷ H
n
(S
n
) and write :
}
= i
}
+
:. Then
(:
1
. :
2
) is a Zbasis of H
n
(S
n
× S
n
). Let ˛ = (˛
1
. ˛
2
): S
n
× S
n
÷ S
n
× S
n
be a map with bidegree (a. b) of ˛
1
and bidegree (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
n÷1
× S
n
÷
D
n÷1
× S
n
the point (.. ,) ÷ S
n
× S
n
in the ﬁrst 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 MVsequence 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
n÷1
× S
n
) ˚H
n
(D
n÷1
× S
n
) ÷H
n
(X).
We use the Zbasis (:
1
. :
2
) as above. The inclusion S
n
÷D
n÷1
×S
n
, . ÷(0. .)
give us as image of : the generators u
1
. u
2
in the summands H
n
(D
n÷1
×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) Š Z Š H
2n÷1
(X) and H
}
(X) = 0 for ¸ ,= 0. n. 2n ÷ 1. We
leave the case c = 0 to the reader. Þ
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 ﬁnite, then the third set is
also ﬁnite 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 deﬁnition
of path components?
2. Let the real projective plane 1 be presented as the union of a Möbius band M and a disk D,
glued together along the common boundary S
1
. Determine the groups and homomorphisms
in the MVsequence of (1: M. D). Do the same for the Klein bottle (1: M. M). (Singular
homology with arbitrary coefﬁcients.)
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
) Ç Y
i
. Suppose that the restriction
_
u÷.
X
u
÷
_
u÷.
Y
u
of ¡ induces a homology isomorphism for each 0 ,= · Ç {1. . . . . n]. Then ¡
270 Chapter 10. Homology
induces a homology isomorphism.
4. Suppose ·T Ç · and ·T Ç T are closed coﬁbrations. Then ¸: N(·. T) ÷· L T is
an hequivalence.
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 L·×1: X ×0 L·×1. X ×1 L·×1) is always excisive. (Excision
of X × 0 L · × Œ0. 1,2Œ and hequivalence.)
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) ¸
k÷1
¡
k
= ¸
k
.
(2) If a
k
: X
k
÷Y is a family of maps such that a
k÷1
¡
k
= a
k
, then there exists
a unique map a: X ÷Y such that a¸
k
= a
k
.
(This deﬁnition 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
Ç X
k÷1
are inclusions, we can take as
colimit the union X =
_
i
X
i
together with the colimit topology: U Ç 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 socalled
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
×Œi. 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
: ¸
k÷1
¡
k
. ¸
k
,
a linear homotopy in X
k
×Œk. k ÷1. Thus the telescope T consists of the mapping
cylinders of the maps ¡
i
glued together. The data deﬁne the homotopy colimit of
the sequence.
10.8. Colimits 271
Given maps a
k
: X
k
÷ Y and homotopies h
k
: X
k
× Œk. k ÷ 1 ÷ Y from
a
k÷1
¡
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
× Œk. k ÷1 ÷T with a is h
k
.
We have subspaces T
k
Ç T , the image of
k1
i=1
X
i
× Œi. 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
+
= ¸
k÷1
+
¡
k
+
: h
n
(X
k
) ÷ h
n
(X
k÷1
) ÷
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
÷
k_1
·
k
. (.
k
) ÷(.
k÷1
÷ 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
2i1
× {2i ÷
1
2
].
(10.8.2) Lemma. The inclusions
i_1
X
2i
× {2i ] ÷·.
i_1
X
2i1
× {2i ÷ 1] ÷T.
i_1
X
i
× {i ] ÷· ¨ T
are hequivalences, and (·. T) is a numerable covering of T .
Because of this lemma we have a Mayer–Vietoris sequence
h
n
(· ¨ T)
h
n
(·) ˚h
n
(T)
h
n
(T )
h
n
(X
}
)
˛
Š
¸
}=0(2)
h
n
(X
}
) ˚
}=1(2)
h
n
(X
}
).
Š
¸
The map ˛ has the form
˛(.
2i
) = (.
2i
. ¡
2i
+
(.
2i
)) and ˛(.
2i÷1
) = (¡
2i÷1
+
(.
2i÷1
). .
2i÷1
)
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 deﬁnition of the colimit.
272 Chapter 10. Homology
For applications we have to ﬁnd conditions under which the homotopy colimit
is hequivalent to the colimit. We consider the case that the ¡
k
: X
k
÷ X
k÷1
are
inclusions, and denote the colimit by X =
_
k
X
k
. We change the deﬁnition of the
telescope slightly and consider it now as the subspace
T =
_
k
X
k
× Œk. k ÷1 Ç X × Œ0. oŒ .
The topology of T may be different, but the proof of (10.8.1) works equally well
in this case. The projection onto the ﬁrst factor yields ¸: T ÷X.
(10.8.3) Example. Let z: X ÷Œ1. oŒ be a function such that
s = (id. z): X ÷X × Œ1. oŒ
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
i1
for each i .
Let (U
i
[ i ÷ N) be a numerable covering of X with locally ﬁnite 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
i1
. Þ
(10.8.4) Proposition. Suppose the inclusions X
k
Ç X
k÷1
are coﬁbrations. Then
T Ç X × Œ1. oŒ is a deformation retract.
Proof. Since X
k
Ç X is a coﬁbration, there exists by (5.1.3) a homotopy
h
k
t
: X × Œ1. oŒ÷X × Œ1. oŒ rel Y
k
= X
k
× Œ1. oŒ LX × Œk ÷1. oŒ
from the identity to a retraction X × Œ0. oŒ÷ Y
k
. The retraction 1
I
acts as
the identity on X
k
× Œ1. oŒ for l > k, and therefore the inﬁnite composition
1
}
= ı1
}÷2
ı1
}÷1
ı1
}
is a welldeﬁned 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) Š ZŒ1,2, 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 ) Š Q?
3. Let X be a CWcomplex and T the telescope of the skeleton ﬁltration. Then the inclusion
T Ç X × Œ1. oŒ 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 deﬁned for a triple (·. T. C) by
¯
d: h
n÷1
(1·. d1· L 1T)
J
÷÷h
n
(d1· L T. d1· L 1C)
Š
÷÷ h
n
(1T. d1T L 1C).
In order to work with this deﬁnition 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
(d1·L1C. d1T L1C) ÷h
n
(d1·L1T. 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) Š h
n
(1T L d1·. d1·) into the exact
sequence of (·. T) for the k
+
groups. Let U Ç T Ç ·and
¨
U Ç T
ı
. 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 Ç (1T L d1·)
ı
and the
usual excision isomorphism. The isomorphisms o
(¹,B)
: h
n
(·. T) ÷k
n÷1
(·. T)
deﬁned 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
n÷1
(·. T)
J
c
(.,!)
h
n
(T. C)
c
(!,C)
k
n÷2
(·. T)
¯
J
k
n÷1
(T. C)
is commutative.
274 Chapter 10. Homology
Proof. By naturality it sufﬁces to consider the case C = 0. The Hexagon Lemma
shows ˛ = ÷ˇ for the maps
˛: h
n÷1
(1·. 1T L d1·) ÷h
n
(1T L d1·. d1·)
Š h
n
(1T L 0·. 0· L 1T) ÷h
n1
(0· L 1T. 0·).
ˇ: h
n÷1
(1·. 1T L d1·) ÷h
n
(1T L d1·. 1T L 0·)
Š h
n
(d1·. 0· L 1T) ÷h
n1
(0· L 1T. 0·).
and the center group h
n
(1T L d1·. 1T L 0·). Let ¸ be the isomorphism
¸ : h
n1
(1T L 0·. 0·)
Š
÷÷ h
n1
(1T) Š h
n1
(T).
By diagram chasing one veriﬁes o
(B)
¸˛ =
¯
d and ¸ˇo
(¹,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
(d1·L1T. d1TL1C).
Here  is induced by the inclusion, and 
¡
by a ÷(v. a).
Proof. This is a consequence of (10.9.1) and (10.7.1).
(10.9.4) Proposition. For each triple (·. T. C) the diagram
h
n
(·. T)
c
(.,!)
˛
h
n÷1
(1·. d1· L 1T)
J
h
n
(·. T) ˚h
n
((·. T) ˚h
n
(1T. d1T L 1C)
ˇ
Š
h
n
(d1· L 1T. d1T L 1C)
is commutative; here ˛(.) = (.. ÷.. ÷o
(B,C)
d.), and the isomorphismˇ 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 veriﬁcation it is helpful
to use the inverse isomorphism of ˇ 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 deﬁnitions 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 ﬁeld – and the reader
may wish to look into the appropriate textbooks.
The main topics are diagrams and exact sequences, chain complexes, derived
functors, universal coefﬁcients and Künneth 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 1linear maps. (The category ABEL of abelian groups can be iden
tiﬁed with Z MOD.) Recall that a sequence of 1modules and 1linear maps
÷ ·
i÷1
o
i¬1
÷÷ ·
i
o
i
÷÷ ·
i1
÷ is exact at ·
i
if Im(a
i÷1
) = 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 1module structure
is given by pointwise addition and scalar multiplication. The canonical projection
¸
k
:
}
M
}
÷ M
k
onto the kth 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 ﬁnite number of .
}
are zero. We have the canonical injection
i
k
: M
k
÷
}
M
}
, deﬁned by ¸
k
i
k
= id and ¸
k
i
I
= 0 for k ,= l. If ¡
}
: M
}
÷N
is a family of 1linear maps, then
(¡
}
):
M
}
÷N
denotes the morphism determined by (¡
}
) ı i
k
= ¡
k
. If g
}
: 1 ÷ M
}
is a family
of 1linear maps, then
(g
}
): 1 ÷
}
M
}
denotes the morphism determined by ¸
k
ı (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 · Ç 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 ﬁnitely generated abelian groups. An
element of ﬁnite 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 ﬁnitely 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 ﬁnitely 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
ﬁnitely generated modules over principal ideal domains.
A linear map ¸: M ÷ M with the property ¸ ı ¸ = ¸ 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
) ı (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
.
278 Chapter 11. Homological Algebra
Proof. The part i. ¸. k satisﬁes 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:
·
˛
o
T
ˇ
b
C
;
c
D
ı
d
1
e
·
t
˛
/
T
t
ˇ
/
C
t
;
/
D
t
ı
/
1
t
.
(1) a surjective, b. J injective =c injective. (Here the 1part 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, ˇ(·) = n for some ·. Since ˇ
t
b(·) = cˇ(·) = 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 = ˇ(·) = 0.
(2) Given n
t
÷ C
t
. Choose . such that J(.) = ;
t
(c
t
). By exactness,
commutativity, and injectivity of e, we see ı(.) = 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) ˇ
t
(·
t
) for some ·
t
. Then c(n ˇ(·)) = c(n) cˇ(·) = c(n) ˇ
t
b(·) =
c(n) ˇ
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 ﬁnite 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
= ı
kl
in this
case). Then (¸
k
) ı (i
k
) = id and (i
k
) ı (¸
k
) =
¸
i
¸
¸
¸
is a projection operator. Hence
the following are equivalent: (1) (i
k
) is an isomorphism. (2) (¸
k
) is an isomorphism. (3)
¸
i
¸
¸
¸
= 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 ﬁnitely generated abelian group. Then · ˝
Z
Q is a Qvector space. Show
that its dimension is the rank of ·.
6. Let M
¸
, ¸ ÷ J be submodules of M. The following assertions are equivalent:
(1)
¸
M
¸
is the direct sum of the M
¸
.
(2) For each i ÷ J, M
i
¨
¸,¸=i
M
¸
= {0].
(3) Suppose
¸
.
¸
= 0, .
¸
÷ M
¸
, almost all .
¸
= 0, then .
¸
= 0 for each ¸ ÷ J.
11.2 Exact Sequences
We start with a commutative diagram of modules.
·
o
˛
T
b
ˇ
C
;
·
t
o
/
T
t
b
/
C
t
It yields two derived diagrams (Ke = kernel, Ko = cokernel, Im = Image).
Ke(˛)
o
(1)
Ke(ˇ)
b
=
Ke(;)
Ke(a
t
˛)
o
Ke(ˇ)
b
Ke(;) ¨ Im(b)
(2)
¸
Ko(˛)
o
/
(3)
Ko(ˇ)
b
/
Ko(;)
·
t
Im(˛) ÷Ke(a
t
)
o
/
T
t
Im(ˇ)
b
/
=
¸
C
t
Im(;b)
(4)
¸
280 Chapter 11. Homological Algebra
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
ı : Ke(;) ¨ Im(b) ÷
·
t
Im(˛) ÷Ke(a
t
)
is deﬁned by the correspondence (a
t
)
1
ˇb
1
.
Proof. For : ÷ Ke(;) ¨ Im(b) there exists , ÷ T such that b(,) = :; since : ÷
Ke(;) and ;b = b
t
ˇ we have ˇ(,) ÷ Ke(b
t
); since Ke(b
t
) Ç Im(a
t
), there exists
.
t
÷ ·
t
such that a
t
(.
t
) = ˇ(,). We set ı(:) = .
t
and show that this assignment
is welldeﬁned. If ¯ , ÷ T, b( ¯ ,) = :, then b(, ÷ ¯ ,) = 0; since Ke(b) Ç Im(a),
there exists . ÷ · such that a(.) = , ÷ ¯ ,. We have ˇ(,) ÷ ˇ( ¯ ,) = a
t
˛(.),
because of a
t
˛ = ˇa, and with a
t
( ¯ .
t
) = ˇ( ¯ ,) 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 veriﬁcation 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. Þ
11.2.3 Let (a
t
. b
t
) be exact. Then
·
t
Im(˛) ÷1e(a
t
)
o
/
÷÷
T
t
Im(ˇ)
b
/
÷÷
C
t
Im(;b)
is exact. If, moreover, b is surjective, then (4) is bijective and therefore
Ko(˛)
o
/
÷÷Ko(ˇ)
b
/
÷÷Ko(;)
is exact. If b
t
is surjective, then the derived b
t
is surjective too. Þ
11.2.4 Let (a. b) be exact. Then
Ke(a
t
˛)
o
÷÷Ke(ˇ)
b
÷÷Ke(;) ¨ Im(b)
is exact. If, moreover, a
t
is injective, then (1) is bijective and therefore
Ke(˛)
o
÷÷Ke(ˇ)
b
÷÷Ke(;)
11.2. Exact Sequences 281
is exact. If a is injective, then the derived a is injective too. Þ
11.2.5 Let (a. b) and (a
t
. b
t
) be exact. Then, as we have seen, ı is deﬁned. Under
these assumptions the bottom lines of the derived diagrams together with ı yield
an exact sequence. (See the special case (11.2.6).) Þ
(11.2.6) Kernel–Cokernel Lemma. If in the original diagrama
t
is injective and b
surjective, then (1). (2). (3), and (4) are bijective and the kernelcokernelsequence
Ke(˛)
o
÷÷Ke(ˇ)
b
÷÷Ke(;)
ı
÷÷Ko(˛)
o
/
÷÷Ko(ˇ)
b
/
÷÷Ko(;)
is exact.
Proof. We show the exactness at places involving ı; the other cases have already
been dealt with. The relations ıb = 0 and a
t
ı = 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
) = ˇ(,). Hence : = b(,) ÷ Ke(;) by commutativity, and ı(:) = .
t
.
Suppose : ÷ Ke(;) is contained in the kernel of ı. Then there exists , such
that : = b(,). ˇ(,) = a
t
(.
t
) and ˇ(:) = ˛(.) ÷ Im(˛). Then b(, ÷ a(.)) = :
and ˇ(, ÷a(.)) = ˇ(,) ÷ˇa(.) = ˇ(,) ÷a
t
˛(.) = ˇ(,) ÷a
t
(.
t
) = 0. Hence
, ÷ a(.) is a preimage of :.
We nowrelate the Kernel–Cokernel Lemma to the Five Lemma (11.2.7); see also
(11.1.4). Given a commutative ﬁveterm diagram of modules and homomorphisms
with exact rows.
·
˛
o
T
ˇ
b
C
;
c
D
ı
d
1
e
·
t
˛
/
T
t
ˇ
/
C
t
;
/
D
t
ı
/
1
t
We have three derived diagrams.
0
Ke(ı)
¯
d
D
d
ı
1
e
·
˛
o
T
b
Ko(˛)
¯
b
0
0
Ke(ı
t
)
D
t
ı
/
1
t
,
·
t
˛
/
T
t
Ko(˛
t
)
0,
0
Ko(˛)
ˇ
¯
b
C
;
c
Ke(ı)
¯
d
0
0
Ko(˛
t
)
ˇ
/
C
t
;
/
Ke(ı
t
)
0.
282 Chapter 11. Homological Algebra
The rows of the ﬁrst 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(ı) ÷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 ﬁveterm 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(ı) = 0. Ko(J) = 0 = Ko(
¯
J) = 0.
(5) a surjective, b. J injective = c injective. (Here the 1part 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 1modules. 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
T
¡
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 deﬁnition 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 socalled 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
ﬁxed ring 1. A family ·
+
= (·
n
[ n ÷ Z) of modules ·
n
is called a Zgraded
module. We call ·
n
the component of degree or dimension n. One sometimes
considers the direct sum
n÷Z
·
n
; then the elements in ·
n
are said to be ho
mogeneous of degree n. Typical examples are polynomial rings; if kŒ.. , is the
polynomial ring in two indeterminates .. , of degree 1 say, then the homogeneous
polynomials of degree n are spanned by .
i
,
ni
for 0 _ i _ n, and in this manner
we consider kŒ.. , as a graded kmodule (actually a graded algebra, as deﬁned
later). One can also consider formal power series; this would correspond to taking
the product
n÷Z
·
n
instead of the sum.
Let ·
+
and T
+
be Zgraded modules. A family ¡
n
: ·
n
÷T
n÷k
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
n1
, called boundary operators or differentials, is said to be a chain
complex, if for each n ÷ Zthe boundary relation d
n1
ıd
n
= 0 holds. We associate
to a chain complex C

the modules
7
n
= 7
n
(C

) = Ker(d
n
: C
n
÷C
n1
).
T
n
= T
n
(C

) = Im(d
n÷1
: C
n÷1
÷C
n
).
H
n
= H
n
(C

) = 7
n
,T
n
.
We call C
n
(7
n
, T
n
) the module of nchains (ncycles, nboundaries) and H
n
the nth homology module of the chain complex. (The boundary relation dd = 0
implies T
n
Ç 7
n
, and therefore H
n
is deﬁned.) Two nchains 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
ı ¡
n
= ¡
n1
ı 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

).
284 Chapter 11. Homological Algebra
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
n1
(C
t
), also called boundary
operator, which is induced by the correspondence ¡
1
n1
ı J
n
ı g
1
n
.
C
n
¿
n
d
n
C
tt
n
÷ :
tt
:
t
÷ C
t
n1
(
n1
C
n1
(11.3.1) Lemma. For a cycle :
tt
÷ C
tt
n
with preimage : under g
n
the relation
g
n1
J
n
: = J
tt
n
g
n
: = J
tt
n
:
tt
= 0 and exactness shows that there exists :
t
with
J
n
(:) = ¡
n1
(:
t
). The assignment :
tt
÷:
t
induces a welldeﬁned homomorphism
d
n
: H
n
(C
tt
) ÷H
n1
(C
t
).
Proof. The relation¡
n2
J
t
n1
:
t
= J
n1
¡
n1
:
t
= J
n1
J
n
: = 0andthe injectivity
of ¡
n2
show that :
t
is a cycle. If we choose another preimage : ÷¡
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
deﬁned. 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
n1
(C
t
) ÷
is exact.
Proof. The boundary operator J
n
: C
n
÷C
n1
induces a homomorphism
J
n
: 1
n
= C
n
,T
n
÷7
n1
.
and its kernel and cokernel are H
n
and H
n1
. By (11.2.3) and (11.2.4) the rows of
the next diagram are exact.
1
t
n
(
n
d
/
n
1
n
¿
n
d
n
1
tt
n
d
//
n
0
0
7
t
n1
(
n1
7
n1
¿
n1
7
tt
n1
The associated sequence (11.2.6) H
t
n
÷H
n
÷H
tt
n
J
÷÷H
t
n1
÷H
n1
÷H
tt
n1
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
n÷1
of homomorphisms which satisfy
J
n÷1
ı s
n
÷s
n1
ı c
n
= g
n
÷ ¡
n
.
(This deﬁnition has two explanations; ﬁrstly, one can deﬁne “chain homotopy” in
analogy to the topological deﬁnition by using the chain complex analogue of the
unit interval; secondly, it codiﬁes 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
n÷1
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
n÷1
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 1modules. Let G be another 1module.
We apply the functor Hom
T
(÷. 1) to C

and obtain a chain complex C

= (C
n
. ı
n
)
of 1modules with C
n
= Hom
T
(C
n
. 1) and the 1linear map
ı
n
: C
n
= Hom
T
(C
n
. 1) ÷Hom
T
(C
n÷1
. 1) = C
n÷1
deﬁned by ı
n
(ç) = (÷1)
n÷1
ç ı d
n÷1
for ç ÷ Hom(C
n
. 1).
For the choice of this sign see 11.7.4. The reader will ﬁnd 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
. ı
n
) is a Z
graded module (C
n
[ n ÷ Z) together with homomorphisms ı
n
: C
n
÷ C
n÷1
,
called coboundary operators or differentials
1
, such that ı
n÷1
ı
n
= 0. We set
7
n
= Ker ı
n
. T
n
= Imı
n1
. H
n
= 7
n
,T
n
and call C
n
. 7
n
. T
n
the module of ncochains, ncocycles, ncoboundaries, and
H
n
the nth 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.
286 Chapter 11. Homological Algebra
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
n1
such that J
T
n1
ıJ
T
n
= 0. Anatural transformation ç
+
: J
+
÷G
+
between such functors is a family of natural transformations ç
n
: J
n
÷ G
n
such
that J
G
n
ç
n
= ç
n1
J
T
n
. A natural chain homotopy s
+
: ç
+
. [
+
from ç
+
to [
+
is
a family s
n
: J
n
÷G
n÷1
of natural transformations such that
J
G
n÷1
ı s
n
÷s
n1
ı J
T
n
= [
n
÷ ç
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 Zbasis of J
n
(X). A natural transformation ç
n
: J
n
÷
G
n
from a free functor J
n
into another functor G
n
is then determined by the values
ç
n
(b
n,}
) and the family of these values can be ﬁxed 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 ¨ ç : H
0
ı J
0
÷ H
0
ı G
0
there exists a natural transformation
ç
+
: J
+
÷ G
+
which induces ¨ ç. Any two natural transformations ç and [ with
this property are naturally chain homotopic ([57]).
Proof. We specify a natural transformation ç
0
by the condition that ç(b
0,}
) rep
resents the homology class ¨ çŒb
0,}
. Let now ç
i
: J
i
÷ G
i
be natural transfor
mations (0 _ i < n) such that J
G
i
ç
i
= ç
i1
J
T
i
for 0 < i < n. Consider the
elements ç
n1
J
T
n
b
n,}
÷ G
n1
(T
n,}
). For n = 1 this element represents 0 in
H
0
, by the construction of ç
0
. For n > 1 we see from the induction hypothesis
that J
G
n1
ç
n1
J
T
n
b
n,}
= ç
n2
J
T
n1
J
T
n
b
n,}
= 0. Since G
+
is acyclic we ﬁnd
g
n,}
÷ G
n
(T
n,}
) such that J
G
n
g
n,}
= ç
n1
J
T
n
b
n,}
. We specify a natural trans
formation ç by the conditions ç(b
n,}
) = g
n,}
. This transformation then satisﬁes
J
G
n
ç
n
= ç
n1
J
T
n
. This ﬁnishes the induction step.
Let now ç
+
and [
+
be given. Then [
0
(b
0,}
) ÷ ç
0
(b
0,}
) = J
G
1
c
0,}
for some
c
0,}
, since [
0
(b
0,}
) and ç
0
(b
0,}
) represent the same homology class. We deﬁne
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
n÷1
are given such that J
G
i÷1
s
i
÷s
i1
J
T
i
= [
i
÷ç
i
for 0 _ i < n
11.6. Chain Equivalences 287
(and s
1
= 0). We compute with the induction hypothesis
J
G
n
([
n
÷ ç
n
÷ s
n1
J
T
n
)
= [
n1
J
T
n
÷ ç
n1
J
T
n
÷ ([
n1
J
T
n
÷ ç
n1
J
T
n
÷ s
n2
J
T
n1
)J
T
n
= 0.
Thus, by acyclicity, we can choose c
n,}
÷ G
n÷1
(T
n,}
) such that
J
G
n÷1
c
n,}
= ([
n
÷ ç
n
÷ s
n1
J
T
n
)(b
n,}
).
We now specify a natural transformation s
n
: J
n
÷ G
n÷1
by s
n
(b
n,}
) = c
n,}
. It
then has the required property J
G
n÷1
s
n
= [
n
÷ ç
n
÷ s
n1
J
T
n
.
Problems
1. Let J
0
÷J
1
÷ be a chain complex of free 1modules J
i
and D
0
÷D
1
÷ an
exact sequence of 1modules. A chain map (ç : J
i
÷ D
i
[ i ÷ N
0
) induces a homomor
phism H
0
(ç
·
): 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 (ç
i
) such that H
0
(ç
·
) = ˛. This can
be obtained as a special case of (11.5.1).
The reader should now study the notion of a projective module (one deﬁnition 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
n1
. J
C(
(,. .) = (J
1
, ÷¡ .. ÷J
1
.).
This can also be written in matrix form
_
,
.
_
÷
_
J
1
¡
0 ÷J
1
__
,
.
_
.
288 Chapter 11. Homological Algebra
The suspension †1 of 1 is deﬁned by (†1)
n
= 1
n1
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
n÷1
(†1) ÷ H
n
(1) equals H
n
(¡ ), if we use the
canonical identiﬁcations H
n÷1
(†1) Š 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(,)) ÷
1˚1 (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
Ç 1
n
is always a
direct summand. Then 1 is contractible.
Proof. We have the exact sequence 0 ÷ 7
n
÷ 1
n
J
÷÷ T
n1
÷ 0, and since
1 is acyclic we conclude 7
n
= T
n
. Moreover there exists t
n1
: T
n1
÷ 1
n
with dt
n1
= id, since 7
n
is a direct summand of 1
n
. We therefore have a direct
decomposition 1
n
= T
n
˚t
n1
T
n1
. We deﬁne s : 1
n
÷ 1
n÷1
by s[T
n
= t
n
and s
n
[t
n1
T
n1
= 0. With these deﬁnitions one veriﬁes separately on T
n
as well
as on t
n1
T
n1
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) Š H
+
(1), then ¡ is a chain equivalence.
Proof. The exact homology sequence and the hypothesis imply that C¡ is acyclic.
Asubmodule of a free 1module is free. Hence the boundary groups of the complex
C¡ are free, and therefore the exact sequence 0 ÷7
n
÷C¡
n
÷T
n1
÷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 1module · = (·
n
) is said to be of ﬁnite type if the modules ·
n
are
ﬁnitely generated 1modules.
(11.6.4) Proposition. Let 1 be a principal ideal domain. Let C = (C
n
) be a chain
complex of free 1modules such that its homology groups are ﬁnitely generated.
Then there exists a free chain complex D of ﬁnite type which is chain equivalent
to C.
Proof. Let J
n
be a ﬁnitely 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). Deﬁne a chain complex D = (D
n
. J
n
) by
D
n
= J
n
˚G
n1
and J
n
(.. ,) = (,. 0). Then D is a free chain complex of ﬁnite
type and H
n
(D) = J
n
,G
n
Š H
n
(C). Since G
n
is a free submodule of T
n
(C) we
can choose for each n a homomorphismç
n
: G
n
÷C
n÷1
such that c
n÷1
ç
n
(,) = ,
for each , ÷ G
n
. Deﬁne [
n
: D
n
= J
n
˚G
n1
÷ C
n
, (.. ,) ÷ . ÷ç
n1
(,).
One veriﬁes 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 Zgraded 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 deﬁned 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) ı (¡
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. Þ
11.7.2 Graded algebras. A Zgraded 1algebra ·

is a Zgraded 1module
(·
n
[ n ÷ Z) together with a family of 1linear maps
·
i
˝
T
·
}
÷·
i÷}
. . ˝, ÷. ,.
290 Chapter 11. Homological Algebra
The algebra is associative, if always . (, :) = (. ,) : holds, and commutative,
if always . , = (÷1)
[x[[,[
, . holds (sign rule). A unit element 1 ÷ ·
0
of the
algebra satisﬁes 1 . = . = . 1. Let M

= (M
n
) be a Zgraded 1module. A
family
·
i
˝M
}
÷M
i÷}
. a ˝. ÷a .
of 1linear 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 1˝1 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.Þ
11.7.3 Tensor product of chain complexes. Let (·

. J
¹
) and (T

. J
B
) be chain
complexes. Then the graded module ·

˝ T

is a chain complex with boundary
operator J = J
¹
˝1 ÷1 ˝J
B
. Here we have to take the sign rule into account,
i.e.,
J(a ˝b) = J
¹
a ˝b ÷(÷1)
[o[
a ˝J
B
b.
One veriﬁes JJ = 0, using this sign rule. Passage to homology induces
H
]
(·

) ˝H
q
(T

) ÷H
]÷q
(·

˝T

). Œa ˝Œb ÷Œa ˝b.
The tensor product of chain complexes is associative. Þ
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 deﬁne the dual graded 1module by ·
+
n
= Hom
T
(·
n
. 1). We require a
boundary operator ı : ·
+
n
÷ ·
+
n1
on the dual module such that the evaluation
c: ·
+

˝·

÷1
c: ·
+
n
˝·
n
÷1. ç ˝a ÷ç(a).
c = 0 otherwise, becomes a chain map. This condition, c(ç ˝a) = 0 and 11.7.3
yield for ç ˝a ÷ ·
+
n1
˝·
n
0 = Jc(ç ˝a) = c(J(ç ˝a)
= c(ıç ˝a ÷(÷1)
[ç[
ç ˝da
= (ıç)(a) ÷(÷1)
[ç[
ç(da).
i.e., we have to deﬁne ıç = (÷1)
[ç[÷1
ç ı d. Þ
11.7. Linear Algebra of Chain Complexes 291
11.7.5 HomComplex. For graded modules ·

and T

we let Hom(·

. T

) be
the module with
o÷Z
Hom(·
o
. T
o÷n
) as component in degree n. On this Hom
module we use the boundary operator
J(¡
i
) = (d ı ¡
i
) ÷ ((÷1)
n
¡
i
ı d)
for (¡
i
: ·
i
÷ T
i÷n
) , i.e., the acomponent pr
o
(J¡ ) ÷ Hom(·
o
. T
o÷n1
) for
¡ = (¡
o
) ÷ Hom(·

. T

)
n
is deﬁned to be
pr
o
(J¡ ) = d ı ¡
o
÷ (÷1)
n
g
o1
ı d.
One veriﬁes JJ = 0. This deﬁnition generalizes our convention about the dual
module. Þ
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÷[¿[
ı g
I
).
(2) The adjunction
ˆ: Hom(·˝T. C) ÷Hom(·. Hom(T. C)). ˆ(¡
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(ç ˝b)(a) = (÷1)
[o[[b[
ç(a)b. Þ
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 nonzero
groups 1
1
Š 1 with basis e, 1
0
Š 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 deﬁne chain homotopies with the cylinder 1
+
˝C. Note
C
n
˚C
n
˚C
n1
Š (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
n¬1
. The h
I
+
are chain maps (t = 0. 1) and Js
n
(,) =
h
1
n
(,) ÷h
0
n
(,) ÷s
n1
c,, i.e., s
+
: h
1
+
. h
0
+
is a chain homotopy in our previous deﬁnition.
3. Imitate the topological deﬁnition of the mapping cone and deﬁne the mapping cone of a
chain map ¡ : C ÷D as a quotient of 1
+
˝C ˚D. The nth chain group is then canonically
isomorphic to C
n1
˚D
n
and the resulting boundary operator is the one we deﬁned in the
section on chain equivalences. Consider also the mapping cylinder from this viewpoint.
292 Chapter 11. Homological Algebra
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 (Zmodules). 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 sufﬁces to require that J
0
is free. Let J(·) denote the
free 1module generated by the set ·. Denote the basis element of J(·) which
belongs to a ÷ · by Œa. We have a surjective homomorphism ¸: J(·) ÷ ·,
n
o
Œa ÷
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
).
Deﬁne ¡
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 deﬁne 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
) ı 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) Š 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) Š {g ÷ G [ ng = 0] Ç G.
(3) If G is torsion free, then Tor(Z,n. G) = 0.
(4) Tor(Z,m. Z,n) Š Z,J with J the greatest common divisor of m. n.
(5) Adirect sumdecomposition · Š ·
1
˚·
2
induces a direct sumdecomposition
Tor(·. G) Š 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 deﬁne Tor
t
(·. T) = Ker(· ˝Q
1
÷· ˝Q
0
).
(11.8.3) Proposition. There exists a canonical isomorphism
Tor(·. T) Š 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
ˇ
1
1
˝T
;
1
0
˝Q
1
1
0
˝Q
0
1
0
˝T
Tor
t
(·. T)
· ˝Q
1
· ˝Q
0
· ˝T.
294 Chapter 11. Homological Algebra
The Kernel–Cokernel Lemma (11.2.6) yields an isomorphism ı of Tor(·. T) =
Ker(;) with the submodule Tor
t
(·. T) of Coker ˛.
Interchangingthe tensor factors yields anisomorphismTor(T. ·) Š 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 deﬁned in analogy to the functor Tor, the tensor product is
replaced by the Homfunctor.
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 deﬁned 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 welldeﬁned
homomorphismCoker(Hom(i. T)) ÷Coker(Hom(i
t
. T)); in particular we obtain
an isomorphism Ext(·. T) Š 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) Š T,nT.
(3) Ext(Z,n. T) = 0 for T = Q. Q,Z. R.
(4) Ext(Z,m. Z,n) Š Z,(m. n).
(5) Ext(·
1
˚·
2
. T) Š 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
1module C and let · be another 1module. 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
n1
÷ ÷T
1
÷T
0
÷C ÷0
is called an nfold extension of A by C. One can obtain Ext
n
T
(C. ·) as certain
congruence classes of nfold extension of · by C, see [120, Chapter III]. Write
1 1
t
if there exists a commutative diagram
1: 0
C
=
T
n1
T
0
·
=
0
1
t
: 0
C
t
T
t
n1
T
t
0
·
0.
The congruence relation is generated by .
11.9. Universal Coefﬁcients 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 nontrivial abelian group · such that · ˝F = 0 for each ﬁeld F?
11.9 Universal Coefﬁcients
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 Coefﬁcients). Let C be a chain complex of free
modules. Then there exists an exact sequence
0 ÷H
q
(C) ˝G
˛
÷÷H
q
(C ˝G)
ˇ
÷÷Tor(H
q1
(C). G) ÷0.
The sequence is natural in C and G and splits. The homomorphism˛ sends Œ: ˝g
for a cycle : to the homology class Œ: ˝g.
Proof. The sequence 0 ÷7
n
÷C
n
c
n
÷÷T
n1
÷0 is exact; T
n1
is a submodule
of C
n1
and hence free. Therefore the sequence splits and the induced sequence
0 ÷7
n
˝G ÷C
n
˝G ÷T
n1
˝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 Tcomplex 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
i˝1
7
n
˝G
H
n
(C ˝G)
T
n1
˝G
i˝1
7
n1
˝G.
One veriﬁes that the boundary operator (11.3.1) of the homology sequence is i ˝1,
where i : T
n
Ç 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
n1
˝ G ÷ 7
n1
˝ G is Tor(H
n1
(C). G),
because 0 ÷ T
n1
÷ 7
n1
÷ H
n1
(C) ÷ 0 is a free resolution. Let r : C
n
÷
7
n
be a splitting of 7
n
Ç C
n
. Then
7
n
(C ˝G) Ç C
n
˝G
i˝1
÷÷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 coefﬁcient sequence.
296 Chapter 11. Homological Algebra
Let again C = (C
n
. c
n
) be a chain complex with free 1modules 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 Coefﬁcients). There exists an exact sequence
0 ÷Ext(H
n1
(C). G) ÷H
n
(C: G)
˛
÷Hom(H
n
(C). G) ÷0.
The map ˛ sends the cohomology class of the cocycle ç : C
n
÷ G to the homo
morphismH
n
(C) ÷G, Œc ÷ç(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
n1
÷0
and the induced exact sequence
0 ÷Hom(7
n
. G) ÷Hom(C
n
. G) ÷Hom(T
n1
. G) ÷0.
We consider the totality of these sequences as an exact sequence of cochain com
plexes, the 7 and the Tcomplex 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
n1
. G) ÷
which induces a short exact sequence
(4) 0 ÷Ker J
n
˛
÷÷ H
n
(C: G) ÷Coker J
n1
÷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 ç : 7
n
÷G be given. Then J
n
(ç) is obtained as follows: Extend ç to
¯ ç : C
n
÷G. Apply ı and ﬁnd a preimage of ı( ¯ ç) = ¯ çc
n÷1
in Hom(T
n
. G). One
veriﬁes that çi is a preimage.
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 veriﬁes that ˛ is as
claimed in the statement (11.9.2). From the free presentation and the deﬁnition
11.9. Universal Coefﬁcients 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
Ç C
n
. Now consider the
diagram
0
7
n
(Hom(C. G))
Ç
Hom(C
n
. G)
ı
i
·
Hom(C
n÷1
. G)
0
Hom(H
n
(C). G)
Hom(7
n
. G)
i
·
Hom(T
n
. G).
If ç ÷ Ker i
+
, then r
+
(ç) = ç ı r ÷ Ker ı. The splitting is induced by Ker i
+
÷
7
n
(Hom(C. G)), ç ÷çt .
Without going into the deﬁnition of Ext we see from the discussion:
(11.9.4) Proposition. Suppose H
n1
(C) is a free 1module. 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
n1
÷ 7
n1
÷ H
n1
÷ 0 splits and therefore the
cokernel of J
n1
is zero.
Given a cochain complex C

= (C
q
. ı
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 deﬁne d
q
: C
q
÷
C
q1
as ı
q
: C
q
÷C
q÷1
. We can now rewrite (11.9.1):
(11.9.5) Proposition. Let C

be a cochain complex of free 1modules. Then we
have a split exact sequence
0 ÷H
q
(C

) ˝G ÷H
q
(C

˝G) ÷Tor(H
q÷1
(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 ﬁnite type or G is ﬁnitely generated. Then there exists a natural
exact sequence
0 ÷H
]
(C) ˝G ÷H
q
(C: G) ÷Tor(H
q÷1
(C). G) ÷0
and this sequence splits.
Proof. If G is ﬁnitely generated we have a canonical isomorphism of the form
Hom(C. 1) ˝G Š Hom(C. G); we use this isomorphism in (11.9.5). If H
+
(C)
is of ﬁnite type we replace C by a chain equivalent complex C
t
of ﬁnite type (see
(11.6.4)). In that case we have again a canonical isomorphism Hom(C
t
. 1) ˝G Š
Hom(C
t
. G). We apply now (11.9.5) to C
t
.
298 Chapter 11. Homological Algebra
(11.9.7) Proposition. Let ¡ : C ÷ D be a chain map between complexes of free
abelian groups. Suppose that for each ﬁeld 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 coefﬁcient sequence. It implies that
Tor(H
+
(C(¡ )). Z,¸) = 0 for each prime ¸. Hence H
+
(C(¡ )) is torsionfree.
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 Künneth Formula
Let C and Dbe chain complexes of 1modules 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). Œ. ˝Œ, ÷Œ. ˝,.
We use the notation + for Tor
T
. The next theorem and its proof generalizes the
universal coefﬁcient formula (11.9.1).
(11.10.1) Theorem (Künneth Formula). Suppose C consists of free 1modules.
Then there exists an exact sequence
0 ÷
i÷}=n
H
i
(C)˝
T
H
}
(D) ÷H
n
(C˝
T
D) ÷
i÷}=n1
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)
n1
)
and
(7(C) ˝T(D))
n
= Im(1 ˝d: (7(C) ˝D)
n÷1
÷(7(C) ˝D)
n
).
and they imply H(7(C) ˝D) Š 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) Š 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 Künneth 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)
i˝1
Š
7(C) ˝H(D)
(2)
Š
H(C) ˝H(D)
H(T(C) ˝D)
(i˝1)
·
H(7(C) ˝D).
Let us use the notation (·Œ÷1)
n
= ·
n1
for a graded object ·. We tensor the
exact sequence of chain complexes 0 ÷ 7(C) ÷ C ÷ T(C)Œ÷1 ÷ 0 with D
and obtain an exact sequence
0 ÷7(C) ˝D ÷C ˝D ÷(T(C) ˝D)Œ÷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)Œ÷1
(1)
÷÷H(7(C) ˝D)Œ÷1 ÷ .
One veriﬁes that (1) is the map (i ˝1)
+
. Hence we obtain the exact sequence
0 ÷Coker(i
+
) ÷H(C ˝D) ÷Ker(i
+
)Œ÷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 ÷ Œr(c) ˝ Œs(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 coefﬁcient theorem we can rewrite (11.10.1) in
terms of cochain complexes. Under suitable ﬁniteness conditions we can then apply
the result to the dual complex of a chain complex and obtain:
(11.10.2) Theorem(Künneth Formula). Let C and D be free chain complexes such
that H
+
(C) or H
+
(D) is of ﬁnite type. Then there exists a functorial exact sequence
0 ÷
i÷}=n
H
i
(C) ˝H
}
(D) ÷H
n
(C ˝D) ÷
i÷}=n÷1
H
i
(C) +H
}
(D) ÷0
and this sequence splits.
Chapter 12
Cellular Homology
In this chapter we ﬁnally 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 CWcomplex. This chain
complex depends on the skeletal ﬁltration, and the boundary operators of the chain
complex are determined by the socalled 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 satisﬁes the dimension axiom). From
this fact one obtains immediately qualitative results and explicit computations of
homology groups. Thus if X has k(n) ncells, then H
n
(X: Z) is a subquotient
of the free abelian group of rank k(n). A ﬁnite cell complex has ﬁnitely 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
ncells via (ˆ. ç):
e÷T
(D
n
e
. S
n1
e
) ÷ (X. ·). The characteristic map of the
cell e is denoted by (ˆ
e
. ç
e
). The index e distinguishes different copies.
(12.1.1) Proposition. The induced map
ˆ
n
= (ˆ
e
+
):
e
h
+
(D
n
e
. S
n1
e
) ÷h
+
(X. ·)
is an isomorphism.
Proof. By (10.4.6), ˆ
+
: h
+
_
e
(D
n
e
. S
n1
e
)
_
÷ h
+
(X. ·) is an isomorphism.
Now apply the additivity isomorphism
e
h
+
(D
n
e
. S
n1
e
) Š h
+
_
e
(D
n
e
. S
n1
e
_
and compose it with ˆ
+
.
The isomorphism inverse to ˆ
n
is obtained as follows. Given : ÷ h
k
(X. ·).
We use the inclusion ¸
e
: (X. ·) Ç (X. X · e) and the relative homeomorphism
ˆ
e
: (D
n
. S
n1
) ÷ (X. X · e). Let :
e
÷ h
k
(D
n
e
. S
n1
e
) denote the image of :
under
: ÷ h
k
(X. ·)
]
c
·
÷÷h
k
(X. X · e)
ˆ
c
·
÷÷ h
k
(D
n
e
. S
n1
e
) ÷ :
e
.
12.1. Cellular Chain Complexes 301
Then : ÷(:
e
[ e ÷ 1) is inverse to ˆ
n
.
Let X be a CWcomplex. The boundary operator d: h
k÷1
(X
n÷1
. X
n
) ÷
h
k
(X
n
. X
n1
) of the triple (X
n÷1
. X
n
. X
n1
) is transformed via the isomorphisms
(12.1.1) into a matrix of linear maps
m(e. ¡ ): h
k÷1
(D
n÷1
(
. S
n
(
) ÷h
k
(D
n
e
. S
n1
e
)
for each pair (¡. e) of an (n÷1)cell ¡ and an ncell e (as always in linear algebra).
Let 
e,(
be the composition
S
n
(
ç
¸
÷÷X
n
q
c
÷÷X
n
,(X
n
· e)
ˆ
c
÷÷ D
n
,S
n1
.
If we compose 
e,(
with an hequivalence k
n
: D
n
,S
n1
÷ S
n
, then k
n

e,(
has
as a selfmap 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 0cells ç
(
(˙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.
(12.1.2) Proposition. The diagram
h
k÷1
(D
n÷1
(
. S
n
(
)
n(e,( )
J
//
h
k
(D
n
e
. S
n1
e
)
]
·
¯
h
k
(S
n
(
)
t
c,¸
·
¯
h
k
(D
n
e
,S
n1
e
)
is commutative.
Proof. Consider the diagram
h
k÷1
(X
n÷1
. X
n
)
J
/
¯
h
k
(X
n
)
}
h
k
(X
n
. X
n1
)
]
c
·
]
·
¯
h
k
(X
n
,X
n1
)
h
k÷1
(D
n÷1
. S
n
)
ˆ
¸
·
¸
J
//
¯
h
k
(S
n
)
ç
¸
·
¸
h
k
(X
n
. X
n
· e)
]
·
¯
h
k
(X
n
,X
n
· e)
h
k
(D
n
. S
n1
)
ˆ
c
·
¸
]
·
¯
h
k
(D
n
,S
n1
).
ˆ
c
·
¸
Given . ÷ h
k÷1
(D
n÷1
. S
n
). Then ¸
+
m(e. ¡ ). is, by deﬁnition of m(e. ¡ ), the
image of . in
¯
h
k
(D
n
,S
n1
). Now use the commutativity of the diagram.
302 Chapter 12. Cellular Homology
(12.1.3) Corollary. Let o : h
k
(D
n
. S
n1
) ÷ h
k÷1
(D
n÷1
. S
n
) be a suspension
isomorphism. Then m(e. ¡ ) ı o is the multiplication by J(e. ¡ ), provided the
relation d
tt
ı o = k
n
+
ı ¸
+
holds.
We now write the isomorphism (12.1.1) in a different form. We use an it
erated suspension isomorphism o
n
: h
kn
÷ h
k
(D
n
. S
n1
) in each summand.
Let C
n
(X) denote the free abelian group on the ncells of (X. ·). Elements in
C
n
(X) ˝
Z
h
kn
will be written as ﬁnite formal sums
e
e ˝u
e
where u
e
÷ h
kn
;
the elements in C
n
(X) ˝ h
kn
are called cellular nchains with coefﬁcients in
h
kn
. We thus have constructed an isomorphism
¸
n
: C
n
(X) ˝
Z
h
kn
÷h
k
(X. ·).
e
e ˝u
e
÷
e
ˆ
e
+
o
n
(u
e
).
The matrix of incidence numbers provides us with the Zlinear map
M(n): C
n÷1
(X) ÷C
n
(X). ¡ ÷
e
J(e. ¡ )e.
The sum is ﬁnite: J(e. ¡ ) can only be nonzero if the image of ç
(
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
k÷1
(X
n÷1
. X
n
)
J
h
k
(X
n
. X
n1
)
C
n÷1
(X) ˝h
kn
(
n¬1
¸
Æ(n)˝id
C
n
(X) ˝h
kn
(
n
¸
is commutative.
The composition of the boundary operators (belonging to the appropriate triples)
h
n÷1
(X
n÷1
. X
n
)
J
÷÷h
n
(X
n
. X
n1
)
J
÷÷h
n1
(X
n1
. X
n2
)
is zero, because the part h
n
(X
n
) ÷ h
n
(X
n
. X
n1
) ÷ h
n1
(X
n1
) of the ex
act sequence of the pair (X
n
. X
n1
) is “contained” in this composition. We set
h
n,k
(X) = h
n÷k
(X
n
. X
n1
). 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(n÷1)M(n) = 0 follows from (12.1.4) applied to the chain
complex H
,0
(X) obtained from singular homology with coefﬁcients in Z.
12.1. Cellular Chain Complexes 303
The cellular chaincomplexhas its algebraicallydeﬁnedhomologygroups. 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 coefﬁcients
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 socalled spectral sequences [130].
The deﬁnition of incidence numbers uses characteristic maps and a homotopy
equivalence k. These data are not part of the structure of a CWcomplex so that
the incidence numbers are not completely determined by the CWcomplex. The
choice of a characteristic map determines, as one says, an orientation of the cell. If
ˆ. ‰: (D
n
. S
n1
) ÷(X
n
. X
n1
) are two characteristic maps of a cell e, then
‰
1
ˆ: D
n
,S
n1
÷X
n
,X
n
· e ÷D
n
,S
n1
is a homeomorphism and hence has degree ˙1. One concludes that the incidence
numbers are deﬁned up to sign by the CWcomplex.
(12.1.6) Proposition. A cellular map ¡ : X ÷ Y induces a chain map with com
ponents ¡
+
: h
n
(X
n
. X
n1
) ÷ h
n
(Y
n
. Y
n1
). Homotopic cellular maps induce
chain homotopic maps.
Proof. The ﬁrst assertion is clear. Let ¡. g: X ÷ Y be cellular maps and let
ç : X×1 ÷Y. ¡ . g be a homotopy between them. By the cellular approximation
theorem we can assume that ç is cellular, i.e., ç((X × 1)
n
) Ç Y
n
. Note that
(X ×1)
n
= X
n
×d1 LX
n1
×1. We deﬁne a chain homotopy as the composition
s
n
: h
n
(X
n
. X
n1
)
c
÷÷h
n÷1
((X
n
. X
n1
) × (1. d1))
ç
·
÷÷h
n÷1
(Y
n÷1
. Y
n
).
In order to verify the relation ds
n
= g
+
÷ ¡
+
÷ s
n1
d we apply (10.9.4) to
(·. T. C) = (X
n
. X
n1
. X
n2
) and compose with ç
+
.
(12.1.7) Proposition. Let j: k
+
(÷) ÷ l
+
(÷) be a natural transformation be
tween additive homology theories such that j induces isomorphisms of the coefﬁ
cient groups j: k
n
(1) Š l
n
(1), n ÷ Z, 1 a