Fuks Rokhlin

Beginner’s Course
in Topology

Springer-Verlag
Berlin Heidelberg New York Tokyo
 NUNC C O G NOSCO EX PARTE

THOMAS J. BATA LIBRARY

TRENT UNIVERSITY
 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/beginnerscourseiOOOOrokh
Universitext
D.B. Fuks V.A. Rokhlin

Beginner’s Course
in Topology
Geometric Chapters

Translated from the Russian by A. lacob

With 17 Figures

Springer-Verlag
Berlin Heidelberg New York Tokyo 1984
Dmitrij Borisovich Fuks
Laboratory of Bioorganic Chemistry
Moscow State University, Moscow, USSR

Vladimir Abramovich Rokhlin

Department of Mathematics and Mechanics
Leningrad State University, Leningrad, USSR

Andrei lacob
Department of Theoretical Mathematics. The Weizmann Institute of
Science. Rehovot 76100, Israel

Title of the Russian original edition: Nachal’nyj Kurs Topologii: Geome

tricheskie Glavy. Publisher Nauka, Moscow 1977

This volume is part of the Springer Series in Soviet Mathematics

Advisers: L.D. Faddeev (Leningrad), R.V. Gamkrelidze (Moscow)

AMS Subject Classification (1980): 54-01, 54A 05, 54Bxx, 54Cxx,

54Dxx, 54Exx, 55Pxx, 5 5 Qxx, 55Rxx, 57-01, 57M xx, 57Nxx,
57Rxx, 57Sxx, 58Axx, 58C 25, 58D10, 58D15, 58E 05

ISBN 3-540-13577-4 Springer-Verlag Berlin Heidelberg New York Tokyo

ISBN 0-387-13577-4 Springer-Verlag New York Heidelberg Berlin Tokyo

Library of Congress Cataloging in Publication Data. Rokhlin, V. A. Beginner’s course

in topology. (Springer series in Soviet mathematics) (Universitext)
Translation of: Nachal’nyi kurs topologii / V. A. Rokhlin, D. B. Fuks. Bibliography: p.
Includes index. 1.Topology. I. Fuks, D. B. II.Title. III. Series.
QA611.R6513 1984 514 84-10657 ISBN 0-387-13577-4 (U.S.)
This work is subject to copyright. All rights are reserved, whether the whole or part of
the material is concerned, specifically those of translation, reprinting, re-use of illustra­
tions, broadcasting, reproduction by photocopying machine or similar means, and
storage in data banks. Under § 54 of the German Copyright Law where copies are made
for other than private use, a fee is payable to "Verwertungsgesellschaft Wort” , Munich.
© Springer-Verlag Berlin Heidelberg 1984
Printed in Germany
Printing and bookbinding: Beltz Offsetdruck, Hemsbach
2141/3140-543210
 Preface

This book is the result of reworking part of a rather lengthy

course of lectures of which we delivered several versions at the
Leningrad and Moscow Universities. In these lectures we presented an
introduction to the fundamental topics of topology: homology theory,
homotopy theory, theory of bundles, and topology of manifolds. The
structure of the course was well determined by the guiding term
elementary topology, whose main significance resides in the fact that
it made us use a rather simple apparatus. ^n this book we have retained
ihose sections of the course where algebra plays a subordinate role.
We plan to publish the more algebraic part of the lectures as a separate
book.
Reprocessing the lectures to produce the book resulted in the
profits and losses inherent in such a situation:,, the rigour has
increased to the detriment of the intuitiveness, the geometric
descriptions have been replaced by formulas needing interpretations,
etc. Nevertheless, it seems to us that the book retains the main
qualities of our lectures: their elementary, systematic, and
pedagogical features. The preparation of the reader is assumed to be
limited to the usual knowledge of set theory, algebra, and calculus
which mathematics students should master after the first year and a
half of studies. The exposition is accompanied by examples and
exercises. We hope that the book can be used as a topology textbook.
The most essential difference between the book and the
corresponding part of our lectures is the arrangement of the material:
here we have followed a much more orderly succesion of topics. However,
from our experience, a lecture course in elementary topology which
exaggerates in the last respect is rather tedious and less efficient
than one which mixes geometry with algebra and applications. This
remark may serve as a warning to the teacher who would like to use our
book as a guide. In fact, it is by no means necessary to read the book
in its order; a reader who is interested in getting to the homotopy
groups or to any other topic sooner, can easily do so.
 VI

Concerning the terminology and notation, we have tried to

stick to standard usage, and have permitted ourselves only a few
reforms. For example, we do not use the terms "simplicial complexes"
or "CW-complexes", but simplicial spaces and cellular spaces? not
"cofibrations", but Borsuk pairs; not "fiber bundles" (or "fibered
products"), but Steenrod bundles. There is even one term which we do
not use in the generally accepted way: for us, a connected space
refers to what usually is called a linearly connected (or path-con-
nected) space (we do not have a special name for the spaces which are
usually called connected). Furthermore, we have avoided using non­
standardized notations for standard objects. In fact, in the majority
of cases, our notation is just an abbreviation of the corresponding
term and can be understood by itself: for example, pr stands for
projection, in - for inclusion, dim - for dimension, ske - for
skeleton, bs - for base, etc.
Topology requires a very precise set-theoretic language, and
this compelled us to devote a special attention to this language; this
is illustrated by pp. 1 “ 4. we emphasize that on these pages we only
list the terms and notations, assuming that the objects themselves are
known.
In this book we rarely refer to the history of topology. We
have even departed from the tradition that some theorems bear the names
of their real or imaginary authors. In return, we willingly have used
names of topologists in the terminology and notations.
The organization of the text and the system of references may
be briefly described as follows. Each Chapter is divided into Sections,
each Section - into Subsections, each Subsection - into Numbers. The
chapters, sections and subsections have numbers and titles, while the
numbers are denoted only by their numbers. Each fact announced without
proof is called Information, and is distinguished from the rest of the
text by this title. To refer to a section, subsection, or number
within the same chapter, we do not indicate the number of the chapter,
and references within a section or subsection are similarly abbreviated.
Examples: the entries §1.2 (Section 2 of Chapter 1), Subsection 1.2.3
(Subsection 3 of Section 2 of Chapter 1), and 1.2.3.4 (No. 4 of Sub­
section 3 of Section 2 of Chapter 1) are abbreviated, within Chapter 1,
as § 2 , Subsection 2.3, and 2.3.4, respectively; the second of these
entries is abbreviated within §1.2 as Subsection 3; the third entry
is abbreviated within §1.2 and Subsection 1.2.3 as 3.4 and 4,
respectively.

The Authors
 Contents

Set-Theoretical Terms and Notations Used in This

Book, but not Generally Adopted................................. 1

Chapter 1
TOPOLOGICAL SPACES ................................................................................................................ 5

§ 1. FUNDAMENTAL CONCEPTS.......................................

1. Topologies............................................. 5
2. M e t r i c s ............................................... 8
3. S u b s p a c e s ............................................. 10
4. Continuous M a p s ....................................... 11
5. Separation Axioms ..................................... 15
6. Countability Axioms ................................ 19
7. Compactness........................................... 21

§2. CONSTRUCTIONS............................................. 26

1. Sums................................................... 26
2. Products............................................... 27
3. Q u o t i e n t s ............................................. 31
4. G l u e i n g ............................................... 34
5. Projective Spaces .................................... 38
6. More Special Constructions............................ 42
7. Spaces of Continuous M a p s ............................ 46
8. The Case of Pointed Spaces............................. 49
9. E x e r c i s e s ............................................. 55

§3. HOMOTOPIES................................................. 56

1. General Definitions ................................ 56

2. P a t h s ................................................. 60
3. Connectedness and k-Connectedness .................... 61
4. Local Properties...................................... 65
5. Borsuk Pairs........................................... 66
6. CN RS-Spaces........................................... 70
7. Homotopy Propertiesof Topological Constructions ... 72
8. E x e r c i s e s ............................................. 78

Chapter 2
CELLULAR S P A C E S ........................................................................................................................... 81

§1. CELLULAR SPACES AND THEIR TOPOLOGICAL PROPERTIES........... 81

1. Fundamental Concepts.................................... 81
2. Glueing Cellular Spaces from Balls.................... 86
 VIII

3. The Canonical Cellular Decompositions of

Spheres, Balls, and Projective Spaces................ 88
4. More Topological Properties of Cellular Spaces . . . . 89
5. Cellular Constructions ............................... 94
6. Exercises.............................................. 98

§2. SIMPLICIAL SPACES.......................................... 99

1. Euclidean Simplices................................... 99
2. Simplicial Spaces and Simplicial Maps................. 101
3. Simplicial Schemes .................................... 105
4. Polyhedra.............................................. 106
5. Simplicial Constructions ............................. 107
6. Stars. Links. Regular Neighborhoods................... 114
7. Simplicial Approximation of Continuous Maps........... 118
8. Exercises.............................................. 119

§3. HOMOTOPY PROPERTIES OF CELLULAR SPACES ................... 120

1. Cellular P a i r s ........................................ 120

2. Cellular Approximation of Continuous Maps............. 123
3. k-Connected Cellular Pairs ........................... 126
4. Simplicial Approximation of Cellular Spaces........... 130
5. Exercises.............................................. 132

Chapter 3
SMOOTH MANIFOLDS ............................................... 133

§1. FUNDAMENTAL CONCEPTS ...................................... 133

1. Topological Manifolds................................. 133

2. Differentiable Structures............................. 141
3. Orientations.......................................... 150
4. The Manifold of Tangent Vectors....................... 156
5. Embeddings, Immersions, and Submersions............... 161
6. Complex Structures .................................... 166
7. Exercises.............................................. 171

§2. STIEFEL AMD GRASSMAN MANIFOLDS ........................... 171

1. Stiefel Manifolds...................................... 171

2. Grassman Manifolds .................................... 177
3. Some Low-Dimensional Stiefel
and Grassman Manifolds ............................... 185
4. Exercises.............................................. 186

§3. A DIGRESSION: THREE THEOREMS FROM CALCULUS ............... 187

1. Polynomial Approximation of Functions................. 187

2. Singular Values........................................ 191
3. Nondegenerate Critical Points......................... 194

§4. EMBEDDINGS. IMMERSIONS. SMOOTHINGS. APPROXIMATIONS . . . . 197

1. Spaces of Smooth Maps.................................. 197

2. The Simplest Embedding Theorems....................... 200
3. Transversalizations and Tubes......................... 202
4. Smoothing Maps in the Case of Closed Manifolds . . . . 205
5. Glueing Manifolds Smoothly ........................... 208
 IX

6. Smoothing Maps in the Presence ofa Boundary.......... 214

7. General Position...................................... 219
8. Maps Transverse to a Submanifold..................... 225
9. Raising the Smoothness Class of aManifold............ 228
10. Approximation of Maps by Embeddings
and Immersions....................................... 233
11. E x e r c i s e s ............................................ 237

§5. THE SIMPLEST STRUCTURE THEOREMS........................... 239

1. Morse F u n c t i o n s ...................................... 239

2. Cobordisms and Surgery................................ 243
3. Two-dimensional Manifolds ........................... 255
4. E x e r c i s e s ............................................ 263

Chapter 4
BUNDLES.......................................................... 265

§1 . BUNDLES WITHOUT GROUP STRUCTURE........................... 265

1. General Definitions................................... 265

2. Locally Trivial Bundles............................... 267
3. Serre Bundles......................................... 269
4. Bundles With Map Spaces asTotal Spaces................ 273
5. Exercises............................................. 276

§2. A DIGRESSION: TOPOLOGICAL GROUPS AND

TRANSFORMATION GROUPS..................................... 276

1. Topological Groups ................................... 276

2. Groups of Homeomorphisms............................. 282
3. Actions............................................... 285
4. Exercises............................................. 296

§3. BUNDLES WITH A GROUP S T R U C T U R E ........................... 297

1. Spaces With F-Structure............................... 297

2. Steenrod Bundles ..................................... 299
3. Associated Bundles ................................... 304
4. Ehresmann-Feldbau Bundles............................. 308
5. Exercises.............................................. 311

§4. THE CLASSIFICATION OF STEENROD BUNDLES ................... 311

1. Steenrod Bundles and Homotopies....................... 311

2. Universal Bundles..................................... 316
3. The Milnor B u n d l e s ................................... 319
4. Reductions of the StructureGroup...................... 321
5. Exercises.............................................. 323

§5. VECTOR B U N D L E S ........................................... 324

1. General Definitions................................... 324

2. Constructions......................................... 331
3. The Classical Universal Vector Bundles ............... 336
4. The Most Important Reductions of
the Structure Group................................... 34 3
5. Exercises.............................................. 345
 X

§6. SMOOTH B U N D L E S ...........................................

1. Fundamental Concepts ................................. 346

2. Smoothings and Approximations......................... 350
3. Smooth Vector Bundles................................. 353
4. Tangent and Normal Bundles ........................... 359
5. D e g r e e ............................................... 364
6. Exercises............................................. 371

Chapter 5
H0M0T0PY GROUPS................................................. 373

§1. THE GENERAL T H E O R Y ....................................... 373

1. Absolute Homotopy Groups............................. 37 3

2. A Digression: Local Systems ......................... 378
3. Local Systems of Homotopy Groups
of a Topological Space............................... 380
4. Relative Homotopy Groups............................. 38 4
5. A Digression: Sequences of Groups and
Homomorphisms, and ^-Sequences....................... 390
6. The Homotopy Sequence of a P a i r ..................... 398
7. The Local System of Homotopy Groups of
the Fibers of a Serre Bundle......................... 402
8. The Homotopy Sequence of a SerreB u n d l e ............. 4 05
9. The Influence of Other Structures
Upon Homotopy Groups................................. 4 11
10. Alternative Descriptions of the Homotopy Groups . . . 416
11. Additional Theorems ................................. 420
12. E x e r c i s e s ........................................... 423

§ 2. THE HOMOTOPY GROUPS OF SPHERES AND

OF CLASSICAL MANIFOLDS ................................... 424

1. Suspension in the Homotopy Groupsof Spheres ......... 4 24

2. The Simplest Homotopy Groups ofSpheres................ 429
3. The Composition Product............................... 4 33
4. Information: Homotopy Groups ofSpheres................ 43 6
5. The Homotopy Groups of Projective
Spaces and Lenses..................................... 438
6 . The Homotopy Groups of Classical Groups............... 439
7. The Homotopy Groups of Stiefel
Manifolds and S p a c e s ................................. 44 1
8 . The Homotopy Groups of Grassman
Manifolds and S p a c e s ................................. 443
9. Exercises.............................................. 444

§3. HOMOTOPY GROUPS OF CELLULAR SPACES ....................... 445

1. The Homotopy Groups of One-dimensional

Cellular Spaces....................................... 445
2. The Effect of Attaching Balls......................... 446
3. The Fundamental Group of a Cellular Space............. 448
4. Homotopy Groups of Compact Surfaces................... 45 1
5. The Homotopy Groups of Bouquets....................... 45 3
6 . The Homotopy Groups of a k-Connected
Cellular Pair......................................... 45 5
7. Spaces With Prescribed Homotopy Groups ............... 459
8 . Eight Instructive Examples ........................... 45O
9. Exercises.............................................. 4^2
 XI

§4. WEAK HOMOTOP Y EQUIVALENCE................................. 463

1. Fundamental Concepts ................................ 463

2. Weak Homotopy EquivalenceandConstructions............ 468
3. Cellular Approximations ofTopological Spaces......... 472
4. Exercises............................................. 478

§5. THE WHITEHEAD PRODUCT..................................... 478

1. The Class w d ( m , n ) .................................. 478

2. Definition and the SimplestProperties
of the Whitehead P r o d u c t ............................. 482
3. Applications......................................... 484
4. Exercises............................................. 486

§6. CONTINUATION OF THE THEORY OF BUNDLES..................... 487

1. Weak Homotopy Equivalence andSteenrod Bundles . . . . 487

2. Theory of Coverings.................................. 49 0
3. Orientations......................................... 500
4. Some Bundles Over Spheres............................ 501
5. Exercises............................................. 503

BIBLIOGRAPHY ................................................... 5 06
INDEX............................................................ 508
GLOSSARY OF SYMBOLS............................................. 517
 Set-Theoretical Terms and Notations Used in this
Book, but not Generally Adopted

Mathematicians manage with a surprisingly modest collection

of set-theoretic terms and notations, which can be roughly divided into
three groups. The first contains terms and notations which have attained
general recognition. The terms and notations in the second group are
equally well-known, but can be understood differently or have varying
connotations. The third group consists of terms and notations used
less frequently.
There is no need todefine terms from the first group. For
example, the notations X U Y, X fl Y, and X, x .. . x x for the
I n
union, intersection, and product of sets, or the notations f: X + Y,
Imf andfI : A -* Y for a map, its image, and its restriction are
i^
understood in the same way by all people. The same is true for the
notation x e X and the terms one-to-one (injective) map and map onto
(surjective map).
For the sake of precision, we must say a few words about our
usage of terms and notations from the second group. We denote the empty
set by 0. We understand the notation X c Y in the broadest sense,
i.e., the equality X = Y is not excluded. The same is true for the
term countable set: we use it both for infinite countable and finite
sets. The identity map of the set X is denoted by id X , or, when
there is no ambiguity about X, simply by id. We shall say that a map
is invertible if it has an inverse, i.e., it is simultaneously injective
and surjective. We let {x E X | ...} denote the set of points x of
the set X which satisfy the condition appearing instead of the three
dots. A family -*-s a maP a set M onto a set of objects
X^ with y £ M, defined by the formula p ^ X ^ .
Next our main task is to list the terms and notations
appearing in this book and belonging to the third group.
 2

Maps

If A is a subset of a set X, then the inclusion of A in

X may be considered as the map defined by the formula xx . We
denote it by in: A -> X. If there isno ambiguity about A and X,
we simply write in.
If A is a subset of X and B is a subset of Y, then
eachmap f: X Y such that f (A) ^ B induces a map ab f : A -* B,
x h- f (x) , and called here the abridgment (or compression) of the map
f to A,B.When there is no ambiguity about A and B, one can write
ab finstead of ab f : A -> B . If B = Y, then ab f is just the
usual restriction of f to A.
By a map of a sequence (X,A ^ ,...,A^) into a sequence
(Y,B1 ,.../Bn ) , where (B^, ...,B^) are subsets of X
(respectively, Y), we mean a sequence of maps

(tp: X Y , tp-Li A 1 -s- tpn : An -►Bn )

such that cp^ = ab cp . Wedenote such a map by

(cp/cp^f •••/cp^): (X ,A^ ,. ..,A^) ->(Y ,B^ , . ..,B^ ) .

If thesubsets A, ,...,A and Bn,... ,B are fixed, then the map

I n 1 ' n ^
f = (cp,cp^,. ../<pn ) and its first component cp are uniquely determined
by each other and usually we do not distinguish between them. For
example, the notation f: (X,A 1 ,...,An ) ^ (Y,B^,...,B^) may be also
used to say that f is a map of X into Y such that f(A^) c= B ^ ,
...,f(An ) c: B^. When we wish to emphasize explicitly this relationship
betweenf and cp, we shall write:

f = rel cp , cp = abs f .

Sometimes we simply write rel instead of rel id .

Quotients

We denote thequotient (orfactor) set ofa set X by a

partition p of X by X / p . The map X+ X/p which takes each point
into the element of p containing it is called projection and is
denoted by pr. A subset of X which is a union of elements of the
partition is said to be saturated. The smallest saturated set
containing a given subset A of X (i.e., pr-1 (pr(A))) is called
 3

the saturation of A.
If p and q are partitions of X and Y, respectively,
then each map f: X + Y which transforms elements of p into elements
of q induces a map X/p Y/q, which takes each element A of p
into the element of q which contains f(A). This map is denoted by
fact f . In particular, factf is defined when q is the partition
of Y into its points, and f is constant on the elements of p.
Thus, for each map f : XY constant on the elements of the partition
p of X we have the corresponding map fact f : X/p Y.
Given a map f: X -+ Y, the partition of X into the nonempty
preimages of the points of Y is denoted by zer(f). The corresponding
map fact f : X/zer(f) Y is injective and is called the injective
factor (or injective quotient) of the map f.

Sums

The sum of the family of sets {x } is the union of

---------------- 1-------- y y EM
disjoint copies of the sets X , i.e., the set of pairs (x,y) such
that x € X . Notation: I I ___X The map of X N (v E M) into
y y -1— LyEM y v
1 ! _ x , defined by the formula x » (x,v), is denoted by in .
-— LyEM y J V
We note that the maps in^ are injective and their images
in^(Xv ) are pairwise disjoint and cover J LyeMxy - Therefore, for any
family (Y } indexed by the same set M, and each family of maps
J y yEM

{V \ " Vv€M' th0re 1 3 3 UniqUS maP f: ^ € M X M-— -y£MYy WhlCh

satisfies the relations f ° iny = iny ° f. f iscalled the sum of
the maps f and is denoted by J *
If M consists of the numbers l,...,n, we use, along with
| Ix and I If , the notations
J— L y J J- V

Products

The product X 1 x ... * Xn is mapped naturally onto its

factor X± following the rule

<X 1 " x i- V
This map is called the i-th projection and is denoted by pr± .
Given maps f-^ X± Y ^ . . . , ^ : Xn + Yn , the rule
defines a map of the product X^ * ... x xn into the product
Y^ x ... x Yn' caH ed the product of the maps and denoted

If p is a partition of the set X and q is a partition

of the set Y, we let p x q denote the partition of the product
X x y into the sets A x b, where A is an element of p and B is
an element of q .
There is a natural map of X into its square X x x, given
by x^ (x,x). This is the diagonal map and is denoted by diag. Its
image diag(X x x) c X x x is called the diagonal of X x x.
 Chapter 1. Topological Spaces

§1. FUNDAMENTAL CONCEPTS

1. Topologies

1. We say that a topological structure or, simply, a topology,

is defined on a set X if there is given a class of subsets of X which
contains the union of any collection in the class and the intersection
of any finite collection in the class. A set endowed with a topological
structure is called a topological space, its elements - points, and the
sets of the given class - open sets.
A collection of sets for which we take the union or the
intersection may be empty. The union of the empty collection is 0,
while the intersection of the empty collection of subsets of X is the
entire set X. Hence 0 and X are open sets (in any topology).
Two examples of topological structures are: the trivial
topology, whose only open sets are 0 and X, and the discrete
topology, in which all the subsets of X are open. If X has more
than one element, it is possible to define other topologies on X. For
example, if X consists of two elements, a and b, then it will admit
two topologies aside from the trivial and discrete ones. In one, the
open sets are 0, a, and X, while in the second the open sets are
0, b, and X. More serious examples will appear in the sequel.

2. A subset of a topological space is closed if its complement

is open. The class of closed sets contains the intersection of any
collection of sets from the class, and it contains the union of any
finite collection of sets from the class. Moreover, given any class of
subsets of a set X with these properties, there exists a unique
topology on X such that the given class is the class of all closed
 6

sets.
3. A neighborhood of a point of a topological space is any
open set containing the given point. A neighborhood of a subset of a
topological space is any open set containing the given subset.

Derived Concepts

4. If we consider the open sets contained in a given subset

A of a topological space X, there is one which is the largest, namely
the union of all such sets. It is called the interior part or, simply,
the interior of the set A. We denote it by IntA or Int^A.
Similarly, among the closed sets containing A, there is one which is
the smallest, namely the intersection of all such sets. It is called
the closure of the set A and is denoted by Cl A or Cl A. The
— x
difference CIA \ IntA can be represented as the intersection of the
closed sets CIA and X \ IntA, and is therefore closed. This set
is called the boundary or frontier of the set A and is denoted by
FrA or Fr A. We remark that X \ IntA = Cl(X \ A) and that A
and X \ A have the same boundary.

5. Relative to the set A, the points of the sets IntA ,

Cl A , Fr A , and X \ Cl A = Int(X \ A) are called interior, adherent
boundary (or frontier) and exterior points, respectively. They can be
characterized more explicitly in terms of neighborhoods. A point is:
an interior point if it has a neighborhood entirely contained in A;
an adherent point if each of its neighborhoods intersects A;
a boundary point if each of its neighborhoods intersects both A and
X \ A; and an exterior point if it has a neighborhood which does not
intersect A.
Clearly, a set is open (closed) if and only if it coincides
with its interior (respectively, closure), i.e., if it consists only of
interior points (respectively, if it contains all its boundary points).

6. A subset A of a topological space X is said to be

dense in X (or everywhere dense) if CIA = X, i.e., if A
intersects any nonempty open set in X. A set A is nowhere dense
if X \ Cl A is everywhere dense.

Bases and Prebases

7. A base of a topological space is a collection of open

 7

sets such that any open set can be represented as a union of sets from
this collection. Equivalently, a collection r of open sets is a base
if for any open set U and any point x £ U there is V G F such that
x € V c U.
A base completely determines the topology: the open sets are
exactly those which can be expressed as union of elements of the base.

8. The following proposition provides us with a standard

method of introducinga topology on a set. Let T be a collection of
subsets of a set X. Then there exists a topology on X with base r
if and only if the intersection of every finite subcollection of sets
from F can be expressed as a union of sets from r.

The intersection of a finite collection of sets belonging to

a base is open; hence the necessity of the condition. Sufficiency
follows from the fact that the class of subsets of X which are
representable as unions of sets from T satisfies the conditions of
Definition 1.

9. The previous proposition can be reformulated in the

following useful way. There is a topology on X with base T if and
only if the sets of T cover X, and for any U,V £ r and any point
x G U n V, there exists W £ T such that x £ W c:U fl V.

10. A collection of subsets of a topological space is said

to be a prebase of the space if the intersections of finite
subcollections of sets from the given collection form a base.
The proposition in 8 shows that any collection r of subsets
of a set X is the prebase of a unique topology on X.

11. A base at the point x of a topological space X is a

collection of neighborhoods of x such that any neighborhood of x
contains a neighborhood from this collection. A prebase at the point x
is a collection of sets such that the intersections of finite
subcollections form a base at x.

Covers

12. As a rule, the covers we shall encounter will be either

covers of a topological space by some of its subsets, or covers of a
subset of a topological space by other subsets of this space. If we
need to emphasize that a certain cover of a subset A of a topological
space X consists of subsets of X which are not necessarily included
in A, we shall refer to it as a cover of the set A in X.
 8

A cover r is a refinement of a cover A if any element of

T is contained in an element of A.
A cover is locally finite if any point of the space has a
neighborhood which intersects only a finite number of elements of the
cover.
A cover is open (closed) if all its elements are open
(respectively, closed) sets.
13. Every open cover of a topological space X has a
refinement whose sets belong to a given base of X.

For example, the sets of the base contained in the sets of

the given cover yield such a refinement.

2. Metrics

1. A nonnegative real function p defined on the square

X x x of the set X is a metric on X if it satisfies three
conditions: p(x,y) =0 if and only if x = y; p(x,y) = p(y,x) for
all x,y E X; p(x,z) £ p(x,y) + p(y,z) for any x, y,z E X . A metric
space is a set equipped with a metric. We use the symbol dist as the
standard notation for metrics.
The values taken by a metric are called distances, and the
inequality they satisfy according to the definition is the triangle
inequality.

2. The standard n-dimensional Euclidean space ]Rn (n ^ 0)

is the fundamental example of metric space. lRn is the set of all
sequences °f real numbers, where the distance between two

sequences ix j_^ and ^i^ is defined as “ Y • The

1
line IR is usually identified with the field of real numbers, denoted
by HR.
We obtain the definition of the standard Hilbert space t
by replacing the n-term sequences ix j_^ with infinite sequences

{x ^ ^ satisfying the condition x^ < and writing instead

rn
of 2,-| i-n the distance formula.

3. The ball with center Xq £ X and radius r > 0 in the

metric space X is the set of points x e X such that dist(xQ ,x) < r.
If we write < (respectively = ) instead of £, we obtain the
definition of the open ball (respectively, of the sphere). The unit
 9

ball and the unit sphere of IRn , i.e., the ball and sphere with center
(0 P. ..,0) and radius 1, are simply called the n-dimensional ball Dn
and the (n-1)-dimensional sphere Sn ^ . In particular, is just a
point, S - a pair of points, and S_1 = 0. Moreover, we set Dn = 0
for n < -1 and Sn = 0 for n £ -2.

4. By definition, the distance between two sets, A and B,

is the number l n f X£ A y£B dist(x,y), and we denote it by Dist(A,B).

In particular, if a is a point, Dist(a,B) = inf _ dist(a,y).

y eB
The diameter of the set A is the number SUPX dist(x,y),

denoted by diamA . A set is bounded if its diameter is finite.

The Metric Topology

5. As a consequence of the triangle inequality, if the open

ball with center at x^ and radius r contains a point x^, then it
also contains the open ball with center at x^ and radius
r - dist (Xg ,x,j ) . Therefore, in any metric space, the intersection of
two open balls contains, together with each point, some open ball
centered at that point. Moreover, since the open balls cover the space,
they constitute the base of a certain topology (see 1.9). In this way,
every metric space becomes a topological one.
The resulting topology is called the metric topology. From
now on we shall tacitly regard the metric spaces as topological spaces,
having in mind the metric topology. In particular, this refers to nRn
and ¿2*
A topological space whose topology is the metric topology
relative to some metric is said to be metrizable.

6. Clearly, the open balls centered at a given point of a

metric space constitute a base at that point. The part of this base
consisting of the balls of radii 1/n (n = 1,2,...) is also a base.

7. If A is a subset of a metric space X, its metric

neighborhood of radius r > 0 is, by definition, the set of all points
x e X such that Dist(A,x) < r. Since this set is the union of all
open balls of radius r centered at the points of A, it is open,
i.e., a genuine neighborhood of A.
 10

3. Subspaces

1. We shall now discuss the relative topology, which

transforms any subset A of atopological space X into an independent
topological space. This topology is defined by taking its open sets
to be those of the form A fl B, where B is an open subset of X. It
is evident that all the conditions of Definition 1.1 are satisfied.
Moreover, the closed subsets of A in this topology are exactly the
intersections A fl B where B is a closed subset of X. The subsets
of the space X, equipped with the relative topology, are called
subspaces of X.

If X is a topological space and A is a subspace of X,

the pair (X,A) is called a topological pair. A topological triple
is a triple (X,A ,B ) consisting of a topological space X and two
subspaces A, B of X, such that B <= A.

2. Let A be a subspace of X. It is clear thatany subset

of A which is open or closed in X has the same property in A. If
A is open, then every set open in A is also open in X. If A is
closed, then any set closed in A is also closed in X. In any case,
if B <= A <= X, then
Cl B = (Cl B) flA.
.A. X
Obviously, if T is a base (prebase) of the space X, then
the sets A fl B with BET yield a base (respectively, prebase) of
the space A.
As a direct consequence of its definition, the relative
topology is transitive: if B is a subset of the subspace A of X,
the topologies induced on B by the inclusions B c= A and B <= X
coincide.

3. If X is a metric space and A is a subset of X, then

the restriction of the function dist to A x A isclearly a metric
on A. Consequently, any subset of a metric space is itself a metric
space. In addition, it is obvious that the metric topology of the
latter coincides with the relative topology induced on A by the
metric topology of the ambient space X.

4. The previous constructions greatly increase the supply

of nontrivial examples of topological spaces: we can now include all
the subsets of nRn and In particular, the balls and spheres of
]Rn are topological spaces. Here we shall add only the cubes of IRn ,
defined by inequalities of the form ou £ x i £ + a (i = 1 ,...,n),
with real numbers a-,...,a ,a, and a>0. If a* = ... = a = 0
1 n 1 n
 11

and a = 1, the cube is called the unit cube ln . The unit segment
1 .
I is also denoted by I.

Fundamental Covers

5. A cover Г of a topological spaceX is fundamental if

eachsubset A of X such that А П В is open in В for all В € Г
is itself open. Equivalently, Г is fundamental if each subset A of
X such that А П В is closed in В for all В E Г is itself closed.
Obviously, a cover which admits a fundamental refinement is
itself fundamental.

6. All open covers, and all finite or locally finite closed

covers are fundamental.

Clearly, the claim is true for both open and finite closed
covers. Now suppose that Г is a locally finite closed cover of a
space X. Consider a cover Д of X consisting of open sets which
intersect only a finite number of elements of Г. Since Д is
fundamental, it is enough to check that, given any set U £ Л, the
cover of U with elementsU П В, В £ Г, is fundamental. But this
results from the fact that the latter cover is finite and closed.

7. A triple (X,A ,В ), where X is a topological space and

A and В are subsets of X which constitute a fundamental cover of
X, is termed a triad. If IntA U Int В = X, or ifA U B = X with
A and В closed, then(X,A,B) is a triad.

4. Continuous Maps

1. A map f of a topological space X into a topological

space Y is continuous if the preimage of each open subset of Y is
open in X. Equivalently, f is continuous if the preimage of each
closed set is closed.
A map f: (X ,A 1 ,...,An > -* (Y ,B1 , ...,B^) , where A 1 , ...,Ar

and B 'i'--e/Bn are sut)sets of the spaces X and Y, respectively,

is said to be continuous, if the map absf : X + Y is continuous.

A useful comment: in order for a map X + Y to be continuous,

it is enough that the preimages of the sets comprising some prebase of
Y be open.
 12

2. If f: X + Y and g: Y Z are continuous, then the

composition g ° f :X Z is obviously continuous. Trivially, the
identity map id X :X X is continuous for any topological space X.
According to the definition of the relative topology, if
f : X ->Y is continuous and A c X, B c Y are subsets with f (A) a B,
then the map ab f :A -> B is also continuous. In particular, the
restriction f |A : A^ Yof a continuous map f: X Y to an arbitrary
subset A of Xis continuous. For example, the inclusion of a sub­
space into its ambient space is always continuous.
If a compression ab f of the map f: X -> Y is defined on
the entire space X, then its continuity is equivalent to the
continuity of f. In particular, f: X -+■ Y is continuousif and only
if the map f: X f (X) is continuous.

3. It is clear that if F is a fundamental cover of X,

then a map f: X Y is continuous whenever all the restrictions
A £ T, are continuous. An equivalent formulation: let F be a
fundamental cover of the topological space X, and assume that for
each A£ T there is a continuous map f : A + Y, such that
fA (x) =fg(x) for all x G A n B (A ,B £ F); then the map f: X + Y
defined by

f (x) = fA (x) for x G A (A £ F)

is continuous.

4. A continuous map is open if the images of the open sets

are open, and closed if the images of the closed sets are closed.
Obviously, a composition of open maps is open, and a
composition of closed maps is closed.
Here we note one useful sufficient condition for a map to be
open: f: X + Y is certainly open if for each x £ X there is a
neighborhood U of f(x) and a continuous map g : u + X such that
x ^x x
the composition f ° g coincides with in: U Y. Proof: if this is
x x
-1
the case, then f (A) = U g (A) for any subset A of X.
X t-X X

Continuity at a Point

5. A map f : X ->Y is continuous at the point x £ X if

for any neighborhood V of the point f(x) thereis a neighborhood U
of x such that f(U) c v.
One can reformulate this definition using fewer open sets.
 13

In fact, let us assume that, along withthe map f: X Y, we are

given an arbitrary prebase at the point x G X,A, and an arbitrary
prebase at the point f(x) £ Y, E. Then one readily sees that f is
continuous at x if and only if each neighborhood V E E contains the
image of some neighborhood U £ A.
When X and Y are metric spaces, and A and E consists
of open balls centered at the points x and f(x), respectively, the
last statement reduces to the usual numerical formulation given in
calculus: the map f : X -* Y is continuous at the point x £ X if for
each e > 0 there is 6 > 0 such that distx (x,xf) < 6 implies
disty (f(x) ,f (x')) < £.

6. A map f : X -* Y iscontinuous if and only if it is

continuous at each point of X.

PROOF. If f is continuous and V is a neighborhood of the

-1
pointf(x), then f (V) is a neighborhood of the point x, and
f(f- 1 (V)) c V.
If fis continuous at each point and V is open in Y,
then each point of the set f (V) is an interior point, since it has
a neighborhood whose image lies in V.

Homeomorphisms and Embeddings

7. Iff : X Y is an invertible continuous map, the inverse

map f ^: Y X is not necessarily continuous. For example, consider
the identity map of a set with the discrete topology onto the same set,
but equippedwith a different topology; its inverse is not continuous.
-1
An invertible map f such that both f and f are
continuous is a homeomorphism. If there is a homeomorphism X ^ Y,
the space Y is said to be homeomorphic to the space X.
The following maps are obviously homeomorphisms: the identity
transformation of a space, the map inverse to a homeomorphism, and the
composition of two homeomorphisms. Thus, the homeomorphism of spaces
is an equivalence relation.

8 (EXAMPLES). The open ball Int Dn is homeomorphic to HRn .

The standardhomeomorphism lRn Int Dn is given by the formula

f 2x arctan (dist (0 ,x)) /tt dist (0 ,x) , if x 0,

x ^ <
0, if x = 0.
 14

The cube in is homeomorphic to Dn ; its interior Int in is

homeomorphic to Int Dn , and its boundary Fr I is homeomorphic to
Fr Dn , i.e., to Sn" 1. The standard homeomorphisms Dn + in ,
Int Dn Int In , and Fr Dn -+ Fr in are realized by translation with
the vector (ort^ + ... + ort^J/2, followed by central projection
(here ort^,...,ortn denote the vectors (1,0,...,0),...,(0,...,0,1)).

The punctured sphere Sn (i.e., sn with one point removed)

is homeomorphic to IRn . A homeomorphism IRn Sn ^ ort^ is given by
the composition of the homeomorphism {x^,...>xn ) ** ,...>xn } °f
IRn onto a subspace of lRn+^ with the stereographic projection, i.e.,
the central projection of this subspace onto Sn \ ort^ from the point
ort^ .

9. A map f: X + Y is an embedding or, more specifically,

a topological embedding, if ab f : X + f(X) is a homeomorphism. For
example, the inclusion of a subspace in its ambient space is an embed­
ding .

Retractions

10. A retraction is a continuous map of a space X onto a

subspace A which is theidentity map on A. A subset onto which a
space can be retracted is a retract of the space.
Each point of a topological space is a retract of this space.
However, a pair of points is already not necessarily a retract. For
example, a segment cannot be retracted onto its boundary since any such
retraction would be a real continuous function taking two values but no
intermediate ones.

11. A subspace A of a topological space X is a retract

of X if and only if every continuous map A Y can be extended to
a continuous map X Y, for any topological space Y.

PROOF. Ifp: X + A is a retraction and f: A + Y is

continuous, then the composition f ° p extends f to X.
If every continuous map A + Yextends to a continuous map
X + Y, then extending the identity map A + A to acontinuous map
X A yields a retraction.
 15

Numerical Functions

12. The well-known theorem of calculus asserting that the

arithmetic operations performed upon continuous functions again produce
continuous functions is obviously true for the numerical functions
defined on an arbitrary topological space. Similarly, the theorem
asserting the continuity of the limit of a uniformly convergent sequence
of continuous functions holds for numerical function on a topological
space.

13. I_f X is a metric space and A is a subset of X,

then the function X + ]R, x ^ Dist(x,A), is continuous.

PROOF. Let x,y £ X and z G A. Then Dist(x,A) £ dist(x,z)

£ dist(x,y) + dist(y,z). Hence Dist(x,A) £ dist(x,y) + Dist(y,A) for
any x,y £ X, and since x and y appear symmetrically, we obtain
|Dist(x,A) - Dist(y,A) | £ dist(x,y).

14. A subset A of a topological space X is said to be

distinguishable if there is a continuous function f : X -* I such that
f (x) = 0 for x € A and f(x) > 0 for x £ X \ A. Any function with
this property is said to distinguish the set A.
A distinguishable set is obviously closed. It is also clear
that any closed subset of a metric space is distinguishable: for
example, the function x *+ min(1,Dist(x,A)) distinguishes the closed
subset A.

5. Separation Axioms

1. In this subsection and the two that follow, we formulate

additional restriction which are often imposed on a topological
structure in order to bring the properties of the corresponding topolo­
gical space closer to those that characterize the subsets of the spaces
]Rn .
2. More than ten "separation axioms" are known. We need the
following four.
T^. Given two arbitrary points a and b, a f b, there is a
neighborhood of a which does not contain b. Equivalent formulations:
each point is a closed set; finite sets are closed.
. Two arbitrary distinct points have disjoint neighborhoods.
 16

. Any point and any closed set not containing this point
have disjoint neighborhoods. An equivalent formulation:
every neighborhood of an arbitrary point contains the closure of a
neighborhood of this point.

. Any two disjoint closed sets have disjoint neighborhoods.

An equivalent formulation: every neighborhood of an arbitrary closed
set contains the closure of a neighborhood of this set. Another
equivalent formulation: given an arbitrary finite collection of pairwise
disjoint closed sets, there are neighborhoods of these sets with
pairwise disjoint closures.

3. Axiom T^ is a consequence of ^2 ' kut simple examples

show that it is not a consequence of or .
Spaces which satisfy axiom are called Hausdorff, those
which satisfy the axioms T^ and - regular, and those which satisfy
the axioms and - normal.
Every normal space is regular, and every regular space is
Hausdorff.

4. Obviously, every subspace of a Hausdorff space is Hausdorff,

every subspace of a regular space is regular, and every closed subspace
of a normal space is normal.

INFORMATION. A nonclosed subspace of a normal space is not

necessarily normal; see [11].

5. Every retract of a Hausdorff space is closed.

PROOF. Let A be a retract of a topological space X and

let p: X + A be a retraction . If b E X \ A, then since X is
Hausdorff and p is continuous, the points b and p(b) have disjoint
neighborhoods, U and V, such that 0(U) c: V. This implies that
P (x ) i x for x e U, i.e., U n A = 0.Thus any point which is not
contained in A is an exterior pointfor A.

6. Every metric space is normal.

PROOF. Clearly, every metric spacesatisfies axiom T ^ . Let

us verify T4 . Suppose A and B are disjoint closed subsets of a
metric space, and set U = {x | Dist(x,A) < Dist(x,B)} and
V = {x | Dist (x,B ) < Dist (x,A) }.Since Dist(x,A) and Dist(x,B)
depend continuously on x (see 4.13), U and V are open. Trivially,
U H V= 0, A C U, and B <= V.
 17

Urysohn Functions

7 (LEMMA). Let A and B be two closed subsets of a

topological space X. Let F be the collection of all neighborhoods
of A which do not intersect B, and let A be the set of dyadic
rational numbers of the interval I (i.e., the numbers m/ 2 q with
arbitrary nonnegative integers m,q satisfying m ^ 2^). _lf X is
normal, then there exists a mapping cp: A -+■ F such that

Cl cp(r ^) c cp(r ^) for r 1 < r2 . (1)

PROOF. Set cp(1 ) = X \ B and let cp(0) be any neighborhood

of A which is contained, together with its closure, in X \ B (see
the second formulation of axiom ). If cp(r)is already defined in
such a way that (1) holds for the numbers r = m/2^ £ A with q = n,
ri.
we can extend the definition to r = m/2 £ A, keeping (1) valid:
n+ 1
if r = m/2 £ A with odd m = 2k + 1 , we take cp(r) to be any open
set containing Clcp(k/2n ) and contained along with its closure in
cp( (k+ 1)/2n ) . This induction yields a mapping cp: A T with property
(1 ) .

8. Given two arbitrary disjoint closed subsets A and B of

a normal space X, there is a continuous function X I, equal to 0
on A and equal to 1 on B.
PROOF. Using the previous lemma and its notations, define a
function f : X -* I by the formula

inf{r | cp(r) 3 x}, if x £ cp(1 ) ,

f(x) =
1, if x € X \ cp(1 ) .

It is evident that f is equal to 0 on A and to 1 on

B. To show that f is continuous, note that the intervals [0,r) and
(r , 1 ] with r £ A constitute a prebase of the segment I, and that

f- 1 ([ 0 ,r )) = Ur ,<r cp(r'), while f” 1 ([0,r]) = nr ,>r cp(r'). Using

property (1), we see that the last intersection is just nr .>r Clcp(r’);

hence f 1 ((r,1]) = X \ f 1 ([ 0 ,r]) = X v. nr ,>r Clcp(r') . Therefore, the

intervals [0 ,r) and (r,1 ] have open preimages, and f is continuous.

9. A continuous function f: X -* I such that f (x) = 0 for

x £ A x and f (x) = 1 for x £ B c X is referred to as a Urysohn
function for the pair A, B .
 18

A Urysohn function for a pair A, B may also take the value

0 outside of A. However, if A is distinguishable, f is any Urysohn
function for the pair A, B, and g distinguishes A, then
x min(f(x)+g(x),1) provides a Urysohn function for A, B which is
positive outside A.
One may note that the proof of Theorem 8 does not use axiom
T^ and conclude that this theorem is true for any T^-space. The
converse is also true: if any pair of disjoint closed subsets of X
admits a Urysohn function, then X satisfies axiom T^.
Finally, we remark that one can prove a slightly stronger
version of Theorem 8, by composing theUrysohn function with a linear
transformation t >+ a + (b-a)t, where a and b are arbitrary real
numbers. The composition is a continuousfunction X [a,b], equal
to a on A and to b on B.

Extension Theorems

10 (LEMMA). Let F be a closed subset of the topological

space X, and let <j): F ]R be a continuous function bounded in
absolute value by a number L > 0. If X is normal, then there exists
a continuous function ip: X IR such that

|^(x)| £ L/3 for x E X,

and (2)

|i);(x) ~ <Mx) | £ 2L/3 for x E F.

PROOF. The subsets of F determined by the inequalities

<f>(x) £ -L/3 and cf>(x) ^ L/3 are closed in F, and hence in X, and
disjoint. Therefore, there is a continuous function \p: X *> [-L/3,L/3],
equal to -L/3 on the first set, and equal to L/3 on the second
(see 9). It is clear thatif; satisfies the requirements (2) .

11 . If_ A is a closed subset of thenormal space X, then

every continuous function A HR extends to a continuous function
X + IR. This claim remains true if one takes an interval instead of the
real line JR.

First, let us show that every continuous map f of A into

an interval can be extended to a continuous map g of X into the same
interval. Without loss of generality, we may take the interval [-1,1].
Define g as the sum of a series of continuous functions q■ X ^ IR
which satisfy the conditions
 19

|gk (x)| £ 2k 1/3k , if x G X, (3)

and
|f(x ) “ lo 9j_(x)| < (2/3)^, if x £ A. (4)

The functions gk are constructed inductively: take gQ = 0

and, assuming that gg,...,g are already constructed and satisfy (3)
and (4) for kg n, define 9n+-| to be the function obtained when
one applies the previous lemma to <j) = f - £]? (g. I ) , F = A, and
u 11 A
L = (2/3) . Inequality (3) shows that the series converges
uniformly on X, and hence its sum g is a continuous function (see
4.12). Inequality (4) implies that g|A = f.

To prove the first part of the theorem, notice that IR is

homeomorphic to the open interval (-1,1). We have just showed that
given any continuous function g: A (-1,1), there exists a continuous
g: X -> [-1,1] such that g(x) = f (x) for all x € A. Let
-1 -1
B - g (-1) U g (1). The sets A and B are closed and disjoint;
hence the pair A, B has a Urysohn function. If we multiply the latter
by g, we get the desired extension X (-1,1) of the function f.

12. I_f A is a closed subset of the normal space X, then

every continuous map A IRn extends to a continuous map X IRn . This
claim remains true if one takes a cube instead of IRn .

To see this, it suffices to apply Theorem 12 to the coordinate

functions of the given map A -*■lRn .

6. Countability Axioms

1. A topological space is said to satisfy the second axiom

of countability (or to be a second countable space) if it has a
countable base. A topological space is said to satisfy the first axiom
of countability (or to be a first countable space) if it has a countable
base at each point. A topological space is separable if it has a
countable dense subset.

2. The second axiom of countability implies the first axiom

of countability and the separability. A metric space is always first
countable, and is second countable if and only if it is separable.

It is immediate that a second countable space is first

countable, and 2.6 shows that every metric space is first countable.
 20

To produce a contable dense set in a space with countable base, just

pick a point in each set of a given countable base. Given a separable
metric space, the open balls centered at the points of a countable dense
set and with radii 1/n (n = 1,2,...) constitute a countable base.

3. nRn and t 2 are separable, and hence have a countable

base.
The collection of all sequences with rational x^'s
is a countable dense set in IRn . The set of all finitely supported
(i.e., having only a finite number of nonzero terms) sequences r 1 00
{x^^
with rational x.'s is countable and dense in
i 2
4. Every subspace of a second countable space is second
countable. In particular, all subspaces of !Rn and t ^ have countable
base.

Indeed, a countable base of the space induces a countable base

of each of its subspaces; see 3.2.

5. In a separable space, every collection of pairwise

disjoint open subsets is countable.

In fact, let S be a countable dense subset of the given

space. For any set in the given collection, pick a point of S
contained in this set. This yields an injective mapping of the given
collection into S.

6. Obviously, a continuous surjective map of topological

spaces carries every dense set into a dense set. It is also clear that
an open map between two topological spaces transforms each base into a
base, and each base at a point into a base at the image of this point.
Therefore, the image of a separable space under a continuous map is
separable, and the image of a first or a second countable space under
an open map is first, respectively second countable.

7. Every regular second countable space is normal.

PROOF. Let A and B be closed disjoint subset of a regular

second countable space. According to the second formulation of axiom
T3 , each point of any of the sets A and B has a neighborhood whose
closure does not intersect the other set. Picking such neighborhoods,
we get open covers of A and B, and we may assume that these covers
are countable; if not, we may refine them by covers made of sets
belonging to a countable base (see 1.13). Let us index these two
covers, writing them as and V^V^..., and then set

uI’
I = UII \ U?I CIV.-L and Vn’ = Vn \ U1?
I Cl U.1 . The sets U = U?1 U*n and
 21

oo
V = U1 Vn are ° P en and clearly disjoint. Since Cl LL fl B = 0 and
Cl n A = 0, we have U => A and V => B.

Embedding and Metrization Theorems

8. Every regular second countable space can be embedded in

V
PROOF. Let X be a regular space with countable base F.
We index the pairs (U,V) , U,V € F, satisfying Cl U <= v, writing
them as a sequence (U^V^, (U2,V2),--- Now define f: X ■> ¿ 2 by

the rule f (x) = {k % ^ ( x ) } ^ f where (f)^ is anarbitrary Urysohn

function for the pair Cl Uk ,X ^ Vk (see 7) . If x f y, then there
exists an index k such that x E U^., y € X \(indeed, X is
regular), and so f is injective. We show next that f is an
embedding.
Since the (f^'s are continuous, given Xq £ X, e > 0 , andn,
there is a neighborhood U of x^ such that ^ (|(f)^(x) - ^ ( X q )|/k)2 <

<c ^ / 2 for all x £ U. Choosing n such that £^+1 ^ ^ < ' we

see that dist(f (x^) , f (x)) <e, and so f is continuous.

Let g denote the inverse of ab f :X f(X) . Given a point

y^ £ f(X)and a neighborhood U of g(yQ>, find n such that
g (yQ ) € Un and vn c ^ ^ ^ f (X ) and dist(yQ,y) < 1/n, then
clearly |4>n (g (y) ) “ (Yq )) I < ^ ' which in turn implies that
g(y) £ V. In conclusion, for y £ f (X)and distfy^y) < 1/n, we have
g(y) £ U,proving the continuity of g.

9. A second countable topological space is metrizable if and

only if it is regular.

The necessity of this condition is contained in 5.6, and its

sufficiency - in 8.

7. Compactness

1. A topological space is compact if any of its open cover

contains a finite cover. For example, a finite set endowed with an
arbitrary topology is compact, whereas an infinite set endowed with the
discrete topology is not compact.
It is clear that a subspace A of a topological space X is
 22

compact if only if from each open cover of A in X one can extract

a finite cover.
2. Every closed subset of a compact space is compact.

PROOF. Let A be a closed subset of the compact space X,

and let A be a cover of A in X. We add the set X ^ A to A,
extract a finite cover from the resulting open cover of X, and then
delete the set X \ A from the latter, if it still remains. This
obviously yields a finite cover of A in X which is contained in A.

3. In a Hausdorff space, any two compact disjoint sets have

disjoint neighborhoods.

PROOF. Let A and B denote the given sets. If B is a

point, then for each point x £ A consider disjoint neighborhoods
and V of x and B, and extract a finite cover U,...,U from
x x1 xs
the open cover of A given by the neighborhoods ; then
U^ ^ and fi^ ^ are disjoint neighborhoods of the setA and the

point B. In the general case, pick for each x £ B disjoint

neighborhoods and of A and of x, and then extract a finite
cover V ,...,V from the resulting cover of B by neighborhoods V ;
X1 xs x
s s
then and are disjoint neighborhoods of A and B.
i i
4. Every compact subset of a Hausdorff space is closed.

Indeed, from 3 we see that a point which is not contained in

a given compact subset of a Hausdorff space has a neighborhood which
does not intersect this subset.

5. Every compact Hausdorff space is normal.

This is a consequence of 2 and 3.

Compactness and Fundamental Covers

6. Suppose A is a compact subset of a T^-space X. Then

from every countable fundamental cover of X one can extract a finite
cover of A .

PROOF. Let be the given cover. If none of the

sets ll™ Ui covers A, pick a point from each set A \ U™ LL , and
denote the set of all these points by Y. It is obvious that Y is
infinite and that each intersection Y n LK is finite. The latter
shows that Y and all its subsets are closed; hence Y is at the same
 23

time compact (see 2) and discrete. But this contradicts the fact that
Y is infinite.

7. The countability assumption in Theorem 6 is essential.

For example, the cover of a segment by all its countable subsets is
fundamental, but one cannot extract from it a countable cover.

Compactness and Maps

8. The image of a compact space under a continuous map is

compact.

PROOF. Let f be a continuous map of the compact space X

onto a topological space Y, and let A be an open cover of Y. The
-1
sets f (V) , VGA, form an open cover of X, and clearly a subcover
U 1,...,U of this cover yields a subcover f(U ),...,f(U ) of A.
i s I s
9. Every continuous map of a compact space into a Hausdorff
space is closed.

This is a corollary of 2, 8, and 4.

10. Every invertible continuous map of a compact space onto

a Hausdorff space is a homeomorphism. Every injective continuous map
of a compact space into a Hausdorff space is an embedding.

These are consequences of proposition 9 and of the obvious

fact that a closed invertible map is a homeomorphism.

Compactness and Metrics

11. Every compact subset of a metric space can be covered by

a finite number of open balls having radius e, for any positive e.

Such a cover can be extracted from any cover consisting of

balls of radius e.
1 2. Every compact metric space has a countable base.

To obtain such a base, it suffices to construct, for each

positive integer n, a finite cover of open balls of radius 1/n, and
then take the union of these covers.

1 3. Every compact metric space is bounded.

This is a consequence of 11.

 24

14. Let X be a compact topological space. Then every

continuous function X + IR attains its absolute maximum and absolute
minimum.

PROOF. Propositions 8 and 13 show that the image of X in

HR is bounded. Proposition 9 shows that this image is closed, which in
turn implies that it contains its adherent points, including its
greatest lower bound and its least upper bound.

15. Let A and B be disjoint subsets of a metric space.

If A is compact and B is closed, then Dist(A,B) > 0.

PROOF. Since A is compact and Dist(x,B) depends

continuously on x E A (see 4.13), there exists a £ A such that
Dist(a,B) = inf^^Dist (x ,B) = Dist(A,B ) (see 14). Since B is closed

and a t B, Dist(a,B) > 0, and thus Dist(A,B) > 0.

16. Suppose that f is a continuous map of a metric space

X into a topological space Yand A is an open cover of Y. I_f X
is compact, then there is e >0 such that for any set A c: xwith
diameter diamA < e, f(A) is contained in some element of A.

It is enough to show that there is an e > 0 such that any

two points x,y E X with dist(x,y) < e are both contained in one of
the sets of the cover T = f (A). For each x E X, pick a ball
centered at x and contained in one of the sets of V, and let U be
x
the concentric ball with half the radius. Now extract a finite cover
U ,...,U from the cover of X by the balls U . Let e. denote
X1 X S X I
the radius of U , and let e = min{e ,...,e }. If x,y £ X and
X 1i i s
dist(x,y) < e, then distix^x) < and dist(x^,y) < dist(x^,x) +
+ dist(x,y) < 2ei for some i. Therefore, x and y belong to one
and the same set of the cover r.

Compactness in Euclidean Space

17. The cubes of lRn are compact.

PROOF. Obviously, any cube in IRn can be divided into 2n

cubes of half the edge, and if some cover r of the original cube by
open subsets of 3Rn does not contain a finite subcover, then it retains
the same property as a cover of one of the smaller cubes. An iteration
of this argument yields a decreasing sequence of cubes Q.|,Q2 ,...,
each of them being half the size of the preceding one, and such that
 25

none of them is covered by a finite collection of sets from I1.But

the point common to all these cubes is certainly covered by some set
fromT, which must also cover all the cubes with k large enough.

18 ♦ A subset of IRn is compact if and only if it is bounded

and closed.

The necessity of these conditions is implied in 13 and 4. The

sufficiency is a consequence of 17 and 2, since any bounded subset of
]Rn is contained in some cube.

Local Compactness

19. A topological space is locally compact if each of its

points has a neighborhood with compact closure.
Compact spaces are obviously locally compact. The most
important examples of noncompact, locally compact spaces are IRn with
n > 0.

20. Every closed subset of a locally compact space is locally

compact.

Indeed, if a is a point of a closed subset A of the

locally compact space X, and U is a neighborhood of a in X with
compact closure Cl U, then U H A is a neighborhood of a in A
x
with compact closure Cl (U H A). [Cl (U D A), being closed in X,
is closed in the compact subset Cl U of X, and hence is compact;
x
see 2].

21. Every open subset of a locally compact Hausdorff space

is locally compact.
PROOF. Let a be a point of the open subset A of the
locally compact space X, and let U be a neighborhood of a in X
with compact closure clxu - Since the space Clxu is regular (see 5),
a has a neighborhood V in Cl^U such that Cl^jj^ U H A. We show
X
that V is a neighborhood of a in A with compact closure C1AV *

The set V is open in ClvU, hence in U fl A, which in

turn implies that V is open in A. To verify that C1AV is compact,
note that the closure Clcl yV is compact (see 2) and contained in
X
U n A. This implies that it equals ClunAV and that the latter is
closed in A (see 4), which finally shows that ClunAV =C1AV.
 26

22. Let U be a neighborhood of the point a of the

locally compact space X. If_ X is Hausdorff, then a has a
neighborhood whose closure is compact and contained in U.
PROOF. Since U is a locally compact space (see 21), a has
a neighborhood V in U with compact closure. Since U is open, V
is open in U. Finally, since is compact and X is Hausdorff,
Clr7V is closed in X (see 4) and thus it coincides with Cl V. We
U X
see that V is the desired neighborhood of a.

23. Locally compact Hausdorff spaces are regular.

This is a consequence of 22.

Information: Paracompactness

24. A Hausdorff space is paracompact if each of its open

covers has a locally finite refinement. The compact Hausdorff spaces
are (obviously) paracompact, and so are all the metric spaces. All
paracompact spaces are normal. For details, see [11].
One can show that a paracompact space which can be covered by
open metrizable sets is metrizable; see [13].

§2. CONSTRUCTIONS

1. Sums

1. The sum J LyGMx u of a family ix^} of topological

spaces becomes a topological space if we declare a subset to be open
if its preimages under all maps inv : x v ^ I lx are open.
Equivalently, a subset of J [x is closed if its preimages under all
maps in are closed. It is evident that each map in : X I lx
v V V -J L 1J
is an embedding and that the images in^ix^) are both open and closed
in | IX .
—1 V
Let {Yy } be another family of topological spaces, indexed
by the same set M, and let fy : X y be continuous maps. Then
the map J !^y: -I ^ J LYy is obviously continuous.

2. If all spaces Xy satisfy one of the axioms T^, , T^,

 27

or T4 , then their sum satisfies the same axiom. The same hold for
the first axiom of countability, and also for the properties of local
compactness and metrizability [if X are metric spaces, one can define
a metric on J j_X^ by the formulas: dist (in^ (x) ,in^ , (x 1) ) = 1 when
v i v1 or when v = v’ and dist(x,x') ^ 1 ; dist (inv (x) ,inv (x1 )) =
= dist (x,x1) when dist(x,x') < 1]. If each X^ has a countable base
and M is countable, then J |_X has a countable base too. Similarly,
when M is countable and each X^ is separable, J [x is separable
too. Finally, J [x is compact whenever all X are compact and M
is finite.

2. Products

1. Let X^,...,Xn be topological spaces. We define a

topology on X^ x ... x x^ by taking as a base the collection ofall
sets U.x...xU
c X. x ... x x , where U. is open in X.,
n 1 1 n 1 r 1
i = 1,...,n. The conditions of 1.1.8 are satisfied by virtue of the
relation (U' x ... x U ’) n (U" x ... x U") = (U' flU") x ... x (u' il U").
1 n 1 n 11 n n
The resulting topological space is called theproduct of the spaces
X 1 ,...,Xn .
If A^ ,...,An are subspaces of X ^ , . . . then the topology
of the product A^ x ... x is obviously identical with the topology
induced by the inclusion ... x A <= X„ x ... x x .
A„ x
n 1 1 n
Actually, we have met with some products already. Indeed,
]Rn is the product of n copies of the real line HR,while ln is
the product of n copies of the unit segment I.
The product X x I, where X is a topological space, is
known as the cylinder over X.

2. Forming the product of spaces is a commutative and

associative operation: there are obvious canonical homeomorphisms
X^ x x 2 X 2 x X^ and (X^ x X2) x X^ + X^ x (X2 x X^) .Similarly,
(X, x ... x x n. , ) V

Moreover, sums and products of spacessatisfy thedistributive

law:there is a canonical homeomorphism X x (J__X Yy* '

3. If the sets ,...,An are open in X^,...,Xn , then

A = A 1 x ... x An is open in X 1 x... x Xr . If A 1 ,...,An are closed,
then A is closed. In all cases, Cl A = Cl A^ x ... xCl A^ .

The first statement is a direct result of the definition of

 28

the topology of x ... x x , while the second is a consequence of

the third; so let us verify the third statement. A point (x^,...,x^) G
£ X^ x ... x X^ is an adherent point of A if and only if each of its
base neighborhoods x ... x has a nonempty intersection with A,
i.e., if and only if for any neighborhoods U^,...,Un of the points
x lf...,xn , LL il A i f 0, i — 1 , .. .,n . That is to say, (x^...,*^
is an adherent point of A if and only if x^ is an adherent point
of A^, i = 1,. ..,n.Therefore, Cl A = Cl A^ x ... x cl A^ .

4. We note that the projections pr^: X^ x ... x Xr X^ are

continuous and open for any topological spaces X ^ ,.../xn -

The sets of the form x® x ... x x ? ^ x x^ x x? + ,| x ... x ,

x° GX. (j f i) are called the fibers of the product X. x ... x x .
j j J c 1 n
Clearly, the restriction of pr^: X^ x .. . x x^ -+ X^ to any fiber
x? X ... X
X X. X x°x ° X ... X x° is a homeomorphism. Hence the
1 i- 1 i i+1 n ^
fibers are canonically homeomorphic to the factors of the product.
For any map ... x x , where Y, X .,..., X
f: Y ^ X. x are
n 1
I n
arbitrary sets, we have the corresponding maps pr^ ° f: Y X^;
conversely, given arbitrary f^: Y X^, there is a unique map
f: Y X. x ... x x such that p r .° f = f.. Clearly, if Y,X. , . ..,X
1 n ^ l i 1 1 n
are topological spaces, then f is continuous if and only if all the
maps pr^ © f are continuous.
In particular, it follows that the map diag: X -> X x x is
continuous for every topological space X. We make the (obvious) remark
that the diagonal diag(X) is closed if and only if X is Hausdorff.

5. Obviously, every product f^ x ... x fn : X^ x ... x x^

Y. x ... x y of continuous maps f .: X „ Y ., ...,f : X Y is
1 n ^ 1 1 1 n n n
continuous. Moreover, f^ x ... x f^ is open whenever are
open.

6 . If X is a metric space, then clearly

(dist (x^j ,x£) - dist(x 1 ,x2)| £ dist (x jj,x ^ ) + dist(x 2 ,x2>for any points
x^,x 2 ,x^,x2 G X. This inequality shows that the function
dist: X x X IR is continuous.

Properties of Products

7. Every product of T^-spaces is a T -space. Every product

of Hausdorff spaces is Hausdorff. Every product of regular spaces is
regular.
 29

The first and second assertions are immediate. We show that

a product of T3-spaces X 1 ,...,Xn is a T 3 ~space. Let U be a
neighborhood of (x^,...»x n)
in X^ x ... x x . Pick neighborhoods
U ,...,U of the points x.,...,x , such that U„ x ... x u c: u,
i n I n 1 n
and fix neighborhoods V.),.. .,Vn of the same points with Cl V 1 c U ,
■■•/ClVn cUn . Since C1(V 1 x ... x Vn ) = Cl V 1 x ... x cl Vn (see 3),
one has Cl (V. x ... x v ) c: U .
\ n
INFORMATION. There are product of normal spaces which are not
normal; see [14].

8. If
S 1f...,S are dense sets in the spaces X „ ,...,X ,
I n c 1 n
then S1 x... x sn is obviously densein X^ x ... x x . Consequently,
a product of separable spaces is separable.

If , . . . , are bases of X ^ , . . . then the sets

U 1 x ... xu with U £ T EF Uform a base of the space
1 n 1 n 1 n ^
X 1 x ... x x . Consequently, a product of second countable spaces is
second countable.

If now
r„,...,r are bases of X„ ,...,X at the points
i n I n
x.,...,x ,then the sets U. x ... x u with U. £ T U £ F ,
i n 1 n l i n n
form a base of X^ x ... x x^ at the point (x^ ,...,x^) . Consequently,
a product of first countable spaces is first countable.

9. Every product of metrizable spaces is metrizable.

In fact, we can say more: if X ^ ,...,XR are metric spaces,

then the formula dist ((x^ , ...,x^) , (x^ ,...,x^) ) = [£ (dist(x^,x|))2]

defines a canonical metric on the product X^ x ... x x .

10. Every product of compact spaces is compact.

It suffices to consider a product of two spaces. So let X

and Y be compact topological spaces, and let T be an open cover of
X x y. Consider an arbitrary refinement A of T, consisting of open
sets of the form U x V (see 1.1.13). Since the fibers x x Y are
homeomorphic to Y and Y is compact, one can find, for each point
x e X, a finite collection Ax = (U^x) x v^x)}*?^ of elements of
A which covers x x y (see 1.7.2); one may assume that x £ U^x) for
all i = 1,...,n(x). Since the sets Ux = U± (x) are open and
cover the compact space X, there exists a finite collection
U ,...,U covering X. It is clear that A 1 = U™ A is a cover
x. x^ 3 D-l x.
1 m J
of X x y. Finally, replacing each set W £ A' by a set of r
containing W, we produce a finite subcover of r.
 30

11. Every product of locally compact spaces is locally compact.

PROOF. Let ke neighborhoods of the points E ,

...,x EX. Then U . x ... x u is a neighborhood of (x.,...,x ) in
n n 1 n ^ I n
X ^ x ... x x^. Furthermore, its closure Cl (U^ x ... x u^) is just
Cl x ... x clU (see 3), and so is compact whenever Cl , ...,C1
are compact (see 1 0 ) .

An Application:
A Method for Constructing Continuous Maps

12. Proposition 14 below allows us to establish continuity

a map in some situations similar to those treated by Theorem 1.4.3, but
where the latter is not applicable.

13 (LEMMA). Suppose that the map f: X x Q + z iscontinuous

and transforms the fiber x^ x q into a point. If the space Q is
compact, then given any neighborhood W of the point f(x^ x q ) , there
is a neighborhood U of_ Xq such that f (U x Q) c w.

PROOF. Given any point q E Q, fix a neighborhood of

Xq and a neighborhood of q with f(U^. x v^) c=w. Since Q is
compact, one can cover it with a finitecollection V ,...,V . Now
s ^1 ^
set U = (1. i U . 1 b
i=1 q±
14. Suppose that X,Y,Z and Q are topological spaces, A
is a subset of X, B is a closed subset of Y, and f: X x q z and
g: Y -> X are continuous maps such that f (x x Q) reduces to a point
for each x E A, and g(B) c A. I_f Q is compact, then for each
continuous map <j>: Y \ B + Q, themap h: Y + Z given by

f (g (y) ) ,<t> (y) ) , for y E Y \ B,

h (y) =
f (g (y) x Q) , for y E B,

is continuous.

PROOF. The map h is clearly continuous at the points of

Y \ B; let us verify its continuity at the points y E B. By virtue
of Lemma 13, given any neighborhood W of the point h(y), there is
a neighborhood U of the point g(y) such that f(U x Q) c: w. The
last inclusion shows that h(g” 1 (U)) c W, and finally note that g~ 1 (U)
is a neighborhood of y.
 31

Information

15. The notion of a product of an infinite number of

topological spaces can be defined in a natural way; in thiscase too a
product of compact spaces remains compact; see [3 ] for details.

3. Quotients

1. The quotient set X/p of a topological space X by any

of its partitions p is equipped with a natural topology: asubset of
X/p is open if its preimage under the map pr: X -> X/p is open.
Equivalently, a subset of X/p is closed if its preimage is closed.
This topology is called the quotient topology, and the set X/p with
the quotient topology is the quotient space of thespace X by its
partition p.
It is clear that pr: X -* X/p is continuous.
In the special case of a partition p whose elements are a
single set A and the points of X \ A, X/p is called the quotient
of the space X by A and is denoted by X/A.

2. Given two topological spaces X and Y with respective

partitions p and q, and a continuous map f: X Y which takes the
elements of p into elements of q, the map factf : X/p Y/q is
continuous. This is a straightforward consequence of the definition of
the quotient topology. Indeed, if U -* Y / q is open, then the set
-1 -1 -1 -1
f (pr (U)) is open in X, and so the identity f (pr (U)) =
= pr ^((factf) ^ (U) ) implies that (factf) ^(U) is open in X/p.

If q is the partition of Y into single points, then

Y/q = Y and p r : Y -* Y / q is the identity map. In this case,
f h- factf defines a one-to-one correspondence between continuous maps
X Y which are constant on the elements of the partition p, and
continuous maps X/p Y.

3. In particular, the discussion above shows that given a

continuous map f: X -* Y, its injective factor factf : X/zer(f) + Y
is continuous too. The converse is also true: every map f : X Y can
Dr fact f
be represented as the composition X --- ► X/zer(f) ------- ► Y, and so
f is continuous whenever factf is continuous.

4. A continuous map whose injective factor is a homeomorphism

will be referred to as a factorial map (or a quotient map).
 32

An equivalent definition: a map f of a topological space X

into a topological space Y is factorial if f(X) = Y , and the preimage
_i
f (B) of a set B c: Y is open if and only if B is open. If we
substitute closed sets for open ones, we obtain another equivalent
definition.
Obviously, the composition of two factorial maps is factorial,
and any injective factorial map is a homeomorphism. Moreover, it is
plain that if f is factorial and the composition g ° f is continuous,
then the map g is continuous. Also, f continuous and g ° f
factorial imply g factorial.
The projections onto quotient spaces form the main class of
factorial maps. A crude necessary condition for a map f : X-> Y with
f (X) = Y to be factorial is that f be an open or a closed map.

5. Taking quotients is a transitive operation: if p is a

partition of X and p* is a partition of X/p, then the quotient
space (X/p)/p' is canonically homeomorphic to X / q , where q
partitions X into the preimages of the elements of p' under the
projection X -> X/p. This canonical homeomorphism is defined as the
injective factor of the composite map X X/p (X/p)p!, and is truly
a homeomorphism, because this composition is factorial (see 4).

6. If the sets A and B constitute a fundamental cover of

the space X , then

fact [in:A -> X] : A/AflB -> X/B

is a homeomorphism.

Given an open subset U of the quotient A/ADB, it is enough

_i
to show that V = [pr: X -+ X/B] (fact in(U)) is open in X. But this
is a consequence of the equalities

V fl A = [pr: A -> A/ADB ] “ 1 (U)

and
B, if pr (A fl B) € U,
V fl B =
', if pr (A fl B) t U.

Properties of Quotient Spaces

7. Obviously, a qutient space X/p satisfies axiom T^ i

and only if the elements of the partition p are closed. Also, X/p
 33

is Hausdorff if and only if any two distinct elements of p have

disjoint saturated neighborhoods. Similarly, X/p is a T^-space
(T^-space) if and only if for any element A of p and any saturated
closed subset B of X (respectively, for any saturated,
closed subsetsA and B of X) such that A fl B = 0, A and B have
disjoint saturated neighborhoods in X.
Moreover, it is readily seen that X/p is second countable
if and only if there isa countable collection of open saturated sets
in X such that any saturated set can be expressed as the union of one
of its subcollections.
It is immediate from 1.6.6 that a quotient of a separable space
is separable.
Similarly, 1.7.8 implies that a quotient of a compact space is
compact.

Closed Partitions

8. A partition p of the space X is closed if pr: X X/p

is a closed map. An equivalent condition: saturations of closed sets
are closed.
Obviously, a partition which has only one element that is not
reduced to a point is closed if and only if this element is closed.

9. The quotient of a T^-space by a closed partition is a

T^-space. The quotient of a normal space by a closed partition is
normal.

Since the first assertion is straightforward, all we have to

show is that the quotient X/p of a T^-space X by a closed partition
p is a T^-space. Let F^ and F^ be disjoint, saturated, closed
subsets of X. Since X is normal, F^ and F^ have disjoint
neighborhoods. Furthermore, since p is a closed partition, the
saturations of the complements of these neighborhoods are closed, and
now it is clear that the complements of these saturations are disjoint
saturated neighborhoods of F^ and F^•

Open Partitions

10. A partition p of the space X is open if pr: X X/p

is an open map. An equivalent condition: saturations of open sets are
open.
 34

If p is an open partition and A is a saturated set, then

the saturation of IntA is open, and hence equals Int A ; passing to
complements, we see thatthe saturation of Cl A is just Cl A .
Therefore, in the case of an open partition, the interior and the
closure of a saturated set are saturated.
As it follows from 1.6.6, the quotient of a first countable
(second countable) space by an open partition is first countable
(respectively, second countable).

11. Let p and q be open partitions of the respective

spaces X and Y. The product (X/p) x (Y/q) is canonically homeo-
morphic to the quotient (X x Y)/(p x q ) .

The injective factor of the map pr x p r : X x y (X/p) x (Y/q)

definesthis canonical map, which is a homeomorphism because pr x pr
is open (see 2.5).

4. Glueing

1. Glueing (or pasting) topological spaces is a composite

operation which consists of taking a sum andsubsequently passing to a
quotient. More precisely, suppose that { isa family of
topological spaces and p is a partition of the space X = J Lxu • Then
we say that the quotient space X/p is obtained by glueing the spaces
Xy hy_ (or according to) p. The composite map

xv X JE2U X/p

is termed the v-th immersion and is denoted by imm^ . Clearly, the

setsimmv (Xy ) yield a fundamental cover of X/p, and a map
f: X/p Y, where Y is an arbitrary topological space, is continuous
if and only if all the compositions f 0 imm^ are continuous.

Unions

2. Let be a family of topological spaces. Suppose

that for each pair (y,y’) £ M x m there is given a subset A c X
. UP’ y
In addition,suppose that for each pair(y,y') € Mx m there is given
an invertible map , , such that:
(i) = Xy and y = id Xy , forany y £M;

(ii) cf>^, (Auy ,fl h v v „) = Ayly flA y ) y l ( and the diagram

 35

is commutative for every y,y',p" £ M. For x £ | |x , denote by B

ii y x
the subset of J [x^ consisting of all the points in^ (cj)^ (x) ) , where
X ^ AVy * The sets Bx are pairwise disjoint and define a partition of
J L V The corresPonding quotient space is called the union of the
spaces Xy by (or a l o n g ) the maps ^yy»-

This construction is a special case of glueing, when all the

immersions i^y are injective. Moreover, assuming that all the maps
cj)^, are homeomorphisms and that the sets Au u » are aH open or all
closed, we see at once that all the maps i™^ are embeddings.
In the general case, a union of T^-spaces is clearly a
T^-space.

3. Often the union construction is employed when all the

spaces X„ are subsets of a set X and cover X, while A . and
^ y yy ’
are given by Ayy> = Xy n Xy » and ^yy• = id- In this situation,
conditions 2 (i) and 2 (ii) are automatically fulfilled, and one may
describe the union of the X y ’s simply as the set X equipped with
the following topology: a set C c: X is open (closed) if and only if
the intersection C n X^ is open (respectively, closed) in X^ for
any y E M.
We devote some special attention to the case where the
topology of each set X is induced by some topology already given on
X. Then our construction produces a new topology on X. It is clear
that the sets open (closed) in the old topology remain open (respectively,
closed) in the new topology. Moreover, if all the intersections
X n X , are open in their sets X (endowed with the initial topology),
y
then the new topology on X induces the initial topology back on each
set X ; the same holds whenever all the intersections X fl X , are
y y y
closed in their sets Xy.

Limits and Filtrations

4. Let X q ,X ^ ,... be topological spaces, and let

 36

X.j, <j>^ -> X ^ , ___ be embeddings. Set

V xo
k-1 t>ki (Xk i) f if k ' < k'
A.
kk'
if k’ ^ k,
V
and
ab (cj)k - 1

if k1 = k,
^kk' id Xk'
[ab ((¡)^, o ... o rf)k )] - 1 , if k 1 > k,

The union of the spaces is well-defined because

conditions 2(i) and 2 (ii) are obviously satisfied. This union is called
the limit of the sequence and denoted by limiX^.,^) or
limX, •
k
A specific property of the limit construction is that the maps
imm, : X, + lim X, are embeddings: indeed, every closed subset A of
K K K.
Xk is the preimage under imm^ of some closed subset of limX^ , for
example, of

U imn^, (Clx (4>k . « 1 ° ••• ° 4>k (A) )

.*=k+1 k

Obviously, if °Pen (closed) in + i f°r aH k'

then all the sets imm^ix^) are open (respectively, closed) in
lim(Xk ,(J)k ) .

Suppose that ^ Xk + 1^ another sequence of

topological spaces and embeddings, and that for each k there is given
a continuous map fk : Xk X£ , so that all the diagrams

f,
X,
xk
vk
k +1
k+1 k +1

are commutative. Then the rule f(imm, (x)) = imm, (f, (x)) defines a
K K K
continuous map f: lim(Xk ,<t>k ) -+■ limiX^,^) (see 3.2); f is called

the limit of the sequence f^,^,..., and is denoted by lim f^ .

5. Ijf X 0 ,Xr ... are T.^-spaces, then every compact subset

of limX^ is contained in one of the sets imm^(X^) .
 37

This is a consequence of 1.7.6.

6. If_ are normal spaces and <|)^(X ) is closed

in Xk + 1 for every k, then lim(Xk ,<|>k ) is normal.

PROOF. Since we already know that lim(X ,cf) ) is a T 1 -space,

K K I
it suffices to show that there exists a Urysohn function for any pair
A ,B of closed disjoint subsets of this space (see 1.5.9). To see this,
we merely have to produce a sequence fk :imm^iX^) I, such that each
fk is a Urysohn function for the pair A fl imm^iX^), Bn imm^iX^),
and f,k a1
+1 I . ^ ,v v = f,k
|lmm^(X^) for each k.

As fg take any Urysohn function for the pair A fl immQ(XQ),

B fl imm^iXQ). Given fk , we define immk+^ (X^ + ^) I as the
(continuous) extension of the function

gk : [immk (Xk )] U [A fl i m m ^ (Xk + 1 )] U [B n immk + 1 (Xk + 1)] I,

defined by the formula

fk (x) , if x e immk (Xk ),

gk (x) = 0, if x £ A fl immk + 1 (Xk + 1
1/ if x € B D immk+ 1 (Xk + 1

(see 1.5.11). The functions gk are continuous because the sets

imm^ (X^) , A n immk+^ (Xk+^), and B fl immk+^ (Xk+ ^) are closed (see
1.4.3), which in turn is a consequence of the fact that ^k^Xk^'
^k+1 *Xk +1 ^ * are cl°sed-

7. A sequence XQ ,X^,... of subsets of a topological space

X is a filtration of X if, firstly, XQ c x^ c ..., and, secondly,
the sets X, form a fundamental cover of X.
k
The first condition shows that the inclusions in: X^ xk + ^
and the limit lim(X, ,in) are meaningful, while the second condition
K
is equivalent to the following: the map X lim(Xk ,in), which equals
imir^: X^ lim(Xk ,in) on each Xk , is a homeoraorphism. Using this
canonical homeoraorphism, we may identify lira (Xk ,in) with X.

Attaching

8. Let X 1 ,X2 ,C, and cp: C -»■ X 2 be two topological space

a subset of X 1, and a continuous map, respectively. Denote by p
the partition of X.J [x2 into the points of in 1 (X1 \ C) and
 38

in2 (X^ ^cp(C)), and the sets in^ ((p ^ (x) ) U in2 (x) with x £ cp(C). The

quotient space (X^J [X^J/p is written X^ . We say that

^2 *^cp "*"S °ktained ky attaching the space X^ to the space X^
(or along) cp.
This construction is clearly a special kind of glueing, and
it is plain that imir^: X 2 + X 2 X^ is an embedding.
If X2 reduces to a point, then X2 X^ is canonically
homeomorphic to the quotient space X^/C; the canonical homeomorphism
is just fact[imm •
. X. Xn U X.].
J 1 1 2 cp 1
9. If: X^ and X2 are normal and C is closed, then
X n U X. is normal.
2 cp 1 ----------
PROOF. Since we already know that X2 X^ is a T^-space
(see 2), it suffices toshow that there exists aUrysohn function for
any pair of closed disjoint subsets A,B of X 2 X^ . Let f2 :X 2
be a Urysohn function for the pair imm 2 (A), imm2 ^(B). Define
— 1 — 1
g: C U [imm^ (A) ] U [imm^(B) ] -* I by

f2 (cp(x) ) , if x €C ,

g(x)= < 0, if x E imm^ (A) ,

-]
-
1, if x e imm^ (B),

and extend it to a continuous function f^: X^ I (see 1.5.11). The

function X n U X. I, defined as
l (p 1
f 1 (x), if y € imm 1 (x) [x € X ],

f 2 (x), if y £ iram2 (x) [x £ x 2^'

is obviously a Urysohn function for the pair A ,B .

5. Projective Spaces

1. In this subsection we shall describe the real, complex,

quaternionic, and Cayley projective spaces. These may be considered as
examples illustrating the previous definitions, but are also important
spaces in their own right.
We denote the field of complex numbers by (C, the field of
quaternions by 1H, and the algebra of Cayley numbers by (Ca. The
corresponding n-dimensional spaces, i.e., the products of n copies
 39

(E x ... x (E,I H X ... X IH, and (Ea x ... x (Ca, are denoted by CDU r
H , and (Ea . Since every complex number is a pair of real numbers,
every quternion a quadruplet of real numbers, and every Caylely number
an octuplet of real numbers, one can naturally identify (En , IHn , and
(Ea with 1R , 3R^n ,and !R^n ,respectively.In particular, the
former are endowed with natural topologies and metrics. The vector
operations in (E , IHn , and (Can (addition of vectors and left or
right multiplication by scalars) are continuous in these topologies.

2. The n-dimensional real projective space lRPn is defined

as the quotient space of Sn by its partition into pairs of diametrical­
ly opposed points. One may equivalently describe HRPn as the quotient
space of Dn by its partition into the points of Int Dn and the pairs
of diametrically opossed points of Fr Dn = Sn 1. The canonical homeo-
morphism permitting us to identify these two quotient spaces is fact f ,
where f: Dn -* Sn is defined by

(x.j,...,xn ) n- (x1 ,.. .,x , (1 -x2-. .*-x^) 1//2) .

One may also identify the points of 3RPn with the lines of
IRn 1 which pass through the point 0 = (0,...,0) [the line passing
through the points x and -x corresponds to the pair of points
x ,-x E Sn ] . The set of all these lines, equipped with the angular
metric (i.e., the distance between two lines is defined as the angle
between them which is less than tt/2 ) , is a metric space, and the above
natural map of lRPn onto this space is clearly a homeomorphism. This
provides a third description of the real projective space. The fourth,
a coordinate description, can be obtained if one remarks that every line
passing through 0 is uniquely determined by any of its nonzero points,
and that the coordinates of any two nonzero points of such a line are
proportional. This enables us to interpret the points of lRPn as
classes of proportional nonzero real sequences (x^,...,x^+^); the
point determined by the sequence (x^,...,x^+^) is denoted by
(x^:...:xn+^), and the numbers x ^ ,...,xn+^ are called its homogeneous
coordinates. The description of the topology of lRPn in terms of the
homogeneous coordinates is plain.

3. All the above discussion of the space ]RPn can be

repeated naturally for the complex case, leading to four equivalent
description of the n-dimensional complex projective space (EPn .
n . 2 n +1
First description: (EP is the quotient space of the unit sphere S
„ j 'i

of (C by its partition into the circles obtained by intersecting

S2n+1 with the (complex) lines of (Cn which pass through the point 0.
The second description: (EPn is the quotient space of the unit ball
 40

D n of (Cn by its partition into the points of Int D and the

circles on Fr D2n = S 2 n _ 1 obtained by intersecting S2n 1 with the
lines of (En passing through 0. The canonical homeomorphism between
these two quotient spaces is fact f , where f: D2n S2n "* is given

by f (x.,,...,x2n) = (X l ,...,x2n, (1-x2 -.•.-x2 n )1 /2 ,0) . The third

description: (CPn is the set of lines of CCn + 1 passing through the

point 0, equipped with the topology induced by the angular metric.
The fourth description: ССРП is the space of classes of (complex)
proportional nonzero complex sequences (x^, .. .,xn + ^) .
The notation
(x„:...:x „) introduced in 2 also extends to
1 n+ 1
the complex case, and the numbers x ^ , . . . are known, as in the
real case, as the homogeneous coordinates of the point (x ,...,xn + ^ .

4. Since the field of quaternions is not commutative, one

has to distinguish between the left and the right lines in ШП. But as
soon as we have chosen one type of lines, we can automatically repeat
the discussion in 3 for the quaternionic case, and so obtain four
equivalent descriptions of the corresponding (left or right)
n-dimensional quaternionic projective space ПНРП . Since the anti-auto-
-1
morphism x ^ x of the multiplicative group of the field Ш takes
left lines into right ones, the right space ШРП is homeomorphic to
the left one.
Henceforth, we shall regard ПНП as a left vector space, and
accordingly ПНРП will be regarded as the left projective space.

5. Since the Cayley algebra is not associative, we cannot

successfully define lines in (Ea11 for n > 2. In the Cayley plane
2
CCa one can define a line passing through the point (0 ,0 ) as a set
{(x^x^) ] x 2 = cx^}, where с E CCa; in addition, there is the
vertical coordinate line { (x ,x0) | x = 0}. If one identifies (Ca2
16
with IR , it becomes clear that these lines are 8 -dimensional subs-
spaces. Moreover, every point different from (0,0) of CCa sits on
exactly one of these lines, and each line intersects S 15 along a
7-dimensional sphere. One can define the projective Cayley line CCaP
15
as the quotient space of S by its partition into these 7-dimensional
spheres. Of course, there are three more description of this projective
line, which are appropriately modified versions of those given in 2 , 3 ,
and 4. In addition, we can define the projective Cayley plane EaP2 as
the quotient of D16 by its partition into the points of Int D 1 6 and
the 7-dimensional spheres just described. However, an attempt to
2
describe the projective plane CCaP in the spirit of the alternatives
of 2, 3, and 4 fails. Projective Cayley spaces of higher dimensions
 41

are not defined.

1 1 1 1
6.
The spaces IRP , (EP , H P , and (EaP are canonically
1 2 4 8 1 1
homeomorphic to S , S , S , and S . The homeomorphism EP -* S
transforms the line x^ = 0 into ort1, and each line x 2 = cx 1 into

the point of the punctured sphere S \ ort which corresponds to c

1 1
via the homeomorphism IR -v s \ ort described in 1.4.8. The homeo-
1 2 1 4 1 8
morphisms (EP -+ S , H P S , and (EaP S are similarly defined,
when we substitute the homeomorphisms (E1 =2 3R S 2 ^ ort ,
1 4 4 1 8 8
]H = IR S \ ort 1 , and la - IR S \ ort for the homeomeorphism
1 1 I 1
IR S \ ort^ .

7. The canonical embedding (x , ...,x ) h- (x ,...,x ,0) of

I JC I JC
k k +1 k
IR in IR permits identification of IR with the subspace x =0
k +1
of IR and may be regarded as an inclusion. This map induces
inclusions D^ + \ ^ -> S^ , and 3RP^ ^3RP^. Similarly, the
k k +1 k k +1 k —1 k
inclusions (E (E and № IH induce inclusions (EP (EP
and B P k ’ 1 -+■ IHPk .
Set
oo k °° k °° k
IR = lim IR , (E = lim (E , IH = l i m l ,

oo k °° k
D = lim D , S = lim S ,
and
HRp” = l i m l P k , (EP°° = lim (EPk , 3HP°° = l i m K P k .

The points of the spaces IR , (E , and-i IH can be naturally

identified with the real, complex, and quaternionic finitely-supported
°o r
(i.e., having only a finite number of nonzero terms) sequences •
Thesphere S™ is included in the ball D°°, which in turn is included
oo oo oo oo
in the space IR . The projective spaces IRP , (EP , and IHP are
constructed from S°° and D°° by taking quotients that are limits of
the quotients described in 2, 3, and 4.

8 .The previous description of (EPn and IHPn , and also of

(Eap\ as quotient spaces of spheres, define projections S2n ^ (EPn ,
^4n +3 f and s 15 (Eap\ which play a distinguishedrole. They
3 1 2
are called Hopf maps. The most important Hopf maps are S + (EP = S ,
n "1 A 1 ^ 1 R
S HP = S , and S -s- (EaP = S .
 42

6 . More Special Constructions

1. Let X be a topological space. The quotient space

(X x I)/ (X x 0) is called the cone over X and is denoted by con X .
The point pr(X x 0) is the vertex of the cone, the set pr(X x 1) is
the base of the cone, and each set pr(x x I), x e X, is a generatrix
of the cone. The base ofcon X is canonically homeomorphic to X and
is usually identified with X. The generatrices are obviouslycanonical­
ly homeomeorphic to I.
For each map f from X into another topological space Y,
we have the map fact(f x id I) : con X con Y , which is continuous
whenever f is so. This map is denoted by con f .

2. The quotient of the product X x I by its partition whose

elements are the sets X x 0 and X x 1 , and the points of the set
(X x I) \ [(X x 0) U (X x 1)], is called the suspension of X and is
denoted by suX . The points pr(X x 0) and pr(X x 1) are the
vertices of the suspension, the set pr(X x 1) is its base, and the
sets pr(x x I), x E X, are its generatrices. The base of suX is
canonically homeomorphic to X, while each of its generatrices is
canonically homeomorphic to I.
For each map f: X Y there is the corresponding map
fact(f x id I) : su X su Y , which we denote by su f ; su f is
continuous whenever f is so.
Notice that the suspension su X can be alternatively
described as con X/X .

3. Let X^ and X^ be topological spaces. The quotient

space of X^ x x^ x I by its partition into the sets x^ x x^ x 0
(with x 1 G X 1), X^ x x2 x 1 (with x 2 E X^), and the points of

(X1 x X 2 x I) \ [(x1 x x 2 x 0) U (X1 x x 2 x 1)],is called the join of

X1 and X 2 and is denoted by X 1 * X2 . The sets pr(X^ x X 2 x 0)
and pr(X^ x x 2 x 1 ) are the bases of the join, and the sets
pr(x 1 X x2 X I) with x 1 E X 1and x2 e X2 are its generatrices.
The bases are obviously canonically homeomorphic to X^ and x2 , and
are usually identified with and x2 - The generatrices are
canonically homeomorphic to I.
For each pair of maps f^. X 1 + Y 1 and . X2 + y2
f2* we have
the map fact(f1 x f 2 x id I) :x 1 * X 2 -> Y 1 * Y2 , denoted by f * f ;
f^ * f 2 is continuous whenever both f and f are so.

The *- operation is commutative, i.e., there exists a

 43

canonical homeomorphism X„ * X -+ X * X_.

^ I I z
We remark that the join X^ * X 2 may be alternatively defined
as <X -|J lX 2 ^ % *X 1 x X2 X I) 'where x -| x x 2 x(0 U 1 ) -> X^j ^x 2
is given by tpix^x^O) =i ^ (x., ) ,cp(Xl ,x2 ,1 ) = in2 (x2) . It is also
clear that the quotient space of X * X2 by its partition whose
elements are the bases X^ and X2 , andthe points of the set
(X 1 * x2) ^ (X^ U X2) is the suspension su(X^ x x2) .

4. The iterated join (...((X1 * X2) * X3)...) * XR maybe

canonically embedded in the product con X. x ... x con X . This
1 n
embedding is denoted by jc or, more precisely, by jc , and
x 1 ,...,xn
is defined inductively: for n= 1 it takes x1 £ X 1into pr(x.j,1 ) £
£ con X^ , while for n ^2 it is given by

jc A 1 r . . . fA
(pr(x,x n ,t)) = ((1 -t)jc A A .
(x),pr(x n ,t)),
i n I n-1

where x £ (... (X1 * X0) * X,)...)* X x £X, and t £ I; the

I J n— I n n
multiplication of a point of con X^ x . . . x conXn_^ by 1 -t is
defined by the rule

(1 -t)(pr(x 1 ,t1 ),...,pr(xn- 1 /tn_1)) =

= (pr (x1 , (1 -t) t j , ...,pr (xn - 1 , (1 -t) tn-1) ) .

Clearly, the image of the embedding jc v is precisely

1,‘'' n
{ (pr(x.,t.) ,...,pr (x ,t )) £ con X. x ... x con X I t.+...+t = 1},
1 1 n n I n \ n

which allows us to identify the iterated join

(...((X 1 * x2) * X )...) * Xn with this set.

5. The ^-operation is associative, meaning, as usual, that

the two joins (X^ * X2) * X^ and * (X2 * X^) are canonical homeo-
morphic. The canonical homeomorphism (X^ * X 2) * X^ -* X^ * (X2 * X^)
is the composition of the canonical homeomorphism
x * (X2 * X^) (X2 * X^) * x -| with the suitablecompression of the
canonical homeomorphism con X 2 x con X^ x con X^
+ con X 2 X con X^ X con X 1 .

A consequence of the associativity of the ^-operation is that

the multiple join X 1 * ... * X r is meaningful for any topological
spaces x ^ /.../Xn .

6. The product con X^ x ... x con Xn is canonically

 44

homeomorphic to con(X^ * ... * Xr ).

The canonical homeomorphism

con(X, * ... * X ) + con X. x ... x con X

I n i n
is defined as

pr(jc“ 1 (pr(x1 ,1^) , . . . ,pr(xn ,tn ) ) ,t) ~

1 '* n
(pr(x^,tt^ /max ( , ..., )), ...,pr (x^,tt^/max (t ^ , ...,t^ ) ) ) .

7. con Sm and su Sm are canonically homeomorphic to D™

*
and o m+1
S

The canonical homeomorphisms con Sm Dm 1 and su Sm -> Sm

are defined by the formulas

pr((X l ,...,Xm + 1 ) ,t) H. (tX l ,...,tXm + 1 )

and
pr ( (x f . . . ^x m+1 ) / t ) » (X1 Sin Tit f . . . /Xm+1 s i n TTt ,COS 771) .

8. The join X * D° is canonically homeomorphic to con X .

The join X * S is canonically homeomorphic to su X . The join
k k+ 1
X * S is canonically homeomorphic to the iterated suspension su X;
m^ m2 m +m2 +1
in particular, S * S is canonically homeomorphic to S

The canonical homeomorphism X * con X is given by

pr(x,0,t) *+ pr(x,t). The canonical homeomorphism X * su X is
given by the formulas pr(x,1,t) *+ pr (x, (1 +t )/2) , pr(x,-1,t) h-
*+ pr(x,(1-t)/2). Finally, the canonical homeomorphism X * Sk suk +1 X
is the composite map
k V— i v n 1<*+1
X * S + su X * S + ... + su X * S + su X,

where the last arrow denotes the canonical homeomorphism, and the r-th
arrow, with r £ k, denotes the composite canonical homeomorphism

sur_1X * Sk_r + 1 ->■ sur_1X * su Sk_r -> sur_1X * Sk_r * S°

- sur"1X * S° * Sk"r - surx * Sk"r .

9. Combining the canonical homeomorphisms constructed in

6 -8 , we obtain the composite homeomorphisms

m m m m m +m +1
s * ... * s s * ... *s n * s n n - ... ^
 “ '1 iun m - —i m -
D x ... x D -*■ con S x ... x con S n

m -i~1 m 1+. ..+m -1 m. + .-.+m

-> con (S * ... * s ) ->■ con S n -*■ D n
and
m m m -1 m -1
D * ... * D -*■ con S * ... * con S n

m1 - 1 0 m _1 n
S *D * . . . * S n * D -*■

m i_1 n i~ 1 n m _1
-* COn(S * D * ... * S * D * s n ) ->-

ml - ^ 0 mn - 1 - ^ n m -1
-*■ con S X con D X ... X con S X con D x con S n -*
m. m m
D x l x . . . x D n x I x D n ->
m1 1 m m m,+...+m +n-1
D X D x ... x D n x D x D n - >- D n

n»1 m m1 m
Th er e fo re , the jo in S * . . . * S , the product D x ... x d n,

and the jo in Dml * . . . * Dmn are c a n o n i c a l l y homeomorphic t o the sphere

m.+ . . . +m +n-1 m + . . . +m m1+...+ m +n-1
S 1 n the b a l l D n , and the b a l l D 1 n ,
respectively.

The Mapping Cylinder and the Mapping Cone

10. Let f :X ^ -> be a continuous map. The result of

attaching the product X^ x I to X^ by the map X^ x 1 4
(x,1 ) ►>f (x),is called themapping cylinder of f, and isdenoted by
Cyl f . The sets imm^ (X^ x 0) and imm2 (X2) are the lower and upper
bases of Cyl f , and the sets imm^(x x I) with x € X^ are its
generatrices. Clearly, the bases are canonically homeomorphic to X^
and X2 # and they are usually identified with these two spaces; the
generatrices are canonically homeomorphic to I. Moreover, there is
a canonical retraction rtf : Cyl f -> X2 , defined on imm^ (X x I) as
rtf (imm^(x,t)) = imm^(x,1) [ = f (x) ] . It is evident that the composite
map

X, ^ 2 — cyl £ x2

equals f.
 46

If X2 = X 1 and f = id X^ , then Cyl f is canonically

homeomorphic to the cylinder over X^, X^ x i.

11. The mapping cone of the continuous map f: X^ X^ is

the space X^ con X , denoted by Con f (do not confuse it with
conf , defined in 1). Equivalent definition: Conf = Cyl f/X^ .

7. Spaces of Continuous Maps

1. Let C(X,Y) be the set of all continuous maps of a

topological space X into a topological space Y. The set of all maps
<J> e C (X,Y ) such that (p(A^ ) c B ^ , ...,cf)(A^) c B^ , where A 1,...,A^ and
B^,...,Bn are given subsets of X and Y, respectively, is denoted by
C (X ,A ^,...,A ^ ;Y ,B ^,...,B^) . It may be interpreted as the set of all
continuous maps (X,A. , ... ,A ) -+ (Y,B. , ...,B ) .
I n I n
We equip C(X,Y) with the compact-open topology: by definition,
this is the topology with the prebase consisting of all sets C(X,A;Y,B)
with A compact and B open. Together with C(X,Y), all the sets
C(X,A^ ,...,A^ ?Y ,B ^ ,...,Bn ) become topological spaces.
If Y is a point, then C(X,Y) reduces to a point. If X
is discrete and consists of the points x„,...,x , then C(X,Y) is
I n
canonically homeomorphic to the product Y x ... x Yof n copies of
the space Y; this homeomorphism is given by cf>-+ ($ (x^) , ..., (x )) .
To each pair of continuous maps f: X ’ -> X and g: Y -> Y 1
there corresponds a mapping C(X,Y) + C(X',Y'), given by therule
(j) h- g o 0 o f . This mapping is continuous, and we shall denote it by
C(f,g).

2. I_f Y is a Hausdorff space, then sois C(X,Y).

Indeed, if (p,ip E C(X,Y) and $ ^ \jj, then there is x £X

such that (J)(x) ? ^(x). Let U and V be disjoint neighborhoods of
the points <J>(x) and i|;(x). Then C(X,x;Y,U) and C(X,x;Y,V) are
disjoint neighborhoods of the points $ and \p.

3. If X is compact and Y is metrizable, then C(X,Y)

is metrizable. Moreover, if Y is equipped with a metric, then
dist ((J>,ip) = suPxex dist (<f>(x ) , ip (x) ) defines a metric on C(X,Y),
compatible with its topology.

PROOF. Given $ £C(X,Y), the set 0(X ) can be covered by

a finite number of balls of an arbitrarily small radius e
(see 1.7.11). It is clear that 0/ = il®= 1 C (X ,<j>“1 (U±) ;Y ,Ui) is a

neighborhood of the point $, contained in the ball of radius 2e

 47

centered at 4). Therefore, every ball in C(X,Y) contains a neighbor­

hood of its center.
On the other hand, if A c X is compact and B c Y is open,
with 0(A) cz B, then C(X,A;Y,B) contains the ball with radius
Dist((j>(A),Y \ B) centered at (J) (see 1 .7.15) . Therefore, every
neighborhood of <J) belonging to the prebase considered in 1 contains a
ball centered at $.

4. For any topological spaces X and Y ,...,Y ^ , the space

C(X,Y^ x ... x y ) is canonically homeomorphic to the product
C (X,Y.) x ... x C(X,Y ) .
I n
This canonical homeomorphism takes each <(> € C(X,Y^ x ... x y )
into (pr1 ° 0 ,.. .,pr ° <(>) G C (X,Y 1) x ... x C(X,Y ) [cf. 2.4].
I n 1 n
5. Let p be a closed partition of the compact Hausdorff
space X, and let Y be an arbitrary topological space. Then
C(pr,idY) : C (X/p,Y) C(X,Y) is an embedding.

It suffices to show that given a compact subset A of X/p

and an open subset B of Y, the setC(pr,idY) [C (X/p ,A; Y ,B) ] is
open in C(pr,idY) [C(X/p,Y)]. Since X/p is Hausdorff (see 3.9),
-1
A is closed. It follows that pr (A) is closed, and hence compact.
Consequently, C(X,pr 1 (A);Y,B) is open in C(X,Y), and it remains to
note that

C (pr ,id Y) [C (X/p ,A; Y,B) ] = C (X ,pr~ 1 (A) ;Y ,B) fl C(pr,idY) [C(X/p;Y)].

The Mappings X x y -* Z and X C(Y,Z)

6. Suppose that X, Y and Z are topological spaces, and

4): X x y Z is continuous. Then the formula [ V (x)] (y) = cj)(x,y)
v
defines a continuous mapping $ : X C(Y,Z).
Let ip: X C(Y,Z) be a continuous mapping, and suppose that
Y is Hausdorff and locally compact. Then the formula ^A (x,y) =
= [^(x)](y) defines a continuous mapping \pA : X x y -> Z .

To prove the first assertion, pick a point X q € X, a compact

set B c Y, and an open set C ci Z . Then it is enough to exhibit a
neighborhood U of xQ such that (pV (U) c: C(Y,B;Z,C).For each point
y € B fix neighborhoods and of x^ and y such that
ch(U x v ) c C, and then extract a finite cover V ,...,V of B
^ Y Y 1 s s
from the collection {Vy y
} t 0r s0 . It is clear that U = H _L-iy.
._1 U is a
 48

neighborhood of xQand that (f>(U x B) c= U ? * cj>(U x v ) <= C .It

1 Yj_ Yj_
remains to remark that the inclusion 6 (U x b) cz C is equivalent to
(PV (U) cz C(Y,B;Z,C) .
To prove the second assertion, pick a point ^x q '^0^ E X x y
and a neighborhood W of the point iJjA (Xq ,Yq )* No w let us find a
neighborhood V of with compact closure Cl V satisfying
-1
Cl V c: [^(Xq )] (W) (see 1.7.22), and then a neighborhood U of xQ
satisfying ijj(U) c C(Y,C1V;Z,W) . Obviously, U x v is a neighborhood
of the point and ’
JjA x V) cz W.
7. The mapping C (X x Y,Z) -> C(X,C(Y,Z)) defined by the r
<j> ^ (pV (see 6) is continuous for any topological spaces X, Y and Z,
If Xis Hausdorff and Y is Hausdorff and locally compact, then this
mapping is a homeomorphism, and its inverse is given by the rule ip ipA ,

The continuity of the mapping 6 *+ (pv results from the fact

that the preimage of C (X,A;C(Y,Z) ,C(Y,B;Z,C)) under this mapping is
just C(X x Y ,A x B;Z,C). Assume that X is Hausdorff and Y is
Hausdorff and locally compact. Consider a point E C (X ,C (Y;Z)), a
compact subset Q of X x Y, a neighborhood W of the set (Q)r
and a point q E Q. Now find a neighborhood U x V of q such that
A ^ ^3
ip* (U
u q x Cl Vq ) <z W. Since Q is compact, its images pr.I (Q) and

pr2 (Q) in X and Y are also compact (see 1.7.8). Moreover, they
are Hausdorff spaces together with X and Y, and hence normal
(see 1.7.5). Consequently, there exist open subsets of pr^ (Q)
and V'q of pr0
2 (0) such that

pr,(q ) £ U ' , Clp V Q ) tr = Uq ,

and
pr2 (q) 6 V', Clpr2,Q)V ^ = V q ,

and it is plain that the intersection (IP x V') n Q is open in Q,

Being compact, Q can be covered by a finite number of such
intersections, say LM x ,..., x Vq * Now set

T = j, C(X'C 1 pr,(QI0 qi:C(V'Z,'C(Y’C 1 pr 2 <Q)Vq1 :Z'W n '

It is clear that T is a neighborhood of q and that the image of T

under the mapping ^ is contained in C (X x Y,Q;Z,W). We conclude
that ip >->- t|j is continuous. It is readily see that the mappings
V A
<J> h- (j) and ip ip are inverses of one another.
 49

A Surprising Application

8. Let f: X ->• X 1 be a factorial map. If the space Y

Hausdorff and locally compact, then the map fxidY:XxY-*X'x Y
is factorial.

One can assume that X' = X/zer(f) and that f is the

projection X -> X/zer(f) . Consider the projection
p r : X x Y ->• (X x Y)/(zer(f) x zer (id Y) ) . The mapping
Pr V : x C (Y , (X x Y)/(zer(f) x zer (id Y) ) is constant on the elements
of the partition zer(f), and hence it induces continuous mappings

fact prV : X 1 C (Y , (X x Y) / (zer (f) x zer(idY))

and
(fact prv)A : X' x y -+ (X x Y)/(zer(f) x zer(idY)).

It it clear that the second of these mappings is the inverse of the

injective factor of f x i d Y i X x Y + X ' x y. Thus the injective
factor of f x i d Y : X x Y ^ X ' x y is a homeomorphism.

9. Let f: X X' and g :Y ->• Y 1 be factorial maps. If X1

and Yare Hausdorff and locally compact, then the map
f x g: x x Y -+ X' x y' is factorial»

In fact, one can express fx g as the composition

f x id . id x q
X x Y — ---- — >X' x Y ------ X' x Y'

and recall that a composition of factorial maps is again factorial.

8. The Case of Pointed Spaces

1. In the sequel, the class of topological spaces equippe

with a simple additional structure - a distinguished point (i.e.,
topological pairs (X,xQ ), where xQ is a point) will play an
important role; we call these spaces pointed spaces, and call the
distinguished point a base point. The constructions described in the
previous subsections must be naturally modified when applied to such
spaces. For some of these construction, the modification entails
merely the addition of a base point to the resulting space: for example,
the quotient space of pointed space (X,xQ) has the natural base point
 50

PIt (Xq ), the product of the pointed spaces (X^ ,x^ ),..., (X^x^) has
the natural base point (x^,...,x ), and the space of continuous maps
from X into a pointed topological space (Y,Yq ) contains the constant
map const: X + Y, x yQ , and hence has the natural base point const.
Other constructions such as the sum, suspension, and join need more
serious modifications.
We shall describe these modified constructions below, and also
introduce a new one - the tensor product of pointed spaces. In every
case, pointed spaces produce pointed spaces, and base point-preserving
maps again produce base point-preserving maps. We remark that the maps
factf , C(f,g), and f^ * ... x fR preserve base points whenever the
initial maps have this property.
We use the symbol bp as a general notation for the base
points.

Bouquets and Tensor Products

2. The construction below replaces the sum construction f

pointed spaces.
Let be a family of topological spaces with base
points . The quotient space of the sum J Lye^ Xy ky the subset
consisting of all points in^(x^) is called the bouquet (or the wedge)
of the spaces , and is denoted by ^ ^ consists of
the numbers 1,...,n, we also write
(X. ,x1 ) V ... V (X ,x ) . The point
II n n c
pr ° in^(x^) E V(X^,x^) does not depend on v; it iscalled the center
of the bouquet V(X ,x ), and is taken as its base point.

The bouquet V(X ,x ) is obviously a union of the spaces X

y y ^ y
(see 4.2), and so there exist the embeddings imm^:X^ V(X ,x ). The
maps Prv : V^Xy'xy) , defined by

xv , if v ’ / v,
prv (immv ,(x)
x, if v 1 = V,

are specific to the bouquet construction. Clearly, pr ° imm = id X

V V v
and pr^ ° imm^, = const if v 1 f v.
If M also indexes another family of pointed spaces (Y ,v )
y y
and a family of continuous maps f^: X^ Y^ such that f (x ) = y ,
then the map fact y ) : V(Xy ,xy) + V(Yy ,y ) is well-defined and^
continuous; we denote it by V f
y
 51

3. Let (x^ ,x^ ( x ^ ,x^) be pointed spaces. The rules

x »(x,x2,- ..,xr ) [x £ X^] , ... , x » (X l ,... /xR_1,x ) [x € X ],
definecanonical embeddings X„ -► X„ x ... x x , .... X -*X x ... x x !
1 1 n' n 1 n1
denoted by in^,...,in^. Moreover, the rule x (pr^ (x) , ...,prn (x))
defines a canonical embedding (XwX. ) V . . .V (X ,x )-> X. x ... x x ,
1 1 n n 1 n
which allows us to regard the bouquet (X„,x ) V ... V (X ,x ) as a
I t n n
subspace of X x ... x x . Clearly, in .: X . -► X. x ... x x is the
I n 1 1 1 n
composition of the embedding imm.: X. -+ (X.,x.) V ... V (X ,x ) with
1 1 1 1 n n
the inclusion (X ,x.) V ... V (X ,x ) -+■ X.x ... x x , while the
1 1 n n 1 n
projection pr : (X ,x. ) V ... V (X ,x ) -v x . is the restriction of
i l l n n 1
p r .: X . x ... x x ■> X ..
I 1 n 1
The quotient space (X. x ... x x )/[(X.,x.) V ... V (X ,x )]
I n i l n n
is called the tensor product of the spaces X^,...,X , and is denoted
by ( X . , x . ) ® ... ®
(X ,x ) . The pointpr[ (X.,x ) V ... V (X ,x )] £
II n n II n n
£ (X^/x^) ® ... ® (X^/X^) is called the center of the tensorproduct
(X ,x ) ® ... ® (X ,x ), and is taken as its base point.
I I n n
The tensor product is a commutative and associative operation:
there are obvious canonical homeomorphisms (X^,x^) 0 (X2 ,x2) -*
■> (X2/x 2) ® (X^x^) and (X^ ,x^ ) ® [ (X2 ,x2) ® (X3 ,x3),bp] ->
t(X^/X^) ® (X2 ,x2),bp] ® (X^,x^); this is also the way we understand
the more general equality [(X^,x^) ® ... ® (X^_^,x^_^),bp] ® (xn 'xR ) =
- (X1 ,x1 ) 0 . .. ® (Xn ,xn ) .

If (Y.,y.),..., (Y ,y ) are other pointed spaces and

1 1 n n
f : X ^ -> Y ^ , ...,f^ : X^ -* Y^ are continuous, base point-preserving maps,
then the map fact(f. x ... x f ): (X.,x.) ® ... ® (X ,x ) ->
c 1 n 1 1 n n
(Y^,y^) 0 ... ® (Y^^y^) is well defined and continuous; we denote it
by f . 0 ... 0 f .
1 n

Cones, Suspensions, and Joins

4. The cone over the pointed space (X,x^) is defined as

quotient of the usual cone con X by its generatrix pr(xQ x I), and
is denoted by con(X,xQ). The image of pr(xQ x I) under the projection
con X -> con(X,xQ) is the vertex of con(X,xQ), and is taken as its
base point. The image of the base of con X under the projection
con X -* con(X,xQ) is the base of con(X,xQ); this projection carries
the first base onto the second one, and thus allows us to identify the
base of con(X,x^) with X.
If (Y ,y q ) is another pointed space and f: X Y is
 52

continuous, with f (xQ ) = y 0 ' then the maP fact con f : con(X,xQ)
con(Y,yQ) is well defined and continuous, and we denote it simply by
con f .
Equivalently, one may describe con(X,xQ) as the quotient
space of the cylinder X x I by (X x 0) U (x^ x I).

5. The suspension of the pointed space (X,x^) is defined

as the quotient of the usual suspension su X by its generatrix
pr(xQ x I), and is denoted by su(X,xQ) . The image of this generatrix
under the projection suX su(X,xQ) is the vertex of su(X,xQ ) and
is taken as its base point.
If (Y, y^) is another pointed space and f: X Y is
continuous, with f(Xg) = yQ , then the map fact su f : su(X,xQ ) -+
s u (Y,Yq ) is well defined and continuous, and we denote it simply by
su f .
Equivalently, we may decribe su (X,X q ) as the quotient space
of the cylinder X x I by (X x (0 U 1)) U (xQ x I), i.e., as
(X,Xq ) 0 (1/(0 U 1) ,bp) = (X,Xq ) ® (s\ort^). Another equivalent
description: su(X,Xg) = con(X,XQ)/X.

6. The join of the pointed spaces (X^ ,x^ ) and (X^fX^) is

defined as the quotient space of the usual join X^ * X^by its
generatrix pr(x^ x x^ xxi) , and is denoted by (X^,x^) * (X^,x^ ) • The
image of pr (x^ x x^ x I) under the projection X^ * X 2 ->
-+ (X^,x^)* (X2 ,x 2 ) is the center of (X^x^) * (X2 ,x2), and is taken
as its base point.
If (Y^,y^) and (Y 2 ,y 2) are another pointed spaces, and
f^ : X^ -* Y^ and f2 : X 2 Y 2 are continuous maps such that f ^ (x^ ) =
= y 1 and f 2 (X 2 ^ = Y 2 'then the maP fact(f 1 * f2> : ( X ^ x ^ * (X2 ,x2) ->
(Y-j/Y-j) * d^fined and continuous, and we denote it
simply by f^ * f2 .

7. For any two pointed spaces (X^x^) and (X2 ,x2), the
bouquet (su(X^,x^) ,bp) V (su(X2 , ^ 2 ) 'bp) is canonically homeomorphic
to su((X1 /x 1) V (X2 /X2 )/bp).

The canonical homeomorphism su((X^,x^) V (X2 ,x2 ),bp)

(su (X1 ,x1) ,bp) V (su (X2 fX2) ,bp) is given by

pr (immi (x) ,t) immi (pr (x,t) ) [x € X± , i = 1,2].

m m
8. con(S ,ort^) , su(S ,ort^) , and (Sm ,ort1) * (Sn ,ort )

are canonically homeomorphic to Dm+1, s m+1, and s m+n+1f

respectively.
 The canonical homeomorphism su(Sm ,ort ) -+■ Dm+1 is defined
as pr ((x1 ,. ..,xm+1) ,t) h- (tXl + (1 -t) ,tx2 ,. ..,txm+1) .

The canonical homeomorphism su (Sm ,ort.j ) Sm+1 transforms

the generatrix passing through the point x G Sm onto the circle on
XT\^
S with center (ort^ + x)/2 (which degenerates to the point ort^
if x = ort^) ; as t varies from 0 to 1, the image of the point
pr(x,t) moves uniformly on this circle, starting from ort^ , and
continuing into the half space xm+>| ^ 0 (see Fig. 1).

First base

Finally, the canonical homeomorphism (Sm ,ort^) * (Sn ,ort^) +

s m+n+1 j_s defined on the bases Sm and sn by the formulas
3x1+1 x /3 x /3 /3(1-x )
/ » / 1 ^ m*t" l 1a /-v \

(x1 '* * *,Xm+ 1 ) ( 4 ' ~2 '**•' 2 /0 /---/0 , 4 )

and 3X.+1 x /3 x /1 /J(xr 1)

( x^/ » » * / X^ +^ ) H ' ( ^ , U , . . . , U , 2 t• • • r 2 9 4 '

and maps the generatrix joining the points x E S™ and x 1 E Sn onto

the arc of the great circle on sm+n+ which joins the images of x
and x 1, in such a way that the lengths are linearly transformed (see
Fig. 2).

9. Since
con(Sm ,ort^) = (Sm ,ort^) ® (1,0) (see 4), and
A
su(Sm ,ort1) = (S (Ort^ 0 (S jort.j) (see 5), the homeomorphisms

con(Sm ,ort1) -> Dm+1 and su(Sm ,ort.|) -> Sm+1 , defined in 8 , lead for
n ^ 1 to the canonical homeomorphisms
 54
and

Dn = (S1 ,ort.) 0 ... 0 (S^ort.) 0 (1,0).

v------ ]-----v------------
n- 1

Now one can define the maps

id ® ... ® id ® pr: Dn = (S ,ort^) ® ... 0 (S ,ort^) ® (I/O) ->

(s\ort^) ® ... (s\ort^) ® (1/(0 U 1) ,bp) = Sn ,

and
(pr ® ... ® pr ® id I) ° pr: In = I x ... x i -v

(1/(0 U 1) ,bp) 0 ... 0 (1/(0 U 1) ,bp) 0 (1,0) = Dn .

We denote them by DS and ID. It is clear that DS takes

Int Dn homeomorphically onto Sn ^ ort^ and takes Fr Dn into ort^ ,

whileID takes homeomorphically Int in onto Int Dn , and Int in ^

ontoSn ^ \ ort^ , and carries Fr ln ^ Int in ^ into ort^ .

Since the map DS is closed (see 1.7.9), its injective factor

fact DS : Dn /Sn ^ Sn is a homeomorphism. Consequently, for n ^ 1
the quotient space Dn /Sn ^ is canonically homeomorphic to Sn .

The Mappings (X,xQ) ® (y /Yq) ^ z and X C(Y,y0 ;Z,zQ)

10. Let (X,Xq ), (Y,Yq ) and (Z,Zq ) be pointed topologica

spaces. If <J): (X,Xq ) ® (Y,yQ) Z is continuous and preserves base

points , then the formula [(J)U (x)](y) = <Mpr(x,y)) defines a continuous,

base point-preserving mapping (f)U : X C (Y ,y^ ?Z ,z^) .

Let ip: X C(Y,Yq;Z,Zq) be a continuous, base point-

preserving mapping, and suppose that Y is Hausdorff and locally
compact. Then the formula ^ (pr(x,y)) = [^(x)](y) defines a continuous,
base point-preserving mapping i(jn : (X,xQ) ® (Y,yQ ) Z.

Indeed,

(f)U = ab [ ((J) o pr: X x y ^ Z)V],

while

o (pr: X X Y (X ,xQ) 0 (Y ,yQ)) =

= [i|> o (in: C(Y,y0 ;Z,z0 ) -► C(Y,Z))]A .

 55

11. Given arbitrary pointed topological spaces (X,Xq ),

(Y ,Y q) , and (Z,Zq), the mapping

C((X,xQ ) ®(Y ,y Q) ,bp ;Z fzQ) -> C (X,xQ ;C (Y,y0 ;Z/z0) ,const)

given bythe formula $ » ^ (see 10) is continuous. If X and Y

are Hausdorff and compact, then this mapping is a homeomorphism and its
inverse is given by the formula iJj *+ ^n .

The preimage of C(X,A,xQ ?C(Y,yQ ;Z ,zQ),C(Y,B,yQ ;Z,C,zQ),const)

under the mapping $ h- <j)U is just C((X,x0) ® (Y,y0 ),pr(A x B) ,bp;Z ,C ,zQ),
which shows that <J> h - $ is continuous. Assume now that X and Y
are Hausdorff and compact, and consider the mapping

C (pr,id Z) : C((X,xQ) ® (Y,yQ),Z) + C (X x Y,Z).

By 7.5, this mapping is an embedding. Consider the diagram

C(X,xQ ;C(Y,y0 ;Z,zQ) ,const) ------ ► C((X,xQ) ® (Y,yQ),bp;Z ,zQ)

C (X , C (Y , Z )) -------------------------------- > C(X X Y , Z)

where the horizontal arrows denote the mappings ip w* and ip » ipA ,

and the vertical arrows the composite mappings

C(X,xQ ;C(Y,y0 ;Z,z0),const) in « C (X,C(Y,yQ ;Z,zQ)) C(ld X,in) ->

---- > C (X,C(Y,Z))

and
C ((X,xQ) 0 (Y,y0) ,bp;Z,z0 ) in > C((X,xQ) ® (Y,yQ> ,Z) ---- ►

C (pr,id_Z) ^ c ^x x Y ,Z).

Since this diagram is commutative, the fact that C(pr,idZ)

is an embedding implies the continuity of ip ip^. That the mappings
<\> h- <f)U and \p \p^ are inverses of one another is plain.

9. Exercises

1. Show that for each topological space X and each compact

topological space Y the map pr^: X x Y + X is closed.

2. Show that the subset of Sn defined by the inequality

 56

9 ? ? 2 ^
x + ... + x, ^ x, + . .. + x in the standard coordinates of IR ,
1 k k +1 n
1c xi k
is homeomorphic to D x S
3.Let M, Xy , and $ , be as in 4.2, and let X denote
the union ofthe spaces X defined by the homeomorphisms ‘f’yy •• Show
that the maps immy : Xy ^ x are topological embeddings whenever M
has only two elements, but when M has three elements this is not
necessarily so.
4. Show that for n i 1 the spaces C (1,0,1;Sn ,ort^,ort2 )
and C(I,0 ,1 ;Sn ,ort1 ,ort1) are homeomorphic.
00
5. Let T be the set of all real sequences with
the topology defined by the prebase consisting of all sets of the form
{{x.}“ | a < x < b}. Further, let S be the quotient space of T ^ 0
1 1 s
(where 0 = {x^=0 }” ) by its partition into rays, i.e., into the sets

{{tx?}^ | 0 < t < 00} with £ T \ 0. Show that T is metrizable,

while S is regular, but has the peculiar property that every

continuous map S JR is constant.

§3. HOMOTOPIES

1. General Definitions

1. A continuous map f': X Y is homotopic to the

continuous map f: X Y if there is a continuous map F: X x I ->» y
such that F(x,0) = f (x) and F(x,1) = f f(x), for all x E X. Every
such map F is called a homotopy from f to f' (or connecting f
to f*). One says also that F is a homotopy of f. A map homotopic
to a constant map is also said to be null homotopic.
Often a homotopy F: X x I Y is interpreted as a family of
continuous maps ft : X -* Y, related to F via ft (x ) = F(x,t)
(0 £ t £ 1). Acoording to 2.7.6, the continuity of F implies that
this family is continuous as a map of thesegment I into C(X,Y).
Moreover, if X is Hausdorff, then the continuity of the family f
is equivalent to that of the map F.
Obviously, the constant homotopy F of a continuous map
f: X + Y, given by F'(x,t) = f (x) , connects f to f; if the
 57

homotopy F connects f to f', then the inverse homotopy, F',

defined by F' (x,t) = F(x, 1 -t) , connects f' to f; if the homotopy
F connects f to f' and the homotopy F' connects f' to f",
then their product F ", defined as

F ( x ,2t ), for t £ 1/2,

F M (x, t)
F'(x,2t-1), for t ^ 1/2,

is a homotopy connecting f to f". Thus homotopy is an equivalence

relation, which yields a partition of C(X,Y) into equivalence classes,
called homotopy classes. We denote the set of these classes by tt(X,Y).

2. Anexample is the rectilinear homotopy. Namely, let f

and f' be continuous maps of a space X into a subspace Y of IRn .
If for each x £ X the segment joining f(x) to f'(x) is entirely
contained in Y, then F(x,t) = (1-t)f(x) + tf'(x) defines a homotopy
from f to f', referred to as rectilinear.
Obviously, any two maps of an arbitrary space into IRn or
Dn are rectilinearly homotopic.

3. Let the maps f ,f ': X ■+ Y be homotopic. Then given any

continuous maps g: Y -> Y' and h: X' X, the maps g o fo h and
g o f' o h are homotopic.

In fact, let F: X x I -* Y be a homotopy from f to f*.

Then g ° F 0 (hx id I) is a homotopy from g ° f ° h to g ° f' ° h.

4. As3 shows, the mapping C(h,g) : C(X,Y) CiX'/Y1)

induced by two continuous maps h: X 1 X and g: Y Y' transforms
homotopy classes into homotopy classes. The resulting mapping
fact C (h,g) : tt(X,Y) + tt (X',Y ') is denoted by ir(h,g), and 3 implies
that it depends only on the homotopy classes of h and g.

Stationary Homotopies

5. Let A be a subset of the space X. A homotopy

F : X x I ■+ Y is said to be stationary on A or, simply, to be an
A-homotopy if F(x,t) = F(x,0) for all x € A and t € I. Two maps
which can be connected by an A-homotopy are A-homotopic.
As with usual homotopy, A-homotopy defines an equivalence
relation, dividing the set of continuous maps X Y which coincide on
A with a given map f: A + Y, into equivalence classes. The latter
are called A-homotopy classes or, in full, homotopy classes of continuous
 58

extensions of the map f to X. We denote the set of these classes by

7T (X, A; f ) .
Notice that a rectilinear homotopy from f to f* (see 2)
is stationary on the set of points where f and g agree.
If one wants to specify that a certain homotopy is ordinary,
i.e., not stationary, then one says that it is free.

Homotopy Equivalence of Spaces

6. A continuous map g: Y X is a homotopy inverse of the

continuous map f: X -* Y if the composition g ° f is homotopic to
id X and the composition f ° g is homotopic to id Y . Acontinuous
map which has a homotopy inverse is called a homotopyequivalence. If
there is a homotopy equivalence X Y, then one says that the space
Y is homotopy equivalent to the space X.
The following are obviously homotopy equivalences: the
identity map of any space, a map which is a homotopy inverse of a
homotopy equivalence, and the composition of two homotopy equivalences.
Thus, homotopy equivalence among topological spaces is an equivalence
relation. It divides the topological spaces into classes called
homotopy types (instead of saying that Y is homotopy equivalent to X,
one says also that X and Y have the same homotopy type) .
Every homeomorphism is clearly a homotopy equivalence.

7. If one of the continuous maps f: X -> Y and g: Y ^ Z,

and their composition g o f: x Z are homotopy equivalences, then
the other map is also a homotopy equivalence.

Indeed, let h be a homotopy inverse of g ° f, and suppose

that f is a homotopy equivalence. Then f o h is a homotopy inverse
of g. Similarly, if g is a homotopy equivalence, then h ° g is a
homotopy inverse of f.

8. 7T(X,Y) is a homotopy invariant. That is to say, if

g: Y + Y' and f : X -* X 1 are homotopy equivalences, then
7r(f,g): tt(X,Y) -> tt(X',Y') is invertible.

Evidently, if f' (respectively, g') is a homotopy

inverse of f (respectively, of g) , then the map tt(f 1 ,g ') is the
inverse of Tr(f,g).
 59

Contractible Spaces

9. A space X is contractible if the map id X is homoto

to a constant map.
]R and Dn are examples of contractible spaces (see 2).

10-. A space is contractible if and only if it is homotopy

equivalent to a point.

If id X is homotopic to a constant map cf), then the map

f: D -> X taking the value (p(X) , and the map g: X are homotopy
inverses of one another: indeed, f ° g = (J) and g ° f = id .
If now f: X and g: X are homotopy inverses of
one another, then id X is homotopic to the constant map fog.

11 . _If X is contractible, then any two continuous maps of

an arbitrary topological space into X are homotopic. In particular,
id X is homotopic to any constant map X X.

This is a straightforward consequence of 10 and 8 .

Deformation Retractions

12. Aretraction p of a topological space X onto one of

its subspaces A (see 1.4.10) is called a deformation (strong

deformation) retraction if the composition X ■ p-> A - — ►X is

homotopic (respectively, A-homotopic) to id X . If the space X admits
a deformation retraction (a strong deformation retraction) onto A,
then A is called a deformation retract (respectively, a strong
deformation retract) of X.
Obviously, if p : X A is a deformation retraction, then
p and the inclusion A X are homotopy equivalences, each being a
homotopy inverse of the other. It is clear also that any space which
admits a deformation retraction onto one of its points is contractible,
and that every point of a contractible space is a deformation retract
of the ambient space.

Relative Homotopies

13. Let X (respectively Y) be a space with a distinguished

sequence of subsets (respectively, B 1 ,...,Bn >. A map
 60

F: (X x I ,A 1 x I,...,An x I) -> (Y ,B ^ ,. .., ) is called a homotopy

connecting the continuous maps f,f! : (X ,A^ ,. ..,A^) -* (Y ,B ^ , .. .,B^)
if abs F is a homotopy connecting the maps abs f and abs f 1 . In
this case, it is evidentthat ababsF: A^ x I -+ B^ is a homotopy
connecting the maps ab abs f , ab abs f 1 : A_^ B^. Moreover, it is
readily seen that the homotopies (X x I,A^ x I,...,An x I) ->
(Y,B,j,...,B ) yield an equivalence relation. This relation divides
C (X ,A ,...,A^ ;Y ,B ^ ,...,B^) into homotopy classes forming a set denoted
by tt(X ,A ^ ,...,A ^ ;Y ,B ^ ,...,B^) . We may give an analogous definition of
the map 7i(h,g) from 4.
A continuous map g: (Y,B^ , .. .,B^) -> (X ,A^ ,.. .,A^) is said
to be ahomotopy inverse of the continuous map f: (X,A ^ ,♦..,Ar ) ■>
(Y,B^ ,..., B ^ ) if g ° f is homotopic to rel id X and fog is
homotopic torel id Y. A continuous map possessing a homotopy inverse
is called a homotopy equivalence. Two sequences (X,A^ ,.. .,A ^ ) and
( Y , B^,...,B^) are said to be homotopy equivalent, or to have the same
homotopy type, if they are related by a homotopy equivalence.
Propositions 7 and 8 , as they stand, apply to the case of relative
homotopy.

14. The situation discussed in 13 encompasses the case whe

X and Y are pointed spaces (in this case A^and B are points,
n = 1 , and the homotopies defined in 13 are just the homotopies
stationary at A^) . Moreover, the definition of a contractible space
given in 9 extends to pointed spaces (however, the homotopy from id X
to a constant map must be stationary at the base point). The same is
true for theorems 1 0 an 1 1 , as well as for the definitions of a
deformation retraction and deformation retract, given in 12 (x and A
must have the same base point, and the homotopy from the composition

X — A —— — > X to id X must be stationary at this point) . Also,

the remarks in 1 2 remain valid, while the definition of strong
deformation retraction is entirely unaffected by the presence of a base
point.

2. Paths

1. A path in a topological space X is any continuous ma

of the closed unit interval I into X. The points s(0) and s(1)
are called the origin and the end of the path s. Closed path are also
termed loops.
Given a path s, the formula t k s(1-t) defines a new path,
 61

_1
called the inverse of s and denoted by s . Given two paths s1 and
with s 1 (1 ) = s 2 (0 ), the formula

s^ (2 t ) , for t £ 1 /2 ,

s2 (2 t- 1 ), for t ^ 1 /2 ,

defines a path, called the product of the paths s^ and s2 , and

denoted by s1s2 * Obviously, (s~ 1 ) ” 1 = s and ( s , ^ ) " ^ s2 1 s”1.

Since I =x i f any path can be considered as a homotopy

of a map D -+ X. If one adopts such an interpretation, then the
inverse path becomes the inverse homotopy, while the product of paths
becomes the product of homotopies.
On the other hand, every homotopy between two continuous maps
f,f' : X -> Y defines a path in C(X,Y), joining f and f* (see
2.7.6), and again the inverse path corresponds to the inverse homotopy,
and the product of paths to the product of homotopies. If X is
Hausdorff and locally compact, then a homotopy connecting two maps
f,f": X Y may be even defined as a path in C(X,Y) joining f and
f '.

3. Since any path is a continuous map, it can be also

subjected to homotopies. Unfortunately, the generally accepted
terminology for such homotopies is not in complete agreement with our
definitions in subsection 1 (which are also generally accepted). More
precisely, when we consider paths, the homotopies and the homotopy
relation are understood always as (0 U 1 )-homotopies (i.e., homotopies
stationary at the extremities of the interval I) and (0 U 1 )-homotopy
relation, respectively. Moreover, a free homotopy of a loop is
understood always as a usual free homotopy whereby the path remains a
loop all the time (i.e., as a continuous map F: I x I X such that
F (0,t ) = F (1,t) for all t £ I).

3. Connectedness and k-Connectedness

1. The properties of topological spaces we study in this

subsection represent weaker versions of the contractibility in the
absolute case, and of deformation retractability in the relative case.
 62

Connectedness

2. A topological space is connected (see the Preface) if

each pair of its points can be joined by a path. Equivalently, X is
connected if the set tt(D°,X) contains just an element; see 2.2.
Since tt(D°,X) is a homotopy invariant, connectedness is
a homotopically invariant property. In particular, all contractible
spaces are connected. For example, 3Rn and Dn are connected for
every n.
For n > 0, Sn is also connected: any two points of S
can be joined by a path, which in fact is contained in Sn \ p, where
p is a third point (recall that the punctured sphere Sn ^ p is
n 0
homeomorphic to 1 ). S is not connected: a path joining -1 and 1
would be a continuous function on [0 ,1 ], taking two distinct values
but no intermediate ones.
The only connected subsets of the real line JR are the empty
set, the finite or infinite intervals, the finite or infinite semi­
intervals, and the closed intervals. Indeed, if a and 6 are the
exact lower and upper bounds of a connected subset A of IR, then A
contains the interval (a, 6).

3. Given an arbitrary topological space X, the property of

being joined by a path defines a relation between its points, which
obviously satisfies all the requirements for an equivalence relation .
This relation defines a partition of X into subsets which are the
maximal connected subsets of X, and are called the components of X.
Clearly, the set of components may be identified with tt (D ,X) . We
denote it by comp X .
Every continuous map f: X Y induces the map
fact f = 7T(idD^,f) : comp X -+ comp Y . This map does not change whenwe
replace f by an arbitrary homotopic map, and factf is invertible
whenever f is a homotopy equivalence (see 1.4 and 1.8). It is also
plain that if f(X) = Y, then factf (comp X) = compY . In particular,
the image of a connected space under a continuous map is connected.

4. X can be written as the union of two connected

subsets A^ and A2 with A^ fl A^ ^ 0, then X is connected.

Indeed, a component of X which contains a point x Q € A^n A 2

contains also A^ and A^r i.e., contains X.

5. Consider a partition of X into open sets. Then every

connected subset of X is contained in one of the elements of this
 63

partition. In particular, every subset of a connected space which is

both open and closed is either empty or the whole space X.

PROOF. Let A be a connected subset of X, and let U be

an element of the partition, such that U n A f 0. Consider the map
f: X which takes U into 1 and X \ U into -1. Since f is
continuous, f(A) is connected, whence f(A) = 1 and A c u.

k-Connectedness

r
6. The following properties of a continuous map f: S
with r ^ 0 are equivalent:

(i) f is homotopic to a constant map;

r +1
(ii) f extends to a continuous map D X;

(iii) the compositions f ° DS+ , f ° DS_: Dr X are

r— 1 r r
S -homotopic, where DS+ and DS_ are the embeddings of D in S ,
defined by
2 2 1/2
DS+ (x1,...,x ) = (x1 ,...,xr , (1—x 1 — ..>-xr ) )
and
2 2 1/2
DS_ (x ,...,xr) = (x1 ,... ,xr ,— ( 1 - x ^ ..*“X r) );

(iv) f i_s ort^-homotopic to a constant map.

The proof follows the following scheme:

(i)
!
l ^ (ii) (iii)
V ^>"7
(iv)

(i) =*> (ii) . A homotopy F: Sr x I X from f to a constant

map takes the upper base of the cylinder S x I into one point.
Consequently, F may be expressed as the composition of the map
Sr x I ■+ Dr + \ defined by ((x^ ,...,x^ +^) ,t ) h- (x^ (1 -t) ,...,x^ +^ (1 -t)),

and a continuous map g: Dr 1 X (see 2.3.4 and 1.7.9), and it is

clear that g = f.
sr
r +1
(ii) => (iii) and (ii) => (iv) . Suppose g: D X is a
continuous extension of f. Then the formulas

((x^ /• •• ) ,t) g(x^,...fxr f(1 ” 2 t)( 1 -x^-... -xr ) )

and
 64

((x1 ,. -./Xr+1) ,1 ) ^ g(t+( 1 -t)x1 , (1 -t)x2, ..., (1 -t)xr + 1 )

define an Sr ^-homotopy Dr x I -> X from f ° DS + to f ° DS_, and

an ort^-homotopy Sr x I -+ X from f to a constant map.

(iii) =* (ii) . An Sr ^-homotopyf : Dr x I -> X from f ° DS +

r
to f ° DS_ takes every generatrix of the cylinder S x i into one
point. Consequently, F can be expressed as the composition of the map
Dr x I Dr + 1 , defined by

((x1 , ...,xr) ,t) n- (x1 , ...,xr , (2 1 — 1 ) (1 -x^— •• •-x^) 1 /' 2 ) ,

r+ 1
and some continuous map g: D -* X (see 2.3.4 and 1.7.9), and it is
clear that g = f.
sr
The implication (iv) => (i) is trivial.

7. A nonempty space X is said to be k-connected

(0 £ k £ °°) , if any continuous map S X with r £ k is homotopic
to a constant map, i.e., satisfies condition 6 (i). Theorem 6 shows that
this definition has three more equivalent formulations, based on
conditions 6 (ii) , 6 (iii) and 6 (iv). Moreover, since for any continuous
r r~ 1
maps D X which agree on S there is a continuous map

f: Sr X such that f ° DS+ = f and f ° DS_ = f ^ r we conclude that

a nonempty space X is k-connected if and only if any continuous maps
r r” 1 r~ 1
f^,f2: D -* X, r ^ k, which agree on S are S -homotopic.
Obviously, for nonempty spaces O-connectedness is nothing else
but connectedness. The 1-connected spaces are usually called simply
connected. Note that a 0-connected space is simply connected if and
only if any two paths with common extremities are homotopic.
The homotopy invariance of the sets tt (S ,X) implies that a
space which is homotopy equivalent to a k-connected space is itself
k-connected. In particular, every contractible space is ^-connected.

The Relative Case

8. The following properties of a continuous map

f: (Dr ,Sr ^) -> (X,A) with r > 0 are equivalent:

(i) f is homotopic to a constant map;

r— 1 r*
(ii) abs f is S -homotopic to a map which carries D
into a subset of A ;
 65

PROOF. (i) =* (ii). If F: (Dr x l,sr x I ) -> (X,A) is a

homotopy from f to a constant map, then the formula

F(x/dist(0 ,x),2 (1 -dist(0 ,x))), if dist(0 ,x) 5 (2 -t)/2 ,

(x,t) H-
F(2x/(2-t) ,t) , if dist(0,x) s; (2-t)/2,
r~ 1 2T
defines an S -homotopy D 'x 1 -> X from abs f to a map which
carries Dr into a subset of A.

(ii) => (i) . If G: Dr x I -> x is a homotopy stationary on

r—1 2T
S from abs f to a map which carries D into a subset of A,
consider the map F: Dr x I -> x given by

G ((x1 ,...,xr ),2 t), if t i 1 /2 ,

G( (2x1 (1-t) ,...,2x (1-t) ,1) , if t 5 1/2

Then rel F : (Dr x I,Sr ^ x i) (X,A) is a homotopy from f to a

constant map.

9. A pair (X,A ) is k-connected (0 £ k £ °°) if for any map

r r— 1 r— 1
f: (D ,S ) -* (X,A) with r £ k, abs f is S -homotopic to a map
whose image is contained in A.
It is clear that the pair (X,A) is 0-connected if and only
if each component of the space X intersects A. If k > 0, then
(X,A) is k-connected if and only if every continuous map
r r— 1
f: (D ,S ) -* (X,A) with r £ k is homotopic to a constant map; see 8 .
A pair which is homotopy equivalent to a k-connected pair is
k-connected. As a consequence, we see that when A is a strong
deformation retract of X, the pair (X,A) is “-connected;indeed,
(X,A) is homotopy equivalent to (X,X). It will be clear later that
the pair (X,A) is already ^-connected if A is a deformation retract
of X, or even when the inclusion A + X is a homotopy equivalence;
see 5.1.6.5.

4. Local Properties

1. A topological space X is locally contractible at the

point Xq € X if each neighborhood U of x^ contains.another
neighborhood V of x^ such that the inclusion V U is homotopic
to the constant map V -> x^. A topological space is locally contractible
if it is locally contractible at any of its points.
 66

If we replace in these definitions the homotopies by

XQ-homotopies, the we get the definitions of a space X which is
strongly locally contractible at the point x^, and of a strongly
locally contractible space X.

IRn , Dn and Sn are examples of strongly locally contractible

spaces.
2. A topological space X is locally connected at the point
Xq E X if each neighborhood U of Xq contains another neighborhood
V of Xq , such that any two points in V can be joined by a path in
U. A topological space is locally connected if it is locally connected
at any of its points.
It is clear that a locally contractible space is locally
connected. As an example of a connected space which is not locally
2
connected we may take the subset of ]R consisting of the lines
m^x^ + m 2 x 2 ' with m -j an(^ m2 integers.
3. A space is locally connected if and only if the components
of its open sets are open. In particular, in a locally connected space
every neighborhood of an arbitrary point contains a connected neighbor­
hood of this point.

PROOF. Suppose that X is locally connected, U is an open

subset of X, A a component of U, and Xq E A an arbitrary point.
Then U contains a neighborhood V of Xq such that any two points
in V can be joined by a path in U. Hence V c A and x Q E IntA .
This proves that in a locally connected space the components of the open
sets are open.

5. Borsuk Pairs

1. A topological pair (X,A) is a Borsuk pair if given any

topological space Y, any continuous map f: X -+ Y, and any homotopy
F: A x I Y of the map f |^' there is a homotopy X x i -* y of f
which extends F.
If (X,A,B) is a topological triple such that (X,A) and
(A,B) are Borsuk pairs, then (X,B) is obviously a Borsuk pair.

2. Let (X,A) be a topological pair. Then in order for

(X,A) to be a Borsuk pair it is necessary that (X x 0) U ( A x i) be
a retract of the cylinder X x i. When A is closed, this condition
is also sufficient.
 67

PROOF OF THE NECESSITY. Any homotopy of the map

in: X = X x 0 -> ( X x O ) u (Ax I ) which extends the homotopy
in: A x I (X x 0) U (A x I) is a retraction of the cylinder X x I
onto ( X x O ) U (A x I).
PROOF OF THE SUFFICIENCY. Let p:XxI+(XxO)U (A x I)
be a retraction. Then given any topological space Y, any continuous
map f: X -* Y, and any homotopy F: A x I y of the map ^[a * the
composition

X x I -P-* ( X x O ) U (A x I) -iL, y,

where G is defined by

f(x) , if t = 0,
G(x,t)
F (x ,t) , if x G A,

is a homotopy of f which extends F.

3. The following statement completes Theorem 2 in an essential

way. I_f X is Hausdorff, then the assumption that (X x 0) U (A x I)
is a retract of the cylinder X x I implies automatically that A is
closed. Indeed, it suffices to note that the above hypothesis implies
that (XxO) U ( A x I) isclosed in X x I (see1.5.5), and that A
is the preimage of this setunder the map X X x i, x *+ (x,1).

4. If the sets A and B form a closed cover of the space

Xand (A,A fl B) is a Borsuk pair, then (X,B) is also a Borsuk pair.

This is a consequence of 2; in fact, any retraction

p: A x I -> [ A x 0] U [ (Afl B) x I]

defines a retraction X x i (x x 0) U (B x i) by

'p(x, t) , if x e A,
(X,t)
(x,t), if x e b.

5. If (X,A) is a Borsuk pair and A is closed, then

(Z x x ,Z x a ) is a Borsuk pair for every topological space Z.

PROOF. If p is a retraction of the cylinder X x I onto

(X x 0) U (A x i), then id Z x p is a retraction of the cylinder
(Z x X) x I = Z x (Xxl) onto [(Z x X) x 0] U [ (Z x A) x I] =
= Z x [ (X x 0) U (A x I) ] .
 68

Borsuk Pairs and Deformation Retractions

6. If (X,A ) is a Borsuk pair and theinclusion A X is

ahomotopy equivalence, then A is a deformation retract of X .

PROOF. Let 7T: X A be a homotopy inverse of the inclusion

A X. Extend the homotopyfrom 71 |^ “ 71 ° in: A A to id A to a
homotopy of the map tt ; this yields a homotopy from tt to a retraction
of X onto A, which we denote by p. Since the composition

X A —— — > Xis homotopic to id X , thecomposition X — A— ^ — >X

is also homotopic to id X , and thus p is adeformation retraction.

7. lf_ A is a deformation retract of X and

(X xI , (x x 0) U (A x i) u (x x 1 )) is a Borsukpair, then A is a
strong deformation retract of X.

PROOF. Let p : X -> A be a deformationretraction, and let

f: X x i x be a homotopy from id X to the composite map

X A — — — > X. Define a homotopy

g: [(X x 0) U (A x i) u (X x 1 )] x i x
by
f x, if t1 = 0 ,
g((x,t1 ),t2) = |f(x,(1-t2 )t1), if x E A,
I f(P(X),1-t2), if t 1 = 1,

and extend it to some homotopy G : (X x I) * I -> X of the map

f: X x i ->■ x. It is clear that (x,t) h- G((x,t),1) yields an
A-homotopy Xx I ->• x fromid X to in ° p.

8. I_f (X,A) is a Borsuk pair and B is a strong deformation

retract of the space A, then the map rel: (X,B) (X,A) is a homotopy
equivalence.

PROOF. Consider a B-homotopy from id A to the composition

of a strong deformation retractionA -> B and the inclusion B -»■A.
Now extend it to a homotopy G of id X . It is clear that the map
(X,A) -+ (X/B), x » G(x,1), is a homotopy inverse of rel.
 69

Local Characteristics of Borsuk Pairs

9. Suppose that (X,A) is a Borsuk pair with X normal,

Y is any topological space, and f: X Y is any continuous map. Then
given anyhomotopy F of the mapand any neighborhood U of A,
there is an (X^ U )-homotopy of f extending F.

PROOF. Let G be a homotopy of f extending F, and let

(J) be anyUrysohn function for the pair X\ U, A. Then the formula
(x,t) *-> G(x,tct>(x)) defines an (X^U)-homotopy of f extending F.

10. I_f (X,A) is a Borsuk pair, then there exists a

neighborhood U of_ A such that the inclusion U X .is A-homotopic
to a map which takes U into a subset of A. Iif Xis normal and A
is distinguishable (in particular, if X is metrizable and A is
closed), then this condition is also sufficient, i.e., the converse
of the above statement is valid.

PROOF OF THE NECESSITY. If a: X x i (X x 0) U (A x I) is

a retraction, then the set U of all points x E X with
a(x,1) € A x (0,1] is open, and the composition
pr
U x I — — > X X I — £-* (x X 0) U (AX I) — » X X I ---- -— > X

is an A-homotopy from the inclusion U X to a map which takes U

into a subset of A.

PROOF OF THE SUFFICIENCY. Let F : Ux I x be an

A-homotopy such that F(x,0) = x and f(x,1) E A for all x E U, and
let (J): X I be a Urysohn function for the pair A,X ^ U, which
distinguishes A (see 1.5.9). The formula

fF (x ,min (t/c{) (x) ,1 )) , if x E U ^ A,

G (x ,t) = <
x, if x E A,

defines amap G: U x I -> X , and Theorem 2.2.14 shows that G is

continuous. This in turn implies the continuity of the map
H: X x I X x I defined by

(G (x ,max (0 ,t-(|> (x) )) ,max (0 ,t-2(J)(x) )) , if x E U,

H (x ,t ) =
(x ,0) , if x E X \ U.

It is readily seen that H(X x I) = (X x 0) U (A x I) and that

 70

ab H :X x i (x x0) U (A x i) is a retraction.

11. Let (X,A ) be a topological pair such that A is a

strong deformation retract of one of its neighborhoods. If X is
normal and A is distinguishable (in particular, if X is metrizable
and Ais closed), then (X,A) is a Borsuk pair.

This is a corollary of 10.

12. If (X,A) is a Borsuk pair, then given anyneighborhood

V of A, there is another neighborhood W of A, such that W c= v
and the inclusion W -> V is A-homotopic to a map which takes W into
a subset of A.

PROOF. By 10, there exists a neighborhood U of Aand an

A-homotopy F: U x I -+ X such that F(x,0) = x and F(x,1) EA for
all x E U. Now 2.2.13 shows that every point x E U has a neighborhood
W in U with F(W x I) c V. Set W =U W . It is clear that
x x xEA x
W c= V, F(W x i) c v, and that ab F : W x I -> v is an A-homotopy from
the inclusion W -> I to a map which takes W into A.

13. I_f X isa topological space and x EX issuch that

(X,x) is a Borsuk pair, then X is strogly locallycontractible at x.
If X is normal and locally contractible at a distinguishablepoint x,
then (X,x) is a Borsuk pair.

This is a consequence of 12 and 10.

6 . CNRS-spaces

1. A subset A of atopological space X is said tobe a

neighborhood retract of X if A is a retract of one of its neighbor­
hoods in X.
The retracts and the open sets are trivial examples of
neighborhood retracts.

2. rf A is aneighborhood retract of X and B is a

neighborhood retract of A, then B is a neighborhood retract of X.

Indeed, let p: U -> A be a neighborhood retraction for A

in Xt and let a: V -* B be a neighborhood retraction for B in A.
Then a° (p |w): W B, where w = p_1 (V) , is a neighborhood
retraction for B in X.

3. A topological space is a CNRS-space or, simply, a CNRS,

if itis compact and can be embedded in a Euclidean space (of a certain
 71

dimension) as a neighborhood retract; CNRS is the abbreviation of

compact neighborhood retract of a sphere.

Dn and Sn are obvious examples of CNRS’s .

4. A compact neighborhood retract of a CNRS is a CNRS.

This is a resultof 2.

5. The image of any embedding of a CNRS in a normal space is

a neighborhood retract.

PROOF. Let f: X + Y be an embedding of the CNRSX in the

normal space Y, and let g: X + TRn be an embedding of X such that
g(X) is a neighborhood retract of JRn . Further, consider
f ^ = [abf: X f (X) ] and g^ = [abg: X + g (X) ] , and let p : V -> g (X)
be a neighborhood retraction. Since f(X) is closed (see 1.7.9),
g ° f .^: f(X) ^ nRn extends to a continuous map h: Y + JRn (see 1.5.12).
1 -1
It is clear that U= h (V) is a neighborhood of f(X) , and that
f^ ° g^ o p o [abh : U -> V ] is a retraction of U onto f (X) .

6. Given anycompact neighborhood retract X of JRn , there

is a number e > 0 such that any two maps, f and g, of an arbitrary
space Y into X which satisfy

sup^EY dist(f(y),g(y)) < e

are homotopic. Moreover, one may choose a homotopy stationary on the

set where f and g agree.

PROOF. Let a: U + X be a neighborhood retraction. We show

that one may take e to be the distance between X and HRn ^ U
(which is positive by 1.7.15).
Let f,g: Y X be continuous and satisfy
SUPyEY dist (y) '9 (y) ) < For anY Point y C Y, the segment with the
extremities f(y) and g(y) lies in U. Consequently, thecomposite

maps Y — X —^ — > U and Y —2-» x —— — > U can be connected by a

rectilinear homotopy F: Y x I -* u, and it is plain that
a ° F : Y x i + x is a homotopy from f to g. Furthermore, a ° F is
is stationary on the set where f and g agree.

7. if A is a neighborhood retract of a CNRS X, then (X,A)

is a Borsuk pair.
PROOF. Let o: U+ A be a neighborhood retraction. Consider
X as a neighborhood retract of ]Rn and pick e as in 6 . Denote by
V the neighborhood of A in X consisting of all the point x E U
for which dist (x,a (x) ) < e, and let <f> be the composition
 72

CM tt
V >A — ► X.

Then dist ((J)(x) ,x) < c for x £ V, and <Mx) = x for x £ A. Hence
the inclusion V X is A-homotopic to <j), and since (j)(V) = A, we
can apply Theorem 5.10.

8. Every CNRS is strongly locally contractible.

This is a consequence of 5, 1 , and 5.13.

Information

9. The converse of Theorem 8 is also true: every locally

contractible compact subspace of a Euclidean space is a neighborhood
retract of this space. For a proof, see [14].

7. Homotopy Properties
of Topological Constructions

1. In this subsection we establish the homotopy invarianc

of some of the constructions described in § 2 and study the homotopy
properties of the resulting spaces.

Products

2. Obviously, two continuous maps f,g: Y + X ^ x ... x

are homotopic if and only if pr^ ° f fpr_. o g : y X^ are homotopic
for all i = 1,...,n (see 2.2.4). In particular, X. x ... x x is
1 n
k-connected if and only if all the X^'s are k-connected (0 ^ k < °o) .
It is also clear that if the maps Y w . ..fq : x -* Y : X„
1 ^n n1 n1
are homotopic to the maps f .: X. -> Y„ ,...,f : X + Y , then
1 1 1 n n n
g^ x ... x g : X. x ... x X + Y. x ... x y and f . x ... x f :
1 n 1 n 1 n 1 n
X 1 x ... x x Y x .. . x y are homotopic; moreover, if f f
i n i n I n
are homotopy equivalences, then so is f„ x ... x f .
1 n
3. If A is a deformation retract (strong deformation
retract) of X, then A x y is a deformation retract (respectively, a
strong deformation retract) of the product Xx Y, for any space Y.
In particular, if X is contractible, thenthe fibers x xY of the
product X x y are deformation retracts of X x y.
 73

Quotients

4. Since the projection X -* X/p is continuous, the quotient

space of a connected space is connected (see 3.3), Moreover, in order
that the quotientX/A of X by its subspace A be connected, it is
even enough that the pair (X,A) be O-connected; if the components of
X are open, then the connectedness of X/A implies the O-connectedness
of (X ,A ) .
We shall see later that neither the k-connectedness with
k > 0 , nor the contractibility are, generally speaking, preserved when
one takes quotients.

5. Let p and q be partitions of the spaces X and Y.

If the maps f^s X Y form a homotopy and take the elements of p
into elements of q , then the maps fact f : X/p Y /q also form a
homotopy.

We have to verify that the map G: (X/p) x I Y /q, given by

G(x,t) = (fact ft ) (x) , is continuous. To do this, it suffices to note
(see 2.3.4) that the composition X x I _ P r x I_) (x/p) x I — — > Y / q
is continuous, and that the map pr x id I is factorial. The first is
a consequence of the commutativity of the diagram

X X I Pr X idI (X/p) x

F I G
y ------ EE-------- , Y/,

where F is F(x,t) = > while the second follows from Theorem

2.7.8.
6. I f_ f ,f 1 : (X,A) -> (Y,B ) are homotopic, then the maps
rel fact f , rel fact f ': (X/A,pr(A)) -► (Y/B,pr(B)) are also homotopic.
If f is a homotopy equivalence, then so is rel fact f.

The first assertion is a corollary of 5, while the second is

a consequence of the first: if g: (Y,B) (X,A) is a homotopy inverse
of f, then rel fact g : (Y/B,pr(B)) -v (X/A,pr(A)) is a homotopy
inverse of rel factf.
7. If (x,A) is a Borsuk pair and A is contractible, then
rel pr : (X,A) (X/A,pr (A) ) is a homotopy equivalence.
PROOF. Let F: A x i A be a homotopy from id A to a
 74

constant map, and let G: X x I X be a homotopy of id X extending

F. Denote by g: X + X the map x G(x,1), which is homotopic to
id X . Since g is constant on A, the map fact g :X/A X is
meaningful; moreover, since fact g (pr (A) ) c A, the map
rel fact g : (X/A;pr(A)) (X,A) is also meaningful. Let us check that
rel fact g is a homotopy inverse of rel pr . Consider the homotopies
rel G : (x X i ,a x i ) - > (X,A ) and rel fact G : ( (X/A) x l,pr(A) x I ) -*
(X/A,pr (A) ) . Thefirst connects the maps

rel id X, rel g : (X,A) + (X,A)

while the second connects the maps

rel id (X/A) , rel fact g: (X/A,pr(A)) -+(X/A,pr(A)).

It is clear that [rel g : (X,A) (X,A) ] =

= [relfactg: (X/A,pr(A)) (X,A)] ° [relpr : (X,A) (X/A,pr(A))] and

[ rel fact g : (X/A,pr (A) ) (X/A,pr (A) ) ] =

= [relpr: (X,A) (X/A,pr(A))] o [relfactg: (X/A,pr(A)) -> (X,A)].

Attachings

8. I_f (X^ ,C) is a Borsuk pair and <£,<£': C ^ X^ are

homotopic, then the spaces X^ X^ and X^ U , X^ are homotopy
equivalent. Moreover, there is a homotopy equivalence
f: X 2 X^ ■+ X^ U , X ^ such that the following diagram is commutative:

imm. lmm. (1 )

X, u X,
2 op 1 X2 V X,
PROOF. Let $: C x i X2 be a homotopy from cp to cp1 ,
and let a: X^ x 1 + (x^ x 0) U (C x 1 ) be a retraction. Define the
maps h: (X^ x 0) U (C x I) + X 2 U X^ and h !: (X ^ x 0) U (C x I) +

* X2 V X1 b*
imm1 (x) , if t = 0 ,
h(x,t) =
imm2 ° <j>(x,t), if x £ C,
 75

and
imm (x) , if t = 0,
h ’ (x,t) =
iiran2 o <!>(x,1-t) , if x £ C.

Now define two more maps, f:X„U X + X U X and

2 cp 1 2 tp' 1
9 = X2 V X 1 ^ X 2 Ucp X 1 V i a

f o imm^ (x) = h* o a(x,1 ), f o imm2 (x) = imm2 (x),

and
g o imm^ (x) =h ° cr(x,1 ), g ° imm2 (x) = imm2 (x) .

It isclear that all thesemaps are continuous and that the diagram (1)
is commutative. Moreover, it is readily seen that the map
( X 2 U<p V X1 ^ X 2 Utp' X1 9 ive n by

(imm^ (x),t) ^ h o a o i|;(a(x,t),t), (imm2 (x) ,t) h- imm2 (x) ,

where : (X^ x I) x I x I, i^((x,u),t) = (x,max (0 ,t-u) ) , is a

homotopy fromid (X^ U X ) to g o f. Thus, g ° f is homotopic to
id(X2 X^ ) and, similarly, f ° g is homotopic to id(X2 U ,X^ ) .

9 (LEMMA) . If the maps f : Y + Y 1 and f *:Y 1 Y are

homotopyinverses of one another, then given any homotopy F ': Y x I Y
from id Y to f* ° f, there is a homotopy F 1: Y' x I y' from
f of', such that the maps f ° F , F 1 ° (f x id I) : Y x I Y1 are
[(YxO)U (Yx 1 )]-homotopic.

PROOF. Let G : Y ' x I -+ Y ' be an arbitrary homotopy from

id Y * to f o f 1, and let F* be the product of the following three
homotopies: G, f ° F ° (f' x id I) , and the inverse of the homotopy
2
G° (f x id I) o (f* x id I) . Divide the square I into eight pieces,
as shown in Fig. 3 (the points A 1 ,A2 ,A3 ,A4 have the abscissae
0,1/2,3/4,1, and ordinate 0, the points B 1 ,-..,Bg the abscissae
0 ,1 /8 ,1 /4 ,1 /2,3/4 ,1 , and ordinate 1/2, and the points ' C2 the
abscissae 0 ,1 , and ordinate 1 ).
2 2
Now define affine maps : I P^ and a^: I p2 with
the following properties: a^(0 ,0 ) = B^, a^( 1 ,0 ) = B^, a^(0 ,1 ) = A 2,
a 2 (0,0) = B5 , a 2 (1,0) = B 4 , and a 2 (0,1) = A 4 . Further, for y E Y
define a map $ : I2 Y' through the conditions:

(a1 (t1 ,t2)) = f (F(F(y,t2) ,t^

and
4> (a2 (t1 ,t2)) = G (f o F(y,t2 ),t1);
 76

if 0 £ t £ <|) (t,0) = G (f (y ) ,21 ) ; the restriction

1/2, then 6
y YI
is constant on all segments parallel to the line A^B^; <(> IQ is

constant on all segments passing through the point is

D^;
"y |Q,
constant on the vertical segments; A I is constant on all segments
y Iq 4
parallel to the line C.Bc; <j> |_ is constant on all segments passing
1 ^ yIQ 5
through the point D ; and, at last, (p I is constant on the
1 y

vertical segments (those segments on which <j) is constant are depicted

in Fig. 4). It is clear that the formula ((y,t^),t2 ) ^

defines an [ (YxO)U (Yx1)]-homotopy (Y x I) x I -> Y ' from f ° F to

F' o (f x id I) .

\ A q
\ b \ b3 b.

w \ \\
\\ \ \

\ P2 \ Q b
\De

Qa 'P'

At Aj A^
Fig. 3 Fig. 4

10. Let (X,C) be aBorsuk pair with C closed, and let

ip: C Y be a continuous map. If f : Y -+ Y ' is a homotopyequivalence,
then

fa c t (id XJ—
I If)
1
: Y U X
tp Y ' U_
f °ip X
is also a homotopy equivalence.

PROOF. Let us fix: (i) a homotopy inverse of f,

f': Y' ->Y; (ii) a homotopy F: Y x I -*■ Y from id Y to f' ° f ;
(iii) a homotopy F': Y' x I Y' from id Y' to f ° f', together
with an [ (YxO) U (Yx1)]-homotopy G: (Y x I) x I y ' from f ° F to
F ' o (f x id I) (see 9) ; (iv) a retraction p : X x I (x x 0) u (C x I)
Further, define g : (X x 0) U (C x i) y U X and
ip
g': [(X x 0) U (C x I) ] x I y ' U£
f otp X by
2
 77

imm (x), if t = 0,
g(x,t)
imm2 o F(ip(x),t), if x G C
and
imm1 (x) , if t1 = 0 ,
g '((x,t1),t2)
imm2 o G ((ip (x) ,t 1 ) ,t2) , if x G C.

Together, these maps yield a map h : Y ' U_ X -* Y U X which

f oip CP
satisf ies

h(imm^(x)) = g o p(x,1), h(imm2 (y')) = imm2 (f'(y')).

Now it is clear that the formulas

H(imm^(x),t) = g o p(x,t), H (imm2 (y),t ) = imm2 (F(y,t ))

define a homotopy H:(YU

<p X) * I Y U
(p X from id(Y U
cp X) to
h ° fact (id X | |f) . Also, we see that the formulas

(imm1 (x) ,t) = g ' (p (x,t) ,1) , (imm2 (y ') ,t) = imm2 (F '(y ',t ))
and
(imm1 (x),t) = g '(p(x,1),t), (imm2 (y'),t) = imm2 (fof'(y)),

define two homotopies H',H': (Y1 U_ X) x I -»• Y' U,. X, whose

1 2 f°(p f °(p
product is a homotopy from id(Y' X) to fact(idx| |f) ° h.
Therefore, h is a homotopy inverse of fact(id XI |f) .

More Special Constructions

11. con X is contractible for anv X. suX is connected

for any X. X^ * X 2 is connected for any X^ and X2 .

The contractibility of the cone is obvious. The connectedness

of the suspension is a consequence of the fact that it is a quotient
space of the cone. To prove that the join * X2 is connected, we
may check that any two points pr (x.j ,x2 ,t) , pr (x.j ,x2 ,t ') G X^ * X2
[(x1 ,x2,t), (x^,x^,t') G X 1 x x2 x i] can be joined by the path

pr (x1 ,x2 ,3 t (1- t ) + t ) , if t £ 1/3,

t i pr (x ^ ,x 2 ,2 - 3 x ) , if 1/3 £ x £ 2/3,

pr (x},x£,t' (3x-2) ) , if T * 2/3.

 78

12.Given twohomotopic maps f,g: X + Y, a straightforward

application of Theorem 5 shows that the maps suf ,sug : su X su Y
are also homotopic. As usual, we may deduce that suf is a homotopy
equivalence whenever f is one.
Moreover, the same Theorem 5 shows that given homotopic maps
f^ ,g^ : X ^ Y^ and homotopic maps ^2'g2 : X2 Y2 ' the maPs
f * f2'g l * g2 : X 1 * X2 Y 1 * Y2 are also homotopic. Similarly, we
conclude that f^ * ^ 2 a homotopy equivalence whenever f^ and f^
are homotopy equivalences.

13. Given a continuous map f: X^ X^r r t f : Cyl f -+ X^

(see 2 .6 .1 0 ) is obviously a strong deformation retraction of the
mapping cylinder Cyl f onto X^. Hence X^ is a strong deformation
retract of Cyl f .
The inclusion X^ Cyl f is a homotopy equivalence if and
only if f is a homotopy equivalence. In fact, the composition

X^ — ►Cyl f — rt * X2 coincides with f, and rtf is a homotopy

equivalence.

The Case of Pointed Spaces

14. The results obtained in 2, 3, 6 , 7, 10, 11, and 12 have

obvious analogs for pointed spaces. Let us add that those theorems which
are the analogs of 1 2 are valid for both bouquets and tensor products:
if f^, g ^ : (X^,x^) Yy) are homotopic for each vi, then
Vfy' (V(X^,x ),bp) (V (Y ,y ) ,bp) are homotopic, and if all the
fy's are homotopy equivalences, then so is V^y- Similarly, if
f ^ g ^ ( X ^ x ^ + (Y 1 ,y 1 ) are homotopic, and f 2 'g 2 : (X2 'x 2 ) ^ (Y2 'X 2 }
are homotopic, then ® f2 'g l 0 g 2 : 0 (X2 #x 2 ),bp) ->
((Y^ ,y ^ ) ® (Y2 ,y2 ),bp) are homotopic; if f and f are homotopy
equivalences, then so is f^ ® f2 .

8. Exercises

00
1. Show that the sphere S is contractible.

2. Show that if (X,A) is a Borsuk pair and X is

contractible, then the quotient X/A is homotopy equivalent to su A

3. Suppose that the product of two topological spaces is

homeomorphic to the suspension of some othertopological space. Prove
 79

that either both factors of the product are contractible, or one of

them reduces to a point.

4. Let f^: be a homotopy equivalence. Show that

X^ is a strong deformation retract of Cyl f .

5. Show that if X is metrizable and x E X is such that

(X,x) is a Borsuk pair, then the projection suX su(X,x) is a
homotopy equivalence.

6. Given an arbitrary connected topological space X and

two arbitrary points x,y £ X, show that the subset of C(I,0;X,x)
consisting of all paths passing through the point y is contractible.
 Chapter 2. Cellular Spaces

§1. CELLULAR SPACES AND THEIR TOPOLOGICAL PROPERTIES

1. Fundamental Concepts

1. A decomposition* p of a topological space X is called

cellular if there is a function d taking the set of elements of p
into the nonnegative integers, such that for every element e of p
there exists a continuous map -* X with the following two
properties: (i) it maps Ind D ^ 0 ^ homeomorphically onto e;
(ii) it maps ^ onto a union of elements of p on which d
takes values smaller than d(e).
The elements of a cellular decomposition and their closures
are called cells and closed cells, respectively. The number d(e) is
the dimension of the cell e and is usually denoted by dim e ; the
n-dimensional cells are also termed n-cells. Any continuous map
die) —
D -> X with the properties (i) and (ii) is said to be characteristic
for e; we use the symbol chae as a standard notation for such a map.

Obviously, cha^ ( D ^ m e ) c C l e , and if X is Hausdorff,

then cha^(D^im e ) = Cl e . In particular, every closed cell in a

cellular decomposition of a Hausdorff space is compact. Moreover, in
the case of a Hausdorff space, given any cell e, Cle \ eis covered
by cells of dimension lower than dim e .

2. A cellular decomposition is said to be rigged (or equipped)

if for each of its cells there is fixed a characteristic, map. The
resulting family {cha^: D d i m 0 x} is called a rigging (or an
equipment) of the decomposition, and the map

* Translator's note: in this chapter we use instead of "partition" the

equivalent term "decomposition", which is generally accepted in the cellular case.
defined bv
-1
the relations cha ° ine = cha_e is called the -----
total
characteristic map.
3. According to the general definitions of Chapter 1 (see
1.2.4.3), the cover of the space X by the closed cells of a cellular
decomposition p defines a new topology on X. A subset of X is
closed in this topology if and only if its intersections with the closed
cells of the decomposition p are closed in the initial topology of X.
The new topology on X is called the weak or cellular topology, and the
process by which we pass from the initial to the cellular topology of
X is called a cellular weakening of the initial topology. The weakening
of a topology can only enlarge the supply of open and closed sets; in
particular, a Hausdorff space remains Hausdorff. In any case, it does
not affect the topology of the closed cells, and so the decomposition
p remains cellular and retains its characteristic maps.
If X is Hausdorff and the decomposition p is endowed with
a rigging{cha^}, then the cellular topology can be described
effectively in terms of the corresponding total characteristic map:
a subset A of X is open (closed) if and only if its preimage
-1
cha (A) is open (respectively, closed). In other words, the cellular
topology is exactly that topology on X which transforms the injective
factor of the map cha into a homeomorphism. Recall that the injective
factor of cha is a map of the quotient space of the sum J Leex/p0^ ™ &
by its partition zer(cha) onto X. The equivalence of these two
definitions of the cellular topology results from the fact that the maps
chae are closed (see 1 .1 .7.9).

4. Acellular space is a Hausdorff topological space endowed

with acellular decomposition which satisfies the following two
conditions: (C) every closed cell intersects only a finite number of
cells ; (W) the closed cells constitute a fundamental cover of the
space. Obviously, condition (W) implies that the cellular topology
coincides with the initial one. The notations (C) and (W) are generally
accepted, and originate from the terms closure finiteness and weak
topology.
Property (C) is clearly preserved under the cellular
weakening of the topology. Therefore, a Hausdorff topological space
having a cellular decomposition satisfying (C) becomes a cellular space
via cellular weakening.
Usually the terminology specific to cellular decompositions
 83

is applied to cellular spaces too. In particular, a cellular space may

be finite, countable, and rigged. Thus*a finite cellular space is one
that has a finite number of cells, but not necessarily a finite number
of points.
The dimension of a cellular space is the supremum of the
dimensions of its cells; the dimension of the empty space (which is not
excluded from the family of cellular spaces) is taken equal to -1 .
The dimension - finite or infinite - of the cellular space X is
denoted by dim X .
We let cell^X denote the set of all r-cells of X.

5. The simplest cellular spaces are the discrete spaces

decomposed into 0-cells (isolated points). It is clear that all
O-dimensional cellular spaces are of this type: a decomposition of a
nondiscrete Hausdorff space into 0-cells does not satisfy condition (W).
The decomposition of the ball Dn (n > 1) into the n-cell
Int Dn and the 0-cells which cover Fr Dn = Sn ^ is an example of
cellular decomposition which satisfies condition (W), but not condition
(C) .

The Locally Finite Case

6. In agreement with the general definitions given in 1.1.1.12,

a cellular decomposition is locally finite if every point of the given
space has a neighborhood which intersects only a finite number of cells.
An equivalent condition: every point has a neighborhood which intersects
only a finite number of closed cells.
Clearly, in a space possessing a locally finite cellular
decomposition, every compact subset has a neighborhood which intersects
only a finite number of cells. As a consequence, every locally finite
cellular decomposition of a Hausdorff space satisfies condition (C).
Theorem 1.1.3.6 shows that condition (W) is satisfied by any locally
finite cellular decomposition. We conclude that every Hausdorff space
endowed with a finite or locally finite cellular decomposition is a
cellular space.
7. A cellular space is locally finite if and only if every
cell intersects only a finite number of closed cells.

PROOF. In a locally finite cellular space the closure of an

arbitrarily given cell has a neighborhood which intersects only a finite
number of cells (see 6 ). It is clear that this neighborhood, and hence
 84

the given cell, do not intersect the closure of other cells.

Conversely, if every cell of a cellular space intersects only
a finite number of closed cells, then the axioms (C) and (W) imply that
the union of any collection of closed cells is closed. Consequently,
the complement of the union of all closed cells which do not intersect
an arbitrarily given cell e is a neighborhood of e. Since this
complement cannot intersect the closed cells which do not intersect e,
it intersects only a finite number of closed cells.

Subspaces

8 (LEMMA). Let A be a subset of the cellular space X with

the following property: if x £ A, then A contains the closure of
the cell in which x lies. Then any part of A whose intersections
with all the closed cells contained in A are closed is itself closed.

Indeed, if B c a is such a part, and e is an arbitrary

cell, then one can write the intersection B fl Cl e as
g
U^_-j [ (B fl Cl e^) fl Cl e ] , where e^ , ..•,eg are all Lhe cells of A
which intersect Cl e , and this shows that B fl Cl e is closed.

9. A subset of a cellular space which contains together with

each point the closure of its cell is called a subspace of the given
cellular space. Every subspace is a cellular space, with the cellular
decomposition induced by the cellular decomposition of the ambient space.
By Lemma 8 , this decomposition satisfies condition (W), and it obviously
satisfies condition (W).
As another consequence of Lemma 8 , we see that every subspace
of a cellular space is closed. Notice also that the union and the
intersection of any collection of subspaces are again subspaces, and
that every cover of a cellular space by subspaces is fundamental.
A pair consisting of a cellular space and one of its subspaces
is called a cellular pair. Cellular triples and cellular triads are
defined similarly.
Warning: a closed cell is not necessarily a subspace. For
1 2
an example, consider the bouquet (D ,0) V (S ,ort^), with the
decomposition into four cells: the 0 -cells imm^-1 ) and imm (1 ) ,
1 2
the 1-cell imm1 (Int D ) , and the 2-cell imm2 (S ^ ort^ . This is
obviously a cellular space; however, the closure of the 2 -cell touches
the 1 -cell, but does not contain it.

10. The most important subspaces of a cellular space X are

 85

its skeletons skeQX, ske^X,..., ske^X, ..., defined as

skerX = ^dime £ r 0# ^ X / 0 , then all the skeletons are nonempty
(since the presence of cells of a certain positive dimension implies
the presence of cells of lower dimension). For formal reasons, we add
the empty skeleton ske^X and the skeleton ske^X = X. The sequence
{skerX}Q<r<oo is clearly a filtration of X.
We remark that any map 0
X which is characteristic
for the cell e takes e 1 ske,.
j_n ^Q .X (in fact, we saw
dim e - 1 '
this already in 1). If X carries a rigging {cha^}, then the map
ab chae : S ^~m 0 1 skedj_m e _ ^ c a l l e d an attaching map for e,
and is denoted by att .
J e

11. Every cell of a cellular space is included in a finite

subspace.

PROOF. Use induction on the dimension of the cell. A 0-cell

is itself a subspace. If e is a cell of positive dimension, then
Cl e \ e may be covered by a finite number of lower dimensional cells,
and the union of e with a collection of finite subspaces which contain
these cells is a finite subspace containing e.

Compact Subsets

12. A compact subset of a cellular space intersects only a

finite number of cells.

PROOF. Every subset A of a cellular space contains a part

B which intersects at only one point each cell intersecting A. Since
B intersects any closed cell at a finite number of points, B and all
its subsets are closed. Therefore, B must be discrete. When A is
compact, B as a discrete, closed, and compact set is finite.

13. Every compact subset of a cellular space is contained in

a finite subspace.
For each cell intersected by the given subset, pick a finite
subspace containing this cell. The union of these subspaces is the
desired finite subspace.
14. Every compact subset of a locally finite cellular space
is contained in the interior of a finite subspace.

Indeed, such a subset has a neighborhood which intersects

only a finite number of cells (cf. 6 ). For each such cell, pick a
 86

finite subspace which contains it; the union of these subspace is again
a finite subspace and contains the above neighborhood.

Cellular Maps

15. A map of a cellular space X into a cellular space Y

is said to be cellular if it is continuous and maps the skeleton ske^X
into ske Y, for each r.
r
A cellular map obviously transforms 0-cells into 0-cells.
However, a cell of positive dimension is not necessarily transformed
into a single cell: for example, consider the identity map of the
i
segment D , decomposed into the 0-cells -1,1 and the 1-cell (-1,1),
onto the same segment, but decomposed now into the 0 -cells - 1 ,0 , 1 and
the 1 -cells (-1 ,0 ), (0 ,1 ); this cellular map takes the 1 -cell (-1 ,1 )
into the union of a 0 -cell and two 1 -cells.

16. A cellular map is a cellular equivalence if it is

invertible and its inverse is also cellular. An equivalent formulation:
a cellular equivalence is a homeomorphism which transforms the cellular
decomposition of the domain space exactly into the cellular decomposition
of the image space. If there is a cellular equivalence between two
cellular spaces, then they are said to be cellular equivalent. Two
rigged cellular spaces related by a cellular equivalence which transforms
one rigging into the other are said to be rigged-equivalent.
If X and Y are cellular spaces, a map f : X -* Y is a
cellular embedding if f(X) is a subspace of Y (defined as in 9) and
ab f : X f (X) is a cellular equivalence.
Warning: there are cellular homeomorphisms which are not
cellular equivalences. An example is the homeomorphism described in 15.

2. Glueing Cellular Spaces From Balls

1. If_ r ^ 0, then the skeleton ske X of the rigged

cellular space X is canonically homeomorphic to the space

(sker- 1 X> % r (De ' Dr" '

where M_ = cell^X and the map cp: | | (S = Sr_1) ■> ske .X is

----- r r e r —1 —
defined by cp o in = att (e E M )
--------- ■*- e e r
This canonical homeomorphism between the above spaces is
 87

the injective factor of the map (ske .X) I | ( I D -> ske X

r- 1 -i— LeGM e r
r
defined by the inclusion ske^^X skerX and the maps
ab cha : D ske X.
e e r

2. The decription in 1.3 of the weak topology in terms of

the total characteristic map shows that one can glue any cellular space
from balls, and even do it in a nice way. Theorem 1 above reduces this
glueing to a sequence of attaching processes: the r-th process
transforms ske^X into skerX (r = 0,1,...), and X is defined,
starting from the sequence {skerX}, as X = lim skerX (see 1.10).

The following formal procedure transforms this description of

cellular spaces into a useful inductive method of constructing such
spaces. First of all, note that if we are given a topological space A
with a rigged cellular decomposition into cells of dimensions <q,
and to A we attach a sum J Ly£M^Dy = of ^“dimensional balls by

some continuous map cp: J Ly£M^Sy = Sg~1) A, then we obtain aspace

endowed with an obvious rigged cellular decomposition into cells of
dimensions < q+1. This space satisfies condition (W) wheneverA
satisfies it, and Theorem 1.2.4.9 shows that itis normal if A is
normal. Moreover, it follows from Theorem 1.12 that this space
satisfies condition (C) provided A is cellular. Finally, we conclude
that if A is a normal rigged cellular space, then A U (II )
cp J— LyEM y
is a normal rigged cellular space too.
These observations form the basis of our inductive
construction. We start with q = 0, i.e., take A = 0, and atthe
r-th step we attach the (D space J_[
= Dr ) to the previously
r y y
constructed normal rigged cellular space xr_-| ' dim X r_i * ' by
r“ 1
a continuous map cpr : J_|_yeM <s u = s > X r-1 ' The r-th Step Y ields

a normal rigged cellular space Xr = (J_ |_y£M ' with

dim X r. The result of the whole process is a sequence
r
0 = x ,X ,X with natural cellular embeddings Xr ->■ xr +1 > and
limit space X =limX . According to Theorem 1.2.4.6 , X is normal
and is endowed with an obvious cellular decomposition satisfying
properties (C) and (W). Therefore, X is a normal rigged cellular
space, and clearly skerX = Xr .
We say that X is an inductively glued cellular space. The
discussion above demonstrates that every rigged cellular space is
rigged-equivalent to an inductively glued cellular space.

3 (COROLLARY). Every cellular space is normal.

 88

3. The Canonical Cellular Decompositions of Spheres,

Balls, and Projective Spaces

1. The spheres, balls, and projective spaces admit canonical

cellular decompositions making them into cellular spaces. These are all
rigged cellular decompositions and will be described in the present
subsection. It will be evident in each case that properties (C) and (W)
are satisfied.

2. The canonical cellular decomposition of the sphere Sn

with 0 £ n < °°, consists of the 0 -cell ort^ and the n-cell Sn ^ ort^.
As the characteristic map of the cell Sn ^ ort^ we take D S : Dn •* S .

3. The canonicalcellular decomposition of the ball Dn with

1 £ n < °°, is given by the 0-cell ort^ , the (n-l)-cell Sn 1 \ ort^ ,
and the n-cell Int Dn . For the characteristic maps of the cells
Sn 1 \ ort^ and Int Dn we take the composite map

Dn-1 _DS ^ sn-1 _in > Dn

and id Dn , respectively.

4. The canonical cellular decomposition of the real projective

n r r— 1
space 3RP (0 c n < co) consists of the r-cells e^ = IRP \ IRP ,
where 0 £ r £ n for n < °°, and 0 £ r < 00 for n = 00. The
composition

or KPr mpn

(where pr is the projection arising from the characterization ofIRPr

as a quotient space of Dr , given in 1.2.5.2) is taken as the
characteristic map of the cell e .
y~— 1 y*_1
It is clear that attg is simply pr: S IRP , and
that ske lRPn = IRPr , r £ n. r
r
5. The canonical cellular decomposition of the complex
projective space tCPn (0 £ n £ °°) consists of the 2r-cells
r r— 1
er = cp \ IP , where 0 £ r £ n for n < °°, and 0 £ r < °° for
n = “>. The composition

D2r (cpH

is taken as the characteristic map of the cell e2r- Obviously,

 89

sker<EPn = I P ^ 2-1, for r ¡? 2n.

The canonical rigged cellular decompositions of the projective

spaces IHP (0 < n £ °°)and (EaP (0 £ n £ 2) are defined similarly.
The decomposition of K P n isgiven by the 4r-cells e^ = lHPr \ M P r_\
where 0 £ r £ n for n < 00, and 0 £ r < 00 for n = 00. For (EaPn
the cells are e = (EaP \ (Eapr ^ with 0 £ r £ n and dime = 8 r.
r r

4. More Topological Properties of Cellular Spaces

1. Our task in this subsection is to examine what connections

exist between properties of cellular spaces such as compactness, local
compactness, separability, second countability, and metrizability, and
properties of their cellular decompositions such as finiteness,
countability, and local finiteness. Incidentally, we prove that
cellular spaces are CNRS 1 s . Moreover, conditions for the connectedness
of a cellular space, as well as its partition into components are
studied.

Compactness and Local Compactness

2. A cellular space is compact if and only if it is finite.

The necessity of this condition is a result of 1.12. Since

every finite cellular space can be covered by a finite number of closed
cells, the condition is also sufficient.

3. A cellular space is locally compact if and only if it is

locally finite.

By Theorem 1.12, every neighborhood with compact closure of an

arbitrarily given point intersects only a finite number of cells, and
hence the condition is necessary. It is also sufficient, because the
closure of a neighborhood which intersects only a finite number of cells
is contained in the union of the closures of these cells.

Embedding Theorems

4. Every cellular space can be embedded in a Euclidean space

of sufficiently high dimension.
We shall proceed by induction. Given a finite cellular space
 90

X of dimension n ^ 0, and an embedding j: ske ^X + IR , we shall

construct an embedding J: X > n 1 (since a cellular space of
dimension - 1 is empty, the first step of the induction is trivial).
Pick a rigging of X
and arrange the n-cells of X in a sequence
n q+n +1 ,
. Now define the maps $ , ...,(p : D -+ 3R by
I s i s

V =

(0 ,...,0 ,y1 ,...,yn- 1 ,yn +2 k , 1 ), if p = (y2 + -..+Y 2 ) 1 / 2 £ 1 /2 ,

ryi 1 Yn - 1 Yn
(2 p-1 )j°cha + (2 - 2 p) 0 , . ..,0 , ^ ------------ +2 k ,1
PJ
if p 5 1/2 ,

and then set

j (x) , if x € ske n - 1„X,

J(x) =
<J>k (y) / if x E Cl e. and x = cha (y).

q+n +1
This yields a continuous map J :X -+ IR
and since the plane y^ = 0 ,. y = 0 ,y „ = 1 contains no lines
Jq q+n +1
parallel to IR^, J is injective. Hence, J is an embedding (see
1.1.7.10).

5. Every finite cellular space is a CNRS.

The proof is a continuation of the previous one and also

requires an induction on n. Namely, we show thatIR^ U J(X) is a
neighborhood retract of IR^+n+1. This is enough: if we assume that
j(sken_^X) is a neighborhood retract of 3Rg , then J(X) is obviously
a neighborhood retract of IR^ U J(X).
Set

Ak = ^k (Int D ) and Bk = A^ fl { (x^ , .*

.*.,xq+n +1 x q+n + 1. f 1}

Since the sets ^ ,...,Ag are pairwise disjoint and closed in

TR^+n +1 ^ Uk = 1 ^ k ^ n they have pairwise disjoint neighborhoods

LL , ...,U in IR ^ uv
1 s kS=
= i*v
1 (Sn
~ 1),
'' and clearly
uiiu j-j
,*
^1 ' ..
.■.'
,Ug
are °Pen^s
q+n +1 n
in IR . Moreover, since A ^ ,...,Ag are all homeomorphic to IR ,
the identity maps A^ A^ , ...,Ag -+ A g extend to continuous maps
(see 1.1.5.12) Now let
v- ui " Ai v us

V7,k. =
= {x
{x e
e U
\,_ | dist (x,ip. (x) ) < Dist(x, IRq )} ,
 91

V = {x e m q +n +1 I Dist(x^Rq ) < 1/2},

and denote by ip the orthogonal projection V ->-:iRq . It is clear that

W = V U U ... U is a neighborhood of lRq U J(X) in IRq+n + ^ and
that the sets

{[(Uk = 1 C l V k ) n Fr V ] U (U^ = 1 Ak )} D w,
( 1)
[(Uk = 1 F r V k ) fl C I V fl W] v Uk= 1 <j>k (Sn_1) ,

are disjoint and closed in Y \ U^_.,<}>, (Sn 1), where

K — I .K

Y = (Uk=1Cl Vk ) n Cl V n W.

Let f: Y \ Uk = 1^k^Sn ^ ^ 1 be a Urysohn function for the

pair (1) . Now given y E Cl B, u and z E 1R^,
be the path in let
q Yz
3R U J(X) defined as the product of the rectilinear path between y
-1 -1
and (y)/dist (0 f(j)^(y) )) with the rectilinear path between

<J>k (y) /dist (0 ,^k "1 (y))) and z. Define

a: (Uk = 1C l B k ) x x I -»■ U J (X)

by a(y,z,t) = u (t) . According to 1.2.2.14, the map x: Y ]Rq U J(X),

yz
n- 1
a (x) ,ip (x) ,f (x) ) , if x E (Cl H Cl V fi W) \ (p^ (S ),
T (x) =

if x e uk = i 4>k (sn 1 ) ,

is continuous. By 1.1.4.3, the map W -* I R* U J(X) defined by the

formula

(x) , if X E ci vk n (W \ V) ,

x h» (x) , if X E w " uk = i V

T (x) , if X E Y,

is also continuous, and it is clearly a retraction.

Connectedness. Components

6. The components of a cellular space are open subspaces,

PROOF. Every closed cell is connected, as the image of a ball

 92

under a characteristic map. Therefore, a component of a cellular space

contains along with each point the closure of the cell in which the point
lies, i.e., it is a subspace. The complement of a component is the union
of the remaining components and so it is a subspace too. Consequently,
this complement is closed, and hence the component is open.

7. If_ r ^ 1, then the r-th skeleton ske^A of a component

A of the cellular space X is a component of ske^X * In particular,
a cellular space X is connected if and only if ske^X is connected.

PROOF. Obviously, ske^A = A fl ske^X for all r, and if B

is another component of X, then ske A and ske^B sit in different
components of ske^X. Therefore, all we have to show is that the
skeletons ske^A with r > 1 are connected or, equivalently, that
given a connected cellular space X, all its skeletons ske^X (r ^ 1)
are connected. But this is plain if one notes that the contruction of
an inductively glued cellular space described in 2 . 2 cannot result in a
connected space if one of the spaces X^ with r ^ 1 is not connected.

8. Every connected locally finite cellular space is countable.

PROOF. Given a connected, locally finite cellular space X,

fix an arbitrary point x^ £ X and consider the set A^ of all points
of X that can be joined to Xq by a path intersecting at most m
cells. Since any path intersects only a finite number of cells (see
CO
1.12), X = U .A , and it is clear that each A consists of whole
m= 1 m m
cells. Therefore, all we have to verify is that each set A with
2 m
m > 1 contains only a finite number of cells, and we do it by induction
on m.
It is readily seen that the closure of any cell contained in
Am+ .i intersects the closure of some cell from Am. On the other hand.
since X is locally finite, the closure of any cell from A^ (being a
compact subset of X) can intersect only a finite number of closed cells.
Thus the number of cells in Am+^ is finite provided that the number
of cells in A^ is so, and to complete the proof note that A^ is just
one cell.

Countability Axioms and Metrizability

9. A cellular space is separable if and only if it is

countable.

PROOF. If the cellular space X is countable, pick a

countable dense set in each cell and then take the union of all these
 93

sets to produce a countable dense set in X.

Now suppose that X is separable. Then every point of X
lies in a finite subspace (see 1.11). Pick such a subspace for each
point of a fixed countable dense set in X. The union of these sub­
spaces is a counatble subspace and actually coincides with X.

10 (LEMMA). If the cellular space X has a countable base

at a point X q € X, then this base contains a neighborhood of x^ which
intersects only afinite number of cells of X.

Suppose that this is not true. Write the elements of the

given base in a sequence and (using a trivial induction),
select a sequence of points x^x^... in X \ x Q, such that:
(i) x^ € LL ; (ii) if i ^ j, then x^ and x^ sit in distinct
cells. Since any closed cell contains only a finite number of the x/s,
the set of all x^'s is closed, and its complement is a neighborhood
of Xq. We reached a contradiction, because this complement contains
none of the neighborhoods U^,U2 ' - • • •

11. A cellular space is first countable if and only if it is

locally finite.

The necessity of this condition is a corollary of Lemma 10.

To prove its sufficiency, use Theorem 1.14 to deduce that every point
Xq of a locally finite cellular space has a neighborhood U contained
in a finite subspace. By Theorem 4, such a subspace, and hence U, are
first countable spaces, and it is obvious that a countable base of U
at xQ is also acountable base of the given space at xQ.

12. A cellular space is second countable if and only if it is

countable and locally finite.

The necessity of this condition is a corollary of 9 and 11.

To prove its sufficiency, for each closed cell we fix a neighborhood
contained in a finite subspace (see 1.14). As a result of Theorem 4,
these neighborhoods are second countable, and it is clear that the union
of their countable bases is a countable base of the given space.

13. A cellular space is metrizable if and only if it is

locally finite.
The necessity of this condition is a corollary of 11. The
metrizability of a connected locally finite cellular space follows from
theorems 8 , 12, 2.3, and 1.1.6 .9. Since every locally finite cellular
space is homeomorphic to the sum of its components, and these are
metrizable, the space is metrizable too (see 6 and 1 .2 .1 .2 ).
 94

5. Cellular Constructions

1. When applied to cellular spaces, the constructions

described in § 1.2 need to be appropriately modified. For certain
constructions the modification consists merely of observing that the
resulting space is endowed with a cellular decomposition and becomes
cellular; anobvious example is the sum. For other examples, such as
the product, the modification also affects the topology of the resulting
space.
Below we describe the main modifications of both types. We
emphasize that all these constructions, when applied to rigged cellular
spaces, produce again rigged cellular spaces.

Cellular Product

2. Let X^ and X2 be topological spaces with cellular

decompositions p^ and p2 . Then the product X^ * X 2 has a natural
cellular decomposition, namely p^ x with dim(e^ * e2 ) = dime^ +
+ dime 2 . As a characteristic map for the cell e^ * e 2 one may take
the composition of the canonical homeomorphism

dime + dime. dime1 dime

D >D x D

dim e dim e
with the product cha x cha : D x d X x x^ of
e e 1 2
dim e
arbitrary characteristic maps cha : D -> x and
dime 2 e1 1
cha : D X9 . When the decompositions p and p0 are rigged,
2 1 Z
p ^ x p2 takes on a canonical rigging.

If both p and p2 fulfil property (C), then (C) holds for

Pi x P2 too. However, there are situations where p x p2 does not
have property (W), even when x^ and X2 are cellular spaces; see
Exercise 6 .6 . The cellular space arising from the product x^ x x 2 of
the cellular spaces X1 and X2 through cellular weakening of its
topology is called the cellular product of X1 and X , and is denoted
by X, xc x2 .

We note that the cellular weakening does not alter the

topology of the compact parts of the space x X2 . Indeed, every
 95

compact subset of x X^ has compact images under the projections

X'j x ^ 2 X 1 and x x X 2 X^, and hence it can be covered by a
finite number of cells.

3. If X^ is locallyfinite, then X„ x x, = X. x x, for

I --------- *------------- 1 c 2 1 2 --
any cellular space X2 .
1 2
PROOF. Let cha and cha be the totalcharacteristic maps
corresponding to some riggings of the cellular decompositions p^ and
P2 of Xl and X 2 - It is clearthat the total characteristic map
corresponding to p^ x P 2 can be expressed as the composition

dim(e xe ) dime. dime,

I Id 1 2 ---- > J |_(D 1 X D 2) =
(2 )
dim e dime, , 1 v ^ 2
= (J |_D 1) x (J [D ) ~Cha • -
* Cha- > Xl x x2 ,

where the first map is the sum of the canonical homeomorphisms

dim(e^xe2) dime^ dim e 2
D D x d . Since p^ and P2 satisfy condition
1 2
(W), the maps cha and cha are factorial (see 1.3). Furthermore,
dim e 2
since | |D and X are locally compact (see 4.3), the map
1 2
cha x cha is factorial too (see 1 .2 .7.9), which in turn implies that
the composite map (2) is factorial. Therefore, the decomposition
x P2 has property (W ) .

4 (INFORMATION). If every point of each of the cellular spaces

X^ and X2 has a neighborhood which intersects only a countable family
of cells, then X. x X, = X. x x, ; see [6 ] for aproof.
1 c Z 1 z

Attaching

5. Consider two cellular spaces X^ and X2 , a subspace C

of X ^ , and a cellular map tp: C + X^- According to 1 .2 .4 .8 , X2 U(p X 1
is a well-defined topological space, while 2.3 and 1.2.4.9 imply that
X 2 Ucp X 1 norma-*-- Now decompose X 2 X^ into the sets imm^e^
and imm2 e 2 , where e^ and e 2 run over the cells in X^ \ C and X2 ,
respectively, and put dim(imm^e^) = dim e^ and dimiimir^^) = dim e 2 •
This is a cellular decomposition: as a characteristic map for imnue^
one may take the composition of an arbitrary characteristic map chae
i
with imitK. Clearly, the only cells that the closure of the cell
imm^e^ intersects are either imm^e^, where is a cell in X^
intersecting Cl e^ , or imir^^' where e^ is a cell in X2
 96

intersecting (p(Cle^ fl C) . Moreover, we see that Cl imm2 e 2

intersects only the cells imm2 f:2 , where 2 is
“ a~ cell in
“ X
“2
intersecting Cl e 2 Consequently, our decomposition has property (C).
To see that it has property (W) too, let F be a subset of X,2 Utp X,1
having closed intersections with all the cells in X U X, The
<P 1
equality
-1
imm^(F) n Cl e 2 = imm2 (F fl Cl(imm2 e2 ))
-1
shows that imm2 (F) is closed, and now the equality

ip 1 (imm0 1 (F) n <p(Cl e ) , if C,

imm. ^ (F) fl Cl e
- 1
imm^ (F fl C l (imm^e^ ) ) , if c X 1 \ C,
'1
- 1
proves that imm 1 (F) is closed too. We conclude that the space
is cellular, It is immediate that imm2 (X2> is a subspace of
X2 U<p X1
that imitu is a cellular embedding, and that the map imm„
X 2 U„ X 1 ' 1
is cellular.
If X2 * D then cp: C ^ X 2 is a cellular map for any
cellular pair (X^,C), and X2 X^ = X^/C. Thus, the previous
definition implies that the quotient space of a cellular space by a
subspace is cellular.

Limits

6. Suppose that Xq,X^,... are cellular spaces and

4>q : X q ->- X^ , (j) : X^ X2 ,... are cellular embeddings. By 1.2.4.4, the
limit lim(Xk ,(f)k ) is a well-defined topological space, which is also
normal (see 2.3 and 1.2.4.6 ). Now consider the decomposition of
lim (X^, (f)^) into the sets imm^e^, where e^ is a cell in
X^ \ ), k = 0 ,1 ,..., and put
dimiimm^e^ ) = dim e, . If we
k
take the composition of an arbitrary characteristic map cha with
ek
imm^ as a characteristic map for the cell imm^e^, we see that this
decomposition is cellular. Since it obviously satisfies conditions (C)
and (W) , lim(X^,c()k ) becomes a cellular space, and imm^ become
cellular embeddings.
Notice that this definition of the limit includes as a special
case the inductive process of glueing a cellular space from balls that
we discussed in subsection 2 .
 97

More Special Constructions

7. Since decomposing the segment I into the cells 0, 1 ,

and Inti makes I into a finite cellular space, the cylinder X x I
is cellular for any cellular space X; see 2 and 3. The bases of
X x I are cellular subspaces (in the sense of 1.9); hence when we pass
to the quotient space con X of X x I, and then to the quotient space
su X of con X , we find ourselves in the situation covered by the
construction 5. Therefore, the cone and the suspension over a cellular
space are also cellular spaces.
If f: X^ X^ is a cellular map, then the attaching processes
which transform X^ x I into Cyl f , and con X^ into Con f fall
again into the category described in 5. Therefore, the mapping cylinder
and the mapping cone of a cellular map are cellular spaces.

8 .The cellular join X^ X^ of two cellular spaces

and X2 is defined as

X- * X = (X.I |X_) U [(X. X X,) x I],

1 c 2 1 -— L 2 (p 1 C Z

where (p:[ (X1 xc X 2) x 0] U [ (X^ xc x2> x 1 ] ^ x -\ I Ix 2 is g iven by

(p(x1 ,x2 ,0) = cp(x1 ,x2 ,1) = in2 (x2); cf. 1.2.6.3. Since ip is
cellular, the space *c X 2 is cellular.

According to 3, when X^ is locally finite, X 1 *c X 2 is

topologically the same as X^ * X2 - In general, the cellular
decomposition of X^ *c X 2 is cellular for X 1 * X 2 too, and so the
cellular weakening of the topology of X^ * X 2 yields X^ *c X 2>
However, this process does not affect the topology of the compact sub-
sets of X^ * X2 ; cf. 2 .

The Case of Pointed Spaces

9. Suppose that X is a cellular space and xQ is a 0-cell

that we take as a base point. The cone c o n ( X , x Q) and the suspension
su(X,Xq) are quotients of con X and su X by subspaces, and as such
they are cellular spaces. Similarly, the bouquet of a family of
cellular spaces with 0 -cells as base points is the quotient of the sum
of this family by a subspace, and hence is a cellular space.
 98

Finally, we define the cellular tensor product and the

cellular join of the cellular spaces and X 2 withthe 0 -cells x^
and x^taken as base points,as the quotient spaces

( X 1 Xc X 2 ) / f (X1 x x2) u (x1 X x2)] and (X1 * c X2 ) / / ( x 1 * X2 ) f

respectively. These are cellular spaces, denoted by (X^,x^) ®c (X2 ,x2>

and (XwX.) * (X„ ,x ) . If X. is locally finite, then they are
1 1 C 2. 2. 1
identical with (X1 ,x1) ® <X 2 'X 2) and ( X 1 'X 1) * (X2'X 2) aS toPolo9 ical
spaces. In the general case, the cellular decompositions of
(X^,x^) ®c (X2 ,x2) and (X^,x ^) * c (X2 ,x2) are cellular for
(X ^ ,x^) ® (X^ , x ^) and (X ^ ,x ^ ) * (X2'X 2) tG>° * Thus '
(X^,x^) ®c (X2 'x 2> and (X ^ ,x ^)*c (X2 ,x2) arise from the cellular
weakening of the topologies of (X^,x^) 0 (X2 ,x2) and
(X^,x^) * (X2 ,x2), respectively, and it is clear that this process
does not affect the topology of the compact subsets of
(X1 ,x^) ® (X2 /X2) and (X^,x ^) * (X2 ,x2>.

6. Exercises

1. Show that given an arbitrary cellular space X and an

arbitrary point x £ X,there exists a cellular space Y together with
a cellular homeomorphism f: X + Y such that f(x) E ske^Y.

2. Show that the sphere S°° and the ball D°° are homeo-
morphic to cellular spaces.

3. Show that every connected, locally finite cellular space

CO
can be topologically embedded in IR .

4. Show that every connected, finite dimensional, locally

finite cellular space can be embedded in 3R^, for large enough q.

5. Show that every finite cellular space admits a cellular

embedding in a cellular space homeomorphic to , for large enough q.

INFORMATION. Every finite celular space of dimension n can

be embedded in a cellular space homeomorphic to D 2n+\

6. Consider the bouquet B = Vt ^ (I =i,0) as a cellular

space (see 5.9) and show that the map id: B x^b B x b is not a
homeomorphism.
 99

§2. SIMPLICIAL SPACES

1. Euclidean Simplices

1 . Let A be a subset of ]Rn consisting of r +1 points

(r ^ 0) which are not contained in any (r-1)-dimensional plane. The
convex hull of A (i.e., the smallest convex set containing A) is
called the Euclidean simplex spanned by A, and is denoted by EsiA .
The points of A are the vertices of the simplex Esi A , and the
number r is its dimension. Esi A is also called a Euclidean r-simplex.
Obviously, a point of Esi A is a vertex if and only if
Esi A contains no nondegenerate segment whose midpoint falls on the
given point. Therefore, the set A is uniquely determined by EsiA .
Every simplex spanned by a subset of A is called a face of
the simplex EsiA . It is clear that Esi A^ DEsi A^ = Esi(A^ D A^) ,
for any A^ ,A^ cz A.
Two faces spanned by complementary subsets A^ and A^ of
A are said to be opposite. In this case, the formula

pr (x1 ,x2 ,t) (1-t)x^ + tx 2 (x^ € Esi A^ , x 2 £ Esi A^ , t € I)

defines a homeomorphism of the join Esi A^ * Esi A^ onto Esi A . Thus,

every Euclidean simplex is canonically homeomorphic to the join of any
of the pairs of its opposite faces.
Since Dr is canonically homeomorphic to any join
with p +q = r - 1 (see 1 .2 .6 .9), a trivial induction proves that both
the spaces Esi A and Drare homeomorphic to a join of r+1 points.
We conclude that every Euclidean r-simplex is homeomorphic to D .
It is clear that the boundary of the simplex Esi A in the
r-plane that it determines is precisely the union of its (r-1 )-faces.
Usually, this boundary and its complement in EsiA are simply referred
to as the boundary and the interior of the simplex EsiA .

2. We may equivalently describe the simplex EsiA as the

set of all sums £a£A t&a , where t& ^ 0 and l a £ A t& = 1. Since
there is no (r-1)-plane containing A, the numbers ta are determined
uniquely for any point x = ^aa? ^a ca-*-^ec^ ^he a-th bary -
 100

centric
— . coordinates
—.— -- -- —. . of x and is denoted by bar a (x).
Obviously, a face Esi B of the simplex Esi A is defined
in the barycentric coordinates of EsiA by the equations bar^ (x) = 0
for a E A \ B. Moreover, if x E Esi B , then the coordinates bar a (x)
computed in EsiA and EsiB coincide for all a E B.
The point of EsiA having all barycentric coordinates equal,
i.e., equal to 1/(r+1), is the center of the simplex EsiA .

3. A map Esi A -+ Esi B is called simplicial if it is affine

and takes A into B. It is clear that such a map takes each face of
Esi A simplicially into a face of Esi B , and takes the interior of
Esi A onto the interior of the simplex which is its image.
Obviously, every map A -+ B extends uniquely to a simplicial
map Esi A Esi B . If the given map A B is injective (invertible),
then its simplicial extension Esi A Esi B is an embedding
(respectively, a homeomorphism).

4. EsiA issaid to be an ordered simplex if the set A is

ordered. Since the subsets of an ordered set inherit a natural order,
all the faces of an ordered simplex are ordered simplices.
If EsiA and EsiB are ordered r-simplices, then the
orders of A and B define an invertible map A + B, and hence a
simplicial homeomorphism EsiA -* EsiB . Consequently, all ordered
Euclidean simplices of the same dimension are canonically simplicial
homeomorphic.
r +1
5. The simplex spanned by the points ort ,...fOrt^ of 3R
is called the unit r-simplex and is denoted by Tr . This simplex is
notable due to the fact that its barycentric coordinates are the usual
r +1 T"
coordinates in IR . The given order of its vertices transforms T
into an ordered simplex, and thus every ordered Euclidean r-simplex is
canonically simplicial homeomorphic to T .
Mote that given an ordered simplex Esi A , the homeomorphism
r
EsiA -+ D discussed in 1 is now canonical. The canonical homeo-
r r
morphism T -* D and its inverse are denoted by TD and DT,
respectively. That TD maps the boundary (the interior) of Tr onto
r— 1 r
S (respectively, onto IntD ) is plain.

Topological Simplices

6. A topological space X is an ordered topological simpl

of dimension r (or an ordered topological r-simplex) if there exists
 101

a homeomorphism Tr -> X; this is called a characteristic homeomorphism

of the simplex X, while X is sometimes referred to as the support
of the simplex. For example, all ordered Euclidean r-simplices and the
jc
ball D are ordered topological r-simplices; see 5.
The standard way to detroy an order is to introduce
simultaneously all possible orders. Accordingly, we say that the
topological space X is a topological simplex of dimension r (or a
topological r-simplex) if there are given (r+1 )! homeomorphisms
r r
T -> X , which can be transformed into each other by simplicial homeo-
r r
morphisms T -+ T . The terms characteristic homeomorphism and support
are employed in this situation too; however, now we have at our disposal
(r+1 )! equally rightful characteristic homeomorphisms.
If X is a topological r-simplex (an ordered topological
r-simplex), and Y is a topological space, then every homeomorphism
X Y transforms Y into a topological r-simplex (respectively, into
an ordered topological r-simplex). Consequently, every homeomorphic
image of a Euclidean r-simplex (ordered Euclidean r-simplex) is a
topological r-simplex (respectively, an ordered topological r-simplex).
The vertices, faces, boundary, interior, barycentric
coordinates, center, and simplicial maps are defined in an obvious
fashion for topological simplices. The faces of a topological simplex
(ordered topological simplex) are topological simplices (respectively,
ordered topological simplices). As with a Euclidean simplex, a
topological simplex becomes an ordered one as soon as we fix an order
of its vertices.

2. Simplicial Spaces and Simplicial Maps

1 . A triangulation of a set X is a cover A of X by

topological simplices such that:
(i) every face of an arbitrary simplex in A is again a
simplex in A;
(ii) if a simplex in A is contained in another simplex of
A, then the first is a face of the second;
(iii) the intersection of the supports of two overlaping
simplices of A is again the support of a simplex in A.

A set X endowed with a triangulation is known as a

simplicial space; the simplices of the triangulation are called
simplices of the space, and the O-simplices are its vertices. The
smallest simplex in the triangulation which contains a given point
 102

x E X is denoted by si x .
According to 1.2.4.3, a triangulation transforms the given set
into a topological space, and 1 .2 .4.1 shows that the supports of the
simplices of the triangulation yield a fundamental cover of this space.
Since the intersection of two simplices in the triangulation is closed
in each of them, the simplices in the triangulation keep the same
topology when considered as subspaces of this topological space (see
1 .2.4.2) .
Let a be a vertex of the simplicial space X. Then the
a-th barycentric coordinate bar (x) is well defined for any point x
a
belonging to any simplex which has a as one of its vertices (see 1 . 2
and 1.6), and we obtain a continuous function bar : X -> IR if we set
a
bar (x) = 0 for those points x G X contained in simplices which do
a
not have a as a vertex.
bar_ is called the a-th barycentric function.
a
Given two arbitrary distinct points x,y E X, there obviously is a
vertex a such that bar_ (x) f bar_ (y). Consequently, every simplicial
a a
space is Hausdorff.

When a set X endowed with a triangulation already has a

topology, it is useful to find conditions ensuring that the topology
defined by the triangulation is identical with the initial one. We
have an immediate necessary and sufficient condition: the topology of
each simplex in the triangulation coincides with the topology induced
by the initial topology of X, and the cover of X by the supports of
these simplices is fundamental in the initial topology. If this
condition is satisfied, then the given triangulation is said to be a
triangulation of the initial topological space X. Example: the cover
of a topological simplex by all its faces is a triangulation of this
simplex.
A simplicial space is ordered if its simplices are ordered in
such a way that the orders of the faces of any simplex agree with the
order of the simplex itself. In particular, this holds whenever the
order of the simplices is induced by some order on the set of all
vertices of the given space, which incidentally shows that a simplicial
space can be always ordered.

2. We shall presently describe a fundamental class of

simplicial spaces. Given an arbitrary nonempty set A, we let Si A
denote the set of all nonnegative, finitely supported functions
$: A IR such that ^ = ^ ^ B c: a , then we identify Si B
with the subset of Si A consisting of all functions <J> £ Si A
such that <J>(x) = 0 for x £ A \ B. If A is finite and has r +1
elements, then Si A is obviously a topological simplex: indeed, Si A
 103

is a subset of the (r+1)-dimensional Euclidean space of all functions

A + JR. Moreover, corresponding to the (r+1)i orders on A there are
(r+ 1) i homeomorphisms Tr Si A , each transforming the point
(x^ ,.. ./xr + <|) into a function taking the values x^ , ...rx r + ^ ? one maY
transform one homeomorphism into another by composition with a simplicial
r r
homeomorphism T T . In the general case, Si A is covered by the
topological simplices Si B corresponding to all finite subsets B of
A, and it is clear that this yields a triangulation of Si A . Si A is
therefore a simplicial space, and we call it the simplex spanned by A.
Ordering Si A is equivalent to ordering the set A.

3. The interiors of the simplices of a simplicial space X

constitute a decomposition of the set X. If we definethe dimensionof
the interior e of the simplex s by dim e = dim s , and take as a
characteristic map for e the composition

Ddims J D T _ Tdims s _in_+ x ^ (1)

where $ is any characteristic homeomorphism for the simplex s, then

the above decomposition becomes cellular. Since conditions (C) and (W)
are clearly satisfied in this situation, and we already know that every
simplicial space is Hausdorff, we see that this cellular decomposition
transforms X into a cellular space. Thus, every simplicial space,
decomposed into the interiors of its simplices, is a cellular space.

Since an r-simplex has (r+1)i characteristic homeomorphisms,

formula (1 ) distinguishes (r+1 )i privileged maps in the family of all
maps that are characteristic for an r-cell in a simplicial space; we
call them simplicial characteristic maps and note that they are
topological embeddings. Fixing a simplicial characteristic map is
equivalent to fixing an order of the simplex s; hence every ordered
simplicial space is canonically rigged.
It is clear that the skeleton skerX of a simplicial space
X is simply the union of all its simplices of dimension £ r, and that
dim(skerX) = r for r £ dim X . In particular, skeQX is the set of
vertices of X. Of course, Xis finite if and only if skeQX is
finite as a set. A simplicial space is locally finite if and only if
each of its vertices is contained in only a finite number of simplices
(see 1.1.7) .
 104

Subspaces

4. A subspace of a simplicial space is any subset which is a

union of whole simplices. Every subspace has a natural triangulation,
and hence is a simplicial space. The subspaces of an ordered simplicial
space are also ordered simplicial spaces.
Obviously, a subset of a simplicial space X is a subspace of
X if and only if it is a cellular subspace of the cellular space X.
A subspace of a simplicial space X is complete if its
intersection with the support of any simplex of X is either the sup­
port of some simplex of X or empty. An equivalent formulation: a
subspace is complete if it contains, along with the vertices of a simplex,
the simplex itself.
Of course, the simplices of a simplicial space are complete
subspaces. A subspace of Si A is complete if and only if it is of the
form Si B , where B c A.

Simplicial Maps

5. A map X Y, where X and Y are simplicial spaces,

is called simplicial if it transforms every simplex of X simplicially
into a simplex of Y. It is clear that such a map is also cellular and
maps X onto a subspace of Y.
The following facts are also immediate.
An invertible simplicial map is a homeomorphism, and its
inverse is also simplicial. Every injective simplicial map is a
topological embedding. A simplicial map f: X ^ Y is uniquely defined
by the map ab f : ske^X ske^Y from the set of vertices of X into
the set of vertices of Y. A map ske^X ->• ske^Y extends to a
simplicial map X ->■ Y if and only if it carries the vertices of each
simplex of X into the vertices of a simplex of Y. A simplicial map
f : X -* Y is injective (invertible) if and only if ab f : ske^X ske^Y
is injective (respectively, invertible).
Two simplicial spaces which can be transformed one into another
by a simplicial homeomorphism are said to be simplicial homeomorphic.

6. A simplicial map f: X Y, where X and Y are ordered

simplicial spaces, is monotone if f(a) < f(b) for any pair of vertices,
a and b, of X which belong to the same simplex and satisfy a < b.
Every simplicial map between simplicial spaces can be made
monotone by suitably ordering the spaces. Moreover, if X and Y are
 105

simplicial spaces and Y is ordered, then one can transform a given

simplicial map f: X -+ Y into a monotone one my suitably ordering X;
indeed, it suffices to order arbitrarily the preimage of each vertex of
Y, and then order the simplices of X by the rule: a < b whenever
f(a) < f(b) or f(a) = f(b) and a < b in f ( f ( a ) ).

3. Simplicial Schemes

1. A simplicial scheme (or schema) is a pair (M,S), where

M is a set and S is a cover of M by finite subsets,such that S
contains, along with each set ACS, all the parts of A.
A map of the simplicial scheme (M,S) into the simplicial
scheme (M',S ') is a pair of maps, <p: M -+ M' and 0: S S', such
that $ (A) = (J)(A) for all AG S. The last condition shows that the
map (<¡>,0) of (M,S ) into (M',S') is actually uniquely determined
by $ . Obviously, (p: M M' defines a map of the scheme (M,S) into
the scheme ( M ^ S 1) if and only if cf>(A) £ S' for all AGS. If
d> and $ are invertible, i.e., <j> is invertible and 0(S) = S',
then the map ($,$) is called an isomorphism. Two simplicial schemes
which can be related by an isomorphism are isomorphic.
A simplicial scheme (M,S) is a subscheme of the simplicial
scheme (M ',S 1 ) if M c M' and S c=s'. The subscheme (M,S) is
complete if A G S ’ and A c= M imply AGS.

2. The simplicial scheme given by the skeleton ske^X of a

simplicial space X and the cover of ske^X by the 0-skeletons of the
simplices of X is termed the scheme of the space X and is denoted by
sch X . For example, the scheme of Si A (see 2.2) consists of the set
A and of the cover of A by all its finite subsets.
The map ab f : skeQX + skeQX' induced by a simplicial map
f: X X* takes the 0-skeleton of each simplex of X into the
0-skeleton of a simplex of X*. Hence it defines a map of schX into
sch X * , called the scheme of the map f and denoted by schf . The
discussion in 2 . 5 implies that a simplicial map is uniquely determined
by its scheme, that every map of sch X into sch X ' is the scheme of
some simplicial map X -* X', for any simplicial spaces X and X*,
and that a simplicial map is invertible if and only if its scheme is an
isomorphism. In particular, two simplicial spaces X and X’ are
simplicial homeomorphic if and only if their schemes schX and sch X '
are isomorphic.
3. If X is a subspace of the simplicial space X', then
 106

sch X is a subscheme of s c h X 1 ,and sch X is complete if and only

if X is complete. Moreover, it is clear that every subscheme of
sch X 1 is the scheme of a subspace of X*.
In particular, let (M,S) bean arbitrary simplicial scheme,
and consider the simplex SiM . Obviously, (M,S) is a subscheme of
schSiM, and so (M,S) is the scheme of a subspace of Si M . Thus,
every simplicial scheme is the scheme of a simplicial space.Moreover,
given an arbitrary simplicial space X, we may take (M,S) to be the
scheme of X and conclude that every simplicial space X can be
simplicially embedded in Si ske^X .

4. A simplicial scheme (M,S) is ordered if the sets of S

are ordered and the order of each set A f. S is compatible with the
orders of the subsets of A. A map (<j>,$) between ordered simplicial
schemes (M,S) and (M',S!) is monotone if <Ma) < <P(b) whenever
a < b. Therefore, ordering the scheme of a simplicial space is
equivalent to ordering the space itself, and the scheme of a simplicial
map between two ordered simplicial spaces is monotone if and only if the
map itself is monotone.

4. Polyhedra

1. A polyhedron is a subset of Euclidean space which admits

a finite triangulation by Euclidean simplices. Of course, the simplest
polyhedra are the Euclidean simplices.
A subspace of a polyhedron is obviously a polyhedron.
Now any simplicial space can be simplicially embedded in the
simplex spanned by its 0-skeleton (see 3.3), and when the initial space
is finite and has q vertices, this skeleton is simplicial homeomorphic
to T^ ^. Therefore, every finite simplicial space admits a simplicial
embedding in a Euclidean simplex with the same number of vertices.
Hence every finite simplicial space is simplicial homeomorphic to a
polyhedron.

2. Every finite n-dimensional simplicial space is simplicial

homeomorphic to a polyhedron contained in IR .

PROOF. Since every finite n-dimensional simplicial space can

be simplicially embedded in ske^T^ for q large enough (see 1 ) , it
suffices to construct, for arbitrarily given q and n, a linear
n _TT^2n+ 1 . . . . . . . Q
mapping f : 3R 3R which is miective on ske T .
n
If q < 2n, we may take f to be the inclusion m q+1 -► IR2n+1.
 107

If q > 2n, define f by

f «Vs--i> ■ { ij=! i S ' i " 1 •

All that remains is to verify that if x = (x„,...,x ) and
cf 1 q +1
X' = (x^,...,x’+1) belong toskenT4 and f(x) =f (x'), then x = x ’.
Since each of the points x and x' lies in an n-dimensional face of
q
T , at most n+1 ofthe numbersx. ,... ,x ., and n+1 ofthe numbers
1 q+ 1
xi x'+ r are different from zero. Consequently, no more that 2n+2
numbers x^-x^,...,x^+^-x^+^ are different from zero, i.e., there are

positive integers j-,'••• »j2n +2 such that ^ < ... < j2n +2 * 2 q +1 and
x-j = X' for j f j, j2nt2. Since

¡? :i <- n — i|:] i 1 «}

for i = 1 ,..., 2 n + 1 , we have

j r (xi ■ x i > = °' 1 = 0 ,...,2 n+ 1 .

Jr Jr

The determinant of the matrix { r- 2n+2 does not vanish, and

L 1 —U f 10 — I
so x. = x *
. for r = 1 f...f2 n +2 f and finally x = x*.
Dr
3 (INFORMATION). For any n there are n-dimensional
polyhedra which cannot be topologically embedded in IR2n. An example
is ske T2n+2; see [10] for a proof.
n

5. Simplicial Constructions

1. Many of the topological and cellular constructions

described in§ 1.2 and Subsection 1.5 can be replaced by parallel
constructions which produce simplicial spaces out of simplicial ones.
The simplest examples are the J and V operations: a sum of
simplicial spaces and a bouquet of pointed simplicial spaces with
vertices as base points are obviously simplicial spaces. There are also
more elaborate constructions, the more important ones being discussed
below. The main one is the barycentric subdivision construction, which
refines triangulations and has no analogs in § 1.2 and Subsection 1.5.

2(LEMMA). Let T be a fundamental cover of the topological

space X by triangulated subspaces. Suppose that for any A,B € r
the intersection A 0 B is a complete subspace of both A and B
(considered as simplicial spaces) and inherits from A and B the
 108

same triangulation. Then there exists a unique triangulation of X

relative to which the elements of r become simplicial subspaces.

This triangulation of X is simply the union of the

triangulations of the elements of r. One may check directly that this
union satisfies conditions (i), (ii), and (iii) in 2 . 1 [the completeness
of the intersections A n B is necessary for (iii)]. Uniqueness is
also evident.

Barycentric Subdivision

3. The construction below produces a new simplicial space,

baX , from any simplicial space X, such that b aX is identical to
X as a topological space, but has a finer triangulation, called the
barycentric subdivision of the initial triangulation.
Consider first a Euclidean simplex X. For an arbitrary
numeration a,*,...,a of the vertices of X, form the set
O r

{x E X I bar (x) £ bar (x) <: ... £ bar (x)}. (2)

a_ a„ a
0 1 r

It is readily seen that (2) is the Euclidean simplex whose

vertices are the centers of the simplices Esi A q ,. ..,Esi A^ , where
Ai = Furthermore, the simplices of the form (2) correspon­
ding to all possible numerations of the vertices of X and their faces
clearly yield a triangulation of X. This is precisely the barycentric
subdivision of the standard triangulation of the simplex X, and it
transforms X into ba X .An obvious property of this construction is
that the inclusion ba X baX1 is a simplicial embedding whenever
X is a face of the simplex X 1.
Now if X is a topological simplex, we define the barycentric
subdivision of its standard triangulation as the image of the barycentric
subdivision of the standard triangulation of the unit simplex Tr ,
r = dim X , under a simplicial homeomorphism Tr -> X . This is clearly
a correct definition, i.e., the triangulation of X thus obtained does
not depend on the choice of the simplicial homeomorphism Tr X among
the (r+1 )! available ones.
Finally, let X be an arbitrary simplicial space, and
consider the cover of X by its simplices, each subdivided as above.
It is easy to verify that this cover satisfies the conditions of Lemma
2, and hence we obtain a new triangulation of X, which is precisely
the barycentric subdivision of the initial triangulation of X.
 109

We note that the barycentric subdivision transforms a finite

(locally finite) simplicial space into a finite (respectively, locally
finite) one. Moreover, if X is a polyhedron, then so is baX .

4. The set of vertices of the space ba X equals exactly

the set of centers of the simplices of X. The centers of the simplices
S1'**‘'Sm °f X are the vertices of a simplex of ba X if and only if
S1'*‘*'Sm Can be reindexed to form an increasing sequence. This
observation enables us to give a concise description of the barycentric
subdivision in the language of schemes: if sch X = (M,S), then
sch ba X = (S,ba S) , where ba S is precisely the collection of those
finite parts of S that can be ordered by inclusion. At the same time,
we obtain a canonical order of ba X : if a,a’ € skeQba X , then a < a1
whenever the simplex (of X) with center a is contained in the simplex
with center a ’.
In particular, the above description of sch ba X shows that
baX is a complete subspace of ba X 1 whenever X is a subspace of
X*. Indeed, sch ba X is clearly a complete subscheme of sch ba X ’.
In general, given a simplicial map f: X X', the map
2
f: ba X ba X' is not simplicial (the simplest example: take X = T ,
X* = t \ f (ort^ ) = ort^ , f (ort^) = f(ort^) = ort^) . However, the
map sch f : sch X sch X' naturally induces a map sch ba X sch ba X ',
and hence a simplicial map ba X ba X * . Thelatter is denoted by
ba f and is clearly always monotone.

5. rf X is a polyhedron, then the maximal diameter of the

simplices of the polyhedron ba X does not exceed the maximal diameter
of the simplices of X times n/(n + 1) , where n = dim X .

It is enough to show that if X is the Euclidean simplex

with vertices a«,...,a , then the diameter of the simplex (2 ) is no
O r
bigger than [r/ (r+1)]diamX . Consider the part X ’ of X defined by
the inequality bar
(x) ^ r/(r+1). X' is the Euclidean simplex
ar
obtained by contracting X towards the vertex a^ by a factor of
r/(r+1). Consequently, diamX' s? [r/(r + 1 )]diam X , and we finally note
that X' contains the simplex (2).

6 (COROLLARY). For any polyhedron X and any e > 0 there

is a positive integer m such that every simplex of the polyhedron
bamX has diameter < e.
 110

Simplicial Products

7. If X^ and are simplicial spaces with dim X^ > 0

and dim X^ >0 , then it is readily seen that their cellular product
X^ X2 does not admit a triangulation such that the interiors of its
simplices are products of interiors of simplices of X^ and X^.
However, we shall presently show that X^ X2 admits triangulations,
and we shall construct a canonical triangulation when X^ and X^ are
ordered. This construction produces a simplicial space out of X^ xc X2 ,
called the simplicial product of X^ and X^, and denoted by X^ x^ x ^ -
To begin with, let X^ be the Euclidean simplex in IRm with
vertices an ,...,a , and let X, be the Euclidean simplex in JRn with
U q Z
vertices bg,...,br . Set for x^ G X^ and x^ G X^:

“i^l* = ^ =0 bara <X 1 >' 6 j (x2 ) = ^ 1 =0 barb 1 (X2 )'

k J 1

and arrange the numbers a 0 (x1 ),...,a^_^(x^),6 0 (x2 ),...,6 r_^(x2) in a

nondecreasing sequence Y-j (x^ ,x2) , .. .,y^ +r (x^ ,x2) . Further, let

denote the collection of subsets with q elements of {1 ,...,q+r},

and let s(y), where y G M , denote the set of all points

qr
(x^ ,x2> G X^ x x 2 such that each of the numbers ^p^x i'x 2 ^' P £ U/
is equal to one of the numbers yQ (x^) ,...,Yq _ 1 (x1) . One may check
directly that there is no (q+r- 1 )-dimensional plane containing the
q+r +1 points

(aQ /bQ) /• ••t (an ,b • — -j) 7

J1
(a1 ,b. (a.,b. _n)?
1 Ji 1 1 J9 z
:3>

(aq-i ,bj q- ( q- i ) ] ' ' ' ' ' (V i ' bj q- q);

(aq'b j - q * (aq'br) '

where G y, j-j < •••< j . Also,one may verify that

^ jk + r (k+1)
k =0 l =i -k[Yk+l + 1 (X1 ,X2 ) " Yk + 1 (x1 ,x2 )] (ak'bl ) = (xi'x 2 ]
Jk

for (x1 fx2) € s(p), jQ = 0 ,j^ +1 = q +r + 1 , Yq (x^,x ^ ) = and

Tq+r +i (x i/x2^ = ‘ Moreover,
and Yp+ 1 <X 1 /^2) - Yp (x ^ fX 2 ) ^ Consequently, the set s(y) is
contained in the Euclidean simplex spanned by the points (3), and since
s (p) is obviously convex and contains all the points (3 ), it equals
this simplex. Now it is clear that the sets s(y) cover x x2 '
and that s(y^) n s (y2^ -*-s I3ust the Euclidean simplex spanned by the
vertices common tothe simplices s(y^) and s(y2), f°r any
y^,y 2 £ Mqr* As a result, the simplices s(y) and their faces
constitute a triangulation of the product X^ x x^, and this
triangulation transforms X^ x x 2 into X x^ x^. Moreover, it is
readily seen that if X^ and X^ are faces of the ordered Euclidean
simplices X' and
X*, then the inclusion X. x xn + X* xx* is a
1 2' 1 s 2 1 s 2
simplicial embedding.
Next let X and X2 be ordered topological simplices. To
define the simplicial product X x
x^, use the previous prescription
^ Q r
to triangulate the product of the unit simplices T ^1 and T , with
q = dimX^ and r = dimX^ , and then employ the product
T^ x Tr X^ x x^ of canonical simplicial homeomorphisms T^ + X^ and
Tr -> X^ to carry this triangulation to X^ xx2 * With the resulting
triangulation, X^ x x^ becomes X^ x^ x 2 .

Finally, let X^ and X^ be arbitrary ordered simplicial

spaces.It is readily verified that the cover of X^ x^ x^ by the
product s^ x of simplices s^ of X^ and s^ of X2 , where
s^ x s2 is triangulated as above, satisfies the conditions of Lemma 2.
This lemma yields a triangulation which transforms X^ x^ x 2 into the
simplicial product X^ X2*
We remark that each cell e of X^ x^ x 2 can be represented
as the union of a finite number of cells ofX^ x^ x2 , having
dimensions £ dime . In particular, the mapid: X^ *c X 2 X^ x^ x 2
is cellular.
8. A straightforward corollary of the simplicial product
construction is that the product f^ x f2 : X 1 xs X 2 ^ X 1 *s X 2 of
two monotone simplicial maps, f^ : X^ X^ and f2 : X 2 -* X2 , is also
simplicial. It is also plain that if and X2 are subspaces of the
ordered simplicial spaces X^j and X 2 , then X^ x^ x 2 is a subspace
 112

Let us conclude with a description of the simplicial product

in terms of schemes. Suppose sch X^ = (M^,S^) and sch = (M^/S^)-
Then sch(X^ X2) = (M^ x M 2 ,S), where S is the collection of sets
A <= X^ x x 2 such that:(i) pr^ (A) E S ^ , pr 2 (A) E S2 ; (ii) if
(a1 ,a2) E A, (ajj,a£) E A, and a 1 < a ’, then a 2 < a £ .

Limits

9. Let Xq,X^ ,X2 , ... be simplicial spaces, together with

simplicial embeddings (J)Q : XQ X , $ ^ : X^ -* X 2 ,. .. . Consider the
cover of limiX^.,^) by the sets imm^ls^), where s^ is a simplex
of X (k = 0,1,...). If we take for a characteristic homeomorphism
K
of immv (s,) the composition of a characteristic homeomorphism
dim s^.
T + sk with the homeomorphism ab imm^ : s^ -> imm^is^) , then
imm^is^) becomes a topological simplex. It is clear that in this way
the cover (imm^is^)} becomes a triangulation of lim (X^,<J>^) ; hence
limlX^.,^) is a simplicial space. As a cellular space, this space
coincides with the limit defined in 1.5.6. Moreover, imiri^ are
simplicial embeddings.
If X^. are ordered spaces and $ are monotone, then the
space limiX^.,^) is ordered, and the embeddings imm^ are monotone.

Joins, Cones, and Suspensions

10. Let X and X2 be topological simplices. Then the

join X^ * x 2 naturally becomes a topological simplex if we define the
characteristic homeomorphisms as
dim X + dim X +1
T = Tl * t 2 i x 1 * x2 ,

dim X + dim X 9 + 1
where and T2 are opposite faces of the simplex T t
and <J> and c|>2 are simplicial homeomorphisms (the equality
dim X^ + dim X 2 + 1
T = T * T2 stands for the simplicial homeomorphism
established in 1.1). Now one may canonically triangulate the cellular
join X 1 *c X 2 of two arbitrary simplicial spaces X1 and X2 : its
simplices are the images of the simplices of X1 and X2 under the
inclusions X 1 ->■ X 1 *c X 2 and X 2 -> X *c X , and also the images of
 113

the simplices * s- under the inclusions in * in: s. * s 0

s X. * X0,
I z 1 z 1 2
where s^ is a simplex of X^ and s 2 a simplex of X2 . The resulting
simplicial space is called the simplicial join of X^ and X2, and is
denoted by x^ *g X2> As a cellular space, *g X 2 is identical to
X 1 *c X 2 -
If sch X 1 = ( M ^ S ^ and sch X 2 = (M2 ,S2>, then clearly
sch (X^ *s X 2> = (M^ | |M2 ,S ), where S is the collection of nonempty
subsets A of M 1J |M2 such that in^ 1 (A) € S 1 or in" 1 (A) = 0, while
— 1 -1
in2 (A) eS2 or in2 (A) = 0 . Here in1 : M 1 M J_¡_M2 and
in^: +M^J [m ^are the canonical mappings.
In particular, since

con X = X * and su X = X *

for any topological space X (see 1.2.6 .8 ), we see that the simplicial
join construction transforms the cone and the suspension over an
arbitrary simplicial space into simplicial spaces.

Simplicial Mapping Cylinders

11. To a given monotone simplicial map f: X^ ^ ' th is

construction associates a simplicial space Scylf , called the
simplicial mapping cylinder of f. Generally speaking, Scylf is not
homeomorphic to the usual mapping cylinder Cyl f of f (see 4.6.6.10),
but has similar properties.
The most suitable language for describing the space Scylf
is that of schemes. Thus, let sch X^ = (M^,S^), sch = (M2 ,S2),
and sch f = ((¡>,$). We defineScyl f by the formula
sch Scyl f = (M1 I |M ,S) , where S is the collection of (finite) sub-
i -1 -1
sets A c M 1J [M2 such that: (i)in^ (A) E S^ or in^ (A) = 0;

(ii) (Min” 1 (A)) U in2 1 (A) E S2 ; (iii) ifl n ^ 1 (A) / 0, then

in” 1 (A) E $ (S 1) ; (iv) if a 1 E in^ 1 (A) and a 2 E in2 1 (A), then

a2 < cj)(a^ ) - Themaps in^ andin2 define twomaps sch X^ sch Scyl f
and sch X 2 sch Scyl f, and hence two simplicial embeddings,
X Scyl f and X 9 -> Scyl f . The images of these embeddings are called
i z
the (lower and upper) bases of the cylinder Scylf and can be
identified with X1 and X 2 - Moreover, the map M 1 x (0 U 1) M 1J_[_M2
defined by (a,0 ) in 1 (a) ,(a,1 ) in2 ° <j>(a), induces acertain map
sch (X„ x I) -* sch Scyl f, and hence a simplicial map X 1 * I -> Scyl f .
1 S I S
Clearly, together with the inclusion X 2 -+■ Scyl f , this simplicial map
 114

yields a continuous map (X^ x I )J Scyl f , which in turn induces

a continuous map csc f : Cyl f Scyl f . Moreover, we see that
csc f (Cyl f ) = Scyl f , and that the canonical retraction
rtf : Cyl f -► X^ (see 1.2.6.10) is constant on the elements of the
partition zer(csc f) . Also, the canonical X^-homotopy from id(Cylf)

to the composite map Cyl f ^ ^ > X^ — — — ► Cyl f is constant on the

elements of the partition zer(csc f)* zer(id I) . Consequently, rtf
defines a strong deformation retraction Scylf X^f and the
composition of the inclusion X^ -> Scylf with this retraction
obviously equals f. We conclude that the inclusion X^ Scyl f is
always a homotopy equivalence, whereas the inclusion X^ ->Scylf is a
homotopy equivalence if and only if f is a homotopy equivalence.

6. Stars. Links. Regular Neighborhoods

1. The star of a simplex s in a simplicial space X is the

union of all simplices of X which contain s. Notation: Sts or
St(s,X). Clearly, Sts is a subspace of X.
The open star of the simplex sis the union of the interiors
of all simplices containing s. Notation: sts or st(s,X). It is
readily seen that st s is the open set defined by the inequalities

bar (x) > 0 ,...,bar (x) > 0,

an a
0 q

where aQ ,...,a^ are the vertices of s. Moreover, C i s t s = Sts .

The link of the simplex s is the union of all simplices in

Sts which do not intersect s.Notation: Iks or lk(s,X).
Clearly, lk s is a subspace of the spaces X and Sts .
The following are obvious facts.
If s' is a face of s, then

St s 1 c s t s , st s' c st s , lk s 1 c Iks .

If a^,...,a^ are vertices which do not sit in the same

simplex,then the intersection sta, is empty. However, if
aQ,...,a
u ^ are vertices of a simplex s, then
q
n r n st a . = st s .
i=0 l

If X is a subspace of X*, then

s t (s,X) = st (s,X *) n X
 115

for any simplex s of X. Moreover, if X is complete, then

St(s,X) = St (s ,X 1) n X and lk(s,X) = lk(s,X') fl X.

Finally, if s 1is a simplex of lk(s,X), then

lk(s' ,1k(s,X)) = lk(s",X ) ,

where s" is the smallest simplex containing s and s f.

2. We can extend the definition of the star, open star, and

link to points of a simplicial space X: for x £ X, the star Stx =
St(x,X ) , the open star stx = st(x,X), and the link lkx = lk(x,X),
are defined as

St x = St six, st x = stsix, and lk x = St six \ stsix.

Obviously, stx is a neighborhood of x, and lk x =

= Fr St x = Frstx. In addition, the star St x is homeomorphic to the
cone over lk x . In fact, the formula

pr (y ,t) ^ <M(1 -t)<j> 1 (x) + t<(> (y)) /

where y £ lk x and t £ I, defines a canonical homeomorphism

con lk x ■+ St x ; here <p is any characteristic homeomorphism of any
simplex containing x and y.

Warning: the equality lk x = lk si x holds only when x is

a vertex.
If s' is a simplex of lk(s,X), then by 1.1 and 5.10, the
join s * s' is canonically simplicial homeomorphic to the smallest
simplex of X containing both s and s'. It is clear that all these
simplicial homeomorphisms together yield a simplicial homeomorphism
s * l k (s ,X ) -* St (s,X) . If follows that the star St(x,X) of any point
x £ X is canonically simplicial homeomorphic to six * lk.(six,X) ,
and we readily see that this simplicial homeomorphism maps the join of
the boundary of six with the link lk(six,X) onto lk(x,X).
_ . ,. ndim si x - 1 _ ,, . .
Moreover, since Fr si x is homeomorphic to S and the join
sdim six-1 ^ lk(six,X) is homeomorphic to the iterated suspension
sudim s i x l k js i x ^X j ^ we conc]_Uc[e that lk(x,X) is homeomorphic to
su dim six.. , . x ,X)
lk (si Vv .
 116

The link of a point is a homotopy invariant

3 (LEMMA). Let A and B be retracts of atopological space

Y. If the inclusions i: A + Y and j:B Y are homotopic to some
maps f: A -> Y and g : B -* Y such thatf (A ) cz B and g (B ) c A, then
A and B are homotopy equivalent.

PROOF. Consider two arbitrary retractions, a: Y B and

p: Y A. Then the restrictions c I and p|D are homotopy equivalen-
ces A B and B A, and inverses of one another. Indeed,
oL = o ° i, pL = p ° j, and the composition p ° j ° a ° i is
A B
homotopic to p ° j 0 a 0 f=p°f, which in turn is homotopic to
p o i = id A . Therefore, pL ° a Iis homotopic to id A , and a
i
similar argument proves that g J^ ° p|^ is homotopic to idB .

4. Let T^ and T^ be subspaces of the topological space

X, and assume that both T^ and are endowed with finite
triangulations. If x^ G X is an interior point for both T^ and
then the links lk(xQ,T^) and lk(xQ,T2> have the same homotopy type.

PROOF. Let F^ denote a homotopy St(x^,T^) x I x from

the inclusion St(XQ,T^) X tothe constant map St(xQ,T_^) x^ X,
such that F^ is rectilinear on each simplex of St(xQ,Tj (i = 1 ,2 ).

Set C^(t) = F^ (St (Xq ,T_^) x t) (i = 1,2). Since st(Xg,T^) is open

and St(x^,T^) is compact (i = 1 ,2 ), there exists e > 0 and 6 > 0

such that C 1 (e) c St(xQ ,T2), C 2 (e) c: St(xQ ,T ), C (6 ) c: C 2 (e), and

C 2 (6 ) c: C^ (e) . Moreover, since C^e) \ x Qis a retract of

St(x^,T^) (i = 1,2), C^ (e) ^ x q and C 2 (e) ^Xq are retracts of

Y = [St(xQ ,T^) U St (Xq ,T2)] \ x Q . Finally, the formulas

(y ,t) F 1 (y,6 t/e) and (y,t) F 2 (y,6t/e)

define homotopies (C^c) \ xQ) x i y and (C2 <e) \ x Q) x i + y

from the inclusions C ^ (e) \ xQ + Y and C 2 (e) ^ x Q -> Y to maps whose
images lie in C 2 ^ ^ x0 and C 1 ^ ^ X0 ' respectively. Consequently,
C 2 (c) \ x Q and C (e) ^ x Q have the same homotopy type (see 3 ),and it
remains to note that C^(e) \ X q has the same homotopy type as
lk(xQ,Ti) (i = 1,2).
 117

Regular Neighborhoods

5. The regular neighborhood of the subspace A of a

simplicial space X is the union of the open stars st(a,X) with
a E A or, equivalently, the union of the open stars st(a,X) with
a £ ske^A.

If the subspace A is complete, then A is a deformation

retract of its regular neighborhood U. In fact, there is even a
canonical A-homotopy h: U x I U from id U to the composition of
a retraction U -> A with the inclusion A -> U . This homotopy is given
by

1 - t I barb (x) bar (x)

bEske^A a
-, if a £ ske^A,
bar, (x)
bEske^A b
bar (h(x,t)) = <
a
t bar (x) , if a £ skenX \ ske A.
a u u

In particular, this shows that every subspace of a simplicial

space X is a deformation retract of its regular neighborhood in ba X
(see 5.4).

Barycentric Stars and Barycentric Links

6 . The barycentric star of the simplex s of a simplicial

space X is the union of all simplices of ba X which have as their
first vertex the center of s. Notation: bsts or bst(s,X). An
equivalent description: bsts is the set of all points x £ X such
that

bar (x) = bar, (x), if a,b £ s D skenX,

a b U
and
bar (x) 5 bar, (x) , if a G ske_X, b G (X \ s) D skenX.
a b u u

It is clear that the barycentric stars of the simplices of X

cover X and are subspaces of baX . Moreover, bst s f bst s'
whenever s f s', and bsts c bsts' whenever s o s'.

7. The union of those simplices of the barycentric star

bst s which do not contain the center of s is the barycentric link
 118

of the simplex s, and is denoted by blk s . The star bsts is

clearly simplicial homeomorphic to the cone over blk s . Moreover,
the rectilinear projection from the center of s induces a homeomorphism
of blk s onto the link lk s of the simplex s in X (and the
barycentric subdivision of lk s transforms this homeomorphism into s
simplicial one). Therefore, the pairs (bst s,blk s) and
(con Iks,Iks) are homeomorphic.

7. Simplicial Approximation of Continuous Maps

1. Let f: X Y be a continuous map, where X and Y are

simplicial spaces. A simplicial map g: X Y is a simplicial
approximation of f if g(x) £ sif(x) for any point x £ X.

Every simplicial approximation g of the map f: X Y is

canonically homotopic to f: the canonical homotopy X x I Y from
f to g is an affine mapping from each generatrix x x I of the
cylinder X x I onto the (possibly degenerate) rectilinear segment
joining f(x) and g(x). It is clear that this homotopy is stationary
on the set of the points x £ X where g(x) = f(x).

2. A simplicial map g: X Y is a simplicial approximation

of the continuous map f: X Y if and only if f(sta) c= stg(a) for
every vertex a of X.

PROOF. Assume first that g is a simplicial approximation

of f, and let x £ st a . Recalling that g(x) £ si f (x) , that g is
simplicial, and that x lies in the interior of a simplex with vertex
a, we conclude that g(x) lies in the interior of a simplex with
vertex g(a) (see 1.3). Thus, g(a) is a vertex of si f (x) , and
hence f (x) c: stg(a).
Now suppose that f(sta) c stg(a) for every vertex a of
X. Pick x £ X; if aQ ,...,a are the vertices of si f (x) , then
q u 4
x £ st a^ , whence

f (x) e f (ni=o s t a ±) c ni=0 flsta.) <= nq =0 Stgla.) .

Therefore, the points g(aQ),...,g(a ) are among the vertices of the

simplex sif(x) , and since g(x) lies in the simplex with vertices
g (aQ),...,g(a ) , g(x) € si f (x) .

3. A continuous map f : X ->■ Y of simplicial spaces has a

simplicial approximation if and only if for each vertex a of X there
 119

is a vertex b of_ Y such that f(sta) cr st b .

The necessity of this condition is an immediate consequence

of 2. To prove its sufficiency, fix a map cp: skeQX -+ skeQY such that
f(sta) c st (p (a )for every vertex a € ske^X. If aA ,...,a are the
a 0 0 q
vertices of a simplex of X, then sta. i 0, and the inclusions

ni = 0 s t ( P(a j_) 3 ni =of ( s t a i ) 3 f ^ni =0 s t a i )

demonstrate that stcp(a^) f0 too. This in turn implies that

cp(aQ ) ,. ../<p (a ) are among the vertices of a simplex of Y (see 6.1),

Therefore, <p extends to a simplicial map X -> Y (see 2.5) and,

applying 2 , this extension is a simplicial approximation of f.

4. For each continuous map f of a finite simplicial spac

X into a simplicial space Y there is a positive integer m such that
the map f : bamX Y admits a simplicial approximation.

Without loss of generality, we may assume that X is a

polyhedron (see 4.1). Since the open stars of the vertices of Y
constitute an open cover, there is c > 0 such that, given any subset
A of X with diamA < e, f(A) is contained in one of these open
stars (see 1.1.7.16). Let m belarge enough so that the simplices of
bamx have diameters less than e /2 (see 5.6). Then given any vertex
of bamX, the diameter of its star is less than e, and 3 shows that
f: bamX -> Y has a simplicial approximation.

8. Exercises

1. Let X be a simplicial space. Show that the formula

dist(x,y) = [ I (bar (y) - bar (x) ) 2 ] 1 ^ 2

a€skeQX

defines a metric on X, and verify that the resulting metric topology

coincides with the initial topology if and only if X is locally finite.

2. Show that for every polyhedron X c IRn there is a

triangulation of IRn by Euclidean simplices, relative to which X
becomes a simplicial subspace of IRn .
3. Show that every connected, locally finite, .n-dimensional
simplicial space can be simplicially embedded in IR triangulated
by Euclidean simplices.
4. Let f: X Y be continuous, where X and Y are
 120

simplicial spaces. Produce a new triangulation of X with the

following two properties: a) each of its simplices is contained in
one of the simplices of the original triangulation; b) f has a
simplicial approximation relative to the new triangulation.

§3. HOMOTOPY PROPERTIES OF CELLULAR SPACES

1. Cellular Pairs

1. Suppose that X is a rigged cellular space and A is

subspace of X. Let h ^ : A U ske^X -+ I denote the function equal to
zero on A and equal to 1 on (A U skeQX) \ A , and define inductively
a sequence of functions h^_: A U ske^X I (r = 1,2,...), such that

hr-1 ^ ' if x E A U sker_^X,

h (x) =
r
1 - t [1 - h r _ 1 (atte (y))], if x = chae <Ty)

r— 1
where e E cell X \ cell A, t £ I, and y E S . Since the functions
r r
hr are continuous and each of them extends the preceding one, together
they yield a continuous function X ^ I. This function is called the
characteristic function of the pair (X,A), and the neighborhood of A
consisting of all points of X where the characteristic function is
less than 1 is called the neat neighborhood of the subspace A.
Obviously, the characteristic function of the pair (X,A)
vanishes on A, and only on A; hence, every subspace of a cellular
space is distinguishable.

If X is a simplicial space, then we may construct a

characteristic function starting with a simplicial rigging of X, and
it is readily seen that such a function does not depend upon the choice
of the rigging. In this case, the neat neighborhood of a subspace A
is simply the regular neighborhood of A in ba X .

2. Every subspace A of a rigged cellular space X is a

strong deformation retract of its neat neighborhood.

PROOF. Let U denote the neat neighborhood of A in X.

Since the products (A U ske^X) x I are subspaces of the cylinder
X x I and cover it, they constitute a fundamental cover of X x I.
 121

Therefore, their intersections with U * I, i.e., the cylinders

Ur x I, where = U D (A U skerX), constitute a fundamental cover
of U x I. Let Gq be the constant homotopy of the inclusion A -*■ X,
and define homotopies Fr : Ur x i -> u, r 5 1, by the formula

if x € U
r- 1 '
Fr (x,t)
cha (((1 —t)i + t)y), if x = cha (ty ),

where e G cell^X ^ cell^A, i E (0,1], and y G Sr ^ . Now construct

homotopies Gr : x I u, r >. Î , by

if 0 g t £ 2 r,

Gr (x,t) = Fr (x,2 rt- 1 ) , if t < 2~r+1 ,

Gr- 1 (Fr (x,1),tl' i£ 2 rt1 i t c 1-

Each homotopy Gr extends the preceding one, and together they yield
an A-homotopy U x i -> u from id U to a map which takes U into A.
The compression of the last map to a map U A is the desired strong
deformation retraction.

3. Every cellular pair is a Borsuk pair.

This is a consequence of 2 combined with 1.3.5.11, because

cellular spaces are normal, and their subspaces are distinguishable.

4. If_ (X,A) is a cellular pair and the inclusion A X

is a homotopy equivalence, then A is a strong deformation retract
of X.

In order toprove this, firstapply theorems 3 and1.3.5 . 6

to the pair (X,A),then apply Theorem 3 to the pair
(X x I , (X x0) U (A x i) u (X x 1 )) and, finally, apply Theorem1.3.5.7
to the pair (X,A).

Cellular Pairs and k-Connectedness

5. Let k be a nonnegative integer or °°. Suppose that

(X,A) is a cellular pair such that all the cells in X ^ A have
dimension at most k, and let (Y,B) be an arbitrary k-connected
topological pair. Then every continuous map f: X Y such that
f(A) B is A-homotopic to a map which takes X into a subset of B.
In particular, every continuous map of a k-dimensional cellular space
 122

into a k-connected topological space is homotopic to a constant map.

PROOF. We exhibit a sequence of A-homotopies

{f ^ : (A U ske^X) x I -* each extending the preceding one, and
satisfying the conditions:
(i) F^(x,0) = f(x) for all x £ A U ske^X;

(ii) F ((A U skerX) x (1-2~r-1)) c: B;

— — 1
(iii) F (x,t) does not depend upon t for t ^ 1- 2

Then the map F : X x I ->Y which equals F^ on

(A U ske^X) x i will be ahomotopy fromf to a map which takes X
into a subset of B.
We proceed by induction. Define F_^ as the constant
homotopy of f|A , and assume that homotopies F_^, . . . ' each
extending its predecessor and satisfying (i)-(iii), are already
constructed. If q > k, then
A U ske X = X and we simply take
q
Fg = • So suppose now that q < k. Since the pair

({A U ske^X) x I,(AU ske^_^X) x I) isBorsuk (see 3), there is a

homotopy G of the map f|AUsk x, suchthat G| (AU ^ x) xl = Fq_., .

Using the fact that F . ( (A U ske .X) x (1-2 q )) c B, the formula

q- 1 q- 1
he (y) = G (cha^ (y) ,1-2defines a map h^: ->Y which takes S^ 1

into B, for each cell e £ cell X \ cell A. Now take advantage of the
q q
k-connectedness of the pair (Y,B) to deduce that, given any cell
e £ cell^X \ cell^A, there is an S^ 1-homotopy x i -y y from
he to a map whose image is a subset of B. We put

F .(x,t), if x £ A U ske .X,

q- 1 q- 1
G (x,t) , if 0 sc t £1-2-q,

F (x,t) = He <y,2 q+1(t - (1 - 2 q )) , if x = chae (y) and

M.
1 - 2 ~q £ t £ 1 -2 - q _ 1 ,

I H {y,1 ), if x = cha (y) and

e e
1 - 2 _ q_ 1 £ t si 1 .

Then it is immediate that the map F

is continuous, extends
q
F _w and fulfils properties(i)-(iii) with r = q.
q *

6. Let k be a nonnegative integer or If the cellular

pair (X ,A) is k-connected and every cell in X \ A is of dimension
at most k, then A is a strong deformation retract of X. In
particular, every k-connected k-dimensional cellular space is
 123

contractible.

Indeed, id X is A-homotopic to a map g: X * X such that

g (X) c: A (see 5) , and hence ab g : X -+ A is a strong deformation
retraction.

2. Cellular Approximation of Continuous Maps

1 (LEMMA) . Let A = A [J Ly£jy^Dy = + ^ )] , where A is

k
a topological space and cp is a continuous map | = s ) + A.
r "l"1 V-
Let f: D ■+ X be a continuous map such that f(S ) c A' = imm2 (A) .
Then:
r
(I) if r < k, f is S -homotopic to a map g such that
r +1
g(Dr ) c A ’;
27
(II) if_ r = k, f is S -homotopic to a map g such that
k +1 ^
there are affine maps ,...,ag: D -+ D with four properties:
k+ 1
(i) the images d^ = (D ) are pairwise disjoint balls lying in
k +1
Int D ; (ii) each of the compositions g ° ou coincides with one
of the composite maps

t in imm
D ----- | I D ---- -— > X ; (1 )
y J— lv GM v
k +1 s
(iii) g(D \ U ± _ 1 Int d^) c= A'; (iv) for k ^ 1, the point of the

ball d^ having the largest value of the first coordinate is just

a^(ort^), and the segment joining this point with ort^ is entirely
k +1 s
contained in D \ Uj= 1 Int ^ = 1/---/S).

(Part (I) of this lemma, i.e., the case r < k, merely

asserts that the pair (X,A*) is k-connected, and this is the only
information that we shall actually use in the present section; part
(II) is needed in § 5.3 .)
PROOF.
Denote the composite map (1) by h , fix an arbitrary
k +1 ^
Euclidean (k+1)-simplex a in D , and thenfix in Inta an
2
arbitrary (k+1)-simplex t of the triangulation ba a. The sets
h (Int a) , y C M, and x ^ uy£M constitute an open cover of X,
r +1
and so there is a triangulation of D which is fine enough to ensure
that the image of any of its simplices under f lies in one of the sets
of this cover (see 2.5.6 and 1.1.7.6 ). Let (respectively, L) be
 124

the union of those simplices whose images are contained in h (Into)

(respectively, in X \ U h (t )). Obviously, the sets K are pairwise
^ J • y y U
disjoint, only a finite number of them are nonempty, K and L are
r +1 ^
simplicial subspaces of the simplicial space D , and L U ( U K ) =
= D
Now apply Theorem 2.7.4 to the composite maps

, _ (abh ) ' 1 ,
K -a- -> h (a) ----- y---- * ba a - (2)
y V

This theorem tells us that there is an m such that the maps (2) admit
simplicial approximations when one replaces by bamK ^ . We let
denote the canonical homotopy from (2 ) to the above simplicial
approximation. Since t is not a face of any other simplex,
F^ ((L D K ) x I) D I n t t = 0, and so together the homotopies F^
define a homotopy F: ( L D ( U K ) ) x I ^ X \ U h (Inti) . By 1.3 and
y y y y
1.3.5.9, F extends to a homotopy G: L x I X \ U h (Int t ) of the
y y
r+1 —1
map ab f : L X \ U^h^(Inti) , stationary on B = D \ Uyf (h^(Intx)).
It is evident that the composite maps

F h.
K x
X iI -— L * a yla -> X
v
and
L x I -JL_» x ^ U h (Int t )
V y

yield together a B-homotopy of f. This homotopy connects f to a map

_ r+1
f^: D X such that, for every y, the composition

ab f (abh ) ' 1
ba K ----- — »h (ct) ----- y > ba a

is simplicial and h (t) c f (k ) c h (a). Since B => Sr , f. is

2~ y I y y \
S -homotopic to f, and to complete the proof of our lemma it suffices
to examine (I) and (II) for f r a t h e r than f.
Consider an arbitrary ball 6 <= Int t , and let ib denote the
k +1
homeomorphism 6 D , (x) = (x — a)/p, where a and p are the
center and the radius of 6. Moreover, let 1': Dk +1 x i -* Dk +1 be
defined as
x - t a
¥(x,t) = U
1 - tQ (1-p)

where is the largest of the numbers 0 e [0 ,t] such that

tQ
k -f-1
(x - 0a)/(1 - 0 (1 —p)) £ D . Now the family of mappings p : X X,
given by
V
 125

X, if x € A' ,
Pt (x) =
° f (h^ 1 (x) ,t) , if x € h (Int D^ +1 ) ,

obviously yields an A '-homotopy of id X such that p^ (X \ Ujh (6 )) c A'

and Pl ° = h^o if r < k, then f.J(Dr + 1) c= X U^h^ (6 ) , and to
complete the proof in this case we only need to note that
(y,t) »-* p^_(f^ (y)) is an S -homotopy from f^ to a map whose image is
included in A'. If r = k, then f ^ 1 (U h^ (<5) ) can be decomposed into
pairwise disjoint ellipsoids (^,...,5^, each being affinely mapped
onto one of the sets h (6 ) by f1. Let {qt> be an Sr-homotopy of
k +1 —i
id D , with the following properties: the preimages d^ = (6 )
are balls; the maps ab q^ : di 6 are affine; and for k ^ 1 the
point of the ball d.having the largest value of the first coordinate
1 -1
is carried by f^ ° into one of the points ^ (ort^)), and the

segment joining this point with ort is entirely contained in

k +1 s 1
D \ Uj = 1 Int • Obviously, the formula (y,t) » (f^ ° qt (y),t)
r
defines an S -homotopy from f^ to a map g satisfying conditions
(i)-(iv) .

2. Every cellular pair (X,A) with

A => ske.X is
K
k-connected (0 < k £ °°) . In particular, every cellular space whose
k-skeleton reduces to a point is k-connected.

PROOF. Since any continuous map of a ball into X takes the

ball into a subset of one of the skeletons ske X, it suffices to show
r
that all the pairs (A U ske^X,A) with r > k are k-connected. But
this is an immediate consequence of the k-connectedness of the pairs
(A U ske >.X,A U ske X) with
.X,A t q ^ k, which in turn follows at once
q
q++11 q
q
from Lemma 1 (see 1.2.1) .

3 (COROLLARIES) . S is (n-1 )-connected (n ^ 1) . (EP is

2
simply connected (0 £ n £ °°) . (EaP is 7-connected.

4. Every continuous map f from a cellular space X into

cellular space Y is homotopic to a cellular map. If, in addition, f
is cellular on a subspace A of X, then f is A-homotopic to a
cellular map.

PROOF. Given f: X -> Y , continuous on X and cellular on

°0
A, we shall construct a sequence of maps {fr : x + Y Jr__i and a
o°
sequence of homotopies iFr : X x I ->• Y)r_Q, such that:

(i) f _ 1 = f;
 126

ii) f is cellular on A ske X;

r r
(iii) F is a homotopy from to ^r ' stationary on
A U ske .X.
r- 1
Then the formula

Fr (x,2 - 2r+ 1 (1-t)) , if 1-2 r £ t S 1-2 r 1

F (x,t), if x E A U ske^X and t = 1,

will define an A-homotopy from f to a cellular map.

We proceed by induction. If f^ and F^ are already
constructed for r <k and satisfy (i)-(iii), then by 2 there is a
(ske^A U skek _ 1 X ) -homotopy from to a maP whose image is
f k_-||s k e
x
k
contained in ske^Y. This homotopy together with the constant homotopy
of fk-1 kuske, „X Yield SOme <A U ske^X)-homotopy of x-
K—1 K
Applying 1.3, the last homotopy extends to some (A U ske^._^X)-homotopy

of the map * which we take as F^. Finally, set = F^.(x,1)f

x E X.

5. Two homotopic cellular maps f,g: X Y are cellular

homotopic. If, in addition, fand g are A-homotopic and A is a
subspace of X, then f and g are cellular A-homotopic.

Surely, every A-homotopy from f to g is a continuous map

of the cylinder X xI into Y, and is cellular on (X x (0 U 1)) U (Ax I).
By Theorem 4, this map is [(Xx(0U1))U(Axi)]-homotopic to a cellularmap.

3. k-Connected Cellular Pairs

1. Every k-connected cellular pair (X,A) (0 £ k £ °°) is

homotopy equivalent to a cellular pair (Y,B) such that B 3 ske.Y.
.K
PROOF. If k = 00, then, by 1.6, A is a strong deformation
retract of X, and hence the pair (X,A) is homotopy equivalent to
(X,X) .
Turning now to the case k < we may assume that
A => ske^_.jX; indeed, one reduces to this case by induction on k,
because for k 5 1 every k-connected pair is also (k-1)-connected,
while the condition A => s k e ^ X is trivially fulfilled. According to
1.5 and2.5, there is a cellular ske^A-homotopy f: ske^X x i -> x from
the inclusion ske^X -> X to a map which takes ske^X into A. Define
 127

F:skekX x I x i + xby Ffx,^,^) = ffx,^), and set

C= (skekX x i x 0) U (skekX x (0 U 1) x I) u (skekA x Ix i)

and
D = ske. X x i x 1 .
k

Obviously, C and D are subspaces of the cellular space

skekXX 1 X I 'and the map F is cellular. Now define Y and B by

Y= X Up j (skekX x i x i), b = imm2 (A) U imm1 (D).

By 1 .5.5, Y is a cellular space, and it is clear that B is

a subspace of Y containing skekY. To verify that (X,A) and (Y,B)
have the same homotopy type, note that imm2 (A) is a strong deformation
retract of B and imm2 (X) is a strong deformation retract of Y. In
fact, the formula

(imm1 (x,t1 , 1 ) ,t) h- imm1 (x,tt1 , 1 ) [x €skekX, t,t1 £ I],

-

(imm2 (x),t) h- imm2 (x) [x € A, t £ I],

defines a homotopy B * I -»■ B,stationary on imm2 (A) , from id B to

a retraction B ->- imm2 (A) . Similarly, the formula

(imm1 (x, t^ ,t2) ,t) h- imm1 (x,t1 ,tt2) [x £ skekX, t,t 1 ,t2 £I],
•<

(imm2 (x),t) imm2 (x) [x £ X, t £ I],

defines a homotopy Y x .I -> Y ,stationary on imm2 (Y), from id Y to

a retraction Y -> imm2 (X). Consequently, the pair (Y,imm2 (A)) is
homotopy equivalent to both the pairs (Y,B) (see 1.3.5.8 ) and
(imm2 (X),imm2 (A)), and it remains to observe that the pairs
(imm2 (X),imm2 (A)) and (X,A) are homeomorphic.

2. Every k-connected cellular space (0 £ k £ °°) is homotopy

equivalent to a cellular space whose k-skeleton reduces to a point.

PROOF. Let X be a k-connected cellular space and choose a

0-cell xQ in X. The pair (X,xQ) is homotopy equivalent to a cel­
lular pair (X,A) such that A => ske^X (see 1). Set Y = X/A. Since
A is contractible, Y has the same homotopy type as X(see 1.3.7.7
and 1.3), and it is clear that ske^Y is just a point.

3. Theorem 2 says nothing about the dimension of the space

Y which replaces the given space X. However, its proof demonstrates
that one can always choose Y to satisfy dimY £ max(dim X ,k +2). Our
 128

next task is to prove that for k = 0 the last equality may be

sharpened to dimY - dim X (see 6 ).
r 00
4 (LEMMA). Let Y
be a topological space, and let iY^^^ =q
be a fundamental cover of Y such that yk n y i 0 whenever k - 1 > 1 .
If Y, „ D Y, is a stronq deformation retract of Y, for all k > 1 ,
— k- 1 k ^----------------------- k — — — —
then Yn is a strong deformation retract of Y .

PROOF. If : Y x I is a homotopy, stationary on

from id Yv to a map which takes k into Yk - 1 n Yk ,
Vi nV
then the formula

if y G Y, and t £ 2
J k
(y /1 )
F i ( F 1 +1 ( ... F k (x,1 ) ...,1),2 t-1 ) , if y € Y. and
J k

2 1 a t < 2 1+1 (1 S k),

defines an Y^-homotopy Y x i y from id Y to a map which takes Y

into Yq .

5. Given any connected cellular space X, there is a

contractible one-dimensional subspace of X containing all the 0-cells.

PROOF. Fix an arbitrary 0-cell x^ in X and let A^ be t

the set of all 0 -cells that can be joined to by a path I ske^X
^0
which touches at most k 1-cells. Since ske^X is connected (see
1.4.7), and a path can touch only a finite number of cells, U
k =0 Ak
= ske^X. Now given any 0-cell x E A^ \ Ak-1 k ^ 1, pick a
closed 1 -cell c (x) joining x to some cell in and set
k- 1 k- 2 '

O' if k = 0,

Yk =
U c (y) , if k > 0,
y e A k'Ak-i
uu
and Y = Uk=Q Yk . Obviously, Y is a one-dimensional subspace of X
containing skeQX, and the cover {Yk } of Y satisfies the conditions
of Lemma 4. Therefore, YQ is a strong deformation retract of Y,
i.e., Y is contractible.

6. Every connected n-dimensional cellular space is homotopy

equivalent to a cellular space of dimension at most n, and having
only one 0-cell. In particular, every connected one-dimensional
cellular space is homotopy equivalent to a bouquet of circles.

This results from 5, 1.3, and 1.3.7.7.

 129

Applications to Cellular Constructions

7. If the cellular space X is k-connected, then su X and

su(X,Xq), where xQ is a 0 -cell, are (k+1 )-connected.

The proof reduces to three remarks. First, since su X and

su(X,xQ) have the same homotopy type (see 1.4.5, 1.3.6 .8 , and 1.3. 7. 7),
the (k+1 )-connectedness of one is equivalent to the (k+1 )-connectedness
of the other. Secondly, according to 2 and 1.3.7.12, it is enough to
verify that su(X,xQ) is (k+1)-connected when ske^X = x Q . And thirdly,
if ske^X = X q , then the (k+1)-connectedness of s u (X,X q ) is a
corollary of Theorem 2, because under this assumption ske, 1 su(X,x )
k+f u
also reduces to a point.

8. Suppose X^ is a k^-connected cellular space and x^ is

a 0-cell of X^, i = 1,2. Then the tensor products (X^,x^) 0 (X^^x^)
and (X.,x ) 0 (Xn ,xn) are (k +k +1)-connected.
1 1 c 2 2 1 2 ----------
Again, the proof reduces to three remarks. First, since
(X^,x^) ® (X^fX^) induces on its compact subsets topologies which are
identical to those induced by the topology of (X^,x^) (X^fX^, the
(k1 +k2 + 1 )-connectedness of one of these spaces implies the
(ki+k2 + 1 )-connectedness of the other. Secondly, using 1 and 1.3.7.12,
it is enough to verify that (X^,x^) (X^/X^) is (k^+k2 + 1 )-connected
when skev X = x and ske, X = x 9 . Thirdly, under these
k^ 1 1 .K2
circumstances, the (k1 +k2 +1 )-connectedness of (X^,x^) ©c (X2 ,x2)
follows from Theorem 2, because skek ®c (X2 'x 2>) also
reduces to a point. ^

9 (LEMMA). For any cellular spaces and X 2 with

0 -cells x^ and x 2 taken as base points, the cellular join
(XwX„) * (X0 ,x„) is homotopy equivalent tosu((X„,x„) ®_ (X0 ,x0 ),bp).
1 1 C z z ------------ — 1 1 I I C z z
PROOF. By definition, the spaces (X^x^) *c (X2 ,x2> and
su((X 1 ,x1) ©c (X2 ,x2 ),bp) are obtained from (X1 x^ X2> x I by taking
quotients two times, and the projection (X1 xc x2^ * 1
-> su ( (x^ ,x ^) ©c (X2 ,x2 ),bp) is constant on the elements of the
partition zer (pr:(X., xc x2 ) x I -> ^ , x ^ ) *c (X2 ,x2)). The resulting
map
f = fact [pr: (X1 xc x2 ) x I -v(X1 ,Xl) *c (X2 ,x2)]:

: su((X ®c (X2 ,x2 ),bp) -> (X1 ,x1 ) *c (X2 ,x2 )

 130

is factorial (see 1.2.3.4). Since the only element of the partition

-1
zer(f) which does not reduce to a point is f (bp), we see that
sudx^x^ (X2 ,x2 ),bp) = [ (X1 /X1 ) *c (X2 ,x2)]/f 1 (bp) . Finally,

note that f_ 1 (bp) = [(X ,x.j) *c (x2 ,x2 >] U [ ( x ^ x ^ *c (X2 ,x2 )], and

since this union is contractible, the quotient space

[(X.jjX.j) *c (X2 ,x2)]/f 1 (bp) is homotopy equivalent to

(X1 'x 1 > *c (X2 'X 2 )*

10. Letthe cellularspaces X^ and X 2 be k^- and
respectively k^-connected. Then the joins X^ * X2 , X^ *c X2 ,
(X1 ,x^ ) * (X2 ,x2), and (X1 /x 1) *c (X2 ,x2 >, where x1 and x 2 are
0 -cells, are (k^ +k2 +2 )-connected.

The proof reduces to four remarks. First, since

(X1 fX1) * (X9 ,x9) i s a q u o t i e n t of X * X 9 by a c o n t r a c t i b l e s p a c e
1 1 C Z Z I c z
(the closed 1 -cell x^ * x 2), (X^,x^) *c (X2 ,x2 ) and X^ *c X 2 have
the same homotopy type. Secondly, X^ *c X 2 induces on its compact
subsets the same topologies as does X^ * x 2 , and hence the (k^+k2 +2 )-
connectedness of one of these spaces implies the (k^ +k2 +2 )-connectedness
of the other. Thirdly, and from the same reason, (X ,x^) * (X2 ,x2) is
(ki+k 2 +2 )-connected if and only if (X^,x^)*c (X2 ,x2) is (k^+k2 +2 )-
connected. Fourthly, the (k^+k2 +2)-connectedness of (X^,x^) *c (X2 ,x2)
is an immediate consequence of 9, 8 , and 7.

4. Simplicial Approximation of Cellular Spaces

1 (LEMMA). Suppose that X and Y are cellular spaces and

00 OO
r r=0 and r r=0 are filtrations of X and Y by subspaces.
Let f : X Y be a cellular map such thatf (X^) c: Y^ (0 £ r £ «>) .
If all the maps ab f : xr ^ Yr are homotopy equivalence, then so is f.

PROOF. Since the = Cyl(abf : X^ + Y ) are cellular sub­

spaces of Z = Cyl f and satisfy the conditions Z cz Z „ and
i r r +1
U n Z = Z, they yield a filtration of Z (see 1.1.9). Thus, the
k
image of any continuous map D Z is contained in one of the sets
Z^ (see 1.2.4.5), and so the pair (Z,X) is «-connected provided that
all the pairs (Z ,X ) are ^-connected. Now note that (Z ,X ) is
r r r r
»-connected if and only if ab f : X^ Y^ is a homotopy equivalence;
similarly, (Z,X) is “-connected if and only if f is a homotopy
equivalence (see 1.3, 1.6, 1.3.3.9, and 1.3.7.13).
 131

2. Given any cellular space X, there is a simplicialspace

which has the same homotopy type and the same dimension as X ,and is
finite or countable together with X .

The proof consists of producing three sequences: one of

simplicial spaces {Y } , one of simplicial embeddings
r• 1 00
11 : Yr + 1 J , and one of cellular homotopy equivalences

{fr *
. skerX Yr }r_Q, with the following four properties:

(l) fr|ske X = lr-1 ° fr-1;

' r- 1
(ii) dim Yr = dim ske^X ;

(iii) if ske^X is finite (countable), then Y^ is finite

(respectively, countable);

(iv) if ske X = ske .X, then Y = Y . and i .

r r- 1 r r- 1 r- 1
* l dYr - 1 •
This will enable us to define the simplicial space limfY^i^)
having dimension dim X , and finite or countable together with X, as
well as a cellular map f : X -> lim(Y ,i ) such that fI , = imm 0 f.
r r sKe x. r
1 r
Finally, we shall use Lemma 1 to show that f is a homotopy equivalence.
Define Yq and f^ as ske^X and id ske^X , and assume
that simplicial spaces Y , cellular homotopy equivalences f^, and
simplicial embeddings ir_'| ' satisfying (i)-(iv), are already
constructed for r < q. By 1.2.1, we may represent ske^X as
ske „X U A, where A = II - n v (D = Dq ) and cp is a continuous
cr-1 cp J— Letcell qX e

map of Z = I I , -- v (S = Sq“1) into ske X. Next triangulate

c - — Le€cell X e q-1
q
A so that Z becomes a complete subspace and the map ° ^ : ^ ^ Yq-1
admits a simplicial approximation g: E -> Y .. Further, order Ya _ 1
q i Si 1
and £ in such a manner that the map g becomes monotone. Applying
successively Theorems 1.3 .7.10, 1.3.7. 8 , and then again 1.3.7.10, we
obtain three homotopy equivalences:
a homotopy equivalence ske X -* Y . Uf A which
q q 1 q - 1 ><p
aqrees with f . on ske .X;
^ q- 1 q- 1
a homotopy equivalence Ya_ 4 - 1 Uqy A
q -| Uf - o^cp A ^ Yq which
is the identity on Yq_i ;
and a homotopy equivalence Yq_i Ug ^ (Scyl g) A,

where in = [in: I ->■ Scyl g ] (see 2.5.11), whichagrees with the

inclusion Y _1 Scyl g on Y .
q i vi
At last, we may define Y as (Scyl g) U. A, f as the
M ^1
 132

composition of the three homotopy equivalences above, and i as the

q *
composite embedding Scyl g -> Y ^ . The triangulations of A and
of the cylinder Scylgyield together a triangulation of (see Y
q
2.5.2). It is plain that -*-s a simplicial embedding and that
Y , f , and i „ satisfy conditions (i)-(iv) for r = q.
q q q-1
3. Let X and Y be cellular spaces with X finite and Y
countable. Then the settt(X,Y) is countable.

PROOF. By theorems 2 and 1.3. 1.8, we need only consider the

case when X and Y are simplicial spaces. Under this assumption,
Theorem 2.4.7 shows that the cardinal of tt (X,Y ) does not exceed the
cardinal of the set of all simplicial mappings bamX Y (m = 0 ,1 , . ..) ,
and the latter is obviously countable.

5. Exercises

1. Suppose that the cellular spaces X^ and X^ are homotopy

equivalent, i = 1,2. Show that X^ x^ x 2 and Xjj ^ x^ are homotopy
equivalent, and that the same is true for the spaces X^ *c X 2 and
X i *c X 2 "
2. Show that every cellular space is homotopy equivalent to
a locally finite cellular space.

3. Show that every cellular pair is homotopy equivalent to

a simplicial pair, and that every finite cellular pair is homotopy
equivalent to a finite simplicial pair.

4. Show that there is no cellular space having the same

homotopy type as the subspace of the real line consisting of the points
0 and 1 /n, n = 1 ,2 ,... .
 Chapter 3. Smooth Manifolds

§1 . FUNDAMENTAL CONCEPTS

1. Topological Manifolds

1. This chapter comprises an elementary introduction to

differential topology. The basic objects of this theory are the smooth
manifolds. They are defined in the next subsection and represent (as do
cellular and simplicial spaces) topological spaces with an additional
structure. The present subsection is devoted to topological manifolds,
which occupy an intermediate position between smooth manifolds and
topological spaces, and do not carry an additional structure.

Locally Euclidean Spaces

2. A topological space is said to be a an n-dimensional

locally Euclidean space if each of its points has a neighborhood homeo-
morphic to the space ]Rn or to the half space ]R^, where ]R^ is the
set of all points (x„ ,...,x ) G ]Rn with x>, £ 0. The half space ]Rn
" I n I -
is defined for n ^ 1 ; we do not define it for n = 0 and, accordingly,
a 0-dimensional locally Euclidean space is simply a topological space
such that each of its points has a neighborhood homeomorphic to ]R^,
i.e., a discrete space.
In a locally Euclidean space X, the points having a
neighborhood homeomorphic to IRn are called interior points, while the
remaining ones are called boundary points. The interior (boundary)
points form the interior (respectively, the boundary) of the locally
Euclidean space X, denoted by intX (respectively 3X). (The dif­
ference between the notations int,9 and Int,Fr should prevent us,
 134

in each context, from confusing the interior and boundary points, and
the interior part and boundary defined here with the interior and
boundary points and the interior part and boundary of a set in a
topological space.) Clearly, the interior of X is a dense open set,
whereas the boundary of X is closed.
If each point of a topological space has a neighborhood homeo-
morphic to an open subset of IRn or , then obviously it already is
an n-dimensional locally Euclidean space. Consequently, every open
subset of an n-dimensional locally Euclidean space is also an
n-dimensional locally Euclidean space. In particular, the interior of
an n-dimensional Euclidean space is an n-dimensional locally Euclidean
space without boundary. Moreover, the interior and boundary of an open
subset U of a locally Euclidean space X are given by int U =
= U n int X and d\J = U fl 3X.
Since a locally Euclidean space is locally connected, its
components are open (see 1 .3.4.3), and hence also closed.
Obvious examples of n-dimensional locally Euclidean spaces
are 3Rn , 3R^, Sn , and Dn . It is clear that 3nRn = 0 and 3Sn = 0.
Furthermore, all the boundary points of the half space lie in the
limiting hyperplane ^, consisting of the points (x ,.../XR ) such
that x^ = 0 , and all the boundary points of the ball D lie in the
limiting sphere Sn 1.
n n2 n i +n2
3. Since the product HR x 3R is homeomorphic to 3R ,
we see that the product X^ x of two locally Euclidean spaces X^
and X^ of dimensions n^ and , and without boundary, is an
(n i+n2)-dimensional locally Euclidean space. This is true in general,
i.e., a product X x ... x x
of boundaryless locally Euclidean spaces
s 1
X i,...,X
s of dimensions n i ,...,n
s , is an (n I+...+n
s )-dimensional
boundaryless locally Euclidean space. Turning to locally Euclidean
spaces with boundary, note that the formula ((x ,...,x ),(y ,...,y ))
1 nl 1 n2
(x ,. ..,x fyw...fy ), which gives the canonical homeomorphism
1 1 2

n1 n2 n l+ n 2 ni n2 n i+n?
1R x hr ]R r also defines a homeomorphism 3R x ]R 3R
for n 1 > 0. Similarly, the formula ((x , .. .,x ) , (y , ...,y ))
n1 n2
ni n2 n i+n?
** (y-i /•••fY ,...,x ) defines a homeomorphism nR x ]R -* 1R
2 1 "

for n^ > 0 , and the formula ((x^,...,x^ ) , (y^,...fy )) ^

2 2
(-2 x^y^,x^ - y^,X2 ,...rxn ,y2 '#’’'yn ^ defines a homeomorphism

R1 n2 n l+ n 2
x ]R_ -+ nR_ for n 1 > 0, n 2 > 0. Thus, each of the products
 135

n ^1 n2 ^ n. +n^
1R_ x IR , JR x 3R_ , and IR x IR_ is homeomorphic to HR- . We
conclude that given locally Euclidean spaces and X^ of dimensions
n^ and n2 , the product X^ x X^ is an (n^+n2)-dimensional locally
Euclidean space. In general, the product X^ x ... x X g of arbitrary
locally Euclidean soaces of dimensions n., ...,n is an (n.+...+n )-
" 1 s 1 s
dimensional locally Euclidean space.

4. The discussion in 2 raises two nontrivial questions.

The first one is whether a nonempty topological space can be
a locally Euclidean space of dimension n and, simultaneously, a
locally Euclidean space of a different dimension n' : n f n1 ? In
Chapter 4 this question is answered negatively (see 4.6.5.10). The
answer is obvious when n = 1, n1 > 1 or n ’ =1, n > 1. In fact,
any connected subset of a one-dimensional locally Euclidean space
becomes disconnected after one removes two suitably choosen points (for
example, two points belonging to an open subset which is homeomorphic
to IR ); in contrast, every nonempty locally Euclidean space of
dimension n' > 1
contains a nonempty open subset which cannot be dis-
connected by removing two points (any open subset homeomorphic to IRn ^
has this property). The picture is cristal clear when n = 0 or
n 1 =0. However, when n > 1, n ’ > 1, the proof requires a technique
which we will develop only later.
The second question is whether we can formulate more efficient
definitions ofthe interior and boundary points, which would permit us
to actually recognize them. For example, consider the half space IR^.
At this point we can show only the trivial inclusion 3IR^ c ir1^ (see
2 ) , and we are forced to settle for one of the extreme equalities
9IR^ = IR1? 1 or 3IR^ = 0
(which obviously are the only possible ones) .
n -n— 1
We shall prove in Chapter 4 that 3lR_ = IR^ (see 4.6.5.12). This
equality is plain for n = 1 (assuming that the point 0 has a
1 1
neighborhood in IR_ which is homeomorphic to IR , then by removing
0 we would disconnect this neighborhood; this is absurd, because the
1
latter cannot happen to a connected neighborhood of 0 in IR_). But
for n > 1, we again need techniques which are to be developed. The
^ •>!
equality 3IR_ = IR^ settles satisfactorily the general problem of
recognizing the interior and boundary points too. Indeed, it follows
that a point x of the n-dimensional locally Euclidean space X which
has a neighborhood U with a homeomorphism U IR^, is. a boundary
point of U, and hence, of X, if and only ifthishomeomorphism
takes into a point of the hyperplane IR1? 1 .
x For Dn this theorem
n n—1
asserts that 3D = S . Finally, we note that the alternative
equality 3IR^ = 0 would obviously imply that 9X = 0 for any
 136

n-dimensional locally Euclidean space.

5. In general, it would be more prudent not to use the

theorems formulated in 4, i.e., the theorem on dimensions and the
equality = IR^ \ as long as they have not been proven. This
indeed is the way we shall deal with the theorem on dimensions - the
only exception is a harmless remark in 2.3. However, we have already
used the equality 33R^ = 3R^ \ and we will take advantage of it again,
before its proof, in 7 and in 2.6, 2.7. But these are the only
instances where these theorems and their corollaries will be used
before their proofs.

6. The boundary of an n-dimensional locally Euclidean space

is an (n-1)-dimensional locally Euclidean space without boundary.

Let x be a boundary point of the locally Euclidean space X

and let U be a neighborhood of x, homeomorphic to IR^. Then the
boundary 9U is a neighborhood of x in 3X, since 3U = U fl 3X,
and is homeomorphic to HRn ^ , since 8 3R^ = 3R^? ^ (here the reference
n n 1
to Chapter 4 for the equality 8nR_ = nR^ is unnecessary: the
alternative dlR = 0 is excluded, because 3X ^ 0).

7. For any locally Euclidean spaces X ^ ,.

'x s
int(X. x ... x X ) = int X. x ... x intX
1 s 1 s
and

9 (X, x ... x x ) = OX, x x o x. ..x x ) U . . . U (X x . ..x X x3X ) .

1 s 1 2 s 1 s- 1 s

It is enough to prove the statement for s = 2. Let x. E X .

i i
and let cp^ be a homeomorphism of a neighborhood LL of onto
n. n.
JR or HR_ , i = 1,2. Then cp^ x ^ is a homeomorphism of the
neighborhood x U2 of the point (x^x^ onto one of theproducts

n1 n2 n1 n2 nl n2 n1 n?
3R x IR , HR x HR_ , IR_ x IR or 3R_ x JR_ ,and composing it
withone of the homeomorphisms exhibited in 3, we obtain a homeomorphism
n 1+n 2 n 1+n2
of x u2 onto !IR or 3R_ . We denote this composition by
nl
cp and analyse the four possible cases. If cp^ (U^ ) = IR and ^2^U2^ =
n2 n 1+n2
- IR , then cp(U^ x U2> - 3R and x^ , x 2 , and (x^ are aH

nl n?
interior points. If cp1 (U1 ) = 3R and <P2 ^U2* = 3R- ' then
n l+ n 2 n i+ n ? “ 1
cp(U^ x U2> = 3R_ , and cp(x^ ,x2l £ 3R^ if and only if
n2 _1
^ 2 ^x 2^ ^ ^1 “ Thus (x ,x ) is an interior (boundary) point if and
 137

only if x2 is an interior (respectively, boundary) point, while

is an interior point. Similarly, if cp^U^ = ]R_ and =^ '

then (x^,x2) is an interior or boundary point simultaneously with x^,

ni
while x2 is an interior point. Finally, if ip^ (U^ ) = ]R_ and
n n +n - 1
(P2 (U2) = 3R_ , then tp(U1 x v ) =®1 z if and only if

n i" 1 n2 _ 1
cp^ix^) € ]R1 or cp2 (x2> € 3R.j . That is to say, (x1 ,x2) is a
boundary point if and only if at least one of the points ,x2 is
boundary. Our conclusion is that in all cases (x^/X^) is an interior
(boundary point) if x^ and x^ areinterior points (respectively,
if x^ or x^ are boundary points).

8. A locally Euclidean space is connected if and only if its

interior is connected.

This condition is obviously sufficient. Now let us show that

it is also necessary. Let X be a connectedlocally Euclidean space,
let A be a component ofint X , and let Bbe the union of the
remaining components. Since the closed sets Cl A and ClB cover X
and X is connected, C IA f) Cl B ^ 0 whenever B f 0, and obviously
Cl A n Cl B c 3X. Let x G CIA (1 C1B , and let U be a neighborhood
of x homeomorphic to ]R^.Since int is connected, its homeo-
morphic image U n int X = intU is also connected, which is impossible
if B ^ 0. Consequently, B = 0 and intX is connected.

9. The topological space X Uin .3x->XX constructed from a

locally Euclidean space X is called the double of X and is denoted
by dopp X . The double of an n-dimensional locally Euclidean space is
an n-dimensional locally Euclidean space without boundary.
From now on, we shall identify imm^ (X) with X and we shall
denote the map ab imm2 : X+ imm2 (X) by cop, and imm2 (X) - by copX.
Note that X and cop X are closed in dopp X .

Manifolds

10. A locally Euclidean space is called a topological

manifold or, briefly, a manifold if it is a Hausdorff topological space
with countable base. A manifold is closed if it is compact and has no
boundaryf and open if it has no compact components.
Comparing what was said in 2, 3, 6 , and 9 with the
 138

corresponding properties of Hausdorff, second countable, and compact

spaces we see that: every open subset of an n-dimensional manifold is
an n-dimensional manifold; the interior of an n-dimensional manifold
is an n-dimensional manifold without boundary; the boundary of an
n-dimensional manifold is an (n-1 )-dimensional manifold without boundary;
the boundary of a compact manifold is a closed manifold; the product
of s manifolds of dimensions n ,...,n is an (n +...+n )-dimensional
I S I S
manifold; the double of an n-dimensional manifold is an n-dimensional
manifold without boundary; and the double of a compact manifold is a
closed manifold.
Since the components of a manifold constitute an open cover,
their number is finite in the compact case and countable in general
(see 1.1 .6.5) .
Clearly HRn , H R S n , and Dn , which we gave above as
examples of locally Euclidean spaces, are manifolds.

11. Manifolds are locally compact.

PROOF. Let x be a point of an n-dimensional manifold X.

Fix a homeomorphism cp of a neighborhood U of x onto 3Rn or IR^,
and a neighborhood V of cp(x) in cp (U) with compact closure CIV .
-1
Let U' = cp (V). Then obviously U* is a neighborhood of x in X,
-1 -1 -1
U* cz cp (CIV) , and cp (Cl V) is compact. Thus cp (Cl V) is closed
and contains both the neighborhood U1 and its closure Cl U ' . We
conclude that Cl U' is compact.

12. Manifolds are metrizable.

PROOF. Every locally compact Hausdorff space is regular

(see 1.1.7.23), and every regular second countable space is metrizable
(see 1 . 1 .6.9) .

13. The following example shows that for n ^ 1 there are

n-dimensional locally Euclidean spaces with countable base which are
not Hausdorff. Consider X = nRn IT IRn , where i = [in: JRn \ ]R^ IRn ] .
Then X is obviously an n-dimensional, second countable, locally
Euclidean space, but for n } 1 and x e IR^ - 1 , any two neighborhoods
of the points imrn.j (x) and imm2 (x) in X intersect.

14 (INFORMATION). For n 5 1 there are connected, Hausdorff,

n-dimensional locally Euclidean spaces that are not second countable.
A two-dimensional example can be found in [5], and a one-dimensional
one - in [11], p. 164 (the transfinite line, or "Alexandrov's line").
Higher-dimensional examples can be constructed from these by taking
direct products with Euclidean spaces.
 139

One-dimensional Manifolds

15. A zero-dimensional connected manifold obviously reduce

to a point. Theorems 17 and 19 below provide the topological
classification of connected one-dimensional manifolds. The two-dimen-
sional case will be analyzed in § 5 (see 5.3). The topological
classification of manifolds of higher dimensions is a very difficult
problem.

16 (LEMMA).If a connected Hausdorff space be X can

i
represented as the union of two open subsets homeomorphic to HR , then
1 1
X is homeomorphic to either HR or S .

PROOF. Let X = UU V be the above representation and

1 1
cp: U + 1R , ^ : V + IR - the corresponding homeomorphisms. We exclude
the trivial cases U <= V and V <= U, where X ishomeomorphic to ]R^ ,
and examine the sets cp(U fl V) and ip(LJ D V) .
Since the intersection U fl V is open in both U and V,
1
cp(U D V) and ^ (U fl V) are open in HR and their components are
intervals. None of these intervals is bounded: indeed, suppose that
-1
cp(U fl V) contains a bounded interval (a,b) . Then cp ((a,b) ) is both
closed in V (as the intersection of the compact,and hence closed set
-1
cp ([a,b]) with V) andopen in V, which implies V = cp-1 ((a,b)) <= U;
contradiction. Moreover, cp(U fl V) / 1r\ because, ifnot, U c= V.
Similarly, (U fl V) ^ JR^ . Finally, weare left with only two possible
cases: (i) each of the sets cp(U flV) and i|MU flV) is an open half
line; (ii) each of the sets cp(U D V) and i|j (U ilV) is the union of
two disjoint open half lines.
Since we may multiply both cp and ip by -1 , we may assume
that, in case (i), cp(U fl V) has the form (-oo,a) , and (U fl V) -
the form (b,°°). Consider the composition

(-°°, a ) = <p(u n V) —>u n V iMU n V) = (b,°°).

This map is injective and continuous, and hence monotone and obviously
-1 -1
increasing (if it were to decrease, the points cp (a) and ip (b)
would haveno disjoint heighborhoods in X). Thus,

X = i|T1 ((-oo,^ (xQ)]) U tp 1 ([(p(xQ) ,°°) ) ,

for somepoint X q £ U 0 V, and so in case (i) X is homeomorphic to

 140

]R1 .
In case (ii), <P(U fl V) = (^>°,a^) U (a2 '00) an<^
i|j(U fl V) = (-0°,^) U ( b 2 ,°o) , for some a ^ a ^ ^ ^ (a1 < a2 , b 1 < b2 >,
and we may assume that the composite homeomorphism

ip(U n V) -- ► U n V Ip (U n V)

maps (-^a^) onto (b2 ,°°), and (a2 ,°o) onto (-°°,b,j). Both functions
(-0° ^ ) + (b2/») and (a2 /°°) + (-oo,^), which represent compressions
of this composite homeomorphism, are increasing (if, for example, the
-1 -1
first were to decrease, then the points cp (a^) and ip (b2) would
have no disjoint neighborhoods in X) . We can thuswrite

X = if; ” 1 ([ip (x2) ,\p (x1 )] ) U cp 1 ([cp (x1 ) ,cp (x2 )] ),

—1 —1 ”1
with some points x^ £ cp ((-°°/a^)) = ip ((b2 r°°) ) and x 2 E cp ((a2 ,«>))
= ip~^ ((-00,b^)) . Therefore, in case (ii) X is homeomorphic to S1 .

17. Every compact, connected, one-dimensional manifold is

homeomorphic to either S1 or D1 .

PROOF. For a start, assume that the given manifold is closed.

Then it can be covered by a finite number of open subsets homeomorphic
1
to 1R , and we may arrange these in a sequence U^,...,Us such that
each union V, = U 1 U ... U U, is connected. According to Lemma 16,
K I -rC -j
the first of the sets V 1,...,V not homeomorphic to IR is homeo-
1
1 i s
morphic to S , and being both open and closed, it is the entire
manifold, which is thus homeomorphic to S1 .
Assume now that the manifold has a boundary. Then its double
is a closed, connected, one-dimensional manifold, and as such is homeo-
1
morphic to S . Therefore, the original manifold is homeomorphic to a
1
subset of S . Since this subset is connected, closed, nonempty,
1
different from S , and not reduced to a point, it is homeomorphic to
D1 .

18 (LEMMA). If a topological space X can be represented

as the union of a nondecreasing sequence of open subsets, all homeo­
morphic to IR1 , then X is homeomorphic to IR1 .

PROOF. Let X = U be the given representation. Clearly,

any homeomorphism of onto some interval (a,b) extends to a
homeomorphism of V j_+-| onto one of the intervals (a,b),(a- 1 ,b) ,
(a,b+1), or (a-1,b+1). Hence one can construct inductively a sequence
of intervals and a sequence of homeomorphisms A ^,
cp2 : V2 ^ A2 ,..., suchthat ^ = abcpi+ 1 . The map of X onto the
interval U Ai , which agrees with on v i , is obviously a homeo-
 141

morphism.

19. Every noncompact, connected, one-dimensional manifold

is homeomoiphic to either IR or IR .

PROOF. First, assume that the given manifold X has no

boundary. Then X can be covered by a countable family of open subsets,
all homeomorphic to IR , and we can arrange these in a sequence
^1'^2'#,‘' such that all unions U ... U are connected. Then all
these unions are homeomorphic to HR1 . Indeed, if not, the first of them
1 1
not homeomorphic to IR is, according to Lemma 16, homeomorphic to S ,
and being open and closed must coincide with X; contradiction.
Therefore, one can apply Lemma 18 to our manifold and deduce that it is
homeomorphic to IR .
Now assume that X has a boundary. Then dopp X is a non­
compact, connected, one-dimensional, boundaryless manifold, and must be
1
homeomorphic to IR . It follows that X is homeomorphic to a connected,
1
closed, noncompact subset of IR , different from IR1 ; as such, it is
1
homeomorphic to IR_ .

2. Differentiable Structures

1.
Recall that a real function defined on an open subset of
n r r
IR is of class C (or a C -function) if it has continuous partial
derivatives of all orders up to and including r. The definition implies
that 0 < r <°°, that Cis the class of all continuous functions,
CO
and that C is the class of all functions which have continuous partial
derivatives of all orders. In addition, we say that the real analytic
functions are of class Ca (orCa-functions). It isconvenient to
consider a > 00 and thusencompass all theclasseslisted above by the
inequality 0 < r £ a.
Obviously, these definitions can be extended to real functions
defined on an open subset of the half space IR^- To do this, we consider
the derivatives with respect to the first coordinate at the points of the
boundary hyperplane IR^ to be left derivatives, and analyticity at such
points is understood as the existence of an analytic continuation to an
open set in IRn . We further extend the definitions to maps of an open
subset of IRn or of IR^, into any subset of IRq : such a map is of
•p

class C , or, simply, a C -map, if its coordinate functions are of

class Cr .
2. A map f of an open subset of IRn or IR^ into an open
 142

subset of 3R^ or jp F is a diffeomorphism if it is invertible and both

f and f_1 are of class C 1. Two sets which can be transformed into
each other by a diffeomorphism are said to be diffeomorphic.
The following facts are contained in well-known theorems of
calculus:
(i) If an open subset of3RP or 3RP is diffeomorphic to
an opensubset of 3Rn or IR^, then p = n.

(ii) An open subset of 3R^ which is diffeomorphic to an

open subset of 3Rn is open in 3Rn .
(iii)
A diffeomorphism which is the inverse of a diffeo-
r r
morphism of class C is itself of class C .

r r
C -structures and C -spaces

3. The definitions below refer to a given set X.

A chartof dimension n on X is an invertible map of a
subset ofX onto an open subset of IRn or 3R^. The domain of a chart
cp is called the support of cp and is denoted by supp ip.
Two charts, (p and ip, are C -compatible (or have a
Cr-overlap) (0 ^ r ^ a) if the set cp(suppcp n supp is open in
Im <p , the set ^(supptp n suppijj) is open in ImiJ; , and the maps

ab “ 1 ab ib
cp(suppcp n supp ip) --- ^— > supp tp n supp ^ iMsuppcp il supp (j)
and
ab , - 1 ab
^(supptp n supp ip) — -— > supp cp n s u p p \p -— cpfsupptp n supp ij),

r r
which are inverses of one another, are of class C (i.e., C -diffeo-
morphisms for r > 1 and homeomorphisms for r = 0). This condition
is trivially satisfied whenever supp cp n supp iii = 0 . If
supp cp nsupp if;^ 0,then the C -compatibility of the charts tp and ip
implies the equality of their dimensions. In fact, this equality results
also from the C^-compatibility of cp and \p, as shown by 1.4.
A collection of charts is an n-dimensional Cr-atlas of the
set X
if these charts cover X, are n-dimensional, and each two of
r r r
them are C -compatible. Two C -atlases of X are C -equivalent if
their union is again a C -atlas. This is clearly an equivalence
relation, and the equivalence classes of n-dimensional Cr-atlases of
the set X are called n-dimensional Cr-structures. The Cr-structures
with r > 0 are called differentiable structures.
Clearly, if 0 £ q £ r, then each n-dimensional Cr-atlas is
 143

also an n-dimensional C^-atlas, and two equivalent Cr-atlases are also

Cq-equivalent. Thus, when 0 £ q £ r every n-dimensional - structure
uniquely extends to a C^-structure.
r
Every C -structure contains a maximal atlas, namely the union
of all its atlases. The latter is called the complete atlas of the
structure, and its charts are called the charts of the structure.
When we pasr. from a Cr“Structure to its C^-extension, the complete
atlas extends too.

4. A set endowed with an n-dimensional Cr-structure is called

an n-dimensional Cr-space. The charts and atlases of the structure are
refered to as the charts and atlases of the space. The complete atlas
27
of a C -space X is denoted by AtlX .
The coordinate functions of a chart cp of the Cr-space X
are called coordinates on supp cp or, alternatively, local coordinates
in X.
We shall denote by C^X the C^-space obtained from the
Cr-space X by extending its Cr-structure to a C^-structure, 0 £ q £ r.
The Cr-spaces with r ^ q are also termed spaces.
For examples of Cr-spaces we may look at all the open subsets
X of IRn or 1R^, with the Cr-structure defined by the atlas reduced
to the single chart id: X -> X. In particular, for any r, the charts
id IRn and id 3R^ transform 3Rn and 3R^ into n-dimensional Cr-spaces.

5. Every n-dimensional locally Euclidean space has an obvious

C^-structure: its complete atlas consists of all possible homeomorphisms
U U ', where U is an open subset of the space and U1 is an open
subset of IRn or JR^. On the other hand, applying the "union of
topological spaces" construction (see 1 .2 .4.3) to the complete atlas of
a given n-dimensional C^-space, we obtain an n-dimensional locally
Euclidean space, and this transition is the inverse of the previous one.
Therefore, C^-spaces are just locally Euclidean spaces.
Since any differentiable structure extends uniquely to a
0
C -structure, every Cr -space with r > 0 is also a locally Euclidean
space. Its topology may be described in a more direct fashion as the
topology of the union constructed from any atlas of the structure.

6. Obviously, every point of an n-dimensional C -space X

can be covered by a chart cp of X such that Im cp = IRn or IR^. The
points with Im cp = IRn are called interior points and form an open
dense set, called the interior of the space X, denoted by intX .
The remaining points are called boundary points and they form a closed
set, called the boundary of X, denoted by 9X. These notations are
 144

in agreement with those introduced at 1.2. In fact, when r - 0, the

previous and present definitions of the interior and boundary points
coincide.
When r > 0, we use Proposition 2 (ii) in order to recognize
the interior and boundary points. According to this proposition, when
r > 0 a point X ofan n-dimensional C -space is a boundary point if
and only if Imp c and cp(x) £ 1R^ 1, where cp is a chart on this
space with x £ supp cp . In particular, if we regard 1R^ as a Cr-space
with r > 0, then 3 ]R^ = ]R^ Recall that the corresponding statement
for r = 0 appeared in 1.4 and its proof was postponed until Chapter 4.
The above characterization of the boundary points shows that
the interior and the boundary of a - space do not change when we extend
its Cr-structure to a C^-structure, for any q £ r. In other words, for
0 £ q £ r, int(C^X) = intX and 3(C^X) =3X. We emphasize that the
equalities int(C^X) = intX and 3 (C^X) = 3X were proved by a refe­
rence to 1.4, i.e., they depended upon results from Chapter 4, whereas
the equalities int(C^X) = intX and 9 (C^X) = 9X for q > 0 need no
such reference.
Using the relation 3X = 3(C°X), we see that Theorem 1.8 is
valid for Cr-spaces with r > 0 too. That is to say, a Cr-space is
connected if and only if its interior is connected. However, this
27
C -variant of Theorem 1.8 can be proved by merely repeating the proof
of the original theorem, and therefore we can eliminate the reference
to Chapter 4.

7. Suppose A is an open subset of an n-dimensional

27
C -space X. Then thecharts of X whose supports are included in A
27
yield a C -atlas of the set A, and define an n-dimensional
Cr-structure on A. In this way, any open subset of an n-dimensional
r r
C -space is an n-dimensional C -space. In particular, the interior of
r r
any n-dimensional C -space is an n-dimensional C -space without boundary.
Moreover, the interior and the boundary of an open subset U of the
Cr-space X are obviously given by intU = U n intX and 3U = U n 3 X.
Suppose tp is a chart on an n-dimensional Cr-space X. Then
ab tp : 3X fl supp tp ->• cp(3X fl supp tp) is an (n-1 )-dimensional chart on
3X. In this way we may construct a Cr-atlas of the set 9X, and so
define a Cr-structure on 3 X. Thus, the boundary of an n-dimensional
r 2T
C -space is an (n-1)-dimensional C -space without boundary.
r 1 r?
If (P1 (tp2) is a chart on the C -space X1 (C -space yi )

n1 n1 n? no
such that Imcp^ = IR or ]R_ (respectively, Im ip^ = ]R or H ),
then the composition of x tp2 with one of the homeomorphisms
 145

1 2 n l+ n 2 n1 n2 n i+n?
]R x ]R ]R and 3R_ x nR -> HR_ , defined in 1.3, provides
an (n^+n2 )“dimensional chart on x X 2 - If = then the charts
constructed as above form a Cr-atlas of the set Xx X~, with
r = min(r 1 ,r2), and hence define a C -structure on X x X„. Thus the
rl
product of the n 1-dimensional C -space X1 and the n 2-dimensional
2
C -space X2 with 3X 2 = 0 is an (n^ +n2>-dimensional Cr-space, where
r is the smallest of
the numbers r. and r„.In general, theproduct
r .

of the n^-dimensional C -spaces X^ i = 1,...,s, such that at most

one of them has a boundary, is an (n^+...+ng)-dimensional Cr-space,
where r = min(r , .. .,r ). Moreover, int (X x ... x x ) = intX x ...
i s I s
x int X g , and if X^ is the only space having a boundary, then
3(X1 X ) = X,
x x ... x X. 1 x 9X. x x .
... x , x ... x x . For r = 0
1 s 1 1-1 i i+1 s
both formulas can be found in 1.7; for r > 0 , they are plain.
It is clear that when we extend the d °-structure of the
r cr r
C -space X to a C -structure, the C -structures induced on the open
subsets of X and on its boundary X also extend to C^-structures,
and that C^(X 1 x ... x X s ) = C^X.1 x ... x C^X s . In particular,
r r
the
27
topology defined by the above induced C -structures coincide with the
r-L
relative topology, and the product of the C -spaces X^, i = 1,...,s,
considered as a topological space, is just the product of the
topological spaces X ,...,X .
I s

Smooth Maps

>r ^r
8 . A continuous map f of a C -space X into a C -space
r r
Y isof class C , or a C -map, if for any chart cpon X and any
chart on Y, the composite map
- 1
-1 , ,x , ab cp ^ j-— 1 , lXabf , ib - .
cp (supp cp n f (supp ip )) supp cp n f (suppiji) >supp ip Im ijj
-

is of class Cr (see 1). Such a composite map is called a local

representative of the map f; we use the notation loc(cp,^)f.
r
Obviously, a map f: X Y of C -spaces is of class C if and only
if the local representatives of constructed for all the charts of f
27
of someatlases of X and Y are of class C .
Note that this general definition of C-maps contains the
definition given in 1. Now, as before, a C _map is just a continuous
1
map. The maps of class C are called smooth, and the maps of class
Ca - (real) analytic.
r r
The composition of two C -maps is obviously a C -map. If A
 146

XT • •
is an open subset or the boundary of a C-space X, then the inclusion
A ->■ X is a Cr-map. If A is open in X or A = 9x, and B is open
in Y or B = 9Y, then the compression A B of any C -map X -+ Y
nr
is a C -map.
^1 ^1
9. A map f of a C " -space X into a C -space Y is a
diffeomorphism if it is invertible and both f and f ^ are smooth.
The space Y is said to be diffeomorphic to the space X if there is
a diffeomorphism X -* Y, and Cr-diffeomorphic to X, if there is a
C -diffeomorphism X -> Y.
r
Of course, the identity map of a C " -space with r ^ 1 is a
Cr-diffeomorphism. Also, the composition of two Cr -diffeomorphisms is
r r
a C -diffeomorphism, and the inverse of a C -diffeomorphism is a
Cr-diffeomorphism itself, as we may easily see from 2 (iii). Therefore,
the property of being C -diffeomorphic is an equivalence relation.
Using 2 (i), we conclude that nonempty diffeomorphic spaces
have the same dimension.

10. Let
f . : X. -+ Y. ,. ..,f : X Y be Cr-maps with r ^ 1,
1 1 1 m m m ^
where no more than one of the spaces X^,...,Xm , andno more than one
of the spaces Y^,...,Y , has a boundary. Then

f A x ... x f : X. x ... x X -- ► Y. x ... x Y

1 m 1 m 1 m
■e
is obviously a C -map. If f^,...,f^ are diffeomorphisms, then so is
f x ... x f .
1 m
The canonical homeomorphism X^ x x^ X^ x x^ is a Cr-dif-
feomorphism for any two Cr-spaces X^ and X2 with r ^ 1, such that
one of them has no boundary. The canonical homeomorphisms
(X- x ... x x m . ) x Xm -+ X. x ... x X and X, x (Xn x . . . x x ) -+
1 m- 1 1 1
m_
r m r 2
„ m
X^ x ... x x^ are C -diffeomorphisms for any C -spaces X^ , .. .,X
with r ^ 1, such that no more that one of them has a boundary.

Subspaces

11. A subset A of an n-dimensional Cr-space X with r

is a k-dimensional subspace of X if each point of A is covered by
a chart <p on X such that the pair (Imip,cp(A n suppcp)) coincides
with one of the pairs (IRn ,]Rk ) , (]Rn ,]R^), or 0R^,nR^). If k > 0,
then this condition is obviously equivalent to the following one: each
point of A is covered by a chart cp of the space X such that
k k
(P(A D supp cp) = Im <p n 3R or Im cp D ]R_ .
Suppose A is a k-dimensional subspace of an n-dimensional
 147

C -space X, and consider the maps ab tp: A fl supp tp -+ tp(A fl supp tp)
corresponding to all charts tp on X such that tp(A D supp tp) =
k k
= Im cp 0 3R or Im cp fl 3R_. Each such map is a k-dimensional chart on
A, and together they yield a k-dimensional Cr-atlas of A. The
27
C -structure defined by this atlas transforms A into a k-dimensional
r r
C -space. The topology of this C -space obviously coincides with the
relative topology.
When we extend the Cr-structure of a space to a C^-structure
with q ^ 1, its subspaces remain subspaces. The subspaces of Cgx
q 2T
are called C -subspaces of the original C -space X.
The codimension of a subspace is the difference between the
dimension of the ambient space and that of the subspace.
27
It is evident that the open subsets of a C -space (with
r ^ 1 ) are among its subspaces; in particular, we cite its interior
27
and its components. Warning: if not empty, the boundary of a C -space
is not a subspace.
It is readily seen that if A is a subspace of codimension
0 of a Cr-space X, then int A = Int A fl int X .
27
A subspace of a C -space is neat if itis closed as a subset
and its boundary is contained in the boundary of the space. Note that
every neat subspace of codimension 0 is made of whole components of
the ambient space.

12* The definition of a subspace given in 11 contains

implicitly the generally used method of defining subspaces through
equations and inequalities. Namely, according to the second variant
27
of our definition, a subset A of an n-dimensional C -space X (r 5 1)
is a subspace of X of positive dimension k if and only if each point
of A has a neighborhood U with coordinates tp^ , ...,tpn , such that
the intersection A fl U is defined in U either by the equations
tpk + 1 (x) = 0 , . ..,tpn (x) = 0 , or by the equations tp^+ 1 (x) = 0 ,...,
tpn (x) = 0 and the inequality tp^ (x) £ 0 .
As a procedure for defining subspaces, this formulation has
an obvious disatvantage; namely, we must assume from the beginning that
cp^ , .. .,tPn are local coordinates. We make use the implicit function
theorem to make it more efficient.
Let X be a Cr-space (r ^ 1), xQ € X, and let f 1 ,...,fm
be real C 1-functions defined on a neighborhood Uq of Xq . We say that
f 1 ,...,fm are independent at the point xQ if there is a chart tp on
X with x„ £ supp tp and such that the functions
<J _1
g 1 ,. . .,gm: n ® defined by gi (y) = f± (<P (y) ) have
linearly independent gradients at the point tp(xQ ). The following
 148

statements are consequences of the implicit function theorem:

(i) if the Cr-functions f ,...,fm are independent at the
interior point X q of the Cr-space X, then they can becompleted to
a system of coordinates ina neighborhood of x^;
(ii) if the Cr-functions f^,...,fm are independent at the
boundary point xQ of the C -space X, and the function f^ is
negative at the interior points and zero on the boundary, then
f^,...,fm can be completed to a system of coordinates in a neighborhood
of Xq having f^ as the first coordinate.
Comparing with the previous, coordinate description of the
r
subspaces of a C -space, we see that a subset A of the n-dimensional
C -space X is a subspace of positive dimension k of X if:
(i) each point x^ E A which is an interior point for X
has a neighborhood U where theintersection A D U is defined either
by the equations cp, * (x) = 0 ,...,cp (x) = 0 , where cp, , .. .,cp are
^ n K^ I n
C -functions independent at x^, or by the equations cp^+^ (x) = 0 ,...,
cp (x) = 0 and the inequality cp (x) £ 0 , where cp1 ,cp , ...,cp are
n T I Jc+ 1 n
C -functions independent at x^;
(ii) each point x^ G A which lies on the boundary of X
has a neighborhood U such that the intersections intX DU, 3x D U,
and A n U are defined in U by the inequality cp^ (x) < 0 , by the
equation cpn (x) = 0 , and by the equations cp, . (x) = 0 ,...,cp (x) = 0 ,
i r n
respectively, where cp^ + r •••r^>n are C -functions independent at
x0.

13. An obvious consequence of the definition of a subspac

is that the interior intA of a subspace A of the Cr-space X is
contained in int X ,and that the intersections 3A fl int X and
3A D 3X are open in 3A, i.e., they consist of whole components of
3A. Moreover, it is clearthat intA is a subspace of both X and
int X, v^iile 3A D int X is a subspace of X. In particular, if A is
a neat subspace of X, then 3A = A n 3X and intA is a neat sub­
space of intX , while 3A is a neat subspace of 3X.
The inclusion A X of a subspace A of a Cr-space X is
obviously a Cr-map. The compression A -> B of any Cg-map X -+ Y,
where A (B) isa subspace ofX (respectively, Y) is a Cq -map.
Let Ai be a subspace of the Cr-space X± , i = 1,...,s,
and assume that at most one of the spaces X ^ ,...,X has a boundary,
the same being true for the subspaces A.,...,A . Then A x ... x a
I s 1 s
is a subspace of X^ x ... x x g , and is a neat subspace if each A.
is neat. For example, the fibers of the product X„ x ... x x are
1 s
neat subspaces.
 149

Let A be a subspace of the Cr-space X, and let B be a

subspace of A. Then, using the description of subspaces in 12, we see
that B is a subspace of X; in particular, if B is a neat subspace
of A and A is a neat subspace of X, then B is also a neat sub­
space of X. In a similar fashion we conclude that a neat subspace B
27
of the C -space X which is contained in a neat subspace A of X is
also a neat subspace of A.

27
C -manifolds

r r T
14. AC -space is a C -manifold, or a manifold of class C ,
if it is a topological manifold, i.e., a second-countable, Hausdorff
space. A manifold of class C° is simply a topological manifold; see 5.
The manifolds of class Cr with r ^ 1 arecalled smooth, or
differentiable. The manifolds of class Ca are called (real) analytic.
27
Since any C -structure is defined by one of its atlases, it
is interesting to discover those properties of an atlas of a Cr-space
which guarantee that the space is Hausdorff and second-countable. Here
we formulate only two obvious conditions: if each pair of points is
covered by a chart of the atlas or by two disjoint charts of the atlas,
then the space is Hausdorff; if the atlas is countable, then the space
is second-countable.
There is no need to check that a space is Hausdorff and
second-countable if the differentiable structure is introduced on a set
which is already a topological manifold, and the topology defined by
the differentiable structure coincides with the initial one. If the
differentiable structure is defined by an atlas then the two
topologies agree if and only if supp cp^ are open and cp^ are homeo-
morphisms; see 5.
A smooth manifold is closed if it is compact and has no
boundary; cf. 1.10. Warning: at the present time the equality
3X = 3(C°X) is not proven (see 6 ). Therefore, we should be careful
to distinguish between the smooth manifold X being closed and the
topological manifold C^X being closed.

16. Reconsidering the statements made in 7 and 11 in the

light of the corresponding properties of Hausdorff spaces, second-
countable spaces, and compact spaces, we may deduce the following:
every open subset of an n-dimensional C -manifold is an n-dimensional
Cr-manifold; the interior of an n-dimensional Cr-manifold is an
n-dimensional Cr-manifold without boundary; the boundary of an
 150

n-dimensional Cr-manifold is an (n-1)-dimensional Cr-manifold without

r r
boundary; the boundary of a compact C -manifold is a closed C -manifold;
the product of the Cr-manifolds X I,...,X
S of dimensions n 1I,...,n
S ,
such that no more than one of them has a boundary, is an (n^+...+ng )-
r r r
dimensional C -manifold; a subspace of a C -manifold is a C -manifold.
The subspaces of smooth manifolds are called submanifolds,
and the neat subspaces - neat submanifolds.
27
16. The basic examples of n-dimensional C -manifolds with
r ^ 1 are again lRn and (see 4) . The submanifolds of these
spaces provide an unlimited supply of examples of C -manifolds. The
simplest are the submanifolds of nRn
defined by one system of equations
r
cpk +i<x) = 0 ,...,tpn (x) = 0 , where ^ +1 ' •••'^n are ^ -functions
defined on an open subset of 3Rn and having linearly independent
gradients on the set of their common zeros. We may add to such a system
27
the inequality tp^ (x) < 0, where cp^ is any C -function defined in a
neighborhood of the set of common zeros of cp^+ ^ , ...,cpn / zero on this
set, and such that the gradients of cp^ ,cp^+^ ,.. .,cp^ are linearly
independent on the same set. For example, the sphere Sn ^ is defined,
2 2
in standard coordinates, by the equation x+ ... + x - 1 = 0 , and
n 2 2 n 1
the ball D - by the inequality x^ + ... + x^ - 1 £ 0. Hence Sn

and Dn are submanifolds of 3R n

and, in particular, Ca-manifolds.
k v
The following facts are clear: 1R , for k £ n, and S ,
n k k
for k < n, are neat submanifolds of 1R ; IR_ and D are not neat
submanifolds of IRn for k £ n; is a neat submanifold of 1rP_
for k £ n; S is a neat submanifold of Sn for k £ n; and Dk is
a neat submanifold of Dn for k £ n.

17.
Finally, we note that every real, n-dimensional vector
27"
space has a natural C -structure for any r (0 < r £ a ) , which makes
27*
it into an n-dimensional C -manifold. This structure is defined by
the linear charts i.e., by the linear maps onto IRn .

3. Orientations

1. Consider a Cr-manifold X, and let CatlX denote th

atlas of X consisting of all the charts with connected support. If
cp and ip are two charts on the smooth manifold X, we denote by
J(cp,i|;) the Jacobian of the map loc (cp,¡p)id, i.e., of the composite
map
- 1
cp(suppcp n supp ip ) — „ supp cp fl supp i> <p (supp cp n supptj;).
 151

An orientation of the smooth manifold X is a function

to : Ca tl X -> S such that to (tp) = [sign J(cp/i^)](y)co(^) for each two
charts £ Catl X , where y is an arbitrary point of
cp(supptp D supp \p) [therefore, the function sign J (tp,if;) must be
constant on cp(suppcp f) supp i|j) ] . A smooth manifold endowed with an
orientation is said to be oriented. A smooth manifold which can be
oriented is orientable.
Clearly, an orientation CatlX -+ is determined by its
values on any subatlas of CatlX . Moreover, every function which
carries a subatlas of CatlX into S° and satisfies the previous
compatibility condition, i.e., its values on two charts tp,\fj of the
given subatlas are obtained one from another multiplying by
[sign J (<p, ip) ] (y) with y £ cp(supp cp H supp if;) , extends to an
orientation Catl X -* S^.
When one extends the Cr-structure of a Cr-manifold X to a
Cq structure with q < r, CatlX becomes a subatlas of Catl Cqx . If
q ^ 1, this establishes a one-to-one correspondence between the
orientations of the manifolds X and C^X.

2. For each orientation to: Catl X -+ S° there exists the

opposite orientation -to, and thus every nonempty orientable manifold
has at least two orientations.
Given two arbitary orientations of the manifold X, the set
covered by the charts on which they agree, and the set covered by the
charts on which they do not agree are open, disjoint, and together they
cover X; hence, each of them is a union of components of the manifold
X. Consequently, the orientation of a connected manifold is uniquely
determined by its value at one chart. Moreover, every smooth, connected,
orientable manifold has exactly two orientations, whereas an orientable
s
smooth manifold with s components has 2 orientations.
The standard way of describing an orientation of a smooth
connected manifold is to indicate the charts where it is positive.
For example, JRn has a natural orientation, positive on the chart
id 3Rn .
3. Since any point of a zero-dimensional manifold X is
covered by only one chart of CatlX , it follows that every zero­
dimensional manifold is orientable, and has actually a natural
orientation, identically equal to +1. We shall see later (in 5.3.1
and 5 .6 .3 .4 ) that all one-dimensional manifolds are orientable,
whereas the manifolds of dimension ^ 2 are not necessarily so.
In Chapter 5 we shall give effective sufficient conditions
for the orientability of a manifold of arbitrary dimension (see 5.6.3).
 152

The crudest of them is that the manifold be simply connected.

4. Let A be an open subset of a smooth manifold X. The

CatlA c CatlX , and every orientation of X induces an orientation
of A; in particular, A is orientable whenever X is. If A
intersects all the components of X, then the orientation of X is
determined by the orientation induced on the submanifold A.
In the case A = intX , we may say more: not only does each
orientation of X restrict to an orientation of the manifold int X ,
but also each orientation of intX extends to an orientation of X.
Indeed, for each connected subset U of X, U D int X is connected
(see 2.6 and 2.7) and so the compression ab cp : supp cp fl int X
cp(suppcp D int X) belongs to Catl (int X) for any cp E Catl X . This
enables us to extend any orientation go : Catl (int X) -*• toan
orientation Catl X through the formula cp u(abcp). Thus, if we
associate to each orientation Catl X -* its restriction
Catl(int X) -* S , we obtain a one-to-one correspondence between the
orientations of the manifolds X and intX . In particular, X is
orientable whenever intX is such.
According to 2.7, for each chart cp on the smooth manifold
X we have the corresponding chart ab cp : supp cp n 9X cp(supp cp fl9X)
on its boundary 3X. It is clear that every chart in Catl 9X is of
the form ab cp , where cp £ Catl X , and that for each orientation
w: CatlX S^ of the manifold X we have an orientation of its
boundary, defined by the rule ab cp v* w (cp) . In particular, 8X is
orientable whenever X is such. If all the components of X have a
boundary, then the orientation of X is uniquely determined by the
orientation it induces on 3X.
The only submanifolds of a smooth manifold X which inherit
a natural orientation when X is oriented are those of codimension
zero. This orientation has been implied in the discussion of the open
subsets of X. If A is an arbitrary submanifold of X of codimension
zero, then the induced orientation is defined by the orientation of its
interior intA , which is open in X. In particular, A is orientable
if X is such. If A intersects all the components of X, then the
orientation of X is uniquely determined by the orientation it induces
on A.
Since the manifold nRn has a natural orientation, all its
n-dimensional submanifolds inherit a natural orientation too. In
particular, Dn carries a natural a natural orientation, which in turn
induces an orientation of its boundary Sn”1. Warning: the orientation
0 1
of S induced by the orientation of D does not coincide with the
 153

canonical orientation of , considered as a zero-dimensional manifold

(see 3).

Orientations and Diffeomorphisms

5. Every diffeomorphism f : X -> Y establishes a one-to-one

correspondence between the orientations of the manifolds X and Y.
If both manifolds are oriented and f transforms the orientation of X
into the orientation of Y (into the opposite orientation of Y),
then we say that f preserves (reverses) the orientation or that f
is orientation preserving (respectively, orientation reversing).
To determine whether a diffeomorphism f: X Y is
orientation preserving or not, we may look at its local representatives;
if X and Y are connected, it is enough to analyze only one local
representative. Namely, suppose that m and w are orientations of
X Y
the connected manifolds X and Y. Let <p € Catl X and £ Catl Y
-1
be two charts and pick x G supp cp 0 f (supp ^) . If the sign of the
Jacobian of loc(cp,ij;)f at the point cp(x) coincides with (is opposite
to) the sign of the product w (tp) oa (^) , then f is clearly orientation
X Y
preserving (respectively, reversing).

6. Of special interest is the case where X = Y is a

connected manifold. It is readily seen that in this situation a dif-
feomorphism which preserves (reverses) one orientation of X, will
preserve (respectively, reverse) all orientations of X. Therefore,
in this case one can talk about an orientation preserving (reversing)
diffeomorphism without fixing an orientation. In particular, every
(auto)diffeomorphism of a smooth, connected, orientable manifold is
either orientation preserving or orientation reversing.
As an example, consider a nonsingular linear transformation
f: ]Rn ->]Rn . It is clear that f is a diffeomorphism and that f is
orientation preserving (reversing) if detf > 0 (respectively,
detf < 0). If the transformation f is orthogonal, then its
compressions ab f : Dn -> Dn and ab f : Sn 1 Sn 1 are meaningful,
are diffeomorphisms, and preserve (reverse) orientation if det f = 1
0 1
(respectively, detf = -1) [here S is oriented as BD ]. If f is
given by f(x) = -x,
then detf = (-1)n . Hence this diffeomorphism
n“1 11“ 1
and the induced antipodal map ab f : S S are orientation
preserving for n even and orientation reversing for n odd.
 154

Orientations and Products of Manifolds

7. We have shown in 2.7 that every product x ... x ^

of charts cpw
I •■s•/<p on the smooth manifolds X 1f...,X
I S without
boundary of dimensions n l M .,nn , may be regarded, using the canonical
l s
n1 n n. + ...+n
identification 3R x ... x hr s = 3R [defined by the formula
((x11 ,x1n (xs1 ,xgn )) H- (x1 1 , ...,x1n ,...,xs1 , ...,xgn ) ] ,
1 s I s
as a chart on the product X 1 x ... x X . Given orientations
I s
of the manifolds , let us define a mapping from
i s I S
thecollection of all charts tp x ... x (p with <p. E Catl X y .. .,
0 s i i
<p G Catl Xg , into S by the formula

cp1 x ... x (p^ h- o)1 (cp ) ... <Dg (cpg) .

It isobvious that the above collection of chartsis a subatlas of

Catl(X^ x ... x x g), and that this mapping satisfies the compatibility
condition introduced in 1. Therefore, it extends to an orientation of
the manifold X^ x ... x X g , called the product of the orientations
,. ..,(jL)g . The latter can be defined even if one of the manifolds
X,j,...,Xs has a boundary: it is the orientation induced by the
orientation of the interior intX^ x ... x int X g . Thus, a product of
smooth, oriented manifolds (such that no more than one of them has a
boundary) is oriented, and a product of smooth, orientable manifolds
is orientable.
We note that the orientability of X^ x ... x Xg implies the
orientability of each factor X 1 , ...,X . In fact, if co is an
1 s
orientation of the product and cp^ , ... + / ••• are fixed
charts of Catl (int X^ ) , ...,Catl (int X^_^ ) ,Catl (int + ^) , .. .,

Catl (int X g) , then the mapping Catl (int X^) given by

(p h- a)((p^ x ... x (p^_<| x <p x cp^+ ^ x ... x cpg) is an orientation of the

manifold intX . .
l

8. Suppose X^ and X2 are smooth oriented manifolds of

dimensions n^ and n2, such that one of them has no boundary. The
canonical diffeomorphism X^ x x 2 -> X 2 x x^ preserves orientation if
n1 n 2 is even and reverses orientation if n '\n 2 is odd.

PROOF. Let cp^ E Catl (int X^) and cp2 E Catl (int X 2) be

n1 n2
charts with Im ^ = 1R and Im cp2 = 1R , respectively. Clearly,
 155

X ^2
^ Catl(int(X^ x x2)), cp2 x € Catl(int(X2 x x^ )) , and the
n +n n +n2
local representative IR -*IR of the canonicaldif feomorphism
X1 xX 2 X2 * X 1 relative to these charts is given by
(x1 .... Xn 1 +n 2> " (xn 1 +1' • • " xn 1 +n 2 'x 1 " - - ' xn 1) • Therefore, the
n1 n2
Jacobian of this local representative is (-1)

9. Suppose X^,...,XS are smooth oriented manifolds of

dimensions n 1 ,...fn _ / such that only one, say X,., has a boundary.
I S 1
Then the canonical orientation of the product

X- x ... x x. , x 3X. x X. „ x ... x X

1 1-1 l 1+1 s

and the natural orientation that this product receives as

n 1 + ...+n ._ 1
3(X^ x ... x xg) differ by a factor of (-1 )

PROOF. Obviously, the above orientations agree if i = 1.

The case i > 1 reduces to i = 1 with the aid of the diffeomorphisms

X. x ... x x -»-X. x X. x ... X X . x X . ,. x ... x X

1 s l 1 1-1 1+1 s
and
X 1 x ... x x i _ 1 x dXi xxi + 1 x ... x Xg -*•

9X. x x„ x ... x X. . x X. .x ... xX .

l 1 1-1 1+1 s
The first of them is the product of the canonical diffeomorphism

(X x ... x X±_1) x X. + X. x (X1 x ... x Xi_1)

with id(Xi+^ x ... x Xg), while the second is the product of the
canonical diffeomorphism
(X1 X ... X Xi_1) X 3Xi -> ax^^ X (X1 x ... X Xi _ 1 )

with id(X^+^ x ... x xs). The first diffeomorphism preserves

(reverses) the orientation if the product (n^+...+n^_^)n^ is even
(respectively, odd), while the second has the same property if the
product (n1 +...+n±_ 1 ) (ni~1) is even (respectively, odd); see 8 . This

n 1 + “ *+ni- 1
explains the factor (-1 )

Orientations of Vector Spaces

10. Real vector spaces are smooth manifolds (see 2.17) a

hence the definition of orientation given in 1 applies. On the other
 156

hand, there is a well-known, purely vectorial definition of an

orientation of a real vector space: an orientation is a mapping from
the set of all bases of the space into S°, which takes the same value
on two bases if and only if the matrix transforming one basis into the
other has a positive determinant. This vectorial definition clearly
agrees with the definition given in 1 and is often more convenient.
In particular, the vectorial definition makes obvious the
following remark. Let f be a linear map of a real vector space V
into another real vector space. Fix a subspace V' <= V that is mapped
isomorphically by f onto Im f . Then we may represent V as the
direct sum of the spaces Imf and Kerf . Consequently, the
orientations of any two of the three spaces V, Im f , and Ker f
determine the orientation of the third- As an immediate result of our
definition, we see that this connection between the orientations of V,
Im f , and Kerf does not depend upon the choice of V 1, but only on
the map f.

4. The Manifold of Tangent Vectors

1. Suppose that X is an n-dimensional C -manifold, r ^ 1.

For each point x £ X, let Atl^X be the collection of all charts
tp E Atl X such that x E supp ; recall that Atl X denotes the
complete atlas of X. If cp,ijj E Atl^X, then at the point tp(x) the
differential of the diffeomorphism

loc (cp,ijj) id: cp(supptp fl suppijj) -* i}j(supptp 0 supp il^)

is meaningful [and is the linear map IRn ]Rn whose matrix is the
Jacobi matrix of the map loc (tp,^) id at the point cp(x)]. We denote
this differential by d^icp,^).
Now consider the real vector space of all the maps
Atl X-> IRn (with the natural operations) . It is clear that those maps
v:Atl^X -* HR which satisfy the relation v(i¡j ) = d (cp,ij;)v (cp) for any
two charts cp,ip £ Atlxx yield a subspace of this vector space, i.e.,
they form a real vector space. The latter is called the tangent space
to X at the point x and is denoted by Tang X. The maps in Tang X
x x
are called tangent vectors at the point x (on X ) , or vectors
tangent to X at x .
Obviously, a tangent vector is completely defined byits
value on an arbitrary chart of Atl X, and for each chart <p £ Atl X
n x x
and each vector u £ 3R there is a vector v £ Tang X with v(cp) = u.
„ X
Therefore, the mapping tp# : TangxX + 3R , tp# (v) = v(ip), defined for a
 157

chart cp E Atl X, is invertible. Moreover, cp„ being linear, it is

an isomorphism; in particular, dim Tang X = n . The isomorphism cp '
^ n
takes the canonical basis ort^,...,ort^ of JR into a basis of
TangxX, which we shall call the cp-basis. The coordinates of a vector
v E Tang^X relative to the cp-basis coincide with the usual coordinates
of the vector v(cp) and are called the cp-coordinates of

2. We denote the union U Tang X, i.e., the space of all

xtx X
vectors tangent to X, by Tang X . The map Tang X + X transforming
Tang^X into x is called projection and is denoted by pr. Thus

pr 1 (x) = Tang^X.
The set Tang X has a natural topology whichmakes it a
topological manifold. Furthermore, for r ^ 2, Tang X has a natural
differentiable structure. In order to describe these structures, we
define, for each chart cp E Atl X , the map tn cp : pr 1 (suppcp) ->
+ Im cp x ]Rn , tncp(v) = (cp ° pr (v) ,cp^ (v) ) . If we identify JRU x JRU
with HR2n, and consider Im cp x nRn to be an open subset of lR^n or
]R2n, then we may interpret tn cp as a 2n-dimensional chart on TangX.
Obviously, if ]p E Atl X is another chart, the composition

cp(suppcp fl supp i|j) x ]Rn =

/
= tn cp (pr (suppcp)\ fl pr " 1 / ,
(suppijj)) ab(tncp) 1 >
----------

-1 z v -1 , -x ab tn :d
pr (suppcp) fl pr (suppy) -----

tn (pr 1 (suppcp) fl pr 1 (supp^)) = suppcp D suppi);) x ]Rn

is given by the formula (a,u) ^ (loc (cp,if;) id (a) ,d (cp,\J;)u).

cp (a)
Equivalently, writing this composite map in coordinates, we have
(a1 ,...,an ;u1 ,...,un ) ^ ( b ^ ,...,bn ;v 1 ,...,vn > , where

bj “ l j (al '•*•'an }

- Ii=1 Di l j (ai'•••'an )ui

and t ,...,£
are the coordinate functions of the map loc (cp,i[i) id
1n
(D. denotes the partial derivative with respect to the i-th coordinate)
l r— 1
Formula (2) shows that the charts tn cp and tnare C compatible
(we set C° °’ 1 = C°° and Ca _ 1 = Ca ) .
Moreover, the charts tn cp ,
r— 1
tp £ Atl X , cover Tang X , and thus yield a C-atlas ofthe set
Tang X . This atlas has a countable subatlas (since AtlX has such a
subatlas). Furthermore, for any two vectors of TangX, it has either
 158

a chart which contains both of them, or a pair of disjoint charts, each

containing one of the vectors (indeed, recall that Atl X contains,
for any two points of X, either a chart containing both of them, or
a pair of disjoint charts, each containing one of the points).
27— 1
Therefore, it makes Tang X into a 2n-dimensional C -manifold, which
we call the total manifold of vectors tangent to the manifold X.

Clearly, the projection Tang X -+X, the inclusions

Tang X Tang X , and the natural map X -+ Tang X which takes each
X — "1
point x into thezero vector of the space Tang^X, are all C maps.
Formula (2) shows that for r > 2 the Jacobian of the
composite map (1 ) at the point (a,u) is equal to the square of the
Jacobian of the map loc(cp,^)id at the point a. We deduce that for
r > 2 the manifold Tang X is always orientable and even carries a
canonical orientation, namely the one which is positive on the charts
tn tp with cp E Catl X .
One more remark: let X^ and X 2 be two arbitrary smooth
manifolds such that SX^ = 0. Then for any two points x^ E X^ and
x 9 E X , Tang. . (X x x«) and Tang X 1 © Tang X are
^ ^ \x ^ /x 2 ^^ ^2
isomorphic as vector spaces, and the isomorphism is natural. In
addition, the isomorphisms corresponding to all pairs (x^,x2) yield
a diffeomorphism of Tang(X^ x x2) onto Tang X^ x Tang X 2 .

The Differential of a Smooth Map

r >r
3. Let f
be a C -map of an m-dimensional C ' -manifold X
>£•
into an n-dimensional C ' -manifold Y, r s 1. For a point x £ X and
two charts tp £ Atl X and \p £ Atlf , .Y, we let d (f;ip,^) denote
x r \X ) x
the differential of the map loc(tp,i{j)f at the point tp(x), regarded
as the linear map 3R -*■ ]Rn whose matrix is the Jacobi matrix of
loc (cprvp)f at tp(x) . If tp' £ Atl X and \p'£ Atl... .Y are two
x r f(x)
other ch a rts , then dX(f ;tp' , ijj' ) = d -I. IXj
. (ij) ) ° d X (f;tp,^) ° dX (tp',tp).
Combining this relation with the equalitiesd (ip',<p) = tp„ o (tp■)— 1
x w w
1 _

and = ° ^ ' we see that

(4>p 1 o d x ( f ; tp ’ ) o ^ o dx (f;tp,i^) o tp ,

i.e., the linear map ijj” 1 o d^. (f ;tp,^ ) ° tp# : TangxX -»■ Tangf(x)Y does

not depend upon the choice of the charts tp and ip . This linear
map is called the differential of the map f atthe point x and is
 159

denoted by d^f. The map Tang X Tang Y which equals d f on

TangxX for all x € X is called the differential of the map f and is
denoted by d f . The resulting diagram

Tang X — =— > Tang Y

pr pr
X

is clearly commutative.
Let be the coordinate functions of the map
loc (cp, f. Then the local representative loc (tn cp,tn ^ )df of the
differential df is given in coordinates by the formula
(a 1 ' “ ‘'am;u i ' * *.'U^) ^ (b-j / •••fb^; v^ ,.. .,v ) , where

bj = V ai V '
j = 1 ,...,n.
v j = ^i = 1 Di'ej ( a 1 am )ui
r—1
Therefore, df is of class C
Of course, d(h o f) = dh ° df for any smooth map h of Y
into a third smooth manifold, and df = id(Tang X) when X = Y and
f = id X . Also, if f is a diffeomorphism, then df is a diffeo-
morphism for r ^ 2 and a homeomorphism for r = 1.
In the special case when X is an open subset of nRm or of
3R™, and Y is an opensubset of nRn or of IR^, in addition to the
differential d f: TangX + Tang,. ,.Y one has the classical
differential of the map f, i.e., the linear map IR + JR whose matrix
is the Jacobi matrix of f at the point x. Let us identify the spaces
Tang^X and 3Rm (Tan9f(x )Y anc^ v^a t^ie l^near isomorphism
(id X) ^ :Tang^X 3Rm (respectively, (id Y) ^ : T a n g ^ ^ Y + 3Rn ) . Then it
is clear that d^f becomes the classical differential.

4. If Ais a submanifold or the boundary of asmooth

manifold X, then the differential dX in: Tang
X A +X Tang X of the
inclusion in: A -> X is a monomorphism for each point x £ A . Thus,
we can identify the tangent space Tang^A with the subspace
d in (Tang A) of Tang X, and Tang A - with din (Tang A).
X X X
If A is a submanifold of 3Rn , then along with the
identification Tang A = d in(Tang A) one has the identification
X X x

Tangx ® n = lRn via the canonical (linear) isomorphism

(id 3Rn ) n : Tang IRn -* ]Rn , and so Tang A becomes a subspace of 3Rn .
tf x x
 160

It is easy to describe this subspace explicitly when A is defined in

a neighborhood of x by the independent functions ^ +1 ' *••'^n' as
2.12. In this situation, Tang A consists of all the vectors of lRn
which are orthogonal to the n-k vectors grad tp^.+ ^ (x) , ...,grad tp^ (x ) .

For example, Tang^S11 1 is the subspace of JRn composed of all

vectors orthogonal to the vector x.

Vector Fields

5. A vector field on a smooth manifold X is a continuous

map X Tang X which takes each point x € X into a vector tangent
to X at x. A trivial example is the zero vector field, whose value
at each point x c X
is the zero vector of the space Tang X (see 2).
^r +1 ^
A smooth, n-dimensional, C -manifold X is
r r
C -parallelizable if there exist n C -vector fields
f^,...,fn : X Tang X such that, at each point x € X, the vectors
f^ (x),...,fn (x) yield a basis of the space Tang^X. For example, HR
(regarded as a Ca-manifold) is Ca-parallelizable: a parallelization is
given by the vector fields which associate to each point x € JRn the
(id 3Rn )-basis of the space lRn . Tang
0 x
C -parallelizability is simply called parallelizability. It
will be shown in Chapter 4 (see 4.6 .4.3) that the parallelizabilility
r +1 r
of a compact C ' -manifold with r £ 00 implies its C -paralleliza­
bility .

INFORMATION. The Cr-parallelizability is a consequence of

<2r +1
the parallelizability of a C ' -manifold even if the manifold is not
compact, or if r = a.

6. Suppose that the smooth manifold X is parallelized by

the Cr-vector fields f^,...,f^: X Tang X . Then the formula
(x, (y1 ,...,yn )) h* y 1 f 1 (x) + ... + y^f^(x) defines a Cr-diffeomorphism

of the product X x jr u onto Tang X . Indeed, the formula

v » (pr(v), (y1 ,...,yn )), where are the coordinates of the
vector v relative to the basis f 1 (pr(v)),...,f (pr(v)) of

Tang^r ^ X , defines the inverse map Tang X -* X x iRn , and obviously

both maps are of class Cr . Thus the total manifold of vectors tangent
to a Cr-parallelizable n-dimensional smooth manifold X is Cr-diffeo-
morphic to the product X x ]Rn .

In particular, we deduce that every C1-parallelizable smooth

 161

manifold is orientable. In fact, Tang X is orientable for any

C -manifold X with r ^ 2 (see 2); hence, in our case, the product
X x HR is orientable, which in turn implies the orientability of X
(see 3.7).

7 (EXAMPLES). For each odd n, there is a vector field

with no zeros on the sphere Sn . For example, such a vector field
S" +3Rn + 1 is given by x = (x , ...,x ) » (y ,...,y ) £ ® n+1 ,
n +>| I nt | | n+ I
X £ S , where y 2k_ 1 = - x 2 k , y 2k = x 2k-1 (k = 1 » • • • # (n + 1 ) /2) ;

here we consider ' ''',yn + 1 * as a vector in Tangos11 (see 4 ).

Note that the same vector field can be defined in a more
concise fashion as the map x ^ xi, regarding nRn + ^ as (E^n +^ >^.
If n+1 is divisible by 4, lRn + ^ may be regarded as +1 )/4 ^ an(^
the formulas x ■+ x ort 2 , x » x ort^ , and x ^ x ort^ (ort^ ort3•
and ort^ are considered here as the imaginary quaternion units) define
three vector fields on Sn which are linearly independent at each point.
If n +1 is divisible by 8 , 3Rn + >* may be regarded as Œa^n+^ ^ , and
the formulas x *+ x ort^ , • • • , x x ortg (ort^, ...>ortg are
considered here as the imaginary Cayley units) define seven vector
fields on Sn which are linearly independent at each point. Since all
the above vector fields are analytic, this construction shows, in
particular, that S^, S^, and S^ are Ca-parallelizable.

INFORMATION. For n i 0,1,3,7 the sphere Sn is not

parallelizable. For a proof, see [1] and [2].

5. Embeddings, Immersions, and Submersions

•£
1. A map f: X Y of smooth manifolds is a C -embedding
if f (X) is a Cr-submanifold of Y and ab f : X f (X) is a
Cr-diffeomorphism. For example, the inclusion of a submanifold into
its ambient C ^ r -manifold is a Cr -embedding. Since every map
f: X -* Y can be written as the composition of its compression
ab f : X + f(X)
with the inclusion f (X) + Y, a Cr-embedding is
27
really a map of class C .
The -embeddings are also termed differentiable embeddings.
Using Theorem 3 we shall prove below, we can see that a differentiable
embedding which is of class Cr is a Cr-embedding. Moreover, it is
evident that differentiable embeddings are topological embeddings. The
latter are sometimes called C^-embeddings.
A differentiable embedding f : X -+ Y is neat if f (X) is a
 162

neat submanifold of Y. For example, the inclusion of a neat submani­

fold into its smooth ambient manifold is such an embedding. Clearly,
if dim X = dimY , Y is connected, and X ^ 0, then every neat
differentiable embedding X Y is a diffeomorphism.
It is obvious that for eachCr-embedding f: X Y andeach
point x€ X, there are charts cp£ Atlx CrX and ^ £ At^f(x)^ Y su°h
that loc(cp,iJj)f coincides with one of the inclusions

JRm + IRn , 1R™ 1bP_ , or 3R™ -+ 3R^ ,

where m = dim X and n = dimY If f is neat,then the second case

must be excluded.

Immersions

2. A smooth map f: X Y of smooth manifolds is an

immersion if: (i) d„f: Tang_X Tang^^^Y is a monomorphism for any
m X X i ixj
x € X; (ii) (df) (Tang 3Y) c: Tang 9X .
We remark that condition (i) implies that dim X £ dim Y ,
and condition (ii) - that f (intX) c int Y . If 3Y = 0, then (ii) is
automatically fulfilled.
Trivially, the composition of two immersions is an immersion.
The differentiable embeddings are examples of immersions.
27
3. If_ f: X Y is an immersion ofclass C ,then each
point of the manifold X hasa neighborhood N such that the
restriction of f toN is a C -embedding.

PROOF. Let xQ£ X be an arbitrary point, and put m = dimX,

n = dim Y . Now pick some charts cp £ Atl X, ip £ Atl . .Y, and denote
x0 0
by the coordinate functions of the map loc (tp, ) f .
According to condition 2(i), the Jacobi matrix of loc(tp,\Jj)f at
tp(x0) has rank m. Using condition 2(ii), one can assume that the
minor M of this matrix, constructed from its first m rows, is not
zero. [If f (Xq ) £ intY , 2(ii) is not necessary: one can achieve
M f 0 by reindexing the local coordinates of the chart
i.e., bypermuting the rows of the matrix. When ^ 3Y' 2 (ii)
ensures that all the elements of the elements of the first row of the
Jacobi matrix, starting with the second, are zero; this allows to
achieve M ^ 0 by reindexing the coordinates ^ ,. ..,\jj .] Next apply
the implicit function theoremto deduce the existence of a neighborhood
W of the point (f (xQ) ) ,.../^m (f (xQ)) ) in and of a
 163

JC
C -embedding h: W Im cp , such that, for i = 1 , .. .,m, one has:

h i ( ^ 1 (f (xQ) ) ,.. .,^m (f (xQ )) ) = Vi (xQ)

and

W y1 I'm» V y1 V = *1 ' < * 1 ............ y m> € w l '

where h,j , ...,h are the coordinate functions of h. Let

N = cp 1 (h(W)). Evidently, N is a neighborhood of xn . Its image

”1 n —m
f (N) in cp (W * IR ) is defined by the equations
~ £j (ip^ , .. .,ipm ) ,.. .,hm (i/^ , ...,^m >) = 0 ,and, possibly, bythe
additional inequality (ip ,... ) c 0. Therefore, f(N) is a
Cr-submanifold of Y. The composite map

f(N) — '
b ^ -» iMf(N)) -- ► W — -* h(W) - (abtP) 1 » N ,

(where the second map is the compression of the orthogonal projection

3Rn ^ IR111) is of class Cr and is the inverse of the map ab f : N f(N).
We conclude that f|N is a Cr-embedding.

4 (COROLLARY). If an immersion of class Cr is a topological

£
embedding, then it is a C -embedding. In particular, every injective
r r
C -immersion of a compact manifold is a C -embedding.

5. Let f: X Y be a smooth map such that

- 1
(df) (Tang 3Y) c Tang 3X, and let A be a compact subset of the
manifold X. If
f|^ is injective and the differential d„f is
I - ^ x
nondegenerate for each point x E A, then f is a differentiable embed­
ding on some neighborhood of A. In particular, if werequire, in
addition to the previous conditions, that dim X = dim Y and
f(9X) c 3Y , then f carries a neighborhood of A diffeomorphically
onto a neighborhood of f(A).

PROOF. Fix for each point x E A a neighborhood such

that fly is a Cr-embedding (see 3), and then cover A by a finite
‘x
number of such neighborhoods, say U ,...,U . Since the set
X1 xs
Y x Y ^ diag Y is open in Y x Y (see 1.2.2.4), its preimage W
under the map f x f ; xxx-^YxY is also open. But f is injective
on A; whence W U [IIs . (U x U )] contains A x A, and is actually
1— I X• X.
1 1
a neighborhood of A * A. Next introduce a metric on X (see 1.1.12),
get the corresponding metric on X x X (see 1 .2.2.9), .and then set

B = {x £ X | Dist(A,x) < Dist((X x X) ^ W,A x A)/2 }.

Since B is open and contains A (see 1.1.7.15), the intersection

 164

B fl (U. „U ) is also ooen and contains A. This intersection contains

1 =1 x .
l
a relatively compact neighborhood U of A, because A is compact.
Moreover, B x B c w, and so f is injective on B. We conclude that
f |Ci y is a topological embedding and f|^ - a differentiable embedding

(see 4).

Submersions

6. A smooth map f: X Y of smooth manifolds is a

submersion if: (i) d f: Tang X -> Tang^ , XY is an epimorphism for any
---------------- X X r (X;
x € X; (ii) ¿xf(Tangx 8X) = Tangf(x)Y for any point x € 3X fl f (intX);
__-1
(iii) 9X fl f (int Y) is a union of whole components of the manifold
9X.
We remark that condition (i) implies dim X ^ dim Y , and that
conditions (ii) and (iii) are automatically fulfilled whenever 9X = 0.
As examples, consider smooth real functions f: X -* 3R. As in
the classical calculus, a point x £ X and the corresponding value
f (x) are said to be critical for f if d f = 0. It it clear that
-------- x
f is a submersion if and only if the functions f and f | have no
critical points.
As additional examples of submersions we cite the projections
of the product of two smooth manifolds (one of them being without
boundary) onto its factors.

7. A map f of a smooth m-dimensional manifold X into a

27
smooth n-dimensional manifold Y is a submersion of class C if and
27
only if for each point x € X there are charts
X-------- _----
cp £ Atl x C X and
---
ip £ Atl^ C Y such that the following holds: f (suppcp) c supp ip ;

the pair (Imcp,Im ip) coincides with one of the pairs (IRm ,IRn ) ,
QR™ ,IRn ) , or 0R^,3R^) , and in each case cp(x) = 0 and ij-(f(x)) = 0;
the corresponding local representative loc(cp,\JJ)f can be described,
in the first case, as the projection of the product HRn * 3Rm n onto
its first factor, in the second case - as the projection of the product
JR^ n x IRn onto its second factor, and in the third case - as the
projection of the product 2R^ x nRm n onto its first factor.

PROOF. The sufficiency of this condition is obvious. Let us

1 1
prove its necessity. Let cp £ Atl^X and {¡ j £ Atlf ^ Y be a r b i t r a r y
1 1 1 1
charts such that f (suppcp ) = supp ip , cp (x) = 0 , and iJj (f(x)) = 0 .
1 1 1 1
We denote by cp ,.. .,ip and ,.. .,\p the corresponding coordinate
 165

functions, and by £,...,£ the coordinate functions of the map

1 1 i n
loc (cp )f. We consider three distinct cases: a) x G int X ,
f(x) £ int Y ; b) x £ 8X, f (x) £ intY ; and c) f(x) £ 9Y.
1 1
Condition 6 (i) says that the Jacobi matrix of the map loc (cp , \p )f,
computed at the point 0, has rank n in each of the three cases.
Condition 6 (ii) says that in case b) this rank does not decrease when
we remove from the matrix the first column. Therefore, in cases a) and
c) we may assume that the minor constructed from the first n columns
does not vanish, while in case b) the same is true for the minor
constructed from the last n columns. Moreover, in case c ) , 0 is a
boundary point of Im cp in nRm , and the function vanishes on
1 1
cp (8X H Im cp ) in a neighborhood of this point. Indeed, the first
part of the last assertion follows from the fact that is a non-
positive function, vanishes at 0 , and has nonzero gradient at 0; now
the second part of the assertion is seen to be a consequence of 6 (iii).
1 1
Wepass from cp ,\p to the required charts cp,i{; through
2 r 2 r
the intermediary charts cp £ Atl C X and if; £ Atl^ { *C Y. In cases
x r IX )
2
a) and c), cp is the chart whose local coordinates are the
1 1 1 1
restrictions of the functions ^ o cp ° ^ /Cpn +1 '***/Cpm tC> a
2
small enough neighborhood U2 of the point x, while ip is the
1 1 2
chart ab ip : V 2 \p (V2> ,where V2 = f (U2> . In case b) , cp and
2 1 1 1
if; are similarly defined, but one replaces ° cp ,..., 1 ^ o cp /<Pn+-1 '

...,cp^ by the functions ' • • • ' ^ 1 ° ^ ' ’’ ’ ° ^

cases f<U 2 ^ = V 2 ' and obviously* in cases a) and c) , the map
2 2 2 2
loc (cp ,if; )f is given in the new coordinates cp^ ,...,cpm and
? 2 2 2 2 2
^ 1 , ...,i^n by the formulas i^ = cp^ ,. ..,i^n = cp^, whereas in case b)
2 2 2 2 . . .
the corresponding formulas are if^ = cPm_n +^ / •••/4fn ~ <Pm - Fix a positive
e and define the subsets U and V of U2 and V 2 by the
inequalities |cp? | < £ (i = 1 ,. ..,m) and |^j| < £ (j = 1 ,.. .,n) ,
respectively. It is clear that for e small enough, the charts cp,if;
with supp cp = U , supp ip = V, and local coordinates
2 . ,
w, (y)
cp. (y) = ---- -— o (i = 1 ,••.,m)
l
e - |tp? (Y)

and
iK(z)
(z) = --- - ^ —2--- 7 (j =
D e - | \p j ( z )

have the desired properties

 166

8 (COROLLARIES) . If_ f: X + Y is a submersion, then

f (int X) c int Y , f_ 1 (3Y) c: 3X, and the maps ab f : int X int Y
- 1
and ab f : f (3Y ) 3X are submersions.
r ”1
If f: X Y is a submersion of class C ,then f (y )
is a neatCr-submanifold of X for y € int Y , and a neat (^-submani­
fold of 3X for y e 3Y.

Every submersion is an open map.

The composition of two submersions is a submersion.

9. A Cr-map f: X Y satisfying condition 6 (iii) is a

submersion if and only if for each point £ X there is a
neighborhood V of the point and a C -map g: V ->■ X, such that
f(g(y)) = Y for all y € V and g(f(xQ)) = x Q .
The sufficiency of this condition is clear, its necessity
results from 7.

10. Let f: X Y be a submersion of class Cr such that

f (X) =Y, and let hbe a map of Y into a third manifold. If the
r r
composition h o f is of class C , then h is of class C too.

PROOF. According to 9, one can find for each point of Y a

neighborhood V and a C -map g: V X such that f°g=[in:V + Y],
and hence h | ^ = (h o f) o g.

6. Complex Structures

1. Recall that a map of an open subset of CCm into a

subset of CCn is holomorphic if its coordinate functions are
holomorphic, and biholomorphic if it is invertible and both the map and
its inverse are holomorphic. Obviously, in the last case we must have
m = n; cf. 2 .2 .
If we regard £Cm and (En as and IR2n , respectively,
then the holomorphic maps become Ca-maps, and the biholomorphic ones -
Ca-diffeomorphisms. A smooth map of an open subset of HR2m into a
2n
subset of 3R is holomorphic relative to the complex structures on
3R2m and ® 2n resulting from the identifications 3R2m = (Em and
]R n = CEn if and only if it satisfies the Cauchy-Riemann conditions.
Evidently, a map which is the inverse of a diffeomorphism satisfying
the Cauchy-Riemann conditions also satisfies these conditions.
Consequently, every holomorphic diffeomorphism is a biholomorphic
map.
 167

2. Suppose that a holomorphic map between subsets of (Cn

has a nondegenerate differential at some point. Then it retains the
same property when considered as a Ca-map, and hence it maps a
neighborhood of the given point diffeomorphically onto its image. Thus,
a holomorphic map whose Jacobian does not vanish at a point maps a
neighborhood of the point biholomorphically onto its image.
This statement is the exact analogue, and also the result of
the theorem concerning local inversion of a smooth map in the real case.
In a similar fashion, one can translate a more general theorem from the
real calculus into the complex language - the implicit function theorem.
Namely, let f be a holomorphic map of an open subset A of (Em x (En
into (En , and let (z^,w^) £ A be such that f(z^,w^) = 0. Suppose
that the Jacobian of f with respect to the second variable does not
vanish at (z^,w^).
The complex implicit function theorem states the
0 m
existence of a neighborhood U of the point zin (E, of a
0 n
neighborhood V of the point w in (C , and of a holomorphic map
g: U -> V, such that U x V c A and the preimage of 0 under f ju x v
is precisely the graph of g.

3. A linear transformation of (En is also linear as a

transformation of 3R^n ; hence to each complex n><n-matrix C one can
associate a real 2nx2n-matrixR. If C = A + iB and z = x + iy
are the decompositions of the matrix C and the vector z € (Cn into
real and imaginary parts, then Cz = (Ax - By) +i(Bx + Ay), and
(A -b '
R =
B A
i i2
In particular, detR = |det C | . Indeed, if we add to the first
block-row of R the second, multiplied by i, and then add to the
second block-column the first, multiplied by -i, we obtain the matrix
'c O'
which has the same determinat as R.
B

Complex Manifolds

4. An n-dimensional complex chart on the set X is an

invertible mapping of a subset of X onto an open subset of (C . Two
complex charts, cp and i/
j, are compatible if the set
cp(supp cp n supp ip) is open in Im cp , the set iMsuppcp H suppijj) is
open in Im iJj , and the composite maps
 168

cp(supp cp n supp ^ ^-- > supp cp n supp ip ^ ^ (supp cp fl supp ^ )

and _^
il) (supp cp fl supp ^ --— > supp cp fl supp ^ -
a— c5_-> cp(supp cp fl supp ),

which are inverses of each other, are holomorphic (here supp is

defined as in the real case). If two overlapping charts of dimensions
m and n are compatible, then m = n.
A collection of complex charts is an n-dimensional holomorphic
atlas of the set X if these charts cover X, are n-dimensional, and
are pairwise compatible. Two holomorphic atlases are holomorphically
equivalent if their union is again an atlas. The family of n-dimensional
holomorphic atlases of X is divided into disjoint classes of holo­
morphically equivalent atlases. These classes are called n-dimensional
complex structures on X.

5. Any n-dimensional complex chart may be regarded as a

2 n-dimensional real chart, i.e., a 2 n-dimensional chart in the sense of
2.3. Furthermore, compatible complex charts yield Ca-compatible charts,
holomorphic atlases yield Ca-atlases, and holomorphically equivalent
atlases yield Ca-equivalent atlases. Therefore, an n-dimensional complex
structure on the set X induces a 2n-dimensional Ca-structure on X.
For us, the most important case occurs when this Ca-structure makes X
into a manifold, i.e., when it defines a Hausdorff, second countable
topology. A set X equipped with an n-dimensional complex structure
enjoying this property is an n-dimensional complex manifold.
We let AtlX denote the complete atlas of the complex
manifold X, i.e., the collection of all charts of all atlases of its
complex structure.

6. A continuous map f : X -> Y between complex manifolds

is holomorphic if all its local representatives are holomorphic, i.e.,
— 1 — *
the maps cp(supp cp fl f (supp ^) ) + Im , x i[>(f (<> (x) ) ) ,
constructed by means of the charts cp E Atl X and ijj G Atl Y , are
holomorphic. A map f: X Y is biholomorphic if it is holomc 'phic,
invertible, and its inverse is also holomorphic. Two complex manifolds
which can be transformed one into another by a biholomorphic map are
said to be biholomorphically equivalent.
If complex manifolds are considered as Ca-manifolds, the
holomorphic maps become Ca-maps, and the biholomorphic maps - Ca-diffeo-
morphisms.

7. Let A be a subset of an n-dimensional complex manifold

X. A is a k-dimensional submanifold of X if for each point x G A
 169

there is a c har t cp e Atl X such that x £ supp cp and

k k
cp(supp cp n A) = Im cp n (E . The charts ab cp : supp cp fl A Im cp fl (E
derived from the c h a r t s cp f . Atl X fo r m a k - d i m e n s i o n a l h ol o m o r p h i c
atl as of the set A, thus t ra n s f o r m i n g A into a complex manifold.
The notion of independent functions defined in 2.12 makes
sense for the complex case too. Therefore, using the implicit function
theorem, we deduce that a subset A of the complex n-dimensional
manifold X is a k-dimensional submanifold of X if and only if for
each point Xq £ A there are a neighborhood U of x^ in X and
holomorphic functions cp^+ ^ ,. ..,cpR : U (E, independent at xQ, and
such that the intersection A fl U is defined in U by the equations
cpk + 1 (x) = 0 , ..., cpn (x) = 0 ; cf. 2 .1 2 .
If the complex manifold X is regarded as a Ca-manifold,
then a submanifold remains a submanifold and its Ca-structure induced
from the Ca-structure of X is identical to the Ca-structure induced
by its own complex structure.
If A is a submanifold of the complex manifold X, then the
inclusion A -* X is holomorphic.
A map f: X Y between complex manifolds is a holomorphic
embedding if ab f : X ■> f (X) is a biholomorphic map of X onto a
submanifold of Y. In this case f is the composition of the biholo­
morphic map X -> f (X) and the inclusion f (X ) -* Y . We conclude that
every holomorphic embedding is a holomorphic map.

8. Suppose X^ ,. ..,Xg are complex manifolds of dimensions

n^,-..,ns . The products cp^ x ... x cpg of all charts cp^ £ Atl X^

form an (n„+...+n )-dimensional holomorphic atlas of the set

1 s
X 1 x ... x x , transforming it into an (n1+...+n )-dimensional complex
I s I S
manifold. Considered as a C -manifold, the latter is just the product
of the Ca-manifolds X ,...,X .
I s
9. The same definitions of tangent vectors, tangent vector
spaces, total manifold of tangent vectors, and differential of a map
(see Subsection 4) apply in the complex case. The space Tang^X
tangent to the n-dimensional complex manifold X at the point x is
an n-dimensional complex vector space; the total manifold Tang X is
a 2n-dimensional complex manifold, and the projection Tang X -* X is
holomorphic. The differential d f of a holomorphic map f: X Y
at the point x € X is a linear mapping Tang^X Tan9f(x )Y of complex
vector spaces, and the differential df is a holomorphic mapping
Tang X -* Tang Y .
Again, the tangent spaces Tang^X and the manifold Tang X
 170

may be considered as real vector spaces and as a Ca-manifold,

respectively. As one may guess, they coincide with the tangent space
and the total manifold of tangent vectors to X, regarded as a
Ca-manifold. The differential of a holomorphic map f, regarded as a
Ca-map, coincides with the differential of the Ca-map f.

10. The simplest examples of complex manifolds are the

spaces CCn themselves. We obtain an unlimited supply of additional
examples by defining submanifolds of £En through systems of equations;
cf. 2.16. However, this method will never produce compact manifolds of
positive dimension: every compact submanifold of (En has dimension
zero.
To convince ourselves that this is true, it is enough to show
that on a compact manifold the only holomorphic functions are the
constants. This is a straightforward consequence of the well-known
theorem stating that a function holomorphic on an open subset of the
complex line (E which attains its maximum modulus is constant. Now
suppose X is a compact, connected, complex manifold, and f: X (E
is holomorphic. Let c be a value of f such that |c! = max If (w) | ,
and let x £ X be such that f (x) = c. Then for each chart cp £ Atl X
with x £ supp cp and Im cp = I n t D ^ ^ m X , and for each point
y £ supp , the set of complex numbers z such that
(1 - z)cp(x) + zcp(y) £ Im cp is an open disc with center 0 and radius
greater than 1. The formula z + f (cp ((1 - z)cp(x) + zcp(y))) defines
a holomorphic function on this disc, which attains its maximum modulus
as 0.
Since such a function is necessarily a constant, we see that
i _
f(y) = c by setting z = 0 and z = 1. Therefore, the set f (c)
-1
is open. Because f (c) is also closed and nonempty, it is all of X.
Examples of compact complex manifolds will appear in §2.

Manifolds of Complex Origin

11. Every complex manifold gives rise to a Ca-manifold when

we pass from complex to real numbers, as described in 5. A Ca-manifold
arising in this way is called a manifold of complex origin.
Clearly, the manifolds of complex origin are even-dimensional
and have no boundary. They are orientable, and if the original complex
structure is known, they receive a canonical orientation, namely that
orientation which is positive on the real connected charts which arise
from the charts of the complete atlas of the complex structure when we
pass to the reals. (Using 3 we see that the compatibility condition
 171

required in 3.1 is satisfied.) It is also clear that the products of

manifolds of complex origin are again manifolds of complex origin.

7. Exercises

1. Show that a nonempty, closed, smooth manifold of dimension

n > 0 cannot be immersed in HRn .
2 2
2. Show that the equation + ... + zr = 1 defines a sub­
manifold of (Cn which is Ca-dif feomorphic to Tang Sn ~ 1 .
2 4
3. Show that the map f: S ->•3R , defined by the formula
2 2
f (x ' x 2 / X 3 ^ = ^X 1 ~ x 2 'X 1X 2 ,X1X 3/x^x^) , is an imerssion and that
2 4 2
f(S ) is a submanifold of ]R diffeomorphic to 1RP .

4. Show that if one of the numbers n. ,...,n is odd and

1
s > 1,then the manifold S x ... x s s is parallelizable.

§2. STIEFEL AND GRASSMAN MANIFOLDS

1. Stiefel Manifolds

1. We denote by
JRV(n,k), or simply by V(n,k) (0 £ k £ n ) ,
k n
the set of linear isometric maps 1R ->■1R . Such a map is uniquely
k
determined by the images of the vectors ort^,...,ort^ E 3R , i.e., by

an orthonormal k-frame in ]Rn . The coordinates of the vectors of this

frame form the matrix of the map, which has n rows and k columns.
In this way, V(n,k) can be interpreted as the set of orthonormal
k-frames in lRn , or as the set of the nxk-matrices|[vsi [[ such that

yn . v .v . = 6 .. (1 £i £ j £ k) . (1 )
L s=^ si sj id J
nk
We may regard a matrix livsill as a P°i-nt of 3R , if we
index its entries v . in dictionary order. Thus V(n,k) becomes
si
nk
the subset of 3R defined by the equations (1). An easy computation
shows that the gradients of the left-hand sides of (1 ) do not vanish
and are pairwise orthogonal on this subset. Hence V(n,k) is an
 172

a nk
[nk-k(k+1)/2]-dimensional C -submanifold without boundary of 3R
(see 1.2.12). V(n,k) is called the Stiefel manifold.
Clearly, V(n,0)reduces to a point, V(n,1) is just the
sphere Sn \ and V(n,2) is thesubmanifold of all vectors of unit
length in Tang Sn 1 .
2. The points of V(n,n) are orthogonal transformations of
3Rn , or orthogonal matrices of order n, and V(n,n) is usually
denoted by 0(n). The composition of transformation (multiplication of
matrices) induces a group structure on 0(n). The subgroup of 0(n)
consisting of all matrices with determinant +1 is denoted by SO(n) .
The two sets SO(n) and 0(n) \ SO(n) are open in 0(n), and hence
are Ca-manifolds. They are actually Ca-diffeomorphic: multiplication
by an arbitrary matrix from 0 (n) \ SO(n) establishes a diffeomorphism.
Moreover, the manifold SO(n) is canonically Ca-diffeomorphic to
V(n,n-1): a matrix from V(n,n-1) is carried by this diffeomorphism
into a matrix from SO(n) through the addition of a column; that is
to say, we complete each orthonormal (n-1)-frame in IRn to a positive
orthonormal n-frame.
We further note that SO (2) = V(2,1) = s \ and the group
structure on S O (2) agrees with the group structure on the circle S ,
considered as the multiplicative group of complex numbers of modulus 1.

3. The inclusion ]Rn -+ nRn+q

induces a Ca-embedding
k n
V(n,k) V(n+q,k), which transforms each map cp: 1R -+ TR into the

composite map lRn i^->nRn+q. There is also a canonical Ca-embedding

k n
V(n,k) -*■ V(n+q,k+q) , which transforms tp: 3R -*■]R into the map

IRk+q = 3Rk x ir^ id.> xEq =

a k —cr k
Finally, the inclusion IR M ->•]R
induces a C -submersion
k n
V(n,k) -*■ V (n,k-q) , which transforms each map cp: 3R -*•1R into the
k —cr in k n
composite map ]R M -- > ]R ■+tr , i.e., each frame (v ,...,v ) is
I K
taken into the frame (v^,...,vk_q )• Using Theorem 1.5.9, it is readily
seen that this is indeed a submersion, because given any frame
.0 0, , .
(v^,...,vk ) € V(n,k) we have an explicit construction of the neighbor­
hood V of the frame (v°,...,vk_q ) in V(n,k-q) and of the Ca-map
g : V ->■ V (n,k) which are required by this theorem. In fact, one can
take as V the set of frames (v 1,...,vk_q ) in V(n,k-q) such that

the vectors V 1 '•*''vk_q'vk_q+ 1 '•••'vk are linearly independent and

then, for ^ ^ ' define g (v^ ,. ..,vk _q ) to be the frame

obtained from (v^,.../vk _q »vk _q+^/•••»vk ) through standard

 173

orthogonalization. Let us add that the preimage of an arbitrary frame

(V 1 > •••'vk _g) Ç V(n,k-q) under this submersion is the submanifold of
V(n,k) consisting of all the frames (v?,..\,v£_ ,v ,...,v, ),
I -K q }c q 1 K
where (vk -q * 1 Vr is an orthonormal q-frame of the (n-k+q)-dimen-
sional subspace of 3R which is orthogonal to the vectors v v,° ;
1 k-q
in particular, this submanifold is diffeomorphic to V(n-k+q,q)

4.
We see from equations (1) that the set V(n,k) is bounded
nk
and closed in 1R . Therefore, V(n,k) is a closed manifold.
The manifold V(n,n-1) = SO(n) is connected: each matrix of
SO(n) can be expressed as cu (^ ,...,(j)^)c ^ where

cos -sin

sin cos

CO S -sm
U ( (¡)
1'
sin cos

(with <i><|*.. .*$ € 3R) and c is an orthogonal matrix. Therefore,

each matrix can be joined to the identity matrix by the path
t *+ cu((1 - t) 4^ ,...,(1 - t ) $ ^ ) c . Since the manifolds V(n,k) with
k < n-1 are the images under continuous maps of V(n,n-1) (see 3),
they are also connected. For n > 0, V(n,n) has two connected
components: S0(n) and 0(n) \ SO(n).

The Complex Case

5. Let (EV (n,k ) , 0 £ k £ n, be the set of linear isometric

V rj
maps (C (C . In other words, (CV(n,k) consists of the orthonormal
k-frames in (En or, equivalently, of the complex n xk-matrices 11 vsj_11
such that

Y-n 1 v . v . = 6 .. (1 £ i £ j £ k).
S = 1 SI
SI SI
SJ 111
J

, v ^ „nk m 2 nk
These equations show that (CV(nfk) is a subset of (E - IR . Now
2
we separate their real and imaginary parts and obtain k real equations
 174

such that the gradients of their left-hand sides do not vanish and are
2
pairwise orthogonal on CEV(n,k) . Thus (EV(n,k) is a (2nk-k )-dimen­
sional Ca-submanifold without boundary of IR^n^ , called the complex
Stiefel manifold.
Warning: (EV(n,k) is nota complex manifold in the sense of
1.6.5. The present definition doesnot equip it with a complex
structure; in fact, such a structure does not exist in general, since
CDV(n,k) is odd-dimensional for k odd.
Clearly, (EV (n,0) reduces to a point, and CCV(n,1) is just
the sphere S^n The points of the manifold CCV(n,n) are unitary
transformations of CCn , and CCV(n,n) is usually denoted by U(n) .
Like 0(n), U(n) is a group under the composition operation o . The
subgroup of U(n) consisting of all matrices with determinant 1 is
denoted by SU(n) and is a Ca-submanifold of U(n), canonically
diffeomorphic to (EV(n,n-1) (cf. 2).
The manifold U(n) is canonically diffeomorphic to
1 1
SU(n) x S : this diffeomorphism takes each pair (u,z) E SU(n) x S
into the matrix obtained from u by multiplying its first row by z.
Warning: this diffeomorphism is notagroup isomorphism between the
direct product of groups SU(n) x s and U(n).
The Ca-embeddings (EV(n,k) (EV(n+q,k) and
(EV(n,k) (EV (n +q,k+q) , and the Ca-submer sion CEV(n,k) CEV(n,k-q) are
defined exactly as in the real case. Moreover, since each linear
k n
isometric map (E + (E may be regarded as a linear isometric map
2k 2n a
H ->•3R , there is a canonical C -embedding (EV(n,k) -*■HRV (2n ,2k ) .
The manifolds (EV(n,k) are compact. Moreover, they are all
connected. Indeed, (EV(n,n) = U (n) is connected, because every unitary
matrix can be expressed as cu ( ^ , ...,<j>n )c , where u (<j>^ , .. .,<J> ) is
the diagonal matrix with entries exp (i<J).j) , ...,exp (i<|> ) , ... ,4>n £ ir)
and c is a unitary matrix, and then joined to the identity matrix
by the path tn-cu((1 - t) <|>1 ,...,(1 - t) $ ) . The manifolds (EV(n,k)
with k < n are connected as continuous images of IV(n,n).

The Quaternionic Case

6 . The discussion in 5 can be repeated almost word for word

if one replaces the field of complex numbers by the field of quaternions.
(Recall that Hn
is considered as a left vector space; see 1.2.5.4;
k n
consequently, a linear map H -*■ U is left-linear here, and the scalar
product of the vectors (u ,...,u ) and (v.,...,v ) is defined as
i n I n
 175

¿i=1 uiv i-> In this way, we obtain: the C -manifold without boundary
HV(n,k) (0 ^ k ^ n) of dimension 4nk-(2k2 -k), called the
quaternionic Stiefel manifold; the Ca-embeddings HV(n,k) +HV(n+q,k)
HV (n ,k ) + KV(n+q,k+q) , and MV(n,k) -v (EV(2n,2k);and the Ca-submersion
HV(n,k) + HV(n,k-q)
Clearly,' HV(n,0) reduces to a point, and UV(n,1) is just
4n— 1
S . The manifold HV(n,n) is usually denoted by Sp(n); its points
are the linear isometric transformations of H n , and the composition
of transformations makes Sp(n) into a group.
All the manifolds living) are compact and connected. Since
the proof of connectedness along the lines in 4 or 5 requires the
normal form of amatrix from Sp(n), which is less known than the
normal forms for SO(n) and U(n), we remark that a different proof
of connectedness is given in Chapter 5 (see 5.2.7.4).

Noncompact Stiefel Manifolds

7. Let
IRV* (n,k), or simply V'(n,k), 0 £ k £ n, denote
k n
the set of linear monomorphisms 1R -*•1R . Alternatively, one may
describe V'(n,k) as the set of nondegenerate k-frames in 3Rn , or as
the set of the real nxk-matrices of rank k. Clearly, this set is open
nk
in the space IRof all real nxk-matrices, and hence V 1 (n,k) is an
nk-dimensional Ca-manifold containing V(n,k) as a submanifold.
Let T(k^IR), or simply T(k), be the set of real upper
triangular matrices of order k with positive diagonal entries.
Trivially, T(k) is open in the space of all upper triangular matrices
of order k, n R ^ ^ +1^ 2 , and is actually diffeomorphic to +
If we orthogonalize a given nondegenerate k-frame in lRn via the
standard procedure, then the matrix corresponding to this frame takes
the form ut, where u £ V(n,k) and t £ T(k). Moreover, it is
obvious that this representation is unique and defines a diffeomorphism
V'(n,k) -* V(n,k) x T(k), transforming V(n,k) into the fiber
V(n,k) x E, where E is the identity matrix. Therefore, the manifold
V'(n,k) is Ca-diffeomorphic to V(n,k) x i R ^ ^ + 1^//^ / and V(n,k) is
its strong deformation retract. In particular, V*(n,k) is connected
for k < n, and V f(n,n) has two components.
Again, V f(n,0) reduces to a point, while V*(n,1) coincides
with nRn ^ 0. The manifold V*(n,n) is usually denoted by GL (n ,1R)
and its points are the nondegenerate linear transformations of IRn or,
equivalently, the nondegenerate matrices of order n. The composition
 176

of transformations (multiplication of matrices) defines a group

structure on GL(n,IR) .Th’e subgroup of GL(n,IR) consisting of all
matrices with positive determinant is denoted by GL+ (nJR).The sets
GL (n^R) and GL (n ,IR) \ GL (n,]R) are open in GL (n ,IR) and Ca-diffeo-

morphic to SO (n) x JRn *n + 1 ^^ 2 ; in fact, they are the components of the

manifold GL (n ,IR) .
k n
Corresponding to each monomorphism cp: IR -*• 3R we have the
., -rJc (p _n in _n +q „k+q id k ^ cp*id . m n ^ m q id, m n+q
composite maps 3R IR ---* IR , 3R *3R x IR ---- * JR * JR ► -IR

and ]R^ ^ ~*~n > IR^ ^ nRn , and so we obtain the Ca-embeddings
V 1 (n,k) V 1 (n+q,k) , V'(n,k) -+ V'(n+q,k+q), and the Ca-submersion
V ' (n,k) -> V' (n,k-q) , respectively; cf. 3.

8.
We let (EV' (n ,k ) (0 £ k £ n) denote the set of linear
k n
monomorphisms (E -+ (E . Alternatively, (EV(n,k) is the set of non­
degenerate k-frames in (En ,
or the set of complex nxk-matrices of rank
nk
k. Again, it it clear that this is an open set in the space (E of
all complex nxk-matrices. Thus (EV!(n,k) is a 2nk-dimensional
Ca-manifold containing (EV(n,k) as a submanifold.
Repeating what was said in 7 (with obvious modifications),
we obtain a Ca-diffeomorphism (EV* (n,k) CEV(n,k) x T(k,(E), where
T(k,(E) is the manifold of complex upper triangular matrices of order k
with positive diagonal elements. Clearly, T(k,(E) is Ca-diffeomorphic
k2 k2
to IR , and hence (EV* (n,k) is diffeomorphic to (EV(n,k) * IR . In
particular, the manifolds CEV*(n,k) are all connected. (EV'(n,0)
reduces to a point, while (EV! (n,1) is just (En \ 0. Usually the
manifold (EV’(n,n) is denoted by GL(n,(E); it consists of all non­
degenerate linear transformations of (En , and is a group with group
operation o. There are also the natural Ca-embeddings
(EV 1 (n,k) -* (EV 1 (n+q,k) and (EV'(n,k) (EV 1 (n+q,k+q) , as well as the
natural Ca-submersion (EV1 (n,k) (EV *(n,k-q) .

9. If we replace the commutative fields IR and (E by the

field in the previous definitions, we obtain: the 4nk-dimensional
Ca-manifold nHV'(n,k) of all linear monomorphisms ->3Hn ; the
2
Ca-diffeomorphism IHV1 (n,k) IV(n,k) x ]R2k ~k ; the Ca-embeddings
W V (n,k) ]HV* (n+q,k) and ]HV1 (n,k) ]HV 1 (n+q,k+q) ; and finally, the
Ca-submersion lV'(n,k) H V ’ (n,k-q) .As before, KV'(n,0) reduces to
a point, while I V ’ (n,1) is just ^ 0. The manifold nHVf(n,n)
consists of all nondegenerate linear transformations of lHn , is a group
with group operation o, and is usually denoted by GL(n,JH) .
 177

2. Grassman Manifolds

1. Let nRG(n,k), or simply G(n,k), 0 s k £ n, be the set

of k-dimensional linear subspaces (or k-planes passing through 0 ) of
JR . If y is such a plane, we let denote the collection of planes
in G(n,k) whose projection on y is nondegenerate. Let us fix an
orthonormal basis e = {e^ , ...,e^. } of y and complete it to an ortho­
normal basis of IRn , adding a frame c = {e ,...,e , }• For each
i n K
y’ € there exists a unique frame u ^ ,...,uk in y' which projects
onto e. Expressing the vectors u^ ,...,u^ in terms of the basis
e 1 ,...,ek ,e 1 ,...,En_k , we obtain

u i = e i + £s = 1 C isGs (i = c is €IR)-

This construction yields a map cp(e,e) from the set U into the space
k (n-k)
HR of real kx(n-k)-matrices. Obviously, <p(e,e) is invertible,
and the collection of maps <p(e,E) obtained from all the pairs e,e
is a Ca-atlas of the set G(n,k). Since any two planes in G(n,k) are
contained in some set , the topology defined by the atlas (<p(e,e)}
is Hausdorff. Moreover, this atlas contains finite subatlases (for
example, the subatlas consisting of the charts cp(e,e) where e and
c are subframes of the standard frame ort^,...,ort of 3Rn ), and
hence the above topology has a countable base. Thus, the atlas (<p(e,e)}
transforms G(n,k) into a k(n-k)-dimensional Ca-manifold, called the
Grassman manifold.
Clearly, G(n,0) and G(n,n) both reduce to points, and
G(n,1), as a set and a topological space, is identical to lRPn So
we see that the space ]RPn 1 has a Ca-structure compatible with its
topology, and hence lRpn is an (n-1)-dimensional Ca-manifold, We
remark that this (^-structure may be described directly and
conveniently as follows: for k = 1, the finite atlas of the manifold
G(n,k) given above consists of n charts cp^ : -> IRn 1 f...,
co : U -> nRn \ defined in homogeneous coordinates by the formulas
n n

U . = { (x : ...:x ) ] x . t 0},
i 1 n i

CPi ( (X 1 : . :xn ) ) = ( x ^ x . .... X . ^ / X ^ X . ^ / X , , . . . ,x i^ "

It is also clear that relative to this Ca-structure the canonical

cl
homeomorphism IRP -> S (see 1.2.5.6 ) becomes a C -dif feomorphism.
 178

2. Obviously, we may modify the definition of the manifold

G(n,k) by replacing the nonoriented planes with oriented ones. More
precisely, G(n,k) is replaced by the set G + (n,k) of oriented
k-dimensional planes (oriented k-planes, for short) of lRn passing
through 0. One has to modify the set accordingly and take it to
be the collection of all planes in G (n,k) whose projections onto the
plane y € G+ (n,k) are nondegenerate and orientation preserving. The
maps <p(e,e): U -*nRn ^n ^ are defined as in 1 and again they form a
a ,
C -atlas possessing finite subatlases. Since any pair of points of
G+ (n,k) is covered by a set of the form U U-, where the plane Y
differs from y only by its orientation (obviously, H u- = 0 ) ,
the topology defined by the atlas (cp(e,e)} is Hausdorff. The
k (n-k)-dimensional Ca-manifold so obtained is termed the upper Grassman
manifold.
The manifolds G+ (n,0) and G+ (n,n) are canonically homeo-
morphic to S^, while G+ (n,1) is canonically diffeomorphic to Sn 1 :
under this diffeomorphism each point x € Sn 1 goes into the oriented
line defined by the pair of points 0 ,x.

3. If one associates to each plane y - G(n,k) its ortho­

gonal complement , one obtains a mapping of G(n,k) onto G(n,n-k)
which is clearly a Ca-diffeomorphism. A Ca-diffeomorphism
G (n,k ) -+ G (n,n-k) is similarly defined, if the orientation given to
1
the orthogonal plane y and the orientation of y behave in accordance
to the classical rule: let a basis of y"*~ which is compatible with this
orientation be written to the right of a basis of y which is compatible
with its orientation; then the resulting basis of lRn should be
compatible with the standard orientation of ]Rn (see 1.3.10). In
particular, G(n,n-1) = nRPn ” 1 and G + (n,n-1) = Sn _ 1 .
The inclusion lRn lRn+q induces obvious Ca-embeddings
G(n,k) G(n+q,k) and G + (n,k) -* G+ (n+q,k). Furthermore, the formula
q 3
Y y x JR defines C -embeddings G(n,k) G(n+q,n+k) and
G + (n,k) -> G+ (n+q,n+k) (the orientation of the product y x is
defined by the orientation of its factors - see 1.3.7). These
embeddings and the previous diffeomorphisms form the following
commutative diagrams:

G(n,k ) -------- > G (n+q,k ) G + (n,k) > G + (n+q,k)

G(n,n-k) --- > G(n+q,n-k+q) f G+ (n,n-k) --- ► G + (n+q,n-k+q) .

 179

The map G+ (n,k) G(n/k) which takes each oriented plane

and "forgets" its orientation is clearly a Ca-submersion such that the
preimage of each point of G(n,k) consists of just two points. For
k = 1we recover the projection Sn _ 1 lRPn ~ 1 .
If to each frame from V(n,k) we associate theoriented
plane that it spans,we obtain a ca-map V(n,k)G (n,k). This is
a submersion, a fact that can be readily checked with the aid of
Theorem 1.5.9, if for a given frame v^ £ V(n,k) one explicitly
indicates the neighborhood V of the oriented plane y° spanned by
v and the map g: V V(n,k) which are needed in this theorem. One
can take V= Un (see 3) and define g(y) for each y £ V to be
y 0
the frame which is obtained from v after projection on y and
standard orthogonalization. The preimage of an oriented plane
y £ G + (n,k)under this submersion is the set of all orthonormal positive
k-frames of yand, in particular, is diffeomorphic to SO(k).
The map of V(n,k) onto G(n,k) which takes each frame into
the nonoriented plane that it spans is also a submersion. In fact, it
is exactly the composition of the two previous submersions. The
preimage of a plane y £ G(n,k) consists of all the orthonormal frames
of y and, in particular, is diffeomorphic to 0 (k).
Finally, the maps V'(n,k) G (n,k) and V'(n,k) + G(n,k),
which transform each frame into the plane that it spans are both sub­
mersions. The preimages of the points of G + (n,k) and G(n,k) under
these maps are dif feomorphic toGL+ (k,]R) and GL(k^IR), respectively.

4. Since the manifolds G(n,k) and G + (n,k) are continu

images of V(n,k), they are compact and, excepting G+ (n,0) and
G+ (n,n), connected.
The family of manifolds G(n,k) (k ^ 0,n) contains both
orientable and nonorientable manifolds. More precisely, for k i 0
G(n,k) is orientable if n is even and nonorientable if n is odd.
To see this, use the atlas made of the charts ip(e,e), where e and
e are complementary subframes of the standard basis of lRn , with the
order of vectors they inherit from this basis (cf. 1). Denote the
indices of the vectors of e by j ^ (e) , ...,j^ (e) (1 £ j (e ) < ... <
< £ n) , and say that two charts cp(e,e) and (p(e',£') of the
K.
atlas are contiguous if there exists I such that j^(e') = j^ ( e ) ± 1
and j (e') = jp<e ) f°r P / Obviously, for k / 0,n we can
always exhibit contiguous charts, and actually any two charts in the
atlas may be connected by a finite chain of charts such that each two
neighboring charts in the chain are contiguous. Suppose that cig and
c' are the coordinate functions of two contiguous charts, ip(e,e) and
is
 180

cp ( e 1 ,£ 1 ) ,
such that j ^ ( e f )= j £ (e > + 1- Thenit: is readily seen that
the coordinate c^ with m= -i + 1 does not vanish on the
intersection of the supportsof the two charts, but takes all the
remaining real values on this intersection. A simple computation shows
that

(c. Co - c. Co ) c n \ if i * s m,
is im lm Is Im
if i = 1, s m,
c ls°lm '
c!
is if i t I, s = m,
cimC£m'
if i -
I, s = m,
,

and thus the Jacobian J (cp(e,c ),cp(e ',c 1 )) [see 1.3.1] is equal to
(-1)nc ^ . If n is odd, this Jacobian takes both positive and negative
values; hence the manifold G(n,k) is not orientable for such n. For
k[j (e)+. ..+jk (e)]
n even, the formula tp(e,e) (-1 ) defines a map of
the atlas under consideration into S°, which satisfies the
compatibility condition in 1.3.1: this condition obviously holds for
contiguous charts, which in turn implies the compatibility for non­
contiguous charts. Hence, for n even G(n,k) is orientable.

5. The next constructions produce Ca-embeddings of G(n,k)

and G+ (n,k) in Euclidean spaces; see also 4.8.
Let us start with G + (n,k). For a matrix v £ V 1 (n,k), let
M. . (v) be the minor constructed from the rows with indices
1 1 ’*•1k
i 1 /•••/i-i_/ and put
. (v) = M. N.
. (v)/y (v) , where y(v) is
1 K k 1 *‘ k1 **
the positive square root of the sum of the squares of all minors having
maximal order. The functions N. . : V* (n,k) -*IR are clearly
i. . . . i,
1 2
analytic. Moreover, if two frames in V'(n,k), v and v , span the
1 2
same k-plane, then N . (v )=N. .(v ).Therefore, the map
( „ , X m1 • • • «1-1k 1 A ♦ ♦ •-L,
1 k
N: V'(n,k) 3R with coordinate functions N. . is the
il...ik
composition of the canonical submersion V'(n,k) ^ g (n,k), defined in

3, with a C -map g+ : G+ (n,k) -* 3R (see 1.5.10). We next show that

g+ is a C -embedding.
Using Theorem 1.5.4, it suffices to verify that g+ is an
injective immersion. To demonstrate that g is injective, one has
1 2 1 + 2
to show that if v ,v € V'(n,k) and N (v ) = N(v ), then the
n* 2 k-matrix constructed by adjoining the matrices v1 and v2 has
rank k. But this is plain, because any (k+1)x(k+1)-minor of this
 181

nx 2 k-matrix, constructed from k columns of

and one column of v
is equal to zero (expand the minor with respect to the column of v ).
To verify that is an immersion, it is enough to show that at each
point Yq £ G+ (n,k), the rank of the differential g+ is k(n-k).
This in turn will be true if we can prove that at each point
Vq £ V' (n,k) , the differential d N has rank ^ k(n-k). Finally, to
0
obtain this property of d
N, we prove that at each point
vo
Vq £ V* (n,k), the differential d M has rank ^ k(n-k) + 1, where
vo
(k)
M: V' (n,k) -* 3R is the map with coordinate functions M. . . So
i 1 ...ik
let A be any k Kk-submatrix of the matrix v £ V*(n,k) such that A
is nondegenerate for v = v 0n /' and let v.. be an element of A
1 o^ 0
such that its cofactor, call it does not vanish for v = v
0*
Next isolate those minors M. (v) of the matrix v which have at
x r . . lk
least k-1 rows in common with A, and then form a submatrix of the
Jacobi matrix of the map M as follows. Take the derivatives of the
chosen minors with respect to those elements of the matrix v which do
not appear in A and the one derivative with respect to v. . . We
1 oDo
obtain a square matrix of order k(n-k) + 1 which has, in a
neighborhood of the point V q (and for a suitable arrangement of the
rows and columns) the form

'(At) 1det A 0 0

0 (At )”1detA . 0

0 0 .. (At) 1det A

^k(n-k)

0 .......... a

where t indicates transposition. The determinant of this matrix is

equal to a (det A) (n-k) ( k _ 1 * and therefore does not vanish.

Now let us turn to G(n,k) and compose the embedding g+

(k) (k) < (k) + 1 > / 2
with the map q: TR + 3R / defined as q (x1 , ...,x j-nj ) -

= <X' , X l X 2 X 1X (n)'X 2'X 2X 3 X 2 X (n} X (g)-1 X (g),X(g)>'

 182

Clearlv, the restriction q is aCa-immersion, and q(x) = q(y)

1 rr^
]R^k ^ \ 0
if and only if x = ±y. Since the equality g + (y') = ~9+ (y ) holds
for planes y,y' £ G+ (n,k) which coincide geometrically (but have
opposite orientations), we see that q ° g+ is a C -immersion of

G+ (n,k) into IR . Moreover, q o g + (y') = q o g + (Y) if

and only if y and y' coincide or differ only by their orientations.
(k) ((k) + 1 ) / 2 _
Therefore, the map fact(q ° g + ) : G(n,k) + IR is well
defined and is a Ca-embedding.

6 (INFORMATION). The coordinate functions gT . of the

(k) r " k
map g+ : G+ (n,k) IR satisfy the relations

+ + rk + + = 0 .

V - V V - A " Zs = 1 gir - - ik-1jsg3 l - - ^ s-1ik js + r - -

Considering these relations as equations relative to the coordinates in

(k) (k)
IR , they define a subset of IR which is exactly g (G+ (n/k)).
See [9] for details.
The numbers
gt (y) are known as the Grassman-Pliicker
il...ik
coordinates of the oriented plane y.

The Complex and Quaternionic Cases

7. To obtain the complex version of the manifold IRGi^k),

one has to take k-dimensional planes of (Cn passing through 0 instead
of k-dimensional planes of IRn passing through 0. The result is a
complex manifold (EG(n,k) of dimension k(n-k), called the complex
Grassman manifold.
Obviously, (CG(n,0) and (LG(n,n) reduce to points, whereas
CCG (n,1) coincides, as a topological space, with (CPn ~ 1 . Thus we equip
the projective space (CPn 1 with a complex structure compatible with
its topology, which makes (EPn 1 into an (n-1)-dimensional complex
manifold. This structure may be given in homogeneous coordinates just
as we described the Ca-structure of ]RPn“1 . Relative to this structure
1 2
the canonical homeomorphism (CP -* S (see 1.2.5.6 ) becomes a
Ca-diffeomorphism.
The complex analogues (EG(n,k) -+ (CG(n,n-k),
CCG (n ,k) -> CCG (n+q,k) , CCG(n,k) + CCG (n+q ,k+q) , (CV(n,k) +CCG(n,k), and
 183

(EV1 (n,k ) -* (EG(n,k) of the maps G(n,k) G(n,n-k), G(n,k) G(n+q,k),

G(n,k) -* G (n+q ,k +q) , V(n,k) -* G(n/k ) / and V'(n,k) + G(n,k),
constructed in 3, are defined in an obvious way. The first is a
biholomorphic map, the second and the third are holomorphic embeddings,
and the fourth and the fifth are Ca-submersions. We mention also the
Ca-embedding (CG(n,k) ->• G + (2n,2k): when we transform (En into IR^n ,
every k-plane becomes an oriented 2 k-plane.
The manifold (EG(n,k) is the image of (EV(n,k) under a
continuous map, and as such it is compact and connected. Moreover, it

readily seen that ŒG(n,k) can be analytically embedded in ]R :

the composition of the embedding (EG(n,k) G (2n,2k) described above
l2kJ a
with the canonical embedding G+ (2n,2k) JR is a C -embedding.

8. Substituting quaternions for complex numbers in these

definitions, we obtain a connected 4k(n-k)-dimensional Ca-manifold
HG(n,k), called the quaternionic Grassman manifold. For k = 0,n,
1G(n,k) reduces to a point. For k = 1,n-1, 3HG(n,k) is topologically
the quaternionic projective space H P n 1. Thus 3HPn 1 becomes a
Ca-manifold, and the canonical homeomorphism 3HP1 S^ becomes a
Ca-diffeomorphism.
The quaternionic analogues of the maps described in 3 and 7
are Ca-maps. In particular ,there is a canonical Ca-embedding
KG(n,k) G (4n,4k). Composing it with the canonical embedding
* ( I K) a , (Ï)
G+ (4n,4k) TR , we obtain a C -embedding 3HG(n,k) JR

9. The maps

S2 n _ 1 = (EV (n ,1) -> ŒG (n,1) = (EPn_ 1

and
S4 n _ 1 = 3HV (n ,1) ->HG(n,1) = ! P n_1,

which are particular cases of the canonical maps in 7 and 8 , obviously

coincide with the corresponding Hopf maps (see 1.2.5.8 ). As a result,
15 8
we see that these Hopf maps are Ca-submersions. The Hopf map S + S
is a Ca-submersion too, a fact that follows directly from its definition.

Projective Cayley Plane

10. The projective Cayley plane is also equipped with a

^-structure which transforms (EaP^ into a Ca-manifold. We next
 184

describe this structure.

Identify ]R16 with CEa2 , and 1R1 7 with la 2 x 1R, and then
define three maps -+ by the formulas

(yr y 2) 1+ (2(1 ~p2 ) 1//2y 1 »2 (1 - p2 ) 1 ^ 2y 2 , 2 p 2 - 1)

(y1/Y2) ^ (2y -jy 2 » 2 (1 - p2 ) 1 / 2 Y i ,1 - 2 | y 1 |2 ),

( y 1#y 2 ) h- (2(1 - p 2 ) 1 / 2 y 2 , 2 y 2y r 1 - 2 |y 2 I2 ) /

2 2 1/2
where y ,y2 € (Ea and P = ( | | + |y2 ! ) <the first map is just
DS, but we do notuse this fact explicitly). These three maps yield
16 16 16 16
a map F : D S x S x S , and one can easily verify that

16 2
zer(F) = zer[pr: D (EaP ].

Sincethe injective factor of F is atopological embedding

of the quotientspace D /zer(F) into S * S x S (see 1.1.7.10),
andsince, according to the last equality, thisquotient space
16 2 2
coincides with D /zer(pr) = (EaP , we canidentify (EaP with
16 2 a
F (D ). To transform (EaP into a C -manifold, it suffices to show
that F(D^) is a Ca-submanifold of S^ x S^ x S^.
16 16
Define f,g: S \ orti7S as

f(u 1 ,u2 ,t) =

u2 ^ 1 - u2 2
(^ _ t >u1 , Y
I I
t - t)

and _ 2 [u^ ,u2 € (Ea, t € 3R]

- u 1 u 2 lU ll
g(u 1 ,u2 ,t) = (u2 ,-1 t >1_ t - t) ,

and consider the three mapsh ^ h ^ h ^ : S \ orti7 "* S x s x s

given by

h 1 (x) = (x,f (x) ,g (x) ) , h 2 (x) = (g (x) ,x ,f (x) ) ,

h 3 (x) = (f(x),g(x) ,x).

Since f and g are analytic, h^ ,h2 and h^ are analytic embeddings,

One can check directly that

F(D16) 0 pr~ 1 (S16 \ ort17) = hj^iS16 ^ ort1?).

Therefore, the left-hand side intersections are Ca-submanifolds of

16 16 16
S x S x s , and it remains to note that

Ui pri 1 *s 1 6 ^ ort17* = ^s ' 6 x s 1 6 x S 16) ^ (ort1 7 ,ort1 7 ,ort1 7),

 185

and that (ort1 7 ,ort1 7 ,ort ) j£F(D16).

Noncompact Grassman Manifolds

1. We denote by 3RG'(n,k), or simply by G 1 (n,k)

(0 £ k £n) the set of allk-dimensional planes of IRn , i.e., the
planes need not pass through 0. It is clear that given a plane
Y 1 £ G 1 (n,k), there is a unique (k+1)-plane y in IRn+ 1 that passes
both through 0 and the k-plane which results by translating y1 by
the vector ortn+^. Moreover, the formula y *^ Y defines an injective
mapping G' (n,k)G(n+1,k+1), which has an open image. So we may
regardG f(n,k) as an open subset of the manifold G(n+1,k+1). In
particular, G 1 (n,k) is a (k + 1) (n-k)-dimensional camanifold. We call
it the noncompact Grassman manifold.
G ’(n,k) can be mapped naturally onto G(n,k): for each
plane Y 1 £ G*(n,k), consider the parallel plane y £ G(n,k). It is
a straightforward consequence of Theorem 1.5.9 that this map is a
Ca-submersion (one can take the entire G(n,k) for the required
neighborhood V of Y in G(n,k), and one can take the map which
takes each plane from G(n,k) into the parallel plane passing through
an arbitrary, but fixed point of y 1, for the required g). The
preimage of a plane y E G(n,k) under this submersion is the set of
all k-planes of 3Rn which are parallel to y, and is canonically
diffeomorphic to the orthogonal complement y"1"of y (a unique k-plane
parallel to y passes through each point of y "*") •
The oriented, complex, and quaternionic versions of these
definitions are immediate.

3. Some Low-Dimensional Stiefel and Grassman Manifolds

1. S0(3) is canonically Ca-diffeomorphic to IRP^.

3 3
We let shi denote the map of 1R into JR^ which takes
(y 1 'Y 2 'Y 3 ^ into ( 0 f Y y Y 2 ' y 3 ) • The canonical C -diffeomorphism
^ 3
]RP + SO (3) takes the line of 1RP passing through the point
4 3 3
x 6 ]R '-■0 into the orthogonal transformation cp: JR HR defined by
.-1 . -1 '
the quaternionic formula tp(y) = shi (x shi(y) x ) . The inverse
3 3 3
dif feomorphism S0(3) ®P takes each transformation ip: HR 3R
into the line consisting of the quaternions of the form
 186

q - shi (cp(ort )) q ort 2 - shi (cp(ort2 ))q ort 3 - shi (cp(ort3 ) ) q ort4 ,

where q is an arbitrary quaternion, and ort2 , ort^ , and ort^ are

regarded as quaternion units. It is routine to check that quaternions
of this form describe precisely a line and that the constructed maps
3 3
KP -> SO (3) and SO (3) -+ 3RP are inverses of one another.
3 2
2. IRV(4,2) is canonically C -diffeomorphic to S * S .
3 2
The canonical Ca-diffeomorphism S x S -*3RV(4,2) takes
3 2
the pair (x,y) E S x S into the frame {x,x shi(y) } .

3. S0(4) is canonically Ca-diffeomorphic to xS0(3).

The canonical Ca-diffeomorphism x S0(3) SO (4) is

defined by the quaternion formula (x,{y,z}) {x,x shi(y) ,x shi(z) }
(here the points of the manifolds S0(3) and S0(4) are interpreted
as frames).

4. G+ (4,2) is canonically Ca-diffeomorphic to S^ x S^.

The canonical Ca-diffeomorphism G+ (4,2) S^ x s^ takes

the oriented plane spanned by the frame {x,y} E V(4,2) into the pair
-1 -1 -1 -1
(shi (xy ),shi (x y)). The inverse diffeomorphism transforms each
2 2
pair (u,v) E S x s into the two-dimensional plane consisting of
quaternions of the form shi(u)q + q shi (v) , where q is an arbitrary
quaternion. Again, it is routine to check that the pair
-1 -1 -1 -1
(shi (xy ),shi (x y)) is uniquely determined by the oriented plane
spanned by the frame {x,y}, that the quaternions shi(u)q + qshi(v)
fill exactly a two-dimensional plane, and that the
2 2 2 2
maps G+ (4,2) -* S x s and S x s ■+ G + (4,2) constructed above are
inverses of one another.

4. Exercises

1. A homogeneous polynomial in n+1 variables and with

real (complex) coefficients is nonsingular if there are no points in

in 3Rn + 1 ^ 0 (respectively, in (En + 1 \ 0) where all its partial

derivatives vanish. Show that the projection 3Rn + 1 \ 0 -»■]Rpn
(respectively, a:n ”* \ 0 ->- IPn )transforsm the set of zeros different
from 0 of such apolynomial into a submanifold of IRPn (respectively,
(EPn ) .

2. Let P (x.j »x 2 ,x^) be a nonsingular homogeneous polynomial

 187

of degree k with real coefficients. Show that the submanifold of the

2
projective plane 3RP defined by the equation pix^x^x^ = 0 has an
orientable neighborhood if and only if k is even.

3. Let p(x^/X 2 /X ) be a nonsingular homogeneous polynomial

of degree 3 with real coefficients. Show that the submanifold of
2
3RP defined by the equation p(x ,x0,x ) = 0 is homeomorphic to
1 1 i i 1
either S or S J |_S , and that both cases are realized.
2 2 2 2
4. Show that the equation x +x +x =0 defines in (EP
2
a submanifold homeomorphic to S .

5. Show that the equation x3 + x 93 + x 3 = 0 defines in (EP2

1 1
a submanifold homeomorphic to S x S .
2 2 2 2
6. Show that the equation +x0 + x
x +x = 0 defines in
3 2 2
(EP a submanifold homeomorphic to S x S .

7. Show that nRG(n,k) ((EG(n,k)) admits a Ca-embedding in

(k)-i
JRP (respectively, (EP ).
n2
8. Show that the map HRG(n,k) R which takes each plane

Y E lRG(n,k) into the matrix of the composite map ]Rn — y —1-n-» 3Rn
(where pr is the orthogonal projection) is a Ca-embedding. Show that
2
the same is true for the map (EG(n,k) ■+ (E which takes each plane

y E (EG(n,k) into the matrix of the composite map (En Pr > y - (En .

9. Show that nRV(8 ,k) is Ca-diffeomorphic to S^ x JRV(7,k-1)

y
(1 ^ k £ 8 ) . Show that (EV(4,k) is Ca-diffeomorphic to S x (EV(3,k-1)
(1 k <: 4) .

§3. A DIGRESSION: THREE THEOREMS FROM CALCULUS

1. Polynomial Approximation of Functions

1. The purpose of this section is to state and prove thre

theorems from calculus, namely theorem 4, 2.3, and 3.3. They differ in
character and we grouped them together here because all three are needed
in this chapter, and none of them is included in the traditional
calculus course.
 188

The main theorem of this subsection, Theorem 4, is a corollary

of Lemma 3, whose proof, in turn, requires Lemma 2.

2 (LEMMA). For any positive 6 < 1

1 d - t2)kdt = 1
lim ,
k^°° k -6
where
■1 9 1c
(1 - t ) dt .
-1

This is an easy consequence of the inequalities

1 (1 _ t 2 ,kdt , 2. (1 - M 1 - 6'
0 < 1 L
ak J- 6
1 - 4

The left inequality is plain, while the right one follows from the
estimates
2 V
(1 - t )Kdt = 2 (1 - t2)kdt < 2 ( 1 - 6 ) (1 - 6 2)k
ak -
-6
and
6/2
(1 - t2)kdt > 6 ( 1 - 4 p)k .
- 6/2

3 (LEMMA). There exists a sequence of mappings

. n T\ 00
{p^: C (I ,1R) ^ C OR ,3R) such that:

(i) is a polynomial for any f E C(in ^R) and any k;

(ii) d-f f equals 0 on Fr I , then the sequence

{pk (f) n^ converges uniformly to f;

(iii) _if f equals 0 on Fr ln and has continuous

partial derivative D^f, then

Pk (Dif) = D.pk (f) .

PROOF. For f € C(ln ,lR) we set

1 n — ,

tp (f) ] (x ,. ..,x ) = — f (t , ...,t )T~ft 1 - (t.-x.) ] dt .. .dt , (2 )

1 n a£ J n 1 n j= 1 11 D 1 n

where a^ is defined by (1). This clearly defines a polynomial and

therefore we must check only properties (ii) and (iii).
To check (ii), we extend the function f to lRn , setting
f (x) = 0 for x E IRn \ ln , and denote by M the maximum of If (x) I.
 189

Given an arbitrary e > 0, we can find 6, 0 < 6 < 1, such that

|f (xjj , ...,x^) - f (x1 , ...,xn ) | < e/2 for |x Jj " | < 6,...,
|x^ - xr | < 6 . Moreover, we can find a number K such that

? v n
1 - (1 - tz)Kdt
4M

for all k > K (see 2). We next show that |[p^(f)](x) - f(x) | < g
for x E In and k > K.
Write [p^(f)](x) - f(x) as

n
n (f(t1+x1,...,tn +xn ) - f(x1,...,xn ))1 fn - tj)kdt1...dtn
n j =1

and then replace the integrand by its absolute value. Now divide the
new integral into two integrals, one over the cube [— 6 ,6 ]n and one
over its complement [—1,1]n \ [-6 ,6 ]n . We obtain

|[p. (f)] (x) - f (x) I *

n
1 f (t1 +x 1 , ...,tn +xn ) - f (x1 ,...,xn ) |J 7 ( 1 “ fcj) dt>| • ••dt^
n
n j= 1
k [-6 ,6 ]
n
n f (t1+x1 ,. . .,tn +xn) - ,xn ) I T TH - fcj) dtr ..dtn
j= 1

The first term is smaller than e/2, since

|f(t1 +x^,. .. ,tn +xn ) - f (x1 ,...,x^) | < e / 2

for (t^...,^) E [-6 ,6 ] and

n
n 6
1
- t")"dt . . .dt (1 - t2)kdt < 1
n n k J-6
ak 6],n j=1 3

The second term is also smaller than e/2, since

|f (t1+x1,...,tn+xn ) - f (X l ,...,xn ) I £ 2M
and

(1 - t2)kdt,...dt
n j 1 n
,n j= 1
*k [-1 ,1 ]n^t- 6 ,6 ]

n n
(1 - tj) dt 1 ...dtn n (1 - t?
t.))^dt 1 .-.dt
n j 1 n
n j= 1 n j= 1
‘k [-1,1] k [-6 ,6 ]
 Therefore, |[p^(f)](x) - f (x) | < e .

To check property (iii) , replace f by D^f in definition

(2 ) and then integrate by parts with respect to t^ in the right-hand
side. We obtain

[p^ (D^f)] / • • •/ ) =

_1_
n f < t , ............‘ „ i f t : f r « ’ - (tj-x i 2)kd t 1. . . d t n . ( 3)
n 1 3 = 1 J J
ak i

Since

91 . I \ j j 9x . I ' j j
1 3=1 J J 1 J=^ J J

the right-hand side of (3) is equal to [D .p, (f)](x ,...,x ).

1 .K. 1 n
4. Suppose X is a compact set in lRn and f is areal
function defined and of class C in a neighborhood ofX. I_f r£
then for any e > 0 and any nonnegative integer s £ r, there is a
polynomial g : lRn JR such that
s s s s
max Id. ...D ng(x) - D. ...D nf (x) I < e
x€X n 1

for any collection s^,...,sn of nonnegative integers with

s -+ •..+s £ s.
1 n
PROOF. Clearly, one can assume that X <= int In . Denote by
U the neighborhood of X mentioned above, and let 6 : lRn 3R be any
Cr-function equal to 0 on Dn and equal to 1 outside the concentric
ball of radius 2. For y € (In ^ U) U Fr In , we let d(y) denote
the ball of center y and radius Dist(y,X)/4. Cover (ln \ U) U Fr In
by a finite number of balls Intd(y) , say Int dty^ ,...,Intd(y ) ,
and define h: in 1R as
P
'£f ( y)
y )~| |B(4(y-y )/Dist(y ,X)), if y C U,
i= 1 1 1
h (y) =
0/ if y £ U.

Then h is of class C , agrees with f on X, and vanishes on a

neighborhood of Fr ln . This shows that one can take g to be p (h) ,
K
with p^. as in Lemma 3 and k large enough.
 191

2. Singular Values

1. In the present subsection we assume that we are given

open subset U of JRn and a C°°-map f: U 3Rq with q Ss 1 . We let
F denote the set of points in U where the rank of the Jacobi matrix
of f is less than q. Our aim is to prove Theorem 3, which is needed
in the next section (see 4.7.4).
We need two auxilliary notations: f^ for the j-th coordinate
function of the mapf (j = 1 ,...,n) and Fg for the set of points in
U where all partial derivatives of order1,...,s of the functions f
vanish. Clearly, Fg is closed in U and F => F^ d F^ ^ ....

2 (LEMMA) . Suppose s > (n/q)-1 and C is any compact part

of the set F . The image f(C) is nowhere dense.

It is enough to show that for any n-dimensional qube Q ci U,

the set f (C fl Q) is nowhere dense. Indeed, one can cover C by a
finite number of such cubes, and use the fact that a finite union of
nowhere dense sets is nowhere dense.
Let a be the edge length of Q. Consider the standard
partition of Q into mn small cubes of edge length a/m (with m
an arbitrary positive integer). Let Q' be a small cube in this
partition which intersects C. Now apply Taylor’s theorem to the
functions fj and use the fact that their partial derivatives of order
s+1 are bounded on Q to show that there is a constant b such that

dist (f (x) ,f (y)) £ bfdist(x,y)]S + 1

for any x € F H Q,£ y £ 0. Since Q' has diameter a/n/m, we see

s s+1
that f(Q') is contained in a ball of radius b(a/n/m) . Therefore,
f(Q1) is contained in a q-dimensional cube with edge length
2b (a/n/m) s + \ and f (C n Q) - in a union of no more than mn such
cubes. The volume of each such cube is [2b(a/n/m)S+1]q , and so the
sum of their volumes does not exceed

mn r , r~ / ,s + 1 ]
[2 b (a/n/m) iq = cmn-q (s+ 1 ) ,

where c is independent of m. By hypothesis, this sum goes to 0

as m ■+ °°. This shows that the set f (C n Q) can have no interior
points. Since the latter is a closed set, it is also nowhere dense.

3. The image f(C) of any compact part of the set F .is

 192

nowhere dense.
We proceed inductively on n. For n = 0 there is nothing
to prove; hence it is enough to show that the theorem holds for
n = k +1 if it holds for n = k.
We start with the special case F^ = 0. Since C is compact,
it suffices to find, for each point x E C, a neighborhood V in U
such that f(C n V) is nowhere dense. D, Assume that
f (x) t 0
K 1 ? k +1
(we can always achieve this by reindexing the coordinates in nR and
k +1
]r") and consider the map g: U -> JR defined by the formula

y = (^ 1 .... yk + 1 > " ( Y 1 Yk'fq (Y))*

^ Its Jacobian at the point x does not vanish (being equal to

Dk +lfq); hence 9 is a C°°-diffeomorphism of a neighborhood W of x
onto a neighborhood of g(x). Now one can take as V any neighborhood
of x with compact closure included in W. To see this, set h to be
the composite map

g(w) w ^abf^q.

It transforms each point from g(W) into a point which has the same
last coordinate. In particular, for any real number u, h can be
compressed to a map

g (W) n pRk x u ] +]Rq ” 1 x u.

k k
If we identify g (W) fl [3R x u] with its orthogonal projection on JR
in the standard fashion, and JR^ 1 x u - with its orthogonal projection
Q—1 oo k cf— 1
on JR, we obtain a C -map h^ of an open subset of JR into JRM
Clearly, the Jacobi matrix of hu at the point (y1 If-*-/yvK.) is
obtained from the Jacobi matrix of the map h at the point
(y^l *•••r r u ) by deleting the last column and the last row, which has
the form 0,...,0,1. Therefore, the rank of the first matrix is less
than q-1 if and only if the rank of the second matrix is less than q,
i.e., if (y^ ,...,y^. ,u) E g(F n W). Applying the induction hypothesis
to h , we deduce that the intersection of h(g(C n CIV)) with each
hyperplane JR^ 1 x u is nowhere dense in IR^ 1 x u. But if this is
the case, h(g(C fl Cl V) ) has no interior points in 3R^ and we need
only note that this set is closed and coincides with f(C fl Cl V) .
Now let us turn to a second special case: C cF and
Fs+1 = ^ (for some s). Again, it is enough to exhibit for each point
x E C a neighborhood V in U such that the set f (C D V ) is nowhere
dense. Let cp be a derivative of order s of one of the functions f.
3
 193

which satisfies the following condition: one of the derivatives D.tp,

say D]C+ '|(P' does not vanish at the point x. Consider the map
g:U + ^ defined as

y = (y1 / • • • /yk+1) » (y-j / • • • / y k / cp( y) ) .

Its Jacobian does not vanish at x (being equal to D, cp) ;hence

oo K+1
g yields a C -diffeomorphism of a neighborhood W of x onto a
neighborhood of g(x). We show, with the aid of the composite map

/T7v (ab g) __ ab f ...

g (W) — ---- >W > HR (4)

that one can take any neighborhood of xwith compact closure included
k
in W for V. To do this, note that g(C) c: IR andrestrict the map
k cr
(4) to a map h: g (W) n 3R HR . Clearly, all the derivatives of order
£ s of the coordinate functions of h vanish on g(C n W). Using
the induction hypothesis, it is evident that h carries the compact
parts of the set g(C n W) into nowhere dense sets. Finally, we
observe that g (C fl CIV) is a compact part of g(C n W) , and that
h (g (C fl CIV)) is just f (C fl Cl V) .
At last, wecome to the general case. According to Lemma 2,
there exists a numberr such that the set f(C fl F^) is nowhere
dense. We shall prove statement 3 by induction on r, i.e., assuming
that f (C fl F ) is nowhere dense, we show that, for r = 1, f (C ) is
nowheredense, and for r > 1, f (C ni-s nowhere dense.
Let G be an open nonempty subset of 3Rq . Since C n F^ is
compact and f (C fl F )is nowhere dense, the set C n has a

neighborhood Nin lRn such that Cl N is compact and Cl N c= U ,

f(Cl N) ^ G. Next replace the map f by its restriction to U \ F^
and C - by the set
C \ N, if r = 1
C'
(C \ Fr ) \ N, if r > 1.

Now we are back to one of the cases covered by the first part
of the proof (namely, in the first case for r = 1 , and in the second
one for r > 1). Therefore, we conclude that f(C') does not cover
G ^ f(Cl N) . Consequently, if r = 1 the set f(C) does not cover
G, while if r > 1 the set f(C n F ) does not cover G. This
completes the proof, because f(C) and f (C n are closed.

4 (INFORMATION). In Theorem 3, the condition that f be

C°°-smooth is unnecessarily strong: in fact, the proof uses only the
 194

fact that f is of class Cr , with r = 2 + max(n-q,0) . A more

precise analysis shows that this r can be decreased by 1 (see, for
example, [2 1 ]), but no further (for q = 1 , this is showed in [23],
and the case q > 1 reduces easily to the case q = 1 ).

3. Nondegenerate Critical Points

2
1. Let f be a real C -function defined on an open subset
of nRn . A critical point y of f is nondegenerate if the second
differential of f at y (considered as a quadratic form) has rank n.
The index of the second differential of f at y (i.e., the number
of negative squares in the diagonal representation of this form) is
called the index of the point y and is denoted by ind^y.
2
We remark that if cp is a C -diffeomorphism of an open
subset U of IRn onto another open subset of nRn and y is a non­
degenerate critical point of f: U 3R, then cp(y) is a nondegenerate
critical point of the function f o cp ^ : cp(U) -* HR, and ind _>|<p(y) =
= ind^y. Both conclusions remain true in the more general
1 -1
situation where cp is only of class C / but the function f ° cp
2
is of class C -
Now consider the function JRn -> HR defined as

(x1 ,...,xn ) H- -Xl - ... - xk + xk + 1 + ... + xn + c , (5 )

where c is a real number (0 £ k £ n ) . This function has a unique

critical point, at 0, which clearly is nondegenerate and of index k.
The main goal in the present subsection is to show that, in a suitably
chosen system of coordinates, any sufficiently smooth function has the
above form (5) in the vicinity of a nondegenerate critical point.

2 (LEMMA). Let V be an open ball in IRn with center 0,

and let f : V IR be a Cr-function, r ^ 1, with f (0) = 0 . There are
r— 1 ~ '
C -functions f
---------------- |f n : V + n R ---------
such that

f(x) = E± = 1 xifi (x> ( 6;

for all points x = (x„,...,x ) e V.

-------- ------ 1 n
To prove the lemma, it is enough to set
r1
fi (x) D.f(tx)dt,
0 1

and then observe that (6 ) is an immediate consequence of the equality

 195

f (tx) = X£ = 1 x.D.f(tx) .

3. Suppose that y is a nondegenerate critical point of

a (^-function f defined on an open subset of !Rn . I_f r ^ 3, then
there exist a neighborhood U of_ y and a diffeomorphism cp of U
onto a neighborhood V of
such that the restriction f 0,
cp /5 )
coincides with the composite map U -> V ---- >1R, where k = ind^y and
c = f (y) .

PROOF. Without loss of generality, we may assume that y = 0

and f(y) = 0 . By Lemma 2,

f(x) = ^ =1 x.f.( X )
r —1
in some neighborhood of 0, where f^ are C -functions.
Differentiating, we obtain D^f(x) = x^D^f^(x), since f^ (0) =
= ... = fn (0). Again we apply Lemma 2 and write, in a neighborhood
VQ of 0,

f i<x ) = Ij=1 x.fij(x) (i = 1,...,n),

r —2
with C -functions f^j. Therefore, for x € Vq ,

fix) ■

where g ^ (x) = (f^ix) + f ji (x)}/2. Clearly, (0) = DjD.. f (0)/2 .

The subsequent constructions mimic the standard reduction of

a quadratic form to canonical form through linear transformations. For
p = 0,...,n, we construct neighborhoods V and W of the point 0
n r- 2 p p r- 2
in 1 , C -diffeomorphisms cp^: V^, and also C -functions

g?.: V -v 1R, i,j = p +1,...,n, such that: (i) W c V 1 ; (ii)

1 "2 P Ir ir
(ii) = 0; (iii) the composite map

V W V W v0 *
p p p- 1 1 0

is represented by the formula

2 2 yn p / , (7 )
> * “ ,* ••• ± x p * n 3 ,x,xix j -•

(iv) . Then we will have finished, since one could take

ij ji
c p " 1 ( . . . (cp‘ 1 ( V n ) ) . . . )
 196

for V, define tp as the composition

ab cp ab cp9 ab cp ab ip ab tt
V ---- ^ cp1 (V) -----1 cp2 (cp1 (V) ) -----i ... ----- > Vr ---- ► n (vn ) ,

where tt is a suitable permutation of the standard coordinates in IRn ,

and set U = cp ^ (V) .
The neighborhood VQ is already given. We let WQ = V Q ,

tp^ = id , q? . = q . ., and assume that we haveconstructed V , W ,

0 0 ID P P
(Pp, and g?j satisfying (i) ,(ii), (iii), and (iv) for p £ q. It
is clear that 0 is a nondegenerate critical point of the function (7)
iik 11 P
with p = q. Hence the matrix G =
g . .(0) I .is nondegenerate
id i,D=q+1
and there exists a nondegenerate (n-q)x(n-q)-matrix A such that the
left upper element of the matrix A^GA is not zero. Let t denote
E 0
the linear transformation of IRn having matrix where E
0 A

is the qxq-identity matrix. The composition of the diffeomorphism

-1
ab I : I (V ) V with the function (3) is given by

x h- ±x 2A ± ... ± x 2 + v).
n \
. h..(x)x.x.
1 q i ,D=q + 1 ID ID
where h . . = h.. and h - .1 (0) ^ 0 . Now consider the subset L of
ID D1 q+ 1 ,q+1
t (V ) consisting of all the points x where h 1 (x) f 0 and
q q +\ tq +I
has the same sign as h 1 1 (0) , and then define : L -> 3Rn as
q + I,q + l

^ ( x) = <X 1 ........... V ^ V l ^ 1 (X )i1/2' V 2 Xn > '

where

£ = X .
, v>
+ V
hs —/q+ 1 (x)
' 1 -----
q+1 s'>q+1 S hq+q>q+ 1 (x) ‘

A simple computation shows that the Jacobian of \]> at the

point 0 does not vanish. Therefore, the compression of ip to a
neighborhood M of 0 and to its image i|j(M) is a Cr- 2 -diffeomorphism.
It is now readily verified that the sets V +1 = (M) and W +1 = I (M) ,
the map <pq + 1 : Wq +1 -*■ Vq + 1 defined by (x) = ^ “ 1 (x) ) , and the

functions : Vq +1 -[R' defined as

q+ 1 ,q + 1

enjoy the properties (i)-(iv).

 197

§4. EMBEDDINGS. IMMERSIONS. SMOOTHINGS.

APPROXIMATIONS

1. Spaces of Smooth Maps

^r-
1. Let X and X1 be C -manifolds
(0 £ r £ a ) . Wedenote
27 r
by C (X,X 1 ) the set of all C -maps X X 1 . If r £ 03, we equip
r . r r
C (X,X') with the C -topology which makes C (X,X') into a topological
space, as follows. Given two arbitrary charts tp G AtlX and
tp1 E Atl X 1 , a sequence of nonnegative integers r^ ,...,r with
n = dim X and r^ + ...+r^ < r, a compact subset A of Im cp , and an
n*
open subset A' of IR , where n* = dim X 1 ,consider the subset of
27
C (X ,X 1) consisting of all maps f such that
r r
[D .. .D nloc (tp,tp1 )f ] (A) CI A 1.
1 n

These subsets form a prebase of the Cr-topology on Cr (X/X').

Clearly, C^(X,X') = C(X,X'), and the C^-topology is simply
the compact-open topology (see 1.2.7.1). Also, for s < r, the
r s
inclusion C (X,X 1) C (X,X 1) isobviously continuous. Another direct
27
consequence of the definition of the C -topology is that all the spaces
27
C (X ,X 1) are regular. In addition, we note that the sets closed
oo
(open) in C (X ,X 1)are exactly those setsclosed(respectively, open)
27
in all the C -topologies with r finite.
For each pair of Cr-maps f: Y X and f 1 : X'■+ Y 1 , there
is a map Cr (X,X') -y Cr (Y,Y*) defined by the formula g + f 1 o g o f,
anddenoted by Cr (f,f'). Obviously, C^(f,f') = C(f,ff) (see 1.2.7.1),
Cr (f,f') = a b C S (f,f') for s < r, and Cr (f,f') is continuous for
all r £ oo.
27
We list some particular subsets of
C (X,Xf) which are
27
important in the sequel. These are the sets of all C -embeddings,
r r r
C -immersions, C -submersions, and C -diffeomeorphisms X + X 1, and
they are denoted by Embr (X,X')/ Immr (X,X')/ Subm (X,X')/ and
Diffr (X,X 1) , respectively (1 £r < a). Moreover, we let C^(X,Xf)
denote the set of all Cr-maps f:X X' such that f (3X ) c= 3X' and
d f(Tang X) c Tang,- / * (3X') for all x EBX. Usually one writes
X X I (X)
 198

Diffrx instead of Diffr (X,X).

For 1 .< r ^ oof the map C^(X,X') -> Cr (3X,3X'), defined as
r r- 1
f *■> ab f is continuous, and the map C (X,X') C (Tang X,Tang X 1 ),
defined as fdf is a topological embedding.

2. If X is compact, then the set Imm (X,X1) is open in

c r (X,X ') (1 < r s “ ).
We have to exhibit, for a given Cr-immersion fq : X X 1,
a neighborhood of f^ in Cr (X,X!) consisting only of immersions.
To do this, pick for each point x E X two charts, cp^ £ Atl^X and
cp' e Atl X 1 , such that fn (supp cp ) c supp cp* and loc(cp ,cp')f equals
X U X X X X
one of the inclusions TRU -* ]Rn , IR^-> 3Rn , or 1R^ 3R^ (see 1.5.3
and 1.5.1; here n = dim X and n 1 = dim X ') . Now cover X with

a finite number of sets U = cp (Int D ) , say U , ...,U , and

x x x^ xg

denote by IL the subset of Cr (X,X')consisting of all the maps f

such that f (Cl U ) c suppcp' and the upper nxn-minor of the Jacobi
Xi i
matrix of the map loc (cp ,cp' )f has no zeros on D . The intersection
"i i
fl ... H Ug is the desired neighborhood of the map f^ .

3. If X is compact and X1 has noboundary, then the set

Submr (X,X') is open in Cr (X,X') (1 £ r £ °°) .
37
We have to exhibit, for a given C -submersion fn : X -> X ',
a neighborhood of f^ in C (X,X') consisting only of submersions.
Again, for each point x £ X we choose charts cp £ Atl X and
X X
cp1 £ Atl X 1 , such that fn (suppcp ) c= suppcp' and loc (cp ,cp1 )f equals
X U X X X X

one of the orthogonal projections IRn -> !Rn , ]R^ -+ nRn (see 1.5.7).
Now cover X with a finite number of sets U = cp 1 (Int Dn ) , say
X X
ux ,...,Ux
„ , and denote by ti the--- ^
subset of C (X,X ') consisting of
XI "s
all maps f such that f (Cl U ) c suppip1 and the left n'^n' -minor
Xi xi
of the Jacobi matrix of the map loc (ip ,cp' )f has no zeros on Dn
x i xi
The intersection U 1 fl ... fl U is the desired neighborhood of the map
V
4.If X is compact, then the set Embr (X,X') is open in
Cr (X,X’) (1 ^ r £ oo) .

PROOF. Given a Cr-embedding fQ : X +X', theorems 2 and 1.5.4

show that it is enough to produce a neighborhood of f^ in Cr (X,X')
consisting only of injective maps. For each point x € X, choose two
 199

charts, cp £ Atl X and tp* £ AtlX' , such that fn (supptp ) c suppcp'

X. X X U X X
and loc (tp^ f coincides with one of the inclusions TRn -> IRn ,

3R , or IR^
(see 1.5.1). Now cover X with a finite
n ~1
number of sets U = tp (Int D ) , say U , ...,U . Let U. be the
x ^x 2 x ' x 1
27 1 s
subset of C (X,Xf) consisting of all maps f such that
f (Cl U ) c: supp cp' and, if we symmetrize the upper nxn-part of the
xi xi
Jacobi matrix of loc (cpX .,cpx
* ,)f and take all the principal minors,
1 1
they are all positive on the ball Dn . (The principal minors are the
left-upper minors; the symmetrized matrix is half the sum of the matrix
with its transpose.) Finally, denote by U that part of Cr (X,Xf)
consisting of all maps such that the preimage of any point of X1 lies
in one of the sets U . Let us show that the intersection
x.
l
1/ = 11^ fl . . . H n U is a neighborhood of f^ with the necessary
property. It is clear that f^ £ 1/ and that all the sets IL are
open. Hence it suffices to verify that: (i) U is open, and
(ii) the maps in IL are injective.
To prove (i), note that U is the preimage of the set

0/ = C ( X X X , ( X X X ) \ U.(U XU ) ;X1 x X ' ,(X' x x') \diag(X') )

1 X. X.

under the continuous mapping C (X,X!) C (X x X,X! x X 1)/ given by

f h* f x f . Since (X x x) ^ U -(U x U ) is compact and
i i
(X* x x 1) \ diag(X1) is open in X* x x 1 (see 1.2.2.4), W is open
in C(X x x,X' x x 1) and U isopen in Cr (X,X').
To prove (ii), given a map f € IL and arbitrary distinct
points y,z £ U , let s: I ->3R be the function which takes each
Xi
point t £ I into the inner product of the vectors v = cp (z) - m (y)
and 1 1

[loc (cp )f]((1-t)cp (y) + tcp (z)) - loc(ip ,<p' )f(y),

xi xi xi xi xi xi
n*
computed in HR . Next denote by J(u) the symmetrized upper nxn-part
of the Jacobi matrix of the map loc (cpx ,ip^ )f at the point u € Int D ,
i i
and by qt - the bilinear form ]Rn x IRn IRn having the matrix
J((1-t)ip (y) + t tp (z) ) . The function s is smooth and its
xi Xi
derivative at the point t is precisely qt (v,v), and is therefore
positive (as a consequence of the definition of U^) . Moreover,
s(0) = 0, which implies that s(1) > 0, i.e., f(y) i- f(z).
 200

5. If X is compact, then the set of neat Cr-embeddings

X + X' is open in C^ (X ,X ') (1 £ r £ «>) .
This is an immediate corollary of 4 since the set in question
is just Embr (X,X') n Cg(X,X').

6. lf_ X is compact, then the set Diffr (X,X') is open in

Cg (X,X' ) (1 £ IT £ °°) .

This is a corollary of 5 (see 1.5.1)

2. The Simplest Embedding Theorems

^r
1. Every compact C -manifold, 1 £ r £ °°, admits a
C -embedding in a Euclidean space of sufficiently high dimension.

PROOF. Let n be the dimension of the given manifold X,

and let a : IRn I be a Cr-f unction equal to 1 on Dn , smaller than
1 outside Dn , and equal to 0 outside the concentric ball of radius
2. For each point x € X, fix a chart cp € Atl X with Im cp = 3Rn
x x x
or 3R^, and =0. Define a map j : X TR x iRn by the formula
(a(<px (y) ) ,a(cpx (y) )<Px (y) , if y£ supp ipx ,

j x <y> =
(0,0) , if y 6 X supp ip
X

N o w cover X by a finite nu mb er of sets Ux = ip ** (Int D n ) , say

Ux , ...,U , and define j: X ->■ OR x lRn ) * ... x qr x hr11) = irs ^n + 1 *

1 s
by j ( y ) = (jx (y) f .••,j (y) ). The ma p j is of cl ass Cr and
1 s
injective: if y £ U and
( y •) / j (y). [Indeed,
y' f y, then j
xixi xi
if y' ? U , then (y1)) < 1 , whereas a (cp (y) ) = 1 ; if
cx(V
i i xi
y' € U , then a (ip (y'))<p (y'> = ip (y ’) i ip (y) = a (tp (y))ip (y).]
"i i i xi xi xi xi
Moreover, j is an immersion, since j is an immersion on U
xi xi
(the second component X 3Rn of the map j agrees with cp on
x. x.
r 1
Ux ). Therefore, j is a C -embedding (see 1.5.4).
i

Supplement for the Case of Nonempty Boundary

2 (LEMMA) . On any compact C^-manifold, 1 < r < °°, there

 201

is a (real) C -function h, equal to 0 on 9X, positive on intX ,

and having no critical points on dX.

PROOF. Let a: 3Rn ^ I , n - dim X , be a Cr-function equal

to 1 on D and equal to 0 outside the concentric ball of radius
2. For each point x e BX fix a chart cp E Atl X such that
n x x
Im cp - IR and cp (x) = 0, and define two functions f ,g : X -* 3R
x ” x x ^x
through the formulas

1 - a (tp (y)), if y G supp tp ,

x x
£x (y> =
1 / if y e x \ supp cp^ ,
and
- 6 (tp (y) ) , if y G supp tp ,
X
g x (y ) =
0, if y € X \ supp cp^

n
Here 6 : HR + IR is given by 3 (t^,..., t^) = t^a(t^, ...,t^) . Covering

dX by a finite number of sets U = cp ^(Int Dn ) , say U , ...,U ,

X X X j* X
and setting s
s s
My) = T T fx (y) + I gx (y) '
i— 1 i i= 1 i

we obtain the needed function h: X IR. In fact, h vanishes

identically on dX, since
f is equal to 0 on U and all the
Xi xi
functions g vanish identically on BX; h is positive on intX ,
i
since all the functions , g f
are nonnegative and g is positive
i i i
at all points of intX , excepting the zeros of f .Finally, h has
i
no critical points on dX, since £g has no critical points on Bx
i
(the derivative with respect to the first coordinate of the local
representative loc (cp ,idIR)g ,i.e., of the composition
xk i
(cp I ) o cp ^ , is negative on Dn fl IR1} ^ for k = i and
x ± |suppcpx xk * 1
k ^ ^
nonpositive on D H IR for all k) , while ~| [ f vanishes
1 xi
identically on UU
J x.
i
>27 .
3. Every compact C ' -manifold, 1 £ r £ °°, admits a neat
Cr-embedding in a Euclidean space of sufficiently high dimension.

The formula x (-h (x) ,j (x) ) , where j is an arbitrary

 202

Cr-embedding in ]Rn (see 1), and h is the function constructed in 2,

r Q +1 *] Q
defines a neat C -embedding in 2R_ = HR_ x 3R .

In fo rm a t i o n

4. The compactness assumption and the condition that r /

may be eliminated from the formulations of 1 and 3. Any smooth manifold
of class C*r, with r £ « or r = a, compact or not, can be
Cr-embedded in Euclidean space, and any smooth manifold of class C
n
,
r
with r £ 00 or r = a, compact or not, admits a neat C -embedding m
Euclidean space. For proofs see [22] and [8 ].
We should m e n t i o n that the case r = a in T h e or em s 1 and 3
is exc ee di ng ly d i f f i c u l t and this is the re a s o n why we e x c l u d e d it here.
In the sequel we shall exclude it from other f or mul a t i o n s too: cf., for
example, 4.2, 5.3, 6.5, and 4.6. 2.7.

3. Transversalizations and Tubes

1. In this subsection, we consider the image in Euclidean

space of a smooth manifold under a differentiable embedding and study
the structure of a neighborhood of this image. The results are
concentrated in theorems 4, 5, and 7, and serve as the technical basis
for the remaining part of the present section.

2. Let j be a differentiable embedding ofthe smooth,

closed, n-dimensional manifold X in 3R^. A transversalization of j
is a c o nt in uo us map t : X G(q,q-n) such that, for eac h p o i n t x E X,
the plane t(x) is transver se to the p l a ne d^j (Tang^X) (i.e., the
two planes intersect at only one point). A basic example is the normal
t r a n s v e r s a l i z a t i o n which a s s o cia te s to each p o int x £ X the
corresponding normal plane (i.e., the orthogonal complement to
dxj (Tang^X) in M q ) ; if j is of class Cr , then its norma l
r— 1
transversalization is obviously of class C (cf. 1.4.2).
Given an embedding j: X and a transversalization
T: X -+■ G (q,q-n) of j, one can construct the natural map
t : X ->■ G' (q,q-n) , which takes each point x into the plane
j(x) + t (x ) (which is parallel to t (x ) and passes through j(x)).
We denote the ball and the sphere w i t h c e n t e r j(x) and r adi us p in
j (x) + i (x) by d T (x,p) and s (x,p), r e s p e c t iv ely . The u n ion s

Ux G X d T (x,p) and Ux £ X [dT (x,p) ^ s T <x »P)i are ca l l e d the tube (or
 203

the tubular neighborhood) and the open tube (or the open tubular
neighborhood) of radius p of the transversalization t , and are
denoted by Tub^_p and tub^_p , respectively.
A tube Tub^_p is said to be neat ifthere is a a > p such
that: (i) the open balls dT (x ,g ) \ sT (x,cr) , x € X, are pairwise
disjoint and the open tube tubT<J they form is a neighborhood of
j(X) in 3Rq ; (ii) the map of this neighborhood onto X, which
transforms all the points of dT (x,cr) \ sT (x,a) into x, is smooth.
The restrictions of the last map to TubTp or tub^_p (which obviously
do not depend on the choice of a) are called projections and are
denoted by PrT •
If Tub^p is neat, then all the tubes Tub p 1 with p' < p
are obviously neat.
Warning: it may happen that the normal transversalization
does not have a neat tube, or even a tube such that the balls d^_(x,p)
are disjoint; see exercise 11.4.

3. The following construction enables us to represent a neat

tubular neighborhood as the image of some ideal model of itself, and is
required in the proofs of theorems 4 and 5.
Let Tu^_ be that subset of the product X x nRqconsisting
of all the pairs (x,t)with t E t (x) .For p > 0, we let Tu^_p and
tu^p denote the pieces of Tu^ such that dist(0 ,t) £ p and
dist(0,t) < p, respectively. We also denote by nat: Tu^ 3Rq the map
given by nat(x,t) = j (x) + t. nat is obviously an isometry of each
plane x x T (x) onto the corresponding plane j(x) + t (x ), and
transforms Tu^p (tu^p) exactly into Tub^_p (respectively, tub^p).
It is also clear thatnat is injective on Tu^_p (tu^_p) if and only
if the balls dT (x,p)(respectively, the open balls d^_(x,p)\ s (x,p))
are pairwise disjoint. In this case nat transforms the restriction
to Tu^p of the projection pr^ : X x nR^ x into the projection
pr^: TubTP -+ X (respectively, the restriction to tu^p of pr^ into
pr : tub p -+ X) .
T T
We are interested only in smooth transversalizations t . If
j and t are Cr-maps with r ^ 1, then X x nRq is a C^r-manifold,
Tu is a neat q-dimensional submanifold of X x ]Rq (without boundary),
and nat is a Cr -map. Moreover, in this case Tu^p
T is a compact
q-dimensional submanifold of Tut such that int(Tu^_p) = tu^p and the
restriction of the projection pr 1 : X x nRq X to each of the manifolds
Tu , Tu p , and tu p is a Cr-submersion.
T T T
4. Let the maps j and t be of class C , r ^ 1. If
 204

r cr
Tub^_p is a neat tube, then it is a C -submanifold of TR with
int (Tub^p) = tubTP, and prT : TubTP -* X is a Cr-submersion.

PROOF. Let a > p be such that the conditions (i) and (ii)
in 2 are satisfied. As 3 shows, the map ab nat : tu a -* tub a is
-1
invertible and its inverse ab nat : tub a ■> tu a is obviously given
-1
by y ** (pr (y),y - j ° prT (y)). This formula shows that ab nat is
smooth and thus a C -diffeomorphism. Now it is evident that the
properties of Tub^p and the projection pr^: Tub^p -> X which we have
to verify are consequences of the properties established in 3 for their
models Tu p and ab pr. : Tu p X.
t 1 T
5. Every smooth transversalization has a neat tube.

PROOF. Let t be a smooth transversalization of the

embedding j: X + IR^. Since the planes t (x ) and d^j(Tang^X) are
transverse, the differential d, .nat is nondegenerate when t = 0 .
ix,t)
Hence, nat defines a diffeomorphism of a neighborhood U of X x 0
in Tu^ onto a neighborhood of j(X)(see 1.5.5). It is clear that
if Tu^p <= u, then Tub^p is a neat tube of the transversalization t .
>27
6 (LEMMA) . Suppose X and X 1 are closed C ' -manifolds,
1 £ r £ 00. If there exists aCr-embedding of X' in Euclidean space
r r
together with a C -transversalization, then the set C (X ,X 1) is dense
in C(X,X*).

PROOF. Fix a Cr-embedding j:X IR^, and a Cr-embedding

cr1 r
3 ': X' together with a C -transversalization x'. It suffices
to show that given an arbitrary continuous map f : X -»■ X ' and an
arbitrary e > 0, there is a Cr-map g: XX' such

max dist(j ' o f(x),j' o g(x)) < e

x€X

(see 1.2.7.3). We construct the neat tube Tub^.p 1 with p' £ e/2
and choose 6 to be the smallest of the two numbers e/ 2 and

Dist(j ' (X'), IRq \ tub^,p '); note that 6 > 0 (see 1.1.7.15). Then,
according to 1.1.5.12, the composition

b j ) _ 1 > x -£-» X' —3_!__>

j (X) -<a- 1

extends to a continuous map f : ]Rq -* ]Rq ' .n o w Theorem 3.1.4.yields

q q'
a map g^ : 1R -+ 1R with polynomial components and such that

max dist(f o j(x),g o j(x)) < 6 .

x£X 1
 205

This in turn shows that ° j(X) c tub ,p 1, and it is clear that

g(x) = pr^.,(g^ o j (x) ) defines the desired map g: X ->• X '.

7. Every Cr-embedding of a closed C*r-manifold, 1 £ r £ a

in Euclidean space admits a Cr-transversalization.

If r = a ,<», the normal transversalization will suffice.

If 1 £ r < 00, the existence of the normal transversalization shows
that the set of all transversalizations of a given Cr-embedding
Q
j: X 1R is not empty. Since the latter set is (trivially) open in
C(X,G(q,q-dim X) ) , it is enough to show that Cr (X,G(q,q-dim X )) is
dense in C(X,G(q,q-dim X) ) . But this is a consequence of Lemma 6 ,
because G(q,q-dim X) can be analytically embedded in Euclidean space.

4. Smoothing Maps in the Case of Closed Manifolds

1. Now we arrive at the main topic of the present section -

approximating maps of one smooth manifold into another by maps which
are more regular in a sense or another as, for example, maps of a higher
differentiability class, or embeddings, or immersions.
In this subsection we consider only approximations which raise
the differentiability class of maps without improving their other
properties, and we restrict ourselves to the simplest case - that of the
closed manifolds.
>^
2. If r £ oof then for any closed C " -manifolds X and X'
andany s < r, the set Cr (X,Xf) is dense in CS (X,X'). The same is
true when r = a provided that X and X' can be Ca-embedded in
Euclidean spaces.
s
We have to show that given a map f £ C (X,Xf) and a
neighborhood U of f in CS (X,X'), there is a Cr-map in U. Fix
Cr-embeddings j: X ->-3Rq and j': X' -* 3Rq , corresponding
transversalizations t and t', and corresponding neat tubes Tub^p
and Tub^,p '. Now consider the mapping

CS (abj,prT,) : CS (TubT (p/2) ,tubT ,p') -+CS (X,X'),

where ab j = [abj : X ->■ TubT (p/2)] (see 1.1). Since it takes Cr-maps
into Cr-maps, it suffices to prove that the preimage 1/ of U under
this mapping intersects Cr (Tubx (p/2),tubx ,p ’). But V is open and
contains the restriction to Tubx <p/2) of the composition

tub p T-* X X' ab ■> tub ,p '. (1)

T T
 206

The image of this restriction is compact, and so it lies at a positive

distance from IRq ^ tub ,p '. Consequently, there exists e > 0 such
that if, at the points of Tub^(p/2), the partial derivatives of order
L s cr*
0 ,1 ,..., s of the coordinate functions of a map g £ C (TubT (p/2) ,IR^ )
differ from the corresponding partial derivatives of the map (1 ) by
less than e, then g (Tub (p/2)) c tub^ ,p 1 and

[abg: Tub^(p/2) tub^ ,p '] £ 1/. Finally, apply Theorem 3.1.4 to the
coordinate functions of (1 ) to deduce that there exists a map g with
polynomial coordinate functions whose partial derivatives have the
property above. Thus I/ (1 Cr (Tub^. (p/2) ,tub^ ,p ') / 0.

3. Comparing Theorem 2 with the Theorems 1.2, 1.3, 1.4, and

1 .6 , we see that for r £ 00 and 1 £ s < r,
and for any given closed
>77 r
C ' -manifolds X and X', the following holds: Iram (X,X') is dense
in ImmS (X,X '), Submr (X,X') is dense in SubmS (X,X'), Embr (X,X') is
s r
dense in Emb (X,X'), and finally Diff (X ,X ') is dense in
DiffS (X,X'). The same is true when r = a provided that X and X'
can be Ca-embedded in Euclidean spaces.
>27
4 (COROLLARY) .
If two closed C " -manifolds, 1 £ r £ 00, are
27
diffeomorphic, then they are C -diffeomorphic. The same is true for
r = a provided that X and X' can be Ca-embedded in Euclidean spaces.

5 (INFORMATION). Two closed homeomorphic Ca-manifolds are

not necessarily diffeomorphic. Historically, the first such examples
a 7
where C -manifolds which are homeomorphic, but not diffeomorphic to S ;
see [15].

Supplements to Theorem 2

6. If r £°°, then Cr (X,X') is dense in CS (X ,X ') r s < r,

in the following more general situation too: X is a closed
>r*
C -manifold and X 1 is an open subset of a closed C " -manifold Y.
The same is true for r = a if X andY can be Ca-embedded in
Euclidean spaces.

The proof reduces to observing that the mapping

CS (id, in) : CS (X,X') -v CS (X,Y)

is a topological embedding with open image which carries Cr (X,X')

into the intersection of this image with Cr (X,Y). Since Cr (X,Y) is
s
dense in C (X,Y), the above intersection is dense in this image,
 207

and Cr (X,Xf) is dense in CS (X,Xf).

7 (LEMMA). Every pair of disjoint closed subsets of a closed

r y
C -manifold with r <» has a Urysohn function of class C .

PROOF. Let (¡>: X -v I be an arbitrary Urysohn function for

the given pair of subsets A, B of X. According to Theorem 6 , there
is a Cr-f unction \p: X -+1R such that max U(x) - $ (x) I < 1/3 . If

now A: 3R I is a Cr-function such that A(y) = 0 for y £ 1/3 and

A(y) = 1 for y ^ 2/3, then A ° \p is obviously a Urysohn Cr-function
for the pair A ,B .
^r
Let X and X 1 be closed C -manifolds and let A be a
closed subset ofX. I_f 0 ^ s < r ^ oo,then that part of CS (X,X')
r
consisting of the C -extensions of a given map (f>: A -* X* is dense in
s
the part of C (X,Xf) consisting of the extensions of $ which are of
2T
class C m a neighborhood of A(the neighborhood depends upon the
extension).
s
Let f EC
(X ,X1) be an extension of (p which is of class
27
C in aneighborhood U of A. Given a neighborhood U of f in
s r
C (X,X 1 ) , we have to show that u contains a C -extension of <j). Fix
r Q* r
a C -embedding j f: X* -* , a C -transversalization x* of j ’f
s
and a neat tube Tub ,p 1 , and denote by 1/ the piece of C (X,X')
consisting of all the maps g such that

max dist (j* o f(x),jf o g(x)) < Dist (j 1 (X) , JRg \ tub ,p 1 ) .
xex T

It is obvious that 1/ is open and that for any g £ 1/, x e X , the

segment with endpoints j f o f (x) , j' o g(x) is contained in tub ,p*.
r ^
Next construct a Urysohn C -function \p for the pair A, X ^ U (see 7)
and consider the map <2>:1/ CS (X,X*) which transforms g into the
map

x ^ prT , ((1 -^ (x) )j ' o f (x) + ip(x)j' o g (x) ) .

One may check directly that <2> is continuous and 0(f)=f; hence
—i —1
the set 0 (U) is open and nonempty. Now Theorem 2 shows that $ (U)
t* r
contains a C -map. Finally, we note that $ takes C -maps into
27
C -extensions of <\>.
>r
9. Suppose that X and X' are closed C' -manifolds and
is a submanifold of X which is itself closed as a manifold. Let
(5 ; A -> X ' be a Cr-map. If 0 ,< s < r ^ then that part of
CS (X,X’)
t S
consisting of the C -extensions of <(> is dense in the part of C (X,X’)
 208

s
consisting of all the C -extensions of d>.

Given a CS-extension of d> and a neighborhood U of this

extension in CS (X,X')/ we have to show that U contains a Cr-extension
of <f>. Fix Cr-embeddings j : X -*■ IR^
and j ': X' -> 3R^ , a
C37-transversalization t of the embedding j•LI .* A IR q and a
Cr-transversalization t ' of j', and corresponding neat tubes Tub^p
and Tub^_,p '- Further, denote
Tub ,p'. denote by I/ the piece of CS (X,X') consisting
T
of all the maps g such that

max dist(j 1 o (}>(x) ,j 1 o g(x)) < Distfj'fX1), 3R^ ^ tub ,P 1 ) .

xex
s
Obviously, I/ is open and contains all the C -extensions of to X.
r —1
Now take any Urysohn C -function \p for the pair X v j (tub p) ,A and
s
consider the mapping $:I/ + C (X,X') which transforms each map g
into the map

PrT »(j°g(x) +^ (x) [j 1 o(J)oprT (j (x) )- j 1 ogoprT (j (x) )]) ,

x h- ^ if j (x) £ TubTP,

g (x) , if j (x) £ Tub P .

It is clear that $ iscontinuous and that 0(g) = g whenever g

-1
extends $. This implies that 0
((J) is an open nonempty set which,
27
according to Theorem 2, contains a C -map. Finally, note that 0 takes
r r
C -maps into C -extensions of $.

5. Glueing Manifolds Smoothly

1. Our main task in this subsection is to make the necessa

preparations for extending the basic approximation theorems given in the
previous subsection, i.e., theorems 4.2-4.4, in their nonanalytic
version, to include compact manifolds with boundary. The main tool
used in the extension is that of smooth doubling of a compact manifold, an
operation which transforms it into a closed manifold. However, we find
it convenient to define and study a more general operation, which is
useful for other purposes too - the smooth glueing of smooth compact
manifolds. To begin with, we need to investigate the structure of a
smooth compact manifold in the vicinity of its boundary.
 209

Collars

2. A collaring of a compact Cr-manifold X (0 £ r £ a) is

a C -embedding of the cylinder 9X x i into X, which takes the point
(x,0) into x,for each x £ 9X. The image of ax x I under such
an embedding is known as a collar (on X).
If X is a smooth manifold (i.e.,r > 1), a collaring is a
differentiable embedding and its image is a submanifold of codimension
0, whose boundary consists of ax and of a submanifold of intX
diffeomorphic to 8X.

3. rf 1 £ r £ °°, every compact C -manifold admits a

collaring.

PROOF. Let X be the given manifold. Pick a neat

Cr-embedding j: X ]Rq (see 2.3), a Cr-transversalization t of the

composite embedding X— ]Rjj — ~*~n-> 3Rq , and a neat tube Tub^p.

-1 1
Consider the map <J): j (tub^p) 3X x HR_, defined as
x (pr^(j(x),j^ (x)),where j^ is the first coordinate function of
j. Since j is neat, the differential d^$ is nondegenerate at each
point x G 3X, so that (j) realizes a diffeomorphism of a neighborhood
of 3X onto a neighborhood of 3X x 0 (see 1.5.5). Now let e > 0
be small enough so that the product 3X x [-£,0] is contained in the
previous neighborhood. Then the formula(x,t) *+ $ (x,-et) obviously
defines a collaring of X.

4 (INFORMATION). The compact topological manifolds (r = 0)

and the compact analytic manifolds (r = a) admit collarings too. The
case r = 0 is considered in [4].

Glueing

5. Suppose that X and X' are compact n-dimensional

Cr-manifolds with r ^ 1, and let C and C' be submanifolds of
3x and 3X', respectively, consisting of whole components of these
boundaries. Assuming that C and C* are diffeomorphic, pick a
Cr-diff eomorphism 4>: C ->■ C* and attach X to X' by the composite

map C C* •-1~ > X 1 (see 1.2.4.8 ). The resulting space

Y = X* U. .X is obviously a compact, n-dimensional, topological
in° (J)
manifold. However, if X and X' have collars then it turns out that
t* r
Y has a natural C -structure that makes it into a collared C -manifold.
 210

The atlas that defines this Cr-structure consists of the charts of

Atl(X ^ C) and Atl(X' ^ C') (we regard X and X' as parts of Y) ,
as well as the charts ¥ constructed from both the charts ^ £ AtlC
and the collarings k: 9X x I + X and k 1 : 8X 1 x I X1 by the
formulas

supp Y = k (supp 11> x [0,1)) U k ' (supp x [o,1))

and

¥(k(z,t)) = (<Mz),-t)
z £ C, t £ [0,1)
V (k ’ (z,t) ) = (ip (z) ,t)

(Im ip<= 3Rn 1 and Im T = Im 4> x( - 1 ,1 ) <= HRn 1 x K = ® n ) . it is readily

seen that these charts are all pairwise compatible. One constructs a
collar on the resulting C -manifold Y from those pieces of the
collarings k and k f which are preserved under the above procedure.
We say that the diffeomorphism <J) glues X and X1 into Y. It is
clear that dY = (3X ^ C) U (8x* ^ C 1) and that X, X 1, and C are
submanifolds of Y. The pieces of the collarings k and k’ which
are related to C and C1 yield a two-sided collaring of the manifold
C in Y, i.e., a Cr-embedding C x [-1,1] -+ Ysuch that (z,0) z;
its image is a two-sided collar of C in Y.

In the particular situation X 1 = X, Cf =C=9X, and

<J> = id 8X , we use the term doubling instead of glueing and denote Y
by doppX . This definition agrees with 1.1.9, i.e., (dopp X) =
= dopp(C^X).

Now suppose that X and X1 are oriented manifolds and C

and C* are equipped with the induced orientations (see 1.3.4). If
<f> is orientation preserving, then Y is orientable and can be actually
oriented in a canonical way. This canonical orientation is that which
induces the original orientation on X and the orientation opposite to
the original one on X 1. In particular, doppX is oriented for any
oriented manifold X.
The simplest examples show that the Cr-structure on Y
depends not only upon the Cr-structures of the manifolds X and X!
and the dif feomorphism (J), but also upon the collarings k and k !.
Our next objective is to demonstrate that for r f a this last
dependence is eliminated if we regard Y as distinct up to
C -diffeomorphisms.
 211

6 (LEMMA). Let X and X’ be closed C^-manifolds,

1 £ r £ oo, and let f: X x [-1,1] + X' x [-1,1] be such that
f (X x [-1,0]) c X' x (-1,0] and f(X x [0,1]) crX' x [0,1). Assume
that ab f : X x [-1,0] -> X ’ x (-1,0] and abf: X x [0,1] -»-X’ x [0,1)
are differentiable embeddings of class Cr , while ab f : X x 0 -s- X ' x 0
is a diffeomorphism. Then there exists a Cr-embedding
g: X x [-1,1] X x [-1,1] which agrees with f on
(X x [—1,—1/2]) U (Xx 0) U (X x [1/2,1]), and such that
g(X x [-1,0]) = f(X x [-1,0]) and g(X x [0,1]) = f(X x [0,1]).

t
In the proof that follows we let f^ and denote the
composite maps

f prl
x X [-1 ,1 ] — ^ X' X [-1 ,1 ] U X’
and
f .......... pr2
X x [-1,1] X ’ x [-1,1] [-1,1]

respectively.
For a start, assume that for some positive e the map f^
is constant on each set x x [-c,e], x e X. In this case, fix positive
numbers 6 and r\, such that 6 £ min(e,1/2) and for any x E X the
derivative of the function t ^ f^ix^) is not less than n on the
intervals [-6,0) and (0,6]. [The existence of such 6 and n
results from the continuity of the functions X x [-1,0] ->1R and
X x [0,1] HR, given by the formula (x,t) h- 8 f 2 ( x , t ) 9 t |T =t (for
t = 0 one takes the left derivative for the first function and the
righ derivative for the second one), and the positivity of both
functions on X x 0. Since X is compact, these functions are bounded
from below by a positive constant on X x 0, and hence on X x [-6 ,6 ]
for some 6 > 0.] To proceed further, pick a C -function
a: [-6 ,6 ] I such that a(t) = 0 for |t| < 6/4 and a(t) = 1 for
Itl ^ 6/2. It is not hard to verify that the formula

(f (x,t) , (1 -a(t) )nt + a (t) f 2 (x,t)), if |t| £ 6 ,

g(x,t) =
f (x,t), if 111 s 6,

defines a map g: X x [-1,1] -> X ' x [-1,1] with the desired properties.
In the general case, choose 0 < £ 1/2 such that the map
(jj : X -*■ X ’, defined as <J)t (x) = f^Xjt), is a dif feomorphism for all
ItI 4 z . [The existence of such an e.
is a consequence of: the
r
continuous dependence of on t in the C-topology, the fact that
^ is a dif feomorphism (recall that ab f : X x 0 - * X * x0 is a
diffeomorphism), and the fact that Diff (X,Xf) is open in C (X,X*)
 212

(see 1.6). Let y: 3R be a nondecreasing C -function such

that y (t ) = 0 for 111 £ e^/4 and y(t) = t for |t| £ c^/2. Now
define f~: X x [-1 ,1 ] > X 1 x [ - 1 ,1 ] by
■1
f (4>t - 4> (t) (x) ,t) , if |t | ^ cr
f (x,t)
f(x,t), if t £ c

Then f~ satisfies all the conditions imposed to f in the statement

of the lemma. Moreover, f~ satisfies the extra conditions under which
the lemma has already been proved, namely that the composite map
pr.
1
X x [-1,1] X' x [-1,1] X’

is constant on the sets x x [-c,c], with c = c^/4. The map g

corresponding to f~ via the above procedure has the needed properties
because f agrees with f on (X x [-1,-1/2])U (X x 0)U(X x [1,1/2]).

7. Suppose that X^ , X^, X2 , and X2 are collared

Cr-manifolds, 1 £ r £ 00, , and are pieces of their
boundaries consisting of whole components, and <i> : C,j and
<j>2 : C 2 C2 are C -dif feomorphisms. Assume that there are

C -diffeomorphisms F: X and F': X' X', such that F (C ) = C

1 '2 '
F' (C Jj) = C2 , and the diagram

C’
^1
ab F ab F 1

C 21

is commutative. If Y^ is the result of glueing and X^

1
and Y is the result of glueing then the X. and X ’ b£ (¡>2 t
manifolds and
Y 2 are C -diffeomorphic. Moreover, there exists
a C -diffeomorphism G: Y ^ Y 2 such that G (X^ ) X2 , G ( X ’
X2 '
G (C, = C 2 ' and [abG: C2] = [ab F
C2 ]

PROOF. Let C 1 x [-1 ,1 ] Y1 and ¿ 2 : C 2 x [-1 , 1

be two-sided collarings Denote by H: Y^ Y2 the map defined by the
formulas H [in: X 2 Y 2 ] o f and [in: X ’ * Y2: o F ’
X
'x i
and choose c > 0 so that H ° x [-c ,e]) <= ? 2 (C2 X Now
apply Lemma 6 to the map f: C 1 x [-1,1] ->■ C 2 x [-1,1] given by
f(z,t) = , - 1 (H t (z,ct)). This lemma guarantees the existence of a
^r
C -embedding g: x [-1 ,1 ] C 2 x [-1,1] which agrees with f on
 213

(C.j x [— 1 ,—1/2] ) U (C1 x 0) U (C1 x [1/2,1]), and satisfies

g(C 1 X [-1,0]) = f(C 1 X [-1,0]), g(C 1 X [0,1]) = f(c X [0,1]).
Clearly,

h (y) , if y i 1 ( C 1 x [_F- ) '

G(y) =
12 ° g (z ,t/e), if y = (z,t) with z1 e C ,
t £ [—e ,e ]

defines the required C -dif feomorphism G: ->• -

Cutting

8. Let Y be a C -manifold, 1 < r £ «>, and let X and X'

be compact submanifolds of Y such that dim X = dim X 1 = dim Y and
Y = X U X 1. rf C = X n X' is a piece of both boundaries 3X and 3X*
consisting of whole components of dX and 9X', then there is a
Cr-embedding ¿: C x [-1,1] Y such that l ( z,0) = z for any point
z E C and t (C x [-1,0)) c= int X, i(C x (0,1]) c int X ' .
r cr r
PROOF. Fix a C -embedding j: Y IR , a C -transversalization
t ofthe embedding j|P : c 3R^, and a neat tube Tub p. Consider
cr— 1
the map $: C which takes each point z £ C into the unit
vector tangent to j(Y) at the point (z), and j (z), contained in t
r—1
pointing towards j(X'). (p is continuous (in fact, of class C )/
r cr— 1
and so Theorem 4.2 yields a C -map cf)^ : C + such that the inner
product <(}>(z) (z)> is positive on C. Denote by ¡¡j : tub^_p C x IR
the map defined by the formula \p(z) = (pr^ (z) ,<z - j ° pr^_ (z) ,(p^ (z)>) .
Clearly, i¡j is of class Cr and for ze C the differential

<3j (z)^ ; Tangj (Z)(tut>TP) + Tang ( z , 0 ) (C x3R) = TangzC) © IR

induces an isomorphism of Tang., .j(C) onto Tang C and carries the

] IZ ) z
vector <f>(z) into <<j> (z) ,4^ (z) > e IR. Moreover, both T a n g ^ zjj(C)

and (p (z ) are contained in Tang.. .j (Y) ; hence d., .^ takes

J \Z) J \Z)

Tangj (z)j (Y ) onto Tang(z,0) (C x :R) * since dim Tang. (z) =

= dimTang(Zi0) (Cx IR) , we see that <*j (z,*Tang j (Y)' i.e.,
U vz )
the linear map dj (z) ^ j j (Y)ntub p} ' ±S an isomorPhism-' B¥ Theorem
15 5.
u»I . - > , defines a dif feomorphism from a neighborhood of
' v |] (Y) fltub^p
j (C) onto a neighborhood of C x 0. Accordingly, C x [-r, ,e] will
 214

lie in the previous neighborhood provided that e > 0 is small enough.

Now it is plain that i ( z,t) = j~1 (^ 1 (z,et)) defines the desired
embedding I: C x [-1,1] ->■ Y .
9 (COROLLARY) . Let Y be a Cr-manifold, 1 £ r £ 00, and
let X and X1 be compact submanifolds of Y such that dim X =
dimX' = dim Y and Y = X U X'. If X il X' is a piece of both
boundaries 3X and 3X!, consisting of whole components of 3X and
27
3X', then id X andid X 1 together define a C -diffeomorphism of Y
onto the manifold obtained from the appropriately collared manifolds
X and X1 glueing X and X 1 b^ id (X fl X').

The Simplest Application

10. Every smooth compact manifold is a CNRS.

When the manifold is closed, this is a consequence of

2.1, 3.7, and 3.5, because the image of a smooth manifold under a
differentiable embedding in Euclidean space is the retract of the
interior of a neat tube corresponding to a smooth transversalization of
the given embedding. Theorem 1.3.6 .4 enables us to reduce the case of
manifolds with boundary to the closed case; namely, any compact smooth
manifold has a smooth closed double (see 5), and is obviously a retract
of this double.

6. Smoothing Maps in the Presence of a Boundary

1. The main results of this subsection are Theorems 5 and

which generalize Theorem 4.2. Lemma 2 is necessary to the proof of
Lemma 3, Lemma 3 - to the proof of Lemma 4, and Lemma 4 - to the proof
of Theorem 5. Finally, lemmas 7 and 8 are necessary to the proof of
Theorem 9.

2 (LEMMA) . Let Y be a C^r-manifold with r < 00, and let

f : Y x 1R_ jr be a C-function. Then the function F :Y x 1R + IR
defined by

f (y /1 ) , if t £ 0,
F (y ,t)
Yr (-1 )^ r + 1 f(y,-kt), if t ^ 0
¿k =0 * u k +1

is also of class Cr .
 215

PROOF. All we must check is that the two expressions defining

F/ as well as their partial derivatives with respect to t and local
coordinates on Y agree for t = 0. To see this, it suffices to note
that the equality

l L n < - i ) k + s f rk++11 k SDS<M0) = DS(M0]

holds for s £ r and any CS-function J

: 3R_ + JR. Indeed, this last
equality is equivalent to
f r+ 1 1
kS = 0 (s s r)
k +1

and this is valid if we interpret the sum as the M (r+1 )-th difference"
of the integral function k computed at k = -1 .

3 (LEMMA) ^r
Let X be a collared C ' -manifold. If
1 £ r < «, thenevery function in C (X,JR) extends to a function in
Cr (dopp X ,JR) .

PROOF. Fix a two-sided collaring of the manifold X in

dopp X , t : 3X x [-1,1] -+ doppX (see 5.5), and pick a Cr-function
a: JR -+ I such that a(t) = 1 for t £ 0 and a(t) = 0 for t ^ 1 /r.
r 1
Let <
(|
J> G C (X^R). Consider the function f: 3X JR_ defined by the
formula

a (t) <$>( I (y,t) ) , if t £ 1 /r,

f(y,t) =
0, if t ^ 1 /r.

Since f is C , Lemma 2 processes it into a C -function

F: 3X x jr jr (take Y = 3X) . Now it is plain that

( 0, if x € dopp X \ (X U ¿ ( d X x [0,1/r]
ip (x) = j<t>(x) , if x GX
F(£_ 1 (x)), if x € K d X x [0,1/r]),

defines a Cr -function \p: dopp X JR extending <j>.

4 (LEMMA). Let X be a collared C^r-manifold, and let X1

be a closed C^r-manif old. If 0 £ r < °°, then every map in Cr (X,X')
extends to a map in C (doppX,X') .

PROOF. This is evident when r = 0. Suppose r > 0 and

f e Cr (X,X 1) . Fix a Cr-embedding j ’t X 1 -»■]Rq , a Cr-transversalization
t1 of j 1, and a neat tube Ttib^.p' . Lemma 3 ensures that the coordinate
functions of j' o f extend to C -functions doppX + 3R, i.e., j' ° f
 216

extends to a Cr-map g: dopp X IRq . Let U be the neighborhood of

3X in X consisting of the points x such that

dist (j 1 o f (x) ,g (cop (x) )) < Dist (j ’ (X’) , ]Rq ^ tubT ,P ') •

Now construct a Urysohn C -function <|>: dopp X Ifor the pair

X, cop(X v u ) . It is clear that for any x £ X the segment with end­
points j 1 ° f (x) and g (cop(x)) lies in tub^,p ’. Moreover, we see
that the formulas

h (x) = j ' ° f (x) , if x G X,

h (cop ( x ) ) = j ' ° f ( x ) , if xeX-'-U,

h (cop (x)) = (1 - <}>(cop(x) ))g (cop(x) ) + <t(cop (x) )j 1 ° f(x),

if x 6 Cl U ,

define a Cr-map h: dopp X -*■ 3Rq which extends j' ° f and satisfies
h(doppX) c tub^ , p '. Finally, the composite map
ab h pr ,
dopp X ----- * tub^ ,p 1 --- -— * X'

is thedesired C -extension of f todopp X .

>■£
5* Let X and X1 be compact C " -manifolds with X'
closed. If 0 £ s < r £ 00, then:
(i) Cr (X,X 1) is dense in
CS (X,X');
E*
(ii) given an arbitrary C -map cj): 3X + X 1, that part of
s r
C (X,X'| consisting of the C -extensions of cj) is dense in the part
s s
of C (X ,X *) consisting of the C -extensions of $ .

PROOF. The m a p p i n g

CS (in: X dopp X ,id X ') : CS (doppX,X’) ^ CS (X,X' )

r r
transforms C -maps into C -maps and, according to Lemma 4, its image
is precisely CS (X,X'). Hence, (i) is a consequence of the fact that
Cr (dopp X ,X 1 ) is dense in CS (doppX,X’) (see 4.2), while (ii)
follows from the fact that the set of all Cr-extensions of cf. isdense
in the part ofCS (doppX,X') consisting of the CS-extensions of $
(see 4.9).

6 (COROLLARY). Let X and X1 be closed C^r-manifolds,

1 £ r £ °°. If two maps in Cr (X,X‘) are homotopic, then they can be
connected by a C -homotopy x x I -> X 1 .

7 (LEMMA) . Let X and X' be compact C "r-manifolds,

 217

1 * r ^ Then the mapping (X,X ') - Cr (3x, 3x ') , f ^ ab f , is open.

PROOF. We have already seen in 1.1 that this mapping is

continuous. We presently show that it is open. Given a map
JO
f £ Cg(X,X'), it is enough to find a neighborhood U of the map
ab f : 3 X -y 3X' and a continuous mapping $: U -+ Cg(X,X') such that
4>(abf) = f and [ab (<3>(g) ) : 3X -v 9X'] = g for all g € U (see 1.1.4.4).
Fix collarings k: 3X x I -* x and k ’: 9X' x I -> X ', a Cr-embedding
Cf1 JO
j ': 9X* -> M * , a C -transveralization t' of j', and a neat tube
Tub^,p *. Now construct a Cr-function a: I + I such that a(t) = 1
for 0 £ t £ 1/3 and a(t) = 0 for 2/3 £ t £ 1, and choose e > 0
with f(k(9X x [0 ,0 ]) ck'(9X' x I). Let f^ and f be the composite
maps

3X x [0 ,e ] -ab k-> k(9X x [0 ,e ]) -ab'f-> k'(9X' x I) - (ab k> )--- ►

Pri
1 9X *,
> 9X' x I
1 •
P 2

We define U as the set of all g € Cr (3 X, 3 X') such that

max dist(j' o f(y),j' o g (y) ) < Dist (j '(9X ') ¿IR^ \ tub ,p'),
y e 3x T

and define $ as

[<4>(g) ] (x) = f(x), if x6X\k(3Xx[0,e]),

[0 (g)] (k(y,t )) -

= k ' (prT , (j ' (f 1 (y,t)) + a (t/e) [j ' (f (y) ) - j ' (g (y) )]) ,f2 ( y, t) ) ,

if y e 3X, t € [0,e].

It isroutine to check that U and $ have the needed properties.

8 (LEMMA). Let X and X1 be compactC^r-manifolds. If

1 s: r si m , then the set of all composite maps X— X' — ln-> dopp X '
with f € cf(X,X') is open in that part of Cr (X,doppX') consisting
------- O
of the extensions of all maps 3X 3X 1

PROOF. Let fQ e Cg(X,X') and denote by gQ the composite

mapX — X' in > doppX' . Fix a collaring of X, k: '3X x I X

anda two-sided collaring of 3X' in doppX, I': 3X 1 x [-1,1] -* doppx'.
Further, take positive 6 ,n such that: (i)
(i) g 0 (k(3X x [0,6]) c i ' ( 3X' x (-1,1));
 218

(ii) for any y E 9X the derivative of the function

t [ pr 2 : 3 x x [- 1, 1] -+■ [- 1 ,1] ](£ 1 ^ (gQ o k ( y ,t) ) , (2 )

[0,5] [-1,1]/ is everywhere less than -n. [The function

9 X x [0 ,1 ] HR, which carries each point (y,t) to the derivative of
the function (2) at the point t, is continuous on 9X x [0,1] and
negative on 3X x 0. Since 8X is compact, this function is bounded
above by a negative constant on 3X x 0, and hence on 3X x [0,6],
for some positive 6. This ensures the existence of 6 and n as
above.] Clearly, the Cr-maps g: X dopp X ' which fulfil conditions
(i) and (ii) (writing g instead of g^) and satisfy
g(X \ k(9X x [0,6))) c int X ' form an open set in Cr (X,doppX') . It
remains to observe that the intersection of this set with that part of
Cr (X,dopp X ') consisting of the extensions of all maps 9X -+ 9X* is

a neighborhood of g^ in the set of all composite maps

X -£-» X' — dopp X' with f G C^(X,X' ) .

d
>
9* Let X and X' be compact C -manifolds. If
1 < s < r £ 00, then:
(i) C^(X,X')
is dense in C~(X,X');
------------------- O
O
s r
(ii) that part of C^(X,X') consisting of all C -extensions
of a given map (p E Cr (BX,9X') is dense in the part of CS (X,X')
3
consisting of all C -extensions of (p.

PROOF. Lemma 7 and Theorem 4.2 (the latter applied to the

manifolds 3X and d X ') show that the set of all maps g E CS (X,X')
such that [abg: 3X -»• 3X’] € Cr (3X,9X') is dense in C®(X,X').
d
Therefore, (i) is a consequence of (ii).
To prove (ii), suppose that f e Cg(X,X') is an extension of
<j>, and let be a neighborhood of f in cf(X,X'). We have to
U
r
show that U contains a C -extension of $. Consider the composite
f in
map X -- » X' --- » dopp X . By virtue of 8 , this map has a neighborhood
1/ in Cs (X ,dopp X ') such that for any gel/ with g(9X) c SX'
one has g(X) <= X' and [abg : 3X -»■ 9X'] £ U. Thus, it suffices to
r
find a C -extension of <p in I/; but such an extension is provided
by 5.

10. Comparing Theorems 5 and 9 with Theorems 1.2 - 1.6, w

arrive at the following statements for 1 £ s < r £ <». Given any
compact C 'r-manif old X and any closed C^-manif old X', Immr (X,X')
is dense in lmmS (X,X'), Submr (X,X') is dense in SubmS (X,X'), and
 219

Emb (X,X*) is dense in EmbS (X,X*)- For any compact C^r-manifolds X

and X', the set of neat embeddings in Cr (X,X!) is dense in the set
of neat embeddings in CS (X,X'), and Diffr (X,X') is dense in
Dif fS (X,X *) .

11 (COROLLARY) . Two compact C^r-manifolds, 1 £ r £ 00, which

are diffeomorphic are Cr-diffeomorphic.

12(INFORMATION). As with Theorem 4.2, Theorems 5 and 9 remain

valid for r = a too (cf. 2.4). We excluded this case in view of its
difficulty.
Theorems 4.2, 5, and 9 (as well as their Ca-variants) also
hold for noncompact X and X'. However, this generalization is of
limited interest. For example, it does not suffice if one wants to
eliminate the compactness assumption in Corollary 11 (which is actually
possible). The appropriate extensions of Theorems 4.2, 5, and 9 to the
noncompact case are related to topologies which are stronger than those
defined in 1.1, and require analytic tools stronger than Theorem 3.1.4.

7. General Position

1. The main result ofthis subsection is the final Theorem 7,

which constitutes the basis of a large part of the material below. We
emphasize that this theorem is formulated and proved only in the
00
C -case. In Subsection 9 we add a statement covering the caseof
Cr-maps with r finite (see 9.10).
The technical part of the subsection is concentrated in
Theorem 2, which establishes the fundamental topological property of
the spaces Cr ( X , X ! ) f and Theorem 4 , which represents the only corollary
of Theorem 3.2.3 that we need.
To simplify the formulations of Theorems 2 and 3, we give a
special name to those topological spaces where the intersection of any
countable collection of dense open sets is dense: we call them Baire

Technicalities

2. C (X,X1) is a Baire space for any C ' -manifolds X an

X' with 0 £ r £

We have to show that given arbitrary open dense subsets

 220

U ^ , U 2 ,... of Cr (x,x’), and an arbitrary open subset W of Cr (x,X'),

CO r °0
the intersection W n is not emPty- Let 1=1 a

{^ i}i_ 1 atlases °f t^ie manifold x/ indexed so that the set

= Cl supp is compact and contained in suppcp^ and = ab cp^

for alli. Similarly, let (cpj}j_ and ^j^j=i atlases °f x ’'

indexed so that K^ = Cl supp \p*c= suppcpj, andip.! = ab cp.! for all j.

We construct a sequence of C -maps f : X + X',f2 : X -> X 1,..., a

sequence of open subsets of Cr (X,X'), and a sequence of
positive integers n (1 ) ,n (2 ) , ... , such that:
(i) f ± e V± ;
(ii) if i ^ 2, then \J ^ ci V ;
(iii) Cl 1/. c W fl U .;
l l
(iv) c Uj=V suPP^j ;
(v) if s £ i and t £ n (s) , then f . (K O f * * (K') ) cz
i s s t
supp cp^ ;
(vi) if i *2 , s £ i, and t £ n(s), then the partial
derivatives of order £ min(r,i-2 ) of the coordinate functions of the
local representatives loc (cp ,cp|)f.and loc (cp ,cp')f. _ 1 differ by less
• S L 1 S t 1 I *

than 2 1 at the points of cpg (Kg fl fg (K^_) ) .

Then we willhave finished the proof. Indeed, (v) and (vi)

imply that for any s,t such that t £ n(s) the sequence

{loc (cp ,cp')f .I / , ^<-” 1 / /

s t i |cps (supp n fs (suppijj^)) i =s

together with all its partial derivatives of order £ r converges

uniformly on cp (suppi|)
—1
D f(suppiV ))
r
to a C -map
s s s t

gst: <PS<SUPP*S n fs (suPP'f’p) ^ ImiJ^ .

Moreover, the composite maps
-1 ab cp _
suppifi fl f (supp \p ' ) ---- ^-» cp (supp i|j n f (supptK)) >
£> o L S S S t

g st (cpt )_1
> Im cp^ -------- ► supp cp^

together define a Cr-map g: X -> X' [ (iv) shows that the sets
-1
supp ips D fg (supp t p p cover X and clearly the maps gg^_ are
compatible on the intersections of these sets]. g is the limit
(in C (X,X') of the sequence f^f^..., and (i)-(iii) show that
 221

g e w n (n“ = 1 u±).

We proceed by induction. Take f t o be any element of

W n , and take to be any neighborhood of f^ (in Cr (X,X'))
such that Cl l/^ c W n (recall that Cr (X,X') is a regular space -
see 1.1). Assume that for some k ^ 2, the maps f. € Cr (X,X'), the
r
open sets \J^ c= C (X,X*), and the positive integers n(i) with i <k
have been defined and satisfy (i)-(vi). We let G denote the set of
all C -maps g: X X* such that for s£ k-1 and t £ n(s)

g(Ks n fs 1 (Kt )) c supp '

and all partial derivative of order £ min(r,k-2 ) of the coordinate

functions of the local representatives loc (cp ,<p')g and loc (cp ,cpMf,
^ S t S t J^. 1 A

differ by less than 2at the points of <PS (KS n fs <K^.) ) • Obviously,
G is open and € G. Therefore, G fl ? 0 and, since Uk is
dense, G n ^ ^ 0• Choose to be any nonempty open set

with Cl l/^ c G H ^ - 1 ^ ^k' ^k tC> be an^ element of an<3 n(k)

to be any positive integer such that fk (K^) <= supp ^ . It is

readily seen that the objects l/^, f^, and n(k) satisfy conditions
(i)- (vi) for i = k.

3. Every topological manifold is a Baire space.

This is a special case of Theorem 2: in fact, the topological

manifold X may be regarded as the space C^(D^,X).
0° r±
4. Let X and X ’ be manifolds of class C or C . If
00
f: X X1 is a C -map and F is the set of all points x € X such
that d_.f(Tang„X) / Tang _ . \x ‘, then f (F) is the union of a countable
X X 1 (X/ "
family of nowhere dense sets.

PROOF. Let $ and $' be arbitrary countable atlases of the

manifolds X and X'. For each pair (<p,<p') € <i> x <j>1 , write
supp (p fl f (suppcp') as the union of a sequence of compact sets
K 1 (tp,(p') ,K2 ((p,cp') , .. ., and let C i ((p,cp') be the set of all points
x £ K^(cp,cp') such that the rank of the Jacobi matrix of loc(tp,tp')f
at tp(x) is less than dimX' .Theorem 3.2.3 shows that the sets
f (Ci (cp,cp ')) are all nowhere dense,and obviously f (F) = L^f (C^ (ip,ip')) .
 222

The Basic Theorem

5. Let X 1, and X1 be smooth manifolds, and let A1

and A 2 be subsetsof X1 and Two smooth mapsf 1: X 1 + X' and
f„: X^ -*■ X 1are said to be transverse (oneto the other) on A 1 ,A0 if
2 2 ' 1 z
for any x 1 G A 1 , x 2 GA 2 with f-,^) = f 2 (x2>,the vector space
Tang,, ,X' isspanned by itssubspaces df (Tang X ) and
f (X^ ) X1 1
d f, (Tang X„),and the following holds: if x G 3X , then
x2 “ x2 ^
Tang, , ,X' is already spanned by its subspaces d f (Tang 3x ) and
f 1 ( x 1) 1 1
d f - (Tang X ); if x9 G 3x , then Tang, , ,X' is already spanned
x2 ^ x2 " 11
by d f (Tang X ) and d f (Tang 3X_); and if x G 3X and
X ^ I X^ l x 2 £ X^ ^ I '

x- G 3X„, then Tang, , >X' is already spanned byd f (Tang 3X )

£ 2. r ^tx ^; X1 i x^i
and d f„ (Tang 3X„) . Two maps f : X X' and f„: X„ -> X' which
x2 “ x2 “
are transverse onX ^ ,X ^ aresimplyreferredto as transverse.

Let us make three obvious remarks. First, if

dimX^ + dimX 2 <dim X ' , then thefact thatf^ and are
transverse on A^, A^ implies that f^(A^) n f 2 (A2 > = 0 . Secondly,
if f^ and are transverse on A^,A 2 , then they are transverse on
some neighborhoods of A^ and And thirdly, if X^, X2 , and X'
are of class Cr with 1 < r < 00,and A^ , A 2 are compact, then
givenf2 £ Cr (X2 ,X '), the set of all f E Cr (X^,X') such that f
and f 2are transverse on A^,A 2 is open in Cr (x^,X*).

6 (LEMMA). Suppose X^ and X 2 are C - or Ca-manifolds

and f ^ : X^ -* and f^ : X 2 areC°°-maps. Then there is a

dense set V in 1RQ 1 such that for each vector v E V, the map
Q1
X1 defined by f^(x^) + v is transverse to f .

PROOF. We may take V to be the set of all v £ IRq which

satisfy the following condition. Consider the four maps
int X 1X int X 2 ■+ IRq , int X 1 X 3X2 -> IRq , 3X1 x int X 2 -> 3Rq , and
q1
3X^ x 9X2 + 3R ,given by (x^ ^ 2 ) ** f 2 <x 2 ) *" Then given any
of these maps, v is not the image of a point where the differential
of that map has rank less than q 1. The map x1 (x ) + v with
v £ V is obviously transverse to f 0, and theorems 4 and 3 show that
q'
V is dense in JR^ .

7. Let X r X2 , and X' be C - or Ca-manifolds with

3x' = 0. If f 2 :* 2 ^ X ' is a C°°-map, then the subset of C^fX^X')
 223

consisting of all maps transverse to f2 is the intersection of a

countable collection of dense open sets.

PROOF. For two sets A 1 c= X 1 and A2 c X2, we let FfA^A^

denote the set of all C°°-maps f 1 : X 1 -> X ' such that f and f2 are

transverse on A^,A2 - If A^ and A2 are compact, then F(A^,A2)

is obviously open in C°°(X,j,X'), and we shall presently show that if

A^ and A2 are compact, then F(A^,A2) is dense in (^(X^X'). These
two facts are enough: express X^ as the union of a sequence of compact
sets K ^ / K 2 ,..., express X 2 as the union of a sequence of compact
sets K2<j,K22, . . .,and then observe that the part in which we are
interested (i.e, F(X1 fX 9)) can be written as fl. ,F(K1 .fK 9 .).
\ z. 1 ,d 11 d
Thus, suppose that A 1 and A? are compact, and let (J be
oo
a neighborhood (in C (X^,X')) of an arbitrarily given map g^ : X^ -> X \
We have to produce a map contained inboth F(A^,A2) and U. We do
this in two steps: first, we compress (Jto a a neighborhood I/ of g^
having a more special form, and then construct a map belonging to
F (A^ ,A 2 ) and to (/.
Let dim X^ = q1 , dim X 2 = q2 , and dimX' = q'. To begin
building the neighborhood 1/, fix for each point x 1 £ f 2 (A2) a chart

cp*t € Atl .X ' withImcp', = ]R^ and (p'.fx1) = 0 . Further, fix

x* x' x' x
2
for each point x2 £ A2 a chart tp2x £ Atlx x 2 suc^ that Im tp2x = 3R
2 2 2
q2
or HR_ , cp2x (x2) = 0, Cl supp cp2x is compact, and f2 (Cl supp cp2x ) cz
2 2 2
—1 a1
c [cpi . .](Int D ) . Nowcover A 9by a finite number of sets
2 2
—1 —1 q2 —1 q2
ip2^ (Int D ), say cp2x (Int D ) / •••/<P2x (Int D *' and denote the

chart tpi , . simply by ip\, and the chart - by ip? ..

2 2j : 2j ^
Continuing, for each point 6 A^ choose a chart cp.^ € Atl^ X^

ql ql 1
such that Im tp. = HR or 3R_ , cp1 (x1 ) - 0, Cl supp (p.. is
Ix ^ IX ^ I 'j

compact, and g (Cl supp cp.. ) is contained in one of the sets

1 *1

X ' ^ f _ (Cl supp i>9 .) , 14>' ■]~1 (Int Dq ) , for any given j = 1 ,. ..,1.
-1 ^1
Finally, cover by a finite number of the sets ip1x .(Int D ), say

tp. (Int D n) ,...,cp1 (Int D ), and denote the chart<p. by

1x11 1k 1i
. . At last, we may define 1/ as the subset ofU consisting of all
 224

maps h 1 :X ■+ X ' such that:h 1 (Cl supp ^ 1 ±) <= X ' \ f 2 (Cl supp ^ 2 j )

whenever g.] (Cl supp ^ <= X ' \f2 (Cl supp \p2 ^ ) , and h 1 (Cl supp ^ ± ) <=

c X ' \ (tjj'.) "*(Int ) whenever (Cl supp <= X ' \ (^^ ) 1 (Int Dq ) .
oo
That V is a neighborhood of g^ in C (X^,X') is plain.

Turning now to the final part of the construction, arrange

the pairs (i,j), i = 1 ,...,k, j = 1 ,...,1 , in a sequence
(i w j „),...,( i ,j ), with m = k-1 . Next construct inductively maps
1 1 m m
h^,...,h™: X ^ X ' with the following properties: (i) h^ E I/;

(ii) if r £ s, then the maps h^ and are transverse on

-1 ^1
-1 ^2 m
), ^2 • (D
(D ). Then h^ will belong to V ( A ^ , A ^ ) 0 I/.
1r ^r
0 s
Put h = f 1 and assume that the maps h satisfying (i)
t- 1
and (ii) are already defined for s < t £ m. If the maps h^ and
-1 ^ _1 ^2 t t- 1
f2 are transverse on ijj^ (D ), (D ), put h^ = h^ . Otherwise,
t- 1 -1 ^1
h1 (i/; . (D )) c= (ip'
. ) (D ), and Lemma 6 guarantees the existence
t
q1
of a set V dense in 1R1 such thatthe composite map

loc(<K )tu_1
Im tp . '-JLJULtV-.jRq'
1lt

is transverse to locfip^ . ,\p . )f_: Im \p~ . ■> 3R^ , for each v £ V.

Dt z Z Jt
CO Q 1 Q *
Pick a C -function a: J R 1 + ]R, equal to 1 on D^1 and equal to 0
outside the concentric ball D^' of radius 2. Now for v E V define
g v : x *, ^ x ’ bY

h^ (x 1) , if x., € X 1 ^ (h* V 1[ ( ^ ) 1 (2Dq ')],

gv (x1 } =
(^4 )_ 1 ( ^ (h^_ 1 (x )) + a(<K (h^ - 1 ( x J D v ) ,
Jt t Jt ' '

| if £ (h^ 1) ^ (supp ip^ ) .

We easily see that gQ = h ^ - 1 , that g^ is C°°, and that the map

q1 00
IR C (X^ ,X') , v -* g^, is continuous. Consequently, there is an
open set U c]Rq , such that 0 e U and for v £ U the map g £ 1/,

-1 qi -1 ^9
and gv and f2 are transverse on
\p ' (D '), ip ! (D ) for r < t.
r
On the other hand, the definition of V shows that for v £ V,9 qTr
 225

-i q-, _ ^ ^2
and f2 are transverse on (D ), ^ (D ), and hence one may
t t t
take h^ = g^ for any v £ U H V.

8. Maps Transverse to a Submanifold

1. Theorem 7.7 is used mainly when X2 is a submanifold of

X1 and f2 is the corresponding inclusion. In such a situation we
use a simpler terminology; namely, instead of saying that the map
f^: X^ X 1 is transverse to the inclusion in: X 2 X *, we say that
f-l is transverse to x2 . Then Theorem 7.7 states that if X^ and X1
are of class C°° or Ca and 3X1 = 0,then the set of all maps in
oo oo
C (X^ ,X 1) which are transverse to X2 is dense in C (X^,X').
The more special case, when both X^ and X2 are
submanifolds of a manifold X 1. and f^ and f2 are the corresponding
inclusions, deserves particular attention. If the maps in: X^ X'
and in: X2 X 1are transverse, we say that the submanifolds X^ and
X2 themselves are transverse. Comparing Theorems 7.7 and 1.4, we see
oo
that given arbitrary submanifolds X and X9 of a closed, C - or
C -manifold X 1,every neighborhood of the inclusion in: X 1 X 1 in
CO

C (X^,X 1) contains embeddings transverse to X2 . The last statement

is frequently formulated in a more geometric and less formal fashion:
two submanifolds can be made transverse through an arbitrarily small
displacement of either of them.
< OO
Of course, the C -complement to Theorem 7.7 that was mentioned
in 7.1 (Theorem 9.10) applies to these special cases too. However, note
that in order to bring two submanifolds into general position in this way
one must displace both of them, and not only one.

2. Suppose X and X' are C^r-manifolds (1 £ r £ a) with

dimX^ = q^ anddimX' = q', and X2 is a submanifold of X* with

dim X0 = q 0 . Let f^: X1 X 1 be a Cr-maptransverse to X0 and such

2. £ I I ^
that f (9X )n SX2 C ax’. Then X 12 =f~1 (X2) is a
r —1
(q^+q2~q')-dimensional C -submanifold of X^ with ^ X ^ = f-j (3X2).

Moreover, this submanifold is neat whenever X2 is neat. Finally, the

linear map

factdx / l : -*■ Tangf1 (x1)X ’/Tangf1 (x1)X2 (3)

is an isomorphism for any point € X^ .

 226

(It is understood that a submanifold of negative dimension is

void.)
PROOF. Note that the second assertion is a straightforward
supplement to the first, while the third assertion becomes evident if
one observes that both quotient spaces appearing in (3) have the same
dimension. Therefore, we have to prove onlythe first assertion, and
in order to do this, we follow the scheme in 1.2.12.
Tobegin with, consider, given x^ E x ^2' Possible
positions of the point f^(x^). Thecase f^(x^) E int D 3X* is
excluded, because int X^ H 3X 1 = 0 (X^is a submanifold of X 1)-
If x^ E intX^ ,the case f^(x^) E 3 X 20 3X* is excluded since it
would contradict the assumption that f^ is transverse to X2 (if
x 1 E int X 1 and f1 (x ) E 3X9 D 3X1, then the spaces Imd f and
1 1 l l z x^ 1
Tang^ ^ )9X 2 are both contained in Tang^ )3X', and so they cannot

span Tang^ ^ jX')). When x^ E 3X^ , the relation f^(x^) E 3 X 2 0 int X 1

is impossible since f ^ (3X ^) D 3X2 ci 3X\ Therefore, we are left with

four possibilities: (i) x^ E intX^ , f (x^) E int X 2 0 int X' ;
(ii) x 1 E int X^ , f ^ (X 1 ) E 3X2 fl int X' ; (iii) E 3X^ ,
f^ix^) E int X 2 D int X' ? and (iv) x^ E 3X^ , f^ix^) E 3X2 fl 3X'.

Fix a chart cp1 E Atl^ , .CrX 1 which transforms the triple

r 1 ,xr
(suppcp* ,X2fl suppcp* ,f^(x^)) into oneof the triples (JR
a '^R q 2 ,0),

q' q2 q' q2
QR ¡JR_ /0), or GR_ /3R_ /0),with corresponding local coordinates

cp',.. .,cp' . Let \\) , ...,ip

,:f1 (supp cp1) 1R be the functions
1 M. ' ~i '
i p ^ f x ) = tp|(f^(x)). In cases (i), (iii), and (iv) above, the intersection

n f/isuppcp') isdefined by the equations ip^ +i (x) = 0,...,

ip , (x) = 0, and in the case (ii) - by the same equations and the

inequality ip^ (x) < 0. Let us verify that for all cases (i)-(iv) , the
above equations and inequality satisfy the independence conditions
displayed in 1.2.12.
In case (i) , we have to show that ...... ,(|i . are
ip
q2 q
independent at the point x 1. This follows from the equality
q1
dim (1j_^^ +^Ker (d^ iJj^) = q f-q2 , which in turn follows from the trivial

inclusion dx ^f1 (nq=q^ + 1Ker(dx ^ j)) C Tangf (j£ ) X2 and the equality

Tangf1 (x1)X ' = Imdx 1f 1 + Tan9f (x )x2 (which is part of the

 227

definition of transversality).
In case (ii) , we have to showthat ilJ^.ip „ , .. .,ip . are
r1 rq2+1 'rq ’

independent at . The proof is a repeat of the previous one, except

that one must replace the equality Tang . . X 1= Im d f + Tang,, , ,X0
/ x^ 1 r ^(x^ ) I

by Tang£1 (xl)X ' = I r " a X l f 1 * Ta"9f1 (x1)3X2-

Finally, in cases (iii) and (iv), we have to produce a

r . —1 —1
C -function ip: f^ (suppcp1) -> 3R which is zero on H f^ (suppcp') ,

negative on Int X 1 n f~1 (suppcp'), and is such thatip ,ip^ +1,...,ip^,

are independent at x .
The existence of such a function is equivalent
i _

to the restrictions of ip +1 ,••- , to 8X1 n f 1 (suppcp’) being

q2 * 1 1
independent at x ^ . The latter can be proved as in (i), employing the
equality Tang, , ,X‘ = Im d (f I ) + Tang, ..X~ rather than
^ \ ^ ^ «j I 10 A ^ X ^ ^

Tangf 1 (x1)X ' = ImdXlf1 + Tangfl (Xl )x2-

3 (COROLLARY). Let X^ and be transverse submanifolds

of a smooth manifold X ’, and assume that X^ is neat. Then X^ D X 2
is a (dim X^ + d i m X 2 - d i m X ’) -dimensional submanifold of X ', and is
neat whenever X2 is neat.

The Simplest Applications

£■
4. Let A be a closed subset of a closed C -manifold X,
and let U be a neighborhood of A. _If 1 £ r £ °°, then thereis in
U a compact submanifold B of codimension 0 such that A c int B .

PROOF. Let (f>: X I be a Urysohn Cr-function for the pair

A, X ^ U (see 4.7), and suppose that c £ (0,1) is not a critical
value of tj> (see 1). Set B = (p ([0,c]) . Then B is the preimage
37
of the submanifold (-°°,c] of JR under the composite C -map

X — I ~*~n > 3R, Since the latter is transverse to (-°°,c], B is a

submanifold of codimension 0. It is immediate from the construction
that B is closed as a subset, that A <= int B , and that B c= U.

5. Every CNRS is homeomorphic to a retract of a closed,

00
orientable, C -manifold.
q
PROOF. Let j be an embedding of the CNRS X m S ,
with q large enough, and let U be a neighborhood of j(X)which
retracts on j(X). Theorem 4 provides a compact submanifold B <= U
 228

such that B 3 j(X), and clearly j(X) is a retract of the double of

B.

9. Raising the Smoothness Class of a Manifold

1. The main results of this subsection are Theorems 6 and

Lemmas 2-5 are needed for the proof of Theorem 6, while Lemma 7 in
conjunction with Theorem 6 yield Theorem 8.

2 (LEMMA) . Let X and X' be C^r-manifolds, 1 £ r £ 00,

X compact, and 3X ' = 0 . Suppose A' is a submanifold of X',
f : X •+ X ' is a Cr-map transverse to A' such that f(3X) c X' \ A ',
—1 r
and p: X f (A') is both a retraction and a C -submersion. Then
there is a neighborhood LI of the map f in Cr (X,X') such that
every g £ U satisfies:
(i) g is transverse to A ';
(ii) g (9X) c X ' \ A ' ;
(iii) 0 : g -1 (A') f (A') is a submersion,
g '(A')

PROOF. We obtain U as the intersection of three open sets,

and U^. U1 is the set of all maps in Cr (X,X') which are
transverse to A'. U2 is the set of all g £ Cr (X ,X 1) with
g(9X) <= X' v A'. Finally, U3 is the set of all g e Cr (X,X ') such
that the intersection of the subspaces Ker d^p and

(dxg) (Tang g ,\x)vA') of Tang x

X reduces to 0 for all points
x £ g (A') .
We alreadyknow that is open(see 7.5). The openness of
U2 is a consequence of the compactness of 8X and the openness of
X' \ A'. To prove that is open, we shall describe it in adifferent
way. Fix a Cr-embedding j: X -*■3Rq and let C c Tang X be the subset
of all vectors u such that dp(u) = 0 and <dj(u),dj(u)> = 1. Then
clearly U3 is just the set ofall g £ Cr (X,X') suchthat
dg(C) c Tang X ' ^ Tang A' . The openness of this last set is a
consequence of the following facts: C is compact, Tang X ' \ Tang A '
is open (in Tang X 1) ,and the mapping Cr (X,X') -* Cr_ 1 (Tang X ,Tang X ') ,
which takes each g £ C (X,X') into dg, is continuous (see 1.1) .
Therefore, U is open, and we see at once that f £ U and
that any map g £ U satisfies (i) and (ii). it is easily checked that
(iii) also holds:indeed, g £ U implies that g_ 1 (A') is a neat
submanifold of X (see 8 . 2 ) ; since g £ U2 , g_ 1 (A') <=int X , and as
 229

such it is closed as an independent manifold. Finally, g E li

implies that the differential d (o * ) is nondegenerate at all
-1 x g (A’)
points x E g (A 1) .

3 (LEMMA). Let X and X1 be smooth closed manifolds of

equal dimensions, and let f: X X ' be a submersion. If X’ is
connected and the preimage under f of one of its points reduces to a
point, then f is a diffeomorphism.

It suffices to show that f is invertible. Let A' = f(X),

i _
and denote by B' the set of all x' E X' such that f (x') consists
of more than one point. Being a submersion, the map f is open (see
1.5.8), so A’ is an open set. Also, B' is open: if f(x^) = f(^2) =
= x' and x^ f x^/ then x^ and x^ have disjoint neighborhoods,
U1 and U^, in X such that f| and f| are differentiable
|U1 I 2
embeddings, and f(U^) 0 fi^) is a neighborhood of x' contained in
B'. On the other hand, A* is closed because X is compact. Also,
B' is closed, because if U^,...,Ug are open sets covering X and
such that the restrictions fly are differentiable embeddings, then
' i
s
B' = n f ( X \ U. ) ,
i= 1 1

and the sets f(X \ U^) are closed. Finally, since X' is connected,
A' 7^0, and B' i X', we have A' = X' and B' = 0, i.e., f is
invertible.
r Q
4 (LEMMA). Suppose X is a compact C-submanifold of M 1,
1 £r £00, U is a neighborhood of X in , and X' is an open
subset of a closed Ca-manifold admitting a Ca-embedding in Euclidean
space. If f E Cr (U,X’), then every neighborhood in Cr (X,X’) ojf
the restriction f|w contains the restriction of some map belonging to
■
------ — ■
—■ x
ca (U,X*) .

PROOF. For a start, suppose that X' is itself a closed

Ca-manifold. Denote by U the given neighborhood of f|x in Cr (X,X'),
and fix a Ca-embedding j ': X ' ->• IRq ,a Ca-transversalization t ' of
j ', and a neat tube TubT ,p '. Since

Cr (id,prT ,): Cr (X,tubT ,p ') + Cr (X,X')

is continuous, the preimage of U under this mapping is open; moreover,

it is not empty because it contains the composite map
[ab j' : X ’ -»■ tub ,p ’] o f. Therefore, Theorem 3.1.4 yields a map
g: U -+ tubT ,p ' with polynomial components, such that g|x belongs to
the above preimage of U. Clearly, the composition prT , ° g is the
 230

desired map belonging to Ca (U,X').

One can reduce the general case to the above situation: if
X' is an open subset of the closed Ca-manifold Y, then

Cr (id,in) : Cr (X,X') -» Cr (X,Y)

is atopological embedding with open image, and transforms C (X,X')

into the intersection of this image with Ca (X,Y) (cf. 4.6).
JO
5 (LEMMA). For each compact, q-dimensional C -submanifold
X of 3Rq , 1 £ r £ a, and each compact subset A of_ lRq , the set
of all Cr-embeddings f: X -* ]Rq with f(IntX) A is open in
Cr (X, TRq ) .
(In the present subsection we shall apply Lemma 5 only in the
case where A is a point. However, we shall need it in full generality
in the next subsection.)

PROOF. For a start,suppose that A is the ball in lRq with

center c and radius p. Let f: XlRq be an embedding of class Cr
such that f(IntX) =3 a, and denote by (J the set of allCr-embeddings
g: X 3Rq satisfying for any x E X the inequality

dist(f(x) ,g (x) ) < min(p,Dist(f(Fr X) ,A) ) .

Since U is open in Cr (X,3Rq ) (see 1.4) and f E U, it is enough

to show that g(IntX) A for any g EU. But if g E U, then
g(FrX) fl A = 0 and thus the two sets g(X) fl A and g(IntX) n A
are equal, while the first is closed in A and the second is open in
-1
A. Moreover, g (Int X) fl A 3 g (f (c ))(since dist(g(f - 1 (c)),c) < p),
so that g(IntX) fl A ^ 0. Consequently, g(IntX) fl A = A, i.e.,
g(IntX) 3 A.
The more general situation where A is the union ofa finite
number of balls reduces to the case already considered. To prove the
theorem in the most general case, it remains to observe that for any
-embedding f: X -* nRq with f (Int X) 3 a, there is a finite number
of balls whose union contains A and is contained in f(IntX) .

6. Every closed Cr-manifold X with 1 £ r £ 00 is

Cr-diffeomorphic to a Ca-submanifold of Euclidean space.

It is sufficient to consider a connected manifold X. Fix

a Cr-embedding j: X IRq , a Cr-transversalization t of j, and a
neat tube TubTP. Consider the map f: TubTP -+ G ’ (q,n=dim X) which
takes each point y € TubTpinto the plane y - j opr^(y) + (To pr^fy))1

(which passes through the point y - j ° prT (y) and is orthogonal to

 231

T ° PrT (y))* It is clear that f is transverse to G(q,n) (even the

maps f| , x C X, are transverse to G(q,n)) and that
'prT (x)
— 1
f (G(q,n)) = j (X) . Pick some point Xq £ X and let V <= G' (q,n) be
the set of all planes transverse to t (Xq) (i.e., intersecting t(Xq)
at only one point) . Finally, let tt be the map V -*■ t(Xq) which
takes each plane belonging to V into its intersection with t(Xq).
According to Lemma 2, f|Tub (p/2) has in Cr (Tubx (p/2),G'(q,n)) a

neighborhood U such that if g 6 U, then:

(i) g is transverse to G(q,n);
(ii) g(9TubT (p/2)) c G'(q,n) \ G(q,n);
“1
(iii) ab pr ^ : g (G'(q,n)) + X is a submersion.

Moreover, as Lemma 5 shows, the set of all Cr-embeddings

<|>: dT (xQ ,p/2) + t (xQ) with ifdntd (xfl, q/ 2 ) ) 3 0 isopen in

Cr (dT (Xq ,p/2) /T (xQ)) • Hence fjTub (p/2) has in Cr (TubT (p/2),G1(q,n))

a neighborhood 1/ such that for each g £ 1/ one has g(d (xQ ,p/2)) c: V
r ^
and tt o [ab g : d^(xQ,p/2) V] is a C -embedding whose image contains
0. Since G'(q,n) is an open subset of a closed Ca-manifold (see
2.2.10), Lemma 4 guarantees the existence of a Ca-map
h: tub^p -+ G' (q,n) whose restriction to Tub^ (p/2) belongs to (J f11/.
i -1
Set Y = th|Tuk (p/2)-^ (G (q,n) ) . By virtue of Theorem 8.2, Y is a

Ca-manifold, and we have only to verify that pr^|Y : Y X is a

diffeomorphism. But this claim follows from Lemma 3. Indeed, since

h Im , / /ox E U fl 1/, the conditions of this lemma are satisfied: the
|Tub^_ (p/2)
fact that h|_ , , /0v £ U shows that pr is a submersion, while
Tub (p/2) t Y
i T i -1
h j^ b (p/2) ^ ^ implies that the preimage (pr^_|Y) (x^) reduces to

a point.

7 (LEMMA) . Let X be a closed Cr-manifold with 1 £ r £ 00,

and let A and B be compact submanifolds of X such that A c: int B
00
and dim B = dim X . If B
C -structurewhich is is endowed with the
OOjr
the restriction of its C -structure, then for any closed C -manifold
X', that part of the space Cr (X,X') consisting of all extensions of
00 r
maps from C (A,X') is dense in C (X,X').

PROOF. Suppose f € Cr (X,X') and U is a neighborhood of

f in Cr (X,X')• We have to produce in U a map extending a map from
C°° (A ,X 1) . Fix a C°°-embedding j': X' JRq , a C°°-transversalization
 232

t ' of j', and a neat tube Tub^,pf. Let 1/ c Cr (B,X') be the set
of all maps g such that

max dist (j ' ° f (x) ,j ' (g (x) )) < Dist (j '(X ') , IRq ' v tuh>T ,p ') .
x£B

Clearly, I/ is open and the segment withendpoints j' ° f(x) and

j ' (g (x) ) lies in tub^,p', for any g £1/ and x £ B. Now choose a
neighborhood U of 9B in B such that Cl U D A = 0, construct a
Urysohn C°°-function <j>: B I for the pair Cl U ,A, and consider the
mapping <J>: 1/ Cr (X,X') which takes g £ V into the map

f(x), if x £ X \ B.

prT ,( (1-<|>(x) )j 1 ° f (x) + <J>(x) j ' (g (x) )) , if x £ B.

Obviously, $ is continuous, and § (f | ) = f. These

-1
properties of $ show that the set $ (Li) is open and nonempty.
— 'I oo
Applying Theorem 6.5, we deduce that $ (U) contains a C -map.
oo
Finally, note that $ transforms C -maps into extensions of maps from
Cro(A,X') .
27
8. Every compact C -manifold with 1 £ r < °° is
27 oo
C -diffeomorphic to a C -manifold. Moreover, if X is a compact
27 oo
C -manifold with 1 £ r < °°, Y is a C -manifold, and Y 3X
27 oo
is a C -diffeomorphism, then there exists a C -manifold X' together
27
with a C -dif feomorphism <j>: X ->• X', such that the composite map

Y _JL 3X ax'

is a C -diffeomorphism.

PROOF. Composing the Cr-diffeomorphism

\p x id: Y x [-1,1] 3X x [-1,1] with an arbitrary two-sided
r r
C -collaring 9X x [-1,1] doppX , we obtain a C -embedding
Y x [-1,1] -+ doppX . Since Y x [-1,1] is a C -manifold, the image B
oo
of this embedding inherits a C -structure which is the restriction of
the induced C -structure, and obviously 3X <= intB . Theorem 6
provides a closed Ca-manifold Z together with a Cr-diffeomorphism
dopp X -+ Z . In addition, Lemma 7 and Theorem 1.6 imply that this
Cr-diffeomorphism may be taken of class C°° on 9X. Now set X' to
be the image of X under the diffeomorphism chosen in this way, and
take tj> to be the compression of this dif feomorphism to a
dif feomorphism X X '. It is immediate that X' and <Ji have the
desired properties.
 233

9 (INFORMATION). Theorems 6 and 8 can be substantially

strengthened: they are valid for noncompact manifolds too, and in
Theorem 8 one may replace oo by a. However, one cannot eliminate the
hypothesis that r ^ 1 from these theorems: there exist closed
topological manifolds which are not homeomorphic to smooth manifolds.
The first example of this kind appeared in [12].

Application: A Supplement to Theorem 7.7

10. Let X^ and X* be compact C" -manifolds with

1 ^ r ^ oo. if
isclosed,then the pairs
x1 (f1,f9) of transverse
r r
maps form a dense set in C (X^X') x C (X2 ,X')-
oo
PROOF. By Theorem 8, there exist C -manifolds Y^, and
Y', together withCr-diffeomorphisms Y^ -* X ^ , Y 2 X2 ' and Y ' X '’
Hence it is enough to prove thatthe pairs of transverse maps are dense
r r
in C (Y ^ ,Y 1) x C (Y^/Y'). By Theorem 7.7, these pairs are dense in
00 00
C (Y^,Y 1) x c (Y^/Y1),and according to Theorem 6.5 this product is
dense in Cr (Y1,Y') x Cr (Y2 ,Y').

10. Approximation of Maps by Embeddings and Immersions

1. In this subsection we complete the program outlined in

4.1. The main results are Theorems 4 and 5. Lemma 3 is our basic tool,
and this same lemma can be used to derive many other corollaries; see,
in particular, 11.11, 11.12, and 11.13.

Auxiliary Manifolds

>27
2. Suppose X is a closed n-dimensional C' -manifold with
1 < r £ a, and let j: X IRqbe anembedding of class Cr . Also,
let m be a positive integer such that 0 < m < q. We shall need two
constructions.
The first construction: we denote by Aux^ or, more
specifically by Aux^(j;m), that subset of Tang X x G(q,m) consisting
of the pairs (u,y) such that <dj (u) ,dj (u)> = 1 and dj (u) G y.
Further, let aux1 : Aux1 ^ G ’(q,m) be the map defined by
aux1 (u,y) = j(pr(u)) + y, where pr = [pr: Tang X + X]. Aux1 is a
 234

[2n-1 + (m-1) (q-m)]-dimensional submanifold of Tang X x G(q,m) [indeed,

it is the preimage of the submanifold ^ x G(q,m) of x g '(q,m)
under the mapping Tang X x G (q,m) 1R^ x G' (q,m) given by
(u,y) h- (dj(u),dj(u) + y ) ; this mapping is transverse to
x G(q,m)]. aux is of class Cr 1 and its image consists of
those m-planes of 3R^ which contain lines tangent to j (X) .

The second construction: we denote by Aux2 or, more

specifically by Aux^ijjm), the subset of X x x x G(q,m) consisting
of the triples (x,x',y) such that xf xf and j (x 1) - j(x) € y.
Further, let aux2 : Aux2 G' (q,m) be the map defined by
aux2 (x,x 1,y ) = j(x) + y. A u x 2 is a [2n+(m-1)(q-m)]-dimensional
submanifold of X x X x G(q,m) [indeed, it is the preimage of the
submanifold G(q,m) of G ’(qfm) under the mapping
((X x X) \ diagX) x G(q,m) ->G 1 (q,m) given by
(x,x',y) j(x') - j (x) + y; this mapping is transverse to G(q,m)].
r cr
aux2 is of class Cand its image consists of those m-planes of 1R^
which intersect j(X) at more than one point.

The Basic Theorems

>if
3 (LEMMA). Let X be a closed n-dimensional C ' -manifold,
q 27
1 £ r £ 00, and let j : X ]R^ be an embedding of class C together
j7
with a C -transversalization
------------------------------------------------------ t : X G(q,q-n) and a neat tube Tub T p.

Then there exist a neighborhood U of the map t~: X G f (q,q-n)

r r
(see 3.2) in C (X,G' (q,q-n) ) and a continuous mapping $ : U -*■ Diff X
such that, for each map g £ U:

(i) [j ° 0(g)] (x) £ g(x) for all x £ X;

(ii) the map T :X -+ G(q,q-n), T„(x ) = g(x) - [j ° 0(g)] (x)

y 9
is a transversalization of the embedding j ° $ (g ) : X IRq ;

(iii) some neat tube of this transversalization contains

TubT (p/2) .

PROOF. Given g £ CT (X ,G ’ (q ,q-n) ) , let h : Tub p -* m q

g t
denote the map which carries each point y E Tub^p into its image
under the orthogonal projection onto the plane g(pr (y)). Obviously,
g n- hg is a continuous mapping Cr (X,G ’(q ,q-n)) -* Cr (Tub^p ^Rq ) , and

if g= t , then h^ = [in: Tub^p -*■]Rq ] . Consequently, t~ has a

neighborhood 1/ inCr (X,G’ (q,q-n) ) such that, for any g £ V, h

g
 235

is a Cr-embedding and hg (tubTP) => TubT (p/2) (see 9.5). For g € 1/,
let i denote the composition
- 1
h • (ab h ) 1 . pr
x — -J--> TubT (p/2) ----- 2--- ► h~ '(TubT (p/2) ) TubTP ---^ X

An obvious verification shows that g *+ i is a continuous mapping

r 9 ^
1/ -+ C (X,X), and that i^~ = id. Therefore, x has a neighborhood
Ü in1/ such that i is a dif feomorphism for all g EU. Set
-1 ^
$ (g) = i for g € U . It is immediate that $ is continuous and that
U and $ satisfy the conditions (i)-(iii).

4. Suppose X
is a compact n-dimensional C' -manifold,
^r
1 £ r £ 00, and X1 is a closed n-dimensional C -manifold. Then for
n' ^ 2n, Immr (X,X') is dense in Cr (X,X*), and for n* ^ 2n +1 ,
Embr (X,X ') is dense in Cr (X,X').

Without loss of generality, we shall prove these statements

in the case r =00; when r < °°, we simply apply Theorems 9.6 and 6.5
to reduce to the first case.
oo
Let fEC (X ,X '), and let U be a neighborhood of f. We
have to show that, for n ! ^ 2n, U contains an immersion, and that
for n 1 ^ 2n+1, U contains an embedding.
OO Q Q 1
Fix C -embeddings j: X 3R1 and j': X 1 + M 1 , and define
an embedding J': X'+IR^+(^ = x 3R^
by the formula
oo
J'(x!) = (j'(x'),0). Further, pick a C -transversalization t ! of J1
and a neat tube TubT ,p '. Then Jfc(x) = (j '(f (x),tj(x)) defines a

C°°-embedding J^: X 3R^+C* , for any fixed t > 0. Pick £ small

enough so that J£ (X) <= TubT ,(p'/2) and denote J£ simply by J.
Applying Lemma 3 (to the manifold X', the embedding J', the
transversalization x' and the tube Tub .p '), we conclude that there
T
are a neighborhood U1 of the map (t ') in C (X',G '(q'+q,q'+q-n'))
and a continuous mapping U' -> Diff X' such that, for each g' £ U':

(i) [J1° $' (g1)] (x1) € g 1 (x1)for all x' 6 X ';

(ii) the map t ^,:X'G (q '+q ,q '+q-n ') , defined as

t ' , (x ' ) = g ' (x ') - [ J ’ o <5 ' (g ' ) ] (x ' ) is a transversalization of the

embedding J' ° 1 (g ’) ;
(iii) some neat tube of this transversalization contains
TubT , (p'/2).
Now consider the mapping V: W -+ Cm ( X , X ' ) given by
[T '(g')](x) = pr , (J(x)). We see that Ÿ’ is continuous and
V
 236

r (x') = f. Hence, (f) 1 (U) is open in C°° (X ',G ’(q '+q, q '+q-n 1) )
- *]
and nonempty. This fact together with Theorem 7.7 show that (H") (U)
contains a map h transverse to each of the maps

aux^ : Aux1 (J; q 1+q-n ') -+ G 1 (q 1+q,q 1+q-n ')

and
a u x 2 : A u x 2 (J; q '+q-n ') -s- G 1 (q 1+q ,q 1+q-n ') .

We shall presently show that r(h) is an immersion for n' 52n, and
a differentiable embedding for n' ^ 2n+1. This will complete the proof,
since ¥ 1 (h) £ U .
The inequality n' ^ 2n is equivalent to

dim X ' + dim Aux1 (J;q'+q-n') < dim G 1 (q '+q,q 1+q-n ') ;

hence for n' ^ 2n the fact that aux^ and h are transverse means
that h(X') does not intersect Imaux^ , i.e., none of the planes
h(x'), x' £ X', contains a line tangent to J(X). The latter, in
turn, means that the differential d (pr ,|T #v x) of the restriction
y Th '
Pr i I-r/v\ is a monomorphism for any point y £ J (X) . Thus pr i|T/vx/
|J(x ) ^h I
and also ¥ 1 (h), are immersions.
Similarly, the inequality n' > 2n+1 is equivalent to

d i m X 1 + dim Aux^ (J; q '+q-n ') < dim G ' (q '+q ,q 1+q-n 1) ;

hence for n* ^ 2n+1 the fact that aux^ and h are transverse means
that h(X') does not intersect Im aux^ , i.e., none of the planes
h(x'), x' £ X 1, intersects J(X) at more than one point. This says
that the restriction pr , |T ,v v is injective. We see at once that
Th I
y ' (h) is also an injective map, and since it is an immersion, ¥ 1 (h)
is a differentiable embedding (see 1.5.4).

Embeddings and Immersions in Euclidean Space

>£•
5. Every compact n-dimensional C' -manifold with 1 £ r £
can be Cr-immersed in IR2n and Cr-embedded in 3R2n + 1.

This is a consequence of Theorem 4 (where we take X to be

the given manifold, and set first X 1 = S2n, and then X 1 = S2n+1;
we disregard the trivial case n = 0) .

6 (INFORMATION). As a matter of fact, every n-dimensional

C^r-manifold with r > 1 admits a Cr-embedding in 3R2n whenever
 237

n * 1, and a Cr -immersion in 3R2n_1 whenever n 5 2. If n > 0 and

is not a power of 2, then every C ^ - m a n i f o l d with r^ 1 admits a
C -embedding in 3R . However, for each n = 2s , s3: 0, there are
smooth, closed, n-dimensional manifolds which cannot be even
2 n —1
topologically embedded in IR (such an example is IRPn ). Every
r
n-dimensional C -manifold, r ^ 1, without closed components, can be
Cr-embedded in HR n 1 . Every orientable n-dimensional C^r-manifold
with r ^ 1 and n / 1 ,4 can be Cr-embedded in lR^n \ [it is not
known whether a smooth, closed, orientable, four-dimensional manifold
admits a differentiable embedding in 1 ; a topological embedding
always exists.]
More information, details, and references can be found in
[20] and [18].

11 . Exercises

>37
1. Show that for any C -manifolds X and X', with
0 s? r £ , the space Cr (X,X') has a countable base.

2. Let X
be a compact C ^ -manifold, 1 £ r£ and let
^r
X' be an arbitrary C -manifold. Show that the set
Submr (X,X') 0(X,X1) is open in C^(X,X').

3. Show that every compact topological manifold is a CNRS.

[cf. 5.10.]
4. Suppose X is the smooth one-dimensional submanifold of
HR', closed as an independent manifold, and containing the graph of the
2
function x X sin(1/x) defined on the interval (0,1). Let x be
2
the normal transversalization of the inclusion X 3R . Show that for
any p the segment d ((0,0),p) intersects a segment d^(x,p) for
some x ^ (0,0). [cf. 3.2.]
5. Suppose X is a compact Cr-manifold with 1 ^ r < co, and
A is a submanifold of X. Show that for a suitable collaring of X,
A U cop(A) is a submanifold of doppX.
6. Let X and Y be compact Cr-manifolds with 1 < r £ 00.
Check that: (i) the product X x Y has a Cr-structure which induces
the usual Cr-structures on int X x Y and X x int Y ; the
Cr-manifold obtained by equipping X x Y with a Cr-structure having
these properties is unique up to a diffeomorphism.
oo
7. Let X 1, X 2 i and X be compact C -manifolds, and let
 238

CO CO
f- £ C _ (X„,X1). Show that the subset of C.(X1,XI) consisting of all
2 3 2. ' _ co d 1
mapstransverse to f2 is dense in Cg(X^,X').

8. Let X^, X2 ,X', and f2 be as inexercise 7. Show that

CO
for any C -map <p: 3X^3X' transverse to ab f2 : dX^ -+ 3X', the
00
subset of C- (X.,X ') consisting of all the extensions of <p which are
o 1 CD
transverse to is dense in the subspace of C^(X^,X') consisting
of all the extensions of $.
>27
9. Let X and X' be compact-manifolds with 1 £ r £ °°,
and let cf): 3X 3X ' be a Cr-immersion. If dimX' > 2 dimX, show
r r
that the subset of C^iX^X') consisting of all C -immersions that
27
extend <p is dense in the subspace of C^(X,X') consisting of all
0
extensions of (p.
>27
10. LetX and X' be compact C " -manifolds with 1 £ r £ 00,
and let <f>: 3X -* 3X' be a Cr-embedding. If dim X ' ^ 2 dim X + 1, show
r r
that the subset of C~(X,X') consisting of all C -embeddings that
27
extend (p is dense in the subspace of C~(X,X') consisting of all
o
extensions of (p. [In particular, every compact n-dimensional
C^r-manifold admits a neat embedding in D2n+\ ]
>27
11. Let X and X* be closed C ' -manifolds with 1£ r £
Show that the set of all C -maps f: X X ' such that
Tang,. , vX' = Im d f + Im d f for any two distinct pointsx ,x € X
I IX<| J X ^ X2 1 z

with f(x^) = f(x2), is dense in Cr (x,xf).

12. Let X and X* be closed C^r-manif olds with 1£ r £ .

If dim X' = 2 dim X - 1, show that the set of all Cr-maps f: X X'
such that rank d^f = dim X for all but a finite number of points
x E X, where rank dxf = dim X - 1, is dense in Cr (X,X')-
>27
13. Let X and X' be closed C " -manifolds with 1£ r £ 00.
If 2 dimX' > 3dimX, show that the set of all Cr-maps f: XX'
such that the preimage of each point of X' under f contains at most
27
two points, is dense in C (X,X')-
 239

§5. THE SIMPLEST STRUCTURE THEOREMS

1. Morse Functions

1. The central result of this section is Theorem 2.10, whose

main conclusion is that every compact n-dimensional C -manifold can be
obtained from an empty n-dimensional manifold through a finite number
of fairly standard operations, namely, by attaching handles. The entire
present subsection and that part of Subsection 2 preceding Theorem 2.10
are essentially devoted to the preparation of its formulation and proof.
The remaining part of Subsection 2 contains corollaries of Theorem 2.10.
In Subsection 3 this theorem is used to effectively classify the compact
smooth two-dimensional manifolds.
It should come as no surprise that, in contrast to the
oo
previous sections, here we consider, in general, only the C -case: the
theorems concerning smoothing of diffeomorphisms and manifolds (i.e.,
Theorems 4.4.4, 4.6.11, and 4.9.6, 4.9.8) show that we may replace the
oo 27
class C by any class C , 1 £ r < 00, without affecting the theory
discussed here.

Cobordisms and Morse Functions

00
2. A compact C -manifold X is called a cobordism if its
boundary 3X is the disjoint union of two parts, 3^X and 3^X, each
consisting of whole components of 3X.Those two parts are termed the
beginning and the end of the cobordism X.Each of them may be empty?
when both are empty, X is closed. In general, given a compact
C°°-manifold, one can transform it into a cobordism in 21 ways, where
1 is the number of components of 3X. Among these cobordisms, there is
one without beginning (3QX = 0, ^ X = 3X) and one without end
(3QX = 3X, 31X = 0) .
Two cobordisms, X and X', are said to be diffeomorphic if
oo
there is a diffeomorphism (and hence a C -diffeomorphism) f: X X1
such that f(3^X) = 9q X' and f (3^X) = 3^X!.
 240

Suppose that X and X'

are two cobordisms such that 3 ^
0°
and are dif feomorphic, and let cp: 3^X 3q X' be a C -diffeo-
morphism. Then one can form a manifold Y by glueing the somehow col­
lared manifolds X and X' with the aid of cp. Now Y naturally
becomes a cobordism if we set 3q Y = 3q X and 3^Y = 3^X'. We say that
the cobordism Y is the result of glueing the cobordismsX and X1
by cp. If the cobordisms X and X' are oriented and cp is
orientation reversing (here the orientations of 3^X and are
those induced by the orientations of X and X 1; see 1.3.4), then one
can orient Y in such a manner that both embeddings, X Y and
X' Y, become orientation preserving. Warning: this definition of
the orientation of the glued cobordism is not in accordance with the
definition of the orientation of a glued manifold, given in 4.5.5.
Two smooth closed manifolds, and , are cobordant if
there is a cobordism with the beginning and the end diffeomorphic to
Vq and , respectively. If, in addition, Vq and are oriented,
and there is an oriented cobordism X such that one of the diffeo-
morphisms V q -> 3q X and -* 3^X preserves orientation, whereas the
other reverses it, then we say that Vq and are oriented
cobordant. Clearly, the cobordism and oriented cobordism relations are
reflexive and symmetric, and since cobordisms can be glued, they are
also transitive, i.e., they are genuine equivalence relations.
2
3. A critical point x of a C -function f: X IR, where
^2 2
X is aC -manifold is nondegenerate if for some chart cp £ Atl^C X
(and hence for any such chart) cp(x) is a nondegenerate critical point
of the function (f I J ° cp 1 : im cp ]R (see 3.3.1). The
|supp cp
corresponding index is independent of the choice of the chart cp (see
3.3.1), and is called the index of the point xrelative to f.

Suppose X is a cobordism, and let f : X + IR be a

°o _1
C -function; f is a Morse function if: Im f c I; f (0) = 3„X, and
0
f (1) = 3^X ; and all critical points of flie in int X and are
nondegenerate.
We say that a Morse function is proper if its values at
distinct critical points are distinct.

The Local Structure of Morse Functions

4. Suppose that X is an n-dimensional cobordism and

f: X-+JR is a Morse function. Then for every point x E X there is a
 241

chart (p£ Atl X with cp(x) = 0, such that therestriction fl

x |supp cp
coincides with the composite map

cp ► im cp -- > 3R ,
supp cp --

where the second arrow denotes one of the following functions:

(t^,...,^) *■> -t^ f if x £ 9qX;

(t^,...,^) ^ , if x £ 9^X ;

(t^,...,t^) *-► f (x) + t , _if x £ int X is not acritical

point of f ;

(t
I n
f (x) - t 2
1
- .. . - t2
k
+
k +1
+ ... + t2 ,
n' —
if X

is a critical point of index k.

To prove the first three cases we need only remark that the
function X 3R, defined by y b» f (y) - x for x £ int X U 9^X

(respectively, by y h- -f (y) for x £ 3QX) can be completed, in a

neighborhood of x, to a system of coordinates (see 1.2.12). For the
fourth case, we refer to Theorem 3.3.3.

5 (COROLLARY). A Morse function has only a finite number of

critical points.

An Existence Theorem

oo y\
6 (LEMMA). Given any C -function f: D + 1R, there exists
an open dense subset A of lRn such that for a £ A the function
Dn 1R defined by

x f (x) - <a ,x> (1 )

has no degenerate critical points.

PROOF. Consider the map gradf : Dn + 3Rn with coordinate

functions D f,...,Dnf. One may take A to be Mn \ [gradf](F) ,

where F = (x £ Dn | rank d^grad f < n}. F is clearly open, and Theorem

4.7.4 implies that itis also dense in 3Rn . Moreover, i’t is evident
that if x£ Dn is a critical point of the function (1), then
[grad f] (x) = a, andthe matrix of the second-order partial derivatives
of (1) at x is precisely the matrix of the differential d^gradf
 242

relative to the standard coordinates in Tang x Dn and Tang a IRn .

Therefore, if a £ A, then this matrix of second-order partial
derivatives is nonsingular.

7. On every cobordism there is a proper Morse function.

First, let us show that if there exists some Morse function

on the cobordism X, then there exists a proper Morse function on X.
Let x„ ,...,x be the critical points of the Morse function f: X -> IR,
1 m
and let ,...,U^ be pairwise disjoint neighborhoods of these points
in intX . Further, let
V . ,...,V be neighborhoods of x . ,...,x_
1 m 1 m
such that Cl V, cz U. , ...,Cl V e U , and let <j) ,.. .,<j> be Urysohn
1 1 m m 1 m
C -functions for the pairs (X \ U. ,Cl V. ) ,.. ., (X \ U ,C1 V ) . (The
1 1 m m
existence of such Urysohn functions results from Theorem 4.7, applied
here to the double of the manifold X.) Clearly, if the numbers
e.j ,..., are small enough, then the function X -> IR defined by

x * f (x) + G.d).(x) + ... + e <b (x) is, together with f, a Morse

1 1 mm
function, and its only critical points are x „ ,...,x . Moreover, its
1 m
values at the points x . ,...,x are f(x.) + e.,...,f(x ) + e , and
1 m 1 1 m m
so, by choosing suitable '•* *'Gm' °ne Can ^o r c e t h e s e values to be
distinct.
Now we show that there exists Morse functions on X. Fix a
03
collaring k: 9X x I -*■ X and a C -function g : I -* I which equals 1
on the segment [0,1/2] and vanishes on a neighborhood of 1. Next
define g: X + JR by the formulas

g(x) = 1 , for x £ X \ k(9X x [0,1)),

g (k (z ,t )) = j + -l(t - 1) (3(t) , for z £ 3qX, t £ I,

g (k (z,t) ) = j + -1(1 - t)g(t), for z £ 3^, t £ I.

It is clear that g is C°°,takes X into I, 3QX into

0, and 9-jX into 1, and has no critical points in k(3X x [0,1/2]).

Pick charts (p ,... ,cp £ Atl X such that Imip = ... = imcp =3Rn ,
i s I s
n = dim X , and the sets (p^ (Int Dn ) ,.. .,(p~1 (Int Dn ) cover

X ^ k(3X x [0,1/2)). Further, choose a C^-function a : IRn -* I which

equals 1 on D and 0 ouside the concentric ball 2Dn of radius
2. Using induction, we shall build functions gQ ,...,g : x IR, such
that g± coincides with g in a neighborhood of 3X, maps intX
into inti , has no critical points in k(3X x [0,1/2]), and has no
degenerate critical points in cp71 (Dn ). Then g will be a Morse
J ' J s
 243

function on X.
Let 9q = 9 and assume that for some k ^ 1 functions
enjoying the required properties are already constructed for i < k.
For each point a £ 3Rn , the formula

gk-1 (x) , if x £ X v. cpk 1 (2Dn ) ,

hci(x) =
gk-1 - <a,ipk (x)>a ° ip^(x), if x £ supp ,

defines a C -function h : X IR which agrees with g. in a

3. A~1
neighborhood of 9X, and for a = 0 coincides with g, 1 on all of
.k i
X. Clearly, the point 0 has a neighborhood U in IRn such that for
a E U the function h^ has no critical points in k(3X x [0,1/2]),
has no degenerate critical points in Uj= ^ ^.^(D11), and maps int X
into inti . Thus, by Lemma 6, we can find a E U such that h also
1 n a
has no degenerate critical points in (D ) , and hence we can take
g, = h for such a value of a.
k a

2. Cobordisms and Surgery

1. This subsection is devoted to two types of special

operations on cobordisms, called attaching of handles and spherical
modifications. The former were already mentioned in 1.1. Spherical
modifications are simpler that the attaching of handles, but their
applications are more limited: we shall define them solely for the
oo
closed case, and starting with a closed C -manifold they are capable of
producing only manifolds cobordant with the given one.

Standard Cobordisms

2. We define standard cobordisms of two kinds: the standa

trivial cobordisms and the standard elementary cobordisms of index k.
On every standard cobordism there is a standard Morse function.
A standard trivial cobordism is constructed by taking an
arbitrary closed C°°-manifold V and simply forming the cylinder V x I,
with 9q (V x I) = V x 0 and x I) = V x 1. The corresponding
standard Morse function is defined as (v,t) » t and has no critical
points.
The standard elementary cobordism of index k is defined for
 244

an arbitrary closed (n-1)-dimensional C -manifold V with n ^ k, and

an arbitrary C -embedding cp: Sk ^ x Dn k V, and is denotedby
El(V,cp). Up to a canonical homeomorphism, El (V,cp) is simply
(V x i) U (Dk x Dn k ) , where the map cp : Sk ^ x o n k ^ V x I is
cp^ I

defined by cp^(x,y) = (cp(x,y),1). To define El(V,tp) as a C -manifold,

we let E(n,k) denote the set

<<*, V I * i * 1' Hi < - s iV>

and let el denote the homeomorphism of the intersection

E (n,k) fl []Rn \ (IRk x lnt(^-Dn k ))]

onto the cylinder

,~k-1 . n-k T x ,1 n-k. . .

cp (S x (d ^ I n t (-^D ))) x 1 1

given by the formula

(t r ...,tn ) » (cp((t1 , . . . , t k ) / ( t 2 + . . . + t 2 ) 1 / 2 , ( t k + 1 , . . . , t n )),

O, 1 4-2 ,2 2 , 2. ,
8 { T6 ~ + \ + 1 + ••• + tn )) •

As a topological space, El (V,cp) is the result of glueing the

k-1 i n—k
cylinder [V \ cp(S x int (-^-D ))] x i and E(n,k) by the homeomor­
phism el (see Fig. 5).

\ /
V ( S ° * V 1)

[V\y(S0* \ n i @ 2 ’))\*/

A5 As B5 B8 A? A8 B^

Fig. 5 (n = 2, k = 1)
 245

The C -structure on El(V,cp) is fixed by an atlas consisting

of the charts of arbitrary atlases of the C°°-manifolds

[V \ cp(Sk 1 X (lDn-k))] X I and E (n ,k) fl (IRk x int Dn_k)

(these manifolds are regarded as parts of the space El(V,<p)). The

beginning of 3QEl(V,(p) is formed by the two sets

[V ^ cp(Sk 1 x lnt(^-Dn k ))] x o

and

{,ti v e E(n'k| i- A - *£i * •••

Similarly, the sets

[V ^ (p(Sk_1 X Int (-^Dn_k))] X 1

and

{(t1 V £ E(n' k) I ’ \ + Vh + +tn = l6}

constitute the end 8^El(V,<p). The standard Morse function on El(V,(p)
is given by the two functions

[V \ <p(Sk_1 x Int (l-Dn-k))] x i + 1 and E(n,k) + HR,

defined by the formulas

(v,t) * t
and

(t1 tn ) 8(116 ••• fck + fck + 1 + “ • + tn l

respectively. It has a unique critical point 0 € E(n,k), of index k.

We use the symbol mo to denote the standard Morse function.
Using Fig. 5 as a guide, it is readily seen that the manifold
k n—k
El(V,cp) is homeomorphic to (V x I) U (D x d ). The formulas
«>1
describing this canonical homeomorphism are cumbersome, and we shall
not burden the reader with their precise form.
The basic property shared by all standard cobordisms
constructed from a given manifold V is that the beginning of each is
CO
canonically C -diffeomorphic to V. For the standard trivial cobordism,
this diffeomorphism is v (v,0). In the case of the cobordism
El(V,(p), the diffeomorphism is defined as v h- (v,0) on the part

V <p(Sk~ 1 x Int(lDn_k)) tV v <p(Sk_1 x Int(^Dn~k ))] x 0,

and as
 246

2,1/2
lp((t1 V ' (tk +1 V ’ " (t1 (T6 + fck + 1 + ••• + tn )

+ t2)1/2 t t )
V T 6 + fck + 1 + V ,fc k + 1 ' • • • ' V

on the part
,^k-1 ^n-k.
cp (S x d )

{ (t1,...,tn ) G E(n,k) - t' n 16

+1 + t 2 = - i > -

3. The construction of the standard elementary cobordism

the simplest when the index k is 0 or n.
If k = 0, then (p embeds the emptyset into V, E(n,k)
is the ball an^ el i-s a homeomorphism of empty sets. In this
case the similarity dilatation (with coefficient 4) of the ball E(n,k)
transforms El(V,cp) into the sum (Vx I)j L0**' with
9n [ ( V x I) I |Dn ] = in.(V x 0) and 3.[ (V x I) I ]dR ] = (V x 1) |
1
it also transforms the standard Morse function mo into the function

V x I ir, (v,t) t ,

n— 1
If k = n, then cp embeds S into V, the image of
this embedding being one of the components of V (see 1.5.1). Again,
E(n,k) is the ball ^-Dn and el is a homeomorphism of empty sets.
Here the similarity dilatation of the ball E(n,k) transforms El(V,cp)
into the sum ([V\cp(Sn ^)] x i)| |Dn , with

30 { ([V cp(Sn 1)] X I)J \_Dn } = ([v^cp(Sn 1)] X 0) [sn 1

and
31{ ([V ^ (p(Sn 1)] X I) I |pn } = ini ([V ^ cp(Sn 1 )] X 1) ,

while the standard Morse function mo is taken into the function

[V ^ (P(sn 1)] X I +1, (v , t ) ^ t ,

D m, (t1 tn ) "

We also remark that k = 0 and k = n are the only values

of the index for which the standard elementary cobordism can have a
connected boundary: for k = 0 the boundary is connected if and only
if the initial manifold V is empty , while for k = n the boundary is
 247

connected if and only if cp is a diffeomorphism. We have seen that in

both cases the cobordism is diffeomorphic to Dn .

Trivial Cobordisms and Elementary Cobordisms

4. A cobordism is said to be trivial if it is diffeomorph

to a standard trivial cobordism. A cobordism is an elementary cobordism
of index k if it is diffeomorphic to a standard elementary cobordism
of index k.
From these definitions it follows that if X is a trivial
cobordism, then the manifolds 3^X and 3^X are diffeomorphic, and
that on a trivial cobordism there is a Morse function without critical
points, whereas on an elementary cobordism of index k there is a Morse
function with a single critical point of index k and no other critical
points.

5 (LEMMA). Suppose X and X1 are two cobordisms such that

the manifolds 3^X and ^^X1 are diffeomorphic♦ If X' is trivial,
the the cobordism obtained by glueing X and X' by an arbitrary
oo
C -diffeomorophism 3^X 9q X ! (and using arbitrary collarings of X
and X 1) is diffeomorphic to X.
oo
Theorem 4.5.9 shows that given a C -diffeomorphism
oo
cp: 3^X -> 9q X !, it suffices to produce two C-embeddings,j: X X
and j ': X' -* X, such that j (X) U j 1 (X' ) = X, j (X) flj ' (X1) =
= j (3^X ) = j M S ^ X 1)/ and the composite dif feomorphism

9QX ' -a^ j ' > j ' (9QX ') -^-3--- ► 9 1X

_i
coincides with cp . In order to accomplish this, let us fix:
OO
a collaring k: 9X * I -*■ X; a C -diffeomorphism f: X' + 9QX' x I
such that f (x 1) = (x',0) for all x' € 9QX'; and an increasing
C°°-function a: I + I, such that a(t) = 1/2 + t/3 for t g 3/4 and
a(t) = t for t ^ 7/8. Further, using the function 6: I I,
8 (t) = (1 - t)/2, set

j (x ) = x , if xex^kO^xtoj)),

j(k(z,t)) = k(z,a(t)), if z £ 9^X, t € I,.

and

j'(x') = k o (cp-1 x B) o f (x') , if x' € X'.

 248

We can verify directly that j and j’ have the required

properties.
6. If on a cobordism there is a Morse function without
critical point, then the cobordism is trivial,

PROOF. Let f : X ->■IR be a Morse function with no critical

points. According to 1.5.8 (or, if one prefers, to 4.8.2), the preimage
-1
f ^ (t) of any point t E (0,1) is a neat submanifold of X; moreover,
-1 (t) is obviously closed as an independent manifold (the preimages
-1 -1
f (0) = 8nX and f (1) = 9 X
are also closed manifolds). By
-1
Theorems 4.5.3 and 4.5.8, the manifold f (t) has a neighborhood U
oo -1
together with a C -submersion tt : U -> f (t) which is the identity
-1
map on f (t), for each fixed t E I. Define, for t E I, a C -map
F_j_: Uj_ f ^ (t) x i by = (7Tt (x) ,f (x) ) . Obviously, the
differential d F is nondegenerate for x E f ^ (t), and F induces
x t _1 X:
a diffeomorphism of f (t) onto f (t) x t. Consequently, F is a
oo -1 11
C -embedding on a neighborhood of f (t) (see 1.5.5) , i.e., there is
a neighborhood A^_ of the point t in I such that F induces a
-1 -1
diffeomorphism of F^_ (At ) onto f(t) x Afc. Let m be large enough

so that if we divide I into m intervals of length 1/m, then each

such interval is contained in one of the sets A . Then all the

cobordisms f 1 ([ (i-1)/m,i/m]) , with 3Qf ([ (i-1)/m,i/m]) =

= f ^((i-1)/m) and 3^f 1 ([(i-1)/m,i/m]) = f 1 (i/m), are trivial, and

now Lemma 5 shows that the entire cobordism X is trivial.

7. If on a cobordism X there is a Morse function with a

single critical point of index k and noother critical points, then
X is an elementary cobordism of index k.

The proof is quite long and we shall begin by constructing an

auxiliary cobordism Y. Fix a Morse function f: X I with a single
critical point of index k, say x, and no other critical points. By
Theorem 1.4, there is a chart <p E Atl X, (p(x) = 0, such that f|
x |supp <p
equals the composition of cp with the function Im <p 1R given by

"=1 V ■* £lx) - + ^ +1 ♦ ••• + ^ •

Obviously, Im <p contains the subset A of nRn determined (in the

standard coordinates) by the inequalities:

2 2 2
t, + ... + t ^ 4c ,
k+1 n '
and
 249

TE
16 ‘ - ‘Î
1 - • '• - k * * 1k+1, * • ■• * t2
n * h16 ■
for some c > 0. Let

B = A n ( 3Rk x ( ]Rn-k \ Inti-^D11 k ))) .

As a topological space, Y is the result of glueing the

subset
-1 2 2 -1
f ([ f (x) - jg,f (x) + yg] ) \ cp (A ^ B)
k 1 n.“k
of X and the product S x 2D x i by the homeomorphism

ip’ 1 (B) -> Sk_1 X [2Dn-k V. Int(-lDn-k)] X I

given by
_i 1 '•••' v •
ip ( t 1 , . . . , t n ) H- , (t, 1 , ••-,t ) ,
2 ,.2. 1/2 ' '‘'k+ l'-’- ' V
(t1+ ---+tk )

_ t2 - - t2 + t2 + + t2)
2 16 r 1 ••• rk k +1 ••• V
c

2 2
£ G
[Here we use the inequalities -r-p < f (x) <1 - ^ , which are immediate
lb lb
consequences of the inclusion A c Imip .]
CO
The C -structure of Y is fixed by the atlas consisting of
oo
arbitrary atlases of the C -manifolds

f ([f (x) - ^,f(x) + ^ Cl cp (A ^ B)

and
Sk 1 x lnt(2Dn k ) x I,

The beginning 3^Y is the union of the two sets

f 1 (f (x) - Jg) ^ [cp 1 (A V B) n f 1 (f (x) - fg) ]

and

Sk 1 x 2 Dn k x 0.

Similarly, the end 9 Y is the union of the two sets

2 -1 -1 2
f (f (x) + yg) ^ [cp (A ^ B) n f (f (x) + fg) ]

and
^k-1 _n-k „
S x 2D x 1

The functions
 250

-1 2 2 -1
f ([f (x) - fg,f(x) + ^g]) ^ cp (A B)
R 2

y ^ j ^~Ys + f iy) — f (x) ) ,

£
and

Sk 1 X Int (2Dn k ) X I m,

(u,v,t) h- t [u e sk_ 1 , v e int(2Dn-k), t e i],

yield together a function g: Y IR. It is readily seen that g is a

Morse function with no critical points. (In virtue of Theorem 6, this
implies that Y is a trivial cobordism; however, in what follows this
property of Y is not used directly.) Now consider the composite
embedding
„k-1 „n-k )H.(Z z ,1/2)
S X D ----------------------» S X D X I >

in ^k-1 _ , ,n n-k. in
►S x Int (2D ) x i ------ > y

k—1 n —k —1
and its compression s S x d g (1/2). To complete the proof
of the theorem we shall presently verify that the cobordism X is
dif feomorphic to El (g ^ (1/2) ,\j;) .
As a preliminary step, we find a number 6 , 0 < 6 < 1/2,
OO
and a C -diffeomorphism

H: g” 1 ( j ) * - « 4 + + 6]) '

such that

H ((u,v,-l) ,t) = (u ,v ,t )

for all u £ sk_1, V £ Dn-k, and t £ +6].

By Theorem 4.5.8, one can find n, 0 < n < 1/2, such that
OO —1 1 1 “1 1
there is a C -submersion it: g ((- - n,^- + n) ) + g (j) which is the
—1 1 oo n _V
identity map on g (-j) • Fix a C -function a : IR 3R, equal to 1

on D and to 0 outside 2Dn k , and define a new C -submersion

-11 1 -11
p: g ( (-^ - n /^ + n )) -*- g ^ y t *i e f ° rm ul a

TT (y) , if y £ Sk 1 X int(2Dn k ) X I,
p (y)
tt (u ,v ,-1+ (t - 1) (1 - a (v) ) ) , if y = (u,v,t) ,

where u £ Sk_1, v £ 2Dn_k, t £ I.

 251

Further, define

G: g n^ ^ g ^ x ^ + n)

by G(y) = (p (y) , g ( y ) ) . Since G induces a diffeomorphism of g ^ ( i )

- 1 1 1 ^
onto g (—) x — and the differential d G is nondegenerate as long
-11
as y € g (-^J , there is 6, 0 < 6 < n , such that the restriction of
1 1 1 oo
G to g ([*2 “ 5 ,*2 - 6 ]) is a C -embedding. Now it is clear that one
can take for H the map

(abG)"1 : g"1 (1) x [1 - 6,1 + 6] - g _1 ([1 - ,1 + 6 ] ) .

- 1 1
Finally, to show that the cobordisms X and El (g i-^) /^)
are diffeomorphic, cut each of them into three cobordisms: X into the
2 2 2
cobordisms f ^([0,f(x) - *^|-] ) , f_ 1 ([f(x) -^|-,f(x) + ]) , and

f ^ ([ f (x) + ,1]), and El(g ^(^)/^) into the cobordisms

mo 1 ([ 0 ,-1 - <5]) , mo 1 ([-1 - 6 ,---+ 6 ]) , and mo 1 ([-^ + 6 ,1 ]) . By

Theorem 6, the first and the third cobordisms in each of these triples
are trivial, and hence the second cobordism is diffeomorphic to the
entire cobordism.Therefore, it is enough to exhibit a diffeomorphism
2 2
mo"1 ([-1 - 6,1 +6]) + f_1 ([f (x) - ^§-/f (x) + £§-] ) .

To do this, consider the composite map

(g 1 (1) v ^(Sk 1 x m t D n k ) ) x [1 - fi,l + 6] ab H

(y ^ (sk_1 x m t Dn’k x D) n g~1 ([^- - 6,1 + 6]) -

ld ■>
2 2
(X V cp-1 ( IRk X Int Dn_k)) n f" 1 ([f (x) - T 5-,f (x) + £§-] ) ,

and the map

E (n,k) fl mo “ 5,-^ + 51) --- >

2 2
cp" 1 (lm<p n ( HRk X eDn_k)) fl f _1 ([f (x) - ^§-,f(x) + -^§— ])
given by the formula

(t1 , .. .,tn ) ►> CP 1 (Gt1 ,...,Etn ) ,

which together provide the desired diffeomorphism.

 252

8 (COROLLARY). Suppose that on a given n-dimensional

cobordism X with connected boundary there exists a Morse function
with a unique critical point. Then X is diffeomorphic to Dn .

A t t a c h i n g Handles

9. Let X be an n-dimensional cobordism and let

ip: S^ ^ x Dn k 9^X be aC^-embedding. The result of glueing X and
the elementary cobordism El(3^Xfcp) by the canonical diffeomorphism
3 ^X 3q E1(8^X,cp) (see 2) is said to be obtained from X byattaching
a handle of index k .

By attaching a handle of index 0 to X we replace, up to a

diffeomorphism, X by x| |Dn ; the new component of the boundary,
n“1
i.e., in2 ^S -*-s added to 3^X. To attach a handle of index n,
we actually glue X and Dn by a diffeomorphism of one of the
components of 8^X onto Sn ^.

10. Every cobordism X can be obtained, up to a diffeomor-

phism, from the standard trivial cobordism 3^X x I, by attaching a
finite number of handles. Moreover, given any proper Morse function
f: X 3R, one may choose these handles so that their number will not
exceed the number of critical points of f.

We prove this statement by induction on the number of critical

points of f. If f has no critical points, Theorem 6 suffices. If
f has m ^ 1 critical points, then there is c E (0,1) such that
one of the critical values of f is greater than c, while the
remaining ones are smaller than c. We cut X into two cobordisms:
-1 -1 -1
f ([0,c]) and f (tc ,1]) .On f (t0,c]) there is a proper Morse
function with m-1 critical points, for example, x » f(x)/c. On
-1
f ([c,1]) there is a Morse function with a unique critical point,
for example, x (f (x) - c)/(1 - c) . Finally, note that by Theorem 7
the second cobordism is elementary.
oo
11. A closed n-dimensional C -manifold X on which there is
Morse function having only two critical points is homeomorphic to Sn .

PROOF. We remark (leaving the trivial case n = 0 aside)

that every Morse function with only two critical points is proper (the
two points are necessarily a maximum and a minimum). Thus Theorem 10
shows that X can be obtained from an empty manifold by attaching
two handles. Obviously, the first handle has index 0, and the second
index n, and hence X actually results from glueing two copies of
 253

Dn by a diffeomorphism of Sn ^ .

AHomotopy Corollary

12(LEMMA). The cobordism El(V,tp) ishomotopyequivalent

k k _1
to V IU D , where f: S V is given by f (y) =tp(y,0). Moreover,
there is a homotopy equivalence V Uf D El(V,tp) which agrees on V
with the inclusion V[= B^El (V,tp) ] + El(V,tp).

One can assemble such a homotopy equivalence from the above

v
inclusion V + El (V ,tp) and the embedding D El(V,tp) which takes
k
each point x E D into the point x/4 £ E(n,k). To complete the proof,
it is enough to remark that the constructed mapping V Dk El(V,cp)
is a topological embedding whose image is a strong deformation retract
of El (V ,tp) .

13. Every compact n-dimensional smooth manifold is homotopy

equivalent to a finite cellular space of dimension £ n.

PROOF. The discussion in 1.2 implies that one may assume that
the given manifold is a cobordism with an empty beginning. Therefore,
all we have to show is that if an n-dimensional cobordism is homotopy
equivalent to a finite cellular space of dimension £ n, then it
retains this property after we attach to it an arbitrary handle; see 10.
But from Lemma 12 it follows that attaching a handle of index k to a
k
cobordism X has the same homotopy effect as attaching D to X by
k- ^
some embedding f: S X. Now replace X by a finite cellular space
Y of dimension £ n with the same homotopy type, replace the map f
by its composition with a homotopy equivalence X Y, and subsequently
k-1
replace this composition by a homotopic cellular map g: S Y (see
2.3.2.4). By Theorem 1.3.7.8, the cobordism which results by attaching
a handle of index k to X is homotopy equivalent to the space
k k
Y Ug D ; according to 2.1.5.5, Y D is a finite cellular space of
dimension £ n.

Spherical Modifications

oo
14. Let V be a closed n-dimensional C -manifold, and let
k— 1 n—k + 1 00
tp: S x D ->• V b e a C -embedding. Fix arbitrary collars on
Y ^ cp(Sk_1 x IntDn_k + ^) and Dk x Sn k , and then glue these manifolds
k—1 n —k k —1 n —k
by the dif feomorphism ab tp : S xs +tp(S xs ) of the
 254

boundary of the second onto the boundary of the first. We say that the
glued manifold is obtained from V by a spherical modification along
the embedding cp.* The number k is the index of the modification.

15. If the cobordism X'

is obtained from the cobordism X
k— 1 n —k
by attaching a handle using an embedding cp: S x D ^ 9 ^X , then
9^Xf is obtained from 9^X by a spherical modification along the same
i • /■'«k 1 n-k « ..
embedding cp: S x D + 9^X .

PROOF. Since 9^X ' = 9 ^El (9^X ,cp) , we actually claim that
9 ^El (9^X ,cp)is obtained from 9^X by a spherical modification along
cp. Recall that El(9^X,cp) is the result of glueing the spaces

[9^X v. cp(S^ ^ x int (^-Dn ^) )] x i and E(n,k) by el (see 2) .

Obviously, [9^X \ cp(S^ ^ x Int Dn ^)] x 1 and E(n,k) D 9 ^El (9 ^X ,cp)

are compact (n-1 )-dimensional submanifolds of the manifold 9^El (9 ^X ,cp) ,
which they cover, and they intersect along their common boundary.
Consider the mappings

pr 1 : [a.jX V tp(Sk 1 x Int Dn k )] x 1 -► 9 X \ cp (Sk 1 x Int Dn k )

and

Ip: E(n,k) n 3 nEl (3 1X ,cp) -* Dk x sn-k-1,

\l>(t1 , ... ,tn ) I6,t t , (tk*1.....tn )

15 1..... k (t?
k +1 n
oo —1
It is clear that ip
is a C -diffeomorphism such that pr^ ( (z fZ^)) =
k—1 _k — 1
= cp(z^,z2) for all z<1 G S and z^ E S . Now applying Theorem
4.5.9, we see that 9^El (9^X ,cp) is really obtained by glueing the

manifolds 9^X \ cp(S^ ^ x Int Dn ^) and x sn ^ ^ by ab cp .

16 (COROLLARY). Given an arbitrary cobordism X, the

manifold 9^X can be obtained from 9^X through a finite number of
spherical modifications. In particular, the boundary of an arbitrary
00
compact C -manifold can be obtained from an empty manifold through a
finite number of spherical modifications.

Translator's note: or by doing a surgery on V using cp.

 255

3. Two-dimensional Manifolds

1. The theory presented in the two previous subsections

represents, first of all, an attempt to somehow visualize, survey, and
classify smooth compact manifolds of a given dimension. To apply this
theory in complex situations, one needs to develop it further, but in
the simplest cases Theorem 2.10 emerges as a sufficiently effective
tool. For example, in the one-dimensional case every standard cobordism
is a sum of segments, and hence Theorem 2.10 implies that every smooth
compact one-dimensional manifold is diffeomorphic to the sum of a finite
number of segments and circles. Warning: although this differentiable
classification coincides with the topological classification given in
1.1.17, it has an entirely different meaning. [We add that also in the
noncompact case (which, due to its relative complexity, is not
considered in our book in any systematic way), the differentiableclas­
sification of one-dimensional manifolds is identical with the topologi­
cal one; see 4.1 .]

The present subsection is devoted to the differentiableclas­

sification of the smooth, compact, two-dimensional manifolds, andagain
Theorem 2.10 is enough. We begin by drawing up a list of model manifolds
(which play the same role as the segments and circles in the one-dimen-
sional case). Next we show that every smooth, compact, connected, two-
dimensional manifold is diffeomorphic to one of the models. No two of
the model manifolds are diffeomorphic or homeomorphic. However, we
prove this fact only in Chapter 5 (see 5.3.4.3); here we merely prepare
the geometric part of the proof, decribing for the closed model manifolds
canonical rigged cellular decomposition, and for the nonclosed model
manifolds - the bouquets of circles to which they are homotopy
equivalent. As we did in the classification of one-dimensional manifolds,
we shall not worry about the differentiability class (cf. 1.1).

Model Surfaces

2. We begin with the elementary model surfaces: the spheres

with holes and the Mobius strips.
— 2
A sphere with 1 holes is S with the interiors of 1
pairwise disjoint sperical caps (segments) removed; its boundary is a
sum of 1 circles and inherits a well-defined orientation. A sphere
2
with one hole is diffeomorphic to D , while a sphere with two holes
 256

1 1
is diffeomorphic to the cylinder S x D ; we call the latter a handle
(do not confuse with the "handles" in Subsection 2 !).
3
A Möbius strip is a submanifold of IR produced by the motion
1
of a segment of length 1 whose middle glides along a circle S m
such a manner that the segment remains normal to the circle and turns
uniformly through a total angle (see Fig. 6). Every Möbius strip
tt
3
is a nonorientable compact submanifold of IR with boundary diffeo-
1
morphi c to S .

The list of all model surfaces is: the empty two-dimensional

manifold, the spheres with handles and holes, and the spheres with
crosscaps.
A sphere with g handles and 1 holes (g ^ 0, 1^0) is
defined as the result of glueing a sphere with 2g+l holes and a sum
of g handles by an orientation preserving diffeomorphism from the
boundary of the sum of handles onto the union of 2g components of the
sphere with holes. This object is a smooth, compact, orientable, two-
dimensional manifold, uniquely defined up to an orientation preserving
diffeomorphism by the numbers g and 1, and whose boundary is diffeo-
2
morphic to a sum of 1 circles. This manifold coincides with S if
2
g = 0, 1 = 0; it is diffeomorphic to D if g = 0, 1 = 1, and to
1 1
S x S if g = 1, 1 = 0, and then it is known as a torus; for g = 2
and 1=0, it is called a double torus or a pretzel. For any g and
1, a sphere with g handles and 1 holes can be differentiably
3
embedded in IR ; for g = 3 and 1=2, the standard embedding is
depicted in Fig. 7.
A sphere with h crosscaps and 1 holes (h ^ 0, 1^0)
is defined as the result of glueing a sphere with h+1 holes and a sum
of h Mobius strips by a diffeomorphism of the boundary of the sum of
strips onto the union of h components of the sphere with holes. One
obtains a smooth, compact, two-dimensional manifold, uniquely defined
up to a diffeomorphism by the numbers h and 1, and whose boundary
is diffeomorphic to a sum of 1 circles. This manifold is orientable
 257

only for h = 0. For h = 1, 1 = 0, it is diffeomorphic to IRP^, and

for h = 1, 1 = 1, to a Möbius strip. For h = 2, 1 = 0, this
manifold is called the Klein bottle and is well known because of its
3
immersion in JR , depicted in Fig. 8; for h = 2, 1 = 1, it is called
a disc with an inverted handle and is depicted in Fig. 9, on the left
(the drawing on the right in Fig. 9 demonstrates that by glueing a disc
with an inverted handle and a usual disc by a diffeomorphism of their
boundaries we actually get a Klein bottle).

Fig. 9

We emphasize that in our list of model manifolds, the

crosscaps do not meet with the handles, i.e., we excluded the case of
a sphere with holes to which we attach both handles and Möbius strips.
We show in 3 that for h £ 1, every sphere with g handles, h cros-
scaps, and 1 holes is diffeomorphic to a sphere with 2g+h crosscaps
and 1 holes.
 258

Auxiliary Propositions

3 (LEMMA). Let and X2 be smooth, compact,

two-dimensional manifolds, and let cp be a diffeomorphism of a
component of BX^ onto a component of ^X^- Denote by X the manifold
obtained by glueing X^ and X^ by cp. Then :

(i) ijf X^ and X^ are diffeomorphic to a sphere with g^

handles and 1^ holes, and to a sphere with g^ handles and 1^
holes, respectively, then X is diffeomorphic to a sphere with ^^^2
handles and holes •

(ii) if X^ and X^ are diffeomorphic to a sphere with h^

crosscaps and 1^ holes, and to a sphere with h2 crosscaps and 1^
holes, respectively, then X is diffeomorphic to a sphere with ^l '\+^ri2
crosscaps and 1^+12“2 holes ;

(iii) _if X^ and X2 are diffeomorphic to a sphere with

g^ handles and 1^ holes, and to a sphere with h 2> 0 crosscaps and
12 holes, respectively, then X is diffeomorphic to a sphere with
2g^+h2 crosscaps and 1^+12~2 holes,

The only case which needs proof is (iii), where we see

immediately that X is diffeomorphic to a sphere with g^ handles,
h2 crosscaps, and l^+l2*-2 holes.
To begin with, suppose that = 1, h 2 = 1, and 1 +12~2 =
= 1. Then X is diffeomorphic to a Möbius strip with a handle attached;
see the left drawing in Fig. 10. The last manifold is obviously diffeo­
morphic to a Möbius strip with an inverted handle, as that in the right
drawing in Fig. 10, and hence can be cut into a sphere with one cross­
cap and a disc with an inverted handle. According to (ii), this implies
that X is diffeomorphic to a sphere with three crosscaps and one hole.

F ig . 10
 259

In the general case, we use induction on q„. If q . = 0,

there is nothing to prove. If g^ > 0, then X can be glued from a
sphere with g^-1 handles,crosscaps, and l^+l^-l holes and a
sphere with one handle, one crosscap, and one hole. But we already
proved that this second manifold is diffeomorphic to a sphere with
three crosscaps and one hole. Consequently, X is diffeomorphic to a
sphere with g^-1 handles, h2+2 crosscaps, and 1^+12~2 holes, and
so diffeomorphic to a sphere with 2g^+h2 crosscaps and 1 +12«2
holes.

4 (LEMMA). Let X be a smooth, compact, connected,

two-dimensional manifold, and let X' be the result of attaching a
handle to X by a diffeomorphism of the boundary of the handle onto
the union of two components of 9x. Then:

(i) if X is diffeomorphic to a sphere with g handles and

1 holes, then X1 is diffeomorphic to a sphere with g+1 handles and
1-2 holes, or to a sphere with 2g+2 crosscaps and 1-2 holes;

(ii) jlf X is diffeomorphic to a sphere with h crosscaps

and 1 holes, then X' is diffeomorphic to a sphere with h+2 cross­
caps and 1-2 holes.

This is a corollary of Lemma 3: indeed, in both cases one can

cut X1 into two manifolds, such that the first one differs from X
by having one hole less, while the second is diffeomorphic to a sphere
with one handle and one hole, or to a disc with an inverted handle.

5. Let X and X2 be smooth, compact, two-dimensional

manifolds, and let cp be a diffeomorphism of a nonempty union of whole
components of 3X2 onto a nonempty union of whole components of 3X^.
Denote by X the manifold glued from X^ and X2 by means of cp. If
both X^ andX2 are diffeomorphic to model surfaces, then X is also
diffeomorphic to one of the model surfaces.

This is a consequence of Lemmas 3 and 4, because glueing by

means of cp is equivalent to first glueing by the diffeomorphism of
one of the components of 9X2 onto the corresponding component of 8X^,
obtained by compressing cp, and subsequently attaching a number of
handles equal to half the number of the components of 9X^ and
which remain to be identified.
 260

The Main Theorem

6. Every smooth, connected, compact, two-dimensional manif

is diffeomorphic to one of the model surfaces.

Applying Theorems 2.10 and 5, all we need to show is that the

components of the elementary two-dimensional cobordisms are diffeomorphic
to model surfaces. And this is not hard to check directly by examining
all possible cases, if we recall that every smooth, closed,
one-dimensional manifold is diffeomorphic to a sum of circles. To spell
it out,every elementary cobordism of index 0 constructed from a sum of
m circles is diffeomorphic to a sum of m spheres with two holes and
a sphere with one hole. Next, every elementary cobordism of index 2
constructed from a sum of m circles (and a differentiable embedding
of a circle in this sum) is diffeomorphic to a sum of m-1 spheres with
two holes and a sphere with one hole. And finally, every elementary
cobordism of index 1constructed from a sum of m circles and a dif-
0 1
ferentiable embedding of S * D in this sum is diffeomorphic to one
of the following three manifolds: a sum of m-2 spheres with two holes
and a sphere with three holes; a sum of m-1 spheres with two holes and
a sphere with three holes; a sum of m-1 spheres with two holes and a
sphere with one crosscap and two holes (for m = 2, one see the three
cases in Fig * 11).
 261

7 (INFORMATION). Every compact, connected, two-dimensional

topological manifold is homeomorphic to one of the model surfaces.

Cellular Decompositions
of the Closed Model Surfaces

8. The closed model surfaces possess standard rigged cellular

decompositions, which generalize the canonical decompositions of the
2 2 1 1
sphere S , the complex projective space (CP , and the torus S xS
into two cells, three cells, and four cells, respectively. Each of
these standard decompositions, except the no-cell decomposition of the
empty model surface, contains only one 0-cell and only one 2-cell, while
the number of 1-cells is 2g for a sphere with g handles, and h
for a sphere with h crosscaps. Therefore, the 1-skeleton of a sphere
with g handles is a bouquet of 2g circles, while the 1-skeleton of
a sphere with h crosscaps is a bouquet of h circles. Moreover, the
description of the entire rigged cellular decomposition reduces to the
characterization of the attaching map for the 2-cell, i.e., of a certain
map of S^ into the aforementioned bouquet.

We disregard the values g = 0,1 and h = 0,1, already

considered, and for the case of a sphere with g handles, we represent
S^ as the contour of aregular polygon with first vertex ort^ and
4g edges, arranged successively as

1
In the case of a sphere with h crosscaps, we represent S
as the contour of a regular polygon with first vertex ort^ and 2h
edges, which are arranged successively as

C1 ' C1 ' •••' ch' ch •

1
In both cases we form a quotient space of S by identifying
each edge with the correspondig "primed" edge, as follows: cu is
identified with a|, and b^ with b through a reflection with
respect to a line (relative to which these edges are symmetric), while
c. is identified with c| through a rotation of the polygon (around
its center). In either of cases the quotient space is a bouquet of
circles: in the first case the number of circles is 2g, and in the
1
second h. The projection of S onto this quotient space is the
required attaching map.
 262

Of course, we still have to convince the reader that the

cellular spaces produced in this manner are homeomorphic to the model
surfaces. To this end, let us divide our 4g-gon into a g-gon and g
pentagons, by drawing diagonals which cut out quadruplets a^f b^,a^,b^.
Similarly, we divide our 2h-gon into a h-gon and h triangles, by
drawing diagonals which cut out pairs ci'ci (see Fig. 12).

Identifying the edges and the way described above,

all the vertices of the remaining g-gon become one and the same point,
thus transforming the g-gon into a sphere with g circular apertures;
at the same time, the edges of the pentagons are identified in such a
manner that each pentagon becomes a torus with a circular aperture.
Attaching these holed tori to the holed sphere in such a way as to
restore all that was destroyed by the auxiliary (diagonal) cuts, we
obtain, up to a homeomorphism, a sphere with g handles. Similarly,
the prescribed identifications of the edges ci'ci take all the
vertices of the h-gon into one and the same point, thus transforming
the h-gon into a sphere with h circular apertures; at the same time,
each triangle becomes a Mobius strip, since two of its edges are
identified. Attaching these strips to the holed sphere, we obtain, up
to a homeomorphism,a sphere with h crosscaps. [Warning: the boundaries
of the previous circular apertures (in the sphere) have a common point,
and for g = 2 or h = 2, they even coincide.]
 263

The Hoinotopy Structure

of the Nonclosed Model Surfaces

9. A sphere with g handles and 1 holes is homotopy

equivalent to a bouquet of 2g+l-1 circles. A sphere with h cross­
caps and 1 holes is homotopy equivalent to a bouquet of h+1-1
circles.

To prove these assertions, we first note that by attaching the

4g-gon to the bouquet of 2g circles as in 8, we produce, up to a
homeomorphism, a sphere with g handles and 1 holes, provided we
first remove the interiors of 1 pairwise disjoint discs from the
interior of the 4g-gon. Now let us arrange these discs such that every
line passing through the first vertex, ort^, intersects no more than
one of them. Denote by A the set consisting of:
(i) the contour of the 4g-gon;
(ii) the 21-2 segments tangent to 1-1 of the removed
discs and passing through ort^;
(iii) the outer arcs of the boundaries of these discs having
as endpoints the tangency points.
Then A is obviously a strong deformation retract of the
holed 4g-gon. If we now project this 4g-gon onto the sphere with g
handles, the above strong deformation retraction is transformed into a
strong deformation retraction of the holed sphere with g handles onto
the image of A under the projection. Finally, note that this image
is manifestly homeomorphic to a bouquet of 2g+l-1 circles.
The proof for a sphere with h crosscaps and 1 holes is a
verbatim repetition of the previous argument, with the 4g-gon replaced
by the 2h-gon.

4. Exercises

1. Show that every smooth, connected, noncompact, one­

dimensional manifold is homeomorphic to a line or a half-line (see 3.1).
2
2. Define a submanifold of CEP in homogeneous coordinates
by the equation z™ + z™ + z™ = °, and show that it is diffeomorphic

to a sphere with (m-1)(m-2)/2 handles (see 2.4.4 and 2..4.5).

1 1
3. Show that the subset of (CP x (CP consisting of the
 264

points ((z :z2),(w1:w2)) such that + w 2* = Z2 ^ ~ W 2* iS 3

manifold diffeomorphic to a sphere with (p-1)(q-1) handles.

4. Show that every smooth, closed, connected, orientable,

three-dimensional manifold can be obtained by glueing two copies of a
handle-body by a diffeomorphism of its boundary. (A handle-body is a
3
part of 3R bounded by a sphere with handles which is standardly
3
embedded in IR ) .

5.
Consider the manifold obtained by glueing two copies of
1 2 1 1
the solid torus S x d by a diffeomorphism of its boundary S x S ,
given by the formula (z^z^) (z^ Z^/Z^ z^) , where a,b,c,d are

integers satisfying ad - be = ±1. Show that this manifold is diffeo-

3 2 1 3
morphic to S for a = 0, to S * S for a = ±1, and to 1RP
for a = ±2.
oo
6. Show that on every connected closed C -manifold there is
a Morse function with a unique local minimum and a unique local maximum.

7. Show that on every connected cobordism X with nonempty

9qX and d ^X there is a proper Morse function with no local maxima
and minima lying in int X .

8. Show that on every cobordism there is a proper Morse

function such that, for any of its critical points x^ and x^, of
indices and k2, k^ < k2 implies f (x^) < f(x2>.

9. Suppose that on a cobordism X there is a Morse function

with no critical points of index 1 and that the manifold 9QX is
orientable. Show that the cobordism X is orientable and that every
orientation of 9^X is induced by some orientation of X.
 Chapter 4. Bundles

§1. BUNDLES WITHOUT GROUP STRUCTURE

1. General Definitions

1. A bundle is a triple (T,p,B), where Tand B are

topological spaces and p: T B is a continuous map.The spaces T
and B are called the total space and the base of thebundle (T,p,B),
respectively, and the map p is its projection. For a bundle £, we
denote its total space, its base, and its projection by tl £, bs£,
and pr £ , respectively: £ = (tl £ ,pr £ ,bs £) .
-i
The preimage pr £ (b)of a point b € bs £ is called the
fiber of the bundle £ over the point b.

A section of the bundle £ is a continuous map s: bs £ tl£

such thatpr ^ ° s = idbs^. Two sections of £ arehomotopic if they
can be connected by a homotopy consisting only of sections, i.e., bya
homotopy h: bs ^ x I + tl £ such that pr £ ° h equals
pr^ : bs C x I bs £ .
The restriction of the bundle £ to a subspace B c bs ^ is
the bundle (pr £ ^ (B) ,ab pr ^,B) , denoted by £ ID •
I^
The product of the bundles ^ and is the bundle
(tl x tl C2 ,pr £ x pr i 2 ' h s ^1 * bs ^2* ' denoted bY £<| x C2 * The
fiber of Ky x £2 over a point (b.j,b2) 6 bs(?1 x ?2) is precisely
the product of the fibers pr (b^ ) and pr lb2) .

2. A map of the bundle £ ’ into the bundle £ is a pair of

continuous maps F: tl £ '-»■ tl £, f: bs £ ' -> bs £ , such that the
diagram
 266

tl E ' tl K

Pr K ' (1 )
bs E bs E,

is commutative. If <f) = (F,f) is such a pair, we write c(>:£ 1 +

F = tl (j), f=bS(|).
A map £1 £ is said to be an isomorphism if tl (p and
bs (pare homeomorphisms, and an equivalence if, in addition, bs E, ’ =
= bs ^ and bs <p = idbs C« If there is an isomorphism (equivalence)
between £ 1 and E,, then the bundles and £ are said to be
isomorphic (respectively, equivalent).

A map (p: £ 1 -* £ is called an inclusion if tl 4> and bs i>

are inclusions. For example, given any subset B of bs f, , the
inclusions in: pr E, ^ (B) ->tl £ and in: B -+ bs E, form the inclusion
of the bundle £ I in £.
IB
3. The commutativity of the diagram (1) implies that F is
a fiber preserving map (or a fibered map), i.e., it takes each fiber of
into a fiber of £. Obviously, if pr E, * (tl £ ') = bs £ 1, then given
an arbitrary fiber preserving map F: tl E,1 tl £ there is a unique
map f: bs£' bs E, which makes diagram (1) commutative. Moreover,
if the map pr £' is factorial, then the continuity of F implies the
continuity of f. Therefore, if £1 is a bundle with factorial
projection, then given any fiber preserving map F : tl £ ' -* tl E, there
is one and only one continuous map <J>: E,1 -* £ such that tl c}> = F.

4. Let f be a continuous map of a topological space B

into the base of a bundle £. We may define a new bundle having base
B, total space {(b,x) £ B x tl £ | f (b) = pr E(x)} , and projection -
the restriction of the projection pr^: B x tl £ B to the last space.
This new bundle is called the bundle induced from E, by f, and is
i —
denoted by f *E,.
It is clear that the restriction of the projection
pr9 : B x t U + tl £ to tl(f’£) defines for each b £ bs(f’£) a
^ i
homeomorphism of the fiber of f *£ over the point b onto the fiber
of E, over the point f (b) £ bs EJ , and determines, together with f,
i
a map f *£ £. This map is called the adjoint of f and is denoted by
ad f .
The following observations also need no proofs or explanations.
i
If f is a homeomorphism, then the adjoint map ad f : f*£ £ is an
isomorphism; if, in addition, f = id bs 5/ then adf is an equivalence.
 267

If f is an inclusion, then ad f : f *£ 5 establishes an equivalence

between f*£ and ^ |g • Finally, given arbitrary continuous maps
f: B -> bs £ and g: B' + B, the bundles (f o q ) 1 Z and g ! (f!?) are
canonically equivalent.

5- If <f>: K 1 K is a map of bundles, then the formula

x (pr £ ' (x) ,tl<J> (x) ) defines a continuous map tl £ 1 tl ((bs cf>) 1£) .
This map defines, together with id bs C ', a map of £' into the bundle
i
(bs (j>) £ ,which we denote by corr cj) ; we say that corr <j> corrects
the map <J>. Obviously, ad(bs({)) o corr <(>=<(>.

2. Locally Trivial Bundles

1. The obvious example of a bundle having a given base B

and fibers homeomorphic to a given space F is the standardtrivial
bundle (or the product bundle) (B x F,pr^,B). Its fibers are the
fibers b x f of the product B x f, and are obviously canonically
homeomorphic to F.
Notice that the sections B B x f of the standard trivial
bundle (B x F ,pr^,B) are in a one-to-one correspondence with the
continuous functions B F: for each function f : B ■+ F there is the
corresponding section s: B -> B x f , s(b) = (b,f(b)); we say that f
and s are associated.

2. A bundle £ is trivial or, more specifically,

topologically trivial, if it is equivalent to a standard trivial bundle.
Any equivalence between a standard trivial bundle and F, is referred
to as a trivialization of £.
A bundle £ is locally trivial or, more specifically,
topologically locally trivial, if every point of bs £ has a neighbor­
hood U such that the bundle is trivial.
Since the projection of a product of topological spaces onto
one of its factors is an open map, the projection of a trivial bundle
is open, and hence so is the projection of a locally trivial bundle.
It is immediate that the product of two trivial (locally
trivial) bundles is a trivial (respectively, locally trivial) bundle.
Furthermore, any bundle induced from a trivial (locally trivial) bundle
is trivial (respectively, locally trivial). If f : B bs£ is
i
constant, then f ‘£ is a trivial bundle, for any £.

3. The fibers of a trivial bundle are, as those of the

standard trivial bundle, homeomorphic to each other. However, in a
 268

trivial, but not standard trivial bundle, these homeomorphisms are not
canonical any longer. If the base of a locally trivial bundle is
connected, then its fibers are also mutually homeomorphic; indeed, the
set of the points of the base having fibers homeomorphic to a given
fiber is open, and the sets of this type form a partition of the base
(see 1.3.3.5). On the other hand, the example of the locally trivial
bundle ( (B x F)J |_(B' x F'),pr^J |_pr -j ^BJ |
_B ') * where B, F, B ' , and
F* are arbitrary topological spaces, demonstrates that in a locally
trivial bundle the fibers over points situated in different components
of the base are not necessarily homeomorphic. Moreover, we see that
there are locally trivial bundles which are not trivial.
A nontrivial, locally trivial bundle may have a connected
base; see 5 and 6.

Coverings

4. A locally trivial bundle is a covering in the broad sense

if all its fibers are discrete spaces. In this case the total space
and the projection are usually called a covering space and a covering
projection, respectively.* Clearly , every point of a covering space has
a neighborhood such that the restriction of the projection to this
neighborhood is a homeomorphism onto its image in the base.
A covering in the broad sense is said to be a covering in the
narrow sense or, simply, a covering, if both the covering and base
spaces are connected and nonempty. According to 3, all the fibers of a
covering have the same cardinality, called the number of sheets (or the
multiplicity) of the given covering.

5. A covering whose number of sheets is greater than one

cannot be trivial.

Indeed, the total space of a trivial bundle is homeomorphic

to the product of the base and a fiber, and hence it cannot beconnected
when the fiber is discrete and has more than one point.

6 (EXAMPLES). The bundle (s\hel ,S1), where hel : S^ -> S^

m m m
is defined as hel^lz) = z, is an m-sheeted covering for any m ± 0.

The bundle OR1 ,hel,S1) , where hel: TO ->■ S1 is defined as

27Tix
hel(x) = e , is a countably-sheeted covering.
If k / 0,n, then the bundle having total space G (n,k),
* Translator's note: Frequently, the terms covering space and covering
projection are themselves used to designate the whole covering .
 269

base G(n,k), and projection equal to the submersion exhibited in

3.2.2.3, is a two-sheeted covering. In particular, so is (Sn ,pr, 3RPn )
for n ^ 1.
Finally, let us show that every sphere with h crosscaps
admits as a two-sheeted covering space a sphere with h-1 handles (see Subsection
3.5.3); here h is an arbitrary positive integer. For h = 1 , we
2 2
already encountered such a covering, namely (S ,pr, IRP ). Generally,
2 2
one may construct it starting with h copies of (S ,pr, IRP ). To do
2 2
this, restrict one copy of (S ,pr, IRP ) to a covering over the
projective plane with h-1 holes, and restrict the remaining h-1
copies to coverings over the projective plane with one hole (i.e., over
the Mobius strip). Now glue the bases of these h restricted coverings
into a sphere with h crosscaps by diffeomorphisms of the boundaries of
the holes. The resulting glued space is thebase of a new covering,
whose total space is obtained by glueing the h total spaces of the
above restricted coverings: one of these total spaces is a sphere with
h-1 pairs of antipodal holes, while the remaining h-1 total spaces
are spheres with two antipodal holes, i.e., cylinders over circles.
(Each of these h-1 cylinders has two possible covering attaching maps,
and we mayuse either one of them.) Since "sealing" a pair of holes by
a cylinder results in replacing this pair by a handle, what we actually
obtain is a two-sheeted covering having a sphere with h crosscaps as
the base and a sphere with h-1 handles as the total space.

3. Serre Bundles

1. A bundle £ is a Serre bundle* if it satisfies Serre1s

condition: for any positive integer r or r = 0, and every
continuous maps f: ir bs £ and f -: ir ^ tl £ , related by
-■ Pr
Pr K ° f0 = f there is a continuous map f : I tl £ such that
ir
2T_ 1
pr £ ° f = f anc^ „ = f„. (We identify the cube I with that
■j-r-1 0

face of Ir whose points have the last coordinate equal to zero; see
1.2.5.7.)
The requirement that pr E, ° f = f appearing in Serre's
condition is fundamental in the theory of bundles, and is encountered
also when f and f are defined on spaces other than cubes. If two
maps, f~: X tl K and f: X -> bs £ , satisfy this last requirement,

* Translator's note: Frequently called a Serre fiber space or a weak

fibration.
 270

we say that f~ covers f (or that f~ is a lifting or lift of f;

we also say that f can be lifted to tl £) ; this terminology is valid
for arbitrary £ and X.
Obviously, the p rod uc t of two Serre b u n d l e s and a b u n d l e
induced from a Serre bu nd le are a g ai n Serre bundles.

2. Examples of bundles which do not satisfy Serre's condit

are (I,p,I), where p(x) = x/2 or p(x) = 4x(1-x). In the first case,
take r = 1, f = id I , f ~ (0) = 0; in the second case, take r = 2,
f(x1,x2) = 4x1 (1-x1) (1—x 2) , f~ = id I . Then there is no continuous map
such that p o f = f and f r-1 f0'

We remark that the first bundle has both empty and nonempty
fibers, while the second has a single connected fiber, the others being
not connected. As we shall see later (see 5.4.3.6), such features of
a bundle are not compatible with Serre 1s condition vfrien the base is connected.

Serre's Condition is Local

3. If every point of the base bs £ of a bundle £ has a

neighborhood Usuch that £ |u is a Serre bundle, then £ itself is
a Serre bundle.

PROOF. Let f : in bs £ and f ^ : in ^ tl £ be continuous

maps satisfying pr £ o f~ = f | . Since bs £ can be covered by open

sets such that the restriction of £ to each of these sets satisfies

Serre's condition, Theorem 1.1.7.16 yields a positive integer N such
that every cube of edge 1/N contained in ln is taken by f into
one of these open sets. Divide ln into Nn cubes of edge 1/N,
arrange these cubes in dictionary order Q ,...,Q n , and set
n — 1 i ^
W± = I U (Uj_<| Qj)and fl . It is clear that

In ^ = WQ c : W^ c= ... c W n = In , and examining all possible cases, we

N
see that each pair (Qi ,Q^) is homeomorphic to (in ,in_1). Now assume

that for some i £ Nn there is a map f~_1 : W±_ -+ tl F, such that

pr £ ° f . -i = f I,, and f . „
1-1 Iwi - 1 1-1 ^ - 1 = f 0' Set gi = f |Qi and

g0i = fi-1 ,. Since the restriction £ satisfies Serre's

i
condition, there is a continuous map g~: tl £ such that
 271

Pr £ o g - g and g - g . Furthermore, because f7 . and g~

x x i ui 1-1
agree on w i_-] n Qi = Q|/ together they form a continuous map
/V
u Q. tH , and obviously pr £ o f~ = f and
~ i
f l. n_-| = fq • This shows that induction on i works, starting with
I
i = 0, and the result is a map f~ = f~ such that pr £ o f~ = f and
f = f'
' N
In-1 r0 ‘

Serre1s Condition and Local Triviality

4. Every locally trivial bundle is a Serre bundle.

By Theorem 3, one need only consider the case (B x F,pr^,B),

where B and F are arbitrary topological spaces. Let f: ln B
rn-1
and fQ : I -> B x f be continuous maps such that pr^ o = f
in-i "
Define f : in B x f as

f (x1,...,xn ) = (f (x1,•••,xn ),pr2 o f0 (X l ,...,xn-1)) .

Clearly, pr^ o f - f and f

In-1 f0

5. The following example demonstrates that there are Serre

2
bundles which are notlocally trivial. Let T be the triangle in JR
with vertices (0,0), (0,1), and (1,0), and let p ^ ,p2 : T -*■ I be
defined by p^ (x ,x2) = x^ , p2 (x^,x2) = x 2 . The bundle (T^^I) is
not locally trivial; indeed, the fibers over the points 0 and 1 are
not homeomorphic. However, (T,p ,1) does satisfy Serre's condition:
n ~ n~ 1 rw
if f: I -*• I and f^: I -*■ T are continuous, and p^ o f^ = f zn-i
then the map f~: in T defined by

f~(x1,...,xn ) = (f(x1,...,xn ),min(1-f(x1,...,xn )/P2 o f 0 (X 1 ^ •' X n - 1 )))

n “1
is continuous, covers f, and equals f^ on I
This example shows also that in a Serre bundle with connected
base there can be nonhomeomorphic fibers. Actually, there are Serre
bundles with connected base and in which some fibers are not even
homotopy equivalent, being instead equivalent in a certain weaker sense
(see 5.4.4.3 and 5.4.3.6) .
 272

The Covering Homotopy Theorem

6. Suppose that £ is a Serre bundle and (X,A) is a

cellular pair. Then for everycontinuous map f~: X -* tl £ , every
homotopy F: X xI -> bs £ of pr C ° f~, and every homotopy
G: A x I tl £ of f ~ L covering F |^ T, there is a homotopy of
A — -------------- A x _L
f~ which covers F and extends G.

PROOF. Assume that X is rigged, and that for some r ^ 0

there is a homotopy : u ske^_^X) x i tl £ of f |AUs^e x

covering FJ (AUske X)xl* If e is an r“ce11 from X ^ A, then

r 1 ^
<p ( x , t ) = F(cha (x) ,t) defines a continuous map $ : D x I -> bs £ ,
e e e
while the formula

f (cha (x)), if t = 0,
e
^0,e(x't)
(cha^ (x) ,t) , if x E Sr ^

r 27” 1
defines a continuous map <J>n : (D x 0) U (S x I) -*tl £ .
u ,e
Obviously, the pairs (Dr x I,(Dr x 0) U (Sr ^ x i) ) and (ir + \ i r )
are homeomorphic , and pr ^ o ^ j
^ = ^e (Drx0)u (S17-^xi)# Consequently,
r
there is a continuous map (J> : D x I -> tl £ covering $ and
^ ~ ^ r—1
extending <\>n . Since <f> (x,t) =F (cha (x),t) for all x ES ,
u ,e e r— i e
the maps <j>~ corresponding to all r-cells from X \ A together with
F~_^ yield a continuous map F~: (A U ske^X) x I + tl £ , and it is

evident that ^p r s£ o F~r = F I|(AUske^X)xi

,_M , _ and r|(AUske^_^X)xi
F ~ | /7VM , _ = r-1
F~ . .

Hence, we may use induction on r, settingF~^ = G, to produce a

sequence (F~: (A U ske^X) x I -> tl £ ^r-_^ homotopies which extend

each other. These homotopies define a homotopy of f~covering F and
extending G.

7. Let X be a cellular space and let f~: X tl £ be

continuous. If £ is a Serre bundle, then every homotopy of pr £ © f~
is covered by ahomotopy of f~.

This is precisely Theorem 6 for the absolute case A = 0,

and hence the proof is immediate. We note t ]
that for X = in , Theorem
7 reduces to Serre1s condition for r = n+1.
 273

The Case of Coverings

8. Suppose that K is a covering in the broad sense, X is

a connected topological space, and f,g: X -> tl £ are continuous maps.
I_f pr£Jo f = pr£o g and f equals g at somepoint, then f = g.

PROOF. Since the set {x E X | f (x) / g(x)} is open and, by

assumption, its complement is not empty, it suffices to show that this
complement is also open. In other words, let us verify that if =
= g(Xg), then x^ has a neighborhood U such that f(x) = g(x) for
all x € U. Let V be a neighborhood of such that
pr £|v : V -> pr E, (V) isa homeomorphism (see 2.4) , and take U to be
anyneighborhood of x^ with f (U) c= v andg(U) c: v. Since
pr £ (f (x) ) = pr£(g(x)) for all x £ X, we have f(x) = g(x) for all
x € U.

9. Suppose that £ is a covering in the broad sense, X is

a connected cellular space with a distinguished Q-cell Xq , and
f ,g: X tl £ are continuous. If the maps pr £ o f and pr £ o g are
XQ-homotopic and f(x^) = g(x^), then f and g are XQ-ho m o t o p i c .

PROOF. By Theorem 6, any x^-homotopy from p r £ © f to

pr K° g is covered by an x^-homotopy from f to some map h. Since
h(xQ) = f(xQ) = g(xQ) and pr ^ ° h = pr g, Proposition 8 yields
h = g.

4. Bundles with Map Spaces as Total Spaces

1. We say that a bundle £ satisfies the strong Serre

condition* if for every topological space X, every continuous map
f~: X -> tl £ , and every homotopy F of pr £ ° f7 there is a homotopy
of f~ which covers F.
If we replace X by a cube of arbitrary dimension, then this
becomes the simple Serre condition; moreover, when X is restricted to
be an arbitrary cellular space, we obtain again a condition equivalent
to the simple Serre condition; see 3.7.
2. Let (X,A) and Y bea Borsuk pairand a topological
space, respectively. If X is Hausdorff and locally compact, then the
bundle (C(X,Y),C(in,id),C(A,Y)) satisfies the strong Serre condition.
* Translator's note: Such a bundle is frequently called a Hurewicz fiber
space or a fibrat i o n .
 274

PROOF. Consider a topological space Z, a continuous map

f~ : Z + C(X,Y), and a homotopy F: Z x I C (A ,Y ) of C(in,id) of .
Since X is Hausdorff and locally compact, the maps g : Z x X ■+ Y
and G: Z x A x i y given by g~(z,x) = (f~ )A (z,x) = [f (z) ] (x) and
G(z,x,t) = [F (z ,t )] (x) are continuous (see 1.2.7.6). It is clear that
G is a homotopy of 9~| z x a ‘ N o w (Z x X,Z x a ) is a Borsuk pair(see
1.3.5.5. and 1.3.5.3), and hence G extends to a homotopy G of g .
Finally, the formula [F (z,t)](x) = G (z,x,t) defines a homotopy
F : Z x i C(X,Y) of f which covers F.

3. In a bundle with connectedbase and satisfying the strong

Serre condition, the fibers are pairwise homotopy equivalent.

Let £
be the given bundle, and let s be a path joining
-1 -1
two given points of bs £ . Set Fq = pr £ (s (0 )) and F^ = pr E, (s (1 ))
Now consider two homotopies : F^ x I -> bs £ and : F^ x i bs £

of the composite maps Fq — — > tl £ --^r ^ •>bs £ and

F^ — ^S— ► tl £ —P.f ■§..-> bs £ , respectively, given by Jg(x,t) = s(t) and

J^(x,t) = s (t). Since £ satisfies the strongSerre condition, Jq

and are covered by two homotopies, Jq : Fq x i tl £ and
J~: F^ x I tl of the maps in: F q -* tl £ and in: F tl £ ,
respectively. Now j q ^f q x c F 1 ' JT<F 1 X 1> C F0 ' andhence there
are well-defined maps fQ : FQ -+ F^ , fQ (x) = J~(x,1), and f 1 : F -+ FQ ,
f^ (x) = J^(x,1). We next show that f^ o f^ is homotopic to id F q ,
and since the construction is symmetric, f^ o f^ will be homotopic to
id F 1 .
The formulas

J~(x,2t), if t £ 1/2,
j(x,t) =
J1 (fQ (x) ,2t-1) , if t * 1/2,
and

H (x,t ,t ) = s ((1- t ) (1 - I1-2tI

define a map j: Fq x I -> tl E, and a homotopy H: (Fq x i) x i bs £

of p r £ o j. Again, using Serre's strong condition, H can be lifted
to a homotopy H~: (FQ x I) x i -*tl £ of j. Since
(1- t ) (1 - |1 — 2 11 ) = 0 f or t = 1 or t = 0,1, we s ee that
H~((FQ x (0 U 1)) x I) u H~((fq x I) x1 ) c F . Therefore, the formula
 275

H ( (x,0) ,3t) , if t * 1/3,

K(x,t) = H~( (x,3t-1) ,1) , if 1/3 £ t * 2/3,
H~( (x,1 ) ,3-3t), if t 5 2/3,

defines a homotopy K: F q x I -* f q . Since K(x,0) = H~((x,0),0) =

= j(x,0) = J~(x,0) = xand K (x,1) = H~((x,1),0) = j (x,1) =
= = ^1 ^ 0 'it follows that K is a homotopy from id F q
t O f l o f Q .

4. Given arbitrary points x^x ,x^x| of a connected

topological space X, thespaces C(I,0,1;X ,x Q ,x 1) and
C (I ,0 ,1 ;X,Xq ,x^j ) have the same homotopy type .

PROOF. C (1,0,1 ;X,Xq,x^ ) and C (1,0 ,1 ;X ,x^ ,x^j ) are the

fibers of the bundle (C (I ,X) ,C (in,id) ,C ( (0 U 1) ,X)) over the points
(xQ ,x^) and (xq ,x'j) of its base C ((0 U 1 ) ,X) = X x X; hence by
theorems 2 and 3, they have the same homotopy type.

The Adjoint Serre Bundle

5. Given an arbitrary bundle £, we let ad £ denote the

bundle with the samebase, total space

{ (x,s) E tl£ x C (I ,bs £) | s (0) = pr£(x) } /

and projection (x,s) -* s(1). We call ad £ the bundle adjoint to £.

Notice that the total spaces tl ad £ and tl £ havethe
same homotopy type: the formulas x b- (x ,u ^) and (x,s) h* x , where
u is the constant path in bs £ with u (0) = pr£(x) , define
X x
homotopy equivalences tl £ + tl ad £ and tl ad £ -* tl £ which are
inverses to one another. Indeed, the composition of the first mapwith
the second one is id tl £, while the composition of thesecond map with
the first one is homotopic to id(tlad^) via the homotopy
((x,s),t) b (x,s ), where s^ is the path in bs £ defined by
st (t ) = S (tT) .
6. The bundle ad £; satisfies the strong Serre condition,
for any bundle E,.

PROOF. Consider a topological space Z, a continuous map

f~: Z -* tl ad £, and a homotopy F: Z x I -> bs ad £ (= bs t ) of
pr ad C o f~. Denote by g^ and g2 the composite maps
 276

Pr i
tl C
Z — — » tl ad K in -> tl Ç x C (I,bs Ç) pr.
C (I,bs Ç) ,

and define a h o m o t o p y g: Z x I -* (I,bs Ç) by

[g2 ( z )]( t (1 +t) ) , if t £ 1/(1+t),

[g (z ,t) ] (T
F ( z , t (1+t) - 1), if T ^ 1/(1+t).

It is readily verified that F (z,t) = (g (z) ,g (z ,t )) defines a homotopy

F~: Z x I tl ad £ of f~ and that F~ covers F.

5. Exercises

1. Show that for any g ^ 1 a sphere with g handles admits

a sphere with 2g-1 handles as a covering space.

2. Show that for any h ^ 1 a sphere with h crosscaps

admits a sphere with 2h-2 crosscaps as a covering space.

3. Show that the spaces C (S^,ort ; IRPn /(1:0:...:0)) and

1 n 0
C (S,ort^;S ,ort^) x s are homeomorphic for any n ^ 1 .

4. Show that the bundle with total space C (1,0;Sn ,ort^) ,

base Sn , and projection s h * s (1) , is locally trivial (see 1.2.9.4).

§2. A DIGRESSION:

TOPOLOGICAL GROUPS AND TRANFORMATION GROUPS

1. Topological Groups

1. A set G is a topological group or a group space if it

is endowed with both a topology and a group structure such that the
group operations, i.e., the maps G x G G, (g,h) g h , and G G,
-1
g h- g , are continuous. Obviously, the continuityof these two maps
is equivalent to the continuity of the single map G x g G,
(g,h) ►> g 1h.

By the definition of the (product) topology on G x g, the

 277

continuity of the map (g,h) h- gh at the point (g0 ,hQ) means that
for every neighborhood W of the point gQh 0 one can find neighborhoods
U and V of gQ and hQ such that UV c w. Similarly, the continuity
-1
of the map (g,h) g h means that for every neighborhood W of the
point g^ h^ there are neighborhoods U and V of g^ and h^ such
that U_1V <= w.
Clearly, every group becomes a topological group if it is
equipped with the discrete topology.

2. The continuity of the group operations implies that the

left and right translations by group elements (i.e., the maps G G
1 _

given by g ag and g h- ga) , and the map g ** g are homeomorphisms

of the space G. In particular, if B c G is open or closed, then so
-1
arethe setsB , aB and Ba, for all a € G.
We note also that if B is open and A is arbitrary, then
AB and BA are open. Indeed, AB = U aB and BA = U Ba.
^ a€A a€A

Subgroups and Quotients

3.Any subset H of the topological group G which is a

subgroup in the algebraic sense inherits both a group structure and a
topology from G, and it is immediate that the group operations in H
are continuous.
A subgroup of a topological group is normal if it is normal
in the algebraic sense. As in ordinary group theory, normal subgroups
are termed also normal divisors or invariant subgroups.
An example of a normal subgroup of a topological group G,
which has no analog in ordinary group theory, is the component of the
identity, i.e., that component of the space G which contains the
identity element, e , of G. This component is obviously a subgroup:
if u and v are paths joining e^ with g and h, respectively,
-1 -1
then the path t u(t) v(t) joins e^ with g h. Since the inner
automorphisms of G are continuous and take eG into itself, this
subgroup is normal. It is also clear that the cosets of this subgroup
in G are exactly the components of the space G, and the corresponding
quotient group coincides, as a set, with compG .

4. Every open subgroup of a topological group ,is also closed.

In fact, the complement of an open subgroup is a union of left

cosets and by 2 each of these cosets is open. Therefore, the complement
is also open.
 278

5. The partitionof a topological group G into the left

cosets of a subgroup H is denoted by zer(G,H), and the corresponding
quotient space is called the space of left cosets of H in G and is
denoted by G/H. We shall not need here right coset spaces.
The basic topological property of the partition zer(G,H)
and of the projection G G/H is their openness. Indeed, the
saturation of a set B relative to zer(G,H) equals BH, which is an
open set whenever B is open (see 2).

6. Given a topological group, the space of left cosets of a

closed subgroup is regular. In particular, a topological group whose
identity element is closed is regular.

PROOF. Since the cosets gH are closed (see 2), G/H

satisfies Axiom T ^ . To show that G/H additionally satisfies Axiom T ^ ,
it suffices to produce, given a coset g^H and a neighborhood U of
g^H which is saturated relative to zer(G,H), a saturated neighborhood
V of g^H such that C IV c U. To see this, suppose that we have such
a neighborhood. Then for every point pr(g^) £ G/H and every
neighborhood pr(U) of pr(g^), there is a neighborhood, pr(V), of
Pr (g0)f such that Cl pr (V) c=pr (U) . The last inclusion follows from
the inclusion pr(V) c= pr(Cl V) together with the fact that pr(Cl V)
is closed (which follows from the fact that Cl V is closed and
saturated; Cl V is saturated because the partition zer(G,H) is open;
see 1.2.3.10).
Now to produce the desired neighborhood V, note that
-1
BG ^0 “ g0 an<^ SO P0:*-nts eQ anc^ gg have neighborhoods W and
-1
WQ such that W W Q c= U. Set V = WQH. If g £ Cl V , then W g ,
being a neighborhood of g, intersects V, i.e., there exist w £ W,
w0 G WQ, and h £ H such that wg = w^h. We have g = w ^w^h, and
thus g = w 1w^h £ W ^WQH c= UH = U. Therefore, Cl V c U.

7. Let H be a normal subgroup of the topological group G.

According to ordinary group theory, the set of cosets G/H is endowed
with a group structure. Let us show that the map G/H * G/H G/H,
'i —
(x,y) x y, is continuous.
First, note that the composition ijj:G x g G/H of the map
G x G G, (g,h) » g 1h, with the projection G -+ G/H is constant on
the elements of the partition zer(G,H) x zer(G/H). Secondly, the map
(x,y) h- x y equals fact ij; o a_1 , where

a-1: G/H x g /H (G x G)/(zer(G,H) x zer(G,H))

is the inverse of the injective factor, a, of the map

 279

pr x pr: G x G -> G/H x G/H,

and
fact i)j : (G x G) / (zer (G ,H) + zer(G,H)) + G/H.

Since the partition zer(G,H) is open, ais a homeomorphism

i _

(see 1.2.3.11), and this implies that the map (x,y) h- x y is

continuous.
We conclude that G/H is a topological group; G/H is called
the quotient of the factor group of the topological group G by H.

Homomo rph isms

8. A map f: G -* G 1, where G and G1 are topological

groups, is a homomorphism if it is an algebraic homomorphism as well as
continuous.
As in ordinary group theory, the kernel Ker f of f is
defined as the preimage of the identity of G'. A homomorphism f is
a monomorphism if it is injective, i.e., if its kernel Kerf is the
identity element of G, and an epimorphism if its image Imf = f(G)
is all of G'.An example of monomorphism (epimorphism) is the
inclusion of a subgroup in a topological group (respectively, the
projection of a topological group onto a factor group).
An invertible homomorphism whose inverse is also a homomor­
phism is called an isomorphism. In other words, an isomorphism of
topological groups is a map which is both an algebraic homomorphismand
a homemorphism.

9. Let f: G + G ' be a homomorphism of topological groups.

Then:
(i) Imf is a subgroup of G 1, and the compression
ab f : G Imf is an epimorphism;
(ii) Kerf is a normal subgroup of G, and the injective
factor of f, fact f : G/Ker f G ', is a monomorphism.

In addition to recognizing that this copies a well-know

statement of ordinary group theory, one has to check that the maps
ab f and factf are continuous, which is trivial.

10. An epimorphism f: G G* is open if and only if its

injective factor, factf : G/Kerf + G 1,is an isomorphism.

The necessity of this condition is obvious. The oppennes of

 280

the projection G G/Kerf shows that the condition is also sufficient.

11. An epimorphism of a compact topological group onto a

topological group with closed identity element is open.

PROOF. Let f: G -** G* be the given epimorphism. Since G

is compact, G/Kerf is compact. Moreover, as the identity element of
G' is closed, G' is Hausdorff (see 6). Finally, factf : G/Ker f G'
is invertible and continuous, and hence in our case a homeomorphism
(see 1.1.7.10). Applying 10, f is open.

Direct Products

12. Let G^ and G^ be topological groups. In ordinary

group theory, the product G^ x G^ is given a group structure, and in
topology it is given a topology (see 1.2.2.1), and it is clear that
these two structures are compatible in the sense of 1 , i.e. , the group
operations in G^ x G^ are continuous. The result is a topological
group G^ x G^ t called the direct product of the topological groups
G^ and G^•

This product operation is both commutative and associative:

there are obvious canonical isomorphisms G^ x G^ G^ x G^ and
(G1 x G2) x G3 + G x (g2 x G3) .

We remark that the inclusions in^: G^ ^ G^ x g ^,

(x^ ,eG ), and in2 ; G^ ^ G^ x g2 , x^ (e ' X2^ ' are monomor-
2 1
phisms (of topological groups) , while the projections pr^ : G^ x g 2 + G
and Pr 2 ; Gi x G 2 G2 are °Pen epimorphisms such that Ker pr^ =
= in2 (G2) and Ker pr2 = in^ (G^) . The last observation together with
Theorem 10 imply that fact pr^ : (G^ x G2)/in2 (G2) G and
fact pr2 : (G^ x G2)/in^ (G^ ) -*• G2 are isomorphisms.

13. We say that the topological group G decomposes into the

direct product of its subgroups G
and G ? if the map G x G + g,
1 2* 1 2
^g 1 'g2^ ** g ig2' is an isomorPh:i-sm of topological groups. If this is
the case, the groups G and G^ x g 2 are usually identified via this
isomorphism.
Recall that a similar definition exists in ordinary group
theory, only there the isomorphism is simply an algebraic isomorphism.
Moreover, in that theory, G decomposes into the direct product of its
subgroups G^ and G2 if and only if G^ and G2 generate G, are
normal subgroups of G, and G 1 n G2 = eG * Consequently, if these
 281

conditions are satisfied in our case, then ^ ^1^2 ^s an

algebraic isomorphism. This map is obviously continuous; however, there
are obvious examples where the algebraic inverse isomorphism is not
continuous. But the algebraic inverse isomorphism is continuous
whenever the space G is compact and Hausdorff. Therefore, every
compact Hausdorff topological group which decomposes algebraically into
the direct product of two subgroups, decomposes also into the direct
product of these subgroups in the sense of our topological definition.

The Simplest Examples

14. The real line IR with addition as the group operation

is a topological group, as is the space 3Rn . Obviously,
IRn = 3R x ... x ir (n factors; the product is understood as in 12).

15. The punctured real line 3R* = IR \ 0, with multiplication

as groupoperation, is a topological group. Its subgroup nR* consisting
ofthe positive reals, is isomorphic to IR: an isomorphism !IR + nR* is
provided by the exponential function x a , with arbitrary a ^ 1.
Another evident subgroup of IR * is S^, and obviously nR* = S^ x nR* .
The punctured complex line (E* = (C ^ 0 and the punctured
quaternionic line 3H* = 1H \ 0 are also topological groups under
1 3
multiplication. Here S is a subgroup of (C*, S is a subgroup of
H* , a n d (E* = s 1 X IR * , E* = s3 X 3R* .

16. The map hel: 3R + S^ (see 1.2.6) is an openepimorphism.

Its kernel is the subgroup of integers, 2Z , of 3R, and hence the
factor group 1R/2Z is isomorphic to S^ .
n 1 n 1 1
The map hel x ... x hel: 3R (S ) = S x . .. x s is also
an open epimorphism. Its kernel is the integer lattice, ZZ x ... x 2 =
= 2Zn , of IRn , and hence the factor group 3Rn /ZZn is isomorphic to
( s V .

17. The subgroup of S^ consisting of the real quaternions

(i.e., of the quaternions (x^ ^ ^ x ^ x ^ ) such that x^ = x^ = x^ = 0)
is simply s°.This is a normal subgroup, and the factor group S'Vs^
3
is, as a topological space, the same as HRP .
3
The sugbroup of S consisting of the complex quaternions
(i.e., of the quaternions (x^,x2 ,x^,x4) such that = 0) is

simply S^. However, this is not a normal subgroup. The coset space
S'Vs^ is canonically homeomorphic to S^:
this canonical homeomorphism
3 2
is provided by the injective factor of the Hopf map S S (obviously,
 282

3 1 3
zer(S ,S ) = zer (S s 2) ) .

2. Groups of Homeomorphisms

1. By 1.1.4. 7, the homeomorphisms of a topological space

are a subgroup of the group SymmX of all invertible transformations
X + X, i.e., they form a group under the ° (composition) operation.
We denote the group of homeomorphisms of X by Top X .
We may define two topologies on Top X . The first one is
induced by the inclusion Top X c C(X,X) (see 1.2.7.1), i.e., is
defined by the prebase consisting of the sets Nb(K,0) =
= C (X,K ;X ,0) fi Top X , with K compact and 0 open. The second
topology is defined by the prebase consisting of the sets U, U \
where U is open in the first topology. Equivalently, the second
topology is generated by the prebase consisting of the sets Nb(K,0),
[Nb(K,0)]~1.

2 (LEMMA) . I_f X is a locally compact Hausdorff space, then

the map Top X x Top X Top X , (g,h) ^ gh (= g o h) , is continuous
in either of the above topologies.

PROOF. If gh £ Nb(K,0), then h(K) c g_ 1 (o), and by

1.1.7.22, every point of h(K) has a neighborhood whose closure is
compact and contained in g ^ (0). Let O' denote the union of a finite
collection of such neighborhoods which cover h(K) . Clearly, Cl 0 1 is
compact, and g E Nb(Cl0',0) , h E Nb(K,0'), Nb (Cl 0 ',0) Nb (K ,0 1) c
e: Nb (K,0) .
Now if gh € [Nb(K,0)]_1, then h_1g_1 £ Nb(K,0), and the
above argument yields two sets, U,V c Top X , open in the first topology,
-1 -1 _1
and such that h E V, g e U, VU c Nb(K,0) . Clearly, g E U ,
h E V”1, and U~1v " 1 c [Nb(K,0)]”1.

3. If X is a locally compact Hausdorff space, then Top X ,

equipped with the second topology, is a topological group.

This is a corollary of 2 and of the obvious continuity of the

-1
map g^ g in the second topology.

4. If X is a compact Hausdorff space, then the first and

the second topologies on Top X are identical.

This is an immediate corollary of the relation

[Nb (K ,0) ]~ 1 = Nb (X \ 0,X v K) .

 283

5 (LEMMA). Let X be a locally compact, locally connected,

Hausdorff space. Then for a prebase of the second topology on Top X
it suffice to take the sets of the form N b (K ,0), where K is the
closure of a connected open set and 0 is open.

PROOF. Given a compact K, an open 0, and a homeomorphism

f £ Nb(K,0), it suffices to produce connected open sets U , ...,U
s I s
with compact closures, such that f £ H^ NbiClU^O) Nb(K,0). For
each point x £ K, fix a connected neighborhood of x, V ,with CIV
compact and Cl Vx <z f ^ (0) (see 1.1.7.2 2 and 1.3.4.3). Now cover K
by a finite number of the V 's, say V ,...,V . It is clear that
x X1 xs
the sets U. = V have the required properties.
i
6. I_f X is a locally compact, locally connected, Hausdorff
space, then the two topologies on Top X are identical.

By Lemma 5, given an open connected U with compact ClU ,

an open 0, and a homeomorphism f £ Nb(ClU,0) , it is enough to
find a subset U c Top X , open in the first topology, and such that
-1
fCU <z Nb (Cl U ,0) . Leaving aside the trivial case U = 0, we fix a
point xq £ f(U) and find a set W with compact closure contained in
0, such that f(Cl U) c W, and then take an open set V satisfying
f(Cl U) c V c Cl V c w. Now set

U = Nb(Xq,U ) n Nb(ClW ^ u , f ~ 1 (0) ^ Cl U) .

-1
The inclusion f £ U is trivial, and all that remains is
-1 -1
to show that U <= Nb(ClU,0) , i.e., that g (Cl U) c: 0 for any
-1
g e U. But g£ U implies g (Cl W ^ U) <= f (O) Cl U , whence
U c g (V) U g (X \ Cl W) .Since g (V) and g (X \ Cl W) are open and
disjoint, and U is connected, we have only two possibilities: either
U n g(U) = 0 or U c g(V). Since g £ U, g ( x Q) £ U; since
x n £ f (U) <= v, g(xn) £ g(V). Consequently, U fl g (V) / 0 , and thus
0 yi -i
U c g (V) , i.e., g (U) c V. Finally, g (Cl U) c CIV c o.

Groups of Diffeomorphisms

7. Let X be a C^r-manifold, 1 £ r £ 00. By 3.1.2.9, the

set of its Cr-diffeomorphisms, DiffrX, is a group under the
composition operation By 3.4.1.1, Diff X can be endowed with the
Cr-topology. We show that these two structures are compatible and
conclude that Diffr x is a topological group.
 284

Obviously, the case r = 00 reduces to r < 00, and so we may

assume from the b e g i n n i n g that r is finite. C o n sid er the m a p p i n g

do ... 0 d: D i f f rX ^ T o p ( Ta ng ... Tang X) .

This is clearly a group monomorphism. Moreover, by 3.4. 1.1,

d o ... o d is a topological embedding when the group
Top (Tang ... Tang X) is e q u ip pe d wit h the se cond topology. However,
the first and the second topo lo gie s on Top(Tang ... Tang X) coincide
(see 6 ) , and hence the operat io ns o and f f ^ are c o n t i n u o u s in
these topologies. Consequently, both o p e r a t i o n s are c o n t i n u o u s in
D if f r x also.
Let us add that the i nc lu si on Diffr x -+ Top X is a m o n o m o r ­
ph is m of to pol ogical groups, and that the same is true for the in cl u s i o n s
D i f f rX + D i f f SX with s < r.

The Classical Groups

10. The analytic manifolds 0(n), SO(n), U(n), SU(n),

Sp (n) , and also GL (n, IR) , GL+ (n, IR) , GL (n ,CE) , GL (n , 3H ) , defined
in Subsection 3.2.1 (see 3.1.1.2 and 3.2.1.5 - 3.2.1.9) and endowed
there with group structures, are obviously topological groups. 0(n)
is called the orthogonal group, SO(n) - the special orthogonal group,
U(n) - the unitary group, SU(n) - the special unitary group, and
Sp(n) - the symplectic group. G L (n, IR) , G L (n,Π) , and G L (n , 3H) are
known as the general linear groups. It is immediate that SO(n) is
the component of the identity of 0(n), while GL+ (n, IR) is the
component of the identity of GL(n, HR).

11. The topological group GL(n, IR) is manifestly a subgroup

of the topological group Top IRn (in the sense of 1.3). In the same
sense, 0 (n) , SO(n), and GL+ (n,lR) are subgroups of TopIRn .
Similarly, GL(n,(E) and its subgroups U(n), SU(n), are
subgroups of Top (En , while GL (n, IH ) and its subgroup Sp(n) are
subgroups of Top HH .
We also note that the inclusions U(n) c= SO(2n),
Sp (n) c SU(2n), GL (n ,Œ ) c GL+ (2n , TR) , and GL(n,K) c=GL(2n,Œ) are
inclusions of a subgroup into a group in the sense of 1.3.
 285

3. Actions

1• An action of the group G on a set X is a map

G x x -y X with the following two properties:
(i) <eG 'x ) X'
*
(ii) if (g^,x) ** x ^ and (g2 ,x^) ►> x ^ , then (g2g^,x) ** x2 •

The image of (g,x) under the given action is usually denoted

by gx, and so one may write conditions (i), (ii) in the form: eçx “ x '
g2 (g1x) = (g2g 1)x.

Every element g £ G defines a map X X, x gx, called

the transformation induced by the element g. We see from (i) and (ii)
that this map is invertible (its inverse is the transformation induced
by g 1) and that the map G -+■ Symm X which takes each g into the
corresponding transformation is a homomorphism. We call it the adjoint
homomorphism of the given action. Actually, this homomorphism uniquely
determines the action, and it is clear that every homomorphism
h: G SymmX is the adjoint homomorphism of a certain action, namely,
of (g,x) *+ (h(g))x. Therefore, an action of a group G on X can be
interpreted as a homomorphism G Symm X .

We are mainly interested in the case where the adjoint

homomorphism is a monomorphism. An action with this property is said
to be effective. Generally, the kernel of the adjoint homomorphism will
be referred to as the noneffectiveness kernel of the given action. If
K is this kernel, then we can write the adjoint homomorphism as the
composition of the projection G G/K with the monomorphism
G/K -> SymmX . Moreover, the action itself may be expressed as the
composition of the map pr x id X :G x X G/K x x with the effective
action G/K x x -> X. We call the action G/K x x X the effective
factor of theaction G xx X.

The image of the set G x xunder a given action G x x X

is a subset of X called the orbit of the point x. Obviously, the
orbits of two points are either identicalor disjoint, and hence the
orbitspartition X. An action with only one orbit is said to be
transitive. In general, we denote the space of orbits by X/G.
If h: G^ G is a group homomorphism, then by composing the
mapping h x id X : G^ x X G x x with an action G x x X of G on
X, we obtain an action of G^ on X. We say that this .new action is
induced by the initial action via the homomorphism h. The adjoint
homomorphism of the induced action is simply the composition of h with
 286

the adjoint homomorphism of the initial action. If h is a monomor­

phism (epimorphism), then an effective (respectively, transitive) action
induces an effective (respectively, a transitive) one.
If is a subgroup of G and h is the inclusion of G^
in G, then we say that the induced action is obtained by restricting
(or by reducing) the group G to G^ ; we also say that the initial
action is obtained by extending (or by prolonging) the groupG^ to G.
The discussion above shows that when one restricts the group, an
effective action remains efective. Also, when one extends the group,
a transitive action remains transitive.
A subset of X is invariant under the action G x X ^ X
if it is saturated with respect to the partition of X into orbits.
If this is the case, then we have an action G x -> X^, and clearly
this is effective whenever the initial action G x x X is effective.
Given two actions, G^ x X^ and G^ x X^ x2 ' their
product is the action (G^ x g ^ ) x (X^ x x^) + X^ x x 2 defined b y
(g^,g2) (x i,x 2 > = (g^x^,g2x 2). A product of effective (transitive)
actions is again effective (respectively, transitive).
Given two actions, G x x
X ', of the same
X and G x x'
group, a map f: X -> X 1 is a G-map (or a G-equivariant map)if
f (gx) = gf (x) for all x £ X and g £ G. We can describe the more
general situation when we are given two actions, G x x -+ X and
G ’ x x 1 + X 1, of different groups, and a homomorphism y: G-+ G '; then
f: X + X ' is called a y-map if f (gx) = y(g)f (x) for all x £ X and
g £ G.
Two actions, G x x -* X and G x x 1 -* X ’, are equivalent if
there is an invertible G-map X X ’.

2. The action defined in 1 should actually be called a

action, to distinguish it from a right action, which is defined as a
map X x g X with the following two properties: (i) (x,e ) ^ x ;
G
(ii) if (x/g 1 ) H- Xl and (x1 ,g2 ) h- x 2, then (x,g1g 2 ) h -x 2>For a
right action, we w r ite xg instead of gx, and p r o p e r t i e s ( i ) , (ii)
can be r e f o r m u l a t e d as: x e Q = x, (xg1 )g2 = x(g g 2 ). F ur th e r m o r e , the
adj oi nt h o m o m o r p h i s m b e c o me s the a d j o i n t a n t i h o m o m o r p h i s m , and the re st
of the d i s c u s s i o n in 1 can be r e p e a t e d wo r d for w o r d for a r i gh t action.
It is c lear that the formula xg = g ^x tr an sf o r m s a left
action into a r ight one, and that the for mu la gx = x g -1 y ie ld s the
inverse transformation. We say that the a c t i ons thus r e l a t e d are
conjugate.

Henceforth, by a ct io n we shall me a n a lef t action, u n l e s s we

m e n ti on e x p l i c i t l y that we are d e a l i n g wit h a r i ght action.
 287

3. If the transformations induced by the elements of the

group G ,acting on X from the left (right), are elements of a sub­
group H of SymmX , then the action (respectively, right action) of
G can be thought of as a homomorphism (respectively, antihomomorphism)
of G into H. In this case we say that G acts on X (respectively,
acts from the right on X) by transformations from H.
The group H is not always indicated explicitly. For example,
if X is a topological space and H =Top X , then one simply says that
G acts on X (acts on X from the right) by homeomorphisms.
Similarly, if X is a group and H is a group of automorphisms of X,
then one says that G acts on X (respectively, acts on X from the
right) by automorphisms; in this case, the action itself will be referred
to as a group-action.

Important special examples are the actions G x G + G given

_ 1
by (g,x) gx or (g,x) ^ gxg ,
and the right actions G x g ■+ G
- 1
given by (x,g) xg and (x,g) ++ g xg (here all the products are
taken in G). These are called,in order: the left canonical action,
the left inner action, the right canonical action, and the right inner
action. The canonical actions are effective and transitive, while the
inner actions are group-actions.
Let us remark that by restricting the left canonical action
G x g G to the action G^ x G G, where G^ is a subgroup of G,
the orbits become the (right) cosets of G^ . Thus, the two interpreta­
tions (the usual group-theoretic one, and that given in 1) of the
notation G/G^ agree (if we denote the right coset space also by G/G^) .
The following generalization of the left canonical action is
already of general importance. Let G^ be a subgroup of G. Since
every left translation takes left cosets into left cosets, the map
(g,x) gx induces a map G x G/G^ G/G^ , and this is clearly an
action, called the canonical action ofthe group G on G/G^ . It is
transitive, and its noneffectiveness kernel is the intersection of all
subgroups of G which are conjugate to G^. The projection G G/G^
is a G-map with respect to the left canonical action of G on G and
the canonical action of G on G/G^ .
It turns out that every transitive action of G is equivalent
to the canonical action on some quotient G/G^. Specifically, let
G x x X be a transitive action, and let x^ G X be an arbitrarily
chosen point. Consider the map f: G X, f ( g) = gx1 . .The preimage
of x1 under f is a subgroup G^ of G, while the preimages under
f of points x G X, x i x 1, are left cosets of G 1- It is routine
 288

to check that the injective factor of f, factf : G/G^ ^ X, is a

G-map.
The subgroup G^ has a special name: it is known as the isotropy
(or stability, or stationary) subgroup of the action G x X X, or of
the group G, at the point x . Obviously, the isotropy subgroup at
-1
the point gx^ is gG^g , and so the isotropy subgroups of a
transitive action of G constitute exactly one of its classes of
conjugate subgroups.

Continuous Actions

5. A continuous action of a topological group G on a

topological space X is a continuous map G x x ->■ X which is an action
in the sense of 1.
For a continuous action, the transformations induced by the
elements of the group are manifestly homeomorphisms. Therefore, the
adjoint homomorphism of a continuous action G x x ->■ X can be compres­
sed to an algebraic isomorphism G -*■ Top X . By 1.2.7.6, this last
homomorphism is continuous in the first topology of the group Top X
(see 2.1). If, in addition, X is Hausdorff and locally compact, then
the existence of a continuous (in the first topology on Top X )
compression G Top X of the adjoint homomorphism of the given action
is equivalent to the continuity of the action. It is clear that an
algebraic homomorphism G Top X which is continuous in one of the
two topologies of Top X is continuous also in the other one, and we
know that if X is locally compact and Hausdorff, then Top X with
the second topology is a topological group (see 2.3). Therefore,
if X is a locally compact Hausdorff space, then a continuousaction of
G on Xmay be defined as a homomorphism G Top Xof topological
groups (see 1.8).

A discrete group which acts by homeomorphisms always acts

continuously. Thus, we may regard the actions of nontopologized groups
which act by homeomorphisms as continuous actions.

6. A G-space is a topological space endowed with a continuous

action of the group G. A G-space is called effective if the action of
G is effective. In the general case, by shifting to the effective
factor of the action of G, the given G-space becomes an effective
(G/K)-space, where K is the noneffectiveness kernel.
When the actionGx x X is continuous, X/G is a
topological space (a quotient space of X ) , known as an orbit space.
 289

Since the saturation of any subset A c: X with respect to the partition

of X into orbits is the union of the sets gA, g € G, this partition
is always open. When the group G is finite, the partition into orbits
is also closed. In particular, X/G is second countable together with
X, and when G is finite, X/G is normal together with X; see
1.2.3.10 and 1.2.3.9. Let us add that the partition into orbits is
again closed whenever X is compact and Hausdorff and G is compact.
Indeed, in this case the action G x X X is a closed map, and since
it transforms every product G x A into the saturation of the set A,
this saturation is closed whenever A is closed.
By restricting the topological group G to a subgroup G^,
we transform every G-space into a G^-space. Any invariant subspace of
a G-space is obviously a G-space; the G-spaces of this type are termed
subspaces of the initial G-space. The product of two continuous actions
is continuous, and hence the product of a G^-space with a G 2 ~space is a
(G 1 x G^J-space.
One can define the notions of G-map and y-map for the case of
continuous actions. To spell it out, a G-map (or a G-equivariant map)
of a G-space into another G-space is any continuous map which is a G-map
in the sense of 1; similarly, a y-map of a G-space into a G T-space is
any continuous map which is also a y-map in the sense of 1 (here
y: G G1 is a homomorphism of topological groups). Two continuous
actions of G are equivalent if the corresponding G-spaces are
G-homeomorphic.
When G is a topological group, the special actions
introduced in 3, i.e., the left canonical and left inner actions, are
continuous.
As in 3, the notation G/G^ can be interpreted in two ways
(as the space of cosets of G^ in G and as an orbit space), but again
the two interpretations agree.

7. The canonical action of a group G on the space G/G^

of cosets of a subgroup G^is also continuous. To see this, consider
the composition ip: G x g G/G^ of the map G x g -+■ G , (g/h) gh,
with the projection G G/G^ . Clearly, \p is constant on the elements
of the p a r t i t i o n zer(G,eG ) x z e r f G ^ ^ . Fu rt he rm ore , the a c tio n
G x G/G>j -+ G/G^ in which we are interested is the composition of the
map G x g /G1 + (G x G )/ (zer (G ,eG ) x z e r f G ^ ) ) , given by the inverse
of the injective factor of id G x pr: G x G G x G/G^,. with
fact \p : (G x G)/(zer(G,e )x z e r f G ^ ^ ) + G/G1 . Since the partitions
zer(G,e ) and zer(GfG j areopen, the above injective factor is a
G •
homeomorphism (see 1.2.3.11), which in turn implies the continuity of
 290

the action of G on G/G^.

Therefore, G/G^ becomes a G-space with respect to the
canonical action G x G/G^ G/G^ , and is called a homogeneous space.
The equivalence between an arbitrary transitive action
G x x X and the canonical action of G on G/G^ , where G^ is an
isotropy subgroup, is not complete in the case of continuous actions.
More precisely, the map f: G X, f(g) = gx1, and its injective
factor, factf : G/G1 X, are indeed continuous, but obvious example^
-r
show that (fact f ) is not n e c e s s a r i l y continuous. However,
(fact f) ^ is continuous provided that G is compact and X is
Hausdorff, i.e., every transitive continuous action of a compact
topological group on a Hausdorff topological space with a distinguished
point is canonically equivalent to the canonical action of the group on
the space of cosets of the isotropy group at the distinguished point.

8. A continuous action G x x X is free if for every point

x E X the map G X given by g *+ gx is an embedding.
Every free action is clearly effective. Moreover, if the
action G x x X is free, then by restricting G to one of its sub­
groups, or by restricting X to one of its G-subspaces, the action
remains free. A product of free actions is free. The canonical action
G x g + G is free, while the canonical action G x g/G^ G/G^ is not
free unless G= e_.
1
Given a free action G x x X, consider the bundle
(X,pr,X/G). If every point x E x U such that
has a neighborhood
gU fl g'U = 0 for all g,g' E G with g ^ g 1 (which happens, in
particular, when G is finite and X is Hausdorff), then the image of
U under the projection p r : X -> X/G is open, and the restriction of
the bundle (X,pr,X/G) to pr(U) is a trivial bundle with discrete
fibers. Therefore, in this case (X,pr,X/G) is a covering in the
broad sense.

9. A continuous right action of a topological group G on a

topological space X is a continuous map X x g X which is also a
right action in the sense of 2.
All definitions and facts discussed in 5, 6, and 8 can be
adapted immediately to the case of right actions. In particular, a
topological space endowed with a right continuous action is called a
right G-space. We keep the simple term G-space only for left G-spaces.

* Translator's note: Think of S as the set of complex numbers of

modulus 1 and let a be irrational. Take G = ]R, X = S* x s* , and G xx X,

2irix 2?riy . 27i'i (x+1) 2Tri(y+at), ,
(t,(e ,e ) ) ••► (e ,e ) (an irrational f l o w 1 on the torus).
 291

Examples

10. Let X be a topological space. The identity homomorphism

Top X Top X defines an effective action of Top X on X, and hence
an effective action of any subgroup of Top X on X. If X is
Hausdorff and locally compact, then Top X is a topological group and
all these actions are continuous. In particular, GL(n, IR) , GL+ (n, HR),
0(n), and SO(n)act effectively and continuously on TRn , while
GL(n,(C), U(n), and SU (n)act in the same manner on CCn , and
GL (n, U) and Sp(n) - on
H n.
>27
Given an arbitrary C' -manifold X with r £ 00, the
r
inclusion Diff X Top X defines an effective and continuous action
of DiffrX on X.

11. Since Sn ^ is invariant under theactionof 0(n) on

IRn , 0(n) and its subgroup SO(n) act continuously on S n \
Similarly, U(n) and SU(n), being subgroups of 0(2n), act
continuously on S2n
\ while Sp(n), being a subgroup of 0(4n),
4n-1
acts continuously onS . All these actions are effective, and, if
0 0 1 1
we exclude the trivial cases SO (1) x s -* S and SU (1 ) x S S ,
transitive.
The isotropy subgroupsof the actions 0(n) x s S
and S0(n) x sn ^ ort Sn ^ at
are exactly 0(n-1) and SO(n-1).
n 2n-1 2n-1
Similarly, the isotropy subgroups of the actions U(n) x s S
and SU (n) x s ^ n ^ -*■ S2n ^ at ort9 are U(n-1) and SU (n-1 ) , while
4n-1 4n-1
the isotropy subgroup of the action Sp(n) x s ■> S at ort.
^ 1
is Sp(n-1). The corresponding homeomorphisms, 0(n)/0(n-1) -*■ S ,
SO (n) /SO (n-1 ) + S n"1, U(n)/U(n-1) -> S2n_1, SU (n) /SU (n-1 ) -*■ S2n_1 ,
4n-1
and Sp(n)/Sp (n-1) + S (see 7) equal the injective factors of the
submersions V(n,n) [= 0(n)] V(n,1) [= SR "*], V(n,n-1) V(n,1),
CEV (n ,n) + CEV (n ,1) , (CV (n,n-1) + CEV(n,1), and lHV(n,n) ->1V( n,1),
defined in Subsection 3.2.1 (see 3.2.1.3, 3.2.1.5, and 3.2.1.6).
If we restrict 0(n), U(n), and Sp(n) (n > 1) to their
subgroups which consists of scalar multiples of the identity matrix,
and which are usually identified with S^, s \ and S3 , respectively,
we obtain continuous actions S^ x s n^ S11 \ S^ x s n + S n ,
and S3 x s 4 n _ 1 S4n_1. These are free actions, and the corresponding
orbit spaces Sn~1/S°, S2n_1/S1, and S4n 1/S3 are simply nRPn ,1,(CPn
CDPn_1 , and H P n‘1 .
We remark also that Dn , D2n, and D4n are invariant under
the actions of 0(n) , U (n) , and Sp(n) on IR , CD , and H . Hence
 292

0(n) and SO(n) act continuously on Dn , U(n) and SU(n) act

continuously on D2n, and Sp(n) acts continuously on D n. All these
are effective actions.
12. The groups 0(n) and SO(n) (0(k) and SO(k)) act
continuously from the left (right) on the Stiefel manifolds V(n,k):
the left actions are defined by (g,v) ^ g o v [g £ 0(n) or SO(n),
v £ V (n ,k ); g andv are regarded as linearmaps]; the right actions
are given by (v,g) v o g . Similarly, U(n) and SU(n) (U(k) and
SU(k)) act continuously from the left (respectively, from the right)
on (CV (n ,k ) , and Sp(n) (Sp(k)) acts continuously from the left
(respectively, from the right) on HV(n,k).
For k ^ 0, all the left actions are effective, and the only
intransitive ones are SO(n) x V(n,n) V(n,n) and SU(n) x (CV(n,n)
(CV(n,n), n > 1. The isotropy subgroups of 0(n) and SO(n) at the
point [ (x^,...,x^) b (0,...,0,x^,...,x^)] £ V(n,k) (the elements of
k n
V(n,k) are considered as linear isometric maps IR ■+IR ) coincide
with O(n-k) and SO(n-k). Similarly, the isotropy subgroups of U(n),
SU(n), and Sp(n) at the points [(x^,...,x^) (0, ... ,0 ,x^ ,.. .,x^)]
of (CV(n,k) and lV(n,k) coincide with U(n-k), SU(n-k), and
Sp(n-k), respectively. The corresponding homeomorphisms

0(n)/0(n-k) -+ V (n ,k) , SO(n)/SO(n-k) ->V(n,k) ,

U(n)/U(n-k) (CV (n ,k ) , SU (n)/SU (n-k) ->CCV(n,k),

and
Sp (n)/Sp (n-k) 3HV (n ,k ) ,

are precisely the injective factors of the maps 0(n) -* V(n,k),

SO (n) -> V (n ,k) , U (n) ->• (CV (n ,k ) , SU(n) + (CV(n,k), and Sp(n) + HV(n,k),
defined in Subsection 3.2.1 (see 3.2. 1.3, 3.2.1.5, and 3.2.1.6). When
k = 1, these actions reduce to those discussed in 11.
All the right actions are free. The corresponding orbit
spaces, V (n,k )/0 (k) , V (n,k )/SO (k) , (CV (n ,k)/U (k) , (CV (n ,k )/SU (k ) ,
and H V (n ,k )/S p (k ), are canonically homeomorphic to the Grassman
manifolds G(n,k), G+ (n,k), CEG(n,k), and ]HG(n,k), respectively;
the corresponding canonical homeomorphisms are the injective factors
of the maps V(n,k) + G(n,k), V(n,k) + G+ (n,k), (CV(n,k) + (CG (n,k),
and 3HV (n ,k) ^]HG(n,k), defined in Subsection 3.2.2 (see 3.2.2.3,
3.2.2.7, and 3.2.2. 8) .

13. The same formulas, i.e., (g,v) g o v, (g,v) v ° g,

define left actions of GL(n,3R) and GL+ (n, IR) on V'(n,k), of
GL(n,(C) on (CV1 (n,k) , and of GL(n, 1H) on 3HV 1 (n,k) , and right
 293

actions of GL(k, IR) and GL+ (k,IR) on V ’(n,k), of GL(k,(E) on

(CV* (n,k) , and of GL(k, IH) on 3HV' (n,k) .
All the left actions are effective and, excepting the action
GL+ (n, 3R) x V 1 (n ,n ) V 1 (n,n) , transitive. The isotropy subgroups of
GL (n , 3R) , GL (n ,(E) , and GL (n , IH) at the points [ (x , ...,x ) *+
I .K
-► (0, . ..,0 ,x1 ,... ,xk )] of v ’(n,k), CCV1 (n ,k ) , and HV' (n,k) are
GL(n-k, HR) , GL(n-k,(C) , and GL(n-k, 3H) , respectively. The
corresponding homeomorphisms

GL(n, IR)/GL(n-k, 3R) + V'(n,k), GL (n ,(C)/GL (n-k ,(C) -»■ E V ’ (n ,k) ,

and
GL (n, IH) /GL (n-k , IH) -> 3HV ' (n,k) ,

are the injective factors of the maps GL (n, IR) -> V'(n,k),
GL(n,(C) + (CV' (n,k) , and GL(n, H) 3HV’ (n,k) , defined in Subsection
3.2.1 (see 3.2.1.7, 3.2.1.8, and 3.2.1.9). The isotropy subgroup of
GL+ (n, IR) at the point [ (x^,...,x^) h- (0 , ...,0 ,x ,. ..,x^)] E V ’(n,k)
is GL+ (n-k , IR) .
All the right actions are free. The corresponding orbit
spaces, V ’ (n,k)/GL (k, IR) , V 1 (n ,k) /GL+ (k , IR) , (CV 1 (n,k) /GL (k ,(C) , and
]HV1 (n,k)/GL(k, ih) , are canonically homeomorphic to the Grassman
manifolds G(n,k),G+ (n,k), £CG(n,k), and 3HG(n,k); the corresponding
canonical homeomorphisms are the injective factors of the maps
V 1 (n,k ) +G(n,k), V 1 (n ,k) G (n,k) , (CV!(n,k)->(CG(n,k),and
H V 1 (n,k) -*IHG(n,k), defined in Subsection 3.2.2 (see 3.2.2.3, 3.2.2.7,
and 3 .2 .2 .8) .

14. GL (n, IR) and its subgroups GL+ (n, IR) , 0(n), and
S0(n) obviously act continuously from the left on the Grassman
manifolds G(n,k), G (n,k). Similarly, GL(n,(C) and its subgroups
U(n) and SU(n) act continuously from the left on (EG(n,k), while
GL(n, IH) and Sp(n) act continuously from the left on IHG(n,k).
For k odd,the actions of 0(n) and SO(n) on G + (n,k) are effective.
The noneffectiveness kernels of the actions GL (n, IR) x G+ (n,k) -> G + (n,k)
and GL+ (n,k) x G+ (n,k) G
+ (n,k) for k odd consist of scalar
matrices with positive diagonal elements. If we exclude the trivial
cases k = 0 and k = n, the noneffectivennes kernels of the remaining
actions consists of all scalar matrices contained in the corresponding
group. The only intransitive actions are GL(n, IR) x G+ (n,0) G+ (n,0) ,
GL+ (n, IR) x G+ (n,0) G + (n,0), 0(n) x G + (n,0) G + (n,0),*
SO (n) x G (n,0) -+G + (n,0), GL+ (n, IR) x G + (n,n) G + (n,n), and
SO (n) x g
+
(n,n) -+G+ (n,n) .Take the plane x =I 0,...,x n ,
Js. = 0
 294

(oriented in the case of G^

+ (n,k)) as a distinguished point in the
manifolds G(n,k), G (n,k), ŒG(n,k), and IG(n,k). Then the isotropy
subgroups of the actions GL(n, 3R) x G(n,k) -> G(n,k),
GL (n ,3R) x G (n,k) + G+ (n/k ) / GL (n,Œ ) x ŒG(n,k) ŒG(n,k), and
GL (n, 3i) xüHG(n,k) ->]HG(n,k) at these distinguished points are the
subgroups of all matrices of the form

where and B are nonsingular matrices of order n-k and k,

respectively, and C is an arbitrary (n-k)xk matrix (and B E GL+(k/IR)
in the case of G (n,k)).
If we restrict the acting group to a subgroup, then the new
isotropy subgroup is the intersection of the original isotropy subgroup
with the new acting group. In particular, for the actions of 0(n) on
G(n,k) and G (n,k), the action of SO(n) on G + (n,k), the action of
U(n) on (EG(n,k), and the action of Sp(n) on ]HG(n,k), the
corresponding isotropy subgroups are the images of the monomorphisms
O(n-k) x 0(k) 0(n), O(n-k) x SO(k) -* 0(n), SO(n-k) x SO(k) SO(n),
U (n-k) x U (k) U (n) , and Sp(n-k) x Sp(k) Sp(n), all defined by
the matrix formula
(A
(A,B) I
lo
If we identify these product with their images, we obtain canonical
homeomorphisms

0 (n) / [0 (n-k) x 0 (k) ] G(n,k), 0(n)/[0(n-k) x SO (k )] G+(n,k),

SO (n) / [SO (n-k ) x SO(k)] -> G (n,k),

U (n) / [U (n-k ) x U (k) ] -v (EG (n ,k ) ,

Sp (n) / [Sp (n-k ) x Sp (k ) ] +IG(n,k) .

15. Let m, be relatively prime positive integers.

The complex-number formula
2ttikL . / m 2itikt /m
(k , (z1 ,.. .,zn )) *+ (z ^e ,.,w zne n ),

where k G 7L , (z^ ./2n ) £ S n ^ , defines an action 7L x S^n ^

2n-1
with noneffectiveness kernel m2Z , which becomes, by shifting
to the effective factor, a free action of the group
 295

s2tl 1 - The orbit space S2n 1/S is denoted by L ) and

m I n
is called a lens (or a lens space).

There are also infinite lenses L(m;£1 , t 2 ,...), with

m , t ^ , ¿ 2 •' • • relatively prime positive integers. The lens
L (m ;£.|/¿2 » •• •) defined as the orbit space of the free action
resulting from passing to the effective factor of the action

2iti k i /m 2iTikl ~ / m
(k, (z1 ,z2 , ...) ) h- (z 1e ,z2e 1 ,...)
OO

of 7L on S . An equivalent decription: L (m ;£ , ¿ 2 > •• •) =

= lim(L(m ; £ . . . , 1 ) ,in: L(m;£1,..., I ) + L ( m ; l . , . . . , 1 )). The

i n i n 1 n +1
inf inite lens L (m ;1 ,1 , ...) is denoted simply by L (in) .
2n-1
According to 8, the triples (S ~ ,pr,L( m ; l , . . . , 1 )) and
00 i n

(S ,pr,L ( m ; l ^ , ¿ 2 '• * - ) ) are coverings.

16. The formula (y,x) ^ yxy , where x and y are

quaternions and y has norm 1, defines a continuous action
3 4 4 3
S x hr ir „ The space HR^ of imaginary quaternions is invariant
3 3 3
under this action, and hence IR , and also HR , are S -spaces.

[We identify JR^j with 1R^ via the map shi: JR^ ->■1R^; see 3.2.3.1.]
3 3 3
The noneffectiveness kernel of the action S x jr ]R is obviously
S°, and now it is clear that the effective action of the factor group
S^/S^ = IRP^ on ]R^ becomes the standard action of S0(3) on IR^
(see 11) under the canonical identification of the spaces IRP^ and
S0(3) (see 3.2.3.1) .

17. Let P be a convex regular polyhedron in IR^

(a tetrahedron, cube, octahedron, dodecahedron, or icosahedron) with
center 0. Let GP be the subgroup of S0(3) consisting of those
rotations which take P into itself, and let G P be the preimage of
GP under the projection S S0(3) (see 16). Obviously, GP and
G~P do not change if we replace P by the dual polyhedron, while they
are transformed into conjugate subgroups of S0(3) and if we
replace P by any convex regular polyhedron with the same number of
faces and center 0. Therefore, in S0(3) (S3) there are exactly
three classes of conjugate subgroups GP (respectively, G P) . The
groups in the first class are called tetrahedral groups (respectively,
binary tetrahedral groups), those in the second class - cube or
octahedral groups (respectively, binary cube or octahedral groups),
and those in the third class - dodecahedral or icosahedral groups
 296

(respectively, binary dodecahedral and icosahedral groups).

To every rotation in GP we may associate the image of a
marked oriented edge of the polyhedron P, and in this way define an
invertible mapping of the group GP onto the set of oriented edges of
P. Consequently, the order of the group GP is twice the number of
edges of P, i.e., 12 when P is a tetrahedron, 24 when P is a
cube or octahedron, and 60when P is a dodecahedron or an
icosahedron. Thecorresponding binarygroups G Phave order 24,
48, and 120.
3 ~
The coset spacesSO(3)/GP and S /G P are orbit spaces of
the free actions induced by the left canonical actions of SO(3) and
3 — 3
S under the inclusions GP -* SO(3) and G P -> S . Therefore, the
3 3 ~
triples (SO(3),pr,SO(3)/GP) and (S ,pr,S /G P) are coverings (see 8).
Obviously, we can write SO(3)/GP = S^/G P.

4. Exercises

1. Show that for any smooth manifold X the first and the
second topologies on Top X coincide.

2. Let X denote the subset of HR consisting of the points

0 and 2n , all n £ 7L . Show that the first and the second topologies
on Top X are distinct.
3
3. Show that the canonical diffeomorphism SU(2) S (see
3.2.1.5) is a group isomorphism.

4. Show that the lenses L (m ;£1 , ..., 1 , ) and L (m ;£ ', .. .,-£/)

I k I k
are homeomorphic whenever for each i the sum or the
difference I , - Z\ is a multiple of m.
2 2
5. Show that the submanifold Tang^ 3RP of Tang IRP
consisting of the unit tangent vectors (i.e., of the images under the
2 2
map d pr : Tang S -* Tang 3RP of the unit tangent vectors) is homeo­
morphic to the lens L(4;7,7).

6. Consider the action of ffi on the manifold V(3,2) of

2
unit vectors tangent to S , where the nonzero element of takes
each vector v into -v. Show that the orbit space V(3,2)/ZZ2 is
homeomorphic to L(4;7,7).
2 9
7. Consider the action ZZ2 x Tang^ 3RP -* Tang^ 3RP (see 5) ,
where the nonzero element of takes each vector v into -v.
2
Show that the orbit space Tang^ ]RP /S^ is homeomorphic to the coset
 297

3 3
space S /H, where H is the subgroup of S consisting of the
quaternions ±ort^, i o r t ^ ±ort^, ±ort^.

2 2
8. Consider the action Z>2 x TP -*(CP , where the nonzero
element of takes each point (z :z :z )into (z : z n : z 0 ) . Show
^ 2 i z J a 1 z 3
that the orbit space (EP / Z 2 is homeomorphic to S .

9. Consider the action of on (EP 1x (EP1 , where the

nonzero element of 7Ltakes each point ((z :z ) , (w :w )) into
^ I A I A
((z^ :z^) f (w^ :w^) ) . Show that the orbit space ((EP^ x (CP^ )/22^ is
4
homeomorphic to S .
2 2
10. Consider the action of 7L^ on S x S, where the
nonzero element of a takes each point (x,y) into (y,x). Show
2 2 2
that the orbit space (S x s )/ ^ 2 is homeomorPhic to (EP .

§3. BUNDLES WITH A GROUP STRUCTURE

1. Spaces With F-Structure

1. The bundles which we encouter most frequently have fibers

thatbesides being merely topological spaces, carry some additional
structure: for example, they may be vector, Euclidean, or Hermitian
spaces. In the present section we shall introduce this concept of
additional structure into the theory of bundles.
We begin by giving an exact description of the necessary type
of structures and then fit them systematically into the basic definitions
of the theory, given in § 1 (see Subsections 1.1 and 1.2).

2. Let G be a topological group, and let F be an effective

G-space. We say that the topological space W is endowed with an
F-structure if there is given a nonempty set A of homeomorphism F W
such that, for an arbitrarily fixed homeomorphism a €A, a given
homeomorphism 6: F -> W belongs to A if and only if 3 o a is the
transformation induced by one of the elements of G. The homeomorphisms
of A are called marked.
Every marked homeomorphism naturally carries the action of G
from F to W. If G is commutative, then the resulting action
 298

G x w W does not depend upon the choice of the marked homeomorphism,

and hence in this case the F-structure reduces to the action of G. If
G is not commutative, then an F-structure does not define a canonical
action of G on W.
We remark that F itself has a canonical F-structure, namely
that whose marked homeomorphisms are the transformations induced by the
elements of G.
In the simplest case when G is the trivial group, a space
with an F-structure is simply a topological space canonically homeo-
morphic to F.
3 (EXAMPLES). If G = GL (n, JR) and F = IRn with the usual
action ofthis group, then a space with an F-structure is nothing else but
an n-dimensional vector space, and fixing a marked homeomorphism is
simply fixing a basis of the space.
If G is one of the groups GL (n,3R), 0(n), or SO(n),
and F is JRn with the usual action of these groups, then a space
with an F-structure is an oriented n-dimensional real vector space,
an n-dimensional Euclidean space, or an oriented n-dimensional Euclidean
space, respectively. When G is GL(n,(C) or U(n), and F is CCn
with the usual action of G, then a space with an F-structure is an
n-dimensional complex vector space, or an n-dimensional Hermitian space,
respectively.
If G = DiffrX, where X = F is a Cr-manifold (1 £ r £ a)
and G acts as usual, then a space with an F-structure is a Cr-manifold
27
which is C -diffeomorphic to X.
If G = Top X , where X is a locally compact Hausdorff space,
and F = X with the usual action of Top X , then a space with an
F-structure is simply a topological space homeomorphic to X.
If G is the group of all simplicial autohomeomorphisms of
the unit simplex Tn , and F = Tn with the standard action of this
group, then a space with an F-structure is simply an n-dimensional
topological simplex.

4. A homeomorphism W -+■ W 1, where W and W1 are spaces

with F-structure, which takes the set of marked homeomorphisms of W
into the set of marked homeomorphisms of W 1, is called an isomorphism
or, more specifically, an F-isomorphism.
In each of the previous examples, the F-isomorphisms form a
well-known class of maps: in the first and the fifth cases they are the
linear isomorphism, in the second - the orientation preserving linear
isomorphisms, in the third and sixth - the linear isometric isomorphisms,
in the fourth - the orientation preserving linear isomorphisms, in the
 299

seventh - the C -diffeomorphisms, in the eighth - the homeomorphisms,

and in the ninth - the simplicial homeomorphisms.

5. Given a space W with an F-structure and a space W'

with an F 1-structure, the product W x W' is obviously a space with an
F x F ’-structure (see 2.3.6); the marked homeomorphisms F x F1 ^ W x W1
are those of the form a x a ' , where a and a ' are marked homeo­
morphisms .
If the G^-space F^ is obtained from the G-space F by
reducing the group G to G^ , then by returning to F from F^ every
space W^ with an F^-structure becomes a space W with an F-structure:
topologically, W is the same as W^, while the new marked homeo­
morphisms are defined as the compositions of the transformations induced
by the elements of G with the old marked homeomorphisms. We say that
W is obtained from W^ by extending (or prolonging) the group G^ to
G.

2. Steenrod Bundles

1. Let G and F be a topological group and an effective

G-space, respectively. A bundle £ is a weak F-bundle, or a W-F-bundle,
if eachof its fibers is endowed with an F-structure. In this case,
F and Gare called the standard fiber and the structure group of £,
respectively. The set of all marked homeomorphisms from F onto the
fibers of £ is denoted by MH(£). The group G acts naturally from
the right on MH(£) by the rule: [ag](y) = a(gy) [a E MH(£), g E G,
y E F] .
If £
is a W-F-bundle and f : B bs £ is continuous, then
i
clearly the induced bundle f ‘£ is a W-F-bundle: the F-structures on
i -1
its fibers are defined via the homeomorphisms ab tl ad f : pr f "£ (b) -*
-+■ pr E, ^ (f (b) ) [b E B] .
Given two W-F-bundles, 5 and n, a map f of £ into n
is called a W-F-map if the maps ab tl f from the fibers of £ into
the fibers of n are isomorphisms (see 1.4). A W-F-map which is an
isomorphism (respectively, equivalence) in the pure topological sense,
i.e., in the sense of 1.1.2, is called a W-F-isomorphism (respectively,
a w-F-equivalence). Two W-F-bundles which can be mapped into each
other by a W-F-isomorphism (W-F-equivalence) are said to be
W - F -isomorphic (respectively, W-F-equivalent).

To each W-F-map f: 5 + n corresponds the map

 300

MH(£) -► MH(n), which takes each marked homeomorphism a: F pr £ 1 (b)

into the composite homeomorphism F — pr C ^ (b) ab tl f ^ pr ^ ^ (bs f (b) ).
Moreover, we see that M H (f ) is a G-map with respect to the natural
right actions of G on MH( D and M H (n )•
The standard trivial bundle, (B x F,pr^,B), with B an
a r bi tr ar y topolo gi ca l space, is o b v i o u s l y a W-F -bu nd le : the F - s t r u c t u r e s
on its fibers are de fi n e d by the h o m e o m o r p h i s m s F -* b x f , y (b,y).
As in Subsection 1.2, every W-F-bundle which is W-F-equivalent to a
standard trivial W-F-bundle is called a W-F-trivial W-F-bundle.

2. A bundle £ is a strong F-bundle or, simply, an F-bundle

if it is a W-F-bundle and MH(£) is endowed with a topology.
If £ is an F- bu ndl e and f: B bs £ is conti nu ou s, then
i
the induced b u n d l e f"£ is also an F-bundle: to i n tr od uce a t o p o l o g y
i i
on M H (f “O , we use the injective mapping MH(f'C) ^ B x MH(£) given
by a »-> (pr f #Uot(F) ) ,[MH(ad f) ] (ot) ) .
A map f: E, n/ where E, and n are F-bundles, is said to
be an F-map if it is a W-F-map and M H (f ) is continuous. An F-map f
is an F-isomorphism (F-equivalence) if it is an isomorphism (respectively,
equivalence) in the pure topological sense and MH(f) is a homeomorphism.
The standard trivial bundle (B x F,pr^,B), with B an
arbitrary topological space, is obviously an F-bundle: the F-structures
of its fibers were already introduced in 1, and one can introduce a
topology on M H ((B x F,pr^,B)) by means of the invertible mapping
B x g -* MH ((B x F,pr^,B)), which takes each pair (b,g) into the
homeomorphism F -> b x f , y (b,gy) . An F-bundle which is F-equivalent
to a standard trivial bundle is called F-trivial, and every such
equivalence is an F-trivialization.

3. The F-bundle E, is locally F-trivial if every point of

bs E, has a neighborhood U such that the restriction E,|^ is
F-trivial. The locally F-trivial bundles are called Steenrod F-bundles.
Steenrod bundles play a major role in what follows, which
accounts also for the importance of the F-bundles. The weak F-bundles
are only auxiliary.
We remark that for Steenrod bundles the canonical right action
of the structure group on the space of marked homeomorphisms is
continuous and free. This is plainly true in the standard trivial case,
to which the general case reduces.

4. If C is a Steenrod F-bundle and f: B bs £ is

continuous, then the induced bundle f "£ is again a Steenrod F-bundle:
i t ,
by 2, f “£ is an F-bundle, and the obvious fact that f ' E , is F-trivial
 301

if £ is soimpliesthe local F-triviality of f*£. Clearly, the

map ad f : f’£ £is an F-map, the canonical equivalence
i _

(idbs^)“^ ^ C/ and the canonical equivalences of the form

i i i
g*(f*U (f © g)*£ (see 1.1.4), are F-equivalences. Moreover, given
any F-map h of £ into another Steenrod F-bundle, the correcting map,
corr h , is an F-map.
The product of a Steenrod F-bundle £ with a Steenrod
F*-bundle is a Steenrod F x F 1-bundle: the F x F 1-structures on
its fibers is defined as in 1.5; the topology on MH(£; x £') is
introduced by means of the invertible mapping MH(U x MH (£ 1) MH(^ x ^')
given by (a,a1) a x a'; the local F x F 1-triviality of the resulting
F x f 1-b un d l e f o llo ws from the fact that it is F x F '- tr i v i a l wh en ev er
£ is F-trivial and is F'-trivial.
If the -space F^ comes from the effective G-space F by
reducing the group G to G^ , then by returning to F from F^,
every Steenrod F^-bundle becomes a Steenrod F-bundle £:
topologically, £ is the same as ; the F-structures on the fibers
of £ are those described in 1.5; further, to define a topology on
MH(£), consider the action of G1 on G x MH(^1)/ defined as
-1
g 1 (g,a) ^ (g^g/ag^), where G^ acts canonically from the right on
MH(£^) (see 1), and then use the invertible mapping
(G x MH ( ) )/G^ -> MH(^), which takes the orbit of the pair (g,a)
into the homeomorphism y a(gy), to transfer the topology of
(G x MH(^^))/G^ to MH(£); finally, the local F-triviality of the
resulting F-bundle is a consequence of its F-triviality in the case
when ^ is F^-trivial.This transformation of F^-bundles into
F-bundles is knownas the extension (or prolongation) of the structure
group. It takes F^-maps into F-maps, and F^-equivalences into
F-equivalences. It is also clear that the extension of the structure
group commutes with the induction operation; that is to say, if £is
obtained from byextension of the structure group and f: B bs £
1 i i
is an arbitrary continuous map, then f ‘£ is obtained from ky
extension of the structure group.

5. E ve ry S t e e n r o d F - b u n d l e with trivial s tructure g r ou p is

F-trivial.

PROOF. Let £ be a Steenrod bundle with standard fiber F

and trivial structure group. Let F be an open cover of bs £ such
that the bundle is F-trivial for any U G T. Set-
n= (bs £ x F/p r^bs F) . Clearly, the F-tr ivialization n |y i|0 is
unique for any U G V, and these F-trivializations together yieldan
F-trivialization n £•
 302

Theorems Ab out F-maps

6. Suppose that £ and £' are St e e n r o d b u n d l e s with

standard fiber F, B is a t op olo gi ca l space, and p: B bs£ is a
factorial m a p . If t : tl £ tl £ ' and 3 : bs £ + bs £ ' are maps
such that ( t ° tl a d p , B ° p) is an F - ma p p ‘Ç + K ' t then (t,3) is
an F-map Ç .

We need only check the c o n t i n u i t y of 3/ t , and M H ( t ,3)«

The c on ti n u i t y of 3 is an immediate c o n s e q u e n c e of the c o n t i n u i t y of
the c omp os i t i o n 3 ° P and the fact that p is f a ct or ial (see 1.2.3.4).
As for t and MH(t,3), it is eno ug h to ve ri f y their c o n t i n u i t y wh en
£ and Ç’ are st andard trivial F-bundles. In this s ituation, t is
given by x(b,y) = (3 (b) , $ (b) y ) , whe re (j) is some m ap from bs £ into
the structure g roup G. M oreover, if we use the h o m e o m o r p h i s m s
В x G MH (p ’Ç ) , bs Ç x G -> MH (Ç ) , and bs Ç ' x G -> MH(Ç') (which
define the topol og ie s on MH(p*£), MH(£), and MH(£'), re s p e c t i v e l y ;
see 2), then the maps

MH(t ° tladp,3 ° p) : MH(p*Ç) -+ MH(Ç')

and
М Н( т ,3) : MH(Ç) MH ( Ç ' )

are t ra ns f o r m e d into the maps

B x G + bs Ç' x g, (b ,g) » (3 ° p(b) , (ф о p (Ь) )g) ,

and
bs Ç x g -> bs Ç ' x g , (b ,g) » (3 (b) ,ф (b)g) ,

respectively. The first formula shows that ф ° p is co nt in u o u s , and

since p is factorial, ф is co nt inuous. F inally, the c o n t i n u i t y of
ф implies the c o n t i n u i t y of т an d M H ( t ,3).

V. Suppose that £ and are Steenrod bundleswith

standard fiber F an d 3 • bs £ -*■ bs £ ’ is c o n t i n u o u s . If
t : tl Ç -> t l Ç ' is a map such that the r e s t r i c t i o n s т . , 3I
pr Ç (U) lU
form a n ‘F- m a p £ j^ , for ea ch el e m e n t U of some f u n d a m e n t a l c ov e r
of bs £ , then (x,3) is an F- ma p

This is a co ro l l a r y of 6: take p to be the m ap

U -> bs E, , where T is the given f u n d a m e n t a l c o v e r of bs C .

8. If the S t e en ro d F - b u n d l e s £ an d % ' have the same

 303

base, then every F-map f: £ e/ with bs f = id bs £ is an

F-equivalence.
—1 —i
All we need to prove is that tlf and MH(f) are
continuous, and it suffices to examine the case when £' = £ and £
is a standard trivial F-bundle. Then tl f ,tl f"1 : bs £ x f bs £ F
are given by

(b,y) b (b,cf)(b) y ) t (b,y) H» (bf^(b)y) [b E bs £ , y € F] ,

where <J> and \fj are some maps from bs £ into the structure group G.
Moreover, if we use the topologizing homeomorphism bs£ x g +
-1
then MH (f) and MH (f) become the maps bs ^ x G ^ b s ^ x g given
by

(b ,g) b (b ,cf>(b)g) and (b,g) (b,ip(b)g ) [b C bs £ , y € F] .

_ *]
Obviously, iJMb) = [cf)(b) ] , and thus the continuity of MH (f ) implies
first the continuity of <j> and ip , and then the continuity of
and MH(f) \

9 (COROLLARY) . The correcting map, corr f , is an

F-equivalence for every F-map f between Steenrod F-bundles.

Principal Bundles

10. A Steenrod bundle is called principal if its standard

fiber is the structure group G which acts canonically from the left
on itself (see 2.3.6). We take the liberty to denote the last G-space
simply by G and, accordingly, the principal bundles with structure
group G will be referred to as Steenrod G-bundles.
A fundamental property of the principal bundles is that their
spaces of marked homeomorphisms can be identified with their total
spaces. More precisely, given a principal G-bundle £, the formula
a *+ a(e^) defines a homeomorphism MH(£) tl £. For a standard trivial
bundle, this is evident, and the general case is readily reduced to the
standard trivial one.
If we identify MH(0 and tl £ via the homeomorphism
a •+ a ( e „ ) , then the natural right action of G on MH(£) (see 1)
G
becomes the free right action of G on tl£ . This free, action,
tl £ x G tl £, can be also described directly: its orbits are exactly
the fibers of £, and on each fiber the action is simply the right
canonical action, transferred from G to the fiber by means of marked
 304

homeomorphisms.
11. This construction of the free right action of G on the
total space of a principal bundle with structure group G can be
partially reversed. Assume that the topological group G acts
continuously and freely from the right on the topological space X, and
consider the bundle (X,pr,X/G).Its fibers (orbits) carry natural
_ 1
G-structures: the marked homeomorphisms G pr (b) (b € X/G) are
_ 1
given by g xg, x £ pr (b). Since to distinct pointsx correspond
distinct homeomorphisms g xg, we obtain also an invertible map of X
onto the set of marked homeomorphisms, and thus we get a topology on the
last set. Therefore, (X,pr,X/G) is a G-bundle.
To explain why we called this last construction a partial
inversion of the original one, apply it now to the right action
tl^ x G ->■ tl £ described in 10; the resulting bundle is exactly £.
More precisely, the injective factor of the projection pr C maps
tl £/G onto bs C , and together with id(tl^) forms a G-isomorphism
(tU,pr,tl £/G) +

12. If the G-bundle (X,pr,X/G), defined by a free right

action of G, has a section, then it is G-trivial. In particular,
every Steenrod G-bundle having a section is G-trivial.

Indeed, if s: X/G X is a section, then themap

f: ((X/G) x G,pr^ ,X/G) ->(X,pr,X/G), given by tlf(b,g) = s(b)g, is
a G-trivialization of the bundle (X,pr,X/G).

13 (COROLLARY). If the G-bundle (X,pr,X/G), defined by a

free right action of G, is topologically trivial, then it is G-trivial.
If (X,pr,G/X) is locally topologically trivial, then it is locally
G-trivial, i.e., it is a Steenrod G-bundle.

3. Associated Bundles

1. Let G be a topological group, and let F and F' be

effective G-spaces. The construction below associates to each Steenrod
F-bundle £ a certain Steenrod F'-bundle having the same base.
-] _

The formula g(a,y) = (ag,g y), where g € G, a £ MH(£),

and y 6 F', defines a right action of G on MH(0 x F' (here G
acts canonically from the right on MH(£); see 2.3). Let denote
the bundle with total space (MH(£) x F')/G, base bs £ , and whose
projection takes the orbit of a pair (a,y) 6 MH(£) x f ’ into the point
pr £ (a(F)) . The fibers of this bundle carry a natural F '-structure:
 305

the marked homeomorphisms F1 (pr ^ (b) are given by y pr(cx,y) ,

where a £ MH (£) is such that a(F) = pr £~1 (b) . Since distinct homeo­
morphisms a yield distinct homeomorphisms y pr(a,y), we obtain at
the same time an invertible map MH(£) MH (5 1) / which we use to
topologize MH(£'), and thus make from an F 1-bundle. Finally,
S'Ju is F '“trivial for each set U such that is F-trivial.
Consequently, £' is locally F*-trivial, i.e., it is a Steenrod
F 1-bundle. We say that is the F 1-bundle associated with £ and
denote it by asso(^,Ff).

2. We add four remarks to the above description of the a

construction:
(i) The map tl £ (MH(£) x F)/G which takes each point
x G tl £ into the orbit consisting of the pairs (a,y) £ MH(£) x F
with a(y) = x, is obviously a homeomorphism; together with id bs £ ,
this map defines an F-equivalence £ ^ asso(£,F). Therefore, the bundle
asso(C,F) is canonically F-equivalent to

(ii) The invertible map MH(£) MH (asso (^,F 1)) that we used
to topologize MH(asso (£,F 1)) , is a G-map with respect to the right
canonical actions of G on MH(£) and MH(asso(£, F 1)). As a corollary,
we may state that, given an arbitrary effective G-space F", the product
of the above invertible G-map with id F" is a G-map MH(£) x f"
-> MH (asso (^ ,F1) ) x F", where G acts from the right on MH(£) x F"
_ -i
and MH(asso(£,F')) x F" by g(a,y) = (ag,g 'y). The resulting homeo-
morphism (MH(£) x F")/G -»• (MH (asso (£ ,F ') ) x F")/G, together with
id bs £ , define an F"-equivalence asso(£,F") asso (asso (£ ,F') ,F" ) .
Therefore, the bundles asso(asso(C,F'),F") and asso(£,F") are
canonically F"-equivalent.

(iii) The bundle asso(£,G), i.e., the principal bundle

associated with is canonically G-isomorphic to the G-bundle
(MH(£),pr,MH(£)/G) defined by the canonical right action of G on
MH(5) (see 2.11). The canonical G-isomorphism (MH (£) ,pr ,MH (£) /G)
asso(£,G) is given by the homeomorphism MH(£) (MH(C) x G)/G
which takes each a 6 MH(£) into the orbit of (a,eG ).

(iv) If F' is a subspace of the G-space F" (see 2.3.6),

then (MH(£) x f')/G is a subset of (MH(£) * F")/G, and the inclusion
(MH(U * F')/G (MH(£) x F")/G together with id bs E, yield an
inclusion of the bundle asso(£,F') into asso(£,F"). Moreover,
MH (asso (5 ,F ')) is exactly the set of maps ab a : F ’ -> a (F ' ) with
a e M H (asso(?,F")).
 306

Behavior With Respect to Maps

3. Let F and F* again be effective G-spaces, and supp

that £ and n are Steenrod bundles with standard fiber F, and
f: £ n is an arbitrary F-map. Define the map

asso(f,F '): asso(£,F ’) + a s s o ( n , F ’)

by the formulas bsasso(f,F') = bs f and tl asso(f ,F' ) =

= [fact (MH (f) x id F 1) : (MH(£) x F ’)/G (MH(n) x F')/G].

It is clear that asso(f,F’) is an F'-map. Moreover,

asso(f,F') is an F '-isomorphism (F'-equivalence) whenever f is an
F-isomorphism (respectively, F-equivalence). Next, consider the
diagrams

Ç --- > asso (Ç ,F) asso (Ç ,F") ----> asso (asso (£ ,F 1) ,F fl

f asso(f,F) asso(f,F") asso(asso(f,F'),F")

n asso(ri,F), asso (rs,F" ) » asso(asso(r»,F’) ,F" ) ,

and

(MH(Ç) ,pr,MH(Ç)/G) --- ► asso(Ç,G)

(MH (f ) ,fact MH (f ) asso(f,G)

(MH(n) t pr,MH(n)/G) --- > asso(n fG) ,

where FM is any effective G-space, and the horizontal arrows denote

successively the canonical F-equivalences described in 2(i) and 2(ii),
and the canonical F-isomorphisms from 2(iii) These diagrams clearly
commute.

4. The asso and induction operations commute. Namely, the

map

corr [asso (ad h ,F 1) ] : a s s o f h ^ F 1) h !asso(Ç,F')

is an F 1-equivalence, for any Steenrod F-bundle Ç and any continuous

map h: B ^ bs Ç ; see 2.9.
Furthermore, the asso operation commutes with the extension
of the structure group. That is to say, let F^ and F^j be the
effective G^-spaces obtained from F and F' by reducing the group G
to a subgroup G^ . If the Steenrod F ^-bundle Ç is taken into £ by
 307

the extension of the group to G, then the map

fact (in x id F 1) : ( M H ^ ) x Fjj)/G1 + (MHU) x F ')/G ,

where in = [in: MH(£^) -*■ MH(£)], defines an F '-equivalence between

the bundle obtained from asso^^F^) by extending G1 to G, and
asso (E, ,F 1) .

Weakly Associated Bundles

5. The construction described in 1 can be generalized to the

situation where the action of G on F1 is not effective: we need only
shift, as a preliminary step, to the effective factor of this action,
and thus transform F' into an effective G/K-space, F^ , where K
is the noneffectiveness kernel. Therefore, asso(£,F') may be defined
for any Steenrod F-bundle E, and any G-space F* and is, in the general
case, a Steenrod bundle with structure group G/K and standard fiber
F^ . We say that asso(£,F') is weakly associated with
The map asso(f,F') defined in 3 remains viable under this
extension of the asso construction, and becomes an F^-map. The
properties of asso discussed in 2 , 3 , and 4 must be modified in an
obvious manner; for example, when F" is not effective (but Ff is
effective), the canonical F"-equivalence asso(£,Ff') asso (asso (£ ,F 1),F")
becomes an F^-equivalence.

Sections Associated with F-Maps

6. Let £ and be Steenrod bundles with structure group

G and standard fiber F, and let f: bs £ bs E, 1 be continuous.
The construction below establishes a one-to-one correspondence between
the F-maps h: E, -> E, ' with bs h = f and the sections of a specially
constructed bundle, Fibr(£,E, 1;f) .
Let G denote the group G endowed with the action of
the group G x G given by ( g ^ g ^ g = 9 ^ 9 9 21 (generally speaking,
this is not an effective action). Set

Fibr (£,£'; f) = asso(diag'(£ x f'£'),GX),

where diag = [diag: bs C ^ bs £ x bs £ ], and asso is taken in the

weak sense of 5. It is clear that for every F-map h such that
bs h = f and every point b € bs £ , the composite homeomorphism
 308

pr £~1 (b) ab t l ^ (prC’) 1 (f (b) ) (ab tL ad f)--

i -1 g-1
► (pr f •£’) (6(F) ) -2--- .

where a £ MH(£), pr£(ct(F)) = b, and 3 € MH(f!£'),p r f !£'(B(F)) =

= b, is simply one of the transformations F + F induced by the
elements of G. We denote the corresponding element by g(a,3), and
note that g i a g ^ S g ^ = g 1g(a,3)g2 for anV V y ?2 e G <here Gacts
canonically from the right on MH(£) and MH(f*£')) • This shows that
the orbit of the pair (a x 3rg(a,3)) E MH(£ x f ’£') x G under the
» x _
right action of G x G on M H (£ x f *£ *) x G ,constructed as in 1,
does not depend upon the choice of a and 3 r provided
h and b
I x
are fixed. When h is f ixed, the map bs C (MH (£ x f '£ ') x g )/ (G x g )
taking b E bs £ into this orbit is continuous and manifestly a section
of the bundle Fibr(£,£ 1;f) . We call it the section associated with h
X X
and denote it by h. The correspondence h h defines an invertible
map from the set of all F-maps h: £ + £ 1 with bs h = f onto the set
of sections of Fibr (£,£';f): the inverse map takes each section
s: bs £ ^ tl Fibr (£ ,£ ';f) into the F-map h: £ £' given by

tlh(x) = tl adf(Bv
x (gv
x (a"1
x (x) )) ) ,

where x £ tl £ , £ M H (£), 6 £ MH (f "£ '), g^ £ G , and

(aX x SX ,g
X ) £ s ° pr £ (x) .
Let us remark that when bs £ ' = bs £ and f = id bs £ , an
F-map h: £ -> £ 1 with bs h = f is simply an F-equivalence (see 2.8).
Therefore, when £ and £' have the same base, the above construction
yields a one-to-one correspondence between the F-equivalences £ -> £ 1
and the sections of the bundle Fibr(£,£';id bs £) .

4. Ehresmann-Feldbau Bundles

1. An Ehresmann-Feldbau bundle is a W-F-bundle which is

locally W-F-trivial; the last means that every point of the base has a
neighborhood such that the restriction of the bundle to this neighborhood
is W-F-trivial.
The theory of Ehresmann-Feldbau bundles is a variant of the
theory of bundles with a group structure; it is simpler than the theory
of Steenrod bundles (there are fewer structures), but also less pithy
(there are no associated bundles). This relative poverty nearly deprives
 309

it of any independent value; however, the fact that it is equivalent,

for a large class of standard fibers which includes the most important
cases, to the theory of Steenrod bundles, makes it useful, as it enables
us to simplify the latter.

The Case of Topologically Effective Actions

2. A continuous effective action G x X X is said to be

topologically effective if given any topological space Y and any map
f: Y G, the continuity of the composite map

Y x X —f X ld > X x G -> X (1 )

implies the continuity of f. In this case we also say that the

G-space X is topologically effective.
Clearly, if we reduce the group, a topologically effective
action remains so, and every G-space which has a topologically effective
subspace is itself topologically effective.
The free actions are immediate examples of topologically
effective actions; in particular, the left canonical action of a
topological group (on itself) is always topologically effective. Also,
the usual actions of GL(n, 3R) and of its subgroups on lRn are all
topologically effective.
In order that a G-space be topologically effective, it is
necessary that the map c : G -> C(X,X), which takes each g £ G into
the transformation induced by g, be a topological embedding; if X is
locally compact and Hausdorff, then this condition is also sufficient.
- 1
The necessity is plain (take Y = c(G) and f = [ (abc) : c (G) G] ).
Now assume that the condition is satisfied. Then the continuity of
f : Y -* G is equivalent to the continuity of the composite map

Y— —►G » C (X,X) . By Theorem 1.2. 7. 6, for X locally compact and

Hausdorff the last map is continuous because so is the map (1). In
particular, for G compact, every effective, locally compact Hausdorff
G-space is topologically effective (see 1.1.7.10 and 1.2.7.2). If X
is Hausdorff, locally compact, and locally connected, then the usual
action of the group Top X on X is topologically effective, and the
same holds when X is Hausdorff and compact (see 2.2.5 and 2.2.6).

3. If the standard fiber F is topologically effective,

then every W-F-map (respectively, W-F-isomorphism, W-F-equivalence)
between Steenrod bundles is an F-map (respectively, F-isomorphism,
 310

F-equivalence).
In fact, let £ and be Steenrod F-bundles. We need only
prove the continuity of the map M H (f ) corresponding to the given
W-F-map f: £ + and we may assume that £ and are standard
trivial bundles. In this case, tl £ = bs £ x F, tl £ 1 = bs £ 1 x F,
and tl f is given by (b,y) ^ (bs f (b) ,<f>(b)y ) , where <p is some map
of bs£ into the structure group G. At the same time,if we use the
canonical homeomorphisms bs £ x g -> MH(£) and bs £ 'x G -> MH ( ^' )
(see 2.2), then MH(f) becomes the map bs ^ xg bs £ ' x g ,
(b,g) (bs f (b) ,<j>(b) g) . The continuity of tl f implies the
continuity of the composition pr^ o tlf : b s £ x p - ^ F , which equals

the composition bs Fy x F ft x — > G x f --- ► F, where the last arrow

denotes the action. Since this action is topologically effective, (p
is continuous, and so is MH(f).

4. Given an Ehresmann-Feldbau bundle with a topologically

effective standard fiber, there is a unique topology on the set of its
marked homeomorphisms which transforms this bundle into a Steenrod bundle.

The uniqueness of this topology is a consequence of Theorem 3.

Let us prove its existence. Let £ be an Ehresmann-Feldbau bundle with
topologically effective fiber F. Cover bs £ by open sets U such
that is W-F-trivial, and fix W-F-equivalences h ^ : (U x F,pr^,U)+
ijy- Now topologize the sets MH (£|^) with the aid of the maps
MHihy): MH((U x F ,pr 1 ,U) ) + MH (£|^) . We obtain a cover of MH(£) by
topological spaces MH (£|
q )/ and Theorem 3 shows that these spaces
induce the same topologies on their intersections, as required for the
construction in 1.2.4.3. The topology on MH(£) produced by this
construction transforms £ into a Steenrod bundle.

Locally Trivial Bundles as

Ehresmann-Feldbau Bundles

5. If F is a locally compact Hausdorff space endowed with

the usual action of the group Top F , then an Ehresmann-Feldbau
W-F-bundle is simply a locally trivial bundle with fibers homeomorphic
to F. Therefore, any ordinary locally trivial bundle whose fibers are
locally compact Hausdorff spaces homeomorphic one to another, may be
regarded as an Ehresmann-Feldbau bundle, and as such it has an implicit
group structure. If* in addition, the fibers are locally connected or
compact, then such bundles can be also regarded as Steenrod bundles.
 311

We remark that the last assertion is also true for all coverings in the
broad sense with connected bases.

5. Exercises

1. Show that all the effective actions listed in 2.3.11,

2.3.12, 2.3.13, and 2.3.14. are topologically effective.
I*
2. Let X be an arbitrary C -manifold (r ^ 1) of positive
dimension. Show that the usual action of Diffrx on X is
topologically effective.

3. Consider IR as a 7L -space with the action (n,t) ^ t + n

(n € 7L , t € IR) , and using the same formula, extend this action to
an action of the additive group ]R, equipped with the discrete topology.
Show that this extension of the structure group takes asso ((ER,hel,S ) JR)
into a bundle which is not trivial as a Steenrod bundle, but is trivial
as an Ehresmann-Feldbau bundle. (Cf. 4.3.)

4. Suppose that G is a connected topological group, F is

an effective G-space,and £ is a nontrivial Steenrod F-bundle with
6 6 6
simply connected base. Denote G , F , and £ the group G equip-
ped with the discrete topology, the space F, regarded as a G -space,
and the bundle £, regarded as a W-F -bundle, respectively. Show that
there is no topology on the set of marked homeomorphisms of £ which
6 6
makes from £ a Steenrod F -bundle. (Cf. 4.4.)

§ 4. THE CLASSIFICATION OF STEENROD BUNDLES

1. Steenrod Bundles and Homotopies

1. We now turn to the problem of classifying the Steenrod

bundles with a given standard fiber F and a given cellular base B
with respect to F-equivalence. Our main achievement in this section
is to establish a canonical one-to-one correspondence between the
classes of F-equivalent bundles over B and the homotopy classes of
maps from B into a specially constructed space that depends only upon
the structure group. This correspondence reduces the given classifi-
 312

cation problem to a problem in ordinary homotopy theory.

Lemmas About F-Trivial Bundles

2. Let £ be a Steenrod bundle with standard fiber F, and

let B1 and B2 be closed subspaces of bs £ , such that B^ U B2 =
= bs£ and B . n B- is a retract of B_ .
If the restrictions £|
\ 2, ' “ " |^
and £ |„ are F-trivial, then £ is also F-trivial.
IB2

PROOF. Choose a retraction p: ^ fl and two

F-trivializations,

h 1 : n1 = (B1 x F fpr1/B1) + I |B
and 1
h 2 : n 2 = (B 2 x F , p r 1 , B 2 ) -► £ | B »

and denote by f the composite F-equivalence

, id | abh1 (abh2 )~1

n2|BinB2 ' ni |B flB2 ” ? |b nB2 ” n2|BinB2 *

Obviously, tl f is given by (b,y) ^ (b,<J)(b)y) [b £ B^ n y £ F] ,

where $ is some map of B^ 0 B^ into the structure group G. More­
over, if we use the canonical homeomorphism (B 0 B0) x g MH(M0 ),
1 2 l B 1n B 2
then MH (f) becomes the map (B^ n B^) x G -+ (B n B2 ) x G, given by
(b,g) h- (b,c()(b)g) [b £ B^ n B ^ , g £ G] . Therefore, the continuity of
MH (f) implies the continuity of <j>, which in turn yields the continuity
of the map B^ x F -> B^ x F , (b,y) (b,d) o p(b)y). But this last map,
together with id B2 , form an F-equivalence f's n2 -> n2 * and the
obvious equality

f * [abf' : n2 |

shows that the composite F-maps

"1 ^ C |b , C and *2 — "2 — i|B2 — 5

coincide on 0 B2 . By 3.2.7, from this it follows that these two

maps yield an F-map (B x F ,pr 1 /B) ->• £. Finnally, by 3.2.8, the last
map is an F-equivalence.

3. Every Steenrod F-bundle with base In is F-trivial.

PROOF. Let n be an arbitrary Steenrod F-bundle with

 313

bs n = In . Find a positive integer N such that n is F-trivial over

any cube of edge 1/N contained in ln . Now divide In , as usual,
into Nn such cubes, arrange them in dictionary order Q ,...,Q ,
± Nn
and set Q j • Induction shows that n is F-trivial over each
of the sets W^,...,W n : to go from to ' aPPly Lemma 2 to

^ = ' B i = w -i' and B 0 = Q..1 . We conclude that n is F-trivial

l^i+l 1 1 z i+I

over W = In .
Nn
4- Let and £ be Steenrod bundles with common standard
fiber F and common base B, and let A be a retract of B. If £
and E, ^ are F-trivial, then for every F-equivalence h: ^ ^2 |A

there is an F-equivalence h ’: ^ £2 such that [abh1 : ^ |a ^ ^2 |a ^ =

= h.

It is enough to prove this assertion for the case where £

is the standard trivial F-bundle (B x F,pr^,B), and = £-j• Let
p: B A be a retraction. Obviously, tl h is given by
(a,y) ^ (a,cf>(a)y) [a E A, y £ F] , where <f> is some map of A into
the structure group G. Moreover, via the canonical homeomorphism
A x G + MH(£^|a ) [= MH (? 2|a )1 'becomes the map A x G A x G,
(a,g) h- (a,({)(a)g) [a € A, g £ G] . Therefore, the continuity of MH (h)
implies the continuity of (p , which in turn implies the continuity of
the map B x F B x F, (b,y) b- (b,cj) ° p(b)y). The latter and id B
yield an F-equivalence h ’: £ which extends h.

The Homotopy Invariance of the Induced Bundle

5. Let £ be a Steenrod bundle with standard fiber F, and

let f and fn becontinuous maps of a cellular space X into bs £ .
1 z , ,
If f^ and f^ are homotopic, then the bundles f ’£ and are
F-equivalent. Moreover, if f1 and f^are A-homotopic, where A is
I z , t
a cellular subspace of X, then there is an F-equivalence f ’£ -*
which is the identity on f ’^j^.

PROOF. Pick an A-homotopy, H: X x I bs £ , from f^ to

f2 , and set ^ =(f!C) x(I,id 1,1) , ^2 = H !?. It is clearthat

?l|(xxO)U(Axl) = ?2 |(XXO)U(AXI) and that the canonical homeomorphism

i 1 1
X -»■ X x 1 transforms £ and ?2 Xx1 into and ,
respectively. Therefore, it suffices to find an F-equivalence £ ^ -* £2
 314

which is the identity over (X x 0) U (A x I).

We produce such an F-equivalence by taking the limit of a
sequence of F-equivalences, hu: C-||c ^ ^ 2 IC ' where
’ i i
where C± - (X x 0) U (A x I) u (skeiX x I), such that each map hi
extends the preceding one. Take h_^ to be the identity map, and
assume that the F-equivalence h^ is already constructed. To get hj_+ 1 ,
suppose that X is rigged, and for each cell e € cell^^X \ cell^ + ^A
consider the bundles

[ab(chae x idl)]l(?1|c ) = [(chae x i d I ) T

* i (Dl+1xO)U(S1xi;
and
[ab (cha x id I )]'(£., I ) = [(cha x id I ) 'g ] I . , ,
e ^'Ci e | (D1 x 0)U (S xi)
where
ab(cha x id I) =[ab(cha x id I ) : (D1+ ^ x 0) U (S1 x I) ->• c .].
e e 1

Let gdenote the F-equivalence of these bundles defined by h.. By

e , , i
Lemma 3, (chae x i d l ) ’?>| and (chae x i d l ) * ^ are F-trivial, and
since(Di+^ x 0) U (S1 x I) is a retract of D^"+ ^ x i / Lemma 4 shows
that g^ extends to an F-equivalence
^ i i
g^: (chae x id I) -* (chae x id I) ‘ ^2 •

Further, note that the map tl g^" is constant on the elements

of the partition zer (tl ad (cha^ x id I)) and apply Theorem 3.2.6

(with B = Di+^ x I and p = [abicha^ xid I) : D"*"+ ^ x I cl e x I] )

to conclude that the composite map

, g~ , ab ad (cha x id I)
(chae x M D - q ----► (chae x idl)'^--------- ® y x

defines an F-map ?-,|cle xi ^jcie x I* By Theorem 3.2.8, this

is an F-equivalence, which we denote by h. Now note that for any
cells e i , e 2 £ cell^ +^X v cell^+^A, tl hg and tl hg agree over

(Cl e. xI) n (Cl e„ x i), and that for each cell e € cell. „X ^ cell. „A,
1 Z 1+1 1+1

tlh C and tlh._L agree over (Cle x i) n C.. Since the sets C. and
Cle x I, e £ celli+1X \ cell^ + 1A, constitute a fundamental cover of
Ci+1 ' WB use 3 *2 *^ to conclude that h^ and hg , e £ ce l l ^ X \cell.+1A,
form together an F-map U -*• L | . We take this map for h. „
l|Li+1 ^ ,Ci+1 1+1
and note that it obviously extends h ^ ; moreover, by 3.2.7, h~ 1-| is
 315

also an F-map.
To check the rest, i.e., that the sequence {h^ : ^ U ^ ^Ic ^
Ii 'i
converges to an F~equivalence £ it is enough to remark that
the sets constitute a fundamental cover of X x I, and then apply
Theorem 3.2.7 to the sequences {h^} and { h ^ }.

The Sets Stee(B,F)

6. We let Stee(B,F) denote the set of F-equivalence classes

of Steenrod F-bundles over B. Below we shall study the mappings of
this set into itself, defined by the induced bundle construction, by the
extension of the structure group, and by the associated bundle
construction.
i
For any continuous map f: B 1 B, the rule Ç f defines
a mapping

f ’: Stee(B,F) +Stee(B',F).
i
If B'is a cellular space, Theorem 5 shows that f"depends only on
the homotopy class of f. In particular, if B and B' are both
i
cellular andf is a homotopy equivalence, then f ‘ is invertible.
The extension of the structure group, which transforms the
-space F into the effective G-space F, defines a mapping

ext: Stee(B,F^) Stee(B,F),

for any topological space B. This mapping is natural: that is to say,

the diagram
pvf
Stee(B,F 1) ----- > Stee(B ,F )
i i
f‘ f*

Stee(B’,F1) -xt— > stee(B',F)

commutes for any continuous map f: B' B; see 3.2.4.

Given another effective G-space, F', the rule Ç h- asso(Ç,F')

defines for any topological space B a mapping

asso: Stee (B,F) Stee(B,F').

This mapping is invertible [its inverse is asso: Stee(B,F')

Stee (B,F) ] and also natural, i.e., the diagram
 316

asso
Stee (B ,F) ------ > Stee(B,F'

f*

Stee (B 1,F) aSS° > SteefB'j')

commutes for any continuous map f: B' + B; see 3.3.4.

Moreover, the diagram

Stee(B,F1) aSS° » Stee(B,F.|)

f•

Stee(B,F) — aSS° » Stee(B,F')

commutes for any topological space B, any effective -spaces F^ and

F,j , and any effective G-spaces, F and F f, obtained from F^ and
F^j by extension of the structure group; see 3.3.4 and 3.3.5.

2. Universal Bundles

1. Let F be an effective G-space. By Theorem 1.5, give

any Steenrod F-bundle £, any cellular space B, and any continuous
i
map f: B -> bs £ , we may consider the bundle f*?. This defines a
mapping TT(B,bs£) -*■ Stee(B,F), which we denote by induz(B,£):
i
induz(B,£)(the homotopy class of f) = the F-equivalence class of f ’5
The following diagram is obviously commutative for any
topological space C and any continuous map g: C bs £

ir(B,C) — induz (B,g *£) , stee(B>p)

7i(idB,g) n. induz(B,£)

7T(B ,bs O

Similarly, the diagram

TMB. bsU induz ( Bfa s s o < S , F ‘ )) ,

induz(B,£) asso

Stee(B,F)

commutes for any effective G-space F 1.

 317

2. A Steenrod F-bundle £ is called universal if the map

induz(B,£) is invertible for any cellular space B. In other words, a
Steenrod F-bundle £ is universal if: (i) givenany Steenrod F-bundle
5 with cellular base, there is a continuous map f : b s £ -* b s £ such
i
that the bundle f*£ is F-equivalent to £, and (ii) if for two
arbitrary continuous maps, f and f , of a cellularspace into bs£
, , I
the bundles f ’£ and f^£ are F-equivalent, then f^ and f^ are
homotopic.
Condition (i) has an equivalent formulation (i1): given any
Steenrod F-bundle £ with cellular base, there is an F-map 5 £•
Indeed, if f :bs £ bs £ is continuous and g: 5 + f *K is an
F-equivalence, then ad f o g: £ £ is an F-map, Conversely, if
h: ^ £ is an F-map, then corr h : 5 (bsh)’£ is an F-equivalence
(see 3.2.9).
Similarly, condition (ii) is equivalent to (ii1): given any
Steenrod F-bundle £ with cellular base and any F-maps hQ ,h^: £ £,
bs hQ
u and bsh i are homotopic. Indeed, suppose
j that hn
,u ,h 1 have
these properties; then both bundles, (bshg) £ and (bsh^)#£ , are
F-equivalent to £, and so, by (ii), bs hQ and bs h^ are homotopic.
Conversely, if fn and f1 are continuous maps of a cellular space
u , i ,
into bs £ and h: f ^ f ’£ is an F-equivalence, then fQ = bs ad fQ ,
f^ = bs(adf^ o h) and, by (ii'), fQ and f are homotopic.

We remark also that both (i 1) and (ii1) (and hence (i) and
(ii)) are consequences of the following condition: given an arbitrary
F-bundle £ with cellular base and an arbitrary subspace A of bs £ ,
every F-map 5 |A * £ extends to an F-map £ -* £. To see that this
condition implies (i1), it suffices to take A = 0. To see that it
implies(ii'), take the F-bundle £ x(I,id 1,1) , thesubspace
A = (bsKx o) U (bs K x 1) of bs(5 x (I,id 1,1) )= bs 5 x I, and
take as the F-map that must be extended

g: £ x (l,idl,l) |( b s £ x o) U (b s £ x 1) ^ ^ '

with
bs g (b,0) = bs hQ (b) , bsg(b,1) = b s h <l (b) [b e bs C ] ,

tl g (x ,0) = t l h Q (x) , tl g (x,1) = tl h 1 (x) [x £ tl £ ] .

3. Every bundle induced from a universal bundle by a homotopy

equivalence is universal.
Indeed, if £
is a universal F-bundle and f: B -* bs £ is a
i
homotopy equivalence, then the map induz(B,f’£) coincides with
induz (B,£) ° 7T(f ,id B) for any cellular space (see 1), and hence it is
 318

invertible.
4. If the F-bundles £ and C' are universal and bs 5 and
bs £ ' are cellular spaces, then bs g : bs £ bs £ ' is a homotopy
equivalence for any F-map g : £ -* C 1•

PROOF. Pick an F-map g *: K• Since g' ° g : £ -> £ and

g o g': £' ' are F-maps, condition 2 (ii ’) implies that the map
bs g' 0 bs g = bs(g' o g) (respectively, b s g ° b s g 1 = bs(g 0 g')) is
homotopic to id(bs £ ) (respectively, to id(bs £ ')) .

5. A bundle associated with a universal bundle is itself

universal.

Indeed, let £ be a universal F-bundle and let F' be another

effective G-space. Then for any cellular space B, induz(B,asso(£,F '))
is precisely the composition of the maps induz(B,£) and
asso: Stee(B,F) + Stee(B,F') (see 1), and hence isinvertible.

Classifying Spaces

6. As Theorem 5 shows, the base of a universal bundle with

structure group G is simultaneously the base of all universal bundles
with structure group G and all possible standard fibers, and so we may
say that it does not depend upon the choice of the standard fiber. This
base is called a classifying space of the group G.

From Theorem 3 it follows that every space which has the same
homotopy type as a classifying space of G is itself a classifying
space of G. Moreover, it results from Theorem 4 that any two cellular
classifying spaces of G have the same homotopy type.

k-Universal Bundles

7. A Steenrod F-bundle £is called k-universal if

induz (B,£) : 7T(B ,bs £ ) Stee(B,F) issurjective foreach cellularspace
B of dimension £ k and injective for each cellular space B of
dimension ^ k—1. In other words, a Steenrod F—bundle ^ is k —universal
if: a) given any Steenrod F-bundle 5 with cellularbase of dimension
£ k, there is a continuous map f : bs ^ bs C such that the bundle
i
f ’£ is F-equivalent to £, and b) any two continuous maps, f and
f^, from a cellular space of dimension £ k-1 into bs £ , such that
 319

i i
the bundles and f *£ are F-equivalent, are homotopic. Equivalent
conditions are: a 1) given any Steenrod F-bundle £ with cellular base
of dimension ^ k, there is an F-map £ £ and b') for any Steenrod
F-bundle £ with cellular base of dimension £ k-1 and any F-maps
g0'g l : ^ ^' bs g0 and bs g 1are homotoP ic-
In all these formulations k is a positive integer, and the
universal bundles are sometimes termed ^-universal. Every k-universal
bundle is obviously 1-universal for 1 £ k. Moreover, Theorems 2.3.2.4
and 2.3.2.5 show that the restriction of a k-universal bundle with
cellular base to a subspace of the base which contains its 1-skeleton
is 1-universal, for any 1 £ k.
Theorems 3 and 5 have immediate analogs for k-universal
bundles: every bundle induced from a k-universal bundle by a homotopy
equivalence is k-universal; a bundle aasociated with a k-universal bundle
is itself k-universal.

3. The Milnor Bundles

1. Below we shall construct for any topological group G a

principal bundle with structure group G, called the Milnor G-bundle
and denoted by Mi G . In 2-6 we shall prove that Mi G is a universal
G-bundle.
Let TG (k) denote the join of k copies of the group G.
Then TG(k) embeds naturally in TG(k+1) (as a base of the join
TG(k) * G = TG(k+1))f and so the limit TG = limTG(k) is meaningful.
The right action G x G G , (g^ ,g) *+ g 1g ^ ,extends to a free,
continuous, right action of G on TG(k). Since the inclusions
TG(k) TG(k+1) are G-maps with respect to this action, G acts also
on TG. Therefore, a G-bundle (TG,p r ,TG/G) results, and this is MiG.
If g^ is a subgroup of G, then there exist the inclusions
in * ... * in: TG1 (k) TG(k) (k = 1,2,...), where in = [in: G 1 + G] ,
and all of them are in-maps. Moreover, they are compatible with the
embeddings TG (k) + TG(k+1) and TG1 (k) + TG^k+1), and hence the
limit map TG1 + TG is meaningful (see 1.2.4.4). This map is also an
in-map, and together with the map TG^G + TG/G that it induces, it
clearly yields an (in, in)-map MiG1 MiG. Therefore, to each
inclusion G^ -*■ G corresponds an (in,in)-map Mi G^ Mi G .
 320

Mi G is Locally Trivial

2. It is convenient to identify TG (k) with that subspace

of the product con G x ... x con G (k factors) which consists of the
points (pr(g.,t.)}k . such that t. + ... + t, = 1, and is canonically
1 1 1— I I K
homeomorphic to TG(k) (see 1.2.6.4; here pr =[pr: G x I -* con G] ) .
After this identification, the points of TG may be represented as
sequences {pr(g^,t^)}?_^ such that £t^ = 1 and only a finite number
of the t.'s are not zero. Now the right action of G on TG (see 1)
1
is described as {pr(g^,t^)}g = {pr(g g^,t^)}.

3. MiG is locally G-trivial.

Let Ug be the collection of sequences ipr(g^,t^)} with

tg ^ 0, and consider the sets pr MiG(U^) , pr Mi GfU^) ,. .. . These

sets are open, cover TG/G,

and over each of them the bundle Mi G is
-i
G-trivial: the G-trivialization pr Mi G(Us) x G (pr Mi G) (pr Mi G(Us))
[= U ] takes each point (x,g) into the sequence (pr(g.,t.)}
s 1 1
determined by the conditions: pr Mi G ({pr(g^,t^)}) = x and gg = g.

Mi G is Universal

4. Mi G is universal.

According to 2.2, it suffices to show that given any G-bundle

£ with cellular base and any subspace A <= bs £ , every G-map
f: C|A MiG extends to a G-map £ Mi G .

We consider first the case b s £ = Dr+1, A = Sr , forsome

r. Then £ is G-trivial (see 1.3), and we may actually assume that C
is the standard trivial G-bundle (Dr+1 x G ,pr^,Dr+1). The desired
extension h: £ -> Mi G has an explicit description: let k be the
smallest number s such that TG(s) 3 tl f (Sr x e ) , and let <j>.
G i
denote the composite map

Sr x G- f— > TG (k) —iS— > con G x . . x con G i— ► con G

" "k ^
r +1
(i = 1 ,... ,k) ; further, define ip : D x G ->• TG(k+1) by

'Mty,g ) = (tcj)n (y) , ... ,t$k (y) ,pr(g, 1-t) ) ,

where y £ Sr , t £ I, and pr = [pr: G * I -+ con G] . Now set

tl h = [in: TG(k+1) -> TG] ° ij).
 321

The general case reduces to this special one. Indeed, assume

that the space bs £ is rigged and that a G-map h : £I , , + M i G
it AllsKe Ds c,
r
extending f is already available. The above argument shows that for
each cell e e c e l l ^ b s £ \ cellr A the G-map

,r
f = h o [ab cha : S a U ske bs Ç ] : cha *Ç ^ MiG
e r e r e
sr

extends to a G-map g : cha1^ ^ M i G , and it is clear that tig is

“ e e
constant on the elements of the partition zer (tl ad cha^) . Applying

Theorem 3.2.6 (with B = Dr+^ and p = [ab cha^ : Dr + ^ Cl e] ) , we see

that g^ defines a G-map K |C1 e Mi G , which we denote by hQ .

Further, note that for any cells e„,e0 E cell ,bs r \ cell „A,
1 2 r+1 r+1
tl h and tl h agree over Cl e fl Cl e0 , and that for any cell
1 2
e E cell^^bs £ \ cell^^A, tl h^ and tl h^ agree over

Cl e fl (A U ske bs£) . This implies that h and h , eE cell ,bs K

r ^ r e r+1 s
^ cellr+1A ' Y ield together a G-map hr+ : ? L Uske bs £ + Mi G
1 r+1
extending hr (see 3.2.7). Therefore, using induction, we can produce

a sequence {h,.: 5|A U s k bsC + M i G }“=_1 of G-maps with h_1 = f,

such that h extends h

. for all s > 0. Since the sets
s s— 1
A U skegbs £ constitute a fundamental cover of bs £ , using again
3.2.7 we conclude that the hg 's yield a G-map ^ + Mi G extending f.

A Promise

5. The base of the bundle MiG is not a cellular space.

However,we shall see in the next chapter that for any topological group
G there are also universal G-bundles with cellular base; see 5.6.1.4.
By 2.7, this will imply the existence of k-universal G-bundles with
cellular base of dimension £ k, for any given topological group G
and any positive integer k.

4. Reductions of the Structure Group

1. We say that the Steenrod F ^-bundle ^ with structure

group G^ is obtained from the Steenrod F-bundle Ç with structure group
G by reducing the group G to G^ if K is obtained from Ç^ by
 322

extending the group to G.

While the extension of the structure group of a Steenrod
bundle is a well-defined operation, the reduction of the structure
group cannot be carried out for every Steenrod bundle, and even when
it is possible, it may produce bundles which are not equivalent with
respect to the reduced group. In other words, the mapping

ext: Stee(B,F ) ->Stee(B,F) (2)

defined in 1.6 may be both nonsurjective and noninjective.

We remark that the set-theoretic properties of the mapping
(2) are uniquely determined by the triple B, G, G^, i.e., they are
preserved when we replace F and F^ by other effective G-spaces and
their corresponding G^-spaces, while keeping B, G, and G^ the same;
this is plain from diagram (1).

2. Recall that given a cellular space B, Stee(B,F) can be

interpreted as the set of homotopy classes of continuous maps from B
into a classifying space of the structure group. Below we describe (2)
in the same homotopy terms.
Let G, G^, F, F^ be as in 1, and let £ and ^ be
universal bundles with standard fibers F and F 1 . A continuous map
i 1
if;: bs ^ -+ bs £ is called classifying if \p*Z is F-equivalent to the
bundle obtained from ^ by extending the structure group G^ to G.
By the definition of a universal bundle, such amap exists whenever
bs^ is a cellular space, and so it certainly exists when £ = Mi G
and C-| = Mi G^ (see 3.1). Our main claim is that the diagram

Stee(B,F1) — ext -> Stee(B,F)

induz(B,£ ^) induz(B,i) (3)

* ( B , b s Cl > ------ . , ( B , b s t ]

commutes for any classifying map ip and any cellular space B.

PROOF. The composition induz(B,C) ° Tr(idB,c) takes the

homotopy class of f : B -* bs i into the class of the bundle
i 1 1
0 f-,)*?' while the same homotopy class is taken by ext ° induz (B,? )
into the class of the bundle obtained from bY extending the
structure group G.) to G. Since the extension of the structure group
and the induction construction commute (see 1.6), the last class
ii l i t
contains f ’(i|j'£), and it remains to observe that f * (if;*^) = (^ o f )*£.

3. The commutativity of the diagram (3) and the invertibility

 323

of its vertical mappings imply that induz(B,ç^) is an injective map­

ping from the set of homotopy classes of maps g : B -> bs z, such that
o g is homotopic to a given map f: B bs ç , onto the set of
classes of -equivalent F^-bundles which are obtained from f 1ç by
reducing the structure group G to G^ . In particular, a Steenrod
F-bundle Ç with cellular base admits the reduction of the group G
to Gn if and only if any continuous map f : bs Ç bs ç such that
i 1
f"ç is F-equivalent to £ is homotopic to the composition of some
continuous map bs Ç -> bs with \Jj.

5. Exercises

1. Given a topological group G and a positive integer k,

denote by Mi(G,k) the restriction of the bundle MiG to TG(k)/G,
i.e., the bundle (TG(k),pr,TG(k)/G). Show that Mi(G,k) is a
(k-1)-universal G-bundle.
oo 00
2. Show that Mi 2Z9 is isomorphic to
(S ,pr,3RP ), while
k-1 k-1
MiiZ^fk) (see Exercise 1) is isomorphic to (S ,pr, ]RP ).
1 OO 00
3. Show that Mi S is isomorphicto (S ,pr,(CP ), while
1 2k-1 k —1
Mi(S ,k) is isomorphic to (S ,pr,(CP ).
-£
4. Let X be a compact n-dimensional C -manifold, 1 £ r £
£ «>. Consider the right action ofDiffrx on Embr (X, IR^) , given by

(j / ft) ^ j ° ft [j £ Embr (X, 3Rq ) , <J> G Diffrx] , and the limit right
action of Diff X on

lim (Embr (X , 3Rq ) ,ab Cr (id X ,in) : Embr (X,TOq ) -> Embr (X, lRq+1)) .

Show that (lim Embr (X,lRq ) ,pr, [lim Embr (X,Kq )]/DiffrX) is a

universal Diffrx-bundle, while (Embr (X, ]Rq ),pr,Embr (X, lRq )/DiffrX)
is a ( q - 2 n - 1 )-universal DiffrX-bundle, for any q 5 2n+1.
 324

§5. VECTOR BUNDLES

1. General Definitions

1. The main objective of this section is to study those

Steenrod bundles whose standard fiber is either IRn with the usual
action of one of the groups GL(n, ]R) , GL+ (n, IR) , 0(n) , or SO(n) ,
or (Cn with the usual action of GL(n,(C) or U (n) .
Since all the standard fibers listed above are topologically
effective, the corresponding bundles may be also regarded as Ehresmann-
-Feldbau bundles with the same standard fibers (see Subsection 3.4).
We shall proceed in this way and ignore completely the topology on the
set of marked homeomorphisms in the course of the entire section.
To simplify the discussion, we introduce a special notation
for the above standard fibers: GL IRn , GL_IRn , OIRn , SOIRn , and GL(C ,

Standard Fiber GLIRn

2. A Steenrod bundle with standard fiber GLIRn is called

an n-dimensional real vector bundle.
Since a space with a GLnRn -structure is simply an n-dimensio­
nal real vector space (see 3.1.3), a W-GL3Rn -bundle is simply a bundle
whose fibers are n-dimensional real vector spaces. Moreover, a
W-GLnRn -equivalence of W-GL]Rn -bundles is an equivalence that is linear
on fibers. Therefore, an n-dimensional real vector bundle is a bundle
whose fibers are n-dimensional real vector spaces, and which is locally
trivial in the natural vector sense: every point of the base has a
neighborhood U over which the given bundle is equivalent to
(U x]Rn fpr>j,U) via an equivalence which is linear on each fiber.

3. A bundle £ whose fibers are n-dimensional real vector

spaces is an n-dimensional real vector bundle (i.e., £ is locally
trivial in the previous vector sense) if and only if:
(i) K is topologically locally trivial;

(ii) the partial vector operations in tl£ , i.e., the maps

]R x tl £ tl K , (X ,x ) b Xx,
and
 325

{(x1,x2) € tl £ x tl £ | pr £ (x1) = pr £ (x2)} -> tl £, (Xl ,x2) ►> x + x2

are continuous.

The necessity of these conditions is obvious. Let us verify

their sufficiency. Let b^ £ b s £ . Fix an arbitrary basis, v^,...,v ,
of the vector space pr £-1 (bQ ) , a neighborhood U of bQ such that

£ [u is topologically trivial, and a trivialization h: (U x 3Rn ,pr ,U) +

?|u" Define a map, h': (U x nRn ,pr^ ,U) -> £, linear on fibers, by

the formula tl h 1 (b, ort. ) = tlh(b,pr9 ° tl h-1 (v .) ), where

-L ^ 1
P r 2 = ^Pr2 : U x lRn 3Rn ] . Now pick disjoint neighborhoods, K and N,

of the set pr^ ° tl h ^ ° tlh' (b^ x Sn ^ ) and of the point

_ 1
P r 2 ° tl h ° t l h 1 (b ,0) in ]R , respectively, and denote by V the

neighborhood of bQ consisting of all b e U such that tlh'(b x Sn~ 1)

c tl h (b x K) and tlh'(b,0) e tl h (b x N) . It is clear that

tlh'(b,0) £ tlh'(b x sn when b E V, and thus the map

ab h ' : (V xnRn fpr^,V) is nondegenerate on each fiber.

Consequently, we can apply Theorem 3.2.8 to abh' , taking F to be

TRn , regarded as a Top 3Rn-space ((V x lRn ,pr^ ,V) and £ |v are thought
of as Steenrod F-bundles; see 3.4.5). We conclude that ab h ' is an
equivalence in the topological sense, and since abh' is also linear
on fibers, the proof is complete.

Standard Fiber 0]Rn

4. A Steenrod bundle with standard fiber 01Rn is called

an n-dimensional Euclidean bundle.
Since a space with an 0!IRn -structure is an n-dimensional
Euclidean space, a W-OIRn -bundle is simply a bundle whose fibers are
n-dimensional Euclidean spaces. Moreover, it is clear that any
W-OIRn -equivalence of W-onRn -bundles is an equivalence which is an
orthogonal map on each fiber. Therefore, an n-dimensional Euclidean
bundle is a bundle whose fibers are n-dimensional Euclidean spaces and
which is locally trivial in the natural Euclidean sense: every point of
the base has a neighborhood U over which the bundle is equivalent to
(U x ]Rn ,pr >j,U ) via an equivalence which is an orthogonal map on each
fiber.
 326

5. A bundle £ whose fibers are n-dimensional Euclidean

spaces is an n-dimensional Euclidean bundle (i.e., £ is locally
trivial in the Euclidean sense) if and only if it satisfies (i),(ii)
of 3 and the following condition:
(iii) the function tl £ -> HR, which takes each vector into
its length, is continuous.

These conditions are obviously necessary. Let us verify

that they are also sufficient. By Theorem 3, (i) and (ii) imply that
every point of bs £ has a neighborhood U together with a triviali-
zation linear on fibers, h: (U x]Rn ,pr^,U) -+ Let v^ (b),..., v^ (b)

be the basis of the vector space pr £ ^ (b) , b £ U, resulting from the

standard orthogonalization of the basis tlh(b,ort^) , .. .,tl h(b,ort^) .

Now (iii) shows that the vectors v^ (b),...,v^(b) depend continuously

on b, and it is clear that the map linear on fibers,
h' : (U x Ш П ,рг1 ,U) £ |и / given by tl h* (Ь,ог^) = v ^ b ) (i = 1,...,n),
is a trivialization, orthogonal on each fiber, of the bundle £|u“

6. Since 0(n) с GL(n,]Rn ), every n-dimensional Euclidean

bundle £ determines a unique n-dimensional real vector bundle £ ',
through extension of the structure group. One may use Theorem 5 to
interpret the reduction of the structure group transforming £' into
£ as enriching the bundle £* with an additional structure: namely, a
Euclidean metric on each fiber, such that the corresponding length
function tl 5 * Ж is continuous. This additional structure is
termed a Euclidean metric on £ *.

Standard Fibers GL+lRn and БОЖП

7. A Steenrod bundle with standard fiber GL ЖП

(respectively, БОЖП ) is called an n-dimensional oriented vector
bundle (respectively, an n-dimensional oriented Euclidean bundle).

Since a space with GL+nRn -structure (S01Rn -structure) is

simply an n-dimensional oriented vector space (respectively, an
n-dimensional oriented Euclidean space), a W-GL+3Rn -bundle
(a W-SOHRn -bundle) is simply a bundle whose fibers are n-dimensional
oriented vector (respectively, Euclidean) spaces. It is also plain
that a W-GL+]Rn -equivalence of W-GL+lRn -bundles (a W-SonRn -equivalence
of W-SO]Rn -bundles) is simply an equivalence which is orientation
preserving and linear (respectively, orthogonal) on fibers.
 327

Consequently, an n-dimensional oriented vector bundle (Euclidean bundle)

is a bundle whose fibers are n-dimensional oriented vector spaces
(respectively, Euclidean spaces), and which is locally trivial in the
following sense: every point of the base has a neighborhood over which
the bundle has a trivialization that is orientation preserving and
linear (respectively, orthogonal) on fibers.

8. To obtain a version of Theorem 3 which is suitable for the

oriented case, note that the orientation existing on each fiber of an
n-dimensional oriented vector bundle £ maps the set of nondegenerate
n-frames of the given fiber into S^. Furthermore, the set of all
nondegenerate n-frames of the fibers of £ is the total space of the
associated bundle asso(£,V 1(n,n)) [where GL+ (n,3R) acts on V'(n,n)
as usual; see 2.3.1.3], and the orientations of the fibers combine to
define a map tl asso (£ ,V 1 (n, n) ) S^. The "oriented” version of
Theorem 3asserts that a bundle £ whose fibers are n-dimensional
oriented real vector spaces is an n-dimensional oriented vector bundle
if and only it satisfies conditions (i) and (ii) of 3 and the condition
(iv): the function tl asso (£ ,V 1 (n,n )) , defined by the orientations
of the fibers of £, is continuous.
Theorem 5 must be modified in a similar fashion. Namely,
given an n-dimensional oriented Euclidean bundle £, the orientation of
each fiber of £ maps the set of orthonormal n-frames of the given fiber
into S . Since the set of all orthonormal n-frames of all fibers of £
equals the total space of the associated bundle asso(£,V(n,n)), we
obtain a function tl asso (£,V (n,n)) -> S° . The "oriented” version of
Theorem 5asserts that a bundle £ whose fibers are n-dimensional
oriented Euclidean spaces is an n-dimensional oriented Euclidean bundle
if and only if it satisfies conditions (i),(ii) of 3, condition (iii)
of 5, and condition (v) : the function tl asso (£ ,V (n,n) ) S^, defined
by the orientations of the fibers of £, is continuous.

9. Since GL+ (n,nR) <= GL (n,1R) , every n-dimensional oriented

real vector bundle £ determines a unique n-dimensional real vector
bundle £', obtained from £ by extending the structure group. As it
follows from the discussion in 8, when we reduce the structure group
and produce £ from £', we are endowing the fibers of £' with
orientations which combine to define a continuous map
asso(C',V 1 (n,n)=GL(n,1R) ) + S°. This additional structure is termed an
orientation of the bundle £ 1.
Similarly, the inclusion SO(n) <= o(n) associates with every
n-dimensional oriented Euclidean bundle £ a unique n-dimensional
 328

Euclidean bundle £ ', obtained from £ by extension of the structure

group. Again, the reduction of the structure group which transforms
into K is seen to provide £1 with an orientation, i.e., as
orienting its fibers in such a manner that the corresponding function
tl asso (K 1,V (n,n) = 0 (n) ) -+ S° is continuous.
The real vector and Euclidean bundles possessing orientations
are reffered to as orientable. Since every orientation may be replaced
by the opposite one, as a result of multiplication by -1, every
orientable bundle has at least two orientations.

Standard Fibers GL(Cn and U(Cn

10. A Steenrod bundle with standard fiber GL(Cn (U(Cn )

is called an n-dimensional complex vector (respectively, Hermitian)
bundle.
Our discussion of real vector bundles in 2 and 3 carries over,
word-for-word, for complex vector bundles, and the same is true for the
discussion in 4, 5, and 6 of Euclidean bundles and the Hermitian bundles.
In particular, a W-GL(Cn-bundle (W-U(Cn-bundle) is simply a bundle whose
fibers are n-dimensional complex vector spaces (respectively, n-dimensio-
nal Hermitian spaces); a W-GL(Cn-equivalence (W-U(Cn-equivalence) of
W-GL(Cn-bundles (respectively, W-U(En-bundles) is simply an equivalence
linear (respectively, Hermitian) on fibers; a W-GL(Cn-bundle (W-U(Cn-bundle)
is locally trivial if and only if it is topologically locally trivial
and, in addition, the vector operations (respectively, the vector
operations and the length function) are continuous; the reduction of the
structure group which turns a given complex vector bundle, into a
Hermitian bundle £, may be interpreted as endowing £' with a
Hermitian metric, i.e., as supplying a Hermitian metric (inner product)
on each fiber of £', in such a manner that the resulting length
function is continuous on tl £ 1 .

11. Given an arbitrary n-dimensional complex vector bundle

£, we may construct an n-dimensional complex vector bundle, conj £ ,
by replacing each marked homeomorphism a with the composition
a ° conj, where conj: (E (Cn is the usual complex conjugation;
conj £ is called the bundle conjugate to £.
This construction carries over to Hermitian bundles and
produces again Hermitian bundles, i.e., to every n-dimensional Hermitian
bundle £ there corresponds the conjugate Hermitian bundle conj £ .

12. The extension of structure group defined by the inclusion

 329

GL(n,(C) <= GL (2n,JR) turns n-dimensional complex vector bundles into

2n-dimensional real vector bundles. Similarly, the extension defined by
the inclusion U(n) c 0(2n) turns n-dimensional Hermitian bundles into
2n-dimensional Euclidean bundles. In both cases we call the extension
of the structure group realification, and we denote by the bundle
obtained from £ by realification.
We recall that the additional structure which turns a given
2n-dimensional real vector space into an n-dimensional complex space
may be described as a linear transformation whose square equals -id
(a "multiplication by -i"). Similarly, the additional structure which
turns a given 2n-dimensional Euclidean space into an n-dimensional
Hermitian space may be described as an orthogonal transformation whose
square equals -id. Accordingly, the additional structure which
distinguishes between the n-dimensional complex bundle £ and
2n
may be regarded as a GLIR -equivalence, I: ^ 3R£ , such that
2
I = -(id!R£) (the minus sign is defined fiberwise), and that which
distinguishes between the n-dimensional Hermitian bundle £ and nR£ -
- as an 03R2n -equivalence, I: 3R£ nR£ , such that I2 = -id(3R£).
Moreover, we may think of the reduction of structure group which turns
3R£ into £ as endowing IRE, with one of the last two equivalences.
Obviously, 3R(conj £) = IR£ for any complex vector or
Hermitian bundle £, and the shift £ *+ conj £ may be described in
the previous language as the shift I -I.

13. Since we have not only GL(n,(C) <= GL(2n,lR) and

U (n) c: 0(2n), but also GL(n,CE) <= GL+ (2n,3R) and U(n) c S0(2n), the
realification of a given n-dimensional complex vector or Hermitian
bundle may be effected in two steps: one may first extend GL(n,(E) to
GL+ (2n,JR) (U(n) to S0(2n)), and then extend GL+ (2n,HR)to GL(2n,3R)
(respectively, S0(2n) to 0(2n)). Therefore, every complex vector or
Hermitian bundle £ provides IR£ with a canonical orientation.

Maps

14. A map of a real or complex vector bundle into another

is said to be linear if it is linear on fibers. Those linear maps which
are nondegenerate (injective) on fibers are called linear monomorphisms.
A map between Euclidean bundles which is both isometric and linear on
fibers is called an orthogonal monomorphism. A map between Hermitian
bundles which is both isometric and linear on fibers is called a unitary
monomorph ism.
 330

Note that a linear monomorphism between n-dimensional vector

bundles is nothing else but a GLJRn -map in the real case and a GLEn-map
in the complex case. Similarly, an orthogonal (unitary) monomorphism
between two n-dimensional Euclidean (respectively, Hermitian) bundles is
simply an 03Rn- map (respectively, a U(Cn-map) .

Vector Fields

15. Sections of vector bundles are called vector fields.

This term is applied equally to the real and complex vector bundles, to
the Euclidean and Hermitian bundles, and to the oriented bundles of both
types.
A sequence of k vector fields is termed a field of k-frames.
A point of the base where the corresponding frame is degenerate is a
singularity of the given field. A field of k-frames with no singula­
rities in an n-dimensional vector bundle may be regarded as a section
of the associated bundle with fiber nRV'(n,k) or (CV'(n,k). Similarly,
a field of orthonormal k-frames with no singularities in an n-dimensional
Euclidean or Hermitian bundle may be thought of as a section of the
associated bundle with fiber 3RV(n,k) or (CV(n,k) [in all cases, the
structure group acts as usual].
For every vector bundle there is the zero vector field, which
takes each point of the base into the zero vector of the fiber over the
given point. As we shall see, there are vector bundles having no
sections without zeros. An n-dimensional real (complex) vector bundle
£ admits a field of n-frames with no singularities if and only if ^
is GL3Rn-trivial (respectively, GLCCn-trivial) ; conversely, any such
field yields a GL3Rn- trivialization (respectively, a GL(Cn-trivialization)
of Similarly, for an n-dimensional Euclidean (Hermitian) bundle £,
giving a field of orthonormal n-frames is equivalent to giving an
03Rn- trivialization (respectively, a U(Cn-trivialization) of
Moreover, given an n-dimensional oriented real vector (Euclidean) bundle
£, a field of n-frames without singularities (respectively, a field of
orthonormal n-frames) such that the orientations of the fibers are
positive on the frames of the field, yields a GL+3Rn -trivialization
(respectively, an S01Rn -trivialization) of
 331

2. Constructions

1. In this subsection we discuss a number of constructions

which shall be applied afterwards to vector, Euclidean, and Hermitian
bundles, and which were not covered by the general theory of Steenrod
bundles from § 3.

Subbundles

2. Let E, be an n-dimensional real vector bundle, and let

g be a section of the bundle asso(£, 3RG (n,k )) weakly associated with
£ [here GL(n,3R) acts on lRG(n,k) in the usual way], i.e., a
continuous function which takes each point b € bs £ into a k-dimensio-
nal subspace of the fiber pr £ ^ (b) . Denote by T the union of all
these k-dimensional subspaces, and let be the bundle
(T,pr £ |T ,bs o . Since the fibers of are the subspaces singled out
k
by g, 5 L is a GL3R -bundle, and it is clear that £ is locally
k
GL3R -trivial. [In fact, let E, be the standard trivial bundle
(B x 3Rn ,pr^ ,B ) . Given b^ 6 B , choose a linear isomorphism
k
t : 3R + g(bg). Then for a small enough neighborhood U of b^, the

restriction of EI to U admits even a canonical GL]Rn -trivialization,

|g
h: (U x IR^ ,pr i fU ) ->■ (C jg ) j0 ^ given by tlh(b,v) = prb (b,£(v)); here

pr^ denotes the projection of the fiber b x ]Rn onto its subspace
g(b)]. Therefore, is a k-dimensional real vector bundle. We call
it the subbundle of t associated with g.

3. The subbundles of Euclidean, complex vector, or Hermitian

bundles are defined similarly. In the Euclidean case, 3RG(n,k) is
regarded as an 0(n)-space, g remains a sectionof asso (E,,3RG (n ,k)) ,
and the resulting subbundle, E, I , turns out to be Euclidean (its Ideal
k ^
03R -triviality is a consequence of its local triviality linear on
fibers, established in 2, and of the continuity of the length; see 1.5).
The discussion in 2 carries over, word-for word, to the complex case:
all we have to do is to replace IR by(C [in particular, the
GL (n,3R) -space ]RG(n,k)must be replaced by the GL(n ,(C)-space CCG(n,k)].
The resulting subbundle, £| , is a complex vector bundle. Finally, in
iy
the Hermitian case, (EG(n,k) is considered as a U(n)-space and the sub-
k
bundle £I is Hermitian (£ islocally U(C -trivial because it is
9 19
 332

locally trivial, linearly on fibers, and the length function is

continuous; see 1.10).
4. It is clear that the inclusion
Ç| K is a linear mono-
IÇT
morphism in both vector cases, an orthogonal monomorphism in the
Euclidean case, and a unitary monomorphism in the Hermitian case.
Conversely, to every linear, orthogonal, or unitary monomorphism,
f: £; ■+ Ç, with bs^ = bs C and bs f = id, there corresponds a
subbundle of Ç, namely the subbundle associated with the section
b tl f (pr (b)) of asso (C ,3RG (n ,dim £ ^ ) ) or asso (Ç ,CCG (n ,dim £ ^ )) .
This subbundle is the image of the monomorphism f, denoted Im f .
By Theorem 3.2.8, abf : 4 Im f is a GL]Rn- , GL(Cn-, OIRn- , or
U(En-equivalence, depending on whether we are in the real vector, complex
vector, Euclidean, or Hermitian case.
i
We note that the correcting map, corr f : (bs f) "Ç ,
which is a linear, orthogonal, or unitary monomorphism together with f
(see 3.2.4), always satisfies bs corr f = id. Therefore, Im corr f is
meaningful for any linear, orthogonal, or unitary monomorphism f.

Orthogonal Complements and Quotient Bundles

5. If Ç is an n-dimensional Euclidean (Hermitian) bundle,

then to each section g of asso (Ç,IRG (n,k) ) (respectively,
of asso (Ç ,(EG (n,k) )) there corresponds the section g“1
“ orthogonal to
g: g^ is the section of asso (Ç ,HRG (n,n-k) ) (respectively, of
asso (Ç ,(CG (n, n-k) )) which takes each b G bs Ç into the orthogonal
complement of the subspace g(b) in the fiber pr C"1 (b) . Therefore,
to every k-dimensional subbundle n = Ç |g of the n-dimensional
Euclidean or Hermitian bundle Ç there corresponds an (n-k)-dimensional
subbundle, ÇI called the orthogonal complement of the subbundle
g 1
n, and denoted n .

6. A subbundle of a vector bundle has no orthogonal

complement, but a corresponding quotient bundle is well defined. Namely,
let n be a k-dimensional subbundle of the n-dimensional real or complex
vector bundle Ç. Consider the bundle (T,fact pr £,bs £ ) , where T is
the quotient space of tl £ by its partition into the sets x+pr£~1 (b),
with x G pr Ç 1 (b) . The fibers of this bundle are the quotient spaces
_ <j -i ,

pr c (b)/prn (b) , and so it is a GL3Rn_ - or a GL(En -bundle.

(T,fact pr Ç ,bs O is called the quotient bundle of Ç b^ n, denoted
by Ç/n-
 333

This construction and the previous one are related: indeed,

if we apply the quotient bundle construction to a Euclidean or Hermitian
bundle, we obtain the result of the orthogonal complement construction.
— 1 — 1
More precisely, in the Euclidean case each quotient pr £ (b)/prn (b)
is an (n-k)-dimensional Euclidean space; hence, £/n is an 01Rn-k-bundle,
and the map h: n ->■ £/n, given by tlh(x) = pr(x), where pr =
= [pr: tl £ -> tl(£/n)], is an 03Rn -equivalence. Similarly, in the
-1 -1
Hermitian case, pr£ (b)/prn (b) are (n-k)-dimensional Hermitian
n —k
spaces, and hence £/n is a U(E -bundle, and the same h is a
n—k
U(E -equivalence. Therefore, the quotient bundle of a Euclidean
(Hermitian) bundle £ by a subbundle n is a Euclidean (respectively,
Hermitian) bundle of dimension dim £ - dim n •
From this it is readily seen that the quotient of a vector
bundle £ by a subbundle n is a vector bundle of dimension
dim £ - dim n ♦ All we have to check is that £/n is locally trivial
linearly on fibers whenever £ is a standard trivial bundle. But this
is immediate from the previous discussion if we note that such a £ may
be assumed to be Euclidean in the real case, and Hermitian in the complex
case .

7.
For real vector or Euclidean bundles, the orientation of
-1 -1 -1 -1
each of the three spaces, pr £ (b) , pr n (b) , pr £ (b)/pr n (b) ,
is uniquely determined by the orientations of the other two (see
3.1.3.10). Consequently, the orientability of two out of the three
bundles, £, r\, £/n/ implies the orientability of the third, and
given orientations of two of them canonically determine the orientation
of the third.

Sums

8. The vector bundle £

is said to decompose into the
-1
(direct) sum of its subbundles ^ and if e^ch fiber pr £ (b)
-1 -1
is the direct sum of its subspaces pr £ (b) and pr ^ (b) • A
Euclidean or Hermitian bundle £ decomposes into the (orthogonal) sum
of its subbundles ^ and if each fiber pr £ (b) is the ortho-
-1 -1
gonal sum of its subspaces pr ^ (b) and pr (b) .

In the Euclidean and Hermitian cases, every subjaundle £ of

the given bundle £ splits £into the sum of its subbundles £ 1 and
We show in Subsection 4 that given any vector bundle K with cel­
lular base and any subbundle of £, there is a subbundle °f
 334

£ such that £ decomposes into the direct sum of £^ and ^2 • see

4.2.
If the vector bundle £ decomposes into the sum of its sub­
bundles E, and £ , then the quotient bundle 5 / ^ is canonically
dim ^ dim E,
GLIR - or GLCC -equivalent to ^2 : this canonical equivalence
h: £>2 5/^ is given by tlh(x) = pr (x) , where pr =
= [pr: tl E, tlf?/^)] . Such an equivalence exists also in the
Euclidean and Hermitian cases, when 0 or U replaces GL (in fact,
we have already established this in 6).
9. We now introduce a construction which reverses the above
process and, in particular, allows us to recover a bundle which
decomposes into a sum of subbundles from its summands.
Let ^ and ¿^2 ke real vector bundles of dimensions n^ and
n2 r and with common base. Then x ^2 is a Steenrod bundle with
base bs ^ x bs ^2 and structure group GL(n^,lR) x GL(n2 ,lR) . The

bundle diag’ (£^ x f where diag: bs bs ^ x bs ^ , has the

same structure group (and base b s £ 1) . Extending this group to
i '
GL(n^+n2 ,IR) , we turn diag’(^ x £2 > into an (n^ +n2 )-dimensional real
vector bundle, called the (direct) sum of the bundles 5^ and ^2 , and
denoted ^ © 5 2* In the Euclidean, complex vector, and Hermitian cases,
the definition and notation of a sum of two bundles are the same [in the
Euclidean case, O(n^) x 0(n2) is extended to 0(n^+n2 ), in the complex
vector case, GL(n^,I) x GL(n2 ,CE) is extended to GL(n^+n2 ,I ), and in
the Hermitian case, U(n^) x U(n2 ) is extended to U(n^+n2); moreover,
a sum of Euclidean, complex vector, or Hermitian bundles is again a
Euclidean, complex vector, or Hermitian bundle, respectively]. In all
cases, b s i ^ ® ?2) = bs ^ (= b s £ 2) and pr(£ © £2 )- 1 (b) =
-1 -1
= pr ^ (b) ® pr £2 (b); the last sum is orthogonal in the Euclidean and
Hermitian situations. These equalities define linear, orthogonal, or
unitary monomorphisms, ^ © ?2 and ^2 ^1 ® ^2' which act as
the identity on the base. These monomorphisms identify £^ and £2
with subbundles of ^ © ?2 and split © C2 into the sum of these
subbundles.
Note that the same identifications allow us to take the
quotients (£^ © and ^-] ® ^2 ^^2' and using the canonical
equivalences defined in 8, we may actually write © £2 )/£^ =
and (£1 © £2)/£;2 = £ . In particular, in the real vector and Euclidean
cases, the orientability of two out of the three bundles, £ , £ , and
® K2 > implies the orientability of the third, and the orientations
of any two of them define canonically an orientation of the third.
 335

Let us add that £ © Ç is always orientable and has a

canonical orientation, for any real vector or Euclidean bundle £.
This canonical orientation is determined on each fiber by arbitrary
orientations of its summands, provided that we take identical
orientations for both summands.

10. The sum of a real vector or Euclidean bundle Ç and the

1 1
one-dimensional trivial GL]R - or OÜR -bundle(bs £ x ]R,pr^ ,bs Ç ) is
called the suspension of Ç,and is denoted by su Ç . Similarly, for
any complex vector or Hermitian bundle Ç, thesuspension of
Ç, suÇ ,
1
is the sum of £ and the one-dimensional standard trivial GLŒ - or
UtE^-bundle (bs £ x (E,pr^,bsÇ) .

Two real vector bundles, Ç^ and with b s = bsÇ j

are said to be stably equivalent if there exist k and k^ such that

dimÇ. + k. = d i m £ n + k n and the bundles k 1su £ .and suk2 are

1 1 2 2 1 "2
dim Ç „+k„
1 1
GLIR -equivalent. Stable equivalence of Euclidean, complex
vector, and Hermitian bundles is similarly defined (replacing
dim Ç +k dim Ç +k d i m ^ 1 +k dim Ç +k
GLIR by 0 1 , GLiE , and UCD ,
respectively). A bundle which is stably equivalent to a standard
trivial bundle is called stably trivial.
Finally, we note that for real vector or Euclidean bundles,
the orientability of one of thebundles £ and su £ implies the
orientability of the other, and any orientation of one of them
canonically defines an orientation of the other; see 9.

Complexification

11. Given an n-dimensional real vector bundle £, consider

the map I: £ 0 £ + Ç © £, given by tll(x,y) = (-y,x) [x and y
2n
sit in the same fiber of Ç]. Obviously, I is a GLÜR -equivalence
2
satisfying I = -id. Therefore, I turns £ © Ç into an n-dimensional
complex vector bundle (see 1.12), called the complexification of Ç,
and denoted (E£.
If Ç is an n-dimensional Euclidean bundle, the same
construction turns Ç © £ into an n-dimensional Hermitian bundle, which
is also called the complexification of £ and is denoted ŒÇ.

The operation £ (EÇ is called complexification in both

cases. Note that IRŒÇ = Ç © £ in both cases.
 336

12. Each of the bundles TRVF and £ © £, appearing in the

last equality,carries a canonical orientation; see 1.13 and 9. It may
be shown that these orientations coincide when n = 0,1 mod 4 , and are
opposite when n = 2,3 mod 4 .
_ 1
Indeed, pick an arbitrary fiber pr £ (b) of £ and an
arbitrary basis (respectively, orthonormal basis) v ,. .. ,v in
-1 -1 1
pr £ (b) . The canonical orientation of the fiber pr F, (b) * pr £ (b)
of ME takes the value +1 on the basis (v ^ ,0), (0,v^),..., (v r ,0) ,
(0,v ) of this fiber, and the same holds for the canonical orientation
n —1 -1
of the fiber pr £ (b) x pr £ (b) of £ © £ and its basis (v^O),
.. (vn ,0), (0,v i),..., (0,vn) . It remains to note that, in order to
shift from one basis to the other, we have to perform n(n-1)/2
permutations of adjacent vectors, and that this number is even for
n = 0,1 mod 4 , and odd for n = 2 ,3 mod 4 .

13. The map conj: (C£ + conj(C£, given by tlconj(x,y) =

= (x,-y) , is a GL(£n-equivalence (U(En-equi valence) for any n-dimensional
realvector (respectively, Euclidean) bundle £.

Indeed, the equivalences I ,I2 :£ © £ - ► £ ; © £ , which turn

£ © £ into (C£ and conj CC£ , are given by tl 1^ (x,y) = (-y,x) and
I^ = anc^ hence I^ ° conj = conj © i^ .

14 . The map K : £ © conj £ (EIR £ , given by tl K (x ,y ) =

1 1
= (x+y) (tl I (y) - tl I (x) )) , where I is the equivalence which

turns 1RF, into £, is a GL(Cn-equivalence (U(Cn-equivalence) for any

n-dimensional complex vector (respectively, Hermitian) bundle.

The equivalences I^I^: 3R£ © JRK -+ 3R? © , which turn

TRF, 0]R£ into F, © conj £ and (C]R£ , are given by tl 1^ (x,y) =
(tl I (x) ,-tl I(y)) and tl X2 (x,y) = (-y,x), and hence 1^ ° K = K ° I .

3. The Classical Universal Vector Bundles

1. The main value of the construction in Subsection 4.3 is

that it establishes the existence of a universal G-bundle for a
completely arbitrary topological group G. However, for the groups
GL (n,IR) , GL+ (n,IR) , 0 (n) , SO(n) , GL (n ,(C) , and U (n) , there are
more convenient, classical constructions, which are described in the
present subsection.
 337

The Grassman Spaces

2. Set

G(°°,n) = lim (G (m, n) ,in : G(m,n) -> G (m+1 #n ) ) #

G+ (°°,n) = lim(G+ (m,n) ,in: G + (m,n) + G+ (m+1,n)),

and
(CG (°°,n ) = lim (EG (m,n) ,in: (CG(m,n) + (CG(m+1,n)).

G(00/n) consists of all n-dimensional planes of nR°° passing

through 0, and is called the n-th (real) Grassman space. Similarly,
the n-th upper Grassman space, G (°°,n) , consists of all oriented
n-dimensional planes in ]R°° passing through 0. Finally, the n-th
complex Grassman space, (CG(°°,n), consists of all n-dimensional planes
oo
in (C passing through 0.

The canonical maps: G (m,n) G(m,n), (CG(m,n) + G(2m,2n),

CCG(m,n) G+ (2m,2n), G(m,n) G(m+q,n+q), G+ (m,n) G+ (m+q,n+q), and
CCG(m,n) (CG(m+q,n+q) (see 3.2.2.3 and 3.2.2.7) define for any n the
following maps:

G+ (°°,n) + G (°°,n ) , (CG (°°,n) + G (°°,2n) , (DG (°°,n) -+ G+ (~,2n),

G (°°,n) + G (°°,n+q ), G+ (°°,n ) + G+ (<»,n+q),

and
£CG(°°,n) -+ (CG (°°,n +q) .

The first of these maps (as the canonical map G+ (m,n) -> G(m,n)
with m < 00) is a two-sheeted covering projection; the second is the
composition of the first and the third; and all these maps, except the
first, are embeddings.

3. The Grassman spaces possess natural cellular decompositions

which we shall presently describe.
First consider G(°°,n). Let denote the set of all
sequences of integers, w = {co(1 ),..., to(n) } , with 0 £ o>(1 ) £ ... £ w(n),
and let us agree to add the term oo(0) = 0 to each sequence w C
Further, let e(to) denote the subset of G(°°,n) consisting of those
oo
n-planes y in ]R (passing through 0) such that

dim(y D HRm ) = max{s|to(s) + s £ ml,

for all m. We show that the sets e(io), uj £ , yield a cellular

decomposition of G(°°,n), with dime(oj) = d(w), where d(u) =
 338

= 03(1) + ... + 03 (n ) •
We have to produce a characteristic map, cha^^: + G(°°,n).
Fix u) and, for points u,v £ sw(n)+n-1 such fchat u + v ^ o , denote
by r(u,v) e SO(w (n)+n) the orthogonal transformation which takes u
into v and keeps fixed all the vectors of n which are ortho­
gonal to u and v. Further, let denote the co(i) -dimensional

hemisphere consisting of all points ^x i '''''xw (i )+i^ e s<' ^ ^ "* such

that x ... . = 0 for j = 1,...,i+1, and x , 5 0. Consider the
Ol3(J)+3 OJllJ+l
map 6: H„ * ... x H -> G(«>,n) which takes each sequence (u,,,...,u )
L 1 n i n
into the plane spanned by the n-frame:

V [ r ( O r t 0 ) ( 1 ) + 1 ' U 1 ) ] ( U2 ) ...........
(1 )
[r (Ort /-V.-/U ) o... O r(ort , . 17U ! )](u )
U)(1)+1 1 co(n-1 )+n-1 n-1 n

(which is obviously orthonormal).Then $ is continuous and maps

int(H„ x ... x H ) into e(u)), and 3 (H. x ... x h ) into a union of
I n I n
sets e(o)') with d(o)1) < d(co). Moreover, its compression
ab <(> : int(H^ x ... x e(o)) is a homeomorphism: its inverse takes
each plane y £ e(o)) into the sequence (u^ ,...,u^) , with

ui = V u 2 “ [r,ort» m * i ' ui)l~1|v2)....

“n = Ir<°rt«)(1).1'U 1) ° “ r lortu(n-1) tn-1 ,un-111"1 lvn ’ '

where v i'‘‘#'vn an orthogonal basis of y, selected in such a way

that v. sits in the hemisphere x ... . > 0 of g^(i)+i-1 (the
1 r 0)(l) +1
vectors v^,...,Vn areuniquelydetermined by these requirements).
Therefore, one maytake cha^^j to be the composition

Dd(u)) ^ D^(1) x _ x Doi(n) + x x Hr I G(«,n),

where the left map is the inverse of the homeomorphism indicated in

1.2.6.9, and the middle map is the product of the homeomorphisms
d “ *1* H. given by (x. ,...,x ) ^
-L I CO \ 1 f

(x1 '• ••,XU)(1 ) ,0,XU)(1)+1 ' •• •,X03(2) ,0'** •'°'xw(i-1 )+1 '•• •'

x (1 - x V /2>
Xw(i), n ¿1 x j' '•

The cellular decomposition of G+ (°°,n) has twice as many

cells as that of G(°°,n). Namely, over each cell e(oo) sit two cells,
e+ (o)) and e_(oj), of G+ (°°,n) , which are homeomorphically mapped
 339

onto e (u)) by the projection G+ (°°,n) G(°°,n): e+ (co) is made up of

planes oriented in such a way that the orientation is positive on the
basis v i'‘**'vn described above, while e_(co) is made up of the same
planes, but with the opposite orientations. The characteristic maps
for e (w) and e (co) are constructed in the same manner as cha , w
+ e (00 )
but now the plane spanned by the frame (1) is oriented [its orientation
is positive (negative) on (1) fore+ (co) (respectively, for e_(co))].

The cellular decomposition of (CG(°°,n) is given by the cells

CCe(w), co € £3 ,which are defined precisely as the efuO's if one
00 m 00 m
replaces 3R and 3R by (C and (C ; dim(Ce(co) = 2d(co), and cha , .
ILe (Cl)/
is the exact complex-Hermitian analog of the map cha^^j.

4. The above cellular decompositions contain only a finite

number of cells of a given dimension, and hence satisfy property (C).
Since each of the manifolds G(m,n), G (m,n), and (CG(m,n) is covered
by a finite number of cells, and since these manifolds constitute
fundamental covers of the spaces G(°°,n), G (°°,n) , and (CG(°°,n),our
cellular decompositions satisfy also property (W). Finally, from
Theorem 1.2.4.6 it follows that G(°°,n), G (°°,n), and (CG(°°,n) are
normal. Therefore, the above cellular decompositions turn G(°o,n),
G+ (°°,n), and CCG(°°,n) into cellular spaces.

It is clear that G(m,n), G + (m,n), and CCG(m,n) are sub­

spaces of G (°°,n ) , G+ (°°,n), and (CG(«,n) in the cellular sense, and
so they are cellular spaces too.
For n = 1, G(m,n), (CG (m, n) ,G(«,n), and (CG(°°,n) are
rn 1 rri— 1 00 00
simplyIRP , (CP , 3RP , and (EP , respectively, andthe above
cellulardecompositions are identical with those described in 2.1.3.4
and 2.1.3.5.
5. Note that the canonical embeddings

G(«>,n) G (°°,n+q) , G+ (°°,n) -> G+ (°°,n+q), (CG(»,n) -»■ GG(«>,n+q)

(see 2) are cellular. The image of the first contains skenG (°°,n+q) ,
the image of the second - ske G+ (cc,n+q), and the image of the third -

ske2n+1tEG (oo'n +c2) •

The inclusions G(m,n) => skem_nG (°°,n) , G+ (m,n) = skem _nG+ (°°,n),
and (EG (m,n) ske2m-2n+110 are ec3ually evident.
 340

The Grassman Bundles

6. Let 0 <r n £ m £ 00 and n < 00. We let T(m,n), T + (m,n),

and (ET(m,n), denote those subsets of the respective products
G (m,n) x ® m , G (m,n) x iRm , and (EG(m,n) x (Em , consiting of all pairs
(y,x) such that x £ y. Consider the bundles (T(m,n),pr ,G (m,n )),
(T (m,n) ,pr,G+ (m,n)), and (£ET (m, n ) ,pr ,(EG (m,n )) , where pr (y,x) = y.
The fibers of the first (second; third) bundle are Euclidean spaces
(respectively, oriented Euclidean spaces; Hermitian spaces). Moreover,
the first (second; third) bundle is locally W-OJRn -trivial (respectively,
locally W-SOJRn -trivial; locally W-U(En-trivial) , and hence it is a
Euclidean (respectively, oriented Euclidean; Hermitian) bundle. We
denote these bundles by Gra(m,0(n)), Gra(m,SO(n)), and Gra(m,U(n)).
In addition, by extending the respective structure groups, 0(n), SO(n),
and U (n) , to GL (n,1R) , GL+ (n,JR) , and GL(n,£E), we obtain bundles
denoted by Gra (m, GL (n ,JR) ) , Gra (m,GL+ (n,JR) ) , and Gra (m, GL (n,(E)) .
The bundles of these six series are called Grassman bundles. For m = °°,
we use the simpler notations GraO(n), Gra SO (n) , GraU(n) ,
Gra GL (n,JR) , Gra GL+ (n,JR) , and Gra GL (n ,(E) .

Note that for m <°°, Gra(m,0(n)) is nothing else but the

subbundle of the standard trivial bundle (G(m,n) x jRm ,pr ,G(m,n))
(viewed as a Euclidean bundle) associated with the diagonal section,
Y ** (Y fy) r of the bundle (G(m,n) x G(m,n),pr^,G(m,n)). The same
holds, with obvious modifications, for the remaining five series.

7. The bundles Gra G with G = GL(n,JR) ,GL (n,JR) ,0(n),

SO(n) ,GL (n,(E) ,U (n) are universal.

The proofs for the different groups G differ only in some

obvious details, and are all very similar to theproof of Theorem 4.3.4.
We shall treat here only the group GL(n,JR) . According to 4.2.2, it
suffices to show that given any n-dimensional real vector bundle £
with cellular base and any subspace A c bs £ , every GLJRn -map
g: £|a ->■ Gra GL (n,JR) extends to a GLJRn -map £ Gra GL (n ,JR) .

Assume first that b s £ = Dr+1 (for some r) and A = Sr .

In this case £ is GLJRn -trivial, and so we may actually assume that
£ is the standard trivial bundle (Dr+1 x jRn ,pr^,Dr+1). The desired
extension of g, f : £ ->■ GraGL(n,JR), can be described explicitly:
let g^ be the composite map

Sr x 3Rn ~ tl ^ > tl Gra GL (n ,JR) — — > G (°°,n) x JR°° --- -— ► hr00

 341

and define f 1 : Dr+1 x ]Rn »■ ir°° by f 1 (ty, (x1 , . . . ,x ) ) =

= tg (y, (x ,...,x )) + (1-t2)1/2(0,...,0,x.,...,x ,0,...),

1 1 n V T— v I n
m
where y E S , t E I, and m is the smallest number s such that

]RS => g 1 (Sr x s n 1 ) ; finally, set tl f (y,x) = (f^ty x ]Rn ) ,f/)(y,x) ) .

The general case reduces to this special situation. Assume

that the cellular space bs £ is rigged and that we already have a
GL]Rn -map f : ^|AlJske bs £ G r a G L (n,:iR) which extends g. The above
I r 5
argument shows that for every cell e E cell „bs £ \ cell „A the
n r+1 s r+1
GLIR -map

r *
g = f oad[abcha : S -*• A U ske bs g] : cha *f Gra GL (n, TR)
^e r e r e
S
n *
extends to a GLIR -map h ^ : °ha^£ -+ Gra GL(n,]R) , and it is clear that
tlh is constant on the elements of the partition zerftlad cha ).
^ r+1 r+1 e
Applying Theorem 3.2.6 (with B = D and p = [ab cha^ : D + Cle] ),

we conclude that h defines a GL!IRn -map -*■ Gra GL (n,3R) , which

e ^ |Cle
we denote by f^. Now note that for any two cells, e^,e^ E cell^ + ^bs

\ cell .A, tl f and tl f agree over Cle. fl Cl e0 , and that

r +1 e^ e -2 1 2
for any cell e E cell .bs £ \ cell .A, tl f and tl f agree over
i r+1 r+1 e r
Cl e fl (A U ske^bs £) . From this compatibility it follows that the maps
f and f^, e E cell^^bs £ \ cell^^A, combine todefine a GL3Rn-map

f • f| . . , >. -> Gra GL (n,IR) which extends f ; see 3.2.7.

r+1 AUske ,bs £ ' r
I r+1
Therefore, using induction, we can produce a sequence of GL]Rn -maps,

ifs : 5 U s k e b s C - O r a G L I n , s u c h that £.,=9 and fs

I s n
extends f ,s ^ 0. Finally, the maps f define a GL3R - map
s— 1 s
£ Gra GL (n,IR) extending g.

8. The bundles Gra (m, GL (n,IR) ) , Gra (m, GL+ (n,IR) ) ,

Gra(m,O(n)), and Gra(m,SO(n)) are (m-n)-universal. The bundles
Gra (m,GL(n,(C) ) and Gra(m,U(n) ) are (2m-2n+1)-universal.

This is a corollary of 7 (see 5 and 4.2.7).

 342

Associated Principal Bundles

9. When m < °°, the total spaces of the principal bundles

associated with the Grassman bundles

Gra (m, GL (n ,1R) ) , Gra (m,GL+ (n ,IR) ) , Gra (m, GL (n,(C) ) ,

and
Gra(m,0(n)), Gra(m,SO(n)), Gra(m,U(n)),

are obviously V' (m,n) , V 1(m,n) , (CV* (m,n) , and V(m,n) , V(m,n) ,
(CV(m,n). The corresponding projections are the maps described in
3.2. 2.3 and 3.2.2.7:

V* (m,n) G (m,n ) , V' (m,n) -> G+ (m,n), CCVf(m,n) (CG(m/n) (2)

and
V(m,n) + G (m,n) , V(m,n) + G+ (m,n), (CV(m,n) (CG(m,n). (3)

The same is true for m = m, if V 1 (°°,n) , (CV1 (°°,n) , V(°°,n),

and CDV (00,n ) are understood as lim (V1(m, n ) ,in) , lim ((CV* (m,n ) ,in) ,
lim (V (m,n) ,in), and lim (CCV (m, n ) ,in) , and the projections (2),(3)
with m = 00 as the limits of the projections (21,(3), m < 00. V't“ ^ ) ,
(CV1 (°°,n) , V(«>,n), and CCV(°°,n) are called Stiefel spaces.
It is clear that for m < 00 the canonical right actions of
the structure groups on the above total spaces (see 3.2.10) are exactly
the right actions described in 2.3.12 and 2.3.13, while for m = °o
they are the limits of the latter.

The Bundles
asso (Gra 0(1) ,0(1) ) and as so (Gra U (1 ) ,U (1 ))

10. The principal bundle associated with Gra 0(1) is

0(1)-isomorphic to Mi0(1) . The principal bundle associated with
Gra U(1) is U (1 )-isomorphic to Mi U (1 ) .

PROOF. It suffices to find an 0(1)-homeomorphism

tl Mi 0(1 ) tl asso (Gra 0 (1 ), 0 (1 )) , when we regard tl Mi 0(1 ) and
OO
tl asso (Gra 0(1),0(1)) = V (°°,1) [= S ] as right 0(1) -spaces ; similarly ,
viewing tlMiU(1) and tl asso (Gra U (1 ), U (1 ) ) = (CV (00,1 ) [= S°°] as
right U (1)-spaces, we need only exhibit a U (1)-homeomorphism
tl Mi U (1 ) + tl asso (Gra U (1 ) ,U (1 )) ; see 3.1.9 and 3.2.10. In both cases
such a homeomorphism is given by the formula

(pr (g± 't±) }”_ 1 - .

 343

The meaning of the left-hand side was explained in 4.3.2, while in the
right-hand side the elements g. of 0(1) or U(1) are thought of as
numbers (the following inclusions are used: V^l) cnR°°, (CV(°°,1) c (E°°,
0(1) = S° c ]R, and U (1) = S1 c (C) .

4. The Most Important

Reductions of the Structure Group

1. The use of Grassman bundles enables us to apply the scheme

presented in Subsection 4.4 to the problems raised in Subsection 1
concerning reductions of the structure group. This is the subject of
the present subsection.
Recall that the reductions corresponding to the inclusions

0 (n) c= GL (n,3R) , SO (n) c GL+ (n,HR) , U(n) c GL(n,<C), (4)

are equivalent to the introduction of a Euclidean or Hermitian metric,

while the reductions resulting from the inclusions

GL (n,3R) <z GL(n,3R) , S0(n) c 0(n), (5)

are equivalent to the introduction of an orientation. Finally, the

reductions resulting from the inclusions

GL(n,<C) c GL (2n,]R) , U(n) c 0(2n), (6)

mean the introduction of a complex structure.

For each of the inclusions (4), (5), and (6), we shall exhibit
a canonical classifying map, and then list the most obvious consequences
of these constructions. Moreover, we shall carry out the same program
for the inclusions

GL (n-s ,JR) c GL (n ,]R) , GL (n-s,3R) <= GL (n,HR) ,

(7)
GL(n-s,CC) c GL (n,(E) ,
and
0 (n-s) c= 0 (n) , SO (n-s) c= SO (n) , U (n-s)c= U(n). (8)

The reductions of the structure group corresponding to the six inclusions

(7) and (8) may be interpreted as the representation of the given
n-dimensional bundle as the s-fold suspension of an (n-s)dimensional
bundle.
2. The outlined program is simple to carry out for inclusions
 344

(4). Indeed,the bundles Gra GL (n,IR) and GraO(n) have the same
base, G(°°,n), and the same is true for GraGL+ (n,]R) and Gra SO (n) ,
with the base G (°°,n), and for Gra GL(n;(C) and GraU(n) , with the
base (EG (°°,n) . It is obvious that in all three cases the identity map
of the base is classifying. Therefore, the mappings

ext: Stee(B,0®n ) Stee (B,GL!IRn ) ,

ext: Stee (B,S03Rn ) •> Stee (B,GL +3Rn ) ,

and
ext: Stee(B,U(Cn ) Stee (B,GL!Cn )

are invertible for any cellular space B; see 4.4.2. In particular,

every real (complex) vector bundle with cellular base admits a Euclidean
(respectively, Hermitian) metric.
As a corollary, we obtain the theorem already formulated in
2.8: given any vector bundle £ with cellular base and any subbundle
there exists a subbundle ^ °f £ such that £ decomposes into
the sum of and

3. Similarly, the projection G+ (°°,n) -> G(°°,n) is classifying

for both inclusions (5), while the inclusion (EG(°°,n) -> G(°°,2n) is
classifying for both inclusions (6). However, a study of the homotopy
properties of these classifying maps is already a quite difficult task.
We shall return to the first of them in §5.6, armed with more sophis­
ticated tools.

4. For the inclusions (7) and (8) there are also obvious
classifying maps: for both left inclusions, such a map is the canonical
embedding G(°°,n-s) G(°°,n), for both middle inclusions - the canonical
embedding G+ (°°,n-s) -+ G (<»,n) , and for both last inclusions - the
canonical embedding (EG(°°,n-s) -> (EG(°°,n) (see 3.2). Identifying
G(°°,n-s), G+ (°°,n-s), and (EG(°°,n-s) with their images under these
embeddings, and using 3.5, we can write:

G (°°,n-s) => ske G(°o,n), G (°°,n-s) z> ske G (°°,n)

n-s + n-s +
and
(EG (°°,n-s) => ske2n_2s+-,(I:G (°°,n) .

From the first inclusion it follows that the pair

(G(°°,n) ,G(°°, n-s)) is (n-s)-connected (see 2.3.2.2), which in turn
implies (by Theorems 2.3.2.4 and 2.3.2.5) that the map

tt ( id,in) : 7T (B,G (°°,n—s )) tt (B,G (°°,n) )

 345

is invertible for any cellular space B with dimB $ n-s, and

surjective for any cellular space B with dimB = n-s. Consequently,

ext: Stee (B,GLlRn-S) -+ Stee (B,GL]Rn )

and
ext: Stee (B,03Rn-s) Stee (B,0]Rn )

are invertible for any cellular B with dimB $ n-s, and surjective
for any cellular B with dimB = n-s. In exactly the same manner the
inclusion G+ (~,n-s) => sken_sG+ (°°,n) leads to the invertibility
(surjectivity) of the mappings

ext: Stee (B ,GL+3Rn_S) -* Stee (B,GL+]Rn )

and
ext: Stee (B, S01Rn-S) -» Stee (B,SOUR11)

for any cellular B with dim B £ n-s (respectively, dimB = n-s),

while the inclusion (CG(o:',n-s) => ske„ __ (CGi^n) implies the
^n ^ S 1
invertibility (surjectivity) of the mappings

ext: Stee (B,GL(Cn-S) -* Stee (B,GL(Cn )

and
ext: Stee (B,U(Cn-S) -+ Stee (B,U(Cn )

for any cellular B with dimB £ 2 (n-s) (respectively, dimB =

= 2n-2s+1). Therefore, every n-dimensional real (complex) vector bundle
with cellular base of dimension £ n-s (respectively, £ 2n-2s+1) is
GLIRn -equivalent (respectively, GL(Cn-equivalent) to the s-fold suspension
of an (n-s)-dimensional bundle; furthermore, if given two (n-s)-dimensio­
nal real (complex) vector bundles with cellular base of dimension < n-s
(respectively, < 2n-2s+1) their s-fold suspensions are GLIRn -equivalent
(respectively, GL(Cn-equivalent), then the bundles themselves are
GL3Rn S -equivalent (respectively, GL(Cn S-equivalent) .

5. Exercises

1. Let Ç be an n-dimensional real vector bundle. Show

asso(£,IRn \ 0) is equivalent (in the sense of 1.1.2) to the bundle
with total space {x G tl K | x f 0} and whose projection is the
restriction of pr Ç to this subspace of tl£ .

Let £ be an n-dimensional Euclidean bundle. Show that

A
asso(Ç,Dn ) and assoit,S ) are equivalent to the bundles whose total
 346

spaces are the subspaces of tl £ consisting of the vectors of length

£ 1 and = 1, respectively, and whose projections are the appropriate
restrictions of pr £ .

2. Let £ be an n-dimensional real (complex) vector bundle.

Show that asso(£,V'(n,k)) (respectively, asso (£ ,(CV1 (n,k )) is
equivalent with the bundle with total space

{(x^ . . . ^ ^ ) G tl £ x ... x tl £ | pr £ (x1 ) = ... = pr£(xk ) ,

k x ,... ,x linearly independent} ,
\ K

and whose projection is the restriction of the composite map

^r l pr£
tl £ x . .. x tl £ --- -— > tl £ ■ ^ > bs £ .

Let £ be an n-dimensional Euclidean (Hermitian) bundle.

Show that asso (£ ,V (n,k) ) (respectively, assso (£ ,(CV (n ,k )) is
equivalent to the bundle with total space

{ (x1 , . ..,xk ) G tl £ x . . x tl £ | pr £ (x1) = ... = pr £ (xk ) ,

k x^,...,xk is an orthonormal frame},

and whose projection is the restriction of the composite map

tl £ X ... X tl £ !— » tl £ ■■Pr 5 , bs £ .

3. Consider the spaces T and S introduced in Exercise

1.2. 9. 5. Now GL+ (1 ,HR) acts on T\ 0 from the right by ({x^},t) {tx^}.
Clearly (T v 0)/GL+ (1,]R) = S. Show that the GL (1,3R) -bundle
defined by this action is locally trivial, but not trivial. Show that
the associated oriented one-dimensional real vector bundle does not
admit a Euclidean metric.

§6. SMOOTH BUNDLES

1. Fundamental Concepts

1. Let 1 £ r £ a. A bundle £ is called a bundle of class

r r r ------------- -— ~
C , or a C -bundle if tl £ and bs £ are C -manifolds, and for each
point bQ € b s £ there are a neighborhood U of b , a Cr-manifold
 347

F with dF - 0 if 9U ф 0 , and a Cr-diffeomorphism h: U x F

-1
-+ pr Ç (U) , such that pr Ç (h(b,x) ) = b for all b € U and x £ F.
The С -bundles with s ^ r will be referred to as bundles of class
^r ^r >1
С , or С -bundles. The C ' -bundles are called smooth.
r
If Ç is a С -bundle, then p r £ is obviously a
r
С -submersion. In particular, the fibers of a smooth bundle are neat
submanifolds of the total manifold tl ç (see 3.1.5.8). Moreover, the
fibers over points belonging to the same component of the base of a
r . r
С -bundle are pairwise С -diffeomorphic. If bs £ is connected and
_ -i
9 b s £ / 0, then the fibers have no boundary, and 3tl £ = p r £ (3bs£),
whereas if 3bs Ç = 0, then 3tl £ = U ^ b s Эрг £ ”*(b) ; in the first

situation, the restriction (3tl Ç,ab pr Ç, 3bs Ç) of the bundle £ to

X;
9bs £ is a С -bundle, whereas in the second (9tl £ ,ab pr £ ,bs £ ) is a
x
C -bundle.
Given two Cr-bundles, Ç. and such that 3bs Ç . = 0
r
and 3tl = 0, the product £ x £2 is a С -bundle.

The restriction of a С -bundle to a neat submanifold of its

£
base is clearly a С -bundle.
r r
Suppose that Ç is a С -bundle, В is a С -manifold, and
r —1
f: В -> bs Ç is a С -map such that the fiber pr £ (f (b )) has no
I xr
boundary for all b £ ЭВ. Then f £ is aС -bundle, and we say that
1 2T‘
the bundle f "£ is neatly induced. For example, given a С -bundle £,
I
in’£ is always neatly induced when in is either the inclusion of a
neat submanifold in bsÇ, or the inclusion 3 b s £ + b s £ ; obviously,
! jo
in *£ coincides, as a С -bundle, with the corresponding restriction of
Ç.
>r
2.Let 0 £ s £ r. A map ф from one C ' -bundle into another
is said to be a ÇS-map (a СS-isomorphism) if tl £ and bs Ç are
СS-maps (respectively, CS-diffeomorphisms for s ^ 1, and homeomorphisms
for s = 0). A CS-isomorphism which is also an equivalence is called a
С S-equivalence.

A C^r-bundle £ is said to be CS-trivial if it is

CS-equivalent to a standard trivial bundle (bsÇ x F,pr^,bsÇ) , where
F is a C^r-manifold (such that 3F = 0 if 9bs Ç i 0 ) . Every Cr-bundle
is obviously locally Cr-trivial, meaning that each point of bs £ has
a neighborhood U such that Ç|y is Cr-trivial; in particular, every
smooth bundle is topologically locally trivial.
1 >r r
If f'£ is neatly induced from the C ' -bundle К by a С -map
f, then ad f : f !Ç + U s a Cr-map.Furthermore, if ф:
Ç' -+ £ is a
Cr-map, where £ and Ç' are C^r-bundles, and (bs<ÿ)'Ç is neatly
 348

i r
induced, then corr (p (bs cp)‘£ is also a C -map.

Smooth Bundles and Submersions

3. Let r * 00,and let f: X -* Y be a Cr-submersion, where

r -1
X and Y are C -manifolds, X is compact, and f (3Y ) = 3X. Then
(X ,f,Y ) is a Cr-bundle. The same holds true when r = a, provided
that X admits a Ca-embedding in Euclidean space.

(See 6.1 for a supplement to this theorem.)

PROOF. It suffices to examine the case f(X) = Y: indeed,

in thegeneral case the set f(X) is both open (see 3.1.5.8) and closed
(see 1.1.7.9), and hence is a union ofwhole components of Y. We show
that for each point y G Y there are a neighborhood U of y ,a
r r —1
closed C -manifold F, and a C -diffeomorphism h: U xf -* f (U) ,
such that f(h(y,x)) = y for all y G U and x G F.
From 3.1.5.8 (or, if it is more convenient, from 3.4.8.2), it
follows that f ^ (y q ) is a neat submanifold of X for y^ G int Y ,
and a neat submanifold of 9X for y^ G 3Y, and in both cases f ^ (Yq )
is closed as an independent manifold. Set F = f ^(Yq ), and pick a

Cr-embedding j: X 3R^, a Cr-transversalization t of the embedding

j|p: F IR^, and a neat tube Tub^p. Now consider the map

(j): j 1 (tub^p) + Y x f , cj>(x) = (f (x) ,prT (j (x) )) .

The following properties of cf> are immediate: (¡> is of class Cr ;

-1
<f>(9(j (tub^p))) c= 3 (Y x f ) ; (p is injective on F; and the
differential d^(p is nondegenerate for all x G F. Since F is
compact, we conclude that cp defines a diffeomorphism of a neighborhood
of Fonto a neighborhood of (p(F) = yQ x f (see 3.1.5.5). Using once
more the compactness of F, we see that the last neighborhood contains
a set of the form V x F, where V is a neighborhood of y (in Y).
— 1
Let U be a smaller neighborhood of yn , such that j (f (u) ) c= tub p.
-1 -1 -1 T
Then cp (U x f ) = f (U) , and we can finally set h = (ab <p) : u x f
+ f'1 (U).

4 (EXAMPLES). The bundles (V(n,k ) ,p r ,G(n,k)),

(V (n,k) ,pr ,G+ (n,k) ) , ((EV (n ,k ) ,pr,(EG (n ,k ) ) , and ®IV (n ,k ) ,pr ,JHG (n ,k )) ,
whose projections are the submersions defined in Subsection 3.2.2, are
principal C -bundles with structure groups 0(k), SO(k), U(k), and
Sp(k), respectively. Similarly, (V(n,k),pr,V(n,k-q)),
((EV (n,k ) ,pr,(EV (n ,k-q)) , and (HIV (n ,k ) ,pr ,HV (n ,k-q) ) ,whose pro ject ions
 349

are the submersions defined in Subsection 3.2.1, are Steenrod Ca-bundles

with structure group 0(n-k+q), U(n-k+q), and Sp(n-k+q), and
standard fibers V(n-k+q,q), ΠV (n-k+q,q), and nHV(n-k+q,q) (on which
the above groups act canonically; see 2.3.12). The coverings
1 1 1
(]R,hel,S ) , (S ,helm ,S ), and (G+ (n,k),pr,G(n,k)), defined in 1.2.6,
are also principal Ca-bundles.
Among the previously listed principal Ca-bundles we find
3 2 7 4
(S ,pr,S ) and (S ,pr,S ), whose projections are the Hopf submersions
(see 3.2.2.9); they are called the Hopf bundles. The Hopf submersion
S + S
defines a Ca-bundle, which is also known as a Hopf bundle;
7
its fibers are diffeomorphic to S (but this bundle is not endowed
with any special structure group).

The Smooth Bundles as Steenrod Bundles

27
5. Let F
be a C -manifold with r ^ 1. According to 2.3.10,
r r
F is an effective Diff F-space, and thus every C -bundle £ whose
fibers are Cr-diffeomorphic to F is a W-F-bundle (see 3.1.3 and 3.2.1).
Moreover, Ç is clearly locally W-F-trivial, i.e., it is a Ehresmann-
-Feldbau bundle. However, the procedure that enabled us in Subsection
3.4 to turn Ehresmann-Feldbau bundles into Steenrod bundles does not
work here: as we already had the ocassion to note (see 3.5.2), when
27
dim F > 0, the natural action of Diff F on F is not topologically
effective. Nevertheless, the set MH(Ç) carries a natural topology,
transferred from Cr (F,tlÇ) with the aid of the injective mapping
MH(£) ** Cr (F,tl Ç ) which takes each dif feomorphism a € MH(Ç) into
the map [in: a(F) -> tl fj ° a. It is clear that with this topology on
MH(Ç), £ becomes a Steenrod F-bundle.

It is instructive to compare the implicit group structures

described above for smooth bundles with the implicit group structures
of locally trivial bundles (see 3.4.5). Here we merely mention two
difficulties encountered in the differential situation. Firstly, not
every Steenrod F-bundle can be smoothed - the fact that its base need
not be a manifold is already an obstruction. Secondly, simple examples
show that there are F-maps of Cr-bundles with fibers diffeomorphic to
27
F, which are not C -maps.
 350

2. Smoothings and Approximations

1. This subsection is similar in character with §3.4: here

we generalize some of the results obtained there for smooth manifolds
to smooth bundles. Although part of these results are indeed rather
important, some problems are not touched upon at all. For the sake of
brevity, we shall consider below only the closed case; the reader can
find some additional information concerning the (more general) compact
case in Subsection 6 (see 6.2-6.5).
We shall need two notations, for 0 £ s £ r: if £ is a
Cr-bundle, we let SecS£ denote the set of all Cs-sections of and
if i and £1 are Cr-bundles, we let CS (£,£') denote the space of
g
all C -maps If s ^ a, both sets carry natural topologies:
SecS£ is a subspace of CS (bs£,tl£), while CS (£,£') is a subspace
of the product CS (tl£,tl£') x CS (bs £ ,bs £ 1) .

5-Transversalizations and Tubes

2. Our immediate task is to adapt the definitions and

theorems of Subsection 3.4.3 for use in the more general setting of
this subsection.
We start with the transversalizations. Let £ be a smooth
bundle such that tl £ is closed, and let j: tl £ 3R^ be a
differentiable embedding. A continuous map t : tl £ G(q,q-n), where
n = dim tl£ - dim bs is called a g-transversalization of the embedding
j if the restriction t | . is a transversalization of the
1pr £ (y)
embedding jI . for all points y € bs £ . A fundamental example
'pr 5 (y)
is the normal g-transversalization, which takes each point x G tl £
into the orthogonal complement of the plane d j(Tang (pr (pr £ (x) ) ))
q x x
in ]R . Our ^-version of Theorem 3.4.3.7 asserts that if E, and j
are of class C (1 £ r £ a) , then j admits a £-transversalization
of class C . The proof is an obvious modification of that of Theorem
3.4.3.7.
Now we move on to tubes. Let t be an arbitrary £-transver-
salization of the embedding j. We define the tube TubTp and the
open tube tubTP as the following subsets of bs E, x lRq :

T u b T p = uxetl E, (P r ^ x > x d T (x,p))

and
 351

tubT P = Ux6tl£ (Pr ^ ( x > x (dT (x,p)^sT (x,p) )),

where d^ (x,p) and (x,p) are the ball, and respectivelythe sphere,
with center j(x) and radius p in the plane j(x) + t ( x ) .

Equivalently,

TUV = V b s £ (y x Tubi p)
pr z 1 (y)
and
tub p = U (y x tub P ).
T yebs £ ■* t
pr Г 1 (y)

The tube Tub^p is neat if there is a a > p such that:

the sets pr £ (x) x (d^_(x,tf) \ s^(x,cr)) are pairwise disjoint, tub^a

is open in bs Ç x r and the map tub^a + tl Ç , which takes

pr Ç(x) x (d^(x,a) ^ s (x,a)) into x, is smooth. The restrictions
of this last map to Tub^p and tub p are called projections and are
denoted by pr (they clearly do not depend upon the choice of a). If
Ç, j, and t are of class C , r ^ 1, then the following are true:

there exists a neat tube; every neat tube Tub^p is a submanifold of

bs Ç x with int Tub p = tub p; and pr : Tub p tl Ç is a
r T T T T
C -submersion. Again, the proof is an obvious modification of the
proofs of Theorems 3.4.3.4 and 3.4.3.5. We must also modify
appropriately the construction in 3.4.3.3: now the model Tu^ is
defined as the subset { (x,t) e tl Ç x ]R^ | t £ x (x) } of tl Ç x IR^,
while nat : Tu^ + bs £ x is given by nat(x,t) = (pr Ç (x) ,j (x)+t)

The Basic Theorems

3. Letandr £ let
«>, £ and £ 1 be C -bundles with
closed total spaces tl£ and tl £ f, and closed bases bs £ and bs £ '.
Then Cr (£,£') is dense in Cs (£,£') for any s < r. The same holds
true for r = a, provided that tl£, tl £1, bs £ , and bs £ 1 admit
Ca-embeddings in Euclidean spaces.

Pick a Cr-embedding j 1: tl £ 1 +
PPOOF. a £ 1-transver-
x*
salization t 1 of j 1 of class С , and a neat tube Tub t *
,p'.
,s s 1
Let U denote the subset of C (tl£,tl£!) x c (bs£,bs£*) consisting
of all the pairs (F: tl £ + tl £ 1, f : bs £ + bs £ 1) such that
(f (pr £ (x) ) ,j 1 (F (x) )) £ tubT ,p * for all x £ tl £ . Then U is open
 352

and contains CS (£,£'). For (F,f) £ U, define 0(F,f): tl £ -> tl £'

by
x ^ pr^, (f(pr £(x)),j ’ (F(x))) .

Obviously, ($(F,f),f) £ CS (£,£') and the map y : U -*■ C (£,£'),

JO
f(F,f) = (<J>(F,f) ,f) , is a retraction which takes U D (C (tl£,tl£') x
x Cr (bs £,bs £’)) into Cr (£,£’). Since Cr (tl£,tl£') x Cr (bs £ ,bs £ ’)
is dense in CS (tl£,tl£’) x CS (bs£,bs£') (see 3.4.4.2), the existence
r s
of such a retraction implies that C (£,£') is dense in C (£,£').
>r
4. Let s < r s; °°, let £ and £' be arbitrary C ' -bundles
such that t l £ , tl £' , bs £ , and bs £' are closed manifolds, and let
f : bs £ -> bs £'be a Cr-map. Then the set {<J>£ Cr (£,£')| bs <J> = f }
is dense in {<J> £ CS (£,£')| bs <}> = f}.The same holds true forr = a,
provided that tl £ , tl £' admit ca-embeddings in Euclidean spaces.
s
PROOF. Let F denote the subspace of C (tl£,tl£')
consisting of the maps tl <J) such that<(> £ CS (£,£’) and b s <f> = f.
We show that F n Cr (£,£') is dense inF-
r cr1
Pick a C -embedding j ’: tl £' -* IR^1 , a £'-transversalization
t ' of j' of class Cr , and a neat tube Tub^,p '. Consider the subset
U <= CS (tl £, tl £ ') consisting of all maps F : tl £ -> tl £ ' such that
(f (pr £ (x),j ' (F(x))) £ tub^,p' for all x £ tl £ . It is clear that U
is open and contains F. Moreover,the mapping U F, transforming
each F £ U into the map

x h- pr^ , (f (pr £ (x) ) ,j ' (F (x) ) ) ,

is a retraction which takes U D Cr (tl£,tl£') into F n Cr (tl£,tl£').

r s
Since C (tl£,tl£*) is dense in C (tl £ ,tl £ ') , the existence of such
jo
a retraction shows that F D C (£,£') is dense in F, as claimed.

5. Let r £ 00,and let £and £' be arbitrary C^r-bundles

such that the manifolds tl £ , tl £ '
, bs £ , and bs £ ' are closed.
r r
If 0 < s < r, then the set of all C -isomorphisms (C -equivalences)
g
£ £' is dense in the subspace of all C -isomorphisms (respectively,
s s
C -equivalences) of C (£,£*). The same holds true for r = a,
provided that tl £ , tl £' , bs £ , and bs £ ' (respectively, tl £ and
t l £ ') admit Ca-embeddings in Euclidean spaces.

This is a consequence of 3, 4, and 3.4.1.6.

6 (COROLLARIES). If two C^r-bundles with closed total

manifolds and bases are C^-isomorphic and r < oo, then they are
jc
C -isomorphic. The same holds true for r = a, provided that the
total manifolds and the bases admit Ca-embeddings in Euclidean spaces.
 353

^r
If two C -bundles with closed total manifolds are
1 r
C -equivalent and r £ then they are C -equivalent. The same holds
true for r = a, provided that the total manifolds admit Ca-embeddings
in Euclidean spaces.
>27
7. I_f r £ 00, then given any C ^ -bundle £ with closed
tl 5 and bs £ , the space Secr£ is dense in SecS£ , for any s < r.
The same holds true for r = a, provided that tl£ and bs£ admit
Ca-embeddings in Euclidean spaces.

This is a result of Theorem 4, applied to thebundles £ and

(bs £ ,id bs £ ,bs £ ) and to the map id bs £ .
>2T
8. Every C -bundle £ such that t l £ and bs£ are
2T cl
closed manifolds is C -isomorphic to a C -bundle n with the property
that tln and bs n can be ca-embedded in Euclidean spaces.

PROOF. By 3.4.9.6, there exist C -manifolds, T and B,

cl
admitting C -embeddings in Euclidean spaces, together with C -diffeo-
2f
morphisms F : T -* tl £ and f: B bs £ . Pick a C -embedding
j: tl £ JR^, a £-transversalization t of j of class Cr , and a
neat tube Tub^_p. Let U denote the subset of Cr (T,B) consisting of
all thesubmersions p: T B such that the image of the composite map

T -di-a-3- , T X T (f °p)x (j °F) , bs 5 x K q

is contained in tub p. Obviously: U is open; f ^ © p r £ © F E U ;

^2T
the mapping $ : U -> C (T,tl£) , which takes each g £ U into the map
x pr (fog (x) ,j ©F(x)) , is continuous and $ ( f ^ ° p r £ © F) = F .
Since ^F is r
a C -diffeomorphism, there is a neighborhood 1/ of
o pr £ © F in Cr (T,B) with the property that 1/ cz U and 0(g)
y 3.
is a C -diffeomorphism for all g € V. Now pick some C -map h
and set n = (T,h ,B ) . By Theorem 1.3, n is a C3-bundle, and it is
plain that$ (h) and f yield a Cr-isomorphism n -> E,.

3. Smooth Vector Bundles

1. Usually, when we encounter a bundle which is smooth, i

carries some additional structures, most frequently a group structure
of Steenrod type. In such cases, smoothness plays the same role as
does the topology in the theory of Steenrod bundles discussed earlier
(§§ 3, 4), and it is natural to try developing this analogy into a
weighty theory of smooth Steenrod bundles.
 354

Unfortunately, such a program is beyond the scope of our book.

Therefore, we shall restrict ourselves to the basic facts concerning
smooth vector bundles, which are of main interest to this study, and can
be derived in a less cumbersome manner that the general theory.
We remark that because the fibers of a vector bundle of
positive dimension are not compact, it is not possible to deduce the
smoothing and approximation theorems below (Theorems 8-12) from the
results of the previous subsection without resorting to additional
devices. However, we prefer to give simple, straightforward proofs of
these theorems, so that this subsection becomes independent of the
previous one.

Fundamental Concepts

27
2. £ is an n-dimensional real vector C -bundle (1 £ r £ a)
27
if it is both an n-dimensional real vector bundle and a C -bundle,
and these two structures are compatible, meaning that the restriction
of £ over a small enough neighborhood of an arbitrary point of bs £
is Cr-GLIRn -trivial (i.e., is Cr-GLIRn -equivalent to a standard trivial
bundle, where, of course, a Cr-GLIRn -equivalence is just a GLIRn -equi­
valence which is simultaneously a (^-equivalence) . The Euclidean,
27
complex vector, and Hermitian C -bundles are similarly defined.
^r
Products of vector, Euclidean, or Hermitian C -bundles, as
well as bundles induced (in particular, obtained by reducing) such
>r
bundles, are again C ' -bundles of the same kind, provided the conditions
imposed by the corresponding definitions from Subsection 1 (see 1.2)
are fulfilled.
We add that the statements and proofs of Theorems 3.2.8 and
3.2.9 carry over, with obvious modifications, to vector, Euclidean, and
Hermitian C^r-bundles. Here we formulate only the Cr-GLnRn -version of
Theorem 3.2.9: let f: Ç n be a Cr-GLmn -map, where Ç and n are
C"'r-GL3Rn -bundles, and suppose that the bundle (bs f) !ti is neatly
induced (see 1.1.); then corrf is a Cr-GLnRn -equivalence.

3. The explicit descriptions of the vector, Euclidean, and

Hermitian bundles, given in Subsection 5.1, have obvious Cr-analogs
(1 s? r £ a) . The Cr-analog of Theorem 5.1.3 asserts that a Cr-bundle
whose fibers are n-dimensional real vector spaces is an n-dimensional
real vector Cr-bundle if and only if the partial vector operations
indicated in 5.1.3 are Cr-maps. Similarly, the Cr-analog of Theorem
27
5.1.5 asserts that a C -bundle whose fibers are n-dimensional Euclidean
 355

is an n-dimensional Euclidean Cr-bundle if and only if the partial

vector operations and the metric (equivalently, the square of the length
of vectors, considered as a function on the total space) are of class
r r*
C . In particular, in order to turn a real vector C -bundle into a
Euclidean C -bundle, one has to equip it with a Euclidean Cr-metric.
The corresponding complex formulations (i.e., the Cr-analogs of the
theorems in 5.1.10) are obtained by replacing Euclidean bundles and
Euclidean metrics by Hermitian bundles and Hermitian metrics.

Smooth Grassman Bundles

4. The Grassman bundles defined in 5.3.6 provide (for m < 00)

fundamental examples of smooth vector, Euclidean, and Hermitian bundles.
Namely, if 0 £ n £ m < °°, then Gra (m, GL (n, IR) ) , Gra (m,GL (n,CC) ) ,
Gra(m,0(n)), and Gra(m,U(n)) are obviously real vector, complex
vector, Euclidean, and Hermitian Ca-bundles, respectively, all of them
n-dimensional. The third (fourth) bundle differs from the first
(respectively, second) by having a Euclidean Ca-metric (respectively, a
Hermitian Ca-metric).
jo

Theorem 5 below may be thought of as a weakened C -analog

(for r f a) of that part of Theorem 5.3.8 concerning Gra(m,GL(n, IR) )
and Gra (m,GL (n, (C) ) .
jo

5. Let £ be an n-dimensional real (complex) vector C -bundle

jo

with compact base. If r i a, then there are a number m and a C -map

f : bs £ -»- G(m,n) (respectively, f : bs £ + CCG(m,n)) such that £ _is
r n ^
C -GLIR -equivalent to the bundle f ‘Gra(m,GL(n, IR) ) (respectively,
r n ^
C -GL(C -equivalent to the bundle f 'Gra (m, GL (n,CD) ) .

Since the proofs of the real and complex cases differ only in
some obvious details, we shall prove only the former. Choose, for every
point b € bs K , a chart E Atlbbs £ such that

^^(supp^^,b) = (HR^,0) or (1R^,0) [q = dimbs£]

and £| , is Cr-GLlRn -trivial, and then fix a Cr-GLlRn -triviali-

Isupp
zation

Tb : (supp^b x 2Rn ,pri ,suppU,b ) - 5|suppti)b •

Now cover bs £ by a finite number of sets (Dq ) , say (D^) ,. . .,

ti/-1 (Dq ) and pick a Cr-function a: lRq IR which equals 1 on Dq

b
s
 356

n
and 0 outside 2D Finally, define H tl £ IR by

— 1 ” 1
a (ip (pr £ (x) )) pr otl Tb (x) , if x £ pr£ (suppi^),
i i i
H± (x) =
-1
0, if x t pr £ (supp ),
l
n
where pr„ = [pr„ : supp ipb . x nR IRn ] , and let H: tl £ IRnx ...x]Rn =
1
= IR be given by H(x) = (H^ (x),... , (x)). Clearly, H is a C -map
and its restrictions H are linear monomorphisms for all
pr £ 1 (b)
b E bs t . Set m = s n and f: bs £ -> G (m,n) , f(b) H (pr £ 1 (b)
i
To verify that the bundles £ and f *Gra(m,GL(n,JR) ) are
Cr-GLIRn -equivalent, it is enough to produce a Cr-GLnRn -map
<f>: £ Gra (m,GL (n,IR) ) with bs (j> = f. Such a ^ is defined by
tl cj) : tl £ -> tl Gra (m,GL (n, ]R) ) , tl cj)(x ) = (f (pr £ (x) ,H (x) ) (recall
that tl Gra (m,GL (n,IR) ) = {(y,y) € G(m,n) x ]Rn | y £ y}).

An Application

27
6. ijE 1 £ r £ °°, then every real (complex) vector C -bun
with compact base has a Euclidean (respectively, Hermitian) C -metric.

Since Gra (m,GL (n,]R) ) (Gra (m, GL (n,(C) ) ) has a Euclidean

(respectively, Hermitian) Ca-metric, 6 is a consequence of 5.

Smoothings and Approximations

>r
7. Given two real or complex vector C ' -bundles, £ and £ !,
and a C S-map f : bs £ ' -* bs £ with r ^ s ^ 0, we let LS (£ ,£ 1;f )
denote, in Theorems 8 and 10 below, the set of all linear CS-maps
cf>: £ 1 + £ such that bs <p = f. If s ^ a, then LS (£,£!;f) inherits
s
a natural topology as a subset of C (£',£).
When dim £ = dim £ ' = n, bs £ = bs £ ' , and f = id, LS (£,£';f)
contains the set of all CS-GL3Rn -equivalences £' -> £ in the real case,
and the set of all CS-GL(Cn-equivalences £' -+ £ in the conplex case. In both
cases this subset is open for any s / a.
Notice that among the spaces L S (£,£';f) we find SecS£,
0 £ s £ r (see 2.1). More precisely, Secs£ is canonically homeo-
morphic to L (£ ,£ ';f) , where £' is the standard trivial bundle
(bs £ x ]R ,pr^,bs £) , and f = id bs £; the canonical homeomorphism
 357

s s
L (£,£'; f) Sec £ takes each map <}>: -> £ into the section
b ^ tl <Mb,1) .

8. Let E, and £' be real or complex vector C^r-bundles

with compact bases. If 0g s < r £ «>, then Lr (£,£’;f) is dense in
LS (£/£';f) for any Cr-map f: bs £' bs E, .

Since the proofs of the real and complex cases differ only in
some obvious details, we shall again prove only the former. By Theorem
5, we may assume that £ = g *Gra (m,-GL (n,HR) ) and = g' !Gra (m' ,GL (n' ,]R),
where g: bs £ -> G(m,n) and g' : bs £ 1 -*• G(m',n') are some Cr-maps.
Then we can identify tl £ with the Cr-submanif old { (b,y ) £ bs £ x i " 1)
y = g (b) } of bs £ x 3Rn and, similarly, tl £ ' = { (b',y') € bs F,1 x iRm |
y' = g(b')}. The orthogonal projections of 3Rm onto its subspaces
g(b) with b £ bs £
combine to define a Cr-map p: bs £ x ]Rm tl £,
x in1
and a C -map p' : bs x ]R £* is similarlydefined. Let Abe the
m* m
Euclidean space of all linear maps IR IR (which is the same as the
space of all real (m><m1)-matrices; c f . 3.2. 1.1. or 3.2. 1.7), and
consider the mappings

LS (£,£' ;f) -* CS (bs £ ',A)

and
V: CS (bs V ,A) -> LS (£,£' ;f) ,
given by

{ [$(<(>) ] (b') } (y' ) = [pr2 : bs £ X 3Rm + 3Rm ] (tl <j> o p' (b* ,y' ) )
and

[tl (T (h))] (b •,y ’) = p (f (b ') , th (b ')] (y 1))

^ TTI * c;
(where <j) € L (£,£'; f) , b ' € bs K ' , y' € TR , and
h € C (bs £ ',A) ) .
r
Clearly, y
is continuous (and so is <i>) and takes C -maps into
r s
C -maps. Moreover, V ° $ = id L (£,£*;f) , and hence ¥ is surjective.
r s
Since C (bs£',A) is densein C (bs£f,A) (see 3.4.6.5), we conclude
that Lr (£,£';f) isdense in LS (£,£!;f).[Explanation: Theorem 3.4.6.5
is applied after we have completed the space A to a sphere by adding
a point; cf. 3.4.4.2 and 3.4.4.6.]
>x
9. Let £
be a real or complex vector C " -bundle with
r s
compact base. If 0 £ s < r £ °°, then Sec £ is dense in Sec £.
r r
This is a consequence of 8: Sec £ = L (£,£*;idbs£) and
SecS£ = LS (£,?'; id bs O , where £ ' = (bs £ x ]R,pr1 ,bs £)'; see 7.

10. Let £ and £' be n-dimensional real (complex)vector

C^r-bundles with common compact base. If 0 £ s < r £ °°, then the set
 358

of Cr-GLIRn -equivalences (respectively, Cr-GL(Cn-equivalences) £'->■£

s
is dense in the subset of L (£,£';id) consisting of all
CS-GLlRn -equivalences (respectively, CS-GL(£n-equivalences) .
s
Since the last subset is open in L (£,£';id), 10 is a
consequence of 8.

11 (COROLLARY). If two n-dimensional real (complex) vector

C^r-bundles with compact base are GL]Rn -equivalent (respectively, GL(Cn-
-equivalent) and r / a, then they are Cr-GLIRn -equivalent
n r
(respectively, C -GL(E -equivalent) .

12. If the base of the n-dimensional real (complex) vector

>r
bundle £ is a compact C -manifold with 1 £ r £ 00, then £ is
GLIRn -equivalent (GL(Cn-equi valent) to a real (complex) vector C^r-bundle.
g
If the base of the n-dimensional real (complex) vector C -bundle £ is
a compact C^r-manifold, where 1 £ s < r £ 00, then 5 is CS-GLIRn -
-equivalent (CS-GL(En-equivalent) to a real (complex) vector C^r-bundle.

We shall prove again only the real case. Let £ be an

>r
n-dimensional real vector bundle with b s £ a compact -manifold.
n 1
By 5.3.8 and 3.5.2.13, £ is GLIR -equivalent to f *Gra (m,GL (n ,JR) ) ,
where m is large enough and f
is some continuous map b s £ G(m,n).
s
If £ is an n-dimensional real vector C -bundle such that bs £ is a
compact C^r-manifold, then by Theorem 5 £ is CS-GLIRn -equivalent to
i g
f *Gra (m, GL (n, IR) ) , where m is large enough and f is some C -map
bs £ G(m,n) . In both cases f is homotopic to a C ' -map
g: bs £ -v G(m,n) (see 3.4.6.5, 1.3.6.6, and 3.4.5.10), so that £
n *
is GLIR -equivalent to g *Gra(m,GL(n,IR) ) (see 4.1.5). This completes
the proof of the first claim; as for the second, we need only add that,
by 11, £ is CS-GLIRn -equivalent to g "Gra(m,GL(n,IR) ) .

Constructions

13. We conclude this subsection with a short review of the

constructions described in § 5.
s
By definition, a C -subbundle of a (real or complex) vector
C -bundle £ is a subbundle of £ in the sense of 5.2.2 or 5.2.3,
whose total space is a CS-submanifold of tl£ . A CS-subbundle is
clearly a vector C —bundle. The C —subbundles of Euclidean or Hermitian
C -bundles are similarly defined. The C -bundles of C -bundles will be
simply referred to as subbundles.
 359

According to 5.2.5, every subbundle n of a Euclidean or

Hermitian bundle £ has an orthogonal complement n , and it is clear
j_ g
that: n is a C -subbundle of £ together with • n; the canonical
equivalence n'L -*■ £/n (see 5.2.6) turns £/n into a Euclidean or
Hermitian CS-bundle (and thus becomes a C-equivalence). We see that in
the (real or complex) vector Cs-case, £/n becomes a vector Cs-bundle
by introducing on £ a (Euclidean or Hermitian) CS-metric. Recall,
however, that we have established the existence of such a metric only
under the assumptions that the base is compact and s ^ a (see 6).
Let £^ and £2 be real vector Cr-bundles with a common
boundaryless base. Then the construction of £ © £. (see 5.2.9)
shows that this sum is again a real vector C -bundle. The difficulty
occuring when the base has a boundary (i.e., the fact that the product
£>1 x ^~snot defined as a Cr-bundle) can be circumvented with the
aid of the formulas

ti(?1 ® ?2) = ti( (Pr h 2)

and
p r ( ^ @ £2) = pr ?2 ° pr ((pr ) 1£2) .

Tf the conditions in 5.2.9. are satisfied, then these formulas are

equivalent to the definition of and this remains valid under
our present circumstances, provided that the base has no boundary; the
same formulas are now taken as the definition of the sum when the
boundary is present. One can repeat the argument for complex vector,
27
Euclidean, and Hermitian C -bundles. In particular, we can define the
27
suspension (see 5.2.10) of a C -bundle.
The C -variants of the other constructions described in §5
and their mutual relations are already evident. In particular, the
conjugate of a complex vector (Hermitian) C -bundle is a complex
27
(respectively, Hermitian) C -bundle; the realification (see 5.1.12) of
a complex vector (Hermitian) Cr-bundle is a real vector (respectively,
Euclidean) Cr-bundle; the complexification (see 5.2.11) of a real vector
(Euclidean) Cr-bundle is a complex vector (respectively, Hermitian)
Cr-bundle; and in the Cr-versions of Theorems 5.2.13 and 5.2.14, the
equivalences conj and K become (^-equivalences.

4. Tangent and Normal Bundles

1. The basic notions of tangent and normal bundles have

actually already been introduced and used in Chapter 3. However, only
 360

now,that we have acquired the ideea of a smooth vector bundle, can we

present the full-fledged definitions of tangent and normal bundles and
give them the general, correct treatment that they deserve.

Tangent Bundles

2. Recall that in Chapter 3 we defined, for an arbitrarily

given Cr-manifold with r ^ 1, the real vector spaces Tang X (x € X) ,
the Cr-1 -manifold Tang X , and the projection p r : Tang X X (see
3.1.4.1 and 3.1.4.2). Comparing these objects with the general
definitions given in 5.1.2 and 3.2, we readily see that (Tang X ,pr,X)
is a real vector bundle of dimension dim X and, for r > 2, a real
r— 1
vector C -bundle of dimension dim X , calledthe tangentbundle of
the manifold X, and is denoted by tang X .
Similarly, confronting the definition of the differential
df: Tang X Tang Y of a Cr-map f: X Y (see3.1.4.3) with the
general definitions given in 5.1.14 and 3.2, we conclude that (df,f)
r— 1 r
is a linear C -map tang X -*■ tang Y . If f is a C -dif feomorphism
r—1
then (df,f) is a linear C -isomorphism.

3. The notion of vector field has been defined twice: once

for smooth manifolds (see 3.1.4.5), and once for vector, Euclidean, and
Hermitian bundles (see 5.1.15). Now it is plain that the second
definition generalizes the first one: a vector field on a smooth
manifold X
is simply a vector field in its tangent bundle tang X .
5
In particular, the parallelizability (C -parallelizability) of an
n-dimensional smooth manifold X is equivalent to the GL3Rn -triviality
s n
(respectively, C -GLIR -triviality) of the bundle tang X . Comparing
this with Theorem 3.11, we see that a parallelizable compact Cr-manifold
r— 1
with r £ 00 is C -parallelizable.
A smooth manifold is stably parallelizable if its tangent
bundle is stably trivial. The discussion in 5.4.4 and Theorem 3.5.2.13
show that if a compact manifold is stably parallelizable, then the
stabilization occurs already at the first step, i.e., the bundle
su tang X is GL}Rn+1 -trivial for any stably parallelizable n-dimensional
manifold X.

4. Recall that, given a point x of the smooth manifold X,

each chart <t> £ Atlxx defines a (j)-basis for the tangent space Tang X,
and the matrix of the transformation from the (j)-basis to the ifj-basis
is just the Jacobi matrix of the map loc (<f), )id, computed at <|>(x) .
 361

From this it follows that the values of any orientation of X (assumed

on the charts of Catl X) are correctly transferred to the bases of
the spaces Tang^X, and in this manner an orientation is defined on
the bundle tang X . This procedure is clearly reversible, and hence
there is a one-to-one correspondence between the orientations of the
smooth manifold X and the orientations of its tangent bundle tang X .
In particular, X is orientable if and only if tang X is orientable,
and tang X is a GL+nRn -bundle for any oriented smooth n-dimensional
manifold X.
We already know that every parallelizable manifold is
orientable (see 3.1.4.6 and 3). Now we can add that every stably
parallelizable manifold is also orientable.

5. One attractive feature of tangent bundles is that they

can naturally be induced from Grassman bundles. Namely, if j : X
r r
is a C -immersion (for example, a C -embedding), then dj maps each
tangent space Tang^X onto an n-dimensional plane of passing
through 0. Thus a map t: X
G(q,n) is defined, and it is clear
i
that tang X is nothing else but t "Gra(q,GL(n,3R) ) (up to a
correcting Cr ^-GL3Rn -equivalence) . [In fact, this obvious observation
is older that the theory of bundles and was one of the factors which
stimulated its creation.] If X is orientable, then one can generalize
this observation and replace G(q,n), the standard fiber GL]Rn , and
the bundle Gra (q ,GL (n,TR) ), by G (q,n), GL+IRn , andGra (q,GL+ (n ,HR) ),
respectively. In all cases t is known as a tangential map. If j is
an embedding and no orientation is involved, thent coincides with the
composition of the normal transversalization X G(q,q-n) of j with
the canonical diffeomorphism G(q,q-n) G(q,n) (see 3.2.2.3).

6. A smooth manifold whose tangent bundle is equipped with a

Euclidean metric is called a Riemannian manifold. In this case the
r
metric is usually termed a Riemannian metric. If X is of class C ,
r— 1
then the metric can be at most of class C . In fact, Theorem 3.6
r— 1
shows that there is a Riemannian C -metric on every compact
Cr-manifold with 1 £ r £ 00. Incidentally, the same result may be
extracted from Theorem 3.4.2.1.
It is customary to look upon the tangent bundle of a Riemannian
g
manifold as a Euclidean bundle. If the metric is of class C , s 5s 1 ,
then this bundle is a Euclidean CS-bundle.
 362

Normal Bundles

7. The initial data involved in the concept of normal bundle

are two smooth manifolds, X and X' , and an immersion j: X ^ X 1.
The most important case is that of an embedding j. In the definitions
below n = dim X and n' = dimX' .
Let us examine first the simplest case: X is a submanifold
of the Riemannian manifold X', and j is the inclusion X X'. One
may naturally view tang X as a subbundle of tang X' I . By 5.2.5,
1
tangX has in tang X'j^ an orthogonal complement, (tangX) This is
an (n'-n)-dimensional Euclidean bundle with base X, called the normal
r
bundle of X, and denoted norm X . If X and X' are of class C '
r—1
with r ^ 2,and the Riemannian metric is of class C , then norm X
>r— 1
is a Euclidean C ' -bundle; see 3.13. In all cases, the sum
n* i
tang X © norm X is canonically OJR -equivalent to tang X 1 ; when
>r 1•A
X and X' are of class C ' with r ^ 2, this equivalence is of
class Cr ^ .
To define the normal bundle for an arbitrary immersion
j: X + X* and without recourse to a Riemannian metric, we have to
i
replace the restriction tang X ' by the induced bundle j ’tangX'
and then pass to a quotient instead of taking an orthogonal complement.
More exactly, the normal bundle, normj , of the immersion j: X X',
with X and X' smooth manifolds, is defined as
i
norm j = j *tang X'/Im corr (dj ,j ) .

This formula defines an (n'-n)-dimensional real vector bundle over X.

The sum tang X © norm X is canonically GL]Rn -equivalent to the
.!
induced bundle j ’tangX' . Unfortunately, there is less to say about
the differentiability class of norm j and of the canonical equivalence
,i
j ’tangX' tang X © norm j : if X' is either compact or dif feomorphic
to an open subset of a compact manifold, and j is Cr with 2 £ r £ »,
>r — 1
then normj is a real vector C ' -bundle, while the above canonical
r— 1
equivalence is C . These limitations are obviously due to the fact
that we have had no theorems which guarantee the existence of a
Riemannian metric in the noncompact and analytic situations.
If j is an inclusion, we may simply write norm X instead
of norm j .
Notice that, as the definition of the normal bundle normj
shows, the orientability of two of the three bundles j 1tangX' , tangX,
and normj implies the orientability of the third, and given
 363

orientations on two of them canonically define the orientation of the

third; see 5.2.7. Comparing this with the discussion in 4, we see that
if X* is orientable, then the orientability of normj is equivalent
to the orientability of the manifold X; moreover, if X' is oriented,
then the orientations of X1 and of normj canonically determine each
other.

8. The introduction of tangent and normal bundles allows us

to better formulate and essentially complete the main result of Sub­
section 3.4.8, i.e., Theorem 3.4.8.2.
We may sharpen the formulation of the third part of this
theorem, which asserts that the linear homomorphisms fact
x 1df1 are
actually isomorphisms. Now we can say that the maps
factd f1 combine
xi
with ab f^ : X ^ to define a linear isomorphism from the normal
bundle of the manifold X^ (taken in X^) onto the normal bundle of
the manifold X9
(taken in X 1). When X' is compact and 2 £ r £ 00,
r-1
this is a linear C -isomorphism normX^ norm X^ .

To complete Theorem 3.4.8.2, we consider orientations. Namely,

suppose that X^ , X', and X^ are orientable (oriented). Then, as an
immediate consequence of 4 and 7, the manifold^ X ^ is orientable
(respectively, canonically oriented). In particular, under the
assumptions of 3.4.8.3, the orientability of X 1, X ^ , and X2 implies
the orientability of X^ il X^, and given orientations of X', X^, and
X^ canonically orient X^ fl X ^ .

9. An n-dimensional smooth compact manifold is stably paral-

lelizable if and only if it admits a differentiable embedding in some
IR^ having a GLIR^ n -trivial normal bundle.

Let j: X 3R^ be an embedding enjoying the above property.

G I Q
Then tang X © normj is GL3R -equivalent to j tang IR , and hence the
condition is sufficient. To prove its necessity, consider an arbitrary
differentiable embedding j : X ->■lRq and the composite embedding

X — > 3Rq — — > ]Rq+n+k ,where k is large enough so that the

k n+k
suspension su tang X is GLIR -trivial (actually, it suffices to
take k =1; see 3). The normal bundle of this composite embedding is
just sun+knorm j , and it is GLIRq+k -trivial, being GLIRq+k -equivalent
to the bundle
k k k 1 q
norm j © su tang X = su (normj © tang X) = su (j'tanglR^1).
 364

The Complex Case

10. The basic definitions of this subsection, i.e., those

the tangent bundle, tang X , Riemannian metric, and normal bundles,
norm X and normj , carry over, word-for-word, to complex manifolds.
The bundles become complex bundles, the Riemannian metric is replaced
by a Hermitian one, while j is assumed to be a holomorphic immersion
(i.e., to be locally a holomorphic embedding). If f: X X' is holo­
morphic, then (df,f) is a linear map tang X -> tang X 1 . The definition
and properties of tangential maps are preserved, but they become less
universal (see 3.1.6.10). Finally, the realification of a complex
manifold (see 3.1.6.9) leads to the realification of its tangent bundle
(see 5.1.12 and 5.13), and turns Hermitian metrics into Riemannian ones.
It is impossible not to notice the eclectic character of these
definitions. The reason for this inconsistency is that, whereas the
notion of differentiable structure, which lies at the heart of the
theory of smooth vector bundles, is specifically a real notion even when
we pass to complex vector bundles, in the complex case the tangent and
normal bundles carry an additional, complex-differentiable structure -
the so-called holomorphic structure. Unfortunately, the theory of
holomorphic vector bundles is beyond the scope of this book.

5. Degree

1. In this subsection we shall apply some of the simplest

results of differential topology to homotopy theory. Namely, given any
oriented, compact, smooth manifold X, any oriented, compact, connected,
smooth manifold Y with dimY = dim X , and any continuous map
f: (X,8X) + (Y ,9Y) , we define an integer which depends only upon the
homotopy class of f. This number is called the degree of the map f
and is denoted by deg f .
Although the degree degf is a global characteristic of f,
and is actually defined for maps which are merely continuous, we shall
approach this notion by infinitesimal methods: we start by assuming that
f £ C (X,9X;Y ,3Y) f) C1 (X,Y) and choose a point y E int Y such that
f is transverse to y. Consider f 1 (y). it consists of a finite
number of points, each of them having a neighborhood which is mapped
diffeomorphically by f onto a neighborhood of y (see 3.4.8.2 and
3.1.5.5), and each of these diffeomorphisms is either orientation
 365

preserving or orientation reversing, where the neighborhoods are oriented

in agreement with the orientations of X and Y (see 4.8). The degree
of the map f at the point y, denoted deq f, is the number of the
-1 y
points of f (y) where the orientation is preserved, minus the number
-1
of points of f (y) where the orientation is reversed. A popular
shorter version of this definition is: deg^f is the algebraic number
of the preimages of the point y.

In this definition of deg^f, the assumption that y £ intY

is essential. However, one may repeat the definition for a boundary
point y, provided that
f £ C^(X,Y) n C (X ,int X ;Y , int Y) and
o
ab f : 3X ->■ 9Y is transverse to y. The degree deg f thus defined
Y
is obviously the same as deg^tabf : f (Z) Z] , where Z is the
component of dY containing y.
We add that if the degree is defined for a point y [i.e.,
either y £ intY , f £ C(X,9X;Y,3Y) H C^(X,Y), and f is transverse
to y, or y € 9Y, f € C^(X,Y) n C (X, int X; Y, int Y) , and ab f : 9X 9Y
O
is transverse to y ] ,then it is defined for any point y' in a
neighborhood of y(in Y), and deg^,f = deg^f.

2 (LEMMA) . Let X and Y be oriented,compact, C°°-manifolds,

with dim X = dimY and Y connected, and let
CO
g,h e Cg(X,Y) fl C (X, int X; Y, int Y) .

Further, let y,z € Y besuch that deg^g and degzh are defined
(see 1). If the maps

rel g ,rel h : (X,9X) -* (Y,9Y)

are homotopic, then deg_h = deg_g.

z y
PROOF. We disregard the trivialcase dimY =0 andassume
for a start that h = g. Using 1, we findneighborhoods U and V of
y and z, such that deg y .g = deg y q and degz ,h = degzh for all
y' € U and z ’ £ V. It is clear that one can join y to z by a
path whichis a C°°-embeddingI -+ Y . By Theorems 3.4.1.4 and 3.4.7.7,
there is a path s : I ->• int Y such that s(0) £ U, s(1) £ V, and
oo “1
s is a C°°-embedding transverse to g. The preimage g(s (I) ) is an
oo
oriented, compact, one-dimensionalC -submanifold ofX, and
9g~1 (s(I) ) = g_1 (s(0)) U g_1 (s(1)) (see 3.4.8.2 and 4.8). Obviously,

de9S (0)g = degs(0)abg ' degs(1)g = degs(1)abg '

where abg = [abg : g- 1 (s(I)) + s(I)]. But the contribution of a point

 366

from g 1 (s (1 )) to deg M ,abg equals the value of the orientation

S (1 ) _ ^
which this point inherits as a component of 3g (s (I ) ) . Similarly,
-1 .
the contribution of a point from g (s(0)) to deg^Q^abg is opposite
to the value of the orientation which this point inherits as a component
of 3g_ 1 (s(I)). Consequently, d e g ^ ^ g - degs ^0 ^g is the number of
points of 3g 1 (s (I) )
which inherit the orientation +1 from g **(s(I)),
-1
minus the number of points of 3g (s (I )) which inherit the orientation
i _
-1 from g (s(I)). However, this difference must be 0, because every
-1 1 1
component of g (s(I)) is diffeomorphic to either S or D (see
3.5.3.1).
Now drop the assumption that h = g, and suppose that Y is
closed. Then X
is also closed, and from the fact that g and h are
CO
homotopic it follows that there exists a C -homotopy F from g to h
(see 3.4.6.6). The map $: X x I Y x I, $(x,t) = (F(x,t),t),
obviously belongs to the intersection
oo
Cg (X x I,Y X I) n C(X x I , int (X X I);Y x I,int(Y x I)).

Furthermore, if we identify X x 0 and X x 1 with X, and Yx 0

and Y x1 with Y, then abO:Xx0-*Yx0 and ab $ : X x 1y x 1
becomeg and h, an we can write deg, nv0 = deq g
and deq , „ v0 =
(Y/0) *y (z ,1 )
= degzh. By the argument above, d e g ^ 1 )^ = deg (y 0) ^ ' and thus
degzh = deg^g.

Finally, if 3Y ^ 0, we find a component Z of 3Y and

points y 1/z 1 G Z such that ab g : 3X -*3Y is transverse to y', while
abh : 3X + 3Y is transverse to z'. We already know that

deg g = deg .g and deg h = deg ,h,

y y z
while, according to 1,

degylg = degy ) [abg: g 1 (Z) Z]

and
deg ¿A,h = deg Z [ab h : h_ 1 (Z) -* Z] .

But

deg,y . [ab g : g 1 (Z) Z]j = deg , [ab h : h 1 (Z ) + Z ] . (1 )

Indeed, since relg and relh are homotopic, h ^ (Z) = g ^ (Z), and
thus (1) follows from that part of the lemma which we have already
proved.
 367

_ oo
3 (LEMMA). For any compact C -manifolds X and Y, the set
oo
C g (X,Y) n C (X, int X;Y, int Y) is d e n s e in C (X,3X;Y,3Y) .

PROOF. C o n s t r u c t the doubles, doppX and d o p p Y , to ge th er

1 1
with two-sided collarings k: 8X * D dopp X and 1: 3Y * D + dopp Y .
OO q 00
Pick a C -embedding j : dopp Y IR * , a C -transversalization x of
j|gy/ and a neat tube Tub^p. Then it suffices, given a map
f € C (X ,8X ;Y ,9Y ) and e > 0, to find g € C*(X,Y) H C(X,i n t X ;Y ,int Y)
such that dist(j(f(x)),j(g(x))) < e for all x G X. To produce such
a g, we shall construct successively three auxiliary maps,
h^h^^h^: dopp X dopp Y .

The map h^ is very simply defined by

h 1 (x) = f (x ) , h^ (cop(x)) = cop(f (x )) [x G X] .

To construct h^/ fix 6, 0 < 6 < 1, such that:

(i) dist(jol(z,t),j°l(z,t*)) < e/4 for |t-tf| < 6 [z £ 3Y,

t,f € D 1] ;
(ii) d i s t ( j ( h 1 ( k (z ,t )) ), j ( h 1 ( k(z,t')))) < e/4 for
for 11— t 1 | < 6 [z £ 3Y, t,t' € d "*] ;
(iii) for any z £ 3Y, the bal l with center j(z) and
radius <5 lies in T u b T P, w hile its image unde the m ap
i o pr : Tub p -*■ 3Rq lies in the ball with center j (z) and radius
T T
e/4. [The e x i s t e n c e of such a 6 is a co ns e q u e n c e of the c on t i n u i t y
of j, k, 1, h^, and p r ^ .] Further, pi c k a C -map <J>: 9X 3Y such
that di s t (j (<Mz) ) ,j of (z) ) < <5 for all z £ 3X (3.4.4.2 gu ar a n t e e s
that such a ma p 4> exists), and define, for each t € I, the map
<|> : 3X -»■ 3Y by <i>t (x) = p r T (tj (<J> (x) ) +(1 -t) j°f (x) ) .Next, pick a

C°°-map a : ]R -*■ 3R such that a (t) = 0 for |t| £ 1/3,and a(t) = 1

for 111 > 2/3, an d de f i n e k^ : X X and 1^ :Y Y by

k 1 (k(z,t)) = k ( z , ( 1 - 6 ) t + 6), if z £ 3X, t £ I,

k 1 (x) = x, if x £ X ^ k( 3 X x I),
and
'l1 ( 1 (z, t) ) = l(z, (1 — 6)t + <5) , if z £ 3Y , t £ I,

1 (y) = y, if y £ Y V 1 ( 3Y x I).

It is clear that k1 and 11 are to po l o g i c a l embeddings.

Now define h2 by
 368

h2 (k(z,t)) = Kcfj^t) (z) ,t) , if z G 9X, |t| < 6,

h2 (x) = 11 (f (k~1 (x) )) ,

if x £ X ^ k (9X x [0 ,6) )
h2 (cop(x)) = cop(11 (f (k11 (x)))

Then the following facts are evident:

h2 € C (dopp X, int X, 9X?dopp Y, int Y, 9Y) ; dist (j (x) ,j °h2 (x) ) < e/2

for all x € doppX ; the restriction of h2 to k(9x * [-6/3,6/3])

OO
is ofclass C ; and if z G 3X, then

I m d z (h2|k(9X x [-6/3 ,6/3] )} ^ Tangh2 (z) 8Y‘

CO . . .
Finally, take h^ to be any C -mapdoppX dopp Y which
equals h2 on k(3X x [-5/6,6/6] and enjoys theproperties:

h3 (X ^ k O X x [0,6/6)) c int Y ;

h^ (cop (X ^ k O X x [0,6/6))) c= cop (int Y ) ;

and
dist(j°h3 (x),j°h2 (x)) < e/2 for all x G X.

[Theorem 3.4.4.8 guarantees that such a h^ exists.]

Clearly, h^(X) <= Y. Now define the desired map g by
g = [ab h^ : X Y],and check directly that it has all the necessary
properties.

4. Let X and Y be (as in 1) oriented, compact, smooth

manifolds with dim X = dim Y , and let f € C (X, 3X;Y , 3Y) . Restrict
the differentiable structures of the manifolds X and Y to
CO oo
C -structures (see 3.4.9.8), and find a map
g G C.(X,Y) n
0
D C (X,int X ;Y, int Y) which is close enough to f in the C -topology,
and such that the maps f,g: X Y, as well as the maps
ab f , ab g : 3 X ^ 3 Y are homotopic (see 3, 3.4.5.10, and 1.3.6.6.) . Now
compute deg^g at some point y G int Y such that g is transverse
to y. By Lemma 2, deg^g does not depend upon the choice of g or
y, while3.4.6.10 and 3.4.1.6. show that deg^g does not depend upon
the modality of restricting the differentiable structures of X and Y
oo
to C -structures (according to the aforementioned theorems, the
°°
C -manifolds resulting from the restriction of the differentiable
structure of a given compact, smooth manifold, effected in two distinct
00 0
ways, are C -diffeomorphic via a diffeomorphism which can be as C -close
to the identity diffeomorphism as we choose). We call deg g the
degree of the map f , denoted deg f .
 369

The main properties of the degree are immediate consequences

of its definition and of Lemmas 2 and 3. We listhere some of them:

(i) if f,f': (X,3X) -+ (Y, 3Y) are homotopic, then deg f =

deg f 1 (by Lemma 2) ;
f g
(ii) the degree of the composite map (X,3X) + (Y,3Y) -> (Z,3Z)
is deg f • deg g (by the definition of deg) ;
(iii) the degree of the identity map is 1 (trivial);

(iv) if f: (X,3X) (Y,3Y) is a homotopy equivalence, then

deg f = 1 (proof: if g is a homotopy inverse of f, then
deg f •deg g = deg (g o f )= deg id(X,3X) = 1) ;

(v) if f:(X,3X) (Y ,3Y ) is such that f(X) t Y,then

deg f = 0 (indeed, one can approximate f as closely asdesired by a
map from C^(x,Y) 0 C(X,int X; Y, int Y) enjoying the same properties);

(vi) if Z is any component of Y,then

deg [f: (X,3X) + (Y,3Y)] = deg tab f : f~1 (Z) ->■ Z]

(a result of the discussion in 1) and,in particular, degf = 0

whenever X is closed but Y is not.

As examples, consider the maps f: (Dn/,Sn ^) (Dn ,Sn ^) and

n“ 1 nw1
ab f : S S ,defined by an orthogonal (nxn)-matrix V (n ^ 2) .
Obviously, deg f = deg ab f = det V , i.e., deg f = deg ab f = 1 if
V € SO(n), anddeg f = deg ab f = -1 if V £ 0(n) ^ SO(n). Thus, the
degree of the antipodal map Sn ^ Sn \ x -xf equals 1 if n is
even and -1 if n is odd.

The Nonoriented Case

5. The discussion in 1, 2, and 3 can be carried over to n

oriented manifolds if we replace integers by integers modulo 2. This
enables us to define deg f € for any continuous map
f: (X, 3X) (Y ,3Y ) , where X and Y are smooth, compact manifolds,
and Y is connected (no orientability needed). All the properties of
the integral degree listed in 4 are preserved. For the case of oriented
manifolds, when both degrees (the integral and mod 2) are defined, we
continue to use the same notation for both, because misunderstandings
are usually eliminated by the context.
 370

Applications

6. Smooth closed manifolds of positive dimension are not

contractible.
This is plain if the given manifold is not connected. In the
connected case, the identity map of a closed manifold is 1, whereas the
degree of any map which takes the whole manifold into one of its points
is zero (here we use the -degree defined in 5).

7. If n i m, then Sn and Sm are not homotopy equivalent.

Indeed, if m < n, then every continuous map Sm + SD is

homotopic to a constant map (see 2.3.2.3 and 2.3.1.6), whereas
id: Sn Sn is not homotopic to a constant map (see 6).

8. The boundary of a nonempty, compact, smooth manifold is

not a retract of the manifold.
It suffices to assume that the given manifold X is compact,
smooth, connected, and with 9X / 0. Let p: X 3X be a retraction,
and let Z be any component of 9X. Consider the composite map

X — X > X. Since its image is not all of X, its degree is 0

(see 4 and 5). On the other hand, this degree equals the degree of
ab(in ° p) : Z -* Z, which is 1, the last map being id Z .

9. Every continuous map Dn Dn has a fixed point.

PROOF. Suppose that f :Dn + Dn is continuous and has no

fixed points. Then the map Dn Sn ^ taking each point x E Dn into
its projection on Sn ^ from the point f(x) is a retraction, and
hence Sn ^ is a retract of Dn : contradiction (see Theorem 8).

10. If an m-dimensional locally Euclidean space is homeo-

morphic to an n-dimensional locally Euclidean space, then n = m.
(Cf. 3.1 .1 .4) .

PROOF. Every point of 3Rq can be covered (in ]Rq ) by a

Euclidean q-simplex. Therefore, every point of a q-dimensional locally
Euclidean space lies in the interior of a finitely-triangulated subset,
and its link in this subset is homeomorphic to Sg_1. By Theorem
2.2.6.4, this link is a homotopy invariant, and 7 shows that the
spheres S and S cannot have the same homotopy type unless m = n.

11. Theorem 10 clarifies not only the definition of a locally

Euclidean space, but also that of a cellular space. Namely, it shows
that the dimension of a cell is uniquely determined by this cell.
 371

Therefore, the dimension function which we introduced into the definition

of the cellular decomposition as an additional element of its structure,
is actually redundant, being completely determined by the decomposition
itself.
n ri“ 1
12. The boundary of the half space ]R_ _is 3R^

(Cf. 3.1.1.4.)

PROOF. It suffices to show that the point 0 has in 3R^ no

neighborhood homeomorphic to 3Rn ; see 3.1.1.4.
Assume that such a neighborhood exists. Then 0 is an
interior point of a finitely-triangulated subset of this neighborhood,
where its link is homeomorphic to S (cf. the proof of Theorem 10).
On the other hand, 0 is an interior point of a finitely-triangulated
subset, where its link is homeomorphic to Dn ^: take any Euclidean
n-simplex which lies in 3R^ and contains 0 in the interior of one
of its (n-l)-faces. Since Sn ^ is not contractible, whereas Dn ^ is,
we contradict Theorem 2. 2. 6. 4.

6. Exercises

1. Let r £ 00, and let f : X -> Y be a Cr-submersion, where

X and Y are Cr-manifolds, X compact and Y closed. Show that
(X,f ,Y ) is a Cr-bundle. (Combined with Theorem 1.3, this result shows
that for r < °°, (X,f ,Y )is a Cr-bundle whenever X is a compact
Cr-manifold and Y is a Cr-manifold, while f: X ■+Y is a
C -submersion.)
2. Let 1 £ r £ °°, and let E, be a Cr-bundle withclosed
base. Show that there isa collaring k: 9tl E, x I -* tl E, such that
k(z x I) c pr £ (pr £ (z)) for every point z £ 9t 1 E. .

3. Let 1 <r £ and let E, be a Cr-bundle with

9tl E, = pr (9bs E,) •Show that there are collarings k: 3bs E, x I -> bs £
and 1: 9tl E, x I -* tl E, such that the diagram

atl E, X I --- 1-- * tl E,

ab pr E, x id I pr K

9bs K x I ------ *■ bs K

commutes.
 372

>r
4. Show that if r £ °°, then for every C ' -bundle £ with
compact bs £ and tl F3, Secr£ is dense in SecS£ for any s < r.
(This generalizes Theorem 2.7 for r f a.)
>r
5. Show that every C ' -bundle £ with compact bs £ and
r 00 _.
tl£ is C -isomorphic to a C -bundle (cf. 2.8).

6. Show that su tang ]RPn is Ca-GLIRn +^- equivalent to the

sum of n+1 copies of Gra (n+1 ,GL (1 ,3R) ) , while su tang CCPn is
Ca-GLCCn+^-equivalent to the sum of n+1 copies of Gra (n+1 ,GL (1 ,CC) ) .

7. Show that the normal bundle of the Ca-embedding

G(m,n) -+G(m+1,n), described in 3.2.2.3, is Ca-GLIRn -equivalent to
Gra(m,GL(n, 3R) ) , while the normal bundle of the Ca-embedding
(CG(m,n) -+ CG(m+1,n), described in 3.2.2.1 , is Ca-GL(Cn-equivalent to
Gra (m, GL (n ,CC) ) .

8. Let P'l ' ***'Pn + 1 ke homogeneous complex polynomials of

degree m in n+1 variables, whose only common zero is the point 0.
Show that the map (EPn CCPn given by

* ,P1 ( Z 1 .....2 n +1 > " ■ • :pn + 1 < Z 1 .....!„ . l "

has degree m11.

9. Show that for n ^ 1 every continuous map Sn -> Sn whose

degree is not (-1)n+ has a fixed point.

10. Show that for n ^ 1 every continuous map Sn -> SR

having odd degree transforms some pair of antipodal points into another
such pair.

11 . Show that for odd n > 1 the degree of any map Sn -+ IRpn
is even.

12. Let f be a simplicial map of the standard 2-simplex

onto the standard 1-simplex. Show that the symplicial mapping cylinder,
Scylf , is not homeomorphic to Cyl f .
 Chapter 5. Homotopy Groups

§1. THE GENERAL THEORY

1. Absolute Homotopy Groups

1. Let (X,Xq ) be a pointed space, and let r 5 0 be an

integer. To simplify the notation, let us agree to write Sphr (X,xQ )
27 x 2T r
for the set C (I ,FrI ;X,xQ) of all continuous maps (I ,Fr I ) (X,Xq

and denote theset of homotopy classes of such maps (i.e.,

r r
tt ( I ,Fr I ; X ,xq ) ) by 7Tr (X,xQ). The elements of Sphr (X,xQ ) will be
referred to as r-dimensional spheroids (or simply r-spheroids) of the
space X with origin x^.

For r > 0 and two arbitrary spheroids (p,\p € Sph (X,Xg), we

define their product, (pip, as the spheroid in Sph^iX^^) given by

<p (2t1 ,t2 ,... ,t^) , if 0 £ t 1 £ 1/2,

(1 )
i(;(2t1-1,t2 ,...,tr ) , if 1/2 $ t1 ^ 1.

-1
For r > 0 and <p £ Sph^(X,XQ), the spheroid <p , called
- 1
the inverse of <p , is defined by <p (t ^ ,t^ ,...,t^) = <p ( 1-t ^ ,t^ /.../1^)

Obviously, if £ Sphr (X,xQ ) are such that <p^ is homotopic

to <p and is homotopic to ip, then the spheroids and M
are homotopic. Therefore, (1) defines a multiplication on t\ ^ (X ,X q ) .
It turns out that this multiplication is associative, that the homotopy
27
class of the constant spheroid const (which takes I into x^) is
a two-sided identity element and that the homotopy classes of jthe
spheroids (p and <p are inverses of one another.
 374

The associativity of the multiplication means that the products

(M)X and <M^X) are homotopic for any spheroids e sPhr (x /x0 )-
Indeed, the formula

((t1 ,tr) ,t) ^

1+1
— ,t2 ,...,tr if 0 * t1 £ ~ ^ r

lf _1+t
_ ,^ tl <„ —2+t (2 )

X if ^ « t 1 £ 1,

defines a homotopy I x I -+ X from (cM)x to <M^x)-

To prove the second claim, we have to show that the products
(J)(const) and (const) (J) are both homotopic to $, for any
$ £ Sphr (X,XQ). The formulas

((t1,t2 ,...,tr ),t) »

2ti
—— t" t" if 0 £ t_ s? 1+t
1+t 2' ' *•' r
(3)
if i t, s 1,
V ¥

and

((t1 ,t2 ,...,tr > ,t) H-

1-t
if 0 £ t„ £
xo ' 1 v 2 '
(4)
2t -1+t
1-t

define homotopies from (j)(const) and (const) cj) to <j).

_i
Finally, the third claim is that the products <jxj) and
~1 _i
4) (j) are both homotopic to const: indeed, a homotopy from $$ to
const is given by

((t1 /12 /•••/tr ),t)

(2 t 1 ,t2 ,.. w tr ), if 0 * t1 <
(1 —tj,12 ,.•* r if < t, c [5)

(2-2t1,t2 ,.../t ), if 1+t

2 * fc1 « 1*

The set tt^(X,Xq ), r > 0, with this group structure is

 375

called the r-th homotopy group of the space X at the point x^ .

If r > 0, then each r-spheroid maps I into the component

Xq of X containingx ^ . Consequently, for r > 0 the groups
TTr (x 'x q ) an<^ are isomorPhic *
By Theorem 2.3.4.3, for a countable cellular space X all
the sets 7Tr (X,x^) are countable.

The Case r = 0

2. Since 1° is a point and Fr 1°= 0, Sph0 (X,xQ ) and

7Tq (X,Xq ) can be identified with X and with the setcomp X of
components of X, respectively. tt0 (X,Xq ) has no natural group
structure. However, it does have a distinguished element which, in
analogy with the higher-dimensional case, will be referred to as an
identity: this is the homotopy class of the 0-spheroid const, i.e.,
the component of X containing x^.
To be able to use the same language for the cases r > 0 and
r = 0,we shall call tTq(X,Xq) the 0-th homotopy group of X at x^,
and we shall apply the group-theoretic terminology to sets with
distinguished elements and their maps. In particular, by a direct
product we understand the usual product, a homomorphism is a map
preserving distinguished elements, the kernel of a homomorphism is the
preimage of the distinguished element, and an isomorphism is an
invertible homomorphism.

The Case r = 1

3. One-dimensional spheroids are nothing else but closed

paths, and the multiplication, inversion, and homotopy of spheroids, as
defined in 1, coincide with the multiplication, inversion, and homotopy
of paths, as defined in 1.3.2.1 and 1.3.2.3. The 1-st homotopy group is
alternatively known as the fundamental group. It was defined some
decades before the higher homotopy groups were introduced, and we shall
see below that it holds a special position amongst the homotopy groups.

4. When r = 1 the homotopies (2)-(5) are defined not only

for loops with common origin: the only condition that the paths
4),^,X must satisfy is that the products involved be meaningful. As in
the case of loops, the homotopy class of a product is uniquely
 376

determined by the homotopy classes of its factors, provided that the

origin of the paths in the second class coincides with the end of the
paths in the first class. This multiplication is associative; the class
of the constant path is a left identity element for the class of paths
with the same origin, and a right identity element for the class of paths
-1
with the same end; and the classes of the paths s and s are
inverses of one another, i.e., their product, taken in any order, is
homotopic with the corresponding identity element.

The Case r > 1

5. For r > 1 the group tt^(X,Xq ) is Abeli<

Lan.

We have to verify that the products (p\p and ¡¡j(p are homotopic
for any spheroids <j),ip £ Sph (X,Xq ), r > 1. Consider the following
r
three homotopies I x I X:

(2t1, (1+t)t2 ,t^,...,t^]

( 2 t l ~1 , (1+t) t 2 - t , t 3 , . . . , t r ) , if J sc t 1 « 1, i: t 2 ^ 1,

0' otherwise;

((t1 ,t2 ,t3 ,...,tr >,t)

*l>(2t1+t-1 ,2t2-1 ,t3 , ...,tr ) , if ~ < t1 <

O' otherwise

and

( ( t 1 , t 2 , t 3 , . . . , t r ) ,t) H-

(2t ^-1, (2-t)t2 ,t3 ,...,t ) , if t, S 1, o s t2 < j L ,

i|j ( 2 t 1 , ( 2 - t ) t 2 + t - 1 , t 3 , . . . , t r ) , if 0 ^ t1 £ j ,

0' otherwise.

Their successive product is a homotopy from (frijj to ipcj).

[These homotopies are pictured in Fig. 13, where the shaded regions
are mapped into xQ .]
 377

Second homotopy

wm

?
9 lfP 9 m
? i 9
B 9
y y y / ?
W //r
9
iü \ 9 w !Ü1
F ir s t hom otopy Third hom otopy

Fig. 13 (r = 2)

Behavior Under Continuous Maps

6. Let f: (X,Xq ) (X',Xg) be a continuous map of pointed

spaces. Then toeachspheroid $: (ir ,Frir ) (X,x^)there corresponds
the spheroid f o $: (ir ,Fr ir ) (X',Xg). This defines a map

= f #r : Sphr (X,x0) -> Sphr (X *>Xq ) ,

and clearly f^ takes homotopic spheroids into homotopic ones, and takes
the constant spheroid into the constant one. Moreover, f„ (M) =
ttr
= f n (ijj) for r > 0. Therefore, f„ defines a homomorphism
fr r tf r fipr
t\ ^ (X ,X q ) tt^X^Xq) for each r > 0, called the homomorphism induced

by the map f, and denoted f* or, more specifically, f*r «

7. For any two continuous maps, f: (X,Xq ) -> (Y,y^) and

g: (Y ,y Q) + (Z,zQ), and any r > 0,

<g ° f )*r = g*r ° f *r -

Tf f = id (X ,Xq ) , then f*r = idTT^jX/Xg).

PROOF. (g ° f)#r = g^r ° f#r and id(X,xQ )^r = id Sphr (X,xQ) .

8. If the continuous maps f,f*: (X,xQ) ■> (Y,yQ) are

homotopic, then f* = f^r for all r. If f: (X,xQ) (Y,yQ ) is a
homotopy equivalence, thenf* isan isomorphism for all r.

PROOF. The spheroids f ° cf> and f' ° § are homotopic for

any cj) € Sphr (X,xQ), which proves the first assertion. The second
assertion follows from the equalities g*r ° f*r = (g ° f)*r = id an^

f*r ° = (f ° g)*r = i d , where g is any homotopy inverse of f

(see 7) .
 378

A Multiplication Theorem

9. Let (X ,X q ) and (Y,yQ ) be arbitrary pointed spaces.

Then for any r ^ 0 the homotopy group tt^(X x Y,(xQ ,y0 )) is canonical­
ly isomorphic to the direct product tt (X,Xg) * TT^(Y,y0 ). The canonical
isomorphism tt^ (X x Y,(xQ ,y0 )) •+ TTr (X,xQ ) x TT^(Y,y0 ) is given by
a *+ (pr1* (a) ,pr2* (a) ) . If (X',x^) and (Y*^) is another pair of
pointed spaces and f: (X,Xq) ->■ (X',Xq) and g: (Y,yg) (Y',yg) are
continuous, then the diagram

7Tr (X x y, (x Q ,y Q ) )
TTr (X'X0 ) X TTr (Y,Y0 :

(f x g)

TTr ( X ’ X Y *, ( x ^ y ^ ) ) TTr (X *,x^) x TIr (Y ’,y'Q,

commutes (the horizontal maps are the canonical homomorphisms)

The proof is immediate.

2. A Digression: Local Systems

1. We say that on the topological space X there is given a

local system of groups if for each point x £ X there is a group G ,
and for each path s: I X there is a homomorphism T : G G ,„ * ,
* s s (0) s (1 ) '
such that three conditions are satisfied:

(i) if s. (0) =s(1), then T = T o t ;

1 ss^ s^ s
(ii) if s is a constant path, then T is the identical
automorphism of g s (q );

(iii) if s and s
are homotopic paths, then T = T
I s s
Condition(iii) shows that we may write T instead of T ,
a s
where a is the homotopy class of the path s. Moreover, from (i)-(iii)
it follows that all the homomorphisms T (T ) are actually isomor-
1 — s o ._

phisms,and that Tg = T_1 (respectively, T = T . ) : indeed, using

s a a
“1 -1
1.4, the paths ss and s s are homotopic to a constant path, and

hence T . Ts - T = idGs and Ts o T = T _ , id G .

b SS s s s
 379

The isomorphism Tg is called the translation along s.

2. In particular, if s is a loop with origin x or,

equivalently, if a is an element of the fundamental group tt^(X,x ),
then Tg = T^ is an automorphism of G^. Comparing this with
conditions (i) and (ii) in 1, we see that the rule a T defines a
o
right group-action of tt^(X,x ) on G^.

To each path s: I -+ X there correponds a natural isomorphism

tg : ir1 (X, s (0) ) -*■ tt1 (X, s (1) ) , g iven by = o cdcy, where o is the
homotopy class of s; sometimes we denote t
by t . One can check
s a
directly that T s : Gs(0) ^ Gs (1 ) is a ta"maP <see 4. 2. 3.1).

3. Let(X ,{G } ,{T }) and (X 1,{G 1} ,{T 1,}) be local systems

X S X s
of groups, given on two spaces, X and X 1, and let f: X + X* be
continuous. Let us assume further that for each point x€ X we are
given a homomorphism h : G -* G* .. We say that the homomorphisms
X X 1\X)
h^ and the map f form a homomorphism of the first local system into
the second if h,„ % ° T = TJ. ° h for any path s: I •* X.
---------- s(1) s f°s s(0) 2 ^
A homomorphism (f ,{h^}) is an isomorphism if f is a homeomorphism

and all h^ are isomorphisms; (f,{h }) is an equivalence if it is an

isomorphism and, in addition, X* = X and f = id X .

If (X1,{G^},{T ^ ,}) is a local system of groups and f: X + X 1

is continuous, then the induced local system (X,{G^},{Ts>) arises on

X: set G
= G* » and T = T* . Obviously, (f,{idG }) is a
X 1 (X ) S 1®S X
homomorphism of the induced local system into the original one.

4. Let X be a connected space with base point x^. Two

local systems of groups, (X ,{G^} ,{Tg}) and (X,{G^} ,{T^,}), are
equivalent if and only if the two corresponding actions of tt^(X,Xq )
on G_ and G/_ are isomorphic, i.e., if and only if there i s a _group
xo xo
isomorphism G„ + G¿ which is also a tt„ (X,x^) -map.
x0 0
That the actions of ir1 (X,xn) on G and G' arising
1 u x0 x0
from equivalent local systems are isomorphic is obvious. To prove the
converse, fix a tt (X,x )-isomorphism h: G -* G' and choose, for
1 0 x0 0
each x e X, some path s^ with origin xQ and end x. It is readily

verified that (id,{h }), where h =T' 0 h ° T 1, is an equivalence.

x x sx sx

5. A local system of groups on a topological space X is

said to be simple if it is equivalent to a canonical simple local system
 380

(X,{G },{T }), where allthe G are equal to some fixed group G,
X s x
and all the homomorphisms T are the identical automorphism of G.
By Theorem 4, a local system of groups on a connected
topological space X with base point x^ is simple if and only if the
induced action of tt (X,xn ) on G
is the identical action. In
x0
particular, a local system is simple whenever (X,xQ ) is trivial or
the groupsare all isomorphic to 7L^ .

6. It is readily seen that the discussion above may be

extended from local systems of groups to local systems of other algebrai
objects, such as vector spaces or rings.
Inthe present section we shall encounter, in addition to
local systems of groups, local systems of sets with an identity (a
distinguished element).

3. Local Systems of
Homotopy Groups of a Topological Space

1. Two spheroids, £ Sphr (X,xQ) and d> £ Sphr (X,x^),

are said to be freely homotopic if the maps abs (J)^,abs <t> : Ir X can
be connected bya homotopy consisting only of spheroids. More precisely
(j)0and (j) are freely homotopic if there is a continuous map
r r
h: I x I -+ X, constant on each set Fr I *t (t £ I) , and such that
h (y,0) = <j)Q (y) , h (y,1) = cj)1 (y) for all y e ir .

An essential element of such a homotopy is the path described

by the origin of the spheroid, i.e., t *+ h(Frir x t) . We say that h
is a free homotopy connecting the spheroids <J)q and $ along this
oath.

2. Every spheroid with origin x^ admits a free homotopy

along any path with origin xQ . Free homotopies of homotopic spheroids
along homotopic paths produce homotopic spheroids.

Let <j) £ Sphr (X,x^), and let s be a path with s (0) = x ^ .

To exhibit a free homotopy of d) along s, is is enough to extend
somehow the homotopy (y,t) ^ s(t) [y £ Fr ir , t £ I] of the constant
map r to a homotopy of 4>: lr -*■ X. (That such an extension
Fr I
exists follows from 2.3.1.3.)
To prove the second claim, let <J> and be homotopic
spheroids, and let h,h': ir x i x be free homotopies of d> and

'0 along the homotopic paths s and s'. To show that the spheroids
 381

and <j)^ are homo topic, where <f>1 (11 ,...,t^) = h ((t ,...,tr ) ,t )
and <J>^ (t^ ,. . .,t^) = h ' ((t^ , .. .,t^) ,t) , pick some homotopies,
rr
f: I x I + X, from (pQ to (p^, and g: I x I X, from s to s 1,

and define a subset K of the cube lr+2 = ir x i x i and a map

H: K X, as follows:

K = Fr r +2Ir +2 ^ t(Int r+1lr+1) x U r

3R ®r 1
and
f (y ,u) , if V = 0 ,

g(u,v), if y £ Fr Ir
IRr
H(y,u ,v) =
h(y,v ) , if u = 0,
h ' (y ,v) , if u = 1
JO
(where y E I , u E I,v E I). There exists a homeomorphism
k: Ir x i -> k such that k(y,u) = (y,u,1) for all (y,u) E Fr(ir x i);
for example, take the inverse of the homeomorphism k ^ : K -*ir x x,

k 1 (y,u,v) = ( y Q , j ) + ^-(1+v) (y-y0 ,u--l) ,

wherey = (ort + ... + ort )/2. Now it is clear that

r
H°k:I x i x is a homotopy from <() to ' .

3. According to the previous theorem, the free homotopies

along a path s: I -* X define a map T : tt (X,s(0)) -> tt (X,s (1) ) for
any r ^ 0. The same theorem demonstrates that the maps Tg fulfil
property 2.1(iii), and it is obvious that they enjoy also the properties
2.1 (i), 2.1 (ii), and are homomorphisms. The resulting local system,
(X,{ (X ,x ) } ,{Tg} ) , is a local system of groups for any r ^ 1, and
a local system of sets with distinguished elements for r = 0;
(X, (X,x) } ,{T^} ) is called the local system of the r-th homotopy
groups of X. In particular, tt^ (X,x ) acts naturally from the right
on tt(X,x) , for any x E X and r ^ 1.

4. If r = 1, then the isomorphism T_ acts by therule

-1 s
Tg =a coa, where a is the homotopy class of the path s (i.e.,
Tg coincides with thehomomorphism tg from 2.2). In particular, the
right action of tt ^ ( X , x ) on (X,x) is the inner right action.

PROOF. Let w be any loop in the class w, and define a

path st : I -* X, t E I, by s^iy) = s(ty). Consider the loop
i
—
= (st w)st> Since Sq is the constant path, belongs to w,

and the formula (y,t)wfc(y) defines a free homotopy I x I -* x from

 382

wQ to the loop w 1 = (s 1w)s along s; w^ belongs to the class

-1
0 0)0 .

5. As 3 shows, for arbitrary fixed r all the homotopy

groups tt^(X,Xq ) of a connected topological space X are isomorphic.
For r = 1, this was already a corollary of 2.2.
A space X is r-simple if it is connected and the local
system of its r-th homotopy groups is simple. In this case, the groups
tt^ (X ,x ) are not only isomorphic, but are manifestly canonically
isomorphic, and hence they may be identified with a unique group, tt^ (X ),
referred to as the r-th homotopy group of X without base point. The
elements of tt^(X) are classes of freely homotopic spheroids. A space
is simple if it is r-simple for all r.
If X is not r-simple, then one cannot use the isomorphisms
Tg to identify the groups tt^(X,x ) with different x. In this
situation one can speak of the group tt^ (X) of X only as an abstract
group.
Obviously, the local system of the 0-th homotopy groups of a
topological space is always simple, and for connectedspaces, it becomes
a local system of sets, each reduced to one point.
According to 4, a space is 1-simple if and only if it is
connected and its fundamental group is Abelian (see 2.5) .

6. As 1.6 shows, every continuous map f: X X1 induces a

a homomorphism f = (f*) : tt (X,x) tt (X’,f(x)), for any x E X. If
x r IT
h is a free homotopy from the spheroid ^ to the spheroid $ along
the path s, then f oh is a free homotopy from f ° <J)q to f ° $^
along the path f o s, and so (f*)g(1) ° Ts = Tfos o (f*)g(0). Thus,
given any r ^ 0,the induced homomorphisms (f. ) combine with f
r X
to define a homomorphism of the local system of the r—th homotopy groups
of X into the local system of the r-th homotopy groups of X ’.
Three special cases deserve to be mentioned: (a) X is
r-simple; (b) X' is r-simple; and (c) X and X' are both r-simple.
In case (a), the homomorphisms )
( f * take the same group, tt (X),
r X 17
into (X1,f (x) ) , x E X, and for any path s : I X the diagram

TTr (X )
 383

commutes. In case (b) , the homomorphisms ^ * r ^x maP the groups

7Tr (x /x )/ x € X, into the same group, tt (X1), and for any path
s: I -+ X the diagram

commutes.Finally, in case (c), the local systems of the r-th homotopy

groups of X and X1 reduce to two groups, tt^ X ) and tt^ X 1), and
the homomorphisms (f*r ^x become one homomorphism f*r : nr (X) -> Ti^iX1).

7. IjE f: X -> X 1 is a homotopy equivalence, then all induced

homomorphisms (f*_)„: tt_(X,x ) tt_ (X 1,f (x) ) are isomorphisms.
r x r r
PROOF. Let f 1: X 1 -+■ X be a homotopy inverse of f. If
H: X x I X is a homotopy from f' ° f to id X , then given any
spheroid cf) € Sphr (X,x), the map T ° x I I , (y,t) +* H(c})(y) ,t) , is
a free homotopy from (j) to the spheroid f 1 o f o $ £ Sph^. (X, f 1of (x) ),
along the path s, s(t) = H(x,t). Therefore, the homomorphism
(fi)f(x, ° (f*)x = H f ' ° f )*)x : (X,x) -> TTr (X, f 1of (X) ) is simply the
translation T g ; in particular, it is an isomorphism, implying that
(f*)f(x) iSan ePimorPhism. On the other hand, ^*^f'of(x) ° ^*^f(x)
is also an isomorphism, and so ^ * ^f(x ) a monomorphism. We conclude
that <f*)f(x )an isomorPhism, and hence so is ^*^x = ^*^f(x)^ ^ ° Ts*
8. Let (X,x^) bea pointed topological space, and let
0 £ k < °°. The homotopy groups tt^(X,Xq ) are trivial for all r £ k
if and only if X is k-connected. The homotopy groups Trr (X,Xg) are
all trivial if and only if X is °°-connected.

If X is k-connected, then tt^(X,Xq ) is trivial for all

r £ k: to see this, compare the definition of tt (X,x ) and that of
r +1 r + 1
k-connectedness (see 1.3.3.7; in 1.3.3.6 one can replace (D ,S )
r +1 r +1
by the homeomorphic pair (I ,Fr I )) . The same two definitions
prove the converse statement,since the triviality of TT^iXfXg) for all
r £ k implies the triviality of trr (X,x) for all r £ k and any
x € X (see 1.2 and 5).
 384

4. R e lat iv e H o m o to py Grou ps

1. Set J r_1 Ir = Fr
Int • G i v e n any t o p o l o g i c a l
IRr 3Rr
pair (X,A) w ith base p o int xQ € A and any p o si ti ve i n t eg er r, we

let Sph^ (X,A , X q ) denote the set C ( l r , F r i r ,Jr 1 ; X ,A ,x Q ) of all

co nt in uo us maps (ir ,Fr ir ,Jr _ 1 ) + (X,A,xQ ). The e l e m e n t s of

Sph (X,A,x ) are ca ll ed r- d i m e n s i o n a l spheroids (or r - s p h e r o i d s ) w i t h
r o r r r—1
origin X q of the pair (X,A) . The set tt (I , Fr I ,J ; X,A ,x Q ) of
h o mo to py classes of such spheroids is simply d e n ot ed by tt^ ( X , A , X q ).
Noti ce that every spheroid cf> G Sph^ (X,A ,X q ), such that
(p (ir ) <= A , is ho mo t o p i c to the const an t spheroid. In fact, there is
even a standard h o m ot op y i r x I -*■ I from $ to the c o n s t a n t spheroid:

((tr . . w t r _ r t r ),t) » (tr . . w tr _ r (1-t)tr + t).

A 1- sp he ro id with o rigin Xq of the pair (X,A) is simp ly

a path with ori gi n in A and end X q . Warning: a homotopy of such a
spheroid is stat io nar y at the point 1, but if A is not re d u c e d to
Xq , it is not n e c e s s a r i l y s t a t i on ary at the point 0.
Wh en r > 2, formula (1) defines a m u l t i p l i c a t i o n on
Sph^ (X, A , x q ) , and this induces a m u l t i p l i c a t i o n on 7t^ ( x ,A,X q ), which
turns 7Tr (X,A,xQ ) into a group. The i dentity e l e m ent of tt^ ( X , A , X q )
is the ho m o t o p y class of the c o n st an t spheroid, w h i l e the class of the
-1 -1
spheroid <J> , w ith cp (t ^ ,t^ , • . • ,t ^ ) = (p (1 -t ^ ,t^ , . . . ,t ^ ) ,is the
inverse of the class of <j>. The p roof of these a s s e r t i o n s is e n t i r e l y
analogous to that given in the case of absolu te h o m o t o p y g r o u p s .
For r ^ 2, tt^ ( X , A , X q ) is called the r -t h h o m o t o p y g r o u p of
the pair (X,A) at the point Xq . The first h o m o t o p y gr o u p of (X,A)
at Xq is d e f in ed to be the set 7t ^ ( X , A , X q ) wi t h an i d e n t i t y (a
di s t i n g u s h e d e l e m e n t ) , na mel y the h o m o t o p y class of the c o n s t a n t
1- s p h e r o i d .
If A = Xq , then tt^ ( X , A , X q ) equals tt^ ( X , X q )
(i.e., tt^ ( X , X q ,X q ) and tt^ ( X , X q ) co in cid e as group s for r ^ 2, and
as sets wit h d i s t i n g u s h e d el eme nt s for r = 1).

For r > 1, tt^ ( X , A , X q ) isc a n o n i c a l l y i so mo r p h i c to

tt^ (Xq ,A q ,X q ) ,where XQ and AQ are the c o m p o n e n t s of X and A
which c o n t a i n xQ.
For r > 2, tt^_(X,A,X q ) is Abelian; the proo f e n t a i l s obvious
modifications of the proof of T h e o r e m 1.5.
 385

2. Every continuous map f: (X,A,xQ ) + (X',A',Xq ) yields

the induced homomorphism f*: tt^ (X/A,xq) tt^ (X 1 ,A 1 ,x ^ ) , r > 1,
defined as in the absolute case. If A = xQ and A' = x ^ , we recover
the absolute induced homomorphism, f *: tt^ ( X , x 0) + tt (X',Xq). As with
the absolute case, <9 ° f )* = g* ° f* and id* = id. If f and f'
are homotopic, then f* = f*. If f is a homotopy equivalence, then
f* is an isomorphism.

The Boundary Homomorphism

3. Given a spheroid (J) € Sph (X,A,x ), its compression,

r- 1 r- 1 r
ab (J) : (I ,Fr I ) (A,Xq ), is a spheroid belonging to Sph^_^ (A,x^) ,
called the boundary of (j), and denoted 3(f). The resulting map,
3: Sphr (X,A,XQ) -5- Sph^_^ (A,Xq ) , takes homotopic spheroids into homotopic
ones, takes the sum of two spheroids into the sum of their boundaries,
and takes the constant spheroid into the constant one. Therefore, it
defines, for every r ^ 1, a homomorphism 3: 7t^(X,A,xq ) 7Tr_^ (A,xQ ) ,
called the boundary homomorphism.

Given any continuous map, f: (X,A,Xq ) (X'^A^x^), the

diagram

11r (X, A,x q ) ----- ► 7Tr_ 1 (A,x0 )

f* (ab f) *

7Tr (X' ,A' , X ^ ) ----- > 7Tr_ 1 (A '/X q )

commutes for any r ^ 1. Indeed, we already know that the similar

diagram with Sph instead of tt , and with f^ and (ab f)^ instead
of f* and (ab f)* , commutes.

Local Systems of
Homotopy Groups of a Topological Pair

4. Two speroids, <J)q £ Sph^(X,A,XQ) and € Sph^ (X,A,x^ ) ,

are said to be freely homotopic if there is a homotopy from <j>0 to <|>
consisting only of spheroids, where (J)q and (J) are viewed as maps
(ir ,Fr Ir ) -»■ (X,A) . In other words, <j>Q and <j>1 are freely homotopic
if there is a map h: I x I I such that h(Fr I x I) cz a, h is
y*_ 1
constant on every set J x t (t € I), and h(y,0) = ^(y),
 386

h (y,1) = <J> (y) for all y e Ir .

We say that h is a free homotopy
r— 1
from 4>q to along the path t *+ h(J x t) .
Every spheroid with origin of (X,A) admits a free
homotopy along any given path, s: I -+ A, with s(0) = X q . Moreover,
free homotopies of homotopic spheroids along homotopic paths of A
produce homotopic spheroids. The proof differs from that of Theorem 3.2
in two details: in the first part of the proof, we have to start with
r— 1 i
ab <(> : J -> A (instead of <H ) and extend its homotopy,
Fr lr
(y, t) s (t) ,initially to a homotopy of ab c() : Fr I -* A, and then
to a homotopy of <J): I+ X; in the second part of the proof, we must
replace g: I x I + Xby g: I x I + A.

This theoremshows that given a path s: I A, the free

homotopies along s define a map T : tt^ (X ,A ,s (0 )) tt^ (X ,A, s (1 ) )
(for each r ^ 1). As in the absolute case, Tg are homomorphisms and
enjoy properties 2.1(i)-(iii). Therefore, a local system,
(A, {tt^ (X, A,x) } ,{Ts>), arises on A, which is a local system of groups
for r ^ 2, and a local system of sets with distingushed elements for
r = 1.This is the local system of the r-th homotopy groups of the pair
(X,A) . In particular, for any x E A and r ^ 1, tt^(A,Xq ) acts
naturally from the right on tt^(X,A,x ); this is a group-action for
r > 1, and it fixes the distingushed element for r = 1.
From the existence of this local system it follows that, for
any r ^ 1, the r-th homotopy groups TTr (X,A,x), x £ A, are all
isomorphic whenever A is connected.
A pair (X,A) with A connected is said to be r-simple if
the local system of its r-th homotopy groups is simple. In this case
all the homotopy groups tt (X,A,x) , x £ A, can be identified with a
single group, the r-th homotopy group of the pair (X,A) without base
point, TTr (X,A); the elements of TTr (X,A) are classes of freely
homotopic spheroids. A pair is simple if it is r-simple for any r ^ 1.
For example, every pointed space is a simple pair.

5. Given any pair (X,A) and any path s: I + A, the

diagram

TTr ( X , A , S (0) ) ---- ---- ► TTr _ 1 (A, S (0) )

T T
s > s
TTr (X,A,s(1) ) -- ^-- > 7Tr_1 (A,s(1 ))

obviously commutes. Therefore, the boundary homomorphisms,

 387

9 = TTr (X/A /x) -> TTr_^(A/x)/ combine with id A to define a homomor­

phism of the local system of the r-th homotopy groups ofthe pair (X,A)
into the local system of the (r-1)-th homotopy groups ofthe space A.
Further, given any continuous map, f: (X,A)( X ^ A 1), the
homomorphisms f* = (f*) x : tt r (X,A,x) tt
IT (X* ,A' ,f (x) ) ,combine with f
to define a homomorphism of the local system of the r-th homotopy groups
of the pair (X,A) into the corresponding local system of (X',A'). As
in 3.6, we mention threespecial cases: (a) (X,A) is r-simple;
(b) (X1,A *) is r-simple; and (c) both (X,A) and ( X ^ A 1) are
r-simple. In cases (a) and (b) , any path s: I A yields commutative
diagrams similar to those in 3.6; in case (c),the localsystems of the
r-th homotopy groups of the pairs (X,A) and (X1,A') reduce to single
groups, Trr (X,A) and TT^iX^A1), while thehomomorphisms reduce
to a single homomorphism, f *: TTr <X,A) 7Tr (X',A?).

If f is a homotopy equivalence, then all the homomorphisms

(f*)x are isomorphisms. The proof is similar to that given in the
absolute case (see 3.7).

6. Let (X,A) be a pair with base point x^ £ A. I_f X, A

are connected, then the triviality of all the homotopy groups 7Tr (X,A,XQ)
is equivalent to the ^-connectedness of (X,A);the triviality of the
homotopy groups tt^ X jA jX q ) for 1 < r < k is equivalent to the
k-connectedness of (X,A).

The proof is a repetition of the proof of Proposition 3.8,

with obvious modifications (instead of referring to 1.3.3.7, we refer to
1.3.3.9) .

The Group 7T2(X,A,Xq )

-1
7. If a ,3 £ ^ 2 (X/A,Xq), then a Ba = T 0a 3.

PROOF (for an alternative proof, see Subsection 10). We have

to check that, given two arbitrary spheroids, <(),ip £ Sph^ (X, A, xQ ) , there
is a free homotopy from ip to c|> %<J> along the loop 3<J). We shall
2
exhibit such a homotopy as a family of maps, xt : 1 ^ x (t £ I) ,
constructed as follows.
1 1 1 2
Set f(t) = -r-p ~ o- t - t-/ and divide I into eight parts,
lb o ¿>
as shown in Fig. 14: the points A 1 (t) , A3 (fc)'' A 4 have
abscissaef(t), t/4, 1 - (t/2), 1-f(t), respectively, while the
pointsB 1 (t), B2 (t) ,B3 (t), B4 (t) lie above these points at the height
 388

1 - f(t). Further, let a^ (t) -*■ I , j = 1,2,3, be the affine maps

defined by the conditions

a (A.,(t)) = (t,0), a1 (A2 (t)) = (0,0), (B1 (t) ) = (t,1),

a2 (A2 (t)) = (0,0), a2 (A3 (t)) = (1,0), a2 (B2 (t)) = (0,1),

a3 (A3 (t)) = (0,0), a3 (A4 (t)) = (t,0), a3 (B3 (t)) = (0,1).

Now set

xt Q i (t) = * ° V XtIQ 2 (t) ^ ° a2 ' t|Q3 (t) 3'

and consider the resulting continuous map Q (t) U Q (t) U Q (t) X.

•y \ a j>
Extend it firstly to a map Q^(t) X which is constant on the
horizontals in Q. (t) U Q c (t) and on the verticals in Q,-(t) U Q_ (t) ,
4 D o /
g
and then to a map Q^(t) X which is constant on the horizontals
in Q q (t). (The latter is possible, since the already extended map
7
U1 (t) X is constant on the segment [B (t) ,B„ (t) ] , and assumes
the same values at those points of the segments [B^ (t ) , (0 ,1 )] and
[B (t) , (1 ,1)] which lie at the same height.) The continuity of the
2 2
map I x I -> x defined by the family Xt : 1 ^ X follows from its
continuity on each of the eight polyhedrons
Ut€I(Qi (t) x t ] ' 1 * 1 * 8
 389

The Action of the Group ^(XjXg) on tt^ X ^ X q)

8. Given a spheroid w E Sph ^ (X,A,x ^ ) and a loop

s E Sph^(X,XQ>f the product ws is well defined, and obviously
ws E Sph^ (X,A,Xq). Moreover, the homotopy class of ws is uniquely
determined by the homotopy classes of w and s, and hence we may
define the product wa for any to E tt^(X,A,Xq) and a E tt^(X,Xq).
Using again the homotopies described in 1.1, we see that w(aa') =
= (a)a)a' and we , » = w for any w E tt.(X,A,xa ) and
7T^ (A ,X q / I U
a,a* E tt^(X,Xq). That is to say, the rule (to,cr) ^ wadefines a right
action Of 7T^ (X ,X q ) On 7T^ (X,A,x q) .

If A = X q , then we clearly recover the canonical right

action of the group tt^ (X,Xq ) (on itself) . It isreadily seen that
the translation T : (X, A, s (0) ) (X ,A, s (1 )) is a
s i 1
[T±nos : iti (X,s (0) ) -+■ tt^ (X, s (1) )] -map for any path s : I ->■ A, i.e.,

T (too) = T (to)T. (a) for all a £ tt.. (X, s (0) ) and u £ tt (X, A, s (0) ) .
s s in°s 1 1
Furthermore,given any continuous map f: (X,A,Xq ) (X',A',Xq ), the
homomorphismf*: tt^(X,A,Xq )tt^ (X ',A ',X q ) is a

[f* : tt1 (X,xQ)tt ^ (X' ,Xq) ]-map, i.e., f*(wo) = f*(to)f*(a) for all

io £ tt.
j (X,A,Xq ) and a£tt^ (X,Xq ) . In particular, if we consider the

map rel = [in: (X,x0 ,x0) -* (X,A,xQ )] , then

(rel*co)o = rel* (uo)

f or all w,a £ tt^ ( X , X q ).

9. For any to £ i t ^ X ^ X q ) and a £ ^(A^q) we have

T^to = a)(in*o) ,

where in = [in: (A,xQ) ■> (X,xQ)].

PROOF. Let w and s be spheroids in the classes to and

a. Consider, for eachfixed t £ I, the path s^: I -> A given by
s (y) = s (ty), and set wfc = w(in ° sfc) . Since sQis a constant path,
w andw„ are homotopic. On the other hand, the formula (y,t) h-w (y)
0
defines a free homotopy I x I I from Wq to w^ = w(in ° s) , along
s.
 390

5. A Digression: Sequences of Groups

and Homomorphisms, and tt-Sequences

1. A sequence of groups and homomorphism is a finite or

infinite (on one or both sides) sequence of groups such that for each
two adjacent groups, G^ and G j_+'i' there is given a homomorphism
G.
l Gi+1
A homomorphism of a sequence of groups and homomorphisms,
{Gi,hi: G^ -+ into another such sequence, {G^,h|: G^
is a sequence of homomorphisms : G^ -> G\] such that the diagram

h.
i-1 G.
i i+1
••• Gi-i l G i+i

H. H
H±-i l i+1
h! h! .
h i-i i 1+1
Gi-i G!l Gi+i

commutes. A homomorphism (HJ such that each H. : G. G!l is an

l l
isomorphism is called an isomorphism.

A sequence of groups and homomorphisms, {G^,h_^}, is exact

if for each group G^, excepting the initial and final ones, the
kernel Ker hu of the homomorphism h^ equals the image Imh^_^ of
the homomorphism h^_^.
The following three properties are common to all exact
sequences {G_^,h^} .

(i) If G^ is an inner (i.e., neither initial, nor final)

term of the given sequence, then h^_^ is trivial if and only if h_^
is a monomorphism, while h^ is trivial if and only if h^ is an
epimorphism; h ^ ^ and hu are both trivial if and only if the group
G. is trivial.

(ii) If G^ and Gi+1 are inner terms, then h^^ and

hi+1 are simultaneously trivial if and only if h^ is an isomorphism.

(iii) In particular, the triviality of Gi_1 and Gi+1

implies the trivialityof G w h i l e the trivialityof G^ and
Gi+2 imPlies th a t h^ is an isomorphism.

3. An exact sequence of the form 1+F +G ^ H + 1 is

called short (here 1 denotes the trivial group). An example is
 391

1 >p >G ► G/F > 1, where F is a normal subgroup of G.

-
This example has a universal character: every short exact sequence,
1 F £ G 2 H + 1iscanonically isomorphic to a sequence of thistype,
namely, to 1 --> Im f — in— > G - Pr— >G/lm f --- >1; the canonical
isomorphism is obviously {id 1 ,ab f : F -* Im f ,id G,h ^ pr (g~1 (h) ) ,id 1 } .

Splitting

4. Let (G^,h_^} be a sequence of groups and homomorphisms.

We say that this sequence is split from the right at the term G^ by
the homomorphism £ :G^+1 G^ if h^ o £ = idG^+1 . Such a splitting
is said to be normal if Im £ is a normal subgroup of G^.

Similarly, {G.,h.} is split from the left at the term G

i i a
by the homomorphism C: G -+ G „if £ o h . = id G „. We sometimes
c a a-1 a-1 a-1
say simply that the given sequence is split, or that it splits (at right
or at left) at G .
a
5 (LEMMA) . Let A and B be groups, and let u : A -> B and
v: B A be homomorphisms. If Im v is a normal subgroup of A and
u o v = id B , then A = Ker u x Im v .

PROOF. Every element a € A can be represented as

[a(v o u(a)) ] (v o u(a)), and obviously a(v o u(a)) € Ker u and
v o u(a) € Im v . If a € Ker u (1 Imv , then u(a) = e , and there is
B
b € B with v(b) = a. Thus, b = u o v(b) = u(a) = e and a = v(b) =
B
= eA ‘
6. If the exact sequence {G ,h^} splits normally from the
right at G^ and splits fromthe right at Ga_3 , then it also splits
from the left at G , and G = G „ * G .„. More precisely, under
-------------------- a a a-1 a+1 ------ c ^ --------
these hypotheses, is a monomorphism, h^ is an epimorphism, and
G^ decomposes into the direct product of Imh^^ and a subgroup which
is mapped isomorphically onto Ga+<1 by_ h . Moreover, every homomorphism
£: G^+^ Ga which splits the sequence from the right is a monomorphism,
and if £ is also normally splitting, then for a direct complement of
Imh^^ one may take Im £ .

PROOF. The equality Ga = Im h ^ x Im £ is a consequence

of Lemma5 and of theexactness of the given sequence. From h^ o £ =
= id G „ it follows that h is an epimorphism and that £ is a
a+1 a
monomorphism. Since {G.,h.}
1 1
splits from the right at GCX J h cx*o
-
is an epimorphism, and now the exactness of {G^,tu} implies that
 392

h ~ is trivial, while h is a monomorphism. Finally, as a

a-2 a- \
homomorphism splitting the given sequence at from the left one can
take the homomorphism which is the inverse of on an<^
equals the identity on Im .
7. If the exact sequence {G^,h^} splits from the left at
G and 'Ja+3 ' then it splits normally from the right at G , and
a ---
G s G , x G More precisely, under these hypotheses, is a
a a-1 a+1*
monomorphism, h^ is an epimorphism, and G^ decomposes into the
direct product of Imh^_^ and a subgroup which is mapped isomorphically
onto b^ Moreover, every homomorphism ç: G which
a+1 a-1
splits the sequence from the left is an epimorphism, and for a direct
complement of Imh^_^ one can take Ker £ .

PROOF. The equality G^ = Imh^_^ x Ker £ is a consequence

of Lemma 5. From ç o h id G it follows that is a mono-
a-1 a-1 a-1
morphism and that ç is an epimorphism. Since {G^,h_^} splits from
the left at G n and is exact, is an epimorphism. As a homo-
a +3
morphism splitting the given sequence normally from the right at G^,
one can take the composition G „ = G /Kerh = G /imh „ =
a+1 . a a a-1
= (Imh . x Ker O / l m h . = Ker Ç — -— ► G .
a-1 a-1 a

An exact sequence 1->-F-*G^H->1 splits at

from
the left if and only if it splits at G normally from the right, and
this happens if and only if the subgroup Im f = Ker g o_f G has a
direct complement.

This is a corollary of Lemma 5.

Five Lemma

9 • 11
h h^ h h
G 1 ---- » G2 > G3 ‘ G4 — ^ S

'1 <!>>

g ; — 1 GJ, G. G. h4 G,

is a homomorphism of exact sequences, and if (j) is an epimorphism,

<t>2' ^4 are isomorphisms, and is a monomorphism, then $ is an
isomorphism.

Let us show first that $ is a monomorphism. If a G Ker (f)^,

 393

then if>4 o h 3 (a) = h^ o <j>3 (a) = eG i' and so h 3 <a) = eQ , i.e.,

a £ Ker h 3 = Imh2 . Let a = h 2 (b), b £G2 . Since h'2 o <J>2 (b) =

= <t>3 o h 2 (b) = <t3 (a) = eQ ,, there is c e G| such that h]j (c) = <t>2 (b).

Therefore, there is d € G1 such that h^j o <j)(d) = <|> (b) . On the

other hand, h^j o ^ (d) = <{>2 o h^ (d) , and hence 4>2 (b) = <j>2 « h^ (d) .

Consequently, b = h (d) and a = h„ o h.(d) = e. .

3
Now let us verify that cf>3 is an epimorphism. Let a £ G3 .

Then <j)j. o h^ o o h 3 <a) = h^ o h 3 <a) = e ^ , and so

-1 -1
^4 ° 04 ° h 3 (a) = eG , i.e., ((>4 o h 3 <a) £ Ker h^ = Imh^ . Let

(f)” 1 o h^ (a) = h 3 (b), b £G3 . Since h^ (a (<(, (b) ) _1 ) =

= (<fr4 o <f, " 1 c h 3 (a)) (h3 o((,3 (b) ) _1 = 4>4 ((0 4 1 ° h^ (a) ) (h3 (b) )_1 ) =eG ,,
_ ^
there is c £ G^ such that h^ (c) = ai^ib)) , and hence there is

d £ G2 such that h^ o ^ (d )= a(cf)^(b)) \ On the other hand,

h^ ° ^ 2 ^) = ^3 ° h 2 (d) , and hence ai^ib)) ^ = ^3 ° h 2 (d) .

Consequently, a = (l^ (d) b ) .

77- Sequences

10. In the next subsections we shall handle the so-called

homotopy sequences. These are rather cumbersome entities which are
similar to sequences of groups and homomorphisms, but possess additional
properties and structures. Such sequences are encountered in various
geometric situations, but they are all algebraically related. The rest
of the present subsection is devoted to a preliminary, purely algebraic
description and study of these sequences.

11. Consider a left-infinite sequence

P^r Pc P4 Pq P? Pi Pq
... n 7 n6 — ^ n5 — n4 — ^ JI3 — n2 — ^ n 1 — ^ n0 , (6 )

where:
V n , II2 are sets with an identity (distinguished element),

n3, n4 , n 5 are groups,

IK, Jln , ... are Abelian groups,

o /
 394

^0' P 1 ' P2 are ^ornornorP^^sras sense °f 1-2,

p^, p^, ... are group homomorphism.

Then (6 ) is called a Ti-sequence if there are given

right group-actions of IT^ on the groups ^3^ with k ^ 2,

right group-actions of TT^ on the groups ^3 ^ + 1 with k ^ 2,

right group-actions of on the groups ^3 ^ - 1 with k ^ 2,

and a right action of II^ on the set IT^ r

such that
(i) is a P3 “homomorphism for all k ^ 2;

(ii) p3 k+i a ^^-homomorphism for all k ^ 2;

(iii) P3 k_i is a ^ 4 “homomorphism for all k ^ 2 , with

respect to the right group-action of II4 on IT^^ induced by the given
action of on IT^ via p^;

(iv) p^ is a II^-homomorphism with respect to the right inner

action of n^;

(v) the transformation of the group 11^ induced by the

image p^ (a) £ II4 of an arbitrary element a £ IT^ is the inner auto-

morphism (B *+ a - 1 3 a; '

(vi) the transformation of the set Jl induced by an

arbitrary element a £ II^ coincides on ^2 ^ 3 ^ with the transformation
P2 (oj) a = p2 (oia) .

A homomorphism of the n-sequence {H.,p.}°° into the

°° 1 1 1 —u
TT-sequence =o a sec3uence °f homomorphisms h^z II -+ IT!
such that

0 = hi ° Pi for all i ^ 0;

h3k' h3k+1' and h3k-1 *k > 2 * are h 3 ~ ' h 4 “ 'and

h^-homomorphisms, respectively;

h 2 (oj)h^(a) = h 2 (00a) for all oj £ and a £ 11^.

An isomorphism is a homomorphism such that all -

h.'s are
isomorphisms.

12. Among conditions (i)-(vi) above, tworefer to p^,

namely (iv) and (v). From (iv) it follows that if H4 acts identically
on IIc-, then Im p^ is contained in the center of the group IT^. From
(v) it follows that if IT4 acts identically on n5 , then H5 is
Abelian, and that the converse is true provided p^ is an epimorphism.
 395

In general, (v) implies that Ker is contained in the center of n^.

13.The Tr-sequence (6 ) is exact if Ker p . = Im p . „ for all

----- 'l i+1
i ^ 0 and, in addition, the preimages of the elements of under p^
are nothing but the orbits of the action of IT3 on n2 -

If the ii-sequence (6 ) is exact and i ^ 0 is arbitrary, then

obviously the homomorphism piis trivial if and only if P^ +1 is an
epimorphism, while Ker p^ is trivial if and only if Pj_+^ ^-s trivial.
In general, wheni ^ 3, Ker p^ is trivial if and only p^ is
injective, because p. is a group homomorphism. Further, the triviality
of Ker p2 means that p2 is injective: if p2 (ct) = p2 (3), then
-1 _1 1 1
P2 (a3 ) = p2 (a) 3 = p2 (3)3 = P2 (3B )= P2 (en ) [see condition (vi)

in 11], and hence a = 3. The triviality of Ker p^ does not imply

the injectivity of p^, and this is also valid for Ker p^ and p^.
However, in thecase of an exact 7T-sequence (6 ), the injectivity of p^
is guaranteed if the group is trivial, or if it acts identically
on TI2 •

The above discussion makes clear that, in the case of an exact

7T-sequence (6 ) and for i ^ 1, the triviality of p^ and P.^+2 -*-s
equivalent to the invertibility of Pj_+i' that the triviality of IK
and H^ + 2 implies the triviality of nj_+ '|' and, finally, that the
triviality of an<^ ^i +2 implies the invertibility of p^ (cf. 2 ).
14. Let (6 ) be an exact iT-sequence. If the action of
on n2 induced by the action of via p^ is identical, then Im p^
is a normal subgroup of n^. The converse is true provided p2 is an
epimorphism.

PROOF. Assume that II^ acts identically on II2 . If a G

3 E IT^ , then p2 (B)p3 (a) = p2 (3), and hence p2 (3p3 (a)B 1) =

= p2 (6 P3 (a))6 ~ 1 = [p2 (6 )P3 (a)]B_1 = p2 (B)B~ 1 = P2 (BfT1) = en [see

-1
condition (vi) in 11]. Therefore, 6 p3 (a)B £ Ker p2 = Im p ^ , and
hence Im p^ is a normal subgroup of IT^ •

Now assume that p2 is an epimorphism and that Im p3 is a

normal subgroup of IT^. If a £ II^ and y £ n2 , then there is 3 £ IT^

such that = Y> and since Bp3 <a)B 1 £ Im p3 = Ker p2 , we have

yp3 (a) = p2 (B)p3 (a) = p2 (Bp3 (a)) = p2 (3p3 <a)B 1 B) = p2 (-Bp3 <a)B 1)S =

= B = P2 (en )3 = P2 (l3) = Y [see again condition (vi) in 1 1 ].

Consequently, II4acts identically on II

 396

15. Every sequence of Abelian groups and group homomorphisms

of the form (6 ) can be viewed as a 7r-sequence, where the action of II^
on ^2'k' k ^ 2' and the actions of n4 on n 3 k +1 and ^3 k - 1 ' k ^ 2,
are identical, while the action of II3 on n2 is given by oia = ojp2 (a)
[w £ ¡¡2 r o £ II^] - Then it is readily seen that a sequence which is
exact in the sense of 2 is also exact as a Ti-sequence, and that the
homomorphisms of sequences in the sense of 1 are also homomorphisms in
the sense of 1 1 .

Splitting of TT-Sequences

16. We say that the Ti-sequence (6 ) is split from the right

at the term n by the homomorphism £: II ^ ^ n if p - o ^ = id II .
------------- a — *--------------- c----- a-1 a "a-1 a-1
This splitting is normal if a = 0,1,2 ora ^ 3 and Im£ is a normal
subgroup of n^. We say that the ir-sequence (6 ) is split from the left
at the term II by the homomorphism £: II II . if £ ° p = id II . .
----------- a — 1------------ c---- a a +1 a a+1
(£ is a group homomorphism when this makes sense, and a homomorphism
of sets with identity elements otherwise.)

17. _If a ^ 5, then any right splitting of the Tr-sequence

(6 ) at. 11^ is normal. If (6 ) is exact and acts identically on n^_ ,
then any right splitting at is normal.

Since for a > 5the groups n are Abelian, we need consider

a
only II ^ and lit-. Suppose that the homomorphism £: II ^ 11^ splits
(6 ) from the right at n^. If a , & £ Tl^, then

Bag- 1 = [3C(p4 (6~1 ))][C(P4 (B))aB_1] =

= [c(p4 (6 ))aB_1] [BC(P4 (6 _1))] = C(P4 (B))a?(P4 (B_1));

_ *1
permuting the factors is permissible because S?(p4 (B )) 6 Ker p4

(as shown by the equality p o x, = id IK), and hence B?(p4 (B~1))

belongs to the center of Hr
(see 12). This representation of SaB
3
_i
shows that if a £ Im x, , then BaB £ Im ^ for all B £ TI .

Now suppose that r,: U^ ■+ n 4 splits the exact ir-sequence (6 )

from the right at n4 , and that n 4 acts identically on IT5 . If
a ,B £ IT4 , then

Ba B_ 1 = [ B C ( p 3 ( B ~ 1 ) ) ] U ( p 3 ( 3) ) a B _ 1 ] =

= U (P3 (B) )aB“1] [Bi(P3 (B_ 1 ) )] = ? (P3 (3) )ac(P3 (B_ 1 ) ) ;

 397

_1
permuting the factors is permissible because 3c(p3 (6 )) e K e r p 3 = Imp^

(as shown by the equality p 3 o £ = i d n 3>, and hence 3 ^ (p3 (3 ~ 1 ))

belongs to the center of II. (see 12). This representation of Bag- 1
^ _1
shows that if a E Im £ , then (3a3 £ Im £ for all B E n4 .

18. Let the Tr-sequence (6) be exact, normally split from

right at and split from the right at n 3. If a ^ 4, then,
according to 6 , sequence (6 ) also splits from the left at II , and IT
a a
decomposes into the product of a subgroup canonically isomorphic to
and a subgroup isomorphic to na _«| • When a = 1,2,3, (6 ) also
splits from the left at (to see this, repeat the arguments of 6 ),
but obvious examples demonstrate that the above isomorphism
II = n , - x II . is not necessarily valid,
a a +1 a- 1
Now let the Tr-sequence (6 ) be exact and split from the left at
11^ and if a ^ 6, then, according to 7, (6 ) is normally split
from the right at n , and decomposes into the product of a sub­
group canonically isomorphic to and a subgroup isomorphic to
. A word-for-word repetition of the arguments in 7 shows that this
holds also for a = 4,5. If a = 3, all we can say is that (6 ) splits
from the right at TI^.

In what follows, we shall often encounter Tr-sequences which

are exact, and split at every third term. The discussion above shows
that if the Tr-sequence (6 ) is exact and normally split from the right
(split from the left) at every term n. with iA +3k ^ 1 , then it
1Q ^K U
also splits from the left at these terms (respectively,it also splits
from the right at IK for iQ +3k ^ 3, and splits normally from the
right at IK for^ in+3k > 4).
1o u

The Ti-Variant of the Five Lemma

°° 00
19. Let =o and i /Pi i=0 be exact Tr-sequences,

and let {hu: IK -+ be a homomorphism of the first sequence into

the second. If h „ and h .„ are isomorphisms and Ker h_. n = e^ ,

— ------------------ a-1 --- a+1 ----------- 1------------ a-z 11a-z0

Imh + 2 = nl+2' then Ker ha = en and I m h a = Ha (and hence h^ is a

a a a
group isomorphism for all a ^ 3 ).

If a ^ 5, then this is contained in 9 . The proof for

a= 2,3,4 is similar.
 398

6. The Homotopy Sequence of a Pair

1. Let (X,A) be a topological pair with base point xQ £ A.

According to Subsections 1 and 4, the homotopy groups tt^ (X ,>:Q ) and
tt (A.xJ are defined for any r ^ 0, whereas the homotopy groups
r 0
7Tr (X,A,xq) are defined for any r ^ 1. Moreover, by 4.3, there are the
homomorphisms 9: 7Tr (X,A,xQ) + TTr_ 1 (A,xQ ). To these we add the homo-
morphisms in*: TTr (A,xQ) + TTr (X,xQ) and rel* : TTr (X,xQ ) TTr (X,A,xQ ),
induced by the inclusions in: A + X and rel: (X,x0 ,xQ ) (X,A,xQ ).
These three series of homotopy groups and three series of homomorphisms
can be assembled into the left-infinite sequence

... tt2 (A,x 0 ) -ln* > ti2 (X,X q ) ■ rel* > tt2 (X,A, x q ) — it 1 (A,xQ ) ln*^
(7)
TI1 (X,xQ) -rel* > ir1 (X,A,xq) — itq (A,xQ ) in* > ir0 (X,x0 )

Here, all the terms, except for the last six, are Abelian
groups, all the terms, except for the last three, are groups, the last
three termsare sets with an identity, all themaps,except for the
last three, are group homomorphisms, and the last three maps are homo­
morphisms of sets with identity. By 3.3 and 4.4, tt^(X,Xq ) acts from
the right on TTr (X,xQ), while tt^(A,x 0 ) acts from the right on
iTr (A,XQ) and 7Tr (X,A,XQ), and all these are group-actions.
Furthermore, the homomorphisms in*, rel*, and 9 are compatible
with these actions, as required in Definition 5.11 (see 3.6, 4.5, and
3.4), and a Ba = T ^ 3 for all a,B £ tt2 (X,A,xq ) (see 4.7). From
4.8 and 4.9 it follows that tt^(X ,x 0 ) acts from the right on tt^(X,A,Xq

in such a manner that rel*(co)a = rel* (wa) for all oo, a £ tt (X,xQ).
Therefore, (7) is a 7T-sequence, called the homotopy sequence of the
pair (X,A) with base point x ^ .

2. Sequence (7) is exact.

The proof is a routine, direct verification of the six

inclusions Im in* cz Ker rel* , Ker rel* cz Im in* , Imrel* cz Ker 3 ,
Ker 9 cz Imrel* , Im 9 cz Ker in* , and Ker in* c Im 3 , plus a
justification of the fact that, given a,B £ (x,A,xQ), there is a
a £ tt(X,xQ ) such that ao = 3 if and only if 9a = 96.

3. Given any path s: I A,the vertical isomorphisms

define an isomorphism of the first sequence onto the second.

The commutativity of the first square has been established in

3.6, while the commutativity of the second and the third in 4.5. From
the fact that the local systems (A, {tt^ (A ,x)} ,{Tg} ), (X, (тг^(Х,х) },{Tg}), and
(АДтгг (X,A,x) },(Ts}) satisfy property 2.1 (i), and from the equality

T Ь (аза) = T S (w)T.
1П оS
(a) [a E тг| (X,s(0)), ш € тт] (X,A, s (0 ))] , deduced in
4.8, it follows that the vertical homomorphisms are compatible with the
actions of the fundamental groups.

4. Given a continuous map f: (X,A,Xq ) + (Х!,А?,х^), the

vertical homomorphisms

rel* /T, ^ x
... 7Tr (A,x0 ) -- n * > TTr (X,x0 ) *-» тг (X,A,x..)
"r - 1 (A' V

(ab f)* (ab f )* (8 )

in* rel^ (А’,хД

• •• ^ (A ',Xg
j . 77r (X ',X q ]
r- 1

induced by f yield a homomorphism of the first sequence into the

second.

The commutativity of the first two squares follows from 1.7

and 4.2. The commutativity of the third square has been established in
4 .3 , and the compatibility of the vertical homomorphisms with the
actions of the fundamental groups - in 3.6, 4.5, and 4.8.

The Most Important Special Cases

5. If X is “-connected, then all the homomorphisms

B: tt^ (X,A, X q ) ^ (A ,X q ) are isomorphisms. If X is k-connected
and k < °°, then 3 : TTr (X,A,x0 ) iTr_ 1 (A,x0) is an isomorphism for
all r £ k, while d: 7rk+ (X,A,xQ ) + Trk (A,x0) is an epimorphism.
The converse of both statements is true provided that X is connected.
If A is “-connected, then all the homomorphisms
rel*: (X,Xq ) TTr (X,A,xQ ) are isomorphisms. If A is k-connected,
 400

and k < 00 then rel.: it (X,xn ) ->- it (X,A,xn ) is an isomorphism for

x i" U r U
all r < k, while rel*: irk+ 1 (X,x0 ) T?k + 1 (X, A,xQ ) is an epimorphism.
If one of the spaces X, A is connected, then again, the converse of
both statements is true.
If the pair (X,A) is “-connected, then all the homomorphisms
in*: tt^(A,Xq) Tr^iX^x^) are isomorphisms. If (X,A) is k-connected
and k < co, then in*: Tr^(A,x0 ) -+ 7T^(X,xq) is an isomorphism for all
r < k, while in*: 7rk (A,xQ ) -> TTk (X,xQ) is an epimorphism. The
converse is true in both cases (with nosupplementary conditions). In
particular, if in: A -+ X is a homotopy equivalence, then the pair
(X,A) is “-connected (see 3.7 and cf. 1.3.3.9).

6. I_f A is a retract of X, then the sequence (7) splits

from the left at tt^(X,Xq), and any retraction p: X A induces
splitting homomorphisms p*: Ti^(X,XQ) -+ tt^(A,Xq).

PROOF. Since p o in = id A , p* o in* = id tt^ (A , X q ) .

7. Suppose that (X,x^) can be contracted to (A,Xq ), i.e.,

id X _is XQ-homotopic to some map h: X X such that h(X) c A. Then
the sequence (7) splits from the right at tt^(A,X q ), and as splitting
homomorphisms one may take (abh)* : tt^ (X , Tr^(A,xQ).

PROOF. The composition in o abh is x^-homotopic to id X ,

and hence in* o (abh)* = i d 7i^(X,XQ> .

8. Suppose that (A,xQ) is contractible in (X,xQ), i.e.,

the inclusion A X ijs XQ-homotopic to the constant map. Then the
sequence (7) splits from the right at tt (X,A,xQ). As splitting homo­
morphisms 77^ (A ,Xq) 7rr+^ (X,A,xq) onemay take those induced by the
maps Yr : Sphr (A,xQ) + Sphr+ 1 (X,A,xQ), given by [yr (<f>) ] (t1 , .. . ,t ) =

= h (tp (t1 , ...,tr ) ,tr + 1 ) tcf> € Sph (A,xq)], where h: A x I -+ X is any

homotopy from in: A X to the constant map.

This is a corollary of the obvious equality

3 o Yr = idSphr (A,xQ) .

9. The following remarks are concerned not with the homotopy

sequences of pairs themselves, but with the homomorphism (8 ) induced by
a map f: (X,A,xQ) -+ (X',A',Xq) between pairs with base point.
The Tr-variant of the Five Lemma (see 5.19) shows that:
if (abf)* : Trr (A,x0) -> irr (A',x^), all r z 0 , and
f*: 7rr iX,A,XQ) ->■ (X' ,A' ,Xq) , all r 5 1, are isomorphisms, then so
are f* : Trr (X,x0> ■+ *r ( x ' , for all r>. 1; if f*: ^ (x,xQ)+ n (X’,x'Q) ,
 401

all r £ 1, and f*: TTr ( X , A , x Q ) + t\^ (X ', A 1, x^ ), all r * 1, are

isomorphisms, then so are (abf)* : 7Tr (A,x0 ) + tt (A',x^), for all
r > 1; finally , if f* : TTr ( X , x Q ) + TTr ( X 1 , x ^ ), all r ^ 0, and
(abf)* : tt^(A,Xq) + tt^ (A1 /xq ) / all r ^ 0, are isomorphisms, then so
are f* : 7Tr ( X / A / x Q ) -* tt ^ (X 1 ,A 1 /xq ) , for all r ^ 2, while
f*: 7t ^ ( X , A , X q ) -> tt ^ (X* , A ' , X q ) is an epimorphism with trivial kernel.

In the last case, f*: tt^(X,A,Xq) -+ tt^ (X 1 ,A 1 ,x^) is not

necessarily injective; see 3.3.8. However, this map is certainly
injective (and hence, an isomorphism) if we assume, in addition, that
all the homomorphisms (abf)*: tt (A,x) + tt (A *,f (x) ) , x G A, are
epimorphic. To see this, let ^ with f* (co^ ) = f*(w2 ),
and let w^ and w2 be spheroids in the classes m and a^. Then
there is a path s': I + A* such that s'(0)= f o w^(1), s'(1) =
= f o w 2 (1 ), and the loop

((f o w ^ ) ( [i n : A 1 -*■ X 1] o s , ) ) ( f o w 2 ) 1 (9)

is homotopic to the constant loop. Since (abf)* : tTq (A,Xq ) tTq (A',Xq )
is an isomorphism and (abf)* : (A,w (1 )) -* tt^ (A ',f (w^ (1 )))isan
epimorphism, there is a path s: I A such that s(0) = w^(1),
s(1) =w 2 (1) / and the path ab f os: I + A* is homotopic to s'. For
such achoice of s, f takes the loop

(w^([in: A -*■ X] ° s))w2^ (10)

into aloop homotopic to (9), and therefore homotopic to theconstant

loop. Finally, from the fact that f*: tt^(X,Xq) ■+ tt^(X',Xq) is an
isomorphism, it follows that (1 0 ) itself is homotopic to the constant
loop, i.e., the spheroids w^ and w 2 are homotopic, and = w2 .

The Homotopy Sequence of a Triple

10. Let (X,A ,B ) be a t o p o l og ic al tr iple wi th b a s e po in t

Xq £ B. According to S u b s e c t i o n 4, w h e n r ^ 1 the h o m o t o p y g roups
TTr (X,A,xQ ), TTr (X,B,xQ ), and TTr (A,B,xQ ), and the h o m o m o r p h i s m s
in* : TTr (A,B,x0 ) + TTr (X,B,xQ ) and rel* : TTr (X,B,xQ ) + TTr (X,A,x0 ),
i n d u c e d by the inclusions in: (A,B) ■+ (X,B) an d rel: (X,B) (X,A) ,
are w el l defined. If r > 2, we de fi ne an ad di t i o n a l h om om orp hi sm ,
9: (X,A ,xQ ) -> TTr _ 1 (A,B,xQ ) , as the composition of the boundary
homomorphism TTr (X,A,x0) + Trr_1 (A,xQ) and the homomorphism
TTr_i(A,xo) TTr_1 (A, B,xQ), induced by the inclusion (A,x0 ,xQ) -> (A,B,xQ ).
 402

N ow we may ass em bl e these three series of g r o u p s and three series of

homomorphisms into a l e f t - i nf in it e sequence

... tt2 (A, B ,xQ ) -— ■*-> 2 (X ,B ,xQ ) - rel*^ tt2 (X,A ,xQ ) —

(11 )
tt1 (A ,B ,xQ ) ^ (X,B,x0 ) rel* » TT1 (X, A,xq )

As was (7), (11) is a ir-sequence: the r i g h t g r o u p - a c t i o n s of

2 (X,A , X q ) on the gr oups tt (X,A fx Q ), and the rig ht g r o u p - a c t i o n s of
tt 2 (X,B ,Xq ) on the groups Trr (A,B,xQ ) and tt^ (X ,B ,x Q ) , are i n du ce d by
the ac tions of n (A,xQ ) and tt1 (B,xQ ) via the h o m o m o r p h i s m s
3: Tr2 (X,A,xQ ) + tt (A,xQ ) and 3: Tr2 (X,B,xQ ) -> tt^ B ^ X q ); similarly, the
right act io n of tt2 (X,A, x q ) on tt,
j (A,B, x 0 ) is ind uced by the a c t i o n
of tt ^ (A,X q ) via the h o m o m o r p h i s m 3: tt2 (X,A, x 0 ) -+ tt^(A, x 0 ); finally,
4.4, 4.5, 4.7, and 4.8 show that the c o n d it io ns i mposed by D e f i n i t i o n
5.11 are satisfied. Tr-Sequence (11) is cal led the h o m o t o p y s e q ue nc e of
the triple (X,A,B) with base poi nt xQ .

Sequence (11) is exact; cf. 2.

Given any path s: I B, the t r a n s l a t i o n s (X,A, s (0) )
TT^ (X, A, s (1 ) ) , (X,B,s (0) ) -+ 7rr (X,B, s (1 )), and tt^ (A, B ,s (0 )) ->
-> it (A,B,s(1)) define an isomorphism of the homotopy sequence of the
triple (X,A,B) with base point s(0) into the homotopy sequence of
the triple (X,A,B) with base point s(1); cf. 3.
Given any continuous map f from a triple (X,A,B) with
base point xQ into a triple (Xf,A,,Bf) with base point x^ (x € B,
Xq G B 1), the homomorphisms f*: Trr (X,A,xQ) tt^ (X ',A 1 ,x^ ) ,
f* : TTr (X,B ,xQ ) tt^_(X 1 ,B 1 ,Xq ) , and (abf)* : TTr (A,B,xQ) -> tt^_ (A 1 ,B 1 ,x^ ),
constitute a homomorphism from the homotopy sequence of the first triple
into the homotopy sequence of the second triple; cf. 4.

7. The Local System of Homotopy

Groups of the Fibers of a Serre Bundle

1. Suppose that £ is a Serre bundle, FQ and F are

fibers of £, and xQ G F Q , x1 G F1 . Two spheroids, <\>Q G Sphr (F0 ,xQ )
and (f)^ G Sph^ (F^ fx ) , are said to be fiber homotopic if the spheroids
[in: Fq tl £] o (J)0 G Sph^ (11 £, x^) and [in: F ^ + tl ?] o <j> G
G Sph^_ (tl £ ,x^) can be connected by a free homotopy consisting of
spheroids of tl £which take ir into fibers of £. In other words,
4>0 and c|) are fiber homotopic if there is a map h: lr x I + tl £
 403

such that: h is constant on each set Fr ir x t, t £ I, h(y,0) =

= c()q (y) ,
h (y,1) = cf) (y) , y G Ir , and the map pr E, ° h is constant
3T
on each set I x t, t G I. We say that h is a fiber homotopy from
^0 — ^1 along the path t h* h(Fr ir x t) .

2. Given any spheroid <J) with origin of the fiber Fq ,

there is afiber homotopy of <j) along any path with origin x^ in
tl£ . Fiber homotopies of homotopic spheroids along homotopic paths of
tl E, always lead to homotopic spheroids. Fiber homotopies of freely
homotopic spheroids along paths which cover homotopic paths of bs £
lead to freely homotopic spheroids.

PROOF. Let s: I tl E, be a path with s(0) = xQ , and let

(J) G Sphr (Fg,x ). Define homotopies H: I x I bs E, and
G: ir x i tl £ by H(y,t) = pr £ o s(t) and G(y,t) = s(t). Since
H (y ,0) = pr£(<j>(y)) for all y G ir , and G(y,0) = <j>(y) for all
y G Fr ir , there is a homotopy, H~: ir x I tl £ , which covers H
and satisfies H~(y,t) = G(y,t) for all y G Fr ir , H (y,0) = (J>(y)
for all y G ir (see 4.1.3.6 ). H~ is manifestly a fiber homotopy of
(j> along s.

To prove the second assertion of the theorem, it suffices to

show that two spheroids, 4>^ G Sphr (F0 ,xQ), which are fiber homotopic
along a loop s: I + tl^ homotopic with the constant loop, are
homotopic in the usual sense. Choose a fiber homotopy $: I x I -* tl E,
from t}) to ip along s, and a homotopy h: I x I -*• tl E, from s
to the constant loop. Now define f : lr + ^ ->• tl E, and homotopies
H: Ir + 1 x I bs E, and G: Fr lr + 1 x I -»• tl E, , by the formulas
f~(tr ...,tr+1) = $((t 1 f...,tr ),tr+1), =
= pr E, ° h (tr + 1 ,t) , and

<f>(t^/...ft^)f if — ^'

^ (t^ / •• • tr+i — 1 t

h (tr+ ^ ,t) , if (t^ , ... G Fr X .

Since H (y,0) =pr£(f~(y)) for Y € Ir+1, and G(y,0) = f~(y) for

y G Fr Ir + 1 , there is a homotopy H~: lr + 1 x I -> tl E, which covers
H and satisfies H~(y,t) = G(y,t) for y G Fr lr + 1 and H~(y,0) =
= f (y) for y G ir ^ . Now let ¥( (t , ... ,tr ) ,t) = H ((t^/..., t^_,t ) ,1 )
and note thaty : ir x i -> fq is a (usual) homotopy from <f> to \p.

Let us prove the last part of the theorem. Suppose that the
spheroids 4>0 € Sphr (FQ ,x0 ) and <{>., G Sph^F^x.,) are fiber homotopic
 404

along the path u: I + tl £ , and that the same holds for the spheroids
^ e Sphr (F0 ,xQ) , I^ e Sphr (F1 ,x1), and the path v: I tl E, . Further,
suppose that the paths pr £ ° u >Pr £ ° v: I bs E, are homotopic, and
that (J)Q and are freely homotopic along a path w: I + FQ . The
last means that there is a fiber homotopy from <J>0 to \J/Q along the
path wQ = [in: FQ tlEj o w. It is clear that the loop

pr £ ° (u ^ (wnv )) : I bs £ is homotopic to the constant loop, which in

-1
turn implies that the path u <w q v ) is homotopic to some path
w 1 : I -> tl £ with w 1 (I)c F (see 4. 1.3.6). By the first part of the
theorem, there exists a fiber homotopy of <b along w^ , and now the
second part guarantees that this homotopy yields a spheroid which is
homotopic to [(K is fiber homotopic to along the path
-1 1 1 i
u (Wq V) ] . Consequently, is fiber homotopic to ^ along a path
in F^ , i.e., the spheroids <i>^ and ^ are freely homotopic.

3. By Theorem 2, the fiber homotopies along a given path

s: I + tl £ define (for any r ^ 0) a map T g : tt^_(Fq ,s (0)) -> tt^(F,s (1)),

where FQ = pr E, 1 (pr £ (s (0 ))) , F 1 = pr ^ 1 (pr £ (s (1 )) ) . The maps Tg

are obviously homomorphisms and satisfy conditions (i)-(iii) in 2 .1 .
Therefore, a local system (tl E,,{tt^ (pr £ (pr E, (x) ) ,x) },{Ts }) arises
on tl 5 . This is a local system of groups for r ^ 1, and a local
system of sets with identity for r = 0. It is called the upper local
system of the r-th homotopy groups of the fibers of £. In particular,
given any x £ tl £ and r ^ 1, there is a natural right group-action
_ 1
of tt1 (tl £,x) on TT^ipr £ (pr E, (x) ) ,x) .

Clearly, by restricting this local system to any fiber of £

we obtain the local system of the r-th homotopy groups of the given
i _
fiber. Moreover, the homomorphisms in*: ^ (pr E, (pr £(x)),x) +
■+ TTr (tl£,x) combine with id tl £ to define a homomorphism of the
upper local system into the local system of the r-th homotopy groups of
tl £ .

4. Suppose now that the fibers of E, are r-simple. Then,

given any point b C bs £ , all the homotopy groups tt (pr (b) ,x) ,
—1 r 1
x € pr E, (b) , may be identified with a unique group, tt (pr E, (b))
(see 3.5). In this case, for any path s: I bs E,, we can define
-1 — 1
TO : tt
-L (pr £ (s (0 ))) -+ ttI (pr F, (s(1 )) ) to be the translation

V : 71 r (Pr £ (s(0)),s (0)) -+ 7T^ (pr E, 1 (S(1)) ,s (1)) along some path

s : I tl E, which covers s; from 2 it follows that T

does not
~ s
depend upon the choice of s . The maps T^ are obviously homomorphisms
and satisfy properties (i)-(iii) in 2.1. Therefore, a local system
 405

(bs £ ,{tt^ (pr £ (b))},{T }) arises on bs £ , which consists of groups

(sets with identity) for any r ^ 1 (respectively, for r = 0). This
is called the lower local system of the r-th homotopy groups of the
fibers of £. In particular, for any r ^1 and any bE bsK ,there
i —
is a natural right group-action of tt^ (bs £, b ) on TT^(pr£ (b) ) .
It is readily seen that the lower local system induces the
upper local system, defined in 3 on tl £ , viathe projection pr £ .

5. Let cp: £ be a map of Serre bundles. Then tl cp

and the homomorphisms
ab tl cp* : TTr (pr£ 1 (pr £ (x)),x) -> (pr £ ^1 (pr £ ^ (tl cp(x))) ,tl cp(x) )
[x E tl £] combine to define a homomorphism of the upper local system
of the r-th homotopy groups of the fibers of £ into the similar system
for Furthermore, if the fibers of £ and arer-simple, then
bs cp and abtlcp*: TT^(pr£ ^ (b) ) (pr £^ (bs cp(b))) [b E bs £]
combine to define a homomorphism of the lower local system of the r-th
homotopy groups of the fibers of £ into the similar system for .

8. The Homotopy Sequence of a Serre Bundle

1 (LEMMA). Let £ be a Serre bundle with a base point

Xq E tl £ , and let B be a subset of bs £ with bQ = pr £ (xQ) E B.
Then pr : TTr (tl£,pr£ 1 (B),xQ) ^ (bs £ ,B ,xQ ) and, in particular,
- 1
pr : TTr (tl£,pr£ (bg) ,bg) 7Tr (bs £ ,bQ) are isomorphisms for any
r > 1.
PROOF. pr is epimorphic.
^ r— 1
Let (p E Sph (bs £,B,bn ) . Define f : I -+ tl £ and two
r — 1 ^" r— 1
homotopies, H: I x I bs £ and G: Fr I x I -> tl £ , by

f~(lr_1) = xQ , H ((t1 , ...,tr _ 1 ) ,t) = 4>(t1 , ... ,1-t) , and

G(FrI r _ 1x i) = X(). SinceH(y,0) = pr£(f~(y)) y £ ir 1 , and

for
^ r—1 ~ r—1
G (y,0) = f (y) for y £ Fr I , there is a homotopy H : I x i -*■ tl E,
r- 1
which covers H and equals G on Fr I x I (see 4.1.3.6 ). Now

the formula (t ,. ..,tr ) = H~ ((t 1 ,...,tj._1),1-tr ) defines a spheroid

ip £ Sphr (tl £ ,pr £ ~ 1 (B) ,Xq ) such that pr^ ^ (ip) = <J>.

2) pr £* is monomorphic.
1 _

Let ip £ Sphr (tl £ ,pr £ (B) ,Xq ) , and suppose that the spheroid
prt^(ip) £ Sphr (bs E, ,B,b0) is homotopic to the constant spheroid.
 406

Choose a homotopy 0: I x I -* bs £ from pr£^(ijj) to the constant

spheroid, and define f : ir -+ tl £ and two homotopies,
H: Ir x I bs £ and G: F r l r x I + tl[ , by f~(lr ) = x,
ko '
h ((t 1 ,...,tr ),t) = $ ( (t1 ,...,tr _ 1 ,1-t),tr ), and

>p (t1, ,tr- 1 ,1 -t), if t = 0,

r
G ((t-j f ••• ^^
if (t1 ,...,t^) G Fr I tr / 0 .
‘0 '

Since H(y,0) = pr£(f~(y)) for y G I , andG(y,0) = f (y)

for G Fr I there exists a homotopy H : I x I -> tl £ which
covers H and equals G on Fr I x I. Now it is plain that
v ( ( t y . . . ,tr ) ,t) = H~( t^ , .. .,tr_-j 't ) ,1 -tr) defines a homotopy

y : ir x I tl £ from ip to the constant spheroid

The Action of tt^ (bs £ ,b Q ) on comp(FQ )

2. Let £ be a Serre bundle with base point bn G bs i .

-1
Set Fq = pr £ (b^) and define a right action of the group 7T^(bs£;,bQ)
on comp(FQ) as follows : for C G comp(FQ) and a G it^ (bs £ ,bQ ) ,
Co is the component of Fq which contains the origins of those paths
which end in C and cover loops in the class a. That this action is
well defined follows from Lemma 1: a path which ends in C and covers
a loop in the class a can be regarded as a spheroid of the pair
(tl £,Fq) with origin in C, and which is carried into a loop in class
a by pr ^ . If s G Sph 1 (tl C/FQ ,x1 )and s2 G Sph^ (tl £,FQ ,x2)
are two such spheroids, and w is a path in C with w(0) = x^ and
w( 1 ) = x2 , then the loops pr^(s ^ w ) and p r ^ ( s 2) are homotopic.
Now Lemma 1 implies that the spheroids s^w,s2 G Sph^ (tl£,FQ ,x2 ) are
homotopic, which, in turn, implies that the components of Fq containin
s^(0) and coincide. It is readily seen that this is indeed a
right action.
This action is compatible with the action of the fundamental
group oftl ^ on the homotopy groups of the fibers of £ (see 7 .3 ),
namely Cpr ^(o) = T^C for all C G comp(FQ) = 7Tq(Fq,Xq),
a G tt1 (tl ?,xQ) , and xQ G FQ . Moreover, if f: £ £;1 is a map of
Serre bundles, then

fact ab tl f : comp (FQ ) -+ comp (pr £ '” 1 (bs f (bQ )) ) ,

whereab tl f = [abtlf: FQ + pr £ '” 1 (bs f (bQ ))] , is a

 407

[bs f* :tt ^ (bs + tt 1 (bs ^ 1 ,bs f (bQ) )]-map.

3. I_f C € compiF^) and X q G C, then the isotropy subgroup

of tt ^ (bs £ /bQ ) at xQ (see 4.2.3.4) equals the image of the homo­
morphism pr£* : TT1 (tl^,x0) + tt 1 (bs ^ ,bQ) .

In fact, the equality Co = C means that there is a path

s: I •+ tl £ such that s(0),s(1) G C and s covers a loop in the class
a. This, in turn, guarantees the existence of a loop with origin x^
which covers a loop in the class a.

Construction of the Sequence

4. Let E, be a Serre bundle with base point xn G t l £ .

-1
Let bQ = pr^(x^) ,F q = pr £ (bQ) , and apply Lemma 1 to transform
the homotopy sequence of the pair (tl^F^) with base point Xq into
a new sequence. Namely, for each r ^ 1, we replace thehomotopy group
TT^_(tl ^,Fq,Xq) by TT^ibs^bg) , the homomorphism rel* : Trr (tl 5 /XQ ) +
7T^(tl ^,Fq,Xq) - by its composition with the isomorphism
pr ¿3*: tt (tl ^,Fq,Xq) -> tt (bs^bg) , and the homomorphism
3: 7T^(tl ^ , F q , X q) ■> TTr _'| ^f q /Xo^ ” the comPos:i-t:i-on
-1
A = 3 o (pr^*) :tt^ (bs £ ,b^ ) + TTr_'| 'X0 ^ # since the composition of the
inclusion rel: (tl^^x^x^) (tl^/F^^x^) with the projection
pr ? : (tl£,F 0 ,xQ) + (bsC/b 0 ,bQ) is simply pr (tl£,x0 ,x0) ■> (bs 5,bQ,b0),

we see that [pr £* : tt (tl^/F^x^) -> tt^ (bs £ ,b^)] © rel* is nothing
else but pr E,* : tt (tl£,xQ) ■> TTr (bs^,b0 ) . Finally, if we attach the
homotopy group TT^(bs^,b0) to the right of the resulting sequence by
means of the homomorphism pr : ^(tl^x^) -* tTq (bs £ ,b^) , we obtain
the sequence

... tt2 (Fq/Xq) > ^2 ^'X 0 ) > ^ 2 ( b3 ^/b0 )

tt1 (Fq ,x0) > tt1 (tl ?/XQ ) P— tt1 (bs S/bQ ) — (12^

^0 (F0 ,x0) ttq (tl g,xQ) Pr ^ * -> ^0 (bs?,b0) .

By 3.3, 7.3, and 2, there are right group-actions of

tt 1 (bs £, bQ ) on TTr (bs£,bQ ) , and of TT.|(tlf;/x0) on TTr (tl^,xQ )
and TTr (FQ,Xp) , and also a right action of ir^ (bs 5,bp) on the set
tt0 (F0 ,x0). The homomorphisms in*, pr , and A are compatible with
these actions, as required by Definition 5.11 (see 3.6, 4.5, 7.3, and 2).
Therefore, (12) is a Tr-sequence, called the homotopy sequence of the
 408

bundle E, with base point xQ .

5. Sequence (12) is exact.

This is a corollary of the exactness of the homotopy sequence
of the pair (tl£,F ) and of two additional and evident facts:
the kernel of pr £* : tt^ (1 1 E,, x^) Ti^bs^h^) equals the image of
in*: 77q (Fq /xq . TTgltl^Xg) ; and given a,B € ^ q ^Fq 'Xq ), there is
a € tt (bs £ ,b^) such that Q = ao if and only if in* (a) = in*(B).

6. Given a map f: E, of Serre bundles, the vertical

homomorphisms
A
•.. V P0 ' V ^ ( t l C,xQ ) Pr irr (bs £/bQ ) 7Tr- 1 (F0 ,X0 ) •**
I
(ab tl f)* tl f* bs f* (ab tl f )

... v (Fo'xo) 7Tr (tl£',X^) ■Pr a > TTr (bs

7Tr - 1 { F 0 ' X0 ]

where x^ = tlf(xQ) , b^ = bsf(bQ ) , and Fg = pr £ 1 1 (b^) ,

constitute a homomorphism of the first homotopy sequence into the second.

The commutativity of the first two squares follows from 1.7,

while the commutativity of the third follows from 4.2 and 4.3. The
compatibility of the vertical homomorphisms with the actions of the
fundamental groups was established in 3.6, 4.5, 7.5, and 2 .

The Most Important Special Cases

7. If tl £ is “-connected, then all the homomorphisms

A: TTr (bs£,b0) -* 7Tr_'| (F q 'xo ^ are isomorphisms • If tl £ is k-connected
and k < °°, then A: TT^fbs^bg) TTr_ ^ FQ'X0^ ‘*‘S an isomorPhism f°r
all r £ k, while A: 7Tk+ ^ (bs E, ,bQ ) ^(Fq'Xq) is an ePimorphism. If
tl £ isconnected, then the converse is true in both cases.
If bs E, is “-connected, then all the homomorphisms
in* : 7Tr ^F0'X0^ TTr ^t l ^'xo^ are isomorphisms. If bs ^ is k-connected
and k < “ , then in*: ^(Fq/Xq) TTr (t l 5 ,x()) is an isomorphism for
all r £ k, while in*: ^ (FQ ,XQ ) ^ (1 1 ^ ,xQ ) is an epimorphism.
If bs E, is connected, then the converse is true in both cases.
If F q is “-connected, then all the homomorphisms
Pr 5* : ]rr (tl £,xQ) ■> 7Tr (bs S,bQ ) are isomorphisms. If FQ is
k-connected and k < “, then pr ^* : 7Tr (tl^,xQ)-+iTr (bs^,b0 ) is an
isomorphism for all r * k, while pr : * k + 1 (tl £ ,xQ ) -* 7Tk+ (bs EJ f b Q)
 409

is an epimorphism. The converse is true in both cases (with no

supplementary conditions).

8. If the bundle £ has a section s such that s ^q) = xq /

then sequence (1 2 ) splits from the right at the terms tt (tl&xQ ) , and
s* : 7Tr (bs £ ,bQ) tt (tl £,xQ ) are splitting homomorphisms for any such
section, s: (bs£,bg) -> (tl£,xQ) .

PROOF. Since pr C ° s = idbs^, pr ^ ° s^ = id (bs £;,b^ ) .

9. I_f Fq is a retract of tl £ ,then sequence (1 2 ) splits

from the left at the terms TT^itl^Xg) , and any retraction p: tl £ FQ
induces splitting homomorphisms p*: 7Tr(tl£,x0) -> 7Tr (Fo'xo^#

PROOF. Since p o in = id FQ , P* ° i-n* = ^dTTr^F0'X0^ *

10. If the inclusion in: Fq -* tl £ is^ XQ-homotopic to the
constant map, then sequence (1 2 ) splits from the right at the terms
7T^(bs £,bg) . Moreover, given any Xp-homotopy h: FQ x I tl £ from
in to the constant map, consider the maps y^: Sph^iF^x^) -*
-»•Sphr + 1 (bs 5,b0) given by [y^ (<f>)] (t1 , ... ,tr + 1 ) = pr E, o h( ,... ,t_) ,tr+1)
(<f> € Sph^ (Fq ,xQ )) . Then the homomorphisms 7rr+ i ^ s i'bo^
induced by y^ split the sequence.

PROOF. Given an arbitrary spheroid <j) G Sph (Fq,Xq), it

suffices to find a spheroid ip G Sph^+ ^ ( tl £ ,FQ ,x^ ) such that Dip = <(>
and pr = y (<(>)- We can set, for example, ip(t1,...,t .) =
. h,

11 . I_f pr E, .is Xg-homotopic to the constant map, then

sequence (12)splits from the left at Trr (FQ ,x0). Moreover, given any
Xp-homotopy h: tl £x I bs E, from pr E, to the constant map,
consider the maps Sphr (F0 ,x0 ) -»■ Sphr+ ^ (bs E,,bQ ) defined by
[Yr (4>) ] (t1 , •. •,tr+1) = h(<f)(t1 ,...,tr ),tr+1) (<> e Sphr (FQ ,x0 )) . Then
the homomorphisms Trr (F0 ,xQ) -*• 7rr+ ^ (bs E, ,bQ) Induced by y^ split the
sequence.
PROOF. Given an arbitrary spheroid ip € Sph ^ (tl£,FQ,Xg) ,
it suffices to show that the spheroids y^ « 9(tp) and prE^(ip) , which
belong to sPhr+^(bsE,bQ ) , are homotopic. Clearly the formula
( (t , .. .,tr + 1 ) ,t) ►+ h(t|»(t1 , ...,tr ,ttr + 1 ) , (1 -t)tr+1) defines such a
homotopy.
12. If £ is a covering in the broad sense, then
Pr 5* : vTTr (bs ^,bQ ) is an isomorphism for r >, 2 , and a
monomorphism for r = 1. _If E, is a covering (in the narrow sense) ,
then, in addition, the map factA : (bs C,bQ)/Im pr C* F0' induced
 410

by A: it (bsC,bQ) > 7To (F0 'X 0 ) = F0 ' is invertible.

This is a corollary of the exactness of the homotopy sequence
of the bundle and ofthe fact that tt^CFq/Xq) = 0 for all r > 0
and TT0 (tl£,x0 ) = 0whenever E, is a covering in the narrow sense.

13. LetE, and E,' be Serre bundles with basepoints

Xq £ t l E, and Xq £ tl E,' ,and let f: E, E' , with tlf(Xg) = x^.
Set bg = pr r,(xQ ) , b^ =pr^'(x^), Fq = prf/ 1 (bQ) , F^ =p r f / ' 1 (b^).
From Propositions 6 and 5.19, we derive the following conclusions.
If bs f * : nr (bs£,b0 ) TTr (bs £ ',b^) , all r 5 1, and
(abtlf)*: tt (F.,x.) i t (Fi,x'), all r ^ 0, are isomorphisms, then
x r u u r u u
so are tlf* : tt^,(tl £ ,xQ) -*•tt (tl E,',x^), for all r ^ 1. If
tl f* :it (tl£,xn) ->• tt (tl £' ,x') , allr ^ 0 , and
IT U IT u
(abtlf).: tt (F.,xn ) tt (F',x'), all r ^ 0, are isomorphisms, then
x r u u r u u
so are bsf * : u ^ ( b s ^ , b ^ ) -+tt^ (bs £ ',b^ ), for all r ^ 1. Finally,
if bs f * :7Tr (bs£,bQ) Trr (bs^',b^) , all r ^ 0 , and
tl f * : TT^(tl£,XQ) tt^ (tl £ ',Xg) , all r ^ 0, are isomorphisms, then
so are (abtlf)*: 7Tr iFo /Xo^ 7Tr^F 0 /X0 ^/ ^°r r while
> ^t
(abtlf)*: '*‘ S an e P im o r P ^ is rn with trivial kernel.

We remark that in the last case, (abtlf).: TTn (Fn ,xn) -*

* u u u
TT0 (Fi ,xi ) is also an isomorphism if we make the additional assumption
that all the homomorphisms tl f * : TT^(tl£,x) tt^ (tl £ 1 ,tl f (x) ) with
x E Fq are epimorphisms. Indeed, let £ FQ be such that
tlf(x^) and tl f (x^) lie in the same component of the fiber F q ,
and let s 1 : I^ F q be a path with s ' (0) = tl f (x^ ) , s ' (1 ) = tlf (x2 ) .
Since tl f * : 7TQ(tl£,XQ) -* tt ( t l ^ ^ x ^ ) is an isomorphism, and
tlf* : tt^ (tl£ ,x^ )-+ TT^(tl^',tlf(x^)) is an epimorphism, there is a
path s: I tl ^ such that s(0) = x^ ,s(1) = x , and
tlf o s: I tl ^ 1is homotopic to the path [in: F q tl £ 1] o s'.
Then the loop pr £ ' ° tlf ° s is homotopic to the constant loop, and
from the fact that bs f * : tt 1 (bs £,bQ) + tt (bs £ ', b '
Q) is an isomorphism
it follows that pr£ o s is also homotopic to the constant loop. Now,
applying Theorem 4.1.3 . 6 to the map s, an arbitrary homotopy from
pr £ o s to the constant loop, and the constant homotopy of the map
s |Fr I 9 weobtain a homotopy from s to a path u: I -+ tl £such that
u(I) c: Fq, u (0 ) = x 1 , and u( 1 ) = x .
 9. The Influence of Other
Structures Upon Homotopy Groups

1. In this subsection we discuss the most elementary

properties of homotopy groups which are due to the presence of an
additional, group-like structure, compatible with the topology of the
space under consideration. The most important such property we shall
consider is simplicity.

The Case of Topological Groups

2. I_f X is a topological group and s: I X is an

arbitrary path, then the translation 7rr (X,s(0 )) + tt^(X,s(1)) coincides,
for any r ^ 0, with the isomorphism induced by the left group
-1
translation by the element [s(1 )][s(0 )]

In fact, there is even a canonical free homotopy from any

- 1
spheroid $ € Sph^(X,s(0)) to [s(1)][s(0)] <J> along s, given by
((t1 , .. .,tr ) ,t) h- [S (tn ts(0 ) ] ' % ( t 1 ,...,tr ).

3 (COROLLARY). The components of a topological group are

simple spaces. In particular, the fundamental groups of these
components are Abelian.

4. If X is a topological group, then, besides the

multiplication on the sets Sph (X,e=e ) defined in 1.1, there is
r X
another one, resulting from the group operation on X [the product of
two spheroids € Sph (X,e) is given by y*-»■ <J>( y ) * My ) ] . Moreover,
the second product makes also sense for r = 0, when the first product
is not even defined. Obviously, this new multiplication turns Sphr (X,e)
r s 0, into a group; the spheroids homotopic to the constant one form
a normal subgroup, and the resulting quotient group equals, as a set,
TTr (X,e). When r = 0, ir0 (X,e) equals the quotient group X/XQ , where
XQ is the component of e. When r £ 1, the new group structure on
•nr (X,e) coincides with the original one; in fact, given <p,ip € Sphr <X,e)
the formula
 412

defines a homotopy ir x i x from the original product of <p and ip

to the new one.
Let us add that the existing multiplication on X also turns
the set U Sph (X,x) of all spheroids of X into a group. The
x£ X r
spheroids homotopic to the constant ones form a normal subgroup, and the
resulting quotient group is canonically isomorphic to Tr^(X,e), for any
r ^ 1.
5. Every inner automorphism of the topological group X
induces automorphisms of the groups 7rr (X,e), and thus the inner right
action of X defines a right group-action of X on each group Trr (X,e)
The transformations induced by the elements of the subgroup Xq (see 4)
are all identical: if w: I X is a path with s (0) = e and s (1) = x,
and <f> E Sph (X,e) , then the formula (y,t) » [w(t)] <t>(y)w(t) defines
rr -1
a homotopy I x I X from <p to the spheroid y » x 4>(y)x.
Therefore, there are natural right group-actions of tt^(X,e) = X/X^ on
the groups tt (X,e).

The Case of Homogeneous Spaces

i
6. Let G be a topological group, and let H be a connected
subgroup of G. I_f (G,pr,X=G/H) is a Serre bundle, then given any
path s: I X, the translation tt (X,s(0)) + tt (X,s(1)) coincides,
for any r ^ 0, with the isomorphism induced by any transformation of
X (under the canonical action) which is given by an element of G
which takes s(0) into s(1).

PROOF. Let g € G be any element such that gs(0) = s(1),

and let s : I G be any path covering s. Since s ~ (1 ) and gs~(0)
lie in the same coset of H, and since the cosets of a connected group
are connected, there is a path w in the coset containing s (1 ) and
gs~(0) such that w(0) = s ~ (1) and w(1) = gs~(0). Now given any
(f> € Sph (X,s(0)), the formula

((t1 ,.. .,tr ) ,t) H-[Si (t) ] [Sl (0)]'1 M t r ----- tr )

where s 1 = s w, defines a free homotopy lr x I x from <p to

[s1 (1 )] [s (0 )] along a path homotopic to s (namely,the product

of s and the constant path).

7 (COROLLARY) . If^ H is a connected subgroup of the topolo­

gical group G and (G,pr,G/H) is a Serre bundle, then the components
of G/H are simple spaces. In particular, the fundamental groups of
 413

the components of G/H are Abelian.

8. Henceforth, (G,pr,G/H) will automatically be a Serre

bundle (as required in Theorems 6 and 7): G and its quotient space
G/H will always be closed smooth manifolds, while the projection
G -+ G/H will be a submersion and, by Theorems 4. 6 .1.3 and 4.1.3.4,
these properties imply that (G,pr,G/H) is a Serre bundle.

The Case of H-Spaces

9. A pointed topological space (X,e) is called an H-space

if there exists a continuous map y: X * X X such that y(e,e) = e,
and the maps X X, given by x ^ y(e,x) and x ♦+ y(x,e), are
e-homotopic to id X . The map y is called a multiplication, and e
is called an identity (or a homotopy identity). Usually, one writes
xy instead of y(x,y).
An H-space X is homotopy associative if the maps
X x x x x + X, given by (x^x^x^ *+ ( x ^ x 2 ) x 3 and (x^x^x-j) »+
x^ (x^x^) , are homotopic, and homotopy commutative (or Abelian) if the
maps X x x -> X, given by (x^x^) h- and (x^,x2 ) h - X£X , are
homotopic.
Given a H-space X, a continuous map v: X + X is said to
be a homotopy inverse if the maps X X, given by x *+ xv(x) and
x ^ v(x)x, are homotopic to the constant map which takes X into e.
-1
Usually, one writes x instead of v(x).
The primary examples of H-spaces are topological groups.
Every topological group (viewed as a H-space) is homotopy associative
and has a homotopy inverse, and every Abelian topological group is
homotopy Abelian.

10. The spaces of spheroids provide another important class

of H-spaces. For a given pointed topological space (X,x^), the sets
Sph (X,xn) with r ^ 1 become H-spaces if we equip them with the
r u ^
topology they inherit as subsets of C (I ,X) and with the usual
multiplication, and take the constant spheroid as the identity. In fact,
the map <J> ^ (const) $ is homotopic to idSphr (X,xQ) via the
const-homotopy Sphr (X,xQ) x I -+ Sphr (X,xQ ), given by (<|>,t) *+ <|>t ,
where c|> is the spheroid whose value at the point (t^,...,tr ) E ir
is given by the right-hand side of formula (3); further,- the map
cj) <f>(const) is homotopic to idSphr (X,xQ ) via the const-homotopy
which is similarly defined by formula (4). The H-spaces Sphr (X,xQ)
are homotopy associative and have a homotopy inverse for r ^ 1, and
 414

are homotopy Abelian for r ^2; the formulas in Subsection 1 again

provide us with the necessary homotopies.
Similarly, given any topological pair (X,A) with base point
xQ , the sets Sphr (X,A,xQ) with r ^ 2 are homotopy associative
H-spaces possessing a homotopy inverse. If r ^ 3, these H-spaces are
homotopy Abelian.

11. Every connected H-space is simple and, in particular, has

an Abelian fundamental group.

PROOF. Suppose X is a connected H-space with identity e,

<p £ Sph (X,e) with arbitrary r, and s E Sph (X,e) is a loop. Then,
the formula (y,t) *->- s(t)<j)(y) defines a free homotopy Ix I X from
the spheroid y h- ecj)(y) (which is homotopic to <j>) to the same spheroid
y e$(y), along the loop t *+ s(t)e (which is homotopic to s) . Thus,
Tr^(X,e) acts identically on the groups Trr (X,e).

12. It is clear that the second description of the homotopy

groups it (X,e), given in 4 for the case of topological groups, carries
over to H-spaces X with identity e. Generally speaking, the
multiplication that each set Sph^(X,e) inherits from X does not turn
this set into a group. However, this multiplication does induce the
usual group structure on tt (X,e), r ^ 1. Moreover, if X is homotopy
associative and has a homotopy inverse, then the set tt ^ (X ,e ) is a
group.

The Local System of Homotopy Groups

of the Total Space of a Principal Bundle

13 (LEMMA). Let £ bea principal bundle with structure

group G, and let u,v: I tl £ be paths satisfying pr £ o u = pr £ ° v.
— g0'g i G G are such that u (0)g0 = vQ and u(1)g = v (1 ) [here G
acts canonically from the right on tl £ ; see 4.3.2.10], then the diagram
T
Trr (tl £,u (0 )) ------ > TTr (tl ^,u (1 ))

T
Trr (tl£,v(0 )) --- 2 -- » Trr (ti C,v( 1 ))

where the vertical isomorphisms are induced by the transformations

x h- xgQ and x >+ xg , commutes.

PROOF. Recall that the canonical right action tl £ x G -> tl E,

 415

is free and its orbits coincide with the fibers of £• Since u(t) and
v(t) lie in the same fiber, for each t €I there is a unique g(t) € G
such that u(t)g(t) = v(t). Therefore, if h : ir x I tl £ is a free
homotopy from <t>Q e Sphr (tl £ ,u (0 )) to <|> € Sph (tl £ ,u(1)) along u,
then (y,t) h(y,t)g(t) yields a free homotopy from the spheroid
y ** ^giyigQ to the spheroid y c|>.j(y)g along v.

14. Suppose that £ is a principalbundle, b € bs £, and

r ^ 0. The right canonical action of the structure group of £ on t l £
induces isomorphisms between the homotopy groups tt(tl£,x) with
—1 r
x £ pr 5 (b) , and so we may identify these groups. We shall call the
resulting group (which actually is a group for r >0 and a set with
identity for r = 0 ) the r-th homotopy group of the space tl£ over
b, and we shall write TT^(tl£,b) .
Given a path s : I + bs £ , we define Tg : (tl £ ,s (0 ))
tt^ (tl £ ,s (1 )) as the translation ^ ( t l ^ s (0 ))
Ts~ : “ n^itl^s (1 ))
along any path s~ which covers s. That this is well defined follows
from Lemma 13. Obviously, Tgare homomorphisms and satisfy conditions
2.1 (i)-(iii). Therefore, we have produced a local system on bs£ ,
(bs £ ,{tt^ (tl £,b)},{Ts}) , which we call the lower local system of the
r-th homotopy groups of the total space of £.
It is clear that the local system on tl£ induced by this
system via pr £ is nothing but the usual local system of the r-th
homotopy groups of tl£ .
Given a monomorphism cp: G -* G' and r ^ 0, every cp-map of
the Steenrod G-bundle £ into the Steenrod G 1-bundle £' induces a
homomorphism of the lower local system of the r-th homotopy groups of
tl £ into the corresponding system of tl £ 1 .

The Homotopy Sequence of a Principal Bundle

15. Let K be a principal G-bundle with base point xQ € tl £.

If in sequence (12) we replace (F 0 'xo)by the Pair (G,e=eQ ) [which
is canonically homeomorphic to (fq'xq^ v^a ^ ^ x 0 ^ an<^ attach 1
to the right of the resulting sequence, we obtain

... 1T2 (G,e) in* > it2 (tl S,x0 ) pr •*"»TT2 (bs £,bQ ) — (G,e) -in*->

TT1 (tl 5,x0 ) TT1 (bs £,bQ ) — TT0 (G,e) • (13)

1T0 (tl 5 ,x0 ) -Pr 7tq (bs 5»b0 ) ---► 1

(bQ = pr £ (xQ ) ) .
 416

Recall that 7T0 (G,e) is a group(see 4) and that 7T^(G,e)

is Abelian (see 3). It is immediate that A: tt^ (bs £ ,bg ) 7iQ(G,e) is
a group homomorphism. Moreover, TT^iG/e) acts from the right on
TTr (G,e) (see 5 ), while tt^ (bs £ ,b^ ) acts similarly on both i\^ (bs £ ,bQ
(see 3.3) and tt (tl£,xQ) = Trr (tl£,b0) (see 14). The canonical right
action tl £ x G tl £ induces a right action of G on TTgitl^Xg) =
= comp(tl£) , and thus a right action of tt^ (G,e ) on tt^ ( 1 1 £ ,x^ ) .
[We regard tt ^ (G ,e ) as the quotient group of G by the component of
e; see 4. The action of this component on tt ^ (11 £ ,x^ ) is identical.]
Finally, it is clear that the homomorphisms in*/ pr*, and A are
compatible with the above actions, as required in Definition 5.11.
Consequently, (13) is a Ti-sequence, called the homotopy sequence of the
G-bundle ^ with base point x^.

Obviously, pr : tTq (tl £ ,xQ ) 7TQ (bs£;,b0) is an epimorphism,

and the partition of tTq (tl £ ,x^ )into the orbits of tt^ (G ,e ) is
exactly zer(pr ^ ) . Therefore, sequence (13) is exact.
Given a monomorphism tp: G 1 -> G, every cp-map f of the
principal G f-bundle £f with base point £ tl £ ' into the principal
G-bundle £ with base point x^ £ tl £ , such that tlf(x^) = x^,
induces a homomorphism of the homotopy sequence of £ 1 into the homotopy
sequence of £•

10. Alternative Descriptions

of the Homotopy Groups

1. The spheroid DS ° ID £ Sph^(Sr ,ort^) (see 1.2.8 .9) is

called the fundamental spheroid of the sphere Sr , denoted IS, and we
let sphr denote the element of 7rr (Sr ,ort ) that it defines. The

spheroid ID £ Sph^(Dr ,Sr \ort^) is called the fundamental spheroid

of the ball Dr , and we let kug^ denote the elementof tt r (Dr ,Sr ~ 1,ort )
that it defines.
Obviously, 9 (ID) = IS, whence 9(kug^) = sph^_^.

We let Sph^(X,x^) denote the set of all continuous maps

from the pointed space (Sr ,ort1) into the pointed space (X,x ), and

defineIS# : Sph°(X,xQ) + Shpr (X,xQ) by IS^(<J>) = <\> o IS. Clearly,

this map is invertible, and 1.3.7 . 6 implies that two maps,
0 u
4>,ip £ Sph^ (X,Xq) , are homotopic if and only if the spheroids IS (<f>) ,

IS ({¡i) are homotopic. Consequently, replacing our "cubic" spheroids

 417

and their homotopies by the "spheric" spheroids from Sph°(X,x0 ) and

their homotopies, we are led to an equivalent description of the set
TTr (X,xQ) .
It is readily seen that the identity spheroid id Sr belongs
to the class s ph^ and that the element of ir^iX^^) given by a
spheroid f: (S^ort^) (X,xQ ) equals f*(sph ).
u
If r £ 1, then IS transfers the multiplication in
Sphr (X,XQ) to Sph^(X,XQ) The resulting multiplication in Spli^iX^^)
may also be described directly: the product

: (Sr ,ort^) = (s\ort^) 0 ... ® (s\ort^) + (X,Xg)

of the spheroids cf>,ip: (S^ort^) = (S1,ort ) ® ... 0 (S^ort^) (X,xQ)

is given by
2
(y1 , Y 2 ’ • ••'Yr > ' if i m y 1 ^ 0,

( Y 1 * Y 2 ' • • * , y r ) =
(14;
2
^ (y1 ' Y 2 '*““ 'yr * ' if i m y i ^ °'

where y-| /y 2 * •••/yr are complex numbers of modulus 1 , and im

denotes the imaginary part. The multiplication that this operation
induces on tt^ îX jX q ) coincides with the existing one. One can use
(14) to study directly the homotopy properties of the multiplication in
Sph^(X,Xg) and get an independent description of the homotopy groups
tt^ Î ^ X q ) in the language of spheric spheroids.
It is particularly simple to describe in this language the
spheroid (f) ^ = (IS^) ^ ( [IS^ (cj))] ^) , that is, the inverse of the
O —1
spheroid 0 € Sph^ (X,Xq): (£> (x^ ,x^ ,x^ ^ •,x^^^ ) ~ x -j/—^2 '^3 '***' 1 ^*

3. Let Sph^(X ,A ,X q ) be the set of all continuous maps

(Dr ,Sr _ 1 ,ort ) -> (X,A, xQ ) , and define ID^: Sph°(X,A,xQ) + Sphr (X,A,xQ )
u
by ID (<J)) = <p 0 ID. This map is invertible, and 1.3.7. 6 implies that
two maps, (p,ip € Sph°(X,A,x ), are homotopic,if and only if the
n r nU jj.
spheroids ID (<t>) and ID (rp) are homotopic. If r 5 2, then ID

transfers the multiplication from Sphr (X,A,xQ) to Sph^(X,A,Xq ).The

resulting multiplication in Sph^(X,A,XQ> may also be described
directly: given two maps,

(Dr ,ort1> = (S1 ,ort^) ® ... ® (S1 ,ort^) (1,1) ■+ (X,Xq ),

formula (14), where Y ‘ • ’ Yr --\ are comPlex numbers of modulus 1

and yr G I, defines a map <t>ip: (Dr ,ort1) (X,xn ), and this map
 418

belongs to Sph°(X,A,xQ) whenever <p,\p £ Sph^(X,A,xQ ). As in the

-1 it -1 # -1
absolute case, <p = (ID ) ([ID (<}>)] ), that is, the inverse of the
spheroid <j) £ Sph°(X,A,xQ) , is given by <J>1 (x1 ,x^ ,x 3 , ..., x ^ ) =
= (p(x^ ,“ x 2 ,x 3 ,•••/ •
Therefore, by replacing the cubic spheroids by spheric ones,
we get an adequate description of thehomotopy groups tt (X,A,Xq).
Obviously, the identity spheroid id D belongs to the class
kug^, and the element of tt (X,A,xQ) given by a spheroid

f: (Dr ,Sr~ 1 ,ort1) + (X,A,xQ ) equals f*(kugr >.

Unlike the cubic sets Sph^(X,x^,x^) and Sph^(X,XQ>, the
sets Sph^(X,X q ,X q ) and Sph^(X,XQ)
are not identical, being merely
If _ 1 #
related through the canonical invertible map (ID ) o (IS ):
: Sph^(X ,X q ) Sph^(X,Xq ,Xq ); this map can be described more directly
as <p (p o D S .

The boundary 9cJ) of a spheric spheroid <p £ Sph°(X,A,x )

0 r—”l
is the element 9(J) £ Sph^^ (A,x^ ) given by d(p = [ab4: (S ,ort^ ) -> (A,Xq)].
U №
Clearly, IS o 9 = 9 ° ID , which demonstrates that this definition
i
of the boundary leads to the same boundary homomorphism 9 : tt^(X,A,Xq) -+
7Tr_-| (a ^xq) as the cubic theory does.
4. For each continuous map, f: (X,x^) -+ (X',Xq) or
f: (X,A ,Xq ) (X ',A 1 ,Xq ) , we have the map f^: Sph^(X,xQ ) -> Sph^(X',Xg)

and respectively f^: Sph^(X,A,xQ) Sph^ (X 1,A ’,x^ ), given by f^(<j)) =

= f O (p. Trivially, IS^ o f = f is^

f„ = f„ o ID^ ,
o and ID^ o
ft ft ft ft
so that these maps f^ lead to the same induced homomorphisms
f* : 7Tr (X,xQ) + Trr (X’,x^) and f* : TTr (X,A,x0> tt^ (X 1 ,A 1 ,x^ ) as the
cubic theory does.
A free homotopy from a spheroid (pn £ Sph°(X,x ) to a spheroi
0 u r 0
<j>1 £ Sphr (X,x ) along a path s: I ->■ X is definedas a usual homotopy

Sr x i x from <j>0 to <J>1 which takes (ort^t) into s(t) for

any t € I. If h issuch a homotopy, then h o (is x id I) is a free
# #
homotopy from IS (<j)Q)£ Sphr (X,xQ) to IS (4>1 ) £ Sphr (X,x ) along s.
Therefore, the spheric free homotopies along s yield the same
isomorphism Tg : ïïr (X,xQ) TTr (X,x1) as do the cubic ones. It is
readily seen that all this carries over to the relative case.
 419

An Alternative Proof of Theorem 4.7

5. Using spheric spheroids and the homotopy sequence of a

pair, one can give a second proof of Theorem 4.7, which is less direct
but, in return, more transparent.
2 2
Consider first a model situation: X = (D ,ort ) V (D ,ort.),
1 1 1 1
A = (S ,ort..) V (S ,ort1), xnis the center of both bouquets,
-1 - 1
a = imm 1 *(kug2), and 6 = imm2 *(kug2>. Since 3 (a 6a) = (3a) OB) Oa) =

= T~ 3 6 (see 3.4) and T~ 3 B = 3T~ 6 (see 4.5), we have 3 (a ^ 6a) =

pa da da
= 3T^S. Moreover, since X is contractible,3 is an isomorphism
(see 6.5), and hence a ^6a = T„ B.
9a
In the general case, pick two arbitray spheroids
(J>,^ E Sph^(X,A,x^) in the classes a , 6/ and define

f: ( (D2 ,ort 1 ) V (D2 ,ort1 ) , (S1 ,ort1 )V ( S ^ o r t ^ ) -* (X,A)

by
2
f(imm^(x)) = (J)(x ) , f(imm2 (x)) = i^(x), x E D .

Since (p = f o imm^ / ip = f o imm2 , we have

a = (f o imm1 )*(kug2), 6 = (f o imm2 )*(kug2),

and hence
-1 -1
a 6 a = f * ([imn^*(kug2 )] [imm2 *(kug2 )] [imm^* (kug2 )] =

= f* (T. . n ximm. .(kug0 )) = T. 6 .

* Soiimn^ * (kug2) 2* 2 3a

Spheroids in Spheroid Spaces

6. Let (X,xQ) be a pointed space. With every spheroid

(j) E Sphr+s(X,XQ> we may associate an r-spheroid of the space
Sphs (X,Xq) [see 9.10] with base point the constant spheroid, by means
of each of the formulas

[^ (t ^ ,...,t^)] (u-j ,•••,ug) = 4)(t^/.-wt^,u^,...,us),

and
[ij;(t^,...,t^)] (u1 , ...,ug) = (J)(u^,.../us ,t>|,..., t^) .

This leads to two maps, Cub,Buc: Sphr +s (X,xQ) -> Sphr (Sphg (X,xQ),const) .
These maps are invertible and both they and their inverses take
homotopic spheroids into homotopic ones. If r > 0, then the
 420

multiplication in Sphr+s(X,xQ ) is transferred by Cub into the usual

multiplication in Sph^(Sphg (X, ) , const) [that is, the multiplication
of spheroids]. If s > 0, then the multiplication in Sphr +s(X,xQ )
is transferred by Buc into the multiplication in Sph^(Sph^(X,X q ),const
which arises from the fact that Sphg (X,XQ) is an H-space with identity
element const (see 9.11 and 9.10). Therefore, when r +s > 0, the maps
Cub and Buc define group isomorphisms, cub,buc: TTr +s (X,x0 ) -+
-> 7Tr (Sphg (X,xQ ),const) [the group structure of ttq (Sphg (X ,xQ), const) ,
s > 0, was explained in 9.12]. Therefore, one can identify TT^iX^Q)
with any of the groups (Sph^ (X,x^), const) with q £ r.

Finally, note that the isomorphism cub: 7Tr (X,Xg) -+

-+ 7Tr_'| (Sph^ (X ,Xq ),const) appears also as the isomorphism
A: (X ,Xq ) Tir _ 1 (Sph^ (X,xQ) ,const) in the homotopy sequence of the
Serre bundle £ = (C(1 ,0;X ,xQ ),ab C ( [in:Fr I -> I] ,id) ,X =C(Fr 1 ,0;X,xQ )) f
whose fiber over the point Xq is Sph^(X,XQ>. That £ is a Serre
bundle follows from 4.1.4.2.

11. Additional Theorems

1. Let d^,...,dm be pairwise disjoint balls in IR such

r O
that d^ , ...,d^ c D , and let g E Sph^ (X,A, x^) be a spheroid with

g(C) c A, where C = Dr \ Um „ Intd. . Let t .: Dr + Dr denote the

^ ----- 1 =1 l l ----------
-

map t± (y) = (center of d^) + (radius of d^)y. Suppose further that

the segments joining the points t^(ort^) , . . . , ( o r ) to ort^ are
contained in C. Then, for r > 2,

Y ■ (T Y ^ I T T2 )...(T V '
1 2 m

where y € irr (X,A,x0) and y^ 6 irr (X, A ^ o t ^ (ort1) ) are the elements
represented by the spheroids g and g ° 6 Sph^(X,A, go-t ^ (ort^ ) ) ,

and s^ is the path In A given by s^(t) = g ((1-t) (ort^ ) + tort^) ,

i = 1,...,m. The same conclusion holds true for r = 2, provided
d 1 ,...,dm are indexed naturally, i.e., each of the 2 -frames
(xi (ort1) - ort 1 ,xi+ 1 (ort1) - ort^) defines the natural orientation

of* IR .

PROOF. We proceed by inductionon m. We make two preliminary

remarks, denoting by t the (rectilinear) path in C given by
¿^(t) = (1 - t ) ( o r t 1) + t o r t 1 .
 421

First remark: for given d„,...,d , it suffices to prove the

r 1 m
theorem when (X,A,x0> = (D , C , o r t ^ ) , g = rel id D , and si = l ± .

Indeed, g*: (Dr ,C ,ort ) -+ iTr (X,A,XQ) takes the class of the spheroid
r
rel id D into y, while it takes the class of the spheroid
translated along £., T y.. into
1
s. 1
l
Second remark: for a given m, it suffices to prove the
theorem for a standard choice of d^,...,d^, namely, for the balls of

radius 1 /2 m centered at 2 L-J-Ort0 , m--^ort0 , ..., — ort., -— —ort0 .

m 2 m 2 m 2 m 2
To see this, consider, along with these standard balls and the
corresponding C, arbitrary balls d*,...,d^ satisfying the
conditions of the theorem, with the corresponding C 1, t !, £!. Clearly,
r r r1
there exists a continuous map h: D D , which is S -homotopic to
id Dr and satisfies h(C) c= C ', h o t. = t!, and h o I . = £!. Then
r r 1 1 1 1
rel h* : tt^.(D ,C,ort^) -* tt^ (D ,C*,ort^) takes the class of the spheroid
r r
id D into the class of id D ,while taking the class of the spheroid
T^f translated along , into the class of the spheroid
translated along i = 1 ,...,m.

Now to our induction. The cases m = 0 and m = 1 are

trivial; consider m = 2.By our remarks, we may assume that
(X,A, X q ) = (Dr ,C,ort ) , g = id, and are the standard b a l l s
(with radius 1/4 and centers ort^/2 and -ort2 /2). Let p denote
r
the rotation of D by an angle <p around the subspace given by the
r r
equations x^ = = 0. Further, consider the homotopies D x I D
given by

(y,t) [(1+t) PTTt/ 2 (y) - 2 ort2 ] / 4

and
(y,t ) » [(1 +t) P1Tt/ 2 (y) + 2ort2]/4

(y € D , t E I). These can be viewed as free homotopies of spheroids

of the pair (D ,C) ; as such, they connect and with two
spheroids with origin 0, a^ and along two paths, which we call
-1 -1
u^ and u2 * Obviously, u^ and u^ are bothhomotopic to the
rectilinear path L which joins 0 to ort^. Consequently, the
product of the classes obtained by translating the classes of and
t2 to ort^ along and is thesame as theproduct of the
classes obtained by translating the classes of and * o^ to ort^
along £, that is, the class of the product of the spheroids a^ and
C$2 / translated to ort^ along I. It remains to observe that there
 422

is a free homotopy from the product a^a 2 to id Dr along Z, for

example, a rectilinear homotopy.
Finally, let m ^ 3. As in the case m = 2, we shall assume
that (X,A,Xq) = (Dr ,C,ort1), g = id, and d 1 f...,dm are the standard

balls (with radius 1 /2 m and centers —— -ort., — ortn , ..., ^ mort9 ,

m 2 m z m ^
-—m-ort„)
2
. Let d be the ball of radius 3/2m centered at —m ^ort20

(note that d „ ,d cd) , and let t : Dr -> Dr bedefined by

m-1 m 2

t (y) = (center of d) + (radius of d)y. Further, let Z, u, and v

be the rectilinear paths joining t (ort,) to ort., t (ort.) to

1 1 m- 1 1
t (ort,), and t (ort„) to t (o r t j , respectively. We let
1 m l 1
r r
6 € tt^(D ,C,t (ort^)) denote the class of t. Since theproducts

uZ , vZ are clearly homotopic to the paths /

1 m^ = Ti /1 r - - Tim —2V t ' V . - i V . ' - 1161

Now apply the theorem, first for the case of two factors, and
then for the case of m- 1 factors, to conclude that

’. V i V , = s- (17)
and

T£ , V " T f ,Tm- 2 T 6 = Y- (,8>

1 m- 2
r
(In the first case, the theorem is applied to (X,A,xQ) = (D ,C,t (ort ))
1
g = t, and the balls (d ) ,t 1 (d ) , while in the second case we
t
— —

m- I m
take (X,A,Xq) = (Dr ,C,ort1), g = id, and the balls d ^ ,...,d^_ 2 ,d ) .
At last, (16), (17), and (18) yield (15).

2. Let X = lim (X^.,c^) , where X^, are -spaces, and le

x £ X, Xq £ Xq, x^ E X^ ,... bepoints suchthat imm^(x^) = x. If for
some r all the homomorphisms ^r ^Xk + 1 'Xk + 1 ^ are
isomorphism, then so are (imm, )+ ..: tt_ (Xn_,x^) tt (X,x) (with the same
' jc r r K. k. r
r) .
PROOF. Notice that, according to Theorem 1.2.4.5, every
27
spheroid I X may be expressed as the composition of a spheroid
27
I + X..
with the embedding imm, , for 1 large enough; similarly,
r
every homotopy I * I -> X is the composition of some homotopy
27
I x I > X^ with imm^, for 1 large enough. Now the fact that
(imm^.)* are epimorphisms and monomorphisms is seen to be a consequence
of the analogous properties of the compositions
 423

{h - l ] * r ° (4>l-2 )*r ° ••• ° (V * r :V W " V W *

12. Exercises

1. Let (X,Xq ) be a pointed space and suppose that there is

given a right group-action of tt^(X,Xq )
on a group G. Show that there
exists a local system of groups, (X,{G },{T }), with G = G, which
x s xQ
determines the given action.

2. Let (X,A) be a cellular pair with base point x^ £ A,

and suppose that X is countable. Show that all the groups 7Tr (X,A,XQ)
are countable.

3. Let (X,A) be a cellular pair with base point x^ £ A,

and suppose that the groups tt^ (X /Xq ) and ^ ( A ^ q) are finitely
generated (for all r ^ 1). Show that if X is simply connected, then
the groups 7Tr (X,A,XQ) with r ^ 2 are also finitely generated.

4. Let £ be a Serre bundle, and let E be a subspace of

tl ^ such that (E,pr5 l^fbs 5) is also a Serre bundle. Show that for
I-k
any point x £ E and any r ^ 1

in* : Tir (pr5 1 (pr ^ (x) ) ,pr 5 1(pr £ (x) )flE,x) + TTr (tl£,E,x)

is an isomorphism.

5. Show that if the base of a covering is k-simple, then its

total space is also k-simple.

6. Let r > 0 and s > 0. Show that the homomorphisms

cub,buc : tt (X,xn) -* tt (Sph (X ,xn ),const)

IT •s u r s u
ITs
differ only by the constant factor (-1 )
 424

§2. THE HOMOTOPY GROUPS OF

SPHERES AND OF CLASSICAL MANIFOLDS

1. Suspension in the Homotopy Groups of Spheres

1. The suspension of a spheroid 6 £ Sph^iXjXp) is the

spheroid su <j> £ Sphr + 1 (su (X,xQ) ,bp) , given by su <{>(t.j /
,•.•
.•./
,tr
r++ 11 ) =
= pr (0 (t1 ,...,tr ) ,tr+ 1) , where pr = [pr: X x I -+ su(X,xQ )]. Obviously,
suspensions of homotopic spheroids are homotopic, the suspension of the
product of two spheroids of positive dimensions equals the product of
their suspensions, and the suspension of the constant spheroid is again
the constant spheroid. Consequently, the mapping ^ su yields a
homomorphism Trr (X,XQ) -> (su (X, ) ,bp) , for any r ^ 0. This
homomorphism is also called suspension and is denoted by su.
Recall that we have already defined the suspension of a
continuous map on two occasions: in 1 .2 .6 .2 , for maps of topological
spaces, and in 1.2.8 .5, for maps of pointed topological spaces. The
present, third definition, is more special; it concerns maps from the
IT
pair (I ,FrI ) into pointed spaces, and ha^ no intersection with the
previous ones. At the same time, it is compatible with the second
definition, in the sense that we may obtain the third definition from
the latter by shifting from spheric spheroids to cubic ones. More
precisely, the spheroids in Sph^(X,x^), being maps between pointed
spaces, have suspensions in the sense of 1 .2 .8 .5, and the diagram

Sphr (X,Xq) ----- > Sphr + 1 (su(X,xQ),bp)

commutes.
Let us add two important, yet obvious remarks. Firstly, if
f: (X,xQ) + (Y,yQ) is continuous, then the diagram

+ T <s u (X,xQ ) ,bp)

f (su f ) *

TTr + 1 (su(Y,y0) ,bp)

(where su f is understood as in 1 .2 .8 .5) commutes for any r ^ 0.

 425

Secondly, since (su (Sn ,ort^ ) ,bp) = (Sn+\ort^), our construction, when
applied to spheres, yields a homomorphism Tr^S^ort^ + Trr + 1 (Sn + 1 ,ort 1 ) .
The main theorem of this subsection, Theorem 4, is devoted precisely to
this homomorphism.

2. Alternatively, one can describe su: TT^iX^^) ->

TTr + 1 (su (X,xQ) ,bp) by means of the map lp: (X,xQ) Sph 1 (su (X ,xQ ),bp)
given by [lp(x)](t) = (x,t). Namely, every spheroid $ £ Sphr (X,xQ)
is taken into su cf) by the composition
lp „
Sphr (X,xQ) ---- > Sphr (Sph1 (su(X,xQ ),bp),const) —Cub > Sphr +1 (su(X,xQ),bp)

(see 1.10.6), and hence the homomorphism su: TTr (X,xQ) nr + 1 (su(X,xQ),bp)
may be defined as su = cub o lp^ (to check this facts is routine).

A new description of the homomorphism su emerges if we view

su(X,xQ) as the quotient space of the cone con(X,xQ) by its base
(which is identified with X). Indeed, given any (p £ Sphr (X,xQ ),
consider the spheroid in Sphr + 1 (con (X,xQ) ,X,xQ) defined as
(t^ ,.. .,t^ +1 ) (4> (t^ , ...,tr ) ,t r + 1 ) . The latter is taken into su <p
by the map pr^: Sphr + 1 (con (X,xQ ),X,xQ) + Sphr + 1 (su(X,xQ),bp), where
pr = [pr: con(X,xQ ) su(X,Xq)] , and is transformed back into <p by
the m^p d: Sphr + 1 (con (X ,xQ ) ,X ,xQ ) -+■ Sphr (X,xQ). Consequently,
su: TTr (X,XQ) ■+ 'ur + ] (su (X,Xq ),bp) is nothing but the composition

-1
TTr (X,xQ) ---- > TTr + 1 (con (X,xQ ) ,X,xQ) Pr* >7T
r + 1 (su (X,xQ) ,bp)

(9: TTr+>|(con (X,Xq ) ,X,Xq ) TTr (X,XQ) is invertible because the cone is
contractible; see 1.6.5).

If, for example, (X,x^) = (Sn ,ort^), then con(X,x^) = Dn+\

s u (X,Xq ) = Sn+\ and pr = D S .

The Suspension Theorem

3 (LEMMA). Let K and L be closed disjoint subsets of im ,

and assume that K is covered by a finite number of k-dimensional
planes, that L is covered by a finite number of 1-dimensional planes,
and that K n Fr im <= im_1 x [0,1/2) and L il F r l ” c im_1 x (1/2,1].
If k+1 £ m-2, then there is a Fr lm -homotopy, F: im x I im , such
that:
(i) the maps im -> im which form F are homeomorphisms,
and their inverses also form a homotopy;
 426

(ii) F connects id lm with a map I™ -+ im which takes

K into im ^ x [0,1/2] and L into im 1 x [1/2,1].

PROOF. Suppose first that L <= im_1 x [1/2,1]. Since the

lines which intersect both K and L constitute a set which can be
covered by a finite number of planes of dimension k+ 1 + 1 £ m- 1 , there
is a point a £ Int im_1 x (0 ,1 /2 ) which does not lie on any such line.
Let us project K and L on Fr lm from a. Their images are closed
disjoint subsets of Fr lm , and so there is a Urysohn function,
a: Fr im + I , for these images. Choose e > 0 such that the
similarity transformation with center a and coefficient e pulls I™
into im x [0 ,1 /2 ], while the similarity transformation with the same
center and coefficient 1/ (1 — e) does not take K fl (lm x [1/2,1]) out
of im . Now let 4> s I ■>■ im denote the rectilinear path which joins
a to a given point x £ Fr I , and let ^ : I -* I denote the homeo-
morphism which takes linearly [0 ,(1 +t)/2 ] onto [0 ,(1 -t)/2 ] and
[(1+t)/2,1] onto [(1-t)/2,1]. The formula

F(<J> (u) ,t) = yx

<p o \rt(
b.t . * , v (u) ,
1 -e)a(x) '

where x £ Fr lm and t,u £ I, defines the desired homotopy.

To reduce everything to this special case, we produce, in the
general case, a Fr im -homotopy F: im x i -* im which satisfies property
(i) and connects id lm with amap which takes L into im ^ x [1/2,1].
We can define such a homotopy F by

F((x,u),t) = (x,i|j_ t 5 (u) ) [x £ Im 1, t ,u £ I],

where 6 £ (0,1) is any numbersuch that L <= im ^ x [(1-5)/2,1].

4. su: Trr (Sn ,ort^) -> tt^ +^ (Sn+\ ort^ ) is an isomorphism for
r £ 2 n- 2 and an epimorphism for r = 2 n- 1 .

PROOF, a) To see that su is epimorphic for r £ 2n-1 , we have

to verify that, given any spheroid <J> £ Sphr + 1 (Sn + 1 ,ort 1 ) , there is a
spheroid \p £ Sph^ (Sn ,ort ) such that su if; is homotopic to <J>.
The proof is quite simple when 4)(Ir x [0,1/2]) is contained
in the upper half fx n + 2 * ofsU + \ while <\>(ir x [1 /2 ,1 ]) is
contained in the lower half {x n £ 0 } of Sn + 1 (here x ,. ..,x
n + 2 n +2 1r n +2
are the standard coordinates in ]R ). In this case, $(I x (1/2))
lies in the intersection of these twohemispheres, i.e.,
c})(ir x (1 / 2 )) c Sn , and the required if; is given by

iM t-j , . . •, t r ) = <p ( t 1 , . . . , t r , 1 / 2 ) .
 427

The formula

((t^,...,t^+^),t) ^

<t> (t1 ,...,tr , y ^ ~ ) , if 0 £ t „ £ 1 -t

r +1 2 '

pr (^(t^,..., t^ ,— ) ,t^+ ^) 1 -t
if £ t
2 r +1 ‘ T*
t --t
(t t r_+4_ ) 1 +t
\ L , , , . . . , L-^_ ,
'1 ' ■
i
1 -t
), if £ t £ 1,
2 " "r+1

where pr = [pr: Sn x i -* Su (Sn ,ort )] , defines a homotopy from to

su .
■1
Consider now the more general case where (ort )c
n +z
<= I x [0,1/2) and (¡) ^(“ortn +2 ^ c ir x (1/2,1]. This case is readily
reduced to the previous one. Indeed, pick e > 0 such that the last
coordinate of (y ) is * 1 -e (>, -( 1 -e )) for all y € I [ 1/ 2 , 1 ]
(respectively, for all y G I [0 , 1 / 2 ] Now define
n +1
(Sn+1rort ^) (S iOrt.j) by h (x ^ ,
'Xn + 2 ^ =
(x ,xn+1'(xn+2 " <1 2 ))1/2)/(1 - (1-e 2 ))1/2 ,
1 '
if X ^ 1 -e ,
n +2
(x ^ 2 J /2 if x n +2 £ 1- E .
1 ' ’x n + 1 ' 0 ) / ( x + + X n +1 ) ‘
.n+1
[This map stretches the polar caps of S"4’ ', defined by the
inequalities x n + 2 ^ and Xn +2 ^ H )/ over the upper and lower
hemispheres, respectively, and contracts the equatorial belt
,n
- (1~e) < Xn + 2 < 1“£ to the equator ] It is clear that is
ort^-homotopic to id Sn+^ [such a homotopy moves each point x G Sn + ^
uniformly towards h(x) along the shortest arc of the great circle
£
passing through x and h(x)]. Moreover, h o <j) takes I x [0,1/2]
into the upper hemisphere, and takes I x [1/2,1] into the lower
hemisphere.
n^ 1
Finally, in the most general case, we triangulate S in
such a way that ort. ort i 0 and -ort i 0
becomes a vertex, while
-1 1 n+z n+z
become interior points of (n+1 )-simplices, and then consider a recti­
linear triangulation of in which ensures that (p has simplicial
approximations. Let be such a approximation. Then the sets
-1 y1
K = <J)1i1 '
( o r"n
t_ 0) and L = 1 ' (-ort_i0) satisfy the conditions of Lemma
+2 ~“ n +2 r +1
3, with m = r +1 and k = 1 = r-n [here the intersections K il Fr I
rr +1 are actually empty] . Let G: lr+^ x r -+ ir+^ be the
and L fl Fr I
homotopy made up of the inverses of the homeomorphisms which form the
homotopy F provided by Lemma 3. The spheroid is obviously
1
 428

2T+ 1
homotopic to (J),and we need only remark that <j>^ o rel G : (I x I,
Fr Ir+1 * ! ) - * (Sn+1,ort^) is a homotopy from tf^ to a spheroid (J>2

such that (f)2 ^ ° rtn+ 2 ^ c ^ x and ^2 ^ ”°rtn+ 2 ^ C 1 *

b) Tosee that su is a monomorphism for r £ 2n+2, we must

show that every spheroid ip : (lr ,Frlr ) -+ (Sn ,ort^) whose suspension
su jp : (Ir+^ fFr Ir+^) ■+ (Sn+\ort^) is homotopic to the constant
spheroid is itself homotopic to the constant spheroid.
Let $: (Ir + 1 x I,Fr Ir+1 x I) (Sn+1fort1) be a homotopy
from su \p to const. Consider a triangulation of Sn+^with the
properties: the equator Sn is a simplicial subspace, ort^ is a
vertex, and ortn+2' ”ortn+2
are interi°r points of (n+1)-simplices.
r +1 r
Further, triangulate rectlinearly I x i = I x I x i so that
I x (1/2) x 0 is a simplicial subspace and $ admits simplicial
approximations. If is such an approximation, then
$ (Ir x (1/2) x 0) cz Sn , and the formula iK (y) = (y,1/2,0) defines
r r n
a spheroid ^ : (I ,Fr I ) (S ,ort^) which is homotopic to ip.
r +2 r +2
Consider the map perm: I -* I , perm (t ^ , ...,t^ +2 ) =
-1
= (t1 , . ..ftr 'tr+2'tr+1 J * Clear ly, K = permiif^ (ort 2>) and
L = perm(ip1 (-ort 9)) satisfy the conditions of Lemma 3, with m = r +2
n r +2 r +2
and k = 1 = r-n+1. Thus, let G: I x I -> I be the homotopy
made up of the inverses of the homeomorphisms which form the homotopy
F provided by this lemma, and let p: Sn+^ \ ^ortn+2 U ^”ortn + 2 ^ ^
be a retraction. One may verify directly that

(y,t) p ( $ 1 opermoG(y,t, 1/2) ,1 )

(where (y,t,1 /2 ) £ lr +^ = ir x i x i) is a homotopy lr x i -> sn

from ^ to const.

The Series { (Sn ,ort )}

5. The main content of Theorem 4 is that each of the seriei

... 7Tr (Sn ,ort1) TTr + 1 (sn+1 ,ort1 ) TTr+2 (Sn+ 2 ,ort1 ) -2H-* ...

of homotopy groups of spheres, connected by the suspension, stabilizes.

That is to say, in the k-th series { tt v (Sn ,ort.)}, the groups
n . n+K 1
ffn+k(S ,ort^) with n 5 k+2 are isomorphic via suspension. This
canonical isomorphism enables us to identify the groups tt , (Sn ,ort ),
n + k 1 r
 429

n 'i k + 2 , with a single group, called the stable group of the series
{iTn+k (Sn ,ort.^ ) } ; we denote it by Stab(k).

2. The Simplest Homotopy Groups of Spheres

1. The groups irr (Sn ) with r < n are trivial. In

particular, Stab(k) = 0 whenever k < 0.

This is a corollary of 2.3.2.2 and 1.3.8.

The Homotopy Groups of the Circle

1 1
2. If r > 1, (S ) is trivial. tt^ (S ,ort^ ) is an
infinite cyclic group with generator sph^.
1
The proof uses the covering (3R,hel,S ) [see 4.1.2.6 ].
First, notice that hel(0) = ort 1 and hel 1 (ort^) = TL . Since the line
H is contractible, its homotopy groups are trivial, and hence, by
1 1
1.8.12 so is 7Tr (S ,ort ^ ) for any r > 1, and A: tt^ (S ^ort^) ■>
7tq (ZZ,0 ) = 2Z is invertible. Moreover, dR,hel,S^) is obviously a
principal bundle with structure group 7L, so that A is a group homo­
morphism (see 1.9.15). Therefore, A is a group isomorphism, and it
remains to observe that A(sph^) = - 1 .

3. The proof above computes the homotopy groups of the circle

as quick as ligthning by applying the general theory to the covering
1
(JR,hel,S ) . Such an approach is natural as the general theory is
already available to us. However, it conceals the fact that the
computation is in fact quite elementary. Since the fundamental group
of the circle has both a historical and intrinsic importance, we shall
not spare a few more lines, and we shall redo the proof of the equality
1
tt^ (S ,ort^) = 7L in an elementary way, which reveals those simple
aspects of the general theory that are really necessary. Namely, we
1
need only the following two facts: every path in S with origin ort^
is covered by a path in JR with origin 0; two paths in JR, with
origin 0, which cover homotopic paths in S^ have the same end.
(Cf. 6.2.1) .
Now to the proof: consider the powers of the fundamental loop
IS: I -* S (with the natural order of multiplication), and let u^
denote the n-th power (n € ffi) . Let u~: I -*3R be the'path with
origin 0 covering the loop u . Obviously, u ^ (1) = n, and so the
loops u^ are pairwise nonhomotopic. Moreover, given any loop
 430

1 ~
u: I -+ S with origin ort , the covering path u : I 1R with origin
0 ends at an integer. Consequently, u~ is homotopic to one of the
paths u~, and thus u is homotopic to one of the loops u^. In other
words, the classes (sph )n with n £ 7L are pairwise distinct and
1
exhaust tt ^ (S ,ort^).

Corollaries

2 1 2 1
4. The pair (D ,S ) is simple. tt (D , S ) is trivial for
2 1 r
any r / 2. ^2 (D ,S ,ort^) is an infinite cyclic group with generator
kug2 .
These are all consequences of Theorem 2 and of the fact that
2 2 1 1
D is contractible, which implies that 0: tt^. (D ,S ,ort^ ) -> TTr _^ (S ,ort^)
is an isomorphism for any r ^ 1 (see 1.6.5).

5. It followsfrom 1 and 2 that for n ^ 1, Sn is a simple

space. Similarly, 1 and 4 show that for n ^ 2 the pair (Dn ,Sn **)
is simple.
In particular, given any point y € Sn with n ^ 1, the
group tt (Sn ,y) can be identified with tt (Sn ,ort ), and given any
n ^ ^ 'j n ^ ^I
point yes with n ^ 2, the group tt (D ,S ,y)can be
n n” 1 ^
identified with ^ ( D 'S ,ort ). Therefore, for r ^ 1, a continuous
17
map f: S -* X, with X a topologicalspace, defines an element of
tt (X,x) for any point x £ f (Sr ) , and not only for x = f (ort ).
r r” 1
Similarly, for r ^ 2,a continuous map f: (D ,S ) -*(X,A) , with
(X,A) a topological pair, defines an element of tt (X,A,x) for any
r— 1 r
x e f(S ), and not only for x = f(ort^).

The Groups tt^ (Sn )

6.
For n i* 1 , su: irn (Sn ,ort.]) irn + 1 (Sn + 1 ,ort1 ) is an
isomorphism and su(sph^) = sph^+^.

PROOF. By Theorem 1.4, su:

(Sn ,ort.) +
tt tt , (Sn + 1 ,ort )
n 1 n +1 ' 1
is an isomorphism for n ^ 2,and an epimorphism for n = 1. To prove
that this epimorphism is alsoa monomorphism, we usethe homotopy
3 2
properties of the Hopf bundle (S ,pr,S ) [see 4. 6 .1.4]: the segment

tt2 (S3 ) = 0 tt2 (S2 ) n^ (S1) -> tt (S3 ) = 0

of its homotopy sequence (see 1) demonstrates that tt (S2 ) = tt (S1 ) .

2 1
 431

Since tt^(S ) - 7L (see 2), su: tt^ ( s \ ort^ ) -* ^ (S2 ,ort^) cannot have
a nontrivial kernel. The equality sufsph^) = sph^+^ is obvious.

7. If n ^ 1, 7Tn (sn /ort^) is an infinite cyclic groupwit

generator sPhn - In particular, Stab(O) = 7L .

For n = 1, this statement is a repetition of a part of

Theorem 2, while for n > 1 it results from 2 and 6 .

8 (COROLLARY). IjE n ^ 2, tt^ (Dn ,Sn \ort^) is an infinite

cyclic group with generator ^u 9 n •

9. For each n > 1, Theorem 7 establishes a canonical

isomorphism 7Tn (Sn ) -* 7L . In particular, it associates with each
continuous map f : sn SR an integer, and it is not hard to see that
this is nothing but the degree deg f , as defined in Subsection 4.6.5.
This is a consequence of three evident facts: deg(suf) = deg f ;
the class k sph^ is represented by the spheroid hel^. (see 4.1.2.6 );
and deg(hel^.) = k.

Similarly, for each n ^ 2, Theorem 8 establishes a canonical

n n_ 1
isomorphism tt (D ,S ) 7L . In particular, it associates with each
n n n“1 n n— 1
continuous map f: (D ,S ) (D ,S ) an integer, which coincides
with deg f , as defined in Subsection 4.6.9.

Further Information Obtained From the Hopf Bundles

3 2
10. If r ^ 3, the homomorphism pr*: tt (S ) tt (S )
3 2 r r
induced by the Hopf map pr: S S is an isomorphism.In particular,
2
tt^ (S ) is canonically isomorphic to 7L , and is generated by
pr*(sph^), i.e., by the class of the Hopf map itself.

This is plain from the segment

^ r (S1 ) = 0 - ^ r (S3 ) -V ^ r ( S 2 ) -> TTr _ 1 ( S 1 ) = 0

3 2
of the homotopv sequence of the Hopf bundle (S ,pr,S ).
7 4
11. If r ^ 1, the homomorphism pr*: tt„ (S ) -* tt_ (S )
7 4 7 r
induced by the Hopf map p r : S -> S mapstt^(S ) isomorphically onto
7
a subgroup of tt (S ) which has a direct complement isomorphic to
----- 3 r 4 3
tt^ ^ (S ). In particular, tt^(S ) = 7L 0 tt^ (S ) .

This is a consequence of Theorem 1.8.10, when applied to the

7 4
Hopf bundle (S ,pr,S ).
 432

15 8
12. If r 5 1, the homomorphism pr*: tt^(S ) -+ tt^IS )
T5 S 15r
induced by the Hopf map pr : S -+ S maps it (S ) isomorphically
8
onto a subgroup of tt^(S ) which has a direct complement isomorphic to
r g
ttr_ 1 (S ). In particular, ti^(S ) ~ 7L © tt^(S ) .

This is a consequence of Theorem 1 .8.10, when applied to the

15 8
Hopf bundle (S ,pr,S ).

13 . The composite maps 7Tr_ -j(s ^) )

SU > ^ r (S 2 ] TTr _ 1 (S1 ) ,

T, . (S3| - S I , „ ( s4 ) TT (s3), TT rs7 ) S U > TT (S 8) T r _ 1 (S7) ,

r- 1 r r- 1 r- 1 r
where the homomorphisms A correspond to the Hopf bundles, coincide
1 3 7
for any r ^ 1 with id 'nr _^ (s ) t id Tir_^ (s ) r and id tt^_^ (S ) ,
respectively.

PROOF. Let q be 2, 4, or 8 , and consider the map

q 2 q- 1
X: D + S given by

2 2 1/2
X (x^,...,x ) = (xlf...,x , ( 1 - x 1 - ... - xc ) ,0 ,...,0 ).
q
,2 q - 1
Its restriction to Sq - 1 is simply the inclusion Sq - 1 while
pr o Xf where pr is the Hopf map p r : S2q 1 is simply
D S : Dq -+ Sq . Therefore, the diagram

^r (S2 <3- 1 .s'3 “ 1 )

(s’ -1)

commutes; here the vertical homomorphisms are induced by

relx : (Dq ,Sq 1) (S2q 1 ,Sq 1). Moreover, rel pr* and the lower 3
are isomorphisms (see 1 .8 . 1 and 1.6.5), and from the above commutativity
it follows that the composite homomorphism

_1 TTr (Dq ,Sq-1) -^e l D S * , v (Sq ) <r e l P r * > ~ 1

1.
3 tt , /(S
„q-1
■ )
r- 1
equals id tt^_^ (Sq ^) To complete the proof, notice that
■1
rel DS* o 9 - su (see 1.2) and 9 ° (rel pr*) - 1 see 1 .8.4) .
14. It follows from 13 and 1.8.7 that su. tt r-(S3)
, -.
v i„
, (S4 )
7 8 5 6
and su: ) -*■ ) are isomorphisms, that is, in the two series
 433

iTrn +2 ^s ^ an<^ + ^as in t*ie s e r ^-e s iTrn (sI1)}] stabilization

begins at least one step earlier than guaranteed by Theorem 1.4.

3. The Composition Product

1. Let X be a space with base point . Given two

spheroids, $ £ Sph^(X,x ) and ip £ Sph^(S^,ort ), the composition
p u q i
<(> o if;: (S^,ort^) ^ (X,Xq ) is a spheroid in Sph^(X,XQ), and the
homotopy class of the latter is uniquely determined by the homotopy
classes of $ and ip. Therefore, for any two classes, a € tt^(X,Xq)
and B € tt^(S^), one may define the composition a o 3 £ tt^(X,Xq).

Equivalently, we can seta o 3 = <|>*(3), where <J> is any representa­

tive of a.
The following facts need no proof: if a € tt^(X,Xq ), then
a o sphp = a; if a £ TT^iX^^), 3 € and y £ tt^ ) , then
a o (3 o y) = (ao 6 ) o y ; if a £ 7T(X,x0 ), 3 £ tt (S^) , and
p u q
f: X Y is continuous, then f* (a o 3 ) = (f*(a)) o 3 ; if a £
and 3 6 then su (a o 8 ) = su a o su 3 ; if a £ TTp(X,XQ) and

G (SP ) , then a o (8 ^ + 8 2) = a 0 6 1 + a 0 B2 [inparticular,

cx o k sphp= k a for all a € 7Tp(X,xQ ) and k £ E ]. The last property
is called right distributivity, to distinguish it from the left
distributivity, which amounts to (a^ + a2) ° 8 = ° 8 + 0^ 0 8 for
any a. 1 ,a 2 £ 1Tp<x »xo^ and $ € irq^sP^* In 9enera1' left distributivity
does not hold usually; see 2 and 5, and also 9.1.

2. Given any ,0^ € TTp(X,XQ> and 8 € ^q-l

(a1 + a2) o su 6 = “ •) 0 su 3 + a 2 o su 8 .

In particular,

(k sphp) o su 8 = k su 8

for anv
------------------------ i -
8 £ 7iq - 1
„ (Sp 1) and integer k.
------------------------------------------- ---------

PROOF. Pick representatives <i>^ ,<i>2 ^ sPhp(X,x^) and

41 £ Sph° _ 1 (SP _ 1 ,ort1) of the classes 0^ ,a2 and 8, respectively,
and let p1 and p2 denote the projections p r : SP 1 x I ->

su (SP _ 1 ,ort1 ) = SP and pr: Sq _ 1 x I -»• su(Sq ^ o r t ^ = Sq . By

definition (see 1 .1 0 .2 ), the class a., + a 2 is represented by the
spheroid
 434

<j> O p (x,2t), if 0 s? t £ 1/2,

P 1 (x , t ) i
o p 2 (x,2 t- 1 ), if 1/2 s t i 1,

while su 3 is represented by the spheroid p 2 <x,t) p^(tjj(x),t)

(see 1.1). This shows that both sides of (1) are represented by the
spheroid

<+>1 0 p 1 (if;(x) ,2 t) , if 0 £ t £ 1 /2 ,
p 2 (x,t) h-
(f>2 O p (lj;(x) ,2t-1 ) , if 1/2 t £ 1.

The Ring Stab

3 (LEMMA). Let k, 1, andn be nonnegative integers,

with n> 0. Then for any 6 tt l1 (Sn ) and 3 € tt n .
a
— n+k +±(Sn ) ,

n n+k , ..kl n 0 n+ 1
su a o su a = (-1 ) su 8 ° su a .

PROOF. Pick representatives <J> £ Sph^+k (Sn ,ort ) and

ip £ Sph^+^ (Sn ,ort^ ) of the classes a and 8, and for nonnegative
integers, p and q, let perm(p,q) be the (auto)homeomorphism of
the sphere S^+C^ = (s\ort^) 0 ... 0 (s\ort^) (see 1 .2 .8 .9) which
permutes the factors according to the rule

( (1, . . . , p+q) w ( p+1, . . . , p+q 1 , . •- , p) . f

One may check directly that the following two compositions are equal:

sun(J) o perm(n,n+k) o sun+^ijj o perm (n+k,n + 1 ) : s^n+k+l

and
perm(n,n) o s u n c() o perm(n,n+l) o s u I1+ '*"i|j : s ^ n + k + l ^ sn ^

Since degperm(p,q) = (-1)pg, this yields

n r, „» n(n+k) , , n+k 0 r / „v (n+k)(n+l)

su a ° [(“ 1 ) sPh 2 n+k] ° su S o [( - 1 ) sPh 2 n +k +l] =
2
= [(-1 )n sph ] o sun 3 o [(-1 )n(n+l)sph_ J o sun+ 1 a.
~ 2n c 2 n+l

Using the right and left distributivities (see 1 and 2),. it is not hard
to reduce this equality to the form
 435

. n vn (n+k) + (n+k) (n+1 ) n n+k 0 , .,n2 +n(n+l) nn n +1

("1) su a o su B = (~1) su B ° su a,

2
and now we note that [n(n+k) + (n+k) (n+ 1 )] - [n +n(n +l)] = kl (mod 2 ) .
oo
4. Vie set Stab = k§^Stab(k) and identify each group
Stab(k) with its image under the natural embedding Stab(k) -* Stab.
The operation o transforms Stab into a ring: if a € and
B € ^H+k+i (sI1) '
then su(a o B) = su a ° su B (see 1 ). Therefore, °
is well defined as a distributive multiplication Stab(k) x Stab(l)
-+ Stab(k+1) (see 1 and 2), and can be extended bidistributively to a
multiplication Stab x Stab -+ Stab. It results from 1, 2, and 3 that
the ring Stab is associative, has the identity sph = sph 1 = sph = ..
kl
and is skew-commutative, meaning that B ° a = (-1 ) a o B for any
a £ Stab(k) and B £ Stab(l).

An Application

5 (LEMMA). The composition (-sph0) o pr+ (sph0), where

3 2 ~
pr: S -+ S is the Hopf map, equals pr*(sph^).

This is a consequence of the commutativity of the diagram

where the horizontal maps are the spheroids (x1 ,x 2 ,x ^ ,x^) »+

(x1 ,-x2 ,x3 ,-x4) and (xi;x 2 ,x3) (x1 ,-x2 ,x3), which represent the
classes sph3 and -sph2 , respectively.

6. The group Stab(1) has at most two elements.

3 2 3
Since (S ) is already stable, and su: ^ ( S ) ^ )
is epimorohic (see 1.4), it suffices to show that Ker su contains the
3 2 2
element 2[pr: S + S ]*(sph3), i.e., twice the generator of tt3 (S )
feee 2.10). Indeed, since

2 pr*(sph3) = pr*(sph3) + pr*(sph3)=sph2 o pr*(sph3) +

+ (-sph2) o pr*(sph3)

(see 5), we get

su (2 pr* (sph3) ) = sph3 o su (pr* (sph3) ) + (-sph^ o (pr*(sph3))
 436

(see 1 ), and we note that the right-hand side of the last equality is
0 (according to 2 ).

4. Information: Homotopy Groups of Spheres

1. For a long time the study and computation of homotopy

groups of spheres was at the center of the attention of topologists.
It was hoped that one could succeed in solving this problem and that
other, more difficult problems in homotopy theory could be reduced to a
considerable extent to it. Deep results have actually been obtained in
both these directions; the initial hopes, however, have not been realize
Gradually it became clear that from the homotopy point of view the spher
is not elementary, but rather an intricate, complicated object. On the
other hand, the informâtiona acquired about the homotopy groups of spher
found unexpected applications, first of all in differential topology.
Below we discuss a (rather small) part of these results:
general results in 2, and those of tabular character in 3. For more
complete information, references, and proofs, see [7].

2. TT4 m-'i (S^m ) , m = 1 ,2 ,..., are the only infinite groups

among TTr (Sn ) with r > n. Each of these infinite groups is isomorphic
to a direct sum 7L 0 (finite group) .
For an odd prime p, the order of the groupStab (2m(p-1 )-1 )
2
with 1 < m £ p- 1 is divisible by p, but not by p , while the order
of Stab(k) with k < 2p(p-1)-2 is not divisible by p if
k t “ 1 mod 2 (p- 1 ) .

3. Among the groups 7Tr (Sn ) which have been computed are
all 7rn+]c(sn) with k £ 22, and all Stab(k) with k £ 37. The group

W (Sn ) with n > 2 and 1 £ k £ 7 are displayed in Table 1.

In Table 2, where pr always denotes one of the Hopf maps

32 7 4 1 5 8
S S , S + S , or S S , we indicate the generators of the
groups Stab(k) with k = 1,...,7.
We add the relations
3 4 5
su (pr*(sph3)) o su (pr*(sph3)) O su (pr*(sph3 )) = 1 2 su (pr*(sph?)),
7 6 9
su (pr*(sph3)) o su (pr*(sph7)) o su (pr*(sph7)) = 120 su(pr* (sph^) ) ,

which give, together with Table 1, a complete description of

7
©k _ 1 Stab (k) , a part of Stab.
 437

Table 1

Groups Generators

Stab(1 ) (SJ) [sa2] su(pr*(sph.J)

2 3
Stab(2) T7g (S4 ) [ = 2 Z2 ] su (pr*(sph3 )) o su (pr*(sph3 ))
Stab(3) V s5) t » a 24] su(pr*(sph7 ) )
Stab (4) ^in(sb ) [=0 ]
Stab(5) •n1 2 (S7) [=0]
4 7
Stab (6 ) tti 4 (S8 ) [= 2Z2 ] su (pr*(sph7 )) o su (pr*(sph7 ))
Stab(7) su(pr*(sph15))
n16 <S ) ^ S 240

Table 2

k Stab(k) k Stab(k)

8 7L2 © S 2 12 0

9 Z 2 ® 7Z>2 © 7Z>2 13 2Z3

10 2Z2 14 s6©s2

11 15 S 480 0 S 2
S 504

Table 3
 438

The groups Stab(k) with k = 8,...,15 are listed in Table 3

5. The Homotopy Groups

of Projective Spaces and Lenses

1. Let 2 £ n £ 00. tt^ (]RPn , (1 :0 :0 :...) ) has twoelements

and is generated bythe class of the loop rp: I ]RPn given by
rp (t) = (cos Tit:sin Tit:0:0 :. ..) . tt^ (HRPn ,(1:0:0:...)) is isomorphic to
7Tr (S ) for all r^ 1 [in particular, this group is trivial for n = 00]
and the isomorphism is induced by the projection Sn -+ ]RPn . lRPn is
simple for n odd, and is not n-simple for n even.

(The casen = 1 has been considered in 2.2.)

PROOF. All assertions concerning the groups

irr (HRPn , (1 :0 :0 :...))follow from Theorem 1.8.12, when applied to the
covering(S , p r , l R P ). Since the fundamental group of ]RPn is Abeliai
IRPn is 1-simple. Nowlet r ^ 2, and consider the automorphism
Trp: n :0:0: ...) ) + tt^.(IRPn ,(1:0:0:...)) . Let rp~ be the path
in Sn which covers rp and has origin ort^. Then from the (obvious)
commutativity of the diagram

(-id S )* /r,n , N
---------- 7Tr (S ,ort ^ )

it follows that T^p is the identity if and only if the composition

(-idSn ).x o t rp~ is the identity. If we make the identification

r (Sn t -ort 1i) = 7Tr (Sn ,ort i) = ttr (Sn ) (see 2.5), then we see that T rp
is theidentity if and only if (-id Sn )* : TT^(Sn ) -+ TT^(Sn ) is the

identity.Finally, note that if n is odd, then -id Sn and id Sn

are homotopic, and hence (-idSn )* : nr (Sn ) frr (Sn ) is the identity
automorphism for all r, while if n is even,

[(-idSn )* : % (Sn ) + V S n >] = "id^n (Sn ) .

2. Let 1 < n < tt^ ((EPn , (1 :0 :0 :. ..) )is isomorphic to 2Z

and is generated by the class of the spheroid in: S 2 =0:P1-+ £EPn .
TTr (CCPn , (1 :0 :0 :. ..)) is isomorphic to TTr (S2n+1) for any r / 2 [in
 439

particular, this group is trivial for n = °°] , and the isomorphism is

induced by the projection S^n+^ -+ (CPn .

(For n = 1 this theorem repeats Theorem 2.10.)

We make two claims: in*: tt ((CP1 , (1 :0 )) тт ((СРП , (1 :0 :0 :. ..

,2 n +1 . rm_n
is an isomorphism; and pr* : t,r
t^ (S ,ог^))
fort тт^ ^
(ССР ,(1:0:0:...)) is
an isomorphism for all r ^ 2. The first follows from the 3-connected-
ТЛ 1
ness of the pair ((CP ,£CP ) (see 2.3.2.2 and 2. 1.3.5), while the second
is a consequence of the homotopy sequence of the bundle (S2 n+\ p r ,(СРП )
and Theorem 2.2.

3. Let 1 £ n £ 00 and r ^ 1 . The homomorphism induced by

the projection S^n+^->• ПНPn maps tt (S^n+^) isomorphically onto a
subgroup of тт(ПНРП
r , (1 ^
:0 :0 :...)) which
has a direct
_ comolement
*
isomorphic to tt ^ (S ) . In particular, tt_(]HP ,(1:0:0:...)) is
ri 3 r
isomorphic to TTr_ ^ s ) for all r ^ 1 .

(For n= 1this theoremrepeats Theorem 2.11.)

For a proof, we need only apply Theorem 1.8.10 to thebundle

, 4n + 3 nx
(S ,pr,HP ) .

4. The lenses L (m ; , . . . , l ^ ) and L( , ¿2 ,.. .) are simple.

tt(L (m;£1 , .. ., t )) and tt (L (m ;£1 ,...)) are isomorphic to 2Z_ . If
i n I I z. 2n 1 ^
r ^ 2, tt (L (m; £ 1 /...,£ )) is isomorphic to tt (S ), and the iso-
r i n 2n - 1 r
morphism is induced by the projection S L (m ;t ^ , .. ., ) .
tt^ (L (m ,¿ 2 , .. .) ) is trivial for all r ^ 2.

The proof is clearly a generalization of the first part of the

proof of Theorem 1.

6. The Homotopy Groups of Classical Groups

1. The inclusion homomorphism Tr^(SO(n)) TTr (SO(n+1)) is

an isomorphism for r £ n-2 and an epimorphism for r = n-1 . Ti^(SO(n))
is isomorphic to ZZ for n = 2 and to TZ>2 for n ^ 3, and is
generated by the class of the inclusion = SO(2) -> SO(n). 7T2 (SO(n))
is trivial for all n. TT3 (SO(n)) is trivial for n £ 2, isomorphic
to ZZ for n = 3, to ZZ 0 ZZ for n = 4, and to the quotient group
of ZZ @ ZZ by a cyclic subgroup for n ^ 5.
PROOF. The triviality of the groups tt2 (S0(2).), tt-^ (SO (2 ) ),
it 2 (SO (3) ) , and tt2 (S0(4)), andthe isomorphisms tt^ SOU)) ~ 7L ,
TT (SO (30 ) £ ZZ2 , TT3 (SO(3)) = ZZ, and tt3 (SO(4)) ~ Z2 0 Z2 all result
 440

from the equalities SO(2) = s \ SO(3) = IRP3 , and SO(4) = ]RP3 x

(see 3.2.1.2, 3.2.3.1, and 3.2.3.3), and Theorems 2.2, 2.7, and 5.1.
The rest is a consequence of the honotopy sequence of the bundle
(SO(n+1),pr ,sn ) with base point id G SO(n+1); see 4. 6 .1.4.

2. The inclusion homomorphism tt (LJ (n)) ffr (U(n+1)) is an

isomorphism for r £ 2n-1 and an epimorphism for r = 2n. If n ^ 1,
TT^(U(n)) is isomorphic to 2Z and is generated by the class of the
inclusion S 1 = U (1) -* U(n). TT2 (U(n)) is trivial for all n. tt^(U(1)
is trivial, while Tr^(U(n)) is Isomorphic to ZZ for all n ^ 2. The
inclusion homomorphism Tr^(U(n)) Tr^(SO(2n)) is epimorphic for all n.

These are corollaries of the equalities [in: U (1)S O (2)] =

1 3
= id and U(2) = S x S , and of the homotopy sequence of the bundle
(U (n+1),pr,S2n+^) with base point id € U(n+1); see 4. 6 .1.4.

3. The inclusion homomorphism (Sp (n )) ^ ^(Spfn+I)) is

an isomorphism for r £ 4n+1 and an epimorphism for r = 4n+2. In
particular, if r £ 5 and n ^ 1, Tr^(Sp(n)) is isomorphic to
irr (Sp(1 )=S3) .

This can be seen from the homotopy sequence of the bundle

(Sp(n+1),pr,S4n+3) with base point id € Sp(n+1); see 4. 6 .1.4.

Stabilization

4. Theorems 1-3 show that for r 5 1 , each series ofgroups

Trr (SO(1)) -+ Tir (SO (2) ) ■+ tt^ (SO (3) ) ...,

Trr (U(1)) + Trr (U(2)) -*■ Trr (U(3)) - ...,

irr (Sp(1)) -> 7Tr (Sp (2) ) + irr (Sp (3) ) -> ...,

stabilizes: the first one, starting with i\^ (SO (r+2)) , the second one,
withTTr (U ([r+2)/2] )) , and the third one, with tt(Sp( [(r+2)/4] ) ) . The
groups 7Tr (SO(n)) with n > r+2, TTr (U(n)) with n 5 [(r+2)/2], and
rrr (Sp (n) ) with n > [(r+2)/4] are said to be stable, and are denoted
by irr (SO), fTr (u), and ir^Sp), respectively. By Theorem 1,
tt1 (SO) s 2Z , tt2 (SO) = 0, and ir3 (SO) s (a © ZZ)/ (cyclic subgroup).
By Theorem 2, (U) = ZZ , tt2 (u) = 0, and tt (U) s a . Finally, by
Theorem 3, (Sp) = 0, Tt2 (Sp) = 0, and tt3 (Sp) = ZZ .

The notations tt^JSO), tt^ (U) , and TTr (Sp) have also a
direct meaning: they represent the ordinary r-th homotopy groups of the
 441

limit spaces SO = limSO(n) , U = limU(n) , and Sp = limSp(n) ,

respectively (see 1 .1 1 .2 ).

Information

5. The homotopy groups tt^SO), irr (U), and ^(Sp) have

been explicitly computed. Namely, for any r i 1 there are canonical
isomorphisms TTr (SO) -* TTr+8 (SO), 7Tr (Sp) -+ TTr +8 (Sp) , and
tvr (U ) -> TTr + 2 (U) , and the first seven homotopy groups of SO and Sp,
together with the first two homotopy groups of U are displayed in the
following tables.

r 1 2 3 4 5 6 7 8 r 1 2

TTr (SO) 0 a 0 0 0 7L 7Tr (U) 7L 0

ffi2 S 2

TTr (Sp) 0 0 2Z 0 7L 0
*2 *2

For a proof, see [17].

There are also many unstable homotopy groups of the manifolds
SO(n), U(n), and Sp(n) which have been computed. For example,

TT2n^U ^n ^ " ffin! ' TT4n +2^S p ^n ^ “ S (2n+1)! for n even' and
71 a l0 (Sp(n) ) = Z20 r / 0 for n odd. For details and references
4n+ z Z [ [zn +1 ) !J
for the proofs, see [7].

7. The Homotopy Groups

of Stiefel Manifolds and Spaces

1 (LEMMA). Let k < n. Then the following are true: the

manifold V(n,k) is simple; the inclusion homomorphism 7r^(V(n,k)) -*■
+ (V (n+1 ,k+1 ) ) is an isomorphism forr < n-1 and an epimorphism for
r = n- 1 ; and if n is odd and k = 1 , the last epimorphism is also an
isomorphism.
The manifolds (CV (n ,k ) and 3HV (n ,k ) are all .simple. The
inclusion homomorphism n (<CV (n,k) ) + tt^ (CCV (n+1 ,k+1 )) is an isomorphism
for r < 2n and an epimorphism for r = 2n. The inclusion homomorphism
tt (3HV(n,k)) + tt^ CHV (n+1 ,k+1 ) ) is an isomorphism for r < 4n +2 and an
 442

epimorphism for r = 4n+2.

The fact that the Stiefel manifolds are simple may be seen
from the equalities V(n,k) = SO (n)/SO (n-k) , £CV(n,k) = U(n)/U(n-k),
and IIV(n,k) = Sp(n)/Sp(n-k) (see 4.2.3.12 and 1.9.7). To prove the
rest, use the homotopy sequences of the bundles (V(n+ 1 ,k+ 1 ),p r ,S ),
((CV(n+1,k+1) ,pr,S2n+1) , and (HV(n +1 ,k +1) ,p r ,S4n +3 ) described in
4.6 .1.4, taking the inclusions TR^ -+ lRn , -► (Cn , and lHn
as base points (in the respective total spaces). In the real case, we
take advantage, in addition, of the fact that for n odd, the bundle
(V(n+1,2),p r ,Sn ) admits a section (see 3.1.4.7); for n odd and
k = 1, this ensures that the first of the aforementioned homotopy
sequences splits from the left at the terms tt (V (n+1 ,k+1 ) ) (see 1.8.8)

2. If k < n, then the manifold V (n ,k ) is (n-k-1)-connected.

tt (V(n,k)) with 0 < k < n is cyclic and is generated by the class
^ n—k
of the inclusion S = V(n-k+1,1) -+ V(n,k); this group isinfinite
whenever n-k is even or k = 1 .

This is a corollary of Lemma 1: when r < n-k,

TTr (V(n,k)) s tt (V(n-1,k-1)) s ... £ tt (V(n-k+1,1)) = TTr (Sn_k) = 0,
n—k
while in the sequence ) = irn_k (V (n-k + 1 ,1 ) ) -»• t ^ (V (n-k + 2 ,2 )) ->

-s- ... ->■ 7Tn K, (V(n,k)) all the maps are isomorphisms,
~
except for the firs-
which is an isomorphism for n-k even and an epimorphism for n-k odd.

3. The manifold (CV (n ,k ) is 2 (n-k)-connected .

TT2 n- 2 k +1 is isomorphic to 7L and is generated by the class
of the inclusion S2n ^k+1 = £v(n-k+1,1) -+ (CV(n,k).

This is a corollary of Lemma 1 : when r £ 2n-2k+1 , in the

sequence tt^ (S 2n ^k+1) = tt^ (EV (n-k+1 ,1) ) -+ tt^ ((CV (n-k+2 ,2)) ->■... -*

■+ tt^ (CCV (n,k) ) all the arrows are isomorphisms.

4. The manifold lV(n,k) is (4n-4k +2)-connected.

tt-n—4k+3 (n 'k) ) is isomorphic to ZZ and is generated by the class
4n —4k + 3
of the inclusion S = HHV (n-k+1 ,1) HV(n,k) .

This is also a corollary of Lemma 1 : if r £ 4n-4k +3

4 r, — 4 V + 7
(actually, if r £ 4n-4k+5), then in the sequence tt^ (S ) =

= TTr №V(n-k+1 ,1 )) -+ TTr (lV(n-k+2,2) ) + ... TTr (H V (n ,k )) all the

arrows are isomorphisms.

5. The spaces V(°°,k) and <CV(°°,k) (see 4.5.3.9), as well

as ]HV (00,k ) = lim (3HV (n,k) ,in : ]HV(n,k) ■+ lHV(n+1,k)) are “-connected.

This follows from Theorems 2, 3, 4, and 1.11.2.

 443

8. The Homotopy Groups

of Grassman Manifolds and Soaces

1. In this subsection the computation of the most important

homotopy groups of the Grassman manifolds G(n,k), G+ (n,k), (CG(n,k),
and EG(n,k) , and of the Grassman spaces G(°°,k), G (°°,k) , (EG(°°,k)
(see 4.5.3.2) and BG(«>,k) = lim (IHG (n,k) ,in : 3HG(n,k) -> IHG(n+1,k))
is reduced to the computation of the homotopy groups of the corresponding
classical groups.
Grassman manifolds and spaces are taken care of together, and
thus n may also take the value «>.

2. If k > 0 and 0 < r < n-k, then tt (G+ (n,k)) is

isomornhic to _^(SO(k)), and the inclusion homomorphism
tt (G (n ,k) ) TTr <G + (n1 ,k)) is an isomorphism for all n* > n.

The first claim results from Theorems 7.2 and 7.5, and the
homotopy sequence of the bundle (V(n,k),pr,G (n,k)), defined in
k n
4.6 .1.4, with the inclusion IR -+ 3R as base point The second claim
results from the commutativity of the diagram

A
Tir (G+ (n,k) (so(k))

in. in* = id

77 r < G + ( n l ' k ) ) ^r_1 (so(k))

(see 1 .8 .6 ]

tt (G(n,k)), with 0 < k < n and r ^ 2, is isomorphic

to tt (G+ (n,k)). tt^ (G(n,k) ) is isomorphic to 7L for n - 2, 1,
and to ZS, for 0 < k < n, n > 3.
,1
Since G (2,1) is homeomorphic to S', Theorem 2.2 yields
^ (G (2 ,1 ) = Z5 . If we now apply Theorem 1.8.12 to the canonical two-
tt

-sheeted covering (G_£n,k ) ,pr ,G (n,k )) , the rest is plain.

4. If 0 < r < 2n-2k+1, then tt^ ((EG (n ,k)) is isomorphic

to TTr ^(U (k )), and the inclusion homomorphism tt^.((EG (n ,k) ) -*•TTr ({CG(nf,k) )
is an isomorphism for all n 1 > n.

The proof repeats that of Theorem 2, with obvious changes.

5. I_f 0 < r < 4n-4k+3, then tt^ (3HG (n,k )) is isomorphic

to ttr ^(Sp (k) ), and the inclusion homomorphism tt^_(3HG (n,k) ) -*
is an isomorphism for all n* > n.
 444

Again, the proof repeats that of Theorem 2, with obvious

changes.

9. Exercises

1. Let q = 2,4,8, and let pr: S 2 q ~ 1 -*Sq be the Hopf map.

Show that for any integer k
2
(k sphg ) o pr*(sph2g_1) = k pr* (sph2g-1) .

2. Show that for any positive integer n, lRPn is

(n+1 )-simple.
3. Let n be even and k be odd. Show that G(n,k) is
simple.

4. Let 3 £ n £ 00.Show that G(n,2) is not 2-simple.

5. Show that the inclusion homomorphisms

7rr (SO(3)) -+ 7Tr (SO (4 )) , Tir (S0(7)) -Trr (S0(8)),

7Tr (U(1)) + TTr (U(2)), 7Tr (U (3) ) ->■ TTr (U(4))f

and
TTr (Sp(1 )) + 77^ (Sp(2) )

are monomorphic for any integer r.

6. Consider the map (EV(n,k) -+ V(2n,2k-1) which takes each

k-frame (v1,...,v ) of En into the frame (v1fiv ,...,v ,iv, ,v )
n 2n “1 .K l X.
of CE , considered as IR . Show that the homomorphism
TT2n-2k+1 (n,k)) ■> * k+1 (V(2n,2k-1)) induced by this map takes the
generator indicated in 7.3 of Tr2n-2k+1 • into the generator
indicated in 7.2 of Tr2n-2k + 1 (V (2n ,2k-1 )) .
 445

§3. HOMOTOPY GROUPS OF CELLULAR SPACES

1. The Homotopy Groups

of One-dimensional Cellular Spaces

1. In this subsection we compute the homotopy groups of a

bouquet B = ,ort^) constructed from an arbitrary family,

{S^=S of circles. As usual, the base point bp will be the

center of the bouquet.
To simplify the exposition, we let u and a denote the
loop defined by the inclusion imm^: S B, i.e., the loop
imm^ o IS: I + B, and the homotopy class of u , i.e., imm *(sph^),
respectively. A loop will be referred to as standard if it is of the
form (...((v.v0 )v0) ...v .)v , where each of the factors v.,...,v
I Z o n-In _^ I n
is either one of the loops u or one of their inverses u and, in
1 y y
addition, two loops u , u^ with the same \i are not allowed to be
adjacent. The case n = 0 is not excluded: then, the product is simply
the constant loop with origin bp.

2 (LEMMA). There is a covering (B~,p,B) with the following

two properties:
(i) B~ is contractible;

(ii) the paths which cover standard loops and originate at

-1
some point x^ of the fiber Fq = p (bp) end at distinct points of
Fq , and Fq is exhausted by the ends of these paths.

PROOF. Let us agree to denote by GF(M), as usual, the free

group generated by the set M. We equip GF(M) with the discrete
topology, form the bouquet A = V (D^ =E)1 ,0) , and then the product
A xGF(M). Further, let p be the partition of A x GF(M) into the
pairs { (imm (1),g),(imm (-1),gy)} with y G M and g G GF(M), and
the points which do not appear in any of these pairs, and denote by p
the composition
pr V (DS = DS)
A x GF(M) --- -— ► A U u------ ► B.

Then p is obviously constant on the elements of p. Now set

B~ = [ Ax GF(M)]/p, p= [factp : B~ + B] , xQ = pr(aQ ,e),

where a^ is the center of the bouquet A, e = eQp (m )' anc^

 446

pr = [pr: A x GF(M) -► B~] . Then it is readily seen that (B~,p,B) is

a covering with F q = pr(a^ x GF(M)) and € Fq .

The contractibility of B~ follows from Lemma 2.3.3.4:

in fact, the subspaces pr(A x [GF (M) \ GF^_^ (M)]) of B , where
GF (Ml is the part of GF(M) consisting of words of length £ n,
n
satisfy the conditions of this lemma. The path with origin X q which
G1 e 2 En
covers the standard path (...(u u )...)u [e ,...,e = ± 1 ], ends
y2 yn
G 1 G2 en
at the point pr(a^,g), with g = y^ y 2 ***^n * Clearly, the ends of
of these paths are pairwise distinct and exhaust Fq.

3. The groups (B ) with r > 1 are trivial, whereas

tt^ (B ,bp) is a free group with free generators a .

The proof is based on Lemma 2 and uses the same notation.

Since B~ is contractible, all its homotopy groups are trivial, and
hence so are the groups tt (B) with r > 1; moreover, the map
A: tt ^ (B,bp) tI q (Fq , x q ) is invertible (see 1.8.12). Combining the
invertibility of A with property (ii) of the covering (B ,p,B), we
see that the homotopy classes of the standard loops are pairwise
distinct and exhaust TT^(B,bp). Consequently, Tr^(B,bp) is a free
group with generators a , y £ M.

4. The fundamental group of a connected one-dimensional

cellular space is free, whereas its higher homotopy groups are trivial.

This is a corollary of Theorem 3, because every connected

one-dimensional cellular space is homotopy equivalent to a bouquet of
circles (see 2 .3.3.6 ).

2. The Effect of Attaching Balls

1. Let X = A [J_____ (Dy= D^+ ^ )] / where A is a connected

topological space, and cp is a continuous mao 1 I ^W (S = S k ) -> A
-*— Ly£M y
(see 2.3.2.1), and let X q € A. In this subsection we exhibit a system
of generators for the group tt (X,A,xn) [ k ^ 1 ] .
K+1 u
We remark that the homotopy groups irr (X,A) with r $k are
trivial (see 2.3.2.1) , whereas for r > k+1, tt^_(X,A) is already a much
more complicated object: in the simplest case, when A is just a point
and the family {D } consists of a single ball, tt (X,A) equals
irr (Sk + 1 ) .
In Theorem 2 below, f denotes the composite map
 447

t_,1 m imm
D ------- | | D U— X,
J— -V V '

and € 'fTk + 1 (X, A, f (ort1 )) is the class of the spheroid

’ (Dk + 1 ,Sk ,ort1) -+(X,A,f (ort1 )) .

fy •

2. Let : I ->• A be an arbitrary path joining thepoints

an(3 Xq.I_f k ^ 1, then nk (X,A,xQ) is generated over
v1 by the classes = Tw [i.e., it is generated, in the
usual sense, by the classes with co € ti^(A,Xq )].

We claim that every element S € tt, (X,A,x ) can be

K+ I u
represented as
in ±1
3 - J~j [T^ _ (3y )] (yi € M, 0). € ^ (A ,Xq )) . (1)

By Lemma 2.3.2.1, there is a spheroid g € Sph, (X,A,xn ) in

k +1
the class y, and similarity transformations , ...,a mapping D
k ■+■1
onto pairwise disjoint balls d ,...,d <= IntD , with the following
properties: the point of d^ having the largest value of the first
coordinate coincides with a.(ort1); the segment joining this point
k+ ^1 I JQ

with ort„ lies in C = D \ u. . Int d. ; the composition

i l= 1 i

Dk + 1 » d. Dk + 1 J2_ X
1

is identical with one of the maps f ; and g (C) cz a. N o w it is clear

^i
that, if we suitably reindex the balls d^ ,. ..,d^, then X, A, x ^ , g,
and d ^ ,..., d^ satisfy the conditions of Theorem 1.11.1, and hence we

have, in the notation of this theorem, y = Tg (y^) . In our case,

±1 ^
y = B and y^ = ? the last equality is a consequence of the fact

that a and y. are the elements of tt, .(X,A,f (ort ) )

U. 1 K+ I ]J. I
1
represented by the spheroids f and g o t ., which are transformed

one into another by the orthogonal transformations of D , a. oabT.,

-1
(ab t .) ° a^ (which are inverses of one another).Consequently,

B = n ™ . Tg (a^) , and to obtain (1), we need only write ok for the

Si -1
class of the loop w s..
^i 1
3 (COROLLARY). Under the hypotheses of Theorem 2, the
inclusion homomorphism tt^ (A ,x q ) -+ tt^(X,Xq ) is an isomorphism for
r £ k-1, and an epimorphism for r = k. The kernel of this epimorphism
 448

is generated over tt^A/Xq) by the classes 3|3^ = Tw (3 ctp) [i.e., by

the classes of the attaching spheroids ^f , translated to xq 1 •

4. Let (X,A) be a cellular pair with base point xQ . I_f A ¿s

connected and A => ske,_X, with k 5 1, then tt_(X,A) is trivial for
- jC r
all r £ k. Moreover, TTk + 1 (X,A,xQ) is generated over tt^ (A,xq ) by
the classes of the characteristic maps of the (k+1)-cells in X ^ A
(regarded as spheroids), translated to xQ along arbitrary paths. The
inclusion homomorphism
----------------- 1------
tt (a,x„)
r 0
tt
r (x,xn)
0
is an isomorphism
------------ for
*--------
r £ k- 1 and an epimorphism for r = k ; the kernel of the latter is
generated over tt ^ (A,Xg) by the classes of the attaching spheroids of
the (k+1)-cells in X ^ A , translated to xQ along arbitrary paths.

When
X ^ A c ske, all these assertions follow from 1, 2,
k+ 1
and 3. The general case is reduced to this special situation by
Theorem 2.3 .2 .4 .

3. The Fundamental Group of a Cellular Space

1. In this subsection we pressenti an effective method for

computing the fundamental group of a cellular space possessing a single
0-cell. This last condition is not a serious limitation, since, firstly,
it is fulfilled in the most important cases and, secondly, every connected
space can be transformed, by taking a rather simple quotient, into
a homotopy equivalent space which meets our requirement (see Subsection
2.3.3). It is by no means difficult to generalize the computation scheme
to arbitrary cellular spaces; however, the exposition is cumbersome.

2. Let X be a cellular space with a single 0-cell xQ .

Since Xq isalso the unique 0-cell of ske^X, this skeleton is
homeomorphic to a bouquet of circles. Consequently, tt ^ (ske ^X ,xQ) is
the free group generated by the homotopy classes of the characteristic
loops, i.e., of the characteristic maps of the 1 -cells (see 1 .3 ).
According to 2.4, in*: 7^ (ske1 X ,xQ ) tt (X,xQ ) is an
epimorphism whose kernel is generated over t^ (ske X,xq) by the homotopy
classes of the attaching maps of the 2-cells of X, translated to xQ
along arbitrary paths. In our case, tt 1 (ske 1X ,x ) acts as a group of
inner automorphisms, and hence Kerin* is the smallest normal subgroup
of tt1 (X,xQ ) containing the above elements. Thus, the fundamental group
that we want to compute is canonically isomorphic to the quotient group
of tt^(s k e X ,X q ) by this normal subgroup.
 449

3. The discussion above shows that in order to compute

tt 1 ) i1: suffices to know the 1-skeleton of X and the attaching
( X ' x q

maps of the 2-cells of X. Given these data, we can exhibit a system

of generators and relations for tt ^ ( X , x q ): to each 1-cell corresponds
a generator, namely the class of the respective characteristic loop;
each 2 -cells defines a relation, namely that the class of the attaching
map of the given 2 -cell, when translated to Xq and expressed in terms
of generators, must be equal to the identity element of tt^(X,Xq). In
a very simplified fashion, we may say that a set of generators of
tt^(X,Xq) consists of the 1-cells of X, while a system of relations
consists of the 2 -cells.
We remark that the system of relations is not entirely
canonical, because it depends upon the choice of the paths along which
we do the translation; consequently, the left-hand sides of the relations
are determined only up to conjugation.

4 (COROLLARY). The fundamental group of a finite connected

cellular space has a presentation given by a finite number of generators
and relations.

An Additional Theorem

5. If A and B are subspaces of the topological space

with inclusions

and xQ G A H B, then the rule a * 3 -> i^(a)j^(6 ) defines a homo­

morphism TT (A,xq ) * tt ^ (B ,xQ ) + TT1 (X,xQ ) [ * denotes the free product],
whose kernel contains all the elements of the form i*(<5) * )
with 6 € tt (A il B,xQ ) . Therefore, the same rule defines a homomorphism

[ tt (A,xQ ) * tt1 (B,x0 )]/vk(X,A,B,xQ ) + tt1 ( X , x q ), (2)

where vk(X,A,B,xQ) designates the smallest normal subgroup of

tt1 (A,xq) * tt ^ (B, X q ) containing the indicated elements (and is known as
the van Kampen subgroup) . Furthermore, homomorphism (2) is natural,
meaning that the diagram
 450

[ Ti 1 (
A ,x Q ) * TT ( B , X Q) ] /vk(X,A,B,x0) ----- ► tt n (X,xQ )

[tt (A',x£) * ir1 (B',x£) ]/vkOi,A',ff,x0) ----- ► tt1 (^^¿)

produced by a continuous map (X,A,B,Xq) -»■ (X1 ,A ',B ',Xq) always

commutes.
6. Let (X,A,B) be a cellular triad (i.e., a cellular sp
X with two subspaces, A and B, such that A U B = X) , and let
Xq £ D = A H B. If_ A, B, and D are connected, then (2) is an
isomorphism.

(This theorem will be generalized in the next section; see

4.3.12.)
PROOF. Let us assume first that X has a single 0-cell x^.
By 3, the fundamental group at x^ of any ofthe spaces A, B, C, or
D admits a presentation by generators and relations corresponding to
its 1-cells and respectively its 2-cells. Therefore, the system of
generators and relations of tt^(X,Xq ) (tt^(D,Xq )) is the union
(respectively, intersection) of the systems of generators and relations
of tt^(A,Xq ) and tt^(B,Xq ), and the homomorphisms and j*
from 5 act as the identity on generators. Using the systems of
generators and relations of tt^(A,Xq ) and tt^(B,Xq ) we may build a
system of generators and relations of the group tt^(A,Xq ) * tt^(B,x 0);
however, the generators corresponding to the 1-cells in D must be
counted twice. With this choice of generators and relations, the
homomorphism (A,xq) * n (B,xQ) *+ i\ (X,xQ) , a * S ^ i* (a) (S) , is
the identity on generators, and its kernel is generated by the elements
obtained by identifying the generators corresponding to the 1 -cells in
D. This completes the proof of the case that we considered.
To reduce the general case to this special one, we shall
transfer the origin from x^ to an arbitrary 0-cell e^ in D (by a
translation inside D), and then replace the quadruplet (X,A,B,eQ ) by
a homotopy equivalent quadruplet (X1 ,A 1 ,B ',e^) with a single 0 -cell,
e^ , and such that A' U B 1 = X'. We exhibit such a quadruplet by
taking quotients twice: first, the quotient of X by a contractible
one-dimensional subspace of D containing skeQD (see 2.3.3.5 and
1.3.3.7, and cf. 2 .3.3.6 ), and subsequently the quotient of the resulting
space by the union (bouquet) of contractible one-dimensjonal subspaces
of the quotients of A and B containing the 0-skeletons of these
quotients.
 451

7 (COROLLARY) . I_f A and B are cellular spaces with 0-cells

a and b as base points, then tt^ ((A,a) (B,b),bp) ~ tt^ (A,a) * tt (B,b) .

4. Homotopy Groups of Compact Surfaces

1. Recall that a sphere with handles and crosscaps and at

one hole is homotopy equivalent to a bouquet of circles, the number of
circles being 2 g+l - 1 when g handles and 1 holes are present, and
h + 1-1 when h crosscaps and 1 holes are present (see 3.5.3.9). Thus,
the fundamental group of such a surface is free, having 2 g+l - 1 or
h+ 1 - 1 generators, respectively, whereas the higher homotopy groups are
trivial; see 1.3.
Below we shall discuss the homotopy groups of closed surfaces,
i.e., of spheres with handles or crosscaps, but no holes. First (using
the cellular decompositions indicated in Subsection 3.5.3, and Theorem
3.3) we compute the fundamental groups, and then (by means of a simple
device) we handle the higher homotopy groups. We disregard the sphere
and the projective space, whose homotopy groups have been computed in the
previous section (see 2.2.7, 2.2.10, 2.4.3, and 2.5.1), and need no
further comment.

The Fundamental Groups of Closed Surfaces

2. The cellular decomposition of a sphere with g handles,

constructed in 3.5.3.8, contains one 0-cell eQ , 2g 1-cells a^b^,...
a ,b , and one 2-cell, whose attaching map takes ort into en . Let
g g I
a ,6 w...,a , 8 denote the generators of the fundamental group of the
1 \ 9 9
1 -skeleton of the given surface which correspond to the 1 -cells a^,b^,
...,a ,b . The the homotopy class of the above attaching map (regarded
9 9
as a loop) is the word

a ^ a ^ e ^ . - . a g B g a g ^ ; 1.

Therefore, the fundamental group of our surface at the point eQ may

be described as the group with generators a 1 ,b1 ,...,a ,bg, and the
relation

S i V i V - ' - V V g V - 1-
 452

The cellular decomposition of a sphere with h crosscaps

(see 3 .5 .3 .8 ) contains one 0 -cell eq, h 1 -cells c^,...,c^, and one
2-cell. Repeating the previous argument with the obvious changes, we
see that the fundamental group of this surface at e^ can be described
as the group with generators c^,...,c^ and the relation

3. It is important to note that the groups computed in 2 are

pairwise nonisomorphic. (To see this, factor each fundamental group by
its commutator subgroup: for a sphere with g handles this yields a
free Abelian group of rank 2g, while in the case of a sphere with h
crosscaps the result is the direct sum of a free Abelian group of rank
h- 1 and a group of order 2 .) In particular, the closed model surfaces
are pairwise nonhomeomorphic.
From this it is readily seen that the compact model surfaces
are also pairwise nonhomeomorphic: it suffices to seal up the holes by
discs. The number of holes is a topological invariant, because it equals
the number of components of the boundary (see 3.1.1.4 and 4.6.5.12).

The Higher Homotopy Groups

4. Let P be a sphere with g handles. If g ^ 1 and

r ^ 2, then ttr (P) = 0.

PROOF. It is clear that P admits as a covering space the

infinitegarland P~ constructed from S 1 * HR by first removing small
open discs centered at the points (ort1 ,2 k) [k = 0 ,±1 ,...] and then
glueing a sphere with g- 1 handles and one hole in the place of each
such disc (see Fig. 15). Denote by P~ the finite garland constructed
in thesame fashion from the product S 1 x [-2n-1,2n+1] . Obviously,
P~ is a sphere with (2n+1) (g-1) handles and two holes, so that
= 0 for all r > 2 (see 1). Since P~ = lim(P~,in: P~ P~ „ ) ,
r n ^ n' n n +1 '
we also have tt^ (P ) for all r *2 (see 1.11.2). Consequently,
tt (P) * 0 for all r > 2 (see 1 .8 .1 2 ).

5. Let P be a sphere with h crosscaps. If h ^ 2 and

r ^ 2, then tt^ (P) = 0 .

Since P admits a sphere with h-1 as covering space (see

4.1.2.6 ), this is a corollary of 4 .
 453

Fig. 15 (g = 3)

5. The Homotopy Groups of Bouquets

1. Suppose that we are given a family { (X ,x ) } of

J y y y£M
pointed T^-spaces, and consider the bouquet B = ^y£j^X y,Xy^ ' -^e
formula Imm({a } c ..) = 7 c .. imm * (a ) defines a homomorphism
y y€M Ly€M y* y

Imm: © ... tt (X ,x ] TTr (B,bp)

y£M r y' y

for any r 5 2. This homomorphism is natural, i.e., if

B ’ = V , £j^i(^yi,Xyi) is another bouquet of T^-spaces, a: M 1 ■+ M is
arbitrary, and f ,: (X^,,x^,) <xa (y <)'xa (y <)> are continuous, then
the following diagram commutes

Imm ttr (B ’,bp)

© , ,tt (X ’ ,,x ' ,
y ' €M' r y y

(3:

Imm
) r ., TT (X ,x ) TTr (B,bp)
yCM r y' y

where the left vertical map is the homomorphism

^“y ’*y ’€ M ’ (I^y Ga

"1/
(y)
CM,
\ <f y , ) ** ( a y1 , ) 1 y€M

and the right vertical map is the homomorphism induced by the map
f y , (Yy*,) [y *y,€ X ’y,, y ' € M ’] .
a (y 1 )
 454

2 (LEMMA). Given any a E тгг (В,Ьр) [r £ 1] , there is а

finite set M' с: M such that: pr^*(a) = 0 ±f_ у E M \ M' ; a lies in
the image of the homomorphism tt (B',bp) -*• ттг (В,Ьр) induced by the
natural embedding of the bouquet В* = V ,(X ,x ) in B.
■- ------------------------------------- ^-------------------------------- ■*-------------------- уt r1 у У "

We only have toobserve that for any spheroid ф E Sph^(B,bp),

ф (Ir ) is covered by a finite number of sets
imm (X ).[Indeed, since
г У "M
ф (I ) is compact and every point of X is closed in X , we can
У r
choose a point in each nonempty intersection ф (I ) Л imm (X^ ^ x^) and
in this way produce a set which, being both discrete and compact, is
finite.]

3. Let r ^ 2 and define the mapP r : 7T^(B,bp) -* ©^ (X^ ,x^)

by Pr (ot) = {pr a.( a ) } _ w (2 shows that this definition is correct).

2 У* уEM
Obviously, Pr is a homomorphism and diagram (3) remains commutative

when we replace —?mm— > by < Pr . If M is finite, then Pr equals

the composition of the homomorphism ттг (В,Ьр) -► тт^.( r{ induced

by the inclusion В -> x x (see 1.2.8.3) with the canonical

yEM у

isomorphism tt (x x ,{x }) © ^хдтт (X ,x ) (see 1.1.9).

^ г yEM у' у yEM г у ' у

4. Pr ° Imm equals the identity automorphism of the group

^ y E M ^ r 'Xy)• In particular, Pr is epimorphic, Imm is monomorphic,
and 7Tr (B,bp) = Ker Pr © Im Imm.

PROOF. Since pr ° imm = id X and pr ° imm (X ) = x

у у у ^ V У у v
for V ф- у,

Pr о imm({a } __.) = Рг[У imm .( а )] =

у yEM LуЕМ у* у

= ilLyEM
£гм(Рг
“ V
, ° у
)** Ь )
у
} vEM = {а v }vEM

for any a E тг (X ,x ) and yEM.

2 у г у у
5. Let (X ,x ) be cellular pairs, and let m, к (у E M)
м и у
be positive integers such that к у +k V > m for v / y, and for any
■
у

тт (X ,x ) = 0 for 1 £ s £ к .
s у у у

If 2 si r «: m, then

Imm: ® ^.тт (X ,x ) -+ tt (B,bp)

уCM г у у г ^
and
Pr: tt (B,bp) -> © ¿-«»тт (X ,x )
г г y£M Г у у
 455

are isomorphisms.
(This theorem will be generalized in the next section; see
4.3.1 .)
PROOF. Suppose first that M is finite. By Theorem 2.3.3.2
and the fact that Imm is natural (see 1), we may assume that
ske, X
ky u
reduces to the point x for all p € M. In this case, B => ske „X,
y m+ 1
where X is the cellular product of the spaces X^. By Theorem
2.3.2.2, in*: 7T^(B,bp) TT^(X,bp) is an isomorphism for r £ m.
According to 1.1.9, in* ° Imm: ©y€MTTr (X ,x ) + 7ir (B,bp) is an
isomorphism for any r. Therefore, Imm is an isomorphism for r £ m.
In the general case, 2 shows that for any a c TTr (B,bp) and
r £ m, there is a finite subbouquet B* = V M ,(Xy ,xy) of B such that
a lies in the image of the homomorphism 7T^(B',bp)7T^(B,bp) induced
by the natural embedding B 1 -+ B. Using 1, this homomorphism is part of
the commutative diagram

© £.M 1 TT (X ,x ) I m P1" > TT ( B 1 ,bp)

y£M r y' y r ^

© ..IT (X ,X ) >TY(B,bp) .
y€M r y y r c

But we have already proved that the upper Imm is an isomorphism, so

that a also lies in the image of our (lower) Imm. Therefore, the
latter is an epimorphism and this, combined with Theorem 4, implies that
Imm and Pr are isomorphisms.

6 (COROLLARY). Let B be a bouquet of n-dimensional spheres,

constructed from a family = Sn ,orti ) H n ^ 2 , then the
groups ^(B) with r < n are trivial, whereas 71^(3,bp) is a free
Abelian group with free generators imm^*(sphn ).

6. The Homotopy Groups

of a k-Connected Cellular Pair

1 (ALGEBRAIC LEMMA). Consider the following commutative

diagram of groups and homomorphisms

□ — 1— >□
aI I6
n —s
 456

If a, 6 , and y are epimorphic and Ker 6 ^ a(Ker y) / then 6 is

epimorphic, Ker 6 = a(Ker y), and Ker 6 = Y(Ker a) .

PROOF. Since the diagram is commutative and a, 6 are epi­

morphic, 6 is epimorphic. The commutativity of the diagram implies
also that alKery) c Ker 6 and Y(Kera) c: Ker 6. Let us verify that
Ker 6 c y(Ker a) • Pick d £ Ker 6 . If d = y(a) , then a(a) £ Ker 6
(again by commutativity), whence a(a) € a(Ker y ) / i.e., there is
- 1
c £ Ker y such that a(c) =a(a). Thelast equality yields ac CKera,
and we have d = y (a) = Y(ac ^) £ y (Ker a) .

2. Let (X,A) be a cellular pair with base point xQ £ A.

If A is connected and, for r £ k, tt (X,A) = 0, then
pr*: TTk + 1 (X,A,xQ) •+(X/A,pr (xQ)) is epimorphic, and for k ^ 1
Ker pr* is the smallest subgroup of ^ (X,A,x^ ) containing all the
"ratios" (T^a)a ^ with a £ ^ + 1 (X,A,Xq) and o £ (A,x^) . For
k = 0, the situation is described by the commutativediagram

tt1 (A, x q ) ln* > tt 1 ( X ,xQ) > tt^ X jA ^ q )

Pr<

•n1 (X/A, pr (Xq ) ) ,

where abs pr* and rel* are also epimorphic, and Ker(abspr*) is
the smallest normal subgroup of Tr^(X,xQ) which contains Ker rel* =
= Im in* .
(This theorem will be generalized in the next section; see
4.3.14.)
PROOF OF THE CASE k ^ 1. Suppose first that X \ A consists
only of (k+1)-cells and, as a consequence, X/A is a bouquet of (k+1)-
-dimensional spheres. For each cell e £ X \ A, consider the homotopy
class of its characteristic map (viewed as a spheroid of the pair (X,A)),
and translate it to xn , denoting the resulting element of ti, (X,A,x )
0 ^ r k+1 0
by a^. Set Be = pr*(ae ).By Theorem 2.4, the classes T0ae with
a £ t^(A,x0 ) form a system of generators of Trk+^ (X,A,xQ ) , and by
Theorem 5.6, the classes B0 form a system of independent generators
of the Abelian group 7Tk+ ^ (X/A,pr (x ) ) . Moreover, it is obvious that
Pr* (Tga ) = Pr*(a) for all a € Trk+ 1 (X,A, x Q) and o e u (A,x ) , and
these facts will suffice to complete the proof of the theorem for
k > 1. Since Se = pr*(ote )generate Trk + 1 (X/A,pr (x )) , pr* is epi­
morphic. Further, since
pr*(T o cx) _
= pr.
- * (a), we have (Tact)cx- 1 G Ker L
pr,.
*
Let us show that the ratios (Toa)a generate Ker pr* . If k > 1
 457

X ( e , a)
and the class £ T1 l 1, (e ,a) (T a
a e
) [with only a finite number of
nonzero integers A( e, a ) j belongs to Ker pr* , then £ A( e; a) = 0
for any cell e (because Y [Y X(e,a)]B pr* (a) o: and thus
Le La e
“1 -iÀ (e ,a )
■
5 = T1
# , [ (T a a e ) a e '] When k = 1, this argument is valid
1( e , a)
only after we factor (X,A,Xq ) by its commutator subgroup, and it
k +1
only demonstrates that every element of Ker pr* is a product of the
above form multiplied by some commutators. However, since in
rr2 (X,A,xq) each commutator y ^ 6y6 ^ equals (T 9 Y 6) 6 1 (see 1.4.7),
-1
we obtain again the desired decomposition of £ into ratios (T a)a
a
In the general situation, we first transform (X,A) into a
k-connected pair, removing those components of X which do not contain
x and then replace it by a homotopy equivalent pair (X ’,A *) such
0'
that ske, X 1 c= A ’ (see 1.4.6 and 2.3.3.1). Thus, we may assume that
k
ske, X c A. Now set Y = A U ske, 1X and consider the commutative
k K +1
diagram
l =in*
Tk + 1 (Y,A,x q )

P ’ = pr, P = prv

7Tk+1 (Y /A,Pr (XQ ) ) — -- -» 7Tk +1 (X/A,pr(xQ)).

Here i , i ', and p' are epimorphic: i because ske^^X c Y,

i' ske, 1 (X/A) c Y/A, and p' because of the proof above.
because
K.+ 1
We claim that our diagram also satisfies the last condition of the
algebraic lemma: Ker i' c p'(Ker i) .
To see this, note that every (k+2)-cell from (X/A) \ (Y/A)
is the image under p of some cell e from X ^ Y, and its
corresponding attaching map can be expressed as pr ° att0 . By Theorem
2.4, this implies that Ker i' c p r j K e r inj , where in* =

= [in*: TTk + 1 (Y,x0 ) ^k+ 1 (X,x0 )] and pr* = [pr* : ^k+ 1 (Y,x0 ) "
-*• TTk + 1 (Y/A,pr (xQ) )] . Since the diagram

in*
Tk + 1 (X'x 0 )

rel^ rel.

^k+i <Y 'A ' V ^k+l^X 'A , x 0 *

commutes and pr* = p ’ ° [rel* : Trk+ 1 (Y,xQ) •+ Trk+ ^ (Y,A, xQ)] , we see that
pr* (Ker in*) c p ’ (Ker i) , whence Ker i ’ c p ’(Keri).
 458

Applying Lemma 1, we conclude that p is epimorphic and

Ker p = i(Kerp') . We have proved already that Kerp' is generated
by the ratios (Toa)a ~ 1 with a £ (Y ,A,xQ ) and a € ^(A^).

Since i ((T a)a_1) = [T (i (a) )] (i (a) ) and i is epimorphic, Kerp

o o _^
is generated by the ratios (T^a)a with a £ (X,A,Xg) and
a € (A,x q) .
PROOF OF THE CASE k = 0. The commutativity of (4) is
obvious, while the fact that rel* is epimorphic results from the
connectedness of A. It remains to verify that abs pr* is an
epimorphism with the indicated kernel. If is the unique 0 -cell of
X, this follows from 3.3: indeed, the system of generators and relations
for (X/A,pr (Xq )) given in 3.3 can be obtained from the system of
generators and relations for ^(X^), also appearing in 3.3, by
deleting the 1-cells and 2-cells of A. When xQ is not the unique
0-cell of X, we may still reduce to this special case by translating
inside A the origin at some 0-cell eQ , and subsequently replacing
the triple (X,A,eQ) by a homotopy equivalent triple (X^A^e^)
having a single 0-cell e^. To produce such a triple, we take quotients
twice: first the quotient of X by a one-dimensional contractible sub­
space of A containing ske^A, and then the quotient of the resulting
space by a one-dimensional contractible subspace which contains all its
0-cells (cf. the proof of Theorem 3.6).

3. Let X be a cellular space, and let A be a simply

connected subspace of X. Then X/A is k-connected (0 £ k £ 00) if
and only if the pair (X,A) is k-connected. If this condition is
fulfilled for some k<°°, then pr* : (X ,A,xQ ) -+ 7ik+ ^ (X/A,pr (xQ ))
is an isomorphism.

The second assertion is an obvious corollary of Theorems 2

and 1.4.6. The first assertion follows from the second by induction on
k. However, note that to deduce the k-connectedness of X/A from
the k-connectedness of (X,A) , Theorems 2.3.3. 1 and 2.3.2.2 suffice.

4. If the cellular space X with the 0-cell xQ as base

point is k-connected, then su: t\^+ ^ (X ,xq ) -+ TTk + 2 (su (X,xQ) ,bp) is an
isomorphism.

This is a corollary of 3 (see 2.1.2).

5 (INFORMATION). Under the assumptions of Theorem 3, if A

is 1-connected (1 £ 1 £ *>), then pr* : TTr (X,A,x0 ) -* ^ (X/A, pr (xQ) )
is an isomorphism for r £ k +1 and an epimorphism for r = k +1 + 1 .
Under the assumptions of Theorem 4, su : tt (X,x„) -*
r 0
 459

7Tr + '1 (su (X,xQ ) ,bp) is an isomorphism for r £ 2k and an epimorphism

for r = 2k+ 1 (cf. 2 .1 .4 ).

7. Spaces With Prescribed Homotopy Groups

1 (LEMMA) . Let tt be a group and n a positive integer.

If tt is Abelian or if n = 1, then there is a connected cellular space
X such that all the groups tt^ ( x ) with r / n are trivial, whereas
TTn (X) ~ TT . *

PROOF. We proceed by induction and construct connected

cellular spaces X q ,X^,..., with base points x^,x ,..., and base-point
preserving cellular embeddings (J)Q : XQ -+ , (p^ : X^ X^, such that:
(i) the groups with r < n and n < r £ n+k are trivial;
(ii) Tin (X ]c 'x k ) is isomorphic to tt ; ^ ( X ^ x ^ -+ % (Xk + 1 ' Xk + 1 J
4>k ^ :

is an isomorphism. Then the space X = lim (X^.,(¡^) will have the

desired properties (see 2 .1.5. 6 and 1 .1 1 .2 ).
To produce (Xq/Xq), write tt as a quotient group F/F1,
where F is a free group if n = 1, and a free Abelian group if n > 1.
Let B and B' be bouquets of n-dimensional spheres such that
TTn (B,bp) = F and ,bp) = F' (see 1.3 and 5.6). Further, let
f: (B',bp) -+ (B,bp) be a continuous map such that f *: TTn (B',bp) -*
TTn (B,bp) equals the inclusion F * -> F (one can construct such a map
out of a family of spheroids whose classes in TT^(B,bp) = F constitute
a free system of generators for F 1). Now replace each sphere in B*
by the ball that it bounds and take X^ to be the result of attaching
this new bouquet (of balls) to B by f. Theorems 1.3, 5.6, and 2.3
show that Xq and xQ = imm^ (bp) [= imm2 (bp)] satisfy conditions (i)
and (ii) for k = 0.
Assume that for some i ^ 1, pointed spaces ^Xk'xk^' k i'
and maps <j>, , k < i-1 , are already constructed and satisfy conditions
iC
(i), (ii), and (iii) . Represent Trn+i (Xi _ 1 / x i _ 1 ) as the quotient group
of a free Abelian group, say G, and then construct a bouquet C of
(n+i)-dimensional spheres, together with a map g: (C,bp) ->• (X± _ 1 )
such that g*:
. (C,bp) + tt
tt (X ,x ) equals the projection
n+ 1 n+ 1 1-1 1 “!
G tt . (X. „,x. „). [To establish the existence of such a C and a
n+i i- 1 i- 1
g, one may procced as in the proof of the existence of B 1 and f
above; however, here Theorem 1.3 is not necessary.] Now .replace each

* Translator's note. A space with such homotopy groups is known as a

cellular K(7T,n) -space or as a cellular space of type (7r,n).
 460

sphere of C by the ball that it bounds and then attach the resulting
bouquet to X. by g to obtain X.. Finally, set x = inutu (X. .)
1^ I ~ 1 1 ¿J_~l
and 4k = imm2 . The fact that (X^x^) satisfies (i), (ii) for
k = i, and (Jk satisfies (iii) for k = i- 1 , is a consequence of
2.3.
2. Given an arbitrary group it and arbitrary Abelian grou
2 , tt3, ..., there exists aconnected cellular space X such that
TTr (X) s 7Tr (r = 1 ,2, ...) .

PROOF. Let X^,X2 ?... be connected cellular spaces with

0-cells x ,x ,...as basepoints, such that the groups tt (X, ) are
1 z r K
trivial for r i k, whereas ^(X^) " 71^ (see 1). Define inductively
cellular spaces Yg,Y ,... and cellular embeddings : ^ Yk +1 ^

Yq = D°, Yk+1= Yk xc xk + 1' and ^ k (y) = (y,xk+1) * A p p ling Theorems

1.1.9 and 1.11.2, the space X = lim(X^,ipk ) has the desired properties.

8. Eight Instructive Examples

1. If r > 1, then the r-th homotopy group of a finite,

connected cellular space is not necessarily finitely generated (cf. 3.4).
r 1
The bouquet (S ,ort^) V (S ,ort^) is a simple illustration of this
phenomenon: its r-th homotopy group (r > 1) is a free Abelian group
r 1
of infinite rank. Indeed, (S jOrt^) V (S jort^ has a covering space
which is homotopy equivalent to an infinite bouquet of r-dimensional
spheres: to produce such a space, attach one copy of Sr in one point
at each integer point of the real line HR .

INFORMATION. The homotopy groups of a finite cellular space

with finite fundamental group are finitely generated. For a proof, see
[19] .

2. Under the conditions of Theorem 1.6.6, the subgroup

Ker rel* = Im in* of tt^(X,Xq ) is not necessarily normal (cf. 1.5.14).
Example: X is the bouquet of two circles, A the first circle, xQ
the center of the bouquet, and p takes the second circle into xQ .

3. Under the conditions of Theorem 1.6.7, the right splitting

of the homotopy sequence of the pair (X,A) at tt (X,xQ) is not
necessarily normal (cf. 1.5.17). Example: X = (D^ort ) V (S1,ort ),
1 1 1 1
A = (S ,ort ^) V(S ,ort1)f Xq is the center of both bouquets, and

h = [imm2 : S 1 X] ° [p^: X ->S 1].

 461

4. For any k ^ 0 there exist k-connected pairs (X,A) with

A connected, which are not (k+1)-simple; moreover, under the conditions
of Theorem 6.2, and for any k ^ 0, the epimornhism pr* : tt (X,A,xn) +
k+ 1 U
^k+l (X/A,pr (x q )) ^-s not necessarily an isomorphism. Example:
X = (Sk+ 1 ,ort1) V (S1 ,ortr), A = imm2 (S1) (cf. 1 and 3).

5. The second homotopy group of a pair (X,A) with connected

A is not necessarily Abelian, even when it (A) is Abelian and X is
1 1 M
simply connected. The simplest example: A = S x S , xQ = (ort^ort^,
2
and X is the result of attaching two copies of D to A by the maps
1
S + A given by y h- (y ,ort ) and y *+ (ort1 ,y) • Then tt (A) = ZZ @ 2Z,
I I 9
i\2 (A) = 0, tt^ (X ) = 0, ^2 (X) = 2Z (Xis homotopy equivalent to S ),
and we have the exact sequence

0 -iS-*— > ZZ rel* > tt2 (X,A,x0) — » E © ZZ in-*— > 0, (5)

which shows, in particular, that 3 is epimorphic. Assuming that

tt2 (X,A,Xq) is Abelian, it follows from 1.4.7 that tt^ (A,Xq) acts
identically on - ^ ( X / A ^ q ), whence, by Theorem 2.2, rank tt^ (X,A,x^) £ 2.
The latter contradicts the exactness of (5).

I 6. There exist 1-connected pairs (X,A) such that

pr* : tt^ (X ,A ,X q ) tt^ (X/A,pr (xQ)) is not even epimorphic. Example:
2 1
X = D ,A = S , xn = ort . Here tt (X,A) = 0for r ^ 2, whereas
r 2
tt2 (X/A) s ZZ(X/A is homeomorphic to S ).

7. For anyk > 2, there are (k—1)-connected but not

k-connectedcellular pairs (X,A) with X and A
connected and X/A
1 k
contractible. To construct an example, let 0 € ir^ ((S ,ort^)V(S ,ort^),bp)
1 k
and a € tt, ((S ,ort ) V (S ,ort ) ,bp) designate the classes of the
K I I
1 1 k k
spheroids imm^ : S (S ,ort^)V (S ,ort^) and imm2 : S -*■
1 k k +1
(S ,ort^) V (S ,ort^), respectively. Next, attach D to
1 k k 1 k
(S ,ort1 ) V (S ,ort^) by an arbitrary spheroid S -+ (S ,ort^)V (S ,ort^)
in the homotopy class 2a - Taa. Take the resulting cellular space as X,
and the circle ske.X as A. The quotient space X/A may be described
k+ 1 k k k
as the result of attaching D to S by a map S -»■ S homotopic
to id Sk , which implies that X/A is contractible (see 1.3.7.8 ). It
is evident that X and A are connected and that (X,A) is (k—1)-con­
nected; therefore, it remains to check that tt, (X) is not trivial. By
1 k
1 , iTk ((S ,ort 1 ) V (S ,ort1) ,bp) is a free Abelian group with free
generators a = T^a (n = 0,±1,...), while (X) is the quotient
 462

1 k
group of tt , ((S ,ort ) V (S ,ort ) ,bp) by its subgroup generated by the
K. I I
elements 2a - a
. (see 6.2). Consequently, tt, (X) is isomorphic to
n+ 1n K
the additive group of binary rational numbers.

8. The homomorphism f* : tt^(X,A,Xq) -> tt^ (X ',A 1 ,x^ ) induc

by a continuous map f: (X,A,xQ) (X^A'^x^) is not necessarily an
isomorphism even if all the homomorphisms f* : Tir (X,XQ) 'nr.(X' ^x^)
and (ab f)* : tt (A,Xg) tt^ (A '<x^) are isomorphisms. Example:

X = X' = (S1,ort ) V (1,0), A = imm 1 (ort2) U imm2 (1),

A ’ = imm^(S^) U imm2 (1 ) , x^ = x^ = imm2 (1 ) , and f = relidX.

9. Exercises

2
1 Consider the subset of (C defined by the equation
p g where p and q are comprime integers, and intersect it with
X1 X2'
,3
the sphere S'". Show that the fundamental group of the complement of
this intersection in S^
is isomorphic to the group with generators
a^,a2, which are connected by the relation p g
= a 2-
(The above intersections (with various p,q) are all homeo-
morphic to a circle and lie on the torus |x1 1 = v/2 /2 , <2 ! = /2 / 2
3
which is contained in S ; they are called torus knots. A torus knot
and the torus which carries it are depicted in Fig. 16 as they lie in
3 3 ,
IR = S \ b p .)

2
2. Consider the submanifold of (CP defined by the equation
X1 2 3 = ' Wlth m a positive integer (cf. 3.5.4.2). Show that
the fundamental group of the complement of this submanifold is
isomorphic to .

3. Consider the submanifold of nonzero vectors in the total

 463

space of the tangent bundle of a sphere with g handles, and show that
its fundamental group is isomorphic to the group with generators
a 1»'* **'aa'k
y 1i'’’ ’ g which are connected by the relations

a b a 1b 1 ...a b a 1b 1 = d 2 2g
1 1 1 1 g g g g
and
a 1d = da , .. .,a d = da^, b1 d = d b 1 ,...,bd=db .
l l g g 1 1 g g

4. Consider the submanifold of nonzero vectors in the tot

space of the tangent bundle of a sphere with h crosscaps, and show
that its fundamental group is isomorphic to the group with generators
c^,...,c^,d, which are connected by the relations

,2 -h
C 1 C 1 * **ch°h ' d
and
-1 “ 1
c.d = d c . , . . . ,c, d = d c, .
I I n n

5. Let tt be any group which can be presented by a finite

number of generators and relations. Show that there exists a smooth,
closed 4-dimensional manifold whose fundamental group is isomorphic to
TT .

6. Show that every smooth, closed, oriented manifold of

dimension / 0,3, is oriented cobordant to a simply connected manifold,

INFORMATION. This is also valid for dimension 3.

7. Show that in Example 8.5, tt2 (X,A,Xq) is isomorphic to

the group with generators a,b,c, which are connected by the relations
-1 -1
ac = ca, be = cb, and aba b = c.

§4. WEAK HOMOTOPY EQUIVALENCE

1. Fundamental Concepts

1. If X and Y are topological spaces, a continuous ma

f : X -> Y is called a weak homotopy equivalence if
f : tt (X,x) tt ( Y ,f (x) ) is an isomorphism for all r ^ 0 and all
* r r
x £ X. To justify this term, we remark that every homotopy equivalence
 464

is a weak homotopy equivalence and that the converse is not true. The
first fact was established in 1.3.7? for the second, see 3.5.
The composition of two weak homotopy equivalences is obviously
a weak homotopy equivalence.
2. Let f : X -> Y be a weak homotopy equivalence. Then for
any cellular pair (K,L) and continuous maps <j>: K -* Y and ijj: L X
with f ° ip = <$>I there is a continuous map y: K + X such that
,rTi_T" |L ■ ..
x |IT
l = ip and f o y is L-homotopic to <p. — The converse is also true;
moreover, if f: X -* Y is continuous and has the property that for any
continuous maps, <j>: Dr Y and ip: S r ^ •+ X (r ^ 0) with f ° ip =
1 , there is a continuous map x* D -+ X such that x „ = ip and f °x
sr 1 Sr - 1 —
r —'I
S -homotopic to (p, then f is a weak homotopy equivalence.

To prove thefirst part, consider the mapping cylinder Cyl f

(see 1.2.6.10). By Theorem 2.3.1.3, there exists a homotopy
0 : K x I -*Cyl f of the composition K Y — 1— — ► Cyl f ,such that
$(z,t) = imm^ (ip (z) ,1-t) for all z £ L, t £ I. Since
rtf o [in: X Cyl f]= f and f is a weak homotopy equivalence, while
rtf is a homotopy equivalence, we conclude that in is a weak homotopy
equivalence. Hence, in*: TTr (X,x) -* 7Tr (Cylf,x) is an isomorphism for
all r ^ 0 and all xE X. Thus, (Cyl f ,X) is an «•-connected pair
(see 1.6.5), and since$ (L x 1) c X, there is an L-homotopy
K x I -*Cyl f such that ^(x,0) = 0 (x,1) for all x £ K, and
m (K x 1) c x (see 2.3.1.6 ). Now define x : K ** X by x(x ) = $(x,1).
Obviously, xIt - ^r and the product of the homotopies rtf ° $ and
I
rtf » ! is an L-homotopy from f ° x t o $■

To prove the second part, we must check thatf* : TTr (x,x)

-»• tt (Y,f(x)) is both epimorphic and monomorphic for all r ^ 0 and
r - 1
x £ X. To see that f* is epimorphic, set ip (S ) = x and take $
to be some spheroid (Dr ,Sr_1) -+ (Y,f(x)) of an arbitrarily given class
3 € nr (Y,f (x) ) ; the resulting spheroid x= (Dr ,Sr-1) -+■ (X,x) defines
a class a € nr (X,x), and it is clear that f* (a) = 3. Finally, to
show that f* is monomorphic, pick a € it (X,x) with f . (a) = 0,
r *
take ip
to be some spheroid of class a, and take <j> to be any
vr+1
continuous map D Y such that <j> - f © \p; the resulting map
r +1 S
X: D X extends ip, and hence a = 0.

3. If f: X Y is a weak homotopy equivalence, then the

mapping 7T(id,f): tt(M,X) -+ tt(M,Y) is invertible for any cellular space
M.
 465

PROOF. The first part of Theorem 2 shows that, given a class

3 € tt(M,Y), there is an a € tt(M,X) such that [tt(id,f)] ( a ) = 3: we
need only set K = M, L = 0, and take (j> to be any map in the class
3. Further, given arbitrary continuous maps M X, it follows
from the same first part of Theorem 2 that if the compositions f o $
and f o (f)^ are homotopic, then so are cJ>Q and <f> : indeed, take
K = M x I, l = (M x 0) U (M x 1 ) , and <j):(M x 0) U (M x 1 ) -> X,
cf)(x,0) = (J)q(x), cf)(x , 1 ) = (j>^ (x) , x € M, and take for ip anyhomotopy
M x I X from f o (J)q to f o (j)^ .

The Case of Cellular Spaces

4. Tf X and Y are cellular spaces, then every weak

homotopy equivalence X «+ Y is a homotopy equivalence.

PROOF. Suppose f: X + Y is a weak homotopy equivalence.

By 3, the mapping tt (id,f ) : tt(Y,X) -* tt(Y,Y) is invertible, and hence
there is a map g : Y -> X whose homotopy class is taken by Tr(id,f)
into the class of idY . That is to say, fog is homotopic to idY ,
and it remains to verify that g ° f is homotopic to id X . The latter
is a consequence of the invertibility of Ti(id,f): tt(X,X) tt(X,Y),
because this mappingtakes the homotopy classes of g 0f and id X
into the same element (indeed, f og o f and f are homotopic) .

5. Theorem 4 states that two connected cellular spaces, X

and Y, are homotopy equivalent whenever there is a continuous map
X -+ Y which induces isomorphisms of the homotopy groups, but it
certainly does not guarantee that X and Y are homotopy equivalent
if theirhomotopy groups arejust isomorphic. In fact, we have simple
examples to show that the latter is not true. Take X = SP x ]RPq ,
Y = Sq x ]RPP , and suppose that 1< p < q . By 2.5.1 and 1.1.9,
Tir (X) ~ f°r r * However, X and Y are not homotopy
equivalent. Indeed, the map pr^ : Sp x nRpq + s p induces a group
isomorphism tt (Sp x3RPq ) -> tt (Sp ) . We next show that there is no
P Q p P p
continuous map f : S^1 x ORP*^ -> S- which induces a group isomorphism
it (Sq x 1RPP ) •> tt (Sp ) . Assuming that such an f exists, the
P P
composition

SP _£2L_ mpP * (ort1rx), sq x mpP gP (1)

induces an automorphism of ir (Sp ) . On the other hand (by 2.3.2.4),

every continuous map 1RPP -> Sp is homotopic to a map which takes

 466

IRpP"1 into ort , and thus(1) ishomotopic to the composition of

the composite projectionSP -+ 3RPP IRPP /IRPP^ = Sp with some
continuous map SP SP . Consequently, (1) cannot have degree ±1,
since the above composite projection has degree 0 when p is even
and degree 2 when p is odd. Contradiction.
The following example illustrates the same phenomenon in the
3 °° 2
simply connected case. Set X = S x (CP and Y = S . By 2.2.10,
2.5.2, and 1.1.9,
(X) ~ tt (Y) for all r.
tt However, X and Y have
^ 3 00 3
not the same homotopy type. Indeed, or : S x EP S is not null
1 3 °°
homotopic (because it induces a group isomorphism tt^ (S x £p ) = S ■>
3 2 3
-> (S ) = 2Z) . On the other hand, every continuous map S -► S is
null homotopic.

6. We say that a topological space is homotopy fit if it is

homotopy equivalent to a cellular space. From Theorem 4 it follows that
if X and Y are homotopy fit, then every weak homotopy equivalence
X -> Y i£ a homotopy equivalence.
By Theorem 3.5.2.13, all smooth compact manifolds are homotopy
fit .

An example of a space whichis not homotopy fit was given in

2.3.5.4. This space is not connected. For an'example of a connected
(and even ^-connected) space which is not homotopy fit, see 4.1 below.
INFORMATION. Every CNRS is homotopy fit, and the same holds
true for every topological manifold (compact or not). A product of
homotopy fit spaces is homotopy fit. If Y is homotopy fit, then
C(X,Y) is homotopy fit for any compact space X. If Y has the homotop
homotopy type of acountable cellular space, then C(X,Y) has the
homotopy type of acountable cellular space, for any compact space X
with countable base. For proofs, see [16].

k-Equivalence

7. Let X and Y be topological spaces. A continuous map

f: X Y is a k-equivalence if, for all x E X, f * : Trr (X,x) tt (Y,f(x))
is an isomorphism for r < k and an epimorphism for r = k. Here k
is a nonnegative integer; sometimes, weak homotopy equivalences are
referred to as “-equivalences.
A composition of two k-equivalences is obviously a
k-equivalence.

8. Let f: X ■+ Y be a k-equivalence. Then for any cellular

 467

pair (K,L) withK \ L c: ske^K and continuous maps (j>: K -> Y and
4*: L -> X with f o\p = (j)|t there is a continuous map y : K X such
,Li • —
that x |L = ^ and f © ^ is L-homotopic to c(>. The converse is also
true; moreover, if f : X -> Y is continuous and has the property that
for any continuous maps cf).: Dr Y and ip: Sr~1 -> X (0 £ r £ k) with
f ° $ = <f> r_ 1 there is a continuous map y : Dr -> X such that
S r- 1
X r _i
sr = ^ and f ° x is S -homotopic to <f>, then f is a
k-equivalence.

The proof repeats that of Theorem 2, with obvious changes:

in the first pair, the pair (Cyl f ,X) is now k-connected; in the
second part, to see that f* is epimorphic (monomorphic), we take r£ k
(respectively, r < k).

9. Ij£ f: X Y is a k-equivalence, then the mapping

Tr(id,f): tt(M,X) 7T(M, Y) is invertible (surjective) for any cellular
space M with dim M < k (respectively, dimM = k) .

The proof repeats that of Theorem 3, except that we need

Theorem 8 instead of Theorem 2.

10. rf X and Y are cellular spaces with dim X < k and

dim Y £ k, then every k-equivalence X Y is a homotopy equivalence.

The proof repeats that of Theorem 4, except that we need

Theorem 9 instead of Theorem 3.

The Relative Case

11. If (X,A) and (Y ,B ) are topological pairs, a conti

map f: (X,A) -+ (Y,B) is said to be a weak homotopy equivalence if
abs f : X + Y and abf (=ababs f) : A -+ B are weak homotopy equivalences.
We remark that if f: (X,A) -+ (Y,B) is a weak homotopy
equivalence, then f * : Tir (X,A,x) -> (Y ,B ,f (x)) is an isomorphism for
all r ^ 1 and all x G A. To see this, apply the 5-Lemma (see 1.5.19)
to the homomorphism induced by f from the homotopy sequence of the pair
(X,A) into the homotopy sequence of the pair (Y,B). As another
corollary of the 5-Lemma, we have the following result: suppose that
f: (X,A^0) -> (Y ,B ) is continuous, that one of the maps abs f : X Y,
ab f : A B is a weak homotopy equivalence, and that for any x £ A all
the homomorphisms f* :TTr (X,A,x) -* (Y ,B ,f (x)) , r > 1, (absf)* :
ttq (X,x ) + 7Tq (Y ,f (x) ) , and (abf)* :ttq (A,x) -> TrQ (B,f(x)) are
isomorphisms. Then f is a weak homotopy equivalence.
 468

12. Ij[ f: (X,A) ■+ (Y,B) is a weak homotopy equivalence,

then the mapping ir(id,f): tt(M,N;X,A) -*■ n (M,N;Y,B) is invertible for
any cellular pair (M,N).
PROOF. Let us show first that every continuous map
<J>:(M,N) + (Y ,B) is homotopic to the composition of some continuous map
(M,N) (Y ,B) with f. By 2, there is a continuous map \p: N + A
whose composition with ab f : A -+ B is homotopic to ab (p : N B.
Using 2.3.1.3, any homotopy from ab cp to ab f °\p may be extended to
a homotopy of <p, and hence there is a (p': (M,N) -► (Y,B) homotopic
to (p and satisfying ab $ * = ab f ° ip . Finally, again using 2, we see
that there is a continuous map x : M -► X extending ij; and such that
f^x is N-homotopic to <J)1. Obviously, x (N) c B and the maps
f 0 x ^ : (M,N) (Y ,B ) are homotopic.
To complete the proof, we have to show, given two continuous
maps (M/N) + (X/A) such that f ° $0 and f 0 (p ^ are homotopic,
that (¡)q and <J> are also homotopic. Let 0: (M x i,n x I) -> (Y,B) be
a homotopy from f o <pQ to f ° <f>^ . By 2, there is a homotopy
y: N x I a from ab <f>Q : N -> A to ab <p ^ A, such that ab f ° V
is [(N x 0) U (Nx 1)]-homotopic to ab 0 :N x I B. Further, by
2.3.1.3, every [(N x 0)U (N x 1)]-homotopy from a b $ to ab f 0V
extends to a [(M x 0) U (M x 1)]-homotopy of 0. Consequently, there is
a homotopy 0 1: (M x I,N x I) + (Y,B) from f ° <p^ to f ° <p^ such
that [ab$f : N x i b] - ab f ° V . Finally, we apply 2 again to deduce
that there exists a continuous map H: M x i ^ x extending ¥, such
that H(x,0) = cJ)q(x), H(x,1) = <}> (x) for all x £ M. In other words,
E is a homotopy from (pQ to cf> .

13. I_f (X,A) and (Y,B)are cellular pairs, then every

weak homotopy equivalence (X,A) (Y,B) is a homotopy equivalence.

The proof repeats that of Theorem 4, but one must refer to

Theorem 12 instead of Theorem 3.

2. Weak Homotopy Equivalence and Constructions

1. Many of the operations on maps which were described in

§1.2 carry weak homotopy equivalences into weak homotopy equivalences.
For example, it is clear that [ |^f ^ : [ |^X^ -+ _[ x^ J is a weak
homotopy equivalence for any family {f : X + X 1} of weak homotopy
y y y y^ M
equivalences; similarly, f. x ... x f : x„ x ... x x + X 1 x . .. x x ‘
1 n 1 n 1 n
is a weak homotopy equivalence for any weak homotopy equivalences
 469

f ^ : Xi -> X,j,...,f^: -> X ^ . Also, the limit of a sequence of of weak

homotopy equivalences is a weak homotopy equivalence.
Similar results for other constructions need supplementary
arguments, or hold only under additional assumptions. The present
subsection is devoted to such results.

2. Let f: (X,A,B)(X1,A',B') be amap of triads such that

ab f : A -> A' , ab f : B B ', and ab f: A fl B A' O B ' are weak
homotopy equivalences.If Int A U Int B = X and Int A' U IntB' = X',
then abs f : X X1 is a weak homotopy equivalence.

PROOF. Using Theorem 1.2, and given continuous maps

4) : Dr -> X 1 and ip : Sr ^ -> Xwith f ° ^ = (p I *r it suffices to
r 'sr
produce a continuous map x: D ** X such that => \p and f
r— 1 sr _ 1
is S -homotopic to <p. Obviously,

U= [4>”1 (IntA’)n Int Dr ] U' 1|T1 (Int A)

and
V = [<t>_ 1 (IntB’)n Int Dr ] U I))‘1 (IntB)
r
are open and cover D . Therefore, there is an e > 0 such that any
subset of D
with diameter less than e is contained in U or V.
r— 1
Now triangulate S so that the diameter of each simplex is less than
e, and then extend the triangulation to D , preserving this property.
Let K (L) be the union of all simplices contained in U (respectively,
V) . It is clear that K and L are simplicial subspaces of D such
that ip (Kn Sr 1) c Int A , \p (L D Sr 1) c Int B , <p (K) cInt A 1 , and
<p (L) c= I n t B 1 . By Theorem 1.2, there is a continuous map Xq:k n L -+
-* A fl B such thatx0 = ab \p and the composition of Xq with
~ KOLOS" r-1
ab f : A il B A 1 il B' is (KOLOS )-homotopic to ab <p : K D L ->
r— 1
A 1 fl B '. By 2.3.1.3, every (K fl L fl S )-homotopy
r— 1
from ab <j> : K n L -*• A 1 fl B 1 to ab f ° xn extends to a (K fl S )-
r—1
-homotopy of ab <t> : K ->■ A ' and to a ( L O S )-homotopy of ab : L -+ B\
r— 1
The two resulting homotopies combine to define an S -homotopy from <j>
to a map <t>' : Dr X’ which satisfies = ^¡AflB* ° x0' c A *'
and 4>’ (L) c B ’. Finally, apply Theorem 1.2 again to deduce that there
are continuous maps x-j: K -»■ A and X2 : L B with the following
properties: xQ = a b x 1 / XQ = ab x2 '> the composition of X-, with
a b f : A -*• A 1 is (Kfl L)-homo topic to ab <f>’ : K -*• A ’; and the
composition of X? with abf : B + B 1 is (K il L)-homotopic to
r
ab <j)’ : L -»■ B ’. Now the desired map x= D -*■ X is obtained by combining
x and x 2•
 470

3.Suppose that (X,C) and (X',C') are Borsuk pairs, Y

and Y* are topological spaces, and cp:C Y, cp' : C1 Y 1,
f: (X,C)-> (X1,C 1) / and g: Y -* Y' are continuous, with g © =
= cp* ° ab f . If f and g areweak homotopy equivalences, then the
formulas

F © [imm. : X Y U X] = [imm. : X ' -> Y 1 U , X f] ° f

1 cp 1 cp
and
F ° [imm2 : Y Y X] = [imm2 : Y* Y 1 U^, X 1] ° g

define a weak homotopy equivalence F: Y X + Y1 t X'.

PROOF. Let us glue X, Y, and Cxi, identifying each point

(c,0) E C x I with c E X, and each point (c,1) E C x I with cp (c ) GY.

Let Z denote the resulting space and, to avoid confusion, denote the
maps imm^ : X -+ Z, imm2 : Y Z, and imm^ : C x i z by a, 8, and y.
First, we want to show that the formulas

h ° a = [imm^ : X YU X] , ho g = [imm2 : Y Y X],

and
h ° y = [imm2 : Y ->■ Y X]© cp ° [pr^ : C x I c]

define a homotopy equivalence h: Z Y X. To this end, extend the

the homotopy y: C x I z to a homotopy r: X x I z of the map a
and define k:: Y U
k X -* Z by k(imm^(x)) = r (x,1 ) [x £ X] , k o imm2 =
= 8. Then the formulas

H (imm^ (x) ,t ) = h (T (x ,t )) for x E X, t E I,

^ H(imm2 (y),t) = imm2 (y) for y E Y, t € I,

and

K(a(x),t) = T(x,t ) for x E X, tel,

K (8 (y) ,t ) = 8 (y) for y E Y, t € I,
L K(y(c,t),u) = y(c,tu-u+1), for c E C, t £ I, u E I,

define a homotopy H: (Y U X) x I + y U X from h ° k to id(Y U X)

<P cp cp
and a homotopy K: Z x I -+ z from k ° h to id Z . Consequently, k
is a homotopy inverse to h.
Now repeat all this for X ', C ', Y ', cp', .... We obtain a
space Z ’, continuous maps a 1: X ’ ■+ Z 1, 3': Y' ■+ Z', y': c ’ x I ->■ z ' ,
and a homotopy equivalence h': Y' U ,X' ■> Z'. Let G: Z -+ Z'denote
the map defined by G ° a = a ' ° f , G ° g = S' ° g, g ° y =
= y 1 ° (ab f x id I) . It is clear that G maps the triad
 471

(Z,a(X) U y (C x [0/1/2])/6(Y) U y ( c x I)) into the triad

(Z', a ' (X*) U y ' ( C x [0,1/2]), 8* (Yf) U y* (C' x i ) ) . At the same time,
all the conditions of Theorem 2 are fulfilled. Thus, applying this
theorem, G is a weak homotopy equivalence. Finally, the commutativity
of the diagram

Y U<p x Y ’ U^, X'

----------------------------------- - z 1

implies that F is a weak homotopy equivalence.

4. Let f: (X,A,B) ( X ^ A ^ B 1)be a map of triads such that

abf: A A 1, abf: B B ', and a b f : A fi B A * H B ' are weak
homotopy equivalences. If (A,A 0 B) and (A1,A 1 fl B f) are Borsuk
pairs, then abs f : X X1 is a weak homotopy equivalence.

This is a corollarv of Theorem 3, because X = B U. A,

in
in = [in: A fl B B] , while X' = B' IK A', in = [in: A 1 fl B 1 B '] •

5. Let (X ,A) and (Xf,A!) be Borsuk pairs. If

f: (X,A ) (X1,A 1) is a weak homotopy equivalence, then so is
fact f : X/A + X 1/A1.

It suffices to apply Theorem 3 for Y = D°, Y* = D°.

6. If f : X -+ X 1 is a weak homotopy equivalence, then so is

su : su X su X 1.

It suffices to apply Theorem 5 to the map relconf: (conX,X)

(con X 1,X 1) .

7. lf_ f: X + X 1 and g: Y Y1 are weak homotopy

equivalences, then so is f * g: X * Y X 1 * Y*.

PROOF. Let A and B be the images of X x Y x [0,2/3]

and X x Y x [1/3,1] under the projection Xxyxl->x*Y.
Similarly, let A' and B1 be the images of X* x Y 1 x [0,2/3] and
X f x Y* x [1/3,1] under the projection X1 x y * x I -> x 1 * Y 1.
Obviously, (f * g) (A) c a 1, (f * g)(B) c B ', and rel (f * g) :
: (X * Y,A,B) (X* * satisfies the conditions of Theorem 2.
Therefore, f * g is a weak homotopy equivalence.

8. I_f f: X ->■ Y is a weak homotopy equivalence and K is a

locally finite cellular space, then C(id,f): C(K,X) C(K,Y) is a weak
homotopy equivalence.
jo
Given arbitrary continuous maps cf>: D C(K,Y) and
 472

<Jj: Sr 1 -> C(K,X) with C (id,f) ° ip = „, we need only exhibit a

sr “ 1
continuous map x : Dr ^ C(K,X) such that r-1 = Ip and C (id, f ) 0 x
S a
is S,r-1 -homotopic to <f) (see 1.2). Consider the maps <(> : D x K Y,
A
tj/': s r_1 x K X (see 1 . 2 . 7 . 6 and 2 . 1 . 4 . 3 ) , which are continuous and
satisfy f ° ip A = <jjA _ . The first part of Theorem 1.2 applies to
Sr x k r
j. A and ipA and yields a continuous map a: D * K -*• X such that
= \pA , as well as an (Sr~1 x K)-homotopy h: (Dr * K) * I Y
Sr , XK
from <j)A to f o a. Denote by H the composition of the canonical
homeomorphism (Dr x I) x k -> (Dr x K) x I with h, and set x = a
(see again 1.2.7.6). It is clear that x . = i(j and that Hv is an
v-*— 1
sr - 1
S -homotopy from $ to C(id,f) ° X-
9. Let Xy and be topological spaces with base points
xy and x^ such that (X^fx^) and are Borsuk pairs. If
f : (Xy^xy) are weak homotopy equivalences, then so is
V f : V (X ,x ) V (X' ,x' ) .
y y y y y y y y
This is a corollary of 5.

3. Cellular Approximations
of Topological Spaces

1. A cellular approximation of the topological space X is

any pair (K,cJ)) consisting of a cellular space K and a weak homotopy
equivalence : K + X.
Example: if X is a Hausdorff space with a cellular
decomposition enjoying the property that each compact subset of X
intersects only a finite number of cells, then (X~,idX) , where X~
is the cellular space obtained from X through the cellular weakening
of its topology, is a cellular approximation of X. In particular, if
X1/...,X are cellular spaces, then (X1 x ... x x ,id) is a
i n I c c n
cellular approximation of X. x ... x x , while (X„ * ... * X ,id)
r 1 n' 1 c c n
is a cellular approximation of X^ * ... * X^.

2. Every topological space admits cellular approximations.

PROOF. We observe first that cellular approximations of the

components of a topological space yield a cellular approximation of the
entire space, while the case of an empty space is trivial. Thus, we may
assume that the space X we want to approximateis connected and
nonempty. We shall construct a sequence KQ ,K^,... of cellular spaces
 473

with O-cells YQ/Y-jf**- as base points, a sequence of continuous maps

^i : (X,Xq ), i = 0,1,... (here x^ £ X is an arbitrarily
fixed point) and, finally, a sequence of cellular embeddings
H i : ^i-i ' ^ i - ^ (Kj_/Yj_K i = 1/2,..., such that <p ^ is an
i-equivalence and $ ± ° n± = (i)i.1 • Then the pair(lim K± ,lim <J> ) will
be a cellular approximation of X (see 1.11.2).

Proceed by induction. Set KQ = D°, yQ = D°, <p^ (KQ ) = xQ ,

and assume that K_^, y^, (¡>^, n^, i < r, have been defined and enjoy
the required properties. Pick a spheroid f : Sr-*■ X in each homotopy
class a £ Trr ( X , X Q ) . Further, for any class 3 £N, where
N = Ker [(4>r_-| )* : 7Ir_ 1 *Kr_i ,yr-i > ^ (x 'x o ^ ' pick a spheroid
r— 1
gs : s K r-1 of class (3, together with a continuous map
27
hR : D X satisfying h
r-1 = ^r-1 0 g B ' The maps f ot' g 3 ' and h i

(with a £ irr (X,XQ) and 3 € N) combine to define three other

continuous maps:

f! #a € ^ ( X , x 0) (Sa = * *•

g: -!— LgeN (S3 s * Kr-1 ' and h: -I— Lg£N(DS d ) ^ X'

I
Now set

Kr ■ < < V l Ug l- i - l BQID6 " ' ira”2 (l' r - i n V

and define nr to be the composition

imrru imm
K . ------ ► K „ U (I L c ^D.) ---- -— ► K ,
r-1 r-1 g J— LBGN 0 r

(p^: -> X to be the map assembled from (p^ ^ , h, and f, and y^

to be nr (yr ^ ) - Applying Theorems 3.2.3 and 3.5.5, we see that <p^ is
an r-equivalence, and it is clear that ° = ^r-1 anC^ t h a t nr is
a cellular embedding.

3. Let (K,<M and (K 1,(j)1) be cellular approximations of

topological spaces X and X 1, and let f: X X1 be an arbitrary
continuous map. Then there is a continuous map g : K ■+ K 1 such that
the diagram

is homotopy commutative (i.e. , the maps f ° <|> and (p1 ° g are

 474

homotopic). This property uniquely defines the homotopy class of g.

PROOF. The mapping tt(id ,4>') : tt(K,K') -* tt(K,X') is

invertible (see 1.3), and thus in tt(K,K') there is a unique element
which is taken bytt ( id , cj)1 ) into the class of f ° (p.

4. Let (K, 4))and (K ',(})1) be two cellular approximations

of the same topological space X . Then there is ahomotopy equivalence
g: K -> K1 such that <p' ° g ishomotopic to (p.

To obtain g, apply Theorem 3 to X' = X and f = id X . To

get a homotopy inverse g' to g, interchange the roles of (K ',(p')
and (K,cf)) . Finally, apply Theorem 3 again to show that g ° g' and
gf ° g are homotopic to id K' and id K, respectively.

Weak Homotopy Equivalence

as an Equivalence Relation

5. Two topological spaces are said to be weakly homotopy

equivalent if they admit cellular approximations (K,<{)) and (L,^)
with K = L. It is clear that this defines an equivalence relation (in
the usual, set-theoretic sense).
Let X and Y be topological spaces such that there is a
weak homotopy equivalence f: X -*Y. Then X and Y are weakly
homotopy equivalent. Indeed, if (K ,4)) is a cellular approximation of
X, then (K,f ° c|)) is a cellular approximation of Y. The converse is
false: there are examples of weakly homotopy equivalent spaces X, Y
such that there is no weak homotopy equivalence X Y and no weak
homotopy equivalence Y X. See 4.2 below for such an example.
Two homotopy equivalent spaces are certainly weakly homotopy
equivalent. The converse is false: for example, every topological space
is weakly homotopy equivalent to a cellular one (see 2), but not every
topological space has the homotopy type of a cellular space (see 1.6).
On the other hand, Theorem 1.4 shows that two weakly homotopy equivalent
cellular spaces are actually homotopy equivalent.

Weak Homotopy Equivalence

of the Fibers of a Serre Bundle

6. Any two fibers of a Serre bundle with connected base are

weakly homotopy equivalent.
 475

PROOF. Let ^ be the given Serre bundle with bs £

connected, and let bQ/b E bs £ . Pick a path s: I -+ bs £ with
s(0) = b^, s (1 ) = b^, and a cellular approximation (K,<|)) of the
fiber pr £_1 (bQ) .The map f~ = [in: pr £_1 (bQ) -»■ tl£] ° <f> and the
homotopy F : K x I-* bs £ , F(x,t) =s(t), satisfy F(x,0) = pr £ o f ~(x)
[x e K] , and hence there is a homotopy F~: K x i -> tl £ of f~ which
covers F (see 4.1.3.6). Now define ip: K pr ^ (b^) by x h- f~( x ,1)
and check that ip is a weak homotopy equivalence.
To do this, given any x £ K and any spheroid g £ Sphr (K,x),
note that the formula (y,t) h- F~(g(y),t) defines a fiber homotopy
(see 1.7.1) from the spheroid <f>^(g) to the spheroid 4>^(g) along the
path s : I ->■ tl £given by t h- F~(g (ort^) ,t ) . Consequently, the
diagram
TTr (K,x)
K
T
Tj.ipr? 1 (bQ ) , (x) ) » TTr (pr £-1 (b1 ) ,ip (x) )

commutes (the translation

T~ is defined in 1.7.3). Since T ~ is an
s s
isomorphism, the invertibility of <J>* implies the invertibility of ip*.

Cellular Approximations of Topological Pairs

7. A cellular approximation of the topological pair (X,A)

is any pair [(K,L ) ,<j)] consisting of a cellular pair (K,L) anda weak
homotopy equivalence (p: (K,L) (X,A).
When A and L are points, a cellular approximation
[(K,L),cJ>] of (X,A) is termed a cellular approximation of the pointed
space (X,A ) .

8. Every topological pair (X,A) admits cellular

approximations. Moreover, given any cellular approximation (L,i/0 of
the subspace A, there is a cellular approximation [(K,L) , ] of
(X,A) with ip = ab cj).

PROOF. Let ( M,x) be a cellular approximation of X (see 2),

and let g: LM be a cellular map such that X ° 9 is homotopic to
[in: A X] ° ip (see 3 and 2.3.2.4). Set K = Cyl g and define <j> to
be the relativization of the map K + X given by x and some homotopy
L x I -> x from [in: A -> X] o ^ to X ° 9- Obviously, c|> is a weak
homotopy equivalence and ab<p = ip.
 476

9. Let [(K ,L ) ,<j)] and [(K ',L ') , 1] be cellular

approximations of the topological pairs (X,A) and (X',A'), and let
f: (X,A) (X',A ') be an arbitrary continuous map.Then there is a
continuous map g: (K,L) (K',L') (unique upto homotopies)such that
the diagram

(K,L) (K ',L ')

*'
'i'
(X, A) (X*,A' )

is homotopy commutative. If [(K,L),c(>] and [(K',L 1) ,<p1] are cellular

approximations of the same topological pair, then there is a homotopy
equivalence g: (K,L) (K',L 1) such that <f>’ ° g is homotopic to $.

The proof repeats the proofs of Theorems 3 and 4, except that

one has to refer to 1.12 instead of 1.3.

Cellular Approximations and Constructions

10. It is clear that if (K ,<j>^) are cellular approximations

of the spaces X (y EM), then ( | | -MK , I I cb ) is a cellular
y J— Ly E M u J— Ly E M y
approximation of J Ly€MXy # Also, applying Theorem 2.9, we see that if
(X^,x^) are Borsuk pairs (x^ £ X^ are base points) and [(K^,y ),<(> ]
are cellular approximations of these pointed spaces, then
[V (K ,y ) ,V 6 ] is a cellular approximation of the bouquet V (X ,x ) .
y y y y y ^ y y y
Further, if (K^,^),..., (K^,<f> ) are cellular approximations
of X^,...,X^, then (K^ ... x^ Kn '(i)'| x ••• x 4>n ) is a cellular
approximation of X 1 x ... x x (see 2.1 and 1). In the same
I n
circumstances, (K * ... * K ,$ * ... * <j> ) is a cellular
I c c n I n
approximation of X 1 * ... * Xn (see 2.7). In particular, (suK,sucf>)
is a cellular approximation of su X whenever (K,d>) is a cellular
approximation of X.
If [(K,L ),(J)] is a cellular approximation of the Borsuk pair
(X,A) , then (K/L,facte}) : K/L X/A) is a cellular approximation of
X/A. This is a corollary of 2.5.
 477

An Application:
Generalization of Theorems 3.3.6, 3.5.5, and 3.6.2

11 (LEMMA). Let (X,A,B) be a triad with the property that

either Int A U Int B = Xor (A,A fl B) is a Borsuk pair. Then there
exist a cellular triad (K,L ,M) and a continuous map f: (K,L,M) +
(X,A ,B ) such that abs f : K X, ab f : L + A, and ab f : L D M
+ A fl B are weak homotopy equivalences.

PROOF. By 2, A fl B has a cellular approximation, say (N,x)/

and the latter can be extended to cellular approximations [(L*,N),(j)]
and [(M',N),iJj] of the pairs (A,A fl B) and (B,B fl A), as shown by 8.
Next, attach the cylinder N x I to L'J |_Mr by the map
(N x 0) u (N x 1) -> lJ [M, (x,0) h*in ^ (x ) , (x ,1 ) *+ in2 (x), and call
the resulting cellular space K. Now identify N x I, L', and M*
with their images in K and set L = (N x I) u Lj M = ( N x i ) U M l.
The composite maps
Pr 1
N x I J »n — *A 0 B — — ► X,

L1 — > A -^2— > X, and M1 ^ > B —^ — > X

jointly define the map f: (K,L,M) (X,A,B) . Obviously, Int L U Int M =

= K and ab f : L -+ A, ab f : M -+ B, and a b f : L fl M + A fl B are weak
homotopy equivalences. Therefore, so is absf : K + X (see 2.2 and
2.4) .

12. The homomorphism

[tt1 (A,xQ ) * tt1 (B,X q )]/vk (X,A, B ,X q ) ->- tt^ X jX q ), (2)

defined in 3 . 3 . 5 , is an isomorphism not only for a cellular triad

(X,A,B) with A, B, A n B connected (as assertedby Theorem 3 . 3 . 6 ) ,
but also for any triad (X,A,B) such that A, B, A n B are connected
and either Int A U Int B = X or (A,A fl B) is a Borsuk pair. In fact,
this follows from Theorem 3 . 3 . 6 and Lemma 11, since the homomorphism (2)
is natural.
In particular, we see that the fundamental group of the
bouquet of two spaces is canonically isomorphic to the free product of
the fundamental groups of these spaces under the only assumption that
each space forms, together with its base point, a Borsuk pair (cf.
3.3.7).

13. Concerning Theorem 3.5.5, we can weaken the demand that

 478

the pairs (X ,x ) be cellular and instead ask only that they be Borsuk
y y
pairs. That this is possible is guaranteed by Theorem 8, the discussion
of bouquets in 10, an the commutativity of diagram (3) in 3.5.1.

14. Theorem 3.6.2 and its corollary 3.6.3 are valid not on
for cellular pairs, but also for arbitrary Borsuk pairs. This
generalization follows from Theorem 8 and the last statement on
quotients in 10.

4. Exercises

1. Consider the union X of the graph of the function

x sin(1/x) on the interval 0 < x £ 1/tt and the broken line made of
the four segments with the successive vertices ( 1 / tt, 0 ) , (1 /tt ,2 ) , (-1,2),
(-1,0), and (0,0). Show that X is «-connected but not homotopy fit. (Cf. 1 .6)

2. Let A={0,2n |nEZ}c= HR (cf. 4.2.4.2) . Show that the

1 1
spaces X = Sj [(A x S ) and Y = Aj [(2Z x S ) are weakly homotopy
equivalent, but there is no weak homotopy equivalence X -> Y, and no
weak homotopy equivalence Y X. (Cf. 3.5.)
3 *
3. Let X denote the subset of IRconsisting of the segment
I and the sequence of segments withendpoints n ort^ , ort^ + ort^/n
(n = 1,2,...). Show that (X,(x^,x2 ,x^) x^,I) is a Serrebundle, but
there exists fibers which are not homotopy equivalent.

4. Suppose (X,Xq), (Y,y^) are pointed topological spaces,

Z is a cellular space with a 0-cell z^ for base point, and f: X Y
is a weak homotopy equivalence with = Yq • Show that
abC(id,f): C(Z,Xq ,*X,Xq ) C(Z,ZQ;Y,yQ) is a weak homotopy equivalence.
(Cf. 2.8.)

§5. THE WHITEHEAD PRODUCT

1. The Class wd(m,n)

1. In this section we define and study some of the propert

of an operation on the elements of homotopy groups. In a certain sense,
this operation generalizes the action of the fundamental group on the
 479

homotopy groups. The definition assumes that a pair m,n of positive

integers is given.
The present subsection is devoted to a very specific
preliminary construction. Recall (see 2. 1.3.2 and 2. 1.5.2) that the
cellular decomposition of Sm x sn , determined by the standard
decompositions of Sm and Sn (each having two cells) consists of
four cells: an (m+n)-cell and three other cells which form the bouquet
(Sm ,ort^) V (Sn ,ort^). We denote this bouquet by B(m,n)or, simply,
by B. The standard characteristic map of the (m+n)-cell is the
composition of the canonical homeomorphism Dm+n Dm x Dn (see 1.2.6.9)
with the map DS x D S ; it takes Sm+n ^ into B, and takes the point
(ort. + ort ^->//2 into bp = (ort.,ortJ. Therefore, this characteris-
1 m+1 1 1
tic map defines an element of the group 7Tm + n ^s x Sn /B,bp) (see 2.2.5),
which we call Wd(m,n) or, simply, Wd. Also, we write wd(m,n) or,
simply, wd, for the element 9 (Wd) € ^m+n-l^B 'kp)/ i.e., the class of
the attaching spheroid Sm+n ^ B.
We need two additional notations: 0 for the homeomorphism
B(m,n) -> B(n,m) which permutes Sm and Sn , and yfor theproduct
of the spheroids imm^ ,imm2 :(S^ort^) + (B(m,n),bp) when m = n.

2. The class wd has infinite order.

It is enough to establish that Wd is of infinite order and

that
3: % +n (sm x Sn 'B 'bP> -

is monomorphic. The first is a consequence of the fact that the homo­

morphism

pr* : % + n (sin x Sn 'B 'bP> ^ m + n ^ 3"1 * sn)/B=Sm+n'Pr (hp) ) =

takes Wd into a generator of the right-hand group. The second claim

follows from the exactness of the homotopy sequence of the pair
(Sm x s n ,B) with base point bp, because in*: iTm + n (B,bp) +
it (Sm x sn ,bp) is epimorphic (see 3.5.4).
m+n
3. The isomorphism 0*: 1Tm+n_-| (m,n) ,bp) 7Tm+n_i (n,m) ,bp)

takes wd(m,n) into (-1)mnwd(n,m).

This results from the commutativity of the diagram

s„m+n-1 . „m+n-1
> ^

0
B (m,n) B (n,m)
 480

where the vertical maps are the attaching spheroids which represent the
classes wd(m,n) and wd (n,m) (see 1), while the upper horizontal map
is given by (x., xm+n) h- (*m + r xm+n 'X 1 ' '’'' V (and itS de^ree
is (-1)mn).
4. If_ m = 1, then

wd(m,n)
' = imm~.(sph
2* ^ n )[T.lmra^(sph^)^imm0
2*. (sphn )]
—1
In particular, wd(1,1) = c^ct^c^ a , where (as in Subsection 3.1),
,a2 denote the elements imm^ * (sph.|) ,imm2* (sph ^) £ tt^ (B (1 ,1 ) ,bp) .

PROOF. According to 1, wd(1,n) is represented by the

spheroid Sn B(1,n),

imm^ o D S f / J x ^ , if |x^| £ 1/ /2,

imm.Z ° DS (/2 (x0

Z
, ...,x
n+1
)) , if |x | 5 1//2.
1

This is obviously homotopic to the product of the spheroid

bp, if x^ £ 1//2,
(1 )
imm2 » DS (/2 (x2 , .. .,xn+1 )) , if x1 5 1 / Æ

with the spheroid obtained by translating the spheroid

imm2 o DS (/2 (x2 ,...,xn+1)) , if x 1 £ -1//2,

(2 )
(X1.... Xn+1> *
bp, if x^ ^ -1//2,

along the path t imm. ° DS(1-2t). Now it remains to observe that the
' _1
class of (1) is imm2*(sphn ), the class of (2) is [imir^* (sph^) ] ,
and the class of the above path is imm^*(sph^).

The class wd(m,n) belongs to the kernel of each of the

following three homomorphisms:

p r l * : ^ m + n - l (B'b P> - rort^),

pr2*: V n - 1 , B ' b P> " % +n - 1 ( S n ' o r t 1 ) '

and

in* : V n - i (B'bp) ■* w i (sm x sn'bp)

For in*/ this results from the exactness of the homotopy

sequence of the pair (Sm x Sn ,B) with base point bp. For the first
 481

and the second homomorphisms, use the equalities [pr^: B Sm ] =

= tPr 1 : Sm x Sn + Sm ] o in and [pr2 : B ->• Sn ] = [pr2 : Sm x S11 + Sn] ° in.

6. The homomorphism (is S V y)*: ^m+n_>| (B (m, n ) ,bp)

^m+n-l((Sm'ortl) ^ (B(n,n),bp),bp) takes wd(m,n) into

[(idSm V imm^ )* (wd (m,n) )] [(id Sm V imm2)*(wd(m,n))] .

PROOF. When m = 1 or n = 1 this follows from 4 and 3;

now let m > 1 and n > 1. The bouquet (Sm ,ort^) V (B(n,n),bp) is
simply connected, and it yields the product Sm x B(n,n) when we add
two (m+n)-cells with attaching spheroids Sm+n ^ + (Sm ,ort^) V (B(n,n),bp)
belonging to the classes (id Sm V imm^)*(wd(m,n)) and
(id Sm V imm^)*(wd(m,n)) . Consequently, the kernel of the homomorphism

in* : ^m+n-l ^(sm^orti) v (B(n,n) ,bp) ,bp) + n +n-1 (S™ x B(n,n),bp)

is generated by the indicated classes. This kernel contains also the

class (id Sm V y )*(wd(m,n)). To see this, note that wd(m,n) sits in
the kernel of the homomorphism induced by the inclusion B(m,n) -> Sm x Sn
(see 5), while id Sm V y is the compression of the map
id Sm x y ; sm x sn - sm x B(n,n). Therefore,

(id Sm Vy)*(w d (m ,n )) =
k k?
= [(id Sm V imm^)* (wd (m,n) )] [(idSm V imm2 )* (wd (m, n) )] , (3)

with k^ ,k^ £ 2Z , and we shall presently show that k^ = = 1.

The compositions (idSm V pr^ ) o (id Sm V y) and

(id Sm V pr2) ° (id Sm V y) , where pr^, pr2 are the projections of
B(n,n) onto Sn , are both homotopic to idB(m,n) . At the same time,
(id Sm V pr^ )© (id Sm V imm1) = (id Sm V pr2) o (id Sm V imm2 ) =

= id B (m,n ) , while both (id S™ V pr1) ° (id Sm V imm2) and

(id Sm V pr2)o (id Sm V imrn^ ) equal the composition

pr imm
B(m,n) --- 1
— >S ---- !
— ► B(m,n).

Now applying the homomorphisms (id Sm V pr^)* and (idS V pr2 )*

to both members of (3) and using Proposition 5, we get wd(m,n) =
k1 k2
= wd(m,n ) , w d (m ,n ) = wd(m,n) . Finally, these equalities yield,
by virtue of 2, k^ = 1, k2 = 1.
7. The class wd(m,n) belongs to the kernelof the
homomorphism
 482

su: tt „ (B(m,n),bp) ■> tt (su (B (m, n ),bp) =B (m+1 ,n+1 ) ,bp) .

m+n-1 m+n

PROOF. By 2.1.1, the diagram

Pr -|* pr
2 . /Cn I )
(S ,ort. %
^m+n_-, (B(m,n),bP ) tt
m+n-1 1
7rm+n- 1 (Sm'ort 1 )
su SU su
,_m+1 pr 1* ^ 21 n +1
. .
r ort 1 )
^m+n 'ort 1 >
— tt , (B (m+1 ,n + 1) ,bp) ---
m+n m+n (S
tt

commutes. This, combined with Theorem 5, shows that su(wd(m,n))

belongs to the kernels of pr^* and pr^*/ and thus to the kernel of
the homomorphism

% +n (B(m+1'n+1)'bp) ^ 1Im+ n (sin+1'ort1) ® TTm + n (Sn+1 'orV

given by pr^* and Pr2*- Finally, recall that the last homomorphism
is an isomorphism (see 3.5.5).

2. Definition and the Simplest

Properties of the Whitehead Product

1. Let (X,Xq ) be a pointed topological space, and let

a £ tt^^Xq), 3 £ tt^^Xq). Clearly, the homotopy class of the map
h: (B,bp) -+ (X,Xq) defined by arbitrary spheroids (Sm ,ort^) -+ (X,Xq)
and (S ,ort1) -+ (X,Xq) representing a and 3 is independent of the
choice of these spheroids. Therefore, the element h* (wd(m,n))£ Tim+n_^ (X,x^)
is determined solely by the classes a and g. This element is called
the Whiteheadproduct of a and 3, denoted [oc,3 ] .

Notice that in terms of this definition, wd(m,n) itself is

the Whitehead product of the classes of the spheroids imm : (Sm ,ort1) -*
n
(B,bp) and imn^ : (S jort^) -* (B,bp), i.e.,

wd(m,n) = [imm. . (sph ),imm0 .(sph )].

1* m z* n

It is readily cheked that f*([a,3]) = [f* (a),f* (3)] for

any a £ ^¡ji(X ,Xq ) , 8 £ iTn (X,x0), and continuous f: (X,xQ ) (Y,yQ ).
Furthermore, T ( [ a , 3] ) = [T a,T 3] for any a £ tt (X,xn ), 3 £ tt (X,x_),
s s m u n U
and any path s : I -+X with s(0) = X q .

2. If a £ tt^ ( X ,Xq ) and 3 £ iTn ( X , x 0 ) , then

[6,a] = ( - 1 ) mn [ a , 3] .
 483

Indeed, if h: (B,bp) -+ (X,Xq ) is the map defined by two

spheroids, (S ,ort^) ■> (X,x^) and (Sn ,ort^) (X,x^) which represent
the classes a and 6, then

[6,a] =(h © 0)*(wd(n,m)) = h * (0*(wd(n,m)) =

= h * ( (-1)mnwd(m,n)) = (-1) mn [a , 8 ]

(see 1.3).

3. If a £ (X,Xq ) and G Un^X,X0^ with n > 1,

then [a,31 + B~] = [a,B ] + [a,8-]. If a ,a„ € tt (x,xn ) with ra> 1
\ z i z i z m o ----
and B £ ^n(X'x q ) , then [a1 + a ^ B ] = [a ,B] + [a2 ,S] .

Because of 2, one has to prove only the first equality.

Consider the map h: ((Sm ,ort.j) V (B(n,n) ,bp) ,bp) (X,xQ ) definedby

arbitrary spheroids, f: (Sm ,ort1) -► (X,xQ) and g ^ ,g 2 : (Sn ,ort1) -►

->■ (X,Xq ) , representing the classes a and 8.j,82 , respectively.
Then the map (B(m,n),bp) -*■ (X,Xq) defined by f and g^ equals
h o (id Sm V imm^), the map defined by f and g2 equals
h ° (idSm V imn^) , and finally the map defined by f and the product
of the spheroids g^ and g2 equals h ° (idSm V y) . Hence,

[a,B1 +B21 = h* o (idSm V y)*(wd(m,n)) =

= h*((idSmV imm^ )* (wd (m,n) ) + (idSm Vimm2)*(wd(m,n))) =

= (h o (idSm V imm^ ))* (wd (m,n) ) +

+ (h o (idSm V imm2 ) )* (wd (m,n) ) = [a,8.|] + [a,B2].

4. If^ a £ tt^ (X,Xq) and 8 £ TTn (X,xQ ) with n S 1, then

-1 -1 -1
[a,8] = B(Ta6) . In particular, [ot,B] = 8aB a for any
a, B £ tt^ (X ,X q ) .

This is a corollary of 1.4.

5.
For any■*- a £ tmt (X,x„)
------ u and
--- B £ tt_(X,
n x „),
u the
----product
-------
[a,B] belongs to the kernel of the homomorphism su: TTm+n_1 (X,xQ ) -+
TT (su (X,xn ) ,bp) .
m+n u
This is a corollary of 1.7.

6. For any a £ TTm (X,xQ) and B £ the product

[imm1* (a),imm2 (B)] belongs to the kernel of each of the homomorphisms:
 484

V n - i ((x'xo) v (Y' V ' bp) + W i (x'x0>'

Pr 2* : V n - 1 (,X ' x 0 ) V <Y ' V ' bp) % +n - 1 ( Y ' V '

and
in*: V ^ d X , ^ ) V (Y,y0),bp) - *m+n_.,(X x Y,bp).

This is a corollary of 1.5.

7. Generally speaking, the Whitehead product is not associ
tive. This was already implicit in 4: if we let (as in Subsection 3.1)
a^,a2 ,a^ denote the elements imm^ * (sph^ ) , imm2*(sph^), imm^*(sph^) of

7T1 ( (S1 ,or 1 1 ) V (S1 ,ort 1) V (S1 ,ort1 ) ,bp) , then [[0^ , 0^ ] , ^ ] =
_1 _^ ^ ^ ^
= a^a2a^a2 a2 , whereas [0^ , [a2 , ]] =
“1 —1 1 1 1
= a 3 a 2 a 3 a2 aia2a3a2 a3 a1 ‘ A second example can be found in 4 .2 below,

The Case of H-Spaces

8. rf X is a H-space, then [a/3]/= 0 for any x^E X,

a 6 TTm ( X, x 0 ) , 6 € Trn (X,x0 ) .

(This generalizes Theorem 1.9.11.)

It is enough to consider the case where x^ is the identity.

Let f: (Sm /ort<l) (X,Xq ) and g: (Sn ,ort>|) (X,Xq ) be spheroids in
the classes a and 3. Define h: (Sm x Sn ,bp) (X,Xq ) and spheroids

f 1 : (Sm /ort1) + (X,X q ), g 1 : ( S ^ o r t ^ -> (X,xQ) by h(x,y) = f(x)g(y),

f1 (y) = f (y)x q / <?1 (y) = g(y)xQ* Obviously, f1 and g are homotopic
to f and g, while the map (B,bp) (X,Xq ) defined by f^ and g^

equals h|D = h o [in: B Sm x Sn] . Consequently, [ a , 6] =

I
= (h o in)* (wd(m,n)) = h* 0 in*(wd(m,n)). Since in*(wd(m,n)) = 0,
we get [a,0] = 0.

3. Applications

1. Let (X,Xq ) and (Y,y^) be pointed spaces, and let

k and 1 be nonnegative integers. If X is k-connected, Y is
1-connected, and (X,xQ), (Y,yQ ) are Borsuk pairs, then the kernel of
the homomorphism
 485

in* : ^k+l+l ( (X,x0 ) V (Y'^o),bp) ^ \ + l +1 (X X Y' bp)

is generated, as a subgroup of \ +l+ 1 ((X,xQ) V (Y,yQ),bp), by the

products [inun1* (a) ,imm2* (3) ] with a £ irk+1(X,x0 ) and 3 € irl+ 1 <Y,y0 )

(Cf. 3.5.5 and 4.3.13.)

PROOF. By 2.6, [imm,,

V
* (a) , imm^* (B) ] £ Kerin*
' ~ ““ ‘2*
for all
aG (X,Xq) and B £ +(Y,Yq) . To see that these products actually
generate Kerin* , note that 4.2.1, 4.2.9, and 2.3.3.2 together guaran­
tee that it suffices to examine the case where (X,Xq ) and (Y ,y^) are
cellular spaces with ske, X = x^ andske,Y= y . . Under these
k 0 1 ^ 0
circumstances, s k e ^ - j ^ (X x Y) c= (X,x^) V (Y,y^), and the classes of
the attaching maps of the (k+1+ 2)-cells in (X x Y) \ [(X,X q ) V (Y,Yq)]
are Whitehead products of the classes of thecharacteristic maps of the
(k+1) -cells in imm^ (X) and the classes of the charactetistic maps of
the (1+1) -cells in imm^(Y) (this is an immediate consequence of
Definition 2.1). Therefore, when k > 0 and 1 > 0, Kerin* is
generated by the classes [imm^* (a),imm2* (B)] with a £ TTk + ^(X,xQ) and
B £ TT1 + -] (see 3.2.4); when k > 0 and 1 = 0, Kerin* is
generated by the classes

T- m / v [imm * (a),imm9* (B) ]

imm2* (a) 1 z
with a £ 7Tk+<1 (X,xQ) and B,a £ Tr^(Y,yQ) (see 3.2.4); when k = 0
and 1 > 0, Kerin* is generated by the classes

T imm^*(a)
t mm 1 * <a > , i m m z * (B) ]

with a,a £ it (X,xQ) and 8 £ tt1+1 (Y,yQ) (see 3.2.4); finally, when
k = 1 = 0, Ker in* is generated by the classes

T. , v. , > . , i > [irtun. .(a) ,imm0 .(8) ]

* (a^)lmn^*(u^)...imm.j* <CTq)imm2* q
^,...,0 € tt1 (X,xQ ) and 8, w 1 , ..., £ tt^Y^) (see
Subsection 3.3). Now all it remains is to observe that

= - [imm^ (a) ,imm2* (8) ] + [imm1* (aa) ,imm2* (6) ]

for any a ,a £ tt^ (X,Xq) , 3 £ 7T]_+ -]{Y »Yg) with 1 > 0;

(ii) T. , . [imm.. (a),imm * (8)] =

imm2*(a) 1* 2*
 486

= - [imm^* (a),imm2*(a)] + [imm^* (a),imir^*(3cr)]

for any a € 7Tk+1(X,x0 ), 3 ,a £ tt^ (Y ,y q ) with k > 0; and

(iii) T. ... , x [imm,.(a),imnu (3)] =

' lmm^ * (a) imm2* (a)) 1* 2*
-1
= [iim^* (a) [imm^(aa) ,imm^^B)] [imm^(aa) jimm^fco)] _ 1 [imm^(a) (w)]

for any a,a £ tt^(X,Xq ), 3,w £ ^ ( Y / Y q ) (see2.4).

The class
2. [sph^sph^] has infinite order in ^ n - l ^ 3^ '
for every even positive integer n. In particular, the groups
2k
TT4k-1 ^ ^ with k ^ 1 are infinite.
(Cf. Subsections 2.2 and 2.4.)

PROOF. Since

y*[sphn ,sphn ] = [y*(sphn ) , y * (sphn )] =

= [imm..(sph )+ imm_.(sph ),imm..(sph ) +imm-.(sph )] =

1X n z* n 1* n Z* n

= [imm..(sph ),imm„.(sph )] + [imm0 .(sph ),immn .(sph )] +

1* n 1* n z* n z* n

+ 2 [imm. .(sph ),imrru.(sph )] =

1* n z* n
= imm..[sph ,sph ] + imnu.[sph ,sph ] + 2 wd(n,n) ,
1 n n Z* n n
we obtain

2 wd (n,n) = y. [sph ,sph ] - imm..[sph ,sph ] - imm„.[sph ,sph ].

n n Ix n n z* n n

Now assuming that [sphn ,sph ] has finite order, the class
wd(n,n) would have finite order too, contradicting 1.2.

3.
The kernel of su: tr„ „ (Sn ) -* tu (S2n + 1 ) is infinite
------------- zn~I Zn -----------
for every positive integer n .

This is a corollary of 2 and 2.5.

4. Exercises

1. Compute the third homotopy group of a bouquet of two-

-dimensional spheres.

2. Show that if

a = imm^ * (sph ^) £ -n (B (1 ,2 ) ,bp)

 487

and
3 = Y = i m m 2 * ( s p h 2 ) € tt2 (B (1 ,2) ,bp) ,
then

[ [a,6] ,y] t [a, [3,y]] •

3. S h o w that

(-1)p m [ [a,3] ,Y] + (-1)m n [[3 ,y ],a ] + (-1)np [[Y ,a] ,3 ] = 0,

for any a £ Trm (X,x0 ), 3 £ TTn (X,x0 ), an d Y € TTp (X,x0 ) wit h m > 1,
n > 1, and p > 1.

§ 6. CONTINUATION OF THE THEORY OF BUNDLES

1. Weak Homotopy Equivalence and Steenrod Bundles

1. Two Steenrod bundles, £ and £2 ,with the same

standard fiber F, are said to be k-equivalent if there exist a
cellular space B and two k-equivalences, c(>1 : B bs ^ and
4>2: B -> bs ^2 f such that the bundles (f)’^ and ^^2 are F~ec3uivalent •
Here 0 £ k £ oo; the most important case is k = «>, and two “-equivalent
Steenrod bundles are also referred to as weakly homotopy equivalent.
Clearly, every bundle induced from a Steenrod bundle by a
k-equivalence is k-equivalent with the original bundle. Moreover, bases
of weakly homotopy equivalent Steenrod bundles are weakly homotopy
equivalent, and by Lemma 1.5.9 this holds true for their total spaces
!
too. [To see this, consider the mapsad ^ and
ad (f>2 : ^2^2 ^2 an(^ h°momorPhisms that they induce from the
homotopy sequences of the bundles an<^ ^2 ^2 into ^ h e homotopy
sequences of the bundles and respectively, and apply to
these homomorphisms Lemma 1.5.19.]

2. Let and ^ be Steenrod bundles with the same

standard fiber F. Suppose that is universal. Then is
universal if and only if it is weakly homotopy equivalent to Ç^.

We first show that the condition is sufficient, i.e., that

given any Steenrod bundle Ç with standard fiber F and cellular base,
any (cellular) subspace B of bs £ , and any continuous map
 488

cj>: B -> bs ç2 such that is F"e(3uivalent to ^|b ' there exists

a continuous map ip: bs ç bs ^ such that ÿ|B = <f> and the bundle
ip1 is F-equivalent to Ç (see 4.4.2.2). By 1 , we can produce a
cellular space K together with weak
homotopy equivalences,
i i
f^ : K bs and f2 : K -► bs ç2 , such that an<^ ^2^2 are
F-equivalent. Using 4.1.2, there is a continuous g : B K such that
f2 © g is homotopic to <|). Therefore, in the chain of F-bundles:

(f1 ° g ) !ç1 = g! = (f2 ° * *ç2' ^|b' the

!
adjacent bundles are F-equivalent, and hence so are (f^ ° g ) “^ and
M . Since ç. is universal, there is a continuous h :bs Ç -+ bs ç
|B 1 i *
such that h|D = f. ° g andthe bundles h'ç and Ç are F-equivalent
IB 1 I
(see again 4.4.2.2). Further, since f is a weak homotopy equivalence,
there is a continuous k: bs £ -> K such that k|B = g and f^ ° k, h
are homotopic (see 4.1.2). The restriction of f ° k : bs Ç bs ^ to
B equals f2 ° g, andhence is homotopic to $. Consequently, there
is a continuous map ty: bs ç bs ^2 homotopic to f2 ° k such that
= (J). Now the fact that the adjacent bundles in the chain ÿ ‘ç2 ,

(f2 o k ) !^2 = k i (f2 ), k 1 (f^^1 ) = (f1 0 k) 1^ , h “^1, Ç , are

F-equivalent implies the F-equivalence of the bundles ip ' ç 2 anc^ K*
We may use what we just proved to show that the condition of
the theorem is also necessary. In fact, assume that ç^ and ^2 are
universal, and let (K,<J) ) and (K,cJ)9) be cellular approximations of
1 t z ,
bs and bs ç2 . Since an<^ ^2^2 are wea^^y homotopy
equivalent, they are both universal. Hence, andK2 are homotopy
equivalent (see 4. 4. 2. 4).

3 (COROLLARY) . I_f X and Y are classifying spaces of the

same topological group, then X and Y are weakly homotopy equivalent.

4. Given any topological group G, there is a universal

G-bundle with cellular base.

This follows from 4.4.3.4, 4.3.2, and 2.

5. A principal bundle is universal if and only if its total

space is “-connected.

PROOF. First, let Ç be a universal G-bundle. Applying 2,

K is weakly homotopy equivalent toMi G . Since tl Mi G is “ -connected
(see 2.3.3.10, 4.2.7, and 1.11.2), tl Ç is also “-connected (see 1).
Now let I be a Steenrod G-bundlewith “-connected total
space tl i .
Pick a cellular approximation(K,cj)) of bs Ç and
i
consider the G-bundle c|>‘£. Since the bundle M i G is universal, there
is a continuous \p: K + bs Mi G such that \p *Ç is G-equivalent to
 489

1 i
\J;#M i G . Furthermore, tl Mi G and tl(<|>-£) are «»-connected, and the
latter implies the ^-connectedness of tl(^'MiG) . Applying Lemma
1.5.19 (to the homomorphism from the homotopy sequence of the bundle
i
\p *Mi G into the homotopy sequence of the bundle M i G induced by the
i
map ad : ifj’Mi G ->■ Mi G ) , we see that is a weak homotopy
equivalence. Thus, £ is weakly homotopy equivalent to MiG , and so
£ is universal (see 2).

6. I_f X is a classifying space of the topological group G,

then for any r ^ 1 the groups tt^(X) and TTr_'j are isomorphic.

Indeed, by Theorem 5, the homomorphisms A figuring in the

homotopy sequence of the universal G-bundle with base X are
isomorphisms (see 1.8.7).

7. _If x and X2 are classifying spaces of the topological

groups G^ and G^, then X^ x X2 is a classifying space of G^ x •

PROOF. Given a universal G^-bundle and a universal

G2~bundle ^2 , it is enough to verify that the (G^ x G^) -bundle
x £;2 is universal. But this follows from 5 and 1.1.9.

8. Theorems 2 and 5 carry over to k-universal bundles. Thus,

if ^ and £2 are Steenrod bundles with the same standard fiber and
£ is k-universal, then ^2 is k-universal if and only if it is
k-equivalent to and a principal bundle is k-universal if and only
if its total space is k-connected. The proofs repeat those of Theorems
2 and 5, with obvious modifications.

An Application: Universal Principal

Bundles for Finitely Generated Abelian Groups

9. For arbitrary n and

00 00
(]R x ... x 3R x s x ... x s ,hel x ... x hel x pr x ... x pr,
v____________/ j j .->
n I n 1
1 1
S x ... x S x L(in1 ) x ... x L ( m 1 ))

n 1
is a universal (7L © ... © '&m )-bundle .In particular, S is

a classifying space for ZZ , while L(m) is a classifying space for

7L .
m
co
Since IR and S are oo-connected, this follows immediately
from 5 and 1.1.9.
 490

2. T h e o ry of C o v e r in g s

1. The main achievement of the present section is a clear

enumeration of the covering spaces of nonpathological connected spaces
and a criterion for the equivalence of two given coverings. Our
instrument will be the fundamental group. The analysis is elementary
enough: in fact, all we use from the whole theory of bundles may be
concentrated in the following two theorems, where it is assumed that a
covering £ together with a basepoint x £ tl i are given. Then:
(i) every path in bs £ with origin pr^(x) is covered by
one and only one path in tl £ with origin x; and

(ii) if two pathsin bs £ with common origin pr S(x) are

homotopic, then the paths in tl £ which cover them are also homotopic
and, in particular, have the same end.
(These assertions are straightforward corollaries of Theorems
4.1.3.6 and 4 .1 .3 .8.)
It is true that here and there we do refer to other facts from
the theory of bundles (for example, to Theorem 1.8.12), but all these
can be readily deduced from (i) and (ii).
At the heart of the theory of coverings lies the definition
of the group of a covering. Recall that, according to 1.8.12,
Pr* : 71-j (tl £/X) tt^ (bs £,pr £(x)) is a monomorphism for any covering
i with base point x C tl £ . We call the image of pr* the group of
the covering Z with base point x, and denote it by gr E(x) .

2. Let pr 5(x ) = pr £(Xq ) and let s be a path in tl £

beginning at x Q and ending at x^ . Then gr K (x^ ) = ^[gr U x q ) ] ^ ’1 ,
where a is the homotopy class of the loop pr C o s . In particular,
if prStx^) = pr £>(Xq) , then gr£(xQ) and grSix^) are conjugate
subgroups of tt ^ (bs ^,Xq) . The converse is also true: the groups
gr Ux) with x £ p r K 1 (pr U x Q )) exhaust the subgroups of
(bs 5,pr K (Xq) ) which are conjugate to gr U x Q) .

In fact, gr £ (x1) = pr S* (tt1 (tl Z ,x<1 )) = pr ^ o Tg (tt (tl £ ,x Q ) )

= Tp r £ ° s ° pr (tl K ’ K 0 ] ] = Tp r C o s (grC(x 0 n = °[9r U V ]a~1 •

These equalities demonstrate that for every o £ tt (bs S,pr£ (xn )) the
1 0
group o[gr Hxglla equals gr Us(1) ) , where s is a path in tl f,
with origin xQ and which covers some path in the class o.
 491

The H ie r a r c h y of C o v e r in g s

3. We say that the covering E, with base point x^ E t l £

is subordinate to the covering EJ with base point x^ £ t l £ 1 ,if
bs £' = bs ^and there exists a continuous map cf): E, 1 £ such that
bs cj) = id bs E, and tl <j)(x^) = x^. In this case, themap $ is called
a subordination.
Obviously, if cj): £ 1 £ is a subordination, then
(tl £ *,tl $ ,tl £) is a covering.

4. If a subordination exists, then it is unique.

PROOF. Suppose that the covering E, with base point x^

is subordinate to the covering E,' with base point x^, and let<J)
and ^ be two such subordinations. Then, if x1 E t l E, 1 is arbitrary
and s: I -*■ tl £ 1 is such that s(0) = x^, s (1 ) = x f, then the paths
tl cj) o s and tl ip ° s cover the same path, pr E,1 ° s , in bs £ and
have the same origin.Therefore, tl(f)0 s = t l ^ 0 s and tlcMx*) =
= tl cf) o s (1 )= tl o s (1) = 11 ip (x1).

5. If two coverings with base points are mutually subordinate,

then* the corresponding subordinations are equivalences which are inverses
of one another.

PROOF. Let (J): £ ' £ and <f)1:5 ^ be subordinations.

Then cp1 ° : £ 1 -*• ^ 1 and cf) ° cf)1: 5 -*■ £ are also subordinations. By
4, $ 1 ° (J) = id ^ 1 and cj) ° cj)1 = id 5 •

6 (LEMMA). Let E, and 5 1 be coverings with base points

xQ E tl £ andX q E tl ^ 1 , and such that bs £ 1 = bs £ , pr £ 1 (x^) =
= pr ^ (Xq ) / and gr E, 1 (x^) c gr ^(xQ ) . If the paths sjj,s£: I tl £ 1
have the common origin x^ and a common end, then the paths in tl £
which cover pr £ 1 ° s^ and pr E,1 ° s^ and have origin x^ also have
a common end.
PROOF. Let s^ and s2 be the path in t l E, with origin
xQ which cover pr E, ' ° s^f and pr i ? ° s^. The class of the loop

(pr E,1 © s ' ) (pr K ' ° $ 2 ) 1 belongs to gr 5 1 (x^) , and hence to gr 5 (xQ ) .

Consequently, the path in tl E, with origin xQ which covers this loop

is closed? moreover, it can be expressed as the product of a path which
covers pr £ 1 o s* (and which, having origin xQ , coincides with s^)
with a path which covers (pr ?f°s^) 1 (and which, having origin xQ ,
coincides with s^1)- Therefore, and s2 have the same end.
 492

7.Let £ and £' be coverings with base points xQ £ tl £

and xq £ tl £ 1, such that bs C = bs 5 andpr £' (x^) = pr K (xQ ) .
(i) UL £ is subordinate to £ 1, then gr (x^J) c gr £ (x^) ;

(ii) If gr £' (Xq ) c gr U X q ) and bs £ islocally connected,

then £ is subordinate to £ 1.
Assertion (i) is trivial, so let us prove (ii). For each point
x' £ t l £ * , consider the common end of all paths in tl £ which start
at Xq and cover paths of the form pr £ 1 ° u f, where u' is a path in
tl £ * such that u ’(0) = x^ , u 1 (1) = x* (see 6). This defines a map
F: tl £ ' + tl £ satisfying pr £° F = pr£ ' , and all we have to check
is the continuity of F. So let x 1£ tl £ ' and U be a neighborhood
ofF(x'). We shall produce a neighborhood U' of x ’ such that
F (U 1) e U.
To do this, let W c bs £ be a neighborhood of p r £ ’(x!)
such that there are neighborhoods, V' of x’ and V of F(xf), which
are homeomorphically mapped by pr £' and respectively pr£ onto W.
Since bs £ is locally connected, we can find a neighborhood <= W
of pr£'(x') such that every point of W^ can be joined to pr£'(x')
by a path in W. Now set Vjj = V 1 fl (pr £ ’) ^ (W^ ), = V il pr ^ 1 (W^),

and U' = Vj il (pr £' ) ^(pr£(U)). Obviously, x' £ U*. Let us check
that F(U1) c u.
Let y' £ U' . Since ab pr £ ': V' W is a homeomorphism,
thereis a path v ': I tl £ ’ such that v ' (I ) c V ' and v ' (0 ) = x ',
v ' (1 ) = y '. Define v: I tl £ by

V ( t) = (pr C|v ) " 1 (pr V O V' (t) ) ,

pick a path u': I tl £ ' with the property that u'(0) = x^ , u'( 1 ) =
= x', and consider the path u :I -*tl £with u(0) = xQ , u(1) = F(x’),
and ucovers the path pr £ 1 o u'. Clearly, (u'v')(0) = xQ,
(u'v ') (1) = y', while the path uv covers p r £ 1 °(u'v') and
(uv) (0) = X q . Therefore, (uv)(1) = F(v'), and hence F(y’)= v(1) =
= (pr £ |y )~1 (pr ^'(y1)) £ V fl pr (pr (U'))c U.

8 (COROLLARY) . Two coverings, £ and E,', with bs £ ' = bs £

a locally connected space, are equivalent if and only if for some points
XQ £ tl £ , x^ £ tl £' , such that pr %' (x^) = pr £ (xQ ) , the groups
gr Z (xQ) and gr £ ' (x^ )are conjugate in (bs £ ,pr £ (xQ ) ) .
 493

The G ro u p of A u t o m o r p h is m s of a C o v e r in g

9. As with the automorphisms (i.e., self-equivalences) of an

arbitrary bundle £, the automorphisms of a covering E, form a group
Aut £.* Its structure is described in Theorem 10 below, where xQ
designates, as usual, a base point in tl£, and two new notations are
used. Namely, let ev be the map from A u t E, into the fiber
-1
pr 5 (pr£(x0 )) given by ev (<f>) = tl 4)(xQ ) , and let Reg be the set
of all points x € p r f 1 (pr U x Q)) such that gr^(x) = gr^(xQ ) .
From 2 it follows that the preimage of Reg under the map
A: tt1 (bs E, ,pr £ (xQ) ) + TrQ (pr C 1 (pr £ (xQ )) ,xQ)= pr (pr £ (xQ )) is
nothing else but the normalizer Nr(gr^(x^)) of the group gr5(x^)
in tt ^ (bs ^ ,pr £ (Xq )) . [Recall that, given a subgroup H of a group G,
its normalizer Nr (H) is the set of all g € G such that gHg_>l = H;
Nr(H) is a subgroup of G and contains H as a normal subgroup.]
Therefore, A induces an invertible map

fact ab A: Nr (gr E, (xQ ))/gr 5 (xQ) + Reg.

10. ev is injective, ev(Aut^) c Reg, and the composition

_ , ab ev _ (fact ab A) , rt r/ *
Aut ------- > Reg ---------------> Nr (gr E, (xQ ))/gr £ (xQ )

is an antihomomorphism. If the base bs £ is locally connected, then

ev (Aut E,) = Reg , and hence the group Aut £ is antiisomorphic to
Nr(gr ^ (Xq ))/gr £ (Xq ) • If, in addition, £ is regular (see 11 below) ,
then Nr (gr ^ (xQ)) = tt^ (bs £ ,pr £ (xQ)) and Aut £ is antiisomorphic to
tt1 (bs £,pr £ (xQ))/gr C (Xq ) .

PROOF. Since every automorphism 4 € Aut £ may be thought

of as a subordination of the covering £ with base point tltMx^) to
the covering E, with base point xQ , the injectivity of ev is
immediate from 5. Similarly, the inclusion ev(Aut^) c Reg follows
from part (i) of Theorem 7, while the equality ev(Aut£) = Reg results,
when bs £ is locally connected, from part (ii) of the same theorem.
_ \
Finally, that (fact ab A) ° ev is an antihomomorphism is plain.

* Translator's note: the elements of Aut£ are also known as covering

transformations or deck transformations.
 494

R e g u la r C o v e r in g s

11. A covering K is regular if for some point E tl E,,

gr5(x^) is a normal subgroup of tt ^ (bs £ ,pr E, (x^)) . In this case
gr Ux) is a normal subgroup of tt ^ (bs E,,pr E? (x)) for all x G tl E,,
as seen from 2.
If E, is regular, then again using 2, the groups gr ^ (x)
i _
with x G pr ^ (b) are all equal for each fixed b E bs £ . ,
We remark that every two-sheeted covering is regular (because
every subgroup of index 2 is normal). More examples of regular
1 1 1
coverings are (3R,hel,S ), (S ,helm ,S ) [see 4. 1.2.6],

(S2n-1,pr,L(m;£1,...,£n )) [see 4 .2 .3 .15] , and (S3 ,p r ,S3/G~P), where

P is a tetrahedron, a cube, or a dodecahedron [see 4.2.3.17]. An
example of a nonregular covering is given in Fig. 17 (where the two
points marked A are identified, as are the two points marked B);
its nonregularity is a result of the following theorem.

12. A covering F, is regular if and only if there is a po

xQ £ tl F, with the property that given any loop s : I -* tl F, with
s(0) = xQ and any point x £ pr £_1 (pr F, (xQ)) , the path with origin x
and covering pr £ ° s is closed.

Indeed, the last condition means that each element of gr £ ( x Q )

also belongs to every group gr£(x) with x € pr £-1 (pr F, (xQ )) , and
this implies that every subgroup of (bs F,,pr F, (xQ)) which is conjugate
with gr F„ (Xq ) actually equals gr£(xQ ) [see 2].
 495

13. A covering £ is regular if and only if it is equivalent

to a principal bundle. The structure group of such a bundle is discrete
and isomorphic to G = tt 1 (bs £ ,pr £ (xQ ))/gr £ (xQ ). Two Steenrod G-bundles
which are equivalent to £ are themselves G-equivalent.

PROOF. By 4.3.2.10 and 4.3.2.11, £ is equivalent to a

Steenrod G-bundle if and only if there exists a free, continuous right
action of G on tl£ whose orbits are precisely the fibers of Z}
in particular, G is necessarily discrete. Moreover, the transformation
of tl£ determined by such an action yield the automorphisms of £.
Finally, by 9, such an action exists if and only if £ is regular and
G is antiisomorphic to Aut £ , i.e., is isomorphic to the group
TT1 (bs £ ,pr £ (xQ))/gr £ ( X Q ) .

Existence of Coverings

14. A topological space X is said to be semilocally simply

connected if every point x 6 X has a neighborhood U such that
in* ; tt^ (U ,x ) tt^(X,x ) is trivial.
Obviously, the class of semilocally simply connected spaces
contains all simply connected spaces and all locally contractible spaces
[and, among the latter, all locally Euclidean spaces, all CNRS’s (see
1.3.6.8), in particular, all cellular spaces (see 2.1.4.5)].

15. Every space which has a simply connected covering space

is semilocally simply connected.

In fact, let £ be a covering projection with tl £ simply

connected and let b 6 bs £ . Then the homomorphism in*: tt^ (U,b) -*•
TT^(bs£,b) is trivial for every neighborhood U of b such that
£ |^ is trivial.
16. Let B be a connected, locally connected, and semilocally
simply connected space, let bQ € B, and let tt be any subgroup of
tt^ (B,bQ) . Then there exists a covering £ with base pointxQ € tl £
such that bs £ = B, pr £ (xQ) = bQ , and gr £ (xQ) = tt.

PROOF. Consider C(I,0;B,bQ ), identify any two paths s1 's2

in this set whenever s^D = s2 (1 ) and the homotopy class of s1s21
belongs to £, and denote the resulting quotient space by E. Further,
given open subsets U of B and V of U, and any path s: I + B,
with s (0) = b Q , s (1) € V, let Nb(U,V;s) be the subset of E
consisting of the equivalence classes of the pathssw with w(I) c U
and w (1) 6 V. The sets Nb(U,V;s) satisfy the conditions of
 496

Proposition 1.1.1.9, and thus yield a base for a topology on E.

[Information: this topology coincides with the quotient topology
that E inherits as aquotient of the topological spaceC (1 ,0;B,bQ ) .]
The map p: E B which takes each point from E into the common end
of the paths which represent it is clearly continuous and open. We set
E, = (E,p,B) and take for xQ the point of E represented by the
constant path. Then p(xQ ) = bQ . Let us show that K is a covering
and that gr£(xQ ) = tt.
We first verify that K is a covering in the broad sense.
Pick an arbitrary point b £ B and find a neighborhood U of b such
that the inclusion homomorphism tt^ (U,b) -> tt^ (B,b) is trivial, and then
a neighborhood V of b in U such that b can be joined to every
point of V by a path in U. We claim that £|v is a trivial bundle
with discrete fiber.
To see this, consider an arbitrary path s: I ■+ B such that
s(0) = bp, s(1) = b, and for each coset a £ tt^ (B,bQ)/TT choose a
loop uOt : I -* B representing some element ofa . The sets Nb(U,V;u
CL
s)
are open and pairwise disjoint [if the paths (u
cl
s)w and (u s)w
a ^ i
with w(I) cz U, w^ (I) c u, define the same point of E, then a = a ^ :
-1 -1
indeed, the loop ((u s)w)((u s)w ) , and h e j i c e the loop u u
0£> 01 T CL CL ^

homotopic to it, are elements of a class that belongs to tt] . Also,

the sets Nb(U,V;uas) cover p 1 (V) [every path s' with s'(0) = b^,
s ' (1) £ V, is homotopic to a path of the form (us)w, where u is a
loop and w(I) c U: and example of such a path is (((s'w ^ )s ^)s)w,
where w : I -+ U is any path with w(0) = b, w(1) = s 1 (1 )]• Finally,
the maps abp : Nb(U,V;u s) + V are open (because p is) and
invertible [if the paths (u^s)w, (u^sjw^ with w(I) c: u, w^(I) c u,
have the same end, then they are homotopic]; that is, ab p are homeo-
morphisms. Consequently, £ Jv is a trivial bundle with discrete fiber,
as asserted.
The last thing to prove is that E is connected and that a
path in E which has origin x^ and covers a loop from a homotopy
class a £ ir^(B,bQ) is closed if and only if a £ tt. Given s: I B
with s(0) = b^, let s : I -> E be the path which takes each t £ I
into the point of Erepresented by the path t *+ s(tx). Clearly,
s (0) = X q , s (1) is the point of E represented by s, and s~
covers s. This hastwo consequences: the first is that every point of
E can be joined to xQ by a path, and the second that given a loop
with origin bQ , the end of the path which starts at xQ and covers
it is precisely the point of E represented by the given loop. The
first shows that E is connected, while the second implies that, given
 497

a loop with origin b^, the path which covers it and has origin b^
ends at Xq if and only if the homotopy class of the given loop belongs
tO 7T.
17. The previous theorem completes the theory of coverings
with a fixed base. Combined with Theorem 7, it establishes a one-to-one
correspondence between the equivalence classes of coverings over a
connected, locally connected, and semilocally simply connected space B
with base point b^ and the classes of conjugate subgroups of tt^ (B,b^).
It transforms the hierarchy of coverings into the usual, set-theoretic
hierarchy of subgroups, and the normal subgroups correspond to the
regular coverings. The covering corresponding to the trivial subgroup
has a simply connected total space. Since every covering is subordinate
to this one, it is called universal. (Warning: do not confuse this
universality with the notion of universality defined in the theory of
Steenrod bundles. Note, however, that among the universal Steenrod
bundles one finds also universal coverings, namely the universal
principal bundles with discrete structure groups; see 1.5 and cf. 18).

An Application:
Classifying Spaces of Discrete Groups

18. In order that a connected topological space X be a

classifying space of a given discrete group tt, it is necessary that
the groups tt^(X) be trivial for all r ^ 2 and that tt^ ( X) be
isomorphic to tt . Ijf X is locally connected and semilocally simply
connected, then this condition is also sufficient.
Necessity results from 1.5, 13, and 1.9.15. Sufficiency
results from 16, 1.8.12, and 1.5.

Covering Maps

19. Let E, and E,1 be coverings with base points xQ E tl E,

and x q € tl£', and let f: (bs £ ',pr £ ’ (x^) ) (bs £ ,pr £ (xQ )) be a
continuous map. Then the inclusion f * (gr £ ' (x^)) c gr £ (Xq ) is a
necessary condition for the existence of a continuous map <j>:£'-»- £
such that bs <t> = f and tl(j)(x^) = xQ . If bs £' is locally
connected, then this condition is also sufficient. If such a <t> exists,
then it is unique.
PROOF. Assume that such a c(> exists. Then from the
 498

commutative diagram (see 1.8.6)

tl Hj (tl Ç,XQ ]
7T1 (tl Ç ' ,X Q ]

pr Ç*

bs ^ (bs Ç,pr Ç (xQ ]

TT1 (bs Ç ',pr Ç' (x¿)

it follows that f* (gr Ç ' (x^)) c gr Ç (xQ) . Now let us prove the convers
Consider the bundle f ‘Ç together with the map ad f : f 'Ç -* Ç. Let
y q £ (pr f !Ç) _1 (pr Ç ' (x^) ) be such that tladf(y^) = xQ . Further, let
Y' be the component of tl(fÇ) containing y^, and let p' be the
restriction of pr fX to Y'. Obviously, (Y',p',bs^) is a coverin
tl ad f and f combine to define a map (Y',p',bsÇ') Ç, and
tl ad f I is injective on every fiber of (Y',p',bsÇ') . In the diagra
Y'
pJl ^, .... ^ > A ,, .,-1 (pr
. Ç' ^¿)) ,y^)
(Y '/Yq V<P'

ab tl ad fj ab tl ad f

.-1
(tl Ç,xQ) Pr £*■>tt 1 (bs ^,pr (xQ) ) îTq (pr Ç (pr Ç (xQ) ) / xQ]

the rows are exact, while the left vertical homomorphism is monomorphic.
i _
Consequently, f* (Im(pr£*)) = Im p* , whence Imp* gr F,1 (x^) . Sine
bs£' is locally connected, the last inclusion shows that the covering
(Y',p',bs£’) with base point y^ is subordinate to the covering
with base point x^. If is such a subordination, then
(tl adf L.) tl ip' and f combine to define the desired map <t>.
Finally, we claim that <p is unique. Indeed, suppose
(py. £ is another map with bs = f and tl^fx^) = x^. Then
for every point x ' £ tl £' and every path s': I -> tl £' with s ’ (0]
= Xq , s'O) = x ', the paths tl s and tl (p cover
f o pr o s' and have the same origin x^. Therefore, in these
circumstances tl <j> ° s * = tl ^ ° s', and hence tl <j>(x ') = t l <p ° s ' (1
= tl s ' (1 ) = tl c()1 (x' )
1
20. Suppose
is a covering with base point £
e t u , x^
Y is a locally connected space with base point y^, and f (Y,y0)
+ (bs £,pr £ (xQ)) is continuous. If f* (tt 1 (Y,yQ)) c gr U x Q) [in
particular, if Y is simply connected], then there is a map
F: (Y,yn) (tl£,xn) which covers f.

(Cf. 4.1 .3.8.)

 499

To see this, apply Theorem 19 to £ and = (Y ,id Y ,Y) .

Coverings and Additional Structures

21. There is an important general principle asserting that

under favorable conditions an additional structure defined on the base
of a given covering can be lifted to the covering space. To conclude
our study of coverings, we apply this principle to three structures:
differentiable, cellular, and simplicial. Further applications appear
in Exercises 5.10 and 5.11.
Concerning differentiable structures, we restrict ourselves
from the onset to manifolds, i.e., we assume that bs£ is a Cr-manifold
(1 £ r £ a) and that the number of sheets of E, is countable. We know
that every chart ch £ Atl bs £ such that the bundle EI is trivial
S |supp4)
has a family of copies in tl E, which cover pr £ (supp(j)). Obviously,
these copies (taken for all £ Atlbs£) constitute a Cr-atlas of the
space tl E, . This atlas provides tl E, with a C -structure and turns
it into a C -manifold. The fundamental property of this lifted structure,
and the one which defines it uniquely, is that relative to this structure
r r
pr E, is a C -submersion, i.e., that E, is a C -bundle. Let us add that
the orientability of bs E, implies the orientability of tl E, .
We can similarly lift a cellular structure. However, to lift
the cells and their characteristic maps we need Theorem 20. As a result,
the total space of the given covering with cellular base becomes a
cellular space (rigged whenever the base is rigged), and the projection
becomes a cellular map.
Lifting a simplicial structure may be viewed as a special case
of lifting a cellular one. The total space of a covering with simplicial
base thus becomes a simplicial space (ordered whenever the base is so),
and the projection becomes a simplicial map.

22. In all the three cases considered above, the lifted

structure is invariant under the automorphism of the given covering.
It is clear, at least if the covering E, is regular, that every
differentiable or cellular structure defined on tl E, and invariant
under the automorphisms of 5 can be lowered to bs E, , i.e., the given
structure on tl £ is the result of lifting a similar structure from
bs E, (this becomes true also for simplicial structures after we effect
the barycentric subdivision twice). For example, the lenses and the
quotient spaces of S by the binary tetrahedral, cube, or icosahedral
groups (see 4.2.3.17) are regularly covered by the corresponding spheres
 500

(see 4.3.2.11 and 13), and hence are Ca-manifolds.

If the given covering K is a C -bundle, then the
Cr-structure on tl£ is automatically invariant under the automorphisms
of In this case, the lowered Cr-structure surely coincides with the
original Cr-structure on bs £ .
This is exactly what happens to all the
1
coverings described in 4.1.2.6 and, in particular, to (IR,hel,S ),
(S1,helm ,S1), and (Sn ,pr,IRPn ) .

3. Orientations

1. In the present subsection the technique developed in the

previous one is applied to the orientability problems considered in
§§ 3.1 and 4.5 (see Subsections 3.1.3, 4.5.1, 4.5.4). We follow the
recipe of Subsection 4.5.1, which applies to bundles over arbitrary
spaces, in constrast to the recipe of Subsection 4.5.4, which yields
the same results, but only for bundles over a cellular base.

2. Let £ be an n-dimensional real vector bundle. Conside

the associated principal bundle, asso(£,GL(n,HR) ) , and construct the
orbit space of the right action of GL+ (n,3R) on tl asso (C,GL(n,IR))
obtained by the restriction to GL+ (n,IR) of the canonical right action
of GL (n,HR) on tl asso (£ ,GL (n ,1R) ) [see 4.3.2.10]. Denote this orbit
space by bs+£, and observe that it is the totalspace of a bundle with
base bs £ and projection fact pr asso(C,GL(n,]R) ) . It is clear that
(bs + £,fact pr asso(£,GL(n,1R) ) ,bs£ ) is a two-sheeted covering in the
broad sense; we denote it by or ^ .
One may view the points of bs+£ as pairs (x,e), where
— 1
x £ bs £ and e is an orientation of the fiber pr £ (x) . This
permits us to look upon a simultaneous orientation of all the fibers of
£ as a map s: bs C + bs+£ such that pr or£ ° s = id bs £ . Clearly,
the compatibility (in the sense of 4.5.1.8) of the orientation given by
s on the fibers of £ is equivalent to the continuity of s. Thus,
the orientations of the bundle £ turn out to be the sections of the
bundle or £ •
We list the most important corollaries of the above discussion:

(i) a bundle £ is orientable if and only if the bundle or £

is trivial;
(ii) if the base bs £ is locally connected and the
fundamental groups of its components have no subgroups of index 2
(the last happens, in particular, whenever the components of bs £ are
simply connected), then £ is orientable.
 501

(iii) if ç is orientable, then the number of its orient

tions equals the number of maps comp bs Ç -+ S° (and, in particular,
equals 2 m if bs ç has m < co components).

3. We would like to make another remark concerning the

previous construction. Given a real vector bundle Ç, consider the
bundle with base bs +Ç, induced from Ç by the projection pr or £.
Here we note that possesses a canonical orientation. Namely, the
orientation of its fiber pr ç” 1 (x,e) (where x £ bs Ç and e is an
-1
orientation of the fiber pr ^ (x)) is simply the orientation e,
1 -
transferred to pr (x,e) by the isomorphism

ab tl ad (pr or Ç ) : p r Ç ~ 1 (x,e) -> pr Ç_1 (x) ;

that these orientations are compatible in the sense of 4.5.1.8 is plain.

The Case of Smooth Manifolds

4. Orienting a smooth manifold is the same as orienting its

tangent bundle (see 4.6.4.4), and hence the discussion in 2 carries over
to the orientability and orientations of smooth manifolds. In
particular, a manifold X is orientable if and only if the bundle
or tang X (which is a two-sheeted covering in the broad sense with base
X) is trivial. Also, a smooth manifold is automatically orientable if
the fundamental groups of its components have no subgroups of index 2
(as happens when all its components are simply connected).
Let us add that, according to 2.2.1, the space tl or tang X
is a smooth manifold. By 3, the latter carries a canonical orientation.
If X is connected and nonorientable, then or tang X is a (two-sheeted)
covering; we call it the orientation-covering of the manifold X.

4. Some Bundles Over Spheres

1. In this subsection we present the most elementary results

concerning orientable real vector bundles and complex vector bundles
over lower-dimensional spheres. This topic has independent interest
and, at the same time, illustrates the general theory.
Recall that the classes of GL+HRn -equivalent GL+3Rn -bundles
over a given cellular base are in a one-to-one correspondence with the
homotopy classes of continuous maps of this base into G (°°,n) .
Similarly, the classes of GL(Cn-equivalent GL(En-bundles over a cellular
 502

base are in a one-to-one correspondence with the homotopy classes of

continuous maps of this base into (EG(°°,n). [See 4.5.3.7.] If the
base is Sr , r ^ 1, then the fact that G+ (°°,n) and (CGi^n) are
simple spaces implies that the aforementioned homotopy classes may be
thought of as elements of the groups tt^ (G+ (°°,n) ) and tt^ ((EG (°°,n )) ,
respectively. Since these groups are isomorphic to tt^ ^ (SO (n) ) and
TTr_^(U(n)) (see 2.8.2 and 2.8.4), it follows that the classes of

GL+]Rn -equivalent GL+]Rn -bundles over Sr (r ^ 1) are in a natural

one-to-one correspondence with the elements of tt^_ (SO(n)), while the
classes of GL(Cn-equivalent GL(Cn-bundles over Sr (r ^ 1) are i_n a natural one
to - one correspondence with the elements of tt^_^ (U (n) ) . Since we
already know Trr_1 (SO(n)) and 7Tr_-| (U (n))for some small values of r
(see Subsection 2.6), we get the classification of the corresponding
bundles, which we shall outline with minimum of details below. Some
supplements appear in Exercises 5.14-5.19.
We note that this method works in a considerably more general
situation. Namely, let G be a topological group, and let F be an
effective G-space. According to the general theory of bundles (see
Subsection 4.4.2), the elements of Stee(S^,F) are in a natural one-to-
-one correspondence with the elements of tt(S^,X), where X is any
classifying space of G. If X is q-simple (for example, if G is
connected), then 7r(S^,X) coincides with tt(X) and, for q ^ 1, with
77 - (G) (see 1.6).
Si '

The Real Oriented Case

2. Since 7T^(SO(n)) is trivial, every GL+]Rn-bundle over S^

is GL+3Rn -trivial, for any positive integer n.
Since tt1 (SO (1) ) is trivial, tt^(SO (2) ) = 7L , and
(SO (n) )= 7L2 for all n ^ 3, it follows that: every GL+3R1-bundle
2 1 2
over S is GL+3R -trivial; the pairwise GL+3R -nonequivalent
2 2
GL+]R -bundles over S form an infinite and countable collection;

for n ^ 3, the number of pairwise GL ]Rn -nonequivalent GL 3Rn -bundles

2
over S is 2.
Since TT2 (SO(n)) is trivial, every GL +nRn -bundle over S3
is GL+]Rn -trivial,for any positive integer n.
Since tt 3 (SO (1 )) and tt ^ (SO (2 )) are trivial, whereas

TT3 (SO(n)), n > 3, is infinite and countable, it follows that: every

 503

GL +3R1 -bundle over S4 is GL+3R1 -trivial; every GL+® 2 -bundle over S4

2
is GL+3R -trivial; for n is 3, the pairwise GL+3R -nonequivalent
n 4
GL+HR -bundles over S form an infinite and countable collection.

The Complex Case

3. Since TT^(U(n)) is trivial, every GL(En-bundle over S^

is GL(Cn-trivial, for any positive integer n.
Since TT^(U(n)) = 2Z for all n ^ 1, the classes of pairwise

GL(Cn-noneguivalent GL(Cn-bundles over S2 form an infinite and countable

collection, for any positive integer n.
Since TT^iUin))is trivial, every GL(Cn-bundle over S^ is
GLCCn-trivial, for any positive integer n.
Since t^(U(1)) is trivial andTT^(U(n)) = ZZ for all n ^ 2,
1 4 1
it follows that: every GLCC -bundle overSis GLCC -trivial; if n ^ 2,

the pairwise GL(Cn -nonequivalent GL(Cn-bundles over S^ form an infinite

and countable collection.

5. Exercises

1. Show that a smooth two-dimensional manifold does not admit

2
the disc D as a covering space (unless the given manifold is homeo-
2
morphic to D , and then the covering projection is the corresponding
homeomorphism).

2. Show that a smooth, compact two-dimensional manifold does

2
not admit the half plane JR_ as a covering space.

3. Show that a sphere with g handles admits asphere with

g~ handles as an m-sheeted covering space if and only if g~ - 1 =
= m(g - 1) . [Cf. 4 .1.5 .1.]

4. Show that a sphere with h crosscaps admits a sphere with

h crosscaps as an m-sheeted covering space if and only if h - 2 =
= m(h-2) . [Cf. 4.1.5.2.]

5. Show that a sphere with h crosscaps admits a sphere with

g handles as a covering space if and only if m is even and g - 1 =
= m(h - 2)/2.
6. Show that the Klein bottle admits a topological space as
 504

a covering space with nonidentical covering projection if and only if

the given space is homeomorphic to the Klein bottle, to the interior of
1 1 1
a Mobius strip, or to one of the products IR x 1R , I R x S , s x S .

7. Let TT-i'TT2'TT3' *** be arbitrary groups, with tt^ Abelian

for k ^ 2, and suppose that there is given a right group action of tt^
on tt, r k ^ 2. Show that there is a cellular space X with base point
Xq together with an isomorphism f ^ : tt^ -> tt^(X,Xq), such that the
group tt^ (X,Xq ) is f-isomorphic to tt^ for all k ^ 2.
8. Let 5 be a covering with bs £ an n-dimensionallocally
Euclidean space. Prove that tl £ is an n-dimensional locally Euclidean
_ -j
space with boundary 3tl £ = pr £ (Bbs FJ .

9. Show that every covering space of a locally finite cellular

space is locally finite (see 2.21).

10. Let £ be
a covering with bs 5a topological group.
1 -

Prove that, given any point x € pr £ ^ebs ' °n ^ ^ there is one

and only one group structure, which turns tl £ into a topological group
with identity x, and turns pr £ into a homomorphism.

11. Let £ be a covering with a finite number of sheets,

such that a connected, compact topological group acts transitively on
bs £ - Prove that one can define a transitive action of a connected,
compact topological group on tl£ .

12. Let 5 be a two-sheeted covering with bs £ a

nonorientable smooth manifold and tl £ orientable. Show that £ is
equivalent to the orientation-covering of bs £ •

13. Let X be a nonorientable smooth manifold and assume

that X admits an orientable manifold Y as a covering space. Show
that Y is also a covering space of tl or tang X.

14. Show that:

(i) for each positive integer n there is a unique (up to
GLlRn -equivalences) GLlRn -nontrivial GL3Rn -bundle over S1 and that
for n = 1 its total space is homeomorphic to the interior of a Mobius
strip;
2 2 2
(ii) the pairwise GL3R -nonequivalent GL3R -bundles over S
form an infinite and countable collection;
(iii) for n s 3, the number of pairwise GLlRn -nonequivalent
GL]Rn -bundles over S2 is equal to 2;
(iv) for any positive integer n, everyGL]Rn -bundle over
3 . n
S is GL3R -trivial;
2 4 2
(v) every GL3R -bundle over S is G L K -trivial;
 505

(vi) for n ^ 3, thepairwise GL]Rn -nonequivalent

n 4
GL3R -bundles over S form an infinite and countable collection.
1 r
15. Show that for r ^ 2 every GL3R -bundle over S is
1
GLIR -trivial.

16. Let r ^ 3. Show that:

2 r 2
(i) every GL]R -bundle over S is GL]R -trivial;
2 r 2
(ii) every GL+IR -bundle over S is GL+IR -trivial;
1 r 1
(iii) every GLE -bundle over S is GLE -trivial.
2 2
17. Show that two GL+]R -bundles over S become equivalent
by extending the structure group GL+ (2,]R) to GL(2/IR) if and only
if the corresponding elements of tt^(SO(2)) [see 4.1] are either equal
or inverses of one another.
2
18. Show that for every nontrivial OJR -bundle £ with
2 1
bs £ = S the space tlasso(£,S ) is homeomorphic to oneof the
lenses L [ h e r e 0(2) acts canonically on S ].

19.
Show that for n ^ 2 the number of pairwise GL(Cn -non-
n 5 -
equivalent GL(E -bundles over S does not exceed 2.

INFORMATION. Actually, the last number is 2 if n= 2,

and 1 if n > 2.
 Bibliography

[1] Adams, J. F.: On the non-existence of elements of Hopf invariant

one, Ann. Math., II. Ser. 7_2, No. 1 , 20-104 (1960).

[2] Adams, J. F., Atiyah, M. F.: K-theory and the Hopf invariant,
0. J. Math., Oxf. II. Ser. V7, 31-38 (1966).

[3] Bourbaki, N.: Topologie Generale, Chap. I/II, Hermann, Paris,

1947; English Transl.: General Topology, Addison-Wesley, Reading,
Mass., 1966.

[4] Brown, M.: Locally flat imbeddings of topological manifolds,

Ann. Math., II. Ser. 7J5, No. 2 , 331-341 (1962).

[5] Calabi, E., Rosenlicht, M.: Complex-analytic manifolds without

countable base, Proc. Am. Math. Soc. _4, 335-340 (1 954).

[6] Dowker, C. H.: Topology of metric complexes, Am. J. Math. 74 ,

555-577 (1952).

[7] Fuks, D. B.: Homotopic topology, Itogi Nauki Tekh., Ser. Algebra
Topologiya Geom., 71-122 (1969); English. Transl.: J. Sov. Math.
1, No. 3, 333-362 (1 973) .

[8] Grauert, H.: On Levi1s problem and the imbedding of real analytic
manifolds, Ann. Math., II. Ser. 6J3, 460-472 (1958).

[9] Hodge, W. V. D., Pedoe, D.: Methods of Algebraic Geometry, Vol. I,

Cambridge Univ. Press, 1947.

[10] Kämpen, E. R., van: Komplexe in Euklidischen Raumen, Abh. Math.

Semin. Univ. Hamb. 9_, 72-78 and 152-1 53 (1 932).

[11] Kelley, J. L.: General Topology, Van Nostrand, New York, 1955;
Reprint Graduate Texts in Math., Springer-Verlag, New York,
Heidelberg, Berlin, 1975.

[12] Kervaire, M.: A manifold which does not admit any differentiable
structure, Comment. Math. Helv. 3_4, 257-270 (1960).

[13] Kuratowski, C.: Topologie, Vol. I, Monografie Matematyczne,

Warszawa, 1948; English Transl.: Topology, Vol. I, Academic Press,
New York, 1966.

[14] Kuratowski, C.: Topologie, Vol. II, Monografie Matematyczne,

Warszawa, 1950; English Transl.: Topology, Vol. II, Academic
Press, New York, 1968.

[15] Milnor, J . : On manifolds homeomorphic to the 7-sphere, Ann. Math.,

II. Ser. 64, No. 2, 399-405 (1956).
 507

[16] Milnor, J.: On spaces having the homotopy type of aCW-complex,

Trans. Am. Math. Soc. 9JD, No. 2, 272-280 (1959).

[17] Milnor, J. : Morse Theory, Princeton Univ. Press, Princeton, N.J.,

1963.

[18] Rokhlin, V. A.: The imbedding of non-orientable three-dimensional

manifolds into a five-dimensional Euclidean space, Dokl. Akad.
Nauk SSSR 160, No. 3, 549-551; English Transl.: Sov. Math. Dokl.
6, No. 1, 153-156 (1965).

[19] Serre, J-P. : Groupes d'homotopie et classes de groupes abeliens,

Ann. Math.,II. Ser. 5 8 , No. 2, 258-294 (1953).

[20] Smale, S.: A survey of some recent developments in differential

topology, Bull. Am. Math. Soc. 69_, No. 2 , 131-1 45 (1963).

[21] Sternberg, S.: Lectures on Differential Geometry, Prentice Hall,

Englewood Cliffs, N. J., 1964.

[22] Whitney, H.: Differentiable mnaifolds, Ann. Math., II. Ser.

37, 645-680 (1936).

[23] Whitney, H . : A function non constant on a connected set of

critical points, Duke Math. J. J_, 514-517 (1935).
 Index

Abridgment of map 2 Binary dodecahedral or

Action 285 icosahedral group 295
Canonical — 287 — tetrahedral group 295
Conjugate — s 286 Borsuk pair 66
Continuous — 288 Boundary 6, 133, 143
Effective — 285 — of spheroid 418
Free — 290 — point 6, 133, 143
Group- — 287 Bounded set 9
Induced — 285 Bouquet 50
Left — 286 Bundle 26 5
Left canonical — 287 Adjoint — 275
Left inner — 287 Associated — 331
Right — 286 Complex vector — 328, 354
Right canonical — 287 Complexification of — 335
Right inner — 28 7 Conjugate — 328
Topologically effective — 309 Ehresmann-Feldbau — 308
Transitive — 285 Euclidean — 325, 354
Adherent point 6 Grassman — 340
Adjoint bundle 275 Hermitian — 328, 354
— homomorphism of action 285 Hopf - 349
— map 266 Induced — 266
Analytic manifold 149 к-universai — 318
— map 145 Locally (topologically)
Approximation, cellular 123, 472 trivial — 26 7
Simplicial — 118, 130 Milnor — 319
Associated bundle 305 Neatly induced — 347
Weakly — - 307 Normal — 36 2
— section 267, 307 — of class С — 346, 354
Atlas 142 Orientable — 328
Complete — 143 Oriented Euclidean — 326
Holomorphic — 168 Oriented vector — 326
Attaching 3 7 Principal — 303
— map 85 Product — 26 7
Quotient — 332
Real vector — 324
Baire space 219 Serre — 269
Ball 8 Smooth — 34 7
n-dimensional — 9 Standard trivial — 267
Open — 8 Steenrod — 300
Barycentric coordinates 100 Tangent — 360
— star 117 Trivial (topologically) — 267
— subdivision 108 Universal — 316
Base 6, 265 Weakly associated — 307
Countable — 20 Bundles, equivalent 266
— at point 7 Isomorphic — 266
— of cone 4 2
— — join 42
— — mapping cylinder 45, 113 Canonical simple local
suspension 4 2 system 379
— point 49 Cell 81
Biholomorphic map 166, 168 Closed — 81
Binary cube or octahedral Cellular approximation — 123,472
group 29 5 — decomposition 81
 509

Cellular decomposition, Complex structure 166

equipped 81 — vector bundle 328
Locally finite — — 83 — vector Cr-bundle354
Rigged — — 81 Complexification of bundle 335
— embedding 86 Component of identity 277
— equivalence 86 Component of topological space 62
Inductively glued — space 87 Compression of map 2
— join 97, 98 Condition, Serre 269
— map 86 Strong Serre — 273
— pair 84 Cone 42, 51
— product 94 Mapping — 46
— space 82 Conjugate bundle 328
— tensor product 98 Constant homotopy 56
— topology 82 Continuous action 288
— triad 84 — map 11
— triple 84 Map — at point 12
— weakening of topology 83 Contractible space 59
Center of bouquet (wedge) 50 Coordinates, barycentric 100
— — Euclidean simplex 100 Grassman-Plucker — 182
~ — join (of pointed spaces) 52 Homogeneous — 39, 40
— — tensor product 51 Local — 143
Characteristic function 120 Correcting map 26 7
— homeomorphism 101 Countability axioms 19
— map 81 Cover 7
Simplicial — — 103 Closed — 8
Total 82 Fundamental — 1 1
Chart 142 Locally finite — 8
Complex — 167 Open — 8
Support of — 142 Refinement of — 8
Class, A-homotopy 57 Covering 26 8
Homotopy — 57 Group of — 490
Classifying map 322 — in broad sense 268
— space 318 — in narrow sense 268
Closed cell 81 Orientation 501
— cover 8 — projection 268
— manifold 137, 149 Regular — 493
— map 12 — space 268
— partition 33 Subordinate — 494
— path 60 Universal — 497
— set 5 Critical point 164, 194
Closure 6 Index of — — 194, 240
Cobordism 239 Nondegenerate — — 194, 2 40
Elementary — 247 — value 164
Standard — — 243 Cube 10
Standard trivial — 243 Binary — group 295
Trivial - 247 — group 295
Codimension 147 Unit — 11
Collar 209 Cylinder 27
Collaring 209 Mapping — 45
Two-sided — 210 Simplicial — — 113
CNRS-space 70
Compact-open topology 46
— space 21 Decomposition, cellular 81
Complement, orthogonal Equipped — — 81
(of subbundle) 332 Locally finite --- 83
Complete subspace 104 Rigged — — 81
Complex chart 167 Deformation retract 59
— Grassman manifold 182 Strong — — 59
space 337 — retraction 59
— manifold 168 Strong — — 59
— projective space 39, 41 Degree of map 36 4
— Stiefel manifold 174 — — — at a point 365
 510

Dense set 6 Face 99

Everywhere — — 6 Opposite — 99
Diagonal 4 Factor, effective 285
— map 4 — group 27 9
Diameter 9 Injective — 3
Diffeomorphism 142, 146 Factorial map 31
Orientation preserving — 153 Fiber 28, 265
Orientation reversing — 153 — homotopy 403
Differentiable embedding 161 — preserving map 26 6
— manifold 149 Standard — 299
— structure 141 Fibered map 266
Differential 158 Field, vector 160, 330
Dimension of cell 81 — of frames 330
cellular space 83 Zero vector — 330
Euclidean simplex 99 Filtration 37
Direct product of First countable space 19
topological groups 280 Frames, field of 330
Disc with inverted handle 257 Free action 290
Discrete topology 5 — homotopy 58, 380
Distance 8 Frontier 6
— between sets 15 — point 6
Distinguishable set 15 Function, characteristic 120
Dodecahedral or Morse — 240
icosahedral group 295 Proper — — 2 40
Binary — — ----- 295 Standard Morse — 24 3
Double 137, 210 Urysohn — 17
— torus 256 Functions independent at point 147
Doubling 210 Fundamental cover 11
— group 37 5
— spheroid 416
Effective action 285
— factor (of action) 285
— G-space 288 General linear group 284
Ehresmann-Feldbau bundle 308 Generatrix of cone 42, 51
Elementary cobordism 247 join 42
Standard 24 3 — — mapping cylinder 45
Embedding, cellular 86 — — suspension 4 2
Differentiable — 161 Glued cellular space, inductively 87
Holomorphic — 169 Glueing 34, 209
Neat — 161 Grassman bundle 340
Topological — 14 — manifold 177
Equivalence, cellular — 86 Complex — — 182
Homotopy — 58 Noncompact — — 185
— of bundles 266 Quaternionic ---- 183
— — local systems 379 Upper — — 178
Stable — of bundles 335 — space 337
Weak homotopy — 463, 467, 474 Complex — — 337
Euclidean bundle 325 Upper — — 337
Oriented — — 326 Grassman-Pliicker coordinates 182
— C -bundle 354 Group, cube or octahedral 29 5
Locally — space 133 Binary — — — — 295
— metric (in vector bundle) 326 Binary dodecahedral or
Ordered — simplex 100 icosahedral — 295
— simplex 99 Dodecahedral or icosahedral — 295
Exact sequence 390 Factor — 37 5
— 7T-sequence 3 95 General linear — 284
Extension of group 286, 299, 301 Homotopy — 37 5
Exterior point 6 — of covering 490
Orthogonal — 284
Relative homotopy — 3 84
— space 276
Special orthogonal — 284
 511

Group, stable homotopy 429, 440 Index of spherical modification 254

Structure — 299 — — standard elementary
Symplectic — 284 cobordism 243
Tetrahedral - 29 5 — — attached handle 252
Topological — 276 Induced action 285
Unitary — 284 — bundle 266
Groups, local system of 378 — homomorphism 385, 377
— local system 379
Inductively glued cellular space 87
Handle 256 Injective factor (quotient) 3
Attaching of — 252 Inner action 287
Hausdorff space 16 Interior 6, 133, 143
Hermitian bundle 328 — part 6, 133, 143
— Cr-bundle 354 — point 6, 133, 143
Holomorphic atlas 168 Invariant set 286
— embedding 169 — subgroup 27 7
— map 166, 168 Inverse homotopy (of another
— structure 364 homotopy) 57
Homogeneous coordinates 39, 40 Homotopy — 58, 413
— space 290 — path 61
Homeomorphism 13 — spheroid 37 3
Characteristic — 101 Isomorphism 105, 298, 379
Marked — of F-structure 297 — of bundles 266
Homomorphism adjoint to action 285 — — topological groups 279
Boundary — 385 Isotropy subgroup 288
— induced by continuous map 377,385
— of local systems 379 Join 42, 52
— — sequences of groups Cellular — 97
and homomorphisms 39 0 Simplicial — 110
— — topological groups 279
Homotopies, product of 57
Homotopy 5 6 Kernel, noneffectiveness 285
— class 57 Klein bottle 257
— commutative H-space 413
Constant — 56
— equivalence 58 Left action 286
Fiber — 403 — canonical action 287
Free - 58, 380 — inner action 287
— group 3 75 Lens (space) 295
— inverse (of map) 58 Lift(ing) 270
— inverse 413 Limit of sequence 36
Inverse — 57 Line, projective Cayley 4 0
Rectilinear — 57 Linear group, general 284
Relative — group 384 — map (of vector bundles) 329
— sequence 393 Link 1 1 4
of bundle 407 Barycentric — 117
----- pair 398 Local coordinates 143
— — — principal bundle 4 1 5 — representative of map 145
----- triple 402 Local system (of groups) 3 78
— , stationary 57 — Canonical simple — — 379
— type 58 Induced — — 379
Weak — equivalence 463, 467, 474 — — of homotopy groups 381, 386
Hopf bundle 34 9 Lower — — — — — of fibers 405
— map 41 — ------— — — total space 415
Simple — — 379
Upper — — of homotopy groups
Identity (homotopy) 413 of fibers 404
Component of — 27 7 Local systems, homomorphism of 379
Immersion 34, 162 Locally compact space 25
Inclusion (of bundles) 266 — connected space 66
Index of critical point 194, 240 — contractible space 65
— — elementary cobordism 24 7 — Euclidean space 133
 512

Locally finite cellular space 83 Mapping cylinder 45

— — cover 8 Simplicial — — 113
— trivial bundle 267 Marked homeomorphism (of
Loop 60 F-structure) 297
Standard — 445 Metric 8
Euclidean (in vector bundle) — 326
Hermitian — 328
Manifold 137 — neighborhood 9
Analytic — 149 Riemannian — 361
Closed — 137, 149 — space 8
Complex — 168 — topology 9
Complex Grassman — 182 Metrizable space 9
Complex Stiefel — 174 Milnor G-bundle 319
Differentiable — 149 Möbius strip 255
Grassman — 177 Model surfaces 255
Noncompact Grassman — 185 Modification, spherical 254
Noncompact Stiefel — 175 Monotone simplicial map 104
— of class Cr 149 Morse function 240
— — complex origin 170 Proper — — 240
Open — 137 Standard — — 240
Orientable — 151 Multiplication (in H-space) 413
Oriented — 151 Multiplicity of covering 268
Parallelizable — 160
Quaternionic Grassman — 183
Ouaternionic Stiefel — 175 Neat embedding 161
Riemannian — 361 — neighborhood 120
Smooth — 149 — subspace 147
Stably parallelizable — 360 — tube 203, 351
Stiefel — 172 Neatly induced bundle 347
Topological — 137 Neighborhood 6
Total — of tangent vectors 158 '"Metric — 9
Upper Grassman — 178 Neat - 120
Map, adjoint 266 Regular — 117
Analytic — 145 — retract 70
Attaching — 85 Tubular - 203, 350
Biholomorphic — 166, 168 Noncompact Grassman manifold 185
Cellular — 86 — Stiefel manifold 175
Characteristic — 81 Noneffectiveness kernel 285
Closed — 12 Normal bundle 36 2
Continuous — 11 — divisor 277
— continuous at point 12 — space 16
Correcting — 267 — splitting 396
Diagonal — 4 — subgroup 2 77
Factorial — 31 — transversalization 202
Fibered — 266 — £>transversalization 350
Fiber preserving — 266 Nowhere dense set 6
Holomorphic — 166, 168 Number of sheets of covering 268
Hopf - 41
Linear — (of vector bundles) 3 29
Monotone simplicial — 104 Open ball 8
— of bundles 265 — cover 8
class Cr 144, 145 — manifold 137
Open — 12 — map 12
Quotient — 31 — partition 33
Simplicial — 100, 104 — set 5
Simplicial characteristic — 103 — star 113
Smooth — 14 5 — tube 20 3
Tangential — 361 Orbit 285
Transverse — 222, 225 — space 288
Mapping cone 4b Ordered Euclidean simplex 100
— simplicial scheme 106
 513

Ordered simplicial space 102 Projection 2, 3, 265

— topological simplex 100 Covering — 268
Orientable manifold 151 Projective, Cayley — line 40
Orientation 151, 327 Cayley — plane
Orientation-covering 501 Complex — space 39, 41
— preserving diffeomorphism 153 Quaternionic — space 40, 47
— reversing diffeomorphism 153 Real — space 39, 41
Oriented Euclidean bundle 326 Prolongation of group 286, 299, 301
— manifold 151
— vector bundle 326
Orthogonal complement Quaternionic Grassman manifold 183
(of subbundle) 332 — projective space 40, 47
— group 28 4 — Stiefel manifold 175
Quotient bundle 332
— group 279
Pair, Borsuk 66 Injective — 3
Cellular — 84 — map 31
k-connected — 6 5 — space 31
r-simple — 386 — topology 31
Simple — 386
Topological — 10
Paracompact space 26 Real projective space 39, 41
Parallelizable manifold 160 — vector bundle 324
Part, interior 6, 133, 143 Cr-bundle 354
Partition, closed 33 Realification 329
Open — 33 Rectilinear homotopy 5 7
Pasting 34 Reduction of group 286, 321
Path 6 0 Refinement of cover 8
Closed — 60 Regular covering 493
End of — 60 — neighborhood 117
Inverse — 61 — space 16
Origin of — 60 Relative homotopy group 384
Paths, product of 61 — topology 10
Point 5 Restriction of bundle 265
Adherent — 6 — — group 286
Base — 49 Retract 14
Boundary “ 6, 133, 143 Deformation — 59
Critical - 164, 194 Neighborhood — 70
Exterior — 6 Strong deformation — 59
Frontier — 6 Retraction 14
Interior — 6,133, 143 Deformation — 59
Nondegenerate critical — 194, 240 Strong 59
Polyhedron 106 Riemannian manifold 361
Prebase 7 — metric 361
— at point 7 Rigged cellular decomposition 81
Pretzel 256 — -equivalent cellular spaces 86
Principal bundle 303 Rigging 81
Product, cellular 94 Right action 286
Direct — (of topological Canonical 287
groups) 280 — G-space 290
— of actions 286 — inner action 287
— — bundles 26 5
— — homotopies 5 7
— — maps 4 Saturated set 2
— — orientations 154 Saturation 3
— — paths 61 Scheme of map 105
— — spheroids 373 — — space 105
topological spaces 27 Ordered simplicial- — 106
Simplicial — 110 Simplicial — 105
Tensor — 50 Section 265
Tensor cellular — 98 Associated — 267, 307
Whitehead — 482 Homotopic — s 26 5
 514

Semilocally simply connected Space, Grassman 337

space 4 95 Group — 276
Separable space 19 Hausdorff — 16
Separation axioms 15 Homeomorphic — s 13
Sequence, exact 390 Homogeneous — 2 90
Homotopy — 39 3 Homotopy associative H— 413
— — of bundle 407 Homotopy commutative H— 4 13
----- pair 398 Homotopy fit — 466
— --- principal bundle 415 k-connected — 64
----- triple 402 Locally compact — 25
Limit of — 36 Locally connected — 66
— of groups and homomorphisms 390 — locally connected at point 66
Short — 390 Locally contractible — 65
Serre bundle 269 — locally contractible at point 65
— condition 269 Locally Euclidean — 131
Strong — — 273 Metric — 8
Set, bounded 9 Metrizable — 9
Closed — 5 Normal — 16
Dense — 6 Orbit - 288
Distinguishable — 15 Ordered simplicial — 102
Everywhere dense — 6 Paracompact — 26
Invariant — 286 Pointed — 49
Nowhere dense — 6 Projective — 38
Open — 5 Quaternionic — — 40, 47
Saturated — 2 Quotient — 31
Short sequence 390 r-simple — 382
Simple local system 379 Real projective — 39, 41
— space 382 Regular — 16
Simplex 103 Rigged cellular — 81
Euclidean — 99 Rigged-equivalent — s 86
Ordered — — 100 Second countable 19
Ordered topological — 100 Semilocally simply connected 495
Topological — 100 Separable 19
Unit - 100 Simple — 382
Simplicial approximation 118, 130 Simplicial — 101
— characteristic map 103 Simply connected — 64
— join 113 Stiefel — 342
— map 100, 104 Strongly locally contractible 66
— mapping cylinder 113 — strongly locally contractible
Monotone — map 104 at point 66
Ordered — space 106 Tangent — 156
— product 109 Topological — 5
— scheme 105 Total - (of bundle) 265
of simplicial space 105 Upper Grassman — 337
— space 1 0 1 Special orthogonal group 284
Simply connected space 64 — unitary group 284
Skeleton 85 Sphere 9
Smooth manifold 149 — with crosscaps and holes 256
— map 1 45 — — handles and holes 256
Space, Baire 219 Spherical modification 243
Cellular — 82 Spheroid 373, 384
Cellular equivalent — s 86 Boundary of — 418
Classifying — 318 Fundamental — 416
CNRS - 7 0 Inverse — 373
Compact — 21 Spheroids, composition of — 433
Complete simplicial — 104 Freely homotopic — 380, 385
Complex Grassman — 33 7 Product of — 373
Complex projective — 39, 41 Split sequence 391
Connected — 62 — 7T-sequence 394
Contractible — 59 Stable equivalence (of bundles) 335
Covering — 268 — homotopy group 4 29, 44 0
First countable — 19 Stably parallelizable manifold 360
 515

Stably trivial bundle 3 3 5 Binary tetrahedral group 295

Standard elementary cobordism 24 3 Topological embedding 14
— fiber 299 — group 2 76
— loop 44 5 — manifold 137
— Morse function 24 3 Metrizable — space 9
— trivial bundle 267 Ordered — simplex 100
— — cobordism 24 7 — pair 10
Star 114 — space 5
Barycentric — 117 — structure 5
Open — 114 — triple 10
Stationary homotopy 57 Topologically effective action 309
— subgroup 288 — trivial bundle 267
Steenrod bundle 300 Topology 5
Stiefel manifold 171 Cellular — 82
Complex — — 174 Cellular weakening of — 83
Noncompact — — 175 Compact-open — 46
Quaternionic — — 175 Discrete — 5
— space 34 2 Metric — 9
Strong deformation retract 59 Quotient — 31
— — retraction 59 Relative — 10
Strongly locally contractible Trivial — 5
space 6 6 Weak — 82
— — — at point, space 66 Torus 256
Structure, differentiable 141 Double — 256
— group 29 9 — knots 462
Holomorphic — 364 Total characteristic map 8 2
Subbundle 331 — manifold of tangent vectors 158
Subdivision, barycentric 108 — space of bundle 265
Subgroup, isotropy 28 8 Transitive action 285
Normal — 277 Translation 379
Stationary — 288 Transversalization 2 0 2 , 350
Van Kampen — 44 9 Normal — 2 0 2 , 350
Submanifold 150 Transverse map 2 2 2 , 225
Neat - 150 Triad 11
Submersion 164 Cellular — 84
Subordinate covering 494 Triangle inequality 8
Subordination 494 Triangulation 101
Subscheme 105 Triple, cellular 84
Complete — 105 Topological — 10
Subspace 10, 84, 104, 146, 289 Trivial bundle 267
Complete — 104 — cobordism 24 7
Neat - 147 Locally — bundle 267
Sum of bundles 333 Standard — bundle 267
— — maps 3 Standard — cobordism 24 3
— — sets 3 — topology 5
— — topological spaces 26 Trivialization of bundle 267
Support of chart 142 Tube 2 0 3 , 350
— — topological simplex 1 0 1 Neat - 2 0 3 , 351
Surfaces, model 255 Open - 2 0 3 , 350
Suspension 42, 52, 335, 424 Tubular neighborhood 20 3, 350
— homomorphism 4 24 Type, homotopy '58
Symplectic group 284
System of groups, local 378
Union of topological spaces 35
Unit cube 1 1
Tangent bundle 360 — simplex 1 00
— space 156 Unitary group 284
— vector 156 Universal bundle 316.
Tangential map 361 — covering 497
Tensor product 50, 51 Urysohn function 17
Cellular — — 98
Tetrahedral group 295
 516

Value, critical 164 k-connected pair 65

Van Kampen subgroup 449 — space 64
Vector bundle 324 k-connectedness 6 3
Complex 328 k-equivalence 466, 487
Oriented --- 326 k-universal bundle 318
Real — — 324 n-cell 81
— field 160, 330 r-simplex 99, 100, 101
Tangent — 156 r-simple space 382
Zero — field 160, 330 — pair 386
Vertex of cone 42, 51 W-F-bundle 299
— — Euclidean simplex 99 W-F-equivalence 29 9
— — suspension 42, 52 W-F-isomorphism 299
W-F-map 299
y-map 286, 289
Weak hamotopy equivalence 463, 467, 474 C-transversalization 350
— topology 82 Normal — 350
Weakening, cellular — of topology 8 3 Ti-sequence 394
Weakly associated bundle 307 Exact — 395
Wedge (of pointed spaces) 50 Split - 396
Whitehead product 482 cp-basis (of tangent space) 157
cp-coordinates (of tangent vector) 157
v-th immersion 34
Zero vector field 160, 330

A-homotopy 5 7
a-th barycentric coordinate 99
a-th barycentric function 1 0 2
C*>atlas 142
C -bundle 34 6
Complex vector — 354
Euclidean — 354
Hermitian — 354
Real vector — 354
C^r-bundle 347
C^-embedding 161
C -equivalence 34 7
Cr-function 141
C -isomorphism 34 7
C -manifold 149
Cr-map 141, 145
Cr-parallelizable 160
C>-space 143
-space 143
C -structure 142
Cr~subspace 147
C -topology 197
F-bundle 300
Locally F-trivial — 300
F-trivial — 300
Weak — 299
F-isomorphism 298, 300
F-map 300
F-structure 297
G-equivalence 299
G-map 286, 289
G-space 288
Effective — 288
Right - 290
G-structure 298
H-space 413
Glossary of Symbols

Con -46 El - 2 4 4
con -42, 51 el - 244
conj - 328 Embr - 197
const — 50 Esi — 99
cop - 13 7 ev _ 4 9 3
corr - 267 ext - 325
CP11 - 39 fact _ 3
oo
115 " 41 Fibr - 317
csc - 114 pr - 6
Cub - 419 G(n,k) - 177
cub - 420 G (°°,k) - 337
(CV(n,k) - 173 G ' (n,k) - 185
I V (00,k ) - 342 G + (n,k) - 178
(CV' (n,k) - 176 G+ (°°,n ) - 337
(EV' (°°,k) - 342 GF - 445
Cy1 _ 45 GL (n,CC) - 176
“ 335 GL(n,M) - 176
d " 159 GL (n,3R) - 175
3 - 133,143, 385 GL+ (n,3R) - 176
D. - 157 n
1 GL +!IR - 324
°n “9 GL(En - 324
dx - 156, 159 GL3Rn - 324
D00 - 41 gr - 490
d T (x,p) - 202 Gra “ 340
deg - 364 9 rad " 241
diag - 4 IH - 38
Diff - 197 m n - 39
00
Dist - 9 3H - 41
dist - 8 M * - 281

dopp - 137, 210 hel ~ 268

DS - 54 helra - 268
DS - 63 MG(n,k) - 183
DS+ - 63 M P n-- 40
DT - 100 4^
e _ 277 HV(n,k)
HV( n , k ) - 175
G
HV'(n,k) - 176
 518

-,]RП -
о
8 supp - 142
OO
Ж - 41 Symm - 282
]R* - 281 (e.g. Afc) - 181
3R^ - 133 Ts - 378 , 381, 386, 4 0 4
3R^ ” 1 - 134 T r - 100
IR* - 281 Tang - 156, 157
IRÇ - 329 tang - 36 0
Reg - 493 TD - 100
rel - 2
TG - 319
IRG (n ,к ) - 177
ti - 265
H G ' (n,к) - 185
tn - 157
Ж Р П - 39 Top - 282
1RP°° - 41 Tu - 203
rt - 45 tu - 203
3RV (n ,к ) - 171
Tub - 2 03
3RV' (n,k) - 175
tub - 203
Sn - 9 и - 441
OO

S - 41 U(n) - 174
S T (x,p) - 202
Uin - 324
Scyl - 113
V (n,к) - 171
SecS - 350
V (°°,k) - 342
sch - 105 V ’ (n,k) - 175
shi - 185
V'(°°,k) - 342
Si - 102
vk - 44 9
si - 1 0 2
Wd - 4 79
ske - 85
wd - 4 79
SO - 441
ZZ - 281
S O (n) - 172
7L - 294
Б ОШ 11 - 324 m
zer - 3, 278
Sp - 441
Л - 407
Sp(n) - 175
TT - 57
Sph - 373, 384
ïï - 373 , 384
sph - 416 r
V
Sph° - 416, 417 (e.g. <f>V ) - 47
Л
St - 114 (e.g. ф л ) - 47
U
st - 114 (e.g. фи ) - 54
n
Stab - 429 (e .g. 'P ) - 54
1 -
L ¡r-Ц
Stee - 315 (e.g. Y / 5 ) - 178 , 332

su - 42, 52, 335, 424

SU(n) - 174 (e.g. f 1) - 266
Submr - 1 9 7
 519

# (e.g. f#)- 156, 377, 418

X (e.g. h X ,GX) - 307, 308

* (e.g. f*) - 377

0 - 1

L I - 3

X - 1 , 2<
V - 50
* 42, !
-

0 51 -

0 - 334

U - 38
Ф
X - 94
c
* - 97,
с
0 - 98
с
X - 110
s
★ - 113
s
 I. M. James

General
Topology and
Homotopy
Theory
1984. 3 figures. Approx. 459 ms-pages.
ISBN 3-540-90970-2

Contents: The Basic Framework. - The

Axioms of Topology. - Spaces Under and
Spaces Over. - Topological Transformation
Groups. - The Notion of Homotopy. - Cofi-
brations and Fibrations. - Numerable Cover­
ings. - Extensors and Neighbourhood
Extensors.

General Topology and Homotopy Theory is a

comprehensive and detailed account of all of
the elements of point set topology that are
useful and needed in algebraic topology.
Springer-Verlag This, and the author’s desire to give complete
Berlin proofs for his results makes the book unique
Heidelberg and extremely useful to the student as well as
New York to the specialist; at the same time, it can serve
Tokyo as an introduction and as a reference.
 888024

D .H .L u e c k in g , L .A .R u b e l

Complex Analysis:
A Functional
Analysis Approach
Universitext

1984. 7 figures. VII, 176 pages.

ISBN 3-540-90993-1

Contents: Introduction. - Preliminaries:

Set Theory and Topology. - Preliminar­
ies: Vector Spaces and Complex
Variables. - Properties of C (G) and
H(G ). - More About C(G) and H(G). -
Duality. - Duality of H (G) - The Case of
the Unit Disc. - The Hahn-Banach
Theorem, and Applications. - More
Applications. - The Dual of H (G). -
Runge’s Theorem. - The Cauchy
Theorem. - Constructive Function
Theory. - Ideals in H (G). - The Riemann
Mapping Theorem. - Carathéodory
Kernels and Farrell’s Theorem. - Ring
Springer-Verlag (not Algebra) Isomorphism of H(G). -
Berlin Dual Space Topologies. - Interpolation. -
Heidelberg Gap-Interpolation Theorems. - First-
New York Order Conformai Invariants. -
Tokyo References.
DATE DUE
 QA611 .R6513 1984 UTLAS
Rokhlin, V. A.
[Nachal'nyi kurs topologii.
English]
B e g i n n e r ’s course in topology :
geometric chapters- - - —
Fuks • Rokhlin Beginner’s Course in Topology

Based on lectures given by the authors at Moscow and Lenin­

grad Universities, this textbook gives a systematic exposition of
the geometric chapters of elementary Topology: Homotopy,
Fibre Bundles and Smooth Manifolds.

One of the main purposes is to present these parts of topology

as a unified whole. As a methodical innovation, a coordinate-
free approach to the theory of fibre bundles and an exposition
of the foundations of differential topology without use of differ­
ential équations are given.

Many descriptive examples and drawings are included; each

section is followed by a set of judiciously chosen exercises. The
reader requires only a basic knowledge of set theory, algebra
and calculus. For first-year graduate students this book provides
a thorough and profound introduction to some of the advanced
aspects of Topology.

ISBN 3-540-13577-4
ISBN 0-387-13577-4