[S<OU_1 62022
m
Call NO.
OSMANIA UNIVERSITY LIBRARY 5 13S* / f *M F Accession No 6
.
Author
Title
^UnUr^.S
Fcuw^aVltvvS
M
last
This book should be returned on or before the date
marked
beloxv.
FOUNDATIONS OF ALGEBRAIC TOPOLOGY
PRINCETON MATHEMATICAL SERIES
Editors:
MARSTON MORSE and A. W. TUCKER
and Representatives. By
1.
The
Classical Groups, Their Invariants
HKKMAXN
WKVI,.
2.
Topological Groups.
By
L. PONTRJAGIN. Translated by
EMMA
LKIIMER.
3.
An
Introduction to Differential Geometry with Use of the Tensor Calculus.
By
LUTHER PFAHLER EISENHABT.
4.
Dimension Theory.
By WITOLD HUREWICZ and HENRY WALLMAN.
Celestial Mechanics.
5.
The Analytical Foundations of
By AUREL WINTNER.
6.
The Laplace Transform. By DAVID VERNON WIDDER.
Integration,
7.
By EDWARD JAMES McSiiANE,
I.
8.
Theory of Lie Groups:
By CLAUDE CHEVALLEY,
Statistics.
9.
Mathematical Methods of
By HARALD CRAMER.
10.
Several Complex Variables.
Introduction to Topology.
By SALOMON BOCIINKR and WILLIAM TED MARTIN.
LI.FSCHETX.
11.
By SOLOMON
12.
Algebraic Geometry and Topology. Edited by R. H. Fox, D. C. SPENCER, and A. W. TUCKER. Algebraic Curves.
13.
By ROBERT
J.
WALKER.
14.
The Topology
of Fibre Bundles.
By NORMAN
STEENHOD.
15.
Foundations of Algebraic Topology.
ROD.
By SAMUEL EILENBERG and NORMAN STEEN
16.
Functionals of Finite Riemann Surfaces.
C. SPENCER.
By MENAHEM SCHIFFER and DONALD
By ALONZO CHURCH.
17.
Introduction to Mathematical Logic, Vol.
I,
18.
Algebraic Geometry.
By SOLOMON LEFSCHETZ.
19.
Homological Algebra. By HENRI CARTAN and SAMUEL EILENBEHU.
20.
The Convolution Transform. By
I.
I.
HIRSCHMAN and D. V. WIDDER.
21.
Geometric Integration Theory. By HASSLER WHITNEY.
FOUNDATIONS
OF
ALGEBRAIC
TOPOLOGY
BY SAMUEL EILENBERG
AND NORMAN STEENROD
PRINCETON, NEW JERSEY
PRINCETON UNIVERSITY PRESS
1952
Copyright
1953, by Princeton University Press
L. C. Card
SMSfl
LONDON: OXFORD UNIVERSITY PRESS
ALL RIGHTS RESERVED
SECOND PRINTING 1957
PRINTED IN THE UNITED STATES OF AMERICA
TO SOLOMON LEFSCHETZ IN ADMIRATION AND GRATITUDE
Preface
1.
PREAMBLE
book
is
The
principal contribution of this
an axiomatic approach to
It is the oldest
the part of algebraic topology called homology theory.
and most extensively developed portion of algebraic topology, and may be regarded as the main body of the subject. The present axiomatization is the first which has been given. The dual theory of cohomology is
likewise axiomatized.
It is assumed that the reader is familiar with the basic concepts of algebra and of point set topology. No attempt is made to axiomatize these subjects. This has been done extensively in the literature. Our
achievement is different in kind. Homology theory is a transition (or It is this transition which is function) from topology to algebra.
axiomatized.
Speaking roughly, a homology theory assigns groups to topological spaces and homomorphisms to continuous maps of one space into another. To each array of spaces and maps is assigned an array of
groups and homomorphisms. In this way, a homology theory is an algebraic image of topology. The domain of a homology theory is the topologist's field of study. Its range is the field of study of the algebraist. Topological problems are converted into algebraic problems. In this respect, homology theory parallels analytic geometry. However, unlike analytic geometry,
it is
not reversible.
The derived
algebraic
system represents only an aspect of the given topological system, and is usually much simpler. This has the advantage that the geometric problem is stripped of inessential features and replaced by a familiar type of problem which one can hope to solve. It has the disadvantage that some essential feature may be lost. In spite of this, the subject has
proved
its
value by a great variety of successful applications.
Our axioms are statements of the fundamental properties of this assignment of an algebraic system to a topological system. The axioms are categorical in the sense that two such assignments give isomorphic
algebraic systems.
2.
THE NEED FOR AXIOMATIZATION
is
The
It is true
construction of a homology theory that the definitions and necessary
vii
exceedingly complicated.
lemmas can be compressed
viii
PREFACE
within ten pages, and the main properties established within a hundred. But this is achieved by disregarding numerous problems raised by the
and ignoring the problem of computing illustrative examThese are serious problems, as is well known to anyone who has taught the subject. There is need for a perspective, and a pattern into which the student can fit the numerous parts. Part of the complexity of the subject is that numerous variants of the basic definitions have appeared, e.g. the singular homology groups of Veblen, Alexander, and Lefschetz, the relative homology groups of Lefschetz, the Vietoris homology groups, the Cech homology groups, the Alexander cohomology groups, etc. The objective of each variant was to extend the validity of some basic theorems, and thereby increase their
construction,
ples.
range of applicability. In spite of this confusion, a picture has gradually evolved of what is and should be a homology theory. Heretofore this has been an imprecise
picture which the expert could use in his thinking but not in his exposition. precise picture is needed. It is at just this stage in the develop
A
ment
of other fields of
mathematics that an axiomatic treatment apair.
peared and cleared the
The
follows:
(1) (2)
(3)
discussion will be advanced
by a rough
outline of the construc
tion of the
homology groups
of a space.
There are four main steps as
space
complex
> >
>
(4)
complex oriented complex groups of chains
In the
first
oriented complex groups of chains
homology groups
(1) it was necessary to place on a space the strucby decomposing it into subsets called cells, each cell being a homeomorph of a euclidean cube of some dimension, and any two cells meeting, if at all, in common faces. It was recognized that
form of
ture of a complex
only certain spaces, called triangulable, admit such a decomposition. There arose the problem of characterizing triangulable spaces by other
properties. This is still unsolved. Special classes of spaces (e.g. differentiable manifolds) have been proved triangulable and these suffice for
many
applications.
different
case, the
The assumption of triangulability was eliminated in three ways by the works of Vietoris, Lefschetz, and Cech. In each
relation that the
complex be a triangulation of the space was replaced by another more complicated relation, and the complex had to be infinite. The gain was made at the cost of effective computability. Step (2) has also been a source of trouble. The problem is to assign
THE NEED FOR AXIOM AT IZ AT ION
ix
integers (incidence numbers) to each pair consisting of a cell and a face (of one lower dimension) so as to satisfy the condition that the boundary
of a cell be a cycle. This is always possible, but the general proof requires the existence of a homology theory. To avoid circularity, it
was necessary to
restrict the class of
complexes to those for which
orientability could be proved directly. The simplicial complexes form such a class. This feature and several others combined to make simplicial
complexes the dominant type. Their sole defect is that computations which use them are excessively long, so much so that they are impractical for the computation of the homology groups of a space as simple as a
torus.
Steps
braic
(3) and and unique.
(4)
have not caused trouble.
They
The only
difficulty a student faces
is
are purely algethe absence of
motivation.
The final major problem is the proof of the topological invariance of the composite assignment of homology groups to a space. Equivalently, one must show that the homology groups are independent of the choices
made
in steps (1)
and
(2).
Some
velopment
of a fully satisfactory proof of invariance.
e.g.
thirty years were required for the deSeveral problems
way have not yet been solved, have isomorphic subdivisions? complexes
arising along the
Do homeomorphic
origin of the present axiomatic treatment was an effort, on the were of the authors, to write a textbook on algebraic topology. part with the problem of presenting two parallel lines of thought. One faced
The
We
was the rigorous and abstract development of the homology groups of a space in the manner of Lofschetz or Cech, a procedure which lacks apparent motivation, and is noneffective so far as calculation is concerned. The other was the nonrigorous, partly intuitive, and computable method of assigning homology groups which marked the early historical development of the subject. In addition the two lines had to be merged eventuThese difficulties made an axiomatic approach. The axioms which we use meet this need in every respect. Their statement requires only the concepts of point set topology and of algebra. The concepts of complex, orientation, and chain do not appear here. However, the axioms lead one to introduce complexes in order to
ally so as to justify the various computations.
clear the need of
calculate the
homology groups
of various spaces.
Furthermore, each
of the steps (2), (3), and (4) is derived from the axioms. These derivations are an essential part of the proof of the categorical nature of the
axioms.
Summarizing, the construction of homology groups
diverse story, with a fairly obscure motivation.
is a long and In contrast, the axioms,
x
PREFACE
which are given in a few pages, state precisely the ultimate goal, and motivate every step of the construction. No motivation is offered for the axioms themselves. The beginning student is asked to take these on faith until the completion of the first three chapters. This should not be difficult, for most of the axioms are
quite natural, and their totality possesses sufficient internal beauty to inspire trust in the least credulous.
3.
COMPARISON WITH OTHER AXIOMATIC SYSTEMS
for
an axiomatic treatment has been felt by topologists for This has resulted in the axiomatization of certain stages in many years. the construction of homology groups. W. Mayer isolated the step (4). He defined the abstract and purely algebraic concept of chain complex,
The need
and showed that
He also it was adequate for the completion of step (4). demonstrated that a number of mixed geometricalgebraic concepts and arguments could be handled with algebra alone. A. W. Tucker axiomatized the notion of an abstract cell complex, i.e. the initial point of step (2), and showed that steps (2), (3), and (4) could be carried through starting with such an object. This had the
effect of relegating
the geometry to step
(1)
alone where, of course,
it is
essential.
Cartan and J. Leray have axiomatized the concept of a grating (carapace) on a space. In essence, it replaces the notion of comTheir associated plex in the fourstep construction outlined in 2. Most important is the invariance theorem has several advantages. inclusion of the de Rham theorem which relates the exterior differential forms in a manifold to the cohomology groups of the manifold. It is to be noted that these various systems are axiomatizations of
Recently
II.
stages in the construction of homology groups. None of them axiomatize a transition from one stage to another. Thus they differ both in scope and in nature from the axioms we shall give. The latter axiomatize the
full
transition
from spaces to homology groups.
4.
NEW METHODS
The great gain of an axiomatic treatment lies in the simplification obtained in proofs of theorems. Proofs based directly on the axioms are usually simple and conceptual. It is no longer necessary for a proof to be burdened with the heavy machinery used to define the homology
groups.
Nor
is
proof
still
hold
if
one faced at the end of a proof by the question, Does the another homology theory replaces the one used? When
STRUCTURE OF THE BOOK
xi
a homology theory has been shown to satisfy the axioms, the machinery of its construction may be dropped. Successful axiomatizations in the past have led invariably to new
techniques of proof and a corresponding new language. The present system is no exception. The reader will observe the presence of numerous
diagrams in the text. Each diagram is a network or linear graph in which each vertex represents a group, and each oriented edge represents a homomorphism connecting the groups at its two ends. A directed path in the network represents the homomorphism which
the composition of the homomorphisms assigned to its edges. Two paths connecting the same pair of vertices usually give the same homois
morphism.
This
is
called
minded individual can regard
a commutativily relation. The combinatorially it as a homology relation due to the preline,
sence of 2dimensional cells adjoined to the graph. If, at some vertex of the graph, two abutting edges are in
one
oriented toward the vertex and the other away, it is frequently the case that the image of the incoming homomorphism coincides with the kernel
of the outgoing homomorphism. This property is called exactness. It asserts that the group at the vertex is determined, up to a group extension, by the two neighboring groups, the kernel of the incoming homo
Exact morphism, and the image of the outgoing homomorphism. Their sequences of groups and homomorphisms occur throughout. algebraic properties are readily established, and are very convenient. The reader will note that there is a vague analogy between the commutativityexactness relations in a diagram and the two Kirchhoff laws for an electrical network. Certain diagrams occur repeatedly in whole or as parts of others. Once the abstract properties of such a diagram have been established, they apply each time it recurs.
The diagrams incorporate a large amount of information. Their use provides extensive savings in space and in mental effort. In the case of many theorems, the setting up of the correct diagram is the major part of the proof. We therefore urge that the reader stop at the end of each
theorem and attempt to construct for himself the relevant diagram before examining the one which is given in the text. Once this is done, the subsequent demonstration can be followed more readily; in fact, the
reader can usually supply
it
himself.
5.
STRUCTURE OF THE BOOK
Chapter i presents the axioms for a homology theory, and a body of general theorems deducible from them. Simplicial complexes and trian
xii
PREFACE
gulable spaces are treated in Chapter n.
geometric.
This chapter is entirely In Chapter in, a homology theory is assumed to be given on We then derive from the axioms the classical triangulable spaces.
algorithms for computing the homology groups of a complex. Using we show that the axioms are categorical for homology theories on
is
these,
triangulable spaces. The first three chapters form a closed unit, but one which
based
on the assumption that a homology theory
consistent in a nontrivial manner.
exists, i.e.
the axioms are
In Chapters iv through x, the exis
tence
is
is
established in four different ways.
The
singular
homology the
ory given in others in x.
Chapter vn, the Cech homology theory in ix, and two
The intervening chapters are preparatory. struction of a homology theory is complicated.
As noted above, the conNot only do we have the
four steps outlined in 2 for the construction of homology groups, but also corresponding constructions of homomorphisms. Then the axioms
Finally the dual cohomology theory must be given in total of eight theories to present, the tendency to avoid constructions and parallel others is nearly irresistible. repeat most of this by presenting a number of steps on a sufficiently abstract
must be
verified.
each case.
With a
We
level to
make them
usable in
all cases.
These are given in Chaps,
iv, v.
Chapter iv presents the ideas and language of category and functor. These concepts formalize a point of view which has dominated the devel
opment of the entire book. We axiomatize here the notion of a homology theory on an abstract category, and formulate a pattern which the subsequent constructions must fulfill. In Chapter v, the step from chain complexes to homology groups is treated. The chapter is entirely
algebraic.
cial
Chapter vi presents the classical homology theory of simplicomplexes. In Chapter vn, the singular homology theory is defined and proved to satisfy the axioms. This chapter is independent of vi except possibly for motivation. A reader interested in the shortest
first
existence proof need only read Chapter iv, the
four articles of v,
and then Chapter vn. Chapter vin treats
limit groups.
This
is
direct and inverse systems of groups and their the algebraic machinery needed for the develop
of the Cech homology theory given in ix. Chapter x presents additional properties of the Cech theory. It is shown that the addition of a single new axiom characterizes the Cech theory on compact spaces. Two additional homology theories are constructed which are extensions
ment
of the
Cech theory on compact spaces. The first is defined on locally compact spaces, and the second on normal spaces. Both are obtained by
processes of compactification.
STRUCTURE OF THE BOOK
Chapter
xi,
xiii
volume, gives a number of the homology theory such as the Brouwer fixedpoint theorem, invariance of domain, and the fundamental theorem of algebra. Homology theory and cohomology theory are dual to one another.
this
classical applications of
which completes
We treat
them in parallel. Throughout Chapters i and m, each section which treats of homology is accompanied by a section on cohomology. The latter contains no proofs. It contains just the list of definitions and theorems dual to those given for homology, and are numbered correspondingly,
e.g.
Definition 4.1c
is
the cohomology form of Definition 4.1.
The
duality between the two theories has only a semiformal status. It is true that, by the use of special "coefficient groups/' Pontrjagin has given a strictly formal duality based on his theory of character groups.
reader
However, the duality appears to persist without such restrictions. The is urged to supply the proofs for the cohomology sections. In
addition to constituting useful exercises, such proofs will familiarize the reader with the language of cohomology.
The device of dual sections occurs rarely in later chapters. The greater parts of the constructions of homology theories and their corresponding cohomology theories deal with mechanisms in which the two aspects are
not differentiated.
treated in
In the remaining parts, the two dual theories are
equal detail.
At the end of each chapter is a list of exercises. These cover material which might well have been incorporated in the text but was omitted as not essential to the main line of thought. There are no footnotes. Instead, comments on the historical development and on the connections with other subjects are gathered together in the form of notes at the ends of various chapters.
A cross reference gives the chapter number first, then the section, and, lastly, the numbered proposition, e.g. x,2.6 refers to Proposition 6 of Section 2 of Chapter x. The chapter number is omitted in the case
where
the one containing the reference. A reference of the form (3) means the displayed formula number 3 of the section at hand.
it is
We
acknowledge with pleasure our indebtedness to Professors
S.
MacLane, T. Rado, and P. Reichelderfer who read large portions of the manuscript and whose suggestions and criticisms resulted in substantial
improvements.
S.
ElLENBERG AND N. STEENROD
August, 1951
Columbia University
Princeton University
Contents
Preface
^
I.
vii
Axioms and general theorems
Simplicial complexes
3
II.
54 76
108
III.
Homology theory
of simplicial complexes
IV.
Categories and functors
V.
VI.
VII.
Chain complexes
124
of simplicial complexes
. .
Formal homology theory
.
162
The
singular
homology theory
limits
185
VIII.
Systems of groups and their
212
233 257
IX.
The Cech homology theory
Special features of the
X.
XI.
Index
Cech theory
Applications to euclidean spaces
298
324
FOUNDATIONS OF ALGEBRAIC TOPOLOGY
CHAPTER
I
Axioms and general theorems
1.
TOPOLOGICAL PRELIMINARIES
The axioms for a homology theory are given in 3. In 1 and 2, we review the language and notation of topology and algebra, and we
introduce a number of definitions and conventions which, as will be seen, are virtually enforced by the nature of our axiomatic system. We define a pair of sets (X,A) to be a set and a subset A of X.
X
is the vacuous subset, the symbol (X,0) In case A = breviated by (X) or, simply, X. A map f of (X,A) into (F,B), in symbols
is
usually ab
/:
(X,A) > (r,B),
to Y such that f(A) B. If also a singlevalued function from (Z,C), then the composition of the two functions is a (Y,B) g: map gf: (X,A) > (Z, C) given by (gf)x = g(fx).
is
X
C
The
relation (X',A')
C
(X,A) means X'
C X
and A'
e
C
X'
A.
is
The
called
map
* (X,A) defined by ix (X',A') the inclusion map and is denoted by
t:
=
x for each x
i:
(X',A')
C
(X,A).
If
(X',A')
It will
=
(A',4),
then the inclusion
map
i
is
called the identity
map
of (X,A).
be important for us to distinguish a function from those obtained from it by seemingly trivial modifications of the domain or Let /: (X,A) > (F,B) be given, and let (X',A'), (K',B') be range. r/ > and /(A') B'. Then such that X' X, Y F, /(X') pairs the unique map /': (X',A ) > (Y',B ) such that /'x = fx for each For a: e X' is called //ie wap defined by /, and / is said /o rfe,/in6 /'.
C
f
C
C
,
C
r
f
example any inclusion map is defined by the identity map of its range. If /: (X A) > (K,B), the map of A into B defined by / is denoted by
9
f\A:
A
3
> B.
AXIOMS AND GENERAL THEOREMS
The
lattice
[CHAP.
I
of
a pair (X,A) consists of the pairs
(X,0)
(0,0)

>
(A,0)
(X,A)

(X,X)
all their all their
identity maps, the inclusion
compositions.
If /:
maps indicated by arrows, and > (X,A) (F,), then / defines a map
of every pair of the lattice of (X,A) into the corresponding pair of the lattice of (F,5). In particular f\A is one of these maps. is a set topological space together with a family of subsets of
A
X
X
X,
called open sets, subject to the following conditions:
(1)
(2) (3)
and the empty set are open, the union of any family of open sets is open, the intersection of a finite family of open sets
the set
of
X
is
open.
The family
plement X
If
U
open of an open
sets of
X
is
called the topology of called closed.
X.
The com
set
U
is
A
is
in
A
a subset of X, then the union of all the open sets contained The intercalled the interior of A and is denoted by Int A.
is
section. of all the closed sets containing
A
is
called the closure of
is
A and
is denoted by A. A topology, called the relative topology, defined in A by the family of intersections A C\ U for all open sets U of X. With this topology, A is called a subspace and (X,A) is called a pair of
topological spaces or, briefly, a pair.
If
X,X' are
spaces, the relation
X'
C X means that X' is a subspace of X. A map /: X Y of one topological space
>
into another
l
is
said to
in
>
be continuous
if,
for every
open set
>
V of
is
A map
defined
of pairs /:
(X<A)
(F,J3)
F, the set f~ (V) continuous if the
is
open
X.
map
X
F
The terms map, mapping, and transformation when applied to topological spaces or pairs will always mean continuous maps. Identity and inclusion maps are continuous.
by /
is
continuous.
Xifa
if, for each pair of distinct points with x v e C/i, x2 e U 2 open sets t/i,t/2 in A space is called compact if it is a Hausdorff space, and if each covering of the space by open sets contains a finite covering. A pair (X,A) is is compact and A is a closed (and therefore compact) called compact if subset of X. The foregoing is intended as a review of some basic definitions. We shall assume a knowledge of the elementary properties of spaces and
A
space
X
is
a Hausdorff space
e
X, there
exist disjoint
X
.
X
1]
TOPOLOGICAL PRELIMINARIES
of Alexandroff
1
5
maps such as can be found, for instance, in the book Hopf (Topologie, J. Springer, Berlin 1935) Chapters
and and 2, or in the book of Lefschetz (Algebraic Topology, Colloq. Pub. Amer. Math. Soc. 1942) Chapter 1. DEFINITION. A family ft of pairs of spaces and maps of such pairs which satisfies the conditions (1) to (5) below is called an admissible category for homology theory. The pairs and maps of ft are called admissible.
(1)
If
(X,A)
e
a, then
all
pairs
in
and inclusion maps of the
ft,
lattice
of
(X,A) are in
(2) If /:
ft.
(X,A) * (Y,B)
all
is
then (X,A) and (Y,B) are in
a
together with
If /,
maps that / defines of members of the lattice of (X,A) into corresponding members of the lattice of (F,B).
(3)
and / 2 are
in
ft,
and
their composition /j/ a
is
defined,
then /,/ a
(4)
e ft.
If 7
=
[0,1] is
the closed unit interval, and (X,A)
e ft,
then the
cartesian product
(X,A)
is in ft
X
/
=
(.Y
X
7,
A X
/)
and the maps
g gi
,
:
(X,A) > (X,A)
X
/
given by
are in
(5)
ft.
There
ft, if
is
in
ft
>
a space
are in
/:
P
X, and
if
P P
consisting of a single point.
is
If
XP
9
a single point, then /
e ft.
The
theory:
ft,
following are examples of admissible categories for
homology
This
is
=
the set of
all
pairs (X,A)
and
all
maps
of such pairs,
the largest admissible category. ft c = the set of all compact pairs and all maps of such pairs. Q LC = the set of pairs (X,A) where is a locally compact Hausdorff space, A is closed in X, and all maps of such pairs having the property
X
that the inverse images of compact sets are compact sets. This last example of an admissible category has the property that
both
X' X'
>
X and X' can be admissible, X' C X may not be admissible. This
an open but not closed subset
of
is
X, and yet the inclusion map the case if X is compact and
is
X.
6
AXIOMS AND GENERAL THEOREMS
DEFINITION.
[CHAP.
I
category
ft
Two maps / ,/i: (X, A) > (F,B) in the admissible are said to be homotopic in ft if there is a map
*:
(X,A)
X
/

(K,B)
in
ft
such that
/
or, explicitly,
=
A0
,
/i
=
7i0i
/
(x)
=
A(*,0),
/,(*)
=
A(*,l).
The map
/i
is
called a homotopy.
2.
ALGEBRAIC PRELIMINARIES
Let R be a ring with a unit element. An abelian group G is callod an Rmodule if for each r e R and each g e G an element rg e G is defined such that
Kffi
+
=
2)
=
rg,
+
rg 2
,
(r,
+
r 2 )g
rfag)
(fy 2 )0,
ig
=
= r^
+ r g,
2
g.
A
of G such that rh & whenever r e /?, h e // is called a subgroup Knear subspace of G. A homomorphism of G into another 72module
is
H
H
G'
called fo'near
if <t>(rg)
=
r<t>(g)
holds for
all r s 7^,
g
e
G.
Only two special cases are of importance to us. In the first, R = F is a field. Then G is a vector space over F. In the second, R = J is of integers. In this case G is an ordinary abelian group (without the ring additional structure). The unifying concept of module saves repetition.
In addition to /2modules, we shall also wish to consider compact abelian groups. To avoid a complete duplication of the discussion, we adopt the following convention:
Unless otherwise stated the word% "group" will be used to
either one of the following
mean
two objects: 1. An 72module (over some ring R with a unit element), or 2. A compact topological abelian group. Whenever, in a discussion, several "groups" appear, the word "group"
is
ring
to be interpreted in a fixed manner. In particular, in case 1, the R is the same for all groups. All groups are written additively. If
G
is
means that G consists of the zero element The word "subgroup" will mean correspondingly
a group,
G =
alone.
1. 2.
A linear subspace, or A closed subgroup.
.
2]
ALGEBRAIC PRELIMINARIES
will
7
The word "homomorphism"
1.
mean
A linear map, or 2. A continuous homomorphism. If G and H are groups, the notation
0:
G
>
H
<t>
means that
subgroup image of 4
is
of elements of
H" means
by "0
is
maps G homomorphically into H. The kernel of is the G mapped into the zero element of H. The the subgroup 0(G) C H. The statement "0 maps G onto
0(G) onto."
= H. We
sometimes abbreviate "0 maps
G
onto
H"
The symbolism

is
used to indicate that the
kernel of
is all of G, or equally well that 0(G) = 0. The expression has kernel zero" means that the kernel of contains only the zero "0
element.
The symbolism
0:
G ~
H
H maps G isomorphically onto #, and ~ G. an isomorphism. Observe that 0: G = implies <f>~ In case 1, this is obvious. In case 2, it follows from the theorem that the inverse of a continuous 11 map between compact spaces is continuous. It is precisely the failure of 0" to be continuous in the noncompact case which prevents our unifying the concepts of TJmodule and compact group by the use of topological /^modules.
means
is
that the
map
0:
G
>
called
H
1
:
H
1
If
i.e.
L
is
a subgroup of G,
G/L
denotes the factor (or difference) group,
the group whose elements are the cosets of
L
in G.
The
natural
homomorphism
77:
G
> G/L
is
the function which attaches to each element of
it: 77(0)
G
the coset of
L which
contains
=
g
+
It
L.
In case
77
that
l
G/L
is
is
an /2module and
is
define r(g linear. In case
1
we
+
L)
=
rg
+ L so
2, we
introduce a
topology into
rj~
G/L
as follows: a subset
(U)
open
in G.
U of G/L is open if and only if can then be seen clearly that G/L is a compact
is
abelian group and that If 0: G > G' and L
coset of
77
continuous.
C G, L' C G' are subgroups such that 0(L) C L',
0:
then the homomorphism
G/L
>
G'/L' induced by
<t>
attaches to each
L
in
G
the coset of L' in G' which contains
77,77'
The
natural
maps
and the
image under 0. homomorphisms 0,0 satisfy the comits
mutativity relation
077
==s
77
0.
8
It asserts that
AXIOMS AND GENERAL THEOREMS
two homomorphisms
of
[CHAP.
1
G into G /L'
f
coincide.
As shown
in the
diagram
*
G
1
>
G'
9 ,
>
1''
G/L
the
G'/L'
first is obtained by moving down and then over, the second, by moving over and then down. If {G a a = 1, n, are groups, their (external) direct sum ^i G a
} ,
,
is
defined, in the usual way, as the set of ntuples {0 a
},
ga
e
Ga
,
with
r{g a
=
]
\rg a
}>
{g a
}
+
{gi}
=
[g.
+
gi\.
G* * s given the product topology. G a > G, a. = 1, n, determine a homomorphisms i a i: G a > G by the rule i({g a }) = ]Ci i a (g a ). If homomorphism ^.i i is an isomorphism of ^ on ^ ^> en the set \i a is called an G as a direct sum, and each component i a is injective representation of called an injection. If, in addition, G a C G and each i a is an inclusion, G is said to decompose into the (internal) direct sum G = ]T) G a A set of homomorphisms j a G > G a a = 1, n, determine a
]
In case 2,
A
set of
:
,
2
^
]
.
:
,
,
homomorphism
isomorphism
projection.
j:
G
>
of (? onto
^
]Ci ^ ^Y
^ e ru e
'
,7(0)
= b^} ^ J
ja
is
ls
an
Ga
,
then the set
{j a
}
is
called a protective called a
representation of
G
as a direct sum, and each
component
Given an injective representation \i a as a direct sum', one constructs a projective representation \j a by defining j a g to be the acoordinate of f l g. Similarly a projective representation determines an injective representation. The advantage of having the two types is that they are dual and one can state the dual of a direct sum theorem by interchanging In Chapter v where we deal with infinite injection and projection.
}
}
direct sums, the distinction
between the two types
will
be more than
formal.
DEFINITION. A lower sequence of groups is a collection {G <,$<,} where, for each integer q (positive, negative, or zero), G Q is a group, and Q G g > G Q .i is a homomorphism. A lower sequence is said to be exact if, for each integer g, the image of # a+1 in G Q coincides with
(f>
:
the kernel of
<
fl
.
DEFINITION. An upper sequence of groups is a collection {G $ q where, for each integer q (positive, negative, or zero), G is a group and
\
9
q
2]
ALGEBRAIC PRELIMINARIES
9
<t>*:
is a homomorphism. An upper sequence is exact if, for ~ each integer q the image of 4> Q l in G Q coincides with the kernel of <f>*. 9 = <_ a for = G_ a If {G Q Q is a lower sequence, and we set G a a is an upper sequence. each q, then {(z This transformation sets up a 11 correspondence between the set of all lower sequences and the set of all upper sequences. In the sequel, definitions are made only for lower sequences. The corresponding definitions for upper sequences are to be obtained using this transformation. DEFINITION. A lower sequence G = {G' q is said to be a subseq quence of the lower sequence G {G q }, written G' G, if, for each G g and q = Q \G' A subsequence is determined by any set Q <?, G' Q of subgroups {G' provided a (GJ) Gi for each q. Q REMARK. The word subsequence is used here in a sense different from the usual one no terms of the original sequence are discarded. It is not to be expected that a subsequence of an exact sequence should be = kernel 4> q is usually not exact. For example, the subsequence q
> (7
y
Gq
a+1
,<t>
\
,
<
,</>
}
f
,<t>
}
,<t>
C
C
<f>
<t>
.
}
C
K
exact.
DEFINITION.
G
q,
>
G'
:
$Q
If (7,<7' are two lower sequences, a homomorphism $: a sequence of homomorphisms {$ Q such that, for each integer G Q * G q and the following commutativity relations hold:
is
}
The subgroups
of
\f/
{kernel
\l/
\I/ Q \
\l/,
and kernel
{image
\f/ \I/ Q \
=
is
form a subsequence of
\I/ Q
G
q.
called the kernel
=
say that
is
onto.
Likewise image a subsequence of G and when G' = image ^, we is said to be If each \f/ g is an isomorphism, then
for each
?/
;
means kernel
=
\{/
an isomorphism. The commutativity
of
relations assert that, in the following
diagram
homomorphisms
the two composite homomorphisms from (j c to GJ_i coincide. DEFINITION. If L is a subsequence of the lower sequence (7, the factor sequence G/L of G by L is the lower sequence composed of the factor groups GJL Q and the homomorphisms 4> q GJL q > G Q i/L^i induced by the q Let ri Q G g > G q /L 9 be the natural homomorphism.
:
<f>
.
:
10
AXIOMS AND GENERAL THEOREMS
<j> q rj Q
[CHAP.
I
Since
=
ij Q
i<t> Q ,
it
follows that
of
rj
=
{ rj Q
}:
G
>
G/L.
It is called
the natural
onto G/L. DEFINITION. If G',G" are lower sequences, their (external) direct sum G' G" is the lower sequence whose groups are G Q = G G",
homomorphism
G
+
+
and whose homomorphisms are defined by
<t> Q
Q (g Q ,g' )
=
(</>(00,<"(00)
DEFINITION. If G',G" are subsequences of the lower sequence G, then G is said to decompose into the (internal) direct sum of G' and G" if, for each g, G decomposes into the direct sum of G' and G". Q
fl
The conventions and notations introduced for pairs of sets (X,A) and their maps will also be used for pairs of groups, pairs of sequences,
and
their
homomorphisms.
3.
AXIOMS FOR HOMOLOGY
A
homology theory
H
on an admissible category
ft is
a collection
of three functions as follows:
The
first is
a function
H
q
(X,A) defined
for each pair (X,A) in a and each integer q (positive, negative, or zero). The value of the function is an abelian group. It is called the qdimen
sional relative homology group of
X modulo A

.
The second function
is
defined for each
map
/:
(X,A)
(Y,B)
in
(1)
Ct
and each integer
q,
and attaches to such a pair a homomorphism
/..:
H,(X,A)
H.(Y,B).
/.
Ct
It is called the
homomorphism induced by
is
The
third function d(g,X,A)
q.
defined for each (X,A) in
and
each integer
(2)
Its value is a
homomorphism
H,(X,A)
d(q,X,A):
 H.^A)
called the boundary operator. Since, in (2), the symbol (q,X,A) is redundant, it will be omitted in the future. Likewise the index q on / +a in (1) will be omitted.
Q (X,A) is According to the convention of the preceding section, an /2module, or always a compact abelian group. The corresponding conventions govern the homomorphisms d and /,. In addition, the three functions are required to have the following
H
either always
properties:
Axiom
1.
If f
identity, then
/#
=
identity.
Explicitly, if / is the identity map of (X,A) e ft for each g, / # is the identity map of q (X,A) on itself.
on
itself,
then,
H
3]
AXIOMS FOR HOMOLOGY
2.
11
Axiom
if /: (X,A) > (Y,B) and g: (Y,B) * (Z,C) arc adthen the composition of the induced homomorphisms /,,,: missible, > and g+: is the induced Q (Y,B) H.(X,A) > q (Y,B) g (Z,C) > homomorphism (gf)^: Q (X,A) Q (Z,C).
Explicitly,
H
H
H
H
H
Axiom
the
into
3.
df
if
Explicitly,
map
defined
H
g
i(B).
(X,A) > (7,B) is admissible and f\A: A > # is Q (X,A) by /, then there are two ways of mapping As shown in the diagram,
/:
H
the composition d/^ is obtained by moving over and then down, the composition (f\A)^d by moving down and over. The axiom requires that the two homomorphisms have the same value on each element of
H.(X,A).
Axiom 4 (EXACTNESS AXIOM).
//
(X,A)
is
admissible
and
i:
A
> X, j: (X,A) are inclusion groups and homomorphisms
X
map,?, then the lower sequence of
i*
$
...
is exact.
<
H
<q ^(A)
H
3+
i*
+g (X,A)
H
<Q (X)
H
a
(X)
^
&
This lower sequence
is called the
homology sequence of (X,A)
(abbreviated:
H.S. of (X,A)).
To make the above statement precise, the groups and homomorphisms of the lower sequence must be indexed by integers. We th = d, G 3 n = = H q (X,A), choose Ho(X,A) as the group, i.e. G H.(X), etc.
3<1
3<I
Axiom 5 (HOMOTOPY AXIOM).
>
//
the
admissible
q,
(Y,B) are homotopic in a, then, for each (X,A) g (Y,B) coincide. /o*,/i* ofH q (X,A) into
the
maps / fii homomorphisms
,
H
Axiom 6 (EXCISION AXIOM).
closure
set
// U U is contained in the interior of A
is
an_open subset of
[7,
X
whose
(i.e,
U C V C A for some open
A
U,
V
of X),
and
it
if the inclusion
map (X
admissible, then for each q.
induces isomorphisms
H
Q
(X
A
U) U)
>
~H
(X,A) is Q (X,A)
12
AXIOMS AND GENERAL THEOREMS
An
inclusion
[CHAP.
is
I
and U is in the an excision. just
in
X
map
i:
(X
(7,
A
will
C7)
interior of
A,
(X,A) where U be called an excision
C
open
or
map
Axiom 7 (DIMENSION AXIOM).
sisting of
a single point, then
list
H (P)
q
//
P
is
an admissible space conq
=
Ofor
all
^
0.
This concludes the
or explicit reference. The reasons for the
of
of theorems, use will be
made of Axioms 1, The remaining axioms
axioms for a homology theory. In proofs 2, and 3 without comment will be referred to by name.
names of Axioms 4, 5, and 6 are apparent. The Axiom 7 the Dimension axiom is not so apparent. Suppose H q (X,A),d,ft ls a h m lgy theory satisfying Axioms 1 through 7. Define H' (X,A) = ff.i(X,A). Define d' and /,, in the natural 9 way. Then the new homology theory satisfies Axioms 1 through 6. This is also true of the homology theory H' '(X,A) = H q (X,A) + q H 9 i(X A). Thus, Axiom 7 tends to insure that the dimensional index
reason for calling
9
q shall have a geometric meaning. The consistency of the axioms
is
easily verified
by choosing each
non
H
q
(X,A)
=
0.
The
ix.
interest, naturally, lies in the existence of
trivial
homology Chapters vn and
theories.
The
existence of such will be proved in
in the following form,
The Homotopy axiom can be put sometimes more convenient:
Axiom
5'.
which
is
// (X,A)
are defined by gQ (x)
=
admissible and g Q ,gii (X,A) = 0,1), then g Q + = g^. (z,0), g^x)
is
5'
>
(X,A)
X
/
Indeed Axiom 5 implies Axiom
since g
and
gi
are homotopic with
the identity map of (X,A) X / as homotopy. On the other hand, if > h is a homotopy of maps / ,/i: (X,A) (F,B), then / = hg Q and = hg and by Axioms 2 and 5' fi
l
/o#
= MO* = h+g^ =
/i*.
The
Excision axiom
6'.
may
be reformulated as follows:
2
l
Axiom
closed
and
X=
2
is admissible,
H
9
(X
l
U X ,X
Xi and X be subsets of a space X such that X is X U Int X If (X X Pi X C (X U X ,X ~ then induces isomorphisms i+: H (Xi,Xi P\ X
Let Int
l
2.
i:
l9
l
a)
l
2
2)
it
q
2)
2)
for each
q.
The equivalence of this axiom with the original formulation is easily = seen by setting A = 2 and Xi. The new formulation of the Excision axiom closely resembles the following theorem of group theory: If GI and <7 2 are subgroups of a group (?, then the inclusion
X
U
X
3c]
AXIOMS FOR COHOMOLOGY
IS
map (G^Gi
P\ G 2 ) > (Gi VJ G 2 ,G 2 ) induces an isomorphism of the factor VJ G 2 /G 2 Here G 2 denotes the least GJ/GX r\ G 2 = groups of G containing both Gi and G 2 subgroup In all the homology theories that will be constructed in the sequel = for q < 0. Because of the various it will be true that Q (X,A) theorems that will be established later, this is also true for uniqueness
d
.
dU
.
H
any homology theory provided that the
pair (X,A) is triangulable (see Because of these facts one could, without limiting the theory in ii,6). any essential fashion, incorporate the requirement that q (X,A) = for q < into the axioms. One could go a step farther and assume that the groups are defined only for q ^ and that d: q (X,A) is defined only for q ^ 1. The Exactness axiom q (XjA) q ^(A) would be the only one that would require modification. Indeed the homology sequence is then no longer infinite in both directions, but terminates on the left with
H
H
H
H
H(X,A)
<
The modified Exactness axiom would assert, in addition to the kernel = image property, that j+ maps H (X) onto HQ (X,A). This is the precise condition that the sequence remain exact when it is completed to a nonterminating sequence by adjoining trivial groups and homomorphisms.
This modification of the basic definition
trivial since
by defining H
q
(X,A)

for q
<
is mathematically one obtains a homology
theory in the unmodified sense. The remainder of Chapter
i is devoted to general theorems conon an admissible category. This hypothesis cerning a homology theory In addition, the will be omitted from the statement of each theorem. assumption that the pairs and maps occurring in theorems are admissible will be omitted unless there is a special reason for calling attention to
this fact.
3c.
AXIOMS FOR COHOMOLOGY
01 is
q
A cohomology theory on an admissible category three functions as follows: The first is a function
each (X,A) in
function
is
a collection of
H
(X,A) defined for
a and
each integer
q.
In each case the value of the
an abelian group.
It is called the qdimensional relative
mod A. cohomology group of The second function, /*, is defined for each admissible
(Y,B)
X
map
U
AXIOMS AND GENERAL THEOREMS
q,
[CHAP.
I
and each integer
and
its
value
is
a homomorphism
/*:
H
9
(Y,B)
 H'(X,A)
/.
e ft
called the
homomorphism induced by
is
The
q.
third function, 6(q,X,A),
defined for (X,A)
and an integer
Its value is
a homomorphism
5:
a /P'CA) * ff C3r,A)
called the coboundary operator.
NOTE: /, and /* have reverse directions, likewise d and index distinction between homology and cohomology groups
remembered by the following
rule:
5.
The
easily
is
H
q
means homology
since the index
below suggests that the corresponding operator (boundarjr) lowers the dimension, H* means cohomology since the index above suggests that
the corresponding operator (coboundary) increase? the dimension. These three functions have the following properties (the statements
in particular, conditions that the appropriate pairs missible are omitted)
:
are abbreviated since they parallel the corresponding homology axioms, and maps be ad
Axiom Axiom
Ic.
If f
(gfi*
identity, then
/*
=
identity.
2c. 3c.
4c.
=
f*g*.
Axiom
t(f\A)*
//
i:
=
f*B.
+
Axiom
the
X, j: (X,A) are inclusion maps, then and homomorphisms upper sequence of groups
i*
d
A
X
> H*~\A)
is exact.

j*
i*
6
H (X,A)
q
>
H (X)
q
> H*(A) ~>
(abbreviated:
This upper sequence C.S. of (X,A)).
5c.
is called the
cohomology sequence of (X,A)
Axiom
// the
maps
f,g:
(X,A)
>
(Y,B) are homotopic, then
r
=
0*.
Axiom
A, then
6c.
//
U
is
the inclusion
map:
open in X, and U U, A (X
is
contained in the interior of
induces isomorphisms:
H
Q
(X,A)
= H (X Q
U)
U,
>
A 
(X
9
A), if admissible,
U).
0.
Axiom
7c.
//
P
is
a point, then H*(P)
=
for q
^
There is a duality relating homology and cohomology. It is based on Q the Pontrjagin theory of character groups. Precisely, let (X,A),f*,d be a cohomology theory satisfying Axioms Ic through 7c; and suppose
H
4]
HOMOMORPHISMS OF HOMOLOGY SEQUENCES
15
H
(X,A) is always a discrete abelian group (R = the ring of integers) or always a compact abelian group. Let be the character q (X,A) q of (X,A) and let f^d be the homomorphisms dual to /*,$. group
q
H
H
y
by the use of standard properties of character that It follows that q (X,A)J+,d satisfy Axioms 1 through 7. groups, the dual of each theorem about \H q ,f+,d\ is a true theorem about
it is
Then
readily shown,
H
passing from a theorem to its dual, arrows are reare replaced by factor groups and vice versa. versed, subgroups Q In case the values of (X,A) are .Rmodules, with R unrestricted, has only a semiformal status. A partial duality is obtained the duality
{#',/*, 6}.
When
H
(X,A) as a discrete abelian group, and ignoring the operations of R. Then the strict duality above shows that the dual of each theorem of homology is a true theorem of cohomology at least in
by treating
H
q
so far as the additive structures of the groups
(X,A) are involved. Because of this duality, theorems on cohomology will not be proved, and will be left to the reader as exercises. At the end of each section on homology in Chapters i and in, the dual definitions and theorems for cohomology will be stated.
4.
H
Q
HOMOMORPHISMS OF HOMOLOGY SEQUENCES
(X,A)
>
A map
/:
(Y,B) defines maps
A:
X
F,
/2
:
A
> B.
The map / 2 is simply the map f\A THEOREM 4.1. The collection
forms a homomorphism of
(Y,B).
the
considered before.
of
,
homomorphisms f^, / u and / homology sequence of (X,A) into that of
2jt
It will be denoted by
f^.
**
PROOF.
Consider the diagram
J*
d
<
H,(X,A)
I/*
1
H
9
(X)
<
H
q
(A)
I/.*
H.(X,A)
I/*
[
[
/,
i;
l
<
Ji
d
,
H,(Y,B)
where
t,
,
H
q
(Y)
H (B)
q
H
q
(Y,B)
<
j, i',
f
are appropriate inclusions.
We
must
verify the
com
mutativity relations
The
i'/a
first
two
relations follow
is
from Axiom 2 since fj
3.
=
j'fi
and f
v
i
The
third relation
a restatement of Axiom
16
AXIOMS AND GENERAL THEOREMS
THEOREM
4.2.
[CHAP.
I
/
:
2ss
Ht(A)
H
///:
(X,A) > (F,B) and / 1#
q,
:
H
q
:
(X)
q
~
all dimensions q (B) for
H
then / #
H (X,A)
H.(Y),
q
(Y,B) for all dimensions q, a?wi /^ is cm isomorphism. This theorem is a consequence of the following group theoretic
lemma:
(THE "FIVE" LEMMA). Let {$<} be a homomorphism of {G Q Q into an exact lower sequence {G q $ q \. If some index q, ^ a + 2 ^ +i> ^ai> and ^a2 are all isomorphisms, then for q is also an isomorphism. This lemma is a consequence of the following two lemmas: LEMMA 4.4. // the kernel of ^ c _! is zero, and q + 2 is onto, then
4.3
LEMMA
an
exact lower sequence
,<t>
}
,
\l/
\l/
kernel
$ q = $ a +i(kernel ^ a +i)'
and
the kernel of
LEMMA
4.5.
//
^<,+i is onto,
^a _ 3
is zero, then
image $ q
PROOFS.
diagram:
/*>
=
tf~
Both lemmas are proved by chasing around the following
/~v
/~v
/~i
/^i
I
W2
i
If MPd1
I
Wo
/
I
/
I
MKd + l
Since
$Q
<t> Q
+i
=
<j+i^c+i,
<
it
follows that
a
+i(kernel
^+1)
C
kernel
\// q
.
prove the inverse inclusion, assume that the hypotheses of 4.4 are satisfied and consider an element g e G q with q g = 0. Then 4>ty q g = 0, = 0. Since the kernel $ q  v is zero, it follows that Q (g) = 0. so q i<t> q g By exactness, there is a g v e G q i such that q +\g\ = g. Then <j+i^ a i(7i = = 0, and, by exactness, there is a g' e G q+2 such that 2 Since ^ c+2 is onto, there is a g 2 e G q+2 such that
\l/ \l/
<t> <t>
To
^a+202
=
02.
Then, on the one hand, $ q +i(gi
^a+i0i
~~
<t> q
+zg2)
0i+2^a4202
<a+i(0i
=
0J+202
=
0,
(7
while,
on the
^+i0i other" hand,
=
""
0+202)
=
0+i0i
is
=
0.
Thus
e^+j (kernel ^ a+ i).
The proof
of 4.5
4.1c.
similar.
THEOREM
byf**.
homomorphism
collection of homomorphisms /*,/*,/* forms a of the C.S. of (Y,B) into that of (X,A). It will be denoted
The
THEOREM
4.2c.
// /:
(X,A) + (Y,B) and ft:
H
q
(Y)
 H
q
(X),
6]
BASE POINT AND COEFFICIENT GROUP
17
/*:
#'(B) H*(X,A) for
= H q (A)
all q,
for
is
all
dimensions
q,
then
/*:
#<(F,B)
=
and /**
an isomorphism.
5.
INVARIANCE OF THE
HOMOLOGY GROUPS
if
Two
admissible pairs (X,A),(Y,B) are homeomorphic
there exist
admissible
maps
/:
(X,A)
>
(F,B) and
both /# and gf are identity maps. 1 morphism, and g = / is the inverse of /. THEOREM 5.1. A homeomorphism f of (X,A) onto (F,B) induces // C (F,B). isomorphisms /: H g (X,A) 1 PROOF. Since /" V = identity, (T /)* = (/"')*/* = identity. = identity. Therefore / has an inverse Of*)" 1 = Similarly /^Cf' )*
1
 (X,A) such that (F,B) The map / is then called a homeogr:
(r
1
),.
A homeomorphism f of (X,A) onto (F,B) induces an 5.2. isomorphism of the H.S. of (X,A) onto that of (F,B). > PROOF. Since / defines homeomorphisms fii F and / 2 A > B, the theorem follows from 4.1 and 5.1.
THEOREM
X
:
THEOREM
THEOREM
5.1c.
isomorphisms /*:
5.2c.
A homeomorphism f of (X,A) onto (F,B) induces C #"(F,B) = H (X,^1). A homeomorphism f of (X,A) onto (F,B) induces an
(F,B) onto
that of
isomorphism of
the C.S. of
(X,A).
6.
THE BASE POINT P AND THE COEFFICIENT GROUP G
6.1.
DEFINITION
Let
P
denote a fixed reference point and also the
(admissible) space consisting of this single point. called the coefficient group of the homology theory
Its elements are
The group
and
is
(P ) is denoted by G.
H
denoted by
f
g,g
,
etc.
One can construct the coefficient group without choosing a base be a set, and, for each a s M, point by the following process: Let
M
let
be a group, and, for every ordered pair a,p e M, let ir be an ~ 0$ such that (1) TT* is the identity map of G a for isomorphism G a each a, and (2) if a,p,y s M, then Tyir^ = TT. Such a collection will be called a transitive system of groups. To such a system we assign a An element g e G is a function assigning to single group G as follows: a each a e an element g a e G a such that a,/3 e implies ir tg a = g^. = ga Addition is defined by adding functional values (g + g') a (7. Then G is a group uniquely isomorphic to each G a under the projection
ft
Ga
a
M
M
+
9*9a.
If
M
is
the collection of
all
spaces in
Ot
each consisting of a single
18
point, the
AXIOMS AND GENERAL THEOREMS
[CHAP.
I
Ga
are the
th
homology groups
of these spaces,
and the
TT
are the isomorphisms induced by the unique maps of one such space into another, then the above conditions are satisfied and the associated
group G could be taken as the coefficient group. DEFINITION 6.1c. The group #(P ) is called the coefficient group of the cohomology theory and is denoted by (?; its elements by g,g', etc.
7.
THE REDUCED 0DIMENSIONAL HOMOLOGY GROUP
DEFINITION 7.1. If x e and g e G, let (gx) x denote the image of H (X) under the homomorphism induced by the map /: P X of G in H (X) under / # is denoted defined by /(P ) = # The image
g in
X
by (Gx) x
.
THEOREM
/*(*)* =
7.2.
(ffy)y.
// /: X > Y, x e X, y = /(x), and Thus f+ maps (Gx) x onto (Gy} Y
.
g e G, then
The proof
space
is trivial.
DEFINITION
7.3.
If
X
the unique
>
is
said to be collapsible.
:
morphism / #
H
Q
(X)
G
>
is
P is admissible, the In such a case the kernel of the homodefined. It is called the reduced 0map
/:
>
X
dimensional homology group of X, and
is
LEMMA
7.4.
// /:
the
X
is
denoted by ff Q (X).
Y
is
admissible and
Y
is collapsible, then
X
is collapsible.
If (X,A) is admissible, and
X
is collapsible,
then
A
an
is collapsible
and
map (X,A)
a
trivial
is
The
is
first
statement
If
(P ,P ) is admissible. consequence of property
is
(3) of
>
admissible category.
admissible.
>
(X,A)
admissible, the inclusion
If,
is
in addition,
X
A
map A
X
collapsible,
A
X
P
admissible; so
is collapsible.
).
the composite map P is adSince
X
missible, so also is (X,X) (P ,Po) is admissible.
* (P ,P
first
Therefore (X,A) > (X,X) >
of admissible cateis
REMARK.
For each of the
is
two examples
gories given in 1, it However in the third
true that each admissible space
collapsible.
example GL LC of locally compact spaces, the only collapsible spaces are the compact spaces. THEOREM 7.5. // P is a space consisting of a single pointj then
#
(P)
(GP) P This follows from 5.1 since P
(P)
= OandH
7.6.
=
.
>
P
is
THEOREM
into
the,
//
X is collapsible and x
a homeomorphism. e X, then (X) decomposes
H
direct
sum
(Gx) x
and
the correspondence g
>
(gx) x
maps G
isomorphically onto (Gx) x
.
7]
REDUCED 0DIMENSIONAL HOMOLOGY GROUPS
Let f l
:
19
let
PROOF.
P
is
>
the composition of  G is the the homomorphisms /,,: G > (X) and / (X) Therefore / G isomorphically onto (Gx) x and / 2jt identity. maps maps (Gx) x isomorphically onto G. Since fi<*(X) = kernel / 2jp it follows that ff (X) and (Go?)* have only zero in common. If h e ff (X),  fe'. It follows immediately that let h' = / u / (A) and A" = A A' e (Gx) x h" e A". (X), and h = h'
. ,
X
X
>
P
be defined by /^Po)
=
x,
and
/3
:
Since f 2 fi
the identity
#
map
of
P
:
2jr
#
lj(t
,
2jt
,
#
+
THEOREM 7.7. // /: Men X is collapsible and
isomorphically onto (Gy) y
.
X >
f^
F,
F is
collapsible,
maps
lJ (X) into
8
X, and ?y = f(x), (Y) and maps (Gx) x
x
e
:
F^P
sition
PROOF.
.
Let
f,:
l
P
is
Then f 2 ff
G > (Gx)^ ^ each homomorphism into / 2 / collapses
be denned by ^(Po) = x and let / 2 the identity map of P Therefore the compo*
.
X
(Gj/) y
is
 G
is
the identity.
By
3)t
definition
and
7.2,
X
P
,
onto; therefore each is an isomorphism. Since = /^(kernel / 2 = (X) = kernel (/ 2 /)
#
J
DEFINITION
7.8.
If /:
X
>
F
and
F
3
Q
(X)
into #
is collapsible,
the
map
of
(F), defined
7.9.
Kernel /, = kernel f+. DEFINITION 7.1c. If x s X, h e H(X), and /: P > is defined by /(P ) = x, let /*(ft) G be denoted by h(x), and let the kernel of /*: H\X) + G be denoted by X (X).
by /^
(see 7.7), is
denoted by
/,,.
COROLLARY
X
8
THEOREM
7.2c.
// /:
X

F, x
e
<A6n (f*K)(x) TAus /* A(y). contains the kernel of /*.
=
X, y
V
=
/(x),
and A
s
maps
8
tf(F),
(Y) into ff t (X), and
#
y
(F)
X > P is admissible (i.e. X is collapsible), If /: 7.3c. G in HQ (X) under /* is denoted by G*. The factor group H(X) = H(X)/G X is called the reduced 0dimensional cohomology group
DEFINITION
the image of
ofX.
THEOREM
7.5c.
#(P) = G P
THEOREM
If
//
and ff(P)
7.6c.
P is a = 0.
space consisting of a single point, then
into the direct
sum
.
8
X (X)
morphism
then
G ~ Gx
7.7c.
X is collapsible and x e.X, then H(X) decomposes + G x and the map X P indwes an iso,
THEOREM
X
///:
X *
F,
F is collapsible,
x
e
is collapsible
and /* maps ffl(V)
into ffl(X)
X, and y and maps
==
/(x),
iso
GY
morphically onto GX.
#(F)
X F and F is collapsible, the map of If /: 7.8c. ff(X) induced by /* is denoted by /*. COROLLARY 7.9c. Under the natural map H(Y) > 8(Y) the kernel
DEFINITION
into
of /* is
mapped
isomorphically onto the kernel of /*.
20
AXIOMS AND GENERAL THEOREMS
8.
[CHAP.
I
THE REDUCED HOMOLOGY SEQUENCE
For each space
i:
>
X and each integer H^X^X) = X X and j: X (X,X) be inclusions. PROOF. Let sider the section of the H.S. of (X,X) around H (X,X):
LEMMA
8.1.
q,
>
q
0.
Con
Since each
=
kernel
H
0. image i^ q = kernel j^ q Since, by exactness, image j+ = kernel d, it follows that H^X^X) = 0. We shall give a second proof of 8.1 which uses the Excision axiom, and uses only the part of the Exactness axiom which asserts that = 0 This proof will be of importance later when the full Exactness J#** axiom is not available. The inclusion map (0,0) C (X,X) is an excision and therefore ~ Q (0). In the homology sequence of (0,0) we have the 9 (X,X)
q
H
(X)
=
i^^ =
i#
is
an isomorphism onto, image d. Therefore
.
it
Q
follows
(X,X) = kernel d. Therefore, image j+ =
H
by exactness that
Similarly
H
homomorphisms
J5T f (0)
^H
(G)
q (0)
^
ff f (0).
i+
Since
i
=
and j are identity maps, we have
0, it
=
follows that
H
g
=
j+
=' identity.
Since
0.
hH
THEOREM 8.2. // X is collapsible, then d maps H^X.A) into 8<>(A). PROOF. By 7.4, the map /: (X,A) > (P ,Po) is admissible. If (X A) then/^/i) ^H^P^P,) = Oby 8.1. Therefore (f\A)J(h) =
l
9
9
Hence, by definition, d(h) e ff (A). If /: is 8.3. (X,A) > (P ,Po) is admissible (i.e. collapsible), the kernel of /^ is called the reduced homology sequence of (X,A). THEOREM 8.4. The reduced homology sequence of (X,A) differs from
^/* (A)
0
=
DEFINITION
X
the
homology sequence of (X,A) only in that
the section
of
^e
latter
Aas been replaced by
J
where J^,i,,3 are
<fte
mops
defined by
j^i^d
respectively.
8]
REDUCED HOMOLOGY SEQUENCE
21
COROLLARY 8.5. // /: (X,A) > (F,J5), then / maps the reduced homology sequence of (X,A) into that of (K,#). PROOF. By 8.1 and the Dimension axiom, the H.S. of (P ,Po) has just two nonzero terms: H Q (P Q ) ^ // (P ). Therefore the kernel of
f^
coincides with the H.S. of (X,A) except for the two corresponding terms which, by 7.3, are ff Q (X) and H Q (A). The corollary follows from
7.7.
THEOREM
(i.e.
8.6.
X is collapsible,
the
// the unique
see 7.4),
map
is
A
f: (X,A) > (P ,P ) is admissible nonvacuous and x e A, then the H.S.
of (X,A) decomposes into the direct sum of two exact subsequences: (1) the reduced H.S. of (X,A) (i.e. the kernel o// and (2) the isomorphic +slt ),
image of
defined by
H.S. of (P ,P
)
under
g(P ) Observe that g
=
g^
where
g:
(P ,P
>
)
* (X,A)
>
is
x.
is
admissible since (P 0) P
)
(A, A)
(X,A) are
admissible.
Since fg is the identity map of (P ,P ), this theorem is a special case of the following one. THEOREM 8.7. Let f: (X,A) > (F,B) and g: (Y,B)  (X,A) be
admissible and such that fg
is
the identity
map
of (Y,B).
Then
the
homology sequence of (X,A) decomposes into the direct sum of two exact subsequences: (1) the kernel of /,,, and (2) the isomorphic image of the homology sequence of (Y B) under g^^. This theorem in turn is a consequence of a purely algebraic proposit
tion:
LEMMA
f
\l/
:
8.8.
C'
>
C
// C,C" are two lower sequences and > C' are homomorphisms such that C'
\l/:
C
>
C",
W:
is the identity,
then
of $,
1
C
decomposes into the
direct
sum
of two subsequences:
r
(1) the kernel
and (2) the isomorphic image of C under f. If C is exact, then C and the subsequences (1) and (2) are exact. PROOF. Let K = kernel ^, and L = image ^'. Since ^' is the identity, ^' maps C' isomorphically onto L. For convenience, suppose C Q > CJ and therefore ^: C( > C a C,C" are so indexed that $ Q
:
.
If
/
e
L
= 0. Thus H 0. Let = *I^(c?). Then s L and ^ (Z) = WrfM = Suppose c e C This proves that Zis in jFC and c = k + ^ (c). Therefore = c C = Kg + L now that C is exact. This implies that Suppose 2 = (4>L) 2 Suppose that = 0); hence (4>\K) 2 = <M>a+i = (i.e. k z K q and <(&) = 0. Since C is exact, there exists ace C<,+i such that = &. Let c = &! + where = e Jff^! and e L^+j. Then <t>(c)  ^(fc,) lies in both K and L There*(*i) + *(Zi). Thus 0(Z,) = fore </>(Zj) = 0, and k = ^(fci). A similar argument shows that, if Z e L a
then
qy
there exists a
c'
e
CJ such that
Z Z
MKO
^(c
fl ,
;
)
=
/.
If
=
.
c'
=
Z
0,
and therefore
^
^
tt
(Z)
= L =
fl
0,
fl
a
A:
a,
I.
Q
.
fl
<t>
.
Zi
fc t
Z!
A;
fc
q
.
fl
22
AXIOMS AND GENERAL THEOREMS
<t>(l)
[CHAP.
I
and and
=
0,
then there
Since
is
an
Zt
e
L
q^
l
such that
that C"
then
<t>(li)
=
q
I.
Thus
K
L are exact. LEMMA 8.1c.
>
l
L ~
C', it follows
is
exact.
For each space
//
X and each integer
GA
8
#,
H
(X,X)
=
0.
THEOREM
8.2c.
X
is collapsible,
lies
in the kernel of
5:
H(A) 8*(A)*H
d:
H
l
(X,A).
If /:
Therefore
induces
a
homomorphism
(i.e.
(X,A).
8.3c.
> (P ,P ) (X,A) (see definition sequence
DEFINITION
is
admissible
X
is
collapsible), the factor
2) of the C.S. of
(X,A)
by the image
of /**
8.4c.
is
called the reduced C.S. of (X,A).
differs
THEOREM
The reduced C.S. of (X,A)
from
the C.S. of
(X,A) only in that the section
j*
i*
5
H(X,A)
>
H(X)  H(A)
*
H (X
l
9
A)
of the latter has been replaced by
r
H*(X,A) > fl(X)
to/iere J*,t*,5
**
5
 5(A) 
ff
l
(*,A)
are
</ie
cose/
</ien /** induces a homoC.S. of (F,B) inio that of (X,A). morphism of the reduced
COROLLARY
8.5c.
mappings induced by (X,A) > (F,B), // /:
(X,A)
>
j*,i*,& respectively.
THEOREM
8.6c.
// /:
(P Q ,P
)
is admissible,
and x
e
A,
then the C.S. of (X,A) decomposes into the direct quences: (1) the isomorphic image of the C.S. of (P ,P
(2) the kernel of g**
sum
of two exact subse)
under /**, and
gf(P
)
where
g:
(P ,P
)
>
(X"A) i
denned 6y
=
x.
the second maps isomorphically onto the reduced C.S. of (X,A) under factorization of the C.S. of (X,A) by the first subsequence. Thus the reduced C.S. of (X,A) is exact. THEOREM 8.7c. Let f: (X,A) * (Y,B) and g: (F,B)  (X,A) Then the be admissible and such that fg is the identity map of (Y,B). C.S. of (X,A) decomposes into the direct sum of two exact subsequences: (1) the kernel of g** and (2) the isomorphic image of the C.S. of (Y,B) under /**.
Furthermore
f
LEMMA
8.8c.
(Replace "lower" by "upper" in 8.8).
9.
HOMOLOGICALLY TRIVIAL SPACES
assumed to be
1.
All spaces in this section are
collapsible
(see 7.3).
However, the
results
stated
hold for noncollapsible spaces except
possibly in the dimensions
and
DEFINITION
9.1.
A
space
X
is
said to be homologically trivial
if
9]
HOMOLOGICALLY TRIVIAL SPACES
23
is
H
a
(X)
=
for q 7*
and
Jflf
(X)
=
0.
If
A
is
= for all q. said to be homologically trivial if Q (X,A) The reason for the exception q = in the definition is found in 7.6. If the coefficient group is not zero, (X) cannot be zero for a nonempty
H
not vacuous, (X,A)
H
space X.
THEOREM
9.2.
then the inclusion
map
is nonvacuous, // (X,A) is homologically trivial and > i: (X,0) induces isomorphisms 04,0)
A
V
THEOREM
X, morphisms
subset of
9.3.
H
Q
(A)
= H
Q
(X)
far all
q,
Conversely, if these relations hold, (X,A) is homologically
trivial.
// homologically then the inclusion map j: (X,0)
A
is
trivial
and
>
is
a norwacuous
iso
(X,A) induces
j+:
H.(X)
;V
# (X)
A
is
H (X,A) H (X,A).
9
forq^O,
trivial.
Conversely, if these relations hold,
homologically
trivial
is
THEOREM
dimension
9.4.
// X
is
homologically
subset, then the
boundary operator of (X,A)
and A is a nonvacuous an isomorphism in each
d:
d:
= H>(X,A) =
H
Q
(X,A)
//._,( A)
forq^l,
ff t (A).
is homologically trivial. Conversely, if these relations hold, and A are homologically trivial, so also is COROLLARY 9.5. // both
X
X
(X,A).
These
q,
results
= 0, thcn<l> + C + = C 0+1 = kernel <^ a+2 = 0. Since = 0, image <^ PROOF. Since C a+3 = 0, image <+, = 0. Therefore C Q+l = kernel + = image + 2 C LEMMA 9.7. If C = {C ,0 is an exact lower sequence, and, for some C a . 2 then C = 0. C +2 = C*+i awd0 .,: C i Q, 0+ = 0, it follows that image = 0. TherePROOF. Since kernel _i = image <+!. But image <<,+ = C^o+i implies fore C Q = kernel = (7^,. Therefore C,, = image a+ = 0. kernel 4> +i
Q
LEMMA 9.6. C = and C
//
+*
depend on two algebraic propositions. C = C ,0 a is an exact lower sequence, and, for some
fl
j
9
3
:
2
.
fl+ ,
Q
<t> Q
}
<t> Q
.
Q

2
:
a
a
a
,
Q
<
a
<t> q
</>
2
Q
i
PROOFS OF
9.2, 3,
AND
4.
The hypothesis
of 9.2
is
that every third
term of the H.S. of (X,A) is zero (i.e. H a (X,A) = 0). By exactness and 9.6, this implies that the remaining adjacent pairs are isomorphic under the mappings of the sequence. The converse of 9.2 is obtained
24
AXIOMS AND GENERAL THEOREMS
[CHAP.
I
from
9.7. Theorems 9.3 and 4 are proved in a similar way using the reduced H.S. of (X,A) instead of the H.S. of (X,A). The exactness of the reduced H.S. was proved in 8.6. is said to be cohomologically trivial if DEFINITION 9.1c. A space
X
H
Q
(X)
=
for q
^
and B*(X)
if
=
0.
If
said to be cohomologically trivial
H
q
(X,A)
all q.
A =
is
nonvacuous, (X,A)
for all
q.
is
THEOREM
vacuous, then
hold,
9.2c.
i*:
// (X,A)
is
Q
cohomologically trivial
and
A
is
non
H*(X)
//
~H
is
(A) for
trivial.
Conversely, if these relations
(X,A)
is
cohomologically
9.3c.
THEOREM
subset of
A
j*:
J*:
cohomologically trivial
and
is
a nonvacuous
X, then
H'(X,A) =H<(X)
H\X,A) 
forq^O,
trivial.
ff(X).
Conversely, if these relations hold,
A
is
cohomologically
THEOREM
subset, then
9.4c.
//
X is cohomologically trivial and A is a nonvacuous
5:
6:
H' (A) ff(A)
l
H (X,A) H (X,A).
9
l
forq^
1,
is cohomologically trivial. Conversely, i$ these relations h^ld, COROLLARY 9.5c. // both and A are cohomologically trivial, so also
X
X
is
(X,A).
10.
THE HOMOLOGY SEQUENCE OF A TRIPLE
(X A,B)
9
Suppose that
i:
X D A D B and that the inclusions (X,B) C (X,A) (A,B) C (X,B), j:
are admissible.
inclusion
Then (X,A,B) is called an admissible triple. The and boundary operators associated with the pairs maps (X,A),(X,B),(A,B) are denoted by
i:
i':
i":
A B + B 
X, X,
A,
j:
j>:
X X
A =
j":
> (X,A), + (X,B), > (A,B),
i
d:
d':
d":
H (X,A) H (X,B) H (A,B)
q
q
Q
~> > *
H H
H^(B)
9
Q .,(B).
A
comparison of these maps with
and j suggests
called the
DEFINITION
triple
10.1.
d
j'd
is
boundary operator of the
(X,A,B):
d:
H.(X,A) *
H.AAM.
10]
HOMOLOGY SEQUENCE OF A TRIPLE
of groups
25
The lower sequence
...
is
d
4
H^A.B)  H
is
J*
g
**
(X,A) < H.(X,B)

H (A,B)
q
<


called the homology sequence of the triple (X,A,B).
It is
indexed so
that
H
(X,A)
the
th
group.
THEOREM 10.2. The homology sequence of a triple is exact. The proof of this theorem is a lengthy and tricky exercise in the use of axioms 1, 2, 3, and 4. The reader may save himself the effort
of following the proof
if he is willing to replace the Exactness axiom by the stronger proposition 10.2. It reduces to the former when B = 0. PROOF. Reference to the diagram below will assist the reader in
following the proof.
*'*
H
q
(A)
>
H
Q
(X)
1*
H.(A,B)
H.(X,B)
Id'
Commutativity relations hold in each square of the diagram. In addition to the homomorphisms displayed in the diagram, we have:
The proof breaks up
(i)
into the proofs of six propositions:
JV'*
If
=
e
o,
(2)
(4)
x
H,(X,B)
r
that ^o:'
(5) If
=
y
= and j> =
dj;
o,
0,
(3)
i;a
=
o.
e
there exists an x'
H,(A,B) such
x.
e
#<,(A ,A) and dy
=
0,
there exists a
t/'
e
H
q
(X,B) such
such
(6)
If z
that dz'
=
= H^A.B)
z.
and ^0
ji:
=
0,
there exists a
2' e
tf a (X,A)
PROOF OF
(1).
The map
(A,B) +
(X A) can be
9
expressed as
the composition of the inclusion
maps
k:
(A,B)
C
(A, A)
and
/:
26
AXIOMS AND GENERAL THEOREMS
[CHAP.
I
(A, A) (X,A). that jjt = 6'0*
C
By
PROOF OP
is
(2).
exact, j'^'^
=
= 0. Therefore k+ = 0. It follows =_W* = 'A = 0. = /'#* = /##'. Since the H.S. of (A,B) dj+
8.1,
H,(A A)
9
0.
PROOF OF
exact, i+d
(3).
0.
=
i+d
=
~ij'+d
= j&d.
Since the H.S. of (X,A)
is
PROOF OF
(4).
Since j^x
=
0, it
follows that
i'tf'x
=
dj+x
=
0.
By
of
the exactness of the H.S. of (A,B), the element d'x of the kernel is the image of an element i"+
xl
eH
=
g
(A,B),
V'xt
== d'x.
Then
d'(x

^Xi)
d'x

d^ =
/> =
x
d'x

d x,
ff
=
0.
i^Xi of the
the exactness of the H.S. of (X,B), the element x the kernel of d' is the image of an element
By
u
e
H (X),
q

t^!.
Then
by (1) and hypothesis. By exactness of the H.S. M of the kernel of j+ is the image of an element
v e
of (X,A), the element
H
q
(A)
y
i^v
=
w.
Define
Then
=
PROOF OF
(5).
i^X!
+ y^w =
i X! #
+
x
i^x,
=
x.
Since
of j'+
by the exactness of the H.S. of (A,B), the element dy of the kernel is the image of an element
u
e
10]
HOMOLOGY SEQUENCE OF A TRIPLE
9
27
Then, by exactness of the H.S. of (X A), i+d
i
=
0.
0.
So
iu
=
V
*w
=
i+dy
=
By
is
exactness of the H.S. of (X,B), the element u of the kernel of i^ the image of an element
y
l
H.(X,B),
d'y,
=
u.
Then
=
By
of d is the
fly
d/*2/i
=
By
t^'d'2/!
=
fy
 i> =
0.
exactness of the H.S. of (X,A), the element y
j^y^ of the kernel
image
of
an element
V*H
Define
g
(X),
jj>
=
y

J+VL
V'
=
2/i
+ J>==
Then
J*y'
=
.7*2/1
+ y*y> =
Since i^z
^2/1
+ j>
^2/1
+
y

jf^yi
=
2/.
PROOF OF
(6).
=
0, it
follows that
By
exactness of the H.S. of (A,/?), the element z of the kernel of d"
is
the image of an element
Then
J
=
W> = V =
t> =

By
is
exactness of the H.S. of (X,B), the element i^u of the kernel of j+
the image of an element
v e ff a i(5),
i+u.
Since t^t^
=
t^, it
l '*(
follows that
/
u
~
*
t;
)
=
l
*u
""
'
** t *v
/
= ^^
"""
**v
^
0.
By
of i+
exactness of the H.S. of (X,A), the element u is the image of an element
i'^v
of the kernel
z'*H.(X,A),
dz'
=
u

t>.
28
AXIOMS AND GENERAL THEOREMS
[CHAP.
I
Then
~A~t uZ
_
J +OZ
Vt\9 9
J +(U
fill
_ Wl I +V)
>
J
i ff<ll
jU
_
J
I"*"*)
j.1
+V
J #U
1*"?/
9 2.
This completes the proof.
A map
map
of
/:
(X^A^BO
Z!
into
X which carries
(X,A,J?) of one triple into another is a A l into A, and 5, into B. The map /
defines
maps
The map / is admissible if / lf / 2 and / 3 are admissible. In this case, the induced homomorphisms /i # / 2 *, and / 3jt map the groups of the H.S. of (X^A^BO into those of the H.S. of (X,A,B).
,
,
(X,A,B) induces a homomorphism homology sequence of (X ,A l ,B ) into the homology sequence of (X,A,B). The proof requires only the verification of three commutativity relations and is left as an exercise for the reader. THEOREM 10.4. // B is nonvacuous, and any one of the three pairs
10.3.
>
THEOREM
A map /: (X^A^BO
of the
l
l
(XjA),(XjB),(A,B) is homologically trivial, then the homology groups of the remaining two pairs are isomorphic under the maps of the H.S. of
(X,A,B).
(1)
9
Explicitly:
(2)
(3)
IfH (X,A) = Ofweaehq,ihmi+: //Hf (X,B) =0 for each q,thm d: IfH Q (A,B) = for each q, then j+:
p
H (X ,B) for each J9T,(A,B) H (X,A)  H^(A B) far each H (X,B) ~ H (X,A) for each
q 9
9
q.
q.
q
q
q.
Conversely
hypothesis.
,
any one of
is
the three conclusions implies the corresponding
The proof THEOREM
10.5.
the same as that of 9.2 except for notation. Let (X,A,B) be an admissible triple. If
B C A
ifACX induces H (A) H (X,B) /or induces H (A,B)
larly,
q Q f
= induces isomorphisms Q (B) (X,A) also induces isomorphisms
H
H (A) for all values of then H (X,B) ~ H (X,A) for all
q
g,
Q
q
(X,B) Simiq.
C
ff.(X) for
all q,
then (A,B)
C
(X,B)
off g.
PROOF. For the first assertion, the hypothesis and the exactness of = for all q. The result follows the H.S. of (A,B) imply that q (A,B)
H
now from
10.4.
is
Another proof (X,B) C (X,A).
obtained by applying 4.2 to the inclusion
is
map
The
proof of the second part of the theorem
similar to that of the
first part.
11]
HOMOTOPY EQUIVALENCE, CONTRACTIBILITY
10. Ic.
5
29
DEFINITION
triple
=
dj"*
is
called the coboundary operator of the
(X,A,B):
:
H
Q
\A,B) >
H (X,A).
Q
The upper sequence
>
is
of groups
*
H
q
~\A,B)  H'(X,A)
10.2c.
10.3c.
J*
i*
 #<(*,) 
#V,)
>
called the cohomology sequence of the triple (X,A,B).
THEOREM THEOREM
homomorphism
of the
10.4c.
THEOREM
The cohomology sequence of a triple is exact. A mapping f: (X^A^BJ  (X,A,B) induces a C.S. of (X,A,B) into that of (X^A^Bj). // B is nonvacuous and any one of the pairs
,
(X,A),(X,B),(A,B) is cohomologically trivial then the cohomology groups of the remaining two pairs are isomorphic under the maps of the C.S. of
(X,A,B).
THEOREM
10.5c.
Let (X,A
q
,J5)
Q
be
an admissible
q,
q
triple.
H (A) ~ H (B) for all then H (X,A) = H (X,B) for all A d X induces isomorphisms H (X) = H (A) for all H (A,B). (X,B) induces isomorphisms H (X,B)
induces isomorphisms induces isomorphisms
9 q q q q
If
B C A
(X,A)
(X B)
y
C
q.
g,
Similarly, if then (A,B)
C
11.
HOMOTOPY EQUIVALENCE AND CONTRACTIBILITY
will
In this article use
axiom.
It is
be made for the
first
time of the
Homotopy
assumed here that all pairs, maps, and homotopies are in Q,. DEFINITION 11.1. Two pairs (X A) and (Y,B) are said to be homoy
topically equivalent (in
/:
Ct) if
there exist
maps
g:
(X,A) > (7,B),
(Y,B) > (X,A)
such that gf is homotopic to the identity map of (X,A), and fg is homotopic to the identity map of (F,B). The pair of maps f,g is called a homotopy equivalence. Frequently each of the maps f,g separately will be referred to as a homotopy equivalence.
THEOREM 11.2. A homotopy equivalence /: H,(X,A) = H g (Y,B), and (/ # )" = g^ for
1
induces
all q.
it
isomorphisms
PROOF.
identity.
Homotopy axiom
Since gf is homotopic to the identity, that (&/% = g+f* = identity. Therefore f+ has the inverse g+.
follows
from the
Similarly f+g+
=
THEOREM 11.3. A homotopy equivalence induces isomorphisms of the ordinary and reduced homology sequences of (X,A) with the corresponding
sequences of (Y,B).
SO
AXIOMS AND GENERAL THEOREMS
PROOF.
[CHAP.
I
homotopy equivalence f,g of (X A) and (F,B) defines homotopy equivalences of X and Y, and of A and B. Hence 11.2 implies that / induces an isomorphism of the H.S. of (XT, A), and that g induces the inverse isomorphism. The proposition for the reduced
y
A
H.S.
X is contraclible on itself X is homotopically equivalent to a point. PROOF. Let h(x,t) be a homotopy h: X X /
=
and
now follows from 7.7. LEMMA 11.4. // a space
to
a point, then
> such that > z for each x e X. Let /: (x ), and #, h(x,l) h(x,Q) = x Then /# = identity map of (x ) > define g: (x ) by g(x Q ) and A is a homotopy connecting gf and the identity map of X. Therefore f,g form a homotopy equivalence. THEOREM 11.5. Every space contractible to a point over itself is
X
=
X
X
.
homologically
trivial.
PROOF.
11.4
and 11.3 show that
Q
it
suffices to
prove that a space
consisting of a single point
and by the Dimension
logically trivial.
P is homologically trivial. By 7.5, /? (P) = for q ^ 0; hence P is homoaxiom H (P) =
11.6. pair (X',A'), contained in a pair (X,A), is r of (X,A) if there exists a map /: called a retract (X,A) > (X',A ) such x for each x s X'. It is called a deformation retract if there that /(x)
DEFINITION
A
=
is
a retraction / and the composition of / and the inclusion map (X,A) is homotopic to the identity map of (X,A). It is called a strong deformation retract if the latter homotopy can be chosen to leave fixed each point of X', i.e. h(x,t) = x for x e X' and all t. (These
(X'j A')
C
depend on the category Ot since all maps must belong to ft). // (X',A') is a deformation retract of (X,A), the in(X',A') C (X,A) and the retraction f form a homotopy equivalence of (X,A) and (X ,A'). By the definition of a deformation retract, gf is homotopic to the identity, and fg is the identity. THEOREM 11.8. // (X',A') is a deformation retract of (X,A\ then f the inclusion map (X',A ) (X,A) induces an isomorphism of the H.S.
definitions
LEMMA 11.7. clusion map g:
f
C
of (X',A') onto that of (X,A).
/*:
THEOREM H\Y,B) THEOREM
This follows immediately from 11.7 and 11.3. 11. 2c. A homotopy equivalence induces
1
the
H*(X,A), and (J*) = g*, for all q. 11. 3c. A homotopy equivalence induces isomorphisms of ordinary and reduced cohomology sequences of (Y,B) with the corre11. 5c.
trivial.
isomorphisms
sponding sequences of (X,A).
THEOREM
Every space contractible
to
a point over
itself is
cohomologically
12]
EXCISION
11. 8c.
f
31
THEOREM
the inclusion
map
the
of
(X,A) onto
) is a deformation retract of (X,A), then (X,A) induces an isomorphism of the C.S. C.S. of (X',A ).
// (X',A
r
(X',A
)
C
f
12.
EXCISION
The
phisms
excision
valid.
Excision axiom asserts that an excision
of the
map
induces isomorInt
homology groups.
relaxed to
it
If
the condition
U C
is
A
on an
map
is
U C
A, then the conclusion
not generally
shall give
However,
here.
does hold in certain useful cases.
discussion of the various
We
two such
A
full
ways
of strengthening
the Excision axiom will be given in x,5. THEOREM_12.1. Let (X,A) be a pair, and
let
U,V
be open subsets of
X such that V C U C
(X
y
A,
the inclusion
maps
g
 UA V).
f U) > (X
U,
 V,A U)
is
V) > (X,A)
are
admissible,
(X
the
V,
A
and (X Then
all
is
A
the inclusion
map
a deformation retract of gf induces isomorphisms of
isomorphic.
homology groups in
dimensions.
is
PROOF.
is
Since g
X
(i)
likewise isomorphic. Hence (gf)+ = g+f+ is isomorphic. be an open subset of THEOREM 12.2. Let (X,A) be a pair and let there exists a subset B of with C. A. containing A such that If
an excision map, g+
By
11.8,
/,
U
U
X
the inclusion
maps
I
\k
m
(X
are admissible, (U)
(iv)

U,B

U)
is
>
(X,B)
A
U C Int B, (Hi) A U is a deformation retract of B
a deformation retract of B, and U, then f induces isomorphisms
in
all
dimensions.
(ii)
The Excision axiom and From (iii), (iv), and 11.8 it morphism.
PROOF.
induce isomorphisms H,(A)
imply that
m+
is
an
iso
follows that the inclusion
H
q
(B)
From
10.5,
it
now
follows that 1+
l
H q (B and H,(A  tf) and k+ are isomorphisms.

maps
U).
This
= k+ m+l+ is an isomorphism. implies that / # is a strong deformation retract of It is to be noted that if A
U
92
AXIOMS AND GENERAL THEOREMS
[CHAP.
I
B V can be extended to [/) X / ;> A U, the homotopy h: (B a homotopy B X / > 4 by defining /i(x,<) = x for x e J7. If the extended homotopy is admissible, then A is a strong deformation retract of B. This is the case for the main applications of 12.2. The theorems of this section hold as stated with cohomology in place
of homology.
13.
THE DIRECT SUM THEOREM
diagram
LEMMA
13.1.
In
the
,
t
G
t
and homomorphisms, assume that commutativity holds in each = kernel j a (<* = 1,2), and ki,k 2 are isomorphisms. triangle, image i a i ly i2 are the components of an injective direct sum representation Then i: GI + G2 ~ G, and j\J 2 are the components of a protective direct sum
of groups
representation j:
G ~
G{
an isomorphism, the image of i 2 isomorphically. Thus (kernel Since image i2 = kernel j2 it follows that
Since jii2
ki is
,
PROOF.
=
+G
2.
it
follows that j\
maps
j\)
C\ (image
i2 )
=
0.
(i)
(kernel j'O C\ (kernel j 2 )
If g e G, let
l
=
l
'g
=
l
i\k 2 j 2 g
+
l
i 2 ki
jig.
kiki jig
=
j t g and similarly j^g
=
j 2 g.
proves that g has a representation g
representation,
=
Then jjg = j\i2 ki Thus g = g by (i).
i^i
jig
=
This
k 2 j 2 g.
l
l
+
=
i2 g 2 .
For any such
we have
Similarly g 2
l
To prove
i\k 2 g 2
.
hence the representation is unique. the second part, consider g{ e GJ, g 2 e G 2y and g = l l Then j\g = jii2 ki g( = k l ki g{ = g{ and similarly
e
= k^j^,
j2 g
=
j2 iigi
+
J2 i2 g 2
k 2 g lt thus g l
=
i2 ki g(
This exhibits an element g
ness of g follows from
(i).
G
with
jagf
=
n g'
(a.
=
^=
+
g*.
1,2).
The unique
The proof as given remains valid if the hypothesis image kernel ji is weakened to image i\ kernel ji. The relation image ii = kernel j l is a consequence.
ii
REMARK.
=
C
13]
DIRECT SUM THEOREM
13.2.
SS
THEOREM
sets
Let
X
=
X W
l
WX
n
be the union of disjoint
each of which
is closed
A =
and
AI
U
U
:
(and therefore open) in X.
that all pairs
all
Let A,
C
An
.
Assume
formed
of the sets
X, and X,,A
%
their
unions are admissible and
Let
ia
inclusion
admissible.
homomorphisms
sentation of
H
i a +:
(X a ,A a ) C (X,A) (a = H q (X a ,A a ) * H q (X,A)
direct
maps
1,
of such pairs are
,
n).
Then
the
yield
an
Q
(X,A) as a
expressed uniquely as u
=
]T] a
PROOF. The theorem is trivial for n = 1. Assume the validity of Xn _ l9 let A' = 4, the theorem for n  1. Let X = X, \J A n , ly and consider the diagram of inclusion maps
1
sum, i.e. each u in Q (X,A) can be i a *u a9 where u a e Q (X ai A a ).
H
injective repre
H
U
U
U
(X,X n
V A)
(X,A)
(X.X'UA
n)
(X'VA
\h'
n ,A)
(X*V
A,A)
The maps hny
the triples
k n hn
,
h'\ k'h'
are excisions, and therefore hnif
,
A'
njc
,
h^
k
are
isomorphisms.
9
Further, the exactness of the homology sequences of
(X X'
UA
=
n ,A)
and (X,Xn
U A, A) implies
kernel j
image l,
kernel j n +
= image
and
/
ln
^.
Thus the conditions
direct
i'\
of 13.1 are fulfilled
and
sum
f
representation of
(X',A ) direct sum. * q (X',A') where
C
H
(X,A). It follows that i and i n# where as a Q (X,A) (X,A), also yield a representation of Q (X a ,A a ) By the inductive hypothesis, the maps i^:
q
,
H
l^
n# yield
an injective
H
H
ii\
(X aj A a )
,
yield
*"*
an
injective representation n 1, ***'a* for i
C of H
1, it
(X',A')
Q
=
9
1,
,
n

1,
=
(X',A') as a direct sum. follows that i a+ (i = 1,
Since
,
n)
form an injective direct sum representation. THEOREM 13. 2c. Under the conditions of 13.2 the homomorphisms Q Q q i*: (X a ,A a ) yield a projective representation of H (X,A) (X,A) > Q as a direct sum i.e. for each sequence u a e H (X a ,A a ), a = 1, n, q n. there is a unique element u e (X,A) with i*u = u aj a = 1,
H
H
y
,
H

,
34
AXIOMS AND GENERAL THEOREMS
14.
[CHAP.
I
TRIADS
consists of a space
,
subsets
X and two X 2 and all X, \J X 2 X pairs formed from these are admissible, and all their inclusion maps are admissible. The triad (X\X ^X 2 ) is called proper if the inclusion maps X2 ) k2 2) 2 ,X 2 ), h: (X2 ,X, (X, (X,,X, X 2 ,XO induce isomorphisms of the homology groups in all (X, dimensions. Note that (X;X 2 ,Xi) is a triad distinct from (X;Xi,X 2 ) unless Xi = X 2 A triad may be proper with respect to one homology theory and not another; but certain triads are proper for all theories. For example, if
DEFINITION
14.1.
A
triad
(X;Xi,X 2 )
Xt,X 2
of
X
such that X, X,,
X
2,
x
H
l
:
n X
.
U
C
U X
H
C
X\
=
Xij
X
2
= X
2,
Xi
l
(Xi i\
2,
X
2)
C\
X
2
(Xi f\
X
2)
==
0,
where closure
is
taken in
X UX
then
fc t
and k 2 are excision maps,
and, by the Excision axiom, the triad is proper relative to any homology Still weaker conditions for independence of the homology theory.
theory can be based on the results of 12. Observe that, if (X;Xi,X 2 ) is a proper triad, then (X^
is
U
2)
X^X^X^
if the in
also proper.
THEOREM
clusion
g,
14.2.
ia
:
A
aj
triad
X X C\ X yield, for (X (X maps an injective representation of H (X U X Xj C\ X as a direct each H (Xi VJ X X P\ X can be expressed sum, if and only if every u H (X a ,Xi (^ X a = 1,2. as u = ii+Ui + uniquely +u for u
X
l
(X]X^X
2)
is
l
C\
X
2)
>
proper \J 2
if
and only
l
,
Q
l
2,
2)
i.e.
e
Q
2
,
l
2)
i2
2
a e
Q
2 ),
PROOF.
(X, \J
Consider the diagram
(X,
XM
29
U X ,X
2
2)
t
r\
ta
(X 2 ,x,
The
nx
2)
(x lt x> r\
x
2)
relation kernel j a +
l
=
image
,
ia
If the triad is proper, then H.S. of the triple (X VJ 2 ). l C\ 2 a and k 2 + are isomorphisms, and 13.1 implies the direct sum conclusion. ki+
,
X X X
^ follows
from the exactness
of the
X
Then
X
l
Conversely assume that the direct sum decomposition holds. i has kernel zero. Thus, in the H.S. of the triple (X! 2 t = 0, and therefore, by exactness, y is onto. C\ 2 ), we have d Now let u e Q (X VJ 2 ,XO. Then u = j\+v for some v e
ljt
UX X
,
,
X
ljt
H
l
X
14]
TRIADS
l
35
H
v
9
(X
UX
+
2,
Xi
nX
2 ).
By
the direct
=
sum assumption, we have
ii^Ui
iz+Ut so that
and
fc
1#
is
onto.
= 0. Assume u e H q (X 2f X, C\ Xa ) and fc 1% w = 0. Then Thus by exactness, there is a e H (X X C\ X 2 ) with i^v = i^u. Then ii^t;) + i^u = 0. By the direct sum assumption, this implies u = 0. Therefore k^ is an isomorphism. By symmetry, the same holds for A; Thus the triad (X]X lt X 2 ) is proper.
j^u
t;
(l
l)
l
.
2sl
DEFINITION
tion of the
14.3.
Let
(X\X^X
2)
be a proper
triad.
The composi
homomorphisms
9
,*,
UX ^
2)
H..!^,
UX ^
2)
fe,
^..(X,
U J^ X
2,
^
2)
*
HUX^ n
triple
where & 2 ,Z 2 are inclusion maps, is called the boundary operator of the proper triad and is denoted, ambiguously, by d. (Note that this new
boundary operator is just the boundary operator of the X 2 X 2 ) followed by an excision.) The lower sequence (X, Xi
U
,
^X^X, H
where
triad.
X,) < HJiX.X,
UX ^
2)
H.(X,XJ
^
H^X^X,
r\
X
2)
i,j are inclusions, is called the
homology sequence of the proper
triad is exact.
THEOREM
PROOF.
14.4.
The homology sequence of a proper
following diagram
a'
As shown by the
t'i
H.(X,XJ <
H (X, U X X
Q
2)
2)
<
H^XJt U J
2)
^
H^(X,X
2)
^.(Xt,^
the H.S. of (XjX^X.)
replacing
is
nJ
q
2)
obtained from that of (X, X,
the
H
q
(X lJ X
l
n
group X,) under
H
is
(X
l
k^
X a X,) by by the isomorphic group and defining a so that k^d = d'. Thus
,
U
}
U X ,^2)
2
the H.S. of (XjX^Xj)
Since the latter
is
exact, so
isomorphic to the H.S. of (X, is the former.
l
,
X UX
2
,.X a ).
Note
that,
if
a proper triad and
(X;X,,X a ) is a triad with X D X a then (X;Xi,X t ) is its homology sequence reduces to the homology
// (X;Xi,X a ) and (Y;Y lt Y 2 ) are proper triads, and
sequence of the triple (X,X!,X 2 ).
THEOREM
14.5.
36
AXIOMS AND GENERAL THEOREMS
(XiX^Xz)
>
1
[CHAP.
I
/:
(F;F ,7 2 ), then f induces a homomorphism f^* of homology sequence of (X;X ,X 2 ) into that of (F;F,,F2 ). In particular,
l
the the
commutes with induced homomorphisms. boundary PROOF. Since i^ j^, i, j are induced by inclusion maps, commutativity obviously holds in the squares not involving d. The latter square extends into
operator for proper triads
,
r\
X
2)
,
H
q
(X,
/2*
/'*
/3*
H^Y^Y,
where
right.
H
F
;
2)
^(F,
by
/.
U
F ,F )
2 2
<
H^Y.Y,
=
ljt
U
F)
2
/i,/ 2 ,/3 are defined
By
10.3,
commutativity holds on the
Since
/c,fc /2*&*fc^/ l 1 are isomorphisms, this yields /i^A;" = k^ f 2 ^.. THEOREM 14.6. Let (X;X ly 2 ) be a proper triad with
Since
are inclusions,
we have
k^k^
VJ
X
X
= Xi
X
2,
A = X
l
C\
X
2
.
Let /x,/ 2 ,/ be
maps (X,A) ~> (F,B) such
that
f*\xt
=
PROOF.
Consider the diagram
AT,
(Y,B)
where
f( is defined
by
/,.
in the lower
two
triangles;
Observe that commutativity does not hold however we have
MAYERVIETORIS SEQUENCE
If
37
we
pass to the induced
homomorphisms on the homology groups, we
observe that the part of the diagram not involving the group q (Y,B) is a diagram of the type considered in 13.1. Thus, for each u e H*(X A),
}
H
we have
Then
./>
= fjijblfau
+ f^Jc^u =
fijcijc^u
+
Sitk
and the proof
is
complete,
The
theory
definitions of a triad
and a proper
of 14.1.
}
triad relative to a
cohomology
for each
2)
is
an obvious analog
14.2c.
THEOREM
q, the
A
triad (X',X
,X 2 )
2,
is
proper if and only
2)
if,
homomorphisms H\
14.3c.
H
9
(X
l
UX
q
yield a projective representation
ofH (X
l
X  H^X^X, C\ X U X X C\ X as a direct sum.
X, C\
2,
l
2)
DEFINITION
P
The coboundary
operator
of
a
proper
triad
(Jf;X 1 ,X 2 )
is
the composition
*r
,,*! n j,)
!
>
1
tf'
^ux J
2,
ff
2)
5
* ^'(x,
uj
2)
>
H
f
(x,X!
and
is
is
also denoted
by
5.
The cohomology
sequence of the proper triad
the upper sequence
j*
i
r\
cohomology sequence of a proper triad is exact. (X]X lt 2 ) > (FjF^F,) is a map of one // /: triad into another, then f induces a homomorphism /** of the proper
14.4c.
T/ie
THEOREM THEOREM
14.5c.
X
cohomology sequence of
the second into that of the first. the conditions of 14.6,
THEOREM
14.6c.
Under
/*
=
/*
+
/*
15.
THE MAYERVIETORIS SEQUENCE OF A TRIAD
Let (X;Xi,Xt) denote a proper triad such that
(1)
X=
X,
UX
A = X,C\
X
2.
S8
AXIOMS AND GENERAL THEOREMS
appellation "MayerVietoris"
is
[CHAP.
I
The
usually applied to a formula ret
numbers (ranks of the homology groups) of X,Xi X 2 ,A. It provides a method of calculating the Betti numbers of a space obtained by assembling two spaces. A purely group theoretic generalizaIt is shown tion of this formula has been given by several authors.
lating the Betti
below that this group theoretic formulation can be simplified to the statement that a certain lower sequence is exact. The purely algebraic part of the construction is based on the following lemma, which we
refer to as the
"hexagonal lemma."
//,
LEMMA
phisms,
15.1.
in the following diagram of groups and homomor
commutativity holds in each triangle, k and k 2 are isomorphisms onto, image i a = kernel j for a = 1,2, andj Qi Q = 0, then the two homomorphv
isms of GO into G' obtained by skirting Q
only.
the sides of the
hexagon
differ in sign
Explicitly
hik^hg =
hjs^l 2 g
for each g
s
GQ
.
PROOF. By 13.1, G decomposes into a direct images of GI and (? 2 and for any g e G
;
sum
of the isomorphic
,
Applying jQ to both
tions,
sides,
and using jQ iQ
=
and commutativity
rela
exactness of the MayerVietoris based on the diagram below, in which all homomorphisms sequence other than d are induced by inclusion maps. We assume that relations (1) hold. Observe that the lower hexagon satisfies the hypotheses of 15.1.
is
we obtain the desired result. The construction and proof of
15]
MAYERVIETORIS SEQUENCE
39
H,(X,X
t
)
*,
**
DEFINITION
15.2.
The MayerVietoris
(X;X,,X 2 )
ti?t//i
X
= X!
A
WX
g
2
and
<t>
A =
Xt C\
sequence of a proper triad 2 is the lower sequence
X
t
.
* ff.^CA) 4
H (X)
4 ff f (XO
+H
(X f )
where A,0,^ are defined as follows
= (hnUth^u), = w^Vi + m ,f2) Aw = dikitli+w, = dzk^lt+w,
^u
24c
u
,
e
/
g
2 e Q
t; 2
Vi e
H (Xi), v H weH (X),
Q
(X 2 ),
by
sequence
15.1.
THEOREM 15.3. TAe MayerVietoris ;X,,X a ) w(A X = X VJ X 2 is exoc^.
l
of a
proper triad
PROOF. (1) If u
<t>\l/u
We must prove the usual &H (A), then
g
six propositions.
=
(t>(h^Uj
h^u)
= m 1+ fe,^w
m
2
^h 2 ^u
=
i^u
t
#u
=
0.
40
(2)
0.
AXIOMS AND GENERAL THEOREMS
If v
[CHAP.
1
I
=
(vi,v s ) eff c
(XO
+H
d
l
g
(X 2 ),
then,
by
exactness, Ii^m ^v
l
=
This implies
v,
=
k^l
litt
m
l
^v l
=
0.
By
15.1,
A =
A<y
d 2 k^l 2 *.
Therefore
=
Q
^(mifVi
+W
=
2jc
y2)
= 0. Thus Am = Am^Vi + Aw
z;
25t
2je
t; 2
=
0.
(3)
If
weH
(X),then
h 2 +&w
hz+dikilli+w
=
we obtain
similarly that
since h 2 ^di
/i
uA =
(4)
0.
If
= by exactness. Hence ^A = 0. w e H q (X) and Aw =
Using
0,
15.1,
then
0,
dkl\l a jjo =
a
that
=
1,2.
By
exactness, there exists v
e
H
Q
(X a } such
where n a becomes
:
Xa C
(X a ,yl).
By
13.1,
j^w
splits into a
sum which now
By 'exactness
of the H.S. of (X,A), there
is
a u
e
H
Q
(A) such that
If
we
let ^i
If v
=
v(
+
h^u and
?^
2
=
^2? it
follows that
</>?;
<t>(v }
,
v2)
=
w.
(5)
=
(v,,v 2 ) e
H^X,)
+
//,(X 2 ), and
=
0,
then
j*<t*>
=
y*^i*^i
+
z
j***"* = *'i*Wi
and
z' 2
+
^'2*^2^2
=
0.
By
13.1,
the images of
i a *n a ^v a
Therefore
zero.
= =
u
(a
=
(a
*
1,2).
have only the zero in common. Again by 13.1, t ajjc has kernel
Therefore n a4( v a
=
1,2).
ua
s
H
Q
(A) such that h a ^u a
=
v a (a
=
By
1,2).
exactness, there exists a
Therefore
Since
<^y
=
.
0,
by exactness there
.
is
an x
e
e
H q+ i(X,A)
(a
u
x
{
=
+
i}
u2
By
13.1, there exists x a
//,+,
(X a ,^4)
=
such that dx = 1,2) such that
^Xi
+
?/
i 2 ^x 2
Then
dx
=
d 2 Xi
Let w
=
5 2 x,.
Then u
=
(w 2
d,
2 ).
It follows
now
that
15]
MAYERVIETORIS SEQUENCE
41
hi+u
fore
=
\l/u
/fci^Wj
=
hi^dzXi
=
hi^Ui
=
vi.
Similarly h 2 ^u
=
v2
.
There0.
v.
(6)
If
u
e
H
q
(A) and $(u)
e
exactness, there exists x a
H
g
= 0, then h^u = ^(X ai A) (a = 1,2)
and h^u
such that
=
By
This implies
By
exactness, there
is
a
w
e
H Q+l (X)
such that
Then
since ki+
**
=
Ji^i 2jc
.
V
THEOREM
**
15.4.
*
Y
1
V ^ " v J 2?
Y V
This completes the proof of 15.3. // (X;X ,X 2 ),(F;F ,F 2 ) are proper triads
]
1
V
*
1
\ ) ^^
V
*
2)
n<nd / UflUt
J
fY'Y Y } ^yv,yVi,W2y
v '
(V'V ijt 2> V \
'

,
induces a homomorphism of the MayerVietoris sequence of (X;Xi,X 2 ) into
If
one observes that / induces a
map
of the
diagram preceding 15.2
into the analogous diagram for (F;Fj,F 2 ), it is seen that the desired commutativity relations follow from standard relations. The proofs
are
left to
the reader.
15.5.
LEMMA
// (X;X,,X 2 )
is
a proper
3
triad, then
commutativity holds
in the diagram
.A Uvr
\
2)
~*
rj *
/
a\^
Y V
1 j^* 1
/^\
'
'
V
\
^*2/
la.
where d
is the
boundary operator of
operator of the
the
H.S. of
the triad
(X;Xi,X2 ) and
A is the boundary (X \J Xa ;Xi,X t).
t
MayerVietoris sequence of the triad
PROOF.
Consider the diagram
H.+^X.X!
UX
2)
^ H.(X
l
WX
k^
<
2)
A* #.(,
W X ,X
2
2)
dt
H,(x
lt
x
l
r\ x,)
 H..,^
n zo
42
AXIOMS AND GENERAL THEOREMS
12
[CHAP.
I
where
and k t are inclusion maps.
Then, by 14.3 and
15.2,
which
is
the desired result.
We now
proper triad
definition
proceed to define the relative MayerVietoris sequence of a = The 2 ). (X;Xi,X t), (it is not assumed that
X
X UX
t
and proof of exactness are based on the diagram below, to the one preceding 15.2. We note that d, j^ are two conanalogous secutive homomorphisms in the H.S. of the triple (X,Xi \J X^,X (~\ X t ), and therefore the lower hexagon satisfies the hypotheses of 15.1.
t
H,(X,X
t
I
H,(X,X,)
[t,
H,(X,X
t
)
H,(X,X, \J X,)
H,. (X
t
t
UX
H*}(Xl \J
^2>
X,,X,)
^1
I
I*.*
*>.
I
NX
V*
H*}(XI,X\ r\ xj)
DEFINITION
triad
15.6.
is
(X;Xi,X 2 )
The relative MayerVietoris sequence of the proper the lower sequence
15]
MAYERVIETORIS SEQUENCE
follows:
i),
where A,<,^ are defined as
,&,
v,
=
THEOREM
is exact.
U^H (X,X r\ x H (X,X,), v H.(X,X weH^XtXtVXJ,
Q
I
2 ),
Q
2
2 ),
n^k^w,
The
relative
by
15.1.
15.7.
MayerVietoris sequence of a proper triad
The proof is analogous The same applies to
to that of 15.3
and
is
left to
the reader.
// /: (X;X l9 2 ) * (Y;Y l9 Y 2 ) is a mapping of then f induces a homomorphism of the relative Mayerproper triads, Vietoris sequence of (X;X,,X 2 ) into that of (Y;Y lf Y z ).
THEOREM
15.8.
X
DEFINITION
proper triad
15.2c.
The MayerVietoris cohomology
with
(X;X,,X 2 )
X
(/>
X
l
VJ
X A = X
2,
l
C\
X
sequence of a 2 is the upper
sequence
A
where
 H\A) * H'(X) 
7/
f
(X
1
)
+
7/'(X 2)

^
//'(A)

= * * 5 w, = (m^,m^), <t*w = vi  >^2, *(i, )
1
2 /b 2
2
by
tr e
i
15.1,
H
q
(X),
v 2 e ff
a
e
^(XO,
(X 8 ).
triad
THEOREM 15. 3c. The MayerVietoris cohomology sequence of a (X\X^X ) with X = X, U X is exact. THEOREM 15.4c. // /: (X;X X 2 ) > (Y;Y Y ) is a map X U X2 Y = Y U F2 proper triad into another, and X
2 2
lt l9
proper
of one then f
2
l
,
}
,
induces a homomorphism of the MayerVietoris cohomology sequence of the second triad into that of the first. LEMMA 15.5c. // (X;X,,X 2 ) is a proper triad, then commutativity holds in the diagram
,
C\
where
8 is from the C.S. of the triad Vietoris sequence of the triad (X\ VJ
X
(X;X ,X 2 ) and A 2 Xi,X 2 ).
}
is
from
the
Mayer
;
44
AXIOMS AND GENERAL THEOREMS
DEFINITION
15.6c.
[CHAP.
I
The
relative
of the proper triad
(X;X,,X
2 ) is
MayerVietoris cohomology sequence the upper sequence
A

> H*\X,XI r\ x,) >
H\X,XI
u
where (using the diagram preceding
15.6)
Aw
= =
l
5 2 fcr
n* w, 2
u
ti>
e
H
q l
(X
J
X
l
Mtf'Vify,
s
H^X.Xi
U
THEOREM
a proper
15.7c.
The
relative
MayerVietoris cohomology sequence of
triad is exact.
15.8c.
> (Y;Y l9 Y2 ) is a mapping of (X\X lt // /: proper triads, then f induces a homomorphism of the relative Mayer
THEOREM
X^
Vietoris cohomology sequence of
(Y\Y l} Y^)
into that of
(X]Xi
9
X
2 ).
16.
n
CELLS AND SPHERES

xn ) will Let R be the euclidean nspace. Coordinates (x sometimes be abbreviated by the vector symbol x, and the norm of x is defined to be \\x\\ = (]? xj)*. Each of the following six symbols will n denote the subset of R defined by the algebraic condition to its right:
t
, ,
ncell
(n
IWI
l)sphere S"
l
^
1*11
1
upper cap lower cap
(n
(n
ET T
/S"'
\X\\

x
2
2)sphere
l)cell
l
En
1
~
x
= = =
1,
1,
1, 1,
1
x
n~ 2
Xn xn xn xn
^ g = =
1
.
0,
0, 0,
0.
Clearly, S*~ = S consists of
l
l
= ET S" and S two points, one of which is E+ and the other E. We also define S~ to be the vacuous set. 72 and E are single points. n n~ l be the euclidean subspace of R defined by xn = 0, and Let R
ET
U ET
H
let
/ be the projection of
/(.,
Rn
,
into
av)
Rn = (*.,
:
~l
'
'
'
,
.., 0).
The
function / defines
maps
and /_ are homeomorphisms.
It is easy to verify that f+
16]
CELLS AND SPHERES
45
We shall assume throughout this section that all the pairs and maps employed are admissible. In particular all spaces will be assumed to be
collapsible.
This
is
the case for
l
all
the admissible categories that
l
we
shall
have occasion to consider in the sequel. LEMMA 16.1. #", E*J and (E*, En ) are homologically trivial. J PROOF. Let x e #", g t g 1. Using vector notation, define = (1  t)x. Then h(x,Q) = x, h(x,l) = 0. Therefore #" is conh(x,t) tractible on itself to a point. Thus, by 11.5, E" is homologically trivial.
,
Since
En J
l
is
homeomorphic to E"~
n
l
l
,
it
is
also homologically trivial.
Finally (ET,
EJ
16.2.
) is
homologically
trivial
l
,
LEMMA
proper.
The
triads (E*',ET
E'
by virtue of 9.5. l l 1 JET ,J0r ) are ) and OS"
1
;
PROOF.
We must
show that the
inclusion
maps
induce isomorphisms of the homology groups in of symmetry, it suffices to consider k 2 only.
all
dimensions.
Because
Let
V V
be the subset of
R n determined by the = 1, x n < i \\x\\
n~
l
conditions
Clearly
is
relative to
S
n~ l
an open subset of S
.
and
V
lies in
the interior of
En J
l
The homotopy
F(x,i)
=
x
xe
shows that
a strong deformation retract of (S l = S n '\ A = En U = E n l J J\ V,E*J 7). Now apply 12.1 with n~2 S to obtain that k 2 induces isomorphisms in all dimensions.
(fi""
1
2
n~ l
,^"
)
is
X
l
THEOREM
reduces
to the
The homology sequence of isomorphism
16.3.
d:
the triad
(E ;El~\E J
n
n
)
H
1
n (E*,S" )

H^Er ,^*)
1
All other groups in the sequence are
called the incidence
n
trivial.
isomorphism and
l ]
is
denoted by [E \E
The above isomorphism is n 1 n ^ ]. The ison
'
morphism [E
:
E"J
16.4.
THEOREM
J El~~ using the triad (E n n1 The homology groups of (/? ,S ) are as follows:
l
is defined similarly
y
En
l
i
).
H (E
n
n
y
Sn
1
)
1
=
0,
H<(E*,Sr
)
=0
l
forq^n.
PROOF OF
thus
16.3
n
l
AND
16.4.
H
q (E ,E J
n
)
=
By
by
0;
and,
16.1, (E",E*J ) is homologically trivial, the exactness of the H,S. of the triad
46
w
1 1
AXIOMS AND GENERAL THEOREMS
[CHAP.
1
I
;#r ,#r ), we have d: 2 the pairs (ET ,^" ) and
(
1
H
n
1
9 (ET,S* )
1
H^ET ,^*).
this
Since
(JB*"
,^"
2
)
are homeomorphic,
we have
yields
l
H,(E*,Sr~
)
1
**
H.
q
l
(E "\S '^.
l
n
Iterating
n (E).
result
is
H
H.H^ 16.4 follows from the Dimension axiom and the
9
(E*,S
)
n (E?,ST )
=
Since
#
a single point,
definition of the co
efficient
group G.
16.5.
This also establishes
16.3.
the
THEOREM
Commutativity holds in
diagram
u>/iere
A
(S
is the
n ~ lt
r
triad
E n En J +
)
~l
boundary operator in the MayerVietoris sequence of the For n > 2 all four homomorphisms are isomor).
l
phisms.
For n
=
2 the same holds provided
H~
n
n~ 2
2
(S
)
=
H
(P)
is
replaced by ff Q (S).
In
particular
A: A:
H (F)
n
n
>
1,
THEOREM
16.6.
The homology groups of S" are as follows:
G,
^(S")
G,
=
0,
n
n
p
> >
0,
=
PROOF.
15.5.
G,
0,
0,
^
n,0.
The commutativity
T and E
1
relation of 16.5
is
a consequence of
it
Since
"' 1
are homologically trivial,
follows from 9.4
that
d2
:
These give
all the conclusions of 16.5. Combining these isomorphisms n with 16.4 yields all the results of 16.6 except those concerning (S ). n Since 5T is nonempty for n ^ 0, it follows from 7.6 that (S ) ~ G. This concludes the proof. 80(8")
H
H
+
The formulations and
are
left to
proofs of analogous results for cohomology
the reader.
NOTES
NOTES
The origins of
the basic concepts.
47
The
basic
machinery for con
structing homology groups (namely: complexes and incidence numbers) is due to Poincar6 [Analysis Situs, Jour, de PEc. Polyt. (2) 1 (1895), 1123]. Using these he denned directly the Betti numbers and torsion
numbers.
They
are the numerical invariants which characterize the
homology groups based on the
coefficient group of integers. For many were the primary source of interest. It was during the years, they period 19251935 that attention shifted from the numerical invariants to the groups themselves. This shift was due in part to the influence of E. Noether. It was also enforced by two directions of generalization:
(1)
are not characterized
from complexes to more general spaces where the homology groups by numerical invariants, and (2) from integer
coefficients to arbitrary coefficients where, again, numerical invariants
are inadequate.
Thus, although Poincar6 did not speak of the homology groups themselves, he is to be credited with the origin of the concept. The concept of relative homology (modulo a subcomplex) is due to Lefschetz [Proc. Nat. Acad. 13 (1927), 614622]. The operator d was
The
used by Lefschetz. It is not clear who first gave it formal recognition. It origin of the induced homomorphism /, is likewise obscure. has been used, in a sense, since the time of Poincar6, but for at least This lack of thirtyfive years it had no name nor any formal status. formal recognition of d and /, is a natural consequence of the failure to accord the homology groups a formal status. Each of our axioms is a theorem of classical homology theory. In
most cases it is not Axioms 1, 2, 3, and 7
and proved them. The and too well understood to warrant such explicit treatment. One must be interested in an axiomatic development before one thinks of writing them down. The first formal recognition of the homology sequence and its exactness is due to Hurewicz [Bull. Amer. Math. Soc. 47 (1941), 562]. It was subsequently exploited by Kelley and Pitcher [Annals of Math. 48 (1947), 682709]. It should be noted though that the six parts of exactness were well known and had occurred frequently in the proofs
clear
first
who
stated
are, perhaps, too basic
of other less obvious propositions.
The excision property was implicit in Lefschetz's construction of He often used the expression "the homology the relative groups. A' instead of "the homology groups of mod A." groups of
X
9
X
The homotopy axiom has had no more formal recognition than f+. However the proposition "homotopic cycles are homologous" has been known and used for many years.
48
NOTES
The
origin of the
is
cohomology groups is thoroughly confused. The due to Whitney [Duke Math. Jour. 3(1937), 35The existence of cohomology groups was implicit in the duality 45]. theorem of Alexander [Trans. Amer. Math. Soc. 23 (1922), 333349]. The complete group invariant form of the duality theorem was proved by Pontrjagin [Annals of Math. 35 (1934), 904914]. Cocycles made their first formal appearance under the name pseudocydes in the book of Lefschetz [Colloq. Publ. Amer. Math. Soc., vol. 12, 1930]. The first intrinsic definition of the cohomology groups was given by Alexander
"co" terminology
at the
Moscow
conference in 1936.
similar to
Homotopy groups. The homotopy groups of Hurewicz are quite homology groups in that there are concepts analogous to ib The q those of homology theory and they obey similar axioms. homotopy group depends not only on a pair (X,A) but also on a reference point XQ e A and is denoted by * q (X,A,x ). It is defined when = 1 and is called q ^ 2. If A is a single point, it is also defined for q the fundamental group. The groups are abelian for q g: 3, and iri(X,Xo) A boundary operator d: 7r (X,A,z ) * TT Q .I(A XQ ) is also abelian. for q ^ 2. If /: exists (X,A,x Q ) (F,B,7/ ), there are induced homo* q (X,A,Xo) 7r a (F,,i/o). morphisms f+: When suitably modified, all of our Axioms 1 through 7 hold for
I
>
homotopy groups with the
is
sole exception of the Excision axiom.
This
the fundamental property distinguishing homology from homotopy. It accounts for the computability of homology groups as contrasted
with our meager knowledge of homotopy groups. n As a simple example, let the nsphere S be divided into upper and n~ l lower hemispheres, E?,E", by an equator S Using a fixed reference 1 point xQ e S"" for all groups, we obtain the diagram
.
where
zero.
All homotopy groups of a cell are i and j are inclusion maps. This and exactness imply that d and j+ are isomorphisms onto.
f?i+d~ Now i is an excision map, and we wish to state properties of i+. We have introduced the equivalent homomorphism E since it can be proved to coincide with the suspension (Einhdngung) homomorphism
.
Then we
define
E
by
E =
l
NOTES
W
introduced by Freudenthal [Composite Math. 5 (1937), 299314]. Since l 2 and v B (S ) is infinite cyclic, it follows that E, and therefore ir 2 ($ ) = i+ is not always an isomorphism onto. Hence the Excision axiom fails for homotopy groups in a very simple case. Freudenthal has shown 1. We may in[ibid] that E is an isomorphism for each q < 2n terpret his results as asserting that the excision property does hold in Most of the little we know about ^(^SP) is certain restricted cases. based on the results of Freudenthal concerning E. The problem of axiomatizing the homotopy groups has not been solved.* One would naturally seek a substitute for the Excision axiom. There is a reasonable candidate. Let B be a fibre bundle over the base I with projection p: B X, let x e X, Y = P~ (XQ), and space
9
X
.

t/o
p+ maps *(/*, Y ,y Q ) isomorphically onto proved using the covering homotopy theorem for bundles. If TT Q is replaced by //, then p^ is usually not an isomorphism onto. We have therefore a simple and useful property of homotopy groups which may serve as a substitute for the excision property. In axiomatizing the homotopy groups one would need an additional
e
ir Q (X,Xo).
Y
Then, This
for q
^
2,
is
basic concept, namely: the isomorphisms ir q (X,A,Xo) to a homotopy class of paths in A from x to assigned
= w^XjA.x^
Xi.
One would
deal not with single groups but with systems of groups connected by isomorphisms assigned to the fundamental groupoid of A. The latter
would include the operations
of 7r,(A,ar
)
on
7r Q
(X,A,x
).
nals of
Cohomotopy groups. The BorsukSpanier cohomotopy groups [AnMath. 50 (1949), 203245] are similar to cohomology groups. th If (X,A) is a compact pair of finite dimension rc, then the <? cohomotopy q group ir (X,A) is defined for each integer q > (n + l)/2 and is an
abelian group.
The elements
a
of Tr*(X,A) are
is
homotopy
classes of
maps
of (X,A) into (S
y
a ^sphere and T/ O is a point. The set Tr*(X A) is defined for q ^ 0; but the addition is only defined for q > A mapping 5: T a1 (A) > ^(X.A) is defined for q > (n l)/2. and is homomorphic when both sides are groups. If /: (X,A) > (Y,B), q > ir then /*: 7r a (F,J5) (X,A) is defined for all q and is homomorphic when both sides are groups. Spanier has shown [ibid] that the co,2/o)
where S*
+
homotopy groups
satisfy the analogs of
aU the cohomology axioms
insofar as they are meaningful.
The axiomatization of the cohomotopy groups has not been attempted. The chief distinction between cohomology and cohomotopy
the absence of a group structure in ir (X,A) for q ^ (n l)/2. This makes it impossible to compute cohomotopy groups by an induction starting with q = 0. Our knowledge of cohomotopy groups is as meager as that of homotopy groups.
is
a
+
*An axiomatization has subsequently been given by Math. 63(1956), 272284].
J.
Milnor [Annals of
50
EXERCISES
EXERCISES
A. ELEMENTARY PROPERTIES OF EXACT SEQUENCES. Establish the following propositions for an exact lower sequence
\G.4,\'.
1. <<,+!
2.
3.
</>,_!
<,+!
= = =
4.
5.
G = Gq =
q
and only if kernel Q = 0. if and only if <<,((?) = (7 a i. if and only if </>,: and <^ G _ = G q = (?_,. = and Q = 0. if and only if Q+l = G+i and kernel Q i if and only if Q+ 2 (G Q +2)
if
<t>
1
<t>
<f>
'
<t>
<f>
=
0.
B.
1.
THE AXIOMS.
Show that Axiom 1 is a consequence of Axioms 2, 3, and 4. Show that Axiom 1 is a consequence of Axioms 2 and 6. Assume that, for every admissible pair (X,A), the map
2. 3.
p:
(XjA)
X
1
/
(XjA) defined by
2,
p(x,t)
=
x
is
admissible.
Show, using
Axioms
and
if
that the
the following
suitable
Homotopy axiom is maps are admissible:
is
equivalent to either of
Axiom Axiom
5".
5'".
p+
is
an isomorphism.
kernel of p^
zero.
is
The
In the following problems, assume that the homology theory defined on the category of all pairs and maps.
C. RETRACTS.
H G and ^ = identity, then has kernel H, H decomposes into a direct sum zero and H = image + kernel X A a retraction, and 2. If A X the inclusion, then H (X) decomposes into the direct sum H (X) = image i+ + kernel r+
1.
If
<f>:
>
\f/:
>
(?,
</>
<f>
\l/.
r:
>
is
i:
>
is
g
Q
q
Q,
and
H
1.
g
(A)
+
H.(X,A).
D. THE 0DIMENSIONAL GROUPS.
If
x
e
and only if H Q (X,A) subset A of X. Assume that H^A) = 0. 3. If X is an arcwise connected space, = (gx')x and (Ox) x  (Gfc')x.
2.
X, then
H
(X,x)
=
x,x' e
Bo(X)
=
if
for each
nonvacuous
e
X, and g
G, then
4. If
X
G.
is
(X)
5.
~
If
a Hausdorff space consisting of just two points, then
G
7*
and
ff (X)
=
0,
then
Z
is
connected.
EXERCISES
E. TRIVIAL MAPS.
1.
51
If /:
q.
(X,A)
 (F,)
is
is
such that /(X)
is
C
#, then /
M =
=
for each
2.
If /:
X >
(F,B)
0.
such that / (X)
for g
^
and
J^ =
a point of F, then /,,
F.
1.
DIRECT SUM.
Prove 13.2 with the condition that the
sets
Xa
is normal, (2) placed by the following conditions: (1 ) A Q have disjoint closures. and (3) the sets a
X
X
are disjoint rea C\ A = A a
,
X
G.
1.
MAYERVIETORIS SEQUENCES. Let (X;Xj,X 2 ) be a proper triad.
(Y
Show
that in the diagram
a
+ lv'**' >**
Y
l
\.
1 ^^
** 2/
Y
\
k *
TT (Y " a\^* Y
'
r\
'
1 j^* 1
**'2/
Y
\
1
d
a
i l
I
where
d 2 are the boundary operators of the triads (X;Xj,X 2 ) and the anticommutativity relation (X;X 2 ,Xi)
d]
and
dd,
+
dd 2
=
holds.
Let (X;Xi,X 2 ) be a proper triad; prove that d maps the relative MayerVietoris sequence into the MayerVietoris sequence of the triad
2.
(X,
U X ;X U X
2
2 ).
3.
For any space X, show that the MayerVietoris
operator
is
A
(defined in 15.6) for the triad
XXOWXX1)=
A = X,
an isomorphism, thereby establishing fl^iCX). Prove this
relative boundary (X X /; X X 0, X X 1) an isomorphism H Q (X X /,
last result
without explicit
use of the MayerVietoris sequence. 4. Let (X;Xi,X 2 ) be a proper triad such that
X=
1)
Xi
UX
2,
let
nX
2
,
7
=
[0,1],
and
let
Y = B =
(X, (Xi
X X
0) 0)
U (A
\J
(X a
X /) U (X X X 1)
2
be the indicated subsets of
X X /. Show, using the isomorphism of problem 3, that the MayerVietoris sequence of (X;Xi,X 2 ) is isomorphic totheH.S. of (K,B).
5.
Show
that the relative MayerVietoris sequence of (X;X,,X 2 )
is
52
EXERCISES
isomorphic to the homology sequence of a triple consisting of
X X
2,
/
and
suitable subspaces.
6.
7.
1
X C\X
its
Examine the MayerVietoris sequence of the Consider a proper triad (X;Xi,X 2 ) with X
2
triad (X;X,X).
=
Xi \J
X A =
7^ 0.
Define the reduced MayerVietoris sequence and prove
exactness.
H. RANKS OP GROUPS. It will be assumed that all groups considered are Dmodules over a domain of integrity D (i.e. a commutative ring D with a unit element, and such that di 5* 0, d 2 ^ imply d^ ^ 0). DEFINITION. The elements #,, g r of a Dmodule G are called
,
linearly independent, provided
where d lt
,
dr
e
D
any relation d g = dr implies d =
l
l
+
=
0.
+
drgf
=
l
The maximum
number of linearly independent elements in G is called the rank of G and is denoted by r(G) (or r D (G) if we wish to indicate D). If this If D is a field, then r(ff) maximum does not exist, we set r(G) =
<
.
is
the dimension of
1.
G over D.
is a submodule of G, show that r(G) = r(H) + r(0/H). Let {G q q } be an exact lower sequence of Dmodules such that each r(G Q ) is finite. Let w q = r(kernel #<,). Then for m ^ n
If
H
2.
,<t>
I.
BETTI NUMBERS.
DEFINITION. Assume in this section that a homology theory is Q (X,A) are all Dmodules and f^d are linear over D. given such that The rank r[H q (X,A)\ is then denoted by R,(X,A) and is called the r th
H
Betti
number of (X,A). Let (X A) be a pair. \R q (X)},{R q (X,A)}. Show
1.
9
Consider the three sequences {R q (A)\,
that,
if
two
of these sequences contain
numbers, then so does the third one. Show that, if in two of these sequences only a finite number of terms are different from zero, then the same holds in the third sequence.
only
finite
2.
Show
that (assuming
all
numbers involved are
finite)
*n
where
w
q
is
the rank of the kernel of the
homomorphism
//<,(;!)
>
H (X)
q
induced by the inclusion
A C
X.
EXERCISES
DEFINITION.
finite
53
If
number
of
them
the integers [R g (X,A)} are all finite and only a are different from zero then the integer
X (X,A)
is
= E(l)'ft.(X
f
A)
called the Euler characteristic of (X,A).
3.
is a pair and two of the Euler characteristics x(4), are defined, then so is the third, and x(X), x(X,A)
If
(X,A)
x(X)
4.
)
=
X (A)
+
X (X,A).
Let
N
= VJ 2 A = Let (X lJ 2 ) be a proper triad with l q be the intersection of the kernels of the homomorphisms
X X
X
X
l
X
,
X OX
2.
H.(A)
^ fl (XO,
f
ff f (A) +
H,(X
2.
2)
induced by inclusion maps Vietoris formula
A C
Xi,
A C X
Prove the Mayer
R,(XJ
and derive from
it
+
r(N t)
+
r(AT,.,)
the formula
8)
,
UX
provided
5.
all
four Euler characteristics involved are defined.
late
Formulate the MayerVietoris formula for cohomology. Formurelative MayerVietoris formulas for homology and cohomology.
CHAPTER
II
Simplicial complexes
1.
INTRODUCTION
This chapter develops the analytical geometric tools needed in subsequent chapters. These are simplex, complex, subcomplex, simplicial map, triangulation, and simplicial approximation. Homology theory is not mentioned; its study is resumed in Chapter in. The reason for this hiatus is the necessity of singling out a class of spaces (triangulable spaces) sufficiently simple that an algorithm can be given for computing their homology groups. The nature of this class is nearly pre:
dictable on the basis of the results of Chapter i. Knowing the groups of a point, the groups of a contractible space are determined.
We
choose a class of contractible spaces (i.e. simplexes) and form more complicated spaces (i.e. complexes) by assembling these in a smooth fashion. Then the groups of the latter spaces can be computed by the use of Mayer Vietoris sequences or similar devices. It is not enough to be able to compute groups: it is also necessary to compute homomorphisms. This requires singling out a simple class
of
maps simplicial maps. Although these are quite restricted, it is shown that any map of one triangulable space in another is homotopic
to one such:
of
all
indeed, simplicial
maps
are dense in the function space
maps. This is achieved by barycentric subdivision and the simplicial approximation theorem. Although triangulable spaces appear to form a rather narrow class, a major portion of the spaces occurring in applications of topology to geometry and analysis are of this type. Furthermore, it is shown in Chapter x that any compact space can be expressed as a limit of triangulable spaces in a reasonable sense. In this sense, triangulable spaces are dense in the family of compact spaces.
2.
SIMPLEXES
1 objects called An nsimplex s is a set of n denoted by {A}, together with the set of all realvalued functions a defined on {A} satisfying
DEFINITION
2.1.
+
vertices, usually
(1)
I>G4) = A
1,
a(A)^Q.
54
2]
SIMPLEXES
55
single function a is called a point of s. The values of a of s are called the barycentric coordinates of the point a.
p(or,/9)
A
on the vertices
The distance
of
two points a,0 of
s is
defined
by
topological space thus defined is denoted by \s\ and is called the space of s. Clearly the barycentric coordinates are continuous functions
The
on
\s\.
vertices
term "barycenter" stems from the fact that, if the a euclidean space, then the point a [A\ to the center of gravity of the system of masses obtained corresponds by assigning to each vertex A the mass a(A). DEFINITION 2.2. A simplex s together with a simple ordering A < < A n of its vertices is called an ordered simplex. Let R n * 1 x n ). The correbe the cartesian space of all coordinates (x n 1 > R** a > (a(A), is then an isometric map s a(A )) spondence
of the
of s are points of
,
The use
,
,
,
called the canonical imbedding of s in
R n+1
.
The image
of s in
Rn *
n
1
is
denoted by A and
n.
n
is
called the unit simplex of
R n+l
.
Clearly
x,
A
is
the
1,
intersection of the plane
,
2"
5.
z
=
1
with the sector
^
for i
=
\s\
Since
An
for
is
closed
and bounded
in
/T*
1
,
it is
compact.
Hence
is
compact
any simplex
Clearly, the dimension of
s,
in the eu
clidean sense, is n. Observe that an unordered nsimplex has (n 1)! 1 canonical imbeddings in R** corresponding to the various orderings of
its vertices.
+
vertices
DEFINITION 2.3. A qface s' of an nsimplex form a subset of the vertices of s.
point a' of
It
s is
a gsimplex whose
A
of
s.
s' is a function defined over a subset of the vertices can be extended to a function a on all vertices by setting
a(A)
=
a.'
(A)
s.
if
A
is
in
s',
and a(A)
=
otherwise.
(1)
Then a
is
clearly
a point of
In view of the conditions
on
a, a', the extension
a of
The map a' > a imbeds s' isometrically in s. Folthe custom, we identify a' with a. so that s' is a subset of \s\. lowing It is surely a closed subset of \s\. It is defined by the equations a(A) =
a' is unique.
for
A A
not in
s'.
It is
= 1. 0simplex has just one vertex A, and just one point a(A) to identify the vertex with the point and to denote customary
either
of
by A. With this convention the vertices A of s are the 0faces and are points of s. As a result, A(B) is defined for any two s, = if A ? B, and A (A) 1 for each A. In the vertices, and A(B) n n* the vertices appear as the unit points on the unit simplex A in R n n*1 In addition A is the smallest convex set in R coordinate axes.
1
56
SIMPLICIAL COMPLEXES
For
[CHAP.
II
containing these unit points.
this reason the simplex is said to
span DEFINITION
its vertices.
2.4.
The
set of points of s having at least is a closed subset of \s\. The \s\
point set boundary of s, denoted by *, is the one coordinate equal to zero. Clearly
open
set
s
\s\
is
called the interior
of s
and
is
If
a,
also referred to as the open simplex. a q are points of an nsimplex s, and
,
w
1,
,
negative real
numbers such that
WQ
+
+w =
g
W Q are nonthen the function
,
a defined by a
is
= W
<x
+
+wa
a
a
again a point of
oi)
s.
For clearly a (A)
^
for each vertex
A
of
s,
and
=
,
LEMMA 2.5. // a, a* are points of a simplex s, and z/ w W Q are positive numbers of sum 1 such that the point w.a* is a vertex A of s, then each a* = A. = A(J5) = t0 PROOF. For any vertex B 9* A of s, we have *(JB) = 0. As this holds for each B 0. Since each w > 0, each a(B) different from A, it follows that a = A.
, ,
^
t
X
t
3.
SIMPLICIAL COMPLEXES
simplicial complex
DEFINITION
a simplex
collection
is
3.1.
A
K
is
a collection of faces of
s satisfying
the condition that every face of a simplex in the is the subset likewise in the collection. The space \K\ of
K
of
s
consisting of those points which belong to simplexes of
simplicial
K.
The same
Si
complex
K
and
*2
.
In such a case
to
$1
K
.
may
lie
in
two
different simplexes
also lies in the simplex s spanning the
Since the topology of s is the subspace is indeit follows that the topology of topology of both \Si\ , pendent of the particular simplex s in terms of which it is defined. The collection of all faces of s including s itself is a simplicial comvertices
common
and s 2 and s 2
K
plex.
This complex
excluding s
is
also denoted
by
s.
The
collection of all faces
of
is denoted by & is said to be ndimensional DEFINITION 3.2. A simplicial complex an ncomplex) provided contains an nsimplex but no (briefly: (n l)simplex (and, therefore, no simplex of dimension >n). DEFINITION 3.3. If K is a simplicial complex, a subcomplex L of
s,
itself, is
a simplicial complex and
K
K
+
K
a subcollection of the simplexes of such that each face of a simplex in L is also in L. Clearly, L is a simplicial complex. is a simplicial LEMMA 3.4. // a, oT are points of where
is
,
K
K
y
K
4]
LINEAR AND SIMPLICIAL MAPS
57
complex, and
w
Qj
,
wn
are positive numbers such that
w
if
+
and only
wn =
then the point a = w a n a are in a simplex of the points a,
1,
+
if
+
+ wa
n
n
is
in
K
w
.
,
K.
PROOF.
If
Let
s
n
be the simplex containing
s
K
and

let s'
,
be the lowest
dimensional face of
,
a, Thus the condition is sufficient. Suppose now that a e \K\. If
a
a
containing the points a, are in a simplex of K, then
4,
' '
a
Then a
and a
E s'.
s'
is
in
K
e \K\.
,
Aq
are the vertices of
s',
then
= a(A)A +
'
+
<*(A
V,
where
a(A')
= E^a'(A*).
;
Since w, > that a(A*)
that s
f
and
>
the least simplex containing a, f and a is in the interior of s Hence
s' is
.
n
,
<x
,
it
follows
a
e
\K\ implies
is in
K and thus a,
3.5.
,
COROLLARY
//
a,
,
a are in a simplex of K. n a e X /tan aZZ Me points
,
n
where
WQ
+

+w =
n
1,
w,
>
are in
K
if
and only
if
a,
,
a*
are in a simplex of K.
DEFINITION
subset st(A) of
3.6.
For each vertex
A
of
K
,
the open star of
all
A
is
the
by the condition a(A) > 0. An equivalent definition is that st(A) is the union of plexes of having A as a vertex.
K defined
//
open sim
K
LEMMA
complex K,
if
A,
,
are distinct vertices of the simplicial then their open stars have a nonempty intersection if and only A n are the vertices of a simplex of K.
3.7.
,
A,
An
PROOF.
interior of s
If
is
A,
E
,
A*
l
in st(A') for
each
are the vertices of a simplex s, then the 0. z, so that P\st(^T) Conversely,
^
suppose that a
therefore
P\ st(A
,
).
Then a(A') >
for i
=
0,
,
n,
and
A,
A
4.
n
are the vertices of a simplex of
K by virtue of 3.4.
LINEAR AND SIMPLICIAL MAPS
>
is
DEFINITION 4.1. If K, K' are simplicial complexes and/: \K\ a map, then we say that / is linear (Notation /: K > K )
1
\K'\
if
/
is
,
linear in terms of the barycentric coordinates. n a are points of \K\ and
Precisely,
if
a,a,
a
=
<
^
0,
58
SIMPLICIAL COMPLEXES
[CHAP. II
then
/(a)
=
UWV)+
K
y
>
+W f(a
n
n
).
A
linear
map which
carries vertices into vertices
is
called simplicial.
K' respectively. By a linear [simLet L,L' be subcomplexes of (K,L) (/',!/) is meant a map of (7,L) into plicial] map /: into K'. which defines a linear [simplicial] map of ( /'),!// 1)
K
Two
trivial
consequences are
THEOREM THEOREM
plicial}.
4.2. 4.3.
The
///:
identity
(K,L) * (K,L) is simplicial. > (#',') and g: (K'.L )  (tf",L") (K,L)
map
1
are both linear [simplicial], then gf:
(K,L)
>
(K",L")
is linear [sim
The
cial
complexes, namely:
following theorem expresses a fundamental property of simplilinear maps are described by their behavior on
4.4.
vertices.
THEOREM
termined by
points of
A
linear
map
to
f:
(K,L)
>
its
r
values on the vertices.
K
A map
map
of the vertices of
f:
(/',//) is uniquely deinto
K
can be extended
a linear
(K,L)
If
>
(/C',Z/)
if
is
and only
if the <t>image of the set of vertices of
contained in a simplex of K' or vertices, then f is simplicial.
of
U
any simplex of
<t>
K
or
L
respectively.
maps
is
vertices into
PROOF. K, then
If /:
(K,L)
>
(/',//)
is
linear,
and a
n n
in a simplex 8
a
where
= a(A)A +
s.
+
<*(A
)A
,
A,
,
A
n
are the vertices of
Then
/(a) so that /(a)
is
= a(A)f(A) +
+ a(AW),
,
is determined by its values on vertices. Further, if for i = 0, s, then a(A') > n; and by 3.4, n f(A), f(A ) are in a simplex of K'. Similarly, if s is a simplex n of L, then /(A ), f(A ) are in a simplex of L'. This proves that the condition stated in 4.4 is necessary for the extension of a vertex
an
interior point of
,
,
map To
Let
<t>.
prove that the condition on
s
<f>
is
also sufficient, consider the
simplexes
and
,
s'
spanning the vertices of
K
and K'
respectively.
</>
be the vertices of s. Given a vertex the conditions of 4.4, for each a. in s, define f(a) by
A,
An
map
satisfying
/(a)
=
(4)*(A)
is
+


+
a(A>(A").
linear
,
It is readily seen that /(a)
in
of
Now A,
let
a
,
s
\K\ (or
An
be the subsequence of those vertices with a(A') > 0. Then consisting
e L)
a
s', and / is a and let B,
map
s
> s'.
B9
4]
LINEAR AND SIMPLICIAL MAPS
50
a
= a() + B are J5,
Q
,
+
=
Q q a(B )B with a(B')
>
0,
and by
3.4,
the vertices
</>CB),
,
0()
those of a simplex of Hence the points (or of L). are all in a simplex of K' (or of L'), so that
K
/(a)
is
a(B)<t>(B)
+
+
<*(<)</>(<)
in
X
7
(or in L').
is
Thus /
defines a linear
map
/:
(tf.L)
> (',/),
and the proof
complete.
consequence of Definition 4.1 is THEOREM 4.5. // /: (K,L) > (K',L') is linear
trivial
f
A
[simplicial],
and
if
(K,L),(K',L') are^subcomplexes of (K,L),(K',L
),
respectively, such that
f maps \K\ into \K'\ and \L\ into L', then the map of (K,L) into (K',U) defined by f is linear [simplicial]. LEMMA 4.6. // /: K > K' is linear and maps \K\ onto \K'\ then each vertex of K' is the image of at least one vertex of K. PROOF. Let B be a vertex of K'. Then B = /(a) where a is a a(A*)A where A' are the vertices of K, it point on K. Since a follows that B = /(a) = X) <*(A )f(A ). This implies that, if a(A') > 0, then B = f(A ) by 2.5.
9
^
l
l
l
l
THEOREM
then f~
11
1
4.7.
// a simplicial
map
f:
K+K' is a homeomorphism,
is simplicial.
from the preceding lemma that / establishes a be the correspondence between the vertices of K and K'. Let vertex map of the vertices of K' onto those of K given by /"*. Consider the vertices B, B n of a simplex of K' and let be a point in the
PROOF.
It follows
</>


,
interior of that simplex.
ft
Then
=
vQ
B
+
l
+vB
n
(/3)
n
where X)
r,
=
1
and
vt
>
0.
Also f'
is
a point of
K
}
and
where
A,
,
Am
are distinct vertices of
/v,
^ w,
). n
)
=
1,
and w>
>
0.
It follows that
ft
= w f(A)
+
+ w mf(A m
,
This implies that B*. Since J5, follows that 0(5),
,
m
A,
,
n and /(A),
,
f(A
is
a permutation of
n
<t>(B
)
are the vertices of a simplex of K, it are the vertices of a simplex of K. Hence,
An
by
l
4.4,
<t>
extends to a simplicial
4.8.
THEOREM
f~
// /:
K
=
map
K'
is
g:
K
r
>
K.
Clearly g
=
l
f~
.
+
a homeomorphism, and both f and
PROOF.
are linear, then f is simplicial. be a vertex of Let
A
K.
Then

f(A)
v
B
+
+
v n B",
60
SIMPLICIAL COMPLEXES
[CHAP. II
)
*>
where
Vi
B,
,
Bn
are the vertices of a simplex of K',
=
1,
and
>
0.
Applying /"* we find
A = ^
Hence, by
2.5,
A =
is
= Bn
,
and /(A)
/"'(B ) = a vertex of K'.
=
/"'(IT),
and therefore
B =
5.
TRIANGULATED SPACES
DEFINITION 5.1. Given a pair CX",A), a triangulation T = of (X",A) consists of a simplicial pair (K ,L) and a homeomorphic
t:
map
The
pair.
pair (Jf ,A) together
If
 (JM). (\K\,\L\) with a triangulation T
is
called a triangulated
is
a triangulation of a pair (X,A)
(X,A),(X',A')
exists, the pair
called tri
angulable.
Let
{t' 9
have triangulations
T =
{*,(#,)},
I"
=
(K',L')} respectively.
A map
l
is
a
called linear [or simplicial], with respect to T,T', if the map t~ f? is linear [or simplicial] map of (K' L') into (K,L). The definition of triangulated triples and their maps is similar.
9
As a
direct consequence of the definition
5.2.
we have
is simplicial.
LEMMA
The
identity
map
of a triangulated pair
The composition of two linear or two simplicial maps is linear or simplicial. As consequences of 4.7 and 4.8 we have LEMMA 5.3. The inverse of a simplicial homeomorphism is simplicial.
LEMMA
simplicial.
If
5.4.
A
linear
homeomorphism, whose inverse
is
linear,
is
{t,(K,L)\ simplicial concepts in
T =
is
K
a triangulation of (X,A), then the various can be carried over into by means of the
X
map
t.
Thus a simplex
etc.
of the triangulation
T
will
mean
the
image of
a simplex of K, DEFINITION
space X
cisely,
is
5.5.
The mesh
the
maximum
of a triangulation T [t,K] of a metric of the diameters of the simplexes of T. Pre
=
mesh T = max [diam
where s is any simplex of K. DEFINITION 5.6. Let T
t(\8\)] 9
=
A function /: X
[t,K] be a triangulation of a space
X.
n
>
fl
,
mapping
X homeomorphically onto a
subset of
6]
n
BARYCENTR1C SUBDIVISION
61
n
euclidean space R is called a linear imbedding of into R (relative to H > R is linear, i.e. provided the cartesian T) provided the map ft:
X
K
coordinates of the point ft(a) are linear functions of the barycentric coordinates of the point a on K, If / is a linear imbedding, then the distance function in defined by
X
is
called a linear metric in
X
(relative to T).
LEMMA
PROOF.
\s\
>
5.7.
Every space
X
with a triangulation
T =
{t,K\ has a
linear imbedding.
K is a subcomplex of a gsimplex s. Let be a canonical imbedding as defined in 2.2. Then / = q* ZT is a linear imbedding of X in R COROLLARY 5.8. Every triangulated space has a linear metric.
The complex
1:
R Q+l
1
1
.
6.
BARYCENTRIC SUBDIVISION
let s
is
Let
vertices
K
be a simplicial complex, and
,
be a simplex of
the point defined
1
K
with
A,
An
"
.
The
1
barycenter b. of s
by
n
+
'
'
1
n
+
'
1
A is a vertex, then 6 X = A. a second complex, denoted by Sd K, defined as to assign follows: Its vertices are the bary centers of the simplexes of K. For each
In particular,
if
We
K
such that s is a face of s + l s a of simplexes of the sequence of corresponding barycenters is the set (i 0, q 1), of vertices of a simplex of Sd K. Only simplexes obtained in this manner
sequence
s
,
i,
,
K
t
t
=
,
are in
of
Sd K. With the notation as above, the vertices 6.., Sd K lie in the simplex S Q of K. Therefore the
Sd
vertices of
K
into
K
a simplex identity map of the a unique linear extension to a map has, by 4.4,
,
&,, of
1
K
:
Sd
K
*
K
Sd
DEFINITION
linear
If
1L
1
6.1.
map K L is a subcomplex
is
l
The pair consisting of the complex called the barycentric subdivision of K.
of
K and the
K
y
then Sd
L
is
=
K \Sd L.
The
7
pair consisting of the pair (Sd
/Q/1
and a subcomplex of Sd Sd L) and the K,
(
K
linear
map
by 1K
is
V
Qt\ T
\
\
K' T
\
defined
,
called the barycentric subdivision of (K,L).
62
S1MPLICIAL COMPLEXES
[CHAP.
II
LEMMA
6.2.
The map 1 K
is 11.
Therefore
Sd
K is
(linearly)
homeo
morphic to K. PROOF. Let
(1)
be a point
b.
, ,
in the simplex of
Sd
K
with vertices
6.
,
,
fe, a .
Since
b >9 are in a simplex of K, the of \K\. Thus, in this symbolism, point
Without loss of and a face of s t+l
same symbol (1) represents a 1 K becomes an identity map. we may assume that each s is an zsimplex generality
t
.
Let then
A,
,
,
A
Q
be the vertices of
s,, i
sa
0,
ordered
,
in such a fashion that
A,
A* are the vertices of
=
q.
Then, by definition
6
of the barycenter,
.iTT* 0+
(1)

+ rTT A
+
(A )A
f a
'
Substituting this in
we
find that
a
= a(A)A +
=
where
(2)
a(A')
E ~: W,
J ~T
are the barycentric coordinates of a in K.
(3)
From
(2)
we deduce
a(A)
^a(A'),
t+1
and that
W
6 to
,
u, t
^
,
= =
(t
(+
in
+
a
l)[a(A')

a(A
)]
for i
=
0,



,
?

1,
1)(A
1:1
G
).
These formulas show that
b, q
//c
maps the simplex
of
Sd
K
with vertices
vertices
A,
,
A
Q
fashion onto the portion of the simplex with determined by conditions (3). Any point a of
(3) for
A
A
a
lies in
If it lies in
some set defined by two such sets, then a(A')
some order
+
1
of the A's.
=
.
a(/l*
)
for
one such order.
(4) is
In this case w,
defined.
It is
=
0,
so the transformation defined
by
uniquely
obviously the inverse of 1 K
is
of
The following lemma Sd K.
an immediate consequence of the definition
is an ncomplex. then Sd {t,(K,L)} be a triangulation of (X,A). Sd L) and its linear homeoConsider the barycentric subdivision (Sd
LEMMA
6.3.
If
K is an ncomplex,
Let
K
DEFINITION
I (K
6.4.
T =
K
,
morphism
,D onto (K,L).
The
triangulation
SdT=
6]
BARYCENTRIC SUBDIVISION
The
ih
68
is
called the baryceniric subdivision of T.
i
barycentric subdivision
of
T
is
defined inductively by
7'=
r,
T
mesh
=
SdC'T),
i
=
1,2, ....
LEMMA
we have
6.5.
For every triangulation
lim
T =
{t,K\ of a metric space
X,
T
=
(if
Since
X
is
compact, the limit in question
equal to zero)
is
inde
pendent
that the metric in
by
6.3,
Hence, by 5.8, Suppose that K is an ncomplex; then, the successive barycentric subdivisions of K also are ncom
of the choice of the metric.
X
we may assume
is
linear.
plexes.
Lemma
6.6.
6.5
is
thus a consequence of the following stronger
be a space with a metric linear relative to
lemma.
LEMMA
angulation
Let
X
a
tri
T =
{t,K\, where
K is an ncomplex.
n
Then
< mesh Sd T ~
n
+
.
,
mesh T.
1
This lemma follows readily from the following succession of elementary lemmas concerning points in a euclidean space. 9 Let po, p n be a sequence of points of a euclidean space R Consider the set C(p Q p n ) of all points x e R of the form (in vector
,
.
,
,
notation)
X
= WQ p +
Q
+
WnPn,
if
where
(1)
w ^
%
and
\y
]T}
w =
t
1.
Observe that,
p,
x,y
s
C, then
i

x
g
\y

for
some
=
0,
,
n,
because
y
*!
= X (w.i/  ^.p.)l ^ 53 w. y  p.l  p,) 53 w = max y ~ P<lg (max
i/
*
Applying
p,.
(1) again,
we
find that \y
x
^
p,
p, for
some
p,
and
Hence
LEMMA LEMMA
6
6.7.
diam C(p
//
,
,
p) = diam
(p
,
,
p n).
6.8.
=

(<D n
4 ... 4 7>J.
6'
=
TTT
ton
+
+
TO,
i
^
n.
I
6 "" b '\
^
~ diam
64
SIMPLICIAL COMPLEXES
Indeed by
(1)
[CHAP. II
we have
\b
b'\
^
6
pf
\
for
some j
=
0,
,
n.
Hence
6

V\
....
(f\.
i.
... _L f\\
/n
.
..
;
n+1
i
n
diam (P>
Pn)
Lemmas 6.7 and 6.8 imply that, if s' is a simplex of the barycentric subdivision of an nsimplex s of T, then, in the linear metric, diam
a'
n/(n
+
1)
diam
s.
This implies
6.6.
7.
SIMPLICIAL APPROXIMATION
and X' be spaces with triangulations DEFINITION 7.1. Let and T' = {*',#'} respectively, and let a map /: X > X' {*,#} X' (with respect to T,T') is called be given. A simplicial map g: a simplicial approximation to f if, for each x e X, g(x) lies on the closed simplex whose interior contains f(x). LEMMA 7.2. A necessary and sufficient condition for a simplicial map * X' to be a simplicial approximation to f is that, for every vertex g:
X
T =
X
X
AofX,
f(Bt(A))
C
*t(g(A)).
i.e.
PROOF. Suppose g is a simplicial approximation to /. Let x e st(A), x has a positive barycentric coordinate relative to A. It follows that g(x) has a positive barycentric coordinate relative to g(A). Since
g(x) lies in the closure of the
open simplex containing /(x), it follows that f(x) has a positive barycentric coordinate relative to g(A), i.e.
Let Conversely, suppose that the condition of 7.2 is fulfilled. and let s be a simplex of (relative to T) containing x in its
x
e
X
X
be a simplex of X' (relative to T') containing f(x) in its interior. For every vertex A of s we have x e st(A), so that f(x) s st(g(A)). Thus g(A) is a vertex of s' and g maps s onto a face
interior.
Similarly let
s'
of
s'.
Thus fif(x) lies in the closed simplex s'. The main result of this section is the following
7.3.
existence theorem:
f
Xbea triangulable metric space, X a triangulated X' be with triangulation T' = {',/'}, and let a map f: X space such that, for any triangulation T of X There is a number c > given. X' relative to T,T of mesh <, there exists a simplicial map g: X
THEOREM
Let
>
f
which
is
a simplicial approximation
to f.
7]
S1MPL1C1AL APPROXIMATION
65
This theorem and 6.5 yield
COROLLARY
T,T' and
let
7.4.
Let
> X' be a map f: There is an integer ho such given. that for every h ^ h* there exists a simplicial map g: X* relative k to T,T' which is a simplicial approximation to f. The proof of 7.3 will be preceded by the following lemma: LEMMA 7.5. // is a compact metric space and a collection of open sets covering X, then there exists a positive number such that each subset
X
X,X'
be topological spaces with triangulations
X
X
of
X of diameter <
upper bound of
is
all
least
contained in at least one set of the family <p. The such numbers is catted the Lebesgue number of
the covering #.
Then, for each positive integer n, 1/n not contained in any the compactness of X, there is a point x e set of 0. Let x n e A n By such that each neighborhood of x contains infinitely many points of {x n j. Let U be an open set of containing x, and let d be the distance  [7. Select n so that n > 2/d and p(x,xn ) < d/2. Then, from x to for each y e A ny
PROOF.
Suppose
this is false.
there exists a subset
An
of
X of diameter <
<t>
.
X
X
p(y,x)
^
p(2/,z)
+
,
p(x,xn)
g

+

<
d.
Hence y
T', let
e
PROOF
U
%
U, a contradiction. l B m be the vertices of the triangulation OF 7.3. Let = st(B') (in T'), and let V, = f' l (U ). Since {U,} is an
U and A n C
B
,
%
open covering of X', \V Lebesgue number of {F.},
angulation of
% }
is
let
an open covering = 77/2, and let
of
X.
Let
17
be the
A with mesh <e. Then, has diameter <rj, so that st(A) V* for angulation T, the set st(A) m. Choose such an i and define g(A) = B\ Then some i = 1,
X
{t,K\ be a trifor every vertex of the tri
T =
C
,
(1)
where st(A)
C\ st(0(A*))
is
in
T and st(g(A))
is
in T'.
If
4
of a simplex of T, then,
5* 0.
st(A ) p* 0. This and (1) imply by 3.7, n 3.7 the vertices g(A), Hence by 0(A ) are in
,
O
A,
,
A n are the vertices
a simplex of
g:
T'.
Therefore by 4.4 the vertex
is
map
g extends to a
X
g:
map
+
X' which
vertex
map
To be precise, the simplicial relative to T,T". l into g defines a vertex map (t')~ gt of the vertices of
K
those of K'.
This vertex
map
extends to a simplicial
g
X /(st(A)) C
and
[t,(K,L)},
inclusion (1) implies st(g(A)); so, by 7.2, g is a simplicial approximation to /. THEOREM 7.6. Let (X,A),(X',A') be pairs with triangulations T =
is
X'
defined
by
=
map
g:
K
K',
1
t'^t"
.
The
T
=
{t' y
(K',L')}.
Consider a
map
f:
(X,A) > (X',A')
66
SIMPLICIAL COMPLEXES
the
[CHAP.
If g^ is f. A1 then g\(A)
II
and
maps
/,:
X
X', f 2
to fi,
:
A
>
A' defined by
T,T')
y
simplicial approximation
gi defines
(relative to
C
any and
maps
g:
(X,A)
on
T
r
(X',A
to T'.
),
g>:
A A
A'.
mation to / 2 relative and most important,
g 2 is a simplicial approxiand A'. Finally, the maps f and g are homotopic in such a way that, during the homotopy, the image of a point x e stays in the simplex whose
is simplicial
The map g
The
map
to the
induced triangulations of
X
interior contains f(x).
A, then /i(x) e A and /i(x) is in a simplex of A'. For each x e X, the points f(x) and g(x) lie in the Hence gi(x) e A same simplex of X', hence the corresponding points /(z),0(x) of K' lie
PROOF.
If
r
a;
e
,
r
.
in the
same simplex
of K'.
Then the
t'[rj(x)
required
(I
homotopy
is
given by
h(x,r)
=
+

r)g(x)],
g
r
g
1.
8.
PRODUCTS OF SIMPLICIAL COMPLEXES
two simplicial complexes is not directly a This fact introduces a difficulty in the study of
The
cartesian product of
simplicial complex. The difficulty is cirthe cartesian product of triangulable spaces. cumvented by the introduction of a simplicial product. This type of product arises naturally in connection with cartesian products of spaces
and their coverings by open sets. DEFINITION 8.1. Given two simplexes
sl
and s2
of dimension
p and q
respectively, the simplicial product s l A s 2 is the simplex of dimension 1 whose vertices are the pairs (A,B) where is a 1) (p l)(q is a vertex of s 2 If KI and vertex of Si and 2 are subcomplexes of
+
+
A
B
.
Si
!
and
$2 ,
then the simplicial product
simplexes
s[
K
l
A
A
s 2 consisting of all
A
s2
with
s[
K K the subcomplex C K^ s C K and
2
is
of of
2
2,
all
faces of such simplexes.
This last addition is due to the fact that a face of Si A $ 2 need not be the simplicial product of faces of s and s 2 because the vertices of Si A s 2 are the cartesian product of the vertices of s and s 2 and a subset of the product need not be a product of subsets.
1
,
t
,
The
tions.
following five
8.2.
lemmas are obvious consequences
s^
of these definin
,
is
LEMMA in KI A
l
A
if
simplex of
if
A

s 2 with vertices
,
K
2
and only
A,
A
n
(A,),
(A ,B
n
)
K
of K
with repetitions allowed, and 2 with repetitions allowed.
B,
,
B
are the vertices of a simplex of n are the vertices of a simplex
8]
PRODUCTS OF SIMPLICIAL COMPLEXES
LEMMA
8.3.
67
//
L
x
and
L
2
then L!
pairs
A L 2 is a subcomplex of K A K 2 The (K ,L ) and (K2 ,L 2 ) is defined then as
l
.
are subcomplexes of
K^ and
K
2
respectively,
simplicial product of the
1
1
(#!,,) A (K 2 ,L 2 )
=
(K,
A
K
2,
L,
A
K U K,
2
A L 2 ).
In reading
this
formula
Iff,:
it
is
understood that the operations
(K(,L{) and / 2
A
precede the
operation W. LEMMA 8.4.
(K l9 LJ >
:
(# 2 ,L 2 ) >
(/STJ,L{)
are simplicial maps, then the vertex
map
(A,B)
(/!(A),/ 2 (J5)) defines
a simplicial
map
A /2
:
/!
(tf^L,)
A (#2 ,L 2 ) >
maps (A,B)
(X{,L{)
>
A
(KJ,LJ).
>
LEMMA
simplicial
8.5.
TTie vertex
A
A
and (A,B)
J5 de/ine
maps
*,:
Ki A
#
2
> X,,
7r 2
:
tf t
#  K
2
2J
which will be referred
its factors.
to
as the projections of the simplicial product onto
LEMMA
8.6.
The
vertex
map A
(A,A) defines a simplicial
map
A:
called the diagonal
(K,L) > (K,L) A (K,L),
map.
DEFINITION
8.7.
A
simplicial
complex
K
is
ordered
if,
for each
simplex, a simple order of its vertices is given such that the order of each simplex agrees with the orders of its faces. The order in a sub
complex in K.
vertices
will
always be assumed to be the order induced by the order
A
can be ordered by selecting a simple order of its ordering the least simplex containing K). is equivalent to a binary relation Clearly an order in a complex ^ A' for the vertices of subject to three conditions: (i) A g B
Every complex
(i.e.
K
K
K
and B ^ A imply A B, (ii) A and B are vertices of a simplex of K if and only if A g B or B g A and (iii) if A,B,C are vertices of a simplex of K, and A ^ B and B ^ C, then A g C. DEFINITION 8.8. Let K and K 2 be ordered simplicial complexes. Define a partial order in the set of vertices of K^ A K 2 by (A,B) ^ (In general this is not an order in (A',B') if A ^ A' and B ^ B'.
,
l
K
}
A
K
2 ).
The
cartesian product
of those simplexes of
relation.
If
K
LI and
L2
/f 2 is the complex consisting vertices are simply ordered by this and 2 then (/f,,L,) are subcomplexes of
}
K X
X K
,
A
K
2
whose
L,
K
l
K
,
X
(/C 2 ,L 2 ) is defined as
(l^
to
X #
2,
2
VJ X,
XL
2 ).
We
proceed
now
compare the cartesian product
JK\
X K
2
of
68
SIMPLICIAL COMPLEXES
[CHAP.
II
ordered complexes 1 and 2 with the cartesian product \Ki\ X \K2 of the spaces \Ki\ and \K 2 \. Let x be any point of KI A Then the 2 projections TTI(X) and ir 2 (x) are points of KI and .K^ respectively. Thus
\
K
K
K
.
TT(X)
=
(iri(x),ir 2 (x))
yields a
TT:
map
A
!#!
K
2
>
\
Xt
X X
#2 .
If
KI and
Jff 2
are ordered, the
17:
map
2
\
TT
defines a
map
\K2
\.
\K,
XK
is
2 \.
2
\
> /q
LEMMA
8.9.
The
map
tj
K
a triangulation of
2y
onto \L t \L 2 \. has the property that, for each vertex angulation
rj
\
then
\Ki\ X \K carries \Li X L
a homeomorphism, and {rj,Ki X If LI and L 2 are subcomplexes of
K K
l
2
\
is
and
tri
X
Furthermore, this
B
spondence x
PROOF. We will be inverse to 17. uniquely in the form
(x,B) is a simplicial shall define a map
map
y:
e
of
K
\.
of K
2,
say, the corre
l
into
2
Let a
e
Jf,j,
j^l \K2
X \K We can
\
\K^
X
\K 2
\.
\K,
X K
2
\
that
ft
express a
and
a
where
A <


'
< A*, B < = 1. Let

.
< BQ
,
a (A
%
)
>
0,
0(B')
>
0,
where
g m g p
and
c
g n ^ ^
gc
5.
p+fl
Now let = cv+Q+1
,

be a single sequence obtained by rearranging the symbols a, a ap ,6, 6 in order of magnitude. For each r = 0, p q let r = (A',B') where z and j are the numbers of a'sand &'s, respectively, C
,
,
+
in c,
1

,
c'"
.
Then i
+j=
r
and
1
.
C r+l is either
+1
depending on whether c is a* or b and they are therefore the vertices of a simplex of
(A' It follows that
,B') or (A\B''

+l
),
C <
l
K XK
2.
< C Now set
,
P+Q
where c~
l
=
0.
Since
S!
2.
r
(c*

c"
1
)
=
c
p+fl

c'
1
=
1,
7
is
a well
defined point of Ki A In the definition of
K
the following ambiguity may occur: If a* = f t+1 l+1 could be either (A ,BO or (A ,B ), depending on 6', t+J>1 +l = 0, c' whether a* or V was labeled as c''*'. In either case c
C
then C**'
+1
'
8]
PRODUCTS OF SIMPLICIAL COMPLEXES
C l+
*
69
so that the choice of
has no effect on 7, which is thus defined without ambiguity. We define TJ by setting rj(a,p) = 7. To prove that rj is the inverse of 77, it suffices to establish the following two propositions:
(1)
"t?
+
l
maps
\K!\
X
\K 2 onto \K,
\
(2)
ipj is
the identity
map
of l/^l
XK X \K
2 \, 2 \.
To prove (1), consider any point 7 in 1 X 2 Then 7 = ]C*o r r r+1 Q d r D where D < D Let A < < A* be the distinct first coD' and similarly let B <  < B q be the second ordinates of D, coordinates. By adjoining extra D's with coefficient zero when necesr = (A\B') then D r+1 = (A\B i+l ) sary, it can be arranged that, if D r r+l i+l = (A ,B ). Define c  ZU>. Let C r = (A',B'), if C"* 1 = orD t+ r r + l = (A'^'* 1 ), then s^t 6' = c r (A VB'), then set a' = c if C p = 1 and 6 This defines a g ^ a g g 6 Q = 1. Finally set
K
K
.
.
,
f
f
,
.
= ZS
1
1
(a
to verify that rj(a,@)
3
a""
)^ and = 7.
,
ft
=
;
?
(6

1
"1
1
ft
)*
.
It is
then easy
To prove (2) it suffices to show that if 7 = i;(a,/3), then ^^7) = r a and 7r 2 (7) = /3. Let C C r+fc be all the vertices in C, r + = (A\B' + ) where r = C*+* with A' as first coordinate. Then C t + j and ^ i g /c, while C^ = (A*' ^') and C^*^ = (A'* ,J3 /+ *). Then from the definitions of TI and of 7,
,
,
l
l
1
1
1
l
/
7T l
\/
yi
(7)(A
)=C )=1^(C C = a  a*' =a(A').
\
X^
4
/ r
+l
r+ll\
r
+k
C
r
1
1
Since this holds for every t, it follows that ^,(7) = 0. This concludes the proof. ^"2(7)
y
=
a.
Similarly
so are
X XB).
COROLLARY 8.10. // (X,A) and (Y B) are triangulabk pairs, then (X X Y, A X B) and (X,A) X (F,B) = (X X Y, A X F VJ
A K"2 The previous discussion shows that the simplicial product contains as subcomplexes the cartesian products X K* for the l and 2 The next lemma implies that \K l A /f a various orderings of
}
K
K
K
K
\
.

and \K l
XX
2

are homotopically equivalent.
\Ki
LEMMA
8.11.
XK
1
2
is
a deformation retract of \Ki
A
K
.
2 \.
xmKiA
fore (1
PROOF. Consider the maps ?r and 77 introduced above. Then r = T7~V is a retraction of \K A ^2! into \Ki X /f 2 Further, for each K 2) the points x and r(x) lie in a simplex of K A K 2 There.
l
the desired homotopy. Both the retraction r and the homotopy remain valid for subt)x
tr(x) yields
+
complexes L! and
L2
of
K
l
and
K
2.
This implies that (K^
X K
2,
70
SIMPLICIAL COMPLEXES
[CHAP.
,
II
a deformation retract of (K v A LI A L 2 ) and that 2 is a deformation retract of (K^L^) A (K 2 ,L 2 }. The closed unit interval 7 = [0,1] may be regarded as an ordered < 1. Thus the above discussion yields the 1simplex with vertices definition of If we write X / for any ordered simplicial complex
LI
2)
X L
is
K
(Ki,Li)
X
(K 2) L 2 )
K
K
.
Ao
where
=
(A,0),
A,
=
(A,l)
of
A
is
any
vertex of
AQ,
K, then each simplex
'
K X
I has vertices
,
AO, Al,
'
'
'
,
Al
,
or
AS,
,
AJ, A
+1
,
A?,
(0
5
t
S
g)
where
A <
< A
9.
a
and are
vertices of a simplex of
K.
REGULAR NEIGHBORHOODS
is
show in this section that, if L is a subcomplex of K, then L smoothly imbedded in \K\ in the sense that L is a strong deformation
We
retract (see 1,11.6) of a closed triangulable neighborhood of L in \K\. DEFINITION 9.1. Let L be a subcomplex of K. The open set
N(L)
where the union
It should
is
=
Ust(A)
the vertices
extended over
all
A
of L,
is
called the
regular neighborhood of
of
L
in
K.
be observed that, in general, L is not a deformation retract = s be a 1simplex and L = s its boundary, For example, let = s; and the fact that \s\ is connected, and s is not, shows then N(L)
N(L).
s
K
that
is
not a deformation retract of
s.
Consideration of this example
leads to the following definition: DEFINITION 9.2. L is called a full subcomplex of each simplex of whose vertices are all in L.
K
K,
if
L
contains
Clearly, in the
LEMMA
define
9.3.
Let
above example L is not a full subcomplex. L be a full subcomplex of K. For each a
e
N(L),
In each case, the sum is taken f is a retraction of N(L) into
/
>
over the vertices of L.
\L\.
Then f(a)
h:
e L,
and
Moreover,
the
homotopy
N(L)
X
N(L), defined by
is
a strong deformation
retraction of
N(L)
into \L\.
Finally, the bary
9]
REGULAR NEIGHBORHOODS
t
71
centric coordinate of h(a,t) with respect to
function of
a vertex A is a nondecreasing in L, and is a nonincreasing function if A is not in L. PROOF. First note that /(a) is a point of the least simplex s containing a. Furthermore its barycentric coordinates are zero for vertices not in L. Thus f(a) belongs to a face 5' of s, all of whose vertices are
if
A
is
in L.
f(ot)
Since
L
is full,
/(a)
is
in L.
If
a
is
=
in L, then v(a)
/(<*)
=
1
a, so that /
s,
is
a retraction.
Since a and
are in the
and same
simplex
the point h(a,f) also lies in s. The last statement of the a direct consequence of the formulas for / and h. It follows that h(a,t) is in N(L) for all a. e N(L) and t s 7. This completes the proof. The following lemma shows that the assumption of 9.3 that L is full is not a severe restriction from the topological point of view.
lemma
is
LEMMA
division, Sd PROOF.
9.4.
// L is a subcomptex of K, then, in L is a full subcomplex of Sd K. A simplex s of Sd K has vertices 6.
,
the barycentric sub
,
b,
q
where
s<
are simplexes of and 5* is a face of s + in Sd L, then the last vertex b s<] of s is in
t
K
,
1
.
If all
the vertices of s are
Sd
L.
Then
s g is also
in L.
Hence s is in Sd S Q and thus in Sd Note that the deformation of
N(L).
retract
L.
9.3
is
not defined on the closure of
To
we
obtain a closed neighborhood of which
resort to further subdivision.
L
is
a deformation
DEFINITION 9.5. If L is a subcomplex of K, the fc regular neighk borhood N (L) is the image in K of the regular neighborhood 7V(Sd* L) N Q (L). under the natural map Sd K K. In particular N(L) LEMMA 9.6. // L is a full subcomplex of K, then a point a of K is in N (L) if and only if there is a vertex A of L such that a(A) > <x(B) for
fc
th
l
each vertex
B
of
K not in L.
A,

PROOF.
a
,
Let a
A may
e K. Then a is in a simplex s whose vertices A, Q be labeled so that a(A) ^ ^ a(A ). Let s, be the
,
simplex spanning the vertices
A*.
The
is
point a
=
1
It
(a) in
Sd K, which corresponds
to
a
(see proof of G.2),
given by
Let j be the least index such that w,
.
>
0.
If
now a
e
N
6a
,
l
(L), that
e
is,
if
t
a
e
JV(Sd L), then, for some index
i,
we must have
are in L.
Sd
L
and
w >
0.
Then
s, is

in
L and A,
J
,
A*
Since
i
^
j, it
follows that
A,
,
A
are in L,
and the
condition of 9.6
is satisfied.
if
Conversely,
in
the condition of 9.6
is full,
is satisfied,
then
A,
is
,
A
1
are
and, since and, since w,
L
L
>
0, it
the simplex .s, is in L. follows that a e AT(Sd L).
Thus b fl Thus a e
in
l
N
Sd L (L) and
the
lemma
is
proved.
72
EXERCISES
LEMMA
9.7.
//
L
is
a subcomplex of K, then
.
the closure of
Nk+l (L)
is
in N"(L) for k
=
0,1,
prove that the closure of (L) is in N(L). If L is full, this follows from 9.6 since, in passing to closure, the inequality a. (A) > a(B) becomes <x(A) ^ a (B). If L is not full, let L' with vertices in L. Then L' be the subcomplex of all simplexes of
It suffices to
is full,
PROOF.
N
l
and
N(L^=
N(L') and
N\L)
K CN
l
(L').
As observed
ct(A)
in 9.3, the
e (L) a(B) for A in L l and L is full, then h(a,f) s (L) for each t. Since the closure of (L) is in N(L), we have LEMMA 9.8. // L is a full subcomplex of K, the homotopy h of 9.3 l (L) into L. defines a strong deformation retraction of Combining this with 9.4 we obtain the main objective of this
>
homotopy h does not disturb the and B not in L. Consequently, if a
relation
N
N N
l
l
N
section:
THEOREM 9.9.
If L
is
a subcomplex of K, then
\L\ is
a strong deforma
tion retract of the closure of the second regular neighborhood N*(L).
EXERCISES
A. SIMPLEXES.
1.
If aft are distinct points of
defined
by
f(t)
=
(1
t)a
+
t).
tp is
ft
a simplex s, show that /: a 11 linear map of I into
in
s,
I
s.
s
It
is
segment in the
its
called the line segment from ratio t: (1
a to
and
f(t) is said to divide this
barycenter point of \s\.
2.
lies
that each point of 5 other than on a unique line segment from the barycenter to a
s' is
)
Show
If s is
an n simplex, and
y
two
triples
n n (M/) and (E S
~l
an (n
1
l)face of
s,
show that the
ET
)
are homeomorphic.
B. EUCLIDEAN COMPLEXES.
DEFINITION.
dimension <q.
1.
The
points
if
A,
,
Aq
of euclidean nspace
Rn
are
called linearly independent
they are not contained in a hyperplane of
Show
that
A,
,
AQ
Oil
are linearly independent
if
and only
if
the matrix
"
* >
O
n
where
2.
a{,
,
a are the coordinates of A', has rank q
if
,
+
1.
Show
that,
A,
a

,
A
Q
are linearly independent, then every
subset of
A,
A
is
also linearly independent.
EXERCISES
3.
73
tt
be a gsimplex with vertices A, A Let /: s q > R* be a linear mapping. Show that / is a linear imbedding if and only if Q the points f(A), f(A ) are linearly independent.
Let
s
9
,
.
,
4.
Show
Using
that
Rn
contains an infinite sequence of points each
n
+
1
of
which are
5.
linearly independent.
4,
in
n
R^\
.
prove that each ncomplex
K admits a linear imbedding
,
,
Let A, A* be linearly independent points of Q convex set a containing A A q is called the eu~ Q clidean qsimpkx with vertices A, A The faces of <r are the euclidean simplexes whose vertices form a subset of A, A*. A = {a} of euclidean simplexes in R n is called a euclidean finite set complex if the intersection of any two simplexes of S is either vacuous or is a common face of both of them. n > R 6. Show that if /: is a linear imbedding of a complex into n R then the images of the simplexes of form a euclidean complex in R n Conversely, if S is a euclidean complex in R n then there is a > R* such that and a linear imbedding /: simplicial complex the simplexes of S coincide with the images of the simplexes of K.
DEFINITION.
,
jR
The
least
,
.
,
K
,
K
.
,
K
K
C. SPACES
WITH OPERATORS.
group (not necessarily abelian, and written said to operate on the space multiplicatively) if, for every w e is defined such that w 2 (w l x) = (w 2 w )x, and x e X, an element wx e
DEFINITION.
is
A
W
X
X
}
X. If and that x wx is a continuous map w: X on X and Y, then a map /: X > Y is called equivariant if operates f(wx) = wf(x) for all w e W, x e X. TF is said to operate on the simplicial complex K if, for each w e IF, /f is given such that the fixed points of w a simplicial map 10: /C form a subcomplex of K and w 2 (wiz) = (w 2 Wi)#, la; = rr for each x
Ix
x,
=
W W
e#.
If
is
TF operates on X, and
if,
T =
{/,/f } is
a triangulation of X, then
the
T
said to be invariant
for every
w
s
W,
map
l
t~ wt:
K
>
K is
simplicial
1.
then
W
w
points form a subcomplex of K. If on X, and an invariant triangulation of operates such that contains an invariant subgroup G
and
its fixed
W
W
X exists,
is finite,
and
x
=
W/W
x for
all t0
e
TT
,
a:
e
X.
then so
is
2. If the triangulation subdivision of T.
T of X is invariant,
the barycentric
and X' that operates on both 3. In theorem 7.3 assume that 7 f X' is the triangulations T and T are invariant, and that /: Show that the simplicial approximation can be chosen equivariant.
',
W
X
X
equivariant.
74
EXERCISES
D. JOINS OF COMPLEXES. DEFINITION. The join K o of two disjoint complexes is the comB* of the A r ,B plex whose vertices form the union of A and vertices of A set of vertices are those of a respectively. and M, if not vacuous, span simplex of the join if its subsets in
M
1
l
K
M
,
,
,
,
K
simplexes there.
1.
K and M are subcomplexes of K
If s t is
o
M.
s v o s 2 is
2.
a psimplex and
its
s2
a gsimplex, then
o s2 )
simplex, and
3.

boundary
is (s t
U
a (p
+q+
1)
(si
o
\
If

is
4.
Let
L
2
a deformation retract of \K o be the subset of points a. of
K o M such that
to \K\
M
2 ).
\M\.
(') =
]^ a(A')
=
that K o
M
V2.

Show
jRC
that
Af

L
is
homeomorphic \M\. homeomorphic to the product of L with
is
X
Show
an open
5.
interval.
2 dim /v. In Sd A', let L(resp. A/) be the subcomwhose vertices are barycenters of simplexes of dimenp(resp. > p). Show that Rd K is isomorphic to a subcomplex
Let
^ p
plex of simplexes
sions
of
^
o
L
6.
M.
Define the join
o Y of two spaces; and to \K o M\. homeomorphic is contractible on itself to a point, 7. If o Y. true of
X
show that
\K\ o \M\ is
is
X X
show that the same
8.
Show
Show that the join of a pcell and a gsphere is a (p that the join of a psphere and a gsphere is a (p q
+ q + l)cell. + + 1) sphere.
E. ELEMENTARY SUBDIVISION OF A COMPLEX. 1 s is the DEFINITION. If s is a simplex of K, the complement that do not have s as a subcomplex consisting of all simplexes of face. The boundary of s, written Bd(s), is the closed subcomplex ~ s which are faces of simplexes not in consisting of simplexes in with respect to s, written The elementary subdivision of s.
K
K
K
K
K
Sd.(#),
1.
is
s.
the complex
(K
s)
W
K
(6.
o Bd(s)),
where
b. is
the baryidentity
center of
Show
that the linear
map of Sd.CK)
is
K defined
,
by the
transformation of the vertices
2.
sf
is
^
a homeomorphism. s k so that dim If the simplexes of are ordered Sj,^, dim Sj+i for each j, show that the bary centric subdivision of
K
K
the result of the succession of elementary subdivisions
Sd.^Sd.^^.Sd.. (K)
F. INFINITE COMPLEXES.
...))
DEFINITION.
infinite
Let
W be an
infinite set of objects called vertices.
An
complex dimensional) simplexes whose vertices are in
is
K with
vertices in
W
an
infinite collection of (finite
W,
subject to the condition
EXERCISES
that a
fa*ce of
75
a simplex in the collection is also in the collection. Each is described by bary centric coordinates a (A) for point a of ^ A e W, where a(A) > only for a finite number of elements of
K
is
W
y
and
^x
oi(A)
=
1.
vertex of
If
a vertex of at most 2 are points of K, then p(a,0) = l(A)  0(A)] )* is a distance function in K, and defines the metric topology. The weak topology in / is obtained by considering as open all sets whose intersection with every (closed) simplex s of is an open subset
K
The complex
said to be locally finite if each a finite number of simplexes of K.
is
K
a and
ft
{^
K
of
*.
Thus
\K\ m
1.
to the infinite complex correspond
\K\ W
.
two topological spaces
and
2.
is
The identity map \K\* \K\ m is continuous. The weak and the metric topologies coincide
if
and only
if
K
locally finite.
3. Linear and simplicial maps (defined as in the finite case) are continuous in both topologies. In particular, for any vertex A, the barycentric coordinate a(A) is continuous in a for both topologies. A subcomplex is closed in either topology. For every vertex A of K, the set stCA) is open in either topology.
consisting of an infinite complex
is
an infinite complex K is a pair K' and a linear map /: K' > K which a homeomorphism in the weak topology. 4. If (K' l) is a subdivision of K, and L is any finite subcomplex
DEFINITION.
A subdivision
(K',l) of
y
of
K
}
then
l
l"
(L)
is
a
finite
5.
is
Given a family {U}
(K',l)
is
a subdivision
in
subcomplex of K'. open sets covering the space \K\ W there of K such that, for every vertex A' of K' the
of
,
t
set l(st(A'))
one
of the sets of the family
infinite
T =
DEFINITION. {t,K} where
An
K
is
an
infinite
morphic mapping of DEFINITION. If
T =
\K\ W
is a pair complex and t: \K\ W > X is a homeoonto X. \t,K} is a triangulation of X, and {U} is a
U\. triangulation of a space
{
X
such that, for every vertex A of K, family of open sets covering the set t(st(A)) is in one of the sets of {[/}, then the triangulation T
is
X
said to be finer than the covering
6.
{
U}
.
be a space which is (Simplicial approximation theorem). Let 1 as an infinite complex, a space with (infinite) triangutriangulable > X' be given. lation T', and let a map /: There is a covering { U\
X
X
X
f
of
X
finer
by open than {[/}, there
If
sets such that, for every triangulation
is
T
a
map
)
g:
X
X
1
which
of
is
which is of a simplicial apis infinite,
X
proximation (relative to T,T
7.
K
and
then the join
KoM
M
to /.
are
is
nonempty complexes one
not locally
finite.
which
CHAPTER
Homology
III
theory of simplicial complexes
1.
INTRODUCTION
In this chapter it is assumed that a homology theory is given such that the groups q (X,A) are defined for all triangulable pairs and that > (F,J3) of one triangulable /, is defined for every map /: (X,A)
H
pair into another.
In other words, let 3 denote the category of triangulable pairs and maps of one such pair into another. It follows from u,8.10 that 3 is an admissible category for homology theory as defined in i,l. It is assumed in this chapter that the homology (cohomology) theory considered is defined on an admissible category Ofc which contains
3.
The
first
main objective
axioms of the
classical algorithm
of this chapter is the derivation from our used to define and compute the
homology groups
of a simplicial complex.
This motivates our use of
a uniqueness theorem which
this algorithm in the existence proofs in later chapters.
The second
asserts that
objective of this chapter
is
any two homology theories, with isomorphic coefficient are isomorphic on the category 3 of triangulable pairs. This groups,
asserts, to
a certain extent, that the axioms are categorical,
i.e.
ad
mitting an interpretation unique up to isomorphisms. This be extended later (Chapter xn) to a larger class of spaces.
result will
As
in
Chapter
i
the statements concerning cohomology are listed at
the end of each section, without proof.
2.
EXCISION AND DIRECT
SUM THEOREMS
pairs,
It will
be shown here that, for triangulable
are valid in
i.
l
excision
and
direct
sum theorems
2.1.
more general
situations than those
encountered in Chapter
THEOREM
the inclusion
//
K
and
K
2
are subcomplexes of a complex
K, then
map
i:
(K^K,
K C
9)
(K, \J
KM
(liL;/iLi,/iC 2 )
induces isomorphisms in each dimension.
is
Consequently
a proper
triad (see i,14.1).
76
2]
EXCISION AND DIRECT SUM THEOREMS
Without
77
PROOF.
K=
KI \J
K
2.
Let
X
loss_jof
=
= C\ JBT 2 Then (X,A) is a trineighborhood of lt and let A and A correspond to subcomplexes of the angulable pair, since both second barycentric subdivision of K. The map i may be factored into
.
K
generality we may assume that N*(Ki) be the closure of the second regular
X
X
inclusions
JT.I)
(x,A)
Since #
\K\
= #'(#0
U (K 4 =
\K,\)
C Int X U Int tf
#2
,
2 ,
it
follows that:
=
X VJ
# 2
,
XH
#
i2
=
Int
X VJ
Int
ff a .
Thus, by the Excision axiom, the
dimensions.
this
map
induces isomorphisms in
all
n,9.9, \Ki\ is a strong deformation retract of X. Throughout deformation each point of \K9 remains in \K2 \. Thus the pair Hence, by 1,11.8, (l/filJ/fj C\ jfiCal) is a deformation retract of (X,A). the homomorphisms induced by ii are isomorphisms. THEOREM 2.2. Let (X,A) be a triangulable pair, and let U be an such that U A. If (X open subset of [7, A U) is a triangulable
By
\
X
C
(X f: an isomorphism in each dimension.
pair, then the inclusion
map
C7,
A
U)
C
(X,A) induces
let
PROOF. Let T be a triangulation of (X,A) (see n,5.1) and V and A be the interior of A (relative to X). Then
V
X
V
are
subcomplexes of
inclusion
X
relative to the triangulation T, so that
by
2.1 the
(X,A) induces isomorphisms.  U, A 1 = A  17, Set X' = be the interior of A' relative  7 ; A'  F ; ) to X'. Then, by the same token, the map i': (X' F') and 7', A' CST',A') induces isomorphisms. Since the pairs (A"'
map
i:
X
(X
F,
A
V) and let
C
V
i'
,
C
=
(X
1
F,
18
A
7)
2.3.
coincide,
and since
=
/i, it follows
that /,
i*i*
an isomorphism.
Let
THEOREM
K
be
a simplicial complex with subcomplexes
K
lf
,
K ,L such that K = K, U
r
U K U L,
r
KiCMttCL,
(Jf,L).
for i
^
j.
Le*
L = J<
f
n L, and
fe*
*,:
(K,,L.)
:
C
/orm an injective representation of H q (\K\ \L\) as a direct sum. L and PROOF. The case r = 1 is covered by 2.1 since K = KI LI = KI C\ L. Assume, inductively, that the theorem has been proved
,
TAen, <Ae homomorphisms k
r)
i)tt
H^K^lL^)
^ ^(l/flJLI)
y
(t
=
1,
U
78
for r
HOMOLOGY OF SIMPLICIAL COMPLEXES
= n ^ ^J L and
1,
[CHAP. Ill
K
and suppose that
r
= n+
maps
J
1.
Set K'
=
KI
U


U
n),
H
consider the inclusion
*'<
(K ,Lj >
t
(K',L) > (K,L)
(t
=
1,
,
ft'
(tfn+1 ,Ln +0
>
1,
(K^
,
U L,L)
n,
/
> (X,L).
+
ljt
= ^ftj^ for i Clearly A; k is an isomorphism. 2.1,
ijt
=
and
ft n
Since (K;K',/f n +j
W L)
= jk+.
is
Further,
by
a proper triad
with
K
it
1
\J
(K n+l \J L) = X,
and
.7"^
K'
n
(X^, VJ L)
=
L,
follows,
from
of
H
a (jfiC,L)
r.14.2, that j^ as a direct sum.
,
1, k(^ (i a direct sum.
=
yield an injective representation the inductive hypothesis, the maps By n) yield an injective representation of // a (#',L) as This implies the conclusion of 2.3.
THEOREM
of H
Q
2.3c.
The maps k* of 2.3
yield
a
projective representation
(\K\,\L\) as a direct sum.
3.
THE INCIDENCE ISOMORPHISM
S
q
Let
s
be a gsimplex,
s
Q
its
Let
in s
A
q
,
boundary, and s^ a "
1
""
,
1
(q
l)face of s
c
.
be the vertex of
i.e.
q
c
Q~
opposite s
s
a
and
1
c
the closed star of
all
A
l
is
cepting
s
and
q~ l
the subcomplex of
.
consisting of
faces of s
c
ex
s
Thus
For each a
e
s, the function
(1

)a
(Is
+
tA,
1
^
!)
t
^
y
1
is
a homotopy contracting the pair Hence, by i,11.5 and 1,9.5 "
],^"
1
into the point pair (A A).
LEMMA
3.1.
Is'lJc
Next we turn
2.1, this is
are homologically trivial. !) '1 1 to the discussion of the triad (Is'l;^ !.!^" )).
!
1
and
(Is*!,)^""
By
1
a proper triad. THEOREM 3.2. The homology sequence of
to the
1
the triad
(Is'l;!**""
!,^*""
!)
reduces
isomorphism
I
d:
H.('
>
A')~H
_ 1 (^
f 
,*
).
All other groups in the sequence are zero.
3]
INCIDENCE ISOMORPHISM
THEOREM
3.3.
79 are as follows:
The homology groups of
ff f (
f

a
a
)
(s ,s
f
f
4
Wl,l*'l) =
Pfeoops.
1
)<7
forp*
for all
q,
q.
By
1
3.1,
H
9
l
9 (\8 \,\c*
\)
=
and
d:
the exactness of the H.S, of the triad,
flp^ds'"" !,!^"
!).
we have
#(l*l>l*l)
3.3 follows.
=
H
Iterating this isomorphism yields
p . Q (\s\).
H # (s,s
p
therefore,
q
\) )
by
9 (\8*\ t \&
a
~ =
Since
l

is
a single point, the assertion of
This also establishes 3.2. Q~ If s 3.4. is a (q  l)face of the ^simplex s'(q the incidence isomorphism
DEFINITION
>
0),
f
[<:']:
is
ff,(l',l*
l)
=
ff..d*''!,!*'"'!)
1
defined to be the isomorphism d of 3.2. defined from the diagram
Explicitly, [s'ls*'
]
may
be
f
ff.(',a
l)
ff..d*'!)
^
1
ff..d^Uc
!)
^ ff..d.'U*
1
!)
where
t,j
are inclusionsTby setting
> // g > 0, c^td /: (Is'Us'P (sj,sj) is such that l l and \c are mapped into \sl~~ and \cl~ \, respectively, then com\s*~ mufativity holds in the diagram
LEMMA
l
\
3.5. q~ l
\
\
q
f
//.(l
,l
)

?/..( I*'"
I,
I*"'
I)
where f\
defined by f. incidence isomorphism is the boundary operator of the 1 triad (Is*!;!^"" !,!^" !), and the commutativity of this boundary operator
is the
map
PROOF.
The
1
with induced homomorphisms
1,14.5.
is
contained in the statement of
Theorem
DEFINITION
3.6.
If
A
is
a 0simplex
.
(i.e.
denote the element of
H
,
(A)
mapped on
g
a vertex) and g s G, let gA by the isomorphism in
duced by the map A > P (In the notation of 1,7.1, gA 1 Let s a be an ordered ^simplex with vertices A < A <
Define
k
=
(gA) A .) < A*.
s
(k
=
0,1,
q
1)
to be the ordered simplex with vertices
80
HOMOLOGY OF SIMPLICIAL COMPLEXES
fc
[CHAP.
S
Ill
A *t < 4^ads'Us'l)
(l)
+i
<
...
<
k
^c
For each g
g
G> the e i ement
Of
is
defined inductively
gs
by
= [<VT
V
1
.
Thus
THEOREM
g
>
a
3.7.
0s
is
an isomorphic mapping
PROOF.
The
the correspondence q q onto the group Q (\s \,\s \). proposition follows from the fact that the incidence
s
,
For every ordered qsimplex
of G
a
H
isomorphisms are isomorphisms, and g
>
Q gA maps G isomorphically
onto# (A') (See i,7,2). THEOREM 3.8. // Si and 2 o.re ordered qsimplexeSj and f = > (s 2 ,s 2 ) preserving the order then f+(gsi) plicial map (s ,Si)
}
is
a sim.
y
gs 2
,
Consider the simplexes s< (i defined in 3.6. Since / is orderpreserving,
PROOF.
=
1,
2;
n =
1,
q) as
it
defines
maps
/:
Consider the diagram
I/a*
I/.1.
1/0,
where the horizontal arrows are the incidence isomorphisms. By 3.5, maps commutativity holds in each square of the diagram. Since / into gA\ (see 1,7.2), it follows that f<^ maps gsl into gs%. gA\ THEOREM 3.2c. The cohomology sequence of the triad (Is'Us*" !,^*" !)
Olt
1
1
reduces to the isomorphism
d:
q
H
is
q" l
q" l
q" l
(\8
\,\8
\)
H
g
a
(*,s
).
THEOREM
3.3c.
H'(\8
\,\?\)
q~ l
G, #
p
(s',s
a
)
= Oforp ^
q.
DEFINITION
3.4c.
If s
a
(q
l)face of
a gsimplex
s
q
,
the
incidence isomorphism
[V]: HXIs
is
1"1
1
1
!)
!,!*
Jsrd^U*
!)
defined as the isomorphism 5 of 3.2c or equivalently as
LEMMA.
3.5c.
Under
the conditions of 3.5
/*[?'!]
=
l
[s" :s']/f
4]
AUTOMORPHISMS OF A SIMPLEX
81
DEFINITION 3.6c. If A is a vertex and g e (?, let gA e H(A) be the of g under the isomorphism induced by the map A P (In q the notation of i,7.1c, gA = g A (A).) If s is an ordered gsimplex with 1 vertices A < A < < A 9 and e G, define 0s a e #'(* a ,') in
image

ductively by
Thus
<
ff5
=
[
5
<~
V]
[4':
V.
a
THEOREM 3.7c. ontoH 9 (\s Q \,\s<\). THEOREM 3.8c.
(s 2 ,s 2 ) is
TAe correspondence g
//
s l} s 2 are
0s
zs
an isomorphism of
(si,$i)
G
>
ordered qsimplexes, and f:
simplicial
and order
preserving, then f*(gs 2 )
=
g$i.
4.
AUTOMORPfflSMS OF A SIMPLEX
q 7
In the preceding section we introduced the symbol gs e H^ds'lJs )), which is a function of g e G and of the order of the vertices of the qQ Our immediate objective is to determine the extent of the simplex $
.
dependence on the order. The result obtained (4.3) is fundamental in connecting our axiomatic system with the classical homology theory, in As a preliminary, particular, with that part dealing with orientation. of a space consisting of two points (the 0we study the group
H
dimensional sphere) in greater
detail.
THEOREM
4.1.
Let
A
and
B
e
1
be the vertices of the lsimplex s
.
We
shall use the simplified notation:
S
Q
notation of 1,7.1, every element h
H
(the Qsphere). Using the \s can be written uniquely as (S)
\
=
l
h
=
if
(gA) s .
+
(g'B) s .
g,g'*G.
The element h
is
in
S
Q
(S)
h
and only
if g
+
g'
=
0, or, equivalently y
The first part of the theorem is a direct consequence of i,7.1, and of the direct sum theorem i,13.1. To prove the second part, con> P where P is the space consisting of a single S sider the map /: point that was used to define the coefficient group G (i,6.1). Then, by
PROOF.
1,7.2,
g'
so that f+h
=
if
and only
if
g
+
g'
=
0.
82
HOMOLOGY OF SIMPLICIAL COMPLEXES
[CHAP.
Ill
DEFINITION 4.2. An automorphism of a simplex is any 11 simplicial map of the simplex on itself. By n,4.4, the automorphisms of a simplex are in 11 correspondence with the elements of the group of permutations of its vertices. Any
permutation is a product of simple permutations, each of which interchanges two vertices and leaves all others fixed. In general, a permutation can be expressed in many ways as a product of simple permutations; but the number of such factors is either always even or always odd. Accordingly the permutation is said to be even or odd. The same adjective is used to describe the corresponding automorphism.
THEOREM
simplex
st
4.3.
e
Let f:
// a (s,s).
(s,s)
(s,s)
be
an automorphism of
the q
and h
Then
f
m
=
\
h
tf
f
is is
w*
odd.
1
h
if
f
PROOF.
Let
For q
l
=
0,
the theorem
is trivial.
Consider the case q
:
=
1.
5 =
by
s

defined
/.
consist of the points A and #, and let / If / = identity, the result is trivial; thus we
S
>
S be
in the
that f(A)
= B and
f(B)
=
may assume
A.
Then commutativity holds
diagram
(see 1,8.4):
o.
Let
ft
eff^lAS
):
then,
by
4.1,
dh
for
=
(gA) s .

(gB) s
<.
some
g
e
G, and,
by
1,7.2,
5/fc
Since
= fojh =
1
(gf(A)) 8 .

(gf(B)) 8 .
is
=
(gB) s .

(gA) a *
=
dh.
^ids
==
!)
0,
the kernel of d
zero;
hence f+h
=
h.
1, Suppose, inductively, that the theorem is true for the integer q q ^ 2. Since every permutation is a product of simple permutations, it is sufficient to consider the case of a simple permutation / on the
vertices of
s.
AQ
such that f(A Q )
4
.
vertex
has more than two vertices, there is a vertex Let s' be the (q l)face of s opposite the Then / maps s' onto itself and defines an automorphism
Since
s
= A
.
/':
(s'X)^(s'/)
4]
AUTOMORPHISMS OF A SIMPLEX
83
Moreover, by 3.5, the commutativity relation [sis']/* = /*[s:s'] holds. Let h e Since /' is a simple permutation of the vertices of a (s,s).
#
s',
and
4.3
is
assumed
for the dimension q
1,
we have
This implies
and, since
the
THEOREM
an isomorphism, f^h = h. Let Si and s 2 be two ordered simplexes same unordered qsimplex s. Then
[sis'] is
4.4.
both carried by
according as the order of the order of s^
s 2 differs
by an even or an odd permutation from
Si
PROOF. Let / be an automorphism of (s,s) which maps $ 2 onto an order preserving fashion. Then, on one hand, 3.8 yields
/*(0$2)
in
=
gs\
On
the other hand, 4.3 implies
Consequently gs
THEOREM
A
9
,
and
let
gs 2 Let s be an ordered simplex with vertices A < k s k be the face obtained by omitting the vertex A and not changing
l
.
=
4.5.

<
the order of the others.
Then
[s:s k ]gs
= (l)V*.
The
s
PROOF. If fc = 0, this is formula (1) of 3.6. as follows: Let be reduced to the case fc = obtained from s by moving the vertex simplex
others.
general case can be the new ordered
in front of all the
Ak
Then
sk
=
SQ
and gs k
=
gs
,
while,
by
4.4, gs
=
(
l)*0s.
Hence
Every element h e H(S?) is uniquely described by = g' of G (see i,7.1c), and every pair the elements h(A) g and h(E) The element h is in the subgroup G s * (see thus be obtained. g,g' may
4.1c.
THEOREM
=
i,7.3c) if
and only
if g
=
g'.
5,
THEOREM
4.3c.
// / is an automorphism of the qsimplex
and
h*H*(\8\,\8\),then
h h
if if
f
is
odd.
84
HOMOLOGY OF SIMPLICIAL COMPLEXES
[CHAP. Ill
THEOREM 4.4c. // $i,$2 ore ordered q$implexes an the same unordered gsimplex s, then gs l = gs 2 according as the orders differ by an even or odd permutation.
THEOREM
4.5c.
//
S,S A are
as in 4.5, then
6.
CHAINS IN A COMPLEX
DEFINITION
skeleton
5.1.
is
If
K
is
a simplicial complex, the qdimensional
K*
of
K
of
K of dimensions g
the subcomplex of
q.
K consisting of all
the simplexes
In this and the subsequent sections, (KjL) will stand for a pair and a subcomplex L. consisting of a simplicial complex Q~ l LEMMA 5.2. If p * q, then H^K* VJ L) = 0. L\, \K
K
V
PROOF.

Let

s if
.
,

sr
9 VJ s VJ K VJ L and s m = s m Then K q VJ L = * VJ (K*~ VJ L) = 1, Thus the direct sum theorem 2.3 can be applied. (m r). = for p 7* q, the proposition follows. Since, by 3.3, l/p (s m ,s m DEFINITION 5.3. The group of qchains of (K,L), denoted by Ca (A:,L), is defined by
1
be the gsimplexes of '
r
K which are not in L.
H
l
,
)
C q (K,L) =
If /:
J5T f (
U L,
(IH
9"
1
\K
U L).
1
(X,L)
>
(KuLi)
is
1
simplicial, the
map
(K
9
U L,^^ U L)
/.:
U L^K; U LO,
f
that / defines, induces homomorphisms
C,(K,L)
 C
5
(K lf L,).
by /. and for q > dim
They
are called the c/iam homomorphisms induced
5.4.
THEOREM
PROOF.
C Q (KjL) = 0/or
0,
<
A".
then C,(K,L) = ff f (L,L), while, if q > dim If, = a (X,JK:). In either case C.(K,L) = by i,8.1. then C<(K,L) A a be a finite sequence of vertices DEFINITION 5.5. Let A, of some simplex of (with possible repetitions). For each g e G define A a of C a (l?,L) as follows: Let s be a simplex with the element gA < B\ Let /: (s,s)  (IT VJ L, IT' 1 ordered vertices B < L) be the simplicial map such that f(B') = A\ Then
If q
<
H
,
K



U
gA
It follows
...
A*
=
from 3,8 that the definition of gA
s.
A
c
is
independent
of the choice of
5]
CHAINS IN A COMPLEX
5.6.
85
THEOREM morphism G
The assignment of g
i.e.
to
gA

Aq
g.2
defines a
homo
C q (K,L),

A =
q
gi
A


A +
9
,
A

A*.
If
*
'
*
iot
>
*f
w
a permutation of
the array 0,
g,
2Aen
0A'
4' f
=

0A
...
Aq
according as the permutation is even or odd. A 9 then gA twice in A, Aq = ,
,
0.
simplex of L, then gA

AQ =
If some vertex occurs at least A* are in a 7/4,
,
0.
PROOF. The and 4.4. If A,
then the
first
,
and the second part follow respectively from 3.7
involve a repetition or are in a simplex of L, "1 q~ l > (K< \J L, \J L. map /: (,) L) carries s into "1 1 VJ L, K'" Thus / may be factored into maps (,)  (A: L) ~l '1 a 9' 1 VJ L). Since VJ L) = (K (see 1,8.1), L, 9 (K L,
A
q
K
U
K
fl
U
U
K
H
,
U
K
it
follows that gA*
A*
=

0.
THEOREM
in L.
<
Let s^ s a , 6e i/ie qsimplexes of which are not each simplex s m an order of its vertices A^ < Suppose that for AZ has been chosen. Then each qchain c of (K,L) can be written
5.7.
K
uniquely in the form
c=
ml
g m Ai
..
Ai,
gm
l
eG.
q
PROOF.
Let

L =
direct
,
U
U
im :
(s m ,s m )
at
UK
"1
C
U
(K
q
U
and
L,
sm
K~
q
\J L).
Since
8' 1
L
=
sm
A
(X
U
K U
L), the
sum theorem
2.3 can be applied to yield a
unique representation
By 3.7, m A^
One
each h m can be written uniquely as h n
=
gm s m
.
Since i m +g m 8 m

=
A*
A^ by
of the consequences of 5.7
Definition 5.5, the conclusion of 5.7 follows. is that the symbols gA
generate the group
Observe further that, using 5.6, any linear combination (with integer coefficients) of such symbols can be brought to the normal form given in 5.7. Thus the group C 9 (K,L) may be
C a (X,L).
A where g regarded as the group generated by symbols gA with relations given in 5.6. and A, A q are in a simplex of
y
,
q
e
G
A*
K
}
THEOREM
is
5.8.
7//:
(K,L)
>
(Ki,Li)
is
simplicial
and gA
Q
a qchain of (K,L), then
f.(gA
A')
= gf(A)
/(A).
PROOF.
Let s be an ordered (/simplex with vertices
B,
,
B*,
86
HOMOLOGY OF SIMPLICIAL COMPLEXES
let h:
l
[CHAP.
Ill
and h(B
(s,s)
(K
9
U
1
L,
K
q~
l
U

)
=
A*.
By
Definition 5.5,
gA
L) be a simplicial A q = h^(gs). Let
map
with
J:
be the
(K< VJ L,/^
U L)
> (K? VJ L,,lCr l
U
1)
map
defined
by
/.
Then /
=
,/,
and
/.(0A
A')
= =
/^(gs)
= JA(0s) = 9f(A) (Jh)*(gs)
Consider the inclusion maps
L The symbol 0A
K
> (#,L).
A
c
then represents an element of the group
C
Q
(K)
as well as of the group
C
Q
(K,L).
q
If in
addition
A,
,
Ac
are in a
simplex of L, then gA To avoid all ambiguity
also represent an element of C Q (L). A Q ) K or (gA A a ) /r,x, > write (gA Q a or (gA A ) L to indicate in which group the symbol gA A 7 is to be taken. From 5.7 follows
A may
we may


(
As an immediate consequence
of 5.65.8
we have
THEOREM
5.9.
The sequence
is exact
C.(L) ~> C.(K) ^ C,(K,L) ~>
LEMMA
image of i q is a direct summand of C q (K). v q 9' 1 = 0. // p * q, then L) (\K L, {K DEFINITION 5.3c. The group of ^cochains of is denoted by
and
the
5.2c.
H
U
U
K
C\K,L) and
is
defined
by
C q (K,L) = H*(\K*
If /:
U L,
the
(If?
a" 1
X
U L).
1
(Jt,L)
>
(JfiC^Li) is simplicial,
1
map
(K VJ L,/^"
that / defines, induces
Q
U L)
>
U L^^r U L,),
homomorphisms
f:
rCK^LO^r^L).
homomorphisms induced by /. = Ofor q < or q > dim K.
e
They
are called the cochain
5.4c.
THEOREM
C
Q
DEFINITION
5.5c.
(K,L) Let c
C 9 (K)
be a ^cochain of (K,L), and
5]
CHAINS IN A COMPLEX
,
87
A,
Aq
a
finite
.
possible repetitions)
sequence of vertices of some simplex of Define
K
(with
c(A,
as follows:
,
A9 =
)
g
E
G
,
be a simplex with ordered vertices B < < Ba ~ a q l and let /: (s,s) > (X L, L) be the simplicial map such s G is the unique element satisfying that f(B*) = A*. Then
Let
s
U
K
U
f*c
=
0s.
s.
By
3.8c, g is
independent of the choice of
5.6c.
,
THEOREM
(c,
2
+ c )(A,
*
>
A
9
}
=
Cl
(A,
.
.
,
A<)
+ c (A,
2
,
A<).
,
A*') = 0, g, Men c(A", as the permutation is even or odd. If some dbc(A, ) according A Q ) =0. A a then c(A, vertex occurs at least twice in A,
If
*o>
%
,
^s
a permutation of
,
A
c

,
,
,
//
A,
,
AQ
THEOREM
not in L.
are in a simplex of L, then c(A, A') = 0. which are Le s l5 s aq be the qsimplexes of 5.7c.
,
,
K
Suppose
c s
that for each s m
an order
of
its vertices
1,
,
A^ <
there exists
<
a
A
has been chosen.
Then,
if g m s
G for
m =
gm
,
a
fl ,
unique qcochain
C(K,L)
such that
,
c(A,
A) =
m=
1,
,
aq
.
C In view of 5.7c and 5.6c the group C (/C,L) may be described as the group of functions with values in G defined for arrays of vertices A Q which are in a simplex of and subject to the conditions A, listed in 5.6c. In view of 5.7c, we may also use the linear form notation
,
K
for cochains;
i.e.
the symbol
A^,
can be interpreted to mean the cochain c which has the value g m on A^ for each m. As the next theorem shows, the functional notation for cochains is much more convenient. THEOREM 5.8c. // /: (K,L) > (K^LJ is simplicial, c e C a (K\,L,)
,
and
A,
,
A
Q
are vertices lying in a simplex of
K, then
,/A).
(fc)(A,
,A) =
c(fA,
..
THEOREM
5.9c.
The sequence
> C'(K,L)
is exac<
 C(K) 
f
i"
C"(L)

and
<Ae
image of
f is a direct summand of C'(K).
88
HOMOLOGY OF SIMPLICIAL COMPLEXES
6.
[CHAP.
Ill
THE BOUNDARY OPERATOR
The boundary operator
for chains
DEFINITION
6.1.
is
defined to be the boundary operator of the triple (\K ~
9
q
U L, \K ~ U L,
9
l
1
2
\K
U L).
Explicitly d Q
is
the composition of the
homomorphisms
d
H (\K* U
Q
1
L, IK'is
U L)
6.2.

1
ff^dK''
U L)
k>
 ff^dK'"
U
2
L,
<
U
where k
the indicated inclusion map.
d,d a+l
THEOREM
PROOF.
=
0.
If
fc
fl!)t
dQ
=
fc c _
ljc
d and d a+1
=
k^d', then d,d fl+
i
=
k^^dk^d'
.
Here d and
are the
homomorphisms
a
ff f (K
f
U L)
JK'"
*<*

ff,(K' U L,
f
!
K
U L)
>
H^K*' U L).
1
Since these are two consecutive homomorphisms in the H.S. of the pair
(K<
U L,
^
1
U L), we have
If /:
afc,,
=
and thus d d a+1
fl
=
0.
THEOREM
6.3.
(K,L) * (K^Lj)
is simplicial, then
f q ^d q
=
dJ
Q
.
This
is
a consequence of corresponding commutativity theorem
(i,10.3) for triples.
In order to give an explicit form to the boundary operator d q for
chains,
we
introduce the following notational convention.
If A
A"
is
an ordered array
of vertices, the
same
series of vertices
with a circumflex over one or more vertices means the ordered array obtained
from
the first array by deleting those vertices with circumflexes^ e.g.
A QA
THEOREM
l
..
A*.
An =
AM
2
A''A**
1
A".
6.4.
The boundary operator for qchains d q
satisfies
d q (gA*
. A<;=
kO
(l)V
&
a
,
A
9
.
PROOF.
^simplex
s
Assume
first
that the complex
K coincides with the ordered
and that
with vertices
q~
l
A <
sk
.
< A
L =
0.
Then
K
9
=
K
vertices
= s and K = s. Let A < A* < Aa
all
denote the ordered simplex with Define c k as the subcomplex of s
consisting of
faces of s except for s
and
sk
.
The
proof
is
based on
6]
BOUNDARY OPERATOR
89
where
all
maps
that occur are either d or are induced by inclusions.
2
).
Now
(1)
d,(gs) is in ^,_,(s,s'"
By
A'
;
theorem
5.8,
d.(j)
=
EM
/*./.'
'
A' =
X
k,
(tjj),,,
for
then
some elements g, s G. Si C c k and the map
Now
may
apply l k + to both sides of (1). If i j be factored into inclusion maps
Since,
by
i,8.1,
^i(c*Uc 4 )
=
0, it
follows that
=
0.
Therefore
(1) gives
(2)
lk+
Now
(3)
factor d a into t^d,
and use
Z*i'
=
z*.
Then
(2)
reduces to
M(p*) = j^(g k s k ).
jf^
is
Now
an isomorphism, and by
3.4,
jk&*d =
is
[s:s fc ]
the incidence isomorphism.
Hence
(3) gives
If this is
k
(
compared with 4.5 and
3.6,
we
Since this holds for each A:, (1) a simplex. A Q be a tfchain in an Returning to the general case, let gA Let s be an ordered ^simplex with vertices arbitrary pair (K,L). B < < B 9 and let /: s > (K,L) be a simplicial map such that
l) g.
obtain immediately that g k = reduces to the required formula
for
/(') =
A*,
i
=
0,
,
q.
Then by
5.8,
A
k
90
HOMOLOGY OF SIMPLICIAL COMPLEXES =
dj,,
[CHAP.
Ill
Since f, t d,
we have
B')
A')
=
dJ,(gB
'
=
f.^gB
=
B")
' ' '
DV
REMARK.
B"
B')
(DV
q~ l
4
l
'
'
'
A".
This completes the proof.
L, (K* L<) VJ L) induces isomorphisms, an isomorphic theory could be ob9~ l Q tained by defining C q (K,L) = L). This alternative Q (\K \K definition has the advantage of employing only skeletons of dimension
Since the excision
K
9 ~*
map (K*,K
\
U
C
U
H
y
U
and thus yielding C q (K,L) = C Q (K,L Q ). On the other hand the boundary operator 3 Q for chains would have to be defined as the composite of three homomorphisms, one of which is the inverse of an isomorphism induced by an excision. DEFINITION 6.1c. The coboundary operator.
q
$':
C Q (K,L)
> C*
9+l
is
the coboundary operator of the triple (\K
,
W
q
1
IK''
U L).
L,
\K
U
Q*l
L,
THEOREM THEOREM
6'/'.
6.2c.
6.3.c.
d
9+1
d
q
=
0.
>
///:
// c
e
(K,L)
(K
V
,L
}
)
is simplicial, then
Q
f
d
=
THEOREM
6.4c.
C a (/f,L),
fcO
then
7.
CYCLES AND
y
HOMOLOGY GROUPS
DEFINITION 7.1. Let (K L) be a pair consisting of a simplicial C Q (K,L) * and a subcomplex L. The kernel of d q C q .i(K,L) is called the group of qcycles of (/C,L) and is denoted by Z q (K,L). The image of d q+l C Q+l (K,L) ^ C Q (K,L) is called the group of qboundaries of (#,L) and is denoted by B q (K,L). Since d d<,4j = 0, B Q (K,L) is a subgroup of Z q (K,L)\ the factor group H Q (K,L) = Z g (K,L)/B g (K,L) is called the qdimensional homology group of (#,L). If /: carries Z Q (K,L) into (jFC,L) > (#,,L,) is simplicial, then / Z^K^Li) and B q (K,L) into B^K^LJ, thereby inducing homomorcomplex
K
:
:
fl
phisms.
REMARK.
The group q (K,L) given by the axioms.
H
is
not to be confused with the group
An
isomorphism of these two groups
7]
will
CYCLES AND HOMOLOGY GROUPS
be established in the next section, and
is
81
the main step in our
uniqueness proof.
According to
5.9,
the inclusion
*
maps
J
L  K induce exact sequences
(K,L)

C,(L) > C,(K)

C.(K,L) * 0.
We
shall explore this fact to define
homomorphisms 3:
H,(K,L)
we define _Z q (K,L)_ = j~\Z,(K,D}, B,(K,L) and H,(K,L) = Z,(K,L)/B.(K,L), then j:\B,(K,L)},
LEMMA
7.2.
Z.(K,L)
=
d.'[Y,C'.
is the
taining
the
operation of forming the smallest subgroup of C q (K) contwo groups. Further, the homomorphism j q induces iso
morphisms
J:
H.(K,L)
=
H.(K,L).
Then c s Z Q (K,L) if and only if jq c e c e C Q (K). and only if dJ Q c = 0. This is equivalent to j q ~id Q c = 0, and since kernel jgi = image z a _i, this is equivalent to d Q c e i q i[C ii(L)]. Suppose c e B q (K,L). Then j q c e B q (K,L) and j a c = d a+1 6 for some 6 e C q+l (K,L). Let rf e C q+l (K) be such that a+l d = b. Then
PROOF.
Let
Z
g
(K,L),
i.e. if
,;
J(c
so that there

a a ^id)
e e
=
jqc
z fl
d q+l j g+l d
iq e
Q
=
y gc

d c+1 6
=
c
0,
is an and therefore c t>,
C
(L)^
with
=
c
e JB a (/f)
O
d^id.
Thus
if
=
fl
d a+1 d
+
i q e,
then_ .;>
=
j q d q+l d
+
j
=
[C (L)]. Conversely, d q+l j q+l d and j a c
c
=
q
d nd
+
e
B
(K,L).
Thus
now a direct consequence of the Noether theorem. isomorphism LEMMA 7.3. The boundary homomorphism d q C q (K) > C q \(K) deThe
last part of 7.2 is
:
fines
homomorphisms
~Z q
(K L) *
}
i. l [Z.
B.(K,L) + i.ilBa
Szncc <Ae kernel of
i q ^. l
is zero, the
composition
i q id q
defines
homomor
phisms
Z
q
(K,L) *
Z C ^(L),
B.(K,L) > ^^(L),
92
HOMOLOGY OF SIMPLICIAL COMPLEXES
[CHAP.
Ill
and thereby induces a homomorphism
A:
H (K,L)
q
>
7.2,
H .,(L).
q
PROOF.
Let
c e
Z
q
(K,L).
t fl  2
By
did
we then have
d q ii q \d
d qc
=
iq
^d
0,
for
some d
and d
c e
fl
e
therefore
C c _i(L). d q ^d =
q
Then
=
=
d q ^d q c
=
e
and
since the kernel of i_2
is zero.
Thus d
e
Z q ^(L)
C<+i
c e ifilZtiOL)].
If c e
B
(K,L), then,
fl
by
7.2, c
=
d a+1 d
C^L). Then d
DEFINITION
c
=
d q d q+l d
+
d q iq e
=
ia
V? for some d ,iV, and d a c e
+
i^B
7.4.
The homomorphism
is
defined to be the composition # = Aj~ THEOREM 7.5. Consider the diagram
l
.
H (K,L)
q
<

Z
q
(K,L)

>
C
q
(K,L)
^~
C q (K)
and v,vi are natural homomorphisms. Let x e q (K,L), y ^H q ^(L). Then y = d#z zf and only if for any c e Z a (K,L) and any d e C q (K) such that x = vc, t/c = j a d, there is ane e Z q i(L) with
in ^AtcA
17,77!
are inclusions
H
d qd
=
tgit;^,
v^
=
y.
The
DEFINITION
q
l
proof follows by inspection of the definitions of j, A, and d l 7.1c. C 9 (K,L) > C*+ (K,L) is called The kernel of 5
.
fl
:
denoted by Z (K,L). The image of d denoted by B (/f,L), is called the group of qcoboundaries. The q factor group H*(K,L) = Z (K,L)/B\K,L) is called the qdimensional > (JfC^LO is simplicial, then cohomology group of (K,L). If /: (K,L) f a Q q (K lf L ) > / (K,L) is the homomorphism induced by /
the group of qcocydes, and ~
a
,
q
is
:
H
l
H
.
LEMMA
7.2c.
7n
#ie sequence
> C'(K,L)
),

C"(K)
=
C'(L) > 0,
l
B"(L)
(f) B<(L),
and
Z'(L)/B'(L).
T/ien
8]
MAIN
Q
ISOMORPHISM
93
and
i
induces isomorphisms
i:
H (L) =
q
ff'(L).
LEMMA
q+
l
7.3c.
Under
6<:
C(K)
>
C q+l (K), Z q (L)
q+l
is carried into
(L) is carried into j j \K,L). Since the kernel l q+l q+l q is zero, the composition (j d induces a homomorphism of j )~
Z
q+l
(K,L) and
B
q
B q+
A:
H (L)
q
>
H
q+l
(K,L).
DEFINITION 7 Ac.
The homomorphism
6*:
H
9
(L)+H*+ (K,L)
f
1
.
l
is
defined to be the composition 5 = At" THEOREM 7.5c. Consider the diagram
~H*CK J\
7 q CK
J\
C**(~K 1\
C^^CK^
i/j
r/!
i
., c
r
and
c e
is
ie
a:
e
H
q~
l
Z
q~ l
(L), y e
H
e
q
(K,L).
(L) and any d
Cq
l
~l
Then 5*y = x if and only if, for any 1 such that x = v,c, ijiC = i*" ^, <fcera (K)
q
an
e
$.
Z q (K,L)
with
d
q~
d
=
j
rje,
ve
=
t/.
8.
TEE MAIN ISOMORPHISM
Let (X,A) be a pair with a triangulation
section
T =
{t,(K,L)\.
In this
we
establish the basic result that
H
Q (X,A) and
H
Q
(K,L) are
isomorphic.
LEMMA
8.1.
In
3*
the
diagram
*
r
(JK,L)
q
< // a (K a
U L,L)
the
> // (K a VJ
fl
Ll,^"
1
U L)
K*~
l
(K U L,L) C (K U L,L) C (K,L) and W L) are inclusion maps, following relations hold:
c
j:
/:
(a)
the kernel of l^ is zero,
(b) the
(c)
image of
is onto,
1+ is
Z
q
(K,L),
l
j\
(d) the kernel of j # is l' [B q (K,L)]. As an immediate consequence of this
lemma we have
94
HOMOLOGY OF SIMPLICIAL COMPLEXES
THEOREM
8.2.
[CHAP.
Ill
(MAIN ISOMORPHISM).
In
the
diagram
H.(X,A)
^
HJi\K\,\L\)
17
^
H,(\K'VL\,\L\)
v the
4
H
C,(K,L)

Z,(K,L) * ff f
is
an inclusion and
are onto
natural map, the homomorphisms t^j^
q
and fif
X
and
their kernels in
zs wnjrfe
(\K*
U
L,L)
are equal.
Therefore 6 T
=
vif\j~*t~t
valued and is an isomorphism
Explicitly,
if
if
x
e
H (X,A) and y H (K,L), then
9
e
9
i/
=
rz
if
and only
there exist elements c
e
Z 9 (K,L) and
/
d
s
H<(\K* U L,L) with
vc
x
= y^d,
now with
#
d
=
T;C,
=
i/.
The
section.
fied
properties of this isomorphism will be established in the next
We
proceed
the proof of
8.1.
The
following simpli
notation will be used:
J=
K,
J'= rUL,
is
A = X~ =
l
L.
Of the theorems
proposition, which
of this chapter, the proof uses only the following
a restatement of
5.2:
}
LEMMA 8.3. Ifp^q, then H q (X',X*~ Two lemmas will be needed. LEMMA 8.4, The homomorphism
)
=
0.
V
V
induced by
the inclusion
map (X
and
,A)
(3)
>
(X *\A)
is (1)
onto
ifq^p
p,p
(2) of kernel zero if q ?* p,
an isomorphism
if q
^
+
1.
PROOF.
Consider the section
homology sequence of the triple (X'*\X ,A). By 8.3, the left term vanishes if q ^ p and the right term vanishes if q ^ p + 1. Thus (1) and (2) follow from the exactness of the H.S. of the triple (see
of the
1,10.2).
V
Statement
8.5.
(3) follows
Q
from
0.
(1)
and
(2).
LEMMA
PROOF.
H,(X "\A) =
8.4
2,
it
From
follows
,
by induction that
Since
H
l
9
(X'~ ,A)
Q
~
=
H (X^
Q
r
,A) for r
=
q
+
1.
H
q
(X~\A) =
H
(A,A)
by
i,8.1,
the proposition follows.
8.1.
PROOF OF
The homomorphism fa
H
q
(X,A)

H,(X,A)
may
be factored into homomorphisms
*
i+l
*
ff f (X
"
r
ff,(X,A)
,A)
H<(F*
,A)
=
H.(X,A)
8]
MAIN ISOMORPHISM
95
where q
8.4, ji^
+
is is
r
= dim
K
,
and where j l9 j 2
* *
,
are inclusion maps.
By
onto, while J2+,JA*9
that j+
(1)
onto, thus proving part
kernel j+
(c)
are isomorphisms. This implies of 8.1. It also follows that
.
=
kernel j u
The
proof of the remaining three parts of 8.1
is
based on the
fol
lowing diagram:
K
The commutativity conditions in the two triangles are obvious consequences of the definitions of the appropriate boundary operators. The middle row is a portion of the H.S. of the triple (X',X'~ ,A). = by 8.5, it follows that the kernel of l^ is zero, Since q (X*~ ,A) thus proving part (a) of 8.1. Furthermore image J # = kernel d l} and
l
H
l
since the kernel of
^
is
zero,
we have
l
kernel
^ =
kernel l+d
=
kernel d q
is
= Z
q
(K,L).
*
Thus image 1+ = Z q (K,L), and part (b) of 8.1 The vertical column is a portion of the H.S. therefore, by (1) and exactness,
proved.
of the triple
(X
q
l
,X*,A);
kernel^
=
kernel j
=
lj(t
imaged.
Thus
Z*[kerneiyj
=
Z*[imaged]
=
image Z*5
is
and
therefore, since the kernel of 1+
= imaged^! = B (KjL) = l~ [B (K L)]. zero, kernel j+
q
l
q
9
This proves
for q
(d) and concludes the proof of 8.1. THEOREM 8.6. // (X,A) is a triangulable pair, then
H
Q
(X,A)
=
<
0.
PROOF.
In view of the main isomorphism,
it suffices
to
show that
for q
H
q (K,L) This implies
=
< 0. By Z (K,L) = 0, and
for q
q
5.4,
we have C q (K,L) =
<
0.
thus also
#
(K,L)
=
0.
LEMMA
8.1c.
In
the
diagram
H'(\K\,\L\) > H'(\K*
96
HOMOLOGY OF SIMPLICIAL COMPLEXES
I
[CHAP.
Ill
where j and
(a)
are inclusions, the following relations hold:
Z* is onto,
(b) the kernel of I* is
(c)
B
q
(K,L)
q
the kernel of j* is zero,
(d) the
image of j*
8.2c.
is
l*Z (K,L).
THEOREM
t*
In
the
diagram
J*
H\X,A) 
y*
ff'(#,L)
>
H (\K
q
q
\J
L
t* is
an isomorphism, the kernel of j* is zero, the images of j* and l*ij = vrj~ l*~ l j*t* coincide, and the kernels of Z*i? and v coincide. Therefore T
l
is
a singlevalued isomorphism
BT
:
H(X,A) =
and y
e
H*(K,L).
Explicitly,
if
x
e
H (X,A)
Q
H
Q
(K,L), then y
=
T x if
and
only
if
there
is
a
c e
Z(K,L) such that
j*t*x
=
then
l*i)C,
vc
P
=
y.
LEMMA LEMMA
8.3c.
8.4c.
// p
*
q,
H
9
The homomorphism
is (1)
Q
by the inclusion map
onto if q 7* p,
(X ,A) induced of kernel zero ifq7*p+l,(2)a homomorphism
if q 7*
(X ,X ) =0. H q (Xv+l ,A) >
p,p
V~ 1
H
q
v
and
(3)
an isomorphism
l
+
q
1.
LEMMA 8.5c. THEOREM 8.6.
9.
H
(X'~ ,A)
=
0.
// (X,A)
is triangulable, then
H
(X,A)
= Ofor q <
0.
PROPERTIES OF THE MAIN ISOMORPHISM
T of
The
the last section will
following three important properties of the isomorphism now be established.
9.1.
THEOREM
In
the
diagram
H<(X,A)
3
I

>
H
q
(K,L)
!'
where T'
=
{t',L} is the triangulation of
A
defined by T, the commutativity
relation d*6 T
=
Q T ,d holds.
THEOREM
9.2.
// (X,A) (Xi,Ai) are pairs with triangulations
9
T =
9]
PROPERTIES OF MAIN ISOMORPHISM
97
>
\t,(K,L)}, T,
=
{(,,(^1,1/1)} respectively,
and
if f:
(X,A)
(X lt Ai)
is simplicial, then,
in the diagram
eT
H,(X,A)
/.
Or,
>
H,(K,L)
Iff
0:
(/?,L)
(K^L
}
)
is the simplicial
mutativity relation g#0 T
=
map
g
=
rfrVX
0rJ*
Ao/ds.
be the space consisting of the single reference point. == C Q (K Q ) 0l we have Z (K ) garding PQ as a simplicial complex
Let
P
Re
K
= G
and So(Po)
= so that H Q (K) = G. THEOREM 9.3. 0: # (Po) ~> #o(#o) is the identity map G > G. The last proposition is an immediate consequence of the definition
Theorem
9.2 follows
of
0.
each square of the following
from standard commutativity relations in diagram
H.(X,A)

ff.(K,L)
< H,(KUL''i)
I'**

Z,(K,L)
I/*
I?*
i?
where h
is
defined
by
^.
The
proof of 9.1 requires some preparation
9.4.
LEMMA
The homomorphism
Vinduced by inclusion,
H
l
t
(\K',\I<'~
\)>IWVL\,\L\)
onto.
fc
is
a homomorphism
#
PROOF.
phisms
The homomorphism
k
may
be factored into homomor
k*
f
H.(\KW
homomorphism
&2 *
l
\)
^H.(K
,L)
H
Q
(\K*
Q
U L,L)
K
9
induced by appropriate inclusion maps.
is
Since
L =
C\ L, the
2.1.
an isomorphism by the excision theorem
To examine u
fc
consider the portion
*,,
H (\K
t
t
\
t~l
t
\L

\)
H,(K',L) * ff^.dL'l.lL'1
1
!)
of the H.S. of the triple
It follows
(K^L^U'
fci,
).
By
8.3,
the end term
is
is
zero.
from exactness that
is
onto.
Thus k^
onto.
8
HOMOLOGY OF SIMPLICIAL COMPLEXES
PROOF OF
9.1.
[CHAP.
Ill
Consider the following diagram
H.(K,L) 1* Z.(K,L)
y*
I
z*
i*i
t*
H.(X,A)
i d

o*
T**
I di
Q~
/*
/i*
1
I
H.(\K\,\L\)
l<>'
t*
ji+
 dg
/nr
\
.
)
>
c
/ tr\
fl
i(/c;
T
6
t'
i
i*
p ^iV
/
T
T7
'l^l
H
t
f
i(L)
In this diagram 77,?;! are inclusions, v,v are natural homomorphisms, and all other homomorphisms are either boundary operators, or are
induced by t and t', or by appropriate inclusion maps. The commutativity relations in each square are easy to verify. Let x e Q (X A). By definition of 6 T there exist elements
H
y
c e
Z
q
(K,L),
b
e
H
q
(\K*
U L,L)
with
By = k^a
9.4,
6.
k^ is onto, and therefore there Define d = f^a. Then
is
an a
e
J7 Q (/
(
J
~1
)
',Z/
with
Thus, by
7.5, there is
an
e e
Z ^(L)
q
with
Since
t+ji*dia
^i'i*^i a
= 3^^a = d^fc = ^<//* a = ^^ = t
follows that
Zi^^ia
and
since the kernel of
i a _i is zero, it
dz
=
^Ji^iOi
=
he.
Thus, by the definition of
T ., it follows that
T >dx
=
v{e
= d8 T x.
9]
PROPERTIES OF MAIN ISOMORPHISM
9.1c.
99
THEOREM
In
the
diagram
6T
H\X,A)
[$
the commutativity relation B T d
=
d*6 T
'
holds.
tri
THEOREM 9.2c. ///: (X,A) > (JSfi^AO is simplicial relative to angulalions T = {^(K>^)}> ^ = (^XK^LO}, Men in Me diagram
BT
H'(X,A)
>
H
a
(K,L)
r
!
'
/)
U Ti
commutativity relation g*0 Tl
T f* holds,
>
where g
=
l
t^ ft.
THEOREM
/C
is
9.3c.
6:
H
(Po)
H
(K
)
is the identity
map G
G,
Ae simplicial complex consisting of the single vertex
9.4c.
fc*:
P
.
LEMMA
The homomorphism
H\\K
\J
kernel zero.
induced by
the inclusion
map, has
proof of 9.1c is not quite dual to that of 9.1, and will therefore be given here. Consider the diagram
The
H*(K,L) T"
Z\K,L)
I
t*
/i*
H'(X,A)
t
a
'*
C\K)
T'
jf
T*.
Tf.
100
HOMOLOGY OF SIMPLICIAL COMPLEXES
e
[CHAP. Ill
l
Let x
H
q
Q~ 1
(A).
By
the definition of
T
,
there exists
ace Z~
(L) with
j*t'*x
=
l*rjiC,
v&
1
=
T >x.
Since
i
is
onto,
we may
e e
b
q
select
d
e
C*"
^)
=
so that i*d
=
?hC.
Then, by
7.5c, there is
an
Z Q (K,L)
d
satisfying
q
=
j
rje,
ve
b*6 T 'X.
Then
k*j*t*8x
=
Stftt'+x
=
M
zero,
it
Since,
by
9.4c, the kernel of
fc* is
follows that
J
t'*/*fir 6 OJU
/*np I 7/t/.
T dx
Thus the
definition of B T implies
=
ve
=
b*Q T >x*
10.
THE UNIQUENESS THEOREM
two homology theories defined on admissible cateThe main triangulable pairs and their maps. gories containing result of this chapter is that every isomorphism of the coefficient groups G and G yields an isomorphism of the theories H and H over the class of triangulable spaces and their maps. THEOREM 10. 1 (UNIQUENESS THEOREM) Given two homology theories, H and H, and given a homomorphism
Let
all
.
H and H be
h
:
G  G
homomorphisms
of their coefficient groups, there exists a unique set of
Tl,(X,A) h(q,X,A): H,(X,A) each triangulable pair (X,A) and each integer defined for
(1) (2) if f:

q,
such that
A(0,P
)
=
h<>
(X,A)
>
(X^AJ
is
a map, and_(X,A) and (X lt Ai) are
tri
angulable, then the commutativity relation f+h(q,X,A) holds in the diagram
=
h(q,Xi,Ai)f+
h
H
__
>
q
(X,A)
H.(X,A)
10]
(3) the
UNIQUENESS THEOREM
commutativity relation dh(q,X,A)
101
l,^4)d holds in the
=
h(q
diagram
H.(X,A)
\d
>
H,(X,A)
// A
:
G ~ G
is
an isomorphism, then each h(q,X,A)
construct
is also
an
isomorphism.
PROOF.
We shall
homomorphisms
t(q,K,L):
for each simplicial pair (K,L) conditions:
(4)
(5)
H,(K,L)
 H,(K,L)
and each integer q satisfying the following
f(0,P
if_/:
)
=
Ac,
(A',L)
*
(/Ci,L) is simplicial,
then the commutativity rela
tion
f{(q,K,L)
=
f (<7,A", ,/>i)/
holds in the diagram
/.
f
(6) the commutativity diagram
relation
df(q,K,L)
=
f (9
l,L)d holds in the
(v j\ n.q\rL)Lj)
u
^ >
H \\.,Lj) n tv ^^
Q
L.
Let
e
,'1
Define
G and
let
gA
a
yl
be a gchain of K.
It follows
from
5.45.7 that
TJ
yields a
homomorphism
102
HOMOLOGY OF SIMPLICIAL COMPLEXES
5.8
[CHAP. Ill
.
From
and
6.4
r;
it
follows that
77
commutes properly with / and d q
Consequently
defines
homomorphisms
Z,(K,L),
Z,(K,L)

B.(K,L)

B,(K,L),
and thus induces homomorphisms
of their quotient groups
satisfying (4)(6).
Now, given a
sider the
pair (X,A) with a triangulation
T =
{f,(K",L)}, con
diagram
H.(X,A) >
where
6T
H (K,L)
a

ff,(ff,L)
< tf,(X,A)
Define
and 0y are defined
for
H and // in 8.2.
=
flj'fflr,
h T (q,X,A)
then
We
will
show that h T
is
independent of the triangulation
:
T and
:
satisfies
(l)(3).
Observe that, if (X,A) = P then T G ~ G and T G = G = f (0,P ) = ^o as desired. are identity maps, so that /i r (0,P ) Now let CX^XCX^AO be pairs with triangulations T = [t,(K,L)} T = {^,(^i,^i)}i and let /: (X,A)  (Xi.AO be a map (not neces,
9
l
sarily
simplicial).
>
In
addition,
we
consider
11,8.6
the
identity
map
i:
(X,A)
integer
(XjA).
of
Theorems u,8.4 and
imply the existence of an
n and
maps
/.:
(X,A)^(X A
lt
l
)
9
in :
(X,A) + (X,A)
relative
angulations T
/,
with the following properties:
/n
l
is
simplicial
to
the
tri
=
{\( K?L)} T =
n
y
{/!,(/iLi,Li)}
and
while
i n is simplicial relative
J,
T =
(t,(K,L)
and
is
= { to the triangulations to i. Define the simplicial homotopic
n
T
is
homotopic to n n n t,( K, L)}
9
maps
g:
n CX, L) > (K^LO,
n
*:
(
X, L)
by
10]
UNIQUENESS THEOREM
the diagram
6 f
103
We now have
H,(X,A)
H
9
_
.
~6
t
(K,L)
H
t
(K,L)
t*.
<
_ H (X,A)
a
11
'
f
/*
g
10*
10*
7*
,
a,
.,
k and ^
B follow
it
The commutativity relations involving f follow from (5) since are simplicial. The commutativity relations involving and
from 9.2 since
follows that
it
and / n are simplicial. Since i n is homotopic to i, and t are identity maps. Since f n is homotopic to i^ follows that / n# = /^ and / nj^ = f^. Thus we obtain
in
njt
/,
(7)
J*h T (q,X,A) Applying
= ^rXg^i^i)/*.
(X,A)
this relation to the special case
=
/
is
=
(Xj,/!,),
and
identity,
we
find h T (q,X,A)
=
h Tl (q,X,A) so that indeed h T (q,X,A}
7
,
independent of Consequently (7) To prove (3) and denote by T'
the triangulation T
yields (2).
and may be written as h(q,X,A).
consider (X,A) with a triangulation T = \t,(K,L)} the induced triangulation of A. In the diagram
H
a,
!
Q
(X,A)
a.
f
commutativity in the center square
is
asserted
tativity in the outside squares follows from 9.1. h(q l,A)d as desired.
by (6) while commuThus dh(q>X>A) =
(l)(3)
We now
are unique.
If s
(8)
is
prove that the homomorphisms h(q,X,A) satisfying Two auxiliary propositions will be established first. an ordered gsimplex and g e (7, then
Afe,a,*')(0)
=
ho(g)s<.
If
g
=
0,
Assume then that q
the proposition is an immediate consequence of (1). 1 in place of q. > and that (8) holds with q
104
HOMOLOGY OF SIMPLICIAL COMPLEXES
<~A 9 A <
1

[CHAP. Ill
s*
Let A < with vertices
be the vertices of

s
q
and
3.7,
l
let s*~
be the face of
< A\
Then, by
Since
q
q~
l ]
[s :s
= #^3,
it
follows from (2)
[
and
f fl
(3)
that
h(q

1,
lOJi''DWl =
]%, mi*'!)
This implies
Since
1
[s'ls""
]
is
Let gA
(9)
an isomorphism, the proposition A' s C,(K,L). Then
1
follows.
h(q,\K< VJ L, IX*
VJ
L)(^
.
A) =
*ofo)A
A.
By 5.5, we have gA and / is a suitable map (2) and (8),
h(gA
as required.
A'
/:
=
f+(g) where s
(s,s)
is an ordered gsimplex, > (K* VJ L, K"" VJ L). Then, by
1
A')
=
fc(/*(<7))
=
7*ft(fl)
=
7*W
=
^o(sr)A
A'
Now
(3).
fc,
let h and fc' be two families of homomorphisms satisfying Let (K,L) be a simplicial pair. We introduce abbreviations:
(1)
=
h(q,\K< \J L\,\K' VJ L),
h(
1
ft 2
=
h(q,\K' VJ L,L),
and similarly define
1:
and
/iz.
1
(K< VJ L,L)
(2)

By
(9)
we have
ht
=
h(.
Let
(K* VJ L, K"'
\J L) be the inclusion map.
Then
by
Since,
by
8.1,
the kernel of
l^
is zero, it
follows that h 2
*
=
hj.
Now
consider the inclusion
map
j:
(K" VJ L,L)
(K,L).
Then
Since,
by
8.1,
^
is
onto,
it
follows that h(q,\K\,\L\)
=
ti (q,\K\,\L\)
.
Let (X,^4) be a triangulable pair and of (X,A). Then
T =
{/,(/,)} a triangulation
Since
^
is
Thus h
=
h'
an isomorphism, it follows that h(q,X,A) and uniqueness has been proved.
=
A'CgjX,^!).
EXERCISES
105
be an_ isomorphism, and let hQ G ~ G be > Let h(q,X,A): g (_X,A) q (X,A) be the homomorphisms satisfying (l)(3) relative to hQ Then
Now
let
hQ
:
G ~ G
:
the inverse isomorphism.
H
H
.
h(q,X,A)h(q,X,A):
are
G.
H (X,A) Q
H,(X,A)
homomorphisms satisfying (l)(3) relative to the identity map G Therefore, by the uniqueness property, we have h(q,X,A)h(q X,A) identity. Similarly h(q,X,A)h(q,X,A) = identity. Thus h(q,X,A)
9
=
is
an isomorphism with ft(g,X,A) as
the theorem.
inverse.
This concludes the proof of
THEOREM
homomorphism
10. Ic.
Given two cohomology theories
H
and
_ H
and a
hQ
:
G  G
homomorphisms
of their coefficient groups, there exists a unique set of
h(q,X,A):
H*(X,A)
9

~H*(X,A)
each integer
g,
defined for each triangulable pair
(1)
(X A) and
)
such that
_
fc(0,P
=
h
,
(2) (3)
///:
If h Q
:
(X,A)

8h(q, A)
=
h(q_+ l,X,A)i.
(X,^,), then
y
j^q^.A,) =
is
h(q,X,A)f\
G ~
G, then each h(q
X,A)
an isomorphism.
EXERCISES
A. INCIDENCE ISOMORPHISMS.
1.
Let sj'^sj"
1
be
(q
l)faces of a gsimplex s
a
and
let s*~*
be
their
common
(q
r.ai
[Si
2)face.
a2ir
.5
J
Verify that
li
.S\
J
a
_
__
r
al
[S
[2
.^~ 2 ]\^.^ 1 1 S [S .02
J
J
Hint:
use 4.5.
B. CELLS
1.
AND SPHERES.
1

Using the notations of i,16, consider the map A,: (fT,^" ) > 1 xn). x n ) = (x i9 x x,, , (J^ST' ) defined by h.(x l9 = z for 2 e n (ET,S" ). Prove that h^z 1 l > S"' S"~ defined by A, in problem 1. 2. Consider the map /,: = z for z e H^S"' ). Prove that /.^z n~ n 3. Consider the map A: (E S ) > (J^,^" ) defined in vector l notation by fc(x) = x. Prove that A^z = fl^z for z s n (E*,S*~ ).
, ,
,
f ,
H
1
1
l
1
y
H
106
4.
EXERCISES
Consider the
Prove that f+z
C.
=
map
n
/:
Sn
~l
(l)
z for z e
Sn " ff^GS
>
11
~l
1
defined by h in problem
).
3.
THE EULER
and
CHARACTERISTIC.
of the homology theory are Dmodules (see and assume that the coefficient group has rank 1. 1. Let (XjA} be a pair with a triangulation T = Let \t,(K,L)}. a q be the number of ^simplexes of K which are not in L. Show that the Euler characteristic \(X,A) exists and satisfies
Assume that the values
i,I),
exercises i,H
x(X,A)
where n
=
E (!)'.
<j0
= dim
K.
(Hint:
prove r[C Q (K,L)]
= a
a)
D. nCiRCuiTs.
(1)
DEFINITION. each point of
A
K
L
is
simplicial pair (K ,L) is called a simple ncircuit if a point of some nsimplex of 7f (2) each (n 1),
simplex of
<r
a face of two nsimplexes of L, and (3) and T are two nsimplexes of there exists an ordered set L,
is
K
K
if
K
(7
=
<7i,
CT 2 ,
'
'
' ,
(7*
=
T
of alternately
n and (n
l)simplexes of
K
L such
that each (n
1)
simplex
a face of the neighboring nsimplexes. 1. Assume that the coefficient group G of the cohomology theory is A n be infinite cyclic (i.e. isomorphic to the integers). Let A,
is
,
the vertices of an nsimplex of If g e G, show that gA A"

K
is
L where
(K,L)
of
is
a simple ncircuit.
an ncocycle
after 5.8c), and show that any ncocycle of n to one of this form. Show that (K,L)
K
is
K mod L (see Remark mod L is cohomologous
either infinite cyclic or an orientable ncircuit,
H
cyclic of order 2. In the first case (K,L) in the second case, nonorientable.
2.
is
called
Let (K,L) be a simple ncircuit and let the coefficient group be If (K,L) is orientable, show that n (K,L) is infinite If (K,L) is cyclic, and H^^K^L) has no elements of finite order. = 0, and the subgroup of // n _!(/C,L) n (K,L) nonorientable, show that of elements of finite order is cyclic of order 2. n 3. Define real projective nspace P to be the space obtained from an nsphere by identifying each point with its antipode. Show that Pn is triangulable as an ncircuit which is orientable or not according
infinite cyclic.
H
H
as
n
is
odd or even.
that the cartesian product of any finite number of spheres can be triangulated so as to be an orientable simple ncircuit.
4.
Show
EXERCISES
5.
107
of sim
Let
K
be a
2circuit.
1,
Show
that the numbers a
,
i,
2
plexes of dimensions 0,
and 2
satisfy the conditions
a,
o
=
3(ao
i (7
2
,

+ V49
___ 2
,
24 X (/0).
that 2sphere S projective plane P and the torus T* can be triangulated as simple 2circuits. Their Euler characteristics are
6.
Show
respectively
2,
1,
and
2 2
0.
Show
that, for
any such triangulation, the
following inequalities hold:
S
P
o^4, o^G,
o
a,
<*!
^
^ ^
6,
15,
r2
Find triangulations
in
^7,
aj
21,
a2 a2 a2
^ ^ ^
4.
10. 14.
which the minimal values are attained.
CHAPTER
Categories
1.
IV
and functors
INTRODUCTION
is
The
first
objective of this chapter
to introduce
and
illustrate the
concepts of category, functor, and related notions. These are needed in subsequent chapters to facilitate the statements of uniqueness and
existence theorems.
in the sequel.
Only as much of the subject is included as is used thorough treatment can be found in a paper of Eilenberg and MacLane [Trans. Amer. Math. Soc. 58 (1945), 231294]. The ideas of category and functor inspired in part the axiomatic treatment of homology theory given in this book. In addition, the point of view that these ideas engender has controlled its development at
A
every stage. The second part of the chapter
fined
is
devoted to homology theories de
on abstract
theory, as the statement of the axioms for a
An admissible category for homology Acategories. defined in i,l, has much more structure than is needed for
homology theory. Abstracting the an Acategory. There is a /ifunctor. It is shown that the composition
essential elements leads to the notion of
corresponding concept of of an ^functor and a homology theory, when defined, is again a homology theory. We thereby obtain a rule for deriving one homology theory from another. It is used frequently in subsequent chapters.
2.
CATEGORIES
The definitions below arise from the consideration of the common properties of collections such as (1) topological spaces and their continuous mappings, (2) groups and their homomorphisms, and (3) simplicial
complexes and their simplicial maps. An examination of the properties of continuous maps, homomorphisms, and simplicial maps
leads to the following definition: DEFINITION 2.1. set 6 of elements {7}
A
is
system
a unit) if 71 = 71 and 7 2 c = 72 whenever *7i and 7 2 c are defined. The multiplicative system is called an abstract category if the following axioms are satisfied:
element
e e
if,
for
some
is
6
pairs 71,72 called an identity (or
e
C, a product 7 27i
called a multiplicative An e C is defined.
108
2]
CATEGORIES
(1)
109
if
The
triple
product 73(7271)
is
is
defined
and only
if
(7372)71 is
defined.
When
either
defined the associative law
73 (727l)
=
(7372)7l
holds.
(2)
This
triple
product
will
The
triple product 7s727i
be written as 7 37 27i. is defined whenever both products
ly
787 2 and 7 27i
(3)
2
are defined.
e
For each 7
2.2.
C
there exist identities
2
e
C such
2
that 7*1 and
7 are defined.
LEMMA
2
For each 7
7*1
e
e, the identities
yi{ are
l
and
such that 761 and
7
are defined, are unique.
PROOF.
ties.
Suppose
(7*1)
(
and
is
both defined with
e^t'i
both identi1
Then
=
7c(
defined.
Hence
i(
is
defined and
=*
,(
=
c(.
Similarly
e 2 is
unique.
DEFINITION 2.3. A category 6 consists of a collection {C} of elements called objects and a collection {7} of elements called mappings. The form an abstract category in the sense of 2.1. The objects mappings
stract category.
are in a 11 correspondence C > Thus to each
ic
with the set of identities of the ab
objects are called the domain and the range of 7 respectively; notations
d
and
C2 such that yi Cl and
C,
mapping 7 there correspond unique i c .y are defined. The objects Ci,C*
C =
v
D(y),
C2 =
R(y), 7:
2.4.
 C
2
.
and only if R(y ) = D(y n ). = D(y ). 2 R(y*), D(y 2y ) J/727i is defined, then In other words the lemma states that, if 7^ Ci > C 2 7 2 C 2 > C8 then 7 27! is defined if and only if C 2 = C 2 and then 7 2 7i: Ci > C8 If T27i is defined, then 7a(7i) = 727i PROOF. Let e = i/z (Tl) Thus 7 2 c is defined and R (yj = D(7 2 ). Conversely, if is defined. = D(y 2 ), then 7 2 and 7i are both defined for e = ^( 7l and R(y^ therefore (7 2 c)7i = 7 2 7i is defined. If 7 2 7i is defined and e = iD (Tl) then 7^ is defined. Hence 7 2 7i* = (7 27i)c is defined so that D(yi) =
LEMMA
The product y 2V\
is defined if
l
R(y yJ =
l
l
:
,
,
.

)
>
Z)(7 2 7i).
Similarly
2.5.
R(y 2yJ
=
^(72).
7:
DEFINITION
if
there exists a
LEMMA
7,
2.6.
an equivalence with (y~ )~ = 7. // > C2 C2 > C3 are equivalences, then y 2y\ is an equivalence yii Ci 1 1 1 Each identity is an equivalence with c" = with (7 27i)~ = 7rVJ = 7"7 = Cl PROOF. Suppose that 7' and 7" satisfy y'y
l
C 2 in C is called an equivalence map 7': C2 > Ci in 6 such that y'y = t Ct 77' = ^c.The map y' of 2.5 is unique and is called the inverse of
Ci
>
,
A map
l
l
l
notation:
,
y~ 72
.
y~
is
also
:

.
,
77'
=
77"
=
ecv
Then
110
CATEGORIES AND FUNCTORS
fact that 7"
1 1 1 1
[CHAP. IV
an equivalence and that (7"" )" = 7 follows directly from the definition. If 7 2 ,7i are equivalences and 7 27i is defined, then 1 = Cl (727i)(7i~ 72 ) is defined and equal to e c> similarly (7i" 72 )(727i) is an identity and e' = Thus (7 27i)" x = 7r 72 If D(0 then ee' Thus e = ee so that e = iD(t) = i R( <) and is defined and e' = ee' = e.
The
is
1
1
,
.
1
1

,
1
e"
=
e.
DEFINITION
that
2.7.
A
c
subcategory
C
of
6
is
a subaggregate of G such
1. 2. 3.
if if
if
C
7
,
e
Co,
72,7, e
s
then i e C C and 7 2 7i
of
,
is
C
,
then D(y) and
defined, then 72(7) are in C
if,
7^
.
e
C
,
and
7
s
A
D(y)
It
subcategory C
s
C
is
s
called full
for each
C, conditions
map
in
Co imply 7 #(7) easy to see that a subcategory 7 s C may be an equivalence in
s
.
C
C
is
is itself a category. However a C without being an equivalence
C The
.
process of obtaining a subcategory
may
be broken up in two
consider the
full
steps.
First select a subset of the objects of
e and
subcategory determined by these objects. Then without further limiting the objects select a subaggregate of maps satisfying 1 and 2.
3.
EXAMPLES OF CATEGORIES
first example of a category is composed of topological spaces continuous maps. The objects are topological spaces, the maps are and continuous maps of one topological space into another. The composi
The
tion (product) of
two maps
is
the usual function of a function.
The
equivalences in this category are the homeomorphic onto another.
maps
of
one space
prime importance in our axiomatic treatment has, is a topological space and A where X\ and has, as its mappings, continuous maps /: (X,A) > (Y,B). This category will be denoted by 0^. The admissible categories discussed in i,l are all subcategories of a In an admissible category ft the
of
The category
as
its objects,
pairs (X,A)
X
t
.
C
equivalences are precisely the homeomorphisms as defined in 1,5. In i,2 we encountered the categories g c and g; the objects in these
categories are, respectively,
a ring R.
The mappings
in the first
morphisms
is
compact abelian groups, and modules over homoWhen R case, and linear maps in the second.
in these categories are the continuous
is
the ring of integers, g 7<
just the category
g
of ordinary abelian
groups and their homomorphisms.
For the categories g c ,g/j we
may
consider the categories Sig<7,Sjg*
4]
FUNCTORS
.
111
whose objects are lower sequences of groups in g c or Q R The mappings are the homomorphisms of one such lower sequence into another. The
exact lower sequences form subcategories 8,g c ,iSThe similarly defined categories of upper sequences and exact upper sequences will be denoted by Su 9c, u g c etc.
,
categories studied in Chapter n are denoted by 3Ci,3C,; the in both these categories are pairs (/C,L) where is a simplicial objects and L is a subcomplex. The mappings in JCj are linear mapcomplex
The
K
pings, while in 3C, they are simplicial. Clearly 3C. is a subcategory of Theorem n,4.8 states that every equivalence in 3d is in 3C.. 3Cj.
In ni,l the important category 3 of triangulable pairs
is
introduced.
The objects are triangulable pairs (X,A), the mappings are continuous maps of one such pair into another. Clearly 3 is a full subcategory of CLi
and
is
an admissible category
for
homology theory.
4.
FUNCTORS
let
Let
(3
and
3D
be categories and
T
be a function which maps the
/
objects of C e G a map
to
into the objects of 3D and, in addition, assigns to each map T(f) e 3D. The map T is called a covariant functor (from 6
3D) if it satisfies
1.
If /:
Ci
 C
the following conditions:
2,
then T(f):
T(d) > T(C t ).
l
2. T(i c ) 3. If / 2 /
=
t
ircc>.
is
The map T
placed by:
1'.
is
defined, then T(f 2 f ) T(f 2 )T(f,). called a contravariant functor if these conditions are re
=
If /:
d
=
is
>
C2
,
then T(f):
T(C 2 ) >
l
r(C,).
2'. 3'.
T(i c )
If
i r( c).
The
RT(f).
= T(f )T(f 2 ). defined, then r(/ 2 /0 7 condition 1 can be rewritten: T(Df) = Z)7 (/)
/2A
Thus T
is
a covariant functor
if it
and T(Rf) = commutes with the opera
a functor T is completely determined by the function T(f) defined for maps / only. Thus a covariant functor T is essentially a homomorphism of the abstract category associated with 6 into the abstract category of 3D subject to the condition that identities
tions of the category. In view of condition 2
be mapped into
identities.
A
contravariant functor yields an anti
homomorphism
of the abstract categories.
If T is a functor from 6 to 3D, and T' is a functor from 3D to 8, they can be composed in the obvious way to form a functor T'T from 6 to 8. If TT have the same (opposite) variance, then T'T is covariant (contra1
variant)
.
112
CATEGORIES AND FUNCTORS
6.
[CHAP. IV
EXAMPLES OF FUNCTORS
Let
given.
Ct
be an admissible category on which a homology theory is Let q be a fixed integer, and define for an admissible map
Then Axioms
functions
1
and 2
for a
is
#
a
(X, A ),#(/)
homology theory assert that the pair of a covariant functor q on the category Ct
H
with values in the category g* or g c Instead of using the category g* we may use the category 8^* of exact lower sequences in g. We then define H(X,A) to be the homology sequence of (X,A}, and H(f) to be the homomorphism /^ of the H.S. of (X,A) into that of (Y,B) induced by / (see 1,4.1). Then is a covariant functor on Ct to Axioms 1, 2, 3, and 4 insure that This functor is referred to briefly as the homology functor. or 8ig c In a similar way, the cohomology functor is a contra variant functor on a to 8ug or Sg c Another covariant functor on Ct, this time with values also in Ct, is
.
H
8^
.
.
obtained by setting
T(X,A)
where f\A
is
=
A,
T(f)
=
f\A
/:
the
map A
B
defined by a
to
Ot c
map
(X,A)
is
>
(Y B).
y
A
covariant functor on
3C,
(compact pairs)
obtained by
setting
T(K,L)
6.
=
(K,L),
T(f)
=
/.
TRANSFORMATIONS OF FUNCTORS
Let T and S be two covariant functors from C to 3D. A function F which assigns to each object C e C a map F(C) e 3D such that 1. T(C): T(C)+S(C), 2. if /: C, > C2 then T(C,)T(f) = S(/)r(d),
,
is
called a natural transformation of the functor
T
into the functor S.
In case
replaced
2'.
T and S
by
/:
are contravariant functors, the condition 2
on F
is
then r(d)r(/) = S(/)r(C,). an equivalence for each C e 6, then F map F(C) natural equivalence of the functors T and /S (notation: F:
if
C, >
C2
,
If the
is
is
called a
T
+
S).
In this case, 2 can be written
7]
EXAMPLES OF TRANSFORMATIONS
condition
1
is
113
The
tions in 2 are always defined.
equivalent to the condition that the composiCondition 2 asserts that commutativity
:
holds in the following diagram
S(C 2 )
It is easily verified that the
tions
is
also natural.
If
T:
T
composition of two natural transformaS is a natural equivalence, then
1
r~
l
:
S > T defined by
IT'
T.
1
and r = T T ^
:
=
r"*(C) = [IXC)]' is also a natural equivalence, T^T has the property r (C) = i r <o. Thus
It follows that the
concept of natural equivalence
is
reflexive,
symmetric, and
transitive.
7.
EXAMPLES OF TRANSFORMATIONS OF FUNCTORS
is
be an admissible category on which a homology functor given, and let T be the functor from Ct to & defined by
Let
Gi
H
T(X,A)
=
A,
T(f)
=
l
f\A.
be the homology functors Let q be a positive integer and let QJ Q^ from Q to the category $ R (or g c ). Then Axiom 3: df+ = (f\A)^d is
precisely the condition 2
d:
H H
that the
homomorphism
l
H
t
(X,A)+H,.i(A) = H,. (T(X,A)}
shall
be a natural transformation of the functor
H
Q
into the composite
functor
H^T.
In Chapter in it was proved that d is an equivalence d: Q over the subcategory of pairs (s,s) where s is a simplex and
H
^H
Q. 1
T
s is its
boundary. The group of ^chains C Q (K) is a covariant functor on JC, to 9 (or = f g for each simplicial map /: (K,L) > 9c), provided we define C,(/) In the same fashion the groups Z Q (K,L) of the ^cycles mod (AT,, LO. L and B q (K,L) of the bounding ^cycles mod L are functors on 3C f to Z g (K,L), the former is called a subS (or Sc). Since B Q (K,L)
C
functor of the latter. The quotient functor The has the same domain and range.
H
Q
(K,L)
= Z
g
(K,L)/B q (K,L)
map
v q (K,L):
Z
Q
H
9
(K,L) which assigns to each cycle
its
homology
class is
> (K,L) a natural
transformation of
Z
Q
into
H
a.
1U
defined on
Ot
CATEGORIES AND FUNCTORS
In order to distinguish this ^functor from the q
to
ib
[CHAP. IV
g* we
shall write fi Q for the q
ih
homology functor homology functor on 3C.
may regard (K,L) as a triwith the triangulation T = \t,(K,L)\ where t is the angulated pair = 6T > Q (\K\,\L\) identity map of (#,L). The isomorphisms 6 defined in 111,8, yield a natural equivalence of two functors q (K,L),
:
For each simplicial pair (K,L) we
H
H
0:
H
Q
T^8
is
q
where
of
0.
T
is
the covariant functor on
3C.
to
Ct
defined
by T(K,L)
=
(/,L), T(f)
=
/.
Theorem
n,8.4
the statement of the naturality
8.
cCATEGORIES AND 3FUNCTORS
In order to achieve uniformity and avoid repetition in the chapters ahead, it is useful to introduce concepts similar to the homology and
cohomology theories of Chapter i but defined on more general cateThe analogy with gories than admissible categories for homology. and cohomology theories is complete as far as the first four homology axioms are concerned. DEFINITION 8.1. A category with couples (briefly: a ccategory) is a category 6 in which certain pairs (a,/3) of maps, called couples, are
distinguished, subject to the sole condition that the composition > defined. If a: A 5, 0: B * C is such a couple, we write
(a,0):
$a
is
A>B>C.
<B
>
A
covariant [contra variant] functor T:
is
3D of
a ccategory, called a cfunctor if for each couple of maps (Ta,TfS) [(Tp,Ta)] is a couple in
>.
a ccategory into (a,/3) in C, the pair
As an example
the sense of
i:
i,l.
of a ccategory consider an admissible category (2 in For each pair (X,A) in a consider the inclusion maps
C.
A C
X,
ji
X
(X,A) as forming a couple
(1,3):
A+XOt
The same
admissible category
gives rise to another ccategory in
which the couples
(f,j):
(A,B)

(X,B) > (X,A)
of
are the appropriate inclusion
maps
Another example of a ccategory (or g c ) of groups by defining couples
is
an admissible triple (X,A,B). obtained from the category
g/j
8]
cCATEGORIES
<t>:
AND dFUNCTORS
G2
115
onto,
whenever
G
l
>
G2
has kernel zero, $:
G3
is
and kernel
^
=
image
<,
i.e.
whenever the sequence
*
>
is
<?!

*
(? 2
>
G3
>
exact.
DEFINITION
J5i
>
8.2.
Let
(a, ft):
A
>
B
>
C and
(a,,/?,):
AI
>
>
d
be couples in a ccategory C.
We define a map
B,,
(a, ft)
(i,ft)
to be a triple of
7i:
maps
AtAt,
72:
B
73
:
C
> C,
in
G such
that commutativity holds in the two squares of the diagram
72
It is
(a,/3)
in
C form
easy to verify that, with the maps thus defined, the couples a category of their own. A cfunctor T: Q > 3D in
duces a functor on the category of couples of
6
into that of
33.
We now
each
come
to the
main objective
of this section.
Let G be a
(2) for
Q
shall consider systems ccategory. for each object A e C and each integer (1)
We
A
:
H =
q, q,
H
{H q (A),a+,d( at p)} where
Q
(A)
is
Q
a group,
>
map
a\
>
B
in
e and each
integer
a+:
H (A)
B
H
(B)
is
a homomorphism, integer q, d (a>$)
H
> > (3) for each couple (a, ft): > // is a homomorphism. _,(A) q (C)
A
C and
.
each
The groups
or g c and homomorphisms belong to just one of the categories Such a system // will be called a covariant dfunctor on the ccategory
6 provided the following four axioms hold: AXIOM 1. If a = identity, then a^ = identity. AXIOM 2. 0&*)* = ** AXIOM 3. If 71 ,72,73 form a map of the couple (a,0):
into the couple (i,ft):
A
B
C
A,
>
B,
>
C,, then commutativity holds in
the diagram
I*.
j.
118
CATEGORIES AND FUNCTORS
AXIOM
4.
[CHAP. IV
the sequence
For every couple
9
(,/3):
fit
A
B
,
>
C
 H...01) 4is
fff (Q < ff,(fi)

ff,(4)

exact.
These four axioms are precise replicas of the first four axioms for a homology theory. = In an analogous fashion we define a contravariant dfunctor Q  B and (A) for a: A {#<(A),V<.*>} where a*: H'(B)  ff' +1 (C) for (a,j8): A > B > C. The axioms are (..: #'(4)
H
H
replicas of the first four
axioms of cohomology. Let be a covariant dfunctor on a ccategory be a covariant cfunctor. The composition
H
3D
and
let
T:
6
>
2D
HT =
is
{H,(TA),(Ta)t,d (Ta
.
T ft
}
}
then clearly a covariant dfunctor on e.
3D
Similarly
if
H
is
a contra
variant 5functor on
and T:
C
> 3D is
covariant, then
.
HT = {ff^),(r)v (a
still is
w
}
a contravariant ^functor on 6.
Thus the composition
HT
is
well defined
whenever
T
is
covariant.
If T is contravariant, HT may still be defined formally as above but the result is neither a covariant dfunctor nor a contravariant 6functor. This suggests the definition of two additional "mixed" types
of functors,
functors.
namely
of covariant 5functors
and
q
of contravariant d)
J
In a covariant 5functor
H =
and
a^ H*(A)
for
(a,8):
a contravariant dfunctor ff = a*: #,(JJ) > #,(4) for a: A > J5 and {^(A), a*,d (a ,^} we have > // a ,(C) for (a,/8): A > /? * C. The axioms red (a ,,,: //,(^l) quired are the obvious reformulations of the four axioms listed above.
>
 H(B) A > B
for a:
A  B
In
{H
(A) a^
:
d (a
^
)
}
we have
jff
(a ., }
H
q
(C) >
aM
(A)
C.
defined.
With the mixed theories at hand, the composition HT is always The situation is summarized in the following theorem, the
is
proof of which
immediate.
Let T:
<B
>
3D be a cfunctor and H a dfunctor The composition HT is then a dfunctor [dfunctor] on and T have the same variance^ and HT 6. Further HT is covariant if and T have opposite variances. is contravariant if Each homology theory on an admissible category a furnishes an example of a covariant dfunctor on either of the ccategories associated with Ct. Similarly each cohomology theory on Ct furnishes an example
THEOREM
8.3.
[dfunctor]
on
3D.
H
H
9]
hCATEGORIES AND hFUNCTORS
Examples
of
117
will
of a contravariant 6functor.
in
mixed functors
occur
Chapter
v.
The connection between the mixed functors and the unmixed ones can be further illuminated by a procedure that will be referred to as "the signchanging trick." Given a covariant functor H =
\H Q (A),a^,d (a> p)\,
as follows:
define a covariant 3functor fL
= {H
Q
(A),a^,8 (a
,
S)
\
H\A) = H.
(precisely, a,,:
Q
(A),
9
a+
=
,
5 (ai/5)
=
d (a ,0)
(B) is denned as a+: H_,(A) > H. Q (B) the jsignchanging trick to we obtain H. etc.). Conversely applying Thus the correspondence establishes a 11 correspondence be
H*(A)
 H
H
H
H
tween covariant dfunctors and covariant 6functors.
cussion applies to contravariant functors.
A
similar dis
9.
ACATEGORIES AND /iFUNCTORS
The introduction of the concept of a ccategory was a step toward the consideration of homology theories defined on categories more general and abstract than on admissible categories of spaces. We saw
that the
first
four axioms can be formulated with ease in this
more
remaining axioms, general setting. not be stated since, in a ccategory, we lack the concepts of "homotopy," "excision," and "point." To complete the picture, we make the following definition:
As
for the
it is
clear that they can
DEFINITION
in C,
9.1.
An
a:
hcategory
6
)
is
is
a ccategory in which
given for
(i)
>
a
binary relation a c^ p(a homotopic to
(ii)
maps a,/3: C are singled out and are called excisions, and (iii) certain objects of C are singled out and are called A covariant dfunctor [contravariant 5functor] on G which points.
certain
A
B
maps
A
B
in
satisfies
the analogs of the Homotopy, Excision, and Dimension axioms
will
be called a homology [cohomology] theory on the /icategory 6. It should be noted that since no properties of homotopies, excisions, and points are required, the above definition has a purely formal character,
and is made only to supply a language convenient for later uses. >Abe two maps in an DEFINITION 9.2. Let a: A B, 0: B B are both homotopic A and aft: B Acategory e. If /to: A to identity maps, then a. and ft are both called homotopy equivalences, ft A map a: A > B is called a homotopy inverse of a and vice versa. in 6 which is a composition of a finite number of excisions and homotopy
equivalences
is
called a generalized excision.
118
CATEGORIES AND FUNCTORS
THEOREM 9.3. // Q and a: A
:
[CHAP. IV
H
is
a homology [cohomology] theory on an h,
category
H,(A)
1
B is a generalized excision = H,(B) [*: H\B)
n
then
H\A)}.
Let a. = a a be a representation of a as a composition it and homotopy equivalences. Since a^ = a+ suffices to show that each a is an isomorphism. If a* is an excision, this follows from the Excision axiom. If a is a homotopy equivalence = (a*0% = identity and with 0* as homotopy inverse, then = identity. Thus a+ is an isomorphism. similarly &*<*+ DEFINITION 9.4. A covariant cfunctor T e > 3D on the Category 6 with values in the /icategory 3D is called an hfunctor if T preserves homotopies, generalized excisions, and points. Explicitly: if a in C, then Ta T0 in 3D; if a is a generalized excision in C, then Tot is a generalized excision in in e, then TA is a ); and if A is a point
PROOF.
of excisions
,
a^
:
~
~
point in
3D.
a 3D be a covariant hfunctor and on 3D. Then the composition HT is a homology [cohomology] theory homology [cohomology} theory on (B. PROOF. By 8.3, is a covariant dfunctor [contravariant 5functor]. There remains to prove that HT satisfies the Homotopy, Excision, and Dimension axioms. This however is an immediate conse9.5.
THEOREM
Let T:
6
H
HT
quence of 9.4 and
9.3.
As
in 8.3, Definition 9.4
and Theorem
9.5 could also be formulated
for contravariant functors
T provided we
consider theories
H of & mixed
character,
i.e.
ing the analogs of the last three axioms.
covariant 6functors and contravariant dfunctors satisfyThese mixed theories do not
later chapters.
have any geometric analog, but will arise occasionally in In particular the following theorem will be used in v,12.
3D be a contravariant hfunctor and a covariant dfunctor [contravariant dfunctor] on 3D satisfying the Homotopy, Excision, and Dimension axioms. Then the composition HT is a cohomology [homology] theory on Q. REMARK. In proving that a ofunctor T is an Afunctor, it suffices to show that T preserves homotopies and points and carries excisions into
THEOREM
9.5c.
Let T:
6

H
generalized excisions. Then it follows immediately that alized excisions into generalized excisions.
T carries
gener
10.
COMPARISON OF dFUNCTORS AND OF HOMOLOGY THEORIES
In this section the discussion will be limited to covariant dfunctors with values in the category R but all that will be said applies equally
,
10]
FUNCTORS AND HOMOLOGY THEORIES
119
well to the contravariant case as well as to 5functors with values in
9
or
gc
.
= {H g (A\a^d (a>$) and H = \H Q (A), JDEFINITION 10.1. Let <**>d<a,/5)} be two covariant dfunctors on the ccategory G with values
}
H
__
_
in
g/j.
A
homomorphism
h:
H  H H
g
is
a family of
homomorphisms
h(q,A):
(A)

H
q
(A)
defined for each
conditions.
relation a^h(q,A)
A
e (B
If a:
A
=
and each integer q, subject to the following two > B is a mapping in C, then the commutativity holds in the diagram h(q,B)a+
h
H
q
(A)
>
_ H (A)
g
h
H
If
(e*,/3):
_
>
q
(B)
II q (B)
A =
>/?
C
d(a.8)h(q,C)
h(q
is a couple in C, then the commutativity relation l,A)d (atft holds in the diagram
)
h
each h(q,A) is an isomorphism H Q (A) ~ H Q (A), then we say that an isomorphism and write h: II ~ II. h The above definition applies automatically to homology theories since a homology theory is a dfunctor defined on an admissible category (or an ftcategory) and satisfying additional axioms. Theorems in, 10.1 and 10. Ic may now be restated as follows: THEOREM 10.2. Let H and H be two homology [cohomology] theories defined on the category 3 of triangulable pairs. Every homomorphism
If
is
h
:
G
>
>
G
homomorphism
of their coefficient groups admits a unique extension to a
h:
H
Hon3.
// h is
an isomorphism, then
so is h,
120
NOTES
DEFINITION
10.3.
An
admissible category
a
is
called a uniqueness
category for homology [cohomology] if, for any two homology [cohomology] theories defined over ft, each homomorphism of the coefficient
groups
hQ
:
G>G
can be extended uniquely to a homomorphism,
h:
ff>#ona.
Whenever h
1
ing
/lo
>
to
G
G,
G
an isomorphism, so also is h. This is shown by extendobserving that h'h and hh' are extensions of identities A', jind * G. Then uniqueness insures that Afe' and h'h are identities.
is
is
may be restated as The category 3 of triangulable pairs and their maps a uniqueness category both for homology and cohomology. In Chapter xn a considerably larger uniqueness category <!! will be
In this terminology 10.2
10.4.
THEOREM
constructed.
With the uniqueness theorem at hand, one can see why the mixed types of homology theories are not considered for categories of spaces.
is a covariant 3functor defined on the Suppose, for example, that 3 satisfying the Homotopy, Excision, and Dimension axioms. category Apply the signchanging trick of 8 to H. There results a homology
H
theory
H
on
3.
It follows that
H
By
Q
the uniqueness theorem,
(X,A)
=
for q
>
0.
for q < 0. (X,A) mixed theories, the Thus,
Q
H
for
positive dimensional groups are trivial.
NOTES
The categories of "all topological spaces" and not handled carefully, lead to the antinomies usually associated with a "set of all sets." To avoid these, one must either restrict the notion of a category and not speak of "the category of all groups," or else place oneself in an axiomatic system in which the
Logical foundations.
if
"all groups,"
"category of all groups" is a legitimate object not leading to antinomies. Best suited for the latter purpose seems to be the system of von Neu
mannBernaysGodel [P. Bernays, A system of axiomatic set theory, Journal of Symbolic Logic 2 (1937), 6577; 6 (1941), 117; 7 (1942), K. Godel, The 6589, 133145; 8 (1943), 89106; 13 (1948) 6579:
consistency of the continuum hypothesis, Annals of Math. Studies 3, Princeton 1940]. This system contains the notions of "class" and "set," each set being also a class, but not vice versa. Then the various cate
gories introduced in Chapters
i
and iv can be proved to
exist as classes.
EXERCISES
One must take
121
care not to perform on these categories certain operations
all
(such as forming the set of but not on classes.
subsets) which can be performed on sets
The transitivity of homotopies. on the binary relation a p (a
no conditions were imposed homotopic to /3). In most of the that will be considered this relation will be reflexive, symAcategories It will also be compositive in the sense that, metric, and transitive. if o,j8: A  B, a',0': B > C, a 0'0. 0, and a' 0', then a' a One may always expand the relation of homotopy in an Acategory C so as to obtain these properties. This is done as follows: As a first ~' 2 if = 0i7i, 2 = &7 2 with 2 and 7! 72 step, define The next step is to define a ~" if there exists a finite sequence a. = ~' a,+i an = i,a 2 (n > 0) such that, for each i < n, either a or cti + ~' a, The new relation ~' is reflexive, symmetric, transitive and compositive. For any homology theory on (B, it is easy to prove that a ~" implies a # = 0^. Thus we may always replace the relation c^ by the relation ~".
In
^
9.1,
is
~
~
~
!
i
l
~
~
.
,
,
t
'
!
.
If
we use
this
homotopy equivalence
transitive.
broadened concept of homotopy, the relation of defined in 9.2 becomes reflexive, symmetric, and
EXERCISES
A.
1.
>, 8, define the concept 2D B C m contra variant in covariant in 6,, with values in 8. Generalize the concepts of natural transformation and natural equivalence to such functors.
,
FUNCTORS OF SEVERAL VARIABLES. Given categories e n <3 m 2Di
, ,
,
of a functor
T
>
,
t ,
,
2.
Exhibit the commutativity and associativity laws of the (external)
direct
sum
G\
G,
+
(G 2
+ +
(? 2
C.,)
~ 
^2
((?i
+ GI +G +G
2)
3
as natural equivalences of functors.
FUNCTORS OF GROUPS. Consider the category of all groups (including nonabelian groups) and homomorphisms. Consider the functions which assign to each
B.
1.
group
gories.
(1) its
commutator subgroup,
automorphisms.
(2) its center, and (3) its group of Convert these into functors on appropriate subcate
122
EXERCISES
DEFINITION. A functor T on one of the categories 9/z, $ c with values in the same or another of these categories is said to be additive = T<t>, if Tfoi G  G'. The functor 2) T<t> 2 whenever 0,,0 2
+
+
:
T
is
said to preserve exactness
If
if it
transforms each exact sequence into
0,
an exact sequence. 2. Let T be additive.
<t>
=
then Tj
=
0.
If
G =
0,
then
TG =
3.
,
0.
Let
n)
TG a
4.
be a covariant additive functor. If i a G a > G (a = 1, an injective representation of G as a direct sum, then Ti a > TG is an injective representation of TG as a direct sum. Let T be a covariant additive functor. If
T
:
is
:
is is
exact,
and the image
of
</>
is
a direct
summand
of G, then the
same
true of
   T<t>
TV
>
TG,
>
TG
TG
2
>
0.
if it
5.
Prove that a functor preserves exactness
G'
>
if
>
and only
carries
each exact sequence of the form such exact sequence.
C. dFUNCTORS AS FUNCTORS.
1.
G
G"
>
into another
Given a covariant [contra variant] d or 5functor
<B,
H
on a
ccate
gory
then
Axiom
4.
assign to each couple (,#) of C the exact sequence given by Show that if (71,72,7.3) maps (,/3) into another couple (ai,0i),
[(7*,7*,7*)]
(7i # ,72*,73 # )
yields a
homomorphism
of the corre
sponding exact sequences.
Show
that one obtains thus a
new
co
variant [contravariant] functor H* defined on the category of pairs of (3 with values in the category of exact lower or upper sequences^ Formuand // are two If late the Axioms 14 in terms of the functor H*.
dfunctors and h:
//
>
H H is_a homomorphism, then h induces a natural
>
transformation h*:
formation T:
2.
H*
/P H*
//*.
What
conditions
it
must a natural
trans
satisfy in order that
be of the form F
=
h*?
that the operation of replacing superscripts by subscripts and changing their sign at the same time carries a 5functor into a
dfunctor of the
3.
Show
same variance.
If
H
is
a covariant dfunctor and
T
is
a covariant [contravariant]
is a functor on groups to groups which preserves exactness, then covariant dfunctor [contravariant 5functor]. Examine the situation
TH
for the remaining three cases for
H.
Show
that the result also applies
EXERCISES
to homology and cohomology on an Acategory.
123
theories on an admissible category, or
D. MODULES AND VECTOR SPACES.
Let G be a module over an integral domain D (see i,Exer. H). Consider pairs (d,g), del), d^O, g zG and define equivalence (d,,0t) to mean that d d 2 g\ d'di<? 2 for some d ^ 0. (e/ 2 ,0 2 )
r
~
f
1.
Show
by
G.
that this
is
equivalence class of (d,g)
classes
2.
a proper equivalence relation. Denote the by [d,g] and the set of all these equivalence a Dmodule under the operations
Show
K0i]
that
G becomes
=
+
[rf,,ff a ]
[e/AAfl.
g
>
+
d,fc],
d'[d,g]
=
[d,d'd.
3.
Show
that the
map
[1,0] is
a linear homomorphism
G
G,
and examine
4.
its kernel.
Consider the case
D = G
(i.e.
D
the obvious composition law).
of quotients of
5.
Show
that I)
regarded as a Dmodule with is a field, namely the field
D.
Show
that
G is
a vector space over
[di,dj[d,0]
D under the
operation
=
[d,d,d 2 0].
6.
Show
that the rank of
G
over
D
is
the same as the rank of
of the
l
G
over D.
7.
G2
,
define
Given a homomorphism <: G! * = [d,<f>g] and show that 4>[d g]
}
Show
that the pair of operations G,^ is gory g of /)modules with values in the category Q over D. Show that this functor preserves exactness.
8.
Dmodules G! and 2 is linear over D. a covariant functor on the cate2
<f>:
G
G
>
G
of vector spaces
Given a homology [cohomology] theory // with values in the category g D of Dmodules, define H as the composition of H with the functor of 7. Show that H again is a homology [cohomology] theory
with values in Q&.
Show
that the Betti
numbers are the same
in
H
and/f.
CHAPTER V
Chain complexes
1.
INTRODUCTION
and the next four chapters
is
The primary aim
tion of
of this
the construc
theories with prescribed coefficient groups. The con> comstruction divides roughly into three steps, namely: (i) space
homology
complex
plex
(ii)
>
groups. This chapter in character.
chain complex, and (iii) chain complex > homology is devoted to the third step and is purely algebraic
A chain complex (called a group system by W. Mayer) is a lower sequence of groups in which the composition of any two successive homomorphisms is zero. The constructions on chain complexes leading
homology groups are suggested by the results of Chapter Analogs of the Axioms of Chapter I are proved for these derived groups. In this way we obtain a "Ilomology theory" for chain comto their
in.
plexes.
Cochain complexes are also defined. formal difference from chain complexes.
However, they bear only a
In order to introduce homology groups over a general coefficient group, two methods of constructing groups out of groups are discussed. These are the tensor product, C (x) G, of two groups, and the group of
homomorphisms Hom(C,G) of C into G. Using these, we assign a new chain complex and a new cochain complex to any given chain complex and a coefficient group G] and we obtain thereby the homology and
cohomology groups
of the chain
complex over G.
2.
CHAIN COMPLEXES
of
DEFINITION 2.1. A chain complex groups and homomorphisms d Q
K
is
is
q
:
C
(K)
a lower sequence {C a (/C),dJ > Cq^K) such that
S q id q
=
for each integer
q.
C
Q
(K)
called the
K, and d Q is called the boundary homomorphism. A map /: K > K' of one chain complex into another is (as in i,2) a sequence of homoC Q (K) > C Q (K') defined for each q and such that morphisms f q
:
group of qchains of
/aA =
*:/.
2.2.
DEFINITION
Let
K =
{C q (K),d q
124
}
be a chain complex.
The
3]
COUPLES
125
kernel,
Z 7 (/), of d q is called the group of qcycles of K. The image, (K) of d +i is called the group of qboundaries of K. Since d Q 3 Q+l = 0, = is a subgroup of Z Q (K), and the factor group Q (K) g (K) Z q (K)/B q (K) is called the qdimensional homology group of K. If  K' is a map, then / a sends Z q (K) into (#') and ,(#) into /:
B B
Q
fl
H
K
B
q
(K')
> 9 (K) Q (K ). thereby inducing homomorphisms f^: The following theorem is an immediate consequence of the defini>
H
H
f
tions:
THEOREM
identity
2.3.
// /:
K
If f:
+
K
>
is the identity
map, then f+
>
is the
homomorphism.
K
K' and
g:
K'
K"
',
then (gf)+
=
Chain complexes K and their maps / constitute a category denoted by dg/j or dg c according as the groups C q (K) are in the category g* or g c Then 2.3 asserts that the pair H q (K)J+ is a covariant functor from d$ R to g* [or from d$ c to g c ]. The concept of a cochain complex K = \C Q (K),d 9 differs from a Q+l 9 > C chain complex in that 5: C (K) (K) and that the dimension Thus an application of the "signchanging is written as a superscript. trick" (iv,8) converts a chain complex into a cochain complex and vice versa. The discussion of cochain complexes therefore differs from the
.
*/*
}
5 discussion of chain complexes in terminology alone. is the group of </cocycles, coboundary operator, Z"(K)
is
called the
q
H
(K)
is
the
pair (K)J* gdimensional cohomology group, a covariant functor on the category dQ K [or 6g c ] of cochain complexes with values in the category R [or g c ].
etc.
The
H
q
constitutes
3.
COUPLES
complexes
In this section
we
shall convert the categories of chain
into ccategories (iv,8) dfunctor.
and we
shall
extend
H
Q
(K),f+ to a covariant
We begin with the observation that if is a chain complex and L a subsequence of 7v (in the sense of i,2), then both L and K/L are again chain complexes called the subcomplex and factor complex reMoreover the inclusion map i: L > X, and the natural spectively.
is
K
map
i)\
K
>
K/L
yield
an exact sequence
i
r?
(i)
>
L
>
K
>
K/L
>
0.
This suggests the following definition: DEFINITION 3.1. Let L,K,M be chain complexes.
The maps
126
0:
CHAIN COMPLEXES
L
>
[CHAP.
:
V
/f,
^:
K
+
M are said to form a couple ($$)
L
>
K
+
M,
provided the sequence
(ii)
0>LKM0
i.e. if
*
is
exact,
If,
the kernel of
of
<
is
</>
zero,
^
is
onto,
and kernel ^
of 7f
=
image
is
<.
further, the image of
is
a direct
all q),
summand
(i.e. if
0(C a (L))
called
is
a direct
It
is
summand
C
Q
(K) for
then the couple (0,^)
direct
and d$ c become
clear that, with couples defined as above, the categories dQ R One obtains different ccategories by ccategories.
taking direct couples only. Circumstances under which it is necessary 11 to consider direct couples rather than all couples will appear in
and
12.
In the subsequent lemmas and definitions,
we assume
q
that an exact
l
sequence
given .__ Let Z a (M) = t~\Z Q (M)\ __ LEMMA_3.2. = Z Q (M)/B a (M). Then q (M)
(ii)
is
_ B
H
(M)
=
t~ (B Q (M)), and
Z
and
\l/
q
(M)
=
d'foC^CL)),
B
(M)
Q
(M)
= B
Q
(K) \J
(C G (L)),
induces isomorphisms
~t:
H
Q
H (M).
q
i.e.
PROOF. Let c e C g (K). Then if and only if tyc = 0. This
c e
is
Z (M)
Q
if
and only
i/^dc
if
$c
0,
s
Z
q
(M),
equivalent to
=
and since
kernel
Suppose c e B q (M). Then fa e B q (M) and i/'C = d6 for some b e C a+1 (M). Let d s C a+1 (/iC) be such that ^d = b. Then ^(c  dd) = dd. ^6 = 0, so that there is an e e C a (L) with </>e = c ^c (tyd = ^c = ad + <t>e, and therefore c e # (K) VJ </>(C (L)). Conversely, Thus c = fad + t<t>e = d^d, and fa e B,(M). Thus if cj= dd + <^e, then fa
^ = image
</>,
this is equivalent to dc e <i>(C Q .i(L)).
The last part of morphism theorem.
3.2
is
a direct consequence of the Noether
the
iso
LEMMA
defines
The boundary homomorphism of homomorphisms
3.3.
chain complex
K
B
//ze A"6r??e^
q
(M)
l
of
Q
<t>
is zero,
<t>~
d defines homomorphisms
Z (M)
> Z.x
and induces a homomorphism
A:
/
3]
COUPLES
PROOF.
Let
c e
<f>dd
127
dc
C
d
Q ,(L).
Then
Z =
g
(M).
d$d
=
By
3.1,
we have
0,
=
<f>d
for
ddc
=
and therefore 3d
=
some d e 0. Thus
e
Z^CZO and
If c e
dc
e*[Z a _,(L)].
B
Q
e e
C
fl
(L).
<t>e for some d e C<,+i(K) and (M), then, by 3.2, c = dd d<e = 0de, and dc e ^B q ^(L). Then dc = ddd
+
+
DEFINITION
3.4.
The homomorphism
defined as the composition d+ L phism of the couple (</>,^)
:
= A^'
>
1
is
/
M
called the boundary
.
homomor
<f>
be useful to have a more direct description of the homomorphism d+. Suppose h s Q (M). Choose z e Z Q (M) belonging to the coset A, and choose c e C q (K) with ^c = 2. Then dc is in the image of and <~ dc e Z q  (L) is a cycle in the coset dji. Another description
It will
H
l
}
of d+
obtained by stating the analog of m,7.5. The star in d+ has been inserted to distinguish it from the boundary operator within the chain complexes. However, in later uses we shall
is
omit the
star.
3.5.
THEOREM
functor on
category
the ccategory
The system H = \H q (K)J^d^\ is a covariant d3Q R [dgc] of chain complexes with values in the
K [g c ].
1
Axioms
correspond to
and 2 of iv,8 are contained in 2.3, while Axioms 3 and 4 Theorems 3.6 and 3.7 that follow.
3.6.
THEOREM
(</>,VO:
Let (gj,h):
L
>
K
*
M
(0,^)
fa',^') be
(<t>',f):
a
map
into the couple
L' >
K
f
of the couple > M', then
commutativity holds in the diagram
H,(M)
7/.(M')
d* a,
ff._,(L)
>
.,(!.')
PROOF.
diagram
This
is
a consequence of the commutativity relations in the
H,(M)
I*.
*
<
H (M')
a
<
 H.W  "f
_
A
^..(L)
_ ^
I/*
_
A
^,(M')
._,(L)
this
where /^ is induced by /. The commutativity relations in are immediate consequences of the definition of ^ and A.
diagram
128
CHAIN COMPLEXES
THEOREM
3.7.
[CHAP.
V
// (0,^):
L
K
>
M
is
a couple, then
the sequence
is (exact.
PROOF.
(1)
*
By
definition,
we must prove
t/^
three propositions:
kernel
(2) (3)
kernel d^ kernel <t>+
= image 0^, = image ^ # = image d^.
,
Suppose that A e //(!/) and z e Z a (L) lies in the coset h. Then ^02 = 0, and therefore ^*0*ft = 0. Thus, image + C kernel ^. = 0. Then, if z e Z(/Q lies in Suppose now that h e q (K) and i/^ft
H
the coset
ft,
it
follows that ^z
e
B
q
(M),
i.e.
^2
e
=
dc for
some
c e
\l/z
Let 6
dc
e
C q+l (K)
be such that $b
dz
=
c.
Then
an a
^(z

d6)
=

C a +,(Af).
d^6
e
=
dfr.
= 0, and Since 06a = d<^a =
dc
If h' s
it
therefore there
is
ddb
=
0, it
= Q (L) with 0a and a follows that aa =
C
</>a
z
Z
(L).
H
q
(L)
is
the coset of
f
a, then, since
and
2 are in
the
same
coset,
follows that
<t>^h
=
h.
e
To prove
(2)
suppose that h
is
H
a
(K) and
let z
Z
Q
(K) be in the
coset A. Then ^z from the definition
0, it follows (M). kernel Therefore image ^^ now that h e q (M) and that d^h = 0. Let z e Z a (M) be Suppose #* in the coset A, and let c s C^/C) be such that \l/c = 2. Then ^T'dc is in Z a _!(L), and is in the coset d^/i. Since d^/i = 0, there is a 6 s C a (L)
in the coset
$Ji
of
H
Q
Since ^2
=
of d+ that d^l/^h
0.
C
H
such that d6
so that c
=
the 06 e Z q (K), and ^(c 06) = h. coset of c 06 in q (K), it follows that $+h To prove (3) suppose that h e q (M), and let 2 e Z q (M) lie in the coset h. Select c e C q (K) with ^c = z. Then there is a 6 e Z a _i(L) with Since 06 e B q .i(K), q ^(L). 06 = dc, and 6 lies in the coset d^h of
z.

0"*dc or
<f)db
=
dc.
Then

d(c
=
$c =
r
</>6)
=
dc
if
0d6
=
0,
Thus
h' is
H
H
H
it
kernel 0*. Suppose now follows that 0^^^^ = 0. Thus image d+ = 0. Let 2 e Z q (L) be in the coset ft. Then that h e q (L) and <j>.Ji
H
C
02
it
e
^02
5 (JK), and 02 = dc for some = 0. Thus ^c e Z a+1 (M). If
a
c e
e
C a+1 (X).
is
ft.
Then d^c =^dc =
ft'
H q+l (M)
d^ft'
follows from the definition of d # that
=
the coset of ^c, then This completes the
proof of 3.7.
The
similar
3.5
is
definitions and theorems for cochain complexes are quite and are obtainable by the signchanging trick. The analog of
3.5c.
THEOREM
The system
H
*=
{H
Q
(K),f^,d*} is a covariant d
4]
HOMOTOP1ES, EXCISIONS, POINTS
[Sg c ] of cochain complexes with
129
functor on the ccategory 6g
category
<
valws in
the
R [g c ].
We
return to the remarks
made
at the beginning of this section
and observe that whenever possible we use the operation of factorization in groups and chain complexes to pass from pairs to single objects. Thus instead of the pair (K,L) where L is a subcomplex of K we consider
the exact sequence
triple
(K,L,N) we
>
K/L
future.
0.
> L > > 0. K/L Similarly instead of the > consider the exact sequence L/N > This point of view will be followed systematically in the
K
K/N
4.
HOMOTOPIES, EXCISIONS, POINTS
we
define homotopies, excisions,
cKj c
,
In this section
ccategories
and points
in the
dQ R and
The system
H
thereby converting them into Acategories. \H Q (K)J^d^} becomes then a homology theory. The
particular definitions of "homotopy," "excision," and "point" adopted are motivated by the applications of the next two chapters.
DEFINITION 4.1. Let two maps of K into K'. D: f 0) is a sequence
K
A
of
Q
:
and K' be chain complexes and
chain homotopy
let f,g
be
^
D
of
/ into g (notation:
homomorphisms
D
such that
C Q (K)>C Q+1 (K')
d g ^D q
If
+D
Q
^d =
q
gq

/,.
such a homotopy D exists, / and g are called homotopic and we write c^ g. If c e C Q (K), then D Q c is called the deformation chain of c. / LEMMA 4.2. The relation f c^ g is reflexive, symmetric, and transitive. PROOF. If D = 0, then D: f c^ f. If D: / g, then D: If D: and D': g D': f then D h. g /. f g A, > L, /',0': L > jl/, / LEMMA 4.3. // f,g: and f g ^,
~
~
~
~
+
K
~
~
~
PROOF.
Let D:
Di'
4.4.
f
q f' ^
^ g and D':
l
C a+1 (M) by
D":
=
D + D^;
q
Define Z>;': C q (K) > /' c^ 0'. then a short computation gives
are chain homotopic, then their in
f'fc^g'g.
THEOREM
// f,g:
K
>
K
f
duced homomorphisms
coincide:
f*
=
s
Q+.
PROOF.
Let D:
fz
fl
Thus
gz

/~
g,
and
let z s
Z
q
(A") and /,
 ^.
(K).
Then
dZ)z
ISO
CHAIN COMPLEXES
DEFINITION 4.5.
L.
>
[CHAP.
V
maps K isomorphically onto THEOREM 4.6. // /: K
A map /: K+L is called an excision if and only if
L
is
an
excision, then
axiom" is of course trivial due to the narrow definition of excision that has been adopted. very = {C Q (K),d} is called pointltke DEFINITION 4.7. A chain complex if d: C Q (K) C Q .i(K) is an isomorphism for q even and >0, and also for q odd and <0.
isomorphically onto q (L). The proof of this ' 'Excision
H
/ # maps
H
q
(K)
K
THEOREM
4.8.
H
//
K
is pointlike, then
H
q
(K)
:
=
Q
(K)
= C
for q
^
0,
and
(X).
Let q be even and >0. Since d Q C g (K) ~ C Q ^(K), it = and B q _,(K) = C q ^(K). Thus #,(#) = follows that Z q (K) = 0. Similarly ff a (X) = 0, #.,(10 = for 9 odd and and Q ^(K) <0. Since did 2 = and cLid = and d 2 and d_i are isomorphisms, it follows that 3, = 0, d = 0. Thus Z Q (K) = C (/Q and B (^:) = 0.
PROOF.
H
Hence H (K) Theorems
= C
(K).
4.4, 4.6,
4.9.
THEOREM
on
<3* [Sc]
and 4.8 combined with 3.5 yield = {H 9 (K)J^d+\ is a homology theory The system
H
the hcategory
d$ R [d$ c ] of chain complexes, with values in
the category
Acategories, one with all couples
be kept in mind that actually d R [d$ c ] represents two and the other with direct couples only. Theorem 4.9 holds with either meaning for d$ R [d$ c ]In the /icategory d$ R [d c ], we have the concepts of homotopy
It should
equivalence and generalized excision as defined in iv,9.2. Since every excision in d$ R [d$ c ] is also a homotopy equivalence, and since the composition of homotopy equivalences is, by 4.3, again a homotopy
equivalence, it follows that in the category d$ R [dg c ] lences and generalized excisions coincide.
homotopy equiva
excisions, and points in the ccategories of cochain complexes are quite similar and are obtainable from ^9/z^Sc those for chain complexes by the signchanging trick. The results for
The definitions of homotopies,
cochain complexes
may
THEOREM
functor on
4.9c.
be summed up in the following analog of 4.9: = {H 9 (K),f+,d*} is a covariant The system
H
6
d$ R [69c] of cochain complexes, satisfying the and Dimension axioms. Homotopy, Excision,
the category
6.
DIRECT SUMS AND PRODUCTS
This section reviews the definitions and basic properties of the
cartesian product of spaces,
and the
direct
sum and
direct product of
groups.
5]
DIRECT SUMS AND PRODUCTS
5.1.
1S1
M
DEFINITION
,
Let
i.e.
for each
a
e
MX
,
[X a be a
]
collection of sets indexed
by a
set
a is
a set of the collection.
onM
The
product
(1)
II *a
\
of the collection
is
and such that x a
element x a
projection
(2)
is
the totality of functions x = \x a defined for the value of x on a is an element of a
ex
.
e
M
X
The
called the acoordinate of x.
For each
s
M,
define the
p,
:
[I ^a
a.Af

X,
by
p,(x)
=
*,.
In case the sets a all coincide with a set X, then the product (1) is M denoted by and is simply the set of all functions from to X. DEFINITION 5.2. If each is a topological space, a topology is a introduced in (1) as follows: If a finite number of a 's are replaced subsets U a the product of the resulting collection is a a by open subset of (1) and is called a rectangular open set of (1). Any union of rectangular open sets is called an open set of the product. The product with this topology is called the cartesian product. It is immediate that
X
X
,
M
X
X
CX
,
the projections
(2)
are continuous.
LEMMA
5.3.
A
function
f:
Y  II X* atM
is
defined on a space
Y
with values in a cartesian product
e
continuous if
and only
if,
for each a
M,
the
function
pj:
is continuous.
Y+X
a
following classical result is due to Tychonoff. For a proof, see Lcfschetz [Algebraic Topology, Colloq. Pub. Amer. Math. Soc., 1942,
p. 19].
The The
proof
is left
as an exercise to the reader.
THEOREM
is
5.4.
The cartesian product of a
collection of
compact spaces
compact.
We recall that the term "compact space" is used here to denote a Hausdorff space in which the Borel covering theorem holds. DEFINITION 5.5. If each a is an abelian group, then an addition
X
is
defined in (1)
by the usual method
(x
of
adding functional values:
(3)
+
x') a
=
xa
+
xi.
In this
way
(1)
of the groups
{X a }.
becomes an abelian group and is called the direct product If each X a is a compact abelian group, then the
18*
CHAIN COMPLEXES
[CHAP.
V
product with the topology of 5.2 and the group operation just defined is a compact abelian group called the direct product. If each a is an jRmodule over the same ring JK, define addition in (1) by means of (3) and multiplication by a scalar r e R by
X
(4)
(rx) a
(1)
=
rx a
.
called the direct product. cases the projections (2) are homomorphisms.. The direct product is thus defined for a collection of groups from either of the categories 9*>Sc, an<i the product belongs to the same
Then
becomes an 72module
In
all
category.
The same symbolism
(1) will
be used to denote each of these
products; the particular product in question can always be determined from the nature of the a 's.
X
In case
all
the groups
X a ,a
e
direct product (1) is denoted with values in G. on
M
M by G and
M coincide with a single group
is
(7,
the
simply the group of functions
DEFINITION
Their direct sum
5.6.
Let {(?} be an indexed collection of ^modules.
atM
is
the subgroup of their direct product Y[ G a consisting of those elements having all but a finite number of coordinates equal to zero, i.e. g a =
e
Ga
for all
but a
finite
number
of
a
e
M
.
For each
e
M,
a a
define the
injection
(5)
*:
G,>
Z0.
is
(a
for for
=
9*
0,
by
(
(0
It is clear that ip is
In case
M
an isomorphism of
finite collection,
a
G onto a subgroup of then of course the direct
ft
sum and
direct product coincide.
In case each G a is a compact group, ]T) G> as a subset of Yl G, has a topology which makes of it a topological group. In addition, each p G a is not a is continuous. As a subspace of the direct product, closed subspace (indeed, it is everywhere dense), and is therefore not compact. It is for this reason that the direct sum is not a useful operaft
^
tion to apply to
compact groups.
5.7.
\
A protective representation of a group G as a direct product consists of a family of groups {G a indexed by a set M, and a homomorphism iy of G onto G a for each a, such that the homomorphism
DEFINITION
rj:
G
I~[
Ga
is
tion, each
Ga
by (770)* = vj a g is an isomorphism. If, in addias a factor group of G by a subgroup, and i\ a is the given
defined
6]
FREE MODULES AND FACTOR GROUPS
1SS
natural homomorphism, then G is said to decompose into the direct product of the factor groups {G a }.
DEFINITION
of
5.8.
Let
G
be an /^module.
An
injective representation
}
sum consists of a family of J?modules {G a indexed by a set M, and a homomorphism a of G into G for each a, such that the homomorphism G, defined by assigning to each element ]T) G a of G the sum ]T) 9<* of the images of its nonzero coordinates in
as a direct
:
G
2
is
G,
an isomorphism.
If,
in addition, each
Ga
is
a subgroup of
G and
the inclusion map, then G is said to decompose into the direct sum of the subgroups {(?}. This is the case if and only if each g e G can
a is
be expressed uniquely in the form
where
If
o^,
is
,
a m are
distinct elements of
M.
clear that ]~lG a any {G a } are represented as the direct product and sum, respectively, of the groups {G a by means of \p a and {i a }.
collection of groups,
it is
}
\
6.
FREE MODULES AND THEIR FACTOR GROUPS
6.1.
DEFINITION
set
Xi,
TJ
Let
G
be an ^module and
if,
X a subset of G.
any
The
X
=
is
,
said to be linearly independent z of X, the relation nz,
for
distinct elements
r t s 72, implies
+
+
= rn = 0. The module of G contains X.
G, then
set
If
X X
is
is
said to
= 0, generate G if
r nx n
linearly independent
no proper suband generates
is said to be a base of G and G is called /ree. In this case every element g e G can be represented uniquely as g = r&i r n x n where r e #, # e X, and x xn are distinct. is a base of G and G' is any 72module, then it is clear that If G' can be extended uniquely to a homoevery function 0:
X
+
+
t
t
t ,
,
X
X
morphism
<f>:
G
>
G'.
It is well
known
that,
if
R = F
is
a
field,
then every vector space
G
over
>
F is LEMMA
free.
6.2.
// G,G' and
H
#
that
0:
homomorphism 6: H > G swc/t tfia <0 = ^. PROOF. Let X be a base of H. For each z
<f>6(x)
G' are homomorphisms,
are Rmodules, <t>: is /ree, is onto and
G Men
G' and ^:
#
there exists
a
so
e
X
select 0(x) e
G
X
=
^.
G
This is possible since ^(x). extends to a homomorphism 6:
6.3.
=
<f>
is
>
onto.
The
function
satisfies
H
G
which then
is free,
<^0
LEMMA
direct
summand
// of G.
H is a
submodule of
G and G/H
then
H
is
a
194
CHAIN COMPLEXES
PROOF.
Consider the natural homomorphism
i:
t\ :
[CHAP.
V
G
>
identity
that
i?0
=
map
i.
G/H
G/H.
This implies that
By 6.2 there is a 0: G///  G such G decomposes into the direct sum of
G/H and
the
the image of 6 and the kernel of rj which is H. Since every vector space is free, we have
vector space is a direct summand. For any set X, let R x be the set of all functions X > R such that fx 7* for at most a finite number of x e X. /: With addition and multiplication defined by
6.4.
COROLLARY
Every subspace of a
DEFINITION
6.5.
(A
+
/,)*
=
fix
+
/,x,
(rflx
=
r(/x),
with the As is customary, we identify each x e on rr and zero elsewhere. Then, if / has the nonzero rn on x lf values riy x n respectively, and is zero elsewhere, we have / = J^i r x It follows that R x is a free 72module having X as a base. It is called the free Rmodule generated by X. If X' is a subset of X, the inclusion map X' C. X extends uniquely is represented to an isomorphism of R x with a submodule of R x If as a union \JX a of disjoint subsets, the inclusion maps X a C X induce an injective representation of R x as a direct sum: R x ~ ]C Rx a DEFINITION 6.6. Let G be an .Rmodule and C G a set generating G. The inclusion map can then be extended to a homomorphism 6: G which maps JR X onto G. Let F be any set generating the Rx kernel of 0. Then we say that G is represented by generators X and relations Y. Note that each element of Y can be written as a formal finite linear combination of elements of X with coefficients in R, and, when evaluated in G, this linear combination yields zero. LEMMA 6.7. Let G be the Rmodule given by generators X and relaA function tions Y and let G' be given by generators X and relations Y G > G' if and only X > X' can be extended to a homomorphism f:
#*
is
an 72module.
is 1
X
function which
,
,
t
% .
>
.
X

X*G
X
1
f
.
<t>:
if for every element
r&i
+
RX
',
>
+
is
PI/ (#1)
+
+
r n f(x n ) e
r n x n e F, r, e R, x, e X, the element a linear combination of elements of Y'.
</>
homomorphism extending f is unique. PROOF: Let ^: R x > R x be the homomorphism extending / and > G'. let N and N' be the kernels of the natural maps R x G, R x Then Y and Y 1 generate TV and N' and the condition of the lemma is
>
If this condition is satisfied, the
equivalent to $(N)
The homomorphism is then induced by ^. a principal ideal ring (i.e. a domain of integrity in which each ideal has the form Rr for some r e 72), then every submodule of a free Rmodule is also a free Rmodule. be a base for G, and let PROOF. Let G be a free J?module, let be a submodule of G. We assume the choice axiom, and may therefore
N'.
<
C
LEMMA
6.8.
// R
is
X
H
7]
FINITELY GENERATED GROUPS
135
x ai are well ordered: x\,x^ suppose the elements of For each ordinal a, let A a be the submodule of those elements of which are linear combinations of the xp with ^ a, and let J3 a be the which are linear combinations of the submodule of those elements of rx a where < a. Then each a e yl a is of the form a = 6 Xfi with The coefficient r is a function of a and provides a homomorphism b s Ba The hypothesis on R implies that I a is of A a onto an ideal / of R.
.
,
X
H
H
+
.
a free Kmodule.
By
6.3,
.
Aa
is
the direct
sum
of
Ba
and a submodule
We will show that is the direct sum of the isomorphic with / a = 0. Aa submodules C a If a < 0, then C a #/*; hence C a C\ C c where c e C ao Let a e //; then a e A ao for some a. Thus a = a! Then a e ^4 0 for some a < a and eh e 5 Iterating this proCa
.
C
H C
ft
+
.
l
l
.
obtain a sequence a n < a n _! < cedure, n such that a = a n+1 e C a for i = 0,1, c,
we
<
,
,
+
cn
+
and elements + CQ where

Since each decreasing sequence of ordinals is finite, it is the direct sum follows that a n +i = for sufficiently large n. Thus
e
.
a n +i
B an
H
of the
modules
is
Ca
.
Since each
Ca
is free, it
follows that
is
H
is free.
When R
group.
the ring of integers, an 72module
an ordinary abelian
Hence 6.8 implies COROLLARY 6.9. Any subgroup
7.
of a free abelian group is free.
FINITELY GENERATED GROUPS
In a great many of the applications of homology theory the spaces involved are triangulable and the coefficient group is the group of In such a case, the homology groups encountered are geneintegers. rated by a finite number of elements. The structure of such a group
is
easily described:
it
is
a direct
sum
of cyclic groups.
Its structure
can be characterized by numerical invariants.
any
finite set of generators and relations, obtaining a second set of generators which gives the direct composition, and, at the same time, the numerical invariants.
Moreover, starting with there is an algorithm for
sum
de
Conse
quently the
finitely
generated groups are of the utmost importance in
applications of
homology theory.
The object of this section is to review this algebraic material and state the applications to chain complexes. recall that a matrix A = (a,,) with integer elements is said to
We
be unimodular if it has the same number of rows as columns, and its 1. In this case the inverse of A, denoted by determinant is +1 or A~ = (a,,), has integer elements and is unimodular. Under ordinary matrix multiplication, the unimodular matrices, of a fixed order, form
l
a group.
1S6
CHAIN COMPLEXES
} ,
, ,
[CHAP.
V
Let F be a free group with a finite base x be a unimodular matrix of integers of order
y,
xn and
let (a M )
=
]C"i
ti*
(*'
=
*
*
'
1?
>
n) al so
f
Then the elements rm a base for F. To prove
n.
x,
this, observe, first,
tion, the
i/'s
2
*,
i
c ^ a ^i x i
j.
are independent. For suppose 2^7 1 c,r/ t = 0. 0. Since the re's are independent, this implies 23 c
that the y's generate F] for
=
2
^*2/*
I*1
addi
Then
t
a,,
=
xn
F'
for each
Therefore
= 23
c t o t; a
;jfc
= 23
c t 5 tjb
=
ck
.
Suppose now that F,F' are free groups with finite bases x respectively (see 6.1 with 72 = integers). If and xi,
,
Xj,
,
/:
F
a homomorphism, then, for each i, /(x t ) = 23? 1 b %i x' for some integers 6,, and / is completely determined by the matrix B = (>,/). X nmatrix B corresponds to a homomorphism. If Conversely, any f the bases (x),(xf) are transformed into bases (y),(y ) by unimodular
is
m
1
.
matrices of integers
A
and A', then the matrix of /
in terms of (y),(y
f
)
becomes ABA'"
Thus, to the homomorphism / corresponds an
equivalence class of matrices of integers. At this point we state for future reference the classical theorem on the reduction of such matrices
to canonical form
by an
If B is an unimodular matrices of integers
THEOREM
(1)
7.1.
equivalence. nmatrix of integers, then there exist A and A' of orders n and m, respectively,
mX
such that
d ly dz
,
in diagonal form, and (2) if and d is a are the diagonal elements of C, then each d ^
the
matrix
C = ABA''
,
1
is
t
1. Furthermore C is unique (m,n) in the sense that the array of diagonal elements is the same for all equivalent diagonal matrices satsifying (2).
divisor of
d l+ i for
i
=
1,
Min
that there exist bases (y\(y
(1)
In terms of the homomorphism /: F F', the theorem asserts f ) of F,F' such that
f(y<)
=
d iy :,
is
(=
number
1,
...
,
P).
The number
p is the rank of
B
and
the
of nonzero diagonal
elements of C.
This result has
Suppose, for example, that / is an isomorphism. Then each d, = 1 = n. Therefore C is unimodular, and B = A~ 1 A' is also uniand
many
consequences.
m
Thus, in terms of any bases sponds to a unimodular matrix.
modular.
If this last result is
(x),(x'),
an isomorphism corre
= identity, applied to the case F = F' and / we obtain that any two bases for F are related by a unimodular matrix. The number n of elements in any base is the same for all bases. It is
easily seen that
n
is
the rank of
F
(see Exercise
i,H for definition of
rank).
THEOREM
7.2.
Suppose
f
of a free group
F
of finite
the group G is isomorphic to the factor group rank by the homomorphic image of a free
m
7]
FINITELY GENERATED GROUPS
137
Then G can be decomposed into the direct sum of and r g m r finite cyclic groups of orders 1 and T where each 0, > divides 0, +1 for i = 1, 0,, r 1. The numbers r,0 1? T are invariants of G. G is decomPrecisely, if posed in two ways as a direct sum of cyclic groups in which the orders
group
r of finite rank.
F
m infinite cyclic groups
, 
%
,
,
r,0,,
of the finite cyclic groups divide one another successively, then the numbers r are the same for the two decompositions.
,
As shown above we may choose bases (y), (y ) in F,F' so that the homomorphism /: F > F has the form (1). Let J' (i = 1, be the subgroup of F' generated by y{, and J. (i = 1, m) Then F decomposes into the p) the subgroup generated by d<y(. direct sum and the image of / decomposes into ^J J This ^<>
PROOF.
f
f
,

'
,
X
t
.
yields
r to be those d's which exceed 1, then the asserted has been demonstrated. decomposition Since r is just the rank of G, its in variance is obvious. Let G' be the subgroup of elements of finite order in G. For any integer n,
If
we
define
0,,
,
ng defines a homomorphism G' G'; let nG' denote the image group, and <t>(n) the number of elements in nG'. Given any decompog
>
sition of
G as described
,.
above, the numbers
t
are easily
T
shown
to satisfy
=
0t
*
2
(0j,n)
(0 2; n)
(0 r ,n)
where
<r
(0,n)
,
=
1,
denotes the greatest T, the equation
common
divisor of
and
n.
For any
=
holds
of
if
1
we
set
n
=
0,,
and
i
is
+
1
.
false for
any
T) is
positive
r
n
less
than
6,.
This
follows since
0,
divides
Therefore
0, (a
n
satisfying 0(n)
=
1;
and
<
the least positive value the least positive value of n
is
satisfying
This provides an inductive definition of T ,0 r _i, function <t>. Since is invariantly defined, the variants. This completes the proof.
,
0j
in terms of the
0's
are likewise in
THEOREM
group
7.3.
The conclusions of
the preceding theorem hold for
G
having a finite number
m of generators.
any
138
CHAIN COMPLEXES
,
[CHAP.
V
Let rr,, x m be a set of generators of G, and let F' be the free group they generate (apply 6.5 with R = the integers). Then the inclusion {x C G extends to a unique homomorphism h of F' onto (?. It follows that G is isomorphic to F /F where F is the kernel of h. By As the rank of F cannot exceed that of F', it 6.9, F is a free group. follows that F has a finite base. Let / be the inclusion map F F'. Then we have shown that G satisfies the hypotheses of 7.2, and the
t j 1
C
proof
is
complete.
REMARK.
tive steps.
An analysis of the proof of 7.3 reveals several nonconstrucOne
is
the nonconstructive nature of the proof that the
integers form a principal ideal ring (see 6.8), and another is the use of the choice axiom in the proof of 6.2. In contrast, the proof of 7.2 the reduction of the matrix to diagonal form) is entirely con(including
structive; the decomposition of G can be found in a finite number of steps starting with bases in F,F' and the matrix for /. In applications of this section to finite chain complexes, we shall use 7.2 rather than
7.3,
and thereby remain within the realm
of effective procedures.
8.
CANONICAL BASES IN A FINITE COMPLEX
8.1.
DEFINITION
q,
A
chain complex
K
Q
is
C
q
(K)
is
a free group on a
finite base.
Then
is
said to be finite, if, for each q (K) also has a finite
H
set of generators. The rank of K. The invariants 0J,
R
,
Q
H (K) 0?, of H (K),
of
Q
called the q
tb
Belli
number
the
th
<?
lorsion
ft q .
numbers of J.
8.2.
The
described in 7.2, are called ranks of C Q (K) and B (K) are denoted
then
by a Q and
THEOREM
If
K is a finite chain complex,
Furthermore, there exists a set of bases, one for each lowing properties: For each q, the base for C q (K)
types of elements,
C q (K),
is
with the folfive
composed of
&
Cg,
l
dQ
,
e7,
= = = m=
j
If 1, 1,
"
i
* ' *
TV
Kq,
T,_I,
Q
K
,
I
,
1,
,
_i
TI.
For each
q,
the
homomorphism d Q
is given
by
aX =
o,
aX =
o,
ad =
o
8]
CANONICAL BASES IN A FINITE COMPLEX
139
Such a
to be
set of bases is called canonical, and the boundary operator is said in diagonal form. PROOF. Since C a ,(/C) is free, we have, by 6.3, that Z q (K), the
,
kernel of d a
is
a direct
summand
of
decomposition of C Q (K) into a direct Then d q defines an isomorphism of
W (K)
q
C q (K). For each sum C (K) = Z
Q
q,
we choose a
9 (K). Since both
q
(K)
+ W
into
Z
q
.i(K).
are free groups on finite bases, there exist bases (y),(y f ) in these groups (see 7.1) such that d q has the form
dy>
Since the kernel of d Q
are relabeled
=
t
d,y[,
i
=
1,
,
j8i.
is Z Q (K), no d is zero. Those y m for which d m = 1 the corresponding y are denoted by a?i. The reel, The maining y's are denoted by d Q) and the corresponding y"& by b Q . are denoted by cj_j. With these choices of base elements remaining y"s
l l
}
.
for each
q,
the boundary operator has diagonal form.
into the direct
decomposes
sum
of cyclic
Then H Q (K) groups generated by the cosets
.
of the base elements b[
scribed in 7.2.
and c*. This decomposition has the form dek Since the Therefore the number of generators c Q is R Q torsion numbers are invariants, the d's associated with the b q s must be the torsion numbers of This establishes the canonical form q (K).
H
boundary operator and the ranges of the indices j, k, and I. Since the elements a* and 6*b q form a base for B Q (K) of rank /3 a the
of the
,
and therefore of m, must be as indicated. The relation on R q follows from the fact that the number of generators of all five types is a Q It is important from the computational standpoint to observe that the reduction to the canonical form of 8.2 can be carried out in a finite number of steps, for any finite number of q's, starting with bases in each group and the matrix of integers describing each d Q in terms of these bases. This follows from the fact that the reduction to diagonal form described in 7.1 is a finite process. One would begin with the least q and reduce d Q to some diagonal form. This provides a new base for C Q including a base for Z Q and a new form for the matrix of d q + which
range of
.
i,
,
l
now
involves only the base elements of Z q The matrix of d^: C + t > Z Q is then reduced to diagonal form. This provides a new base for C a +i C a + 2 * Z a n, etc. including a base for Z Q+l and a new matrix for d a4 2
.
.
:
,
The analogs
the reader.
of 8.1
and
8.2 for finite cochain complexes are left to
UO
9.
CHAIN COMPLEXES
THE TENSOR PRODUCT
[CHAP.
V
DEFINITION
G
(1)
is
The tensor product C (x) G of two /^modules C and 9.1.* the 72module generated by the set of all pairs (c,g), c s C, g s G with
(c t
relations
(c,flfi
(2)
(rc,flf)
 (c!,0)  (c ,0) = 0, = 0, a )  r(c,0) =  r(c,0) = 0, (c,rg)
+c +
G
2) g)
2
2
(c,fifi)
(c,flf )
0.
obtained then as follows: Let R(C,Q) be let Y(C,G) the free Bmodule "generated by the set of pairs (c,g) and all the elements of the form be the least subgroup of B(C,0) containing
Following
6.5,
C
is
(rc,flf)
r(c,fif),
(c f rg)

then
CG =
will
R(C,ff)/Y(C,G).
is
g.
The element
R(C,G)
of C (g) G which be denoted by c
relations are
the image of the generator (c,g) of These elements generate the group
C
G
and the
QM
(2')
fe
c
+c
(ffi
2 ) (g)
+
ff a )
= ci = c
(g)
g
+
(g)
c2
(g)
(g)
gf,
(re) (g)
gr
=
r(c
g)
^ + = c
c
2
,
(r0).
If
gr
is
a fixed element of G, the correspondence
>
>
c
>
c
(g)
gr
gr
is is
morphism C morphism G
(3) (4)
C (g) G. Similarly, if c is C (g) G. Thus (!') implies
9
fixed,
^
^ c
a homoa homo
(2 O
LEMMA
9.2.
=
Z
(c<
gr
9),
c c
(g)
=
X)
0.
0,
=
= Z)
(c
ff,),
0.
//
^)
G is an Rmodule,
defined by f(r
=
the correspondence f:
R
G
G
/?
>
G
G.
r^ is
7n
/Ae segi^eZ fro^A
B
(g)
G
an isomorphism. Similarly, G and G 72 wiB &6 identified with
the
=
by these
isomorphisms. PROOF. Consider
/'(TJ?)
/' is
first
=
map
rgr.
Since R(R,G)
is free,
/': R(R,G) and the pairs
G
defined
by
a homomorphism.
It is easily seen that /'
(r,0) form a base, has the value on each
of the four relations of (1)
and (2). Therefore /' maps Y(R,G) and thereby induces a homomorphism / of the factor group R
into 0, G into
* Assume throughout that R is commutative. Tensor products over noncommutative rings are treated in the book of H. Cartan and S. Eilenberg: Homological Algebra (Princeton Press) 1956.
9]
TENSOR PRODUCTS
g)
141
G
such that f(r
=
rg.
Since /(I
0r)
=
</,
/
is
onto.
Applying
(2')
and
(3),
we have
Therefore each element of
If /(I
0)
R
0.
G
can be written
1
in the
form
1
is
0.
=
0,
then g
=
This implies
>
=
>
0.
Hence /
an
isomorphism.
DEFINITION
9.3.
If/:
(g)
C
A:
C"
and ft:
G
G' are homomorphisms,
the correspondence c
(/c)
(g) (ftp)
defines a
homomorphism
/
called the
C
G
> C'
G'
homomorphism
h.
of the tensor product induced
by the homodefines a
morphisms / and
Precisely, the
R(C',6') which carries F(C,G) into 7(C"G'), homomorphism J?(C,G) and thus induces a homomorphism / (g) A of the tensor products. In case G = G' and A is the identity, we shall speak of / (g) h as

map
(c,g)
>
(fc,hg)
the
homomorphism C
(g)
G
>
C'
G
induced by / and will denote
is
>
it
by a symbol such as
/'.
The proof
of the following proposition
9.4.
immediate:
ij: CG>C
ft:
THEOREM
//
i\
C
>
C, j:
G
G
are identity maps, then
Iff: CC',/': C' > C", = (/' (g) ft')(/ h). V: G' > G", tfien (/'/) (fc'fc) This theorem states that is a covariant functor of two variables, in the category Q with values in g*. THEOREM 9.5. Let C and G be represented as the direct sums (see
is the identity.
G
G > G',
5.8)
C =
6t/
a*M
E
Ca
,
G 0Y
G,
means of
the injections
Then, for a
e
M and
ft
e
N,
the
homomorphisms
ia
je:
Ca
G^^CG
sum
representation
have zero kernels, and they provide a direct
CG =
PROOF.
identity,
/8'
(a, ft)
Z
Ca
>
Gp.
Let p a
:
C
>
Ca
,
r^:
G
G/j
be maps such that p a i a
=
for
p a 'i a
=
for a'
^
a, rft jft rft )(i a
=
.;
identity,
5^
ft.
It follows that (p a
=
and r ,j = identity and (p a
ft
ft
,
r^O
142
(i<*
jft)
CHAIN COMPLEXES
[CHAP.
V
=
for (a,0)
^
(<*',')
be written uniquely as a
ponents.
/:
finite
Any element ft of 23 <? G, can sum h = a ^ of its nonzero com
^
/i
Define
23 C
(p a
G,
 C
c
G
by
/(ft)
= 23
ft
(t.
Then
23
?>)/ projects h into its
component
a 0.
It follows that
has kernel zero.
*aC
= and and the proof onto,
LEMMA
9.6.
g be any generator of = c and /(23 c 90) 23 JMt*
Let
is
C
(g) 0.
G.
Then
c
=
/
Therefore /
is
complete. G is // C is a free Rmodule with base X, then C x elements x with relations x 2) g (gi g \generated by = x rgr. // G zs aZso a /ree Rmodule with base F, x 2 awd r(x gr) <7 G is a /ree Rmodule with base {x y}, x z X, y e Y. 2ften C
+
l
an easy consequence of 9.5 and 9.2. 9.7. If f is a homomorphism of B onto C, then the induced C (g) G is also onto. homomorphism B (x) G G is a finite sum 21 c PROOF. An element of C ^ For each
This
is
LEMMA
*
c t select
a b
%
e J5
such that
//
/(&,)
=
/
ct
.
Then /(23
h
^
^)
=
S
c
^
LEMMA
9.8.
0A>#>C0
zs
an
exact sequence of Rmodules
and homomorphismSj then
the
induced
sequence
AG>BG>CG~>0
is also exact.
r
w
further, the
no statement is made about the kernel of /'.) a direct summand of B, then the sequence image of f
(Note:
is
//,
>
n AG r>Gr
C
G
>
is exact,
and the image of f is a direct summand of B G. PROOF. The second part of the lemma follows from the first and from 9.5. By 9.7, h is onto. It therefore remains to show that image = kernel h'. Let F denote the image of /'. If a e A, g s G, then /'
h'f'(a
g)
=
(hfd)
=
/i'
0<7 =
0.
Therefore T
is
contained in the
kernel of
h'.
Hence
induces a
homomorphism
ft":
B
17:
G/T
*
>
C
G
such that the composition of the natural
<f>:
map
B
G
B
G/r followed by ft" is ft'. Since ft is onto, there is a function
ft<
C
*
B
(not necessarily a
homomorphism) such that
*':
is
the identity.
Define
(C,G)
+BQQ/T
by
0'(c,0)
=
9]
TENSOR PRODUCTS
US
obvious that <'
r(c,g)
Since R(C,G)
(c,0i
+
2)

is free,
</>'
is
(c,00
e
~
a homomorphism.
It is
(c,0 2 ), r(c,g)

(c,r0),
and

maps
into
(rc,gr)
zero.
Suppose c u c 2
fc(*(ci
2)
C and
<ci
g
e
G.
Since
+c 
4>c 2 )
= =
e
jty(d
(d
+ +c
c 2)
d
fod
c2
/i^c 2
2)
=
0,
exactness implies there Hence 0c 2 <j>Ci
.
is
an a
A
such that /(a)
=
0(c,
+
c2 )
c 2)
c2 )
 (d) g  to  4>c
?
g
2 ) (x)
^)
foc 2 )
(x) g)
=
r/(/(a) (x) ^)
=
i>/'(a (x) g)
=
0.
Thus
0":
</>'
CG*BG/Y.
<i>hb
maps F(C,6 )
into zero.
Therefore
<f>'
induces a
homomorphism
Furthermore
But
6
is
<f>"h" identity, and, therefore, kernel h" = 0. follows that T is the kernel of h'. This proves 9.8.
=
in / 04);
hence
i)((<t>hb)
g)
=
rj(b
g).
This implies
r
Since h
=
h"rj, it
where J
be an ordinary abelian group, or equivalently, a Jmodule the ring of integers. If p is an integer, the operation g * pg is a homomorphism of G into G whose image is denoted by pG. LEMMA 9.9. Let G be an abelian group, p an integer, and let rj\ G * G/pG be the natural homomorphism. Then the correspondence n (x) g > Let
is
G
ij(ng) defines
an isomorphism
(J/pJ)
G  G/pG.
PROOF.
the natural Therefore
Define/:
J by /(n) = pn\ and let h: J J/pJ be homomorphism. Then fji satisfy the hypothesis of 9.8.
J
>
/'
h'
J(g>G>J(?is
(J/pJ)
G
>
;
exact.
By
9.2,
we may
identify
J
homomorphism
g p0; hence kernel onto, the stated isomorphism is proved.
A'
G with G. Then / becomes the = image /' = pG. Since A' is
is their greatest
LEMMA
9.10.
If p and q are integers, and r
common
divisor, then
(J/pJ)
(J/qJ)
=
J/rJ.
This follows from 9.9 in view of the wellknown proposition
(J/tJ)/p(J/qJ)
J/rJ.
144
CHAIN COMPLEXES
[CHAP.
>
V
J
As an example, let G be a cyclic group of order 4, and let /: J be defined by f(ri) = 2n. Then J (x) G is cyclic of order 4, and
2(n (g) g). Thus the kernel This example shows that the concept "subgroup" is not preserved in general under tensor products. This means that the symbol c (g) g is ambiguous in a situation where c e
J(g)G
>JGis given by f(n
G
/':
(g)
0)
=
of / is zero, while that of /' is not.
C
C
C" or g
far
e
C
G
f
.
(g) as a functor (of two variables) defined However we may also consider (g) as and g* with values in g a functor defined on g ( = the category of ordinary abelian groups) and g* with values in g*. Indeed let C be an abelian group and G an Rmodule. Define C (g) G by regarding G as an abelian group. Then convert C (g) G into an jRmodule by setting
So
we have
considered
on g
fl
.
We
must
verify that,
is
if
/:
C
>
C',
/i:
G
G' are homomorphisms,
then /
(g) /i
a homomorphism of 72modules.
(g)
Indeed we have
(/c) (g) r(hg)
r(/
h)(c
(g) 0)
=
r[(/c)
(g)
(%)]
=
(g)
With
this verified all the previously stated results
(g)
remain valid for
this
modified
product.
product to a functor on g' and g c with values in g c where g' is the subcategory of g consisting of This refinitely generated abelian groups and their homomorphisms.
shall also generalize the tensor
,
We
quires
preliminaries. Let C be a free group on a finite base c if element of C (g) G can be written uniquely in the form ]jTj c
some
LEMMA
9.11.
,
cn
.
t
(g)
gt
.
Each The
function defined by ^ c
is
Q\
=
}
==
'
*
(gi
*
j
>
<7>)
f.J
an isomorphism of
)
C
(g)
G
with the direct
sum Gn (=
the direct product
G
11
of nfactors equal to G.
PROOF.
into
]
If
.
Ck
is
Ck
Applying
the subgroup generated by ck> then C decomposes ~ Ck 9.5, we obtain an isomorphism C (g) G
G.
Then
9.2 provides
an isomorphism
Ck
(g)
G ~
^
G.
Combining these
yields the isomorphism /. DEFINITION 9.12. Let
G
/:
be a compact group and
the base c ly group.
,
cn
.
The
direct product
Gn of
nfactors
is
C a free group on G is a compact
to carry
The isomorphism
C (g) G
=
GT of 9.11
now used
9]
n
TENSOR PRODUCTS
C
(g)
145
over the topology of G into a topology for C (x) G. Then compact group and / is continuous. LEMMA 9.13. The topology of C (x) G is independent of
the base in C.
G
is
a
the choice of
LEMMA 9.14. Let C,D be free groups on finite bases, and f: C > D a homomorphism. Let G,H be compact groups and h: G a homoi s continuous. morphism. Then f () h: C (x) G > D (g) PROOF. We shall prove 9.13 and 9.14 simultaneously. Let d cn d m be bases in C and D used to define the topologies in and d u C (g) G and D <g> H. Then / is given by /(c ) = Z7i a ,d, (i = 1, n) where (a,,) is a matrix of integers. It follows that
H
H
,
,
,



t
t
,
Z
c.
0.
= =
Z Z a" ZX Z
ji
n
( \
a.ifcflU/
ti
Using the isomorphisms C (x) G h corresponds to a map that f
<j>:
= G D H = # of Gn * # m given by
,
m
9.11,
we
see
=
( \
Z
ti
a.i
Z
a.
Since h
continuous and the a's are integers, it follows that is conh is continuous; and 9.14 is proved. Now let Hence / D and G G', and let f,h be the identities. Then / (g) h is the C identity and is continuous. This proves 9.13.
is
<
tinuous.
DEFINITION and let
9.15.
Let
C
be a group with
finite set of generators,
i
n
R
>
F  C
be a representation of C as a factor group of a free group F on a finite base (i is the inclusion map, and rj is the natural homomorphism). Let G be a compact group. In the induced diagram
G
R
(g) G and and kernel
F
y'
(g)
G
are compact,
i' is
continuous, and
by
9.8,
77'
is
onto
image i'. In this way C (x) G is isomorphic to the compact group F (g) G/image i'. Using this isomorphism we carry over the topology of the factor group to provide a topology in C (g) G. This is equivalent to defining a set U C C (g) G to be open if i)'~ (U) is open
l
=
in
F (g)
G.
146
CHAIN COMPLEXES
LEMMA
9.16.
[CHAP.
V
the
//
C
has a
finite set of generators
and
topology of
C
(x)
G
is
independent of the representation
G is compact, C = F/R
.
LEMMA
9.17.
//
C,D
>
compact, and f:
CG*DHis
PROOF.
C
D,
h:
have finite sets of generators, and G,H are G > are homomorphisms, then f (g) h:
H
continuous.
The two lemmas
i
TJ
are proved simultaneously.
ii
Let
rh
R
>
F >
a
C,
R, > F, >
D
F
>
be the representations used
in defining the topologies of
D
f*l
(x)
H.
i?i/i
By
6.2, there is
homomorphism
f^.
C G and F such that
l
Therefore commutativity holds in the diagram
CG
By
9.15,
17
77'
and
/i)
?/[
are continuous.
By
9.14,
/,()*
is
(g)
continuous.
/I)TJ'
Then
J(/,
is
continuous.
is
This implies that (/
is
conl
open in D (x) #, it follows that 77'' '(/ h)~ (U) is open in F (g) G. By the remark at the end of 9.15, this last implies that (/ (g) K)~ (U) is open. Hence / (g) h is continuous, and 9.17 holds. If we specialize / and h to be identities, we obtain 9.16. It is necessary to review previous results under the assumption that C (g) G has a topology and observe the effect on the conclusions. In 9.2, the function / is continuous since it is the one used in defining the topology. The same remark applies to the function / of 9.11, and to
tinuous.
Thus,
l
if
U
the isomorphism of 9.9. If the hypotheses of 9.4,7,8 are strengthened by assuming the continuity of the appropriate functions, the corresponding functions in the conclusions are also continuous by 9.16. In
9.5, if
G
is
to be
ranges
M
and
N
compact and C to be finitely generated,, the index must be finite, each C a must be finitely generated,
ft
and each Gp compact. Assuming each j to be continuous, then is continuous and the final isomorphism likewise. Summarizing we have
i a (x) i p
THEOREM
and
their
9.18.
Let
9
homomorphisms, and
denote the category of ordinary abelian groups let g' denote the subcategory of groups
having finite bases.
(1)
Then
the tensor product is defined in the following cases:
(2) (3)
<?eg, C e g, C e g',
<?eg,
.then
then then
Gs Ge
g*,
gc
,
C G e g, C (g) G s g*, C G e gc
.
10]
GROUPS OF HOMOMORPHISMS
147
All the preceding results of this section are valid in cases 1 and 2. The same is true in case 3 except for the direct sum Theorem 9.5 which is valid
when
restricted to finite
sums.
10.
GROUPS OF HOMOMORPHISMS
Given two ^modules
all
DEFINITION
10.1.
C and
G,
denote by
Hom(C,G)
the flmodule of
homomorphisms
0:
C
>
G
with addition
<t>i
+
<f> 2
denned by
and with the product
r<
defined
by
In a sense, IIom(C,G) is dual to C (x) G. This duality becomes apparent if the results of this section are compared with the similarly
numbered
>
results of
9.
LEMMA 10.2. // G is an Rmodule, the correspondence G defined by /(<) = 0(1) is an isomorphism.
PROOF.
Since
(0,
f:
Hom(J?,G)
+
2 )(1)
=
</>i(l)
+ 0a(l),
</>
/
is
Since
is
1
generates /, /(<)
=
a homomorphism.
72
is
implies
*
=
0.
Since
free,
there
a
having any prescribed 0(1). DEFINITION 10.3. If /: C'
C and
A:
phisms, the correspondence
/K/>/
defines a
homomorhomomorphism
>
G
G' are
Hom(/,A):
Hom(C,G) > HomCC",^)is
The proof THEOREM
of the following proposition
10.4.
immediate:
identity
//
i:
C
G',
then Hom(i,j) is the identity
C and j: G > G are map of Hom(C,G). // /:
h':
C'

maps,
C, /':
C
/;
*
C',
ft':
G
^
G'
>
G",
is
Briefly, the
theorem asserts that Horn
,
a functor contravariant in
the
first
variable in Q R co variant in the second variable in g* and with
.
values in $ R
THEOREM
10.5.
Let
C
be represented as
:
a
direct
sum C
:
~ 2Z,M C a
Let G be represented as a direct Ca by means of the injections i a G * G/j. Then product G ~ n/s.Ar G/s by means of the projections p &
+ C.
the projections
Hom(i a ,p,):
Hom(C,G)

148
CHAIN COMPLEXES
[CHAP.
V
represent Hom(C,(?) as a direct product
Hom(C,(?)
II Hom(C a ,G,),
(a,0) e
MX
AT..
/:
The Horn (i a ,Pp) are the components of a homomorphism > fl Hom(C,(?0). Suppose e Hom(C,(r) is not zero. Then for some a and c a e C a we have a c a 5^ 0. Then there is a is not zero. such that P00tc a ^ 0. Thus Hom(i a ,p/j) applied to and the kernel of / is zero. Suppose a^ e Hom(C a ,G0) Hence f<f> 7*
PROOF.
Hom(C,(?)
<t>
,
<t>i
<
sum
e X N. For any c e C, write c as a finite and define <foc = X) 0a0C. Then e Hom(C,G) e Hom(C, fl ^) for each ft e Af These are the components of a The corresponding e Hom(C,(?) has the property Hom(i a ,p0)< =: <t>ap* Thus / is an isomorphism. LEMMA 10.6. // C is a free Rmodule with base X, then Hom(C,(7) is x > G the isomorphism isomorphic with the module G of all functions
is
given for each
c
=
2
(,)
M
<f>
*
c
<t> ft
.
</>'
<t>
X
y
being obtained by assigning
by<t>.
to
each
</>:
C
>
G
the function
X
>
G defined
The
proof
is
obvious.
// / is a
LEMMA
PROOF.
10.7.
of the induced
homomorphism of homomorphism /': Hom(C,G)
<t>
B
*
onto C, then the kernel
Horn (#,(?)
c e
is zero.
Hom(C,G Then choose b e B such that Hence /'< is not zero.
If
e
f
) is
not zero, choose
f(b)
=
c.
C such that <f>(c) ^ Then (/'0)6 = 0/(6) ^
0.
0.
LEMMA
10.8.
//
0*A*B*C+Q
is
/
h
an
exact sequence of Rmodules
and homomorphisms, then
the
induced
sequence
h'
f
is
> Hom(C,G) > Hom(B,G) ^ Kom(A,G)
is exact.
If, further, the
image of f
h'
a
direct
summand
of B, then the
sequence
is exact
Hom(C,G) > Hom(B,G)  Hom(A,G) *
image of
h' is
r
and
the
a
direct
summand
of
Hom(B,G).
part of the lemma follows from the first and from 10.5. By 10.7, the kernel of h' is zero. It therefore remains to e Hom(C,<?), then (/'A'</>)(a) = prove that image h' = kernel /'. If = since hf = 0. Therefore image h' kernel /'. Suppose <t>hf(a)
PROOF,
The second
C
10]
e
GROUPS OF HOMOMORPHISMS
/'.
140
for b e kernel h. A, and <(&) = Since /i of B/kernel h into G. is onto, B/kernel A is naturally isomorphic to C, and <' can be regarded as an element of Hom(C,G). Clearly h'tf = <; and we have image
kernel
<
Then
</>/(a)
=
for
a
e
Hence
'induces a
homomorphism
</>'
h'
=
kernel
/'.
*
the
LEMMA 10.9. If p is an integer, the correspondence f: Hom(//p7,G) G defined by /(<) = <(!) is an isomorphism of Hom(<7/pJ,G) onto subgroup of G of elements of order p.
The proof
is
LEMMA
10.10.
immediate (compare with 10.2). // p and q are integers, and r is
their greatest
common
divisor, then
Hom(J/pJ,J/qJ)
J/rJ.
This follows from 10.9 and the isomorphism
(J/qJ)
=
J/rJ.
and Q R with values Horn as a functor 9/j. on g (= the category of ordinary abelian groups) and $ R with values in 9^. Indeed, let C be an abelian group and G an 72module. Define Hom(C,G) regarding G as an abelian group. Then convert Hom(C,G) into an 7?module by setting for C > G and re/?,
So
far
in
As
we have regarded Horn in the case of (g), we
as a functor on g*
shall also consider
<t>:
(r*)(c)
If /:
=
rfoc).
is
C'
>
C and
A:
G
G',
then Hom(/,0)
a
homomorphism
of
72modules.
[r
Indeed,
Hom(/,A)]
=
r(^/)
=
fc(i^/)
=
Hom(/,/i)(r0).
We
in gc.
shall also consider
Horn as a functor on 9 and g c with values
Indeed, let C be an ordinary abelian group and let G be a compact abelian group. Define Hom(C,G) regarding G as an abelian group (without topology). Then convert Hom(C,G) into a topological abelian
group by treating
functions
\f/:
it
as a subgroup of the
compact group
G
of all
C
>
continuous, $(c) condition \l/(ci
closed subset of
sets of
It
is
Since the projections of a cartesian product are c a continuous function of \l/ in G and therefore the
G.
+
G
Gc
,
is
Thus Hom(C,G), as an intersection a closed subgroup and hence is compact.
.
ca) c
V( C
~*
^fe)
=
0, for fixed Ci,c 2 e C, defines
a
of closed sub
C must be checked that Hom(/,/i) is continuous for /: C' and h: G > G' with G,G' compact and h continuous. Let U be a
rectangular open set in Hom(C',G') defined
by the conditions
<c, e V\
150
CHAIN COMPLEXES
c, e C",
[CHAP.
V
l
where
is
open in G' for i = 1, then a rectangular open set defined by \f/fc Summarizing we have THEOREM 10.11. The group Hom(C,G)
V
t
is
,
n.
e
The
l
t
set
Hom(f,h)~
U
t
h~ (V
), (i
=
1,
,
n).
is defined
in the following
cases:
(1)
(2)
(3)
C C
e
e
g*,
g,
(7 e
g,
G s g, G e g*, G e g c>
tfen
tf*n
tf^n
Hom(C,G) Hom(C,G) Hom(C,G)
e e
e
g*, g*,
gc
.
Tta preceding
11.
results of this section hold in all these cases.
HOMOLOGY GROUPS OF A CHAIN COMPLEX OVER A
COEFFICIENT GROUP
The
DEFINITION
group, define
d' is q
If f:
tensor product operation extends to chain complexes as follows: = [C q (K),d q ] is a chain complex and G is a 11.1. If
K
K
G
induced by d qy * K' is a map of chain complexes, let /': (x) G > #' (g) G be the map induced by /, i.e. /J = / The resulting functor from i. chain complexes to chain complexes is denoted by G. By 9.18 there
K
to be the chain complex {C Q (K) <g) G,d} where i.e. d q = d q (g) i where i is the identity map of G.
K
are three cases:
(1) (2)
e9,
Geg*,
then tfen
G: G:
dQ R > dg*, dg * dg^,
(3)
Geg c
,
Naturally one must show that X" (g) G is a chain complex. This is with f, and applying 9.4 to obtain proved by tensoring d a .id a = d'id = 0. Similarly f q ^d q = d c / g implies /J.^J = dJ/J; hence /' is a map.
THEOREM 11.2. // dg*,dS>d9' are regarded as hcategories in the sense G is a covariant hof direct couples (see 3.1), ^,en, in all three cases,
functor.
PROOF.
If (<,iA)
:
L
>
K
>
Af
is
a direct couple, then by definition
is is
exact and the image of is a direct true of the induced sequence
<
summand
of
K.
By
9.8 the
same
so that (0',^')
is
again a direct couple.
Thus
G
is
a cfunctor.
11]
HOMOLOGY GROUPS
151
Since the functor (x)G carries an isomorphism into an isomorphism,
follows immediately that (gX? carries excisions into excisions and points into points. It remains to show that it carries homotopies into
it
> homotopies. Let f,g: L, and is a sequence of homomorphisms D Q
:
K
let
D:
(K)
f
~
C
Q
Q
By definition D g. C Q+ i(L) such that
(4)
d^D + D
Q
q
^d = g,~
/,.
Let D' = D Q (g) i where i is the identity map of G. If we tensor (4) Q with i and apply the relations of 9.4, we obtain (4) with primes on each homomorphism. Therefore D': /' c^ g', and the proof is complete.
DEFINITION
a
11.3.
/ifunctor (x)G with the
According to iv,9.5, the composition of the on d$ R [or SQ C (see 4.9) is homology theory
H
]
domain of (x)(j. It is called the with coefficient group G. For any chain complex K, homology theory is customarily written H V (K',G) and is called the the group H g (K G)
qdimensional homology group of
new homology theory
defined on the
mology group of K over G.
(1)
(2)
As
with coefficients in G, or the q in 11.1, there are three cases:
then then
K
th
ho
Geg*,
(3)
GeS, Geg c
B
,
KidS*, Kedg, Keag',
*n
H (K;G) H (K;GF) H (K',G)
q
9
e
9*, g*,
e
e
g
gc
.
CONVENTIONS AND INTERPRETATIONS 11.4. The group of chains, cycles, and boundaries of K with coefficients in G are written C Q (K',G),
Z
9
(K,G),
and
Q
(K]G) rather than
(x) g,
C
a
(/v (g)
G), etc.
this notation the chain c
gc.
where
of
c e
C
Then, by
9.1,
any element
C
Q
(K'fl)
is
(/Q and g e a linear combination
is
In keeping with G, will be written
^
<7,c,
of elements of
C
Q
(K) with coefficients in G, and any relation
a conse
quence of relations of the form
(0i
+
02)0
=
g,c
+
2 c,
gr(d
1,
+
c2)
=
gc
l
+
0c 2
,
and
(rg)c
r(gc)
=
g(rc) in case
or (rg)c
is
r(gc) in case 2.
Like
wise the boundary operator of
K
K
(g)
G
given
by
and,
if
/:
K A
r/
,
then
/':
G
+ K'
G
is
given by
REMARK.
in
In the case
when G = J
is
the group of integers,
C.
9 to identify
fl
C (x) J with C for any group
Consequently
we agreed K J=
A'
and 7/ (/v;J) = //<,(/). Thus we shall regard the ordinary homology groups as being those based on integer coefficients. An analogous discussion for cochain complexes may be obtained by
152
CHAIN COMPLEXES
[CHAP.
V
plex,
a simple application of the signchanging trick. If is a cochain com= {H(K (x) <?),/*, 5*} is a mixed theory, i.e. a covariant then
K
H
5functor satisfying the
Homotopy,
Excision,
and Dimension axioms.
12.
COHOMOLOGY GROUPS OF A CHAIN COMPLEX OVER A
COEFFICIENT GROUP
The operation Horn of 10 extends to chain complexes as follows: DEFINITION 12.1. If K = {C q (K),d Q is a chain complex and G is a a group, define Horn (#",(?) to be the cochain complex {Hom(C (K'),(j),6 where
}
fl
)
d
is
q
:
Hom(C,(/0,(?)
:

Horn (C c+l (#),(?)
i.e. d
r
induced by d Q+i
C Q+l (K)
If /:
>
C q (K),
q
= Hom(d
a +!,i)
where
i is
the identity
map
of G.
K
K
is
map
of chain complexes, let
be the
map
of cochain complexes induced
by
Q
/, i.e. f'
= Hom(/
a ,z).
The
resulting contravariant functor from chain complexes to cochain complexes is denoted by Hom( ,(?). By 10.11, there are three cases:
(1)
G
e
SB,
then then
<ften
(2)
(3)
Geg, Geg c
,
Hom( Hom( Hom(
,G):
,(?):
,<3):
d$ R > $g, dg > 5g, dg > 5g c
.
If we apply Hom( and ,i) to the relation d 9+l d q ^ 2 we obtain 5 q+l d Q = 0; hence Hom(K,(?) is a cochain complex. ~ " ~
use 10.4,
Similarly
the sense
f Q ~id a
= dj
q
implies
12.2.
THEOREM
= f fq d hence /' is a map. / // d$ R 5g are regarded as hcategories in
5
a
1
/a
1
q
l
'
9
,
of direct couples (see 3.1), then, in variant hfunctor.
all three cases,
Hom(
,G) is
a contra
PROOF.
If
(<t>,\p)
:
L*K+Mis& direct couple, then by definition
is
exact and the image of
is
a direct
summand
of
K.
By
10.8, the
induced sequence
+'
*'
*
is
Hom(M,G) > Hom(K,G) > Hom(L,G) >
and the image
of
\l/'
likewise exact
is
is
a direct summand.
Hence
(^',<//)
a direct couple.
This proves that
Hom(
,(?)
is
a contra
variant cfunctor.
12]
COHOMOLOGY GROUPS
is
1SS
Since Hom(/,i)
it
an isomorphism whenever / a ,t are isomorphisms,
,G) carries excisions into excisions.
follows that
If the
Hom(
is pointlike, i.e. 6 Q C q (K) = C Q .i(K) for 1 q even and >0, and for q odd and <0, then d*' is likewise an isomorphism under the same conditions. Therefore Hom(jff,<7) is a pointlike cochain complex since, after application of the signchanging trick, we
chain complex
K
:
obtain a pointlike chain complex. It remains to show that Hom(
topies.
,G) carries
Let
q
f,g:
K
+
L,
and
let
D:
~ f
gq
g.
homotopies into homoThen D is a sequence
D
q
:
C
(K) >
C Q+l (L)
such that
(4)
a, +1 Z>.
+D
9 .,d Q
=

fQ
.
Let
D" = YLom(D
D'
9
:
Q  ly i)
where
i is
the identity
map
of G.
Then
Hom(C c (L),G) ^HomCC^^/q,^).
to (4)
Applying
Hom(
,i)
and using
10.4,
we
obtain
Therefore D':
/'
c^
0',
and the proof
is
complete.
According to 9.5c, the composition of the contravariant /ifunctor Hom( ,G) with the covariant 5functor on 6g [or 5$c] (see 4.9c) is a cohomology theory defined on the domain of
12.3.
DEFINITION
H
Hom(
is
,G).
It is called the cohomology theory of chain complexes with
coefficient
H K with coefficient group G, or the q homology group of of K over G. As in 12.1, there are three cases:
(1)
group G. For any chain complex K, the group /P (Horn (/,(?)) Q (K;G) and is called the qdimensional cocustomarily written
ih
cohomology group
GeS,
Geg*,
tfed9,
tfedS, tfedg,
then
then
(2)
(3)
#<(K;G) H\K',G)
e
e
g,
g*,
Gsg c
,
CONVENTIONS AND INTERPRETATIONS 12.4. The group of cochains, a q cocycles, and coboundaries of Hom(K,G) are written C ( fC;G), Z (K;G), a Q and B (K]G) rather than C (Horn (X ;(?)), etc. By definition, a ^cochain
J
of
(g
K +
is
a homomorphism 0:
<t>d g+l
:
(?<,(#)
>
>
G.
Its
coboundary 6V
<t>
is
the
l)cochain
d
Q
<f>
C
Q
+i(K)
if
<t>
G.
It follows that
q
is
a gcocycle
(i.e.
=
0)
if
and only
maps
B
(K) into zero.
It also follows
that a ^coboundary maps Z q (K) into zero; however, in general, this > K' property is not sufficient to characterize a coboundary. If /:
K
map Hom( K",G)
is
J
a
of chain complexes, then the induced /':
is
Hom(?C',G)
*
given
by
(/'V)z
= *7A
*' ^
r(K';G),
x
e
154
CHAIN COMPLEXES
[CHAP.
V
An analogous discussion for cochain complexes may be obtained by a simple application of the signchanging trick. If is a cochain = {# 9 (Hom(#,G)),/*,<3*} is a contravariant dfunctor complex, then
&
H
satisfying the
Homotopy,
Excision,
and Dimension axioms.
13.
COMPARISON OF VARIOUS COEFFICIENT GROUPS
we
shall establish various results bearing
In this section
efficient groups.
on the
homology and cohomology groups
of a chain
The
socalled "universal coefficient
complex theorem" has not
for various co
(x)
been included since its statement requires new group operations, besides and Horn, which are not of sufficient importance in the sequel to be
studied here (see Exercises v,G). LEMMA 13.1. If is a chain complex composed of free abelian groups [or of vector spaces over afield F], then Z Q (K) is a direct summand of C q (K). PROOF. As a subgroup of a free group, the group B Q .i(K) is free
K
(see 6.8).
Since d q
maps C a (K) onto
B ^(K)
q
with
of
follows from 6.3 that
(K) LEMMA 13.2. // is a chain complex such summand of C Q (K) and such that q (K) = for C q (K) > C q+i (K) such that homomorphisms D Q
Q
Z
is
a direct
summand
K
H
Z (K) as kernel, it C (K). that Z (K) is a direct
g Q
q
all q,
then there exist
:
d q+]
for all x
e
D
q
v
x
+D
q
^d q x =
x
C
Q
(K).
PROOF.
dq
Let
C
Q
Then q isomorphically ojito B maps we have B q .,(K) = Z Q ^(K). Let d q
:
W
(K)
= Z
(K)
+ W
q
be a direct
(J
sum
Since
decomposition.
defined
by d q
.
Define
D
q
:
i(K) Z q .,(K) denote the q C Q (K) > C g+l (K) as follows:
Q
W
^ l (K).
H
=
0,
+
map
qy
D D
For x
e
q
x
gx
= =
for
d'l.x
l
for
x x
e e
W
Z
q
(K).
W
e
dDx
+
q we have dDx Ddx = dd~ x =
l
+
x.
Ddx
d~ dx
=
x,
while for x
e
Z q we have
x holds
Thus the
and
relation
dDx
}
Ddx =
for all x
C,(K).
13.3.
THEOREM
Let
K
L
be chain complexes
abelian groups [or of vector spaces over a field F] and let f: map. In order that f be a homotopy equivalence (iv,9.2),
composed of free > L be a
K
it
is
necessary
and
sufficient that
JV
for all dimensions
q.
H
g
(K)
= H
Q
(L)
PROOF.
The
necessity of the condition
is
an immediate consequence
13]
VARIOUS COEFFICIENT GROUPS
In order to prove the sufficiency
155
of 4.4.
we
construct a
new chain
complex / as follows:
C<(7)
=
C^K) +
C
Q
(L)
(direct sum),
Since dd(x,y) = d(dx,dy /x) = (ddx.ddy definition indeed yields a chain complex /.
+
+
dfx

/dx)
=
0, this
Next we show that the condition
for all
q.
Indeed,
let (x,y) e
e
Z(/).
= theorem implies q (f) = and dy = /x. Then Then dx
of the
it
H
x
e
Z
a
_!(/0 and /x
e
B^L).
fdx'
that x
(y
B_i(/0.
Thus there
exists
Since the kernel of /^ is zero, an x' e C(/C) with ax' =
follows
x.
*
Then
Q
+
/*')
=
dy
+
=
dy
/x'.
+
/x
=
0,
so that y
s
+
and
fx'
y'
Since /^ maps such that /x"
^(K)
onto ff^(L), there exist x"
Z
fl
(/T)
+
d?/
d(x"

x',7/')
=
Z (L). s C ^(L)
a
y
+
Then
(dx'

dx", dy'
+
/x"

/x')
=
(x,7/),
and
ff f (/)
=0.
C
Q
In view of 13.1, the groups Z (/) are direct summands of therefore 13.2 implies the existence of homomorphisms D q C a+1 (/) such that
0)
di+iD.(*,y)
of the
(f)^
and
>
:
C q (f)
+D
q
^d
q
(x,y)
=
(x,t/).
Each
homomorphisms
Z) a yields
four
homomorphisms
>,'_>: C.^C/0 ^ C f (K), fc f C.(L) > C.(K),
:
D'J:
?
a
 i:
C (L)  C f+l (L), C.^dO > C..t(L),
f
such that
^(^,2/)
=
(D^iX
(i)
+
fc
a y,
15,tX
+
D' 'y). q
Upon computation,
(ii)
condition
yields
= D'dx + Aay + A/x  dD'x  dhy, (in) y = Edx + D"dy + D"fx + dEx + dD"y + fD'x + fhy. /iC is a L in (ii) yields hdy = dhy so that h: Substituting x = = in (ii) and x = in (iii) yields map. Substituting y dD'x + D'dx = A/x  x, dD"y + D'^y = y  Ay.
x
>
Thus D':
of
hf
~ ^ and
D":
fh
~
z2
where
is
i,
and
fa
are the identity
X and L respectively, so that / a homotopy equivalence. maps THEOREM 13.4. Let K and L be chain complexes composed of free K L be abelian groups [or of vector spaces over a field F], and f:
let
>
a
map
such that
/.:
H,(K)
=
H.(L)
1B6
for all q.
CHAIN COMPLEXES
Then
[CHAP.
V
/.:
H (KG) ~ H
Q
9
(L&,
/*:
H*(L;G)
=
for any coefficient group
G [or any vector space G over F]. PROOF. By 13.3, / is a homotopy equivalence. Since the functors (g)G and Hom( ,(?) carry homotopies into homotopies, it follows
homotopy
equivalences.
that they carry homotopy equivalences into Thus the maps
/':
KG*LG,
/":
Hom(L,(?)

Hom(tf,G)
are
homotopy equivalences. and /* are isomorphisms.
Therefore the induced homomorphisms /^
NOTE
Abstract cell complexes. One often encounters chain complexes K in which each group C Q (K) is a free abelian group with a given base (cr?). In defining such an "abstract cell complex/' it suffices to indicate the cells in each dimension and the boundary
of each "cell"
q
<r
.
The
1
coefficients
[o^?"
]
are called "incidence
num
bers"; for a fixed i, only a finite number of The condition dd = is equivalent to
them
are different from zero.
a)
for
2
.
zw^rMw
j
1
:']
=
<>
any pair <r*,crl~ th Let A* denote the matrix whose (i,j) element is [ajicrj" ] Then Q Q each row of A corresponds to a gcell and each column of A correCondition (1) becomes l)cell. sponds to a (q
1
A
If
q
A*~
l
=
0.
then a
)/
K and L are two such complexes with cells and r respectively, = in L determined by the coefficients map /: K
<r
>
is
cj,
f<r*
c^rj.
If
we denote by C 9
that the condition df
=
fd
is
the matrix of the integers equivalent to
l
cj/,
we
find
~ C qB q = A q C q
where {B
q
}
are the incidence matrices of the complex L.
EXERCISES
EXERCISES
A.
157
HOMOMORPHISMS OF EXACT SEQUENCES.
/:
be a map of the exact lower sequence into another and K' as chain complexes and such sequence Regard both denote their homomorphisms by d and d'. The kernel of / is a subof K'. This complex L of K, and the image of / is a subcomplex
Let
>
1
.
K
K K
r
K
K
U
gives couples
K  K/L 1.
0,
K'
K'/L'
0.
For each index
g,
the various subgroups of
C q (K) form
a lattice
as follows:
Z(K)
= B(K)
C(L)
1 holds with K',L' in place of K,L. that the above lattices of subgroups of C q (K)j C q (K') form l 1 a closed family under the six operations d, d~\ d', d'' /, and f~
2.
3.
Show Show
that Exercise
.
,
4.
For each index
#,
we have
the isomorphisms
H Q+l (K'/L')
where
5.
{/<,}
H
q
(U)
=
H.(K/L)
 H^(L)
is induced by /. Derive from 4 the following proposition due to R. H. Fox: If into another is a homomorphism of the exact sequence {C Q1 Q
g:
K/L
L
f
<t>
}
exact sequence {C^^JJ, then
#;i(image / a )/(image f Q+l )
W (image
(kernel /
fl
4>J+ 2 )
~
_,)
H (kernel ^O/
158
EXERCISES
B. CHAIN HOMOTOPIES. r > be a 1. Let /:
K
K
map
of chain complexes,
and
r
let
).
D =
Show
{D q
}
be a sequence of homomorphisms
D
Q:
C Q (K)
>
C q +i(K
that
defines a
and that D: / g. > K be maps of chain complexes. X" 2. Let /,/': K', 0,0': /' Prove /'. Suppose that g is the homotopy inverse of / and that / that g' is a homotopy inverse of /' if and only if g' g. K > #', 0: K > K''. If any two of the three maps /, 3. Let /: g, and 0/ are homotopy equivalences, so is the third. C. FREE GROUPS AND COMPLEXES. 1. Show that, for abelian groups, a group H is free if and only if
map
0:
K
>
K
f
~
~
~
r
6.2 holds.
2.
If
F is free
and
> (?'
G
>
G"
>
is
exact, then the
same
is
true in the induced sequences
 Hom(F,(?0
3.
>
0>FG'F(g)G>F(g)G"0,
> Hom(F,G) ~> Hom(F,G") >
0.
Let
G'
>
K
G
be a chain complex composed of free abelian groups, and > > G" an exact sequence of groups. Show that the
induced sequences
> >
K
down
G' >
X
G
*
K
G" >
0,
Hom^G'O
^ Hom(/f,G) >
Hom(X,GO
* 0.
are exact and write
represent.
the exact sequences of the couples that they Give an explicit definition of the homomorphisms
H
4.
q
(KG") * H^Kfi'),
their properties.
H
q
(K,G')
> H*+ (K,G")
l
and examine
Let
>L
>
K
+
M
^Obean
>
composed
of free abelian groups
arise sequence of coefficient these sequences in a single diagram. Arrange
and There groups.
(?'
exact sequence of complexes ;/ > G > > G an exact
six
homology sequences. Carry out a similar dis
cussion of the cohomology sequences. 5. Let and be chain complexes composed of free abelian groups > onto Establish the equivalence and let /: be a map of
K K
L
M M
K
M
.
of the following conditions:
(a) /,:
II g (K)
is
=
a
H
Q
(M)
g:
for all
q.
(b)
(c)
If
the kernel of
/,
There
exists
map
M
for all q. then H,(L) = * such that fh = iK and there
K
,
is
EXERCISES
a chain homotopy D: hf D are in N). Here iK and
159
~
i
i
M such that /D
=
(i.e.
the values of
M denote appropriate identity maps.
D. HOMOLOGY GROUPS OF MAPS. L of chain complexes, consider DEFINITION. Given a map /: the chain complex / denned in the proof of 13.3. The groups g (f) may be called the homology groups of the map / and denoted 1. Condider the maps k: C C (L) > C g (f) and I: C Q (J) x. Show that A; is a map k: L > /, denned by ky = (0,i/), l(x,y) = and that J lowers the dimension by 1 and commutes with d. Establish
K
H
the exactness of the sequence
/
fc*
**
where
^
all
is
induced by
/
in the
obvious manner.
E. PROPERTIES OF In
.
the subsequent exercises it is assumed that all groups are in the category Q. The restatements for other categories are left to the
reader.
1.
Establish the natural isomorphisms
A
2.
B ~ B
A,
(AB)C~A(BC)
(?,
for ordinary abelian groups .4,,C.
If
C"
C C and
G'
C
then the natural homomorphisms
(?
C"
I
CG'>CG>C
I
(0/GO
I
(C/C")
arise.
G 1
Show
that
C
G
> (C/C')
(G/0
)
is
onto,
and
its
kernel
is
the group
image
3.
(C'
G) VJ image (C
G').
If all
elements of
CG.
DEFINITION. every integer n
4. If
5.
C
or of
G
are of finite order, the
same
is
true of
If
if, for every g e G and such that ng' = g. C or G is infinitely divisible, then so is C G. all elements of C are of finite order and G is infinitely divisible,
A
group
G
is
infinitely divisible
1
7* 0,
there
is
a g
e
G
then
6.
C
Let
G = 0. G = I~I
G^ be a direct product and p
\
ft
G
>
G$ the pro
160
jection.
EXERCISES
The induced homomorphisms p'p: C (x) G > C (g) Gp are the components of a homomorphism p': C (x) G > Y[ C G0 Show that p' has kernel zero, and show, by an example, that p need not be onto. If C is finitely generated, then p is an isomorphism. If, further, the Op are compact groups, then p' is continuous. 7. State and prove the analogs of 9.5 and 10.5 for direct products is finite, and each C a is finitely instead of direct sums assuming that
r r
M
<t>
generated.
Include the case of compact groups
G>
F. RELATIONS
BETWEEN
(g)
AND Horn.
of the groups
e
DEFINITION. values in a group
A
G
bihomomorphism
is
d
c2 e
and C2 with
a function <(ci,c 2 )
G,
Ci
s
C
I;
C
2
such that
Let Hom(Ci,C2 ;G) denote the group of all such bihomomorphisms. We assume that Ci and C 2 are in the category g while both G and Hom(Ci,C2 ;G) are in one of the categories g/z or g c
.
1.
Establish the following natural isomorphisms:
Hom(C
1
,C 2 ;G)
=:
Hom(d C2 ,G) HomCd, Hom(C 2 ,G)) = Hom(C
a
,
2. 3.
If
K
is
Let
J
C a chain complex, then (K (x) C,G). (K, Hom(C,G)) be the group of integers, and define a homomorphism
H
Q
^
9:
C<?
(x)
by
setting for each c
e
C
(x)
G and
each 0:
C
Prove that ^ is a natural transformation of the appropriate functors. Prove that, if C is a free group on a finite base, then 6 is an isomorphism. 4. Let K be a chain complex such that each of thejgroups C q (K) is a free group on a finite base.__Define a cochain complex^ ='Hom(/f, J). Prove that Hom(# ,G) and Hom(X,G) G, where G is (g) G
K
K
a group in g, g c or g F Consequently the homology [cohomology] of over any coefficient group coincide with those homology groups
,
.
K
[cohomology] groups of K.
G. UNIVERSAL COEFFICIENT THEOREMS. 1. Let F be a free group, R a subgroup of F, and
sider the
G any group.
Con
homomorphisms
i'\
#<g)G>FG,
i":
Hom(F ,G) 
EXERCISES
induced by the inclusion
kernel
i:
161
R C
F.
Show
that the groups
i"
i',
Horn (ft, G) /image
are essentially functions of the groups
H =
F/R and
G.
Denote these
new groups by
2.
Let
K
Tor(//,G) and Ext(.//,G) and examine their properties. be a chain complex composed of free groups. Establish
exact sequences
B
>
(Hint:
a
G>Z
>
a
(g)(7>
H
q
(K\G) >
,_!
GHom(2
>
HomtZ'^G)
Hom^'SG)
> //<(K,G) >
>
a
,G)
*
>
Consider the exact sequence
Z
*
7v
7v/Z
where
z =
3.
{z.(/o,o}.) Using the results of 2 establish exact sequences
a
*
#(#)
G
> #,(K;G) >
13
Tor^.^G)
5
>
 Ext^.^K),^)
and show that the images
> //(A';G) > Hom(// a (:),G) >
of
a and
/?
are direct
summands.
Examine
> K'. the behavior of these exact sequences under maps /: The above results yield the following isomorphisms for the chain
K
complex K:
H
g
(K;G)
 H
==
q
(K)
v
G
H*(K,G)
llom(H (K),G)
+
These are known as the "Universal Coefficient Theorems" since they Q q (K\G) and H (K',G) over any coefficients in terms of H g (K) express and q i(K) with integral coefficients. For more details see 8. Eilenberg and S. MacLane, [Group extensions and homology, Annals of Math. 43 (1942), 757831]; see also H. Tartan and S. Eilenberg, Homological
H
H
Algebra (Princeton University Press) 1950. > K' is a 4. Using the above results show that, if /: free chain complex into another such complex, and if /^:
K
map
of a
H
Q
H
Q (K') and /: and /*: H a (K;G)
H^K)
~ H
(K)
~
9 .,(K'), then
~ H\K',G)
/:
H
q
(K;G)

7/.(K;G)
for
any
coefficient
domain
G.
CHAPTER
Formal homology
In this chapter
VI
theory of simplicial complexes
1.
INTRODUCTION
the formal homology and cohomology This is achieved by associating a chain complex with each simplicial complex and then using the definitions and results of Chapter v. The formalism leading from simplicial to chain
theories of simplicial complexes.
we develop
complexes
definitions point the
The
strongly influenced by the results of Chapter iv. These way to the existence proofs of Chapters vn and ix. homology theory of simplicial complexes is constructed in two
is
ways. The classical procedure attaches to each simplicial complex a chain complex a which we call the alternating chain complex of K. The definition of a is completely motivated by the results of Chapter the ordered chain complex in. The other procedure attaches to each > There is a natural mapping a which induces isomorphisms of their homology groups. The use of is advantageous in proofs of theorems. It is formally simpler since one does not need to general worry about relations in groups and possible degeneracy. On the other hand, the groups C a (K ) are unnecessarily large. Consequently, the chain complex a is used whenever it is necessary to compute the groups of a complex. Since this chapter deals almost exclusively with the formal relations in a complex K, and the underlying space \K\ is not utilized, the assumption that the complex is finite may be dropped. To this end we reproduce a definition already made in the Exercises of Chapter u. DEFINITION 1.1. Let be an infinite set of objects called vertices. A complex with vertices in is a collection of (finite dimensional) simplexes whose vertices are in W, subject to the condition that a face of simplex in the collection is also in the collection. The concepts of "subcomplex" and 'simplicial map" are introduced in the obvious way. The category of pairs of infinite complexes and simplicial maps will be denoted by 3C,. The term "infinite" is always used in the sense of "finite or infinite," so that the finite complexes form a subcategory OCi of 3C t
K
K K
K
K
.
K
K K
K
W
K
W
'
.
2.
THE ORDERED CHAIN COMPLEX OF A SIMPLICIAL COMPLEX
2.1.
DEFINITION
(q
If
K
is
^
0) of vertices of /C, included
A* a simplicial complex, an array A among the vertices of some simplex 162
2]
ORDERED CHAIN COMPLEX
,
163
called an elementary qchain of K. Precisely an elementary qa function which, to each integer i = 0, q, assigns a vertex A* all lie in a simplex of K. The free group A* of such that A, generated by this set of elementary ^chains of (see v,6.5) will be denoted by C Q (K 9 ). By definition, C Q (K ) = for q < 0.
of
K
is
chain
is
,
K
,
K
For each elementary
a o (A
.
<?chain
g
A

A
q
(q
>
0), define
A
)
= Z(1)'A 0
Q
..
A*
A
Q
where the circumflex over a vertex indicates that the vertex
is
omitted.
Having defined d g
for the generators of
C (K
),
a
homomorphism
is
uniquely determined.
If q
g
0,
then d q
=
0,
by
definition.
LEMMA
PROOF.
that
q
<?
2.2.
If q
2.
d a _,d
=
0.
^
fc
1,
then 3 Q .i
Z
=
0,
by
definition.
Suppose therefore
^
It is sufficient to verify that d q .id g (A
A c) =
for
^
2.
Let
^
<
^
^.
Since
the symbol
.4
(l)XiG4
the
tion.
A* A*

A
A
a )
z
A
Q
will
occur in the expression for
with the coefficient
(i)^!)
1

1
and
in
expression
1)*.
(!)'(
l)^ _i(A Hence the two terms
(
for
A
1

A
Q
)
with
the
sign
cancel,
which proves the proposi
from 2.2 that (C Q (K e \d q is a chain complex. This chain If L is a subcomplex of A", then C Q (L ) be denoted by is generated by a subset of the set generating C (/v ), and d Q on L and is a direct Therefore L is a subcomplex of agrees with d Q on L to denote that Given c e C Q (K ) we shall write c summand of
It follows
will
\
complex
K
.
K
.
K
K
.
C
c e
C(L
/L
a ).
DEFINITION
2.3.
For each
K
is
called the ordered chain
)
simplicial pair (K,L), the chain complex complex of the pair (/C,L). The groups
if
C Q (K /L
complex
/.:
Q
are free groups, and
K
r
is
finite,
K /L
is
a
finite
chain
in the sense of v,8.1.
LEMMA 2.4. ///: (K,L) > (K ,L') is simplicial, the homomorphisms C (K ) ~> C.(K'.) and f C Q (L ) > C q (L' ) defined by
g
:
f Q (A
..
A<)
=/(4)

a
map
f
:
K/l,
+ K' /L>
.
164
FORMAL HOMOLOGY THEORY
[CHAP. VI
Moreover, if f:
and, iff:
The
tion #
> (K,L) (K,L) is the identity, then f is the identity, (K,L) > (K',U), g: (K',U) > (K",L"), then (gf) = gJ proof only requires the verification of the commutativity rela.
/c
=
/<,,#
which
The lemma
on the category
states that
3C,
an immediate consequence of the definitions. /L and f a form a covariant functor of simplicial pairs and simplicial maps with values
is
K
in the category dg of chain complexes (recall that g is the category of ordinary abelian groups, i.e. g = Q K where R = the ^integers). We now convert 3C, into a ccategory by defining couples (i,f) to consist of the inclusion maps i: > L > K, j: (K,L) for each pair (K ,L) * iff in 3C. Since i is the inclusion map L it follows that the sequence
K
,
Ov
is
>
T L
>
IT K
>
v
TT K
/L
1
T
>
v
A
exact, thus
observed
couple
(i
a couple on the category dg. Since, as was a direct summand of it follows that the earlier, ,j 6 ) is direct in the sense of v,3,l.
(i"
j L
)
is
is
K
,
Summarizing we have THEOREM 2.5. The pair
the ccategory
K /L J
9
forms a covariant cfunctor
simplicial
on
the category 3C, of simplicial complexes
and
maps
with values in
dg of chain complexes (with only
direct couples considered).
3.
HOMOTOPIES, EXCISIONS, POINTS
DEFINITION
3.1.
if,
Two
simplicial
maps
f,g\
f
called contiguous
for every simplex s of
[of g(\s\) are faces of a single simplex of play the role of homotopy in the category 3C.
and
K [of K
+ (K,L) (K ,L') are the simplexes /() L], This relation will L'].
f
We
fusion with the
use the term "contiguity" instead of homotopy to avoid conhomotopy of the maps f,g: (\K\,\L\) > (/',L') of
the associated topological spaces (when and are finite complexes). Indeed if / and g are contiguous, then they are also homotopic, and
h(djf)
K
K
f
=
(1
0/()
is
+
*0()
f
r
The converse
generally
false:
I^L / and g
a
= may
^ = 1 ls a homotopy. be homotopic without
being contiguous.
// the simplicial maps f,g: contiguous, then the induced chain maps f oj g 9
3.2.
:
THEOREM
(K,L)
*
>
(/',!/) are
are chain
K /L
A

K' /L'
in
homotopic.
PROOF.
Given an elementary chain
c
= An
1
q
K, define
Dc =
q
E (iO
f
1)
/A

fA'gA
gA\
3]
HOMOTOPIES, EXCISIONS, POINTS
165
f
f
This defines homomorphisms
D
Q
:
C (K /L
Q
)
>
C Q+l (K /L
).
It re
mains to verify that
OD
For simplicity,
let
Qc
=
g c
=
fa c
gA*.

D..dc.
B =
l
/A*, C"
f
Then
B*C"
r
'
.i
# E(i)* Z(i)
&
..
c
+1
+ +
(ij'js
%+
'. ..
B''C"
C
Q
+
(1)'
B
.
+l
J?'C"
C'
Z
1
+1
(1)'
#
B'C"
C
1
C"
J
line,
The terms on
for the initial
the middle
when summed on
fl
i,
cancel in pairs except
and end terms:
C
If
...
r
=
g.c,
B<
= / c.
O
the order of summation of the terms on the
l
first
and third
lines are
D a . dc. interchanged, one obtains A second and more conceptual proof of 3.2
of
5.
will
be given at the end
DEFINITION
3.3.
If
K, we denote by K'
C\
K"
K' and K" are subcomplexes of a chain complex K" the subcomplexes of K defined and K'
U
by
C.(K' C\ C Q (K' VJ
C.(K"), K") = C Q (K') K") = C Q (K') \J C (K")
q
where the last group is the least subgroup of C Q (K) containing C g (K') and C q (K") (sometimes written as C g (K ) + C Q (K")). The following lemma is an immediate consequence of the definitions:
f
LEMMA
then
3.4.
//
K'^K"
are subcomplexes of the simplicial complex
K,
(K'
n K").
= K'
C\
K
9
.',
(K' VJ K").
=
K'.
U
K'.'.
DEFINITION 3.5. Let K' and complex K. The inclusion map
i:
K"
be subcomplexes of a simplicial
(K',K' C\ K") * (K
f
U K",K")
1
is
called
an
excision.
3.6.
THEOREM
cision, then
//
i:
(K ,K C\ K") > (K \J K",K")
r
f
is
an
ex
i.:
K'./(K' C\
K"} > (K'
U K")./K'.'
is
an isomorphism.
166
FORMAL HOMOLOGY THEORY
PROOF.
In view of
q
[CHAP. VI
3.4,
the
map
i
consists of the
homomorphisms
V
By
3.6.
C
(K')/C q (K') C\ C,(K")
 C q (K')
V
r
C (K")/C,(K")
V
induced by the inclusion homomorphism C Q (K ) C g (K') C g (K"). the Noether isomorphism theorem z a is an isomorphism. This proves
C
U
THEOREM
vertex, then the
3.7.
// P is a simplicial complex consisting of a single chain complex P is pointlike in the sense of v,4.7.
9
PROOF.
A = d C
Q
q:
= A =
Q
The only elementary
P.
(P
)
where For q even and >0, we have d Q c v = c q _, so that we have C q (P )  0. Thus the C,,(P ). For q <
gchain of
is
P
cq
= A
Aq
conditions of v,4.7 are fulfilled. Having denned the concepts of homotopy, excision, and point in the ccategory 3C t it becomes an /icategory in the sense of iv,9.1.
,
Then, by
iv,9.4,
Theorems
is
2.5, 3.2, 3.6,
and
3.7 are
summarized by
// dg regarded as an hcategory in the sense of direct then 0: is a covariant hfunctor. 3C, couples, dQ DEFINITION 3.9. According to iv,9.5, the composition of the h3.8.
THEOREM
with the homology theory of dQ with coefficient group G (see v,11.3) yields an homology theory on 3C, called the homology theory of with the Likewise the composition of 3C. with coefficient group G. of dg with coefficient group G (see v,l2.3) yields cohomology theory a cohomology theory on 3C, called the cohomology theory of 3C, with
functor
coefficient
group
(?.
cohomology groups H (K /L.;G), H (K /L ]G) (see v,11.4, 12.4) written H (K,L',G) and H (K L;G), respectively, and are called
Q
g
For any simplicial pair (K,L) the homology and
will
be
ih
q
Q
t
the q
homology and cohomology groups of (K,L) with coefficient group G. cording to v,11.3 and 12.3, we have the following cases:
(1)
Ac
(2)
(3)
G G G
s
S*,
g,, Sc,
finite
(K,L) (K,L) (K,L)
e JC.,
s
s
e 3C a , e JC.,
then then then
H.(K,L\G)
e
8,
g*,
H\K,L'G] H\K,L]G)
e
e
gc
;
.
When
that
K K /L
is
a
is
a
finite
complex, i.e. (K,L) e 3Cl, chain complex; hence
we have observed in 2.3 K /L e dg Thus, by
.
v,11.3,
(4)
we have a
fourth case:
G e gc
4.
,
(K,L)
e JCJ,
then
H
q
(K,L'G)
e
gc
.
DIRECT DESCRIPTION OF THE BASIC CONCEPTS
The customary procedure for defining the groups give a direct definition of the chain groups C q (K,L\G) =
H
q
(K,L]G)
is
<g)
to
Cq (K /L
G)
4]
BASIC CONCEPTS
167
and
of the boundary operator, and then to set g equal to the gcycles reduced by the ^boundaries. The procedure we have followed presents Q as the result of composing three functors: 0, G, and the homology group of a chain complex. The advantages of our procedure are twofold: it provides an analysis of the construction, and it enables us to handle a variety of cases with a minimum of repetition. The disad
H
H
vantage
is
that the direct definition has been obscured.
is
of this section
scriptions.
The
The objective to rectify this situation by presenting the direct destatements to be made hardly require any proof since
they are the results of assembling definitions. The group C Q (K,L]G) is by definition the tensor product C Q (K /L ) (x) G. By 2.1, a base for C Q (K ) is provided by the elementary
chains
c's
c
= A
not on L.
will
A\ By 2.3, a base for C (K /L ) is provided by those By v,9.6, C (K,L]G) is generated by the elements c (x) g
Q
Q
which
be written gc
4.1.
in
accordance with the convention v,11.4.
the elements
all
Thus we have
THEOREM
gA

Aq
where g
e
The group C (/C,L;G) is generated by A Q are vertices of /, G, and A,
,
contained
in a simplex of K, with the relations
and
gA
whenever
given by
A =
Q
A,
the
,
A
Q
are contained in a simplex of L.
The boundary
is
formula
d q (gA" .. A')
= X)(1)W iO
the
A'
A'.
C
For a simplicial map f: (K,L) (7v',L') > C g (K',L';G) is given by Q (K,L;G)
f,(gA
If this
chain transformation f Q
:
..
A')
= gf(A)

/(A*).
theorem were adopted as a definition, it would be necessary and 3 Q are compatible with the relations and commute. The remainder of the description is a strict paraphrase of what was done in in, and v,2. The group Z (/v,L;G) of cycles mod L, consists of chains c e C with d Q c 0, and the group B g (K,L]G) of boundaries
to verify that / Q
mod
L,
consists
g
f
of
Q
.
the boundaries of elements
of
C Q+l
.
Finally
7/ (7f,L;G)
= Z /B
)
The homomorphisms / c
associated
with /:
(K,L) > (K',L
carry Z,(K,L;G) and
B
Q
(K,L',G) into
Z
a
(/f',L';G)
and
B
q
(K',L',G)
respectively,
and
thereby
induce
the
homomorphism
168
/*:
FORMAL HOMOLOGY THEORY
[CHAP. VI
H
The direct interpretation of d: (K,L\G) > H<(K',L']G).  Q _,(L;G) is as follows: If h e Q (K,L,G), choose 2 e (K,L;G) Z Q (K,L;G) in the coset (= homology class) A. Write 2 as a formal sum
Q
H
H
H
Then d x
and interpret the resulting expression as an element x is in Z _ ,(!/;(?) and is in the coset dh.
of
C
Q
(K\G).
If (? = J is the group of integers, then stead of C,(K,L;G), etc.
we
write C<j(K,L),
etc., in
C Q (K /L
Passing to cohomology \G) of cochains of
we
K mod L over
A, A )
q
begin with the group C*(K,L\G)
=
the group
Hom(C
a
(7
fl
/L
),(j),
(?. This group is by definition and therefore v,10.6 and v,12 imply
THEOREM
4.1c.
The group
C q (K,L\G)
,
,
is the
defined for each array of vertices simplex of K, the value 0(4,
AQ
(?,
all
group of functions <f> of which are on some
is zero if
is
d
in
q
<f>
and
A,
,
AQ
are in a simplex of L.
The coboundary
is defined
by
For a simplicial map f: (K,L) > (K',L ), the cochain transformation f: C" (#',!/ ;G)  C a (/C,L;6) w given for <f>z C"(#',L';G) by the formula
f
group Z (7v,L;G) of cocycles mod L consists of cochains e C (/C,L;<7) such that 6 a = 0; and the Q group B (K,L]G) of coboundaries mod L consists of coboundaries of Q' Q q elements of C The direct in(K L]GF). Then H*(K,L;G) = Z /B terpretation of 6: H*(L]G) > H*+ (K,L\G) is the following: If u is in
Continuing in the
classical vein, the
a
l
9
.
a
l
H*(L',G), choose
in
Z(L;G) belonging
to the coset w.

Extend
</>
to a
function 0'
e
A,
l
,
A3
C*(K]G) by defining 0'(A, are not on a simplex of L. Then
Its
,
A
<j>'
a
)
arbitrarily
when
and
is
b
q
is
zero on L,
in Z** (K,L;G).
cohomology
class is 5w.
The reason
for adopting the linear
form notation for chains and the
functional notation for cochains appears here for the first time: the formulas for d, 6, / c f q have closed forms. For example, if we were to
,
use the functional notation for chains, then d
is
given by
(a*)(A,
,
A')
=
Z E (DVU
A
0
,
,
A* ,A,A\
l
...
,
A')
which
where the sum on
include
A
q
extends over
all
vertices of all simplexes
A,
,
A
~\
_
Following inj.2 and v,3.2, we could also consider the groups Z 9 (K,L\G) and B Q (K,L',G) which are the counterimages of Z Q and B under the natural mapping C y (K;G) > C(K,L;Or). Thus, by v,3.2, Z a consists of all chains c e C q (K' G) with dc e C a i(L;G), while B,
t
5]
REDUCED GROUPS, ACYCLICITY, MAPPINGS
169
B
Q
H
= ZJB, is isomorphic with VJ C,(L;G). The group Q This alternative description of the relative homology q (K,L]G).
(K',G)
H
groups
is very common, and has the advantage of allowing a much simpler definition of the operator d: a (K,L',G) q ~i(L;G). Most of the time we shall write d and 6 for d Q and 6 C and / for /
H
H
,
and /.
If
P
is
a simplicial complex consisting of a single vertex, then by
4.1
and 4.1c
C
and
d,
(P;(?)

(?,
C(P;G)

G,
=
0,
d
=
0.
Thus
H
and C and
#
= C.
Consequently
This shows that the homology [cohomology] theory on simplicial complexes constructed over a group G actually has the group G as a coefficient group in the sense of i,6.1.
5.
REDUCED GROUPS, ACYCLICITY, ALGEBRAIC MAPPINGS
theories for simplicial complexes
With the homology and cohomology
fully established, we turn to the discussion of some special features of these groups. First we discuss the question of reduced homology and
cohomology groups. These could be defined exactly as in i,7; however a more direct and explicit definition is useful for the applications. Since the space consisting of a single point P is a simplicial complex and for every simplicial complex K the map /: K > P is simplicial, th it would be in accord with 1,7.3 to define the reduced homology group fio(K\G) as the kernel of the induced homomorphism /^: = (K;G) > (P;G). If c g.A is any 0chain in C (K;G), then
H
#
1
/( c )
= (S
Qi)P*
is
DEFINITION
index In(c)
5.1.
This suggests the following definitions: For each element c = ] g,A of
{
CQ (K ;G),
the
0. e
G.
Clearly In:
C (K ,G)
5.2.
is
G
is
a homomorphism, and Ind
If
=
0.
DEFINITION
The augmented ordered chain complex
defined as follows.
simplicial complex K If K 7* 0, then
K
=
0,
then
K of the K = K. = 0.
of integers),
= C (K ) forq* 1, C_!(K ) = J (= additive group d = d< for q 7* 0,1, = In: C (K)  /, do 5.! = o.
C
g
(K.)
Q
a
170
FORMAL HOMOLOGY THEORY
reduced
th
[CHAP. VI
The
homology and cohomology groups
of
K are defined
as
Clearly
H
In the dimension
0,
Q (Ko',G)
=
H
Q
(K
;G)
=
is
#,(#;(?)
for q
>
0.
Z Q (K,G) = Z
(K',G)
a subgroup of
Z (K
;G)
=
C Q (K \G) and consists of the 0chains c with In(c) 0. Hence, by the argument preceding 5.1, we have th THEOREM 5.3. The reduced homology group ft (K;G) is the kernel > // (/v;<7) of the homomorphism / # H^P'jG) where f: K > P and
:
=
P consist
of one point. find that
For cohomology we
H
while
</>
Q
(K.;G)
= H\K
G}
= H\K]G)
.
f or
q
*
#(#;G) = Z?(K\G)/&(Kff), where #(K;G) consists of those coQ chains e C (K;G) which are constant on the vertices of 7t This implies th Q THEOREM 5.3c. The reduced cohomology group fl (K,G) is the factor group ofH(K',G) by the image of the homomorphism /*: H(P,G) > P anrf P consists of one point. H(KjG) where f: K > / which So far we have considered only those maps /: /<"
>
were induced by a simplicial
/:
/C
K'
.
Such maps
satisfy
the
condition In(/c) = In(c) for every c e C (K). In the sequel we shall have occasion to consider maps which are not necessarily induced by
simplicial
maps
(e.g.
the subdivision operator for chains in
f
7 below).
map f:
DEFINITION 5.4. Let K,K be K > K' is a chain map /:
e
simplicial complexes.
An
In(/r)
algebraic
K
>
K' such that
=
In(c)
will
for each c
C
(K).
A
chain homotopy
D
between algebraic maps
/:
K > K' can be extended to a map augmented ordered chain complexes by defining ~ C_,(K ) = Ci(K. /_! to be the identity map of J Similarly an D can be extended by setting D_ = 0. algebraic homotopy DEFINITION 5.5. A function C which to each simplex s of a simplicial K
>
be called an algebraic homotopy. Clearly an algebraic map /:
K'
of the
t
complex K assigns a nonempty subcomplex C(s) of a simplicial complex K' is called a carrier function if, for every face s' of s, C(s') is a subK is an algebraic map such that c e C Q (K) complex of C(s). If /: K and c C s imply fc C C(s) then C is called a carrier of /. Similarly, if D: s f g is an algebraic homotopy such that c e C V (K) and c
f
~
t
C
imply 7>c C C(s), then (7 is called a carrier of DEFINITION 5.6. A simplicial complex
7^).
K
is
called
acyclic
if
5]
REDUCED GROUPS, ACYCLICITY, MAPPINGS
=
if,
171
H
g
(K)
acyclic
and fl<>(K) = 0. for q carrier function for each simplex s, the complex f(,s) is acyclic.
^
A
C
let
is
called
THEOREM
5.7.
Let
K
and K'
be simplicial complexes,
C
be
an
with values in K', and let L be a K' with carrier C can be subcomplex of K. Any algebraic map L 1 > > K' extended to an algebraic map with carrier C. If f,g:
acyclic carrier function defined on
K
K
K
K
maps with carrier C, then any algebraic homotopy between and g\L with carrier C can be extended to an algebraic homotopy f\L between f and g with carrier C.
PROOF. Suppose /: L > K' is an algebraic map. For each vertex which is not in L we select a vertex f(A) of C(A). This extends to an algebraic map /: K VJ L K'. Let A Q A be an elementary / 1 chain of K and let s be the least simplex containing If s is in
are algebraic
A
of
K
l
>
AA
l
.
L, then
f(AA
l
) is
already defined.
1
If s is
l
not in L, then
In(/d(A/l
))
=
In(/A
Since
)

In(/A) =
0.
Thus fd(AA
l
)
e
Z
(C(s)).
(C(s))
1
f(AA
to
l
)
e
d(C(*)) such that
an algebraic
map
>
/:
K
1
1
d/OlM \J L > K'
C
)
.
= 0, there is a chain = /0(AM ). This extends /
!
From
here
we proceed by induction and assume
that an algebraic
map
c
/:
K* VJ L
K
with carrier
of
is
be an elementary
(q
of
L, then fc is already defined. If s is not in containing c. If s = fddc = 0, so that fdc s Z,(C(s)). Since q (C(s)) = L, then dfdc we may choose /c s C Q ^(C(s)) so that d/c  fdc. This extends / to
K
+ l)ehain C
Q *
1
K
already given (q > 0). Let and let s be the least simplex
H
VJ L > /' with carrier C. K' be algebraic maps with carrier C, and let D: f\L g\L be an algebraic homotopy with carrier C. For each vertex A of K which is not in L, we have ln(gA fA) = 0. Thus gA = 0, we may choose D (A) e e Z (C(A)), and since # (C(A)) /A  fA. This extends D to an algebraic C,(C(/1)) so that a/)^l = gA
an algebraic Let /,0:
map
/:
>
K
~
K
homotopy D:
g\K U L with carrier C. f\K VJ L From here we proceed by induction and assume that an algebraic Q~ U L c^ ^/v a1 U L with carrier C is already given homotopy D: f\K
l
~
(q
>
0).
Let
c
containing c. simplex of s is not in L, then define z
dz
K
be an elementary gchain of /C, and let s be the least If s L, then DC is already defined. If = gc Ddc. Clearly z C(s) and fc
C
C
=
fyc

dfc

dDdc = gdc

fdc

(</dc

fdc
+
Dddc)
=
0.
Thus
z e
Z
a
(C()).
Since
H
such that dDc
=
2.
Then
0, there g (C(s)) D^c = gc dZ)c
=
is
a chain
+
fc and
DC e C a+l (C(s)) we find that D is
11*
FORMAL HOMOLOGY THEORY
f\K* \J
[CHAP. VI
cs=:
extended to an algebraic homotopy D:
carrier C.
L
q g\K VJ
L
with
THEOREM
carrier C.
5.8.
Let f,g:
K
>
K
t
f
be algebraic
f
maps with an
acyclic
L Let L,L' be subcomplexes of K,K respectively such that s L f Then the maps f g induce chain maps fo,g<>: KJL. implies C(s) In particular /. = g,+ and ft = g* o/L'. which are chain homotopic. or homology and cohomology groups over any coefficient group.
C
C
.
K
,
g with By 5.7 there exists an algebraic homotopy D: f carries C a (L) Each homomorphism D 9 C 9 (K) C^i(K') into C a +i(Z/) thus inducing a homomorphism D 9q C 9 (K)/C q (L) > C + i(JfiL )/C t+i(^' )' The homomorphisms Z> yield the desired homotopy Z),. DEFINITION 5.9. Given a vertex A of a simplicial complex K, the
PROOF.
~
carrier C.
/
:
:
f
/
star St(A) of A in is the subcomplex of consisting of the simplexes which have A as a vertex, and all their faces. (This is to be distinguished from the open star st(A) defined in n,3.6 which is an open
K
K
subset of \K\)
DEFINITION
define a (q follows: If
5.10.
Given a chain
c e
C^(K) such that
c
C
St(A),
c,
+
l)chain
Ac
e
C Q+ i(K),
called the join of
A
with
as
then
LEMMA
5.11.
// c
s
C
a
(#) and
d(Ac) d(Ac)
c
C
>
St(A),
A(dc)
if
= =
c
c
g q
In(c)A
if
> =
0,
0.
q
PROOF.
Because of
linearity,
we may assume
0,
that c
= A
A
4
is
an elementary gchain.
Then,
)
for q
dAc
= d(AA A = c Adc,
ff
fl
= A


A'
tO
(1)''AA

A
A*
and, for
=
0,
dAc
= dAA = A  A =
For any
vertex
c

In(c)A.
THEOREM
5.12.
A
of a simplicial complex
K,
the sub
complex St(A) is PROOF. Let
acyclic.
z e
Z a (St(A))
for q
>
0.
Then, by
5.11, d(Az)
=
2;
6]
ALTERNATING CHAIN COMPLEX
178
hence z
0,
e B,(St(A)) and so that dAz = z and 2
H (St(A)) = 0.
g
If z e
Z
e
B
(St(A))
Thus
# (St(A)) =
(St(A)), then In(s)
0.

THEOREM
,
5.13.
A
simplex
is acyclic.
This follows from 5.12 and the fact that, if A is any vertex of the then s = St(A). simplex Using 5.8 and 5.13 we can give a second and less computational proof * Let f,g: (K,L) of 3.2. (K',L ) be contiguous simplicial maps. For each simplex s of K, let C(s) be the least simplex of K' containing both fs and gs. By 5.13, C(s) is acyclic. Further if s is in L then C(s) is in L'. Then C is an acyclic carrier for the algebraic maps f.,g..
f
Thus, by
5.8, /.
~
g..
6.
THE ALTERNATING CHAIN COMPLEX OF A SIMPLICIAL COMPLEX
DEFINITION 6.1. Let K be a simplicial complex. A chain c e be called an elementary degenerate chain if it has either of the two l Q A 9 where A = A 1 or c = v + t/ following forms: c = A A q = A A is an elementary chain, and v' is the elementary where v chain obtained from v by interchanging two neighboring vertices 1 = 0,1, A chain which is a linear combi1. q A', A** for some i
will
, ,
nation of elementary degenerate chains will be called degenerate. subgroup of degenerate chains is denoted by Q (K 9 ).
The
D
LEMMA LEMMA
6.2.
6.3.
d,[D 9 (K.)]
//
C
)
D^(K.).
L
is
a subcomplex of K, then
D,(L
= D (KJ
Q
H C.(L.).
then
Q
LEMMA
6.4.
// /:
K
K
f
is simplicial,
f,(D.(K.)]
CD
(K' ).
These lemmas are direct consequences of the definitions. The following three lemmas follow directly from the fact that every permutation is a product of transpositions (i.e. permutations interchanging two
adjacent elements).
LEMMA
and
e
=
6.5. q i* is a permutation of the integers 0, // i , =tl according as this permutation is even or odd, then the chain
,
,
A
is degenerate.
A  cA"
Q

A"
= A
AQ
,
LEMMA
6.6.
// some vertex occurs at least twice in c // the vertices of each qsimpkx s { of
then c is degenerate.
LEMMA
6.7.
K have been ordered
174
FORMAL HOMOLOGY THEORY
way A*
[CHAP. VI
e
in a definite
is
<
c'
<
C q (K
)
a unique qchain
c' e
A', then for each qchain c of the form
C Q (K
)
there
=
o.A?
c'
A\
such that
c
c'
e
D
is
q
(K
).
Moreover
=
Q
if
.
and only
if c
e
D
9
(K
).
If c C. L, where
a subcomplex, then c! The analogy of the last three lemmas with m,5.5,
L
C
5.6,
and
5.8 sug
gests the consideration of the groups
C Q (K a ) = C (K )/D Q (K
Q
).
induces a {C a (/f a ),d a is a homomorphism chain complex. If L is a subcomplex of K, then L a may be regarded, in a natural fashion, as a subcomplex of K a and is then a direct summand
In view of 6.2 the homomorphisms d q
da
:
:
C Q (K
)
CVi(/Q
C q (K a )
>
(?<,!
(/)
so that
X =
a
}
of
K
a
a.
DEFINITION
6.8.
For each
K
/L
a
is
called the alternating chain
simplicial pair (K L) the chain complex complex of the pair (K,L).
9
All the statements
made
in
complexes K
out.
/L replaced by The functor K a /L a J a is
the alternating complexes
now be repeated with the ordered K a /L a throughdenoted by A. Theorem 2.5 and the
2 can
replaced by the functor subsequent discussion apply with the functor A. There are almost no changes to be made in 3. The proof of 3.2 as are simply ordered given in 3 remains valid provided the vertices of are always written with chains of and the elementary (alternating) the vertices in order. The second proof given in 5 carries over without
K
K
modification.
even easier; for, if P is the simplicial = for q ^ 0. complex consisting of a single vertex, then C a (P ) The discussion of 4 carries over to the alternating language with
The
proof of 3.7
is
the following two modifications: In 4.1 the set of relations that the symbols gA be augmented by the following two kinds of relations:

A
q
satisfy should
gA
if
A =
Q
A,

,
AQ
are not
all distinct,
and
to
gA"
if *o,
,
...
A9 =
0,
egA
,

A' f
e is
i q is
a permutation of
q
and
the sign of this per
mutation.
The corresponding
additions in 4.1c are:
<t>(A,
,
4)
=
if
6]
ALTERNATING CHAIN COMPLEX
,
175
l

A,
if z
,
A
,
q
are not
all
distinct
and
<t>(A\
,
,
A")
=
t>(A
,
,
A")
a permutation of 0, q of sign e. The above remarks show that a second collection of homology and cohomology theories on the Acategory 3C, is obtained by replacing the functor O by the functor A. It will be shown below (see 6.9) that
i a is
or the other
these two approaches yield isomorphic theories, so that the use of one is entirely a matter of convenience. The alternating approach is closer to the geometry as is indicated by the alternating char
acter of the chains and cochains obtained in m,5. The alternating is also more convenient for the actual computation of the approach
homology groups,
ordered approach
is
since the groups C Q (K,L;G) are "smaller." The closer to the singular homology theory of Chapter
vn and
is also useful in many problems connected with abstract algebra. In the sequel, unless specific mention is made, either of the two approaches could be used. To compare the ordered and tho alternating theories, for each sim
plicial pair
(/,L), define a
map
:
(1)
K./L.^K./L.
as follows:
Let
be the natural homomorphism. By 6.2, a q and d commute, thus yielding * This map carries L Q into L a and thus induces a map a: K a
K
.
,
the
map (1). THEOREM
6.9.
The map a induces isomorphisms
a*:
H\K
a
/L a \G)
H\K
/L^G)
These isomorphisms yield an isomorphism for any coefficient group G. between the homology (or cohomology) theories derived from the ordered
and
the alternating approach.
We first verify that a^ is a homomorphism of the ordered homology theory into the alternating one. This requires the verification
PROOF:
of commutativity relations in the diagrams
<**
oc*
H
r
Q
(K a /L a ;G)
H
Q
(K /L
'G)
>
H
Q
(K a /La ;G)
*
a*
176
FORMAL HOMOLOGY THEORY
/:
f (K,L) * (K',L )
[CHAP. VI
where
is
simplicial.
These commutativity relations
are consequences of commutativity relations in the diagrams
K /L
a
>
K /L
a
a
L
U
J.
where
i:
L
C K and
j:
KC
(K,L).
yields a
A
similar
argument shows that a*
homomorphism
of co
homology
theories.
The
fact that a+
and a* are isomorphisms
The
is
a consequence of the
a is
following theorem:
THEOREM
PROOF.
vertices of
6.10.
map
a:
K /L
K /L
a
a homotopy equiva
lence (see iv,9.2).
Select a partial order for the vertices of
K
such that the
are simply ordered. A <?chain of will any simplex of A Q with be called normal if it is a linear combination of chains A
K
if
K
A <
< A
q
.
Clearly,
c is
normal, so
is
d Q c.
Given a gchain c e C (/C ), there exists in view of 6.7 a unique normal chain a q c e C Q (K ) such that a Qa Q c = c. Since d qa q c is normal, and a^d^c = d a a a a c = d Q c, it follows that d q a q c = 5 _id a c. Further,
fl
fl fl
if
c
C
I/ (i.e. if c e
:
C
fl
(L.)), then,
by
6,7,
aac
C
L.
Hence the homo
morphisms a q
C (Ka )  C a (/iL.)
q
define a
map
5:
KJL.+K./L.
and aa
$ is
= identity. Now consider the map aa:
K
*.
>
#
.
C is
then aac a simplex of Thus, a common carrier for aa and the identity
,
K
C
C a (K) and c C 5 where = s, we find that setting C(s)
If c e
map
z:
K
i
>
>
K
/L
.
Since
the carrier
C
is
acyclic
by
5.13,
and
since both
aa and
are algebraic
maps,
it
follows from 5.8 that the
map
aa:
K /L
K
is
ho
motopic to the identity. Thus a is a homotopy inverse of a, proving that a: /L C > KJL* is a homotopy equivalence.
K
Note
that, in the
the ordering of the vertices of K.
above proof, the construction of a depended on However the induced homomorphisms
a#
(or a*) are
independent of this ordering since they are the inverses
of a^ (or a*).
7]
EFFECT OF BARYCENTRIC SUBDIVISION
7.
177
EFFECT OF BARYCENTRIC SUBDIVISION
In this section
we
shall
K
be a
finite simplicial
assume that the complexes are finite. Let complex and (1 K Sd K) its barycentric sub,
division (n,6).
We
begin with the construction of an algebraic
Sd:
map
K 
(Sd
K)
and
satisfying the following
called the subdivision operator for chains properties for c e C Q (K):
(1)
(2)
If c
C
s,
where
s is
a simplex of K, then Sd
c
C
Sd
5.
a
Sd
c
= Sd
dc.
for q < 0. Since every vertex of K is also a vertex Define Sd c = Sd K, the group C (K) is a subgroup of C (Sd K). Define Sd: C Q (K) > C (Sd K) to be the inclusion homomorphism. From here proceed by induction and assume that Sd has been defined for dimensions i < q Let c be an elementary (/chain of K, (q ^ 1) so that (1) and (2) hold. and let s be the least simplex of A' such that c C s. Then dc C s and Since b, is a vertex of Sd s, and Sd s C St b,, therefore Sd dc C Sd s.
of
the join operation of 5.10
may
Sd
c
be used to define
=
b.Sd
(dc).
Clearly Sd
c
C
Sd
s.
Further, by 5.11
dSdc =
Sd
(dc)
(dc)
= Sd
 b.dSd (dc)  b.Sd (ddc) =
Sd
(dc).
Having defined Sd for the generators of C C (A) so that (1) and (2) hold, C c (Sd A') is well defined and satisfies the homomorphism Sd: C q (K)
If L is a subcomplex (1) and (2). There results a proper map
of A', then
Sd
carries
L9
into (Sd L) 9
.
Sd:
It
K
/L > (Sd K)./(Sd
L)..
should be noted that, although the subdivision operator for chains
was defined by induction, the procedure is quite definite and could be replaced by a closed formula. Sd K > /v, we shall define simIn addition to the linear map 1 K
\
plicial
maps
TT:
Sd
K
>
K
Sd K,
,
For every vertex b. of called projections, as follows: be one of the vertices of the simplex s of K. If b.
,
let *(b.)
b. t
are vertices
178
of a simplex of
*"(ko)>
FORMAL HOMOLOGY THEORY
"
Sd K, with
s,
[CHAP. VI
,
a face of
s. +
i,
i
=
0,
q
1,
then
' >
s.
of
If
Thus, by extends in a unique fashion to a simplicial map TT: Sd L is a subcomplex of If, then TT carries Sd L into L so that
K(b, q ) are contained in the set of vertices of the simplex into the vertices n,4.4, the vertex map ?r of vertices of Sd
K
K
K
>
K.
TT:
(Sd K, Sd L)

(K,L).
pair (K,L) the subdivision
THEOREM
operator
7.1.
For every
Sd:
(finite) simplicial
K
/L > (Sd X)./(Sd L).
The
is
a homotopy equivalence.
TT O :
map
/,)<,
(Sd ff)./(Sd
TT:

tf
/L.
>
induced by any projection
inverse of Sd.
(Sd
7f,
Sd L)
(K,L)
is
a homotopy
PROOF.
(3)
We
shall establish chain
TT O
homotopies
i.,
(4)
Sd ~ Sd7T ~^
identity
where
(Sd
i
and
j
are
the
chain
maps
of
K
/L
and
7r<,
K) /(Sd L) respectively. First observe that, if c s C q (K) and c s, then Sd c s. Hence C(s) = s defines a carrier C for both Sd c
C
C
TT O
C
Sd s and Sd and i
.
Since the carrier
acyclic, (3) follows from 5.11. For every simplex s of Sd K, let s denote the least simplex of containing all the vertices of s, and let C(s) = Sd s. If c e (^(Sd K)
is
K
and
c
C
s,
then
a
TT^C
C
s
and Sd
TT^C
C
C(s).
Since
s
C
C(,s), it
follows
that C(s)
(the star
is
is is
common
taken in
Sd ir and j. Since C(s) = Sd s = St (s) the complex Sd s), it follows from 5.12 that C(s)
carrier for
acyclic. Thus (4) is a consequence of 5.8. TT Sd c is a degenerate chain. REMARK. It is easily seen that c on the alternating chain complex Hence TT Sd c induces the identity map Ka However this is not true of the operation which Sd TT induces in
.
(Sd
K) a As an immediate consequence of COROLLARY 7.2. A projection
.
7.1
we have
(K,L)
TT:
(Sd K, Sd L)

induces isomorphisms
TT,,:
#
a
TT*:
H\K,L;G) =
(Sd K, Sd L;G) = H,(K,L;G), 8 (Sd K, Sd L;G).
H
TAese isomorphisms are independent of the choice of the projection.
8]
UNIQUENESS
8.
179
UNIQUENESS OF SIMPLICIAL HOMOLOGY THEORIES
be a homology theory defined on the /icategory 3Cj of finite It will be shown how a uniqueness theorem may be established for such simplicial homology
that
all
Let
H
simplicial complexes. analogous to in, 10.1
theories.
First
we observe
the results of Chapter
I
with the exception
11 carry over to simplicial homology theories with the obvious modifications in formulation. Proposition 1,11.5 that a contractible space is homologically trivial has the following analog:
of those of
// the complex
K
is the star (sec 5.9) of
one of
its vertices,
then
K
is
homologically trivial
Indeed,
let
>
K=
K.
St
A
.
Consider the
map /:
K
A and the identity
map
is
H
> fg: to the identity. Hence f+ is an isomorphism //<,(/) contiguous A is homologically trivial, so is K. Q (A), and since
g:
A
Then
A
>
A
is
the identity while gf:
K
K
=
With
this done, all the propositions of in, 2
to simplicial
The
first
through in,9 carry over any change except in notation. homology of the proof of the Uniqueness theorem in, 10.1 (the conpart
theories without
struction of the
homomorphism
THEOREM
8.1.
f) yields the following theorem for si?nplicial homology theories). (Uniqueness
on the hcategory and Given two simplicial homology theories and given a homomorphism, finite simplicial complexes,
h
:
H
H
3Cj
of
G+G
homomorphism
3C:
of their coefficient groups, there exists a unique
h:
7/^77
.
on
which
is
When h is an isomorphism, so also is h. is an extension of h Thus, as far as simplicial homology theories are concerned, the theory complete: both existence and uniqueness are established. This theorem casts light on the structure of the proof of the Unique
on JCO by setting H ttQ (K,L) = H q (\K\,\L\). The axioms for //. follow directly from the corresponding axioms for // except for the Excision axiom which is the Theorem iu,2.2. _ G > G is If //,// are two homology theories (defined on 3) and h Q
:
Each homology theory ness theorem of in, 10.1. (in the sense of Chapter i) defined on an admissible category ft containing the category 3 of triangulable pairs, leads to a simplicial homology theory H. (defined
H
a homomorphism of their coefficient groups, then 8.1 yields a unique extension of h to a homomorphism of the associated simplicial theories;
h,:
H.+H.
on
X'.,
180
FORMAL HOMOLOGY THEORY
[CHAP. VI
The second part of the proof of in, 10.1 then shows how one can pass from the homomorphism h, to a homomorphism
h:
H *~H
on
3.
This
is
makes use
the only stage in the proof of the Uniqueness theorem which of the Simplicial approximation theorem (n,7).
An
are
left
analogous discussion can be given for cohomology; the details to the reader.
NOTES
The
classical
invariant theorem.
We
shall give here a
f
modernized
version of the classical in variance theorem.
Let (X,A) be a triangulable pair. If T,T are triangulations of (X,A), we shall write T < T if, for each vertex B' of (or of A) in the triangulation T', there is a vertex B of X*(or of A) in the triangulation T such that
f
X
st(B')
C
st(B).
We
note the following formal properties: T' < T and T" < T' imply T" < T. th is the n (2) "T < T where barycentric subdivision of
(1)
T
T
For any T and T ', there exists an integer n such that < T (follows from n,6.5). (4) For any 5P,,T a> there is a T with T < T ,, T < T 2 (follows from (2) and (3)). < T. If, for each vertex B of T', we select a vertex Suppose T' * T called a p(B') = B as above, we obtain a simplicial map p: two projections are contiguous, and therefore, by projection. Any 3.2 and v,4.4, they induce the same homomorphism
7
f
(3)
T
(n
^
0).
7
T
f
a(T,T'):
H
Q
(T') >
H
q
(T).
Here
H
9
(T),
where
T =
the homology group
H
q
\t,(K,L)} is a triangulation of (X,A), denotes (K,L) in either the ordered or the alternating
approach.
If
T"
If
< T < T
1
and
also
p:
T
1
> T7
,
p';
T" >
T' are projections,
then pp'\
(5)
T" T"
*
T
f
is
<T
<T,
is
a projection. This yields: then a(T,T')a(T',T") = a(T,T").
The
(6)
crucial fact
the following:
1
< T. a(T,T') is an isomorphism for any In the special case when T f = T is the barycentric subdivision of T, = T. Thus (5) implies that (6) is valid for (6) follows from 7.2.
T
T
NOTES
For any T'
181
<
n
T, choose
n so that
T
<
T'
<
T.
Then a(T,T
f
n
)
=
iso
a(T,T')a(T',T ) is an isomorphism. It follows that a(T,T n Therefore a(T',T ) is also onto. This implies that ot(T,T')
)
is
onto.
is
an
morphism as
Let
desired.
l9
now T T 2 be any two
triangulations of (X,A).
Choose
T
with
T <
T,,
T < T and
2
define
If
T'
<
T, then
This fact, together with (4), implies that /3 r (T 2 ,T ) is independent of the choice of T. We thus obtain unique isomorphisms
1
(7)
0(T a ,7\):
tf^O =
//,(r a ).
1
l
In particular
(8)
,T ) if 7 T with T triangulations, then, choosing
2
}
0(?VA) = a(T
0(r 3 ,7\)
KTMKTM.
H
g
< TV If T^TV^ < T, (i = 1,2,3), we
are three
find
It follows
that the groups
(T) together with the isomorphisms (7)
form a transitive system
groups H g (T) denoted by g (X,A).
H
of groups as defined in 1,6. In this sense, the 1 are independent of the triangulation 7 and may be
,
The
and
classical in variance proof limited itself to
showing that
H
g
(Ti)
H
Q
(T 2 )
were isomorphic.
It
did not exhibit a "unique" isomorphism
(7) satisfying (8).
Having constructed a group 7/ g (AVl), we could continue in the same vein and construct d and f+, and then verify the axioms. This would
of triangulable spaces.
provide a proof of the existence of a homology theory on the category It is not worthwhile to do this in view of sub
sequent existence proofs for
much
larger categories.
of
The development of the invar iance theorem. Much of the development homology theory during the period 18951925 centered around the question of the topological invariance of the homology groups. The first fully satisfactory solution was achieved by Alexander near the end of this period through the development of simplicial complexes and their techniques [Trans. Amer. Math. Soc. 28 (1926), 301329]. Prior to this a broader concept of cell complex was used (see Chapter xni of the second volume) Homology groups were defined by choosing
.
a cellular decomposition of the space, orienting the cells (i.e. selecting incidence numbers the coefficients in the boundary relations), and
taking the homology groups of the resulting chain complex.
This pro
182
NOTES
rise to
cedure gave
cell
a series of problems:
(1)
a clearcut definition of
complex was needed; (2) the orientability must be proved (i.e. show that dd = 0); (3) the independence of the homology groups on the choice of orientation must be shown; and (4) the homology groups must be proved independent of the choice of the cellular decomposition. The
last is the invariance
By
came
problem. using too broad a definition of complex, the second problem betoo difficult. Problems (1), (2), and (3) were solved by using
polyhedra for complexes. A polyhedron is a collection of cells in a euclidean space, each cell is defined by a system of linear equalities and A step in solving (4) was the proof of invariance under inequalities.
subdivision.
This led to the Hauptvermutung: // two polyhedra are homeomorphic then they have isomorphic subdivisions. This plus invariance under subdivision would imply topological invariance. But it remains unproved to this day.
The wording
of the invariance
problem
is
such that one
feels
con
strained to use only topological maps in its solution. However the are invariant under homotopy equivalences. It was homology groups
perhaps the realization of this fact which led Alexander to abandon the
refined techniques of topological equivalence, and to replace them by the rough techniques of simplicial approximation and deformation, and
thereby achieve a satisfactory solution.
Although simplicial complexes appear to eliminate most theoretical difficulties, they are highly inefficient for the computation of homology
triangulation of a space usually demands a large number of simplexes. As noted in in, Exer. D6, a triangulation of a torus requires at least 42 simplexes. 1 The nsimplex has 2 n
groups of simple spaces.
A
faces.
(see
Polyhedral and cellular decompositions are far more Chapters xin and xiv of the second volume).
efficient
The Mayer homology groups. Let be a simplicial complex. Consider the free abelian group n (K) whose generators are sequences n Q A n are in a simplex (A, ) of vertices of K, such that ,
K
M
,
A
A


,
such sequences are regarded as equal if they differ only by > a permutation of coordinates. Define the homomorphism F: n (K)
of
'.
F
Two
M
Mn^K) by
setting
F(A\
Let p be a prime and
all
,
A")
=
t0
n
(A
..
,
,
i',
..
,
A").
g s G).
Define
n
morphism F:
M (K\G) = M (K) G and consider the homoM (K;G) M ..i(K;G) induced by the operator F above.
n
n
G
an abelian group with pG
(x)
>
=
(i.e.
pg
for
n
NOTES
Then the
iteration
183
for each integer
Fv
of
F
is
zero.
Thus
p(l
<
q
<
p
the composition
PVQ
M
is
n
<
p^Q
>
M
n
>
M
n_ q
zero.
q
The Mayer homology group
H
n>Q
(K]G)
is
defined to be (kernel
These groups were defined by W. Mayer [A new ). homology theory i, n, Annals of Math. 43 (1942) 370380, 594605]. He established a large number of their properties, but was unable to determine their relation with the ordinary homology groups. This was done by E. Spanier [The Mayer hornoloyy theory, Bulletin A. M.S. 55 (1949) 102112], making essential use of our Uniqueness theorem vi,8.1. In this order, Spanier defines (1) relative groups 7/ n a (/v,L;(7) where
)f (image F""
F
L
II n
a subcomplex of 7v, (2) a homomorphism /^: // niQ (/i,L;G) each simplicial map /: (K,L) ^ (K',L'), and (3) a Q > 7/ _ He then shows n Q (K,L',G) n boundary operator F .,,_ (L;G). that the groups // n a suitably reindexed, provide several homology
is
,
q
(K',L'fi) for
:
H
,
,
theories on the /icategory of simplicial complexes. The coefficient or G depending on arithgroup of each theory is shown to be either metic properties of the indices involved. Thus the Uniqueness theorem
implies that II n
.
Q
(K,L',G)
r
is
either
homology group
H
H
n>q (KJj\G)
~
1
1
(KJj',G] for a suitable
or isomorphic with the ordinary r. The precise result is that
// r (/v,L;(7) in the following
two
cases:
n n
In
all
+ +
= =
q
(mod (mod
is
p), p),
and and
r r
= =
2(n 2(n
+ +
1
</)/p,
l)/p

1.
other cases 7/ n
zero.
EXERCISES
A. SlMPLlCI\L APPROXIMATIONS.
KI arc both sim})licial approximations to a map 2 are contiguous. /: 0! and \K\ \Ki\, Sd K > K (in the sense of 7) is a 2. Show that a projection TT: K involved in to the linear map 1 K Sd K simplicial approximation
1.
If 0i,02>
K~
>
then
:
the definition of subdivision. B. LOCALLY FINITE COMPLEXES.
DEFINITION. Let K be a (possibly infinite) simplicial complex. A" in C Q (K ), define an integral coGiven an elementary chain .4 A 9 )  1 and </>(r) = for all chain e C"(K ;J) so that 0(.4,
</>
,
The cochains other elementary chains in ( w (/v ). thus obtained Q Q generate a free subgroup of C (K ;J) denoted by C (K) and called the
T
rt
</>
group of
finite integral
cochains of K.
184
1.
EXERCISES
Show
that
K
is
locally finite
if
and only
if
every finite cochain is again a cochain complex by K.
2.
finite cochain.
the coboundary of Denote the resulting
A
simplicial
map
/:
l
K
>
K
l
of a locally finite
complex
K
into
is called locally finite if, for every finite subanother such complex l complex L of #,, J~ (L) is a finite subcomplex of K. Show that a locally K. finite simplicial map induces a map f: Kl 3. Let L be a subcomplex of the locally finite complex K, and let
i:
K
L
*
K,
j:
K
>
(K ,L) be
i:
inclusion maps.
>
Denote by
K
LI
L
>
the kernel of the
plicial
map
K
l9
L.
Show
that a locally finite simpairs of
K
<
(K Li) induces a map /*: Kl map /: CK,L) L. Show that with suitable definitions the category of
and locally finite maps constitutes an ^category and that the pair K L,f then yields a contravariant Munctor 0' on with values in the category of 6Q of cochain complexes (direct
locally finite complexes
couples).
Use the Afunctor 0' to define homology and cohomology theories q (K,L;G). Give a detailed description of these theories to that of 3. What are the limitations on the coefficient analogous G? group 5. Adopt the alternating approach and define a functor A' analogous to 0'. Show that the homology and cohomology theories obtained using 0' and A are isomorphic. 6. Compare the and 3C theories on finite complexes. Correlate
4.
3C g (jfiL,L;G), 3
1
H
the result
7.
with
Show
v, Exer. F4. that, in the 3C theories, the reduced groups are defined only
for finite complexes.
The homology groups a of complex based on the chain complexes K (or K n ) will be called the direct homology groups, while the corresponding cohomology Q will be called the inverse cohomology groups. In a locally groups a finite complex the cohomology groups 3C based on the chain complexes K (or Ka ) will be called the direct cohomology groups while the corresponding homology groups 3C will be called the inverse homology groups.
an
infinite
REMARKS CONCERNING TERMINOLOGY.
H
H
fl
The The
reasons for this terminology will become clear in Exercises vm,F. direct homology and cohomology groups are not defined for comIf
pact coefficients.
the linear form notation were used for
all
chains and
cochains, then the direct groups would be based on finite linear forms while the inverse groups would utilize infinite linear forms. For reasons
explained in the remark following 4.1c, it is convenient to use the linear form notation for chains and the functional notation for cochains, regardless of whether the chain or cochain is finite or infinite.
CHAPTER
VII
theory
The singular homology
1.
INTRODUCTION
The
mology and cohomology
objective of this chapter is to establish the existence of hotheories on the largest admissible category,
the category GI of all pairs (X,A) and all maps of such pairs. namely: For homology [cohomology] theory, the existence is established for any prescribed coefficient group in any of the categories R [Q or g c ], and A singular homology the theory has values in the same category. with values in g c is not constructed. theory The method is the following: Using mappings of ordered simplexes into the space X, a chain complex S(X) is obtained, called the singular complex of the space X. Then the homology and cohomology groups of a pair (X,A) over any coefficient group are defined as the appropriate groups of the complex 8(X)/S(A), using the methods of Chapter v. The verifications of all the axioms, save for Homotopy and Excision, are trivial consequences of the corresponding theorems about chain
ft
complexes. To prove the Homotopy axiom, a chain homotopy from the given homotopy.
it is
The
necessary to construct proof of the Excision
axiom requires some preliminary construction.
2.
THE SINGULAR COMPLEX OF A SPACE
X Q ) consider space with coordinates (j q d are the unit the unit simplex A< (see n,2.2). The vertices rf, q * 9 If we regard R as the subset points on the coordinate axes of R Q+l of R given by x g = 0, then A _, is a face of A< and has vertices d,
Tn the euclidean
,
R
Q +
l
,
,
1
.
fl
,
d'
1
.
In
A the bary centric
fl
coordinates of a point coincide \vith
its
cartesian coordinates.
Consider the simplicial maps
el:
A,.! >
A
,
i
=
0,
,
q,
given by the following vortex assignments:
rW) = rX) =
185
d>
1
if
j
i
<
i,
"
1
rf
^
^J
186
SINGULAR HOMOLOGY THEORY
simplicially onto the (q
l)face of
[CHAP. VII
Thus e*'maps A a _i
A a not conj
taining d* and preserves the order of the vertices.
LEMMA
The
2.1.
is
eX_i
2.2.
=
e\e\1\
for
g
<
i
^
q.
proof
obvious.
DEFINITION
A
continuous mapping
T:
A a >
X X
is
of the unit ^simplex A c into a topological space l)simplex qsimplex in X. The singular (</
called a singular
is
called the
th
i
/ace of T, for
t
=
0,
,
q.
By 2.1, LEMMA
we have
2.3.
(T
2.4.
(0
)
(J)
= (r>) u ~
l >
for
<>
j
<
i
^
q.
group generated by the singular is denoted by C Q (X) and is called the group of singular in ^simplexes If q < 0, there are no <?simplexes and (integral) qchains of X.
free abelian
DEFINITION
The
X
C
Q
(X)
=
0.
The boundary homomorphism
dQ
is
:
C
a
(X) >
0,
C
g
defined as follows:
If q
g
then d q
=
0.
If
(7
>
and
7
1
is
a
singular (/simplex in X, then
LEMMA
PROOF.
that q
2.5.
If q
d _,a a
=
0.
<
1,
then
#<_!
=
by
definition.
Assume
therefore
of
^
2.
It is sufficient to verify 2.5 for
each generator
T
(0
C a (.Y).
y
By
definition
a.i,r= ZCiyaoiZ*
4
=
Z S (i) (i) (r
t
J
(J)
)
i
In view of
2.3,
the
first
sum
equals
1
OSJ<Sa
Z
(i)**'^ )'"
i
'
Replacing
i
1
by
y
and y by
gives
Thus the two sums
cancel,
0^t^;<</
Z
(i)
i+>
(l)
(>)
(y
is
)
.
and the lemma
proved,
2]
It follows
will
SINGULAR COMPLEX Of A SPACE
]
187
from 2.5 that \C Q (X),d Q is a chain complex. This chain be denoted by 8(X). If A is a subspace of X, then C g (A) is generated by a subset of the set generating C Q (X) and d q on C q (A) agrees with 3 Q on C Q (X). Therefore S(A) is a subcomplex of S(X) and it is a direct summarid of S(X). A Given c e C(X) we shall write c
complex
C
to indicate c
e
C,CA).
DEFINITION 2.6. For each pair (X,A) consisting of a topological space X and a subset A, the chain complex S(X)/S(A) is called the The groups C q (S(X)/S(A)) = singular complex of the pair (X,A). C g (X)/C q (A) are free groups. REMARK. Strictly speaking S(A) is not a subcomplex of S(X), because a generator of S(A) is a map T: A, > A and not a map A a > X. The inclusion map i: A C X induces an isomorphism of S(A) with a direct summand of *S(X), and S(X)/S(A) should be written S(X)/iS(A). LEMMA 2.7. // /: (X,A) (F,B), ffce homomorphisms / > C.(F C,(X)
:
for T:
A q >X(orT:
/.:
& g + A)
,
yields a
map
S(X)/S(A)
the identity,
Moreover, if
and, iff:
> (X,A) (X,.4) is //te identify, then f s is * (F,JB), ^: (F,)  (Z,C), //ien (^/) s (X,A)
f:
proof requires only the verification of the commutativity relation 6 Q f q = f q id q which is an immediate consequence of the relation
(0 (/TV = JTe\  /(T ). The lemma states that S(X)/S(A) and /
The
=^ f /,.
tS
form a covariant functor
the category fti of pairs of topological spaces with values in the category #9 of chain complexes (recall that g is the category of ordinary abelian groups, i.e. S = 9 where R = the integers). We now convert
d, into a ccategory by defining couples
(i,j)
S on
to consist of the inclusion
maps
i:
A
>
X,
j:
X
is is the inclusion
map
(X,A) for each pair (X,A) in a^ Since S(A) > S(X), it follows that the sequence
is
js
> S(A) * S(X) > S(X)/S(4) >
a couple in the category d Since, as was a direct summand of S(X), it follows that the couple (isjs) is direct in the sense of v,3.1. Thus S is a cfunctor. We now go a step further. The ccategory dg has been converted in v,4 into an /icategory. We shall do the same with the category a,. To this end we define homotopies and points in a to be actual homois
exact, thus (isjs)
earlier,
is
observed
S(A)
is
x
188
topics
SINGULAR HOMOLOGY THEORY
and actual
(X' ,A')
(2)
[CHAP. VII
map
map,
/:
XA =
=
points, while excisions in Cfci are defined as follows: > (X,A) is called an excision if (1) / is an inclusion
A
X'

A',
and
(3)
the closure of
XX
1
lies in
the
interior of A.
The
last
two conditions may be replaced by the more
symmetric conditions
A'
X' C\
A
y
X=
Int
(X
f
)
U Int
(A).
i,3 in
Note that
this
U = X
X'
is
concept of excision is broader than that of not required to be open.
that
With these
definitions
^ becomes
is treated
an ^category, and we can state
the main result of this chapter.
THEOREM
2.8.
// dg
as
an hcategory in
is
the sense of direct
couples, then the functor S:
a covariant hfunctor (see iv,9.4). &i dg To prove the It has already been shown that S is a cfunctor. theorem we must show that S preserves homotopies, generalized exciThis will be done in Theorems 4.7, 7.1, and 9.1. sions, and points. Granting that this has been done, we apply iv,9.5 to obtain our chief
objective:
DEFINITION 2.9. The composition of the /ifunctor S with the homology theory of dg with coefficient group G (see v,11.3) yields a homology theory on QI called the singular homology theory of Ct with The composition of S with the cohomology theory coefficient group G. of dg with coefficient group G (see v,12.3) yields a cohomology theory on GI called the singular cohomology theory of &! with coefficient group G. For any given pair (X,A) e G^, the homology and cohomology groups
v
H H
(S(X)/S(A)}G), (S(X)/S(A),G) (see v,11.4, 12.4) will be written ih Q singular Q (X,A]G), (X,A',G), respectively, and are called the q homology and cohomology groups of (X,A) with coefficient group G. According to v,11.3 and 12.3 we have the following cases:
q
H
Q
H
(1)
Gs
g*,
(2) (3)
Geg*, Gegc,
then then then
H (X,A;G) H (X,A>G)
g Q
e
e e
g,
g*,
H'(X,A;G)
gc
.
Observe that singular homology groups q (X,A]G) are not defined is compact. The reason for this is that C Q (X) usually does not have a finite base (see v,11.3).
H
when G
note here the following alternative definition of the singular < A*. simplexes. Let s be an ordered (/simplex with vertices A < A map T: \s\ will be called a singular ^simplex in (in the > < B q and T new sense). If s' has vertices B < X, then \s'\
We
X
X
1
:
* s' is that T and T' are equivalent if T = T'B where B: s B i = 0, In this definition, the simplicial map with B(A ) q. the group C Q (X) is generated not by the individual singular #simplexes,
we say
l
l
,
,
3]
BASIC CONCEPTS
189
but by the equivalence classes of such gsimplexes. Clearly every such equivalence class contains precisely one singular gsimplex in the sense
of 2.2.
3.
DIRECT DESCRIPTION OF THE BASIC CONCEPTS
vi.
This section serves the same purpose as Section 4 of Chapter
A
direct description of the singular
homology and cohomology theories
will
be given. The opening remarks of vi,4 apply equally well here. The group C Q (X,A',G) of singular ^chains of X mod A is by defini
tion the tensor product [C Q (X) / C Q (A)] G. > A form a base for C Q (X). T: simplexes
X
generated by the elements T g which will with the convention v,11.4. Thus we have
By 2.4, the singular By v,9.6, C (X,A]G) is be written gT in accordance
Q
THEOREM
where g
s
3.1.
The group
A<,
C (X,A\G)
Q
is
generated by the elements
gT
y
G and
T:
>
X is a singular qsimplex in X.
<7,r
These generators
are subject to the relations
+
g2 T
and
gT =
The boundary
is given
for
r(A a )
C
A.
by the formula
For
> (Y,H), any map f: (X,A)  C a (Y,B;G) is given by C.(X,A;G)
the
chain
transformation
/,:
f,(T) = g(fT).
to verify that J Q
theorem were adopted as a definition, it would be necessary and d are compatible with the relations and commute. The descriptions of Z q (X,A;G), B,(X,A;G), H q (X A' G)J^ and d+ are as in vi,4 with (K L) replaced by (X,A).
If this
9 9 9
t
= Passing to cohomology we begin with the group C (X,A\G) q This group is by of A" mod A over G. C (S(X)/S(A)\G) of cochains definition the group Hom(C Q (X)/C Q (A),G), and therefore v,10 and v,12
Q
imply
THEOREM
3.1c.
The group
defined on singular qsimplexes Q The coboundary 8 <t> zero if T is in A.
C (X,A;G) is the T in X, the value
is
9
group of functions
<f>(T)
<t>
is
in G,
and
is
defined by
190
SINGULAR HOMOLOGY THEORY
any
[CHAP. VII
transformation
*Ae
For
/":
C
f
(X,A) (F,B;G) > C*(X,A$F)
f:
map
is
(Y,B),
given for
<f>
the
e
cochain
C'(F,#;G) by
/ormufa
The
descriptions of
Z q (X.A;G), B a (X,A;G),
H
Q
(X,A;G), /*, and
6* are as in vi,4
with (K,L) replaced by (X,A).
4.
PRESERVATION OF POINTS, REDUCED GROUPS
4.1.
THEOREM
PROOF.
If
P
is
a space consisting of a single point, then the
in the sense of v,4.7.
chain complex S(P)
is pointlike
T
Q
:
:
dQ
For each q ^ there is only one singular gsimplex A g > P. For q even and > 0, we have dT Q = so that C Q ^(X). For q < 0, we have C 9 (X) = 0. Thus condiC Q (X)
T^
tions of v,4.7 are fulfilled. Observe further that by 3.1
and
G,
3.1c
we have
C
and
a,
(P;(?)
7/
=
0,
d
=
0.
Thus
= C
G,
and
7/
= C.
Consequently

This shows that the singular homology and cohomolpgy theories constructed over a group G actually have the group G as coefficient group
in the sense of i,6.1.
> Observe that, for any space X, a singular 0simplex T: A determined by the image point x = T(A ). If we agree completely to write x instead of T, then every 0chain c e Co(X,G) will have the form of a finite formal sum
X
is
c
=
S ^^
e
where
</>:
g<
>
e
G
?
,
o: t
e ^T.
Every cochain
C(X,G)
will
be a
the
unction
X
G
(with no continuity assumption).
In analogy with vi,5.1
definitions:
and
vi,5.2,
we introduce
c
following
DEFINITION
index In(c)
is
4.2.
For each element
= X) ^*
G.
of
CQ (X,G),
the
In(c)
=
Z^
e
Clearly In:
C
(X;G) ~> G.
4.3.
DEFINITION
The augmented
singular complex
S(X)
of the space
5]
LINEAR COMPLEX OF A CONVEX SET
defined as follows:
If
191
If
X
is
X=
0,
then S(X)
= S(X) =
0.
X^
0,
then
C
= C (X) C_i(5(X)) = J = d 5, C (S(X)) do = In: 3! = 0.
f
(SCY))
Q
forq* 1,
(=
additive group of integers),
for 2
q
^
0,1,
J,
The
reduced
th
homology and cohomology groups
of
X
over
G
are
defined as
/7
()
(X;G)
=
H
Q
(S(X);G),
B(X;G) = H(S(X);G).
Clearly
H
Q
(S(X)]G)
=
H
q
(X\G] for ?
^
0.
In the dimension
0,
Z
(X;G)
sists
(S(X);G) is a subgroup of Z (X;G) of 0chains c with In(c) = 0. This implies
4.4.
= Z
= C(X;G) and
is the
con
THEOREM
of the
The reduced
homomorphism /:
H
Q
homology group fi Q (X\G) Q (P\G) where f: (X\G) *
th
kernel
H
X P
and
P
is
a spacf? consisting of one, point. Therefore the concept of reduced group coincides with that defined
in i,7 for
any homology theory.
q
For cohomology, we find that H'(S(X);G) = H (X;G) for q ^ 0, and that #(X;G)  Z(A";G)/B(A';G), where B\X\G) = B(S(X);G) e C(X;G) which are constant on the singular consists of those cochains the points) of A". This implies 0sirnplexes (i.e. THEOREM 4.4c. The reduced O tb cohomology group //(X;G) is the
factor group
ofH(X;G)
/:
H(X;G) where
The
by the image of the homomorphism /*: 7/(P;G) > P a/zrf P zs a A' 6>'pacc consisting of one point.
>
definition of "acyclic space"
and "algebraic map
carrier of /
is
jS(F)" are as in vi,5.6 and vi,5.4.
to each singular simplex
A
S(X) > a function C which
/:
T
)
in
X
assigns a subset C(T) in
Y
such that
f(T)
C
C(T) and C(7
T(0
C
0(7').
The statement and
proof of vi,5.7
carry over with only formal changes.
6.
THE LINEAR COMPLEX OF A CONVEX SET
in
We
functor
7
proceed to develop the tools to be used
S
preserves homotopies
5.1.
and generalized
the proofs that the excisions (given in
and 9). DEFINITION
Let
V
be a convex set in a euclidean space.
A
192
SINGULAR HOMOLOGY THEORY
Aa
>
[CHAP. VII
1
singular simplex T:
V
is
said to be linear,
if
a
,a 2 e
Aa
,
Xi
^
0,
X2
and X
t
+
X2
=
1
imply
X 2 a 2)
=
The The
i
v v
=
form a sub complex Q(F) of S(V). linear simplex T is completely determined by the points = 0, and we shall therefore use the symbol T(d*), i 7 v to designate the linear simplex T. With this notation we have
linear simplexes
,
,
DEFINITION
join vT
is
5.2.
If v e
V and T =
v


V
a
.
Q
is
a linear gsimplex, the
v fixed, this
the linear (q l)simplex vv tion on generators extends uniquely to a
+
With
opera>
homomorphism
c
we of
C
a (Q(F)) into
C^CQCF)).
LEMMA
5.3.
IfveVandcz C (Q(V)), then = c v(dc) d(vc)
Q
if q
d(vc)
=
c
ln(c)v
if q
> =
0,
0.
The
proof
is
the same as that of vi,5.11.
6.
PRISMS
The
unit qprism (q
>
0)
is
7T a
the cartesian product
=
is
/
X
A,,,
where A a _
t
t
is
the unit (q
l)simplex in
The prism W Q g ^ Consider the maps
1 in
.
R
l
and I is the closed interval +l Q a convex subset of R X R = R"
,
R
Q
l
.
rq
:
7r a _,

7r fl;
i
=
0,
,
g

1
defined for
(J,t>)
e ir q i
by
and the maps
1Q
:
A a .i
> TT O ,
^g
:
A a _!
>
TT O
defined for v
e
A a _! by
As an immediate consequence
of these definitions
for
,
we have
LEMMA
6.1.
rXi
= r Xi
for i
^y < = 0,
t
^
,
g
q

1,
1
.
6]
PRISMS
6.2.
I
OS
DEFINITION
A
continuous mapping
P:
WQ

X
of the unit (/prism TT, into the topological space The singular (q 0). qprism in (q l)prism
X
is
called a singular
X
>
P (0 =
is called the has no faces.
tb
Prl:
0,
*,_!
>
,
X
1.
z
/ace of
P
t
for
i
q
A
t
singular 1prism
The
singular (q
l)simplexes
P,
=
PI.:
A,_
* X,
Pu =
Pii.:
Aa _ 
X
are called the lower and the upper base of P, respectively.
By 6.1, we have LEMMA 6.3. (p '>) (l) = (p<'>)<(0 (0 (P )u = (Pu) (P ), = (P,)"',
( (
for
,
^
j
<
0,

t

for i
=
g

g q

1,
1,
,
and we may write Pi,P{ without ambiguity. DEFINITION 6.4. The free group generated by the singular <?simplexes and the singular (/prisms in X is denoted by C Q (SP(X)) and is called the group of singular (integral) prismatic qchains in X. If q < 0, If q then C 9 (SP(X)) = 0. 0, there are no singular 0prisms so
f)
CQ (SP(X) = C
(X).
The boundary homomorphism
is
defined as follows:
If q
g
0,
then 3,
=
0.
If
q
>
and
P is a singular
gprism, then
d<t
p = pu _
T
p,
t0
(l)'P'".
For a singular
(/simplex
LEMMA
PROOF.
6.5.
dQ ldq
=
0.
Clearly singular #prism P.
definition
it is
sufficient to verify that d Q _id Q
P =
for every
Since d
=
0,
we may assume
that q
>
1.
By

P,

E (1)^
tO
(i)'p:
pi
10
194
SINGULAR HOMOLOGY THEORY
single
is
[CHAP. VII
6.3.
The sum
first
zero
sums cancel because of the second part of by the same calculation as in the proof
6.6.
The double
of 2.5, using the
part of 6.3.
DEFINITION
The chain complex
SP(X) = {C,(PS(X)),d Q
}
is
called the singular prismatic complex of the space X.
Clearly S(X) is a subcomplex of SP(X). If then SP(A) is a subcomplex of SP(X) and S(A) Hence the inclusions S(A) S(X) and SP(A)
A is & subset of X, = S(X) Pi SP(A).
C
C
SP(X) induce a
homomorphism
77:
S(X)/S(A) > SP(X)/SP(A)
whose kernel
It
will
is
zero
by
77
be convenient to regard
that
is
virtue of the Noether isomorphism theorem. S(X)/S(A) as a subcomplex of
SP(X)/SP(A) so THEOREM 6.7.
an inclusion map.
There exists a
map
r:
(SP(X),SP(A)) > (S(X),S(A
which
is the identity
is
(S(X),S(A))
on S(X). In other words the pair of chain complexes a retract of the pair (SP(X),SP(A)).
n
PROOF.
(n
\
For any linear nsimplex l)chain in W Q by
v*

v
of
A
ff
_i,
define a linear
Then
)
extends uniquely to a
homomorphism
_,))
Dl.
C n (Q(A
l Qc
C. +1 (Q(T.)).
The same
(2)
calculation as in the proof vi,3.2 shows that
dDlc
=
U QC

 DI&,
(1),
for c
e
C n (Q(A n _0).
If
we use
the second set of relations of 6.1 with
we obtain immedi
ately
l
(3)
r\Dl'
c
=
Z>XiC,
for c
e (7n
?(A n _ 2 ))
.
Define the homomorphism
as follows.
If
T is a singular gsimplex,
rQ
then
rQ
T =
1
T.
If
P is a singular
#prism, then
(4)
P = PsDl.^d
..cT
),
7]
PRESERVATION OF HOMOTOPIES
195
where
Ps
is
defined
by
2.7.
Clearly r Q
It
maps C Q (8P(A)) into C 9 (A). This is obvious when applied
prove that
C (X), and it and d commute. to a singular simplex. Hence it suffices to
is
the identity on
r
remains to prove that
dr Q P
=
r q _,dP.
Apply 8
(5)
to both sides of (4), use
dP s = P s d and
g
then use
'
(1) to
obtain
dr Q
definition
P =
(P s u Q
 Pa
l
 P s DlPa
2 d)(d"
O=
PI.
By
(6)
P su
P M ,P/
q
(f
O
Pu =
= P
MJ
Z
a
(d
O
ff
Since
(7)
are singular simplexes,
r.^P,,,
P,
=
r .,P,.
Using
(3)
we obtain
Since,
by
0.2,
P s r\ = 7^, we
have
(8)
;:;
= E(i)v..,p (0 tO
Combining
dr Q
(5), (6), (7),
.
and
(8) gives
P = r^Pu 
r a .,P,

E (l)r
0
fl
_1
P (0 =
This completes the proof.
7.
PRESERVATION OF HOMOTOPIES
THEOREM
7.1.
Two
homotopic maps
f,g:
(X,A)

(Y,R)
induce chain homotopic maps
f 8 ,g f
:
S(X)/S(A)
 S(Y)/S(B).
196
SINGULAR HOMOLOGY THEORY
PROOF.
Let
h:
[CHAP. VII
(X
X
/,
e
A
X
/)
> (F,B)
be a
map
such that for each x
h(x,Q)
X
fc(x,l)
=
/(x),
=
0(x).
(q
For every singular gsimplex T: prism
Aa
T, +I
>
X, consider the singular
+
1)
DT:
defined
 F
by
(0T)(x,0
=
A(r(x),0
for
x
e
A,,
g
i
1.
This defines homomorphisms
D:
which
carries C g (A) into C a+1 (SP(B)). C Q+l (SP(Y))/C ^(SP(B)). Clearly
Q
Thus D:
C q (X)/C<(A)
>
(DT),
=
fT,
(0
(Z)r)
= D(r
(DT) U
(t)
=
)
ffT,
fort
=
0,
,
q.
Hence
=
(DT) U

= gT  fT  DdT.
Let
now
6.7.
by
r: SP(Y)/SP(B) > S(Y)/S(B) be a Then
retraction as asserted
with
= rdDT = r^T  rfT  rDdT = 07  fT  rDdT.
7
7
Thus the operator rD
establishes the desired chain
homotopy.
8.
A COVERING
THEOREM
DEFINITION
8.1.
Let
F
>
singular simplex T:
Au
be a family of subsets of the space X. A will be said to fee/ongr to P if the set
X
8]
A COVERING THEOREM
is
197
contained in at least one of the sets of the family F.
Clearly
then the faces of
singular simplexes of that belong to F make up a subcomplex S(X,F) of the singular complex S(X). If A is a subset of X, then S(A,F) is the subcomplex of S(X,F)
T
also belong to F.
The
X
made up
of the singular simplexes of
A
that belong to F,
i.e.
S(A.F)
=
S(A) C\ S(X,F).
The
inclusions
S(X,F)
77:
C
S(X), S(A,F)
C
S(A) define a homomorphism
S(X,F)/S(A,F) > S(X)/S(A)
which, in view of the Noether isomorphism theorem, has kernel zero. It will be convenient to regard S(X,F)/S(A,F) as a subgroup of S(X)/S(A) so that rj is an inclusion map.
THEOREM
every point
8.2.
is
of X
such that If F is a family of subsets of the space in the interior of at least one set of F, then, for every
X
A
C
X,
the inclusion
map
S(X,F)/S(A,F)
9:
 S(X)/S(A)
is
a homotopy equivalence (see iv,9.2).
The proof
Given a
is
requires several preliminary constructions.
r
linear rsimplex a
v
v
in
A
fl
the barycenter
6, of a
the point
T
A
i
i
*
r
"'
r
+
of
1"
'
'
r
+
1
We
define
two sequences
Sd:
homomorphisms
r
R:
C C
(Q(A.)) >
> r (Q(Aj)
C C
r
(Q(A,))
r
by recurrence as follows: If r For every linear rsimplex a
operation of 5.2
(1)
=
in
0,
then Sd
r
is
the identity
0,
map and R =
0.
A a with
>
we
define using the join
Sd a
Rff
(2)
= =
6,
Sd
((r
6

da,
ff
Sd
<r

flda).
r
We
(3)
(4)
shall
prove by induction with respect to
d Sd
c
that
a/2c
= =
. fl
Sd
c
If
ac,
Sd
c

Rdc.
for every linear rchain c in
A
^
^
0,
the propositions are obvious.
198
SINGULAR HOMOLOGY THEORY
(3)
[CHAP. VII
i
Assuming that
d Sd
and
(4)
hold for linear ichains with
<
r (r
>
0),
consider a linear rsimplex a in
(r
Aa
.
From
5.4
we deduce
= = = = = =
db,
<r
<r
d<r = Sd d<r  b.d Sd do b Sd dd<r = Sd da, Sd d<r  Sd  flda) d6,(<7 Sd  fldo  b.(d<j  a Sd d  dfldtr)  Sd  Rda  b,(d(T  Sdd<r  d(r + Sd  Sd 
Sd
<r
<r
<r
da
+
Rdda)
(T
<7
Formula
that
(3) asserts
that Sd
>
is
a
map Q(A
)
Q(A a ), and
(4) asserts
R is a chain homotopy of Sd into the identity map. Now let T: A X be a singular gsimplex in X. Define chains Sd T and RT by
a
singular
Sd T
where
= T s Sd
(d
ef ),
RT = T a R(f
R
d),
defined on the
T s is defined by 2.7. The operations Sd and base elements extend uniquely to homomorphisms
Sd:
It is
C
g
(X)
 C
g
(X),
R:
C g (X)
> C, +l (X).
easy to see that formulas
singular gchain c in
setting
X.
n
Sd
c
=
c,
Sd
c
(3) and (4) are still valid for every n As before we define the iterates Sd of Sd by " = Sd Sd c for n > 0. Formulas (3) and (4)
71
1
then yield
(5)
d Sd c
n
=
Sd
n
n
dc,
(6)
d
0
R
Sd* c
=
c

Sd
c
*0
#
Sd
1
flc.
If F is any family of subsets of and c is a singular chain in the complex S(X,F), then Sd c and ftc are both in the complex S(X,F). Let now F be a family of subsets of having the property stated
X
X
in 8.2.
there prove that for every singular simplex T in n is an integer n such that Sd T is in the complex S(X,F). Let T: A, > and let F T be the family of subsets of A a of the l form T~ (B) for B e F. Since every point x e is in the interior of
shall
We
X
X
X
one of the sets in F, it follows that every point v e A a in the interior of one of the sets of F T Since the set A a is compact, it follows from such that every subset of A a n,7.5 that there exists a number e > of diameter < is in one of the sets of F T The argument just concluded shows that it suffices to construct an n q d ) is a linear combination of linear integer n such that Sd (d, q v simplexes of diameter < c. To verify this observe that, if or = v
.
.
,
,
9]
INVARIANCE UNDER EXCISION
199
then Sd <r has the form o + is a face of a,. By
t
i
^
it r
where
r
=
b ao


b aq with
,
<T O
=
<r,
and
11,6.8,
diam
r
< ~
9
p +
diam
1
a.
Hence Sd
diameter
large.
n
(d
d
a )
is
n
a linear combination of linear simplexes of
(d
d"),
g (<?/(</
+
l))
diam
and
this
is
< c for n sufficiently
We
are
now ready
and K(X,F)/S(A,F) are composed
v,13.3 to prove that
to prove 8.2. Since the complexes S(X)/S(A) of free groups, it suffices in view of
V
maps
in the
II,(S(X,F)/S(A,F))
H
g
(S(X)/S(A
q.
for integer coefficients
and
for all dimensions
Since
77^
is
one of the
mapping
of the
homology
sequence of the couple
> S(A,F) > S(X,F) > S(X,F)/Fi(A,F) >
into the
homology sequence
of (X,.4),
it
follows from the five
lemma
(i,4.3) that it suffices to
prove that
V
for
(7)
all q.
H H
g
(S(X,F))
is

/
By
exactness this
g
equivalent to showing that
(S(X)/S(X,F))
c g
=
for all dimensions
q.
Let
C
g
(S(X)) be a singular chain such that
K(X,F). By the result previously established there is an n integer n such that Sd c is in S(X,F). Since dc is in R(X,F), the chains 7^ Sd' ac are also in S(X,F). Thus formula (6) yields
dc d
o
=
mod
R
mod
Sd* c
s
c
mod S(X,F)
This proves
(7)
so that c
is
a bounding cycle
S(X,F).
and con
cludes the proof of 8.2.
9.
INVARIANCE UNDER EXCISION
THEOREM
9.1.
Let the inclusion
i:
map
U)
(X Int A.
U,
A the
C
(X,A)
be given where
UC
ia
:
Then
induced
map
S(X  U)/S(A 
U) * S(X)/S(A)
is
a homotopy equivalence.
200
SINGULAR HOMOLOGY THEORY
PROOF.
V.
[CHAP. VII
X
A and Consider^ the family F consisting of the two sets Int (A) = X. Int (A) implies Int (X  E7) Condition U Thus 8.2 applies. Observe that
C
U
S(A
S(X,F)

= S(X 
U)
= S(X C7)
U) C\ S(A), VJ S(A), S(A,F) = S(A).
Thus the map
j:
rj:
is
may
be factored into two maps
> S(X,F)/S(A,F) E7) S(X,F)/S(A,F) > S(X)/S(A)
S(X 
U)/S(A

both of which are induced by inclusions. The map equivalence by 8.2. The map j may be written as
j:
17
is
a
homotopy
S(X  U)/S(X  U)
H 5(4)
 S(X 
E7)
W S(A)/S(A)
which, by the Noether isomorphism theorem, is an isomorphism. Consequently j also is a homotopy equivalence. Since, in the category dg of chain complexes, the composition of homotopy equivalences is again a homotopy equivalence, it follows that i s = 77,7 is a homotopy equivalence.
Theorem
9.1 together
statement of 2.8 that S:
theories.
with 4.1 and 7.1 conclude the proof of the (X,A) > S(X)/S(A) is an Afunctor. This
concludes the construction of the singular homology and cohomology
It should be pointed out that the excision axiom that has been obtained for the singular theories is stronger than that required in i,3, was not made. namely, the assumption that U be an open subset of
X
10.
SINGULAR THEORY OF TRIANGULATED SPACES
Now
theories have been verified,
that the axioms for the singular homology and cohomology it follows from the results of Chapter in
that, in a triangulated space, the singular groups are isomorphic with the simplicial groups computed from the triangulation using one of the two procedures of Chapter vi. Nevertheless, it is useful to have a direct definition of an isomorphism between the singular and the
simplicial groups.
{
ty
In this section such an isomorphism is exhibited. Consider a triangulated pair (X,A) with a triangulation T = We define a map (K,L)
}
.
0:
K /L
L
> S(X)/S(A)
c
as follows:
Given an elementary #cha5n
= AQ

Aq
in
K
,
consider
the linear
map
e
:
A, *
K
10]
TRIANGULATED SPACES
L
c
201
defined by
(d')
singular gsimplex.
= A*, i = We define
0,
,
q.
Then
tL f
:
A,
 X
is
a
0c
=
tL e
.
This defines homomorphisms
0:
C
Q
(K) *
C
f
(X)
carrying
c
(l)
C
fl
(L) into
C
= A
A 1
To prove that /3 and d commute, consider q (A). (l) = /3ce; = (0c) (0 Thus A*. Then 0c
.
The main THEOREM
result of this section
10.1.
is
the following
The
0:
map
K./L. > S(X)/S(A)
induces isomorphisms
0:
over
H,(K,L;G)
coefficient
=
H.(X,A;GF),
0*:
H\K,L;G)
= H\X,A;G)
any
group G.
chain complex
K
/L
the groups of the vi while the groups of (X,A) are the of Chapter
The groups of (K,L) are
singular homology and cohomology groups. In view of the "five lemma" i,4.3, it suffices to prove the case ^4 The proof leans heavily on the concepts introduced in n,9.
=
0.
For
we shall consider the first regular neighborhood every simplex s of N (s). In view of n,9.6 a point a of is in (s) if and only if there is a vertex A of s such that a (A) > a(B) for each vertex B of K not
l
K
K
N
l
in
s.
This implies the following properties of (1) (s) is an open set containing s.
N
1
(s):
N
l
(2)
(3)
Every vertex
l
jv (*0
K that is in N n N\s = N n
of
l
l
(s) is
a vertex of
s.
2)
(s,
2 ).
l
X" for all simplexes s of Let F be the family of open sets tN (s) K. This family satisfies the condition stated in 8.2, and therefore the
C
inclusion
map
:
S(X,F)
C
S(X)
induces homomorphisms
further that in virtue of
0c
is
77^
and
77*
which are isomorphisms. Observe
(1), for each chain c of
K
,
the singular chain
in
S(X,F).
Thus
ft
may
be factored
ft
=
rjy
where
K
i? 7 > S(X,F) > S(X)
202
SINGULAR HOMOLOGY THEORY
[CHAP. VII
iso
with yc
=
ftc.
It suffices therefore to
show that 7^ and 7* are
morphisms. For each point a of
the sense that a(n(a))
If
If
K select
g
a vertex n(a) of
K
a(B) for any vertex
B
of
which is nearest in K. We note the
following properties of n(a):
(4) (5)
a a
is
e
N
a vertex, then n(a) = a. l (s), then n(d) is a vertex of
(4) is
s.
Proposition
of
obvious while
(5) follows
from the characterization
N
(s) quoted above. Given a singular simplex T:
l
A
>
fl
X, consider the vertices
T<T),
i
l
A*
of
= n(r
l
l
=

0,
,
q
5
K.
(5).
If
T(A a )
if
C
T
fcV'OO, then t~
is
Td
l
e
N
(s)
and A*
is
a vertex of
by
Thus,
a singular simplex in S(X,F), then
yT = A
is
.


A*
C
a
This yields homomorphisms 7: an elementary (/chain in K > C (/C,) which clearly commute with d. Thus we have Q (S(X,F))
fl
map
7:
S(X,F)
 K
Let
Q
.
.
We examine the composed map 77.
gchain of
B
l



Ba
be an elementary
K
=
is
have Td
so that
(6)
%
A and let 77 T /B and A* = n(r Td
,



4
l
l
)
.
A Then by definition of we = n(B ). Thus, by (4), ;T = B'
We
that
of (3).
Consequently map of 7*7* = identity. identity, 7^7^ now turn our attention to the map 77. Let T:
77
the identity
K
=
singular simplex of S(X,F),
T(A
tN ) Denote
C
l
(s).
> X be a A such be the least simplex of Such a simplex s exists and is unique by virtue
and
let s
K
C(T)
=
tN
l
(s).
We
shall establish the following properties of the sets
(l) )
C(T)
for
T
in
(7)
(8) (9)
l
C(T
C
are singular simplexes in C(T).
for q > and g (C(T)) = (C(T}) = obvious and so is the first half of (8). Let C(T)
i.e.
T and 77T
C(T)
is
acyclic,
H
Q
H
0.
Td s N (s), and, by (5), of s. Thus yT is in s and 77^ is in t(\s\). Since that t(\s\) C C(T). Thus yyT is in C(T).
l
Proposition (7) is tN (s). It follows that
=
l
l
l
t~
n(t~
Td
1
l
)
is
a vertex
follows
&
CN
(s) ) it
11]
TRIADS
prove
(9)
l
203
To
observe
9
first
that, in
\s\
retract of
N
view of n,9.8,
\s\
is
a deformation
(s)
and, since
is
contractible to point,
is
N
l
(s)
also
is
contractible to a point. Thus C(T) fore homologically trivial by 1,11.5.
contractible to a point
and there
With
(7)(9) established,
a chain homotopy of the
an exact replica of the proof of vi,5.7 yields map 77 with the identity map of S(X F).
9
Consequently
7*7*
which, combined with
=
identity,
7*7*
=
identity,
(6),
shows that 7^ and 7* are isomorphisms with
7* and 7* as
inverses.
11.
TRIADS
Let (XjX^Xz) be a
are subspaces of X.
triad,
i.e.
X
is
a topological space and Xi,X 2
of a triad was defined in Within the setting of the singular homology, we shall define new homology groups /7 a (X;X,,X 2 ) called the homology groups of the triad. We shall imbed these groups in an exact sequence
i,14 for proper triads only.
The homology sequence
called the tnadic homology sequence. Whenever the triad the sequence reduces to the homology sequence of the triad.
is
proper,
DEFINITION
//
fl
11.1.
(X;X,,X 2 ;G) and
H
l
9
The homology and cohomology groups (X;X X.2 }G) of the triad (XjX^X,) are defined
lt
9
as the appropriate
triads induces a
S(X)/S(X ) U S(X 2 ) > S(Y)/S(Y U S(Y 2 ) which in turn induces homomorphisms H (X]X^X^G) > f+: H g (Y;Y Y 2 ;G) and /*: H'(Y;Y Y 29 G) > H'(X;X X,;G).
homology and cohomology groups of the free chain complex S(X)/S(X ) S(X 2 ). A map /: (X;X,,X 2 ) * (Y Y lt Y 2 ) of
U
map
/:
l
l
)
q
l9
l9
l9
Note that the analogs
identity, (gf)
of
Axioms
1
and 2 are
Now
f
= gj^.
homomorphisms
satisfied: (identity)^
=
consider the
induced by inclusions.
Since
i'
may
V
be factored into
Off Y \ QY V ^ O^A^// O^AI^I
S~\
/
i
o^A
Qf
V
\
2>;
_
>
CV V \ o^AjJ Vy
1
O^A^/O^A^
tf/
V
\ /
O/V
\
>
O/ V\ o^Ajy
where
z(
is
direct couple,
the Noether isomorphism, and since the maps i2 ,f form a it follows that (i' f) form a direct couple.
9
204
SINGULAR HOMOLOGY THEORY
DEFINITION
11.2.
[CHAP. VII
The
triadic
homology and cohomology sequences
of (X;X,,X,) over
G
H,(X,X 2 ;G) 
/*
are defined as the appropriate sequences of the couple (i',j A map /: (X;X,,X a )  (F;Ki,F 2 ) induces a map /:
f
).
(i',j') x
>
Therefore the homomorphisms (i'if)Y of the appropriate couples. a map of the triadic homology [cohomology] [/**] /* [/*] yield
/^
sequences.
should be noted that, although the homology groups of the triads (X' X^X 2 ) and (X\X^X^) are the same, their triadic homology sequences
It
t
are distinct.
If,
in the triad
(X;X ,X 2 ), we have Xi
l
DX
2,
then the triadic ho
mology sequence
of (X;Jf 1 ,X 2 ) is easily seen to coincide with the hornology sequence of the triple (X,Xi,X 2 ).
THEOREM
11.3.
For each
is
triad
(X;Xi,X 2 ) and each
coefficient
group
G G
the following conditions are equivalent:
(i)
The
triad
(X;X lt X 2 )
proper in the singular homology theory with
as coefficient group. (ii) The inclusion
isomorphisms k 2 +:
(iii)
(X VJ X 2 ,X 2 ) induces VJ X 2 ,X 2 ;G) for all q. q (X (X l9 X C\ X 2 \G) X2 ) The homomorphism S(X)/S(X ) \J S(X 2 ) * S(X)/S(X
H
map
q
/c 2
:
(X ly X C\
l
l
X H
2)
>
l
l
l
l
U
induced
H^X.X,
(iv)
by VJ
inclusion
induces
q.
isomorphisms
H
q
X ]G) for all
2
(X\X^X 2 ]G)
~
H.(X, VJ
X^X^X^G) =
map S(XJ
0/or
(v)
The inclusion
all the
U
a// 9.
S(X 2 ) > S(X 2
VJ
X
2)
induces
isomorphisms of
homology groups over G.
this proof
A
similar statement holds for the cohomology groups over G.
PROOF.
formulas.
(ii)
Throughout
(iii).
we
shall
omit the symbol
6:
G
in all
<+
Consider the inclusion
map
(X\X ly X 2 )
>
(X;Xi
U
X
2
,X 2 ).
The
triadic
homology
sequence of the second triad coincides
with the homology sequence of the triple (X,Xi VJ the diagram
X ,X
2
2 ).
This yields
11]
TRIADS
205
\
/
H
/
Q
(X,X
2)
\
U
where the upper edge is the triadic homology sequence of (X;X lf Xz) and the lower one is the H.S. of the triple (X,Xi \J X 2 ,X 2 ). An application of the "five lemma" (i,4.3) yields the equivalence of (ii) and (iii).
(i)
(ii).
(ii)
The
condition that the triad
(X^X^X^
be proper con
sists of
and the condition (ii)' obtained from (ii) by interchanging > On the other hand (ii) is the indices 1 and 2. Therefore (i) (ii). equivalent to (iii) which is symmetric in the indices 1 and 2. Thus > (iii) > (ii)'. Hence (ii)  (i). (ii)
(iii)
<>
(iv).
Consider the direct couple
2
l
>
5(X
t
VJ
X )/S(X
)
> S(X)/S(X l
induced by inclusions.

The H.S.
4
of this couple
is
< H.. l (X
l
U X,;^,^)
^(X,
^
H^X,
U XajX^XO
< ..
Thus
(iii)
<>
(iv)
by exactness.
proof
is
(iv) <
(v).
The
the same as the preceding using the direct
couple
o >
s(xj
u s(x
a)
> S(X!
ux
a)
>
s(x
l
u jr )/s(xo u S(X
a
2)

o.

We
(iii),
return for a
moment
to the
diagram used
is
in the proof of
(ii)
and assume that (X',X l9 X 2 )
0+ are isomorphisms.
l l
and
X 2 ) * H q (X lJ X C\ X 2 ) in the diagram. homomorphism H Q ^ (X;X In this way the above diagram contains also the homology sequence of
l
U
Using either
a proper triad. l or ^* we fci~*
Then both
A 2j(t
may
insert the
the proper triad (X;Xi,X 2 ) as defined in i,14.3. The fact that a given triad (X\X ly 2 ) is proper in the singular theory may depend on the coefficient group and whether homology or
X
cohomology is employed. The following theorem shows that, in a sense, homology groups with integer coefficients are universal: THEOREM 11.4. // the triad (X;X,,X 2 ) is proper in the singular
homology theory with integer
homology and cohomology
coefficients,
then
it is
also proper in the singular
theory for arbitrary coefficients.
206
NOTES
PROOF. By 11.3 (i) and (iv), the triad (X ^X^X*) is proper for homology with integer coefficients if and only if the chain complex S(Xi \J Xz)/S(Xi) S(X 2 ) has vanishing integral homology groups. Since the complex in question is a free chain complex, it follows from VJ X 2 v,13.5 that all the homology and cohomology groups Q (X X ,X2 ]G) and H q (X \J X 2 ',X l9 X9 ;G) vanish. Thus, again by 11.3, the triad (X;Xi,X 2 ) is proper for singular homology and cohomology with any coefficients. For the use of triads in homotopy theory, see note below.
U
H
'
l
f
l
l
NOTES
The development of the singular theory. Perhaps the basic idea of is a collection conthe singular theory is that a singular chain on > and an ordinary chain on K. of a complex K, a map X, sisting
X
K
This can be found in the book of Veblen [Analysis Situs, Colloq. Pub. Amer. Math. Soc, v (1921)]. Elaborations of this approach have been given (Hurewicz, Dowker, and Dugundji, Annals of Math. 49 (1948),
391406); but they have the defect that the homology groups are not obtained from chain complexes; for the chains do not form groups. It was Lefschetz [Bull. Amer. Math. Soc. 39 (1933), 124129] who
first defined a group of singular chains. He considered only the > of oriented simplexes into X. Two maps /: Si X, g: s 2 called equivalent if there exists a bary centric equivalence h: s^
maps
>
X were
s2
preserving orientation such that gh
a singular oriented simplex.
equivalence class is called Each such has a negative obtained by ref.
=
An
versing the orientation of the original simplex. Generators for the chain groups are found by choosing one of the two orientations of each singular
simplex.
It
was Cech who pointed out a
difficulty in this
approach:
is
the
two
orientations of a singular simplex could coincide.
An example
obtained
folding a 1simplex about its midpoint and then mapping it into X. This meant that the groups of chains contained elements of order 2,
by
and so were not
free abelian.
As we have
seen, the free abelian char
acter of the chain groups is very useful in proving that the functor S is an Afunctor. defined homomorphisms of S(X) by prescribing
We
on the generators, namely, the singular simplexes. If S(X) were not free abelian, one would have to verify that each such assignment could be extended to a homomorphism. The final step in the development was the substitution of the ordered simplex for the oriented simplex [S. Eilenberg, Annals of Math. 45 (1944), 407447]. The intuitive justification is contained in the theorem
their values
NOTES
207
(vi,6.10) that the ordered chain complex and the alternating chain complex assigned to a simplicial complex are chain equivalent.
Homotopy groups of triads. The groups and sequences of 11 have a natural counterpart in homotopy theory. Given a triad (X;X,,X 2 ) with base point x Blakers and Massey [The homotopy groups of a triad /, Annals of Math. 53 (1951), 161205] introduce homotopy groups > based upon the consideration of maps (A;/?"" ,/?!" ) ir Q (X;A,B)
.
1
1
(X;X!,X 2 ). The groups are defined for n > 2 and are abelian for n > 3. In analogy with the triadic sequences of 11.2, they construct an exact
sequence
<?r Q
(X,X 2 ) < ^(X^X,
if
nX
X
2
,
a)

ir a+1
(X;X lf X 2 ) < TT^GY,^) <homomor
In particular,
X =
X, VJ
it
follows that the "excision"
phism
TT^Y^Y,
nx
2)
+ TT^Y!
u x ,x
a
a)
belongs to an exact sequence in which it is preceded by ir Q +i(Xi VJ VJ Thus the latter groups 2 ;X,,X 2 ). 2 ;Xi,X 2 ) and followed by ir Q (X
X
}
X
measure the extent by which the Excision axiom fails for homotopy groups. Using the homotopy groups of a triad, Blakers and Massey have proved the suspension theorems of Freudenthal (see Note to Chapter i on homotopy groups) and various other results of the same
general nature.
Proof of theorem
8.2.
The proof given
in the text uses v,13.3 to
deduce the fact that ry is a chain equivalence from the fact that rj+ is an isomorphism (for integral homology groups). We shall give here a somewhat longer but more direct proof, not using v,13.3. The advantage of this proof is that it applies in some situations (e.g. for local coefficients) in which the proof of the text breaks down. For a given singular simplex T in X, let {T a be the family of
}
singular simplexes consisting of T, its faces, their faces, and so on. Let nm n(T) be the least integer such that Sd (T a ) is in S(X,F) for every
Ta
in
\T a }.
This integer exists and has the following properties:
n(T ) ^ n(T). nm Sd TisinS(X,F). If T is in S(X,F), O^i n(T) = 0, nm T is in // T is in A, then Sd S(A,F).
(0
208
NOTES
Now
(1)
define for every singular ^simplex
T
in
X
Sd' (T
(l)
rT
=
Sd
nm T +
Q
n(T)l
2 (1)* E
0
B
),
n<r<>)
n(T)l
(2)
DT =
observe that
Y,
70
R
Sd' r.
We
in
rf is a gchain Thus homomorphisms SCX).
r:
in
8(X,F) and
D7
7
is
a
(<?
+
l)chain
C.(X)

C,(S(X,F)),
D:
Ct (X) 
are obtained.
Further
r[C.(A)]
C
C.(S(A,F)),
Z)[C.(A)]
C
C,
From
(2)
we deduce
r  sd n<r) r 
Q
n(T)l
(i)'
E
(1),
sd'
r (0
,
Adding these two formulas and comparing with
we
find
dDT + DdT = T Hence
(3)
rT.
for every singular gchain c in
X
TC
dDc =
c


Ddc.
This implies
arc
= =
d(c
is
dDc
ac

Dae)
=
>
dc
ac
r
+
rac
+
Daac
 aa/)c  dDdc = rac,
Since r carries
which proves that
into S(A,F)
9
a
it
defines a
r:
map S(X) map
S(X,F).
S(X)/S(A) * S(X,F)/S(A,F),
and the composition TT; is the identity map of S(X,F*)/S(A,F). On the other hand ryrc = re, so that formula (3) may be rewritten as dDc + Ddc = c VJTC. Thus IJT is homotopic to the identity map of S(X)/S(A). Hence r is a homotopy inverse of 17.
be a topological space operated on by a Spaces with operators. Let Since each w e (see H, Exer. C for a definition). group yields a > it induces an isomorphism w 5 w: homeomorphism X, *S(.Y)
X
W
W
X
:
EXERCISES
S(X)
If
209
so that the chain
is
A
a subspace of
then we say that
W
as a group of (left) operators. such that the operators of carry A into A operates on the pair (X,A). Then operates also
complex has
W
X
W
W
y
on S(X)/S(A). Let G be an abelian group with as a group of right operators. We define the tensor product G W S(X)/8(A) as the factor group of G S(X)/S(A) obtained by identifying gw (g) c with g (x) we. The homology are called the equivariant homology groups of this tensor product over
W
W
groups H?(X,A;G). If G is an abelian group with as left operators we consider the cochain complex Hom w (S(X)/S(A),G) which is the subcomplex of
W
of equivariant homomorphisms (i.e. The cohomology groups of this subcomplex operator homomorphisms). are called the equivariant cohomology groups Hw(X,A;G).
Hom(S(X)/S(A),G)
consisting
If /: (X,A) > (Y,B) is an equivariant map of pairs each having > as operator group, then the map f s S(X)/S(A) S(Y)/S(B) is and induces homomorphisms /^ and /* of the equivariant equivariant
:
W
homology and cohomology groups.
d,:
The operators
6*:
H?(X,A;G) > H^Atf),
basic concepts defined,
H'W (A\G) 
HV
l
(X,A;G)
can also be defined
in the usual fashion.
it is
With the
tains in this
possible to prove that one ob
way homology and cohomology theories on the category Q. w as operators; follows: the objects are pairs (X,A) with defined as * A" > are equivariant mappings; couples are ij: A maps (X,A); F: (X X /, A X /) * (Y,B) which homotopies are provided by maps
W
are equivariant,
i.e.
Int A and w(U) = U for all w e W', U) > (AVI) where ~U U, a point is a discrete space A" on which operates transitively and withitself is a point). Of course, the resultout fixed points (in particular,
(X 
A
satisfy F(wx_J)
=
wF(x,f)', excisions are inclusions
C
W
W
ing category
If
(i
w
is
W operates simplicially on the simplicial pair (K,L) (see u,Exer. C)
K
,
,
not an admissible category in the sense of
i,l.
,
then equivariant homology and cohomology groups may be defined iL llom w (K /L 0) G) or the comusing either the complexes G (x) w w K a /L a llom w (K a /L a ,G). The resulting groups are isoplexes G morphic with the equivariant homology and cohomology groups of the
pair (A',L) defined using the singular theory.
EXERCISES
A. PATHWISE CONNECTEDNESS.
DEFINITION.
Let
0:0,0:,
e
X.
A mapping/:
/
*
X of the closed unit
210
interval / into
in
EXERCISES
X
such that /(O)
Xi.
=
rr
and /(I)
=
Xi, is
called a
X joining x
1.
and
that each space decomposes into disjoint path wise concalled the path wise connected components of X, such that two points belonging to distinct components cannot be joined by
Show
X
nected sets
{X a },
a path in X.
2.
Establish the isomorphisms
H
where [X a
3.
g
(X;G)
Z H (X
v
a
,G),
II
g
(X;G)
fl //'(*
a
;C)
a
\
are the pathwise connected components of X.
Compute the groups
H
(X,G),
#
(X;(7),
//(X,G), and lT(X',G).
4.
Carry out an analogous discussion
for the groups of a pair (X,A).
B. INFINITE COMPLEXES.
1.
Let
of
K
be an
infinite
complex.
In analogy with the transforma
tion
/3
10, define chain transformations
ft.:
K
> S(\K\ W ),
m
:
K
> S(7C m )
where
/ w and \K\ m have the weak and the metric topology respectively Show that both ft w and /3 m induce isomorphisms of (see n, Exer. F).
the homology and of the cohomology groups. > 2. Consider the map i: \K\ m induced by the identity map \K\ W > K. Show that this map defines S(/v u ,) as a subcomplex of
K
morphisms
S(/v m ) induces isoS(\K\ m )j and that the inclusion map S(\K\ W ) Show by of the homology and the cohomology groups.
C
examples that S(\K\ W )
C.
may
be a proper subcomplex of
THE MAPPING
(see n, Exer.
e
DEFINITION.
CYLINDER. > Let /:
X
Y
be a continuous map.
In the join
X
o
Y
intervals joining x cylinder of /.
1.
X
D6) consider the set C consisting of X, Y, and all with f(x) e Y. The set C is called the mapping
Establish the exactness of the sequence
.


+
H
>
q
/* (X) *
is
H (Y)
q
J*
+
H
g
(C,X) >
H
9
^(X) >

where
2.
j:
Y
(C,X)
the inclusion map.
:
Show the existence of homomorphisms D Q such that dDx + Ddz = /x  x for a: e C a (X).
:
C
Q
(X)
>
C a4.i(C)
3. Consider the map 8 S(X) > S(Y) induced by /: X +Y, and consider the complex f s as introduced in v,13.3. The groups H Q (f s ) will be called the homology groups of / and will be denoted by H Q (f).
EXERCISES
Using a homotopy operator D: fa > S(C)/S(X) by setting
*(x,y)
211
>
S(X)
S(C) as in
2,
define a
map
(Y).
*:
= Dx
+
y
mod S(X)
g
for x
e
C.^X),
t/
e
C
g
Show
that <,:
//<,(/)
= H
(C,X)
is
an isomorphism.
Hint:
use the
diagram
/*
**
i*
>H.(X)
>H
g
(Y)
//,(/)
>#.
!*
4>
k
4
k
+
/i
7/.W
^
//.(C)
^. 7/.(C,X)
7/ f
where the upper row is the exact sequence of v, Exer. Dl, the lower row is the homology sequence of (C,X), the map m is the identity, and n: is the inclusion map. 4. Carry out a similar discussion for a map /: (X,A) > (K,B).
FCC
CHAPTER
Systems of groups
1.
VIII
their limits
and
INTRODUCTION
The material of this chapter is preparatory for the second existence proof of Chapter ix. It is concerned almost entirely with the algebraic aspects of the limiting process used in that proof. The notions of direct
and inverse systems
of groups are due to Pontrjagin. In Cech's original construction of his homology theory, these ideas appeared only in implicit form. There are several advantages to their explicit use. First,
the algebraic and geometric difficulties are separated. Secondly, the reasons for the various limitations on the Cech theory become apparent.
Finally, direct
and inverse systems are of use
is
in other connections, so
that an independent treatment
not amiss.
2.
DIRECT AND INVERSE SYSTEMS
DEFINITION
if it is
2.1.
A relation a <
transitive.
reflexive
0.
and
M M for which a < y and < A directed M, pair a M' implies a A/ and a < a subset of M (M C M) set M A subset M for each in M. in M' implies a < cofinal in M M and N are directed M' such that a < a M there exists a M N an orderpreserving function from M to TV a map in M implies in N). a < < M A DEFINITION system of {X,ir} over a directed a function which attaches to each a M a set X and, to each pair
imply a
r
A
directed set
is called a quasiorder ft in a set In general a < and < a. does not is a quasiordered set such that for each
/3
M
a,/3 e
is
there exists a y
e
7.
1
if
s
e
ft
'
is
if,
e
y
ft
z
ft.
If
sets,
(i.e.
<f>:
is
ft
<t>a
<t>/3
2.2.
direct
sets
set
a
is
e
,
,
such that a
<
/5
in
M, a map
IT a
. .
A =
rrof
>
A
rr/9
such that, for each a
e
M,
TT"
identity,
and
for
a
<
ft
<
y
in
M,
a
==:
^o
212
2]
DlKtiCT
{
AND INVERSE SYSTEMS
21*
An inverse system of sets X,ir\ over a directed set attaches to each a e a set aj and to each pair
M
M
is
a function which
X
a,/3
such that a
<
@
in
M,
a
map
such that
= =
jections of the system.
identity,
Trl,
a a
e
M,
/3
<
<
7
in
Af
.
For both inverse and direct systems, the maps ?rf are called proa If each X [X a is a topological space, or an ftmodule, or a topological group, and each projection is, respectively, continuous, or an 7^homomorphism, or a continuous homomorphism, then \X,ir\ is called a direct [inverse] system of, respectively, spaces,
]
ftmodules, or topological groups. A directed set becomes a category
M
if
each relation a
<
is
reis
garded as a
map a
>
/3.
Then a
direct [inverse] system over
M
to the category of simply a covariant [contravariant] functor from sets and maps, or to the category of 72modules and homomorphisms,
etc.
M
DEFINITION
2.3.
M'
respectively.
Let {A ,7r), JX'XI be direct systems over Then a map
*:
r
M and
A> 
{X',*'\
consists of a
map
</>:
M
in 3/,
M', and, for each a
a
:
M,
a
map
A
"\7~ot
>
A
x r /<fl
such that,
if
a
<
/3
then commutativity holds in the diagram
7T
<t>
X'/4>a

f
J\
\r,<t>0
Now
let
{X,ir\
y
\X'y\ be
inverse systems over Af,AT respectively.
Then a map
consists of a
map
<t>:
M'
3/, and, for each a'
s
M', a
map
214
SYSTEMS OF GROUPS AND THEIR LIMITS
if
[CHAP. VIII
such that,
a'
<
/3'
in
M', then commutativity holds in the diagram
of the direct [inverse] systems are systems of topological spaces, or groups, etc., the components <t>" [$ a >] of the map $ are re
Whenever both
quired to be continuous, or homomorphic, etc.
*:
{X,7rJ
In case
>
{XV},
*':
{XV}
> {X'V'J
are
two maps of
direct [inverse] systems, their composition
is
defined to consist of the compositions
</>'<
and
tt
flr
0'*
,
a
e
M
[<txj
*i4>t. a ..," *M"\. It is a simple exercise to verify that direct or inverse systems of any specified type and their maps form a category. The category of direct Similar [Invga], [inverse] systems of jf?modules is denoted by DirQ
notations are used for the other categories inverse systems of compact groups).
It is to
(e.g.
Invgc
=
is
category of
be noted that the identity
<t>:
the identities
M
>
If
M and
M,M'
a
<t>
:
X  X
a
map
<
of {X,7r}
tt
.
composed
(see 2.1),
of
DEFINITION
{X,7rJ
2.4.
are directed sets,
M'
C
system which correspond to elements and relations in [inverse] system {X'X} over M', and is called the subsystem of (X,ir over M'. If M' is cofinal in the subsystem is called cofinal. The ' > and identity maps identity map
<t>:
{X,7rJ is a direct [inverse]
over A/, then the sets
and and maps of M' form a direct
M
M
M
M
,
X'
form a
a
 Xa
,
fo a
:
X
a
> X.' .,
map
<!>:
{XV}
>
{X,7r},
[*:
{X,xJ >
{XV}]
called the injection of the
subsystem into the system [system into the
subsystem].
It is clear that the identity
map
of {X,7r}
is
an
injection.
Also the
composition of two injections It is to be observed that,
is itself
if
an
injection.
the indexing set
M
reduces to a single
3]
INVERSE LIMITS
215
reduces to a single element, the inverse or direct system indexed by set or group. Conversely, every set or group may be regarded as an inverse or direct system indexed by a set consisting of a single
M
M
element.
we
In particular, see that a mapping
if
in 2.3
we
select
M'
to reduce to one element,
<*>:
[X,*]
 X'
sot
of a direct of
system
:
maps
a
</>
X"  X'
[X,ir}
indexed by
M into the set X'
is
a family
such that
Similarly,
if
in 2.3
M reduces to a point, we obtain a mapping
*:
X
> {A>'}.
If \X',ir'} is
an inverse system, then
$
is
a family of
maps
<
a
:
X
>
X' a
such that
We could
also consider maps A",7r) A"' with A^,TT) an inverse system, r and maps X > {X',ir'\ where {A ',7r'l is a direct system. However, these two typos of maps aro not of much interest to us.
> { {
3.
INVERSE LIMITS
r
DEFINITION
directed set
3.1.
Lot {A
,7rj
be an inverse system of sots over the
is
M.
The
inverse limit A~ w (briefly: limit) of X,ir)
the
subset of the product Yl A' a (soo v,5.1) consisting of those functions .r = in M, such that, for each relation a < j.r
rt
)
(1)
Trf(^)

xa
.
Define the projection
(2)
7T0:
7T0 is
Xm
>
X
ft
by
v
ft
(x)
=
x
ft ,
so that
A' oo.
If
tho
samo
as tho function p^ of v,5.1 restricted to the subset
an inverse system of spaces, then m is assigned the If IA'.TT) is an inverse system a topology it has as a subspaco of J~J and of ftmodules, it is easily soon that A"oo is a subgroup of J~J a X m is assigned this structure of an R modulo. Similarly, an inverse
A",7r)
is
X
X
.
X
;
limit of topological groups
is
a topologioal group.
ir a
LEMMA
PROOF.
xn
3.2.
If a
<
ft,
then
.i>
=
TT^TT^.
By
by
3.3.
(2), Vft (jc) (2).
=
By
(1), ir(.i>)
=
xn
.
Hence irf^Cx)
=
=
Tr a
(x)
LEMMA
//
{A",7r}
is
an
inverse system of spaces [Rmodules or
216
SYSTEMS OF GROUPS AND THEIR LIMITS
then each
ir a
[CHAP. VIII
topological groups],
is
continuous [homomorphic or con
tinuously homomorphic]. These follow from the corresponding properties of the projection p a
(see v,5.2,5.4).
is enlarged to a directed set M' by greater than each a e M, then the inverse m and the projections ir a form an inverse system {X,*} together with system of M' The inverse limit of this enlarged system is in a natural
REMARK.
If
the directed set
M
adjoining an element
X
.
11
EXAMPLE.
correspondence with X*. As it stands the notion of inverse limit
ia
derived from
that of product.
Conversely, the product of an infinite number of sets can be represented as an inverse limit of finite products. Let {X,} be
sets,
a collection of
of finite subsets of
be the collection indexed by an infinite set J. Let J ordered by inclusion. For each a e define
M
M
If Y a = IT,,** ^ K y e ^0 then V s a function defined over a < 0, the function y defines a function on a, denoted by TT(?/), and Y a It is easily seen that the Thus TT: Yp the latter is in Y a inverse limit Y M of {Y,ir\ coincides with J"J X,.
*
ft.
.
>
.
THEOREM
relation
3.4.
// \X,w}
is
an
inverse system over
of
M
a
ir a
<
is
ft
in
a
a
e
M,
a 11
M, map ofX?rf is
e
a 11
map
Xp
into (onto)
a
.
Xa
,
and, for each then, for each
into (onto)
X
>
PROOF.
11 into,
it
Suppose x,y
X^
follows, for every ft rected set, for any y e M, there
and, for some a, x a > a, that Xp = y
is
=
.
ft
Since TT is ya is a diSince
.
M
a
ft
7, a.
Then x
=
ft
y$ implies
xy
TT
=
is
l
(ir%)~
Thus x = y, and 7r a is 11 11 onto, and for a fixed a, x a e (x a ) For any 7 e M, choose a ft
yy
.
It
l
remains to prove that x
t
.
=
Suppose now that each X a For ft > a, define x = > a,y and define x y = wf(jcp).
into.
.
{x y
}
lies in
Xa
.
Suppose
71
<
ft
y2

Let
.
> a,7 and ft 2 > ,7 2 be the elements used in defining x T ,,x 7l /3 = TT^(X^,) Choose ft > j8i,0 2 Then x = Va(x ) ^d7rj.(^)> and x = 1,2). Since TT^ is 11, it follows that ir^(x ) x0l Therefore (z
ft
.
TT.(^) = *?>?. fo) = 7,. Finally, ^(x^) = TT^TT^^) = TT^^) x 7l and the proof is complete. REMARK. If the hypothesis of 3.4 is weakened by requiring that
,
=
each projection be only a mapping onto, one might hope to prove that has a is a countable set, or if each 7r a is a mapping onto. If countable cofinal subset, this can be proved. It is also proved for inverse systems of compact Hausdorff spaces in 3.9. That the result
M
M
does not hold in the general case
L.
Henkin
the
is shown by an example found by Amer. Math. Soc. 1 (1950), 224225]. LEMMA 3.5. // \X,ir} is an inverse system of Hausdorff spaces, then limit space X* is a closed subspace ofY[ ^a
[Proc.
3]
INVERSE LIMITS
PROOF.
217
a
<
ft
ft
in
M
Suppose x
y
e
Y[ X*
xa
.
ls
n
t
m ^U a ,U0
is
Then
for
some
relation
Wa(x0)
sets
^
Since
X
a is
disjoint
U = (O'^a.
The product
in
open
U a ,V a
containing x a
Replace
of the resulting collection
PI X a (see v,5.2). Thus X m is closed since
in the collection {X\. a rectangular neighborhood This open set contains z, but no point of X.
a
X
a Hausdorff space, there are and irf(x^) respectively. Let
,X^ by
its
THEOREM
is
3.6.
The
limit space
complement is open. of' an inverse system
of compact spaces
is
a compact space.
then the limit space is
If each space of the inverse system nonvacuous.
nonvacuouSy
a is a compart PROOF. Since m is closed in Y\ A" (3.5), and JJ space (v,5.4), it follows immediately that X* is a compact space. For the second part, (Recall that "compact" includes "Hausdorff").
X
X
define, for
each
/3
s
M,
the subset
Y
all
of J~J
X
/3.
a
to consist of those eleft
ments x such that
for
Xfl
TT^(X^)
=
x y for
y
<
can be specified arbitrarily in
Next,
X/j,
then
Now Y is nonvacuous; set x y = iry(x^) for y < 0,
all
and
finally use the choice
axiom to extend x over
other sets of the
system.
Y
is
closed.
0,
The proof
then
of this
and is omitted. If a < from TT? = TyTTa where y
Ya
/3.
D
Y^
similar to that of 3.5, This follows immediately
is
<
a
<
Since Af
is
a directed
set, it follows
now
any
that
finite
a collection of closed sets in a compact space and number of them have a nonvacuous intersection. It follows
\
Ya
}
is
section belongs to
is nonvacuous. Clearly, any point of this interand the proof is complete. X, THEOREM 3.7. // A",TT} is an inverse system of compact spaces over M, a e My and U is an open set of X a containing T a (X), then there exists a > a such that ir(X0) C U. PROOF. For each > a let Y = X  (O'^f/). Otherwise,
that their intersection
{
f
ft
be the map defined by *}. Then an inverse system of compact spaces. If the conclusion of 3.7 were false, each Yp would be nonvacuous. Hence Y m X* would be nonvacuous (3.6). If y e F,, then y a = p a (y) Tc a (y) would lie in both Y a and U. As this is a contradiction, 3.7 is proved. COROLLARY 3.8. Under the hypotheses of 3.7,
Yp
=
X
.
ft
If
ft
<
y
let p}:
Yy
>
Y
ft
{F,s,p^}
is
C
PROOF.
If
xa
e
X
a is
/3.
not in
7r rt
(X
00 ),
then
its
complement
U
con
tains Va(Xft) for
some
Therefore x a
is
not in the intersection.
This
shows that the
Tra(X ) for each the right.
ft
left side
>
a.
contains the right side. By 3.2, T a (X g ) This proves that the left side is included in
C
218
SYSTEMS OF GROUPS AND THEIR LIMITS
COROLLARY
3.9.
//, to the
[CHAP. VIII
the condition
hypothesis of 3.7,
we adjoin
a mapping onto, then each ir a is a mapping onto. This follows immediately from 3.8. DEFINITION 3.10. Let $: [X,ir\ > [X y\ be a map of one inverse of 4> is a map system into another (see 2.3). The inverse limit
that each ir% is
f </>
0oo*
X<x>
>
Xm
set
defined as follows:
in
If
x
e
X*,
and a
M
e Af',
x
=
<<,(*(<*))
If
<
x'.
/8
x.
follows from the commutativity condition of 2.3 that Define <t> <*>(x) Therefore x' = jx) is an element of
',
it
?r'f (x)
=
X.
=
LEMMA
3.11.
//
<f>:
{X^}
f
>
{
X',ir
\
and a
e
M'
,
//^M commutativity
holds in the diagram
This follows immediately from the definition.
LEMMA 3.12. // {X,ir} is an inverse system of spaces, then the sets 1 runs over all open sets of a and a over M, form a f^a (/)), where base for the open sets of be open in X*, and x e W. By v,5.2, there exists a PROOF. Let
U
X
X.
W
rectangular neighborhood
V
of x in J"!
Xa
such that
T
V
>
C\
Xm C
W.
,
Now V is determined
and consists of
set,
all
y
e
by specifying open sets tl %a such that ?/, e
7
,
CX
.
a
(i
=
is
is
1,
n),
F
t
Since J/
TT!,
a directed
there exists a
ft
>
a,
(f
=
1,
,
n).
Since
continuous,
and
ir ai (xfi}
e
T,,
it
follows that
u =
is
A (TT^.rxFo
i
.
an open
e
set of
X$ containing x
ft
Hence x
?/
e
ir^(U).
y
TTft\U).
Then
^
s
e
U.
this in turn implies y
F
H
This implies
a
X.
C
W.
V, (z Thus x e
,
e
=
!
1,
,
7i>
((7)
C
}
Suppose n); and W, and
are 6o/A
the
lemma is THEOREM
proved.
3.13.
//
<l>:
\X,ir}
> {X'^
f
}
and
{X>},{XV
1
inverse systems of spaces [Rmodules or topological groups], then the limit 4>co of $ is continuous [homomorphic or continuously homomorphic].
continuity it suffices to show that <t>* maps each set of a base for the open sets of X* into an open set of X*. By open
PROOF.
To prove
3]
INVERSE LIMITS
l
219
3.12, the sets ir'~ (U),
where a
s
M', and
(7 is
open
in
X,
form such a
base.
By
3.11,
Since
<f>
a
and
7r^ (a)
are continuous, this
<f>
is
an open
set.
The proof
that,
in the case of /^modules,
a is
homomorphic
is left
to the reader.
THEOREM 3.14. Let ft be any one of the following categories: sets and maps, spaces and continuous maps, compact spaces and continuous maps, Rmoduks and homomorphisms, or compact groups and continuous homomorphisms. Let Inv(a) denote the category of inverse systems and maps of such all of whose elements belong to ft. Then the operations of assigning an inverse limit X* to each inverse system X,TT and an inverse
{
\
limit ^oo to each
map
4>:
{X,7r
{X',7r'J
form a covariant functor from
except those
<
Inv(ft) to a.
PROOF.
Condition
1
The
fact that
X
this
e
d
is
trivial in all cases
involving compactness, and
was proved
in 3.8.
That
e ft is
3.13.
and
this
>
of iv,4 for a covariant functor requires that <,: follows from its definition. Condition 2 reads:
(X,7r)
is
X
X<,
If
<:
of
fX,7r)
the identity, so also
is
<,.
But
4> is
composed
The final condition 3 requires identity maps, so this is immediate. that the limit of a composition of maps 4>,<J>' shall be the composition
of their limits.
This follows readily from the fact that
</>(x) is defined
</>,
by mapping and the
the coordinates of x
by means
4>'<f>
of the coordinate functions
fact that the composition
was defined by composing
coordinate functions.
Propositions 3.10, 3.11, and 3.13 can be applied in the special case when the system {X,v\ = A" is a system indexed by a set consisting of a single point. The limit of a system consisting of a single set being the
set itself,
we
>
see that a
<,:
X
X
mapping such that *.> =
Let
<f>:
j
<i>:
X
>
{X'XJ
induces a mapping
</>
THEOREM
sets 3f,Af',
3.15.
let
and
directed subset
N
of
M'
be inverse systems over directed > \X,w\ X',ir'} be a map. If there exists a r is cofinal in such that (1) (see 2.1), (2)
Ar ,7rj,{Ar ',7r'}
W
M
onto Xjj for each 0(AO is cofinal in M, and (3) <t>p is a 11 map of X^ is a 11 map of X*, onto XL ft s N, then e abbreviate <f>(0) by 0'. Suppose x,y e X* are PROOF. If x a 7* y a distinct. Then, for some a e By 2, there is a e W such that a < 0'. Then x ^ y^\ and by 3, fyfa.) 7* ^(y/f ) But these It reis 11 into. are the ^coordinates of 0(x) and </>(t/). Thus mains to prove that <> is onto. Suppose x' e X. For each a e M, choose a /3 e W such that a < 0', and define
(/5)
<
M
r
,
M
.
y
.
fi
<
(3)
xa
= T*V(*;).
MO
If
SYSTEMS OF GROUPS AND THEIR LIMITS
[CHAP. VIII
<
y
in
N, commutativity
in the
diagram
Xy>
k
1
1
1
and the
11 properties of
<fo,<
7
yield
nV^X^) =
l
<t>p
(x'i*).
.
It follows
Since AT is that (3) with 7 in place of yields the same value of x a directed, it follows that (3) is independent of the choice of ft. We must show that x = \x^\ lies in X*. Suppose a < 7 in M. Choose ft e
N
such that 7
<
ft'.
As
just
shown
Since
ir*
= T!^,
ft
it
To prove
for each
l
this it suffices to
e
M
= x'. follows that Trl(x y ) = x a Finally, <(#) show that they have the same 0coordinate
.
r
.
By 1,
it
suffices to
definition,
<l>]i
(*,(x))^
=
<fo(#0')fl
(Xfi)*
Therefore^/j(x ') = a:J, COROLLARY 3.16. Let M' be a cofinal subset of the directed
the
e N. By prove this for each As shown above, ov = ^^(x'ft) = and the proof is complete.
set
M.
Let {X,ir} be an inverse system over A/,
(see 2.4),
and
<f>
{X'XJ subsystem over the injection of {X,7rj into {X',7r'j. Then, the inverse
M
f
limit
a 11 map of m onto X. of = M'. This follows from 3.15 with
</>
$
is
X
W
COROLLARY
system over
for each a
e
3.17.
Let
M'
be cofinal in
e
M, and
let
M. Then
M'.
two elements x,y
X*
coincide if
[X,*} be an inverse and only ifx a = y a
LEMMA 3.18. Let 4> and ^ be maps each a e M' we have </>() < ^(a), and PROOF. By 3.11 and 3.2, we have
,
{X,ir}
a
[X',*'} such that, for
\l/
= * a ^t()
Then
</>
= ^.
fl"a0oo
=
*a**(a)
=
s
07T ^
Af ',
(
a )7T^ ( a)
=
^ a ^"^(a)
=
^"a^oo.
Since this holds for each a
we have <, = ^.
4.
DIRECT LIMITS
DEFINITION
M where each G
^
G
Let {G,7r} be a direct system over the directed set an /2module, and each TT is homomorphic. Let denote the direct sum (v,5.5) of the ftmodules of {G ir\. If
4.1.
a
is
t
4]
DIRECT LIMITS
we
shall agree to identify g
221
in
g
ia
e
G",
with
its
of v,5.6.
For each a
<
ft
in
M and each g
<f
image
a
J^
G under the map
e
Ga
the element
(1)
T.V ~
of
^G
is
called a relation.
Let
Q
be the subgroup of
)
G
generated
by
all relations.
The
direct limit of {G,7rJ is the factor
group
G)/Q.
The
natural
map
^G
*
G
00
defines
homomorphisms
called projections.
Since each relation
(1) is
00
mapped
TI^TT^
into zero,
we have
a
e
LEMMA LEMMA
a
Tr a g
4.2. 4.3.
If a
//
u
< 0, e G
then
=
wa
.
,
then there exists
an a and a g
Ga
such that
=
u.
first
e
Now
A/
irl
k
PROOF. This is the u is an image of v
ah
place where the directedness of
M
s
is
needed.
Since
^
k
.
is
directed, there exists
j
and v = ^" an a > a* for k
G,
a*
where

a
*
Ga
v'
*.
=
it
1,
,
n.
Let g k
v e
=
Q.
g
and
r'
let r'
= X^ g
?/.
Since
a
<;*
.
*
^
s
Q,
follows that
Hence
*lg
also
maps on
If g
e
LEMMA = o.
PROOF.
4.4.
But G y and
r' s
Ga ir y g =
0,
then there is a 8
>
y such
thai
By
a
of elements of the
G is a linear combination hypothesis the element g of form (1). Since a multiple of (1) also has the same
of elements of the
^
form, g
a,/3
is
sum
form
(1).
Choose a
5
>
7 and
all
entering such a sum.
Then
*}ff
=
(*'v

0)
+
Q
is
also such a
sum and
9
5
exceeds
all a,/3
entering the sum.
Since
.V it
=
is
Or.'?"
 9) +
6rS(T.V)

(T.V)),
follows that ?r^
a
sum
of terms (1)
where each
=
5.
All terms
with a
common a
can be lumped into a single term.
Thus
Since X^ G is a direct sum, any relation is a consequence of relations in a for each a 7* 5. the individual summands. It follows then that g =
Clearly TT^

1
g
=
ir a
0,
a
so w'g
=
if
0.
LEMMA
that
a
4.5.
g
= TT^
and only
if there exists
a y
>
aft such
*lg
= *V
222
SYSTEMS OF GROUPS AND THEIR LIMITS
PROOF.
If
[CHAP. VIII
Q.
such a 7
if ir a g
exists,
a
then wlg
is
a
a
g
Hence
7T0/.
their difference
/a
g
also in Q.
and TT^ / are in a But this implies Tr a g
=
Conversely,
=
TT/J/,
choose a 7
* fa*.
By
4.2,
7r 7
=
0.
Then by
> aft and let =; wig" 4.4 there exists a 8 > 7 such that
*h =
It is to
TV
 Tj/ =
0.
be observed that the directedness of
M
is
not used in the
4.5,
definition of the direct limit.
As a consequence
of 4.3
and
the following theorem, which provides an alternative definition of using the directedness of M.
we have G"
shall
THEOREM
a
4.6.
Let {G,ir} be a direct system of Rmodules.
We
say that g
Trig
e
5
Ga

is equivalent to g* e
G&
if there exists
a
a y
e
<7
= ^V
This relation divides
the elements g
</
t
disjoint equivalence classes.
The sum
a
+
g\.
~g.
2
=
<7i,02
> aft such that G, for all a, into
classes is obtained by choosing
2 g",g"
Ga
in which
of two equivalence have representatives
and defining
~g
to be the class
a
of g\
+
to be the class of rg
a
where g
is
in the class
to
The product fg is defined Then the equivalence classes
the
form an Rmodule isomorphic
attaches to each class the
G
of
under
correspondence which
image in
is
*
G
its representatives.
THEOREM 4.7 a < in My ?rf
then, for each a,
// {G,7r}
:
ir a
:
Ga Ga
G
>
a direct system over M, and, for each has kernel zero [is a homomorphism onto]
PROOF.
The
first
half
is
onto, there exists a g such that a (g ) It follows that ir a (g*) = u; hence ir a is onto. ir}(g ). COROLLARY 4.8. // each ?rf is an isomorphism, so also is each ir a
is
jr
ft
u e G, and a e M. By 4.3, Choose a 7 > a,/3. Since 7r
kernel zero [is a homomorphism onto]. an immediate consequence of 4.4. Suppose & there exists a /3 and a g such that T (g ) = u.
a
y
G* has
a
=
.
be compared with the corresponding ones for inverse systems (see 3.4 and the subsequent remark; also 3.9). the collection of all finite EXAMPLE 4.9. Let G be a group and a subsets of G. For each a e M, let G be the subgroup of G generated a G* and let by the elements of a. Define a < ft to mean that G is directed, Ga G^ be the inclusion. It is easily proved that TT
results of 4.7 should
The
M
C
9
:
M
,
limit of its subgroups
G. Thus, any group is the direct on finite bases. EXAMPLE 4.10. Let [H be a family of groups indexed by a set J, be the collection of finite subsets be their direct sum. Let and let
{G,7r}
is
,
a direct system, and
G
=
)
H
M
If a ordered by inclusion. If a e M, define G = ]C,, 0, G a > G^ in the obvious way (the coordinates of TT (g) are define TT
of
J
a
tt
H
9
.
C
:
those of g for j
e
a and are zero
direct
for j
e ft
a).
It follows
G ~
//.
Thus, a
sum
<l>:
is
a
direct limit of finite direct
>
quickly that sums.
of
DEFINITION
4.11.
Let
{G,7r}
{G'X} be a map
one direct
4]
DIRECT LIMITS
of
22S
a
The homomorphisms G' G are the components of a homomorphism G 2Z G'. ] Because of the commutativity condition of 3.3, maps each relation of G into a relation of G'. Hence induces a homomorphism (linear map) of the factor group
system
a
>
Rmodules
into
another.
<
:
0(a)
<f>:
<l>
2
^
<f>
<t>~:
G" > G'
00
called the dzred Zz>m7 of $.
It satisfies
</>V a
=
^
a
(
a)0
,
a
a
e
M
g
.
Referring to the alternative definition of 4.6, note that g
a
<t>
g" implies Q*into one in (G',7r'j. This
<l>
^
~
$
ft
Hence
<$>
maps an equivalence
of classes
is
class in
{G,?r}
mapping
precisely
0.
<J>:
Definition 4.11
{G,7r
>
may be applied in the special case of a map G' of a direct system {G,irJ into a single fimodule G'.
of the following
The proof
to the reader.
theorem
is
straightforward and
is
left
THEOREM
maps
all
a direct
4.12. Let Dir(S*) be the category of direct systems and whose elements belong to 9 Then the operations of assigning of limit to each direct system and a direct limit to each map of one
is
such into another
from Dir(Q/ ) to <^ R be noted that we do not attempt to assign a topology to a direct limit of topological groups. This can be done in a However the natural way using the topology of the product space. G even in the case of is usually not closed in group Q of relations compact groups. As a consequence the topology induced in G is not a Hausdorff topology. This means that no analog of 4.12 would hold
e
.
a coranant functor
REMARK.
It is to
^
for
any reasonable category
of topological groups.
(G',7r'l
>
THEOREM
spectively,
4.13.
let
T
Let G,7r),
<i>:
be direct systems over be a map.
is
M,M
G'"
e
e

f
re
and
\G',ir'}
f
{G,7r)
If there exists
a
directed subset
cofinal in
A
of
M
r
such that (1) .V
G'*
M, and (3)
If
/3
0':
 G M) for each
cofinal in J/',
ft
(3'.
(2)
<f>(N) is
e
A
r
,
then
0:
G.
PROOF.
e
M', abbreviate 0(0) by
is
^(g) = 0. Since A that rtf = g (see
there exists a
6
cofinal, there exists
a
(3
Suppose g N and a </*
0.
G'* and
G'* such
4.3).
>
e
ft'
in
7
M such that 7i><A/)
7'
By
4.11,
*vr0V) =
=
y
<l>
*"(^) Since
=
BY
By
44

<f>(N) is cofinal,
there exists a 7
is
A
such that
>
5.
Then irV/to*) =
IT? (g*)
0.
the
7
commutativity condition
isomorphic,*
of 2.3,
we have
=
7
0.
Since
 0. Hence g 0. This TT ir^J ^) TJ(^) that 4> has kernel zero. Suppose now that g e G. Since j>(N) proves = ^. is cofinal, there exists a ft e A' and a /' e G^' such that *>(0*') 10 = /'. Let e G'* such that /(0'0 is onto, there exists a g Since
V)
=
=
=
224
f
SYSTEMS OF GROUPS AND THEIR LIMITS
ft
[CHAP. VIII
of
Q
=
*i(9'
)
Then
g.
r
g
e
G
/0
,
and by the commutativity condition
be cofinal in
4.11,
ftf) =
COROLLARY
4.14.
Let
M
r
subsystem over G' jection (see 2.4). Then 4": This follows from 4.13 with
the
M, (G'X}
M
M,
jG,?r}
a
direct system over
*
f
,
and $:
(G'X)
{G,ir\
the in
G. # = M'.
*
f
LEMMA
a
e
4.15.
Let
By
= T^faT^" Then = ^. we have <t>(a) < ^(a) and a PROOF. If g e G, choose an a and a g' e G such that TT^' 4.11, we have
a
y
M
$,V
be
maps \G^\
\l/
{G',ir
}
such
that,
for each
=
0.
6.
SYSTEMS OF EXACT SEQUENCES
DEFINITION
directed set
5.1.
is
M
An inverse system of lower sequences {8,*} over the a lower a function which attaches to each a e
M
sequence
Sa
(see i,2),
=
{G ai<n
<
a
,
c
}
and to each
such that
TT"
relation
a
<
ft
in
if
M, a homomorphism T:
a
.
< < 7, TT^ = TT^. identity, and, the groups and homomorphisms {G a Q ,7r ia g, any Then, form an inverse system of groups over M. Its limit group is denoted Again, for a fixed q, the homomorphism {</><,,<,) together with by <7oo,
S0
*
Sa
=
for
fixed

.
fl
the identity
map
<
of
The
limit of
a
is
}
S
=
{
G\*, c ,0> l0
{G a ,^i,7rf. iJ. The lower sequence denoted by <a,, c Goo. a i. (?,<, so obtained is called the inverse limit of the system
<t>
M form a map
:
Q
:
!Ga, c ,7rf. c
>

}
tf
upper sequences and its limit sequence are defined in an analogous manner. In the above definition it is to be understood that all groups and
{$,TT).
A
direct system of
homomorphisms
fixed
in an inverse system of lower sequences belong to a one of the standard categories S,Sc That the limit sequence is of the same type is proved in the preceding sections. In the case of a
direct
if
system of upper sequences we shall use only the categories g/j. DEFINITION 5.2. A lower [upper] sequence is said to be of order 2 the composition of any two successive homomorphisms of the sequence
zero:
i.e.
is
kernel
D image.
Although
this notion coincides
with that
of chain [cochain] complex (v,2), we shall prefer the "order 2" language whenever the sequences are not to be treated as chain complexes.
5.3. // each sequence of an inverse [direct] system of lower sequences is of order 2, then the limit sequence is also of order 2. [upper]
THEOREM
5]
SYSTEMS OF EXACT SEQUENCES
The composition
limit of
4> _i4> a
225
into (Ga.oa,
a ,<ji<a,<i
PROOF.
*a.<,2\ consists of
the identity
map
is
of
mapping (G a>(/ ,7rf and the maps
M
Q
<
\
=
0
Hence the inverse
tion
#cD.
fl
tfv,^
zero.
By
3.14, this
is
the composi
i0oo.fl.
The corresponding
to the reader.
proof for a direct system of upper
sequences
is exact,
is left
THEOREM
// each sequence of a direct system of upper sequences then the limit sequence is also exact. PROOF. One half of exactness follows from 5.3. Suppose u e G'
5.4.
Q
and
<t>'
(u)
=
in G"'"*
a
)
1
.
By
4.9,
tt
4.3, there exists
an a and a u
a e
Ga
'
Q
such that Wa(u
=
u.
By
,rrv"(M
=,
)
0x(o =
0.
o.
By
4.4, there exists
a
ft
>
a such that
irt^V" V) =
Let u*
q
=
a
7rl'(u
0. u'.
).
= /<V) =
4>*'
(u*)
By
from the cominutativity of TT and Q~ exactness of S*, there exists a v* * G*' ^ Then  u. Let v = *y\v*) eCr (r)
It follows
1
that
l
such This
1
1</
" 1
.
proves the other half of exactness.
EXAMPLE 5.5. The limit sequence of an inverse system of exact lower sequences need not be exact as the following example shows. Consider the diagram
j
7
,
j
IT
,
j 2J
J
>
J
>
J/2J
e is
where J is the group of integers, r(n) = In, y(n) = 3^, map, and is the natural map. It is easy to see that
rj
the identity
T
rj
S:
is
* J
 J
> J/2J >
an exact sequence, and that the homomorphisms 7,7,6 together yield a homomorphism </>: AS > *S. be the set of positive integers with the usual order. For Let Define all other 7r for ft < a define S a = S and 7r + = each a e by composition: TT = (<t>Y~ Then {>S a ,7rf is an inverse system of
M M
1
<t>.
(
exact lower sequences.
S.:
Let
>
G
>
G
>
H 
226
SYSTEMS OF GROUPS AND THEIR LIMITS
[CHAP. VIII
be the limit sequence of this inverse system. Since e is the identity = J/2J. An element of G is a sequence {n a }, map, it follows that It follows that n a is divisible a e of integers such that n a = 3n a +
M
H
1
.
by
arbitrarily high
>
powers
Let
of 3;
i.e.
na
=
and
is
(7
=
0.
Thus S*
is
the sequence
J/2/
{
which
not exact.
S,TT} be an inverse system of exact lower sequences and homomorphisms of S^\ belong to the category groups $c of compact groups. Then the limit sequence S* of {S,7r} is also exact. PROOF. Onehalf of exactness follows from 5.3. Suppose g e (?,<, and m Q (g) = 0. Let g a = v a Q ((j) be the coordinate of g in (?.<,.
THEOREM
5.6.
over
M where
<t>
,
all
{
,
Since
exact,
<t><, g
(g)
=
0, it
follows that
}
X =
a
.
(<t>
a .<ni)~
</>
,
(g a )
is
= for each a. Since S a is a .(0a) a closed non vacuous subset of G a q +\'
.
From the relation a Q+1 7rf, u+l = 7Ta,<,<Va+i> it follows that Xp into X a Let pf be the map so defined. Then {X a ,pf
}
Trf.^j
is
maps
an inverse
nonvacuous, it follows compact spaces. system from 3.6 that the limit space X^ is nonvacuous. It is easily seen that This proves the other half A" C G a g +i and 0co,<,+i maps A^ into g.
of
is
,
Since each space
of exactness.
Let {,TT} be an inverse system of exact lower sequences terms of {S,ir} belong to the category of finite dimensional Then the limit sequence S of {S,ir\ is vector spaces over a fixed field F.
5.7.
THEOREM
over
M where
all
also exact.
a vector space over F, a variety V in G means a coset The dimension of V is that of //. It some is easily seen that, under a linear map, the image (inverse image) of a variety is a variety. It is also clear that the intersection of a collection of varieties, if not vacuous, is also a variety. We interrupt the proof
PROOF.
If
G
is
of
linear subspace // of G.
for a basic
lemma.
5.8.
LEMMA
is
//
G
is
a
finite
a
collection of varieties in
G
(j
dimensional vector space over F, and F, e J) such that any finite number have a
{
}
nonvacuous
Since
intersection, then C\, tJ F,
^
0.
dimension, any subspace has a finite dimension, and therefore the same holds for any variety. By hypothesis, any finite subcollection of {F,J has an intersection of dimension ^ 0. Since all
G
has
finite
such dimensions are integers, there is a smallest such integer A:, and a h corresponding V = C\ ^ V ){ with dim V = k. For any j e J, V C\ Vj 7* by hypothesis. Also dim V C\ V, = k by the minimal property of fc. Thus dim V = dim V C\ V, which implies V = V C\ V,, and
l
this in turn implies
FC
We
ga
is
return
now
F,. This proves the lemma. As before, onehalf of to the proof proper of 5.7.
exactness follows from 5.3.
Suppose g
e
<?,<,
and
0,
0.
a ((7)
=
0.
If
the coordinate of g in
G 0iQ
,
then
<. <,((/) =
Define
F
=
6]
l
COMMUTATIVITY OF OPERATIONS
(Qa)
227
(<t><*.Q+i)~
Since
Sa
is
is
exact,
Fa
<,+!, V* functions v = {V a
kernel of
}
a variety.
if
is not vacuous. As a coset of the Consider now the family ^ of all
defined for a
e
M
such that
Va
is
a variety in
t>
G...+I,
V a C VI
and,
a
<
ft,
7rf(F,)
C
v
V..
Clearly
= V
for each
belongs to ^. If u,t> e ^, the relation u a e This defines a partial order in ^.
the partially ordered set SF contains a maximal simply ordered subset ^'. For each is '. Since a, define V* to be the intersection of all V a for V e
.
M
C
means
Ua C Fa
By Zorn's lemma,
'
simply ordered, any
5.8
finite
subset has a smallest element.
and the assumption that
It is easily
1
Ga
.
g +i
has
1
finite
vacuous.
proved that 7rf(T0)
Since
v
C
t;
1
.
Hence, by dimension, F is not V* whenever a < ft.
'
'
is and e XP'. Thus we have found an element it follows that v maximal, v e ^ such that v C f in ^ implies y = v It remains to show that
Therefore
1
v
=
{V^}
lies in M*.
1
C
for each v
e
1
Consider any fixed element T a. s M. The collection (7rf(T J)} for all ft > a has the property that any finite subcollection has a non vacuous intersection (M is a directed Hence by 5.8 there is an element g' in their intersection. For a set).
each
is
Fa
l
a single element of
Ga
.
Q+
}
.
any y
e
M,
choose a
ft
>
V,
a,7
and define
r\ (irftrfr
1
 V*
ft.
^)).
it
The
choice
made
of g
insures that
Vy
5.
is
not vacuous.
r
independent of the choice of
From
7
this
Clearly follows that 7
Va =
and
<7
gZ,
and
1
t'
.
ir'(T'a)
C
Vy
is
if
<
Thus
1
=
{Vy}
v
V y is F C Vy, belongs to ^
=
*/;
v
C
V^.
By
the minimal property of
^7
v
,
we have
hence
=
Thus
=
j<7a)
an element of
(j,. g41
and 0.
a+ i(^)
=
g.
This completes the proof of
5.7.
REMARK.
The
failure
of exactness
under inverse limits can be
>
described with greater precision. Let 4>: {G',ir'} be a map \G<ir\ of one inverse system of groups into another. Then the following relations always hold:
kernel
<
=
lim
{
kernel
<
a ),
image
</>
C Km
{image
</>).
In the second case equality holds if all groups are compact or finite dimensional vector spaces over a field. The example 5.5 shows that equality does not hold in general. In the case of direct systems equality holds in both cases.
6.
COMMUTATIVITY OF THE OPERATIONS OF LIMIT AND FACTOR GROUP
DEFINITION
6.1.
Let
e
{G,ir}
be an inverse system of groups over
M.
Suppose, for each
a
^/, that
Ha
is
a subgroup of
<7 a ,
and suppose,
228
for each
SYSTEMS OF GROUPS AND THEIR LIMITS
a
[CHAP. VIII
<
in
Af that
,
TT
maps
H
ft
into
H
a.
Let pf
:
H
>
ft
H
a
be
the
map
It is called
Ka =
the
by TT. Clearly {H,p} is an inverse system over Af. a system of subgroups of {(?,*}. For each a e Af, define > / G a /H a and, for each a < in Af define af a to be
defined
:
,
,
^
map
induced by TT.
Then {K,a}
is
an inverse system over Af called
the system of factor groups of \G,ir} by \H,p}. <J>: {G,7rJ j#,p} {(7,7r) and the natural map ^:
in the obvious
The
inclusion
map
{#,0} are defined
of {//,p}
is
way. By definition, each element of the limit group // an element of (?, and #: // G> is the inclusion. The
left
corresponding definitions for direct systems are
to the reader.
THEOREM
6.2.
Let
{
G,?r
} , {
H,p
J
,
j
subgroups, and factor groups over
M where
>
K,a
}
be
all
an
inverse system of groups,
to the
terms belong
category
gc of compact groups or
over a fixed field F\ then
to the
category of finite dimensional vector spaces
^: G*
a, adjoin
K m induces an isomorphism (?///, =
infinite set of trivial
PROOF.
For each
an
groups and
maps
to
II a
*
Ga
>
Ka
so as to obtain a lower sequence
<t>
Sa
t
>
It is clear that
>
Ha
 Ga
Ka
>
^
Sa
is
an exact sequence.
is
For each a
<
/3
in
Af adjoin
,
to 7rf,pf,crf
Sft
>
an
infinite set of trivial
maps
so as to obtain a
map
rf:
Sa
.
Then
{S,r\
an inverse system
of exact sequences.
It is
also clear that the limit sequence
Sm
/~Y
consists of
f\
TT
TT
f\
.
By
5.6
and
its
5.7, *S> is
an exact sequence.
It follows that
\l/
m
must be
onto and
It
if it is
kernel
is
H^.
This completes the proof.
is
to be observed that example 5.5 shows that 6.2 does not hold merely required that all terms belong to the category 9.
THEOREM 6.3. Let {G,7r},{H,p},{K,<r} be a direct system of groups, where all terms belong to the category subgroups, and factor groups over then \l/ m m m induces an isomorphism G m /H m G* 9^; The proof is analogous to that of 6.2 and uses 5.4 in place of 5.6
M
:
K
~ K
.
and
5.7.
EXERCISES
EXERCISES
A. INVERSE LIMITS OF SPACES.
1.
229
Prove that every countable directed
set contains a cofinal sub
sequence (sequence
integers).
2.
=
directed set isomorphic with the set of positive
Show
and
If
if
sets,
3.
M
that,
is
if
each
X a ,a
e
M, has
a countable base for
its
open
{X a ,Tr\
a has a countable base. countable, then lia.A/ is an inverse system indexed by M, each a has a
X
countable base and
M has a cofinal subsequence, then the limit X
X
a
has
a countable base.
4.
An
hibit
an inverse sequence
inverse limit of compact connected spaces is connected. Exof connected spaces with a disconnected limit.
B. LIMITS OF GROUPS.
G be a group and {G a a system of subgroups such that U G = G and for any G* fi* there is a G y with G" C G\ G* C G\ Show that G is naturally isomorphic with the limit of the direct system
1.
Let
}
a
j(r",7rf
2.

whore the
?rf
are inclusions.
)
be a compact group and (<7 a a system of (closed) subsuch that C\ G a is the identity and for any G a ,G there is G y groups  G/G a and for each G a G let G a C\ G Let // with G\
Let
ft
G
C
.
ft
,
C
ft
7T0:
be the natural map. Show that (7 is naturally isoa to the limit of the inverse system \H a ,Tr p\. morphic 3. Show that the above results remain valid for nonabelian groups,
a
>
H
lift
provided the definitions of Chapter vin are suitably interpreted.
C.
1.
DIRECT SUMS AND PRODUCTS.
Let
>
r
(7,7r),
(7," 7r),
7
y a
e
<f>:
same directed
for each
set
a
11
e
A/,
M. Let y the maps
.
<f>
:
F be direct systems of groups over the y y > {Gjir} be maps such that, G, w\ a y G G a form an injective representa\
>
tion of
y
,<*>.
G
7/T
Cr
as a direct sum.
__> k
,!*>
Show
the
y
same holds
{G,ir\
>
{
for the limiting
y
maps
<p
.
Cr
2.
In the above assume that
y
p:
G,
y
ir}
and that the
>
maps
3.
4>:
G"
>
y
G
a
yield a projective representation of G* as a
direct product.
Show
that the
same holds
1
for
V
G
y
G
.
Establish propositions similar to
and 2
for inverse systems.
D. SPLIT EXACT SEQUENCES. DEFINITION. An exact lower sequence
{(?,</>}
and a sequence
of
homomorphisms
230
EXERCISES
^3,
+ 103, +
2
such that
is
the identity
map
of
(7 3lf2
for each
i,
is
called a
split exact lower sequence.
n
ss 1
mod
3) are called
The homomorphisms $* (defined only for the splitting homomorphisms of the exact
is
sequence. A map of a split exact sequence into another one in the obvious fashion.
1.
if
defined
Show
and only
(a) 03.
if
that {0 n and {^ 3t + yield a split exact lower sequence the following conditions hold (for all i):
}
1 j
=
0,
== 0, (b) 03* + 103* + 2
(c)
^3 .+i03t+2
=
identity,
>
(d) 03. n:
63. 41
(? 3l+1
G 3t and
^3. + r.
G3
+
i
>
G
3
,+ 2
yield a projective
representation of
2.
as a direct sum.
in
Let
(S,7r)
be a direct [inverse] system
which each S
is
a
split
exact sequence.
3.
Show
that the limit
is
also a split exact sequence.
Reformulate the above results for upper sequences. > A a retraction. Show that Let (X,A) be a pair and r: the homology sequence of (X,A) together with the homomorphisms r+ form a split exact sequence.
4.
X
E. pADIC GROUPS AND SOLENOIDS. n 1. Let G n be a cyclic group of order p with generator g n
.
Consider
the inverse system
01
02
0n
(?j
<
<
GI
<
G2
gn
.
<
On
<
G
is
n
+i
4
where
<t>
n (g n+ i)
=
is
The
inverse limit
G,
called the padic group.
Show
it
compact, totally disconnected, and perfect. Show that has a countable base for its open sets, and, therefore, is homeomorphic
that
(?
with the Cantor
is
set.
Show
that for each integer n the subgroup p G<
of
n
open and closed and yields a fundamental sequence
Let
\z\
neighborhoods
all
of zero.
DEFINITION. numbers z with
S be
1.
l
=
l
the multiplicative group of Consider the inverse sequence
l
complex
s < s <...< s < s 4l
l
...
where 0(2)
2.
=
z*.
The
Show
that there
zero,
the addive group R and the image of ^ is dense in 2 P 3. Show that the padic solenoid contains a subgroup isomorphic with the padic group.
\l/
.
S p is called the padic solenoid. a continuous homomorphism \l/: R > S p of is of real numbers into 2 P such that the kernel of
inverse limit
is
4.
In euclidean 3space R* consider the disc
D =
{(x l
+
2
2)
+
EXERCISES
xl
231
^
1,
#3
=
0)
and the
solid torus
T
the o^axis.
In
D
obtained by revolving
D
around
consider the disc
Do
= M*i +
\\
o) 1
+
(i
xl
^
<?,
z3
Z)
,
=
0^
}
where
c
< ~sin2,
p
center
and the
discs
D, obtained from
by revolving
D
around
its
by the angle (2vi)/p
=
1,
p
,
1).
The
choice of c insures the
Now assume that as disjointness of the discs QJ Dj,,^ revolves around the o^axis it also revolves around its own center in
}
,
D D
D
such a fashion that, as one revolution around the Aaxis
}
is
complete,
,
D p i Do becomes D ly D becomes D, etc. Then the discs Z> ,/)i, sweep out a torus T which runs p times around the inside of the torus T. To describe the situation arithmetically, represent T by means of pairs of complex numbers (z,s),J2 ^ 1, s = 1 (z describes a point on D and s describes the angle of revolution around the x r axis). Then
l
define a
mapping
0:
T
T by
where
solenoid
<
is
c
<
} sin */p.
Then T
1
=
homeomorphic with the inverse
6(T). Show that the padic limit of the sequence
which whore
in
turn
is
6T =
Let
homeomorphic with the ~
intersection
PC.
n
Q
T
T,
0T =
n
l
0(6
T).
F. LIMITING
1.
GROUPS FOR INFINITE COMPLEXES.
K
complex. C\ L;G) ordered by inclusion. Show that the groups Q (K a ,K a a ,K a C\ L;G)] together with the homomorphisms of these groups [H"(K
}
be a (possibly infinite) simplicial complex and L a subbe the family of all finite subcomplexes of Let {K a
K
H
induced by inclusions form a direct [inverse] system of groups. The c resulting limit groups are denoted by # a (/C,L;G) [// (/f,L;G)] and are
called the direct limit homology [inverse limit cohomology] groups of (K,L)
(cf. vi,
Exer. B).
tf
2.
H
A
Show thatJ/
(/v,L;G)
and
H
_^
Q
(K,L]G) are isomorphic.
if
Show
that
Q
(K,L,G)
and H'(K.L;G) are isomorphic
G
is
compact or a vector
space over a field. 3. Let be a locally
K
finite simplicial
is
complex, and
if
L
subcomplex
Ka
of
K
called counterfinite
Ka
contains
a subcomplex. all but a
finite
number
of simplexes of
K.
Show
that with a suitable ordering
2S2
of the counterfinite
Q
EXERCISES
the groups 3C q (K,L a \G) a subcomplexes Exer. B) with suitable homomorphisms in(see vi, duced by inclusions, form an inverse [direct] system of groups. The resulting limit groups are denoted by 3C, Q (K L' G) [3C*(K,L;G)] and are
,
K
U K
[3Q,
(K,L \J
K
a ;G)]
t
9
called the inverse limit homology [direct limit cohomology] groups of (K,L).
4.
Show
thatJJC'(#,L;G)
m
and
3C'(#,L;(?) are isomorphic.
if
Show
that
3C q (K,L;G)
and
3C Q (K,L ,G) are isomorphic
G
is
compact or a vector
space over a
field.
CHAPTER
IX
theory
The Cech homology
1.
INTRODUCTION
In this chapter the Cech homology and cohomology theories are
defined and the axioms for such are verified.
of the results has been greatly increased
The degree of generality the use of modifications by introduced by Dowker [Annals of Math. 51 (1950), 278292]. The Cech cohomology theory is defined on the category ftj of arto be
The coefficient group G is taken Q an Rmodule for any ring If, and the resulting H (X,A) are in g*. The axioms are verified without exception. Cech cohomology groups
bitrary pairs (X,A) and their maps.
with compact coefficients are not defined.
are defined under the same circumstances Cech cohomology groups. Further, if (X,A) is a compact pair, then the homology group H Q (X,A) is also defined for G e gc, and is The axioms are valid without restrictions except for the itself in g c Exactness axiom, which is valid only after drastic restrictions. The homology sequence of any pair is defined, and it is proved that the composition of any two homomorphisms is zero. To obtain the full Exactness axiom, we must restrict (X,A) to be a compact pair and G must be either compact or a vector space over a field. In case (X,A) is triangulable, exactness is proved without this restriction on G. This is
The Cech homology groups
as the

natural
accomplished by showing directly that H 9 (X,A) is isomorphic in a way to the group based on the chains of a simplicial division.
The
axiom
is
failure of the
Cech homology theory
to satisfy the Exactness
directly traceable to the facts established in vm,6 concerning the interchangeability of factorization and inverse limits.
2.
COVERINGS, NERVES, AND PROJECTIONS
DEFINITION 2.1. An indexed family of sets in a space X is a function a defined on a set V a of indices such that, for each v e V a a, (the value of a on v) is a subset of X. If X = (J a,, then a is called a covering of X. It is called an open (closed) covering of X if each a, is open (dosed) in X. The set of all open coverings of X is denoted by Cov(X)
,
f
234.
THE CECH HOMOLOGY THEORY
is e
[CHAP. IX
for v
a subset of X, and V* is a subset of V a such that A \J . F, then we say that a is a covering of the pair (X,A) with (V a ,Va) as indexing pair. The set of all open coverings of (X,A) is denoted by Cov(X,A). One should be careful not to confuse the sets Cov(^T) and Cov(^T,0). However one may regard Cov(X) as the subset of Cov(X,0) consisting of coverings a indexed by (F a ,0). DEFINITION 2.2. Let a be an indexed family of sets in a space X. Let s a be the simplicial complex consisting of all simplexes whose vertices are elements of V a (if V a is finite, then s a is itself a simplex). If s is a simplex of s a the carrier of s, denoted by Car a (.s), is the inThe tersection of those sets a?, which correspond to vertices v of s.
If
,
A
C
nerve of a, denoted
by
X
aj
is
the subcomplex of s a consisting of
all
If a is a covering of the pair simplexes with nonvacuous carriers. (X,A) indexed by (F a ,F), then we denote by A a the subcomplex of such that A C\ a consisting of all simplexes s with vertices in V
X
Car (s) T* 0. The pair (X a ,A a ) is then called the Note that a, is the carrier of the vertex v.
nerve of a.
LEMMA
PROOF.
2.3.
one of s, it follows that each term of the intersection defining Car a (,s') is a term of that defining Car a (s). It follows from 2.3 that each face of a simplex of the nerve X a is
s' is
// ,s' is a face of Since each vertex of
s,
then
Car(')
3
Car a (s).
also a simplex of
X
a
.
Therefore
If/:
X
a is
 (F,7?) and ft is a covering of (F,#), (X,A) l then f~ fi is the covering a of (X,A) with the same indexing pair, B (V a ,Vi) = (Vfi,V fi), and defined by a, = /"'(ft) for each v e V,. It follows from the continuity of / that, if ft is an open (closed) covering,
DEFINITION
2.4.
l
a simplicial complex.
so also
is
f~
ft.
LEMMA
X
a is
(X,A) a subcomplex of Y and A
// /:
ft
2.5.
>
(Y,B) and a = J" ^, then the nerve is a subcomplex of B The inclusion
1
. .
map (X a ,A a )
simplex
of
denoted by f This follows from the fact that /(Car(s))
>
(YB ,Bp)
Sp.
is
ft
C
Car^(s)
for each
s of s a
(X>A) * (X A) is the identity, and a is a covering a and f a is the identity map of (X ay A ). LEMMA 2.7. // /: (X,A) > (F,B), g: (Y,B) ~> (Z,C), 7 w a = g y fp covering of (Z,C), and ft = g' y, then f~*g~ y = (gf)~ y and (gf) y DEFINITION 2.8. Let a and ft be two coverings of (X,A). The covering ft is called a refinement of <* (notation: a < /?) if every set of ft is contained in some set of a, and every set of ft indexed by an element of V $ is contained in some set of a indexed by V*. If a < ft, a function A (Vff V f) (F ,F) is called a projection if a p D /3 for each p:
2.6.
LEMMA
}
///:
l
9
(X A),
then f~
a
=
l
l
l
.
,
r
2]
COVERINGS, NERVES, PROJECTIONS
$.
235
v e
fy
>
The
which
vertex
is
s
also denoted
mapping p extends uniquely by p.
relation
to a simplicial
map
a.
LEMMA
ft
2.9.
a.
The
<
is
a quasiorder,
i.e.
a
<
a,
and
<
<
7 implies
<
7.
This follows from the corresponding properties of the relation C. Note that a < and ft < a do not imply a = ft. For example A a < 13 and /3 < a holds if A = V a = Fj = and both a and ft have a set
=
X.
2.10.
LEMMA
PROOF.
a,
ft ft
The
2.9
set
Cov(X,A)
it
of open coverings of
(X A)
y
is
a
directed set with respect to the relation
<
is e
.
By
and vm,2.1
e
Cov(X,/l), there exists a 7
Define
v
< =
7.
(vi,v 2 )
where
Vy = Va X v e Va v s
l
,
2
prove that, for any such that a < 7 and Cov(X,/l) A A If v e 7 T then V, and V* = V a X V Define TV
ft
.
sufficient to
,
7,
It
is
=
a fl
H
0...
clear that
7 has the desired properties.
of
a,/3.
Note that the covering 7 is a "smallest" common refinement That is, if 6 is any common refinement of a,/3, then 7 < 6.
LEMMA
If P
pp'\
:
2.11.
s
>
>
6V
For any P /: s y
a, the identity
>
map
sa
>
s a is
a projection.
fy are projections,
then their composition
suffices
sy
s a is
The
first
a projection. statement follows from a,
3
a,.
For the second
it
to note that
a ptt
.
9
D
/3 P >,
D
7, implies
a pp <,
3 7,.
LEMMA
p:
$0
>
sa
nerve of
(3
If a < ft are coverings of (X,A), then a projection This simplicial map of the nuips (X^A^) into (X a ,A a ). into thai of a. is also called a projection and is denoted by the
2.12.
same symbol p. PROOF. Since a
j)9
D
/?
any intersection
of /3/s
is
contained in
the intersection of the corresponding sets of a. Hence, for any face s Car (p(s)) Consequently if the Car (s) is nonvacuous of Sfi, Car^(s)
C
/3
(meets
/I),
so also
2.13.
Car a (p(s)).
If a
THEOREM
(see vi,3.1).
<
ft
projections p,p'\
(X
s
,Ap)
>
(X a ,/l a )
are two coverings of (X,A), then any two are contiguous simplicial maps
PROOF.
vertex v of
in
Let
s,
be a simplex of
e /?,.
ot p > 9 .
X
fty
and
/3 f
let
x
r
e Car/3(5).
Then,
for
any
lies
we have x
Since
C
A
.
<* P ,
and
sa
ft,
C
P ',,
^
both a pv and
two images of and p'(s) are faces of s' in X a The same argument shows that
.
Hence the simplex s of the vertices of s has a nonvacuous
If s is in
s' is
ft ,
spanned by the carrier. Thus, p(s) choose x e A Cards').
O
in
Aa
COROLLARY
2.14.
Using homology and cohomology groups of
COTTI
23*
THE CECH HOMOLOGY THEORY
we
have, for
[CHAP. IX
plexes in the sense of vi,3.9,
any
coefficient
group
(?,
that the
homomorphisms
induced by a projection (Xp,Ap)
of the projection
(X a ,A a )
are independent of the choice
and are
// /:
therefore uniquely associated with the relation
a
<
ft.
LEMMA
(Y a ,B a )
the
2.15.
(X,A)
ft'
>
l
(Y,B), then, if a'
map
Ifp: (Y ,Bp) > is a projection, then p maps (Xp^Ap.) into (X^^A^.). If p' is so defined by p, then p' is a projection and commutativity holds
f~ a,
f~
ft,
=
l
=
(Y,B), and a we have a' <
<
ft'.
ft
are coverings of
ft
in the diagram
P'
(X a .,A a .)
f.
<
(Xf .,A f .)
P
PROOF.
projection.
By
definition, Vp>
ft v
= V
and, for each
l
v e Vp, p(v)
=
p'(v).
f
Therefore a v ,
(X a ',A a *).
j8j. V9 f~ 0,, i.e. a' Thus, p is a implies /"* It follows from now 2.14, that p' maps (Xp*,Ap*) into Commutativity in the diagram follows from the fact
D
D
D
that f at fp are inclusion the proof.
maps
(2.5),
and p defines
p'
.
This completes
Beginning with 2.4 the discussion of this chapter was limited to coverings of pairs. However the definitions and results may be duplicated for coverings of spaces (without a distinguished subset) and to indexed families of sets which are not necessarily coverings.
3.
THE CECH GROUPS
We are now prepared to define the Cech homology and cohomology groups of an arbitrary pair (X,A) over any coefficient group G in 9 The homology groups of complexes, used in the definition, are those
defined in vi,3.9.
DEFINITION
nerve
(2.2),
3.1.
ings of (X,A) (see 2.10).
Let Cov(Z,A) be the directed set of all open coverFor each a e Cov(X,A), let (X a ,A.) be its
and
let
H Qta = H (X
Q
a
,A a >G),
H* = H\X a ,A&.
3]
CECH GROUPS
<
ft
237
For each relation a
7T a
I
in
Cov(X,A),
>
let
H
qt
p
H Qtaj
7T ftl
Ha
Hp
ft
be the homomorphisms induced by any projection (X^A ) (X a ,A a ) The collections {#,,* ,{#aVSJ are called ^e g th CecA (see 2.14). homology and cohomology systems of (X,A) over G. THEOREM 3.2. The q ib Cech homology [cohomology] system of (X,A) over G is an inverse [direct] system of groups defined on the directed set
PROOF.
TTa
is
By
2.11, the identity
the identity
map
>
of
H
map
q
,
a
.
(X a ,A a ) is a projection; hence Again, by 2.11, if a < ft < y, the
of
projection
(X T ,A 7 )
it prescribed projections follows that 71^0 = Tri for homology, and TT^TT ^ = TT" for cohomology. tb DEFINITION 3.3. The inverse [direct] limit of the g Cech homology [cohomology] system of (X,A) over G is denoted by q (X,A;G)
ft
(X a ,A a ) can be chosen as the composition (X ,Afi) (X a ,A a ). From this (X 7 ,A y )
>
of
>
H
Cech homology \cohomology] group of [H (X,A ;(?)] over G. The group G may belong to any one of the categories (X,A) 9, and by vm,3.14 and vm,4.12, the groups g (X,A;G) and H*(X,A;G)
is
9
and
called the #
th
H
are in the
same category
e
Q
as G.
9c the situation is as follows: Each of the cohomology groups H (X a ,A',G) is in g c and the groups form a direct system. Since the limit of a direct system of compact groups is not defined (see Remark vui,4), the cohomology groups for G e Q c are not defined. For homology, the Cech system is an inverse system and the passage to the
For
(?
Q (X a ,A a ,G) themselves are permissible. However the groups One not defined for G e Sc since in general the complex a is infinite. could try to avoid this difficulty by replacing the directed set Cov(X,A)
limit
is
H
m
X
consisting only of finite coverings (i.e. coverThe definitions and results of the preceding ings a with V a finite). f section remain valid for the directed set Cov (X,A) so that 3.13.3
by
its
subset
Cov f (X,A)
H
could be repeated with Cov(X,A) replaced by Cov (Jf,A). Of course, f the resulting limiting group may not be isomorphic with q (X,A\G) However, if the pair (X,A) is compact, then (see 9 (X,A;G) (see x,9).
r
H
below)
Cov f (X,A)
vin,3.16 the limits
we may
is
a cofinal subset of Cov(3f,A) and therefore by Thus, for compact pairs, q are isomorphic. limit our attention to finite coverings, and thereby define the
is
H[ and
q
H
homology groups
also in g^.
H
(X,A\G) with
G
e
gc
.
Then the group
H
Q
(X,A;G)
The
group Hi(K,L;G)
situation resembles the one of Chapter vi where the is finite, for G s 9c was defined only when
K
To
justify the above discussion
w e need
r
2S8
THE CECH HOMOLOGY THEORY
LEMMA
PROOF.
3.4.
[CHAP. IX
set
// the pair (X,A)
is
compact, then the
Cov f (X,A)
consisting of finite coverings is
(V a ,V*).
U C V,
WC
a cofinal subset of Cov(X,A). Let a e Cov(X,A) be a covering indexed by the pair Since both and A are compact, there exist finite subsets Vi such that
X
VtW
Then the covering a
Since
ft
ft with indexing set (U TF,T7). the proposition follows. In the definition of Cov(X,A), the indexing pair (F ,y) for a
defines a covering
U
is finite
and a
<
0,
ft
covering a is any pair of sets. Since the set of "all" sets involves one of the usual logical difficulties, the same is true of Cov(X ,yl). We shall
now show how
Let
y
M
this difficulty
be any
infinite set.
Cov(X A) consisting of all that the set has the same cardinal as
M
X
M
can be avoided. Consider the subset Cov M (X,A) of M. We now observe coverings a with V a
C
M
,
and
therefore, as in
2.10,
the space has a base for open sets of cardinal power <a(X). If and has power at least u(X), then the set Cov Af(X,A) is cofinal in Cov N (X,A). PROOF. Let a e Cov N (X A) and let be a family of sets in X, indexed by the set M, and such that ft m m e M, runs through a base for the open sets of X. Let V y (V*) be the subset of consisting of all m s such that ftm a* for some v e Y (v e F). Then the indexed family ft defines a covering y with (V y ,Vy) as indexing pair and a < 7. Since y e Cov M (X A) it follows that the latter set is cofinal in
we can prove that Cov M (X,A) is a directed set. LEMMA 3.5. Let w(X) be the least cardinal such that
MCN
M
X
1
f
,
M
M
C
ft
9
t
The homology and cohomology groups
of
(X,A)
may now
be defined
as limits of appropriate directed systems defined on Cov M (X, A) where is a set of cardinal power at least co(X). This definition seems to depend
M
on the choice of M. However, if is another such set, then by 3.5 both sets Cov M and Cov^ are cofinal in the set Cov^^. This yields an isomorphism of the limit groups defined using and those defined
N
M
using the set N.
It is easy to see that these
transitive system of groups as defined in 1,6. groups may be regarded as independent of the choice of the set
isomorphisms define a Consequently the limit
M.
We have carried out the discussion for a single pair
the
(X,A); however
co(X),
#
fl
any admissible category a such that the cardinals a have an upper bound. THEOREM 3.4. (Dimension axiom). If P is a single point, then  H q (P',G) forq^Q and H (P',G) ~ G  #(P;G). (P;G) 
same
applies to
X
e
4]
INDUCED HOMOMORPHISMS
Then
239
PROOF.
itself.
the set C"
C
Let a be the covering of P consisting of the single set P a. is a refinement of any covering of P. This means that Cov(P) consisting of the single element a is cofinal in
inverse system
Cov(P).
The
H qa
,
a
e
C',
has but a single term.
By
the definition of inverse limit (vm,3.1), the inverse limit of the single term Qa is itself Qa By vm,3.16, it follows that qa is isomorphic
H
H
.
H
to 7/ a (P;G). result follows
The nerve P a consists of a single now from vi,3.8 (see end of vi,4).
4.
vertex.
The
desired
INDUCED HOMOMORPHISMS
(X,A) > (F,B),
>
1
THEOREM
a.
4.1.
Let f:
let
f'
:
Cov(Ffl)
of
Cov(^,A) be the associated map of e Cov(Y,B) let f a (X a ,A a *)
:
the coverings (2.1, 2.4), and, for each
(F a
,
a)
be the inclusion
map
the nerve of
a
l
f~
a
into that of
a
(see 2.5).
Then
the
induced homo
rnorphisms
f a +:
[/*:
H
q
(X a .,A a .;G) >
II q (Y a ,B a ;G),
9
H'(Y a ,B a ;G) 1
H
(X a .,A a .;G)]
for all a
e
Cov(F,7?) together with f' form a
map
$(/) of the
th
<?
&ech
homology [cohomology] system of (X,A)
[(Y,B)] over
G
into that of (Y,B)
PROOF.
vin,2.3, 4>(/)
By
is
l
2.15,
/
if
is
order preserving.
Therefore, according to
in the
a
map
commutativity holds
diagram
where a
defines
Cov(F,). By 2.15, a projection p: (Y ,B ) * (F a ,B a ) > Hence / a p' = p//j. a projection p': (X^,B 0f ) (Jf a ,,A a ,).
<
ft
in
ft
ft
Therefore, by vi,3.8, / a *P*
DEFINITION
4.2.
= pje* By 3.1, ?rf = p^ and irf = p^. th The limit of the map $(/) of the g Cech homology
f
[cohomology] system of (X,A) [(r,B)].into that (K B) [(X,A)] noted by
is
de
and
is
called <Ae
homomorphism induced by
If
f.
The
is
coefficient
group
G
belongs to one of the categories 8*; then / #
and /* belong to the same
compact, then
category (see vin,3.14,4.12).
the pair (X,A)
we
HO
THE CECH HOMOLOGY THEORY
7
[CHAP. IX
replace Cov by Cov throughout, and as above derive the definition of f+ (homology only) for G in the category g<>
THEOREM 4.3. (Axiom 1). // /: (X A) (X A) is the identity, then f^ and /* are identities. It is only necessary to observe that <(/) is the identity.
t
9
THEOREM 4.4.
(Z,O, then (<7J% PROOF. It is
l
= gj^
(Axiom 2). ///: (X,A) and (gf)* = /V1
:

(Y,B), and
g:
(F,B) >
Cov (Z,C) > Cov(X,4) is the If a s Cov(Z,C), j8 = g"*a, and 7 = f~ ft = composition of 0" and /~ = (<//) a because all three maps g a (Y^Bp) a f0 (9f)~ <*> then (Z.,C a ), /* (*T,4,) > (y,,B,) and ( /) a (X YI A Y ) > (Z a ,C a ) are ? inclusion maps (2.5). Therefore $(<//) is the composition of &(g) and
clear that (g/)"
l
.
1
l
\
:
$(/) (see vin,2.3). The desired result follows now from the functional properties of the limit homomorphism (vm,3.14,4.12).
6.
THE HOMOTOPY AXIOM
Homotopy axiom
in the alter
It will be convenient to establish the
native form (Axiom
5')
given in
i,3.
THEOREM
ani/ coefficient
5.1.
be defined by g Q (x)
=
The
its
group G for which the appropriate Cech groups are defined. proof will be preceded by several definitions and lemmas. LEMMA 5.2. // a is a finite open covering of I by connected sets, then
(Homotopy axiom). Let g^g^\ (X,A) > (X,A) X / = (x,l); then = ^*/or (x,0), g^x) ^o^= flfi* a^ 0o
nerve I a is acyclic (see vi,5.6).
PROOF. shall first reduce the general case to the case when no inclusion holds between the sets of the covering. Indeed suppose Let ft be the covering ob.i <*, for some two indices v^v 2 e F
We
C
.
from V a Then a < ft and ft < a. > > Let T^. 1 a > I a and 7r 2 / /$ be projections. Then w^ii Ip /^ > / and ir 2 iri: / are projections and therefore induce identity maps on the homology groups. It follows that I a and I have isomorphic homology groups. Since the covering ft has fewer sets than a, it follows by induction that we can limit our attention to coverings for which no inclusion holds. If a is such a covering, an order VQ < Vi < v n of the elements of V a can be chosen so that the left end points l and the right end points r of a ti satisfy
tained by removing the element
:
Vi
.
t
t
= k ^
li
< k <

<
I*,
r
< n <
>

<
r n _,
i
g
0,
rn
=
,
1.
Consider the simplicial maps
/,:
/
/
/
defined for
=
n by
/<(<)
/.(^,)
=
for
jf
== v<
for
j
g ^
t,
t\
5]
HOMOTOPY AXIOM
t
241
let s be any simplex then f +i(s) = /(s). then /<($) has the form
t
We assert that /
of I a
If
.
+
i
and
/, are contiguous.
s
Indeed
l
If all
the vertices of
$
are less than v
+l,
some of the vertices of s'vt, and /,+i(s) is either
simplex with vertices
vertices of
s'
are larger than y f of the form s'v.v.+i or s Vi+i where
t ,
s'
t
is
a
<v
it
t
.
In the
/,(s)
first
case /,()
is
a face of /
i.
+i(s).
Thus we may assume that
are
=
r
s'y,,/, + i($)
=
is
s'v. +
Since
all
the
<v
t ,
follows that Cards')
I
a connected set with
t
end points
it
l,r
satisfying
g
lt
,
^
r,.
Since Cards') P\ a f< +
5^ 0,
follows that
we have
a tifl 7^ so that s'v i>, + is a simplex Consequently Car a (s') C\ a 9i both f (s) and f + (s). Thus /,+! and /, are contiguous. It containing follows from vi,3.2 that /, + !, = /,,. However f n is the identity map of I a while /o maps 7 a into the single vertex V Q Since / n<t = / it follows that / is acyclic. = DEFINITION 5.3. A covering a of 7 indexed by the set is called regular if the sets a are open and conn) (n ^ 0) (0,1,
t i
O
t
i
l
.
Ojt
,
N
,
t
nected, and
if
e
a
,
a,
Oa
a
t
Ononeai,
%
1
e
,
lnonea n _i,
0,
,
+
1
7^
for
i
Ha, =
a.
for
i
= <
n
j

1,
1.
LEMMA
PROOF.
5.4.
TTie regular coverings
form a
<f>
cofinal subset of Cov(7).
Let
be any open
covering of 7.
*
of
components
3>
of the sets of a.
Then
<
sets covering 7,
and by compactness
select a
Consider the family a family of connected open contains a finite subfamily 4>o
is
<!>!
of
covering 7. In covers 7. <t>! covering which
minimal subset
such that no proper subset
Then $
is
{
after suitable indexing constitutes a regular
a refinement of a. Let a e Cov(X,A) be a covering indexed by the pair (V a ,Vi). Suppose to each v s V a there corresponds a regular cover9 9 = (0,1, Consider the set of n'). ing ft of 7 indexed by 9 and let IF' be the subset consisting of pairs all pairs (y,z), v e V, i s The covering 7 e Cov(X X 7, A X 7) indexed by e 71 (v,i) with
DEFINITION
5.5.
N N
,
W
t;
(W,W) and
defined
by
7.,
=
a,
X
8J
is
called a stacked covering over a.
The
coverings 0' are called the
stacfcs
of 7.
LEMMA
5.6.
Stacked coverings form a cofinal subset of
Cov(X
X
7,
4 X
7).
242
THE CECH HOMOLOGY THEORY
PROOF.
Let
[CHAP. IX
each
For 6 e Cov(X X /, A X /) be indexed by (V h V*). open sets U(x () C X, V(x,t) C / such that U(x,t) X F(o:,0 contains (a:,/) and is contained in one of the sets ,, and if x e A, then U(x,t) X V(x,t) is contained in one of the sets 6. with v e V\. For each fixed x e X, the sets V (#,0 constitute a covering of /, which therefore by 5.3 has a regular refinement fi*. For each set p* there is an open such that U(x,i) X 0* is in one of the sets of $., and if set U(x,i) in x e A, then U(x,i) X 0t is in one of the sets 6, with v e Fl. Define Then a is a covering of (^T,^l) indexed by (X,A), a, = f\ U(x,i). and the covering 7 stacked over a with /3* as stacks is a refinement of 6. LEMMA 5.7. Ld y be a stacked covering over a e Cov(X). // the
(x,t) select
)
X
nerve
Xa
7,
is
a
(finite)
PROOF.
Without
let
simplex, then the nerve (X loss of generality we may
X
/) 7 is acyclic.
assume that none of the
d of
sets of the covering of
a and
and
a is empty. v @ be the stacks.
Let
V and
W be the indexing sets of
/ indexed by
Define a covering
W as follows:
Let
s
be any simplex with vertices
(f
,t'
),
,
(v n ,i u )
in
W.
Then
Car 7 (s)
= =
H n
Tc.,..,)
=
n
Sf ,
a.,
X
H
ff}
<*.,
X
H
f<l
=
is
na
B/
X
It follows that
Since
Xa
is
is
a simplex, the set r\,a,
if
t
not empty.
empty and only if Car a (s) is empty. Thus (X X /) 7 = / Since 7 6 is acyclic by 5.2, it follows that (X X I) y is acyclic. We shall abbreviate (X X /) T by X X 7 Y LEMMA 5.8. Let y be a covering stacked over the covering a. of (X,A) indexed by (V a ,V*). Consider the simplicial maps
Car 7 (s)
.
l,u:
(X a) A a ) *
(XXl
u(v)
y
,A
X/
f
7)
defined for v
e
Va
by
l(v)
=
(t,,0),
=
(i;,n ).
=
PROOF.
w,,
s of a consider the subcomplex C(s) of / 7 consisting of all simplexes whose vertices have the form (v,i) where v is a vertex of s. The simplex s is the nerve of a covering a' of and C(s) is the nerve of a covering 7' stacked over of. a subset X' of
and Z* = u*. For each simplex
X
X X
X
Thus by
if s is
5.7, C(s) is acyclic.
If s' is
a face of
.
5,
in
Aa
,
then C(s)
is
in
A X
/7
If c is
then C(s') C(s), and a chain in the simplex s,
C
6]
INVARIANCE UNDER EXCISION
l(c)
24*
then
and
u(c) are chains in C(s).
carrier for the
maps I and u and by
PROOF OF 5.1.
= u^ and I* = u*. l^ Let D be the subset of Cov(X X /, A X /) consisting
vi,5.8
Thus C(s) we have
is
a
common
acyclic
of stacked coverings. Since the Cech groups of (X /,
D
g
is
cofinal,
/).
we can
Let 7
e
use the set
D
to define
a.
X
A X
0*7,
D
be stacked over
Consider the coverings y
=
71
= 0^7
of OY,A)
and the inclusion
maps
g 0j
p l7
:
:
7 7 ), (X y .,A y .) > (X X I y (X 7l ,A 7l ) 4 (XX/ 7 A XI,).
,
,
AX
The map w defined in 5.8 may be factored into u = g ly u' where f (X a ,A a ) > (X Tl ,A Yl ) is an isomorphism defined by u'(v) (v,n ).
Observe that y n
u':
<
7i
and that the map
defined
by
7r(y,i)
=
(y,0)
is
a projection.
Further observe that
I
It follows
that
Since by
5.8, Z^
=
u^,
and
since
u^
is
an isomorphism,
it
follows that
Thus by vm,3.18 we have
larly gfy
=
gQ ^
=
g
}
+.
For cohomology we prove simi
ir*g$y
and apply vm,4.15.
6.
INVARIANCE UNDER EXCISION
_
f:
THEOREM
is
U
6.1. (Excision axiom). If contained in the interior of A
U
is
open in
X and
its
closure
C

X",
then the inclusion
map
(X
U,
A
/,:
/*:
U)
J? f
a
>
(X,A) induces isomorphisms
U,
(X
A // (J^
fl
U)
C7,
//.(*, A),
// (,Y,A)

A 
C7),
/or ani/ coefficient group G for which the respective Cech groups are defined. U. Let PROOF. For notational brevity let X' = U, A' = A
X
D
be the subset of Cov(A'",A) consisting of all coverings a indexed by such that (F a U T* 0, then v e V A and a, A. a (1) if a, By vin, 3. 15, the conclusion of the theorem is a consequence of the
,O
H
C
following three propositions:
(2)
D
is
l
cofinal in
is
Cov(X,A).
(3)
f~ (D)
cofinal in Cov(X',A').
244
(4)
THE CECH HOMOLOGY THEORY
If
[CHAP. IX
a
e
D and
ft
=
l
f~
a,
then
(2) consider any covering a of (X ,A) with indexing pair Let F' be a set disjoint with V a and in a 11 correspondence with F. For each v e F a the corresponding element of F' will be denoted by v'. Consider the covering 7 of (X,A) indexed by (V a V, Vi \J V) and defined as follows
To prove
(F a ,F).
,
U
7. Tf
,
= =
,
,
U
for
v e
v' e
Fa
F'
,
n Int A
7
is
f
for
.
Since
f/
C
e
Int A,
s
it
follows that
a covering of (X,A).
Clearly
a
<
7 and 7
D.
consider
To prove
Define a
any covering ft of (X',A ) indexed by (F^FjT). Cov(X,A) indexed by the same pair (F0,F ') as follows:
(3)
.
=
0.
U
C/.
Then
ft
=
l
/"'a.
Choose 7
e
# so that a <
view of
,
7.
Then
ft
=
f' a
l
<
l
f~ y,
so that f~
D is cofinal in Cov(X',A').
(4), it suffices, in
To prove
(5)
vi,3.5, 3.6, to
prove
.
Xa
is
= Xi U A a
AJ =
JTJ
n
Aa
Since (X^,A^)
clusions
a subcomplex of
(X a ,A a )
(see 2.5),
we have
the in
X DX',V A
a
a,
AlCX' r\
ft
A..
It
(6)
thus remains to prove
Let
s
be a
X a CX'p\JA a AJDXfnA.. not in X' Then simplex of X a which
,
is
p
.
Car(s)
Car(s)
9^
and Car a (s) Pi X'
therefore, since
=
Car^(s)
=
s
0.
This implies that, for every vertex
v of
Consequently s, we have a,
Since
C7
.
^
H
C
0,
U.
U ^
and
a
e Z),
that v
is
F.
Aa
C
Car(s) C\
part of (6).
A
5*
so that s
a simplex of
.
Aa
A, it follows that This proves the first
Let
are in
s
be a simplex of Xp C\
It follows that the vertices of s
F
and that
Car(s)
H X'
A'
=
Car,()
*
0,
Car a (s) C\
A ^
0.
If
Car a () CST', then
)
=
Car a (s)
O X' H A
=
Car a (s)
nA
^
7]
BOUNDARY OPERATOR AND EXACTNESS
is in
245
follows that
and s a9 C
A'
.
ft
If
Car(s) r\
1
U ^
s.
0, then, since
a
is
in D,
it
A
for every vertex v of
Thus Car a (s)
CA
6.1.
and
Cai>() r\
so that
s is in
A = Car a (s)
.
n A H JT
= Car a (s)
H X' ^
0,
A'
ft
This concludes the proof of
7.
THE BOUNDARY OPERATOR AND EXACTNESS
The Cech homology and cohomology groups of (X,A), A, and are defined as limits of suitable systems of groups defined over the directed sets Cov(X,A), Cov(A,0), and Cov(X,0) respectively. In order to
define the
X
boundary operator and discuss exactness, it will be convenient to have equivalent definitions in which all these systems are defined over the same directed set. It appears that the directed set Cov(X,A)
is
most suitable DEFINITION
S0
>
>
for this purpose.
7.1.
If
a
a
e
Cov(X,A),
a
let
Sa
If
a
[S
]
[cohomology] sequence of
IT*'
(X
>
,A
a)
over G.
be a
a
<
is
be the homology in Cov(X",A), let
called the adjusted
Sa
(X
[TT:
(Xfi,Ap)
a ,A a ).
S The
S*]
map
induced by a projection
resulting limit sequence
homology [cohomology] sequence of (X,A). phisms of the adjusted sequences are written
...
The groups and homomor

7/.(X,A)
X  H (X)
a
(X A
.
,
^
'*
ii
ff.(A) (X M>

*
H.. (X,A) <t
...

j'*
H'(A',A) * H'(X)< X
.
A
,

&'
H(A) <X
.
A)
//'
+1
CY,A)

...
To compare the groups with the subscript (X,A) with the groups without this subscript, we introduce two maps,
0:
*:
Cov(X,A) * Cov(A,0), Cov(Z,A) > Cov(X,0).
<f>a is
Cov(X,A) be indexed by the pair (V a ,V*). Then by (^a,0) and (<), = A P\ a, for y e y. The covering by (Va,0) and satisfies (^a), = a,. Observe that
Let a
e
indexed
\f/a is
indexed
Aa =
The maps
<t>
Aj a
,
X
a
Xj, a
.
and ^ and the appropriate identity maps
of the
homology
groups yield maps of inverse systems
$:
{7/,(A a ;G),7r^
M
(
,o)
*:
H a (X tt<?),T5}
x.o)
> >
a
{II 9 (A a ;G),* ,\ {X .A
a
.
}
{// a (X a ;G),7r , (I A)
where the subscript indicates the directed
set
on which these inverse
846
THE CECH HOMOLOGY THEORY
The
inverse limits of the
[CHAP. IX
<l>
systems are defined.
maps
and
^
.
are
homomorphisms
0.:
HM\G)  H
4>
q
(A;G) (X
.
A}
,
*:
ff.(X;G) >
H
Q
(X;G) (X
A)
.
For cohomology,
are
and
^
are
maps
of direct
systems and their limits
LEMMA
phisms.
7.2.
T/ie
homomorphisms
0o>,
^,,
</>",
and ^"
are isomor
PROOF. image of
In view of vm,3.15 and vm,4.13, it suffices to show that the a cofinal subset of Cov(Afi) and that the image of ^ is a cofinal subset of Cov(X,0). Let a e Cov(A,0) be indexed by the pair (V,W). Let V* be a set consisting of V and a single element V Q not in V. For each v e 7 select an open set , in so that a, = A /3 f = X. Then e Cov(Z,A) and has (7*,F) as indexing set. Define f . The covering 4>{l agrees with a but has (F,0) as indexing set, thus
<t>
is
X
O
.
and the image of is cofinal. Then a defines a covering e Cov(X,0) be indexed by (V,W). The covering $$ then agrees with 8 e Cov(X,A) indexed by (V,V). a but is indexed by (F,0). Thus a < \f/p and the image of ^ is cofinal. DEFINITION 7.3. The homomorphisms
<t>&
<t>
a
<
Let a
d:
H (X,AG)
Q
*
H^AfT),
d:
H\A]G)
are defined from the diagrams
d'
9 <
H.(X A;G) > ff^UjCO^.x, < H.
as a
=
0;^' and
6
= ^(O"
(4rwom
1
*
THEOREM
(/A) # d
7.4.
3).
Le^
/:
(X,A)
=
df* and 6(/A)*
=
/*5.
is
PROOF.
The formula
for
homology
an immediate consequence of
the commutativity relations in the diagram
d'
H
Q
(X,A;G)
I/*
7]
BOUNDARY OPERATOR AND EXACTNESS
is
247
where (f\A)
tivity
definitions.
relations
defined as the appropriate limit map. The commutain each square are immediate consequences of the
of a pair (X,A) with the adjusted homology isomorphic
is
q
The proof for cohomology is similar. THEOREM 7.5. The homology [cohomology] sequence
any
coefficient
over
[cohomology] sequence.
group The isomorphism
G
is
(X,A;G) [H itself. We shall only carry out the proof for homology, the proof for cohomology is quite similar. In view of 7.2 we need only verify comQ
[0V"] and PROOF.
the identity
map of H
given by the (X,A\G)] onto
maps 0,^
mutativity relations in the diagram
H
Q
(A)
.
>
H.(X)
0.
H
Commutativity
the definition of
discuss the
limits of
d.
9
(A) (X
A)
(
>
H (X) (X A>
q
.
in the lefthand triangle is a direct
To prove commutativity
in the
maps maps
T,?/:
0^
Q
and
z^
in greater detail.
consequence of middle square, we These maps are
a
[H (A a ),7T0}
x,0)
{H Q (Xa),Tr
ft}
(X.A)
where the subscript is used to indicate the directed set over which the The map r carries each covering a e inverse systems are defined. indexed by (F a ,~T) into the covering TO. = 0a e Cov(A,0) Cov(X,A) indexed by the set (7^,0) and satisfying (TO), = A C\ ,. Further r > H induced by the inclusion A ra C X<* is the map H Q (A ra ;G) Q (X a ;G) The map q assigns to each covering a e Cov(X,.4) indexed by (V a ,Vi) the covering yet = i~ $a e Cov(A,0) indexed by (F,0) and satisfying
l
= A C\ a,. Again r] a is the map H q (A^ a \G) > H Q (X a ;G) in(tia) < ra; furthermore the induced by inclusion A^ a C X a Clearly a clusion map A ra C A^ a is a projection. It follows that ir\ a is the map H 9 (A ra ]G) > H Q (A^ a ]G) induced by inclusion. Thus we have
9
.
iy
rja
<
rot,
Ta
=
rjair*,,*.
Thus by vm,3.l8 we have
rm
=
rj
m,
i.e.
0^
=
i*^oo,
248
THE CECH HOMOLOGY THEORY
is
[CHAP. IX
The situation in the righthand triangle j$a> and j+ are limits of maps
T,U:
similar.
The maps
{#,(*
a ;G),7r ,} (X ,o)
* {#,(*, A a ;G),7r ,J
a
(
*,x).
For each a e Cov(X,A) indexed by (V a ,V*) the coverings ra, ija e Cov(X,Q) coincide with a but are indexed by (V a ,Q) and (V a ,V*) a are the ra = X, a = X a and both r a and respectively. We have > fl" Again we have c (X" a ,yl a ;G) induced by inclusion. map q (X a \G) > X a a is a projection of ra into yet. ija < ra and the identity map
X
?y
H
X
Thus again
so that T m
=
r?.
With the proof of 7.5 concluded, the question of the exactness of the homology and cohomology sequence is replaced by the question of the exactness of the adjusted sequences. The adjusted sequences are
however
set
Cov(XjA).
limits of systems of exact sequences defined over the directed Thus the results of viu,5 may be applied.
7.6.
For any pair (X,A) and any G cohomology sequence is exact while the homology sequence is a sequence of order 2 (see vm,5.2). // (X,A) is compact and G is either in g c or in g F (the category of vector spaces over a field F),
THEOREM
(Exactness axiom).
in a category
9>
the
then the homology sequence is also exact. PROOF. The first part of the theorem
is
a consequence of viu,5.3
and
If (X,A) is compact, then, in defining the groups ocvin,5.4. curring in the homology sequence, we may limit our attention to finite coverings. If G is compact, then, for each finite covering a, the homology
sequence of (X a ,A a ) over G is composed of compact groups and thereIf G is in g/? and is finite fore by vin, 5.6 the limit sequence is exact. dimensional (over F), then the groups of the homology sequence of
(X aj A a ) over F
limit follows
are
all
finite
If
from vm,5.7.
G
dimensional and the exactness of the e g^ is not finite dimensional, then G
may
sional.
be represented as a direct sum Gp where each Gp is finite dimenThis decomposition yields a decomposition of the homology sequence of (X,A) over G into a direct sum of the homology sequences over Gft. Since each of these is exact, it follows that the homology
(9.4) that the homology sequence of a triangulable pair is exact without restriction on the coefficient group. In x,4 we shall construct a compact pair for which the homology sequence
^
sequence over G is exact. It will be shown later
with integer coefficients
is
not exact.
8]
CLOSED SUBSETS
8.
249
CLOSED SUBSETS
It will be shown here that the reason for using coverings a e Cov(X,A) indexed by pairs (7 a ,V) in the definition of is due to the g (X,A) fact that A was not assumed to be closed. If A is closed, we can use the set Cov(X) consisting of coverings a indexed by a single set V a
H
.
DEFINITION
called proper
if
Va
8.1. A
is
A
covering a
all v
s
the set of
Cov(X,A) indexed by (V a ,Vi) e V with a, C\ A 7* 0.
e
is
LEMMA
where
8.2.
For each covering a
e
Cov(X) indexed by a
set
Va
,
con
sider the covering pa
V
is the set
of
all
Cov(X,A) defined by a and indexed by (V a ,V) v e V a with a v C\ A ^0. The map p: Cov(X) >
in
Cov(X,A) and the set of all proper coverings The proof is obvious.
is then
a 11 order preserving correspondence between Cov(X)
Cov(X,A).
LEMMA 8.3. // A is a closed subset of X, then the proper coverings iorm a coflnal subset of Cov(X,A). If is a T^space, then the converse
X
fs also true.
V
Cov(X,A) be a covering indexed by (V a ,V ). Let A and consider the covering be the set of all v E V a with a, C\ A 7^ indexed by (V aj V) and defined by Cov(X,A)
PROOF.
s
f
Let a
A
P,
= =
a,
a,
 A
for
v E v e
V a  V,
7'.
for
proper and a < /3. Let x e is a TVspace and A is not closed. __ Suppose now that = of the sets <*i = X, A A. Consider the covering a consisting 2  (x) and indexed by V a = (1,2), V A = (2). Let ft e Cov(X,A) a
Then
/3
is
X
X
be any refinement of a. Then Thus z e 0, for some v z Vp
j8 t
P\
A
7^
so that
is
X" (x). ^ we must have ft, follows that V p. Since 0. is open, it not a proper covering. Thus the proper cover
for v
r
e T
C
A
ings are not cofinal.
These lemmas show
that,
if
A
is
closed in
X, then,
in defining the
adjusted homology sequence of (X,A), \ve may replace the set Cov(X,A) on which the direct and inverse systems are defined by the subset consisting only of proper coverings. By 8.2, this set may in turn be
replace
by the set Cov(X) of coverings of X. For each covering a e Cov(X) we then consider the homology [cohomology] sequence of the pair (X a ,A a ) where A a consists of those simplexes of X a whose carriers meet A. This is precisely the nerve of the covering pa e Cov(X,A). The groups of the thus modified adjusted homology sequence will be
indicated
If
by the subscript X. the pair (X,A) is compact,
we
can, in addition to the
above
250
THE CECH HOMOLOGY THEORY
[CHAP. IX
Thus the homology changes, restrict ourselves to finite coverings. defined as a limit taken over the directed [cohomology] sequence can be
set
Cov f (X)
of finite coverings of
X.
9.
CECH GROUPS OF TRIANGULABLE SPACES
9.1.
DEFINITION
Let
T = of A
(X,A)
be
a
\t,(K,L)\. For every vertex A of in as defined in n,3.6 and define
K
pair with a triangulation consider the open star st(A)
K
TA
=
{A} = \K*\, V T = \L\. We shall say that r is the covering associated with the triangulation T. LEMMA 9.2. The nerve of (X,A) in the covering r is (K,L), i.e.
There
results a covering r of
(X,A) with
V =
r
A
X X
r
K and
PROOF.
A = L. The vertices
T
T
vertices of
are subcomplexes of K. Then, since
and of X r are the same; thus K and A n be distinct the same simplex. Let A,
of
K
,
t
is
a homeomorphism,
*(st(A'))
(i)
n
H
if
TA
.
=
n
if
if Pi st(A') 0, and, by n,3.7, are vertices of a simplex 5 of K.
It follows that
T A i 7*
if
and only
,
^
this holds
Thus
X =
r
and only K. From
A,
An
(1)
we deduce
that
n
,0
,
TA .
n
A =
n
A,
L
n
,
so that the simplex s with vertices A n are vertices of L, and A,
An
is
in
A
r
if
and only
if
(2)
n
0
it
L
Pi st(A')
^
0.
Now
follows directly from the definition of st(A')
where st/XA*) is the open star of the vertex A* constructed relative to the complex L. Thus, by n,3.7, condition (2) is equivalent to s being
a simplex of L.
Thus A T =
Let
L.
THEOREM
(X,A).
9.3.
T =
\t,(K,L)\
be a triangulation of the pair
(
The homology [cohomology] sequence S K,D of (K,L) (in the simplicial theory) and the adjusted homology [cohomology] sequence S of The isomorphism is obtained (in the Cech theory) are isomorphic.
9]
TRIANGULABLE SPACES
T and
T
r
,
251
by taking the covering r associated with
regarding
S (K ,D
maps
as the
homology [cohomology] sequence of
TT T
:
the nerve of r.
[TT
:
This yields
S\
S > S
r S >
which are isomorphisms. This result is valid for
all coefficient
groups
G
for
which the ap
propriate Cech groups are denned.
(X,A)
a
9.4. The Cech homology sequence of a triangulable pair for any coefficient group G. The proof of 9.3 will be preceded by a definition and two lemmas. is the DEFINITION 9.5. The mesh of a covering a of a metric space
is exact
COROLLARY
X
maximum of the diameters of the sets a,. LEMMA 9.0. A subset C of the directed set Cov(X)
of a compact metric space
set
of all open coverings
is
> the if for every contains a covering of mesh <e. PROOF. The condition is clearly necessary. To prove that it also Let e be the Lebesgue number sufficient, consider a covering a of X.
is cofinal if
X
and only
C
of this covering (see n,8.5)
and
1
let
ft
e
C
be a covering with mesh P
<
*.
Then a <
ft
and
C
is
cofinal.
LEMMA
9.7.
Let T'
=
{t ',(K'',L')1
be the barycentric subdivision of
r'
the triangulalion
ciated with 7
V
is
Then the covering {t,(K,L)\ of (X,A). a refinement of the covering T associated with T
T
'
T =
assothe
',
and
projections
TT T
:
S
T
.
~ S
rt
[irl
:
S ~ S
r
r
]
are isomorphisms.
PROOF.
We recall that
map
f
(KjL) be the linear
f (K',L ) = (Sd K, Sd L). Let I: (K',L ) * connected with this subdivision. Then t' = tl.
f
Any
vertex
,
B
it
of
K
is
the barycenter of
s.
A,
open
A
n
be the vertices of
follows that
K. Let Then, from the definition of I and the
s of
some simplex
stars,
Consequently
(1)
ri
=
t'sk(B)
=
tlst,(B)
=
O
tfl
tet(A')
=
H
0
T A <.
This implies that T' Let TT: (K',U)
TT
is
>
a refinement of r. (K,L) be a projection as defined in
to every vertex
is
vi,7,6,
i.e.
is
a simplicial
map which
B
of the simplex of
K
whose barycenter
B.
K' assigns a vertex Hence (1) implies that TT
of
252
also
is
THE CECH HOMOLOGY THEORY
a projection of the covering
T'
[CHAP. IX
last part of
into
r.
Thus the
the
the
lemma is a consequence of vi,7.2. = (%( n K, n L)} be PROOF OF 9.3. Let
T
n tb barycentric sub
division of the triangulation T, and let { r} be the corresponding sequence of coverings of X. It follows from u,7.5 that
n
lim
n+oo
mesh
n
r
=
0.
Thus,
they
if
we
n
regard
{
r} as coverings of
if
X, they form a
cofinal subset of
Cov(X).
will
Consequently,
we regard
C =
form a
By 8.3,
Cov(X,A). T Let S' be the inverse [direct] limit of the system {S T ,7r} [{$ ,7r)] T defined over the directed set C. The homomorphism w r [TT can then be represented as the composition
]
the set
R R is cofinal in Cov(X,A). Thus C is cofinal in
cofinal subset of the set
{V} as coverings of (X,A), of proper coverings of (X,A).
T
S where ^> [^]
*.
TT;
*.
TT'
5' >
ST
[S <
S' <
S
T
]
is the limit map of the injection of the system defined over Cov(X,A) into the subsystem defined over C. Since C is cofinal m is an isomorphism (vm,3.16, 4.14). in Cov(X,A), it follows that m ] In the system over C, all the maps are isomorphisms by 9.8. Thus
\l/
[\l/
vm,3.4 implies that irj: This shows that the map
10.
S'
in 9.3
ST
is
/T
[?r
:
S
T >
S'] is
an isomorphism.
an isomorphism.
PARTIALLY EXACT
seen, the
HOMOLOGY AND COHOMOLOGY THEORIES
As we have
cumstances.
larly for
except for the Exactness axiom,
Cech homology groups satisfy all the axioms which holds only in very special cirNevertheless the Cech groups are very useful (particu
compact pairs) because of various features that will be discussed in the following chapters. In later chapters we shall encounter systems other than the Cech homology theory which fail to satisfy the Exactness
axiom
in its present form.
It is therefore useful to generalize the con
cept of a
homology theory so as to include the the other theories that we shall encounter.
Cech theory as
well as
begin by modifying the axioms for an admissible category & (see 1,1) by replacing condition (5) by the following: (X,A) > (F,fi) (5') If (X,A) is triangulable, then (X,A) e d. If /:
We
and (X,A), (Y,B) are triangulable, then / e Ct. This axiom asserts that the category 3 of triangulable
category of a: this
is satisfied
pairs
is
a sub
by
all
the admissible categories that
we
have encountered.
NOTE
253
one:
is
Now
we replace the Exactness axiom by a weaker AXIOM 4' (Partial Exactness axiom). If (X,A)
admissible
,
the
homology [cohomology] sequence of (X,A) is a sequence of order 2 (see vin, 5.2). // (X,A) is triangulable, then the sequence is exact. A system // = [H g (X A) f^d\ [// = {ff<(X,A),/*,6}] satisfying axioms 13, 57, and 4' is called a partially exact homology [cohomology]
9
9
The Cech homology theory (on the category Cti) is a typical example of a partially exact homology theory. Others will appear later. In applying the results of Chapter i to a partially exact homology theory, we must check the extent to which the exactness axiom is used in the proof. For instance in proving that H g (X,X) = (i,8.1) the
theory.
proof makes full use of exactness: however, the alternative proof uses only the order 2 property of the homology sequence. Thus this proposition remains valid for partially exact theories. An inspection of
first
the proof of i,10.2 shows that the homology and cohomology sequences
of a triple are of order 2 even for a partially exact theory. Similarly, inspection of the proof of 1,8.6 shows that the reduced homology and cohomology sequences of a pair are of order 2.
NOTE
The development of the Cech theory. The first definition of homology groups of the Cech type was made by Vietoris [Math. Ann. 97 (1927), He restricted himself to compact metric spaces and used a 454472].
specific metric to define his cycles.
About the same time Alexandroff 30 (1928), 101187] introduced the concept of ap[Annals of Math. proximating a compact metric space by an inverse sequence of complexes (called:
numbers.
a projection spectrum), and successfully defined Betti Pontrjagin [Math. Ann. 105 (1931), 165205] added to this
the notion of an inverse sequence of groups, and obtained homology groups. It was Cech [Fund. Math. 19 (1932), 149183] who first defined the nerve of a finite covering by open sets, and used such complexes as
approximations to a space. By using inverse systems instead of seWhen it quences, he defined homology groups of arbitrary spaces. became clear that the Cech definitions for noncompact spaces did not
give a fully satisfactory theory, Dowker [Annals of Math. 51 (1950), 278292] found the satisfactory modification based on pairs of infinite
coverings.
parallel development has taken place based on an idea of Alexander [Proc. Nat. Acad. Sci. 21 (1935), 509512] that a gcochain may be 1 points in the space to defined as a function from ordered sets of q
A
+
the coefficient group.
He
subsequently modified this approach by in
254
EXERCISES
troducing the notion of a grating [Bull. Amer. Math. Soc. 53 (1947), Using another modification suggested by Wallace, Spanier [Annals of Math. 49 (1948), 407427] showed that the cohomology theory obtained from such cochains satisfies our axioms on compact
201233].
spaces and
this
is isomorphic there to the Cech theory. The advantage of approach lies in the simplicity of the definition of cochain. A disadvantage is that there appears to be no equally simple and dual con
struction of chains
and
their
homology groups.
EXERCISES
A.
CONNECTEDNESS AND QUASICOMPONENTS.
DEFINITION. Let component of X Q in
taining x
.
X
is
X
be a topological space and let x e X. the union of all connected subsets of
of z
is
The
conthe
X
all
The quasicomponent
the intersection of
.
simultaneously open and closed subsets of X containing X Q 1. Show that: (1) components are closed, connected, and disjoint; (2) quasicomponents are closed and disjoint; (3) eash quasicomponent is a union of components; (4) a quasicomponent is a component if and only if it is connected; (5) if X is compact or if the number of quasi
components is finite, then each quasicomponent is a component. Show by an example that a quasicomponent need not be connected. 2. Let X Q ,XI e X, g e G, and g ^ 0. Show that the elements (gx ) x H (X;G), as defined in i,7.1, are equal if and (gxj x of the Cech group and only if XQ and x belong to the same quasicomponent. 3. For each element h of the Cech group H(X\G) where G is in
l
* G as defined in the category Q R consider the function i,7.1c. Show that this establishes an isomorphism between //(X;G) and the > 72module of all continuous functions G, where G is taken with the
,
X
X
discrete topology.
4.
Show
that
H
Describe the group fi(X;G) in a similar fashion. Q (X;G) is isomorphic with Hom(/f(X;J);G) where
J
is
the group of integers.
B. 0DIMENSIONAL SETS.
ing of
is called Qdimensional if every open coverDEFINITION. A space has a refinement consisting of disjoint sets. 1. Show that a compact space is 0dimensional if and only if it is totally disconnected (i.e. each component reduces to a single point). 2. Show that every compact 0dimensional space is the limit space
X
X
of
an inverse system of
3.
finite sets.
Show that,
if
X
is
0dimensional, then
H
a (X,G)
=
0,
H
Q
(X]G)
=
for q
>
(Cech groups).
EXERCISES
4.
255
where
Let (X 1X^X2) be a proper triad (with respect to the Cech groups) = VJ A" 2 and let A = C\ Show 2 be 0dimensional.
X
X
l
,
X
}
X
that ff.(X;C)

//,(*i ;C)
+
connected and II^X^G)
C. LIMITING GROUPS.
=
II q (X 2 ;G) for q
>
1.
If
X\ and
X
2
are
= II^X^G),
then //^XjG)
ff (A,G).
We
pairs.
1.
a Hausdorff space and
shall consider here the category d// of pairs (AT,^4), where is is a subspace of X, and of all maps of such
X
A
Given a
Ct r
(partially exact)
homology [cohomology] theory
an extension
the category
of
compact
pairs, define
H on H [II] of II to the
Q category G/, using direct limits of the groups II q (X a ,A a ) [H (X at A a )] where (X a ,A a ) is any compact pair contained in (X,A). Verify the axioms for the limiting theory // [//].
Assume in 1 that H is given on the category Ct//. Define a natural //>// [//+//]. map 3. Show that, if H is the singular homology theory on the category > H is an G//, then // isomorphism. Show that, if the coefficient group
2.
field, then, for singular cohomology, an isomorphism. 4. Show that, if (X,A) is in ft/, and A is closed, the groups q (X,A) ^_ Q [H (X,A)] may be defined equivalently as limits of the groups 7/ a (Y af /l X a ) [H q (X ai A X a )] where X a is any compact subset of X. 5. Let be a (possibly infinite) simplicial complex and L a subcomplex of K. Show that the inverse limiting cohomology group
is
compact or a vector space over a
>
//
//
is
H
n
H
K
H
Q
(K,L;G) (see vin, Exer. F)
group topology
H
Q
naturally isomorphic with the limiting (\K\,\L\;G) using singular cohomology and using the weak
is
in \K\.
D. HOMOLOGY WITH COMPACT CARRIERS. DEFINITION. A partially exact homology theory H defined on the category a// is said to have compact carrier* if, for each u e H Q (X,A), there is a compact pair (X',A ) C (X,A) such that u is in the image of
f
the
homomorphism 1. Assume that
ft//
H (X',A') H H exact and
g
>
g
(X A) induced by
t
inclusion.
is
be a pair in
and (X',A
f
)
has compact carriers. Let (X,A) a compact pair contained in (X,A). If
u
e
H
g
(X',A')
is
in the kernel of the
homomorphism
H
g
(X',A')
>
H
show that there exists a compact pair V (X,A) induced by inclusion, (X",A") such that (X',A ) C (X",A") C (A',.4), and that u is the
f
kernel of
> g (X",A"). (Hint: consider the triple (X,A,A') 9 (X',A') then consider the triple to reduce the general case to the case A = A
1
H
H
',
256
2.
EXERCISES
Show
that for any exact homology theory
H
on
&H
the following
conditions are equivalent:
(a)
(b)
(c)
3.
H has compact carriers, the map H * H of C,2 H of C,2 the map H
>
is
is
an isomorphism,
onto.
Show
Let
that singular homology has compact carriers.
E. ADMISSIBLE
1.
AND UNIQUENESS CATEGORIES.
a partially exact homology [cohomology] theory on an admissible category a. Let a' be the (full) subcategory of a deter
H be
sequence
which the homology [cohomology] an admissible category and that is an exact homology [cohomology] theory on 0,'. 2. Let $ H be the category consisting of all compact pairs (X,A) which have the homotopy type of a triangulable pair and of all maps
pairs
is
mined by those
exact.
(X A) Show that
9
for
<2/ is
H
of such pairs.
3.
Show that $ H is an admissible category. that any partially exact homology [cohomology] theory on the category Q c is an exact theory on the subcategory <0//. Show
4.
Show
that $ H
is
homology. second volume.)
(This result will be generalized
a uniqueness category for homology and coin Chapter xii, of the
CHAPTER X
Special features of the Cech theory
1.
INTRODUCTION
The discussion in this chapter is centered around the continuity property of the Cech groups. Roughly speaking this property may be If the compact pair (X,A) is the inverse limit of the stated as follows:
compact
pairs (X a ,Aa), then the Cech groups of (X,A) are limits (inverse for homology, and direct for cohomology) of the groups of (X a ,A a ). At the end of the chapter it is shown that among all the
,
on the category Ct c the Cech theory is " the only one satisfying this ''continuity axiom. Thus a essentially axiomatic description of the Cech groups is obtained (for complete
partially exact theories (ix,10)
compact pairs) despite the lack of exactness. As an application of the continuity property, we show that the Cech groups satisfy a much stronger form of the Excision axiom. In particular, in the Cech theory, every compact triad is a proper triad
very useful in applications. Sections 69 deal with homology theories for noncompact spaces obtained by compactifying the spaces (in some standard fashion), and then taking the Cech groups of the compactified space. Two methods
(see i,14.1).
is
This fact
of compactification are discussed.
The first is the compactification of a locally compact space by adding a point at infinity; the other one is the Tychonoff compactification of a normal space. The first yields a
theory on the category
especially interesting for
This theory is pairs. In the case of a locally finite (1) infinite simplicial complex, the groups are isomorphic with those based on finite cochains and infinite chains. (2) The theory does not admit
Ct L c
of locally
compact
two reasons:
a reduced theory because in the category Ct LC only compact spaces are collapsible (in the sense of i,7.3). The theory for normal spaces obtained by using the Tychonoff compactification turns out to be isomorphic with the Cech theory based on finite open coverings.
Throughout the chapter, the Cech groups will only be applied to compact pairs (X,A) and therefore (see ix,8) we may assume that the f set Cov (X) of finite open coverings of X is the directed set used in the definition of the Cech groups of (X,A).
257
258
SPECIAL FEATURES OF THE CECH THEORY
2.
[CHAP.
X
FORMULATION OF CONTINUITY
Let Ct denote a category of pairs (X,A) and maps, and let InvCt denote the category of inverse systems having values in ft. Elements of Invft will be denoted by boldface symbols. Thus, (X,A) e Invft
means
(X,A)
is
=
\(X m ,A m),^\}
in
with values (Jf m ,yl m ),7r) an inverse system over some directed set * s an inverse ft. if (Y,B) = Similarly, {(Fn ,B n ),PnM system over
to Invft, a
M
N
and belongs
map
f:
(X,A) > (Y,B)
of Invft
> is composed of a map /: and maps fn (X fn ,A fn ) > as in the definition vm,2.3. (YnjBn) The operation of taking the inverse (or direct) limit will be denoted is the limit of the by lim. Thus lim (X,A) is a pair (X,A) where
:
N
M
X
inverse system
{X m ,Trl}
y
and
f
A
e
is
the limit of the
f:
system of subspaces
lim (X,A)
>
{A m ,7r;A mi }.
(Y,B)
is
Likewise, if defined as vm,3.10.
Invft, then lim
lim
We
shall
assume that the category
ft is
such that the operation of
inverse limit
maps
Invft into
ft:
lim:
Invft
>
ft.
For example, this is true if ft is the category of compact pairs. observed in vm,3.14 > lim is a covariant functor.
As
H
is
Suppose now that a partially exact homology [cohomology] theory can be applied defined on a with values in g or g c [9^]. Then
H
to (X,A)
e
InvOl to yield inverse [direct] systems
and maps
d:
// a (X,A)
>
H
a
. l (A),
[5:
H*~
if f:
l
(A)
> //(X,A)].
>
defined in the obvious way. maps / mje [/*] form a map
f
Similarly,
(X,A)
(Y,B), then the
#
:
ff,(X,A) +
H.(Y.B\
[f*:
c H*(Y,B) > ff (X,A)].
thus a kind of "homology theory on the category Invtt. values are, corresponding to those of //, in Invg* or Invg c
We have
"
Its
2]
FORMULATION OF CONTINUITY
259
It follows that the two composite functors Q lim and lim q are both defined on Inva and both have values in Q R or g c They are comparable as follows: THEOREM 2.1. Let (X,A) = lim (X,A) and let ir m (X,A) > (Xm ,A m ) be the projections. Then ir m = ?r 7r m for m < m! and 7r m# = Thus the homomorphisms \Tr m ^} constitute a homomorphism fl"m*7Tm'*.
.
H
H
:
>
H
7'Az's defines,
9
(lim (X,A)) > //,(X,A).
by vm,3.10,3.13, a limit homomorphism
Z(?,X,A):
7/ Q (lim (X,A)) > lim ff,(X,A).
lim into lim // in the following sense: a natural transformation of e Invd then commutativity holds in the diagram If (X,A)
lis is
I
>
H
lim
#
Q
(X,A)
Ilim d
*
I
//^(lim
Likewise, iff:
the
(A))
>
lim //,.,( A)
//icn
(X,A)
>
(Y,B)
^
in Invd,
commutativity holds in
diagram
I
7/ (lim
(X,A
>
lim 7/,(X,A)
//.(lim (Y,B))

>
lim tf C (Y,B)
The
proofs follow from vin,3.11.
2.2.
A
similar result holds for co
homology:
THEOREM
be the projections.
Let (X,A) = lim (X,A) and ir m (X,A) > (X m ,A m ) Then the maps {TT*} constitute a homomorphism
:
thereby defining a limit
homomorphism
lim // (X,A)
G
Z,X,A):
260
SPECIAL FEATURES OF THE CECH THEORY
I
[CHAP.
X
The homomorphism
so defined is a natural transformation of lim
H into
H lim in the sense that commutativity holds in the two diagrams
I
lim #'(X,A)
t
lim
5
lim ff''(A)
lim
#(X,A)
lim(f*)
'
I
>
H(lim
(X,A))
(limf)*
lim
DEFINITION
2.3.
A
or
H with values in g^
Q c [g]
I
partially exact homology [cohomology] theory is said to be continuous on the category
01 if the transformation each (X,A) e InvCL
of 2.1 [2.2]
is
a natural equivalence,
i.e.
for
lim
lim
# (X,A)
Q
#V,A)
ff(lim
(
We emphasize that the concept of continuity is not defined for cohomology theories with values in the category g<> DEFINITION 2.4. An inverse system {(X a ,A a ),7rf is called a nested a is a subspace of some fixed space Z and each TT^ is system if each an inclusion map (X^A^ (X a ,A a ). The intersection (X,A) = (H a ,n A a ) is called the intersection of the nested system.
}
X
C
X
THEOREM
2.5.
{(X a ,A a })iTa }, /ien eac/i point x s X regarded as an element of each X a The mapping yields an element <t>(x) of the inverse limit (X^^A^). ~ (Xa>,Aa>). established a homeomorphism (X,A) and The verification that is a 11 correspondence. Obviously
<t>:
<^>
<j>
// (3T,A)
is
the
intersection
of the nested system
<t>
</>"*
are continuous
is left
to the reader.
be a continuous
THEOREM
2.6.
Let
H
and
partially exact
homology
[cohomology} theory with values in $ R or g c [g^]. Let (X,A) be the intersection of the nested system {(X,A a ),7rf } of compact pairs in a compact Let i a (X a ,A a ). Then each system \u a where (X,A) space Z.
:
(% a ,A a ) and TT^+UP such that u = i a ^u and
ua
e
Q
f
H
u a determines a unique element u e H q (X,A) 9 vice versa. [Each u e H (X,A) is of the form
C
]
3]
CONTINUITY OF THE CECH THEORY
e
261
i*u a for some a and some u a
for some
/3
H
Q
(X a ,A a ).
t
If i*u a
=
0,
then
w^u a =
limit
of
>
a].
{
we identify (X A) with (X a ,A a ),7r} using the mapping of 2.5, the
PROOF.
If
<t>
the
inverse
conclusion of 2.6 becomes
a direct consequence of the continuity of the homology [cohomology]
theory
//.
3.
CONTINUITY OF THE CECH THEORY
which
The Cech homology theory based on a coefficient group is continuous on the category of compact pairs. The same conclusion holds for the Cech cohomology theory based on a coefficient
3.1.
is
THEOREM
in
9/2
or
Qc
group in Q K
.
number of lemmas, useful later, will precede the proof. Since we deal with compact pairs, we can limit our attention to finite coverings only. Since the subset A of a compact pair (X,A) is always closed, we may assume (see ix,8) that the Cech groups of (X,A)
are defined as limits of inverse or direct systems of groups defined over rather then on the directed set Co\r/ (X) of finite open coverings of
A
X
on the set Cov'(X,A). DEFINITION 3.2. If
indexing set
a, ft are families of sets in
,
Va = V
If a is an indexed an enlargement of ft, family of sets in X, a denotes the family of closures of sets of a. LEMMA 3.3. // X is a normal space and a is a finite open covering of X, then there exists a closed covering ft of X which is a redaction of a. PROOF. If V a has but one element v, then a, = X. Let ft, = X. Then ft is the desired covering. Proceeding by induction, assume the lemma holds in any normal space if V a has k elements. Let a be a 1 elements. such that V a has k Select a fixed v e V a covering of Since X is normal, there is an open and let A = a, for v 9^j) Q
and, for each v e V a a, and ft is called a reduction of a.
fty
X defined on the same then a called D
ft,,
is
X X X
1
U
+
.
U, and let an open covering of the normal space X' defined on a_set of k elements. Let ft' be a closed reduction Then ft is clearly a of a. Define ft,. = U, and ft, = # for v ^ v
set
U
such that
A
C U
^
and
U C
a...
Let X'
=
X 
a(
=
a, C\
for v
VQ
.
Then
a'
is
.
closed reduction of a.
LEMMA
exists
3.4.
Let
X
be a
normal space, a
sets
e
finite indexed family of closed
in
X
.
such
Cov / (X), and let ft that a < ft. Then
be a
there
an open enlargement y of ft such that a < y < 7 < ft and the nerves of 7,7 and ft coincide: Xy = X y = PROOF. For convenience, we can suppose that Vp consists of the k. Choose a v e V a such that & C <*,. Let B be integers 1,2, the union of those intersections of sets of ft 2 ftk which do not
X
ft
,
,
,
262
SPECIAL FEATURES OF THE CECH THEORY
ft.
[CHAP.
ft
X
7i>
It
meet
7i
Since
X
.
is
normal, there
0.
is
an open
0,
,
set y\
such that
71
C
0).
C
and
71 P\
S =
(If ft
=
we must choose
=
follows that the indexed family 71, ft, ft is > a, and its nerve coincides with Repeat the foregoing process on the element ft of
X
ft obtaining 7 2 and a new indexed family 7i,7 2 ,ft, obvious induction completes the proof. and A DEFINITION 3.5. If a is an indexed family of sets in X, then a is said to be regular relative to A if the following two conditions = 0, v e V a then a, C\ A = 0, and (2) if are satisfied: (1) if a, C\ A
the family
,
71, ft,
,
ft.
An
X
C
,
r\
D
of
P
Xa
A 7^ A 7* 0.
v is
for
i
=
1,
,
k,
and
D =
a Pl
P
,
P a,
fc
^
0,
then
The two
(1)
if
conditions can be restated in terms of nerves as follows:
a vertex of
all its
A &1
a
then
v is
a vertex of
it lies
Aa
.
and (2)
,
has
vertices in
is
Aa
then
in
Aa
or,
if a simplex s using the termi
nology of
ii,9.1,
Aa
//
full
subcomplex of
Xa
X
to
a normal space, and A',X' are dosed which are regular with A' X', then the finite open coverings of r both A' and X' form a cofinal family in Cov (X).
3.6.
is
LEMMA
X
sets
in
C
X
relative
PROOF. Let a e Cov (X). By 3.3, there is a closed covering )8 of A' (consisting of closed sets of A') such that a < 0. By 3.4, there is an enlargement 7 of consisting of open sets of X' such that a < 7 X y = X Let U be the union of the sets of y. Then B = and X f
.
r
ft
X'
sets
U is a closed set. By 3.3, there is a covering d of B by closed of B such that < d. Form the composite covering = (7,6) of
a.
e
X' by closed
sets of X'.
Clearly a
<
e.
such t rj of rj n consisting of open sets of A 1 = for each ij v which is an enlargerequire in addition that TJ V B ment of a set of 5. If this is not the case, choose an open set
.
P
X
By 3.4, there is an enlargement and X = X that a < We
such that
WP
X,,
s is
A'
=
W3
and replace
still
a
<
TI
and
= X<
hold.
rj.
properties In addition, condition (1) holds for
y,
rj v
by
C\
W.
The
regularity relative to A' of
Suppose each vertex
7.
s
v of s,
a simplex of X^ and its vertices lie in A(. Then, for ^ 0. Thus 77, is an enlargement of a set of 17, C\ A
f
t,
.
Since X, = Since Xy it follows that s belongs to X^ A' Pi Car,(s). Therefore Car^(s) ^ 0. But Car^(s) belongs to It follows that Car,(s) meets A'. Therefore 6* lies in A(. Suppose now that s is a simplex of X, and its vertices are in XJ. = () s belongs to X.. Therefore Car(s) Since 0. But Car (s) n
X
X
fti
X
.
ft
C
X
X
^
C
X' Pi Car,(s). It follows that Car,() meets X'. Therefore s lies in XJ. Thus is regular relative to A' and satisfies condition (2) for regu77
larity relative to X'.
W be an open
set
Let U' be the union of thejypen sets of  U' and C\ X' = such that
WDX
W
rj.
0.
Let Let
3]
OL
CONTINUITY OF THE CECH THEORY
263
be the covering a. cut down to If we adjoin a to 77, a composite covering >a which is regular relative to A' and
.
W
we obtain
X
1
'.
LEMMA
let
3.7.
Let
jX m ,7r;} be
Then
an
X=
m
inverse system of compact spaces,
and
lim
e
where
M
,
{X m ,7r;}. e Cov'(X m ), form a
,
the elements of
c ofinal
Cov'(X)
.
of the
form
Tr'X/S),
PROOF. Suppose a e Cov'(X). ^(U), where U is open in X m form a base for the open sets of (vm,3.12), it follows from the compactness of X (vm,3.6) that there is a covering a consisting of sets Tr'^f/,) (i Since 1, fc) such that a < a. is directed, there is an m > m, for i = 1, k. Define a covering = (ir:,)= X m  w m (X). of X m by for i = 1, fc, and & +1 = 0. It follows  v~}(U.) ( i = 1, Then 7^(0,) K) and TT^A + I)
family Since sets of the form
X
M
,

,
1
t
^)


,
,
that
ir~ \ff) m
>
3.8.
a.
LEMMA
and
let
Let
{
(X mj A
,
WI
),7r
(X,A)
=
lim {(X m
A
),*%}.
e
form a
=
j)
ir^GS)
(m
e
M,
e
an inverse system of compact pairs, Then elements of Cov'(X) of the Cov'(XJ) <md suc/i tfia* (X a ,A a ) =
1 )
1
be
(X m 0,^l m/ form
PROOF.
a cofinal family.
Suppose 7
e
Co\r/ (X).
By
3.7, there is
an
w<>
and a
6
e
Cov(X mo ) such that By 3.6, there is an
and
ir^(6)
>
7
Ict Y'
=
is
7r mo
(^) and A'

7r mo
(^).
Cov(X mo ) which
[7
A and
f
5
<
f
e.
Let
(W)
regular relative to both be the union of the sets of c which do
X
f
not meet X' (A
t/',TF' are open,
it
),
and
1
1
let
r
f
X C
f
also meets
m > m
(A ). such that ir.(XJ
X
and, if a ^', A C H two applications By
1
,
Then mo ,]V be their complements in T/ set of meets [/' (ir), of vm,3.7, there is an
.
X
C U
f
and *l.(A m )
C W.
Define
=
Ou.rXO.
Then
Since a
=
Suppose
(Car^(s)) v of s, e p
,9
is
Hence for each vertex is regular relative to X', Car (s) meets X'. A"'. meets Since e But Car a (s) = ^l(Car.(s)). It follows that Car a () ^ 0. Therefore s belongs to X a This proves that X = X m ^. The same argument shows that A a = A mft with X, X m X mo U' replaced by A, A mj A mt This proves the lemma. PROOF OF 3.1. The proof will be given only for the case of homology;
.
C
TT^^), we have, by ix,2.5, that a simplex of Then Car^(s) ^ nft we have that Car (s) meets U U',
(X a ,A a )
0.
C
e
(X mft ,A mft ).
f
X
.
Since Car
(s)
D T.
f
.
,
,
,
W
.
the case of cohomology is left to the reader. Suppose then that (X,A) = {(X m ,A m ),ir\\ is an inverse system of compact pairs. Let (X,A) = lim (X,A). We must prove two propositions concerning the
is
homomorphism
it is
/(</,X,A) of 2.1.
First, its kernel
zero; and, second,
onto.
26*
SPECIAL FEATURES OF THE CECH THEORY
Suppose u
E
[CHAP.
is
X
H^X^A)
lies in
f
the kernel of
it
I.
Since
7r
m5e
ir
coordinate of l(u) in
m
is
e
M
H
Q
(X mj A m )
follows that
7r
(w)
=
m ^(u)
the
for each
.
If
If
ft
E
Cov(X m ),
1
then the coordinate of
ffl+
(w) in
it
zero.
a
=
Tr"
^) and (X a ,A)
= (X
m/3 ,yi mi8 ),
q (X m ^A m& ) follows from the
H
definition of
By
(X,A) such that Z(u) = v. It suffices to construct the coordinates u a of u for a in some cofinal family D in Cov'CX"). Let D be the cofinal family described in 3.8. For each 1 7 such a choose an m e and a e Cov (X m ) such that a = Tr" ^) and
g
H in H
?r that the coordinate of u in Q (X a ,A a ) is zero. mjt (ix,4.2) / such a's form a cofinal family in Cov (X). Therefore, u = 0. Let v m be the coordinate of v C (X,A). Suppose now that v & lim / in For any e Cov (X m ), let be the coordinate of v m Q (X m ,A m ).
H
3.8,
#
(X mft ,A m p).
We
must
find
awe H
^
g
M
3
(X a ,A
i
= (X mf ,A mf ). Define u a = v Suppose a, < a in D, and Wi,j3i and w ,0 are the = (^m!)"" ^) and a Choose w > m, and define
tt
)
mft .
2
2
2
choices for
(i
1
2.
/3<
=
f${
1,2).
By
02*
3.6, there exists
is
an
e
Cov (X mj ) which
/
is
a refinement of
and
regular relative to both 7r m ,(X) and v m ^(A). the proof of 3.8, that there is an ra 4 > ra 3 such that,
and
It follows, as in
if
=
(ir"J)
^(e),
1
and
for i
i
=
= 7r;l(/3), then (JT ,A a = (X m A m J. = 1,2. Then (X a% ,A ai ) = (Xm ^ ,A m y
)
. ftt
%
Let 7,
>)
=
(ir::)"
^,)
for
.
= (X m ^A mtftt )
1,2,
because each
the third.
and a
>
contained in the following and the first equals > 7i/y 2 Therefore u ai = v m ^ Since e > &{,&*, we have Let u = v m4 Since commutativity holds in the i,a 2
is
%
. .
/j.
diagram
H
g
(X a ,AJ
1
= H (X m ^A m ^
q
I
u projects into u at (i = 1,2). Therefore u a projects This proves that the elements u a for a e D are the coordinates of an element u e H^X^A). It shows also that w a is independent of the choice of m and 0. = v m for each It remains to show that l(u) v, or that ^^(u)
it
follows that
.
,
into w ai
This follows if (7r mj(c (w)) = y me for c in a cofinal family in c be regular relative to ir m (X) and ir m (A). By 3.6, such coverings form a cofinal family. As in the proof of 3.8, there is an > such that, if = (nCT'W and a = 7r^(/3), then (X a ,A a ) = l (^mt/9,A mi/9 ). Because w a is independent of the choice of the m,/5 used in its definition,, it follows that u a = v mft Therefore u a and v mift have > the same image in 7/ c (X m ,A m< ) under the inclusion map (X miftj A mtft ) (Xm ,A m< ). These images are (7r m% (u)) and v mt respectively. This
m m
e
M
.
Cov(X m ). Let
m
.
completes the proof of
3.1.
4]
CONTINUITY VERSUS EXACTNESS
4.
265
CONTINUITY VERSUS EXACTNESS
efficients is
be shown that the Cech "homology theory" with integer conot exact on all compact pairs. The proof is based on the continuity of the Cech theory. This and Theorem 12.2 below imply that the (essentially unique) homology theory with integer coefficients on triangulable pairs does not admit an extension to compact pairs which is simultaneously continuous and exact. EXAMPLE 4.1. Let (E,S) be the 2cell and its boundary defined in
It will
the complex plane by the conditions Consider the maps
<f>:
\z\
^
1
and
\z\
1
respectively.
S >
5,
f:
E
*
E
defined
by
If for
each
z e
S we
is
(
identity z
and
<(z), the pair (E,S)
becomes a pair
(P,C) where
Since
f<t>(z)
P
the projective plane and
3
=
z)
>
=
(P,C).
3
z
=
</>/(e), it
a projective line in P. follows that the map / induces
is
C
a
map
/:
(P,C)
In xiv, Example
9.7,
the reduced homology
sequence with integer coefficients of (P,C) will be calculated, and also the endomorphfsm of this homology sequence induced by /. It will be shown that the only nontrivial groups are #i(P), ffi(C), and 2 (P,C)
H
and that the diagram
isomorphic with the diagram of vm,5.5. Let be the set of positive integers with the usual order. For each a e define (P,C) = (P,C) and 7r + 1 = /. Define all other ir
is
M M
for
ft
<
a by
transitivity:
TT
=
1
irf^iTa"
.
Then {(P,C),irf
}
is
an
inverse system of spaces indexed
by M.
Let (P,,C) be the limit pair.
By vin,3.6, (P, Coo) is a compact pair. If we employ integer coefficients and the Cech homology theory (or any other continuous homology theory), then the (reduced) homology sequence of (P >,<*) is isomorphic with the inverse limit of the homology sequences of (P,C) under the
266
SPECIAL FEATURES OF THE CECH THEORY
.
[CHAP.
X
maps (7rf) #Je This limit sequence is then precisely the sequence obtained as the limit in vni,5.5. Thus in the reduced homology sequence of (P,C ro ) the group H^(P^) is cyclic of order 2 while all the other
groups are trivial. Thus the reduced homology sequence of (Poo,Coo) is not exact. The nonreduced homology sequence differs only in the positions 7/o(P>) and // (Cco) (which are cyclic infinite) and also is not exact. The pair (P w \C m ) may be described alternatively as follows: Consider the pair (E,S)
where S is the 3adic solenoid and E is the join of Using the antipodism of /S define an antipodism on S by a passage to the limit. Identifying antipodic pairs of points on S, one obtains a pair homeomorphic with (Poo, Co,).
S with
a point.
1
,
6.
RELATIVE
5.1.
if
HOMEOMORPHISMS AND EXCISIONS
> (X,A) (Y,B) is called a relative B. homeomorphically onto Y > is a map of compact pairs and f (X,A) (Y,B) #, then f is a relative homeofashion onto Y
DEFINITION
homeomorphism
/maps X
A map
/:
A
LEMMA
maps
5.2.
X
A
A.
A
// /: in a 11
morphism. PROOF.
>
We need only verify the
defined
continuity of the
X
X
map
3
g:
Y
B
by
g(y)
=
l
f~
(y).
Let
U
be any open
subset of
Then
g\U) = f(U) = Y
Since
open.
 B  f(X is
U).
l
X
is
U
g
is is
compact, f(X
continuous.
5,3.
U)
closed,
and therefore g~ (U)
//
if
is
Thus
DEFINITION
inexact)
A
homology [cohomology] theory
/:
(exact or
said to be invariant under relative horn eomor phi sms
admissible relative
homeomorphism
(X,A)
>
for every (Y,B) the homomor
phisms / #
[/*]
are isomorphisms in
5.4.
all
dimensions.
The Cech homology and cohomology theories on the category <jt c of compact pairs are invariant under relative homeomorphism s. > PROOF. Let /: (X,A) (F,J3) be a relative homeomorphism of compact paiis, and let \B a be the collection of all closed sets of Y such that 7?CInt/? a If B a and Bp are in the collection, then B a C\ B is also in the collection. Therefore {(F,# a ),pf} where pf: (Y,Bp) Let A a = (F,/? a ) form a nested system with (Y,B) as intersection. /" (J5 a ). Then {(X,AJ,rf}, where rf: (X,A ) C (X,A a ), is also a > nested system with (X,A) as intersection. Let /: (X,A a ) (F,J5 a ) be the map defined by /. We show first that f a ^ is an isomorphism.
}
.
THEOREM
ft
C
l
ft
5]
RELATIVE HOMEOMORPHISMS, EXCISIONS
Let
267
V
be an open set of
Y
such that
U =
f~\V).
Then
A C UC U
C
U)
Int
B C V C ~V C Int B a Let A a Consider the diagram
. .
H
g
(X
 U,A a 
>
H
g
(X,AJ
I/*
II g (Y

V,B a

V)
>
H
q
(Y,B a )
and g is defined by /. Since / is a and A C U, B C V, it follows that g is a homeohomomorphism morphism. Thus g^ is an isomorphism. The maps i # and i'^ are Thus commutativity in the isomorphisms by the excision axiom. is an isomorphism. that / diagram implies
where
i
i
and
f
are excision maps,
relative
ast
The commutativity
relations in the
>
diagram
)
>
H
g
(X,A)
I/*
U
g
(X,A
/^*
ft
II g (X,A a )
/*
>
H
An
g
(Y,B)
>
H.(Y,Bj)
//(!', B a )
where the horizontal maps are induced by imply that / + is an isomorphism.
inclusions, together with 2.6
analogous proof applies to cohomology. The above proof did not make any appeal to the specific definition of the Cech groups and remains valid for any partially exact homology or cohomology theory satisfying the continuity axiom. This
REMARK.
gain in generality is however to a large extent illusory, since it will be shown in 12.2 that any such theory is isomorphic with the Cech theory. Theorem 5.4 is of course a generalization of the excision axiom for
compact
redefine
This theorem becomes the excision axiom pairs. an excision to be a relative homeomorphism.
CLi
itself if
we
In general in the categories
following types of excisions: (E) Inclusion maps /:
or
dc
or
Q LC we may
U)
C7)
consider the
(X (X (X

U,
A 
C
C
(X,A) where (X,A) where
U
is
open
in
X and U C
Int A.
C7,
(E,) Inclusion maps /: and C. A. open in
A A 
U
is
X
U
(E 2 ) Inclusion maps Int A.
/:

U,
U)
C
_
(X,A) where
UC
(E 3 ) Relative homeomorphisms.
Type
(E) consists of the excisions introduced in Chapter
i.
Type
268
SPECIAL FEATURES OF THE &ECH THEORY
[CHAP.
X
(E 2 ) in the categories Q c and CL LC coincides with type (E) since the map / is not admissible unless U is open. Excisions of type (E 2 ) in the category CLi are applicable in the singular homology and cohomology theory (see vn,9.1). Theorem 5.4 asserts that excisions of type (E 3) are applicable in the Cech theories on the category Ct c The following example shows that excisions of type (EO are not
.
applicable in the singular homology theory (even in the case of compact metric spaces).
EXAMPLE
5.5.
Let
R
be the region in the
sin
1 
(z,2/)plane defined
by
X
1
< <
y
<
<
2 2
for
Q x
y
for
< =
\x\
<
2
,
07T
0.
Let
X"
be the rectangle given by
and
It is
let
C
be the boundary of X.
Clearly
R C
X.
Set
A =
X 
R.
easy to see that
implies
H (C) = H
Q
by 1,10.5 H*(X,A)
Let
H
q
 0. B = R
A
(X,C)
Pi
a strong deformation retract of A. This Q (A) (induced by the inclusion map), and therefore Since (X,C) is a 2cell, it follows that g (X,A).
is
C
H
A B
be the boundary of R.
singular theory,
we have
where
U
=
shall show that, in the since (R,B) = (X U) U, A 0; (R,B) Int A, it will follow that excisions of type (E,)
We
H
2
=
are not applicable in the singular theory. Since R is contractible over itself to a point,
H (R)
2
0.
Therefore,
by exactness,
d:
H
2
(R,B)
we have H^R) = H,(B). Thus it
__
remains to show that Hi(B) = (in the singular theory.) 1 g y ^ 1, and let Let D be the part of B on the 7/axis with D' = B D. By a simple argument involving local connectedness,
any singular simplex
in
B
lies
wholly in
D
is
or wholly in D'.
Thus the
total singular complex of divides into and that of D'. Hence plexes that of
B
two
D
disjoint singular com#i(B) decomposes into the
it is
direct
contractible to a point. So Hi(D) = 0. Similarly D' is homeomorphic to an open line segment; so it too is contractible to a point, Hi(D') = 0.
sum H^D)
+
H^D').
Since
D
a closed line segment,
Thus #i(B) = The above
flo(B)
as required. analysis of the singular complex of
G,
B
implies also that
+
#
(B)
(B)
G.
Thus we have
G,
ffi(B)
5
*

6]
LOCALLY COMPACT SPACES
we assume
that
269
in the singular theory.
theory
Hi(B)
H
0.
Thus 2 (R,B) 2 (X,A) by 5.4. (R,B) by exactness and Since B is connected, the nerve B a for any finite open and fi Q (B) = covering a of B also will be connected. Thus /? (B a ) =
2
l
H
In order to compute the groups of B in the Cech G is a field, thus^ insuring exactness. We have
H
H
(B)
~
G.
This yields
#
(5)

0,
H^B) =
G.
This exhibits the difference between the Cech and singular theories. Incidentally it shows that the singular theory is not continuous on the In the above discussion we assumed that G was a field category Ct c
.
in order to
compute the Cech group Hi(B) using exactness. argument using continuity could be used to establish Hi(E) any coefficient group.
6.
A direct ~ G for
HOMOLOGY THEORIES FOR LOCALLY COMPACT SPACES
recall here that in i,2, the category
We
and
& LC was
defined to consist
of pairs
(X A) such
y
A
is
a locally compact (Hausdorff) space, closed in X, and to consist of maps / such that the inverse
that
is
X
image of any compact set under / is compact. DEFINITION 6.1. A subset A of a locally compact space (abbreviated:
LCspace)
X
is
called bounded
if
if
A
is
compact.
A
subset
U
of
X
is
called countercompact
X
U
is
compact.
be an LCspace and let w be a point not in DEFINITION 6.2. Let w to consist of the the space X. We define a topological space \J w and the open sets (i) the open sets of X, and (ii) the union set a? of 0} with the countercompact sets of X. It is easy to see that
X
X+
X
X+
co
is
a compact (Hausdorff) space.
LEMMA
and w
x
e
6.3.
Suppose (X,A), Y, and
Set g(x)
>
f:
X
A
>
Y
are in
Q, LC ,
is not
in Y.
g:
=
f(x)
is
a continuous extension of /. // / is a homeomorphism, then g is a relative homeomorphism. If f is a homeois compact, and A is a single point, then g is a homeomorphism, morphism. l l PROOF. To prove that g is continuous observe that g~ (U) = f~ (U) then for U Y, and if U =
A.
Then
X
forxeX
A
and
g(x)
=
for
Y +
w
X
C
if
WU
l
,
l
9~ (U)
=
f' (W)
UA
l

XW)
l
f' (Y
is

W).
l
Thus,
open.
W
is
countercompact, f~ (Y
part of the
compact and g~ (U)
is
The second
lemma
is
obvious.
To prove
the last part
270
SPECIAL FEATURES OF THE CECH THEORY
if
[CHAP.
X
a homeomorphism and A is a single point, then g 1; thus, compact, g is a homeomorphism. It will be convenient to pick a fixed point co and assume that co is not contained in any space of the category Q, LC This actually amounts to replacing & LC by a subcategory. With this assumption it is possible
observe that,
is 1
/
is
if
X
is
.
to "compactify"
all
breviate
in
X+
jf:
the spaces of
If
GL LC
w by X.
(X,A)
ac
.
If
is
(X,A) (Y,B)
is
is
= f(x) for x e and /() .= co, is in a covariant functor on & LC with values in Ot c We shall refer to this functor as "the single point compactification." Note that 6 = co, so that A is never vacuous in the symbol (X,A). To show that is an ftfunctor (iv,9), we convert Ct c and Ct LC into
(X,A) > (7,5), defined by f(x)
It is clear that
Ct c .
.
shall abby adjoining w. a pair in GL LC then (X,A) is a pair in a LC then by 6.3 the map /:
, ,
We
X
/icategories Ctc
and a/ c
in the following
way:
Couples
(i,f)
are in
clusions
i
(A,B) > (X,B)
 (X,A)
3
where
and
J5
and
a
A is in Ct c (or &LC), B X",A,B) is a triple, i.e. where X, A are closed in X. Excisions are defined to be the relative
map
/
F:
is
X
C
C
homeomorphisms.
there
is
F(a;,0)
=
(x),
where
a single point. THEOREM 6.4. Le^ // 6e a homology [cohomology] theory on & c or Q LC which is invariant with respect to relative homeomorphisms. If we regard the boundary [coboundary] operator of triples as the basic boundary [cobecomes a homology [cohomology} theory on QC boundary] operator, then
X
Two maps / ,/i: (X,A) (F,Z?) are homotopic if X f A X /) (F,#) in G c (or a LC ) such that F(x,l) = / (x). Poin/s are defined to be pairs (X,A)
>
(Y
>
x
A
H
or
die
respectively.
theories.
The same holds for
partially exact homology
and
co
homology
PROOF. The validity of Axioms 1, 2, 3 is clear. The Exactness axiom is the statement of the exactness of the homology [cohomology]
sequence of a triple. If the given theory is partially exact, i.e. the homology [cohomology] sequence of a pair is a sequence of order 2, then the same holds for the sequence of a triple (see ix,10). The Homotopy axiom is satisfied since the homotopies in d c (or GLLC) coincide with those of QC (or GLC). The Excision axiom holds since, by assumption, the given theory is invariant under relative homeomorphisms. Finally, A is a single point, and the if the pair (X,A) is a "point", then is a relative homeomorphism. A Thus inclusion map i: (X,A) are isomorphisms, which implies the Dimension axiom. i+ [t*] THEOREM 6.5. The single point compactification is a covariant h
X
X
*
C
functor
:
Qic
&C
6]
LOCALLY COMPACT SPACES
271
It is clear that carries couples into couples and thus is a Let / ,/,: (X,/l) > (Y,B) be in ai c and let F: (X X /, A X /) > (F,#) be a map in &Lc such that F(x,Q) = / (z), * CM) = P^xtend f to a map F: /i(x). (X X /, A X /) > (]?VB) be setting
PROOF.
cfunctor.
7
F(,0 =
tive
.
Then F
is
continuous by
6.3, If
/i(x).
Thus / and
y
j\ are
homotopic.
homomorphism,
,
then, since
X X
and F(z,0) (X,A) /:
=
f
(x),
F(x,l)
is
=
A =
(F,B)
a rela
X
A
and
Y
B =
B it follows that / is also a relative homeomorphism. If (X,A) A = A is a single point and (X,A) a ''point" in C/' c then is a "point" in GC. DEFINITION 6.0. Let // be a partially exact homology [cohomology] theory on the category (i c invariant with respect to relative homeoas a theory on (ic and denote by // the compomorphisms. Regard There results a partially with the /ifunctor sition of c. Ct/c ^ on GiLc or, what amounts to the exact homology [cohomology] theory same, a theory on GL L c invariant with respect to relative homeomorwill be called the LCtheory associated with H. phisms. The theory
Y
is
X
H
H
:
H
If // is exact, so also is //.
Note
z:
that,
(X,A)
C
(X,A) in Ct/, c is a compact pair, then the inclusion map (X,A) is admissible and is a relative homeomorphism.
if
Thus we have
H
Q
(X,A)
= H
q
(X,A)
 H
**
q
(X,A).
.
Thus H ~ II on Ct c and // may be regarded as an "extension" of H From now on we shall limit ourselves to the case when H is the Cech The objective will be to give a direct description of the theory on Ct c
,
.
associated LCtheory
terms of coverings. f Let (X,A) be a pair in CL LC and a e Cov (X) a Let 12 a be the subcomplex of the nerve finite open covering of X. a of simplexes s whose carrier Car a (s) is not bounded (i.e. the consisting
H in
DEFINITION
6.7.
X
not compact). We shall consider the groups of > is a refinement of a and ir: a is a pro(X then TT maps A VJ 12^ into A a \J fi a jection, r DEFINITION 6.8. Let Cov^X) denote the subset of Cov (X) conis either bounded or sisting of those coverings a such that each set a,
closure of
a ,A a
W
Car a (s)
fi a ).
is
If
X
X
.
H
Clearly Covj(X) is a directed subset of Cov (X). The inverse [direct] limit over the directed set Cov,(X) of the groups n ) [H q (X a ,A a 12J] with the homomorphisms induced Q (X a ,A a
r
countercompact.
U
tt
U
denoted by II*(X,A) [ff^(X,A)]. The provision conby projections the categories in which the coefficient group may lie are the cerning same as for the Cech groups of compact pairs (see ix,3.3 and the subis
sequent remarks).
272
SPECIAL FEATURES OF THE CECH THEORY
THEOREM
q
[CHAP.
X
6.9.
For every
locally
compact pair (X,A) we have
H (X,A)
of coverings coverings for which either co e a, or co non e / a. for each v (see 3.5). By 3.6, Cov^X) is a cofinal subset of Cov (X), and therefore we may assume that the groups of (X,A) are defined as
where the groups of (X,A) are the Cech groups. PROOF. In Cov'(i) we consider the subsets
Cov (J)
r
a which are regular at
co, i.e.
limits over
r
Cov (X). Next we Cov (X)  Cov^X) as follows:
r
define
an order preserving
map
<f>:
=
ct 9
C\
X
=
o? f
co.
For the nerves we then have the inclusion maps
<t>
a
:
(Xi a9 Ata \J fy a )
C (X,A
The system
groups
H
Q
(X ay
H X
<I> {<,<*} constitutes a map of the inverse system of (Xp,A0 \J Q ) indexed by Covi(X) into the system of groups r The limit map a ) indexed by Cov (X).
=
q
ft
</>:
H'.(X,A) * H.(X,A)
is
is thus defined. Similarly for cohomology {0,0 *} systems, yielding a limit map
a
map
of direct
and assert that are isomorphisms. The proof breaks up two cases according as the pair (X,A) is compact or not. r Further If the pair (X,A) is compact, then Cov'(X) = Cov (X). the inclusion map i: (X,A) C (X,A) is a relative homeomorphism and therefore i # and i* are isomorphisms. Comparing the respective definitions shows that i+ = 0oo and t* = </>", which concludes the proof in
</>
We
into
this case.
If (X A) is not a compact pair, then it is easy to see that maps Cov (X) onto CoviCX") and the inclusion maps a become identities. and <t>* are isomorphisms, and it follows from vm,3.15 and Thus
t
<t>
r
<t>
ajic
vni,4.13 that
<t>a,
and
a
<t>
was incomplete since we did not define induced homomorphisms and boundary [coboundary]; the missing definitions are quite analogous to those made for the Cech
1.
REMARK
The
also are isomorphisms. discussion of the A groups
groups. After these definitions are made it is trivial to check that the isomorphisms <> obtained above commute with /, and 5, thus yielding an isomorphism between homology theories. REMARK 2. Let be locally compact but not compact. Let
X
7]
LCTHEORIES
Covi(X).
Since every
intersection,
finite
it
27S
of
a
e
number
an unbounded
follows that
quently is homologically trivial. It the reduced homology sequence that // a (X a ,ft a ) ~ ft Q (X a ). Thus in defining the (absolute) groups H*(X) for a locally compact, non
countercompact sets has is a simplex and consefollows then from the exactness of
12 a
Note Thus
Q (X a ) rather then compact, space we can, use the groups 9 (X ay tt a ). = Q (X a ). if is compact, then fi a = and a ,ti a ) that, Q (X in this case H*(X a ) uses the groups Q (X a ) instead of ff Q (X a ). The difference between these two cases is of course limited to the di
G
H
X
H
H
H
mension
zero.
3.
A
similar
is
REMARK
It
remark applies to cohomology. easily shown by examples that the functor
is
does
not preserve excisions of type (E). Thus the singular theory able for composition with this functor.
not suit
7.
LCTHEORIES IN TERMS OF A SINGLE SPACE
Let
H
of
be a homology theory on
to relative
map a
X
homeomorphisms. A has an extension
a:
Q c which is invariant with respect Given a compact pair (X,A), the identity
(X,A) * ((X

A),u)
which
is
a relative homeomorphism.
Consequently
H,(X,A)
~
H,((X

A)'*,)
= H,(X 
A).
This suggests that one can describe the original homology theory using nonrelative homology groups alone, but defined over a suitable LC
category.
This leads to a new type of axiomatic system which we shall
proceed to describe. Let (BJLC be the subcategory of
/:
is
X
'
Y
and all maps GL LC of all LCspaces such that for every compact subset C of Y the set f~ (C)
l
X
H
?
compact. By a 'single space" homology theory on ($> LC we shall mean a system = {H q (X),f+,r,d} where the primitive terms are as follows: e & LC and each integer q (X) is an abelian group defined for each
H
X
If /:
X Y
is
in (B L c,
then
/:
H
q
(X) > ff.(F)
is
a homo
morphism.
is
> If U is an open subset of the LCspace X, then r: q (X) q (U) a homomorphism. We shall sometimes use the notation r (X .u) to and U. indicate the spaces
H
H
X
274
If
SPECIAL FEATURES OF THE CECH THEORY
[CHAP.
X
> an open subset of the LCspace X, then d: q (U) U) is a homomorphism. //<,_! (.X" = {// a (JK'),/ ,r,d} is subjected to the following The system
U
is
H
H
Jt
axioms:
AXIOM AXIOM AXIOM AXIOM
1.
1'.
// /:
// /:
X + X is the identity, then f+
the identity.
jr:
is the identity.
T<JT. *) i
2.
2'.
If
F > Z, tfien (0J% = JV X > F, V d U C, X and V and U are open in
and
X, then
AXIOM 3. // mutativity holds in the diagram
V C U C X
V
and
U
are open in
X, then com
H
Q
(V)
H^(X  U) + H^(X  F) where h: X  U C X  V. AXIOM 3'. ///: X > F, and U and V are open subsets of X and Y, U) C (Y V), then comrespectively, such that f(U) C V and f(X
mutativity holds in the diagram
H.(X)
/*
>
H.(Y)
H,(U)
where / t
:
H,(V)
C/
*
F is defined by f.
Under
the
AXIOM
3".
same conditions as in Axiom
3',
commutativity
holds in the diagram
H^X to/wre
U)
ff.^F

7)
/2
:
X
U
>
7
V is defined by f.
7]
LCTHEOR1ES
275
AXIOM 4
j:
X
...
UC
<
is an open subset of (Exactness axiom). If X, then the homology sequence
U
X
and
H
q
.,(X

17)
<
H
g
(U) < H,(X)
 H.(X 
U) <



is exact.
that the
for partially exact homology theories, we assume only homology sequence is of order 2, and is exact whenever (X,X U)
(Notq:
.
is triangulable)
AXIOM
g
5 (Homotopy axiom).
If g^g\:
ljr
X
.
>
X X
= (x) (x,0), 0,(z) (x,l), theng^ AXIOM 7 (Dimension axiom.) If P
= =
/ are defined by
is
a point, then
H
Q
(P)
=
for
q T* 0.
Note the total absence of any analog of the excision axiom. The main results about the connection between ordinary and
space" homology theories
is
"single
incorporated in the following four theorems:
THEOREM
7.1.
Let
H
be
variant with respect
to relative
H
is
Q
(X)
=
H
9
(X,o>).
If f:
X
a homology theory on the category Q, c ine (B L c define homeomorphisms. For each > Y is in & LC and f: (JS > (>>)
X
an extension of /, then Let U be an open subset
be defined by f(x)
define
f:
H
Q
(X)
>
H
*
q
(Y) by setting f
(X,w)
*
=
/,.
of the
= x for x s TTien cfe/inc f/. > H' (U) by T = /. AnaHy / r: fll(X) g  C7) ') ff a .i(X  [/) consider /fee map 0: (X,(X and g+ is an isomorphism. Define Tfeen ^ is a relative homeomorphism  U)') > ^^((X  C7), w ) is rfee where d: d = q (X,(X
1
LCspace X, and let f: U, f(x) = w for x e X
(C/,o>)
dgl
H
}
boundary operator in
the
is
H
=
{Hl(X),f+,T
.
,d
TAeri /fee system C7)',w). triple (X,(X a "single space" homology theory defined over
the category (& LC
THEOREM
the category
7.2.
Let II be a "single space" homology theory defined over
(B/,
c
.
For each (X A) in a c
(in
define
l
Letf:
(X,A)+(Y,B)
a c ),
let
U =
a
H\(X,A) = H Q (X  B),andte// f~ (Y
1
:

A).
C7
>
Y
B
be defined by f.
>
Then
fi is
map
in
($>
LC and we define /*:
H\(Y,B)
to be the
composition
>
ff a (F

B).
to be the
// (X,4)
is
an
ac
d:
,
/fee^
^ de^ne
A)
^^:
>
H\(X,A) > H+.^A)
g
^(A). Then the system H* homomorphism is a homology theory on a c invariant with respect {/ft(X,^),/i,^^} relative homeomorphisms.
Q
H
(X
H
=
to
276
SPECIAL FEATURES OF THE CECH THEORY
THEOREM 7.3.
Let
[CHAP.
X
Hbea homology theory on QC invariant with respect
Then
to relative
homeomorphisms.
H* is again such a theory
,
and
H?(X,A) = Hl(X  A) = H.((X <
A).
X
H
 4) ',) 66 denned by <f>(x) = x /or x e  A Let 0: (X,A) > ((X = w /or x e 4. TTien zs a relative homeomorphism and </,: and 4>(x) = //'"""(XT,;!) 2/zeWs an isomorphism ~ o/ tf*e homology q (X,A)
H
H
" "
1
theories.
THEOREM
Then
in
7.4.
Le
H
be a "single space
19
homology theory on
(B LC .
H*
coincides with
is straightforward but lengthy The proof of view of the large number of propositions that have to be checked. We therefore leave these proofs as an exercise for the reader. The results remain valid for partially exact homology theories, and with
H. the above four theorems
suitable reformulations also for cohomology.
If
(B LC
H
is
on
then the "single space" theory description using finite open coverings.
ac
H
on
may
the Cech theory be given a direct
8.
TYCHONOFF IMBEDDING AND COMPACTIFICATION
8.1.
/,
DEFINITION
functions a:
A
Given any where / =
set
A
[0,1] is
consider the space I of all the closed unit interval, with
4
A
the product topology defined as in v,5.2. The space I is called the cube A indexed by the set A. By v,5.4, I is compact. If B is a subset of A, A B then we identify the cube I with a subset of I by extending each
function a:
B
:
*
I to a function a:
this convention D
>
A
we
a:
If
>
x
is
e
A
B.
With
I
A
call
mapping p B
I
which to each
A
A
/ such that a'(x) = u A I a subcube of I * I assigns a\B: B
,
for
The
>
/
is
called the projection of I
onto I
B
.
the set
A
is finite,
then I
A
called a finite cube.
DEFINITION
every closed set
function /:
It follows
X
for A T^space X is called completely regular 8.2. A C X and every point XQ not in A, there is a continuous = 0, f(x) = 1 for x e A. I such that f(x
if,
>
)
from Urysohn's lemma that normal spaces are completely
Every subspace of a completely regular space
is
regular.
LEMI^A
PROOF.
8.3.
combe a
pletely regular.
Let
closed subset of closed subset of
that
g(y
)
/(i/
)
=
X andj/ X
e
Y C X, where Y and y e Y
=
1.
X
is
completely regular.
Let
A
=
and f(A)
A^ Then the closure A of A in X is a A. Thus there is an /: A" > 7 such The function g = f\Y: Y * I satisfies
and g(A)
=
1,
thus showing that
Y
is
completely regular.
8]
TYCHONOFF IMBEDDING, COMPACTIF1CAT10N
8.4.
277
DEFINITION
of all
maps
/:
X
Given a topological space
*
/.
X
map
consider the set
T:
Define the Tychonoff
X
*
/*
by
setting
(*)
for
**.
then
The
continuity of
8.5.
T
follows from v,5.3.
LEMMA
there is
If
X
is
completely regular and
A
C X
is closed,
a subcube Q of 7* such that T(A) = Q C\ T(X). PROOF. Let a be the subset of consisting of all maps /: which are not identically zero on A. Let x e X] the condition T(x) is equivalent to f(x) = X * 7 such that /(A) = 0. for all /:
X
7
e
I
a
complete regularity this is equivalent to x e A. T(X). THEOREM 8.6. For every topological space
ditions are equivalent:
(i)
(ii)
Thus T(A)
=
By
a
I
C\
X
the following three con
X is completely regular,
the
Tychonoff
map
T:
X
>
7* defines
a homeomorphism of
X
with T(X),
(iii)
PROOF.
* is ) points It follows that T(x )(f) Tfa). Thus T(x )(f) so that T(xQ ) If A is any closed subset of X, it follows from 8.5 that T(A) 11. closed subset of T(X. Thus the inverse of T is continuous and
X is homeomorphic with a subset of a compact space. are two distinct Suppose X completely regular. If X = 0, /(zi) = of X, then there 7 with /(x an /: X
is
Q ,XI
1.
^
l
^
T
is
is
a
is
T
a homeomorphism.
that
it
(ii) >
(iii).
Thus
(iii)
(i)
*
(ii).
Since 7*
is
is
compact,
it
follows
Finally, since a
>
compact space
then
completely regular,
follows from 8.3 that
(i).
LEMMA
extended
to
8.7.
//
X is any space,
7*
>
any map
is
f:
T(X)
.
*
I can be
a
map
/':
7.
PROOF.
y
Clearly g
=
7* define f'(y)
a:
T(x) for some
s
Then /': y(g). and we have X,
y(g)
=
fT:
X
>
7
an element of
7.
7*

If
y
s
For each then y = T(X),
/'(*/)
=
=
r(x)to)
=
^(x)
= /r() =
/(y).
Hence
of
X
Y
/' is an extension of /. DEFINITION 8.8. Let X~ be a compact space and X C X~ a subspace X~. We shall say that X~ is a Tychonoff compactification of X if X > 7 admits is dense in X~ (i.e. X = X~), and if every map /:
an extension /~:
LEMMA
8.9.
Let
X~ X~
>
7.
and Y~
be Tychonoff compactifications of
f:
X and
respectively.
Then
every
map
X
>
Y
admits a unique extension
278
SPECIAL FEATURES OF THE CECH THEORY
PROOF.
[CHAP.
X
In view of 8.6, we may assume without loss of generality A Then the map / is described by a subset of some cube I > I for a e A. Each / admits an means of coordinate maps / and together they yield an extension_/~: X~ extension fl: /, JT > I A of /. Since = JT, it follows that f(X~) = f~(X) = f(X) Y~. The uniqueness of /~ also follows Y =_Y~. Thus f: X~
that
F~
is
.
:
^ X
X
C
from
X=
JT.
8.10.
THEOREM
2?ven/ completely regular space
X admits a Tychonoff
The compactification is unique in the following compactification X~. X~ and X* are two Tychonoff compactifications of X, then there sense: If X~ ~ X* which is the identity on X. is a unique homeomorphism h: > /* be the Tychonoff map of X. It follows PROOF. Let T: from 8.7 that T(X) is a Tychonoff compactification of T(X). Since X is completely regular, T(X) and X are homeomorphic, so that X also
X
has a Tychonoff compactification. Let X~ and X* be two Tychonoff compactifications of X. By 8.9, > X* g: admits extensions h: X~ the identity map i: + > X+ are both the JT ~> JT. Then ^: X~  X~ and %:
X
X
t
X
identity
h~ and
l
Thus gh = identity and hg = identity. Hence a homeomorphism. This completes the proof. In view of 8.10, we may regard the Tychonoff compactification
on
X
.
=
ft
is
JT
the
as a well defined space. Then of a completely regular space from 8.9 that the functions JT,/~ form a covariant functor
X
~ on
it
follows
category of completely regular spaces and continuous maps with values in the category of compact spaces and continuous maps.
closure
a normal space and A is closed, then the Tychonoff compactification A~ of A. /. PROOF. Let /: A By Tietze's extension theorem, there is 7 of /. The^map g injburn admits an extension an extension g: X~ > /. This extends f to A so that A = A~, and the proof is g~\
LEMMA
8.11.
If
X is
CX
A
of
A
in
X~
is the
X
complete.
If
__
X
is
normal and
A
C
closure of
in
A
in
X and the notation A~ for the closure of A
X, then_we
shall use the notation
A
for the
in
JT, which,
(A)"".
view of
8.11, coincides
with the Tychonoff compactification
Clearly
_
A = XC\
__
A~.
y
A is a closed subset of the normal space Jf, then the pair (X A) be called a normal pair. The category of normal pairs and their continuous maps is denoted by &# If (X A) is a normal pair, then by If /: (X,A) > (Y,B) is a map of normal 8.11, (JT,A~) is a pair. then also f: (JT,A~) (F~,B~). Thus may be regarded pairs, as a covariant functor ~: Q N > &<?
If
will
t
~
9]
HOMOLOGY THEORIES FOR NORMAL SPACES
9.
279
HOMOLOGY THEORIES FOR NORMAL SPACES
Our next objective is to convert the category Q, N of normal pairs into an /icategory in such a manner that the Tychonoff compactification becomes an /ifunctor (see iv,9). Several preliminary results are needed.
then
LEMMA 9.1. If A and B are closed subsets of the normal space X, A~ Pi B~ = (A C\ B)~ and A~ \J B~ = (A \J )~. PROOF. Since A C\ B C A~ C\ B~, it follows that (A H B)~ C
A~ P\ B~. Suppose x is not in (A B)~. Let V be a closed neighA and borhood of x in JT such that V C\ (A C\ B)~ = 0. Then 7 B are disjoint closed sets in X. Let /: X > I be a Urysohn function such that J(V C\ A) = and /(B) = 1. Then f: X~  I satisfies = and /~(IT) = 1. If X Q e A", then x e (F A)" f((V C\ and /~($ ) = 0> thus x is not in J3~, and consequently x is not in
O
H
AD
H
A^
the
J5^". This proves A~ P\ B~ C (A C\ B)". The second part of lemma is a general property of the closure operation. THEOREM 9.2. // the inclusion map i: (X^AO C CX",A) of normal
Pi
pairs is
an excision (type (E) of 5), then the same is true of C. PROOF. By definition, X ,A are closed in X, and
l
l
A
The
l
= X,
C\
A
,
X=
Int
X, \J
Int
A
last condition is equivalent to
xWe
By
will
9.1,
x, r\
x
A =
o.
show that the same
relations hold for the compactified sets.
we have
A? = xr
Since
n
A~,
x  xr n x X
C\
A~ =
o.
A
is
closed in
X, we have
is
X
A C X" dense in X"
A~. Since X A~. Therefore
A, and this implies A is dense in X", we have that
A" =
X
XSince the same holds with
A~ = X~  A~
replaced by
A
X
lt
we have
o.
x~  xr
Thus
i" is
n x~
 A~ =
i:
an
>
excision.
LEMMA
9.3.
//
X
is
normal and
X C X~
9
then the
map
t"
1
:
Cov^X)
is onto.
280
SPECIAL FEATURES OF THE CECH THEORY
Let a be a
e
[CHAP.
X
PROOF. For each v
vj
finite
7
f
= X~  (X  ,)~. Then, by 9.1, = x~  n (x  a.)*" = x  (n (x  <*,))~ = x~  (x  vj a.r  Jr.
define 0.
open covering
of
X indexed by the set V.
Therefore
X
a

JC
n
j8
is
CY 
an open covering of
a.)"
X~
=
Xe
CST

indexed by V.
Since
X C\
l
(3,
=
a.
a.)
=
a,, it
e
follows that i'
ft
=
DEFINITION 9.4, If a dexed by the set V and
W
Cov(X) and
j8
Cov(F) are coverings
in
respectively, then
we
X
ft
e
CovCY X F)
indexed by the set
(
V X
W as follows:
define the covering
X0)<,,,)
=
a.
X
ft.
9.5. and Y are compact, then coverings of the form // a e Cov'X, ft e Cov'Y farm a cofinal subset of Cov(X X F). ft PROOF. Let y e Cov(X X Y). Since X X F is compact, and sets of the form U X W, U C F, 7 and open, form a base for X, the open sets in there is a finite covering 6 of X X F such that X Y, For each x e X and 9 7 < d and each set 8, is of the form 5, = C7, X
LEMMA
X
a
X
C
W
W
X
W
.
t/
e
F
define
U(x)
W(y]
= =
C\
U
v
n TF.
X
Thus y
where where
a;
e
e
U.,
t/
W,
.
The
ft
distinct sets
of
X
among
and
9
F
some
set
U X
9.6.
Each set 17(z) respectively. a TT. = 5. so that 5 ft
the t/(x) and W(y) form finite coverings a and X W(y) is contained in
<
>
<
a
X
ft
LEMMA
subset of the
Le t: A C space X. Let f:
X be an inclusion map where A
A
is
a dense
F
where
l
Y
is
compact.
ft
l
The
Y if and only if, for every covering e Cov r (F), f e Cov (X) such that f~ < i~ a. PROOF. Assume that the extension F: X > F of / exists. Given r any 7 e Cov (F), define a = F~ (y). Then, since / = Fi, we have = i~ F~ y = i~ a. Thus the condition is necessary. f~ y
has an extension F: there is a covering a
>
l
X
map f
l
l
l
l
To prove
t'ke
the sufficiency of the condition, consider for each x
all
e
X
family N(x) of
S v = f(V
H A).
neighborhoods of
x.
For each
If
V

19
Fw
S Vl
s
#(x), and r\ C\
>S r
C
S Vl
H
,
N(x), then C\ S Vn Since S v
e
.
Vn
V e N(x) V = 7 H
X
define

O
that
}
^
0, it follows
S Vn
j* 0.
has a non vacuous intersection.
F is compact,
is
itr
follows
number of subsets of {S v S v are closed in F, and that the intersection F(x) = C\S Y for V e N(x)
Thus any
finite
Since the sets
nonvacuous. Indeed let y^y2
We
shall
F(x).
Select a covering
prove that F(x) consists of a single point. r ft e Cov (F) consisting of
9]
HOMOLOGY THEORIES FOR NORMAL SPACES
l
281
two sets ft,ft such that y e ft l such that f~ ft < i~ a. Let a: family N(x), and
l
ft,
e
;
y2
e
ft
ft.
a f then
a,
= V
is
Let a e Cov'(X) be an element of the
Since /(a, C\ A) ft. f(a t r\ A)
is contained^ in either ft or ft, we may assume Then S v C ft so that y 2 is not in *S r and therefore y 2 is not in F(x). Thus F(x) is a single point. If x e A, then /(x) e S v for all V e N(x) so that /(x) e P\ S v and therefore F(x) = /(x). Thus F is an extension of /. It remains to
C
,
,
be proved that F is continuous. Let U be any open set in Y containing F(x). Choose a covering e Cov'(F) consisting of two sets ft and ft such that F(x) e ft ft ft, l and ft C f/. Let a e Cov'(X) be such that f~ ft < i~ a and let x e a..
l
Since a, = /(a, P\ yi).
/(a, P\ A), it follows that F(x) e contained in either ft or ft, since  ft, it follows that /(a, P\ A) F(x) e /(a. A), and since F(x) e fa For every x' e a,, we have a, = V e W(x') so that F(xO e S K = ft. U. Thus F(a,) ft A) (/a, U, which proves that F is cone
F
AT(x)
and S v
=
n
C
Since f(a v C\ A)
is
C
n
C
C
tinuous.
LEMMA
and let f: if and only
PROOF.
9.7.
Let
X,Y
>
be completely regular,
XX Y
if for every
l
Z.
e
An
extension F:
there exist
let Z be a normal space, X~ X Y~ > Z~ of f exists
y
Cov r (Z)
a
Cov'(X) and ft
e
Cov r (F)
such that f~ y
<
a
X
ft.
Consider the inclusion maps
t\:
X
X^,
i9 :
Y
> r',
t:
Z
> ZT.
/ Assume that an extension F: X~ X Y~ > 7T is given. If 7 e Cov (Z), r then by 9.3 there is a y' e Cov fZ~) such that 7 = i~V By 9.5 there r such tiiat F'V < exist coverings a' e Cov'(X~), ft' e Cov X jS'. r Let a = zTV, )8 = ia y. Then a e Cov'(X), ft e Cov (F), and since = F(Z! X i 2 ), we have i'/
(O
l
'
l
f~ y
= f~ i~*y' = zVV X
l
l
(i\
X
a
l
i2
)F~ y'
ft.
<
1
(t'J"
X
tjOC"'
X
j80
z7^'
=
X
Conversely, assume that the condition of the lemma is satisfied. Y  Z~, and let 7 e Cov r (Z~), then Consider the map g = if: / / r i~ 7 e Cov (Z), and there exist coverings a e Cov (X) and ft e Cov (F)
XX
!
y < a X ft. Then (/"V = f~ i~ y < a X ft, and by Z~. Clearly F X~ X Y~ 9.6, the map g admits an extension F: is also an extension of /, which concludes the proof. DEFINITION 9.8. Let (X,A) and (Y,B) be normal pairs. A homosuch that
r
l
l
l
l
i'
282
SPECIAL FEATURES OF THE CECH THEORY
[CHAP.
X
(Y,B) is called uniform if, for every finite topy F: (X X /, A X /) / / open covering 7 e Cov (F), there exist coverings a e Cov (X) and ft e Cov'(7) such that F~V < a X 0. As an immediate consequence of 9.7 we obtain THEOREM 9.9. Let / ,/r. (X,A) (F,J5) 66 maps of normal pairs. A homotopy F: (X X /, A X /) > (F,B) between /<> and fi is uniform (X~ X /, A~ X /) if and only if F can be extended to a homotopy G:
 (F~ B~). (F~ B~) 6etoeen tfe maps fafr. (X~,A~) Theorems 9.2 and 9.9 indicate clearly how to convert the category The couples in a# are inclusions i: A d X, Ofctf into an /icategory.
(X,A) where (X,A) is a normal couple. Excisions are defined j: as usual, namely, type (E) of 5. Homotopies are defined to be uniform now can state homotopies. Points are defined in the ordinary way.
XC
We
THEOREM
functor
9.10.
Gi N
>
The Tychonoff compactification
<2 C .
is
a covariant h
~:
DEFINITION 9.11. Let be a partially exact homology [cohomology] as a theory on Ct c and denote by If theory on the category GL C Regard the composition of with the /ifunctor ~: Q N > Ct c The result is a partially exact homology [cohomology] theory on G^, called the normal
>
H
H
H
.
space theory associated with H. If // is exact, so also is If. Since Q c is clearly a subcategory of Ct^ and since (X~,A~) from for compact pairs, it is clear that If is an extension of
=
&c
(X,A)
to Oty.
H
is the Cech theory on Q c then the associated theory If on be given a direct description in terms of nerves of coverings. may We recall that in ix,3, in addition to the Cech groups Q (XjA), we f f defined also the groups Q (X,A) using the finite coverings Cov (X,A) for the* passage to the limit. We will show that, if the pair (X,A) is g (X~,A~). normal, this group is essentially identical with H~(X,A) = THEOREM 9.12. Let (X,A) be a normal pair and let H((X A) 9 [H f (X,A)] be the homology [cohomology] groups defined using the Cech method and finite open coverings. Then the inclusion map i: (X,A)
If
H
,
01^
H
H
H
y
C
(X~ A~) induces isomorphisms
9
V
in ix,8,
H ,(X,A) =
f
H,(X~,A~)
[i*:
is
H(X~,A~)
and
f
MX, A)].
PROOF.
Cov'(X).
Since the subset
A
closed,
in
view of the discussion
we may
= a e Cov r (X~) and and A is dense in A~,
that
If
By
replace the directed set Cov (X,A) by the directed set 1 Cov^JO > Cov'(X) is onto. If 9.3, the map
T
:
i"
it
# CX7,A~) = H
a
Q
(X
ft
is dense in X~ then, since follows readily that (X7,A) = (X^A ) so Thus the fact that i+ [i*] is an iso,A0).
l
a
e
Cov r (X"),
X
ft
morphism
is
we combine
a consequence of vm,3.15 and vm,4.13. the above result with the obvious fact that the
9]
HOMOLOGY THEORIES FOR NORMAL SPACES
i
283
//'
isomorphism  IT.
#
commutes with /^ and d we obtain an isomorphism
Let f:
THEOREM
morphism
that
if
9.13.
(X,A)
>
(F,Z?) be
>
a
relative
is
homeomorphism
set
of normal pairs.
Then
if
f:
f(C)
is
(X~,A~)
is closed
(F~,ZT)
a re/atoe feomeo
C
CX
and only
A.
in
Y for
each closed
C
of
X such
A
PROOF:
is
Suppose /~
a relative homeomorphism, and
C\
C
CX
closed in X.
By
9.1,
C"
A~ =
0.
Hence /~ maps C~ topologically
into
F~
JS~,
Since
C =
and /~
is
XH
and
(T = (X
(7~,

A)
H (T, n
is
11
on
X
A
we have
= (F As
5)
n rtfn
= Y
(7" is compact, f~(C~) is closed in F~. This proves the necessity of the condition.
Hence /(C)
closed in 7.
Assuming the condition fulfilled, we will show that /~ is a relative homeomorphism. Since Y is dense in Y~ and B C B~, it follows that Y  B is dense in Y~  /T. Therefore Y~  B~ C .T(X~). As
C
B~, we have
be closures of Let Q ,N 4", and x ?^ x,. Suppose J,,,^! e Y" neighborhoods in A"~ of J ^i respectively, such that A,N Q ,Ni are pairis wise disjoint. Let C, = AT, Pi .Y and 7) = /(C ) for i = 0,1. Since
l
N
t
f
X
dense
Since
in Y.
in
X~,
is
it
follows that
(7 % is
N
t
closed,
J
T
l
t
is
closed in
r N,, and therefore (T, = A X. Then, by hypothesis, D, is closed
dense
in
i
.
Thus B
D D
()J
are closed disjoint sets of F.
By
9.1,
S",D^,D7 are
disjoint.
Since
/~(*.)
e
HAT.)
l
=
/~(C7)
C
D~,
i
=
0,1,
it
follows that /"(x
)
and f(x
)
are distinct points of
F"
B~.
This
combined with the
defines a 11
is
result of the preceding
paragraph shows that /~
.
map
of
X~
A~
onto F~
B~
Then
5.2 states that
f
a relative homeomorphism. on Q c is invariant under relaREMARK. If the homology theory tive homeomorphisms, e.g., the Cech theory, then we may use in a^ the larger class of "excisions" described in 9.13. Also, if // is invariant
H
under excisions of type
(E,), the
same holds
for
H~
.
234
10.
SPECIAL FEATURES OF THE CECH THEORY
[CHAP.
X
COMPACT PAIRS AS LIMITS OF TRIANGULABLE PAIRS
The objective of this section is to prove the following theorem: THEOREM 10.1. Every compact pair is homeomorphic with the inverse
limit of
an
The proof
inverse system of triangulable pairs. will be preceded by two lemmas.
10.2.
LEMMA
(Q,R)
//
Q
is
a finite cube,
R
is
a subcube of Q, then the pair
is triangulable.
PROOF.
tion of
;
Since Q is a finite product of intervals an iterated applican 8.9 yields a triangulation of Q in which R is a subcomplex. LEMMA 10.3. Let Q be a finite cube, R a subcube, X a closed subset
of Q,
Y
z
th
C U such that X C
Q Then there exists a set (Y,Y C\R) is triangulable. PROOF. Let T be a triangulation of (Q,R) and let T denote the barycentric subdivision of T. Select a closed set X' such that
and
U
an open
set
such that
XC UC
the pair
Int
Y
and
%
XC
%
Int X'
CX C
1
f
U,
interand choose i large enough so that every closed simplex of X' is in U. Such an i exists by n,6.5. Let Y be the union of secting Then Y and Y C\ R are suball those simplexes of T which meet X in the triangulation *T hence (Y,Y C\ R) is triangulable. complexes Since X' C Y, it follows that X C Int Y, and the lemma is proved. PROOF OF 10.1. Let (X,A) be a compact pair. By 8.6 and 8.5, we may assume that X is a closed subset of a cube 7* and that A = X C\ R where R = I" is a subcube of /*, a C For each subset a of we shall consider the subcube 7 and the 7* 7. projection p a Let denote the set of all pairs m = [a,N] where
'.
r
T
,
,
:
M
a
(1)
a
is
a
finite
subset of
,
(2)
(3)
N C 1, p X C Int N
(relative to
O,
m
2
the pair (N,N r\ R) is triangulable. A quasiorder is defined in as follows: = [a 2 ,N 2 ], then m^ < 2 provided a2 (5) a
(4)
M
If
mi
=
[a^A^] and
m
l
(6)
Pai
C N C
,
2
To prove
define
#1. that
M
is
directed, let
m =
l
[a^TVj],
m
a
2
[a 2 ,N 2 ],
and
a
=
a,
Ua
2,
B =
pa X
7
n p[(Ifj H p' l(N
B C
7
a
,
2 ).
Then
C
Int
M
and by
.
10.3 there
l
is
a set
NC
2
Then
m < m and ra <
= [a,N] 7 such that is directed. m hence
M
m
is
an element of
11]
MAPPINGS OF SPACES INTO NERVES
285
We now define an inverse system of triangiilable pairs
indexed by the set (N,N C\ R). If mi
the
M
as follows:
[ai,Ni],
If
m =
=
m =
2
l
[a,N],
then
,
(Xm ,A m ) =
[a 2 ,N 2 ]
map
I*
is
of
>
p
fll
:
(N 2 ,N 2 P\ R) into (Ni,N ai 7 The verification that
.
and mi < ra 2 then 7r is R) defined by the projection this defines an inverse system of
H
spaces
Let (X^.A^) be the limit of this inverse system. map f m (X,A) > (X m ,A m ) defined = m and 7* I* which carries into by the projection p a A C\ R into TV C\ R = A m The system of maps {f n constitutes a map of the pair (X,A) into the system (X n ,A m ), and defines a limit map (X A) > (X.jA.) (see vm,2 and vm,3). /,:
immediate.
For each
m =
[a, AT],
consider the
:
:
X
N
}
X
,
X
.
9
We
that
shall
.
prove that /
B
]
is
a homeomorphism.
an
)
Let a be any
it
finite
subset of
Since, by 10.1, the pair (7,7
is
is
triangulable,
follows
m =
is
[a,7
where a
an element of M. Let x e m and set x(a) = ir m (x) finite subset of and m = [a, 7]. We prove the folany
X
,
lowing properties of x(a): (7) x(a) e Po X.
A., then x (a) e 72. a 2 then p ai x(a 2 ) = x(a ). <7j (9) If m' = [a,N], then ir m .(z) = x(d). (10) To prove (7), assume that x(a) is not in p Q X. Then, by 10.2, there is but o*(a) = 7r^'x m = x m e A7 which an m' = [a,N] with x(a) not in provides a contradiction. Propositions (8), (9), and (10) are obvious. Condition (9) implies the existence of an element x' e 7* such that = x(a) for all finite subsets a of It is easy to see that x' e X, Pa(x') = x, and that x' e A if x e A*,. This shows that / maps (X,A) ) /^(x onto (Xv,A m ). and Xj 5^ x 2 Then, for some finite subset Suppose that Xi,x 2 s a of p a x! 7^ p a x 2 Hence ir m Xi ^ ir m x 2 where m = [a,7], and therefore /co(xO 7* /oo(.T 2 ). This concludes the proof.
(8) If
If
a:
e
C
>
}
N
m
,
,
,
.
7
X
.
.
,
11.
CANONICAL MAPPINGS OF SPACES INTO NERVES
with nerve For DEFINITION 11.1. Let a be a covering of a the vertices v of a such that x e a, form a simplex of a every x e that is denoted by A a (x) and is called the simplex determined by x in a a such that x is in the carrier Clearly A a (x) is the largest simplex s of
.
X
X
X
y
X
X X
.
X
of
s.
LEMMA
l
11.2.
Let f:
e
X
>
Y,
let ft
be
a covering of Y, and
let
a
=
f~
fi.
For every x
X,
A a (x) =
286
SPECIAL FEATURES OF THE CECH THEORY
This follows from the fact that, for every vertex l s /3, and x e a, = f~ (& 9 ) are equivalent.
11.3.
v of
[CHAP.
0J
X
Y
the rela
tions f(x)
DEFINITION
Let a be a covering of
*:
X
with nerve
X
a.
A
mapping
Xif,
Z.
e
s called canonical relative to
a
>
for every #
X, the point 0(x)
F0 ts
tf*aZ
is
in
the closed simplex
A a (#).
// /:
LEMMA
and a
<f>f
11.4.
l
X
Y and
</>:
F
>
a canonical
map
and
=
f~
(0), tfien
</ maps
X
into
Xa
(recaM
X C
a
Y
ft
),
is
a canonical map relative to a, This follows directly from 11.2.
As a
\
LEMMA
defines
11.5.
If
<t>:
X
>
\X a
is
.
corollary of 11.4, canonical and A
we have
X, then
<t>
C
map*: LEMMA
a.
a canonical map of A into (X,A) + (\X a \,\A a \).
11.6.
Aa
We
then speak of the canonical
be a refinement of
Let
a,/3
be coverings of
ft
X, and
let /3
If
ft ,
</>:
p:
is
(X A
PROOF.
(,),
 (X a ,A a
Let x
e
(X,A)
*
)
(\X \,\Afi\) is canonical is a projection, then p<t>:
let v
to 0) and (X,A) > (X a ,A a )
(relative
canonical (relative
it
to a).
X
and
be a vertex of A^(x).
Since x
e fl,
C
follows that p(v) is a vertex of A a (x). Hence p maps the (closed) simplex &p(x) into the (closed) simplex A (x), so that p<t>(x) e A(z), and p<t> is canonical.
ft
LEMMA
11.7.
Any
two canonical
maps
<
,<i*
(X,A)
>
(X a ,^4
a
)
are homotopic. This follows at once
x
e
A), the points
</>
(x)
THEOREM
PROOF.
11.8.
X there exists a canonical map
By
reduction of a.
from the observation that, for every x e X (or and <t>i(x) are in the same simplex of X a (or A a ). For every finite open covering a of a normal space
<t>:
3.3, there exists
> \X a a closed covering
\.
X
of
X
which
is
a
By
Urysohn's
:
Lemma
Vp, a continuous function / f such that
X
f v (x)
f,(x)
there exists, for each v e V a = >/(/ = closed unit interval [0,1])
= =
for
1
x
e
X
</>,
a,,
for
ze/?,.
Since VJ 0,
/,//.
=
Then
X, the sum /(x) = J^ t /,(z) is positive. Define = 1, and 4>,(z) = for a; not <f>,(x) ^ 0, X) *.W
=
in a,.
Therefore
4>(x)
=
tt .
E *.(>
is
a canonical
map
<#>:
X
>
X
12]
CONTINUITY UNIQUENESS THEOREM
287
Using the concept of a canonical map we prove the following approximation theorem that will be used in the next section: THEOREM 11.9. Let (X,A) be the inverse limit of the inverse system of compact pairs {(X m ,A m ),ir"\\. For every mapping f: (X,A)  (Y,B) where (F,B) is a triangulable pair, there exists an index m and a map
fm (X m ,A m ) (F,B) such that the maps f and fm ir m of (X,A) into are homotopic. (Y,B) PROOF. Let T = {t,(K L)} be a triangulation of (7,5), and let r be the associated covering of Y as defined in ix,9.1. Then (Yr ,B T ) (K,L), and the map r (Y,B) > (\K\,\L\) is canonical. Let a =
t
'
1
:
/"V; then (X a ,A a )
contained in (K,L), and, by 11.4, the map t~ f is canonical. We now apply 3.8, and find an index and a covering is a refinement of a. and m such that the covering 7 = Tr" /? of 8 of
is
l
m
X
1
X
(X 7 ,A T ) = (X^A^).
Let
p:
(X y ,A y ) > (X.,A a )
be a projection, and
let
be a canonical map which exists by virtue of 11.8 and from 11.4 and 11.5 that the map
11.5.
It follows
*w m
is
:
(X,A) > (\X,\,\A yi \)
canonical (relative to 7), and therefore, by 11.6, the
p07r m
:
map
(X,A) + (\X a \,\A a \)
1
canonical (relative to a). Since t" f is also canonical (relative to a), 11.7 implies that the maps p07r m and r / of (X,A) into (/C,Z/) are homotopic. Since t is a homeomorphism, the maps tp^ir m and / of
is
1
homotopic. conditions of the theorem.
(X,A) into
(F,JB) are
Thus
the
map f m =
tp<f>
satisfies
the
12.
THE CONTINUITY UNIQUENESS THEOREM
be proved in this section that the Cech theory is essentially the only one satisfying the continuity requirement. be two partially exact homology [coand THEOREM 12.1. Let on the category & c of compact pairs, and let homology] theories defined
It will
H
H
H
be continuous.
Let G,G Then any homomorphism
be the coefficient groups of
H,H
respectively.
h
:
G  G
[ft
;
<?><?]
288
SPECIAL FEATURES OF THE CECH THEORY
to
[CHAP.
X
can be extended in just one way
h:
a homomorphism
H
>
H
:
[h:
H H
+ H]
[h Q
:
on
ac
is
.
If H is also continuous and h
then h is also
G ~ G
[h:
G =
on
G]
an isomorphism,
an isomorphism
h:
H ~H
Any
H]
ac
.
Taking
H to be the Cech theory,
12.2.
Ct c
one obtains
THEOREM
continuous partially exact
mology] theory on the category
homology [cohoof compact pairs is isomorphic with the
Cech theory over the same coefficient group. In this theorem the values of the homology theory are either in Q R or g<7. The cohomology theory has values in g^. The Cech cohomology theory with values in g c are in general not defined, and the concept of continuity for a cohomology theory with values in g c was not
defined either (see
2).
The proof will be restricted to homology. The proof for cohomology is dual. On the category 3 of triangulable pairs, both and are exact theories. By the Uniqueness theorem iv,10.2, we may assume that the
12.1.
PROOF OF
H
H
homomorphism
h:
is
H
>
H
on
3
already given. Let (X,A) be a compact pair. By 10.1 there exists an inverse > system (X,A) in Inv3 and a homeomorphism <: lim (X,A) (X,A). The pair consisting of (X,A) and of will be called a development of (X,A) and will be denoted by the letter Z>.
<f>
If
M
is
the indexing set of (X,A), then, for every
m
e
M, we have
a
homomorphism
A(^,X m ,^4 m):
Their totality
(1)
is
H (X mt A m
g
)
H (X m ,A m
Q
).
a homomorphism of inverse systems
%,X,A):
2.
H.(X,A) >
# (X,A)
fl
as defined in
Using the homomorphisms of
2.1,
we
obtain the
diagram
H (X,A)
9
<
#
(lim (X,A))
>
lim
# (X,A)
a
12]
CONTINUITY UNIQUENESS THEOREM
</>
289
where /io,(</,X,A) isjthe inverse limit of the map (1). Since is a homeoare isomorphisms. Finally I is an isomorphism by morphism, <^ and virtue of the assumption that // is a continuous homology theory.
^
Define the homomorphism
h(q,X,A,D):
H
q
(X,A)
>H
9
(X,A)
by
setting
h(q,X,A,D)
= ^r'JLZ*;
1
.
We
(2)
shall establish the following properties of h(q<X,A,D):
If If
(X,A)
(3)
D
1
is
then h(q,X,A,D) = h(q,X,A). another development of (X,A), then h(q,X,A,D)
e 3,
=
A( 9 ,X,A,D').
(4) If /:
and (F,B)
and D' are developments of (X,A) (X,A) * (7,J3) and respectively, then commutativity holds in the diagram
h(q,X,A,D)
D
H,(X,A)
>
H.(X,A)
h(q,Y,B,D')
H
(5)
t
(Y,B)
>
H
a
(Y,B)
Let
D
f
denote the development of
A
induced by the development
D
of (X,A), then commutativity holds in the diagram
h(q,X,A,D)
H (X,A)
Q
>
H.(X,A)
d I
h(q

1,A,Z>')
Propositions (2)(5) imply that_setting /i(#,AV4) > extending A yield a homomorphism A: //
=
A(},X,.4,D) does
H
.
Without any
loss of generality
we may assume
and
(F,B) actually are
the inverse limits of (X,A)
<f>
(Y,B),
that (X,A) and and therefore
that the homeomorphisms
and
0! are identity
maps.
We
begin by
proving two auxiliary
propositions.
290
If
SPECIAL FEATURES OF THE CECH THEORY
(6)
?r
[CHAP.
X
m
:
(X,A)
>
(Xm ,A m )
holds in the diagram
H.(X,A)
is
a projection, then commutativity
h(q,X,A,D)
>
H,(X,A)
H,(X a ,A m)
Consider the diagram
lim#,(X,A)
H,(X,A)
\r m
H
in
v
(X m A ,)
,

\r m
H,(X A)
9
_
>
//
(X W
,
.4
)
groups lim H q and Commutativity in the square is a consequence of the definition of /i*, and commutativity in the triangles follows from the definition of / and of /. Hence
which r m and
r m are the projections of the limit
lim
H
a
into their
m
th
coordinates.
=
This proves
(7) If /:
(6).
h(q,X m ,A m )T m l
=
h( q>
X m ,A m
)ir n *.
(X,A)

(F,B) and (F,B)
is
in 3,
then commutativity
holds in the diagram
h(q,X,A,D)
H (X,A)
Q
>
.(X^)
h(q,Y,B)
H,(Y,B)
By
11.9
we may
select
an element
m
of the indexing set
M of the
system (X,A), and a map
fm
:
(X m ,A m ) > (F,B)
12]
CONTINUITY UNIQUENESS THEOREM
291
such that
i J m tr m
~ 
f J
where
ir
m
\
(X,A)
>
(X m ,A m )
is
the projection.
Consider the diagram
H
q
(X m ,A m )
h(q,Y,B)
Commutativity in the square, commutativity
is
homomorphism by the Homotopy axiom,
a
square follows from (6); and, in the right > a consequence of the fact that h: on 3 and both (X m ,A m ) and (Y,B) are in 3. Since,
left
is
H
H
proposition
(7) follows.
PROOFS OF
(X,A)
that
(2)(5).
Proposition
(2)
follows from
(7)
by taking
=
(Y,B) and /
=
identity.
(7)
implies h(q,X,A,D)
=
Then /^ and /^
are identities so
h(q,X,A).
Next we prove (4). Let n }>e an element of the indexing set N of and let Pn >', (Y,B) > (Y n ,B n be the projection. Since (F n ,B n ) is in 3, proposition (7) applied to the maps p n and pj yields
:
)
Consequently
Since this _holds for every n e A the^ above relation holds with p n + replaced by I Since, by assumption, 77 is a continuous homology theory,
,
r
I
is
an isomorphism, and therefore
7 % /i(g,X,A,7>)
=
h(q,Y,B,D')f+
which proves (4). To prove (3) apply (4) with (X,A) = (F,B) and/ = /, and f+ are both identities and (4) yields h(q,X,A,D)
identity.
Then
;
=
A(g,X,A
D ).
x
292
NOTES
(5) is
Proposition the diagram
a consequence of the commutativity relations in
where d>
#>.
is
the limit of
d:
H
>
Q
(X n ,A m )
>
#<,_! (^l m )
and similarly
__
for
This concludes the proof
(2)(5).
h:
We now
unique.
show that the homomorphism
h':
H
+
H
extending h Q
is
Let then
H
by iv,10.2, we have h
loss of generality
=
h'
we may
be another such homomorphism. Then, on 3. Let (X,A) be a compact pair. Without assume that (X,A) is the inverse limit of an
:
H
{(Xm ,A m \^\] of triangulable pairs. Let ir m (Xm ,A m ) be the projections. Since (Xm ,A m ) is triangulable and h and h are homomorphisms of homology theories, we have
inverse system (X,A)
=
(X A)
9
>
r
Since this holds for every element
m of the indexing set,
it
follows from
the definition of the
map
J(g,X,A) that
Since
H
is
continuous,
/
is
an isomorphism, and therefore h(q,X,A)
:
=
h'(q,X,A).
Finally assume that
G
h:
~_(?
is
H
H H
H is also a Continuous theory and that h Q an isomorphism. Let hoi G = G be the inverse^ of h and Jet > ff Then hh: be the homomorphism on a<? extending A
.
is
a homomorphism extending the identity
by the uniqueness condition, Thus h is an isomorphism and h
M
=
map G_*G. Thus
hh
is its
identity. Similarly inverse.
=
identity.
NOTES
The Cech process. The method used in defining the Cech homology and cohomology groups is not restricted to these two instances. The general situation may be described somewhat as follows: Suppose that a functor // (covariant or contra variant) on the category of simplicial pairs and simplicial maps with values in the category of groups and homomorphisms is given. Suppose further that, if /o,/*:
EXERCISES
l
29$
tions of
H
(K,L) > (#i,Li) are contiguous, then H(f ) = H(f ). Then the defini24 of Chapter ix can be applied to yield a similar functor
satisfies
For compact pairs the functor // ) fi(f\) if / and fi are homotopic) and the continuity axiom. This procedure was used by Spanier to show that the cohomotopy groups satisfy the continuity axiom (see
of pairs (X,A).
(i.e.
on the category
the
homotopy axiom
fi(f
note to Chapter
i).
Compactifications.
In defining the compactification
X
of a locally
compact space X, we required in 6 that the compactification be made by adding a point w which does not belong to any of the spaces in the This actually amounts to replacing the category by a category CL LC This highly artificial convention was caused by the resubcategory.
.
quirement that the compactification (X,A) of a pair (X,A) again be a pair; this forces us to compactify A using the same point as for X. The proper solution in this situation calls for the generalization of the concept of a pair. Let a generalized pair be defined as a triple > and A are topological spaces and <t>: A is a (X,A,<t>) where of A onto a subset of X. If X,A and are in a LC homeomorphism then we say that (X,A,<I>) is in GL LC Now the compactification (X,A,<) is a generalized couple even if distinct points are used to compactify and A. There remains the question of a suitable natural choice of a point w for each locally compact space X. To this end we remark that, in the von NeumannBernaysGodel axiomatics, a set may be an element of another set but never is an element of itself. Thus if [X] denotes the Thus we may set set whose only element is 0. we have [X] C\ = with the topology of 6.2. \X}
X
X
</>
,
.
X
X
XW
X
X
An
analogous discussion can be carried out for the Tychonoff com
pactification.
EXERCISES
A.
1.
WEAK
Show
CONTINUITY.
that every compact pair may be represented as a nested intersection of pairs in the category 3//, i.e. pairs having the homotopy
type of triangulable
10.1.)
pairs.
(Hint:
Use method employed
to prove
H
2. The conclusion of 2.6 will be referred to as the weak continuity property of a homology [cohomology] theory. Prove 12.1 assuming that is weakly continuous. (Hint: use preceding exercise and the fact
that
3/f is
a uniqueness category
(ix,
Exer. E4).)
&U
EXERCISES
3. Use the preceding exercise to prove that continuity and weak continuity are equivalent properties. 4. Show that, in the category G c the Homotopy axiom is a conse,
quence of Axioms 1, 2 and continuity. Math. 54 (1951), 247249.)
B. DIRECT
1.
(See J.
W.
Keesee, Annals of
Show
that the direct
SUM THEOREMS. sum theorem
9
i,13.2
remains valid in the
X OX
l
Cech theory (arbitrary pairs). 2. Let (X' Xi X 2 ) be a compact
9
triad with
X
=
X W X A =
v
2,
2.
Show
that the
maps
H,(X.,A)
yield
 H
9
(X,A)
9
i
=
1,2
an injective representation of
H
q
(X A)
as a direct sum, while the
i
maps
H.(X,A)

7/ f (X,X.)
=
1,2
yield a protective representation. (All maps are induced by inclusions, and Cech groups are used.) Establish similar results for cohomology.
3. Let a be a compact space, and let U a with union U. pairwise disjoint open subsets of
{
} ,
X
X
M, be a Show that
family of the maps
H
9
(X,X
[/)
ff a (X,X

C/ a )
induced by inclusion yield a projective representation of the Cech U) as a direct product. Similarly show that homology group H q (X,X
#'(X,X  U a )
yield
4.

a
Jf/
(X,X

J7)
an
injective representation of
H
q
(X,X
U) as a direct sum.
7.
Transcribe 3 into the "single space notation" of
C. EXACTNESS.
1.
Let
A
be an open and closed subset of a space X.
is
Show
is split
Q
that
(see
the Cech homology sequence of the pair (X,A) vm, Exer. D) by the homomorphisms \l/:
exact and
H
t = *;X>
2.
I:
{J X^ A = 2 ) be a compact triad with C\ Show that the Cech homology sequence of the triple (X,X29 A) 2 Xi > is exact and is split by the homomorphisms \l/: Q (X 2 ,A) Q (X,A)
Let (X]Xi
.
xC X
9
(X,X

Q
(X)
>
H
v
A),
fc:
A
C
(X,X

(A) where
A).
X
X
X
H
H
where
3.
(X,A) (X,XO, (X a ,A) (X,XO. fc; ^, Formulate similar results for the Cech cohomology sequences.
/:
fc:
^ =
1
C
C
D.
1.
LCGROUPS
Show
that, in the definition of
H*(X A)
9
9
the directed set
EXERCISES
295
may
resulting limit group
(a)
be replaced by any of the directed sets listed below and that the is isomorphic with H*(X,A).
Cov 2 (X) =
the set of
is
all finite
open coverings a of
X such that
at least one of the sets a,
all
countercompact.
such that (b) Cov (X) = the set of all finite open coverings a of the sets a v except perhaps one are bounded. Hint: Show Cov Cov 2 and Cov is cofinal in Cov 2 Cov!
X
C
C
a
,
.
(c)
Cov 3 (X),
sets of
the set of
all finite
/3
open
coverings of
ordering
defined as follows:
(3
means a
<
ft
but with the and the union of the
(Hint:
X
bounded
set
(d)
(e)
2.
contains the union of the bounded sets of a.
up a map Cov 3 (X)
Cov'(X).
Cov
(X)).
The
sets
s
Given a
Cov (x), Cov,(X), Cov 2 (X) with the ordering Cov f (X,A) define, in addition to the subcomplex
.
fi a
,
the subcomplex 12^ which is the full subcomplex of the vertices v for which a p is not bounded. Thus
X
'
a
determined by
is
fl a
the least
full
subcomplex
containing any coverings named above, one may replace the groups A used in the limiting process, by the groups Q (X aj
fi a
.
of
Xa
Show
that, for
of the families of
Q
H
a
(X aj A a \J
VJ
fi a ')
ft a ),
H
without
altering (up to
an isomorphism) the
limit groups.
The limit groups of Q (X a ,A a fi) over the directed set are those treated in detail by Alexandroff [Trans. Amer. Math. Cov'(X) Soc. 49(1941), 41105]. Thus the Alexandroff groups are isomorphic
REMARK.
Let [X a be the family countercompact subsets of X. Show that the Cech groups g J7 a (X,A X a ;G)] together with the homomorphisms ;(?) [H (X,A induced by inclusions form an inverse [direct] system of groups with a
]
H
U
with the Agroups. 3. Let (X,A) be a locally compact pair.
of all
UX
K
rt
U
limit
4.
group isomorphic with II*(X,A\G) [#^(X,4;G)]. be a locally finite simplicial complex and Let
of
K.
Show
that the pair
(/v,L) is locally
L a subcomplex compact and establish the
natural isomorphisms
with 3C a and 5C as defined in vin, Exer. F3. 5. Transcribe the relative MayerVietoris sequence of a compact triad (X;A lf A 2 ) into the single space language of 7.
E.
1.
TYCHONOFF COMPACTIFICATION. Show that, for any topological space X, any map
/:
X
I can
be factored uniquely into maps
property describes
> I. > Show that this T(X) T(X) among compact spaces up to a homeomorphism.
X T
296
2.
EXERCISES
Replace the interval 7 in exercise
Let
1
by any completely
regular
space.
3.
_ X be a normal subset of a compact space F such that X = Y and for any two subsets A,B of X the relation A C\ B C\ X = implies
A
nB
4.
=
0.
Show
that
F
is
Let
that a
X be normal and homotopy F: X X
there
is
t 2 ))
Y
/
a Tychonoff compactification of X. compact metric with a metric p.
>
Show
Y
is
uniform
t ly t 2
if
and only
t2
\
if
for each
5
>
p(F(x
5.
)
a
5
>
for all
tl)
)
F(x
y
<
such that x e X.
e
7,
^
<
implies
Let points x e
X be a normal space and let w(X) denote the set of those X such that x has a countable set N of neighborhoods with
the property that any neighborhood of x contains an element of N. Show that (X) = (JT). 6. Show that the Cech homology and cohomology theories of normal
pairs obtained using finite open coverings obtained using finite closed coverings.
is
isomorphic with the theory
F. SOLENOIDS.
fk
:
In these exercises use the fact (established in xi,4.5) that the map l 1 S > S defined by f k (z) = z* satisfies f t +u = ku for u e ff^S^G)
and /t(u)
for
ku for u e (S ;G). 1. Let S p be the padic solenoid the Cech groups
=
H
l
l
(see
vm,
Exer. E).
Show
that
Ho(2 p ;G)
# (S
a
p
;G)
= =
0,
0,
fl(S F ;G) H*(2 V ;G)
= =
for
g
>
1.
H^^G)
G
<
is
isomorphic with the limit of the inverse system
G
<
<
G
<
where
H
l
(2j,;G) is
isomorphic with the limit of the direct system
GG>
00
00 G >
1
.
(L v ]J) is Using the above results, prove that #i(S p ;J) = 0, with the group of all rational numbers of the form m/p* isomorphic l where m and n are integers, and that /^(SpjS ) = S p where S is the
l
H
multiplicative group of
complex numbers of absolute value
1
(or
equivalently, the addive group of real numbers reduced modulo integers). 2. Let p and q be two integers. Consider the exact sequence
T
1\
S:
> J *
J  J/$J
>
EXERCISES
where r(n)
7:
297
=
qn and
17
is
S
S
defined
by 7(2)
=
the natural map. Consider the mapping px. Analyze the inverse limit S* of the
inverse sequence
S < S
For what values of p and
vin, 5. 5).
77 <*7
<?
S *(see
is
S m an
exact sequence?
Example
3. Let (E,S) be a 2cell and its boundary defined in the complex plane by the conditions \z\ ^ 1 and \z\ = 1 respectively. Consider the 2 " /q Let (P,C) be the pair obmap <t>: S > S defined by <t>(z) = ze tained from (,S) by identifying each z e 5 with 0(z). Show that the
.
homology sequence
isomorphic with the exact sequence S of the preceding exercise. Show that the map /: (E,S) > (E,S), /(*) = z" induces a map /: (P,C) * (P,C) such that /*(w) = pu. Let (P^C m ) be the limit of the inverse sequence
(integer coefficients) of the pair (P,C)
is
(P,O
.
Show
(see
coefficients) is
that the homology sequence of (P^Cco) (Cech group with integer isomorphic with the sequence S m of the preceding exercise
x,4.1).
Example
Show
that
C w and S p
and
q
are homeomorphic.
G. MISCELLANEOUS. 1. Let (X,A) be a compact
natural
pair,
>
let
u be
in the kernel of the
homomorphism (X) (A) (Cech cohomology groups). Show that there exists a closed neighborhood V of A such that u is in 9 9 the kernel of (X) > (V).
H
q
H
H
H
> Let a be a covering of X. Show that a map </>: \X a canonical relative to a if and only if, for each v e V a we have <f>~\st(v))
2.
X
is
\
,
C
CHAPTER
Applications
1.
XI
to
Euclidean Spaces
INTRODUCTION
The objective of this chapter is twofold. First we derive a number of theorems concerning Euclidean space among which are some of the most classical and widely used ones such as the Brouwer fixedpoint
theorem and the in variance of domain. Secondly we show how such theorems can be derived using the axioms without appeal to any concretely defined
The
first five articles
section (6)
= notations of i,16 are used throughout the chapter: R En = the ncell z g 1, ^S""" 1 = the (n cartesian nspace, l)sphere
The
n
o:
homology or cohomology theory. depend only on Chapters i and n. The uses, in addition, the continuity axiom of Chapter x.
last
=
1,
etc.
still
valid if arc replaced by homeomorphs. For ,E ,S the Brouwer fixed point theorem 3.3 holds for any set example,
These spaces appear ~
in
most of the theorems.
The
theorems are
Rn
n
n
l
X
homeomorphic with
En
.
For
morphism
there
is
X
let /'
be a
l
>
FT.
a fixed point,
Then / = hfh~ maps E n into itself. By 3.3, say x, of /. Then h'\x) is a fixed point of /'.
map
X
X, and h a homeo
The
proofs of invariance of the other theorems of this chapter are
equally trivial.
2.
MAPPINGS INTO SPHERES
is
LEMMA
closed in
2.1.
// (X,A)
f:
a normal pair
,
(i.e.
X
.
is
normal, and
A
is
X) and
A
*
Sn
then there is an open
n
set
U
of
X containing
of /.
S of f. f: U n+1 PROOF. Regard / as a map of A into R extension theorem, there exists an extension g: X U X g~*(G) where is the origin of R n and
A and an
extension
Then, by Tietze's
*
R n+l
Define
* 1
set
''<*>
ifn
X
""
X
This yields the desired extension of /. LEMMA 2.2. Let (X,A) be a normal pair such that X I is normal. A X / > S n admits an extension F: X X / Every map f:
X
W
2]
MAPPINGS INTO SPHERES
2M
PROOF. Let B = A X /. Since the pair (X X /,B) is X normal, there is, by 2.1, a neighborhood U of B and an extension <: n > S f/ of /. Since /I X / U and / is compact, we may construct a neighborhood V of A in such that V X / U. Select a Urysohn
X
U
C
X
C
function
for
0:
z
each x
X X
7 such that
is
on
X
V
e
and
?/.
is
1
on A.
s
Then
and
t
s
/,
we have
4>(.r,0)
<f>(x,t)
(x,0(x)l)
Define /*"(XO
=
tf>0,0(*)0.
6(x)
Then FfoO) so that F(x,t)
2.3.
=
/(x,0),
and
for
x
=
A, we have
I
=

/(x,Q.
Thus F
is
the desired exI is normal, a homotopy
tension.
LEMMA
and
off.
let
f:
X
>
Sn
Let (X,A) be a normal pair such thai Every homotopy of f\A can be extended
.
XX
to
PHOOF.
map
A':
X
Let h: X \J
A X A X
/
/
>
S be a homotopy
n
by 2.2, h' admits an homotopy. DEFINITION 2.4.
homotopic
to a
by setting h(xfl) extension //: /
=
>
of f\A.
f(x) for x
*ST,
X X
Extend h to a e X. Then,
is
which
the desired
A map
of
/:
X
?'.s
n
is
called inessential
<ST.
map
///:
X
n
into a single point of
if / Otherwise /
is
is
called essential.
LEMMA
2.5.
X >
e >S
?/,
S"
essential then
f(X)
?/
= S\
contractible to a
PROOF. Suppose y n point and f(X) C #
/(X).
it
Since
A?"
is
follows that /
H
>
is
inessential.
LEMMA
2.6.
Let (X,A) be a normal pair such that
X X
/ is normal.
V A
Every inessential
map
/i:
/T
A
n
>
*S
admits an inessential cxlcnsion f:
into a single point of S existence of /' follows now from 2.3.
of
^
Oi
^S
.
PROOF.
Let
2.7.
X
T
iS
be a
map
X
n
.
Then h\A
is
homotopic with
Let
/.
The
X such that dim n Then every map f: X > *S is inessential. n PROOF. Select a triangulation T' = {t',K'\ of \vith dim K = n. the theorem on simplicial approximation there exists a map By X > *S n homotopic with / and which is simplicial relative to the g: k k = k t, k K\ is a sufficiently high barytriangulations T,T' where T
LEMMA
n.
{t,K\ be a triangulation of
K <
f
{
ccntric subdivision of T.
Since the dimension of a simplicial complex
k
does not change under bary centric subdivision, we have dim n onto S Since g is simplicial, it follows that g does not map
K <
n.
X
>
.
Thus,
by
2.5, g is inessential, and therefore / is inessential. m COROLLA HY 2.8. // n, every f:
m <
map
.
S
Sn
is inessential
> fl and admi's an extension f: E is inessential follows from PROOF. The fact that / of an extension/' follows from 2.6.
m+v
n
2.7.
The existence
LEMMA
2.9.
Let
T =
{t,K\ be a triangulation of
X with dim K
g
n,
SOO
let
APPLICATIONS TO EUCLIDEAN SPACES
[CHAP. XI
a closed subset of X, and > S n extension f:
be
A
let
f:
A
Sn
.
Then f admits an
X
.
LEMMA
2.10.
1, let
Let
K
g n
a
+
A
F
be
exists
finite set
CX
with dim T = [t,K] be a triangulation of Sn There a closed subset of X, and let f: A A such that f admits an extension /':
.
X
X
F* S\
PROOF OP
2.9
AND
2.10.
By
/':
2.1,
there
>
is
an open
If
k
{
set
U
in
X
con
taining A, and an extension
U
S
n
k
of /.
we
k
replace
T by
a
= sufficiently high barycentric subdivision T k A' of to T) such that (relative subcomplex
t,
K}, we can
A.'
find a
X
A
C
C
C7.
Since
dim
1
= dim
*K,
it
follows that
a subcomplex of relative to T. Q Let relative to T. Clearly denote the gdimensional skeleton of A > S" admits an extension f: X VJ A Sn Suppose in/: n Q Q A > S is already given for ductively that an extension / some q < n. Let s r be the (g + l)dimensional simplexes of 9 A. Then / a is defined on s s r and, by 2.8, f may be exG+1 s r thus yielding / tended over every one of the simplexes s ly n c+1 A > $ extending /. Therefore there exists an extension n n n = X, which proves VJ A > S of /. If dim g n, then f: 2.9. If dim s r be the (ft + l)simplexes of ^ n 1, let s ly A, and let F ={&!,, 6 r be the set of their barycenters. A radial projection in each simplex s s r yields a retraction h: Q  F > 5 n is then an extension of F > n A. The map / A:
is
A
X
we may assume
in 2.9
and 2.10 that
X
X
.
:
X U
,
X
X
t ,
,
,
W
,
,
:
X
K

X
K
+

,
X
/:
}


A  Sn A
X U
{ ,
,
X
X
.
Thus 2.10
Let
1
,
is
proved.
LEMMA
every f:
f':
2.11.
A
be a closed subset of
taining exactly one point out of each
S and let B be a set conn A. Then for component of S
y
n
S"'
there exists
a
Sn
 F^Sn ~
finite subset
F
of
B
and an extension
l
off.
,
PROOF. By 2.10, there exists a finite subset (x ly x k ) of S" A n n~ and an extension S xk ) > S of /. For each x let b { (x l9 denote the point of B in the component of S A containing x and let F be the set (b ly b k ). To prove that / admits an extension n F > S"' we shall prove inductively that, if / admits an exS tension S" x k ) * S"" then / also admits an (61, _i, s n 6 extension S x t+1 Since z, and 6 (6^ z*) ~> S""
l
,
t
,
t ,
,
1
,
1
,
fc t
% ,
,
,
1
,
t ,
,
,
.

t
lie
in the
same component
x*
of
S
n
A, there
,
is
a
finite
sequence
=
2/o,
Vi
=
&
of points
and a sequence of convex
ncells
8]
THEOREMS OF BROUWER AND BORSUK
A
such that
t/,i,2/,
SOI
in 5"
/S,
s
E, for j
=
1,
,
Z,
and the boundary
of E, contains
none of the
6's, z's,
and
y's.
It clearly suffices to
/<.,.!!
show
,
that,
,
if
/ admits an extension
,
S
n

(6 t
6<i,,i,x, +1
,
a:*)
> S"" 1
,
then / also admits an extension
Let
r:
S"
y,
>
Sn
,
be a retraction.
e fiT
Then
,
setting
.
/<.i(z)
=
fi ti ir(x) for
x

(b lf
6 i . l ,y .,x i +1 ,

,
z
ft
)
yields the desired extension /,,,.
This concludes the proof.
3.
THEOREMS OF BROUWER AND BORSUK
The nsphere S
n
THEOREM
point,
i.e.
3.1.
is not contractible (over itself) to
a
the identity
PROOF.
Then, by
is
Let
H be any homology (or cohomology) theory defined on
map
of
S
n
is essential.
the category 5 of triangulable pairs and with a coefficient group G j& 0. n Thus, by i,11.5, S" 1,16.6, S is not homologically trivial.
not contractible to a point. REMARK. The above statement
is
the only one of this section that
requires the use of homology theory in the proof. All the other propositions follow from 3.1 using the results of 2 and other simple geometric
arguments.
THEOREM
PROOF.
3.2.
r:
Sn
~l
is not
Let
E"
 Sn ~
a retract of
1
JET.
be a retraction.
Then
x
e
F(z,0
yields a
=
r((l

f)x)
S"
1
,
t e
7
homotopy contracting S
3.3.
THEOREM
En
>
En
to a point, contrary to 3.1. Each map f. (Brouwer's Fixed Point Theorem). n such that f(x) = x: has at least one fixed point, i.e. a point x e
n~ l
E
Let r(x) be the point of PROOF. Assume f(x) j x for all x s E 1 The conS""" such that x lies on the line segment from f(x) to r(x). tinuity of r is proved using that of / and some elementary geometry. ~ n Sn contradicting 3.2. Since r(x) = x when a; e S", r is a retraction E THEOREM 3.4. Let T = [t K\ be a triangulation of a space X. In order that dim K ^ n, it is necessary and sufficient that for every closed A > S n there exists an extension subset A of X, and every map f:
n
.
l
y
,
f:
X>S
PROOF.
n
.
The
that dim
K>
necessity of the condition follows from 2.9. of dimension n n, and let s be a simplex of
X
+
Suppose 1. Let
302
APPLICATIONS TO EUCLIDEAN SPACES
s
[CHAP. XI
A =
and
is
let /:
/':
A
an extension
fore s
X
Sn be a homeomorphism. Suppose that / admits is a retraction. ThereSn Then /~V: X
.
^4.
a retract of the closed simplex
3.5.
s
contrary to
If
3.2.
THEOREM
(Invariance of Dimension).
T = [t^K^ and
l
T2 =
dim
1/2^2} are triangulations of the same space X, then dim KI
2.
=
K
This
is
an immediate consequence of
3.6.
3.4.
THEOREM
(Borsuk's Separation Criterion).
Let
X be a compact
subset of cartesian space
Rn
and
n
let
x
e
Rn
it
in the unbounded component of R n~ > S the mat) p: given by
X,
X
Ti
is
In order that x n lie necessary and sufficient that
X.
X
l
mx)
be inessential.
/
\
=
Xo
rr
x
e
A
CT
PROOF. By a translation and a similarity transformation of R", it n can be arranged that X lies in the interior of E and x is the origin. The map p is then defined by p(x) x/ \\x\\. n X. Since x lies in the unbounded component C of R Suppose C is arcwise connected, there exists a mapping /: 7 > C such that ~ = x n ,/(l) = Xj e/J"  #n Consider the mapping F: X /> Sn /(O)
,
() .
X
l
defined
by
Then
and
F(x,Q)
=
p(x)
and F(x,l)
is
=
(x

x,)/x
5^

x,.
Since x
It
e
#n
X! is
not in
#
n
,
it is
easy to see that F(x,l)
inessential.
Xi/UxJI.
therefore
follows from 2.5 that p
X containing x is Assume nowjthat the component C of 7 bounded. Then C C #" and C \J X is closed. Suppose p is inessential. n C VJ X /S "\ Define a Then, by 2.6, p admits tin extension p " n E > S by setting r: map
n
1
:
7
1
r(x)
=
p'(x)
for
f r
x
x
e
C
^m
x
VJ X,
""
r ( x)
=
x TTTT
1
e
C
f

1^11
The two we have
definitions agree on x, so that r r(x)
X, hence
is
r is continuous.
When
n1
.
e
Sn
~l
,
=
x
a retraction of
Let
E
n
onto S
This con
tradicts 3.2.
THEOREM n The set S
inessential.
3.7 (Borsuk's Theorem).
X
be
a closed subset of
Sn
1
.
X
is
connected if and only if every
n is
map
n
f:
X
If
AS""
is
PROOF.
points of
AS"
Suppose S
X not connected and X lying in distinct components of S
let Xo,x t
X.
be two we regard
3]
THEOREMS OF BROVWER AND BORSVK
Xi as
n
,
303
the cartesian space R it follows that X Q lies in a bounded n X. Thus, by 3.6, there exists an essential map component of R
p:
Sn
X>S
Sn
n
~
l
.
z
is
X is connected and let /: X > S Suppose now that S n n e S X. By 2.11, / admits an extension /': S  x
n
n~
l
.
Let
Since
x
is
contractible,
it
follows that /'
is
inessential.
S n ~\ Thus /
>
also inessential.
~l
LEMMA 3.8. Let (X,A) be a pair in S n homemomorphic with (En ,Sn ). n A decomposes into two components, namely Sn X and X A. Then S In particular, X A is an open subset of 8 n
.
PROOF. Since E does not admit essential maps into S*~ it follows n from 3.7 that 8 X is connected. On the other hand ^S"" does admit n~ n .4 is not an essential map into S (e.g. the identity map), thus S ~ n A is homeomorphic with E Sn connected. The set X and is
}
n
,
1
l
l
,
therefore connected.
Since
S
n
n
 A =
n
(8

X) \J (X
n

A),
it
A are connected, A is not connected, and S and S X, n A. follows that the latter two sets are the components of *S
THEOREM
subsets of
n
X
*S
,
and
3.9 (Invariancc of Domain}. If Ui,U 2 arc is open, then U2 is also open.
homeomorphic
U
}
PROOF.
X]
Let /:
U\
>
U
2
be a homeomorphism.
=
I
= V  T^then (V V,) and (E ,S = (fV.JV^. Then (X,A) and pairs. Let (X,A) A is open. Since x morphic, and, by 3.8, X
n
/'
(x 2 ).
Select a spherical_ neighborhood V, of ~
n
l
l
A
Let x.2 e C/^and such that V
l
C
17,.
Let Vi
lt
homeomorphic n (E ,S* ) are homeo)
1
are
2
e
X
A C
^2,
it
follows that
U
2 is
open.
A space dimension n if every point x e n morphic with K
DEFINITION
3.10.
.
M M
will
be called locally euclidean of possesses a neighborhood homeo
THEOREM
A/!
n.
and
If
M
i
2,
Let U^ and U 2 be homeomorphic subsets of spaces respectively, both of which are locally euclidean of dimension
3.11.
U
is
open, then
l
U
2
is
also open.
,
let
U 2 be a homeomorphism. Let x2 e f/ 2 and PROOF. Let /: Ui x f~ (x 2 ). Select neighborhoods V ,V 2 of Xi,x 2 respectively such
}
l
that
x
x 2 sV 2 C M,, c V, C U lt n Vi,V 2 are homeomorphic w ith R
t
r
/(F.)
.
CV
,
t,
homeomorphic with an open subset of S we may select a n n map 0i*. Vi > S 2 V 2 > S mapping Vi and V 2 homeomorphically n Then 2 f<i>i maps homeoonto open subsets Wi and 2 of S the set <^>2/</>r (W i) is onto a subset of 2 Thus, by 3.9, morphically
Since
is
,
Rn
n
</>
:
W W
l
.
<f>
W
l
1
/r
.
SOU
APPLICATIONS TO EUCLIDEAN SPACES
in
is
[CHAP. XI
l
open
/(y,)
S n and
open
in
2
therefore also in
V2
is
Therefore the set f<t>~ (Wi) U2 Since x 2 e f(V^ and therefore also in 2
2.
W
=
,
M
.
C
it
follows that
U
open.
COROLLARY
3.12.
Two
locally euclidean spaces
MI and M
2
of different
let m = m +
l
dimensions are not homeomorphic. PROOF. Let m,,m 2 be the dimensions of
2
n,
n
>
0.
Then M
M
l
and
M
2
respectively
and
2
X Rn
is
locally euclidean of
sion mi,
by
3.11,
and contains a nonopen subset homeomorphic with M! and 2 cannot be homeomorphic.
M
M
dimenThus,
2.
4.
DEGREES OF MAPS
In this section and the following one, we assume that a homology is given with a coefficient group G isomorphic to the group of integers J. Analogous results also hold for a cohomology theory
theory
H
with a coefficient group G ~ J. DEFINITION 4.1. Consider a
H
n (S")
~ J
map
n
/:
/S"
>
,
n
>
0.
Since
for
all
u
e
H
(see i,16.6) there is
n
an integer d such that f^u
= du
n (S ).
This integer d
is
called the degree of /, written d
=
degree (/). n n > S have the same Clearly two homotopic mappings of S degree. The converse is also true but will not be proved until Chapter xv. It is easy to see that the degree of the composition of two maps is the
product of their degrees.
LEMMA
therefore
4.2.
n
A mapping
onto S".
f:
Sn
*
Sn
of degree 5^
is essential,
and
maps S
PROOF.
H
homotopic
n (xo)
=
0, it
LEMMA
into itself:
follows that f = 0. Thus the degree of / is zero. n Let n 4.3. 1, and let f be a map of the triad (S ;E2,E")
map
Suppose / is inessential. Then we may replace / by a n Since /' such that /'(") is a single point X Q e S
.
>
Let /i:
>Sr

S",
/2
:
Sn
~l
AS"'
1
be the
maps
defined by /.
Then
degree(/ x )
=
degree(/ 2 ).
n
PROOF.
diagram
By
1,16.2,
the triad (S ;E?,E)
is
proper.
Consider the
A
H
n
n (S )
I/**
I'
.
4]
DEGREES OF MAPS
is
SOS
where A
triad.
1,10.5,
By
A
is
the boundary operator in the MayerVietoris sequence of the Further, by i,15.4, commutativity holds in this diagram.
an isomorphism.
degree(/ 2 )Au
Let u
e
// n (S
H
),
u
7^ 0.
Then
Since
Au 9* 0, it LEMMA 4.4.
= Af^u = degree(f J Au. = degree(/i). follows that degree (/ n n Sn Suppose that a function T: S X S
=
/ 2 *Aw
2)
,
n
>
0,
is
given such that there exists
an element
e s
Sn
(unit element) satisfying
(i)
T(e,x)
e
,
=
x
=
T(x,e)
for all x
Sn
S
n

(see note at
S
n
define f:
n n S > S
end of this chapter). by setting
Given two maps f\J*
degree(/)
=
degree(/0
+
degree(/ 2 ).
g\ f g 2
PROOF.
(ii)
We
first
replace /i,/ 2
gi (E*)
by homotopic maps
g*(El)
such that
=
e,
=
e.
To show
to a point, there = ^ for x F(o:,l)
maps exist, consider fi\E2. Since E2 is contractible n a homotopy F: EH X /  S of /i# such that e #". By 2.3, this homotopy may be extended to a F: Sn X / > >S n of /,. Set gr^x) = F(x,l). The existence homotopy of g 2 is proved similarly. The map / is now homotopic with the map Sn > S n defined by g:
that such
is
so that
it suffices
to prove that
9*
in the
=
0i*
+
02*
dimension
n.
From
(i)(iii) it
follows that
for
*
e
El,
for x
If
e
El.
we denote by
g',gi,gi
the
maps (S
n
n~ l
,S
)
>
(fiT,6)
defined
respectively, then the conditions of i,14.6 are satisfied;
by 0,0i,0 2 and therefore
0*
=
0f*
+
02*.
306
APPLICATIONS TO EUCLIDEAN SPACES
i:
[CHAP. XI
n
n
Consider the inclusion maps
Since jg
Sn
=
C
n
n~
(S ,S
1
0'z
and similarly with subscripts
.7*0*
and j: S C (S and 2, it follows that
l
)
,e).
=
,7*0i*
+
J*02*.
Since n
>
0,
we have
4.5.
zero in the dimension n.
THEOREM
\z\
=
= 0. Hence, by exactness, j^ has kernel in the dimension n. This implies g# = Oi* 2* l S as the set of all complex numbers z with Regard
n (e)
H
+
1.
For any
k.
integer
fc,
the
map f k
:
S
1
l
>
S
1
defined by
f k (z)
rry.
=
k
z
has degree
l PROOF. Define T: S X S * S by setting rfoy) any two integers fc,Z, we have
l
=
For
Thus
4.4 yields
degree(/jn.i)
=
it
degree(/*)
+
1.
degree(/i).
Since /i
is
the identity map,
has degree
This implies that f k has
degree
k.
REMARK.
k
f +u = ku for each u e Hi(S J\u = kuforueH (S ).
l l
The proof above
actually yields a stronger result, namely:
l
)
in
any homology theory.
Similarly
THEOREM 4.6. For any n > and any integer k there exist maps S n  Sn of degree k. f: PROOF. The proposition has been established for n = 1. Assume n~ S S"' (n > 1) is a map of degree k. Extend inductively that g: n n to a map /: S S such that f(E$ C E? and /(_) C (such g
l
1
extensions obviously exist).
same
degree.
Then Thus / has degree k.
4.3 implies that /
and g have the
6.
THE FUNDAMENTAL THEOREM OF ALGEBRA
n
Consider the compactification R of the cartesian space R obtained n n > R by adjoining a point w at infinity (see x,6.2). A mapping /: R is admissible (in the category CL LC of locally compact pairs) if its exn n tension /: R + R defined by /(w) = w is continuous.
n
DEFINITION
5.1.
The
degree of
defined as the degree of the map latter degree is n n defined since R is homeomorphic with the nsphere S n n LEMMA 5.2. // /: R n + R n has degree 0, then f maps R onto R
. .
an admissible map R n + R n The /:
/:
Rn
*
Rn
is
^
.
PROOF.
4.2 implies that /
LEMMA 5.3. \\x\\ ^ 1. ///:
Let
Dn = R n  En
n
>
maps R onto R\ Thus f(R
n
.
n
n
)
= Rn
.
R
n
/2
,
n
>
1, is
Thus D is defined by the condition n an admissible map such that f(E ) C
5]
FUNDAMENTAL THEOREM OF ALGEBRA
n
>
,
307
E n f(D n ) C n~
,
g:
S
l
 Sn ~
t
then the degree of f coincides with the degree of the
n
map
n
l
PROOF. from 4.3 since the triads (R ;E ,D ) n and (S ;E? E") are homeomorphic. LEMMA 5.4. Let /<,,/!.* R n > R n and let F: R n X /  R n be a homotopy between f and fi such that, for every real number A, there is a real number B such that
defined by f. The result follows
n
\\F(x,0\\>A
if
\\x\\
>B,
<e7.
Then the maps / and fi are both admissible and have the same degree. PROOF. The condition of the lemma implies that setting F(w,J) = w n n > R of F. This implies that / and yields an extension F: R X / /! are admissible and that F is a homotopy between / and fi. Therefore / and /! have the same degree.
THEOREM
5.5.
A
polynomial of degree k
k
>
a,z
f:
f(z)
=
k.
ak z
+
is
a^z*'
1
+


+
+
a
2
,
ak j
>
with complex coefficients
topological degree
an admissible map
R
R
2
which has
COROLLARY
f(z)
5.6 (Fundamental
least
Theorem of Algebra).
The equation
=
has at
one solution.
PROOF.
Let
g(z)
=
a k ^z
2
k~
l
+
Let
a
.
Define F:
R X = F(z,t)
/
 R 2 by
k
+
+

a,2
+
a and a
=
a t
,
+
setting
t)g(z),
akz
+
(1
zeR
2
,teI.
A
be any real number, and
z
let
>
^
(A
+
a),
\z\
>
1.
Since k
>
0,
we have
\aj\
\F(z,l)\
^

\g(z)\
\aj\

o*'
> Klk 5.4.
/':
a
>
A.
Thus the homotopy F
is
satisfies
the condition of
admissible and has the same degree as the
/'() = F(,l) = ^. 2 Let 6: I > R  (0) be such that 9(0) = a*, 9(1) = 1. The ho2 2 defined by G(z,0 = 9(02* again satisfies the motopy G: R X / > Therefore /' has the same degree as the map condition of 5.4.
map
Consequently / R 2 > R 2 where
/"()
=
G(z,l)
=
z.
Since
;/
/
(z)
=
k
\z\
,
the conditions of 5.3 are
l
l
satisfied
* S defined and /" has the same degree as the map /*: S by /". Since, by 4.5, f k has degree fc, the proof is complete. We now proceed to establish an analog of the fundamental theorem of algebra in a setting more general than that of complex numbers.
SOS
APPLICATIONS TO EUCLIDEAN SPACES
[CHAP. XI
given.
We
shall
assume that a mapping
shall denote T(x,y)
T:
Rn X Rn
>
Rn
is
For
is
simplicity subject to the following conditions:
(i)
(ii)
we
by the product
0.
e
xy.
This product
x 7* 0, y 5* imply xy 9* There exists an element e
Rn
such that ex
=
x
=
xe for
all
(iii)
number
The relation (tx)y = J(:n/) = x(ty) holds of each The last condition implies by continuity: Ox
t.
real positive xO.
=
numbers (n = 1), comTypical examples numbers (n = 2), quaternions (n = 4), Cayley numbers (n = 8). plex It is not known for what values of n a product such as described above
division algebra.
are:
These properties are
satisfied
if
Rn
carries the structure of
real
a real
exists (see note at
end of
this chapter).
We define inductively a monomial m (in one variable) of degree k as A constant in R n different from zero is a monomial of degree follows:
zero.
w
2
The identity function are monomials of degree
fcj
a;
fcj
is
a monomial and k 2 then
,
of degree
1.
If
Wi and
m m^
v
is
a monomial of
degree
+
>
fc 2 .
It follows that,
if
m
is
a monomial and x
n e fl
,
then m(x)
e
R* and
m:
R
n
R
n
is
continuous.
5.7. Assume n > 1. Le f be a polynomial of algebraic k > is a monomial of degree k and g g where degree of the form n n > R a finite sum of monomials of degree < k. Then the map f: R is is admissible and has topological degree k.
THEOREM
m+
m
has at least one solution. The equation f(x) = e = e = 0, we find and xa Applying this to the equations ax COROLLARY 5.9. Each a j* has at least one left inverse and at least
COROLLARY
5.8.
one right inverse.
Applying 5.8 to the polynomial xx + e yields COROLLARY 5.10. There exists an element whose square is The proof of 5.7 will be preceded by some lemmas. LEMMA 5.11. There exist real numbers < A < B such
A\\x\\
\\y\\
e.
that
^
\\xy\\
g
n
B\\x\\
\\y\\.
1 > R be defined by y(x,y) = xy. Let y: S"' X S 1 Then y(S X S"" ) is a compact subset of R n not containing zero. Thus there exist real numbers < A < B such that A ^ \\xy\\ ^ B 1 Then (iii) yields the conclusion of the lemma. for x y e S"" LEMMA 5.12. // m is a monomial of degree k, then there exist real
PROOF.
n' 1
n" 1
.
t
numbers
< A m < Bm
such that
ro(*)
A m \\x\\ k S
^ B m \\x\\\
5]
FUNDAMENTAL THEOREM OF ALGEBRA
309
PROOF. The conclusion is obviously valid for a monomial of degree 0, and for the monomial x. Suppose now that m = m^ ra 2 is a product of monomials of degrees k lt k 2 respectively with k = ki + fc 2 and that the conclusion of the lemma is valid for mi and ra 2 with the constants A mt ,B mi ,A mty B mt respectively. Then, by 5.11,
,
where
B m = BB mi B mt
.
.
Similarly
A m \\x\\ k
\\m(x)\\
where
>
Am =
AA mi A mt
PROOF OF
5.7.
Consider the homotopy F:
F(x,t)
Rn X
/
I defined by
=
m(x)
+
(1

t)g(x),
x
Rn
,
t
e J.
Let g
=m
l
+
<
fc,
+ m\ be a decomposition of
and
let
into a
sum of monomials
of degrees
If
A'
is
a
real
number, and
then
fc
>
implies
\\F(x,t)\\
*
lm(a)
> A n \\x\\
Thus the homotopy F
 9 (x) ^ A w  5' > A'.
fc

JL,
B mt \\x\\
h~l
satisfies
the condition of
5.4.
admissible and has the same topological degree as the F(x,l). This reduces the proof to the case when / =
of degree k.
Consequently / is map m(x) = m is a monomial
1
S
n~ l
With each monomial m of degree k we associate a function
defined
"" 1
g:
/S*
+
by
ctnl
It follows
from 5.11 that m(x)
continuous function.
topological degree.
logical degree 0.
If
We
fc
shall
0,
0.
=
Let
fc
>
0, thus g is a welldefined and g have the same that prove both and g are constant and have topon / R* by Define a homotopy F; R
^
for
x
^
m
m
X
setting
F(x,t)
=
F(0,<)
=
ji^lij
0.
[(1

0IM*)II
+
<x*] f
xefl"

(0),
<e
J,
S10
APPLICATIONS TO EUCLIDEAN SPACES
5.12
[(1
[CHAP. XI
By

t)A m
+
t]
\\x\\"
g
fc
\\F(x,t)\\
((I

t)B M
+
t]
IMP,
and therefore
AF
where
that
z
^
\\F(x,t)\\
g BF
k
\\x\\
A F = min
is
(l,A m ),
it
F
continuous at x
B F = max (l,# m ). This inequality = 0, and that F satisfies the condition
map
shows
of 5.4.
Since F(xfl)
=
m(x),
follows from 5.4 that the
M(x)
=
=
F(x,l)
=
m(z)
TT,
\\m\x)\\
x
*
0,
M(0)
is
M
it
admissible and has the same topological degree as m. The function satisfies the condition of 5.3 and therefore has the same degree as
the
map
S"'
1
> S n
~l
defined
g
follows that
m and
by M. Since M(x) have the same degree.
=
g(x) for
x
e
S ~\
n
Consider the mapping
y:
S
n~ l
X Sn
~l
> S" 1
defined
by
Setting
e'
=
c/c
we have
y(e',x)
=
S
x
=
l
7(x,cO,
2
^
s
S""
1
.
Now
ffiffitfo
consider the product be the associated maps of
m = mim n~
into
1
of
two monomials and
let
itself.
Then
llm^x)^ \\m 2 (x)\\
___m
Since
(x)m 2 (x)
n
1
>
0,
4.4 gives
degree(^)
=
degree(^i)
+
degree(0 2 ).
is
This implies that the topological degree of m
.
the
sum of the
topological
degrees of mi and ra 2 Thus the topological degree of a monomial obeys the same additivity rule as the algebraic degree. Since for a monomial
6]
MANIFOLDS
311
of algebraic degree zero (i.e. a constant 5^ 0) the topological degree is also zero, and the monomial x has both algebraic and topological degree
1,
it
follows that the algebraic
and topological degrees coincide
for all
monomials.
This concludes the proof.
6.
MANIFOLDS
a compact space,
DEFINITION
6.1.
If
X
is
A
is
closed in
X
(X,A)
its
X, and
A
is
is
locally euclidean of dimension n (see 3.10), then the pair called a relative nmanifold. If there exists a relative homeo/:
morphism
> (E,S) (X,A) (see x,5.1), where (E,S) is an ncell and then (X,A) is called a relative ncell. boundary, The objective of this article is to show that the homology and cohomology groups of a relative nmanifold (X,A) are zero for dimensions greater than n, and to establish results concerning the structure of the
ndimensional groups. In case (X,A) is triangulable, the Uniqueness theorem of Chapter in shows that the results are independent of the choice of the homology
theory. It has been proved that any compact, differentiate, absolute manifold can be triangulated [S. Cairns, Bull. Amer. Math. Soc. 41
(1935), 549552].
It is
is
an unsolved problem of long standing to de
termine
valid without the assumption of differentiability. It is easy to construct relative nmanifolds which are not triangulable. For example, let be a closed subset of be an nsphere, and let
if
the result
X
A
X
is
which
is
not triangulable, then (X,A)
is
a relative nmanifold and
not triangulable. A less trivial example is obtained as follows: Let Y be a locallyeuclidean space which is not separable. Let be the
X
compact space obtained by adjoining a point at
infinity to Y,
and
let
A
consist of this point.
Since
Y
is
not separable,
X
is
not triangulable.
will
We shall use the Cech homologycohomology theory. Its continuity play a vital role in the computations. We repeat that the results
are independent of this choice if (X,A) is triangulable. If the singular theory were used, there would be no correspondingly simple results valid in the nontriangulable case.
Since the results
we
will
obtain are more significant in the case of
cohomology, we
sults for
restrict ourselves to that case.
The corresponding
re
homology are stated at the end of
this article.
LEMMA
n,
and
let
6.2. Let Y be a space having a triangulation 9 (X,A) (X,A) be a compact pair in Y. Then
T
of dimension
H
is zero
for each
q
>
t
n.
X
PROOF. Let and Ai be the
T denote the
least
barycentric subdivision of T. Let subcomplexes of F, in the triangulation T, coni
tb
312
taining
APPLICATIONS TO EUCLIDEAN SPACES
[CHAP. XI
X
and
A
respectively.
and (Xt+i,Am)
C
tends to infinity, it follows that (X,A) = (C\ X<, A<). By applying we obtain that Q (X A) is zero, and the proof is complete. x,2.6, Since a relative homeomorphism induces isomorphisms of the Cech
(X
t
,A>).
Clearly Since the
(X i1 A^) is zero for q > n, mesh of T tends to zero as i
q
%
H
H
H
t
groups
(x,5.4), it follows that the
i.e. all
Cech groups
are those of an ncell,
groups are zero except
of a relative ncell (X,A) n G. (X,A)
H
~
// (X,A) and (X,E) are relative ncells such that (X,A) then the inclusion map induces an isomorphism (X,B),
LEMMA 6.3.
C
H
PROOF.
ncell
n
(X,A)
H
n
(X,B).
is the euclidean Consider first the special case where E: ]? x? g 1, and A is the sphere S: 2? x* = 1. Also, let x be a point of E S, let U be the interior of a sphere with center E  S, and let B = E  U. Then (E,E) is a relative XQ such that U
X
C
ncell containing (E,S).
r:
B
Radial projection from XQ defines a retraction
S.
If
we
set
r(*,0
=
(1

t)x
+
tr(x),
we obtain a
strong deformation retraction of
B
into 5.
Then
i,10.5c
applies to give the desired result in this case. For the next case, let (X,A) be a relative ncell, let /: (E,S) be the interior of a sphere (X,A) be a relative homeomorphism, let
contained in
ncell,
E j:
S,
and
let
B =
X
U
/([/).
Then (X,5)
is
a relative
and /
Let
*/>
defines a relative
(fi,S)
(X,fi).
C
=
fij
=
w^ have
j*/*
[/) homeomorphism /ji (,#  U) and h: (X,A) C (X,B). Since (E,E /*A*. As /j,/ are relative homeomorphisms,
f\ and /* are isomorphisms. By the preceding case, j* is an isomorphism. It follows that h* is an isomorphism. > Consider now the general case. Let /: (E,S) (X,A) and g: l B) is open (EjS) (X,B) be relative homeomorphisms. Now f~ (X = in E S\ hence it contains a sphere with interior U. Let A' 1 S; hence it contains a sphere A') is open in E /([/). Then ^ (X with interior V. Let B' = Inclusion relations induce /(F).
X
X
homomorphisms
y*
fc*
Z*
H (X,B')
n
> ffn (X,A')
By
Hence k*j* and
implies that
I* is
construction, the preceding case applies to the inclusions jk and kl. f*fc* are isomorphisms. It is easily seen that this
an isomorphism.
This completes the proof.
6]
MANIFOLDS
6.4.
313
DEFINITION
pair of relative ncells (X,B),(X,B
Let (X,A) be a relative nmanifold. ) such that
;
An
ordered
(X,A)
will
C
(X,B)
C
(X,B'),
or
(X,A)
C
(X,B')
C
(X,B)
 A. By 6.3, the inclusion map induces an be called a s/ep in n = n (X,B') called the isomorphism of the step. isomorphism (X,B) A finite sequence P of relative ncells
X
H
H
such that each pair of adjacent terms is a step in X  A, will be called a path in A from (Z,B,) to (X,B ). The composition of the iso
X
ft
morphisms of the successive steps is an isomorphism /P(X,Bi) = H n (X,B k ) called the isomorphism P* o/ Me paffc. In case (X,Bj) = (X B k ) the path is said to be dosed, and the isomorphism of the path becomes an automorphism. The manifold (X,A) is said to be orientable if the automorphism of each closed path in X  A is the identity;
9
otherwise (X,A) is called nonorientable. pends on the coefficient group).
(Note that
this concept de
LEMMA
6.5.
Let
X C
/
X, and X'

A'
(X,A),(X',A') be relative nmanifolds such that  A. If (X,A) is orientable, so also is
C X
,
(X',A').
PROOF.
Let (X',BO,
(X',BO be a closed path
X'),
;
in
X'
1,
A'.
Set
B<
Since X'
in
is

=BIU(X r/
t
i
=
,
ft.
B^ = X  J5 is open in A  .4 by 3.11, it is also open A; hence B is closed. It follows that (X',BJ) C (X,BJ a relative homeomorphism, (X,B.) is a relative ncell, and (X,B,), A. We obtain the diagram (X,BO is a closed path in

X ,
,
t
t
:

X
/r(X',BO

H^X'tB'z) **...** H*(X',BQ
The isomorphisms of morphism
<j>
of the
upper
[<t>']
the path in
X
[lower] line
.4
[X'

compose
A'].
to give the auto0. is
Since
a relative
homeomorphism, 0* is an isomorphism. Since all isomorphisms are induced by inclusions, commutativity holds in each square. It follows
that 0t*
=
0'0*
Since the original path
is
closed, BJ
=
Bi; hence
314,
APPLICATIONS TO EUCLIDEAN SPACES
0*
[CHAP. XI
<f>'
0i
is
=
Since (X,A)
is
orientable,
<t>
is
the identity.
It follows that
the identity; therefore (X',A') is orientable.  A from (X,B) to (X,B f ), and LEMMA 6.6. // P is a path in is the isomorphism of P, then commutativity holds in the diagram
X
=
P*
P*
H (X,B)
n
H\X,B')
H\X,A)
where j,f are
the indicated inclusions.
PROOF.
all
If
P consists of a single step, commutativity is obvious since
we
In the general case, are induced by inclusions. the diagram by inserting the individual steps and the homoenlarge n n > Commumorphisms (X,B>) (X,A) induced by inclusions.
homomorphisms
H
H
tativity in
nected.
each small triangle implies the desired result. LEMMA 6.7. Let (X,A) be a relative nmanifold with
Let (X,B),(X,B
f
X
and
A
A
conB'.
)
be relative ncells such that
Then there exists a path in X A from (X,B) to (X,B'). PROOF. Let F be the family of those relative ncells (X,C) which A starting from (X,B). Let are the end terms of paths in X be the union of the sets X C for (X,C) s F. Clearly is an open A be a limit point of W. As connected set in X A. Let x e X
ACS
C
W
W
X
A
e
is
locally euclidean, there is a relative ncell
e
(X,D) such that
'.
DDA
there
Then y
is
Then
e W X A locally euclidean, a relative ncell (X,C such that C' D C \J D and y X  C'. a path in X  A from (X,C) to (X D). (X,C),(X,C'),(X,D)
and x
X
X
C
for
As X some (X,C)
D.
f
D
is
open,
it
contains a point y
is
e
F.
As
)
e
is
9
Combining this with a path from (X,B) to (X,C), we obtain a path from (X,B) to (X,D). Thus a; e W, and TF is also closed in X  A. = X 4 is connected, we must have As X A. Therefore X J5 contains a point z eW. Treating (X,B'),z in the same manner as we treated (X,Z)),i/ above, we obtain the desired path from (X,B)
W
;
to (X,B'). are
We
now prepared
6.8.
to state
and prove the main
results of this
section.
THEOREM
nected]
let
Let (X,A) be a relative nmanifold with
relative ncell containing
X
A
con
(X,B) be a
Q
(X,A), and
let j:
(X,A)
C
(X,B).
(i)
(ii)
Then we have
the following five propositions:
H
j*
(X A) = Oforq> n. n n maps H (X,B) onto H (X,A).
y
6]
MANIFOLDS
(X,B) and (iv) // (X,A)
(\\\) n
815
The kernel
of j* contains all elements of the
is
u
e
H
form u
P*u where
P
any
closed path in
is orientable, then j*:
X A starting with (X,B). H (X,B) H*(X,A).
n
(v) // the coefficient
group
G
is the
group of
integers,
and (X,A)
is
nonorientable, then
H
n
(X,A)
(iii)
is cyclic of order 2.
PROOF. hence j*(u
To prove
P*u)
k)
we apply
6.6 with j
= j', to obtain j*P* =
j*}
=
0.
We
,
shall say that
(X,A)
(ii),
is finite if
there exists a finite set (X,50,
(X,B
of relative ncells such that
(i),
X  A =
VJj. t
will
X  B
%
.
The
will
propositions
(iv), (v), restricted to the case
(iv)', (v)'
(X,A)
finite,
first
be denoted by
(i)',
(ii)',
respectively.
We
show
that the finite case implies the general case. Let {(X,B a )}, s M, be an indexed collection of relative ncells such
that
A =
O atA
f
Ba
.
The
existence of such a collection
is
is
implicit in
.
the assumption that For each finite set
X C
be the set of those 's such that W^ is connected. Inclusion is a Delations in provide a partial order in N. We will show that set with respect to this order. If x e be the directed A, let C(x) union of all W^ such that e AT and x e TF$. Then C(x) is open. If y is a limit point of C(x) in A, choose a F a containing y. Then F a P\ C(x) contains a point of some W$ which contains x. Hence V a VJ W$ is connected, contains x and y, and is a W, for vj e Af. ThereLet
N
locally euclidean. Let define W^ = a and Af, ai{
A
U
V
Ba Va = X = X  W^. A^
M
N
X
X
fore y
As C(a;) is open and A. connected, we have C(x) = X
e
C(x).
closed in
If
,7;
X
e AT,
A, and choose x e
X
A
T/
is
.
W
.
{,
e
TF n
Since C(x) 9 VJ f
=
.
X
U
e
A, there
tt
is
a
f e A^
Then TF
o>
=
TTj VJ IF,
TJ.
U
Let w = such that x,y s r Tf f is a connected set. Hence
is
W
co
N
It
9
and
contains
and
This proves that A^
directed.
the intersection of the nested system {(X,A f ), e N. By x,2.6, and the continuity of the Cech groups, the Q 9 (X,A) is the limit group of the groups {H (X A^)}. group
follows that
(X,A)
is
H
y
Now
Then the
(i)'
asserts that each
H
q
(X^A^)
=
for q
>
n and each
e
N.
group H*(X,A) is also zero; hence (i)' implies (i). n If u e (X,A), then by definition of direct limit, there is an element n Let v e (X,At). (X,At) such that g*v = u where g: (X,A) f (X,B') be a relative ncell such that (X,A k ) (X,B ), and let / denote
limit
H
H
C
C
the inclusion map. such that j'*w  v.
contains u.
of j*.
Then Thus
6.6,
it
(ii)'
asserts the existence of
w
e
H (X,B')
n
By
6.7
and
the image of (j'g)*: (X,B')*H*(X,A) the image of (j'g)* coincides with the image
follows that
(ii)'
H
9
As u is arbitrary, Choose a e M, let
of those
implies
(ii).
e
AT consist of the single element
e
a
>
and
let
N' be the subset
N such that D
.
Clearly AT'
is
cofinal
316
in
APPLICATIONS TO EUCLIDEAN SPACES
[CHAP. XI
e
N] hence
H
n
(X,A)
is
C
(1)
J?
in
W. Then we
/*
the limit group of have the diagram
\H (X
n
y
A{)} for
N'.
Let
0*
/i*
where
/,<7,&
are inclusions.
Suppose now that (X,A)
orientable.
is
orientable.
By
6.5, CX",A{)
and (X,A,) are
Then
y
(iv)' asserts that /*
g*f* are isomorphisms. It follows that g* is an isomorphism. n projections of the direct system {H (X A^)}^ e N', are isomorphisms.
It follows
and Thus all
from vm,4.8 that (fgh)*
B', it follows that j*
is
is
with
(iv).
B ao =
an isomorphism. If we apply 6.6 an isomorphism. Thus (iv)' implies
Suppose (X,A) is nonorientable relative to integer coefficients. Let be a path such that JP* is not the identity. We may suppose that the collection {(X,B a )} includes the relative ncells of P (adjoin them denote the set of indices of these cells. otherwise), and let s e N. is connected, so Let N' be the set of those Clearly W^ now as in the preceding paragraph, and Proceed such that Now P is a path in e &> obtain the diagram (1) where A^ e N'; hence (X,At) is nonorientable. for each Then (v)' asserts that
P
C M
N
D

X
H
n
(X,At) and
H
n
(X,A^

are cyclic of order 2;
is is
and
(ii)'
asserts that
It follows that g* /* and flf*/* are onto n h* is an isomorphism. Hence (X,A)
an isomorphism.
cyclic of order 2.
H
By vui,4.8,
Thus
(ii)'
and
imply (v). It remains to establish the propositions in the finite case, i.e. X" A = Uj.! (X  B ). If k = 1, then (X,A) is a relative ncell; so (i) Hence (ii) holdo. holds, and 6.3 implies that j* is an isomorphism. Since (iii) has been proved, and the kernel of j* is zero, it follows that (X,A) is orientable, and then (iv) holds. Proposition (v) does not apply if k = 1. Suppose inductively that (i), (ii), (iv), (v) have been proved for finite (X,A) such that k g m. Consider now the case k = m 1. B is A is connected, the union of some m of the sets Since connected. Let the indexing be such that VJ7 (X B,) is connected.  A' is connected, and the Set A' = HT # and B' = B m+l Then
(v)'
t
+
X
X
t
t ,
.
X
inductive hypotheses apply to (X,A'). The relative MayerVietoris sequence (i,15.6) of the triad (X;A',B ) contains the portion
f
(2)
H'(X,A'
U B')

7/'(X,A')
+ H (X,B')
Q
}
>
A
H
q
(X,A) >
^
+1
(^,A' UB')
6]
MANIFOLDS
f
S17
To prove (i) for (X,A), choose a relative homeomorphism /: (E,S) > (X,B ). Let C = f'\A' VJ B'). Then C is closed, and / > defines a relative homeomorphism g: #') (X,A' By 6.2, (E,C)
U
(E C) = (X,A' \J Q that H (X,A)
9^
1
9
H H+
for #
^
?i.
By
x,5.4, g*
n.
is
an isomorphism.
Hence
q
f
l
B
is
f
)
=
for g
^
Then the exactness
of (2) implies
and and
H
(i)
the image of *. By the inductive hypothesis, (X,A ) 9 9 = *(0) = 0; are zero for q > n. Therefore (X,A) (X,B')
H
H
has been proved.
H
n
(X,A')
/*
a
A;
(3)
H\X,D.)
For each component a of
X
^4'
U B',
3
let
.
Ca =
X
a;
and
let
C a We obtain the diagram (X,D a ) be a relative ncell such that D a in which all homomorphisms are induced by inclusions. By 6.3, (3)
(4)
g*
is
an isomorphism.
By
(5)
the inductive hypothesis
(ii)
on (XjA')
t
H*(X,A')
= =
image/*.
As shown
(6)
in the preceding paragraph,
H
e
n
(X,A)
image*.
e
Then, if u> such that
H*(X,A), we can choose u
ty
//"(X,^') and
t;
e
H
n
(X,B')
=
ifr(u,v)
f*u
n
g*v,
see i,15.6.
By
g*v'
(4)
and
v.
=
we can choose u',v' e H (X,D a ) such Since / a / = g a g, we have
(5)
that f*u
f
=
u and
^(7*(u'

i/)
= /*/V //
n
j*j*f;'
=
f*u

g*v
=
w.
This proves
(7)
(^,^)

image
g*.
By
6.7
and
6.6,
H
n
(X,B') and
H
n
(X,B) have the same image in
H
n
(X,A).
Thus
(7) implies
(ii).
318
APPLICATIONS TO EUCLIDEAN SPACES
Having proved
(ii), it
[CHAP. XI
follows that A*
is
onto.
Since g' *h* a
=
g*, (4)
yields
(8)
h*
is
n
an isomorphism.
According to x, Exer. B3,
H
(X A'
y
U B')
a)
is
the injective direct
sum
H*(X,A' VJ B')
Z jr(X,C
a

Z H\X,DJ.
a
{
Referring to the definition of <, it follows that of components of a homomorphism
(/ *,0*)
)
is
the collection
having the same image as
kernel
</>.
Since the sequence
(2) is exact,
0'.
^
n
=
image
<t>
=
n
image
Hence by
(6)
H
n
(X,A)
n
f
[# (X,4')
)
+#
(X,B')
Let us regard (X,B the fact that ^(0,i>) =
(9)
H
as
imbedded
in the direct
sum.
Using
(7)
and
g*v,
we obtain
n
Hn (X,A) = H
n
y
(X,B')/K
where
K = H (X B') C\ image a sum Z w Now an element of Z H (X,D a
n
) is
<*/.
where w
e
H
n
(X,D a )
is
nonzero for only a
finite
number
of a's.
By
definition
Hence w
(10)
e liC if
and only
if
w has the form u
=
Z ^w
where
Z /2X
X
=
0.
A' VJ B'; and, for any component Choose a fixed component 7 of X  A' from (X,Z) a ) to (X,D y ). The path in a, let Q a be a path  A (X D^,(X,B ),(X,D a ) followed by Q a is a closed path P a in beginning and ending with (Jf,Z) 7 ). Set
X
=
f
t
l
va
g* glu a
tHn (X,DT).
1
Using 6.6 we have
E/SU. = E/^Qiffr
Then
(11)
(10)
^.

/?
Z/*
becomes
1*
=
0?
Z
^
where
Z ^^
=
0.
6]
MANIFOLDS
To prove
(iv), let
319
(X,A) be orientable.
Then
By
6.5,
f
(X A')
y
is
orientable.
By
the inductive hypothesis
(iv)
for
(X,A ), f* is an isomorphism. Hence (11) reduces to u = 0. By (10), we have K = 0, so g*: H n (X,B') Hn (X,A). Using 6.6 and a path P from (X,B) to (X,B') it follows that j* is an isomorphism. This
t
proves
(iv) for
(X,A).
To prove
nontrivial
(v), let
G =
integers,
and (X,A) nonorientable.
,
The only
.
automorphism
of
G is
va
Let x be the sum of those v a such that PJt; a = va
.
the reversal of sign; hence P%v a = =bv a v a and y the sum of those such P*v a
(11)
Then
becomes
y)
(12)
u
=
0*(z
+
where
f*(x
t/)
f
=
0.
If CX",A') is orientable, the inductive hypothesis (iv) for
(X,A
)
gives
y = 0; hence u = 20*z. If (X,<A ) is nonorientable, the inductive hypothesis (v) for (X,A ) states that x y has the form 2z. Hence u = 2g$(x In either case u = mod 2. Thus consists of the z).
x
;
f
K
even elements of (X,B'). It follows that (X,A) is cyclic of order This proves (v) for (X,A), and thereby completes the proof of the 2. theorem.
H
n
H
n
REMARK.
is
The
not necessary:
isomorphic to
If
restriction to integer coefficients in proposition (v) n (X,A) is nonorientable, then, for any G, (X,A)
H
is
if
G reduced mod
any closed
we know,
is
for
The proof just given holds in general path P, that P* = db/ where / = identity.
2.
This
we
true by virtue of the universality of integer coefficients which have not proved (see v, Exer. G3). Specifically, if P* = /, re/,
spectively for any G.
for integer coefficients, then
P* =
/,
respectively
7,
It is therefore
customary to use the terms orientable and
n
nonorientable always in the sense of integer coefficients.
THEOREM
group.
6.9.
The nsphere S
e
is orientable relative to
any
coefficient
PROOF.
If
z
relative ncell.
a sterographic projection shows that (S ,^) is a n n n Since H"(S ,x ) = (S ) under the inclusion, it follows
&",
11
H
from
(6.8)
(iii)
that
Sn
is
orientable.
is THEOREM 6.10. Let (X,A) be a relative nmanifold such that homeomorphic to a subset of the nsphere. For each connected component a of a. Then each (X,A a ) is orientable, so that Ay let A a = n A a ) ~ G] and the homomorphisms n (X,A a ) > H*(X,A) induced (X by inclusions provide an injective direct sum representation
X
X
y
X
H
H
H"(X,A)
**
IT(X,AJ.
S20
NOTE
The orientability of (X,A ) follows from 6.9 and 6.5. The direct sum d.ecom posit ion follows from x, Exer. B3. We turn now to the corresponding results for homology. Up through
6.7,
the exactness property was not used.
Hence
all
statements and
proofs through 0.7 hold with cohomology replaced by homology. The proof of 6.8 uses exactness. The Cech homology theory on compact pairs is not generally exact. To obtain exactness we add to
the hypotheses of 6.8 that G is compact or a vector space over a Then the duals of (i)(iv) of p.8 are valid. These are
(i)
field.
H
X
(ii)
(iii)
for q > n. g (X,A) The kernel of j+: H n (X,A) > n (X,B) is zero. = P*u where // u is in the image of j^> then u
=0
H
P
is
any
closed
path in
starting with (X,B). n (X,A) n (X,B). (iv) // (X,A) is orientable, then j+: The duals of 6.9 and 6.10 hold with the same restriction on G.
A
H
H
The
dual of the conclusion of 6.10 reads:
H
The homomorphisms
n
(X,A a ) provide a protective representation of
H
n
n (X,A) (X,A) as a direct
H
product
H
n
(X,A)
TlH (X,A
n
a ).
Cech homology theory is exact for triThis suggests an alternative procedure; namely, add angulable pairs. to 6.8 the hypothesis that (X,A) is triangulable, then show that the proof carries through using only triangulable pairs at each stage. One would use triangulable relative ncells having triangulable unions and For this it would suffice to use a single triangulation of intersections. and its bary centric subdivision. Then the duals of (i) to (iv), (X,A) as stated above, would hold; and, in addition, the dual of (v), which is
restricting G, the
(v) If
Without
G =
integers,
and (X,A)
is nonorientable, then
H
n
(X,A)
=
0.
NOTE
The
ditions
In a multiplication F: (A) There exists
existence of multiplications.
(i),
(ii),
5
we assumed
>
Rn X Rn
Rn
the proposition: satisfying con
and (iii) of 5. remarked that examples of such a multiplication, for n = 1,2,4, and 8, were provided by the real, complex, quaternionic, and Cayley number systems respectively. No other examples are known. If we require in addition that T(x,y) be bilinear and F(z,2/) = \x\ \y\ then Hurwitz has proved [Nachr. Ges. d. Wiss. Gottingen (1898), 309316] that ft = 1,2,4, or 8, and T is isomorphic to the corresponding abovementioned number system. If we drop the norm condition, but retain bilinearity, then Hopf has shown [Comm. Math. Helv. 13 (1941), 219239] that n must be a power of 2.
We
also
f
EXERCISES
It
321
can be shown that (A)
is
equivalent to each of the propositions
For the equivalence of (C) and (D) (B), (C), and (D) listed below. as well as the definition of the terms involved see S. Eilenberg, Ann. of
with a twox = xe for all x e S " ~ '1 X S 1 > Sn of type (1,1). (C) There exists a map S 2n ~ n S having Hopf invariant 1. (D) There exists a map S is of considerable importance in the study of hoProposition (D) motopy groups of spheres, and much effort has been expended in trying to determine the values of n for which it is valid. Hopf has shown Math. 25 (1935), 427440] that, if n > 1, then (D) implies that [Fund. n is even. G. W. Whitehead has proved [Ann. of Math. 51 (1950), 192237] that, if n > 2, then (D) implies that n is divisible by 4. J. Adem has announced [Proc. Nat. Acad. Sci. 38 (1952)] that (D) imof 2. H. Toda has announced [C. R. Acad. Sci. plies that n is a power Paris 241(1955), 849850] that D is false for n = 10.
S""
l
Math. 41 (1940), 662673. n~ (B) There exists a multiplication S
e:
l
sided unit
ex
=
n~ l
X
1
Sn
~l
.
11
71
l
EXERCISES
DEGREES OF MAPS. Define the degree of a map /: group fI (S). Show that this degree 1. that 4.3 remains valid for n
A.
1.
S
is
>
S by
0,
always
using the reduced Show 1. 1, or
n n~
l
l
Define the degree of a map /: (",5""') > (E ,S ), and show ~ n~ S n defined by /. that it is equal to the degree of the map g: S  1. 3. Show that a map /: (E ,S) has degree 0, 1, or (E\S) 4. Prove that, for each n > 1 and each integer k, there exists a
2.
l

l
map
n
n~
l
/:
(E ,S
)

(E S
f
n
n~
l
)
of degree k.
B. INVARIANCE OF DOMAIN. be a closed subset of 1. Let
X
Sn and
x e Int X, then H (X,X borhoods U of x 2. Prove the converse of
n
.
U)
1
^
Show that, if let X Q e X. for all sufficiently small neigh
assuming both continuity and exactness.
a family of closed subsets of X. A a and i a (X,A) Let A = (X,A a ). Assume that an exact and continuous homology theory (on the category & c of compact pairs) is given and that, for a fixed integer q,
\
C. CARTAN'S MATCHING PROCESS. be a compact space and [A a Let
X
C
:
C
H Q+l (X,B\J A a =
)
for all
a
a
where
B
is
any
finite intersections of
the sets \A
}.
322
1.
EXERCISES
Prove
that,
if
u
q
e
H
Q
(X,A)
is
such that
ia
u
=
2.
^u
=
for all a, then
0.
i ajt
(X,A) determines a family {u a \ where u a = > k ft>a u where fc ai/} q (X,A q (X,A a ) satisfying k a ^u a is induced by inclusion. Show that each family {u a Q (X,A a A0) = k &a u^ is obtained in this fashion from some u e satisfying k a ^u a
Each u
e
H
w
eH
H
U
a)
:
ft
H
}
H
f
(X,A.).
3.
A and A a form a base for X Assume that the open sets X H Q (X,A a ) are such that k a0 u a = % whenever A a C A that u a s Show that there is some u e H Q (X,A) with t a+ u = w a for all a.
.
ft
4.
Transcribe the results of
6 and of the preceding exercises into
the "single space notation" of x,7. 5. Assume that the coefficient group
G
is
a domain of integrity, and
l
deduce from 6 results concerning coBetti numbers. In particular show ~ n n A has R n (A ;G) + J that if A is a proper closed subset of S then S
components.
Index
Index
abstract, category, 108; cell complex, 156 acyclic, carrier, 171; complex, 170 additive functor, 121
coefficient group, 17, 18
cofinal, 212;
subsystem, 214
adjusted homology sequence, 245
admissible, category, 5; map, 306 algebraic, homotopy, 170; map, 170 alternating chain complex, 174 associated LCtheory, 271
cohomology functor, 112 cohomology group, 13; Cech, 237; of a
chain complex, 153; of a complex, 92, 166; singular, 188
cohomology sequence
triad, 37; triple,
of
a,
pair,
14;
29
augmented, complex, 169; singular complex, 190
cohomology theory,
automorphism of a simplex, 82 axioms for cohomology, 14 axioms for homology, 10
bary center, 61
barycentric, coordinates, 55; subdivision,
61, 63, 177
13; Cech, 237; of chain complexes, 153; of complexes, 166; singular, 188 cohomotopy groups, 49 collapsible, 18
compact,
4; carriers,
255
compactification, 270, 277
complement, 74
completely regular, 276 complex, finite, 56; infinite, locally finite, 75, 183 component, 254 composition of functions, 3 contiguous, 164
74,
base, 133; point, 17 Betti number, 52, 138
162,
bihomomorphism, 160
Borsuk's separation boundary, 56, 74, 90
criterion,
302
boundary of a chain, 88 boundary operator, 10, of a couple, 127;
of a traid, 35; of a bounded, 269
triple,
24
continuous, 4 continuous homology theory, 260 contractible to a point, 30
contravariant, functor, 116; dfunotor, 117 coordinate, 131
111;
5functor,
Brouwer
fixed point theorem, 301
canonical, base, 139; imbedding, 55; map into a nerve, 286 carrier 170, 234; compact, 255 cartesian product, 67, 131 category, 109; of all pairs, 5; of compact
pairs, 5; of locally
countercompact, 269 couple, 114, 126
covariant, functor, 111; dfunctor, 115;
covering, 233; 187
C.S., 14
5functor, 116 of a pair, 234; theorem,
compact
pairs, 5;
of triangulable pairs, 76 category with couples, 114
cube, 276
cycle,
Ccch homology groups, 237; system, 237
44; relative, 311; complex, 156 cfunctor, 114 chain, 84; complex, 124; homomorphism, 84; homotopy, 129
cell,
90
cylinder, 210
circuit,
106
deformation, chain, 129; retract, 30 degenerate chain, 173 degree of a map, 304, 306
closed, covering, 233; path, 313; sot, 4 closure, 4
development, 288
diagonal form, 139; map, 67 dimension, axiom, 12; of a complex, 56 direct, couple, 126; homology groups, 184; limit, 220; product, 131; sum, 8, 10, 32, 132; system, 212; system of upper sequences, 224
coboundary, 92; of a cochain, 90 coboundary operator, 14; of a triad, 37; of a triple, 29 cochain, 86; complex, 125 cocycle, 92
S25
826
directed set, 212 distance in a simplex, 55 domain, 109; of integrity, 52 duality, 14
INDEX
homomorphism,
groups,
9,
7;
of
a sequence of
15 homotopic, 6; chain maps, 129 homotopy, 6; axiom, 11, 12; equivalence, 29, 117; groups, 48, 207; inverse, 117
elementary, chain, 163; subdivision, 74, 75 enlargement, 261 equivalence, 109
equivariant, 73;
essential,
H.
S., 11
identity, 108;
map, 3
homology groups, 209
299
euclidean complex and simplex, 73
Euler characteristic, 53 exact sequence of groups, 8 exactness axiom, 11; partial, 253 excision axiom, 11, 12
excision
map, 12, 130; for complexes, 165; generalized, 117 excisions of various types, 267 Ext, 161
face, of a simplex, 55; of a singular sim
image, 7, 9 imbedding, 61 incidence isomorphism, 79, 80; number, 156 inclusion map, 3 index, 169; of a singular chain, 190 indexed family of sets, 233 induced homomorphism, 10, 14, 239 inessential, 299 infinite, complex, 74; triangulation, 75 infinitely divisible, 159
injection, 8, 132,
interior,
214
injective representation, 8, 133
plex, 186
factor, complex, 125; sequence, 9 finite cochains, 183 five lemma, 16 fixed point theorem, 301 formal homology theory, 166 free, base, 133; module, 134
full,
4
intersection of a nested system, 260
Inv, 258 invariance, 180 in variance, of dimension, 302; of domain,
303
invariant,
subcomplex, 70; subcategory, 110
functor, 111
fundamental theorem of algebra, 307
generalized excision, 117 generate, 133
grating, 254 group, 6; of
73; under relative homeomorphisms, 266 inverse homology group, 184 inverse limit, 215 inverse system, 213; of factor groups, 228; of lower sequences, 224; of subgroups, 228
isomorphism,
7,
9
homomorphisms, 147;
of
join, 74, 172,
operators, 73
192
Hausdorff space, 4 /icategory, 117
kernel, 7, 9
lattice of
hexagonal lemma, 38 Wunctor, 118 Horn, 147, 152
homeomorphic, 17
homologically
trivial,
a pair, 4 LCtheory, 271 Lebesgue number, 65 line segment, 72
linear, 6;
22
imbedding, 61; independence,
homology functor, 112 homology group, 10; Cech, 237; equivariant, 209; of a chain complex, 125, 151; of a chain mapping, 169; of a complex, 90, 166; of a triad, 203; singular, 188 homology sequence of a, pair, 11; triad,
map, 57, 60; metric, 61; singular simplex, 192; subspace, 6 locally euclidean, 303
52, 72, 133;
locally finite complex, 75, 183
lower cap, 44 lower sequence of groups, 8
35; triple, 25
homology theory,
tinuous,
10; Cech, 237; conequivariant, 209, for locally compact spaces, 271 for normal spaces, 282; singular, 188
main isomorphism, 94
manifold, 311
260;
;
map,
3; linear, 57, 60; simplicial, 58,
60
map
defined
by a map, 3
INDEX
mapping
cylinder, 210
327
mappings, 109
maps of inverse and direct systems, 213 Mayer homology groups, 182
MayerVietoris, formula, 53; sequence,
37, 39, 42, 43 mesh, 60; of a covering, 251 metric topology, 75 module over a ring, 6 monomial, 308
reduction, 261 refinement, 234 regular, covering, 241; family relative to a subspace, 262; neighborhood, 70, 71
relative, cell, 311;
homeomorphism, 266; manifold, 311; Mayer Vietoris sequence, 42; topology, 4 representation by generators and relations, 134
retract,
multiplicative system, 108
natural,
30
equivalence,
112;
homomor
phism, 7, 10; transformation, 112 nerve, 234 nested system, 260 nonorien table, 313; circuit, 106 norm, 44 normal pair, 278 normal space homology theory, 282
objects, 109
sequence of groups, 8; of order two, 224 sign changing trick, 117 single point compactification, 270 single space homology theory, 273 187; 186; complex, singular, chain,
188; prism, 193; prismatic chain, 193; prismatic complex, 194; simplex, 186 simple circuit, 106 simplex, 54; singular, 186
simplicial, approximation, 64; complex,
homology groups,
onto,
7,
9
open covering, 233 open set, 4 open simplex, 56 open star, 57
operator group, 73 ordered, chain complex, 163; complex, 67, simplex, 55 orientable, 313; circuit, 106
padic group, 230; solenoid, 230 pair of, sets, 3; groups, 10; sequences, 10; topological spaces, 4 partial exactness, 253 path, 210, 313 point of a simplex, 55
pointlike, 130 principal ideal ring, 134
56, map, 58, 60; product, 66 skeleton of a complex, 84 solenoid, 230, 296 sphere, 44 split exact sequence, 230 stacked covering, 241
172 313 strong deformation retract, 30 subcategory, 110 subcomplex, 56, 125 subcube, 276 subdivision, 61, 63, 74, 75, 177
star, 57,
step,
subfunctor, 113 subgroup, 6
prism, 192 product, 66, 67, 131
projection, 8, 67,
131, 177, 234; spec
subsequence of groups, 9 subsystem, 214 suspension, 48
tensor product, 140, 150 theory, 282 topological, invariance, 180; space, 4 Tor, 161
trum, 253
projective, representation, 8, 132; space,
106 proper, coverings, 249; triad, 34
quasicomponent, 254 quasiorder, 212 quotient functor, 113
range, 109 rank, 52
number, 138 system of groups, 17 triad, 34; homology groups, 203 triangulation, 60 triple, 24
torsion
transitive
TychonofT, compactification, 277; map, 277 types of excisions, 263
18, 19,
reduced homology, group, 191; sequence, 20
170,
uniform homotopy, 282
828
unimodular, 135
INDEX
variety, 226
uniqueness, 179; category, 120; theorem, 100 unit, 108 unit simplex, 55 universal coefficient theorems, 160
vector space, 6
vertex, 54 Vietoris cycles, 253
weak, continuity, 293; topology, 75
zero dimensional, 254
upper cap, 44 upper sequence of groups, 8