Maclane Homology
Maclane Homology
Maclane Homology
DR. S A U N D E R S M A C L A N E
MAX MASON DISTINGUISHED SERVICE PROFESSOR OF MATHEMATICS
AT THE UNIVERSITY OF CHICAGO
WITH 7 FIGURES
SPRINGER-VERLAG
BERLIN G U T T I N G E N H E I D E L B E R G
1963
Geschaftsfhnde Herausgeber:
Prof. Dr. B. ECRMANN,
EidgenossischeTechnische Hochxhule Zilrich
Prof. Dr. B. L. VAN nru WAERDEN,
Mathematisches Institut der Universitst Ziirich
Alle Rechte,
insbesondere das der Ubersetzung in fremde Sprachen,
vorbebalten
Ohne ausdriickliche Genehmigung des Verlages
ist es auch nicht gestattet, dieses Buch oder Teile daraus
auf photomechanixhem Wege (Photokopie, Mikrokopie) oder auf andere Art
zu vervielfaltigen
TO DOROTHY
Preface
In presenting this treatment of homological algebra, it is a pleasure
to acknowledge the help and encouragement which I have had from
all sides. Homological algebra arose from many sources in algebra and
topology. Decisive examples came from the study of group extensions
and their factor sets, a subject I learned in joint work with OTTO SCHILLING. A further- development of homological ideas, with a view to their
topological applications, came in my long collaboration with SAMUEL
EILENBERG; to both collaborators, especial thanks. For many years
the Air Force Office of Scientific Research supported my research
projects on various subjects now summarized here; it is a pleasure to
acknowledge their lively understanding of basic science.
BAER and JOSEF SCHMID read and commented on
Both REINHOLD
my entire manuscript; their advice has led to many improvements.
ANDERS
KOCK and JACQUES RIGUET have read the entire galley proof
and caught many slips and obscurities. Among the others whose suggestions have served me well, I note F RANK ADAMS, LOUIS AUSLANDER,
WILFRED COCKCROFT, ALBRECHT DOLD, GEOFFREY HORROCKS, FRIEDRICH KASCH, J O H A NN LEICHT, ARUNAS LIULEVICIUS, J O H N MOORE, DIETER P UPPE, JOSEPH YAO, and a number of my current students at the
University of Chicago - not to mention the auditors of my lectures
at Chicago, Heidelberg, Bonn, Frankfurt, and Aarhus. My wife, DOROTHY,
has cheerfully typed more versions of more chapters than she would
like to count. Messrs. SPRINGER have been unfailingly courteous in the
preparation of the book; in particular, I am grateful to F. K. SCHMIDT,
the Editor of this series, for his support. To all these and others who
have helped me, I express my best thanks.
Chicago, 17. February 1963
SAUNDERS
MAC LANE
Table of Contents
Introduction
............................
Page
. . . . . . . . . . . . . .
. . . .
2. Modules . . . . . . . . .
3. Diagrams . . . . . . . . .
4. Direct Sums . . . . . . .
5. Free and Projective Modules
6. The Functor Hom . . . . .
7. Categories . . . . . . . .
8. Functors . . . . . . . . .
.
I . Differential Groups . . . . .
2. Complexes . . . . . . . . .
3. Cohomology . . . . . . . .
4. The Exact Homology Sequence
5. Some Diagram Lemmas . . .
6. Additive Relations . . . . . .
7. Singular Homology . . . . .
8. Homotopy . . . . . . . . .
g. Axioms for Homology . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
Chapter 111. Extensions and Resolutions . . . . . . . . . . . . . . . .
I
. Extensions of Modules . . . . . . . . . . . . . . . . . . . . .
2. Addition of Extensions . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
34
35
39
42
44
49
51
54
57
61
63
63
67
. . . . . . . 72
. . . . . . . 76
. . . . . . . 82
. . . . . . . 87
. . . . . . . 92
. . . . . . . 95
. . . . . . . 96
. . . . . . . 99
. . . . . . . 102
.
Chapter IV. Cohomology of Groups . . . . . . . . . . . . . . . . . . 103
1. The Group Ring . . . . . . . . . . . . . . . . . . . . . . . . 104
2. Crossed Homomorphisms . . . . . . . . . . . . . . . . . . . . 105
3. Group Extensions . . . . . . . . . . . . . . . . . . . . . . . 108
4. Factor Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5. The Bar Resolution . . . . . . . . . . . . . . . . . . . . . . 114
6. The Characteristic Class of a Group Extension . . . . . . . . . . 120
Table of Contents
IX
. . . . . . . . . . . . . . Pace
121
. . . . . . . . . . . . . . . . . . . 124
. . . . . . . . . . . . . . . . . . . 129
. . . . . . . . . . . . . . . . . . . . . . . 131
. . . . . . . . . . . . . . . . . . . . . 134
Chapter V. Tensor and Torsion Products . . . . . . . . . . . . . . . . 138
1. Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . 138
2. Modules over Commutative Rings . . . . . . . . . . . . . . . . . 141
3. Bimodules . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4. Dual Modules . . . . . . . . . . . . . . . . . . . . . . . . . 146
5. Right Exactness of Tensor Products . . . . . . . . . . . . . . . 148
6. Torsion Products of Groups . . . . . . . . . . . . . . . . . . . 150
7. Torsion Products of Modules . . . . . . . . . . . . . . . . . . 154
8. Torsion Products by Resolutions . . . . . . . . . . . . . . . . 159
9. The Tensor Product of Complexes . . . . . . . . . . . . . . . . 163
10. The KUNNETH
Formula . . . . . . . . . . . . . . . . . . . . 166
11. Universal Coefficient Theorems . . . . . . . . . . . . . . . . . 170
Chapter VI. Types of Algebras . . . . . . . . . . . . . . . . . . . . . 173
1. Algebras by Diagrams . . . . . . .
2. Graded Modules . . . . . . . . . .
3. Graded Algebras . . . . . . . . .
4. Tensor Products of Algebras . . . .
5. Modules over Algebras . . . . . . .
6. Cohomology of free Abelian Groups .
7. Differential Graded Algebras . . . .
8. Identities on Hom and @ . . . . .
9. Coalgebras and HOPFAlgebras . . . .
Chapter VII. Dimension . . . . . . . . . . . . . . . . . . . . . . . U)O
i. Homological Dimension . . . . . . . . . . . . . . . . . . . . 200
2. Dimensions in Polynomial Rings . . . . . . . . . . . . . . . . . 204
3. Ext and Tor for Algebras . . . . . . . . . . . . . . . . . . . . 206
4. Global Dimensions of Polynomial Rings . . . . . . . . . . . . . 210
5.SeparableAlgebra.s . . . . . . . . . . . . . . . . . . . . . . 211
6.GradedSyzygies . . . . . . . . . . . . . . . . . . . . . . . 215
.
.
.
.
.
7. Local Rings .
.........................
Chapter VIII. Products . . . . . . . . . . . . . . . . . . . . . . . .
1. Homology Products . . . . . .
2. The Torsion Product of Algebras
3. A Diagram Lemma . . . . . .
4. External Products for Ext . . .
5. Simplicial Objects . . . . . . .
6. Normalization . . . . . . . . .
7. Acyclic Models . . . . . . . .
8. The EIL~NBERGZILBER
Theorem
9. Cup Products . . . . . . . . .
220
. . . . . . . . . . . . . . . 248
. . . . . . . . . . . . . . . . . . . . . . 249
. . . . . . . . . . . . . . . . . . . . . . . 254
218
Table of Contents
page
. Categories of Diagrams . . . . . . . . . . . . . . . . . . . . . 257
. Comparison of Allowable Resolutions . . . . . . . . . . . . . . . 260
. Relative Abelian Categories . . . . . . . . . . . . . . . . . . . 262
Relative Resolutions . . . . . . . . . . . . . . . . . . . . . . 265
. The Categorical Bar Resolution . . . . . . . . . . . . . . . . . 270
Relative Torsion Products . . . . . . . . . . . . . . . . . . . 273
3
4
5
6.
7
8.
g.
. . . . . . . . . . . . . . . . . . . 278
. . . . . . . . . . . . . . 280
. . . . . . . . . . . . . . . . . . . . . . . . . 280
. . . . . . . . . . . . . . . . . 280
. . . . . . . . . . . . . . . . . 283
. . . . . . . . . . . 288
.
. . . . . . . . . . . 290
. . . . . . . . . . . 293
.
. . . . . . . . . . . . . . . . . . 295
.
. . . . . . . . . . . . . . . . . . 298
. . . . . . . . . . . . . . . . . . . . 301
. . . . . . . . . . . 303
. . . . . . . . . . . . . 308
. . . . . . . . . . . 311
. . . . . . . . . . . 315
Chapter XI. Spectral Sequences . . . . . . . . . . . . . . . . . . . . 318
I. Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . 318
2. Fiber Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 322
3. Filtered Modules . . . . . . . . . . . . . . . . . . . . . . . 326
4.Transgression. . . . . . . . . . . . . . . . . . . . . . . . . 332
5. Exact Couples . . . . . . . . . . . . . . . . . . . . . . . . . 336
6. Bicomplexes . . . . . . . . . . . . . . . . . . . . . . . . . 340
7. The Spectral Sequence of a Covering . . . . . . . . . . . . . . . 342
8. Cohomology Spectral Sequences . . . . . . . . . . . . . . . . . 345
g. Restriction, Inflation, and Connection . . . . . . . . . . . . . . 347
I 0. The Lyndon Spectral Sequence . . . . . . . . . . . . . . . . . 351
1I
. The Comparison Theorem. . . . . . . . . . . . . . . . . . . . 355
Chapter XI1. Derived Functors . . . . . . . . . . . . . . . . . . . . 358
I. squares . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
2. Subobjects and Quotient Objects . . . . . . . . . . . . . . . . 361
3. Diagram Chasing . . . . . . . . . . . . . . . . . . . . . . . 364
4. Proper Exact Sequences . . . . . . . . . . . . . . . . . . . . 367
5. Ext without Projectives . . . . . . . . . . . . . . . . . . . . 371
6. The Category of Short Exact Sequences . . . . . . . . . . . . . 375
7. Connected Pairs of Additive Functors . . . . . . . . . . . . . . 379
8. Connected Sequences of Functors . . . . . . . . . . . . . . . . 386
g. Derived Functors . . . . . . . . . . . . . . . . . . . . . . . 389
10. Products by Universality . . . . . . . . . . . . . . . . . . . . 394
II. Proper Projective Complexes . . . . . . . . . . . . . . . . . . 397
. . . . . . . . . . . . . . . . . 400
12. The Spectral K ~ ~ N N EFormula
TH
Direct Products of Rings
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
List of Standard Symbols . . . . . . . . . . . . . . . . . . . . . . . 413
I n d e x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Bibliography
Introduction
Our subject starts with homology, homomorphisms, and tensors.
Homology provides an algebraic "picture" of topological spaces,
assigning to each space X a family of abelian groups H,(X), ..., H , ( X ) ,
. . . , to each continuous map f : X+Y a family of group homomorphisms
f,: H,(X) +H, (Y). Properties of the space or the map can often be
effectively found from properties of the groups H, or the homomorphisms
f,. A similar process associates homology groups to other Mathematical
objects; for example, to a group n o r to an associative algebra A. Homology in all such cases is our concern.
Complexes provide a means of calculating homology. Each %-dimensional "singular" simplex T in a topological space X has a boundary
consistini of singular simplices of dimension .n- 1. If K, is the free
abelian group generated by all these %-simplices,the function a assigning
to each T the alternating sum aT of its boundary simplices determines a
This yields (Chap.11) a "complex" which
homomorphism a:K,+K,-,.
consists of abelian groups K, and boundary homomorphisms a, in the
form
a
a
a
a
O+K,+K,+K,+K,+...
.
Moreover, aa= 0, so the kernel C, of a: K, +K,-, contains the image
aK,,,.
The factor group H, (K) = C,/aK,+, is the %-th homology
group of the complex K or of the underlying space X. Often a smaller
or simpler complex will suffice to compute the same homology groups for
X. Given a group IT, there is a corresponding complex whose homology
is that appropriate to the group. For example, the one dimensional
homology of ll is its factor commutator group 17/[U,IT].
Homomorphisms of appropriate type are associated with each type
of algebraic system; under composition of homomorphisms the systems
and their homomorphisms constitute a "category - (Chap.1). If C and
A are abelian groups, the set Hom (C,A) of all group homomorphisms
f : C -+A is also an abelian group. For C fixed, it is a covariant "functor"
on the category of all abelian groups A; each homomorphism a:A+A1
induces the map u, : Hom (C,A) -+Horn (C,A') which carries each f into
its composite u f with u. For A fixed, Hom is contravariant: Each
y:Cf-+C induces the map y* in the opposite direction, Hom(C,A) +Hom (C', A), sending f to the composite f y. Thus Hom ( ?,A) applied
0
Introduction
involving the functor Extl just introduced. If the K,, are not free groups,
there is a more complex answer, involving the spectral sequences to be
described in Chap. XI.
Tensors arise from vector spaces U, V, and W and bilinear functions
B (u, v) on U x V to W. Manufacture the vector space U @ V generated
by symbols u@v which are bilinear in U E Uand veV and nothing
more. Then u@v is a universal bilinear function; to any bilinear B
there is a unique linear transformation T: U@ V+W with B (zl, v) =
T(u @v). The elements of V@V turn out to be just the classical tensors
(in two indices) associated with the vector space V. Two abelian groups
A and G have a tensor product A @G generated by bilinear symbols
a @ g ; it is an abelian group, and a functor covariant in A and G. In
particular, if K is a complex, so is A @ K: A @K,t A @Kl+. Question: Does the homology of K determine that of A @ K ?
Answer: Almost yes; if each K is free, there is an exact sequence
Here Tor, (A,G) is a new covariant functor of the abelian groups A and
G, called the "torsion product"; it depends (Chap.V) on the elements
of finite order in A and G and is generated, subject to suitable relations,
Introduction
by pairs of elements aeA and geG for which there is an integer m with
ma=O=mg.
' Take the cartesian product X x Y of two spaces. Can we calculate
its homology from that of X and Y ? A study of complexes constructed
from simplices (Chap.VII1) reduces this question to the calculation of
the homology of a tensor product K @ L of two complexes. This calculation again involves the torsion product, via an exact sequence (the
Kiinneth Thm, Chap. V)
'
Now generalize; replace abelian groups by modules over any commutative ring R. Then Ext l (A,G) ist still defined as an R-module, but
the longer sequence may now fail of exactness at Extl(A,G). There is
a new functor Exta(A,G),a new connectinghomomorphismE*:Extl (A,G)
+Exta(C,G), and an exact sequence extending indefinitely to the right
as
Ext"+l (C, G) +.
1
.. .
Introduction
(n,
11
if.
if,
i:
o t A t Z $ +-q'+ttt.. .
Since P, is projective, the map P, +A lifts to f,:P, +Pi. The composite
map &-+Pi lifts in turn to an f,:P,-+P; with af,=f,,a, and so on by
Introduction
Introduction
Introduction
C h a p t e r one
2. Modules
(rrf)a=r(r'a),
It follows that Oa = O and (-1)a = - a. Some authors define an Rmodule without requiring that l a =a, and call a module with this
property unitary. In this book, every ring has an identity and every
module is unitary.
Our treatment of left R-modules will apply, mutatis mutandis, to
right R-modzcles. They are abelian groups A with ar E A defined so as to
satisfy the corresponding four identities; for example a (rr') = (ar)r'.
Modules appear in many connections. In case R is a field or a skew
field, a left R-module is a left vector space over R. If F is a field and
R=F[x] the polynomial ring in one indeterminate x with coefficients
in F, then an R-module is simply a vector space V over F together with
10
When a: A-tB, call A the domain and B the range of a. The image
Im (a) =aA consists of all elements a a for a E A ;it is a submodule of the
range B; the kernel Ker(a) consists of all a in A with a a =0; it is a
submodule of the domain A. If aA = B, we say that a is an e$imor#hism
and write a: A +B, while if Kera = 0 we say that a is a monomorfihism
and write a : A *B. Finally, a is an isomorphism if and only if a is both a
monomorphism and an epimorphism. For each module A, the identity
function I A : A-tA is an isomorphism. For any A and B, the zero or
"trivial" function 0 with every O(a)= O is a homomorphism 0: A-tB.
A homomorphism w : A 4 A with range and domain equal is called an
endomorphism.
If al, a, : A+B are homomorphisms with the same domain A and the
same range B, their sum al f a2, defined by (al a,) a = al a a,a, is an
R-module homomorphism al f a , : A+B.
If a : A-tB and /?: B+C are R-module homomorphisms, the composite function /fa is also an R-module homomorphism /fa: A-tC; but
2.
Modules
11
note that this composite is defined only when Range a =Domain ,!I.
The composition of homomorphisms is associative when defined. A
(two-sided) inverse of a: A-tB is a homomorphism a-l: B+A such that
both aa-l=IB and a-la = I A . Moreover, a has an inverse if and only if
it is an isomorphism, and the inverse is then unique. We write a: A B
when a is an isomorphism. A left inverse of a is any homomorphism y:
B-tA with ya =IA ; it need not exist or be unique.
A pair of homomorphisms(a,B) with Range a =Domain B = B,
+ + +
+ +
Proof. Set B' (b T) =Bb; since T< Ker B, this is well defined. In
particular, if 8: B+Bf is an epimorphism with kernel T,Pf: BITEB'.
This result may be worded: Each B with /3 (T)= 0 factors uniquely
through the projection q. This property characterizes q: B+B/T up
to an isomorphism of BIT, in the following sense:
Proposition 2.2. If T < B and 5 : B+D is such that c(T)=O and
each B: B 3 B' with P (T)= 0 factors uniquely through 5, there is an isomorphism 0 : B I T s D withc=Bq.
12
Coker a = B/Im a ,
called the coimage and the cokernel of a. This definition gives two short
exact sequences
Ker a
A+ Coim a,
Im a -B+
Coker a ,
(2.2)
3. Diagrams
13
inverse image of T. I n particular, Ker a = c l O , where 0 denotes the submodule of B consisting only of the zero element.
For K<S<A the module S/K is called a subquotient of A ; it is a
quotient module of the submodule S of A , and simultaneously a submodule of the quotient module A/K. Furthermore, if K<K1<S'<S<A,
then K1/K is a submodule of S1/K and the composite projection St+
S1/K+ (S1/K)/(K'/K)
has kernel K t , hence the familiar isomorphism
(S1/K)/(K'/K)
G S1/K'. This allows us to write each subquotient
(S1/K)/(K'/K)
of a subquotient S/K directly as a subquotient of A.
Let S/K be a subquotient of A , S'IK' one of A'. If a : A-tA' has
a S < S' and a K < K', then a s K' is a coset of S1/K' uniquely determined
by the coset s+ K of S/K. Hence a, (s+ K ) =as+ K' defines a homomorphism
a, : S/K -+S1/K'
(a S < S', a K < K')
(2.4)
3. Diagrams
O+AAB$C+O
1. .Is b
o +A'ZB
I cl+o
~
(3.1)
i4
Proof. Clearly (ii) and (iii) together yield (i). To prove (ii), take
b E Ker @. The right square is commutative, so y a b =a'p b = 0 ; as y is a
monomorphism, this means that a b = 0. Since the top row is exact, there
is an element a with xa =b. Now the left square is commutative, so
x'aa =pxa = @b=O. But the bottom row is exact at A', so aa = 0. Since
a is a monomorphism, a = 0, and hence b=xa =0. This proves @ a
monomorphism.
To prove (iii), consider any b' in B'. Since y is an epimorphism there
B
is a C E Cwith yc =arb'; since the top row is exact, there is a ~ E with
a b = c. Then a' Gg b- b') = 0 in C'. The exactness of the bottom row
yields an a' E A' with x'a' =@ b - b'. Since a is an epimorphism, there is
an aEA with aa=af and hence with Bxa=xfaa=@b-b'. Then
b' =@ (b -xu) is in the image of @, q. e. d.
This type of proof is called "diagram chasing". Inspection shows
that the chase succeeds just as well if the groups are non-abelian (multiplicative) groups.
By the same method, the reader should verify the following more
general results (as formulated by J. LEICHT):
have exact rows. If al, a,, a,, a, are isomorphims, so is Q. I n more detail,
(i) If al is an efiimorfihism and a, and a, monomor~hisms,then a,is a
monomorfihism,
I5
4. Direct Sums
Proof. Chase the diagram, or apply Lemma 3.2 twice to the lefthand and right-hand portions.
4. Direct Sums
The external direct sum A,@A, of two R-modules A , and A , is the
R-module consisting of all ordered pairs (a,,a,), for aie A i , with module
operations defined by
The functions L and n defined by L, a,= (a,,0 ), &,a,= (0,a,), n, (a,,a,) =q,
n,(a,, a,) = a, are homomorphisms
A,& A,BA,,C
Zl
11
A,
(4.1)
81
n2hz0,
3~2~2=1~,1
(4.2)
h%tl+h%=1~,t~~~.
Call 1, and r, the injections and q,n, the projections of the direct sum.
The diagram (4.1) contains partial diagrams, to wit:
A@
%
A,-.
~ , * f r -A,.
A,ZA,@ A 2 2A,,
A,@ A,= A,,
A,-!$ A,@ A , Z A , ;
of the form (4.1) and satisfying the five iderttities like (4.2) i s isomorphic to
the direct s m diagram. I n more detail, there i s exactly one isomorphism
16
for i = 1 , 2 .
(4-3)
.-
A short exact sequence with one (and hence all) of these properties
is said to split (some authors say instead that the sequence is inessential).
Proof. We just observed that (iii) implies (i) and (ii). Conversely,
exactness shows that L' gives the isomorphism Al=Kernu, so (i) implies
(iii) by Prop. 4.2. Similarly, (ii) implies (iii).
Now consider pairs of coterminal homomorphisms a,,a,, as in the
diagram
D: A , ~ B ~ A , .
(4.4)
Such a diagram is said to be universal with ends A, and A, if to every
diagram D': A1+B1+-A, with the same ends there exists a unique
17
4. Direct Sums
with D as first row and end maps the identities, there is a unique way
of inserting the middle dotted arrow so that the whole diagram becomes
commutative @%=a;, p a 2 = 4 ) .
with D as first row and with vertical maps 1 on each Ai, there is a unique
way of inserting the middle dotted arrow to make the diagram commutative. The reader should prove
Mac Lane, Homology
18
n,
n,
Then
A, is an R-module. The homomorphisms n,: 17, A, + A
defined by n,f = f (t) are called the projections of the cartesian product.
For given A , , let {y,: B+A,} be a diagram with one additional module
B and one homomorphism y, for each t E T . This diagram is couniversal
with ends A, if to each diagram {y;: B1+ A, 1 t E T } there exists a unique
B: B1-+Bsuch that y; = y, for all t. The projectionsof the cartesian product
A, yield such a couniversal diagram, and any two such diagrams
are isomorphic, as before.
The externd direct sum A, of the same modules A, is that submodule
of 17,A, which consists of all those functions f with but a finite number
of non-zero values. The homomorphisms 6,: A t + z t At are defined for
each a A , by letting it (a) be the function on T with [ i t (a)](t)= a ,
[ 4 (a)](s)=0 for s +t. These homomorphisms are called the injections of
the direct sum. As in the case of two summands, the diagram
{ i t : A t + z t A,} is universal for given ends A,, and is determined up to
isomorphism by this fact.
For a finite number of summands the external direct sum is identical
with the cartesian product. This implies that any finite universal
diagram ai: A -+ B, for j = 1, ... , n, yields a couniversal diagram
{yi: B+Aj}. More explicitly, each yi is that map which is uniquely
determined (since B is universal) by the conditions yi aj= 1 A , , yi ak= 0
for j k. Dually, the reader should obtain a universal diagram from the
couniversal one.
Direct sums may be treated in terms of submodules. If S, is any
family of submodules of B indexed by a set T, their union US, is the set
of all finite sums s,+ ... s, with each sj in some St ; it is a submodule
of B containing all the St and contained in any submodule which contains all the s,. Their intersection n S , is the intersection of the sets S, ;
n,
z,
19
it is a submodule of B contained in all the S, which contains every submodule contained in every S t . We also write S 1 u S, or Sl nS, for the
union or intersection of two submodules S l , S,.
Proof. Given (i), B is isomorphic to 2 S t , which satisfies (ii). Conversely, given (ii), the condition B = U S t states that each b + O can be
written as a finite sum b = sl+ . s, of elements si 0 belonging to
different submodules St,, i = I , . .., n ; the second condition of (ii) states
that this representation is unique. For any other diagram {a,: St+B1} the
s,) =at, s1 .. . ah s,
homomorphism /I: B-tB' defined by /I (s,+ ..
is the unique homomorphism with /?j, =a, ; hence the universality.
When these conditions hold, B is called the internal direct sum of its
submodules S t . Therefore an internal direct sum is isomorphic to the
external direct sum 2 S t . In particular, B is the internal direct sum of
two submodules Sl and S, if and only if S,n S, = 0 and S1u S, = B ;these
conditions imply B rS,@ S , .
+ +
Exercises
nl L, = I , n, I, = 1 , n, L,= 0,and ( L , , n,) exact
20
z,
'
:
'
B*,
1.
(5.1)
bC
21
22
(6.1)
(6.3)
a*
HornR(A, B) ;
one shows that a*al* = (ala)*. This reversal of order generalizes the
fact that the transpose of the product of two matrices is the product of
their transposes in ofiposite order. Because of this reversal we shall say
that HomR(A,B), for B fixed, is a co.ntravaria.nt functor of A.
Now vary both A and B. Given a : A+A1 and /?:B-tB', each
f : A'+B determinesa composite #Ifa: A+B'; the correspondence f -+#I f a
is a homomorphism
Hom (a, /?): Hom (A', B) +Hom (A, B')
of abelian groups, with a*/?, = Hom(a, /?) =/?,a*. It has the properties
Hom (I, 1') =the identity,
Hom (aa',
@/?I)
23
a+
(6.4)
These
Similarly, Horn (a,
B2)= Horn (a, /I,) Horn (a, /I,)
. two properties state that Horn is an "additive" functor.
For B fixed, apply Horn(--, B) to a direct sum diagram (4.1). The
6:
result
Horn (A,, B) 7 H o m (A, @A,, B)
om (A,, B)
n~
r:
changes injections to projections 6 7 , but by (6.4) still satisfies the identities (4.2) for a direct sum diagram. Similarly, for A fixed, a direct sum
diagram on modules B, and B, is carried by Horn ( A , -) to a direct sum
diagram (injections to injections). Thus
Hom(A,@A,, B)sHom(A,, B)@Hom(A,, B),
Horn (A, Bl@ B,) r H o m (A, B,) @ Horn (A, B,) .
In particular, Horn (A, B,)
exact
>
(6.7)
i s exact.
A sequence M+B+C+O exact at B and C is called a short right
exact sequence. This theorem states that the functor Horn,(-, D) for
fixed D turns each short right exact sequence into a short left exact
sequence ;by the previous theorem, Horn, (D,-) carries a short left exact
sequence into a short left exact sequence. If A - B I C is a short exact
24
A)+Hom,(D,
(6.6')
o-tHomR(C,D)-+HomR(B,D)+HornR(A,D)-+?.
(6.7')
By the two theorems above, each is exact except perhaps at the right
end. With 0 for ? on the right, these would not usually be exact. For
example, exactness of (6.6') at Hom, (D, C) would assert that each
h: D-tC has the form h =crhlfor some h': D+B; i. e., that each map hinto
the quotient C = B/xA could be lifted to a map h' into B (as would be
possible were D projective). To see that this need not be so, take R=Z
and D=Z, the cyclic group of order m. For the short exact sequence
Z-Z+Z,,
with first map x the operation of multiplication by m, the
sequence (6.6') becomes 0 -to -+O+Horn (Z,,Z,)-tO, and is manifestly
inexact. Similarly, (6.7') can be inexact with a zero at ?, since for A <B a
homomorphism f : A+D cannot in general be extended to one of B into D.
I t will be possible to describe an object which is the "obstruction" to the
problem of extending such an f. The group of these objects, placed at
"?" in (6.7'), will restore exactness. This construction, done for both
(6.6') and (6.7'), is one of the objectives of homological algebra.
We now can formulate several characterizations of projective modules.
is exact.
Suppose La+ L.
(i) The sequence (6.63 need not be exact with zero for ? on the right. Show this
for D =R/L by proving that Hom, (RIL,R) +Hom, (RIL,R/L) is not an epimorphism
(1 is not an image !).
7. Categories
25
(ii)The sequence (6.7') need not be exact with zero for ? on the right. Show this
for D = L b y proving that HomR(R,L) +HomR(L, L) is not an epimorphism ( I not
an image I).
2. For any set T o f indices establish an isomorphism
2 At+B
sntHornR( A t ,B)
17 HornR( A , Bt).
t
7. Categories
(i) A family of disjoint sets hom(A, B), one for each pair of objects;
(ii) For each triple of objects A , B, C a function which assigns to
a E hom (A, B) and p E hom (B, C) an element p a E hom (A, C) ;
(iii) A function which assigns to each object A an element
I,~hom(A,A);
all subject to the two axioms:
Associativity : If a E hom (A, B), p E hom (B, C), and y E horn (C, D),
then y (Ba)= (yS)a ;
Idedity. If a E hom (A, B), then a l A = a = 1, a.
, and call a a morphism of V with
Write a : A+B for a ~ h o m ( AB)
domain A and ralzge B. By (ii), the composite p a is defined if and only if
range a =domain /?; the triple composite ypa is associative whenever it
is defined. Call a morphism x an identity of V if both xa = a whenever xcr
is defined and ,9x =p whenever px is defined. Each I, is an identity.
Conversely, if x is an identity, then x: A+A for some object A, and
x = X I A =1 A : Each identity of V has the form 1 A for a unique object A.
In other words, the identities of V determine the objects of V. I t is
possible to describe a category simply as a class of morphisms, with a
composite sometimes defined and subject to suitable axioms (Ex. 3
below).
A morphism 8: A+B is called an eqzlivalelzce in V if there is in V
another morphism p,: B+ A such that p, 8 =1 A and 8 p, = 1,. Then p, is
unique,forif alsop,'8=lA, thenp,=fAp,=p,'8p,=y11B=p,'. Callp,the
ilzverse p, = 8-1 of the equivalence 8. The composite of two equivalences,
when defined, is an equivalence.
26
r=
is commutative.
(Ply11 ...>P,Y,).
r:
is
7. Categories
One may also have the category of exact sequences infinite to the
right or infinite to the left, or both. Another example is the category of
y) of
short exact sequences E: A-B+ C, with morphisms all triples (a,#?,
module homomorphisms for which the appropriate diagram (3.1) is
commutative. I t is by now amply clear how more examples can be
constructed ad libitum -a category of sequences of exact sequencesof.. .
I t is also clear that a number of concepts applicable to modules will
apply to the objects of any category - provided the definition of the
concept makes reference not to the elements of the modules but only to
modules and their homomorphisms. Thus, in any category V , a diagram
consisting of morphisms a,: A, +C of V ,one for each t in a given set T ,
is universal for the given objects A, (or, a direct sum diagram for the A,)
if to each diagram {a;: A, -+ CfIt E T} on the same A, there exists a unique
C+Ct of V with #?a,=a; for each ~ E T (For
.
T ={I, 21,
morphism #?:
this is exactly the property formulated in (4.5)). The previous
uniqueness proof for the direct sum of two modules carries over verbatim
to give
28
product which does exist in the category of all abelian groups is nondenumerable).
To each category one may construct an opposite category WP. Take
the objects of Vpto be a class in 1-1 correspondence A*@A with the
objects A of V . Take the morphisms to be a class in 1-1 correspondence
a** a with the morphisms of W. Decree that a* : A*+B* if and only if
a : B+A, and that a*p* isdefined and is @a)*,exactly whenpais defined.
Then V P is a category, and any statement G* about the category V is
the same as the original statement G about the category WP. This,
again, shows the dual B* of a provable statement G provable. The 1-1
function T with T(A) =A* and T(a) =a* is an "anti-isomorphism" of
Q to VP, since T@a)=T(a) TP).
Subsequently, we shall define a special sort of category, called an
"abelian category ', by requiring essentially that hom (A, B) be an abelian
group and that kernels and cokernelsexist, as in the case of the category of
modules. It turns out that many theorems about modules remain true
when the modules and their homomorphisms are replaced by the objects
and the morphisms of any abelian category. The interested reader may
turn at once to Chaps. I X and XII.
J
Exercises
I . In the category of topological spaces, show that the disjoint union of two
spaces provides a direct sum diagram, and that the cartesian product X x Y of two
spaces, with its usual topology and with the natural projections on X and Y, provides a direct product diagram.
2. Show that any two objects in the category of groups may be ends of a direct
product diagram and of a direct sum diagram. (Note: The "direct sum" for not
necessarily abelian groups is more often known as the "free product".)
8. Functors
Let V and 9 be categories. A covariant fulzctor T on %?to 9 is a pair
of functions (both denoted by the same letter T): An "object function"
which assigns to each object C E V an object T(C)E 9, and a "mapping
8. Functors
29
T v y )=T@) T ( y ),
(8.2)
30
T v y )= T(y)T@),
By defined in V .
(8.5)
which assigns to each vector v that function hv: DV+F with (hv)f = f (v)
for f c DV . For finite dimensional V , h ( V ) is the familiar isomorphism
31
8. Functors
of V to its double dual. One verifies readily that h is a natural transformation h: I+Da (where I denotes the identity functor).
There is a similar natural isomorphism of a finite abelian group to
its double character group.
As an example of a non-natural isomorphism, recall that there is an
isomorphism k: VYD(V) for any finite dimensional vector space V.
Specifically, for each such V choose a fixed basis v,, . . . , vn, construct in
D(V) the dual basis vl, ... , vn, with vi defined by the requirement that
vi (vi) is 0 or 1 according as i j or i =j, and set kvi = vi. This linear
transformation k =k(V): V+D(V) is defined for each V; it maps the
covariant identity functor I to the contravariant functor D. If we restrict
attention to the category whose objects are finite dimensional vector
spaces and whose morphisms are isomorphisms a of such, we may replace
D by a covariant functor
with B(V)=D(V), D(a)=D(a-l). But
k(V): V--fD(V) is not natural. For example, if V is I-dimensional and
a: V+V is defined by avl = cvl for some scalar c E F with 0 c / l , then
D(a) k (V)vl = (I/c)v1 ;however k (V)avl= cvl, SO (8.3)isnot commutative.
Functors in several variables may be covariant in some of their
arguments and contravariant in others. Two arguments, contra and co,
suffice to illustrate. Let 8, V, and 9 be three categories. A bifunctor T
on 8 x w to 9,contravariant in and covariant in %,' is a pair of functions: An object function which assigns to B E 99 and C E V an object
T(B, C) 9,and a mapping function which assigns to morphisms
j3: BJB' and y: C+C1 a morphism
(8.7)
the latter to hold whenever both composites p'j3 and y'y are defined.
The composite on the right is then defined, for j3': B1+B" and
y': C1+C" with (8.7) give
T(5'
T(B", C) -%T
C") .
32
LT(B,c)
T(B.c)~
(8.10)
S (B, C')
-1
T(B,Y)
f (B C')
(8.11)
T(B,C')
34
Notes. The idea of a module goes back a t least to KRONECKER, who considered
modules over polynomial rings; only in the last twenty years has this idea taken on
its present central role in algebra. Projective modules were first used effectively in
CARTAN-EILENBERG; now i t is clear that they provide for linear algebra the appropriate generalization of a vector space (which is always a free module). EMMY
NOETHER,
in lectures a t Giittingen, emphasized the importance of homomorphisms.
The initial restriction to homomorphisms a : A+B with a (A)= B, as in VAN DER
WAERDEN'S
influential Moderne Algebra, soon proved to be needlessly restrictive,
and was dropped. By now i t is expected that each definition of a type of Mathematical system be accompanied by a definition of the morphisms of this system.
The arrow notation developed in topological investigations about 1940, probably
starting with the use for correspondences and then for continous maps. Exact
sequences were first noted in HUREWICZ [i940]. The functor "Hom" was long
known, but apparently first appeared by this name in EILENBERG-MACLANE
[1942]. Categories and functors were introduced by the same authors in 1945.
They have proved useful in the formulation of axiomatic homology (Chap. I1 below),
in the cohomology of a sheaf over a topological space [GODEMENT 19581, in differential geometry [EHRESMANN 19581, and in algebraic geometry (GROTHENDIECKDIEUDONN
k [I 9601, cf. also the review by LANG [1961]). Foundational questions
about the theory of categories, using sets and classes, are formulated in M A C LANE
[19611.
Chapter two
Homology of Complexes
Here we first meet the basic notions of homology in simple geometric
cases where the homology group arises from a boundary operator. In
general, an abelian group with a boundary operator is called a "differential group" or, when provided with dimensions, a "chain complex".
1. Differential
35
Groups
1. Differential Groups
A differential grot@ C is an abelian group C together with an endomorphism d: C+C such that dz= 0; call d the "differential"or "boundary
operator" of C. Elements of C are often called chains, elements of Kerd,
cycles, and elements of Imd, boundaries. The requirement that dZ=O is
equivalent to the inclusion Imd<Kerd. The homology group of the
differential group C is defined to be the factor group of cycles modulo
boundaries,
H(C)= Ker d/Im d = Ker d/d C.
(1-1)
Its elements are the cosets c+ Imd of cycles c; we call them homology
classes and write them as
cls(c)=c+dC~H(C).
(14
Two cycles c and c' in the same homology class are said to be homologous;
in symbols c c'.
a
As first examples we shall give a number
of specific differential groups with their homology. Most of these examples will be found
by dissecting a simple geometric figure into
cells and taking d to be the operator which q
P
assigns to each cell the sum of its boundary
cells, each affected with a suitable sign.
Exaqtle I . Take two points p and q on
a circle S1 which divide the circle into two
b
semicircular arcs a and b. The "boundary"
Fig. I
or "ends" of the arc a are the points q and p.
Hence introduce the free abelian group C(S1) with the four free generators
a, b, p, and q, and define an endomorphism d of C(S1) by setting
Any element of C(S1) is represented uniquely as a linear combination
%a +m2b +%p +m4q with integral coefficients q,
m2, %, and m,,
while
d(~a+m2b+m,~+m4q)=%(q-p)+mz(~-q)=(%-mz) (q-P).
Thus C(S1) is a differential group. Its cycles are all the integral linear
combinations of p, q, and a+ b, while its boundaries are all the multiples
3*
36
H(f): H(C)+H(C')
(for f : C+C1).
(1.5)
37
Differential Groups
qu
-----+-------*'
Fig. 2
Define d : C+ C by setting
ayax
38
d ( a x I ) = % - (qxI)-a,+
(+XI)
and d(bxI)=bl- bo+ (qxI)-(#XI).
39
2. Complexes
for all c in C(S1). This equation may be read: The boundary d (sc) of the
cylinder sc over c consists of the top flc minus the bottom f0c minus the
cylinder s(dc) over the boundary of c. This equation implies that the
homomorphisms H(fl) and H(fo)are equal, for if c is any cycle (dc=O),
then (1.7) gives f,c-f,,c=d(sc), whence flc-foe.
Maps with the property (1.7) will appear frequently under the name
of "chain homotopies".
Exercises
I . L e t C b e a differential group. T h e definition H ( C ) = KerdlImd can be
written as H ( C )=Coker (d': C+ K e r d ) ,where d'is induced b y d . UsingC/Kerd= I m d ,
show that H ( C ) has a dual description as Ker (d": (Cokerd)-+C).
2. For a family C t , tET, o f differential groups, define the direct sum
Ct and
the direct product 17 Ct and prove that H ( z C t ) ~Hx( C t ) ,H ( n C t ) s U H ( C t ) .
2. Complexes
, ,
40
much of the literature. Two n-cycles in the same homology class (clsc=
clsc') are said to be homologous (c-c'); this is the case if and only if
c-c'E~K,+~.
If K and K' are complexes, a chain transformation f : K+Kf is a
family of module homomorphisms f,: K,+KL, one for each n, such that
a:f,= f,-,a, for all n. This last condition asserts the commutativity of
the diagram (neglect the dotted arrows)
K:
...-
a':
f'rr
g':
2.
Complexes
41
,-,.
ef=1,
a l ~ O + f ~ = I ~ an+l~n+~n-la,,=l
e,
(n>o).
(2.5)
42
Exercises
1. Call a complex S q-seecia1 if Sn= 0 for n+ q, q+ 1 and 8: Sqtl+ Sq is a monomorphism. Prove that any complex K of free abelian groups K, is a direct sum of
q-special complexes (one for each q).
2 . Call a q-special complex S of abelian groups elementary if either Sq= Sq+,= Z
or S q = Z , S4+1= O . Prove that each special S with Sq, Sq+l finitely generated free
groups is a direct sum of elementary complexes. (Hint: use row and column operations on matrices of integers to choose new bases for Sq and Sq+,.)
3. Prove that any complex with each K,, a finitely generated free abelian group
is a direct sum of elementary complexes.
3. Cohomology
3. Cohomology
43
-,
(aKk)
k ~ K p , f ~ H o m n , (3.5)
44
(aH+,=aLsp+ sp-laK,
n=i .
(4.4)
of chain complexes and chain transformations x, a. The first transformation x has kernel zero, but the induced map Hn (x): H,(K) +H,(L) on
homology may have a non-trivial kernel, as in Example 1.3. To study
when this can happen, identify K with the subcomplex x K of L and
consider a cycle c of K , whose homology class becomes zero in L. This
means that c=aZ for some (n+ 1)-chain ZcL, and hence that the coset
I + K,,, is a cycle of the quotient complex L I K r M . Conversely, any
homology class of H,+, (LIK)consists of cycles I+ K,,, with a1= C E K , ,
hence yields a homology class clsc in H, ( K ) which is in the kernel of
H, (x). This correspondence of I+ K,+, to c is a homomorphism
H,,, (LIK)+Hn ( K ) which we now describe systematically.
In (4.1),let m be a cycle in Mn+,. Since o is an epimorphism, one
can choose 1~ L,+, with ol=m. Since am=O, one has oal=O; since E
is exact, there is a unique cycle C E K with
,
xc=al, as in
45
The homology class cls (c)E H, (K) is independent of the choice of 1 with
ol=m, depends only on the homology class of m, and is additive in m.
Hence 8, (clsm)= clsc defines a homomorphism
ol=m
for some 1.
(4.3)
This suggests the notation c=x-lao-lm; or regard cls as a homomorphism cls,: C,(K)-+H,(K); then aE is defined by a "switchback"
formula a, = (cls,) x-la a-I (cls,) -l - even though the inverses cls-l,
x-l, a-l are not strictly defined (but see $6 below).
a,,
x, =
This sequence (4.4) is infinite in both directions, but is zero for n < 0
when the complexes are positive. I t gives the desired description of the
kernel and cokernel of H, (x): H,, (K) +H, (L) when x is a monomorphism ;
namely the kernel is a,H,+,(M), and the cokernel is isomorphic to
o* H, (4.
46
.I
E': o+K'+L'-+M'+o
.If
.Ig
aE: H,+,
-.-+ H,,,
(L) -% H,,,
k*
ig*
. . -+H,+,(L')%
H,+,(M')%
if*
ig*
H,(K')% H,,(L')
+... .
(4.7)
Here the rows are the exact homology sequences of Thm.4.1 for E and
E' and the whole diagram is commutative; for example, the left hand
square because a: g, = (a'g) , h, a, = (ha), and a'g = ha by the commutativity of (4.5). The conclusion may be formulated thus: A morphism
of E to E' induces a morphism of the exact homology sequence of E
to that of E'.
The mapping cone of a chain transformation f : K -+Kt gives an example of this exact sequence. The problem is that of fitting the induced
maps f , : H,(K) -+H,(K1) on homology into an exact sequence. For
this purpose, construct a complex M = M ( f ) , called the mapping cone
of f (or sometimes, with less accuracy, the mappifig cylinder off), with
47
is exact.
Proof. To construct the cohomology of E, first apply the contravariant functor Horn,(-,G)
to E to get the reversed sequence of
complexes
E*: 0 +Hom (M,G) +Horn (L,G) +Hom (K, G) +O .
Since the given sequence splits as a sequence of modules, E* is exact. The
connecting homomorphism a,,: H-,+,(Horn (K,G)) +H-,(Horn (M,G))
for E*, when written with upper indices Hu-l=H-u+l, is the desired
connecting homomorphism dE. By Prop.4.2 it follows that 6, is natural when the arguments Hu(K,G) and Hn+l(AC,G) are regarded as
48
(4.9)
o +Hom
is exact, and yields 6, as 8,. , with the usual shift to upper indices, and
with (4.11) as a consequence of Thm. 4.1.
Note the explicit rule for constructing ds. Let f : K, +G" be a cocycle. Since S, is exact, we may write f=zg for g: Kn+ G a cochain;
since f is a cocycle, ga=ilh, where h : Kn+,-+G1 is a cocycle. Then
49
Exercises
1. If f , g: K -+ K' are chain homotopic, show that the associated exact sequences
for the mapping cones M(f) and M(g) are isomorphic.
2. (The BOCKSTEIN Operator.) Let K be a complex of free abelian groups,
Z p the additive group of integers modulo the prime p, and S= (A, r): Z n Z +Zg
the short exact sequence with A multiplication by 9 . Construct the corresponding
exact sequence (4.11 ) and show that p= rt ds: H n ( K , Z p )+H"+'(K,Z,) can be
described as follows. Lift each n-cocycle c : K n + Z p to an n-cochain a: K n + - 2 ;
then Ga= pb for some b: K,+l + Z , and p(cls c)= cls ( t b ) . This p is known as
the BOCKSTEIN
cohomology operator [cf. BROWDER
19611.
exact, where r ] =
4. Show that the exact sequence of Ex. 3, that of Prop. 4.3, and those for F
and G all appear in a "braid" diagram
..
Hn+,(X') - H n + , ( N )
/
"
/'
H%+1( f qHn+l(IM)
/
" I
Hn-I (L) +
/"
-..
Hn ( f K )
/"
which is commutative except for a sign (- 1 ) in the middle diamond [MAC LANE
1960 b].
P P Y
su##ose that all three columns and the first two rows (or the last two rows)
are short exact sequences. Then the remaining row is exact.
Mac Lane, Homology
50
.1.
with exact rows; the kernel and the cokernel of this morphism are short
sequences, but need not be exact (example: map 0 -A = A to A =A +0
with /?=
I,). The horizontal maps of the diagram do induce maps which
give exact sequences
O+Keru+Ker/?-tKery
and
Cokeru -tCoker/? +Cokery +O .
They can be combined in a long exact sequence:
Lemma 5.2. For any commutative diagram
with exact rows there is a map D,: Kery +Cokeru, lzatural for functors
of the diagram D, such that the sequence
6. Additive Relations
51
Proof. Let L : Kery -tC be the injection, 7 : A1+A'/aA, the projection. The switchback formula D, =7 %'-It9a-, L then defines D,
without ambiguity. To prove the exactness of (5.2), say at Cokera,
suppose x i (a' + a A ) = O for some a'. Then x1a'=,9b for some b and
a'x'a = yab = 0, so a b c Kery has D, ab = a' a A , which is the required
exactness. The rest of the proof is similar.
We call D, the connecting homomor~hismof the diagram D.
Now we prove Thm.4.f for the short exact sequence E of complexes
K - L +M. Let C, ( K ) denote the module of n-cycles of K and form
the diagram
KnIaKs+l -tLn/aLn+l +MnIaM,+l +0
a
a
D ( E ):
0
C,-l(K) + C,-l(L) + C,-l(M)
.l *
.la*
.l *
-+
with exact rows and vertical maps induced by a. The first kernel is
C, (K)/aK,,,= H, ( K ) ,and the first cokernel is Cn-, (K)/aK,= H,-l ( K ) ,
so the Ker-Coker sequence (5.2) is
homology sequence.
2. If in the hypotheses of the 3 x 3 lemma one assumes only the first and third
rows exact, show that the second row need not be exact, but will be exact if PIP2= 0.
3. Under the hypotheses of the 3 x 3 lemma, establish exact sequences
6. Additive Relations
The "switchback" formulas can be justified in terms of "additive
relations". They will appear later in the treatment of spectral sequences.
An additive relation r : A- B is defined to be a submodule of the direct
sum A@ B; in other words, r is a set of pairs (a,b) closed under addition
4"
52
Indr = Kerr-l.
(6.1)
<
Here Kerr <Def r A and Indr < Imr < B. Def r is the domain of definition of r, while Ind is the "indeterminancy" of r, and consists of all b
with (0,b) ~ r Moreover,
.
r is the graph of a homomorphism if and only
if Defr= A and Indr=O.
For example, the converse of a homomorphism p: B +A is an additive relation P-l with Defp-l=Imp, Indp-l=Ker,!?. In a complex K
the set of pairs (c, clsc) for C E C, (K) is an additive relation cls: K,,H,(K) with Def (cls)= C, (K). With these observations, the "switchback" formula for the connecting homomorphism appears as the composite of additive relations.
Any additive relation can be regarded as a "many-valued homomorphism; more exactly, as a homomorphism of a submodule to a
quotient module:
j ,
j: Defr+A,
B determines a homo-
n : B+B/Indr,
(6.2)
Proof. Given a Def r, (a, b) Er and (a,b') cr imply (0,b- b') cr, hence
b- b ' Indr.
~
Then P (a)= b+ Indr defines a homomorphism P with the
desired form (6.2). Conversely, given P, r is the set of all pairs (s,b)
with b@(s).
A similar argument shows that each additive relation r induces an
isomorphism 13,: (Def r)/(Kerr) z (Imr)/(Indr); conversely, each isomorphism of a subquotient of A to a subquotient of B arises in this
way from a unique additive relation r.
6. Additive Relations
53
(6-3
defined to be the set of all pairs (s+K, sl+K') of cosets with scS,
sfS , and s f = as. This includes the previous notion of induced homomorphisms.
For an equivalence one can detqmine the inverse of an induced
relation.
f
provided (i) either a K > K' or /? K' < K" and (ii) either a S < S' or/?-lS"< Sf.
Proof. Suppose first that (s+ K, s"+ K") E /I
By definition of
the composite of two relations, there are s; and s, m S' with s;+ K t =
s;+K1 and as=s;, Bs;=s1'. Thus s;-s;=kf~Kf, and pas=s"+/?kl.
In case either /? Kt<K" or K < a K this gives (s+ K, s"+ K") E (/?a)#, so
hypothesis (i) gives /?#a#<@a)#. Similarly, (ii) gives the opposite inclusion.
Exercises
1. For each additive relation v: A- B, prove vrlv=v.
2. For additive relations v and s, prove (vs)-l=s-lrl.
3. If u = A - A is an additive relation with u-l= u = u2, prove that there are
submodules K < S < A with u = [(s,s + k) I s E S, k E K] . Establish the converse.
4. For each additive relation v: A- B, describe v r l and r l r .
5. Under the hypotheses of the strong Four Lemma (Lemma I.3.2), prove
E a-l= /I-'?.
f
54
7. Singular Homology
The use of complexes may be illustrated by a brief description of
the singular homology groups of a topological space. We first introduce
affine simplices.
Let E be an n-dimensional Euclidean space; that is, an n-dimensional
vector space over the field of real numbers in which there is given a
symmetric, bilinear, and positive definite inner product (u,v) for each
pair of vectors u, v~ E. The usual distance function Q (u,v) = (u-v, u- v)h
makes E a metric space and hence a topological space. In particular,
E may be the space Emof all n-tuples u = (a,, . . . , a,) of real numbers a,,
. . . , a,)
with termwise addition and with the standard inner product (6,
(bl, . . . , b,) = 2 ai b, .
The line segment joining two points u , v ~ Eis the set of all points
tu+(l--t) v, for t real and O s t s l ; that is, of all points xou+xlv,
where x,, xl are real numbers with xo+ xl = 1, x, 2 0, x, 2 0. A subset
C of E is convex if it contains the line segment joining any two of its
points. If uO,. . ., u, are m+ 1 points of E, the set of all points
is a convex set containing u,, .. ., urn and in fact the smallest convex
set containing these points; it is called the convex huIl of u,, ..., u, .
The points u,, . . ., u, are said to be affine independent if every point
of this convex hull has a unique representation in the form (7.1); the
real numbers xi are then the barycentric coordinates of u relative to
u,, . .., u,. I t can be shown that the points u,, ..., u, are affine independent if and only if the vectors y- u,, . . . , ~ d , - uo are linearly
independent.
An affine m-simplex is by definition the convex hull of m+ 1 affine
independent points. These points are the vertices of the simplex. Thus
a 1-simplex is a line segment, a 2-simplex is a triangle (with interior),
a 3-simplex is a tetrahedron, etc. For each dimension n we will take a
standard affine n-simplex An in the space E n , and we will label the vertices of An as (0, 3 , . . . , n). (For example, take 0 to be the origin and
1, . . . , n a basis of n orthogonal vectors in En.)
For any topological space X, a singular n-simplex T in X is a continuous map T: An+X. Thus a singular 0-simplex of X is just a point
of X, or, more accurately, a map of the standard point A O into (a point
of) X. We first construct certain singular simplices in convex subsets
of E.
Let E and E' be Euclidean spaces, L : E +E' a linear transformation
and u; a fixed vector of E'. The function f (u) =a; L(u) on E to E'
is called an affine transformation f : E+E'. As the composite of the
linear transformation L with the translation by u;, f is continuous.
7. Singular Homology
55
An+C.
(7.2)
8. Homotopy
57
,-,,
Exercises
8. Homotopy
Two continuous maps of a space X into a space Y are said to be
"homoto~ic"if it is possible to continuously deform the first map ihto
the second. Consider the deformation as taking place in a unit interval
58
F(x,I)=f,(x).
(8.1
59
8. Homotopy
In other words, the segment in An+' joining vo to each point u' of the
opposite face is mapped by S T linearly onto the segment joining WEC
to T(u)E C. In particular, since An is compact, TAn is compact and hence
bounded, so ST: An+'-+C is continuous at xo=O.
This s: Sn(C)+Sn+,(C) provides a contracting homotopy for the
augmented complex 8: S (X) +Z. In the notation of (2.5), let f : Z +
S(X) be the chain transformation which carries I EZ t o the singular
0-simplex T, at the chosen point w E C. The i-th face di (sT) is the singular
n-simplex obtained from (8.3) by setting xi= 0. Hence do(ST)=T,
while dj+,(sT)=sdiT if n>0 and dlsT=To if n=O. This gives a(sT)=
T-s(aT) for n>O, asT=T-feT for n=0, and & f = l , all as in (2.5).
Hence S (X) is acyclic, as required.
More generally, consider any homotopy F: X x I +Y. Regard X x I
as a cylinder on the base X ; the boundary of this cylinder is the top
(where F= f,) minus the bottom (where F= fo) minus the sides (i.e.,
minus F on (3X)xI). The resulting schematic formula a F = fl- f,,-Fa
suggests the definition as=fl-fo-sa
of a chain homotopy. These
indications can be made precise, as follows:
Theorem 8.2. If f o rfi: X - f Y are homotopic continuous maps, the
induced chain transformations S (f,), S (f,): S (X) +S (Y) are chain homotopic.
We reduce this theorem to the special case of the cylinder X x I .
By the base b and the top t of this cylinder we mean the continuous
maps b, t : X + X x I defined by b ( x )= (x,0) and t (x)= (x, I) ; they are
clearly homotopic.
Lemma 8.3. For any cylinder there is a chain homotopy u : S (t)s~ S (b).
The lemma implies the theorem. For let F: X x I + Y be any homotopy F: fO=fl. Then F b = fo,F t = f,, and S(F)is a chain transformation.
Define s as the composite
S=
Then a s + s a = S ( F ) ( a ~ + u a ) = S ( F ) ( S ( t ) - S ( b ) ) = S ( f l ) - S ( f , ) .
To prove the lemma, we prove more: That u = u, : S (X)
+S ( X xI)
can be chosen simultaneously for all topological spaces X so as to be
natural. For each continuous map g: X + X of spaces, naturality
requires that the diagram
f
S,(X)ZS
(XxI)
(g)
n+[s ( g x 1)
sn(Xf)% s,,+l(X'x I)
(8.4)
I
60
auL=s(t)J,- s(b)L-un-1aA-
(8.5)
a u T = S ( T x I ) a a = S ( T x l ) [ S ( tJn) S(B)J,-21,-,aJ,],
where t and b are top and base for A n x I . But t, b, and us-, are all
natural, hence (8.5) gives auT= S (t)T - S (b)T - un-,aT, as desired.
This type of proof consists in first constructing the desired object
(here, the desired chain homotopy) on a model chain such as J, by
observing that the space A n x I in which the model lies is acyclic, and
in then carrying the object around to other spaces by the maps T .
It is an old method in topology; it will reappear later (Chap.VII1) as
the method of acyclic models. Here it has the merit of avoiding an
explicit formula for the homotopy u.
..., n',
d',
if,
those of the top. Show that an explicit homotopy u forX= AninLemma 8.3
is given, in the notation for affine singular simplices, by
61
3. For n = I , 2 as in Ex.2,show that the terms of uJ, correspond to a "triangulation" of the prism A n x I (Draw a figure).
4. Show that A n x I can be "triangulated" a s follows. Partly order the vertices
of A n x { O ) and A " X { I )by the rule that (i,.s)$(j,v) for s.q=O, 1 , if i l j and
e s v . Take as simplices of the triangulation all those formed by a linearly ordered
subset of the whole set of vertices, and show that the resulting n-simplices are
those appearing in uJ, in Ex. 2.
H,(X,A) =C,(X,A)/B,(X,A)
(9.2)
...,
(9.3)
62
Two maps f,, f,: (X,A) +(Y, B) of pairs are homotopic if there is
a homotopy F: f o ef,: X+Y with F ( A x I ) < B ; this last condition
means that F cut down to A x 1 is a homotopy between fo and f, cut
down to maps of A into B. An extension of the argument for Thm. 8.2
shows that homotopic maps f,, f, of pairs have Hn(f,) = Hn(f,): Hn(X,A)
+Hn (Y, B).
The singular homology theory for pairs of spaces thus gives:
1. Functors Hn(X,A) of pairs of spaces to abelian groups, n=0,
1, ... .
2. Natural homomorphisms a, : Hn(X, A) +Hn-, (A), lz = 1, 2, . . . .
These data satisfy the following additional conditions:
3. If X consists of a single point, H,(X) s.2 and Hn(X)= O for
%>O.
4. For any pair (X, A) the relative homology sequence (9.3) is exact.
5. Homotopic maps of pairs induce equal homomorphismson each H, .
6. (Excision.) If X>A>M are spaces such that the closure of M
is contained in the interior of A, let X- M>A- M denote the subspaces obtained by removing all points of M from X and from A,
respectively. Then the injection k of X- M into X induces isomorphisms
on the relative homology groups
(9.4)
Our discussion has indicated the proofs of all except the sixth
property; a proof of this uses "barycentric subdivisions"; it may be
found in EILENBERG-STEENROD [1952], WALLACE [195 71, or HILTONWYLIE [ I 9601.
These six properties may be taken as axioms for homology. It can
be proved that when the pair (X,A) can be "triangulated" by a finite
number of affine simplices, any relative homology groups satisfying
the axioms must agree with the singular homology groups. Moreover,
from the axioms alone one can calculate the singular homology groups
of elementary spaces to agree with those calculated from "naive"
subdivisions in $1. In particular, if Sn is the %-sphere,one deduces
that H n (S n ) GZ, H, (Sn)SZ and Hi(Sn)= O for O+ i+%. This, and
other striking geometric properties (Brouwer fixed point theorem, etc.)
are presented in EILENBERG-STEENROD [1952], Chap. XI.
We have now completed our too brief indication of the use of homology theory in topology.
Notes. "Complex" originally meant simplicia1 complex; in topology "complex"
has various geometric meanings, such as "cell complex" or "CW-complex". ~ h k
chain complex in our purely algebraic sense was introduced by MAYER [1929,
I. Extensions of Modules
63
Chapter three
O-+A+Jo+ J,+
r:
64
triple r = ( a ,
8, y)
Proposition 1.1. For any abelian group A and for Zm(c)the cyclic
group of order m and generator c there is a I - 1 corresfiondence
where mA is the subgroup of A consisting of all ma for acA.
1. Extensions of Modules
65
E l c =E ,
E(y y')
= (Ey) y'.
(r,
the sides and the bottom are given; we wish to fill in the module at
" ? " and the dotted arrows so as to make the diagram commutative
and the top row exact. To do so, put at ? that subgroup B'< B e C'
which consists of the pairs (b, c') with a b =y c'; define a' and #
as
I
ul(b, c') =c', g (b, c') =b. This choice insures commutativity in the righthand square of (1.3). With the definition %'a= (xu,0) the diagram is
completed; the remaining conditions may be verified.
Uniqueness proof: Take any other such E" with a morphism r"=
(IA,g", y): EU--+E.If B" is the middle module ' E", defineg': B"+ B'
by g'b" = (/Y"'", a"bU); then G= (IA,B', jC,): "+E' is a congruence
E'+E
and the composite Eu-+E'+E is r",so t b t the diagram
is unique up to a congruence 4 of El, as asserted.
We call El= E y the composite of the extension E and the homomorphism y; the type of construction involved occurs repeatedly, for
r:
66
T;=
r:
at the question mark and the dot'ted arrows so as to make the diagram
commutative and the bottom row exact. To do so, take in A'@ B
the subgroup N of all elements (- ua,xu) for agA. At " ? " in the
diagram put the quotient group (A'@ B)/N, and write elements of this
quotient group as cosets (a', b) N. Then the equations x'al= (a', 0 ) N,
u'[(al,b) N ]=ub and pb= (0,b) N define m s which satisfy the
required conditions. That the E' so constructed is nique may be proved
directly or deduced from the following "univers~l" property of E'.
Lemma 1.5. Under the hy$otheses of Lemwa 1.4, any morpkism
I;= (al, p,, 7,) : E +El of extensions with a, = ci c@nbe writtelz ulziquely
El.
More briefly, T; can be "factored througlz" E+aE.
67
2. Addition of Extensions
p':
Hence Ext (C,A) is a covariant functor of A. The fact that it is a bifunctor (of A and C) is demonstrated by the following result:
Lemma 1.6. For a, y, and E as in Lemmas 1.2 and 1.4 there is a
congr~enceof extensions a ( E y)r (aE ) y.
Proof. By the definitions of E y and a E there are morphisms
&,Y)
E y -(1E
G a E(a B
1)
a,
r:
2. Addition of Extensions
The direct sum A @ C of two modules may be regarded as a covariant
bifunctor of A and C, since there is for any two homomorphisms
i.-II/
s*
68
J..
iae.
AftA'@C'+C'
i.
V=VA:A$A+A,
V(%,a2)=%+a2;
(2.47
the reader should verify that (f+g)c is still fc+gc under this formula.
Given two extensions E, = (xi, a,): A , B, -a C, for i = 1 , 2 , we
define their direct sum to be the extension
69
2. Addition of Extensions
The class of the sfilit extension A -A@ C+C is the zero element of this
groufi, while the inverse of any E is the extension (-1,)E. For homommfihisms a : A+A1 and y : C'+C one hus
a(El+E2)=aEl+aE2,
(El+E2)y=Ely+E,y,
(al+ad E=a1E+a2E,
E(y1+y2)= Eyl+Ey2.
(2.5)
(2.6)
The composition (2.4) is known as the Baer sum; the rules (2.5)
state that the maps a, : Ext (C. A ) +Ext (C,A') and y* : Ext (C,A ) +
Ext (C', A ) are group homomorphisms.
We give two different proofs. The first is "computational"; it is
like the calculation made in 3 I to show that Extz(Z,, A ) is the group
A / m A.
Take any extension E = (x,a) of A by C, with a : B +C. To each c
in C choose a representative u(c); that is, an element u(c)E B with
au(c)=c. For each ~ E Rthe
, exactness of E gives r u ( c ) - u ( r c ) ~x A ;
similarly, c, d~C have u (c+ 4 - u (c)- u ( d )xA.
~ Hence there are elements f (c,d) and g (r,c) E A with
. (4 u (c+ a) , c, d C ,
(2.7a)
ru(c)=xg(r,c)+u(rc), ~ E RC, E C .
(2.7b)
Call the pair of functions (f,g) a factor system for E. Let FR(C,A ) denote,
during this proof, the set of all pairs (f,g) of functions f on C x C to A
and g on R x C to A . Each factor system is an element of FR(C,A),
and FR is a group under termwise addition; that is, with (fl+ f,) (c,d)=
f1(c,d)+fs(cJ).
The factor system for E is not unique. For any different choice
of representatives ul(c) we must have ul(c)= x h (c)+ u (c) for some
function h on C to A. One calculates that
21 (c)+>d)=
. =xf
(dch)(c,a)=h(c)+h(d)-h(c+d),
(dRh)(r,c)=rh(c)-h(rc).
The factor system f', g' for representatives u' then has the form (f',g') =
( f ,g) (dc h, dR h) Conversely, any such function h can be used to change
representatives in an extension. Thus, if we denote by SR(C,A) the
subgroup of all those pairs of functions in FR(C,A) of the form (dch,
dR h), the factor system ( f , g) of E is uniquely defined modulo SR(C,A ) .
70
,
X
. .
+I
The first follows by calculating a factor system for El@ E,, and thence
for El+ E,. The second follows by observing (draw the diagram!) that
(-1,) E is obtained from E just by changing the si'sjd%f the map
x : A n B and hence by changing the signs off a n d g i n t h ~ k t osystem.
r
Finally, the split extension E, has (0,O) as one of its factor systems,
hence is the zero of this addition.
It is also possible (see the exercises) to characterize directly those
pairs of functions (f,g) which can occur as factor systems for an extension, and hence to show that ExtR(C,A) is an abelian group without
using the Baer sum at all.
(2.5) is easy; FR(C,A)/SR(C,A)is a bifunctor, and o
is a n ural homomorphism. The proof of (2.6) is similar.
We now turn to the second (conceptual) proof of the theorem. For
the direct sum (2.3) of two extensions Ed the congruences
"r
Now we can prove the assertion (2.5) of the theorem by the string of
congruences
a(El+E2)=aV(El@E,) A=V(u@a) (El@ E,)A
rV(aEl@aE,)A=aEl+aE2;
the second half is similar. The proof of (2.6) is parallel to this once we
know that
AE=(E@E)A, EV=V(E@E).
(2.11)
2. Addition of Extensions
71
To show that the split extension E, acts as the zero for the Baer sum,
t A ) there is a commutative diagram
first observe that for any E ~ E x(C,
E : o+A-B-%C+O
.1Y
E,: o + A + A @ C + C + O ,
where v is the map v b= (0, o b)=bob. This diagram asserts that the
split extension E, can be written as the composite E,=O, E, with 0, :
A+A the zero homomorphism. Now the distributive law gives E+ E,=
l,E+OAE=(l,+OA)
E=I,E=E.
A similar argument shows that
(-1,)
E acts as the additive inverse of E under the Baer sum. Our
second proof of the theorem is complete.
The second distributive law (2.6) contained in this theorem may be
expressed as follows. For each a : A+ A' let a, : Ext (C,A ) +Ext (C,A')
be the induced homomorphism, and similarly set y*E=Ey. Then
( y f yz) E=yf E+ y; E, so (2.6) may now be written
72
+. +
where v,:
fl
Exercises
3. Do the analogue of Ex. 1 for any ring (identities on factor systems consisting
of two functions f and g).
73
where E' is the external direct sum with injection tl, projection n,.
Fill in the dotted arrow with the map b+(&b, a b ) =sldb+ r,ab. The
resulting diagram is commutative, hence yields a morphism E +El.
According to Lemma 1.4, E' = a E. Since E' splits, so does a E.
Conversely, assume that a E splits. The diagram
used to define la^ yields a map q@:B+G which extends a. The lemma
is proved.
The assignment to each a : A+G of its obstruction a E is, by (2.6),
a group homomorphism
E*: Hom, (A, G)
(C, G)
74
a,
(3.3)
75
i ix
E,: o-+K+F-+B+O
with exact rows El, E , and hence E p E , x . This yields a commutative
diagram
Hom ( K ,G) Et Ext ( B ,G)
b'
E': O+A
.... ? ....
>
>
G=G
rp
+
D +O
76
A diagram chase then shows the middle column exact. This middle
column provides an element of Ext (G, B) mapping on the last column
S E Ext (G, C), as desired.
Note that the diagram above is symmetric: Given exactness of the
top row and the right column, exactness of the middle row is equivalent
to exactness of the middle column. The case of Thm.3.2 asserts that
the diagram can be filled out so that the middle row is exact, while the
case of Thm. 3.4 asserts that the diagram can be filled out so the middle
column is exact. The same fact can be stated in subgroup language as
follows:
Corollary 3.8. Given abdian grou#s D a& A< B a d a monomor#hism
p : BIA -D, there exists arc abelian grou# M > B and an extension of p
to an isomor#hism MIA GD.
This amounts to the construction of a group M from a given subgroup B and an "overlapping" quotient group D.
Exercises
( 1 n e & & a Ext on the right.) Let R = K [x, y ] be the polynomial ring
in two indeterminates x and y with coefficients in a field K and (x,y) the ideal
of all polynomials with constant term 0. The quotient module R/(x,y) is isomorphic
to K, where K is regarded as an R-module with xk= O= y k, for all k K, and
E: (x,y) H R a K is an exact sequence of R-modules. Show that ExtR(R, G) +
ExtR((x,y),G) is not an epimorphism for all G, by choosing an extension on
the right in which (x,y) is represented as the quotient of a free module in two
generators.
1.
2. Show similarly that the sequence of Thm. 3.4 cannot be completed with
a zero on the right.
7. For p a prime, P the additive group of all rational numbers of the form
m/p6, m, eZ, and ~ ( f ' ) the additive group of p-adic integers, prove
ExtZ(P,Z) G z(~)/z
[EILENBERG-MAC LANE 1942, Appendix B] .
77
i the ppo&tion,
T,: C,nK,+D,,
S,: D,+~~c,+H,,(K)
(4.3)
are exact, the second by the definition of I& as C,/aK, . The coboundary
in the complex Hom ( K ,G) is 6= fa*, where a*: Hom (K,-,, G) +
Hom (K,, G) is induced by a. This complex appears as the middle row
in the diagram
0 -+
.1
This diagram is commutative up to a sign (that involved in the definition 6= fa*). In the diagram the fundamental exact sequence (Thm.3.2)
for Hom and Ext appears several times. The top row is the exact sequence for S,,, the bottom that for S,with the right-hand zero
standing for Ext (C,-,, G) which vanishes because Cn-,< K,-, is free.
The columns are (parts of) the exact sequences for T,-,, T,,and Tn+,;
the zero at the middle top is Ext (D,, G), zero because D, is free.
78
The cohomology of the middle row is Ker 6/Im 6. Since j* is a monomorphism and i* an epimorphism, it is Ker (al*i*)/Im(j*a'*), and is
mapped by i* onto Ker a'*, isomorphic to Hom (H,, G) by exactness
of the top row. The combined map is a. Its kernel is Im j*/Im (j*a' *) ;
as j* is a monomorphism, this is Ext (H,-,, G), by exactness of the
bottom row. This proves (4.1) exact, with /I described in "switchback"
notation as j* (S,*_,)-l, hence natural.
To split the sequence (4.1), observe that D n sBn-,< Kn-, is free, so
the sequence T, of (4.3) splits by a homomorphism p: Dn+Kn with
j p = ID. Then p* j* = 1, so S2-1p* is a left inverse for /I= j* (S:-,)-l,
as desired. This left inverse depends on the choice of the maps p splitting
T,. Such a choice cannot be made uniformly for all free complexes K,
hence p* is not natural in K (but is natural in G for K fixed).
This proof uses several times the fact that subgroups of free abelian
groups are free.
analogous statement holds for free modules over a
principal ideal domal ; hence the theorem holds for K a complex of
free modules over such a domain D (and G a D-module). The most
useful case is that for vector spaces over a field. Here Thm.4.1 gives
-?
IT
79
the products has at most one non-zero term, and (4.5) becomes (4.1).
In general, if we shift the indices of L by n (and change the sign of the
boundaries in L by (-1)") we shift H, (Hom (K,L)) to H, (Hom (K, L));
hence it suffices to prove the theorem when n=0. Now a O-cycle of
Hom (K,L) is by (4.4) just a chain transformation f : K - t L ; as such
it induces for each dimension a homomorphism (f*), : Hp(K) +H# (L).
The family of these homomorphisms is an element f,={(f*),)~
17 Hom (H*(K),H*(L)). Any f' homotopic to f induces the same homo-
&J
<
E: O . + H ~ ~ ( D , L ) ~ H ~ ~ ( K , L ) - % H O ~ ( C , L ) - +
where the zero on the right stands for Ext (D,L), which vanishes because
D, B,-,< K,-, is a subgroup of a free group, hence free. The exact
80
a,,
4%
ITHom(Cp, Hp+n(L))2
P
l~
II
Hom(Dp+l>Hp+n(L)).
P
81
Substituting these values in (4.7) gives the desired exact sequence (4.5)
of Thm.4.3. The homomorphism a thereby is the composite
Ho(Hom (K. L))5 Ho(Hom (C,L)) -% 17 Hom (Cp, HP(L))
b
here the last arrow stands for the additive relation which is the converse
(or, the "inverse") of the first monomorphism of (4.8). This composite a
assigns to each f : K + L the family of induced maps of homology
classes, so is the map already described. The homomorphism /? is the
composite
\-
82
Exercises
1. Give a direct proof of Cor. 4.2.
2. Show that Thms.4.1 and 4.3 hold for complexes of R-modules if the hypothesis that K,, is free is replaced by the assumption that C,,(K) and KJC,,(K)
5. Composition of Extensions
Return now to the study of the extensions of modules. Two short
exact sequences
where the second starts, may be sfiliced
the first ending at the
B,-tK+B,
to give a longer exact
together by the
sequence
EoE': o+A+B,~B,+c+-o
(5.1)
called the Yoneda coqbosite of E and E'. Conversely, any exact sequence
A w Bl-t B,+ C has such a factorization, with K = Ker (B, +C) =
Im (Bl +B,).
Longer exact sequences work similarly. Consider
IJ
E: A H
Bl
.1=
+ K,
E;: K @ M - B o @ M - + C
E':
.1=
Bo
11
+C
and the definitions of composites show that El= E n and E'= nE;;
in the top row, the composite
is not the same as (5.1) ; in other words (En)o E;+Eo (nE;), and the
associative law fails.
5. Composition of Extensions
83
whenever the composites involved are all defined; that is, for E" ending
at some module K, for /3: L+K for some L, and E' starting at L. Let us
then define congruence as the weakest reflexive, symmetric, and transitive relation including (5.2) and the previous congruencefor short exact
sequences. This definition can be restated as follows. Write any n-fold
exact sequence S as the composite of n exact sequences Ei in the form
We say that F starts with the homomorphism a and ends with y. Now
aE was defined by just such a diagram E+aE, so our definition of
a S above yields a morphism S +a S starting with a and ending with 1,
as well as S y +S starting with 1 and ending with y. More generally
we have, as in Prop. 1.8,
6*
84
r:
r:
rulzning alternately to the left and to the right, all startifig with I A and
ending with 1c.
This proposition states that S r S' is the weakest reflexive, symmetric, and transitive relation such that
S +S' with ends 1 implies
S= S f .
Proof. First suppose S= S f . In the elementary congruence (5.2),
the definition of E"/? yields a morphism E"B+EU, while the definition
of BE' yields a morphism E1+/?E' of exact sequences. Placing these
morphisms side by side yields a diagram
r:
5. Composition of Extensions
85
and the associative law Q (a z) = (Qa)z all hold when the addition and
com+osition involved are defined.
86
In brief, ExtR is like a ring, except that the sum a + z and the
product a z are not always defined.
This theorem clearly includes the previous Thm.2.1 on Ext l =Ext,
and the proof is exactly like the "conceptual" proof of that theorem.
That proof rested on certain rules for "direct sums". In the present
case these rules (and their prior counterparts) are
(a $a') (a63 z') =at. $ a'z',
(5.6)
V(El @ El)
87
6. Resolutions
n>0,
P projective.
(5.10)
(6.1
Theorem 6.1. (Comparison Theorem.) If y: C+C1 is a homomorphism of modules, while E: X-tC is a projective com,plex over C and
E' : X1-t C' is a resolution of C', then there is a chain transformationf : X +X'
with &'/=YE and any two such chain transformations are chain homotopic.
We say that such an f lifts y.
The proof uses only categorical properties of projectives and of
exactness. Since X, is projective and E' an epirnorphism, ye: X0+C1
can be lifted to f,: X,-tX; with ~'f,=ys. By induction it then suffices
88
Jh-1
..., fo
Ltm-s
a;,
&-I ,
x:,--tx;-,-x,-,+
Jfe
JY
... -+x;-+cl
(f,-,
6. Resolutions
89
t: Extn(C,A)sHn(X,A),
E:
X + C a firojective
n=O,l, ...,
(6.3)
(6.4)
This isomor#hism 5 is natzlral in A. I t is also natural in C in the following
sense: I f y : C1-+C,E' : X1-+C' is a #rejective resolution of C', and f : X1+X
lifts y, then
cy*= f* 5: Ext n (C,A) +Hn (X', A).
(6.5)
Proof. First observe that 5 is well defined. Since g,a= 0, g, is an
n-cocycle, as stated. Replace g by any other chain transformation g'
lifting lc, as in (6.2). By Thm.6.1, there is a chain homotopy s such
that gL-g,= as,+~,-~t3. But s,: Xn+O, so s,=O, gA-g,=s,-,a=
( - - 1 ) " 6 ~ , - ~ this
,
by the definition (11.3.1) of the coboundary in
Hom (X,A). This states that the cocycles g,' and g, are cohomologous, so
cls gL= cls g,. Next replace S by any congruent exact sequence S'.
According to the description of the congruence relation S = S' given
in Prop. 5.2, it will suffice to consider the case when there is a morphism
S -+S' starting and ending with 1. In this case any g: X -+S yields
r g: X -+S' with the same cocycle g, = &
I( ;
hence cls g, is well defined
as a function of clsS. Thus 5 is defined; its naturality properties as
asserted follow at once, using suitable compositions of chain transformations.
Rather than proving directly that 5 is an isomorphism we construct
a.
its inverse. Given a resolution X, factor a: X,-+X,-,
as X,+ BX,
AX,-,, with x the injection; this yields an n-fold exact sequence
r:
g0
composite hS, of h with the n-fold exact sequence S,; this fills in the
bottom row of this diagram. Define 7: H" (X, A) +Ext n (C,A) by setting
By the distributive law for the composition in Ext, the right hand
side is additive in h. Hence to show 7 well defined it suffices to show
that 7 (cls h a') vanishes when ha' is the coboundary of some cochain
8' means that L=f kx
k: Xu-,+A. But ha'=dk=(-I)"ka=(-1)"kx
and hence that h S, = fkx S, , where x S, r 0 by Prop. 1.7. Hence 17 is
well defined and is a homomorphism. Comparison of the diagrams (6.2)
and (6.6) now shows that 7 =C-l.
This theorem states that the groups ExtU(C,A)may be computed
from any projective resolution E : X+C; in particular, (6.5) shows how
to compute induced homomorphisms y*: Extn(C, A) +ExtU (C', A) from
resolutions.
Alternatively, many authors define the functor Ext" without using
long exact sequences, setting Ext n (C,A)= H" (X, A) = HU(Hom(X, A)).
This gives a covariant functor of A, while for y: C'+C the induced
maps y*: Ext n (C,A) +Ext n (C', A) are defined by lifting y to a comparison X1+X.
Another consequence is a "canonical form" for sequences under
congruence:
Corollary 6.5. If SEE EX^" (C, A) with n > l , then t k e i s a T s S
of the form T: O+A +B,-,+B,-,+...
+Bo+C+O in which the
modules B,-,, ..., Bo are free.
Proof. Take T = h Sn(C, X) for a suitable h:
-+A, and X any
free resolution of C.
Corollary 6.6. For abelian grozlps A and C, Ext; (C,A) = O if %>I.
Proof. Write C=F/R for F free abelian. Since the subgroup R of
is a free resolution
the free abelian group F is free, O+R+F+C+O
which vanishes (with its cohomology) in dimensions above 1.
Consider now the effect of a ring homomorphism e: R1+R (with
e 1 = i ) . Any left R-module A becomes a left R'-module when the
operators are defined by rla= ($)a; we say that A has been pulled
back along e to become the R'-module ,A. Any R-module homomorphism
a : C + A is also an R'-module homomorphism ,C-t, A. By the same
token, any long exact sequence S of R-modules pulls back to a long
exact sequence S of R'-modules, and congruent sequences remain
congruent. Hence e#a =a, e# (clsS)= cls ,S define homomorphisms .
e#:
n=o, I ,
...
(6.8)
6. Resolutions
91
These homomorphisms may be calculated from projective resolutions E : X+C and E': XI-+& by R and R'-modules, respectively.
(Hom, (X,A)) for Hu(X,A).
To exhibit the ring R, write
Theorem 6.7. The change of rings Q#, via the isommphism
Thm.6.4, is given by the composite map
O ~ (,X,
~ ,
5 of
Now read off the maps: The isomorphism carries clsS to clsg,, Q#
regards gn as a R'-module homomorphism, f* maps clsg, to cls(gufu),
which is exactly c(cls,S) because gf lifts 1. Hence the result, which will
be of use in the treatment of products.
Exercises
i . If E : Y+ C is a projective complex over C and e': X +C a resolution of C,
construct natural homomorphisms
5 : E X ~ ~ ( C , A ) + H ~ ( Y , Aq) ,: ~ ( ~ , ~ ) + ~ ~ t ; ( ~ , ~ ) .
2. (Calculation of Yoneda product by resolutions.) If X +C and Y+A are
projective resolutions, g~ Homu(X, A) and h Homm(Y,D) are cocycles, write g
as goa' for go: BXn+A, lift go to f as in the diagram
show hf an (m+ n)-cocycle of Homm+"(X,D), and prove that the Yoneda product
q (cls h) o q (clsg) is q (cls hf).
t : B+A1, show
92
r:
r:
7. Injective Modules
with horizontal row exact can be filled in (on the dotted arrow) so as
to be commutative. The characterization of projective modules in
Thm. 1.6.3 and Thm. 3.5 dualizes at once to give
<
7. Injective Modules
93
(Si, yi) of these pairs there is an upper bound (T, z) with (Si, yi) 5 (T, z),
for take T to be the union of the submodules Sdwith z defined for each
Hence, by Zorn's lemma, there is a maximal
t by zt=y,t when
such pair (S,, y,). We need only prove S,= B. If not, there is an element b~ B not in S,; take the submodule U of B generated by b and S,.
Then r+rb+ S, is an epimorphism R+ U/S,, the kernel of the epimorphism is a left ideal L in R, and R/L=U/S,. Since the sequence
S,- U-9 U/Sw is exact, so is the sequence
Hom(U,J)- + H o m ( S , , ~ + E x t (U/S,, J).
(7.4
94
C E C,
(7.4)
Homz (B,D)
HomZ(A,D)
8. Injective Resolutions
95
in other words, that f (a, rs) =f (sa, r) always. In particular, f (a, s)=
f (sa, 1) so the function f is determined by g(a)=f (a,1). Clearly g: A+G.
Now qAand its inverse are defined by
The maps r;l~ and qT1 are clearly homomorphisms and natural (in A
and in G ) .
This idea of regarding a function f (a,r) of two variables as a function
of a whose values are functions of r will reappear more formally later
(V.3), and this lemma will turn out to be a special case of a more general
natural isomorphism, called "adjoint associativity". Injective modules
will be studied further in $11.
Exercises
I . If R is an integral domain, show that the field of quotients of R is a divisible
R-module. If, in addition, R is a principal ideal domain, show that this field is
injective as an R-module.
2. If A is a left R-module and L a left ideal in R, each a E A defines an R-module
homomorphism fa: L + A by fa(l)= la. Prove that A is injective if and only if,
for all L every f : L -+A is fa for some a.
8. Injective Resolutions
96
(8.2)
f,r.
(Hom (C,Y)) .
(8.5)
The rest of the proof, like the definition, is dual to the proof of Thm. 6.4.
We can summarize the theorems of $6 and $8 in the scheme
Hn(Hom (RespC,A)) r E x t n (C, A) =Hn(Hom (C, R ~ s ~ A ) ) ,
where Res,C denotes an arbitrary projective resolution of C, ResJA
an arbitrary injective coresolution of A. A symmetric formula Ext" (C, A )
~HU(Hom(RespC
ResJA))
,
can be established (Ex.V.9.3).
Exercises
Carry out the construction of g in (8.4) and of the inverse of
2. State and prove the dual of Lemma 6.2.
3. For direct sums and products establish the isomorphisms
1.
~ x t ("C c t ,A )
n Extn(Ct,A ) ,
Extn(C,n A t ) C:
5.
nExtn(C, At)
E*:
G) ,
E, :
97
r=
are exact. These sequences start at the left with 0 +Hom (C,G)= Ext 0 (C,G)
and with O+Hom (G,A), res#ectively, and continue to the right for all
n=O, 1 , 2, ... . The maps in th6se sequences are defined for arguments
e E Ext" (C,G), o e Ext" (B, G), zc Ext" (A,G), ... by composition with
X , a, E as follows:
.It
.It
,-I,,,
.It
H"+' (X,A),
7
(9.5)
98
om (c,Y))- om ( B , Y ) ) ~ H * ( (A,Y))
H ~ ~ d- ~ " + l ( ~ (c.Y))
om
tf
Ext" (C,G)
Ext" (B,G)
Ext" (A,G)
ExtWf1(C, G)
The desired proof that the bottom row is exact requires now only
the commutativity of the diagram. Note that the connecting homomorphism 6, is defined by switchback as 8, =cis o*-l6x*-lcls-l, and no
trouble with signs occurs. Given this definition, the proof that commutativity holds is now like that given above for the dual case - though
Ext
99
since the proof manipulates not only arrows, but also elements, we
cannot say that the proof is exactly dual. Thm.9.1, though formulated
in the language of exact sequences, can also be regarded as a statement
about annihilators in the "pseudo-ring" ExtR of Thm. 5.3. Indeed, if
E = (x,a) is a short exact sequence of R-modules, then
and these equations indicate the whole left and right annihilator in
Ext, of each of x , E, and a, as follows. The right annihilator of x consists of multiples of E; whenever e ~ E x t ,is such that the composite
x~ is defined and is 0,then either Q = O or Q = E z for a suitable z ~ E x t , .
Similarly, e x = 0 implies e= z a for some z, etc. In other words, the left
annihilator of x is the principal left ideal (ExtR)a.
Exercises
I . Given the usual short exact sequence E of modules and given projective
resolutions E ' : X - t A and E": Z+C of the end modules A and C, construct a
projective resolution E : Y-t B of the middle module B and chain transformations
f: X + Y , g: Y+Z lifting x and a, respectively, such that X-Y +Z is an exact
sequence of complexes. (Hint: for each n, take Yn= Xn@Znand define E and i3
so that ( Y , E ) is a complex.)
2. Use the result of Ex. 1 to give a proof of the exactness of (9.1) by projective resolutions.
3. Deduce Prop. 3.7 from Thm. 9.1 and Cor. 6.6.
4. For A a finite abelian group, Q the additive group of rational numbers,
prove ExtZ( A ,2)GE HornZ(A, Q/Z).
Theorem 10.1. For each n = 0, 1 , .. ., let there be given a contravariant functor Ex n (A) of the module A, taking abelian grozcps as values,
and for each n and each short exact seqzcence E : AH B +C let there be given
a homomorphism E": Ex n ( A )+Exn+l (C) which is natural for mor+hisms
E-tE' of short exact sequences. Suppose that there is a fixed moduZe G
such that
Ex0 (A)= Hom (A,G )
for all A ,
(10.1)
r:
Exn (F)= 0
(10.2)
100
b"
Ext n ( A ,G)
1-
Extn+' (C,G)
(10.4)
6bEx
(C)
0.
(C,G ),
which is a special case of (10.4). To show that I& is natural for any
y: C +Cf, pick an exact Ec,: Kf F'+Cf. The comparison theorem
lifts y to /I: F-tF', which induces a morphism
(a, /I, y): Ec+ Ec,.
Since both connecting homomorphisms El and E* are natural with
it follows that y*p'&E&=y* E$= E&*
respect to such morphisms
= &EL a*= & * E:, . But E$,is an epimorphism, so
r=
r,
101
(because E* is natural),
(definition of tp) ,
(because El is natural).
and Ex*(C)rEx*-l (K). Using the similar sequence for Ext*, we define
tpu by
tp; = E&;-'(%-~)-~ : Exu (C) r E x t u(C, G)
and establish naturality, independence of the choice of F, and the
commutativity (10.4) much as in the case n =1.
There is a dual characterization for Ext (C,A) as a functor of A,
using the second exact sequence (9.2).
for aU A,
(10.5)
(10.6)
(10.7)
Proof. Observe first that Extu does have the property (10.6), for
which
an injective module J has the injective coresolution O+J+J+O,
vanishes in all dimensions above 0. Conversely, the proof that these
three properties characterize the Extu(G,A) as functors of A is dual
to the proof we have just given.
Exercises
(S. SCHANUIL.)
Given two short exact sequences K H P+C and Kt* P'+C
with P and P' projective, K < P, K'< P', and the same end module C : Construct
an isomorphism P @ P ' a P @ P' which maps K @ P' isomorphically on P @ Kt.
2. Call two modules C and C' projectively equivalent if there are projective
modules Q and Q' and an isomorphism C e Q ' e r C'@ Q. Let S : K- P,-,+
+P,+C be an n-fold exact sequence with all 4 projective. Using Ex.1, show
that the projective equivalence class of K depends only on that of C and not on
the choice of S.
1.
102
103
Chapter four
Cohomology of Groups
The cohomology of a group 17 provides our first example of the
functors Extz(C,A) - with R the group ring and C=Z. These cohomology groups may be defined directly in terms of a standard "bar
resolution". In low dimensions they arise in problems of group extensions
by I7;in all dimensions they have a topological interpretation (51I).
104
(a,
2. Crossed Homomorphisms
105
One proves that this is a group with the "identity" element 0 = (0,i)
and inverse - (a, x)= (- x-la, x-1) and that there is a short exact
sequence
o+A-%AxJI-%17+1,
(1.5)
where x is the homomorphism given by x a = (a, I), a is a(a, x) =x,
and 1 denotes the trivial multiplicative group. Also a has a right inverse
v defined by V X = (0,
X) for all x; it is a homomorphism of the multiplicative group 17 to the additive group A x p .
Exercises
I.A
2. Crossed Homomorphisms
If A is a 17-module, a crossed homorfihism of 17to A is a function f
on 17to A such that
f(xy)=xf(y)+f(x),
YE^.
(2.1)
106
2. Crossed Homomorphisms
107
Proof. When E (s) = O the identity (2.3) for g becomes g (rs) =rg (s),
so g restricted to the kernel of E is a module homomorphism, as stated.
Conversely, any module homomorphism h: I(IT)+A, when composed
with the special crossed homomorphism p r = r- E (r)1, yields a crossed
homomorphism hp on Z(IT) whose restriction to I(17) is exactly h.
Finally, the principal homomorphisms behave as stated.
For 17fixed, Zi(17, A) and H i are covariant functors of A ; for each
module homomorphism u: A+ B, (a,f) (x) is defined as a [f (x)]. For a
fixed abelian group A with the trivial 17-module structure one can
make Zf, and Hf, contravariant functors of 17; for a group homomorphism 5: n+IT and a crossed homomorphism f on IT define the induced map [* : Zf,(n'
A)
,+Zi (17,A) by ([* f) (x)=f ([x). This will not
do when A is a non-trivial 17- or IT-module. However, if [: n+IT and
A' is a n'-module via p': IT+Aut A', then A' is also a 17-module via
qf[: 17+Aut A', and we may define induced homomorphisms
[*:
[*:
108
(r,
3. Group Extensions
A grot@ extension is a short exact sequence
E: o+-G%BB~~-+~
(3.1)
Suppose G= A abelian; then 8 ( A )= 1, so that 8 induces a homomorphism y : 17-tAutA with p,a=6. Thus y is defined by
3. Group Extensions
109
r:
r=
-+n-+I
L -1~
E: O + A + B
E': o-+At-+B'-+IT'+l
r:
r:
r=
110
With these equations, the sum of any two elements of the form a + i u
can be put in the same form. By associativity, u + m u = ( m + I ) u =
mu+ u, so u+ a,.= a,+ u. Therefore a,= ta,, so a, is "invariant" under t.
This element a, is not unique; if u'= %+ u is a different representative
for t in B, then, by (3.5) and induction on m,
(3.6)
This is 1-1; given any invariant a,, take B to be all symbols a+ i u
with 0si< nz and addition given by (3.5). The invariance of a, proves
this addition associative, and B is an extension of A by C, with the
given operators. In particular, if A has trivial operators ( t a = a always),
the expression on the right of (3.6)is the group AIm A - in agreement
with the result already found in the case of abelian extensions in Prop.
111.1.1. In this case, all extensions of A by C, are abelian.
be the free multiplicative abelian group on
Again, let 17=C,xC,
two generators t, and t,. In any extension by 17,take representatives
ui of ti, i = 1 , 2. There is then a constant a, in A with u,+ y = a,+ y+u, ,
all elements of the extension can be written uniquely as a+
y+ m,u,
with integral coefficients m, and m,, and the addition in B is determined
by the addition in A and the rules
% + a = h a + ~ , , u,+a=t,a+u,,
u,+u,=a,+u,+u,.
This addition is always associative and makes the collection of elements
a+ m,ul+m u a group. If the representatives u, and u, are replaced
by any other y = % + u , , u;=a,+u,, for 6, a,cA, the constant a,
is replaced by a,+a,-ha,--%+t,a,.
Hence, if S is the subgroup of
A generated by all sums a,-- t,a,-- %+ t,a,, we have a 1-1 correspondence,
Opext (C,X C, , A , q)* A / S .
(3.7)
,,
Exercises
I . Describe Opext (C,
x C, x C,, A,
v).
v).
4. Factor Sets
I11
4. Factor Sets
The calculations just made suggest that Opext ( n , A, cp), like Ext,
is a group. This group structure can be described by means of certain
factor sets.
Let E be an extension (3.1) in Opext (l7,A , c p ) : For convenience,
identify each a with x a . To each x in lI choose a "representative"
u ( x ) in B; that is, an element u ( x ) with o u ( x ) = x. In particular, choose
u ( l ) =O. Now each coset of A in B contains exactly one u ( x ) , and the
elements of B can be represented uniquely as a+ u ( x ) for a c A, ~ 1 7 .
We write the operators as cp(x)a= x a : Then (3.2) for b=u ( x ) becomes
On the other hand, the sum u ( x ) u ( y ) must lie in the same coset as
.u( x y ) , so there are unique elements f ( x , y ) E A such that always
(x)+u(y)=f (x,Y)+~(xY).
(4.4
r ~ ( ~ ) + u ( ~ ) l + u ( z ) = f ( ~ , y )Z )++ Uf ((X~Y ~ 4, ,
(~)+[u(Y)+~(z)I=xf(r>z)+f(~,~z)+~(xYz).
Their equality (associative law!) gives
xf(y,z)+f(x,yz)=f(x,y)+f(xy,z),
x,y,zcn.
(4.5)
The factor set f for an extension depends on a choice of representatives; if u f ( x ) is a second set of representatives with u l ( l )= 0 , then
u l ( x ) and u ( x ) lie in the same coset, so there is a function g on I7 to A
with g ( I )= 0 such that u l ( x )=g ( x ) u ( x ) . Thus
One verfies that this function dg does satisfy the identity (4.5), with f
replaced by dg there.
112
(4.7)
between the set Opext of all congruence classes of such externions and the
2-dimensional cohomology group. Under this correspondence the semi-direct
product corresponds to the zero element of P,.
Since H: is an abelian group, this correspondence w imposes the
desired group structure on Opext. This group structure can also be
described conceptually via the Baer product, as set forth in the exercises
below.
Proof. Since the factor set of an extension is well defined modulo
the subgroup B:, and since congruent extensions have the same factor
sets, we know that the correspondence w is well defined. The semidirect product A x & clearly has the trivial function f ( x , y) =O as one
of its factor sets. If two extensions yield factor sets whose difference
is some function Bg(x,y), then a change of representatives in one extension will make the factor sets equal and the extensions congruent.
Therefore (4.7) is a 1-1 correspondence of Opext with part of H a . Finally,
given any f satisfying (4.5) and (4.3), one may define a group B to consist of pairs (a,x) with a sum given as in (4.4) by
( a , ~ ) + ( % > ~ ) = ( a + x % + f ( x > ~ ) # x ~ ) a,beA.
>
The module rules and the condition (4.5) show that this composition is associative; it clearly yields an extension with representatives
u ( x )= (0,x ) and factor set f . This completes the proof of the theorem.
If A is abelian, a central group extension of A by 17 is an extension
E as in (3.1) in which x A is in the center of B. In other words, a central
extension is one with operators p, = 1. This theorem thus includes the
fact that the set of congruence classes of central extensions of A by 17
4. Factor Sets
113
then aa=O, and the coboundary of g is the function given by the formula (4.6). Thus Ha, (17,A) is Ha(Homz(,, (X,A)). We get the analogous
result for H i if we take X, to be Z(17) and set a[x] =x- 1~Z(17).
This complex also defines a 0-dimensional cohomology group as
(17,A)= HO(Homz(,, (X,A)). A 0-dimensional cochain is a module
homomorphism f : Z ( m +A ; it is determined by its value f (4) .=a A.
I t is a cocycle if -(dl) [x] = f a [x] =f (x-I ) = xu- a is zero. Hence
the 0-cocycles correspond to the elements acA invariant under 17
(xa=a for all x):
Exercises
The Baer sum, introduced for extensions of modules in Chap.111, can also be
applied to group extensions, as indicated in the following sequence of exercises.
I . Prove: If E is an extension of G by L!and y : Il'+L!, there exists an extension
E' of G by 17' and a morphism I l = ( I C , B, y ) : E'+E. The pair (I:E') is unique up
to a congruence of E'. If G is abelian and has operators q : n+AutG, then E'
has operators q y . Define E y = E'.
Mac Lane, Homology
114
3. For E EOpext (L7,A , q), q': lI+AutA', and a: A+A' a L7-module homomorphism, prove that there exists an extension E1Opext(L7,A', q') and a morphism
O = (a,8,i n ) : E +E', unique up to a congruence of E'. Define a E to be E'.
4. Under the hypotheses of Ex. 3, prove that for E l Opext (q,
A', qi) each
morphism (a,, yl): E -t El with al= a and q;yl = q' can be factored uniquely
through O.
a,
7. Show that Opext (lT,A , q) is an abelian group under the Baer sum defined
by El+ E,= 6 (El% E,) An, and show that this composition agrees with that
given by factor sets.
The boundary formulas (4.8) and (4.9) for the complex X of the
last section can be generalized to higher dimensions. Specifically, for
any group 17 we construct a certain chain complex of 17-modules
B,(Z(n)). Take B , to be the free 17-module with generators [x,] . . . 1 x,]
all %-tuplesof elements x1+ 4 , .. . , x,+ 4 of 17. Operation on a generator
with an x d l yields an element x [x, 1 . . . I x,] in B , , so B , may be described
as the free abelian group generated by all x [xll .. . I x,]. To give a meaning
to every symbol [xlJ.. . x,], set
[xll
...I x,]=o
ifanyone x i = 4 ;
(5.4)
a: B,+B,-,
for %>O by
115
Theorem 5.1. For alzy group I7 the bar resolution B(Z(17)) with
azcgmentatiolz E is a free resolution of the trivial 17-module Z.
Proof. The B, are free modules, by construction, so we must show
that the sequence of solid arrows in (5.4), with zero adjoined on the left,
is exact. We will prove more: That this sequence is a complex of abelian
groups with s as a contracting homotopy. The latter statement means
that
ES-,=I,
~ s , + s - , & = I , ~ s , + s , - , a = ~ , (n>o).
(5.5)
Each of these equations is immediate from the definition; for example,
by (5.3), as,(x [xll . . . I x,]) starts with x [xll ... x,] while the remaining
terms are those of s,-,ax[x1l ... Ix,], each with sign changed; this
proves the last equation of (5.5). Moreover, these equations determine
E and a,,+,: B,,, -+ B, uniquely by recursion on n, for B,+, is generated
as a 17-module by the subgroup s, B,, and the equations (5.5) give
a,,, on this subgroup as a,,,,sn = I - s,-, a,; thus the formula (5.3)
for a can be deduced from (5.5) and (5.2) for s. By the same recursion
argument it follows that E a,= 0 and a,a,+, = 0, for
gives a 2 = O by induction. This can also be proved, directly but laboriously, from the formula (5.3) for a. Either argument shows B ( Z ( Q )
a complex and a resolution of Z, as stated in the theorem.
The same theorem holds for the "non-normalized" bar resolution
B(Z(17)). Here /?, is the free I7-module generated by all the n-tuples
xl @ ... @ x, of elements of I7 (no normalization condition) and E, a,
s are given by the same formulas as for B. Thus B , s / ? , / D , , where D,
is the submodule generated by all xl @ - . . @ x, with one xi = 1. The
symbol @ is used here because this description makes /?, the (n+ 1)fold "tensor product" Z(17) @ . .. @Z(17) of the abelian groups Z(17);
these tensor products are defined in Chap.V and applied to the bar resolution in Chap. IX.
For any I7-module A we define the cohomology groufis of ll with
coefficients A by the formula
'
116
in keeping with the special cases treated in the previous section (where
the subscript g, was used to record explicitly the structure of A as a
IT-module). The cohomology groups H" (IT,A) are thus those of the
cochain complex B (17, A )= Hom, [B(Z(n)),A], where Hornn is short
for
Since B, is a free module with generators [xll .. . I xu]
(no xi =I), an n-cochain f : B, -tA is a IT-module homomorphism which
is uniquely determined by its values on these generators. Therefore the
group B" (IT,A) of n-cochains may be identified with the set of all those
functions f of n arguments xi in IT, with values in A, which satisfy the
"normalization" conditions
The sum of two cochains fl and f , is given by addition of values as
Under this addition the set Bn of all such f is an abelian group. The
coboundary homomorphism 8: P + P - I is defined by
df(x1, ..., %+I)= (- I)"+l[Xlf(xp, . . . , %,+1)+
n
+z
... ,%)I.
(54
(-l)if(~l?...tXiXi+l>...,~,+l)+(-~)n+lf(~l
,
i-1
117
Add the identities (5.8) for all x=x,+, in Il. The last term is independent of x; in the next to the last term, for x, fixed,
2 Gf(~1,..., x,,
%En
x)=-Gg(xl,
By Cor. 5.4, the two outside groups vanish; since P has trivial module
structure, Hl (IT, P)=Hom (17,P). Hence the connecting homomorphism
is the desired isomorphism.
To illustrate the use of resolutions, consider the operation of conjugation by a fixed element t d I . Let 8,: IT-tIT denote the inner automorphism B,x=t-lxt, while, for any IT-module A, a,: A+A is the
118
Observation shows g, a= ag,, so g, is a chain transformation of resolutions which lifts the identity Z+Z. By the comparison theorem for
resolutions, g, is homotopic to the identity, so the induced map on
cohomology is the identity. But this induced map carries any 12-cochain
f into g: f where
while conversely
(5.11)
119
remaining such elements are free generators of B,. The 17-module structure
is given by
Y(YO,
YI, . . . ~ Y ~ ) = ( Y Y o.~. .Y* Y IY~~ )
(5.12)
and the boundary
a: B,+ B,-,
is determined by
%.A=
.
y,
Y4-7-Yl
Xl=Yi
Yt
Exercises
1. Show that fi (2(If))
- the non-normalized bar resolution - with a suitable
augmentation is a free 17-module resolution of 2.
3. For n= 2 in Prop. 5.6, show explicitly that the cohomologous factor sets f
and g; f determine congruent elements of Opext (lT,A,
q).
4. Show that Ext;(rn ( Z , A ) is a contravariant functor on t h e category 9of changes of groups, and prove the isomorphism 8 of Cor. 5.2 natural.
120
0: Ext&,,(Z,A)rHa(l7,A).
(6.1)
Lemma 6.1. As a chain complex of abelialz groufls, (6.2) has a colztrading homoto+y.
Proof. A contracting homotopy s would consist of homomorphisms
s: Z+Z(n), s: Z ( Q +M, and s: M +A of abelian groups such that
E S = I ~/,~ s + s E = I ~ (as+s/?=lM,
~),
and sa=l,. The first condition
is satisfied by setting s 1= 1, and the second by s x= [u (x)] L, where
u (x)E B is a representative of x in B with au (x)= x and u (1)= 0. For
all x and 6, zc (x)+ b- u(x(o;b)) is in the kernel of a, so there are elements h (x, b) E A with
-+
I21
Exercises
Show that a and B as defined for (6.2) are indeed module homomorphisms.
2. Complete the proof of Lemma 6.1, in particular showing that the function h
there introduced satisfies h ( x , b,+ b,) = h ( x (ab,), b,) h ( x , b,) and hence that
s: M +A is well defined.
3. Express the function h in terms of the factor set f.
4. Verify that (6.4) gives a chain transformation as claimed.
1.
(a
Since Ha
A) =Exti (n,(2,A ) , we may calculate the cohomology
of a particular group I7 by using a IT-module resolution of Z suitably
adapted to the structure of the group I7.
Let 17=Cm(t) be the multiplicative cyclic group of order m with
generator t. The group ring r = Z ( C m ( t ) )is the ring of all polynomials
m-1
r are
122
r,
ai=O
+ (am-l+...+aO)tm-l].Thi~meansthat
r-4
N*
A-A-A-A-....,
Theorem 7.1. For a finite cyclic group Cmof order m and generator t
and a Cm-module A , the cohomology groups are
Note that these groups for n>O repeat with period two.
Next we consider free groups.
123
We verify that these formulas hold even when the word e,x or e;lx
is not reduced, and hence that the f so defined is a crossed homomorphism. This completes the proof.
Consider now the exact sequence (2.4) of Z(F)-modules
Theorem 7.3. For a free groyb F, H" (F,A ) =0, for n > l .
Exercises
F free.
2. Without using crossed homomorphisms, prove I(F)a free module.
3. Find a resolution for Z as a trivial module over the free abelian group IZ
on two generators, and calculate the cohomology of IZ.
4. Determine the Yoneda products for the cohomology groups H k (C m ,Z),
showing that
sari: o+zN,r3,rj... + r > z + O
1. Describe H1 (F, A) for
124
6. Let 5 : Cm+Ch be a homomorphism of cyclic groups. For A a trivial 17module in Thm. 7.1, calculate the induced map [* in cohomology.
8. Obstructions to Extensions
The 3-dimensional cohomology groups appear in the study of extensions of a non-abelian group G. We write the composition in G as
addition, even though G is not abelian.
For any element h of G, denote by p (h) or ph the ilzrter automor#hism
phg= h+ g- h given by conjugation with h. The map p : G +Aut G
is a homomorphism of the additive group G to the multiplicative group
Aut G of all automorphisms of G; its image p G is the group In G of
inner automorphisms of G. This image is a normal subgroup of Aut G,
t then always
for if q ~ A uG,
q(~hg)=q(~+g-~)=)7~+qg-'17~=~~h(~g)
and hence
(8.1
The factor group Aut G/InG is called the group of automor#hism classes
or of outer automor#hisms of G; it is the cokernel of p : G+AutG. The
kernel of p is the celzter C of G; it consists of all ccG such that c+ g=
g+ c for all G. The sequence
is therefore exact.
Any group extension
E : o-+G-L~,BL~+I
of G by lI determines, via conjugation in the additive group B, a homomorphism 8: B +Aut G for which 8 (xG)< In G. It hence determines an
induced homomorphism p: lI+AutG/InG. In other words, for each
b E B the automorphism g -+ b+ g- b of G is in the automorphism class
~ ( a b ) We
. say that the extension E has colzjugatiolt class y: thus y
records the fashion in which G appears as a normal subgroup of B.
Conversely, call a pair of groups IT, G together with a homomorphism
y: L!-tAutG/InG an abstract kernel. The general problem of group
extensions is that of constructing all extensions E to a given abstract
kernel (lI,G,y); that is, of constructing all short exact sequences E
with given end groups G and 17 and given conjugation class v. As in
9 3, congruent extensions have the same conjugation class.
A given extension E may be described as follows. Identify each
g~G with x g B.
~ To each x d 7 choose u ( x ) E B with au ( x )= x, choosing
8. Obstructions to Extensions
125
126
8. Obstructions to Extensions
127
in two ways. In the first way, apply (8.5') to the inside terms beginning
y ( y ) f (2, t ) ; upon application of the homomorphism y ( x ) to the result
there are terms y ( x ) f ( y , z) and cp ( x ) f ( y z , t ) to each of which (8.5')
may again be applied. When the terms k in the center are put in front,
the result reads
L=[xk(y,z,t)l+k(x,yz,t)+k(x,y,z)+ U ,
(8.8)
Using (8.5') on each term involving y , the fact that all values of k lie
in the center gives
L = k ( x ~ , z , t ) + k ( x , ~zt)+
,
U
(8.9)
with U as before. But the terms added to U in (8.8) and (8.9) are respectively the positive and the negative terms in 6 k ( x , y, z, t ); hence comparison of (8.8) and (8.9) gives 6k= 0 , q.e. d.
We now investigate the effect of different choices of g, and f in the
construction of an obstruction to a given kernel.
h(x,l)=O=h(lj~)
(8.10)
where the function h has values in C, hence may be viewed as a 2-dimensional normalized cocycle of 17 with values in C. Now the definition
(8.5') states essentially that the obstruction k is the coboundary k=6f.
The obstruction k' of f' is thus k' = 6 (h+ f ) . The values of h lie in the
center, so we may write 6 (h+ f ) = (6h) ( 6 / ); the new obstruction thus
has the asserted form; since in (8.10) h may be chosen arbitrarily in C,
we can indeed replace the obstruction k by any cohomologous cocycle.
128
As the new function f'(x, y) we may then select the expression in brackets.
We write this definition as
This definition has the form f'= (6g)+ f, so that we should have 6 f'=
(8Bg) B f =6 f, modulo troubles with commutativity. If one, in fact,
successively transforms the expression y' (x) f'(y,z) f'(x,y z) g (xyz)
by (8.1 I ) and (8.6) one obtains k (x,y,z) f'(x,y) f'(xy,z)+g(xyz),
which shows that the obstruction k is the same as before.
These results may be summarized as follows.
I
I
9. Realization oi Obstructions
129
P [f'@J~)l-
This states that f ( x ,y)- f'(x, y) lies in the kernel of p, that is, in the center
of G. If h is defined as h (x,y)=- f (x,y) f'(x,y), then (8.5) for f and f'
shows dh=O, hence h a cocycle with f'=h+ f , as desired.
The operations of Ha on Opext may also be defined in invariant
terms, without using factor sets. Represent an element of Ha(17,C),
according to Thm. 4.1, as an extension D of C by 17 with the indicated
operators. Let C x G be the cartesian product of the groups C and G.
Define a "codiagonal" map V : C x G +G by setting V(c, g)= c+ g;
since C is the center of G, this is a homomorphism. The result of operating
with D on an extension E in Opext(D,G,y) may then be written as
V ( D x E ) A,. Exactly as in the case of the Baer sum (Ex.4.7) this does
yield an extension of G by 17 with the operators y ; if we calculate the
factor set for this extension we find that it is given, just as above, by
a map f +h+ f .
9. Realization of Obstructions
We have proved that the obstruction to an extension problem is
an element of H3(17,C). If C=O, the obstruction vanishes, hence the
extension problem has a solution. The result is
130
B ( ~ ) B ( Y[z,) t l = f
( x , y ) + k ( x y , z , t ) + f ( x y , z ) + f (xyz,t)--f ( x y , z t ) - f ( x , ~ )
= f(x>Y)+B(xY)[z, t l - f (%Y)
=luV(%y)I B ( x Y ) [ z . t l .
This proves (9.2).
We claim that each B ( x ) is an automorphism of G. Indeed, (9.2)
proves that B ( x ) B ( ~ - l ) = p [ f ( x , x - ~ ) ] ~ ( 1 ) = p [ f ( x , x - l ) is
] an inner
automorphism. Hence B ( x - l ) has kernel 0 and B ( x ) has image G. Since
x is arbitrary, this gives the result.
Denote by y ( x ) the automorphism class containing B ( x ) . By (9.2),
y is a homomorphism y: lI-+AutG/InG, hence (IT,G,y) is an abstract
kernel. Since II is not cyclic of order 2, we can assume that lI contains
more than two elements. The free group F then has more than one
131
132
+1 has a center C+ 1.
+x
10.
Schur's Theorem
133
yB\
y
1
I
1
134
Exercises
I . (The first Sylow Theorem.) If the order of a finite group G is divisible by
pk, p
135
136
p* : Hn (XlL7,A ) rHn(Hom, ( S ( X ) A, ) ) .
(11.1)
137
138
simple algebras over a number field. The cohomology of groups has been applied
extensively in class-field theory: HOCHSCHILD
[1950], TATE [1952], ARTIN-TATE
[I 9601.
Exercises
and only if p-I V is
I . Show that a set V is
asserts that X / n has the standard "quotient space" topology.)
2. Construct an explicit homeomorphism of A n with I".
Chapter five
f (gr,a)= f(g,4 .
(1-4)
1.
Tensor Products
139
140
Q#:
If
is an efiimorphism, both
$1
Q#
Q#
(1.8)
are isomorfihisms.
This holds also for negative m, since f (- g,a)=- f (g,a)= f (g, - a).
Hence the middle associativity condition (1.2) may be omitted in
defining 8,.
Tensor products of finite abelian groups can be explicitly computed.
For each positive integer m, let Zm(go) be the cyclic group with generator
go of order m, while m A denotes that subgroup of A which consists
of all multiples ma, aeA. We claim that an isomorphism
141
Exercises
I.
Prove Z,,,@Z,rZ(,,,),
2. Show that G
IX G~ B ~ A A~ . ~
3. Show that the tensor product of two free modules is a free abelian group.
4. If Q is the additive group of rational numbers, Q @ Q r Q .
(or =a@kb).
(2.1)
I 42
5': (A@B)@CrA@(B@C),
t: A@B=B@A
(2.2)
of K-modules, with @ short for @K . The function (a@b) @ c is Ktrilinear (i.e., K-linear in each argument separately) and is universal
for K-trilinear functions on A x B x C to a K-module. The same holds
for K-multilinear functions in any number of arguments.
Similarly, (cf. 1.6) HomK(A,B) becomes a K-module if to each
f : A+ B the multiple kf : A+ B is defined as (kf)(a)= k (fa).
A module over a field F is simply a vector space V, and Hom,(V, W)
is the vector space of all linear transformations f : V+W. Suppose
that V and W have finite bases (e,, .. . , em}and (4,..., h,,}, respectively.
This means that V is a direct sum 2 Fe, of copies Fei of the field F.
Since "Hom" carries finite direct sums to direct sums, Hom,(V, W)
is a vector space of dimension mn, as in the usual representation of
linear transformations f : V+W by m x n matrices. Since the tensor
product is additive, V@,W has a basis of mn vectors e@$, hence
has dimension mn. In particular, any vector u of V@,V has a unique
expression as u= zxij(ei@ej); the ma constants xiicF are known
as the "components" of the "tensor" u relative to the basis (ei). From
a change of bases one calculates the corresponding change in these
components xii. Classical tensor analysis, lacking a proper conceptual
definition of the tensor product, described twice covariant tensors
(elements u of V@,V) strictly in terms of such components and their
transformations under change of basis. A tensor with one covariant
and one contravariant index is, by definition, an element of V@,V*,
where V* = Hom,(V, F) is the dual space. Now the given basis {ej}
for V determines a dual basis (ei} for V*. Any tensor in V@,V* has a
unique representation as a sum 2 xij (ei@ ej), so is determined by components xSj,for i, j = 4 , . .., n.
Exercises
i . If a new basis {el} in the finite dimensional space V is given by the formulas
e; = Zi ti jej, calculate the resulting transformation in the components of
3. Bimodules
If R and S are two rings, an R-S-bimodule A - in symbols .As is an abelian group which is both a left R-module and a right S-module,
with always (ra)s= r (as). For example, any ring R is an R-R-bimodule;
3. Bimodules
143
An elzdomorPhism of the right R-module A is by definition an Rmodule homomorphism f : A-tA. Under addition and composition the
R-endomorphisms of A form a ring End,(A) =Hom,(A,A)
set of
with identity element l A. The equation (fa) r = f (ar) which states that
f is a homomorphism of right R-modules also states that A is an End, (A)R-bimodule. If ,AR is a bimodule, the left multiplication 1, by SES,
defined by l,a=sa, is an R-endomorphism of A, and the correspondence
s+l, is a ring homomorphism S-+EndRA. Conversely, given A, and
a ring homomorphism S -+EndRA, pull-back along this homomorphism
yields a bimodule ,AR. In our treatment of ExtR(C,A) (Chap.111),
we showed how to multiply an element S,EEEX~;(C,
A) on the left
by a homomorphism a: A+A' and on the right by a y: C'+C, and we
144
proved (Lemma 111.1.6) the congruence (a So)y =ci (Soy ) . For endomorphisms a and y, this means that Extl (C,A) is an End, (A)-End,(C)bimodule. If we pull back this bimodule structure along T+End,C
and S+End,A, we get the implication
Theorem 3.1. I f R and S are rings with A, B, and C modules irc the
situation A,, ,BS, CS, there is a natural isomorphism
q : Hom, (A@,B, C) =HornR ( A , Horns ( B ,C))
(3.5)
The proof is mechanical. First check that (3.6) assigns to each acA
and each S-module homomorphism f : A@,B-+C a function F =
[(qo a] which, as a function of b, is an S-homomorphism [(qn a] : B -.C.
Next check that qf, as a function of a, is an R-module homomorphism
of A into Hom, (B,C). Finally check that q (I,+ f,) =qfi+qf,, so that
q is a group homomorphism, as asserted.
To show that q is an isomorphism, construct an inverse map (.
To this end, take any right R-module homomorphism g: A+ Horn, (B,C),
and consider the function (ga)b of aeA, bcB. Any r in R operates on
a on the right and on b on the left, and
One checks that (: Hom, (A, Hom, (B,C)) -+Horn, (A@* B, C), and that
both composites cq and q( are the identity. Both domain and range
3. Bimodules
145
q': Hom,,
(3.7)
Proof. The right U-module structure on the terms of (3.5) and the
description of these terms as functors of A are given by identical formulas. Hence the fact that 7 is natural (in A and C) implies that 7 is
a module homomorphism in U and T. In case U= T, this yields (3.7).
As another application we prove
Corollary 3.3. I f PRis projective as an R-modde, while the bimodzlle
,P,' is firojective as an S-modzlle, then P@,P1 is a firojective S-module.
Proof. To say that P' is S-projective means that to each epimor-
GORA is an EndR(G)-EndR(A)-bimodule.
10
146
4. Dual Modules
The dual or conjugate of a left R-module A is the right R-module
A*= Hom, (A,R). Thus an element of A* is an R-module map f : A-tR,
while f r: A-tR is the R-module map defined for each a A by (fr)a =
(fa)r. The dual of an R-module homomorphism u: A+A1 is u*=
Hom(u, I): A1*+A*, so the dual is a contravariant functor on left
modules to right modules. Similarly, the dual of a right R-module G
is a left R-module G*.
For left modules A and B there is a natural isomorphism
(A@ B)*gA*@ B*.
(4.1)
(B/A)*sAnnihA<B*,
B*/AnnihA
A*.
(4.2)
4. Dual Modules
147
f=z
=cL
148
of 5 of (4.4) with adjoint associativity. The latter is always an isomorphism, and so is 5 when A = L is finitely generated and projective.
Note. For further discussions of duality see DIEUDONN*
[1958], MORITA
[1958], Bass [1960], or JANS[1961].
Exercises
I. For each
0 : ABZA*+R.
RA,
is the identity.
149
so g @ 2 is in the kernel of 1 8 % .
This example can be reformulated thus. For a submodule A< B
one cannot assume that G @ A<G @ B, because an element g@a of
G @ A may be non-zero while the "same" element g@a becomes zero
in G@B. For these and related reasons we have insisted from the
start that the inclusion A< B be represented by a map x : A+B.
In this example, the integer 2 can be replaced by any integer m.
Thus we can describe certain elements in Ker(l@x) for R=Z and
(%,a)
: A w B +C any short exact sequence of abelian groups. These
elements g @a arise whenever there is an element b with both xa=mb
and mg=O for the same integer m, for then
Now xu, and hence a, is determined by b, while g@a depends only on
a b C.~ Indeed, a b =a b' by exactness implies b'= b+ xu, for some a,,
whence x(a+ma0)=mb1 and g@(a+ma,)=g@a+g@ma,=g@a.
The kernel element g @a depends on g, mEZ, and ab= c ; furthermore
mc = m (ab)=a (mb)= axa= 0, by exactness. By way of notation, set
whenever mc,= O=mc,. For any two integers m and n, one calculates
that
k (g, mn, c ) = k (g,m, nc)
(5.4)
whenever mg=O, mnc= 0, and that
whenever gmn=O, nc=O. Here we have written gm for mg because
we can consider the abelian group G as a right module over 2. These
relations will now be used to define a new group.
150
Exercises
I.
If A n B, show that each element in Ker (A @ Z m-+ B @Zm) has the form
k (c, m, I ) for 1 the generator of 2 , .
2. If J is a two-sided ideal in the ring R show that the map a @(Y+ J) 4ar for
a E J yields an epimorphism
(R/J) -3J/JZ of R-modules. If Jz+ J prove that
the injection J -tR induces a map JBR(R/J)+RQR (R/
J ) which is not a monomorphism.
3. If (y, t ) :G + H + K and (B, a ) :D - t B - 3 C are exact, then t @ a :H @ R B +
KBRC has kernel y + (G @ B ) u ~ (, H @D).
JaR
i 51
o +Tor
5 G@A*G@B+G@C+O.
The maps are those induced by x and a except for E,, which i s defined
on each generator of Tor (G,C) by the formda
for k as ilz (5.2). This map E, i s natural whert its arguments are considered
as bifunctors of E and G.
152
commutes, and we are allowed to assume the top row exact. Since Go
contains all the elements hjeG used to show x,u=O, these same elements will show x, uo= 0 in Go@ B. By exactness of the top row, there is
atoTor(G0,C) with E,to=uo. But E, L * ~ ~ = ( LE @
* t oI =~()~ @ l ) u o = u ;
this proves the bottom row exact at G@A.
This argument depends not on the particular form of the definitions
of Tor and @, but only on the fact that these groups were described
by generators and relations.
Return to the proof of exactness. Now G is finitely generated, hence
representable as a direct sum of cyclic groups. Since both Tor and @
carry direct sums into direct sums, the sequence (6.6) is the direct sum
of the corresponding sequences for cyclic groups G. If G=Z is cyclic
infinite, the torsion products are all zero and the sequence is isomorphic
to the given sequence E. If G =Zqis finite cyclic, the various terms have
been calculated in (1.9) and (6.5); the calculations amount to a diagram
in which the central portion is
Exactness here may be verified from the definitions of the terms and
the exactness of E. For example, if x (a+ qA) = O in B/q B, then xa=qb
for some be B. Thus a (qb) =o, hence ab= ccqC; the very definition of
the switchback yields E#c =a + q A.
We leave the reader to prove
(iv) Any short exact sequelzce remains exact upon tensor mdtiplication
by G;
(v) Any exact sequence remains exact when tensored with G.
153
Q:
L+Lf.
Exercises
I . For both sides defined, prove ( g , %+ ma, a ) = ( g , mx,a )
+ ( g , ma, a ) .
2. Let Q > Z be the additive group of rational numbers, and let T ( A ) (the "torsion subgroup ") be the subgroup of A consisting of all elements of finite order in A .
Establish a natural isomorphism Tor (Q/Z,A ) T ( A ) .
Y,
a)for
gY = 0 = r a and relations (6.1) through (6.4). Show that the sequence (6.6) with
154
length n,
L : Lo+-L,+...-L,-,+L,,
L*: L:+L,i,+...+L:+-L:.
resembling the definition of the congruence relation on Ext n by morphisms of long exact sequences. Formally, the equality relation on
Tor, is to be the weakest equivalence relation in which (7.1) holds;
this means that two triples in Tor, are equal if the second is obtained
from the first by a finite succession of applications of the rule (7.1).
This describes Tor, as a set.
This set is a functor. Indeed, for maps 7 : G +GI, y : C 4 C ' the rules
155
with VG: G@ G+G and Vc the codiagonal maps (111.2.1'). The proof
of the group axioms is direct. The associative law follows from the
associativity of the codiagonal maps. The commutative law follows from
(aG)*( h e t,) = (aC)*(t2@tl) and VG a, = VG. As a zero for the addition
we may take (O,O,0), where the middle zero designates the zero complex
of length n, while the inverse - (p,L,v) is (-p,L,v). The maps 17, and
y , , defined as in (7.2), respect this addition, so Tor, is a bifunctor to
the category of abelian groups. The same formulas (7.2) show that if
the modules G and C are bimodules =GR,RCSfor other rings T and S ,
then Tor, is a bimodule T(T~r,)S,much as in (3.1).
Proposition 7.2. Every element of Tor, (G,C) has the form (p, F, v)
where p : F+G, v: F*+C, and F is a chain complex of length n of finitely
generated free right modules. Hence the functor Tor, defined using complexes
of finitely generated free moddes is naturally isomorphic to the functor
Tor, defined using complexes of finitely generated projective modules Lj.
Proof. The construction above, using only free modules instead of
projectives, yields a functor Torf,(G,C). Since each free complex F
of length N is also projective, each element (p, F, v) of Torf, is also an
element of Tor,. This map Torf +Tor has a two-sided inverse. For take
any (p, L, v) Torn. Each L, can be written as a direct summand of
some finitely generated free module F,=L,@ M,. Make F a complex
with boundary a@ 0: L,@ M, -+L,-,@ M,-I . The injection L : L+F
i 56
and the projection IG: F+L are chain transformations with n &=I,
6*n*' 1. B y our equality rule,
(p,L, v)= ( p , L, v L* lt*) = ( p lt, F , v L*) ;
this is an element of Torf, for F is a free compIex of length n. By this
process, triples equal in the sense (7.1) are turned into equal triples
of Torf ; hence the natural isomorphism Tor,rTorf, .
For n= 0, Tor, may be identified with @:
@ ( p , F , v ) = , Z p ( e i ) @ v ( e i )E G@RC,=1
@ ( p l e ,F , v)
=s
, (C,pl(e,! rji)@ v ei)
9
=C ( p ' e ~ @ C v ( r j i e i ) ) = O ( pF',ve*).
',
i
This shows O well defined for the equality in Tor; also, if F=F1 and
e,! is a different basis in F, it shows the definition of O independent of
the choice of the basis in F. Since O is a two-sided inverse of the previous
map, the proof is complete.
157
..-+L,~~-L,+:C
l-
E:O+A
+B+C+O.
(7.5)
Theorem 7.6. For E E Extl (C,A) and t Tor, (G,C) the Product E t
is a well defined element of Tor,-,(G, A ) which satisfies the associative
laws
( E t ) = ( a E ) t , (Ey)t1=E(y,t'), E(q,t)=q*(Et),
(7.7)
158
... +Tor,
+ a .
..
(7.9)
n 2 m.
(7.10)
A),
nam.
(7.11)
159
can be taken to be symbols ((gl, ..., g,), x, (el, ...,c,)) with giEG, cjEC and x an
m x n matrix of entries from R such that (g,, ..., g,) x = 0= x (c,, ..., c,)'; here
the prime denotes the transpose. Describe the addition of such symbols and show
t h a t the equality of such is given by sliding matrix factors of x right and left.
2. Obtain a similar definition of Tor, (G, C).
3. Prove that Tor, (P,C) = 0 for n > 0 and P projective. (Hint: show first that
i t suffices t o prove this when P is finitely generated.)
The exactness of (7.9) can be proved directly (i.e., without homology) as in the
followipg sequence of exercises.
4. Show that the composite of two successive maps in (7.9) is zero and that the
exactness of (7.9) for G finitely generated implies that for all G.
5 . For E'= (1, t): G n H -+K exact with H free show that E: : Tor, (K, C) 4
Tor,-, (G,C) is an isomorphism for n> I and amonomorphism with image Ker (181-,)
for n = I.(Hint: construct an inverse map.) Show that E; maps the displayed
portion of (7.9) for n= Iisomorphically onto the Ker-Coker sequence of the 2 x 3
diagram with rows G BE and H BE.
6. Prove by induction on n that the displayed portion of (7.9) is exact.
161
( I , X i , v hk*)=a(l,
5)=(1, X i , 5
a*),
so the uniqueness assertion of Cor. 7.4 yields v hi*= C a*. Let ",XIbe
the part of X' from X i to XL inclusive and ,+:X' the part from X i
to X;+, , so that h': L+;X1 and a: "+:X'+:X' are chain transformations.
The original element t of Tor,(G, A ) becomes
11
162
163
Proof. Clearly (iii) + (ii). Conversely, given (ii), Thm. 5.4 implies
that any short exact sequence remains exact upon tensor multiplication
by G; since a long exact sequence is a composite of short ones, this gives
(iii).
Given (i), the long exact sequences for Tor, yield (ii) and (iv). Conversely, given (ii), represent C as a quotient C r P / A of a projective P ,
so that
O=Tor,(G, P ) +Tor,(G,C) - + G @ A + G @ P
is exact with G@A+G@ P a monomorphism by (ii), and therefore
Tor, (G,C) = 0. The proof that (iv) + (i) is analogous.
Finally, (v) + (i); conversely CrP/A and exactness of
O=Tor, (G, P)+Tor, (G,C) +Tor,-,
(G,A)
Exercises
1 . If ? I : Y + A
is a projective resolution, establish an isomorphism
w': ~ o r(G,
t A ) cr Hn(G BRY).For E': G-H + K exact prove that wfE' = aEBYw'.
2. For a projective resolution X of G let S,(G,X) be the n-fold exact sequence
0 +ax, + X , - , -+
+G+O. Show that the isomorphism w of Thm. 8.1 is
w t = CIS a-1[S, (G, X) t].
,
. ..
(94
11*
164
given as u = s @gl+
~ ( k @ l ) = s k @ g ~ l + ( - l ) ~f 2~kg@~ t l .
This is in accord with the sign convention, since t of degree 1 has
been commuted past k.
165
ZBY
Exercises
I.For complexes K, L, M over a commutative ring, establish the adjoint
associativity Horn (K @ L ,M ) r Hom(K, Hom (L, M)).
166
167
has exact rows and columns. Indeed, the exactness of the long column
while the exactness of the
states that C, is the kernel of a: L,-+L,-,,
short column gives B , as aL,+, , and the exact row defines H , as C,/B, .
Now take the tensor product of this diagram with G. Since G is flat,
the new diagram is exact, and states that G @ H , is isomorphic, under
p, to the homology group H,(G @ L ) , thus proving the lemma.
To prove Thm. 10.2, we regard the families C,= C, ( K ) and D,=
K,/C,rB,-,(K)
as complexes of flat modules with zero boundary,
so that C H K +D is an exact sequence of complexes. As D , r B,-, ( K )
is flat by hypothesis, Tor, (D,, L,) = 0 , so the sequence E : C @ L
K @ L + D @ L is also an exact sequence of complexes. The usual
exact homology sequence for E reads
168
Since Dm+,and C, are flat, the vertical maps are isomorphisms by the
lemma. If we know that the square part of the diagram is commutative, we get Ker (E,,,) from Ker (a' @ 1 ) ; explicitly, Ker (E,+,) g
Ker (a' @ 1 ) CTor, ( H ( K ) ,H ( L ) ) , and Coker En+, rCoker (a' @ I ) r
C H ( K )@ H(L). Thereby (10.4) becomes the desired sequence (10.3).
One checks also that the first map of (10.3) is indeed the homology
product, while the second map ,9 is described by the commutativity of
the diagram
Hs ( K @ L ) L C TorI(Hrn-1 ( K ) Hq( L ) )
m+q=n
H,((KIC) 8
LS*
~
2m+q=n
c) ( K I C @) ~H ,(L)
( 1 0.6)
169
+z
+z
commutes. By the choice of f and g, the outside vertical maps are isomorphisms. Hence by the Short Five Lemma the middle vertical map
is also an isomorphism. The bottom exact sequence is thereby isomorphic to the top exact sequence, which has just been shown to split.
Therefore the bottom sequence splits.
This proof, due to A. DOLD,depends on the following useful lemma.
170
(I/m)a(k@l)=u@Z+ (-l)^+lk@v,
dim u = n
is a (Type 11) cycle. One may verify that its homology class is determined, module Im #, by cls u and clsv. This yields an expression for 8
in the KUNNETH
formula, as follows:
474
is a free resolution of H,,-,; its tensor product with A then has homology
0 in dimension 2, Tor (H,-, , A) in dimension 1, and Hn-, @A in dimensiod zero. The first states that C, @A can be regarded as a subgroup
of &,, @ A :indeed
The second states that Ker a&, @ A =Tor (H,-, , A); the third (with n
replaced by n + l ) that C,@A/Im a,,+, =H,@A. Therefore H,(K @A) =
Ker a,,/Im a,,+, is an extension of *,@A by Tor (Hn-,, A ) , as asserted
by the exact sequence (11 .I).
172
Corollary 11.2. If K am2 K' are compZexes of abeZialz grou$s, each with
no elements of finite order, while f : K+K' is a chain tralzsformatiolz
with f* : H, (K) r H , (K') an isomorphism for each n, then f , : H, (K @ A )
+H, (K' @ A ) is an isomorphism for every abelian group A and every n.
Proof. Write the sequences (11.1) for K and K' and apply the Five
Lemma, as in the proof of Cor. 111.4.6.
These universal coefficient theorems express the homology and
cohomology of K with any coefficientsin terms of the so-called "integral "
homology H,(K), a,t least when the K, are free. If the K, are free and
finitely generated abelian groups, there are corresponding expressions
in terms of the "integral" cohomology Hn (K, Z), as in Ex. 2 below.
Exercises
1. For abelian groups K and A construct natural homomorphisms
Hom (K,Z) @A +Horn (K, A) and K @A +Hom (Hom (K,Z),A). Show them
isomorphisms when K is a finitely generated free group, and chain transformations
when K is a complex.
2. Let K be a complex of abelian groups, with each K, a finitely generated free
abelian group. Write H"(K) for H" (Hom (K,Z)). Using Ex. I and the universal
coefficient theorems, establish natural exact sequences
+,
4. For K as in Ex. 3, and Zp the field of integers modulo calculate the dimension
of the vector space H, ( K @Zp) from the Betti numbers and torsion coefficients of K.
5. If 69 is a complex of vector spaces over a field F, write K* for its dual
Hom (K, F). If each K, is finite dimensional, establish the natural isomorphisms
H"(K*)= [H,(K)I*.
Notes: Tensor products were long used implicitly; for example, via GmRXRe,
xGei or V@ W E Hom (V*, W). Their central role in multilinear algebra was
highlighted by BOURBAKI'S
[I9481 treatise on this subject. The tensor product for
abelian groups was first explicitly defined by WHITNEY[1938]. The universal
coefficient theorem 11.1 was first proved by CECH [I9351 who thereby first introduced (but did not name) the torsion product Torl. CARTAN-EILENBERG used
resolutions to define the higher torsion products. The description ( 8 6) of Tor, for
[1954. 1123) is
abelian groups by generators and relations (EILENBERGMACLANE
spectrum of a complex K of abelian groups (the
useful in treating the BOCKSTEIN
[19581 ; PALERMO
[I 95 71).
various H, (K,2,) and their interrelations - BOCKSTEIN
A similar description (Ex. 7.1, 7.2) of Tor, by generators and relations (MACLANE
[1955]) involves some rather mysterious new functors, the "slide products" (e.g.,
T in Ex. 6.7) and leads to the conceptual characterization ( 3 7) of the elements of
Tor, as triples ( p , L, v ) .
I . Algebras by Diagrams
173
Chapter six
Types of Algebras
1. Algebras by Diagrams
k(3,1,)=(k&)A2=iZ,(kA2),
~ E K , A1, A 2 d .
are commutative. Indeed, the first diagram asserts that the product
is associative, while the deft and right halves of the second diagram state
that I ( I K ) is a left and right identity element for the product in A and
that n ( I k @ A ) = k 1 * n ( A @ I k ) .
In case K is the ring 2 of integers, a 2-algebra is simply a ring, so
this gives a diagrammatic definition of a ring, via tensor products of
abelian groups. The dual diagrams define a "coring" or a "coalgebra".
Algebras may be graded by degrees such that deg &A2) = deg A,+ deg A,,
or may have a differential a, with a ( & A , ) = ( a ~ ) 1 , + A l ( ~ 1 , )This
.
chapter will give a uniform treatment of these various types of algebras
and the modules over them. As an illustration of algebras with a differential, we first consider certain resolutions over a polynomial ring.
Let P=F [x] be the usual ring of polynomials in an indeterminate x
with coefficients in a field F; actually, P can be regarded as an F-algebra,
but for the moment we consider it just as a commutative ring. Since
F=F [ x ] / ( xis
) the quotient of P by the principal ideal (x)of all multiples
of x, we can regard F as a P-module so that E (x)= 0 defines a P-module
homomorphism 8 : P+F. Form the sequence
174
2. Graded Modules
175
2. Graded Modules
An (externally) 2-graded K-modzcLe is a family M ={M,, , n = 0,
fI , f2, .. .) of K-modules M,,; an element m of M,, is also said to be
an element of degree n in M (briefly, deg m= n). A graded szcbmodzcle
S<M is a family of submodules S,,<M,,, one for each n. For two
2-graded modules L and M a homomorfihism f : L +M of degree d is a
family f = {f,: L, +M,,+,; nZ) of K-module homomorphisms f,,. The
set of all f : L +M of fixed degree d is a K-module Horn, (L, M). The
composite of homomorphisms of degrees d and d' has degree d+d'.
A 2-graded module M may also be written with upper indices as
Mn=M-,; in particular, Homd(L, M)=Hom-,(L, M).
A graded K-module M is a 2-graded module with M,,= 0 for n <0.
These graded modules are of most frequent occurrence, and will be
studied below, leaving the reader to formulate the corresponding facts
for 2-graded modules. Warning: Many authors use "graded" for our
2-graded and "positively graded" for our graded modules.
A trivially graded module M has M, = 0 for n#=0.
The graded K-modules M, with morphisms hom (L, M) =Homo (L, M)
the homomorphismsof degree 0, form a category. Each f : L +M of degree
o has kernel, image, cokernel, and coimage defined as expected (i.e.,
termwise for each n) ;they are graded modules with the usual properties.
For fixed degree d, Horn, (L, M) is a bifunctor on this category, contravariant in L and covariant in M. Alternatively, the family Hom (L, M)=
{Hom, (L, M)) is a bifunctor on this category to the category of 2-graded
K-modules. Both bifunctors are left exact, in the sense of Thm. 1.6.1
and Thm. 1.6.2.
The tensor product of two graded modules L and M is the graded
module given by
W@M),,= 2 L,@M,;
(2.1)
P+q=n
176
(2.2)
Here @ = gK,
and the ground ring K is regarded as the trivially graded
. We regard these isomorph'i;ms
module K with K,, = K, Kn=O for m
i s identities. This we can do because they are manifestly consistent with
each other: Given any two iterated tensor products of the same modules
M I , ... , M s , a suitable combination of these isomorphisms provides a
canonical map of the first tensor product into the second-deleting or
adding factors K a t will, and with sign according to the sign conventions,
as in (2.4).
hold for 2-graded modules
The same properties of HornKand gK
and in a variety of other cases, as follows.
A bigraded K-module B is a family B = { B p , , /P,q c Z ) of K-modules
with Bp,,= 0 when p <O or q < 0 ; a homomorphism f : B -+ B' of
hidegree (d, e) is a family {f*,,: Bp,q-+B~+,,q+,}
of K-module homomorphisms. For example, the tensor product of two graded modules
L and M is initially a bigraded module {L* @ Mq},which the summation
(2.1) has turned into a singly graded module. Similarly, the tensor
product of two bigraded modules B and C is a 4-graded module which
yields a bigraded module by
An element of B*,, is said to have total degree ++q. The natural isomorphisms (2.3), (2.4), and (2.5) hold for bigraded modules when the
total degrees are used in the sign of the transposition z.
Trigraded modules, 2-bigraded modules, and the like are defined
similarly.
3. Graded Algebras
177
3. Graded Algebras
The first asserts that the "product" il,u=n(l @,u)is associative, and
=IA is a two-sided identity for this product.
the second that IA(IK)
.A homomor+hism f : A+Af between two graded algebras over the same
K is a homomorphism of degree 0 of graded K-modules such that the
diagrams
A B A ~ A
K%I
are commutative.
These definitions may be restated in terms of elements. A graded
algebra A is a family of K-modules {A,, n = 0, I, . . .) with a distinguished
Mac Lane, Homology
12
I7 8
element ~ E Aand
, a function which assigns to each pair of elements
I , p a product I p which is K-bilinear and such that always
<
3. Graded Algebras
179
that is, if ZA=ZAT: A @ A - t A , with z the transposition (2.4). In consequence, the elements of even degree commute in the ordinary sense.
If, in addition, Ia=O for every element I of odd degree, A is called
strict1 y commutative. If the ground ring K is a field of characteristic not 2 ,
then any commutative graded K-algebra is strictly commutative, for
(3.4) with deg I odd gives AI= -11, 2 Aa= 0, so 2-I in K implies l a = 0 .
For example, a graded #olynomial algebra P = P K [ x ]may be defined
for an "indeterminate " x of any degree d 2 0 . If d= 0, P is the ordinary
ring of polynomials in x with coefficients in K. For d>,o, P is the
graded module with P,=O for n+ 0 (mod d ) , while P,, is the free
K-module on one generator x4 for each q 1 0 ; the product is defined by
xfiX4= x $ + ~ .If d is even, this polynomial algebra is commutative. P is
characterized up to isomorphism by the fact that it is free in x: For any
graded K-algebra A with a selected element Id of degree d there is a
unique homomorphism f : P-+A of graded algebras with f x= Ad.
The exterior algebra E = E K [u] on one symbol u of odd degree d is
constructed from the free K-module Ku with one generator u as the
graded algebra E with Eo= K, Ed= Ku, En= 0 for O=/= n+ d, and with
product determined by I u = u =uI, ua= 0 . It is strictly commutative.
We may also define E as the quotient algebra PK[ x ] / ( x a )where
,
x is an
indeterminate of degree d and (xa) denotes the (two-sided) ideal in P
generated by xa. The algebra E may be characterized as the strictly
commutative algebra free on u : Given any strictly commutative
with a selected element AdcAd, there is a unique homomorphism
f : EK[u]+ A of graded algebras with f (u) = I d .
The tensor algebra T ( M )of a K-module M is the graded K-module
T,(M)=K, T,(M)=Ms=M@
... @ M
(n factors),
where the second sum is taken over all di with 4+ . .. +a,=%. For
M =MI this includes the previous case. The graded algebra T ( M )with
the obvious K-module injection M -+ T ( M ) (of degree 0 ) is characterized
up to isomorphism by the following "universal" property:
12*
180
Note that a graded algebra isn't an algebra, but that an internally graded
algebra may be regarded simply as an algebra (ignore the grading). The
internally graded ideals, defined as above, are usually called homogeneous
ideals; they are among the ideals of the associated ungraded algebra.
Exercises
I . Describe the free graded K-module on any graded set of generators.
2. Describe the bigraded tensor algebra of a bigraded module, and prove the
analogue of Prop. 3.1.
181
3. Let S be a set of elements in a graded algebra A. Show that the set of all
homogeneous sums of products I s A', for sES, is a graded ideal in A and is the
smallest ideal containing S . It is called the ideal generated by S (or, sflanned by S ) .
4. Show that a graded K-algebra may be described as a graded ring R equipped
with a homomorphism I : K -+ R of graded rings such that always (Ik) r = v ( I k ).
l@r@l
AA
(4.1)
where t is the (signed) transposition (2.4) of Z and A, and with identityelement map given by I @ I : K = K @ K +A @Z. In terms of elements,
the product is given by
@@a) (A'@ar)=(--l)degodega'
AAr B a a r
and the identity of A @ Z is In@ lz. The axioms for a graded algebra
all hold. If f : A-+Ar and g: Z+Zr are homomorphisms of graded
algebras, so is f @g: A@Z-+A1@Z'. Also, A+A@l,, cr+l,@a define
homomorphisms
A-+A@ZtZ
of graded algebras. With these mappings, the tensor product A @ Z is
characterized up to isomorphism by the following property:
182
is called the graded polynomial algebra on the given xi. In each dimension
m , PK[x,, . . . , x,] is the free K-module on all
x:@..
. @q
with
%&+...
+ end,=m
two such generators are multiplied by adding the corresponding exponents. This polynomial algebra is the free commutative algebra on the
generators xi of even degree, in the sense of the following characterization.
5 : {x,, ..., xn}+A with deg (Exi)= deg xi for aU i extends to a unique
183
the xi have degree zero, PK[%, ..., x,] is trivially graded. We may
regard it as an ungraded algebra, and denote it as K [x,, .., x,] ; it is
the ordinary polynomial algebra in n indeterminates over K. For n
given constants kicK, Prop. 4.2 yields a unique homomorphism
f : PK+K with f xi = ki , i=l , ..., n. This is the homomorphism obtained
by the familiar process of "substituting ki for xi, i = l , ..., n".
We next construct a similar free strictly commutative algebra with
generators q of odd degree (degree1 will suffice). For n letters y , .. , u,,,
each of degree 1, the tensor product (over K)
,... U i p = U ~ l A t 4 i , A ... A % ,
UiUi
< ip5 n.
with 15 <<i,<
where
The tensor product K @, K' of two commutative rings is a commutative ring, and the definition of E shows that
There are similar isomorphisms for more u's, more factors, or for E
replaced by P. The polynomials on n commuting indeterminates with
coefficients in a not necessarily commutative (ungraded) K-algebra A
may be defined as
P ~ [ x l...,
, x,]=A&P~[xl,
..., x,].
(4.7)
184
Exercises
C = C(A) be the ideal spanned (cf. Ex. 3.3) by
all differences Ap - (- 1 ) m " p A for m = deg A, n = deg p. Show that A/C is commutative, and that any homomorphism of A into a commutative algebra factors
uniquely through the projection A-tAlC.
2. The symmetvic algebra S ( M ) is defined from the tensor algebra as S (M) =
T (M)/C( T (M)),for C as in Ex. 1. Show that Prop. 3.1 holds when A is commutative
and T(M) is replaced by S (M), and that, for M free on a finite set of generators of
even degrees, S (M) is the polynomial algebra.
3. Make a similar construction of the exterior algebra on any graded K-module
M consisting of elements all of degree 1.
4. I n E x . 2, show S ( M @ N ) = S ( M ) O S ( N ) .
5. Show that free strictly commutative graded algebra on any finite graded set
of generators may be constructed as a tensor product of polynomial and exterior
algebras.
6. If P = K[x] and Q = P[y], show that Q, as an (ungraded) K-algebra, is
isomorphic to K[x, y ] . Extend this result to the graded case with more indeterminate~.
1. In any graded algebra A let
) ( ~ ~ g f ) 1( (~f C~) g; ~ )
(5.4')
185
the usual sign arises from the definition (2.2) of l @ f . The set of all
C,
such f of degree d is a K-module which we denote as H o ~ - ~ (A).
The class ,A of all left A-modules is a category with morphisms
hom, (C, A) = Horn! (C, A) those of degree 0. In ,A, direct sums, suband quotient-modules, kernel, image, coimage, and cokernel are defined
as expected, with the usual properties. For each n, Homn (C, A) is an
additive bifunctor on A to K-modules, contravariant in C and covariant in A. The fam' y Hom, (C,A) = {Hom; (C, A), n = 0, fI, & 2, . . .)
is a similar bifunctor on ,A to 2-graded K-modules. According to the
definition (5.4) of a A-module homomorphism, we can also describe
Hom, (C,A) as that 2-graded K-module which is the kernel of the natural
homomorphism
186
187
AAz-(A@~o~)
A
(5.91
viaoOPb=(-~)(~~g~)(~~
(il@aOP)a=(-l)(deg~(d*a)Iau.
g~)bu,
This reduction carries with it the definitions of Hom and @ for bimodules. Thus
for bimodules ,GA and .Az the bimodzlle tensor product
G@A-zA=G@(A,~)A
(5.10)
is by (5.7) the quotient of G gK
A by the graded K-submodule spanned
by all
gI@a-g@la,
ag@a- ( - ~ ) ( ~ ~ g " ) ( ~ q g + ~ ~ g ~ ) ~ @ a a .
The vanishing of the first expression is A-middle associativity; that of
the second is Z-outside associativity. Similarly the graded K-module of
birnodule homomorphisms of .C, into .Az is written HornA-, (C,A) =
Hom(~g,,) (C,A).
Exercises
1. An (ungraded) K-algebra A is a ring R equipped with a ring homomorphism
I: K+R with I (K) in the center of R. Show that a left A-module A is just a left
R-module, with the K-module structure of A given by pull-back along I. Show
A.
also that HornA(C, A) = HomR(C, A) and G @A A = G aR
2. As in Ex. i , reduce modules over the graded algebra A to modules over A,
regarded as a graded ring (cf. Ex. 3.4).
I88
b
For Z(17,) is characterized (Prop. IV.1.1) by the fact that any multiplicative map pi of Hi into a ring S, with pi (I)=is, extends to a ring
homomorphism Z (17,)+S. By Prop. 4.4, a multiplicative map
p : fix& +S with p (1)=1 then extends to a unique ring homomorphism
Z (17,)@Z (17,)+S, so that Z
@Z (17,)satisfies this characterization
of the group ring Z (ax17,).
Let C, be the infinite (multiplicative) cyclic group with generator t,
and R=Z(C,) its group ring. Any element of R is a polynomial in
positive, negative, and zero powers of t, hence may be written as tm+(t)
where is an (ordinary) polynomial in positive powers of t with integral
coefficients. The kernel of the augmentation E: R+Z is the set of all
multiples of t-1, hence the exact sequence
(a)
with R u the free R-module with one generator u and au= t-I . Thus
a : R t R u is a free R-module resolution of ,Z; it is a special case of the
resolution found in (IV.7.3) for any free group, and is analogous to the
resolution (1.3) for a polynomial ring. For any R-module A , H1 (C,, A )
may be calculated from this resolution to be the factor group
A/[ta-ala~A], while Hn(CoJ,A)=O if n>1.
The free abelian group 17 on generators t,, .. . , t, is the cartesian
product of n infinite cyclic groups. By (6.1) the group ring Z ( n ) is
R1 @ ... @Rn, where each R%s the group ring Z (C, (ti)), while the
augmentation E: Z (17)+Z is the tensor product @I .. @ E" of the
Ri+Z. For each index i form the Ri-projective
augmentations
resolution X b R R i t R i u as
i in (6.2). Form the tensor product complex
189
Exercises
II free abelian as above, show that IP ( n ,A) is the quotient LIM, where
L is the subgroup of A@ ... $A (n summands) consisting of all (al, ..., a,) with
1. For
tiai-t.a.=ai-ai
3 *
always, while M i s all (tla-a, ..., t,a-a)
for aEA. Interpret
this result in terms of classes of crossed homomorphisms.
2. Obtain a similar formula for Ha (II, A) and compare this with the result found
for two generators in IV.3.7.
3. Determine Hn(17, A) for II free abelian on n generators.
190
(7.2)
-1 with
491
but f need not commute with 8. The set of all such f is a K-module
Hom",X, Y). The family Hom, (X, Y) = {Homc(X, Y)) becomes a
DG,-module over K when the differential 8, : Homn - t ~ o m " + is
l defined
by the usual formula
aHf= a,f+ (-qU+lf a,.
(7.6)
Thus Hom, with capital "H" differs from hom,, with lower case "h":
Hom, (X, Y) is a DG,-module over K; elements all f : X +Y;
hom, (X, Y) is an (ungraded) K-module; elements all 6 : X + Y.
Moreover, hom, is the K-module of cycles of degree 0 in the complex
Hom,.
Let X be a right U-module, Y a left U-module. Considered jusi as
modules over the graded algebra U, they define a graded K-module
X @,Y which becomes a DG-module over K when the differential is
defined by (7.1); for, by that formula, a (xu @ y) = a (x @u y) (U-middle
associativity). Thus the elements of Hom, and @, are defined from
192
with r,= deg x,+ ... deg xi-, , in accord with the sign convention.
The analogue of Prop. 3 .I holds for this T(X).
One may construct universal DG-algebras on given generators. Thus
if x has degree 2 and u degree I , there is exactly one DG-algebra structure
on V = P[x] @ E[u] for which ax= u, for by the Leibniz rule (7.2) the
differential is given on the free K-module generators of the algebra V as
If u2 is a selected element of degree 2 in any strictly commutative DGalgebra U, there is a unique homomorphism f : V - t U of DG-algebras
with f x = u2 (and hence with f u = azc,).
Similar considerations will define differential internally graded and
differential 2-graded algebras.
4 93
Exercises
1. For DG-modules over K,show that the exact homology sequence (Thm. 11.4.1)
for a short exact sequence E : W-X-t Y of DG-homomorphisms x and a takes the
form of an exact triangle
H(W) 3 H(X)
&
d*
H ( Y)
(kernel = image a t each vertex), with x+ and a+ homomorphisms of graded modules
of degree 0, while the connecting homomorphism aE has degree -I. [The usual
long exact sequence spirals around this triangle, dropping one level with each aE .]
2. Prove that a DG-algebra U is a DG-module over K with homomorphisms
n: U @ U - + U and I: K + U of DG-modules, of degree 0, satisfying (3.1). Give a
similar definition of U-modules by (5.1), and show that Homu and @U may be
obtained from HornKand BK,for DG-modules, by the analogues of (5.5) and (5.7).
3. For V as in (7.7) determine the graded homology algebra H(V) when K=Z
and when K = Z p (the field of integers modulo p).
4. Construct a universal strictly commutative DG-algebra on a given finite set
of generators (of odd and even degrees).
5. For degxi=2, degui=l, the graded algebra P[xl, ..., xnl@E[ul. ..., u-I
is isomorphic to the tensor product of n algebras 5 = P[xi] @E[ui] like that treated
in the text, hence has a unique differential with axj=
ui,
i= I, .., n. For any
.
.
polynominal p in the xi, show that ap = z-aP
a
a xi @u,, where-denotes
a xi
the usual
partial derivative. Hence show directly that @ p = 0. Note that a p is the usual
differential of the function p of n variables if we replace ui by a symbol axi.
194
@e6)
and ,C,,
defined for f : A @A B+C by
for modules I)AA, ,,B,,
[(qf ) a] b = f (a @ b), just as in th'e case of rings (V.3.5). In the DG-case
one checks that q commutes with the differentials defined on both sides,
hence is an isomorphism of DG,-modules over K. In particular, taking
the cycles of degree zero on each side of (8.7) gives the natural isomorphism
hom~-,(A @ A BJ C) ~ O ~ Q (A,
- A Hom=(B, C))
(8.8)
I95
of the maps (8.5), (8.9), and (8.5). I t would be instructive to know the
various identities holding between composites of the assorted natural
homomorphisms (8.1)-(8.10) described above.
As an application; consider free and projective A-modules over a
graded algebra A. A left A-module P is projective, as usual, if each
epimorphism a: B+ C of left A-modules, of degree 0, induces an epimorphism homA(P,B) +hOmA(P,c). The free A-module on the graded
set S of generators is the A-module C containing S and characterized up
to isomorphism by the usual property (Prop. 1.5.1) that each set map
13*
196
Proposition 8.3. I f B and B' are free left A- and A'-modzdes of filzite
type, the Hom- 31
interchange is a natural isomorphism
a direct proof of the middle four interchange; that is, show that t as
specified is well defined and has an inverse.
2. Deduce the middle four interchange by repeated applications of the associativity (8.3) and A @K B s% B BKA.
3. For modules ACZand AAn describe the bimodule structure on H o m (C,
~ A)
(Attention to signs !).
4. Describe the behavior of composition (8.9) for a map B +B'.
4 97
5. Show that 5 of (8.10) may have non-zero kernel (Hint: use finite cyclic
groups).
6. Construct a natural homomorphism
A @ a H o m(B,
~ C) +Homn (Homn(A,B), C)
are commutative. The first diagram gives the associative law for the
diagonal map (or coproduct) y; the second diagram states that E is a
cozmit. Coalgebras which are not associative or which have no counit
are sometimes useful, but will not occur in this book. A homomor~hism
p : W -+W' of coalgebras is a K-module homomorphism of degree 0 such
that the diagrams
W LW@W
WLK
b
11.
(9.2)
W
'
, W1-K
ll
1@7@1
W@W'-W@W@W'@W'-----+
(W @ w 1@ (W @ '
for z as in (9.3), and with counit E@E': W@IW'+K@K=K.
(9.4)
98
pe.
1.
pi
I1
8 0 1
(9.5)
v01
W@C-W@W@C,
W@C-K@C
both commute.
A graded Ho#/ algebra V is a graded K-module V = {K} which, with
this grading, is both a graded algebra for a product map n: V @V -tV
and unit I : K -+V and a graded coalgebra for a diagonal y and a counit E ,
and such that
(i) I : K -tV is a homomorphism of graded coalgebras;
V@V
V @ V @ V @ V -l @ @ l
V@V@V@V
J.0.
v
>
V@V ,
(9.6)
199
c:
200
(n)
C h a p t e r seven
Dimension
This chapter is a brief introduction to the extensive applications of
homological algebra to ring theory and algebraic geometry. We define
various dimensions, use them in polynomial rings and separable algebras,
and in the Hilbert theorem on syzygies. Subsequent chapters are independent of this material, except for the description ($3) of Ext and Tor for
algebras and the direct product and ground ring extensions for algebras.
1. Homological Dimension
I . Homological Dimension
20 1
202
Proof. The first four conditions are equivalent by Thm. 1.1. The
sequence S in (v) gives a connecting homomorphism which is an isomorphism S, : Extl (C,A,) z E x t n + l(C,A) for each C. But Extl (C, A,) =O
I . Homological Dimension
203
204
Exercises
I . State and prove the analogue of Thm. 1.1 for left injective dimensions.
2. I f 1.gl. dim R 2 I , then
1. gl. dim R = 1
205
Thus P= P" @ P', E" and E' are DG-algebras over P" and P' respectively, their tensor product E" @E' is a P" @ PI-algebra, and (VI.4.6)
gives an isomorphism E r E U @ E ' of DG-algebras over P. Moreover,
C = d'@d and 7 =ql'@ 7'.
For n=1, the contracting homotopy
s' : E; -+E; may be defined on a polynomial f = 2 a, xi of degree k in
x=xn with coefficients sic K by setting
Proof. Suppose first that K is the P-module ,K. The Koszul resolution (2.1) stops with degree n. Hence the homological dimension of K
is at most n.
We may calculate TorP(K, K) from the resolution (2.1) as the homology of the complex
with boundary a (k @ H ~=
) k @xi. But under the isomorphisms above,
k @xi -+ (k @, xi) @ 1-+k E (xi)@ 1= 0, since by definition E (xi)= 0. Thus
the differential on the complex is zero, so TorP(K, K) is the exterior
algebra over K in lz generators. In particular, To< (K, K) =K+o, so
h. dim, K is exactly n. Similarly, Extp(K, K) is calculated from the
resolution as the cohomology of the complex
206
O t J t El+
. . a +
E,t 0
Exercises
1 . Calculate
..., xk)= k - I .
3. Examine Thm.2.2 when K is a skew field.
207
with (Ik) r = r (Ik) ;that is, with I (K) in the center of R. A left A-module
(say as defined by (VI.5.2) and (VI.5.3)) is just a left R-module A; by
pull-back along I: K +R it is a K-module, hence also an R- K-bimodule
,AK. A homomorphism a: A -+Af of left A-modules is defined to be a
homomorphism of left R-modules, and then is automatically a homomorphism of R-K-bimodules.
208
of K-modules. For lz= 0, Ext; (C, A) = Hom, (C,A) = Hom, (A @*C, A),
and 17 is the ordinary adjoilzt associativity.
In (3.2), HornK(C,A) is a left A-module via the left A-module structure of A and a right A-module via the left A-module structure of the
contravariant argument C.
209
Corollary 3.4. For any long exact sequence SEE EX^;^ (C, A) with
n> 0 the isomorfihism g of (3.2) carries the class of S into the class of
is exact, and Horn is short for HornK, then Hom (C, S) is the sequence
Exercises
1. If A is an algebra over a field, P a projective A-A-bimodule, and B a left
A-module, show that PBAB is a projective left A-module.
2. If TEE
Lane, Homology
14
4. For K-algebras A and 2,modules GA and AAz and an injective right Zmodule J use Ex.III.7.3 to establish the isomorphism ("duality"; CARTAN-EILEN
BERG
vI$/
Ext2(G, Homz(A, J))a:~ o m ~ ( ~(G,oA),
$ J).
Here the ExtKon the right are K-modules, hence P-modules by pullback.
212
re
its identity is thus (Ir, I=). The projections n, (y, a) =y, n2(7, a) = a
are algebra homomorphisms
r 3rx.z-%
Z,
(5.2)
(the injections tl, l2are not ; they do not map identity to identity). With
these maps the algebra r x Z is couniversal for r and 2 in the category
of algebras. This is why we calI r x Z the direct "product ", even though
it is often called the direct "sum" of F and 2.
Any r-bimodule becomes a (TxZ)-bimodule by pull-back along nl
(on both left and right sides); similarly any Z-bimodule or any ( r - 2 ) bimodule becomes a (rxC)-bimodule. In particular, the definition (5.1)
shows that A = r x Z , regarded as a A-bimodule, is the direct sum
Z of the A-bimodules T and 2. Since the tensor product is additive,
A @ A = (P@2)@ (P@ Z) is the direct sum of four A-bimodules
re
A@A=(r@r)@(r@Z)@(2@q@(Z@Z).
Proposition 5.2. For algebras F and Z, bidim
i w l i e s bidim ( T x Z )= 0.
(5.3)
r=0 = bidim Z
5 . Separable Algebras
213
fz
The squares are commutative; the composite of the top row is a composite of A-bimodule maps, hence is a bimodule map a': A+A@A.
214
21 5
6. Graded Syzygies
I.
rR@RZX,( ~ X Z
1
.r)R~x Z R ,
6. Graded Syzygies
which stops with the term X,. The Hilbert syzygy theorem asserts that
a graded P-module A has such a resolution with X, free graded modules
stopping at the same point. Though closely related, we cannot deduce
this syzygy theorem from our previous result, because we do not know
that a projective module must be free.
In this section we regard P as an internally graded algebra over F ;
the homogeneous elements of degree m are thus the ordinary homogeneous polynomials of that degree. We work in the category of all
internally graded P-modules with morphisms all P-module homomorphisms of degree 0 ; the kernels and cokernels of such morphisms
are again internally graded P-modules. Each internally graded P-module A = 2 A, is also an ungraded module over the ungraded algebra P.
If G is a second such module, we use G @,A and TO~;(G,A ) to denote
the ordinary tensor and torsion products, constructed without regard to
the grading. This use of internal grading has the advantage of suiting
216
where the left hand zero stands for Tor, (M, F ) , which vanishes since M
is free. By (6.1) again, B @, FsTor,P(A,F), which vanishes by assumption. Hence B @, F =O, so B=O by another application of the
previous lemma. Our exact sequence has collapsed to O-tM +A +O,
showing A isomorphic to the free module M.
6. Graded Syzygies
21 7
of graded P-modules with the Xi free has its n-th term A, free.
Such a sequence can always be constructed, by choosing X, free on
a set of homogeneous generators of A, XI similarly for generators of
Ker [X,+A], and so on. The theorem implies that h. dim,A $ n.
Proof. Since the Xi are free, the connecting homomorphism of the
given exact sequence T provides an isomorphism TO$+, (A, F)
Tor: (A,, F). But the Koszul resolution for F showed h. dim, F 5 n ,
so Tor:+, (A, F)= 0. Then by Lemma 6.2 A , is free, as asserted.
Any ideal J of P is a submodule of P ; as in VI.3 it is called a homogeneous ideal if it is a graded submodule; that is, if J is generated by its
homogeneous elements.
2.18
o t J t X o c . . . t X ,
Exercises
I . For P= F [ x , y, z ] construct an ungraded P-module which has no internal
grading consistent with this P-module structure.
2. Show that the HILBERTSyzygy Theorem holds with P replaced by any
internally graded ring G for which Go is a field.
..
..
7. Local Rings
In this section we summarize without proofs some of the accomplishments of homological algebra for the study of local rings. All rings will
be commutative.
A prime ideal P in a ring K is an ideal such that rs c P implies re P
or sc P; it is equivalent to require that the quotient ring KIP has no
divisors of zero. Any ring K has as ideals the set (0) consisting of 0
alone and the set K; a proper ideal J of K is an ideal with (O)+ J f K.
A unit u of K is an element with an inverse v ( v u = l ) in K . Clearly no
proper ideal can contain an unit.
7. Local Rings
219
<
220
each n, and the intersection of the M u is zero. An instructive presentation of this argument may be found in EILENBERG [1956].
In a noetherian ring K the KrulJ dimemion k is the largest integer
for which there is a properly ascending sequence of prime ideals
P,<P,< .-<P,< K; it can be shown that this dimension is always
finite. In a local ring L with maximal ideal M, the quotient M/Ma is a
vector space over the residue field LIM; since M is a finitely generated
ideal, the vector space dimension n=dimLIMM/Mais finite. It can be
shown that the Krull dimension of L is at most n. The local ring is
said to be regular if its Krull dimension is exactly n=dimLlMM/M2.
These are the local rings of greatest geometric interest.
Using homological methods, SERRE [1956] and later AUSLANDERBUCHSBAUM
[1956] have proved (see also A s s ~ u s[1959]) :
Chapter eight
Products
1. Homology Products
Throughout the study of products there is an interplay between
"external" and "internal" products. This relation may be illustrated
in the case of homology products. If XR and .
Y are chain complexes of
I.
Homology Products
221
n: H(U) @H(U)-+H(U)
already defined (VI .7) as
n(c1s u@cls v)=cls(uv);
we call this the internal homology poduct. The internal product may be
obtained from the external product via the product map nu: U@U -+ U,
as the composite
The external homology product can be defined with coefficient modules. Take (ungraded) K-algebras A and A', complexes ,X and A,X'
of K-modules, and right modules GA, GAT,and set O=A@Af. The
external homology firoduct is the composite map fiH= zfi in the diagram
222
Theorem 1.1. For algebras A and /I' over a field, the homology product
is an isomorphism
Proof. All modules over a field are free, so the Kunneth tensor
formula makes fi an isomorphism, while t is always an isomorphism.
the external cohomology prodzlct is the
For left modules AA and
composite map pH=[p in the diagram
4@Xi+A@A1
p+q=*
defined for n=k+m, xeXp, x'eX; as
ah'+
Then 6 (h @Jh') = 8 h
(-1)" @ 6 h' and
classes as pH(c1s h @CISh') =c1s (h @h') .
#H
is given on cohomology
Theorem 1.2. For algebras A and A' over a field and positive complexes X and X' with each Xk a d each XL a free A- or A'-module of
finite type, the cohomology product is an isomorphism
I. Homology Products
223
(1.5)
L, +M, is split.
224
pH and
Theorem 2.1. For four K-modzlles B,A, B', A', the product (2.3) is a
homomor~hism
E"
225
r).
or:
r))
r)
for n and
226
+r,
r) @r,
r)
r)
r)
3. A Diagram Lemma
227
Our product definition (2.3) is new, but the external product PT which i t
(Chap. XI.4).
defines is exactly the product m defined by CARTAN-EILENBERG
Their definition uses resolutions of A and A', but this is irrelevant (Ex. 3).
Exercises
1. If U is a DG-algebra, A a K-module, and p: A+U, a module homomorphism
with amp= 0,show that the graded algebra U@T(A) has a unique DG-structure
with alU=a,,
a l A = q , and A of degree n + I .
4. For K-algebras A and A' and modules BA, Bh,, AA, n*A' show that the
formula (2.3) provides an external product
T o d (B,A) @TorA'(B', A') + T o ~ A @ ~ ' (@
BB', A@A'),
describe its properties, and show that i t commutes with all four connecting homomorphisms. This is the product T of CARTAN-EILENBERG X1.I.
3. A Diagram Lemma
In the next section we need the following anticommutative rule on
the splicing of exact sequences.
228
with v b = (06, p b ) and p(c, 6") =yc- zb" (note the sign), while the
other unlabelled arrows are maps or composites from (3.1). The diagram
is commutative; a diagram chase shows the middle row exact. The top
row is the composite EAo E"; by the vertical maps with - I,,,: C"4C"
at the right, it is congruent to the negative of the middle row, which in
turn is congruent to the bottom row E'o Ec. This is the desired result.
A related and frequently used result is
Lemma 3.2. For right R-modules A< B, left R-modules A'< B',
(BIA) @R(Bf/A')=CB@RB'I/[im(A@RB')u i m ( B @ d f ) ] .
(3.3)
Proof. The first image here is that of A@ Bf+ B @ B'. This and the
symmetric map yield the exact sequence
A g R B f@ B @, A'+ B @RB'+(B/A) @,(B1/A') -to.
This sequence can also be derived (cf. Ex.2 below) from a diagram
like (3.1) with first row A@Af, A@Bf, A@ (Bf/A').
Exercises
I. I n (3.1) assume only that the rows and columns are right exact, with the
third row and the third column short exact. Prove that (3.2),with the left hand
zeros omitted, is commutative with exact rows.
2. Prove Lemma 3.1 by a diagram like (3.2) with vertical arrows reversed and
middle row A' H B'@A -t B +Cf'.
By Thm. 111.5.3, this makes Ext, (A, A) a graded ring; indeed (by
VII.3.1) a graded K-algebra. In this algebra, the elements of degree
zero form the K-algebra of A-module endomorphisms of A. We now
describe how this product can sometimes be obtained from the cohomology product for resolutions.
Let A and A' be algebras over a commutative ring K, while C and
A are left A-modules, C' and A' are left A'-modules. Write Q for A@A1,
where @ is short for
and note that C @Cf and A@Af are left
Q-modules. We wish to define a K-module homomorphism
v : E x t i (C, A) @Ext?, (C', A') +Exti+" (C @Cf, A@Af)
(4.1)
called the external or wedge product; for a Ext, and a' Ext,, we will
write v (a @a') as a v a'. Take free resolutions 8: X -t C and E' : X'-+Cf
229
Theorem 4.1. [YONEDA 1958.1 For algebras A and A' over a field
and U E Ext;(C, A), U'E Extz (C', A') the wedge product is given by
Here a @A' has an evident meaning, as follows. If k = 0, o is a homomorphism C+A; let a@A1 mean a@lA,: C@A'+A@Af. If k>O
and m>Ol a and a' are the congruence classes of long exact sequences
Since K is a field,
230
which extends in the same fashion left and right, ending with a column
C@C1on the right. The first row of vertical maps projects each (X@X1),
to the indicated one of its direct summands. The bottom row is the
composite long sequence T = (S@A') o (C @S t ), with the splice at
C@A1 displayed. The top squares do not commute, but erase the
middle row; the resulting diagram is commutative, even at the splice.
Hence the composite vertical map is a chain transformation h: X @X'
+ T which lifts the identity on C@Ct. To read off the cohomology
class of X@X' corresponding to T, take h on dimension k+m. But
h there is just
( ~ ~ x ~ ) ~ + ~ A
- ~+ Ax I ~; ~ x : ~
the cohomology class of this cocycle is exactly the one obtained from
cls f k @cls f: by the cohomology product $H. Since cls fk and cls f:,
represent S and S', respectively, this proves the first equation of the
theorem for k>O and m>O. The proof for k=O (or m=o) uses a similar
diagram, with splicing of sequences replaced by the action of a homomorphism a: C -+ A on a sequence.
The second equality in (4.2) is an (anti-) commutation rule. I t is
~ m=O. Since any long exact
immediate from the definition if k = or
sequence is a composite of short ones, it suffices to give a proof in the
case k= m= I , for short exact sequences E and E'. Here the commutative
square diagram
A@Et: A@A'-+A@Bt+A@C'
I
23 1
232
Proof. These wedge products are still given in terms of composition of long exact sequences by the formulas (4.2), where the modules
in these long exact sequences are pulled back to V-modules by y. The
Lemma asserts that the resulting V-module structure is independent
of y.
In particular, let all modules in sight be ,K; then K @K= K, and the
external wedge product becomes an internal product
ExtV(K.K) @ExtV(K,K) +ExtV (K, K)
(4.7)
which makes Extv(K, K) a graded K-algebra. Since a @ K=a, the formula (4.4) shows this algebra commutative.
Note. The external product for Tor arises from the middle four interchange
and agrees with that map for Toro= 8 ;
i t may be obtained, as in (2.5), by replacing
suitable arguments by resolutions, and composing with the homology product
and a comparison of resolutions. The external product for Ext arises similarly
from the Horn-@ interchange. Various other "products" involving Tor and E x t
arise by the same mechanism from identities on Horn and 8;for example, there
is one arising from the mixed adjoint associativity
Hom ( A @A', Horn (C, C')) -t Hom (C 8A , Horn (A', C'))
These are given in detail, via resolutions, in CARTAN-EILENBERG Chap.XI. Description in terms of the invariant definition of Tor and Ext would be of interest.
Other types of products will appear in Chap.X below.
Exercises
1. Describe how the external product in Ext commutes with connecting
homomorphisms.
3. Show that the wedge product for K a ring can still be defined, using projective resolutions, provided A and A' are projective as K-modules and TO$ (C, C') = 0
for n > 0.If, in addition, A and A' are K-flat, show that Thm.4.1 still.holds.
5. Simplicia1 Objects
233
5. Simplicia1 Objects
The cohomology H(X, 2) of a topological space X with coefficients
Z is a graded ring under a product known as the cup product. This
product can be defined not only for spaces but for other complexes
with a "simplicial" structure. Hence we now analyze the combinatorial
structure of a simplex; more exactly of a #-dimensional simplex A P
with ordered vertices.
For each non-negative integer #, let [#I denote the set {o, 1, . . ., #)
of integers in their usual order. A (weakly) monotonic map p : [q] +[#I
is a function on [q] to [#] such that i 5 j implies p i 5 p j. The objects
[#I with morphisms all weakly monotonic maps p constitute a category
A (for monotonic). Note that a monotonic p is determined by the
pq in [#] where p,=pO, .. . ;
sequence of q+ 1 integers p, 2 p15..
hence we regard p as the affine simplex (p,, . . ., pq)determined by the
on the standard psimplex AP.
vertices
Let V be any category. A contravariant functor S: d + V will be
called a simplicial object in%. Specifically, S assigns to each non-negative
integer q (to each object of A) an object S, of %, and to each monotonic
p : [q]
a morphism p* = S (p): Sp+Sq of V, with S (1)= 1 and
S ( p v) = S (v) S (p). By a sim#licial set is meant a simplicial object
in the category of sets; by a simplicial A-module is meant a simplicial
object in the category of all A-modules.
If F : V + 9 is a covariant functor, each simplicial object S in V
determines a simplicial object F S in 9,with (FS), =F(Sq), F S (p)=
F(Sp). In particular, if A is an algebra, and 4 the functor which assigns
to each set Y the free (left) A-module with generators Y, then each
simplicial set S determines a simplicial A-module 4 S.
The singular simplices (11.7) of a topological space X constitute a
simplicial set 3 (X). In detail, let Sp(X) be the set of all singular #simplices T of X ; each T is a continuous map T: AP+X defined on
the standard affine p-simplex A+. Now each monotonic p : [q]
determines a unique affine map p : Aq+AP carrying, vertex i of Aq
onto vertex pi of AP; the composite p* T= T p : Aq+X defines a map
p*
(p):
(X) +Sq (X) which makes 3 a functor on A and hence
a simplicial set. For Z the ring of integers, S'=F,~ is a simplicial
abelian group with SL the free abelian group generated by all singular
+-simplices of X. In other words, S; is just the usual group of singular
#-chains of the space X. We shall soon see that the usual boundary
of a singular #-chain is also determined by the simplicial structure of
S1(X).
I t is convenient to use two special families of monotonic maps
.s
+[#I
+[#I
=s
sp
234
ci(j)=j
=j+l
in the case of
(
)
=j-I
E')
by
for j S i ,
for j>i.
q
7:+1=
qi-l~;
i<j,
rlF1.
&j,
j-1
= Eq-17q-8,
i<j,
= I,
- i-1 j
- Eq-lqq-a
z=y,
. .
i=j+l,
i> j+ I
and q.
Lemma 5.1. Any monotonic p : [q] --+ [fi] has a unique factorization
p =& * I . . .
. .
&'vp... ?p,
(5.5)
Proof. Let the elements of [#I not in ,a[q] be il, ..., is in reverse
order, while those elements j of [q] with p (j)= p (j+ I) are jl, . . ., j,
in order. Then (5.5) holds, and presents p as the composite of a monotonic epimorphism (the product of the 7's) with a monotonic monomorphism (the product of the e's).
This lemma allows an alternative definition of a simplicia1 object.
Theorem 5.2. A simfilicial object S in a category Q is a family (Sq}
of objects of Q together with two families of morphisms of Q,
di: Sq+Sq-l,
si: Sq+Sq+l,
i=O, ..., q,
(and with q>O in the case of d,) which satisfy the identities
5. Simplicia1 Objects
235
Conversely, given the di and the s,, write any monotonic p in the unique
form ( 5 . 5 ) and define
S ( p )= sjt.. . ~ jdi..
, . . dil : Sp-t Sq.
The identities (5.6)-(5.8) suffice to commute any two of d,, s f , hence
to calculate the factorization of a composite p v from that of p and
of v , hence to prove that S ( pv)= S (v)S ( p ) . This makes S : A 4 W
contravariant.
We call di the i-th face operator and si the j-th degelzeracy operator
of S. Note that (5.6) and (5.7) imply
did,=d,di+l,
sc. s7. = s 1. sr.- 1 ,
i zj,
(5.9)
(5.10)
For example, let V be any partly ordered set (1.8);call an ordered
(q+ 1)-tuple (v,, ..., vq) with elements v , S - -.5 vq in the given partial
order of V a q-simplex of V. Let Sq( V ) be the set of all q-simplices of V.
Then S ( V ) is a simplicial set under the face and degeneracy operators
defined by
d, (v,, . . ., vq)= (v,, ..., Gi, . . . , %I
(omit vi),
(5.1 1)
(double v,). ( 5 . 1 2 )
si (vO
, . . ., vq)= (v,, ..., vi, v,, . .., v,)
Geometrically, V may be regarded as a schematic description of a
polyhedron with partly ordered vertices v i .
If S and S' are simplicial objects in a category Q, a simplicial map
a: S + S 1 is a natural transformation of the contravariant functors
S, S t : A+V. In other words, a simplicial map a is a family of morphisms a, : S, -t Si of V such that a, S ( p )= S 1 ( p ap
) for each monotonic
p : [q] -+[ p ] , or, equivalently, such that ad, = d,a and as, = s,o for every i.
The simplicial objects in V form a category with morphisms the simplicia1 maps.
Each simplicial module S determines a (positive) chain complex
K = K ( S ) with Kq= Sq and with boundary homomorphism a: K q - + K q - ,
the alternating sum of the face homomorphisms:
a=d,-h+...+
(-l)qdq: K q - + K q - l .
(5.13)
The identities (5.6) for d,dj imply that aa=O. This allows us to speak
of the homology or cohomology modules of a sirnplicial module S,
meaning those of the associated chain complex K ( S ) . For a topological
space X, (5.13) gives the usual boundary operator a in the singular
complex S ( X ) . More formally, X determines the simplicial set 3 (X)
described above, hence the simplicial abelian group F,S ( X ) , hence the
chain complex K F , (~X ); with boundary 8, this complex is the usual
singular complex S ( X ).
236
6. Normalization
Let S be a simplicial module. In each dimension n, define ( D S ) ,
to be the submodule of S , generated by all degenerate elements; that
is, set (D S ) ,= o and
( D S ) , = S ~ S ~u-...
~ u s ~ - ~ S ~ - n>O.
~ .
By the identities (5.8) for disj, D S is closed under 3, so is a subcomplex
of the associated chain complex K S of S . The quotient K SID S = KN S
is known as the normalized chain complex of the simplicial module S .
Theorem 6.1. (Normalization Theorem.) For each simplicial module S the canonical projection n: K S+KN S= K SID S is a chain
equivalence.
For the proof, we interpret the degeneracies si as homotopies. For
each non negative k, let D,S be the graded submodule of S generated
k ; that is, set
by all degenerate elements s,a with
is
( D k S ) n = ~ o S n -. l. ~- uS , - ~ S , - ~ ,
. - .VS,S,,-~,
n-ISk,
n-1>k.
t,a=(--l)kska,
=0 ,
k l d i m a , a S ,
k>dima, a e S ,
and set hk= 1- atk- t,a. This makes h,: K ( S )+ K ( S ) a chain transformation and t, : 1 = h, a chain homotopy. Since t, S Dk and aD, D,
<
h,a=a(modDS),
aeS.
<
(6.1
7. Acyclic Models
237
=z(-
~ ) ~these
d ~combine
,
to give (atk+ t, a)ska- ska (modDk-, S )
With a
for k s d i m a and hence the first inclusion of (6.2). In particular,
hoDOS=o.
Now set h=hot,h, . . . h, ... . Since h,a=a for k>dima, this composite is finite in each dimension, and defines a chain transformation
h : K S + K S . By (6.1), h k D S < D S , so an iteration of (6.1) gives
ha= a (mod D S ) .
(6.3)
7. Acyclic Models
238
239
for ~ E A , ~, E B , and
, n>0 in the case of d,. To avoid confusion with
the tensor product of complexes we shall write a x b for the element
a @ b of A,@ B,. For simplicial sets U and V , this definition insures
that there is a natural isomorphism of simplicial modules
The proof, as recorded in the following lemmas, will use the method
of acyclic models. Note that KO( A xB) =Ao @ Bo= KO( A )@KO( B );
hence we can choose maps f and g in (8.4) to be the identity in dimension
zero.
240
complexes (Prop. 7.1). Hence there is in this complex a chain c of dimension n with a c = e. We set f , (x"xxn)= c, so that
Now consider aA,, bc B, . By Prop. 7.2, there are simplicia1 maps
a : Mn+A, ,!?: M n +B with uxn=a, ,!?xn=b. Then K ( a ) : K(Mn)+
K ( A ) is a chain transformation which we again denote as a, and
a @,!?: K(M n )@ K(M n )-tK ( A )@K ( B ) is a chain transformation. Set
fn ( a xb)= (u@I,!?) c, for c as in (8.6); since the simplicial maps a and @
are unique, the right hand side is bilinear in a and b, so defines
fn : K , ( A x B) -t [ K ( A )@ K ( B ) ] ., Moreover,
Now fn-,
is natural, so
at, ( z n x x n =
) ~ ( x ~ x x* ~) ' ( x ~ x-xtn-l
~ ) a ( x ~ x x.~ )
By the induction assumption, a(/- f'- ta)=o, so the right hand side
is a cycle in an acyclic complex, hence is the boundary of some chain d.
Set t,(xnxxn)=d, tn(axb)=(u@,!?)dfor a, @ with ax*=a, jxn=b.
The previous type of argument then shows t, natural and at,+ t,-,a=
f - f' for all a x b .
24 1
the models Mfi and Mq and maps a : Mfi -+ A , /3: M4 -+B with a xfi= a,
Now the complex K(MfixMq)is acyclic, for the homotopies s
of (7.1) for Mfi and Mq yield a contracting homotopy s ( a x b )= s a x s b
on K(MfixMq).Using this acyclicity, the construction of g proceeds
as before.
We now have the chain transformations f and g of the theorem;
it remains to establish homotopies = f g, 1 =g f . These are done by
exactly the same method; for instance the homotopy 1=gf in K ( A x B )
is obtained, using the acyclicity of K(MfixMP),by comparing h=1
with hl=gf as follows.
/3 d = b .
In af ( x n x x n ) the last face of each first factor cancels with the term
arising from the initial face of the second factor, and the remaining
terms assemble to give f i3(xnxxn),as required.
The chain transformation f of (8.7) is known as the AlexanderWhitney map, since it appeared in the simultaneous and independent
definition of the cup product in topology by these authors. The explicit
map f calculated from our contracting homotopy differs from the
Mac Lane, Homology
16
242
Corollary 8.6. The Alexander-Whitlzey map f induces a chain transformation on the associated aormalized chain complexes,
243
(-I)"(")
v)
(i0,
...,i,+,)x(i0,
..., ifi+,),
where the first factor is degenerate at those indices k for which the
second factor is not degenerate. This symbol may be read as the (#+ q)dimensional affine simplex with vertices (i,,jk) in the product A k A q .
16*
244
Corollary 8.9. The shuffle map g of (8.9) induces a chain transformation ON the normalized chain com$lexes
Exercises
Exhibit a second explicit formula for f, with first and last faces interchanged
in (8.7).
2. Establish associativity for the shuffle map g.
3. Prove the normalization theorem of 6 by the method of acyclic models.
4. Show that the EILENBERG-ZILBER theorem holds for A a simplicial right
R-module, B a simplicial left R-module, and R any ring.
5. Calculate the integral homology of a torus S1xS1from that of a circle S1
(EILENBERG-ZILBER plus KUNNETH).
1.
9. Cup Products
For any simplicial set U, Au = u x u defines a simplicial map A: U-t
U x U called the simplicial diagolzal map. Now U determines the simplicial abelian group F, U and hence the chain complex K(F,U) which
we write simply as K(U); each K,, (U) is the free abelian group generated
by the set U,, with a = (- l)'di. The diagonal induces a chain transformation K(U)+ K ( U x U), also denoted by A . If f is any one of the
natural maps from the EILENBERG-ZILBER theorem the composite
9. Cup Products
245
Hk(U,A)@H'"(U,A')
lfiE
Y?
(9.3)
H ~ + " ( K ( U@K(U),
)
A@A')% H ~ + " ( uA
, @Ar),
where p H is the cohomology product of (1.3), is called the (external)
simplicial cup product. With cochains h and h', the definition reads
(cls h) u (cls h') = cls (h u h'), where
( h u h1)u=(h@h')f Au,
(9.4)
for h@h' as in (1.4). In particular, if U = S ( V ) is the simplicial set
associated with a partly ordered set V of vertices and f is the AlexanderWhitney map, while heHk, ~ ' E H " -then
~,
(h u h') (vo,..., v,) = h(vO,..., vk)@h'(vk,.. ., v,) .
(9.5)
If A =A' is the additive group of a commutative ring R with product
z: R @R +R, the composite n, u is a map
Hk (U, R) @ HM( U , R) +H~+" (u,R)
called the internal simplicial cup product.
(9.6)
246
O S i S n , n>O,
0 Si l
;n,
r=O
We have recovered the homogeneous description (IV.5.13) of the unnormalized bar resolution B (17)= KF, (B (17)),while K,F, (B(17)) is the
normalized bar resolution B (17).
Now recall that the group ring Z(17) is a Hopf algebra with coproduct
y : ZO+Z(17)@Z(17),
y(x)=x@x.
By pull-back along the corresponding diagonal map 17+ 17xl7, the
cartesian product 8 (17) x B (17) of two II-sets is a II-set. The diagonal
map w for (17)= KB (4is the composite
(9.7)
9. Cup Products
247
248
Exercise
1. Show that 8 (17) with non-homogeneous generators (IV.5. I I ) has degeneracies and faces given by
s i ( x [x,,
. . ., x,])
= x [x,,
d i ( x [ x l , ..., x n l ) = z 2 - ,
=x[xl.
=x[x1.
. . ., xi-1> I , x i , . . . , xn1,
OdiSn,
[ x 2 , ..., %,I,
i=o,
O<i<n,
i=n
Notes. For topological discussion of the cup product (in contrary terminology)
see HILTON-WYLIE [I9601 .For the cup product for groups see EILENBERG-MAC
LANE [1947]. ECKMANN [1945-19461, [ I 9 5 4 1 A fiber space may be regarded
as a sort of "twisted" cartesian product; there is a corresponding twisted version
of the EILENBERG-ZILBER theorem (BROWN [ 1 9 5 9 ] , GUGENHEIM [1960], SZCZARBA
[ 1 9 6 1 ] ) . Simplicia1 fiber bundles are treated in BARRATT-GUGENHEIM-MOORE
[I9593
Chapter nine
I. Additive Categories
249
(1.1)
B : B +C,
Identities: If a : A+ B, then
Zero: There is an object 0' such that hom(ol, 0') is the zero group.
250
A,&B=A,
with
n,
% h = l ~ l r na~a=1.4,, h q + L , ~ , = I ~ .
(l.4)
To avoid foundational difficulties, two further axioms of a set-theoretic
character are required; they will be stated at the end of this section.
These axioms are like those for a category (1.7). Indeed, an additive
category may be defined as a category with zero and direct sums, as
above, in which each set hom(A, B) of morphisms has the structure
of an abelian group such that the distributive laws
are valid whenever both sides are defined. (This insures that composition is bilinear, as required by (1.I).)
If the existence of direct sums is not required, we speak of a preadditive category. As in the case of categories, we can omit the objects
and work only with the morphisms, using the identity morphisms jA
in place of objects. The axioms are then like the axioms for a ring in
which the compositions al+a, and /?a are not always defined but,
whenever defined, satisfy the usual ring axioms such as (1.2), (1.?),
and (1.5). Thus HILTON-LEDERMANN [1958] call a preadditive category
[1954].
a rimgoid, following the terminology of BARRATT
By 0 we denote (ambiguously) the zero element of any group
hom (A, B); then 0 cr = 0 = /? 0 whenever defined (proof: 0 a = (0+ 0) a ;
use the distributive law). An object 0' with hom(O', 0') the zero group
is called a zero object. Then 1 ,= 0, hence hom (A, 0') and hom (O', B)
are the zero groups whatever the objects A and B, and any two zero
objects are equivalent.
Examine next the consequences of the finite direct sum axiom.
By (1.4),
n , ~ , = n ~ ( h q + ~~~=17d1~,+n,dl~al
~n~)
=nlb+%c11s1
hence 7611,=0 and n,~,=o, as usual. Props.4.1-4.5 of Chap.1 now
follow; in particular, the diagram (1.4) determines the object B up to
equivalence, and we usually write such a B as A, @ A,. Each morphism
y: A,@ A2+C determines a pair of morphisms yj=y L,.:Ai+C; the
correspondence g,(y)= (y, , y,) is an isomorphism
I . Additive Categories
251
Given two direct sums A, @A, and A; @ A; and morphisms ai: A,. -+A;
there is a unique morphism ul @a,: A, @ A, +A; @A; with
252
while y a = 0
implies y = y l a
(1.11)
for some y', necessarily unique. The left annihilators of a are thus the
left multiples of a cokernel a of a. Any two cokernels of a are left equivalent; if a has a cokernel, the class of all cokernels of a is a quotient
1. Additive Categories
253
a B epic + a epic,
x monic + ker (x a)= ker a , a epic + coker (a a)= coker a .
lAEker0 and I B E
This implies a rnonic and ,!? epic, so we may read "xlla" as " x and a
are the morphisms of a short exact sequence".
To keep the foundations in order we wish the collection of all subobjects of an object A and the collection of all extensions of A by C
both to be sets and not classes. Hence, for an additive category we
assume two additional axioms:
Sets of sub- and quotient objects. For each object A there is a set
of morphisms x, each rnonic with range A, which contains a representative of every subobject of A and dually, for quotient objects of A.
Set of extensions. For each pair of objects C, A and each n 2 1 there
is a set of n-fold exact sequences from A to C containing a representative
of every congruence class of such sequences (with "congruence"defined
as in 111.5).
Both axioms hold in all the relevant examples.
Exercises
1. Prove: If 0: A+ B is rnonic, then A is a zero object, and conversely.
q,-I
(yl, yz)=
vc (yl @ yn).
254
2. Abelian Categories
d is a ficnctor.
2. Abelian Categories
25 5
with row and column short exact sequences. Here "ker a" of course
stands for any morphism in the class ker a. With the same convention
we may read off the relations
(ker a) 1 (coim a),
coim a = coker (ker a),
ker a =ker (coim a),
(2.4)
(2.5)
(2.6)
256
with xllo and x'lla', we wish to prove that a, y monic imply /3 monic,
and dually. But take ,UE ker /3. Then /3 ,u= 0 gives 0=01/3 ,u= y a ,u;
since y is rnonic, a p = 0. This implies that ,u factors through x~ ker a
as ,u= x Y , for a v which is necessarily rnonic. Then x'a v =/3 x v =/3 ,u= 0.
But x' and a are rnonic, so v = 0 and thus ker /3=p =x v=0, so /3 is
rnonic, as desired.
The Five Lemma, the Four Lemma, and the 3 x3 Lemma also hold
in an abelian category. The proofs, which depend on certain additional
techniques, will be given in Chap. XII.
Call an abelian category selective if
(Select I). There is a function assigning to each pair of objects
A,, A,, a direct sum diagram, of the form specified in (1.4).
(Select 2). There is a function selecting a unique representative x
for each subobject and a unique representative o for each quotient
object.
3. Categories of Diagrams
257
Exercises
I . Given (Abel-2) and (Abel-3). show that (Abel-I) may be replaced by the
weaker statement that each epic has a kernel and each monic a cokernel.
2. I n ( 2 4 , show
El
monic implies
r],
monic and
lamonic
implies r ] , rnonic.
3. Categories of Diagrams
17
258
C, C' and three morphisms lc, iC,,and y: C +C', then each T: yb+d
is determined by a morphism T(y) in d , so that Dgram (Wo,d)is the
category Morph(d) of $2, with objects the morphisms of d.
Proposition 3.1. (GROTHENDIECK [19571.) If the category Q is small
and d is abelian, 9=Dgram (V, d)is an abelian category. I f f and g
are mmfihisms of 9 , f Ug ilc 93 if and onZy if, for each C, f (C)llg (C) in d.
Proof. Let f : T+S be natural. Since is small, we can choose
for each object C a monic k ( C ) ~ k efr(C), with domain, say, K(C).
Thus k (C): K(C)-t T(C) is a morphism of d . Since f is natural, each
y: C +C' gives a commutative diagram
with exact rows. Since f(Ct)[T(y) k(C)]=O, and k(Ct) is the kernel
of f (C'), there is a unique K(y) (dotted arrow) with T(y) k (C)= k (C') K ( y ).
It follows that K: W + d with mapping function K(y) is a functor
and k: K+T natural. As a morphism of 9 , k is rnonic, for if kh=O,
then (k h) (C)= k (C) h (C)= O for each object C; since k (C) is monic
in d , h (C)=O. Furthermore, if g: R +T is natural with fg=O, each
g (C) factors uniquely through k (C) as g (C)= k (C) h (C), h: R +K is
natural, and g= k h. Therefore k E ker9 f. This argument with its dual
proves (Abel-1) in 9 and also gives
f rnonic in 9 w each f (C) monic in d,
k ~ k ef r ~ each k(C)~ker,,f (C).
These statements with their duals prove (Abel-2).
To get a standard factorization (Abelq) for f : T+S, choose for
each C a standard factorization f (C)=I (C) t (C); the range R (C) of
t (C) yields a functor R : V - t d , t: T+R is epic and I: R +S monic
in 9 , and f =I t. Since Q is small, we can also select for each T a set
of representatives of the subobjects of T and for each S and T a set
of representatives for the extensions of S by T, thus proving that 9
satisfies the supplementary set-theoretical axioms ( 1) for an additive
category.
Next consider the diagrams which involve zero objects. In any category V call an object N a null object if for each object C of V there is
exactly one morphism C+N and exactly one morphism N-tC; write
' oc: C + N
and oC: N -tC for these morphisms. Any two null objects
in V are equivalent, and any object equivalent to a null object is null.
3. Categories of Diagrams
259
the objects of O are all integers n plus a null object N ; the morphisms
are all identities, the null morphisms n+N, N - w , and n-tm, and
morphisms a, : n +(n- 1). The composite of morphisms is defined by
requiring a,-,a, to be null. Take any abelian category d.A normalized
covariant functor T: O - t d is then given by
tT,-,tT, -+.. ., a
sequence of objects and morphisms of d , with 8,-,a,=O,
so is just
A natua chain complex of objects from a?' (in brief, an &'-complex).
ral transformation f : T-tS is a chain transformation. Therefore
Dgram,(O, d ) is the category of all d-complexes; by Prop.).l it is
an abelian category. If a?' is selective, the homology objects H, (T)=
Ker a,/Im a,, may be defined as usual; the reader should show that
each f : T+S induces f, : H,(T) +H, (S), so that H, is a covariant
functor on this category, and that homotopies have the usual properties.
Exercises
1. If d is abelian, show that the category of graded objects of d is abelian.
2. Describe an abelian category whose objects include the analyses (2.1).
5 ( VA@ I A ) AA= 0
that
260
261
262
and E' : Xi +C' an allowable epic, ye: Xo-+C factors as ~ ' f ,for some fo.
We next wish to construct f, so that the diagram
will be commutative. Take a standard factorization a'=l a as displayed; since X' is allowable, I be'. But E' fOa= y E a= 0,SO fOa factors
' foa=l$ for some p. Now a is an allowable epic
through l ~ k e r e as
and Xl an allowable projective, so = a fl for some fl, and a' fl= l a f, =
l p= foa, as desired. The construction of f,, f,, ... and of the homotopy
is similar.
Note on injective envelopes. A family {at) of subobjects of A is directed by
inclusion if each pair of subobjects a,, at of the family is contained in a third
subobject of the family (in the obvious sense of "contained, as defined in Chap. XI1
below). An abelian category d satisfies GROTHENDIECK'S axiom AB-5 if for each
A, each subobject b, and each family at directed by inclusion, bn(Utat)=Ul(bnal)
holds in the lattice of subobjects of &and if d h a s infinite direct sums. An object
U is a generator of d if t o each non-zero morphism a : A+ B there is a morphism
5 : U + A with
0. Both conditions hold in the category of all A-modules,
[1957, Thm.l.lO.l] shows that an abelian
with A a generator. GROTHENDIECK
category with AB-5 and a generator has enough injectives; MITCHELL [I9621
constructs the ECKMANN-SCHOPF injective envelope under these hypotheses. I n
particular this shows that there are enough injectives in the category of sheaves
over a fixed topological space (though in this case there are not enough projectives) :
See GROTHENDIECK
[I 95 71. GODEMENT
[I 9581.
Exercise
1. (Characterization of allowable short exact sequences by allowable projectives [HELLER 19581.) If d is an allowable class of short exact sequences satisfying the condition U B E ~ ~a&,,
J and if there are enough allowable projectives,
show that an epic a: B +C is allowable if and only if hom (P. B) +hom (P,C)
is an epimorphism for all allowable projectives P.
263
AZ.BGC
(5.2)
(solid arrows for the & part, solid and broken arrows for A).Similarly
the equations
264
Proof. The equivalence of (i), (ii), and the allowability of a is immediate, by Prop.4.1. If til= 1 and a k = 1, as in
and v a v = k t l o k t =
then v = k t : B.+A has ava=iloktila=ila=a
k t = v ; this proves (i) +(iv). Trivially, (iv)+(iii). Finally, to get (iii) +
(i), assume aua=cc. and set a=Ao. Thus ilouila=ilo with il monic
(in A!) and o epic implies ou il= 1, so 3, has a left inverse ou in A
and o a right inverse uil.
If X is an &-complex in the sense of 3 (X, objects and a, morphisms in &) then X is an A-complex; since is exact, it follows that
q [H, ( X ) ]=H, (0X).
Theorem 5.2. If X i s a n d - c o q 5 l e x (not necessarily fiositive) then
X has a contracting homotofiy s with as+ s a= 1 (and each s,: X,---9
X,+, a morfihism of A) if and only if all H,(X) vanish and all boundary
homomor+hisms 8 are allowable. W h e n these conditions hold, s m a y be
chosen so that sa= 0.
6. Relative Resolutions
265
a,:
Exercises
(The first three exercises deal with the absolute case d = J . )
*+
X,+
is a direct sum representation of each X, .
and only if (im a,+,, coim a,) :
When these conditions hold, there is an s with sa= 0.
2. A complex X of modules has a contracting homotopy if and only if, for
each n, the module of n-cycles is a direct summand of X,.
3. A (not necessarily positive) complex X of free abelian groups has homomorphisms s : X, +X,+l with as s a = 1 if and only if all H, (X) vanish.
4. Deduce Thm. 5.2 from the result of Ex. 1.
6. Relative Resolutions
266
(M,
A) ;
this last property states that the functor F: A+& is a left adjoint
[KAN 19581 of the functor 0 : &-+A (see note below).
Conversely, the conditions (i) and (ii) for a resolvent pair may be
has a left adjoint F.
replaced by the requirement that the functor
Indeed, this requirement means that there is a natural isomorphism
in
(of abelian groups). Take A = F M in this isomorphism; then I,
the group on the left gives q (IFM)= eM: M--+17F M . That e: I, -- 17F
is a natural transformation follows by taking any p : M-tM' and applying g, to the diagram
homd(F M, F M)% h o m d ( ~M, ~ ~ ' ) K h o r n . . , , ( ~M', F M') .
Next take any A and any a : F M +A. Since g, is natural, the diagram
homd(F M, FM)% hornd (M,
k*
F M)
k*
Again, for each K-algebra A there is a resolvent pair with d the left
M (Prop. VI.8.2). Other
A-modules, A the K-modules, F(M)=A BK
examples of resolvent pairs appear in Ex. 2 below.
6. Relative Resolutions
267
268
i n d as s-,=e(OC)
and sn=e(o/3,C).
of d which make p (92, C )= {p, (92, C ) } a relatively free allowable resolution of C with s as contracting homotofiy in d. This resolution, with its
contracting homotofiy, i s a covariant f ~ n c t o rof C.
We do not claim sa=O - because it usually isn't so.
6 . Relative Resolutions
269
The comparison theorem shows that we can equally well use any O-split
relatively projective resolution E : X +C to calculate Extg = ExtL as
Extg (C, A) s Hn(homd (X, A)) ;
(6.4)
note that, in each dimension n, homd (X,, A) stands for the group of
all morphisms 6:Xn+A in a2 - not just the allowable ones. In particular, Ext; (C, A) =hornd (C, A). Replacement of C by a short exact
sequence E will give the usual long exact sequence for Ext n , as in
Thm.III.9.1, provided E is 0-split. The analogous result holds if A
is replaced by a O-split short exact sequence; the proof uses either an
exact sequence of resolutions (Ex.III.9.1) or the assumption that there
are enough relative injectives. These long exact sequences are actually
valid in any relative abelian category without the assumption that
there are enough relative projectives or injectives. The proof, to be
given in Chap.XI1, depends on the interpretation of Extb (C, A) as
congruence classes of n-fold, O-split exact sequences from A to C.
In particular, Extb , unlike ExtL , depends on .
Note on Adjoints. If Q and d are categories, a functor T : Q + d is called
a right adjoint of S : d+ Q if there is a natural equivalence
horn& (A. T(C))
=hornye ( S (A), C) ;
here both sides are bifunctors of A and C with values in the category of sets (or,
if Q and d are additive, in the category of abelian groups). For example, adjoint
associativity
Horn ( A B B , C) =Horn (A, Horn (B, C))
states for fixed B that T(C)= Hom(B, C) is a right adjoint t o S(A)= A@B. There
are many other examples [K A N 19581.
Exercises
1. If the relative abelian category q is a resolvent pair of categories for two
functors F and F', show that there is a unique natural isomorphism 7 :F-tF'
with r] e = e'.
3. I n case (b) of Ex. 2 show that the allowable exact sequences and the relative
E x t functor are identical with those for W = (R, S) when S = QR'.
270
F (M)=
(7.1)
hornn (M,
A).
271
of A by s-,= e (O C) and
(7.5)
since s,-, = e p with # epic, (1- s,-, a,) e = 0. By Lemma 7.1, 1- s,-, a,
factors as ( I - s,-, a,) =as,, which gives a,, = a satisfying (7.7). These
morphisms E , a, are uniquely determined, again by Lemma 7.1. Moreover, (7.7) gives
272
To show that this description of the bar resolution agrees with the
previous usage for a group IT, take a? to be the category of left 17modules, d that of abelian groups, F ( M )= Z (IT)@ M , and e, (m)=
1 @ m for each mc M . This gives a resolvent pair of categories. Now the
sequence Z - Z ( I 7 ) +2(17)/Z of free abelian groups is exact, hence so
is its tensor product with M :
Therefore F ( M )r[Z ( D ) / Z ]@ M . Take for C the trivial 17-module Z .
Then
Fn(2)= [Z (17)/Z]@. . - @ [Z (IT)/Z], n factors.
But Z(IT)/Z is the free abelian group with generators all xS.1 in IT.
Hence F S ( Z )may be identified with the free abelian group generated
by all symbols [x,( .. . I x,], with no xi in l7 equal to 1 . Then B , (W, Z )=
Z (17)@Fn (2)is the free abelian group with generators all x [x,) . . I x,]
with x d 7 , while the map s= e p : B,--+ B,,, defined above becomes
Exercises
I . Show that the long sequence (7.4) is exact in A.
With this result the following exercises, suggested t o me by J. SCHMID, will yield
the cup product reduction theorem as stated in VlII.9 (cf. SCHMID [1963]).
273
7. From a presentation I7= FIR with F a free group obtain a group extension
E : Ro n B'+I7, where [ R , R ] is t h e commutator subgroup o f R. Ro = R / [ R ,R ]
and B'= F / [ R , R ] .
8. T h e characteristic class x o f t h e group extension E, described as i n IV.6.
is a two-fold 2-split extension o f Ro b y Z . Show t h a t t h e intermediate module M
i n x is free; specifically, let F be t h e free group o n generators g, S t h e free I7module o n corresponding generators g' and show t h a t g'+ [cls g]E M is an isomorphism S z M . ( H i n t : Use Lemma IV.7.2 t o construct an inverse.)
9. Let A be a IT-module. Show t h a t t h e iterated connecting homomorphism
, ) for t h e relative e x t
for x yields an isomorphism E x t n ( R o , A ) ~ E x t " + a ( Z A
functor and n > o and hence, b y Ex.6, isomorphisms
H n + 2 ( 1 7 , A ) ~ H n ( L 7 , H o m Z ( R o , A ) )n>O,
,
H a (IT, A ) GE Coker[Homn(M, A ) +Homn(R0, A ) ] .
10. For
>o
(R)
.
R ~ B(R)
O +A (R)
by @n(R)=R@Rn@R=R"+2,with n + 2 factors, ~ ( r ~ @ r ~ ) =@
r~r~
short for &, and
+a
) =
i=O
(- l i 0
@rjri+l
(8.1)
s(~o@..*@Y~+I)=~@~o@...@~~+I*
18
2 74
Corollary 8.2 For each left R-module C, j3(R) @,C is the bar resolution P(C) for the resolvent pair (R, S). Symmetrically, for each right
R-module G, G @, j3 (R) with contracting homotopy t is the (right) bar
resolution j3 (G).
Proof. Since R @, C s C , one forms j3 (R) @, C simply by replacing
C
the last argument rn+, in (8.1) and (8.2) above by c E C. Then j3, (R)
= F F ( C ) , the contracting homotopy s of (8.2) is that of (6.2), and a
is the unique boundary with s as contracting homotopy, by Thm.6.3.
In particular, j3(R) itself is just the bar resolution (left or right) of the
R-module R.
Observe that the boundary operator (8.1) in @(R)is the alternating
sum of the face operators di: j3,,+/?,-, defined by
(8.3)
di(ro@...@rn+l)=ro@...@ri-1@riri+l@ri+~@...@rn+l,
are
(8.4)
the usual identities for di and s j hold, and j3 (R) is a simplicial R-bimodule
in the sense of VIII.5. The reader may show that the simplicial normalization of j3 (R) yields the normalized bar resolution B (R).
Take R-modules GRandRC. The (absolute) torsion products T o e (G, C)
are calculated from a projective resolution E : X + C as Hn(G g R X ) .
In the present relative case, j3(R) @,C provides a canonical and functorial resolution, so we define the n-th relative torsion product as:
si(~~@...@rn+~)=r~@.*.@ri@l@ri+l@...@rn+l,
G@Rj3(R)@RA-G@RB(R) @RB+G@RB(R)@RC.
The resulting connecting homomorphisms
n >0,
Em
275
just as for the absolute torsion product, except that here E must be
S-split. If E': G n K -wL is an S-split short exact sequence of right
R-modules, the same argument (interchanging left and right) gives
natural connecting homomorphisms
EA: ToriRpS)(L, C) +Tori%:) (G, C),
n>O,
(G, C)
(natural) ,
(8.7)
(8.8)
n >0,
when P' and P are (R, S)-#rojective right and left modules, respectively.
If E' and E are S-split short exact sequences of right and left R-modules,
respectively, the corresponding connecting homomorphisms are natural and
yield long exact sequences (8.6)) and symmetrically.
In particular, this Theorem leads to a characterization of the relative
torsion products as functors of the second argument, by properties
(8.7), ( 8 4 , and (8.6), just as in Thm.V.8.5; for this purpose we may
replace "(R, S)-projective" by "(R, S)-free" in (8.8).
We need only prove (8.7) and (8.8). First Bl (R) +So (R) +R -to is
exact; since tensor products carry right exact sequences into right
exact sequences, so is
276
Theorem 8.5. If E : Y+G is an S-split resolution of the right Rmodule G by (R, S)-projectivemodules Y,, there is a canonical isomorphism
c,
(8.10)
+.
277
EX^:^,^,
Exercises
The first six o f the following exercises are taken from HOCHSCHILD
[1956].
1.
-t
9. Show that B(R) is the (unnormalized) bar resolution for the resolvent
pair 9Z' with d = R-bimodules, A = S-R-bimodules, F ( M ) = R B S M and e (m)
I Bm.
10. For 1' as in Ex. 9, show that ExtTR,~)
(C, A ) r Ext%,(R, HomZ(C,A ) ) and
~ o r f .(G,
~ C)
)
~ o r ( fR ,C QZ G ) . Here C BZG is the bimodule with r (c B g ) = rc a g ,
(cB g ) r = c Bgr.
278
The direct product R=R1xR" of two rings is the ring with the
additive group R f @R" and multiplication (r;, r;') (ri, r',')= (ri ri, ryry).
(This is just the direct product of R' and R" as 2-algebras,as in (VII.5 . I ) ) .
Each left R'-module A' is an R-module ,lA' b y pull-back along the projection n,: R ' x R"+ R', and similarly for R"-modules. In particular, R'
and R" are left R-modules, while the termwise definition o f the product
in R shows that R r R 1 @R" is an isomorphism of R-modules, and
hence that R' and R" are projective R-modules.
Lemma 9.1. If the R'-modules C' and G' and the R"-module A"
are regarded as R =R ' xR" modules, then
Gf@,A"=O,
HomR(C1,A")= O .
(9.1)
(9.2)
279
(9.4)
(9.5)
valid for all n. The same isomor#hisms hold with S, S', and S" omitted.
Proof. First observe that an (R', S1)-freemodule R'@s>M' is also
(R, S)-projective (though not necessarily (R, S)-free). For, the left
St-module M' is a left S-module by pull-back, and, using the pull-back
lemma,
R @, M1rR'@,M' $ R"BS M'gR'@s,M' $ R"@s.. M'.
Since R @, M' is (R, S)-projective, so is its R-direct summand Rf@,, M'.
Now choose relatively free split resolutions E' : X'+C1 and E" : X + C"
of the components of C. Then E'$ E " : X'$ X"-+C1$ C" is a resolution
of the R-module C1$C" which is S-split by the direct sum of the S'
and S" contracting homotopies for X' and X . By the first observation,
each term XA$X: is an (R, S)-projective. By (9.2) and (9.3) for X =
Xf@XI',
Hom, (X, A )=HomRt(X1,A') $ Hom,,,(X", A"),
280
Chapter ten
(14
These equivalent descriptions are presented in terms of the bar resolution for algebras, which is given explicitly in $ 2 - it is a special case
of the bar resolution (IX.7) for a resolvent pair of categories. This
chapter examines the properties of H , and H" and develops similar
(co)-homology for graded and for differential graded algebras, as well
as for monoids and for abelian groups.
2. The Bar Resolution for Algebras
Let A be an algebra over K. The identity element lAgives a Kmodule map I: K+A; its cokernel A/I(K)=A/(K I A ) will be denoted
(simply but inaccurately) as A/K, with elements the cosets A+ K. For
each left A-module C construct the relatively free A-module (@=
281
So
B,---+B,+, are
Theorem 2.1. For each left A-module C, E : B (A, C) +C is a resolution of C b y (A, K)-relatively free left A-modules which is K-s@it b y the
contracting homotofiy s with s2=0. Moreover, B(A, C) is a covariant
functor of C.
This can be proved directly from the formulas above. Alternatively,
apply the resolution of IX.7 for the resolvent pair of categories 9 with
d =left A-modules, A= K-modules, F(M) = A @M , e (m)= 1@m. Since
K ---+A
---+A/K--+
o is a right exact sequence of K-modules, each K-module
M yields a right exact sequence
282
/? (B,
C) =/?
(A, C)
for i=O, ..., n. With these si and di as in (2.7), /?(A,C) is the associated
chain complex of a simplicia1A-module and 11 is the simplicial normalization of Thm. VIII.6.1.
For the bimodule bar resolution B(A, A), take C above to be A.
Each B, is then a A-bimodule; formula (2.5) with c replaced by ~ ' E A
shows that E and each a, is a A-bimodule homomorphism. Similarly,
s of (2.4) becomes a homomorphism of right A-modules. Hence
283
s (A,). Thus B (A) +,K is a K-split, (A, K)-free resolution of the left
A-module ,K. In particular, when K=Z and A = Z ( n ) , this is the bar
resolution of IV. 5.
The redzcced bar resolution for an augmented algebra A is the complex
B (A)= K, @,gB(A), so B, (A)= K and 73, (A)= (All<)" for n>O. The
contracting homotopy does not apply toB, but the formula for the boundary still applies, with c and the left operator A omitted: In (2.5) replace
) A,c by E (A,,).The "reduced bar resolution"
the operator A, by ~ ( 4and
is not a resolution, but is useful for computations. The left and the reduced bar resolution can also be formed without normalization.
Exercises
I. For an augmented algebra A, let X be any relatively free K-split resolution
of ,K by left A-modules. Show that the canonical comparisons (Thm.IX.6.2)
q,: B (A) +X, y : X +B (A) over the identity satisfy q, y = 1. 2. (CARTAN.) For A as in Ex. I, show that the left bar resolution B(A) is
characterized up to isomorphism as a K-split resolution X of ,K with a contracting
homotopy s such that sB= 0 and X, EA@sX,-, .
3. The normalization theorem can be proved directly. Show that a bimodule
chain transformation 5: B (A, C) + B(A, C) with 6 5 = E can be defined recursively
where e,=e(FOC). Prove that T ~ = Iand
, by
with [,=I, [,e,=s[,-lae,,
similar means construct a chain homotopy [q a I , all for r] as in Cor. 2.2.
4. For left A-modules C and A , show that the I-cocycles of the cochain complex HomA(i?(A, C), A ) can be regarded as factor sets for K-split A-module extensions of A by G.
n =0 I
... ;
(3.1)
it is a covariant functor of A. Here HornA-,., stands for bimodule homomorphisms. According to the normalization theorem we can replace
the bimodule bar resolution B(A, A) here by the un-normalized bar
resolution /? (A, A). Both B (A, A) and /?(A,A) are right A-split (11-11,
K-A) relative projective resolutions of the bimodule A, and also are
K-split (A-A, K) relative projective resolutions of A, so HU(A,A) is
the n-th relative Ext functor in either case, as stated in (1.3).
We call H n (A, A) the Hochschild cohomology modules of A, since
[I9451 using exactly the
they were originally defined by HOCHSCHILD
formulas given by the bar resolution with K a field.
The complex Horn,,-, (B (A, A), A) used in (3.I) may be described
more directly. Consider K-multilinear functions f on the n-fold cartesian
product A x . .. x A to A ; call f rtormalized if f (il,, . . ., A,) = 0 whenever
284
one 4 is 1. For example, the function [All . . . An] c of (2.3) is K-multilinear and normalized. The universal property of the tensor product
B, ( A , A)=A @ (A/K)"@ A states that each normalized K-multilinear f
determines a unique bimodule homomorphism
B n ( A ,A)-+A such
that always
j ( m [ i l 1 l ... /A,I
..., 2,).
7:
m)=f(~
H O ( A A)=AA=
,
[a1ila=aA for all i l A ] .
(3.3)
r,
r.
285
(k,&+k,Az)=k,~(&)+kzu(A,),
kieK-
With the factor set f for u defined by (3.9, the rule (3.6) gives
(3.7)
286
As the product in
r is associative,
r;
r
The projection # has kernel Ja, while a' has kernel J/J8, hence is a singular extension of A. By the case n=2, a' is cleft by some g ~ .Now
f1 (p,A) =I" is a subalgebra of
and # induces fi' : I " + g ~ A n i with
kernel Ja. Since (J8)"-l< J"= 0, the induction assumption shows #'
cleft by some I#, so a is cleft by L ~ I ' Y .
This result includes the Principal Theorem of Wedderburn for an
algebra I' of finite dimension (as a vector space) over a field. Each
such algebra has a two-sided nilpotent ideal R, called the radical, such
that r / R is semi-simple. The Wedderburn Theorem asserts that if r / R
is separable, then the extension F - t r / R is cleft. This follows from
r,
287
Thm. 3.2, for FIR separable implies (Thm.VII. 5.6) bidim r / R= 0 , hence
bidim r / R S I , hence H2(r/R,A) = O for all (r/R)-bimodules A , hence
r + r / R cleft.
Note. For algebras of finite dimension over a field, Thm. 3.2 is also valid
without the hypothesis that the kernel is nilpotent [HOCHSCHILD 1945, Prop.6.11;
[ROSENBERG-ZELINSKY 19561. The obstruction problem for the construction of
non-singular K-split extensions with a given kernel [HOCHSCHILD 19471 leads to
an interpretation of -(A, A) parallel to that for groups (IV.8). Extensions which
are not K-split require a second, additive, factor set in place of the linearity of u
in (3.7); we return to this question in 3 13.
Hn (A, A),E
~ t z (A,
- ~A )
(if A is K-projective)
288
Using B for p, the same result holds for A/K projective as a K-module. CARTANEILENBERG define the "Hochschild" cohomology by the absolute E x t functor
in all cases, so their definition does not always agree with ours.
Exercises
Show that A" is a sub-A-bimodule of A when A is commutative.
2. Construct a "Baer sum" of extensions of A by A so t h a t the correspondence
of Thm.3.i maps the Baer sum into the sum in H2(A, A ) .
3. Show t h a t *(A, A ) is the group of congruence classes of those bimodule
extensions A w B +A which K-split.
4. Show explicitly t h a t each short exact sequence A-B +A of bimodules
which is K-split is also split as a sequence of right A-modules.
5. If A is an augmented algebra and M a K-module then the cohomology of
M , pulled back to be a bimodule, may be calculated from the reduced bar resolution
as Hn(A, ,Me) 1.~ " ( H o m(B(A),M)).
1.
I3 (
0
,I . . ., .
AB ( AA)),
(
A A) (
(4.2)
As for cohomology, this is an instance (1.4) of the relative torsion
functor, for sequences of A-bimodules split either as sequences of right
A-modules or as sequences of K-modules.
In the definition (4.2) we may replace B by the un-normalized bar
resolutionj3 (A, A) with& (A, A) =A @An@A. By (4.1), A
(&I)
r A @ A n . Hence Hn(A, A ) is the n-th homology module of the complex
of K-modules A@An with a boundary a =do- &+ ...+ (- I)"&, where
the di are "simplicial" faces:
di(a@Al@...@An)= aAl@A,@..-@An,
i=0,
= a@[email protected].@AiAi+,@...@A n , 0 c i < n ,
=Ana@Al@-..@A,-,,
z=n;
(4.3)
289
in the last term, A, appears in front in virtue of the "outside" associativity rule. In particular a (a @A)= a A- A a, so that Ho is the quotient
of A,
Ho(A,A)rA/{A a- a 21 AEA,UEA),
(4.4)
by the sub-K-module generated by all differences A a- a A.
Much as in Thm. 3.3 we have
n>o,LaK-modzcle.
+-
Exercises
1. Show that the isomorphism (1.1) is natural over the category I-.
Hn(-4, $4
Mac Lane, Homology
= H, ( M @A
B(4).
H, (A, ,GJ
= H, (G @B(A)).
19
290
a(,,.
Z=J, we may also write this definition in terms of the relative torsion
product as
H , (M, G ) =T o ~ ~ ( (G,
~ ) Z)
J)
(G, Z).
(5.2)
In particular, Ho (M, G)= G BMZ. We leave to the reader the description of the cohomology of a monoid.
For a free module the higher torsion products vanish, hence
Proposition 5.1. For lI a grot@ and F a free IT-modzlle
Note that if F is the free IT-module on generators {t),then F&Z is
the free abelian group on the generators {t @I).
The commutator subgroup [IT, IT] is the subgroup of lI generated by
all commutators x y x-l y-l for x , y in IT. It is a normal subgroup of IT;
the factor .group IT/[17, n] is abelian, and any homomorphism of IT
into any abelian group has kernel containing [IT, ITJ.
Proposition 5.2. For 17 a grot@ and Z the trivial 17-module
Proof. The homology of Z is that of the complex Z@,B(Z(IT)
which is the reduced bar resolution B(Z(l7)) of 9 2, with B,=Z,
29 1
Hn(n,
A ) r H n (Z(IT),A,) .
Proof. This is a consequence of Prop.3.4, for the cohomology of
the group IT on the left was defined by B ( Z ( n ) ) , that of the algebra
z(n)by B ( Z O Z V ) )
These propositions reduce the (co)homology of groups to that of
algebras. Conversely, the (co)homology of the Z-algebra Z(L7) reduces
to that of the group IT. This reduction depends on two special properties
of the group ring Z(l7). First, y x = x@ x defines a ring homomorphism
y: Z(n)+Z(n) @ Z ( n ) ; indeed, y is the coproduct which makes Z ( n )
a Hopf algebra (VI.9). Second, Z ( n ) is canonically isomorphic to its
opposite ring. Indeed, if the opposite ring Z(17)OPconsists as usual of
elements roPfor r Z ( n )with product PPsoP=( S r)Op,then the function
5 ( x )= (x-l)OP on IT to Z(n)OP has 5 (1)= 1 , 5 ( x y)= (5x) (5 y), hence
extends (Prop. 1V.I . I ) to a ring homomorphism 5 : Z ( n ) -tZ(1T)OPwhich
is clearly an isomorphism. Composition with the coproduct gives a ring
homomorphism
J
x (x)= X @ (x-yap.
2 92
(a,C) ,
Hn( z ( n ) , c)r H n
(5.5)
O ~(B,~C) -+Hornn
~
(,B, ,C)
h*
--t
293
Exercises
I. (CARTAN-EILENBERG,
p. 201.) For an abelian group G regarded as a trivial
IT-module the homology and cohomology can be calculated from the reduced
bar resolution. Establish the exact sequences
m).
A),
H,(M, G ) = H , ( ~ ,G).
294
Consider the ground ring extension from K to a commutative Kalgebra R. Each K-algebra A yields an R-algebra AR= R @A; there are
ring homomorphisms jK: K+R and jA: A+AR given by jK(k) = k IR
and jA(A) = IR@il, so that (jK,jA): (K, A) -t (R, d R ) is a change of algebras. Each AR-module or birnodule pulls back along jA to be a
A-module or bimodule. There is also a passage in the opposite direction.
Each K-module M determines an R-module M R =R@M and a homomorphism jM: M+MR of K-modules given as jM(m)=l@m. Each
K-module homomorphism ,u: M-tN determines an R-module homomorphism ,uR: M ~ + Nby~ ,uR(r@m)=r@,um, SO that ,uRjM=jNP.
Thus T R (M) =MR, TR(,u)=,uR is a covariant functor on K-modules to
R-modules. This functor preserves tensor products (with @ for @lK,
as always), since ,j and jN yield a natural isomorphism
with an inverse given by q-l [(r@m) BR(rl@n)]= r rt@m@n. We
regard q~as an identification.
For any R-module U and any K-module M there is a natural isomorphism
v: u @ M s u @ ~ M ~ , y ~ ( ~ @ m ) = ~ @ ~ i ~ (m6 4
of R-modules, where U@M on the left is an R-module via the Rmodule structure of the left factor U. The inverse of y~ is given by
@(u @, (r@m))= u r @m. There is a similar natural isomorphism of
R-modules
x:
(X f) (r @m) = r f (m)
(6.3)
29 5
H,, (AR, A). The argument for cohomology is analogous, using x in place
of p.
The direct product A = h Z of two K-algebras may be treated as
a special case of the direct product of two rings (IX.9). The Hochschild
cohomology H" (A, A) is Ext?', (A, A), where R =A @AOpis
,,
(x,
Exercises
IT is a group and K a commutative ring, give a direct description of
augmented K-algebra Z (WK. (It is called the gvou* algeiwa of IT over K.)
I . If
the
296
7 B (A@A1,A @ A 1 )
E
(7.4
i n which the maps are chain transformations of complexes of left A @ A 1 modules cornmzcting with e and E'.
Proof. B y Prop. 7.1, both sides are K-split relatively free resolutions
o f the l e f t A @A -module A @ A 1 ; apply the comparison theorem.
A n explicit chain transformation is given b y the following natural
map
f { l @ [&@&I
ll
..-IA,@Gla@a')
u
= 2 l [A,/ ... I li]
&+l ... &a @All; ... 1:[1:+11
... I &] a'; ( 7 4
1
i= 0
B (A@A1,A @ A 1 )
B (A,A)@ B (A',A')
o f A @A'-bimodules; the map f is again given as in (7.2).
(7.3)
(7.4)
#A:Hk(A,A)@Hrn(A',A')+Hk+rn(A@A',A@A') (7.5)
of K-modules, natural i n the bimodules A and A' and commzcting with
connecting homomorphisms for K-split short exact sequences of bimodules
A or A'. For k = rn =0,these products are induced by the identity map
of A @ A'. The products are associative.
297
Theorem 7.4. I f A and A' are algebras over the same field, the homology product for bimodules A and A' yields for each n a natural isomorphism
If in addition A and A' are K-moddes of finite type the cohomology product is a natural isomorphism
298
This yields a fancier proof of the result (Prop. VII.S.2) that bidim
bidim (TxZ)= 0.
o =bidim Z implies
r=
Exercises
For G a right module and A a left module over A, the k-th relative torsion
product is Hk(G @AB
a A A ) , with B short for B (A,A). The external product for
the relative torsion functor is the map
1.
PT: TorLAlK)(G, A )
tor^^ K, (G',
A3 -+~orL"+$9,"3K)(G@G', A@A1)
defined as the composite of the homology product for complexes, the chain transformation
given by two applications of the middle-four interchange, and the chain equivalence g of (7.3). Show that PT is natural, commutes with connecting homornorphisms in all four arguments, and reduces for k = m = 0 to the middle-four interchange.
2. For K a field, show that the relative torsion product of Ex. i gives an iso-
299
300
lmL(M)n+l.
(8.3)
The sign is the usual one for the commutation of morphisms L(p) and
I of degrees d and 1 . Since L ( p fp)= L(,ul)L ( p ) , L is a covariant functor
on the category of graded K-modules with morphisms of degree 0,
while I : M +L(M) is a natural transformation. A left A-module A
is a graded K-module with operators A @ A + A , so L ( A ) is also a left
A-module with operators
These functors Ext have proved useful for the Steenrod aIgebra for a
fixed prime number f i ; this is the algebra over the field ZPof integers
modulo fi consisting of all primary cohomology operations, modulo # ADAMS[1960], LIULEVICIUS
[ I 9601.
Exercise
I. For A graded, regard the corresponding internally graded algebra A, =&I,
simply a s an ungraded K-algebra. Similarly A-modules G and C yield &-modules
G* and C,. Prove that
~ort;l*.
K) (G*, C*) E 2 ~ori&K)(G, C).
P
9, Complexes of Complexes
301
9. Complexes of Complexes
In any abelian category we may construct complexes; in particular,
there are complexes in the category whose objects are themselves
complexes and whose morphisms are chain transformations. These will
occur in our study of DG-algebras in the next section.
A complex X of complexes may be displayed as a diagram
with additional rows below and above. Each row Xp is a complex with
boundary d , while the successive rows form a complex under another
boundary a' which is a chain transformation a': Xp+Xp-,. Hence
a'd=dal. Adjust the sign of d by setting aWxp,,=(- 4)f'dxP,,. This
gives two families of boundary operators
(x0),=2
x*,,,
p=al+au
P+q=n
is a (single) complex. We say that XO is obtained from X by condensation; its degree is the sum of the two given degrees; its boundary
a0 the sum of the two given boundaries, with sign adjustment. This
sign adjustment may be made plausible by a more systematic presentation.
Let & be any abelian category. Recall that a (positive) &-complex
X is a family (Xp} of objects of d with Xp=O for # < 0, together with
morphisms a: Xp+Xp-, of d such that aa= 0. These X are the objects
of &-complexes. The morphisms of % ( d ) are the
of the category I(&)
chain transformations f : X + Y ; they are families {f*: Xp+YP) of dmorphisms with afp= fp-,8 for all A chain homotopy s: f = f ' : X+Y
is a family sp: Xp-tYp+, of d-morphisms with as+ s a= f- f'. We also
use chain ma+ h: X+Y of degree d ; that is, families {hp:Xp+YP+,)
of d-morphisms with ah =(a. We do not explicitly introduce
the category with morphisms all such chain "maps" because our discussion of abelian categories is adapted only to the case of morphisms
of degree 0.
The lifting functor L of 8 gives a covariant functor on I ( & ) to
X ( d ) , which assigns to each complex X the complex L(X) with L(X),+,
=Xn and differential L(a). The identity induces a chain map I: X-t
L(X) of degree 1 ; as in (8.3), L(a)l=-la. In brief, L raises all degrees
by 1 and changes the sign of the boundary operator.
+.
302
aP a
where each
is a chain map of degree
Lfi (X,,). The chain maps
Xi=
ail$+,
=o. Each X i is an
as Z-lLP(ap)= LP-I (ap)1-I. Then
define
&-complex with a boundary operator which we denote as a". Therefore X m = x X i is an &-complex with boundary a". On the other hand,
a; : X i
has degree - I , hence gives another boundary operator
in X m . Now a' is a chain map of degree - 1 for the boundary a", so
a'' at= - a' a". Therefore P= a'+ a" satisfies P P =O, so ( X m , P ) is
an &-complex, called the condensation of X . This description of X m
agrees with the initial description, since the boundary a" of X i is that
of X,, with + sign changes due to p applications of L. Since X~,,=O
for #>n, only finite direct sums are involved in the construction of X m .
Now let f : X - t Y be a chain transformation. I t is a family of chain
transformations {f,,: Xp+YP) and determines f': X'+Yf as the family
f' - LP(f,,): Xi+Yi. Thus
a'' f; and a'f;= fi-,a'. Hence p=x
PT
satisfies Pp =p P , so is a chain transformation f? Xe-+ Y m . This shows
condensation a functor, as stated.
fia"=
if
r=f:
303
The
Now let X and Y be complexes of complexes (i.e., in $($(at))).
complex of complexes X @ Y has (X@ Y),=2 Xp@Y,, so the yP,, for
p+ q=n give a chain isomorphism of single complexes
The complex (X @ Y)* is the direct sum of the Lu((X@ Y),) ,with boundary a'+ a". The complexXe @ Ye is (2LPXp)@ (2LqT) with boundary
determined by the usual tensor product formula (9.1) from the boundaries P= a'+ a" in Xe and in Ye. By construction, y, commutes with
the a
'' part of the boundary; a straightforward calculation shows
that it commutes with a', and hence with the total boundary P.
Note. The notion of a complex of complexes is not usually distinguished
from the closely related notion of a "bicomplex", which will be discussed in XI.6.
The superficial difference is just one of sign, in the formula
(- l)f'dxp,,.
304
F ( M )= U@M, and e(m)= 1@meF(M). Write E: U+K for the augmentation of U; by pull-back, ,K is a left U-module. An augmentation
of A or of M is a morphism
Proposition 10.1. Each left U-module A determines a DG-modzcle
E:
U+K. If A is augmented, so is A.
305
+c.
Theorem 10.2. Condensatiorz is a covariant functor on A-s@it resolutions X of ,K by U-modules to constructions X O for U. I f X i s relatively
free, so is X O.
Proof. Let E X : X+,K be a resolution by U-modules Xp. By neglect
each U-module Xp is a DG-module; that is, a positive complex. By the
same neglect, X is a complex of complexes, so has a condensation X O =
2 LP(Xp) which is a DG-module under boundary operators a', a",
and aO=a'+ a". But if A is a U-module, then L(A) is a U-module
with zl(lm) = (- l)deg"l
(um). Hence LP (Xp) is a U-module with differential a", while a': LP(Xp)+LP-'(Xp-,)
is a map of U-modules of
degree -1, so that, writing a u for the differential of U E U,
If X is relatively free, each X, has the form U@M, for some DGmodule M, . Thus LP (Xp)r U Q9 LP (M*), so X O sU Q9 Lp (Mp) shows
X O relatively free.
Next we condense the canonical comparison (Thm.IX.6.2).
Theorem 10.3. (Comparisort theorem.) I f X+,K is a relatively free
resolution and Y+,K arz d-s#lit resolution, both by U-modules, there
is a uniqzce homomor#hism p : X O +Y O of augmented U-modules with
20
306
The left bar resolution B(U) is an .&-split resolution of ,K by relatively free left U-modules, so its condensation Bm(U)is a construction
called the bar construction. Specifically, Bm(U)is the graded K-module
2 U@Lp((U/K)f') ; as a tensor product, it is generated by elements
which we write in the usual form as
an element is multiplied by
The augmentation is
U'E
E:
(u [Im)= E (21) ,
(10.5)
[Im and
The normalization insures that smsm=O. The formulas for the two
boundary operators a' and a" are most easily found from that for sm
by recursion on @, using (10.3) and (10.2); they are
Except for sign, a" is the boundary of a tensor product, and a' like
that of the bar resolution. Incidentally, the signs in (10.7) and (10.8)
can be read as cases of our usual sign conventions.
Thus Thm. 10.2 gives
307
This theorem can also be proved directly from the formulas above,
,
atat'+ al'a'=o en route.
with proofs of (10.2), ( I o . ~ )and
In the sequel we use only the condensed bar construction for a
DGA-algebra, so we shall drop the now superfluous dot. The curious
reader may note that the signs occurring in this boundary formula
are not those arising in the bar resolution of $ 2 for an algebra. The
change of signs can be deduced from the lifting operation L P; we have
avoided the meticulous control of this change by deriving the signs from
(10.2) and (10.3).
As for any U-module, the reduced bar construction B ( U ) has the
,
B(U) is regarded as a graded K-sdbmodule of
iorm K, @ I ~ B ( U )and
B (27).
Corollary 10.5. For each DGA-algebra U the reduced bar construction E(U) i s a DG-module over K with B(U)
L*((U/K)P). If elements
are denoted by [ull . . . Iu*] for uic U, the degree of these elements i s given
by (10.4) with u omitted, the boundary a=at+aft by (10.7) and (10.8)
with u= 1 and with u, refilaced by E (a,) in the first term on the right of
(10.8).
Note also that the projection fi : B (U) +B ( U ) rB (U)/JB (U) is
given by P (u[ull ... I u p ]=
) E (u)[yl . .. 1 lip] ; it is a morphism of DGmodules of degree zero. The isomorphism tp: B (U)G U @B (U) is given
) u @ [u,] ... I u p ]; it is an isomorphism of modules
by p, (u[y1 ... I u p ]=
over the graded algebra of U, but does not respect the differential,
because tp a'+ 89.
The bar construction has the convenient property
=z
Corollary 10.6. Both B(U) and B(U), the latter with its contracting
homotopy, are covariant functors of the DGA-algebra U with values in
the category of DG-modules over K. Moreover, f i : B + B and i : B-t B
are natural transformations of fulzctors.
Proof. If p : U - t V is a homomorphism of DGA-algebras, then
B (V) pulled back along p is a U-module, still with a K-module contracting homotopy. Hence the canonical comparison of Thm.10.3 gives a
unique homomorphism
B(P): B(U)-tpB(V)
(10.11)
of U-modules with
E'
B (p) = E .
20*
308
Exercises
1. Describe the bar construction explicitly when K = Z g is the field of integers
modulo and U= E ( x ) is the exterior algebra on a generator of odd degree.
(x*)<
For stage one, the left bar construction B (U) with its total boundary 8.
is a left U-module while G is a U-module by pull-back, so Hom, (B(U),,G)
and G,
B (U) are DG-modules with (co)homology the K-modules
H k (U, 1; G)= H k (Hom, (B(U),,G)) ,
(11 .I)
Since B (U) +,K arises from a resolution, the definition of Hk(U, 1;G)
resembles that of the (U, K)-relative torsion product Tork(Ge,,K), but
it is not a relative torsion product because it uses the total boundary
operator a* of B(U) and not just the boundary operator a' arising
from the resolution.
A homomorphism p : (U, E) -t (V, E') of two DGA-algebras over a
E : U-+K.
fixed K is a homomorphism of DG-algebras with ~ ' p =
Thus B (V) is an augmented U-module by pull-back, and p induces
B (p): B (U) +,B (V), a homomorphism of augmented U-modules which
I + G)
commutes with the contracting homotopy. I t follows that Hk(U,
is a covariant bifunctor of U and G and that HA(U, 1 ; G) is a bifunctor
covariant in G and contravariant in U.The reduced (condensed) bar
construction is also a covariant functor of DGA-algebras to DG-modules.
The (co)homology modules of U may be expressed by the reduced
bar construction. Indeed, since G is a K-module, each U-module homomorphism B (U) +,G must annihilate JB(U), where J is the kernel
I I.
309
(11.4)
If X+,K is anyd-split resolution by relatively free U-modules, standard comparison arguments give
(11.6)
@ . . a
@L(U/K)
(p factors).
(11.7)
310
for the sign exponent ei is that of the tensor product boundary formula,
and the minus sign in front of the summation is that introduced in
L(U/K) by the definition L(a)1u=-- Z au.
A chain transformation p : X -t Y of complexes is called a homology
isomorphism if, for each dimension n, H, (p): H,, (X) H,, (Y).
The proof is an exercise in the use of filtration and the Five Lemma.
First, p carries f U to I", hence induces a chain transformation
U/K+ V/K. We claim that this map is a homology isomorphism. Indeed,
the special assumptions (K a field or K=Z, Uo free) show that I: K+U
is a monomorphism ; hence K U + U/K is an exact sequence of complexes, which is mapped by p into the corresponding exact sequence for V.
Therefore ,LL maps the exact homology sequence of the first into that of
the second. For n 2 2, Hn-, (K) = 0 and the exact homology sequence
reduces to the isomorphism H,, (U)GH,, (U/K). For n = 1 it becomes
I 2.
3fl
312
<
t-iK
g[B(U)@B(V)]
<
s-,K
t(B(U)@B(V)),
u
sB(U@V).
[I
I 2.
313
=+
B< B; on
this sub-
as desired. This completes the proof. Note that the products in B and
U determine that in B; indeed, since nBis a homomorphism of (U@ U)modules, we have
I ) ( ~ ~ (deg61)
~ % )
% ~ u(J1~@zJa)~ (12.3)
Since g is canonical, it can be given by an explicit formula; the formula is (except for signs) just the explicit map g of the EILENBERGZILBER theorem, as given by the simplicia1 structure of B(U). As in
that case (VIII.8), let t be a (+, q)-shuffle, regarded as a suitable permutation of the integers {I, . . .,
q}. For elements
+-+
bi=[%l
Iufi], h2=[~11 Iv,] E B(V),
define a bilinear map (the shuffle product) * : (U) @ B (V) +B (U @ V)
by labelling the elements u, @I, ..., ufi@1, 1 @vl, .. ., l @vq of U@ V
in order as wl, ..., wfi+,and setting
[~i(...1~pI*[~il...(~~]=C(-1)~(*)[wt-~(i)(...(~rl(p
(12.4)
t
where the sum is taken over all (+, q)-shuffles t and the sign exponent
e(t) is given in terms of the total degrees as
t(i)>t(++j) i i P , j S q .
(12.5)
This sign is exactly that given by the sign convention, since the sum
is taken over all those pairs of indices (i, j ) for which ui of degree
deg[ui] has been shuffled past vi of degree deg[vj].
e(t)=;I:(deg[ui])(deg[vj]),
314
Evlwl ~ 1 .
B(U) is strictly
When deg 6 is odd, the signs are opposite, so 6 * 6= 0,as required for
strict commutativity.
The essential observation is that each commutative DGA-algebra U
yields a commutative DGA-algebra B(U),so allows an iteration to
form a commutative DGA-algebra p ( U ) for each positive n. This
gives an n-th stage cohomology (or homology) of U with coefficients
in the K-module G as
H k (U, n ; G)= Hk (Hom (B"(U),G)).
This may be applied when U=Z (17)is the group ring of a commutative
multiplicative group 17. The fi-th stage homology and cohomology
groups of this group 17, with coefficients in the abelian group GIare
thus
Hk(n, f l ; G)=Hk(G@B'(Z(n)))l
(12.6)
(12.7)
Hk (17, n ; G) = H k (Hom (B"(Z
G)) ;
(n)),
315
Hk(K(17,n), G)rHk(17,12; G ) ,
with the corresponding result for homology.
Explicit calculations of these groups can be made effectively by
using iterated alternative resolutions X, so chosen that X has an algebra
[195 51).
structure (CARTAN
Exercises
1. Show that the image of the contracting homotopy in B (U) @ B (V) properly
contains B(u)@B(V).
2. Prove Thm. 12.1 from the explicit formula for the product s.
3. Show that Bn(Z(IT)) vanishes in dimensions between 0 and n, and hence
that H P ( I I , n ; G ) = o = H , ( I I , f l ; G ) for o < p < n .
4. Show that H n (IT, n ; G) E Horn (IT, G) for n 2 1 and that, for n 2 2,
H " + ~(I7,n ; G) E E x t i (IT, G).
5. (The suspension theorem [EILENBERG-MACLANE
1953b, Thm.20.41.) For
p< n, show that S* and S , in (12.8) and (12.9) are isomorphisms, while for p = n,
S* is a monomorphism and S , an epimorphism. (Hint: Compare the complexes
B"+l(U) and En(U) in the indicated dimensions.)
6. For any K-split relatively free resolution X+,K, written as X = U @ X as
in Thm. 10.2, let j: U-tX be given by j(u) = % @ I (assume iEU=X,). Show
that the composite psi: U - t x with s the contracting homotopy, gives the
suspension.
7. For any X as in Ex.6 find a product X @ X + X associative up to a homo~OPY.
317
318
C h a p t e r eleven
Spectral Sequences
n/r.
<
P+PP*
with the usual degree - I , and H({En},d) is the
I . Spectral Sequences
319
r=2,3,
of bidegree (- r , r - I), and with isomorphisms
:
, q + r , q + , ,
...,
(14
More briefly, each Er+l is the bigraded homology module of the preceding (E', 6). Thus Er and d' determine Er+l, but not necessarily
dr+l. The bigraded module Ea is the initial term of the spectral sequence
(occasionally it is convenient to start the spectral sequence with r = 4
and initial term El).
If E' is a second spectral sequence, a homomorphism f : E +E' is
a family of homomorphisms
f : Er-+E",
r = 2 , 3, ...
of bigraded modules, each of bidegree (0, 0), with drf'= f'dr and such
that each f'+' is the map induced by /' on homology (use the isomorphisms (1.3)).
It is instructive to describe a spectral sequence in terms of submodules
of Ea (or of El, if this be present). First identify each Er+' with H(E r , dr)
via the given isomorphism (1.3). This makes E 8 = H(Ea, da) a subquotient Ca/Baof Ea, where Ca=Ker d2 and Ba=Im da. In turn, E4=
H (E 3 , d s ) is a subquotient of Ca/Ba and so is isomorphic to Cs/B" where
C3/Ba= Ker d8, B8/Ba= Im d3,and B3 C s . Upon iteration, the spectral
sequence is presented as a tower
<
320
We regard the terms Er of the spectral sequence as successive approximations (via successive formation of subquotients) to E m . In this
representation (1.4), a homomorphism f : E + E f of spectral sequences
is a homomorphism f : E a 4 E ' a of bigraded modules, of bidegree (0, o),
such that f (C r )< C", f (B r )< Bw, and such that all the diagrams
The {outgoing dr ends outside the quadrant if r > p , the incoming d'
starts outside if r>q+ I, so that
I . Spectral Sequences
321
The terms EP,, on the p-axis are called the base terms. Each arrow
dr ending on the base comes from below, hence from 0, so each E;?
is a submodule of E;,, , namely the kernel of 6 :E;,, +E;-,,-, . This
gives a sequence of monomorphisms
The terms E,,, on the q-axis are called the fiber terms. Each arrow
from a fiber term ends to the left at zero, hence E;,, consists of cycles
and the next fiber term is a quotient of E;,, (the quotient by the image
This gives a sequence of epimorphisms
of a 6).
,++,
it
.I
Exercises
=x
322
2. Fiber Spaces
Before studying various algebraic examples of spectral sequences
it is illuminating to exhibit some of the formal arguments which can
be made directly from the definition of a spectral sequence. For this
purpose we cite without proof the important topological example of
the spectral sequence of a fibration.
Let I denote the unit interval and P any finite polyhedron; recall
that a homotopy is a continuous map H : P x I + - B . A continuous
map f : E -t B of topological spaces with f (E)= B is called a fiber map
if any commutative diagram of the following form
h
P -+E
p 4..* if
P X I ~ B ,
(all maps continuous) can always be filled in at L so as to be commutative. This is the "covering homotopy" property for f : Any homotopy
H of P in B whose initial values H(x, 0) can be "lifted" to a map h: P + E
with fh(x)=H(x, 0) can itself be lifted to a homotopy L of P in E
with f L = H and h ( x )= L(x, 0). If b is any point in BJ its inverse image
F= f-lb is called the fiber of f over b. If B is pathwise connected, it
can be shown that any two such fibers (over different points b) have
isomorphic (singular) homology groups. Hence one may form the singular
homology groups Hp(B, Hq(F))of B with coefficients in the homology
groups Hq(F) of "the" fiber. Strictly speaking, we should use "local
coefficients" which display the action of the fundamental group of
B on Hq(F) ; this we avoid by assuming B simply connected. Since B
is path*
connected, its Odimensional singular homology is
Theorem (LERAY-SERRE).
If f : E - t B is a fiber map with base B
pathwise cmnected and s i w l y connected and fiber F pathwise con~cted,
there is for each n a nested family of s u b g r e s of the singdar homology
group H" (EL
2.
323
Fiber Spaces
In particular, if all H,,-, (B) are torsion-free, E:, = H,, (B) @H, (F).
Assuming this result, we deduce several consequences so as to
illustrate how information can be extracted from a spectral sequence.
theorem holds when all homology groups (of B,
The LERAY-SERRE
F , and E) are interpreted to be homology groups over the field Q of
rational numbers. Write dimV for the dimension of a Q-vector space
V over Q. For any space X the n-th Betti number bn(X) and the Eder
characteristic x (X) are defined by
dim E,:
,= bP (B) b, (F)<
00
and x(Ea)=x(B)x(F). Write Cis, and q,,for the cycles and the
boundaries of E,,,, under C.The short exact sequences
324
= dim EL,
Here the last term has total degree 9-r+ q+r- 1 = (p+ q)- 1, so
(E r ); by induction x (E r )= X ( E 2 ) Since EL, vanishes for p
and q large, Ew= E r for r large, and x (E m )= X (E 2 ). NOW b y (2.1) and
(2.4,
dim H, ( E )= r, dim ( H p ,/Hp-,,
,
,+,) = 2 dim EE, ,
~+q=n
$+q=n
so x ( E ) = X ( E w ) = ~ ( E 2 ) = ~ ( 6 j ~ as
( F asserted.
),
x (E'+') = X
The non-zero terms of E:, all lie on the vertical lines = 0 and $= k,
so--__
the only differential dr with r >= 2 which is not zero has r = k. Therefore
-.=E k , Ek+l= Ek+a= ...=Em. The description of
E~+'=E" as the homology of (E R , d k ) amounts to the exactness of the
sequence
dk
O-tEcq-tE:,q-+
E~,q+k-l-fEOq)4+k-1-+o.
On the other hand, the tower (2.1) has only two non-vanishing quotient
modules, so collapses to 0 <Hop,= Hk-l,,-k+l< Hk,,-,= H , . With the
isomorphisms for E w in (2.4, this amounts to a short exact sequence
with H,(E) in the middle. Now set q=n- k in (2.4), put in the values
of E 2 in terms of H(F) and splice the sequences (2.4) and (2.5) together:
325
2. Fiber Spaces
326
- -.+ H,,(E)
H,,(B)
rptl
Hn-,-,(B) +Hn-, ( E )
-to.-.
0 ~ ~ , q b ~ ~ ~ , o - - - * ~ u - k - l , k ~ ~ ~ - k - l , k ~ 0 t
Exercises
Sm+ sk is a fiber map with k z 2 and with fiber a sphere S' prove
that one must have m= 2k - I and I = k - I . (For k = 2, 4, and 8 there are indeed
such fiber maps; they are the Hopj fibvations; HOPF [1931, 19351, STEENROD
[19511, Hu r-1959, p.661.)
In the following three exercises, f : E - t B is a fiber space with B pathwise
connected and simply connected and fiber F pathwise connected.
2. If H j ( F ) = o for O < j < t and H i ( B ) = O for O<i<s, obtain the exact
sequence
H,+t-1 (F)--%+,-I (El +Hs+t-1 ( B )+Hs+t-,(F)
I. If f :
+...
+ H , ( B ) +Hx(!,(F)+%(El
+H,(B)
+o-
3. If Hi ( B )= 0 for all i#
prove that H,, ( F ) H,, ( E ) for all n.
4. I f Hi(- = 0 for a l H > 0, prove that H,,(E) = H n ( B ) for all n.
5 . Given the LER&-SERRE
spectral sequence E and Q the field of rational
numbers, define a s$ectral sequence E'= Q @ E of vector spaces over Q and show
that E& = . H p ( B , Q ) @ H ( F , Q) and E i z = H i , q / H i - l , q + l , where the H' appear
in a tower hke (2.1) with
( E ) replaced by H, ( E , Q).
d,,
3. Filtered Modules
A fiZtyation F of a module A is a family of submodules F p A , one for
each p 2,with
. .C F ~ - ~ A < I $ A < F ~ + ~ A <
(3-1)
a * . .
3. Filtered Modules
327
<
:"."
i.e., E $ , , E H ~ + , ( F ~ A / F ~ _ ~ A (3.2)
).
q,q=
E l p = ( Z l p ~ F p - ~ A ) l ( a 2 ; : ~ - ~ ~ ~ r- =
~ IA, 2) , ,...
328
E;+?
a,
E;
-i- Ei-,
-.
b= 0
Next, a, cons% of the pairs (qp+?
b, q pab) for ~EZ;$+,
, while
if also b~2$,+,_,A; that is, if ~ E Z & ; _. , Hence (where "Ind" means
the "indeterminacy ')
J
induces homomorphisms
The tower on the first line, taken modulo l&,A, gives that on the
second line, with B;=% aZ;;T;-, and Ci=qpZg. B y 11.6 the additive
329
3. Filtered Modules
relation a,:
to an isomorphism
with n the projection, L the injection. This yields the spectral sequence,
much as in Ex. l .I (except that C' there is Cr+' here).
To describe FpHIFp-,H, write C = Ker a and B= a A for the cycles
and boundaries, respectively. in A . Then F induces on C and B filtrations
FpC=CnF,A, FPB= BnFPA. By definition, F P ( H A=
) (FPCuB)/B.
Hence
A.
represents FPHI&-, H as a subquotient of F9A/FPV1
The numerator of Ei in (3.6) is ( Z ~ U I $ - ~ A ) / I $ - , AA/I$-,A
<F~ ;
the denominator is (aZi7:-luF,-1A)/Fp_,A, so
330
union of all FpA and bouded below if for each degree n there is an
integer s=s(n) such that F,A,=O.
Proposition 3.2. If F is bounded below and convergent above, then
O=F-lH,A<F,H,A<F,H,A<~-~<F,H,A=H,A
d t h successive quotiertts l$H,/l$+, H,zEE,-,, , under isomor+hisms
induced by i A .For exattqble, the LERAY-SERRE theorem arises from a
canonically bounded filtration of the singular chains of a fiber space.
Since El A = 0, E i = H(FpAll$-, A ) =o for p< 0. Since
Fp-I AP+qand hence E:,, = o for q< 0.
Therefore all non-zero E;,, lie in the first quadrant of the (p, 9)-plane.
and the induced filtration of H,(A) is finite as displayed.
For n=1 the filtration of HI amounts to a description of HI as the
middle term of a short exact sequence
Proof.
,
+
A
H n ( A )+E:o*
(3.10)
each induced by iA.In general, the spectral sequence of F determines
not H ( A ) but its subquotients FpH/Fp-,HI asserting that each is in
its turn a subquotient of Ei= H(FpA/Fp-,A).
3. Filtered Modules
331
Proof. Since both spectral sequences are bounded below, the previous
mapping theorem (Thm. 1 .1) shows ar and a": E" +Elw isomorphisms.
Consider the induced map a,: H,,(A) -+H,,(Af) on homology for a fixed
HL . Since both filtradegree n, and the corresponding a#,,,: l$H,+$
tions are bounded below there is an s with F, H,, =o =F,' H; . The convergence isomorphisms (3.3) give the horizontal sequences in the commutative diagram
,/'o-+
( A )-+FpH,(A) +EOO,-p-+O
1+1..
332
p,
3. (The exact sequence of "terms of low degree"; cf. Ex.2.2.) In Ex.2 suppose
E ~ , ~ = Owheneither o<q<t
or
o<p<s.
4. (The two-row exact sequence.) In Thm. 3.3 suppose there are two indices
< b such that E;, = o for q a, b and all p. Derive the exact sequence
0 4a
2I
induced by the identity. (This description may be used to define the spectral
sequence of a filtration; see FADELL-HUREWICZ
[1958, p. 3181.)
4. Transgression
with exact row and columns. When (as we have assumed here) the
spectral sequence starts with r = I,the additive relation
4. Transgression
333
is called the transgression. Any additive relation (Prop. 11.6.1) is a homomorphism from a submodule of its domain (here called the module of
transgressive elements) to a quotient module of its range; in this case,
(4.1) represents z as the homomorphism & from the submodule E$,,
of E;,, to the quotient module Egf'-, of E:,,-l. Replacing El by E2
in this definition of z gives an additive relation z': E : , o y Ei, #-,, also
called transgression. Each transgression uniquely determines the other
via the edge homomorphism e: E$,*-, +Eo,2 p - l , for z = e-lt'; since e
is an epimorphism, e e-l= 1, so z'=e t.
a: A + A .
z: E;,,,-E$,,-~
334
where
B and
Proof. The hypotheses L,-,< M,-, and L,< M, show that the
identity induces homomorphisms /I and y as displayed. By the equivalence principle (Prop.II.6.2), /I-I and y-I are the additive relations
induced by 1. By the composition principle (Prop.II.6.3), each of
8, B-I and y-I 8, turns out to be the additive relation induced by a 1= la;
hence the result.
This result shows that Def Q = Im B and Ind Q = Ker y.
In 8 10 we need information as to the effect of a chain equivalence
on the connecting relations, as follows.
i s commtative.
This result computes
Q'
from
e as ef=g,
h;l, or conversely.
4. Transgression
335
But h*
Since the diagram commutes, B'F, = h,B, or B-l h;' =i&l/?'-'.
and y , are isomorphisms, so e)* B-I =,9' -l h* . NOW g* Q = g* aLB-l =
aL.o)*B-l= aL.B' -l h, = e'h,, as desired.
But the maps (., and e, are induced by the identity, whence the first
result. Similarly, each of z and Q is the additive relation EE,o--E;,
induced by a, so z= Q, as desired.
The situation may be visualized in terms of the complexes
,-,
AIM.
Since M,) (F,A), for n 2 1, the transgression can also be described in
terms of ordinary connecting homomorphisms, as in Prop.4.2. This
theorem shows how additive relations clarify a result of SERRE(loc. cit..
1.3 ;his notation R=F,A, S= AIM). In the case of a fiber map f : E +B,
H(A)= H(E), H(F,A) is the homology of the fiber, HP(AIM)= Eip0=
H,,(B, 2) that of the base. Thus Prop.4.2 gives for transgression the
following "geometric" description (in which it originated): A homology
class of the base is transgressive if it can be represented by a cycle z
such that z= fc, for c a chain of the total space with a c in the fiber.
An image of cls z under transgression is then the homology class, in
the fiber, of any such ac.
336
5. Exact Couples
An alternative description of spectral sequences can be given via
"exact couples"(MASSEY [I 9521). Though not necessary to the sequel,
they throw some light on the origin and nature of spectral sequences.
An exact coufile &= {D, E ;i, j, k} is a pair of modules D, E together
with three homomorphisms i, j, k ,
which form an exact triangle in the sense that kernel = image at each
vertex. The modules D and E in an exact couple may be graded or
Z-bigraded; in the latter case each of i,j, k has some bidegree.
The exactness of & shows that the composite j k : E - t E has square
module
zero, hence is a differential on E. Form the
for this differential. Construct the triangle
i'
i D ------+ i D
\k
(54
K 4
W E ,j k )
where i' is induced by i and j' and K' are given by
Observe that id=O implies d c k E, so j d c jk E and j' is well defined.
Similarly jke= 0 implies k e c i D, so k' is well defined. Call &' the
derived cou+Ze of &; it is a functor of & under the evident definition
of homomorphisms for exact couples. A diagram chase proves
Then i, jil-',
D?2Dr
6'
&
Jr
E'
called the r-th derived cou#Ze of &.
r=1,2
,...,
5. Exact Couples
337
Theorem 5.2. The r-th derived cozlple 6' is exact with Q1=Q, Qa= Q',
and Q'+lis the derived cozlple of Qr.
Proof. For r = l , E1=E. For r=2, exactness of Q gives iD=j-lo,
ker i= k E, hence Ea= k-l j-lO/jkE = H(E, jk) and thus Qa the derived
we need only show that
couple of 6. For r>2, D'+l=iD'=i,D';
E'+I is the homology of E' under the differential j, k,: E r +ET. To exhibit this differential, write the definition (5.3) of E r as
Ef=C/B,
C=k-l(iT-lD),
B=j(Kerir-l).
kc=ir-ld.
(5.4)
First j, k, (c+ B)= 0 gives jd= ja for some a D with ;'-la =o. By the
exactness of Q, d-a=id' for some a', so kc=iTd' and c~k-l(irD).
Conversely, k c= i'd' gives j, k, (c B)= 0; the kernel is as stated.
Similarly Im (irkr)consists by (5.4) of elements jd+ B with i'd = i k c =0,
and conversely i'd =O implies ir-la= kc for some c; this gives the
stated image. Since Qr+l is the derived couple of Q', it is exact by
Thm.5.1.
Corollary 5.3. An exact couple of 2-bigraded modules D, E with
maps of bidegrees
d e g i = ( l , -I),
degj=(O,O),
(5.5)
degk=(--1,0)
... -fEp,q+i
0
+Ep+l,q
k
--'
k
+
"
'
-+EP+2,g-1+
ji
DP-i,q+l
ii
i
+
-j
.li
k .li
A EPPq
DP-l,q
.li
i
k .li
DP+l,g-l+ EP+l,q-l-j DP,g-l
.li
Li
DPjq
E p - - ~ , ~ + iDP-a,q+i+ '..
-+
i
+
. .'
...
338
and with degrees of i, j, k as in (5.5). Call this the exact cou#Ze of the
filtration F.
Theorem 5.4. The s$ectral seqwence of F is isonzor$hic to that of the
exact cm$Ze of F.
Proof. The spectral sequence of the exact couple (5.6) of F has
E r =k-1 (Im ir-l)/j(Ker ir-I )
i: H(I$-,A) -+H(%A).
339
5. Exact Couples
k
-+
Torq(G, KpWl)
-)
340
Exercises
I.For an exact couple 6 with "first quadrant" term E, show that Dp-l,q=
Dp,p-l for p < 0 and q < 0. Describe the upper and lower edges of the corresponding
diagram for 6.
2. Show that the exactness of the derived couple 6' can be deduced from
the Ker-coker sequence for the diagram
6. Bicomplexes
a':
K,,,,;
H~,-+Hr-,,,
+H~-l,q)/a'H~+l,q
(6.3)
induced
(6.4)
6 . Bicomplexes
341
K ~ , ~ a=af+
,
aft: x,+x,-,.
(6.5)
fi+q=n
The assumptions (6.2) imply that a2=0; if K is positive, so is X , and
in this case each direct sum in (6.5) is finite. This totalization operator
has already been used. Thus, if X and Y are complexes of K-modules
with boundary operators a' and a", respectively, X @ Y is naturally
a bicomplex ( X f i@YJ with two boundaries
X,=Z
342
,,:'
where we have grouped terms of the same bidegree. Hence a&
a f f a , , = o and acZ;,, if
if
where
Therefore E$,= Lp,q/Mp,q,
Exercises
I . Let X and Y be complexes of abelian groups, with each X, a free group.
In the first spectra1 sequence of the bicomplex K = X B Y , show that
E L ~
for Tor, applied as in Prop.V.10.6, to show that @ = ds= ...= 0 and hence that
F = Em in this case.
2. Describe E;,, by a quotient LIM, as in the second proof of the text.
343
344
(7.3)
This gives the spectral sequence of the theorem. As for any canonically
bounded filtration, it converges to H(Tot K) as given in (7.3) by the
first spectral sequence. Hence the conclusion (7.1).
This proof is a typical case of two spectral sequences, one of which
collapses so as to determine the limit of the second.
Corollary 7.2. If 17 oflerates #roflerZy on the flathwise connected
acyclic sflace X there is a natural isomorflhism H,(D, C)rH,(X/lT, C)
for each fl, where C is any abelian groufl regarded as a trivial IT-module.
Proof. Since X is acyclic, Hq (X, C) = o for q j o and is C for q = 0 ,
so the (second) spectral sequence collapses, so has E2 isomorphic to the
limit, as asserted.
This result is the homology parallel of Thm. IV.11.5 on the cohomology of Xl17. As in that case, this corollary could be proved directly
without the use of spectral sequences. Put differently, the spectral
sequences allow us to generalize Thm. IV.11.5 to apply to spaces which
are not acyclic. For example:
Corollary 7.3. If the sflace X has Ho ( X ) s Z and Hq (X)
= 0 for
O<q<t and if 17 operates flroperly on X , then
~r~
345
o +EFt
while the description of EEt as the homology of E:,, under dt+l is the
exact sequence
Replacing E;+l,, by its value Ht+, (IT, C), using (X.5.2) to calculate
E$,,=H,, (17,Ht (X, C)) r H t (X, C) @,Z, and splicing these sequences
gives the result. This exact sequence is a particular case of the "exact
sequence of terms of low degree" (Ex.3.3).
This result determines H, (X/W for n< t and Ht (XI17) up to a certain
group extension. A complete determination of Ht(X/IT) in terms of
H(17) and H(X) requires an additional invariant, a cohomology class
k E Ht+l (IT, H, (x)), as introduced by EILENBERG-MACLANE
[I 949,19501.
The spectral sequence of a covering is due to CARTAN-LERAY [I9491 and to
CARTAN[I9481 For further applications, see CARTAN-EILENBERG,
p.356; Hu
[1959], p. 287ff. ; HILTON-WYLIE [1960], p. 467.
Exercise
1 . Show that the use of the first spectral sequence in the proof above may
-+ C @ S (X) BnZ is a homobe replaced by proving that 1 BE:C @ S (X) BnB (n)
&:
346
- often called a descending filtration, though it's really the same filtration in a different notation. Thm. 3.1 applies directly (only the notation is changed): Each such F yields a spectral sequence {E,, d,) with
@= H(FPA/FP+~A)
and
~ ,FP+'AP+gf '1, and d, is induced by 6.
where Z ? = [a\ a E ~ A P +6 a
If F is bounded, there are natural isomorphisms E L ~ F P H
A/P+'HA,
where p H denotes the filtration of HA induced by F. These isomorphisms also hold if F is convergent above (U P A =A) and bounded
below (for each .n there is an s with FSAn=O). Note that bounded
"below" appears as a bound at the right in the descending filtration (8.2).
The filtration F is canonically cobounded if P A =A and F"+'An= 0
(note that this is not the same as canonically bounded). This implies
that the complex A is positive in upper indices (An=O for n<O). An
argument like that for Thm. 4.4 proves
347
Exercises
I.Under the hypotheses of Thm. 7.1, obtain a third quadrant spectral sequence
E@q EHP (lI,Hq ( X , C)) pj H n (X/17,
C)
r.
called restriction, which is natural in A. Also A < r < 1 7 gives res; resg=
resz. Let Ar denote, as usual, the subgroup of those elements a in A
If is a normal subgroup of 17, Ar is a left
with t a= a for every t
(17/r)-module. The projection a : ll-+17/r and the injection j : Ar+A
give a change of groups (a, j):
A) +(17/r, Ar) which induces a
homomorphism
i n g r : H" (II/T, A 3 -+H" (ll, A)
(9.2)
~ r .r
(a
(r,
348
For g E Hornnlr (B, (ITIT), A ) , inflation is induced by the cochain transformation a* with
(9.7)
Indeed, since 17 is a union of cosets r y of r , the free l7-module Z(17)
on one generator is the direct sum of the free r-modules Z ( r )y. Hence
any free l7-module is also a free r-module, so E : B (IT)-+Z is also a
free r-resolution of the trivial r-module Z. The map B (x): B ( r ) -+B (17)
is a chain transformation lifting the identity
hence by the comparison
theorem gives an isomorphism (9.7).
Next, if r i s a normal subgroup of 17 and A a I7-module, each H" ( r ,A)
is a (17/r)-module. First, for any =B, Hom,(B, A ) is a (17/r)-module
under the definition (Hopf algebra structure!)
( x f ) ( b ) = x f ( x l b ) forf: B+A, ~17,
~ E B .
(9.8)
349
(r)
Lemma 9.1. For F normal in 17, the image of the restriction lies ilz
H" ( r , A)".
Proof. By (9.6), restriction is the composite y = B(x)*i. For each
17-module homomorphism f : B(17)+ A , (9.8) gives x f = f for each
x d 7 . Hence, if f is a cocycle, clsf in H" ( r , A) is invariant under each
operator of 17.
In the diagram (9.6), the definitions (9.5) and (9.4) show a*: L-tK
a monomorphism and y : K +S an epimorphism, with composite y a*
zero in dimensions greater than 0. Hence we are in the situation of
a complex K with two given subcomplexes a* L and M = Ker y, with
(a*L)"<Mn for n>0 and S r K / M ; in this situation (4.4) defines a
homology connecting relation
e = e ( K ; a*L, K e r y ) : Hn(S)-Hn+l (L)
Take this to be the connection ,!e; of (9.3). Explicitly,
relation consisting of all pairs of cohomology classes
is the additive
where the second term on the right in effect implicitly uses the contracting
homotopy in B(17/r). Since the values of g lie in Ar, this function h
is indeed a r-module homomorphism h: B, (17) A. A calculation
with the boundary formula in B (U), using G f = a* g and dg= 0, shows
6 h= 0. Moreover, B (x): B (r)
+ B (17) carries h in S into y f in S,
so any clss y f in Def Q is represented by clss,h in H" (S ). In this complex
S we can compute the action of any ~17.
Let k, be the cochain with
k, (x,, , ..., xn-,) =g (a x, I , a x,, ..., a x,-,). The coboundary formula
and the definition (9.8) show that
f
Q:
H" ( r , A)"-
Hn+l(17/T, A) .
3 50
(9.9)
where Hom, has operators as in (9.8) and C @ B has "diagonal" operators x (c@ b) = (x c @ x b). The map in (9.9) is given by adjoint associativity. To check that it respects the operators indicated, consider any
group homomorphism f : C @ B-tA. This lies in Hom, on the right if
f(xc@xb)=xf(c@b),
CEC,bcB, XEIT.
(9.10)
(9.11)
Proof. (Cf. Ex. 6.) I t suffices to take for F the free 17-module Z (17)
on one generator. The cohomology in question is that of the complex
Exercises
1 . Show how the restriction homomorphism may be calculated from any free
L7-module resolution of Z.
2. If L 7 = T x A , identify
lI/A with
10.
351
4. For
normal in 17, each 17-module A is a r-module under the induced
q': r+Aut A. For each ~ 1 7show
,
that the definitions 5,t = r l t x for t e r and
a,a = x u yield a change o f groups
ex=(5,,a,):
(r.A,v')-t(r,A,q'l
ti e r , x 17.
HP+q(17, A) ;
(Wr)@ Bq(U), A)
as by (9.9), and with two differentials given, with the standard signs
for a coboundary and a differential in B, @ Bq, for f e Kfl'q by
The condition 6'6"+ 6"6'=0 is readily verified. The first and second
filtrations of this bicomplex yield corresponding spectral sequences E'
and E", both converging to H(Tot K).
For the second spectral sequence E" the filtration index is still to
be denoted as p, so we write Kqlf'= Homnlr(Bq, Homr(Bp, A)) for the
terms of K, with second degree labelled as p. As for any bicomplex,
EiP1q=H"pH'q(K). But H1q(K) is the cohomology ot n/P with coefficients in Homr(Bp, A ) . By Lemma 9.3, this is zero for q>O; it is
3 52
(a
(5,
The non-zero terms all lie in the base q=O, so the spectral sequence
collapses. For each total degree n there is only one non-zero quotient
in the filtration of H n (Tot K), hence an isomorphism
H n ( n , A )G H n (Tot K) .
(10.1)
The proof shows that this isomorphism is induced by the chain transformation
(: Homn(B (D), A ) +Tot K
which assigns to each f : B,(L?)-+ A the element
c ~ E K O Idefined
"
by
(mr), Hq ( S t ) )~
o (Bp
m (17/r),
~ ~Hq (~r , A)).
(r,A )
353
The isomorphisms (10.3) are special cases of (10.2). Note that the
edge homomorphism on the fiber has its image in E2n=H"(I', A)?
exactly as for the restriction map (Lemma 9.1) and that the transgression
z has its domain of definition contained in Hn-' ( r , A)n, exactly as for
the connecting relation p (Lemma 9.2).
Proof. For the spectral sequence E of the first filtration the edge
effects are calculated by Thm.8.1 from the subcomplexes F I K and L
of Tot K, where LP=ZPO, using the injection L : L-tTot K and the
projection n: Tot K-tTot K/FIK. This gives the first line of the following diagram, in which the second line presents the complexes used in
5 9 in the calculation of res, inf, and Q:
L
i"
Tot K
5 (Tot K)/FIK
,J;
(10.5)
om=(^ (17/r),
A ) 2 om,(^ (IT), A ) -%H O ~ , ( B(r),
A).
The maps A, 17, p, comparing these two lines will be defined in terms
Specifically,
of the homogeneous generators (xo, ..., x,) of B
Lp=ZPO< P K consists of all gc K9.O with G ~ Kfi+l*O;
E
that is, with
dUg=0. Since B1(IT) is the free abelian group on generators (x, y) with
a(% y)= (Y)- ( 4 ,
(m.
<=
<
23
3 54
restriction induced by y. Similarly, Lemma 4.3 shows that the transgression, regarded as the connecting relation for the top line, agrees
with the group-theoretic transgression computed (as in 9 9) from the
bottom line.
The terms of low degree in this spectral sequence yield an exact
sequence
r.
As an application, we prove
[n:
[r:
355
Exercises
(All the exercises refer to the Lyndon spectral sequence)
I . In the filtration of Hn (R; A) show that p H n may be characterized as
the image of the inflation map, and F I H n as the kernel of the restriction.
2. Establish the exact sequence
O+PH~F=H=-+E$~-+E~O-+HS(~,
A).
(n,
4. (HOCHSCHILD-SERRE
[I9531; HATTORI
[1960].) Suppose m 2 I and I% ( r , A)
< n t m. For O < n < m establish the exact sequence
= 0 for I
5. For C a right I7-module, establish a first quadrant spectral sequence converging to the homology of
n,
H p (Wr,
H , (r,c))=E;, 0 ? H ( n , c).
11. The Comparison Theorem
for the term E a . This hypothesis frequently holds. For example, in the
LERAY-SERRE spectral sequence of a fiber space with simply connected
base space, (2.2) gives E;,,=Hp(B, Hq(F)), which by the universal
coefficient theorem yields the exact sequence
Theorem 11.1. (Comparisolz Theorem.) Let f : E +E' be a homomorphism of first quadrant spectral sequences of modztles over a commutative
23*
356
ring, each of which satisfies (11.1), such that f commutes with the ma#s
a' in (11 .I). Write G,,: EL,, +Egg. Then a n y two of the following
conditions i m p l y the third (and hence that f i s a n isomor#hism) :
x , o, x ' ,
In view of the geometric applications, we read (i) as " f is an isomorphism on the base", while (ii) is "f is an isomorphism on the fiber",
and (iii) is " f is an isomorphism on the total space".
has exact rows and is commutative. Conditions (i) and (ii) imply that
the outside vertical maps are isomorphisms. By the short Five Lemma,
so is the middle vertical map f;,,. This isomorphism of the complexes
(E2, d2), (Eta, dl2) implies that of their homologies E 3, Ef3, and so on
by induction to give (iii), since each Ei,, is ultimately constant.
The other cases of the proof exploit the fact that a spectral sequence
can be regarded as an elaborate congeries of exact sequences in the
bigraded modules
E T,
CT = ker d',
BT =im d'
In the application of the Five Lemma (in its refined form, Lemma 1.3.3)
we shall write down only the first row of commutative diagrams like
(11- 2 ) .
To prove that (i) and (iii) imply (ii), consider the property
(ii,) f$,,: E~,,-+E~:,is an isomorphism for O s q g m .
Since E;,,= E?,, (iii) implies (ii,). Hence it will suffice to prove
by induction on m that (i), (iii), and (ii,) imply (i&+,). Given (ii,), the
diagram (11.2) shows that fi,, is an isomorphism for q s m . By a subsidiary induction on 7 2 2 , we prove that
a monomorphism for q S m and all # ,
an isomorphism for q 5 m- r+ 2 and all # .
11.
357
This holds for r = 2 ; assume it for some r . The Five Lemma for the
commutative diagram on the exact sequence
dr
O--+C;,q+E;,q -+ EL-r,q+r-I
which defines the kernel Cr of d' shows for the map cr induced by f' that
, ,--+ C;lq is
c;, : C;,
a monomorphism for q l m ,
an isomorphism for q s m- r+ 1.
isanepimorphismforqSm.
,-,+,
(11.5)
is an epimorphism for r 2 2 2 .
(11.7)
df
0--+C:,m-r+2--+E:,m-r+2--z
E&rn+1--+Ek$+l+o-
+,
358
Notes. .Spectral sequences were discovered by LERAY [1946, 19501 for the
case of cohomology; their essential features were noted independently by LYNDON
[1946, 19481 in the case of the spectral sequence for the cohomology of a group.
The algebraic properties of spectral sequences were effectively codified by K o s z u ~
[1947]. Their utility in calculations for the homotopy groups of spheres was decisively demonstrated by SERRE [1951]. The equivalent formulation by exact
couples is due to MASSEY[I9521; for still another formulation see CARTAN-EILENBERG, XV.7. The LERAY-SERRE theorem has been proved by acyclic models
[GUGENHEIM-MOORE
19571; for other proofs see H u [1959, Chap. 1x1, HILTONWYLIE [1960, Chap. XI. and, with a slightly different notion of fiber space, FADELLHUREWICZ
[1958]. L YNDON' S spectral sequence was originally defined by a filtration of Hom(B(II), A ) ; his sequence satisfies Thm.lO.1, but i t is a t present
not known whether i t is isomorphic to the spectral sequence we define, which
uses a filtration due to HOCHSCHILD-SERRE [1953]. These authors established the
edge effects (Prop. 10.2) only for the Lyndon filtration; our proof direct from the
Hochschild-Serre filtration depends upon our description of connecting relations,
which was concocted for this purpose. The LYNDON spectral sequence has been
used by GREEN [I9561 to prove for a finite p-group II of order pn that H, (II, Z)
[I9591 proved topologically
has order p k with k 5n (n- 1)/2. For lI finite, VENKOV
that the cohomology ring H(II, Z) is finitely generated as a ring; the algebraic
proof of this result by EVENS [1961] uses the product structure of the L YNDON
spectral sequence. Among many other applications of spectral sequences, we
note BOREL'S[I9551 proof of the SMITH fixed point theorem and FEDERER'S
application to function spaces [I9561 In the comparison theorem, due to MOORE
[CARTAN
seminar 1954- 195 51, we follow the proof of KUDOand ARAKI[ I 9561;
a closely related proof by ZEEMAN [I9571 includes the case where the given isomorphisms are assumed only up to specified dimensions. EILENBERG-MOORE [I9621
study convergence and duality properties of spectral sequences in an abelian
category.
Chapter twelve
Derived Functors
This chapter will place our previous developments in a more general
setting. First, we have already noted that modules may be replaced
by objects in an abelian category; our first three sections develop this
technique and show how those ideas of homological algebra which do
not involve tensor products can be carried over to 'any abelian category.
Second, the relative and the absolute Ext functors can be treated together, as cases of the general theory of "proper" exact sequences
developed here in $5 4-7. The next sections describe the process of
forming "derived" functors: Horn, leads to the functors Extft, @,
to the T o e , and any additive functor T to a sequence of "satellite"
functors. Finally, an application of these ideas to the category of complexes yields a generalized KUNNETHformula in which the usual exact
sequence is replaced by a spectral sequence.
I. Squares
359
1. Squares
formed with the given edges a and j9. Call the left hand square couniversal, for given a and j9, if to each right hand square there exists a
unique morphism y : DU+D with p = j 9 ' y , a"=afy. A couniversal
square (also called a "pull-back diagram), if it exists, is unique up
to an equivalence of D, so that a and j9 together determine ar' and p'
up to a right equivalence. GABRIEL [I9621 calls D a fibred product.
Such couniversal squares are familiar in many branches of Mathematics and under more general assumptions (than those made in an
abelian category). In the category of sets, if a and j9 are injections,
D is just the intersection of the subsets A and B of C. In the category
of topological spaces, if /Iis a fiber map and a : A+C a continuous
map into the base space of j9, then j9' is the so-called "induced" fiber
map. In any abelian category, the couniversal square for C =0 is
Theorem 1.1. (Sqware Construction.) To given coterminal morphisms a, p in an abelian category there exists a couniversal sqware (1.1).
In terms of the direct sum A @ B with its projections nl and n,, D may
be described as the domain of v E ker (a nl- 8 n,), with a' = n,v, p' = nlv.
Proof. For D, v, a', and /If as described, consider
The two triangles are commutative, by definition of a' and j9'. The
square (better, the diamond) on D is commutative, for
Moreover, for any second commutative square on a and p, with upper
corner D" as in (1.I), the couniversality of A @ B provides : D"+ A@ B
360
2 . For x rnonic, o epic, and x, a coterminal, prove that x' and a' in the square
e,
3. If
Q (ker
a ) = 0 with
361
pQ=au.
4. I n a commutative diagram
tl
iTi
.*.+.
let both squares be couniversal. Show that the square with top and bottom edges
E, y is also couniversal.
r]
al o= a,o
362
5,=0;
5 is epic @ 5: is surjective;
(iii) 5 is monic @ 5, is injective @ Kernel 5, = 0.
I f the com#osite q 5 is defined, (q E), =qs 5, and
(ii)
Kernelq,=[blb~B, and b s k e r q ] ;
(2.1)
363
If the cowqbosite
Exercises
1 . Verify directly that each of the assertions of Thm.2.3 holds in the abelian
category of all R-modules.
2. If r] l is defined, show that im 5 5 ker r] if and only if coker
and that ker r] 5 im if and only if coim r] 2 coker 5.
l z coim
r]
364
3. An anti-isomorphism g,: S+ T of partly ordered sets S and T is a 1-1 correspondence such t h a t s 5 s ' implies g, s
sf. Prove A, anti-isomorphic t o A9
under the correspondence cls x +coker x.
zg,
5. For 6 : A+ B define ES: B,-t A, by ES im j? = ker [Eq (coker j?)] (in the notation
of Ex. 3, ES = g,-lPg,).Prove that ES is characterized by the properties E, (5' im B) 5
im j?, E, im a $ im j? implies im a 5 6' im 8. For modules, conclude that SS(im /?)
is the inverse image of the submodule im j? under 5.
6. Restate Thm.2.3 in terms of the maps
ES.
3. Diagram Chasing
with exact rows (i.e., with rows exact at B , C , B', and C') satisfies
monic,
3. Diagram Chasing
365
Proof. Consider the diagram (zeros on the top and sides omitted)
366
Now make the added assumptions and use the diagram of the
corresponding sets of quotient objects, with all mappings reversed. To
prove epic, by Thm. 2.3 part (iii), consider c ' ~C'q with image 0 in B'q.
By (ii) of the same theorem, there is a c which maps to c'. Let c also map
to be Bq. Since b then maps to 0 in B'q, exactness of the middle column
at B gives a b" with image b. But b and hence b" go to 0 in A . By the
short exactness of the first column, b" already goes to 0 in A". Exactness of the row at B" gives a c" with image b". Let c" map to c, in Cq.
Then c and c; have the same image in Bq, so c,=c by exactness. The
original c', as the image of c", is now zero, so /?is epic as desired.
Again, the proof uses quotients to avoid subtraction. For completeness, we adjoin
Exercises
The first two exercises use the "subtraction" device noted in the proof of
the Four Lemma.
1. Prove the strong Four Lemma (Lemma 1.3.2) in an abelian category.
367
two non-zero morphisms in the middle row is zero, then the middle row is short
exact (cf. Ex. 11.5.2).
3. An epic
5. Prove:
6. Prove :
7. If
368
Proposition 4.1. The direct sum of two proper short exact seqzcences
i s #roper exact.
Proof. Morphisms a,: Ai +Bi have a direct sum
(4.1
where ni: A,@A2+Ai and l j : Bj+Bl@ B,. If xlla and 1112, an easy
argument shows (x@il)ll( a @ z ) . Hence it is enough to show that
x, AeBm imply x@AeBm. Since x@ il= (x@1 ) (1@ A), it suffices by
(P-3) to prove x@ 1 ~ 9 Thus
~ . we wish to prove for each D that
(%,a):A-B+C
proper exact implies ( x @ l , a l ) :A@D-B@D+C
proper exact. Here we have a' =a n , where n: B @ D +B is a projection of the direct sum, hence proper by (P-2). Therefore al=a n
is proper by (P-37, hence x@ 1 Eker a' proper, as required.
Two proper short exact sequences E= (x, a) and E'= (x', a') from
A to C are called congruent if there is a morphism 8 with 8 x = x', a' 8=a.
By the short Five Lemma, any such 8 is necessarily an equivalence.
369
Theorem 4.3. For each proper class 9' of short exact sequences in
an abelian category d,Ext>(C, A) is a bifunctor on d.The addition
El+ E, = & (El@ E2)AC makes it a bifunctor to abelian groups.
The proof is like that for R-modules. The essential step is the demonstration that ~ x t is
> a contravariant functor of C; as in Lemma
111.1.2,we must construct to each proper E and each morphism y : C1+C
of d a unique commutative diagram
with first row E' proper exact (here 0 is the zero object 0'). First build
the right-hand square by the square construction of Thm. 1.1. By
Thm. 1.2, o' is epic. Form a second square
The couniversal property of the first square provides x': A+D with
B x ' = x and o'xl=O. The diagram (4.2) is now constructed and is
commutative.
To prove E' exact, consider any t with a'[ = O . Thus o p 5 = y 0'5 = 0,
so 5 factors through x~ ker a as /3 5=x a = p x'a for some a. But also
o1,$=0=o'x'a, so the couniversality of D for the coinitial maps 5 and
x'a with range D gives 5=x'u. Since any 5 with 0'5=0 factors through
x', and a'%'= 0, we have x ' ~
ker a'.
The proof that E' is proper uses a direct sum. By the square
construction, D, p, and o' are defined by the left exact sequence
on -yn,
O + D B@C1~
Mac Lane, Homology
C,
n,v=p, n , v = o l .
24
370
a,
371
Exercises
I , [BUCHSBAUM.]
Show that (P-2)may be replaced by the requirement that
u /?= 1
, implies PEPm.
2. [HELLER.] If %EPm and x u is a proper morphism, u is proper.
5. If 9 and 9
' are proper classes of short exact sequences, so is Bn 9'.
6. [HARRISON.] If S is a fixed module, show that the class of all short exact
A H- B +C with Hom (S, B) 4 Hom (S, C ) an epimorphism is a proper class.
372
The dual of this theorem asserts the exactness of the usual long
sequence with E placed in the second argument, as in Thm. 111.9.1.
For n = 0 and for EOld, the proof is that for modules, with minor variants.
To show ollxl, consider E'EEx~>(B,G) with E'x-0. This states
that E'x splits, so the definition (4.2) of E'x amounts to a commutative
diagram
E'X: O + G ~ ' - G @ A % A + O
Lemma 5.2. If xnl E" for all proper E , the% En]on+' and on+'l xn+l.
3 73
rill.
Next consider xl(El, which asserts that if FEE
Extl ( A ,G) has F E r 0,
then F r F 1 x E for some F'. To deal with this we must enter into the
several-step definition of the congruence relation FE=O. We actually
prove a little more :
Lemma 5.3. For FEExtl (A, G) and E E Extl (C, A ) , the followirzg
three properties are equivalent:
(i) F ~ F ' x E f o r ~ o m e F ' ~ E x t l ;
(ii) E= oFEf for some E'eExtl;
(iii) F E = o .
To prove that (i) implies (ii), write the commutative diagram for
the morphism F+Ff defining F f x Eas
..Lo,1%.
F': o-tG-+*--+
B-tO
with last column E. Here ,LL is monic by the square construction for
F'xE. Insert oEol at the dotted arrow. This morphism is proper epic and
also in coker p, by a proof like that for oo' in the previous diagram. The
middle column is now a proper short exact sequence E', and the
diagram states that oFE1=E, as required. The proof that (ii) implies
(i) is dual to this one.
The hypotheses of the lemma insure that FEcExt2(C,G) is defined, and (i) implies that F E = (F' xE)E =Ff(xEE) =F ' O = 0, which is
(iii). Dually, (ii) implies (iii). To prove the converse, let F # E denote
the property of F and E given by the equivalent statements (i) and (ii).
Now the zero of Ext2(C, G) has the factorization O=F,E,, with
374
The implication (iii)+ (i) will show xnl E n and complete the proof
of the theorem.
To prove that (i) implies (ii), write Sf as a composite TF', with
F'EExtl. This gives S- SixE= T(F1xE).Apply Lemma 5.3 to F= FixE
and E ; it proves E- a,Ef with S a,- T ( F u ~=) 0, which is (ii).
To prove that (ii) implies (i), use the induction assumption. Given
E E U E' and S a=O, write S as a composite T F with T~cExt"-l.
Now T(Fu)-0, so by induction [(iii) implies (i)] there is a T f ~ ~ E x t n - l
375
r:E +El:
r=
All triples
(a, @, y) of morphisms of d
which yield a commutative diagram
k B'-+
P C1+O.
b
,
E': O+Af+
,,,
r=
3 76
r,
r:D +Dl:
r=
All triples
(a, B, y) of morphisms of d
which yield a commutative 2 x 3 diagram, as above.
Since 9(d)
is a category of diagrams in an abelian category, it is
if and only if a, p, and y
abelian; moreover, (a, p, y) is epic in 9(d)
are all epic in d, and likewise for rnonics. A short exact sequence
D' H D +D" in Y ( d ) then corresponds to a commutative 3 x 3 diagram
r=
377
Wd)'
Proof. Given any commutative diagram in d ,
r:
c':
Theorem 6.3. If the abelian category d has enough proper projectives, the% to each proper short exact seqzcence E : A B+C of d there
i s a n allowable projective F and a n allowable epic Z= (5,7,5): F+E
of q d ) .
We will construct an F of the form given by Thm. 6.2. Since d has
enough proper projectives, we can find proper projectives P and Q
and proper epics 5 : P - t A , w : Q-tB. The composite 5 = o o : Q+C
is proper epic, while 7 = x 5 nl+ o n,: P @ Q +B provides a morphism
Z= (5,7 ,c): F+E. But 5 and epic, by the short Five Lemma, imply
7 epic. Hence Z is allowable by Prop. 6.1, provided only that 7 is proper.
378
3 79
Exercises
1. If A is the category of all left R-modules, show that every monic in Ses (A)
has a cokernel in S e s ( d ) , and dually. (Use the Ker-coker sequence.)
r=
2. A morphism
(a, p, y) : D +D' is allowable in Y ( d )if and only if D
and D' are proper short exact sequences of d and the induced map ker B -+ ker y
is proper
is proper epic in d (or, dually, the induced map coker a-tcoker
monic in d ) .
The systematic treatment of functors T: d - + W in the next sections ($5 7-9) will assume
(i) d is an abelian category,
(ii) B is a proper class of short exact sequences in d,
(iii) W is a selective abelian category ( 1 x 4 .
This formulation includes both relative homological algebra (e.g., with
B the class of suitably split exact sequences) and "absolute ' homological
algebra, with B all short exact sequences in d . In W we use the class
of all short exact sequences. For the applications intended, W might as
well be the category of all modules over some ring or algebra.
An additive functor T: d + W is a functor (covariant or contravariant) with T(a+/?) = T(u) T(B) whenever ct+/? is defined. This condition implies T(0)=O, T(-a)=- T(u), and T(A@ B) = T(A)@T(B).
Henceforth we assume: all fzlnctors are additive.
Study the effect of a covariant T upon all the firofier short exact
sequences ( x , o): A w B -+C of d. Call T
J
( u ) = i m ( T )
T (coim a) = coim ( T u) .
380
S (C) 3 T(A)
381
If a : A+A1, then
E,
=E
i,
(a E),=a,E,,
(7.2a)
a,= T(a).
(7.2~)
Indeed, (7.2) with a = I and y = I gives (a). If (I, /?, y) : E +El, then
E'y is by definition E , so (7.2) gives (b). Dually, (7.2) with y = I gives (c).
Conversely, given (a), (b), and (c) with (a, p, y): E +El, the congruence
a E = E'y of Prop. 111.1.8 gives (7.2).
If E, splits, then (E,), = 0. For, if E, splits, the morphism (IA,nl, 0 )
maps E, to the sequence A-A+O.
Since S is additive, S(0) =O, so
(7.2) gives O = S (0)=T(I) (E,), = (E,), .
For each proper E: A B +C, the long sequence
p)
3 82
Theorem 7.2. If d has enough proper firojectives, the following conditions on a 9-connected pair (S, E,, T) of covariant functors are
equivalent :
(i) (S, E, , T) i s left 9-couniversal,
(ii) For each proper short exact sequence K n P + C the sequence
0 -+ S (C) -+T(K)+T(P) ,
9,
(7.5)
with top row a complex, has its bottom row exact, by hypothesis.
Hence E, is monic, so that f (C), if it exists, is unique. On the other
hand, T(x) g(K) E#=g(P) Tf(x)E#=O, so g(K) E# factors through
E, E ker (T(x)) as g (K)E#= E, E for some unique f : Sf(C)+S (C). Take
f (C)= f . This fills in the dotted arrow to make the diagram commute.
Now take any proper short exact sequence E f = (x', a'): A'-Bf+C'
and any morphism y: C +C' of d . The diagram
E: o+K+P+C+O
El: o+A'+B'-+C'-+O,
d.
(7.8)
383
W.
(7.9)
r=
(r)
= T(a) p. Moreover HI
depends
which is characterized by p' Hl (r)
only on y, E, and E', and not on a and P. For, let = (a,, Po,y) : E +Ef
be any other morphism with the same y. In the diagram
r,
(r) (r,).
384
We now have
r=
385
T ( P )+ T ( K ) + S ( C ) + O ,
9.
25
386
5E (y) f (GI)= (- 1
(deg f,
f (G,) 51(y)
387
388
9. Derived Functors
389
5E ( y ) f (G,)= ( - I ) cdeg"
(deg f,
f (GI)T ( y )
9. Derived Functors
3 90
9. Derived Functors
394
(9.1)
in 9.
This sequence {T,, En) may also be described as a contravariant
+Wf. Given the right exact functor T: d + W , its
functor 5:
left derived functors may be characterized by their construction from
injective coresolutions or by either of the properties:
(Ha) To= T and 5 is couniversal; that is, given 5': 13~(d)
+9+,
each natural fo: Ti -+To extends to a unique natural f : 5'+5,
(I1b) To= T, (9.2) is always exact, and T, (J)= 0 for n>0 and each
proper injective J.
392
...
(9.3)
(9.4)
I
I1
I11
IV
Variance Derived
-
Right exact
Right exact
Left exact
Left exact
Co
Contra
Co
Contra
Resolution
Type, Tn(A)
couniversal
couniversal
universal
universal
projective
injective
injective
projective
Tor,, (G, A )
Left
Left
Right
Right
Extn(G, A)
Extn(A,G )
9. Derived Functors
393
394
a "directed" class; the direct limit of the kernels of T(A) -+ T(B) taken over the
directed class gives the right satellite of T [BUCHSBAUM 19601, defined in this
way without assuming that there are enough projectives. This construction has
[I9611 ; ROHRL[I9621 gives an existence theorem
been studied further by AMITSUR
for satellites of half exact functors, with applications to the theory of sheaves.
For any additive functor which is not half-exact one must distinguish the derived
functors, the satellites, and the cosatellites; their interrelations are studied in
BUTLER-HORROCKS [I 9611.
Derived functors of non-additive functors have been studied by DOLD-PUPPE
[I9611 using iterated bar constructions. Indeed, the homology groups H,+k (17,
n ; G)
of
provide many examples of non-additive functors (EILENBERG-MACLANE
[1954a]). The classical example is the functor of J. H. C. WHITEHEAD [1950].
For each abelian group A, r ( A ) is the abelian group with generators [y(a) / aEA],
relations y (- a)= y (a) and
y(a+b+c)-y(a+b)-y(a+c)-y(b+c)+y(a)+y(b)+y(c)=o.
These are the relations valid for a "square" y (a)= as.
395
defined (say) by Yoneda composition, and giving the usual long exact
sequence. Moreover, H P ( J )=O for $>O and J proper injective (any
extension of a proper injective splits). Hence the Hf'(C) are the right
derived functors of HO (C)= C".
Lemma 10.2. For each fixed integer q and each fixed 17-module C',
the functors HP(C) @Hq(C1) constitute the coqbonents of a universal
sequence of functors with 9-connecting homomor#hisms E , @ I .
is exact and 2-split, hence gives the required (natural) connecting maps.
Similarly, for 9 and C fixed, the functors HP(C) @Hq(C1)constitute
a universal 9-connected sequence, when the connecting homomorphisms
1@ E; are defined with the usual sign :
Moreover, the Hp+q(C @C1)constitute a @-connected sequence of functors of C' with connecting homomorphisms (C B E ' ) , .
For #= 0, H0 (C)= C" is the subgroup of 17-invariant elements of C.
Now ceCn and c ' E C ' ~give c BC'E
(C @C1)", so the identity induces
a homomorphism Cn @ Cfn+ (C @ C')".
396
(10.2)
fPIQ,
fi 2 0, q 2 0.
p 20, 920,
fi 1 0 , q 2 0 ;
The last two conditions assert that the maps f commute with the
connecting homomorphisms.
Proof. We are given /"PO. For q=O and C' fixed, the left hand
side of (10.2) is 8-universal, while the right hand side is @connected.
Hence the maps fP,O, natural in C, exist and are unique subject to (iii)
for q=O. These maps are also natural in C'. For consider y: C1+D'.
Then yf*>Oand ffi>Oy are two natural transformations of the 8-universal
functor HP (C) @HO(C') to the 8-connected functor HP (C @D') which
agree for fi = O and hence for all 9.
Now hold 9 and C fixed. In (10.2), f0.Q is given for q=O, and by
(iv) must be a natural transformation of a universal to a connected
sequence. Hence it exists and is unique; as before it is also natural
in C.
Our construction gives (iii) only for q=O; it remains to prove it
for q>O. For + fixed, let qP be the left-hand side and yQbe the righthand side of (iii). Both are maps
yQ,yQ: HP (C) @HQ(C') -+HP+Q+l(A
C')
(A@E'),~pq=(A@E'),ffi+l~Q(E,@
=l ffi+l?Q+'(~
)
BE;) (E* @I),
397
Let X be the abelian category of positive complexes K (of left modules over some ring), with morphisms all chain transformations
f
f : K +L. Call a short sequence of complexes ~4LAMfirofier exact
if, for all lz,
(i) O-+K,+L,+M,+O
is exact, and
(ii) 0 -t C, (K) +C, (L) +C, (M) +0 is exact,
where C,(K) denotes the module of n-cycles of K. Since (i) implies
that (ii)is left exact, (ii) may be replaced by
(ii') C, (L) +C, (M) is an epimorphism for all n .
In other words, a chain epimorphism g: L + M is proper if to each
mcM with a m = o there exists an ZEL with g l = m and al=0. Equivalently, a chain monomorphism f : K - t L is proper if to each ZEL
with a1 E f K there is a k c K with a1= a f k . With these characterizations,
the reader may verify that this class of proper short exact sequences
satisfies the axioms of $ 4 for propriety. Since a long exact sequence
is a Yoneda composite of short exact sequences, we have
Lemma 11.1. A sequence of comfilexes ... +K +L -+M +N -+. .. is
firofier exact if and only if, for every dimension n 2 0 , both . .. +K,+
L,+M,+N,+...
and .-.+C,(K)-+C,(L)-+C,(M)+C,(N)+.-are
exact.
Proposition 11.2. If K n L + M is a firofier short exact sequelzce
of comfilexes, then each of the following sequences is exact for all n:
(iii) 0 -+ B, (K) -t B, (L) +B, (M) +O,
(iv) o +Hn (K)+Hn (L) -tH, (M) +O,
(v) 0 +KnIBn (K)+Ln/Bn (L) +MnIBn (M)4 0 ,
(vi) o +K,/C, (K) -tL,/C, (L)+M,/C, (M) 4 0 .
398
with a induced by the identity Q,+Q,, so H(S) and B(S) are as stated.
The last assertion follows by induction from the fact that every extension by a projective module splits.
399
We can now prove that there are enough proper projective complexes.
i n which each
Y , is
{Y,,,} with
400
Remark. The often used fact that the tensor product of two complexes
is a bicomplex applies to functors other than the tensor product. Let
T ( A ,B) be a covariant bifunctor of modules A and B with values in
some additive category%. If K and L are positive complexes of modules,
T applies to give a bigraded object T(K*,L,) in % and the boundary
homomorphisms of K and L induce morphisms
a t = T ( % , 1 ) : T ( K p ,L,)-+T(K*-l, L,),
a t 1 = ( - 1 ) @ ~a,):
( ~ , T(K,, L,)+T(K,, L,J
which satisfy a' a' =0, at' a" =0, and a' a'' =- atla', the latter because
T is a bifunctor. Therefore T ( K ,L )= ( T ( K 9 ,L,), a', a") is a bicomplex
in V with an associated total complex Tot [ T ( K ,L ) ] . If homotopies
are to be treated, one assumes T biadditive; that is, additive in each
variable separately. When T is the tensor product, T ( K ,L) is the familiar bicomplex K @L.
Exercises
f
1. Let K-+L-M
be a sequence of complexes with gf = 0. Show that it is
a proper short exact sequence if an only if both (iii) and (iv) of Prop. 11.2 hold,
and also if and only if both (ii) and (iii) hold. Find other sufficient pairs of conditions.
2. Show that every proper projective positive complex has the form given in
Lemma 11.4.
3, Show that Zlt,(K, L) is independent of the choice of the resolution of L,
and prove that it can also be computed from a proper projective resolution of K,
or from resolutions of both K and L.
4. Study proper exact sequences for complexes not necessarily positive.
5. Let 4" be a proper class of short exact sequences in an abelian category d.
Study the corresponding proper class in the abelian category of positive complexes
in d .
6. Each additive functor T: d-4 induces a functor T on d-complexes K
to 9-complexes. For S left exact and T right exact, construct natural maps
Extend to bifunctors, and obtain the homology product as a special case for T = 8.
(12.1)
Formula
401
The hypothesis (12.1) for this theorem requires that each of the
complexes Tor, (K, L), defined as in the remark of 9 11, has zero homology for m>o. The stronger hypothesis that each K, is flat would
imply that each Tor, (K, L)= 0 for m>O, hence (12.1).
Theorem (Thm. V.10.2)
For positive complexes, the previous KWNNETH
is included in this one. In detail, the hypotheses of that theorem required
that C, (K) and B,(K) be flat; i.e., that Torp(C,, G) =O=Torp (B,, G)
for all G and 9 >O. Since C, (K) H K, +B,-, (K) is exact, the following
portion of the standard exact sequence for the torsion product
E;:,
EX: = s+t=q
2 Tor* (Hs(K), H, (L)).
(12.4)
Under the previous hypothesis (12. I), the first sequence collapses
to the base, gives 8, s E:, =H, (K @ L), hence yields the result of the
first theorem.
26
402
By totalizing the first and third indices, construct a double complex with
EX:.
403
Proof. Over 2,Torp vanishes for + > I , so the first spectral sequence
has only two non-vanishing rows (q=O, q= 1) and only one non-zero
differential d2: E:,, +E:-a, ,;hence the exact sequence
d'
Bibliography
(With most listings the citation X.8 (to Chap. X, 8) indicates where the article
a t issue is quoted in this book.)
ADAMS,J. F.: On the Cobar Construction. Proc. NAS USA 42, 409-412 (1956).
x.13
- On the non-Existence of Elements of Hopf Invariant One. Ann. of Math. 72,
20-104 (1960). VII.6; X.8
AMITSUR, S. A.: Derived Functors in Abelian Categories. J. Math. and Mech. 10,
971-994 (1961). XII.9
ARTIN,E . : Galois Theory. Notre Dame Math Lectures, No. 2, Notre Dame (Ind.),
2nd ed. 1944. IV.2
-, and J. T ATE: Class Field Theory. (Mimeographed notes) Cambridge : Harvard University 1960. IV. 11
ASANO, K., and K. SHODA:Zur Theorie der Darstellungen einer endlichen Gruppe
durch Kollineationen. Comp. Math. 2, 230-240 (1935). IV.11
ASSMUS,
E. F., Jr.: On the Homology of Local Rings. Ill. J. Math. 3, 187-199
(1959). VII.7
ATIYAH,M. F.: On the Krull-Schmidt Theorem with Application t o Sheaves.
Bull. Soc. Math. France 84, 307-317 (1956). IX.2
- Characters and Cohomology of Finite Groups. Pub. Math. No. 9, Inst. Hautes
Etudes, Paris.
AUSLANDER, M.: On the Dimension of Modules and Algebras. 111. Nagoya Math.
J. 9. 67-77 (1955). VII.1
-, and D. A. B UCHSBAUM: Homological Dimension in Noetherian Rings. Proc.
NAS USA 42, 3 6 3 8 (1956). VII.7
- - Homological Dimension in Local Rings. Trans. AMS 85, 390-405 (1957).
- - Homological Dimension in Noetherian Rings. 11. Trans. AMS 88, 194-206
(1958). VII.7
- - Codimension and Multiplicity. Ann. of Math. 68, 625-657 (1958). VII.7
- - Unique Factorization in Regular Local Rings. R o c . NAS USA 45, 733734 (1959). VII.7
BAER, R.: Erweiterung von Gruppen und ihren Isomorphismen. Math. 2. 38,
375-416 (1934). 111.11; IV.11
- Automorphismen von Erweiterungsgruppen. Actualit& Scientifiques e t Industrielles, No. 205. Paris 1935. 111.11; 1V.11
- Abelian Groups t h a t Are Direct Summands of Every Containing Abelian Group.
Bull. AMS 46, 800-806 (1940). 111.1 I
BARRATT, M. G. : Homotopy Ringoids and Homotopy Groups. Q. J. Math. Oxon
(2) 5, 271-290
(1954). Ix.1
-, V. K. A. M. G UGENHEIM and J. C. M OORE: On Semisimplicial Fibre Bundles.
Am. J. Math. 81, 639-657 (1959). VIII.9
BASS, H. : Finitistic Dimension and a Homological Generalization of Semi-primary
Rings. Trans. AMS 95, 466-488 (1960). V.4; VII.l
BOCKSTEIN, M. : Sur le spectre d'homologie d'un complexe. C. R. Acad. Sci. Paris
247, 259-261 (1958). V.ll
Bibliography
40 5
BOCKSTEIN, M. : Sur la formule des coefficients universels pour les groupes d'homologie. C. R. Acad. Sci. Paris 247, 3 9 6 3 9 8 (1958). V.11
BOREL, A.: Sur la cohomologie des espaces fibres principaux e t des espaces homogenes de groups de Lie compacts. Ann. of Math. 57, 115-207 (1953). VI.9
- Nouvelle demonstration d'un theoreme de P. A. Smith. Comment. Math. Helv.
29, 27-39 (1955). X I . l l
BOURBAKI, N. : Algebre MultilinBare. (Chap. I11 of Book I1 in Elements de Mathematique) Hermann, Paris 1948. 2nd ed. 1951. V . l l
BRAUER, R.: Untersuchungen iiber die arithmetischen Eigenschaften von Gruppen linearer Substitutionen. I. Math. 2. 28, 677-696 (1928). 111.11
BROWDER, W.: Torsion in H-Spaces. Ann. of Math. 74, 24-51 (1961). 11.4; XI.5
BROWN, E. H., Jr.: Twisted Tensor Products. I. Ann. of Math. 69,223-246 (1959).
VIII.9
BUCHSBAUM, D. A.: Exact Categories and Duality. Trans. AMS 80, 1-34 (1955).
IX.2
- A Note on Homology in Categories. Ann. of Math. 69, 66-74 (1959). XII.4;
XII.5
- Satellites and Universal Functors. Ann. of Math. 71, 199-209 (1960). XII.4;
XII.9
BUTLER, M. C. R., and G. HORROCKS: Classes of Extensions and Resolutions.
Phil. Trans. Roy. Soc. London 254, 155-222 (1961). XII.4
CARTAN,
H. : Sur la cohomologie des espaces oh opere un groupe. C . R. Acad. Sci.
Paris 226, 148-1 50, 303-305 (1948). XI.7
- SBminaire de Topologie ~ l ~ B b r i q u;e950-1
,
95 1 (l?cole Norm. Sup.). Paris 195I .
111.11
- Extension du theoreme des "chatnes de syzgyies". Rend. Matem. Appl. Roma
11, 156-166 (1952). VII.6
- Sur les groupes d'Eilenberg-MacLane H ( n , lz) : I. MBthode des constructions;
11. Proc. NAS USA 40, 467-471, 704-707 (1954). X.13
- Shminaire (&ole Norm. Sup.). Algebres d'Eilenberg-MacLane et Homotopie.
Paris 1955. X.12; XI.91
--, and S. EILENBERG: Homological Algebra. Princeton 1956.
-, and J. L ERAY : Relations entre anneaux d'homologie e t groupe de PoincarB.
Colloque Topologie Algbbrique, Paris 1949, p. 83-85. XI.7
CECH, E.: Les groupes de Betti d'un complex infini. Fund. Math. 25, 33-44
(1935). V . l l
CHEVALLEY, C. : Fundamental Concepts of Algebra. Academic Press, New York
1956. VI.9
-, and S. E ILENBERG: Cohomology Theory of Lie Groups and Lie Algebras.
Trans. AMS 63,85--124 (1948). X.13
COHN, P. M.: On the Free Product of Associative Rings. Math. 2. 71, 380-398
(1959) ; 73, 433-456 (1960). VI.4
DIEUDONNB,
J.: Remarks on quasi-Frobenius Rings. Ill. J. Math. 2, 346-354
(2958). V.4
DIXMIER, J.: Homologie des anneaux de Lie. Ann. Sci. Ecole Norm. Sup. 74,
25-83 (1957). X.13
DOLD, A.: Homology of Symmetric Products and Other Functors of Complexes.
Ann. of Math. 68, 54-80 (1958). VIII.5
- Zur Homotopietheorie der Kettenkomplexe. Math. Ann. 140, 278-298 (1960).
11.4
- Uber die Steenrodschen Kohomologieoperationen. Ann. of Math. 73, 258-294
(1961). VIII.8
406
Bibliography
Bibliography
407
EILENBERG, S., and J. MOORE: Limits and Spectral Sequences. Topology 1, 123 (1962). xI.11
-, and N. STEENROD: Foundations of Algebraic Topology. Princeton 1952. 11.9
-, and J. A. ZILBER: Semi-simplicial Complexes and Singular Homology. Ann.
of Math. 51, 499-513 (1950). VIII.5
- - On Products of Complexes. Am. J. Math. 75, 200-204 (1953). VIII.5
EVENS, L.: The Cohomology Ring of a Finite Group. Trans. AMS 101, 224-239
(1961). X I . l l
EVERETT, C. J.: An Extension Theory for Rings. Am. J. Math. 64, 363-370
(1942). x.13
FADELL,
E., and W. HUREWICZ: On the Structure of Higher Differential Operators
in Spectral Sequences. Ann. of Math. 68, 314-347 (1958). XI.3; XI.11
FEDERER, H.: A Study of Function Spaces by Spectral Sequences. Trans. AMS
82, 340-361 (1956). XI.11
FITTING, H.: Beitrage zur Theorie der Gruppen endlicher Ordnung. Jber. DMV
48, 77-141 (1938). IV.ll
FREUDENTHAL,
H. : Der EinfluD der Fundamentalgruppe auf die Bettischen Gruppen. Ann. of Math. 47, 274-316 (1946). IV.11
FREYD, P.: Functor Theory, Dissertation, Princeton University, 1960. XII.3
FR~HLICH,
A.: On Groups over a d.g. Near Ring. 11. Categories and Functors.
Q. J. Math. Oxon (2) 11, 211-228 (9960).
- Non Abelian Homological Algebra. I. Derived Functors and Satellites. Proc.
London Math. Soc. (3) 11, 239-275 (1961); 11. Varieties. Proc. London
Math. Soc. (3) 12, 1-28 (1961).
FRUCHT, R. : Zur Darstellung endlicher Abelscher Gruppen durch Kollineationen.
Math. Z. 63, 145-155 (1955). IV.11
FUCHS,
L.: Abelian Groups, Budapest, 1958. Also published by Pergamon Press,
Oxford and New York, I 960.
- Notes on Abelian Groups. I. Ann. Univ. Sci., Budapest, Sect. Math. 2, 5-23
(1959). XII.4
GABRIEL, P.: Des categories abeliennes. Bull. Soc. Math. France 90, 323-448.
(1962) IX.2
GODEMENT,
R.: Theorie des Faisceaux. Hermann, Paris 1958. 1.8; IX.4
GODEL,K. : The Consistency of the Continuum Hypothesis. Ann. of Math. Studies3.
Princeton 1940. 1.7
GRAY, J. W. : Extensions of Sheaves of Algebras. Ill. J. Math. 5, 159-1 74 (1961).
X.13
- Extensions of Sheaves of Associative Algebras by non-trivial Kernels. Pac.
J . Math. 11, 90+-917 (1961). X.13
GREEN, J . A.: On the Number of Automorphisms of a Finite Group. Proc. Roy.
Soc. London, Ser. A 237, 574-581 (1956). XI.11
GROBNER,
W.: uber die Syzgyien-Theorie der Polynomideale. Monatsh. Math.
53, 1-16 (1949). VII.6
GROTHENDIECK, A. : Sur quelques Points d'Alg8bre Homologique. Tohoku Math. J.
9, l i e 2 2 1 (1957). 1x.2; 1 x . 4 ; x11.8
- (with J. DIEUDONN*)
: l&ments de Mometrie Algebrique. I, 11. Pub. Math.
Inst. des Hautes Etudes. Paris 1960, 1961. Nos. 4 and 8. 1.8
GRUENBERG, K. W.: Resolutions by Relations. J. London Math. Soc. 35, 481494 (1960). X.5
GUGENHEIM,
V. K. A. M.: On a Theorem of E. H. Brown. Ill. J. Math. 4, 292311 (1960). VIII.9
408
Bibliography
GUGGENHEIM,
V. K. A. M. : On Extensions of Algebras, Co-algebras, and Hopf
Algebras I. Am. J. Math. 84, 349-382 (1962).
-, and J. C. MOORE: Acyclic Models and Fibre Spaces. Trans. AMS 85, 265-3136
(1957). VIII.7; X I . l l
HALL, M.: Group Rings and Extensions. I. Ann. of Math. 39, 220-234 (1938).
IV.1.1
HALPERN, E.: Twisted Polynomial Hyperalgebras. Memoir AMS 29. Providence,
1958. VI.9
HASSE, H., R. BRAUER, and E. NOETHER:Beweis eines Hauptsatzes in der Theorie
04 (1932). 111.1 I
der Algebren. J. reine angew. Math. 167, 39HARRIS, B. : Cohomology of Lie Triple Systems and Lie Algebras with Involution.
Trans. AMS 98, 148-162 (1961). X.13
HARRISON, D. K.: Infinite Abelian Groups and Homological Methods. Ann. of
Math. 69, 3 6 6 3 9 1 (1959); 71, 197 (1960). XII.4
- Commutative Algebras and Cohomology. Trans. AMS 104,191-204; (1962).
X.13
HATTORI, A.: On Fundamental Exact Sequences. J. Math. Soc. Japan 12, 65-80
(1960). XI.11, ex. 4
HELLER, A. : Homological Algebra in Abelian Categories. Ann. of Math. 68, 484525 (1958). XII.4; XII.6
HILBERT, D. : uber die Theorie der Algebraischen Formen. Math. Ann. 36, 473534 (1890). V11.6
HILTON,P. J., and W. LEDERMANN: Homology and Ringoids. I. R o c . Camb.
Phil. SOC.54, 152-167 (1958). IX.l
-, and D. REES: Natural Maps of Extension Functors and a Theorem of R. G.
Swan. Proc. Camb. Phil. Soc. 57, 489-502 (1961). XII.9
- ,and S. W YLIE: Homology Theory. Cambridge 1960. 11.9; VIII.9; XI.7
HOCHSCHILD,
G.: On the Cohomology Groups of a n Associative Algebra. Ann. of
Math. 46, 58-67 (1945). VII.4; VII.7
- On the Cohomology Theory for Associative Algebras. Ann. of Math. 47, 568579 (1946). VII.7
- Cohomology and Representations of Associative Algebras. Duke M. J. 14,
921-948 (1947). VII.4; IX.3; X.13
- Local Class Field Theory. Ann. of Math. 51, 331-347 (1950). IV.ll
- Cohomology Classes of Finite Type and Finite Dimensional Kernels for Lie
Algebras. Am. J. Math. 76, 763-778 (1954). X.13
- Lie Algebra Kernels and Cohomology. Am. J. Math. 76,698-716 (1954). X. 13
- Relative Homological Algebra. Trans. AMS 82, 2 4 6 2 6 9 (1956). 1 x 3 ; XII.4
-, and J.-P. SERRE: Cohomology of Group Extensions. Trans. AMS 74, 110-134
(1953). XI.10; X I . l l
HOLDER, 0.: Die Gruppen der Ordnungen pS, p qa, p q r, p4. Math. Ann. 43,
3 0 1 4 1 2 (1893) (especially 3 18). 111.11
HOPF, H. : u b e r die Abbildungen der dreidimensionalen Sphare auf die Kugelflache. Math. Ann. 104, 637-665 (1931). XI.2
- uber die Abbildungen von Spharen auf Spharen niedrigerer Dimension. Fund.
Math. 25, 427-440 (1935). XI.2
- Fundamentalgruppe und zweite Bettische Gruppe. Comment. Math. Helv. 14,
257-309 (1941142). I V . l l
- Relations between the Fundamental Group and the Second Betti Group.
Lectures in Topology. Ann. Arbor 1942. IV.ll
- uber die Bettischen Gruppen, die zu einer beliebigen Gruppe gehoren. Comment.
(1944145). 111.11 ; I V . l l
Math. Helv. 17, 3-79
Bibliography
409
Hu, S.-T. : Homotopy Theory. Academic Press, New York and London 1959. XI.2;
x1.7
HUBER, P. J.: Homotopy Theory in General Categories. Math. Ann. 144, 361-385
(1961).
HUREWICZ, W. : Beitrage zur Topologie der Deformationen. Proc. Akad. Amsterdam 38, 112-119, 521-538 (1935); 39, 117-125, 215-224 (1936). I V . l l
- On Duality Theorems. Bull. AMS, abstract 47-7-329 47. 562-563 (1941). 1.8
JACOBSON, N. : Structure of Rings. AMS, Providence, 1956.
- Lie Algebras. New York: [Interscience] John Wiley & Sons 1962. X. 13
JANS,J. P.: Duality in Noetherian Rings. Proc. AMS 82+-835 (1961). V.4;
v11.7
KAN, D. M.: Adjoint Functors. Trans. AMS 87, 294-329 (1958). IX.6
- A Combinatorial Definition of Homotopy Groups. Ann. of Math. 67, 282-312
(1958). VIII.5
KAPLANSKY,
I . : On the Dimension of Modules and Algebras. X . Nagoya Math.
J. 13, 85-88 (1958). VII.l
KASCH,F.: Dualitatseigenschaften von Frobenius-Erweiterungen. Math. Z. 77,
21-227
(1961). VII.7
KELLEY, J. L., and E. PITCHER: Exact Homomorphism Sequences in Homology
Theory. Ann. of Math. 48, 682-709 (1947). 1.8
KLEISLI, H.: Homotopy Theory in Abelian Categories. Can. J. Math. 14, 139169 (1962). XII.4
KOCHENDORFFER,
R.: Uber den Multiplikator einer Gruppe. Math. Z. 63, 507513 (1956). Iv.11
K o s z u ~ J.
, L. : Sur les opbrateurs de dbrivation dans un anneau. C. R. Acad. Sci.
Paris225 217-219(1947).
xI.11
- Sur un type dlalg&bresdiffkrentiellesen rapport avec la transgression. Colloque
de Topologie, Brussels 1950, p. 73-41. VII.2
- Homologie et cohomologie des algebres de Lie. Bull. Soc. Math. France 78,
65-127 (1950). X.13
KUDO, T., and S. ARAKI: Topology of H,-Spaces and H-Squaring Operations.
Mem. Fac. Sci. Kyusyu Univ., Ser. A 10, 85-120 (1956). XI.11
KWNNETH,
H.: Uber die Bettischen Zahlen einer Produktmannigfaltigkeit. Math.
Ann. 90. 65-85 (1923). V.10
- Uber die Torsionszahlen von Produktmannigfaltigkeiten. Math. Ann. 91, 125134 (1924). V.10
LAMBEK, J.: Goursat's Theorem and the Zassenhaus Lemma. Can. J . Math. 10,
45-56 (1958). 11.9
LANG, S.: Review of "$lbments de gbombtrie algbbrique". Bull. AMS 67, 239246 (1961). 1.8
LEFSCHETZ,
S. : Algebraic Topology. AMS, New York 1942.
LERAY, J. : Structure de l'anneau d'homologie d'une reprksentation. C. R. Acad.
Sci. Paris 222, 1419-1422 (1946). XI.2; X I . l l
- L'anneau spectral et l'anneau filtrb d'homologie d'un espace localement compact et d'une application continue. J . Math. Pures Appl. 29, 1-139 (1950).
XI.2; X I . l l
- L'homologie d'un espace fibrb dont la fibre est connexe. J. Math. P u r e Appl.
29, 16+-213 (1950). XI.2; XI.11
LIULEVICIUS,
A.: The Factorization of Cyclic Reduced Powers by Secondary
Cohomology Operations. Proc. NAS USA 46, 978-981 (1960). X.8
LORENZEN, P.: Uber die Korrespondenzen einer Struktur. Math. 2. 60, 61-65
(1954). 11.9
410
Bibliography
412
Bibliography
SZCZARBA,
R. H.: The Homology of Twisted Cartesian Products. Trans. AMS
100, 197-216 (1961). VIII.9
S ZENDREI , J.: On Schreier Extension of Rings without Zero-divisors. Publ. Math.
Debrecen 2, 2 7 6 2 8 0 (1952). X.13
TATE, J.: The Higher Dimensional Cohomology Groups of Class Field Theory.
Ann. of Math. 56, 294-297 (1952). IV.ll
- Homology of Noetherian Rings and Local Rings. Ill. J. Math. 1, 14-27 (1957).
v11.7
TEICHMULLER,
0.: Uber die sogenannte nicht kommutative Galoissche Theorie
und die Relation .. . . Deutsche Math. 5, 138-149 (1940). IV.ll
VENKOV, B. B.: Cohomology Algebras for some Classifying Spaces [Russian].
Dokl. Akad. Nauk. SSSR 127, 943-944 (1959). XI.11
WALL, C. T. C.: Resolutions for Extensions of Groups. Proc. Camb. Phil. Soc.
57, 251-255 (1961). X.5
- On the Cohomology of Certain Groups. Proc. Camb. Phil. Soc. 57, 731-733
(1961). x . 5
WALLACE, A. H.: An Introduction to Algebraic Topology. Pergamon, London,
1957. 11.9
WATTS, C. E.: Intrinsic Characterizations of some Additive Functors. Proc. AMS
11, 5-8 (1960). XII.9
WEDDERBURN, J. H. M.: Homomorphisms of Groups. Ann. of Math. 42, 486-487
(1941). 11.9
WHITEHEAD, J. H. C. : A Certain Exact Sequence. Ann. of Math. 52, 51-1 10
(1950). XII.9
- On Group Extensions with Operators. Q. J. Math. Oxon (2) 1, 219-228
(1950).
WHITNEY, H.: Tensor Products of Abelian Groups. Duke Math. J . 4, 495-528
(1938). V.ll
YAMAGUTI, K.: On the Cohomology Space of a Lie Triple System. Kumamoto J.
Sci. A 5, 44-51 (1960). X.13
YONEDA, N.: On the Homology Theory of Modules. J. Fac. Sci. Tokyo, Sec. I
7, 193-227 (1954). 111.11
- Notes on Products in Ext. Proc. AMS 9, 873-875 (1958). VIII.4
- On Ext and Exact Sequences. J. Fac. Sci. Tokyo, Sec. I 8, 507-526 (1960).
111.11; XII.4; XII.9
ZASSENHAUS, H.: The Theory of Groups. 2nd ed., New York, 1958 (especially
p. 237: Morphisms). 11.9
ZEEMAN,
E. C.: A Proof of the Comparison Theorem for Spectral Sequences.
Proc. Camb. Phil. Soc. 53, 57-62 (1957). XI.ll
*.
z,
v,
a, r
di
bidim
Dimension as a bimodule
(VII.5)
cls
(Homology) class of
Coim, coim Coimage
Coker, coker Cokernel
Def
deg
Ext, ext
414
h. dim
Homological dimension
(VII.1)
Hom, hom Group of homomorphisms
Im, im
Image
Ind
Indeterminacy (of an additive relation) ; (11.6)
u
T,I
Ker, ker
1.gl.dim.
Tor
Kernel
Left global dimension of a
ring (VII.1)
Torsion product
Isomorphism; Equivalence in a
category (IX.1)
Injection of direct sum (1.4.1)
L
x
Monomorphism
A
Monomorphism
ni Projection of a direct sum (1.4.1)
n, n~ Product map of an algebra (VI. $2)
Q
Epimorphism
a, z Epimorphism
z
Middle four exchange (VI.2.4);
(VI.8.4)
y
Codiagonal map of coalgebra or
Hopf algebra (VI.9)
Index
Abelian category 254
Relative - - 263
Selective - - 256
Abelian group I0
Divisible - - 93
Acyclic space 5 7
Additive
- category 249
- functor 23, 71, 263, 379
- relation 51, 366 (Note)
Adjoint associativity 95, 144, 194
Adjoint, left 266, 269 (Note)
Affine independence 54
- simplex 54
- transformation 54
Alexander-Whitney map 241
Algebra 173
Augmented
180
Bigraded - 180
Exterior - I79, 183
Graded - 177
Hopf - I 9 8
Opposite - 182
Polynomial - 179, 182
Separable - 214
Symmetric - 184 (Ex.)
Tensor - 179
Total matrix - 214
Ungraded - 180
~llowableclass (in a category) 260
Analysis (of a morphism) 254
Annihilator 146
Anti-isomorphism 364 (Ex.)
Associative law I73
- - for diagonal map I97
- - for tensor product 142, 145, 194
Associativity 25, 249
Adjoint - 95, 144, 194
Middle - 138, 186
Outside - 187
Augmentation 180
- of DG-algebra 192
- of DG-module 304
- of graded algebra I80
- of group ring 104
Augmentation
- of singular complex
57
Automorphism 104
Inner - 124
Outer - 124
- class 124
Axioms
- for cohomology of algebras 287
- for cohomology of groups 117
- for Ext 99, 101, 269
- for homology of algebras 289
- for Tor 162, 275
Index
Canonical map II
comparison 267
Cartesian product 18
- - of modules 18
- - of sets 8
- - of simplicia1 sets 238
Category 25
Abelian - 254
Abstract - 28 (Ex.)
Additive - 249
Graded additive - 370
Opposite - 28
Preadditive - 250
Selective - 256
Small - 26
Cauchy's Theorem 131
Center of a group 124
Chain 35
- complex 39
- equivalence 40
- homotopy 40
- transformation 40
Change
- of algebras 289
- of groups 108
- of rings 90, 276, 293
Class 26
Automorphism - 124
Characteristic - 120
Conjugation - 224
Cleft extension 284
Coalgebra 197
Coboundary 42, 43, 116
Cochain 42
Cocycle 43
Codiagonal map 68, 251
Codimension 218 (Note)
Cohomology 42
Equivariant - 136
- of a g r o u p 115
- product 222, 296
Coimage 12, 41, 255
Coinitial maps 17
Cobounded filtration 346
Cokernel 12. 41, 252
Commutation rule for signs 164
Commutative
Skew and anti - 178
Strictly - 179
- coalgebra 179
- DGA-algebra 312
- diagram 13
Commutator subgroup 290
Comodule 198
Comparison Theorem 87, 355
Allowable - - 261
Canonical - - 267
Condensed canonical - - 305
Complex I, 39
Chain - 39
Cochain - 42
Complete semi-simplicial - 236
(Note)
Elementary special - 42 (Ex.)
Free - 87
Injective - 95
Negative - 41
Positive - 41
Projective - 87, 261
Quotient - 41
Singular - 56
Special - 42 (Ex.)
- of complexes 301
- over a module 87
- under a module 95
Composite
Yoneda - 82
- extension 65, 82
- function 9
- homomorphism 10, 195
- morphism 25, 249
- relation 52
- sequence 82
Composition principle 53
Condensation 302
Congruence
- of extensions 64, 109, 285, 368
- of n-fold sequences 83, 370
Conjugate module 146
Conjugation 124, I3I
Connected
- DG-algebra 192
- pair of functors 380
- sequence of functors 386
Connecting homomorphism 45, 51, 96,
I62
Iterated - - 97
Connecting relation (for complexes) 333
Connection (cohomology of groups) 347,
349
Construction 304
Relatively free - 304
Contractible space 58
Contravariant functor 22, 29
Convergence (of spectral sequence) 327,
329
417
Index
Converse (of a relation) 52
Convex hull 54
Coresolution 95
Correspondence (additive relation)
366 (Note)
Coset (of a subgroup) 11
Coterminal homomorphisms 16, 359
Counit of coalgebra 197
Couniversal 17
- functor 381, 387
- square 359
Couple, exact 336
Covariant functor 28, 386
Cycle 35, 39
Relative - 61
Cyclic module 20
Cylinder 59
DG-Algebra 190
DGA-Algebra I92
Degeneracy operator 235
Degree 175
Complementary - 327
Filtration - 327
Total - 176, 318
Derived
- couple 336
- functor 389, 393 (Note)
Diagonal
Simplicid - 244
- homomorphism 68
- map 197
- morphism 251
Diagram 33
Category of - 257
Commutative - I3
direct product - 27
direct sum - 27
Schematic - 263
Diagram chasing 14, 364
Differential 35, 318
- graded algebra 1go
- graded augmented algebra 192
- graded module 189
- group 35
- module 10
- on Hom (K, L) 43, 190
- on Hornu 191
Dimension 39
Finitistic - 203
Homological - 201
Krull - 220
Left global - 202
Mac Lane, Homologie
Dimension
Left injective - 203
Right - 203
Weak - 203
Direct product 27, 32
Semi - - 105
- - of algebras 212, 295
- - of morphisms 33
Direct sum 15, 27
External - - I5, 18
Internal - - 19
Semi - - 286
Divisible abelian group 93
Domain
- of definition of relation 52
- of function 9
- of homomorphism 10
- of morphism 25
Dual 27
- basis 147
- module 146
- statement 27
Edge homomorphisms 321, 335
Eilenberg-Zilber Theorem 239, 241
Endomorphism 10, 143
Ring of - 21, 143
Enough projectives 261, 367 (Ex.),
377
Envelope, injective 103
Enveloping algebra 317
Epic 252
Epimorphism 10, 251
Equivalence 25, 252
Chain
40
Left - 252
Right - 252
- principle 53
Equivariant cohomology 136
Essential extension 102
Euler characteristic 323
Exact
- couple 336
- functor 263, 379
- homology sequence 45
- sequence 11, 256
- triangle 193 (Ex.), 336
Half - functor 379
Left - functor 379
Left - sequence 23, 365
Proper - sequence 367,
3 70
Right-sequence 23, 379
27
418
Excision axiom 62
Extension
Algebra - 284
Central group - 112
Cleft - 284
Crossed product - 125
Essential - 102
Ground ring - 213, 294
Group - 108
Module - 63
Morphism of - 63, 109
Operator - 108
Singular - 284
Exterior algebra 174, 179, 183
Face operator 23 5
Factor set I11, 285
Factor system 69
Factorization, standard 254
- through 12, 66, 252
Fiber
- map 322
- terms (of spectral sequence) 321
Fibred product 359
Filtration 165, 309, 326
Bounded - 327, 330
Canonically bounded - 330
Convergent - 329
Descending - 346
First - 341
Second - 342
- of bar construction 309
- of bicomplex 341
- of tensor product 165
Finite type (module) 20, 219
First quadrant spectral sequence 320
Five Lemma 14, 365 .
Flat module 163
Four Lemma 14, 364
Free
- graded module 195
- group 122
- module 19
- ring (over a group) 104
Relatively - module 196
Function 8
Mapping - 29, 31
Object - 28, 31
Functor 28
Additive - 23, 71, 263, 379
Contravariant - 22, 29
Covariant - 22, 28, 379, 386
Exact - 263, 379
Index
Functor
Faithful - 263
Forgetful - 262
Half exact - 3 79
Left derived - 389
Left exact - 379
Normalized - 259
Right exact - 3 79
Generators (of a module) 20
Graded algebras I77
Internally - 180, 21 5
Tensor product of - I 8 1
Graded modules 175
Associated - 326
Internally - 177
Positively - 175
Trivially - 175
Graded object 177 (Note)
Graded set 177 (Note)
Graph of homomorphism 52
Group
Change of - 108
Cohomology - 115
Differential - 35
Free - 122, 123
Relative homology - 61
Singular cohomology - 57
- algebra 199. 295 (Ex.)
- ring 104
Group extension 108
Pure - - 367
Gysin sequence - 326
Hilbert Syzygy theorem 21 7
Hochschild (co)homology modules 283,
288 '
Holomorph of a group 105 (Ex.)
Homogeneous
- elements 177
- generator 119
- ideals 178
non - generator 118, 1 19
Homological dimension 201
Homologous cycles 35, 40
Homology 3 5
- classes 35, 39
- group 35, 57
- isomorphism 310
- modules 39
- product 166, 221, 296
Homomorphism 10
Boundary
35, 56, 235
419
Index
Homomorphism
Composition of - 1I,195
Connecting - 45, 5I,96
Crossed
105, 284
Diagonal - 68
Edge - 321
Induced - 13, 36
Principal crossed - 106, 284
- of algebras 177
- of bimodules 143
- of coalgebras 197
- of DGalgebras 190
- of differential groups 36
- of graded algebras I77
- of graded modules 175
- of Hopf algebras I98
- of A-modules 184
- of modules 10
- of spectral sequences 319
Homotopy 39. 57
Chain - 40
Contracting - 41, 265, 267
Homotopy classification theorem 78
Hyperhomology groups 400, 402
Ideal 10
Graded (two-sided) - I78
Graded left - I78
Graded right - 178
Homogeneous - I80
Nilpotent - 286
Prime - 218
Proper
218
Identity
- element 9
- function 9, I 0
- morphism 25, 249
Image 10, 255
Inverse
13, 363
- of morphism 255
Indeterminacy 52
Induced
- additive relation 53
- homomorphism 13, 36, 255
- relation 53
Inessential extension I 0 2
Inflation homomorphism 347
Injection (identity) 9
Injective
Allowable - 261
- complex 95
- envelope 103
- function 251, 362
Injective
- module 92
- object (in a category) 261
- resolution 95
Interchange
Hom-8 - 195
Middle four - I94
Internal
- direct sum 19
- grading 177. 180, 215
- homology product 221
Intersection 13, 18, 364 (Ex.)
Invariant element 122, 284
Inverse 11, 25
Left - II
- image 13
Isomorphism 10
Modular Noether - 318
Natural - 29
Noether - 13
Iterated connecting homomorphism 97
Ker-coker sequence 50
Kernel 10, 252, 362, 363
Abstract
124
Koszul Resolution 205, 218 (Ex.)
Kronecker product I82
Kiinneth
- Formula 166
- Formula for abelian groups 168
- spectral sequence 400
Left
Index
Map
Simplicia1
23 5
Mapping
- cone 46, 47
- cylinder 46
- function 29, 31
- theorem 321
Middle
- associative 138
- four interchange 194
- linear 138
Module 9
Bigraded
176
Cyclic - 20
Differential graded - 189
Flat - 163
Free
19, 195
Relatively - - 196
Free graded - 215
Graded K- - 175
Graded U-- 191
Injective - 92
Internally graded
1 77
Left A- - 184
Left R- - 9
Monogenic - 20
Projective - 20
Quotient - 11
Reduced - 304
Relative projective - 265, 273
Right - 9, 138
Semi-simple
203
Simple - 203
Simplicia1- 233
Submodule - 10
Trigraded - 176
Trivial - 105
Trivially graded - 175
Unitary - 9
Monic 251
Monogenic module 20
Monoid 2 w
Monomorphism 10, 251
Monotonic maps 233
Morphism 25, 249
Allowable - 260
Connecting - 380
Null - 259
Proper - 368
- of connected pairs 381
- of exact sequence 63, 83
- of group extension 109
Multiplicator 137 (Note)
Natural
- isomorphism 29
- transformation 29, 30, 32, 386
Negative complex 41
n-fold exact sequence 82, 83
Nilpotent ideal 286
Noether, Emmy 63 (Note)
Noether isomorphism 13, 318
Noetherian module 219
Norm 110
Normalization I14
- Theorem 236, 282
Normalized
- function 281, 283
- functor 259
- simplicia1 complex 236
Null
- morphism 259
- object 258
Object 25
Allowable projective - 261
Null - 258
Quotient
252, 363
Relatively projective - 265
Simplicia1 - 233
- Function 28, 31
Obstruction
- of abstract kernel 126
- of homomorphisms 72, 74
Opposite
- algebra 182
- category 28
- ring 157
Orbit 131
Order (of homotopy) 331
p-Group 132
Pair of spaces 61
Pairing 148 (Ex.), 247
Partly ordered set 33. 362
Pathwise connected space 57 (Ex.)
Poincarb 63 (Note)
Pointed set 362
Positive
- bicomplex 340
- complex 41
- grading 175
peadditive category 250
Product 173
Crossed
125
CUP- 245, 394ff.
External
220, 221
Index
Product
Fibred - 359
Homology - 166, 221, 296
Hopf wedge - 231
Internal - 221, 232
Kronecker - 182
Relative torsion - 274, 275
Semi-direct - 105
Simplicia1 cup - 245
Tensor - I38
Torsion - 150, 154
Wedge - 228
Projection
- of cartesian product 18
- on quotient 11
Projective
Allowable - object 261
Relative - 265, 273
- equivalence I01 (Ex.)
- module 20
Proper
- class 367
- long exact sequence 370
- morphism 368
- open set 134
- operators 134
- projective object 376
- right exact sequence 379
- sequence of complexes 397
- short exact sequence 367, 379
- subobject 368
.
Pull-back 90, 143
- Lemma 140
Pure subgroup 367
Quotient
- algebra 178
- group II
- module 11
- object 252, 363
- space 134
Range
- of a function 9
- of a homomorphism
- of a morphism 25
Reduced
- bar resolution 283
- module 304
Relative
- abelian category 263
- boundary 61
10
421
Relative
- cycle 61
- ext functor 269
- homology group 61
- projective object 265
- torsion product 274, 299
Relatively free
- - complex 267
- - module 1%
Residue field (of a local ring) 219
Resolution 87
Allowable - 261
Allowable projective - 261, 378
Bar - 115, 268, 271, 280
Free - 87
Injective - 95
KOSZU~
- 205, 218 (EX.)
Minimal 217
Projective - 87
Short projective - 382
Resolvent pair (of categories) 265
Restriction homomorphism 347
Right
- equivalent 252
- exact functor 379
- exact sequence 23, 379
- module 9, 138
- satellite 384
Ring
Change of - 90, 276
Integral group - 104
Local - 219
Opposite - I5 7
Regular - 220
Semi-simple - 203
- of quotients 219
Ringoid 250
Satellite
Left - 382
Right - 384
Selective abelian category 256
Semi-direct product I05
Sequence
Allowable exact - 261
Left exact - 23, 365
Negative ana positive - 386
Proper exact - 3 70, 375
Right exact - 23, 379
s-split - 275
Short exact - 12, 16
Split exact - 16, 260
Weakly split - 260
Index
Sets
Category of - 26
Pointed - 362
Simplicia1- 233
Sheets of a covering 134
Short Five Lemma 13, 256
Short projective resolution 382
Shuffle 243, 313
Signature (of a shuffle) 243
Signs (commutation rule) 164
Simplex 54
Simplicia1
- cup product 245
- map 235
- module 233
- object 233
- set 233
Singular
- complex (of a space) 56
- extension (of an algebra) 284
- homology 57
Space
Acyclic - 5 7
Contractible - 58
Covering - 134
Pathwise connected - 57 (Ex.)
- with operators 134
Spanned 181 (Ex.)
Spectral sequence 318ff.
Cohomology - 345
Convergent - 327, 329
First - of a bicomplex 341
First quadrant
320
Second - of a bicomplex 342
Third quadrant - 345
- - of a covering 343
- - of a filtration 327
- - of an exact couple 33 7
Split
Relatively - 263
s - 273
Weakly - 260
- extension 67, I 0 8
- sequence of complexes 47
- short exact sequence 16, 260
Squares 359
Standard affine simplex 54
Strong Four Lemma 14, 366 (Ex.)
Subalgebra 178
Subcomplex 41
Submodule 10
Graded - 175
Subobject 252, 361
Subobjekt
Proper - 368
Subquotient 13
Subring 273
Summands (direct) 18
Surjective function 25 1, 362
Suspension homomorphism 309,
315 (Ex.)
Switchback 45, 52, 98
Symmetric algebra I 8 4 (Ex.)
Syzygy, Hilvert Theorem 217
Tensor product 138 ff.
- - of algebras 295
- - of bimodules 143, I 8 7
- - of complexes I 6 3
- - of DG-algebras 190
- - of graded algebras 181
- - of graded modules I 7 6
- - of modules 138, I 8 6
3 x 3 Lemma 49, 365, 366
3 x 3 splice 227
Torsion coefficients 42
Torsion product 150, 154, 224
Relative - 274, 299
Transformation
Chain - 40
Natural - 29, 386
- of bifunctors 32
Transgression 333
Tricomplex 399
Trilinear function 142
Union of submodules 13, 18
Unitary module 9
Universal
- Coefficient Theorem 77, 1'/Off.
- covering space 135
- diagram 16, 27
- pair of functors 382, 384
- square 360
Vertices of simplex 54, 57 (Ex.)
Wang Sequence Theorem 324
Weak Four Lemma 14, 364
Weakly split sequence 260
Word (in free group) 122
Yoneda composite 82
2-graded
- algebra
- module
180
175
Zero
33