Homology, Cohomology, and Sheaf Cohomology

for Algebraic Topology, Algebraic Geometry, and

Di↵erential Geometry

Jean Gallier and Jocelyn Quaintance

Department of Computer and Information Science
University of Pennsylvania
Philadelphia, PA 19104, USA
e-mail: [email protected]

c Jean Gallier
Please, do not reproduce without permission of the authors

March 12, 2022

2
Preface

The main topics of this book are cohomology, sheaves, and sheaf cohomology. Why? Mostly
because for more than thirty years the senior author has been trying to learn algebraic
geometry. To his dismay, he realized that since 1960, under the influence and vision of A.
Grothendieck and his collaborators, in particular Serre, the foundations of algebraic geometry
were built on sheaves and cohomology. But the invasion of these theories was not restricted
to algebraic geometry. Cohomology was already a pillar of algebraic topology but sheaves
and sheaf cohomology sneaked in too.
For a novice the situation seems hopeless. Even before one begins to discuss curves or
surfaces, one has to spend years learning

(1) Some algebraic topology (especially homology and cohomology).

(2) Some basic homological algebra (chain complexes, cochain complexes, exact sequences,
chain maps, etc.). Some commutative algebra (injective and projective modules, injec-
tive and projective resolutions).

(3) Some sheaf theory.

This book represents the result of an unfinished journey in attempting to accomplish

the above. What we discovered on the way is that algebraic topology is a captivating and
beautiful subject. We also believe that it is hard to appreciate sophisticated concepts such as
sheaf cohomology without prior exposure to fundamentals of algebraic topology, simplicial
homology, singular homology, and CW complexes, in particular.
With the above motivation in mind, this book consists of two parts. The first part con-
sisting of the first seven chapters gives a crash-course on the homological and cohomological
aspects of algebraic topology, with a bias in favor of cohomology. Unfortunately homotopy
theory is omitted. Generally we do not provide proofs, with the exception of the homolog-
ical tools needed later in the second part (such as the “zig-zag lemma”). Instead we try to
provide intuitions and motivations, but we still provide rigorous definitions.
We conclude this overview of algebraic topology with a presentation of Poincaré duality,
one of the jewels of algebraic topology. We follow Milnor and Stashe↵’s exposition [45]
using the cap product, occasionally supplemented by Massey [41]. Contrary to the previous
chapters we provide almost all proofs.

3
4

Hopefully this approach will not be frustrating to the reader. Our advice is to keep a
copy of Hatcher [31] or Munkres [48] and Massey [41] at hand. Omitted details will be found
in these references. Spanier [59] may also be helpful for some of the more advanced topics.
The second part is devoted to presheaves, sheaves, Čech cohomology, derived functors,
sheaf cohomology, and spectral sequences.
Every book on algebraic geometry that goes beyond the classical material known before
1960 discusses sheaves and cohomology. The classic on the subject is Hartshorne [30]. The
joke in certain circles is that most people are so exhausted after reading Chapters II and III
that they never get to read the subsequent chapters on curves and surfaces.
It appears that after almost seventy five years it is not easy to find thorough expositions of
sheaf cohomology designed for a “general” audience, with the exception of Rotman (second
edition) [52]. Godement was already lamenting about this in the preface of his book [24]
published in 1958. He says that ironically, someone with expertise in functional analysis
(him) was compelled to give a complete exposition, that is, less incomplete than the other
existing expositions of sheaf theory.
Godement writes in French in the Bourbaki style, which means that the exposition is
terse, motivations are missing, and examples are few. This is very unfortunate because
Godement’s book contains some interesting material that is not easily found elsewhere, such
as the spectral sequence of a di↵erential sheaf and the spectral sequence of Čech cohomology.
We discuss these topics in Chapter 15.
Our own experience is that the process of learning sheaves is facilitated by proceeding in
stages. The first stage is to just define presheaves and sheaves and to give several examples.
We do this in Chapter 8.
The second stage is to define the Čech cohomology of sheaves. Čech cohomology is combi-
natorial in nature and quite concrete so one can see how sheaves provide varying coefficients.
This is the approach followed by Bott and Tu [4]. It is even possible without getting too
technical to explain why De Rham cohomology is equivalent to Čech cohomology with coef-
ficients in R by introducing the double complex known as the Čech–de–Rham complex . This
material is discussed in Chapter 9.
The third stage is to explain the sheafification process, making a presheaf into a sheaf,
and the approach to sheaves in terms of stalk spaces due to Lazard and Cartan. One would
like to define the notion of exact sequence of sheaves, but unfortunately the obvious notion of
image of a sheaf is not a sheaf in general, so the sheafification process can’t be avoided. The
right way to define the image of a sheaf is to define the notion of cokernel map of a sheaf and
to define the image as the kernel of the cokernel map. These gymnastics are inspired by the
notion of image of a map in an abelian categories, so we proceed with a basic presentation
of the notions of categories, additive categories, and abelian categories. This way we can
rightly claim that sheaves form an abelian category.
Personally, we find Serre’s explanation of the sheafification process to be one of the
clearest and we borrowed much from his famous paper FAC [55] (actually, his presentation
 5

of Čech cohomology of sheaves is also very precise and clear). The above material is presented
in Chapter 10.
Having the machinery of sheaves at our disposal, the next step is to introduce sheaf
cohomology. This can be done in two ways:

(1) In terms of resolutions by injectives.

(2) In terms of resolutions by flasque sheaves, a method invented by Godement [24].

In either case it is not possible to escape discussing the concept of resolution. We decided
that we might as well go further and present some notions of homological algebra, namely
projective and injective resolutions, as well as the notion of derived functor. Given a module
A, a resolution is an exact sequence starting with A involving projective and injective mod-
ules. Projective and injective modules are modules satisfying certain extension properties.
Given an additive functor T and an object A, it is possible to define uniquely some homology
groups Ln T (A) induced by projective resolutions of A and independent of such resolutions.
It is also possible to define uniquely some cohomology groups Rn T (A) induced by injective
resolutions of A and independent of such resolutions; see Chapter 11. As special cases we
obtain the Tor modules (associated with the tensor product) and the Ext modules (associ-
ated with the Hom functor). The modules Tor and Ext play a crucial role in the universal
coefficient theorems; see Chapter 12. Our presentation of the homological algebra given in
Chapters 11 and 12 is heavily inspired by Rotman’s excellent exposition [52]. Although Mac
Lane’s presentation is more concise it is still a very reliable and elegantly written source
which also contains historical sections [37].
Having gone that far, we also discuss Grothendieck’s notion of -functors and universal
-functors. The significance of this notion is that the machinery of universal -functors can
be used to prove that di↵erent kinds of cohomology theories yield isomorphic groups. This
technique will be used in Chapter 13 to prove that sheaf cohomology and Čech cohomology
are isomorphic for paracompact spaces.
Grothendieck’s legendary Tohoku paper [27] is written in French in a very terse style
and many proof details are omitted (there are also quite a few typos). We are not aware
of any source that gives detailed proofs of the main results about -functors (in particular,
Proposition 2.2.1 on Page 141 of [27]). Lang [35] gives a fairly complete proof but omits the
proof that the construction of the required morphism is unique. We fill in this step using an
argument communicated to us by Steve Shatz; see Chapter 11.
Having the machinery of resolutions and derived functors at our disposal we are in the
position to discuss sheaf cohomology quite thoroughly in Chapter 13. We show that the
definition of sheaf cohomology in terms of derived functors is equivalent to the definition
in terms of resolutions by flasque sheaves (due to Godement). We prove the equivalence of
sheaf cohomology and Čech cohomology for paracompact spaces. We also discuss soft and
fine sheaves, and prove that for a paracompact topological space, singular cohomology, Čech
6

cohomology, Alexander–Spanier cohomology, and sheaf cohomology (for a suitable constant

sheaf) are equivalent.
The purpose of Chapter 14 is to present various generalizations of Poincaré duality. These
versions of duality involve taking direct limits of direct mapping families of singular cohomol-
ogy groups which, in general, are not singular cohomology groups. However, such limits are
isomorphic to Alexander–Spanier cohomology groups, and thus to Čech cohomology groups.
These duality results also require relative versions of homology and cohomology.
The last chapter of our book (Chapter 15) is devoted to spectral sequences. A spectral
sequence is a tool of homological algebra whose purpose is to approximate the cohomology
(or homology) H(M ) of a module M endowed with a family (F p M )p2Z of submodules called
a filtration. The module M is also equipped with a linear map d : M ! M called di↵erential
such that d d = 0, so that it makes sense to define

H(M ) = Ker d/Im d.

We say that (M, d) is a di↵erential module. To be more precise, the filtration induces
cohomology submodules H(M )p of H(M ), the images of H(F p M ) in H(M ), and a spectral
sequence is a sequence of modules Erp (equipped with a di↵erential dpr ), for r 1, such that
Erp approximates the “graded piece” H(M )p /H(M )p+1 of H(M ).
Actually, to be useful, the machinery of spectral sequences must be generalized to filtered
cochain complexes. Technically this implies dealing with objects Erp,q involving three indices,
which makes its quite challenging to follow the exposition.
Many presentations jump immediately to the general case, but it seems pedagogically
advantageous to begin with the simpler case of a single filtered di↵erential module. This is
the approach followed by Serre in his dissertation [56] (Pages 24–104, Annals of Mathematics,
54 (1951), 425–505), Godement [24], and Cartan and Eilenberg [10].
Spectral sequences were first introduced by Jean Leray in 1945 and 1946. Paraphrazing
Jean Dieudoné [11], Leray’s definitions were cryptic and proofs were incomplete. Koszul
was the first to give a clear definition of spectral sequences in 1947. Independently, in his
dissertation (1946), Lyndon introduced spectral sequences in the context of group extensions.
Detailed expositions of spectral sequences do not seem to have appeared until 1951, in
lecture notes by Henri Cartan and in Serre’s dissertation [56], which we highly recommend for
its clarity (Serre defines homology spectral sequences, but the translation to cohomology is
immediate). A concise but very clear description of spectral sequences appears in Dieudonné
[11] (Chapter 4, Section 7, Parts D, E, F). More extensive presentations appeared in Cartan
and Eilenberg [10] and Godement [24] around 1955.
There are several methods for defining spectral sequences, including the following three:

(1) Koszul’s original approach as described by Serre [56] and Godement [24]. In our opinion
it is the simplest method to understand what is going on.
 7

(2) Cartan and Eilenberg’s approach [10]. This is a somewhat faster and slicker method
than the previous method.

(3) Exact couples of Massey (1952). This somewhat faster method for defining spectral
sequences is adopted by Rotman [50, 52] and Bott and Tu [4]. Mac Lane [37], Weibel
[63], and McCleary [44] also present it and show its equivalence with the first approach.
It appears to be favored by algebraic topologists. This approach leads to spectral
sequences in a quicker fashion and is more general because exact couples need not
arise from a filtration, but our feeling is that it is even more mysterious to a novice
than the first two approaches.

We will primarily follow Method (1) and present Method (2) and Method (3) in starred
sections (Method (2) in Section 15.15 and Method (3) in Section 15.14). All three methods
produce isomorphic sequences, and we will show their equivalence. We will also discuss
the spectral sequences induced by double complexes and give as illustrations the spectral
sequence of a di↵erential sheaf and the spectral sequence of Čech cohomology. These spectral
sequences are discussed in Godement [24].
We hope that the reader who read this book, especially the second part, will be well
prepared to tackle Hartshorne [30] or comparable books on algebraic geometry. But we will
be even happier if our readers found the topics of algebraic topology and homological algebra
presented lovable (as Rotman hopes in his preface), and even beautiful.
In the second part of our book, except for a few exceptions we provide complete proofs.
We did so to make this book self-contained, but also because we believe that no deep knowl-
edge of this material can be acquired without working out some proofs. However, our advice
is to skip some of the proofs upon first reading, especially if they are long and intricate.
The chapters or sections marked with the symbol ~ contain material that is typically
more specialized or more advanced, and they can be omitted upon first (or second) reading.
Acknowledgement: We would like to thank Ching-Li Chai, Ron Donagi, Herman Gluck,
David Harbater, Alexander Kirillov, Julius Shaneson, Jim Stashe↵, and Wolfgang Ziller for
their encouragement, advice, and what they taught us. Special thanks to Pascal Adjamagbo
and Steve Shatz for reporting typos. Steve Shatz also provided several proofs.
Contents

Contents 9

1 Introduction 13
1.1 Exact Sequences, Chain Complexes, Homology, Cohomology . . . . . . . . . 15
1.2 Relative Homology and Cohomology . . . . . . . . . . . . . . . . . . . . . . 24
1.3 Duality; Poincaré, Alexander, Lefschetz . . . . . . . . . . . . . . . . . . . . 26
1.4 Presheaves, Sheaves, and Čech Cohomology . . . . . . . . . . . . . . . . . . 32
1.5 Sheafification and Stalk Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.6 Cokernels and Images of Sheaf Maps . . . . . . . . . . . . . . . . . . . . . . 41
1.7 Injective and Projective Resolutions; Derived Functors . . . . . . . . . . . . 42
1.8 Universal -Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.9 Universal Coefficient Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.10 Sheaf Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.11 Alexander and Alexander–Lefschetz Duality . . . . . . . . . . . . . . . . . . 55
1.12 Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.13 Suggestions On How to Use This Book . . . . . . . . . . . . . . . . . . . . . 56

2 Homology and Cohomology 59

2.1 Exact Sequences and Short Exact Sequences . . . . . . . . . . . . . . . . . . 59
2.2 The Five Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3 Duality and Exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4 The Functors Hom( , A), Hom(A, ), and ⌦ A . . . . . . . . . . . . . . . 72
2.5 Abstract Cochain Complexes and Their Cohomology . . . . . . . . . . . . . 78
2.6 Chain Maps and Chain Homotopies . . . . . . . . . . . . . . . . . . . . . . 82
2.7 The Long Exact Sequence of Cohomology or Zig-Zag Lemma . . . . . . . . 84
2.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3 de Rham Cohomology 95
3.1 Review of de Rham Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2 The Mayer–Vietoris Argument . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3 Poincaré Duality on an Orientable Manifold . . . . . . . . . . . . . . . . . . 104
3.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

4 Singular Homology and Cohomology 109

8
CONTENTS 9

4.1 Singular Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.2 Homotopy Equivalence and Homology . . . . . . . . . . . . . . . . . . . . . 117
4.3 Relative Singular Homology Groups . . . . . . . . . . . . . . . . . . . . . . 120
4.4 Good Pairs and Reduced Homology . . . . . . . . . . . . . . . . . . . . . . 125
4.5 Excision and the Mayer–Vietoris Sequence . . . . . . . . . . . . . . . . . . . 128
4.6 Some Applications of Singular Homology . . . . . . . . . . . . . . . . . . . . 134
4.7 Singular Homology with G-Coefficients . . . . . . . . . . . . . . . . . . . . . 143
4.8 Singular Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.9 Relative Singular Cohomology Groups . . . . . . . . . . . . . . . . . . . . . 155
4.10 The Cup Product and the Cohomology Ring . . . . . . . . . . . . . . . . . 159
4.11 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

5 Simplicial Homology and Cohomology 165

5.1 Simplices and Simplicial Complexes . . . . . . . . . . . . . . . . . . . . . . 167
5.2 Simplicial Homology Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.3 Simplicial and Relative Homology with G-Coefficients . . . . . . . . . . . . 187
5.4 Equivalence of Simplicial and Singular Homology . . . . . . . . . . . . . . . 189
5.5 The Euler–Poincaré Characteristic of a Simplicial Complex . . . . . . . . . 195
5.6 Simplicial Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6 Homology and Cohomology of CW Complexes 205

6.1 CW Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.2 Homology of CW Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.3 The Euler–Poincaré Characteristic of a CW Complex . . . . . . . . . . . . . 224
6.4 Cohomology of CW Complexes . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

7 Poincaré Duality 237

7.1 Orientations of a Manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
7.2 The Cap Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
7.3 Cohomology with Compact Support . . . . . . . . . . . . . . . . . . . . . . 256
7.4 The Poincaré Duality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 258
7.5 The Poincaré Duality Theorem with Coefficients in G . . . . . . . . . . . . 269
7.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

8 Presheaves and Sheaves; Basics 275

8.1 Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
8.2 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
8.3 Direct Mapping Families and Direct Limits . . . . . . . . . . . . . . . . . . 288
8.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

9 Čech Cohomology with Values in a Presheaf 297

10 CONTENTS

9.1 Čech Cohomology of a Cover . . . . . . . . . . . . . . . . . . . . . . . . . . 298

9.2 Čech Cohomology with Values in a Presheaf . . . . . . . . . . . . . . . . . . 305
9.3 Equivalence of Čech Cohomology to Other Cohomologies . . . . . . . . . . . 309
9.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

10 Presheaves and Sheaves; A Deeper Look 319

10.1 Stalks and Maps of Stalks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
10.2 Sheafification of a Presheaf . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
10.3 Stalk Spaces (or Sheaf Spaces) . . . . . . . . . . . . . . . . . . . . . . . . . 333
10.4 The Equivalence of Sheaves and Stalk Spaces . . . . . . . . . . . . . . . . . 339
10.5 Stalk Spaces of Modules or Rings . . . . . . . . . . . . . . . . . . . . . . . . 342
10.6 Kernels of Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . . 344
10.7 Cokernels of Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . 347
10.8 Presheaf and Sheaf Isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . 351
10.9 Exact Sequences of Presheaves and Sheaves . . . . . . . . . . . . . . . . . . 354
10.10 Categories, Functors, Additive Categories . . . . . . . . . . . . . . . . . . . 357
10.11 Abelian Categories and Exactness . . . . . . . . . . . . . . . . . . . . . . . 367
10.12 Ringed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
10.13 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

11 Derived Functors, -Functors, and @-Functors 379

11.1 Projective, Injective, and Flat Modules . . . . . . . . . . . . . . . . . . . . . 385
11.2 Projective and Injective Resolutions . . . . . . . . . . . . . . . . . . . . . . 396
11.3 Comparison Theorems for Resolutions . . . . . . . . . . . . . . . . . . . . . 403
11.4 Left and Right Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . 413
11.5 Left-Exact and Right-Exact Derived Functors . . . . . . . . . . . . . . . . . 423
11.6 Long Exact Sequences Induced by Derived Functors . . . . . . . . . . . . . 426
11.7 T -Acyclic Resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
11.8 Universal -Functors and @-Functors . . . . . . . . . . . . . . . . . . . . . . 439
11.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454

12 Universal Coefficient Theorems 457

12.1 Universal Coefficient Theorems for Homology . . . . . . . . . . . . . . . . . 458
12.2 Computing Tor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
12.3 Universal Coefficient Theorems for Cohomology . . . . . . . . . . . . . . . . 471
12.4 Computing Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
12.5 Künneth Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
12.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493

13 Cohomology of Sheaves 497

13.1 Cohomology Groups of a Sheaf of Modules . . . . . . . . . . . . . . . . . . . 498
13.2 Flasque Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
13.3 Comparison of Čech Cohomology and Sheaf Cohomology . . . . . . . . . . . 507
CONTENTS 11

13.4 Singular Cohomology and Sheaf Cohomology . . . . . . . . . . . . . . . . . 515

13.5 Soft Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
13.6 Fine Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
13.7 de Rham Cohomology and Sheaf Cohomology . . . . . . . . . . . . . . . . . 525
13.8 Alexander–Spanier Cohomology and Sheaf Cohomology . . . . . . . . . . . 526
13.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529

14 Alexander and Alexander–Lefschetz Duality 531

14.1 Relative Alexander–Spanier Cohomology . . . . . . . . . . . . . . . . . . . . 531
14.2 Alexander–Spanier Cohomology as a Direct Limit . . . . . . . . . . . . . . . 535
14.3 Alexander–Spanier Cohomology with Compact Support . . . . . . . . . . . 540
14.4 Relative Classical Čech Cohomology . . . . . . . . . . . . . . . . . . . . . . 542
14.5 Alexander–Lefschetz Duality . . . . . . . . . . . . . . . . . . . . . . . . . . 549
14.6 Alexander Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
14.7 Alexander–Lefschetz Duality for Cohomology with Compact Support . . . . 556
14.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557

15 Spectral Sequences 559

15.1 Case 1: Filtered Di↵erential Modules . . . . . . . . . . . . . . . . . . . . . . 561
15.2 Graded Modules and Their Cohomology . . . . . . . . . . . . . . . . . . . . 566
15.3 Construction of the Spectral Sequence . . . . . . . . . . . . . . . . . . . . . 569
15.4 Case 2: Filtered Di↵erential Complexes . . . . . . . . . . . . . . . . . . . . 575
15.5 Some Graded Modules of a Filtered and Graded Complex . . . . . . . . . . 581
15.6 Construction of a Spectral Sequence; Serre–Godement . . . . . . . . . . . . 589
15.7 Convergence of Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . 597
15.8 Degenerate Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . . 607
15.9 Spectral Sequences Defined by Double Complexes . . . . . . . . . . . . . . . 612
15.10 Spectral Sequences of a Di↵erential Sheaf . . . . . . . . . . . . . . . . . . . 627
15.11 Spectral Sequences of Čech Cohomology, I . . . . . . . . . . . . . . . . . . . 632
15.12 Spectral Sequences of Čech Cohomology, II . . . . . . . . . . . . . . . . . . 642
15.13 Grothendieck’s Spectral Sequences; Composed Functors ~ . . . . . . . . . . 647
15.14 Exact Couples ~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652
15.15 Construction of a Spectral Sequence; Cartan–Eilenberg ~ . . . . . . . . . . 659
15.16 More on the Degeneration of Spectral Sequences ~ . . . . . . . . . . . . . . 669
15.17 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677

Bibliography 679

Index 683
12 CONTENTS
Chapter 1

Introduction

One of the main problems, if not “the” problem of topology, is to understand when two
spaces X and Y are similar or dissimilar. A related problem is to understand the connectivity
structure of a space in terms of its holes and “higher-order” holes. Of course, one has to
specify what “similar” means. Intuitively, two topological spaces X and Y are similar if
there is a “good” bijection f : X ! Y between them. More precisely, “good” means that
f is a continuous bijection whose inverse f 1 is also continuous; in other words, f is a
homeomorphism. The notion of homeomorphism captures the notion proposed in the mid
1860s that X can be deformed into Y without tearing or overlapping. The problem then
is to describe the equivalence classes of spaces under homeomorphism; it is a classification
problem.
The classification problem for surfaces was investigated as early as the mid 1860s by
Möbius and Jordan. These authors discovered that two (compact) surfaces are equivalent i↵
they have the same genus (the number of holes) and orientability type. Their “proof” could
not be rigorous since they did not even have a precise definition of what a 2-manifold is! We
have to wait until 1921 for a complete and rigorous proof of the classification theorem for
compact surfaces; see Gallier and Xu [22] for a historical as well as technical account of this
remarkable result.
What if X and Y do not have the nice structure of a surface or if they have higher-
order dimension? In the words of Dieudonné, the problem is a “hopeless undertaking;” see
Dieudonné’s introduction [11].
The reaction to this fundamental difficulty was the creation of algebraic and di↵erential
topology, whose major goal is to associate “invariant” objects to various types of spaces, so
that homeomorphic spaces have “isomorphic” invariants. If two spaces X and Y happen to
have some distinct invariant objects, then for sure they are not homeomorphic.
Poincaré was one of the major pioneers of this approach. At first these invariant objects
were integers (Betti numbers and torsion numbers), but it was soon realized that much more
information could be extracted from invariant algebraic structures such as groups, ring, and
modules.

13
14 CHAPTER 1. INTRODUCTION

Three types of invariants can be assigned to a topological space:

(1) Homotopy groups.

(2) Homology groups.

(3) Cohomology groups.

The above are listed in the chronological order of their discovery. It is interesting that
the first homotopy group ⇡1 (X) of the space X, also called fundamental group, was invented
by Poincaré (Analysis Situs, 1895), but homotopy basically did not evolve until the 1930s.
One of the reasons is that the first homotopy group is generally nonabelian, so harder to
study.
On the other hand, homology and cohomology groups (or rings, or modules) are abelian,
so results about commutative algebraic structures can be leveraged. This is true in particular
if the ring R is a PID, where the structure of the finitely generated R-modules is completely
determined.
There are di↵erent kinds of homology groups. They usually correspond to some geometric
intuition about decomposing a space into simple shapes such as triangles, tetrahedra, etc.
Cohomology is more abstract because it usually deals with functions on a space. However,
we will see that it yields more information than homology precisely because certain kinds of
operations on functions can be defined (cup and cap products).
As often in mathematics, some machinery that is created to solve a specific problem, here
a problem in topology, unexpectedly finds fruitful applications to other parts of mathematics
and becomes a major component of the arsenal of mathematical tools, in the present case
homological algebra and category theory. In fact, category theory, invented by Mac Lane and
Eilenberg, permeates algebraic topology and is really put to good use, rather than being a
fancy attire that dresses up and obscures some simple theory, as often is the case.
In view of the above discussion, it appears that algebraic topology might involve more
algebra than topology. This is great if one is quite proficient in algebra, but not so good
news for a novice who might be discouraged by the abstract and arcane nature of homological
algebra. After all, what do the zig-zag lemma and the five lemma have to do with topology?
Unfortunately, it is true that a firm grasp of the basic concepts and results of homological
algebra is essential to really understand what are the homology and the cohomology groups
and what are their roles in topology.
One of our goals is to attempt to demystify homological algebra. For those of us fond of
puns, keep this simple analogy in mind and all trepidation will (hopefully) fade. Homology
groups describe what man does in his home; in French, l’homme au logis. Cohomology
groups describe what co-man does in his home; in French, le co-homme au logis, that is, la
femme au logis. Obviously this is not politically correct, so cohomology should be renamed.
The big question is: what is a better name for cohomology?
1.1. EXACT SEQUENCES, CHAIN COMPLEXES, HOMOLOGY, COHOMOLOGY 15

In the following sections we give a brief description of the topics that we are going to
discuss in this book, and we try to provide motivations for the introduction of the concepts
and tools involved. These sections introduce topics in the same order in which they are
presented in the book. All historical references are taken from Dieudonné [11]. This is a
remarkable account of the history of algebraic and di↵erential topology from 1900 to the
1960s which contains a wealth of information.

1.1 Exact Sequences, Chain Complexes, Homology

and Cohomology
There are various kinds of homology groups (simplicial, singular, cellular, etc.), but they all
arise the same way, namely from a (possibly infinite) sequence called a chain complex

d0 d1 dp 1 dp dp+1
0o C0 o C1 o ··· o Cp 1
o Cp o Cp+1 o ··· ,

in which the Cp are vector spaces, or more generally abelian groups (typically freely gen-
erated), and the maps dp : Cp ! Cp 1 are linear maps (homomorphisms of abelian groups)
satisfying the condition
dp dp+1 = 0 for all p 0. (⇤1 )
The elements of Cp are called p-chains and the maps dp are called boundary operators (or
boundary maps). The intuition behind Condition (⇤1 ) is that elements of the form dp (c) 2
Cp 1 with c 2 Cp are boundaries, and “a boundary has no boundary.” For example, in R2 ,
the points on the boundary of a closed unit disk form the unit circle, and the points on the
unit circle have no boundary.
Since dp dp+1 = 0, we have Bp (C) = Im dp+1 ✓ Ker dp = Zp (C) so the quotient
Zp (C)/Bp (C) = Ker dp /Im dp+1 makes sense. The quotient module

Hp (C) = Zp (C)/Bp (C) = Ker dp /Im dp+1

is the p-th homology module of the chain complex C. Elements of Zp are called p-cycles and
elements of Bp are called p-boundaries; see Figure 1.1.
A condition stronger that Condition (⇤1 ) is that

Im dp+1 = Ker dp for all p 0. (⇤⇤1 )

A sequence satisfying Condition (⇤⇤1 ) is called an exact sequence. Thus, we can view the
homology groups as a measure of the failure of a chain complex to be exact. Surprisingly,
exact sequences show up in various areas of mathematics, especially abstract algebra.
In the case of many homology theories, chain complexes are constructed by “nicely”
mapping simple geometric objects into a given topological space X. For singular homology
16 CHAPTER 1. INTRODUCTION

Figure 1.1: Let X be the surface of the torus. Elements of Z1 are geometrically represented
by curves which are homeomorphic to S 1 . Thus both the red and blue curves are 1-cycles.
The red curve is also a 1-boundary since it is the boundary of a region in X which is
homeomorphic to the closed unit disk.

the Cp ’s arePthe abelian groups Cp = Sp (X; Z) consisting of all (finite) linear combinations
of the form ni i , where ni 2 Z and each i , a singular p-simplex , is a continuous function
p
i: ! X from the p-simplex p to the space X. A 0-simplex is a single point, a 1-simplex
is a line segment, a 2-simplex is a triangle, a 3-simplex is a tetrahedron, and a p-simplex is
a higher-order generalization of a tetrahedron; see Figure 1.2.
A p-simplex p has p + 1 faces, and the ith face is a (p 1)-singular simplex
p 1
i : p 1 ! X defined in terms of a certain function pi 1 : p 1 ! p ; see Section 4.1.
In the framework of singular homology, the boundary map dp is denoted by @p , and for any
singular p-simplex , @ is the singular (p 1)-chain given by
p 1 p 1
@ = 0 1 + · · · + ( 1)p p 1
p .

A simple calculation confirms that @p @p+1 = 0. Consequently the free abelian groups
Sp (X; Z) together with the boundary maps @p form a chain complex denoted S⇤ (X; Z) called
the simplicial chain complex of X. Then the quotient module

Hp (X; Z) = Hp (S⇤ (X; Z)) = Ker @p /Im @p+1 ,

also denoted Hp (X), is called the p-th homology group of X. Singular homology is discussed
in Chapter 4, especially in Section 4.1.
Historically, singular homology did not come first. According to Dieudonné [11], sin-
gular homology emerged around 1925 in the work of Veblen, Alexander and Lefschetz (the
“Princeton topologists,” as Dieudonné calls them), and was defined rigorously and in com-
plete generality by Eilenberg (1944). The definition of the homology modules Hp (C) in terms
1.1. EXACT SEQUENCES, CHAIN COMPLEXES, HOMOLOGY, COHOMOLOGY 17

0-simplex 1-simplex
2-simplex

3-simplex

Figure 1.2: The top row illustrates lower order p-simplicies while the bottom figure illustrates
a singular 2-simplex within the 2-dimensional torus.

of sequences of abelian groups Cp and boundary homomorphisms dp : Cp ! Cp 1 satisfying

the condition dp dp+1 = 0 as quotients Ker dp /Im dp+1 seems to have been suggested to H.
Hopf by Emmy Noether while Hopf was visiting Göttingen in 1925. Hopf used this definition
in 1928, and independently so did Vietoris in 1926, and then Mayer in 1929.
The first occurrence of a chain complex is found in Poincaré’s papers of 1900, although
he did not use the formalism of modules and homomorphisms as we do now, but matrices
instead. Poincaré introduced the homology of simplicial complexes, which are combinatorial
triangulated objects objects made up of simplices; see Figure 1.3.
Given a simplicial complex K, we have free abelian groups Cp (K) consisting of Z-linear
combinations of oriented p-simplices, and the boundary maps @p : Cp (K) ! Cp 1 (K) are
defined by
p
X
@p = ( 1)i [↵0 , . . . , ↵bi , . . . , ↵p ],
i=0

for any oriented p-simplex, = [↵0 , . . . , ↵p ], where [↵0 , . . . , ↵bi , . . . , ↵p ] denotes the oriented
(p 1)-simplex obtained by deleting vertex ↵i . Then we have a simplicial chain complex
(Cp (K), @p ) denoted C⇤ (K), and the corresponding homology groups Hp (C⇤ (K)) are denoted
Hp (K) and called the simplicial homology groups of K. Simplicial homology is discussed in
Chapter 5. We discussed singular homology first because it subsumes simplicial homology,
as shown in Section 5.2.
A simplicial complex K is a purely combinatorial object, thus it is not a space, but it has
a geometric realization Kg , which is a (triangulated) topological space. This brings up the
18 CHAPTER 1. INTRODUCTION

Figure 1.3: The surface of a cube as a simplicial complex consisting of 12 triangles (2-
simplicies), 18 edges (1-simplicies), and 8 vertices (0-simplices).

following question: if K1 and K2 are two simplicial complexes whose geometric realizations
(K1 )g and (K2 )g are homeomorphic, are the simplicial homology groups Hp (K1 ) and Hp (K2 )
isomorphic?
Poincaré conjectured that the answer was “yes,” and this conjecture was first proved by
Alexander. The proof is nontrivial, and we present a version of it in Section 5.2.
The above considerations suggest that it would be useful to understand the relationship
between the homology groups of two spaces X and Y related by a continuous map f : X ! Y .
For this, we define mappings between chain complexes called chain maps.
Given two chain complexes C and C 0 , a chain map f : C ! C 0 is a family f = (fp )p 0 of
homomorphisms fp : Cp ! Cp0 such that all the squares of the following diagram commute:

d0 d1 dp 1 dp dp+1
0o C0 o C1 o ··· o Cp 1
o Cp o Cp+1 o ···
f0 f1 fp 1 fp fp+1
✏ ✏ ✏ ✏ ✏
0o C00 o C10 o ··· o Cp0 1
o Cp0 o 0
Cp+1 o ··· ,
d00 d01 d0p 1 d0p d0p+1

that is, fp dp+1 = d0p+1 fp+1 , for all p 0.

A chain map f : C ! C 0 induces homomorphisms of homology
Hp (f ) : Hp (C) ! Hp (C 0 )
for all p 0. Furthermore, given three chain complexes C, C 0 , C 00 and two chain maps
f : C ! C and g : C 0 ! C 00 , we have
0

Hp (g f ) = Hp (g) Hp (f ) for all p 0

1.1. EXACT SEQUENCES, CHAIN COMPLEXES, HOMOLOGY, COHOMOLOGY 19

and
Hp (idC ) = idHp (C) for all p 0.
We say that the map C 7! (Hp (C))p 0 is functorial (to be more precise, it is a functor
from the category of chain complexes and chain maps to the category of abelian groups and
groups homomorphisms).
For example, in singular homology, a continuous function f : X ! Y between two topo-
logical spaces X and Y induces a chain map f] : S⇤ (X; Z) ! S⇤ (Y ; Z) between the two
simplicial chain complexes S⇤ (X; Z) and S⇤ (Y ; Z) associated with X and Y , which in turn
yield homology homomorphisms usually denoted f⇤,p : Hp (X; Z) ! Hp (Y ; Z). Thus the map
X 7! (Hp (X; Z))p 0 is a functor from the category of topological spaces and continuous
maps to the category of abelian groups and groups homomorphisms. Functoriality implies
that if f : X ! Y is a homeomorphism, then the maps f⇤,p : Hp (X; Z) ! Hp (Y ; Z) are iso-
morphisms. Thus, the singular homology groups are topological invariants. This is one of
the advantages of singular homology; topological invariance is basically obvious.
This is not the case for simplicial homology where it takes a fair amount of work to prove
that if K1 and K2 are two simplicial complexes whose geometric realizations (K1 )g and (K2 )g
are homeomorphic, then the simplicial homology groups Hp (K1 ) and Hp (K2 ) isomorphic.
One might wonder what happens if we reverse the arrows in a chain complex? Abstractly,
this is how cohomology is obtained, although this point of view was not considered until at
least 1935.
A cochain complex is a sequence

d 1 d0 d1 p 1 dp dp+1
0 / C0 / C1 / ··· / Cp 1 d / Cp / C p+1 / C p+2 / ··· ,

in which the C p are abelian groups, and the maps dp : C p ! C p+1 are homomorphisms of
abelian groups satisfying the condition

dp+1 dp = 0 for all p 0 (⇤2 )

The elements of C p are called cochains and the maps dp are called coboundary maps. This
time it is not clear how coboundary maps arise naturally. Since dp+1 dp = 0, we have
B p = Im dp ✓ Ker dp+1 = Z p+1 , so the quotient Z p /B p = Ker dp+1 /Im dp makes sense and
the quotient module
H p (C) = Z p /B p = Ker dp+1 /Im dp
is the pth cohomology module of the cochain complex C. Elements of Z p are called p-cocycles
and elements of B p are called p-coboundaries.
There seems to be an unwritten convention that when dealing with homology we use
subscripts, and when dealing with cohomology we use with superscripts. Also, the “dual” of
any “notion” is the “co-notion.”
20 CHAPTER 1. INTRODUCTION

As in the case of a chain complex, a condition stronger that Condition (⇤2 ) is that

Im dp = Ker dp+1 for all p 0. (⇤⇤2 )

A sequence satisfying Condition (⇤⇤2 ) is also called an exact sequence. Thus, we can view
the cohomology groups as a measure of the failure of a cochain complex to be exact.
Given two cochain complexes C and C 0 , a (co)chain map f : C ! C 0 is a family f =
(f p )p 0 of homomorphisms f p : C p ! C 0 p such that all the squares of the following diagram
commute:
d 1 d0 d1 p 1 dp dp+1
1 d
0 / C0 / C1 / ··· / Cp / Cp / C p+1 / ···
f0 f1 fp 1 fp f p+1
✏ ✏ ✏ ✏ ✏
0 / C 00 / C 01 / ··· / C 0p 1 / C 0p / C 0 p+1 / ··· ,
d0 1
d0 0 d0 1 d0 p 1 d0 p d0 p+1

that is, f p+1 dp = d0 p f p for all p 0. A chain map f : C ! C 0 induces homomorphisms

of cohomology
H p (f ) : H p (C) ! H p (C 0 )
for all p 0. Furthermore, this assignment is functorial (more precisely, it is a functor from
the category of cochain complexes and chain maps to the category of abelian groups and
their homomorphisms).
At first glance cohomology appears to be very abstract so it is natural to look for explicit
examples. A way to obtain a cochain complex is to apply the operator (functor) HomZ ( , G)
to a chain complex C, where G is any abelian group. Given a fixed abelian group A, for any
abelian group B we denote by HomZ (B, A) the abelian group of all homomorphisms from
B to A. Given any two abelian groups B and C, for any homomorphism f : B ! C, the
homomorphism HomZ (f, A) : HomZ (C, A) ! HomZ (B, A) is defined by

HomZ (f, A)(') = ' f for all ' 2 HomZ (C, A);

see the commutative diagram below:

f
B / C
'
HomZ (f,A)(') ✏
A.

The map HomZ (f, A) is also denoted by HomZ (f, idA ) or even HomZ (f, id). Observe that
the e↵ect of HomZ (f, id) on ' is to precompose ' with f .
If f : B ! C and g : C ! D are homomorphisms of abelian groups, a simple computation
shows that
HomR (g f, id) = HomR (f, id) HomR (g, id).
1.1. EXACT SEQUENCES, CHAIN COMPLEXES, HOMOLOGY, COHOMOLOGY 21

Observe that HomZ (f, id) and HomZ (g, id) are composed in the reverse order of the compo-
sition of f and g. It is also immediately verified that

HomZ (idA , id) = idHomZ (A,G) .

We say that HomZ ( , id) is a contravariant functor (from the category of abelian groups
and group homomorphisms to itself). Then given a chain complex

d0 d1 dp 1 dp dp+1
0o C0 o C1 o ··· o Cp 1
o Cp o Cp+1 o ··· ,

we can form the cochain complex

HomZ (d0 ,id) HomZ (dp+1 ,id)
0 / HomZ (C0 , G) / ··· / HomZ (Cp , G) / HomZ (Cp+1 , G) / ···

obtained by applying HomZ ( , G), and denoted HomZ (C, G). The coboundary map dp is
given by
dp = HomZ (dp+1 , id),
which means that for any f 2 HomZ (Cp , G), we have

dp (f ) = f dp+1 .

Thus, for any (p + 1)-chain c 2 Cp+1 we have

(dp (f ))(c) = f (dp+1 (c)).

We obtain the cohomology groups H p (HomZ (C, G)) associated with the cochain complex
HomZ (C, G). The cohomology groups H p (HomZ (C, G)) are also denoted H p (C; G).
This process was applied to the simplicial chain complex C⇤ (K) associated with a sim-
plicial complex K by Alexander and Kolmogoro↵ to obtain the simplicial cochain com-
plex HomZ (C⇤ (K); G) denoted C ⇤ (K; G) and the simplicial cohomology groups H p (K; G)
of the simplicial complex K; see Section 5.6. Soon after, this process was applied to
the singular chain complex S⇤ (X; Z) of a space X to obtain the singular cochain complex
HomZ (S⇤ (X; Z); G) denoted S ⇤ (X; G) and the singular cohomology groups H p (X; G) of the
space X; see Section 4.8.
Given a continuous map f : X ! Y , there is an induced chain map f ] : S ⇤ (Y ; G) !
S (X; G) between the singular cochain complexes S ⇤ (Y ; G) and S ⇤ (X; G), and thus homo-
⇤

morphisms of cohomology f ⇤ : H p (Y ; G) ! H p (X; G). Observe the reversal: f is a map from

X to Y , but f ⇤ maps H p (Y ; G) to H p (X; G). We say that the map X 7! (H p (X; G))p 0 is
a contravariant functor from the category of topological spaces and continuous maps to the
category of abelian groups and their homomorphisms.
So far our homology groups have coefficients in Z, but the process of forming a cochain
complex HomZ (C, G) from a chain complex C allows the use of coefficients in any abelian
22 CHAPTER 1. INTRODUCTION

group G, not just the integers. Actually, it is a trivial step to define chain complexes con-
sisting of R-modules in any commutative ring R with a multiplicative identity element 1,
and such complexes yield homology modules Hp (C; R) with coefficients in R. This process
immediately applies to the singular homology groups Hp (X; R) and to the simplicial ho-
mology groups Hp (K; R). Also, given a chain complex C where the Cp are R-modules, for
any R-module G we can form the cochain complex HomR (C, G) and we obtain cohomology
modules H p (C; G) with coefficients in any R-module G; see Section 4.8 and Chapter 12.
We can generalize homology with coefficients in a ring R to modules with coefficients in
a R-module G by applying the operation (functor) ⌦R G to a chain complex C, where
the Cp ’s are R-modules, to get the chain complex denoted C ⌦R G. The homology groups
of this complex are denoted Hp (C, G). We will discuss this construction in Section 4.7 and
Chapter 12.
If the ring R is a PID, given a chain complex C where the Cp are R-modules, the homology
groups Hp (C; G) of the complex C ⌦R G are determined by the homology groups Hp 1 (C; R)
and Hp (C; R) via a formula called the Universal Coefficient Theorem for Homology; see
Theorem 12.1. This formula involves a term TorR 1 (Hn 1 (C); G) that corresponds to the fact
that the operation ⌦R G on linear maps generally does not preserve injectivity ( ⌦R G
is not left-exact). These matters are discussed in Chapter 11.
Similarly, if the ring R is a PID, given a chain complex C where the Cp are R-modules, the
cohomology groups H p (C; G) of the complex HomR (C, G) are determined by the homology
groups Hp 1 (C; R) and Hp (C; R) via a formula called the Universal Coefficient Theorem
for Cohomology; see Theorem 12.6. This formula involves a term Ext1R (Hn 1 (C); G) that
corresponds to the fact that if the linear map f is injective, then HomR (f, id) is not necessarily
surjective (HomR ( , G) is not right-exact). These matters are discussed in Chapter 11.
One of the advantages of singular homology (and cohomology) is that it is defined for all
topological spaces, but one of its disadvantages is that in practice it is very hard to compute.
On the other hand, simplicial homology (and cohomology) only applies to triangulable spaces
(geometric realizations of simplicial complexes), but in principle it is computable (for finite
complexes). One of the practical problems is that the triangulations involved may have a
large number of simplices. J.H.C. Whiteahead invented a class of spaces called CW complexes
that are more general than triangulable spaces and for which the computation of the singular
homology groups is often more tractable. Unlike a simplicial complex, a CW complex is
obtained by gluing spherical cells as shown in Figure 1.4. CW complexes are discussed in
Chapter 6.
There are at least four other ways of defining cohomology groups of a space X by di-
rectly forming a cochain complex without first forming a homology chain complex and then
dualizing by applying HomZ ( , G):

(1) If X is a smooth manifold, then there is the de Rham complex which uses the modules
of smooth p-forms Ap (X) and the exterior derivatives dp : Ap (X) ! Ap+1 (X). The
1.1. EXACT SEQUENCES, CHAIN COMPLEXES, HOMOLOGY, COHOMOLOGY 23

0-cell 1-cell

2-cell

“glued” 1-cell to 0-cell

attached 2-cell

Figure 1.4: The spherical hemisphere is a CW complex consisting of a point (0-cell), a line
segment (1-cell), and a closed unit disk (2-cell).

p
corresponding cohomology groups are the de Rham cohomology groups HdR (X). These
are actually real vector spaces. de Rham cohomology is discussed in Chapter 3.

(2) If X is any space and U = (Ui )i2I is any open cover of X, we can define the Čech
cohomology groups Ȟ p (X, U ) in a purely combinatorial fashion. Then we can define
the notion of refinement of a cover and define the Čech cohomology groups Ȟ p (X, G)
with values in an abelian group G using a limiting process known as a direct limit (see
Section 8.3, Definition 8.10). Čech cohomology is discussed in Chapter 9.

(3) If X is any space, then there is the Alexander–Spanier cochain complex which yields the
Alexander–Spanier cohomology groups ApA-S (X; G). Alexander–Spanier cohomology is
discussed in Section 13.8 and in Chapter 14.

(4) Sheaf cohomology, based on derived functors and injective resolutions. This is the
most general kind of cohomology of a space X, where cohomology groups H p (X, F)
with values in a sheaf F on the space X are defined. Intuitively, this means that the
modules F(U ) of “coefficients” in which these groups take values may vary with the
open domain U ✓ X. Sheaf cohomology is discussed in Chapter 13, and the algebraic
machinery of derived functors is discussed in Chapter 11.

We will see that for topological manifolds, all these cohomology theories are equivalent;
see Chapter 13. For paracompact spaces, Čech cohomology, Alexander–Spanier cohomology,
and derived functor cohomology (for constant sheaves) are equivalent (see Chapter 13). In
fact, Čech cohomology and Alexander–Spanier cohomology are equivalent for any space; see
Chapter 14.
24 CHAPTER 1. INTRODUCTION

1.2 Relative Homology and Cohomology

In general, computing homology groups is quite difficult so it would be helpful if we had
techniques that made this process easier. Relative homology and excision are two such tools
that we discuss in this section.
Lefschetz (1928) introduced the relative homology groups Hp (K, L; Z), where K is a
simplicial complex and L is a subcomplex of K. The same idea immediately applies to
singular homology and we can define the relative singular homology groups Hp (X, A; R)
where A is a subspace of X. The intuition is that the module of p-chains of a relative chain
complex consists of chains of K modulo chains of L. For example, given a space X and
a subspace A ✓ X, the singular chain complex S⇤ (X, A; R) of the pair (X, A) is the chain
complex in which each R-module Sp (X, A; R) is the quotient module

Sp (X, A; R) = Sp (X; R)/Sp (A; R).

It is easy to see that Sp (X, A; R) is actually a free R-module; see Section 4.3.
Although this is not immediately apparent, the motivation is that the groups Hp (A; R)
and Hp (X, A; R) are often “simpler” than the groups Hp (X; R), and there is an exact se-
quence called the long exact sequence of relative homology that can often be used to come
up with an inductive argument that allows the determination of Hp (X; R) from Hp (A; R)
and Hp (X, A; R). Indeed, we have the following exact sequence as shown in Section 4.3 (see
Theorem 4.9):

··· / Hp+2 (X, A; R)

@⇤p+2

i⇤ j⇤
/ Hp+1 (A; R) / Hp+1 (X; R) / Hp+1 (X, A; R)
@⇤p+1

i⇤ j⇤
/ Hp (A; R) / Hp (X; R) / Hp (X, A; R)
@⇤p

/ Hp 1 (A; R) / ···
ending in
H0 (A; R) / H0 (X; R) / H0 (X, A; R) / 0.
Furthermore, if (X, A) is a “good pair,” then there is an isomorphism

Hp (X, A; R) ⇠
= Hp (X/A, {pt}; R),

where the space X/A, called a quotient space, is obtained from X by identifying A with a
single point, and where pt stands for any point in X.
1.2. RELATIVE HOMOLOGY AND COHOMOLOGY 25

The long exact sequence of relative homology is a corollary of one the staples of homology
theory, the “zig-zag lemma.” The zig-zag lemma says that for any short exact sequence

f g
0 !X !Y !Z !0

of chain complexes X, Y, Z there is a long exact sequence of cohomology

··· / H p 1 (Z)
p 1

f⇤ g⇤
/ H p (X) / H p (Y ) / H p (Z)
p

f⇤ g⇤
/ H p+1 (X) / H p+1 (Y ) / H p+1 (Z)
p+1

/ H p+2 (X) / ···

The zig-zag lemma is fully proven in Section 2.7; see Theorem 2.22. There is also a homology
version of this theorem.
Another very important aspect of relative singular homology is that it satisfies the ex-
cision axiom, another useful tool to compute homology groups. This means that removing
a subspace Z ✓ A ✓ X which is clearly inside of A, in the sense that Z is contained in
the interior of A, does not change the relative homology group Hp (X, A; R). More precisely,
there is an isomorphism

Hp (X Z, A Z; R) ⇠
= Hp (X, A; R);

see Section 4.5 (Theorem 4.14). A good illustration of the use of excision and of the long
exact sequence of relative homology is the computation of the homology of the sphere S n ;
see Section 4.6. Relative singular homology also satisfies another important property: the
homotopy axiom, which says that if two spaces are homotopy equivalent, then their homology
is isomorphic; see Theorem 4.8.
Following the procedure for obtaining cohomology from homology described in Section
1.1, by applying HomR ( , G) to the chain complex S⇤ (X, A; R) we obtain the cochain com-
plex S ⇤ (X, A; G) = HomR (S⇤ (X, A; R), G), and thus the singular relative cohomology groups
H p (X, A; G); see Section 4.9. In this case we can think of the elements of S p (X, A; G) as lin-
ear maps (with values in G) on singular p-simplices in X that vanish on singular p-simplices
in A.
Fortunately, since each Sp (X, A; R) is a free R-module, it can be shown that there is a
long exact sequence of relative cohomology (see Theorem 4.36):
26 CHAPTER 1. INTRODUCTION

··· / H p 1 (A; G)
⇤
p 1

(j > )⇤ (i> )⇤
/ H p (X, A; G) / H p (X; G) / H p (A; G)
⇤
p

(j > )⇤ (i> )⇤
/ H p+1 (X, A; G) / H p+1 (X; G) / H p+1 (A; G)
⇤
p+1

/ H p+2 (X, A; G) / ···

Relative singular cohomology also satisfies the excision axiom and the homotopy axioms (see
Section 4.9).

1.3 Duality; Poincaré, Alexander, Lefschetz

Roughly speaking, duality is a kind of symmetry between the homology and the cohomology
groups of a space. Historically, duality was formulated only for homology, but it was later
found that more general formulations are obtained if both homology and cohomology are
considered. We will discuss two duality theorems: Poincaré duality, and Alexander–Lefschetz
duality. Original versions of these theorems were stated for homology and applied to special
kinds of spaces. It took at least thirty years to obtain the versions that we will discuss.
The result that Poincaré considered as the climax of his work in algebraic topology
was a duality theorem (even though the notion of duality was not very clear at the time).
Since Poincaré was working with finite simplicial complexes, for him duality was a con-
struction which, given a simplicial complex K of dimension n, produced a “dual” complex
K ⇤ ; see Munkres [48] (Chapter 8, Section 64). If done the right way, the matrices of the
boundary maps @ : Cp (K) ! Cp 1 (K) are transposes of the matrices of the boundary maps
@ ⇤ : Cn p+1 (K ⇤ ) ! Cn p (K ⇤ ). As a consequence, the homology groups Hp (K) and Hn p (K ⇤ )
are isomorphic. Note that this type of duality relates homology groups, not homology and
cohomology groups as it usually does nowadays, for the good reason that cohomology did
not exist until about 1935.
Around 1930 de Rham gave a version of Poincaré duality for smooth orientable, compact
manifolds. If M is a smooth, oriented, and compact n-manifold, then there are isomorphisms
p
HdR (M ) ⇠ n p
= (HdR (M ))⇤ ,

where (H n p (M ))⇤ is the dual of the vector space H n p (M ). This duality is actually induced
by a nondegenerate pairing
p n p
h , i : HdR (M ) ⇥ HdR (M ) ! R
1.3. DUALITY; POINCARÉ, ALEXANDER, LEFSCHETZ 27

given by integration, namely Z

h[!], [⌘]i = ! ^ ⌘,
M
where ! is a di↵erential p-form and ⌘ is a di↵erential (n p)-form. For details, see Chapter
3, Theorem 3.8. The proof uses several tools from the arsenal of homological algebra: the
zig-zag lemma (in the form of Mayer–Vietoris sequences), the five lemma, and an induction
on finite “good covers.”
Around 1935, inspired by Pontrjagin’s duality theorem and his introduction of the no-
tion of nondegenerate pairing (see the end of this section), Alexander and Kolmogoro↵
independently started developing cohomology, and soon after this it was realized that be-
cause cohomology primarily deals with functions, it is possible to define various products.
Among those, the cup product is particularly important because it induces a multiplication
operation on what is called the cohomology algebra H ⇤ (X; R) of a space X, and the cap
product yields a stronger version of Poincaré duality.
L
Recall that S ⇤ (X; R) is the R-module p 0 S p (X; R), where the S p (X; R) are the sin-
gular cochain modules. For all p, q 0, it possible to define a function
^ : S p (X; R) ⇥ S q (X; R) ! S p+q (X; R),
called cup product. These functions induce a multiplication on S ⇤ (X; R) also called the
cup product, which is bilinear, associative, and has an identity element. The cup product
satisfies the following equation
(c ^ d) = ( c) ^ d + ( 1)p c ^ ( d),
reminiscent of a property of the wedge product. (In the above equation is the coboundary
map, i.e. p : S p (X; R) ! S p+1 (X; R).) This equation can be used to show that the cup
product is a well defined on cohomology classes:
^ : H p (X; R) ⇥ H q (X; R) ! H p+q (X; R).
L
These operations induce a multiplication operation on H ⇤ (X; R) = p 0 H p (X; R) which is
bilinear and associative. Together with the cup product, H ⇤ (X; R) is called the cohomology
ring of X. For details, see Section 4.10.
The cup product for simplicial cohomology was invented independently by Alexander
and Kolmogoro↵ (in addition to simplicial cohomology) and presented at a conference held
in Moscow in 1935. Alexander’s original definition was not quite correct and he modified
his definition following a suggestion of Čech (1936). This modified version was discovered
independently by Whitney (1938), who introduced the notation ^. Eilenberg extended the
definition of the cup product to singular cohomology (1944).
The significance of the cohomology ring is that two spaces X and Y may have isomorphic
cohomology modules but nonisomorphic cohomology rings. Therefore the cohomology ring
is an invariant of a space X that is finer than its cohomology.
28 CHAPTER 1. INTRODUCTION

Another product related to the cup product is the cap product. The cap product combines
cohomology and homology classes, it is an operation

_ : H p (X; R) ⇥ Hn (X; R) ! Hn p (X; R);

see Section 7.2.

The cap product was introduced by Čech (1936) and independently by Whitney (1938),
who introduced the notation _ and the name cap product. Again Eilenberg generalized the
cap product to singular homology and cohomology.
The cup product and the cap product are related by the following equation:

a(b _ ) = (a ^ b)( )

for all a 2 S n p (X; R), all b 2 S p (X; R), and all 2 Sn (X; R), or equivalently using the
bracket notation for evaluation as

ha, b _ i = ha ^ b, i,

which shows that _ is the adjoint of ^ with respect to the evaluation pairing h , i.
The reason why the cap product is important is that it can be used to state a sharper
version of Poincaré duality. But first we need to talk about orientability.
If M is a topological manifold of dimension n, it turns out that for every x 2 M the
relative (singular) homology groups Hp (M, M {x}; Z) are either (0) if p 6= n, or equal to
Z if p = n. An orientation of M is a choice of a generator µx 2 Hn (M, M {x}; Z) ⇠ =Z
for each x 2 M which varies “‘continuously” with x. A manifold that has an orientation is
called orientable.
Technically, this means that for every x 2 M , locally on some small open subset U of M
containing x there is some homology class µU 2 Hn (M, M U ; Z) such that all the chosen
µx 2 Hn (M, M {x}; Z) for all x 2 U are obtained as images of µU ; see Figure 1.5.
If such a µU can be found when U = M , we call it a fundamental class of M and denote
it by µM ; see Section 7.3. Readers familiar with di↵erential geometry may think of the
fundamental class as a discrete analog to the notion of volume form. The crucial result is
that a compact manifold of dimension n is orientable i↵ it has a unique fundamental class
µM ; see Theorem 7.7.
The notion of orientability can be generalized to the notion of R-orientability. One of
the advantages of this notion is that every manifold is Z/2Z orientable. We are now in a
position to state the Poincaré duality theorem in terms of the cap product.
If M is compact and orientable, then there is a fundamental class µM . In this case (if
0  p  n) we have a map

DM : H p (M ; Z) ! Hn p (M ; Z)
1.3. DUALITY; POINCARÉ, ALEXANDER, LEFSCHETZ 29

μU

μx μx
μx
μx
U

Figure 1.5: A schematic representation which shows µx as the image of µU .

given by
DM (!) = ! _ µM .
Poincaré duality asserts that the map

DM : ! 7! ! _ µM

is an isomorphism between H p (M ; Z) and Hn p (M ; Z); see Theorem 7.16.

Poincaré duality can be generalized to compact R-orientable manifolds for any commuta-
tive ring R, and to coefficients in any R-module G. It can also be generalized to noncompact
manifolds if we replace cohomology by cohomology with compact support (the modules
Hcp (X; R)); see Sections 7.3, 7.4, and 7.5. If R = Z/2Z, Poincaré duality holds for all
manifolds, orientable or not.
Another kind of duality was introduced by Alexander in 1922. Alexander considered a
compact proper subset A of the sphere S n (n 2) which is a curvilinear cell complex (A has
some type of generalized triangulation). For the first time he defined the homology groups
of the open subset S n A with coefficients in Z/2Z (so that he did not have to bother with
signs), and he proved that for p  n 2 there are isomorphisms

Hp (A; Z/2Z) ⇠
= Hn p 1 (S
n
A; Z/2Z);

see Figure 1.6. Since cohomology did not exist yet, the original version of Alexander duality
was stated for homology.
Around 1928 Lefschetz started investigating homology with coefficients in Z, Z/mZ, or
Q, and defined relative homology. In his book published in 1930, using completely di↵erent
methods from Alexander, Lefschetz proved a version of Alexander’s duality in the case where
30 CHAPTER 1. INTRODUCTION

2
S

H p (A; Z/2Z) y H 2-p-1 (S 2 - A; Z/2Z)

Figure 1.6: Let A be the peach spherical triangle in S 2 . The original version of Alexander
duality compares the homology of the peach spherical triangle with the homology of the
surface consisting of S 2 A.

A is a subcomplex of S n . Soon after he obtained a homological version of what we call the

Lefschetz duality theorem in Section 14.5 (Theorem 14.9):

H p (M, L; Z) ⇠
= Hn p (M L; Z),

where M and L are complexes and L is a subcomplex of M ; see Figure 1.7.

Both Alexander and Lefschetz duality can be generalized to the situation where in Alexan-
der duality A is an arbitrary closed subset of S n , and in Lefschetz duality L is any compact
subset of M and M is orientable, but new kinds of cohomology need to be introduced: Čech
cohomology and Alexander–Spanier cohomology, which turn out to be equivalent. This is a
nontrivial theorem due to Dowker [13]. Then a duality theorem generalizing both Poincaré
duality and Alexander–Lefschetz duality can be proven. These matters are discussed in
Chapter 9, Section 13.8, and Chapter 14.
Proving the general version of Alexander–Lefschetz duality takes a significant amount
of work because it requires defining relative versions of Čech cohomology and Alexander–
Spanier cohomology, and to prove their equivalence as well as their equivalence to another
definition in terms of direct limits of singular cohomology groups (see Definition 14.13 and
Proposition 14.7).
When discussing the notion of duality, we would be remiss if we did not mention the
1.3. DUALITY; POINCARÉ, ALEXANDER, LEFSCHETZ 31

M
L

H p (M, L, Z) ¥ Hn-p (M-L; Z) M-L

Figure 1.7: A representation of Lefschetz duality when M is the simplicial complex consisting
of two solid tetrahedra while L is the subcomplex consisting of the front left peach face, the
back right pink face, and the solid red edge.

important contributions of Pontrjagin. In a paper published in 1931 Pontrjagin investigates

the duality between a closed subset A of Rn homeomorphic to a simplicial complex and
Rn A. Pontrjagin introduces for the first time the notion of a nondegenerate pairing
' : U ⇥ V ! G between two finitely abelian groups U and V , where G is another abelian
group (he uses G = Z or G = Z/mZ). This is a bilinear map ' : U ⇥ V ! G such that
if '(u, v) = 0 for all v 2 V , then u = 0, and if '(u, v) = 0 for all u 2 U , then v = 0.
Pontrjagin proves that U and V are isomorphic for his choice of G, and applies the notion of
nondegenerate pairing to Poincaré duality and to a version of Alexander duality for certain
subsets of Rn . Pontrjagin also introduces the important notion of direct limit (see Section
8.3, Definition 8.10) which, among other things, plays a crucial role in the definition of Čech
cohomology and in the construction of a sheaf from a presheaf (see Chapter 10).
In another paper published in 1934, Pontrjagin states and proves his famous duality
theory between discrete and compact abelian topological groups. In this situation, U is a
discrete group, G = R/Z ⇠= S 1 , and V = U b = Hom(U, R/Z) (with the topology of simple
convergence). Pontrjagin applies his duality theorem to a version of Alexander duality for
compact subsets of Rn and for a version of Čech homology (cohomology had not been defined
yet).
One notion that we still need to address, especially since it has appeared numerous times
in our aforementioned discussions, is Čech cohomology. We will do so in the next section.
It turns out that Čech cohomology accommodates very general types of coefficients, namely
presheaves and sheaves. In Chapters 8, 9 and 10 we introduce these notions that play a major
role in many area of mathematics, especially algebraic geometry and algebraic topology.
One can say that from a historical point of view, all the notions we presented so far are
32 CHAPTER 1. INTRODUCTION

discussed in the landmark book by Eilenberg and Steenrod [15] (1952). This is a beautiful
book well worth reading, but it is not for the beginner. The next landmark book is Spanier’s
[59] (1966). It is easier to read than Eilenberg and Steenrod but still quite demanding.
The next era of algebraic topology begins with the introduction of the notion of sheaf by
Jean Leray around 1946.

1.4 Presheaves, Sheaves, and Čech Cohomology

The machinery of sheaves is applicable to problems designated by the vague notion of “pas-
sage from local to global properties.” When some mathematical object attached to a topo-
logical space X can be “restricted” to any open subset U of X, and that restriction is known
for sufficiently small U , what can be said about that “global” object? For example, consider
the continuous functions defined over R2 and their restrictions to open subsets of R2 .
Problems of this type had arisen since the 1880s in complex analysis in several variables
and had been studied by Poincaré, Cousin, and later H. Cartan and Oka. Beginning in 1942,
Leray considered
L a similar problem in cohomology. Given a space X, when the cohomology
H ⇤ (U ; G) = L p 0 H p (U ; G) is known for sufficiently small U , what can be said about
H ⇤ (X; G) = p 0 H p (X; G)?
Leray devised some machinery in 1946 that was refined and generalized by H. Cartan,
M. Lazard, A. Borel, Koszul, Serre, Godement, and others to yield the notions of presheaves
and sheaves.
Given a topological space X and a class C of structures (a category), say sets, vector
spaces, R-modules, groups, commutative rings, etc., a presheaf on X with values in C consists
of an assignment of some object F(U ) in C to every open subset U of X and of a map
F(i) : F(U ) ! F(V ) of the class of structures in C to every inclusion i : V ! U of open
subsets V ✓ U ✓ X, such that

F(i j) = F(j) F(i)

F(idU ) = idF (U ) ,

for any two inclusions i : V ! U and j : W ! V , with W ✓ V ✓ U ; see Figure 1.8.

Note that the order of composition is switched in F(i j) = F(j) F(i).
Intuitively, the map F(i) : F(U ) ! F(V ) is a restriction map if we think of F(U ) and
F(V ) as sets of functions (which is often the case). For this reason, the map F(i) : F(U ) !
F(V ) is also denoted by ⇢UV : F(U ) ! F(V ), and the first equation of the above definition
is expressed by
⇢UW = ⇢VW ⇢UV .
Presheaves, as defined above and in Section 8.1, are typically used to keep track of local
information assigned to a global object (the space X). It is usually desirable to use consistent
1.4. PRESHEAVES, SHEAVES, AND ČECH COHOMOLOGY 33

F(U) F(U)
F(V)

F(U)

U
V
U

X = R2

Figure 1.8: A schematic representation of the presheaf of continuous real valued function on
X = R2 . An open set U is a circle in the plane while F(U ) is the “balloon” of functions
floating above U .

local information to recover some global information, but this requires a sharper notion, that
of a sheaf.
As stated at the beginning of Section 8.2, the motivation for the extra condition that a
sheaf should satisfy is this. Suppose we consider the presheaf of continuous functions on a
topological space X. If U is any open subset of X and if (Ui )i2I is an open cover of U , for
any family (fi )i2I of continuous functions fi : Ui ! R, if fi and fj agree on every overlap
Ui \ Uj , then the fi patch to a unique continuous function f : U ! R whose restriction to
Ui is fi .
Given a topological space X and a class C of structures (a category), say sets, vector
spaces, R-modules, groups, commutative rings, etc., a sheaf on X with values in C is a
presheaf F onSX such that for any open subset U of X, for every open cover (Ui )i2I of U
(that is, U = i2I Ui for some open subsets Ui ✓ U of X), the following conditions hold:
(G) (Gluing condition) For every family (fi )i2I with fi 2 F(Ui ), if the fi are consistent,
which means that
U
⇢UUii \Uj (fi ) = ⇢Uji \Uj (fj ) for all i, j 2 I,
then there is some f 2 F(U ) such that ⇢UUi (f ) = fi for all i 2 I; see Figure 1.9.

(M) (Monopresheaf condition) For any two elements f, g 2 F(U ), if f and g agree on all
the Ui , which means that

⇢UUi (f ) = ⇢UUi (g) for all i 2 I,

then f = g.
34 CHAPTER 1. INTRODUCTION

f
fj

fi

Uj
Ui
U

X = R2

Figure 1.9: Let F be the sheaf of continuous real valued functions on X = R2 . Let U =
U1 [ U2 . The graph of the pink function f1 and the peach function f2 glue together over
U1 \ U2 to form the continuous function f : U ! R2 .

Many (but not all) objects defined on a manifold are sheaves: the smooth functions
C 1 (U ), the smooth di↵erential p-forms Ap (U ), the smooth vector fields X(U ), where U is
any open subset of M .
Given any commutative ring R and a fixed R-module G, the constant presheaf GX is
defined such that GX (U ) = G for all nonempty open subsets U of X, and GX (;) = (0). The
constant sheaf GeX is the sheaf given by GeX (U ) = the set of locally constant functions on
U (the functions f : U ! G such that for every x 2 U there is some open subset V of U
containing x such that f is constant on V ), and GeX (;) = (0); see Figure 1.10.

In general a presheaf is not a sheaf. For example, the constant presheaf is not a sheaf.
However, there is a procedure for converting a presheaf to a sheaf. We will return to this
process in Section 1.5.
Čech cohomology with values in a presheaf of R-modules involves open covers of the
topological space X; see Chapter 9.
Apparently, Čech himself did not introduce Čech cohomology, but he did introduce Čech
homology using the notion of open cover (1932). Dowker, Eilenberg, and Steenrod introduced
Čech cohomology in the early 1950s.
X, a family U = (Uj )j2J is an open cover of X if the Uj are
Given a topological space S
open subsets of X and if X = j2J Uj . Given any finite sequence I = (i0 , . . . , ip ) of elements
of some index set J (where p 0 and the ij are not necessarily distinct), we let

UI = Ui0 ···ip = Ui0 \ · · · \ Uip .

1.4. PRESHEAVES, SHEAVES, AND ČECH COHOMOLOGY 35

z=4

z=2

z=1 U3
U2
U1

X = R2

Figure 1.10: Let F be the sheaf of continuous real valued functions over X = R2 , and let
U = U1 [ U2 [ U3 , a disjoint union. The function f is locally constant over U since it takes
a constant value over each Ui , where 1  i  3.

Note that it may happen that UI = ;. We denote by Ui0 ···ibj ···ip the intersection

ci \ · · · \ Uip
Ui0 ···ibj ···ip = Ui0 \ · · · \ U j

of the p subsets obtained by omitting Uij from Ui0 ···ip = Ui0 \ · · · \ Uip (the intersection of
the p + 1 subsets); see Figure 1.11.
Now given a presheaf F of R-modules, the R-module of Čech p-cochains C p (U , F) is the
set of all functions f with domain J p+1 such that f (i0 , . . . , ip ) 2 F(Ui0 ···ip ); in other words,
Y
C p (U , F) = F(Ui0 ···ip ),
(i0 ,...,ip )2J p+1

the set of all J p+1 -indexed families (fi0 ,...,ip )(i0 ,...,ip )2J p+1 with fi0 ,...,ip 2 F(Ui0 ···ip ). Observe
that the coefficients (the modules F(Ui0 ···ip )) can “vary” from open subset to open subset.
We have p + 1 inclusion maps
p
j: Ui0 ···ip ! Ui0 ···ibj ···ip , 0  j  p.
p
Each inclusion map j: Ui0 ···ip ! Ui0 ···ibj ···ip induces a map

F( jp ) : F(Ui0 ···ibj ···ip ) ! F(Ui0 ···ip )

Ui ···ib ···ip
which is none other that the restriction map ⇢Ui0 ···ipj which, for the sake of notational
0

simplicity, we also denote by ⇢ji0 ···ip .

36 CHAPTER 1. INTRODUCTION

U1 U2

U123

U1 U2
U3 U123^

U1

U123
^
U2

U^123 U3

U3

Figure 1.11: An illustration of the notation U123 = U1 \ U2 \ U3 and the three cases of
ci \ · · · \ Uip , where i0 = 1 and ip = 3.
Ui0 ···ibj ···ip = Ui0 \ · · · \ U j

Given a topological space X, an open cover U = (Uj )j2J of X, and a presheaf of R-

modules F on X, the coboundary maps Fp : C p (U , F) ! C p+1 (U , F) are given by
p+1
p
X p+1
F = ( 1)j F( j ), p 0.
j=0

More explicitly, for any p-cochain f 2 C p (U , F), for any sequence (i0 , . . . , ip+1 ) 2 J p+2 , we
have
p+1
p
X
( F f )i0 ,...,ip+1 = ( 1)j ⇢ji0 ···ip+1 (fi0 ,...,ibj ,...,ip+1 ).
j=0

Unravelling the above definition for p = 0 we have

0
( F f )i,j = ⇢0ij (fj ) ⇢1ij (fi ),

and for p = 1 we have

1
( F f )i,j,k = ⇢0ijk (fj,k ) ⇢1ijk (fi,k ) + ⇢2ijk (fi,j ).
p+1 p
It is easy to check that F F = 0 for all p 0, so we have a chain complex of
cohomology
1 0 p 1 p p+1
F / C 0 (U , F) F F F F
0 / C 1 (U , F) / ··· / C p (U , F) / C p+1 (U , F) / ···
1.5. SHEAFIFICATION AND STALK SPACES 37

and we can define the Čech cohomology groups as follows.

Given a topological space X, an open cover U = (Uj )j2J of X, and a presheaf of R-
modules F on X, the Čech cohomology groups Ȟ p (U , F) of the cover U with values in F are
defined by
Ȟ p (U , F) = Ker Fp /Im Fp 1 , p 0.
The classical Čech cohomology groups Ȟ p (U ; G) of the cover U with coefficients in the R-
module G are the groups Ȟ p (U , GX ), where GX is the constant sheaf on X with values in
G.
The next step is to define Čech cohomology groups that do not depend on the open
cover U . This is achieved by defining a notion of refinement on covers and by taking direct
limits (see Section 8.3, Definition 8.10). Čech had used such a method in defining his Čech
homology groups, by introducing the notion of inverse limit (which, curiously, was missed
by Pontrjagin whose introduced direct limits!).
Without going into details, given two covers U = (Ui )i2I and V = (Vj )j2J of a space X,
we say that V is a refinement of U , denoted U V, if there is a function ⌧ : J ! I such that

Vj ✓ U⌧ (j) for all j 2 J.

Under this notion of refinement, the open covers of X form a directed preorder, and the
family (Ȟ p (U , F))U is what is called a direct mapping family so its direct limit

lim Ȟ p (U , F)
!U

makes sense. We define the Čech cohomology groups Ȟ p (X, F) with values in F by

Ȟ p (X, F) = lim Ȟ p (U , F).

!U

The classical Čech cohomology groups Ȟ p (X; G) with coefficients in the R-module G are the
groups Ȟ p (X, GX ) where GX is the constant presheaf with value G. All this is presented in
Chapter 9.
A natural question to ask is how does the classical Čech cohomology of a space com-
pare with other types of cohomology, in particular singular cohomology. In general Čech
cohomology can di↵er from singular cohomology, but for manifolds it agrees. Classical Čech
cohomology also agrees with de Rham cohomology of the constant presheaf RX . These
results are hard to prove; see Chapter 13.

1.5 Sheafification and Stalk Spaces

One of the major goals of this book is to introduce sheaf cohomology. This means we need to
develop a deeper understanding of mappings between sheaves. A map (or morphism) ' : F !
38 CHAPTER 1. INTRODUCTION

G of presheaves (or sheaves) F and G on X consists of a family of maps 'U : F(U ) ! G(U )
of the class of structures in C, for any open subset U of X, such that

'V (⇢F )UV = (⇢G )UV 'U

for every pair of open subsets U, V such that V ✓ U ✓ X. Equivalently, the following
diagrams commute for every pair of open subsets U, V such that V ✓ U ✓ X
'U
F(U ) / G(U )

(⇢F )U
V (⇢G )U
V
✏ ✏
F(V ) 'V
/ G(V ).

The notion of kernel Ker ' and image Im ' of a presheaf or sheaf map ' : F ! G is easily
defined. The presheaf Ker ' is defined by (Ker ')(U ) = Ker 'U , and the presheaf Im ' is
defined by (Im ')(U ) = Im 'U . In the case of presheaves, they are also presheaves, but in
the case of sheaves, the kernel Ker ' is indeed a sheaf, but the image Im ' is not a sheaf in
general.
This failure of the image of a sheaf map to be a sheaf is a problem that causes significant
technical complications. In particular, it is not clear what it means for a sheaf map to be
surjective, and a “good” definition of the notion of an exact sequence of sheaves is also
unclear.
Fortunately, there is a procedure for converting a presheaf F into a sheaf Fe which is
reasonably well-behaved. This procedure is called sheafification. There is a sheaf map
⌘ : F ! Fe which is generally not injective.
The sheafification process is universal in the sense that given any presheaf F and any
sheaf G, for any presheaf map ' : F ! G, there is a unique sheaf map ' b : Fe ! G such that

b ⌘F
'='

as illustrated by the following commutative diagram

⌘F
F e
/F

b
'
'
✏
G;

see Theorem 10.12.

The sheafification process involves constructing a topological space SF from the presheaf
F that we call the stalk space of F; see Figure 1.12. Godement calls it the espace étalé.
The stalk space is the disjoint union of sets (modules) Fx called stalks. Each stalk Fx is
1.5. SHEAFIFICATION AND STALK SPACES 39

F(U)

Fx Fx Fy Fz
SF

x x y z
U X=R X=R

Figure 1.12: Let X = R and F be the sheaf of real valued continuous functions. An element
F(U ) is represented by the floating balloon. By “collapsing” the balloon (via the direct
limiting process), we form the stalk Fx , which is represented as a vertical line.

the direct limit lim(F(U ))U 3x of the family of modules F(U ) for all “small” open sets U
!
containing x (see Definition 10.1).
There is a surjective map p : SF ! X which, under the topology given to SF, is a local
homeomorphism, which means that for every y 2 SF, there is some open subset V of SF
containing y such that the restriction of p to V is a homeomorphism. The sheaf Fe consists
of the continuous sections of p, that is, the continuous functions s : U ! SF such that
p s = idU , for any open subset U of X. This construction is presented in detail in Sections
10.1, 10.2, and 10.4.
The construction of the pair (SF, p) from a presheaf F suggests another definition of a
sheaf as a pair (E, p), where E is a topological space and p : E ! X is a surjective local
homeomorphism onto another space X. Such a pair (E, p) is often called a sheaf space,
but we prefer to call it a stalk space. This is the definition that was given by H. Cartan
and M. Lazard around 1950. The sheaf E associated with the stalk space (E, p) is defined
as follows: for any open subset U or X, the sections of E are the continuous sections
s : U ! E, that is, the continuous functions such that p s = id; see Figure 1.13. We can
also define a notion of map between two stalk spaces. Stalk spaces are discussed in Section
10.3.
As this stage, given a topological space X we have three categories (classes of objects):
(1) The category Psh(X) of presheaves and their morphisms.

(2) The category Sh(X) of sheaves and their morphisms.

40 CHAPTER 1. INTRODUCTION

s4 E
s3
s2 p

s1
X
U

Figure 1.13: A schematic representation of a stalk space (E, p). We drew four sections over
U , where each section is a colored curve such that p s = id.

(3) The category StalkS(X) of stalk spaces and their morphisms.

There is also a functor
S : PSh(X) ! StalkS(X)
from the category PSh(X) to the category StalkS(X) given by the construction of a stalk
space SF from a presheaf F, (S(F) = SF), and a functor

: StalkS(X) ! Sh(X)

from the category StalkS(X) to the category Sh(X), given by the sheaf E of continuous
sections of E. Here we are using the term functor in an informal way. A more precise
definition is given in Sections 1.7 and 10.10.
Note that every sheaf F is also a presheaf, and that every map ' : F ! G of sheaves is
also a map of presheaves. Therefore, we have an inclusion map

i : Sh(X) ! PSh(X),

which is a functor. As a consequence, S restricts to an operation (functor)

S : Sh(X) ! StalkS(X).

There is also a map ⌘ which maps a presheaf F to the sheaf S(F) = F. e This map ⌘ is
a natural isomorphism between the functors id (the identity functor) and S from Sh(X)
1.6. COKERNELS AND IMAGES OF SHEAF MAPS 41

to itself. In other words, if we take F, form the stalk space SF, then turn this stalk space
into the sheaf of continuous sections SF, this new sheaf is isomorphic to F.
We can also define a map ✏ which takes a stalk space (E, p) and makes the stalk space
S E. The map ✏ is a natural isomorphism between the functors id (the identity functor)
and S from StalkS(X) to itself. In other words, if we take the stalk space (E, p), form the
sheaf of continuous sections E, then form the stalk space of E, namely S E, this new
stalk space is isomorphic to (E, p).
Then we see that the two operations (functors)

S : Sh(X) ! StalkS(X) and : StalkS(X) ! Sh(X)

are almost mutual inverses, in the sense that there is a natural isomorphism ⌘ between S
and id and a natural isomorphism ✏ between S and id. In such a situation, we say that the
classes (categories) Sh(X) and StalkS(X) are equivalent. The upshot is that it is basically
a matter of taste (or convenience) whether we decide to work with sheaves or stalk spaces.
In fact, for the aspects of sheaf cohomology that deal with soft and fine sheaves (Sections
13.5 and 13.6), it is best to use the stalk space construction of a sheaf.
We also have the operator (functor)

S : PSh(X) ! Sh(X)
e Theorem 10.12 can be restated as saying
which “sheafifies” a presheaf F into the sheaf F.
that there is an isomorphism

HomPSh(X) (F, i(G)) ⇠ e G),

= HomSh(X) (F,
between the set (category) of maps between the presheaves F and i(G) and the set (category)
of maps between the sheaves Fe and G. In fact, such an isomorphism is natural, so in
categorical terms, i and e = S are adjoint functors.
All this is explained in Sections 10.3 and 10.4.

1.6 Cokernels and Images of Sheaf Maps

We still need to define the image of a sheaf map in such a way that the notion of exact
sequence of sheaves makes sense. Recall that if f : A ! B is a homomorphism of modules,
the cokernel Coker f of f is defined by B/Im f . It is a measure of the surjectivity of f . We
also have the projection map coker(f ) : B ! Coker f , and observe that

Im f = Ker coker(f ).

The above suggests defining notions of cokernels of presheaf maps and sheaf maps. For a
presheaf map ' : F ! G this is easy, and we can define the presheaf cokernel PCoker('). It
comes with a presheaf map pcoker(') : G ! PCoker(').
42 CHAPTER 1. INTRODUCTION

If F and G are sheaves, we define the sheaf cokernel SCoker(') as the sheafification of
PCoker('). It also comes with a presheaf map scoker(') : G ! SCoker(').
Then it can be shown that if ' : F ! G is a sheaf map, SCoker(') = (0) i↵ the stalk
maps 'x : Fx ! Gx are surjective for all x 2 X; see Proposition 10.19.
It follows that the “correct” definition for the image SIm ' of a sheaf map ' : F ! G is
SIm ' = Ker scoker(').
With this definition, a sequence of sheaves
'
F / G /H

is said to be exact if SIm ' = Ker . Then it can be shown that

'
F / G /H

is an exact sequence of sheaves i↵ the sequence

'x x
Fx / Gx / Hx

is an exact sequence of R-modules (or rings) for all x 2 X; see Proposition 10.24. This
second characterization of exactness (for sheaves) is usually much more convenient than the
first condition.
The definitions of cokernels and images of presheaves and sheaves as well as the notion
of exact sequences of presheaves and sheaves are discussed in Sections 10.6, 10.7, 10.8, 10.9,
and 10.10.

1.7 Injective and Projective Resolutions;

Derived Functors
In order to define, even informally, the concept of derived functor, we need to describe what
are functors and exact functors.
Suppose we have two types of structures (categories) C and D (for concreteness, think
of C as the class of R-modules over some commutative ring R with an identity element 1
and of D as the class of abelian groups), and we have a transformation T (a functor) which
works as follows:
(i) Each object A of C is mapped to some object T (A) of D.
f
(ii) Each map A / B between two objects A and B in C (of example, an R-linear map)
T (f )
is mapped to some map T (A) / T (B) between the objects T (A) and T (B) in D
(for example, a homomorphism of abelian groups) in such a way that the following
properties hold:
1.7. INJECTIVE AND PROJECTIVE RESOLUTIONS; DERIVED FUNCTORS 43

f g
(a) Given any two maps A / B and B /C between objects A, B, C in C such
g f f g
that the composition A /C = A /B / C makes sense, the composition
T (f )
T (A) / T (B) T (g) / T (C) makes sense in D, and
T (g f ) = T (g) T (f ).

(b) If A
idA
/ A is the identity map of the object A in C, then T (A) T (idA )/ T (A) is the
identity map of T (A) in D; that is,
T (idA ) = idT (A) .

Whenever a transformation T : C ! D satisfies the Properties (i), (ii), (a), (b), we call it a
(covariant) functor from C to D.
If T : C ! D satisfies Properties (i), (b), and if Properties (ii) and (a) are replaced by
the Properties (ii’) and (a’) below
f
(ii’) Each map A / B between two objects A and B in C is mapped to some map
T (f )
T (B) / T (A) between the objects T (B) and T (A) in D in such a way that the
following properties hold:
f g
(a’) Given any two maps A / B and B /C between objects A, B, C in C such
g f f g
that the composition A /C = A /B / C makes sense, the composition
T (g) T (f )
T (C) / T (B) / T (A) makes sense in D, and
T (g f ) = T (f ) T (g),

then T is called a contravariant functor from C to D.

An example of a (covariant) functor is the functor Hom(A, ) (for a fixed R-module A)
from R-modules to R-modules which maps a module B to the module Hom(A, B) and a mod-
ule homomorphism f : B ! C to the module homomorphism Hom(A, f ) from Hom(A, B)
to Hom(A, C) given by
Hom(A, f )(') = f ' for all ' 2 Hom(A, B).
Another example is the functor T from R-modules to R-modules such that T (A) = A ⌦R M
for any R-module A, and T (f ) = f ⌦R idM for any R-linear map f : A ! B.
An example of a contravariant functor is the functor Hom( , A) (for a fixed R-module A)
from R-modules to R-modules which maps a module B to the module Hom(B, A) and a mod-
ule homomorphism f : B ! C to the module homomorphism Hom(f, A) from Hom(C, A) to
Hom(B, A) given by
Hom(f, A)(') = ' f for all ' 2 Hom(C, A).
44 CHAPTER 1. INTRODUCTION

Categories and functors were introduced by Eilenberg and Mac Lane, first in a paper
published in 1942, and then in a more complete paper published in 1945.
Given a type of structures (category) C, let us denote the set of all maps from an object
A to an object B by HomC (A, B). For all the types of structures C that we will dealing
with, each set HomC (A, B) has some additional structure; namely it is an abelian group.
Intuitively speaking an abelian category is a category in which the notion of kernel and
cokernel of a map makes sense. Then we can define the notion of image of a map f as the
kernel of the cokernel of f , so the notion of exact sequence makes sense, as we did in Section
1.6. The categories of R-modules and the categories of sheaves (or presheaves) are abelian
categories. For more details, see Sections 10.10 and 10.11.
A sequence of R-modules and R-linear maps (more generally objects and maps between
objects in an abelian category)
f g
0 /A / B / C /0 (⇤)

is a short exact sequence if

(1) f is injective.
(2) Im f = Ker g.
(3) g is surjective.

According to Dieudonné [11], the notion of exact sequence first appeared in a paper of
Hurewicz (1941), and then in a paper of Eilenberg and Steenrod and a paper of H. Cartan,
both published in 1945. In 1947, Kelly and Pitcher generalized the notion of exact sequence
to chain complexes, and apparently introduced the terminology exact sequence. In their 1952
treatise [15], Eilenberg and Steenrod took the final step of allowing a chain complex to be
indexed by Z (as we do in Section 2.5).
Given two types of structures (categories) C and D in each of which the concept of
exactness is defined (abelian categories), given an additive functor T : C ! D, by applying
T to the short exact sequence (⇤) we obtain the sequence
T (f ) T (g)
0 / T (A) / T (B) / T (C) / 0, (⇤⇤)

which is a chain complex (since T (g) T (f ) = 0). Then the following question arises:
Is the sequence (⇤⇤) also exact?
In general, the answer is no, but weaker forms of preservation of exactness suggest
themselves.
A functor T : C ! D, is said to be exact if whenever the sequence

0 /A / B / C /0
1.7. INJECTIVE AND PROJECTIVE RESOLUTIONS; DERIVED FUNCTORS 45

is exact in C, then the sequence

0 / T (A) / T (B) / T (C) / 0

is exact in D; left exact if whenever the sequence

0 / A / B /C

is exact in C, then the sequence

0 / T (A) / T (B) / T (C)

is exact; and right exact if whenever the sequence

A / B /C / 0

is exact in C, then the sequence

T (A) / T (B) / T (C) / 0

is exact.
If T : C ! D is a contravariant functor, then T is said to be exact if whenever the
sequence
0 /A /B /C /0

is exact in C, then the sequence

0 / T (C) / T (B) / T (A) / 0

is exact in D; left exact if whenever the sequence

A / B /C / 0

is exact in C, then the sequence

0 / T (C) / T (B) / T (A)

is exact; and right exact if whenever the sequence

0 / A / B /C

is exact in C, then the sequence

T (C) / T (B) / T (A) / 0

is exact.
46 CHAPTER 1. INTRODUCTION

For example, the functor Hom( , A) is (contravariant) left-exact but not exact in general
(see Section 2.1). Similarly, the functor Hom(A, ) is left-exact but not exact in general (see
Section 2.4).
Modules for which the functor Hom(A, ) is exact play an important role. They are
called projective modules. Similarly, modules for which the functor Hom( , A) is exact are
called injective modules.
The functor ⌦R M is right-exact but not exact in general (see Section 2.4). Modules
M for which the functor ⌦R M is exact are called flat.
A good deal of homological algebra has to do with understanding how much a module
fails to be projective or injective (or flat).
Injective and projective modules are also characterized by extension properties. As we
will see later, these extension characterizations can be used to define injective and projective
objects in an abelian category.

(1) A module P is projective i↵ for any surjective linear map h : A ! B and any linear
map f : P ! B, there is some linear map fb: P ! A lifting f : P ! B in the sense
that f = h fb, as in the following commutative diagram:

P
fb
f
 ✏
A h
/ B / 0.

(2) A module I is injective i↵ for any injective linear map h : A ! B and any linear map
f : A ! I, there is some linear map fb: B ! I extending f : A ! I in the sense that
f = fb h, as in the following commutative diagram:
h
0 / A / B
f
✏ ~ fb
I.
See Section 11.1.

Injective modules were introduced by Baer in 1940 and projective modules by Cartan
and Eilenberg in the early 1950s. Every free module is projective. Injective modules are
more elusive. If the ring R is a PID, an R-module M is injective i↵ it is divisible (which
means that for every nonzero 2 R, the map given by u 7! u for u 2 M is surjective).
One of the most useful properties of projective modules is that every module M is the
image of some projective (even free) module P , which means that there is a surjective
homomorphism ⇢ : P ! M . Similarly, every module M can be embedded in an injective
1.7. INJECTIVE AND PROJECTIVE RESOLUTIONS; DERIVED FUNCTORS 47

module I, which means that there is an injective homomorphism i : M ! I. This second

fact is harder to prove (see Baer’s embedding theorem, Theorem 11.6).
The above properties can be used to construct inductively projective and injective resolu-
tions of a module M , a process that turns out to be remarkably useful. Intuitively, projective
resolutions measure how much a module deviates from being projective, and injective reso-
lutions measure how much a module deviates from being injective.
Hopf introduced free resolutions in 1945. A few years later Cartan and Eilenberg defined
projective and injective resolutions.
Given any R-module A, a projective resolution of A is any exact sequence

··· / Pn
dn / Pn 1 d n 1
/ ··· / P1
d1
/ P0
p0
/ A / 0 (⇤1 )

in which every Pn is a projective module. The exact sequence

··· / Pn
dn / Pn 1 d n 1
/ ··· / P1
d1
/ P0

obtained by truncating the projective resolution of A after P0 is denoted by PA , and the

projective resolution (⇤1 ) is denoted by
p0
PA /A / 0.

Given any R-module A, an injective resolution of A is any exact sequence

i0
/ I0 d0 / I1 d1 / In dn / I n+1
0 / A / ··· / ··· (⇤⇤1 )

in which every I n is an injective module. The exact sequence

d0 d1 dn
I0 / I1 / ··· / In / I n+1 / ···

obtained by truncating the injective resolution of A before I 0 is denoted by IA , and the

injective resolution (⇤⇤1 ) is denoted by
i0
0 / A / IA .

Now suppose that we have a functor T : C ! D, where C is the category of R-modules

and D is the category of abelian groups. If we apply T to PA we obtain the chain complex
T (d1 ) T (d2 ) T (dn )
0o T (P0 ) o T (P1 ) o ··· o T (Pn 1 ) o T (Pn ) o ··· , (Lp)

denoted T (PA ). The above is no longer exact in general but it defines homology groups
Hp (T (PA )).
48 CHAPTER 1. INTRODUCTION

Similarly If we apply T to IA we obtain the cochain complex

T (d0 ) T (d1 ) T (dn )

0 / T (I 0 ) / T (I 1 ) / ··· / T (I n ) / T (I n+1 ) / ··· , (Ri)

denoted T (IA ). The above is no longer exact in general but it defines cohomology groups
H p (T (IA )).
The reason why projective resolutions are so special is that even though the homology
groups Hp (T (PA )) appear to depend on the projective resolution PA , in fact they don’t; the
groups Hp (T (PA )) only depend on A and T . This is proven in Theorem 11.28.
Similarly, the reason why injective resolutions are so special is that even though the
cohomology groups H p (T (IA )) appear to depend on the injective resolution IA , in fact they
don’t; the groups H p (T (IA )) only depend on A and T . This is proven in Theorem 11.27.
Proving the above facts takes some work; we make use of the comparison theorems; see
Section 11.2, Theorem 11.17 and Theorem 11.21. In view of the above results, given a functor
T as above, Cartan and Eilenberg were led to define the left derived functors Ln T of T by

Ln T (A) = Hn (T (PA )),

for any projective resolution PA of A, and the right derived functors Rn T of T by

Rn T (A) = H n (T (IA )),

for any injective resolution IA of A. The functors Ln T and Rn T can also be defined on maps.
If T is right-exact, then L0 T is isomorphic to T (as a functor), and if T is left-exact, then
R0 T is isomorphic to T (as a functor).
For example, the left derived functors of the right-exact functor TB (A) = A ⌦ B (with
B fixed) are the “Tor” functors. We have TorR ⇠ R
0 (A, B) = A ⌦ B, and the functor Tor1 ( , G)
plays an important role in comparing the homology of a chain complex C and the homology
of the complex C ⌦R G; see Chapter 12. Čech introduced the functor TorR 1 ( , G) in 1935
in terms of generators and relations. It is only after Whitney defined tensor products of
arbitrary Z-modules in 1938 that the definition of Tor was expressed in the intrinsic form
that we are now familar with.
There are also versions of left and right derived functors for contravariant functors.
For example, the right derived functors of the contravariant left-exact functor TB (A) =
HomR (A, B) (with B fixed) are the “Ext” functors. We have Ext0R (A, B) ⇠ = HomR (A, B),
1
and the functor ExtR ( , G) plays an important role in comparing the homology of a chain
complex C and the cohomology of the complex HomR (C, G); see Chapter 12. The Ext
functors were introduced in the context of algebraic topology by Eilenberg and Mac Lane
(1942).
Everything we discussed so far is presented in Cartan and Eilenberg’s groundbreaking
book, Cartan–Eilenberg [10], published in 1956. It is in this book that the name homological
1.8. UNIVERSAL -FUNCTORS 49

algebra is introduced. MacLane [37] (1975) and Rotman [50, 52] give more “gentle” pre-
sentations (see also Weibel [63] and Eisenbud [16]). A more sophisticated presentation of
homological algebra is found in Gelfand and Manin [23].
Derived functors can be defined for functors T : C ! D, where C or D is a more general
category than the category of R-modules or the category of abelian groups. For example, in
sheaf cohomology, the category C is the category of sheaves of rings. In general, it suffices
that C and D are abelian categories.
We say that C has enough projectives if every object in C is the image of some projec-
tive object in C, and that C has enough injectives if every object in C can be embedded
(injectively) into some injective object in C.
There are situations (for example, when dealing with sheaves) where it is useful to know
that right derived functors can be computed by resolutions involving objects that are not
necessarily injective, but T -acyclic, as defined below.
Given a left-exact functor T : C ! D, an object J 2 C is T -acyclic if Rn T (J) = (0) for
all n 1.
The following proposition shows that right derived functors can be computed using T -
acyclic resolutions.
Proposition Given an additive left-exact functor T : C ! D, for any A 2 C suppose there
is an exact sequence
✏ d0 d1 d2
0 / A / J0 / J1 / J2 / ··· (†)

in which every J n is T -acyclic (a right T -acyclic resolution JA formed by truncating (†)

before J 0 ). Then for every n 0 we have an isomorphism between Rn T (A) and H n (T (JA )).
The above proposition is used several times in Chapter 13.

1.8 Universal -Functors

The most important property of derived functors is that short exact sequences yield long
exact sequences of homology or cohomology. This property was proven by Cartan and
Eilenberg, but Grothendieck realized how crucial it was and this led him to the fundamental
concept of universal -functor . Since we will be using right derived functors much more than
left derived functors we state the existence of the long exact sequences of cohomology for
right derived functors.
Theorem Assume the abelian category C has enough injectives, let 0 ! A0 ! A !
A00 ! 0 be an exact sequence in C, and let T : C ! D be a left-exact (additive) functor.
(1) Then for every n 0, there is a map
n
(Rn T )(A00 ) ! (Rn+1 T )(A0 ),
50 CHAPTER 1. INTRODUCTION

and the sequence

0 / T (A0 ) / T (A) / T (A00 )

0

/ (R1 T )(A0 ) / ··· / ···

/ (Rn T )(A0 ) / (Rn T )(A) / (Rn T )(A00 )

n

/ (Rn+1 T )(A0 ) / ··· / ··· / ···

is exact. This property is similar to the property of the zig-zag lemma from Section 1.2.
(2) If 0 ! B 0 ! B ! B 00 ! 0 is another exact sequence in C, and if there is a
commutative diagram
0 / A0 / A / A00 / 0

✏ ✏ ✏
0 / B0 / B / B 00 / 0,
then the induced diagram beginning with
0
T (A0 ) T (A00 )
A
0 / / T (A) / /

✏ ✏ ✏
0 / T (B 0 ) / T (B) / T (B 00 ) /
0
B

and continuing with

n
Rn T (A0 ) Rn T (A00 ) (Rn+1 T )(A0 )
A
··· / / Rn T (A) / / / ···

✏ ✏ ✏ ✏
0 00
··· / n
R T (B ) / n
R T (B) / n
R T (B ) n
/ (R n+1
T )(B 0 ) / ···
B

is also commutative.
The proof of this result (Theorem 11.31) is fairly involved and makes use of the horseshoe
lemma (Theorem 11.25).
The previous theorem suggests the definition of families of functors originally proposed by
Cartan and Eilenberg [10] and then investigated by Grothendieck in his legendary “Tohoku”
paper [27] (1957).
1.8. UNIVERSAL -FUNCTORS 51

A -functors consists of a countable family T = (T n )n 0 of functors T n : C ! D that

satisfy the two conditions of the previous theorem. There is a notion of map, also called
morphism, between -functors.
Given two -functors S = (S n )n 0 and T = (T n )n 0 , a morphism ⌘ : S ! T between S
and T is a family ⌘ = (⌘ n )n 0 of natural transformations ⌘ n : S n ! T n such that a certain
diagram commutes; see Definition 11.21.
Grothendieck also introduced the key notion of universal -functor; see Grothendieck [27]
(Chapter II, Section 2.2).
A -functor T = (T n )n 0 is universal if for every -functor S = (S n )n 0 and every natural
transformation ' : T 0 ! S 0 , there is a unique morphism ⌘ : T ! S such that ⌘ 0 = '; we say
that ⌘ lifts '.
The reason why universal -functors are important is the following kind of uniqueness
property that shows that a universal -functor is completely determined by the component
T 0 ; see Proposition 11.38.
Proposition Suppose S = (S n )n 0 and T = (T n )n 0 are both universal -functors and
there is an isomorphism ' : S 0 ! T 0 (a natural transformation ' which is an isomorphism).
Then there is a unique isomorphism ⌘ : S ! T lifting '.
One might wonder whether (universal) -functors exist. Indeed there are plenty of them;
see Theorem 11.39.
Theorem Assume the abelian category C has enough injectives. For every additive left-
exact functor T : C ! D, the family (Rn T )n 0 of right derived functors of T is a -functor.
Furthermore T is isomorphic to R0 T .
In fact, the -functors (Rn T )n 0 are universal.
Grothendieck came up with an ingenious sufficient condition for a -functor to be univer-
sal: the notion of an erasable functor. Since Grothendieck’s paper is written in French, this
notion defined in Section 2.2 (Page 141) of [27] is called e↵açable, and many books and paper
use it. Since the English translation of “e↵açable” is “erasable,” as advocated by Lang, we
will use the the English word.
A functor T : C ! D is erasable (or e↵açable) if for every object A 2 C there is some
object MA and an injection u : A ! MA such that T (u) = 0. In particular this will be the
case if T (MA ) is the zero object of D. If the category C has enough injectives, it can be
shown that T is erasable i↵ T (I) = (0) for all injectives I.
Our favorite functors, namely the right derived functors Rn T , are erasable by injectives
for all n 1. The following result due to Grothendieck is crucial:
Theorem Assume the abelian category C has enough injectives. Let T = (T n )n 0 be a
-functor between two abelian categories C and D. If T n (I) = (0) for every injective I, for
all n 1, then T is a universal -functor.
52 CHAPTER 1. INTRODUCTION

Finally, by combining the previous results, we obtain the most important theorem about
universal -functors:
Theorem Assume the abelian category C has enough injectives. For every left-exact func-
tor T : C ! D, the right derived functors (Rn T )n 0 form a universal -functor such that T
is isomorphic to R0 T . Conversely, every universal -functor T = (T n )n 0 is isomorphic to
the right derived -functor (Rn T 0 )n 0 .
After all, the mysterious universal -functors are just the right derived functors of left-
exact functors. As an example, the functors ExtnR (A, ) constitute a universal -functor (for
any fixed R-module A).
The machinery of universal -functors can be used to prove that di↵erent kinds of co-
homology theories yield isomorphic groups. If two cohomology theories (HSn ( ))n 0 and
(HTn ( ))n 0 defined for objects in a category C (say, topological spaces) are given by univer-
sal -functors S and T in the sense that the cohomology groups HSn (A) and HTn (A) are given
by HSn (A) = S n (A) and HTn (A) = T n (A) for all objects A 2 C, and if HS0 (A) and HT0 (A)
are isomorphic, then HSn (A) and HTn (A) are isomorphic for all n 0. This technique will
be used in Chapter 13 to prove that sheaf cohomology and Čech cohomology are isomorphic
for paracompact spaces.
In Section 1.10 we will further see how the machinery of right derived functors can be
used to define sheaf cohomology (where the category C is the category of sheaves of R-
modules, the category D is the category of abelian groups, and T is the left exact “global
section functor”).

1.9 Universal Coefficient Theorems

Suppose we have a homology chain complex

d0 d1 dp 1 dp dp+1
0o C0 o C1 o ··· o Cp 1
o Cp o Cp+1 o ··· ,

where the Ci are R-modules over some commutative ring R with a multiplicative identity
element (recall that di di+1 = 0 for all i 0). Given another R-module G we can form the
homology complex

d0 ⌦id d1 ⌦id dp ⌦id

0o C0 ⌦ R G o C1 ⌦ R G o ··· o Cp ⌦ R G o ··· ,

obtained by tensoring with G, denoted C ⌦R G, and the cohomology complex

HomR (d0 ,G) HomR (dp+1 ,G)

0 / HomR (C0 , G) / ··· / HomR (Cp , G) / HomR (Cp+1 , G) / ···

obtained by applying HomR ( , G), and denoted HomR (C, G).

1.10. SHEAF COHOMOLOGY 53

The question is: what is the relationship between the homology groups Hp (C ⌦R G) and
the original homology groups Hp (C) in the first case, and what is the relationship between
the cohomology groups H p (HomR (C, G)) and the original homology groups Hp (C) in the
second case?
The ideal situation would be that

Hp (C ⌦R G) ⇠
= Hp (C) ⌦R G and H p (HomR (C, G)) ⇠
= HomR (Hp (C), G),
but this is generally not the case. If the ring R is nice enough and if the modules Cp are nice
enough, then Hp (C ⌦R G) can be expressed in terms of Hp (C) ⌦R G and TorR 1 (Hp 1 (C), G),
where TorR 1 ( , G) is a one of the left-derived functors of ⌦ R G, and H p
(Hom R (C, G)) can
1 1
be expressed in terms of HomR (Hp (C), G)) and ExtR (Hp 1 (C), G), where ExtR ( , G) is one
of the right-derived functors of HomR ( , G); both derived functors are defined in Section
11.2 and further discussed in Example 11.1. These formulae known as universal coefficient
theorems are discussed in Chapter 12.

1.10 Sheaf Cohomology

Given a topological space X, we define the global section functor (X, ) such that for every
sheaf of R-modules F,
(X, F) = F(X).
This is a functor from the category Sh(X) of sheaves of R-modules over X to the category
of abelian groups.
A sheaf I is injective if for any injective sheaf map h : F ! G and any sheaf map
f : F ! I, there is some sheaf map fb: G ! I extending f : F ! I in the sense that
f = fb h, as in the following commutative diagram:
h
0 / F / G
f
✏  fb
I.
This is the same diagram that we used to define injective modules in Section 1.7, but here,
the category involved is the category of sheaves.
A nice feature of the category of sheaves of R-modules is that its has enough injectives.
Proposition For any sheaf F of R-modules, there is an injective sheaf I and an injective
sheaf homomorphism ' : F ! I.
As in the case of modules, the fact that the category of sheaves has enough injectives
implies that any sheaf has an injective resolution.
On the other hand, the category of sheaves does not have enough projectives. This is the
reason why projective resolutions of sheaves are of little interest.
54 CHAPTER 1. INTRODUCTION

Another good property is that the global section functor is left-exact. Then as in the
case of modules in Section 1.7, the cohomology groups induced by the right derived functors
Rp (X, ) are well defined.
The cohomology groups of the sheaf F (or the cohomology groups of X with values in F),
denoted by H p (X, F), are the groups Rp (X, )(F) induced by the right derived functor
Rp (X, ) (with p 0).
To compute the sheaf cohomology groups H p (X, F), pick any resolution of F
d0 d1 d2
0 / F / I0 / I1 / I2 / ···

by injective sheaves I n , apply the global section functor (X, ) to obtain the complex of
R-modules
1 0 1 2
0 / I 0 (X) / I 1 (X) / I 2 (X) / ··· ,

and then
H p (X, F) = Ker p
/Im p 1
.
By Theorem 11.47 (stated in the previous section) the right derived functors Rp (X, )
constitute a universal -functor, so all the properties of -functors apply.
In principle, computing the cohomology groups H p (X, F) requires finding injective reso-
lutions of sheaves. However injective sheaves are very big and hard to deal with. Fortunately,
there is a class of sheaves known as flasque sheaves (due to Godement) which are (X, )-
acyclic, and every sheaf has a resolution by flasque sheaves. Therefore, by Proposition 11.34
(stated in the previous section), the cohomology groups H p (X, F) can be computed using
flasque resolutions.
Then we compare sheaf cohomology (defined by derived functors) to the other kinds of
eX ).
cohomology defined so far: de Rham, singular, Čech (for the constant sheaf G
If the space X is paracompact, then it turns out that for any sheaf F, the Čech cohomol-
ogy groups Ȟ p (X, F) are isomorphic to the cohomology groups H p (X, F). Furthermore, if
e are isomorphic,
F is a presheaf, then the Čech cohomology groups Ȟ p (X, F) and Ȟ p (X, F)
where Fe is the sheafification of F. Several other results (due to Leray and Henri Cartan)
about the relationship between Čech cohomology and sheaf cohomology will be stated.
When X is a topological manifold (thus paracompact), for every R-module G, we will
show that the singular cohomology groups H p (X; G) are isomorphic to the cohomology
eX ) of the constant sheaf G
groups H p (X, G eX . Technically, we will need to define soft and
fine sheaves.
We will also define Alexander–Spanier cohomology and prove that it is equivalent to sheaf
cohomology (and Čech cohomology) for paracompact spaces and for the constant sheaf G eX .
In summary, for manifolds, singular cohomology, Čech cohomology, Alexander–Spanier
cohomology, and sheaf cohomology all agree (for the constant sheaf G eX ). For smooth mani-
folds, we can add de Rham cohomology to the above list of equivalent cohomology theories,
e X . All these results are presented in Chapter 13.
for the constant sheaf R
1.11. ALEXANDER AND ALEXANDER–LEFSCHETZ DUALITY 55

1.11 Alexander and Alexander–Lefschetz Duality

The goal of Chapter 14 is to present various generalizations of Poincaré duality. These ver-
sions of duality involve taking direct limits of direct mapping families of singular cohomology
groups which, in general, are not singular cohomology groups. However, such limits are iso-
morphic to Alexander–Spanier cohomology groups, and thus to Čech cohomology groups.
These duality results also require relative versions of homology and cohomology.

1.12 Spectral Sequences

A spectral sequence is a tool of homological algebra whose purpose is to approximate the
cohomology (or homology) H(M ) of a module M endowed with a family (F p M )p2Z of sub-
modules such that F p+1 M ✓ F p M for all p and
[
M= F p M,
p2Z

called a filtration. The module M is also equipped with a linear map d : M ! M called
di↵erential such that d d = 0, so that it makes sense to define

H(M ) = Ker d/Im d.

We say that (M, d) is a di↵erential module. To be more precise, the filtration induces
cohomology submodules H(M )p of H(M ), the images of H(F p M ) in H(M ), and a spectral
sequence is a sequence of modules Erp (equipped with a di↵erential dpr ), for r 1, such that
Erp approximates the “graded piece” H(M )p /H(M )p+1 of H(M ).
Actually, to be useful, the machinery of spectral sequences must be generalized to filtered
cochain complexes. Technically this implies dealing with objects Erp,q involving three indices,
which makes its quite challenging to follow the exposition.
Many presentations jump immediately to the general case, but it seems pedagogically
advantageous to begin with the simpler case of a single filtered di↵erential module. This the
approach followed by Serre in his dissertation [56] (Pages 24–104, Annals of Mathematics,
54 (1951), 425–505), Godement [24], and Cartan and Eilenberg [10]. Spectral sequences are
discussed in great detail in Chapter 15.
There are several methods for defining spectral sequences, including the following three:

(1) Koszul’s original approach as described by Serre [56] and Godement [24]. In our opinion
it is the simplest method to understand what is going on.

(2) Cartan and Eilenberg’s approach [10]. This is a somewhat faster and slicker method
than the previous method.
56 CHAPTER 1. INTRODUCTION

(3) Exact couples of Massey (1952). This somewhat faster method for defining spectral
sequences is adopted by Rotman [50, 52] and Bott and Tu [4]. Mac Lane [37], Weibel
[63], and McCleary [44] also present it and show its equivalence with the first approach.
It appears to be favored by algebraic topologists. This approach leads to spectral
sequences in a quicker fashion and is more general because exact couples need not
arise from a filtration, but our feeling is that it is even more mysterious to a novice
than the first two approaches.

We will primarily follow Method (1) and present Method (2) and Method (3) in starred
sections (Method (2) in Section 15.15 and Method (3) in Section 15.14). All three methods
produce isomorphic sequences, and we will show their equivalence.

1.13 Suggestions On How to Use This Book

This book basically consists of two parts. The first part covers fairly basic material presented
in the first seven chapters. The second part deals with more sophisticated material including
sheaves, derived functors, sheaf cohomology, and spectral sequences.
Chapter 3 on de Rham cohomology, Chapter 5 on simplicial homology and cohomology,
and Chapter 6 on CW-complexes, are written in such a way that they are pretty much
independent of each other and of the rest of book, and thus can be safely skipped. Readers
who have never heard about di↵erential forms can skip Chapter 3, although of course they
will miss a nice facet of the global picture. Chapter 5 on simplicial homology and cohomology
was included mostly for historical sake, and because they have a strong combinatorial and
computational flavor. Chapter 6 on CW-complexes was included to show that there are
tools for computing homology goups and to compensate for the lack of computational flavor
of singular homology. However, CW-complexes can’t really be understood without a good
knowledge of singular homology.
Our feeling is that singular homology is simpler to define than the other homology the-
ories, and since it is also more general, we decided to choose it as our first presentation of
homology.
Our main goal is really to discuss cohomology, but except for de Rham cohomology, we
feel that a two step process where we first present singular homology, and then singular
cohomology as the result of applying the functor Hom( , G), is less abrupt than discussing
Čech cohomology (or Alexander–Spanier cohomology) first. If the reader prefers, he/she
may to go directly to Chapter 9.
In any case, we highly recommend first reading the first four sections of Chapter 2.
Sections 2.7 and 2.2 can be skipped upon first reading. Next, either proceed with Chapter
3, or skip it, but read Chapter 4 entirely.
After this, we recommend reading Chapter 7 on Poincaré duality, since this is one of the
jewels of algebraic topology.
1.13. SUGGESTIONS ON HOW TO USE THIS BOOK 57

Knowledge about manifolds is not necessary to read this book but definitely useful since
manifolds form a large class of spaces for which all the main cohomology theories are equiva-
lent. Among the many books that cover manifolds, we suggest (in alphabetic order) Lee [36],
Morita [46], Tu [61], and Warner [62]. A detailed presentation, first at a basic level and then
at a more advanced level is also provided in Gallier and Quaintance [20]. Chapter 3 requires
knowledge of di↵erential forms on smooth manifolds. Di↵erential forms are discussed in Tu
[61], Morita [46], Madsen and Tornehave [39], and Bott and Tu [4]. A detailed exposition,
including an extensive review of tensor algebra, is also provided in Gallier and Quaintance
[21]. A firm grasp of linear algebra and of some commutative algebra, at the level discussed
in texts such as Artin [2] and Dummit and Foote [14], is required.
The second part, starting with presheaves and sheaves in Chapter 8, relies on more
algebra, especially Chapter 11 on derived functors and Chapter 15 on spectral sequences.
However, this is some of the most beautiful material, so do not be discouraged if the going
is tough. Skip proofs upon first reading and try to plow through as much as possible. Stop
to take a break, and go back!
One of our goals is to fully prepare the reader to read books like Hartshorne [30] (Chapter
III). Others have expressed the same goal, and we hope to be more successful.
We have borrowed some proofs of Steve Shatz from Shatz and Gallier [58], and many
proofs in Chapter 11 are borrowed from Rotman [50, 52]. Generally, we relied heavily on
Bott and Tu [4], Bredon [7], Godement [24], Hatcher [31], Milnor and Stashe↵ [45], Munkres
[48], Serre [55], Spanier [59], Tennison [60], and Warner [62]. These are wonderful books,
and we hope that reading our book will prepare the reader to study them. We express our
gratitude to these authors, and to all the others that have inspired us (including, of course,
Dieudonné).
Since we made the decision not to include all proofs (this would have doubled if not
tripled the size of the book!), we tried very hard to provide precise pointers to all omitted
proofs. This may be irritating to the expert, but we believe that a reader with less knowledge
will appreciate this. The reason for including a proof is that we feel that it presents a type
of argument that the reader should be exposed to, but this often subjective and a reflection
of our personal taste. When we omitted a proof, we tried to give an idea of what it would
be, except when it was a really difficult proof. This should be an incentive for the reader to
dig into these references.
58 CHAPTER 1. INTRODUCTION